]>
Commit | Line | Data |
---|---|---|
53e1b683 | 1 | /* SPDX-License-Identifier: LGPL-2.1+ */ |
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" |
45c2e068 | 11 | #include "btrfs-util.h" |
6592b975 | 12 | #include "bus-error.h" |
fdb3deca | 13 | #include "cgroup-setup.h" |
03a7b521 | 14 | #include "cgroup-util.h" |
3ffd4af2 LP |
15 | #include "cgroup.h" |
16 | #include "fd-util.h" | |
0d39fa9c | 17 | #include "fileio.h" |
77601719 | 18 | #include "fs-util.h" |
3a0f06c4 | 19 | #include "limits-util.h" |
6bedfcbb | 20 | #include "parse-util.h" |
9eb977db | 21 | #include "path-util.h" |
03a7b521 | 22 | #include "process-util.h" |
c36a69f4 | 23 | #include "procfs-util.h" |
9444b1f2 | 24 | #include "special.h" |
74c48bf5 | 25 | #include "stat-util.h" |
906c06f6 | 26 | #include "stdio-util.h" |
8b43440b | 27 | #include "string-table.h" |
07630cea | 28 | #include "string-util.h" |
cc6271f1 | 29 | #include "virt.h" |
8e274523 | 30 | |
10f28641 | 31 | #define CGROUP_CPU_QUOTA_DEFAULT_PERIOD_USEC ((usec_t) 100 * USEC_PER_MSEC) |
9a054909 | 32 | |
39b9fefb LP |
33 | /* Returns the log level to use when cgroup attribute writes fail. When an attribute is missing or we have access |
34 | * problems we downgrade to LOG_DEBUG. This is supposed to be nice to container managers and kernels which want to mask | |
35 | * out specific attributes from us. */ | |
36 | #define LOG_LEVEL_CGROUP_WRITE(r) (IN_SET(abs(r), ENOENT, EROFS, EACCES, EPERM) ? LOG_DEBUG : LOG_WARNING) | |
37 | ||
3a0f06c4 ZJS |
38 | uint64_t tasks_max_resolve(const TasksMax *tasks_max) { |
39 | if (tasks_max->scale == 0) | |
40 | return tasks_max->value; | |
41 | ||
42 | return system_tasks_max_scale(tasks_max->value, tasks_max->scale); | |
43 | } | |
44 | ||
611c4f8a | 45 | bool manager_owns_host_root_cgroup(Manager *m) { |
cc6271f1 LP |
46 | assert(m); |
47 | ||
48 | /* Returns true if we are managing the root cgroup. Note that it isn't sufficient to just check whether the | |
49 | * group root path equals "/" since that will also be the case if CLONE_NEWCGROUP is in the mix. Since there's | |
50 | * appears to be no nice way to detect whether we are in a CLONE_NEWCGROUP namespace we instead just check if | |
51 | * we run in any kind of container virtualization. */ | |
52 | ||
28cfdc5a LP |
53 | if (MANAGER_IS_USER(m)) |
54 | return false; | |
55 | ||
cc6271f1 LP |
56 | if (detect_container() > 0) |
57 | return false; | |
58 | ||
57ea45e1 | 59 | return empty_or_root(m->cgroup_root); |
cc6271f1 LP |
60 | } |
61 | ||
611c4f8a | 62 | bool unit_has_host_root_cgroup(Unit *u) { |
f3725e64 LP |
63 | assert(u); |
64 | ||
cc6271f1 LP |
65 | /* Returns whether this unit manages the root cgroup. This will return true if this unit is the root slice and |
66 | * the manager manages the root cgroup. */ | |
f3725e64 | 67 | |
611c4f8a | 68 | if (!manager_owns_host_root_cgroup(u->manager)) |
f3725e64 LP |
69 | return false; |
70 | ||
cc6271f1 | 71 | return unit_has_name(u, SPECIAL_ROOT_SLICE); |
f3725e64 LP |
72 | } |
73 | ||
293d32df LP |
74 | static int set_attribute_and_warn(Unit *u, const char *controller, const char *attribute, const char *value) { |
75 | int r; | |
76 | ||
77 | r = cg_set_attribute(controller, u->cgroup_path, attribute, value); | |
78 | if (r < 0) | |
79 | log_unit_full(u, LOG_LEVEL_CGROUP_WRITE(r), r, "Failed to set '%s' attribute on '%s' to '%.*s': %m", | |
80 | strna(attribute), isempty(u->cgroup_path) ? "/" : u->cgroup_path, (int) strcspn(value, NEWLINE), value); | |
81 | ||
82 | return r; | |
83 | } | |
84 | ||
2b40998d | 85 | static void cgroup_compat_warn(void) { |
128fadc9 TH |
86 | static bool cgroup_compat_warned = false; |
87 | ||
88 | if (cgroup_compat_warned) | |
89 | return; | |
90 | ||
cc6271f1 LP |
91 | log_warning("cgroup compatibility translation between legacy and unified hierarchy settings activated. " |
92 | "See cgroup-compat debug messages for details."); | |
93 | ||
128fadc9 TH |
94 | cgroup_compat_warned = true; |
95 | } | |
96 | ||
97 | #define log_cgroup_compat(unit, fmt, ...) do { \ | |
98 | cgroup_compat_warn(); \ | |
99 | log_unit_debug(unit, "cgroup-compat: " fmt, ##__VA_ARGS__); \ | |
2b40998d | 100 | } while (false) |
128fadc9 | 101 | |
4ad49000 LP |
102 | void cgroup_context_init(CGroupContext *c) { |
103 | assert(c); | |
104 | ||
de8a711a | 105 | /* Initialize everything to the kernel defaults. */ |
4ad49000 | 106 | |
de8a711a LP |
107 | *c = (CGroupContext) { |
108 | .cpu_weight = CGROUP_WEIGHT_INVALID, | |
109 | .startup_cpu_weight = CGROUP_WEIGHT_INVALID, | |
110 | .cpu_quota_per_sec_usec = USEC_INFINITY, | |
10f28641 | 111 | .cpu_quota_period_usec = USEC_INFINITY, |
66ebf6c0 | 112 | |
de8a711a LP |
113 | .cpu_shares = CGROUP_CPU_SHARES_INVALID, |
114 | .startup_cpu_shares = CGROUP_CPU_SHARES_INVALID, | |
d53d9474 | 115 | |
de8a711a LP |
116 | .memory_high = CGROUP_LIMIT_MAX, |
117 | .memory_max = CGROUP_LIMIT_MAX, | |
118 | .memory_swap_max = CGROUP_LIMIT_MAX, | |
da4d897e | 119 | |
de8a711a | 120 | .memory_limit = CGROUP_LIMIT_MAX, |
b2f8b02e | 121 | |
de8a711a LP |
122 | .io_weight = CGROUP_WEIGHT_INVALID, |
123 | .startup_io_weight = CGROUP_WEIGHT_INVALID, | |
13c31542 | 124 | |
de8a711a LP |
125 | .blockio_weight = CGROUP_BLKIO_WEIGHT_INVALID, |
126 | .startup_blockio_weight = CGROUP_BLKIO_WEIGHT_INVALID, | |
d53d9474 | 127 | |
3a0f06c4 | 128 | .tasks_max = TASKS_MAX_UNSET, |
de8a711a | 129 | }; |
4ad49000 | 130 | } |
8e274523 | 131 | |
4ad49000 LP |
132 | void cgroup_context_free_device_allow(CGroupContext *c, CGroupDeviceAllow *a) { |
133 | assert(c); | |
134 | assert(a); | |
135 | ||
71fda00f | 136 | LIST_REMOVE(device_allow, c->device_allow, a); |
4ad49000 LP |
137 | free(a->path); |
138 | free(a); | |
139 | } | |
140 | ||
13c31542 TH |
141 | void cgroup_context_free_io_device_weight(CGroupContext *c, CGroupIODeviceWeight *w) { |
142 | assert(c); | |
143 | assert(w); | |
144 | ||
145 | LIST_REMOVE(device_weights, c->io_device_weights, w); | |
146 | free(w->path); | |
147 | free(w); | |
148 | } | |
149 | ||
6ae4283c TH |
150 | void cgroup_context_free_io_device_latency(CGroupContext *c, CGroupIODeviceLatency *l) { |
151 | assert(c); | |
152 | assert(l); | |
153 | ||
154 | LIST_REMOVE(device_latencies, c->io_device_latencies, l); | |
155 | free(l->path); | |
156 | free(l); | |
157 | } | |
158 | ||
13c31542 TH |
159 | void cgroup_context_free_io_device_limit(CGroupContext *c, CGroupIODeviceLimit *l) { |
160 | assert(c); | |
161 | assert(l); | |
162 | ||
163 | LIST_REMOVE(device_limits, c->io_device_limits, l); | |
164 | free(l->path); | |
165 | free(l); | |
166 | } | |
167 | ||
4ad49000 LP |
168 | void cgroup_context_free_blockio_device_weight(CGroupContext *c, CGroupBlockIODeviceWeight *w) { |
169 | assert(c); | |
170 | assert(w); | |
171 | ||
71fda00f | 172 | LIST_REMOVE(device_weights, c->blockio_device_weights, w); |
4ad49000 LP |
173 | free(w->path); |
174 | free(w); | |
175 | } | |
176 | ||
177 | void cgroup_context_free_blockio_device_bandwidth(CGroupContext *c, CGroupBlockIODeviceBandwidth *b) { | |
178 | assert(c); | |
8e274523 | 179 | assert(b); |
8e274523 | 180 | |
71fda00f | 181 | LIST_REMOVE(device_bandwidths, c->blockio_device_bandwidths, b); |
4ad49000 LP |
182 | free(b->path); |
183 | free(b); | |
184 | } | |
185 | ||
186 | void cgroup_context_done(CGroupContext *c) { | |
187 | assert(c); | |
188 | ||
13c31542 TH |
189 | while (c->io_device_weights) |
190 | cgroup_context_free_io_device_weight(c, c->io_device_weights); | |
191 | ||
6ae4283c TH |
192 | while (c->io_device_latencies) |
193 | cgroup_context_free_io_device_latency(c, c->io_device_latencies); | |
194 | ||
13c31542 TH |
195 | while (c->io_device_limits) |
196 | cgroup_context_free_io_device_limit(c, c->io_device_limits); | |
197 | ||
4ad49000 LP |
198 | while (c->blockio_device_weights) |
199 | cgroup_context_free_blockio_device_weight(c, c->blockio_device_weights); | |
200 | ||
201 | while (c->blockio_device_bandwidths) | |
202 | cgroup_context_free_blockio_device_bandwidth(c, c->blockio_device_bandwidths); | |
203 | ||
204 | while (c->device_allow) | |
205 | cgroup_context_free_device_allow(c, c->device_allow); | |
6a48d82f DM |
206 | |
207 | c->ip_address_allow = ip_address_access_free_all(c->ip_address_allow); | |
208 | c->ip_address_deny = ip_address_access_free_all(c->ip_address_deny); | |
fab34748 KL |
209 | |
210 | c->ip_filters_ingress = strv_free(c->ip_filters_ingress); | |
211 | c->ip_filters_egress = strv_free(c->ip_filters_egress); | |
047f5d63 PH |
212 | |
213 | cpu_set_reset(&c->cpuset_cpus); | |
214 | cpu_set_reset(&c->cpuset_mems); | |
4ad49000 LP |
215 | } |
216 | ||
74b5fb27 CD |
217 | static int unit_get_kernel_memory_limit(Unit *u, const char *file, uint64_t *ret) { |
218 | _cleanup_free_ char *raw_kval = NULL; | |
219 | uint64_t kval; | |
220 | int r; | |
221 | ||
222 | assert(u); | |
223 | ||
224 | if (!u->cgroup_realized) | |
225 | return -EOWNERDEAD; | |
226 | ||
227 | r = cg_get_attribute("memory", u->cgroup_path, file, &raw_kval); | |
228 | if (r < 0) | |
229 | return r; | |
230 | ||
231 | if (streq(raw_kval, "max")) { | |
232 | *ret = CGROUP_LIMIT_MAX; | |
233 | return 0; | |
234 | } | |
235 | ||
236 | r = safe_atou64(raw_kval, &kval); | |
237 | if (r < 0) | |
238 | return r; | |
239 | ||
240 | *ret = kval; | |
241 | ||
242 | return 0; | |
243 | } | |
244 | ||
245 | static int unit_compare_memory_limit(Unit *u, const char *property_name, uint64_t *ret_unit_value, uint64_t *ret_kernel_value) { | |
246 | CGroupContext *c; | |
247 | CGroupMask m; | |
248 | const char *file; | |
249 | uint64_t unit_value; | |
250 | int r; | |
251 | ||
252 | /* Compare kernel memcg configuration against our internal systemd state. Unsupported (and will | |
253 | * return -ENODATA) on cgroup v1. | |
254 | * | |
255 | * Returns: | |
256 | * | |
257 | * <0: On error. | |
258 | * 0: If the kernel memory setting doesn't match our configuration. | |
259 | * >0: If the kernel memory setting matches our configuration. | |
260 | * | |
261 | * The following values are only guaranteed to be populated on return >=0: | |
262 | * | |
263 | * - ret_unit_value will contain our internal expected value for the unit, page-aligned. | |
264 | * - ret_kernel_value will contain the actual value presented by the kernel. */ | |
265 | ||
266 | assert(u); | |
267 | ||
268 | r = cg_all_unified(); | |
269 | if (r < 0) | |
270 | return log_debug_errno(r, "Failed to determine cgroup hierarchy version: %m"); | |
271 | ||
272 | /* Unsupported on v1. | |
273 | * | |
274 | * We don't return ENOENT, since that could actually mask a genuine problem where somebody else has | |
275 | * silently masked the controller. */ | |
276 | if (r == 0) | |
277 | return -ENODATA; | |
278 | ||
279 | /* The root slice doesn't have any controller files, so we can't compare anything. */ | |
280 | if (unit_has_name(u, SPECIAL_ROOT_SLICE)) | |
281 | return -ENODATA; | |
282 | ||
283 | /* It's possible to have MemoryFoo set without systemd wanting to have the memory controller enabled, | |
284 | * for example, in the case of DisableControllers= or cgroup_disable on the kernel command line. To | |
285 | * avoid specious errors in these scenarios, check that we even expect the memory controller to be | |
286 | * enabled at all. */ | |
287 | m = unit_get_target_mask(u); | |
288 | if (!FLAGS_SET(m, CGROUP_MASK_MEMORY)) | |
289 | return -ENODATA; | |
290 | ||
291 | c = unit_get_cgroup_context(u); | |
292 | assert(c); | |
293 | ||
294 | if (streq(property_name, "MemoryLow")) { | |
295 | unit_value = unit_get_ancestor_memory_low(u); | |
296 | file = "memory.low"; | |
297 | } else if (streq(property_name, "MemoryMin")) { | |
298 | unit_value = unit_get_ancestor_memory_min(u); | |
299 | file = "memory.min"; | |
300 | } else if (streq(property_name, "MemoryHigh")) { | |
301 | unit_value = c->memory_high; | |
302 | file = "memory.high"; | |
303 | } else if (streq(property_name, "MemoryMax")) { | |
304 | unit_value = c->memory_max; | |
305 | file = "memory.max"; | |
306 | } else if (streq(property_name, "MemorySwapMax")) { | |
307 | unit_value = c->memory_swap_max; | |
308 | file = "memory.swap.max"; | |
309 | } else | |
310 | return -EINVAL; | |
311 | ||
312 | r = unit_get_kernel_memory_limit(u, file, ret_kernel_value); | |
313 | if (r < 0) | |
314 | return log_unit_debug_errno(u, r, "Failed to parse %s: %m", file); | |
315 | ||
316 | /* It's intended (soon) in a future kernel to not expose cgroup memory limits rounded to page | |
317 | * boundaries, but instead separate the user-exposed limit, which is whatever userspace told us, from | |
318 | * our internal page-counting. To support those future kernels, just check the value itself first | |
319 | * without any page-alignment. */ | |
320 | if (*ret_kernel_value == unit_value) { | |
321 | *ret_unit_value = unit_value; | |
322 | return 1; | |
323 | } | |
324 | ||
325 | /* The current kernel behaviour, by comparison, is that even if you write a particular number of | |
326 | * bytes into a cgroup memory file, it always returns that number page-aligned down (since the kernel | |
327 | * internally stores cgroup limits in pages). As such, so long as it aligns properly, everything is | |
328 | * cricket. */ | |
329 | if (unit_value != CGROUP_LIMIT_MAX) | |
330 | unit_value = PAGE_ALIGN_DOWN(unit_value); | |
331 | ||
332 | *ret_unit_value = unit_value; | |
333 | ||
334 | return *ret_kernel_value == *ret_unit_value; | |
335 | } | |
336 | ||
bc0623df CD |
337 | #define FORMAT_CGROUP_DIFF_MAX 128 |
338 | ||
339 | static char *format_cgroup_memory_limit_comparison(char *buf, size_t l, Unit *u, const char *property_name) { | |
340 | uint64_t kval, sval; | |
341 | int r; | |
342 | ||
343 | assert(u); | |
344 | assert(buf); | |
345 | assert(l > 0); | |
346 | ||
347 | r = unit_compare_memory_limit(u, property_name, &sval, &kval); | |
348 | ||
349 | /* memory.swap.max is special in that it relies on CONFIG_MEMCG_SWAP (and the default swapaccount=1). | |
350 | * In the absence of reliably being able to detect whether memcg swap support is available or not, | |
351 | * only complain if the error is not ENOENT. */ | |
352 | if (r > 0 || IN_SET(r, -ENODATA, -EOWNERDEAD) || | |
353 | (r == -ENOENT && streq(property_name, "MemorySwapMax"))) { | |
354 | buf[0] = 0; | |
355 | return buf; | |
356 | } | |
357 | ||
358 | if (r < 0) { | |
359 | snprintf(buf, l, " (error getting kernel value: %s)", strerror_safe(r)); | |
360 | return buf; | |
361 | } | |
362 | ||
363 | snprintf(buf, l, " (different value in kernel: %" PRIu64 ")", kval); | |
364 | ||
365 | return buf; | |
366 | } | |
367 | ||
368 | void cgroup_context_dump(Unit *u, FILE* f, const char *prefix) { | |
85c3b278 | 369 | _cleanup_free_ char *disable_controllers_str = NULL, *cpuset_cpus = NULL, *cpuset_mems = NULL; |
13c31542 TH |
370 | CGroupIODeviceLimit *il; |
371 | CGroupIODeviceWeight *iw; | |
6ae4283c | 372 | CGroupIODeviceLatency *l; |
4ad49000 LP |
373 | CGroupBlockIODeviceBandwidth *b; |
374 | CGroupBlockIODeviceWeight *w; | |
375 | CGroupDeviceAllow *a; | |
bc0623df | 376 | CGroupContext *c; |
c21c9906 | 377 | IPAddressAccessItem *iaai; |
fab34748 | 378 | char **path; |
bc0623df | 379 | char q[FORMAT_TIMESPAN_MAX]; |
10f28641 | 380 | char v[FORMAT_TIMESPAN_MAX]; |
4ad49000 | 381 | |
bc0623df CD |
382 | char cda[FORMAT_CGROUP_DIFF_MAX]; |
383 | char cdb[FORMAT_CGROUP_DIFF_MAX]; | |
384 | char cdc[FORMAT_CGROUP_DIFF_MAX]; | |
385 | char cdd[FORMAT_CGROUP_DIFF_MAX]; | |
386 | char cde[FORMAT_CGROUP_DIFF_MAX]; | |
387 | ||
388 | assert(u); | |
4ad49000 LP |
389 | assert(f); |
390 | ||
bc0623df CD |
391 | c = unit_get_cgroup_context(u); |
392 | assert(c); | |
393 | ||
4ad49000 LP |
394 | prefix = strempty(prefix); |
395 | ||
25cc30c4 AZ |
396 | (void) cg_mask_to_string(c->disable_controllers, &disable_controllers_str); |
397 | ||
047f5d63 PH |
398 | cpuset_cpus = cpu_set_to_range_string(&c->cpuset_cpus); |
399 | cpuset_mems = cpu_set_to_range_string(&c->cpuset_mems); | |
400 | ||
4ad49000 | 401 | fprintf(f, |
6dfb9282 CD |
402 | "%sCPUAccounting: %s\n" |
403 | "%sIOAccounting: %s\n" | |
404 | "%sBlockIOAccounting: %s\n" | |
405 | "%sMemoryAccounting: %s\n" | |
406 | "%sTasksAccounting: %s\n" | |
407 | "%sIPAccounting: %s\n" | |
408 | "%sCPUWeight: %" PRIu64 "\n" | |
409 | "%sStartupCPUWeight: %" PRIu64 "\n" | |
410 | "%sCPUShares: %" PRIu64 "\n" | |
411 | "%sStartupCPUShares: %" PRIu64 "\n" | |
412 | "%sCPUQuotaPerSecSec: %s\n" | |
413 | "%sCPUQuotaPeriodSec: %s\n" | |
414 | "%sAllowedCPUs: %s\n" | |
415 | "%sAllowedMemoryNodes: %s\n" | |
416 | "%sIOWeight: %" PRIu64 "\n" | |
417 | "%sStartupIOWeight: %" PRIu64 "\n" | |
418 | "%sBlockIOWeight: %" PRIu64 "\n" | |
419 | "%sStartupBlockIOWeight: %" PRIu64 "\n" | |
420 | "%sDefaultMemoryMin: %" PRIu64 "\n" | |
421 | "%sDefaultMemoryLow: %" PRIu64 "\n" | |
bc0623df CD |
422 | "%sMemoryMin: %" PRIu64 "%s\n" |
423 | "%sMemoryLow: %" PRIu64 "%s\n" | |
424 | "%sMemoryHigh: %" PRIu64 "%s\n" | |
425 | "%sMemoryMax: %" PRIu64 "%s\n" | |
426 | "%sMemorySwapMax: %" PRIu64 "%s\n" | |
6dfb9282 CD |
427 | "%sMemoryLimit: %" PRIu64 "\n" |
428 | "%sTasksMax: %" PRIu64 "\n" | |
429 | "%sDevicePolicy: %s\n" | |
430 | "%sDisableControllers: %s\n" | |
431 | "%sDelegate: %s\n", | |
4ad49000 | 432 | prefix, yes_no(c->cpu_accounting), |
13c31542 | 433 | prefix, yes_no(c->io_accounting), |
4ad49000 LP |
434 | prefix, yes_no(c->blockio_accounting), |
435 | prefix, yes_no(c->memory_accounting), | |
d53d9474 | 436 | prefix, yes_no(c->tasks_accounting), |
c21c9906 | 437 | prefix, yes_no(c->ip_accounting), |
66ebf6c0 TH |
438 | prefix, c->cpu_weight, |
439 | prefix, c->startup_cpu_weight, | |
4ad49000 | 440 | prefix, c->cpu_shares, |
95ae05c0 | 441 | prefix, c->startup_cpu_shares, |
bc0623df | 442 | prefix, format_timespan(q, sizeof(q), c->cpu_quota_per_sec_usec, 1), |
10f28641 | 443 | prefix, format_timespan(v, sizeof(v), c->cpu_quota_period_usec, 1), |
85c3b278 LP |
444 | prefix, strempty(cpuset_cpus), |
445 | prefix, strempty(cpuset_mems), | |
13c31542 TH |
446 | prefix, c->io_weight, |
447 | prefix, c->startup_io_weight, | |
4ad49000 | 448 | prefix, c->blockio_weight, |
95ae05c0 | 449 | prefix, c->startup_blockio_weight, |
7ad5439e | 450 | prefix, c->default_memory_min, |
c52db42b | 451 | prefix, c->default_memory_low, |
bc0623df CD |
452 | prefix, c->memory_min, format_cgroup_memory_limit_comparison(cda, sizeof(cda), u, "MemoryMin"), |
453 | prefix, c->memory_low, format_cgroup_memory_limit_comparison(cdb, sizeof(cdb), u, "MemoryLow"), | |
454 | prefix, c->memory_high, format_cgroup_memory_limit_comparison(cdc, sizeof(cdc), u, "MemoryHigh"), | |
455 | prefix, c->memory_max, format_cgroup_memory_limit_comparison(cdd, sizeof(cdd), u, "MemoryMax"), | |
456 | prefix, c->memory_swap_max, format_cgroup_memory_limit_comparison(cde, sizeof(cde), u, "MemorySwapMax"), | |
