]>
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 | ||
cc6271f1 LP |
29 | bool manager_owns_root_cgroup(Manager *m) { |
30 | assert(m); | |
31 | ||
32 | /* Returns true if we are managing the root cgroup. Note that it isn't sufficient to just check whether the | |
33 | * group root path equals "/" since that will also be the case if CLONE_NEWCGROUP is in the mix. Since there's | |
34 | * appears to be no nice way to detect whether we are in a CLONE_NEWCGROUP namespace we instead just check if | |
35 | * we run in any kind of container virtualization. */ | |
36 | ||
37 | if (detect_container() > 0) | |
38 | return false; | |
39 | ||
57ea45e1 | 40 | return empty_or_root(m->cgroup_root); |
cc6271f1 LP |
41 | } |
42 | ||
f3725e64 LP |
43 | bool unit_has_root_cgroup(Unit *u) { |
44 | assert(u); | |
45 | ||
cc6271f1 LP |
46 | /* Returns whether this unit manages the root cgroup. This will return true if this unit is the root slice and |
47 | * the manager manages the root cgroup. */ | |
f3725e64 | 48 | |
cc6271f1 | 49 | if (!manager_owns_root_cgroup(u->manager)) |
f3725e64 LP |
50 | return false; |
51 | ||
cc6271f1 | 52 | return unit_has_name(u, SPECIAL_ROOT_SLICE); |
f3725e64 LP |
53 | } |
54 | ||
2b40998d | 55 | static void cgroup_compat_warn(void) { |
128fadc9 TH |
56 | static bool cgroup_compat_warned = false; |
57 | ||
58 | if (cgroup_compat_warned) | |
59 | return; | |
60 | ||
cc6271f1 LP |
61 | log_warning("cgroup compatibility translation between legacy and unified hierarchy settings activated. " |
62 | "See cgroup-compat debug messages for details."); | |
63 | ||
128fadc9 TH |
64 | cgroup_compat_warned = true; |
65 | } | |
66 | ||
67 | #define log_cgroup_compat(unit, fmt, ...) do { \ | |
68 | cgroup_compat_warn(); \ | |
69 | log_unit_debug(unit, "cgroup-compat: " fmt, ##__VA_ARGS__); \ | |
2b40998d | 70 | } while (false) |
128fadc9 | 71 | |
4ad49000 LP |
72 | void cgroup_context_init(CGroupContext *c) { |
73 | assert(c); | |
74 | ||
75 | /* Initialize everything to the kernel defaults, assuming the | |
76 | * structure is preinitialized to 0 */ | |
77 | ||
66ebf6c0 TH |
78 | c->cpu_weight = CGROUP_WEIGHT_INVALID; |
79 | c->startup_cpu_weight = CGROUP_WEIGHT_INVALID; | |
80 | c->cpu_quota_per_sec_usec = USEC_INFINITY; | |
81 | ||
d53d9474 LP |
82 | c->cpu_shares = CGROUP_CPU_SHARES_INVALID; |
83 | c->startup_cpu_shares = CGROUP_CPU_SHARES_INVALID; | |
d53d9474 | 84 | |
da4d897e TH |
85 | c->memory_high = CGROUP_LIMIT_MAX; |
86 | c->memory_max = CGROUP_LIMIT_MAX; | |
96e131ea | 87 | c->memory_swap_max = CGROUP_LIMIT_MAX; |
da4d897e TH |
88 | |
89 | c->memory_limit = CGROUP_LIMIT_MAX; | |
b2f8b02e | 90 | |
13c31542 TH |
91 | c->io_weight = CGROUP_WEIGHT_INVALID; |
92 | c->startup_io_weight = CGROUP_WEIGHT_INVALID; | |
93 | ||
d53d9474 LP |
94 | c->blockio_weight = CGROUP_BLKIO_WEIGHT_INVALID; |
95 | c->startup_blockio_weight = CGROUP_BLKIO_WEIGHT_INVALID; | |
96 | ||
97 | c->tasks_max = (uint64_t) -1; | |
4ad49000 | 98 | } |
8e274523 | 99 | |
4ad49000 LP |
100 | void cgroup_context_free_device_allow(CGroupContext *c, CGroupDeviceAllow *a) { |
101 | assert(c); | |
102 | assert(a); | |
103 | ||
71fda00f | 104 | LIST_REMOVE(device_allow, c->device_allow, a); |
4ad49000 LP |
105 | free(a->path); |
106 | free(a); | |
107 | } | |
108 | ||
13c31542 TH |
109 | void cgroup_context_free_io_device_weight(CGroupContext *c, CGroupIODeviceWeight *w) { |
110 | assert(c); | |
111 | assert(w); | |
112 | ||
113 | LIST_REMOVE(device_weights, c->io_device_weights, w); | |
114 | free(w->path); | |
115 | free(w); | |
116 | } | |
117 | ||
6ae4283c TH |
118 | void cgroup_context_free_io_device_latency(CGroupContext *c, CGroupIODeviceLatency *l) { |
119 | assert(c); | |
120 | assert(l); | |
121 | ||
122 | LIST_REMOVE(device_latencies, c->io_device_latencies, l); | |
123 | free(l->path); | |
124 | free(l); | |
125 | } | |
126 | ||
13c31542 TH |
127 | void cgroup_context_free_io_device_limit(CGroupContext *c, CGroupIODeviceLimit *l) { |
128 | assert(c); | |
129 | assert(l); | |
130 | ||
131 | LIST_REMOVE(device_limits, c->io_device_limits, l); | |
132 | free(l->path); | |
133 | free(l); | |
134 | } | |
135 | ||
4ad49000 LP |
136 | void cgroup_context_free_blockio_device_weight(CGroupContext *c, CGroupBlockIODeviceWeight *w) { |
137 | assert(c); | |
138 | assert(w); | |
139 | ||
71fda00f | 140 | LIST_REMOVE(device_weights, c->blockio_device_weights, w); |
4ad49000 LP |
141 | free(w->path); |
142 | free(w); | |
143 | } | |
144 | ||
145 | void cgroup_context_free_blockio_device_bandwidth(CGroupContext *c, CGroupBlockIODeviceBandwidth *b) { | |
146 | assert(c); | |
8e274523 | 147 | assert(b); |
8e274523 | 148 | |
71fda00f | 149 | LIST_REMOVE(device_bandwidths, c->blockio_device_bandwidths, b); |
4ad49000 LP |
150 | free(b->path); |
151 | free(b); | |
152 | } | |
153 | ||
154 | void cgroup_context_done(CGroupContext *c) { | |
155 | assert(c); | |
156 | ||
13c31542 TH |
157 | while (c->io_device_weights) |
158 | cgroup_context_free_io_device_weight(c, c->io_device_weights); | |
159 | ||
6ae4283c TH |
160 | while (c->io_device_latencies) |
161 | cgroup_context_free_io_device_latency(c, c->io_device_latencies); | |
162 | ||
13c31542 TH |
163 | while (c->io_device_limits) |
164 | cgroup_context_free_io_device_limit(c, c->io_device_limits); | |
165 | ||
4ad49000 LP |
166 | while (c->blockio_device_weights) |
167 | cgroup_context_free_blockio_device_weight(c, c->blockio_device_weights); | |
168 | ||
169 | while (c->blockio_device_bandwidths) | |
170 | cgroup_context_free_blockio_device_bandwidth(c, c->blockio_device_bandwidths); | |
171 | ||
172 | while (c->device_allow) | |
173 | cgroup_context_free_device_allow(c, c->device_allow); | |
6a48d82f DM |
174 | |
175 | c->ip_address_allow = ip_address_access_free_all(c->ip_address_allow); | |
176 | c->ip_address_deny = ip_address_access_free_all(c->ip_address_deny); | |
4ad49000 LP |
177 | } |
178 | ||
179 | void cgroup_context_dump(CGroupContext *c, FILE* f, const char *prefix) { | |
13c31542 TH |
180 | CGroupIODeviceLimit *il; |
181 | CGroupIODeviceWeight *iw; | |
6ae4283c | 182 | CGroupIODeviceLatency *l; |
4ad49000 LP |
183 | CGroupBlockIODeviceBandwidth *b; |
184 | CGroupBlockIODeviceWeight *w; | |
185 | CGroupDeviceAllow *a; | |
c21c9906 | 186 | IPAddressAccessItem *iaai; |
9a054909 | 187 | char u[FORMAT_TIMESPAN_MAX]; |
4ad49000 LP |
188 | |
189 | assert(c); | |
190 | assert(f); | |
191 | ||
192 | prefix = strempty(prefix); | |
193 | ||
194 | fprintf(f, | |
195 | "%sCPUAccounting=%s\n" | |
13c31542 | 196 | "%sIOAccounting=%s\n" |
4ad49000 LP |
197 | "%sBlockIOAccounting=%s\n" |
198 | "%sMemoryAccounting=%s\n" | |
d53d9474 | 199 | "%sTasksAccounting=%s\n" |
c21c9906 | 200 | "%sIPAccounting=%s\n" |
66ebf6c0 TH |
201 | "%sCPUWeight=%" PRIu64 "\n" |
202 | "%sStartupCPUWeight=%" PRIu64 "\n" | |
d53d9474 LP |
203 | "%sCPUShares=%" PRIu64 "\n" |
204 | "%sStartupCPUShares=%" PRIu64 "\n" | |
b2f8b02e | 205 | "%sCPUQuotaPerSecSec=%s\n" |
13c31542 TH |
206 | "%sIOWeight=%" PRIu64 "\n" |
207 | "%sStartupIOWeight=%" PRIu64 "\n" | |
d53d9474 LP |
208 | "%sBlockIOWeight=%" PRIu64 "\n" |
209 | "%sStartupBlockIOWeight=%" PRIu64 "\n" | |
48422635 | 210 | "%sMemoryMin=%" PRIu64 "\n" |
da4d897e TH |
211 | "%sMemoryLow=%" PRIu64 "\n" |
212 | "%sMemoryHigh=%" PRIu64 "\n" | |
213 | "%sMemoryMax=%" PRIu64 "\n" | |
96e131ea | 214 | "%sMemorySwapMax=%" PRIu64 "\n" |
4ad49000 | 215 | "%sMemoryLimit=%" PRIu64 "\n" |
03a7b521 | 216 | "%sTasksMax=%" PRIu64 "\n" |
a931ad47 LP |
217 | "%sDevicePolicy=%s\n" |
218 | "%sDelegate=%s\n", | |
4ad49000 | 219 | prefix, yes_no(c->cpu_accounting), |
13c31542 | 220 | prefix, yes_no(c->io_accounting), |
4ad49000 LP |
221 | prefix, yes_no(c->blockio_accounting), |
222 | prefix, yes_no(c->memory_accounting), | |
d53d9474 | 223 | prefix, yes_no(c->tasks_accounting), |
c21c9906 | 224 | prefix, yes_no(c->ip_accounting), |
66ebf6c0 TH |
225 | prefix, c->cpu_weight, |
226 | prefix, c->startup_cpu_weight, | |
4ad49000 | 227 | prefix, c->cpu_shares, |
95ae05c0 | 228 | prefix, c->startup_cpu_shares, |
b1d6dcf5 | 229 | prefix, format_timespan(u, sizeof(u), c->cpu_quota_per_sec_usec, 1), |
13c31542 TH |
230 | prefix, c->io_weight, |
231 | prefix, c->startup_io_weight, | |
4ad49000 | 232 | prefix, c->blockio_weight, |
95ae05c0 | 233 | prefix, c->startup_blockio_weight, |
48422635 | 234 | prefix, c->memory_min, |
da4d897e TH |
235 | prefix, c->memory_low, |
236 | prefix, c->memory_high, | |
237 | prefix, c->memory_max, | |
96e131ea | 238 | prefix, c->memory_swap_max, |
4ad49000 | 239 | prefix, c->memory_limit, |
03a7b521 | 240 | prefix, c->tasks_max, |
a931ad47 LP |
241 | prefix, cgroup_device_policy_to_string(c->device_policy), |
242 | prefix, yes_no(c->delegate)); | |
4ad49000 | 243 | |
02638280 LP |
244 | if (c->delegate) { |
245 | _cleanup_free_ char *t = NULL; | |
246 | ||
247 | (void) cg_mask_to_string(c->delegate_controllers, &t); | |
248 | ||
47a78d41 | 249 | fprintf(f, "%sDelegateControllers=%s\n", |
02638280 LP |
250 | prefix, |
251 | strempty(t)); | |
252 | } | |
253 | ||
4ad49000 LP |
254 | LIST_FOREACH(device_allow, a, c->device_allow) |
255 | fprintf(f, | |
256 | "%sDeviceAllow=%s %s%s%s\n", | |
257 | prefix, | |
258 | a->path, | |
259 | a->r ? "r" : "", a->w ? "w" : "", a->m ? "m" : ""); | |
260 | ||
13c31542 TH |
261 | LIST_FOREACH(device_weights, iw, c->io_device_weights) |
262 | fprintf(f, | |
6ae4283c | 263 | "%sIODeviceWeight=%s %" PRIu64 "\n", |
13c31542 TH |
264 | prefix, |
265 | iw->path, | |
266 | iw->weight); | |
267 | ||
6ae4283c TH |
268 | LIST_FOREACH(device_latencies, l, c->io_device_latencies) |
269 | fprintf(f, | |
270 | "%sIODeviceLatencyTargetSec=%s %s\n", | |
271 | prefix, | |
272 | l->path, | |
273 | format_timespan(u, sizeof(u), l->target_usec, 1)); | |
274 | ||
13c31542 TH |
275 | LIST_FOREACH(device_limits, il, c->io_device_limits) { |
276 | char buf[FORMAT_BYTES_MAX]; | |
9be57249 TH |
277 | CGroupIOLimitType type; |
278 | ||
279 | for (type = 0; type < _CGROUP_IO_LIMIT_TYPE_MAX; type++) | |
280 | if (il->limits[type] != cgroup_io_limit_defaults[type]) | |
281 | fprintf(f, | |
282 | "%s%s=%s %s\n", | |
283 | prefix, | |
284 | cgroup_io_limit_type_to_string(type), | |
285 | il->path, | |
286 | format_bytes(buf, sizeof(buf), il->limits[type])); | |
13c31542 TH |
287 | } |
288 | ||
4ad49000 LP |
289 | LIST_FOREACH(device_weights, w, c->blockio_device_weights) |
290 | fprintf(f, | |
d53d9474 | 291 | "%sBlockIODeviceWeight=%s %" PRIu64, |
4ad49000 LP |
292 | prefix, |
293 | w->path, | |
294 | w->weight); | |
295 | ||
296 | LIST_FOREACH(device_bandwidths, b, c->blockio_device_bandwidths) { | |
297 | char buf[FORMAT_BYTES_MAX]; | |
298 | ||
979d0311 TH |
299 | if (b->rbps != CGROUP_LIMIT_MAX) |
300 | fprintf(f, | |
301 | "%sBlockIOReadBandwidth=%s %s\n", | |
302 | prefix, | |
303 | b->path, | |
304 | format_bytes(buf, sizeof(buf), b->rbps)); | |
305 | if (b->wbps != CGROUP_LIMIT_MAX) | |
306 | fprintf(f, | |
307 | "%sBlockIOWriteBandwidth=%s %s\n", | |
308 | prefix, | |
309 | b->path, | |
310 | format_bytes(buf, sizeof(buf), b->wbps)); | |
4ad49000 | 311 | } |
c21c9906 LP |
312 | |
313 | LIST_FOREACH(items, iaai, c->ip_address_allow) { | |
314 | _cleanup_free_ char *k = NULL; | |
315 | ||
316 | (void) in_addr_to_string(iaai->family, &iaai->address, &k); | |
317 | fprintf(f, "%sIPAddressAllow=%s/%u\n", prefix, strnull(k), iaai->prefixlen); | |
318 | } | |
319 | ||
320 | LIST_FOREACH(items, iaai, c->ip_address_deny) { | |
321 | _cleanup_free_ char *k = NULL; | |
322 | ||
323 | (void) in_addr_to_string(iaai->family, &iaai->address, &k); | |
324 | fprintf(f, "%sIPAddressDeny=%s/%u\n", prefix, strnull(k), iaai->prefixlen); | |
325 | } | |
4ad49000 LP |
326 | } |
327 | ||
fd870bac YW |
328 | int cgroup_add_device_allow(CGroupContext *c, const char *dev, const char *mode) { |
329 | _cleanup_free_ CGroupDeviceAllow *a = NULL; | |
330 | _cleanup_free_ char *d = NULL; | |
331 | ||
332 | assert(c); | |
333 | assert(dev); | |
334 | assert(isempty(mode) || in_charset(mode, "rwm")); | |
335 | ||
336 | a = new(CGroupDeviceAllow, 1); | |
337 | if (!a) | |
338 | return -ENOMEM; | |
339 | ||
340 | d = strdup(dev); | |
341 | if (!d) | |
342 | return -ENOMEM; | |
343 | ||
344 | *a = (CGroupDeviceAllow) { | |
345 | .path = TAKE_PTR(d), | |
346 | .r = isempty(mode) || !!strchr(mode, 'r'), | |
347 | .w = isempty(mode) || !!strchr(mode, 'w'), | |
348 | .m = isempty(mode) || !!strchr(mode, 'm'), | |
349 | }; | |
350 | ||
351 | LIST_PREPEND(device_allow, c->device_allow, a); | |
352 | TAKE_PTR(a); | |
353 | ||
354 | return 0; | |
355 | } | |
356 | ||
45c2e068 | 357 | static int lookup_block_device(const char *p, dev_t *ret) { |
4ad49000 | 358 | struct stat st; |
45c2e068 | 359 | int r; |
4ad49000 LP |
360 | |
361 | assert(p); | |
45c2e068 | 362 | assert(ret); |
4ad49000 | 363 | |
b1c05b98 | 364 | if (stat(p, &st) < 0) |
45c2e068 | 365 | return log_warning_errno(errno, "Couldn't stat device '%s': %m", p); |
8e274523 | 366 | |
4ad49000 | 367 | if (S_ISBLK(st.st_mode)) |
45c2e068 LP |
368 | *ret = st.st_rdev; |
369 | else if (major(st.st_dev) != 0) | |
370 | *ret = st.st_dev; /* If this is not a device node then use the block device this file is stored on */ | |
371 | else { | |
372 | /* If this is btrfs, getting the backing block device is a bit harder */ | |
373 | r = btrfs_get_block_device(p, ret); | |
374 | if (r < 0 && r != -ENOTTY) | |
375 | return log_warning_errno(r, "Failed to determine block device backing btrfs file system '%s': %m", p); | |
376 | if (r == -ENOTTY) { | |
377 | log_warning("'%s' is not a block device node, and file system block device cannot be determined or is not local.", p); | |
378 | return -ENODEV; | |
379 | } | |
4ad49000 | 380 | } |
8e274523 | 381 | |
45c2e068 LP |
382 | /* If this is a LUKS device, try to get the originating block device */ |
383 | (void) block_get_originating(*ret, ret); | |
384 | ||
385 | /* If this is a partition, try to get the originating block device */ | |
386 | (void) block_get_whole_disk(*ret, ret); | |
8e274523 | 387 | return 0; |
