]>
Commit | Line | Data |
---|---|---|
53e1b683 | 1 | /* SPDX-License-Identifier: LGPL-2.1+ */ |
8e274523 | 2 | |
c6c18be3 | 3 | #include <fcntl.h> |
e41969e3 | 4 | #include <fnmatch.h> |
8c6db833 | 5 | |
afcfaa69 LP |
6 | #include "sd-messages.h" |
7 | ||
b5efdb8a | 8 | #include "alloc-util.h" |
18c528e9 | 9 | #include "blockdev-util.h" |
d8b4d14d | 10 | #include "bpf-devices.h" |
906c06f6 | 11 | #include "bpf-firewall.h" |
45c2e068 | 12 | #include "btrfs-util.h" |
6592b975 | 13 | #include "bus-error.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" |
d8b4d14d | 19 | #include "nulstr-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 | ||
611c4f8a | 38 | bool manager_owns_host_root_cgroup(Manager *m) { |
cc6271f1 LP |
39 | assert(m); |
40 | ||
41 | /* Returns true if we are managing the root cgroup. Note that it isn't sufficient to just check whether the | |
42 | * group root path equals "/" since that will also be the case if CLONE_NEWCGROUP is in the mix. Since there's | |
43 | * appears to be no nice way to detect whether we are in a CLONE_NEWCGROUP namespace we instead just check if | |
44 | * we run in any kind of container virtualization. */ | |
45 | ||
28cfdc5a LP |
46 | if (MANAGER_IS_USER(m)) |
47 | return false; | |
48 | ||
cc6271f1 LP |
49 | if (detect_container() > 0) |
50 | return false; | |
51 | ||
57ea45e1 | 52 | return empty_or_root(m->cgroup_root); |
cc6271f1 LP |
53 | } |
54 | ||
611c4f8a | 55 | bool unit_has_host_root_cgroup(Unit *u) { |
f3725e64 LP |
56 | assert(u); |
57 | ||
cc6271f1 LP |
58 | /* Returns whether this unit manages the root cgroup. This will return true if this unit is the root slice and |
59 | * the manager manages the root cgroup. */ | |
f3725e64 | 60 | |
611c4f8a | 61 | if (!manager_owns_host_root_cgroup(u->manager)) |
f3725e64 LP |
62 | return false; |
63 | ||
cc6271f1 | 64 | return unit_has_name(u, SPECIAL_ROOT_SLICE); |
f3725e64 LP |
65 | } |
66 | ||
293d32df LP |
67 | static int set_attribute_and_warn(Unit *u, const char *controller, const char *attribute, const char *value) { |
68 | int r; | |
69 | ||
70 | r = cg_set_attribute(controller, u->cgroup_path, attribute, value); | |
71 | if (r < 0) | |
72 | log_unit_full(u, LOG_LEVEL_CGROUP_WRITE(r), r, "Failed to set '%s' attribute on '%s' to '%.*s': %m", | |
73 | strna(attribute), isempty(u->cgroup_path) ? "/" : u->cgroup_path, (int) strcspn(value, NEWLINE), value); | |
74 | ||
75 | return r; | |
76 | } | |
77 | ||
2b40998d | 78 | static void cgroup_compat_warn(void) { |
128fadc9 TH |
79 | static bool cgroup_compat_warned = false; |
80 | ||
81 | if (cgroup_compat_warned) | |
82 | return; | |
83 | ||
cc6271f1 LP |
84 | log_warning("cgroup compatibility translation between legacy and unified hierarchy settings activated. " |
85 | "See cgroup-compat debug messages for details."); | |
86 | ||
128fadc9 TH |
87 | cgroup_compat_warned = true; |
88 | } | |
89 | ||
90 | #define log_cgroup_compat(unit, fmt, ...) do { \ | |
91 | cgroup_compat_warn(); \ | |
92 | log_unit_debug(unit, "cgroup-compat: " fmt, ##__VA_ARGS__); \ | |
2b40998d | 93 | } while (false) |
128fadc9 | 94 | |
4ad49000 LP |
95 | void cgroup_context_init(CGroupContext *c) { |
96 | assert(c); | |
97 | ||
de8a711a | 98 | /* Initialize everything to the kernel defaults. */ |
4ad49000 | 99 | |
de8a711a LP |
100 | *c = (CGroupContext) { |
101 | .cpu_weight = CGROUP_WEIGHT_INVALID, | |
102 | .startup_cpu_weight = CGROUP_WEIGHT_INVALID, | |
103 | .cpu_quota_per_sec_usec = USEC_INFINITY, | |
10f28641 | 104 | .cpu_quota_period_usec = USEC_INFINITY, |
66ebf6c0 | 105 | |
de8a711a LP |
106 | .cpu_shares = CGROUP_CPU_SHARES_INVALID, |
107 | .startup_cpu_shares = CGROUP_CPU_SHARES_INVALID, | |
d53d9474 | 108 | |
de8a711a LP |
109 | .memory_high = CGROUP_LIMIT_MAX, |
110 | .memory_max = CGROUP_LIMIT_MAX, | |
111 | .memory_swap_max = CGROUP_LIMIT_MAX, | |
da4d897e | 112 | |
de8a711a | 113 | .memory_limit = CGROUP_LIMIT_MAX, |
b2f8b02e | 114 | |
de8a711a LP |
115 | .io_weight = CGROUP_WEIGHT_INVALID, |
116 | .startup_io_weight = CGROUP_WEIGHT_INVALID, | |
13c31542 | 117 | |
de8a711a LP |
118 | .blockio_weight = CGROUP_BLKIO_WEIGHT_INVALID, |
119 | .startup_blockio_weight = CGROUP_BLKIO_WEIGHT_INVALID, | |
d53d9474 | 120 | |
de8a711a LP |
121 | .tasks_max = CGROUP_LIMIT_MAX, |
122 | }; | |
4ad49000 | 123 | } |
8e274523 | 124 | |
4ad49000 LP |
125 | void cgroup_context_free_device_allow(CGroupContext *c, CGroupDeviceAllow *a) { |
126 | assert(c); | |
127 | assert(a); | |
128 | ||
71fda00f | 129 | LIST_REMOVE(device_allow, c->device_allow, a); |
4ad49000 LP |
130 | free(a->path); |
131 | free(a); | |
132 | } | |
133 | ||
13c31542 TH |
134 | void cgroup_context_free_io_device_weight(CGroupContext *c, CGroupIODeviceWeight *w) { |
135 | assert(c); | |
136 | assert(w); | |
137 | ||
138 | LIST_REMOVE(device_weights, c->io_device_weights, w); | |
139 | free(w->path); | |
140 | free(w); | |
141 | } | |
142 | ||
6ae4283c TH |
143 | void cgroup_context_free_io_device_latency(CGroupContext *c, CGroupIODeviceLatency *l) { |
144 | assert(c); | |
145 | assert(l); | |
146 | ||
147 | LIST_REMOVE(device_latencies, c->io_device_latencies, l); | |
148 | free(l->path); | |
149 | free(l); | |
150 | } | |
151 | ||
13c31542 TH |
152 | void cgroup_context_free_io_device_limit(CGroupContext *c, CGroupIODeviceLimit *l) { |
153 | assert(c); | |
154 | assert(l); | |
155 | ||
156 | LIST_REMOVE(device_limits, c->io_device_limits, l); | |
157 | free(l->path); | |
158 | free(l); | |
159 | } | |
160 | ||
4ad49000 LP |
161 | void cgroup_context_free_blockio_device_weight(CGroupContext *c, CGroupBlockIODeviceWeight *w) { |
162 | assert(c); | |
163 | assert(w); | |
164 | ||
71fda00f | 165 | LIST_REMOVE(device_weights, c->blockio_device_weights, w); |
4ad49000 LP |
166 | free(w->path); |
167 | free(w); | |
168 | } | |
169 | ||
170 | void cgroup_context_free_blockio_device_bandwidth(CGroupContext *c, CGroupBlockIODeviceBandwidth *b) { | |
171 | assert(c); | |
8e274523 | 172 | assert(b); |
8e274523 | 173 | |
71fda00f | 174 | LIST_REMOVE(device_bandwidths, c->blockio_device_bandwidths, b); |
4ad49000 LP |
175 | free(b->path); |
176 | free(b); | |
177 | } | |
178 | ||
179 | void cgroup_context_done(CGroupContext *c) { | |
180 | assert(c); | |
181 | ||
13c31542 TH |
182 | while (c->io_device_weights) |
183 | cgroup_context_free_io_device_weight(c, c->io_device_weights); | |
184 | ||
6ae4283c TH |
185 | while (c->io_device_latencies) |
186 | cgroup_context_free_io_device_latency(c, c->io_device_latencies); | |
187 | ||
13c31542 TH |
188 | while (c->io_device_limits) |
189 | cgroup_context_free_io_device_limit(c, c->io_device_limits); | |
190 | ||
4ad49000 LP |
191 | while (c->blockio_device_weights) |
192 | cgroup_context_free_blockio_device_weight(c, c->blockio_device_weights); | |
193 | ||
194 | while (c->blockio_device_bandwidths) | |
195 | cgroup_context_free_blockio_device_bandwidth(c, c->blockio_device_bandwidths); | |
196 | ||
197 | while (c->device_allow) | |
198 | cgroup_context_free_device_allow(c, c->device_allow); | |
6a48d82f DM |
199 | |
200 | c->ip_address_allow = ip_address_access_free_all(c->ip_address_allow); | |
201 | c->ip_address_deny = ip_address_access_free_all(c->ip_address_deny); | |
fab34748 KL |
202 | |
203 | c->ip_filters_ingress = strv_free(c->ip_filters_ingress); | |
204 | c->ip_filters_egress = strv_free(c->ip_filters_egress); | |
4ad49000 LP |
205 | } |
206 | ||
207 | void cgroup_context_dump(CGroupContext *c, FILE* f, const char *prefix) { | |
25cc30c4 | 208 | _cleanup_free_ char *disable_controllers_str = NULL; |
13c31542 TH |
209 | CGroupIODeviceLimit *il; |
210 | CGroupIODeviceWeight *iw; | |
6ae4283c | 211 | CGroupIODeviceLatency *l; |
4ad49000 LP |
212 | CGroupBlockIODeviceBandwidth *b; |
213 | CGroupBlockIODeviceWeight *w; | |
214 | CGroupDeviceAllow *a; | |
c21c9906 | 215 | IPAddressAccessItem *iaai; |
fab34748 | 216 | char **path; |
9a054909 | 217 | char u[FORMAT_TIMESPAN_MAX]; |
10f28641 | 218 | char v[FORMAT_TIMESPAN_MAX]; |
4ad49000 LP |
219 | |
220 | assert(c); | |
221 | assert(f); | |
222 | ||
223 | prefix = strempty(prefix); | |
224 | ||
25cc30c4 AZ |
225 | (void) cg_mask_to_string(c->disable_controllers, &disable_controllers_str); |
226 | ||
4ad49000 LP |
227 | fprintf(f, |
228 | "%sCPUAccounting=%s\n" | |
13c31542 | 229 | "%sIOAccounting=%s\n" |
4ad49000 LP |
230 | "%sBlockIOAccounting=%s\n" |
231 | "%sMemoryAccounting=%s\n" | |
d53d9474 | 232 | "%sTasksAccounting=%s\n" |
c21c9906 | 233 | "%sIPAccounting=%s\n" |
66ebf6c0 TH |
234 | "%sCPUWeight=%" PRIu64 "\n" |
235 | "%sStartupCPUWeight=%" PRIu64 "\n" | |
d53d9474 LP |
236 | "%sCPUShares=%" PRIu64 "\n" |
237 | "%sStartupCPUShares=%" PRIu64 "\n" | |
b2f8b02e | 238 | "%sCPUQuotaPerSecSec=%s\n" |
10f28641 | 239 | "%sCPUQuotaPeriodSec=%s\n" |
13c31542 TH |
240 | "%sIOWeight=%" PRIu64 "\n" |
241 | "%sStartupIOWeight=%" PRIu64 "\n" | |
d53d9474 LP |
242 | "%sBlockIOWeight=%" PRIu64 "\n" |
243 | "%sStartupBlockIOWeight=%" PRIu64 "\n" | |
7ad5439e | 244 | "%sDefaultMemoryMin=%" PRIu64 "\n" |
c52db42b | 245 | "%sDefaultMemoryLow=%" PRIu64 "\n" |
48422635 | 246 | "%sMemoryMin=%" PRIu64 "\n" |
da4d897e TH |
247 | "%sMemoryLow=%" PRIu64 "\n" |
248 | "%sMemoryHigh=%" PRIu64 "\n" | |
249 | "%sMemoryMax=%" PRIu64 "\n" | |
96e131ea | 250 | "%sMemorySwapMax=%" PRIu64 "\n" |
4ad49000 | 251 | "%sMemoryLimit=%" PRIu64 "\n" |
03a7b521 | 252 | "%sTasksMax=%" PRIu64 "\n" |
a931ad47 | 253 | "%sDevicePolicy=%s\n" |
25cc30c4 | 254 | "%sDisableControllers=%s\n" |
a931ad47 | 255 | "%sDelegate=%s\n", |
4ad49000 | 256 | prefix, yes_no(c->cpu_accounting), |
13c31542 | 257 | prefix, yes_no(c->io_accounting), |
4ad49000 LP |
258 | prefix, yes_no(c->blockio_accounting), |
259 | prefix, yes_no(c->memory_accounting), | |
d53d9474 | 260 | prefix, yes_no(c->tasks_accounting), |
c21c9906 | 261 | prefix, yes_no(c->ip_accounting), |
66ebf6c0 TH |
262 | prefix, c->cpu_weight, |
263 | prefix, c->startup_cpu_weight, | |
4ad49000 | 264 | prefix, c->cpu_shares, |
95ae05c0 | 265 | prefix, c->startup_cpu_shares, |
b1d6dcf5 | 266 | prefix, format_timespan(u, sizeof(u), c->cpu_quota_per_sec_usec, 1), |
10f28641 | 267 | prefix, format_timespan(v, sizeof(v), c->cpu_quota_period_usec, 1), |
13c31542 TH |
268 | prefix, c->io_weight, |
269 | prefix, c->startup_io_weight, | |
4ad49000 | 270 | prefix, c->blockio_weight, |
95ae05c0 | 271 | prefix, c->startup_blockio_weight, |
7ad5439e | 272 | prefix, c->default_memory_min, |
c52db42b | 273 | prefix, c->default_memory_low, |
48422635 | 274 | prefix, c->memory_min, |
da4d897e TH |
275 | prefix, c->memory_low, |
276 | prefix, c->memory_high, | |
277 | prefix, c->memory_max, | |
96e131ea | 278 | prefix, c->memory_swap_max, |
4ad49000 | 279 | prefix, c->memory_limit, |
03a7b521 | 280 | prefix, c->tasks_max, |
a931ad47 | 281 | prefix, cgroup_device_policy_to_string(c->device_policy), |
f4c43a81 | 282 | prefix, strempty(disable_controllers_str), |
a931ad47 | 283 | prefix, yes_no(c->delegate)); |
4ad49000 | 284 | |
02638280 LP |
285 | if (c->delegate) { |
286 | _cleanup_free_ char *t = NULL; | |
287 | ||
288 | (void) cg_mask_to_string(c->delegate_controllers, &t); | |
289 | ||
47a78d41 | 290 | fprintf(f, "%sDelegateControllers=%s\n", |
02638280 LP |
291 | prefix, |
292 | strempty(t)); | |
293 | } | |
294 | ||
4ad49000 LP |
295 | LIST_FOREACH(device_allow, a, c->device_allow) |
296 | fprintf(f, | |
297 | "%sDeviceAllow=%s %s%s%s\n", | |
298 | prefix, | |
299 | a->path, | |
300 | a->r ? "r" : "", a->w ? "w" : "", a->m ? "m" : ""); | |
301 | ||
13c31542 TH |
302 | LIST_FOREACH(device_weights, iw, c->io_device_weights) |
303 | fprintf(f, | |
6ae4283c | 304 | "%sIODeviceWeight=%s %" PRIu64 "\n", |
13c31542 TH |
305 | prefix, |
306 | iw->path, | |
307 | iw->weight); | |
308 | ||
6ae4283c TH |
309 | LIST_FOREACH(device_latencies, l, c->io_device_latencies) |
310 | fprintf(f, | |
311 | "%sIODeviceLatencyTargetSec=%s %s\n", | |
312 | prefix, | |
313 | l->path, | |
314 | format_timespan(u, sizeof(u), l->target_usec, 1)); | |
315 | ||
13c31542 TH |
316 | LIST_FOREACH(device_limits, il, c->io_device_limits) { |
317 | char buf[FORMAT_BYTES_MAX]; | |
9be57249 TH |
318 | CGroupIOLimitType type; |
319 | ||
320 | for (type = 0; type < _CGROUP_IO_LIMIT_TYPE_MAX; type++) | |
321 | if (il->limits[type] != cgroup_io_limit_defaults[type]) | |
322 | fprintf(f, | |
323 | "%s%s=%s %s\n", | |
324 | prefix, | |
325 | cgroup_io_limit_type_to_string(type), | |
326 | il->path, | |
327 | format_bytes(buf, sizeof(buf), il->limits[type])); | |
13c31542 TH |
328 | } |
329 | ||
4ad49000 LP |
330 | LIST_FOREACH(device_weights, w, c->blockio_device_weights) |
331 | fprintf(f, | |
d53d9474 | 332 | "%sBlockIODeviceWeight=%s %" PRIu64, |
4ad49000 LP |
333 | prefix, |
334 | w->path, | |
335 | w->weight); | |
336 | ||
337 | LIST_FOREACH(device_bandwidths, b, c->blockio_device_bandwidths) { | |
338 | char buf[FORMAT_BYTES_MAX]; | |
339 | ||
979d0311 TH |
340 | if (b->rbps != CGROUP_LIMIT_MAX) |
341 | fprintf(f, | |
342 | "%sBlockIOReadBandwidth=%s %s\n", | |
343 | prefix, | |
344 | b->path, | |
345 | format_bytes(buf, sizeof(buf), b->rbps)); | |
346 | if (b->wbps != CGROUP_LIMIT_MAX) | |
347 | fprintf(f, | |
348 | "%sBlockIOWriteBandwidth=%s %s\n", | |
349 | prefix, | |
350 | b->path, | |
351 | format_bytes(buf, sizeof(buf), b->wbps)); | |
4ad49000 | 352 | } |
c21c9906 LP |
353 | |
354 | LIST_FOREACH(items, iaai, c->ip_address_allow) { | |
355 | _cleanup_free_ char *k = NULL; | |
356 | ||
357 | (void) in_addr_to_string(iaai->family, &iaai->address, &k); | |
358 | fprintf(f, "%sIPAddressAllow=%s/%u\n", prefix, strnull(k), iaai->prefixlen); | |
359 | } | |
360 | ||
361 | LIST_FOREACH(items, iaai, c->ip_address_deny) { | |
362 | _cleanup_free_ char *k = NULL; | |
363 | ||
364 | (void) in_addr_to_string(iaai->family, &iaai->address, &k); | |
365 | fprintf(f, "%sIPAddressDeny=%s/%u\n", prefix, strnull(k), iaai->prefixlen); | |
366 | } | |
fab34748 KL |
367 | |
368 | STRV_FOREACH(path, c->ip_filters_ingress) | |
369 | fprintf(f, "%sIPIngressFilterPath=%s\n", prefix, *path); | |
370 | ||
371 | STRV_FOREACH(path, c->ip_filters_egress) | |
372 | fprintf(f, "%sIPEgressFilterPath=%s\n", prefix, *path); | |
4ad49000 LP |
373 | } |
374 | ||
fd870bac YW |
375 | int cgroup_add_device_allow(CGroupContext *c, const char *dev, const char *mode) { |
376 | _cleanup_free_ CGroupDeviceAllow *a = NULL; | |
377 | _cleanup_free_ char *d = NULL; | |
378 | ||
379 | assert(c); | |
380 | assert(dev); | |
381 | assert(isempty(mode) || in_charset(mode, "rwm")); | |
382 | ||
383 | a = new(CGroupDeviceAllow, 1); | |
384 | if (!a) | |
385 | return -ENOMEM; | |
386 | ||
387 | d = strdup(dev); | |
388 | if (!d) | |
389 | return -ENOMEM; | |
390 | ||
391 | *a = (CGroupDeviceAllow) { | |
392 | .path = TAKE_PTR(d), | |
490c5a37 LP |
393 | .r = isempty(mode) || strchr(mode, 'r'), |
394 | .w = isempty(mode) || strchr(mode, 'w'), | |
395 | .m = isempty(mode) || strchr(mode, 'm'), | |
fd870bac YW |
396 | }; |
397 | ||
398 | LIST_PREPEND(device_allow, c->device_allow, a); | |
399 | TAKE_PTR(a); | |
400 | ||
401 | return 0; | |
402 | } | |
403 | ||
6264b85e CD |
404 | #define UNIT_DEFINE_ANCESTOR_MEMORY_LOOKUP(entry) \ |
405 | uint64_t unit_get_ancestor_##entry(Unit *u) { \ | |
406 | CGroupContext *c; \ | |
407 | \ | |
408 | /* 1. Is entry set in this unit? If so, use that. \ | |
409 | * 2. Is the default for this entry set in any \ | |
410 | * ancestor? If so, use that. \ | |
411 | * 3. Otherwise, return CGROUP_LIMIT_MIN. */ \ | |
412 | \ | |
413 | assert(u); \ | |
414 | \ | |
415 | c = unit_get_cgroup_context(u); \ | |
c5322608 | 416 | if (c && c->entry##_set) \ |
6264b85e CD |
417 | return c->entry; \ |
418 | \ | |
c5322608 | 419 | while ((u = UNIT_DEREF(u->slice))) { \ |
6264b85e | 420 | c = unit_get_cgroup_context(u); \ |
c5322608 | 421 | if (c && c->default_##entry##_set) \ |
6264b85e CD |
422 | return c->default_##entry; \ |
423 | } \ | |
424 | \ | |
425 | /* We've reached the root, but nobody had default for \ | |
426 | * this entry set, so set it to the kernel default. */ \ | |
427 | return CGROUP_LIMIT_MIN; \ | |
c52db42b CD |
428 | } |
429 | ||
6264b85e | 430 | UNIT_DEFINE_ANCESTOR_MEMORY_LOOKUP(memory_low); |
7ad5439e | 431 | UNIT_DEFINE_ANCESTOR_MEMORY_LOOKUP(memory_min); |
6264b85e | 432 | |
0d2d6fbf CD |
433 | static void cgroup_xattr_apply(Unit *u) { |
434 | char ids[SD_ID128_STRING_MAX]; | |
435 | int r; | |
436 | ||
437 | assert(u); | |
438 | ||
439 | if (!MANAGER_IS_SYSTEM(u->manager)) | |
440 | return; | |
441 | ||
442 | if (sd_id128_is_null(u->invocation_id)) | |
443 | return; | |
444 | ||
445 | r = cg_set_xattr(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path, | |
446 | "trusted.invocation_id", | |
447 | sd_id128_to_string(u->invocation_id, ids), 32, | |
448 | 0); | |
449 | if (r < 0) | |
450 | log_unit_debug_errno(u, r, "Failed to set invocation ID on control group %s, ignoring: %m", u->cgroup_path); | |
451 | } | |
452 | ||
45c2e068 | 453 | static int lookup_block_device(const char *p, dev_t *ret) { |
f5855697 YS |
454 | dev_t rdev, dev = 0; |
455 | mode_t mode; | |
45c2e068 | 456 | int r; |
4ad49000 LP |
457 | |
458 | assert(p); | |
45c2e068 | 459 | assert(ret); |
4ad49000 | 460 | |
f5855697 | 461 | r = device_path_parse_major_minor(p, &mode, &rdev); |
d5aecba6 | 462 | if (r == -ENODEV) { /* not a parsable device node, need to go to disk */ |
f5855697 | 463 | struct stat st; |
d5aecba6 LP |
464 | if (stat(p, &st) < 0) |
465 | return log_warning_errno(errno, "Couldn't stat device '%s': %m", p); | |
f5855697 YS |
466 | rdev = (dev_t)st.st_rdev; |
467 | dev = (dev_t)st.st_dev; | |
468 | mode = st.st_mode; | |
d5aecba6 LP |
469 | } else if (r < 0) |
470 | return log_warning_errno(r, "Failed to parse major/minor from path '%s': %m", p); | |
471 | ||
f5855697 | 472 | if (S_ISCHR(mode)) { |
d5aecba6 LP |
473 | log_warning("Device node '%s' is a character device, but block device needed.", p); |
474 | return -ENOTBLK; | |
f5855697 YS |
475 | } else if (S_ISBLK(mode)) |
476 | *ret = rdev; | |
477 | else if (major(dev) != 0) | |
478 | *ret = dev; /* If this is not a device node then use the block device this file is stored on */ | |
45c2e068 LP |
479 | else { |
480 | /* If this is btrfs, getting the backing block device is a bit harder */ | |
481 | r = btrfs_get_block_device(p, ret); | |
