]>
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); | |
4ad49000 LP |
202 | } |
203 | ||
204 | void cgroup_context_dump(CGroupContext *c, FILE* f, const char *prefix) { | |
13c31542 TH |
205 | CGroupIODeviceLimit *il; |
206 | CGroupIODeviceWeight *iw; | |
6ae4283c | 207 | CGroupIODeviceLatency *l; |
4ad49000 LP |
208 | CGroupBlockIODeviceBandwidth *b; |
209 | CGroupBlockIODeviceWeight *w; | |
210 | CGroupDeviceAllow *a; | |
c21c9906 | 211 | IPAddressAccessItem *iaai; |
9a054909 | 212 | char u[FORMAT_TIMESPAN_MAX]; |
10f28641 | 213 | char v[FORMAT_TIMESPAN_MAX]; |
4ad49000 LP |
214 | |
215 | assert(c); | |
216 | assert(f); | |
217 | ||
218 | prefix = strempty(prefix); | |
219 | ||
220 | fprintf(f, | |
221 | "%sCPUAccounting=%s\n" | |
13c31542 | 222 | "%sIOAccounting=%s\n" |
4ad49000 LP |
223 | "%sBlockIOAccounting=%s\n" |
224 | "%sMemoryAccounting=%s\n" | |
d53d9474 | 225 | "%sTasksAccounting=%s\n" |
c21c9906 | 226 | "%sIPAccounting=%s\n" |
66ebf6c0 TH |
227 | "%sCPUWeight=%" PRIu64 "\n" |
228 | "%sStartupCPUWeight=%" PRIu64 "\n" | |
d53d9474 LP |
229 | "%sCPUShares=%" PRIu64 "\n" |
230 | "%sStartupCPUShares=%" PRIu64 "\n" | |
b2f8b02e | 231 | "%sCPUQuotaPerSecSec=%s\n" |
10f28641 | 232 | "%sCPUQuotaPeriodSec=%s\n" |
13c31542 TH |
233 | "%sIOWeight=%" PRIu64 "\n" |
234 | "%sStartupIOWeight=%" PRIu64 "\n" | |
d53d9474 LP |
235 | "%sBlockIOWeight=%" PRIu64 "\n" |
236 | "%sStartupBlockIOWeight=%" PRIu64 "\n" | |
c52db42b | 237 | "%sDefaultMemoryLow=%" PRIu64 "\n" |
48422635 | 238 | "%sMemoryMin=%" PRIu64 "\n" |
da4d897e TH |
239 | "%sMemoryLow=%" PRIu64 "\n" |
240 | "%sMemoryHigh=%" PRIu64 "\n" | |
241 | "%sMemoryMax=%" PRIu64 "\n" | |
96e131ea | 242 | "%sMemorySwapMax=%" PRIu64 "\n" |
4ad49000 | 243 | "%sMemoryLimit=%" PRIu64 "\n" |
03a7b521 | 244 | "%sTasksMax=%" PRIu64 "\n" |
a931ad47 LP |
245 | "%sDevicePolicy=%s\n" |
246 | "%sDelegate=%s\n", | |
4ad49000 | 247 | prefix, yes_no(c->cpu_accounting), |
13c31542 | 248 | prefix, yes_no(c->io_accounting), |
4ad49000 LP |
249 | prefix, yes_no(c->blockio_accounting), |
250 | prefix, yes_no(c->memory_accounting), | |
d53d9474 | 251 | prefix, yes_no(c->tasks_accounting), |
c21c9906 | 252 | prefix, yes_no(c->ip_accounting), |
66ebf6c0 TH |
253 | prefix, c->cpu_weight, |
254 | prefix, c->startup_cpu_weight, | |
4ad49000 | 255 | prefix, c->cpu_shares, |
95ae05c0 | 256 | prefix, c->startup_cpu_shares, |
b1d6dcf5 | 257 | prefix, format_timespan(u, sizeof(u), c->cpu_quota_per_sec_usec, 1), |
10f28641 | 258 | prefix, format_timespan(v, sizeof(v), c->cpu_quota_period_usec, 1), |
13c31542 TH |
259 | prefix, c->io_weight, |
260 | prefix, c->startup_io_weight, | |
4ad49000 | 261 | prefix, c->blockio_weight, |
95ae05c0 | 262 | prefix, c->startup_blockio_weight, |
c52db42b | 263 | prefix, c->default_memory_low, |
48422635 | 264 | prefix, c->memory_min, |
da4d897e TH |
265 | prefix, c->memory_low, |
266 | prefix, c->memory_high, | |
267 | prefix, c->memory_max, | |
96e131ea | 268 | prefix, c->memory_swap_max, |
4ad49000 | 269 | prefix, c->memory_limit, |
03a7b521 | 270 | prefix, c->tasks_max, |
a931ad47 LP |
271 | prefix, cgroup_device_policy_to_string(c->device_policy), |
272 | prefix, yes_no(c->delegate)); | |
4ad49000 | 273 | |
02638280 LP |
274 | if (c->delegate) { |
275 | _cleanup_free_ char *t = NULL; | |
276 | ||
277 | (void) cg_mask_to_string(c->delegate_controllers, &t); | |
278 | ||
47a78d41 | 279 | fprintf(f, "%sDelegateControllers=%s\n", |
02638280 LP |
280 | prefix, |
281 | strempty(t)); | |
282 | } | |
283 | ||
4ad49000 LP |
284 | LIST_FOREACH(device_allow, a, c->device_allow) |
285 | fprintf(f, | |
286 | "%sDeviceAllow=%s %s%s%s\n", | |
287 | prefix, | |
288 | a->path, | |
289 | a->r ? "r" : "", a->w ? "w" : "", a->m ? "m" : ""); | |
290 | ||
13c31542 TH |
291 | LIST_FOREACH(device_weights, iw, c->io_device_weights) |
292 | fprintf(f, | |
6ae4283c | 293 | "%sIODeviceWeight=%s %" PRIu64 "\n", |
13c31542 TH |
294 | prefix, |
295 | iw->path, | |
296 | iw->weight); | |
297 | ||
6ae4283c TH |
298 | LIST_FOREACH(device_latencies, l, c->io_device_latencies) |
299 | fprintf(f, | |
300 | "%sIODeviceLatencyTargetSec=%s %s\n", | |
301 | prefix, | |
302 | l->path, | |
303 | format_timespan(u, sizeof(u), l->target_usec, 1)); | |
304 | ||
13c31542 TH |
305 | LIST_FOREACH(device_limits, il, c->io_device_limits) { |
306 | char buf[FORMAT_BYTES_MAX]; | |
9be57249 TH |
307 | CGroupIOLimitType type; |
308 | ||
309 | for (type = 0; type < _CGROUP_IO_LIMIT_TYPE_MAX; type++) | |
310 | if (il->limits[type] != cgroup_io_limit_defaults[type]) | |
311 | fprintf(f, | |
312 | "%s%s=%s %s\n", | |
313 | prefix, | |
314 | cgroup_io_limit_type_to_string(type), | |
315 | il->path, | |
316 | format_bytes(buf, sizeof(buf), il->limits[type])); | |
13c31542 TH |
317 | } |
318 | ||
4ad49000 LP |
319 | LIST_FOREACH(device_weights, w, c->blockio_device_weights) |
320 | fprintf(f, | |
d53d9474 | 321 | "%sBlockIODeviceWeight=%s %" PRIu64, |
4ad49000 LP |
322 | prefix, |
323 | w->path, | |
324 | w->weight); | |
325 | ||
326 | LIST_FOREACH(device_bandwidths, b, c->blockio_device_bandwidths) { | |
327 | char buf[FORMAT_BYTES_MAX]; | |
328 | ||
979d0311 TH |
329 | if (b->rbps != CGROUP_LIMIT_MAX) |
330 | fprintf(f, | |
331 | "%sBlockIOReadBandwidth=%s %s\n", | |
332 | prefix, | |
333 | b->path, | |
334 | format_bytes(buf, sizeof(buf), b->rbps)); | |
335 | if (b->wbps != CGROUP_LIMIT_MAX) | |
336 | fprintf(f, | |
337 | "%sBlockIOWriteBandwidth=%s %s\n", | |
338 | prefix, | |
339 | b->path, | |
340 | format_bytes(buf, sizeof(buf), b->wbps)); | |
4ad49000 | 341 | } |
c21c9906 LP |
342 | |
343 | LIST_FOREACH(items, iaai, c->ip_address_allow) { | |
344 | _cleanup_free_ char *k = NULL; | |
345 | ||
346 | (void) in_addr_to_string(iaai->family, &iaai->address, &k); | |
347 | fprintf(f, "%sIPAddressAllow=%s/%u\n", prefix, strnull(k), iaai->prefixlen); | |
348 | } | |
349 | ||
350 | LIST_FOREACH(items, iaai, c->ip_address_deny) { | |
351 | _cleanup_free_ char *k = NULL; | |
352 | ||
353 | (void) in_addr_to_string(iaai->family, &iaai->address, &k); | |
354 | fprintf(f, "%sIPAddressDeny=%s/%u\n", prefix, strnull(k), iaai->prefixlen); | |
355 | } | |
4ad49000 LP |
356 | } |
357 | ||
fd870bac YW |
358 | int cgroup_add_device_allow(CGroupContext *c, const char *dev, const char *mode) { |
359 | _cleanup_free_ CGroupDeviceAllow *a = NULL; | |
360 | _cleanup_free_ char *d = NULL; | |
361 | ||
362 | assert(c); | |
363 | assert(dev); | |
364 | assert(isempty(mode) || in_charset(mode, "rwm")); | |
365 | ||
366 | a = new(CGroupDeviceAllow, 1); | |
367 | if (!a) | |
368 | return -ENOMEM; | |
369 | ||
370 | d = strdup(dev); | |
371 | if (!d) | |
372 | return -ENOMEM; | |
373 | ||
374 | *a = (CGroupDeviceAllow) { | |
375 | .path = TAKE_PTR(d), | |
490c5a37 LP |
376 | .r = isempty(mode) || strchr(mode, 'r'), |
377 | .w = isempty(mode) || strchr(mode, 'w'), | |
378 | .m = isempty(mode) || strchr(mode, 'm'), | |
fd870bac YW |
379 | }; |
380 | ||
381 | LIST_PREPEND(device_allow, c->device_allow, a); | |
382 | TAKE_PTR(a); | |
383 | ||
384 | return 0; | |
385 | } | |
386 | ||
c52db42b CD |
387 | uint64_t unit_get_ancestor_memory_low(Unit *u) { |
388 | CGroupContext *c; | |
389 | ||
390 | /* 1. Is MemoryLow set in this unit? If so, use that. | |
391 | * 2. Is DefaultMemoryLow set in any ancestor? If so, use that. | |
392 | * 3. Otherwise, return CGROUP_LIMIT_MIN. */ | |
393 | ||
394 | assert(u); | |
395 | ||
396 | c = unit_get_cgroup_context(u); | |
397 | ||
398 | if (c->memory_low_set) | |
399 | return c->memory_low; | |
400 | ||
401 | while (UNIT_ISSET(u->slice)) { | |
402 | u = UNIT_DEREF(u->slice); | |
403 | c = unit_get_cgroup_context(u); | |
404 | ||
405 | if (c->default_memory_low_set) | |
406 | return c->default_memory_low; | |
407 | } | |
408 | ||
409 | /* We've reached the root, but nobody had DefaultMemoryLow set, so set it to the kernel default. */ | |
410 | return CGROUP_LIMIT_MIN; | |
411 | } | |
412 | ||
0d2d6fbf CD |
413 | static void cgroup_xattr_apply(Unit *u) { |
414 | char ids[SD_ID128_STRING_MAX]; | |
415 | int r; | |
416 | ||
417 | assert(u); | |
418 | ||
419 | if (!MANAGER_IS_SYSTEM(u->manager)) | |
420 | return; | |
421 | ||
422 | if (sd_id128_is_null(u->invocation_id)) | |
423 | return; | |
424 | ||
425 | r = cg_set_xattr(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path, | |
426 | "trusted.invocation_id", | |
427 | sd_id128_to_string(u->invocation_id, ids), 32, | |
428 | 0); | |
429 | if (r < 0) | |
430 | log_unit_debug_errno(u, r, "Failed to set invocation ID on control group %s, ignoring: %m", u->cgroup_path); | |
431 | } | |
432 | ||
45c2e068 | 433 | static int lookup_block_device(const char *p, dev_t *ret) { |
f5855697 YS |
434 | dev_t rdev, dev = 0; |
435 | mode_t mode; | |
45c2e068 | 436 | int r; |
4ad49000 LP |
437 | |
438 | assert(p); | |
45c2e068 | 439 | assert(ret); |
4ad49000 | 440 | |
f5855697 | 441 | r = device_path_parse_major_minor(p, &mode, &rdev); |
d5aecba6 | 442 | if (r == -ENODEV) { /* not a parsable device node, need to go to disk */ |
f5855697 | 443 | struct stat st; |
d5aecba6 LP |
444 | if (stat(p, &st) < 0) |
445 | return log_warning_errno(errno, "Couldn't stat device '%s': %m", p); | |
f5855697 YS |
446 | rdev = (dev_t)st.st_rdev; |
447 | dev = (dev_t)st.st_dev; | |
448 | mode = st.st_mode; | |
d5aecba6 LP |
449 | } else if (r < 0) |
450 | return log_warning_errno(r, "Failed to parse major/minor from path '%s': %m", p); | |
451 | ||
f5855697 | 452 | if (S_ISCHR(mode)) { |
d5aecba6 LP |
453 | log_warning("Device node '%s' is a character device, but block device needed.", p); |
454 | return -ENOTBLK; | |
f5855697 YS |
455 | } else if (S_ISBLK(mode)) |
456 | *ret = rdev; | |
457 | else if (major(dev) != 0) | |
458 | *ret = dev; /* If this is not a device node then use the block device this file is stored on */ | |
45c2e068 LP |
459 | else { |
460 | /* If this is btrfs, getting the backing block device is a bit harder */ | |
461 | r = btrfs_get_block_device(p, ret); | |
462 | if (r < 0 && r != -ENOTTY) | |
463 | return log_warning_errno(r, "Failed to determine block device backing btrfs file system '%s': %m", p); | |
464 | if (r == -ENOTTY) { | |
465 | log_warning("'%s' is not a block device node, and file system block device cannot be determined or is not local.", p); | |
466 | return -ENODEV; | |
467 | } | |
4ad49000 | 468 | } |
8e274523 | 469 | |
45c2e068 LP |
470 | /* If this is a LUKS device, try to get the originating block device */ |
471 | (void) block_get_originating(*ret, ret); | |
472 | ||
473 | /* If this is a partition, try to get the originating block device */ | |
474 | (void) block_get_whole_disk(*ret, ret); | |
8e274523 | 475 | return 0; |
8e274523 LP |
476 | } |
477 | ||
084c7007 | 478 | static int whitelist_device(BPFProgram *prog, const char *path, const char *node, const char *acc) { |
f5855697 YS |
479 | dev_t rdev; |
480 | mode_t mode; | |
8c6db833 | 481 | int r; |
8e274523 | 482 | |
4ad49000 LP |
483 | assert(path); |
484 | assert(acc); | |
8e274523 | 485 | |
74c48bf5 LP |
486 | /* Some special handling for /dev/block/%u:%u, /dev/char/%u:%u, /run/systemd/inaccessible/chr and |
487 | * /run/systemd/inaccessible/blk paths. Instead of stat()ing these we parse out the major/minor directly. This | |
488 | * means clients can use these path without the device node actually around */ | |
f5855697 | 489 | r = device_path_parse_major_minor(node, &mode, &rdev); |
74c48bf5 LP |
490 | if (r < 0) { |
491 | if (r != -ENODEV) | |
492 | return log_warning_errno(r, "Couldn't parse major/minor from device path '%s': %m", node); | |
b200489b | 493 | |
f5855697 | 494 | struct stat st; |
74c48bf5 LP |
495 | if (stat(node, &st) < 0) |
496 | return log_warning_errno(errno, "Couldn't stat device %s: %m", node); | |
e7330dfe | 497 | |
74c48bf5 LP |
498 | if (!S_ISCHR(st.st_mode) && !S_ISBLK(st.st_mode)) { |
499 | log_warning("%s is not a device.", node); | |
500 | return -ENODEV; | |
501 | } | |
f5855697 YS |
502 | rdev = (dev_t) st.st_rdev; |
503 | mode = st.st_mode; | |
4ad49000 LP |
504 | } |
505 | ||
084c7007 RG |
506 | if (cg_all_unified() > 0) { |
507 | if (!prog) | |
508 | return 0; | |
4ad49000 | 509 | |
f5855697 YS |
510 | return cgroup_bpf_whitelist_device(prog, S_ISCHR(mode) ? BPF_DEVCG_DEV_CHAR : BPF_DEVCG_DEV_BLOCK, |
511 | major(rdev), minor(rdev), acc); | |
b9839ac9 | 512 | |
084c7007 RG |
513 | } else { |
514 | char buf[2+DECIMAL_STR_MAX(dev_t)*2+2+4]; | |
515 | ||
516 | sprintf(buf, | |
517 | "%c %u:%u %s", | |
f5855697 YS |
518 | S_ISCHR(mode) ? 'c' : 'b', |
519 | major(rdev), minor(rdev), | |
084c7007 RG |
520 | acc); |
521 | ||
8c838407 LP |
522 | /* Changing the devices list of a populated cgroup might result in EINVAL, hence ignore EINVAL here. */ |
523 | ||
084c7007 RG |
524 | r = cg_set_attribute("devices", path, "devices.allow", buf); |
525 | if (r < 0) | |
2c74e12b | 526 | return log_full_errno(IN_SET(r, -ENOENT, -EROFS, -EINVAL, -EACCES, -EPERM) ? LOG_DEBUG : LOG_WARNING, |
b9839ac9 | 527 | r, "Failed to set devices.allow on %s: %m", path); |
4ad49000 | 528 | |
b9839ac9 LP |
529 | return 0; |
530 | } | |
8e274523 LP |
531 | } |
532 | ||
084c7007 | 533 | static int whitelist_major(BPFProgram *prog, const char *path, const char *name, char type, const char *acc) { |
90060676 | 534 | _cleanup_fclose_ FILE *f = NULL; |
8e8b5d2e | 535 | char buf[2+DECIMAL_STR_MAX(unsigned)+3+4]; |
90060676 | 536 | bool good = false; |
8e8b5d2e | 537 | unsigned maj; |
90060676 LP |
538 | int r; |
539 | ||
540 | assert(path); | |
541 | assert(acc); | |
4c701096 | 542 | assert(IN_SET(type, 'b', 'c')); |
90060676 | 543 | |
8e8b5d2e LP |
544 | if (streq(name, "*")) { |
545 | /* If the name is a wildcard, then apply this list to all devices of this type */ | |
546 | ||
547 | if (cg_all_unified() > 0) { | |
548 | if (!prog) | |
549 | return 0; | |
550 | ||
551 | (void) cgroup_bpf_whitelist_class(prog, type == 'c' ? BPF_DEVCG_DEV_CHAR : BPF_DEVCG_DEV_BLOCK, acc); | |
552 | } else { | |
553 | xsprintf(buf, "%c *:* %s", type, acc); | |
554 | ||
555 | r = cg_set_attribute("devices", path, "devices.allow", buf); | |
556 | if (r < 0) | |
557 | log_full_errno(IN_SET(r, -ENOENT, -EROFS, -EINVAL, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, | |
558 | "Failed to set devices.allow on %s: %m", path); | |
559 | return 0; | |
560 | } | |
561 | } | |
562 | ||
563 | if (safe_atou(name, &maj) >= 0 && DEVICE_MAJOR_VALID(maj)) { | |
564 | /* The name is numeric and suitable as major. In that case, let's take is major, and create the entry | |
565 | * directly */ | |
566 | ||
567 | if (cg_all_unified() > 0) { | |
568 | if (!prog) | |
569 | return 0; | |
570 | ||
571 | (void) cgroup_bpf_whitelist_major(prog, | |
572 | type == 'c' ? BPF_DEVCG_DEV_CHAR : BPF_DEVCG_DEV_BLOCK, | |
573 | maj, acc); | |
574 | } else { | |
575 | xsprintf(buf, "%c %u:* %s", type, maj, acc); | |
576 | ||
577 | r = cg_set_attribute("devices", path, "devices.allow", buf); | |
578 | if (r < 0) | |
579 | log_full_errno(IN_SET(r, -ENOENT, -EROFS, -EINVAL, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, | |
580 | "Failed to set devices.allow on %s: %m", path); | |
581 | } | |
582 | ||
583 | return 0; | |
584 | } | |
585 | ||
90060676 | 586 | f = fopen("/proc/devices", "re"); |
4a62c710 MS |
587 | if (!f) |
588 | return log_warning_errno(errno, "Cannot open /proc/devices to resolve %s (%c): %m", name, type); | |
90060676 | 589 | |
c66e60a8 LP |
590 | for (;;) { |
591 | _cleanup_free_ char *line = NULL; | |
8e8b5d2e | 592 | char *w, *p; |
90060676 | 593 | |
c66e60a8 LP |
594 | r = read_line(f, LONG_LINE_MAX, &line); |
595 | if (r < 0) | |
596 | return log_warning_errno(r, "Failed to read /proc/devices: %m"); | |
597 | if (r == 0) | |
598 | break; | |
90060676 LP |
599 | |
600 | if (type == 'c' && streq(line, "Character devices:")) { | |
601 | good = true; | |
602 | continue; | |
603 | } | |
604 | ||
605 | if (type == 'b' && streq(line, "Block devices:")) { | |
606 | good = true; | |
607 | continue; | |
608 | } | |
609 | ||
610 | if (isempty(line)) { | |
611 | good = false; | |
612 | continue; | |
613 | } | |
614 | ||
615 | if (!good) | |
616 | continue; | |
617 | ||
618 | p = strstrip(line); | |
619 | ||
620 | w = strpbrk(p, WHITESPACE); | |
621 | if (!w) | |
622 | continue; | |
623 | *w = 0; | |
624 | ||
625 | r = safe_atou(p, &maj); | |
626 | if (r < 0) | |