8e274523 LP |
388 | } |
389 | ||
084c7007 | 390 | static int whitelist_device(BPFProgram *prog, const char *path, const char *node, const char *acc) { |
4ad49000 | 391 | struct stat st; |
b200489b | 392 | bool ignore_notfound; |
8c6db833 | 393 | int r; |
8e274523 | 394 | |
4ad49000 LP |
395 | assert(path); |
396 | assert(acc); | |
8e274523 | 397 | |
b200489b DR |
398 | if (node[0] == '-') { |
399 | /* Non-existent paths starting with "-" must be silently ignored */ | |
400 | node++; | |
401 | ignore_notfound = true; | |
402 | } else | |
403 | ignore_notfound = false; | |
404 | ||
4ad49000 | 405 | if (stat(node, &st) < 0) { |
b200489b | 406 | if (errno == ENOENT && ignore_notfound) |
e7330dfe DP |
407 | return 0; |
408 | ||
409 | return log_warning_errno(errno, "Couldn't stat device %s: %m", node); | |
4ad49000 LP |
410 | } |
411 | ||
412 | if (!S_ISCHR(st.st_mode) && !S_ISBLK(st.st_mode)) { | |
413 | log_warning("%s is not a device.", node); | |
414 | return -ENODEV; | |
415 | } | |
416 | ||
084c7007 RG |
417 | if (cg_all_unified() > 0) { |
418 | if (!prog) | |
419 | return 0; | |
4ad49000 | 420 | |
b9839ac9 LP |
421 | return cgroup_bpf_whitelist_device(prog, S_ISCHR(st.st_mode) ? BPF_DEVCG_DEV_CHAR : BPF_DEVCG_DEV_BLOCK, |
422 | major(st.st_rdev), minor(st.st_rdev), acc); | |
423 | ||
084c7007 RG |
424 | } else { |
425 | char buf[2+DECIMAL_STR_MAX(dev_t)*2+2+4]; | |
426 | ||
427 | sprintf(buf, | |
428 | "%c %u:%u %s", | |
429 | S_ISCHR(st.st_mode) ? 'c' : 'b', | |
430 | major(st.st_rdev), minor(st.st_rdev), | |
431 | acc); | |
432 | ||
433 | r = cg_set_attribute("devices", path, "devices.allow", buf); | |
434 | if (r < 0) | |
b9839ac9 LP |
435 | return log_full_errno(IN_SET(r, -ENOENT, -EROFS, -EINVAL, -EACCES) ? LOG_DEBUG : LOG_WARNING, |
436 | r, "Failed to set devices.allow on %s: %m", path); | |
4ad49000 | 437 | |
b9839ac9 LP |
438 | return 0; |
439 | } | |
8e274523 LP |
440 | } |
441 | ||
084c7007 | 442 | static int whitelist_major(BPFProgram *prog, const char *path, const char *name, char type, const char *acc) { |
90060676 LP |
443 | _cleanup_fclose_ FILE *f = NULL; |
444 | char line[LINE_MAX]; | |
084c7007 | 445 | char *p, *w; |
90060676 LP |
446 | bool good = false; |
447 | int r; | |
448 | ||
449 | assert(path); | |
450 | assert(acc); | |
4c701096 | 451 | assert(IN_SET(type, 'b', 'c')); |
90060676 LP |
452 | |
453 | f = fopen("/proc/devices", "re"); | |
4a62c710 MS |
454 | if (!f) |
455 | return log_warning_errno(errno, "Cannot open /proc/devices to resolve %s (%c): %m", name, type); | |
90060676 LP |
456 | |
457 | FOREACH_LINE(line, f, goto fail) { | |
90060676 LP |
458 | unsigned maj; |
459 | ||
460 | truncate_nl(line); | |
461 | ||
462 | if (type == 'c' && streq(line, "Character devices:")) { | |
463 | good = true; | |
464 | continue; | |
465 | } | |
466 | ||
467 | if (type == 'b' && streq(line, "Block devices:")) { | |
468 | good = true; | |
469 | continue; | |
470 | } | |
471 | ||
472 | if (isempty(line)) { | |
473 | good = false; | |
474 | continue; | |
475 | } | |
476 | ||
477 | if (!good) | |
478 | continue; | |
479 | ||
480 | p = strstrip(line); | |
481 | ||
482 | w = strpbrk(p, WHITESPACE); | |
483 | if (!w) | |
484 | continue; | |
485 | *w = 0; | |
486 | ||
487 | r = safe_atou(p, &maj); | |
488 | if (r < 0) | |
489 | continue; | |
490 | if (maj <= 0) | |
491 | continue; | |
492 | ||
493 | w++; | |
494 | w += strspn(w, WHITESPACE); | |
e41969e3 LP |
495 | |
496 | if (fnmatch(name, w, 0) != 0) | |
90060676 LP |
497 | continue; |
498 | ||
084c7007 RG |
499 | if (cg_all_unified() > 0) { |
500 | if (!prog) | |
501 | continue; | |
90060676 | 502 | |
913c898c LP |
503 | (void) cgroup_bpf_whitelist_major(prog, |
504 | type == 'c' ? BPF_DEVCG_DEV_CHAR : BPF_DEVCG_DEV_BLOCK, | |
505 | maj, acc); | |
084c7007 RG |
506 | } else { |
507 | char buf[2+DECIMAL_STR_MAX(unsigned)+3+4]; | |
508 | ||
509 | sprintf(buf, | |
510 | "%c %u:* %s", | |
511 | type, | |
512 | maj, | |
513 | acc); | |
514 | ||
515 | r = cg_set_attribute("devices", path, "devices.allow", buf); | |
516 | if (r < 0) | |
517 | log_full_errno(IN_SET(r, -ENOENT, -EROFS, -EINVAL, -EACCES) ? LOG_DEBUG : LOG_WARNING, | |
518 | r, "Failed to set devices.allow on %s: %m", path); | |
519 | } | |
90060676 LP |
520 | } |
521 | ||
522 | return 0; | |
523 | ||
524 | fail: | |
25f027c5 | 525 | return log_warning_errno(errno, "Failed to read /proc/devices: %m"); |
90060676 LP |
526 | } |
527 | ||
66ebf6c0 TH |
528 | static bool cgroup_context_has_cpu_weight(CGroupContext *c) { |
529 | return c->cpu_weight != CGROUP_WEIGHT_INVALID || | |
530 | c->startup_cpu_weight != CGROUP_WEIGHT_INVALID; | |
531 | } | |
532 | ||
533 | static bool cgroup_context_has_cpu_shares(CGroupContext *c) { | |
534 | return c->cpu_shares != CGROUP_CPU_SHARES_INVALID || | |
535 | c->startup_cpu_shares != CGROUP_CPU_SHARES_INVALID; | |
536 | } | |
537 | ||
538 | static uint64_t cgroup_context_cpu_weight(CGroupContext *c, ManagerState state) { | |
539 | if (IN_SET(state, MANAGER_STARTING, MANAGER_INITIALIZING) && | |
540 | c->startup_cpu_weight != CGROUP_WEIGHT_INVALID) | |
541 | return c->startup_cpu_weight; | |
542 | else if (c->cpu_weight != CGROUP_WEIGHT_INVALID) | |
543 | return c->cpu_weight; | |
544 | else | |
545 | return CGROUP_WEIGHT_DEFAULT; | |
546 | } | |
547 | ||
548 | static uint64_t cgroup_context_cpu_shares(CGroupContext *c, ManagerState state) { | |
549 | if (IN_SET(state, MANAGER_STARTING, MANAGER_INITIALIZING) && | |
550 | c->startup_cpu_shares != CGROUP_CPU_SHARES_INVALID) | |
551 | return c->startup_cpu_shares; | |
552 | else if (c->cpu_shares != CGROUP_CPU_SHARES_INVALID) | |
553 | return c->cpu_shares; | |
554 | else | |
555 | return CGROUP_CPU_SHARES_DEFAULT; | |
556 | } | |
557 | ||
558 | static void cgroup_apply_unified_cpu_config(Unit *u, uint64_t weight, uint64_t quota) { | |
559 | char buf[MAX(DECIMAL_STR_MAX(uint64_t) + 1, (DECIMAL_STR_MAX(usec_t) + 1) * 2)]; | |
560 | int r; | |
561 | ||
562 | xsprintf(buf, "%" PRIu64 "\n", weight); | |
563 | r = cg_set_attribute("cpu", u->cgroup_path, "cpu.weight", buf); | |
564 | if (r < 0) | |
565 | log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, | |
566 | "Failed to set cpu.weight: %m"); | |
567 | ||
568 | if (quota != USEC_INFINITY) | |
569 | xsprintf(buf, USEC_FMT " " USEC_FMT "\n", | |
570 | quota * CGROUP_CPU_QUOTA_PERIOD_USEC / USEC_PER_SEC, CGROUP_CPU_QUOTA_PERIOD_USEC); | |
571 | else | |
572 | xsprintf(buf, "max " USEC_FMT "\n", CGROUP_CPU_QUOTA_PERIOD_USEC); | |
573 | ||
574 | r = cg_set_attribute("cpu", u->cgroup_path, "cpu.max", buf); | |
575 | ||
576 | if (r < 0) | |
577 | log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, | |
578 | "Failed to set cpu.max: %m"); | |
579 | } | |
580 | ||
581 | static void cgroup_apply_legacy_cpu_config(Unit *u, uint64_t shares, uint64_t quota) { | |
582 | char buf[MAX(DECIMAL_STR_MAX(uint64_t), DECIMAL_STR_MAX(usec_t)) + 1]; | |
583 | int r; | |
584 | ||
585 | xsprintf(buf, "%" PRIu64 "\n", shares); | |
586 | r = cg_set_attribute("cpu", u->cgroup_path, "cpu.shares", buf); | |
587 | if (r < 0) | |
588 | log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, | |
589 | "Failed to set cpu.shares: %m"); | |
590 | ||
591 | xsprintf(buf, USEC_FMT "\n", CGROUP_CPU_QUOTA_PERIOD_USEC); | |
592 | r = cg_set_attribute("cpu", u->cgroup_path, "cpu.cfs_period_us", buf); | |
593 | if (r < 0) | |
594 | log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, | |
595 | "Failed to set cpu.cfs_period_us: %m"); | |
596 | ||
597 | if (quota != USEC_INFINITY) { | |
598 | xsprintf(buf, USEC_FMT "\n", quota * CGROUP_CPU_QUOTA_PERIOD_USEC / USEC_PER_SEC); | |
599 | r = cg_set_attribute("cpu", u->cgroup_path, "cpu.cfs_quota_us", buf); | |
600 | } else | |
601 | r = cg_set_attribute("cpu", u->cgroup_path, "cpu.cfs_quota_us", "-1"); | |
602 | if (r < 0) | |
603 | log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, | |
604 | "Failed to set cpu.cfs_quota_us: %m"); | |
605 | } | |
606 | ||
607 | static uint64_t cgroup_cpu_shares_to_weight(uint64_t shares) { | |
608 | return CLAMP(shares * CGROUP_WEIGHT_DEFAULT / CGROUP_CPU_SHARES_DEFAULT, | |
609 | CGROUP_WEIGHT_MIN, CGROUP_WEIGHT_MAX); | |
610 | } | |
611 | ||
612 | static uint64_t cgroup_cpu_weight_to_shares(uint64_t weight) { | |
613 | return CLAMP(weight * CGROUP_CPU_SHARES_DEFAULT / CGROUP_WEIGHT_DEFAULT, | |
614 | CGROUP_CPU_SHARES_MIN, CGROUP_CPU_SHARES_MAX); | |
615 | } | |
616 | ||
508c45da | 617 | static bool cgroup_context_has_io_config(CGroupContext *c) { |
538b4852 TH |
618 | return c->io_accounting || |
619 | c->io_weight != CGROUP_WEIGHT_INVALID || | |
620 | c->startup_io_weight != CGROUP_WEIGHT_INVALID || | |
621 | c->io_device_weights || | |
6ae4283c | 622 | c->io_device_latencies || |
538b4852 TH |
623 | c->io_device_limits; |
624 | } | |
625 | ||
508c45da | 626 | static bool cgroup_context_has_blockio_config(CGroupContext *c) { |
538b4852 TH |
627 | return c->blockio_accounting || |
628 | c->blockio_weight != CGROUP_BLKIO_WEIGHT_INVALID || | |
629 | c->startup_blockio_weight != CGROUP_BLKIO_WEIGHT_INVALID || | |
630 | c->blockio_device_weights || | |
631 | c->blockio_device_bandwidths; | |
632 | } | |
633 | ||
508c45da | 634 | static uint64_t cgroup_context_io_weight(CGroupContext *c, ManagerState state) { |
64faf04c TH |
635 | if (IN_SET(state, MANAGER_STARTING, MANAGER_INITIALIZING) && |
636 | c->startup_io_weight != CGROUP_WEIGHT_INVALID) | |
637 | return c->startup_io_weight; | |
638 | else if (c->io_weight != CGROUP_WEIGHT_INVALID) | |
639 | return c->io_weight; | |
640 | else | |
641 | return CGROUP_WEIGHT_DEFAULT; | |
642 | } | |
643 | ||
508c45da | 644 | static uint64_t cgroup_context_blkio_weight(CGroupContext *c, ManagerState state) { |
64faf04c TH |
645 | if (IN_SET(state, MANAGER_STARTING, MANAGER_INITIALIZING) && |
646 | c->startup_blockio_weight != CGROUP_BLKIO_WEIGHT_INVALID) | |
647 | return c->startup_blockio_weight; | |
648 | else if (c->blockio_weight != CGROUP_BLKIO_WEIGHT_INVALID) | |
649 | return c->blockio_weight; | |
650 | else | |
651 | return CGROUP_BLKIO_WEIGHT_DEFAULT; | |
652 | } | |
653 | ||
508c45da | 654 | static uint64_t cgroup_weight_blkio_to_io(uint64_t blkio_weight) { |
538b4852 TH |
655 | return CLAMP(blkio_weight * CGROUP_WEIGHT_DEFAULT / CGROUP_BLKIO_WEIGHT_DEFAULT, |
656 | CGROUP_WEIGHT_MIN, CGROUP_WEIGHT_MAX); | |
657 | } | |
658 | ||
508c45da | 659 | static uint64_t cgroup_weight_io_to_blkio(uint64_t io_weight) { |
538b4852 TH |
660 | return CLAMP(io_weight * CGROUP_BLKIO_WEIGHT_DEFAULT / CGROUP_WEIGHT_DEFAULT, |
661 | CGROUP_BLKIO_WEIGHT_MIN, CGROUP_BLKIO_WEIGHT_MAX); | |
662 | } | |
663 | ||
f29ff115 | 664 | static void cgroup_apply_io_device_weight(Unit *u, const char *dev_path, uint64_t io_weight) { |
64faf04c TH |
665 | char buf[DECIMAL_STR_MAX(dev_t)*2+2+DECIMAL_STR_MAX(uint64_t)+1]; |
666 | dev_t dev; | |
667 | int r; | |
668 | ||
669 | r = lookup_block_device(dev_path, &dev); | |
670 | if (r < 0) | |
671 | return; | |
672 | ||
673 | xsprintf(buf, "%u:%u %" PRIu64 "\n", major(dev), minor(dev), io_weight); | |
f29ff115 | 674 | r = cg_set_attribute("io", u->cgroup_path, "io.weight", buf); |
64faf04c | 675 | if (r < 0) |
f29ff115 TH |
676 | log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, |
677 | "Failed to set io.weight: %m"); | |
64faf04c TH |
678 | } |
679 | ||
f29ff115 | 680 | static void cgroup_apply_blkio_device_weight(Unit *u, const char *dev_path, uint64_t blkio_weight) { |
64faf04c TH |
681 | char buf[DECIMAL_STR_MAX(dev_t)*2+2+DECIMAL_STR_MAX(uint64_t)+1]; |
682 | dev_t dev; | |
683 | int r; | |
684 | ||
685 | r = lookup_block_device(dev_path, &dev); | |
686 | if (r < 0) | |
687 | return; | |
688 | ||
689 | xsprintf(buf, "%u:%u %" PRIu64 "\n", major(dev), minor(dev), blkio_weight); | |
f29ff115 | 690 | r = cg_set_attribute("blkio", u->cgroup_path, "blkio.weight_device", buf); |
64faf04c | 691 | if (r < 0) |
f29ff115 TH |
692 | log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, |
693 | "Failed to set blkio.weight_device: %m"); | |
64faf04c TH |
694 | } |
695 | ||
6ae4283c TH |
696 | static void cgroup_apply_io_device_latency(Unit *u, const char *dev_path, usec_t target) { |
697 | char buf[DECIMAL_STR_MAX(dev_t)*2+2+7+DECIMAL_STR_MAX(uint64_t)+1]; | |
698 | dev_t dev; | |
699 | int r; | |
700 | ||
701 | r = lookup_block_device(dev_path, &dev); | |
702 | if (r < 0) | |
703 | return; | |
704 | ||
705 | if (target != USEC_INFINITY) | |
706 | xsprintf(buf, "%u:%u target=%" PRIu64 "\n", major(dev), minor(dev), target); | |
707 | else | |
708 | xsprintf(buf, "%u:%u target=max\n", major(dev), minor(dev)); | |
709 | ||
710 | r = cg_set_attribute("io", u->cgroup_path, "io.latency", buf); | |
711 | if (r < 0) | |
712 | log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, | |
713 | "Failed to set io.latency on cgroup %s: %m", u->cgroup_path); | |
714 | } | |
715 | ||
17ae2780 | 716 | static void cgroup_apply_io_device_limit(Unit *u, const char *dev_path, uint64_t *limits) { |
64faf04c TH |
717 | char limit_bufs[_CGROUP_IO_LIMIT_TYPE_MAX][DECIMAL_STR_MAX(uint64_t)]; |
718 | char buf[DECIMAL_STR_MAX(dev_t)*2+2+(6+DECIMAL_STR_MAX(uint64_t)+1)*4]; | |
719 | CGroupIOLimitType type; | |
720 | dev_t dev; | |
64faf04c TH |
721 | int r; |
722 | ||
723 | r = lookup_block_device(dev_path, &dev); | |
724 | if (r < 0) | |
17ae2780 | 725 | return; |
64faf04c | 726 | |
17ae2780 LP |
727 | for (type = 0; type < _CGROUP_IO_LIMIT_TYPE_MAX; type++) |
728 | if (limits[type] != cgroup_io_limit_defaults[type]) | |
64faf04c | 729 | xsprintf(limit_bufs[type], "%" PRIu64, limits[type]); |
17ae2780 | 730 | else |
64faf04c | 731 | xsprintf(limit_bufs[type], "%s", limits[type] == CGROUP_LIMIT_MAX ? "max" : "0"); |
64faf04c TH |
732 | |
733 | xsprintf(buf, "%u:%u rbps=%s wbps=%s riops=%s wiops=%s\n", major(dev), minor(dev), | |
734 | limit_bufs[CGROUP_IO_RBPS_MAX], limit_bufs[CGROUP_IO_WBPS_MAX], | |
735 | limit_bufs[CGROUP_IO_RIOPS_MAX], limit_bufs[CGROUP_IO_WIOPS_MAX]); | |
f29ff115 | 736 | r = cg_set_attribute("io", u->cgroup_path, "io.max", buf); |