482 | if (r < 0 && r != -ENOTTY) | |
483 | return log_warning_errno(r, "Failed to determine block device backing btrfs file system '%s': %m", p); | |
484 | if (r == -ENOTTY) { | |
485 | log_warning("'%s' is not a block device node, and file system block device cannot be determined or is not local.", p); | |
486 | return -ENODEV; | |
487 | } | |
4ad49000 | 488 | } |
8e274523 | 489 | |
45c2e068 LP |
490 | /* If this is a LUKS device, try to get the originating block device */ |
491 | (void) block_get_originating(*ret, ret); | |
492 | ||
493 | /* If this is a partition, try to get the originating block device */ | |
494 | (void) block_get_whole_disk(*ret, ret); | |
8e274523 | 495 | return 0; |
8e274523 LP |
496 | } |
497 | ||
084c7007 | 498 | static int whitelist_device(BPFProgram *prog, const char *path, const char *node, const char *acc) { |
f5855697 YS |
499 | dev_t rdev; |
500 | mode_t mode; | |
8c6db833 | 501 | int r; |
8e274523 | 502 | |
4ad49000 LP |
503 | assert(path); |
504 | assert(acc); | |
8e274523 | 505 | |
74c48bf5 LP |
506 | /* Some special handling for /dev/block/%u:%u, /dev/char/%u:%u, /run/systemd/inaccessible/chr and |
507 | * /run/systemd/inaccessible/blk paths. Instead of stat()ing these we parse out the major/minor directly. This | |
508 | * means clients can use these path without the device node actually around */ | |
f5855697 | 509 | r = device_path_parse_major_minor(node, &mode, &rdev); |
74c48bf5 LP |
510 | if (r < 0) { |
511 | if (r != -ENODEV) | |
512 | return log_warning_errno(r, "Couldn't parse major/minor from device path '%s': %m", node); | |
b200489b | 513 | |
f5855697 | 514 | struct stat st; |
74c48bf5 LP |
515 | if (stat(node, &st) < 0) |
516 | return log_warning_errno(errno, "Couldn't stat device %s: %m", node); | |
e7330dfe | 517 | |
74c48bf5 LP |
518 | if (!S_ISCHR(st.st_mode) && !S_ISBLK(st.st_mode)) { |
519 | log_warning("%s is not a device.", node); | |
520 | return -ENODEV; | |
521 | } | |
f5855697 YS |
522 | rdev = (dev_t) st.st_rdev; |
523 | mode = st.st_mode; | |
4ad49000 LP |
524 | } |
525 | ||
084c7007 RG |
526 | if (cg_all_unified() > 0) { |
527 | if (!prog) | |
528 | return 0; | |
4ad49000 | 529 | |
f5855697 YS |
530 | return cgroup_bpf_whitelist_device(prog, S_ISCHR(mode) ? BPF_DEVCG_DEV_CHAR : BPF_DEVCG_DEV_BLOCK, |
531 | major(rdev), minor(rdev), acc); | |
b9839ac9 | 532 | |
084c7007 RG |
533 | } else { |
534 | char buf[2+DECIMAL_STR_MAX(dev_t)*2+2+4]; | |
535 | ||
536 | sprintf(buf, | |
537 | "%c %u:%u %s", | |
f5855697 YS |
538 | S_ISCHR(mode) ? 'c' : 'b', |
539 | major(rdev), minor(rdev), | |
084c7007 RG |
540 | acc); |
541 | ||
8c838407 LP |
542 | /* Changing the devices list of a populated cgroup might result in EINVAL, hence ignore EINVAL here. */ |
543 | ||
084c7007 RG |
544 | r = cg_set_attribute("devices", path, "devices.allow", buf); |
545 | if (r < 0) | |
2c74e12b | 546 | return log_full_errno(IN_SET(r, -ENOENT, -EROFS, -EINVAL, -EACCES, -EPERM) ? LOG_DEBUG : LOG_WARNING, |
b9839ac9 | 547 | r, "Failed to set devices.allow on %s: %m", path); |
4ad49000 | 548 | |
b9839ac9 LP |
549 | return 0; |
550 | } | |
8e274523 LP |
551 | } |
552 | ||
084c7007 | 553 | static int whitelist_major(BPFProgram *prog, const char *path, const char *name, char type, const char *acc) { |
90060676 | 554 | _cleanup_fclose_ FILE *f = NULL; |
8e8b5d2e | 555 | char buf[2+DECIMAL_STR_MAX(unsigned)+3+4]; |
90060676 | 556 | bool good = false; |
8e8b5d2e | 557 | unsigned maj; |
90060676 LP |
558 | int r; |
559 | ||
560 | assert(path); | |
561 | assert(acc); | |
4c701096 | 562 | assert(IN_SET(type, 'b', 'c')); |
90060676 | 563 | |
8e8b5d2e LP |
564 | if (streq(name, "*")) { |
565 | /* If the name is a wildcard, then apply this list to all devices of this type */ | |
566 | ||
567 | if (cg_all_unified() > 0) { | |
568 | if (!prog) | |
569 | return 0; | |
570 | ||
571 | (void) cgroup_bpf_whitelist_class(prog, type == 'c' ? BPF_DEVCG_DEV_CHAR : BPF_DEVCG_DEV_BLOCK, acc); | |
572 | } else { | |
573 | xsprintf(buf, "%c *:* %s", type, acc); | |
574 | ||
575 | r = cg_set_attribute("devices", path, "devices.allow", buf); | |
576 | if (r < 0) | |
577 | log_full_errno(IN_SET(r, -ENOENT, -EROFS, -EINVAL, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, | |
578 | "Failed to set devices.allow on %s: %m", path); | |
579 | return 0; | |
580 | } | |
581 | } | |
582 | ||
583 | if (safe_atou(name, &maj) >= 0 && DEVICE_MAJOR_VALID(maj)) { | |
584 | /* The name is numeric and suitable as major. In that case, let's take is major, and create the entry | |
585 | * directly */ | |
586 | ||
587 | if (cg_all_unified() > 0) { | |
588 | if (!prog) | |
589 | return 0; | |
590 | ||
591 | (void) cgroup_bpf_whitelist_major(prog, | |
592 | type == 'c' ? BPF_DEVCG_DEV_CHAR : BPF_DEVCG_DEV_BLOCK, | |
593 | maj, acc); | |
594 | } else { | |
595 | xsprintf(buf, "%c %u:* %s", type, maj, acc); | |
596 | ||
597 | r = cg_set_attribute("devices", path, "devices.allow", buf); | |
598 | if (r < 0) | |
599 | log_full_errno(IN_SET(r, -ENOENT, -EROFS, -EINVAL, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, | |
600 | "Failed to set devices.allow on %s: %m", path); | |
601 | } | |
602 | ||
603 | return 0; | |
604 | } | |
605 | ||
90060676 | 606 | f = fopen("/proc/devices", "re"); |
4a62c710 MS |
607 | if (!f) |
608 | return log_warning_errno(errno, "Cannot open /proc/devices to resolve %s (%c): %m", name, type); | |
90060676 | 609 | |
c66e60a8 LP |
610 | for (;;) { |
611 | _cleanup_free_ char *line = NULL; | |
8e8b5d2e | 612 | char *w, *p; |
90060676 | 613 | |
c66e60a8 LP |
614 | r = read_line(f, LONG_LINE_MAX, &line); |
615 | if (r < 0) | |
616 | return log_warning_errno(r, "Failed to read /proc/devices: %m"); | |
617 | if (r == 0) | |
618 | break; | |
90060676 LP |
619 | |
620 | if (type == 'c' && streq(line, "Character devices:")) { | |
621 | good = true; | |
622 | continue; | |
623 | } | |
624 | ||
625 | if (type == 'b' && streq(line, "Block devices:")) { | |
626 | good = true; | |
627 | continue; | |
628 | } | |
629 | ||
630 | if (isempty(line)) { | |
631 | good = false; | |
632 | continue; | |
633 | } | |
634 | ||
635 | if (!good) | |
636 | continue; | |
637 | ||
638 | p = strstrip(line); | |
639 | ||
640 | w = strpbrk(p, WHITESPACE); | |
641 | if (!w) | |
642 | continue; | |
643 | *w = 0; | |
644 | ||
645 | r = safe_atou(p, &maj); | |
646 | if (r < 0) | |
647 | continue; | |
648 | if (maj <= 0) | |
649 | continue; | |
650 | ||
651 | w++; | |
652 | w += strspn(w, WHITESPACE); | |
e41969e3 LP |
653 | |
654 | if (fnmatch(name, w, 0) != 0) | |
90060676 LP |
655 | continue; |
656 | ||
084c7007 RG |
657 | if (cg_all_unified() > 0) { |
658 | if (!prog) | |
659 | continue; | |
90060676 | 660 | |
913c898c LP |
661 | (void) cgroup_bpf_whitelist_major(prog, |
662 | type == 'c' ? BPF_DEVCG_DEV_CHAR : BPF_DEVCG_DEV_BLOCK, | |
663 | maj, acc); | |
084c7007 | 664 | } else { |
084c7007 RG |
665 | sprintf(buf, |
666 | "%c %u:* %s", | |
667 | type, | |
668 | maj, | |
669 | acc); | |
670 | ||
8c838407 LP |
671 | /* Changing the devices list of a populated cgroup might result in EINVAL, hence ignore EINVAL |
672 | * here. */ | |
673 | ||
084c7007 RG |
674 | r = cg_set_attribute("devices", path, "devices.allow", buf); |
675 | if (r < 0) | |
2c74e12b | 676 | log_full_errno(IN_SET(r, -ENOENT, -EROFS, -EINVAL, -EACCES, -EPERM) ? LOG_DEBUG : LOG_WARNING, |
084c7007 RG |
677 | r, "Failed to set devices.allow on %s: %m", path); |
678 | } | |
90060676 LP |
679 | } |
680 | ||
681 | return 0; | |
90060676 LP |
682 | } |
683 | ||
66ebf6c0 TH |
684 | static bool cgroup_context_has_cpu_weight(CGroupContext *c) { |
685 | return c->cpu_weight != CGROUP_WEIGHT_INVALID || | |
686 | c->startup_cpu_weight != CGROUP_WEIGHT_INVALID; | |
687 | } | |
688 | ||
689 | static bool cgroup_context_has_cpu_shares(CGroupContext *c) { | |
690 | return c->cpu_shares != CGROUP_CPU_SHARES_INVALID || | |
691 | c->startup_cpu_shares != CGROUP_CPU_SHARES_INVALID; | |
692 | } | |
693 | ||
694 | static uint64_t cgroup_context_cpu_weight(CGroupContext *c, ManagerState state) { | |
695 | if (IN_SET(state, MANAGER_STARTING, MANAGER_INITIALIZING) && | |
696 | c->startup_cpu_weight != CGROUP_WEIGHT_INVALID) | |
697 | return c->startup_cpu_weight; | |
698 | else if (c->cpu_weight != CGROUP_WEIGHT_INVALID) | |
699 | return c->cpu_weight; | |
700 | else | |
701 | return CGROUP_WEIGHT_DEFAULT; | |
702 | } | |
703 | ||
704 | static uint64_t cgroup_context_cpu_shares(CGroupContext *c, ManagerState state) { | |
705 | if (IN_SET(state, MANAGER_STARTING, MANAGER_INITIALIZING) && | |
706 | c->startup_cpu_shares != CGROUP_CPU_SHARES_INVALID) | |
707 | return c->startup_cpu_shares; | |
708 | else if (c->cpu_shares != CGROUP_CPU_SHARES_INVALID) | |
709 | return c->cpu_shares; | |
710 | else | |
711 | return CGROUP_CPU_SHARES_DEFAULT; | |
712 | } | |
713 | ||
10f28641 FB |
714 | usec_t cgroup_cpu_adjust_period(usec_t period, usec_t quota, usec_t resolution, usec_t max_period) { |
715 | /* kernel uses a minimum resolution of 1ms, so both period and (quota * period) | |
716 | * need to be higher than that boundary. quota is specified in USecPerSec. | |
717 | * Additionally, period must be at most max_period. */ | |
718 | assert(quota > 0); | |
719 | ||
720 | return MIN(MAX3(period, resolution, resolution * USEC_PER_SEC / quota), max_period); | |
721 | } | |
722 | ||
723 | static usec_t cgroup_cpu_adjust_period_and_log(Unit *u, usec_t period, usec_t quota) { | |
724 | usec_t new_period; | |
725 | ||
726 | if (quota == USEC_INFINITY) | |
727 | /* Always use default period for infinity quota. */ | |
728 | return CGROUP_CPU_QUOTA_DEFAULT_PERIOD_USEC; | |
729 | ||
730 | if (period == USEC_INFINITY) | |
731 | /* Default period was requested. */ | |
732 | period = CGROUP_CPU_QUOTA_DEFAULT_PERIOD_USEC; | |
733 | ||
734 | /* Clamp to interval [1ms, 1s] */ | |
735 | new_period = cgroup_cpu_adjust_period(period, quota, USEC_PER_MSEC, USEC_PER_SEC); | |
736 | ||
737 | if (new_period != period) { | |
738 | char v[FORMAT_TIMESPAN_MAX]; | |
527ede0c | 739 | log_unit_full(u, u->warned_clamping_cpu_quota_period ? LOG_DEBUG : LOG_WARNING, 0, |
10f28641 FB |
740 | "Clamping CPU interval for cpu.max: period is now %s", |
741 | format_timespan(v, sizeof(v), new_period, 1)); | |
527ede0c | 742 | u->warned_clamping_cpu_quota_period = true; |
10f28641 FB |
743 | } |
744 | ||
745 | return new_period; | |
746 | } | |
747 | ||
52fecf20 LP |
748 | static void cgroup_apply_unified_cpu_weight(Unit *u, uint64_t weight) { |
749 | char buf[DECIMAL_STR_MAX(uint64_t) + 2]; | |
66ebf6c0 TH |
750 | |
751 | xsprintf(buf, "%" PRIu64 "\n", weight); | |
293d32df | 752 | (void) set_attribute_and_warn(u, "cpu", "cpu.weight", buf); |
52fecf20 LP |
753 | } |
754 | ||
10f28641 | 755 | static void cgroup_apply_unified_cpu_quota(Unit *u, usec_t quota, usec_t period) { |
52fecf20 | 756 | char buf[(DECIMAL_STR_MAX(usec_t) + 1) * 2 + 1]; |
66ebf6c0 | 757 | |
10f28641 | 758 | period = cgroup_cpu_adjust_period_and_log(u, period, quota); |
66ebf6c0 TH |
759 | if (quota != USEC_INFINITY) |
760 | xsprintf(buf, USEC_FMT " " USEC_FMT "\n", | |
10f28641 | 761 | MAX(quota * period / USEC_PER_SEC, USEC_PER_MSEC), period); |
66ebf6c0 | 762 | else |
10f28641 | 763 | xsprintf(buf, "max " USEC_FMT "\n", period); |
293d32df | 764 | (void) set_attribute_and_warn(u, "cpu", "cpu.max", buf); |
66ebf6c0 TH |
765 | } |
766 | ||
52fecf20 LP |
767 | static void cgroup_apply_legacy_cpu_shares(Unit *u, uint64_t shares) { |
768 | char buf[DECIMAL_STR_MAX(uint64_t) + 2]; | |
66ebf6c0 TH |
769 | |
770 | xsprintf(buf, "%" PRIu64 "\n", shares); | |
293d32df | 771 | (void) set_attribute_and_warn(u, "cpu", "cpu.shares", buf); |
52fecf20 LP |
772 | } |
773 | ||
10f28641 | 774 | static void cgroup_apply_legacy_cpu_quota(Unit *u, usec_t quota, usec_t period) { |
52fecf20 | 775 | char buf[DECIMAL_STR_MAX(usec_t) + 2]; |
66ebf6c0 | 776 | |
10f28641 FB |
777 | period = cgroup_cpu_adjust_period_and_log(u, period, quota); |
778 | ||
779 | xsprintf(buf, USEC_FMT "\n", period); | |
293d32df | 780 | (void) set_attribute_and_warn(u, "cpu", "cpu.cfs_period_us", buf); |
66ebf6c0 TH |
781 | |
782 | if (quota != USEC_INFINITY) { | |
10f28641 | 783 | xsprintf(buf, USEC_FMT "\n", MAX(quota * period / USEC_PER_SEC, USEC_PER_MSEC)); |
293d32df | 784 | (void) set_attribute_and_warn(u, "cpu", "cpu.cfs_quota_us", buf); |
66ebf6c0 | 785 | } else |
589a5f7a | 786 | (void) set_attribute_and_warn(u, "cpu", "cpu.cfs_quota_us", "-1\n"); |
66ebf6c0 TH |
787 | } |
788 | ||
789 | static uint64_t cgroup_cpu_shares_to_weight(uint64_t shares) { | |
790 | return CLAMP(shares * CGROUP_WEIGHT_DEFAULT / CGROUP_CPU_SHARES_DEFAULT, | |
791 | CGROUP_WEIGHT_MIN, CGROUP_WEIGHT_MAX); | |
792 | } | |
793 | ||
794 | static uint64_t cgroup_cpu_weight_to_shares(uint64_t weight) { | |
795 | return CLAMP(weight * CGROUP_CPU_SHARES_DEFAULT / CGROUP_WEIGHT_DEFAULT, | |
796 | CGROUP_CPU_SHARES_MIN, CGROUP_CPU_SHARES_MAX); | |
797 | } | |
798 | ||
508c45da | 799 | static bool cgroup_context_has_io_config(CGroupContext *c) { |
538b4852 TH |
800 | return c->io_accounting || |
801 | c->io_weight != CGROUP_WEIGHT_INVALID || | |
802 | c->startup_io_weight != CGROUP_WEIGHT_INVALID || | |
803 | c->io_device_weights || | |
6ae4283c | 804 | c->io_device_latencies || |
538b4852 TH |
805 | c->io_device_limits; |
806 | } | |
807 | ||
508c45da | 808 | static bool cgroup_context_has_blockio_config(CGroupContext *c) { |
538b4852 TH |
809 | return c->blockio_accounting || |
810 | c->blockio_weight != CGROUP_BLKIO_WEIGHT_INVALID || | |
811 | c->startup_blockio_weight != CGROUP_BLKIO_WEIGHT_INVALID || | |
812 | c->blockio_device_weights || | |
813 | c->blockio_device_bandwidths; | |
814 | } | |
815 | ||
508c45da | 816 | static uint64_t cgroup_context_io_weight(CGroupContext *c, ManagerState state) { |
64faf04c TH |
817 | if (IN_SET(state, MANAGER_STARTING, MANAGER_INITIALIZING) && |
818 | c->startup_io_weight != CGROUP_WEIGHT_INVALID) | |
819 | return c->startup_io_weight; | |
820 | else if (c->io_weight != CGROUP_WEIGHT_INVALID) | |
821 | return c->io_weight; | |
822 | else | |
823 | return CGROUP_WEIGHT_DEFAULT; | |
824 | } | |
825 | ||
508c45da | 826 | static uint64_t cgroup_context_blkio_weight(CGroupContext *c, ManagerState state) { |
64faf04c TH |
827 | if (IN_SET(state, MANAGER_STARTING, MANAGER_INITIALIZING) && |
828 | c->startup_blockio_weight != CGROUP_BLKIO_WEIGHT_INVALID) | |
829 | return c->startup_blockio_weight; | |
830 | else if (c->blockio_weight != CGROUP_BLKIO_WEIGHT_INVALID) | |
831 | return c->blockio_weight; | |
832 | else | |
833 | return CGROUP_BLKIO_WEIGHT_DEFAULT; | |
834 | } | |
835 | ||
508c45da | 836 | static uint64_t cgroup_weight_blkio_to_io(uint64_t blkio_weight) { |
538b4852 TH |
837 | return CLAMP(blkio_weight * CGROUP_WEIGHT_DEFAULT / CGROUP_BLKIO_WEIGHT_DEFAULT, |
838 | CGROUP_WEIGHT_MIN, CGROUP_WEIGHT_MAX); | |
839 | } | |
840 | ||
508c45da | 841 | static uint64_t cgroup_weight_io_to_blkio(uint64_t io_weight) { |
538b4852 TH |
842 | return CLAMP(io_weight * CGROUP_BLKIO_WEIGHT_DEFAULT / CGROUP_WEIGHT_DEFAULT, |
843 | CGROUP_BLKIO_WEIGHT_MIN, CGROUP_BLKIO_WEIGHT_MAX); | |
844 | } | |
845 | ||
f29ff115 | 846 | static void cgroup_apply_io_device_weight(Unit *u, const char *dev_path, uint64_t io_weight) { |
64faf04c TH |
847 | char buf[DECIMAL_STR_MAX(dev_t)*2+2+DECIMAL_STR_MAX(uint64_t)+1]; |
848 | dev_t dev; | |
849 | int r; | |
850 | ||
851 | r = lookup_block_device(dev_path, &dev); | |
852 | if (r < 0) | |
853 | return; | |
854 | ||
855 | xsprintf(buf, "%u:%u %" PRIu64 "\n", major(dev), minor(dev), io_weight); | |
293d32df | 856 | (void) set_attribute_and_warn(u, "io", "io.weight", buf); |
64faf04c TH |
857 | } |
858 | ||
f29ff115 | 859 | static void cgroup_apply_blkio_device_weight(Unit *u, const char *dev_path, uint64_t blkio_weight) { |
64faf04c TH |
860 | char buf[DECIMAL_STR_MAX(dev_t)*2+2+DECIMAL_STR_MAX(uint64_t)+1]; |
861 | dev_t dev; | |
862 | int r; | |
863 | ||
864 | r = lookup_block_device(dev_path, &dev); | |
865 | if (r < 0) | |
866 | return; | |
867 | ||
868 | xsprintf(buf, "%u:%u %" PRIu64 "\n", major(dev), minor(dev), blkio_weight); | |
293d32df | 869 | (void) set_attribute_and_warn(u, "blkio", "blkio.weight_device", buf); |
64faf04c TH |
870 | } |
871 | ||
6ae4283c TH |
872 | static void cgroup_apply_io_device_latency(Unit *u, const char *dev_path, usec_t target) { |
873 | char buf[DECIMAL_STR_MAX(dev_t)*2+2+7+DECIMAL_STR_MAX(uint64_t)+1]; | |
874 | dev_t dev; | |
875 | int r; | |
876 | ||
877 | r = lookup_block_device(dev_path, &dev); | |
878 | if (r < 0) | |
879 | return; | |
880 | ||
881 | if (target != USEC_INFINITY) | |
882 | xsprintf(buf, "%u:%u target=%" PRIu64 "\n", major(dev), minor(dev), target); | |
883 | else | |
884 | xsprintf(buf, "%u:%u target=max\n", major(dev), minor(dev)); | |
885 | ||
293d32df | 886 | (void) set_attribute_and_warn(u, "io", "io.latency", buf); |
6ae4283c TH |
887 | } |
888 | ||
17ae2780 | 889 | static void cgroup_apply_io_device_limit(Unit *u, const char *dev_path, uint64_t *limits) { |
64faf04c TH |
890 | char limit_bufs[_CGROUP_IO_LIMIT_TYPE_MAX][DECIMAL_STR_MAX(uint64_t)]; |
891 | char buf[DECIMAL_STR_MAX(dev_t)*2+2+(6+DECIMAL_STR_MAX(uint64_t)+1)*4]; | |
892 | CGroupIOLimitType type; | |
893 | dev_t dev; | |
64faf04c TH |
894 | int r; |
895 | ||
896 | r = lookup_block_device(dev_path, &dev); | |
897 | if (r < 0) | |
17ae2780 | 898 | return; |
64faf04c | 899 | |
17ae2780 LP |
900 | for (type = 0; type < _CGROUP_IO_LIMIT_TYPE_MAX; type++) |
901 | if (limits[type] != cgroup_io_limit_defaults[type]) | |
64faf04c | 902 | xsprintf(limit_bufs[type], "%" PRIu64, limits[type]); |
17ae2780 | 903 | else |
64faf04c | 904 | xsprintf(limit_bufs[type], "%s", limits[type] == CGROUP_LIMIT_MAX ? "max" : "0"); |
64faf04c TH |
905 | |
906 | xsprintf(buf, "%u:%u rbps=%s wbps=%s riops=%s wiops=%s\n", major(dev), minor(dev), | |
907 | limit_bufs[CGROUP_IO_RBPS_MAX], limit_bufs[CGROUP_IO_WBPS_MAX], | |
908 | limit_bufs[CGROUP_IO_RIOPS_MAX], limit_bufs[CGROUP_IO_WIOPS_MAX]); | |
293d32df | 909 | (void) set_attribute_and_warn(u, "io", "io.max", buf); |
64faf04c TH |
910 | } |
911 | ||
17ae2780 | 912 | static void cgroup_apply_blkio_device_limit(Unit *u, const char *dev_path, uint64_t rbps, uint64_t wbps) { |
64faf04c TH |