627 | continue; | |
628 | if (maj <= 0) | |
629 | continue; | |
630 | ||
631 | w++; | |
632 | w += strspn(w, WHITESPACE); | |
e41969e3 LP |
633 | |
634 | if (fnmatch(name, w, 0) != 0) | |
90060676 LP |
635 | continue; |
636 | ||
084c7007 RG |
637 | if (cg_all_unified() > 0) { |
638 | if (!prog) | |
639 | continue; | |
90060676 | 640 | |
913c898c LP |
641 | (void) cgroup_bpf_whitelist_major(prog, |
642 | type == 'c' ? BPF_DEVCG_DEV_CHAR : BPF_DEVCG_DEV_BLOCK, | |
643 | maj, acc); | |
084c7007 | 644 | } else { |
084c7007 RG |
645 | sprintf(buf, |
646 | "%c %u:* %s", | |
647 | type, | |
648 | maj, | |
649 | acc); | |
650 | ||
8c838407 LP |
651 | /* Changing the devices list of a populated cgroup might result in EINVAL, hence ignore EINVAL |
652 | * here. */ | |
653 | ||
084c7007 RG |
654 | r = cg_set_attribute("devices", path, "devices.allow", buf); |
655 | if (r < 0) | |
2c74e12b | 656 | log_full_errno(IN_SET(r, -ENOENT, -EROFS, -EINVAL, -EACCES, -EPERM) ? LOG_DEBUG : LOG_WARNING, |
084c7007 RG |
657 | r, "Failed to set devices.allow on %s: %m", path); |
658 | } | |
90060676 LP |
659 | } |
660 | ||
661 | return 0; | |
90060676 LP |
662 | } |
663 | ||
66ebf6c0 TH |
664 | static bool cgroup_context_has_cpu_weight(CGroupContext *c) { |
665 | return c->cpu_weight != CGROUP_WEIGHT_INVALID || | |
666 | c->startup_cpu_weight != CGROUP_WEIGHT_INVALID; | |
667 | } | |
668 | ||
669 | static bool cgroup_context_has_cpu_shares(CGroupContext *c) { | |
670 | return c->cpu_shares != CGROUP_CPU_SHARES_INVALID || | |
671 | c->startup_cpu_shares != CGROUP_CPU_SHARES_INVALID; | |
672 | } | |
673 | ||
674 | static uint64_t cgroup_context_cpu_weight(CGroupContext *c, ManagerState state) { | |
675 | if (IN_SET(state, MANAGER_STARTING, MANAGER_INITIALIZING) && | |
676 | c->startup_cpu_weight != CGROUP_WEIGHT_INVALID) | |
677 | return c->startup_cpu_weight; | |
678 | else if (c->cpu_weight != CGROUP_WEIGHT_INVALID) | |
679 | return c->cpu_weight; | |
680 | else | |
681 | return CGROUP_WEIGHT_DEFAULT; | |
682 | } | |
683 | ||
684 | static uint64_t cgroup_context_cpu_shares(CGroupContext *c, ManagerState state) { | |
685 | if (IN_SET(state, MANAGER_STARTING, MANAGER_INITIALIZING) && | |
686 | c->startup_cpu_shares != CGROUP_CPU_SHARES_INVALID) | |
687 | return c->startup_cpu_shares; | |
688 | else if (c->cpu_shares != CGROUP_CPU_SHARES_INVALID) | |
689 | return c->cpu_shares; | |
690 | else | |
691 | return CGROUP_CPU_SHARES_DEFAULT; | |
692 | } | |
693 | ||
10f28641 FB |
694 | usec_t cgroup_cpu_adjust_period(usec_t period, usec_t quota, usec_t resolution, usec_t max_period) { |
695 | /* kernel uses a minimum resolution of 1ms, so both period and (quota * period) | |
696 | * need to be higher than that boundary. quota is specified in USecPerSec. | |
697 | * Additionally, period must be at most max_period. */ | |
698 | assert(quota > 0); | |
699 | ||
700 | return MIN(MAX3(period, resolution, resolution * USEC_PER_SEC / quota), max_period); | |
701 | } | |
702 | ||
703 | static usec_t cgroup_cpu_adjust_period_and_log(Unit *u, usec_t period, usec_t quota) { | |
704 | usec_t new_period; | |
705 | ||
706 | if (quota == USEC_INFINITY) | |
707 | /* Always use default period for infinity quota. */ | |
708 | return CGROUP_CPU_QUOTA_DEFAULT_PERIOD_USEC; | |
709 | ||
710 | if (period == USEC_INFINITY) | |
711 | /* Default period was requested. */ | |
712 | period = CGROUP_CPU_QUOTA_DEFAULT_PERIOD_USEC; | |
713 | ||
714 | /* Clamp to interval [1ms, 1s] */ | |
715 | new_period = cgroup_cpu_adjust_period(period, quota, USEC_PER_MSEC, USEC_PER_SEC); | |
716 | ||
717 | if (new_period != period) { | |
718 | char v[FORMAT_TIMESPAN_MAX]; | |
527ede0c | 719 | log_unit_full(u, u->warned_clamping_cpu_quota_period ? LOG_DEBUG : LOG_WARNING, 0, |
10f28641 FB |
720 | "Clamping CPU interval for cpu.max: period is now %s", |
721 | format_timespan(v, sizeof(v), new_period, 1)); | |
527ede0c | 722 | u->warned_clamping_cpu_quota_period = true; |
10f28641 FB |
723 | } |
724 | ||
725 | return new_period; | |
726 | } | |
727 | ||
52fecf20 LP |
728 | static void cgroup_apply_unified_cpu_weight(Unit *u, uint64_t weight) { |
729 | char buf[DECIMAL_STR_MAX(uint64_t) + 2]; | |
66ebf6c0 TH |
730 | |
731 | xsprintf(buf, "%" PRIu64 "\n", weight); | |
293d32df | 732 | (void) set_attribute_and_warn(u, "cpu", "cpu.weight", buf); |
52fecf20 LP |
733 | } |
734 | ||
10f28641 | 735 | static void cgroup_apply_unified_cpu_quota(Unit *u, usec_t quota, usec_t period) { |
52fecf20 | 736 | char buf[(DECIMAL_STR_MAX(usec_t) + 1) * 2 + 1]; |
66ebf6c0 | 737 | |
10f28641 | 738 | period = cgroup_cpu_adjust_period_and_log(u, period, quota); |
66ebf6c0 TH |
739 | if (quota != USEC_INFINITY) |
740 | xsprintf(buf, USEC_FMT " " USEC_FMT "\n", | |
10f28641 | 741 | MAX(quota * period / USEC_PER_SEC, USEC_PER_MSEC), period); |
66ebf6c0 | 742 | else |
10f28641 | 743 | xsprintf(buf, "max " USEC_FMT "\n", period); |
293d32df | 744 | (void) set_attribute_and_warn(u, "cpu", "cpu.max", buf); |
66ebf6c0 TH |
745 | } |
746 | ||
52fecf20 LP |
747 | static void cgroup_apply_legacy_cpu_shares(Unit *u, uint64_t shares) { |
748 | char buf[DECIMAL_STR_MAX(uint64_t) + 2]; | |
66ebf6c0 TH |
749 | |
750 | xsprintf(buf, "%" PRIu64 "\n", shares); | |
293d32df | 751 | (void) set_attribute_and_warn(u, "cpu", "cpu.shares", buf); |
52fecf20 LP |
752 | } |
753 | ||
10f28641 | 754 | static void cgroup_apply_legacy_cpu_quota(Unit *u, usec_t quota, usec_t period) { |
52fecf20 | 755 | char buf[DECIMAL_STR_MAX(usec_t) + 2]; |
66ebf6c0 | 756 | |
10f28641 FB |
757 | period = cgroup_cpu_adjust_period_and_log(u, period, quota); |
758 | ||
759 | xsprintf(buf, USEC_FMT "\n", period); | |
293d32df | 760 | (void) set_attribute_and_warn(u, "cpu", "cpu.cfs_period_us", buf); |
66ebf6c0 TH |
761 | |
762 | if (quota != USEC_INFINITY) { | |
10f28641 | 763 | xsprintf(buf, USEC_FMT "\n", MAX(quota * period / USEC_PER_SEC, USEC_PER_MSEC)); |
293d32df | 764 | (void) set_attribute_and_warn(u, "cpu", "cpu.cfs_quota_us", buf); |
66ebf6c0 | 765 | } else |
589a5f7a | 766 | (void) set_attribute_and_warn(u, "cpu", "cpu.cfs_quota_us", "-1\n"); |
66ebf6c0 TH |
767 | } |
768 | ||
769 | static uint64_t cgroup_cpu_shares_to_weight(uint64_t shares) { | |
770 | return CLAMP(shares * CGROUP_WEIGHT_DEFAULT / CGROUP_CPU_SHARES_DEFAULT, | |
771 | CGROUP_WEIGHT_MIN, CGROUP_WEIGHT_MAX); | |
772 | } | |
773 | ||
774 | static uint64_t cgroup_cpu_weight_to_shares(uint64_t weight) { | |
775 | return CLAMP(weight * CGROUP_CPU_SHARES_DEFAULT / CGROUP_WEIGHT_DEFAULT, | |
776 | CGROUP_CPU_SHARES_MIN, CGROUP_CPU_SHARES_MAX); | |
777 | } | |
778 | ||
508c45da | 779 | static bool cgroup_context_has_io_config(CGroupContext *c) { |
538b4852 TH |
780 | return c->io_accounting || |
781 | c->io_weight != CGROUP_WEIGHT_INVALID || | |
782 | c->startup_io_weight != CGROUP_WEIGHT_INVALID || | |
783 | c->io_device_weights || | |
6ae4283c | 784 | c->io_device_latencies || |
538b4852 TH |
785 | c->io_device_limits; |
786 | } | |
787 | ||
508c45da | 788 | static bool cgroup_context_has_blockio_config(CGroupContext *c) { |
538b4852 TH |
789 | return c->blockio_accounting || |
790 | c->blockio_weight != CGROUP_BLKIO_WEIGHT_INVALID || | |
791 | c->startup_blockio_weight != CGROUP_BLKIO_WEIGHT_INVALID || | |
792 | c->blockio_device_weights || | |
793 | c->blockio_device_bandwidths; | |
794 | } | |
795 | ||
508c45da | 796 | static uint64_t cgroup_context_io_weight(CGroupContext *c, ManagerState state) { |
64faf04c TH |
797 | if (IN_SET(state, MANAGER_STARTING, MANAGER_INITIALIZING) && |
798 | c->startup_io_weight != CGROUP_WEIGHT_INVALID) | |
799 | return c->startup_io_weight; | |
800 | else if (c->io_weight != CGROUP_WEIGHT_INVALID) | |
801 | return c->io_weight; | |
802 | else | |
803 | return CGROUP_WEIGHT_DEFAULT; | |
804 | } | |
805 | ||
508c45da | 806 | static uint64_t cgroup_context_blkio_weight(CGroupContext *c, ManagerState state) { |
64faf04c TH |
807 | if (IN_SET(state, MANAGER_STARTING, MANAGER_INITIALIZING) && |
808 | c->startup_blockio_weight != CGROUP_BLKIO_WEIGHT_INVALID) | |
809 | return c->startup_blockio_weight; | |
810 | else if (c->blockio_weight != CGROUP_BLKIO_WEIGHT_INVALID) | |
811 | return c->blockio_weight; | |
812 | else | |
813 | return CGROUP_BLKIO_WEIGHT_DEFAULT; | |
814 | } | |
815 | ||
508c45da | 816 | static uint64_t cgroup_weight_blkio_to_io(uint64_t blkio_weight) { |
538b4852 TH |
817 | return CLAMP(blkio_weight * CGROUP_WEIGHT_DEFAULT / CGROUP_BLKIO_WEIGHT_DEFAULT, |
818 | CGROUP_WEIGHT_MIN, CGROUP_WEIGHT_MAX); | |
819 | } | |
820 | ||
508c45da | 821 | static uint64_t cgroup_weight_io_to_blkio(uint64_t io_weight) { |
538b4852 TH |
822 | return CLAMP(io_weight * CGROUP_BLKIO_WEIGHT_DEFAULT / CGROUP_WEIGHT_DEFAULT, |
823 | CGROUP_BLKIO_WEIGHT_MIN, CGROUP_BLKIO_WEIGHT_MAX); | |
824 | } | |
825 | ||
f29ff115 | 826 | static void cgroup_apply_io_device_weight(Unit *u, const char *dev_path, uint64_t io_weight) { |
64faf04c TH |
827 | char buf[DECIMAL_STR_MAX(dev_t)*2+2+DECIMAL_STR_MAX(uint64_t)+1]; |
828 | dev_t dev; | |
829 | int r; | |
830 | ||
831 | r = lookup_block_device(dev_path, &dev); | |
832 | if (r < 0) | |
833 | return; | |
834 | ||
835 | xsprintf(buf, "%u:%u %" PRIu64 "\n", major(dev), minor(dev), io_weight); | |
293d32df | 836 | (void) set_attribute_and_warn(u, "io", "io.weight", buf); |
64faf04c TH |
837 | } |
838 | ||
f29ff115 | 839 | static void cgroup_apply_blkio_device_weight(Unit *u, const char *dev_path, uint64_t blkio_weight) { |
64faf04c TH |
840 | char buf[DECIMAL_STR_MAX(dev_t)*2+2+DECIMAL_STR_MAX(uint64_t)+1]; |
841 | dev_t dev; | |
842 | int r; | |
843 | ||
844 | r = lookup_block_device(dev_path, &dev); | |
845 | if (r < 0) | |
846 | return; | |
847 | ||
848 | xsprintf(buf, "%u:%u %" PRIu64 "\n", major(dev), minor(dev), blkio_weight); | |
293d32df | 849 | (void) set_attribute_and_warn(u, "blkio", "blkio.weight_device", buf); |
64faf04c TH |
850 | } |
851 | ||
6ae4283c TH |
852 | static void cgroup_apply_io_device_latency(Unit *u, const char *dev_path, usec_t target) { |
853 | char buf[DECIMAL_STR_MAX(dev_t)*2+2+7+DECIMAL_STR_MAX(uint64_t)+1]; | |
854 | dev_t dev; | |
855 | int r; | |
856 | ||
857 | r = lookup_block_device(dev_path, &dev); | |
858 | if (r < 0) | |
859 | return; | |
860 | ||
861 | if (target != USEC_INFINITY) | |
862 | xsprintf(buf, "%u:%u target=%" PRIu64 "\n", major(dev), minor(dev), target); | |
863 | else | |
864 | xsprintf(buf, "%u:%u target=max\n", major(dev), minor(dev)); | |
865 | ||
293d32df | 866 | (void) set_attribute_and_warn(u, "io", "io.latency", buf); |
6ae4283c TH |
867 | } |
868 | ||
17ae2780 | 869 | static void cgroup_apply_io_device_limit(Unit *u, const char *dev_path, uint64_t *limits) { |
64faf04c TH |
870 | char limit_bufs[_CGROUP_IO_LIMIT_TYPE_MAX][DECIMAL_STR_MAX(uint64_t)]; |
871 | char buf[DECIMAL_STR_MAX(dev_t)*2+2+(6+DECIMAL_STR_MAX(uint64_t)+1)*4]; | |
872 | CGroupIOLimitType type; | |
873 | dev_t dev; | |
64faf04c TH |
874 | int r; |
875 | ||
876 | r = lookup_block_device(dev_path, &dev); | |
877 | if (r < 0) | |
17ae2780 | 878 | return; |
64faf04c | 879 | |
17ae2780 LP |
880 | for (type = 0; type < _CGROUP_IO_LIMIT_TYPE_MAX; type++) |
881 | if (limits[type] != cgroup_io_limit_defaults[type]) | |
64faf04c | 882 | xsprintf(limit_bufs[type], "%" PRIu64, limits[type]); |
17ae2780 | 883 | else |
64faf04c | 884 | xsprintf(limit_bufs[type], "%s", limits[type] == CGROUP_LIMIT_MAX ? "max" : "0"); |
64faf04c TH |
885 | |
886 | xsprintf(buf, "%u:%u rbps=%s wbps=%s riops=%s wiops=%s\n", major(dev), minor(dev), | |
887 | limit_bufs[CGROUP_IO_RBPS_MAX], limit_bufs[CGROUP_IO_WBPS_MAX], | |
888 | limit_bufs[CGROUP_IO_RIOPS_MAX], limit_bufs[CGROUP_IO_WIOPS_MAX]); | |
293d32df | 889 | (void) set_attribute_and_warn(u, "io", "io.max", buf); |
64faf04c TH |
890 | } |
891 | ||
17ae2780 | 892 | static void cgroup_apply_blkio_device_limit(Unit *u, const char *dev_path, uint64_t rbps, uint64_t wbps) { |
64faf04c TH |
893 | char buf[DECIMAL_STR_MAX(dev_t)*2+2+DECIMAL_STR_MAX(uint64_t)+1]; |
894 | dev_t dev; | |
64faf04c TH |
895 | int r; |
896 | ||
897 | r = lookup_block_device(dev_path, &dev); | |
898 | if (r < 0) | |
17ae2780 | 899 | return; |
64faf04c | 900 | |
64faf04c | 901 | sprintf(buf, "%u:%u %" PRIu64 "\n", major(dev), minor(dev), rbps); |
293d32df | 902 | (void) set_attribute_and_warn(u, "blkio", "blkio.throttle.read_bps_device", buf); |
64faf04c | 903 | |
64faf04c | 904 | sprintf(buf, "%u:%u %" PRIu64 "\n", major(dev), minor(dev), wbps); |
293d32df | 905 | (void) set_attribute_and_warn(u, "blkio", "blkio.throttle.write_bps_device", buf); |
64faf04c TH |
906 | } |
907 | ||
c52db42b CD |
908 | static bool unit_has_unified_memory_config(Unit *u) { |
909 | CGroupContext *c; | |
910 | ||
911 | assert(u); | |
912 | ||
913 | c = unit_get_cgroup_context(u); | |
914 | assert(c); | |
915 | ||
916 | return c->memory_min > 0 || unit_get_ancestor_memory_low(u) > 0 || | |
917 | c->memory_high != CGROUP_LIMIT_MAX || c->memory_max != CGROUP_LIMIT_MAX || | |
918 | c->memory_swap_max != CGROUP_LIMIT_MAX; | |
da4d897e TH |
919 | } |
920 | ||
f29ff115 | 921 | static void cgroup_apply_unified_memory_limit(Unit *u, const char *file, uint64_t v) { |
589a5f7a | 922 | char buf[DECIMAL_STR_MAX(uint64_t) + 1] = "max\n"; |
da4d897e TH |
923 | |
924 | if (v != CGROUP_LIMIT_MAX) | |
925 | xsprintf(buf, "%" PRIu64 "\n", v); | |
926 | ||
293d32df | 927 | (void) set_attribute_and_warn(u, "memory", file, buf); |
da4d897e TH |
928 | } |
929 | ||
0f2d84d2 | 930 | static void cgroup_apply_firewall(Unit *u) { |
0f2d84d2 LP |
931 | assert(u); |
932 | ||
acf7f253 | 933 | /* Best-effort: let's apply IP firewalling and/or accounting if that's enabled */ |
906c06f6 | 934 | |
acf7f253 | 935 | if (bpf_firewall_compile(u) < 0) |
906c06f6 DM |
936 | return; |
937 | ||
938 | (void) bpf_firewall_install(u); | |
906c06f6 DM |
939 | } |
940 | ||
941 | static void cgroup_context_apply( | |
942 | Unit *u, | |
943 | CGroupMask apply_mask, | |
906c06f6 DM |
944 | ManagerState state) { |
945 | ||
f29ff115 TH |
946 | const char *path; |
947 | CGroupContext *c; | |
52fecf20 | 948 | bool is_host_root, is_local_root; |
4ad49000 LP |
949 | int r; |
950 | ||
f29ff115 TH |
951 | assert(u); |
952 | ||
906c06f6 | 953 | /* Nothing to do? Exit early! */ |
17f14955 | 954 | if (apply_mask == 0) |
4ad49000 | 955 | return; |
8e274523 | 956 | |
52fecf20 LP |
957 | /* Some cgroup attributes are not supported on the host root cgroup, hence silently ignore them here. And other |
958 | * attributes should only be managed for cgroups further down the tree. */ | |
959 | is_local_root = unit_has_name(u, SPECIAL_ROOT_SLICE); | |
960 | is_host_root = unit_has_host_root_cgroup(u); | |
f3725e64 LP |
961 | |
962 | assert_se(c = unit_get_cgroup_context(u)); | |
963 | assert_se(path = u->cgroup_path); | |
964 | ||
52fecf20 | 965 | if (is_local_root) /* Make sure we don't try to display messages with an empty path. */ |
6da13913 | 966 | path = "/"; |
01efdf13 | 967 | |
be2c0327 LP |
968 | /* We generally ignore errors caused by read-only mounted cgroup trees (assuming we are running in a container |
969 | * then), and missing cgroups, i.e. EROFS and ENOENT. */ | |
714e2e1d | 970 | |
be2c0327 LP |
971 | /* In fully unified mode these attributes don't exist on the host cgroup root. On legacy the weights exist, but |
972 | * setting the weight makes very little sense on the host root cgroup, as there are no other cgroups at this | |
973 | * level. The quota exists there too, but any attempt to write to it is refused with EINVAL. Inside of | |
4e1dfa45 | 974 | * containers we want to leave control of these to the container manager (and if cgroup v2 delegation is used |
be2c0327 LP |
975 | * we couldn't even write to them if we wanted to). */ |
976 | if ((apply_mask & CGROUP_MASK_CPU) && !is_local_root) { | |
8e274523 | 977 | |
b4cccbc1 | 978 | if (cg_all_unified() > 0) { |
be2c0327 | 979 | uint64_t weight; |
b2f8b02e | 980 | |
be2c0327 LP |
981 | if (cgroup_context_has_cpu_weight(c)) |
982 | weight = cgroup_context_cpu_weight(c, state); | |
983 | else if (cgroup_context_has_cpu_shares(c)) { | |
984 | uint64_t shares; | |
66ebf6c0 | 985 | |
be2c0327 LP |
986 | shares = cgroup_context_cpu_shares(c, state); |
987 | weight = cgroup_cpu_shares_to_weight(shares); | |
66ebf6c0 | 988 | |
be2c0327 LP |
989 | log_cgroup_compat(u, "Applying [Startup]CPUShares=%" PRIu64 " as [Startup]CPUWeight=%" PRIu64 " on %s", |