4ad49000 | 457 | prefix, c->memory_limit, |
3a0f06c4 | 458 | prefix, tasks_max_resolve(&c->tasks_max), |
a931ad47 | 459 | prefix, cgroup_device_policy_to_string(c->device_policy), |
f4c43a81 | 460 | prefix, strempty(disable_controllers_str), |
a931ad47 | 461 | prefix, yes_no(c->delegate)); |
4ad49000 | 462 | |
02638280 LP |
463 | if (c->delegate) { |
464 | _cleanup_free_ char *t = NULL; | |
465 | ||
466 | (void) cg_mask_to_string(c->delegate_controllers, &t); | |
467 | ||
6dfb9282 | 468 | fprintf(f, "%sDelegateControllers: %s\n", |
02638280 LP |
469 | prefix, |
470 | strempty(t)); | |
471 | } | |
472 | ||
4ad49000 LP |
473 | LIST_FOREACH(device_allow, a, c->device_allow) |
474 | fprintf(f, | |
6dfb9282 | 475 | "%sDeviceAllow: %s %s%s%s\n", |
4ad49000 LP |
476 | prefix, |
477 | a->path, | |
478 | a->r ? "r" : "", a->w ? "w" : "", a->m ? "m" : ""); | |
479 | ||
13c31542 TH |
480 | LIST_FOREACH(device_weights, iw, c->io_device_weights) |
481 | fprintf(f, | |
6dfb9282 | 482 | "%sIODeviceWeight: %s %" PRIu64 "\n", |
13c31542 TH |
483 | prefix, |
484 | iw->path, | |
485 | iw->weight); | |
486 | ||
6ae4283c TH |
487 | LIST_FOREACH(device_latencies, l, c->io_device_latencies) |
488 | fprintf(f, | |
6dfb9282 | 489 | "%sIODeviceLatencyTargetSec: %s %s\n", |
6ae4283c TH |
490 | prefix, |
491 | l->path, | |
bc0623df | 492 | format_timespan(q, sizeof(q), l->target_usec, 1)); |
6ae4283c | 493 | |
13c31542 TH |
494 | LIST_FOREACH(device_limits, il, c->io_device_limits) { |
495 | char buf[FORMAT_BYTES_MAX]; | |
9be57249 TH |
496 | CGroupIOLimitType type; |
497 | ||
498 | for (type = 0; type < _CGROUP_IO_LIMIT_TYPE_MAX; type++) | |
499 | if (il->limits[type] != cgroup_io_limit_defaults[type]) | |
500 | fprintf(f, | |
6dfb9282 | 501 | "%s%s: %s %s\n", |
9be57249 TH |
502 | prefix, |
503 | cgroup_io_limit_type_to_string(type), | |
504 | il->path, | |
505 | format_bytes(buf, sizeof(buf), il->limits[type])); | |
13c31542 TH |
506 | } |
507 | ||
4ad49000 LP |
508 | LIST_FOREACH(device_weights, w, c->blockio_device_weights) |
509 | fprintf(f, | |
6dfb9282 | 510 | "%sBlockIODeviceWeight: %s %" PRIu64, |
4ad49000 LP |
511 | prefix, |
512 | w->path, | |
513 | w->weight); | |
514 | ||
515 | LIST_FOREACH(device_bandwidths, b, c->blockio_device_bandwidths) { | |
516 | char buf[FORMAT_BYTES_MAX]; | |
517 | ||
979d0311 TH |
518 | if (b->rbps != CGROUP_LIMIT_MAX) |
519 | fprintf(f, | |
6dfb9282 | 520 | "%sBlockIOReadBandwidth: %s %s\n", |
979d0311 TH |
521 | prefix, |
522 | b->path, | |
523 | format_bytes(buf, sizeof(buf), b->rbps)); | |
524 | if (b->wbps != CGROUP_LIMIT_MAX) | |
525 | fprintf(f, | |
6dfb9282 | 526 | "%sBlockIOWriteBandwidth: %s %s\n", |
979d0311 TH |
527 | prefix, |
528 | b->path, | |
529 | format_bytes(buf, sizeof(buf), b->wbps)); | |
4ad49000 | 530 | } |
c21c9906 LP |
531 | |
532 | LIST_FOREACH(items, iaai, c->ip_address_allow) { | |
533 | _cleanup_free_ char *k = NULL; | |
534 | ||
535 | (void) in_addr_to_string(iaai->family, &iaai->address, &k); | |
6dfb9282 | 536 | fprintf(f, "%sIPAddressAllow: %s/%u\n", prefix, strnull(k), iaai->prefixlen); |
c21c9906 LP |
537 | } |
538 | ||
539 | LIST_FOREACH(items, iaai, c->ip_address_deny) { | |
540 | _cleanup_free_ char *k = NULL; | |
541 | ||
542 | (void) in_addr_to_string(iaai->family, &iaai->address, &k); | |
6dfb9282 | 543 | fprintf(f, "%sIPAddressDeny: %s/%u\n", prefix, strnull(k), iaai->prefixlen); |
c21c9906 | 544 | } |
fab34748 KL |
545 | |
546 | STRV_FOREACH(path, c->ip_filters_ingress) | |
6dfb9282 | 547 | fprintf(f, "%sIPIngressFilterPath: %s\n", prefix, *path); |
fab34748 KL |
548 | |
549 | STRV_FOREACH(path, c->ip_filters_egress) | |
6dfb9282 | 550 | fprintf(f, "%sIPEgressFilterPath: %s\n", prefix, *path); |
4ad49000 LP |
551 | } |
552 | ||
fd870bac YW |
553 | int cgroup_add_device_allow(CGroupContext *c, const char *dev, const char *mode) { |
554 | _cleanup_free_ CGroupDeviceAllow *a = NULL; | |
555 | _cleanup_free_ char *d = NULL; | |
556 | ||
557 | assert(c); | |
558 | assert(dev); | |
559 | assert(isempty(mode) || in_charset(mode, "rwm")); | |
560 | ||
561 | a = new(CGroupDeviceAllow, 1); | |
562 | if (!a) | |
563 | return -ENOMEM; | |
564 | ||
565 | d = strdup(dev); | |
566 | if (!d) | |
567 | return -ENOMEM; | |
568 | ||
569 | *a = (CGroupDeviceAllow) { | |
570 | .path = TAKE_PTR(d), | |
490c5a37 LP |
571 | .r = isempty(mode) || strchr(mode, 'r'), |
572 | .w = isempty(mode) || strchr(mode, 'w'), | |
573 | .m = isempty(mode) || strchr(mode, 'm'), | |
fd870bac YW |
574 | }; |
575 | ||
576 | LIST_PREPEND(device_allow, c->device_allow, a); | |
577 | TAKE_PTR(a); | |
578 | ||
579 | return 0; | |
580 | } | |
581 | ||
6264b85e CD |
582 | #define UNIT_DEFINE_ANCESTOR_MEMORY_LOOKUP(entry) \ |
583 | uint64_t unit_get_ancestor_##entry(Unit *u) { \ | |
584 | CGroupContext *c; \ | |
585 | \ | |
586 | /* 1. Is entry set in this unit? If so, use that. \ | |
587 | * 2. Is the default for this entry set in any \ | |
588 | * ancestor? If so, use that. \ | |
589 | * 3. Otherwise, return CGROUP_LIMIT_MIN. */ \ | |
590 | \ | |
591 | assert(u); \ | |
592 | \ | |
593 | c = unit_get_cgroup_context(u); \ | |
c5322608 | 594 | if (c && c->entry##_set) \ |
6264b85e CD |
595 | return c->entry; \ |
596 | \ | |
c5322608 | 597 | while ((u = UNIT_DEREF(u->slice))) { \ |
6264b85e | 598 | c = unit_get_cgroup_context(u); \ |
c5322608 | 599 | if (c && c->default_##entry##_set) \ |
6264b85e CD |
600 | return c->default_##entry; \ |
601 | } \ | |
602 | \ | |
603 | /* We've reached the root, but nobody had default for \ | |
604 | * this entry set, so set it to the kernel default. */ \ | |
605 | return CGROUP_LIMIT_MIN; \ | |
c52db42b CD |
606 | } |
607 | ||
6264b85e | 608 | UNIT_DEFINE_ANCESTOR_MEMORY_LOOKUP(memory_low); |
7ad5439e | 609 | UNIT_DEFINE_ANCESTOR_MEMORY_LOOKUP(memory_min); |
6264b85e | 610 | |
0d2d6fbf CD |
611 | static void cgroup_xattr_apply(Unit *u) { |
612 | char ids[SD_ID128_STRING_MAX]; | |
613 | int r; | |
614 | ||
615 | assert(u); | |
616 | ||
617 | if (!MANAGER_IS_SYSTEM(u->manager)) | |
618 | return; | |
619 | ||
3288ea8f LP |
620 | if (!sd_id128_is_null(u->invocation_id)) { |
621 | r = cg_set_xattr(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path, | |
622 | "trusted.invocation_id", | |
623 | sd_id128_to_string(u->invocation_id, ids), 32, | |
624 | 0); | |
625 | if (r < 0) | |
626 | log_unit_debug_errno(u, r, "Failed to set invocation ID on control group %s, ignoring: %m", u->cgroup_path); | |
627 | } | |
0d2d6fbf | 628 | |
3288ea8f LP |
629 | if (unit_cgroup_delegate(u)) { |
630 | r = cg_set_xattr(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path, | |
631 | "trusted.delegate", | |
632 | "1", 1, | |
633 | 0); | |
634 | if (r < 0) | |
635 | log_unit_debug_errno(u, r, "Failed to set delegate flag on control group %s, ignoring: %m", u->cgroup_path); | |
636 | } else { | |
637 | r = cg_remove_xattr(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path, "trusted.delegate"); | |
638 | if (r != -ENODATA) | |
639 | log_unit_debug_errno(u, r, "Failed to remove delegate flag on control group %s, ignoring: %m", u->cgroup_path); | |
640 | } | |
0d2d6fbf CD |
641 | } |
642 | ||
45c2e068 | 643 | static int lookup_block_device(const char *p, dev_t *ret) { |
f5855697 YS |
644 | dev_t rdev, dev = 0; |
645 | mode_t mode; | |
45c2e068 | 646 | int r; |
4ad49000 LP |
647 | |
648 | assert(p); | |
45c2e068 | 649 | assert(ret); |
4ad49000 | 650 | |
f5855697 | 651 | r = device_path_parse_major_minor(p, &mode, &rdev); |
d5aecba6 | 652 | if (r == -ENODEV) { /* not a parsable device node, need to go to disk */ |
f5855697 | 653 | struct stat st; |
d5aecba6 LP |
654 | if (stat(p, &st) < 0) |
655 | return log_warning_errno(errno, "Couldn't stat device '%s': %m", p); | |
f5855697 YS |
656 | rdev = (dev_t)st.st_rdev; |
657 | dev = (dev_t)st.st_dev; | |
658 | mode = st.st_mode; | |
d5aecba6 LP |
659 | } else if (r < 0) |
660 | return log_warning_errno(r, "Failed to parse major/minor from path '%s': %m", p); | |
661 | ||
f5855697 | 662 | if (S_ISCHR(mode)) { |
d5aecba6 LP |
663 | log_warning("Device node '%s' is a character device, but block device needed.", p); |
664 | return -ENOTBLK; | |
f5855697 YS |
665 | } else if (S_ISBLK(mode)) |
666 | *ret = rdev; | |
667 | else if (major(dev) != 0) | |
668 | *ret = dev; /* If this is not a device node then use the block device this file is stored on */ | |
45c2e068 LP |
669 | else { |
670 | /* If this is btrfs, getting the backing block device is a bit harder */ | |
671 | r = btrfs_get_block_device(p, ret); | |
672 | if (r < 0 && r != -ENOTTY) | |
673 | return log_warning_errno(r, "Failed to determine block device backing btrfs file system '%s': %m", p); | |
674 | if (r == -ENOTTY) { | |
675 | log_warning("'%s' is not a block device node, and file system block device cannot be determined or is not local.", p); | |
676 | return -ENODEV; | |
677 | } | |
4ad49000 | 678 | } |
8e274523 | 679 | |
45c2e068 LP |
680 | /* If this is a LUKS device, try to get the originating block device */ |
681 | (void) block_get_originating(*ret, ret); | |
682 | ||
683 | /* If this is a partition, try to get the originating block device */ | |
684 | (void) block_get_whole_disk(*ret, ret); | |
8e274523 | 685 | return 0; |
8e274523 LP |
686 | } |
687 | ||
66ebf6c0 TH |
688 | static bool cgroup_context_has_cpu_weight(CGroupContext *c) { |
689 | return c->cpu_weight != CGROUP_WEIGHT_INVALID || | |
690 | c->startup_cpu_weight != CGROUP_WEIGHT_INVALID; | |
691 | } | |
692 | ||
693 | static bool cgroup_context_has_cpu_shares(CGroupContext *c) { | |
694 | return c->cpu_shares != CGROUP_CPU_SHARES_INVALID || | |
695 | c->startup_cpu_shares != CGROUP_CPU_SHARES_INVALID; | |
696 | } | |
697 | ||
698 | static uint64_t cgroup_context_cpu_weight(CGroupContext *c, ManagerState state) { | |
699 | if (IN_SET(state, MANAGER_STARTING, MANAGER_INITIALIZING) && | |
700 | c->startup_cpu_weight != CGROUP_WEIGHT_INVALID) | |
701 | return c->startup_cpu_weight; | |
702 | else if (c->cpu_weight != CGROUP_WEIGHT_INVALID) | |
703 | return c->cpu_weight; | |
704 | else | |
705 | return CGROUP_WEIGHT_DEFAULT; | |
706 | } | |
707 | ||
708 | static uint64_t cgroup_context_cpu_shares(CGroupContext *c, ManagerState state) { | |
709 | if (IN_SET(state, MANAGER_STARTING, MANAGER_INITIALIZING) && | |
710 | c->startup_cpu_shares != CGROUP_CPU_SHARES_INVALID) | |
711 | return c->startup_cpu_shares; | |
712 | else if (c->cpu_shares != CGROUP_CPU_SHARES_INVALID) | |
713 | return c->cpu_shares; | |
714 | else | |
715 | return CGROUP_CPU_SHARES_DEFAULT; | |
716 | } | |
717 | ||
10f28641 FB |
718 | usec_t cgroup_cpu_adjust_period(usec_t period, usec_t quota, usec_t resolution, usec_t max_period) { |
719 | /* kernel uses a minimum resolution of 1ms, so both period and (quota * period) | |
720 | * need to be higher than that boundary. quota is specified in USecPerSec. | |
721 | * Additionally, period must be at most max_period. */ | |
722 | assert(quota > 0); | |
723 | ||
724 | return MIN(MAX3(period, resolution, resolution * USEC_PER_SEC / quota), max_period); | |
725 | } | |
726 | ||
727 | static usec_t cgroup_cpu_adjust_period_and_log(Unit *u, usec_t period, usec_t quota) { | |
728 | usec_t new_period; | |
729 | ||
730 | if (quota == USEC_INFINITY) | |
731 | /* Always use default period for infinity quota. */ | |
732 | return CGROUP_CPU_QUOTA_DEFAULT_PERIOD_USEC; | |
733 | ||
734 | if (period == USEC_INFINITY) | |
735 | /* Default period was requested. */ | |
736 | period = CGROUP_CPU_QUOTA_DEFAULT_PERIOD_USEC; | |
737 | ||
738 | /* Clamp to interval [1ms, 1s] */ | |
739 | new_period = cgroup_cpu_adjust_period(period, quota, USEC_PER_MSEC, USEC_PER_SEC); | |
740 | ||
741 | if (new_period != period) { | |
742 | char v[FORMAT_TIMESPAN_MAX]; | |
527ede0c | 743 | log_unit_full(u, u->warned_clamping_cpu_quota_period ? LOG_DEBUG : LOG_WARNING, 0, |
10f28641 FB |
744 | "Clamping CPU interval for cpu.max: period is now %s", |
745 | format_timespan(v, sizeof(v), new_period, 1)); | |
527ede0c | 746 | u->warned_clamping_cpu_quota_period = true; |
10f28641 FB |
747 | } |
748 | ||
749 | return new_period; | |
750 | } | |
751 | ||
52fecf20 LP |
752 | static void cgroup_apply_unified_cpu_weight(Unit *u, uint64_t weight) { |
753 | char buf[DECIMAL_STR_MAX(uint64_t) + 2]; | |
66ebf6c0 TH |
754 | |
755 | xsprintf(buf, "%" PRIu64 "\n", weight); | |
293d32df | 756 | (void) set_attribute_and_warn(u, "cpu", "cpu.weight", buf); |
52fecf20 LP |
757 | } |
758 | ||
10f28641 | 759 | static void cgroup_apply_unified_cpu_quota(Unit *u, usec_t quota, usec_t period) { |
52fecf20 | 760 | char buf[(DECIMAL_STR_MAX(usec_t) + 1) * 2 + 1]; |
66ebf6c0 | 761 | |
10f28641 | 762 | period = cgroup_cpu_adjust_period_and_log(u, period, quota); |
66ebf6c0 TH |
763 | if (quota != USEC_INFINITY) |
764 | xsprintf(buf, USEC_FMT " " USEC_FMT "\n", | |
10f28641 | 765 | MAX(quota * period / USEC_PER_SEC, USEC_PER_MSEC), period); |
66ebf6c0 | 766 | else |
10f28641 | 767 | xsprintf(buf, "max " USEC_FMT "\n", period); |
293d32df | 768 | (void) set_attribute_and_warn(u, "cpu", "cpu.max", buf); |
66ebf6c0 TH |
769 | } |
770 | ||
52fecf20 LP |
771 | static void cgroup_apply_legacy_cpu_shares(Unit *u, uint64_t shares) { |
772 | char buf[DECIMAL_STR_MAX(uint64_t) + 2]; | |
66ebf6c0 TH |
773 | |
774 | xsprintf(buf, "%" PRIu64 "\n", shares); | |
293d32df | 775 | (void) set_attribute_and_warn(u, "cpu", "cpu.shares", buf); |
52fecf20 LP |
776 | } |
777 | ||
10f28641 | 778 | static void cgroup_apply_legacy_cpu_quota(Unit *u, usec_t quota, usec_t period) { |
52fecf20 | 779 | char buf[DECIMAL_STR_MAX(usec_t) + 2]; |
66ebf6c0 | 780 | |
10f28641 FB |
781 | period = cgroup_cpu_adjust_period_and_log(u, period, quota); |
782 | ||
783 | xsprintf(buf, USEC_FMT "\n", period); | |
293d32df | 784 | (void) set_attribute_and_warn(u, "cpu", "cpu.cfs_period_us", buf); |
66ebf6c0 TH |
785 | |
786 | if (quota != USEC_INFINITY) { | |
10f28641 | 787 | xsprintf(buf, USEC_FMT "\n", MAX(quota * period / USEC_PER_SEC, USEC_PER_MSEC)); |
293d32df | 788 | (void) set_attribute_and_warn(u, "cpu", "cpu.cfs_quota_us", buf); |
66ebf6c0 | 789 | } else |
589a5f7a | 790 | (void) set_attribute_and_warn(u, "cpu", "cpu.cfs_quota_us", "-1\n"); |
66ebf6c0 TH |
791 | } |
792 | ||
793 | static uint64_t cgroup_cpu_shares_to_weight(uint64_t shares) { | |
794 | return CLAMP(shares * CGROUP_WEIGHT_DEFAULT / CGROUP_CPU_SHARES_DEFAULT, | |
795 | CGROUP_WEIGHT_MIN, CGROUP_WEIGHT_MAX); | |
796 | } | |
797 | ||
798 | static uint64_t cgroup_cpu_weight_to_shares(uint64_t weight) { | |
799 | return CLAMP(weight * CGROUP_CPU_SHARES_DEFAULT / CGROUP_WEIGHT_DEFAULT, | |
800 | CGROUP_CPU_SHARES_MIN, CGROUP_CPU_SHARES_MAX); | |
801 | } | |
802 | ||
2cea199e | 803 | static void cgroup_apply_unified_cpuset(Unit *u, const CPUSet *cpus, const char *name) { |
047f5d63 PH |
804 | _cleanup_free_ char *buf = NULL; |
805 | ||
2cea199e | 806 | buf = cpu_set_to_range_string(cpus); |
c259ac9a LP |
807 | if (!buf) { |
808 | log_oom(); | |
809 | return; | |
810 | } | |
047f5d63 PH |
811 | |
812 | (void) set_attribute_and_warn(u, "cpuset", name, buf); | |
813 | } | |
814 | ||
508c45da | 815 | static bool cgroup_context_has_io_config(CGroupContext *c) { |
538b4852 TH |
816 | return c->io_accounting || |
817 | c->io_weight != CGROUP_WEIGHT_INVALID || | |
818 | c->startup_io_weight != CGROUP_WEIGHT_INVALID || | |
819 | c->io_device_weights || | |
6ae4283c | 820 | c->io_device_latencies || |
538b4852 TH |
821 | c->io_device_limits; |
822 | } | |
823 | ||
508c45da | 824 | static bool cgroup_context_has_blockio_config(CGroupContext *c) { |
538b4852 TH |
825 | return c->blockio_accounting || |
826 | c->blockio_weight != CGROUP_BLKIO_WEIGHT_INVALID || | |
827 | c->startup_blockio_weight != CGROUP_BLKIO_WEIGHT_INVALID || | |
828 | c->blockio_device_weights || | |
829 | c->blockio_device_bandwidths; | |
830 | } | |
831 | ||
508c45da | 832 | static uint64_t cgroup_context_io_weight(CGroupContext *c, ManagerState state) { |
64faf04c TH |
833 | if (IN_SET(state, MANAGER_STARTING, MANAGER_INITIALIZING) && |
834 | c->startup_io_weight != CGROUP_WEIGHT_INVALID) | |
835 | return c->startup_io_weight; | |
836 | else if (c->io_weight != CGROUP_WEIGHT_INVALID) | |
837 | return c->io_weight; | |
838 | else | |
839 | return CGROUP_WEIGHT_DEFAULT; | |
840 | } | |
841 | ||
508c45da | 842 | static uint64_t cgroup_context_blkio_weight(CGroupContext *c, ManagerState state) { |
64faf04c TH |
843 | if (IN_SET(state, MANAGER_STARTING, MANAGER_INITIALIZING) && |
844 | c->startup_blockio_weight != CGROUP_BLKIO_WEIGHT_INVALID) | |
845 | return c->startup_blockio_weight; | |
846 | else if (c->blockio_weight != CGROUP_BLKIO_WEIGHT_INVALID) | |
847 | return c->blockio_weight; | |
848 | else | |
849 | return CGROUP_BLKIO_WEIGHT_DEFAULT; | |
850 | } | |
851 | ||
508c45da | 852 | static uint64_t cgroup_weight_blkio_to_io(uint64_t blkio_weight) { |
538b4852 TH |
853 | return CLAMP(blkio_weight * CGROUP_WEIGHT_DEFAULT / CGROUP_BLKIO_WEIGHT_DEFAULT, |
854 | CGROUP_WEIGHT_MIN, CGROUP_WEIGHT_MAX); | |
855 | } | |
856 | ||
508c45da | 857 | static uint64_t cgroup_weight_io_to_blkio(uint64_t io_weight) { |
538b4852 TH |
858 | return CLAMP(io_weight * CGROUP_BLKIO_WEIGHT_DEFAULT / CGROUP_WEIGHT_DEFAULT, |
859 | CGROUP_BLKIO_WEIGHT_MIN, CGROUP_BLKIO_WEIGHT_MAX); | |
860 | } | |
861 | ||
f29ff115 | 862 | static void cgroup_apply_io_device_weight(Unit *u, const char *dev_path, uint64_t io_weight) { |
64faf04c TH |
863 | char buf[DECIMAL_STR_MAX(dev_t)*2+2+DECIMAL_STR_MAX(uint64_t)+1]; |
864 | dev_t dev; | |
865 | int r; | |
866 | ||
867 | r = lookup_block_device(dev_path, &dev); | |
868 | if (r < 0) | |
869 | return; | |
870 | ||
871 | xsprintf(buf, "%u:%u %" PRIu64 "\n", major(dev), minor(dev), io_weight); | |
293d32df | 872 | (void) set_attribute_and_warn(u, "io", "io.weight", buf); |
64faf04c TH |
873 | } |
874 | ||
f29ff115 | 875 | static void cgroup_apply_blkio_device_weight(Unit *u, const char *dev_path, uint64_t blkio_weight) { |
64faf04c TH |
876 | char buf[DECIMAL_STR_MAX(dev_t)*2+2+DECIMAL_STR_MAX(uint64_t)+1]; |
877 | dev_t dev; | |
878 | int r; | |
879 | ||
880 | r = lookup_block_device(dev_path, &dev); | |
881 | if (r < 0) | |
882 | return; | |
883 | ||
884 | xsprintf(buf, "%u:%u %" PRIu64 "\n", major(dev), minor(dev), blkio_weight); | |
293d32df | 885 | (void) set_attribute_and_warn(u, "blkio", "blkio.weight_device", buf); |
64faf04c TH |
886 | } |
887 | ||
6ae4283c TH |
888 | static void cgroup_apply_io_device_latency(Unit *u, const char *dev_path, usec_t target) { |
889 | char buf[DECIMAL_STR_MAX(dev_t)*2+2+7+DECIMAL_STR_MAX(uint64_t)+1]; | |
890 | dev_t dev; | |
891 | int r; | |
892 | ||
893 | r = lookup_block_device(dev_path, &dev); | |
894 | if (r < 0) | |
895 | return; | |
896 | ||
897 | if (target != USEC_INFINITY) | |
898 | xsprintf(buf, "%u:%u target=%" PRIu64 "\n", major(dev), minor(dev), target); | |
