]>
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" |
6592b975 | 10 | #include "bus-error.h" |
03a7b521 | 11 | #include "cgroup-util.h" |
3ffd4af2 LP |
12 | #include "cgroup.h" |
13 | #include "fd-util.h" | |
0d39fa9c | 14 | #include "fileio.h" |
77601719 | 15 | #include "fs-util.h" |
6bedfcbb | 16 | #include "parse-util.h" |
9eb977db | 17 | #include "path-util.h" |
03a7b521 | 18 | #include "process-util.h" |
c36a69f4 | 19 | #include "procfs-util.h" |
9444b1f2 | 20 | #include "special.h" |
906c06f6 | 21 | #include "stdio-util.h" |
8b43440b | 22 | #include "string-table.h" |
07630cea | 23 | #include "string-util.h" |
cc6271f1 | 24 | #include "virt.h" |
8e274523 | 25 | |
9a054909 LP |
26 | #define CGROUP_CPU_QUOTA_PERIOD_USEC ((usec_t) 100 * USEC_PER_MSEC) |
27 | ||
cc6271f1 LP |
28 | bool manager_owns_root_cgroup(Manager *m) { |
29 | assert(m); | |
30 | ||
31 | /* Returns true if we are managing the root cgroup. Note that it isn't sufficient to just check whether the | |
32 | * group root path equals "/" since that will also be the case if CLONE_NEWCGROUP is in the mix. Since there's | |
33 | * appears to be no nice way to detect whether we are in a CLONE_NEWCGROUP namespace we instead just check if | |
34 | * we run in any kind of container virtualization. */ | |
35 | ||
36 | if (detect_container() > 0) | |
37 | return false; | |
38 | ||
57ea45e1 | 39 | return empty_or_root(m->cgroup_root); |
cc6271f1 LP |
40 | } |
41 | ||
f3725e64 LP |
42 | bool unit_has_root_cgroup(Unit *u) { |
43 | assert(u); | |
44 | ||
cc6271f1 LP |
45 | /* Returns whether this unit manages the root cgroup. This will return true if this unit is the root slice and |
46 | * the manager manages the root cgroup. */ | |
f3725e64 | 47 | |
cc6271f1 | 48 | if (!manager_owns_root_cgroup(u->manager)) |
f3725e64 LP |
49 | return false; |
50 | ||
cc6271f1 | 51 | return unit_has_name(u, SPECIAL_ROOT_SLICE); |
f3725e64 LP |
52 | } |
53 | ||
2b40998d | 54 | static void cgroup_compat_warn(void) { |
128fadc9 TH |
55 | static bool cgroup_compat_warned = false; |
56 | ||
57 | if (cgroup_compat_warned) | |
58 | return; | |
59 | ||
cc6271f1 LP |
60 | log_warning("cgroup compatibility translation between legacy and unified hierarchy settings activated. " |
61 | "See cgroup-compat debug messages for details."); | |
62 | ||
128fadc9 TH |
63 | cgroup_compat_warned = true; |
64 | } | |
65 | ||
66 | #define log_cgroup_compat(unit, fmt, ...) do { \ | |
67 | cgroup_compat_warn(); \ | |
68 | log_unit_debug(unit, "cgroup-compat: " fmt, ##__VA_ARGS__); \ | |
2b40998d | 69 | } while (false) |
128fadc9 | 70 | |
4ad49000 LP |
71 | void cgroup_context_init(CGroupContext *c) { |
72 | assert(c); | |
73 | ||
74 | /* Initialize everything to the kernel defaults, assuming the | |
75 | * structure is preinitialized to 0 */ | |
76 | ||
66ebf6c0 TH |
77 | c->cpu_weight = CGROUP_WEIGHT_INVALID; |
78 | c->startup_cpu_weight = CGROUP_WEIGHT_INVALID; | |
79 | c->cpu_quota_per_sec_usec = USEC_INFINITY; | |
80 | ||
d53d9474 LP |
81 | c->cpu_shares = CGROUP_CPU_SHARES_INVALID; |
82 | c->startup_cpu_shares = CGROUP_CPU_SHARES_INVALID; | |
d53d9474 | 83 | |
da4d897e TH |
84 | c->memory_high = CGROUP_LIMIT_MAX; |
85 | c->memory_max = CGROUP_LIMIT_MAX; | |
96e131ea | 86 | c->memory_swap_max = CGROUP_LIMIT_MAX; |
da4d897e TH |
87 | |
88 | c->memory_limit = CGROUP_LIMIT_MAX; | |
b2f8b02e | 89 | |
13c31542 TH |
90 | c->io_weight = CGROUP_WEIGHT_INVALID; |
91 | c->startup_io_weight = CGROUP_WEIGHT_INVALID; | |
92 | ||
d53d9474 LP |
93 | c->blockio_weight = CGROUP_BLKIO_WEIGHT_INVALID; |
94 | c->startup_blockio_weight = CGROUP_BLKIO_WEIGHT_INVALID; | |
95 | ||
96 | c->tasks_max = (uint64_t) -1; | |
4ad49000 | 97 | } |
8e274523 | 98 | |
4ad49000 LP |
99 | void cgroup_context_free_device_allow(CGroupContext *c, CGroupDeviceAllow *a) { |
100 | assert(c); | |
101 | assert(a); | |
102 | ||
71fda00f | 103 | LIST_REMOVE(device_allow, c->device_allow, a); |
4ad49000 LP |
104 | free(a->path); |
105 | free(a); | |
106 | } | |
107 | ||
13c31542 TH |
108 | void cgroup_context_free_io_device_weight(CGroupContext *c, CGroupIODeviceWeight *w) { |
109 | assert(c); | |
110 | assert(w); | |
111 | ||
112 | LIST_REMOVE(device_weights, c->io_device_weights, w); | |
113 | free(w->path); | |
114 | free(w); | |
115 | } | |
116 | ||
117 | void cgroup_context_free_io_device_limit(CGroupContext *c, CGroupIODeviceLimit *l) { | |
118 | assert(c); | |
119 | assert(l); | |
120 | ||
121 | LIST_REMOVE(device_limits, c->io_device_limits, l); | |
122 | free(l->path); | |
123 | free(l); | |
124 | } | |
125 | ||
4ad49000 LP |
126 | void cgroup_context_free_blockio_device_weight(CGroupContext *c, CGroupBlockIODeviceWeight *w) { |
127 | assert(c); | |
128 | assert(w); | |
129 | ||
71fda00f | 130 | LIST_REMOVE(device_weights, c->blockio_device_weights, w); |
4ad49000 LP |
131 | free(w->path); |
132 | free(w); | |
133 | } | |
134 | ||
135 | void cgroup_context_free_blockio_device_bandwidth(CGroupContext *c, CGroupBlockIODeviceBandwidth *b) { | |
136 | assert(c); | |
8e274523 | 137 | assert(b); |
8e274523 | 138 | |
71fda00f | 139 | LIST_REMOVE(device_bandwidths, c->blockio_device_bandwidths, b); |
4ad49000 LP |
140 | free(b->path); |
141 | free(b); | |
142 | } | |
143 | ||
144 | void cgroup_context_done(CGroupContext *c) { | |
145 | assert(c); | |
146 | ||
13c31542 TH |
147 | while (c->io_device_weights) |
148 | cgroup_context_free_io_device_weight(c, c->io_device_weights); | |
149 | ||
150 | while (c->io_device_limits) | |
151 | cgroup_context_free_io_device_limit(c, c->io_device_limits); | |
152 | ||
4ad49000 LP |
153 | while (c->blockio_device_weights) |
