]>
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 | |
b5efdb8a | 6 | #include "alloc-util.h" |
18c528e9 | 7 | #include "blockdev-util.h" |
906c06f6 | 8 | #include "bpf-firewall.h" |
45c2e068 | 9 | #include "btrfs-util.h" |
084c7007 | 10 | #include "bpf-devices.h" |
6592b975 | 11 | #include "bus-error.h" |
03a7b521 | 12 | #include "cgroup-util.h" |
3ffd4af2 LP |
13 | #include "cgroup.h" |
14 | #include "fd-util.h" | |
0d39fa9c | 15 | #include "fileio.h" |
77601719 | 16 | #include "fs-util.h" |
6bedfcbb | 17 | #include "parse-util.h" |
9eb977db | 18 | #include "path-util.h" |
03a7b521 | 19 | #include "process-util.h" |
c36a69f4 | 20 | #include "procfs-util.h" |
9444b1f2 | 21 | #include "special.h" |
906c06f6 | 22 | #include "stdio-util.h" |
8b43440b | 23 | #include "string-table.h" |
07630cea | 24 | #include "string-util.h" |
cc6271f1 | 25 | #include "virt.h" |
8e274523 | 26 | |
9a054909 LP |
27 | #define CGROUP_CPU_QUOTA_PERIOD_USEC ((usec_t) 100 * USEC_PER_MSEC) |
28 | ||
39b9fefb LP |
29 | /* Returns the log level to use when cgroup attribute writes fail. When an attribute is missing or we have access |
30 | * problems we downgrade to LOG_DEBUG. This is supposed to be nice to container managers and kernels which want to mask | |
31 | * out specific attributes from us. */ | |
32 | #define LOG_LEVEL_CGROUP_WRITE(r) (IN_SET(abs(r), ENOENT, EROFS, EACCES, EPERM) ? LOG_DEBUG : LOG_WARNING) | |
33 | ||
611c4f8a | 34 | bool manager_owns_host_root_cgroup(Manager *m) { |
cc6271f1 LP |
35 | assert(m); |
36 | ||
37 | /* Returns true if we are managing the root cgroup. Note that it isn't sufficient to just check whether the | |
38 | * group root path equals "/" since that will also be the case if CLONE_NEWCGROUP is in the mix. Since there's | |
39 | * appears to be no nice way to detect whether we are in a CLONE_NEWCGROUP namespace we instead just check if | |
40 | * we run in any kind of container virtualization. */ | |
41 | ||
42 | if (detect_container() > 0) | |
43 | return false; | |
44 | ||
57ea45e1 | 45 | return empty_or_root(m->cgroup_root); |
cc6271f1 LP |
46 | } |
47 | ||
611c4f8a | 48 | bool unit_has_host_root_cgroup(Unit *u) { |
f3725e64 LP |
49 | assert(u); |
50 | ||
cc6271f1 LP |
51 | /* Returns whether this unit manages the root cgroup. This will return true if this unit is the root slice and |
52 | * the manager manages the root cgroup. */ | |
f3725e64 | 53 | |
611c4f8a | 54 | if (!manager_owns_host_root_cgroup(u->manager)) |
f3725e64 LP |
55 | return false; |
56 | ||
cc6271f1 | 57 | return unit_has_name(u, SPECIAL_ROOT_SLICE); |
f3725e64 LP |
58 | } |
59 | ||
293d32df LP |
60 | static int set_attribute_and_warn(Unit *u, const char *controller, const char *attribute, const char *value) { |
61 | int r; | |
62 | ||
63 | r = cg_set_attribute(controller, u->cgroup_path, attribute, value); | |
64 | if (r < 0) | |
65 | log_unit_full(u, LOG_LEVEL_CGROUP_WRITE(r), r, "Failed to set '%s' attribute on '%s' to '%.*s': %m", | |
66 | strna(attribute), isempty(u->cgroup_path) ? "/" : u->cgroup_path, (int) strcspn(value, NEWLINE), value); | |
67 | ||
68 | return r; | |
69 | } | |
70 | ||
2b40998d | 71 | static void cgroup_compat_warn(void) { |
128fadc9 TH |
72 | static bool cgroup_compat_warned = false; |
73 | ||
74 | if (cgroup_compat_warned) | |
75 | return; | |
76 | ||
cc6271f1 LP |
77 | log_warning("cgroup compatibility translation between legacy and unified hierarchy settings activated. " |
78 | "See cgroup-compat debug messages for details."); | |
79 | ||
128fadc9 TH |
80 | cgroup_compat_warned = true; |
81 | } | |
82 | ||
83 | #define log_cgroup_compat(unit, fmt, ...) do { \ | |
84 | cgroup_compat_warn(); \ | |
85 | log_unit_debug(unit, "cgroup-compat: " fmt, ##__VA_ARGS__); \ | |
2b40998d | 86 | } while (false) |
128fadc9 | 87 | |
4ad49000 LP |
88 | void cgroup_context_init(CGroupContext *c) { |
89 | assert(c); | |
90 | ||
de8a711a | 91 | /* Initialize everything to the kernel defaults. */ |
4ad49000 | 92 | |
de8a711a LP |
93 | *c = (CGroupContext) { |
94 | .cpu_weight = CGROUP_WEIGHT_INVALID, | |
95 | .startup_cpu_weight = CGROUP_WEIGHT_INVALID, | |
96 | .cpu_quota_per_sec_usec = USEC_INFINITY, | |
66ebf6c0 | 97 | |
de8a711a LP |
98 | .cpu_shares = CGROUP_CPU_SHARES_INVALID, |
99 | .startup_cpu_shares = CGROUP_CPU_SHARES_INVALID, | |
d53d9474 | 100 | |
de8a711a LP |
101 | .memory_high = CGROUP_LIMIT_MAX, |
102 | .memory_max = CGROUP_LIMIT_MAX, | |
103 | .memory_swap_max = CGROUP_LIMIT_MAX, | |
da4d897e | 104 | |
de8a711a | 105 | .memory_limit = CGROUP_LIMIT_MAX, |
b2f8b02e | 106 | |
de8a711a LP |
107 | .io_weight = CGROUP_WEIGHT_INVALID, |
108 | .startup_io_weight = CGROUP_WEIGHT_INVALID, | |
13c31542 | 109 | |
de8a711a LP |
110 | .blockio_weight = CGROUP_BLKIO_WEIGHT_INVALID, |
111 | .startup_blockio_weight = CGROUP_BLKIO_WEIGHT_INVALID, | |
d53d9474 | 112 | |
de8a711a LP |
113 | .tasks_max = CGROUP_LIMIT_MAX, |
114 | }; | |
4ad49000 | 115 | } |
8e274523 | 116 | |
4ad49000 LP |
117 | void cgroup_context_free_device_allow(CGroupContext *c, CGroupDeviceAllow *a) { |
118 | assert(c); | |
119 | assert(a); | |
120 | ||
71fda00f | 121 | LIST_REMOVE(device_allow, c->device_allow, a); |
4ad49000 LP |
122 | free(a->path); |
123 | free(a); | |
124 | } | |
125 | ||
13c31542 TH |
126 | void cgroup_context_free_io_device_weight(CGroupContext *c, CGroupIODeviceWeight *w) { |
127 | assert(c); | |
128 | assert(w); | |
129 | ||
130 | LIST_REMOVE(device_weights, c->io_device_weights, w); | |
131 | free(w->path); | |
132 | free(w); | |
133 | } | |
134 | ||
6ae4283c TH |
135 | void cgroup_context_free_io_device_latency(CGroupContext *c, CGroupIODeviceLatency *l) { |
136 | assert(c); | |
137 | assert(l); | |
138 | ||
139 | LIST_REMOVE(device_latencies, c->io_device_latencies, l); | |
140 | free(l->path); | |
141 | free(l); | |
142 | } | |
143 | ||
13c31542 TH |
144 | void cgroup_context_free_io_device_limit(CGroupContext *c, CGroupIODeviceLimit *l) { |
145 | assert(c); | |
146 | assert(l); | |
147 | ||
148 | LIST_REMOVE(device_limits, c->io_device_limits, l); | |
149 | free(l->path); | |
150 | free(l); | |
151 | } | |
152 | ||
4ad49000 LP |
153 | void cgroup_context_free_blockio_device_weight(CGroupContext *c, CGroupBlockIODeviceWeight *w) { |
154 | assert(c); | |
155 | assert(w); | |
156 | ||
71fda00f | 157 | LIST_REMOVE(device_weights, c->blockio_device_weights, w); |
4ad49000 LP |
158 | free(w->path); |
159 | free(w); | |
160 | } | |
161 | ||
162 | void cgroup_context_free_blockio_device_bandwidth(CGroupContext *c, CGroupBlockIODeviceBandwidth *b) { | |
163 | assert(c); | |
8e274523 | 164 | assert(b); |
8e274523 | 165 | |
71fda00f | 166 | LIST_REMOVE(device_bandwidths, c->blockio_device_bandwidths, b); |
4ad49000 LP |
167 | free(b->path); |
168 | free(b); | |
169 | } | |
170 | ||
171 | void cgroup_context_done(CGroupContext *c) { | |
172 | assert(c); | |
173 | ||
13c31542 TH |
174 | while (c->io_device_weights) |
175 | cgroup_context_free_io_device_weight(c, c->io_device_weights); | |
176 | ||
6ae4283c TH |
177 | while (c->io_device_latencies) |
178 | cgroup_context_free_io_device_latency(c, c->io_device_latencies); | |
179 | ||
13c31542 TH |
180 | while (c->io_device_limits) |
181 | cgroup_context_free_io_device_limit(c, c->io_device_limits); | |
182 | ||
4ad49000 LP |
183 | while (c->blockio_device_weights) |
184 | cgroup_context_free_blockio_device_weight(c, c->blockio_device_weights); | |
185 | ||
186 | while (c->blockio_device_bandwidths) | |
187 | cgroup_context_free_blockio_device_bandwidth(c, c->blockio_device_bandwidths); | |
188 | ||
189 | while (c->device_allow) | |
190 | cgroup_context_free_device_allow(c, c->device_allow); | |
6a48d82f DM |
191 | |
192 | c->ip_address_allow = ip_address_access_free_all(c->ip_address_allow); | |
193 | c->ip_address_deny = ip_address_access_free_all(c->ip_address_deny); | |
4ad49000 LP |
194 | } |
195 | ||
196 | void cgroup_context_dump(CGroupContext *c, FILE* f, const char *prefix) { | |
13c31542 TH |
197 | CGroupIODeviceLimit *il; |
198 | CGroupIODeviceWeight *iw; | |
6ae4283c | 199 | CGroupIODeviceLatency *l; |
4ad49000 LP |
200 | CGroupBlockIODeviceBandwidth *b; |
201 | CGroupBlockIODeviceWeight *w; | |
202 | CGroupDeviceAllow *a; | |
c21c9906 | 203 | IPAddressAccessItem *iaai; |
9a054909 | 204 | char u[FORMAT_TIMESPAN_MAX]; |
4ad49000 LP |
205 | |
206 | assert(c); | |
207 | assert(f); | |
208 | ||
209 | prefix = strempty(prefix); | |
210 | ||
211 | fprintf(f, | |
212 | "%sCPUAccounting=%s\n" | |
13c31542 | 213 | "%sIOAccounting=%s\n" |
4ad49000 LP |
214 | "%sBlockIOAccounting=%s\n" |
215 | "%sMemoryAccounting=%s\n" | |
d53d9474 | 216 | "%sTasksAccounting=%s\n" |
c21c9906 | 217 | "%sIPAccounting=%s\n" |
66ebf6c0 TH |
218 | "%sCPUWeight=%" PRIu64 "\n" |
219 | "%sStartupCPUWeight=%" PRIu64 "\n" | |
d53d9474 LP |
220 | "%sCPUShares=%" PRIu64 "\n" |
221 | "%sStartupCPUShares=%" PRIu64 "\n" | |
b2f8b02e | 222 | "%sCPUQuotaPerSecSec=%s\n" |
13c31542 TH |
223 | "%sIOWeight=%" PRIu64 "\n" |
224 | "%sStartupIOWeight=%" PRIu64 "\n" | |
d53d9474 LP |
225 | "%sBlockIOWeight=%" PRIu64 "\n" |
226 | "%sStartupBlockIOWeight=%" PRIu64 "\n" | |
48422635 | 227 | "%sMemoryMin=%" PRIu64 "\n" |
da4d897e TH |
228 | "%sMemoryLow=%" PRIu64 "\n" |
229 | "%sMemoryHigh=%" PRIu64 "\n" | |
230 | "%sMemoryMax=%" PRIu64 "\n" | |
96e131ea | 231 | "%sMemorySwapMax=%" PRIu64 "\n" |
4ad49000 | 232 | "%sMemoryLimit=%" PRIu64 "\n" |
03a7b521 | 233 | "%sTasksMax=%" PRIu64 "\n" |
a931ad47 LP |
234 | "%sDevicePolicy=%s\n" |
235 | "%sDelegate=%s\n", | |
4ad49000 | 236 | prefix, yes_no(c->cpu_accounting), |
13c31542 | 237 | prefix, yes_no(c->io_accounting), |
4ad49000 LP |
238 | prefix, yes_no(c->blockio_accounting), |
239 | prefix, yes_no(c->memory_accounting), | |
d53d9474 | 240 | prefix, yes_no(c->tasks_accounting), |
c21c9906 | 241 | prefix, yes_no(c->ip_accounting), |
66ebf6c0 TH |
242 | prefix, c->cpu_weight, |
243 | prefix, c->startup_cpu_weight, | |
4ad49000 | 244 | prefix, c->cpu_shares, |
95ae05c0 | 245 | prefix, c->startup_cpu_shares, |
b1d6dcf5 | 246 | prefix, format_timespan(u, sizeof(u), c->cpu_quota_per_sec_usec, 1), |
13c31542 TH |
247 | prefix, c->io_weight, |
248 | prefix, c->startup_io_weight, | |
4ad49000 | 249 | prefix, c->blockio_weight, |
95ae05c0 | 250 | prefix, c->startup_blockio_weight, |
48422635 | 251 | prefix, c->memory_min, |
da4d897e TH |
252 | prefix, c->memory_low, |
253 | prefix, c->memory_high, | |
254 | prefix, c->memory_max, | |
96e131ea | 255 | prefix, c->memory_swap_max, |
4ad49000 | 256 | prefix, c->memory_limit, |
03a7b521 | 257 | prefix, c->tasks_max, |
a931ad47 LP |
258 | prefix, cgroup_device_policy_to_string(c->device_policy), |
259 | prefix, yes_no(c->delegate)); | |
4ad49000 | 260 | |
02638280 LP |
261 | if (c->delegate) { |
262 | _cleanup_free_ char *t = NULL; | |
263 | ||
264 | (void) cg_mask_to_string(c->delegate_controllers, &t); | |
265 | ||
47a78d41 | 266 | fprintf(f, "%sDelegateControllers=%s\n", |
02638280 LP |
267 | prefix, |
268 | strempty(t)); | |
269 | } | |
270 | ||
4ad49000 LP |
271 | LIST_FOREACH(device_allow, a, c->device_allow) |
272 | fprintf(f, | |
273 | "%sDeviceAllow=%s %s%s%s\n", | |
274 | prefix, | |
275 | a->path, | |
276 | a->r ? "r" : "", a->w ? "w" : "", a->m ? "m" : ""); | |
277 | ||
13c31542 TH |
278 | LIST_FOREACH(device_weights, iw, c->io_device_weights) |
279 | fprintf(f, | |
6ae4283c | 280 | "%sIODeviceWeight=%s %" PRIu64 "\n", |
13c31542 TH |
281 | prefix, |
282 | iw->path, | |
283 | iw->weight); | |
284 | ||
6ae4283c TH |
285 | LIST_FOREACH(device_latencies, l, c->io_device_latencies) |
286 | fprintf(f, | |
287 | "%sIODeviceLatencyTargetSec=%s %s\n", | |
288 | prefix, | |
289 | l->path, | |
290 | format_timespan(u, sizeof(u), l->target_usec, 1)); | |
291 | ||
13c31542 TH |
292 | LIST_FOREACH(device_limits, il, c->io_device_limits) { |
293 | char buf[FORMAT_BYTES_MAX]; | |
9be57249 TH |
294 | CGroupIOLimitType type; |
295 | ||
296 | for (type = 0; type < _CGROUP_IO_LIMIT_TYPE_MAX; type++) | |
297 | if (il->limits[type] != cgroup_io_limit_defaults[type]) | |
298 | fprintf(f, | |
299 | "%s%s=%s %s\n", | |
300 | prefix, | |
301 | cgroup_io_limit_type_to_string(type), | |
302 | il->path, | |
303 | format_bytes(buf, sizeof(buf), il->limits[type])); | |
13c31542 TH |
304 | } |
305 | ||
4ad49000 LP |
306 | LIST_FOREACH(device_weights, w, c->blockio_device_weights) |
307 | fprintf(f, | |
d53d9474 | 308 | "%sBlockIODeviceWeight=%s %" PRIu64, |
4ad49000 LP |
309 | prefix, |
310 | w->path, | |
311 | w->weight); | |
312 | ||
313 | LIST_FOREACH(device_bandwidths, b, c->blockio_device_bandwidths) { | |
314 | char buf[FORMAT_BYTES_MAX]; | |
315 | ||
979d0311 TH |
316 | if (b->rbps != CGROUP_LIMIT_MAX) |
317 | fprintf(f, | |
318 | "%sBlockIOReadBandwidth=%s %s\n", | |
319 | prefix, | |
320 | b->path, | |
321 | format_bytes(buf, sizeof(buf), b->rbps)); | |
322 | if (b->wbps != CGROUP_LIMIT_MAX) | |
323 | fprintf(f, | |
324 | "%sBlockIOWriteBandwidth=%s %s\n", | |
325 | prefix, | |
326 | b->path, | |
327 | format_bytes(buf, sizeof(buf), b->wbps)); | |
4ad49000 | 328 | } |
c21c9906 LP |
329 | |
330 | LIST_FOREACH(items, iaai, c->ip_address_allow) { | |
331 | _cleanup_free_ char *k = NULL; | |
332 | ||
333 | (void) in_addr_to_string(iaai->family, &iaai->address, &k); | |
334 | fprintf(f, "%sIPAddressAllow=%s/%u\n", prefix, strnull(k), iaai->prefixlen); | |
335 | } | |
336 | ||
337 | LIST_FOREACH(items, iaai, c->ip_address_deny) { | |
338 | _cleanup_free_ char *k = NULL; | |
339 | ||
340 | (void) in_addr_to_string(iaai->family, &iaai->address, &k); | |
341 | fprintf(f, "%sIPAddressDeny=%s/%u\n", prefix, strnull(k), iaai->prefixlen); | |
342 | } | |
4ad49000 LP |
343 | } |
344 | ||
fd870bac YW |
345 | int cgroup_add_device_allow(CGroupContext *c, const char *dev, const char *mode) { |
346 | _cleanup_free_ CGroupDeviceAllow *a = NULL; | |
347 | _cleanup_free_ char *d = NULL; | |
348 | ||
349 | assert(c); | |
350 | assert(dev); | |
351 | assert(isempty(mode) || in_charset(mode, "rwm")); | |
352 | ||
353 | a = new(CGroupDeviceAllow, 1); | |
354 | if (!a) | |
355 | return -ENOMEM; | |
356 | ||
357 | d = strdup(dev); | |
358 | if (!d) | |
359 | return -ENOMEM; | |
360 | ||