899 | else | |
900 | xsprintf(buf, "%u:%u target=max\n", major(dev), minor(dev)); | |
901 | ||
293d32df | 902 | (void) set_attribute_and_warn(u, "io", "io.latency", buf); |
6ae4283c TH |
903 | } |
904 | ||
17ae2780 | 905 | static void cgroup_apply_io_device_limit(Unit *u, const char *dev_path, uint64_t *limits) { |
64faf04c TH |
906 | char limit_bufs[_CGROUP_IO_LIMIT_TYPE_MAX][DECIMAL_STR_MAX(uint64_t)]; |
907 | char buf[DECIMAL_STR_MAX(dev_t)*2+2+(6+DECIMAL_STR_MAX(uint64_t)+1)*4]; | |
908 | CGroupIOLimitType type; | |
909 | dev_t dev; | |
64faf04c TH |
910 | int r; |
911 | ||
912 | r = lookup_block_device(dev_path, &dev); | |
913 | if (r < 0) | |
17ae2780 | 914 | return; |
64faf04c | 915 | |
17ae2780 LP |
916 | for (type = 0; type < _CGROUP_IO_LIMIT_TYPE_MAX; type++) |
917 | if (limits[type] != cgroup_io_limit_defaults[type]) | |
64faf04c | 918 | xsprintf(limit_bufs[type], "%" PRIu64, limits[type]); |
17ae2780 | 919 | else |
64faf04c | 920 | xsprintf(limit_bufs[type], "%s", limits[type] == CGROUP_LIMIT_MAX ? "max" : "0"); |
64faf04c TH |
921 | |
922 | xsprintf(buf, "%u:%u rbps=%s wbps=%s riops=%s wiops=%s\n", major(dev), minor(dev), | |
923 | limit_bufs[CGROUP_IO_RBPS_MAX], limit_bufs[CGROUP_IO_WBPS_MAX], | |
924 | limit_bufs[CGROUP_IO_RIOPS_MAX], limit_bufs[CGROUP_IO_WIOPS_MAX]); | |
293d32df | 925 | (void) set_attribute_and_warn(u, "io", "io.max", buf); |
64faf04c TH |
926 | } |
927 | ||
17ae2780 | 928 | static void cgroup_apply_blkio_device_limit(Unit *u, const char *dev_path, uint64_t rbps, uint64_t wbps) { |
64faf04c TH |
929 | char buf[DECIMAL_STR_MAX(dev_t)*2+2+DECIMAL_STR_MAX(uint64_t)+1]; |
930 | dev_t dev; | |
64faf04c TH |
931 | int r; |
932 | ||
933 | r = lookup_block_device(dev_path, &dev); | |
934 | if (r < 0) | |
17ae2780 | 935 | return; |
64faf04c | 936 | |
64faf04c | 937 | sprintf(buf, "%u:%u %" PRIu64 "\n", major(dev), minor(dev), rbps); |
293d32df | 938 | (void) set_attribute_and_warn(u, "blkio", "blkio.throttle.read_bps_device", buf); |
64faf04c | 939 | |
64faf04c | 940 | sprintf(buf, "%u:%u %" PRIu64 "\n", major(dev), minor(dev), wbps); |
293d32df | 941 | (void) set_attribute_and_warn(u, "blkio", "blkio.throttle.write_bps_device", buf); |
64faf04c TH |
942 | } |
943 | ||
c52db42b CD |
944 | static bool unit_has_unified_memory_config(Unit *u) { |
945 | CGroupContext *c; | |
946 | ||
947 | assert(u); | |
948 | ||
949 | c = unit_get_cgroup_context(u); | |
950 | assert(c); | |
951 | ||
7c9d2b79 | 952 | return unit_get_ancestor_memory_min(u) > 0 || unit_get_ancestor_memory_low(u) > 0 || |
c52db42b CD |
953 | c->memory_high != CGROUP_LIMIT_MAX || c->memory_max != CGROUP_LIMIT_MAX || |
954 | c->memory_swap_max != CGROUP_LIMIT_MAX; | |
da4d897e TH |
955 | } |
956 | ||
f29ff115 | 957 | static void cgroup_apply_unified_memory_limit(Unit *u, const char *file, uint64_t v) { |
589a5f7a | 958 | char buf[DECIMAL_STR_MAX(uint64_t) + 1] = "max\n"; |
da4d897e TH |
959 | |
960 | if (v != CGROUP_LIMIT_MAX) | |
961 | xsprintf(buf, "%" PRIu64 "\n", v); | |
962 | ||
293d32df | 963 | (void) set_attribute_and_warn(u, "memory", file, buf); |
da4d897e TH |
964 | } |
965 | ||
0f2d84d2 | 966 | static void cgroup_apply_firewall(Unit *u) { |
0f2d84d2 LP |
967 | assert(u); |
968 | ||
acf7f253 | 969 | /* Best-effort: let's apply IP firewalling and/or accounting if that's enabled */ |
906c06f6 | 970 | |
acf7f253 | 971 | if (bpf_firewall_compile(u) < 0) |
906c06f6 DM |
972 | return; |
973 | ||
fab34748 | 974 | (void) bpf_firewall_load_custom(u); |
906c06f6 | 975 | (void) bpf_firewall_install(u); |
906c06f6 DM |
976 | } |
977 | ||
8b139557 ZJS |
978 | static int cgroup_apply_devices(Unit *u) { |
979 | _cleanup_(bpf_program_unrefp) BPFProgram *prog = NULL; | |
980 | const char *path; | |
981 | CGroupContext *c; | |
982 | CGroupDeviceAllow *a; | |
45669ae2 | 983 | CGroupDevicePolicy policy; |
8b139557 ZJS |
984 | int r; |
985 | ||
986 | assert_se(c = unit_get_cgroup_context(u)); | |
987 | assert_se(path = u->cgroup_path); | |
988 | ||
45669ae2 ZJS |
989 | policy = c->device_policy; |
990 | ||
8b139557 | 991 | if (cg_all_unified() > 0) { |
45669ae2 | 992 | r = bpf_devices_cgroup_init(&prog, policy, c->device_allow); |
8b139557 ZJS |
993 | if (r < 0) |
994 | return log_unit_warning_errno(u, r, "Failed to initialize device control bpf program: %m"); | |
995 | ||
996 | } else { | |
997 | /* Changing the devices list of a populated cgroup might result in EINVAL, hence ignore | |
998 | * EINVAL here. */ | |
999 | ||
45669ae2 | 1000 | if (c->device_allow || policy != CGROUP_DEVICE_POLICY_AUTO) |
8b139557 ZJS |
1001 | r = cg_set_attribute("devices", path, "devices.deny", "a"); |
1002 | else | |
1003 | r = cg_set_attribute("devices", path, "devices.allow", "a"); | |
1004 | if (r < 0) | |
1005 | log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EINVAL, -EACCES, -EPERM) ? LOG_DEBUG : LOG_WARNING, r, | |
1006 | "Failed to reset devices.allow/devices.deny: %m"); | |
1007 | } | |
1008 | ||
45669ae2 ZJS |
1009 | bool whitelist_static = policy == CGROUP_DEVICE_POLICY_CLOSED || |
1010 | (policy == CGROUP_DEVICE_POLICY_AUTO && c->device_allow); | |
1011 | if (whitelist_static) | |
8b139557 ZJS |
1012 | (void) bpf_devices_whitelist_static(prog, path); |
1013 | ||
45669ae2 | 1014 | bool any = whitelist_static; |
8b139557 ZJS |
1015 | LIST_FOREACH(device_allow, a, c->device_allow) { |
1016 | char acc[4], *val; | |
1017 | unsigned k = 0; | |
1018 | ||
1019 | if (a->r) | |
1020 | acc[k++] = 'r'; | |
1021 | if (a->w) | |
1022 | acc[k++] = 'w'; | |
1023 | if (a->m) | |
1024 | acc[k++] = 'm'; | |
8b139557 ZJS |
1025 | if (k == 0) |
1026 | continue; | |
8b139557 ZJS |
1027 | acc[k++] = 0; |
1028 | ||
1029 | if (path_startswith(a->path, "/dev/")) | |
45669ae2 | 1030 | r = bpf_devices_whitelist_device(prog, path, a->path, acc); |
8b139557 | 1031 | else if ((val = startswith(a->path, "block-"))) |
45669ae2 | 1032 | r = bpf_devices_whitelist_major(prog, path, val, 'b', acc); |
8b139557 | 1033 | else if ((val = startswith(a->path, "char-"))) |
45669ae2 ZJS |
1034 | r = bpf_devices_whitelist_major(prog, path, val, 'c', acc); |
1035 | else { | |
8b139557 | 1036 | log_unit_debug(u, "Ignoring device '%s' while writing cgroup attribute.", a->path); |
45669ae2 ZJS |
1037 | continue; |
1038 | } | |
1039 | ||
1040 | if (r >= 0) | |
1041 | any = true; | |
1042 | } | |
1043 | ||
1044 | if (prog && !any) { | |
1045 | log_unit_warning_errno(u, SYNTHETIC_ERRNO(ENODEV), "No devices matched by device filter."); | |
1046 | ||
1047 | /* The kernel verifier would reject a program we would build with the normal intro and outro | |
1048 | but no whitelisting rules (outro would contain an unreachable instruction for successful | |
1049 | return). */ | |
1050 | policy = CGROUP_DEVICE_POLICY_STRICT; | |
8b139557 ZJS |
1051 | } |
1052 | ||
45669ae2 | 1053 | r = bpf_devices_apply_policy(prog, policy, any, path, &u->bpf_device_control_installed); |
8b139557 ZJS |
1054 | if (r < 0) { |
1055 | static bool warned = false; | |
1056 | ||
1057 | log_full_errno(warned ? LOG_DEBUG : LOG_WARNING, r, | |
1058 | "Unit %s configures device ACL, but the local system doesn't seem to support the BPF-based device controller.\n" | |
1059 | "Proceeding WITHOUT applying ACL (all devices will be accessible)!\n" | |
1060 | "(This warning is only shown for the first loaded unit using device ACL.)", u->id); | |
1061 | ||
1062 | warned = true; | |
1063 | } | |
1064 | return r; | |
1065 | } | |
1066 | ||
906c06f6 DM |
1067 | static void cgroup_context_apply( |
1068 | Unit *u, | |
1069 | CGroupMask apply_mask, | |
906c06f6 DM |
1070 | ManagerState state) { |
1071 | ||
f29ff115 TH |
1072 | const char *path; |
1073 | CGroupContext *c; | |
52fecf20 | 1074 | bool is_host_root, is_local_root; |
4ad49000 LP |
1075 | int r; |
1076 | ||
f29ff115 TH |
1077 | assert(u); |
1078 | ||
906c06f6 | 1079 | /* Nothing to do? Exit early! */ |
17f14955 | 1080 | if (apply_mask == 0) |
4ad49000 | 1081 | return; |
8e274523 | 1082 | |
52fecf20 LP |
1083 | /* Some cgroup attributes are not supported on the host root cgroup, hence silently ignore them here. And other |
1084 | * attributes should only be managed for cgroups further down the tree. */ | |
1085 | is_local_root = unit_has_name(u, SPECIAL_ROOT_SLICE); | |
1086 | is_host_root = unit_has_host_root_cgroup(u); | |
f3725e64 LP |
1087 | |
1088 | assert_se(c = unit_get_cgroup_context(u)); | |
1089 | assert_se(path = u->cgroup_path); | |
1090 | ||
52fecf20 | 1091 | if (is_local_root) /* Make sure we don't try to display messages with an empty path. */ |
6da13913 | 1092 | path = "/"; |
01efdf13 | 1093 | |
be2c0327 LP |
1094 | /* We generally ignore errors caused by read-only mounted cgroup trees (assuming we are running in a container |
1095 | * then), and missing cgroups, i.e. EROFS and ENOENT. */ | |
714e2e1d | 1096 | |
be2c0327 LP |
1097 | /* In fully unified mode these attributes don't exist on the host cgroup root. On legacy the weights exist, but |
1098 | * setting the weight makes very little sense on the host root cgroup, as there are no other cgroups at this | |
1099 | * level. The quota exists there too, but any attempt to write to it is refused with EINVAL. Inside of | |
4e1dfa45 | 1100 | * containers we want to leave control of these to the container manager (and if cgroup v2 delegation is used |
be2c0327 LP |
1101 | * we couldn't even write to them if we wanted to). */ |
1102 | if ((apply_mask & CGROUP_MASK_CPU) && !is_local_root) { | |
8e274523 | 1103 | |
b4cccbc1 | 1104 | if (cg_all_unified() > 0) { |
be2c0327 | 1105 | uint64_t weight; |
b2f8b02e | 1106 | |
be2c0327 LP |
1107 | if (cgroup_context_has_cpu_weight(c)) |
1108 | weight = cgroup_context_cpu_weight(c, state); | |
1109 | else if (cgroup_context_has_cpu_shares(c)) { | |
1110 | uint64_t shares; | |
66ebf6c0 | 1111 | |
be2c0327 LP |
1112 | shares = cgroup_context_cpu_shares(c, state); |
1113 | weight = cgroup_cpu_shares_to_weight(shares); | |
66ebf6c0 | 1114 | |
be2c0327 LP |
1115 | log_cgroup_compat(u, "Applying [Startup]CPUShares=%" PRIu64 " as [Startup]CPUWeight=%" PRIu64 " on %s", |
1116 | shares, weight, path); | |
1117 | } else | |
1118 | weight = CGROUP_WEIGHT_DEFAULT; | |
66ebf6c0 | 1119 | |
be2c0327 | 1120 | cgroup_apply_unified_cpu_weight(u, weight); |
10f28641 | 1121 | cgroup_apply_unified_cpu_quota(u, c->cpu_quota_per_sec_usec, c->cpu_quota_period_usec); |
66ebf6c0 | 1122 | |
52fecf20 | 1123 | } else { |
be2c0327 | 1124 | uint64_t shares; |
52fecf20 | 1125 | |
be2c0327 LP |
1126 | if (cgroup_context_has_cpu_weight(c)) { |
1127 | uint64_t weight; | |
52fecf20 | 1128 | |
be2c0327 LP |
1129 | weight = cgroup_context_cpu_weight(c, state); |
1130 | shares = cgroup_cpu_weight_to_shares(weight); | |
52fecf20 | 1131 | |
be2c0327 LP |
1132 | log_cgroup_compat(u, "Applying [Startup]CPUWeight=%" PRIu64 " as [Startup]CPUShares=%" PRIu64 " on %s", |
1133 | weight, shares, path); | |
1134 | } else if (cgroup_context_has_cpu_shares(c)) | |
1135 | shares = cgroup_context_cpu_shares(c, state); | |
1136 | else | |
1137 | shares = CGROUP_CPU_SHARES_DEFAULT; | |
66ebf6c0 | 1138 | |
be2c0327 | 1139 | cgroup_apply_legacy_cpu_shares(u, shares); |
10f28641 | 1140 | cgroup_apply_legacy_cpu_quota(u, c->cpu_quota_per_sec_usec, c->cpu_quota_period_usec); |
66ebf6c0 | 1141 | } |
4ad49000 LP |
1142 | } |
1143 | ||
047f5d63 | 1144 | if ((apply_mask & CGROUP_MASK_CPUSET) && !is_local_root) { |
2cea199e ZJS |
1145 | cgroup_apply_unified_cpuset(u, &c->cpuset_cpus, "cpuset.cpus"); |
1146 | cgroup_apply_unified_cpuset(u, &c->cpuset_mems, "cpuset.mems"); | |
047f5d63 PH |
1147 | } |
1148 | ||
4e1dfa45 | 1149 | /* The 'io' controller attributes are not exported on the host's root cgroup (being a pure cgroup v2 |
52fecf20 LP |
1150 | * controller), and in case of containers we want to leave control of these attributes to the container manager |
1151 | * (and we couldn't access that stuff anyway, even if we tried if proper delegation is used). */ | |
1152 | if ((apply_mask & CGROUP_MASK_IO) && !is_local_root) { | |
1153 | char buf[8+DECIMAL_STR_MAX(uint64_t)+1]; | |
1154 | bool has_io, has_blockio; | |
1155 | uint64_t weight; | |
13c31542 | 1156 | |
52fecf20 LP |
1157 | has_io = cgroup_context_has_io_config(c); |
1158 | has_blockio = cgroup_context_has_blockio_config(c); | |
13c31542 | 1159 | |
52fecf20 LP |
1160 | if (has_io) |
1161 | weight = cgroup_context_io_weight(c, state); | |
1162 | else if (has_blockio) { | |
1163 | uint64_t blkio_weight; | |
128fadc9 | 1164 | |
52fecf20 LP |
1165 | blkio_weight = cgroup_context_blkio_weight(c, state); |
1166 | weight = cgroup_weight_blkio_to_io(blkio_weight); | |
128fadc9 | 1167 | |
67e2ea15 | 1168 | log_cgroup_compat(u, "Applying [Startup]BlockIOWeight=%" PRIu64 " as [Startup]IOWeight=%" PRIu64, |
52fecf20 LP |
1169 | blkio_weight, weight); |
1170 | } else | |
1171 | weight = CGROUP_WEIGHT_DEFAULT; | |
13c31542 | 1172 | |
52fecf20 LP |
1173 | xsprintf(buf, "default %" PRIu64 "\n", weight); |
1174 | (void) set_attribute_and_warn(u, "io", "io.weight", buf); | |
538b4852 | 1175 | |
2dbc45ae KK |
1176 | /* FIXME: drop this when distro kernels properly support BFQ through "io.weight" |
1177 | * See also: https://github.com/systemd/systemd/pull/13335 */ | |
1178 | xsprintf(buf, "%" PRIu64 "\n", weight); | |
1179 | (void) set_attribute_and_warn(u, "io", "io.bfq.weight", buf); | |
1180 | ||
52fecf20 LP |
1181 | if (has_io) { |
1182 | CGroupIODeviceLatency *latency; | |
1183 | CGroupIODeviceLimit *limit; | |
1184 | CGroupIODeviceWeight *w; | |
128fadc9 | 1185 | |
52fecf20 LP |
1186 | LIST_FOREACH(device_weights, w, c->io_device_weights) |
1187 | cgroup_apply_io_device_weight(u, w->path, w->weight); | |
128fadc9 | 1188 | |
52fecf20 LP |
1189 | LIST_FOREACH(device_limits, limit, c->io_device_limits) |
1190 | cgroup_apply_io_device_limit(u, limit->path, limit->limits); | |
6ae4283c | 1191 | |
52fecf20 LP |
1192 | LIST_FOREACH(device_latencies, latency, c->io_device_latencies) |
1193 | cgroup_apply_io_device_latency(u, latency->path, latency->target_usec); | |
6ae4283c | 1194 | |
52fecf20 LP |
1195 | } else if (has_blockio) { |
1196 | CGroupBlockIODeviceWeight *w; | |
1197 | CGroupBlockIODeviceBandwidth *b; | |
13c31542 | 1198 | |
52fecf20 LP |
1199 | LIST_FOREACH(device_weights, w, c->blockio_device_weights) { |
1200 | weight = cgroup_weight_blkio_to_io(w->weight); | |
17ae2780 | 1201 | |
67e2ea15 | 1202 | log_cgroup_compat(u, "Applying BlockIODeviceWeight=%" PRIu64 " as IODeviceWeight=%" PRIu64 " for %s", |
52fecf20 | 1203 | w->weight, weight, w->path); |
538b4852 | 1204 | |
52fecf20 LP |
1205 | cgroup_apply_io_device_weight(u, w->path, weight); |
1206 | } | |
538b4852 | 1207 | |
17ae2780 | 1208 | LIST_FOREACH(device_bandwidths, b, c->blockio_device_bandwidths) { |
538b4852 TH |
1209 | uint64_t limits[_CGROUP_IO_LIMIT_TYPE_MAX]; |
1210 | CGroupIOLimitType type; | |
1211 | ||
1212 | for (type = 0; type < _CGROUP_IO_LIMIT_TYPE_MAX; type++) | |
1213 | limits[type] = cgroup_io_limit_defaults[type]; | |
1214 | ||
1215 | limits[CGROUP_IO_RBPS_MAX] = b->rbps; | |
1216 | limits[CGROUP_IO_WBPS_MAX] = b->wbps; | |
1217 | ||
67e2ea15 | 1218 | log_cgroup_compat(u, "Applying BlockIO{Read|Write}Bandwidth=%" PRIu64 " %" PRIu64 " as IO{Read|Write}BandwidthMax= for %s", |
128fadc9 TH |
1219 | b->rbps, b->wbps, b->path); |
1220 | ||
17ae2780 | 1221 | cgroup_apply_io_device_limit(u, b->path, limits); |
538b4852 | 1222 | } |
13c31542 TH |
1223 | } |
1224 | } | |
1225 | ||
906c06f6 | 1226 | if (apply_mask & CGROUP_MASK_BLKIO) { |
52fecf20 | 1227 | bool has_io, has_blockio; |
4ad49000 | 1228 | |
52fecf20 LP |
1229 | has_io = cgroup_context_has_io_config(c); |
1230 | has_blockio = cgroup_context_has_blockio_config(c); | |
1231 | ||
1232 | /* Applying a 'weight' never makes sense for the host root cgroup, and for containers this should be | |
1233 | * left to our container manager, too. */ | |
1234 | if (!is_local_root) { | |
64faf04c TH |
1235 | char buf[DECIMAL_STR_MAX(uint64_t)+1]; |
1236 | uint64_t weight; | |
64faf04c | 1237 | |
7d862ab8 | 1238 | if (has_io) { |
52fecf20 | 1239 | uint64_t io_weight; |
128fadc9 | 1240 | |
52fecf20 | 1241 | io_weight = cgroup_context_io_weight(c, state); |
538b4852 | 1242 | weight = cgroup_weight_io_to_blkio(cgroup_context_io_weight(c, state)); |
128fadc9 | 1243 | |
67e2ea15 | 1244 | log_cgroup_compat(u, "Applying [Startup]IOWeight=%" PRIu64 " as [Startup]BlockIOWeight=%" PRIu64, |
128fadc9 | 1245 | io_weight, weight); |
7d862ab8 TH |
1246 | } else if (has_blockio) |
1247 | weight = cgroup_context_blkio_weight(c, state); | |
1248 | else | |
538b4852 | 1249 | weight = CGROUP_BLKIO_WEIGHT_DEFAULT; |
64faf04c TH |
1250 | |
1251 | xsprintf(buf, "%" PRIu64 "\n", weight); | |
293d32df | 1252 | (void) set_attribute_and_warn(u, "blkio", "blkio.weight", buf); |
4ad49000 | 1253 | |
7d862ab8 | 1254 | if (has_io) { |
538b4852 TH |
1255 | CGroupIODeviceWeight *w; |
1256 | ||
128fadc9 TH |
1257 | LIST_FOREACH(device_weights, w, c->io_device_weights) { |
1258 | weight = cgroup_weight_io_to_blkio(w->weight); | |
1259 | ||
67e2ea15 | 1260 | log_cgroup_compat(u, "Applying IODeviceWeight=%" PRIu64 " as BlockIODeviceWeight=%" PRIu64 " for %s", |
128fadc9 TH |
1261 | w->weight, weight, w->path); |
1262 | ||
1263 | cgroup_apply_blkio_device_weight(u, w->path, weight); | |
1264 | } | |
7d862ab8 TH |
1265 | } else if (has_blockio) { |
1266 | CGroupBlockIODeviceWeight *w; | |
1267 | ||
7d862ab8 TH |
1268 | LIST_FOREACH(device_weights, w, c->blockio_device_weights) |
1269 | cgroup_apply_blkio_device_weight(u, w->path, w->weight); | |
538b4852 | 1270 | } |
4ad49000 LP |
1271 | } |
1272 | ||
5238e957 | 1273 | /* The bandwidth limits are something that make sense to be applied to the host's root but not container |
52fecf20 LP |
1274 | * roots, as there we want the container manager to handle it */ |
1275 | if (is_host_root || !is_local_root) { | |
1276 | if (has_io) { | |
1277 | CGroupIODeviceLimit *l; | |
538b4852 | 1278 | |
52fecf20 | 1279 | LIST_FOREACH(device_limits, l, c->io_device_limits) { |
67e2ea15 | 1280 | log_cgroup_compat(u, "Applying IO{Read|Write}Bandwidth=%" PRIu64 " %" PRIu64 " as BlockIO{Read|Write}BandwidthMax= for %s", |
52fecf20 | 1281 | l->limits[CGROUP_IO_RBPS_MAX], l->limits[CGROUP_IO_WBPS_MAX], l->path); |
128fadc9 | 1282 | |
52fecf20 LP |
1283 | cgroup_apply_blkio_device_limit(u, l->path, l->limits[CGROUP_IO_RBPS_MAX], l->limits[CGROUP_IO_WBPS_MAX]); |
1284 | } | |
1285 | } else if (has_blockio) { | |
1286 | CGroupBlockIODeviceBandwidth *b; | |
7d862ab8 | 1287 | |
52fecf20 LP |
1288 | LIST_FOREACH(device_bandwidths, b, c->blockio_device_bandwidths) |
1289 | cgroup_apply_blkio_device_limit(u, b->path, b->rbps, b->wbps); | |
1290 | } | |
d686d8a9 | 1291 | } |
8e274523 LP |
1292 | } |
1293 | ||
be2c0327 LP |
1294 | /* In unified mode 'memory' attributes do not exist on the root cgroup. In legacy mode 'memory.limit_in_bytes' |