154 | cgroup_context_free_blockio_device_weight(c, c->blockio_device_weights); | |
155 | ||
156 | while (c->blockio_device_bandwidths) | |
157 | cgroup_context_free_blockio_device_bandwidth(c, c->blockio_device_bandwidths); | |
158 | ||
159 | while (c->device_allow) | |
160 | cgroup_context_free_device_allow(c, c->device_allow); | |
6a48d82f DM |
161 | |
162 | c->ip_address_allow = ip_address_access_free_all(c->ip_address_allow); | |
163 | c->ip_address_deny = ip_address_access_free_all(c->ip_address_deny); | |
4ad49000 LP |
164 | } |
165 | ||
166 | void cgroup_context_dump(CGroupContext *c, FILE* f, const char *prefix) { | |
13c31542 TH |
167 | CGroupIODeviceLimit *il; |
168 | CGroupIODeviceWeight *iw; | |
4ad49000 LP |
169 | CGroupBlockIODeviceBandwidth *b; |
170 | CGroupBlockIODeviceWeight *w; | |
171 | CGroupDeviceAllow *a; | |
c21c9906 | 172 | IPAddressAccessItem *iaai; |
9a054909 | 173 | char u[FORMAT_TIMESPAN_MAX]; |
4ad49000 LP |
174 | |
175 | assert(c); | |
176 | assert(f); | |
177 | ||
178 | prefix = strempty(prefix); | |
179 | ||
180 | fprintf(f, | |
181 | "%sCPUAccounting=%s\n" | |
13c31542 | 182 | "%sIOAccounting=%s\n" |
4ad49000 LP |
183 | "%sBlockIOAccounting=%s\n" |
184 | "%sMemoryAccounting=%s\n" | |
d53d9474 | 185 | "%sTasksAccounting=%s\n" |
c21c9906 | 186 | "%sIPAccounting=%s\n" |
66ebf6c0 TH |
187 | "%sCPUWeight=%" PRIu64 "\n" |
188 | "%sStartupCPUWeight=%" PRIu64 "\n" | |
d53d9474 LP |
189 | "%sCPUShares=%" PRIu64 "\n" |
190 | "%sStartupCPUShares=%" PRIu64 "\n" | |
b2f8b02e | 191 | "%sCPUQuotaPerSecSec=%s\n" |
13c31542 TH |
192 | "%sIOWeight=%" PRIu64 "\n" |
193 | "%sStartupIOWeight=%" PRIu64 "\n" | |
d53d9474 LP |
194 | "%sBlockIOWeight=%" PRIu64 "\n" |
195 | "%sStartupBlockIOWeight=%" PRIu64 "\n" | |
48422635 | 196 | "%sMemoryMin=%" PRIu64 "\n" |
da4d897e TH |
197 | "%sMemoryLow=%" PRIu64 "\n" |
198 | "%sMemoryHigh=%" PRIu64 "\n" | |
199 | "%sMemoryMax=%" PRIu64 "\n" | |
96e131ea | 200 | "%sMemorySwapMax=%" PRIu64 "\n" |
4ad49000 | 201 | "%sMemoryLimit=%" PRIu64 "\n" |
03a7b521 | 202 | "%sTasksMax=%" PRIu64 "\n" |
a931ad47 LP |
203 | "%sDevicePolicy=%s\n" |
204 | "%sDelegate=%s\n", | |
4ad49000 | 205 | prefix, yes_no(c->cpu_accounting), |
13c31542 | 206 | prefix, yes_no(c->io_accounting), |
4ad49000 LP |
207 | prefix, yes_no(c->blockio_accounting), |
208 | prefix, yes_no(c->memory_accounting), | |
d53d9474 | 209 | prefix, yes_no(c->tasks_accounting), |
c21c9906 | 210 | prefix, yes_no(c->ip_accounting), |
66ebf6c0 TH |
211 | prefix, c->cpu_weight, |
212 | prefix, c->startup_cpu_weight, | |
4ad49000 | 213 | prefix, c->cpu_shares, |
95ae05c0 | 214 | prefix, c->startup_cpu_shares, |
b1d6dcf5 | 215 | prefix, format_timespan(u, sizeof(u), c->cpu_quota_per_sec_usec, 1), |
13c31542 TH |
216 | prefix, c->io_weight, |
217 | prefix, c->startup_io_weight, | |
4ad49000 | 218 | prefix, c->blockio_weight, |
95ae05c0 | 219 | prefix, c->startup_blockio_weight, |
48422635 | 220 | prefix, c->memory_min, |
da4d897e TH |
221 | prefix, c->memory_low, |
222 | prefix, c->memory_high, | |
223 | prefix, c->memory_max, | |
96e131ea | 224 | prefix, c->memory_swap_max, |
4ad49000 | 225 | prefix, c->memory_limit, |
03a7b521 | 226 | prefix, c->tasks_max, |
a931ad47 LP |
227 | prefix, cgroup_device_policy_to_string(c->device_policy), |
228 | prefix, yes_no(c->delegate)); | |
4ad49000 | 229 | |
02638280 LP |
230 | if (c->delegate) { |
231 | _cleanup_free_ char *t = NULL; | |
232 | ||
233 | (void) cg_mask_to_string(c->delegate_controllers, &t); | |
234 | ||
47a78d41 | 235 | fprintf(f, "%sDelegateControllers=%s\n", |
02638280 LP |
236 | prefix, |
237 | strempty(t)); | |
238 | } | |
239 | ||
4ad49000 LP |
240 | LIST_FOREACH(device_allow, a, c->device_allow) |
241 | fprintf(f, | |
242 | "%sDeviceAllow=%s %s%s%s\n", | |
243 | prefix, | |
244 | a->path, | |
245 | a->r ? "r" : "", a->w ? "w" : "", a->m ? "m" : ""); | |
246 | ||
13c31542 TH |
247 | LIST_FOREACH(device_weights, iw, c->io_device_weights) |
248 | fprintf(f, | |
249 | "%sIODeviceWeight=%s %" PRIu64, | |
250 | prefix, | |
251 | iw->path, | |
252 | iw->weight); | |
253 | ||
254 | LIST_FOREACH(device_limits, il, c->io_device_limits) { | |
255 | char buf[FORMAT_BYTES_MAX]; | |
9be57249 TH |
256 | CGroupIOLimitType type; |
257 | ||
258 | for (type = 0; type < _CGROUP_IO_LIMIT_TYPE_MAX; type++) | |
259 | if (il->limits[type] != cgroup_io_limit_defaults[type]) | |
260 | fprintf(f, | |
261 | "%s%s=%s %s\n", | |
262 | prefix, | |
263 | cgroup_io_limit_type_to_string(type), | |
264 | il->path, | |
265 | format_bytes(buf, sizeof(buf), il->limits[type])); | |
13c31542 TH |
266 | } |
267 | ||
4ad49000 LP |
268 | LIST_FOREACH(device_weights, w, c->blockio_device_weights) |
269 | fprintf(f, | |
d53d9474 | 270 | "%sBlockIODeviceWeight=%s %" PRIu64, |
4ad49000 LP |
271 | prefix, |
272 | w->path, | |
273 | w->weight); | |
274 | ||
275 | LIST_FOREACH(device_bandwidths, b, c->blockio_device_bandwidths) { | |
276 | char buf[FORMAT_BYTES_MAX]; | |
277 | ||
979d0311 TH |
278 | if (b->rbps != CGROUP_LIMIT_MAX) |
279 | fprintf(f, | |
280 | "%sBlockIOReadBandwidth=%s %s\n", | |
281 | prefix, | |
282 | b->path, | |
283 | format_bytes(buf, sizeof(buf), b->rbps)); | |
284 | if (b->wbps != CGROUP_LIMIT_MAX) | |
285 | fprintf(f, | |
286 | "%sBlockIOWriteBandwidth=%s %s\n", | |
287 | prefix, | |
288 | b->path, | |
289 | format_bytes(buf, sizeof(buf), b->wbps)); | |
4ad49000 | 290 | } |
c21c9906 LP |
291 | |
292 | LIST_FOREACH(items, iaai, c->ip_address_allow) { | |
293 | _cleanup_free_ char *k = NULL; | |
294 | ||
295 | (void) in_addr_to_string(iaai->family, &iaai->address, &k); | |
296 | fprintf(f, "%sIPAddressAllow=%s/%u\n", prefix, strnull(k), iaai->prefixlen); | |