361 | *a = (CGroupDeviceAllow) { | |
362 | .path = TAKE_PTR(d), | |
490c5a37 LP |
363 | .r = isempty(mode) || strchr(mode, 'r'), |
364 | .w = isempty(mode) || strchr(mode, 'w'), | |
365 | .m = isempty(mode) || strchr(mode, 'm'), | |
fd870bac YW |
366 | }; |
367 | ||
368 | LIST_PREPEND(device_allow, c->device_allow, a); | |
369 | TAKE_PTR(a); | |
370 | ||
371 | return 0; | |
372 | } | |
373 | ||
45c2e068 | 374 | static int lookup_block_device(const char *p, dev_t *ret) { |
4ad49000 | 375 | struct stat st; |
45c2e068 | 376 | int r; |
4ad49000 LP |
377 | |
378 | assert(p); | |
45c2e068 | 379 | assert(ret); |
4ad49000 | 380 | |
b1c05b98 | 381 | if (stat(p, &st) < 0) |
45c2e068 | 382 | return log_warning_errno(errno, "Couldn't stat device '%s': %m", p); |
8e274523 | 383 | |
4ad49000 | 384 | if (S_ISBLK(st.st_mode)) |
45c2e068 LP |
385 | *ret = st.st_rdev; |
386 | else if (major(st.st_dev) != 0) | |
387 | *ret = st.st_dev; /* If this is not a device node then use the block device this file is stored on */ | |
388 | else { | |
389 | /* If this is btrfs, getting the backing block device is a bit harder */ | |
390 | r = btrfs_get_block_device(p, ret); | |
391 | if (r < 0 && r != -ENOTTY) | |
392 | return log_warning_errno(r, "Failed to determine block device backing btrfs file system '%s': %m", p); | |
393 | if (r == -ENOTTY) { | |
394 | log_warning("'%s' is not a block device node, and file system block device cannot be determined or is not local.", p); | |
395 | return -ENODEV; | |
396 | } | |
4ad49000 | 397 | } |
8e274523 | 398 | |
45c2e068 LP |
399 | /* If this is a LUKS device, try to get the originating block device */ |
400 | (void) block_get_originating(*ret, ret); | |
401 | ||
402 | /* If this is a partition, try to get the originating block device */ | |
403 | (void) block_get_whole_disk(*ret, ret); | |
8e274523 | 404 | return 0; |
8e274523 LP |
405 | } |
406 | ||
084c7007 | 407 | static int whitelist_device(BPFProgram *prog, const char *path, const char *node, const char *acc) { |
4ad49000 | 408 | struct stat st; |
b200489b | 409 | bool ignore_notfound; |
8c6db833 | 410 | int r; |
8e274523 | 411 | |
4ad49000 LP |
412 | assert(path); |
413 | assert(acc); | |
8e274523 | 414 | |
b200489b DR |
415 | if (node[0] == '-') { |
416 | /* Non-existent paths starting with "-" must be silently ignored */ | |
417 | node++; | |
418 | ignore_notfound = true; | |
419 | } else | |
420 | ignore_notfound = false; | |
421 | ||
4ad49000 | 422 | if (stat(node, &st) < 0) { |
b200489b | 423 | if (errno == ENOENT && ignore_notfound) |
e7330dfe DP |
424 | return 0; |
425 | ||
426 | return log_warning_errno(errno, "Couldn't stat device %s: %m", node); | |
4ad49000 LP |
427 | } |
428 | ||
429 | if (!S_ISCHR(st.st_mode) && !S_ISBLK(st.st_mode)) { | |
430 | log_warning("%s is not a device.", node); | |
431 | return -ENODEV; | |
432 | } | |
433 | ||
084c7007 RG |
434 | if (cg_all_unified() > 0) { |
435 | if (!prog) | |
436 | return 0; | |
4ad49000 | 437 | |
b9839ac9 LP |
438 | return cgroup_bpf_whitelist_device(prog, S_ISCHR(st.st_mode) ? BPF_DEVCG_DEV_CHAR : BPF_DEVCG_DEV_BLOCK, |
439 | major(st.st_rdev), minor(st.st_rdev), acc); | |
440 | ||
084c7007 RG |
441 | } else { |
442 | char buf[2+DECIMAL_STR_MAX(dev_t)*2+2+4]; | |
443 | ||
444 | sprintf(buf, | |
445 | "%c %u:%u %s", | |
446 | S_ISCHR(st.st_mode) ? 'c' : 'b', | |
447 | major(st.st_rdev), minor(st.st_rdev), | |
448 | acc); | |
449 | ||
8c838407 LP |
450 | /* Changing the devices list of a populated cgroup might result in EINVAL, hence ignore EINVAL here. */ |
451 | ||
084c7007 RG |
452 | r = cg_set_attribute("devices", path, "devices.allow", buf); |
453 | if (r < 0) | |
2c74e12b | 454 | return log_full_errno(IN_SET(r, -ENOENT, -EROFS, -EINVAL, -EACCES, -EPERM) ? LOG_DEBUG : LOG_WARNING, |
b9839ac9 | 455 | r, "Failed to set devices.allow on %s: %m", path); |
4ad49000 | 456 | |
b9839ac9 LP |
457 | return 0; |
458 | } | |
8e274523 LP |
459 | } |
460 | ||
084c7007 | 461 | static int whitelist_major(BPFProgram *prog, const char *path, const char *name, char type, const char *acc) { |
90060676 | 462 | _cleanup_fclose_ FILE *f = NULL; |
084c7007 | 463 | char *p, *w; |
90060676 LP |
464 | bool good = false; |
465 | int r; | |
466 | ||
467 | assert(path); | |
468 | assert(acc); | |
4c701096 | 469 | assert(IN_SET(type, 'b', 'c')); |
90060676 LP |
470 | |
471 | f = fopen("/proc/devices", "re"); | |
4a62c710 MS |
472 | if (!f) |
473 | return log_warning_errno(errno, "Cannot open /proc/devices to resolve %s (%c): %m", name, type); | |
90060676 | 474 | |
c66e60a8 LP |
475 | for (;;) { |
476 | _cleanup_free_ char *line = NULL; | |
90060676 LP |
477 | unsigned maj; |
478 | ||
c66e60a8 LP |
479 | r = read_line(f, LONG_LINE_MAX, &line); |
480 | if (r < 0) | |
481 | return log_warning_errno(r, "Failed to read /proc/devices: %m"); | |
482 | if (r == 0) | |
483 | break; | |
90060676 LP |
484 | |
485 | if (type == 'c' && streq(line, "Character devices:")) { | |
486 | good = true; | |
487 | continue; | |
488 | } | |
489 | ||
490 | if (type == 'b' && streq(line, "Block devices:")) { | |
491 | good = true; | |
492 | continue; | |
493 | } | |
494 | ||
495 | if (isempty(line)) { | |
496 | good = false; | |
497 | continue; | |
498 | } | |
499 | ||
500 | if (!good) | |
501 | continue; | |
502 | ||
503 | p = strstrip(line); | |
504 | ||
505 | w = strpbrk(p, WHITESPACE); | |
506 | if (!w) | |
507 | continue; | |
508 | *w = 0; | |
509 | ||
510 | r = safe_atou(p, &maj); | |
511 | if (r < 0) | |
512 | continue; | |
513 | if (maj <= 0) | |
514 | continue; | |
515 | ||
516 | w++; | |
517 | w += strspn(w, WHITESPACE); | |
e41969e3 LP |
518 | |
519 | if (fnmatch(name, w, 0) != 0) | |
90060676 LP |
520 | continue; |
521 | ||
084c7007 RG |
522 | if (cg_all_unified() > 0) { |
523 | if (!prog) | |
524 | continue; | |
90060676 | 525 | |
913c898c LP |
526 | (void) cgroup_bpf_whitelist_major(prog, |
527 | type == 'c' ? BPF_DEVCG_DEV_CHAR : BPF_DEVCG_DEV_BLOCK, | |
528 | maj, acc); | |
084c7007 RG |
529 | } else { |
530 | char buf[2+DECIMAL_STR_MAX(unsigned)+3+4]; | |
531 | ||
532 | sprintf(buf, | |
533 | "%c %u:* %s", | |
534 | type, | |
535 | maj, | |
536 | acc); | |
537 | ||
8c838407 LP |
538 | /* Changing the devices list of a populated cgroup might result in EINVAL, hence ignore EINVAL |
539 | * here. */ | |
540 | ||
084c7007 RG |
541 | r = cg_set_attribute("devices", path, "devices.allow", buf); |
542 | if (r < 0) | |
2c74e12b | 543 | log_full_errno(IN_SET(r, -ENOENT, -EROFS, -EINVAL, -EACCES, -EPERM) ? LOG_DEBUG : LOG_WARNING, |
084c7007 RG |
544 | r, "Failed to set devices.allow on %s: %m", path); |
545 | } | |
90060676 LP |
546 | } |
547 | ||
548 | return 0; | |
90060676 LP |
549 | } |
550 | ||
66ebf6c0 TH |
551 | static bool cgroup_context_has_cpu_weight(CGroupContext *c) { |
552 | return c->cpu_weight != CGROUP_WEIGHT_INVALID || | |
553 | c->startup_cpu_weight != CGROUP_WEIGHT_INVALID; | |
554 | } | |
555 | ||
556 | static bool cgroup_context_has_cpu_shares(CGroupContext *c) { | |
557 | return c->cpu_shares != CGROUP_CPU_SHARES_INVALID || | |
558 | c->startup_cpu_shares != CGROUP_CPU_SHARES_INVALID; | |
559 | } | |
560 | ||
561 | static uint64_t cgroup_context_cpu_weight(CGroupContext *c, ManagerState state) { | |
562 | if (IN_SET(state, MANAGER_STARTING, MANAGER_INITIALIZING) && | |
563 | c->startup_cpu_weight != CGROUP_WEIGHT_INVALID) | |
564 | return c->startup_cpu_weight; | |
565 | else if (c->cpu_weight != CGROUP_WEIGHT_INVALID) | |
566 | return c->cpu_weight; | |
567 | else | |
568 | return CGROUP_WEIGHT_DEFAULT; | |
569 | } | |
570 | ||
571 | static uint64_t cgroup_context_cpu_shares(CGroupContext *c, ManagerState state) { | |
572 | if (IN_SET(state, MANAGER_STARTING, MANAGER_INITIALIZING) && | |
573 | c->startup_cpu_shares != CGROUP_CPU_SHARES_INVALID) | |
574 | return c->startup_cpu_shares; | |
575 | else if (c->cpu_shares != CGROUP_CPU_SHARES_INVALID) | |
576 | return c->cpu_shares; | |
577 | else | |
578 | return CGROUP_CPU_SHARES_DEFAULT; | |
579 | } | |
580 | ||
581 | static void cgroup_apply_unified_cpu_config(Unit *u, uint64_t weight, uint64_t quota) { | |
582 | char buf[MAX(DECIMAL_STR_MAX(uint64_t) + 1, (DECIMAL_STR_MAX(usec_t) + 1) * 2)]; | |
66ebf6c0 TH |
583 | |
584 | xsprintf(buf, "%" PRIu64 "\n", weight); | |
293d32df | 585 | (void) set_attribute_and_warn(u, "cpu", "cpu.weight", buf); |
66ebf6c0 TH |
586 | |
587 | if (quota != USEC_INFINITY) | |
588 | xsprintf(buf, USEC_FMT " " USEC_FMT "\n", | |
589 | quota * CGROUP_CPU_QUOTA_PERIOD_USEC / USEC_PER_SEC, CGROUP_CPU_QUOTA_PERIOD_USEC); | |
590 | else | |
591 | xsprintf(buf, "max " USEC_FMT "\n", CGROUP_CPU_QUOTA_PERIOD_USEC); | |
592 | ||
293d32df | 593 | (void) set_attribute_and_warn(u, "cpu", "cpu.max", buf); |
66ebf6c0 TH |
594 | } |
595 | ||
596 | static void cgroup_apply_legacy_cpu_config(Unit *u, uint64_t shares, uint64_t quota) { | |
597 | char buf[MAX(DECIMAL_STR_MAX(uint64_t), DECIMAL_STR_MAX(usec_t)) + 1]; | |
66ebf6c0 TH |
598 | |
599 | xsprintf(buf, "%" PRIu64 "\n", shares); | |
293d32df | 600 | (void) set_attribute_and_warn(u, "cpu", "cpu.shares", buf); |
66ebf6c0 TH |
601 | |
602 | xsprintf(buf, USEC_FMT "\n", CGROUP_CPU_QUOTA_PERIOD_USEC); | |
293d32df | 603 | (void) set_attribute_and_warn(u, "cpu", "cpu.cfs_period_us", buf); |
66ebf6c0 TH |
604 | |
605 | if (quota != USEC_INFINITY) { | |
606 | xsprintf(buf, USEC_FMT "\n", quota * CGROUP_CPU_QUOTA_PERIOD_USEC / USEC_PER_SEC); | |
293d32df | 607 | (void) set_attribute_and_warn(u, "cpu", "cpu.cfs_quota_us", buf); |
66ebf6c0 | 608 | } else |
293d32df | 609 | (void) set_attribute_and_warn(u, "cpu", "cpu.cfs_quota_us", "-1"); |
66ebf6c0 TH |
610 | } |
611 | ||
612 | static uint64_t cgroup_cpu_shares_to_weight(uint64_t shares) { | |
613 | return CLAMP(shares * CGROUP_WEIGHT_DEFAULT / CGROUP_CPU_SHARES_DEFAULT, | |
614 | CGROUP_WEIGHT_MIN, CGROUP_WEIGHT_MAX); | |
615 | } | |
616 | ||
617 | static uint64_t cgroup_cpu_weight_to_shares(uint64_t weight) { | |
618 | return CLAMP(weight * CGROUP_CPU_SHARES_DEFAULT / CGROUP_WEIGHT_DEFAULT, | |
619 | CGROUP_CPU_SHARES_MIN, CGROUP_CPU_SHARES_MAX); | |
620 | } | |
621 | ||
508c45da | 622 | static bool cgroup_context_has_io_config(CGroupContext *c) { |
538b4852 TH |
623 | return c->io_accounting || |
624 | c->io_weight != CGROUP_WEIGHT_INVALID || | |
625 | c->startup_io_weight != CGROUP_WEIGHT_INVALID || | |
626 | c->io_device_weights || | |
6ae4283c | 627 | c->io_device_latencies || |
538b4852 TH |
628 | c->io_device_limits; |
629 | } | |
630 | ||
508c45da | 631 | static bool cgroup_context_has_blockio_config(CGroupContext *c) { |
538b4852 TH |
632 | return c->blockio_accounting || |
633 | c->blockio_weight != CGROUP_BLKIO_WEIGHT_INVALID || | |
634 | c->startup_blockio_weight != CGROUP_BLKIO_WEIGHT_INVALID || | |
635 | c->blockio_device_weights || | |
636 | c->blockio_device_bandwidths; | |
637 | } | |
638 | ||
508c45da | 639 | static uint64_t cgroup_context_io_weight(CGroupContext *c, ManagerState state) { |
64faf04c TH |
640 | if (IN_SET(state, MANAGER_STARTING, MANAGER_INITIALIZING) && |
641 | c->startup_io_weight != CGROUP_WEIGHT_INVALID) | |
642 | return c->startup_io_weight; | |
643 | else if (c->io_weight != CGROUP_WEIGHT_INVALID) | |
644 | return c->io_weight; | |
645 | else | |
646 | return CGROUP_WEIGHT_DEFAULT; | |
647 | } | |
648 | ||
508c45da | 649 | static uint64_t cgroup_context_blkio_weight(CGroupContext *c, ManagerState state) { |
64faf04c TH |
650 | if (IN_SET(state, MANAGER_STARTING, MANAGER_INITIALIZING) && |
651 | c->startup_blockio_weight != CGROUP_BLKIO_WEIGHT_INVALID) | |
652 | return c->startup_blockio_weight; | |
653 | else if (c->blockio_weight != CGROUP_BLKIO_WEIGHT_INVALID) | |
654 | return c->blockio_weight; | |
655 | else | |
656 | return CGROUP_BLKIO_WEIGHT_DEFAULT; | |
657 | } | |
658 | ||
508c45da | 659 | static uint64_t cgroup_weight_blkio_to_io(uint64_t blkio_weight) { |
538b4852 TH |
660 | return CLAMP(blkio_weight * CGROUP_WEIGHT_DEFAULT / CGROUP_BLKIO_WEIGHT_DEFAULT, |
661 | CGROUP_WEIGHT_MIN, CGROUP_WEIGHT_MAX); | |
662 | } | |
663 | ||
508c45da | 664 | static uint64_t cgroup_weight_io_to_blkio(uint64_t io_weight) { |
538b4852 TH |
665 | return CLAMP(io_weight * CGROUP_BLKIO_WEIGHT_DEFAULT / CGROUP_WEIGHT_DEFAULT, |
666 | CGROUP_BLKIO_WEIGHT_MIN, CGROUP_BLKIO_WEIGHT_MAX); | |
667 | } | |
668 | ||
f29ff115 | 669 | static void cgroup_apply_io_device_weight(Unit *u, const char *dev_path, uint64_t io_weight) { |
64faf04c TH |
670 | char buf[DECIMAL_STR_MAX(dev_t)*2+2+DECIMAL_STR_MAX(uint64_t)+1]; |
671 | dev_t dev; | |
672 | int r; | |
673 | ||
674 | r = lookup_block_device(dev_path, &dev); | |
675 | if (r < 0) | |
676 | return; | |
677 | ||
678 | xsprintf(buf, "%u:%u %" PRIu64 "\n", major(dev), minor(dev), io_weight); | |
293d32df | 679 | (void) set_attribute_and_warn(u, "io", "io.weight", buf); |
64faf04c TH |
680 | } |
681 | ||
f29ff115 | 682 | static void cgroup_apply_blkio_device_weight(Unit *u, const char *dev_path, uint64_t blkio_weight) { |
64faf04c TH |
683 | char buf[DECIMAL_STR_MAX(dev_t)*2+2+DECIMAL_STR_MAX(uint64_t)+1]; |
684 | dev_t dev; | |
685 | int r; | |
686 | ||
687 | r = lookup_block_device(dev_path, &dev); | |
688 | if (r < 0) | |
689 | return; | |
690 | ||
691 | xsprintf(buf, "%u:%u %" PRIu64 "\n", major(dev), minor(dev), blkio_weight); | |
293d32df | 692 | (void) set_attribute_and_warn(u, "blkio", "blkio.weight_device", buf); |
64faf04c TH |
693 | } |
694 | ||
6ae4283c TH |
695 | static void cgroup_apply_io_device_latency(Unit *u, const char *dev_path, usec_t target) { |
696 | char buf[DECIMAL_STR_MAX(dev_t)*2+2+7+DECIMAL_STR_MAX(uint64_t)+1]; | |
697 | dev_t dev; | |
698 | int r; | |
699 | ||
700 | r = lookup_block_device(dev_path, &dev); | |
701 | if (r < 0) | |
702 | return; | |
703 | ||
704 | if (target != USEC_INFINITY) | |
705 | xsprintf(buf, "%u:%u target=%" PRIu64 "\n", major(dev), minor(dev), target); | |
706 | else | |
707 | xsprintf(buf, "%u:%u target=max\n", major(dev), minor(dev)); | |
708 | ||
293d32df | 709 | (void) set_attribute_and_warn(u, "io", "io.latency", buf); |
6ae4283c TH |
710 | } |
711 | ||
17ae2780 | 712 | static void cgroup_apply_io_device_limit(Unit *u, const char *dev_path, uint64_t *limits) { |