1295 | * exists on the root cgroup, but any writes to it are refused with EINVAL. And if we run in a container we | |
4e1dfa45 | 1296 | * want to leave control to the container manager (and if proper cgroup v2 delegation is used we couldn't even |
be2c0327 LP |
1297 | * write to this if we wanted to.) */ |
1298 | if ((apply_mask & CGROUP_MASK_MEMORY) && !is_local_root) { | |
efdb0237 | 1299 | |
52fecf20 | 1300 | if (cg_all_unified() > 0) { |
be2c0327 LP |
1301 | uint64_t max, swap_max = CGROUP_LIMIT_MAX; |
1302 | ||
c52db42b | 1303 | if (unit_has_unified_memory_config(u)) { |
be2c0327 LP |
1304 | max = c->memory_max; |
1305 | swap_max = c->memory_swap_max; | |
1306 | } else { | |
1307 | max = c->memory_limit; | |
efdb0237 | 1308 | |
be2c0327 LP |
1309 | if (max != CGROUP_LIMIT_MAX) |
1310 | log_cgroup_compat(u, "Applying MemoryLimit=%" PRIu64 " as MemoryMax=", max); | |
128fadc9 | 1311 | } |
da4d897e | 1312 | |
64fe532e | 1313 | cgroup_apply_unified_memory_limit(u, "memory.min", unit_get_ancestor_memory_min(u)); |
c52db42b | 1314 | cgroup_apply_unified_memory_limit(u, "memory.low", unit_get_ancestor_memory_low(u)); |
be2c0327 LP |
1315 | cgroup_apply_unified_memory_limit(u, "memory.high", c->memory_high); |
1316 | cgroup_apply_unified_memory_limit(u, "memory.max", max); | |
1317 | cgroup_apply_unified_memory_limit(u, "memory.swap.max", swap_max); | |
128fadc9 | 1318 | |
afcfaa69 LP |
1319 | (void) set_attribute_and_warn(u, "memory", "memory.oom.group", one_zero(c->memory_oom_group)); |
1320 | ||
be2c0327 LP |
1321 | } else { |
1322 | char buf[DECIMAL_STR_MAX(uint64_t) + 1]; | |
1323 | uint64_t val; | |
52fecf20 | 1324 | |
c52db42b | 1325 | if (unit_has_unified_memory_config(u)) { |
be2c0327 LP |
1326 | val = c->memory_max; |
1327 | log_cgroup_compat(u, "Applying MemoryMax=%" PRIi64 " as MemoryLimit=", val); | |
1328 | } else | |
1329 | val = c->memory_limit; | |
78a4ee59 | 1330 | |
be2c0327 LP |
1331 | if (val == CGROUP_LIMIT_MAX) |
1332 | strncpy(buf, "-1\n", sizeof(buf)); | |
1333 | else | |
1334 | xsprintf(buf, "%" PRIu64 "\n", val); | |
1335 | ||
1336 | (void) set_attribute_and_warn(u, "memory", "memory.limit_in_bytes", buf); | |
da4d897e | 1337 | } |
4ad49000 | 1338 | } |
8e274523 | 1339 | |
4e1dfa45 | 1340 | /* On cgroup v2 we can apply BPF everywhere. On cgroup v1 we apply it everywhere except for the root of |
52fecf20 LP |
1341 | * containers, where we leave this to the manager */ |
1342 | if ((apply_mask & (CGROUP_MASK_DEVICES | CGROUP_MASK_BPF_DEVICES)) && | |
8b139557 ZJS |
1343 | (is_host_root || cg_all_unified() > 0 || !is_local_root)) |
1344 | (void) cgroup_apply_devices(u); | |
03a7b521 | 1345 | |
00b5974f LP |
1346 | if (apply_mask & CGROUP_MASK_PIDS) { |
1347 | ||
52fecf20 | 1348 | if (is_host_root) { |
00b5974f LP |
1349 | /* So, the "pids" controller does not expose anything on the root cgroup, in order not to |
1350 | * replicate knobs exposed elsewhere needlessly. We abstract this away here however, and when | |
1351 | * the knobs of the root cgroup are modified propagate this to the relevant sysctls. There's a | |
1352 | * non-obvious asymmetry however: unlike the cgroup properties we don't really want to take | |
1353 | * exclusive ownership of the sysctls, but we still want to honour things if the user sets | |
1354 | * limits. Hence we employ sort of a one-way strategy: when the user sets a bounded limit | |
1355 | * through us it counts. When the user afterwards unsets it again (i.e. sets it to unbounded) | |
1356 | * it also counts. But if the user never set a limit through us (i.e. we are the default of | |
1357 | * "unbounded") we leave things unmodified. For this we manage a global boolean that we turn on | |
1358 | * the first time we set a limit. Note that this boolean is flushed out on manager reload, | |
5238e957 | 1359 | * which is desirable so that there's an official way to release control of the sysctl from |
00b5974f LP |
1360 | * systemd: set the limit to unbounded and reload. */ |
1361 | ||
3a0f06c4 | 1362 | if (tasks_max_isset(&c->tasks_max)) { |
00b5974f | 1363 | u->manager->sysctl_pid_max_changed = true; |
3a0f06c4 | 1364 | r = procfs_tasks_set_limit(tasks_max_resolve(&c->tasks_max)); |
00b5974f LP |
1365 | } else if (u->manager->sysctl_pid_max_changed) |
1366 | r = procfs_tasks_set_limit(TASKS_MAX); | |
1367 | else | |
1368 | r = 0; | |
00b5974f | 1369 | if (r < 0) |
39b9fefb | 1370 | log_unit_full(u, LOG_LEVEL_CGROUP_WRITE(r), r, |
00b5974f | 1371 | "Failed to write to tasks limit sysctls: %m"); |
52fecf20 | 1372 | } |
03a7b521 | 1373 | |
52fecf20 LP |
1374 | /* The attribute itself is not available on the host root cgroup, and in the container case we want to |
1375 | * leave it for the container manager. */ | |
1376 | if (!is_local_root) { | |
3a0f06c4 ZJS |
1377 | if (tasks_max_isset(&c->tasks_max)) { |
1378 | char buf[DECIMAL_STR_MAX(uint64_t) + 1]; | |
03a7b521 | 1379 | |
3a0f06c4 | 1380 | xsprintf(buf, "%" PRIu64 "\n", tasks_max_resolve(&c->tasks_max)); |
293d32df | 1381 | (void) set_attribute_and_warn(u, "pids", "pids.max", buf); |
00b5974f | 1382 | } else |
589a5f7a | 1383 | (void) set_attribute_and_warn(u, "pids", "pids.max", "max\n"); |
00b5974f | 1384 | } |
03a7b521 | 1385 | } |
906c06f6 | 1386 | |
17f14955 | 1387 | if (apply_mask & CGROUP_MASK_BPF_FIREWALL) |
0f2d84d2 | 1388 | cgroup_apply_firewall(u); |
fb385181 LP |
1389 | } |
1390 | ||
16492445 LP |
1391 | static bool unit_get_needs_bpf_firewall(Unit *u) { |
1392 | CGroupContext *c; | |
1393 | Unit *p; | |
1394 | assert(u); | |
1395 | ||
1396 | c = unit_get_cgroup_context(u); | |
1397 | if (!c) | |
1398 | return false; | |
1399 | ||
1400 | if (c->ip_accounting || | |
1401 | c->ip_address_allow || | |
fab34748 KL |
1402 | c->ip_address_deny || |
1403 | c->ip_filters_ingress || | |
1404 | c->ip_filters_egress) | |
16492445 LP |
1405 | return true; |
1406 | ||
1407 | /* If any parent slice has an IP access list defined, it applies too */ | |
1408 | for (p = UNIT_DEREF(u->slice); p; p = UNIT_DEREF(p->slice)) { | |
1409 | c = unit_get_cgroup_context(p); | |
1410 | if (!c) | |
1411 | return false; | |
1412 | ||
1413 | if (c->ip_address_allow || | |
1414 | c->ip_address_deny) | |
1415 | return true; | |
1416 | } | |
1417 | ||
1418 | return false; | |
1419 | } | |
1420 | ||
c52db42b | 1421 | static CGroupMask unit_get_cgroup_mask(Unit *u) { |
efdb0237 | 1422 | CGroupMask mask = 0; |
c52db42b CD |
1423 | CGroupContext *c; |
1424 | ||
1425 | assert(u); | |
1426 | ||
1427 | c = unit_get_cgroup_context(u); | |
8e274523 | 1428 | |
c710d3b4 CD |
1429 | assert(c); |
1430 | ||
fae9bc29 | 1431 | /* Figure out which controllers we need, based on the cgroup context object */ |
8e274523 | 1432 | |
fae9bc29 | 1433 | if (c->cpu_accounting) |
f98c2585 | 1434 | mask |= get_cpu_accounting_mask(); |
fae9bc29 LP |
1435 | |
1436 | if (cgroup_context_has_cpu_weight(c) || | |
66ebf6c0 | 1437 | cgroup_context_has_cpu_shares(c) || |
3a43da28 | 1438 | c->cpu_quota_per_sec_usec != USEC_INFINITY) |
fae9bc29 | 1439 | mask |= CGROUP_MASK_CPU; |
ecedd90f | 1440 | |
047f5d63 PH |
1441 | if (c->cpuset_cpus.set || c->cpuset_mems.set) |
1442 | mask |= CGROUP_MASK_CPUSET; | |
1443 | ||
538b4852 TH |
1444 | if (cgroup_context_has_io_config(c) || cgroup_context_has_blockio_config(c)) |
1445 | mask |= CGROUP_MASK_IO | CGROUP_MASK_BLKIO; | |
ecedd90f | 1446 | |
4ad49000 | 1447 | if (c->memory_accounting || |
da4d897e | 1448 | c->memory_limit != CGROUP_LIMIT_MAX || |
c52db42b | 1449 | unit_has_unified_memory_config(u)) |
efdb0237 | 1450 | mask |= CGROUP_MASK_MEMORY; |
8e274523 | 1451 | |
a931ad47 | 1452 | if (c->device_allow || |
084870f9 | 1453 | c->device_policy != CGROUP_DEVICE_POLICY_AUTO) |
084c7007 | 1454 | mask |= CGROUP_MASK_DEVICES | CGROUP_MASK_BPF_DEVICES; |
4ad49000 | 1455 | |
03a7b521 | 1456 | if (c->tasks_accounting || |
3a0f06c4 | 1457 | tasks_max_isset(&c->tasks_max)) |
03a7b521 LP |
1458 | mask |= CGROUP_MASK_PIDS; |
1459 | ||
fae9bc29 | 1460 | return CGROUP_MASK_EXTEND_JOINED(mask); |
8e274523 LP |
1461 | } |
1462 | ||
53aea74a | 1463 | static CGroupMask unit_get_bpf_mask(Unit *u) { |
17f14955 RG |
1464 | CGroupMask mask = 0; |
1465 | ||
fae9bc29 LP |
1466 | /* Figure out which controllers we need, based on the cgroup context, possibly taking into account children |
1467 | * too. */ | |
1468 | ||
17f14955 RG |
1469 | if (unit_get_needs_bpf_firewall(u)) |
1470 | mask |= CGROUP_MASK_BPF_FIREWALL; | |
1471 | ||
1472 | return mask; | |
1473 | } | |
1474 | ||
efdb0237 | 1475 | CGroupMask unit_get_own_mask(Unit *u) { |
4ad49000 | 1476 | CGroupContext *c; |
8e274523 | 1477 | |
442ce775 LP |
1478 | /* Returns the mask of controllers the unit needs for itself. If a unit is not properly loaded, return an empty |
1479 | * mask, as we shouldn't reflect it in the cgroup hierarchy then. */ | |
1480 | ||
1481 | if (u->load_state != UNIT_LOADED) | |
1482 | return 0; | |
efdb0237 | 1483 | |
4ad49000 LP |
1484 | c = unit_get_cgroup_context(u); |
1485 | if (!c) | |
1486 | return 0; | |
8e274523 | 1487 | |
c52db42b | 1488 | return (unit_get_cgroup_mask(u) | unit_get_bpf_mask(u) | unit_get_delegate_mask(u)) & ~unit_get_ancestor_disable_mask(u); |
02638280 LP |
1489 | } |
1490 | ||
1491 | CGroupMask unit_get_delegate_mask(Unit *u) { | |
1492 | CGroupContext *c; | |
1493 | ||
1494 | /* If delegation is turned on, then turn on selected controllers, unless we are on the legacy hierarchy and the | |
1495 | * process we fork into is known to drop privileges, and hence shouldn't get access to the controllers. | |
19af675e | 1496 | * |
02638280 | 1497 | * Note that on the unified hierarchy it is safe to delegate controllers to unprivileged services. */ |
a931ad47 | 1498 | |
1d9cc876 | 1499 | if (!unit_cgroup_delegate(u)) |
02638280 LP |
1500 | return 0; |
1501 | ||
1502 | if (cg_all_unified() <= 0) { | |
a931ad47 LP |
1503 | ExecContext *e; |
1504 | ||
1505 | e = unit_get_exec_context(u); | |
02638280 LP |
1506 | if (e && !exec_context_maintains_privileges(e)) |
1507 | return 0; | |
a931ad47 LP |
1508 | } |
1509 | ||
1d9cc876 | 1510 | assert_se(c = unit_get_cgroup_context(u)); |
fae9bc29 | 1511 | return CGROUP_MASK_EXTEND_JOINED(c->delegate_controllers); |
8e274523 LP |
1512 | } |
1513 | ||
efdb0237 | 1514 | CGroupMask unit_get_members_mask(Unit *u) { |
4ad49000 | 1515 | assert(u); |
bc432dc7 | 1516 | |
02638280 | 1517 | /* Returns the mask of controllers all of the unit's children require, merged */ |
efdb0237 | 1518 | |
bc432dc7 | 1519 | if (u->cgroup_members_mask_valid) |
26a17ca2 | 1520 | return u->cgroup_members_mask; /* Use cached value if possible */ |
bc432dc7 | 1521 | |
64e844e5 | 1522 | u->cgroup_members_mask = 0; |
bc432dc7 LP |
1523 | |
1524 | if (u->type == UNIT_SLICE) { | |
eef85c4a | 1525 | void *v; |
bc432dc7 LP |
1526 | Unit *member; |
1527 | Iterator i; | |
1528 | ||
eef85c4a | 1529 | HASHMAP_FOREACH_KEY(v, member, u->dependencies[UNIT_BEFORE], i) { |
cb5e3bc3 CD |
1530 | if (UNIT_DEREF(member->slice) == u) |
1531 | u->cgroup_members_mask |= unit_get_subtree_mask(member); /* note that this calls ourselves again, for the children */ | |
bc432dc7 LP |
1532 | } |
1533 | } | |
1534 | ||
1535 | u->cgroup_members_mask_valid = true; | |
6414b7c9 | 1536 | return u->cgroup_members_mask; |
246aa6dd LP |
1537 | } |
1538 | ||
efdb0237 | 1539 | CGroupMask unit_get_siblings_mask(Unit *u) { |
4ad49000 | 1540 | assert(u); |
246aa6dd | 1541 | |
efdb0237 LP |
1542 | /* Returns the mask of controllers all of the unit's siblings |
1543 | * require, i.e. the members mask of the unit's parent slice | |
1544 | * if there is one. */ | |
1545 | ||
bc432dc7 | 1546 | if (UNIT_ISSET(u->slice)) |
637f421e | 1547 | return unit_get_members_mask(UNIT_DEREF(u->slice)); |
4ad49000 | 1548 | |
64e844e5 | 1549 | return unit_get_subtree_mask(u); /* we are the top-level slice */ |
246aa6dd LP |
1550 | } |
1551 | ||
4f6f62e4 CD |
1552 | CGroupMask unit_get_disable_mask(Unit *u) { |
1553 | CGroupContext *c; | |
1554 | ||
1555 | c = unit_get_cgroup_context(u); | |
1556 | if (!c) | |
1557 | return 0; | |
1558 | ||
1559 | return c->disable_controllers; | |
1560 | } | |
1561 | ||
1562 | CGroupMask unit_get_ancestor_disable_mask(Unit *u) { | |
1563 | CGroupMask mask; | |
1564 | ||
1565 | assert(u); | |
1566 | mask = unit_get_disable_mask(u); | |
1567 | ||
1568 | /* Returns the mask of controllers which are marked as forcibly | |
1569 | * disabled in any ancestor unit or the unit in question. */ | |
1570 | ||
1571 | if (UNIT_ISSET(u->slice)) | |
1572 | mask |= unit_get_ancestor_disable_mask(UNIT_DEREF(u->slice)); | |
1573 | ||
1574 | return mask; | |
1575 | } | |
1576 | ||
efdb0237 LP |
1577 | CGroupMask unit_get_subtree_mask(Unit *u) { |
1578 | ||
1579 | /* Returns the mask of this subtree, meaning of the group | |
1580 | * itself and its children. */ | |
1581 | ||
1582 | return unit_get_own_mask(u) | unit_get_members_mask(u); | |
1583 | } | |
1584 | ||
1585 | CGroupMask unit_get_target_mask(Unit *u) { | |
1586 | CGroupMask mask; | |
1587 | ||
1588 | /* This returns the cgroup mask of all controllers to enable | |
1589 | * for a specific cgroup, i.e. everything it needs itself, | |
1590 | * plus all that its children need, plus all that its siblings | |
1591 | * need. This is primarily useful on the legacy cgroup | |
1592 | * hierarchy, where we need to duplicate each cgroup in each | |
1593 | * hierarchy that shall be enabled for it. */ | |
6414b7c9 | 1594 | |
efdb0237 | 1595 | mask = unit_get_own_mask(u) | unit_get_members_mask(u) | unit_get_siblings_mask(u); |
84d2744b ZJS |
1596 | |
1597 | if (mask & CGROUP_MASK_BPF_FIREWALL & ~u->manager->cgroup_supported) | |
1598 | emit_bpf_firewall_warning(u); | |
1599 | ||
efdb0237 | 1600 | mask &= u->manager->cgroup_supported; |
c72703e2 | 1601 | mask &= ~unit_get_ancestor_disable_mask(u); |
efdb0237 LP |
1602 | |
1603 | return mask; | |
1604 | } | |
1605 | ||
1606 | CGroupMask unit_get_enable_mask(Unit *u) { | |
1607 | CGroupMask mask; | |
1608 | ||
1609 | /* This returns the cgroup mask of all controllers to enable | |
1610 | * for the children of a specific cgroup. This is primarily | |
1611 | * useful for the unified cgroup hierarchy, where each cgroup | |
1612 | * controls which controllers are enabled for its children. */ | |
1613 | ||
1614 | mask = unit_get_members_mask(u); | |
6414b7c9 | 1615 | mask &= u->manager->cgroup_supported; |
c72703e2 | 1616 | mask &= ~unit_get_ancestor_disable_mask(u); |
6414b7c9 DS |
1617 | |
1618 | return mask; | |
1619 | } | |
1620 | ||
5af88058 | 1621 | void unit_invalidate_cgroup_members_masks(Unit *u) { |
bc432dc7 LP |
1622 | assert(u); |
1623 | ||
5af88058 LP |
1624 | /* Recurse invalidate the member masks cache all the way up the tree */ |
1625 | u->cgroup_members_mask_valid = false; | |
bc432dc7 | 1626 | |
5af88058 LP |
1627 | if (UNIT_ISSET(u->slice)) |
1628 | unit_invalidate_cgroup_members_masks(UNIT_DEREF(u->slice)); | |
6414b7c9 DS |
1629 | } |
1630 | ||
6592b975 | 1631 | const char *unit_get_realized_cgroup_path(Unit *u, CGroupMask mask) { |
03b90d4b | 1632 | |
6592b975 | 1633 | /* Returns the realized cgroup path of the specified unit where all specified controllers are available. */ |
03b90d4b LP |
1634 | |
1635 | while (u) { | |
6592b975 | 1636 | |
03b90d4b LP |
1637 | if (u->cgroup_path && |
1638 | u->cgroup_realized && | |
d94a24ca | 1639 | FLAGS_SET(u->cgroup_realized_mask, mask)) |
03b90d4b LP |
1640 | return u->cgroup_path; |
1641 | ||
1642 | u = UNIT_DEREF(u->slice); | |
1643 | } | |
1644 | ||
1645 | return NULL; | |
1646 | } | |
1647 | ||
6592b975 LP |
1648 | static const char *migrate_callback(CGroupMask mask, void *userdata) { |
1649 | return unit_get_realized_cgroup_path(userdata, mask); | |
1650 | } | |
1651 | ||
303ee601 | 1652 | char *unit_default_cgroup_path(const Unit *u) { |
efdb0237 LP |
1653 | _cleanup_free_ char *escaped = NULL, *slice = NULL; |
1654 | int r; | |
1655 | ||
1656 | assert(u); | |
1657 | ||
1658 | if (unit_has_name(u, SPECIAL_ROOT_SLICE)) | |
1659 | return strdup(u->manager->cgroup_root); | |
1660 | ||
1661 | if (UNIT_ISSET(u->slice) && !unit_has_name(UNIT_DEREF(u->slice), SPECIAL_ROOT_SLICE)) { | |
1662 | r = cg_slice_to_path(UNIT_DEREF(u->slice)->id, &slice); | |
1663 | if (r < 0) | |
1664 | return NULL; | |
1665 | } | |
1666 | ||
1667 | escaped = cg_escape(u->id); | |
1668 | if (!escaped) | |
1669 | return NULL; | |
1670 | ||
657ee2d8 | 1671 | return path_join(empty_to_root(u->manager->cgroup_root), slice, escaped); |
efdb0237 LP |
1672 | } |
1673 | ||
1674 | int unit_set_cgroup_path(Unit *u, const char *path) { | |
1675 | _cleanup_free_ char *p = NULL; | |
1676 | int r; | |
1677 | ||
1678 | assert(u); | |
1679 | ||
5210387e LP |
1680 | if (streq_ptr(u->cgroup_path, path)) |
1681 | return 0; | |
1682 | ||
efdb0237 LP |
1683 | if (path) { |
1684 | p = strdup(path); | |
1685 | if (!p) | |
1686 | return -ENOMEM; | |
5210387e | 1687 | } |
efdb0237 LP |
1688 | |
1689 | if (p) { | |
1690 | r = hashmap_put(u->manager->cgroup_unit, p, u); | |
1691 | if (r < 0) | |
1692 | return r; | |
1693 | } | |
1694 | ||
1695 | unit_release_cgroup(u); | |
ae2a15bc | 1696 | u->cgroup_path = TAKE_PTR(p); |
efdb0237 LP |
1697 | |
1698 | return 1; | |
1699 | } | |
1700 | ||
1701 | int unit_watch_cgroup(Unit *u) { | |
ab2c3861 | 1702 | _cleanup_free_ char *events = NULL; |
efdb0237 LP |
1703 | int r; |
1704 | ||
1705 | assert(u); | |
1706 | ||
0bb814c2 LP |
1707 | /* Watches the "cgroups.events" attribute of this unit's cgroup for "empty" events, but only if |
1708 | * cgroupv2 is available. */ | |
1709 | ||
efdb0237 LP |
1710 | if (!u->cgroup_path) |
1711 | return 0; | |
1712 | ||
0bb814c2 | 1713 | if (u->cgroup_control_inotify_wd >= 0) |
efdb0237 LP |
1714 | return 0; |
1715 | ||
1716 | /* Only applies to the unified hierarchy */ | |
c22800e4 | 1717 | r = cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER); |
b4cccbc1 LP |
1718 | if (r < 0) |
1719 | return log_error_errno(r, "Failed to determine whether the name=systemd hierarchy is unified: %m"); | |
1720 | if (r == 0) | |
efdb0237 LP |
1721 | return 0; |
1722 | ||
0bb814c2 | 1723 | /* No point in watch the top-level slice, it's never going to run empty. */ |
efdb0237 LP |
1724 | if (unit_has_name(u, SPECIAL_ROOT_SLICE)) |
1725 | return 0; | |
1726 | ||
0bb814c2 | 1727 | r = hashmap_ensure_allocated(&u->manager->cgroup_control_inotify_wd_unit, &trivial_hash_ops); |
efdb0237 LP |
1728 | if (r < 0) |
1729 | return log_oom(); | |
1730 | ||
ab2c3861 | 1731 | r = cg_get_path(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path, "cgroup.events", &events); |
efdb0237 LP |
1732 | if (r < 0) |
1733 | return log_oom(); | |
1734 | ||
0bb814c2 LP |
1735 | u->cgroup_control_inotify_wd = inotify_add_watch(u->manager->cgroup_inotify_fd, events, IN_MODIFY); |
1736 | if (u->cgroup_control_inotify_wd < 0) { | |