913 | char buf[DECIMAL_STR_MAX(dev_t)*2+2+DECIMAL_STR_MAX(uint64_t)+1]; |
914 | dev_t dev; | |
64faf04c TH |
915 | int r; |
916 | ||
917 | r = lookup_block_device(dev_path, &dev); | |
918 | if (r < 0) | |
17ae2780 | 919 | return; |
64faf04c | 920 | |
64faf04c | 921 | sprintf(buf, "%u:%u %" PRIu64 "\n", major(dev), minor(dev), rbps); |
293d32df | 922 | (void) set_attribute_and_warn(u, "blkio", "blkio.throttle.read_bps_device", buf); |
64faf04c | 923 | |
64faf04c | 924 | sprintf(buf, "%u:%u %" PRIu64 "\n", major(dev), minor(dev), wbps); |
293d32df | 925 | (void) set_attribute_and_warn(u, "blkio", "blkio.throttle.write_bps_device", buf); |
64faf04c TH |
926 | } |
927 | ||
c52db42b CD |
928 | static bool unit_has_unified_memory_config(Unit *u) { |
929 | CGroupContext *c; | |
930 | ||
931 | assert(u); | |
932 | ||
933 | c = unit_get_cgroup_context(u); | |
934 | assert(c); | |
935 | ||
936 | return c->memory_min > 0 || unit_get_ancestor_memory_low(u) > 0 || | |
937 | c->memory_high != CGROUP_LIMIT_MAX || c->memory_max != CGROUP_LIMIT_MAX || | |
938 | c->memory_swap_max != CGROUP_LIMIT_MAX; | |
da4d897e TH |
939 | } |
940 | ||
f29ff115 | 941 | static void cgroup_apply_unified_memory_limit(Unit *u, const char *file, uint64_t v) { |
589a5f7a | 942 | char buf[DECIMAL_STR_MAX(uint64_t) + 1] = "max\n"; |
da4d897e TH |
943 | |
944 | if (v != CGROUP_LIMIT_MAX) | |
945 | xsprintf(buf, "%" PRIu64 "\n", v); | |
946 | ||
293d32df | 947 | (void) set_attribute_and_warn(u, "memory", file, buf); |
da4d897e TH |
948 | } |
949 | ||
0f2d84d2 | 950 | static void cgroup_apply_firewall(Unit *u) { |
0f2d84d2 LP |
951 | assert(u); |
952 | ||
acf7f253 | 953 | /* Best-effort: let's apply IP firewalling and/or accounting if that's enabled */ |
906c06f6 | 954 | |
acf7f253 | 955 | if (bpf_firewall_compile(u) < 0) |
906c06f6 DM |
956 | return; |
957 | ||
fab34748 | 958 | (void) bpf_firewall_load_custom(u); |
906c06f6 | 959 | (void) bpf_firewall_install(u); |
906c06f6 DM |
960 | } |
961 | ||
962 | static void cgroup_context_apply( | |
963 | Unit *u, | |
964 | CGroupMask apply_mask, | |
906c06f6 DM |
965 | ManagerState state) { |
966 | ||
f29ff115 TH |
967 | const char *path; |
968 | CGroupContext *c; | |
52fecf20 | 969 | bool is_host_root, is_local_root; |
4ad49000 LP |
970 | int r; |
971 | ||
f29ff115 TH |
972 | assert(u); |
973 | ||
906c06f6 | 974 | /* Nothing to do? Exit early! */ |
17f14955 | 975 | if (apply_mask == 0) |
4ad49000 | 976 | return; |
8e274523 | 977 | |
52fecf20 LP |
978 | /* Some cgroup attributes are not supported on the host root cgroup, hence silently ignore them here. And other |
979 | * attributes should only be managed for cgroups further down the tree. */ | |
980 | is_local_root = unit_has_name(u, SPECIAL_ROOT_SLICE); | |
981 | is_host_root = unit_has_host_root_cgroup(u); | |
f3725e64 LP |
982 | |
983 | assert_se(c = unit_get_cgroup_context(u)); | |
984 | assert_se(path = u->cgroup_path); | |
985 | ||
52fecf20 | 986 | if (is_local_root) /* Make sure we don't try to display messages with an empty path. */ |
6da13913 | 987 | path = "/"; |
01efdf13 | 988 | |
be2c0327 LP |
989 | /* We generally ignore errors caused by read-only mounted cgroup trees (assuming we are running in a container |
990 | * then), and missing cgroups, i.e. EROFS and ENOENT. */ | |
714e2e1d | 991 | |
be2c0327 LP |
992 | /* In fully unified mode these attributes don't exist on the host cgroup root. On legacy the weights exist, but |
993 | * setting the weight makes very little sense on the host root cgroup, as there are no other cgroups at this | |
994 | * level. The quota exists there too, but any attempt to write to it is refused with EINVAL. Inside of | |
4e1dfa45 | 995 | * containers we want to leave control of these to the container manager (and if cgroup v2 delegation is used |
be2c0327 LP |
996 | * we couldn't even write to them if we wanted to). */ |
997 | if ((apply_mask & CGROUP_MASK_CPU) && !is_local_root) { | |
8e274523 | 998 | |
b4cccbc1 | 999 | if (cg_all_unified() > 0) { |
be2c0327 | 1000 | uint64_t weight; |
b2f8b02e | 1001 | |
be2c0327 LP |
1002 | if (cgroup_context_has_cpu_weight(c)) |
1003 | weight = cgroup_context_cpu_weight(c, state); | |
1004 | else if (cgroup_context_has_cpu_shares(c)) { | |
1005 | uint64_t shares; | |
66ebf6c0 | 1006 | |
be2c0327 LP |
1007 | shares = cgroup_context_cpu_shares(c, state); |
1008 | weight = cgroup_cpu_shares_to_weight(shares); | |
66ebf6c0 | 1009 | |
be2c0327 LP |
1010 | log_cgroup_compat(u, "Applying [Startup]CPUShares=%" PRIu64 " as [Startup]CPUWeight=%" PRIu64 " on %s", |
1011 | shares, weight, path); | |
1012 | } else | |
1013 | weight = CGROUP_WEIGHT_DEFAULT; | |
66ebf6c0 | 1014 | |
be2c0327 | 1015 | cgroup_apply_unified_cpu_weight(u, weight); |
10f28641 | 1016 | cgroup_apply_unified_cpu_quota(u, c->cpu_quota_per_sec_usec, c->cpu_quota_period_usec); |
66ebf6c0 | 1017 | |
52fecf20 | 1018 | } else { |
be2c0327 | 1019 | uint64_t shares; |
52fecf20 | 1020 | |
be2c0327 LP |
1021 | if (cgroup_context_has_cpu_weight(c)) { |
1022 | uint64_t weight; | |
52fecf20 | 1023 | |
be2c0327 LP |
1024 | weight = cgroup_context_cpu_weight(c, state); |
1025 | shares = cgroup_cpu_weight_to_shares(weight); | |
52fecf20 | 1026 | |
be2c0327 LP |
1027 | log_cgroup_compat(u, "Applying [Startup]CPUWeight=%" PRIu64 " as [Startup]CPUShares=%" PRIu64 " on %s", |
1028 | weight, shares, path); | |
1029 | } else if (cgroup_context_has_cpu_shares(c)) | |
1030 | shares = cgroup_context_cpu_shares(c, state); | |
1031 | else | |
1032 | shares = CGROUP_CPU_SHARES_DEFAULT; | |
66ebf6c0 | 1033 | |
be2c0327 | 1034 | cgroup_apply_legacy_cpu_shares(u, shares); |
10f28641 | 1035 | cgroup_apply_legacy_cpu_quota(u, c->cpu_quota_per_sec_usec, c->cpu_quota_period_usec); |
66ebf6c0 | 1036 | } |
4ad49000 LP |
1037 | } |
1038 | ||
4e1dfa45 | 1039 | /* The 'io' controller attributes are not exported on the host's root cgroup (being a pure cgroup v2 |
52fecf20 LP |
1040 | * controller), and in case of containers we want to leave control of these attributes to the container manager |
1041 | * (and we couldn't access that stuff anyway, even if we tried if proper delegation is used). */ | |
1042 | if ((apply_mask & CGROUP_MASK_IO) && !is_local_root) { | |
1043 | char buf[8+DECIMAL_STR_MAX(uint64_t)+1]; | |
1044 | bool has_io, has_blockio; | |
1045 | uint64_t weight; | |
13c31542 | 1046 | |
52fecf20 LP |
1047 | has_io = cgroup_context_has_io_config(c); |
1048 | has_blockio = cgroup_context_has_blockio_config(c); | |
13c31542 | 1049 | |
52fecf20 LP |
1050 | if (has_io) |
1051 | weight = cgroup_context_io_weight(c, state); | |
1052 | else if (has_blockio) { | |
1053 | uint64_t blkio_weight; | |
128fadc9 | 1054 | |
52fecf20 LP |
1055 | blkio_weight = cgroup_context_blkio_weight(c, state); |
1056 | weight = cgroup_weight_blkio_to_io(blkio_weight); | |
128fadc9 | 1057 | |
67e2ea15 | 1058 | log_cgroup_compat(u, "Applying [Startup]BlockIOWeight=%" PRIu64 " as [Startup]IOWeight=%" PRIu64, |
52fecf20 LP |
1059 | blkio_weight, weight); |
1060 | } else | |
1061 | weight = CGROUP_WEIGHT_DEFAULT; | |
13c31542 | 1062 | |
52fecf20 LP |
1063 | xsprintf(buf, "default %" PRIu64 "\n", weight); |
1064 | (void) set_attribute_and_warn(u, "io", "io.weight", buf); | |
538b4852 | 1065 | |
2dbc45ae KK |
1066 | /* FIXME: drop this when distro kernels properly support BFQ through "io.weight" |
1067 | * See also: https://github.com/systemd/systemd/pull/13335 */ | |
1068 | xsprintf(buf, "%" PRIu64 "\n", weight); | |
1069 | (void) set_attribute_and_warn(u, "io", "io.bfq.weight", buf); | |
1070 | ||
52fecf20 LP |
1071 | if (has_io) { |
1072 | CGroupIODeviceLatency *latency; | |
1073 | CGroupIODeviceLimit *limit; | |
1074 | CGroupIODeviceWeight *w; | |
128fadc9 | 1075 | |
52fecf20 LP |
1076 | LIST_FOREACH(device_weights, w, c->io_device_weights) |
1077 | cgroup_apply_io_device_weight(u, w->path, w->weight); | |
128fadc9 | 1078 | |
52fecf20 LP |
1079 | LIST_FOREACH(device_limits, limit, c->io_device_limits) |
1080 | cgroup_apply_io_device_limit(u, limit->path, limit->limits); | |
6ae4283c | 1081 | |
52fecf20 LP |
1082 | LIST_FOREACH(device_latencies, latency, c->io_device_latencies) |
1083 | cgroup_apply_io_device_latency(u, latency->path, latency->target_usec); | |
6ae4283c | 1084 | |
52fecf20 LP |
1085 | } else if (has_blockio) { |
1086 | CGroupBlockIODeviceWeight *w; | |
1087 | CGroupBlockIODeviceBandwidth *b; | |
13c31542 | 1088 | |
52fecf20 LP |
1089 | LIST_FOREACH(device_weights, w, c->blockio_device_weights) { |
1090 | weight = cgroup_weight_blkio_to_io(w->weight); | |
17ae2780 | 1091 | |
67e2ea15 | 1092 | log_cgroup_compat(u, "Applying BlockIODeviceWeight=%" PRIu64 " as IODeviceWeight=%" PRIu64 " for %s", |
52fecf20 | 1093 | w->weight, weight, w->path); |
538b4852 | 1094 | |
52fecf20 LP |
1095 | cgroup_apply_io_device_weight(u, w->path, weight); |
1096 | } | |
538b4852 | 1097 | |
17ae2780 | 1098 | LIST_FOREACH(device_bandwidths, b, c->blockio_device_bandwidths) { |
538b4852 TH |
1099 | uint64_t limits[_CGROUP_IO_LIMIT_TYPE_MAX]; |
1100 | CGroupIOLimitType type; | |
1101 | ||
1102 | for (type = 0; type < _CGROUP_IO_LIMIT_TYPE_MAX; type++) | |
1103 | limits[type] = cgroup_io_limit_defaults[type]; | |
1104 | ||
1105 | limits[CGROUP_IO_RBPS_MAX] = b->rbps; | |
1106 | limits[CGROUP_IO_WBPS_MAX] = b->wbps; | |
1107 | ||
67e2ea15 | 1108 | log_cgroup_compat(u, "Applying BlockIO{Read|Write}Bandwidth=%" PRIu64 " %" PRIu64 " as IO{Read|Write}BandwidthMax= for %s", |
128fadc9 TH |
1109 | b->rbps, b->wbps, b->path); |
1110 | ||
17ae2780 | 1111 | cgroup_apply_io_device_limit(u, b->path, limits); |
538b4852 | 1112 | } |
13c31542 TH |
1113 | } |
1114 | } | |
1115 | ||
906c06f6 | 1116 | if (apply_mask & CGROUP_MASK_BLKIO) { |
52fecf20 | 1117 | bool has_io, has_blockio; |
4ad49000 | 1118 | |
52fecf20 LP |
1119 | has_io = cgroup_context_has_io_config(c); |
1120 | has_blockio = cgroup_context_has_blockio_config(c); | |
1121 | ||
1122 | /* Applying a 'weight' never makes sense for the host root cgroup, and for containers this should be | |
1123 | * left to our container manager, too. */ | |
1124 | if (!is_local_root) { | |
64faf04c TH |
1125 | char buf[DECIMAL_STR_MAX(uint64_t)+1]; |
1126 | uint64_t weight; | |
64faf04c | 1127 | |
7d862ab8 | 1128 | if (has_io) { |
52fecf20 | 1129 | uint64_t io_weight; |
128fadc9 | 1130 | |
52fecf20 | 1131 | io_weight = cgroup_context_io_weight(c, state); |
538b4852 | 1132 | weight = cgroup_weight_io_to_blkio(cgroup_context_io_weight(c, state)); |
128fadc9 | 1133 | |
67e2ea15 | 1134 | log_cgroup_compat(u, "Applying [Startup]IOWeight=%" PRIu64 " as [Startup]BlockIOWeight=%" PRIu64, |
128fadc9 | 1135 | io_weight, weight); |
7d862ab8 TH |
1136 | } else if (has_blockio) |
1137 | weight = cgroup_context_blkio_weight(c, state); | |
1138 | else | |
538b4852 | 1139 | weight = CGROUP_BLKIO_WEIGHT_DEFAULT; |
64faf04c TH |
1140 | |
1141 | xsprintf(buf, "%" PRIu64 "\n", weight); | |
293d32df | 1142 | (void) set_attribute_and_warn(u, "blkio", "blkio.weight", buf); |
4ad49000 | 1143 | |
7d862ab8 | 1144 | if (has_io) { |
538b4852 TH |
1145 | CGroupIODeviceWeight *w; |
1146 | ||
128fadc9 TH |
1147 | LIST_FOREACH(device_weights, w, c->io_device_weights) { |
1148 | weight = cgroup_weight_io_to_blkio(w->weight); | |
1149 | ||
67e2ea15 | 1150 | log_cgroup_compat(u, "Applying IODeviceWeight=%" PRIu64 " as BlockIODeviceWeight=%" PRIu64 " for %s", |
128fadc9 TH |
1151 | w->weight, weight, w->path); |
1152 | ||
1153 | cgroup_apply_blkio_device_weight(u, w->path, weight); | |
1154 | } | |
7d862ab8 TH |
1155 | } else if (has_blockio) { |
1156 | CGroupBlockIODeviceWeight *w; | |
1157 | ||
7d862ab8 TH |
1158 | LIST_FOREACH(device_weights, w, c->blockio_device_weights) |
1159 | cgroup_apply_blkio_device_weight(u, w->path, w->weight); | |
538b4852 | 1160 | } |
4ad49000 LP |
1161 | } |
1162 | ||
5238e957 | 1163 | /* The bandwidth limits are something that make sense to be applied to the host's root but not container |
52fecf20 LP |
1164 | * roots, as there we want the container manager to handle it */ |
1165 | if (is_host_root || !is_local_root) { | |
1166 | if (has_io) { | |
1167 | CGroupIODeviceLimit *l; | |
538b4852 | 1168 | |
52fecf20 | 1169 | LIST_FOREACH(device_limits, l, c->io_device_limits) { |
67e2ea15 | 1170 | log_cgroup_compat(u, "Applying IO{Read|Write}Bandwidth=%" PRIu64 " %" PRIu64 " as BlockIO{Read|Write}BandwidthMax= for %s", |
52fecf20 | 1171 | l->limits[CGROUP_IO_RBPS_MAX], l->limits[CGROUP_IO_WBPS_MAX], l->path); |
128fadc9 | 1172 | |
52fecf20 LP |
1173 | cgroup_apply_blkio_device_limit(u, l->path, l->limits[CGROUP_IO_RBPS_MAX], l->limits[CGROUP_IO_WBPS_MAX]); |
1174 | } | |
1175 | } else if (has_blockio) { | |
1176 | CGroupBlockIODeviceBandwidth *b; | |
7d862ab8 | 1177 | |
52fecf20 LP |
1178 | LIST_FOREACH(device_bandwidths, b, c->blockio_device_bandwidths) |
1179 | cgroup_apply_blkio_device_limit(u, b->path, b->rbps, b->wbps); | |
1180 | } | |
d686d8a9 | 1181 | } |
8e274523 LP |
1182 | } |
1183 | ||
be2c0327 LP |
1184 | /* In unified mode 'memory' attributes do not exist on the root cgroup. In legacy mode 'memory.limit_in_bytes' |
1185 | * exists on the root cgroup, but any writes to it are refused with EINVAL. And if we run in a container we | |
4e1dfa45 | 1186 | * want to leave control to the container manager (and if proper cgroup v2 delegation is used we couldn't even |
be2c0327 LP |
1187 | * write to this if we wanted to.) */ |
1188 | if ((apply_mask & CGROUP_MASK_MEMORY) && !is_local_root) { | |
efdb0237 | 1189 | |
52fecf20 | 1190 | if (cg_all_unified() > 0) { |
be2c0327 LP |
1191 | uint64_t max, swap_max = CGROUP_LIMIT_MAX; |
1192 | ||
c52db42b | 1193 | if (unit_has_unified_memory_config(u)) { |
be2c0327 LP |
1194 | max = c->memory_max; |
1195 | swap_max = c->memory_swap_max; | |
1196 | } else { | |
1197 | max = c->memory_limit; | |
efdb0237 | 1198 | |
be2c0327 LP |
1199 | if (max != CGROUP_LIMIT_MAX) |
1200 | log_cgroup_compat(u, "Applying MemoryLimit=%" PRIu64 " as MemoryMax=", max); | |
128fadc9 | 1201 | } |
da4d897e | 1202 | |
be2c0327 | 1203 | cgroup_apply_unified_memory_limit(u, "memory.min", c->memory_min); |
c52db42b | 1204 | cgroup_apply_unified_memory_limit(u, "memory.low", unit_get_ancestor_memory_low(u)); |
be2c0327 LP |
1205 | cgroup_apply_unified_memory_limit(u, "memory.high", c->memory_high); |
1206 | cgroup_apply_unified_memory_limit(u, "memory.max", max); | |
1207 | cgroup_apply_unified_memory_limit(u, "memory.swap.max", swap_max); | |
128fadc9 | 1208 | |
afcfaa69 LP |
1209 | (void) set_attribute_and_warn(u, "memory", "memory.oom.group", one_zero(c->memory_oom_group)); |
1210 | ||
be2c0327 LP |
1211 | } else { |
1212 | char buf[DECIMAL_STR_MAX(uint64_t) + 1]; | |
1213 | uint64_t val; | |
52fecf20 | 1214 | |
c52db42b | 1215 | if (unit_has_unified_memory_config(u)) { |
be2c0327 LP |
1216 | val = c->memory_max; |
1217 | log_cgroup_compat(u, "Applying MemoryMax=%" PRIi64 " as MemoryLimit=", val); | |
1218 | } else | |
1219 | val = c->memory_limit; | |
78a4ee59 | 1220 | |
be2c0327 LP |
1221 | if (val == CGROUP_LIMIT_MAX) |
1222 | strncpy(buf, "-1\n", sizeof(buf)); | |
1223 | else | |
1224 | xsprintf(buf, "%" PRIu64 "\n", val); | |
1225 | ||
1226 | (void) set_attribute_and_warn(u, "memory", "memory.limit_in_bytes", buf); | |
da4d897e | 1227 | } |
4ad49000 | 1228 | } |
8e274523 | 1229 | |
4e1dfa45 | 1230 | /* On cgroup v2 we can apply BPF everywhere. On cgroup v1 we apply it everywhere except for the root of |
52fecf20 LP |
1231 | * containers, where we leave this to the manager */ |
1232 | if ((apply_mask & (CGROUP_MASK_DEVICES | CGROUP_MASK_BPF_DEVICES)) && | |
1233 | (is_host_root || cg_all_unified() > 0 || !is_local_root)) { | |
084c7007 | 1234 | _cleanup_(bpf_program_unrefp) BPFProgram *prog = NULL; |
4ad49000 | 1235 | CGroupDeviceAllow *a; |
8e274523 | 1236 | |
084c7007 RG |
1237 | if (cg_all_unified() > 0) { |
1238 | r = cgroup_init_device_bpf(&prog, c->device_policy, c->device_allow); | |
1239 | if (r < 0) | |
1240 | log_unit_warning_errno(u, r, "Failed to initialize device control bpf program: %m"); | |
1241 | } else { | |
8c838407 | 1242 | /* Changing the devices list of a populated cgroup might result in EINVAL, hence ignore EINVAL |
084c7007 | 1243 | * here. */ |
714e2e1d | 1244 | |
084c7007 RG |
1245 | if (c->device_allow || c->device_policy != CGROUP_AUTO) |
1246 | r = cg_set_attribute("devices", path, "devices.deny", "a"); | |
1247 | else | |
1248 | r = cg_set_attribute("devices", path, "devices.allow", "a"); | |
1249 | if (r < 0) | |
2c74e12b LP |
1250 | log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EINVAL, -EACCES, -EPERM) ? LOG_DEBUG : LOG_WARNING, r, |
1251 | "Failed to reset devices.allow/devices.deny: %m"); | |
084c7007 | 1252 | } |
fb385181 | 1253 | |
4ad49000 LP |
1254 | if (c->device_policy == CGROUP_CLOSED || |
1255 | (c->device_policy == CGROUP_AUTO && c->device_allow)) { | |
1256 | static const char auto_devices[] = | |
7d711efb LP |
1257 | "/dev/null\0" "rwm\0" |
1258 | "/dev/zero\0" "rwm\0" | |
1259 | "/dev/full\0" "rwm\0" | |
1260 | "/dev/random\0" "rwm\0" | |
1261 | "/dev/urandom\0" "rwm\0" | |
1262 | "/dev/tty\0" "rwm\0" | |
5a7f87a9 | 1263 | "/dev/ptmx\0" "rwm\0" |
0d9e7991 | 1264 | /* Allow /run/systemd/inaccessible/{chr,blk} devices for mapping InaccessiblePaths */ |
74c48bf5 LP |
1265 | "/run/systemd/inaccessible/chr\0" "rwm\0" |
1266 | "/run/systemd/inaccessible/blk\0" "rwm\0"; | |
4ad49000 LP |
1267 | |
1268 | const char *x, *y; | |
1269 | ||
1270 | NULSTR_FOREACH_PAIR(x, y, auto_devices) | |
913c898c | 1271 | (void) whitelist_device(prog, path, x, y); |
7d711efb | 1272 | |
5a7f87a9 | 1273 | /* PTS (/dev/pts) devices may not be duplicated, but accessed */ |
913c898c | 1274 | (void) whitelist_major(prog, path, "pts", 'c', "rw"); |
4ad49000 LP |
1275 | } |
1276 | ||
1277 | LIST_FOREACH(device_allow, a, c->device_allow) { | |
fb4650aa | 1278 | char acc[4], *val; |
4ad49000 LP |
1279 | unsigned k = 0; |
1280 | ||
1281 | if (a->r) | |
1282 | acc[k++] = 'r'; | |
1283 | if (a->w) | |
1284 | acc[k++] = 'w'; | |
1285 | if (a->m) | |
1286 | acc[k++] = 'm'; | |
fb385181 | 1287 | |
4ad49000 LP |
1288 | if (k == 0) |
1289 | continue; | |
fb385181 | 1290 | |
4ad49000 | 1291 | acc[k++] = 0; |
90060676 | 1292 | |
27458ed6 | 1293 | if (path_startswith(a->path, "/dev/")) |
913c898c | 1294 | (void) whitelist_device(prog, path, a->path, acc); |