64faf04c TH |
713 | char limit_bufs[_CGROUP_IO_LIMIT_TYPE_MAX][DECIMAL_STR_MAX(uint64_t)]; |
714 | char buf[DECIMAL_STR_MAX(dev_t)*2+2+(6+DECIMAL_STR_MAX(uint64_t)+1)*4]; | |
715 | CGroupIOLimitType type; | |
716 | dev_t dev; | |
64faf04c TH |
717 | int r; |
718 | ||
719 | r = lookup_block_device(dev_path, &dev); | |
720 | if (r < 0) | |
17ae2780 | 721 | return; |
64faf04c | 722 | |
17ae2780 LP |
723 | for (type = 0; type < _CGROUP_IO_LIMIT_TYPE_MAX; type++) |
724 | if (limits[type] != cgroup_io_limit_defaults[type]) | |
64faf04c | 725 | xsprintf(limit_bufs[type], "%" PRIu64, limits[type]); |
17ae2780 | 726 | else |
64faf04c | 727 | xsprintf(limit_bufs[type], "%s", limits[type] == CGROUP_LIMIT_MAX ? "max" : "0"); |
64faf04c TH |
728 | |
729 | xsprintf(buf, "%u:%u rbps=%s wbps=%s riops=%s wiops=%s\n", major(dev), minor(dev), | |
730 | limit_bufs[CGROUP_IO_RBPS_MAX], limit_bufs[CGROUP_IO_WBPS_MAX], | |
731 | limit_bufs[CGROUP_IO_RIOPS_MAX], limit_bufs[CGROUP_IO_WIOPS_MAX]); | |
293d32df | 732 | (void) set_attribute_and_warn(u, "io", "io.max", buf); |
64faf04c TH |
733 | } |
734 | ||
17ae2780 | 735 | static void cgroup_apply_blkio_device_limit(Unit *u, const char *dev_path, uint64_t rbps, uint64_t wbps) { |
64faf04c TH |
736 | char buf[DECIMAL_STR_MAX(dev_t)*2+2+DECIMAL_STR_MAX(uint64_t)+1]; |
737 | dev_t dev; | |
64faf04c TH |
738 | int r; |
739 | ||
740 | r = lookup_block_device(dev_path, &dev); | |
741 | if (r < 0) | |
17ae2780 | 742 | return; |
64faf04c | 743 | |
64faf04c | 744 | sprintf(buf, "%u:%u %" PRIu64 "\n", major(dev), minor(dev), rbps); |
293d32df | 745 | (void) set_attribute_and_warn(u, "blkio", "blkio.throttle.read_bps_device", buf); |
64faf04c | 746 | |
64faf04c | 747 | sprintf(buf, "%u:%u %" PRIu64 "\n", major(dev), minor(dev), wbps); |
293d32df | 748 | (void) set_attribute_and_warn(u, "blkio", "blkio.throttle.write_bps_device", buf); |
64faf04c TH |
749 | } |
750 | ||
da4d897e | 751 | static bool cgroup_context_has_unified_memory_config(CGroupContext *c) { |
48422635 | 752 | return c->memory_min > 0 || c->memory_low > 0 || c->memory_high != CGROUP_LIMIT_MAX || c->memory_max != CGROUP_LIMIT_MAX || c->memory_swap_max != CGROUP_LIMIT_MAX; |
da4d897e TH |
753 | } |
754 | ||
f29ff115 | 755 | static void cgroup_apply_unified_memory_limit(Unit *u, const char *file, uint64_t v) { |
da4d897e | 756 | char buf[DECIMAL_STR_MAX(uint64_t) + 1] = "max"; |
da4d897e TH |
757 | |
758 | if (v != CGROUP_LIMIT_MAX) | |
759 | xsprintf(buf, "%" PRIu64 "\n", v); | |
760 | ||
293d32df | 761 | (void) set_attribute_and_warn(u, "memory", file, buf); |
da4d897e TH |
762 | } |
763 | ||
0f2d84d2 | 764 | static void cgroup_apply_firewall(Unit *u) { |
0f2d84d2 LP |
765 | assert(u); |
766 | ||
acf7f253 | 767 | /* Best-effort: let's apply IP firewalling and/or accounting if that's enabled */ |
906c06f6 | 768 | |
acf7f253 | 769 | if (bpf_firewall_compile(u) < 0) |
906c06f6 DM |
770 | return; |
771 | ||
772 | (void) bpf_firewall_install(u); | |
906c06f6 DM |
773 | } |
774 | ||
775 | static void cgroup_context_apply( | |
776 | Unit *u, | |
777 | CGroupMask apply_mask, | |
906c06f6 DM |
778 | ManagerState state) { |
779 | ||
f29ff115 TH |
780 | const char *path; |
781 | CGroupContext *c; | |
01efdf13 | 782 | bool is_root; |
4ad49000 LP |
783 | int r; |
784 | ||
f29ff115 TH |
785 | assert(u); |
786 | ||
906c06f6 | 787 | /* Nothing to do? Exit early! */ |
17f14955 | 788 | if (apply_mask == 0) |
4ad49000 | 789 | return; |
8e274523 | 790 | |
f3725e64 | 791 | /* Some cgroup attributes are not supported on the root cgroup, hence silently ignore */ |
611c4f8a | 792 | is_root = unit_has_host_root_cgroup(u); |
f3725e64 LP |
793 | |
794 | assert_se(c = unit_get_cgroup_context(u)); | |
795 | assert_se(path = u->cgroup_path); | |
796 | ||
797 | if (is_root) /* Make sure we don't try to display messages with an empty path. */ | |
6da13913 | 798 | path = "/"; |
01efdf13 | 799 | |
714e2e1d LP |
800 | /* We generally ignore errors caused by read-only mounted |
801 | * cgroup trees (assuming we are running in a container then), | |
802 | * and missing cgroups, i.e. EROFS and ENOENT. */ | |
803 | ||
906c06f6 DM |
804 | if ((apply_mask & CGROUP_MASK_CPU) && !is_root) { |
805 | bool has_weight, has_shares; | |
806 | ||
807 | has_weight = cgroup_context_has_cpu_weight(c); | |
808 | has_shares = cgroup_context_has_cpu_shares(c); | |
8e274523 | 809 | |
b4cccbc1 | 810 | if (cg_all_unified() > 0) { |
66ebf6c0 | 811 | uint64_t weight; |
b2f8b02e | 812 | |
66ebf6c0 TH |
813 | if (has_weight) |
814 | weight = cgroup_context_cpu_weight(c, state); | |
815 | else if (has_shares) { | |
816 | uint64_t shares = cgroup_context_cpu_shares(c, state); | |
b2f8b02e | 817 | |
66ebf6c0 TH |
818 | weight = cgroup_cpu_shares_to_weight(shares); |
819 | ||
8d33dca2 | 820 | log_cgroup_compat(u, "Applying [Startup]CPUShares %" PRIu64 " as [Startup]CPUWeight %" PRIu64 " on %s", |
66ebf6c0 TH |
821 | shares, weight, path); |
822 | } else | |
823 | weight = CGROUP_WEIGHT_DEFAULT; | |
824 | ||
825 | cgroup_apply_unified_cpu_config(u, weight, c->cpu_quota_per_sec_usec); | |
826 | } else { | |
827 | uint64_t shares; | |
828 | ||
7d862ab8 | 829 | if (has_weight) { |
66ebf6c0 TH |
830 | uint64_t weight = cgroup_context_cpu_weight(c, state); |
831 | ||
832 | shares = cgroup_cpu_weight_to_shares(weight); | |
833 | ||
8d33dca2 | 834 | log_cgroup_compat(u, "Applying [Startup]CPUWeight %" PRIu64 " as [Startup]CPUShares %" PRIu64 " on %s", |
66ebf6c0 | 835 | weight, shares, path); |
7d862ab8 TH |
836 | } else if (has_shares) |
837 | shares = cgroup_context_cpu_shares(c, state); | |
838 | else | |
66ebf6c0 TH |
839 | shares = CGROUP_CPU_SHARES_DEFAULT; |
840 | ||
841 | cgroup_apply_legacy_cpu_config(u, shares, c->cpu_quota_per_sec_usec); | |
842 | } | |
4ad49000 LP |
843 | } |
844 | ||
906c06f6 | 845 | if (apply_mask & CGROUP_MASK_IO) { |
538b4852 TH |
846 | bool has_io = cgroup_context_has_io_config(c); |
847 | bool has_blockio = cgroup_context_has_blockio_config(c); | |
13c31542 TH |
848 | |
849 | if (!is_root) { | |
64faf04c TH |
850 | char buf[8+DECIMAL_STR_MAX(uint64_t)+1]; |
851 | uint64_t weight; | |
13c31542 | 852 | |
538b4852 TH |
853 | if (has_io) |
854 | weight = cgroup_context_io_weight(c, state); | |
128fadc9 TH |
855 | else if (has_blockio) { |
856 | uint64_t blkio_weight = cgroup_context_blkio_weight(c, state); | |
857 | ||
858 | weight = cgroup_weight_blkio_to_io(blkio_weight); | |
859 | ||
860 | log_cgroup_compat(u, "Applying [Startup]BlockIOWeight %" PRIu64 " as [Startup]IOWeight %" PRIu64, | |
861 | blkio_weight, weight); | |
862 | } else | |
538b4852 | 863 | weight = CGROUP_WEIGHT_DEFAULT; |
13c31542 TH |
864 | |
865 | xsprintf(buf, "default %" PRIu64 "\n", weight); | |
293d32df | 866 | (void) set_attribute_and_warn(u, "io", "io.weight", buf); |
13c31542 | 867 | |
538b4852 TH |
868 | if (has_io) { |
869 | CGroupIODeviceWeight *w; | |
870 | ||
538b4852 | 871 | LIST_FOREACH(device_weights, w, c->io_device_weights) |
f29ff115 | 872 | cgroup_apply_io_device_weight(u, w->path, w->weight); |
538b4852 TH |
873 | } else if (has_blockio) { |
874 | CGroupBlockIODeviceWeight *w; | |
875 | ||
128fadc9 TH |
876 | LIST_FOREACH(device_weights, w, c->blockio_device_weights) { |
877 | weight = cgroup_weight_blkio_to_io(w->weight); | |
878 | ||
879 | log_cgroup_compat(u, "Applying BlockIODeviceWeight %" PRIu64 " as IODeviceWeight %" PRIu64 " for %s", | |
880 | w->weight, weight, w->path); | |
881 | ||
882 | cgroup_apply_io_device_weight(u, w->path, weight); | |
883 | } | |
538b4852 | 884 | } |
6ae4283c TH |
885 | |
886 | if (has_io) { | |
887 | CGroupIODeviceLatency *l; | |
888 | ||
889 | LIST_FOREACH(device_latencies, l, c->io_device_latencies) | |
890 | cgroup_apply_io_device_latency(u, l->path, l->target_usec); | |
891 | } | |
13c31542 TH |
892 | } |
893 | ||
538b4852 | 894 | if (has_io) { |
17ae2780 LP |
895 | CGroupIODeviceLimit *l; |
896 | ||
897 | LIST_FOREACH(device_limits, l, c->io_device_limits) | |
898 | cgroup_apply_io_device_limit(u, l->path, l->limits); | |
538b4852 | 899 | |
538b4852 | 900 | } else if (has_blockio) { |
17ae2780 | 901 | CGroupBlockIODeviceBandwidth *b; |
538b4852 | 902 | |
17ae2780 | 903 | LIST_FOREACH(device_bandwidths, b, c->blockio_device_bandwidths) { |
538b4852 TH |
904 | uint64_t limits[_CGROUP_IO_LIMIT_TYPE_MAX]; |
905 | CGroupIOLimitType type; | |
906 | ||
907 | for (type = 0; type < _CGROUP_IO_LIMIT_TYPE_MAX; type++) | |
908 | limits[type] = cgroup_io_limit_defaults[type]; | |
909 | ||
910 | limits[CGROUP_IO_RBPS_MAX] = b->rbps; | |
911 | limits[CGROUP_IO_WBPS_MAX] = b->wbps; | |
912 | ||
128fadc9 TH |
913 | log_cgroup_compat(u, "Applying BlockIO{Read|Write}Bandwidth %" PRIu64 " %" PRIu64 " as IO{Read|Write}BandwidthMax for %s", |
914 | b->rbps, b->wbps, b->path); | |
915 | ||
17ae2780 | 916 | cgroup_apply_io_device_limit(u, b->path, limits); |
538b4852 | 917 | } |
13c31542 TH |
918 | } |
919 | } | |
920 | ||
906c06f6 | 921 | if (apply_mask & CGROUP_MASK_BLKIO) { |
538b4852 TH |
922 | bool has_io = cgroup_context_has_io_config(c); |
923 | bool has_blockio = cgroup_context_has_blockio_config(c); | |
4ad49000 | 924 | |
01efdf13 | 925 | if (!is_root) { |
64faf04c TH |
926 | char buf[DECIMAL_STR_MAX(uint64_t)+1]; |
927 | uint64_t weight; | |
64faf04c | 928 | |
7d862ab8 | 929 | if (has_io) { |
128fadc9 TH |
930 | uint64_t io_weight = cgroup_context_io_weight(c, state); |
931 | ||
538b4852 | 932 | weight = cgroup_weight_io_to_blkio(cgroup_context_io_weight(c, state)); |
128fadc9 TH |
933 | |
934 | log_cgroup_compat(u, "Applying [Startup]IOWeight %" PRIu64 " as [Startup]BlockIOWeight %" PRIu64, | |
935 | io_weight, weight); | |
7d862ab8 TH |
936 | } else if (has_blockio) |
937 | weight = cgroup_context_blkio_weight(c, state); | |
938 | else | |
538b4852 | 939 | weight = CGROUP_BLKIO_WEIGHT_DEFAULT; |
64faf04c TH |
940 | |
941 | xsprintf(buf, "%" PRIu64 "\n", weight); | |
293d32df | 942 | (void) set_attribute_and_warn(u, "blkio", "blkio.weight", buf); |
4ad49000 | 943 | |
7d862ab8 | 944 | if (has_io) { |
538b4852 TH |
945 | CGroupIODeviceWeight *w; |
946 | ||
128fadc9 TH |
947 | LIST_FOREACH(device_weights, w, c->io_device_weights) { |
948 | weight = cgroup_weight_io_to_blkio(w->weight); | |
949 | ||
950 | log_cgroup_compat(u, "Applying IODeviceWeight %" PRIu64 " as BlockIODeviceWeight %" PRIu64 " for %s", | |
951 | w->weight, weight, w->path); | |
952 | ||
953 | cgroup_apply_blkio_device_weight(u, w->path, weight); | |
954 | } | |
7d862ab8 TH |
955 | } else if (has_blockio) { |
956 | CGroupBlockIODeviceWeight *w; | |
957 | ||
7d862ab8 TH |
958 | LIST_FOREACH(device_weights, w, c->blockio_device_weights) |
959 | cgroup_apply_blkio_device_weight(u, w->path, w->weight); | |
538b4852 | 960 | } |
4ad49000 LP |
961 | } |
962 | ||
7d862ab8 | 963 | if (has_io) { |
17ae2780 | 964 | CGroupIODeviceLimit *l; |
538b4852 | 965 | |
17ae2780 | 966 | LIST_FOREACH(device_limits, l, c->io_device_limits) { |
128fadc9 TH |
967 | log_cgroup_compat(u, "Applying IO{Read|Write}Bandwidth %" PRIu64 " %" PRIu64 " as BlockIO{Read|Write}BandwidthMax for %s", |
968 | l->limits[CGROUP_IO_RBPS_MAX], l->limits[CGROUP_IO_WBPS_MAX], l->path); | |
969 | ||
17ae2780 | 970 | cgroup_apply_blkio_device_limit(u, l->path, l->limits[CGROUP_IO_RBPS_MAX], l->limits[CGROUP_IO_WBPS_MAX]); |
538b4852 | 971 | } |
7d862ab8 | 972 | } else if (has_blockio) { |
17ae2780 | 973 | CGroupBlockIODeviceBandwidth *b; |
7d862ab8 | 974 | |
17ae2780 LP |
975 | LIST_FOREACH(device_bandwidths, b, c->blockio_device_bandwidths) |
976 | cgroup_apply_blkio_device_limit(u, b->path, b->rbps, b->wbps); | |
d686d8a9 | 977 | } |
8e274523 LP |
978 | } |
979 | ||
906c06f6 | 980 | if ((apply_mask & CGROUP_MASK_MEMORY) && !is_root) { |
b4cccbc1 LP |
981 | if (cg_all_unified() > 0) { |
982 | uint64_t max, swap_max = CGROUP_LIMIT_MAX; | |
efdb0237 | 983 | |
96e131ea | 984 | if (cgroup_context_has_unified_memory_config(c)) { |
da4d897e | 985 | max = c->memory_max; |
96e131ea WC |
986 | swap_max = c->memory_swap_max; |
987 | } else { | |
da4d897e | 988 | max = c->memory_limit; |
efdb0237 | 989 | |
128fadc9 | 990 | if (max != CGROUP_LIMIT_MAX) |
73fe5314 | 991 | log_cgroup_compat(u, "Applying MemoryLimit=%" PRIu64 " as MemoryMax=", max); |
128fadc9 TH |
992 | } |
993 | ||
48422635 | 994 | cgroup_apply_unified_memory_limit(u, "memory.min", c->memory_min); |
f29ff115 TH |
995 | cgroup_apply_unified_memory_limit(u, "memory.low", c->memory_low); |
996 | cgroup_apply_unified_memory_limit(u, "memory.high", c->memory_high); | |
997 | cgroup_apply_unified_memory_limit(u, "memory.max", max); | |
96e131ea | 998 | cgroup_apply_unified_memory_limit(u, "memory.swap.max", swap_max); |
efdb0237 | 999 | } else { |
da4d897e | 1000 | char buf[DECIMAL_STR_MAX(uint64_t) + 1]; |
7d862ab8 | 1001 | uint64_t val; |
da4d897e | 1002 | |
7d862ab8 | 1003 | if (cgroup_context_has_unified_memory_config(c)) { |
78a4ee59 | 1004 | val = c->memory_max; |
73fe5314 | 1005 | log_cgroup_compat(u, "Applying MemoryMax=%" PRIi64 " as MemoryLimit=", val); |
7d862ab8 TH |
1006 | } else |
1007 | val = c->memory_limit; | |
128fadc9 | 1008 | |
78a4ee59 DM |
1009 | if (val == CGROUP_LIMIT_MAX) |
1010 | strncpy(buf, "-1\n", sizeof(buf)); | |
1011 | else | |
1012 | xsprintf(buf, "%" PRIu64 "\n", val); | |
1013 | ||
293d32df | 1014 | (void) set_attribute_and_warn(u, "memory", "memory.limit_in_bytes", buf); |
da4d897e | 1015 | } |
4ad49000 | 1016 | } |
8e274523 | 1017 | |
084c7007 RG |
1018 | if ((apply_mask & (CGROUP_MASK_DEVICES | CGROUP_MASK_BPF_DEVICES)) && !is_root) { |
1019 | _cleanup_(bpf_program_unrefp) BPFProgram *prog = NULL; | |
4ad49000 | 1020 | CGroupDeviceAllow *a; |
8e274523 | 1021 | |
084c7007 RG |
1022 | if (cg_all_unified() > 0) { |
1023 | r = cgroup_init_device_bpf(&prog, c->device_policy, c->device_allow); | |
1024 | if (r < 0) | |
1025 | log_unit_warning_errno(u, r, "Failed to initialize device control bpf program: %m"); | |
1026 | } else { | |
8c838407 | 1027 | /* Changing the devices list of a populated cgroup might result in EINVAL, hence ignore EINVAL |
084c7007 | 1028 | * here. */ |
714e2e1d | 1029 | |
084c7007 RG |