efdb0237 | 1737 | |
0bb814c2 LP |
1738 | if (errno == ENOENT) /* If the directory is already gone we don't need to track it, so this |
1739 | * is not an error */ | |
efdb0237 LP |
1740 | return 0; |
1741 | ||
0bb814c2 | 1742 | return log_unit_error_errno(u, errno, "Failed to add control inotify watch descriptor for control group %s: %m", u->cgroup_path); |
efdb0237 LP |
1743 | } |
1744 | ||
0bb814c2 | 1745 | r = hashmap_put(u->manager->cgroup_control_inotify_wd_unit, INT_TO_PTR(u->cgroup_control_inotify_wd), u); |
efdb0237 | 1746 | if (r < 0) |
0bb814c2 | 1747 | return log_unit_error_errno(u, r, "Failed to add control inotify watch descriptor to hash map: %m"); |
efdb0237 LP |
1748 | |
1749 | return 0; | |
1750 | } | |
1751 | ||
afcfaa69 LP |
1752 | int unit_watch_cgroup_memory(Unit *u) { |
1753 | _cleanup_free_ char *events = NULL; | |
1754 | CGroupContext *c; | |
1755 | int r; | |
1756 | ||
1757 | assert(u); | |
1758 | ||
1759 | /* Watches the "memory.events" attribute of this unit's cgroup for "oom_kill" events, but only if | |
1760 | * cgroupv2 is available. */ | |
1761 | ||
1762 | if (!u->cgroup_path) | |
1763 | return 0; | |
1764 | ||
1765 | c = unit_get_cgroup_context(u); | |
1766 | if (!c) | |
1767 | return 0; | |
1768 | ||
1769 | /* The "memory.events" attribute is only available if the memory controller is on. Let's hence tie | |
1770 | * this to memory accounting, in a way watching for OOM kills is a form of memory accounting after | |
1771 | * all. */ | |
1772 | if (!c->memory_accounting) | |
1773 | return 0; | |
1774 | ||
1775 | /* Don't watch inner nodes, as the kernel doesn't report oom_kill events recursively currently, and | |
1776 | * we also don't want to generate a log message for each parent cgroup of a process. */ | |
1777 | if (u->type == UNIT_SLICE) | |
1778 | return 0; | |
1779 | ||
1780 | if (u->cgroup_memory_inotify_wd >= 0) | |
1781 | return 0; | |
1782 | ||
1783 | /* Only applies to the unified hierarchy */ | |
1784 | r = cg_all_unified(); | |
1785 | if (r < 0) | |
1786 | return log_error_errno(r, "Failed to determine whether the memory controller is unified: %m"); | |
1787 | if (r == 0) | |
1788 | return 0; | |
1789 | ||
1790 | r = hashmap_ensure_allocated(&u->manager->cgroup_memory_inotify_wd_unit, &trivial_hash_ops); | |
1791 | if (r < 0) | |
1792 | return log_oom(); | |
1793 | ||
1794 | r = cg_get_path(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path, "memory.events", &events); | |
1795 | if (r < 0) | |
1796 | return log_oom(); | |
1797 | ||
1798 | u->cgroup_memory_inotify_wd = inotify_add_watch(u->manager->cgroup_inotify_fd, events, IN_MODIFY); | |
1799 | if (u->cgroup_memory_inotify_wd < 0) { | |
1800 | ||
1801 | if (errno == ENOENT) /* If the directory is already gone we don't need to track it, so this | |
1802 | * is not an error */ | |
1803 | return 0; | |
1804 | ||
1805 | return log_unit_error_errno(u, errno, "Failed to add memory inotify watch descriptor for control group %s: %m", u->cgroup_path); | |
1806 | } | |
1807 | ||
1808 | r = hashmap_put(u->manager->cgroup_memory_inotify_wd_unit, INT_TO_PTR(u->cgroup_memory_inotify_wd), u); | |
1809 | if (r < 0) | |
1810 | return log_unit_error_errno(u, r, "Failed to add memory inotify watch descriptor to hash map: %m"); | |
1811 | ||
1812 | return 0; | |
1813 | } | |
1814 | ||
a4634b21 LP |
1815 | int unit_pick_cgroup_path(Unit *u) { |
1816 | _cleanup_free_ char *path = NULL; | |
1817 | int r; | |
1818 | ||
1819 | assert(u); | |
1820 | ||
1821 | if (u->cgroup_path) | |
1822 | return 0; | |
1823 | ||
1824 | if (!UNIT_HAS_CGROUP_CONTEXT(u)) | |
1825 | return -EINVAL; | |
1826 | ||
1827 | path = unit_default_cgroup_path(u); | |
1828 | if (!path) | |
1829 | return log_oom(); | |
1830 | ||
1831 | r = unit_set_cgroup_path(u, path); | |
1832 | if (r == -EEXIST) | |
1833 | return log_unit_error_errno(u, r, "Control group %s exists already.", path); | |
1834 | if (r < 0) | |
1835 | return log_unit_error_errno(u, r, "Failed to set unit's control group path to %s: %m", path); | |
1836 | ||
1837 | return 0; | |
1838 | } | |
1839 | ||
efdb0237 LP |
1840 | static int unit_create_cgroup( |
1841 | Unit *u, | |
1842 | CGroupMask target_mask, | |
0d2d6fbf CD |
1843 | CGroupMask enable_mask, |
1844 | ManagerState state) { | |
efdb0237 | 1845 | |
65be7e06 | 1846 | bool created; |
27adcc97 | 1847 | int r; |
64747e2d | 1848 | |
4ad49000 | 1849 | assert(u); |
64747e2d | 1850 | |
27c4ed79 | 1851 | if (!UNIT_HAS_CGROUP_CONTEXT(u)) |
0cd385d3 LP |
1852 | return 0; |
1853 | ||
a4634b21 LP |
1854 | /* Figure out our cgroup path */ |
1855 | r = unit_pick_cgroup_path(u); | |
1856 | if (r < 0) | |
1857 | return r; | |
b58b8e11 | 1858 | |
03b90d4b | 1859 | /* First, create our own group */ |
efdb0237 | 1860 | r = cg_create_everywhere(u->manager->cgroup_supported, target_mask, u->cgroup_path); |
23bbb0de | 1861 | if (r < 0) |
efdb0237 | 1862 | return log_unit_error_errno(u, r, "Failed to create cgroup %s: %m", u->cgroup_path); |
490c5a37 | 1863 | created = r; |
efdb0237 LP |
1864 | |
1865 | /* Start watching it */ | |
1866 | (void) unit_watch_cgroup(u); | |
afcfaa69 | 1867 | (void) unit_watch_cgroup_memory(u); |
efdb0237 | 1868 | |
65be7e06 | 1869 | /* Preserve enabled controllers in delegated units, adjust others. */ |
1fd3a10c | 1870 | if (created || !u->cgroup_realized || !unit_cgroup_delegate(u)) { |
27adcc97 | 1871 | CGroupMask result_mask = 0; |
65be7e06 ZJS |
1872 | |
1873 | /* Enable all controllers we need */ | |
27adcc97 | 1874 | r = cg_enable_everywhere(u->manager->cgroup_supported, enable_mask, u->cgroup_path, &result_mask); |
65be7e06 | 1875 | if (r < 0) |
27adcc97 LP |
1876 | log_unit_warning_errno(u, r, "Failed to enable/disable controllers on cgroup %s, ignoring: %m", u->cgroup_path); |
1877 | ||
1878 | /* If we just turned off a controller, this might release the controller for our parent too, let's | |
1879 | * enqueue the parent for re-realization in that case again. */ | |
1880 | if (UNIT_ISSET(u->slice)) { | |
1881 | CGroupMask turned_off; | |
1882 | ||
1883 | turned_off = (u->cgroup_realized ? u->cgroup_enabled_mask & ~result_mask : 0); | |
1884 | if (turned_off != 0) { | |
1885 | Unit *parent; | |
1886 | ||
1887 | /* Force the parent to propagate the enable mask to the kernel again, by invalidating | |
1888 | * the controller we just turned off. */ | |
1889 | ||
1890 | for (parent = UNIT_DEREF(u->slice); parent; parent = UNIT_DEREF(parent->slice)) | |
1891 | unit_invalidate_cgroup(parent, turned_off); | |
1892 | } | |
1893 | } | |
1894 | ||
1895 | /* Remember what's actually enabled now */ | |
1896 | u->cgroup_enabled_mask = result_mask; | |
65be7e06 | 1897 | } |
03b90d4b LP |
1898 | |
1899 | /* Keep track that this is now realized */ | |
4ad49000 | 1900 | u->cgroup_realized = true; |
efdb0237 | 1901 | u->cgroup_realized_mask = target_mask; |
4ad49000 | 1902 | |
1d9cc876 | 1903 | if (u->type != UNIT_SLICE && !unit_cgroup_delegate(u)) { |
0cd385d3 LP |
1904 | |
1905 | /* Then, possibly move things over, but not if | |
1906 | * subgroups may contain processes, which is the case | |
1907 | * for slice and delegation units. */ | |
1908 | r = cg_migrate_everywhere(u->manager->cgroup_supported, u->cgroup_path, u->cgroup_path, migrate_callback, u); | |
1909 | if (r < 0) | |
efdb0237 | 1910 | log_unit_warning_errno(u, r, "Failed to migrate cgroup from to %s, ignoring: %m", u->cgroup_path); |
0cd385d3 | 1911 | } |
03b90d4b | 1912 | |
0d2d6fbf CD |
1913 | /* Set attributes */ |
1914 | cgroup_context_apply(u, target_mask, state); | |
1915 | cgroup_xattr_apply(u); | |
1916 | ||
64747e2d LP |
1917 | return 0; |
1918 | } | |
1919 | ||
6592b975 LP |
1920 | static int unit_attach_pid_to_cgroup_via_bus(Unit *u, pid_t pid, const char *suffix_path) { |
1921 | _cleanup_(sd_bus_error_free) sd_bus_error error = SD_BUS_ERROR_NULL; | |
1922 | char *pp; | |
7b3fd631 | 1923 | int r; |
6592b975 | 1924 | |
7b3fd631 LP |
1925 | assert(u); |
1926 | ||
6592b975 LP |
1927 | if (MANAGER_IS_SYSTEM(u->manager)) |
1928 | return -EINVAL; | |
1929 | ||
1930 | if (!u->manager->system_bus) | |
1931 | return -EIO; | |
1932 | ||
1933 | if (!u->cgroup_path) | |
1934 | return -EINVAL; | |
1935 | ||
1936 | /* Determine this unit's cgroup path relative to our cgroup root */ | |
1937 | pp = path_startswith(u->cgroup_path, u->manager->cgroup_root); | |
1938 | if (!pp) | |
1939 | return -EINVAL; | |
1940 | ||
1941 | pp = strjoina("/", pp, suffix_path); | |
858d36c1 | 1942 | path_simplify(pp, false); |
6592b975 LP |
1943 | |
1944 | r = sd_bus_call_method(u->manager->system_bus, | |
1945 | "org.freedesktop.systemd1", | |
1946 | "/org/freedesktop/systemd1", | |
1947 | "org.freedesktop.systemd1.Manager", | |
1948 | "AttachProcessesToUnit", | |
1949 | &error, NULL, | |
1950 | "ssau", | |
1951 | NULL /* empty unit name means client's unit, i.e. us */, pp, 1, (uint32_t) pid); | |
7b3fd631 | 1952 | if (r < 0) |
6592b975 LP |
1953 | return log_unit_debug_errno(u, r, "Failed to attach unit process " PID_FMT " via the bus: %s", pid, bus_error_message(&error, r)); |
1954 | ||
1955 | return 0; | |
1956 | } | |
1957 | ||
1958 | int unit_attach_pids_to_cgroup(Unit *u, Set *pids, const char *suffix_path) { | |
1959 | CGroupMask delegated_mask; | |
1960 | const char *p; | |
1961 | Iterator i; | |
1962 | void *pidp; | |
1963 | int r, q; | |
1964 | ||
1965 | assert(u); | |
1966 | ||
1967 | if (!UNIT_HAS_CGROUP_CONTEXT(u)) | |
1968 | return -EINVAL; | |
1969 | ||
1970 | if (set_isempty(pids)) | |
1971 | return 0; | |
7b3fd631 | 1972 | |
fab34748 KL |
1973 | /* Load any custom firewall BPF programs here once to test if they are existing and actually loadable. |
1974 | * Fail here early since later errors in the call chain unit_realize_cgroup to cgroup_context_apply are ignored. */ | |
1975 | r = bpf_firewall_load_custom(u); | |
1976 | if (r < 0) | |
1977 | return r; | |
1978 | ||
6592b975 | 1979 | r = unit_realize_cgroup(u); |
7b3fd631 LP |
1980 | if (r < 0) |
1981 | return r; | |
1982 | ||
6592b975 LP |
1983 | if (isempty(suffix_path)) |
1984 | p = u->cgroup_path; | |
1985 | else | |
270384b2 | 1986 | p = prefix_roota(u->cgroup_path, suffix_path); |
6592b975 LP |
1987 | |
1988 | delegated_mask = unit_get_delegate_mask(u); | |
1989 | ||
1990 | r = 0; | |
1991 | SET_FOREACH(pidp, pids, i) { | |
1992 | pid_t pid = PTR_TO_PID(pidp); | |
1993 | CGroupController c; | |
1994 | ||
1995 | /* First, attach the PID to the main cgroup hierarchy */ | |
1996 | q = cg_attach(SYSTEMD_CGROUP_CONTROLLER, p, pid); | |
1997 | if (q < 0) { | |
1998 | log_unit_debug_errno(u, q, "Couldn't move process " PID_FMT " to requested cgroup '%s': %m", pid, p); | |
1999 | ||
2000 | if (MANAGER_IS_USER(u->manager) && IN_SET(q, -EPERM, -EACCES)) { | |
2001 | int z; | |
2002 | ||
2003 | /* If we are in a user instance, and we can't move the process ourselves due to | |
2004 | * permission problems, let's ask the system instance about it instead. Since it's more | |
2005 | * privileged it might be able to move the process across the leaves of a subtree who's | |
2006 | * top node is not owned by us. */ | |
2007 | ||
2008 | z = unit_attach_pid_to_cgroup_via_bus(u, pid, suffix_path); | |
2009 | if (z < 0) | |
2010 | log_unit_debug_errno(u, z, "Couldn't move process " PID_FMT " to requested cgroup '%s' via the system bus either: %m", pid, p); | |
2011 | else | |
2012 | continue; /* When the bus thing worked via the bus we are fully done for this PID. */ | |
2013 | } | |
2014 | ||
2015 | if (r >= 0) | |
2016 | r = q; /* Remember first error */ | |
2017 | ||
2018 | continue; | |
2019 | } | |
2020 | ||
2021 | q = cg_all_unified(); | |
2022 | if (q < 0) | |
2023 | return q; | |
2024 | if (q > 0) | |
2025 | continue; | |
2026 | ||
2027 | /* In the legacy hierarchy, attach the process to the request cgroup if possible, and if not to the | |
2028 | * innermost realized one */ | |
2029 | ||
2030 | for (c = 0; c < _CGROUP_CONTROLLER_MAX; c++) { | |
2031 | CGroupMask bit = CGROUP_CONTROLLER_TO_MASK(c); | |
2032 | const char *realized; | |
2033 | ||
2034 | if (!(u->manager->cgroup_supported & bit)) | |
2035 | continue; | |
2036 | ||
2037 | /* If this controller is delegated and realized, honour the caller's request for the cgroup suffix. */ | |
2038 | if (delegated_mask & u->cgroup_realized_mask & bit) { | |
2039 | q = cg_attach(cgroup_controller_to_string(c), p, pid); | |
2040 | if (q >= 0) | |
2041 | continue; /* Success! */ | |
2042 | ||
2043 | 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", | |
2044 | pid, p, cgroup_controller_to_string(c)); | |
2045 | } | |
2046 | ||
2047 | /* So this controller is either not delegate or realized, or something else weird happened. In | |
2048 | * that case let's attach the PID at least to the closest cgroup up the tree that is | |
2049 | * realized. */ | |
2050 | realized = unit_get_realized_cgroup_path(u, bit); | |
2051 | if (!realized) | |
2052 | continue; /* Not even realized in the root slice? Then let's not bother */ | |
2053 | ||
2054 | q = cg_attach(cgroup_controller_to_string(c), realized, pid); | |
2055 | if (q < 0) | |
2056 | log_unit_debug_errno(u, q, "Failed to attach PID " PID_FMT " to realized cgroup %s in controller %s, ignoring: %m", | |
2057 | pid, realized, cgroup_controller_to_string(c)); | |
2058 | } | |
2059 | } | |
2060 | ||
2061 | return r; | |
7b3fd631 LP |
2062 | } |
2063 | ||
906c06f6 DM |
2064 | static bool unit_has_mask_realized( |
2065 | Unit *u, | |
2066 | CGroupMask target_mask, | |
17f14955 | 2067 | CGroupMask enable_mask) { |
906c06f6 | 2068 | |
bc432dc7 LP |
2069 | assert(u); |
2070 | ||
d5095dcd LP |
2071 | /* Returns true if this unit is fully realized. We check four things: |
2072 | * | |
2073 | * 1. Whether the cgroup was created at all | |
4e1dfa45 CD |
2074 | * 2. Whether the cgroup was created in all the hierarchies we need it to be created in (in case of cgroup v1) |
2075 | * 3. Whether the cgroup has all the right controllers enabled (in case of cgroup v2) | |
d5095dcd LP |
2076 | * 4. Whether the invalidation mask is currently zero |
2077 | * | |
2078 | * If you wonder why we mask the target realization and enable mask with CGROUP_MASK_V1/CGROUP_MASK_V2: note | |
4e1dfa45 CD |
2079 | * that there are three sets of bitmasks: CGROUP_MASK_V1 (for real cgroup v1 controllers), CGROUP_MASK_V2 (for |
2080 | * real cgroup v2 controllers) and CGROUP_MASK_BPF (for BPF-based pseudo-controllers). Now, cgroup_realized_mask | |
2081 | * is only matters for cgroup v1 controllers, and cgroup_enabled_mask only used for cgroup v2, and if they | |
d5095dcd LP |
2082 | * differ in the others, we don't really care. (After all, the cgroup_enabled_mask tracks with controllers are |
2083 | * enabled through cgroup.subtree_control, and since the BPF pseudo-controllers don't show up there, they | |
2084 | * simply don't matter. */ | |
2085 | ||
906c06f6 | 2086 | return u->cgroup_realized && |
d5095dcd LP |
2087 | ((u->cgroup_realized_mask ^ target_mask) & CGROUP_MASK_V1) == 0 && |
2088 | ((u->cgroup_enabled_mask ^ enable_mask) & CGROUP_MASK_V2) == 0 && | |
17f14955 | 2089 | u->cgroup_invalidated_mask == 0; |
6414b7c9 DS |
2090 | } |
2091 | ||
4f6f62e4 CD |
2092 | static bool unit_has_mask_disables_realized( |
2093 | Unit *u, | |
2094 | CGroupMask target_mask, | |
2095 | CGroupMask enable_mask) { | |
2096 | ||
2097 | assert(u); | |
2098 | ||
2099 | /* Returns true if all controllers which should be disabled are indeed disabled. | |
2100 | * | |
2101 | * Unlike unit_has_mask_realized, we don't care what was enabled, only that anything we want to remove is | |
2102 | * already removed. */ | |
2103 | ||
2104 | return !u->cgroup_realized || | |
2105 | (FLAGS_SET(u->cgroup_realized_mask, target_mask & CGROUP_MASK_V1) && | |
2106 | FLAGS_SET(u->cgroup_enabled_mask, enable_mask & CGROUP_MASK_V2)); | |
2107 | } | |
2108 | ||
a57669d2 CD |
2109 | static bool unit_has_mask_enables_realized( |
2110 | Unit *u, | |
2111 | CGroupMask target_mask, | |
2112 | CGroupMask enable_mask) { | |
2113 | ||
2114 | assert(u); | |
2115 | ||
2116 | /* Returns true if all controllers which should be enabled are indeed enabled. | |
2117 | * | |
2118 | * Unlike unit_has_mask_realized, we don't care about the controllers that are not present, only that anything | |
2119 | * we want to add is already added. */ | |
2120 | ||
2121 | return u->cgroup_realized && | |
c72703e2 CD |
2122 | ((u->cgroup_realized_mask | target_mask) & CGROUP_MASK_V1) == (u->cgroup_realized_mask & CGROUP_MASK_V1) && |
2123 | ((u->cgroup_enabled_mask | enable_mask) & CGROUP_MASK_V2) == (u->cgroup_enabled_mask & CGROUP_MASK_V2); | |
a57669d2 CD |
2124 | } |
2125 | ||
27adcc97 | 2126 | void unit_add_to_cgroup_realize_queue(Unit *u) { |
2aa57a65 LP |
2127 | assert(u); |
2128 | ||
2129 | if (u->in_cgroup_realize_queue) | |
2130 | return; | |
2131 | ||
2132 | LIST_PREPEND(cgroup_realize_queue, u->manager->cgroup_realize_queue, u); | |
2133 | u->in_cgroup_realize_queue = true; | |
2134 | } | |
2135 | ||
2136 | static void unit_remove_from_cgroup_realize_queue(Unit *u) { | |
2137 | assert(u); | |
2138 | ||
2139 | if (!u->in_cgroup_realize_queue) | |
2140 | return; | |
2141 | ||
2142 | LIST_REMOVE(cgroup_realize_queue, u->manager->cgroup_realize_queue, u); | |
2143 | u->in_cgroup_realize_queue = false; | |
2144 | } | |
2145 | ||
a57669d2 CD |
2146 | /* Controllers can only be enabled breadth-first, from the root of the |
2147 | * hierarchy downwards to the unit in question. */ | |
2148 | static int unit_realize_cgroup_now_enable(Unit *u, ManagerState state) { | |
2149 | CGroupMask target_mask, enable_mask, new_target_mask, new_enable_mask; | |
2150 | int r; | |
2151 | ||
2152 | assert(u); | |
2153 | ||
2154 | /* First go deal with this unit's parent, or we won't be able to enable | |
2155 | * any new controllers at this layer. */ | |
2156 | if (UNIT_ISSET(u->slice)) { | |
2157 | r = unit_realize_cgroup_now_enable(UNIT_DEREF(u->slice), state); | |
2158 | if (r < 0) | |
2159 | return r; | |
2160 | } | |
2161 | ||
2162 | target_mask = unit_get_target_mask(u); | |
2163 | enable_mask = unit_get_enable_mask(u); | |
2164 | ||
2165 | /* We can only enable in this direction, don't try to disable anything. | |
2166 | */ | |
2167 | if (unit_has_mask_enables_realized(u, target_mask, enable_mask)) | |
2168 | return 0; | |
2169 | ||
2170 | new_target_mask = u->cgroup_realized_mask | target_mask; | |
2171 | new_enable_mask = u->cgroup_enabled_mask | enable_mask; | |
2172 | ||
c72703e2 | 2173 | return unit_create_cgroup(u, new_target_mask, new_enable_mask, state); |
a57669d2 CD |
2174 | } |
2175 | ||
4f6f62e4 CD |
2176 | /* Controllers can only be disabled depth-first, from the leaves of the |
2177 | * hierarchy upwards to the unit in question. */ | |
2178 | static int unit_realize_cgroup_now_disable(Unit *u, ManagerState state) { | |
2179 | Iterator i; | |