297 | } | |
298 | ||
299 | LIST_FOREACH(items, iaai, c->ip_address_deny) { | |
300 | _cleanup_free_ char *k = NULL; | |
301 | ||
302 | (void) in_addr_to_string(iaai->family, &iaai->address, &k); | |
303 | fprintf(f, "%sIPAddressDeny=%s/%u\n", prefix, strnull(k), iaai->prefixlen); | |
304 | } | |
4ad49000 LP |
305 | } |
306 | ||
fd870bac YW |
307 | int cgroup_add_device_allow(CGroupContext *c, const char *dev, const char *mode) { |
308 | _cleanup_free_ CGroupDeviceAllow *a = NULL; | |
309 | _cleanup_free_ char *d = NULL; | |
310 | ||
311 | assert(c); | |
312 | assert(dev); | |
313 | assert(isempty(mode) || in_charset(mode, "rwm")); | |
314 | ||
315 | a = new(CGroupDeviceAllow, 1); | |
316 | if (!a) | |
317 | return -ENOMEM; | |
318 | ||
319 | d = strdup(dev); | |
320 | if (!d) | |
321 | return -ENOMEM; | |
322 | ||
323 | *a = (CGroupDeviceAllow) { | |
324 | .path = TAKE_PTR(d), | |
325 | .r = isempty(mode) || !!strchr(mode, 'r'), | |
326 | .w = isempty(mode) || !!strchr(mode, 'w'), | |
327 | .m = isempty(mode) || !!strchr(mode, 'm'), | |
328 | }; | |
329 | ||
330 | LIST_PREPEND(device_allow, c->device_allow, a); | |
331 | TAKE_PTR(a); | |
332 | ||
333 | return 0; | |
334 | } | |
335 | ||
45c2e068 | 336 | static int lookup_block_device(const char *p, dev_t *ret) { |
4ad49000 | 337 | struct stat st; |
45c2e068 | 338 | int r; |
4ad49000 LP |
339 | |
340 | assert(p); | |
45c2e068 | 341 | assert(ret); |
4ad49000 | 342 | |
b1c05b98 | 343 | if (stat(p, &st) < 0) |
45c2e068 | 344 | return log_warning_errno(errno, "Couldn't stat device '%s': %m", p); |
8e274523 | 345 | |
4ad49000 | 346 | if (S_ISBLK(st.st_mode)) |
45c2e068 LP |
347 | *ret = st.st_rdev; |
348 | else if (major(st.st_dev) != 0) | |
349 | *ret = st.st_dev; /* If this is not a device node then use the block device this file is stored on */ | |
350 | else { | |
351 | /* If this is btrfs, getting the backing block device is a bit harder */ | |
352 | r = btrfs_get_block_device(p, ret); | |
353 | if (r < 0 && r != -ENOTTY) | |
354 | return log_warning_errno(r, "Failed to determine block device backing btrfs file system '%s': %m", p); | |
355 | if (r == -ENOTTY) { | |
356 | log_warning("'%s' is not a block device node, and file system block device cannot be determined or is not local.", p); | |
357 | return -ENODEV; | |
358 | } | |
4ad49000 | 359 | } |
8e274523 | 360 | |
45c2e068 LP |
361 | /* If this is a LUKS device, try to get the originating block device */ |
362 | (void) block_get_originating(*ret, ret); | |
363 | ||
364 | /* If this is a partition, try to get the originating block device */ | |
365 | (void) block_get_whole_disk(*ret, ret); | |
8e274523 | 366 | return 0; |
8e274523 LP |
367 | } |
368 | ||
4ad49000 LP |
369 | static int whitelist_device(const char *path, const char *node, const char *acc) { |
370 | char buf[2+DECIMAL_STR_MAX(dev_t)*2+2+4]; | |
371 | struct stat st; | |
b200489b | 372 | bool ignore_notfound; |
8c6db833 | 373 | int r; |
8e274523 | 374 | |
4ad49000 LP |
375 | assert(path); |
376 | assert(acc); | |
8e274523 | 377 | |
b200489b DR |
378 | if (node[0] == '-') { |
379 | /* Non-existent paths starting with "-" must be silently ignored */ | |
380 | node++; | |
381 | ignore_notfound = true; | |
382 | } else | |
383 | ignore_notfound = false; | |
384 | ||
4ad49000 | 385 | if (stat(node, &st) < 0) { |
b200489b | 386 | if (errno == ENOENT && ignore_notfound) |
e7330dfe DP |
387 | return 0; |
388 | ||
389 | return log_warning_errno(errno, "Couldn't stat device %s: %m", node); | |
4ad49000 LP |
390 | } |
391 | ||
392 | if (!S_ISCHR(st.st_mode) && !S_ISBLK(st.st_mode)) { | |
393 | log_warning("%s is not a device.", node); | |
394 | return -ENODEV; | |
395 | } | |
396 | ||
397 | sprintf(buf, | |
398 | "%c %u:%u %s", | |
399 | S_ISCHR(st.st_mode) ? 'c' : 'b', | |
400 | major(st.st_rdev), minor(st.st_rdev), | |
401 | acc); | |
402 | ||
403 | r = cg_set_attribute("devices", path, "devices.allow", buf); | |
1aeab12b | 404 | if (r < 0) |
077ba06e | 405 | log_full_errno(IN_SET(r, -ENOENT, -EROFS, -EINVAL, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, |
714e2e1d | 406 | "Failed to set devices.allow on %s: %m", path); |
4ad49000 LP |
407 | |
408 | return r; | |
8e274523 LP |
409 | } |
410 | ||
90060676 LP |
411 | static int whitelist_major(const char *path, const char *name, char type, const char *acc) { |
412 | _cleanup_fclose_ FILE *f = NULL; | |
413 | char line[LINE_MAX]; | |
414 | bool good = false; | |
415 | int r; | |
416 | ||
417 | assert(path); | |
418 | assert(acc); | |
4c701096 | 419 | assert(IN_SET(type, 'b', 'c')); |
90060676 LP |
420 | |
421 | f = fopen("/proc/devices", "re"); | |
4a62c710 MS |
422 | if (!f) |
423 | return log_warning_errno(errno, "Cannot open /proc/devices to resolve %s (%c): %m", name, type); | |
90060676 LP |
424 | |
425 | FOREACH_LINE(line, f, goto fail) { | |
426 | char buf[2+DECIMAL_STR_MAX(unsigned)+3+4], *p, *w; | |
427 | unsigned maj; | |
428 | ||
429 | truncate_nl(line); | |
430 | ||
431 | if (type == 'c' && streq(line, "Character devices:")) { | |
432 | good = true; | |
433 | continue; | |
434 | } | |
435 | ||
436 | if (type == 'b' && streq(line, "Block devices:")) { | |
437 | good = true; | |
438 | continue; | |
439 | } | |
440 | ||
441 | if (isempty(line)) { | |
442 | good = false; | |
443 | continue; | |
444 | } | |
445 | ||
446 | if (!good) | |
447 | continue; | |
448 | ||
449 | p = strstrip(line); | |
450 | ||
451 | w = strpbrk(p, WHITESPACE); | |
452 | if (!w) | |
453 | continue; | |
454 | *w = 0; | |
455 | ||
456 | r = safe_atou(p, &maj); | |
457 | if (r < 0) | |
458 | continue; | |
459 | if (maj <= 0) | |
460 | continue; | |