fb4650aa | 1295 | else if ((val = startswith(a->path, "block-"))) |
913c898c | 1296 | (void) whitelist_major(prog, path, val, 'b', acc); |
fb4650aa | 1297 | else if ((val = startswith(a->path, "char-"))) |
913c898c | 1298 | (void) whitelist_major(prog, path, val, 'c', acc); |
90060676 | 1299 | else |
8e8b5d2e | 1300 | log_unit_debug(u, "Ignoring device '%s' while writing cgroup attribute.", a->path); |
4ad49000 | 1301 | } |
084c7007 RG |
1302 | |
1303 | r = cgroup_apply_device_bpf(u, prog, c->device_policy, c->device_allow); | |
1304 | if (r < 0) { | |
1305 | static bool warned = false; | |
1306 | ||
1307 | log_full_errno(warned ? LOG_DEBUG : LOG_WARNING, r, | |
1308 | "Unit %s configures device ACL, but the local system doesn't seem to support the BPF-based device controller.\n" | |
1309 | "Proceeding WITHOUT applying ACL (all devices will be accessible)!\n" | |
1310 | "(This warning is only shown for the first loaded unit using device ACL.)", u->id); | |
1311 | ||
1312 | warned = true; | |
1313 | } | |
4ad49000 | 1314 | } |
03a7b521 | 1315 | |
00b5974f LP |
1316 | if (apply_mask & CGROUP_MASK_PIDS) { |
1317 | ||
52fecf20 | 1318 | if (is_host_root) { |
00b5974f LP |
1319 | /* So, the "pids" controller does not expose anything on the root cgroup, in order not to |
1320 | * replicate knobs exposed elsewhere needlessly. We abstract this away here however, and when | |
1321 | * the knobs of the root cgroup are modified propagate this to the relevant sysctls. There's a | |
1322 | * non-obvious asymmetry however: unlike the cgroup properties we don't really want to take | |
1323 | * exclusive ownership of the sysctls, but we still want to honour things if the user sets | |
1324 | * limits. Hence we employ sort of a one-way strategy: when the user sets a bounded limit | |
1325 | * through us it counts. When the user afterwards unsets it again (i.e. sets it to unbounded) | |
1326 | * it also counts. But if the user never set a limit through us (i.e. we are the default of | |
1327 | * "unbounded") we leave things unmodified. For this we manage a global boolean that we turn on | |
1328 | * the first time we set a limit. Note that this boolean is flushed out on manager reload, | |
5238e957 | 1329 | * which is desirable so that there's an official way to release control of the sysctl from |
00b5974f LP |
1330 | * systemd: set the limit to unbounded and reload. */ |
1331 | ||
1332 | if (c->tasks_max != CGROUP_LIMIT_MAX) { | |
1333 | u->manager->sysctl_pid_max_changed = true; | |
1334 | r = procfs_tasks_set_limit(c->tasks_max); | |
1335 | } else if (u->manager->sysctl_pid_max_changed) | |
1336 | r = procfs_tasks_set_limit(TASKS_MAX); | |
1337 | else | |
1338 | r = 0; | |
00b5974f | 1339 | if (r < 0) |
39b9fefb | 1340 | log_unit_full(u, LOG_LEVEL_CGROUP_WRITE(r), r, |
00b5974f | 1341 | "Failed to write to tasks limit sysctls: %m"); |
52fecf20 | 1342 | } |
03a7b521 | 1343 | |
52fecf20 LP |
1344 | /* The attribute itself is not available on the host root cgroup, and in the container case we want to |
1345 | * leave it for the container manager. */ | |
1346 | if (!is_local_root) { | |
00b5974f LP |
1347 | if (c->tasks_max != CGROUP_LIMIT_MAX) { |
1348 | char buf[DECIMAL_STR_MAX(uint64_t) + 2]; | |
03a7b521 | 1349 | |
00b5974f | 1350 | sprintf(buf, "%" PRIu64 "\n", c->tasks_max); |
293d32df | 1351 | (void) set_attribute_and_warn(u, "pids", "pids.max", buf); |
00b5974f | 1352 | } else |
589a5f7a | 1353 | (void) set_attribute_and_warn(u, "pids", "pids.max", "max\n"); |
00b5974f | 1354 | } |
03a7b521 | 1355 | } |
906c06f6 | 1356 | |
17f14955 | 1357 | if (apply_mask & CGROUP_MASK_BPF_FIREWALL) |
0f2d84d2 | 1358 | cgroup_apply_firewall(u); |
fb385181 LP |
1359 | } |
1360 | ||
16492445 LP |
1361 | static bool unit_get_needs_bpf_firewall(Unit *u) { |
1362 | CGroupContext *c; | |
1363 | Unit *p; | |
1364 | assert(u); | |
1365 | ||
1366 | c = unit_get_cgroup_context(u); | |
1367 | if (!c) | |
1368 | return false; | |
1369 | ||
1370 | if (c->ip_accounting || | |
1371 | c->ip_address_allow || | |
fab34748 KL |
1372 | c->ip_address_deny || |
1373 | c->ip_filters_ingress || | |
1374 | c->ip_filters_egress) | |
16492445 LP |
1375 | return true; |
1376 | ||
1377 | /* If any parent slice has an IP access list defined, it applies too */ | |
1378 | for (p = UNIT_DEREF(u->slice); p; p = UNIT_DEREF(p->slice)) { | |
1379 | c = unit_get_cgroup_context(p); | |
1380 | if (!c) | |
1381 | return false; | |
1382 | ||
1383 | if (c->ip_address_allow || | |
1384 | c->ip_address_deny) | |
1385 | return true; | |
1386 | } | |
1387 | ||
1388 | return false; | |
1389 | } | |
1390 | ||
c52db42b | 1391 | static CGroupMask unit_get_cgroup_mask(Unit *u) { |
efdb0237 | 1392 | CGroupMask mask = 0; |
c52db42b CD |
1393 | CGroupContext *c; |
1394 | ||
1395 | assert(u); | |
1396 | ||
1397 | c = unit_get_cgroup_context(u); | |
8e274523 | 1398 | |
c710d3b4 CD |
1399 | assert(c); |
1400 | ||
fae9bc29 | 1401 | /* Figure out which controllers we need, based on the cgroup context object */ |
8e274523 | 1402 | |
fae9bc29 | 1403 | if (c->cpu_accounting) |
f98c2585 | 1404 | mask |= get_cpu_accounting_mask(); |
fae9bc29 LP |
1405 | |
1406 | if (cgroup_context_has_cpu_weight(c) || | |
66ebf6c0 | 1407 | cgroup_context_has_cpu_shares(c) || |
3a43da28 | 1408 | c->cpu_quota_per_sec_usec != USEC_INFINITY) |
fae9bc29 | 1409 | mask |= CGROUP_MASK_CPU; |
ecedd90f | 1410 | |
538b4852 TH |
1411 | if (cgroup_context_has_io_config(c) || cgroup_context_has_blockio_config(c)) |
1412 | mask |= CGROUP_MASK_IO | CGROUP_MASK_BLKIO; | |
ecedd90f | 1413 | |
4ad49000 | 1414 | if (c->memory_accounting || |
da4d897e | 1415 | c->memory_limit != CGROUP_LIMIT_MAX || |
c52db42b | 1416 | unit_has_unified_memory_config(u)) |
efdb0237 | 1417 | mask |= CGROUP_MASK_MEMORY; |
8e274523 | 1418 | |
a931ad47 LP |
1419 | if (c->device_allow || |
1420 | c->device_policy != CGROUP_AUTO) | |
084c7007 | 1421 | mask |= CGROUP_MASK_DEVICES | CGROUP_MASK_BPF_DEVICES; |
4ad49000 | 1422 | |
03a7b521 | 1423 | if (c->tasks_accounting || |
8793fa25 | 1424 | c->tasks_max != CGROUP_LIMIT_MAX) |
03a7b521 LP |
1425 | mask |= CGROUP_MASK_PIDS; |
1426 | ||
fae9bc29 | 1427 | return CGROUP_MASK_EXTEND_JOINED(mask); |
8e274523 LP |
1428 | } |
1429 | ||
53aea74a | 1430 | static CGroupMask unit_get_bpf_mask(Unit *u) { |
17f14955 RG |
1431 | CGroupMask mask = 0; |
1432 | ||
fae9bc29 LP |
1433 | /* Figure out which controllers we need, based on the cgroup context, possibly taking into account children |
1434 | * too. */ | |
1435 | ||
17f14955 RG |
1436 | if (unit_get_needs_bpf_firewall(u)) |
1437 | mask |= CGROUP_MASK_BPF_FIREWALL; | |
1438 | ||
1439 | return mask; | |
1440 | } | |
1441 | ||
efdb0237 | 1442 | CGroupMask unit_get_own_mask(Unit *u) { |
4ad49000 | 1443 | CGroupContext *c; |
8e274523 | 1444 | |
442ce775 LP |
1445 | /* Returns the mask of controllers the unit needs for itself. If a unit is not properly loaded, return an empty |
1446 | * mask, as we shouldn't reflect it in the cgroup hierarchy then. */ | |
1447 | ||
1448 | if (u->load_state != UNIT_LOADED) | |
1449 | return 0; | |
efdb0237 | 1450 | |
4ad49000 LP |
1451 | c = unit_get_cgroup_context(u); |
1452 | if (!c) | |
1453 | return 0; | |
8e274523 | 1454 | |
c52db42b | 1455 | return (unit_get_cgroup_mask(u) | unit_get_bpf_mask(u) | unit_get_delegate_mask(u)) & ~unit_get_ancestor_disable_mask(u); |
02638280 LP |
1456 | } |
1457 | ||
1458 | CGroupMask unit_get_delegate_mask(Unit *u) { | |
1459 | CGroupContext *c; | |
1460 | ||
1461 | /* If delegation is turned on, then turn on selected controllers, unless we are on the legacy hierarchy and the | |
1462 | * process we fork into is known to drop privileges, and hence shouldn't get access to the controllers. | |
19af675e | 1463 | * |
02638280 | 1464 | * Note that on the unified hierarchy it is safe to delegate controllers to unprivileged services. */ |
a931ad47 | 1465 | |
1d9cc876 | 1466 | if (!unit_cgroup_delegate(u)) |
02638280 LP |
1467 | return 0; |
1468 | ||
1469 | if (cg_all_unified() <= 0) { | |
a931ad47 LP |
1470 | ExecContext *e; |
1471 | ||
1472 | e = unit_get_exec_context(u); | |
02638280 LP |
1473 | if (e && !exec_context_maintains_privileges(e)) |
1474 | return 0; | |
a931ad47 LP |
1475 | } |
1476 | ||
1d9cc876 | 1477 | assert_se(c = unit_get_cgroup_context(u)); |
fae9bc29 | 1478 | return CGROUP_MASK_EXTEND_JOINED(c->delegate_controllers); |
8e274523 LP |
1479 | } |
1480 | ||
efdb0237 | 1481 | CGroupMask unit_get_members_mask(Unit *u) { |
4ad49000 | 1482 | assert(u); |
bc432dc7 | 1483 | |
02638280 | 1484 | /* Returns the mask of controllers all of the unit's children require, merged */ |
efdb0237 | 1485 | |
bc432dc7 | 1486 | if (u->cgroup_members_mask_valid) |
26a17ca2 | 1487 | return u->cgroup_members_mask; /* Use cached value if possible */ |
bc432dc7 | 1488 | |
64e844e5 | 1489 | u->cgroup_members_mask = 0; |
bc432dc7 LP |
1490 | |
1491 | if (u->type == UNIT_SLICE) { | |
eef85c4a | 1492 | void *v; |
bc432dc7 LP |
1493 | Unit *member; |
1494 | Iterator i; | |
1495 | ||
eef85c4a | 1496 | HASHMAP_FOREACH_KEY(v, member, u->dependencies[UNIT_BEFORE], i) { |
cb5e3bc3 CD |
1497 | if (UNIT_DEREF(member->slice) == u) |
1498 | u->cgroup_members_mask |= unit_get_subtree_mask(member); /* note that this calls ourselves again, for the children */ | |
bc432dc7 LP |
1499 | } |
1500 | } | |
1501 | ||
1502 | u->cgroup_members_mask_valid = true; | |
6414b7c9 | 1503 | return u->cgroup_members_mask; |
246aa6dd LP |
1504 | } |
1505 | ||
efdb0237 | 1506 | CGroupMask unit_get_siblings_mask(Unit *u) { |
4ad49000 | 1507 | assert(u); |
246aa6dd | 1508 | |
efdb0237 LP |
1509 | /* Returns the mask of controllers all of the unit's siblings |
1510 | * require, i.e. the members mask of the unit's parent slice | |
1511 | * if there is one. */ | |
1512 | ||
bc432dc7 | 1513 | if (UNIT_ISSET(u->slice)) |
637f421e | 1514 | return unit_get_members_mask(UNIT_DEREF(u->slice)); |
4ad49000 | 1515 | |
64e844e5 | 1516 | return unit_get_subtree_mask(u); /* we are the top-level slice */ |
246aa6dd LP |
1517 | } |
1518 | ||
4f6f62e4 CD |
1519 | CGroupMask unit_get_disable_mask(Unit *u) { |
1520 | CGroupContext *c; | |
1521 | ||
1522 | c = unit_get_cgroup_context(u); | |
1523 | if (!c) | |
1524 | return 0; | |
1525 | ||
1526 | return c->disable_controllers; | |
1527 | } | |
1528 | ||
1529 | CGroupMask unit_get_ancestor_disable_mask(Unit *u) { | |
1530 | CGroupMask mask; | |
1531 | ||
1532 | assert(u); | |
1533 | mask = unit_get_disable_mask(u); | |
1534 | ||
1535 | /* Returns the mask of controllers which are marked as forcibly | |
1536 | * disabled in any ancestor unit or the unit in question. */ | |
1537 | ||
1538 | if (UNIT_ISSET(u->slice)) | |
1539 | mask |= unit_get_ancestor_disable_mask(UNIT_DEREF(u->slice)); | |
1540 | ||
1541 | return mask; | |
1542 | } | |
1543 | ||
efdb0237 LP |
1544 | CGroupMask unit_get_subtree_mask(Unit *u) { |
1545 | ||
1546 | /* Returns the mask of this subtree, meaning of the group | |
1547 | * itself and its children. */ | |
1548 | ||
1549 | return unit_get_own_mask(u) | unit_get_members_mask(u); | |
1550 | } | |
1551 | ||
1552 | CGroupMask unit_get_target_mask(Unit *u) { | |
1553 | CGroupMask mask; | |
1554 | ||
1555 | /* This returns the cgroup mask of all controllers to enable | |
1556 | * for a specific cgroup, i.e. everything it needs itself, | |
1557 | * plus all that its children need, plus all that its siblings | |
1558 | * need. This is primarily useful on the legacy cgroup | |
1559 | * hierarchy, where we need to duplicate each cgroup in each | |
1560 | * hierarchy that shall be enabled for it. */ | |
6414b7c9 | 1561 | |
efdb0237 | 1562 | mask = unit_get_own_mask(u) | unit_get_members_mask(u) | unit_get_siblings_mask(u); |
84d2744b ZJS |
1563 | |
1564 | if (mask & CGROUP_MASK_BPF_FIREWALL & ~u->manager->cgroup_supported) | |
1565 | emit_bpf_firewall_warning(u); | |
1566 | ||
efdb0237 | 1567 | mask &= u->manager->cgroup_supported; |
c72703e2 | 1568 | mask &= ~unit_get_ancestor_disable_mask(u); |
efdb0237 LP |
1569 | |
1570 | return mask; | |
1571 | } | |
1572 | ||
1573 | CGroupMask unit_get_enable_mask(Unit *u) { | |
1574 | CGroupMask mask; | |
1575 | ||
1576 | /* This returns the cgroup mask of all controllers to enable | |
1577 | * for the children of a specific cgroup. This is primarily | |
1578 | * useful for the unified cgroup hierarchy, where each cgroup | |
1579 | * controls which controllers are enabled for its children. */ | |
1580 | ||
1581 | mask = unit_get_members_mask(u); | |
6414b7c9 | 1582 | mask &= u->manager->cgroup_supported; |
c72703e2 | 1583 | mask &= ~unit_get_ancestor_disable_mask(u); |
6414b7c9 DS |
1584 | |
1585 | return mask; | |
1586 | } | |
1587 | ||
5af88058 | 1588 | void unit_invalidate_cgroup_members_masks(Unit *u) { |
bc432dc7 LP |
1589 | assert(u); |
1590 | ||
5af88058 LP |
1591 | /* Recurse invalidate the member masks cache all the way up the tree */ |
1592 | u->cgroup_members_mask_valid = false; | |
bc432dc7 | 1593 | |
5af88058 LP |
1594 | if (UNIT_ISSET(u->slice)) |
1595 | unit_invalidate_cgroup_members_masks(UNIT_DEREF(u->slice)); | |
6414b7c9 DS |
1596 | } |
1597 | ||
6592b975 | 1598 | const char *unit_get_realized_cgroup_path(Unit *u, CGroupMask mask) { |
03b90d4b | 1599 | |
6592b975 | 1600 | /* Returns the realized cgroup path of the specified unit where all specified controllers are available. */ |
03b90d4b LP |
1601 | |
1602 | while (u) { | |
6592b975 | 1603 | |
03b90d4b LP |
1604 | if (u->cgroup_path && |
1605 | u->cgroup_realized && | |
d94a24ca | 1606 | FLAGS_SET(u->cgroup_realized_mask, mask)) |
03b90d4b LP |
1607 | return u->cgroup_path; |
1608 | ||
1609 | u = UNIT_DEREF(u->slice); | |
1610 | } | |
1611 | ||
1612 | return NULL; | |
1613 | } | |
1614 | ||
6592b975 LP |
1615 | static const char *migrate_callback(CGroupMask mask, void *userdata) { |
1616 | return unit_get_realized_cgroup_path(userdata, mask); | |
1617 | } | |
1618 | ||
303ee601 | 1619 | char *unit_default_cgroup_path(const Unit *u) { |
efdb0237 LP |
1620 | _cleanup_free_ char *escaped = NULL, *slice = NULL; |
1621 | int r; | |
1622 | ||
1623 | assert(u); | |
1624 | ||
1625 | if (unit_has_name(u, SPECIAL_ROOT_SLICE)) | |
1626 | return strdup(u->manager->cgroup_root); | |
1627 | ||
1628 | if (UNIT_ISSET(u->slice) && !unit_has_name(UNIT_DEREF(u->slice), SPECIAL_ROOT_SLICE)) { | |
1629 | r = cg_slice_to_path(UNIT_DEREF(u->slice)->id, &slice); | |
1630 | if (r < 0) | |
1631 | return NULL; | |
1632 | } | |
1633 | ||
1634 | escaped = cg_escape(u->id); | |
1635 | if (!escaped) | |
1636 | return NULL; | |
1637 | ||
657ee2d8 | 1638 | return path_join(empty_to_root(u->manager->cgroup_root), slice, escaped); |
efdb0237 LP |
1639 | } |
1640 | ||
1641 | int unit_set_cgroup_path(Unit *u, const char *path) { | |
1642 | _cleanup_free_ char *p = NULL; | |
1643 | int r; | |
1644 | ||
1645 | assert(u); | |
1646 | ||
5210387e LP |
1647 | if (streq_ptr(u->cgroup_path, path)) |
1648 | return 0; | |
1649 | ||
efdb0237 LP |
1650 | if (path) { |
1651 | p = strdup(path); | |
1652 | if (!p) | |
1653 | return -ENOMEM; | |
5210387e | 1654 | } |
efdb0237 LP |
1655 | |
1656 | if (p) { | |
1657 | r = hashmap_put(u->manager->cgroup_unit, p, u); | |
1658 | if (r < 0) | |
1659 | return r; | |
1660 | } | |
1661 | ||
1662 | unit_release_cgroup(u); | |
ae2a15bc | 1663 | u->cgroup_path = TAKE_PTR(p); |
efdb0237 LP |
1664 | |
1665 | return 1; | |
1666 | } | |
1667 | ||
1668 | int unit_watch_cgroup(Unit *u) { | |
ab2c3861 | 1669 | _cleanup_free_ char *events = NULL; |
efdb0237 LP |
1670 | int r; |
1671 | ||
1672 | assert(u); | |
1673 | ||
0bb814c2 LP |
1674 | /* Watches the "cgroups.events" attribute of this unit's cgroup for "empty" events, but only if |
1675 | * cgroupv2 is available. */ | |
1676 | ||
efdb0237 LP |
1677 | if (!u->cgroup_path) |
1678 | return 0; | |
1679 | ||
0bb814c2 | 1680 | if (u->cgroup_control_inotify_wd >= 0) |
efdb0237 LP |
1681 | return 0; |
1682 | ||
1683 | /* Only applies to the unified hierarchy */ | |
c22800e4 | 1684 | r = cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER); |
b4cccbc1 LP |
1685 | if (r < 0) |
1686 | return log_error_errno(r, "Failed to determine whether the name=systemd hierarchy is unified: %m"); | |
1687 | if (r == 0) | |
efdb0237 LP |
1688 | return 0; |
1689 | ||
0bb814c2 | 1690 | /* No point in watch the top-level slice, it's never going to run empty. */ |
efdb0237 LP |
1691 | if (unit_has_name(u, SPECIAL_ROOT_SLICE)) |
1692 | return 0; | |
1693 | ||
0bb814c2 | 1694 | r = hashmap_ensure_allocated(&u->manager->cgroup_control_inotify_wd_unit, &trivial_hash_ops); |
efdb0237 LP |
1695 | if (r < 0) |
1696 | return log_oom(); | |
1697 | ||
ab2c3861 | 1698 | r = cg_get_path(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path, "cgroup.events", &events); |
efdb0237 LP |
1699 | if (r < 0) |
1700 | return log_oom(); | |
1701 | ||
0bb814c2 LP |
1702 | u->cgroup_control_inotify_wd = inotify_add_watch(u->manager->cgroup_inotify_fd, events, IN_MODIFY); |
1703 | if (u->cgroup_control_inotify_wd < 0) { | |
efdb0237 | 1704 | |
0bb814c2 LP |
1705 | if (errno == ENOENT) /* If the directory is already gone we don't need to track it, so this |
1706 | * is not an error */ | |
efdb0237 LP |
1707 | return 0; |
1708 | ||
0bb814c2 | 1709 | return log_unit_error_errno(u, errno, "Failed to add control inotify watch descriptor for control group %s: %m", u->cgroup_path); |
efdb0237 LP |
1710 | } |
1711 | ||
0bb814c2 | 1712 | r = hashmap_put(u->manager->cgroup_control_inotify_wd_unit, INT_TO_PTR(u->cgroup_control_inotify_wd), u); |
efdb0237 | 1713 | if (r < 0) |
0bb814c2 | 1714 | return log_unit_error_errno(u, r, "Failed to add control inotify watch descriptor to hash map: %m"); |
efdb0237 LP |
1715 | |
1716 | return 0; | |
1717 | } | |
1718 | ||
afcfaa69 LP |
1719 | int unit_watch_cgroup_memory(Unit *u) { |
1720 | _cleanup_free_ char *events = NULL; | |
1721 | CGroupContext *c; | |
1722 | int r; | |
1723 | ||
1724 | assert(u); | |
1725 | ||
1726 | /* Watches the "memory.events" attribute of this unit's cgroup for "oom_kill" events, but only if | |
1727 | * cgroupv2 is available. */ | |
1728 | ||
1729 | if (!u->cgroup_path) | |
1730 | return 0; | |
1731 | ||
1732 | c = unit_get_cgroup_context(u); | |
1733 | if (!c) | |
1734 | return 0; | |
1735 | ||
