]>
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 | ||
b5efdb8a | 7 | #include "alloc-util.h" |
18c528e9 | 8 | #include "blockdev-util.h" |
d8b4d14d | 9 | #include "bpf-devices.h" |
906c06f6 | 10 | #include "bpf-firewall.h" |
506ea51b | 11 | #include "bpf-foreign.h" |
45c2e068 | 12 | #include "btrfs-util.h" |
6592b975 | 13 | #include "bus-error.h" |
fdb3deca | 14 | #include "cgroup-setup.h" |
03a7b521 | 15 | #include "cgroup-util.h" |
3ffd4af2 LP |
16 | #include "cgroup.h" |
17 | #include "fd-util.h" | |
0d39fa9c | 18 | #include "fileio.h" |
77601719 | 19 | #include "fs-util.h" |
d9e45bc3 | 20 | #include "io-util.h" |
3a0f06c4 | 21 | #include "limits-util.h" |
d9e45bc3 | 22 | #include "nulstr-util.h" |
6bedfcbb | 23 | #include "parse-util.h" |
9eb977db | 24 | #include "path-util.h" |
1ead0b2a | 25 | #include "percent-util.h" |
03a7b521 | 26 | #include "process-util.h" |
c36a69f4 | 27 | #include "procfs-util.h" |
a8e5eb17 | 28 | #include "socket-bind.h" |
9444b1f2 | 29 | #include "special.h" |
74c48bf5 | 30 | #include "stat-util.h" |
906c06f6 | 31 | #include "stdio-util.h" |
8b43440b | 32 | #include "string-table.h" |
07630cea | 33 | #include "string-util.h" |
cc6271f1 | 34 | #include "virt.h" |
8e274523 | 35 | |
10f28641 | 36 | #define CGROUP_CPU_QUOTA_DEFAULT_PERIOD_USEC ((usec_t) 100 * USEC_PER_MSEC) |
9a054909 | 37 | |
39b9fefb LP |
38 | /* Returns the log level to use when cgroup attribute writes fail. When an attribute is missing or we have access |
39 | * problems we downgrade to LOG_DEBUG. This is supposed to be nice to container managers and kernels which want to mask | |
40 | * out specific attributes from us. */ | |
41 | #define LOG_LEVEL_CGROUP_WRITE(r) (IN_SET(abs(r), ENOENT, EROFS, EACCES, EPERM) ? LOG_DEBUG : LOG_WARNING) | |
42 | ||
3a0f06c4 ZJS |
43 | uint64_t tasks_max_resolve(const TasksMax *tasks_max) { |
44 | if (tasks_max->scale == 0) | |
45 | return tasks_max->value; | |
46 | ||
47 | return system_tasks_max_scale(tasks_max->value, tasks_max->scale); | |
48 | } | |
49 | ||
611c4f8a | 50 | bool manager_owns_host_root_cgroup(Manager *m) { |
cc6271f1 LP |
51 | assert(m); |
52 | ||
53 | /* Returns true if we are managing the root cgroup. Note that it isn't sufficient to just check whether the | |
54 | * group root path equals "/" since that will also be the case if CLONE_NEWCGROUP is in the mix. Since there's | |
55 | * appears to be no nice way to detect whether we are in a CLONE_NEWCGROUP namespace we instead just check if | |
56 | * we run in any kind of container virtualization. */ | |
57 | ||
28cfdc5a LP |
58 | if (MANAGER_IS_USER(m)) |
59 | return false; | |
60 | ||
cc6271f1 LP |
61 | if (detect_container() > 0) |
62 | return false; | |
63 | ||
57ea45e1 | 64 | return empty_or_root(m->cgroup_root); |
cc6271f1 LP |
65 | } |
66 | ||
611c4f8a | 67 | bool unit_has_host_root_cgroup(Unit *u) { |
f3725e64 LP |
68 | assert(u); |
69 | ||
cc6271f1 LP |
70 | /* Returns whether this unit manages the root cgroup. This will return true if this unit is the root slice and |
71 | * the manager manages the root cgroup. */ | |
f3725e64 | 72 | |
611c4f8a | 73 | if (!manager_owns_host_root_cgroup(u->manager)) |
f3725e64 LP |
74 | return false; |
75 | ||
cc6271f1 | 76 | return unit_has_name(u, SPECIAL_ROOT_SLICE); |
f3725e64 LP |
77 | } |
78 | ||
293d32df LP |
79 | static int set_attribute_and_warn(Unit *u, const char *controller, const char *attribute, const char *value) { |
80 | int r; | |
81 | ||
82 | r = cg_set_attribute(controller, u->cgroup_path, attribute, value); | |
83 | if (r < 0) | |
8ed6f81b YW |
84 | log_unit_full_errno(u, LOG_LEVEL_CGROUP_WRITE(r), r, "Failed to set '%s' attribute on '%s' to '%.*s': %m", |
85 | strna(attribute), isempty(u->cgroup_path) ? "/" : u->cgroup_path, (int) strcspn(value, NEWLINE), value); | |
293d32df LP |
86 | |
87 | return r; | |
88 | } | |
89 | ||
2b40998d | 90 | static void cgroup_compat_warn(void) { |
128fadc9 TH |
91 | static bool cgroup_compat_warned = false; |
92 | ||
93 | if (cgroup_compat_warned) | |
94 | return; | |
95 | ||
cc6271f1 LP |
96 | log_warning("cgroup compatibility translation between legacy and unified hierarchy settings activated. " |
97 | "See cgroup-compat debug messages for details."); | |
98 | ||
128fadc9 TH |
99 | cgroup_compat_warned = true; |
100 | } | |
101 | ||
102 | #define log_cgroup_compat(unit, fmt, ...) do { \ | |
103 | cgroup_compat_warn(); \ | |
104 | log_unit_debug(unit, "cgroup-compat: " fmt, ##__VA_ARGS__); \ | |
2b40998d | 105 | } while (false) |
128fadc9 | 106 | |
4ad49000 LP |
107 | void cgroup_context_init(CGroupContext *c) { |
108 | assert(c); | |
109 | ||
de8a711a | 110 | /* Initialize everything to the kernel defaults. */ |
4ad49000 | 111 | |
de8a711a LP |
112 | *c = (CGroupContext) { |
113 | .cpu_weight = CGROUP_WEIGHT_INVALID, | |
114 | .startup_cpu_weight = CGROUP_WEIGHT_INVALID, | |
115 | .cpu_quota_per_sec_usec = USEC_INFINITY, | |
10f28641 | 116 | .cpu_quota_period_usec = USEC_INFINITY, |
66ebf6c0 | 117 | |
de8a711a LP |
118 | .cpu_shares = CGROUP_CPU_SHARES_INVALID, |
119 | .startup_cpu_shares = CGROUP_CPU_SHARES_INVALID, | |
d53d9474 | 120 | |
de8a711a LP |
121 | .memory_high = CGROUP_LIMIT_MAX, |
122 | .memory_max = CGROUP_LIMIT_MAX, | |
123 | .memory_swap_max = CGROUP_LIMIT_MAX, | |
da4d897e | 124 | |
de8a711a | 125 | .memory_limit = CGROUP_LIMIT_MAX, |
b2f8b02e | 126 | |
de8a711a LP |
127 | .io_weight = CGROUP_WEIGHT_INVALID, |
128 | .startup_io_weight = CGROUP_WEIGHT_INVALID, | |
13c31542 | 129 | |
de8a711a LP |
130 | .blockio_weight = CGROUP_BLKIO_WEIGHT_INVALID, |
131 | .startup_blockio_weight = CGROUP_BLKIO_WEIGHT_INVALID, | |
d53d9474 | 132 | |
3a0f06c4 | 133 | .tasks_max = TASKS_MAX_UNSET, |
4d824a4e AZ |
134 | |
135 | .moom_swap = MANAGED_OOM_AUTO, | |
136 | .moom_mem_pressure = MANAGED_OOM_AUTO, | |
4e806bfa | 137 | .moom_preference = MANAGED_OOM_PREFERENCE_NONE, |
de8a711a | 138 | }; |
4ad49000 | 139 | } |
8e274523 | 140 | |
4ad49000 LP |
141 | void cgroup_context_free_device_allow(CGroupContext *c, CGroupDeviceAllow *a) { |
142 | assert(c); | |
143 | assert(a); | |
144 | ||
71fda00f | 145 | LIST_REMOVE(device_allow, c->device_allow, a); |
4ad49000 LP |
146 | free(a->path); |
147 | free(a); | |
148 | } | |
149 | ||
13c31542 TH |
150 | void cgroup_context_free_io_device_weight(CGroupContext *c, CGroupIODeviceWeight *w) { |
151 | assert(c); | |
152 | assert(w); | |
153 | ||
154 | LIST_REMOVE(device_weights, c->io_device_weights, w); | |
155 | free(w->path); | |
156 | free(w); | |
157 | } | |
158 | ||
6ae4283c TH |
159 | void cgroup_context_free_io_device_latency(CGroupContext *c, CGroupIODeviceLatency *l) { |
160 | assert(c); | |
161 | assert(l); | |
162 | ||
163 | LIST_REMOVE(device_latencies, c->io_device_latencies, l); | |
164 | free(l->path); | |
165 | free(l); | |
166 | } | |
167 | ||
13c31542 TH |
168 | void cgroup_context_free_io_device_limit(CGroupContext *c, CGroupIODeviceLimit *l) { |
169 | assert(c); | |
170 | assert(l); | |
171 | ||
172 | LIST_REMOVE(device_limits, c->io_device_limits, l); | |
173 | free(l->path); | |
174 | free(l); | |
175 | } | |
176 | ||
4ad49000 LP |
177 | void cgroup_context_free_blockio_device_weight(CGroupContext *c, CGroupBlockIODeviceWeight *w) { |
178 | assert(c); | |
179 | assert(w); | |
180 | ||
71fda00f | 181 | LIST_REMOVE(device_weights, c->blockio_device_weights, w); |
4ad49000 LP |
182 | free(w->path); |
183 | free(w); | |
184 | } | |
185 | ||
186 | void cgroup_context_free_blockio_device_bandwidth(CGroupContext *c, CGroupBlockIODeviceBandwidth *b) { | |
187 | assert(c); | |
8e274523 | 188 | assert(b); |
8e274523 | 189 | |
71fda00f | 190 | LIST_REMOVE(device_bandwidths, c->blockio_device_bandwidths, b); |
4ad49000 LP |
191 | free(b->path); |
192 | free(b); | |
193 | } | |
194 | ||
b894ef1b JK |
195 | void cgroup_context_remove_bpf_foreign_program(CGroupContext *c, CGroupBPFForeignProgram *p) { |
196 | assert(c); | |
197 | assert(p); | |
198 | ||
199 | LIST_REMOVE(programs, c->bpf_foreign_programs, p); | |
200 | free(p->bpffs_path); | |
201 | free(p); | |
202 | } | |
203 | ||
b18e9fc1 JK |
204 | void cgroup_context_remove_socket_bind(CGroupSocketBindItem **head) { |
205 | CGroupSocketBindItem *h; | |
206 | ||
207 | assert(head); | |
208 | ||
209 | while (*head) { | |
210 | h = *head; | |
211 | LIST_REMOVE(socket_bind_items, *head, h); | |
212 | free(h); | |
213 | } | |
214 | } | |
215 | ||
4ad49000 LP |
216 | void cgroup_context_done(CGroupContext *c) { |
217 | assert(c); | |
218 | ||
13c31542 TH |
219 | while (c->io_device_weights) |
220 | cgroup_context_free_io_device_weight(c, c->io_device_weights); | |
221 | ||
6ae4283c TH |
222 | while (c->io_device_latencies) |
223 | cgroup_context_free_io_device_latency(c, c->io_device_latencies); | |
224 | ||
13c31542 TH |
225 | while (c->io_device_limits) |
226 | cgroup_context_free_io_device_limit(c, c->io_device_limits); | |
227 | ||
4ad49000 LP |
228 | while (c->blockio_device_weights) |
229 | cgroup_context_free_blockio_device_weight(c, c->blockio_device_weights); | |
230 | ||
231 | while (c->blockio_device_bandwidths) | |
232 | cgroup_context_free_blockio_device_bandwidth(c, c->blockio_device_bandwidths); | |
233 | ||
234 | while (c->device_allow) | |
235 | cgroup_context_free_device_allow(c, c->device_allow); | |
6a48d82f | 236 | |
b18e9fc1 JK |
237 | cgroup_context_remove_socket_bind(&c->socket_bind_allow); |
238 | cgroup_context_remove_socket_bind(&c->socket_bind_deny); | |
239 | ||
6a48d82f DM |
240 | c->ip_address_allow = ip_address_access_free_all(c->ip_address_allow); |
241 | c->ip_address_deny = ip_address_access_free_all(c->ip_address_deny); | |
fab34748 KL |
242 | |
243 | c->ip_filters_ingress = strv_free(c->ip_filters_ingress); | |
244 | c->ip_filters_egress = strv_free(c->ip_filters_egress); | |
047f5d63 | 245 | |
b894ef1b JK |
246 | while (c->bpf_foreign_programs) |
247 | cgroup_context_remove_bpf_foreign_program(c, c->bpf_foreign_programs); | |
248 | ||
047f5d63 PH |
249 | cpu_set_reset(&c->cpuset_cpus); |
250 | cpu_set_reset(&c->cpuset_mems); | |
4ad49000 LP |
251 | } |
252 | ||
74b5fb27 | 253 | static int unit_get_kernel_memory_limit(Unit *u, const char *file, uint64_t *ret) { |
74b5fb27 CD |
254 | assert(u); |
255 | ||
256 | if (!u->cgroup_realized) | |
257 | return -EOWNERDEAD; | |
258 | ||
613328c3 | 259 | return cg_get_attribute_as_uint64("memory", u->cgroup_path, file, ret); |
74b5fb27 CD |
260 | } |
261 | ||
262 | static int unit_compare_memory_limit(Unit *u, const char *property_name, uint64_t *ret_unit_value, uint64_t *ret_kernel_value) { | |
263 | CGroupContext *c; | |
264 | CGroupMask m; | |
265 | const char *file; | |
266 | uint64_t unit_value; | |
267 | int r; | |
268 | ||
269 | /* Compare kernel memcg configuration against our internal systemd state. Unsupported (and will | |
270 | * return -ENODATA) on cgroup v1. | |
271 | * | |
272 | * Returns: | |
273 | * | |
274 | * <0: On error. | |
275 | * 0: If the kernel memory setting doesn't match our configuration. | |
276 | * >0: If the kernel memory setting matches our configuration. | |
277 | * | |
278 | * The following values are only guaranteed to be populated on return >=0: | |
279 | * | |
280 | * - ret_unit_value will contain our internal expected value for the unit, page-aligned. | |
281 | * - ret_kernel_value will contain the actual value presented by the kernel. */ | |
282 | ||
283 | assert(u); | |
284 | ||
285 | r = cg_all_unified(); | |
286 | if (r < 0) | |
287 | return log_debug_errno(r, "Failed to determine cgroup hierarchy version: %m"); | |
288 | ||
289 | /* Unsupported on v1. | |
290 | * | |
291 | * We don't return ENOENT, since that could actually mask a genuine problem where somebody else has | |
292 | * silently masked the controller. */ | |
293 | if (r == 0) | |
294 | return -ENODATA; | |
295 | ||
296 | /* The root slice doesn't have any controller files, so we can't compare anything. */ | |
297 | if (unit_has_name(u, SPECIAL_ROOT_SLICE)) | |
298 | return -ENODATA; | |
299 | ||
300 | /* It's possible to have MemoryFoo set without systemd wanting to have the memory controller enabled, | |
301 | * for example, in the case of DisableControllers= or cgroup_disable on the kernel command line. To | |
302 | * avoid specious errors in these scenarios, check that we even expect the memory controller to be | |
303 | * enabled at all. */ | |
304 | m = unit_get_target_mask(u); | |
305 | if (!FLAGS_SET(m, CGROUP_MASK_MEMORY)) | |
306 | return -ENODATA; | |
307 | ||
806a9362 | 308 | assert_se(c = unit_get_cgroup_context(u)); |
74b5fb27 CD |
309 | |
310 | if (streq(property_name, "MemoryLow")) { | |
311 | unit_value = unit_get_ancestor_memory_low(u); | |
312 | file = "memory.low"; | |
313 | } else if (streq(property_name, "MemoryMin")) { | |
314 | unit_value = unit_get_ancestor_memory_min(u); | |
315 | file = "memory.min"; | |
316 | } else if (streq(property_name, "MemoryHigh")) { | |
317 | unit_value = c->memory_high; | |
318 | file = "memory.high"; | |
319 | } else if (streq(property_name, "MemoryMax")) { | |
320 | unit_value = c->memory_max; | |
321 | file = "memory.max"; | |
322 | } else if (streq(property_name, "MemorySwapMax")) { | |
323 | unit_value = c->memory_swap_max; | |
324 | file = "memory.swap.max"; | |
325 | } else | |
326 | return -EINVAL; | |
327 | ||
328 | r = unit_get_kernel_memory_limit(u, file, ret_kernel_value); | |
329 | if (r < 0) | |
330 | return log_unit_debug_errno(u, r, "Failed to parse %s: %m", file); | |
331 | ||
332 | /* It's intended (soon) in a future kernel to not expose cgroup memory limits rounded to page | |
333 | * boundaries, but instead separate the user-exposed limit, which is whatever userspace told us, from | |
334 | * our internal page-counting. To support those future kernels, just check the value itself first | |
335 | * without any page-alignment. */ | |
336 | if (*ret_kernel_value == unit_value) { | |
337 | *ret_unit_value = unit_value; | |
338 | return 1; | |
339 | } | |
340 | ||
341 | /* The current kernel behaviour, by comparison, is that even if you write a particular number of | |
342 | * bytes into a cgroup memory file, it always returns that number page-aligned down (since the kernel | |
343 | * internally stores cgroup limits in pages). As such, so long as it aligns properly, everything is | |
344 | * cricket. */ | |
345 | if (unit_value != CGROUP_LIMIT_MAX) | |
346 | unit_value = PAGE_ALIGN_DOWN(unit_value); | |
347 | ||
348 | *ret_unit_value = unit_value; | |
349 | ||
350 | return *ret_kernel_value == *ret_unit_value; | |
351 | } | |
352 | ||
bc0623df CD |
353 | #define FORMAT_CGROUP_DIFF_MAX 128 |
354 | ||
355 | static char *format_cgroup_memory_limit_comparison(char *buf, size_t l, Unit *u, const char *property_name) { | |
356 | uint64_t kval, sval; | |
357 | int r; | |
358 | ||
359 | assert(u); | |
360 | assert(buf); | |
361 | assert(l > 0); | |
362 | ||
363 | r = unit_compare_memory_limit(u, property_name, &sval, &kval); | |
364 | ||
365 | /* memory.swap.max is special in that it relies on CONFIG_MEMCG_SWAP (and the default swapaccount=1). | |
366 | * In the absence of reliably being able to detect whether memcg swap support is available or not, | |
367 | * only complain if the error is not ENOENT. */ | |
368 | if (r > 0 || IN_SET(r, -ENODATA, -EOWNERDEAD) || | |
369 | (r == -ENOENT && streq(property_name, "MemorySwapMax"))) { | |
370 | buf[0] = 0; | |
371 | return buf; | |
372 | } | |
373 | ||
374 | if (r < 0) { | |
375 | snprintf(buf, l, " (error getting kernel value: %s)", strerror_safe(r)); | |
376 | return buf; | |
377 | } | |
378 | ||
379 | snprintf(buf, l, " (different value in kernel: %" PRIu64 ")", kval); | |
380 | ||
381 | return buf; | |
382 | } | |
383 | ||
384 | void cgroup_context_dump(Unit *u, FILE* f, const char *prefix) { | |
85c3b278 | 385 | _cleanup_free_ char *disable_controllers_str = NULL, *cpuset_cpus = NULL, *cpuset_mems = NULL; |
13c31542 TH |
386 | CGroupIODeviceLimit *il; |
387 | CGroupIODeviceWeight *iw; | |
6ae4283c | 388 | CGroupIODeviceLatency *l; |
4ad49000 LP |
389 | CGroupBlockIODeviceBandwidth *b; |
390 | CGroupBlockIODeviceWeight *w; | |
b894ef1b | 391 | CGroupBPFForeignProgram *p; |
4ad49000 | 392 | CGroupDeviceAllow *a; |
bc0623df | 393 | CGroupContext *c; |
b18e9fc1 | 394 | CGroupSocketBindItem *bi; |
c21c9906 | 395 | IPAddressAccessItem *iaai; |
fab34748 | 396 | char **path; |
bc0623df | 397 | char q[FORMAT_TIMESPAN_MAX]; |
10f28641 | 398 | char v[FORMAT_TIMESPAN_MAX]; |
4ad49000 | 399 | |
bc0623df CD |
400 | char cda[FORMAT_CGROUP_DIFF_MAX]; |
401 | char cdb[FORMAT_CGROUP_DIFF_MAX]; | |
402 | char cdc[FORMAT_CGROUP_DIFF_MAX]; | |
403 | char cdd[FORMAT_CGROUP_DIFF_MAX]; | |
404 | char cde[FORMAT_CGROUP_DIFF_MAX]; | |
405 | ||
406 | assert(u); | |
4ad49000 LP |
407 | assert(f); |
408 | ||
806a9362 | 409 | assert_se(c = unit_get_cgroup_context(u)); |
bc0623df | 410 | |
4ad49000 LP |
411 | prefix = strempty(prefix); |
412 | ||
25cc30c4 AZ |
413 | (void) cg_mask_to_string(c->disable_controllers, &disable_controllers_str); |
414 | ||
047f5d63 PH |
415 | cpuset_cpus = cpu_set_to_range_string(&c->cpuset_cpus); |
416 | cpuset_mems = cpu_set_to_range_string(&c->cpuset_mems); | |
417 | ||
4ad49000 | 418 | fprintf(f, |
6dfb9282 CD |
419 | "%sCPUAccounting: %s\n" |
420 | "%sIOAccounting: %s\n" | |
421 | "%sBlockIOAccounting: %s\n" | |
422 | "%sMemoryAccounting: %s\n" | |
423 | "%sTasksAccounting: %s\n" | |
424 | "%sIPAccounting: %s\n" | |
425 | "%sCPUWeight: %" PRIu64 "\n" | |
426 | "%sStartupCPUWeight: %" PRIu64 "\n" | |
427 | "%sCPUShares: %" PRIu64 "\n" | |
428 | "%sStartupCPUShares: %" PRIu64 "\n" | |
429 | "%sCPUQuotaPerSecSec: %s\n" | |
430 | "%sCPUQuotaPeriodSec: %s\n" | |
431 | "%sAllowedCPUs: %s\n" | |
432 | "%sAllowedMemoryNodes: %s\n" | |
433 | "%sIOWeight: %" PRIu64 "\n" | |
434 | "%sStartupIOWeight: %" PRIu64 "\n" | |
435 | "%sBlockIOWeight: %" PRIu64 "\n" | |
436 | "%sStartupBlockIOWeight: %" PRIu64 "\n" | |
437 | "%sDefaultMemoryMin: %" PRIu64 "\n" | |
438 | "%sDefaultMemoryLow: %" PRIu64 "\n" | |
bc0623df CD |
439 | "%sMemoryMin: %" PRIu64 "%s\n" |
440 | "%sMemoryLow: %" PRIu64 "%s\n" | |
441 | "%sMemoryHigh: %" PRIu64 "%s\n" | |
442 | "%sMemoryMax: %" PRIu64 "%s\n" | |
443 | "%sMemorySwapMax: %" PRIu64 "%s\n" | |
6dfb9282 CD |
444 | "%sMemoryLimit: %" PRIu64 "\n" |
445 | "%sTasksMax: %" PRIu64 "\n" | |
446 | "%sDevicePolicy: %s\n" | |
447 | "%sDisableControllers: %s\n" | |
4d824a4e AZ |
448 | "%sDelegate: %s\n" |
449 | "%sManagedOOMSwap: %s\n" | |
450 | "%sManagedOOMMemoryPressure: %s\n" | |
d9d3f05d | 451 | "%sManagedOOMMemoryPressureLimit: " PERMYRIAD_AS_PERCENT_FORMAT_STR "\n" |
0f6bb1ed | 452 | "%sManagedOOMPreference: %s\n", |
4ad49000 | 453 | prefix, yes_no(c->cpu_accounting), |
13c31542 | 454 | prefix, yes_no(c->io_accounting), |
4ad49000 LP |
455 | prefix, yes_no(c->blockio_accounting), |
456 | prefix, yes_no(c->memory_accounting), | |
d53d9474 | 457 | prefix, yes_no(c->tasks_accounting), |
c21c9906 | 458 | prefix, yes_no(c->ip_accounting), |
66ebf6c0 TH |
459 | prefix, c->cpu_weight, |
460 | prefix, c->startup_cpu_weight, | |
4ad49000 | 461 | prefix, c->cpu_shares, |
95ae05c0 | 462 | prefix, c->startup_cpu_shares, |
bc0623df | 463 | prefix, format_timespan(q, sizeof(q), c->cpu_quota_per_sec_usec, 1), |
10f28641 | 464 | prefix, format_timespan(v, sizeof(v), c->cpu_quota_period_usec, 1), |
85c3b278 LP |
465 | prefix, strempty(cpuset_cpus), |
466 | prefix, strempty(cpuset_mems), | |
13c31542 TH |
467 | prefix, c->io_weight, |
468 | prefix, c->startup_io_weight, | |
4ad49000 | 469 | prefix, c->blockio_weight, |
95ae05c0 | 470 | prefix, c->startup_blockio_weight, |
7ad5439e | 471 | prefix, c->default_memory_min, |
c52db42b | 472 | prefix, c->default_memory_low, |
bc0623df CD |
473 | prefix, c->memory_min, format_cgroup_memory_limit_comparison(cda, sizeof(cda), u, "MemoryMin"), |
474 | prefix, c->memory_low, format_cgroup_memory_limit_comparison(cdb, sizeof(cdb), u, "MemoryLow"), | |
475 | prefix, c->memory_high, format_cgroup_memory_limit_comparison(cdc, sizeof(cdc), u, "MemoryHigh"), | |
476 | prefix, c->memory_max, format_cgroup_memory_limit_comparison(cdd, sizeof(cdd), u, "MemoryMax"), | |
477 | prefix, c->memory_swap_max, format_cgroup_memory_limit_comparison(cde, sizeof(cde), u, "MemorySwapMax"), | |
4ad49000 | 478 | prefix, c->memory_limit, |
3a0f06c4 | 479 | prefix, tasks_max_resolve(&c->tasks_max), |
a931ad47 | 480 | prefix, cgroup_device_policy_to_string(c->device_policy), |
f4c43a81 | 481 | prefix, strempty(disable_controllers_str), |
4d824a4e AZ |
482 | prefix, yes_no(c->delegate), |
483 | prefix, managed_oom_mode_to_string(c->moom_swap), | |
484 | prefix, managed_oom_mode_to_string(c->moom_mem_pressure), | |