461 | ||
462 | w++; | |
463 | w += strspn(w, WHITESPACE); | |
e41969e3 LP |
464 | |
465 | if (fnmatch(name, w, 0) != 0) | |
90060676 LP |
466 | continue; |
467 | ||
468 | sprintf(buf, | |
469 | "%c %u:* %s", | |
470 | type, | |
471 | maj, | |
472 | acc); | |
473 | ||
474 | r = cg_set_attribute("devices", path, "devices.allow", buf); | |
1aeab12b | 475 | if (r < 0) |
077ba06e | 476 | log_full_errno(IN_SET(r, -ENOENT, -EROFS, -EINVAL, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, |
714e2e1d | 477 | "Failed to set devices.allow on %s: %m", path); |
90060676 LP |
478 | } |
479 | ||
480 | return 0; | |
481 | ||
482 | fail: | |
25f027c5 | 483 | return log_warning_errno(errno, "Failed to read /proc/devices: %m"); |
90060676 LP |
484 | } |
485 | ||
66ebf6c0 TH |
486 | static bool cgroup_context_has_cpu_weight(CGroupContext *c) { |
487 | return c->cpu_weight != CGROUP_WEIGHT_INVALID || | |
488 | c->startup_cpu_weight != CGROUP_WEIGHT_INVALID; | |
489 | } | |
490 | ||
491 | static bool cgroup_context_has_cpu_shares(CGroupContext *c) { | |
492 | return c->cpu_shares != CGROUP_CPU_SHARES_INVALID || | |
493 | c->startup_cpu_shares != CGROUP_CPU_SHARES_INVALID; | |
494 | } | |
495 | ||
496 | static uint64_t cgroup_context_cpu_weight(CGroupContext *c, ManagerState state) { | |
497 | if (IN_SET(state, MANAGER_STARTING, MANAGER_INITIALIZING) && | |
498 | c->startup_cpu_weight != CGROUP_WEIGHT_INVALID) | |
499 | return c->startup_cpu_weight; | |
500 | else if (c->cpu_weight != CGROUP_WEIGHT_INVALID) | |
501 | return c->cpu_weight; | |
502 | else | |
503 | return CGROUP_WEIGHT_DEFAULT; | |
504 | } | |
505 | ||
506 | static uint64_t cgroup_context_cpu_shares(CGroupContext *c, ManagerState state) { | |
507 | if (IN_SET(state, MANAGER_STARTING, MANAGER_INITIALIZING) && | |
508 | c->startup_cpu_shares != CGROUP_CPU_SHARES_INVALID) | |
509 | return c->startup_cpu_shares; | |
510 | else if (c->cpu_shares != CGROUP_CPU_SHARES_INVALID) | |
511 | return c->cpu_shares; | |
512 | else | |
513 | return CGROUP_CPU_SHARES_DEFAULT; | |
514 | } | |
515 | ||
516 | static void cgroup_apply_unified_cpu_config(Unit *u, uint64_t weight, uint64_t quota) { | |
517 | char buf[MAX(DECIMAL_STR_MAX(uint64_t) + 1, (DECIMAL_STR_MAX(usec_t) + 1) * 2)]; | |
518 | int r; | |
519 | ||
520 | xsprintf(buf, "%" PRIu64 "\n", weight); | |
521 | r = cg_set_attribute("cpu", u->cgroup_path, "cpu.weight", buf); | |
522 | if (r < 0) | |
523 | log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, | |
524 | "Failed to set cpu.weight: %m"); | |
525 | ||
526 | if (quota != USEC_INFINITY) | |
527 | xsprintf(buf, USEC_FMT " " USEC_FMT "\n", | |
528 | quota * CGROUP_CPU_QUOTA_PERIOD_USEC / USEC_PER_SEC, CGROUP_CPU_QUOTA_PERIOD_USEC); | |
529 | else | |
530 | xsprintf(buf, "max " USEC_FMT "\n", CGROUP_CPU_QUOTA_PERIOD_USEC); | |
531 | ||
532 | r = cg_set_attribute("cpu", u->cgroup_path, "cpu.max", buf); | |
533 | ||
534 | if (r < 0) | |
535 | log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, | |
536 | "Failed to set cpu.max: %m"); | |
537 | } | |
538 | ||
539 | static void cgroup_apply_legacy_cpu_config(Unit *u, uint64_t shares, uint64_t quota) { | |
540 | char buf[MAX(DECIMAL_STR_MAX(uint64_t), DECIMAL_STR_MAX(usec_t)) + 1]; | |
541 | int r; | |
542 | ||
543 | xsprintf(buf, "%" PRIu64 "\n", shares); | |
544 | r = cg_set_attribute("cpu", u->cgroup_path, "cpu.shares", buf); | |
545 | if (r < 0) | |
546 | log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, | |
547 | "Failed to set cpu.shares: %m"); | |
548 | ||
549 | xsprintf(buf, USEC_FMT "\n", CGROUP_CPU_QUOTA_PERIOD_USEC); | |
550 | r = cg_set_attribute("cpu", u->cgroup_path, "cpu.cfs_period_us", buf); | |
551 | if (r < 0) | |
552 | log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, | |
553 | "Failed to set cpu.cfs_period_us: %m"); | |
554 | ||
555 | if (quota != USEC_INFINITY) { | |
556 | xsprintf(buf, USEC_FMT "\n", quota * CGROUP_CPU_QUOTA_PERIOD_USEC / USEC_PER_SEC); | |
557 | r = cg_set_attribute("cpu", u->cgroup_path, "cpu.cfs_quota_us", buf); | |
558 | } else | |
559 | r = cg_set_attribute("cpu", u->cgroup_path, "cpu.cfs_quota_us", "-1"); | |
560 | if (r < 0) | |
561 | log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, | |
562 | "Failed to set cpu.cfs_quota_us: %m"); | |
563 | } | |
564 | ||
565 | static uint64_t cgroup_cpu_shares_to_weight(uint64_t shares) { | |
566 | return CLAMP(shares * CGROUP_WEIGHT_DEFAULT / CGROUP_CPU_SHARES_DEFAULT, | |
567 | CGROUP_WEIGHT_MIN, CGROUP_WEIGHT_MAX); | |
568 | } | |
569 | ||
570 | static uint64_t cgroup_cpu_weight_to_shares(uint64_t weight) { | |
571 | return CLAMP(weight * CGROUP_CPU_SHARES_DEFAULT / CGROUP_WEIGHT_DEFAULT, | |
572 | CGROUP_CPU_SHARES_MIN, CGROUP_CPU_SHARES_MAX); | |
573 | } | |
574 | ||
508c45da | 575 | static bool cgroup_context_has_io_config(CGroupContext *c) { |
538b4852 TH |
576 | return c->io_accounting || |
577 | c->io_weight != CGROUP_WEIGHT_INVALID || | |
578 | c->startup_io_weight != CGROUP_WEIGHT_INVALID || | |
579 | c->io_device_weights || | |
580 | c->io_device_limits; | |
581 | } | |
582 | ||
508c45da | 583 | static bool cgroup_context_has_blockio_config(CGroupContext *c) { |
538b4852 TH |
584 | return c->blockio_accounting || |
585 | c->blockio_weight != CGROUP_BLKIO_WEIGHT_INVALID || | |
586 | c->startup_blockio_weight != CGROUP_BLKIO_WEIGHT_INVALID || | |
587 | c->blockio_device_weights || | |
588 | c->blockio_device_bandwidths; | |
589 | } | |
590 | ||
508c45da | 591 | static uint64_t cgroup_context_io_weight(CGroupContext *c, ManagerState state) { |
64faf04c TH |
592 | if (IN_SET(state, MANAGER_STARTING, MANAGER_INITIALIZING) && |