990 | shares, weight, path); | |
991 | } else | |
992 | weight = CGROUP_WEIGHT_DEFAULT; | |
66ebf6c0 | 993 | |
be2c0327 | 994 | cgroup_apply_unified_cpu_weight(u, weight); |
10f28641 | 995 | cgroup_apply_unified_cpu_quota(u, c->cpu_quota_per_sec_usec, c->cpu_quota_period_usec); |
66ebf6c0 | 996 | |
52fecf20 | 997 | } else { |
be2c0327 | 998 | uint64_t shares; |
52fecf20 | 999 | |
be2c0327 LP |
1000 | if (cgroup_context_has_cpu_weight(c)) { |
1001 | uint64_t weight; | |
52fecf20 | 1002 | |
be2c0327 LP |
1003 | weight = cgroup_context_cpu_weight(c, state); |
1004 | shares = cgroup_cpu_weight_to_shares(weight); | |
52fecf20 | 1005 | |
be2c0327 LP |
1006 | log_cgroup_compat(u, "Applying [Startup]CPUWeight=%" PRIu64 " as [Startup]CPUShares=%" PRIu64 " on %s", |
1007 | weight, shares, path); | |
1008 | } else if (cgroup_context_has_cpu_shares(c)) | |
1009 | shares = cgroup_context_cpu_shares(c, state); | |
1010 | else | |
1011 | shares = CGROUP_CPU_SHARES_DEFAULT; | |
66ebf6c0 | 1012 | |
be2c0327 | 1013 | cgroup_apply_legacy_cpu_shares(u, shares); |
10f28641 | 1014 | cgroup_apply_legacy_cpu_quota(u, c->cpu_quota_per_sec_usec, c->cpu_quota_period_usec); |
66ebf6c0 | 1015 | } |
4ad49000 LP |
1016 | } |
1017 | ||
4e1dfa45 | 1018 | /* The 'io' controller attributes are not exported on the host's root cgroup (being a pure cgroup v2 |
52fecf20 LP |
1019 | * controller), and in case of containers we want to leave control of these attributes to the container manager |
1020 | * (and we couldn't access that stuff anyway, even if we tried if proper delegation is used). */ | |
1021 | if ((apply_mask & CGROUP_MASK_IO) && !is_local_root) { | |
1022 | char buf[8+DECIMAL_STR_MAX(uint64_t)+1]; | |
1023 | bool has_io, has_blockio; | |
1024 | uint64_t weight; | |
13c31542 | 1025 | |
52fecf20 LP |
1026 | has_io = cgroup_context_has_io_config(c); |
1027 | has_blockio = cgroup_context_has_blockio_config(c); | |
13c31542 | 1028 | |
52fecf20 LP |
1029 | if (has_io) |
1030 | weight = cgroup_context_io_weight(c, state); | |
1031 | else if (has_blockio) { | |
1032 | uint64_t blkio_weight; | |
128fadc9 | 1033 | |
52fecf20 LP |
1034 | blkio_weight = cgroup_context_blkio_weight(c, state); |
1035 | weight = cgroup_weight_blkio_to_io(blkio_weight); | |
128fadc9 | 1036 | |
67e2ea15 | 1037 | log_cgroup_compat(u, "Applying [Startup]BlockIOWeight=%" PRIu64 " as [Startup]IOWeight=%" PRIu64, |
52fecf20 LP |
1038 | blkio_weight, weight); |
1039 | } else | |
1040 | weight = CGROUP_WEIGHT_DEFAULT; | |
13c31542 | 1041 | |
52fecf20 LP |
1042 | xsprintf(buf, "default %" PRIu64 "\n", weight); |
1043 | (void) set_attribute_and_warn(u, "io", "io.weight", buf); | |
538b4852 | 1044 | |
52fecf20 LP |
1045 | if (has_io) { |
1046 | CGroupIODeviceLatency *latency; | |
1047 | CGroupIODeviceLimit *limit; | |
1048 | CGroupIODeviceWeight *w; | |
128fadc9 | 1049 | |
52fecf20 LP |
1050 | LIST_FOREACH(device_weights, w, c->io_device_weights) |
1051 | cgroup_apply_io_device_weight(u, w->path, w->weight); | |
128fadc9 | 1052 | |
52fecf20 LP |
1053 | LIST_FOREACH(device_limits, limit, c->io_device_limits) |
1054 | cgroup_apply_io_device_limit(u, limit->path, limit->limits); | |
6ae4283c | 1055 | |
52fecf20 LP |
1056 | LIST_FOREACH(device_latencies, latency, c->io_device_latencies) |
1057 | cgroup_apply_io_device_latency(u, latency->path, latency->target_usec); | |
6ae4283c | 1058 | |
52fecf20 LP |
1059 | } else if (has_blockio) { |
1060 | CGroupBlockIODeviceWeight *w; | |
1061 | CGroupBlockIODeviceBandwidth *b; | |
13c31542 | 1062 | |
52fecf20 LP |
1063 | LIST_FOREACH(device_weights, w, c->blockio_device_weights) { |
1064 | weight = cgroup_weight_blkio_to_io(w->weight); | |
17ae2780 | 1065 | |
67e2ea15 | 1066 | log_cgroup_compat(u, "Applying BlockIODeviceWeight=%" PRIu64 " as IODeviceWeight=%" PRIu64 " for %s", |
52fecf20 | 1067 | w->weight, weight, w->path); |
538b4852 | 1068 | |
52fecf20 LP |
1069 | cgroup_apply_io_device_weight(u, w->path, weight); |
1070 | } | |
538b4852 | 1071 | |
17ae2780 | 1072 | LIST_FOREACH(device_bandwidths, b, c->blockio_device_bandwidths) { |
538b4852 TH |
1073 | uint64_t limits[_CGROUP_IO_LIMIT_TYPE_MAX]; |
1074 | CGroupIOLimitType type; | |
1075 | ||
1076 | for (type = 0; type < _CGROUP_IO_LIMIT_TYPE_MAX; type++) | |
1077 | limits[type] = cgroup_io_limit_defaults[type]; | |
1078 | ||
1079 | limits[CGROUP_IO_RBPS_MAX] = b->rbps; | |
1080 | limits[CGROUP_IO_WBPS_MAX] = b->wbps; | |
1081 | ||
67e2ea15 | 1082 | log_cgroup_compat(u, "Applying BlockIO{Read|Write}Bandwidth=%" PRIu64 " %" PRIu64 " as IO{Read|Write}BandwidthMax= for %s", |
128fadc9 TH |
1083 | b->rbps, b->wbps, b->path); |
1084 | ||
17ae2780 | 1085 | cgroup_apply_io_device_limit(u, b->path, limits); |
538b4852 | 1086 | } |
13c31542 TH |
1087 | } |
1088 | } | |
1089 | ||
906c06f6 | 1090 | if (apply_mask & CGROUP_MASK_BLKIO) { |
52fecf20 | 1091 | bool has_io, has_blockio; |
4ad49000 | 1092 | |
52fecf20 LP |
1093 | has_io = cgroup_context_has_io_config(c); |
1094 | has_blockio = cgroup_context_has_blockio_config(c); | |
1095 | ||
1096 | /* Applying a 'weight' never makes sense for the host root cgroup, and for containers this should be | |
1097 | * left to our container manager, too. */ | |
1098 | if (!is_local_root) { | |
64faf04c TH |
1099 | char buf[DECIMAL_STR_MAX(uint64_t)+1]; |
1100 | uint64_t weight; | |
64faf04c | 1101 | |
7d862ab8 | 1102 | if (has_io) { |
52fecf20 | 1103 | uint64_t io_weight; |
128fadc9 | 1104 | |
52fecf20 | 1105 | io_weight = cgroup_context_io_weight(c, state); |
538b4852 | 1106 | weight = cgroup_weight_io_to_blkio(cgroup_context_io_weight(c, state)); |
128fadc9 | 1107 | |
67e2ea15 | 1108 | log_cgroup_compat(u, "Applying [Startup]IOWeight=%" PRIu64 " as [Startup]BlockIOWeight=%" PRIu64, |
128fadc9 | 1109 | io_weight, weight); |
7d862ab8 TH |
1110 | } else if (has_blockio) |
1111 | weight = cgroup_context_blkio_weight(c, state); | |
1112 | else | |
538b4852 | 1113 | weight = CGROUP_BLKIO_WEIGHT_DEFAULT; |
64faf04c TH |
1114 | |
1115 | xsprintf(buf, "%" PRIu64 "\n", weight); | |
293d32df | 1116 | (void) set_attribute_and_warn(u, "blkio", "blkio.weight", buf); |
4ad49000 | 1117 | |
7d862ab8 | 1118 | if (has_io) { |
538b4852 TH |
1119 | CGroupIODeviceWeight *w; |
1120 | ||
128fadc9 TH |
1121 | LIST_FOREACH(device_weights, w, c->io_device_weights) { |
1122 | weight = cgroup_weight_io_to_blkio(w->weight); | |
1123 | ||
67e2ea15 | 1124 | log_cgroup_compat(u, "Applying IODeviceWeight=%" PRIu64 " as BlockIODeviceWeight=%" PRIu64 " for %s", |
128fadc9 TH |
1125 | w->weight, weight, w->path); |
1126 | ||
1127 | cgroup_apply_blkio_device_weight(u, w->path, weight); | |
1128 | } | |
7d862ab8 TH |
1129 | } else if (has_blockio) { |
1130 | CGroupBlockIODeviceWeight *w; | |
1131 | ||
7d862ab8 TH |
1132 | LIST_FOREACH(device_weights, w, c->blockio_device_weights) |
1133 | cgroup_apply_blkio_device_weight(u, w->path, w->weight); | |
538b4852 | 1134 | } |
4ad49000 LP |
1135 | } |
1136 | ||
52fecf20 LP |
1137 | /* The bandwith limits are something that make sense to be applied to the host's root but not container |
1138 | * roots, as there we want the container manager to handle it */ | |
1139 | if (is_host_root || !is_local_root) { | |
1140 | if (has_io) { | |
1141 | CGroupIODeviceLimit *l; | |
538b4852 | 1142 | |
52fecf20 | 1143 | LIST_FOREACH(device_limits, l, c->io_device_limits) { |
67e2ea15 | 1144 | log_cgroup_compat(u, "Applying IO{Read|Write}Bandwidth=%" PRIu64 " %" PRIu64 " as BlockIO{Read|Write}BandwidthMax= for %s", |
52fecf20 | 1145 | l->limits[CGROUP_IO_RBPS_MAX], l->limits[CGROUP_IO_WBPS_MAX], l->path); |
128fadc9 | 1146 | |
52fecf20 LP |
1147 | cgroup_apply_blkio_device_limit(u, l->path, l->limits[CGROUP_IO_RBPS_MAX], l->limits[CGROUP_IO_WBPS_MAX]); |
1148 | } | |
1149 | } else if (has_blockio) { | |
1150 | CGroupBlockIODeviceBandwidth *b; | |
7d862ab8 | 1151 | |
52fecf20 LP |
1152 | LIST_FOREACH(device_bandwidths, b, c->blockio_device_bandwidths) |
1153 | cgroup_apply_blkio_device_limit(u, b->path, b->rbps, b->wbps); | |
1154 | } | |
d686d8a9 | 1155 | } |
8e274523 LP |
1156 | } |
1157 | ||
be2c0327 LP |
1158 | /* In unified mode 'memory' attributes do not exist on the root cgroup. In legacy mode 'memory.limit_in_bytes' |
1159 | * exists on the root cgroup, but any writes to it are refused with EINVAL. And if we run in a container we | |
4e1dfa45 | 1160 | * want to leave control to the container manager (and if proper cgroup v2 delegation is used we couldn't even |
be2c0327 LP |
1161 | * write to this if we wanted to.) */ |
1162 | if ((apply_mask & CGROUP_MASK_MEMORY) && !is_local_root) { | |
efdb0237 | 1163 | |
52fecf20 | 1164 | if (cg_all_unified() > 0) { |
be2c0327 LP |
1165 | uint64_t max, swap_max = CGROUP_LIMIT_MAX; |
1166 | ||
c52db42b | 1167 | if (unit_has_unified_memory_config(u)) { |
be2c0327 LP |
1168 | max = c->memory_max; |
1169 | swap_max = c->memory_swap_max; | |
1170 | } else { | |
1171 | max = c->memory_limit; | |
efdb0237 | 1172 | |
be2c0327 LP |
1173 | if (max != CGROUP_LIMIT_MAX) |
1174 | log_cgroup_compat(u, "Applying MemoryLimit=%" PRIu64 " as MemoryMax=", max); | |
128fadc9 | 1175 | } |
da4d897e | 1176 | |
be2c0327 | 1177 | cgroup_apply_unified_memory_limit(u, "memory.min", c->memory_min); |
c52db42b | 1178 | cgroup_apply_unified_memory_limit(u, "memory.low", unit_get_ancestor_memory_low(u)); |
be2c0327 LP |
1179 | cgroup_apply_unified_memory_limit(u, "memory.high", c->memory_high); |
1180 | cgroup_apply_unified_memory_limit(u, "memory.max", max); | |
1181 | cgroup_apply_unified_memory_limit(u, "memory.swap.max", swap_max); | |
128fadc9 | 1182 | |
afcfaa69 LP |
1183 | (void) set_attribute_and_warn(u, "memory", "memory.oom.group", one_zero(c->memory_oom_group)); |
1184 | ||
be2c0327 LP |
1185 | } else { |
1186 | char buf[DECIMAL_STR_MAX(uint64_t) + 1]; | |
1187 | uint64_t val; | |
52fecf20 | 1188 | |
c52db42b | 1189 | if (unit_has_unified_memory_config(u)) { |
be2c0327 LP |
1190 | val = c->memory_max; |
1191 | log_cgroup_compat(u, "Applying MemoryMax=%" PRIi64 " as MemoryLimit=", val); | |
1192 | } else | |
1193 | val = c->memory_limit; | |
78a4ee59 | 1194 | |
be2c0327 LP |
1195 | if (val == CGROUP_LIMIT_MAX) |
1196 | strncpy(buf, "-1\n", sizeof(buf)); | |
1197 | else | |
1198 | xsprintf(buf, "%" PRIu64 "\n", val); | |
1199 | ||
1200 | (void) set_attribute_and_warn(u, "memory", "memory.limit_in_bytes", buf); | |
da4d897e | 1201 | } |
4ad49000 | 1202 | } |
8e274523 | 1203 | |
4e1dfa45 | 1204 | /* On cgroup v2 we can apply BPF everywhere. On cgroup v1 we apply it everywhere except for the root of |
52fecf20 LP |
1205 | * containers, where we leave this to the manager */ |
1206 | if ((apply_mask & (CGROUP_MASK_DEVICES | CGROUP_MASK_BPF_DEVICES)) && | |
1207 | (is_host_root || cg_all_unified() > 0 || !is_local_root)) { | |
084c7007 | 1208 | _cleanup_(bpf_program_unrefp) BPFProgram *prog = NULL; |
4ad49000 | 1209 | CGroupDeviceAllow *a; |
8e274523 | 1210 | |
084c7007 RG |
1211 | if (cg_all_unified() > 0) { |
1212 | r = cgroup_init_device_bpf(&prog, c->device_policy, c->device_allow); | |
1213 | if (r < 0) | |
1214 | log_unit_warning_errno(u, r, "Failed to initialize device control bpf program: %m"); | |
1215 | } else { | |
8c838407 | 1216 | /* Changing the devices list of a populated cgroup might result in EINVAL, hence ignore EINVAL |
084c7007 | 1217 | * here. */ |
714e2e1d | 1218 | |
084c7007 RG |
1219 | if (c->device_allow || c->device_policy != CGROUP_AUTO) |
1220 | r = cg_set_attribute("devices", path, "devices.deny", "a"); | |
1221 | else | |
1222 | r = cg_set_attribute("devices", path, "devices.allow", "a"); | |
1223 | if (r < 0) | |
2c74e12b LP |
1224 | log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EINVAL, -EACCES, -EPERM) ? LOG_DEBUG : LOG_WARNING, r, |
1225 | "Failed to reset devices.allow/devices.deny: %m"); | |
084c7007 | 1226 | } |
fb385181 | 1227 | |
4ad49000 LP |
1228 | if (c->device_policy == CGROUP_CLOSED || |
1229 | (c->device_policy == CGROUP_AUTO && c->device_allow)) { | |
1230 | static const char auto_devices[] = | |
7d711efb LP |
1231 | "/dev/null\0" "rwm\0" |
1232 | "/dev/zero\0" "rwm\0" | |
1233 | "/dev/full\0" "rwm\0" | |
1234 | "/dev/random\0" "rwm\0" | |
1235 | "/dev/urandom\0" "rwm\0" | |
1236 | "/dev/tty\0" "rwm\0" | |
5a7f87a9 | 1237 | "/dev/ptmx\0" "rwm\0" |
0d9e7991 | 1238 | /* Allow /run/systemd/inaccessible/{chr,blk} devices for mapping InaccessiblePaths */ |
74c48bf5 LP |
1239 | "/run/systemd/inaccessible/chr\0" "rwm\0" |
1240 | "/run/systemd/inaccessible/blk\0" "rwm\0"; | |
4ad49000 LP |
1241 | |
1242 | const char *x, *y; | |
1243 | ||
1244 | NULSTR_FOREACH_PAIR(x, y, auto_devices) | |
913c898c | 1245 | (void) whitelist_device(prog, path, x, y); |
7d711efb | 1246 | |
5a7f87a9 | 1247 | /* PTS (/dev/pts) devices may not be duplicated, but accessed */ |
913c898c | 1248 | (void) whitelist_major(prog, path, "pts", 'c', "rw"); |
4ad49000 LP |
1249 | } |
1250 | ||
1251 | LIST_FOREACH(device_allow, a, c->device_allow) { | |
fb4650aa | 1252 | char acc[4], *val; |
4ad49000 LP |
1253 | unsigned k = 0; |
1254 | ||
1255 | if (a->r) | |
1256 | acc[k++] = 'r'; | |
1257 | if (a->w) | |
1258 | acc[k++] = 'w'; | |
1259 | if (a->m) | |
1260 | acc[k++] = 'm'; | |
fb385181 | 1261 | |
4ad49000 LP |
1262 | if (k == 0) |
1263 | continue; | |
fb385181 | 1264 | |
4ad49000 | 1265 | acc[k++] = 0; |
90060676 | 1266 | |
27458ed6 | 1267 | if (path_startswith(a->path, "/dev/")) |
913c898c | 1268 | (void) whitelist_device(prog, path, a->path, acc); |
fb4650aa | 1269 | else if ((val = startswith(a->path, "block-"))) |
913c898c | 1270 | (void) whitelist_major(prog, path, val, 'b', acc); |
fb4650aa | 1271 | else if ((val = startswith(a->path, "char-"))) |
913c898c | 1272 | (void) whitelist_major(prog, path, val, 'c', acc); |
90060676 | 1273 | else |
8e8b5d2e | 1274 | log_unit_debug(u, "Ignoring device '%s' while writing cgroup attribute.", a->path); |
4ad49000 | 1275 | } |
084c7007 RG |
1276 | |
1277 | r = cgroup_apply_device_bpf(u, prog, c->device_policy, c->device_allow); | |
1278 | if (r < 0) { | |
1279 | static bool warned = false; | |
1280 | ||
1281 | log_full_errno(warned ? LOG_DEBUG : LOG_WARNING, r, | |
1282 | "Unit %s configures device ACL, but the local system doesn't seem to support the BPF-based device controller.\n" | |
1283 | "Proceeding WITHOUT applying ACL (all devices will be accessible)!\n" | |
1284 | "(This warning is only shown for the first loaded unit using device ACL.)", u->id); | |
1285 | ||
1286 | warned = true; | |
1287 | } | |
4ad49000 | 1288 | } |
03a7b521 | 1289 | |
00b5974f LP |
1290 | if (apply_mask & CGROUP_MASK_PIDS) { |
1291 | ||
52fecf20 | 1292 | if (is_host_root) { |
00b5974f LP |
1293 | /* So, the "pids" controller does not expose anything on the root cgroup, in order not to |
1294 | * replicate knobs exposed elsewhere needlessly. We abstract this away here however, and when | |
1295 | * the knobs of the root cgroup are modified propagate this to the relevant sysctls. There's a | |
1296 | * non-obvious asymmetry however: unlike the cgroup properties we don't really want to take | |
1297 | * exclusive ownership of the sysctls, but we still want to honour things if the user sets | |
1298 | * limits. Hence we employ sort of a one-way strategy: when the user sets a bounded limit | |
1299 | * through us it counts. When the user afterwards unsets it again (i.e. sets it to unbounded) | |
1300 | * it also counts. But if the user never set a limit through us (i.e. we are the default of | |
1301 | * "unbounded") we leave things unmodified. For this we manage a global boolean that we turn on | |
1302 | * the first time we set a limit. Note that this boolean is flushed out on manager reload, | |
1303 | * which is desirable so that there's an offical way to release control of the sysctl from | |
1304 | * systemd: set the limit to unbounded and reload. */ | |
1305 | ||
1306 | if (c->tasks_max != CGROUP_LIMIT_MAX) { | |
1307 | u->manager->sysctl_pid_max_changed = true; | |
1308 | r = procfs_tasks_set_limit(c->tasks_max); | |
1309 | } else if (u->manager->sysctl_pid_max_changed) | |
1310 | r = procfs_tasks_set_limit(TASKS_MAX); | |
1311 | else | |
1312 | r = 0; | |
00b5974f | 1313 | if (r < 0) |
39b9fefb | 1314 | log_unit_full(u, LOG_LEVEL_CGROUP_WRITE(r), r, |
00b5974f | 1315 | "Failed to write to tasks limit sysctls: %m"); |