64faf04c | 737 | if (r < 0) |
f29ff115 TH |
738 | log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, |
739 | "Failed to set io.max: %m"); | |
64faf04c TH |
740 | } |
741 | ||
17ae2780 | 742 | static void cgroup_apply_blkio_device_limit(Unit *u, const char *dev_path, uint64_t rbps, uint64_t wbps) { |
64faf04c TH |
743 | char buf[DECIMAL_STR_MAX(dev_t)*2+2+DECIMAL_STR_MAX(uint64_t)+1]; |
744 | dev_t dev; | |
64faf04c TH |
745 | int r; |
746 | ||
747 | r = lookup_block_device(dev_path, &dev); | |
748 | if (r < 0) | |
17ae2780 | 749 | return; |
64faf04c | 750 | |
64faf04c | 751 | sprintf(buf, "%u:%u %" PRIu64 "\n", major(dev), minor(dev), rbps); |
f29ff115 | 752 | r = cg_set_attribute("blkio", u->cgroup_path, "blkio.throttle.read_bps_device", buf); |
64faf04c | 753 | if (r < 0) |
f29ff115 TH |
754 | log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, |
755 | "Failed to set blkio.throttle.read_bps_device: %m"); | |
64faf04c | 756 | |
64faf04c | 757 | sprintf(buf, "%u:%u %" PRIu64 "\n", major(dev), minor(dev), wbps); |
f29ff115 | 758 | r = cg_set_attribute("blkio", u->cgroup_path, "blkio.throttle.write_bps_device", buf); |
64faf04c | 759 | if (r < 0) |
f29ff115 TH |
760 | log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, |
761 | "Failed to set blkio.throttle.write_bps_device: %m"); | |
64faf04c TH |
762 | } |
763 | ||
da4d897e | 764 | static bool cgroup_context_has_unified_memory_config(CGroupContext *c) { |
48422635 | 765 | 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 |
766 | } |
767 | ||
f29ff115 | 768 | static void cgroup_apply_unified_memory_limit(Unit *u, const char *file, uint64_t v) { |
da4d897e TH |
769 | char buf[DECIMAL_STR_MAX(uint64_t) + 1] = "max"; |
770 | int r; | |
771 | ||
772 | if (v != CGROUP_LIMIT_MAX) | |
773 | xsprintf(buf, "%" PRIu64 "\n", v); | |
774 | ||
f29ff115 | 775 | r = cg_set_attribute("memory", u->cgroup_path, file, buf); |
da4d897e | 776 | if (r < 0) |
f29ff115 TH |
777 | log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, |
778 | "Failed to set %s: %m", file); | |
da4d897e TH |
779 | } |
780 | ||
0f2d84d2 | 781 | static void cgroup_apply_firewall(Unit *u) { |
0f2d84d2 LP |
782 | assert(u); |
783 | ||
acf7f253 | 784 | /* Best-effort: let's apply IP firewalling and/or accounting if that's enabled */ |
906c06f6 | 785 | |
acf7f253 | 786 | if (bpf_firewall_compile(u) < 0) |
906c06f6 DM |
787 | return; |
788 | ||
789 | (void) bpf_firewall_install(u); | |
906c06f6 DM |
790 | } |
791 | ||
792 | static void cgroup_context_apply( | |
793 | Unit *u, | |
794 | CGroupMask apply_mask, | |
906c06f6 DM |
795 | ManagerState state) { |
796 | ||
f29ff115 TH |
797 | const char *path; |
798 | CGroupContext *c; | |
01efdf13 | 799 | bool is_root; |
4ad49000 LP |
800 | int r; |
801 | ||
f29ff115 TH |
802 | assert(u); |
803 | ||
906c06f6 | 804 | /* Nothing to do? Exit early! */ |
17f14955 | 805 | if (apply_mask == 0) |
4ad49000 | 806 | return; |
8e274523 | 807 | |
f3725e64 LP |
808 | /* Some cgroup attributes are not supported on the root cgroup, hence silently ignore */ |
809 | is_root = unit_has_root_cgroup(u); | |
810 | ||
811 | assert_se(c = unit_get_cgroup_context(u)); | |
812 | assert_se(path = u->cgroup_path); | |
813 | ||
814 | if (is_root) /* Make sure we don't try to display messages with an empty path. */ | |
6da13913 | 815 | path = "/"; |
01efdf13 | 816 | |
714e2e1d LP |
817 | /* We generally ignore errors caused by read-only mounted |
818 | * cgroup trees (assuming we are running in a container then), | |
819 | * and missing cgroups, i.e. EROFS and ENOENT. */ | |
820 | ||
906c06f6 DM |
821 | if ((apply_mask & CGROUP_MASK_CPU) && !is_root) { |
822 | bool has_weight, has_shares; | |
823 | ||
824 | has_weight = cgroup_context_has_cpu_weight(c); | |
825 | has_shares = cgroup_context_has_cpu_shares(c); | |
8e274523 | 826 | |
b4cccbc1 | 827 | if (cg_all_unified() > 0) { |
66ebf6c0 | 828 | uint64_t weight; |
b2f8b02e | 829 | |
66ebf6c0 TH |
830 | if (has_weight) |
831 | weight = cgroup_context_cpu_weight(c, state); | |
832 | else if (has_shares) { | |
833 | uint64_t shares = cgroup_context_cpu_shares(c, state); | |
b2f8b02e | 834 | |
66ebf6c0 TH |
835 | weight = cgroup_cpu_shares_to_weight(shares); |
836 | ||
837 | log_cgroup_compat(u, "Applying [Startup]CpuShares %" PRIu64 " as [Startup]CpuWeight %" PRIu64 " on %s", | |
838 | shares, weight, path); | |
839 | } else | |
840 | weight = CGROUP_WEIGHT_DEFAULT; | |
841 | ||
842 | cgroup_apply_unified_cpu_config(u, weight, c->cpu_quota_per_sec_usec); | |
843 | } else { | |
844 | uint64_t shares; | |
845 | ||
7d862ab8 | 846 | if (has_weight) { |
66ebf6c0 TH |
847 | uint64_t weight = cgroup_context_cpu_weight(c, state); |
848 | ||
849 | shares = cgroup_cpu_weight_to_shares(weight); | |
850 | ||
851 | log_cgroup_compat(u, "Applying [Startup]CpuWeight %" PRIu64 " as [Startup]CpuShares %" PRIu64 " on %s", | |
852 | weight, shares, path); | |
7d862ab8 TH |
853 | } else if (has_shares) |
854 | shares = cgroup_context_cpu_shares(c, state); | |
855 | else | |
66ebf6c0 TH |
856 | shares = CGROUP_CPU_SHARES_DEFAULT; |
857 | ||
858 | cgroup_apply_legacy_cpu_config(u, shares, c->cpu_quota_per_sec_usec); | |
859 | } | |
4ad49000 LP |
860 | } |
861 | ||
906c06f6 | 862 | if (apply_mask & CGROUP_MASK_IO) { |
538b4852 TH |
863 | bool has_io = cgroup_context_has_io_config(c); |
864 | bool has_blockio = cgroup_context_has_blockio_config(c); | |
13c31542 TH |
865 | |
866 | if (!is_root) { | |
64faf04c TH |
867 | char buf[8+DECIMAL_STR_MAX(uint64_t)+1]; |
868 | uint64_t weight; | |
13c31542 | 869 | |
538b4852 TH |
870 | if (has_io) |
871 | weight = cgroup_context_io_weight(c, state); | |
128fadc9 TH |
872 | else if (has_blockio) { |
873 | uint64_t blkio_weight = cgroup_context_blkio_weight(c, state); | |
874 | ||
875 | weight = cgroup_weight_blkio_to_io(blkio_weight); | |
876 | ||
877 | log_cgroup_compat(u, "Applying [Startup]BlockIOWeight %" PRIu64 " as [Startup]IOWeight %" PRIu64, | |
878 | blkio_weight, weight); | |
879 | } else | |
538b4852 | 880 | weight = CGROUP_WEIGHT_DEFAULT; |
13c31542 TH |
881 | |
882 | xsprintf(buf, "default %" PRIu64 "\n", weight); | |
883 | r = cg_set_attribute("io", path, "io.weight", buf); | |
884 | if (r < 0) | |
f29ff115 TH |
885 | log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, |
886 | "Failed to set io.weight: %m"); | |
13c31542 | 887 | |
538b4852 TH |
888 | if (has_io) { |
889 | CGroupIODeviceWeight *w; | |
890 | ||
538b4852 | 891 | LIST_FOREACH(device_weights, w, c->io_device_weights) |
f29ff115 | 892 | cgroup_apply_io_device_weight(u, w->path, w->weight); |
538b4852 TH |
893 | } else if (has_blockio) { |
894 | CGroupBlockIODeviceWeight *w; | |
895 | ||
128fadc9 TH |
896 | LIST_FOREACH(device_weights, w, c->blockio_device_weights) { |
897 | weight = cgroup_weight_blkio_to_io(w->weight); | |
898 | ||
899 | log_cgroup_compat(u, "Applying BlockIODeviceWeight %" PRIu64 " as IODeviceWeight %" PRIu64 " for %s", | |
900 | w->weight, weight, w->path); | |
901 | ||
902 | cgroup_apply_io_device_weight(u, w->path, weight); | |
903 | } | |
538b4852 | 904 | } |
6ae4283c TH |
905 | |
906 | if (has_io) { | |
907 | CGroupIODeviceLatency *l; | |
908 | ||
909 | LIST_FOREACH(device_latencies, l, c->io_device_latencies) | |
910 | cgroup_apply_io_device_latency(u, l->path, l->target_usec); | |
911 | } | |
13c31542 TH |
912 | } |
913 | ||
538b4852 | 914 | if (has_io) { |
17ae2780 LP |
915 | CGroupIODeviceLimit *l; |
916 | ||
917 | LIST_FOREACH(device_limits, l, c->io_device_limits) | |
918 | cgroup_apply_io_device_limit(u, l->path, l->limits); | |
538b4852 | 919 | |
538b4852 | 920 | } else if (has_blockio) { |
17ae2780 | 921 | CGroupBlockIODeviceBandwidth *b; |
538b4852 | 922 | |
17ae2780 | 923 | LIST_FOREACH(device_bandwidths, b, c->blockio_device_bandwidths) { |
538b4852 TH |
924 | uint64_t limits[_CGROUP_IO_LIMIT_TYPE_MAX]; |
925 | CGroupIOLimitType type; | |
926 | ||
927 | for (type = 0; type < _CGROUP_IO_LIMIT_TYPE_MAX; type++) | |
928 | limits[type] = cgroup_io_limit_defaults[type]; | |
929 | ||
930 | limits[CGROUP_IO_RBPS_MAX] = b->rbps; | |
931 | limits[CGROUP_IO_WBPS_MAX] = b->wbps; | |
932 | ||
128fadc9 TH |
933 | log_cgroup_compat(u, "Applying BlockIO{Read|Write}Bandwidth %" PRIu64 " %" PRIu64 " as IO{Read|Write}BandwidthMax for %s", |
934 | b->rbps, b->wbps, b->path); | |
935 | ||
17ae2780 | 936 | cgroup_apply_io_device_limit(u, b->path, limits); |
538b4852 | 937 | } |
13c31542 TH |
938 | } |
939 | } | |
940 | ||
906c06f6 | 941 | if (apply_mask & CGROUP_MASK_BLKIO) { |
538b4852 TH |
942 | bool has_io = cgroup_context_has_io_config(c); |
943 | bool has_blockio = cgroup_context_has_blockio_config(c); | |
4ad49000 | 944 | |
01efdf13 | 945 | if (!is_root) { |
64faf04c TH |
946 | char buf[DECIMAL_STR_MAX(uint64_t)+1]; |
947 | uint64_t weight; | |
64faf04c | 948 | |
7d862ab8 | 949 | if (has_io) { |
128fadc9 TH |
950 | uint64_t io_weight = cgroup_context_io_weight(c, state); |
951 | ||
538b4852 | 952 | weight = cgroup_weight_io_to_blkio(cgroup_context_io_weight(c, state)); |
128fadc9 TH |
953 | |
954 | log_cgroup_compat(u, "Applying [Startup]IOWeight %" PRIu64 " as [Startup]BlockIOWeight %" PRIu64, | |
955 | io_weight, weight); | |
7d862ab8 TH |
956 | } else if (has_blockio) |
957 | weight = cgroup_context_blkio_weight(c, state); | |
958 | else | |
538b4852 | 959 | weight = CGROUP_BLKIO_WEIGHT_DEFAULT; |
64faf04c TH |
960 | |
961 | xsprintf(buf, "%" PRIu64 "\n", weight); | |
01efdf13 | 962 | r = cg_set_attribute("blkio", path, "blkio.weight", buf); |
1aeab12b | 963 | if (r < 0) |
f29ff115 TH |
964 | log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, |
965 | "Failed to set blkio.weight: %m"); | |
4ad49000 | 966 | |
7d862ab8 | 967 | if (has_io) { |
538b4852 TH |
968 | CGroupIODeviceWeight *w; |
969 | ||
128fadc9 TH |
970 | LIST_FOREACH(device_weights, w, c->io_device_weights) { |
971 | weight = cgroup_weight_io_to_blkio(w->weight); | |
972 | ||
973 | log_cgroup_compat(u, "Applying IODeviceWeight %" PRIu64 " as BlockIODeviceWeight %" PRIu64 " for %s", | |
974 | w->weight, weight, w->path); | |
975 | ||
976 | cgroup_apply_blkio_device_weight(u, w->path, weight); | |
977 | } | |
7d862ab8 TH |
978 | } else if (has_blockio) { |
979 | CGroupBlockIODeviceWeight *w; | |
980 | ||
7d862ab8 TH |
981 | LIST_FOREACH(device_weights, w, c->blockio_device_weights) |
982 | cgroup_apply_blkio_device_weight(u, w->path, w->weight); | |
538b4852 | 983 | } |
4ad49000 LP |
984 | } |
985 | ||
7d862ab8 | 986 | if (has_io) { |
17ae2780 | 987 | CGroupIODeviceLimit *l; |
538b4852 | 988 | |
17ae2780 | 989 | LIST_FOREACH(device_limits, l, c->io_device_limits) { |
128fadc9 TH |
990 | log_cgroup_compat(u, "Applying IO{Read|Write}Bandwidth %" PRIu64 " %" PRIu64 " as BlockIO{Read|Write}BandwidthMax for %s", |
991 | l->limits[CGROUP_IO_RBPS_MAX], l->limits[CGROUP_IO_WBPS_MAX], l->path); | |
992 | ||
17ae2780 | 993 | cgroup_apply_blkio_device_limit(u, l->path, l->limits[CGROUP_IO_RBPS_MAX], l->limits[CGROUP_IO_WBPS_MAX]); |
538b4852 | 994 | } |
7d862ab8 | 995 | } else if (has_blockio) { |
17ae2780 | 996 | CGroupBlockIODeviceBandwidth *b; |
7d862ab8 | 997 | |
17ae2780 LP |
998 | LIST_FOREACH(device_bandwidths, b, c->blockio_device_bandwidths) |
999 | cgroup_apply_blkio_device_limit(u, b->path, b->rbps, b->wbps); | |
d686d8a9 | 1000 | } |
8e274523 LP |
1001 | } |
1002 | ||
906c06f6 | 1003 | if ((apply_mask & CGROUP_MASK_MEMORY) && !is_root) { |
b4cccbc1 LP |
1004 | if (cg_all_unified() > 0) { |
1005 | uint64_t max, swap_max = CGROUP_LIMIT_MAX; | |
efdb0237 | 1006 | |
96e131ea | 1007 | if (cgroup_context_has_unified_memory_config(c)) { |
da4d897e | 1008 | max = c->memory_max; |
96e131ea WC |
1009 | swap_max = c->memory_swap_max; |
1010 | } else { | |
da4d897e | 1011 | max = c->memory_limit; |
efdb0237 | 1012 | |
128fadc9 TH |
1013 | if (max != CGROUP_LIMIT_MAX) |
1014 | log_cgroup_compat(u, "Applying MemoryLimit %" PRIu64 " as MemoryMax", max); | |
1015 | } | |
1016 | ||
48422635 | 1017 | cgroup_apply_unified_memory_limit(u, "memory.min", c->memory_min); |
f29ff115 TH |
1018 | cgroup_apply_unified_memory_limit(u, "memory.low", c->memory_low); |
1019 | cgroup_apply_unified_memory_limit(u, "memory.high", c->memory_high); | |
1020 | cgroup_apply_unified_memory_limit(u, "memory.max", max); | |
96e131ea | 1021 | cgroup_apply_unified_memory_limit(u, "memory.swap.max", swap_max); |
efdb0237 | 1022 | } else { |
da4d897e | 1023 | char buf[DECIMAL_STR_MAX(uint64_t) + 1]; |
7d862ab8 | 1024 | uint64_t val; |
da4d897e | 1025 | |
7d862ab8 | 1026 | if (cgroup_context_has_unified_memory_config(c)) { |
78a4ee59 | 1027 | val = c->memory_max; |
7d862ab8 TH |
1028 | log_cgroup_compat(u, "Applying MemoryMax %" PRIi64 " as MemoryLimit", val); |
1029 | } else | |
1030 | val = c->memory_limit; | |
128fadc9 | 1031 | |
78a4ee59 DM |
1032 | if (val == CGROUP_LIMIT_MAX) |
1033 | strncpy(buf, "-1\n", sizeof(buf)); | |
1034 | else | |
1035 | xsprintf(buf, "%" PRIu64 "\n", val); | |
1036 | ||
da4d897e TH |
1037 | r = cg_set_attribute("memory", path, "memory.limit_in_bytes", buf); |
1038 | if (r < 0) | |
f29ff115 TH |
1039 | log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, |
1040 | "Failed to set memory.limit_in_bytes: %m"); | |
da4d897e | 1041 | } |
4ad49000 | 1042 | } |
8e274523 | 1043 | |
084c7007 RG |
1044 | if ((apply_mask & (CGROUP_MASK_DEVICES | CGROUP_MASK_BPF_DEVICES)) && !is_root) { |
1045 | _cleanup_(bpf_program_unrefp) BPFProgram *prog = NULL; | |
4ad49000 | 1046 | CGroupDeviceAllow *a; |
8e274523 | 1047 | |
084c7007 RG |
1048 | if (cg_all_unified() > 0) { |
1049 | r = cgroup_init_device_bpf(&prog, c->device_policy, c->device_allow); | |
1050 | if (r < 0) | |
1051 | log_unit_warning_errno(u, r, "Failed to initialize device control bpf program: %m"); | |
1052 | } else { | |
1053 | /* Changing the devices list of a populated cgroup | |
1054 | * might result in EINVAL, hence ignore EINVAL | |
1055 | * here. */ | |
714e2e1d | 1056 | |
084c7007 RG |
1057 | if (c->device_allow || c->device_policy != CGROUP_AUTO) |
1058 | r = cg_set_attribute("devices", path, "devices.deny", "a"); | |