593 | c->startup_io_weight != CGROUP_WEIGHT_INVALID) | |
594 | return c->startup_io_weight; | |
595 | else if (c->io_weight != CGROUP_WEIGHT_INVALID) | |
596 | return c->io_weight; | |
597 | else | |
598 | return CGROUP_WEIGHT_DEFAULT; | |
599 | } | |
600 | ||
508c45da | 601 | static uint64_t cgroup_context_blkio_weight(CGroupContext *c, ManagerState state) { |
64faf04c TH |
602 | if (IN_SET(state, MANAGER_STARTING, MANAGER_INITIALIZING) && |
603 | c->startup_blockio_weight != CGROUP_BLKIO_WEIGHT_INVALID) | |
604 | return c->startup_blockio_weight; | |
605 | else if (c->blockio_weight != CGROUP_BLKIO_WEIGHT_INVALID) | |
606 | return c->blockio_weight; | |
607 | else | |
608 | return CGROUP_BLKIO_WEIGHT_DEFAULT; | |
609 | } | |
610 | ||
508c45da | 611 | static uint64_t cgroup_weight_blkio_to_io(uint64_t blkio_weight) { |
538b4852 TH |
612 | return CLAMP(blkio_weight * CGROUP_WEIGHT_DEFAULT / CGROUP_BLKIO_WEIGHT_DEFAULT, |
613 | CGROUP_WEIGHT_MIN, CGROUP_WEIGHT_MAX); | |
614 | } | |
615 | ||
508c45da | 616 | static uint64_t cgroup_weight_io_to_blkio(uint64_t io_weight) { |
538b4852 TH |
617 | return CLAMP(io_weight * CGROUP_BLKIO_WEIGHT_DEFAULT / CGROUP_WEIGHT_DEFAULT, |
618 | CGROUP_BLKIO_WEIGHT_MIN, CGROUP_BLKIO_WEIGHT_MAX); | |
619 | } | |
620 | ||
f29ff115 | 621 | static void cgroup_apply_io_device_weight(Unit *u, const char *dev_path, uint64_t io_weight) { |
64faf04c TH |
622 | char buf[DECIMAL_STR_MAX(dev_t)*2+2+DECIMAL_STR_MAX(uint64_t)+1]; |
623 | dev_t dev; | |
624 | int r; | |
625 | ||
626 | r = lookup_block_device(dev_path, &dev); | |
627 | if (r < 0) | |
628 | return; | |
629 | ||
630 | xsprintf(buf, "%u:%u %" PRIu64 "\n", major(dev), minor(dev), io_weight); | |
f29ff115 | 631 | r = cg_set_attribute("io", u->cgroup_path, "io.weight", buf); |
64faf04c | 632 | if (r < 0) |
f29ff115 TH |
633 | log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, |
634 | "Failed to set io.weight: %m"); | |
64faf04c TH |
635 | } |
636 | ||
f29ff115 | 637 | static void cgroup_apply_blkio_device_weight(Unit *u, const char *dev_path, uint64_t blkio_weight) { |
64faf04c TH |
638 | char buf[DECIMAL_STR_MAX(dev_t)*2+2+DECIMAL_STR_MAX(uint64_t)+1]; |
639 | dev_t dev; | |
640 | int r; | |
641 | ||
642 | r = lookup_block_device(dev_path, &dev); | |
643 | if (r < 0) | |
644 | return; | |
645 | ||
646 | xsprintf(buf, "%u:%u %" PRIu64 "\n", major(dev), minor(dev), blkio_weight); | |
f29ff115 | 647 | r = cg_set_attribute("blkio", u->cgroup_path, "blkio.weight_device", buf); |
64faf04c | 648 | if (r < 0) |
f29ff115 TH |
649 | log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, |
650 | "Failed to set blkio.weight_device: %m"); | |
64faf04c TH |
651 | } |
652 | ||
17ae2780 | 653 | static void cgroup_apply_io_device_limit(Unit *u, const char *dev_path, uint64_t *limits) { |
64faf04c TH |
654 | char limit_bufs[_CGROUP_IO_LIMIT_TYPE_MAX][DECIMAL_STR_MAX(uint64_t)]; |
655 | char buf[DECIMAL_STR_MAX(dev_t)*2+2+(6+DECIMAL_STR_MAX(uint64_t)+1)*4]; | |
656 | CGroupIOLimitType type; | |
657 | dev_t dev; | |
64faf04c TH |
658 | int r; |
659 | ||
660 | r = lookup_block_device(dev_path, &dev); | |
661 | if (r < 0) | |
17ae2780 | 662 | return; |
64faf04c | 663 | |
17ae2780 LP |
664 | for (type = 0; type < _CGROUP_IO_LIMIT_TYPE_MAX; type++) |
665 | if (limits[type] != cgroup_io_limit_defaults[type]) | |
64faf04c | 666 | xsprintf(limit_bufs[type], "%" PRIu64, limits[type]); |
17ae2780 | 667 | else |
64faf04c | 668 | xsprintf(limit_bufs[type], "%s", limits[type] == CGROUP_LIMIT_MAX ? "max" : "0"); |
64faf04c TH |
669 | |
670 | xsprintf(buf, "%u:%u rbps=%s wbps=%s riops=%s wiops=%s\n", major(dev), minor(dev), | |
671 | limit_bufs[CGROUP_IO_RBPS_MAX], limit_bufs[CGROUP_IO_WBPS_MAX], | |
672 | limit_bufs[CGROUP_IO_RIOPS_MAX], limit_bufs[CGROUP_IO_WIOPS_MAX]); | |
f29ff115 | 673 | r = cg_set_attribute("io", u->cgroup_path, "io.max", buf); |
64faf04c | 674 | if (r < 0) |
f29ff115 TH |
675 | log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, |
676 | "Failed to set io.max: %m"); | |
64faf04c TH |
677 | } |
678 | ||
17ae2780 | 679 | static void cgroup_apply_blkio_device_limit(Unit *u, const char *dev_path, uint64_t rbps, uint64_t wbps) { |
64faf04c TH |
680 | char buf[DECIMAL_STR_MAX(dev_t)*2+2+DECIMAL_STR_MAX(uint64_t)+1]; |
681 | dev_t dev; | |
64faf04c TH |
682 | int r; |
683 | ||
684 | r = lookup_block_device(dev_path, &dev); | |
685 | if (r < 0) | |
17ae2780 | 686 | return; |
64faf04c | 687 | |
64faf04c | 688 | sprintf(buf, "%u:%u %" PRIu64 "\n", major(dev), minor(dev), rbps); |
f29ff115 | 689 | r = cg_set_attribute("blkio", u->cgroup_path, "blkio.throttle.read_bps_device", buf); |
64faf04c | 690 | if (r < 0) |
f29ff115 TH |
691 | log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, |
692 | "Failed to set blkio.throttle.read_bps_device: %m"); | |
64faf04c | 693 | |
64faf04c | 694 | sprintf(buf, "%u:%u %" PRIu64 "\n", major(dev), minor(dev), wbps); |
f29ff115 | 695 | r = cg_set_attribute("blkio", u->cgroup_path, "blkio.throttle.write_bps_device", buf); |
64faf04c | 696 | if (r < 0) |
f29ff115 TH |
697 | log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, |
698 | "Failed to set blkio.throttle.write_bps_device: %m"); | |
64faf04c TH |
699 | } |
700 | ||
da4d897e | 701 | static bool cgroup_context_has_unified_memory_config(CGroupContext *c) { |
48422635 | 702 | 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 |
703 | } |
704 | ||
f29ff115 | 705 | static void cgroup_apply_unified_memory_limit(Unit *u, const char *file, uint64_t v) { |
da4d897e TH |
706 | char buf[DECIMAL_STR_MAX(uint64_t) + 1] = "max"; |
707 | int r; | |
708 | ||
709 | if (v != CGROUP_LIMIT_MAX) | |