1736 | /* The "memory.events" attribute is only available if the memory controller is on. Let's hence tie | |
1737 | * this to memory accounting, in a way watching for OOM kills is a form of memory accounting after | |
1738 | * all. */ | |
1739 | if (!c->memory_accounting) | |
1740 | return 0; | |
1741 | ||
1742 | /* Don't watch inner nodes, as the kernel doesn't report oom_kill events recursively currently, and | |
1743 | * we also don't want to generate a log message for each parent cgroup of a process. */ | |
1744 | if (u->type == UNIT_SLICE) | |
1745 | return 0; | |
1746 | ||
1747 | if (u->cgroup_memory_inotify_wd >= 0) | |
1748 | return 0; | |
1749 | ||
1750 | /* Only applies to the unified hierarchy */ | |
1751 | r = cg_all_unified(); | |
1752 | if (r < 0) | |
1753 | return log_error_errno(r, "Failed to determine whether the memory controller is unified: %m"); | |
1754 | if (r == 0) | |
1755 | return 0; | |
1756 | ||
1757 | r = hashmap_ensure_allocated(&u->manager->cgroup_memory_inotify_wd_unit, &trivial_hash_ops); | |
1758 | if (r < 0) | |
1759 | return log_oom(); | |
1760 | ||
1761 | r = cg_get_path(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path, "memory.events", &events); | |
1762 | if (r < 0) | |
1763 | return log_oom(); | |
1764 | ||
1765 | u->cgroup_memory_inotify_wd = inotify_add_watch(u->manager->cgroup_inotify_fd, events, IN_MODIFY); | |
1766 | if (u->cgroup_memory_inotify_wd < 0) { | |
1767 | ||
1768 | if (errno == ENOENT) /* If the directory is already gone we don't need to track it, so this | |
1769 | * is not an error */ | |
1770 | return 0; | |
1771 | ||
1772 | return log_unit_error_errno(u, errno, "Failed to add memory inotify watch descriptor for control group %s: %m", u->cgroup_path); | |
1773 | } | |
1774 | ||
1775 | r = hashmap_put(u->manager->cgroup_memory_inotify_wd_unit, INT_TO_PTR(u->cgroup_memory_inotify_wd), u); | |
1776 | if (r < 0) | |
1777 | return log_unit_error_errno(u, r, "Failed to add memory inotify watch descriptor to hash map: %m"); | |
1778 | ||
1779 | return 0; | |
1780 | } | |
1781 | ||
a4634b21 LP |
1782 | int unit_pick_cgroup_path(Unit *u) { |
1783 | _cleanup_free_ char *path = NULL; | |
1784 | int r; | |
1785 | ||
1786 | assert(u); | |
1787 | ||
1788 | if (u->cgroup_path) | |
1789 | return 0; | |
1790 | ||
1791 | if (!UNIT_HAS_CGROUP_CONTEXT(u)) | |
1792 | return -EINVAL; | |
1793 | ||
1794 | path = unit_default_cgroup_path(u); | |
1795 | if (!path) | |
1796 | return log_oom(); | |
1797 | ||
1798 | r = unit_set_cgroup_path(u, path); | |
1799 | if (r == -EEXIST) | |
1800 | return log_unit_error_errno(u, r, "Control group %s exists already.", path); | |
1801 | if (r < 0) | |
1802 | return log_unit_error_errno(u, r, "Failed to set unit's control group path to %s: %m", path); | |
1803 | ||
1804 | return 0; | |
1805 | } | |
1806 | ||
efdb0237 LP |
1807 | static int unit_create_cgroup( |
1808 | Unit *u, | |
1809 | CGroupMask target_mask, | |
0d2d6fbf CD |
1810 | CGroupMask enable_mask, |
1811 | ManagerState state) { | |
efdb0237 | 1812 | |
65be7e06 | 1813 | bool created; |
27adcc97 | 1814 | int r; |
64747e2d | 1815 | |
4ad49000 | 1816 | assert(u); |
64747e2d | 1817 | |
27c4ed79 | 1818 | if (!UNIT_HAS_CGROUP_CONTEXT(u)) |
0cd385d3 LP |
1819 | return 0; |
1820 | ||
a4634b21 LP |
1821 | /* Figure out our cgroup path */ |
1822 | r = unit_pick_cgroup_path(u); | |
1823 | if (r < 0) | |
1824 | return r; | |
b58b8e11 | 1825 | |
03b90d4b | 1826 | /* First, create our own group */ |
efdb0237 | 1827 | r = cg_create_everywhere(u->manager->cgroup_supported, target_mask, u->cgroup_path); |
23bbb0de | 1828 | if (r < 0) |
efdb0237 | 1829 | return log_unit_error_errno(u, r, "Failed to create cgroup %s: %m", u->cgroup_path); |
490c5a37 | 1830 | created = r; |
efdb0237 LP |
1831 | |
1832 | /* Start watching it */ | |
1833 | (void) unit_watch_cgroup(u); | |
afcfaa69 | 1834 | (void) unit_watch_cgroup_memory(u); |
efdb0237 | 1835 | |
65be7e06 | 1836 | /* Preserve enabled controllers in delegated units, adjust others. */ |
1fd3a10c | 1837 | if (created || !u->cgroup_realized || !unit_cgroup_delegate(u)) { |
27adcc97 | 1838 | CGroupMask result_mask = 0; |
65be7e06 ZJS |
1839 | |
1840 | /* Enable all controllers we need */ | |
27adcc97 | 1841 | r = cg_enable_everywhere(u->manager->cgroup_supported, enable_mask, u->cgroup_path, &result_mask); |
65be7e06 | 1842 | if (r < 0) |
27adcc97 LP |
1843 | log_unit_warning_errno(u, r, "Failed to enable/disable controllers on cgroup %s, ignoring: %m", u->cgroup_path); |
1844 | ||
1845 | /* If we just turned off a controller, this might release the controller for our parent too, let's | |
1846 | * enqueue the parent for re-realization in that case again. */ | |
1847 | if (UNIT_ISSET(u->slice)) { | |
1848 | CGroupMask turned_off; | |
1849 | ||
1850 | turned_off = (u->cgroup_realized ? u->cgroup_enabled_mask & ~result_mask : 0); | |
1851 | if (turned_off != 0) { | |
1852 | Unit *parent; | |
1853 | ||
1854 | /* Force the parent to propagate the enable mask to the kernel again, by invalidating | |
1855 | * the controller we just turned off. */ | |
1856 | ||
1857 | for (parent = UNIT_DEREF(u->slice); parent; parent = UNIT_DEREF(parent->slice)) | |
1858 | unit_invalidate_cgroup(parent, turned_off); | |
1859 | } | |
1860 | } | |
1861 | ||
1862 | /* Remember what's actually enabled now */ | |
1863 | u->cgroup_enabled_mask = result_mask; | |
65be7e06 | 1864 | } |
03b90d4b LP |
1865 | |
1866 | /* Keep track that this is now realized */ | |
4ad49000 | 1867 | u->cgroup_realized = true; |
efdb0237 | 1868 | u->cgroup_realized_mask = target_mask; |
4ad49000 | 1869 | |
1d9cc876 | 1870 | if (u->type != UNIT_SLICE && !unit_cgroup_delegate(u)) { |
0cd385d3 LP |
1871 | |
1872 | /* Then, possibly move things over, but not if | |
1873 | * subgroups may contain processes, which is the case | |
1874 | * for slice and delegation units. */ | |
1875 | r = cg_migrate_everywhere(u->manager->cgroup_supported, u->cgroup_path, u->cgroup_path, migrate_callback, u); | |
1876 | if (r < 0) | |
efdb0237 | 1877 | log_unit_warning_errno(u, r, "Failed to migrate cgroup from to %s, ignoring: %m", u->cgroup_path); |
0cd385d3 | 1878 | } |
03b90d4b | 1879 | |
0d2d6fbf CD |
1880 | /* Set attributes */ |
1881 | cgroup_context_apply(u, target_mask, state); | |
1882 | cgroup_xattr_apply(u); | |
1883 | ||
64747e2d LP |
1884 | return 0; |
1885 | } | |
1886 | ||
6592b975 LP |
1887 | static int unit_attach_pid_to_cgroup_via_bus(Unit *u, pid_t pid, const char *suffix_path) { |
1888 | _cleanup_(sd_bus_error_free) sd_bus_error error = SD_BUS_ERROR_NULL; | |
1889 | char *pp; | |
7b3fd631 | 1890 | int r; |
6592b975 | 1891 | |
7b3fd631 LP |
1892 | assert(u); |
1893 | ||
6592b975 LP |
1894 | if (MANAGER_IS_SYSTEM(u->manager)) |
1895 | return -EINVAL; | |
1896 | ||
1897 | if (!u->manager->system_bus) | |
1898 | return -EIO; | |
1899 | ||
1900 | if (!u->cgroup_path) | |
1901 | return -EINVAL; | |
1902 | ||
1903 | /* Determine this unit's cgroup path relative to our cgroup root */ | |
1904 | pp = path_startswith(u->cgroup_path, u->manager->cgroup_root); | |
1905 | if (!pp) | |
1906 | return -EINVAL; | |
1907 | ||
1908 | pp = strjoina("/", pp, suffix_path); | |
858d36c1 | 1909 | path_simplify(pp, false); |
6592b975 LP |
1910 | |
1911 | r = sd_bus_call_method(u->manager->system_bus, | |
1912 | "org.freedesktop.systemd1", | |
1913 | "/org/freedesktop/systemd1", | |
1914 | "org.freedesktop.systemd1.Manager", | |
1915 | "AttachProcessesToUnit", | |
1916 | &error, NULL, | |
1917 | "ssau", | |
1918 | NULL /* empty unit name means client's unit, i.e. us */, pp, 1, (uint32_t) pid); | |
7b3fd631 | 1919 | if (r < 0) |
6592b975 LP |
1920 | return log_unit_debug_errno(u, r, "Failed to attach unit process " PID_FMT " via the bus: %s", pid, bus_error_message(&error, r)); |
1921 | ||
1922 | return 0; | |
1923 | } | |
1924 | ||
1925 | int unit_attach_pids_to_cgroup(Unit *u, Set *pids, const char *suffix_path) { | |
1926 | CGroupMask delegated_mask; | |
1927 | const char *p; | |
1928 | Iterator i; | |
1929 | void *pidp; | |
1930 | int r, q; | |
1931 | ||
1932 | assert(u); | |
1933 | ||
1934 | if (!UNIT_HAS_CGROUP_CONTEXT(u)) | |
1935 | return -EINVAL; | |
1936 | ||
1937 | if (set_isempty(pids)) | |
1938 | return 0; | |
7b3fd631 | 1939 | |
fab34748 KL |
1940 | /* Load any custom firewall BPF programs here once to test if they are existing and actually loadable. |
1941 | * Fail here early since later errors in the call chain unit_realize_cgroup to cgroup_context_apply are ignored. */ | |
1942 | r = bpf_firewall_load_custom(u); | |
1943 | if (r < 0) | |
1944 | return r; | |
1945 | ||
6592b975 | 1946 | r = unit_realize_cgroup(u); |
7b3fd631 LP |
1947 | if (r < 0) |
1948 | return r; | |
1949 | ||
6592b975 LP |
1950 | if (isempty(suffix_path)) |
1951 | p = u->cgroup_path; | |
1952 | else | |
270384b2 | 1953 | p = prefix_roota(u->cgroup_path, suffix_path); |
6592b975 LP |
1954 | |
1955 | delegated_mask = unit_get_delegate_mask(u); | |
1956 | ||
1957 | r = 0; | |
1958 | SET_FOREACH(pidp, pids, i) { | |
1959 | pid_t pid = PTR_TO_PID(pidp); | |
1960 | CGroupController c; | |
1961 | ||
1962 | /* First, attach the PID to the main cgroup hierarchy */ | |
1963 | q = cg_attach(SYSTEMD_CGROUP_CONTROLLER, p, pid); | |
1964 | if (q < 0) { | |
1965 | log_unit_debug_errno(u, q, "Couldn't move process " PID_FMT " to requested cgroup '%s': %m", pid, p); | |
1966 | ||
1967 | if (MANAGER_IS_USER(u->manager) && IN_SET(q, -EPERM, -EACCES)) { | |
1968 | int z; | |
1969 | ||
1970 | /* If we are in a user instance, and we can't move the process ourselves due to | |
1971 | * permission problems, let's ask the system instance about it instead. Since it's more | |
1972 | * privileged it might be able to move the process across the leaves of a subtree who's | |
1973 | * top node is not owned by us. */ | |
1974 | ||
1975 | z = unit_attach_pid_to_cgroup_via_bus(u, pid, suffix_path); | |
1976 | if (z < 0) | |
1977 | log_unit_debug_errno(u, z, "Couldn't move process " PID_FMT " to requested cgroup '%s' via the system bus either: %m", pid, p); | |
1978 | else | |
1979 | continue; /* When the bus thing worked via the bus we are fully done for this PID. */ | |
1980 | } | |
1981 | ||
1982 | if (r >= 0) | |
1983 | r = q; /* Remember first error */ | |
1984 | ||
1985 | continue; | |
1986 | } | |
1987 | ||
1988 | q = cg_all_unified(); | |
1989 | if (q < 0) | |
1990 | return q; | |
1991 | if (q > 0) | |
1992 | continue; | |
1993 | ||
1994 | /* In the legacy hierarchy, attach the process to the request cgroup if possible, and if not to the | |
1995 | * innermost realized one */ | |
1996 | ||
1997 | for (c = 0; c < _CGROUP_CONTROLLER_MAX; c++) { | |
1998 | CGroupMask bit = CGROUP_CONTROLLER_TO_MASK(c); | |
1999 | const char *realized; | |
2000 | ||
2001 | if (!(u->manager->cgroup_supported & bit)) | |
2002 | continue; | |
2003 | ||
2004 | /* If this controller is delegated and realized, honour the caller's request for the cgroup suffix. */ | |
2005 | if (delegated_mask & u->cgroup_realized_mask & bit) { | |
2006 | q = cg_attach(cgroup_controller_to_string(c), p, pid); | |
2007 | if (q >= 0) | |
2008 | continue; /* Success! */ | |
2009 | ||
2010 | 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", | |
2011 | pid, p, cgroup_controller_to_string(c)); | |
2012 | } | |
2013 | ||
2014 | /* So this controller is either not delegate or realized, or something else weird happened. In | |
2015 | * that case let's attach the PID at least to the closest cgroup up the tree that is | |
2016 | * realized. */ | |
2017 | realized = unit_get_realized_cgroup_path(u, bit); | |
2018 | if (!realized) | |
2019 | continue; /* Not even realized in the root slice? Then let's not bother */ | |
2020 | ||
2021 | q = cg_attach(cgroup_controller_to_string(c), realized, pid); | |
2022 | if (q < 0) | |
2023 | log_unit_debug_errno(u, q, "Failed to attach PID " PID_FMT " to realized cgroup %s in controller %s, ignoring: %m", | |
2024 | pid, realized, cgroup_controller_to_string(c)); | |
2025 | } | |
2026 | } | |
2027 | ||
2028 | return r; | |
7b3fd631 LP |
2029 | } |
2030 | ||
906c06f6 DM |
2031 | static bool unit_has_mask_realized( |
2032 | Unit *u, | |
2033 | CGroupMask target_mask, | |
17f14955 | 2034 | CGroupMask enable_mask) { |
906c06f6 | 2035 | |
bc432dc7 LP |
2036 | assert(u); |
2037 | ||
d5095dcd LP |
2038 | /* Returns true if this unit is fully realized. We check four things: |
2039 | * | |
2040 | * 1. Whether the cgroup was created at all | |
4e1dfa45 CD |
2041 | * 2. Whether the cgroup was created in all the hierarchies we need it to be created in (in case of cgroup v1) |
2042 | * 3. Whether the cgroup has all the right controllers enabled (in case of cgroup v2) | |
d5095dcd LP |
2043 | * 4. Whether the invalidation mask is currently zero |
2044 | * | |
2045 | * If you wonder why we mask the target realization and enable mask with CGROUP_MASK_V1/CGROUP_MASK_V2: note | |
4e1dfa45 CD |
2046 | * that there are three sets of bitmasks: CGROUP_MASK_V1 (for real cgroup v1 controllers), CGROUP_MASK_V2 (for |
2047 | * real cgroup v2 controllers) and CGROUP_MASK_BPF (for BPF-based pseudo-controllers). Now, cgroup_realized_mask | |
2048 | * is only matters for cgroup v1 controllers, and cgroup_enabled_mask only used for cgroup v2, and if they | |
d5095dcd LP |
2049 | * differ in the others, we don't really care. (After all, the cgroup_enabled_mask tracks with controllers are |
2050 | * enabled through cgroup.subtree_control, and since the BPF pseudo-controllers don't show up there, they | |
2051 | * simply don't matter. */ | |
2052 | ||
906c06f6 | 2053 | return u->cgroup_realized && |
d5095dcd LP |
2054 | ((u->cgroup_realized_mask ^ target_mask) & CGROUP_MASK_V1) == 0 && |
2055 | ((u->cgroup_enabled_mask ^ enable_mask) & CGROUP_MASK_V2) == 0 && | |
17f14955 | 2056 | u->cgroup_invalidated_mask == 0; |
6414b7c9 DS |
2057 | } |
2058 | ||
4f6f62e4 CD |
2059 | static bool unit_has_mask_disables_realized( |
2060 | Unit *u, | |
2061 | CGroupMask target_mask, | |
2062 | CGroupMask enable_mask) { | |
2063 | ||
2064 | assert(u); | |
2065 | ||
2066 | /* Returns true if all controllers which should be disabled are indeed disabled. | |
2067 | * | |
2068 | * Unlike unit_has_mask_realized, we don't care what was enabled, only that anything we want to remove is | |
2069 | * already removed. */ | |
2070 | ||
2071 | return !u->cgroup_realized || | |
2072 | (FLAGS_SET(u->cgroup_realized_mask, target_mask & CGROUP_MASK_V1) && | |
2073 | FLAGS_SET(u->cgroup_enabled_mask, enable_mask & CGROUP_MASK_V2)); | |
2074 | } | |
2075 | ||
a57669d2 CD |
2076 | static bool unit_has_mask_enables_realized( |
2077 | Unit *u, | |
2078 | CGroupMask target_mask, | |
2079 | CGroupMask enable_mask) { | |
2080 | ||
2081 | assert(u); | |
2082 | ||
2083 | /* Returns true if all controllers which should be enabled are indeed enabled. | |
2084 | * | |
2085 | * Unlike unit_has_mask_realized, we don't care about the controllers that are not present, only that anything | |
2086 | * we want to add is already added. */ | |
2087 | ||
2088 | return u->cgroup_realized && | |
c72703e2 CD |
2089 | ((u->cgroup_realized_mask | target_mask) & CGROUP_MASK_V1) == (u->cgroup_realized_mask & CGROUP_MASK_V1) && |
2090 | ((u->cgroup_enabled_mask | enable_mask) & CGROUP_MASK_V2) == (u->cgroup_enabled_mask & CGROUP_MASK_V2); | |
a57669d2 CD |
2091 | } |
2092 | ||
27adcc97 | 2093 | void unit_add_to_cgroup_realize_queue(Unit *u) { |
2aa57a65 LP |
2094 | assert(u); |
2095 | ||
2096 | if (u->in_cgroup_realize_queue) | |
2097 | return; | |
2098 | ||
2099 | LIST_PREPEND(cgroup_realize_queue, u->manager->cgroup_realize_queue, u); | |
2100 | u->in_cgroup_realize_queue = true; | |
2101 | } | |
2102 | ||
2103 | static void unit_remove_from_cgroup_realize_queue(Unit *u) { | |
2104 | assert(u); | |
2105 | ||
2106 | if (!u->in_cgroup_realize_queue) | |
2107 | return; | |
2108 | ||
2109 | LIST_REMOVE(cgroup_realize_queue, u->manager->cgroup_realize_queue, u); | |
2110 | u->in_cgroup_realize_queue = false; | |
2111 | } | |
2112 | ||
a57669d2 CD |
2113 | /* Controllers can only be enabled breadth-first, from the root of the |
2114 | * hierarchy downwards to the unit in question. */ | |
2115 | static int unit_realize_cgroup_now_enable(Unit *u, ManagerState state) { | |
2116 | CGroupMask target_mask, enable_mask, new_target_mask, new_enable_mask; | |
2117 | int r; | |
2118 | ||
2119 | assert(u); | |
2120 | ||
2121 | /* First go deal with this unit's parent, or we won't be able to enable | |
2122 | * any new controllers at this layer. */ | |
2123 | if (UNIT_ISSET(u->slice)) { | |
2124 | r = unit_realize_cgroup_now_enable(UNIT_DEREF(u->slice), state); | |
2125 | if (r < 0) | |
2126 | return r; | |
2127 | } | |
2128 | ||
2129 | target_mask = unit_get_target_mask(u); | |
2130 | enable_mask = unit_get_enable_mask(u); | |
2131 | ||
2132 | /* We can only enable in this direction, don't try to disable anything. | |
2133 | */ | |
2134 | if (unit_has_mask_enables_realized(u, target_mask, enable_mask)) | |
2135 | return 0; | |
2136 | ||
2137 | new_target_mask = u->cgroup_realized_mask | target_mask; | |
2138 | new_enable_mask = u->cgroup_enabled_mask | enable_mask; | |
2139 | ||
c72703e2 | 2140 | return unit_create_cgroup(u, new_target_mask, new_enable_mask, state); |
a57669d2 CD |
2141 | } |
2142 | ||
4f6f62e4 CD |
2143 | /* Controllers can only be disabled depth-first, from the leaves of the |
2144 | * hierarchy upwards to the unit in question. */ | |
2145 | static int unit_realize_cgroup_now_disable(Unit *u, ManagerState state) { | |
2146 | Iterator i; | |
2147 | Unit *m; | |
2148 | void *v; | |
2149 | ||
2150 | assert(u); | |
2151 | ||
2152 | if (u->type != UNIT_SLICE) | |
2153 | return 0; | |
2154 | ||
2155 | HASHMAP_FOREACH_KEY(v, m, u->dependencies[UNIT_BEFORE], i) { | |
2156 | CGroupMask target_mask, enable_mask, new_target_mask, new_enable_mask; | |
2157 | int r; | |
2158 | ||
2159 | if (UNIT_DEREF(m->slice) != u) | |
2160 | continue; | |
2161 | ||
2162 | /* The cgroup for this unit might not actually be fully | |
2163 | * realised yet, in which case it isn't holding any controllers | |
2164 | * open anyway. */ | |
2165 | if (!m->cgroup_path) | |
2166 | continue; | |
2167 | ||
2168 | /* We must disable those below us first in order to release the | |
2169 | * controller. */ | |