1059 | else | |
1060 | r = cg_set_attribute("devices", path, "devices.allow", "a"); | |
1061 | if (r < 0) | |
1062 | log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EINVAL, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, | |
1063 | "Failed to reset devices.list: %m"); | |
1064 | } | |
fb385181 | 1065 | |
4ad49000 LP |
1066 | if (c->device_policy == CGROUP_CLOSED || |
1067 | (c->device_policy == CGROUP_AUTO && c->device_allow)) { | |
1068 | static const char auto_devices[] = | |
7d711efb LP |
1069 | "/dev/null\0" "rwm\0" |
1070 | "/dev/zero\0" "rwm\0" | |
1071 | "/dev/full\0" "rwm\0" | |
1072 | "/dev/random\0" "rwm\0" | |
1073 | "/dev/urandom\0" "rwm\0" | |
1074 | "/dev/tty\0" "rwm\0" | |
5a7f87a9 | 1075 | "/dev/ptmx\0" "rwm\0" |
0d9e7991 | 1076 | /* Allow /run/systemd/inaccessible/{chr,blk} devices for mapping InaccessiblePaths */ |
e7330dfe DP |
1077 | "-/run/systemd/inaccessible/chr\0" "rwm\0" |
1078 | "-/run/systemd/inaccessible/blk\0" "rwm\0"; | |
4ad49000 LP |
1079 | |
1080 | const char *x, *y; | |
1081 | ||
1082 | NULSTR_FOREACH_PAIR(x, y, auto_devices) | |
913c898c | 1083 | (void) whitelist_device(prog, path, x, y); |
7d711efb | 1084 | |
5a7f87a9 | 1085 | /* PTS (/dev/pts) devices may not be duplicated, but accessed */ |
913c898c | 1086 | (void) whitelist_major(prog, path, "pts", 'c', "rw"); |
4ad49000 LP |
1087 | } |
1088 | ||
1089 | LIST_FOREACH(device_allow, a, c->device_allow) { | |
fb4650aa | 1090 | char acc[4], *val; |
4ad49000 LP |
1091 | unsigned k = 0; |
1092 | ||
1093 | if (a->r) | |
1094 | acc[k++] = 'r'; | |
1095 | if (a->w) | |
1096 | acc[k++] = 'w'; | |
1097 | if (a->m) | |
1098 | acc[k++] = 'm'; | |
fb385181 | 1099 | |
4ad49000 LP |
1100 | if (k == 0) |
1101 | continue; | |
fb385181 | 1102 | |
4ad49000 | 1103 | acc[k++] = 0; |
90060676 | 1104 | |
27458ed6 | 1105 | if (path_startswith(a->path, "/dev/")) |
913c898c | 1106 | (void) whitelist_device(prog, path, a->path, acc); |
fb4650aa | 1107 | else if ((val = startswith(a->path, "block-"))) |
913c898c | 1108 | (void) whitelist_major(prog, path, val, 'b', acc); |
fb4650aa | 1109 | else if ((val = startswith(a->path, "char-"))) |
913c898c | 1110 | (void) whitelist_major(prog, path, val, 'c', acc); |
90060676 | 1111 | else |
f29ff115 | 1112 | log_unit_debug(u, "Ignoring device %s while writing cgroup attribute.", a->path); |
4ad49000 | 1113 | } |
084c7007 RG |
1114 | |
1115 | r = cgroup_apply_device_bpf(u, prog, c->device_policy, c->device_allow); | |
1116 | if (r < 0) { | |
1117 | static bool warned = false; | |
1118 | ||
1119 | log_full_errno(warned ? LOG_DEBUG : LOG_WARNING, r, | |
1120 | "Unit %s configures device ACL, but the local system doesn't seem to support the BPF-based device controller.\n" | |
1121 | "Proceeding WITHOUT applying ACL (all devices will be accessible)!\n" | |
1122 | "(This warning is only shown for the first loaded unit using device ACL.)", u->id); | |
1123 | ||
1124 | warned = true; | |
1125 | } | |
4ad49000 | 1126 | } |
03a7b521 | 1127 | |
00b5974f LP |
1128 | if (apply_mask & CGROUP_MASK_PIDS) { |
1129 | ||
1130 | if (is_root) { | |
1131 | /* So, the "pids" controller does not expose anything on the root cgroup, in order not to | |
1132 | * replicate knobs exposed elsewhere needlessly. We abstract this away here however, and when | |
1133 | * the knobs of the root cgroup are modified propagate this to the relevant sysctls. There's a | |
1134 | * non-obvious asymmetry however: unlike the cgroup properties we don't really want to take | |
1135 | * exclusive ownership of the sysctls, but we still want to honour things if the user sets | |
1136 | * limits. Hence we employ sort of a one-way strategy: when the user sets a bounded limit | |
1137 | * through us it counts. When the user afterwards unsets it again (i.e. sets it to unbounded) | |
1138 | * it also counts. But if the user never set a limit through us (i.e. we are the default of | |
1139 | * "unbounded") we leave things unmodified. For this we manage a global boolean that we turn on | |
1140 | * the first time we set a limit. Note that this boolean is flushed out on manager reload, | |
1141 | * which is desirable so that there's an offical way to release control of the sysctl from | |
1142 | * systemd: set the limit to unbounded and reload. */ | |
1143 | ||
1144 | if (c->tasks_max != CGROUP_LIMIT_MAX) { | |
1145 | u->manager->sysctl_pid_max_changed = true; | |
1146 | r = procfs_tasks_set_limit(c->tasks_max); | |
1147 | } else if (u->manager->sysctl_pid_max_changed) | |
1148 | r = procfs_tasks_set_limit(TASKS_MAX); | |
1149 | else | |
1150 | r = 0; | |
03a7b521 | 1151 | |
00b5974f LP |
1152 | if (r < 0) |
1153 | log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, | |
1154 | "Failed to write to tasks limit sysctls: %m"); | |
03a7b521 | 1155 | |
00b5974f LP |
1156 | } else { |
1157 | if (c->tasks_max != CGROUP_LIMIT_MAX) { | |
1158 | char buf[DECIMAL_STR_MAX(uint64_t) + 2]; | |
03a7b521 | 1159 | |
00b5974f LP |
1160 | sprintf(buf, "%" PRIu64 "\n", c->tasks_max); |
1161 | r = cg_set_attribute("pids", path, "pids.max", buf); | |
1162 | } else | |
1163 | r = cg_set_attribute("pids", path, "pids.max", "max"); | |
1164 | if (r < 0) | |
1165 | log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, | |
1166 | "Failed to set pids.max: %m"); | |
1167 | } | |
03a7b521 | 1168 | } |
906c06f6 | 1169 | |
17f14955 | 1170 | if (apply_mask & CGROUP_MASK_BPF_FIREWALL) |
0f2d84d2 | 1171 | cgroup_apply_firewall(u); |
fb385181 LP |
1172 | } |
1173 | ||
efdb0237 LP |
1174 | CGroupMask cgroup_context_get_mask(CGroupContext *c) { |
1175 | CGroupMask mask = 0; | |
8e274523 | 1176 | |
4ad49000 | 1177 | /* Figure out which controllers we need */ |
8e274523 | 1178 | |
b2f8b02e | 1179 | if (c->cpu_accounting || |
66ebf6c0 TH |
1180 | cgroup_context_has_cpu_weight(c) || |
1181 | cgroup_context_has_cpu_shares(c) || | |
3a43da28 | 1182 | c->cpu_quota_per_sec_usec != USEC_INFINITY) |
efdb0237 | 1183 | mask |= CGROUP_MASK_CPUACCT | CGROUP_MASK_CPU; |
ecedd90f | 1184 | |
538b4852 TH |
1185 | if (cgroup_context_has_io_config(c) || cgroup_context_has_blockio_config(c)) |
1186 | mask |= CGROUP_MASK_IO | CGROUP_MASK_BLKIO; | |
ecedd90f | 1187 | |
4ad49000 | 1188 | if (c->memory_accounting || |
da4d897e TH |
1189 | c->memory_limit != CGROUP_LIMIT_MAX || |
1190 | cgroup_context_has_unified_memory_config(c)) | |
efdb0237 | 1191 | mask |= CGROUP_MASK_MEMORY; |
8e274523 | 1192 | |
a931ad47 LP |
1193 | if (c->device_allow || |
1194 | c->device_policy != CGROUP_AUTO) | |
084c7007 | 1195 | mask |= CGROUP_MASK_DEVICES | CGROUP_MASK_BPF_DEVICES; |
4ad49000 | 1196 | |
03a7b521 | 1197 | if (c->tasks_accounting || |
8793fa25 | 1198 | c->tasks_max != CGROUP_LIMIT_MAX) |
03a7b521 LP |
1199 | mask |= CGROUP_MASK_PIDS; |
1200 | ||
4ad49000 | 1201 | return mask; |
8e274523 LP |
1202 | } |
1203 | ||
17f14955 RG |
1204 | CGroupMask unit_get_bpf_mask(Unit *u) { |
1205 | CGroupMask mask = 0; | |
1206 | ||
1207 | if (unit_get_needs_bpf_firewall(u)) | |
1208 | mask |= CGROUP_MASK_BPF_FIREWALL; | |
1209 | ||
1210 | return mask; | |
1211 | } | |
1212 | ||
efdb0237 | 1213 | CGroupMask unit_get_own_mask(Unit *u) { |
4ad49000 | 1214 | CGroupContext *c; |
8e274523 | 1215 | |
efdb0237 LP |
1216 | /* Returns the mask of controllers the unit needs for itself */ |
1217 | ||
4ad49000 LP |
1218 | c = unit_get_cgroup_context(u); |
1219 | if (!c) | |
1220 | return 0; | |
8e274523 | 1221 | |
17f14955 | 1222 | return cgroup_context_get_mask(c) | unit_get_bpf_mask(u) | unit_get_delegate_mask(u); |
02638280 LP |
1223 | } |
1224 | ||
1225 | CGroupMask unit_get_delegate_mask(Unit *u) { | |
1226 | CGroupContext *c; | |
1227 | ||
1228 | /* If delegation is turned on, then turn on selected controllers, unless we are on the legacy hierarchy and the | |
1229 | * process we fork into is known to drop privileges, and hence shouldn't get access to the controllers. | |
19af675e | 1230 | * |
02638280 | 1231 | * Note that on the unified hierarchy it is safe to delegate controllers to unprivileged services. */ |
a931ad47 | 1232 | |
1d9cc876 | 1233 | if (!unit_cgroup_delegate(u)) |
02638280 LP |
1234 | return 0; |
1235 | ||
1236 | if (cg_all_unified() <= 0) { | |
a931ad47 LP |
1237 | ExecContext *e; |
1238 | ||
1239 | e = unit_get_exec_context(u); | |
02638280 LP |
1240 | if (e && !exec_context_maintains_privileges(e)) |
1241 | return 0; | |
a931ad47 LP |
1242 | } |
1243 | ||
1d9cc876 | 1244 | assert_se(c = unit_get_cgroup_context(u)); |
02638280 | 1245 | return c->delegate_controllers; |
8e274523 LP |
1246 | } |
1247 | ||
efdb0237 | 1248 | CGroupMask unit_get_members_mask(Unit *u) { |
4ad49000 | 1249 | assert(u); |
bc432dc7 | 1250 | |
02638280 | 1251 | /* Returns the mask of controllers all of the unit's children require, merged */ |
efdb0237 | 1252 | |
bc432dc7 LP |
1253 | if (u->cgroup_members_mask_valid) |
1254 | return u->cgroup_members_mask; | |
1255 | ||
64e844e5 | 1256 | u->cgroup_members_mask = 0; |
bc432dc7 LP |
1257 | |
1258 | if (u->type == UNIT_SLICE) { | |
eef85c4a | 1259 | void *v; |
bc432dc7 LP |
1260 | Unit *member; |
1261 | Iterator i; | |
1262 | ||
eef85c4a | 1263 | HASHMAP_FOREACH_KEY(v, member, u->dependencies[UNIT_BEFORE], i) { |
bc432dc7 LP |
1264 | |
1265 | if (member == u) | |
1266 | continue; | |
1267 | ||
d4fdc205 | 1268 | if (UNIT_DEREF(member->slice) != u) |
bc432dc7 LP |
1269 | continue; |
1270 | ||
31604970 | 1271 | u->cgroup_members_mask |= unit_get_subtree_mask(member); /* note that this calls ourselves again, for the children */ |
bc432dc7 LP |
1272 | } |
1273 | } | |
1274 | ||
1275 | u->cgroup_members_mask_valid = true; | |
6414b7c9 | 1276 | return u->cgroup_members_mask; |
246aa6dd LP |
1277 | } |
1278 | ||
efdb0237 | 1279 | CGroupMask unit_get_siblings_mask(Unit *u) { |
4ad49000 | 1280 | assert(u); |
246aa6dd | 1281 | |
efdb0237 LP |
1282 | /* Returns the mask of controllers all of the unit's siblings |
1283 | * require, i.e. the members mask of the unit's parent slice | |
1284 | * if there is one. */ | |
1285 | ||
bc432dc7 | 1286 | if (UNIT_ISSET(u->slice)) |
637f421e | 1287 | return unit_get_members_mask(UNIT_DEREF(u->slice)); |
4ad49000 | 1288 | |
64e844e5 | 1289 | return unit_get_subtree_mask(u); /* we are the top-level slice */ |
246aa6dd LP |
1290 | } |
1291 | ||
efdb0237 LP |
1292 | CGroupMask unit_get_subtree_mask(Unit *u) { |
1293 | ||
1294 | /* Returns the mask of this subtree, meaning of the group | |
1295 | * itself and its children. */ | |
1296 | ||
1297 | return unit_get_own_mask(u) | unit_get_members_mask(u); | |
1298 | } | |
1299 | ||
1300 | CGroupMask unit_get_target_mask(Unit *u) { | |
1301 | CGroupMask mask; | |
1302 | ||
1303 | /* This returns the cgroup mask of all controllers to enable | |
1304 | * for a specific cgroup, i.e. everything it needs itself, | |
1305 | * plus all that its children need, plus all that its siblings | |
1306 | * need. This is primarily useful on the legacy cgroup | |
1307 | * hierarchy, where we need to duplicate each cgroup in each | |
1308 | * hierarchy that shall be enabled for it. */ | |
6414b7c9 | 1309 | |
efdb0237 LP |
1310 | mask = unit_get_own_mask(u) | unit_get_members_mask(u) | unit_get_siblings_mask(u); |
1311 | mask &= u->manager->cgroup_supported; | |
1312 | ||
1313 | return mask; | |
1314 | } | |
1315 | ||
1316 | CGroupMask unit_get_enable_mask(Unit *u) { | |
1317 | CGroupMask mask; | |
1318 | ||
1319 | /* This returns the cgroup mask of all controllers to enable | |
1320 | * for the children of a specific cgroup. This is primarily | |
1321 | * useful for the unified cgroup hierarchy, where each cgroup | |
1322 | * controls which controllers are enabled for its children. */ | |
1323 | ||
1324 | mask = unit_get_members_mask(u); | |
6414b7c9 DS |
1325 | mask &= u->manager->cgroup_supported; |
1326 | ||
1327 | return mask; | |
1328 | } | |
1329 | ||
17f14955 | 1330 | bool unit_get_needs_bpf_firewall(Unit *u) { |
906c06f6 DM |
1331 | CGroupContext *c; |
1332 | Unit *p; | |
1333 | assert(u); | |
1334 | ||
906c06f6 DM |
1335 | c = unit_get_cgroup_context(u); |
1336 | if (!c) | |
1337 | return false; | |
1338 | ||
1339 | if (c->ip_accounting || | |
1340 | c->ip_address_allow || | |
1341 | c->ip_address_deny) | |
1342 | return true; | |
1343 | ||
1344 | /* If any parent slice has an IP access list defined, it applies too */ | |
1345 | for (p = UNIT_DEREF(u->slice); p; p = UNIT_DEREF(p->slice)) { | |
1346 | c = unit_get_cgroup_context(p); | |
1347 | if (!c) | |
1348 | return false; | |
1349 | ||
1350 | if (c->ip_address_allow || | |
1351 | c->ip_address_deny) | |
1352 | return true; | |
1353 | } | |
1354 | ||
1355 | return false; | |
1356 | } | |
1357 | ||
6414b7c9 DS |
1358 | /* Recurse from a unit up through its containing slices, propagating |
1359 | * mask bits upward. A unit is also member of itself. */ | |
bc432dc7 | 1360 | void unit_update_cgroup_members_masks(Unit *u) { |
efdb0237 | 1361 | CGroupMask m; |
bc432dc7 LP |
1362 | bool more; |
1363 | ||
1364 | assert(u); | |
1365 | ||
1366 | /* Calculate subtree mask */ | |
efdb0237 | 1367 | m = unit_get_subtree_mask(u); |
bc432dc7 LP |
1368 | |
1369 | /* See if anything changed from the previous invocation. If | |
1370 | * not, we're done. */ | |
1371 | if (u->cgroup_subtree_mask_valid && m == u->cgroup_subtree_mask) | |
1372 | return; | |
1373 | ||
1374 | more = | |
1375 | u->cgroup_subtree_mask_valid && | |
1376 | ((m & ~u->cgroup_subtree_mask) != 0) && | |
1377 | ((~m & u->cgroup_subtree_mask) == 0); | |
1378 | ||
1379 | u->cgroup_subtree_mask = m; | |
1380 | u->cgroup_subtree_mask_valid = true; | |
1381 | ||
6414b7c9 DS |
1382 | if (UNIT_ISSET(u->slice)) { |
1383 | Unit *s = UNIT_DEREF(u->slice); | |
bc432dc7 LP |
1384 | |
1385 | if (more) | |
1386 | /* There's more set now than before. We | |
1387 | * propagate the new mask to the parent's mask | |
1388 | * (not caring if it actually was valid or | |
1389 | * not). */ | |
1390 | ||
1391 | s->cgroup_members_mask |= m; | |
1392 | ||
1393 | else | |
1394 | /* There's less set now than before (or we | |