52fecf20 | 1316 | } |
03a7b521 | 1317 | |
52fecf20 LP |
1318 | /* The attribute itself is not available on the host root cgroup, and in the container case we want to |
1319 | * leave it for the container manager. */ | |
1320 | if (!is_local_root) { | |
00b5974f LP |
1321 | if (c->tasks_max != CGROUP_LIMIT_MAX) { |
1322 | char buf[DECIMAL_STR_MAX(uint64_t) + 2]; | |
03a7b521 | 1323 | |
00b5974f | 1324 | sprintf(buf, "%" PRIu64 "\n", c->tasks_max); |
293d32df | 1325 | (void) set_attribute_and_warn(u, "pids", "pids.max", buf); |
00b5974f | 1326 | } else |
589a5f7a | 1327 | (void) set_attribute_and_warn(u, "pids", "pids.max", "max\n"); |
00b5974f | 1328 | } |
03a7b521 | 1329 | } |
906c06f6 | 1330 | |
17f14955 | 1331 | if (apply_mask & CGROUP_MASK_BPF_FIREWALL) |
0f2d84d2 | 1332 | cgroup_apply_firewall(u); |
fb385181 LP |
1333 | } |
1334 | ||
16492445 LP |
1335 | static bool unit_get_needs_bpf_firewall(Unit *u) { |
1336 | CGroupContext *c; | |
1337 | Unit *p; | |
1338 | assert(u); | |
1339 | ||
1340 | c = unit_get_cgroup_context(u); | |
1341 | if (!c) | |
1342 | return false; | |
1343 | ||
1344 | if (c->ip_accounting || | |
1345 | c->ip_address_allow || | |
1346 | c->ip_address_deny) | |
1347 | return true; | |
1348 | ||
1349 | /* If any parent slice has an IP access list defined, it applies too */ | |
1350 | for (p = UNIT_DEREF(u->slice); p; p = UNIT_DEREF(p->slice)) { | |
1351 | c = unit_get_cgroup_context(p); | |
1352 | if (!c) | |
1353 | return false; | |
1354 | ||
1355 | if (c->ip_address_allow || | |
1356 | c->ip_address_deny) | |
1357 | return true; | |
1358 | } | |
1359 | ||
1360 | return false; | |
1361 | } | |
1362 | ||
c52db42b | 1363 | static CGroupMask unit_get_cgroup_mask(Unit *u) { |
efdb0237 | 1364 | CGroupMask mask = 0; |
c52db42b CD |
1365 | CGroupContext *c; |
1366 | ||
1367 | assert(u); | |
1368 | ||
1369 | c = unit_get_cgroup_context(u); | |
8e274523 | 1370 | |
fae9bc29 | 1371 | /* Figure out which controllers we need, based on the cgroup context object */ |
8e274523 | 1372 | |
fae9bc29 | 1373 | if (c->cpu_accounting) |
f98c2585 | 1374 | mask |= get_cpu_accounting_mask(); |
fae9bc29 LP |
1375 | |
1376 | if (cgroup_context_has_cpu_weight(c) || | |
66ebf6c0 | 1377 | cgroup_context_has_cpu_shares(c) || |
3a43da28 | 1378 | c->cpu_quota_per_sec_usec != USEC_INFINITY) |
fae9bc29 | 1379 | mask |= CGROUP_MASK_CPU; |
ecedd90f | 1380 | |
538b4852 TH |
1381 | if (cgroup_context_has_io_config(c) || cgroup_context_has_blockio_config(c)) |
1382 | mask |= CGROUP_MASK_IO | CGROUP_MASK_BLKIO; | |
ecedd90f | 1383 | |
4ad49000 | 1384 | if (c->memory_accounting || |
da4d897e | 1385 | c->memory_limit != CGROUP_LIMIT_MAX || |
c52db42b | 1386 | unit_has_unified_memory_config(u)) |
efdb0237 | 1387 | mask |= CGROUP_MASK_MEMORY; |
8e274523 | 1388 | |
a931ad47 LP |
1389 | if (c->device_allow || |
1390 | c->device_policy != CGROUP_AUTO) | |
084c7007 | 1391 | mask |= CGROUP_MASK_DEVICES | CGROUP_MASK_BPF_DEVICES; |
4ad49000 | 1392 | |
03a7b521 | 1393 | if (c->tasks_accounting || |
8793fa25 | 1394 | c->tasks_max != CGROUP_LIMIT_MAX) |
03a7b521 LP |
1395 | mask |= CGROUP_MASK_PIDS; |
1396 | ||
fae9bc29 | 1397 | return CGROUP_MASK_EXTEND_JOINED(mask); |
8e274523 LP |
1398 | } |
1399 | ||
53aea74a | 1400 | static CGroupMask unit_get_bpf_mask(Unit *u) { |
17f14955 RG |
1401 | CGroupMask mask = 0; |
1402 | ||
fae9bc29 LP |
1403 | /* Figure out which controllers we need, based on the cgroup context, possibly taking into account children |
1404 | * too. */ | |
1405 | ||
17f14955 RG |
1406 | if (unit_get_needs_bpf_firewall(u)) |
1407 | mask |= CGROUP_MASK_BPF_FIREWALL; | |
1408 | ||
1409 | return mask; | |
1410 | } | |
1411 | ||
efdb0237 | 1412 | CGroupMask unit_get_own_mask(Unit *u) { |
4ad49000 | 1413 | CGroupContext *c; |
8e274523 | 1414 | |
442ce775 LP |
1415 | /* Returns the mask of controllers the unit needs for itself. If a unit is not properly loaded, return an empty |
1416 | * mask, as we shouldn't reflect it in the cgroup hierarchy then. */ | |
1417 | ||
1418 | if (u->load_state != UNIT_LOADED) | |
1419 | return 0; | |
efdb0237 | 1420 | |
4ad49000 LP |
1421 | c = unit_get_cgroup_context(u); |
1422 | if (!c) | |
1423 | return 0; | |
8e274523 | 1424 | |
c52db42b | 1425 | return (unit_get_cgroup_mask(u) | unit_get_bpf_mask(u) | unit_get_delegate_mask(u)) & ~unit_get_ancestor_disable_mask(u); |
02638280 LP |
1426 | } |
1427 | ||
1428 | CGroupMask unit_get_delegate_mask(Unit *u) { | |
1429 | CGroupContext *c; | |
1430 | ||
1431 | /* If delegation is turned on, then turn on selected controllers, unless we are on the legacy hierarchy and the | |
1432 | * process we fork into is known to drop privileges, and hence shouldn't get access to the controllers. | |
19af675e | 1433 | * |
02638280 | 1434 | * Note that on the unified hierarchy it is safe to delegate controllers to unprivileged services. */ |
a931ad47 | 1435 | |
1d9cc876 | 1436 | if (!unit_cgroup_delegate(u)) |
02638280 LP |
1437 | return 0; |
1438 | ||
1439 | if (cg_all_unified() <= 0) { | |
a931ad47 LP |
1440 | ExecContext *e; |
1441 | ||
1442 | e = unit_get_exec_context(u); | |
02638280 LP |
1443 | if (e && !exec_context_maintains_privileges(e)) |
1444 | return 0; | |
a931ad47 LP |
1445 | } |
1446 | ||
1d9cc876 | 1447 | assert_se(c = unit_get_cgroup_context(u)); |
fae9bc29 | 1448 | return CGROUP_MASK_EXTEND_JOINED(c->delegate_controllers); |
8e274523 LP |
1449 | } |
1450 | ||
efdb0237 | 1451 | CGroupMask unit_get_members_mask(Unit *u) { |
4ad49000 | 1452 | assert(u); |
bc432dc7 | 1453 | |
02638280 | 1454 | /* Returns the mask of controllers all of the unit's children require, merged */ |
efdb0237 | 1455 | |
bc432dc7 | 1456 | if (u->cgroup_members_mask_valid) |
26a17ca2 | 1457 | return u->cgroup_members_mask; /* Use cached value if possible */ |
bc432dc7 | 1458 | |
64e844e5 | 1459 | u->cgroup_members_mask = 0; |
bc432dc7 LP |
1460 | |
1461 | if (u->type == UNIT_SLICE) { | |
eef85c4a | 1462 | void *v; |
bc432dc7 LP |
1463 | Unit *member; |
1464 | Iterator i; | |
1465 | ||
eef85c4a | 1466 | HASHMAP_FOREACH_KEY(v, member, u->dependencies[UNIT_BEFORE], i) { |
cb5e3bc3 CD |
1467 | if (UNIT_DEREF(member->slice) == u) |
1468 | u->cgroup_members_mask |= unit_get_subtree_mask(member); /* note that this calls ourselves again, for the children */ | |
bc432dc7 LP |
1469 | } |
1470 | } | |
1471 | ||
1472 | u->cgroup_members_mask_valid = true; | |
6414b7c9 | 1473 | return u->cgroup_members_mask; |
246aa6dd LP |
1474 | } |
1475 | ||
efdb0237 | 1476 | CGroupMask unit_get_siblings_mask(Unit *u) { |
4ad49000 | 1477 | assert(u); |
246aa6dd | 1478 | |
efdb0237 LP |
1479 | /* Returns the mask of controllers all of the unit's siblings |
1480 | * require, i.e. the members mask of the unit's parent slice | |
1481 | * if there is one. */ | |
1482 | ||
bc432dc7 | 1483 | if (UNIT_ISSET(u->slice)) |
637f421e | 1484 | return unit_get_members_mask(UNIT_DEREF(u->slice)); |
4ad49000 | 1485 | |
64e844e5 | 1486 | return unit_get_subtree_mask(u); /* we are the top-level slice */ |
246aa6dd LP |
1487 | } |
1488 | ||
4f6f62e4 CD |
1489 | CGroupMask unit_get_disable_mask(Unit *u) { |
1490 | CGroupContext *c; | |
1491 | ||
1492 | c = unit_get_cgroup_context(u); | |
1493 | if (!c) | |
1494 | return 0; | |
1495 | ||
1496 | return c->disable_controllers; | |
1497 | } | |
1498 | ||
1499 | CGroupMask unit_get_ancestor_disable_mask(Unit *u) { | |
1500 | CGroupMask mask; | |
1501 | ||
1502 | assert(u); | |
1503 | mask = unit_get_disable_mask(u); | |
1504 | ||
1505 | /* Returns the mask of controllers which are marked as forcibly | |
1506 | * disabled in any ancestor unit or the unit in question. */ | |
1507 | ||
1508 | if (UNIT_ISSET(u->slice)) | |
1509 | mask |= unit_get_ancestor_disable_mask(UNIT_DEREF(u->slice)); | |
1510 | ||
1511 | return mask; | |
1512 | } | |
1513 | ||
efdb0237 LP |
1514 | CGroupMask unit_get_subtree_mask(Unit *u) { |
1515 | ||
1516 | /* Returns the mask of this subtree, meaning of the group | |
1517 | * itself and its children. */ | |
1518 | ||
1519 | return unit_get_own_mask(u) | unit_get_members_mask(u); | |
1520 | } | |
1521 | ||
1522 | CGroupMask unit_get_target_mask(Unit *u) { | |
1523 | CGroupMask mask; | |
1524 | ||
1525 | /* This returns the cgroup mask of all controllers to enable | |
1526 | * for a specific cgroup, i.e. everything it needs itself, | |
1527 | * plus all that its children need, plus all that its siblings | |
1528 | * need. This is primarily useful on the legacy cgroup | |
1529 | * hierarchy, where we need to duplicate each cgroup in each | |
1530 | * hierarchy that shall be enabled for it. */ | |
6414b7c9 | 1531 | |
efdb0237 LP |
1532 | mask = unit_get_own_mask(u) | unit_get_members_mask(u) | unit_get_siblings_mask(u); |
1533 | mask &= u->manager->cgroup_supported; | |
c72703e2 | 1534 | mask &= ~unit_get_ancestor_disable_mask(u); |
efdb0237 LP |
1535 | |
1536 | return mask; | |
1537 | } | |
1538 | ||
1539 | CGroupMask unit_get_enable_mask(Unit *u) { | |
1540 | CGroupMask mask; | |
1541 | ||
1542 | /* This returns the cgroup mask of all controllers to enable | |
1543 | * for the children of a specific cgroup. This is primarily | |
1544 | * useful for the unified cgroup hierarchy, where each cgroup | |
1545 | * controls which controllers are enabled for its children. */ | |
1546 | ||
1547 | mask = unit_get_members_mask(u); | |
6414b7c9 | 1548 | mask &= u->manager->cgroup_supported; |
c72703e2 | 1549 | mask &= ~unit_get_ancestor_disable_mask(u); |
6414b7c9 DS |
1550 | |
1551 | return mask; | |
1552 | } | |
1553 | ||
5af88058 | 1554 | void unit_invalidate_cgroup_members_masks(Unit *u) { |
bc432dc7 LP |
1555 | assert(u); |
1556 | ||
5af88058 LP |
1557 | /* Recurse invalidate the member masks cache all the way up the tree */ |
1558 | u->cgroup_members_mask_valid = false; | |
bc432dc7 | 1559 | |
5af88058 LP |
1560 | if (UNIT_ISSET(u->slice)) |
1561 | unit_invalidate_cgroup_members_masks(UNIT_DEREF(u->slice)); | |
6414b7c9 DS |
1562 | } |
1563 | ||
6592b975 | 1564 | const char *unit_get_realized_cgroup_path(Unit *u, CGroupMask mask) { |
03b90d4b | 1565 | |
6592b975 | 1566 | /* Returns the realized cgroup path of the specified unit where all specified controllers are available. */ |
03b90d4b LP |
1567 | |
1568 | while (u) { | |
6592b975 | 1569 | |
03b90d4b LP |
1570 | if (u->cgroup_path && |
1571 | u->cgroup_realized && | |
d94a24ca | 1572 | FLAGS_SET(u->cgroup_realized_mask, mask)) |
03b90d4b LP |
1573 | return u->cgroup_path; |
1574 | ||
1575 | u = UNIT_DEREF(u->slice); | |
1576 | } | |
1577 | ||
1578 | return NULL; | |
1579 | } | |
1580 | ||
6592b975 LP |
1581 | static const char *migrate_callback(CGroupMask mask, void *userdata) { |
1582 | return unit_get_realized_cgroup_path(userdata, mask); | |
1583 | } | |
1584 | ||
303ee601 | 1585 | char *unit_default_cgroup_path(const Unit *u) { |
efdb0237 LP |
1586 | _cleanup_free_ char *escaped = NULL, *slice = NULL; |
1587 | int r; | |
1588 | ||
1589 | assert(u); | |
1590 | ||
1591 | if (unit_has_name(u, SPECIAL_ROOT_SLICE)) | |
1592 | return strdup(u->manager->cgroup_root); | |
1593 | ||
1594 | if (UNIT_ISSET(u->slice) && !unit_has_name(UNIT_DEREF(u->slice), SPECIAL_ROOT_SLICE)) { | |
1595 | r = cg_slice_to_path(UNIT_DEREF(u->slice)->id, &slice); | |
1596 | if (r < 0) | |
1597 | return NULL; | |
1598 | } | |
1599 | ||
1600 | escaped = cg_escape(u->id); | |
1601 | if (!escaped) | |
1602 | return NULL; | |
1603 | ||
1604 | if (slice) | |
605405c6 ZJS |
1605 | return strjoin(u->manager->cgroup_root, "/", slice, "/", |
1606 | escaped); | |
efdb0237 | 1607 | else |
605405c6 | 1608 | return strjoin(u->manager->cgroup_root, "/", escaped); |
efdb0237 LP |
1609 | } |
1610 | ||
1611 | int unit_set_cgroup_path(Unit *u, const char *path) { | |
1612 | _cleanup_free_ char *p = NULL; | |
1613 | int r; | |
1614 | ||
1615 | assert(u); | |
1616 | ||
5210387e LP |
1617 | if (streq_ptr(u->cgroup_path, path)) |
1618 | return 0; | |
1619 | ||
efdb0237 LP |
1620 | if (path) { |
1621 | p = strdup(path); | |
1622 | if (!p) | |
1623 | return -ENOMEM; | |
5210387e | 1624 | } |
efdb0237 LP |
1625 | |
1626 | if (p) { | |
1627 | r = hashmap_put(u->manager->cgroup_unit, p, u); | |
1628 | if (r < 0) | |
1629 | return r; | |
1630 | } | |
1631 | ||
1632 | unit_release_cgroup(u); | |
ae2a15bc | 1633 | u->cgroup_path = TAKE_PTR(p); |
efdb0237 LP |
1634 | |
1635 | return 1; | |
1636 | } | |
1637 | ||
1638 | int unit_watch_cgroup(Unit *u) { | |
ab2c3861 | 1639 | _cleanup_free_ char *events = NULL; |
efdb0237 LP |
1640 | int r; |
1641 | ||
1642 | assert(u); | |
1643 | ||
0bb814c2 LP |
1644 | /* Watches the "cgroups.events" attribute of this unit's cgroup for "empty" events, but only if |
1645 | * cgroupv2 is available. */ | |
1646 | ||
efdb0237 LP |
1647 | if (!u->cgroup_path) |
1648 | return 0; | |
1649 | ||
0bb814c2 | 1650 | if (u->cgroup_control_inotify_wd >= 0) |
efdb0237 LP |
1651 | return 0; |
1652 | ||
1653 | /* Only applies to the unified hierarchy */ | |
c22800e4 | 1654 | r = cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER); |
b4cccbc1 LP |
1655 | if (r < 0) |
1656 | return log_error_errno(r, "Failed to determine whether the name=systemd hierarchy is unified: %m"); | |
1657 | if (r == 0) | |
efdb0237 LP |
1658 | return 0; |
1659 | ||
0bb814c2 | 1660 | /* No point in watch the top-level slice, it's never going to run empty. */ |
efdb0237 LP |
1661 | if (unit_has_name(u, SPECIAL_ROOT_SLICE)) |
1662 | return 0; | |
1663 | ||
0bb814c2 | 1664 | r = hashmap_ensure_allocated(&u->manager->cgroup_control_inotify_wd_unit, &trivial_hash_ops); |
efdb0237 LP |
1665 | if (r < 0) |
1666 | return log_oom(); | |
1667 | ||
ab2c3861 | 1668 | r = cg_get_path(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path, "cgroup.events", &events); |
efdb0237 LP |
1669 | if (r < 0) |
1670 | return log_oom(); | |
1671 | ||
0bb814c2 LP |
1672 | u->cgroup_control_inotify_wd = inotify_add_watch(u->manager->cgroup_inotify_fd, events, IN_MODIFY); |
1673 | if (u->cgroup_control_inotify_wd < 0) { | |
efdb0237 | 1674 | |
0bb814c2 LP |
1675 | if (errno == ENOENT) /* If the directory is already gone we don't need to track it, so this |
1676 | * is not an error */ | |
efdb0237 LP |
1677 | return 0; |
1678 | ||
0bb814c2 | 1679 | return log_unit_error_errno(u, errno, "Failed to add control inotify watch descriptor for control group %s: %m", u->cgroup_path); |
efdb0237 LP |
1680 | } |
1681 | ||
0bb814c2 | 1682 | r = hashmap_put(u->manager->cgroup_control_inotify_wd_unit, INT_TO_PTR(u->cgroup_control_inotify_wd), u); |
efdb0237 | 1683 | if (r < 0) |
0bb814c2 | 1684 | return log_unit_error_errno(u, r, "Failed to add control inotify watch descriptor to hash map: %m"); |
efdb0237 LP |
1685 | |
1686 | return 0; | |
1687 | } | |
1688 | ||
afcfaa69 LP |
1689 | int unit_watch_cgroup_memory(Unit *u) { |
1690 | _cleanup_free_ char *events = NULL; | |
1691 | CGroupContext *c; | |
1692 | int r; | |
1693 | ||
1694 | assert(u); | |
1695 | ||
1696 | /* Watches the "memory.events" attribute of this unit's cgroup for "oom_kill" events, but only if | |
1697 | * cgroupv2 is available. */ | |
1698 | ||
1699 | if (!u->cgroup_path) | |
1700 | return 0; | |
1701 | ||
1702 | c = unit_get_cgroup_context(u); | |
1703 | if (!c) | |
1704 | return 0; | |
1705 | ||
1706 | /* The "memory.events" attribute is only available if the memory controller is on. Let's hence tie | |
1707 | * this to memory accounting, in a way watching for OOM kills is a form of memory accounting after | |
1708 | * all. */ | |
1709 | if (!c->memory_accounting) | |
1710 | return 0; | |
1711 | ||
1712 | /* Don't watch inner nodes, as the kernel doesn't report oom_kill events recursively currently, and | |
1713 | * we also don't want to generate a log message for each parent cgroup of a process. */ | |
1714 | if (u->type == UNIT_SLICE) | |
1715 | return 0; | |
1716 | ||
1717 | if (u->cgroup_memory_inotify_wd >= 0) | |
1718 | return 0; | |
1719 | ||
1720 | /* Only applies to the unified hierarchy */ | |
1721 | r = cg_all_unified(); | |
1722 | if (r < 0) | |
1723 | return log_error_errno(r, "Failed to determine whether the memory controller is unified: %m"); | |
1724 | if (r == 0) | |
1725 | return 0; | |
1726 | ||
1727 | r = hashmap_ensure_allocated(&u->manager->cgroup_memory_inotify_wd_unit, &trivial_hash_ops); | |
1728 | if (r < 0) | |
1729 | return log_oom(); | |
1730 | ||