d9d3f05d | 485 | prefix, PERMYRIAD_AS_PERCENT_FORMAT_VAL(UINT32_SCALE_TO_PERMYRIAD(c->moom_mem_pressure_limit)), |
4e806bfa | 486 | prefix, managed_oom_preference_to_string(c->moom_preference)); |
4ad49000 | 487 | |
02638280 LP |
488 | if (c->delegate) { |
489 | _cleanup_free_ char *t = NULL; | |
490 | ||
491 | (void) cg_mask_to_string(c->delegate_controllers, &t); | |
492 | ||
6dfb9282 | 493 | fprintf(f, "%sDelegateControllers: %s\n", |
02638280 LP |
494 | prefix, |
495 | strempty(t)); | |
496 | } | |
497 | ||
4ad49000 LP |
498 | LIST_FOREACH(device_allow, a, c->device_allow) |
499 | fprintf(f, | |
6dfb9282 | 500 | "%sDeviceAllow: %s %s%s%s\n", |
4ad49000 LP |
501 | prefix, |
502 | a->path, | |
503 | a->r ? "r" : "", a->w ? "w" : "", a->m ? "m" : ""); | |
504 | ||
13c31542 TH |
505 | LIST_FOREACH(device_weights, iw, c->io_device_weights) |
506 | fprintf(f, | |
6dfb9282 | 507 | "%sIODeviceWeight: %s %" PRIu64 "\n", |
13c31542 TH |
508 | prefix, |
509 | iw->path, | |
510 | iw->weight); | |
511 | ||
6ae4283c TH |
512 | LIST_FOREACH(device_latencies, l, c->io_device_latencies) |
513 | fprintf(f, | |
6dfb9282 | 514 | "%sIODeviceLatencyTargetSec: %s %s\n", |
6ae4283c TH |
515 | prefix, |
516 | l->path, | |
bc0623df | 517 | format_timespan(q, sizeof(q), l->target_usec, 1)); |
6ae4283c | 518 | |
13c31542 TH |
519 | LIST_FOREACH(device_limits, il, c->io_device_limits) { |
520 | char buf[FORMAT_BYTES_MAX]; | |
9be57249 TH |
521 | CGroupIOLimitType type; |
522 | ||
523 | for (type = 0; type < _CGROUP_IO_LIMIT_TYPE_MAX; type++) | |
524 | if (il->limits[type] != cgroup_io_limit_defaults[type]) | |
525 | fprintf(f, | |
6dfb9282 | 526 | "%s%s: %s %s\n", |
9be57249 TH |
527 | prefix, |
528 | cgroup_io_limit_type_to_string(type), | |
529 | il->path, | |
530 | format_bytes(buf, sizeof(buf), il->limits[type])); | |
13c31542 TH |
531 | } |
532 | ||
4ad49000 LP |
533 | LIST_FOREACH(device_weights, w, c->blockio_device_weights) |
534 | fprintf(f, | |
6dfb9282 | 535 | "%sBlockIODeviceWeight: %s %" PRIu64, |
4ad49000 LP |
536 | prefix, |
537 | w->path, | |
538 | w->weight); | |
539 | ||
540 | LIST_FOREACH(device_bandwidths, b, c->blockio_device_bandwidths) { | |
541 | char buf[FORMAT_BYTES_MAX]; | |
542 | ||
979d0311 TH |
543 | if (b->rbps != CGROUP_LIMIT_MAX) |
544 | fprintf(f, | |
6dfb9282 | 545 | "%sBlockIOReadBandwidth: %s %s\n", |
979d0311 TH |
546 | prefix, |
547 | b->path, | |
548 | format_bytes(buf, sizeof(buf), b->rbps)); | |
549 | if (b->wbps != CGROUP_LIMIT_MAX) | |
550 | fprintf(f, | |
6dfb9282 | 551 | "%sBlockIOWriteBandwidth: %s %s\n", |
979d0311 TH |
552 | prefix, |
553 | b->path, | |
554 | format_bytes(buf, sizeof(buf), b->wbps)); | |
4ad49000 | 555 | } |
c21c9906 LP |
556 | |
557 | LIST_FOREACH(items, iaai, c->ip_address_allow) { | |
558 | _cleanup_free_ char *k = NULL; | |
559 | ||
560 | (void) in_addr_to_string(iaai->family, &iaai->address, &k); | |
6dfb9282 | 561 | fprintf(f, "%sIPAddressAllow: %s/%u\n", prefix, strnull(k), iaai->prefixlen); |
c21c9906 LP |
562 | } |
563 | ||
564 | LIST_FOREACH(items, iaai, c->ip_address_deny) { | |
565 | _cleanup_free_ char *k = NULL; | |
566 | ||
567 | (void) in_addr_to_string(iaai->family, &iaai->address, &k); | |
6dfb9282 | 568 | fprintf(f, "%sIPAddressDeny: %s/%u\n", prefix, strnull(k), iaai->prefixlen); |
c21c9906 | 569 | } |
fab34748 KL |
570 | |
571 | STRV_FOREACH(path, c->ip_filters_ingress) | |
6dfb9282 | 572 | fprintf(f, "%sIPIngressFilterPath: %s\n", prefix, *path); |
fab34748 KL |
573 | |
574 | STRV_FOREACH(path, c->ip_filters_egress) | |
6dfb9282 | 575 | fprintf(f, "%sIPEgressFilterPath: %s\n", prefix, *path); |
b894ef1b JK |
576 | |
577 | LIST_FOREACH(programs, p, c->bpf_foreign_programs) | |
578 | fprintf(f, "%sBPFProgram: %s:%s", | |
579 | prefix, bpf_cgroup_attach_type_to_string(p->attach_type), p->bpffs_path); | |
b18e9fc1 JK |
580 | |
581 | if (c->socket_bind_allow) { | |
582 | fprintf(f, "%sSocketBindAllow:", prefix); | |
583 | LIST_FOREACH(socket_bind_items, bi, c->socket_bind_allow) | |
584 | cgroup_context_dump_socket_bind_item(bi, f); | |
585 | fputc('\n', f); | |
586 | } | |
587 | ||
588 | if (c->socket_bind_deny) { | |
589 | fprintf(f, "%sSocketBindDeny:", prefix); | |
590 | LIST_FOREACH(socket_bind_items, bi, c->socket_bind_deny) | |
591 | cgroup_context_dump_socket_bind_item(bi, f); | |
592 | fputc('\n', f); | |
593 | } | |
594 | } | |
595 | ||
596 | void cgroup_context_dump_socket_bind_item(const CGroupSocketBindItem *item, FILE *f) { | |
f80a206a LP |
597 | const char *family = |
598 | item->address_family == AF_INET ? "ipv4:" : | |
599 | item->address_family == AF_INET6 ? "ipv6:" : ""; | |
b18e9fc1 JK |
600 | |
601 | if (item->nr_ports == 0) | |
602 | fprintf(f, " %sany", family); | |
603 | else if (item->nr_ports == 1) | |
604 | fprintf(f, " %s%" PRIu16, family, item->port_min); | |
605 | else { | |
606 | uint16_t port_max = item->port_min + item->nr_ports - 1; | |
607 | fprintf(f, " %s%" PRIu16 "-%" PRIu16, family, item->port_min, port_max); | |
608 | } | |
4ad49000 LP |
609 | } |
610 | ||
fd870bac YW |
611 | int cgroup_add_device_allow(CGroupContext *c, const char *dev, const char *mode) { |
612 | _cleanup_free_ CGroupDeviceAllow *a = NULL; | |
613 | _cleanup_free_ char *d = NULL; | |
614 | ||
615 | assert(c); | |
616 | assert(dev); | |
617 | assert(isempty(mode) || in_charset(mode, "rwm")); | |
618 | ||
619 | a = new(CGroupDeviceAllow, 1); | |
620 | if (!a) | |
621 | return -ENOMEM; | |
622 | ||
623 | d = strdup(dev); | |
624 | if (!d) | |
625 | return -ENOMEM; | |
626 | ||
627 | *a = (CGroupDeviceAllow) { | |
628 | .path = TAKE_PTR(d), | |
490c5a37 LP |
629 | .r = isempty(mode) || strchr(mode, 'r'), |
630 | .w = isempty(mode) || strchr(mode, 'w'), | |
631 | .m = isempty(mode) || strchr(mode, 'm'), | |
fd870bac YW |
632 | }; |
633 | ||
634 | LIST_PREPEND(device_allow, c->device_allow, a); | |
635 | TAKE_PTR(a); | |
636 | ||
637 | return 0; | |
638 | } | |
639 | ||
b894ef1b JK |
640 | int cgroup_add_bpf_foreign_program(CGroupContext *c, uint32_t attach_type, const char *bpffs_path) { |
641 | CGroupBPFForeignProgram *p; | |
642 | _cleanup_free_ char *d = NULL; | |
643 | ||
644 | assert(c); | |
645 | assert(bpffs_path); | |
646 | ||
647 | if (!path_is_normalized(bpffs_path) || !path_is_absolute(bpffs_path)) | |
648 | return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "Path is not normalized: %m"); | |
649 | ||
650 | d = strdup(bpffs_path); | |
651 | if (!d) | |
652 | return log_oom(); | |
653 | ||
654 | p = new(CGroupBPFForeignProgram, 1); | |
655 | if (!p) | |
656 | return log_oom(); | |
657 | ||
658 | *p = (CGroupBPFForeignProgram) { | |
659 | .attach_type = attach_type, | |
660 | .bpffs_path = TAKE_PTR(d), | |
661 | }; | |
662 | ||
663 | LIST_PREPEND(programs, c->bpf_foreign_programs, TAKE_PTR(p)); | |
664 | ||
665 | return 0; | |
666 | } | |
667 | ||
6264b85e CD |
668 | #define UNIT_DEFINE_ANCESTOR_MEMORY_LOOKUP(entry) \ |
669 | uint64_t unit_get_ancestor_##entry(Unit *u) { \ | |
670 | CGroupContext *c; \ | |
671 | \ | |
672 | /* 1. Is entry set in this unit? If so, use that. \ | |
673 | * 2. Is the default for this entry set in any \ | |
674 | * ancestor? If so, use that. \ | |
675 | * 3. Otherwise, return CGROUP_LIMIT_MIN. */ \ | |
676 | \ | |
677 | assert(u); \ | |
678 | \ | |
679 | c = unit_get_cgroup_context(u); \ | |
c5322608 | 680 | if (c && c->entry##_set) \ |
6264b85e CD |
681 | return c->entry; \ |
682 | \ | |
c5322608 | 683 | while ((u = UNIT_DEREF(u->slice))) { \ |
6264b85e | 684 | c = unit_get_cgroup_context(u); \ |
c5322608 | 685 | if (c && c->default_##entry##_set) \ |
6264b85e CD |
686 | return c->default_##entry; \ |
687 | } \ | |
688 | \ | |
689 | /* We've reached the root, but nobody had default for \ | |
690 | * this entry set, so set it to the kernel default. */ \ | |
691 | return CGROUP_LIMIT_MIN; \ | |
c52db42b CD |
692 | } |
693 | ||
6264b85e | 694 | UNIT_DEFINE_ANCESTOR_MEMORY_LOOKUP(memory_low); |
7ad5439e | 695 | UNIT_DEFINE_ANCESTOR_MEMORY_LOOKUP(memory_min); |
6264b85e | 696 | |
4e806bfa AZ |
697 | void cgroup_oomd_xattr_apply(Unit *u, const char *cgroup_path) { |
698 | CGroupContext *c; | |
699 | int r; | |
700 | ||
701 | assert(u); | |
702 | ||
703 | c = unit_get_cgroup_context(u); | |
704 | if (!c) | |
705 | return; | |
706 | ||
707 | if (c->moom_preference == MANAGED_OOM_PREFERENCE_OMIT) { | |
708 | r = cg_set_xattr(SYSTEMD_CGROUP_CONTROLLER, cgroup_path, "user.oomd_omit", "1", 1, 0); | |
709 | if (r < 0) | |
710 | log_unit_debug_errno(u, r, "Failed to set oomd_omit flag on control group %s, ignoring: %m", cgroup_path); | |
711 | } | |
712 | ||
713 | if (c->moom_preference == MANAGED_OOM_PREFERENCE_AVOID) { | |
714 | r = cg_set_xattr(SYSTEMD_CGROUP_CONTROLLER, cgroup_path, "user.oomd_avoid", "1", 1, 0); | |
715 | if (r < 0) | |
716 | log_unit_debug_errno(u, r, "Failed to set oomd_avoid flag on control group %s, ignoring: %m", cgroup_path); | |
717 | } | |
718 | ||
719 | if (c->moom_preference != MANAGED_OOM_PREFERENCE_AVOID) { | |
720 | r = cg_remove_xattr(SYSTEMD_CGROUP_CONTROLLER, cgroup_path, "user.oomd_avoid"); | |
721 | if (r != -ENODATA) | |
722 | log_unit_debug_errno(u, r, "Failed to remove oomd_avoid flag on control group %s, ignoring: %m", cgroup_path); | |
723 | } | |
724 | ||
725 | if (c->moom_preference != MANAGED_OOM_PREFERENCE_OMIT) { | |
726 | r = cg_remove_xattr(SYSTEMD_CGROUP_CONTROLLER, cgroup_path, "user.oomd_omit"); | |
727 | if (r != -ENODATA) | |
728 | log_unit_debug_errno(u, r, "Failed to remove oomd_omit flag on control group %s, ignoring: %m", cgroup_path); | |
729 | } | |
730 | } | |
731 | ||
0d2d6fbf CD |
732 | static void cgroup_xattr_apply(Unit *u) { |
733 | char ids[SD_ID128_STRING_MAX]; | |
734 | int r; | |
735 | ||
736 | assert(u); | |
737 | ||
738 | if (!MANAGER_IS_SYSTEM(u->manager)) | |
739 | return; | |
740 | ||
3288ea8f LP |
741 | if (!sd_id128_is_null(u->invocation_id)) { |
742 | r = cg_set_xattr(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path, | |
743 | "trusted.invocation_id", | |
744 | sd_id128_to_string(u->invocation_id, ids), 32, | |
745 | 0); | |
746 | if (r < 0) | |
747 | log_unit_debug_errno(u, r, "Failed to set invocation ID on control group %s, ignoring: %m", u->cgroup_path); | |
748 | } | |
0d2d6fbf | 749 | |
3288ea8f LP |
750 | if (unit_cgroup_delegate(u)) { |
751 | r = cg_set_xattr(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path, | |
752 | "trusted.delegate", | |
753 | "1", 1, | |
754 | 0); | |
755 | if (r < 0) | |
756 | log_unit_debug_errno(u, r, "Failed to set delegate flag on control group %s, ignoring: %m", u->cgroup_path); | |
757 | } else { | |
758 | r = cg_remove_xattr(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path, "trusted.delegate"); | |
759 | if (r != -ENODATA) | |
760 | log_unit_debug_errno(u, r, "Failed to remove delegate flag on control group %s, ignoring: %m", u->cgroup_path); | |
761 | } | |
4e806bfa AZ |
762 | |
763 | cgroup_oomd_xattr_apply(u, u->cgroup_path); | |
0d2d6fbf CD |
764 | } |
765 | ||
45c2e068 | 766 | static int lookup_block_device(const char *p, dev_t *ret) { |
f5855697 YS |
767 | dev_t rdev, dev = 0; |
768 | mode_t mode; | |
45c2e068 | 769 | int r; |
4ad49000 LP |
770 | |
771 | assert(p); | |
45c2e068 | 772 | assert(ret); |
4ad49000 | 773 | |
f5855697 | 774 | r = device_path_parse_major_minor(p, &mode, &rdev); |
d5aecba6 | 775 | if (r == -ENODEV) { /* not a parsable device node, need to go to disk */ |
f5855697 | 776 | struct stat st; |
57f1030b | 777 | |
d5aecba6 LP |
778 | if (stat(p, &st) < 0) |
779 | return log_warning_errno(errno, "Couldn't stat device '%s': %m", p); | |
57f1030b | 780 | |
f5855697 | 781 | mode = st.st_mode; |
a0d6590c LP |
782 | rdev = st.st_rdev; |
783 | dev = st.st_dev; | |
d5aecba6 LP |
784 | } else if (r < 0) |
785 | return log_warning_errno(r, "Failed to parse major/minor from path '%s': %m", p); | |
786 | ||
57f1030b LP |
787 | if (S_ISCHR(mode)) |
788 | return log_warning_errno(SYNTHETIC_ERRNO(ENOTBLK), | |
789 | "Device node '%s' is a character device, but block device needed.", p); | |
790 | if (S_ISBLK(mode)) | |
f5855697 YS |
791 | *ret = rdev; |
792 | else if (major(dev) != 0) | |
793 | *ret = dev; /* If this is not a device node then use the block device this file is stored on */ | |
45c2e068 LP |
794 | else { |
795 | /* If this is btrfs, getting the backing block device is a bit harder */ | |
796 | r = btrfs_get_block_device(p, ret); | |
57f1030b LP |
797 | if (r == -ENOTTY) |
798 | return log_warning_errno(SYNTHETIC_ERRNO(ENODEV), | |
799 | "'%s' is not a block device node, and file system block device cannot be determined or is not local.", p); | |
800 | if (r < 0) | |
45c2e068 | 801 | return log_warning_errno(r, "Failed to determine block device backing btrfs file system '%s': %m", p); |
4ad49000 | 802 | } |
8e274523 | 803 | |
b7cf4b4e BB |
804 | /* If this is a LUKS/DM device, recursively try to get the originating block device */ |
805 | while (block_get_originating(*ret, ret) > 0); | |
45c2e068 LP |
806 | |
807 | /* If this is a partition, try to get the originating block device */ | |
808 | (void) block_get_whole_disk(*ret, ret); | |
8e274523 | 809 | return 0; |
8e274523 LP |
810 | } |
811 | ||
66ebf6c0 TH |
812 | static bool cgroup_context_has_cpu_weight(CGroupContext *c) { |
813 | return c->cpu_weight != CGROUP_WEIGHT_INVALID || | |
814 | c->startup_cpu_weight != CGROUP_WEIGHT_INVALID; | |
815 | } | |
816 | ||
817 | static bool cgroup_context_has_cpu_shares(CGroupContext *c) { | |
818 | return c->cpu_shares != CGROUP_CPU_SHARES_INVALID || | |
819 | c->startup_cpu_shares != CGROUP_CPU_SHARES_INVALID; | |
820 | } | |
821 | ||
822 | static uint64_t cgroup_context_cpu_weight(CGroupContext *c, ManagerState state) { | |
823 | if (IN_SET(state, MANAGER_STARTING, MANAGER_INITIALIZING) && | |
824 | c->startup_cpu_weight != CGROUP_WEIGHT_INVALID) | |
825 | return c->startup_cpu_weight; | |
826 | else if (c->cpu_weight != CGROUP_WEIGHT_INVALID) | |
827 | return c->cpu_weight; | |
828 | else | |
829 | return CGROUP_WEIGHT_DEFAULT; | |
830 | } | |
831 | ||
832 | static uint64_t cgroup_context_cpu_shares(CGroupContext *c, ManagerState state) { | |
833 | if (IN_SET(state, MANAGER_STARTING, MANAGER_INITIALIZING) && | |
834 | c->startup_cpu_shares != CGROUP_CPU_SHARES_INVALID) | |
835 | return c->startup_cpu_shares; | |
836 | else if (c->cpu_shares != CGROUP_CPU_SHARES_INVALID) | |
837 | return c->cpu_shares; | |
838 | else | |
839 | return CGROUP_CPU_SHARES_DEFAULT; | |
840 | } | |
841 | ||
10f28641 FB |
842 | usec_t cgroup_cpu_adjust_period(usec_t period, usec_t quota, usec_t resolution, usec_t max_period) { |
843 | /* kernel uses a minimum resolution of 1ms, so both period and (quota * period) | |
844 | * need to be higher than that boundary. quota is specified in USecPerSec. | |
845 | * Additionally, period must be at most max_period. */ | |
846 | assert(quota > 0); | |
847 | ||
848 | return MIN(MAX3(period, resolution, resolution * USEC_PER_SEC / quota), max_period); | |
849 | } | |
850 | ||
851 | static usec_t cgroup_cpu_adjust_period_and_log(Unit *u, usec_t period, usec_t quota) { | |
852 | usec_t new_period; | |
853 | ||
854 | if (quota == USEC_INFINITY) | |
855 | /* Always use default period for infinity quota. */ | |
856 | return CGROUP_CPU_QUOTA_DEFAULT_PERIOD_USEC; | |
857 | ||
858 | if (period == USEC_INFINITY) | |
859 | /* Default period was requested. */ | |
860 | period = CGROUP_CPU_QUOTA_DEFAULT_PERIOD_USEC; | |
861 | ||
862 | /* Clamp to interval [1ms, 1s] */ | |
863 | new_period = cgroup_cpu_adjust_period(period, quota, USEC_PER_MSEC, USEC_PER_SEC); | |
864 | ||
865 | if (new_period != period) { | |
866 | char v[FORMAT_TIMESPAN_MAX]; | |
8ed6f81b | 867 | log_unit_full(u, u->warned_clamping_cpu_quota_period ? LOG_DEBUG : LOG_WARNING, |
10f28641 FB |
868 | "Clamping CPU interval for cpu.max: period is now %s", |
869 | format_timespan(v, sizeof(v), new_period, 1)); | |
527ede0c | 870 | u->warned_clamping_cpu_quota_period = true; |
10f28641 FB |
871 | } |
872 | ||
873 | return new_period; | |
874 | } | |
875 | ||
52fecf20 LP |
876 | static void cgroup_apply_unified_cpu_weight(Unit *u, uint64_t weight) { |
877 | char buf[DECIMAL_STR_MAX(uint64_t) + 2]; | |
66ebf6c0 TH |
878 | |
879 | xsprintf(buf, "%" PRIu64 "\n", weight); | |
293d32df | 880 | (void) set_attribute_and_warn(u, "cpu", "cpu.weight", buf); |
52fecf20 LP |
881 | } |
882 | ||
10f28641 | 883 | static void cgroup_apply_unified_cpu_quota(Unit *u, usec_t quota, usec_t period) { |
52fecf20 | 884 | char buf[(DECIMAL_STR_MAX(usec_t) + 1) * 2 + 1]; |
66ebf6c0 | 885 | |
10f28641 | 886 | period = cgroup_cpu_adjust_period_and_log(u, period, quota); |
66ebf6c0 TH |
887 | if (quota != USEC_INFINITY) |
888 | xsprintf(buf, USEC_FMT " " USEC_FMT "\n", | |
10f28641 | 889 | MAX(quota * period / USEC_PER_SEC, USEC_PER_MSEC), period); |
66ebf6c0 | 890 | else |
10f28641 | 891 | xsprintf(buf, "max " USEC_FMT "\n", period); |
293d32df | 892 | (void) set_attribute_and_warn(u, "cpu", "cpu.max", buf); |
66ebf6c0 TH |
893 | } |
894 | ||
52fecf20 LP |
895 | static void cgroup_apply_legacy_cpu_shares(Unit *u, uint64_t shares) { |
896 | char buf[DECIMAL_STR_MAX(uint64_t) + 2]; | |
66ebf6c0 TH |
897 | |
898 | xsprintf(buf, "%" PRIu64 "\n", shares); | |
293d32df | 899 | (void) set_attribute_and_warn(u, "cpu", "cpu.shares", buf); |
52fecf20 LP |
900 | } |
901 | ||
10f28641 | 902 | static void cgroup_apply_legacy_cpu_quota(Unit *u, usec_t quota, usec_t period) { |
52fecf20 | 903 | char buf[DECIMAL_STR_MAX(usec_t) + 2]; |
66ebf6c0 | 904 | |
10f28641 FB |
905 | period = cgroup_cpu_adjust_period_and_log(u, period, quota); |
906 | ||
907 | xsprintf(buf, USEC_FMT "\n", period); | |
293d32df | 908 | (void) set_attribute_and_warn(u, "cpu", "cpu.cfs_period_us", buf); |
66ebf6c0 TH |
909 | |
910 | if (quota != USEC_INFINITY) { | |
10f28641 | 911 | xsprintf(buf, USEC_FMT "\n", MAX(quota * period / USEC_PER_SEC, USEC_PER_MSEC)); |
293d32df | 912 | (void) set_attribute_and_warn(u, "cpu", "cpu.cfs_quota_us", buf); |
66ebf6c0 | 913 | } else |
589a5f7a | 914 | (void) set_attribute_and_warn(u, "cpu", "cpu.cfs_quota_us", "-1\n"); |
66ebf6c0 TH |
915 | } |
916 | ||
917 | static uint64_t cgroup_cpu_shares_to_weight(uint64_t shares) { | |
918 | return CLAMP(shares * CGROUP_WEIGHT_DEFAULT / CGROUP_CPU_SHARES_DEFAULT, | |
919 | CGROUP_WEIGHT_MIN, CGROUP_WEIGHT_MAX); | |
920 | } | |
921 | ||
922 | static uint64_t cgroup_cpu_weight_to_shares(uint64_t weight) { | |
923 | return CLAMP(weight * CGROUP_CPU_SHARES_DEFAULT / CGROUP_WEIGHT_DEFAULT, | |
924 | CGROUP_CPU_SHARES_MIN, CGROUP_CPU_SHARES_MAX); | |
925 | } | |
926 | ||
2cea199e | 927 | static void cgroup_apply_unified_cpuset(Unit *u, const CPUSet *cpus, const char *name) { |
047f5d63 PH |
928 | _cleanup_free_ char *buf = NULL; |
929 | ||
2cea199e | 930 | buf = cpu_set_to_range_string(cpus); |
c259ac9a LP |
931 | if (!buf) { |
932 | log_oom(); | |
933 | return; | |
934 | } | |
047f5d63 PH |
935 | |
936 | (void) set_attribute_and_warn(u, "cpuset", name, buf); | |
937 | } | |
938 | ||
508c45da | 939 | static bool cgroup_context_has_io_config(CGroupContext *c) { |
538b4852 TH |
940 | return c->io_accounting || |
941 | c->io_weight != CGROUP_WEIGHT_INVALID || | |
942 | c->startup_io_weight != CGROUP_WEIGHT_INVALID || | |
943 | c->io_device_weights || | |
6ae4283c | 944 | c->io_device_latencies || |
538b4852 TH |
945 | c->io_device_limits; |
946 | } | |
947 | ||
508c45da | 948 | static bool cgroup_context_has_blockio_config(CGroupContext *c) { |
538b4852 TH |
949 | return c->blockio_accounting || |
950 | c->blockio_weight != CGROUP_BLKIO_WEIGHT_INVALID || | |
951 | c->startup_blockio_weight != CGROUP_BLKIO_WEIGHT_INVALID || | |
952 | c->blockio_device_weights || | |
953 | c->blockio_device_bandwidths; | |
954 | } | |
955 | ||
508c45da | 956 | static uint64_t cgroup_context_io_weight(CGroupContext *c, ManagerState state) { |
64faf04c TH |
957 | if (IN_SET(state, MANAGER_STARTING, MANAGER_INITIALIZING) && |
958 | c->startup_io_weight != CGROUP_WEIGHT_INVALID) | |
959 | return c->startup_io_weight; | |
960 | else if (c->io_weight != CGROUP_WEIGHT_INVALID) | |
961 | return c->io_weight; | |
962 | else | |
963 | return CGROUP_WEIGHT_DEFAULT; | |
964 | } | |
965 | ||
508c45da | 966 | static uint64_t cgroup_context_blkio_weight(CGroupContext *c, ManagerState state) { |
64faf04c TH |
967 | if (IN_SET(state, MANAGER_STARTING, MANAGER_INITIALIZING) && |
968 | c->startup_blockio_weight != CGROUP_BLKIO_WEIGHT_INVALID) | |
969 | return c->startup_blockio_weight; | |
970 | else if (c->blockio_weight != CGROUP_BLKIO_WEIGHT_INVALID) | |
971 | return c->blockio_weight; | |
972 | else | |
973 | return CGROUP_BLKIO_WEIGHT_DEFAULT; | |
974 | } | |
975 | ||
508c45da | 976 | static uint64_t cgroup_weight_blkio_to_io(uint64_t blkio_weight) { |