1395 | * don't know), we need to recalculate | |
1396 | * everything, so let's invalidate the | |
1397 | * parent's members mask */ | |
1398 | ||
1399 | s->cgroup_members_mask_valid = false; | |
1400 | ||
1401 | /* And now make sure that this change also hits our | |
1402 | * grandparents */ | |
1403 | unit_update_cgroup_members_masks(s); | |
6414b7c9 DS |
1404 | } |
1405 | } | |
1406 | ||
6592b975 | 1407 | const char *unit_get_realized_cgroup_path(Unit *u, CGroupMask mask) { |
03b90d4b | 1408 | |
6592b975 | 1409 | /* Returns the realized cgroup path of the specified unit where all specified controllers are available. */ |
03b90d4b LP |
1410 | |
1411 | while (u) { | |
6592b975 | 1412 | |
03b90d4b LP |
1413 | if (u->cgroup_path && |
1414 | u->cgroup_realized && | |
d94a24ca | 1415 | FLAGS_SET(u->cgroup_realized_mask, mask)) |
03b90d4b LP |
1416 | return u->cgroup_path; |
1417 | ||
1418 | u = UNIT_DEREF(u->slice); | |
1419 | } | |
1420 | ||
1421 | return NULL; | |
1422 | } | |
1423 | ||
6592b975 LP |
1424 | static const char *migrate_callback(CGroupMask mask, void *userdata) { |
1425 | return unit_get_realized_cgroup_path(userdata, mask); | |
1426 | } | |
1427 | ||
efdb0237 LP |
1428 | char *unit_default_cgroup_path(Unit *u) { |
1429 | _cleanup_free_ char *escaped = NULL, *slice = NULL; | |
1430 | int r; | |
1431 | ||
1432 | assert(u); | |
1433 | ||
1434 | if (unit_has_name(u, SPECIAL_ROOT_SLICE)) | |
1435 | return strdup(u->manager->cgroup_root); | |
1436 | ||
1437 | if (UNIT_ISSET(u->slice) && !unit_has_name(UNIT_DEREF(u->slice), SPECIAL_ROOT_SLICE)) { | |
1438 | r = cg_slice_to_path(UNIT_DEREF(u->slice)->id, &slice); | |
1439 | if (r < 0) | |
1440 | return NULL; | |
1441 | } | |
1442 | ||
1443 | escaped = cg_escape(u->id); | |
1444 | if (!escaped) | |
1445 | return NULL; | |
1446 | ||
1447 | if (slice) | |
605405c6 ZJS |
1448 | return strjoin(u->manager->cgroup_root, "/", slice, "/", |
1449 | escaped); | |
efdb0237 | 1450 | else |
605405c6 | 1451 | return strjoin(u->manager->cgroup_root, "/", escaped); |
efdb0237 LP |
1452 | } |
1453 | ||
1454 | int unit_set_cgroup_path(Unit *u, const char *path) { | |
1455 | _cleanup_free_ char *p = NULL; | |
1456 | int r; | |
1457 | ||
1458 | assert(u); | |
1459 | ||
1460 | if (path) { | |
1461 | p = strdup(path); | |
1462 | if (!p) | |
1463 | return -ENOMEM; | |
1464 | } else | |
1465 | p = NULL; | |
1466 | ||
1467 | if (streq_ptr(u->cgroup_path, p)) | |
1468 | return 0; | |
1469 | ||
1470 | if (p) { | |
1471 | r = hashmap_put(u->manager->cgroup_unit, p, u); | |
1472 | if (r < 0) | |
1473 | return r; | |
1474 | } | |
1475 | ||
1476 | unit_release_cgroup(u); | |
1477 | ||
ae2a15bc | 1478 | u->cgroup_path = TAKE_PTR(p); |
efdb0237 LP |
1479 | |
1480 | return 1; | |
1481 | } | |
1482 | ||
1483 | int unit_watch_cgroup(Unit *u) { | |
ab2c3861 | 1484 | _cleanup_free_ char *events = NULL; |
efdb0237 LP |
1485 | int r; |
1486 | ||
1487 | assert(u); | |
1488 | ||
1489 | if (!u->cgroup_path) | |
1490 | return 0; | |
1491 | ||
1492 | if (u->cgroup_inotify_wd >= 0) | |
1493 | return 0; | |
1494 | ||
1495 | /* Only applies to the unified hierarchy */ | |
c22800e4 | 1496 | r = cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER); |
b4cccbc1 LP |
1497 | if (r < 0) |
1498 | return log_error_errno(r, "Failed to determine whether the name=systemd hierarchy is unified: %m"); | |
1499 | if (r == 0) | |
efdb0237 LP |
1500 | return 0; |
1501 | ||
1502 | /* Don't watch the root slice, it's pointless. */ | |
1503 | if (unit_has_name(u, SPECIAL_ROOT_SLICE)) | |
1504 | return 0; | |
1505 | ||
1506 | r = hashmap_ensure_allocated(&u->manager->cgroup_inotify_wd_unit, &trivial_hash_ops); | |
1507 | if (r < 0) | |
1508 | return log_oom(); | |
1509 | ||
ab2c3861 | 1510 | r = cg_get_path(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path, "cgroup.events", &events); |
efdb0237 LP |
1511 | if (r < 0) |
1512 | return log_oom(); | |
1513 | ||
ab2c3861 | 1514 | u->cgroup_inotify_wd = inotify_add_watch(u->manager->cgroup_inotify_fd, events, IN_MODIFY); |
efdb0237 LP |
1515 | if (u->cgroup_inotify_wd < 0) { |
1516 | ||
1517 | if (errno == ENOENT) /* If the directory is already | |
1518 | * gone we don't need to track | |
1519 | * it, so this is not an error */ | |
1520 | return 0; | |
1521 | ||
1522 | return log_unit_error_errno(u, errno, "Failed to add inotify watch descriptor for control group %s: %m", u->cgroup_path); | |
1523 | } | |
1524 | ||
1525 | r = hashmap_put(u->manager->cgroup_inotify_wd_unit, INT_TO_PTR(u->cgroup_inotify_wd), u); | |
1526 | if (r < 0) | |
1527 | return log_unit_error_errno(u, r, "Failed to add inotify watch descriptor to hash map: %m"); | |
1528 | ||
1529 | return 0; | |
1530 | } | |
1531 | ||
a4634b21 LP |
1532 | int unit_pick_cgroup_path(Unit *u) { |
1533 | _cleanup_free_ char *path = NULL; | |
1534 | int r; | |
1535 | ||
1536 | assert(u); | |
1537 | ||
1538 | if (u->cgroup_path) | |
1539 | return 0; | |
1540 | ||
1541 | if (!UNIT_HAS_CGROUP_CONTEXT(u)) | |
1542 | return -EINVAL; | |
1543 | ||
1544 | path = unit_default_cgroup_path(u); | |
1545 | if (!path) | |
1546 | return log_oom(); | |
1547 | ||
1548 | r = unit_set_cgroup_path(u, path); | |
1549 | if (r == -EEXIST) | |
1550 | return log_unit_error_errno(u, r, "Control group %s exists already.", path); | |
1551 | if (r < 0) | |
1552 | return log_unit_error_errno(u, r, "Failed to set unit's control group path to %s: %m", path); | |
1553 | ||
1554 | return 0; | |
1555 | } | |
1556 | ||
efdb0237 LP |
1557 | static int unit_create_cgroup( |
1558 | Unit *u, | |
1559 | CGroupMask target_mask, | |
17f14955 | 1560 | CGroupMask enable_mask) { |
efdb0237 | 1561 | |
0cd385d3 | 1562 | CGroupContext *c; |
bc432dc7 | 1563 | int r; |
65be7e06 | 1564 | bool created; |
64747e2d | 1565 | |
4ad49000 | 1566 | assert(u); |
64747e2d | 1567 | |
0cd385d3 LP |
1568 | c = unit_get_cgroup_context(u); |
1569 | if (!c) | |
1570 | return 0; | |
1571 | ||
a4634b21 LP |
1572 | /* Figure out our cgroup path */ |
1573 | r = unit_pick_cgroup_path(u); | |
1574 | if (r < 0) | |
1575 | return r; | |
b58b8e11 | 1576 | |
03b90d4b | 1577 | /* First, create our own group */ |
efdb0237 | 1578 | r = cg_create_everywhere(u->manager->cgroup_supported, target_mask, u->cgroup_path); |
23bbb0de | 1579 | if (r < 0) |
efdb0237 | 1580 | return log_unit_error_errno(u, r, "Failed to create cgroup %s: %m", u->cgroup_path); |
65be7e06 | 1581 | created = !!r; |
efdb0237 LP |
1582 | |
1583 | /* Start watching it */ | |
1584 | (void) unit_watch_cgroup(u); | |
1585 | ||
65be7e06 ZJS |
1586 | /* Preserve enabled controllers in delegated units, adjust others. */ |
1587 | if (created || !unit_cgroup_delegate(u)) { | |
1588 | ||
1589 | /* Enable all controllers we need */ | |
1590 | r = cg_enable_everywhere(u->manager->cgroup_supported, enable_mask, u->cgroup_path); | |
1591 | if (r < 0) | |
1592 | log_unit_warning_errno(u, r, "Failed to enable controllers on cgroup %s, ignoring: %m", | |
1593 | u->cgroup_path); | |
1594 | } | |
03b90d4b LP |
1595 | |
1596 | /* Keep track that this is now realized */ | |
4ad49000 | 1597 | u->cgroup_realized = true; |
efdb0237 | 1598 | u->cgroup_realized_mask = target_mask; |
ccf78df1 | 1599 | u->cgroup_enabled_mask = enable_mask; |
4ad49000 | 1600 | |
1d9cc876 | 1601 | if (u->type != UNIT_SLICE && !unit_cgroup_delegate(u)) { |
0cd385d3 LP |
1602 | |
1603 | /* Then, possibly move things over, but not if | |
1604 | * subgroups may contain processes, which is the case | |
1605 | * for slice and delegation units. */ | |
1606 | r = cg_migrate_everywhere(u->manager->cgroup_supported, u->cgroup_path, u->cgroup_path, migrate_callback, u); | |
1607 | if (r < 0) | |
efdb0237 | 1608 | log_unit_warning_errno(u, r, "Failed to migrate cgroup from to %s, ignoring: %m", u->cgroup_path); |
0cd385d3 | 1609 | } |
03b90d4b | 1610 | |
64747e2d LP |
1611 | return 0; |
1612 | } | |
1613 | ||
6592b975 LP |
1614 | static int unit_attach_pid_to_cgroup_via_bus(Unit *u, pid_t pid, const char *suffix_path) { |
1615 | _cleanup_(sd_bus_error_free) sd_bus_error error = SD_BUS_ERROR_NULL; | |
1616 | char *pp; | |
7b3fd631 | 1617 | int r; |
6592b975 | 1618 | |
7b3fd631 LP |
1619 | assert(u); |
1620 | ||
6592b975 LP |
1621 | if (MANAGER_IS_SYSTEM(u->manager)) |
1622 | return -EINVAL; | |
1623 | ||
1624 | if (!u->manager->system_bus) | |
1625 | return -EIO; | |
1626 | ||
1627 | if (!u->cgroup_path) | |
1628 | return -EINVAL; | |
1629 | ||
1630 | /* Determine this unit's cgroup path relative to our cgroup root */ | |
1631 | pp = path_startswith(u->cgroup_path, u->manager->cgroup_root); | |
1632 | if (!pp) | |
1633 | return -EINVAL; | |
1634 | ||
1635 | pp = strjoina("/", pp, suffix_path); | |
858d36c1 | 1636 | path_simplify(pp, false); |
6592b975 LP |
1637 | |
1638 | r = sd_bus_call_method(u->manager->system_bus, | |
1639 | "org.freedesktop.systemd1", | |
1640 | "/org/freedesktop/systemd1", | |
1641 | "org.freedesktop.systemd1.Manager", | |
1642 | "AttachProcessesToUnit", | |
1643 | &error, NULL, | |
1644 | "ssau", | |
1645 | NULL /* empty unit name means client's unit, i.e. us */, pp, 1, (uint32_t) pid); | |
7b3fd631 | 1646 | if (r < 0) |
6592b975 LP |
1647 | return log_unit_debug_errno(u, r, "Failed to attach unit process " PID_FMT " via the bus: %s", pid, bus_error_message(&error, r)); |
1648 | ||
1649 | return 0; | |
1650 | } | |
1651 | ||
1652 | int unit_attach_pids_to_cgroup(Unit *u, Set *pids, const char *suffix_path) { | |
1653 | CGroupMask delegated_mask; | |
1654 | const char *p; | |
1655 | Iterator i; | |
1656 | void *pidp; | |
1657 | int r, q; | |
1658 | ||
1659 | assert(u); | |
1660 | ||
1661 | if (!UNIT_HAS_CGROUP_CONTEXT(u)) | |
1662 | return -EINVAL; | |
1663 | ||
1664 | if (set_isempty(pids)) | |
1665 | return 0; | |
7b3fd631 | 1666 | |
6592b975 | 1667 | r = unit_realize_cgroup(u); |
7b3fd631 LP |
1668 | if (r < 0) |
1669 | return r; | |
1670 | ||
6592b975 LP |
1671 | if (isempty(suffix_path)) |
1672 | p = u->cgroup_path; | |
1673 | else | |
1674 | p = strjoina(u->cgroup_path, "/", suffix_path); | |
1675 | ||
1676 | delegated_mask = unit_get_delegate_mask(u); | |
1677 | ||
1678 | r = 0; | |
1679 | SET_FOREACH(pidp, pids, i) { | |
1680 | pid_t pid = PTR_TO_PID(pidp); | |
1681 | CGroupController c; | |
1682 | ||
1683 | /* First, attach the PID to the main cgroup hierarchy */ | |
1684 | q = cg_attach(SYSTEMD_CGROUP_CONTROLLER, p, pid); | |
1685 | if (q < 0) { | |
1686 | log_unit_debug_errno(u, q, "Couldn't move process " PID_FMT " to requested cgroup '%s': %m", pid, p); | |
1687 | ||
1688 | if (MANAGER_IS_USER(u->manager) && IN_SET(q, -EPERM, -EACCES)) { | |
1689 | int z; | |
1690 | ||
1691 | /* If we are in a user instance, and we can't move the process ourselves due to | |
1692 | * permission problems, let's ask the system instance about it instead. Since it's more | |
1693 | * privileged it might be able to move the process across the leaves of a subtree who's | |
1694 | * top node is not owned by us. */ | |
1695 | ||
1696 | z = unit_attach_pid_to_cgroup_via_bus(u, pid, suffix_path); | |
1697 | if (z < 0) | |
1698 | log_unit_debug_errno(u, z, "Couldn't move process " PID_FMT " to requested cgroup '%s' via the system bus either: %m", pid, p); | |
1699 | else | |
1700 | continue; /* When the bus thing worked via the bus we are fully done for this PID. */ | |
1701 | } | |
1702 | ||
1703 | if (r >= 0) | |
1704 | r = q; /* Remember first error */ | |
1705 | ||
1706 | continue; | |
1707 | } | |
1708 | ||
1709 | q = cg_all_unified(); | |
1710 | if (q < 0) | |
1711 | return q; | |
1712 | if (q > 0) | |
1713 | continue; | |
1714 | ||
1715 | /* In the legacy hierarchy, attach the process to the request cgroup if possible, and if not to the | |
1716 | * innermost realized one */ | |
1717 | ||
1718 | for (c = 0; c < _CGROUP_CONTROLLER_MAX; c++) { | |
1719 | CGroupMask bit = CGROUP_CONTROLLER_TO_MASK(c); | |
1720 | const char *realized; | |
1721 | ||
1722 | if (!(u->manager->cgroup_supported & bit)) | |
1723 | continue; | |
1724 | ||
1725 | /* If this controller is delegated and realized, honour the caller's request for the cgroup suffix. */ | |
1726 | if (delegated_mask & u->cgroup_realized_mask & bit) { | |
1727 | q = cg_attach(cgroup_controller_to_string(c), p, pid); | |
1728 | if (q >= 0) | |
1729 | continue; /* Success! */ | |
1730 | ||
1731 | 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", | |
1732 | pid, p, cgroup_controller_to_string(c)); | |
1733 | } | |
1734 | ||
1735 | /* So this controller is either not delegate or realized, or something else weird happened. In | |
1736 | * that case let's attach the PID at least to the closest cgroup up the tree that is | |
1737 | * realized. */ | |
1738 | realized = unit_get_realized_cgroup_path(u, bit); | |
1739 | if (!realized) | |
1740 | continue; /* Not even realized in the root slice? Then let's not bother */ | |
1741 | ||
1742 | q = cg_attach(cgroup_controller_to_string(c), realized, pid); | |
1743 | if (q < 0) | |
1744 | log_unit_debug_errno(u, q, "Failed to attach PID " PID_FMT " to realized cgroup %s in controller %s, ignoring: %m", | |
1745 | pid, realized, cgroup_controller_to_string(c)); | |
1746 | } | |
1747 | } | |
1748 | ||
1749 | return r; | |
7b3fd631 LP |
1750 | } |
1751 | ||
4b58153d LP |
1752 | static void cgroup_xattr_apply(Unit *u) { |
1753 | char ids[SD_ID128_STRING_MAX]; | |
1754 | int r; | |
1755 | ||
1756 | assert(u); | |
1757 | ||
1758 | if (!MANAGER_IS_SYSTEM(u->manager)) | |
1759 | return; | |
1760 | ||
1761 | if (sd_id128_is_null(u->invocation_id)) | |
1762 | return; | |
1763 | ||
1764 | r = cg_set_xattr(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path, | |
1765 | "trusted.invocation_id", | |
1766 | sd_id128_to_string(u->invocation_id, ids), 32, | |
1767 | 0); | |
1768 | if (r < 0) | |
0fb84499 | 1769 | log_unit_debug_errno(u, r, "Failed to set invocation ID on control group %s, ignoring: %m", u->cgroup_path); |
4b58153d LP |
1770 | } |
1771 | ||
906c06f6 DM |
1772 | static bool unit_has_mask_realized( |
1773 | Unit *u, | |
1774 | CGroupMask target_mask, | |
17f14955 | 1775 | CGroupMask enable_mask) { |