1731 | r = cg_get_path(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path, "memory.events", &events); | |
1732 | if (r < 0) | |
1733 | return log_oom(); | |
1734 | ||
1735 | u->cgroup_memory_inotify_wd = inotify_add_watch(u->manager->cgroup_inotify_fd, events, IN_MODIFY); | |
1736 | if (u->cgroup_memory_inotify_wd < 0) { | |
1737 | ||
1738 | if (errno == ENOENT) /* If the directory is already gone we don't need to track it, so this | |
1739 | * is not an error */ | |
1740 | return 0; | |
1741 | ||
1742 | return log_unit_error_errno(u, errno, "Failed to add memory inotify watch descriptor for control group %s: %m", u->cgroup_path); | |
1743 | } | |
1744 | ||
1745 | r = hashmap_put(u->manager->cgroup_memory_inotify_wd_unit, INT_TO_PTR(u->cgroup_memory_inotify_wd), u); | |
1746 | if (r < 0) | |
1747 | return log_unit_error_errno(u, r, "Failed to add memory inotify watch descriptor to hash map: %m"); | |
1748 | ||
1749 | return 0; | |
1750 | } | |
1751 | ||
a4634b21 LP |
1752 | int unit_pick_cgroup_path(Unit *u) { |
1753 | _cleanup_free_ char *path = NULL; | |
1754 | int r; | |
1755 | ||
1756 | assert(u); | |
1757 | ||
1758 | if (u->cgroup_path) | |
1759 | return 0; | |
1760 | ||
1761 | if (!UNIT_HAS_CGROUP_CONTEXT(u)) | |
1762 | return -EINVAL; | |
1763 | ||
1764 | path = unit_default_cgroup_path(u); | |
1765 | if (!path) | |
1766 | return log_oom(); | |
1767 | ||
1768 | r = unit_set_cgroup_path(u, path); | |
1769 | if (r == -EEXIST) | |
1770 | return log_unit_error_errno(u, r, "Control group %s exists already.", path); | |
1771 | if (r < 0) | |
1772 | return log_unit_error_errno(u, r, "Failed to set unit's control group path to %s: %m", path); | |
1773 | ||
1774 | return 0; | |
1775 | } | |
1776 | ||
efdb0237 LP |
1777 | static int unit_create_cgroup( |
1778 | Unit *u, | |
1779 | CGroupMask target_mask, | |
0d2d6fbf CD |
1780 | CGroupMask enable_mask, |
1781 | ManagerState state) { | |
efdb0237 | 1782 | |
65be7e06 | 1783 | bool created; |
27adcc97 | 1784 | int r; |
64747e2d | 1785 | |
4ad49000 | 1786 | assert(u); |
64747e2d | 1787 | |
27c4ed79 | 1788 | if (!UNIT_HAS_CGROUP_CONTEXT(u)) |
0cd385d3 LP |
1789 | return 0; |
1790 | ||
a4634b21 LP |
1791 | /* Figure out our cgroup path */ |
1792 | r = unit_pick_cgroup_path(u); | |
1793 | if (r < 0) | |
1794 | return r; | |
b58b8e11 | 1795 | |
03b90d4b | 1796 | /* First, create our own group */ |
efdb0237 | 1797 | r = cg_create_everywhere(u->manager->cgroup_supported, target_mask, u->cgroup_path); |
23bbb0de | 1798 | if (r < 0) |
efdb0237 | 1799 | return log_unit_error_errno(u, r, "Failed to create cgroup %s: %m", u->cgroup_path); |
490c5a37 | 1800 | created = r; |
efdb0237 LP |
1801 | |
1802 | /* Start watching it */ | |
1803 | (void) unit_watch_cgroup(u); | |
afcfaa69 | 1804 | (void) unit_watch_cgroup_memory(u); |
efdb0237 | 1805 | |
65be7e06 | 1806 | /* Preserve enabled controllers in delegated units, adjust others. */ |
1fd3a10c | 1807 | if (created || !u->cgroup_realized || !unit_cgroup_delegate(u)) { |
27adcc97 | 1808 | CGroupMask result_mask = 0; |
65be7e06 ZJS |
1809 | |
1810 | /* Enable all controllers we need */ | |
27adcc97 | 1811 | r = cg_enable_everywhere(u->manager->cgroup_supported, enable_mask, u->cgroup_path, &result_mask); |
65be7e06 | 1812 | if (r < 0) |
27adcc97 LP |
1813 | log_unit_warning_errno(u, r, "Failed to enable/disable controllers on cgroup %s, ignoring: %m", u->cgroup_path); |
1814 | ||
1815 | /* If we just turned off a controller, this might release the controller for our parent too, let's | |
1816 | * enqueue the parent for re-realization in that case again. */ | |
1817 | if (UNIT_ISSET(u->slice)) { | |
1818 | CGroupMask turned_off; | |
1819 | ||
1820 | turned_off = (u->cgroup_realized ? u->cgroup_enabled_mask & ~result_mask : 0); | |
1821 | if (turned_off != 0) { | |
1822 | Unit *parent; | |
1823 | ||
1824 | /* Force the parent to propagate the enable mask to the kernel again, by invalidating | |
1825 | * the controller we just turned off. */ | |
1826 | ||
1827 | for (parent = UNIT_DEREF(u->slice); parent; parent = UNIT_DEREF(parent->slice)) | |
1828 | unit_invalidate_cgroup(parent, turned_off); | |
1829 | } | |
1830 | } | |
1831 | ||
1832 | /* Remember what's actually enabled now */ | |
1833 | u->cgroup_enabled_mask = result_mask; | |
65be7e06 | 1834 | } |
03b90d4b LP |
1835 | |
1836 | /* Keep track that this is now realized */ | |
4ad49000 | 1837 | u->cgroup_realized = true; |
efdb0237 | 1838 | u->cgroup_realized_mask = target_mask; |
4ad49000 | 1839 | |
1d9cc876 | 1840 | if (u->type != UNIT_SLICE && !unit_cgroup_delegate(u)) { |
0cd385d3 LP |
1841 | |
1842 | /* Then, possibly move things over, but not if | |
1843 | * subgroups may contain processes, which is the case | |
1844 | * for slice and delegation units. */ | |
1845 | r = cg_migrate_everywhere(u->manager->cgroup_supported, u->cgroup_path, u->cgroup_path, migrate_callback, u); | |
1846 | if (r < 0) | |
efdb0237 | 1847 | log_unit_warning_errno(u, r, "Failed to migrate cgroup from to %s, ignoring: %m", u->cgroup_path); |
0cd385d3 | 1848 | } |
03b90d4b | 1849 | |
0d2d6fbf CD |
1850 | /* Set attributes */ |
1851 | cgroup_context_apply(u, target_mask, state); | |
1852 | cgroup_xattr_apply(u); | |
1853 | ||
64747e2d LP |
1854 | return 0; |
1855 | } | |
1856 | ||
6592b975 LP |
1857 | static int unit_attach_pid_to_cgroup_via_bus(Unit *u, pid_t pid, const char *suffix_path) { |
1858 | _cleanup_(sd_bus_error_free) sd_bus_error error = SD_BUS_ERROR_NULL; | |
1859 | char *pp; | |
7b3fd631 | 1860 | int r; |
6592b975 | 1861 | |
7b3fd631 LP |
1862 | assert(u); |
1863 | ||
6592b975 LP |
1864 | if (MANAGER_IS_SYSTEM(u->manager)) |
1865 | return -EINVAL; | |
1866 | ||
1867 | if (!u->manager->system_bus) | |
1868 | return -EIO; | |
1869 | ||
1870 | if (!u->cgroup_path) | |
1871 | return -EINVAL; | |
1872 | ||
1873 | /* Determine this unit's cgroup path relative to our cgroup root */ | |
1874 | pp = path_startswith(u->cgroup_path, u->manager->cgroup_root); | |
1875 | if (!pp) | |
1876 | return -EINVAL; | |
1877 | ||
1878 | pp = strjoina("/", pp, suffix_path); | |
858d36c1 | 1879 | path_simplify(pp, false); |
6592b975 LP |
1880 | |
1881 | r = sd_bus_call_method(u->manager->system_bus, | |
1882 | "org.freedesktop.systemd1", | |
1883 | "/org/freedesktop/systemd1", | |
1884 | "org.freedesktop.systemd1.Manager", | |
1885 | "AttachProcessesToUnit", | |
1886 | &error, NULL, | |
1887 | "ssau", | |
1888 | NULL /* empty unit name means client's unit, i.e. us */, pp, 1, (uint32_t) pid); | |
7b3fd631 | 1889 | if (r < 0) |
6592b975 LP |
1890 | return log_unit_debug_errno(u, r, "Failed to attach unit process " PID_FMT " via the bus: %s", pid, bus_error_message(&error, r)); |
1891 | ||
1892 | return 0; | |
1893 | } | |
1894 | ||
1895 | int unit_attach_pids_to_cgroup(Unit *u, Set *pids, const char *suffix_path) { | |
1896 | CGroupMask delegated_mask; | |
1897 | const char *p; | |
1898 | Iterator i; | |
1899 | void *pidp; | |
1900 | int r, q; | |
1901 | ||
1902 | assert(u); | |
1903 | ||
1904 | if (!UNIT_HAS_CGROUP_CONTEXT(u)) | |
1905 | return -EINVAL; | |
1906 | ||
1907 | if (set_isempty(pids)) | |
1908 | return 0; | |
7b3fd631 | 1909 | |
6592b975 | 1910 | r = unit_realize_cgroup(u); |
7b3fd631 LP |
1911 | if (r < 0) |
1912 | return r; | |
1913 | ||
6592b975 LP |
1914 | if (isempty(suffix_path)) |
1915 | p = u->cgroup_path; | |
1916 | else | |
1917 | p = strjoina(u->cgroup_path, "/", suffix_path); | |
1918 | ||
1919 | delegated_mask = unit_get_delegate_mask(u); | |
1920 | ||
1921 | r = 0; | |
1922 | SET_FOREACH(pidp, pids, i) { | |
1923 | pid_t pid = PTR_TO_PID(pidp); | |
1924 | CGroupController c; | |
1925 | ||
1926 | /* First, attach the PID to the main cgroup hierarchy */ | |
1927 | q = cg_attach(SYSTEMD_CGROUP_CONTROLLER, p, pid); | |
1928 | if (q < 0) { | |
1929 | log_unit_debug_errno(u, q, "Couldn't move process " PID_FMT " to requested cgroup '%s': %m", pid, p); | |
1930 | ||
1931 | if (MANAGER_IS_USER(u->manager) && IN_SET(q, -EPERM, -EACCES)) { | |
1932 | int z; | |
1933 | ||
1934 | /* If we are in a user instance, and we can't move the process ourselves due to | |
1935 | * permission problems, let's ask the system instance about it instead. Since it's more | |
1936 | * privileged it might be able to move the process across the leaves of a subtree who's | |
1937 | * top node is not owned by us. */ | |
1938 | ||
1939 | z = unit_attach_pid_to_cgroup_via_bus(u, pid, suffix_path); | |
1940 | if (z < 0) | |
1941 | log_unit_debug_errno(u, z, "Couldn't move process " PID_FMT " to requested cgroup '%s' via the system bus either: %m", pid, p); | |
1942 | else | |
1943 | continue; /* When the bus thing worked via the bus we are fully done for this PID. */ | |
1944 | } | |
1945 | ||
1946 | if (r >= 0) | |
1947 | r = q; /* Remember first error */ | |
1948 | ||
1949 | continue; | |
1950 | } | |
1951 | ||
1952 | q = cg_all_unified(); | |
1953 | if (q < 0) | |
1954 | return q; | |
1955 | if (q > 0) | |
1956 | continue; | |
1957 | ||
1958 | /* In the legacy hierarchy, attach the process to the request cgroup if possible, and if not to the | |
1959 | * innermost realized one */ | |
1960 | ||
1961 | for (c = 0; c < _CGROUP_CONTROLLER_MAX; c++) { | |
1962 | CGroupMask bit = CGROUP_CONTROLLER_TO_MASK(c); | |
1963 | const char *realized; | |
1964 | ||
1965 | if (!(u->manager->cgroup_supported & bit)) | |
1966 | continue; | |
1967 | ||
1968 | /* If this controller is delegated and realized, honour the caller's request for the cgroup suffix. */ | |
1969 | if (delegated_mask & u->cgroup_realized_mask & bit) { | |
1970 | q = cg_attach(cgroup_controller_to_string(c), p, pid); | |
1971 | if (q >= 0) | |
1972 | continue; /* Success! */ | |
1973 | ||
1974 | 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", | |
1975 | pid, p, cgroup_controller_to_string(c)); | |
1976 | } | |
1977 | ||
1978 | /* So this controller is either not delegate or realized, or something else weird happened. In | |
1979 | * that case let's attach the PID at least to the closest cgroup up the tree that is | |
1980 | * realized. */ | |
1981 | realized = unit_get_realized_cgroup_path(u, bit); | |
1982 | if (!realized) | |
1983 | continue; /* Not even realized in the root slice? Then let's not bother */ | |
1984 | ||
1985 | q = cg_attach(cgroup_controller_to_string(c), realized, pid); | |
1986 | if (q < 0) | |
1987 | log_unit_debug_errno(u, q, "Failed to attach PID " PID_FMT " to realized cgroup %s in controller %s, ignoring: %m", | |
1988 | pid, realized, cgroup_controller_to_string(c)); | |
1989 | } | |
1990 | } | |
1991 | ||
1992 | return r; | |
7b3fd631 LP |
1993 | } |
1994 | ||
906c06f6 DM |
1995 | static bool unit_has_mask_realized( |
1996 | Unit *u, | |
1997 | CGroupMask target_mask, | |
17f14955 | 1998 | CGroupMask enable_mask) { |
906c06f6 | 1999 | |
bc432dc7 LP |
2000 | assert(u); |
2001 | ||
d5095dcd LP |
2002 | /* Returns true if this unit is fully realized. We check four things: |
2003 | * | |
2004 | * 1. Whether the cgroup was created at all | |
4e1dfa45 CD |
2005 | * 2. Whether the cgroup was created in all the hierarchies we need it to be created in (in case of cgroup v1) |
2006 | * 3. Whether the cgroup has all the right controllers enabled (in case of cgroup v2) | |
d5095dcd LP |
2007 | * 4. Whether the invalidation mask is currently zero |
2008 | * | |
2009 | * If you wonder why we mask the target realization and enable mask with CGROUP_MASK_V1/CGROUP_MASK_V2: note | |
4e1dfa45 CD |
2010 | * that there are three sets of bitmasks: CGROUP_MASK_V1 (for real cgroup v1 controllers), CGROUP_MASK_V2 (for |
2011 | * real cgroup v2 controllers) and CGROUP_MASK_BPF (for BPF-based pseudo-controllers). Now, cgroup_realized_mask | |
2012 | * is only matters for cgroup v1 controllers, and cgroup_enabled_mask only used for cgroup v2, and if they | |
d5095dcd LP |
2013 | * differ in the others, we don't really care. (After all, the cgroup_enabled_mask tracks with controllers are |
2014 | * enabled through cgroup.subtree_control, and since the BPF pseudo-controllers don't show up there, they | |
2015 | * simply don't matter. */ | |
2016 | ||
906c06f6 | 2017 | return u->cgroup_realized && |
d5095dcd LP |
2018 | ((u->cgroup_realized_mask ^ target_mask) & CGROUP_MASK_V1) == 0 && |
2019 | ((u->cgroup_enabled_mask ^ enable_mask) & CGROUP_MASK_V2) == 0 && | |
17f14955 | 2020 | u->cgroup_invalidated_mask == 0; |
6414b7c9 DS |
2021 | } |
2022 | ||
4f6f62e4 CD |
2023 | static bool unit_has_mask_disables_realized( |
2024 | Unit *u, | |
2025 | CGroupMask target_mask, | |
2026 | CGroupMask enable_mask) { | |
2027 | ||
2028 | assert(u); | |
2029 | ||
2030 | /* Returns true if all controllers which should be disabled are indeed disabled. | |
2031 | * | |
2032 | * Unlike unit_has_mask_realized, we don't care what was enabled, only that anything we want to remove is | |
2033 | * already removed. */ | |
2034 | ||
2035 | return !u->cgroup_realized || | |
2036 | (FLAGS_SET(u->cgroup_realized_mask, target_mask & CGROUP_MASK_V1) && | |
2037 | FLAGS_SET(u->cgroup_enabled_mask, enable_mask & CGROUP_MASK_V2)); | |
2038 | } | |
2039 | ||
a57669d2 CD |
2040 | static bool unit_has_mask_enables_realized( |
2041 | Unit *u, | |
2042 | CGroupMask target_mask, | |
2043 | CGroupMask enable_mask) { | |
2044 | ||
2045 | assert(u); | |
2046 | ||
2047 | /* Returns true if all controllers which should be enabled are indeed enabled. | |
2048 | * | |
2049 | * Unlike unit_has_mask_realized, we don't care about the controllers that are not present, only that anything | |
2050 | * we want to add is already added. */ | |
2051 | ||
2052 | return u->cgroup_realized && | |
c72703e2 CD |
2053 | ((u->cgroup_realized_mask | target_mask) & CGROUP_MASK_V1) == (u->cgroup_realized_mask & CGROUP_MASK_V1) && |
2054 | ((u->cgroup_enabled_mask | enable_mask) & CGROUP_MASK_V2) == (u->cgroup_enabled_mask & CGROUP_MASK_V2); | |
a57669d2 CD |
2055 | } |
2056 | ||
27adcc97 | 2057 | void unit_add_to_cgroup_realize_queue(Unit *u) { |
2aa57a65 LP |
2058 | assert(u); |
2059 | ||
2060 | if (u->in_cgroup_realize_queue) | |
2061 | return; | |
2062 | ||
2063 | LIST_PREPEND(cgroup_realize_queue, u->manager->cgroup_realize_queue, u); | |
2064 | u->in_cgroup_realize_queue = true; | |
2065 | } | |
2066 | ||
2067 | static void unit_remove_from_cgroup_realize_queue(Unit *u) { | |
2068 | assert(u); | |
2069 | ||
2070 | if (!u->in_cgroup_realize_queue) | |
2071 | return; | |
2072 | ||
2073 | LIST_REMOVE(cgroup_realize_queue, u->manager->cgroup_realize_queue, u); | |
2074 | u->in_cgroup_realize_queue = false; | |
2075 | } | |
2076 | ||
a57669d2 CD |
2077 | /* Controllers can only be enabled breadth-first, from the root of the |
2078 | * hierarchy downwards to the unit in question. */ | |
2079 | static int unit_realize_cgroup_now_enable(Unit *u, ManagerState state) { | |
2080 | CGroupMask target_mask, enable_mask, new_target_mask, new_enable_mask; | |
2081 | int r; | |
2082 | ||
2083 | assert(u); | |
2084 | ||
2085 | /* First go deal with this unit's parent, or we won't be able to enable | |
2086 | * any new controllers at this layer. */ | |
2087 | if (UNIT_ISSET(u->slice)) { | |
2088 | r = unit_realize_cgroup_now_enable(UNIT_DEREF(u->slice), state); | |
2089 | if (r < 0) | |
2090 | return r; | |
2091 | } | |
2092 | ||
2093 | target_mask = unit_get_target_mask(u); | |
2094 | enable_mask = unit_get_enable_mask(u); | |
2095 | ||
2096 | /* We can only enable in this direction, don't try to disable anything. | |
2097 | */ | |
2098 | if (unit_has_mask_enables_realized(u, target_mask, enable_mask)) | |
2099 | return 0; | |
2100 | ||
2101 | new_target_mask = u->cgroup_realized_mask | target_mask; | |
2102 | new_enable_mask = u->cgroup_enabled_mask | enable_mask; | |
2103 | ||
c72703e2 | 2104 | return unit_create_cgroup(u, new_target_mask, new_enable_mask, state); |
a57669d2 CD |
2105 | } |
2106 | ||
4f6f62e4 CD |
2107 | /* Controllers can only be disabled depth-first, from the leaves of the |
2108 | * hierarchy upwards to the unit in question. */ | |
2109 | static int unit_realize_cgroup_now_disable(Unit *u, ManagerState state) { | |
2110 | Iterator i; | |
2111 | Unit *m; | |
2112 | void *v; | |
2113 | ||
2114 | assert(u); | |
2115 | ||
2116 | if (u->type != UNIT_SLICE) | |
2117 | return 0; | |
2118 | ||
2119 | HASHMAP_FOREACH_KEY(v, m, u->dependencies[UNIT_BEFORE], i) { | |