710 | xsprintf(buf, "%" PRIu64 "\n", v); | |
711 | ||
f29ff115 | 712 | r = cg_set_attribute("memory", u->cgroup_path, file, buf); |
da4d897e | 713 | if (r < 0) |
f29ff115 TH |
714 | log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, |
715 | "Failed to set %s: %m", file); | |
da4d897e TH |
716 | } |
717 | ||
0f2d84d2 | 718 | static void cgroup_apply_firewall(Unit *u) { |
0f2d84d2 LP |
719 | assert(u); |
720 | ||
acf7f253 | 721 | /* Best-effort: let's apply IP firewalling and/or accounting if that's enabled */ |
906c06f6 | 722 | |
acf7f253 | 723 | if (bpf_firewall_compile(u) < 0) |
906c06f6 DM |
724 | return; |
725 | ||
726 | (void) bpf_firewall_install(u); | |
906c06f6 DM |
727 | } |
728 | ||
729 | static void cgroup_context_apply( | |
730 | Unit *u, | |
731 | CGroupMask apply_mask, | |
732 | bool apply_bpf, | |
733 | ManagerState state) { | |
734 | ||
f29ff115 TH |
735 | const char *path; |
736 | CGroupContext *c; | |
01efdf13 | 737 | bool is_root; |
4ad49000 LP |
738 | int r; |
739 | ||
f29ff115 TH |
740 | assert(u); |
741 | ||
906c06f6 DM |
742 | /* Nothing to do? Exit early! */ |
743 | if (apply_mask == 0 && !apply_bpf) | |
4ad49000 | 744 | return; |
8e274523 | 745 | |
f3725e64 LP |
746 | /* Some cgroup attributes are not supported on the root cgroup, hence silently ignore */ |
747 | is_root = unit_has_root_cgroup(u); | |
748 | ||
749 | assert_se(c = unit_get_cgroup_context(u)); | |
750 | assert_se(path = u->cgroup_path); | |
751 | ||
752 | if (is_root) /* Make sure we don't try to display messages with an empty path. */ | |
6da13913 | 753 | path = "/"; |
01efdf13 | 754 | |
714e2e1d LP |
755 | /* We generally ignore errors caused by read-only mounted |
756 | * cgroup trees (assuming we are running in a container then), | |
757 | * and missing cgroups, i.e. EROFS and ENOENT. */ | |
758 | ||
906c06f6 DM |
759 | if ((apply_mask & CGROUP_MASK_CPU) && !is_root) { |
760 | bool has_weight, has_shares; | |
761 | ||
762 | has_weight = cgroup_context_has_cpu_weight(c); | |
763 | has_shares = cgroup_context_has_cpu_shares(c); | |
8e274523 | 764 | |
b4cccbc1 | 765 | if (cg_all_unified() > 0) { |
66ebf6c0 | 766 | uint64_t weight; |
b2f8b02e | 767 | |
66ebf6c0 TH |
768 | if (has_weight) |
769 | weight = cgroup_context_cpu_weight(c, state); | |
770 | else if (has_shares) { | |
771 | uint64_t shares = cgroup_context_cpu_shares(c, state); | |
b2f8b02e | 772 | |
66ebf6c0 TH |
773 | weight = cgroup_cpu_shares_to_weight(shares); |
774 | ||
775 | log_cgroup_compat(u, "Applying [Startup]CpuShares %" PRIu64 " as [Startup]CpuWeight %" PRIu64 " on %s", | |
776 | shares, weight, path); | |
777 | } else | |
778 | weight = CGROUP_WEIGHT_DEFAULT; | |
779 | ||
780 | cgroup_apply_unified_cpu_config(u, weight, c->cpu_quota_per_sec_usec); | |
781 | } else { | |
782 | uint64_t shares; | |
783 | ||
7d862ab8 | 784 | if (has_weight) { |
66ebf6c0 TH |
785 | uint64_t weight = cgroup_context_cpu_weight(c, state); |
786 | ||
787 | shares = cgroup_cpu_weight_to_shares(weight); | |
788 | ||
789 | log_cgroup_compat(u, "Applying [Startup]CpuWeight %" PRIu64 " as [Startup]CpuShares %" PRIu64 " on %s", | |
790 | weight, shares, path); | |
7d862ab8 TH |
791 | } else if (has_shares) |
792 | shares = cgroup_context_cpu_shares(c, state); | |
793 | else | |
66ebf6c0 TH |
794 | shares = CGROUP_CPU_SHARES_DEFAULT; |
795 | ||
796 | cgroup_apply_legacy_cpu_config(u, shares, c->cpu_quota_per_sec_usec); | |
797 | } | |
4ad49000 LP |
798 | } |
799 | ||
906c06f6 | 800 | if (apply_mask & CGROUP_MASK_IO) { |
538b4852 TH |
801 | bool has_io = cgroup_context_has_io_config(c); |
802 | bool has_blockio = cgroup_context_has_blockio_config(c); | |
13c31542 TH |
803 | |
804 | if (!is_root) { | |
64faf04c TH |
805 | char buf[8+DECIMAL_STR_MAX(uint64_t)+1]; |
806 | uint64_t weight; | |
13c31542 | 807 | |
538b4852 TH |
808 | if (has_io) |
809 | weight = cgroup_context_io_weight(c, state); | |
128fadc9 TH |
810 | else if (has_blockio) { |
811 | uint64_t blkio_weight = cgroup_context_blkio_weight(c, state); | |
812 | ||
813 | weight = cgroup_weight_blkio_to_io(blkio_weight); | |
814 | ||
815 | log_cgroup_compat(u, "Applying [Startup]BlockIOWeight %" PRIu64 " as [Startup]IOWeight %" PRIu64, | |
816 | blkio_weight, weight); | |
817 | } else | |
538b4852 | 818 | weight = CGROUP_WEIGHT_DEFAULT; |
13c31542 TH |
819 | |
820 | xsprintf(buf, "default %" PRIu64 "\n", weight); | |
821 | r = cg_set_attribute("io", path, "io.weight", buf); | |
822 | if (r < 0) | |
f29ff115 TH |
823 | log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, |
824 | "Failed to set io.weight: %m"); | |
13c31542 | 825 | |
538b4852 TH |
826 | if (has_io) { |
827 | CGroupIODeviceWeight *w; | |
828 | ||
829 | /* FIXME: no way to reset this list */ | |
830 | LIST_FOREACH(device_weights, w, c->io_device_weights) | |
f29ff115 | 831 | cgroup_apply_io_device_weight(u, w->path, w->weight); |
538b4852 TH |
832 | } else if (has_blockio) { |
833 | CGroupBlockIODeviceWeight *w; | |
834 | ||
835 | /* FIXME: no way to reset this list */ | |
128fadc9 TH |
836 | LIST_FOREACH(device_weights, w, c->blockio_device_weights) { |
837 | weight = cgroup_weight_blkio_to_io(w->weight); | |
838 | ||
839 | log_cgroup_compat(u, "Applying BlockIODeviceWeight %" PRIu64 " as IODeviceWeight %" PRIu64 " for %s", | |
840 | w->weight, weight, w->path); | |
841 | ||
842 | cgroup_apply_io_device_weight(u, w->path, weight); | |
843 | } | |
538b4852 | 844 | } |
13c31542 TH |
845 | } |
846 | ||
64faf04c | 847 | /* Apply limits and free ones without config. */ |
538b4852 | 848 | if (has_io) { |
17ae2780 LP |
849 | CGroupIODeviceLimit *l; |
850 | ||
851 | LIST_FOREACH(device_limits, l, c->io_device_limits) | |