1030 | if (c->device_allow || c->device_policy != CGROUP_AUTO) |
1031 | r = cg_set_attribute("devices", path, "devices.deny", "a"); | |
1032 | else | |
1033 | r = cg_set_attribute("devices", path, "devices.allow", "a"); | |
1034 | if (r < 0) | |
2c74e12b LP |
1035 | log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EINVAL, -EACCES, -EPERM) ? LOG_DEBUG : LOG_WARNING, r, |
1036 | "Failed to reset devices.allow/devices.deny: %m"); | |
084c7007 | 1037 | } |
fb385181 | 1038 | |
4ad49000 LP |
1039 | if (c->device_policy == CGROUP_CLOSED || |
1040 | (c->device_policy == CGROUP_AUTO && c->device_allow)) { | |
1041 | static const char auto_devices[] = | |
7d711efb LP |
1042 | "/dev/null\0" "rwm\0" |
1043 | "/dev/zero\0" "rwm\0" | |
1044 | "/dev/full\0" "rwm\0" | |
1045 | "/dev/random\0" "rwm\0" | |
1046 | "/dev/urandom\0" "rwm\0" | |
1047 | "/dev/tty\0" "rwm\0" | |
5a7f87a9 | 1048 | "/dev/ptmx\0" "rwm\0" |
0d9e7991 | 1049 | /* Allow /run/systemd/inaccessible/{chr,blk} devices for mapping InaccessiblePaths */ |
e7330dfe DP |
1050 | "-/run/systemd/inaccessible/chr\0" "rwm\0" |
1051 | "-/run/systemd/inaccessible/blk\0" "rwm\0"; | |
4ad49000 LP |
1052 | |
1053 | const char *x, *y; | |
1054 | ||
1055 | NULSTR_FOREACH_PAIR(x, y, auto_devices) | |
913c898c | 1056 | (void) whitelist_device(prog, path, x, y); |
7d711efb | 1057 | |
5a7f87a9 | 1058 | /* PTS (/dev/pts) devices may not be duplicated, but accessed */ |
913c898c | 1059 | (void) whitelist_major(prog, path, "pts", 'c', "rw"); |
4ad49000 LP |
1060 | } |
1061 | ||
1062 | LIST_FOREACH(device_allow, a, c->device_allow) { | |
fb4650aa | 1063 | char acc[4], *val; |
4ad49000 LP |
1064 | unsigned k = 0; |
1065 | ||
1066 | if (a->r) | |
1067 | acc[k++] = 'r'; | |
1068 | if (a->w) | |
1069 | acc[k++] = 'w'; | |
1070 | if (a->m) | |
1071 | acc[k++] = 'm'; | |
fb385181 | 1072 | |
4ad49000 LP |
1073 | if (k == 0) |
1074 | continue; | |
fb385181 | 1075 | |
4ad49000 | 1076 | acc[k++] = 0; |
90060676 | 1077 | |
27458ed6 | 1078 | if (path_startswith(a->path, "/dev/")) |
913c898c | 1079 | (void) whitelist_device(prog, path, a->path, acc); |
fb4650aa | 1080 | else if ((val = startswith(a->path, "block-"))) |
913c898c | 1081 | (void) whitelist_major(prog, path, val, 'b', acc); |
fb4650aa | 1082 | else if ((val = startswith(a->path, "char-"))) |
913c898c | 1083 | (void) whitelist_major(prog, path, val, 'c', acc); |
90060676 | 1084 | else |
f29ff115 | 1085 | log_unit_debug(u, "Ignoring device %s while writing cgroup attribute.", a->path); |
4ad49000 | 1086 | } |
084c7007 RG |
1087 | |
1088 | r = cgroup_apply_device_bpf(u, prog, c->device_policy, c->device_allow); | |
1089 | if (r < 0) { | |
1090 | static bool warned = false; | |
1091 | ||
1092 | log_full_errno(warned ? LOG_DEBUG : LOG_WARNING, r, | |
1093 | "Unit %s configures device ACL, but the local system doesn't seem to support the BPF-based device controller.\n" | |
1094 | "Proceeding WITHOUT applying ACL (all devices will be accessible)!\n" | |
1095 | "(This warning is only shown for the first loaded unit using device ACL.)", u->id); | |
1096 | ||
1097 | warned = true; | |
1098 | } | |
4ad49000 | 1099 | } |
03a7b521 | 1100 | |
00b5974f LP |
1101 | if (apply_mask & CGROUP_MASK_PIDS) { |
1102 | ||
1103 | if (is_root) { | |
1104 | /* So, the "pids" controller does not expose anything on the root cgroup, in order not to | |
1105 | * replicate knobs exposed elsewhere needlessly. We abstract this away here however, and when | |
1106 | * the knobs of the root cgroup are modified propagate this to the relevant sysctls. There's a | |
1107 | * non-obvious asymmetry however: unlike the cgroup properties we don't really want to take | |
1108 | * exclusive ownership of the sysctls, but we still want to honour things if the user sets | |
1109 | * limits. Hence we employ sort of a one-way strategy: when the user sets a bounded limit | |
1110 | * through us it counts. When the user afterwards unsets it again (i.e. sets it to unbounded) | |
1111 | * it also counts. But if the user never set a limit through us (i.e. we are the default of | |
1112 | * "unbounded") we leave things unmodified. For this we manage a global boolean that we turn on | |
1113 | * the first time we set a limit. Note that this boolean is flushed out on manager reload, | |
1114 | * which is desirable so that there's an offical way to release control of the sysctl from | |
1115 | * systemd: set the limit to unbounded and reload. */ | |
1116 | ||
1117 | if (c->tasks_max != CGROUP_LIMIT_MAX) { | |
1118 | u->manager->sysctl_pid_max_changed = true; | |
1119 | r = procfs_tasks_set_limit(c->tasks_max); | |
1120 | } else if (u->manager->sysctl_pid_max_changed) | |
1121 | r = procfs_tasks_set_limit(TASKS_MAX); | |
1122 | else | |
1123 | r = 0; | |
03a7b521 | 1124 | |
00b5974f | 1125 | if (r < 0) |
39b9fefb | 1126 | log_unit_full(u, LOG_LEVEL_CGROUP_WRITE(r), r, |
00b5974f | 1127 | "Failed to write to tasks limit sysctls: %m"); |
03a7b521 | 1128 | |
00b5974f LP |
1129 | } else { |
1130 | if (c->tasks_max != CGROUP_LIMIT_MAX) { | |
1131 | char buf[DECIMAL_STR_MAX(uint64_t) + 2]; | |
03a7b521 | 1132 | |
00b5974f | 1133 | sprintf(buf, "%" PRIu64 "\n", c->tasks_max); |
293d32df | 1134 | (void) set_attribute_and_warn(u, "pids", "pids.max", buf); |
00b5974f | 1135 | } else |
293d32df | 1136 | (void) set_attribute_and_warn(u, "pids", "pids.max", "max"); |
00b5974f | 1137 | } |
03a7b521 | 1138 | } |
906c06f6 | 1139 | |
17f14955 | 1140 | if (apply_mask & CGROUP_MASK_BPF_FIREWALL) |
0f2d84d2 | 1141 | cgroup_apply_firewall(u); |
fb385181 LP |
1142 | } |
1143 | ||
efdb0237 LP |
1144 | CGroupMask cgroup_context_get_mask(CGroupContext *c) { |
1145 | CGroupMask mask = 0; | |
8e274523 | 1146 | |
fae9bc29 | 1147 | /* Figure out which controllers we need, based on the cgroup context object */ |
8e274523 | 1148 | |
fae9bc29 | 1149 | if (c->cpu_accounting) |
f98c2585 | 1150 | mask |= get_cpu_accounting_mask(); |
fae9bc29 LP |
1151 | |
1152 | if (cgroup_context_has_cpu_weight(c) || | |
66ebf6c0 | 1153 | cgroup_context_has_cpu_shares(c) || |
3a43da28 | 1154 | c->cpu_quota_per_sec_usec != USEC_INFINITY) |
fae9bc29 | 1155 | mask |= CGROUP_MASK_CPU; |
ecedd90f | 1156 | |
538b4852 TH |
1157 | if (cgroup_context_has_io_config(c) || cgroup_context_has_blockio_config(c)) |
1158 | mask |= CGROUP_MASK_IO | CGROUP_MASK_BLKIO; | |
ecedd90f | 1159 | |
4ad49000 | 1160 | if (c->memory_accounting || |
da4d897e TH |
1161 | c->memory_limit != CGROUP_LIMIT_MAX || |
1162 | cgroup_context_has_unified_memory_config(c)) | |
efdb0237 | 1163 | mask |= CGROUP_MASK_MEMORY; |
8e274523 | 1164 | |
a931ad47 LP |
1165 | if (c->device_allow || |
1166 | c->device_policy != CGROUP_AUTO) | |
084c7007 | 1167 | mask |= CGROUP_MASK_DEVICES | CGROUP_MASK_BPF_DEVICES; |
4ad49000 | 1168 | |
03a7b521 | 1169 | if (c->tasks_accounting || |
8793fa25 | 1170 | c->tasks_max != CGROUP_LIMIT_MAX) |
03a7b521 LP |
1171 | mask |= CGROUP_MASK_PIDS; |
1172 | ||
fae9bc29 | 1173 | return CGROUP_MASK_EXTEND_JOINED(mask); |
8e274523 LP |
1174 | } |
1175 | ||
17f14955 RG |
1176 | CGroupMask unit_get_bpf_mask(Unit *u) { |
1177 | CGroupMask mask = 0; | |
1178 | ||
fae9bc29 LP |
1179 | /* Figure out which controllers we need, based on the cgroup context, possibly taking into account children |
1180 | * too. */ | |
1181 | ||
17f14955 RG |
1182 | if (unit_get_needs_bpf_firewall(u)) |
1183 | mask |= CGROUP_MASK_BPF_FIREWALL; | |
1184 | ||
1185 | return mask; | |
1186 | } | |
1187 | ||
efdb0237 | 1188 | CGroupMask unit_get_own_mask(Unit *u) { |
4ad49000 | 1189 | CGroupContext *c; |
8e274523 | 1190 | |
efdb0237 LP |
1191 | /* Returns the mask of controllers the unit needs for itself */ |
1192 | ||
4ad49000 LP |
1193 | c = unit_get_cgroup_context(u); |
1194 | if (!c) | |
1195 | return 0; | |
8e274523 | 1196 | |
17f14955 | 1197 | return cgroup_context_get_mask(c) | unit_get_bpf_mask(u) | unit_get_delegate_mask(u); |
02638280 LP |
1198 | } |
1199 | ||
1200 | CGroupMask unit_get_delegate_mask(Unit *u) { | |
1201 | CGroupContext *c; | |
1202 | ||
1203 | /* If delegation is turned on, then turn on selected controllers, unless we are on the legacy hierarchy and the | |
1204 | * process we fork into is known to drop privileges, and hence shouldn't get access to the controllers. | |
19af675e | 1205 | * |
02638280 | 1206 | * Note that on the unified hierarchy it is safe to delegate controllers to unprivileged services. */ |
a931ad47 | 1207 | |
1d9cc876 | 1208 | if (!unit_cgroup_delegate(u)) |
02638280 LP |
1209 | return 0; |
1210 | ||
1211 | if (cg_all_unified() <= 0) { | |
a931ad47 LP |
1212 | ExecContext *e; |
1213 | ||
1214 | e = unit_get_exec_context(u); | |
02638280 LP |
1215 | if (e && !exec_context_maintains_privileges(e)) |
1216 | return 0; | |
a931ad47 LP |
1217 | } |
1218 | ||
1d9cc876 | 1219 | assert_se(c = unit_get_cgroup_context(u)); |
fae9bc29 | 1220 | return CGROUP_MASK_EXTEND_JOINED(c->delegate_controllers); |
8e274523 LP |
1221 | } |
1222 | ||
efdb0237 | 1223 | CGroupMask unit_get_members_mask(Unit *u) { |
4ad49000 | 1224 | assert(u); |
bc432dc7 | 1225 | |
02638280 | 1226 | /* Returns the mask of controllers all of the unit's children require, merged */ |
efdb0237 | 1227 | |
bc432dc7 LP |
1228 | if (u->cgroup_members_mask_valid) |
1229 | return u->cgroup_members_mask; | |
1230 | ||
64e844e5 | 1231 | u->cgroup_members_mask = 0; |
bc432dc7 LP |
1232 | |
1233 | if (u->type == UNIT_SLICE) { | |
eef85c4a | 1234 | void *v; |
bc432dc7 LP |
1235 | Unit *member; |
1236 | Iterator i; | |
1237 | ||
eef85c4a | 1238 | HASHMAP_FOREACH_KEY(v, member, u->dependencies[UNIT_BEFORE], i) { |
bc432dc7 LP |
1239 | |
1240 | if (member == u) | |
1241 | continue; | |
1242 | ||
d4fdc205 | 1243 | if (UNIT_DEREF(member->slice) != u) |
bc432dc7 LP |
1244 | continue; |
1245 | ||
31604970 | 1246 | u->cgroup_members_mask |= unit_get_subtree_mask(member); /* note that this calls ourselves again, for the children */ |
bc432dc7 LP |
1247 | } |
1248 | } | |
1249 | ||
1250 | u->cgroup_members_mask_valid = true; | |
6414b7c9 | 1251 | return u->cgroup_members_mask; |
246aa6dd LP |
1252 | } |
1253 | ||
efdb0237 | 1254 | CGroupMask unit_get_siblings_mask(Unit *u) { |
4ad49000 | 1255 | assert(u); |
246aa6dd | 1256 | |
efdb0237 LP |
1257 | /* Returns the mask of controllers all of the unit's siblings |
1258 | * require, i.e. the members mask of the unit's parent slice | |
1259 | * if there is one. */ | |
1260 | ||
bc432dc7 | 1261 | if (UNIT_ISSET(u->slice)) |
637f421e | 1262 | return unit_get_members_mask(UNIT_DEREF(u->slice)); |
4ad49000 | 1263 | |
64e844e5 | 1264 | return unit_get_subtree_mask(u); /* we are the top-level slice */ |
246aa6dd LP |
1265 | } |
1266 | ||
efdb0237 LP |
1267 | CGroupMask unit_get_subtree_mask(Unit *u) { |
1268 | ||
1269 | /* Returns the mask of this subtree, meaning of the group | |
1270 | * itself and its children. */ | |
1271 | ||
1272 | return unit_get_own_mask(u) | unit_get_members_mask(u); | |
1273 | } | |
1274 | ||
1275 | CGroupMask unit_get_target_mask(Unit *u) { | |
1276 | CGroupMask mask; | |
1277 | ||
1278 | /* This returns the cgroup mask of all controllers to enable | |
1279 | * for a specific cgroup, i.e. everything it needs itself, | |
1280 | * plus all that its children need, plus all that its siblings | |
1281 | * need. This is primarily useful on the legacy cgroup | |
1282 | * hierarchy, where we need to duplicate each cgroup in each | |
1283 | * hierarchy that shall be enabled for it. */ | |
6414b7c9 | 1284 | |
efdb0237 LP |
1285 | mask = unit_get_own_mask(u) | unit_get_members_mask(u) | unit_get_siblings_mask(u); |
1286 | mask &= u->manager->cgroup_supported; | |
1287 | ||
1288 | return mask; | |
1289 | } | |
1290 | ||
1291 | CGroupMask unit_get_enable_mask(Unit *u) { | |
1292 | CGroupMask mask; | |
1293 | ||
1294 | /* This returns the cgroup mask of all controllers to enable | |
1295 | * for the children of a specific cgroup. This is primarily | |
1296 | * useful for the unified cgroup hierarchy, where each cgroup | |
1297 | * controls which controllers are enabled for its children. */ | |
1298 | ||
1299 | mask = unit_get_members_mask(u); | |
6414b7c9 DS |
1300 | mask &= u->manager->cgroup_supported; |
1301 | ||
1302 | return mask; | |
1303 | } | |
1304 | ||
17f14955 | 1305 | bool unit_get_needs_bpf_firewall(Unit *u) { |
906c06f6 DM |
1306 | CGroupContext *c; |
1307 | Unit *p; | |
1308 | assert(u); | |
1309 | ||
906c06f6 DM |
1310 | c = unit_get_cgroup_context(u); |
1311 | if (!c) | |
1312 | return false; | |
1313 | ||
1314 | if (c->ip_accounting || | |
1315 | c->ip_address_allow || | |
1316 | c->ip_address_deny) | |
1317 | return true; | |
1318 | ||
1319 | /* If any parent slice has an IP access list defined, it applies too */ | |
1320 | for (p = UNIT_DEREF(u->slice); p; p = UNIT_DEREF(p->slice)) { | |
1321 | c = unit_get_cgroup_context(p); | |
1322 | if (!c) | |
1323 | return false; | |
1324 | ||
1325 | if (c->ip_address_allow || | |
1326 | c->ip_address_deny) | |
1327 | return true; | |
1328 | } | |
1329 | ||
1330 | return false; | |
1331 | } | |
1332 | ||
6414b7c9 DS |
1333 | /* Recurse from a unit up through its containing slices, propagating |
1334 | * mask bits upward. A unit is also member of itself. */ | |
bc432dc7 | 1335 | void unit_update_cgroup_members_masks(Unit *u) { |
efdb0237 | 1336 | CGroupMask m; |
bc432dc7 LP |
1337 | bool more; |
1338 | ||
1339 | assert(u); | |
1340 | ||
1341 | /* Calculate subtree mask */ | |
efdb0237 | 1342 | m = unit_get_subtree_mask(u); |
bc432dc7 LP |
1343 | |
1344 | /* See if anything changed from the previous invocation. If | |
1345 | * not, we're done. */ | |
1346 | if (u->cgroup_subtree_mask_valid && m == u->cgroup_subtree_mask) | |
1347 | return; | |
1348 | ||
1349 | more = | |
1350 | u->cgroup_subtree_mask_valid && | |
1351 | ((m & ~u->cgroup_subtree_mask) != 0) && | |
1352 | ((~m & u->cgroup_subtree_mask) == 0); | |
1353 | ||
1354 | u->cgroup_subtree_mask = m; | |
1355 | u->cgroup_subtree_mask_valid = true; | |
1356 | ||
6414b7c9 DS |
1357 | if (UNIT_ISSET(u->slice)) { |
1358 | Unit *s = UNIT_DEREF(u->slice); | |
bc432dc7 LP |
1359 | |
1360 | if (more) | |
1361 | /* There's more set now than before. We | |
1362 | * propagate the new mask to the parent's mask | |