2120 | CGroupMask target_mask, enable_mask, new_target_mask, new_enable_mask; | |
2121 | int r; | |
2122 | ||
2123 | if (UNIT_DEREF(m->slice) != u) | |
2124 | continue; | |
2125 | ||
2126 | /* The cgroup for this unit might not actually be fully | |
2127 | * realised yet, in which case it isn't holding any controllers | |
2128 | * open anyway. */ | |
2129 | if (!m->cgroup_path) | |
2130 | continue; | |
2131 | ||
2132 | /* We must disable those below us first in order to release the | |
2133 | * controller. */ | |
2134 | if (m->type == UNIT_SLICE) | |
2135 | (void) unit_realize_cgroup_now_disable(m, state); | |
2136 | ||
2137 | target_mask = unit_get_target_mask(m); | |
2138 | enable_mask = unit_get_enable_mask(m); | |
2139 | ||
2140 | /* We can only disable in this direction, don't try to enable | |
2141 | * anything. */ | |
2142 | if (unit_has_mask_disables_realized(m, target_mask, enable_mask)) | |
2143 | continue; | |
2144 | ||
2145 | new_target_mask = m->cgroup_realized_mask & target_mask; | |
2146 | new_enable_mask = m->cgroup_enabled_mask & enable_mask; | |
2147 | ||
2148 | r = unit_create_cgroup(m, new_target_mask, new_enable_mask, state); | |
2149 | if (r < 0) | |
2150 | return r; | |
2151 | } | |
2152 | ||
2153 | return 0; | |
2154 | } | |
a57669d2 | 2155 | |
6414b7c9 DS |
2156 | /* Check if necessary controllers and attributes for a unit are in place. |
2157 | * | |
a57669d2 CD |
2158 | * - If so, do nothing. |
2159 | * - If not, create paths, move processes over, and set attributes. | |
2160 | * | |
2161 | * Controllers can only be *enabled* in a breadth-first way, and *disabled* in | |
2162 | * a depth-first way. As such the process looks like this: | |
2163 | * | |
2164 | * Suppose we have a cgroup hierarchy which looks like this: | |
2165 | * | |
2166 | * root | |
2167 | * / \ | |
2168 | * / \ | |
2169 | * / \ | |
2170 | * a b | |
2171 | * / \ / \ | |
2172 | * / \ / \ | |
2173 | * c d e f | |
2174 | * / \ / \ / \ / \ | |
2175 | * h i j k l m n o | |
2176 | * | |
2177 | * 1. We want to realise cgroup "d" now. | |
c72703e2 | 2178 | * 2. cgroup "a" has DisableControllers=cpu in the associated unit. |
a57669d2 CD |
2179 | * 3. cgroup "k" just started requesting the memory controller. |
2180 | * | |
2181 | * To make this work we must do the following in order: | |
2182 | * | |
2183 | * 1. Disable CPU controller in k, j | |
2184 | * 2. Disable CPU controller in d | |
2185 | * 3. Enable memory controller in root | |
2186 | * 4. Enable memory controller in a | |
2187 | * 5. Enable memory controller in d | |
2188 | * 6. Enable memory controller in k | |
2189 | * | |
2190 | * Notice that we need to touch j in one direction, but not the other. We also | |
2191 | * don't go beyond d when disabling -- it's up to "a" to get realized if it | |
2192 | * wants to disable further. The basic rules are therefore: | |
2193 | * | |
2194 | * - If you're disabling something, you need to realise all of the cgroups from | |
2195 | * your recursive descendants to the root. This starts from the leaves. | |
2196 | * - If you're enabling something, you need to realise from the root cgroup | |
2197 | * downwards, but you don't need to iterate your recursive descendants. | |
6414b7c9 DS |
2198 | * |
2199 | * Returns 0 on success and < 0 on failure. */ | |
db785129 | 2200 | static int unit_realize_cgroup_now(Unit *u, ManagerState state) { |
efdb0237 | 2201 | CGroupMask target_mask, enable_mask; |
6414b7c9 | 2202 | int r; |
64747e2d | 2203 | |
4ad49000 | 2204 | assert(u); |
64747e2d | 2205 | |
2aa57a65 | 2206 | unit_remove_from_cgroup_realize_queue(u); |
64747e2d | 2207 | |
efdb0237 | 2208 | target_mask = unit_get_target_mask(u); |
ccf78df1 TH |
2209 | enable_mask = unit_get_enable_mask(u); |
2210 | ||
17f14955 | 2211 | if (unit_has_mask_realized(u, target_mask, enable_mask)) |
0a1eb06d | 2212 | return 0; |
64747e2d | 2213 | |
4f6f62e4 CD |
2214 | /* Disable controllers below us, if there are any */ |
2215 | r = unit_realize_cgroup_now_disable(u, state); | |
2216 | if (r < 0) | |
2217 | return r; | |
2218 | ||
2219 | /* Enable controllers above us, if there are any */ | |
6414b7c9 | 2220 | if (UNIT_ISSET(u->slice)) { |
a57669d2 | 2221 | r = unit_realize_cgroup_now_enable(UNIT_DEREF(u->slice), state); |
6414b7c9 DS |
2222 | if (r < 0) |
2223 | return r; | |
2224 | } | |
4ad49000 | 2225 | |
0d2d6fbf CD |
2226 | /* Now actually deal with the cgroup we were trying to realise and set attributes */ |
2227 | r = unit_create_cgroup(u, target_mask, enable_mask, state); | |
6414b7c9 DS |
2228 | if (r < 0) |
2229 | return r; | |
2230 | ||
c2baf11c LP |
2231 | /* Now, reset the invalidation mask */ |
2232 | u->cgroup_invalidated_mask = 0; | |
6414b7c9 | 2233 | return 0; |
64747e2d LP |
2234 | } |
2235 | ||
91a6073e | 2236 | unsigned manager_dispatch_cgroup_realize_queue(Manager *m) { |
db785129 | 2237 | ManagerState state; |
4ad49000 | 2238 | unsigned n = 0; |
db785129 | 2239 | Unit *i; |
6414b7c9 | 2240 | int r; |
ecedd90f | 2241 | |
91a6073e LP |
2242 | assert(m); |
2243 | ||
db785129 LP |
2244 | state = manager_state(m); |
2245 | ||
91a6073e LP |
2246 | while ((i = m->cgroup_realize_queue)) { |
2247 | assert(i->in_cgroup_realize_queue); | |
ecedd90f | 2248 | |
2aa57a65 LP |
2249 | if (UNIT_IS_INACTIVE_OR_FAILED(unit_active_state(i))) { |
2250 | /* Maybe things changed, and the unit is not actually active anymore? */ | |
2251 | unit_remove_from_cgroup_realize_queue(i); | |
2252 | continue; | |
2253 | } | |
2254 | ||
db785129 | 2255 | r = unit_realize_cgroup_now(i, state); |
6414b7c9 | 2256 | if (r < 0) |
efdb0237 | 2257 | log_warning_errno(r, "Failed to realize cgroups for queued unit %s, ignoring: %m", i->id); |
0a1eb06d | 2258 | |
4ad49000 LP |
2259 | n++; |
2260 | } | |
ecedd90f | 2261 | |
4ad49000 | 2262 | return n; |
8e274523 LP |
2263 | } |
2264 | ||
91a6073e | 2265 | static void unit_add_siblings_to_cgroup_realize_queue(Unit *u) { |
4ad49000 | 2266 | Unit *slice; |
ca949c9d | 2267 | |
4ad49000 LP |
2268 | /* This adds the siblings of the specified unit and the |
2269 | * siblings of all parent units to the cgroup queue. (But | |
2270 | * neither the specified unit itself nor the parents.) */ | |
2271 | ||
2272 | while ((slice = UNIT_DEREF(u->slice))) { | |
2273 | Iterator i; | |
2274 | Unit *m; | |
eef85c4a | 2275 | void *v; |
8f53a7b8 | 2276 | |
eef85c4a | 2277 | HASHMAP_FOREACH_KEY(v, m, u->dependencies[UNIT_BEFORE], i) { |
6414b7c9 DS |
2278 | /* Skip units that have a dependency on the slice |
2279 | * but aren't actually in it. */ | |
4ad49000 | 2280 | if (UNIT_DEREF(m->slice) != slice) |
50159e6a | 2281 | continue; |
8e274523 | 2282 | |
6414b7c9 DS |
2283 | /* No point in doing cgroup application for units |
2284 | * without active processes. */ | |
2285 | if (UNIT_IS_INACTIVE_OR_FAILED(unit_active_state(m))) | |
2286 | continue; | |
2287 | ||
2288 | /* If the unit doesn't need any new controllers | |
2289 | * and has current ones realized, it doesn't need | |
2290 | * any changes. */ | |
906c06f6 DM |
2291 | if (unit_has_mask_realized(m, |
2292 | unit_get_target_mask(m), | |
17f14955 | 2293 | unit_get_enable_mask(m))) |
6414b7c9 DS |
2294 | continue; |
2295 | ||
91a6073e | 2296 | unit_add_to_cgroup_realize_queue(m); |
50159e6a LP |
2297 | } |
2298 | ||
4ad49000 | 2299 | u = slice; |
8e274523 | 2300 | } |
4ad49000 LP |
2301 | } |
2302 | ||
0a1eb06d | 2303 | int unit_realize_cgroup(Unit *u) { |
4ad49000 LP |
2304 | assert(u); |
2305 | ||
35b7ff80 | 2306 | if (!UNIT_HAS_CGROUP_CONTEXT(u)) |
0a1eb06d | 2307 | return 0; |
8e274523 | 2308 | |
4ad49000 LP |
2309 | /* So, here's the deal: when realizing the cgroups for this |
2310 | * unit, we need to first create all parents, but there's more | |
2311 | * actually: for the weight-based controllers we also need to | |
2312 | * make sure that all our siblings (i.e. units that are in the | |
73e231ab | 2313 | * same slice as we are) have cgroups, too. Otherwise, things |
4ad49000 LP |
2314 | * would become very uneven as each of their processes would |
2315 | * get as much resources as all our group together. This call | |
2316 | * will synchronously create the parent cgroups, but will | |
2317 | * defer work on the siblings to the next event loop | |
2318 | * iteration. */ | |
ca949c9d | 2319 | |
4ad49000 | 2320 | /* Add all sibling slices to the cgroup queue. */ |
91a6073e | 2321 | unit_add_siblings_to_cgroup_realize_queue(u); |
4ad49000 | 2322 | |
6414b7c9 | 2323 | /* And realize this one now (and apply the values) */ |
db785129 | 2324 | return unit_realize_cgroup_now(u, manager_state(u->manager)); |
8e274523 LP |
2325 | } |
2326 | ||
efdb0237 LP |
2327 | void unit_release_cgroup(Unit *u) { |
2328 | assert(u); | |
2329 | ||
8a0d5388 LP |
2330 | /* Forgets all cgroup details for this cgroup — but does *not* destroy the cgroup. This is hence OK to call |
2331 | * when we close down everything for reexecution, where we really want to leave the cgroup in place. */ | |
efdb0237 LP |
2332 | |
2333 | if (u->cgroup_path) { | |
2334 | (void) hashmap_remove(u->manager->cgroup_unit, u->cgroup_path); | |
2335 | u->cgroup_path = mfree(u->cgroup_path); | |
2336 | } | |
2337 | ||
0bb814c2 LP |
2338 | if (u->cgroup_control_inotify_wd >= 0) { |
2339 | if (inotify_rm_watch(u->manager->cgroup_inotify_fd, u->cgroup_control_inotify_wd) < 0) | |
2340 | 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 | 2341 | |
0bb814c2 LP |
2342 | (void) hashmap_remove(u->manager->cgroup_control_inotify_wd_unit, INT_TO_PTR(u->cgroup_control_inotify_wd)); |
2343 | u->cgroup_control_inotify_wd = -1; | |
efdb0237 | 2344 | } |
afcfaa69 LP |
2345 | |
2346 | if (u->cgroup_memory_inotify_wd >= 0) { | |
2347 | if (inotify_rm_watch(u->manager->cgroup_inotify_fd, u->cgroup_memory_inotify_wd) < 0) | |
2348 | 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); | |
2349 | ||
2350 | (void) hashmap_remove(u->manager->cgroup_memory_inotify_wd_unit, INT_TO_PTR(u->cgroup_memory_inotify_wd)); | |
2351 | u->cgroup_memory_inotify_wd = -1; | |
2352 | } | |
efdb0237 LP |
2353 | } |
2354 | ||
2355 | void unit_prune_cgroup(Unit *u) { | |
8e274523 | 2356 | int r; |
efdb0237 | 2357 | bool is_root_slice; |
8e274523 | 2358 | |
4ad49000 | 2359 | assert(u); |
8e274523 | 2360 | |
efdb0237 LP |
2361 | /* Removes the cgroup, if empty and possible, and stops watching it. */ |
2362 | ||
4ad49000 LP |
2363 | if (!u->cgroup_path) |
2364 | return; | |
8e274523 | 2365 | |
fe700f46 LP |
2366 | (void) unit_get_cpu_usage(u, NULL); /* Cache the last CPU usage value before we destroy the cgroup */ |
2367 | ||
efdb0237 LP |
2368 | is_root_slice = unit_has_name(u, SPECIAL_ROOT_SLICE); |
2369 | ||
2370 | r = cg_trim_everywhere(u->manager->cgroup_supported, u->cgroup_path, !is_root_slice); | |
dab5bf85 | 2371 | if (r < 0) { |
f29ff115 | 2372 | log_unit_debug_errno(u, r, "Failed to destroy cgroup %s, ignoring: %m", u->cgroup_path); |
dab5bf85 RL |
2373 | return; |
2374 | } | |
8e274523 | 2375 | |
efdb0237 LP |
2376 | if (is_root_slice) |
2377 | return; | |
2378 | ||
2379 | unit_release_cgroup(u); | |
0a1eb06d | 2380 | |
4ad49000 | 2381 | u->cgroup_realized = false; |
bc432dc7 | 2382 | u->cgroup_realized_mask = 0; |
ccf78df1 | 2383 | u->cgroup_enabled_mask = 0; |
084c7007 RG |
2384 | |
2385 | u->bpf_device_control_installed = bpf_program_unref(u->bpf_device_control_installed); | |
8e274523 LP |
2386 | } |
2387 | ||
efdb0237 | 2388 | int unit_search_main_pid(Unit *u, pid_t *ret) { |
4ad49000 | 2389 | _cleanup_fclose_ FILE *f = NULL; |
4d051546 | 2390 | pid_t pid = 0, npid; |
efdb0237 | 2391 | int r; |
4ad49000 LP |
2392 | |
2393 | assert(u); | |
efdb0237 | 2394 | assert(ret); |
4ad49000 LP |
2395 | |
2396 | if (!u->cgroup_path) | |
efdb0237 | 2397 | return -ENXIO; |
4ad49000 | 2398 | |
efdb0237 LP |
2399 | r = cg_enumerate_processes(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path, &f); |
2400 | if (r < 0) | |
2401 | return r; | |
4ad49000 | 2402 | |
4ad49000 | 2403 | while (cg_read_pid(f, &npid) > 0) { |
4ad49000 LP |
2404 | |
2405 | if (npid == pid) | |
2406 | continue; | |
8e274523 | 2407 | |
4d051546 | 2408 | if (pid_is_my_child(npid) == 0) |
4ad49000 | 2409 | continue; |
8e274523 | 2410 | |
efdb0237 | 2411 | if (pid != 0) |
4ad49000 LP |
2412 | /* Dang, there's more than one daemonized PID |
2413 | in this group, so we don't know what process | |
2414 | is the main process. */ | |
efdb0237 LP |
2415 | |
2416 | return -ENODATA; | |
8e274523 | 2417 | |
4ad49000 | 2418 | pid = npid; |
8e274523 LP |
2419 | } |
2420 | ||
efdb0237 LP |
2421 | *ret = pid; |
2422 | return 0; | |
2423 | } | |
2424 | ||
2425 | static int unit_watch_pids_in_path(Unit *u, const char *path) { | |
b3c5bad3 | 2426 | _cleanup_closedir_ DIR *d = NULL; |
efdb0237 LP |
2427 | _cleanup_fclose_ FILE *f = NULL; |
2428 | int ret = 0, r; | |
2429 | ||
2430 | assert(u); | |
2431 | assert(path); | |
2432 | ||
2433 | r = cg_enumerate_processes(SYSTEMD_CGROUP_CONTROLLER, path, &f); | |
2434 | if (r < 0) | |
2435 | ret = r; | |
2436 | else { | |
2437 | pid_t pid; | |
2438 | ||
2439 | while ((r = cg_read_pid(f, &pid)) > 0) { | |
f75f613d | 2440 | r = unit_watch_pid(u, pid, false); |
efdb0237 LP |
2441 | if (r < 0 && ret >= 0) |
2442 | ret = r; | |
2443 | } | |
2444 | ||
2445 | if (r < 0 && ret >= 0) | |
2446 | ret = r; | |
2447 | } | |
2448 | ||
2449 | r = cg_enumerate_subgroups(SYSTEMD_CGROUP_CONTROLLER, path, &d); | |
2450 | if (r < 0) { | |
2451 | if (ret >= 0) | |
2452 | ret = r; | |
2453 | } else { | |
2454 | char *fn; | |
2455 | ||
2456 | while ((r = cg_read_subgroup(d, &fn)) > 0) { | |
2457 | _cleanup_free_ char *p = NULL; | |
2458 | ||
605405c6 | 2459 | p = strjoin(path, "/", fn); |
efdb0237 LP |
2460 | free(fn); |
2461 | ||
2462 | if (!p) | |
2463 | return -ENOMEM; | |
2464 | ||
2465 | r = unit_watch_pids_in_path(u, p); | |
2466 | if (r < 0 && ret >= 0) | |
2467 | ret = r; | |
2468 | } | |
2469 | ||
2470 | if (r < 0 && ret >= 0) | |
2471 | ret = r; | |
2472 | } | |
2473 | ||
2474 | return ret; | |
2475 | } | |
2476 | ||
11aef522 LP |
2477 | int unit_synthesize_cgroup_empty_event(Unit *u) { |
2478 | int r; | |
2479 | ||
2480 | assert(u); | |
2481 | ||
2482 | /* Enqueue a synthetic cgroup empty event if this unit doesn't watch any PIDs anymore. This is compatibility | |
2483 | * support for non-unified systems where notifications aren't reliable, and hence need to take whatever we can | |
2484 | * get as notification source as soon as we stopped having any useful PIDs to watch for. */ | |
2485 | ||
2486 | if (!u->cgroup_path) | |
2487 | return -ENOENT; | |
2488 | ||
2489 | r = cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER); | |
2490 | if (r < 0) | |
2491 | return r; | |
2492 | if (r > 0) /* On unified we have reliable notifications, and don't need this */ | |
2493 | return 0; | |
2494 | ||
2495 | if (!set_isempty(u->pids)) | |
2496 | return 0; | |
2497 | ||
2498 | unit_add_to_cgroup_empty_queue(u); | |
2499 | return 0; | |
2500 | } | |
2501 | ||
efdb0237 | 2502 | int unit_watch_all_pids(Unit *u) { |
b4cccbc1 LP |
2503 | int r; |
2504 | ||
efdb0237 LP |
2505 | assert(u); |
2506 | ||
2507 | /* Adds all PIDs from our cgroup to the set of PIDs we | |
2508 | * watch. This is a fallback logic for cases where we do not | |
2509 | * get reliable cgroup empty notifications: we try to use | |
2510 | * SIGCHLD as replacement. */ | |
2511 | ||
2512 | if (!u->cgroup_path) | |
2513 | return -ENOENT; | |
2514 | ||
c22800e4 | 2515 | r = cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER); |
b4cccbc1 LP |
2516 | if (r < 0) |
2517 | return r; | |
2518 | if (r > 0) /* On unified we can use proper notifications */ | |
efdb0237 LP |
2519 | return 0; |
2520 | ||
2521 | return unit_watch_pids_in_path(u, u->cgroup_path); | |
2522 | } | |
2523 | ||
09e24654 LP |
2524 | static int on_cgroup_empty_event(sd_event_source *s, void *userdata) { |
2525 | Manager *m = userdata; | |
2526 | Unit *u; | |
efdb0237 LP |
2527 | int r; |
2528 | ||
09e24654 LP |
2529 | assert(s); |
2530 | assert(m); | |
efdb0237 | 2531 | |
09e24654 LP |
2532 | u = m->cgroup_empty_queue; |
2533 | if (!u) | |
efdb0237 LP |
2534 | return 0; |
2535 | ||
09e24654 LP |
2536 | assert(u->in_cgroup_empty_queue); |
2537 | u->in_cgroup_empty_queue = false; | |
2538 | LIST_REMOVE(cgroup_empty_queue, m->cgroup_empty_queue, u); | |
2539 | ||
2540 | if (m->cgroup_empty_queue) { | |