538b4852 TH |
977 | return CLAMP(blkio_weight * CGROUP_WEIGHT_DEFAULT / CGROUP_BLKIO_WEIGHT_DEFAULT, |
978 | CGROUP_WEIGHT_MIN, CGROUP_WEIGHT_MAX); | |
979 | } | |
980 | ||
508c45da | 981 | static uint64_t cgroup_weight_io_to_blkio(uint64_t io_weight) { |
538b4852 TH |
982 | return CLAMP(io_weight * CGROUP_BLKIO_WEIGHT_DEFAULT / CGROUP_WEIGHT_DEFAULT, |
983 | CGROUP_BLKIO_WEIGHT_MIN, CGROUP_BLKIO_WEIGHT_MAX); | |
984 | } | |
985 | ||
f29ff115 | 986 | static void cgroup_apply_io_device_weight(Unit *u, const char *dev_path, uint64_t io_weight) { |
64faf04c TH |
987 | char buf[DECIMAL_STR_MAX(dev_t)*2+2+DECIMAL_STR_MAX(uint64_t)+1]; |
988 | dev_t dev; | |
989 | int r; | |
990 | ||
991 | r = lookup_block_device(dev_path, &dev); | |
992 | if (r < 0) | |
993 | return; | |
994 | ||
995 | xsprintf(buf, "%u:%u %" PRIu64 "\n", major(dev), minor(dev), io_weight); | |
293d32df | 996 | (void) set_attribute_and_warn(u, "io", "io.weight", buf); |
64faf04c TH |
997 | } |
998 | ||
f29ff115 | 999 | static void cgroup_apply_blkio_device_weight(Unit *u, const char *dev_path, uint64_t blkio_weight) { |
64faf04c TH |
1000 | char buf[DECIMAL_STR_MAX(dev_t)*2+2+DECIMAL_STR_MAX(uint64_t)+1]; |
1001 | dev_t dev; | |
1002 | int r; | |
1003 | ||
1004 | r = lookup_block_device(dev_path, &dev); | |
1005 | if (r < 0) | |
1006 | return; | |
1007 | ||
1008 | xsprintf(buf, "%u:%u %" PRIu64 "\n", major(dev), minor(dev), blkio_weight); | |
293d32df | 1009 | (void) set_attribute_and_warn(u, "blkio", "blkio.weight_device", buf); |
64faf04c TH |
1010 | } |
1011 | ||
6ae4283c TH |
1012 | static void cgroup_apply_io_device_latency(Unit *u, const char *dev_path, usec_t target) { |
1013 | char buf[DECIMAL_STR_MAX(dev_t)*2+2+7+DECIMAL_STR_MAX(uint64_t)+1]; | |
1014 | dev_t dev; | |
1015 | int r; | |
1016 | ||
1017 | r = lookup_block_device(dev_path, &dev); | |
1018 | if (r < 0) | |
1019 | return; | |
1020 | ||
1021 | if (target != USEC_INFINITY) | |
1022 | xsprintf(buf, "%u:%u target=%" PRIu64 "\n", major(dev), minor(dev), target); | |
1023 | else | |
1024 | xsprintf(buf, "%u:%u target=max\n", major(dev), minor(dev)); | |
1025 | ||
293d32df | 1026 | (void) set_attribute_and_warn(u, "io", "io.latency", buf); |
6ae4283c TH |
1027 | } |
1028 | ||
17ae2780 | 1029 | static void cgroup_apply_io_device_limit(Unit *u, const char *dev_path, uint64_t *limits) { |
4c1f9343 ZJS |
1030 | char limit_bufs[_CGROUP_IO_LIMIT_TYPE_MAX][DECIMAL_STR_MAX(uint64_t)], |
1031 | buf[DECIMAL_STR_MAX(dev_t)*2+2+(6+DECIMAL_STR_MAX(uint64_t)+1)*4]; | |
64faf04c | 1032 | dev_t dev; |
64faf04c | 1033 | |
4c1f9343 | 1034 | if (lookup_block_device(dev_path, &dev) < 0) |
17ae2780 | 1035 | return; |
64faf04c | 1036 | |
4c1f9343 | 1037 | for (CGroupIOLimitType type = 0; type < _CGROUP_IO_LIMIT_TYPE_MAX; type++) |
17ae2780 | 1038 | if (limits[type] != cgroup_io_limit_defaults[type]) |
64faf04c | 1039 | xsprintf(limit_bufs[type], "%" PRIu64, limits[type]); |
17ae2780 | 1040 | else |
64faf04c | 1041 | xsprintf(limit_bufs[type], "%s", limits[type] == CGROUP_LIMIT_MAX ? "max" : "0"); |
64faf04c TH |
1042 | |
1043 | xsprintf(buf, "%u:%u rbps=%s wbps=%s riops=%s wiops=%s\n", major(dev), minor(dev), | |
1044 | limit_bufs[CGROUP_IO_RBPS_MAX], limit_bufs[CGROUP_IO_WBPS_MAX], | |
1045 | limit_bufs[CGROUP_IO_RIOPS_MAX], limit_bufs[CGROUP_IO_WIOPS_MAX]); | |
293d32df | 1046 | (void) set_attribute_and_warn(u, "io", "io.max", buf); |
64faf04c TH |
1047 | } |
1048 | ||
17ae2780 | 1049 | static void cgroup_apply_blkio_device_limit(Unit *u, const char *dev_path, uint64_t rbps, uint64_t wbps) { |
64faf04c TH |
1050 | char buf[DECIMAL_STR_MAX(dev_t)*2+2+DECIMAL_STR_MAX(uint64_t)+1]; |
1051 | dev_t dev; | |
64faf04c | 1052 | |
4c1f9343 | 1053 | if (lookup_block_device(dev_path, &dev) < 0) |
17ae2780 | 1054 | return; |
64faf04c | 1055 | |
64faf04c | 1056 | sprintf(buf, "%u:%u %" PRIu64 "\n", major(dev), minor(dev), rbps); |
293d32df | 1057 | (void) set_attribute_and_warn(u, "blkio", "blkio.throttle.read_bps_device", buf); |
64faf04c | 1058 | |
64faf04c | 1059 | sprintf(buf, "%u:%u %" PRIu64 "\n", major(dev), minor(dev), wbps); |
293d32df | 1060 | (void) set_attribute_and_warn(u, "blkio", "blkio.throttle.write_bps_device", buf); |
64faf04c TH |
1061 | } |
1062 | ||
c52db42b CD |
1063 | static bool unit_has_unified_memory_config(Unit *u) { |
1064 | CGroupContext *c; | |
1065 | ||
1066 | assert(u); | |
1067 | ||
806a9362 | 1068 | assert_se(c = unit_get_cgroup_context(u)); |
c52db42b | 1069 | |
7c9d2b79 | 1070 | return unit_get_ancestor_memory_min(u) > 0 || unit_get_ancestor_memory_low(u) > 0 || |
c52db42b CD |
1071 | c->memory_high != CGROUP_LIMIT_MAX || c->memory_max != CGROUP_LIMIT_MAX || |
1072 | c->memory_swap_max != CGROUP_LIMIT_MAX; | |
da4d897e TH |
1073 | } |
1074 | ||
f29ff115 | 1075 | static void cgroup_apply_unified_memory_limit(Unit *u, const char *file, uint64_t v) { |
589a5f7a | 1076 | char buf[DECIMAL_STR_MAX(uint64_t) + 1] = "max\n"; |
da4d897e TH |
1077 | |
1078 | if (v != CGROUP_LIMIT_MAX) | |
1079 | xsprintf(buf, "%" PRIu64 "\n", v); | |
1080 | ||
293d32df | 1081 | (void) set_attribute_and_warn(u, "memory", file, buf); |
da4d897e TH |
1082 | } |
1083 | ||
0f2d84d2 | 1084 | static void cgroup_apply_firewall(Unit *u) { |
0f2d84d2 LP |
1085 | assert(u); |
1086 | ||
acf7f253 | 1087 | /* Best-effort: let's apply IP firewalling and/or accounting if that's enabled */ |
906c06f6 | 1088 | |
acf7f253 | 1089 | if (bpf_firewall_compile(u) < 0) |
906c06f6 DM |
1090 | return; |
1091 | ||
fab34748 | 1092 | (void) bpf_firewall_load_custom(u); |
906c06f6 | 1093 | (void) bpf_firewall_install(u); |
906c06f6 DM |
1094 | } |
1095 | ||
a8e5eb17 JK |
1096 | static void cgroup_apply_socket_bind(Unit *u) { |
1097 | assert(u); | |
1098 | ||
1099 | (void) socket_bind_install(u); | |
1100 | } | |
1101 | ||
8b139557 ZJS |
1102 | static int cgroup_apply_devices(Unit *u) { |
1103 | _cleanup_(bpf_program_unrefp) BPFProgram *prog = NULL; | |
1104 | const char *path; | |
1105 | CGroupContext *c; | |
1106 | CGroupDeviceAllow *a; | |
45669ae2 | 1107 | CGroupDevicePolicy policy; |
8b139557 ZJS |
1108 | int r; |
1109 | ||
1110 | assert_se(c = unit_get_cgroup_context(u)); | |
1111 | assert_se(path = u->cgroup_path); | |
1112 | ||
45669ae2 ZJS |
1113 | policy = c->device_policy; |
1114 | ||
8b139557 | 1115 | if (cg_all_unified() > 0) { |
45669ae2 | 1116 | r = bpf_devices_cgroup_init(&prog, policy, c->device_allow); |
8b139557 ZJS |
1117 | if (r < 0) |
1118 | return log_unit_warning_errno(u, r, "Failed to initialize device control bpf program: %m"); | |
1119 | ||
1120 | } else { | |
1121 | /* Changing the devices list of a populated cgroup might result in EINVAL, hence ignore | |
1122 | * EINVAL here. */ | |
1123 | ||
45669ae2 | 1124 | if (c->device_allow || policy != CGROUP_DEVICE_POLICY_AUTO) |
8b139557 ZJS |
1125 | r = cg_set_attribute("devices", path, "devices.deny", "a"); |
1126 | else | |
1127 | r = cg_set_attribute("devices", path, "devices.allow", "a"); | |
1128 | if (r < 0) | |
8ed6f81b YW |
1129 | log_unit_full_errno(u, IN_SET(r, -ENOENT, -EROFS, -EINVAL, -EACCES, -EPERM) ? LOG_DEBUG : LOG_WARNING, r, |
1130 | "Failed to reset devices.allow/devices.deny: %m"); | |
8b139557 ZJS |
1131 | } |
1132 | ||
6b000af4 | 1133 | bool allow_list_static = policy == CGROUP_DEVICE_POLICY_CLOSED || |
45669ae2 | 1134 | (policy == CGROUP_DEVICE_POLICY_AUTO && c->device_allow); |
6b000af4 LP |
1135 | if (allow_list_static) |
1136 | (void) bpf_devices_allow_list_static(prog, path); | |
8b139557 | 1137 | |
6b000af4 | 1138 | bool any = allow_list_static; |
8b139557 ZJS |
1139 | LIST_FOREACH(device_allow, a, c->device_allow) { |
1140 | char acc[4], *val; | |
1141 | unsigned k = 0; | |
1142 | ||
1143 | if (a->r) | |
1144 | acc[k++] = 'r'; | |
1145 | if (a->w) | |
1146 | acc[k++] = 'w'; | |
1147 | if (a->m) | |
1148 | acc[k++] = 'm'; | |
8b139557 ZJS |
1149 | if (k == 0) |
1150 | continue; | |
8b139557 ZJS |
1151 | acc[k++] = 0; |
1152 | ||
1153 | if (path_startswith(a->path, "/dev/")) | |
6b000af4 | 1154 | r = bpf_devices_allow_list_device(prog, path, a->path, acc); |
8b139557 | 1155 | else if ((val = startswith(a->path, "block-"))) |
6b000af4 | 1156 | r = bpf_devices_allow_list_major(prog, path, val, 'b', acc); |
8b139557 | 1157 | else if ((val = startswith(a->path, "char-"))) |
6b000af4 | 1158 | r = bpf_devices_allow_list_major(prog, path, val, 'c', acc); |
45669ae2 | 1159 | else { |
8b139557 | 1160 | log_unit_debug(u, "Ignoring device '%s' while writing cgroup attribute.", a->path); |
45669ae2 ZJS |
1161 | continue; |
1162 | } | |
1163 | ||
1164 | if (r >= 0) | |
1165 | any = true; | |
1166 | } | |
1167 | ||
1168 | if (prog && !any) { | |
1169 | log_unit_warning_errno(u, SYNTHETIC_ERRNO(ENODEV), "No devices matched by device filter."); | |
1170 | ||
1171 | /* The kernel verifier would reject a program we would build with the normal intro and outro | |
6b000af4 | 1172 | but no allow-listing rules (outro would contain an unreachable instruction for successful |
45669ae2 ZJS |
1173 | return). */ |
1174 | policy = CGROUP_DEVICE_POLICY_STRICT; | |
8b139557 ZJS |
1175 | } |
1176 | ||
45669ae2 | 1177 | r = bpf_devices_apply_policy(prog, policy, any, path, &u->bpf_device_control_installed); |
8b139557 ZJS |
1178 | if (r < 0) { |
1179 | static bool warned = false; | |
1180 | ||
1181 | log_full_errno(warned ? LOG_DEBUG : LOG_WARNING, r, | |
1182 | "Unit %s configures device ACL, but the local system doesn't seem to support the BPF-based device controller.\n" | |
1183 | "Proceeding WITHOUT applying ACL (all devices will be accessible)!\n" | |
1184 | "(This warning is only shown for the first loaded unit using device ACL.)", u->id); | |
1185 | ||
1186 | warned = true; | |
1187 | } | |
1188 | return r; | |
1189 | } | |
1190 | ||
29eb0eef ZJS |
1191 | static void set_io_weight(Unit *u, const char *controller, uint64_t weight) { |
1192 | char buf[8+DECIMAL_STR_MAX(uint64_t)+1]; | |
1193 | const char *p; | |
1194 | ||
1195 | p = strjoina(controller, ".weight"); | |
1196 | xsprintf(buf, "default %" PRIu64 "\n", weight); | |
1197 | (void) set_attribute_and_warn(u, controller, p, buf); | |
1198 | ||
1199 | /* FIXME: drop this when distro kernels properly support BFQ through "io.weight" | |
1200 | * See also: https://github.com/systemd/systemd/pull/13335 and | |
1201 | * https://github.com/torvalds/linux/commit/65752aef0a407e1ef17ec78a7fc31ba4e0b360f9. | |
1202 | * The range is 1..1000 apparently. */ | |
1203 | p = strjoina(controller, ".bfq.weight"); | |
1204 | xsprintf(buf, "%" PRIu64 "\n", (weight + 9) / 10); | |
1205 | (void) set_attribute_and_warn(u, controller, p, buf); | |
1206 | } | |
1207 | ||
506ea51b JK |
1208 | static void cgroup_apply_bpf_foreign_program(Unit *u) { |
1209 | assert(u); | |
1210 | ||
1211 | (void) bpf_foreign_install(u); | |
1212 | } | |
1213 | ||
906c06f6 DM |
1214 | static void cgroup_context_apply( |
1215 | Unit *u, | |
1216 | CGroupMask apply_mask, | |
906c06f6 DM |
1217 | ManagerState state) { |
1218 | ||
f29ff115 TH |
1219 | const char *path; |
1220 | CGroupContext *c; | |
52fecf20 | 1221 | bool is_host_root, is_local_root; |
4ad49000 LP |
1222 | int r; |
1223 | ||
f29ff115 TH |
1224 | assert(u); |
1225 | ||
906c06f6 | 1226 | /* Nothing to do? Exit early! */ |
17f14955 | 1227 | if (apply_mask == 0) |
4ad49000 | 1228 | return; |
8e274523 | 1229 | |
52fecf20 LP |
1230 | /* Some cgroup attributes are not supported on the host root cgroup, hence silently ignore them here. And other |
1231 | * attributes should only be managed for cgroups further down the tree. */ | |
1232 | is_local_root = unit_has_name(u, SPECIAL_ROOT_SLICE); | |
1233 | is_host_root = unit_has_host_root_cgroup(u); | |
f3725e64 LP |
1234 | |
1235 | assert_se(c = unit_get_cgroup_context(u)); | |
1236 | assert_se(path = u->cgroup_path); | |
1237 | ||
52fecf20 | 1238 | if (is_local_root) /* Make sure we don't try to display messages with an empty path. */ |
6da13913 | 1239 | path = "/"; |
01efdf13 | 1240 | |
be2c0327 LP |
1241 | /* We generally ignore errors caused by read-only mounted cgroup trees (assuming we are running in a container |
1242 | * then), and missing cgroups, i.e. EROFS and ENOENT. */ | |
714e2e1d | 1243 | |
be2c0327 LP |
1244 | /* In fully unified mode these attributes don't exist on the host cgroup root. On legacy the weights exist, but |
1245 | * setting the weight makes very little sense on the host root cgroup, as there are no other cgroups at this | |
1246 | * level. The quota exists there too, but any attempt to write to it is refused with EINVAL. Inside of | |
4e1dfa45 | 1247 | * containers we want to leave control of these to the container manager (and if cgroup v2 delegation is used |
be2c0327 LP |
1248 | * we couldn't even write to them if we wanted to). */ |
1249 | if ((apply_mask & CGROUP_MASK_CPU) && !is_local_root) { | |
8e274523 | 1250 | |
b4cccbc1 | 1251 | if (cg_all_unified() > 0) { |
be2c0327 | 1252 | uint64_t weight; |
b2f8b02e | 1253 | |
be2c0327 LP |
1254 | if (cgroup_context_has_cpu_weight(c)) |
1255 | weight = cgroup_context_cpu_weight(c, state); | |
1256 | else if (cgroup_context_has_cpu_shares(c)) { | |
1257 | uint64_t shares; | |
66ebf6c0 | 1258 | |
be2c0327 LP |
1259 | shares = cgroup_context_cpu_shares(c, state); |
1260 | weight = cgroup_cpu_shares_to_weight(shares); | |
66ebf6c0 | 1261 | |
be2c0327 LP |
1262 | log_cgroup_compat(u, "Applying [Startup]CPUShares=%" PRIu64 " as [Startup]CPUWeight=%" PRIu64 " on %s", |
1263 | shares, weight, path); | |
1264 | } else | |
1265 | weight = CGROUP_WEIGHT_DEFAULT; | |
66ebf6c0 | 1266 | |
be2c0327 | 1267 | cgroup_apply_unified_cpu_weight(u, weight); |
10f28641 | 1268 | cgroup_apply_unified_cpu_quota(u, c->cpu_quota_per_sec_usec, c->cpu_quota_period_usec); |
66ebf6c0 | 1269 | |
52fecf20 | 1270 | } else { |
be2c0327 | 1271 | uint64_t shares; |
52fecf20 | 1272 | |
be2c0327 LP |
1273 | if (cgroup_context_has_cpu_weight(c)) { |
1274 | uint64_t weight; | |
52fecf20 | 1275 | |
be2c0327 LP |
1276 | weight = cgroup_context_cpu_weight(c, state); |
1277 | shares = cgroup_cpu_weight_to_shares(weight); | |
52fecf20 | 1278 | |
be2c0327 LP |
1279 | log_cgroup_compat(u, "Applying [Startup]CPUWeight=%" PRIu64 " as [Startup]CPUShares=%" PRIu64 " on %s", |
1280 | weight, shares, path); | |
1281 | } else if (cgroup_context_has_cpu_shares(c)) | |
1282 | shares = cgroup_context_cpu_shares(c, state); | |
1283 | else | |
1284 | shares = CGROUP_CPU_SHARES_DEFAULT; | |
66ebf6c0 | 1285 | |
be2c0327 | 1286 | cgroup_apply_legacy_cpu_shares(u, shares); |
10f28641 | 1287 | cgroup_apply_legacy_cpu_quota(u, c->cpu_quota_per_sec_usec, c->cpu_quota_period_usec); |
66ebf6c0 | 1288 | } |
4ad49000 LP |
1289 | } |
1290 | ||
047f5d63 | 1291 | if ((apply_mask & CGROUP_MASK_CPUSET) && !is_local_root) { |
2cea199e ZJS |
1292 | cgroup_apply_unified_cpuset(u, &c->cpuset_cpus, "cpuset.cpus"); |
1293 | cgroup_apply_unified_cpuset(u, &c->cpuset_mems, "cpuset.mems"); | |
047f5d63 PH |
1294 | } |
1295 | ||
4e1dfa45 | 1296 | /* The 'io' controller attributes are not exported on the host's root cgroup (being a pure cgroup v2 |
52fecf20 LP |
1297 | * controller), and in case of containers we want to leave control of these attributes to the container manager |
1298 | * (and we couldn't access that stuff anyway, even if we tried if proper delegation is used). */ | |
1299 | if ((apply_mask & CGROUP_MASK_IO) && !is_local_root) { | |
52fecf20 LP |
1300 | bool has_io, has_blockio; |
1301 | uint64_t weight; | |
13c31542 | 1302 | |
52fecf20 LP |
1303 | has_io = cgroup_context_has_io_config(c); |
1304 | has_blockio = cgroup_context_has_blockio_config(c); | |
13c31542 | 1305 | |
52fecf20 LP |
1306 | if (has_io) |
1307 | weight = cgroup_context_io_weight(c, state); | |
1308 | else if (has_blockio) { | |
1309 | uint64_t blkio_weight; | |
128fadc9 | 1310 | |
52fecf20 LP |
1311 | blkio_weight = cgroup_context_blkio_weight(c, state); |
1312 | weight = cgroup_weight_blkio_to_io(blkio_weight); | |
128fadc9 | 1313 | |
67e2ea15 | 1314 | log_cgroup_compat(u, "Applying [Startup]BlockIOWeight=%" PRIu64 " as [Startup]IOWeight=%" PRIu64, |
52fecf20 LP |
1315 | blkio_weight, weight); |
1316 | } else | |
1317 | weight = CGROUP_WEIGHT_DEFAULT; | |
13c31542 | 1318 | |
29eb0eef | 1319 | set_io_weight(u, "io", weight); |
2dbc45ae | 1320 | |
52fecf20 LP |
1321 | if (has_io) { |
1322 | CGroupIODeviceLatency *latency; | |
1323 | CGroupIODeviceLimit *limit; | |
1324 | CGroupIODeviceWeight *w; | |
128fadc9 | 1325 | |
52fecf20 LP |
1326 | LIST_FOREACH(device_weights, w, c->io_device_weights) |
1327 | cgroup_apply_io_device_weight(u, w->path, w->weight); | |
128fadc9 | 1328 | |
52fecf20 LP |
1329 | LIST_FOREACH(device_limits, limit, c->io_device_limits) |
1330 | cgroup_apply_io_device_limit(u, limit->path, limit->limits); | |
6ae4283c | 1331 | |
52fecf20 LP |
1332 | LIST_FOREACH(device_latencies, latency, c->io_device_latencies) |
1333 | cgroup_apply_io_device_latency(u, latency->path, latency->target_usec); | |
6ae4283c | 1334 | |
52fecf20 LP |
1335 | } else if (has_blockio) { |
1336 | CGroupBlockIODeviceWeight *w; | |
1337 | CGroupBlockIODeviceBandwidth *b; | |
13c31542 | 1338 | |
52fecf20 LP |
1339 | LIST_FOREACH(device_weights, w, c->blockio_device_weights) { |
1340 | weight = cgroup_weight_blkio_to_io(w->weight); | |
17ae2780 | 1341 | |
67e2ea15 | 1342 | log_cgroup_compat(u, "Applying BlockIODeviceWeight=%" PRIu64 " as IODeviceWeight=%" PRIu64 " for %s", |
52fecf20 | 1343 | w->weight, weight, w->path); |
538b4852 | 1344 | |
52fecf20 LP |
1345 | cgroup_apply_io_device_weight(u, w->path, weight); |
1346 | } | |
538b4852 | 1347 | |
17ae2780 | 1348 | LIST_FOREACH(device_bandwidths, b, c->blockio_device_bandwidths) { |
538b4852 TH |
1349 | uint64_t limits[_CGROUP_IO_LIMIT_TYPE_MAX]; |
1350 | CGroupIOLimitType type; | |
1351 | ||
1352 | for (type = 0; type < _CGROUP_IO_LIMIT_TYPE_MAX; type++) | |
1353 | limits[type] = cgroup_io_limit_defaults[type]; | |
1354 | ||
1355 | limits[CGROUP_IO_RBPS_MAX] = b->rbps; | |
1356 | limits[CGROUP_IO_WBPS_MAX] = b->wbps; | |
1357 | ||
67e2ea15 | 1358 | log_cgroup_compat(u, "Applying BlockIO{Read|Write}Bandwidth=%" PRIu64 " %" PRIu64 " as IO{Read|Write}BandwidthMax= for %s", |
128fadc9 TH |
1359 | b->rbps, b->wbps, b->path); |
1360 | ||
17ae2780 | 1361 | cgroup_apply_io_device_limit(u, b->path, limits); |
538b4852 | 1362 | } |
13c31542 TH |
1363 | } |
1364 | } | |
1365 | ||
906c06f6 | 1366 | if (apply_mask & CGROUP_MASK_BLKIO) { |
52fecf20 | 1367 | bool has_io, has_blockio; |
4ad49000 | 1368 | |
52fecf20 LP |
1369 | has_io = cgroup_context_has_io_config(c); |
1370 | has_blockio = cgroup_context_has_blockio_config(c); | |
1371 | ||
1372 | /* Applying a 'weight' never makes sense for the host root cgroup, and for containers this should be | |
1373 | * left to our container manager, too. */ | |
1374 | if (!is_local_root) { | |
64faf04c | 1375 | uint64_t weight; |
64faf04c | 1376 | |
7d862ab8 | 1377 | if (has_io) { |
52fecf20 | 1378 | uint64_t io_weight; |
128fadc9 | 1379 | |
52fecf20 | 1380 | io_weight = cgroup_context_io_weight(c, state); |
538b4852 | 1381 | weight = cgroup_weight_io_to_blkio(cgroup_context_io_weight(c, state)); |
128fadc9 | 1382 | |
67e2ea15 | 1383 | log_cgroup_compat(u, "Applying [Startup]IOWeight=%" PRIu64 " as [Startup]BlockIOWeight=%" PRIu64, |
128fadc9 | 1384 | io_weight, weight); |
7d862ab8 TH |
1385 | } else if (has_blockio) |
1386 | weight = cgroup_context_blkio_weight(c, state); | |
1387 | else | |
538b4852 | 1388 | weight = CGROUP_BLKIO_WEIGHT_DEFAULT; |
64faf04c | 1389 | |
29eb0eef | 1390 | set_io_weight(u, "blkio", weight); |
35e7a62c | 1391 | |
7d862ab8 | 1392 | if (has_io) { |
538b4852 TH |
1393 | CGroupIODeviceWeight *w; |
1394 | ||
128fadc9 TH |
1395 | LIST_FOREACH(device_weights, w, c->io_device_weights) { |
1396 | weight = cgroup_weight_io_to_blkio(w->weight); | |
1397 | ||
67e2ea15 | 1398 | log_cgroup_compat(u, "Applying IODeviceWeight=%" PRIu64 " as BlockIODeviceWeight=%" PRIu64 " for %s", |
128fadc9 TH |
1399 | w->weight, weight, w->path); |
1400 | ||
1401 | cgroup_apply_blkio_device_weight(u, w->path, weight); | |
1402 | } | |
7d862ab8 TH |
1403 | } else if (has_blockio) { |
1404 | CGroupBlockIODeviceWeight *w; | |
1405 | ||
7d862ab8 TH |
1406 | LIST_FOREACH(device_weights, w, c->blockio_device_weights) |
1407 | cgroup_apply_blkio_device_weight(u, w->path, w->weight); | |
538b4852 | 1408 | } |
4ad49000 LP |
1409 | } |
1410 | ||
5238e957 | 1411 | /* The bandwidth limits are something that make sense to be applied to the host's root but not container |
52fecf20 LP |
1412 | * roots, as there we want the container manager to handle it */ |
1413 | if (is_host_root || !is_local_root) { | |
1414 | if (has_io) { | |
1415 | CGroupIODeviceLimit *l; | |