1363 | * (not caring if it actually was valid or | |
1364 | * not). */ | |
1365 | ||
1366 | s->cgroup_members_mask |= m; | |
1367 | ||
1368 | else | |
1369 | /* There's less set now than before (or we | |
1370 | * don't know), we need to recalculate | |
1371 | * everything, so let's invalidate the | |
1372 | * parent's members mask */ | |
1373 | ||
1374 | s->cgroup_members_mask_valid = false; | |
1375 | ||
1376 | /* And now make sure that this change also hits our | |
1377 | * grandparents */ | |
1378 | unit_update_cgroup_members_masks(s); | |
6414b7c9 DS |
1379 | } |
1380 | } | |
1381 | ||
6592b975 | 1382 | const char *unit_get_realized_cgroup_path(Unit *u, CGroupMask mask) { |
03b90d4b | 1383 | |
6592b975 | 1384 | /* Returns the realized cgroup path of the specified unit where all specified controllers are available. */ |
03b90d4b LP |
1385 | |
1386 | while (u) { | |
6592b975 | 1387 | |
03b90d4b LP |
1388 | if (u->cgroup_path && |
1389 | u->cgroup_realized && | |
d94a24ca | 1390 | FLAGS_SET(u->cgroup_realized_mask, mask)) |
03b90d4b LP |
1391 | return u->cgroup_path; |
1392 | ||
1393 | u = UNIT_DEREF(u->slice); | |
1394 | } | |
1395 | ||
1396 | return NULL; | |
1397 | } | |
1398 | ||
6592b975 LP |
1399 | static const char *migrate_callback(CGroupMask mask, void *userdata) { |
1400 | return unit_get_realized_cgroup_path(userdata, mask); | |
1401 | } | |
1402 | ||
efdb0237 LP |
1403 | char *unit_default_cgroup_path(Unit *u) { |
1404 | _cleanup_free_ char *escaped = NULL, *slice = NULL; | |
1405 | int r; | |
1406 | ||
1407 | assert(u); | |
1408 | ||
1409 | if (unit_has_name(u, SPECIAL_ROOT_SLICE)) | |
1410 | return strdup(u->manager->cgroup_root); | |
1411 | ||
1412 | if (UNIT_ISSET(u->slice) && !unit_has_name(UNIT_DEREF(u->slice), SPECIAL_ROOT_SLICE)) { | |
1413 | r = cg_slice_to_path(UNIT_DEREF(u->slice)->id, &slice); | |
1414 | if (r < 0) | |
1415 | return NULL; | |
1416 | } | |
1417 | ||
1418 | escaped = cg_escape(u->id); | |
1419 | if (!escaped) | |
1420 | return NULL; | |
1421 | ||
1422 | if (slice) | |
605405c6 ZJS |
1423 | return strjoin(u->manager->cgroup_root, "/", slice, "/", |
1424 | escaped); | |
efdb0237 | 1425 | else |
605405c6 | 1426 | return strjoin(u->manager->cgroup_root, "/", escaped); |
efdb0237 LP |
1427 | } |
1428 | ||
1429 | int unit_set_cgroup_path(Unit *u, const char *path) { | |
1430 | _cleanup_free_ char *p = NULL; | |
1431 | int r; | |
1432 | ||
1433 | assert(u); | |
1434 | ||
1435 | if (path) { | |
1436 | p = strdup(path); | |
1437 | if (!p) | |
1438 | return -ENOMEM; | |
1439 | } else | |
1440 | p = NULL; | |
1441 | ||
1442 | if (streq_ptr(u->cgroup_path, p)) | |
1443 | return 0; | |
1444 | ||
1445 | if (p) { | |
1446 | r = hashmap_put(u->manager->cgroup_unit, p, u); | |
1447 | if (r < 0) | |
1448 | return r; | |
1449 | } | |
1450 | ||
1451 | unit_release_cgroup(u); | |
1452 | ||
ae2a15bc | 1453 | u->cgroup_path = TAKE_PTR(p); |
efdb0237 LP |
1454 | |
1455 | return 1; | |
1456 | } | |
1457 | ||
1458 | int unit_watch_cgroup(Unit *u) { | |
ab2c3861 | 1459 | _cleanup_free_ char *events = NULL; |
efdb0237 LP |
1460 | int r; |
1461 | ||
1462 | assert(u); | |
1463 | ||
1464 | if (!u->cgroup_path) | |
1465 | return 0; | |
1466 | ||
1467 | if (u->cgroup_inotify_wd >= 0) | |
1468 | return 0; | |
1469 | ||
1470 | /* Only applies to the unified hierarchy */ | |
c22800e4 | 1471 | r = cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER); |
b4cccbc1 LP |
1472 | if (r < 0) |
1473 | return log_error_errno(r, "Failed to determine whether the name=systemd hierarchy is unified: %m"); | |
1474 | if (r == 0) | |
efdb0237 LP |
1475 | return 0; |
1476 | ||
1477 | /* Don't watch the root slice, it's pointless. */ | |
1478 | if (unit_has_name(u, SPECIAL_ROOT_SLICE)) | |
1479 | return 0; | |
1480 | ||
1481 | r = hashmap_ensure_allocated(&u->manager->cgroup_inotify_wd_unit, &trivial_hash_ops); | |
1482 | if (r < 0) | |
1483 | return log_oom(); | |
1484 | ||
ab2c3861 | 1485 | r = cg_get_path(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path, "cgroup.events", &events); |
efdb0237 LP |
1486 | if (r < 0) |
1487 | return log_oom(); | |
1488 | ||
ab2c3861 | 1489 | u->cgroup_inotify_wd = inotify_add_watch(u->manager->cgroup_inotify_fd, events, IN_MODIFY); |
efdb0237 LP |
1490 | if (u->cgroup_inotify_wd < 0) { |
1491 | ||
1492 | if (errno == ENOENT) /* If the directory is already | |
1493 | * gone we don't need to track | |
1494 | * it, so this is not an error */ | |
1495 | return 0; | |
1496 | ||
1497 | return log_unit_error_errno(u, errno, "Failed to add inotify watch descriptor for control group %s: %m", u->cgroup_path); | |
1498 | } | |
1499 | ||
1500 | r = hashmap_put(u->manager->cgroup_inotify_wd_unit, INT_TO_PTR(u->cgroup_inotify_wd), u); | |
1501 | if (r < 0) | |
1502 | return log_unit_error_errno(u, r, "Failed to add inotify watch descriptor to hash map: %m"); | |
1503 | ||
1504 | return 0; | |
1505 | } | |
1506 | ||
a4634b21 LP |
1507 | int unit_pick_cgroup_path(Unit *u) { |
1508 | _cleanup_free_ char *path = NULL; | |
1509 | int r; | |
1510 | ||
1511 | assert(u); | |
1512 | ||
1513 | if (u->cgroup_path) | |
1514 | return 0; | |
1515 | ||
1516 | if (!UNIT_HAS_CGROUP_CONTEXT(u)) | |
1517 | return -EINVAL; | |
1518 | ||
1519 | path = unit_default_cgroup_path(u); | |
1520 | if (!path) | |
1521 | return log_oom(); | |
1522 | ||
1523 | r = unit_set_cgroup_path(u, path); | |
1524 | if (r == -EEXIST) | |
1525 | return log_unit_error_errno(u, r, "Control group %s exists already.", path); | |
1526 | if (r < 0) | |
1527 | return log_unit_error_errno(u, r, "Failed to set unit's control group path to %s: %m", path); | |
1528 | ||
1529 | return 0; | |
1530 | } | |
1531 | ||
efdb0237 LP |
1532 | static int unit_create_cgroup( |
1533 | Unit *u, | |
1534 | CGroupMask target_mask, | |
17f14955 | 1535 | CGroupMask enable_mask) { |
efdb0237 | 1536 | |
0cd385d3 | 1537 | CGroupContext *c; |
bc432dc7 | 1538 | int r; |
65be7e06 | 1539 | bool created; |
64747e2d | 1540 | |
4ad49000 | 1541 | assert(u); |
64747e2d | 1542 | |
0cd385d3 LP |
1543 | c = unit_get_cgroup_context(u); |
1544 | if (!c) | |
1545 | return 0; | |
1546 | ||
a4634b21 LP |
1547 | /* Figure out our cgroup path */ |
1548 | r = unit_pick_cgroup_path(u); | |
1549 | if (r < 0) | |
1550 | return r; | |
b58b8e11 | 1551 | |
03b90d4b | 1552 | /* First, create our own group */ |
efdb0237 | 1553 | r = cg_create_everywhere(u->manager->cgroup_supported, target_mask, u->cgroup_path); |
23bbb0de | 1554 | if (r < 0) |
efdb0237 | 1555 | return log_unit_error_errno(u, r, "Failed to create cgroup %s: %m", u->cgroup_path); |
490c5a37 | 1556 | created = r; |
efdb0237 LP |
1557 | |
1558 | /* Start watching it */ | |
1559 | (void) unit_watch_cgroup(u); | |
1560 | ||
65be7e06 ZJS |
1561 | /* Preserve enabled controllers in delegated units, adjust others. */ |
1562 | if (created || !unit_cgroup_delegate(u)) { | |
1563 | ||
1564 | /* Enable all controllers we need */ | |
1565 | r = cg_enable_everywhere(u->manager->cgroup_supported, enable_mask, u->cgroup_path); | |
1566 | if (r < 0) | |
1567 | log_unit_warning_errno(u, r, "Failed to enable controllers on cgroup %s, ignoring: %m", | |
1568 | u->cgroup_path); | |
1569 | } | |
03b90d4b LP |
1570 | |
1571 | /* Keep track that this is now realized */ | |
4ad49000 | 1572 | u->cgroup_realized = true; |
efdb0237 | 1573 | u->cgroup_realized_mask = target_mask; |
ccf78df1 | 1574 | u->cgroup_enabled_mask = enable_mask; |
4ad49000 | 1575 | |
1d9cc876 | 1576 | if (u->type != UNIT_SLICE && !unit_cgroup_delegate(u)) { |
0cd385d3 LP |
1577 | |
1578 | /* Then, possibly move things over, but not if | |
1579 | * subgroups may contain processes, which is the case | |
1580 | * for slice and delegation units. */ | |
1581 | r = cg_migrate_everywhere(u->manager->cgroup_supported, u->cgroup_path, u->cgroup_path, migrate_callback, u); | |
1582 | if (r < 0) | |
efdb0237 | 1583 | log_unit_warning_errno(u, r, "Failed to migrate cgroup from to %s, ignoring: %m", u->cgroup_path); |
0cd385d3 | 1584 | } |
03b90d4b | 1585 | |
64747e2d LP |
1586 | return 0; |
1587 | } | |
1588 | ||
6592b975 LP |
1589 | static int unit_attach_pid_to_cgroup_via_bus(Unit *u, pid_t pid, const char *suffix_path) { |
1590 | _cleanup_(sd_bus_error_free) sd_bus_error error = SD_BUS_ERROR_NULL; | |
1591 | char *pp; | |
7b3fd631 | 1592 | int r; |
6592b975 | 1593 | |
7b3fd631 LP |
1594 | assert(u); |
1595 | ||
6592b975 LP |
1596 | if (MANAGER_IS_SYSTEM(u->manager)) |
1597 | return -EINVAL; | |
1598 | ||
1599 | if (!u->manager->system_bus) | |
1600 | return -EIO; | |
1601 | ||
1602 | if (!u->cgroup_path) | |
1603 | return -EINVAL; | |
1604 | ||
1605 | /* Determine this unit's cgroup path relative to our cgroup root */ | |
1606 | pp = path_startswith(u->cgroup_path, u->manager->cgroup_root); | |
1607 | if (!pp) | |
1608 | return -EINVAL; | |
1609 | ||
1610 | pp = strjoina("/", pp, suffix_path); | |
858d36c1 | 1611 | path_simplify(pp, false); |
6592b975 LP |
1612 | |
1613 | r = sd_bus_call_method(u->manager->system_bus, | |
1614 | "org.freedesktop.systemd1", | |
1615 | "/org/freedesktop/systemd1", | |
1616 | "org.freedesktop.systemd1.Manager", | |
1617 | "AttachProcessesToUnit", | |
1618 | &error, NULL, | |
1619 | "ssau", | |
1620 | NULL /* empty unit name means client's unit, i.e. us */, pp, 1, (uint32_t) pid); | |
7b3fd631 | 1621 | if (r < 0) |
6592b975 LP |
1622 | return log_unit_debug_errno(u, r, "Failed to attach unit process " PID_FMT " via the bus: %s", pid, bus_error_message(&error, r)); |
1623 | ||
1624 | return 0; | |
1625 | } | |
1626 | ||
1627 | int unit_attach_pids_to_cgroup(Unit *u, Set *pids, const char *suffix_path) { | |
1628 | CGroupMask delegated_mask; | |
1629 | const char *p; | |
1630 | Iterator i; | |
1631 | void *pidp; | |
1632 | int r, q; | |
1633 | ||
1634 | assert(u); | |
1635 | ||
1636 | if (!UNIT_HAS_CGROUP_CONTEXT(u)) | |
1637 | return -EINVAL; | |
1638 | ||
1639 | if (set_isempty(pids)) | |
1640 | return 0; | |
7b3fd631 | 1641 | |
6592b975 | 1642 | r = unit_realize_cgroup(u); |
7b3fd631 LP |
1643 | if (r < 0) |
1644 | return r; | |
1645 | ||
6592b975 LP |
1646 | if (isempty(suffix_path)) |
1647 | p = u->cgroup_path; | |
1648 | else | |
1649 | p = strjoina(u->cgroup_path, "/", suffix_path); | |
1650 | ||
1651 | delegated_mask = unit_get_delegate_mask(u); | |
1652 | ||
1653 | r = 0; | |
1654 | SET_FOREACH(pidp, pids, i) { | |
1655 | pid_t pid = PTR_TO_PID(pidp); | |
1656 | CGroupController c; | |
1657 | ||
1658 | /* First, attach the PID to the main cgroup hierarchy */ | |
1659 | q = cg_attach(SYSTEMD_CGROUP_CONTROLLER, p, pid); | |
1660 | if (q < 0) { | |
1661 | log_unit_debug_errno(u, q, "Couldn't move process " PID_FMT " to requested cgroup '%s': %m", pid, p); | |
1662 | ||
1663 | if (MANAGER_IS_USER(u->manager) && IN_SET(q, -EPERM, -EACCES)) { | |
1664 | int z; | |
1665 | ||
1666 | /* If we are in a user instance, and we can't move the process ourselves due to | |
1667 | * permission problems, let's ask the system instance about it instead. Since it's more | |
1668 | * privileged it might be able to move the process across the leaves of a subtree who's | |
1669 | * top node is not owned by us. */ | |
1670 | ||
1671 | z = unit_attach_pid_to_cgroup_via_bus(u, pid, suffix_path); | |
1672 | if (z < 0) | |
1673 | log_unit_debug_errno(u, z, "Couldn't move process " PID_FMT " to requested cgroup '%s' via the system bus either: %m", pid, p); | |
1674 | else | |
1675 | continue; /* When the bus thing worked via the bus we are fully done for this PID. */ | |
1676 | } | |
1677 | ||
1678 | if (r >= 0) | |
1679 | r = q; /* Remember first error */ | |
1680 | ||
1681 | continue; | |
1682 | } | |
1683 | ||
1684 | q = cg_all_unified(); | |
1685 | if (q < 0) | |
1686 | return q; | |
1687 | if (q > 0) | |
1688 | continue; | |
1689 | ||
1690 | /* In the legacy hierarchy, attach the process to the request cgroup if possible, and if not to the | |
1691 | * innermost realized one */ | |
1692 | ||
1693 | for (c = 0; c < _CGROUP_CONTROLLER_MAX; c++) { | |
1694 | CGroupMask bit = CGROUP_CONTROLLER_TO_MASK(c); | |
1695 | const char *realized; | |
1696 | ||
1697 | if (!(u->manager->cgroup_supported & bit)) | |
1698 | continue; | |
1699 | ||
1700 | /* If this controller is delegated and realized, honour the caller's request for the cgroup suffix. */ | |
1701 | if (delegated_mask & u->cgroup_realized_mask & bit) { | |
1702 | q = cg_attach(cgroup_controller_to_string(c), p, pid); | |
1703 | if (q >= 0) | |
1704 | continue; /* Success! */ | |
1705 | ||
1706 | 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", | |
1707 | pid, p, cgroup_controller_to_string(c)); | |
1708 | } | |
1709 | ||
1710 | /* So this controller is either not delegate or realized, or something else weird happened. In | |
1711 | * that case let's attach the PID at least to the closest cgroup up the tree that is | |
1712 | * realized. */ | |
1713 | realized = unit_get_realized_cgroup_path(u, bit); | |
1714 | if (!realized) | |
1715 | continue; /* Not even realized in the root slice? Then let's not bother */ | |
1716 | ||
1717 | q = cg_attach(cgroup_controller_to_string(c), realized, pid); | |
1718 | if (q < 0) | |
1719 | log_unit_debug_errno(u, q, "Failed to attach PID " PID_FMT " to realized cgroup %s in controller %s, ignoring: %m", | |
1720 | pid, realized, cgroup_controller_to_string(c)); | |
1721 | } | |
1722 | } | |
1723 | ||
1724 | return r; | |
7b3fd631 LP |
1725 | } |
1726 | ||
4b58153d LP |
1727 | static void cgroup_xattr_apply(Unit *u) { |
1728 | char ids[SD_ID128_STRING_MAX]; | |
1729 | int r; | |
1730 | ||
1731 | assert(u); | |
1732 | ||
1733 | if (!MANAGER_IS_SYSTEM(u->manager)) | |
1734 | return; | |
1735 | ||
1736 | if (sd_id128_is_null(u->invocation_id)) | |
1737 | return; | |
1738 | ||
1739 | r = cg_set_xattr(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path, | |
1740 | "trusted.invocation_id", | |
1741 | sd_id128_to_string(u->invocation_id, ids), 32, | |
1742 | 0); | |
1743 | if (r < 0) | |
0fb84499 | 1744 | log_unit_debug_errno(u, r, "Failed to set invocation ID on control group %s, ignoring: %m", u->cgroup_path); |
4b58153d LP |
1745 | } |
1746 | ||
906c06f6 DM |