2170 | if (m->type == UNIT_SLICE) | |
2171 | (void) unit_realize_cgroup_now_disable(m, state); | |
2172 | ||
2173 | target_mask = unit_get_target_mask(m); | |
2174 | enable_mask = unit_get_enable_mask(m); | |
2175 | ||
2176 | /* We can only disable in this direction, don't try to enable | |
2177 | * anything. */ | |
2178 | if (unit_has_mask_disables_realized(m, target_mask, enable_mask)) | |
2179 | continue; | |
2180 | ||
2181 | new_target_mask = m->cgroup_realized_mask & target_mask; | |
2182 | new_enable_mask = m->cgroup_enabled_mask & enable_mask; | |
2183 | ||
2184 | r = unit_create_cgroup(m, new_target_mask, new_enable_mask, state); | |
2185 | if (r < 0) | |
2186 | return r; | |
2187 | } | |
2188 | ||
2189 | return 0; | |
2190 | } | |
a57669d2 | 2191 | |
6414b7c9 DS |
2192 | /* Check if necessary controllers and attributes for a unit are in place. |
2193 | * | |
a57669d2 CD |
2194 | * - If so, do nothing. |
2195 | * - If not, create paths, move processes over, and set attributes. | |
2196 | * | |
2197 | * Controllers can only be *enabled* in a breadth-first way, and *disabled* in | |
2198 | * a depth-first way. As such the process looks like this: | |
2199 | * | |
2200 | * Suppose we have a cgroup hierarchy which looks like this: | |
2201 | * | |
2202 | * root | |
2203 | * / \ | |
2204 | * / \ | |
2205 | * / \ | |
2206 | * a b | |
2207 | * / \ / \ | |
2208 | * / \ / \ | |
2209 | * c d e f | |
2210 | * / \ / \ / \ / \ | |
2211 | * h i j k l m n o | |
2212 | * | |
2213 | * 1. We want to realise cgroup "d" now. | |
c72703e2 | 2214 | * 2. cgroup "a" has DisableControllers=cpu in the associated unit. |
a57669d2 CD |
2215 | * 3. cgroup "k" just started requesting the memory controller. |
2216 | * | |
2217 | * To make this work we must do the following in order: | |
2218 | * | |
2219 | * 1. Disable CPU controller in k, j | |
2220 | * 2. Disable CPU controller in d | |
2221 | * 3. Enable memory controller in root | |
2222 | * 4. Enable memory controller in a | |
2223 | * 5. Enable memory controller in d | |
2224 | * 6. Enable memory controller in k | |
2225 | * | |
2226 | * Notice that we need to touch j in one direction, but not the other. We also | |
2227 | * don't go beyond d when disabling -- it's up to "a" to get realized if it | |
2228 | * wants to disable further. The basic rules are therefore: | |
2229 | * | |
2230 | * - If you're disabling something, you need to realise all of the cgroups from | |
2231 | * your recursive descendants to the root. This starts from the leaves. | |
2232 | * - If you're enabling something, you need to realise from the root cgroup | |
2233 | * downwards, but you don't need to iterate your recursive descendants. | |
6414b7c9 DS |
2234 | * |
2235 | * Returns 0 on success and < 0 on failure. */ | |
db785129 | 2236 | static int unit_realize_cgroup_now(Unit *u, ManagerState state) { |
efdb0237 | 2237 | CGroupMask target_mask, enable_mask; |
6414b7c9 | 2238 | int r; |
64747e2d | 2239 | |
4ad49000 | 2240 | assert(u); |
64747e2d | 2241 | |
2aa57a65 | 2242 | unit_remove_from_cgroup_realize_queue(u); |
64747e2d | 2243 | |
efdb0237 | 2244 | target_mask = unit_get_target_mask(u); |
ccf78df1 TH |
2245 | enable_mask = unit_get_enable_mask(u); |
2246 | ||
17f14955 | 2247 | if (unit_has_mask_realized(u, target_mask, enable_mask)) |
0a1eb06d | 2248 | return 0; |
64747e2d | 2249 | |
4f6f62e4 CD |
2250 | /* Disable controllers below us, if there are any */ |
2251 | r = unit_realize_cgroup_now_disable(u, state); | |
2252 | if (r < 0) | |
2253 | return r; | |
2254 | ||
2255 | /* Enable controllers above us, if there are any */ | |
6414b7c9 | 2256 | if (UNIT_ISSET(u->slice)) { |
a57669d2 | 2257 | r = unit_realize_cgroup_now_enable(UNIT_DEREF(u->slice), state); |
6414b7c9 DS |
2258 | if (r < 0) |
2259 | return r; | |
2260 | } | |
4ad49000 | 2261 | |
0d2d6fbf CD |
2262 | /* Now actually deal with the cgroup we were trying to realise and set attributes */ |
2263 | r = unit_create_cgroup(u, target_mask, enable_mask, state); | |
6414b7c9 DS |
2264 | if (r < 0) |
2265 | return r; | |
2266 | ||
c2baf11c LP |
2267 | /* Now, reset the invalidation mask */ |
2268 | u->cgroup_invalidated_mask = 0; | |
6414b7c9 | 2269 | return 0; |
64747e2d LP |
2270 | } |
2271 | ||
91a6073e | 2272 | unsigned manager_dispatch_cgroup_realize_queue(Manager *m) { |
db785129 | 2273 | ManagerState state; |
4ad49000 | 2274 | unsigned n = 0; |
db785129 | 2275 | Unit *i; |
6414b7c9 | 2276 | int r; |
ecedd90f | 2277 | |
91a6073e LP |
2278 | assert(m); |
2279 | ||
db785129 LP |
2280 | state = manager_state(m); |
2281 | ||
91a6073e LP |
2282 | while ((i = m->cgroup_realize_queue)) { |
2283 | assert(i->in_cgroup_realize_queue); | |
ecedd90f | 2284 | |
2aa57a65 LP |
2285 | if (UNIT_IS_INACTIVE_OR_FAILED(unit_active_state(i))) { |
2286 | /* Maybe things changed, and the unit is not actually active anymore? */ | |
2287 | unit_remove_from_cgroup_realize_queue(i); | |
2288 | continue; | |
2289 | } | |
2290 | ||
db785129 | 2291 | r = unit_realize_cgroup_now(i, state); |
6414b7c9 | 2292 | if (r < 0) |
efdb0237 | 2293 | log_warning_errno(r, "Failed to realize cgroups for queued unit %s, ignoring: %m", i->id); |
0a1eb06d | 2294 | |
4ad49000 LP |
2295 | n++; |
2296 | } | |
ecedd90f | 2297 | |
4ad49000 | 2298 | return n; |
8e274523 LP |
2299 | } |
2300 | ||
91a6073e | 2301 | static void unit_add_siblings_to_cgroup_realize_queue(Unit *u) { |
4ad49000 | 2302 | Unit *slice; |
ca949c9d | 2303 | |
4ad49000 LP |
2304 | /* This adds the siblings of the specified unit and the |
2305 | * siblings of all parent units to the cgroup queue. (But | |
2306 | * neither the specified unit itself nor the parents.) */ | |
2307 | ||
2308 | while ((slice = UNIT_DEREF(u->slice))) { | |
2309 | Iterator i; | |
2310 | Unit *m; | |
eef85c4a | 2311 | void *v; |
8f53a7b8 | 2312 | |
eef85c4a | 2313 | HASHMAP_FOREACH_KEY(v, m, u->dependencies[UNIT_BEFORE], i) { |
6414b7c9 DS |
2314 | /* Skip units that have a dependency on the slice |
2315 | * but aren't actually in it. */ | |
4ad49000 | 2316 | if (UNIT_DEREF(m->slice) != slice) |
50159e6a | 2317 | continue; |
8e274523 | 2318 | |
6414b7c9 DS |
2319 | /* No point in doing cgroup application for units |
2320 | * without active processes. */ | |
2321 | if (UNIT_IS_INACTIVE_OR_FAILED(unit_active_state(m))) | |
2322 | continue; | |
2323 | ||
2324 | /* If the unit doesn't need any new controllers | |
2325 | * and has current ones realized, it doesn't need | |
2326 | * any changes. */ | |
906c06f6 DM |
2327 | if (unit_has_mask_realized(m, |
2328 | unit_get_target_mask(m), | |
17f14955 | 2329 | unit_get_enable_mask(m))) |
6414b7c9 DS |
2330 | continue; |
2331 | ||
91a6073e | 2332 | unit_add_to_cgroup_realize_queue(m); |
50159e6a LP |
2333 | } |
2334 | ||
4ad49000 | 2335 | u = slice; |
8e274523 | 2336 | } |
4ad49000 LP |
2337 | } |
2338 | ||
0a1eb06d | 2339 | int unit_realize_cgroup(Unit *u) { |
4ad49000 LP |
2340 | assert(u); |
2341 | ||
35b7ff80 | 2342 | if (!UNIT_HAS_CGROUP_CONTEXT(u)) |
0a1eb06d | 2343 | return 0; |
8e274523 | 2344 | |
4ad49000 LP |
2345 | /* So, here's the deal: when realizing the cgroups for this |
2346 | * unit, we need to first create all parents, but there's more | |
2347 | * actually: for the weight-based controllers we also need to | |
2348 | * make sure that all our siblings (i.e. units that are in the | |
73e231ab | 2349 | * same slice as we are) have cgroups, too. Otherwise, things |
4ad49000 LP |
2350 | * would become very uneven as each of their processes would |
2351 | * get as much resources as all our group together. This call | |
2352 | * will synchronously create the parent cgroups, but will | |
2353 | * defer work on the siblings to the next event loop | |
2354 | * iteration. */ | |
ca949c9d | 2355 | |
4ad49000 | 2356 | /* Add all sibling slices to the cgroup queue. */ |
91a6073e | 2357 | unit_add_siblings_to_cgroup_realize_queue(u); |
4ad49000 | 2358 | |
6414b7c9 | 2359 | /* And realize this one now (and apply the values) */ |
db785129 | 2360 | return unit_realize_cgroup_now(u, manager_state(u->manager)); |
8e274523 LP |
2361 | } |
2362 | ||
efdb0237 LP |
2363 | void unit_release_cgroup(Unit *u) { |
2364 | assert(u); | |
2365 | ||
8a0d5388 LP |
2366 | /* Forgets all cgroup details for this cgroup — but does *not* destroy the cgroup. This is hence OK to call |
2367 | * when we close down everything for reexecution, where we really want to leave the cgroup in place. */ | |
efdb0237 LP |
2368 | |
2369 | if (u->cgroup_path) { | |
2370 | (void) hashmap_remove(u->manager->cgroup_unit, u->cgroup_path); | |
2371 | u->cgroup_path = mfree(u->cgroup_path); | |
2372 | } | |
2373 | ||
0bb814c2 LP |
2374 | if (u->cgroup_control_inotify_wd >= 0) { |
2375 | if (inotify_rm_watch(u->manager->cgroup_inotify_fd, u->cgroup_control_inotify_wd) < 0) | |
2376 | 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 | 2377 | |
0bb814c2 LP |
2378 | (void) hashmap_remove(u->manager->cgroup_control_inotify_wd_unit, INT_TO_PTR(u->cgroup_control_inotify_wd)); |
2379 | u->cgroup_control_inotify_wd = -1; | |
efdb0237 | 2380 | } |
afcfaa69 LP |
2381 | |
2382 | if (u->cgroup_memory_inotify_wd >= 0) { | |
2383 | if (inotify_rm_watch(u->manager->cgroup_inotify_fd, u->cgroup_memory_inotify_wd) < 0) | |
2384 | 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); | |
2385 | ||
2386 | (void) hashmap_remove(u->manager->cgroup_memory_inotify_wd_unit, INT_TO_PTR(u->cgroup_memory_inotify_wd)); | |
2387 | u->cgroup_memory_inotify_wd = -1; | |
2388 | } | |
efdb0237 LP |
2389 | } |
2390 | ||
2391 | void unit_prune_cgroup(Unit *u) { | |
8e274523 | 2392 | int r; |
efdb0237 | 2393 | bool is_root_slice; |
8e274523 | 2394 | |
4ad49000 | 2395 | assert(u); |
8e274523 | 2396 | |
efdb0237 LP |
2397 | /* Removes the cgroup, if empty and possible, and stops watching it. */ |
2398 | ||
4ad49000 LP |
2399 | if (!u->cgroup_path) |
2400 | return; | |
8e274523 | 2401 | |
fe700f46 LP |
2402 | (void) unit_get_cpu_usage(u, NULL); /* Cache the last CPU usage value before we destroy the cgroup */ |
2403 | ||
efdb0237 LP |
2404 | is_root_slice = unit_has_name(u, SPECIAL_ROOT_SLICE); |
2405 | ||
2406 | r = cg_trim_everywhere(u->manager->cgroup_supported, u->cgroup_path, !is_root_slice); | |
0219b352 DB |
2407 | if (r < 0) |
2408 | /* One reason we could have failed here is, that the cgroup still contains a process. | |
2409 | * However, if the cgroup becomes removable at a later time, it might be removed when | |
2410 | * the containing slice is stopped. So even if we failed now, this unit shouldn't assume | |
2411 | * that the cgroup is still realized the next time it is started. Do not return early | |
2412 | * on error, continue cleanup. */ | |
2413 | log_unit_full(u, r == -EBUSY ? LOG_DEBUG : LOG_WARNING, r, "Failed to destroy cgroup %s, ignoring: %m", u->cgroup_path); | |
8e274523 | 2414 | |
efdb0237 LP |
2415 | if (is_root_slice) |
2416 | return; | |
2417 | ||
2418 | unit_release_cgroup(u); | |
0a1eb06d | 2419 | |
4ad49000 | 2420 | u->cgroup_realized = false; |
bc432dc7 | 2421 | u->cgroup_realized_mask = 0; |
ccf78df1 | 2422 | u->cgroup_enabled_mask = 0; |
084c7007 RG |
2423 | |
2424 | u->bpf_device_control_installed = bpf_program_unref(u->bpf_device_control_installed); | |
8e274523 LP |
2425 | } |
2426 | ||
efdb0237 | 2427 | int unit_search_main_pid(Unit *u, pid_t *ret) { |
4ad49000 | 2428 | _cleanup_fclose_ FILE *f = NULL; |
4d051546 | 2429 | pid_t pid = 0, npid; |
efdb0237 | 2430 | int r; |
4ad49000 LP |
2431 | |
2432 | assert(u); | |
efdb0237 | 2433 | assert(ret); |
4ad49000 LP |
2434 | |
2435 | if (!u->cgroup_path) | |
efdb0237 | 2436 | return -ENXIO; |
4ad49000 | 2437 | |
efdb0237 LP |
2438 | r = cg_enumerate_processes(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path, &f); |
2439 | if (r < 0) | |
2440 | return r; | |
4ad49000 | 2441 | |
4ad49000 | 2442 | while (cg_read_pid(f, &npid) > 0) { |
4ad49000 LP |
2443 | |
2444 | if (npid == pid) | |
2445 | continue; | |
8e274523 | 2446 | |
4d051546 | 2447 | if (pid_is_my_child(npid) == 0) |
4ad49000 | 2448 | continue; |
8e274523 | 2449 | |
efdb0237 | 2450 | if (pid != 0) |
4ad49000 LP |
2451 | /* Dang, there's more than one daemonized PID |
2452 | in this group, so we don't know what process | |
2453 | is the main process. */ | |
efdb0237 LP |
2454 | |
2455 | return -ENODATA; | |
8e274523 | 2456 | |
4ad49000 | 2457 | pid = npid; |
8e274523 LP |
2458 | } |
2459 | ||
efdb0237 LP |
2460 | *ret = pid; |
2461 | return 0; | |
2462 | } | |
2463 | ||
2464 | static int unit_watch_pids_in_path(Unit *u, const char *path) { | |
b3c5bad3 | 2465 | _cleanup_closedir_ DIR *d = NULL; |
efdb0237 LP |
2466 | _cleanup_fclose_ FILE *f = NULL; |
2467 | int ret = 0, r; | |
2468 | ||
2469 | assert(u); | |
2470 | assert(path); | |
2471 | ||
2472 | r = cg_enumerate_processes(SYSTEMD_CGROUP_CONTROLLER, path, &f); | |
2473 | if (r < 0) | |
2474 | ret = r; | |
2475 | else { | |
2476 | pid_t pid; | |
2477 | ||
2478 | while ((r = cg_read_pid(f, &pid)) > 0) { | |
f75f613d | 2479 | r = unit_watch_pid(u, pid, false); |
efdb0237 LP |
2480 | if (r < 0 && ret >= 0) |
2481 | ret = r; | |
2482 | } | |
2483 | ||
2484 | if (r < 0 && ret >= 0) | |
2485 | ret = r; | |
2486 | } | |
2487 | ||
2488 | r = cg_enumerate_subgroups(SYSTEMD_CGROUP_CONTROLLER, path, &d); | |
2489 | if (r < 0) { | |
2490 | if (ret >= 0) | |
2491 | ret = r; | |
2492 | } else { | |
2493 | char *fn; | |
2494 | ||
2495 | while ((r = cg_read_subgroup(d, &fn)) > 0) { | |
2496 | _cleanup_free_ char *p = NULL; | |
2497 | ||
95b21cff | 2498 | p = path_join(empty_to_root(path), fn); |
efdb0237 LP |
2499 | free(fn); |
2500 | ||
2501 | if (!p) | |
2502 | return -ENOMEM; | |
2503 | ||
2504 | r = unit_watch_pids_in_path(u, p); | |
2505 | if (r < 0 && ret >= 0) | |
2506 | ret = r; | |
2507 | } | |
2508 | ||
2509 | if (r < 0 && ret >= 0) | |
2510 | ret = r; | |
2511 | } | |
2512 | ||
2513 | return ret; | |
2514 | } | |
2515 | ||
11aef522 LP |
2516 | int unit_synthesize_cgroup_empty_event(Unit *u) { |
2517 | int r; | |
2518 | ||
2519 | assert(u); | |
2520 | ||
2521 | /* Enqueue a synthetic cgroup empty event if this unit doesn't watch any PIDs anymore. This is compatibility | |
2522 | * support for non-unified systems where notifications aren't reliable, and hence need to take whatever we can | |
2523 | * get as notification source as soon as we stopped having any useful PIDs to watch for. */ | |
2524 | ||
2525 | if (!u->cgroup_path) | |
2526 | return -ENOENT; | |
2527 | ||
2528 | r = cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER); | |
2529 | if (r < 0) | |
2530 | return r; | |
2531 | if (r > 0) /* On unified we have reliable notifications, and don't need this */ | |
2532 | return 0; | |
2533 | ||
2534 | if (!set_isempty(u->pids)) | |
2535 | return 0; | |
2536 | ||
2537 | unit_add_to_cgroup_empty_queue(u); | |
2538 | return 0; | |
2539 | } | |
2540 | ||
efdb0237 | 2541 | int unit_watch_all_pids(Unit *u) { |
b4cccbc1 LP |
2542 | int r; |
2543 | ||
efdb0237 LP |
2544 | assert(u); |
2545 | ||
2546 | /* Adds all PIDs from our cgroup to the set of PIDs we | |
2547 | * watch. This is a fallback logic for cases where we do not | |
2548 | * get reliable cgroup empty notifications: we try to use | |
2549 | * SIGCHLD as replacement. */ | |
2550 | ||
2551 | if (!u->cgroup_path) | |
2552 | return -ENOENT; | |
2553 | ||
c22800e4 | 2554 | r = cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER); |
b4cccbc1 LP |
2555 | if (r < 0) |
2556 | return r; | |
2557 | if (r > 0) /* On unified we can use proper notifications */ | |
efdb0237 LP |
2558 | return 0; |
2559 | ||
2560 | return unit_watch_pids_in_path(u, u->cgroup_path); | |
2561 | } | |
2562 | ||
09e24654 LP |
2563 | static int on_cgroup_empty_event(sd_event_source *s, void *userdata) { |
2564 | Manager *m = userdata; | |
2565 | Unit *u; | |
efdb0237 LP |
2566 | int r; |
2567 | ||
09e24654 LP |
2568 | assert(s); |
2569 | assert(m); | |
efdb0237 | 2570 | |
09e24654 LP |
2571 | u = m->cgroup_empty_queue; |
2572 | if (!u) | |
efdb0237 LP |
2573 | return 0; |
2574 | ||
09e24654 LP |
2575 | assert(u->in_cgroup_empty_queue); |
2576 | u->in_cgroup_empty_queue = false; | |
2577 | LIST_REMOVE(cgroup_empty_queue, m->cgroup_empty_queue, u); | |
2578 | ||
2579 | if (m->cgroup_empty_queue) { | |
2580 | /* More stuff queued, let's make sure we remain enabled */ | |
2581 | r = sd_event_source_set_enabled(s, SD_EVENT_ONESHOT); | |
2582 | if (r < 0) | |
19a691a9 | 2583 | log_debug_errno(r, "Failed to reenable cgroup empty event source, ignoring: %m"); |
09e24654 | 2584 | } |
efdb0237 LP |
2585 | |
2586 | unit_add_to_gc_queue(u); | |
2587 | ||
2588 | if (UNIT_VTABLE(u)->notify_cgroup_empty) | |
2589 | UNIT_VTABLE(u)->notify_cgroup_empty(u); | |
2590 | ||
2591 | return 0; | |
2592 | } | |
2593 | ||
09e24654 LP |
2594 | void unit_add_to_cgroup_empty_queue(Unit *u) { |
2595 | int r; | |
2596 | ||
2597 | assert(u); | |
2598 | ||
2599 | /* Note that there are four different ways how cgroup empty events reach us: | |
2600 | * | |
2601 | * 1. On the unified hierarchy we get an inotify event on the cgroup | |
2602 | * | |
2603 | * 2. On the legacy hierarchy, when running in system mode, we get a datagram on the cgroup agent socket | |
2604 | * | |
2605 | * 3. On the legacy hierarchy, when running in user mode, we get a D-Bus signal on the system bus | |
2606 | * | |
2607 | * 4. On the legacy hierarchy, in service units we start watching all processes of the cgroup for SIGCHLD as | |
2608 | * soon as we get one SIGCHLD, to deal with unreliable cgroup notifications. | |
2609 | * | |
2610 | * Regardless which way we got the notification, we'll verify it here, and then add it to a separate | |
2611 | * queue. This queue will be dispatched at a lower priority than the SIGCHLD handler, so that we always use | |
2612 | * SIGCHLD if we can get it first, and only use the cgroup empty notifications if there's no SIGCHLD pending | |
2613 | * (which might happen if the cgroup doesn't contain processes that are our own child, which is typically the | |
2614 | * case for scope units). */ | |
2615 | ||
2616 | if (u->in_cgroup_empty_queue) | |
2617 | return; | |
2618 | ||
2619 | /* Let's verify that the cgroup is really empty */ | |
2620 | if (!u->cgroup_path) | |
2621 | return; | |
2622 | r = cg_is_empty_recursive(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path); | |