538b4852 | 1416 | |
52fecf20 | 1417 | LIST_FOREACH(device_limits, l, c->io_device_limits) { |
67e2ea15 | 1418 | log_cgroup_compat(u, "Applying IO{Read|Write}Bandwidth=%" PRIu64 " %" PRIu64 " as BlockIO{Read|Write}BandwidthMax= for %s", |
52fecf20 | 1419 | l->limits[CGROUP_IO_RBPS_MAX], l->limits[CGROUP_IO_WBPS_MAX], l->path); |
128fadc9 | 1420 | |
52fecf20 LP |
1421 | cgroup_apply_blkio_device_limit(u, l->path, l->limits[CGROUP_IO_RBPS_MAX], l->limits[CGROUP_IO_WBPS_MAX]); |
1422 | } | |
1423 | } else if (has_blockio) { | |
1424 | CGroupBlockIODeviceBandwidth *b; | |
7d862ab8 | 1425 | |
52fecf20 LP |
1426 | LIST_FOREACH(device_bandwidths, b, c->blockio_device_bandwidths) |
1427 | cgroup_apply_blkio_device_limit(u, b->path, b->rbps, b->wbps); | |
1428 | } | |
d686d8a9 | 1429 | } |
8e274523 LP |
1430 | } |
1431 | ||
be2c0327 LP |
1432 | /* In unified mode 'memory' attributes do not exist on the root cgroup. In legacy mode 'memory.limit_in_bytes' |
1433 | * exists on the root cgroup, but any writes to it are refused with EINVAL. And if we run in a container we | |
4e1dfa45 | 1434 | * want to leave control to the container manager (and if proper cgroup v2 delegation is used we couldn't even |
be2c0327 LP |
1435 | * write to this if we wanted to.) */ |
1436 | if ((apply_mask & CGROUP_MASK_MEMORY) && !is_local_root) { | |
efdb0237 | 1437 | |
52fecf20 | 1438 | if (cg_all_unified() > 0) { |
be2c0327 LP |
1439 | uint64_t max, swap_max = CGROUP_LIMIT_MAX; |
1440 | ||
c52db42b | 1441 | if (unit_has_unified_memory_config(u)) { |
be2c0327 LP |
1442 | max = c->memory_max; |
1443 | swap_max = c->memory_swap_max; | |
1444 | } else { | |
1445 | max = c->memory_limit; | |
efdb0237 | 1446 | |
be2c0327 LP |
1447 | if (max != CGROUP_LIMIT_MAX) |
1448 | log_cgroup_compat(u, "Applying MemoryLimit=%" PRIu64 " as MemoryMax=", max); | |
128fadc9 | 1449 | } |
da4d897e | 1450 | |
64fe532e | 1451 | cgroup_apply_unified_memory_limit(u, "memory.min", unit_get_ancestor_memory_min(u)); |
c52db42b | 1452 | cgroup_apply_unified_memory_limit(u, "memory.low", unit_get_ancestor_memory_low(u)); |
be2c0327 LP |
1453 | cgroup_apply_unified_memory_limit(u, "memory.high", c->memory_high); |
1454 | cgroup_apply_unified_memory_limit(u, "memory.max", max); | |
1455 | cgroup_apply_unified_memory_limit(u, "memory.swap.max", swap_max); | |
128fadc9 | 1456 | |
afcfaa69 LP |
1457 | (void) set_attribute_and_warn(u, "memory", "memory.oom.group", one_zero(c->memory_oom_group)); |
1458 | ||
be2c0327 LP |
1459 | } else { |
1460 | char buf[DECIMAL_STR_MAX(uint64_t) + 1]; | |
1461 | uint64_t val; | |
52fecf20 | 1462 | |
c52db42b | 1463 | if (unit_has_unified_memory_config(u)) { |
be2c0327 LP |
1464 | val = c->memory_max; |
1465 | log_cgroup_compat(u, "Applying MemoryMax=%" PRIi64 " as MemoryLimit=", val); | |
1466 | } else | |
1467 | val = c->memory_limit; | |
78a4ee59 | 1468 | |
be2c0327 LP |
1469 | if (val == CGROUP_LIMIT_MAX) |
1470 | strncpy(buf, "-1\n", sizeof(buf)); | |
1471 | else | |
1472 | xsprintf(buf, "%" PRIu64 "\n", val); | |
1473 | ||
1474 | (void) set_attribute_and_warn(u, "memory", "memory.limit_in_bytes", buf); | |
da4d897e | 1475 | } |
4ad49000 | 1476 | } |
8e274523 | 1477 | |
4e1dfa45 | 1478 | /* On cgroup v2 we can apply BPF everywhere. On cgroup v1 we apply it everywhere except for the root of |
52fecf20 LP |
1479 | * containers, where we leave this to the manager */ |
1480 | if ((apply_mask & (CGROUP_MASK_DEVICES | CGROUP_MASK_BPF_DEVICES)) && | |
8b139557 ZJS |
1481 | (is_host_root || cg_all_unified() > 0 || !is_local_root)) |
1482 | (void) cgroup_apply_devices(u); | |
03a7b521 | 1483 | |
00b5974f LP |
1484 | if (apply_mask & CGROUP_MASK_PIDS) { |
1485 | ||
52fecf20 | 1486 | if (is_host_root) { |
00b5974f LP |
1487 | /* So, the "pids" controller does not expose anything on the root cgroup, in order not to |
1488 | * replicate knobs exposed elsewhere needlessly. We abstract this away here however, and when | |
1489 | * the knobs of the root cgroup are modified propagate this to the relevant sysctls. There's a | |
1490 | * non-obvious asymmetry however: unlike the cgroup properties we don't really want to take | |
1491 | * exclusive ownership of the sysctls, but we still want to honour things if the user sets | |
1492 | * limits. Hence we employ sort of a one-way strategy: when the user sets a bounded limit | |
1493 | * through us it counts. When the user afterwards unsets it again (i.e. sets it to unbounded) | |
1494 | * it also counts. But if the user never set a limit through us (i.e. we are the default of | |
1495 | * "unbounded") we leave things unmodified. For this we manage a global boolean that we turn on | |
1496 | * the first time we set a limit. Note that this boolean is flushed out on manager reload, | |
5238e957 | 1497 | * which is desirable so that there's an official way to release control of the sysctl from |
00b5974f LP |
1498 | * systemd: set the limit to unbounded and reload. */ |
1499 | ||
3a0f06c4 | 1500 | if (tasks_max_isset(&c->tasks_max)) { |
00b5974f | 1501 | u->manager->sysctl_pid_max_changed = true; |
3a0f06c4 | 1502 | r = procfs_tasks_set_limit(tasks_max_resolve(&c->tasks_max)); |
00b5974f LP |
1503 | } else if (u->manager->sysctl_pid_max_changed) |
1504 | r = procfs_tasks_set_limit(TASKS_MAX); | |
1505 | else | |
1506 | r = 0; | |
00b5974f | 1507 | if (r < 0) |
8ed6f81b YW |
1508 | log_unit_full_errno(u, LOG_LEVEL_CGROUP_WRITE(r), r, |
1509 | "Failed to write to tasks limit sysctls: %m"); | |
52fecf20 | 1510 | } |
03a7b521 | 1511 | |
52fecf20 LP |
1512 | /* The attribute itself is not available on the host root cgroup, and in the container case we want to |
1513 | * leave it for the container manager. */ | |
1514 | if (!is_local_root) { | |
3a0f06c4 ZJS |
1515 | if (tasks_max_isset(&c->tasks_max)) { |
1516 | char buf[DECIMAL_STR_MAX(uint64_t) + 1]; | |
03a7b521 | 1517 | |
3a0f06c4 | 1518 | xsprintf(buf, "%" PRIu64 "\n", tasks_max_resolve(&c->tasks_max)); |
293d32df | 1519 | (void) set_attribute_and_warn(u, "pids", "pids.max", buf); |
00b5974f | 1520 | } else |
589a5f7a | 1521 | (void) set_attribute_and_warn(u, "pids", "pids.max", "max\n"); |
00b5974f | 1522 | } |
03a7b521 | 1523 | } |
906c06f6 | 1524 | |
17f14955 | 1525 | if (apply_mask & CGROUP_MASK_BPF_FIREWALL) |
0f2d84d2 | 1526 | cgroup_apply_firewall(u); |
506ea51b JK |
1527 | |
1528 | if (apply_mask & CGROUP_MASK_BPF_FOREIGN) | |
1529 | cgroup_apply_bpf_foreign_program(u); | |
a8e5eb17 JK |
1530 | |
1531 | if (apply_mask & CGROUP_MASK_BPF_SOCKET_BIND) | |
1532 | cgroup_apply_socket_bind(u); | |
fb385181 LP |
1533 | } |
1534 | ||
16492445 LP |
1535 | static bool unit_get_needs_bpf_firewall(Unit *u) { |
1536 | CGroupContext *c; | |
1537 | Unit *p; | |
1538 | assert(u); | |
1539 | ||
1540 | c = unit_get_cgroup_context(u); | |
1541 | if (!c) | |
1542 | return false; | |
1543 | ||
1544 | if (c->ip_accounting || | |
1545 | c->ip_address_allow || | |
fab34748 KL |
1546 | c->ip_address_deny || |
1547 | c->ip_filters_ingress || | |
1548 | c->ip_filters_egress) | |
16492445 LP |
1549 | return true; |
1550 | ||
1551 | /* If any parent slice has an IP access list defined, it applies too */ | |
1552 | for (p = UNIT_DEREF(u->slice); p; p = UNIT_DEREF(p->slice)) { | |
1553 | c = unit_get_cgroup_context(p); | |
1554 | if (!c) | |
1555 | return false; | |
1556 | ||
1557 | if (c->ip_address_allow || | |
1558 | c->ip_address_deny) | |
1559 | return true; | |
1560 | } | |
1561 | ||
1562 | return false; | |
1563 | } | |
1564 | ||
506ea51b JK |
1565 | static bool unit_get_needs_bpf_foreign_program(Unit *u) { |
1566 | CGroupContext *c; | |
1567 | assert(u); | |
1568 | ||
1569 | c = unit_get_cgroup_context(u); | |
1570 | if (!c) | |
1571 | return false; | |
1572 | ||
1573 | return !LIST_IS_EMPTY(c->bpf_foreign_programs); | |
1574 | } | |
1575 | ||
a8e5eb17 JK |
1576 | static bool unit_get_needs_socket_bind(Unit *u) { |
1577 | CGroupContext *c; | |
1578 | assert(u); | |
1579 | ||
1580 | c = unit_get_cgroup_context(u); | |
1581 | if (!c) | |
1582 | return false; | |
1583 | ||
1584 | return c->socket_bind_allow != NULL || c->socket_bind_deny != NULL; | |
1585 | } | |
1586 | ||
c52db42b | 1587 | static CGroupMask unit_get_cgroup_mask(Unit *u) { |
efdb0237 | 1588 | CGroupMask mask = 0; |
c52db42b CD |
1589 | CGroupContext *c; |
1590 | ||
1591 | assert(u); | |
1592 | ||
806a9362 | 1593 | assert_se(c = unit_get_cgroup_context(u)); |
c710d3b4 | 1594 | |
fae9bc29 | 1595 | /* Figure out which controllers we need, based on the cgroup context object */ |
8e274523 | 1596 | |
fae9bc29 | 1597 | if (c->cpu_accounting) |
f98c2585 | 1598 | mask |= get_cpu_accounting_mask(); |
fae9bc29 LP |
1599 | |
1600 | if (cgroup_context_has_cpu_weight(c) || | |
66ebf6c0 | 1601 | cgroup_context_has_cpu_shares(c) || |
3a43da28 | 1602 | c->cpu_quota_per_sec_usec != USEC_INFINITY) |
fae9bc29 | 1603 | mask |= CGROUP_MASK_CPU; |
ecedd90f | 1604 | |
047f5d63 PH |
1605 | if (c->cpuset_cpus.set || c->cpuset_mems.set) |
1606 | mask |= CGROUP_MASK_CPUSET; | |
1607 | ||
538b4852 TH |
1608 | if (cgroup_context_has_io_config(c) || cgroup_context_has_blockio_config(c)) |
1609 | mask |= CGROUP_MASK_IO | CGROUP_MASK_BLKIO; | |
ecedd90f | 1610 | |
4ad49000 | 1611 | if (c->memory_accounting || |
da4d897e | 1612 | c->memory_limit != CGROUP_LIMIT_MAX || |
c52db42b | 1613 | unit_has_unified_memory_config(u)) |
efdb0237 | 1614 | mask |= CGROUP_MASK_MEMORY; |
8e274523 | 1615 | |
a931ad47 | 1616 | if (c->device_allow || |
084870f9 | 1617 | c->device_policy != CGROUP_DEVICE_POLICY_AUTO) |
084c7007 | 1618 | mask |= CGROUP_MASK_DEVICES | CGROUP_MASK_BPF_DEVICES; |
4ad49000 | 1619 | |
03a7b521 | 1620 | if (c->tasks_accounting || |
3a0f06c4 | 1621 | tasks_max_isset(&c->tasks_max)) |
03a7b521 LP |
1622 | mask |= CGROUP_MASK_PIDS; |
1623 | ||
fae9bc29 | 1624 | return CGROUP_MASK_EXTEND_JOINED(mask); |
8e274523 LP |
1625 | } |
1626 | ||
53aea74a | 1627 | static CGroupMask unit_get_bpf_mask(Unit *u) { |
17f14955 RG |
1628 | CGroupMask mask = 0; |
1629 | ||
fae9bc29 LP |
1630 | /* Figure out which controllers we need, based on the cgroup context, possibly taking into account children |
1631 | * too. */ | |
1632 | ||
17f14955 RG |
1633 | if (unit_get_needs_bpf_firewall(u)) |
1634 | mask |= CGROUP_MASK_BPF_FIREWALL; | |
1635 | ||
506ea51b JK |
1636 | if (unit_get_needs_bpf_foreign_program(u)) |
1637 | mask |= CGROUP_MASK_BPF_FOREIGN; | |
1638 | ||
a8e5eb17 JK |
1639 | if (unit_get_needs_socket_bind(u)) |
1640 | mask |= CGROUP_MASK_BPF_SOCKET_BIND; | |
1641 | ||
17f14955 RG |
1642 | return mask; |
1643 | } | |
1644 | ||
efdb0237 | 1645 | CGroupMask unit_get_own_mask(Unit *u) { |
4ad49000 | 1646 | CGroupContext *c; |
8e274523 | 1647 | |
442ce775 LP |
1648 | /* Returns the mask of controllers the unit needs for itself. If a unit is not properly loaded, return an empty |
1649 | * mask, as we shouldn't reflect it in the cgroup hierarchy then. */ | |
1650 | ||
1651 | if (u->load_state != UNIT_LOADED) | |
1652 | return 0; | |
efdb0237 | 1653 | |
4ad49000 LP |
1654 | c = unit_get_cgroup_context(u); |
1655 | if (!c) | |
1656 | return 0; | |
8e274523 | 1657 | |
12b975e0 | 1658 | return unit_get_cgroup_mask(u) | unit_get_bpf_mask(u) | unit_get_delegate_mask(u); |
02638280 LP |
1659 | } |
1660 | ||
1661 | CGroupMask unit_get_delegate_mask(Unit *u) { | |
1662 | CGroupContext *c; | |
1663 | ||
1664 | /* If delegation is turned on, then turn on selected controllers, unless we are on the legacy hierarchy and the | |
1665 | * process we fork into is known to drop privileges, and hence shouldn't get access to the controllers. | |
19af675e | 1666 | * |
02638280 | 1667 | * Note that on the unified hierarchy it is safe to delegate controllers to unprivileged services. */ |
a931ad47 | 1668 | |
1d9cc876 | 1669 | if (!unit_cgroup_delegate(u)) |
02638280 LP |
1670 | return 0; |
1671 | ||
1672 | if (cg_all_unified() <= 0) { | |
a931ad47 LP |
1673 | ExecContext *e; |
1674 | ||
1675 | e = unit_get_exec_context(u); | |
02638280 LP |
1676 | if (e && !exec_context_maintains_privileges(e)) |
1677 | return 0; | |
a931ad47 LP |
1678 | } |
1679 | ||
1d9cc876 | 1680 | assert_se(c = unit_get_cgroup_context(u)); |
fae9bc29 | 1681 | return CGROUP_MASK_EXTEND_JOINED(c->delegate_controllers); |
8e274523 LP |
1682 | } |
1683 | ||
d9ef5944 MK |
1684 | static CGroupMask unit_get_subtree_mask(Unit *u) { |
1685 | ||
1686 | /* Returns the mask of this subtree, meaning of the group | |
1687 | * itself and its children. */ | |
1688 | ||
1689 | return unit_get_own_mask(u) | unit_get_members_mask(u); | |
1690 | } | |
1691 | ||
efdb0237 | 1692 | CGroupMask unit_get_members_mask(Unit *u) { |
4ad49000 | 1693 | assert(u); |
bc432dc7 | 1694 | |
02638280 | 1695 | /* Returns the mask of controllers all of the unit's children require, merged */ |
efdb0237 | 1696 | |
bc432dc7 | 1697 | if (u->cgroup_members_mask_valid) |
26a17ca2 | 1698 | return u->cgroup_members_mask; /* Use cached value if possible */ |
bc432dc7 | 1699 | |
64e844e5 | 1700 | u->cgroup_members_mask = 0; |
bc432dc7 LP |
1701 | |
1702 | if (u->type == UNIT_SLICE) { | |
eef85c4a | 1703 | void *v; |
bc432dc7 | 1704 | Unit *member; |
bc432dc7 | 1705 | |
90e74a66 | 1706 | HASHMAP_FOREACH_KEY(v, member, u->dependencies[UNIT_BEFORE]) |
cb5e3bc3 CD |
1707 | if (UNIT_DEREF(member->slice) == u) |
1708 | u->cgroup_members_mask |= unit_get_subtree_mask(member); /* note that this calls ourselves again, for the children */ | |
bc432dc7 LP |
1709 | } |
1710 | ||
1711 | u->cgroup_members_mask_valid = true; | |
6414b7c9 | 1712 | return u->cgroup_members_mask; |
246aa6dd LP |
1713 | } |
1714 | ||
efdb0237 | 1715 | CGroupMask unit_get_siblings_mask(Unit *u) { |
4ad49000 | 1716 | assert(u); |
246aa6dd | 1717 | |
efdb0237 LP |
1718 | /* Returns the mask of controllers all of the unit's siblings |
1719 | * require, i.e. the members mask of the unit's parent slice | |
1720 | * if there is one. */ | |
1721 | ||
bc432dc7 | 1722 | if (UNIT_ISSET(u->slice)) |
637f421e | 1723 | return unit_get_members_mask(UNIT_DEREF(u->slice)); |
4ad49000 | 1724 | |
64e844e5 | 1725 | return unit_get_subtree_mask(u); /* we are the top-level slice */ |
246aa6dd LP |
1726 | } |
1727 | ||
d9ef5944 | 1728 | static CGroupMask unit_get_disable_mask(Unit *u) { |
4f6f62e4 CD |
1729 | CGroupContext *c; |
1730 | ||
1731 | c = unit_get_cgroup_context(u); | |
1732 | if (!c) | |
1733 | return 0; | |
1734 | ||
1735 | return c->disable_controllers; | |
1736 | } | |
1737 | ||
1738 | CGroupMask unit_get_ancestor_disable_mask(Unit *u) { | |
1739 | CGroupMask mask; | |
1740 | ||
1741 | assert(u); | |
1742 | mask = unit_get_disable_mask(u); | |
1743 | ||
1744 | /* Returns the mask of controllers which are marked as forcibly | |
1745 | * disabled in any ancestor unit or the unit in question. */ | |
1746 | ||
1747 | if (UNIT_ISSET(u->slice)) | |
1748 | mask |= unit_get_ancestor_disable_mask(UNIT_DEREF(u->slice)); | |
1749 | ||
1750 | return mask; | |
1751 | } | |
1752 | ||
efdb0237 | 1753 | CGroupMask unit_get_target_mask(Unit *u) { |
a437c5e4 | 1754 | CGroupMask own_mask, mask; |
efdb0237 | 1755 | |
a437c5e4 LP |
1756 | /* This returns the cgroup mask of all controllers to enable for a specific cgroup, i.e. everything |
1757 | * it needs itself, plus all that its children need, plus all that its siblings need. This is | |
1758 | * primarily useful on the legacy cgroup hierarchy, where we need to duplicate each cgroup in each | |
efdb0237 | 1759 | * hierarchy that shall be enabled for it. */ |
6414b7c9 | 1760 | |
a437c5e4 | 1761 | own_mask = unit_get_own_mask(u); |
84d2744b | 1762 | |
a437c5e4 | 1763 | if (own_mask & CGROUP_MASK_BPF_FIREWALL & ~u->manager->cgroup_supported) |
84d2744b ZJS |
1764 | emit_bpf_firewall_warning(u); |
1765 | ||
a437c5e4 LP |
1766 | mask = own_mask | unit_get_members_mask(u) | unit_get_siblings_mask(u); |
1767 | ||
efdb0237 | 1768 | mask &= u->manager->cgroup_supported; |
c72703e2 | 1769 | mask &= ~unit_get_ancestor_disable_mask(u); |
efdb0237 LP |
1770 | |
1771 | return mask; | |
1772 | } | |
1773 | ||
1774 | CGroupMask unit_get_enable_mask(Unit *u) { | |
1775 | CGroupMask mask; | |
1776 | ||
1777 | /* This returns the cgroup mask of all controllers to enable | |
1778 | * for the children of a specific cgroup. This is primarily | |
1779 | * useful for the unified cgroup hierarchy, where each cgroup | |
1780 | * controls which controllers are enabled for its children. */ | |
1781 | ||
1782 | mask = unit_get_members_mask(u); | |
6414b7c9 | 1783 | mask &= u->manager->cgroup_supported; |
c72703e2 | 1784 | mask &= ~unit_get_ancestor_disable_mask(u); |
6414b7c9 DS |
1785 | |
1786 | return mask; | |
1787 | } | |
1788 | ||
5af88058 | 1789 | void unit_invalidate_cgroup_members_masks(Unit *u) { |
bc432dc7 LP |
1790 | assert(u); |
1791 | ||
5af88058 LP |
1792 | /* Recurse invalidate the member masks cache all the way up the tree */ |
1793 | u->cgroup_members_mask_valid = false; | |
bc432dc7 | 1794 | |
5af88058 LP |
1795 | if (UNIT_ISSET(u->slice)) |
1796 | unit_invalidate_cgroup_members_masks(UNIT_DEREF(u->slice)); | |
6414b7c9 DS |
1797 | } |
1798 | ||
6592b975 | 1799 | const char *unit_get_realized_cgroup_path(Unit *u, CGroupMask mask) { |
03b90d4b | 1800 | |
6592b975 | 1801 | /* Returns the realized cgroup path of the specified unit where all specified controllers are available. */ |
03b90d4b LP |
1802 | |
1803 | while (u) { | |
6592b975 | 1804 | |
03b90d4b LP |
1805 | if (u->cgroup_path && |
1806 | u->cgroup_realized && | |
d94a24ca | 1807 | FLAGS_SET(u->cgroup_realized_mask, mask)) |
03b90d4b LP |
1808 | return u->cgroup_path; |
1809 | ||
1810 | u = UNIT_DEREF(u->slice); | |
1811 | } | |
1812 | ||
1813 | return NULL; | |
1814 | } | |
1815 | ||
6592b975 | 1816 | static const char *migrate_callback(CGroupMask mask, void *userdata) { |
7b639614 MK |
1817 | /* If not realized at all, migrate to root (""). |
1818 | * It may happen if we're upgrading from older version that didn't clean up. | |
1819 | */ | |
1820 | return strempty(unit_get_realized_cgroup_path(userdata, mask)); | |
6592b975 LP |
1821 | } |
1822 | ||
303ee601 | 1823 | char *unit_default_cgroup_path(const Unit *u) { |
efdb0237 LP |
1824 | _cleanup_free_ char *escaped = NULL, *slice = NULL; |
1825 | int r; | |
1826 | ||
1827 | assert(u); | |
1828 | ||
1829 | if (unit_has_name(u, SPECIAL_ROOT_SLICE)) | |
1830 | return strdup(u->manager->cgroup_root); | |
1831 | ||
1832 | if (UNIT_ISSET(u->slice) && !unit_has_name(UNIT_DEREF(u->slice), SPECIAL_ROOT_SLICE)) { | |
1833 | r = cg_slice_to_path(UNIT_DEREF(u->slice)->id, &slice); | |
1834 | if (r < 0) | |
1835 | return NULL; | |
1836 | } | |
1837 | ||
1838 | escaped = cg_escape(u->id); | |
1839 | if (!escaped) | |
1840 | return NULL; | |
1841 | ||
657ee2d8 | 1842 | return path_join(empty_to_root(u->manager->cgroup_root), slice, escaped); |
efdb0237 LP |
1843 | } |
1844 | ||
1845 | int unit_set_cgroup_path(Unit *u, const char *path) { | |
1846 | _cleanup_free_ char *p = NULL; | |
1847 | int r; | |
1848 | ||
1849 | assert(u); | |
1850 | ||
5210387e LP |
1851 | if (streq_ptr(u->cgroup_path, path)) |
1852 | return 0; | |
1853 | ||
efdb0237 LP |
1854 | if (path) { |
1855 | p = strdup(path); | |
1856 | if (!p) | |
1857 | return -ENOMEM; | |
5210387e | 1858 | } |
efdb0237 LP |
1859 | |
1860 | if (p) { | |
1861 | r = hashmap_put(u->manager->cgroup_unit, p, u); | |
1862 | if (r < 0) | |
1863 | return r; | |
1864 | } | |
1865 | ||
1866 | unit_release_cgroup(u); | |
ae2a15bc | 1867 | u->cgroup_path = TAKE_PTR(p); |
efdb0237 LP |
1868 | |
1869 | return 1; | |
1870 | } | |
1871 | ||
1872 | int unit_watch_cgroup(Unit *u) { | |
ab2c3861 | 1873 | _cleanup_free_ char *events = NULL; |
efdb0237 LP |
1874 | int r; |
1875 | ||
1876 | assert(u); | |
1877 | ||
0bb814c2 LP |
1878 | /* Watches the "cgroups.events" attribute of this unit's cgroup for "empty" events, but only if |
1879 | * cgroupv2 is available. */ | |
1880 | ||
efdb0237 LP |
1881 | if (!u->cgroup_path) |
1882 | return 0; | |
1883 | ||
0bb814c2 | 1884 | if (u->cgroup_control_inotify_wd >= 0) |
efdb0237 LP |
1885 | return 0; |
1886 | ||
1887 | /* Only applies to the unified hierarchy */ | |
c22800e4 | 1888 | r = cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER); |
b4cccbc1 LP |
1889 | if (r < 0) |
1890 | return log_error_errno(r, "Failed to determine whether the name=systemd hierarchy is unified: %m"); | |
1891 | if (r == 0) | |
efdb0237 LP |
1892 | return 0; |
1893 | ||
0bb814c2 | 1894 | /* No point in watch the top-level slice, it's never going to run empty. */ |
efdb0237 LP |
1895 | if (unit_has_name(u, SPECIAL_ROOT_SLICE)) |
1896 | return 0; | |
1897 | ||
0bb814c2 | 1898 | r = hashmap_ensure_allocated(&u->manager->cgroup_control_inotify_wd_unit, &trivial_hash_ops); |
efdb0237 LP |