852 | cgroup_apply_io_device_limit(u, l->path, l->limits); | |
538b4852 | 853 | |
538b4852 | 854 | } else if (has_blockio) { |
17ae2780 | 855 | CGroupBlockIODeviceBandwidth *b; |
538b4852 | 856 | |
17ae2780 | 857 | LIST_FOREACH(device_bandwidths, b, c->blockio_device_bandwidths) { |
538b4852 TH |
858 | uint64_t limits[_CGROUP_IO_LIMIT_TYPE_MAX]; |
859 | CGroupIOLimitType type; | |
860 | ||
861 | for (type = 0; type < _CGROUP_IO_LIMIT_TYPE_MAX; type++) | |
862 | limits[type] = cgroup_io_limit_defaults[type]; | |
863 | ||
864 | limits[CGROUP_IO_RBPS_MAX] = b->rbps; | |
865 | limits[CGROUP_IO_WBPS_MAX] = b->wbps; | |
866 | ||
128fadc9 TH |
867 | log_cgroup_compat(u, "Applying BlockIO{Read|Write}Bandwidth %" PRIu64 " %" PRIu64 " as IO{Read|Write}BandwidthMax for %s", |
868 | b->rbps, b->wbps, b->path); | |
869 | ||
17ae2780 | 870 | cgroup_apply_io_device_limit(u, b->path, limits); |
538b4852 | 871 | } |
13c31542 TH |
872 | } |
873 | } | |
874 | ||
906c06f6 | 875 | if (apply_mask & CGROUP_MASK_BLKIO) { |
538b4852 TH |
876 | bool has_io = cgroup_context_has_io_config(c); |
877 | bool has_blockio = cgroup_context_has_blockio_config(c); | |
4ad49000 | 878 | |
01efdf13 | 879 | if (!is_root) { |
64faf04c TH |
880 | char buf[DECIMAL_STR_MAX(uint64_t)+1]; |
881 | uint64_t weight; | |
64faf04c | 882 | |
7d862ab8 | 883 | if (has_io) { |
128fadc9 TH |
884 | uint64_t io_weight = cgroup_context_io_weight(c, state); |
885 | ||
538b4852 | 886 | weight = cgroup_weight_io_to_blkio(cgroup_context_io_weight(c, state)); |
128fadc9 TH |
887 | |
888 | log_cgroup_compat(u, "Applying [Startup]IOWeight %" PRIu64 " as [Startup]BlockIOWeight %" PRIu64, | |
889 | io_weight, weight); | |
7d862ab8 TH |
890 | } else if (has_blockio) |
891 | weight = cgroup_context_blkio_weight(c, state); | |
892 | else | |
538b4852 | 893 | weight = CGROUP_BLKIO_WEIGHT_DEFAULT; |
64faf04c TH |
894 | |
895 | xsprintf(buf, "%" PRIu64 "\n", weight); | |
01efdf13 | 896 | r = cg_set_attribute("blkio", path, "blkio.weight", buf); |
1aeab12b | 897 | if (r < 0) |
f29ff115 TH |
898 | log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, |
899 | "Failed to set blkio.weight: %m"); | |
4ad49000 | 900 | |
7d862ab8 | 901 | if (has_io) { |
538b4852 TH |
902 | CGroupIODeviceWeight *w; |
903 | ||
904 | /* FIXME: no way to reset this list */ | |
128fadc9 TH |
905 | LIST_FOREACH(device_weights, w, c->io_device_weights) { |
906 | weight = cgroup_weight_io_to_blkio(w->weight); | |
907 | ||
908 | log_cgroup_compat(u, "Applying IODeviceWeight %" PRIu64 " as BlockIODeviceWeight %" PRIu64 " for %s", | |
909 | w->weight, weight, w->path); | |
910 | ||
911 | cgroup_apply_blkio_device_weight(u, w->path, weight); | |
912 | } | |
7d862ab8 TH |
913 | } else if (has_blockio) { |
914 | CGroupBlockIODeviceWeight *w; | |
915 | ||
916 | /* FIXME: no way to reset this list */ | |
917 | LIST_FOREACH(device_weights, w, c->blockio_device_weights) | |
918 | cgroup_apply_blkio_device_weight(u, w->path, w->weight); | |
538b4852 | 919 | } |
4ad49000 LP |
920 | } |
921 | ||
64faf04c | 922 | /* Apply limits and free ones without config. */ |
7d862ab8 | 923 | if (has_io) { |
17ae2780 | 924 | CGroupIODeviceLimit *l; |
538b4852 | 925 | |
17ae2780 | 926 | LIST_FOREACH(device_limits, l, c->io_device_limits) { |
128fadc9 TH |
927 | log_cgroup_compat(u, "Applying IO{Read|Write}Bandwidth %" PRIu64 " %" PRIu64 " as BlockIO{Read|Write}BandwidthMax for %s", |
928 | l->limits[CGROUP_IO_RBPS_MAX], l->limits[CGROUP_IO_WBPS_MAX], l->path); | |
929 | ||
17ae2780 | 930 | cgroup_apply_blkio_device_limit(u, l->path, l->limits[CGROUP_IO_RBPS_MAX], l->limits[CGROUP_IO_WBPS_MAX]); |
538b4852 | 931 | } |
7d862ab8 | 932 | } else if (has_blockio) { |
17ae2780 | 933 | CGroupBlockIODeviceBandwidth *b; |
7d862ab8 | 934 | |
17ae2780 LP |
935 | LIST_FOREACH(device_bandwidths, b, c->blockio_device_bandwidths) |
936 | cgroup_apply_blkio_device_limit(u, b->path, b->rbps, b->wbps); | |
d686d8a9 | 937 | } |
8e274523 LP |
938 | } |
939 | ||
906c06f6 | 940 | if ((apply_mask & CGROUP_MASK_MEMORY) && !is_root) { |
b4cccbc1 LP |
941 | if (cg_all_unified() > 0) { |
942 | uint64_t max, swap_max = CGROUP_LIMIT_MAX; | |
efdb0237 | 943 | |
96e131ea | 944 | if (cgroup_context_has_unified_memory_config(c)) { |
da4d897e | 945 | max = c->memory_max; |
96e131ea WC |
946 | swap_max = c->memory_swap_max; |
947 | } else { | |
da4d897e | 948 | max = c->memory_limit; |
efdb0237 | 949 | |
128fadc9 TH |
950 | if (max != CGROUP_LIMIT_MAX) |
951 | log_cgroup_compat(u, "Applying MemoryLimit %" PRIu64 " as MemoryMax", max); | |
952 | } | |
953 | ||
48422635 | 954 | cgroup_apply_unified_memory_limit(u, "memory.min", c->memory_min); |
f29ff115 TH |
955 | cgroup_apply_unified_memory_limit(u, "memory.low", c->memory_low); |
956 | cgroup_apply_unified_memory_limit(u, "memory.high", c->memory_high); | |
957 | cgroup_apply_unified_memory_limit(u, "memory.max", max); | |
96e131ea | 958 | cgroup_apply_unified_memory_limit(u, "memory.swap.max", swap_max); |
efdb0237 | 959 | } else { |
da4d897e | 960 | char buf[DECIMAL_STR_MAX(uint64_t) + 1]; |
7d862ab8 | 961 | uint64_t val; |
da4d897e | 962 | |
7d862ab8 | 963 | if (cgroup_context_has_unified_memory_config(c)) { |
78a4ee59 | 964 | val = c->memory_max; |
7d862ab8 TH |
965 | log_cgroup_compat(u, "Applying MemoryMax %" PRIi64 " as MemoryLimit", val); |
966 | } else | |
967 | val = c->memory_limit; | |
128fadc9 | 968 | |
78a4ee59 DM |
969 | if (val == CGROUP_LIMIT_MAX) |
970 | strncpy(buf, "-1\n", sizeof(buf)); | |
971 | else | |
972 | xsprintf(buf, "%" PRIu64 "\n", val); | |
973 | ||
da4d897e TH |
974 | r = cg_set_attribute("memory", path, "memory.limit_in_bytes", buf); |