2541 | /* More stuff queued, let's make sure we remain enabled */ | |
2542 | r = sd_event_source_set_enabled(s, SD_EVENT_ONESHOT); | |
2543 | if (r < 0) | |
19a691a9 | 2544 | log_debug_errno(r, "Failed to reenable cgroup empty event source, ignoring: %m"); |
09e24654 | 2545 | } |
efdb0237 LP |
2546 | |
2547 | unit_add_to_gc_queue(u); | |
2548 | ||
2549 | if (UNIT_VTABLE(u)->notify_cgroup_empty) | |
2550 | UNIT_VTABLE(u)->notify_cgroup_empty(u); | |
2551 | ||
2552 | return 0; | |
2553 | } | |
2554 | ||
09e24654 LP |
2555 | void unit_add_to_cgroup_empty_queue(Unit *u) { |
2556 | int r; | |
2557 | ||
2558 | assert(u); | |
2559 | ||
2560 | /* Note that there are four different ways how cgroup empty events reach us: | |
2561 | * | |
2562 | * 1. On the unified hierarchy we get an inotify event on the cgroup | |
2563 | * | |
2564 | * 2. On the legacy hierarchy, when running in system mode, we get a datagram on the cgroup agent socket | |
2565 | * | |
2566 | * 3. On the legacy hierarchy, when running in user mode, we get a D-Bus signal on the system bus | |
2567 | * | |
2568 | * 4. On the legacy hierarchy, in service units we start watching all processes of the cgroup for SIGCHLD as | |
2569 | * soon as we get one SIGCHLD, to deal with unreliable cgroup notifications. | |
2570 | * | |
2571 | * Regardless which way we got the notification, we'll verify it here, and then add it to a separate | |
2572 | * queue. This queue will be dispatched at a lower priority than the SIGCHLD handler, so that we always use | |
2573 | * SIGCHLD if we can get it first, and only use the cgroup empty notifications if there's no SIGCHLD pending | |
2574 | * (which might happen if the cgroup doesn't contain processes that are our own child, which is typically the | |
2575 | * case for scope units). */ | |
2576 | ||
2577 | if (u->in_cgroup_empty_queue) | |
2578 | return; | |
2579 | ||
2580 | /* Let's verify that the cgroup is really empty */ | |
2581 | if (!u->cgroup_path) | |
2582 | return; | |
2583 | r = cg_is_empty_recursive(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path); | |
2584 | if (r < 0) { | |
2585 | log_unit_debug_errno(u, r, "Failed to determine whether cgroup %s is empty: %m", u->cgroup_path); | |
2586 | return; | |
2587 | } | |
2588 | if (r == 0) | |
2589 | return; | |
2590 | ||
2591 | LIST_PREPEND(cgroup_empty_queue, u->manager->cgroup_empty_queue, u); | |
2592 | u->in_cgroup_empty_queue = true; | |
2593 | ||
2594 | /* Trigger the defer event */ | |
2595 | r = sd_event_source_set_enabled(u->manager->cgroup_empty_event_source, SD_EVENT_ONESHOT); | |
2596 | if (r < 0) | |
2597 | log_debug_errno(r, "Failed to enable cgroup empty event source: %m"); | |
2598 | } | |
2599 | ||
afcfaa69 LP |
2600 | static int unit_check_oom(Unit *u) { |
2601 | _cleanup_free_ char *oom_kill = NULL; | |
2602 | bool increased; | |
2603 | uint64_t c; | |
2604 | int r; | |
2605 | ||
2606 | if (!u->cgroup_path) | |
2607 | return 0; | |
2608 | ||
2609 | r = cg_get_keyed_attribute("memory", u->cgroup_path, "memory.events", STRV_MAKE("oom_kill"), &oom_kill); | |
2610 | if (r < 0) | |
2611 | return log_unit_debug_errno(u, r, "Failed to read oom_kill field of memory.events cgroup attribute: %m"); | |
2612 | ||
2613 | r = safe_atou64(oom_kill, &c); | |
2614 | if (r < 0) | |
2615 | return log_unit_debug_errno(u, r, "Failed to parse oom_kill field: %m"); | |
2616 | ||
2617 | increased = c > u->oom_kill_last; | |
2618 | u->oom_kill_last = c; | |
2619 | ||
2620 | if (!increased) | |
2621 | return 0; | |
2622 | ||
2623 | log_struct(LOG_NOTICE, | |
2624 | "MESSAGE_ID=" SD_MESSAGE_UNIT_OUT_OF_MEMORY_STR, | |
2625 | LOG_UNIT_ID(u), | |
2626 | LOG_UNIT_INVOCATION_ID(u), | |
2627 | LOG_UNIT_MESSAGE(u, "A process of this unit has been killed by the OOM killer.")); | |
2628 | ||
2629 | if (UNIT_VTABLE(u)->notify_cgroup_oom) | |
2630 | UNIT_VTABLE(u)->notify_cgroup_oom(u); | |
2631 | ||
2632 | return 1; | |
2633 | } | |
2634 | ||
2635 | static int on_cgroup_oom_event(sd_event_source *s, void *userdata) { | |
2636 | Manager *m = userdata; | |
2637 | Unit *u; | |
2638 | int r; | |
2639 | ||
2640 | assert(s); | |
2641 | assert(m); | |
2642 | ||
2643 | u = m->cgroup_oom_queue; | |
2644 | if (!u) | |
2645 | return 0; | |
2646 | ||
2647 | assert(u->in_cgroup_oom_queue); | |
2648 | u->in_cgroup_oom_queue = false; | |
2649 | LIST_REMOVE(cgroup_oom_queue, m->cgroup_oom_queue, u); | |
2650 | ||
2651 | if (m->cgroup_oom_queue) { | |
2652 | /* More stuff queued, let's make sure we remain enabled */ | |
2653 | r = sd_event_source_set_enabled(s, SD_EVENT_ONESHOT); | |
2654 | if (r < 0) | |
2655 | log_debug_errno(r, "Failed to reenable cgroup oom event source, ignoring: %m"); | |
2656 | } | |
2657 | ||
2658 | (void) unit_check_oom(u); | |
2659 | return 0; | |
2660 | } | |
2661 | ||
2662 | static void unit_add_to_cgroup_oom_queue(Unit *u) { | |
2663 | int r; | |
2664 | ||
2665 | assert(u); | |
2666 | ||
2667 | if (u->in_cgroup_oom_queue) | |
2668 | return; | |
2669 | if (!u->cgroup_path) | |
2670 | return; | |
2671 | ||
2672 | LIST_PREPEND(cgroup_oom_queue, u->manager->cgroup_oom_queue, u); | |
2673 | u->in_cgroup_oom_queue = true; | |
2674 | ||
2675 | /* Trigger the defer event */ | |
2676 | if (!u->manager->cgroup_oom_event_source) { | |
2677 | _cleanup_(sd_event_source_unrefp) sd_event_source *s = NULL; | |
2678 | ||
2679 | r = sd_event_add_defer(u->manager->event, &s, on_cgroup_oom_event, u->manager); | |
2680 | if (r < 0) { | |
2681 | log_error_errno(r, "Failed to create cgroup oom event source: %m"); | |
2682 | return; | |
2683 | } | |
2684 | ||
2685 | r = sd_event_source_set_priority(s, SD_EVENT_PRIORITY_NORMAL-8); | |
2686 | if (r < 0) { | |
2687 | log_error_errno(r, "Failed to set priority of cgroup oom event source: %m"); | |
2688 | return; | |
2689 | } | |
2690 | ||
2691 | (void) sd_event_source_set_description(s, "cgroup-oom"); | |
2692 | u->manager->cgroup_oom_event_source = TAKE_PTR(s); | |
2693 | } | |
2694 | ||
2695 | r = sd_event_source_set_enabled(u->manager->cgroup_oom_event_source, SD_EVENT_ONESHOT); | |
2696 | if (r < 0) | |
2697 | log_error_errno(r, "Failed to enable cgroup oom event source: %m"); | |
2698 | } | |
2699 | ||
efdb0237 LP |
2700 | static int on_cgroup_inotify_event(sd_event_source *s, int fd, uint32_t revents, void *userdata) { |
2701 | Manager *m = userdata; | |
2702 | ||
2703 | assert(s); | |
2704 | assert(fd >= 0); | |
2705 | assert(m); | |
2706 | ||
2707 | for (;;) { | |
2708 | union inotify_event_buffer buffer; | |
2709 | struct inotify_event *e; | |
2710 | ssize_t l; | |
2711 | ||
2712 | l = read(fd, &buffer, sizeof(buffer)); | |
2713 | if (l < 0) { | |
47249640 | 2714 | if (IN_SET(errno, EINTR, EAGAIN)) |
efdb0237 LP |
2715 | return 0; |
2716 | ||
2717 | return log_error_errno(errno, "Failed to read control group inotify events: %m"); | |
2718 | } | |
2719 | ||
2720 | FOREACH_INOTIFY_EVENT(e, buffer, l) { | |
2721 | Unit *u; | |
2722 | ||
2723 | if (e->wd < 0) | |
2724 | /* Queue overflow has no watch descriptor */ | |
2725 | continue; | |
2726 | ||
2727 | if (e->mask & IN_IGNORED) | |
2728 | /* The watch was just removed */ | |
2729 | continue; | |
2730 | ||
afcfaa69 LP |
2731 | /* Note that inotify might deliver events for a watch even after it was removed, |
2732 | * because it was queued before the removal. Let's ignore this here safely. */ | |
2733 | ||
0bb814c2 | 2734 | u = hashmap_get(m->cgroup_control_inotify_wd_unit, INT_TO_PTR(e->wd)); |
afcfaa69 LP |
2735 | if (u) |
2736 | unit_add_to_cgroup_empty_queue(u); | |
efdb0237 | 2737 | |
afcfaa69 LP |
2738 | u = hashmap_get(m->cgroup_memory_inotify_wd_unit, INT_TO_PTR(e->wd)); |
2739 | if (u) | |
2740 | unit_add_to_cgroup_oom_queue(u); | |
efdb0237 LP |
2741 | } |
2742 | } | |
8e274523 LP |
2743 | } |
2744 | ||
17f14955 RG |
2745 | static int cg_bpf_mask_supported(CGroupMask *ret) { |
2746 | CGroupMask mask = 0; | |
2747 | int r; | |
2748 | ||
2749 | /* BPF-based firewall */ | |
2750 | r = bpf_firewall_supported(); | |
2751 | if (r > 0) | |
2752 | mask |= CGROUP_MASK_BPF_FIREWALL; | |
2753 | ||
084c7007 RG |
2754 | /* BPF-based device access control */ |
2755 | r = bpf_devices_supported(); | |
2756 | if (r > 0) | |
2757 | mask |= CGROUP_MASK_BPF_DEVICES; | |
2758 | ||
17f14955 RG |
2759 | *ret = mask; |
2760 | return 0; | |
2761 | } | |
2762 | ||
8e274523 | 2763 | int manager_setup_cgroup(Manager *m) { |
9444b1f2 | 2764 | _cleanup_free_ char *path = NULL; |
10bd3e2e | 2765 | const char *scope_path; |
efdb0237 | 2766 | CGroupController c; |
b4cccbc1 | 2767 | int r, all_unified; |
17f14955 | 2768 | CGroupMask mask; |
efdb0237 | 2769 | char *e; |
8e274523 LP |
2770 | |
2771 | assert(m); | |
2772 | ||
35d2e7ec | 2773 | /* 1. Determine hierarchy */ |
efdb0237 | 2774 | m->cgroup_root = mfree(m->cgroup_root); |
9444b1f2 | 2775 | r = cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER, 0, &m->cgroup_root); |
23bbb0de MS |
2776 | if (r < 0) |
2777 | return log_error_errno(r, "Cannot determine cgroup we are running in: %m"); | |
8e274523 | 2778 | |
efdb0237 LP |
2779 | /* Chop off the init scope, if we are already located in it */ |
2780 | e = endswith(m->cgroup_root, "/" SPECIAL_INIT_SCOPE); | |
0d8c31ff | 2781 | |
efdb0237 LP |
2782 | /* LEGACY: Also chop off the system slice if we are in |
2783 | * it. This is to support live upgrades from older systemd | |
2784 | * versions where PID 1 was moved there. Also see | |
2785 | * cg_get_root_path(). */ | |
463d0d15 | 2786 | if (!e && MANAGER_IS_SYSTEM(m)) { |
9444b1f2 | 2787 | e = endswith(m->cgroup_root, "/" SPECIAL_SYSTEM_SLICE); |
15c60e99 | 2788 | if (!e) |
efdb0237 | 2789 | e = endswith(m->cgroup_root, "/system"); /* even more legacy */ |
0baf24dd | 2790 | } |
efdb0237 LP |
2791 | if (e) |
2792 | *e = 0; | |
7ccfb64a | 2793 | |
7546145e LP |
2794 | /* And make sure to store away the root value without trailing slash, even for the root dir, so that we can |
2795 | * easily prepend it everywhere. */ | |
2796 | delete_trailing_chars(m->cgroup_root, "/"); | |
8e274523 | 2797 | |
35d2e7ec | 2798 | /* 2. Show data */ |
9444b1f2 | 2799 | r = cg_get_path(SYSTEMD_CGROUP_CONTROLLER, m->cgroup_root, NULL, &path); |
23bbb0de MS |
2800 | if (r < 0) |
2801 | return log_error_errno(r, "Cannot find cgroup mount point: %m"); | |
8e274523 | 2802 | |
415fc41c TH |
2803 | r = cg_unified_flush(); |
2804 | if (r < 0) | |
2805 | return log_error_errno(r, "Couldn't determine if we are running in the unified hierarchy: %m"); | |
5da38d07 | 2806 | |
b4cccbc1 | 2807 | all_unified = cg_all_unified(); |
d4c819ed ZJS |
2808 | if (all_unified < 0) |
2809 | return log_error_errno(all_unified, "Couldn't determine whether we are in all unified mode: %m"); | |
2810 | if (all_unified > 0) | |
efdb0237 | 2811 | log_debug("Unified cgroup hierarchy is located at %s.", path); |
b4cccbc1 | 2812 | else { |
c22800e4 | 2813 | r = cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER); |
b4cccbc1 LP |
2814 | if (r < 0) |
2815 | return log_error_errno(r, "Failed to determine whether systemd's own controller is in unified mode: %m"); | |
2816 | if (r > 0) | |
2817 | log_debug("Unified cgroup hierarchy is located at %s. Controllers are on legacy hierarchies.", path); | |
2818 | else | |
2819 | log_debug("Using cgroup controller " SYSTEMD_CGROUP_CONTROLLER_LEGACY ". File system hierarchy is at %s.", path); | |
2820 | } | |
efdb0237 | 2821 | |
09e24654 LP |
2822 | /* 3. Allocate cgroup empty defer event source */ |
2823 | m->cgroup_empty_event_source = sd_event_source_unref(m->cgroup_empty_event_source); | |
2824 | r = sd_event_add_defer(m->event, &m->cgroup_empty_event_source, on_cgroup_empty_event, m); | |
2825 | if (r < 0) | |
2826 | return log_error_errno(r, "Failed to create cgroup empty event source: %m"); | |
2827 | ||
cbe83389 LP |
2828 | /* Schedule cgroup empty checks early, but after having processed service notification messages or |
2829 | * SIGCHLD signals, so that a cgroup running empty is always just the last safety net of | |
2830 | * notification, and we collected the metadata the notification and SIGCHLD stuff offers first. */ | |
09e24654 LP |
2831 | r = sd_event_source_set_priority(m->cgroup_empty_event_source, SD_EVENT_PRIORITY_NORMAL-5); |
2832 | if (r < 0) | |
2833 | return log_error_errno(r, "Failed to set priority of cgroup empty event source: %m"); | |
2834 | ||
2835 | r = sd_event_source_set_enabled(m->cgroup_empty_event_source, SD_EVENT_OFF); | |
2836 | if (r < 0) | |
2837 | return log_error_errno(r, "Failed to disable cgroup empty event source: %m"); | |
2838 | ||
2839 | (void) sd_event_source_set_description(m->cgroup_empty_event_source, "cgroup-empty"); | |
2840 | ||
2841 | /* 4. Install notifier inotify object, or agent */ | |
10bd3e2e | 2842 | if (cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER) > 0) { |
c6c18be3 | 2843 | |
09e24654 | 2844 | /* In the unified hierarchy we can get cgroup empty notifications via inotify. */ |
efdb0237 | 2845 | |
10bd3e2e LP |
2846 | m->cgroup_inotify_event_source = sd_event_source_unref(m->cgroup_inotify_event_source); |
2847 | safe_close(m->cgroup_inotify_fd); | |
efdb0237 | 2848 | |
10bd3e2e LP |
2849 | m->cgroup_inotify_fd = inotify_init1(IN_NONBLOCK|IN_CLOEXEC); |
2850 | if (m->cgroup_inotify_fd < 0) | |
2851 | return log_error_errno(errno, "Failed to create control group inotify object: %m"); | |
efdb0237 | 2852 | |
10bd3e2e LP |
2853 | r = sd_event_add_io(m->event, &m->cgroup_inotify_event_source, m->cgroup_inotify_fd, EPOLLIN, on_cgroup_inotify_event, m); |
2854 | if (r < 0) | |
2855 | return log_error_errno(r, "Failed to watch control group inotify object: %m"); | |
efdb0237 | 2856 | |
cbe83389 LP |
2857 | /* Process cgroup empty notifications early. Note that when this event is dispatched it'll |
2858 | * just add the unit to a cgroup empty queue, hence let's run earlier than that. Also see | |
2859 | * handling of cgroup agent notifications, for the classic cgroup hierarchy support. */ | |
2860 | r = sd_event_source_set_priority(m->cgroup_inotify_event_source, SD_EVENT_PRIORITY_NORMAL-9); | |
10bd3e2e LP |
2861 | if (r < 0) |
2862 | return log_error_errno(r, "Failed to set priority of inotify event source: %m"); | |
efdb0237 | 2863 | |
10bd3e2e | 2864 | (void) sd_event_source_set_description(m->cgroup_inotify_event_source, "cgroup-inotify"); |
efdb0237 | 2865 | |
611c4f8a | 2866 | } else if (MANAGER_IS_SYSTEM(m) && manager_owns_host_root_cgroup(m) && !MANAGER_IS_TEST_RUN(m)) { |
efdb0237 | 2867 | |
10bd3e2e LP |
2868 | /* On the legacy hierarchy we only get notifications via cgroup agents. (Which isn't really reliable, |
2869 | * since it does not generate events when control groups with children run empty. */ | |
8e274523 | 2870 | |
10bd3e2e | 2871 | r = cg_install_release_agent(SYSTEMD_CGROUP_CONTROLLER, SYSTEMD_CGROUP_AGENT_PATH); |
23bbb0de | 2872 | if (r < 0) |
10bd3e2e LP |
2873 | log_warning_errno(r, "Failed to install release agent, ignoring: %m"); |
2874 | else if (r > 0) | |
2875 | log_debug("Installed release agent."); | |
2876 | else if (r == 0) | |
2877 | log_debug("Release agent already installed."); | |
2878 | } | |
efdb0237 | 2879 | |
09e24654 | 2880 | /* 5. Make sure we are in the special "init.scope" unit in the root slice. */ |
10bd3e2e LP |
2881 | scope_path = strjoina(m->cgroup_root, "/" SPECIAL_INIT_SCOPE); |
2882 | r = cg_create_and_attach(SYSTEMD_CGROUP_CONTROLLER, scope_path, 0); | |
aa77e234 MS |
2883 | if (r >= 0) { |
2884 | /* Also, move all other userspace processes remaining in the root cgroup into that scope. */ | |
2885 | r = cg_migrate(SYSTEMD_CGROUP_CONTROLLER, m->cgroup_root, SYSTEMD_CGROUP_CONTROLLER, scope_path, 0); | |
2886 | if (r < 0) | |
2887 | log_warning_errno(r, "Couldn't move remaining userspace processes, ignoring: %m"); | |
c6c18be3 | 2888 | |
aa77e234 MS |
2889 | /* 6. And pin it, so that it cannot be unmounted */ |
2890 | safe_close(m->pin_cgroupfs_fd); | |
2891 | m->pin_cgroupfs_fd = open(path, O_RDONLY|O_CLOEXEC|O_DIRECTORY|O_NOCTTY|O_NONBLOCK); | |
2892 | if (m->pin_cgroupfs_fd < 0) | |
2893 | return log_error_errno(errno, "Failed to open pin file: %m"); | |
0d8c31ff | 2894 | |
638cece4 | 2895 | } else if (!MANAGER_IS_TEST_RUN(m)) |
aa77e234 | 2896 | return log_error_errno(r, "Failed to create %s control group: %m", scope_path); |
10bd3e2e | 2897 | |
09e24654 | 2898 | /* 7. Always enable hierarchical support if it exists... */ |