1899 | if (r < 0) |
1900 | return log_oom(); | |
1901 | ||
ab2c3861 | 1902 | r = cg_get_path(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path, "cgroup.events", &events); |
efdb0237 LP |
1903 | if (r < 0) |
1904 | return log_oom(); | |
1905 | ||
0bb814c2 LP |
1906 | u->cgroup_control_inotify_wd = inotify_add_watch(u->manager->cgroup_inotify_fd, events, IN_MODIFY); |
1907 | if (u->cgroup_control_inotify_wd < 0) { | |
efdb0237 | 1908 | |
0bb814c2 LP |
1909 | if (errno == ENOENT) /* If the directory is already gone we don't need to track it, so this |
1910 | * is not an error */ | |
efdb0237 LP |
1911 | return 0; |
1912 | ||
0bb814c2 | 1913 | return log_unit_error_errno(u, errno, "Failed to add control inotify watch descriptor for control group %s: %m", u->cgroup_path); |
efdb0237 LP |
1914 | } |
1915 | ||
0bb814c2 | 1916 | r = hashmap_put(u->manager->cgroup_control_inotify_wd_unit, INT_TO_PTR(u->cgroup_control_inotify_wd), u); |
efdb0237 | 1917 | if (r < 0) |
0bb814c2 | 1918 | return log_unit_error_errno(u, r, "Failed to add control inotify watch descriptor to hash map: %m"); |
efdb0237 LP |
1919 | |
1920 | return 0; | |
1921 | } | |
1922 | ||
afcfaa69 LP |
1923 | int unit_watch_cgroup_memory(Unit *u) { |
1924 | _cleanup_free_ char *events = NULL; | |
1925 | CGroupContext *c; | |
1926 | int r; | |
1927 | ||
1928 | assert(u); | |
1929 | ||
1930 | /* Watches the "memory.events" attribute of this unit's cgroup for "oom_kill" events, but only if | |
1931 | * cgroupv2 is available. */ | |
1932 | ||
1933 | if (!u->cgroup_path) | |
1934 | return 0; | |
1935 | ||
1936 | c = unit_get_cgroup_context(u); | |
1937 | if (!c) | |
1938 | return 0; | |
1939 | ||
1940 | /* The "memory.events" attribute is only available if the memory controller is on. Let's hence tie | |
1941 | * this to memory accounting, in a way watching for OOM kills is a form of memory accounting after | |
1942 | * all. */ | |
1943 | if (!c->memory_accounting) | |
1944 | return 0; | |
1945 | ||
1946 | /* Don't watch inner nodes, as the kernel doesn't report oom_kill events recursively currently, and | |
1947 | * we also don't want to generate a log message for each parent cgroup of a process. */ | |
1948 | if (u->type == UNIT_SLICE) | |
1949 | return 0; | |
1950 | ||
1951 | if (u->cgroup_memory_inotify_wd >= 0) | |
1952 | return 0; | |
1953 | ||
1954 | /* Only applies to the unified hierarchy */ | |
1955 | r = cg_all_unified(); | |
1956 | if (r < 0) | |
1957 | return log_error_errno(r, "Failed to determine whether the memory controller is unified: %m"); | |
1958 | if (r == 0) | |
1959 | return 0; | |
1960 | ||
1961 | r = hashmap_ensure_allocated(&u->manager->cgroup_memory_inotify_wd_unit, &trivial_hash_ops); | |
1962 | if (r < 0) | |
1963 | return log_oom(); | |
1964 | ||
1965 | r = cg_get_path(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path, "memory.events", &events); | |
1966 | if (r < 0) | |
1967 | return log_oom(); | |
1968 | ||
1969 | u->cgroup_memory_inotify_wd = inotify_add_watch(u->manager->cgroup_inotify_fd, events, IN_MODIFY); | |
1970 | if (u->cgroup_memory_inotify_wd < 0) { | |
1971 | ||
1972 | if (errno == ENOENT) /* If the directory is already gone we don't need to track it, so this | |
1973 | * is not an error */ | |
1974 | return 0; | |
1975 | ||
1976 | return log_unit_error_errno(u, errno, "Failed to add memory inotify watch descriptor for control group %s: %m", u->cgroup_path); | |
1977 | } | |
1978 | ||
1979 | r = hashmap_put(u->manager->cgroup_memory_inotify_wd_unit, INT_TO_PTR(u->cgroup_memory_inotify_wd), u); | |
1980 | if (r < 0) | |
1981 | return log_unit_error_errno(u, r, "Failed to add memory inotify watch descriptor to hash map: %m"); | |
1982 | ||
1983 | return 0; | |
1984 | } | |
1985 | ||
a4634b21 LP |
1986 | int unit_pick_cgroup_path(Unit *u) { |
1987 | _cleanup_free_ char *path = NULL; | |
1988 | int r; | |
1989 | ||
1990 | assert(u); | |
1991 | ||
1992 | if (u->cgroup_path) | |
1993 | return 0; | |
1994 | ||
1995 | if (!UNIT_HAS_CGROUP_CONTEXT(u)) | |
1996 | return -EINVAL; | |
1997 | ||
1998 | path = unit_default_cgroup_path(u); | |
1999 | if (!path) | |
2000 | return log_oom(); | |
2001 | ||
2002 | r = unit_set_cgroup_path(u, path); | |
2003 | if (r == -EEXIST) | |
2004 | return log_unit_error_errno(u, r, "Control group %s exists already.", path); | |
2005 | if (r < 0) | |
2006 | return log_unit_error_errno(u, r, "Failed to set unit's control group path to %s: %m", path); | |
2007 | ||
2008 | return 0; | |
2009 | } | |
2010 | ||
7b639614 | 2011 | static int unit_update_cgroup( |
efdb0237 LP |
2012 | Unit *u, |
2013 | CGroupMask target_mask, | |
0d2d6fbf CD |
2014 | CGroupMask enable_mask, |
2015 | ManagerState state) { | |
efdb0237 | 2016 | |
7b639614 MK |
2017 | bool created, is_root_slice; |
2018 | CGroupMask migrate_mask = 0; | |
27adcc97 | 2019 | int r; |
64747e2d | 2020 | |
4ad49000 | 2021 | assert(u); |
64747e2d | 2022 | |
27c4ed79 | 2023 | if (!UNIT_HAS_CGROUP_CONTEXT(u)) |
0cd385d3 LP |
2024 | return 0; |
2025 | ||
a4634b21 LP |
2026 | /* Figure out our cgroup path */ |
2027 | r = unit_pick_cgroup_path(u); | |
2028 | if (r < 0) | |
2029 | return r; | |
b58b8e11 | 2030 | |
03b90d4b | 2031 | /* First, create our own group */ |
efdb0237 | 2032 | r = cg_create_everywhere(u->manager->cgroup_supported, target_mask, u->cgroup_path); |
23bbb0de | 2033 | if (r < 0) |
efdb0237 | 2034 | return log_unit_error_errno(u, r, "Failed to create cgroup %s: %m", u->cgroup_path); |
490c5a37 | 2035 | created = r; |
efdb0237 LP |
2036 | |
2037 | /* Start watching it */ | |
2038 | (void) unit_watch_cgroup(u); | |
afcfaa69 | 2039 | (void) unit_watch_cgroup_memory(u); |
efdb0237 | 2040 | |
7b639614 MK |
2041 | |
2042 | /* For v2 we preserve enabled controllers in delegated units, adjust others, | |
2043 | * for v1 we figure out which controller hierarchies need migration. */ | |
1fd3a10c | 2044 | if (created || !u->cgroup_realized || !unit_cgroup_delegate(u)) { |
27adcc97 | 2045 | CGroupMask result_mask = 0; |
65be7e06 ZJS |
2046 | |
2047 | /* Enable all controllers we need */ | |
27adcc97 | 2048 | r = cg_enable_everywhere(u->manager->cgroup_supported, enable_mask, u->cgroup_path, &result_mask); |
65be7e06 | 2049 | if (r < 0) |
27adcc97 LP |
2050 | log_unit_warning_errno(u, r, "Failed to enable/disable controllers on cgroup %s, ignoring: %m", u->cgroup_path); |
2051 | ||
27adcc97 LP |
2052 | /* Remember what's actually enabled now */ |
2053 | u->cgroup_enabled_mask = result_mask; | |
7b639614 MK |
2054 | |
2055 | migrate_mask = u->cgroup_realized_mask ^ target_mask; | |
65be7e06 | 2056 | } |
03b90d4b LP |
2057 | |
2058 | /* Keep track that this is now realized */ | |
4ad49000 | 2059 | u->cgroup_realized = true; |
efdb0237 | 2060 | u->cgroup_realized_mask = target_mask; |
4ad49000 | 2061 | |
7b639614 MK |
2062 | /* Migrate processes in controller hierarchies both downwards (enabling) and upwards (disabling). |
2063 | * | |
2064 | * Unnecessary controller cgroups are trimmed (after emptied by upward migration). | |
2065 | * We perform migration also with whole slices for cases when users don't care about leave | |
2066 | * granularity. Since delegated_mask is subset of target mask, we won't trim slice subtree containing | |
2067 | * delegated units. | |
2068 | */ | |
2069 | if (cg_all_unified() == 0) { | |
2070 | r = cg_migrate_v1_controllers(u->manager->cgroup_supported, migrate_mask, u->cgroup_path, migrate_callback, u); | |
2071 | if (r < 0) | |
873446f2 | 2072 | log_unit_warning_errno(u, r, "Failed to migrate controller cgroups from %s, ignoring: %m", u->cgroup_path); |
0cd385d3 | 2073 | |
7b639614 MK |
2074 | is_root_slice = unit_has_name(u, SPECIAL_ROOT_SLICE); |
2075 | r = cg_trim_v1_controllers(u->manager->cgroup_supported, ~target_mask, u->cgroup_path, !is_root_slice); | |
0cd385d3 | 2076 | if (r < 0) |
873446f2 | 2077 | log_unit_warning_errno(u, r, "Failed to delete controller cgroups %s, ignoring: %m", u->cgroup_path); |
0cd385d3 | 2078 | } |
03b90d4b | 2079 | |
0d2d6fbf CD |
2080 | /* Set attributes */ |
2081 | cgroup_context_apply(u, target_mask, state); | |
2082 | cgroup_xattr_apply(u); | |
2083 | ||
64747e2d LP |
2084 | return 0; |
2085 | } | |
2086 | ||
6592b975 LP |
2087 | static int unit_attach_pid_to_cgroup_via_bus(Unit *u, pid_t pid, const char *suffix_path) { |
2088 | _cleanup_(sd_bus_error_free) sd_bus_error error = SD_BUS_ERROR_NULL; | |
2089 | char *pp; | |
7b3fd631 | 2090 | int r; |
6592b975 | 2091 | |
7b3fd631 LP |
2092 | assert(u); |
2093 | ||
6592b975 LP |
2094 | if (MANAGER_IS_SYSTEM(u->manager)) |
2095 | return -EINVAL; | |
2096 | ||
2097 | if (!u->manager->system_bus) | |
2098 | return -EIO; | |
2099 | ||
2100 | if (!u->cgroup_path) | |
2101 | return -EINVAL; | |
2102 | ||
2103 | /* Determine this unit's cgroup path relative to our cgroup root */ | |
2104 | pp = path_startswith(u->cgroup_path, u->manager->cgroup_root); | |
2105 | if (!pp) | |
2106 | return -EINVAL; | |
2107 | ||
2108 | pp = strjoina("/", pp, suffix_path); | |
858d36c1 | 2109 | path_simplify(pp, false); |
6592b975 LP |
2110 | |
2111 | r = sd_bus_call_method(u->manager->system_bus, | |
2112 | "org.freedesktop.systemd1", | |
2113 | "/org/freedesktop/systemd1", | |
2114 | "org.freedesktop.systemd1.Manager", | |
2115 | "AttachProcessesToUnit", | |
2116 | &error, NULL, | |
2117 | "ssau", | |
2118 | NULL /* empty unit name means client's unit, i.e. us */, pp, 1, (uint32_t) pid); | |
7b3fd631 | 2119 | if (r < 0) |
6592b975 LP |
2120 | return log_unit_debug_errno(u, r, "Failed to attach unit process " PID_FMT " via the bus: %s", pid, bus_error_message(&error, r)); |
2121 | ||
2122 | return 0; | |
2123 | } | |
2124 | ||
2125 | int unit_attach_pids_to_cgroup(Unit *u, Set *pids, const char *suffix_path) { | |
2126 | CGroupMask delegated_mask; | |
2127 | const char *p; | |
6592b975 LP |
2128 | void *pidp; |
2129 | int r, q; | |
2130 | ||
2131 | assert(u); | |
2132 | ||
2133 | if (!UNIT_HAS_CGROUP_CONTEXT(u)) | |
2134 | return -EINVAL; | |
2135 | ||
2136 | if (set_isempty(pids)) | |
2137 | return 0; | |
7b3fd631 | 2138 | |
fab34748 KL |
2139 | /* Load any custom firewall BPF programs here once to test if they are existing and actually loadable. |
2140 | * Fail here early since later errors in the call chain unit_realize_cgroup to cgroup_context_apply are ignored. */ | |
2141 | r = bpf_firewall_load_custom(u); | |
2142 | if (r < 0) | |
2143 | return r; | |
2144 | ||
6592b975 | 2145 | r = unit_realize_cgroup(u); |
7b3fd631 LP |
2146 | if (r < 0) |
2147 | return r; | |
2148 | ||
6592b975 LP |
2149 | if (isempty(suffix_path)) |
2150 | p = u->cgroup_path; | |
2151 | else | |
270384b2 | 2152 | p = prefix_roota(u->cgroup_path, suffix_path); |
6592b975 LP |
2153 | |
2154 | delegated_mask = unit_get_delegate_mask(u); | |
2155 | ||
2156 | r = 0; | |
90e74a66 | 2157 | SET_FOREACH(pidp, pids) { |
6592b975 LP |
2158 | pid_t pid = PTR_TO_PID(pidp); |
2159 | CGroupController c; | |
2160 | ||
2161 | /* First, attach the PID to the main cgroup hierarchy */ | |
2162 | q = cg_attach(SYSTEMD_CGROUP_CONTROLLER, p, pid); | |
2163 | if (q < 0) { | |
2164 | log_unit_debug_errno(u, q, "Couldn't move process " PID_FMT " to requested cgroup '%s': %m", pid, p); | |
2165 | ||
065b4774 | 2166 | if (MANAGER_IS_USER(u->manager) && ERRNO_IS_PRIVILEGE(q)) { |
6592b975 LP |
2167 | int z; |
2168 | ||
2169 | /* If we are in a user instance, and we can't move the process ourselves due to | |
2170 | * permission problems, let's ask the system instance about it instead. Since it's more | |
2171 | * privileged it might be able to move the process across the leaves of a subtree who's | |
2172 | * top node is not owned by us. */ | |
2173 | ||
2174 | z = unit_attach_pid_to_cgroup_via_bus(u, pid, suffix_path); | |
2175 | if (z < 0) | |
2176 | log_unit_debug_errno(u, z, "Couldn't move process " PID_FMT " to requested cgroup '%s' via the system bus either: %m", pid, p); | |
2177 | else | |
2178 | continue; /* When the bus thing worked via the bus we are fully done for this PID. */ | |
2179 | } | |
2180 | ||
2181 | if (r >= 0) | |
2182 | r = q; /* Remember first error */ | |
2183 | ||
2184 | continue; | |
2185 | } | |
2186 | ||
2187 | q = cg_all_unified(); | |
2188 | if (q < 0) | |
2189 | return q; | |
2190 | if (q > 0) | |
2191 | continue; | |
2192 | ||
2193 | /* In the legacy hierarchy, attach the process to the request cgroup if possible, and if not to the | |
2194 | * innermost realized one */ | |
2195 | ||
2196 | for (c = 0; c < _CGROUP_CONTROLLER_MAX; c++) { | |
2197 | CGroupMask bit = CGROUP_CONTROLLER_TO_MASK(c); | |
2198 | const char *realized; | |
2199 | ||
2200 | if (!(u->manager->cgroup_supported & bit)) | |
2201 | continue; | |
2202 | ||
2203 | /* If this controller is delegated and realized, honour the caller's request for the cgroup suffix. */ | |
2204 | if (delegated_mask & u->cgroup_realized_mask & bit) { | |
2205 | q = cg_attach(cgroup_controller_to_string(c), p, pid); | |
2206 | if (q >= 0) | |
2207 | continue; /* Success! */ | |
2208 | ||
2209 | log_unit_debug_errno(u, q, "Failed to attach PID " PID_FMT " to requested cgroup %s in controller %s, falling back to unit's cgroup: %m", | |
2210 | pid, p, cgroup_controller_to_string(c)); | |
2211 | } | |
2212 | ||
2213 | /* So this controller is either not delegate or realized, or something else weird happened. In | |
2214 | * that case let's attach the PID at least to the closest cgroup up the tree that is | |
2215 | * realized. */ | |
2216 | realized = unit_get_realized_cgroup_path(u, bit); | |
2217 | if (!realized) | |
2218 | continue; /* Not even realized in the root slice? Then let's not bother */ | |
2219 | ||
2220 | q = cg_attach(cgroup_controller_to_string(c), realized, pid); | |
2221 | if (q < 0) | |
2222 | log_unit_debug_errno(u, q, "Failed to attach PID " PID_FMT " to realized cgroup %s in controller %s, ignoring: %m", | |
2223 | pid, realized, cgroup_controller_to_string(c)); | |
2224 | } | |
2225 | } | |
2226 | ||
2227 | return r; | |
7b3fd631 LP |
2228 | } |
2229 | ||
906c06f6 DM |
2230 | static bool unit_has_mask_realized( |
2231 | Unit *u, | |
2232 | CGroupMask target_mask, | |
17f14955 | 2233 | CGroupMask enable_mask) { |
906c06f6 | 2234 | |
bc432dc7 LP |
2235 | assert(u); |
2236 | ||
d5095dcd LP |
2237 | /* Returns true if this unit is fully realized. We check four things: |
2238 | * | |
2239 | * 1. Whether the cgroup was created at all | |
4e1dfa45 CD |
2240 | * 2. Whether the cgroup was created in all the hierarchies we need it to be created in (in case of cgroup v1) |
2241 | * 3. Whether the cgroup has all the right controllers enabled (in case of cgroup v2) | |
d5095dcd LP |
2242 | * 4. Whether the invalidation mask is currently zero |
2243 | * | |
2244 | * If you wonder why we mask the target realization and enable mask with CGROUP_MASK_V1/CGROUP_MASK_V2: note | |
4e1dfa45 CD |
2245 | * that there are three sets of bitmasks: CGROUP_MASK_V1 (for real cgroup v1 controllers), CGROUP_MASK_V2 (for |
2246 | * real cgroup v2 controllers) and CGROUP_MASK_BPF (for BPF-based pseudo-controllers). Now, cgroup_realized_mask | |
2247 | * is only matters for cgroup v1 controllers, and cgroup_enabled_mask only used for cgroup v2, and if they | |
d5095dcd LP |
2248 | * differ in the others, we don't really care. (After all, the cgroup_enabled_mask tracks with controllers are |
2249 | * enabled through cgroup.subtree_control, and since the BPF pseudo-controllers don't show up there, they | |
2250 | * simply don't matter. */ | |
2251 | ||
906c06f6 | 2252 | return u->cgroup_realized && |
d5095dcd LP |
2253 | ((u->cgroup_realized_mask ^ target_mask) & CGROUP_MASK_V1) == 0 && |
2254 | ((u->cgroup_enabled_mask ^ enable_mask) & CGROUP_MASK_V2) == 0 && | |
17f14955 | 2255 | u->cgroup_invalidated_mask == 0; |
6414b7c9 DS |
2256 | } |
2257 | ||
4f6f62e4 CD |
2258 | static bool unit_has_mask_disables_realized( |
2259 | Unit *u, | |
2260 | CGroupMask target_mask, | |
2261 | CGroupMask enable_mask) { | |
2262 | ||
2263 | assert(u); | |
2264 | ||
2265 | /* Returns true if all controllers which should be disabled are indeed disabled. | |
2266 | * | |
2267 | * Unlike unit_has_mask_realized, we don't care what was enabled, only that anything we want to remove is | |
2268 | * already removed. */ | |
2269 | ||
2270 | return !u->cgroup_realized || | |
2271 | (FLAGS_SET(u->cgroup_realized_mask, target_mask & CGROUP_MASK_V1) && | |
2272 | FLAGS_SET(u->cgroup_enabled_mask, enable_mask & CGROUP_MASK_V2)); | |
2273 | } | |
2274 | ||
a57669d2 CD |
2275 | static bool unit_has_mask_enables_realized( |
2276 | Unit *u, | |
2277 | CGroupMask target_mask, | |
2278 | CGroupMask enable_mask) { | |
2279 | ||
2280 | assert(u); | |
2281 | ||
2282 | /* Returns true if all controllers which should be enabled are indeed enabled. | |
2283 | * | |
2284 | * Unlike unit_has_mask_realized, we don't care about the controllers that are not present, only that anything | |
2285 | * we want to add is already added. */ | |
2286 | ||
2287 | return u->cgroup_realized && | |
c72703e2 CD |
2288 | ((u->cgroup_realized_mask | target_mask) & CGROUP_MASK_V1) == (u->cgroup_realized_mask & CGROUP_MASK_V1) && |
2289 | ((u->cgroup_enabled_mask | enable_mask) & CGROUP_MASK_V2) == (u->cgroup_enabled_mask & CGROUP_MASK_V2); | |
a57669d2 CD |
2290 | } |
2291 | ||
fb46fca7 | 2292 | static void unit_add_to_cgroup_realize_queue(Unit *u) { |
2aa57a65 LP |
2293 | assert(u); |
2294 | ||
2295 | if (u->in_cgroup_realize_queue) | |
2296 | return; | |
2297 | ||
a479c21e | 2298 | LIST_APPEND(cgroup_realize_queue, u->manager->cgroup_realize_queue, u); |
2aa57a65 LP |
2299 | u->in_cgroup_realize_queue = true; |
2300 | } | |
2301 | ||
2302 | static void unit_remove_from_cgroup_realize_queue(Unit *u) { | |
2303 | assert(u); | |
2304 | ||
2305 | if (!u->in_cgroup_realize_queue) | |
2306 | return; | |
2307 | ||
2308 | LIST_REMOVE(cgroup_realize_queue, u->manager->cgroup_realize_queue, u); | |
2309 | u->in_cgroup_realize_queue = false; | |
2310 | } | |
2311 | ||
a57669d2 CD |
2312 | /* Controllers can only be enabled breadth-first, from the root of the |
2313 | * hierarchy downwards to the unit in question. */ | |
2314 | static int unit_realize_cgroup_now_enable(Unit *u, ManagerState state) { | |
2315 | CGroupMask target_mask, enable_mask, new_target_mask, new_enable_mask; | |
2316 | int r; | |
2317 | ||
2318 | assert(u); | |
2319 | ||
2320 | /* First go deal with this unit's parent, or we won't be able to enable | |
2321 | * any new controllers at this layer. */ | |
2322 | if (UNIT_ISSET(u->slice)) { | |
2323 | r = unit_realize_cgroup_now_enable(UNIT_DEREF(u->slice), state); | |
2324 | if (r < 0) | |
2325 | return r; | |
2326 | } | |
2327 | ||
2328 | target_mask = unit_get_target_mask(u); | |
2329 | enable_mask = unit_get_enable_mask(u); | |
2330 | ||
2331 | /* We can only enable in this direction, don't try to disable anything. | |
2332 | */ | |
2333 | if (unit_has_mask_enables_realized(u, target_mask, enable_mask)) | |
2334 | return 0; | |
2335 | ||
2336 | new_target_mask = u->cgroup_realized_mask | target_mask; | |
2337 | new_enable_mask = u->cgroup_enabled_mask | enable_mask; | |
2338 | ||
7b639614 | 2339 | return unit_update_cgroup(u, new_target_mask, new_enable_mask, state); |
a57669d2 CD |
2340 | } |
2341 | ||
4f6f62e4 CD |
2342 | /* Controllers can only be disabled depth-first, from the leaves of the |
2343 | * hierarchy upwards to the unit in question. */ | |
2344 | static int unit_realize_cgroup_now_disable(Unit *u, ManagerState state) { | |
4f6f62e4 CD |
2345 | Unit *m; |
2346 | void *v; | |
2347 | ||
2348 | assert(u); | |
2349 | ||
2350 | if (u->type != UNIT_SLICE) | |
2351 | return 0; | |
2352 | ||
90e74a66 | 2353 | HASHMAP_FOREACH_KEY(v, m, u->dependencies[UNIT_BEFORE]) { |
4f6f62e4 CD |
2354 | CGroupMask target_mask, enable_mask, new_target_mask, new_enable_mask; |
2355 | int r; | |
2356 | ||
2357 | if (UNIT_DEREF(m->slice) != u) | |
2358 | continue; | |
2359 | ||
2360 | /* The cgroup for this unit might not actually be fully | |
2361 | * realised yet, in which case it isn't holding any controllers | |
2362 | * open anyway. */ | |
d9ef5944 | 2363 | if (!m->cgroup_realized) |
4f6f62e4 CD |
2364 | continue; |
2365 | ||
2366 | /* We must disable those below us first in order to release the | |
2367 | * controller. */ | |
2368 | if (m->type == UNIT_SLICE) | |
2369 | (void) unit_realize_cgroup_now_disable(m, state); | |
2370 | ||
2371 | target_mask = unit_get_target_mask(m); | |
2372 | enable_mask = unit_get_enable_mask(m); | |
2373 | ||
2374 | /* We can only disable in this direction, don't try to enable | |
2375 | * anything. */ | |
2376 | if (unit_has_mask_disables_realized(m, target_mask, enable_mask)) | |
2377 | continue; | |
2378 | ||