2180 | Unit *m; | |
2181 | void *v; | |
2182 | ||
2183 | assert(u); | |
2184 | ||
2185 | if (u->type != UNIT_SLICE) | |
2186 | return 0; | |
2187 | ||
2188 | HASHMAP_FOREACH_KEY(v, m, u->dependencies[UNIT_BEFORE], i) { | |
2189 | CGroupMask target_mask, enable_mask, new_target_mask, new_enable_mask; | |
2190 | int r; | |
2191 | ||
2192 | if (UNIT_DEREF(m->slice) != u) | |
2193 | continue; | |
2194 | ||
2195 | /* The cgroup for this unit might not actually be fully | |
2196 | * realised yet, in which case it isn't holding any controllers | |
2197 | * open anyway. */ | |
2198 | if (!m->cgroup_path) | |
2199 | continue; | |
2200 | ||
2201 | /* We must disable those below us first in order to release the | |
2202 | * controller. */ | |
2203 | if (m->type == UNIT_SLICE) | |
2204 | (void) unit_realize_cgroup_now_disable(m, state); | |
2205 | ||
2206 | target_mask = unit_get_target_mask(m); | |
2207 | enable_mask = unit_get_enable_mask(m); | |
2208 | ||
2209 | /* We can only disable in this direction, don't try to enable | |
2210 | * anything. */ | |
2211 | if (unit_has_mask_disables_realized(m, target_mask, enable_mask)) | |
2212 | continue; | |
2213 | ||
2214 | new_target_mask = m->cgroup_realized_mask & target_mask; | |
2215 | new_enable_mask = m->cgroup_enabled_mask & enable_mask; | |
2216 | ||
2217 | r = unit_create_cgroup(m, new_target_mask, new_enable_mask, state); | |
2218 | if (r < 0) | |
2219 | return r; | |
2220 | } | |
2221 | ||
2222 | return 0; | |
2223 | } | |
a57669d2 | 2224 | |
6414b7c9 DS |
2225 | /* Check if necessary controllers and attributes for a unit are in place. |
2226 | * | |
a57669d2 CD |
2227 | * - If so, do nothing. |
2228 | * - If not, create paths, move processes over, and set attributes. | |
2229 | * | |
2230 | * Controllers can only be *enabled* in a breadth-first way, and *disabled* in | |
2231 | * a depth-first way. As such the process looks like this: | |
2232 | * | |
2233 | * Suppose we have a cgroup hierarchy which looks like this: | |
2234 | * | |
2235 | * root | |
2236 | * / \ | |
2237 | * / \ | |
2238 | * / \ | |
2239 | * a b | |
2240 | * / \ / \ | |
2241 | * / \ / \ | |
2242 | * c d e f | |
2243 | * / \ / \ / \ / \ | |
2244 | * h i j k l m n o | |
2245 | * | |
2246 | * 1. We want to realise cgroup "d" now. | |
c72703e2 | 2247 | * 2. cgroup "a" has DisableControllers=cpu in the associated unit. |
a57669d2 CD |
2248 | * 3. cgroup "k" just started requesting the memory controller. |
2249 | * | |
2250 | * To make this work we must do the following in order: | |
2251 | * | |
2252 | * 1. Disable CPU controller in k, j | |
2253 | * 2. Disable CPU controller in d | |
2254 | * 3. Enable memory controller in root | |
2255 | * 4. Enable memory controller in a | |
2256 | * 5. Enable memory controller in d | |
2257 | * 6. Enable memory controller in k | |
2258 | * | |
2259 | * Notice that we need to touch j in one direction, but not the other. We also | |
2260 | * don't go beyond d when disabling -- it's up to "a" to get realized if it | |
2261 | * wants to disable further. The basic rules are therefore: | |
2262 | * | |
2263 | * - If you're disabling something, you need to realise all of the cgroups from | |
2264 | * your recursive descendants to the root. This starts from the leaves. | |
2265 | * - If you're enabling something, you need to realise from the root cgroup | |
2266 | * downwards, but you don't need to iterate your recursive descendants. | |
6414b7c9 DS |
2267 | * |
2268 | * Returns 0 on success and < 0 on failure. */ | |
db785129 | 2269 | static int unit_realize_cgroup_now(Unit *u, ManagerState state) { |
efdb0237 | 2270 | CGroupMask target_mask, enable_mask; |
6414b7c9 | 2271 | int r; |
64747e2d | 2272 | |
4ad49000 | 2273 | assert(u); |
64747e2d | 2274 | |
2aa57a65 | 2275 | unit_remove_from_cgroup_realize_queue(u); |
64747e2d | 2276 | |
efdb0237 | 2277 | target_mask = unit_get_target_mask(u); |
ccf78df1 TH |
2278 | enable_mask = unit_get_enable_mask(u); |
2279 | ||
17f14955 | 2280 | if (unit_has_mask_realized(u, target_mask, enable_mask)) |
0a1eb06d | 2281 | return 0; |
64747e2d | 2282 | |
4f6f62e4 CD |
2283 | /* Disable controllers below us, if there are any */ |
2284 | r = unit_realize_cgroup_now_disable(u, state); | |
2285 | if (r < 0) | |
2286 | return r; | |
2287 | ||
2288 | /* Enable controllers above us, if there are any */ | |
6414b7c9 | 2289 | if (UNIT_ISSET(u->slice)) { |
a57669d2 | 2290 | r = unit_realize_cgroup_now_enable(UNIT_DEREF(u->slice), state); |
6414b7c9 DS |
2291 | if (r < 0) |
2292 | return r; | |
2293 | } | |
4ad49000 | 2294 | |
0d2d6fbf CD |
2295 | /* Now actually deal with the cgroup we were trying to realise and set attributes */ |
2296 | r = unit_create_cgroup(u, target_mask, enable_mask, state); | |
6414b7c9 DS |
2297 | if (r < 0) |
2298 | return r; | |
2299 | ||
c2baf11c LP |
2300 | /* Now, reset the invalidation mask */ |
2301 | u->cgroup_invalidated_mask = 0; | |
6414b7c9 | 2302 | return 0; |
64747e2d LP |
2303 | } |
2304 | ||
91a6073e | 2305 | unsigned manager_dispatch_cgroup_realize_queue(Manager *m) { |
db785129 | 2306 | ManagerState state; |
4ad49000 | 2307 | unsigned n = 0; |
db785129 | 2308 | Unit *i; |
6414b7c9 | 2309 | int r; |
ecedd90f | 2310 | |
91a6073e LP |
2311 | assert(m); |
2312 | ||
db785129 LP |
2313 | state = manager_state(m); |
2314 | ||
91a6073e LP |
2315 | while ((i = m->cgroup_realize_queue)) { |
2316 | assert(i->in_cgroup_realize_queue); | |
ecedd90f | 2317 | |
2aa57a65 LP |
2318 | if (UNIT_IS_INACTIVE_OR_FAILED(unit_active_state(i))) { |
2319 | /* Maybe things changed, and the unit is not actually active anymore? */ | |
2320 | unit_remove_from_cgroup_realize_queue(i); | |
2321 | continue; | |
2322 | } | |
2323 | ||
db785129 | 2324 | r = unit_realize_cgroup_now(i, state); |
6414b7c9 | 2325 | if (r < 0) |
efdb0237 | 2326 | log_warning_errno(r, "Failed to realize cgroups for queued unit %s, ignoring: %m", i->id); |
0a1eb06d | 2327 | |
4ad49000 LP |
2328 | n++; |
2329 | } | |
ecedd90f | 2330 | |
4ad49000 | 2331 | return n; |
8e274523 LP |
2332 | } |
2333 | ||
91a6073e | 2334 | static void unit_add_siblings_to_cgroup_realize_queue(Unit *u) { |
4ad49000 | 2335 | Unit *slice; |
ca949c9d | 2336 | |
4ad49000 LP |
2337 | /* This adds the siblings of the specified unit and the |
2338 | * siblings of all parent units to the cgroup queue. (But | |
2339 | * neither the specified unit itself nor the parents.) */ | |
2340 | ||
2341 | while ((slice = UNIT_DEREF(u->slice))) { | |
2342 | Iterator i; | |
2343 | Unit *m; | |
eef85c4a | 2344 | void *v; |
8f53a7b8 | 2345 | |
eef85c4a | 2346 | HASHMAP_FOREACH_KEY(v, m, u->dependencies[UNIT_BEFORE], i) { |
6414b7c9 DS |
2347 | /* Skip units that have a dependency on the slice |
2348 | * but aren't actually in it. */ | |
4ad49000 | 2349 | if (UNIT_DEREF(m->slice) != slice) |
50159e6a | 2350 | continue; |
8e274523 | 2351 | |
6414b7c9 DS |
2352 | /* No point in doing cgroup application for units |
2353 | * without active processes. */ | |
2354 | if (UNIT_IS_INACTIVE_OR_FAILED(unit_active_state(m))) | |
2355 | continue; | |
2356 | ||
2357 | /* If the unit doesn't need any new controllers | |
2358 | * and has current ones realized, it doesn't need | |
2359 | * any changes. */ | |
906c06f6 DM |
2360 | if (unit_has_mask_realized(m, |
2361 | unit_get_target_mask(m), | |
17f14955 | 2362 | unit_get_enable_mask(m))) |
6414b7c9 DS |
2363 | continue; |
2364 | ||
91a6073e | 2365 | unit_add_to_cgroup_realize_queue(m); |
50159e6a LP |
2366 | } |
2367 | ||
4ad49000 | 2368 | u = slice; |
8e274523 | 2369 | } |
4ad49000 LP |
2370 | } |
2371 | ||
0a1eb06d | 2372 | int unit_realize_cgroup(Unit *u) { |
4ad49000 LP |
2373 | assert(u); |
2374 | ||
35b7ff80 | 2375 | if (!UNIT_HAS_CGROUP_CONTEXT(u)) |
0a1eb06d | 2376 | return 0; |
8e274523 | 2377 | |
4ad49000 LP |
2378 | /* So, here's the deal: when realizing the cgroups for this |
2379 | * unit, we need to first create all parents, but there's more | |
2380 | * actually: for the weight-based controllers we also need to | |
2381 | * make sure that all our siblings (i.e. units that are in the | |
73e231ab | 2382 | * same slice as we are) have cgroups, too. Otherwise, things |
4ad49000 LP |
2383 | * would become very uneven as each of their processes would |
2384 | * get as much resources as all our group together. This call | |
2385 | * will synchronously create the parent cgroups, but will | |
2386 | * defer work on the siblings to the next event loop | |
2387 | * iteration. */ | |
ca949c9d | 2388 | |
4ad49000 | 2389 | /* Add all sibling slices to the cgroup queue. */ |
91a6073e | 2390 | unit_add_siblings_to_cgroup_realize_queue(u); |
4ad49000 | 2391 | |
6414b7c9 | 2392 | /* And realize this one now (and apply the values) */ |
db785129 | 2393 | return unit_realize_cgroup_now(u, manager_state(u->manager)); |
8e274523 LP |
2394 | } |
2395 | ||
efdb0237 LP |
2396 | void unit_release_cgroup(Unit *u) { |
2397 | assert(u); | |
2398 | ||
8a0d5388 LP |
2399 | /* Forgets all cgroup details for this cgroup — but does *not* destroy the cgroup. This is hence OK to call |
2400 | * when we close down everything for reexecution, where we really want to leave the cgroup in place. */ | |
efdb0237 LP |
2401 | |
2402 | if (u->cgroup_path) { | |
2403 | (void) hashmap_remove(u->manager->cgroup_unit, u->cgroup_path); | |
2404 | u->cgroup_path = mfree(u->cgroup_path); | |
2405 | } | |
2406 | ||
0bb814c2 LP |
2407 | if (u->cgroup_control_inotify_wd >= 0) { |
2408 | if (inotify_rm_watch(u->manager->cgroup_inotify_fd, u->cgroup_control_inotify_wd) < 0) | |
2409 | 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 | 2410 | |
0bb814c2 LP |
2411 | (void) hashmap_remove(u->manager->cgroup_control_inotify_wd_unit, INT_TO_PTR(u->cgroup_control_inotify_wd)); |
2412 | u->cgroup_control_inotify_wd = -1; | |
efdb0237 | 2413 | } |
afcfaa69 LP |
2414 | |
2415 | if (u->cgroup_memory_inotify_wd >= 0) { | |
2416 | if (inotify_rm_watch(u->manager->cgroup_inotify_fd, u->cgroup_memory_inotify_wd) < 0) | |
2417 | 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); | |
2418 | ||
2419 | (void) hashmap_remove(u->manager->cgroup_memory_inotify_wd_unit, INT_TO_PTR(u->cgroup_memory_inotify_wd)); | |
2420 | u->cgroup_memory_inotify_wd = -1; | |
2421 | } | |
efdb0237 LP |
2422 | } |
2423 | ||
2424 | void unit_prune_cgroup(Unit *u) { | |
8e274523 | 2425 | int r; |
efdb0237 | 2426 | bool is_root_slice; |
8e274523 | 2427 | |
4ad49000 | 2428 | assert(u); |
8e274523 | 2429 | |
efdb0237 LP |
2430 | /* Removes the cgroup, if empty and possible, and stops watching it. */ |
2431 | ||
4ad49000 LP |
2432 | if (!u->cgroup_path) |
2433 | return; | |
8e274523 | 2434 | |
fe700f46 LP |
2435 | (void) unit_get_cpu_usage(u, NULL); /* Cache the last CPU usage value before we destroy the cgroup */ |
2436 | ||
efdb0237 LP |
2437 | is_root_slice = unit_has_name(u, SPECIAL_ROOT_SLICE); |
2438 | ||
2439 | r = cg_trim_everywhere(u->manager->cgroup_supported, u->cgroup_path, !is_root_slice); | |
0219b352 DB |
2440 | if (r < 0) |
2441 | /* One reason we could have failed here is, that the cgroup still contains a process. | |
2442 | * However, if the cgroup becomes removable at a later time, it might be removed when | |
2443 | * the containing slice is stopped. So even if we failed now, this unit shouldn't assume | |
2444 | * that the cgroup is still realized the next time it is started. Do not return early | |
2445 | * on error, continue cleanup. */ | |
2446 | log_unit_full(u, r == -EBUSY ? LOG_DEBUG : LOG_WARNING, r, "Failed to destroy cgroup %s, ignoring: %m", u->cgroup_path); | |
8e274523 | 2447 | |
efdb0237 LP |
2448 | if (is_root_slice) |
2449 | return; | |
2450 | ||
2451 | unit_release_cgroup(u); | |
0a1eb06d | 2452 | |
4ad49000 | 2453 | u->cgroup_realized = false; |
bc432dc7 | 2454 | u->cgroup_realized_mask = 0; |
ccf78df1 | 2455 | u->cgroup_enabled_mask = 0; |
084c7007 RG |
2456 | |
2457 | u->bpf_device_control_installed = bpf_program_unref(u->bpf_device_control_installed); | |
8e274523 LP |
2458 | } |
2459 | ||
efdb0237 | 2460 | int unit_search_main_pid(Unit *u, pid_t *ret) { |
4ad49000 | 2461 | _cleanup_fclose_ FILE *f = NULL; |
4d051546 | 2462 | pid_t pid = 0, npid; |
efdb0237 | 2463 | int r; |
4ad49000 LP |
2464 | |
2465 | assert(u); | |
efdb0237 | 2466 | assert(ret); |
4ad49000 LP |
2467 | |
2468 | if (!u->cgroup_path) | |
efdb0237 | 2469 | return -ENXIO; |
4ad49000 | 2470 | |
efdb0237 LP |
2471 | r = cg_enumerate_processes(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path, &f); |
2472 | if (r < 0) | |
2473 | return r; | |
4ad49000 | 2474 | |
4ad49000 | 2475 | while (cg_read_pid(f, &npid) > 0) { |
4ad49000 LP |
2476 | |
2477 | if (npid == pid) | |
2478 | continue; | |
8e274523 | 2479 | |
4d051546 | 2480 | if (pid_is_my_child(npid) == 0) |
4ad49000 | 2481 | continue; |
8e274523 | 2482 | |
efdb0237 | 2483 | if (pid != 0) |
4ad49000 LP |
2484 | /* Dang, there's more than one daemonized PID |
2485 | in this group, so we don't know what process | |
2486 | is the main process. */ | |
efdb0237 LP |
2487 | |
2488 | return -ENODATA; | |
8e274523 | 2489 | |
4ad49000 | 2490 | pid = npid; |
8e274523 LP |
2491 | } |
2492 | ||
efdb0237 LP |
2493 | *ret = pid; |
2494 | return 0; | |
2495 | } | |
2496 | ||
2497 | static int unit_watch_pids_in_path(Unit *u, const char *path) { | |
b3c5bad3 | 2498 | _cleanup_closedir_ DIR *d = NULL; |
efdb0237 LP |
2499 | _cleanup_fclose_ FILE *f = NULL; |
2500 | int ret = 0, r; | |
2501 | ||
2502 | assert(u); | |
2503 | assert(path); | |
2504 | ||
2505 | r = cg_enumerate_processes(SYSTEMD_CGROUP_CONTROLLER, path, &f); | |
2506 | if (r < 0) | |
2507 | ret = r; | |
2508 | else { | |
2509 | pid_t pid; | |
2510 | ||
2511 | while ((r = cg_read_pid(f, &pid)) > 0) { | |
f75f613d | 2512 | r = unit_watch_pid(u, pid, false); |
efdb0237 LP |
2513 | if (r < 0 && ret >= 0) |
2514 | ret = r; | |
2515 | } | |
2516 | ||
2517 | if (r < 0 && ret >= 0) | |
2518 | ret = r; | |
2519 | } | |
2520 | ||
2521 | r = cg_enumerate_subgroups(SYSTEMD_CGROUP_CONTROLLER, path, &d); | |
2522 | if (r < 0) { | |
2523 | if (ret >= 0) | |
2524 | ret = r; | |
2525 | } else { | |
2526 | char *fn; | |
2527 | ||
2528 | while ((r = cg_read_subgroup(d, &fn)) > 0) { | |
2529 | _cleanup_free_ char *p = NULL; | |
2530 | ||
95b21cff | 2531 | p = path_join(empty_to_root(path), fn); |
efdb0237 LP |
2532 | free(fn); |
2533 | ||
2534 | if (!p) | |
2535 | return -ENOMEM; | |
2536 | ||
2537 | r = unit_watch_pids_in_path(u, p); | |
2538 | if (r < 0 && ret >= 0) | |
2539 | ret = r; | |
2540 | } | |
2541 | ||
2542 | if (r < 0 && ret >= 0) | |
2543 | ret = r; | |
2544 | } | |
2545 | ||
2546 | return ret; | |
2547 | } | |
2548 | ||
11aef522 LP |
2549 | int unit_synthesize_cgroup_empty_event(Unit *u) { |
2550 | int r; | |
2551 | ||
2552 | assert(u); | |
2553 | ||
2554 | /* Enqueue a synthetic cgroup empty event if this unit doesn't watch any PIDs anymore. This is compatibility | |
2555 | * support for non-unified systems where notifications aren't reliable, and hence need to take whatever we can | |
2556 | * get as notification source as soon as we stopped having any useful PIDs to watch for. */ | |
2557 | ||
2558 | if (!u->cgroup_path) | |
2559 | return -ENOENT; | |
2560 | ||
2561 | r = cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER); | |
2562 | if (r < 0) | |
2563 | return r; | |
2564 | if (r > 0) /* On unified we have reliable notifications, and don't need this */ | |
2565 | return 0; | |
2566 | ||
2567 | if (!set_isempty(u->pids)) | |
2568 | return 0; | |
2569 | ||
2570 | unit_add_to_cgroup_empty_queue(u); | |
2571 | return 0; | |
2572 | } | |
2573 | ||
efdb0237 | 2574 | int unit_watch_all_pids(Unit *u) { |
b4cccbc1 LP |
2575 | int r; |
2576 | ||
efdb0237 LP |
2577 | assert(u); |
2578 | ||
2579 | /* Adds all PIDs from our cgroup to the set of PIDs we | |
2580 | * watch. This is a fallback logic for cases where we do not | |
2581 | * get reliable cgroup empty notifications: we try to use | |
2582 | * SIGCHLD as replacement. */ | |
2583 | ||
2584 | if (!u->cgroup_path) | |
2585 | return -ENOENT; | |
2586 | ||
c22800e4 | 2587 | r = cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER); |
b4cccbc1 LP |
2588 | if (r < 0) |
2589 | return r; | |
2590 | if (r > 0) /* On unified we can use proper notifications */ | |
efdb0237 LP |
2591 | return 0; |
2592 | ||
2593 | return unit_watch_pids_in_path(u, u->cgroup_path); | |
2594 | } | |
2595 | ||
09e24654 LP |
2596 | static int on_cgroup_empty_event(sd_event_source *s, void *userdata) { |
2597 | Manager *m = userdata; | |
2598 | Unit *u; | |
efdb0237 LP |
2599 | int r; |
2600 | ||
09e24654 LP |
2601 | assert(s); |
2602 | assert(m); | |
efdb0237 | 2603 | |
09e24654 LP |
2604 | u = m->cgroup_empty_queue; |
2605 | if (!u) | |
efdb0237 LP |
2606 | return 0; |
2607 | ||
09e24654 LP |
2608 | assert(u->in_cgroup_empty_queue); |
2609 | u->in_cgroup_empty_queue = false; | |
2610 | LIST_REMOVE(cgroup_empty_queue, m->cgroup_empty_queue, u); | |
2611 | ||
2612 | if (m->cgroup_empty_queue) { | |
2613 | /* More stuff queued, let's make sure we remain enabled */ | |
2614 | r = sd_event_source_set_enabled(s, SD_EVENT_ONESHOT); | |
2615 | if (r < 0) | |
19a691a9 | 2616 | log_debug_errno(r, "Failed to reenable cgroup empty event source, ignoring: %m"); |
09e24654 | 2617 | } |
efdb0237 LP |
2618 | |
2619 | unit_add_to_gc_queue(u); | |
2620 | ||
2621 | if (UNIT_VTABLE(u)->notify_cgroup_empty) | |
2622 | UNIT_VTABLE(u)->notify_cgroup_empty(u); | |
2623 | ||
2624 | return 0; | |
2625 | } | |
2626 | ||
09e24654 LP |
2627 | void unit_add_to_cgroup_empty_queue(Unit *u) { |
2628 | int r; | |
2629 | ||
2630 | assert(u); | |
2631 | ||
2632 | /* Note that there are four different ways how cgroup empty events reach us: | |
2633 | * | |
2634 | * 1. On the unified hierarchy we get an inotify event on the cgroup | |
2635 | * | |
2636 | * 2. On the legacy hierarchy, when running in system mode, we get a datagram on the cgroup agent socket | |
2637 | * | |
2638 | * 3. On the legacy hierarchy, when running in user mode, we get a D-Bus signal on the system bus | |
2639 | * | |
2640 | * 4. On the legacy hierarchy, in service units we start watching all processes of the cgroup for SIGCHLD as | |
2641 | * soon as we get one SIGCHLD, to deal with unreliable cgroup notifications. | |
2642 | * | |
2643 | * Regardless which way we got the notification, we'll verify it here, and then add it to a separate | |
2644 | * queue. This queue will be dispatched at a lower priority than the SIGCHLD handler, so that we always use | |
2645 | * SIGCHLD if we can get it first, and only use the cgroup empty notifications if there's no SIGCHLD pending | |
2646 | * (which might happen if the cgroup doesn't contain processes that are our own child, which is typically the | |
2647 | * case for scope units). */ | |
2648 | ||
2649 | if (u->in_cgroup_empty_queue) | |