1747 | static bool unit_has_mask_realized( |
1748 | Unit *u, | |
1749 | CGroupMask target_mask, | |
17f14955 | 1750 | CGroupMask enable_mask) { |
906c06f6 | 1751 | |
bc432dc7 LP |
1752 | assert(u); |
1753 | ||
906c06f6 DM |
1754 | return u->cgroup_realized && |
1755 | u->cgroup_realized_mask == target_mask && | |
1756 | u->cgroup_enabled_mask == enable_mask && | |
17f14955 | 1757 | u->cgroup_invalidated_mask == 0; |
6414b7c9 DS |
1758 | } |
1759 | ||
2aa57a65 LP |
1760 | static void unit_add_to_cgroup_realize_queue(Unit *u) { |
1761 | assert(u); | |
1762 | ||
1763 | if (u->in_cgroup_realize_queue) | |
1764 | return; | |
1765 | ||
1766 | LIST_PREPEND(cgroup_realize_queue, u->manager->cgroup_realize_queue, u); | |
1767 | u->in_cgroup_realize_queue = true; | |
1768 | } | |
1769 | ||
1770 | static void unit_remove_from_cgroup_realize_queue(Unit *u) { | |
1771 | assert(u); | |
1772 | ||
1773 | if (!u->in_cgroup_realize_queue) | |
1774 | return; | |
1775 | ||
1776 | LIST_REMOVE(cgroup_realize_queue, u->manager->cgroup_realize_queue, u); | |
1777 | u->in_cgroup_realize_queue = false; | |
1778 | } | |
1779 | ||
6414b7c9 DS |
1780 | /* Check if necessary controllers and attributes for a unit are in place. |
1781 | * | |
1782 | * If so, do nothing. | |
1783 | * If not, create paths, move processes over, and set attributes. | |
1784 | * | |
1785 | * Returns 0 on success and < 0 on failure. */ | |
db785129 | 1786 | static int unit_realize_cgroup_now(Unit *u, ManagerState state) { |
efdb0237 | 1787 | CGroupMask target_mask, enable_mask; |
6414b7c9 | 1788 | int r; |
64747e2d | 1789 | |
4ad49000 | 1790 | assert(u); |
64747e2d | 1791 | |
2aa57a65 | 1792 | unit_remove_from_cgroup_realize_queue(u); |
64747e2d | 1793 | |
efdb0237 | 1794 | target_mask = unit_get_target_mask(u); |
ccf78df1 TH |
1795 | enable_mask = unit_get_enable_mask(u); |
1796 | ||
17f14955 | 1797 | if (unit_has_mask_realized(u, target_mask, enable_mask)) |
0a1eb06d | 1798 | return 0; |
64747e2d | 1799 | |
4ad49000 | 1800 | /* First, realize parents */ |
6414b7c9 | 1801 | if (UNIT_ISSET(u->slice)) { |
db785129 | 1802 | r = unit_realize_cgroup_now(UNIT_DEREF(u->slice), state); |
6414b7c9 DS |
1803 | if (r < 0) |
1804 | return r; | |
1805 | } | |
4ad49000 LP |
1806 | |
1807 | /* And then do the real work */ | |
17f14955 | 1808 | r = unit_create_cgroup(u, target_mask, enable_mask); |
6414b7c9 DS |
1809 | if (r < 0) |
1810 | return r; | |
1811 | ||
1812 | /* Finally, apply the necessary attributes. */ | |
17f14955 | 1813 | cgroup_context_apply(u, target_mask, state); |
4b58153d | 1814 | cgroup_xattr_apply(u); |
6414b7c9 | 1815 | |
c2baf11c LP |
1816 | /* Now, reset the invalidation mask */ |
1817 | u->cgroup_invalidated_mask = 0; | |
6414b7c9 | 1818 | return 0; |
64747e2d LP |
1819 | } |
1820 | ||
91a6073e | 1821 | unsigned manager_dispatch_cgroup_realize_queue(Manager *m) { |
db785129 | 1822 | ManagerState state; |
4ad49000 | 1823 | unsigned n = 0; |
db785129 | 1824 | Unit *i; |
6414b7c9 | 1825 | int r; |
ecedd90f | 1826 | |
91a6073e LP |
1827 | assert(m); |
1828 | ||
db785129 LP |
1829 | state = manager_state(m); |
1830 | ||
91a6073e LP |
1831 | while ((i = m->cgroup_realize_queue)) { |
1832 | assert(i->in_cgroup_realize_queue); | |
ecedd90f | 1833 | |
2aa57a65 LP |
1834 | if (UNIT_IS_INACTIVE_OR_FAILED(unit_active_state(i))) { |
1835 | /* Maybe things changed, and the unit is not actually active anymore? */ | |
1836 | unit_remove_from_cgroup_realize_queue(i); | |
1837 | continue; | |
1838 | } | |
1839 | ||
db785129 | 1840 | r = unit_realize_cgroup_now(i, state); |
6414b7c9 | 1841 | if (r < 0) |
efdb0237 | 1842 | log_warning_errno(r, "Failed to realize cgroups for queued unit %s, ignoring: %m", i->id); |
0a1eb06d | 1843 | |
4ad49000 LP |
1844 | n++; |
1845 | } | |
ecedd90f | 1846 | |
4ad49000 | 1847 | return n; |
8e274523 LP |
1848 | } |
1849 | ||
91a6073e | 1850 | static void unit_add_siblings_to_cgroup_realize_queue(Unit *u) { |
4ad49000 | 1851 | Unit *slice; |
ca949c9d | 1852 | |
4ad49000 LP |
1853 | /* This adds the siblings of the specified unit and the |
1854 | * siblings of all parent units to the cgroup queue. (But | |
1855 | * neither the specified unit itself nor the parents.) */ | |
1856 | ||
1857 | while ((slice = UNIT_DEREF(u->slice))) { | |
1858 | Iterator i; | |
1859 | Unit *m; | |
eef85c4a | 1860 | void *v; |
8f53a7b8 | 1861 | |
eef85c4a | 1862 | HASHMAP_FOREACH_KEY(v, m, u->dependencies[UNIT_BEFORE], i) { |
4ad49000 LP |
1863 | if (m == u) |
1864 | continue; | |
8e274523 | 1865 | |
6414b7c9 DS |
1866 | /* Skip units that have a dependency on the slice |
1867 | * but aren't actually in it. */ | |
4ad49000 | 1868 | if (UNIT_DEREF(m->slice) != slice) |
50159e6a | 1869 | continue; |
8e274523 | 1870 | |
6414b7c9 DS |
1871 | /* No point in doing cgroup application for units |
1872 | * without active processes. */ | |
1873 | if (UNIT_IS_INACTIVE_OR_FAILED(unit_active_state(m))) | |
1874 | continue; | |
1875 | ||
1876 | /* If the unit doesn't need any new controllers | |
1877 | * and has current ones realized, it doesn't need | |
1878 | * any changes. */ | |
906c06f6 DM |
1879 | if (unit_has_mask_realized(m, |
1880 | unit_get_target_mask(m), | |
17f14955 | 1881 | unit_get_enable_mask(m))) |
6414b7c9 DS |
1882 | continue; |
1883 | ||
91a6073e | 1884 | unit_add_to_cgroup_realize_queue(m); |
50159e6a LP |
1885 | } |
1886 | ||
4ad49000 | 1887 | u = slice; |
8e274523 | 1888 | } |
4ad49000 LP |
1889 | } |
1890 | ||
0a1eb06d | 1891 | int unit_realize_cgroup(Unit *u) { |
4ad49000 LP |
1892 | assert(u); |
1893 | ||
35b7ff80 | 1894 | if (!UNIT_HAS_CGROUP_CONTEXT(u)) |
0a1eb06d | 1895 | return 0; |
8e274523 | 1896 | |
4ad49000 LP |
1897 | /* So, here's the deal: when realizing the cgroups for this |
1898 | * unit, we need to first create all parents, but there's more | |
1899 | * actually: for the weight-based controllers we also need to | |
1900 | * make sure that all our siblings (i.e. units that are in the | |
73e231ab | 1901 | * same slice as we are) have cgroups, too. Otherwise, things |
4ad49000 LP |
1902 | * would become very uneven as each of their processes would |
1903 | * get as much resources as all our group together. This call | |
1904 | * will synchronously create the parent cgroups, but will | |
1905 | * defer work on the siblings to the next event loop | |
1906 | * iteration. */ | |
ca949c9d | 1907 | |
4ad49000 | 1908 | /* Add all sibling slices to the cgroup queue. */ |
91a6073e | 1909 | unit_add_siblings_to_cgroup_realize_queue(u); |
4ad49000 | 1910 | |
6414b7c9 | 1911 | /* And realize this one now (and apply the values) */ |
db785129 | 1912 | return unit_realize_cgroup_now(u, manager_state(u->manager)); |
8e274523 LP |
1913 | } |
1914 | ||
efdb0237 LP |
1915 | void unit_release_cgroup(Unit *u) { |
1916 | assert(u); | |
1917 | ||
1918 | /* Forgets all cgroup details for this cgroup */ | |
1919 | ||
1920 | if (u->cgroup_path) { | |
1921 | (void) hashmap_remove(u->manager->cgroup_unit, u->cgroup_path); | |
1922 | u->cgroup_path = mfree(u->cgroup_path); | |
1923 | } | |
1924 | ||
1925 | if (u->cgroup_inotify_wd >= 0) { | |
1926 | if (inotify_rm_watch(u->manager->cgroup_inotify_fd, u->cgroup_inotify_wd) < 0) | |
5e1ee764 | 1927 | log_unit_debug_errno(u, errno, "Failed to remove cgroup inotify watch %i for %s, ignoring: %m", u->cgroup_inotify_wd, u->id); |
efdb0237 LP |
1928 | |
1929 | (void) hashmap_remove(u->manager->cgroup_inotify_wd_unit, INT_TO_PTR(u->cgroup_inotify_wd)); | |
1930 | u->cgroup_inotify_wd = -1; | |
1931 | } | |
1932 | } | |
1933 | ||
1934 | void unit_prune_cgroup(Unit *u) { | |
8e274523 | 1935 | int r; |
efdb0237 | 1936 | bool is_root_slice; |
8e274523 | 1937 | |
4ad49000 | 1938 | assert(u); |
8e274523 | 1939 | |
efdb0237 LP |
1940 | /* Removes the cgroup, if empty and possible, and stops watching it. */ |
1941 | ||
4ad49000 LP |
1942 | if (!u->cgroup_path) |
1943 | return; | |
8e274523 | 1944 | |
fe700f46 LP |
1945 | (void) unit_get_cpu_usage(u, NULL); /* Cache the last CPU usage value before we destroy the cgroup */ |
1946 | ||
efdb0237 LP |
1947 | is_root_slice = unit_has_name(u, SPECIAL_ROOT_SLICE); |
1948 | ||
1949 | r = cg_trim_everywhere(u->manager->cgroup_supported, u->cgroup_path, !is_root_slice); | |
dab5bf85 | 1950 | if (r < 0) { |
f29ff115 | 1951 | log_unit_debug_errno(u, r, "Failed to destroy cgroup %s, ignoring: %m", u->cgroup_path); |
dab5bf85 RL |
1952 | return; |
1953 | } | |
8e274523 | 1954 | |
efdb0237 LP |
1955 | if (is_root_slice) |
1956 | return; | |
1957 | ||
1958 | unit_release_cgroup(u); | |
0a1eb06d | 1959 | |
4ad49000 | 1960 | u->cgroup_realized = false; |
bc432dc7 | 1961 | u->cgroup_realized_mask = 0; |
ccf78df1 | 1962 | u->cgroup_enabled_mask = 0; |
084c7007 RG |
1963 | |
1964 | u->bpf_device_control_installed = bpf_program_unref(u->bpf_device_control_installed); | |
8e274523 LP |
1965 | } |
1966 | ||
efdb0237 | 1967 | int unit_search_main_pid(Unit *u, pid_t *ret) { |
4ad49000 LP |
1968 | _cleanup_fclose_ FILE *f = NULL; |
1969 | pid_t pid = 0, npid, mypid; | |
efdb0237 | 1970 | int r; |
4ad49000 LP |
1971 | |
1972 | assert(u); | |
efdb0237 | 1973 | assert(ret); |
4ad49000 LP |
1974 | |
1975 | if (!u->cgroup_path) | |
efdb0237 | 1976 | return -ENXIO; |
4ad49000 | 1977 | |
efdb0237 LP |
1978 | r = cg_enumerate_processes(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path, &f); |
1979 | if (r < 0) | |
1980 | return r; | |
4ad49000 | 1981 | |
df0ff127 | 1982 | mypid = getpid_cached(); |
4ad49000 LP |
1983 | while (cg_read_pid(f, &npid) > 0) { |
1984 | pid_t ppid; | |
1985 | ||
1986 | if (npid == pid) | |
1987 | continue; | |
8e274523 | 1988 | |
4ad49000 | 1989 | /* Ignore processes that aren't our kids */ |
6bc73acb | 1990 | if (get_process_ppid(npid, &ppid) >= 0 && ppid != mypid) |
4ad49000 | 1991 | continue; |
8e274523 | 1992 | |
efdb0237 | 1993 | if (pid != 0) |
4ad49000 LP |
1994 | /* Dang, there's more than one daemonized PID |
1995 | in this group, so we don't know what process | |
1996 | is the main process. */ | |
efdb0237 LP |
1997 | |
1998 | return -ENODATA; | |
8e274523 | 1999 | |
4ad49000 | 2000 | pid = npid; |
8e274523 LP |
2001 | } |
2002 | ||
efdb0237 LP |
2003 | *ret = pid; |
2004 | return 0; | |
2005 | } | |
2006 | ||
2007 | static int unit_watch_pids_in_path(Unit *u, const char *path) { | |
b3c5bad3 | 2008 | _cleanup_closedir_ DIR *d = NULL; |
efdb0237 LP |
2009 | _cleanup_fclose_ FILE *f = NULL; |
2010 | int ret = 0, r; | |
2011 | ||
2012 | assert(u); | |
2013 | assert(path); | |
2014 | ||
2015 | r = cg_enumerate_processes(SYSTEMD_CGROUP_CONTROLLER, path, &f); | |
2016 | if (r < 0) | |
2017 | ret = r; | |
2018 | else { | |
2019 | pid_t pid; | |
2020 | ||
2021 | while ((r = cg_read_pid(f, &pid)) > 0) { | |
2022 | r = unit_watch_pid(u, pid); | |
2023 | if (r < 0 && ret >= 0) | |
2024 | ret = r; | |
2025 | } | |
2026 | ||
2027 | if (r < 0 && ret >= 0) | |
2028 | ret = r; | |
2029 | } | |
2030 | ||
2031 | r = cg_enumerate_subgroups(SYSTEMD_CGROUP_CONTROLLER, path, &d); | |
2032 | if (r < 0) { | |
2033 | if (ret >= 0) | |
2034 | ret = r; | |
2035 | } else { | |
2036 | char *fn; | |
2037 | ||
2038 | while ((r = cg_read_subgroup(d, &fn)) > 0) { | |
2039 | _cleanup_free_ char *p = NULL; | |
2040 | ||
605405c6 | 2041 | p = strjoin(path, "/", fn); |
efdb0237 LP |
2042 | free(fn); |
2043 | ||
2044 | if (!p) | |
2045 | return -ENOMEM; | |
2046 | ||
2047 | r = unit_watch_pids_in_path(u, p); | |
2048 | if (r < 0 && ret >= 0) | |
2049 | ret = r; | |
2050 | } | |
2051 | ||
2052 | if (r < 0 && ret >= 0) | |
2053 | ret = r; | |
2054 | } | |
2055 | ||
2056 | return ret; | |
2057 | } | |
2058 | ||
11aef522 LP |
2059 | int unit_synthesize_cgroup_empty_event(Unit *u) { |
2060 | int r; | |
2061 | ||
2062 | assert(u); | |
2063 | ||
2064 | /* Enqueue a synthetic cgroup empty event if this unit doesn't watch any PIDs anymore. This is compatibility | |
2065 | * support for non-unified systems where notifications aren't reliable, and hence need to take whatever we can | |
2066 | * get as notification source as soon as we stopped having any useful PIDs to watch for. */ | |
2067 | ||
2068 | if (!u->cgroup_path) | |
2069 | return -ENOENT; | |
2070 | ||
2071 | r = cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER); | |
2072 | if (r < 0) | |
2073 | return r; | |
2074 | if (r > 0) /* On unified we have reliable notifications, and don't need this */ | |
2075 | return 0; | |
2076 | ||
2077 | if (!set_isempty(u->pids)) | |
2078 | return 0; | |
2079 | ||
2080 | unit_add_to_cgroup_empty_queue(u); | |
2081 | return 0; | |
2082 | } | |
2083 | ||
efdb0237 | 2084 | int unit_watch_all_pids(Unit *u) { |
b4cccbc1 LP |
2085 | int r; |
2086 | ||
efdb0237 LP |
2087 | assert(u); |
2088 | ||
2089 | /* Adds all PIDs from our cgroup to the set of PIDs we | |
2090 | * watch. This is a fallback logic for cases where we do not | |
2091 | * get reliable cgroup empty notifications: we try to use | |
2092 | * SIGCHLD as replacement. */ | |
2093 | ||
2094 | if (!u->cgroup_path) | |
2095 | return -ENOENT; | |
2096 | ||
c22800e4 | 2097 | r = cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER); |
b4cccbc1 LP |
2098 | if (r < 0) |
2099 | return r; | |
2100 | if (r > 0) /* On unified we can use proper notifications */ | |
efdb0237 LP |
2101 | return 0; |
2102 | ||
2103 | return unit_watch_pids_in_path(u, u->cgroup_path); | |
2104 | } | |
2105 | ||
09e24654 LP |
2106 | static int on_cgroup_empty_event(sd_event_source *s, void *userdata) { |
2107 | Manager *m = userdata; | |
2108 | Unit *u; | |
efdb0237 LP |
2109 | int r; |
2110 | ||
09e24654 LP |
2111 | assert(s); |
2112 | assert(m); | |
efdb0237 | 2113 | |
09e24654 LP |
2114 | u = m->cgroup_empty_queue; |
2115 | if (!u) | |
efdb0237 LP |
2116 | return 0; |
2117 | ||
09e24654 LP |
2118 | assert(u->in_cgroup_empty_queue); |
2119 | u->in_cgroup_empty_queue = false; | |
2120 | LIST_REMOVE(cgroup_empty_queue, m->cgroup_empty_queue, u); | |
2121 | ||
2122 | if (m->cgroup_empty_queue) { | |
2123 | /* More stuff queued, let's make sure we remain enabled */ | |
2124 | r = sd_event_source_set_enabled(s, SD_EVENT_ONESHOT); | |
2125 | if (r < 0) | |
19a691a9 | 2126 | log_debug_errno(r, "Failed to reenable cgroup empty event source, ignoring: %m"); |