2379 | new_target_mask = m->cgroup_realized_mask & target_mask; | |
2380 | new_enable_mask = m->cgroup_enabled_mask & enable_mask; | |
2381 | ||
7b639614 | 2382 | r = unit_update_cgroup(m, new_target_mask, new_enable_mask, state); |
4f6f62e4 CD |
2383 | if (r < 0) |
2384 | return r; | |
2385 | } | |
2386 | ||
2387 | return 0; | |
2388 | } | |
a57669d2 | 2389 | |
6414b7c9 DS |
2390 | /* Check if necessary controllers and attributes for a unit are in place. |
2391 | * | |
a57669d2 CD |
2392 | * - If so, do nothing. |
2393 | * - If not, create paths, move processes over, and set attributes. | |
2394 | * | |
2395 | * Controllers can only be *enabled* in a breadth-first way, and *disabled* in | |
2396 | * a depth-first way. As such the process looks like this: | |
2397 | * | |
2398 | * Suppose we have a cgroup hierarchy which looks like this: | |
2399 | * | |
2400 | * root | |
2401 | * / \ | |
2402 | * / \ | |
2403 | * / \ | |
2404 | * a b | |
2405 | * / \ / \ | |
2406 | * / \ / \ | |
2407 | * c d e f | |
2408 | * / \ / \ / \ / \ | |
2409 | * h i j k l m n o | |
2410 | * | |
2411 | * 1. We want to realise cgroup "d" now. | |
c72703e2 | 2412 | * 2. cgroup "a" has DisableControllers=cpu in the associated unit. |
a57669d2 CD |
2413 | * 3. cgroup "k" just started requesting the memory controller. |
2414 | * | |
2415 | * To make this work we must do the following in order: | |
2416 | * | |
2417 | * 1. Disable CPU controller in k, j | |
2418 | * 2. Disable CPU controller in d | |
2419 | * 3. Enable memory controller in root | |
2420 | * 4. Enable memory controller in a | |
2421 | * 5. Enable memory controller in d | |
2422 | * 6. Enable memory controller in k | |
2423 | * | |
2424 | * Notice that we need to touch j in one direction, but not the other. We also | |
2425 | * don't go beyond d when disabling -- it's up to "a" to get realized if it | |
2426 | * wants to disable further. The basic rules are therefore: | |
2427 | * | |
2428 | * - If you're disabling something, you need to realise all of the cgroups from | |
2429 | * your recursive descendants to the root. This starts from the leaves. | |
2430 | * - If you're enabling something, you need to realise from the root cgroup | |
2431 | * downwards, but you don't need to iterate your recursive descendants. | |
6414b7c9 DS |
2432 | * |
2433 | * Returns 0 on success and < 0 on failure. */ | |
db785129 | 2434 | static int unit_realize_cgroup_now(Unit *u, ManagerState state) { |
efdb0237 | 2435 | CGroupMask target_mask, enable_mask; |
6414b7c9 | 2436 | int r; |
64747e2d | 2437 | |
4ad49000 | 2438 | assert(u); |
64747e2d | 2439 | |
2aa57a65 | 2440 | unit_remove_from_cgroup_realize_queue(u); |
64747e2d | 2441 | |
efdb0237 | 2442 | target_mask = unit_get_target_mask(u); |
ccf78df1 TH |
2443 | enable_mask = unit_get_enable_mask(u); |
2444 | ||
17f14955 | 2445 | if (unit_has_mask_realized(u, target_mask, enable_mask)) |
0a1eb06d | 2446 | return 0; |
64747e2d | 2447 | |
4f6f62e4 CD |
2448 | /* Disable controllers below us, if there are any */ |
2449 | r = unit_realize_cgroup_now_disable(u, state); | |
2450 | if (r < 0) | |
2451 | return r; | |
2452 | ||
2453 | /* Enable controllers above us, if there are any */ | |
6414b7c9 | 2454 | if (UNIT_ISSET(u->slice)) { |
a57669d2 | 2455 | r = unit_realize_cgroup_now_enable(UNIT_DEREF(u->slice), state); |
6414b7c9 DS |
2456 | if (r < 0) |
2457 | return r; | |
2458 | } | |
4ad49000 | 2459 | |
0d2d6fbf | 2460 | /* Now actually deal with the cgroup we were trying to realise and set attributes */ |
7b639614 | 2461 | r = unit_update_cgroup(u, target_mask, enable_mask, state); |
6414b7c9 DS |
2462 | if (r < 0) |
2463 | return r; | |
2464 | ||
c2baf11c LP |
2465 | /* Now, reset the invalidation mask */ |
2466 | u->cgroup_invalidated_mask = 0; | |
6414b7c9 | 2467 | return 0; |
64747e2d LP |
2468 | } |
2469 | ||
91a6073e | 2470 | unsigned manager_dispatch_cgroup_realize_queue(Manager *m) { |
db785129 | 2471 | ManagerState state; |
4ad49000 | 2472 | unsigned n = 0; |
db785129 | 2473 | Unit *i; |
6414b7c9 | 2474 | int r; |
ecedd90f | 2475 | |
91a6073e LP |
2476 | assert(m); |
2477 | ||
db785129 LP |
2478 | state = manager_state(m); |
2479 | ||
91a6073e LP |
2480 | while ((i = m->cgroup_realize_queue)) { |
2481 | assert(i->in_cgroup_realize_queue); | |
ecedd90f | 2482 | |
2aa57a65 LP |
2483 | if (UNIT_IS_INACTIVE_OR_FAILED(unit_active_state(i))) { |
2484 | /* Maybe things changed, and the unit is not actually active anymore? */ | |
2485 | unit_remove_from_cgroup_realize_queue(i); | |
2486 | continue; | |
2487 | } | |
2488 | ||
db785129 | 2489 | r = unit_realize_cgroup_now(i, state); |
6414b7c9 | 2490 | if (r < 0) |
efdb0237 | 2491 | log_warning_errno(r, "Failed to realize cgroups for queued unit %s, ignoring: %m", i->id); |
0a1eb06d | 2492 | |
4ad49000 LP |
2493 | n++; |
2494 | } | |
ecedd90f | 2495 | |
4ad49000 | 2496 | return n; |
8e274523 LP |
2497 | } |
2498 | ||
4c591f39 MK |
2499 | void unit_add_family_to_cgroup_realize_queue(Unit *u) { |
2500 | assert(u); | |
2501 | assert(u->type == UNIT_SLICE); | |
ca949c9d | 2502 | |
4c591f39 MK |
2503 | /* Family of a unit for is defined as (immediate) children of the unit and immediate children of all |
2504 | * its ancestors. | |
2505 | * | |
2506 | * Ideally we would enqueue ancestor path only (bottom up). However, on cgroup-v1 scheduling becomes | |
2507 | * very weird if two units that own processes reside in the same slice, but one is realized in the | |
2508 | * "cpu" hierarchy and one is not (for example because one has CPUWeight= set and the other does | |
2509 | * not), because that means individual processes need to be scheduled against whole cgroups. Let's | |
2510 | * avoid this asymmetry by always ensuring that siblings of a unit are always realized in their v1 | |
2511 | * controller hierarchies too (if unit requires the controller to be realized). | |
e1e98911 | 2512 | * |
4c591f39 MK |
2513 | * The function must invalidate cgroup_members_mask of all ancestors in order to calculate up to date |
2514 | * masks. */ | |
2515 | ||
2516 | do { | |
4ad49000 | 2517 | Unit *m; |
eef85c4a | 2518 | void *v; |
8f53a7b8 | 2519 | |
4c591f39 MK |
2520 | /* Children of u likely changed when we're called */ |
2521 | u->cgroup_members_mask_valid = false; | |
f23ba94d | 2522 | |
90e74a66 | 2523 | HASHMAP_FOREACH_KEY(v, m, u->dependencies[UNIT_BEFORE]) { |
65f6b6bd | 2524 | /* Skip units that have a dependency on the slice but aren't actually in it. */ |
4c591f39 | 2525 | if (UNIT_DEREF(m->slice) != u) |
fb46fca7 | 2526 | continue; |
8e274523 | 2527 | |
65f6b6bd | 2528 | /* No point in doing cgroup application for units without active processes. */ |
6414b7c9 DS |
2529 | if (UNIT_IS_INACTIVE_OR_FAILED(unit_active_state(m))) |
2530 | continue; | |
2531 | ||
e1e98911 LP |
2532 | /* We only enqueue siblings if they were realized once at least, in the main |
2533 | * hierarchy. */ | |
2534 | if (!m->cgroup_realized) | |
2535 | continue; | |
2536 | ||
65f6b6bd LP |
2537 | /* If the unit doesn't need any new controllers and has current ones realized, it |
2538 | * doesn't need any changes. */ | |
906c06f6 DM |
2539 | if (unit_has_mask_realized(m, |
2540 | unit_get_target_mask(m), | |
17f14955 | 2541 | unit_get_enable_mask(m))) |
6414b7c9 DS |
2542 | continue; |
2543 | ||
91a6073e | 2544 | unit_add_to_cgroup_realize_queue(m); |
50159e6a LP |
2545 | } |
2546 | ||
4c591f39 MK |
2547 | /* Parent comes after children */ |
2548 | unit_add_to_cgroup_realize_queue(u); | |
2549 | } while ((u = UNIT_DEREF(u->slice))); | |
4ad49000 LP |
2550 | } |
2551 | ||
0a1eb06d | 2552 | int unit_realize_cgroup(Unit *u) { |
4ad49000 LP |
2553 | assert(u); |
2554 | ||
35b7ff80 | 2555 | if (!UNIT_HAS_CGROUP_CONTEXT(u)) |
0a1eb06d | 2556 | return 0; |
8e274523 | 2557 | |
4c591f39 MK |
2558 | /* So, here's the deal: when realizing the cgroups for this unit, we need to first create all |
2559 | * parents, but there's more actually: for the weight-based controllers we also need to make sure | |
2560 | * that all our siblings (i.e. units that are in the same slice as we are) have cgroups, too. On the | |
2561 | * other hand, when a controller is removed from realized set, it may become unnecessary in siblings | |
2562 | * and ancestors and they should be (de)realized too. | |
2563 | * | |
2564 | * This call will defer work on the siblings and derealized ancestors to the next event loop | |
2565 | * iteration and synchronously creates the parent cgroups (unit_realize_cgroup_now). */ | |
ca949c9d | 2566 | |
4c591f39 MK |
2567 | if (UNIT_ISSET(u->slice)) |
2568 | unit_add_family_to_cgroup_realize_queue(UNIT_DEREF(u->slice)); | |
4ad49000 | 2569 | |
6414b7c9 | 2570 | /* And realize this one now (and apply the values) */ |
db785129 | 2571 | return unit_realize_cgroup_now(u, manager_state(u->manager)); |
8e274523 LP |
2572 | } |
2573 | ||
efdb0237 LP |
2574 | void unit_release_cgroup(Unit *u) { |
2575 | assert(u); | |
2576 | ||
8a0d5388 LP |
2577 | /* Forgets all cgroup details for this cgroup — but does *not* destroy the cgroup. This is hence OK to call |
2578 | * when we close down everything for reexecution, where we really want to leave the cgroup in place. */ | |
efdb0237 LP |
2579 | |
2580 | if (u->cgroup_path) { | |
2581 | (void) hashmap_remove(u->manager->cgroup_unit, u->cgroup_path); | |
2582 | u->cgroup_path = mfree(u->cgroup_path); | |
2583 | } | |
2584 | ||
0bb814c2 LP |
2585 | if (u->cgroup_control_inotify_wd >= 0) { |
2586 | if (inotify_rm_watch(u->manager->cgroup_inotify_fd, u->cgroup_control_inotify_wd) < 0) | |
2587 | log_unit_debug_errno(u, errno, "Failed to remove cgroup control inotify watch %i for %s, ignoring: %m", u->cgroup_control_inotify_wd, u->id); | |
efdb0237 | 2588 | |
0bb814c2 LP |
2589 | (void) hashmap_remove(u->manager->cgroup_control_inotify_wd_unit, INT_TO_PTR(u->cgroup_control_inotify_wd)); |
2590 | u->cgroup_control_inotify_wd = -1; | |
efdb0237 | 2591 | } |
afcfaa69 LP |
2592 | |
2593 | if (u->cgroup_memory_inotify_wd >= 0) { | |
2594 | if (inotify_rm_watch(u->manager->cgroup_inotify_fd, u->cgroup_memory_inotify_wd) < 0) | |
2595 | log_unit_debug_errno(u, errno, "Failed to remove cgroup memory inotify watch %i for %s, ignoring: %m", u->cgroup_memory_inotify_wd, u->id); | |
2596 | ||
2597 | (void) hashmap_remove(u->manager->cgroup_memory_inotify_wd_unit, INT_TO_PTR(u->cgroup_memory_inotify_wd)); | |
2598 | u->cgroup_memory_inotify_wd = -1; | |
2599 | } | |
efdb0237 LP |
2600 | } |
2601 | ||
e08dabfe AZ |
2602 | bool unit_maybe_release_cgroup(Unit *u) { |
2603 | int r; | |
2604 | ||
2605 | assert(u); | |
2606 | ||
2607 | if (!u->cgroup_path) | |
2608 | return true; | |
2609 | ||
2610 | /* Don't release the cgroup if there are still processes under it. If we get notified later when all the | |
2611 | * processes exit (e.g. the processes were in D-state and exited after the unit was marked as failed) | |
2612 | * we need the cgroup paths to continue to be tracked by the manager so they can be looked up and cleaned | |
2613 | * up later. */ | |
2614 | r = cg_is_empty_recursive(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path); | |
2615 | if (r < 0) | |
2616 | log_unit_debug_errno(u, r, "Error checking if the cgroup is recursively empty, ignoring: %m"); | |
2617 | else if (r == 1) { | |
2618 | unit_release_cgroup(u); | |
2619 | return true; | |
2620 | } | |
2621 | ||
2622 | return false; | |
2623 | } | |
2624 | ||
efdb0237 | 2625 | void unit_prune_cgroup(Unit *u) { |
8e274523 | 2626 | int r; |
efdb0237 | 2627 | bool is_root_slice; |
8e274523 | 2628 | |
4ad49000 | 2629 | assert(u); |
8e274523 | 2630 | |
efdb0237 LP |
2631 | /* Removes the cgroup, if empty and possible, and stops watching it. */ |
2632 | ||
4ad49000 LP |
2633 | if (!u->cgroup_path) |
2634 | return; | |
8e274523 | 2635 | |
fe700f46 LP |
2636 | (void) unit_get_cpu_usage(u, NULL); /* Cache the last CPU usage value before we destroy the cgroup */ |
2637 | ||
efdb0237 LP |
2638 | is_root_slice = unit_has_name(u, SPECIAL_ROOT_SLICE); |
2639 | ||
2640 | r = cg_trim_everywhere(u->manager->cgroup_supported, u->cgroup_path, !is_root_slice); | |
0219b352 DB |
2641 | if (r < 0) |
2642 | /* One reason we could have failed here is, that the cgroup still contains a process. | |
2643 | * However, if the cgroup becomes removable at a later time, it might be removed when | |
2644 | * the containing slice is stopped. So even if we failed now, this unit shouldn't assume | |
2645 | * that the cgroup is still realized the next time it is started. Do not return early | |
2646 | * on error, continue cleanup. */ | |
8ed6f81b | 2647 | log_unit_full_errno(u, r == -EBUSY ? LOG_DEBUG : LOG_WARNING, r, "Failed to destroy cgroup %s, ignoring: %m", u->cgroup_path); |
8e274523 | 2648 | |
efdb0237 LP |
2649 | if (is_root_slice) |
2650 | return; | |
2651 | ||
e08dabfe AZ |
2652 | if (!unit_maybe_release_cgroup(u)) /* Returns true if the cgroup was released */ |
2653 | return; | |
0a1eb06d | 2654 | |
4ad49000 | 2655 | u->cgroup_realized = false; |
bc432dc7 | 2656 | u->cgroup_realized_mask = 0; |
ccf78df1 | 2657 | u->cgroup_enabled_mask = 0; |
084c7007 RG |
2658 | |
2659 | u->bpf_device_control_installed = bpf_program_unref(u->bpf_device_control_installed); | |
8e274523 LP |
2660 | } |
2661 | ||
efdb0237 | 2662 | int unit_search_main_pid(Unit *u, pid_t *ret) { |
4ad49000 | 2663 | _cleanup_fclose_ FILE *f = NULL; |
4d051546 | 2664 | pid_t pid = 0, npid; |
efdb0237 | 2665 | int r; |
4ad49000 LP |
2666 | |
2667 | assert(u); | |
efdb0237 | 2668 | assert(ret); |
4ad49000 LP |
2669 | |
2670 | if (!u->cgroup_path) | |
efdb0237 | 2671 | return -ENXIO; |
4ad49000 | 2672 | |
efdb0237 LP |
2673 | r = cg_enumerate_processes(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path, &f); |
2674 | if (r < 0) | |
2675 | return r; | |
4ad49000 | 2676 | |
4ad49000 | 2677 | while (cg_read_pid(f, &npid) > 0) { |
4ad49000 LP |
2678 | |
2679 | if (npid == pid) | |
2680 | continue; | |
8e274523 | 2681 | |
4d051546 | 2682 | if (pid_is_my_child(npid) == 0) |
4ad49000 | 2683 | continue; |
8e274523 | 2684 | |
efdb0237 | 2685 | if (pid != 0) |
4ad49000 LP |
2686 | /* Dang, there's more than one daemonized PID |
2687 | in this group, so we don't know what process | |
2688 | is the main process. */ | |
efdb0237 LP |
2689 | |
2690 | return -ENODATA; | |
8e274523 | 2691 | |
4ad49000 | 2692 | pid = npid; |
8e274523 LP |
2693 | } |
2694 | ||
efdb0237 LP |
2695 | *ret = pid; |
2696 | return 0; | |
2697 | } | |
2698 | ||
2699 | static int unit_watch_pids_in_path(Unit *u, const char *path) { | |
b3c5bad3 | 2700 | _cleanup_closedir_ DIR *d = NULL; |
efdb0237 LP |
2701 | _cleanup_fclose_ FILE *f = NULL; |
2702 | int ret = 0, r; | |
2703 | ||
2704 | assert(u); | |
2705 | assert(path); | |
2706 | ||
2707 | r = cg_enumerate_processes(SYSTEMD_CGROUP_CONTROLLER, path, &f); | |
2708 | if (r < 0) | |
2709 | ret = r; | |
2710 | else { | |
2711 | pid_t pid; | |
2712 | ||
2713 | while ((r = cg_read_pid(f, &pid)) > 0) { | |
f75f613d | 2714 | r = unit_watch_pid(u, pid, false); |
efdb0237 LP |
2715 | if (r < 0 && ret >= 0) |
2716 | ret = r; | |
2717 | } | |
2718 | ||
2719 | if (r < 0 && ret >= 0) | |
2720 | ret = r; | |
2721 | } | |
2722 | ||
2723 | r = cg_enumerate_subgroups(SYSTEMD_CGROUP_CONTROLLER, path, &d); | |
2724 | if (r < 0) { | |
2725 | if (ret >= 0) | |
2726 | ret = r; | |
2727 | } else { | |
2728 | char *fn; | |
2729 | ||
2730 | while ((r = cg_read_subgroup(d, &fn)) > 0) { | |
2731 | _cleanup_free_ char *p = NULL; | |
2732 | ||
95b21cff | 2733 | p = path_join(empty_to_root(path), fn); |
efdb0237 LP |
2734 | free(fn); |
2735 | ||
2736 | if (!p) | |
2737 | return -ENOMEM; | |
2738 | ||
2739 | r = unit_watch_pids_in_path(u, p); | |
2740 | if (r < 0 && ret >= 0) | |
2741 | ret = r; | |
2742 | } | |
2743 | ||
2744 | if (r < 0 && ret >= 0) | |
2745 | ret = r; | |
2746 | } | |
2747 | ||
2748 | return ret; | |
2749 | } | |
2750 | ||
11aef522 LP |
2751 | int unit_synthesize_cgroup_empty_event(Unit *u) { |
2752 | int r; | |
2753 | ||
2754 | assert(u); | |
2755 | ||
2756 | /* Enqueue a synthetic cgroup empty event if this unit doesn't watch any PIDs anymore. This is compatibility | |
2757 | * support for non-unified systems where notifications aren't reliable, and hence need to take whatever we can | |
2758 | * get as notification source as soon as we stopped having any useful PIDs to watch for. */ | |
2759 | ||
2760 | if (!u->cgroup_path) | |
2761 | return -ENOENT; | |
2762 | ||
2763 | r = cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER); | |
2764 | if (r < 0) | |
2765 | return r; | |
2766 | if (r > 0) /* On unified we have reliable notifications, and don't need this */ | |
2767 | return 0; | |
2768 | ||
2769 | if (!set_isempty(u->pids)) | |
2770 | return 0; | |
2771 | ||
2772 | unit_add_to_cgroup_empty_queue(u); | |
2773 | return 0; | |
2774 | } | |
2775 | ||
efdb0237 | 2776 | int unit_watch_all_pids(Unit *u) { |
b4cccbc1 LP |
2777 | int r; |
2778 | ||
efdb0237 LP |
2779 | assert(u); |
2780 | ||
2781 | /* Adds all PIDs from our cgroup to the set of PIDs we | |
2782 | * watch. This is a fallback logic for cases where we do not | |
2783 | * get reliable cgroup empty notifications: we try to use | |
2784 | * SIGCHLD as replacement. */ | |
2785 | ||
2786 | if (!u->cgroup_path) | |
2787 | return -ENOENT; | |
2788 | ||
c22800e4 | 2789 | r = cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER); |
b4cccbc1 LP |
2790 | if (r < 0) |
2791 | return r; | |
2792 | if (r > 0) /* On unified we can use proper notifications */ | |
efdb0237 LP |
2793 | return 0; |
2794 | ||
2795 | return unit_watch_pids_in_path(u, u->cgroup_path); | |
2796 | } | |
2797 | ||
09e24654 LP |
2798 | static int on_cgroup_empty_event(sd_event_source *s, void *userdata) { |
2799 | Manager *m = userdata; | |
2800 | Unit *u; | |
efdb0237 LP |
2801 | int r; |
2802 | ||
09e24654 LP |
2803 | assert(s); |
2804 | assert(m); | |
efdb0237 | 2805 | |
09e24654 LP |
2806 | u = m->cgroup_empty_queue; |
2807 | if (!u) | |
efdb0237 LP |
2808 | return 0; |
2809 | ||
09e24654 LP |
2810 | assert(u->in_cgroup_empty_queue); |
2811 | u->in_cgroup_empty_queue = false; | |
2812 | LIST_REMOVE(cgroup_empty_queue, m->cgroup_empty_queue, u); | |
2813 | ||
2814 | if (m->cgroup_empty_queue) { | |
2815 | /* More stuff queued, let's make sure we remain enabled */ | |
2816 | r = sd_event_source_set_enabled(s, SD_EVENT_ONESHOT); | |
2817 | if (r < 0) | |
19a691a9 | 2818 | log_debug_errno(r, "Failed to reenable cgroup empty event source, ignoring: %m"); |
09e24654 | 2819 | } |
efdb0237 LP |
2820 | |
2821 | unit_add_to_gc_queue(u); | |
2822 | ||
2823 | if (UNIT_VTABLE(u)->notify_cgroup_empty) | |
2824 | UNIT_VTABLE(u)->notify_cgroup_empty(u); | |
2825 | ||
2826 | return 0; | |
2827 | } | |
2828 | ||
09e24654 LP |
2829 | void unit_add_to_cgroup_empty_queue(Unit *u) { |
2830 | int r; | |
2831 | ||
2832 | assert(u); | |
2833 | ||
2834 | /* Note that there are four different ways how cgroup empty events reach us: | |
2835 | * | |
2836 | * 1. On the unified hierarchy we get an inotify event on the cgroup | |
2837 | * | |
2838 | * 2. On the legacy hierarchy, when running in system mode, we get a datagram on the cgroup agent socket | |
2839 | * | |
2840 | * 3. On the legacy hierarchy, when running in user mode, we get a D-Bus signal on the system bus | |
2841 | * | |
2842 | * 4. On the legacy hierarchy, in service units we start watching all processes of the cgroup for SIGCHLD as | |
2843 | * soon as we get one SIGCHLD, to deal with unreliable cgroup notifications. | |
2844 | * | |
2845 | * Regardless which way we got the notification, we'll verify it here, and then add it to a separate | |
2846 | * queue. This queue will be dispatched at a lower priority than the SIGCHLD handler, so that we always use | |
2847 | * SIGCHLD if we can get it first, and only use the cgroup empty notifications if there's no SIGCHLD pending | |
2848 | * (which might happen if the cgroup doesn't contain processes that are our own child, which is typically the | |
2849 | * case for scope units). */ | |
2850 | ||
2851 | if (u->in_cgroup_empty_queue) | |
2852 | return; | |
2853 | ||
2854 | /* Let's verify that the cgroup is really empty */ | |
2855 | if (!u->cgroup_path) | |
2856 | return; | |
e1e98911 | 2857 | |
09e24654 LP |
2858 | r = cg_is_empty_recursive(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path); |
2859 | if (r < 0) { | |
2860 | log_unit_debug_errno(u, r, "Failed to determine whether cgroup %s is empty: %m", u->cgroup_path); | |
2861 | return; | |
2862 | } | |
2863 | if (r == 0) | |
2864 | return; | |
2865 | ||
2866 | LIST_PREPEND(cgroup_empty_queue, u->manager->cgroup_empty_queue, u); | |
2867 | u->in_cgroup_empty_queue = true; | |
2868 | ||
2869 | /* Trigger the defer event */ | |