638cece4 | 2899 | if (!all_unified && !MANAGER_IS_TEST_RUN(m)) |
10bd3e2e | 2900 | (void) cg_set_attribute("memory", "/", "memory.use_hierarchy", "1"); |
c6c18be3 | 2901 | |
17f14955 | 2902 | /* 8. Figure out which controllers are supported */ |
efdb0237 LP |
2903 | r = cg_mask_supported(&m->cgroup_supported); |
2904 | if (r < 0) | |
2905 | return log_error_errno(r, "Failed to determine supported controllers: %m"); | |
17f14955 RG |
2906 | |
2907 | /* 9. Figure out which bpf-based pseudo-controllers are supported */ | |
2908 | r = cg_bpf_mask_supported(&mask); | |
2909 | if (r < 0) | |
2910 | return log_error_errno(r, "Failed to determine supported bpf-based pseudo-controllers: %m"); | |
2911 | m->cgroup_supported |= mask; | |
2912 | ||
2913 | /* 10. Log which controllers are supported */ | |
efdb0237 | 2914 | for (c = 0; c < _CGROUP_CONTROLLER_MAX; c++) |
eee0a1e4 | 2915 | log_debug("Controller '%s' supported: %s", cgroup_controller_to_string(c), yes_no(m->cgroup_supported & CGROUP_CONTROLLER_TO_MASK(c))); |
9156e799 | 2916 | |
a32360f1 | 2917 | return 0; |
8e274523 LP |
2918 | } |
2919 | ||
c6c18be3 | 2920 | void manager_shutdown_cgroup(Manager *m, bool delete) { |
8e274523 LP |
2921 | assert(m); |
2922 | ||
9444b1f2 LP |
2923 | /* We can't really delete the group, since we are in it. But |
2924 | * let's trim it. */ | |
f6c63f6f | 2925 | if (delete && m->cgroup_root && m->test_run_flags != MANAGER_TEST_RUN_MINIMAL) |
efdb0237 LP |
2926 | (void) cg_trim(SYSTEMD_CGROUP_CONTROLLER, m->cgroup_root, false); |
2927 | ||
09e24654 LP |
2928 | m->cgroup_empty_event_source = sd_event_source_unref(m->cgroup_empty_event_source); |
2929 | ||
0bb814c2 | 2930 | m->cgroup_control_inotify_wd_unit = hashmap_free(m->cgroup_control_inotify_wd_unit); |
afcfaa69 | 2931 | m->cgroup_memory_inotify_wd_unit = hashmap_free(m->cgroup_memory_inotify_wd_unit); |
efdb0237 LP |
2932 | |
2933 | m->cgroup_inotify_event_source = sd_event_source_unref(m->cgroup_inotify_event_source); | |
2934 | m->cgroup_inotify_fd = safe_close(m->cgroup_inotify_fd); | |
8e274523 | 2935 | |
03e334a1 | 2936 | m->pin_cgroupfs_fd = safe_close(m->pin_cgroupfs_fd); |
c6c18be3 | 2937 | |
efdb0237 | 2938 | m->cgroup_root = mfree(m->cgroup_root); |
8e274523 LP |
2939 | } |
2940 | ||
4ad49000 | 2941 | Unit* manager_get_unit_by_cgroup(Manager *m, const char *cgroup) { |
acb14d31 | 2942 | char *p; |
4ad49000 | 2943 | Unit *u; |
acb14d31 LP |
2944 | |
2945 | assert(m); | |
2946 | assert(cgroup); | |
acb14d31 | 2947 | |
4ad49000 LP |
2948 | u = hashmap_get(m->cgroup_unit, cgroup); |
2949 | if (u) | |
2950 | return u; | |
acb14d31 | 2951 | |
8e70580b | 2952 | p = strdupa(cgroup); |
acb14d31 LP |
2953 | for (;;) { |
2954 | char *e; | |
2955 | ||
2956 | e = strrchr(p, '/'); | |
efdb0237 LP |
2957 | if (!e || e == p) |
2958 | return hashmap_get(m->cgroup_unit, SPECIAL_ROOT_SLICE); | |
acb14d31 LP |
2959 | |
2960 | *e = 0; | |
2961 | ||
4ad49000 LP |
2962 | u = hashmap_get(m->cgroup_unit, p); |
2963 | if (u) | |
2964 | return u; | |
acb14d31 LP |
2965 | } |
2966 | } | |
2967 | ||
b3ac818b | 2968 | Unit *manager_get_unit_by_pid_cgroup(Manager *m, pid_t pid) { |
4ad49000 | 2969 | _cleanup_free_ char *cgroup = NULL; |
8e274523 | 2970 | |
8c47c732 LP |
2971 | assert(m); |
2972 | ||
62a76913 | 2973 | if (!pid_is_valid(pid)) |
b3ac818b LP |
2974 | return NULL; |
2975 | ||
62a76913 | 2976 | if (cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER, pid, &cgroup) < 0) |
b3ac818b LP |
2977 | return NULL; |
2978 | ||
2979 | return manager_get_unit_by_cgroup(m, cgroup); | |
2980 | } | |
2981 | ||
2982 | Unit *manager_get_unit_by_pid(Manager *m, pid_t pid) { | |
62a76913 | 2983 | Unit *u, **array; |
b3ac818b LP |
2984 | |
2985 | assert(m); | |
2986 | ||
62a76913 LP |
2987 | /* Note that a process might be owned by multiple units, we return only one here, which is good enough for most |
2988 | * cases, though not strictly correct. We prefer the one reported by cgroup membership, as that's the most | |
2989 | * relevant one as children of the process will be assigned to that one, too, before all else. */ | |
2990 | ||
2991 | if (!pid_is_valid(pid)) | |
8c47c732 LP |
2992 | return NULL; |
2993 | ||
2ca9d979 | 2994 | if (pid == getpid_cached()) |
efdb0237 LP |
2995 | return hashmap_get(m->units, SPECIAL_INIT_SCOPE); |
2996 | ||
62a76913 | 2997 | u = manager_get_unit_by_pid_cgroup(m, pid); |
5fe8876b LP |
2998 | if (u) |
2999 | return u; | |
3000 | ||
62a76913 | 3001 | u = hashmap_get(m->watch_pids, PID_TO_PTR(pid)); |
5fe8876b LP |
3002 | if (u) |
3003 | return u; | |
3004 | ||
62a76913 LP |
3005 | array = hashmap_get(m->watch_pids, PID_TO_PTR(-pid)); |
3006 | if (array) | |
3007 | return array[0]; | |
3008 | ||
3009 | return NULL; | |
6dde1f33 | 3010 | } |
4fbf50b3 | 3011 | |
4ad49000 LP |
3012 | int manager_notify_cgroup_empty(Manager *m, const char *cgroup) { |
3013 | Unit *u; | |
4fbf50b3 | 3014 | |
4ad49000 LP |
3015 | assert(m); |
3016 | assert(cgroup); | |
4fbf50b3 | 3017 | |
09e24654 LP |
3018 | /* Called on the legacy hierarchy whenever we get an explicit cgroup notification from the cgroup agent process |
3019 | * or from the --system instance */ | |
3020 | ||
d8fdc620 LP |
3021 | log_debug("Got cgroup empty notification for: %s", cgroup); |
3022 | ||
4ad49000 | 3023 | u = manager_get_unit_by_cgroup(m, cgroup); |
5ad096b3 LP |
3024 | if (!u) |
3025 | return 0; | |
b56c28c3 | 3026 | |
09e24654 LP |
3027 | unit_add_to_cgroup_empty_queue(u); |
3028 | return 1; | |
5ad096b3 LP |
3029 | } |
3030 | ||
3031 | int unit_get_memory_current(Unit *u, uint64_t *ret) { | |
3032 | _cleanup_free_ char *v = NULL; | |
3033 | int r; | |
3034 | ||
3035 | assert(u); | |
3036 | assert(ret); | |
3037 | ||
2e4025c0 | 3038 | if (!UNIT_CGROUP_BOOL(u, memory_accounting)) |
cf3b4be1 LP |
3039 | return -ENODATA; |
3040 | ||
5ad096b3 LP |
3041 | if (!u->cgroup_path) |
3042 | return -ENODATA; | |
3043 | ||
1f73aa00 | 3044 | /* The root cgroup doesn't expose this information, let's get it from /proc instead */ |
611c4f8a | 3045 | if (unit_has_host_root_cgroup(u)) |
c482724a | 3046 | return procfs_memory_get_used(ret); |
1f73aa00 | 3047 | |
efdb0237 | 3048 | if ((u->cgroup_realized_mask & CGROUP_MASK_MEMORY) == 0) |
5ad096b3 LP |
3049 | return -ENODATA; |
3050 | ||
b4cccbc1 LP |
3051 | r = cg_all_unified(); |
3052 | if (r < 0) | |
3053 | return r; | |
3054 | if (r > 0) | |
efdb0237 | 3055 | r = cg_get_attribute("memory", u->cgroup_path, "memory.current", &v); |
b4cccbc1 LP |
3056 | else |
3057 | r = cg_get_attribute("memory", u->cgroup_path, "memory.usage_in_bytes", &v); | |
5ad096b3 LP |
3058 | if (r == -ENOENT) |
3059 | return -ENODATA; | |
3060 | if (r < 0) | |
3061 | return r; | |
3062 | ||
3063 | return safe_atou64(v, ret); | |
3064 | } | |
3065 | ||
03a7b521 LP |
3066 | int unit_get_tasks_current(Unit *u, uint64_t *ret) { |
3067 | _cleanup_free_ char *v = NULL; | |
3068 | int r; | |
3069 | ||
3070 | assert(u); | |
3071 | assert(ret); | |
3072 | ||
2e4025c0 | 3073 | if (!UNIT_CGROUP_BOOL(u, tasks_accounting)) |
cf3b4be1 LP |
3074 | return -ENODATA; |
3075 | ||
03a7b521 LP |
3076 | if (!u->cgroup_path) |
3077 | return -ENODATA; | |
3078 | ||
c36a69f4 | 3079 | /* The root cgroup doesn't expose this information, let's get it from /proc instead */ |
611c4f8a | 3080 | if (unit_has_host_root_cgroup(u)) |
c36a69f4 LP |
3081 | return procfs_tasks_get_current(ret); |
3082 | ||
1f73aa00 LP |
3083 | if ((u->cgroup_realized_mask & CGROUP_MASK_PIDS) == 0) |
3084 | return -ENODATA; | |
3085 | ||
03a7b521 LP |
3086 | r = cg_get_attribute("pids", u->cgroup_path, "pids.current", &v); |
3087 | if (r == -ENOENT) | |
3088 | return -ENODATA; | |
3089 | if (r < 0) | |
3090 | return r; | |
3091 | ||
3092 | return safe_atou64(v, ret); | |
3093 | } | |
3094 | ||
5ad096b3 LP |
3095 | static int unit_get_cpu_usage_raw(Unit *u, nsec_t *ret) { |
3096 | _cleanup_free_ char *v = NULL; | |
3097 | uint64_t ns; | |
3098 | int r; | |
3099 | ||
3100 | assert(u); | |
3101 | assert(ret); | |
3102 | ||
3103 | if (!u->cgroup_path) | |
3104 | return -ENODATA; | |
3105 | ||
1f73aa00 | 3106 | /* The root cgroup doesn't expose this information, let's get it from /proc instead */ |
611c4f8a | 3107 | if (unit_has_host_root_cgroup(u)) |
1f73aa00 LP |
3108 | return procfs_cpu_get_usage(ret); |
3109 | ||
f98c2585 CD |
3110 | /* Requisite controllers for CPU accounting are not enabled */ |
3111 | if ((get_cpu_accounting_mask() & ~u->cgroup_realized_mask) != 0) | |
3112 | return -ENODATA; | |
3113 | ||
92a99304 LP |
3114 | r = cg_all_unified(); |
3115 | if (r < 0) | |
3116 | return r; | |
b4cccbc1 | 3117 | if (r > 0) { |
66ebf6c0 TH |
3118 | _cleanup_free_ char *val = NULL; |
3119 | uint64_t us; | |
5ad096b3 | 3120 | |
b734a4ff | 3121 | r = cg_get_keyed_attribute("cpu", u->cgroup_path, "cpu.stat", STRV_MAKE("usage_usec"), &val); |
b734a4ff LP |
3122 | if (IN_SET(r, -ENOENT, -ENXIO)) |
3123 | return -ENODATA; | |
d742f4b5 LP |
3124 | if (r < 0) |
3125 | return r; | |
66ebf6c0 TH |
3126 | |
3127 | r = safe_atou64(val, &us); | |
3128 | if (r < 0) | |
3129 | return r; | |
3130 | ||
3131 | ns = us * NSEC_PER_USEC; | |
3132 | } else { | |
66ebf6c0 TH |
3133 | r = cg_get_attribute("cpuacct", u->cgroup_path, "cpuacct.usage", &v); |
3134 | if (r == -ENOENT) | |
3135 | return -ENODATA; | |
3136 | if (r < 0) | |
3137 | return r; | |
3138 | ||
3139 | r = safe_atou64(v, &ns); | |
3140 | if (r < 0) | |
3141 | return r; | |
3142 | } | |
5ad096b3 LP |
3143 | |
3144 | *ret = ns; | |
3145 | return 0; | |
3146 | } | |
3147 | ||
3148 | int unit_get_cpu_usage(Unit *u, nsec_t *ret) { | |
3149 | nsec_t ns; | |
3150 | int r; | |
3151 | ||
fe700f46 LP |
3152 | assert(u); |
3153 | ||
3154 | /* Retrieve the current CPU usage counter. This will subtract the CPU counter taken when the unit was | |
3155 | * started. If the cgroup has been removed already, returns the last cached value. To cache the value, simply | |
3156 | * call this function with a NULL return value. */ | |
3157 | ||
2e4025c0 | 3158 | if (!UNIT_CGROUP_BOOL(u, cpu_accounting)) |
cf3b4be1 LP |
3159 | return -ENODATA; |
3160 | ||
5ad096b3 | 3161 | r = unit_get_cpu_usage_raw(u, &ns); |
fe700f46 LP |
3162 | if (r == -ENODATA && u->cpu_usage_last != NSEC_INFINITY) { |
3163 | /* If we can't get the CPU usage anymore (because the cgroup was already removed, for example), use our | |
3164 | * cached value. */ | |
3165 | ||
3166 | if (ret) | |
3167 | *ret = u->cpu_usage_last; | |
3168 | return 0; | |
3169 | } | |
5ad096b3 LP |
3170 | if (r < 0) |
3171 | return r; | |
3172 | ||
66ebf6c0 TH |
3173 | if (ns > u->cpu_usage_base) |
3174 | ns -= u->cpu_usage_base; | |
5ad096b3 LP |
3175 | else |
3176 | ns = 0; | |
3177 | ||
fe700f46 LP |
3178 | u->cpu_usage_last = ns; |
3179 | if (ret) | |
3180 | *ret = ns; | |
3181 | ||
5ad096b3 LP |
3182 | return 0; |
3183 | } | |
3184 | ||
906c06f6 DM |
3185 | int unit_get_ip_accounting( |
3186 | Unit *u, | |
3187 | CGroupIPAccountingMetric metric, | |
3188 | uint64_t *ret) { | |
3189 | ||
6b659ed8 | 3190 | uint64_t value; |
906c06f6 DM |
3191 | int fd, r; |
3192 | ||
3193 | assert(u); | |
3194 | assert(metric >= 0); | |
3195 | assert(metric < _CGROUP_IP_ACCOUNTING_METRIC_MAX); | |
3196 | assert(ret); | |
3197 | ||
2e4025c0 | 3198 | if (!UNIT_CGROUP_BOOL(u, ip_accounting)) |
cf3b4be1 LP |
3199 | return -ENODATA; |
3200 | ||
906c06f6 DM |
3201 | fd = IN_SET(metric, CGROUP_IP_INGRESS_BYTES, CGROUP_IP_INGRESS_PACKETS) ? |
3202 | u->ip_accounting_ingress_map_fd : | |
3203 | u->ip_accounting_egress_map_fd; | |
906c06f6 DM |
3204 | if (fd < 0) |
3205 | return -ENODATA; | |
3206 | ||
3207 | if (IN_SET(metric, CGROUP_IP_INGRESS_BYTES, CGROUP_IP_EGRESS_BYTES)) | |
6b659ed8 | 3208 | r = bpf_firewall_read_accounting(fd, &value, NULL); |
906c06f6 | 3209 | else |
6b659ed8 LP |
3210 | r = bpf_firewall_read_accounting(fd, NULL, &value); |
3211 | if (r < 0) | |
3212 | return r; | |
3213 | ||
3214 | /* Add in additional metrics from a previous runtime. Note that when reexecing/reloading the daemon we compile | |
3215 | * all BPF programs and maps anew, but serialize the old counters. When deserializing we store them in the | |
3216 | * ip_accounting_extra[] field, and add them in here transparently. */ | |
3217 | ||
3218 | *ret = value + u->ip_accounting_extra[metric]; | |
906c06f6 DM |
3219 | |
3220 | return r; | |
3221 | } | |
3222 | ||
3223 | int unit_reset_cpu_accounting(Unit *u) { | |
5ad096b3 LP |
3224 | nsec_t ns; |
3225 | int r; | |
3226 | ||
3227 | assert(u); | |
3228 | ||
fe700f46 LP |
3229 | u->cpu_usage_last = NSEC_INFINITY; |
3230 | ||
5ad096b3 LP |
3231 | r = unit_get_cpu_usage_raw(u, &ns); |
3232 | if (r < 0) { | |
66ebf6c0 | 3233 | u->cpu_usage_base = 0; |
5ad096b3 | 3234 | return r; |
b56c28c3 | 3235 | } |
2633eb83 | 3236 | |
66ebf6c0 | 3237 | u->cpu_usage_base = ns; |
4ad49000 | 3238 | return 0; |
4fbf50b3 LP |
3239 | } |
3240 | ||
906c06f6 DM |
3241 | int unit_reset_ip_accounting(Unit *u) { |
3242 | int r = 0, q = 0; | |
3243 | ||
3244 | assert(u); | |
3245 | ||
3246 | if (u->ip_accounting_ingress_map_fd >= 0) | |
3247 | r = bpf_firewall_reset_accounting(u->ip_accounting_ingress_map_fd); | |
3248 | ||
3249 | if (u->ip_accounting_egress_map_fd >= 0) | |
3250 | q = bpf_firewall_reset_accounting(u->ip_accounting_egress_map_fd); | |
3251 | ||
6b659ed8 LP |
3252 | zero(u->ip_accounting_extra); |
3253 | ||
906c06f6 DM |
3254 | return r < 0 ? r : q; |
3255 | } | |
3256 | ||
e7ab4d1a LP |
3257 | void unit_invalidate_cgroup(Unit *u, CGroupMask m) { |
3258 | assert(u); | |
3259 | ||
3260 | if (!UNIT_HAS_CGROUP_CONTEXT(u)) | |
3261 | return; | |
3262 | ||
3263 | if (m == 0) | |
3264 | return; | |
3265 | ||
538b4852 TH |
3266 | /* always invalidate compat pairs together */ |
3267 | if (m & (CGROUP_MASK_IO | CGROUP_MASK_BLKIO)) | |
3268 | m |= CGROUP_MASK_IO | CGROUP_MASK_BLKIO; | |
3269 | ||
7cce4fb7 LP |
3270 | if (m & (CGROUP_MASK_CPU | CGROUP_MASK_CPUACCT)) |
3271 | m |= CGROUP_MASK_CPU | CGROUP_MASK_CPUACCT; | |
3272 | ||
e00068e7 | 3273 | if (FLAGS_SET(u->cgroup_invalidated_mask, m)) /* NOP? */ |
e7ab4d1a LP |
3274 | return; |
3275 | ||
e00068e7 | 3276 | u->cgroup_invalidated_mask |= m; |
91a6073e | 3277 | unit_add_to_cgroup_realize_queue(u); |
e7ab4d1a LP |
3278 | } |
3279 | ||
906c06f6 DM |
3280 | void unit_invalidate_cgroup_bpf(Unit *u) { |
3281 | assert(u); | |
3282 | ||
3283 | if (!UNIT_HAS_CGROUP_CONTEXT(u)) | |
3284 | return; | |
3285 | ||
17f14955 | 3286 | if (u->cgroup_invalidated_mask & CGROUP_MASK_BPF_FIREWALL) /* NOP? */ |
906c06f6 DM |
3287 | return; |
3288 | ||
17f14955 | 3289 | u->cgroup_invalidated_mask |= CGROUP_MASK_BPF_FIREWALL; |
91a6073e | 3290 | unit_add_to_cgroup_realize_queue(u); |
906c06f6 DM |
3291 | |
3292 | /* If we are a slice unit, we also need to put compile a new BPF program for all our children, as the IP access | |
3293 | * list of our children includes our own. */ | |
3294 | if (u->type == UNIT_SLICE) { | |
3295 | Unit *member; | |
3296 | Iterator i; | |
eef85c4a | 3297 | void *v; |
906c06f6 | 3298 | |
eef85c4a | 3299 | HASHMAP_FOREACH_KEY(v, member, u->dependencies[UNIT_BEFORE], i) { |
cb5e3bc3 CD |
3300 | if (UNIT_DEREF(member->slice) == u) |
3301 | unit_invalidate_cgroup_bpf(member); | |
906c06f6 DM |
3302 | } |
3303 | } | |
3304 | } | |
3305 | ||
1d9cc876 LP |
3306 | bool unit_cgroup_delegate(Unit *u) { |
3307 | CGroupContext *c; | |
3308 | ||
3309 | assert(u); | |
3310 | ||
3311 | if (!UNIT_VTABLE(u)->can_delegate) | |
3312 | return false; | |
3313 | ||
3314 | c = unit_get_cgroup_context(u); | |
3315 | if (!c) | |
3316 | return false; | |
3317 | ||
3318 | return c->delegate; | |
3319 | } | |
3320 | ||
e7ab4d1a LP |
3321 | void manager_invalidate_startup_units(Manager *m) { |
3322 | Iterator i; | |
3323 | Unit *u; | |
3324 | ||
3325 | assert(m); | |
3326 | ||
3327 | SET_FOREACH(u, m->startup_units, i) | |
13c31542 | 3328 | unit_invalidate_cgroup(u, CGROUP_MASK_CPU|CGROUP_MASK_IO|CGROUP_MASK_BLKIO); |
e7ab4d1a LP |
3329 | } |
3330 | ||
4ad49000 LP |
3331 | static const char* const cgroup_device_policy_table[_CGROUP_DEVICE_POLICY_MAX] = { |
3332 | [CGROUP_AUTO] = "auto", | |
3333 | [CGROUP_CLOSED] = "closed", | |
3334 | [CGROUP_STRICT] = "strict", | |
3335 | }; | |
4fbf50b3 | 3336 | |
4ad49000 | 3337 | DEFINE_STRING_TABLE_LOOKUP(cgroup_device_policy, CGroupDevicePolicy); |