906c06f6 | 1776 | |
bc432dc7 LP |
1777 | assert(u); |
1778 | ||
906c06f6 DM |
1779 | return u->cgroup_realized && |
1780 | u->cgroup_realized_mask == target_mask && | |
1781 | u->cgroup_enabled_mask == enable_mask && | |
17f14955 | 1782 | u->cgroup_invalidated_mask == 0; |
6414b7c9 DS |
1783 | } |
1784 | ||
2aa57a65 LP |
1785 | static void unit_add_to_cgroup_realize_queue(Unit *u) { |
1786 | assert(u); | |
1787 | ||
1788 | if (u->in_cgroup_realize_queue) | |
1789 | return; | |
1790 | ||
1791 | LIST_PREPEND(cgroup_realize_queue, u->manager->cgroup_realize_queue, u); | |
1792 | u->in_cgroup_realize_queue = true; | |
1793 | } | |
1794 | ||
1795 | static void unit_remove_from_cgroup_realize_queue(Unit *u) { | |
1796 | assert(u); | |
1797 | ||
1798 | if (!u->in_cgroup_realize_queue) | |
1799 | return; | |
1800 | ||
1801 | LIST_REMOVE(cgroup_realize_queue, u->manager->cgroup_realize_queue, u); | |
1802 | u->in_cgroup_realize_queue = false; | |
1803 | } | |
1804 | ||
6414b7c9 DS |
1805 | /* Check if necessary controllers and attributes for a unit are in place. |
1806 | * | |
1807 | * If so, do nothing. | |
1808 | * If not, create paths, move processes over, and set attributes. | |
1809 | * | |
1810 | * Returns 0 on success and < 0 on failure. */ | |
db785129 | 1811 | static int unit_realize_cgroup_now(Unit *u, ManagerState state) { |
efdb0237 | 1812 | CGroupMask target_mask, enable_mask; |
6414b7c9 | 1813 | int r; |
64747e2d | 1814 | |
4ad49000 | 1815 | assert(u); |
64747e2d | 1816 | |
2aa57a65 | 1817 | unit_remove_from_cgroup_realize_queue(u); |
64747e2d | 1818 | |
efdb0237 | 1819 | target_mask = unit_get_target_mask(u); |
ccf78df1 TH |
1820 | enable_mask = unit_get_enable_mask(u); |
1821 | ||
17f14955 | 1822 | if (unit_has_mask_realized(u, target_mask, enable_mask)) |
0a1eb06d | 1823 | return 0; |
64747e2d | 1824 | |
4ad49000 | 1825 | /* First, realize parents */ |
6414b7c9 | 1826 | if (UNIT_ISSET(u->slice)) { |
db785129 | 1827 | r = unit_realize_cgroup_now(UNIT_DEREF(u->slice), state); |
6414b7c9 DS |
1828 | if (r < 0) |
1829 | return r; | |
1830 | } | |
4ad49000 LP |
1831 | |
1832 | /* And then do the real work */ | |
17f14955 | 1833 | r = unit_create_cgroup(u, target_mask, enable_mask); |
6414b7c9 DS |
1834 | if (r < 0) |
1835 | return r; | |
1836 | ||
1837 | /* Finally, apply the necessary attributes. */ | |
17f14955 | 1838 | cgroup_context_apply(u, target_mask, state); |
4b58153d | 1839 | cgroup_xattr_apply(u); |
6414b7c9 DS |
1840 | |
1841 | return 0; | |
64747e2d LP |
1842 | } |
1843 | ||
91a6073e | 1844 | unsigned manager_dispatch_cgroup_realize_queue(Manager *m) { |
db785129 | 1845 | ManagerState state; |
4ad49000 | 1846 | unsigned n = 0; |
db785129 | 1847 | Unit *i; |
6414b7c9 | 1848 | int r; |
ecedd90f | 1849 | |
91a6073e LP |
1850 | assert(m); |
1851 | ||
db785129 LP |
1852 | state = manager_state(m); |
1853 | ||
91a6073e LP |
1854 | while ((i = m->cgroup_realize_queue)) { |
1855 | assert(i->in_cgroup_realize_queue); | |
ecedd90f | 1856 | |
2aa57a65 LP |
1857 | if (UNIT_IS_INACTIVE_OR_FAILED(unit_active_state(i))) { |
1858 | /* Maybe things changed, and the unit is not actually active anymore? */ | |
1859 | unit_remove_from_cgroup_realize_queue(i); | |
1860 | continue; | |
1861 | } | |
1862 | ||
db785129 | 1863 | r = unit_realize_cgroup_now(i, state); |
6414b7c9 | 1864 | if (r < 0) |
efdb0237 | 1865 | log_warning_errno(r, "Failed to realize cgroups for queued unit %s, ignoring: %m", i->id); |
0a1eb06d | 1866 | |
4ad49000 LP |
1867 | n++; |
1868 | } | |
ecedd90f | 1869 | |
4ad49000 | 1870 | return n; |
8e274523 LP |
1871 | } |
1872 | ||
91a6073e | 1873 | static void unit_add_siblings_to_cgroup_realize_queue(Unit *u) { |
4ad49000 | 1874 | Unit *slice; |
ca949c9d | 1875 | |
4ad49000 LP |
1876 | /* This adds the siblings of the specified unit and the |
1877 | * siblings of all parent units to the cgroup queue. (But | |
1878 | * neither the specified unit itself nor the parents.) */ | |
1879 | ||
1880 | while ((slice = UNIT_DEREF(u->slice))) { | |
1881 | Iterator i; | |
1882 | Unit *m; | |
eef85c4a | 1883 | void *v; |
8f53a7b8 | 1884 | |
eef85c4a | 1885 | HASHMAP_FOREACH_KEY(v, m, u->dependencies[UNIT_BEFORE], i) { |
4ad49000 LP |
1886 | if (m == u) |
1887 | continue; | |
8e274523 | 1888 | |
6414b7c9 DS |
1889 | /* Skip units that have a dependency on the slice |
1890 | * but aren't actually in it. */ | |
4ad49000 | 1891 | if (UNIT_DEREF(m->slice) != slice) |
50159e6a | 1892 | continue; |
8e274523 | 1893 | |
6414b7c9 DS |
1894 | /* No point in doing cgroup application for units |
1895 | * without active processes. */ | |
1896 | if (UNIT_IS_INACTIVE_OR_FAILED(unit_active_state(m))) | |
1897 | continue; | |
1898 | ||
1899 | /* If the unit doesn't need any new controllers | |
1900 | * and has current ones realized, it doesn't need | |
1901 | * any changes. */ | |
906c06f6 DM |
1902 | if (unit_has_mask_realized(m, |
1903 | unit_get_target_mask(m), | |
17f14955 | 1904 | unit_get_enable_mask(m))) |
6414b7c9 DS |
1905 | continue; |
1906 | ||
91a6073e | 1907 | unit_add_to_cgroup_realize_queue(m); |
50159e6a LP |
1908 | } |
1909 | ||
4ad49000 | 1910 | u = slice; |
8e274523 | 1911 | } |
4ad49000 LP |
1912 | } |
1913 | ||
0a1eb06d | 1914 | int unit_realize_cgroup(Unit *u) { |
4ad49000 LP |
1915 | assert(u); |
1916 | ||
35b7ff80 | 1917 | if (!UNIT_HAS_CGROUP_CONTEXT(u)) |
0a1eb06d | 1918 | return 0; |
8e274523 | 1919 | |
4ad49000 LP |
1920 | /* So, here's the deal: when realizing the cgroups for this |
1921 | * unit, we need to first create all parents, but there's more | |
1922 | * actually: for the weight-based controllers we also need to | |
1923 | * make sure that all our siblings (i.e. units that are in the | |
73e231ab | 1924 | * same slice as we are) have cgroups, too. Otherwise, things |
4ad49000 LP |
1925 | * would become very uneven as each of their processes would |
1926 | * get as much resources as all our group together. This call | |
1927 | * will synchronously create the parent cgroups, but will | |
1928 | * defer work on the siblings to the next event loop | |
1929 | * iteration. */ | |
ca949c9d | 1930 | |
4ad49000 | 1931 | /* Add all sibling slices to the cgroup queue. */ |
91a6073e | 1932 | unit_add_siblings_to_cgroup_realize_queue(u); |
4ad49000 | 1933 | |
6414b7c9 | 1934 | /* And realize this one now (and apply the values) */ |
db785129 | 1935 | return unit_realize_cgroup_now(u, manager_state(u->manager)); |
8e274523 LP |
1936 | } |
1937 | ||
efdb0237 LP |
1938 | void unit_release_cgroup(Unit *u) { |
1939 | assert(u); | |
1940 | ||
1941 | /* Forgets all cgroup details for this cgroup */ | |
1942 | ||
1943 | if (u->cgroup_path) { | |
1944 | (void) hashmap_remove(u->manager->cgroup_unit, u->cgroup_path); | |
1945 | u->cgroup_path = mfree(u->cgroup_path); | |
1946 | } | |
1947 | ||
1948 | if (u->cgroup_inotify_wd >= 0) { | |
1949 | if (inotify_rm_watch(u->manager->cgroup_inotify_fd, u->cgroup_inotify_wd) < 0) | |
1950 | log_unit_debug_errno(u, errno, "Failed to remove cgroup inotify watch %i for %s, ignoring", u->cgroup_inotify_wd, u->id); | |
1951 | ||
1952 | (void) hashmap_remove(u->manager->cgroup_inotify_wd_unit, INT_TO_PTR(u->cgroup_inotify_wd)); | |
1953 | u->cgroup_inotify_wd = -1; | |
1954 | } | |
1955 | } | |
1956 | ||
1957 | void unit_prune_cgroup(Unit *u) { | |
8e274523 | 1958 | int r; |
efdb0237 | 1959 | bool is_root_slice; |
8e274523 | 1960 | |
4ad49000 | 1961 | assert(u); |
8e274523 | 1962 | |
efdb0237 LP |
1963 | /* Removes the cgroup, if empty and possible, and stops watching it. */ |
1964 | ||
4ad49000 LP |
1965 | if (!u->cgroup_path) |
1966 | return; | |
8e274523 | 1967 | |
fe700f46 LP |
1968 | (void) unit_get_cpu_usage(u, NULL); /* Cache the last CPU usage value before we destroy the cgroup */ |
1969 | ||
efdb0237 LP |
1970 | is_root_slice = unit_has_name(u, SPECIAL_ROOT_SLICE); |
1971 | ||
1972 | r = cg_trim_everywhere(u->manager->cgroup_supported, u->cgroup_path, !is_root_slice); | |
dab5bf85 | 1973 | if (r < 0) { |
f29ff115 | 1974 | log_unit_debug_errno(u, r, "Failed to destroy cgroup %s, ignoring: %m", u->cgroup_path); |
dab5bf85 RL |
1975 | return; |
1976 | } | |
8e274523 | 1977 | |
efdb0237 LP |
1978 | if (is_root_slice) |
1979 | return; | |
1980 | ||
1981 | unit_release_cgroup(u); | |
0a1eb06d | 1982 | |
4ad49000 | 1983 | u->cgroup_realized = false; |
bc432dc7 | 1984 | u->cgroup_realized_mask = 0; |
ccf78df1 | 1985 | u->cgroup_enabled_mask = 0; |
084c7007 RG |
1986 | |
1987 | u->bpf_device_control_installed = bpf_program_unref(u->bpf_device_control_installed); | |
8e274523 LP |
1988 | } |
1989 | ||
efdb0237 | 1990 | int unit_search_main_pid(Unit *u, pid_t *ret) { |
4ad49000 LP |
1991 | _cleanup_fclose_ FILE *f = NULL; |
1992 | pid_t pid = 0, npid, mypid; | |
efdb0237 | 1993 | int r; |
4ad49000 LP |
1994 | |
1995 | assert(u); | |
efdb0237 | 1996 | assert(ret); |
4ad49000 LP |
1997 | |
1998 | if (!u->cgroup_path) | |
efdb0237 | 1999 | return -ENXIO; |
4ad49000 | 2000 | |
efdb0237 LP |
2001 | r = cg_enumerate_processes(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path, &f); |
2002 | if (r < 0) | |
2003 | return r; | |
4ad49000 | 2004 | |
df0ff127 | 2005 | mypid = getpid_cached(); |
4ad49000 LP |
2006 | while (cg_read_pid(f, &npid) > 0) { |
2007 | pid_t ppid; | |
2008 | ||
2009 | if (npid == pid) | |
2010 | continue; | |
8e274523 | 2011 | |
4ad49000 | 2012 | /* Ignore processes that aren't our kids */ |
6bc73acb | 2013 | if (get_process_ppid(npid, &ppid) >= 0 && ppid != mypid) |
4ad49000 | 2014 | continue; |
8e274523 | 2015 | |
efdb0237 | 2016 | if (pid != 0) |
4ad49000 LP |
2017 | /* Dang, there's more than one daemonized PID |
2018 | in this group, so we don't know what process | |
2019 | is the main process. */ | |
efdb0237 LP |
2020 | |
2021 | return -ENODATA; | |
8e274523 | 2022 | |
4ad49000 | 2023 | pid = npid; |
8e274523 LP |
2024 | } |
2025 | ||
efdb0237 LP |
2026 | *ret = pid; |
2027 | return 0; | |
2028 | } | |
2029 | ||
2030 | static int unit_watch_pids_in_path(Unit *u, const char *path) { | |
b3c5bad3 | 2031 | _cleanup_closedir_ DIR *d = NULL; |
efdb0237 LP |
2032 | _cleanup_fclose_ FILE *f = NULL; |
2033 | int ret = 0, r; | |
2034 | ||
2035 | assert(u); | |
2036 | assert(path); | |
2037 | ||
2038 | r = cg_enumerate_processes(SYSTEMD_CGROUP_CONTROLLER, path, &f); | |
2039 | if (r < 0) | |
2040 | ret = r; | |
2041 | else { | |
2042 | pid_t pid; | |
2043 | ||
2044 | while ((r = cg_read_pid(f, &pid)) > 0) { | |
2045 | r = unit_watch_pid(u, pid); | |
2046 | if (r < 0 && ret >= 0) | |
2047 | ret = r; | |
2048 | } | |
2049 | ||
2050 | if (r < 0 && ret >= 0) | |
2051 | ret = r; | |
2052 | } | |
2053 | ||
2054 | r = cg_enumerate_subgroups(SYSTEMD_CGROUP_CONTROLLER, path, &d); | |
2055 | if (r < 0) { | |
2056 | if (ret >= 0) | |
2057 | ret = r; | |
2058 | } else { | |
2059 | char *fn; | |
2060 | ||
2061 | while ((r = cg_read_subgroup(d, &fn)) > 0) { | |
2062 | _cleanup_free_ char *p = NULL; | |
2063 | ||
605405c6 | 2064 | p = strjoin(path, "/", fn); |
efdb0237 LP |
2065 | free(fn); |
2066 | ||
2067 | if (!p) | |
2068 | return -ENOMEM; | |
2069 | ||
2070 | r = unit_watch_pids_in_path(u, p); | |
2071 | if (r < 0 && ret >= 0) | |
2072 | ret = r; | |
2073 | } | |
2074 | ||
2075 | if (r < 0 && ret >= 0) | |
2076 | ret = r; | |
2077 | } | |
2078 | ||
2079 | return ret; | |
2080 | } | |
2081 | ||
11aef522 LP |
2082 | int unit_synthesize_cgroup_empty_event(Unit *u) { |
2083 | int r; | |
2084 | ||
2085 | assert(u); | |
2086 | ||
2087 | /* Enqueue a synthetic cgroup empty event if this unit doesn't watch any PIDs anymore. This is compatibility | |
2088 | * support for non-unified systems where notifications aren't reliable, and hence need to take whatever we can | |
2089 | * get as notification source as soon as we stopped having any useful PIDs to watch for. */ | |
2090 | ||
2091 | if (!u->cgroup_path) | |
2092 | return -ENOENT; | |
2093 | ||
2094 | r = cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER); | |
2095 | if (r < 0) | |
2096 | return r; | |
2097 | if (r > 0) /* On unified we have reliable notifications, and don't need this */ | |
2098 | return 0; | |
2099 | ||
2100 | if (!set_isempty(u->pids)) | |
2101 | return 0; | |
2102 | ||
2103 | unit_add_to_cgroup_empty_queue(u); | |
2104 | return 0; | |
2105 | } | |
2106 | ||
efdb0237 | 2107 | int unit_watch_all_pids(Unit *u) { |
b4cccbc1 LP |
2108 | int r; |
2109 | ||
efdb0237 LP |
2110 | assert(u); |
2111 | ||
2112 | /* Adds all PIDs from our cgroup to the set of PIDs we | |
2113 | * watch. This is a fallback logic for cases where we do not | |
2114 | * get reliable cgroup empty notifications: we try to use | |
2115 | * SIGCHLD as replacement. */ | |
2116 | ||
2117 | if (!u->cgroup_path) | |
2118 | return -ENOENT; | |
2119 | ||
c22800e4 | 2120 | r = cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER); |
b4cccbc1 LP |
2121 | if (r < 0) |
2122 | return r; | |
2123 | if (r > 0) /* On unified we can use proper notifications */ | |
efdb0237 LP |
2124 | return 0; |
2125 | ||
2126 | return unit_watch_pids_in_path(u, u->cgroup_path); | |
2127 | } | |
2128 | ||
09e24654 LP |
2129 | static int on_cgroup_empty_event(sd_event_source *s, void *userdata) { |
2130 | Manager *m = userdata; | |
2131 | Unit *u; | |
efdb0237 LP |
2132 | int r; |
2133 | ||
09e24654 LP |
2134 | assert(s); |
2135 | assert(m); | |
efdb0237 | 2136 | |
09e24654 LP |
2137 | u = m->cgroup_empty_queue; |
2138 | if (!u) | |
efdb0237 LP |
2139 | return 0; |
2140 | ||
09e24654 LP |
2141 | assert(u->in_cgroup_empty_queue); |
2142 | u->in_cgroup_empty_queue = false; | |
2143 | LIST_REMOVE(cgroup_empty_queue, m->cgroup_empty_queue, u); | |
2144 | ||
2145 | if (m->cgroup_empty_queue) { | |
2146 | /* More stuff queued, let's make sure we remain enabled */ | |
2147 | r = sd_event_source_set_enabled(s, SD_EVENT_ONESHOT); | |
2148 | if (r < 0) | |
19a691a9 | 2149 | log_debug_errno(r, "Failed to reenable cgroup empty event source, ignoring: %m"); |
09e24654 | 2150 | } |
efdb0237 LP |
2151 | |
2152 | unit_add_to_gc_queue(u); | |
2153 | ||