2870 | r = sd_event_source_set_enabled(u->manager->cgroup_empty_event_source, SD_EVENT_ONESHOT); | |
2871 | if (r < 0) | |
2872 | log_debug_errno(r, "Failed to enable cgroup empty event source: %m"); | |
2873 | } | |
2874 | ||
d9e45bc3 MS |
2875 | static void unit_remove_from_cgroup_empty_queue(Unit *u) { |
2876 | assert(u); | |
2877 | ||
2878 | if (!u->in_cgroup_empty_queue) | |
2879 | return; | |
2880 | ||
2881 | LIST_REMOVE(cgroup_empty_queue, u->manager->cgroup_empty_queue, u); | |
2882 | u->in_cgroup_empty_queue = false; | |
2883 | } | |
2884 | ||
fe8d22fb AZ |
2885 | int unit_check_oomd_kill(Unit *u) { |
2886 | _cleanup_free_ char *value = NULL; | |
2887 | bool increased; | |
2888 | uint64_t n = 0; | |
2889 | int r; | |
2890 | ||
2891 | if (!u->cgroup_path) | |
2892 | return 0; | |
2893 | ||
2894 | r = cg_all_unified(); | |
2895 | if (r < 0) | |
2896 | return log_unit_debug_errno(u, r, "Couldn't determine whether we are in all unified mode: %m"); | |
2897 | else if (r == 0) | |
2898 | return 0; | |
2899 | ||
e3038333 | 2900 | r = cg_get_xattr_malloc(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path, "user.oomd_kill", &value); |
fe8d22fb AZ |
2901 | if (r < 0 && r != -ENODATA) |
2902 | return r; | |
2903 | ||
2904 | if (!isempty(value)) { | |
2905 | r = safe_atou64(value, &n); | |
2906 | if (r < 0) | |
2907 | return r; | |
2908 | } | |
2909 | ||
2910 | increased = n > u->managed_oom_kill_last; | |
2911 | u->managed_oom_kill_last = n; | |
2912 | ||
2913 | if (!increased) | |
2914 | return 0; | |
2915 | ||
2916 | if (n > 0) | |
c2503e35 RH |
2917 | log_unit_struct(u, LOG_NOTICE, |
2918 | "MESSAGE_ID=" SD_MESSAGE_UNIT_OOMD_KILL_STR, | |
2919 | LOG_UNIT_INVOCATION_ID(u), | |
2920 | LOG_UNIT_MESSAGE(u, "systemd-oomd killed %"PRIu64" process(es) in this unit.", n)); | |
fe8d22fb AZ |
2921 | |
2922 | return 1; | |
2923 | } | |
2924 | ||
2ba6ae6b | 2925 | int unit_check_oom(Unit *u) { |
afcfaa69 LP |
2926 | _cleanup_free_ char *oom_kill = NULL; |
2927 | bool increased; | |
2928 | uint64_t c; | |
2929 | int r; | |
2930 | ||
2931 | if (!u->cgroup_path) | |
2932 | return 0; | |
2933 | ||
2934 | r = cg_get_keyed_attribute("memory", u->cgroup_path, "memory.events", STRV_MAKE("oom_kill"), &oom_kill); | |
2935 | if (r < 0) | |
2936 | return log_unit_debug_errno(u, r, "Failed to read oom_kill field of memory.events cgroup attribute: %m"); | |
2937 | ||
2938 | r = safe_atou64(oom_kill, &c); | |
2939 | if (r < 0) | |
2940 | return log_unit_debug_errno(u, r, "Failed to parse oom_kill field: %m"); | |
2941 | ||
2942 | increased = c > u->oom_kill_last; | |
2943 | u->oom_kill_last = c; | |
2944 | ||
2945 | if (!increased) | |
2946 | return 0; | |
2947 | ||
c2503e35 RH |
2948 | log_unit_struct(u, LOG_NOTICE, |
2949 | "MESSAGE_ID=" SD_MESSAGE_UNIT_OUT_OF_MEMORY_STR, | |
2950 | LOG_UNIT_INVOCATION_ID(u), | |
2951 | LOG_UNIT_MESSAGE(u, "A process of this unit has been killed by the OOM killer.")); | |
afcfaa69 LP |
2952 | |
2953 | if (UNIT_VTABLE(u)->notify_cgroup_oom) | |
2954 | UNIT_VTABLE(u)->notify_cgroup_oom(u); | |
2955 | ||
2956 | return 1; | |
2957 | } | |
2958 | ||
2959 | static int on_cgroup_oom_event(sd_event_source *s, void *userdata) { | |
2960 | Manager *m = userdata; | |
2961 | Unit *u; | |
2962 | int r; | |
2963 | ||
2964 | assert(s); | |
2965 | assert(m); | |
2966 | ||
2967 | u = m->cgroup_oom_queue; | |
2968 | if (!u) | |
2969 | return 0; | |
2970 | ||
2971 | assert(u->in_cgroup_oom_queue); | |
2972 | u->in_cgroup_oom_queue = false; | |
2973 | LIST_REMOVE(cgroup_oom_queue, m->cgroup_oom_queue, u); | |
2974 | ||
2975 | if (m->cgroup_oom_queue) { | |
2976 | /* More stuff queued, let's make sure we remain enabled */ | |
2977 | r = sd_event_source_set_enabled(s, SD_EVENT_ONESHOT); | |
2978 | if (r < 0) | |
2979 | log_debug_errno(r, "Failed to reenable cgroup oom event source, ignoring: %m"); | |
2980 | } | |
2981 | ||
2982 | (void) unit_check_oom(u); | |
2983 | return 0; | |
2984 | } | |
2985 | ||
2986 | static void unit_add_to_cgroup_oom_queue(Unit *u) { | |
2987 | int r; | |
2988 | ||
2989 | assert(u); | |
2990 | ||
2991 | if (u->in_cgroup_oom_queue) | |
2992 | return; | |
2993 | if (!u->cgroup_path) | |
2994 | return; | |
2995 | ||
2996 | LIST_PREPEND(cgroup_oom_queue, u->manager->cgroup_oom_queue, u); | |
2997 | u->in_cgroup_oom_queue = true; | |
2998 | ||
2999 | /* Trigger the defer event */ | |
3000 | if (!u->manager->cgroup_oom_event_source) { | |
3001 | _cleanup_(sd_event_source_unrefp) sd_event_source *s = NULL; | |
3002 | ||
3003 | r = sd_event_add_defer(u->manager->event, &s, on_cgroup_oom_event, u->manager); | |
3004 | if (r < 0) { | |
3005 | log_error_errno(r, "Failed to create cgroup oom event source: %m"); | |
3006 | return; | |
3007 | } | |
3008 | ||
3009 | r = sd_event_source_set_priority(s, SD_EVENT_PRIORITY_NORMAL-8); | |
3010 | if (r < 0) { | |
3011 | log_error_errno(r, "Failed to set priority of cgroup oom event source: %m"); | |
3012 | return; | |
3013 | } | |
3014 | ||
3015 | (void) sd_event_source_set_description(s, "cgroup-oom"); | |
3016 | u->manager->cgroup_oom_event_source = TAKE_PTR(s); | |
3017 | } | |
3018 | ||
3019 | r = sd_event_source_set_enabled(u->manager->cgroup_oom_event_source, SD_EVENT_ONESHOT); | |
3020 | if (r < 0) | |
3021 | log_error_errno(r, "Failed to enable cgroup oom event source: %m"); | |
3022 | } | |
3023 | ||
d9e45bc3 MS |
3024 | static int unit_check_cgroup_events(Unit *u) { |
3025 | char *values[2] = {}; | |
3026 | int r; | |
3027 | ||
3028 | assert(u); | |
3029 | ||
3030 | r = cg_get_keyed_attribute_graceful(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path, "cgroup.events", | |
3031 | STRV_MAKE("populated", "frozen"), values); | |
3032 | if (r < 0) | |
3033 | return r; | |
3034 | ||
3035 | /* The cgroup.events notifications can be merged together so act as we saw the given state for the | |
3036 | * first time. The functions we call to handle given state are idempotent, which makes them | |
3037 | * effectively remember the previous state. */ | |
3038 | if (values[0]) { | |
3039 | if (streq(values[0], "1")) | |
3040 | unit_remove_from_cgroup_empty_queue(u); | |
3041 | else | |
3042 | unit_add_to_cgroup_empty_queue(u); | |
3043 | } | |
3044 | ||
3045 | /* Disregard freezer state changes due to operations not initiated by us */ | |
3046 | if (values[1] && IN_SET(u->freezer_state, FREEZER_FREEZING, FREEZER_THAWING)) { | |
3047 | if (streq(values[1], "0")) | |
3048 | unit_thawed(u); | |
3049 | else | |
3050 | unit_frozen(u); | |
3051 | } | |
3052 | ||
3053 | free(values[0]); | |
3054 | free(values[1]); | |
3055 | ||
3056 | return 0; | |
3057 | } | |
3058 | ||
efdb0237 LP |
3059 | static int on_cgroup_inotify_event(sd_event_source *s, int fd, uint32_t revents, void *userdata) { |
3060 | Manager *m = userdata; | |
3061 | ||
3062 | assert(s); | |
3063 | assert(fd >= 0); | |
3064 | assert(m); | |
3065 | ||
3066 | for (;;) { | |
3067 | union inotify_event_buffer buffer; | |
3068 | struct inotify_event *e; | |
3069 | ssize_t l; | |
3070 | ||
3071 | l = read(fd, &buffer, sizeof(buffer)); | |
3072 | if (l < 0) { | |
47249640 | 3073 | if (IN_SET(errno, EINTR, EAGAIN)) |
efdb0237 LP |
3074 | return 0; |
3075 | ||
3076 | return log_error_errno(errno, "Failed to read control group inotify events: %m"); | |
3077 | } | |
3078 | ||
3079 | FOREACH_INOTIFY_EVENT(e, buffer, l) { | |
3080 | Unit *u; | |
3081 | ||
3082 | if (e->wd < 0) | |
3083 | /* Queue overflow has no watch descriptor */ | |
3084 | continue; | |
3085 | ||
3086 | if (e->mask & IN_IGNORED) | |
3087 | /* The watch was just removed */ | |
3088 | continue; | |
3089 | ||
afcfaa69 LP |
3090 | /* Note that inotify might deliver events for a watch even after it was removed, |
3091 | * because it was queued before the removal. Let's ignore this here safely. */ | |
3092 | ||
0bb814c2 | 3093 | u = hashmap_get(m->cgroup_control_inotify_wd_unit, INT_TO_PTR(e->wd)); |
afcfaa69 | 3094 | if (u) |
d9e45bc3 | 3095 | unit_check_cgroup_events(u); |
efdb0237 | 3096 | |
afcfaa69 LP |
3097 | u = hashmap_get(m->cgroup_memory_inotify_wd_unit, INT_TO_PTR(e->wd)); |
3098 | if (u) | |
3099 | unit_add_to_cgroup_oom_queue(u); | |
efdb0237 LP |
3100 | } |
3101 | } | |
8e274523 LP |
3102 | } |
3103 | ||
17f14955 RG |
3104 | static int cg_bpf_mask_supported(CGroupMask *ret) { |
3105 | CGroupMask mask = 0; | |
3106 | int r; | |
3107 | ||
3108 | /* BPF-based firewall */ | |
3109 | r = bpf_firewall_supported(); | |
3110 | if (r > 0) | |
3111 | mask |= CGROUP_MASK_BPF_FIREWALL; | |
3112 | ||
084c7007 RG |
3113 | /* BPF-based device access control */ |
3114 | r = bpf_devices_supported(); | |
3115 | if (r > 0) | |
3116 | mask |= CGROUP_MASK_BPF_DEVICES; | |
3117 | ||
506ea51b JK |
3118 | /* BPF pinned prog */ |
3119 | r = bpf_foreign_supported(); | |
3120 | if (r > 0) | |
3121 | mask |= CGROUP_MASK_BPF_FOREIGN; | |
3122 | ||
a8e5eb17 JK |
3123 | /* BPF-based bind{4|6} hooks */ |
3124 | r = socket_bind_supported(); | |
3125 | if (r > 0) | |
3126 | mask |= CGROUP_MASK_BPF_SOCKET_BIND; | |
3127 | ||
17f14955 RG |
3128 | *ret = mask; |
3129 | return 0; | |
3130 | } | |
3131 | ||
8e274523 | 3132 | int manager_setup_cgroup(Manager *m) { |
9444b1f2 | 3133 | _cleanup_free_ char *path = NULL; |
10bd3e2e | 3134 | const char *scope_path; |
efdb0237 | 3135 | CGroupController c; |
b4cccbc1 | 3136 | int r, all_unified; |
17f14955 | 3137 | CGroupMask mask; |
efdb0237 | 3138 | char *e; |
8e274523 LP |
3139 | |
3140 | assert(m); | |
3141 | ||
35d2e7ec | 3142 | /* 1. Determine hierarchy */ |
efdb0237 | 3143 | m->cgroup_root = mfree(m->cgroup_root); |
9444b1f2 | 3144 | r = cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER, 0, &m->cgroup_root); |
23bbb0de MS |
3145 | if (r < 0) |
3146 | return log_error_errno(r, "Cannot determine cgroup we are running in: %m"); | |
8e274523 | 3147 | |
efdb0237 LP |
3148 | /* Chop off the init scope, if we are already located in it */ |
3149 | e = endswith(m->cgroup_root, "/" SPECIAL_INIT_SCOPE); | |
0d8c31ff | 3150 | |
efdb0237 LP |
3151 | /* LEGACY: Also chop off the system slice if we are in |
3152 | * it. This is to support live upgrades from older systemd | |
3153 | * versions where PID 1 was moved there. Also see | |
3154 | * cg_get_root_path(). */ | |
463d0d15 | 3155 | if (!e && MANAGER_IS_SYSTEM(m)) { |
9444b1f2 | 3156 | e = endswith(m->cgroup_root, "/" SPECIAL_SYSTEM_SLICE); |
15c60e99 | 3157 | if (!e) |
efdb0237 | 3158 | e = endswith(m->cgroup_root, "/system"); /* even more legacy */ |
0baf24dd | 3159 | } |
efdb0237 LP |
3160 | if (e) |
3161 | *e = 0; | |
7ccfb64a | 3162 | |
7546145e LP |
3163 | /* And make sure to store away the root value without trailing slash, even for the root dir, so that we can |
3164 | * easily prepend it everywhere. */ | |
3165 | delete_trailing_chars(m->cgroup_root, "/"); | |
8e274523 | 3166 | |
35d2e7ec | 3167 | /* 2. Show data */ |
9444b1f2 | 3168 | r = cg_get_path(SYSTEMD_CGROUP_CONTROLLER, m->cgroup_root, NULL, &path); |
23bbb0de MS |
3169 | if (r < 0) |
3170 | return log_error_errno(r, "Cannot find cgroup mount point: %m"); | |
8e274523 | 3171 | |
d4d99bc6 | 3172 | r = cg_unified(); |
415fc41c TH |
3173 | if (r < 0) |
3174 | return log_error_errno(r, "Couldn't determine if we are running in the unified hierarchy: %m"); | |
5da38d07 | 3175 | |
b4cccbc1 | 3176 | all_unified = cg_all_unified(); |
d4c819ed ZJS |
3177 | if (all_unified < 0) |
3178 | return log_error_errno(all_unified, "Couldn't determine whether we are in all unified mode: %m"); | |
3179 | if (all_unified > 0) | |
efdb0237 | 3180 | log_debug("Unified cgroup hierarchy is located at %s.", path); |
b4cccbc1 | 3181 | else { |
c22800e4 | 3182 | r = cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER); |
b4cccbc1 LP |
3183 | if (r < 0) |
3184 | return log_error_errno(r, "Failed to determine whether systemd's own controller is in unified mode: %m"); | |
3185 | if (r > 0) | |
3186 | log_debug("Unified cgroup hierarchy is located at %s. Controllers are on legacy hierarchies.", path); | |
3187 | else | |
3188 | log_debug("Using cgroup controller " SYSTEMD_CGROUP_CONTROLLER_LEGACY ". File system hierarchy is at %s.", path); | |
3189 | } | |
efdb0237 | 3190 | |
09e24654 LP |
3191 | /* 3. Allocate cgroup empty defer event source */ |
3192 | m->cgroup_empty_event_source = sd_event_source_unref(m->cgroup_empty_event_source); | |
3193 | r = sd_event_add_defer(m->event, &m->cgroup_empty_event_source, on_cgroup_empty_event, m); | |
3194 | if (r < 0) | |
3195 | return log_error_errno(r, "Failed to create cgroup empty event source: %m"); | |
3196 | ||
cbe83389 LP |
3197 | /* Schedule cgroup empty checks early, but after having processed service notification messages or |
3198 | * SIGCHLD signals, so that a cgroup running empty is always just the last safety net of | |
3199 | * notification, and we collected the metadata the notification and SIGCHLD stuff offers first. */ | |
09e24654 LP |
3200 | r = sd_event_source_set_priority(m->cgroup_empty_event_source, SD_EVENT_PRIORITY_NORMAL-5); |
3201 | if (r < 0) | |
3202 | return log_error_errno(r, "Failed to set priority of cgroup empty event source: %m"); | |
3203 | ||
3204 | r = sd_event_source_set_enabled(m->cgroup_empty_event_source, SD_EVENT_OFF); | |
3205 | if (r < 0) | |
3206 | return log_error_errno(r, "Failed to disable cgroup empty event source: %m"); | |
3207 | ||
3208 | (void) sd_event_source_set_description(m->cgroup_empty_event_source, "cgroup-empty"); | |
3209 | ||
3210 | /* 4. Install notifier inotify object, or agent */ | |
10bd3e2e | 3211 | if (cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER) > 0) { |
c6c18be3 | 3212 | |
09e24654 | 3213 | /* In the unified hierarchy we can get cgroup empty notifications via inotify. */ |
efdb0237 | 3214 | |
10bd3e2e LP |
3215 | m->cgroup_inotify_event_source = sd_event_source_unref(m->cgroup_inotify_event_source); |
3216 | safe_close(m->cgroup_inotify_fd); | |
efdb0237 | 3217 | |
10bd3e2e LP |
3218 | m->cgroup_inotify_fd = inotify_init1(IN_NONBLOCK|IN_CLOEXEC); |
3219 | if (m->cgroup_inotify_fd < 0) | |
3220 | return log_error_errno(errno, "Failed to create control group inotify object: %m"); | |
efdb0237 | 3221 | |
10bd3e2e LP |
3222 | r = sd_event_add_io(m->event, &m->cgroup_inotify_event_source, m->cgroup_inotify_fd, EPOLLIN, on_cgroup_inotify_event, m); |
3223 | if (r < 0) | |
3224 | return log_error_errno(r, "Failed to watch control group inotify object: %m"); | |
efdb0237 | 3225 | |
cbe83389 LP |
3226 | /* Process cgroup empty notifications early. Note that when this event is dispatched it'll |
3227 | * just add the unit to a cgroup empty queue, hence let's run earlier than that. Also see | |
3228 | * handling of cgroup agent notifications, for the classic cgroup hierarchy support. */ | |
3229 | r = sd_event_source_set_priority(m->cgroup_inotify_event_source, SD_EVENT_PRIORITY_NORMAL-9); | |
10bd3e2e LP |
3230 | if (r < 0) |
3231 | return log_error_errno(r, "Failed to set priority of inotify event source: %m"); | |
efdb0237 | 3232 | |
10bd3e2e | 3233 | (void) sd_event_source_set_description(m->cgroup_inotify_event_source, "cgroup-inotify"); |
efdb0237 | 3234 | |
611c4f8a | 3235 | } else if (MANAGER_IS_SYSTEM(m) && manager_owns_host_root_cgroup(m) && !MANAGER_IS_TEST_RUN(m)) { |
efdb0237 | 3236 | |
10bd3e2e LP |
3237 | /* On the legacy hierarchy we only get notifications via cgroup agents. (Which isn't really reliable, |
3238 | * since it does not generate events when control groups with children run empty. */ | |
8e274523 | 3239 | |
ce906769 | 3240 | r = cg_install_release_agent(SYSTEMD_CGROUP_CONTROLLER, SYSTEMD_CGROUPS_AGENT_PATH); |
23bbb0de | 3241 | if (r < 0) |
10bd3e2e LP |
3242 | log_warning_errno(r, "Failed to install release agent, ignoring: %m"); |
3243 | else if (r > 0) | |
3244 | log_debug("Installed release agent."); | |
3245 | else if (r == 0) | |
3246 | log_debug("Release agent already installed."); | |
3247 | } | |
efdb0237 | 3248 | |
09e24654 | 3249 | /* 5. Make sure we are in the special "init.scope" unit in the root slice. */ |
10bd3e2e LP |
3250 | scope_path = strjoina(m->cgroup_root, "/" SPECIAL_INIT_SCOPE); |
3251 | r = cg_create_and_attach(SYSTEMD_CGROUP_CONTROLLER, scope_path, 0); | |
aa77e234 MS |
3252 | if (r >= 0) { |
3253 | /* Also, move all other userspace processes remaining in the root cgroup into that scope. */ | |
3254 | r = cg_migrate(SYSTEMD_CGROUP_CONTROLLER, m->cgroup_root, SYSTEMD_CGROUP_CONTROLLER, scope_path, 0); | |
3255 | if (r < 0) | |
3256 | log_warning_errno(r, "Couldn't move remaining userspace processes, ignoring: %m"); | |
c6c18be3 | 3257 | |
aa77e234 MS |
3258 | /* 6. And pin it, so that it cannot be unmounted */ |
3259 | safe_close(m->pin_cgroupfs_fd); | |
3260 | m->pin_cgroupfs_fd = open(path, O_RDONLY|O_CLOEXEC|O_DIRECTORY|O_NOCTTY|O_NONBLOCK); | |
3261 | if (m->pin_cgroupfs_fd < 0) | |
3262 | return log_error_errno(errno, "Failed to open pin file: %m"); | |
0d8c31ff | 3263 | |
638cece4 | 3264 | } else if (!MANAGER_IS_TEST_RUN(m)) |
aa77e234 | 3265 | return log_error_errno(r, "Failed to create %s control group: %m", scope_path); |
10bd3e2e | 3266 | |
09e24654 | 3267 | /* 7. Always enable hierarchical support if it exists... */ |
638cece4 | 3268 | if (!all_unified && !MANAGER_IS_TEST_RUN(m)) |
10bd3e2e | 3269 | (void) cg_set_attribute("memory", "/", "memory.use_hierarchy", "1"); |
c6c18be3 | 3270 | |
17f14955 | 3271 | /* 8. Figure out which controllers are supported */ |
0fa7b500 | 3272 | r = cg_mask_supported_subtree(m->cgroup_root, &m->cgroup_supported); |
efdb0237 LP |
3273 | if (r < 0) |
3274 | return log_error_errno(r, "Failed to determine supported controllers: %m"); | |
17f14955 RG |
3275 | |
3276 | /* 9. Figure out which bpf-based pseudo-controllers are supported */ | |
3277 | r = cg_bpf_mask_supported(&mask); | |
3278 | if (r < 0) | |
3279 | return log_error_errno(r, "Failed to determine supported bpf-based pseudo-controllers: %m"); | |
3280 | m->cgroup_supported |= mask; | |
3281 | ||
3282 | /* 10. Log which controllers are supported */ | |
efdb0237 | 3283 | for (c = 0; c < _CGROUP_CONTROLLER_MAX; c++) |
eee0a1e4 | 3284 | log_debug("Controller '%s' supported: %s", cgroup_controller_to_string(c), yes_no(m->cgroup_supported & CGROUP_CONTROLLER_TO_MASK(c))); |
9156e799 | 3285 | |
a32360f1 | 3286 | return 0; |
8e274523 LP |
3287 | } |
3288 | ||
c6c18be3 | 3289 | void manager_shutdown_cgroup(Manager *m, bool delete) { |
8e274523 LP |
3290 | assert(m); |
3291 | ||
9444b1f2 LP |
3292 | /* We can't really delete the group, since we are in it. But |
3293 | * let's trim it. */ | |
f6c63f6f | 3294 | if (delete && m->cgroup_root && m->test_run_flags != MANAGER_TEST_RUN_MINIMAL) |
efdb0237 LP |
3295 | (void) cg_trim(SYSTEMD_CGROUP_CONTROLLER, m->cgroup_root, false); |
3296 | ||
09e24654 LP |
3297 | m->cgroup_empty_event_source = sd_event_source_unref(m->cgroup_empty_event_source); |
3298 | ||
0bb814c2 | 3299 | m->cgroup_control_inotify_wd_unit = hashmap_free(m->cgroup_control_inotify_wd_unit); |
afcfaa69 | 3300 | m->cgroup_memory_inotify_wd_unit = hashmap_free(m->cgroup_memory_inotify_wd_unit); |
efdb0237 LP |
3301 | |
3302 | m->cgroup_inotify_event_source = sd_event_source_unref(m->cgroup_inotify_event_source); | |
3303 | m->cgroup_inotify_fd = safe_close(m->cgroup_inotify_fd); | |
8e274523 | 3304 | |
03e334a1 | 3305 | m->pin_cgroupfs_fd = safe_close(m->pin_cgroupfs_fd); |
c6c18be3 | 3306 | |
efdb0237 | 3307 | m->cgroup_root = mfree(m->cgroup_root); |
8e274523 LP |
3308 | } |
3309 | ||
4ad49000 | 3310 | Unit* manager_get_unit_by_cgroup(Manager *m, const char *cgroup) { |
acb14d31 | 3311 | char *p; |
4ad49000 | 3312 | Unit *u; |
acb14d31 LP |
3313 | |
3314 | assert(m); | |
3315 | assert(cgroup); | |
acb14d31 | 3316 | |
4ad49000 LP |
3317 | u = hashmap_get(m->cgroup_unit, cgroup); |
3318 | if (u) | |
3319 | return u; | |
acb14d31 | 3320 | |
8e70580b | 3321 | p = strdupa(cgroup); |
acb14d31 LP |
3322 | for (;;) { |
3323 | char *e; | |
3324 | ||
3325 | e = strrchr(p, '/'); | |
efdb0237 LP |
3326 | if (!e || e == p) |
3327 | return hashmap_get(m->cgroup_unit, SPECIAL_ROOT_SLICE); | |
acb14d31 LP |
3328 | |
3329 | *e = 0; | |
3330 | ||
4ad49000 LP |
3331 | u = hashmap_get(m->cgroup_unit, p); |
3332 | if (u) | |
3333 | return u; | |
acb14d31 LP |
3334 | } |
3335 | } | |
3336 | ||
b3ac818b | 3337 | Unit *manager_get_unit_by_pid_cgroup(Manager *m, pid_t pid) { |
4ad49000 | 3338 | _cleanup_free_ char *cgroup = NULL; |
8e274523 | 3339 | |
8c47c732 LP |
3340 | assert(m); |
3341 | ||
62a76913 | 3342 | if (!pid_is_valid(pid)) |
b3ac818b LP |
3343 | return NULL; |
3344 | ||
62a76913 | 3345 | if (cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER, pid, &cgroup) < 0) |
b3ac818b LP |
3346 | return NULL; |
3347 | ||