2650 | return; | |
2651 | ||
2652 | /* Let's verify that the cgroup is really empty */ | |
2653 | if (!u->cgroup_path) | |
2654 | return; | |
2655 | r = cg_is_empty_recursive(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path); | |
2656 | if (r < 0) { | |
2657 | log_unit_debug_errno(u, r, "Failed to determine whether cgroup %s is empty: %m", u->cgroup_path); | |
2658 | return; | |
2659 | } | |
2660 | if (r == 0) | |
2661 | return; | |
2662 | ||
2663 | LIST_PREPEND(cgroup_empty_queue, u->manager->cgroup_empty_queue, u); | |
2664 | u->in_cgroup_empty_queue = true; | |
2665 | ||
2666 | /* Trigger the defer event */ | |
2667 | r = sd_event_source_set_enabled(u->manager->cgroup_empty_event_source, SD_EVENT_ONESHOT); | |
2668 | if (r < 0) | |
2669 | log_debug_errno(r, "Failed to enable cgroup empty event source: %m"); | |
2670 | } | |
2671 | ||
2ba6ae6b | 2672 | int unit_check_oom(Unit *u) { |
afcfaa69 LP |
2673 | _cleanup_free_ char *oom_kill = NULL; |
2674 | bool increased; | |
2675 | uint64_t c; | |
2676 | int r; | |
2677 | ||
2678 | if (!u->cgroup_path) | |
2679 | return 0; | |
2680 | ||
2681 | r = cg_get_keyed_attribute("memory", u->cgroup_path, "memory.events", STRV_MAKE("oom_kill"), &oom_kill); | |
2682 | if (r < 0) | |
2683 | return log_unit_debug_errno(u, r, "Failed to read oom_kill field of memory.events cgroup attribute: %m"); | |
2684 | ||
2685 | r = safe_atou64(oom_kill, &c); | |
2686 | if (r < 0) | |
2687 | return log_unit_debug_errno(u, r, "Failed to parse oom_kill field: %m"); | |
2688 | ||
2689 | increased = c > u->oom_kill_last; | |
2690 | u->oom_kill_last = c; | |
2691 | ||
2692 | if (!increased) | |
2693 | return 0; | |
2694 | ||
2695 | log_struct(LOG_NOTICE, | |
2696 | "MESSAGE_ID=" SD_MESSAGE_UNIT_OUT_OF_MEMORY_STR, | |
2697 | LOG_UNIT_ID(u), | |
2698 | LOG_UNIT_INVOCATION_ID(u), | |
2699 | LOG_UNIT_MESSAGE(u, "A process of this unit has been killed by the OOM killer.")); | |
2700 | ||
2701 | if (UNIT_VTABLE(u)->notify_cgroup_oom) | |
2702 | UNIT_VTABLE(u)->notify_cgroup_oom(u); | |
2703 | ||
2704 | return 1; | |
2705 | } | |
2706 | ||
2707 | static int on_cgroup_oom_event(sd_event_source *s, void *userdata) { | |
2708 | Manager *m = userdata; | |
2709 | Unit *u; | |
2710 | int r; | |
2711 | ||
2712 | assert(s); | |
2713 | assert(m); | |
2714 | ||
2715 | u = m->cgroup_oom_queue; | |
2716 | if (!u) | |
2717 | return 0; | |
2718 | ||
2719 | assert(u->in_cgroup_oom_queue); | |
2720 | u->in_cgroup_oom_queue = false; | |
2721 | LIST_REMOVE(cgroup_oom_queue, m->cgroup_oom_queue, u); | |
2722 | ||
2723 | if (m->cgroup_oom_queue) { | |
2724 | /* More stuff queued, let's make sure we remain enabled */ | |
2725 | r = sd_event_source_set_enabled(s, SD_EVENT_ONESHOT); | |
2726 | if (r < 0) | |
2727 | log_debug_errno(r, "Failed to reenable cgroup oom event source, ignoring: %m"); | |
2728 | } | |
2729 | ||
2730 | (void) unit_check_oom(u); | |
2731 | return 0; | |
2732 | } | |
2733 | ||
2734 | static void unit_add_to_cgroup_oom_queue(Unit *u) { | |
2735 | int r; | |
2736 | ||
2737 | assert(u); | |
2738 | ||
2739 | if (u->in_cgroup_oom_queue) | |
2740 | return; | |
2741 | if (!u->cgroup_path) | |
2742 | return; | |
2743 | ||
2744 | LIST_PREPEND(cgroup_oom_queue, u->manager->cgroup_oom_queue, u); | |
2745 | u->in_cgroup_oom_queue = true; | |
2746 | ||
2747 | /* Trigger the defer event */ | |
2748 | if (!u->manager->cgroup_oom_event_source) { | |
2749 | _cleanup_(sd_event_source_unrefp) sd_event_source *s = NULL; | |
2750 | ||
2751 | r = sd_event_add_defer(u->manager->event, &s, on_cgroup_oom_event, u->manager); | |
2752 | if (r < 0) { | |
2753 | log_error_errno(r, "Failed to create cgroup oom event source: %m"); | |
2754 | return; | |
2755 | } | |
2756 | ||
2757 | r = sd_event_source_set_priority(s, SD_EVENT_PRIORITY_NORMAL-8); | |
2758 | if (r < 0) { | |
2759 | log_error_errno(r, "Failed to set priority of cgroup oom event source: %m"); | |
2760 | return; | |
2761 | } | |
2762 | ||
2763 | (void) sd_event_source_set_description(s, "cgroup-oom"); | |
2764 | u->manager->cgroup_oom_event_source = TAKE_PTR(s); | |
2765 | } | |
2766 | ||
2767 | r = sd_event_source_set_enabled(u->manager->cgroup_oom_event_source, SD_EVENT_ONESHOT); | |
2768 | if (r < 0) | |
2769 | log_error_errno(r, "Failed to enable cgroup oom event source: %m"); | |
2770 | } | |
2771 | ||
efdb0237 LP |
2772 | static int on_cgroup_inotify_event(sd_event_source *s, int fd, uint32_t revents, void *userdata) { |
2773 | Manager *m = userdata; | |
2774 | ||
2775 | assert(s); | |
2776 | assert(fd >= 0); | |
2777 | assert(m); | |
2778 | ||
2779 | for (;;) { | |
2780 | union inotify_event_buffer buffer; | |
2781 | struct inotify_event *e; | |
2782 | ssize_t l; | |
2783 | ||
2784 | l = read(fd, &buffer, sizeof(buffer)); | |
2785 | if (l < 0) { | |
47249640 | 2786 | if (IN_SET(errno, EINTR, EAGAIN)) |
efdb0237 LP |
2787 | return 0; |
2788 | ||
2789 | return log_error_errno(errno, "Failed to read control group inotify events: %m"); | |
2790 | } | |
2791 | ||
2792 | FOREACH_INOTIFY_EVENT(e, buffer, l) { | |
2793 | Unit *u; | |
2794 | ||
2795 | if (e->wd < 0) | |
2796 | /* Queue overflow has no watch descriptor */ | |
2797 | continue; | |
2798 | ||
2799 | if (e->mask & IN_IGNORED) | |
2800 | /* The watch was just removed */ | |
2801 | continue; | |
2802 | ||
afcfaa69 LP |
2803 | /* Note that inotify might deliver events for a watch even after it was removed, |
2804 | * because it was queued before the removal. Let's ignore this here safely. */ | |
2805 | ||
0bb814c2 | 2806 | u = hashmap_get(m->cgroup_control_inotify_wd_unit, INT_TO_PTR(e->wd)); |
afcfaa69 LP |
2807 | if (u) |
2808 | unit_add_to_cgroup_empty_queue(u); | |
efdb0237 | 2809 | |
afcfaa69 LP |
2810 | u = hashmap_get(m->cgroup_memory_inotify_wd_unit, INT_TO_PTR(e->wd)); |
2811 | if (u) | |
2812 | unit_add_to_cgroup_oom_queue(u); | |
efdb0237 LP |
2813 | } |
2814 | } | |
8e274523 LP |
2815 | } |
2816 | ||
17f14955 RG |
2817 | static int cg_bpf_mask_supported(CGroupMask *ret) { |
2818 | CGroupMask mask = 0; | |
2819 | int r; | |
2820 | ||
2821 | /* BPF-based firewall */ | |
2822 | r = bpf_firewall_supported(); | |
2823 | if (r > 0) | |
2824 | mask |= CGROUP_MASK_BPF_FIREWALL; | |
2825 | ||
084c7007 RG |
2826 | /* BPF-based device access control */ |
2827 | r = bpf_devices_supported(); | |
2828 | if (r > 0) | |
2829 | mask |= CGROUP_MASK_BPF_DEVICES; | |
2830 | ||
17f14955 RG |
2831 | *ret = mask; |
2832 | return 0; | |
2833 | } | |
2834 | ||
8e274523 | 2835 | int manager_setup_cgroup(Manager *m) { |
9444b1f2 | 2836 | _cleanup_free_ char *path = NULL; |
10bd3e2e | 2837 | const char *scope_path; |
efdb0237 | 2838 | CGroupController c; |
b4cccbc1 | 2839 | int r, all_unified; |
17f14955 | 2840 | CGroupMask mask; |
efdb0237 | 2841 | char *e; |
8e274523 LP |
2842 | |
2843 | assert(m); | |
2844 | ||
35d2e7ec | 2845 | /* 1. Determine hierarchy */ |
efdb0237 | 2846 | m->cgroup_root = mfree(m->cgroup_root); |
9444b1f2 | 2847 | r = cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER, 0, &m->cgroup_root); |
23bbb0de MS |
2848 | if (r < 0) |
2849 | return log_error_errno(r, "Cannot determine cgroup we are running in: %m"); | |
8e274523 | 2850 | |
efdb0237 LP |
2851 | /* Chop off the init scope, if we are already located in it */ |
2852 | e = endswith(m->cgroup_root, "/" SPECIAL_INIT_SCOPE); | |
0d8c31ff | 2853 | |
efdb0237 LP |
2854 | /* LEGACY: Also chop off the system slice if we are in |
2855 | * it. This is to support live upgrades from older systemd | |
2856 | * versions where PID 1 was moved there. Also see | |
2857 | * cg_get_root_path(). */ | |
463d0d15 | 2858 | if (!e && MANAGER_IS_SYSTEM(m)) { |
9444b1f2 | 2859 | e = endswith(m->cgroup_root, "/" SPECIAL_SYSTEM_SLICE); |
15c60e99 | 2860 | if (!e) |
efdb0237 | 2861 | e = endswith(m->cgroup_root, "/system"); /* even more legacy */ |
0baf24dd | 2862 | } |
efdb0237 LP |
2863 | if (e) |
2864 | *e = 0; | |
7ccfb64a | 2865 | |
7546145e LP |
2866 | /* And make sure to store away the root value without trailing slash, even for the root dir, so that we can |
2867 | * easily prepend it everywhere. */ | |
2868 | delete_trailing_chars(m->cgroup_root, "/"); | |
8e274523 | 2869 | |
35d2e7ec | 2870 | /* 2. Show data */ |
9444b1f2 | 2871 | r = cg_get_path(SYSTEMD_CGROUP_CONTROLLER, m->cgroup_root, NULL, &path); |
23bbb0de MS |
2872 | if (r < 0) |
2873 | return log_error_errno(r, "Cannot find cgroup mount point: %m"); | |
8e274523 | 2874 | |
d4d99bc6 | 2875 | r = cg_unified(); |
415fc41c TH |
2876 | if (r < 0) |
2877 | return log_error_errno(r, "Couldn't determine if we are running in the unified hierarchy: %m"); | |
5da38d07 | 2878 | |
b4cccbc1 | 2879 | all_unified = cg_all_unified(); |
d4c819ed ZJS |
2880 | if (all_unified < 0) |
2881 | return log_error_errno(all_unified, "Couldn't determine whether we are in all unified mode: %m"); | |
2882 | if (all_unified > 0) | |
efdb0237 | 2883 | log_debug("Unified cgroup hierarchy is located at %s.", path); |
b4cccbc1 | 2884 | else { |
c22800e4 | 2885 | r = cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER); |
b4cccbc1 LP |
2886 | if (r < 0) |
2887 | return log_error_errno(r, "Failed to determine whether systemd's own controller is in unified mode: %m"); | |
2888 | if (r > 0) | |
2889 | log_debug("Unified cgroup hierarchy is located at %s. Controllers are on legacy hierarchies.", path); | |
2890 | else | |
2891 | log_debug("Using cgroup controller " SYSTEMD_CGROUP_CONTROLLER_LEGACY ". File system hierarchy is at %s.", path); | |
2892 | } | |
efdb0237 | 2893 | |
09e24654 LP |
2894 | /* 3. Allocate cgroup empty defer event source */ |
2895 | m->cgroup_empty_event_source = sd_event_source_unref(m->cgroup_empty_event_source); | |
2896 | r = sd_event_add_defer(m->event, &m->cgroup_empty_event_source, on_cgroup_empty_event, m); | |
2897 | if (r < 0) | |
2898 | return log_error_errno(r, "Failed to create cgroup empty event source: %m"); | |
2899 | ||
cbe83389 LP |
2900 | /* Schedule cgroup empty checks early, but after having processed service notification messages or |
2901 | * SIGCHLD signals, so that a cgroup running empty is always just the last safety net of | |
2902 | * notification, and we collected the metadata the notification and SIGCHLD stuff offers first. */ | |
09e24654 LP |
2903 | r = sd_event_source_set_priority(m->cgroup_empty_event_source, SD_EVENT_PRIORITY_NORMAL-5); |
2904 | if (r < 0) | |
2905 | return log_error_errno(r, "Failed to set priority of cgroup empty event source: %m"); | |
2906 | ||
2907 | r = sd_event_source_set_enabled(m->cgroup_empty_event_source, SD_EVENT_OFF); | |
2908 | if (r < 0) | |
2909 | return log_error_errno(r, "Failed to disable cgroup empty event source: %m"); | |
2910 | ||
2911 | (void) sd_event_source_set_description(m->cgroup_empty_event_source, "cgroup-empty"); | |
2912 | ||
2913 | /* 4. Install notifier inotify object, or agent */ | |
10bd3e2e | 2914 | if (cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER) > 0) { |
c6c18be3 | 2915 | |
09e24654 | 2916 | /* In the unified hierarchy we can get cgroup empty notifications via inotify. */ |
efdb0237 | 2917 | |
10bd3e2e LP |
2918 | m->cgroup_inotify_event_source = sd_event_source_unref(m->cgroup_inotify_event_source); |
2919 | safe_close(m->cgroup_inotify_fd); | |
efdb0237 | 2920 | |
10bd3e2e LP |
2921 | m->cgroup_inotify_fd = inotify_init1(IN_NONBLOCK|IN_CLOEXEC); |
2922 | if (m->cgroup_inotify_fd < 0) | |
2923 | return log_error_errno(errno, "Failed to create control group inotify object: %m"); | |
efdb0237 | 2924 | |
10bd3e2e LP |
2925 | r = sd_event_add_io(m->event, &m->cgroup_inotify_event_source, m->cgroup_inotify_fd, EPOLLIN, on_cgroup_inotify_event, m); |
2926 | if (r < 0) | |
2927 | return log_error_errno(r, "Failed to watch control group inotify object: %m"); | |
efdb0237 | 2928 | |
cbe83389 LP |
2929 | /* Process cgroup empty notifications early. Note that when this event is dispatched it'll |
2930 | * just add the unit to a cgroup empty queue, hence let's run earlier than that. Also see | |
2931 | * handling of cgroup agent notifications, for the classic cgroup hierarchy support. */ | |
2932 | r = sd_event_source_set_priority(m->cgroup_inotify_event_source, SD_EVENT_PRIORITY_NORMAL-9); | |
10bd3e2e LP |
2933 | if (r < 0) |
2934 | return log_error_errno(r, "Failed to set priority of inotify event source: %m"); | |
efdb0237 | 2935 | |
10bd3e2e | 2936 | (void) sd_event_source_set_description(m->cgroup_inotify_event_source, "cgroup-inotify"); |
efdb0237 | 2937 | |
611c4f8a | 2938 | } else if (MANAGER_IS_SYSTEM(m) && manager_owns_host_root_cgroup(m) && !MANAGER_IS_TEST_RUN(m)) { |
efdb0237 | 2939 | |
10bd3e2e LP |
2940 | /* On the legacy hierarchy we only get notifications via cgroup agents. (Which isn't really reliable, |
2941 | * since it does not generate events when control groups with children run empty. */ | |
8e274523 | 2942 | |
10bd3e2e | 2943 | r = cg_install_release_agent(SYSTEMD_CGROUP_CONTROLLER, SYSTEMD_CGROUP_AGENT_PATH); |
23bbb0de | 2944 | if (r < 0) |
10bd3e2e LP |
2945 | log_warning_errno(r, "Failed to install release agent, ignoring: %m"); |
2946 | else if (r > 0) | |
2947 | log_debug("Installed release agent."); | |
2948 | else if (r == 0) | |
2949 | log_debug("Release agent already installed."); | |
2950 | } | |
efdb0237 | 2951 | |
09e24654 | 2952 | /* 5. Make sure we are in the special "init.scope" unit in the root slice. */ |
10bd3e2e LP |
2953 | scope_path = strjoina(m->cgroup_root, "/" SPECIAL_INIT_SCOPE); |
2954 | r = cg_create_and_attach(SYSTEMD_CGROUP_CONTROLLER, scope_path, 0); | |
aa77e234 MS |
2955 | if (r >= 0) { |
2956 | /* Also, move all other userspace processes remaining in the root cgroup into that scope. */ | |
2957 | r = cg_migrate(SYSTEMD_CGROUP_CONTROLLER, m->cgroup_root, SYSTEMD_CGROUP_CONTROLLER, scope_path, 0); | |
2958 | if (r < 0) | |
2959 | log_warning_errno(r, "Couldn't move remaining userspace processes, ignoring: %m"); | |
c6c18be3 | 2960 | |
aa77e234 MS |
2961 | /* 6. And pin it, so that it cannot be unmounted */ |
2962 | safe_close(m->pin_cgroupfs_fd); | |
2963 | m->pin_cgroupfs_fd = open(path, O_RDONLY|O_CLOEXEC|O_DIRECTORY|O_NOCTTY|O_NONBLOCK); | |
2964 | if (m->pin_cgroupfs_fd < 0) | |
2965 | return log_error_errno(errno, "Failed to open pin file: %m"); | |
0d8c31ff | 2966 | |
638cece4 | 2967 | } else if (!MANAGER_IS_TEST_RUN(m)) |
aa77e234 | 2968 | return log_error_errno(r, "Failed to create %s control group: %m", scope_path); |
10bd3e2e | 2969 | |
09e24654 | 2970 | /* 7. Always enable hierarchical support if it exists... */ |
638cece4 | 2971 | if (!all_unified && !MANAGER_IS_TEST_RUN(m)) |
10bd3e2e | 2972 | (void) cg_set_attribute("memory", "/", "memory.use_hierarchy", "1"); |
c6c18be3 | 2973 | |
17f14955 | 2974 | /* 8. Figure out which controllers are supported */ |
efdb0237 LP |
2975 | r = cg_mask_supported(&m->cgroup_supported); |
2976 | if (r < 0) | |
2977 | return log_error_errno(r, "Failed to determine supported controllers: %m"); | |
17f14955 RG |
2978 | |
2979 | /* 9. Figure out which bpf-based pseudo-controllers are supported */ | |
2980 | r = cg_bpf_mask_supported(&mask); | |
2981 | if (r < 0) | |
2982 | return log_error_errno(r, "Failed to determine supported bpf-based pseudo-controllers: %m"); | |
2983 | m->cgroup_supported |= mask; | |
2984 | ||
2985 | /* 10. Log which controllers are supported */ | |
efdb0237 | 2986 | for (c = 0; c < _CGROUP_CONTROLLER_MAX; c++) |
eee0a1e4 | 2987 | log_debug("Controller '%s' supported: %s", cgroup_controller_to_string(c), yes_no(m->cgroup_supported & CGROUP_CONTROLLER_TO_MASK(c))); |
9156e799 | 2988 | |
a32360f1 | 2989 | return 0; |
8e274523 LP |
2990 | } |
2991 | ||
c6c18be3 | 2992 | void manager_shutdown_cgroup(Manager *m, bool delete) { |
8e274523 LP |
2993 | assert(m); |
2994 | ||
9444b1f2 LP |
2995 | /* We can't really delete the group, since we are in it. But |
2996 | * let's trim it. */ | |
f6c63f6f | 2997 | if (delete && m->cgroup_root && m->test_run_flags != MANAGER_TEST_RUN_MINIMAL) |
efdb0237 LP |
2998 | (void) cg_trim(SYSTEMD_CGROUP_CONTROLLER, m->cgroup_root, false); |
2999 | ||
09e24654 LP |
3000 | m->cgroup_empty_event_source = sd_event_source_unref(m->cgroup_empty_event_source); |
3001 | ||
0bb814c2 | 3002 | m->cgroup_control_inotify_wd_unit = hashmap_free(m->cgroup_control_inotify_wd_unit); |
afcfaa69 | 3003 | m->cgroup_memory_inotify_wd_unit = hashmap_free(m->cgroup_memory_inotify_wd_unit); |
efdb0237 LP |
3004 | |
3005 | m->cgroup_inotify_event_source = sd_event_source_unref(m->cgroup_inotify_event_source); | |
3006 | m->cgroup_inotify_fd = safe_close(m->cgroup_inotify_fd); | |
8e274523 | 3007 | |
03e334a1 | 3008 | m->pin_cgroupfs_fd = safe_close(m->pin_cgroupfs_fd); |
c6c18be3 | 3009 | |
efdb0237 | 3010 | m->cgroup_root = mfree(m->cgroup_root); |
8e274523 LP |
3011 | } |
3012 | ||
4ad49000 | 3013 | Unit* manager_get_unit_by_cgroup(Manager *m, const char *cgroup) { |
acb14d31 | 3014 | char *p; |
4ad49000 | 3015 | Unit *u; |
acb14d31 LP |
3016 | |
3017 | assert(m); | |
3018 | assert(cgroup); | |
acb14d31 | 3019 | |
4ad49000 LP |
3020 | u = hashmap_get(m->cgroup_unit, cgroup); |
3021 | if (u) | |
3022 | return u; | |
acb14d31 | 3023 | |
8e70580b | 3024 | p = strdupa(cgroup); |
acb14d31 LP |
3025 | for (;;) { |
3026 | char *e; | |
3027 | ||
3028 | e = strrchr(p, '/'); | |
efdb0237 LP |
3029 | if (!e || e == p) |
3030 | return hashmap_get(m->cgroup_unit, SPECIAL_ROOT_SLICE); | |
acb14d31 LP |
3031 | |
3032 | *e = 0; | |
3033 | ||
4ad49000 LP |
3034 | u = hashmap_get(m->cgroup_unit, p); |
3035 | if (u) | |
3036 | return u; | |
acb14d31 LP |
3037 | } |
3038 | } | |
3039 | ||
b3ac818b | 3040 | Unit *manager_get_unit_by_pid_cgroup(Manager *m, pid_t pid) { |
4ad49000 | 3041 | _cleanup_free_ char *cgroup = NULL; |
8e274523 | 3042 | |
8c47c732 LP |
3043 | assert(m); |
3044 | ||
62a76913 | 3045 | if (!pid_is_valid(pid)) |
b3ac818b LP |
3046 | return NULL; |
3047 | ||
62a76913 | 3048 | if (cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER, pid, &cgroup) < 0) |
b3ac818b LP |
3049 | return NULL; |
3050 | ||
3051 | return manager_get_unit_by_cgroup(m, cgroup); | |
3052 | } | |
3053 | ||
3054 | Unit *manager_get_unit_by_pid(Manager *m, pid_t pid) { | |
62a76913 | 3055 | Unit *u, **array; |
b3ac818b LP |
3056 | |
3057 | assert(m); | |
3058 | ||
62a76913 LP |
3059 | /* Note that a process might be owned by multiple units, we return only one here, which is good enough for most |
3060 | * cases, though not strictly correct. We prefer the one reported by cgroup membership, as that's the most | |
3061 | * relevant one as children of the process will be assigned to that one, too, before all else. */ | |
3062 | ||
3063 | if (!pid_is_valid(pid)) | |
8c47c732 LP |
3064 | return NULL; |
3065 | ||
2ca9d979 | 3066 | if (pid == getpid_cached()) |
efdb0237 LP |
3067 | return hashmap_get(m->units, SPECIAL_INIT_SCOPE); |
3068 | ||