2623 | if (r < 0) { | |
2624 | log_unit_debug_errno(u, r, "Failed to determine whether cgroup %s is empty: %m", u->cgroup_path); | |
2625 | return; | |
2626 | } | |
2627 | if (r == 0) | |
2628 | return; | |
2629 | ||
2630 | LIST_PREPEND(cgroup_empty_queue, u->manager->cgroup_empty_queue, u); | |
2631 | u->in_cgroup_empty_queue = true; | |
2632 | ||
2633 | /* Trigger the defer event */ | |
2634 | r = sd_event_source_set_enabled(u->manager->cgroup_empty_event_source, SD_EVENT_ONESHOT); | |
2635 | if (r < 0) | |
2636 | log_debug_errno(r, "Failed to enable cgroup empty event source: %m"); | |
2637 | } | |
2638 | ||
2ba6ae6b | 2639 | int unit_check_oom(Unit *u) { |
afcfaa69 LP |
2640 | _cleanup_free_ char *oom_kill = NULL; |
2641 | bool increased; | |
2642 | uint64_t c; | |
2643 | int r; | |
2644 | ||
2645 | if (!u->cgroup_path) | |
2646 | return 0; | |
2647 | ||
2648 | r = cg_get_keyed_attribute("memory", u->cgroup_path, "memory.events", STRV_MAKE("oom_kill"), &oom_kill); | |
2649 | if (r < 0) | |
2650 | return log_unit_debug_errno(u, r, "Failed to read oom_kill field of memory.events cgroup attribute: %m"); | |
2651 | ||
2652 | r = safe_atou64(oom_kill, &c); | |
2653 | if (r < 0) | |
2654 | return log_unit_debug_errno(u, r, "Failed to parse oom_kill field: %m"); | |
2655 | ||
2656 | increased = c > u->oom_kill_last; | |
2657 | u->oom_kill_last = c; | |
2658 | ||
2659 | if (!increased) | |
2660 | return 0; | |
2661 | ||
2662 | log_struct(LOG_NOTICE, | |
2663 | "MESSAGE_ID=" SD_MESSAGE_UNIT_OUT_OF_MEMORY_STR, | |
2664 | LOG_UNIT_ID(u), | |
2665 | LOG_UNIT_INVOCATION_ID(u), | |
2666 | LOG_UNIT_MESSAGE(u, "A process of this unit has been killed by the OOM killer.")); | |
2667 | ||
2668 | if (UNIT_VTABLE(u)->notify_cgroup_oom) | |
2669 | UNIT_VTABLE(u)->notify_cgroup_oom(u); | |
2670 | ||
2671 | return 1; | |
2672 | } | |
2673 | ||
2674 | static int on_cgroup_oom_event(sd_event_source *s, void *userdata) { | |
2675 | Manager *m = userdata; | |
2676 | Unit *u; | |
2677 | int r; | |
2678 | ||
2679 | assert(s); | |
2680 | assert(m); | |
2681 | ||
2682 | u = m->cgroup_oom_queue; | |
2683 | if (!u) | |
2684 | return 0; | |
2685 | ||
2686 | assert(u->in_cgroup_oom_queue); | |
2687 | u->in_cgroup_oom_queue = false; | |
2688 | LIST_REMOVE(cgroup_oom_queue, m->cgroup_oom_queue, u); | |
2689 | ||
2690 | if (m->cgroup_oom_queue) { | |
2691 | /* More stuff queued, let's make sure we remain enabled */ | |
2692 | r = sd_event_source_set_enabled(s, SD_EVENT_ONESHOT); | |
2693 | if (r < 0) | |
2694 | log_debug_errno(r, "Failed to reenable cgroup oom event source, ignoring: %m"); | |
2695 | } | |
2696 | ||
2697 | (void) unit_check_oom(u); | |
2698 | return 0; | |
2699 | } | |
2700 | ||
2701 | static void unit_add_to_cgroup_oom_queue(Unit *u) { | |
2702 | int r; | |
2703 | ||
2704 | assert(u); | |
2705 | ||
2706 | if (u->in_cgroup_oom_queue) | |
2707 | return; | |
2708 | if (!u->cgroup_path) | |
2709 | return; | |
2710 | ||
2711 | LIST_PREPEND(cgroup_oom_queue, u->manager->cgroup_oom_queue, u); | |
2712 | u->in_cgroup_oom_queue = true; | |
2713 | ||
2714 | /* Trigger the defer event */ | |
2715 | if (!u->manager->cgroup_oom_event_source) { | |
2716 | _cleanup_(sd_event_source_unrefp) sd_event_source *s = NULL; | |
2717 | ||
2718 | r = sd_event_add_defer(u->manager->event, &s, on_cgroup_oom_event, u->manager); | |
2719 | if (r < 0) { | |
2720 | log_error_errno(r, "Failed to create cgroup oom event source: %m"); | |
2721 | return; | |
2722 | } | |
2723 | ||
2724 | r = sd_event_source_set_priority(s, SD_EVENT_PRIORITY_NORMAL-8); | |
2725 | if (r < 0) { | |
2726 | log_error_errno(r, "Failed to set priority of cgroup oom event source: %m"); | |
2727 | return; | |
2728 | } | |
2729 | ||
2730 | (void) sd_event_source_set_description(s, "cgroup-oom"); | |
2731 | u->manager->cgroup_oom_event_source = TAKE_PTR(s); | |
2732 | } | |
2733 | ||
2734 | r = sd_event_source_set_enabled(u->manager->cgroup_oom_event_source, SD_EVENT_ONESHOT); | |
2735 | if (r < 0) | |
2736 | log_error_errno(r, "Failed to enable cgroup oom event source: %m"); | |
2737 | } | |
2738 | ||
efdb0237 LP |
2739 | static int on_cgroup_inotify_event(sd_event_source *s, int fd, uint32_t revents, void *userdata) { |
2740 | Manager *m = userdata; | |
2741 | ||
2742 | assert(s); | |
2743 | assert(fd >= 0); | |
2744 | assert(m); | |
2745 | ||
2746 | for (;;) { | |
2747 | union inotify_event_buffer buffer; | |
2748 | struct inotify_event *e; | |
2749 | ssize_t l; | |
2750 | ||
2751 | l = read(fd, &buffer, sizeof(buffer)); | |
2752 | if (l < 0) { | |
47249640 | 2753 | if (IN_SET(errno, EINTR, EAGAIN)) |
efdb0237 LP |
2754 | return 0; |
2755 | ||
2756 | return log_error_errno(errno, "Failed to read control group inotify events: %m"); | |
2757 | } | |
2758 | ||
2759 | FOREACH_INOTIFY_EVENT(e, buffer, l) { | |
2760 | Unit *u; | |
2761 | ||
2762 | if (e->wd < 0) | |
2763 | /* Queue overflow has no watch descriptor */ | |
2764 | continue; | |
2765 | ||
2766 | if (e->mask & IN_IGNORED) | |
2767 | /* The watch was just removed */ | |
2768 | continue; | |
2769 | ||
afcfaa69 LP |
2770 | /* Note that inotify might deliver events for a watch even after it was removed, |
2771 | * because it was queued before the removal. Let's ignore this here safely. */ | |
2772 | ||
0bb814c2 | 2773 | u = hashmap_get(m->cgroup_control_inotify_wd_unit, INT_TO_PTR(e->wd)); |
afcfaa69 LP |
2774 | if (u) |
2775 | unit_add_to_cgroup_empty_queue(u); | |
efdb0237 | 2776 | |
afcfaa69 LP |
2777 | u = hashmap_get(m->cgroup_memory_inotify_wd_unit, INT_TO_PTR(e->wd)); |
2778 | if (u) | |
2779 | unit_add_to_cgroup_oom_queue(u); | |
efdb0237 LP |
2780 | } |
2781 | } | |
8e274523 LP |
2782 | } |
2783 | ||
17f14955 RG |
2784 | static int cg_bpf_mask_supported(CGroupMask *ret) { |
2785 | CGroupMask mask = 0; | |
2786 | int r; | |
2787 | ||
2788 | /* BPF-based firewall */ | |
2789 | r = bpf_firewall_supported(); | |
2790 | if (r > 0) | |
2791 | mask |= CGROUP_MASK_BPF_FIREWALL; | |
2792 | ||
084c7007 RG |
2793 | /* BPF-based device access control */ |
2794 | r = bpf_devices_supported(); | |
2795 | if (r > 0) | |
2796 | mask |= CGROUP_MASK_BPF_DEVICES; | |
2797 | ||
17f14955 RG |
2798 | *ret = mask; |
2799 | return 0; | |
2800 | } | |
2801 | ||
8e274523 | 2802 | int manager_setup_cgroup(Manager *m) { |
9444b1f2 | 2803 | _cleanup_free_ char *path = NULL; |
10bd3e2e | 2804 | const char *scope_path; |
efdb0237 | 2805 | CGroupController c; |
b4cccbc1 | 2806 | int r, all_unified; |
17f14955 | 2807 | CGroupMask mask; |
efdb0237 | 2808 | char *e; |
8e274523 LP |
2809 | |
2810 | assert(m); | |
2811 | ||
35d2e7ec | 2812 | /* 1. Determine hierarchy */ |
efdb0237 | 2813 | m->cgroup_root = mfree(m->cgroup_root); |
9444b1f2 | 2814 | r = cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER, 0, &m->cgroup_root); |
23bbb0de MS |
2815 | if (r < 0) |
2816 | return log_error_errno(r, "Cannot determine cgroup we are running in: %m"); | |
8e274523 | 2817 | |
efdb0237 LP |
2818 | /* Chop off the init scope, if we are already located in it */ |
2819 | e = endswith(m->cgroup_root, "/" SPECIAL_INIT_SCOPE); | |
0d8c31ff | 2820 | |
efdb0237 LP |
2821 | /* LEGACY: Also chop off the system slice if we are in |
2822 | * it. This is to support live upgrades from older systemd | |
2823 | * versions where PID 1 was moved there. Also see | |
2824 | * cg_get_root_path(). */ | |
463d0d15 | 2825 | if (!e && MANAGER_IS_SYSTEM(m)) { |
9444b1f2 | 2826 | e = endswith(m->cgroup_root, "/" SPECIAL_SYSTEM_SLICE); |
15c60e99 | 2827 | if (!e) |
efdb0237 | 2828 | e = endswith(m->cgroup_root, "/system"); /* even more legacy */ |
0baf24dd | 2829 | } |
efdb0237 LP |
2830 | if (e) |
2831 | *e = 0; | |
7ccfb64a | 2832 | |
7546145e LP |
2833 | /* And make sure to store away the root value without trailing slash, even for the root dir, so that we can |
2834 | * easily prepend it everywhere. */ | |
2835 | delete_trailing_chars(m->cgroup_root, "/"); | |
8e274523 | 2836 | |
35d2e7ec | 2837 | /* 2. Show data */ |
9444b1f2 | 2838 | r = cg_get_path(SYSTEMD_CGROUP_CONTROLLER, m->cgroup_root, NULL, &path); |
23bbb0de MS |
2839 | if (r < 0) |
2840 | return log_error_errno(r, "Cannot find cgroup mount point: %m"); | |
8e274523 | 2841 | |
d4d99bc6 | 2842 | r = cg_unified(); |
415fc41c TH |
2843 | if (r < 0) |
2844 | return log_error_errno(r, "Couldn't determine if we are running in the unified hierarchy: %m"); | |
5da38d07 | 2845 | |
b4cccbc1 | 2846 | all_unified = cg_all_unified(); |
d4c819ed ZJS |
2847 | if (all_unified < 0) |
2848 | return log_error_errno(all_unified, "Couldn't determine whether we are in all unified mode: %m"); | |
2849 | if (all_unified > 0) | |
efdb0237 | 2850 | log_debug("Unified cgroup hierarchy is located at %s.", path); |
b4cccbc1 | 2851 | else { |
c22800e4 | 2852 | r = cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER); |
b4cccbc1 LP |
2853 | if (r < 0) |
2854 | return log_error_errno(r, "Failed to determine whether systemd's own controller is in unified mode: %m"); | |
2855 | if (r > 0) | |
2856 | log_debug("Unified cgroup hierarchy is located at %s. Controllers are on legacy hierarchies.", path); | |
2857 | else | |
2858 | log_debug("Using cgroup controller " SYSTEMD_CGROUP_CONTROLLER_LEGACY ". File system hierarchy is at %s.", path); | |
2859 | } | |
efdb0237 | 2860 | |
09e24654 LP |
2861 | /* 3. Allocate cgroup empty defer event source */ |
2862 | m->cgroup_empty_event_source = sd_event_source_unref(m->cgroup_empty_event_source); | |
2863 | r = sd_event_add_defer(m->event, &m->cgroup_empty_event_source, on_cgroup_empty_event, m); | |
2864 | if (r < 0) | |
2865 | return log_error_errno(r, "Failed to create cgroup empty event source: %m"); | |
2866 | ||
cbe83389 LP |
2867 | /* Schedule cgroup empty checks early, but after having processed service notification messages or |
2868 | * SIGCHLD signals, so that a cgroup running empty is always just the last safety net of | |
2869 | * notification, and we collected the metadata the notification and SIGCHLD stuff offers first. */ | |
09e24654 LP |
2870 | r = sd_event_source_set_priority(m->cgroup_empty_event_source, SD_EVENT_PRIORITY_NORMAL-5); |
2871 | if (r < 0) | |
2872 | return log_error_errno(r, "Failed to set priority of cgroup empty event source: %m"); | |
2873 | ||
2874 | r = sd_event_source_set_enabled(m->cgroup_empty_event_source, SD_EVENT_OFF); | |
2875 | if (r < 0) | |
2876 | return log_error_errno(r, "Failed to disable cgroup empty event source: %m"); | |
2877 | ||
2878 | (void) sd_event_source_set_description(m->cgroup_empty_event_source, "cgroup-empty"); | |
2879 | ||
2880 | /* 4. Install notifier inotify object, or agent */ | |
10bd3e2e | 2881 | if (cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER) > 0) { |
c6c18be3 | 2882 | |
09e24654 | 2883 | /* In the unified hierarchy we can get cgroup empty notifications via inotify. */ |
efdb0237 | 2884 | |
10bd3e2e LP |
2885 | m->cgroup_inotify_event_source = sd_event_source_unref(m->cgroup_inotify_event_source); |
2886 | safe_close(m->cgroup_inotify_fd); | |
efdb0237 | 2887 | |
10bd3e2e LP |
2888 | m->cgroup_inotify_fd = inotify_init1(IN_NONBLOCK|IN_CLOEXEC); |
2889 | if (m->cgroup_inotify_fd < 0) | |
2890 | return log_error_errno(errno, "Failed to create control group inotify object: %m"); | |
efdb0237 | 2891 | |
10bd3e2e LP |
2892 | r = sd_event_add_io(m->event, &m->cgroup_inotify_event_source, m->cgroup_inotify_fd, EPOLLIN, on_cgroup_inotify_event, m); |
2893 | if (r < 0) | |
2894 | return log_error_errno(r, "Failed to watch control group inotify object: %m"); | |
efdb0237 | 2895 | |
cbe83389 LP |
2896 | /* Process cgroup empty notifications early. Note that when this event is dispatched it'll |
2897 | * just add the unit to a cgroup empty queue, hence let's run earlier than that. Also see | |
2898 | * handling of cgroup agent notifications, for the classic cgroup hierarchy support. */ | |
2899 | r = sd_event_source_set_priority(m->cgroup_inotify_event_source, SD_EVENT_PRIORITY_NORMAL-9); | |
10bd3e2e LP |
2900 | if (r < 0) |
2901 | return log_error_errno(r, "Failed to set priority of inotify event source: %m"); | |
efdb0237 | 2902 | |
10bd3e2e | 2903 | (void) sd_event_source_set_description(m->cgroup_inotify_event_source, "cgroup-inotify"); |
efdb0237 | 2904 | |
611c4f8a | 2905 | } else if (MANAGER_IS_SYSTEM(m) && manager_owns_host_root_cgroup(m) && !MANAGER_IS_TEST_RUN(m)) { |
efdb0237 | 2906 | |
10bd3e2e LP |
2907 | /* On the legacy hierarchy we only get notifications via cgroup agents. (Which isn't really reliable, |
2908 | * since it does not generate events when control groups with children run empty. */ | |
8e274523 | 2909 | |
10bd3e2e | 2910 | r = cg_install_release_agent(SYSTEMD_CGROUP_CONTROLLER, SYSTEMD_CGROUP_AGENT_PATH); |
23bbb0de | 2911 | if (r < 0) |
10bd3e2e LP |
2912 | log_warning_errno(r, "Failed to install release agent, ignoring: %m"); |
2913 | else if (r > 0) | |
2914 | log_debug("Installed release agent."); | |
2915 | else if (r == 0) | |
2916 | log_debug("Release agent already installed."); | |
2917 | } | |
efdb0237 | 2918 | |
09e24654 | 2919 | /* 5. Make sure we are in the special "init.scope" unit in the root slice. */ |
10bd3e2e LP |
2920 | scope_path = strjoina(m->cgroup_root, "/" SPECIAL_INIT_SCOPE); |
2921 | r = cg_create_and_attach(SYSTEMD_CGROUP_CONTROLLER, scope_path, 0); | |
aa77e234 MS |
2922 | if (r >= 0) { |
2923 | /* Also, move all other userspace processes remaining in the root cgroup into that scope. */ | |
2924 | r = cg_migrate(SYSTEMD_CGROUP_CONTROLLER, m->cgroup_root, SYSTEMD_CGROUP_CONTROLLER, scope_path, 0); | |
2925 | if (r < 0) | |
2926 | log_warning_errno(r, "Couldn't move remaining userspace processes, ignoring: %m"); | |
c6c18be3 | 2927 | |
aa77e234 MS |
2928 | /* 6. And pin it, so that it cannot be unmounted */ |
2929 | safe_close(m->pin_cgroupfs_fd); | |
2930 | m->pin_cgroupfs_fd = open(path, O_RDONLY|O_CLOEXEC|O_DIRECTORY|O_NOCTTY|O_NONBLOCK); | |
2931 | if (m->pin_cgroupfs_fd < 0) | |
2932 | return log_error_errno(errno, "Failed to open pin file: %m"); | |
0d8c31ff | 2933 | |
638cece4 | 2934 | } else if (!MANAGER_IS_TEST_RUN(m)) |
aa77e234 | 2935 | return log_error_errno(r, "Failed to create %s control group: %m", scope_path); |
10bd3e2e | 2936 | |
09e24654 | 2937 | /* 7. Always enable hierarchical support if it exists... */ |
638cece4 | 2938 | if (!all_unified && !MANAGER_IS_TEST_RUN(m)) |
10bd3e2e | 2939 | (void) cg_set_attribute("memory", "/", "memory.use_hierarchy", "1"); |
c6c18be3 | 2940 | |
17f14955 | 2941 | /* 8. Figure out which controllers are supported */ |
efdb0237 LP |
2942 | r = cg_mask_supported(&m->cgroup_supported); |
2943 | if (r < 0) | |
2944 | return log_error_errno(r, "Failed to determine supported controllers: %m"); | |
17f14955 RG |
2945 | |
2946 | /* 9. Figure out which bpf-based pseudo-controllers are supported */ | |
2947 | r = cg_bpf_mask_supported(&mask); | |
2948 | if (r < 0) | |
2949 | return log_error_errno(r, "Failed to determine supported bpf-based pseudo-controllers: %m"); | |
2950 | m->cgroup_supported |= mask; | |
2951 | ||
2952 | /* 10. Log which controllers are supported */ | |
efdb0237 | 2953 | for (c = 0; c < _CGROUP_CONTROLLER_MAX; c++) |
eee0a1e4 | 2954 | log_debug("Controller '%s' supported: %s", cgroup_controller_to_string(c), yes_no(m->cgroup_supported & CGROUP_CONTROLLER_TO_MASK(c))); |
9156e799 | 2955 | |
a32360f1 | 2956 | return 0; |
8e274523 LP |
2957 | } |
2958 | ||
c6c18be3 | 2959 | void manager_shutdown_cgroup(Manager *m, bool delete) { |
8e274523 LP |
2960 | assert(m); |
2961 | ||
9444b1f2 LP |
2962 | /* We can't really delete the group, since we are in it. But |
2963 | * let's trim it. */ | |
f6c63f6f | 2964 | if (delete && m->cgroup_root && m->test_run_flags != MANAGER_TEST_RUN_MINIMAL) |
efdb0237 LP |
2965 | (void) cg_trim(SYSTEMD_CGROUP_CONTROLLER, m->cgroup_root, false); |
2966 | ||
09e24654 LP |
2967 | m->cgroup_empty_event_source = sd_event_source_unref(m->cgroup_empty_event_source); |
2968 | ||
0bb814c2 | 2969 | m->cgroup_control_inotify_wd_unit = hashmap_free(m->cgroup_control_inotify_wd_unit); |
afcfaa69 | 2970 | m->cgroup_memory_inotify_wd_unit = hashmap_free(m->cgroup_memory_inotify_wd_unit); |
efdb0237 LP |
2971 | |
2972 | m->cgroup_inotify_event_source = sd_event_source_unref(m->cgroup_inotify_event_source); | |
2973 | m->cgroup_inotify_fd = safe_close(m->cgroup_inotify_fd); | |
8e274523 | 2974 | |
03e334a1 | 2975 | m->pin_cgroupfs_fd = safe_close(m->pin_cgroupfs_fd); |
c6c18be3 | 2976 | |
efdb0237 | 2977 | m->cgroup_root = mfree(m->cgroup_root); |
8e274523 LP |
2978 | } |
2979 | ||
4ad49000 | 2980 | Unit* manager_get_unit_by_cgroup(Manager *m, const char *cgroup) { |
acb14d31 | 2981 | char *p; |
4ad49000 | 2982 | Unit *u; |
acb14d31 LP |
2983 | |
2984 | assert(m); | |
2985 | assert(cgroup); | |
acb14d31 | 2986 | |
4ad49000 LP |
2987 | u = hashmap_get(m->cgroup_unit, cgroup); |
2988 | if (u) | |
2989 | return u; | |
acb14d31 | 2990 | |
8e70580b | 2991 | p = strdupa(cgroup); |
acb14d31 LP |
2992 | for (;;) { |
2993 | char *e; | |
2994 | ||
2995 | e = strrchr(p, '/'); | |
efdb0237 LP |
2996 | if (!e || e == p) |
2997 | return hashmap_get(m->cgroup_unit, SPECIAL_ROOT_SLICE); | |
acb14d31 LP |
2998 | |
2999 | *e = 0; | |
3000 | ||
4ad49000 LP |
3001 | u = hashmap_get(m->cgroup_unit, p); |
3002 | if (u) | |
3003 | return u; | |
acb14d31 LP |
3004 | } |
3005 | } | |
3006 | ||
b3ac818b | 3007 | Unit *manager_get_unit_by_pid_cgroup(Manager *m, pid_t pid) { |
4ad49000 | 3008 | _cleanup_free_ char *cgroup = NULL; |
8e274523 | 3009 | |
8c47c732 LP |
3010 | assert(m); |
3011 | ||
62a76913 | 3012 | if (!pid_is_valid(pid)) |
b3ac818b LP |
3013 | return NULL; |
3014 | ||
62a76913 | 3015 | if (cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER, pid, &cgroup) < 0) |
b3ac818b LP |
3016 | return NULL; |
3017 | ||
3018 | return manager_get_unit_by_cgroup(m, cgroup); | |
3019 | } | |
3020 | ||
3021 | Unit *manager_get_unit_by_pid(Manager *m, pid_t pid) { | |
62a76913 | 3022 | Unit *u, **array; |
b3ac818b LP |
3023 | |
3024 | assert(m); | |
3025 | ||
62a76913 LP |
3026 | /* Note that a process might be owned by multiple units, we return only one here, which is good enough for most |
3027 | * cases, though not strictly correct. We prefer the one reported by cgroup membership, as that's the most | |