3348 | return manager_get_unit_by_cgroup(m, cgroup); | |
3349 | } | |
3350 | ||
3351 | Unit *manager_get_unit_by_pid(Manager *m, pid_t pid) { | |
62a76913 | 3352 | Unit *u, **array; |
b3ac818b LP |
3353 | |
3354 | assert(m); | |
3355 | ||
62a76913 LP |
3356 | /* Note that a process might be owned by multiple units, we return only one here, which is good enough for most |
3357 | * cases, though not strictly correct. We prefer the one reported by cgroup membership, as that's the most | |
3358 | * relevant one as children of the process will be assigned to that one, too, before all else. */ | |
3359 | ||
3360 | if (!pid_is_valid(pid)) | |
8c47c732 LP |
3361 | return NULL; |
3362 | ||
2ca9d979 | 3363 | if (pid == getpid_cached()) |
efdb0237 LP |
3364 | return hashmap_get(m->units, SPECIAL_INIT_SCOPE); |
3365 | ||
62a76913 | 3366 | u = manager_get_unit_by_pid_cgroup(m, pid); |
5fe8876b LP |
3367 | if (u) |
3368 | return u; | |
3369 | ||
62a76913 | 3370 | u = hashmap_get(m->watch_pids, PID_TO_PTR(pid)); |
5fe8876b LP |
3371 | if (u) |
3372 | return u; | |
3373 | ||
62a76913 LP |
3374 | array = hashmap_get(m->watch_pids, PID_TO_PTR(-pid)); |
3375 | if (array) | |
3376 | return array[0]; | |
3377 | ||
3378 | return NULL; | |
6dde1f33 | 3379 | } |
4fbf50b3 | 3380 | |
4ad49000 LP |
3381 | int manager_notify_cgroup_empty(Manager *m, const char *cgroup) { |
3382 | Unit *u; | |
4fbf50b3 | 3383 | |
4ad49000 LP |
3384 | assert(m); |
3385 | assert(cgroup); | |
4fbf50b3 | 3386 | |
09e24654 LP |
3387 | /* Called on the legacy hierarchy whenever we get an explicit cgroup notification from the cgroup agent process |
3388 | * or from the --system instance */ | |
3389 | ||
d8fdc620 LP |
3390 | log_debug("Got cgroup empty notification for: %s", cgroup); |
3391 | ||
4ad49000 | 3392 | u = manager_get_unit_by_cgroup(m, cgroup); |
5ad096b3 LP |
3393 | if (!u) |
3394 | return 0; | |
b56c28c3 | 3395 | |
09e24654 LP |
3396 | unit_add_to_cgroup_empty_queue(u); |
3397 | return 1; | |
5ad096b3 LP |
3398 | } |
3399 | ||
3400 | int unit_get_memory_current(Unit *u, uint64_t *ret) { | |
5ad096b3 LP |
3401 | int r; |
3402 | ||
3403 | assert(u); | |
3404 | assert(ret); | |
3405 | ||
2e4025c0 | 3406 | if (!UNIT_CGROUP_BOOL(u, memory_accounting)) |
cf3b4be1 LP |
3407 | return -ENODATA; |
3408 | ||
5ad096b3 LP |
3409 | if (!u->cgroup_path) |
3410 | return -ENODATA; | |
3411 | ||
1f73aa00 | 3412 | /* The root cgroup doesn't expose this information, let's get it from /proc instead */ |
611c4f8a | 3413 | if (unit_has_host_root_cgroup(u)) |
c482724a | 3414 | return procfs_memory_get_used(ret); |
1f73aa00 | 3415 | |
efdb0237 | 3416 | if ((u->cgroup_realized_mask & CGROUP_MASK_MEMORY) == 0) |
5ad096b3 LP |
3417 | return -ENODATA; |
3418 | ||
b4cccbc1 LP |
3419 | r = cg_all_unified(); |
3420 | if (r < 0) | |
3421 | return r; | |
5ad096b3 | 3422 | |
613328c3 | 3423 | return cg_get_attribute_as_uint64("memory", u->cgroup_path, r > 0 ? "memory.current" : "memory.usage_in_bytes", ret); |
5ad096b3 LP |
3424 | } |
3425 | ||
03a7b521 | 3426 | int unit_get_tasks_current(Unit *u, uint64_t *ret) { |
03a7b521 LP |
3427 | assert(u); |
3428 | assert(ret); | |
3429 | ||
2e4025c0 | 3430 | if (!UNIT_CGROUP_BOOL(u, tasks_accounting)) |
cf3b4be1 LP |
3431 | return -ENODATA; |
3432 | ||
03a7b521 LP |
3433 | if (!u->cgroup_path) |
3434 | return -ENODATA; | |
3435 | ||
c36a69f4 | 3436 | /* The root cgroup doesn't expose this information, let's get it from /proc instead */ |
611c4f8a | 3437 | if (unit_has_host_root_cgroup(u)) |
c36a69f4 LP |
3438 | return procfs_tasks_get_current(ret); |
3439 | ||
1f73aa00 LP |
3440 | if ((u->cgroup_realized_mask & CGROUP_MASK_PIDS) == 0) |
3441 | return -ENODATA; | |
3442 | ||
613328c3 | 3443 | return cg_get_attribute_as_uint64("pids", u->cgroup_path, "pids.current", ret); |
03a7b521 LP |
3444 | } |
3445 | ||
5ad096b3 | 3446 | static int unit_get_cpu_usage_raw(Unit *u, nsec_t *ret) { |
5ad096b3 LP |
3447 | uint64_t ns; |
3448 | int r; | |
3449 | ||
3450 | assert(u); | |
3451 | assert(ret); | |
3452 | ||
3453 | if (!u->cgroup_path) | |
3454 | return -ENODATA; | |
3455 | ||
1f73aa00 | 3456 | /* The root cgroup doesn't expose this information, let's get it from /proc instead */ |
611c4f8a | 3457 | if (unit_has_host_root_cgroup(u)) |
1f73aa00 LP |
3458 | return procfs_cpu_get_usage(ret); |
3459 | ||
f98c2585 CD |
3460 | /* Requisite controllers for CPU accounting are not enabled */ |
3461 | if ((get_cpu_accounting_mask() & ~u->cgroup_realized_mask) != 0) | |
3462 | return -ENODATA; | |
3463 | ||
92a99304 LP |
3464 | r = cg_all_unified(); |
3465 | if (r < 0) | |
3466 | return r; | |
b4cccbc1 | 3467 | if (r > 0) { |
66ebf6c0 TH |
3468 | _cleanup_free_ char *val = NULL; |
3469 | uint64_t us; | |
5ad096b3 | 3470 | |
b734a4ff | 3471 | r = cg_get_keyed_attribute("cpu", u->cgroup_path, "cpu.stat", STRV_MAKE("usage_usec"), &val); |
b734a4ff LP |
3472 | if (IN_SET(r, -ENOENT, -ENXIO)) |
3473 | return -ENODATA; | |
d742f4b5 LP |
3474 | if (r < 0) |
3475 | return r; | |
66ebf6c0 TH |
3476 | |
3477 | r = safe_atou64(val, &us); | |
3478 | if (r < 0) | |
3479 | return r; | |
3480 | ||
3481 | ns = us * NSEC_PER_USEC; | |
613328c3 AZ |
3482 | } else |
3483 | return cg_get_attribute_as_uint64("cpuacct", u->cgroup_path, "cpuacct.usage", ret); | |
5ad096b3 LP |
3484 | |
3485 | *ret = ns; | |
3486 | return 0; | |
3487 | } | |
3488 | ||
3489 | int unit_get_cpu_usage(Unit *u, nsec_t *ret) { | |
3490 | nsec_t ns; | |
3491 | int r; | |
3492 | ||
fe700f46 LP |
3493 | assert(u); |
3494 | ||
3495 | /* Retrieve the current CPU usage counter. This will subtract the CPU counter taken when the unit was | |
3496 | * started. If the cgroup has been removed already, returns the last cached value. To cache the value, simply | |
3497 | * call this function with a NULL return value. */ | |
3498 | ||
2e4025c0 | 3499 | if (!UNIT_CGROUP_BOOL(u, cpu_accounting)) |
cf3b4be1 LP |
3500 | return -ENODATA; |
3501 | ||
5ad096b3 | 3502 | r = unit_get_cpu_usage_raw(u, &ns); |
fe700f46 LP |
3503 | if (r == -ENODATA && u->cpu_usage_last != NSEC_INFINITY) { |
3504 | /* If we can't get the CPU usage anymore (because the cgroup was already removed, for example), use our | |
3505 | * cached value. */ | |
3506 | ||
3507 | if (ret) | |
3508 | *ret = u->cpu_usage_last; | |
3509 | return 0; | |
3510 | } | |
5ad096b3 LP |
3511 | if (r < 0) |
3512 | return r; | |
3513 | ||
66ebf6c0 TH |
3514 | if (ns > u->cpu_usage_base) |
3515 | ns -= u->cpu_usage_base; | |
5ad096b3 LP |
3516 | else |
3517 | ns = 0; | |
3518 | ||
fe700f46 LP |
3519 | u->cpu_usage_last = ns; |
3520 | if (ret) | |
3521 | *ret = ns; | |
3522 | ||
5ad096b3 LP |
3523 | return 0; |
3524 | } | |
3525 | ||
906c06f6 DM |
3526 | int unit_get_ip_accounting( |
3527 | Unit *u, | |
3528 | CGroupIPAccountingMetric metric, | |
3529 | uint64_t *ret) { | |
3530 | ||
6b659ed8 | 3531 | uint64_t value; |
906c06f6 DM |
3532 | int fd, r; |
3533 | ||
3534 | assert(u); | |
3535 | assert(metric >= 0); | |
3536 | assert(metric < _CGROUP_IP_ACCOUNTING_METRIC_MAX); | |
3537 | assert(ret); | |
3538 | ||
2e4025c0 | 3539 | if (!UNIT_CGROUP_BOOL(u, ip_accounting)) |
cf3b4be1 LP |
3540 | return -ENODATA; |
3541 | ||
906c06f6 DM |
3542 | fd = IN_SET(metric, CGROUP_IP_INGRESS_BYTES, CGROUP_IP_INGRESS_PACKETS) ? |
3543 | u->ip_accounting_ingress_map_fd : | |
3544 | u->ip_accounting_egress_map_fd; | |
906c06f6 DM |
3545 | if (fd < 0) |
3546 | return -ENODATA; | |
3547 | ||
3548 | if (IN_SET(metric, CGROUP_IP_INGRESS_BYTES, CGROUP_IP_EGRESS_BYTES)) | |
6b659ed8 | 3549 | r = bpf_firewall_read_accounting(fd, &value, NULL); |
906c06f6 | 3550 | else |
6b659ed8 LP |
3551 | r = bpf_firewall_read_accounting(fd, NULL, &value); |
3552 | if (r < 0) | |
3553 | return r; | |
3554 | ||
3555 | /* Add in additional metrics from a previous runtime. Note that when reexecing/reloading the daemon we compile | |
3556 | * all BPF programs and maps anew, but serialize the old counters. When deserializing we store them in the | |
3557 | * ip_accounting_extra[] field, and add them in here transparently. */ | |
3558 | ||
3559 | *ret = value + u->ip_accounting_extra[metric]; | |
906c06f6 DM |
3560 | |
3561 | return r; | |
3562 | } | |
3563 | ||
fbe14fc9 LP |
3564 | static int unit_get_io_accounting_raw(Unit *u, uint64_t ret[static _CGROUP_IO_ACCOUNTING_METRIC_MAX]) { |
3565 | static const char *const field_names[_CGROUP_IO_ACCOUNTING_METRIC_MAX] = { | |
3566 | [CGROUP_IO_READ_BYTES] = "rbytes=", | |
3567 | [CGROUP_IO_WRITE_BYTES] = "wbytes=", | |
3568 | [CGROUP_IO_READ_OPERATIONS] = "rios=", | |
3569 | [CGROUP_IO_WRITE_OPERATIONS] = "wios=", | |
3570 | }; | |
3571 | uint64_t acc[_CGROUP_IO_ACCOUNTING_METRIC_MAX] = {}; | |
3572 | _cleanup_free_ char *path = NULL; | |
3573 | _cleanup_fclose_ FILE *f = NULL; | |
3574 | int r; | |
3575 | ||
3576 | assert(u); | |
3577 | ||
3578 | if (!u->cgroup_path) | |
3579 | return -ENODATA; | |
3580 | ||
3581 | if (unit_has_host_root_cgroup(u)) | |
3582 | return -ENODATA; /* TODO: return useful data for the top-level cgroup */ | |
3583 | ||
3584 | r = cg_all_unified(); | |
3585 | if (r < 0) | |
3586 | return r; | |
3587 | if (r == 0) /* TODO: support cgroupv1 */ | |
3588 | return -ENODATA; | |
3589 | ||
3590 | if (!FLAGS_SET(u->cgroup_realized_mask, CGROUP_MASK_IO)) | |
3591 | return -ENODATA; | |
3592 | ||
3593 | r = cg_get_path("io", u->cgroup_path, "io.stat", &path); | |
3594 | if (r < 0) | |
3595 | return r; | |
3596 | ||
3597 | f = fopen(path, "re"); | |
3598 | if (!f) | |
3599 | return -errno; | |
3600 | ||
3601 | for (;;) { | |
3602 | _cleanup_free_ char *line = NULL; | |
3603 | const char *p; | |
3604 | ||
3605 | r = read_line(f, LONG_LINE_MAX, &line); | |
3606 | if (r < 0) | |
3607 | return r; | |
3608 | if (r == 0) | |
3609 | break; | |
3610 | ||
3611 | p = line; | |
3612 | p += strcspn(p, WHITESPACE); /* Skip over device major/minor */ | |
3613 | p += strspn(p, WHITESPACE); /* Skip over following whitespace */ | |
3614 | ||
3615 | for (;;) { | |
3616 | _cleanup_free_ char *word = NULL; | |
3617 | ||
3618 | r = extract_first_word(&p, &word, NULL, EXTRACT_RETAIN_ESCAPE); | |
3619 | if (r < 0) | |
3620 | return r; | |
3621 | if (r == 0) | |
3622 | break; | |
3623 | ||
3624 | for (CGroupIOAccountingMetric i = 0; i < _CGROUP_IO_ACCOUNTING_METRIC_MAX; i++) { | |
3625 | const char *x; | |
3626 | ||
3627 | x = startswith(word, field_names[i]); | |
3628 | if (x) { | |
3629 | uint64_t w; | |
3630 | ||
3631 | r = safe_atou64(x, &w); | |
3632 | if (r < 0) | |
3633 | return r; | |
3634 | ||
3635 | /* Sum up the stats of all devices */ | |
3636 | acc[i] += w; | |
3637 | break; | |
3638 | } | |
3639 | } | |
3640 | } | |
3641 | } | |
3642 | ||
3643 | memcpy(ret, acc, sizeof(acc)); | |
3644 | return 0; | |
3645 | } | |
3646 | ||
3647 | int unit_get_io_accounting( | |
3648 | Unit *u, | |
3649 | CGroupIOAccountingMetric metric, | |
3650 | bool allow_cache, | |
3651 | uint64_t *ret) { | |
3652 | ||
3653 | uint64_t raw[_CGROUP_IO_ACCOUNTING_METRIC_MAX]; | |
3654 | int r; | |
3655 | ||
3656 | /* Retrieve an IO account parameter. This will subtract the counter when the unit was started. */ | |
3657 | ||
3658 | if (!UNIT_CGROUP_BOOL(u, io_accounting)) | |
3659 | return -ENODATA; | |
3660 | ||
3661 | if (allow_cache && u->io_accounting_last[metric] != UINT64_MAX) | |
3662 | goto done; | |
3663 | ||
3664 | r = unit_get_io_accounting_raw(u, raw); | |
3665 | if (r == -ENODATA && u->io_accounting_last[metric] != UINT64_MAX) | |
3666 | goto done; | |
3667 | if (r < 0) | |
3668 | return r; | |
3669 | ||
3670 | for (CGroupIOAccountingMetric i = 0; i < _CGROUP_IO_ACCOUNTING_METRIC_MAX; i++) { | |
3671 | /* Saturated subtraction */ | |
3672 | if (raw[i] > u->io_accounting_base[i]) | |
3673 | u->io_accounting_last[i] = raw[i] - u->io_accounting_base[i]; | |
3674 | else | |
3675 | u->io_accounting_last[i] = 0; | |
3676 | } | |
3677 | ||
3678 | done: | |
3679 | if (ret) | |
3680 | *ret = u->io_accounting_last[metric]; | |
3681 | ||
3682 | return 0; | |
3683 | } | |
3684 | ||
906c06f6 | 3685 | int unit_reset_cpu_accounting(Unit *u) { |
5ad096b3 LP |
3686 | int r; |
3687 | ||
3688 | assert(u); | |
3689 | ||
fe700f46 LP |
3690 | u->cpu_usage_last = NSEC_INFINITY; |
3691 | ||
0bbff7d6 | 3692 | r = unit_get_cpu_usage_raw(u, &u->cpu_usage_base); |
5ad096b3 | 3693 | if (r < 0) { |
66ebf6c0 | 3694 | u->cpu_usage_base = 0; |
5ad096b3 | 3695 | return r; |
b56c28c3 | 3696 | } |
2633eb83 | 3697 | |
4ad49000 | 3698 | return 0; |
4fbf50b3 LP |
3699 | } |
3700 | ||
906c06f6 DM |
3701 | int unit_reset_ip_accounting(Unit *u) { |
3702 | int r = 0, q = 0; | |
3703 | ||
3704 | assert(u); | |
3705 | ||
3706 | if (u->ip_accounting_ingress_map_fd >= 0) | |
3707 | r = bpf_firewall_reset_accounting(u->ip_accounting_ingress_map_fd); | |
3708 | ||
3709 | if (u->ip_accounting_egress_map_fd >= 0) | |
3710 | q = bpf_firewall_reset_accounting(u->ip_accounting_egress_map_fd); | |
3711 | ||
6b659ed8 LP |
3712 | zero(u->ip_accounting_extra); |
3713 | ||
906c06f6 DM |
3714 | return r < 0 ? r : q; |
3715 | } | |
3716 | ||
fbe14fc9 LP |
3717 | int unit_reset_io_accounting(Unit *u) { |
3718 | int r; | |
3719 | ||
3720 | assert(u); | |
3721 | ||
3722 | for (CGroupIOAccountingMetric i = 0; i < _CGROUP_IO_ACCOUNTING_METRIC_MAX; i++) | |
3723 | u->io_accounting_last[i] = UINT64_MAX; | |
3724 | ||
3725 | r = unit_get_io_accounting_raw(u, u->io_accounting_base); | |
3726 | if (r < 0) { | |
3727 | zero(u->io_accounting_base); | |
3728 | return r; | |
3729 | } | |
3730 | ||
3731 | return 0; | |
3732 | } | |
3733 | ||
9b2559a1 | 3734 | int unit_reset_accounting(Unit *u) { |
fbe14fc9 | 3735 | int r, q, v; |
9b2559a1 LP |
3736 | |
3737 | assert(u); | |
3738 | ||
3739 | r = unit_reset_cpu_accounting(u); | |
fbe14fc9 LP |
3740 | q = unit_reset_io_accounting(u); |
3741 | v = unit_reset_ip_accounting(u); | |
9b2559a1 | 3742 | |
fbe14fc9 | 3743 | return r < 0 ? r : q < 0 ? q : v; |
9b2559a1 LP |
3744 | } |
3745 | ||
e7ab4d1a LP |
3746 | void unit_invalidate_cgroup(Unit *u, CGroupMask m) { |
3747 | assert(u); | |
3748 | ||
3749 | if (!UNIT_HAS_CGROUP_CONTEXT(u)) | |
3750 | return; | |
3751 | ||
3752 | if (m == 0) | |
3753 | return; | |
3754 | ||
538b4852 TH |
3755 | /* always invalidate compat pairs together */ |
3756 | if (m & (CGROUP_MASK_IO | CGROUP_MASK_BLKIO)) | |
3757 | m |= CGROUP_MASK_IO | CGROUP_MASK_BLKIO; | |
3758 | ||
7cce4fb7 LP |
3759 | if (m & (CGROUP_MASK_CPU | CGROUP_MASK_CPUACCT)) |
3760 | m |= CGROUP_MASK_CPU | CGROUP_MASK_CPUACCT; | |
3761 | ||
e00068e7 | 3762 | if (FLAGS_SET(u->cgroup_invalidated_mask, m)) /* NOP? */ |
e7ab4d1a LP |
3763 | return; |
3764 | ||
e00068e7 | 3765 | u->cgroup_invalidated_mask |= m; |
91a6073e | 3766 | unit_add_to_cgroup_realize_queue(u); |
e7ab4d1a LP |
3767 | } |
3768 | ||
906c06f6 DM |
3769 | void unit_invalidate_cgroup_bpf(Unit *u) { |
3770 | assert(u); | |
3771 | ||
3772 | if (!UNIT_HAS_CGROUP_CONTEXT(u)) | |
3773 | return; | |
3774 | ||
17f14955 | 3775 | if (u->cgroup_invalidated_mask & CGROUP_MASK_BPF_FIREWALL) /* NOP? */ |
906c06f6 DM |
3776 | return; |
3777 | ||
17f14955 | 3778 | u->cgroup_invalidated_mask |= CGROUP_MASK_BPF_FIREWALL; |
91a6073e | 3779 | unit_add_to_cgroup_realize_queue(u); |
906c06f6 DM |
3780 | |
3781 | /* If we are a slice unit, we also need to put compile a new BPF program for all our children, as the IP access | |
3782 | * list of our children includes our own. */ | |
3783 | if (u->type == UNIT_SLICE) { | |
3784 | Unit *member; | |
eef85c4a | 3785 | void *v; |
906c06f6 | 3786 | |
90e74a66 | 3787 | HASHMAP_FOREACH_KEY(v, member, u->dependencies[UNIT_BEFORE]) |
cb5e3bc3 CD |
3788 | if (UNIT_DEREF(member->slice) == u) |
3789 | unit_invalidate_cgroup_bpf(member); | |
906c06f6 DM |
3790 | } |
3791 | } | |
3792 | ||
1d9cc876 LP |
3793 | bool unit_cgroup_delegate(Unit *u) { |
3794 | CGroupContext *c; | |
3795 | ||
3796 | assert(u); | |
3797 | ||
3798 | if (!UNIT_VTABLE(u)->can_delegate) | |
3799 | return false; | |
3800 | ||
3801 | c = unit_get_cgroup_context(u); | |
3802 | if (!c) | |
3803 | return false; | |
3804 | ||
3805 | return c->delegate; | |
3806 | } | |
3807 | ||
e7ab4d1a | 3808 | void manager_invalidate_startup_units(Manager *m) { |
e7ab4d1a LP |
3809 | Unit *u; |
3810 | ||
3811 | assert(m); | |
3812 | ||
90e74a66 | 3813 | SET_FOREACH(u, m->startup_units) |
13c31542 | 3814 | unit_invalidate_cgroup(u, CGROUP_MASK_CPU|CGROUP_MASK_IO|CGROUP_MASK_BLKIO); |
e7ab4d1a LP |
3815 | } |
3816 | ||
da8e1782 MO |
3817 | static int unit_get_nice(Unit *u) { |
3818 | ExecContext *ec; | |
3819 | ||
3820 | ec = unit_get_exec_context(u); | |
3821 | return ec ? ec->nice : 0; | |
3822 | } | |
3823 | ||
3824 | static uint64_t unit_get_cpu_weight(Unit *u) { | |
3825 | ManagerState state = manager_state(u->manager); | |
3826 | CGroupContext *cc; | |
3827 | ||
3828 | cc = unit_get_cgroup_context(u); | |
3829 | return cc ? cgroup_context_cpu_weight(cc, state) : CGROUP_WEIGHT_DEFAULT; | |
3830 | } | |
3831 | ||
3832 | int compare_job_priority(const void *a, const void *b) { | |
3833 | const Job *x = a, *y = b; | |
3834 | int nice_x, nice_y; | |
3835 | uint64_t weight_x, weight_y; | |
3836 | int ret; | |
3837 | ||
217b7b33 ZJS |
3838 | if ((ret = CMP(x->unit->type, y->unit->type)) != 0) |
3839 | return -ret; | |
3840 | ||
da8e1782 MO |
3841 | weight_x = unit_get_cpu_weight(x->unit); |
3842 | weight_y = unit_get_cpu_weight(y->unit); | |
3843 | ||
217b7b33 ZJS |
3844 | if ((ret = CMP(weight_x, weight_y)) != 0) |
3845 | return -ret; | |
da8e1782 MO |
3846 | |
3847 | nice_x = unit_get_nice(x->unit); | |
3848 | nice_y = unit_get_nice(y->unit); | |
3849 | ||
3850 | if ((ret = CMP(nice_x, nice_y)) != 0) | |
3851 | return ret; | |
3852 | ||
da8e1782 MO |
3853 | return strcmp(x->unit->id, y->unit->id); |
3854 | } | |
3855 | ||
d9e45bc3 MS |
3856 | int unit_cgroup_freezer_action(Unit *u, FreezerAction action) { |
3857 | _cleanup_free_ char *path = NULL; | |
3858 | FreezerState target, kernel = _FREEZER_STATE_INVALID; | |
3859 | int r; | |
3860 | ||
3861 | assert(u); | |
3862 | assert(IN_SET(action, FREEZER_FREEZE, FREEZER_THAW)); | |
3863 | ||
9a1e90ae MS |
3864 | if (!cg_freezer_supported()) |
3865 | return 0; | |
3866 | ||
d9e45bc3 MS |
3867 | if (!u->cgroup_realized) |
3868 | return -EBUSY; | |
3869 | ||
3870 | target = action == FREEZER_FREEZE ? FREEZER_FROZEN : FREEZER_RUNNING; | |
3871 | ||
3872 | r = unit_freezer_state_kernel(u, &kernel); | |
3873 | if (r < 0) | |
3874 | log_unit_debug_errno(u, r, "Failed to obtain cgroup freezer state: %m"); | |
3875 | ||
3876 | if (target == kernel) { | |
3877 | u->freezer_state = target; | |
3878 | return 0; | |
3879 | } | |
3880 | ||
3881 | r = cg_get_path(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path, "cgroup.freeze", &path); | |
3882 | if (r < 0) | |
3883 | return r; | |
3884 | ||
3885 | log_unit_debug(u, "%s unit.", action == FREEZER_FREEZE ? "Freezing" : "Thawing"); | |
3886 | ||
3887 | if (action == FREEZER_FREEZE) | |
3888 | u->freezer_state = FREEZER_FREEZING; | |
3889 | else | |
3890 | u->freezer_state = FREEZER_THAWING; | |
3891 | ||
3892 | r = write_string_file(path, one_zero(action == FREEZER_FREEZE), WRITE_STRING_FILE_DISABLE_BUFFER); | |
3893 | if (r < 0) | |
3894 | return r; | |
3895 | ||
d910f4c2 | 3896 | return 1; |
d9e45bc3 MS |
3897 | } |
3898 | ||
047f5d63 PH |
3899 | int unit_get_cpuset(Unit *u, CPUSet *cpus, const char *name) { |
3900 | _cleanup_free_ char *v = NULL; | |
3901 | int r; | |
3902 | ||
3903 | assert(u); | |
3904 | assert(cpus); | |
3905 | ||
3906 | if (!u->cgroup_path) | |
3907 | return -ENODATA; | |
3908 | ||
3909 | if ((u->cgroup_realized_mask & CGROUP_MASK_CPUSET) == 0) | |
3910 | return -ENODATA; | |
3911 | ||
3912 | r = cg_all_unified(); | |
3913 | if (r < 0) | |
3914 | return r; | |
3915 | if (r == 0) | |
3916 | return -ENODATA; | |
48fd01e5 LP |
3917 | |
3918 | r = cg_get_attribute("cpuset", u->cgroup_path, name, &v); | |
047f5d63 PH |
3919 | if (r == -ENOENT) |
3920 | return -ENODATA; | |
3921 | if (r < 0) | |
3922 | return r; | |
3923 | ||
3924 | return parse_cpu_set_full(v, cpus, false, NULL, NULL, 0, NULL); | |
3925 | } | |
3926 | ||
4e806bfa AZ |
3927 | static const char* const cgroup_device_policy_table[_CGROUP_DEVICE_POLICY_MAX] = { |
3928 | [CGROUP_DEVICE_POLICY_AUTO] = "auto", | |
3929 | [CGROUP_DEVICE_POLICY_CLOSED] = "closed", | |
3930 | [CGROUP_DEVICE_POLICY_STRICT] = "strict", | |
3931 | }; | |
3932 | ||
4ad49000 | 3933 | DEFINE_STRING_TABLE_LOOKUP(cgroup_device_policy, CGroupDevicePolicy); |
d9e45bc3 MS |
3934 | |
3935 | static const char* const freezer_action_table[_FREEZER_ACTION_MAX] = { | |
3936 | [FREEZER_FREEZE] = "freeze", | |
3937 | [FREEZER_THAW] = "thaw", | |
3938 | }; | |
3939 | ||
3940 | DEFINE_STRING_TABLE_LOOKUP(freezer_action, FreezerAction); |