62a76913 | 3069 | u = manager_get_unit_by_pid_cgroup(m, pid); |
5fe8876b LP |
3070 | if (u) |
3071 | return u; | |
3072 | ||
62a76913 | 3073 | u = hashmap_get(m->watch_pids, PID_TO_PTR(pid)); |
5fe8876b LP |
3074 | if (u) |
3075 | return u; | |
3076 | ||
62a76913 LP |
3077 | array = hashmap_get(m->watch_pids, PID_TO_PTR(-pid)); |
3078 | if (array) | |
3079 | return array[0]; | |
3080 | ||
3081 | return NULL; | |
6dde1f33 | 3082 | } |
4fbf50b3 | 3083 | |
4ad49000 LP |
3084 | int manager_notify_cgroup_empty(Manager *m, const char *cgroup) { |
3085 | Unit *u; | |
4fbf50b3 | 3086 | |
4ad49000 LP |
3087 | assert(m); |
3088 | assert(cgroup); | |
4fbf50b3 | 3089 | |
09e24654 LP |
3090 | /* Called on the legacy hierarchy whenever we get an explicit cgroup notification from the cgroup agent process |
3091 | * or from the --system instance */ | |
3092 | ||
d8fdc620 LP |
3093 | log_debug("Got cgroup empty notification for: %s", cgroup); |
3094 | ||
4ad49000 | 3095 | u = manager_get_unit_by_cgroup(m, cgroup); |
5ad096b3 LP |
3096 | if (!u) |
3097 | return 0; | |
b56c28c3 | 3098 | |
09e24654 LP |
3099 | unit_add_to_cgroup_empty_queue(u); |
3100 | return 1; | |
5ad096b3 LP |
3101 | } |
3102 | ||
3103 | int unit_get_memory_current(Unit *u, uint64_t *ret) { | |
3104 | _cleanup_free_ char *v = NULL; | |
3105 | int r; | |
3106 | ||
3107 | assert(u); | |
3108 | assert(ret); | |
3109 | ||
2e4025c0 | 3110 | if (!UNIT_CGROUP_BOOL(u, memory_accounting)) |
cf3b4be1 LP |
3111 | return -ENODATA; |
3112 | ||
5ad096b3 LP |
3113 | if (!u->cgroup_path) |
3114 | return -ENODATA; | |
3115 | ||
1f73aa00 | 3116 | /* The root cgroup doesn't expose this information, let's get it from /proc instead */ |
611c4f8a | 3117 | if (unit_has_host_root_cgroup(u)) |
c482724a | 3118 | return procfs_memory_get_used(ret); |
1f73aa00 | 3119 | |
efdb0237 | 3120 | if ((u->cgroup_realized_mask & CGROUP_MASK_MEMORY) == 0) |
5ad096b3 LP |
3121 | return -ENODATA; |
3122 | ||
b4cccbc1 LP |
3123 | r = cg_all_unified(); |
3124 | if (r < 0) | |
3125 | return r; | |
3126 | if (r > 0) | |
efdb0237 | 3127 | r = cg_get_attribute("memory", u->cgroup_path, "memory.current", &v); |
b4cccbc1 LP |
3128 | else |
3129 | r = cg_get_attribute("memory", u->cgroup_path, "memory.usage_in_bytes", &v); | |
5ad096b3 LP |
3130 | if (r == -ENOENT) |
3131 | return -ENODATA; | |
3132 | if (r < 0) | |
3133 | return r; | |
3134 | ||
3135 | return safe_atou64(v, ret); | |
3136 | } | |
3137 | ||
03a7b521 LP |
3138 | int unit_get_tasks_current(Unit *u, uint64_t *ret) { |
3139 | _cleanup_free_ char *v = NULL; | |
3140 | int r; | |
3141 | ||
3142 | assert(u); | |
3143 | assert(ret); | |
3144 | ||
2e4025c0 | 3145 | if (!UNIT_CGROUP_BOOL(u, tasks_accounting)) |
cf3b4be1 LP |
3146 | return -ENODATA; |
3147 | ||
03a7b521 LP |
3148 | if (!u->cgroup_path) |
3149 | return -ENODATA; | |
3150 | ||
c36a69f4 | 3151 | /* The root cgroup doesn't expose this information, let's get it from /proc instead */ |
611c4f8a | 3152 | if (unit_has_host_root_cgroup(u)) |
c36a69f4 LP |
3153 | return procfs_tasks_get_current(ret); |
3154 | ||
1f73aa00 LP |
3155 | if ((u->cgroup_realized_mask & CGROUP_MASK_PIDS) == 0) |
3156 | return -ENODATA; | |
3157 | ||
03a7b521 LP |
3158 | r = cg_get_attribute("pids", u->cgroup_path, "pids.current", &v); |
3159 | if (r == -ENOENT) | |
3160 | return -ENODATA; | |
3161 | if (r < 0) | |
3162 | return r; | |
3163 | ||
3164 | return safe_atou64(v, ret); | |
3165 | } | |
3166 | ||
5ad096b3 LP |
3167 | static int unit_get_cpu_usage_raw(Unit *u, nsec_t *ret) { |
3168 | _cleanup_free_ char *v = NULL; | |
3169 | uint64_t ns; | |
3170 | int r; | |
3171 | ||
3172 | assert(u); | |
3173 | assert(ret); | |
3174 | ||
3175 | if (!u->cgroup_path) | |
3176 | return -ENODATA; | |
3177 | ||
1f73aa00 | 3178 | /* The root cgroup doesn't expose this information, let's get it from /proc instead */ |
611c4f8a | 3179 | if (unit_has_host_root_cgroup(u)) |
1f73aa00 LP |
3180 | return procfs_cpu_get_usage(ret); |
3181 | ||
f98c2585 CD |
3182 | /* Requisite controllers for CPU accounting are not enabled */ |
3183 | if ((get_cpu_accounting_mask() & ~u->cgroup_realized_mask) != 0) | |
3184 | return -ENODATA; | |
3185 | ||
92a99304 LP |
3186 | r = cg_all_unified(); |
3187 | if (r < 0) | |
3188 | return r; | |
b4cccbc1 | 3189 | if (r > 0) { |
66ebf6c0 TH |
3190 | _cleanup_free_ char *val = NULL; |
3191 | uint64_t us; | |
5ad096b3 | 3192 | |
b734a4ff | 3193 | r = cg_get_keyed_attribute("cpu", u->cgroup_path, "cpu.stat", STRV_MAKE("usage_usec"), &val); |
b734a4ff LP |
3194 | if (IN_SET(r, -ENOENT, -ENXIO)) |
3195 | return -ENODATA; | |
d742f4b5 LP |
3196 | if (r < 0) |
3197 | return r; | |
66ebf6c0 TH |
3198 | |
3199 | r = safe_atou64(val, &us); | |
3200 | if (r < 0) | |
3201 | return r; | |
3202 | ||
3203 | ns = us * NSEC_PER_USEC; | |
3204 | } else { | |
66ebf6c0 TH |
3205 | r = cg_get_attribute("cpuacct", u->cgroup_path, "cpuacct.usage", &v); |
3206 | if (r == -ENOENT) | |
3207 | return -ENODATA; | |
3208 | if (r < 0) | |
3209 | return r; | |
3210 | ||
3211 | r = safe_atou64(v, &ns); | |
3212 | if (r < 0) | |
3213 | return r; | |
3214 | } | |
5ad096b3 LP |
3215 | |
3216 | *ret = ns; | |
3217 | return 0; | |
3218 | } | |
3219 | ||
3220 | int unit_get_cpu_usage(Unit *u, nsec_t *ret) { | |
3221 | nsec_t ns; | |
3222 | int r; | |
3223 | ||
fe700f46 LP |
3224 | assert(u); |
3225 | ||
3226 | /* Retrieve the current CPU usage counter. This will subtract the CPU counter taken when the unit was | |
3227 | * started. If the cgroup has been removed already, returns the last cached value. To cache the value, simply | |
3228 | * call this function with a NULL return value. */ | |
3229 | ||
2e4025c0 | 3230 | if (!UNIT_CGROUP_BOOL(u, cpu_accounting)) |
cf3b4be1 LP |
3231 | return -ENODATA; |
3232 | ||
5ad096b3 | 3233 | r = unit_get_cpu_usage_raw(u, &ns); |
fe700f46 LP |
3234 | if (r == -ENODATA && u->cpu_usage_last != NSEC_INFINITY) { |
3235 | /* If we can't get the CPU usage anymore (because the cgroup was already removed, for example), use our | |
3236 | * cached value. */ | |
3237 | ||
3238 | if (ret) | |
3239 | *ret = u->cpu_usage_last; | |
3240 | return 0; | |
3241 | } | |
5ad096b3 LP |
3242 | if (r < 0) |
3243 | return r; | |
3244 | ||
66ebf6c0 TH |
3245 | if (ns > u->cpu_usage_base) |
3246 | ns -= u->cpu_usage_base; | |
5ad096b3 LP |
3247 | else |
3248 | ns = 0; | |
3249 | ||
fe700f46 LP |
3250 | u->cpu_usage_last = ns; |
3251 | if (ret) | |
3252 | *ret = ns; | |
3253 | ||
5ad096b3 LP |
3254 | return 0; |
3255 | } | |
3256 | ||
906c06f6 DM |
3257 | int unit_get_ip_accounting( |
3258 | Unit *u, | |
3259 | CGroupIPAccountingMetric metric, | |
3260 | uint64_t *ret) { | |
3261 | ||
6b659ed8 | 3262 | uint64_t value; |
906c06f6 DM |
3263 | int fd, r; |
3264 | ||
3265 | assert(u); | |
3266 | assert(metric >= 0); | |
3267 | assert(metric < _CGROUP_IP_ACCOUNTING_METRIC_MAX); | |
3268 | assert(ret); | |
3269 | ||
2e4025c0 | 3270 | if (!UNIT_CGROUP_BOOL(u, ip_accounting)) |
cf3b4be1 LP |
3271 | return -ENODATA; |
3272 | ||
906c06f6 DM |
3273 | fd = IN_SET(metric, CGROUP_IP_INGRESS_BYTES, CGROUP_IP_INGRESS_PACKETS) ? |
3274 | u->ip_accounting_ingress_map_fd : | |
3275 | u->ip_accounting_egress_map_fd; | |
906c06f6 DM |
3276 | if (fd < 0) |
3277 | return -ENODATA; | |
3278 | ||
3279 | if (IN_SET(metric, CGROUP_IP_INGRESS_BYTES, CGROUP_IP_EGRESS_BYTES)) | |
6b659ed8 | 3280 | r = bpf_firewall_read_accounting(fd, &value, NULL); |
906c06f6 | 3281 | else |
6b659ed8 LP |
3282 | r = bpf_firewall_read_accounting(fd, NULL, &value); |
3283 | if (r < 0) | |
3284 | return r; | |
3285 | ||
3286 | /* Add in additional metrics from a previous runtime. Note that when reexecing/reloading the daemon we compile | |
3287 | * all BPF programs and maps anew, but serialize the old counters. When deserializing we store them in the | |
3288 | * ip_accounting_extra[] field, and add them in here transparently. */ | |
3289 | ||
3290 | *ret = value + u->ip_accounting_extra[metric]; | |
906c06f6 DM |
3291 | |
3292 | return r; | |
3293 | } | |
3294 | ||
fbe14fc9 LP |
3295 | static int unit_get_io_accounting_raw(Unit *u, uint64_t ret[static _CGROUP_IO_ACCOUNTING_METRIC_MAX]) { |
3296 | static const char *const field_names[_CGROUP_IO_ACCOUNTING_METRIC_MAX] = { | |
3297 | [CGROUP_IO_READ_BYTES] = "rbytes=", | |
3298 | [CGROUP_IO_WRITE_BYTES] = "wbytes=", | |
3299 | [CGROUP_IO_READ_OPERATIONS] = "rios=", | |
3300 | [CGROUP_IO_WRITE_OPERATIONS] = "wios=", | |
3301 | }; | |
3302 | uint64_t acc[_CGROUP_IO_ACCOUNTING_METRIC_MAX] = {}; | |
3303 | _cleanup_free_ char *path = NULL; | |
3304 | _cleanup_fclose_ FILE *f = NULL; | |
3305 | int r; | |
3306 | ||
3307 | assert(u); | |
3308 | ||
3309 | if (!u->cgroup_path) | |
3310 | return -ENODATA; | |
3311 | ||
3312 | if (unit_has_host_root_cgroup(u)) | |
3313 | return -ENODATA; /* TODO: return useful data for the top-level cgroup */ | |
3314 | ||
3315 | r = cg_all_unified(); | |
3316 | if (r < 0) | |
3317 | return r; | |
3318 | if (r == 0) /* TODO: support cgroupv1 */ | |
3319 | return -ENODATA; | |
3320 | ||
3321 | if (!FLAGS_SET(u->cgroup_realized_mask, CGROUP_MASK_IO)) | |
3322 | return -ENODATA; | |
3323 | ||
3324 | r = cg_get_path("io", u->cgroup_path, "io.stat", &path); | |
3325 | if (r < 0) | |
3326 | return r; | |
3327 | ||
3328 | f = fopen(path, "re"); | |
3329 | if (!f) | |
3330 | return -errno; | |
3331 | ||
3332 | for (;;) { | |
3333 | _cleanup_free_ char *line = NULL; | |
3334 | const char *p; | |
3335 | ||
3336 | r = read_line(f, LONG_LINE_MAX, &line); | |
3337 | if (r < 0) | |
3338 | return r; | |
3339 | if (r == 0) | |
3340 | break; | |
3341 | ||
3342 | p = line; | |
3343 | p += strcspn(p, WHITESPACE); /* Skip over device major/minor */ | |
3344 | p += strspn(p, WHITESPACE); /* Skip over following whitespace */ | |
3345 | ||
3346 | for (;;) { | |
3347 | _cleanup_free_ char *word = NULL; | |
3348 | ||
3349 | r = extract_first_word(&p, &word, NULL, EXTRACT_RETAIN_ESCAPE); | |
3350 | if (r < 0) | |
3351 | return r; | |
3352 | if (r == 0) | |
3353 | break; | |
3354 | ||
3355 | for (CGroupIOAccountingMetric i = 0; i < _CGROUP_IO_ACCOUNTING_METRIC_MAX; i++) { | |
3356 | const char *x; | |
3357 | ||
3358 | x = startswith(word, field_names[i]); | |
3359 | if (x) { | |
3360 | uint64_t w; | |
3361 | ||
3362 | r = safe_atou64(x, &w); | |
3363 | if (r < 0) | |
3364 | return r; | |
3365 | ||
3366 | /* Sum up the stats of all devices */ | |
3367 | acc[i] += w; | |
3368 | break; | |
3369 | } | |
3370 | } | |
3371 | } | |
3372 | } | |
3373 | ||
3374 | memcpy(ret, acc, sizeof(acc)); | |
3375 | return 0; | |
3376 | } | |
3377 | ||
3378 | int unit_get_io_accounting( | |
3379 | Unit *u, | |
3380 | CGroupIOAccountingMetric metric, | |
3381 | bool allow_cache, | |
3382 | uint64_t *ret) { | |
3383 | ||
3384 | uint64_t raw[_CGROUP_IO_ACCOUNTING_METRIC_MAX]; | |
3385 | int r; | |
3386 | ||
3387 | /* Retrieve an IO account parameter. This will subtract the counter when the unit was started. */ | |
3388 | ||
3389 | if (!UNIT_CGROUP_BOOL(u, io_accounting)) | |
3390 | return -ENODATA; | |
3391 | ||
3392 | if (allow_cache && u->io_accounting_last[metric] != UINT64_MAX) | |
3393 | goto done; | |
3394 | ||
3395 | r = unit_get_io_accounting_raw(u, raw); | |
3396 | if (r == -ENODATA && u->io_accounting_last[metric] != UINT64_MAX) | |
3397 | goto done; | |
3398 | if (r < 0) | |
3399 | return r; | |
3400 | ||
3401 | for (CGroupIOAccountingMetric i = 0; i < _CGROUP_IO_ACCOUNTING_METRIC_MAX; i++) { | |
3402 | /* Saturated subtraction */ | |
3403 | if (raw[i] > u->io_accounting_base[i]) | |
3404 | u->io_accounting_last[i] = raw[i] - u->io_accounting_base[i]; | |
3405 | else | |
3406 | u->io_accounting_last[i] = 0; | |
3407 | } | |
3408 | ||
3409 | done: | |
3410 | if (ret) | |
3411 | *ret = u->io_accounting_last[metric]; | |
3412 | ||
3413 | return 0; | |
3414 | } | |
3415 | ||
906c06f6 | 3416 | int unit_reset_cpu_accounting(Unit *u) { |
5ad096b3 LP |
3417 | int r; |
3418 | ||
3419 | assert(u); | |
3420 | ||
fe700f46 LP |
3421 | u->cpu_usage_last = NSEC_INFINITY; |
3422 | ||
0bbff7d6 | 3423 | r = unit_get_cpu_usage_raw(u, &u->cpu_usage_base); |
5ad096b3 | 3424 | if (r < 0) { |
66ebf6c0 | 3425 | u->cpu_usage_base = 0; |
5ad096b3 | 3426 | return r; |
b56c28c3 | 3427 | } |
2633eb83 | 3428 | |
4ad49000 | 3429 | return 0; |
4fbf50b3 LP |
3430 | } |
3431 | ||
906c06f6 DM |
3432 | int unit_reset_ip_accounting(Unit *u) { |
3433 | int r = 0, q = 0; | |
3434 | ||
3435 | assert(u); | |
3436 | ||
3437 | if (u->ip_accounting_ingress_map_fd >= 0) | |
3438 | r = bpf_firewall_reset_accounting(u->ip_accounting_ingress_map_fd); | |
3439 | ||
3440 | if (u->ip_accounting_egress_map_fd >= 0) | |
3441 | q = bpf_firewall_reset_accounting(u->ip_accounting_egress_map_fd); | |
3442 | ||
6b659ed8 LP |
3443 | zero(u->ip_accounting_extra); |
3444 | ||
906c06f6 DM |
3445 | return r < 0 ? r : q; |
3446 | } | |
3447 | ||
fbe14fc9 LP |
3448 | int unit_reset_io_accounting(Unit *u) { |
3449 | int r; | |
3450 | ||
3451 | assert(u); | |
3452 | ||
3453 | for (CGroupIOAccountingMetric i = 0; i < _CGROUP_IO_ACCOUNTING_METRIC_MAX; i++) | |
3454 | u->io_accounting_last[i] = UINT64_MAX; | |
3455 | ||
3456 | r = unit_get_io_accounting_raw(u, u->io_accounting_base); | |
3457 | if (r < 0) { | |
3458 | zero(u->io_accounting_base); | |
3459 | return r; | |
3460 | } | |
3461 | ||
3462 | return 0; | |
3463 | } | |
3464 | ||
9b2559a1 | 3465 | int unit_reset_accounting(Unit *u) { |
fbe14fc9 | 3466 | int r, q, v; |
9b2559a1 LP |
3467 | |
3468 | assert(u); | |
3469 | ||
3470 | r = unit_reset_cpu_accounting(u); | |
fbe14fc9 LP |
3471 | q = unit_reset_io_accounting(u); |
3472 | v = unit_reset_ip_accounting(u); | |
9b2559a1 | 3473 | |
fbe14fc9 | 3474 | return r < 0 ? r : q < 0 ? q : v; |
9b2559a1 LP |
3475 | } |
3476 | ||
e7ab4d1a LP |
3477 | void unit_invalidate_cgroup(Unit *u, CGroupMask m) { |
3478 | assert(u); | |
3479 | ||
3480 | if (!UNIT_HAS_CGROUP_CONTEXT(u)) | |
3481 | return; | |
3482 | ||
3483 | if (m == 0) | |
3484 | return; | |
3485 | ||
538b4852 TH |
3486 | /* always invalidate compat pairs together */ |
3487 | if (m & (CGROUP_MASK_IO | CGROUP_MASK_BLKIO)) | |
3488 | m |= CGROUP_MASK_IO | CGROUP_MASK_BLKIO; | |
3489 | ||
7cce4fb7 LP |
3490 | if (m & (CGROUP_MASK_CPU | CGROUP_MASK_CPUACCT)) |
3491 | m |= CGROUP_MASK_CPU | CGROUP_MASK_CPUACCT; | |
3492 | ||
e00068e7 | 3493 | if (FLAGS_SET(u->cgroup_invalidated_mask, m)) /* NOP? */ |
e7ab4d1a LP |
3494 | return; |
3495 | ||
e00068e7 | 3496 | u->cgroup_invalidated_mask |= m; |
91a6073e | 3497 | unit_add_to_cgroup_realize_queue(u); |
e7ab4d1a LP |
3498 | } |
3499 | ||
906c06f6 DM |
3500 | void unit_invalidate_cgroup_bpf(Unit *u) { |
3501 | assert(u); | |
3502 | ||
3503 | if (!UNIT_HAS_CGROUP_CONTEXT(u)) | |
3504 | return; | |
3505 | ||
17f14955 | 3506 | if (u->cgroup_invalidated_mask & CGROUP_MASK_BPF_FIREWALL) /* NOP? */ |
906c06f6 DM |
3507 | return; |
3508 | ||
17f14955 | 3509 | u->cgroup_invalidated_mask |= CGROUP_MASK_BPF_FIREWALL; |
91a6073e | 3510 | unit_add_to_cgroup_realize_queue(u); |
906c06f6 DM |
3511 | |
3512 | /* If we are a slice unit, we also need to put compile a new BPF program for all our children, as the IP access | |
3513 | * list of our children includes our own. */ | |
3514 | if (u->type == UNIT_SLICE) { | |
3515 | Unit *member; | |
3516 | Iterator i; | |
eef85c4a | 3517 | void *v; |
906c06f6 | 3518 | |
eef85c4a | 3519 | HASHMAP_FOREACH_KEY(v, member, u->dependencies[UNIT_BEFORE], i) { |
cb5e3bc3 CD |
3520 | if (UNIT_DEREF(member->slice) == u) |
3521 | unit_invalidate_cgroup_bpf(member); | |
906c06f6 DM |
3522 | } |
3523 | } | |
3524 | } | |
3525 | ||
1d9cc876 LP |
3526 | bool unit_cgroup_delegate(Unit *u) { |
3527 | CGroupContext *c; | |
3528 | ||
3529 | assert(u); | |
3530 | ||
3531 | if (!UNIT_VTABLE(u)->can_delegate) | |
3532 | return false; | |
3533 | ||
3534 | c = unit_get_cgroup_context(u); | |
3535 | if (!c) | |
3536 | return false; | |
3537 | ||
3538 | return c->delegate; | |
3539 | } | |
3540 | ||
e7ab4d1a LP |
3541 | void manager_invalidate_startup_units(Manager *m) { |
3542 | Iterator i; | |
3543 | Unit *u; | |
3544 | ||
3545 | assert(m); | |
3546 | ||
3547 | SET_FOREACH(u, m->startup_units, i) | |
13c31542 | 3548 | unit_invalidate_cgroup(u, CGROUP_MASK_CPU|CGROUP_MASK_IO|CGROUP_MASK_BLKIO); |
e7ab4d1a LP |
3549 | } |
3550 | ||
da8e1782 MO |
3551 | static int unit_get_nice(Unit *u) { |
3552 | ExecContext *ec; | |
3553 | ||
3554 | ec = unit_get_exec_context(u); | |
3555 | return ec ? ec->nice : 0; | |
3556 | } | |
3557 | ||
3558 | static uint64_t unit_get_cpu_weight(Unit *u) { | |
3559 | ManagerState state = manager_state(u->manager); | |
3560 | CGroupContext *cc; | |
3561 | ||
3562 | cc = unit_get_cgroup_context(u); | |
3563 | return cc ? cgroup_context_cpu_weight(cc, state) : CGROUP_WEIGHT_DEFAULT; | |
3564 | } | |
3565 | ||
3566 | int compare_job_priority(const void *a, const void *b) { | |
3567 | const Job *x = a, *y = b; | |
3568 | int nice_x, nice_y; | |
3569 | uint64_t weight_x, weight_y; | |
3570 | int ret; | |
3571 | ||
217b7b33 ZJS |
3572 | if ((ret = CMP(x->unit->type, y->unit->type)) != 0) |
3573 | return -ret; | |
3574 | ||
da8e1782 MO |
3575 | weight_x = unit_get_cpu_weight(x->unit); |
3576 | weight_y = unit_get_cpu_weight(y->unit); | |
3577 | ||
217b7b33 ZJS |
3578 | if ((ret = CMP(weight_x, weight_y)) != 0) |
3579 | return -ret; | |
da8e1782 MO |
3580 | |
3581 | nice_x = unit_get_nice(x->unit); | |
3582 | nice_y = unit_get_nice(y->unit); | |
3583 | ||
3584 | if ((ret = CMP(nice_x, nice_y)) != 0) | |
3585 | return ret; | |
3586 | ||
da8e1782 MO |
3587 | return strcmp(x->unit->id, y->unit->id); |
3588 | } | |
3589 | ||
4ad49000 | 3590 | static const char* const cgroup_device_policy_table[_CGROUP_DEVICE_POLICY_MAX] = { |
084870f9 ZJS |
3591 | [CGROUP_DEVICE_POLICY_AUTO] = "auto", |
3592 | [CGROUP_DEVICE_POLICY_CLOSED] = "closed", | |
3593 | [CGROUP_DEVICE_POLICY_STRICT] = "strict", | |
4ad49000 | 3594 | }; |
4fbf50b3 | 3595 | |
047f5d63 PH |
3596 | int unit_get_cpuset(Unit *u, CPUSet *cpus, const char *name) { |
3597 | _cleanup_free_ char *v = NULL; | |
3598 | int r; | |
3599 | ||
3600 | assert(u); | |
3601 | assert(cpus); | |
3602 | ||
3603 | if (!u->cgroup_path) | |
3604 | return -ENODATA; | |
3605 | ||
3606 | if ((u->cgroup_realized_mask & CGROUP_MASK_CPUSET) == 0) | |
3607 | return -ENODATA; | |
3608 | ||
3609 | r = cg_all_unified(); | |
3610 | if (r < 0) | |
3611 | return r; | |
3612 | if (r == 0) | |
3613 | return -ENODATA; | |
3614 | if (r > 0) | |
3615 | r = cg_get_attribute("cpuset", u->cgroup_path, name, &v); | |
3616 | if (r == -ENOENT) | |
3617 | return -ENODATA; | |
3618 | if (r < 0) | |
3619 | return r; | |
3620 | ||
3621 | return parse_cpu_set_full(v, cpus, false, NULL, NULL, 0, NULL); | |
3622 | } | |
3623 | ||
4ad49000 | 3624 | DEFINE_STRING_TABLE_LOOKUP(cgroup_device_policy, CGroupDevicePolicy); |