3028 | * relevant one as children of the process will be assigned to that one, too, before all else. */ | |
3029 | ||
3030 | if (!pid_is_valid(pid)) | |
8c47c732 LP |
3031 | return NULL; |
3032 | ||
2ca9d979 | 3033 | if (pid == getpid_cached()) |
efdb0237 LP |
3034 | return hashmap_get(m->units, SPECIAL_INIT_SCOPE); |
3035 | ||
62a76913 | 3036 | u = manager_get_unit_by_pid_cgroup(m, pid); |
5fe8876b LP |
3037 | if (u) |
3038 | return u; | |
3039 | ||
62a76913 | 3040 | u = hashmap_get(m->watch_pids, PID_TO_PTR(pid)); |
5fe8876b LP |
3041 | if (u) |
3042 | return u; | |
3043 | ||
62a76913 LP |
3044 | array = hashmap_get(m->watch_pids, PID_TO_PTR(-pid)); |
3045 | if (array) | |
3046 | return array[0]; | |
3047 | ||
3048 | return NULL; | |
6dde1f33 | 3049 | } |
4fbf50b3 | 3050 | |
4ad49000 LP |
3051 | int manager_notify_cgroup_empty(Manager *m, const char *cgroup) { |
3052 | Unit *u; | |
4fbf50b3 | 3053 | |
4ad49000 LP |
3054 | assert(m); |
3055 | assert(cgroup); | |
4fbf50b3 | 3056 | |
09e24654 LP |
3057 | /* Called on the legacy hierarchy whenever we get an explicit cgroup notification from the cgroup agent process |
3058 | * or from the --system instance */ | |
3059 | ||
d8fdc620 LP |
3060 | log_debug("Got cgroup empty notification for: %s", cgroup); |
3061 | ||
4ad49000 | 3062 | u = manager_get_unit_by_cgroup(m, cgroup); |
5ad096b3 LP |
3063 | if (!u) |
3064 | return 0; | |
b56c28c3 | 3065 | |
09e24654 LP |
3066 | unit_add_to_cgroup_empty_queue(u); |
3067 | return 1; | |
5ad096b3 LP |
3068 | } |
3069 | ||
3070 | int unit_get_memory_current(Unit *u, uint64_t *ret) { | |
3071 | _cleanup_free_ char *v = NULL; | |
3072 | int r; | |
3073 | ||
3074 | assert(u); | |
3075 | assert(ret); | |
3076 | ||
2e4025c0 | 3077 | if (!UNIT_CGROUP_BOOL(u, memory_accounting)) |
cf3b4be1 LP |
3078 | return -ENODATA; |
3079 | ||
5ad096b3 LP |
3080 | if (!u->cgroup_path) |
3081 | return -ENODATA; | |
3082 | ||
1f73aa00 | 3083 | /* The root cgroup doesn't expose this information, let's get it from /proc instead */ |
611c4f8a | 3084 | if (unit_has_host_root_cgroup(u)) |
c482724a | 3085 | return procfs_memory_get_used(ret); |
1f73aa00 | 3086 | |
efdb0237 | 3087 | if ((u->cgroup_realized_mask & CGROUP_MASK_MEMORY) == 0) |
5ad096b3 LP |
3088 | return -ENODATA; |
3089 | ||
b4cccbc1 LP |
3090 | r = cg_all_unified(); |
3091 | if (r < 0) | |
3092 | return r; | |
3093 | if (r > 0) | |
efdb0237 | 3094 | r = cg_get_attribute("memory", u->cgroup_path, "memory.current", &v); |
b4cccbc1 LP |
3095 | else |
3096 | r = cg_get_attribute("memory", u->cgroup_path, "memory.usage_in_bytes", &v); | |
5ad096b3 LP |
3097 | if (r == -ENOENT) |
3098 | return -ENODATA; | |
3099 | if (r < 0) | |
3100 | return r; | |
3101 | ||
3102 | return safe_atou64(v, ret); | |
3103 | } | |
3104 | ||
03a7b521 LP |
3105 | int unit_get_tasks_current(Unit *u, uint64_t *ret) { |
3106 | _cleanup_free_ char *v = NULL; | |
3107 | int r; | |
3108 | ||
3109 | assert(u); | |
3110 | assert(ret); | |
3111 | ||
2e4025c0 | 3112 | if (!UNIT_CGROUP_BOOL(u, tasks_accounting)) |
cf3b4be1 LP |
3113 | return -ENODATA; |
3114 | ||
03a7b521 LP |
3115 | if (!u->cgroup_path) |
3116 | return -ENODATA; | |
3117 | ||
c36a69f4 | 3118 | /* The root cgroup doesn't expose this information, let's get it from /proc instead */ |
611c4f8a | 3119 | if (unit_has_host_root_cgroup(u)) |
c36a69f4 LP |
3120 | return procfs_tasks_get_current(ret); |
3121 | ||
1f73aa00 LP |
3122 | if ((u->cgroup_realized_mask & CGROUP_MASK_PIDS) == 0) |
3123 | return -ENODATA; | |
3124 | ||
03a7b521 LP |
3125 | r = cg_get_attribute("pids", u->cgroup_path, "pids.current", &v); |
3126 | if (r == -ENOENT) | |
3127 | return -ENODATA; | |
3128 | if (r < 0) | |
3129 | return r; | |
3130 | ||
3131 | return safe_atou64(v, ret); | |
3132 | } | |
3133 | ||
5ad096b3 LP |
3134 | static int unit_get_cpu_usage_raw(Unit *u, nsec_t *ret) { |
3135 | _cleanup_free_ char *v = NULL; | |
3136 | uint64_t ns; | |
3137 | int r; | |
3138 | ||
3139 | assert(u); | |
3140 | assert(ret); | |
3141 | ||
3142 | if (!u->cgroup_path) | |
3143 | return -ENODATA; | |
3144 | ||
1f73aa00 | 3145 | /* The root cgroup doesn't expose this information, let's get it from /proc instead */ |
611c4f8a | 3146 | if (unit_has_host_root_cgroup(u)) |
1f73aa00 LP |
3147 | return procfs_cpu_get_usage(ret); |
3148 | ||
f98c2585 CD |
3149 | /* Requisite controllers for CPU accounting are not enabled */ |
3150 | if ((get_cpu_accounting_mask() & ~u->cgroup_realized_mask) != 0) | |
3151 | return -ENODATA; | |
3152 | ||
92a99304 LP |
3153 | r = cg_all_unified(); |
3154 | if (r < 0) | |
3155 | return r; | |
b4cccbc1 | 3156 | if (r > 0) { |
66ebf6c0 TH |
3157 | _cleanup_free_ char *val = NULL; |
3158 | uint64_t us; | |
5ad096b3 | 3159 | |
b734a4ff | 3160 | r = cg_get_keyed_attribute("cpu", u->cgroup_path, "cpu.stat", STRV_MAKE("usage_usec"), &val); |
b734a4ff LP |
3161 | if (IN_SET(r, -ENOENT, -ENXIO)) |
3162 | return -ENODATA; | |
d742f4b5 LP |
3163 | if (r < 0) |
3164 | return r; | |
66ebf6c0 TH |
3165 | |
3166 | r = safe_atou64(val, &us); | |
3167 | if (r < 0) | |
3168 | return r; | |
3169 | ||
3170 | ns = us * NSEC_PER_USEC; | |
3171 | } else { | |
66ebf6c0 TH |
3172 | r = cg_get_attribute("cpuacct", u->cgroup_path, "cpuacct.usage", &v); |
3173 | if (r == -ENOENT) | |
3174 | return -ENODATA; | |
3175 | if (r < 0) | |
3176 | return r; | |
3177 | ||
3178 | r = safe_atou64(v, &ns); | |
3179 | if (r < 0) | |
3180 | return r; | |
3181 | } | |
5ad096b3 LP |
3182 | |
3183 | *ret = ns; | |
3184 | return 0; | |
3185 | } | |
3186 | ||
3187 | int unit_get_cpu_usage(Unit *u, nsec_t *ret) { | |
3188 | nsec_t ns; | |
3189 | int r; | |
3190 | ||
fe700f46 LP |
3191 | assert(u); |
3192 | ||
3193 | /* Retrieve the current CPU usage counter. This will subtract the CPU counter taken when the unit was | |
3194 | * started. If the cgroup has been removed already, returns the last cached value. To cache the value, simply | |
3195 | * call this function with a NULL return value. */ | |
3196 | ||
2e4025c0 | 3197 | if (!UNIT_CGROUP_BOOL(u, cpu_accounting)) |
cf3b4be1 LP |
3198 | return -ENODATA; |
3199 | ||
5ad096b3 | 3200 | r = unit_get_cpu_usage_raw(u, &ns); |
fe700f46 LP |
3201 | if (r == -ENODATA && u->cpu_usage_last != NSEC_INFINITY) { |
3202 | /* If we can't get the CPU usage anymore (because the cgroup was already removed, for example), use our | |
3203 | * cached value. */ | |
3204 | ||
3205 | if (ret) | |
3206 | *ret = u->cpu_usage_last; | |
3207 | return 0; | |
3208 | } | |
5ad096b3 LP |
3209 | if (r < 0) |
3210 | return r; | |
3211 | ||
66ebf6c0 TH |
3212 | if (ns > u->cpu_usage_base) |
3213 | ns -= u->cpu_usage_base; | |
5ad096b3 LP |
3214 | else |
3215 | ns = 0; | |
3216 | ||
fe700f46 LP |
3217 | u->cpu_usage_last = ns; |
3218 | if (ret) | |
3219 | *ret = ns; | |
3220 | ||
5ad096b3 LP |
3221 | return 0; |
3222 | } | |
3223 | ||
906c06f6 DM |
3224 | int unit_get_ip_accounting( |
3225 | Unit *u, | |
3226 | CGroupIPAccountingMetric metric, | |
3227 | uint64_t *ret) { | |
3228 | ||
6b659ed8 | 3229 | uint64_t value; |
906c06f6 DM |
3230 | int fd, r; |
3231 | ||
3232 | assert(u); | |
3233 | assert(metric >= 0); | |
3234 | assert(metric < _CGROUP_IP_ACCOUNTING_METRIC_MAX); | |
3235 | assert(ret); | |
3236 | ||
2e4025c0 | 3237 | if (!UNIT_CGROUP_BOOL(u, ip_accounting)) |
cf3b4be1 LP |
3238 | return -ENODATA; |
3239 | ||
906c06f6 DM |
3240 | fd = IN_SET(metric, CGROUP_IP_INGRESS_BYTES, CGROUP_IP_INGRESS_PACKETS) ? |
3241 | u->ip_accounting_ingress_map_fd : | |
3242 | u->ip_accounting_egress_map_fd; | |
906c06f6 DM |
3243 | if (fd < 0) |
3244 | return -ENODATA; | |
3245 | ||
3246 | if (IN_SET(metric, CGROUP_IP_INGRESS_BYTES, CGROUP_IP_EGRESS_BYTES)) | |
6b659ed8 | 3247 | r = bpf_firewall_read_accounting(fd, &value, NULL); |
906c06f6 | 3248 | else |
6b659ed8 LP |
3249 | r = bpf_firewall_read_accounting(fd, NULL, &value); |
3250 | if (r < 0) | |
3251 | return r; | |
3252 | ||
3253 | /* Add in additional metrics from a previous runtime. Note that when reexecing/reloading the daemon we compile | |
3254 | * all BPF programs and maps anew, but serialize the old counters. When deserializing we store them in the | |
3255 | * ip_accounting_extra[] field, and add them in here transparently. */ | |
3256 | ||
3257 | *ret = value + u->ip_accounting_extra[metric]; | |
906c06f6 DM |
3258 | |
3259 | return r; | |
3260 | } | |
3261 | ||
fbe14fc9 LP |
3262 | static int unit_get_io_accounting_raw(Unit *u, uint64_t ret[static _CGROUP_IO_ACCOUNTING_METRIC_MAX]) { |
3263 | static const char *const field_names[_CGROUP_IO_ACCOUNTING_METRIC_MAX] = { | |
3264 | [CGROUP_IO_READ_BYTES] = "rbytes=", | |
3265 | [CGROUP_IO_WRITE_BYTES] = "wbytes=", | |
3266 | [CGROUP_IO_READ_OPERATIONS] = "rios=", | |
3267 | [CGROUP_IO_WRITE_OPERATIONS] = "wios=", | |
3268 | }; | |
3269 | uint64_t acc[_CGROUP_IO_ACCOUNTING_METRIC_MAX] = {}; | |
3270 | _cleanup_free_ char *path = NULL; | |
3271 | _cleanup_fclose_ FILE *f = NULL; | |
3272 | int r; | |
3273 | ||
3274 | assert(u); | |
3275 | ||
3276 | if (!u->cgroup_path) | |
3277 | return -ENODATA; | |
3278 | ||
3279 | if (unit_has_host_root_cgroup(u)) | |
3280 | return -ENODATA; /* TODO: return useful data for the top-level cgroup */ | |
3281 | ||
3282 | r = cg_all_unified(); | |
3283 | if (r < 0) | |
3284 | return r; | |
3285 | if (r == 0) /* TODO: support cgroupv1 */ | |
3286 | return -ENODATA; | |
3287 | ||
3288 | if (!FLAGS_SET(u->cgroup_realized_mask, CGROUP_MASK_IO)) | |
3289 | return -ENODATA; | |
3290 | ||
3291 | r = cg_get_path("io", u->cgroup_path, "io.stat", &path); | |
3292 | if (r < 0) | |
3293 | return r; | |
3294 | ||
3295 | f = fopen(path, "re"); | |
3296 | if (!f) | |
3297 | return -errno; | |
3298 | ||
3299 | for (;;) { | |
3300 | _cleanup_free_ char *line = NULL; | |
3301 | const char *p; | |
3302 | ||
3303 | r = read_line(f, LONG_LINE_MAX, &line); | |
3304 | if (r < 0) | |
3305 | return r; | |
3306 | if (r == 0) | |
3307 | break; | |
3308 | ||
3309 | p = line; | |
3310 | p += strcspn(p, WHITESPACE); /* Skip over device major/minor */ | |
3311 | p += strspn(p, WHITESPACE); /* Skip over following whitespace */ | |
3312 | ||
3313 | for (;;) { | |
3314 | _cleanup_free_ char *word = NULL; | |
3315 | ||
3316 | r = extract_first_word(&p, &word, NULL, EXTRACT_RETAIN_ESCAPE); | |
3317 | if (r < 0) | |
3318 | return r; | |
3319 | if (r == 0) | |
3320 | break; | |
3321 | ||
3322 | for (CGroupIOAccountingMetric i = 0; i < _CGROUP_IO_ACCOUNTING_METRIC_MAX; i++) { | |
3323 | const char *x; | |
3324 | ||
3325 | x = startswith(word, field_names[i]); | |
3326 | if (x) { | |
3327 | uint64_t w; | |
3328 | ||
3329 | r = safe_atou64(x, &w); | |
3330 | if (r < 0) | |
3331 | return r; | |
3332 | ||
3333 | /* Sum up the stats of all devices */ | |
3334 | acc[i] += w; | |
3335 | break; | |
3336 | } | |
3337 | } | |
3338 | } | |
3339 | } | |
3340 | ||
3341 | memcpy(ret, acc, sizeof(acc)); | |
3342 | return 0; | |
3343 | } | |
3344 | ||
3345 | int unit_get_io_accounting( | |
3346 | Unit *u, | |
3347 | CGroupIOAccountingMetric metric, | |
3348 | bool allow_cache, | |
3349 | uint64_t *ret) { | |
3350 | ||
3351 | uint64_t raw[_CGROUP_IO_ACCOUNTING_METRIC_MAX]; | |
3352 | int r; | |
3353 | ||
3354 | /* Retrieve an IO account parameter. This will subtract the counter when the unit was started. */ | |
3355 | ||
3356 | if (!UNIT_CGROUP_BOOL(u, io_accounting)) | |
3357 | return -ENODATA; | |
3358 | ||
3359 | if (allow_cache && u->io_accounting_last[metric] != UINT64_MAX) | |
3360 | goto done; | |
3361 | ||
3362 | r = unit_get_io_accounting_raw(u, raw); | |
3363 | if (r == -ENODATA && u->io_accounting_last[metric] != UINT64_MAX) | |
3364 | goto done; | |
3365 | if (r < 0) | |
3366 | return r; | |
3367 | ||
3368 | for (CGroupIOAccountingMetric i = 0; i < _CGROUP_IO_ACCOUNTING_METRIC_MAX; i++) { | |
3369 | /* Saturated subtraction */ | |
3370 | if (raw[i] > u->io_accounting_base[i]) | |
3371 | u->io_accounting_last[i] = raw[i] - u->io_accounting_base[i]; | |
3372 | else | |
3373 | u->io_accounting_last[i] = 0; | |
3374 | } | |
3375 | ||
3376 | done: | |
3377 | if (ret) | |
3378 | *ret = u->io_accounting_last[metric]; | |
3379 | ||
3380 | return 0; | |
3381 | } | |
3382 | ||
906c06f6 | 3383 | int unit_reset_cpu_accounting(Unit *u) { |
5ad096b3 LP |
3384 | int r; |
3385 | ||
3386 | assert(u); | |
3387 | ||
fe700f46 LP |
3388 | u->cpu_usage_last = NSEC_INFINITY; |
3389 | ||
0bbff7d6 | 3390 | r = unit_get_cpu_usage_raw(u, &u->cpu_usage_base); |
5ad096b3 | 3391 | if (r < 0) { |
66ebf6c0 | 3392 | u->cpu_usage_base = 0; |
5ad096b3 | 3393 | return r; |
b56c28c3 | 3394 | } |
2633eb83 | 3395 | |
4ad49000 | 3396 | return 0; |
4fbf50b3 LP |
3397 | } |
3398 | ||
906c06f6 DM |
3399 | int unit_reset_ip_accounting(Unit *u) { |
3400 | int r = 0, q = 0; | |
3401 | ||
3402 | assert(u); | |
3403 | ||
3404 | if (u->ip_accounting_ingress_map_fd >= 0) | |
3405 | r = bpf_firewall_reset_accounting(u->ip_accounting_ingress_map_fd); | |
3406 | ||
3407 | if (u->ip_accounting_egress_map_fd >= 0) | |
3408 | q = bpf_firewall_reset_accounting(u->ip_accounting_egress_map_fd); | |
3409 | ||
6b659ed8 LP |
3410 | zero(u->ip_accounting_extra); |
3411 | ||
906c06f6 DM |
3412 | return r < 0 ? r : q; |
3413 | } | |
3414 | ||
fbe14fc9 LP |
3415 | int unit_reset_io_accounting(Unit *u) { |
3416 | int r; | |
3417 | ||
3418 | assert(u); | |
3419 | ||
3420 | for (CGroupIOAccountingMetric i = 0; i < _CGROUP_IO_ACCOUNTING_METRIC_MAX; i++) | |
3421 | u->io_accounting_last[i] = UINT64_MAX; | |
3422 | ||
3423 | r = unit_get_io_accounting_raw(u, u->io_accounting_base); | |
3424 | if (r < 0) { | |
3425 | zero(u->io_accounting_base); | |
3426 | return r; | |
3427 | } | |
3428 | ||
3429 | return 0; | |
3430 | } | |
3431 | ||
9b2559a1 | 3432 | int unit_reset_accounting(Unit *u) { |
fbe14fc9 | 3433 | int r, q, v; |
9b2559a1 LP |
3434 | |
3435 | assert(u); | |
3436 | ||
3437 | r = unit_reset_cpu_accounting(u); | |
fbe14fc9 LP |
3438 | q = unit_reset_io_accounting(u); |
3439 | v = unit_reset_ip_accounting(u); | |
9b2559a1 | 3440 | |
fbe14fc9 | 3441 | return r < 0 ? r : q < 0 ? q : v; |
9b2559a1 LP |
3442 | } |
3443 | ||
e7ab4d1a LP |
3444 | void unit_invalidate_cgroup(Unit *u, CGroupMask m) { |
3445 | assert(u); | |
3446 | ||
3447 | if (!UNIT_HAS_CGROUP_CONTEXT(u)) | |
3448 | return; | |
3449 | ||
3450 | if (m == 0) | |
3451 | return; | |
3452 | ||
538b4852 TH |
3453 | /* always invalidate compat pairs together */ |
3454 | if (m & (CGROUP_MASK_IO | CGROUP_MASK_BLKIO)) | |
3455 | m |= CGROUP_MASK_IO | CGROUP_MASK_BLKIO; | |
3456 | ||
7cce4fb7 LP |
3457 | if (m & (CGROUP_MASK_CPU | CGROUP_MASK_CPUACCT)) |
3458 | m |= CGROUP_MASK_CPU | CGROUP_MASK_CPUACCT; | |
3459 | ||
e00068e7 | 3460 | if (FLAGS_SET(u->cgroup_invalidated_mask, m)) /* NOP? */ |
e7ab4d1a LP |
3461 | return; |
3462 | ||
e00068e7 | 3463 | u->cgroup_invalidated_mask |= m; |
91a6073e | 3464 | unit_add_to_cgroup_realize_queue(u); |
e7ab4d1a LP |
3465 | } |
3466 | ||
906c06f6 DM |
3467 | void unit_invalidate_cgroup_bpf(Unit *u) { |
3468 | assert(u); | |
3469 | ||
3470 | if (!UNIT_HAS_CGROUP_CONTEXT(u)) | |
3471 | return; | |
3472 | ||
17f14955 | 3473 | if (u->cgroup_invalidated_mask & CGROUP_MASK_BPF_FIREWALL) /* NOP? */ |
906c06f6 DM |
3474 | return; |
3475 | ||
17f14955 | 3476 | u->cgroup_invalidated_mask |= CGROUP_MASK_BPF_FIREWALL; |
91a6073e | 3477 | unit_add_to_cgroup_realize_queue(u); |
906c06f6 DM |
3478 | |
3479 | /* If we are a slice unit, we also need to put compile a new BPF program for all our children, as the IP access | |
3480 | * list of our children includes our own. */ | |
3481 | if (u->type == UNIT_SLICE) { | |
3482 | Unit *member; | |
3483 | Iterator i; | |
eef85c4a | 3484 | void *v; |
906c06f6 | 3485 | |
eef85c4a | 3486 | HASHMAP_FOREACH_KEY(v, member, u->dependencies[UNIT_BEFORE], i) { |
cb5e3bc3 CD |
3487 | if (UNIT_DEREF(member->slice) == u) |
3488 | unit_invalidate_cgroup_bpf(member); | |
906c06f6 DM |
3489 | } |
3490 | } | |
3491 | } | |
3492 | ||
1d9cc876 LP |
3493 | bool unit_cgroup_delegate(Unit *u) { |
3494 | CGroupContext *c; | |
3495 | ||
3496 | assert(u); | |
3497 | ||
3498 | if (!UNIT_VTABLE(u)->can_delegate) | |
3499 | return false; | |
3500 | ||
3501 | c = unit_get_cgroup_context(u); | |
3502 | if (!c) | |
3503 | return false; | |
3504 | ||
3505 | return c->delegate; | |
3506 | } | |
3507 | ||
e7ab4d1a LP |
3508 | void manager_invalidate_startup_units(Manager *m) { |
3509 | Iterator i; | |
3510 | Unit *u; | |
3511 | ||
3512 | assert(m); | |
3513 | ||
3514 | SET_FOREACH(u, m->startup_units, i) | |
13c31542 | 3515 | unit_invalidate_cgroup(u, CGROUP_MASK_CPU|CGROUP_MASK_IO|CGROUP_MASK_BLKIO); |
e7ab4d1a LP |
3516 | } |
3517 | ||
da8e1782 MO |
3518 | static int unit_get_nice(Unit *u) { |
3519 | ExecContext *ec; | |
3520 | ||
3521 | ec = unit_get_exec_context(u); | |
3522 | return ec ? ec->nice : 0; | |
3523 | } | |
3524 | ||
3525 | static uint64_t unit_get_cpu_weight(Unit *u) { | |
3526 | ManagerState state = manager_state(u->manager); | |
3527 | CGroupContext *cc; | |
3528 | ||
3529 | cc = unit_get_cgroup_context(u); | |
3530 | return cc ? cgroup_context_cpu_weight(cc, state) : CGROUP_WEIGHT_DEFAULT; | |
3531 | } | |
3532 | ||
3533 | int compare_job_priority(const void *a, const void *b) { | |
3534 | const Job *x = a, *y = b; | |
3535 | int nice_x, nice_y; | |
3536 | uint64_t weight_x, weight_y; | |
3537 | int ret; | |
3538 | ||
217b7b33 ZJS |
3539 | if ((ret = CMP(x->unit->type, y->unit->type)) != 0) |
3540 | return -ret; | |
3541 | ||
da8e1782 MO |
3542 | weight_x = unit_get_cpu_weight(x->unit); |
3543 | weight_y = unit_get_cpu_weight(y->unit); | |
3544 | ||
217b7b33 ZJS |
3545 | if ((ret = CMP(weight_x, weight_y)) != 0) |
3546 | return -ret; | |
da8e1782 MO |
3547 | |
3548 | nice_x = unit_get_nice(x->unit); | |
3549 | nice_y = unit_get_nice(y->unit); | |
3550 | ||
3551 | if ((ret = CMP(nice_x, nice_y)) != 0) | |
3552 | return ret; | |
3553 | ||
da8e1782 MO |
3554 | return strcmp(x->unit->id, y->unit->id); |
3555 | } | |
3556 | ||
4ad49000 LP |
3557 | static const char* const cgroup_device_policy_table[_CGROUP_DEVICE_POLICY_MAX] = { |
3558 | [CGROUP_AUTO] = "auto", | |
3559 | [CGROUP_CLOSED] = "closed", | |
3560 | [CGROUP_STRICT] = "strict", | |
3561 | }; | |
4fbf50b3 | 3562 | |
4ad49000 | 3563 | DEFINE_STRING_TABLE_LOOKUP(cgroup_device_policy, CGroupDevicePolicy); |