2154 | if (UNIT_VTABLE(u)->notify_cgroup_empty) | |
2155 | UNIT_VTABLE(u)->notify_cgroup_empty(u); | |
2156 | ||
2157 | return 0; | |
2158 | } | |
2159 | ||
09e24654 LP |
2160 | void unit_add_to_cgroup_empty_queue(Unit *u) { |
2161 | int r; | |
2162 | ||
2163 | assert(u); | |
2164 | ||
2165 | /* Note that there are four different ways how cgroup empty events reach us: | |
2166 | * | |
2167 | * 1. On the unified hierarchy we get an inotify event on the cgroup | |
2168 | * | |
2169 | * 2. On the legacy hierarchy, when running in system mode, we get a datagram on the cgroup agent socket | |
2170 | * | |
2171 | * 3. On the legacy hierarchy, when running in user mode, we get a D-Bus signal on the system bus | |
2172 | * | |
2173 | * 4. On the legacy hierarchy, in service units we start watching all processes of the cgroup for SIGCHLD as | |
2174 | * soon as we get one SIGCHLD, to deal with unreliable cgroup notifications. | |
2175 | * | |
2176 | * Regardless which way we got the notification, we'll verify it here, and then add it to a separate | |
2177 | * queue. This queue will be dispatched at a lower priority than the SIGCHLD handler, so that we always use | |
2178 | * SIGCHLD if we can get it first, and only use the cgroup empty notifications if there's no SIGCHLD pending | |
2179 | * (which might happen if the cgroup doesn't contain processes that are our own child, which is typically the | |
2180 | * case for scope units). */ | |
2181 | ||
2182 | if (u->in_cgroup_empty_queue) | |
2183 | return; | |
2184 | ||
2185 | /* Let's verify that the cgroup is really empty */ | |
2186 | if (!u->cgroup_path) | |
2187 | return; | |
2188 | r = cg_is_empty_recursive(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path); | |
2189 | if (r < 0) { | |
2190 | log_unit_debug_errno(u, r, "Failed to determine whether cgroup %s is empty: %m", u->cgroup_path); | |
2191 | return; | |
2192 | } | |
2193 | if (r == 0) | |
2194 | return; | |
2195 | ||
2196 | LIST_PREPEND(cgroup_empty_queue, u->manager->cgroup_empty_queue, u); | |
2197 | u->in_cgroup_empty_queue = true; | |
2198 | ||
2199 | /* Trigger the defer event */ | |
2200 | r = sd_event_source_set_enabled(u->manager->cgroup_empty_event_source, SD_EVENT_ONESHOT); | |
2201 | if (r < 0) | |
2202 | log_debug_errno(r, "Failed to enable cgroup empty event source: %m"); | |
2203 | } | |
2204 | ||
efdb0237 LP |
2205 | static int on_cgroup_inotify_event(sd_event_source *s, int fd, uint32_t revents, void *userdata) { |
2206 | Manager *m = userdata; | |
2207 | ||
2208 | assert(s); | |
2209 | assert(fd >= 0); | |
2210 | assert(m); | |
2211 | ||
2212 | for (;;) { | |
2213 | union inotify_event_buffer buffer; | |
2214 | struct inotify_event *e; | |
2215 | ssize_t l; | |
2216 | ||
2217 | l = read(fd, &buffer, sizeof(buffer)); | |
2218 | if (l < 0) { | |
47249640 | 2219 | if (IN_SET(errno, EINTR, EAGAIN)) |
efdb0237 LP |
2220 | return 0; |
2221 | ||
2222 | return log_error_errno(errno, "Failed to read control group inotify events: %m"); | |
2223 | } | |
2224 | ||
2225 | FOREACH_INOTIFY_EVENT(e, buffer, l) { | |
2226 | Unit *u; | |
2227 | ||
2228 | if (e->wd < 0) | |
2229 | /* Queue overflow has no watch descriptor */ | |
2230 | continue; | |
2231 | ||
2232 | if (e->mask & IN_IGNORED) | |
2233 | /* The watch was just removed */ | |
2234 | continue; | |
2235 | ||
2236 | u = hashmap_get(m->cgroup_inotify_wd_unit, INT_TO_PTR(e->wd)); | |
2237 | if (!u) /* Not that inotify might deliver | |
2238 | * events for a watch even after it | |
2239 | * was removed, because it was queued | |
2240 | * before the removal. Let's ignore | |
2241 | * this here safely. */ | |
2242 | continue; | |
2243 | ||
09e24654 | 2244 | unit_add_to_cgroup_empty_queue(u); |
efdb0237 LP |
2245 | } |
2246 | } | |
8e274523 LP |
2247 | } |
2248 | ||
17f14955 RG |
2249 | static int cg_bpf_mask_supported(CGroupMask *ret) { |
2250 | CGroupMask mask = 0; | |
2251 | int r; | |
2252 | ||
2253 | /* BPF-based firewall */ | |
2254 | r = bpf_firewall_supported(); | |
2255 | if (r > 0) | |
2256 | mask |= CGROUP_MASK_BPF_FIREWALL; | |
2257 | ||
084c7007 RG |
2258 | /* BPF-based device access control */ |
2259 | r = bpf_devices_supported(); | |
2260 | if (r > 0) | |
2261 | mask |= CGROUP_MASK_BPF_DEVICES; | |
2262 | ||
17f14955 RG |
2263 | *ret = mask; |
2264 | return 0; | |
2265 | } | |
2266 | ||
8e274523 | 2267 | int manager_setup_cgroup(Manager *m) { |
9444b1f2 | 2268 | _cleanup_free_ char *path = NULL; |
10bd3e2e | 2269 | const char *scope_path; |
efdb0237 | 2270 | CGroupController c; |
b4cccbc1 | 2271 | int r, all_unified; |
17f14955 | 2272 | CGroupMask mask; |
efdb0237 | 2273 | char *e; |
8e274523 LP |
2274 | |
2275 | assert(m); | |
2276 | ||
35d2e7ec | 2277 | /* 1. Determine hierarchy */ |
efdb0237 | 2278 | m->cgroup_root = mfree(m->cgroup_root); |
9444b1f2 | 2279 | r = cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER, 0, &m->cgroup_root); |
23bbb0de MS |
2280 | if (r < 0) |
2281 | return log_error_errno(r, "Cannot determine cgroup we are running in: %m"); | |
8e274523 | 2282 | |
efdb0237 LP |
2283 | /* Chop off the init scope, if we are already located in it */ |
2284 | e = endswith(m->cgroup_root, "/" SPECIAL_INIT_SCOPE); | |
0d8c31ff | 2285 | |
efdb0237 LP |
2286 | /* LEGACY: Also chop off the system slice if we are in |
2287 | * it. This is to support live upgrades from older systemd | |
2288 | * versions where PID 1 was moved there. Also see | |
2289 | * cg_get_root_path(). */ | |
463d0d15 | 2290 | if (!e && MANAGER_IS_SYSTEM(m)) { |
9444b1f2 | 2291 | e = endswith(m->cgroup_root, "/" SPECIAL_SYSTEM_SLICE); |
15c60e99 | 2292 | if (!e) |
efdb0237 | 2293 | e = endswith(m->cgroup_root, "/system"); /* even more legacy */ |
0baf24dd | 2294 | } |
efdb0237 LP |
2295 | if (e) |
2296 | *e = 0; | |
7ccfb64a | 2297 | |
7546145e LP |
2298 | /* And make sure to store away the root value without trailing slash, even for the root dir, so that we can |
2299 | * easily prepend it everywhere. */ | |
2300 | delete_trailing_chars(m->cgroup_root, "/"); | |
8e274523 | 2301 | |
35d2e7ec | 2302 | /* 2. Show data */ |
9444b1f2 | 2303 | r = cg_get_path(SYSTEMD_CGROUP_CONTROLLER, m->cgroup_root, NULL, &path); |
23bbb0de MS |
2304 | if (r < 0) |
2305 | return log_error_errno(r, "Cannot find cgroup mount point: %m"); | |
8e274523 | 2306 | |
415fc41c TH |
2307 | r = cg_unified_flush(); |
2308 | if (r < 0) | |
2309 | return log_error_errno(r, "Couldn't determine if we are running in the unified hierarchy: %m"); | |
5da38d07 | 2310 | |
b4cccbc1 | 2311 | all_unified = cg_all_unified(); |
d4c819ed ZJS |
2312 | if (all_unified < 0) |
2313 | return log_error_errno(all_unified, "Couldn't determine whether we are in all unified mode: %m"); | |
2314 | if (all_unified > 0) | |
efdb0237 | 2315 | log_debug("Unified cgroup hierarchy is located at %s.", path); |
b4cccbc1 | 2316 | else { |
c22800e4 | 2317 | r = cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER); |
b4cccbc1 LP |
2318 | if (r < 0) |
2319 | return log_error_errno(r, "Failed to determine whether systemd's own controller is in unified mode: %m"); | |
2320 | if (r > 0) | |
2321 | log_debug("Unified cgroup hierarchy is located at %s. Controllers are on legacy hierarchies.", path); | |
2322 | else | |
2323 | log_debug("Using cgroup controller " SYSTEMD_CGROUP_CONTROLLER_LEGACY ". File system hierarchy is at %s.", path); | |
2324 | } | |
efdb0237 | 2325 | |
09e24654 LP |
2326 | /* 3. Allocate cgroup empty defer event source */ |
2327 | m->cgroup_empty_event_source = sd_event_source_unref(m->cgroup_empty_event_source); | |
2328 | r = sd_event_add_defer(m->event, &m->cgroup_empty_event_source, on_cgroup_empty_event, m); | |
2329 | if (r < 0) | |
2330 | return log_error_errno(r, "Failed to create cgroup empty event source: %m"); | |
2331 | ||
2332 | r = sd_event_source_set_priority(m->cgroup_empty_event_source, SD_EVENT_PRIORITY_NORMAL-5); | |
2333 | if (r < 0) | |
2334 | return log_error_errno(r, "Failed to set priority of cgroup empty event source: %m"); | |
2335 | ||
2336 | r = sd_event_source_set_enabled(m->cgroup_empty_event_source, SD_EVENT_OFF); | |
2337 | if (r < 0) | |
2338 | return log_error_errno(r, "Failed to disable cgroup empty event source: %m"); | |
2339 | ||
2340 | (void) sd_event_source_set_description(m->cgroup_empty_event_source, "cgroup-empty"); | |
2341 | ||
2342 | /* 4. Install notifier inotify object, or agent */ | |
10bd3e2e | 2343 | if (cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER) > 0) { |
c6c18be3 | 2344 | |
09e24654 | 2345 | /* In the unified hierarchy we can get cgroup empty notifications via inotify. */ |
efdb0237 | 2346 | |
10bd3e2e LP |
2347 | m->cgroup_inotify_event_source = sd_event_source_unref(m->cgroup_inotify_event_source); |
2348 | safe_close(m->cgroup_inotify_fd); | |
efdb0237 | 2349 | |
10bd3e2e LP |
2350 | m->cgroup_inotify_fd = inotify_init1(IN_NONBLOCK|IN_CLOEXEC); |
2351 | if (m->cgroup_inotify_fd < 0) | |
2352 | return log_error_errno(errno, "Failed to create control group inotify object: %m"); | |
efdb0237 | 2353 | |
10bd3e2e LP |
2354 | r = sd_event_add_io(m->event, &m->cgroup_inotify_event_source, m->cgroup_inotify_fd, EPOLLIN, on_cgroup_inotify_event, m); |
2355 | if (r < 0) | |
2356 | return log_error_errno(r, "Failed to watch control group inotify object: %m"); | |
efdb0237 | 2357 | |
10bd3e2e LP |
2358 | /* Process cgroup empty notifications early, but after service notifications and SIGCHLD. Also |
2359 | * see handling of cgroup agent notifications, for the classic cgroup hierarchy support. */ | |
09e24654 | 2360 | r = sd_event_source_set_priority(m->cgroup_inotify_event_source, SD_EVENT_PRIORITY_NORMAL-4); |
10bd3e2e LP |
2361 | if (r < 0) |
2362 | return log_error_errno(r, "Failed to set priority of inotify event source: %m"); | |
efdb0237 | 2363 | |
10bd3e2e | 2364 | (void) sd_event_source_set_description(m->cgroup_inotify_event_source, "cgroup-inotify"); |
efdb0237 | 2365 | |
638cece4 | 2366 | } else if (MANAGER_IS_SYSTEM(m) && !MANAGER_IS_TEST_RUN(m)) { |
efdb0237 | 2367 | |
10bd3e2e LP |
2368 | /* On the legacy hierarchy we only get notifications via cgroup agents. (Which isn't really reliable, |
2369 | * since it does not generate events when control groups with children run empty. */ | |
8e274523 | 2370 | |
10bd3e2e | 2371 | r = cg_install_release_agent(SYSTEMD_CGROUP_CONTROLLER, SYSTEMD_CGROUP_AGENT_PATH); |
23bbb0de | 2372 | if (r < 0) |
10bd3e2e LP |
2373 | log_warning_errno(r, "Failed to install release agent, ignoring: %m"); |
2374 | else if (r > 0) | |
2375 | log_debug("Installed release agent."); | |
2376 | else if (r == 0) | |
2377 | log_debug("Release agent already installed."); | |
2378 | } | |
efdb0237 | 2379 | |
09e24654 | 2380 | /* 5. Make sure we are in the special "init.scope" unit in the root slice. */ |
10bd3e2e LP |
2381 | scope_path = strjoina(m->cgroup_root, "/" SPECIAL_INIT_SCOPE); |
2382 | r = cg_create_and_attach(SYSTEMD_CGROUP_CONTROLLER, scope_path, 0); | |
aa77e234 MS |
2383 | if (r >= 0) { |
2384 | /* Also, move all other userspace processes remaining in the root cgroup into that scope. */ | |
2385 | r = cg_migrate(SYSTEMD_CGROUP_CONTROLLER, m->cgroup_root, SYSTEMD_CGROUP_CONTROLLER, scope_path, 0); | |
2386 | if (r < 0) | |
2387 | log_warning_errno(r, "Couldn't move remaining userspace processes, ignoring: %m"); | |
c6c18be3 | 2388 | |
aa77e234 MS |
2389 | /* 6. And pin it, so that it cannot be unmounted */ |
2390 | safe_close(m->pin_cgroupfs_fd); | |
2391 | m->pin_cgroupfs_fd = open(path, O_RDONLY|O_CLOEXEC|O_DIRECTORY|O_NOCTTY|O_NONBLOCK); | |
2392 | if (m->pin_cgroupfs_fd < 0) | |
2393 | return log_error_errno(errno, "Failed to open pin file: %m"); | |
0d8c31ff | 2394 | |
638cece4 | 2395 | } else if (!MANAGER_IS_TEST_RUN(m)) |
aa77e234 | 2396 | return log_error_errno(r, "Failed to create %s control group: %m", scope_path); |
10bd3e2e | 2397 | |
09e24654 | 2398 | /* 7. Always enable hierarchical support if it exists... */ |
638cece4 | 2399 | if (!all_unified && !MANAGER_IS_TEST_RUN(m)) |
10bd3e2e | 2400 | (void) cg_set_attribute("memory", "/", "memory.use_hierarchy", "1"); |
c6c18be3 | 2401 | |
17f14955 | 2402 | /* 8. Figure out which controllers are supported */ |
efdb0237 LP |
2403 | r = cg_mask_supported(&m->cgroup_supported); |
2404 | if (r < 0) | |
2405 | return log_error_errno(r, "Failed to determine supported controllers: %m"); | |
17f14955 RG |
2406 | |
2407 | /* 9. Figure out which bpf-based pseudo-controllers are supported */ | |
2408 | r = cg_bpf_mask_supported(&mask); | |
2409 | if (r < 0) | |
2410 | return log_error_errno(r, "Failed to determine supported bpf-based pseudo-controllers: %m"); | |
2411 | m->cgroup_supported |= mask; | |
2412 | ||
2413 | /* 10. Log which controllers are supported */ | |
efdb0237 | 2414 | for (c = 0; c < _CGROUP_CONTROLLER_MAX; c++) |
eee0a1e4 | 2415 | log_debug("Controller '%s' supported: %s", cgroup_controller_to_string(c), yes_no(m->cgroup_supported & CGROUP_CONTROLLER_TO_MASK(c))); |
9156e799 | 2416 | |
a32360f1 | 2417 | return 0; |
8e274523 LP |
2418 | } |
2419 | ||
c6c18be3 | 2420 | void manager_shutdown_cgroup(Manager *m, bool delete) { |
8e274523 LP |
2421 | assert(m); |
2422 | ||
9444b1f2 LP |
2423 | /* We can't really delete the group, since we are in it. But |
2424 | * let's trim it. */ | |
f6c63f6f | 2425 | if (delete && m->cgroup_root && m->test_run_flags != MANAGER_TEST_RUN_MINIMAL) |
efdb0237 LP |
2426 | (void) cg_trim(SYSTEMD_CGROUP_CONTROLLER, m->cgroup_root, false); |
2427 | ||
09e24654 LP |
2428 | m->cgroup_empty_event_source = sd_event_source_unref(m->cgroup_empty_event_source); |
2429 | ||
efdb0237 LP |
2430 | m->cgroup_inotify_wd_unit = hashmap_free(m->cgroup_inotify_wd_unit); |
2431 | ||
2432 | m->cgroup_inotify_event_source = sd_event_source_unref(m->cgroup_inotify_event_source); | |
2433 | m->cgroup_inotify_fd = safe_close(m->cgroup_inotify_fd); | |
8e274523 | 2434 | |
03e334a1 | 2435 | m->pin_cgroupfs_fd = safe_close(m->pin_cgroupfs_fd); |
c6c18be3 | 2436 | |
efdb0237 | 2437 | m->cgroup_root = mfree(m->cgroup_root); |
8e274523 LP |
2438 | } |