09e24654 | 2127 | } |
efdb0237 LP |
2128 | |
2129 | unit_add_to_gc_queue(u); | |
2130 | ||
2131 | if (UNIT_VTABLE(u)->notify_cgroup_empty) | |
2132 | UNIT_VTABLE(u)->notify_cgroup_empty(u); | |
2133 | ||
2134 | return 0; | |
2135 | } | |
2136 | ||
09e24654 LP |
2137 | void unit_add_to_cgroup_empty_queue(Unit *u) { |
2138 | int r; | |
2139 | ||
2140 | assert(u); | |
2141 | ||
2142 | /* Note that there are four different ways how cgroup empty events reach us: | |
2143 | * | |
2144 | * 1. On the unified hierarchy we get an inotify event on the cgroup | |
2145 | * | |
2146 | * 2. On the legacy hierarchy, when running in system mode, we get a datagram on the cgroup agent socket | |
2147 | * | |
2148 | * 3. On the legacy hierarchy, when running in user mode, we get a D-Bus signal on the system bus | |
2149 | * | |
2150 | * 4. On the legacy hierarchy, in service units we start watching all processes of the cgroup for SIGCHLD as | |
2151 | * soon as we get one SIGCHLD, to deal with unreliable cgroup notifications. | |
2152 | * | |
2153 | * Regardless which way we got the notification, we'll verify it here, and then add it to a separate | |
2154 | * queue. This queue will be dispatched at a lower priority than the SIGCHLD handler, so that we always use | |
2155 | * SIGCHLD if we can get it first, and only use the cgroup empty notifications if there's no SIGCHLD pending | |
2156 | * (which might happen if the cgroup doesn't contain processes that are our own child, which is typically the | |
2157 | * case for scope units). */ | |
2158 | ||
2159 | if (u->in_cgroup_empty_queue) | |
2160 | return; | |
2161 | ||
2162 | /* Let's verify that the cgroup is really empty */ | |
2163 | if (!u->cgroup_path) | |
2164 | return; | |
2165 | r = cg_is_empty_recursive(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path); | |
2166 | if (r < 0) { | |
2167 | log_unit_debug_errno(u, r, "Failed to determine whether cgroup %s is empty: %m", u->cgroup_path); | |
2168 | return; | |
2169 | } | |
2170 | if (r == 0) | |
2171 | return; | |
2172 | ||
2173 | LIST_PREPEND(cgroup_empty_queue, u->manager->cgroup_empty_queue, u); | |
2174 | u->in_cgroup_empty_queue = true; | |
2175 | ||
2176 | /* Trigger the defer event */ | |
2177 | r = sd_event_source_set_enabled(u->manager->cgroup_empty_event_source, SD_EVENT_ONESHOT); | |
2178 | if (r < 0) | |
2179 | log_debug_errno(r, "Failed to enable cgroup empty event source: %m"); | |
2180 | } | |
2181 | ||
efdb0237 LP |
2182 | static int on_cgroup_inotify_event(sd_event_source *s, int fd, uint32_t revents, void *userdata) { |
2183 | Manager *m = userdata; | |
2184 | ||
2185 | assert(s); | |
2186 | assert(fd >= 0); | |
2187 | assert(m); | |
2188 | ||
2189 | for (;;) { | |
2190 | union inotify_event_buffer buffer; | |
2191 | struct inotify_event *e; | |
2192 | ssize_t l; | |
2193 | ||
2194 | l = read(fd, &buffer, sizeof(buffer)); | |
2195 | if (l < 0) { | |
47249640 | 2196 | if (IN_SET(errno, EINTR, EAGAIN)) |
efdb0237 LP |
2197 | return 0; |
2198 | ||
2199 | return log_error_errno(errno, "Failed to read control group inotify events: %m"); | |
2200 | } | |
2201 | ||
2202 | FOREACH_INOTIFY_EVENT(e, buffer, l) { | |
2203 | Unit *u; | |
2204 | ||
2205 | if (e->wd < 0) | |
2206 | /* Queue overflow has no watch descriptor */ | |
2207 | continue; | |
2208 | ||
2209 | if (e->mask & IN_IGNORED) | |
2210 | /* The watch was just removed */ | |
2211 | continue; | |
2212 | ||
2213 | u = hashmap_get(m->cgroup_inotify_wd_unit, INT_TO_PTR(e->wd)); | |
2214 | if (!u) /* Not that inotify might deliver | |
2215 | * events for a watch even after it | |
2216 | * was removed, because it was queued | |
2217 | * before the removal. Let's ignore | |
2218 | * this here safely. */ | |
2219 | continue; | |
2220 | ||
09e24654 | 2221 | unit_add_to_cgroup_empty_queue(u); |
efdb0237 LP |
2222 | } |
2223 | } | |
8e274523 LP |
2224 | } |
2225 | ||
17f14955 RG |
2226 | static int cg_bpf_mask_supported(CGroupMask *ret) { |
2227 | CGroupMask mask = 0; | |
2228 | int r; | |
2229 | ||
2230 | /* BPF-based firewall */ | |
2231 | r = bpf_firewall_supported(); | |
2232 | if (r > 0) | |
2233 | mask |= CGROUP_MASK_BPF_FIREWALL; | |
2234 | ||
084c7007 RG |
2235 | /* BPF-based device access control */ |
2236 | r = bpf_devices_supported(); | |
2237 | if (r > 0) | |
2238 | mask |= CGROUP_MASK_BPF_DEVICES; | |
2239 | ||
17f14955 RG |
2240 | *ret = mask; |
2241 | return 0; | |
2242 | } | |
2243 | ||
8e274523 | 2244 | int manager_setup_cgroup(Manager *m) { |
9444b1f2 | 2245 | _cleanup_free_ char *path = NULL; |
10bd3e2e | 2246 | const char *scope_path; |
efdb0237 | 2247 | CGroupController c; |
b4cccbc1 | 2248 | int r, all_unified; |
17f14955 | 2249 | CGroupMask mask; |
efdb0237 | 2250 | char *e; |
8e274523 LP |
2251 | |
2252 | assert(m); | |
2253 | ||
35d2e7ec | 2254 | /* 1. Determine hierarchy */ |
efdb0237 | 2255 | m->cgroup_root = mfree(m->cgroup_root); |
9444b1f2 | 2256 | r = cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER, 0, &m->cgroup_root); |
23bbb0de MS |
2257 | if (r < 0) |
2258 | return log_error_errno(r, "Cannot determine cgroup we are running in: %m"); | |
8e274523 | 2259 | |
efdb0237 LP |
2260 | /* Chop off the init scope, if we are already located in it */ |
2261 | e = endswith(m->cgroup_root, "/" SPECIAL_INIT_SCOPE); | |
0d8c31ff | 2262 | |
efdb0237 LP |
2263 | /* LEGACY: Also chop off the system slice if we are in |
2264 | * it. This is to support live upgrades from older systemd | |
2265 | * versions where PID 1 was moved there. Also see | |
2266 | * cg_get_root_path(). */ | |
463d0d15 | 2267 | if (!e && MANAGER_IS_SYSTEM(m)) { |
9444b1f2 | 2268 | e = endswith(m->cgroup_root, "/" SPECIAL_SYSTEM_SLICE); |
15c60e99 | 2269 | if (!e) |
efdb0237 | 2270 | e = endswith(m->cgroup_root, "/system"); /* even more legacy */ |
0baf24dd | 2271 | } |
efdb0237 LP |
2272 | if (e) |
2273 | *e = 0; | |
7ccfb64a | 2274 | |
7546145e LP |
2275 | /* And make sure to store away the root value without trailing slash, even for the root dir, so that we can |
2276 | * easily prepend it everywhere. */ | |
2277 | delete_trailing_chars(m->cgroup_root, "/"); | |
8e274523 | 2278 | |
35d2e7ec | 2279 | /* 2. Show data */ |
9444b1f2 | 2280 | r = cg_get_path(SYSTEMD_CGROUP_CONTROLLER, m->cgroup_root, NULL, &path); |
23bbb0de MS |
2281 | if (r < 0) |
2282 | return log_error_errno(r, "Cannot find cgroup mount point: %m"); | |
8e274523 | 2283 | |
415fc41c TH |
2284 | r = cg_unified_flush(); |
2285 | if (r < 0) | |
2286 | return log_error_errno(r, "Couldn't determine if we are running in the unified hierarchy: %m"); | |
5da38d07 | 2287 | |
b4cccbc1 | 2288 | all_unified = cg_all_unified(); |
d4c819ed ZJS |
2289 | if (all_unified < 0) |
2290 | return log_error_errno(all_unified, "Couldn't determine whether we are in all unified mode: %m"); | |
2291 | if (all_unified > 0) | |
efdb0237 | 2292 | log_debug("Unified cgroup hierarchy is located at %s.", path); |
b4cccbc1 | 2293 | else { |
c22800e4 | 2294 | r = cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER); |
b4cccbc1 LP |
2295 | if (r < 0) |
2296 | return log_error_errno(r, "Failed to determine whether systemd's own controller is in unified mode: %m"); | |
2297 | if (r > 0) | |
2298 | log_debug("Unified cgroup hierarchy is located at %s. Controllers are on legacy hierarchies.", path); | |
2299 | else | |
2300 | log_debug("Using cgroup controller " SYSTEMD_CGROUP_CONTROLLER_LEGACY ". File system hierarchy is at %s.", path); | |
2301 | } | |
efdb0237 | 2302 | |
09e24654 LP |
2303 | /* 3. Allocate cgroup empty defer event source */ |
2304 | m->cgroup_empty_event_source = sd_event_source_unref(m->cgroup_empty_event_source); | |
2305 | r = sd_event_add_defer(m->event, &m->cgroup_empty_event_source, on_cgroup_empty_event, m); | |
2306 | if (r < 0) | |
2307 | return log_error_errno(r, "Failed to create cgroup empty event source: %m"); | |
2308 | ||
2309 | r = sd_event_source_set_priority(m->cgroup_empty_event_source, SD_EVENT_PRIORITY_NORMAL-5); | |
2310 | if (r < 0) | |
2311 | return log_error_errno(r, "Failed to set priority of cgroup empty event source: %m"); | |
2312 | ||
2313 | r = sd_event_source_set_enabled(m->cgroup_empty_event_source, SD_EVENT_OFF); | |
2314 | if (r < 0) | |
2315 | return log_error_errno(r, "Failed to disable cgroup empty event source: %m"); | |
2316 | ||
2317 | (void) sd_event_source_set_description(m->cgroup_empty_event_source, "cgroup-empty"); | |
2318 | ||
2319 | /* 4. Install notifier inotify object, or agent */ | |
10bd3e2e | 2320 | if (cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER) > 0) { |
c6c18be3 | 2321 | |
09e24654 | 2322 | /* In the unified hierarchy we can get cgroup empty notifications via inotify. */ |
efdb0237 | 2323 | |
10bd3e2e LP |
2324 | m->cgroup_inotify_event_source = sd_event_source_unref(m->cgroup_inotify_event_source); |
2325 | safe_close(m->cgroup_inotify_fd); | |
efdb0237 | 2326 | |
10bd3e2e LP |
2327 | m->cgroup_inotify_fd = inotify_init1(IN_NONBLOCK|IN_CLOEXEC); |
2328 | if (m->cgroup_inotify_fd < 0) | |
2329 | return log_error_errno(errno, "Failed to create control group inotify object: %m"); | |
efdb0237 | 2330 | |
10bd3e2e LP |
2331 | r = sd_event_add_io(m->event, &m->cgroup_inotify_event_source, m->cgroup_inotify_fd, EPOLLIN, on_cgroup_inotify_event, m); |
2332 | if (r < 0) | |
2333 | return log_error_errno(r, "Failed to watch control group inotify object: %m"); | |
efdb0237 | 2334 | |
10bd3e2e LP |
2335 | /* Process cgroup empty notifications early, but after service notifications and SIGCHLD. Also |
2336 | * see handling of cgroup agent notifications, for the classic cgroup hierarchy support. */ | |
09e24654 | 2337 | r = sd_event_source_set_priority(m->cgroup_inotify_event_source, SD_EVENT_PRIORITY_NORMAL-4); |
10bd3e2e LP |
2338 | if (r < 0) |
2339 | return log_error_errno(r, "Failed to set priority of inotify event source: %m"); | |
efdb0237 | 2340 | |
10bd3e2e | 2341 | (void) sd_event_source_set_description(m->cgroup_inotify_event_source, "cgroup-inotify"); |
efdb0237 | 2342 | |
611c4f8a | 2343 | } else if (MANAGER_IS_SYSTEM(m) && manager_owns_host_root_cgroup(m) && !MANAGER_IS_TEST_RUN(m)) { |
efdb0237 | 2344 | |
10bd3e2e LP |
2345 | /* On the legacy hierarchy we only get notifications via cgroup agents. (Which isn't really reliable, |
2346 | * since it does not generate events when control groups with children run empty. */ | |
8e274523 | 2347 | |
10bd3e2e | 2348 | r = cg_install_release_agent(SYSTEMD_CGROUP_CONTROLLER, SYSTEMD_CGROUP_AGENT_PATH); |
23bbb0de | 2349 | if (r < 0) |
10bd3e2e LP |
2350 | log_warning_errno(r, "Failed to install release agent, ignoring: %m"); |
2351 | else if (r > 0) | |
2352 | log_debug("Installed release agent."); | |
2353 | else if (r == 0) | |
2354 | log_debug("Release agent already installed."); | |
2355 | } | |
efdb0237 | 2356 | |
09e24654 | 2357 | /* 5. Make sure we are in the special "init.scope" unit in the root slice. */ |
10bd3e2e LP |
2358 | scope_path = strjoina(m->cgroup_root, "/" SPECIAL_INIT_SCOPE); |
2359 | r = cg_create_and_attach(SYSTEMD_CGROUP_CONTROLLER, scope_path, 0); | |
aa77e234 MS |
2360 | if (r >= 0) { |
2361 | /* Also, move all other userspace processes remaining in the root cgroup into that scope. */ | |
2362 | r = cg_migrate(SYSTEMD_CGROUP_CONTROLLER, m->cgroup_root, SYSTEMD_CGROUP_CONTROLLER, scope_path, 0); | |
2363 | if (r < 0) | |
2364 | log_warning_errno(r, "Couldn't move remaining userspace processes, ignoring: %m"); | |
c6c18be3 | 2365 | |
aa77e234 MS |
2366 | /* 6. And pin it, so that it cannot be unmounted */ |
2367 | safe_close(m->pin_cgroupfs_fd); | |
2368 | m->pin_cgroupfs_fd = open(path, O_RDONLY|O_CLOEXEC|O_DIRECTORY|O_NOCTTY|O_NONBLOCK); | |
2369 | if (m->pin_cgroupfs_fd < 0) | |
2370 | return log_error_errno(errno, "Failed to open pin file: %m"); | |
0d8c31ff | 2371 | |
638cece4 | 2372 | } else if (!MANAGER_IS_TEST_RUN(m)) |
aa77e234 | 2373 | return log_error_errno(r, "Failed to create %s control group: %m", scope_path); |
10bd3e2e | 2374 | |
09e24654 | 2375 | /* 7. Always enable hierarchical support if it exists... */ |
638cece4 | 2376 | if (!all_unified && !MANAGER_IS_TEST_RUN(m)) |
10bd3e2e | 2377 | (void) cg_set_attribute("memory", "/", "memory.use_hierarchy", "1"); |
c6c18be3 | 2378 | |
17f14955 | 2379 | /* 8. Figure out which controllers are supported */ |
efdb0237 LP |
2380 | r = cg_mask_supported(&m->cgroup_supported); |
2381 | if (r < 0) | |
2382 | return log_error_errno(r, "Failed to determine supported controllers: %m"); | |
17f14955 RG |
2383 | |
2384 | /* 9. Figure out which bpf-based pseudo-controllers are supported */ | |
2385 | r = cg_bpf_mask_supported(&mask); | |
2386 | if (r < 0) | |
2387 | return log_error_errno(r, "Failed to determine supported bpf-based pseudo-controllers: %m"); | |
2388 | m->cgroup_supported |= mask; | |
2389 | ||
2390 | /* 10. Log which controllers are supported */ | |
efdb0237 | 2391 | for (c = 0; c < _CGROUP_CONTROLLER_MAX; c++) |
eee0a1e4 | 2392 | log_debug("Controller '%s' supported: %s", cgroup_controller_to_string(c), yes_no(m->cgroup_supported & CGROUP_CONTROLLER_TO_MASK(c))); |
9156e799 | 2393 | |
a32360f1 | 2394 | return 0; |
8e274523 LP |
2395 | } |
2396 | ||
c6c18be3 | 2397 | void manager_shutdown_cgroup(Manager *m, bool delete) { |
8e274523 LP |
2398 | assert(m); |
2399 | ||
9444b1f2 LP |
2400 | /* We can't really delete the group, since we are in it. But |
2401 | * let's trim it. */ | |
f6c63f6f | 2402 | if (delete && m->cgroup_root && m->test_run_flags != MANAGER_TEST_RUN_MINIMAL) |
efdb0237 LP |
2403 | (void) cg_trim(SYSTEMD_CGROUP_CONTROLLER, m->cgroup_root, false); |
2404 | ||
09e24654 LP |
2405 | m->cgroup_empty_event_source = sd_event_source_unref(m->cgroup_empty_event_source); |
2406 | ||
efdb0237 LP |
2407 | m->cgroup_inotify_wd_unit = hashmap_free(m->cgroup_inotify_wd_unit); |
2408 | ||
2409 | m->cgroup_inotify_event_source = sd_event_source_unref(m->cgroup_inotify_event_source); | |
2410 | m->cgroup_inotify_fd = safe_close(m->cgroup_inotify_fd); | |
8e274523 | 2411 | |
03e334a1 | 2412 | m->pin_cgroupfs_fd = safe_close(m->pin_cgroupfs_fd); |
c6c18be3 | 2413 | |
efdb0237 | 2414 | m->cgroup_root = mfree(m->cgroup_root); |