2439 | ||
4ad49000 | 2440 | Unit* manager_get_unit_by_cgroup(Manager *m, const char *cgroup) { |
acb14d31 | 2441 | char *p; |
4ad49000 | 2442 | Unit *u; |
acb14d31 LP |
2443 | |
2444 | assert(m); | |
2445 | assert(cgroup); | |
acb14d31 | 2446 | |
4ad49000 LP |
2447 | u = hashmap_get(m->cgroup_unit, cgroup); |
2448 | if (u) | |
2449 | return u; | |
acb14d31 | 2450 | |
8e70580b | 2451 | p = strdupa(cgroup); |
acb14d31 LP |
2452 | for (;;) { |
2453 | char *e; | |
2454 | ||
2455 | e = strrchr(p, '/'); | |
efdb0237 LP |
2456 | if (!e || e == p) |
2457 | return hashmap_get(m->cgroup_unit, SPECIAL_ROOT_SLICE); | |
acb14d31 LP |
2458 | |
2459 | *e = 0; | |
2460 | ||
4ad49000 LP |
2461 | u = hashmap_get(m->cgroup_unit, p); |
2462 | if (u) | |
2463 | return u; | |
acb14d31 LP |
2464 | } |
2465 | } | |
2466 | ||
b3ac818b | 2467 | Unit *manager_get_unit_by_pid_cgroup(Manager *m, pid_t pid) { |
4ad49000 | 2468 | _cleanup_free_ char *cgroup = NULL; |
8e274523 | 2469 | |
8c47c732 LP |
2470 | assert(m); |
2471 | ||
62a76913 | 2472 | if (!pid_is_valid(pid)) |
b3ac818b LP |
2473 | return NULL; |
2474 | ||
62a76913 | 2475 | if (cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER, pid, &cgroup) < 0) |
b3ac818b LP |
2476 | return NULL; |
2477 | ||
2478 | return manager_get_unit_by_cgroup(m, cgroup); | |
2479 | } | |
2480 | ||
2481 | Unit *manager_get_unit_by_pid(Manager *m, pid_t pid) { | |
62a76913 | 2482 | Unit *u, **array; |
b3ac818b LP |
2483 | |
2484 | assert(m); | |
2485 | ||
62a76913 LP |
2486 | /* Note that a process might be owned by multiple units, we return only one here, which is good enough for most |
2487 | * cases, though not strictly correct. We prefer the one reported by cgroup membership, as that's the most | |
2488 | * relevant one as children of the process will be assigned to that one, too, before all else. */ | |
2489 | ||
2490 | if (!pid_is_valid(pid)) | |
8c47c732 LP |
2491 | return NULL; |
2492 | ||
2ca9d979 | 2493 | if (pid == getpid_cached()) |
efdb0237 LP |
2494 | return hashmap_get(m->units, SPECIAL_INIT_SCOPE); |
2495 | ||
62a76913 | 2496 | u = manager_get_unit_by_pid_cgroup(m, pid); |
5fe8876b LP |
2497 | if (u) |
2498 | return u; | |
2499 | ||
62a76913 | 2500 | u = hashmap_get(m->watch_pids, PID_TO_PTR(pid)); |
5fe8876b LP |
2501 | if (u) |
2502 | return u; | |
2503 | ||
62a76913 LP |
2504 | array = hashmap_get(m->watch_pids, PID_TO_PTR(-pid)); |
2505 | if (array) | |
2506 | return array[0]; | |
2507 | ||
2508 | return NULL; | |
6dde1f33 | 2509 | } |
4fbf50b3 | 2510 | |
4ad49000 LP |
2511 | int manager_notify_cgroup_empty(Manager *m, const char *cgroup) { |
2512 | Unit *u; | |
4fbf50b3 | 2513 | |
4ad49000 LP |
2514 | assert(m); |
2515 | assert(cgroup); | |
4fbf50b3 | 2516 | |
09e24654 LP |
2517 | /* Called on the legacy hierarchy whenever we get an explicit cgroup notification from the cgroup agent process |
2518 | * or from the --system instance */ | |
2519 | ||
d8fdc620 LP |
2520 | log_debug("Got cgroup empty notification for: %s", cgroup); |
2521 | ||
4ad49000 | 2522 | u = manager_get_unit_by_cgroup(m, cgroup); |
5ad096b3 LP |
2523 | if (!u) |
2524 | return 0; | |
b56c28c3 | 2525 | |
09e24654 LP |
2526 | unit_add_to_cgroup_empty_queue(u); |
2527 | return 1; | |
5ad096b3 LP |
2528 | } |
2529 | ||
2530 | int unit_get_memory_current(Unit *u, uint64_t *ret) { | |
2531 | _cleanup_free_ char *v = NULL; | |
2532 | int r; | |
2533 | ||
2534 | assert(u); | |
2535 | assert(ret); | |
2536 | ||
2e4025c0 | 2537 | if (!UNIT_CGROUP_BOOL(u, memory_accounting)) |
cf3b4be1 LP |
2538 | return -ENODATA; |
2539 | ||
5ad096b3 LP |
2540 | if (!u->cgroup_path) |
2541 | return -ENODATA; | |
2542 | ||
1f73aa00 LP |
2543 | /* The root cgroup doesn't expose this information, let's get it from /proc instead */ |
2544 | if (unit_has_root_cgroup(u)) | |
2545 | return procfs_memory_get_current(ret); | |
2546 | ||
efdb0237 | 2547 | if ((u->cgroup_realized_mask & CGROUP_MASK_MEMORY) == 0) |
5ad096b3 LP |
2548 | return -ENODATA; |
2549 | ||
b4cccbc1 LP |
2550 | r = cg_all_unified(); |
2551 | if (r < 0) | |
2552 | return r; | |
2553 | if (r > 0) | |
efdb0237 | 2554 | r = cg_get_attribute("memory", u->cgroup_path, "memory.current", &v); |
b4cccbc1 LP |
2555 | else |
2556 | r = cg_get_attribute("memory", u->cgroup_path, "memory.usage_in_bytes", &v); | |
5ad096b3 LP |
2557 | if (r == -ENOENT) |
2558 | return -ENODATA; | |
2559 | if (r < 0) | |
2560 | return r; | |
2561 | ||
2562 | return safe_atou64(v, ret); | |
2563 | } | |
2564 | ||
03a7b521 LP |
2565 | int unit_get_tasks_current(Unit *u, uint64_t *ret) { |
2566 | _cleanup_free_ char *v = NULL; | |
2567 | int r; | |
2568 | ||
2569 | assert(u); | |
2570 | assert(ret); | |
2571 | ||
2e4025c0 | 2572 | if (!UNIT_CGROUP_BOOL(u, tasks_accounting)) |
cf3b4be1 LP |
2573 | return -ENODATA; |
2574 | ||
03a7b521 LP |
2575 | if (!u->cgroup_path) |
2576 | return -ENODATA; | |
2577 | ||
c36a69f4 LP |
2578 | /* The root cgroup doesn't expose this information, let's get it from /proc instead */ |
2579 | if (unit_has_root_cgroup(u)) | |
2580 | return procfs_tasks_get_current(ret); | |
2581 | ||
1f73aa00 LP |
2582 | if ((u->cgroup_realized_mask & CGROUP_MASK_PIDS) == 0) |
2583 | return -ENODATA; | |
2584 | ||
03a7b521 LP |
2585 | r = cg_get_attribute("pids", u->cgroup_path, "pids.current", &v); |
2586 | if (r == -ENOENT) | |
2587 | return -ENODATA; | |
2588 | if (r < 0) | |
2589 | return r; | |
2590 | ||
2591 | return safe_atou64(v, ret); | |
2592 | } | |
2593 | ||
5ad096b3 LP |
2594 | static int unit_get_cpu_usage_raw(Unit *u, nsec_t *ret) { |
2595 | _cleanup_free_ char *v = NULL; | |
2596 | uint64_t ns; | |
2597 | int r; | |
2598 | ||
2599 | assert(u); | |
2600 | assert(ret); | |
2601 | ||
2602 | if (!u->cgroup_path) | |
2603 | return -ENODATA; | |
2604 | ||
1f73aa00 LP |
2605 | /* The root cgroup doesn't expose this information, let's get it from /proc instead */ |
2606 | if (unit_has_root_cgroup(u)) | |
2607 | return procfs_cpu_get_usage(ret); | |
2608 | ||
b4cccbc1 LP |
2609 | r = cg_all_unified(); |
2610 | if (r < 0) | |
2611 | return r; | |
2612 | if (r > 0) { | |
66ebf6c0 TH |
2613 | _cleanup_free_ char *val = NULL; |
2614 | uint64_t us; | |
5ad096b3 | 2615 | |
66ebf6c0 TH |
2616 | if ((u->cgroup_realized_mask & CGROUP_MASK_CPU) == 0) |
2617 | return -ENODATA; | |
5ad096b3 | 2618 | |
b734a4ff | 2619 | r = cg_get_keyed_attribute("cpu", u->cgroup_path, "cpu.stat", STRV_MAKE("usage_usec"), &val); |
66ebf6c0 TH |
2620 | if (r < 0) |
2621 | return r; | |
b734a4ff LP |
2622 | if (IN_SET(r, -ENOENT, -ENXIO)) |
2623 | return -ENODATA; | |
66ebf6c0 TH |
2624 | |
2625 | r = safe_atou64(val, &us); | |
2626 | if (r < 0) | |
2627 | return r; | |
2628 | ||
2629 | ns = us * NSEC_PER_USEC; | |
2630 | } else { | |
2631 | if ((u->cgroup_realized_mask & CGROUP_MASK_CPUACCT) == 0) | |
2632 | return -ENODATA; | |
2633 | ||
2634 | r = cg_get_attribute("cpuacct", u->cgroup_path, "cpuacct.usage", &v); | |
2635 | if (r == -ENOENT) | |
2636 | return -ENODATA; | |
2637 | if (r < 0) | |
2638 | return r; | |
2639 | ||
2640 | r = safe_atou64(v, &ns); | |
2641 | if (r < 0) | |
2642 | return r; | |
2643 | } | |
5ad096b3 LP |
2644 | |
2645 | *ret = ns; | |
2646 | return 0; | |
2647 | } | |
2648 | ||
2649 | int unit_get_cpu_usage(Unit *u, nsec_t *ret) { | |
2650 | nsec_t ns; | |
2651 | int r; | |
2652 | ||
fe700f46 LP |
2653 | assert(u); |
2654 | ||
2655 | /* Retrieve the current CPU usage counter. This will subtract the CPU counter taken when the unit was | |
2656 | * started. If the cgroup has been removed already, returns the last cached value. To cache the value, simply | |
2657 | * call this function with a NULL return value. */ | |
2658 | ||
2e4025c0 | 2659 | if (!UNIT_CGROUP_BOOL(u, cpu_accounting)) |
cf3b4be1 LP |
2660 | return -ENODATA; |
2661 | ||
5ad096b3 | 2662 | r = unit_get_cpu_usage_raw(u, &ns); |
fe700f46 LP |
2663 | if (r == -ENODATA && u->cpu_usage_last != NSEC_INFINITY) { |
2664 | /* If we can't get the CPU usage anymore (because the cgroup was already removed, for example), use our | |
2665 | * cached value. */ | |
2666 | ||
2667 | if (ret) | |
2668 | *ret = u->cpu_usage_last; | |
2669 | return 0; | |
2670 | } | |
5ad096b3 LP |
2671 | if (r < 0) |
2672 | return r; | |
2673 | ||
66ebf6c0 TH |
2674 | if (ns > u->cpu_usage_base) |
2675 | ns -= u->cpu_usage_base; | |
5ad096b3 LP |
2676 | else |
2677 | ns = 0; | |
2678 | ||
fe700f46 LP |
2679 | u->cpu_usage_last = ns; |
2680 | if (ret) | |
2681 | *ret = ns; | |
2682 | ||
5ad096b3 LP |
2683 | return 0; |
2684 | } | |
2685 | ||
906c06f6 DM |
2686 | int unit_get_ip_accounting( |
2687 | Unit *u, | |
2688 | CGroupIPAccountingMetric metric, | |
2689 | uint64_t *ret) { | |
2690 | ||
6b659ed8 | 2691 | uint64_t value; |
906c06f6 DM |
2692 | int fd, r; |
2693 | ||
2694 | assert(u); | |
2695 | assert(metric >= 0); | |
2696 | assert(metric < _CGROUP_IP_ACCOUNTING_METRIC_MAX); | |
2697 | assert(ret); | |
2698 | ||
2e4025c0 | 2699 | if (!UNIT_CGROUP_BOOL(u, ip_accounting)) |
cf3b4be1 LP |
2700 | return -ENODATA; |
2701 | ||
906c06f6 DM |
2702 | fd = IN_SET(metric, CGROUP_IP_INGRESS_BYTES, CGROUP_IP_INGRESS_PACKETS) ? |
2703 | u->ip_accounting_ingress_map_fd : | |
2704 | u->ip_accounting_egress_map_fd; | |
906c06f6 DM |
2705 | if (fd < 0) |
2706 | return -ENODATA; | |
2707 | ||
2708 | if (IN_SET(metric, CGROUP_IP_INGRESS_BYTES, CGROUP_IP_EGRESS_BYTES)) | |
6b659ed8 | 2709 | r = bpf_firewall_read_accounting(fd, &value, NULL); |
906c06f6 | 2710 | else |
6b659ed8 LP |
2711 | r = bpf_firewall_read_accounting(fd, NULL, &value); |
2712 | if (r < 0) | |
2713 | return r; | |
2714 | ||
2715 | /* Add in additional metrics from a previous runtime. Note that when reexecing/reloading the daemon we compile | |
2716 | * all BPF programs and maps anew, but serialize the old counters. When deserializing we store them in the | |
2717 | * ip_accounting_extra[] field, and add them in here transparently. */ | |
2718 | ||
2719 | *ret = value + u->ip_accounting_extra[metric]; | |
906c06f6 DM |
2720 | |
2721 | return r; | |
2722 | } | |
2723 | ||
2724 | int unit_reset_cpu_accounting(Unit *u) { | |
5ad096b3 LP |
2725 | nsec_t ns; |
2726 | int r; | |
2727 | ||
2728 | assert(u); | |
2729 | ||
fe700f46 LP |
2730 | u->cpu_usage_last = NSEC_INFINITY; |
2731 | ||
5ad096b3 LP |
2732 | r = unit_get_cpu_usage_raw(u, &ns); |
2733 | if (r < 0) { | |
66ebf6c0 | 2734 | u->cpu_usage_base = 0; |
5ad096b3 | 2735 | return r; |
b56c28c3 | 2736 | } |
2633eb83 | 2737 | |
66ebf6c0 | 2738 | u->cpu_usage_base = ns; |
4ad49000 | 2739 | return 0; |
4fbf50b3 LP |
2740 | } |
2741 | ||
906c06f6 DM |
2742 | int unit_reset_ip_accounting(Unit *u) { |
2743 | int r = 0, q = 0; | |
2744 | ||
2745 | assert(u); | |
2746 | ||
2747 | if (u->ip_accounting_ingress_map_fd >= 0) | |
2748 | r = bpf_firewall_reset_accounting(u->ip_accounting_ingress_map_fd); | |
2749 | ||
2750 | if (u->ip_accounting_egress_map_fd >= 0) | |
2751 | q = bpf_firewall_reset_accounting(u->ip_accounting_egress_map_fd); | |
2752 | ||
6b659ed8 LP |
2753 | zero(u->ip_accounting_extra); |
2754 | ||
906c06f6 DM |
2755 | return r < 0 ? r : q; |
2756 | } | |
2757 | ||
e7ab4d1a LP |
2758 | void unit_invalidate_cgroup(Unit *u, CGroupMask m) { |
2759 | assert(u); | |
2760 | ||
2761 | if (!UNIT_HAS_CGROUP_CONTEXT(u)) | |
2762 | return; | |
2763 | ||
2764 | if (m == 0) | |
2765 | return; | |
2766 | ||
538b4852 TH |
2767 | /* always invalidate compat pairs together */ |
2768 | if (m & (CGROUP_MASK_IO | CGROUP_MASK_BLKIO)) | |
2769 | m |= CGROUP_MASK_IO | CGROUP_MASK_BLKIO; | |
2770 | ||
7cce4fb7 LP |
2771 | if (m & (CGROUP_MASK_CPU | CGROUP_MASK_CPUACCT)) |
2772 | m |= CGROUP_MASK_CPU | CGROUP_MASK_CPUACCT; | |
2773 | ||
60c728ad | 2774 | if ((u->cgroup_realized_mask & m) == 0) /* NOP? */ |
e7ab4d1a LP |
2775 | return; |
2776 | ||
2777 | u->cgroup_realized_mask &= ~m; | |
91a6073e | 2778 | unit_add_to_cgroup_realize_queue(u); |
e7ab4d1a LP |
2779 | } |
2780 | ||
906c06f6 DM |
2781 | void unit_invalidate_cgroup_bpf(Unit *u) { |
2782 | assert(u); | |
2783 | ||
2784 | if (!UNIT_HAS_CGROUP_CONTEXT(u)) | |
2785 | return; | |
2786 | ||
17f14955 | 2787 | if (u->cgroup_invalidated_mask & CGROUP_MASK_BPF_FIREWALL) /* NOP? */ |
906c06f6 DM |
2788 | return; |
2789 | ||
17f14955 | 2790 | u->cgroup_invalidated_mask |= CGROUP_MASK_BPF_FIREWALL; |
91a6073e | 2791 | unit_add_to_cgroup_realize_queue(u); |
906c06f6 DM |
2792 | |
2793 | /* If we are a slice unit, we also need to put compile a new BPF program for all our children, as the IP access | |
2794 | * list of our children includes our own. */ | |
2795 | if (u->type == UNIT_SLICE) { | |
2796 | Unit *member; | |
2797 | Iterator i; | |
eef85c4a | 2798 | void *v; |
906c06f6 | 2799 | |
eef85c4a | 2800 | HASHMAP_FOREACH_KEY(v, member, u->dependencies[UNIT_BEFORE], i) { |
906c06f6 DM |
2801 | if (member == u) |
2802 | continue; | |
2803 | ||
2804 | if (UNIT_DEREF(member->slice) != u) | |
2805 | continue; | |
2806 | ||
2807 | unit_invalidate_cgroup_bpf(member); | |
2808 | } | |
2809 | } | |
2810 | } | |
2811 | ||
1d9cc876 LP |
2812 | bool unit_cgroup_delegate(Unit *u) { |
2813 | CGroupContext *c; | |
2814 | ||
2815 | assert(u); | |
2816 | ||
2817 | if (!UNIT_VTABLE(u)->can_delegate) | |
2818 | return false; | |
2819 | ||
2820 | c = unit_get_cgroup_context(u); | |
2821 | if (!c) | |
2822 | return false; | |
2823 | ||
2824 | return c->delegate; | |
2825 | } | |
2826 | ||
e7ab4d1a LP |
2827 | void manager_invalidate_startup_units(Manager *m) { |
2828 | Iterator i; | |
2829 | Unit *u; | |
2830 | ||
2831 | assert(m); | |
2832 | ||
2833 | SET_FOREACH(u, m->startup_units, i) | |
13c31542 | 2834 | unit_invalidate_cgroup(u, CGROUP_MASK_CPU|CGROUP_MASK_IO|CGROUP_MASK_BLKIO); |
e7ab4d1a LP |
2835 | } |
2836 | ||
4ad49000 LP |
2837 | static const char* const cgroup_device_policy_table[_CGROUP_DEVICE_POLICY_MAX] = { |
2838 | [CGROUP_AUTO] = "auto", | |
2839 | [CGROUP_CLOSED] = "closed", | |
2840 | [CGROUP_STRICT] = "strict", | |
2841 | }; | |
4fbf50b3 | 2842 | |
4ad49000 | 2843 | DEFINE_STRING_TABLE_LOOKUP(cgroup_device_policy, CGroupDevicePolicy); |