8e274523 LP |
2415 | } |
2416 | ||
4ad49000 | 2417 | Unit* manager_get_unit_by_cgroup(Manager *m, const char *cgroup) { |
acb14d31 | 2418 | char *p; |
4ad49000 | 2419 | Unit *u; |
acb14d31 LP |
2420 | |
2421 | assert(m); | |
2422 | assert(cgroup); | |
acb14d31 | 2423 | |
4ad49000 LP |
2424 | u = hashmap_get(m->cgroup_unit, cgroup); |
2425 | if (u) | |
2426 | return u; | |
acb14d31 | 2427 | |
8e70580b | 2428 | p = strdupa(cgroup); |
acb14d31 LP |
2429 | for (;;) { |
2430 | char *e; | |
2431 | ||
2432 | e = strrchr(p, '/'); | |
efdb0237 LP |
2433 | if (!e || e == p) |
2434 | return hashmap_get(m->cgroup_unit, SPECIAL_ROOT_SLICE); | |
acb14d31 LP |
2435 | |
2436 | *e = 0; | |
2437 | ||
4ad49000 LP |
2438 | u = hashmap_get(m->cgroup_unit, p); |
2439 | if (u) | |
2440 | return u; | |
acb14d31 LP |
2441 | } |
2442 | } | |
2443 | ||
b3ac818b | 2444 | Unit *manager_get_unit_by_pid_cgroup(Manager *m, pid_t pid) { |
4ad49000 | 2445 | _cleanup_free_ char *cgroup = NULL; |
8e274523 | 2446 | |
8c47c732 LP |
2447 | assert(m); |
2448 | ||
62a76913 | 2449 | if (!pid_is_valid(pid)) |
b3ac818b LP |
2450 | return NULL; |
2451 | ||
62a76913 | 2452 | if (cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER, pid, &cgroup) < 0) |
b3ac818b LP |
2453 | return NULL; |
2454 | ||
2455 | return manager_get_unit_by_cgroup(m, cgroup); | |
2456 | } | |
2457 | ||
2458 | Unit *manager_get_unit_by_pid(Manager *m, pid_t pid) { | |
62a76913 | 2459 | Unit *u, **array; |
b3ac818b LP |
2460 | |
2461 | assert(m); | |
2462 | ||
62a76913 LP |
2463 | /* Note that a process might be owned by multiple units, we return only one here, which is good enough for most |
2464 | * cases, though not strictly correct. We prefer the one reported by cgroup membership, as that's the most | |
2465 | * relevant one as children of the process will be assigned to that one, too, before all else. */ | |
2466 | ||
2467 | if (!pid_is_valid(pid)) | |
8c47c732 LP |
2468 | return NULL; |
2469 | ||
2ca9d979 | 2470 | if (pid == getpid_cached()) |
efdb0237 LP |
2471 | return hashmap_get(m->units, SPECIAL_INIT_SCOPE); |
2472 | ||
62a76913 | 2473 | u = manager_get_unit_by_pid_cgroup(m, pid); |
5fe8876b LP |
2474 | if (u) |
2475 | return u; | |
2476 | ||
62a76913 | 2477 | u = hashmap_get(m->watch_pids, PID_TO_PTR(pid)); |
5fe8876b LP |
2478 | if (u) |
2479 | return u; | |
2480 | ||
62a76913 LP |
2481 | array = hashmap_get(m->watch_pids, PID_TO_PTR(-pid)); |
2482 | if (array) | |
2483 | return array[0]; | |
2484 | ||
2485 | return NULL; | |
6dde1f33 | 2486 | } |
4fbf50b3 | 2487 | |
4ad49000 LP |
2488 | int manager_notify_cgroup_empty(Manager *m, const char *cgroup) { |
2489 | Unit *u; | |
4fbf50b3 | 2490 | |
4ad49000 LP |
2491 | assert(m); |
2492 | assert(cgroup); | |
4fbf50b3 | 2493 | |
09e24654 LP |
2494 | /* Called on the legacy hierarchy whenever we get an explicit cgroup notification from the cgroup agent process |
2495 | * or from the --system instance */ | |
2496 | ||
d8fdc620 LP |
2497 | log_debug("Got cgroup empty notification for: %s", cgroup); |
2498 | ||
4ad49000 | 2499 | u = manager_get_unit_by_cgroup(m, cgroup); |
5ad096b3 LP |
2500 | if (!u) |
2501 | return 0; | |
b56c28c3 | 2502 | |
09e24654 LP |
2503 | unit_add_to_cgroup_empty_queue(u); |
2504 | return 1; | |
5ad096b3 LP |
2505 | } |
2506 | ||
2507 | int unit_get_memory_current(Unit *u, uint64_t *ret) { | |
2508 | _cleanup_free_ char *v = NULL; | |
2509 | int r; | |
2510 | ||
2511 | assert(u); | |
2512 | assert(ret); | |
2513 | ||
2e4025c0 | 2514 | if (!UNIT_CGROUP_BOOL(u, memory_accounting)) |
cf3b4be1 LP |
2515 | return -ENODATA; |
2516 | ||
5ad096b3 LP |
2517 | if (!u->cgroup_path) |
2518 | return -ENODATA; | |
2519 | ||
1f73aa00 | 2520 | /* The root cgroup doesn't expose this information, let's get it from /proc instead */ |
611c4f8a | 2521 | if (unit_has_host_root_cgroup(u)) |
1f73aa00 LP |
2522 | return procfs_memory_get_current(ret); |
2523 | ||
efdb0237 | 2524 | if ((u->cgroup_realized_mask & CGROUP_MASK_MEMORY) == 0) |
5ad096b3 LP |
2525 | return -ENODATA; |
2526 | ||
b4cccbc1 LP |
2527 | r = cg_all_unified(); |
2528 | if (r < 0) | |
2529 | return r; | |
2530 | if (r > 0) | |
efdb0237 | 2531 | r = cg_get_attribute("memory", u->cgroup_path, "memory.current", &v); |
b4cccbc1 LP |
2532 | else |
2533 | r = cg_get_attribute("memory", u->cgroup_path, "memory.usage_in_bytes", &v); | |
5ad096b3 LP |
2534 | if (r == -ENOENT) |
2535 | return -ENODATA; | |
2536 | if (r < 0) | |
2537 | return r; | |
2538 | ||
2539 | return safe_atou64(v, ret); | |
2540 | } | |
2541 | ||
03a7b521 LP |
2542 | int unit_get_tasks_current(Unit *u, uint64_t *ret) { |
2543 | _cleanup_free_ char *v = NULL; | |
2544 | int r; | |
2545 | ||
2546 | assert(u); | |
2547 | assert(ret); | |
2548 | ||
2e4025c0 | 2549 | if (!UNIT_CGROUP_BOOL(u, tasks_accounting)) |
cf3b4be1 LP |
2550 | return -ENODATA; |
2551 | ||
03a7b521 LP |
2552 | if (!u->cgroup_path) |
2553 | return -ENODATA; | |
2554 | ||
c36a69f4 | 2555 | /* The root cgroup doesn't expose this information, let's get it from /proc instead */ |
611c4f8a | 2556 | if (unit_has_host_root_cgroup(u)) |
c36a69f4 LP |
2557 | return procfs_tasks_get_current(ret); |
2558 | ||
1f73aa00 LP |
2559 | if ((u->cgroup_realized_mask & CGROUP_MASK_PIDS) == 0) |
2560 | return -ENODATA; | |
2561 | ||
03a7b521 LP |
2562 | r = cg_get_attribute("pids", u->cgroup_path, "pids.current", &v); |
2563 | if (r == -ENOENT) | |
2564 | return -ENODATA; | |
2565 | if (r < 0) | |
2566 | return r; | |
2567 | ||
2568 | return safe_atou64(v, ret); | |
2569 | } | |
2570 | ||
5ad096b3 LP |
2571 | static int unit_get_cpu_usage_raw(Unit *u, nsec_t *ret) { |
2572 | _cleanup_free_ char *v = NULL; | |
2573 | uint64_t ns; | |
2574 | int r; | |
2575 | ||
2576 | assert(u); | |
2577 | assert(ret); | |
2578 | ||
2579 | if (!u->cgroup_path) | |
2580 | return -ENODATA; | |
2581 | ||
1f73aa00 | 2582 | /* The root cgroup doesn't expose this information, let's get it from /proc instead */ |
611c4f8a | 2583 | if (unit_has_host_root_cgroup(u)) |
1f73aa00 LP |
2584 | return procfs_cpu_get_usage(ret); |
2585 | ||
b4cccbc1 LP |
2586 | r = cg_all_unified(); |
2587 | if (r < 0) | |
2588 | return r; | |
f98c2585 CD |
2589 | |
2590 | /* Requisite controllers for CPU accounting are not enabled */ | |
2591 | if ((get_cpu_accounting_mask() & ~u->cgroup_realized_mask) != 0) | |
2592 | return -ENODATA; | |
2593 | ||
b4cccbc1 | 2594 | if (r > 0) { |
66ebf6c0 TH |
2595 | _cleanup_free_ char *val = NULL; |
2596 | uint64_t us; | |
5ad096b3 | 2597 | |
b734a4ff | 2598 | r = cg_get_keyed_attribute("cpu", u->cgroup_path, "cpu.stat", STRV_MAKE("usage_usec"), &val); |
66ebf6c0 TH |
2599 | if (r < 0) |
2600 | return r; | |
b734a4ff LP |
2601 | if (IN_SET(r, -ENOENT, -ENXIO)) |
2602 | return -ENODATA; | |
66ebf6c0 TH |
2603 | |
2604 | r = safe_atou64(val, &us); | |
2605 | if (r < 0) | |
2606 | return r; | |
2607 | ||
2608 | ns = us * NSEC_PER_USEC; | |
2609 | } else { | |
66ebf6c0 TH |
2610 | r = cg_get_attribute("cpuacct", u->cgroup_path, "cpuacct.usage", &v); |
2611 | if (r == -ENOENT) | |
2612 | return -ENODATA; | |
2613 | if (r < 0) | |
2614 | return r; | |
2615 | ||
2616 | r = safe_atou64(v, &ns); | |
2617 | if (r < 0) | |
2618 | return r; | |
2619 | } | |
5ad096b3 LP |
2620 | |
2621 | *ret = ns; | |
2622 | return 0; | |
2623 | } | |
2624 | ||
2625 | int unit_get_cpu_usage(Unit *u, nsec_t *ret) { | |
2626 | nsec_t ns; | |
2627 | int r; | |
2628 | ||
fe700f46 LP |
2629 | assert(u); |
2630 | ||
2631 | /* Retrieve the current CPU usage counter. This will subtract the CPU counter taken when the unit was | |
2632 | * started. If the cgroup has been removed already, returns the last cached value. To cache the value, simply | |
2633 | * call this function with a NULL return value. */ | |
2634 | ||
2e4025c0 | 2635 | if (!UNIT_CGROUP_BOOL(u, cpu_accounting)) |
cf3b4be1 LP |
2636 | return -ENODATA; |
2637 | ||
5ad096b3 | 2638 | r = unit_get_cpu_usage_raw(u, &ns); |
fe700f46 LP |
2639 | if (r == -ENODATA && u->cpu_usage_last != NSEC_INFINITY) { |
2640 | /* If we can't get the CPU usage anymore (because the cgroup was already removed, for example), use our | |
2641 | * cached value. */ | |
2642 | ||
2643 | if (ret) | |
2644 | *ret = u->cpu_usage_last; | |
2645 | return 0; | |
2646 | } | |
5ad096b3 LP |
2647 | if (r < 0) |
2648 | return r; | |
2649 | ||
66ebf6c0 TH |
2650 | if (ns > u->cpu_usage_base) |
2651 | ns -= u->cpu_usage_base; | |
5ad096b3 LP |
2652 | else |
2653 | ns = 0; | |
2654 | ||
fe700f46 LP |
2655 | u->cpu_usage_last = ns; |
2656 | if (ret) | |
2657 | *ret = ns; | |
2658 | ||
5ad096b3 LP |
2659 | return 0; |
2660 | } | |
2661 | ||
906c06f6 DM |
2662 | int unit_get_ip_accounting( |
2663 | Unit *u, | |
2664 | CGroupIPAccountingMetric metric, | |
2665 | uint64_t *ret) { | |
2666 | ||
6b659ed8 | 2667 | uint64_t value; |
906c06f6 DM |
2668 | int fd, r; |
2669 | ||
2670 | assert(u); | |
2671 | assert(metric >= 0); | |
2672 | assert(metric < _CGROUP_IP_ACCOUNTING_METRIC_MAX); | |
2673 | assert(ret); | |
2674 | ||
2e4025c0 | 2675 | if (!UNIT_CGROUP_BOOL(u, ip_accounting)) |
cf3b4be1 LP |
2676 | return -ENODATA; |
2677 | ||
906c06f6 DM |
2678 | fd = IN_SET(metric, CGROUP_IP_INGRESS_BYTES, CGROUP_IP_INGRESS_PACKETS) ? |
2679 | u->ip_accounting_ingress_map_fd : | |
2680 | u->ip_accounting_egress_map_fd; | |
906c06f6 DM |
2681 | if (fd < 0) |
2682 | return -ENODATA; | |
2683 | ||
2684 | if (IN_SET(metric, CGROUP_IP_INGRESS_BYTES, CGROUP_IP_EGRESS_BYTES)) | |
6b659ed8 | 2685 | r = bpf_firewall_read_accounting(fd, &value, NULL); |
906c06f6 | 2686 | else |
6b659ed8 LP |
2687 | r = bpf_firewall_read_accounting(fd, NULL, &value); |
2688 | if (r < 0) | |
2689 | return r; | |
2690 | ||
2691 | /* Add in additional metrics from a previous runtime. Note that when reexecing/reloading the daemon we compile | |
2692 | * all BPF programs and maps anew, but serialize the old counters. When deserializing we store them in the | |
2693 | * ip_accounting_extra[] field, and add them in here transparently. */ | |
2694 | ||
2695 | *ret = value + u->ip_accounting_extra[metric]; | |
906c06f6 DM |
2696 | |
2697 | return r; | |
2698 | } | |
2699 | ||
2700 | int unit_reset_cpu_accounting(Unit *u) { | |
5ad096b3 LP |
2701 | nsec_t ns; |
2702 | int r; | |
2703 | ||
2704 | assert(u); | |
2705 | ||
fe700f46 LP |
2706 | u->cpu_usage_last = NSEC_INFINITY; |
2707 | ||
5ad096b3 LP |
2708 | r = unit_get_cpu_usage_raw(u, &ns); |
2709 | if (r < 0) { | |
66ebf6c0 | 2710 | u->cpu_usage_base = 0; |
5ad096b3 | 2711 | return r; |
b56c28c3 | 2712 | } |
2633eb83 | 2713 | |
66ebf6c0 | 2714 | u->cpu_usage_base = ns; |
4ad49000 | 2715 | return 0; |
4fbf50b3 LP |
2716 | } |
2717 | ||
906c06f6 DM |
2718 | int unit_reset_ip_accounting(Unit *u) { |
2719 | int r = 0, q = 0; | |
2720 | ||
2721 | assert(u); | |
2722 | ||
2723 | if (u->ip_accounting_ingress_map_fd >= 0) | |
2724 | r = bpf_firewall_reset_accounting(u->ip_accounting_ingress_map_fd); | |
2725 | ||
2726 | if (u->ip_accounting_egress_map_fd >= 0) | |
2727 | q = bpf_firewall_reset_accounting(u->ip_accounting_egress_map_fd); | |
2728 | ||
6b659ed8 LP |
2729 | zero(u->ip_accounting_extra); |
2730 | ||
906c06f6 DM |
2731 | return r < 0 ? r : q; |
2732 | } | |
2733 | ||
e7ab4d1a LP |
2734 | void unit_invalidate_cgroup(Unit *u, CGroupMask m) { |
2735 | assert(u); | |
2736 | ||
2737 | if (!UNIT_HAS_CGROUP_CONTEXT(u)) | |
2738 | return; | |
2739 | ||
2740 | if (m == 0) | |
2741 | return; | |
2742 | ||
538b4852 TH |
2743 | /* always invalidate compat pairs together */ |
2744 | if (m & (CGROUP_MASK_IO | CGROUP_MASK_BLKIO)) | |
2745 | m |= CGROUP_MASK_IO | CGROUP_MASK_BLKIO; | |
2746 | ||
7cce4fb7 LP |
2747 | if (m & (CGROUP_MASK_CPU | CGROUP_MASK_CPUACCT)) |
2748 | m |= CGROUP_MASK_CPU | CGROUP_MASK_CPUACCT; | |
2749 | ||
60c728ad | 2750 | if ((u->cgroup_realized_mask & m) == 0) /* NOP? */ |
e7ab4d1a LP |
2751 | return; |
2752 | ||
2753 | u->cgroup_realized_mask &= ~m; | |
91a6073e | 2754 | unit_add_to_cgroup_realize_queue(u); |
e7ab4d1a LP |
2755 | } |
2756 | ||
906c06f6 DM |
2757 | void unit_invalidate_cgroup_bpf(Unit *u) { |
2758 | assert(u); | |
2759 | ||
2760 | if (!UNIT_HAS_CGROUP_CONTEXT(u)) | |
2761 | return; | |
2762 | ||
17f14955 | 2763 | if (u->cgroup_invalidated_mask & CGROUP_MASK_BPF_FIREWALL) /* NOP? */ |
906c06f6 DM |
2764 | return; |
2765 | ||
17f14955 | 2766 | u->cgroup_invalidated_mask |= CGROUP_MASK_BPF_FIREWALL; |
91a6073e | 2767 | unit_add_to_cgroup_realize_queue(u); |
906c06f6 DM |
2768 | |
2769 | /* If we are a slice unit, we also need to put compile a new BPF program for all our children, as the IP access | |
2770 | * list of our children includes our own. */ | |
2771 | if (u->type == UNIT_SLICE) { | |
2772 | Unit *member; | |
2773 | Iterator i; | |
eef85c4a | 2774 | void *v; |
906c06f6 | 2775 | |
eef85c4a | 2776 | HASHMAP_FOREACH_KEY(v, member, u->dependencies[UNIT_BEFORE], i) { |
906c06f6 DM |
2777 | if (member == u) |
2778 | continue; | |
2779 | ||
2780 | if (UNIT_DEREF(member->slice) != u) | |
2781 | continue; | |
2782 | ||
2783 | unit_invalidate_cgroup_bpf(member); | |
2784 | } | |
2785 | } | |
2786 | } | |
2787 | ||
1d9cc876 LP |
2788 | bool unit_cgroup_delegate(Unit *u) { |
2789 | CGroupContext *c; | |
2790 | ||
2791 | assert(u); | |
2792 | ||
2793 | if (!UNIT_VTABLE(u)->can_delegate) | |
2794 | return false; | |
2795 | ||
2796 | c = unit_get_cgroup_context(u); | |
2797 | if (!c) | |
2798 | return false; | |
2799 | ||
2800 | return c->delegate; | |
2801 | } | |
2802 | ||
e7ab4d1a LP |
2803 | void manager_invalidate_startup_units(Manager *m) { |
2804 | Iterator i; | |
2805 | Unit *u; | |
2806 | ||
2807 | assert(m); | |
2808 | ||
2809 | SET_FOREACH(u, m->startup_units, i) | |
13c31542 | 2810 | unit_invalidate_cgroup(u, CGROUP_MASK_CPU|CGROUP_MASK_IO|CGROUP_MASK_BLKIO); |
e7ab4d1a LP |
2811 | } |
2812 | ||
4ad49000 LP |
2813 | static const char* const cgroup_device_policy_table[_CGROUP_DEVICE_POLICY_MAX] = { |
2814 | [CGROUP_AUTO] = "auto", | |
2815 | [CGROUP_CLOSED] = "closed", | |
2816 | [CGROUP_STRICT] = "strict", | |
2817 | }; | |
4fbf50b3 | 2818 | |
4ad49000 | 2819 | DEFINE_STRING_TABLE_LOOKUP(cgroup_device_policy, CGroupDevicePolicy); |