975 | if (r < 0) | |
f29ff115 TH |
976 | log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, |
977 | "Failed to set memory.limit_in_bytes: %m"); | |
da4d897e | 978 | } |
4ad49000 | 979 | } |
8e274523 | 980 | |
906c06f6 | 981 | if ((apply_mask & CGROUP_MASK_DEVICES) && !is_root) { |
4ad49000 | 982 | CGroupDeviceAllow *a; |
8e274523 | 983 | |
714e2e1d LP |
984 | /* Changing the devices list of a populated cgroup |
985 | * might result in EINVAL, hence ignore EINVAL | |
986 | * here. */ | |
987 | ||
4ad49000 LP |
988 | if (c->device_allow || c->device_policy != CGROUP_AUTO) |
989 | r = cg_set_attribute("devices", path, "devices.deny", "a"); | |
990 | else | |
991 | r = cg_set_attribute("devices", path, "devices.allow", "a"); | |
1aeab12b | 992 | if (r < 0) |
f29ff115 TH |
993 | log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EINVAL, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, |
994 | "Failed to reset devices.list: %m"); | |
fb385181 | 995 | |
4ad49000 LP |
996 | if (c->device_policy == CGROUP_CLOSED || |
997 | (c->device_policy == CGROUP_AUTO && c->device_allow)) { | |
998 | static const char auto_devices[] = | |
7d711efb LP |
999 | "/dev/null\0" "rwm\0" |
1000 | "/dev/zero\0" "rwm\0" | |
1001 | "/dev/full\0" "rwm\0" | |
1002 | "/dev/random\0" "rwm\0" | |
1003 | "/dev/urandom\0" "rwm\0" | |
1004 | "/dev/tty\0" "rwm\0" | |
5a7f87a9 | 1005 | "/dev/ptmx\0" "rwm\0" |
0d9e7991 | 1006 | /* Allow /run/systemd/inaccessible/{chr,blk} devices for mapping InaccessiblePaths */ |
e7330dfe DP |
1007 | "-/run/systemd/inaccessible/chr\0" "rwm\0" |
1008 | "-/run/systemd/inaccessible/blk\0" "rwm\0"; | |
4ad49000 LP |
1009 | |
1010 | const char *x, *y; | |
1011 | ||
1012 | NULSTR_FOREACH_PAIR(x, y, auto_devices) | |
1013 | whitelist_device(path, x, y); | |
7d711efb | 1014 | |
5a7f87a9 | 1015 | /* PTS (/dev/pts) devices may not be duplicated, but accessed */ |
7d711efb | 1016 | whitelist_major(path, "pts", 'c', "rw"); |
4ad49000 LP |
1017 | } |
1018 | ||
1019 | LIST_FOREACH(device_allow, a, c->device_allow) { | |
fb4650aa | 1020 | char acc[4], *val; |
4ad49000 LP |
1021 | unsigned k = 0; |
1022 | ||
1023 | if (a->r) | |
1024 | acc[k++] = 'r'; | |
1025 | if (a->w) | |
1026 | acc[k++] = 'w'; | |
1027 | if (a->m) | |
1028 | acc[k++] = 'm'; | |
fb385181 | 1029 | |
4ad49000 LP |
1030 | if (k == 0) |
1031 | continue; | |
fb385181 | 1032 | |
4ad49000 | 1033 | acc[k++] = 0; |
90060676 | 1034 | |
27458ed6 | 1035 | if (path_startswith(a->path, "/dev/")) |
90060676 | 1036 | whitelist_device(path, a->path, acc); |
fb4650aa ZJS |
1037 | else if ((val = startswith(a->path, "block-"))) |
1038 | whitelist_major(path, val, 'b', acc); | |
1039 | else if ((val = startswith(a->path, "char-"))) | |
1040 | whitelist_major(path, val, 'c', acc); | |
90060676 | 1041 | else |
f29ff115 | 1042 | log_unit_debug(u, "Ignoring device %s while writing cgroup attribute.", a->path); |
4ad49000 LP |
1043 | } |
1044 | } | |
03a7b521 | 1045 | |
00b5974f LP |
1046 | if (apply_mask & CGROUP_MASK_PIDS) { |
1047 | ||
1048 | if (is_root) { | |
1049 | /* So, the "pids" controller does not expose anything on the root cgroup, in order not to | |
1050 | * replicate knobs exposed elsewhere needlessly. We abstract this away here however, and when | |
1051 | * the knobs of the root cgroup are modified propagate this to the relevant sysctls. There's a | |
1052 | * non-obvious asymmetry however: unlike the cgroup properties we don't really want to take | |
1053 | * exclusive ownership of the sysctls, but we still want to honour things if the user sets | |
1054 | * limits. Hence we employ sort of a one-way strategy: when the user sets a bounded limit | |
1055 | * through us it counts. When the user afterwards unsets it again (i.e. sets it to unbounded) | |
1056 | * it also counts. But if the user never set a limit through us (i.e. we are the default of | |
1057 | * "unbounded") we leave things unmodified. For this we manage a global boolean that we turn on | |
1058 | * the first time we set a limit. Note that this boolean is flushed out on manager reload, | |
1059 | * which is desirable so that there's an offical way to release control of the sysctl from | |
1060 | * systemd: set the limit to unbounded and reload. */ | |
1061 | ||
1062 | if (c->tasks_max != CGROUP_LIMIT_MAX) { | |
1063 | u->manager->sysctl_pid_max_changed = true; | |
1064 | r = procfs_tasks_set_limit(c->tasks_max); | |
1065 | } else if (u->manager->sysctl_pid_max_changed) | |
1066 | r = procfs_tasks_set_limit(TASKS_MAX); | |
1067 | else | |
1068 | r = 0; | |
03a7b521 | 1069 | |
00b5974f LP |
1070 | if (r < 0) |
1071 | log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, | |
1072 | "Failed to write to tasks limit sysctls: %m"); | |
03a7b521 | 1073 | |
00b5974f LP |
1074 | } else { |
1075 | if (c->tasks_max != CGROUP_LIMIT_MAX) { | |
1076 | char buf[DECIMAL_STR_MAX(uint64_t) + 2]; | |
03a7b521 | 1077 | |
00b5974f LP |
1078 | sprintf(buf, "%" PRIu64 "\n", c->tasks_max); |
1079 | r = cg_set_attribute("pids", path, "pids.max", buf); | |
1080 | } else | |
1081 | r = cg_set_attribute("pids", path, "pids.max", "max"); | |
1082 | if (r < 0) | |
1083 | log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, | |
1084 | "Failed to set pids.max: %m"); | |
1085 | } | |
03a7b521 | 1086 | } |
906c06f6 DM |
1087 | |
1088 | if (apply_bpf) | |
0f2d84d2 | 1089 | cgroup_apply_firewall(u); |
fb385181 LP |
1090 | } |
1091 | ||
efdb0237 LP |
1092 | CGroupMask cgroup_context_get_mask(CGroupContext *c) { |
1093 | CGroupMask mask = 0; | |
8e274523 | 1094 | |
4ad49000 | 1095 | /* Figure out which controllers we need */ |
8e274523 | 1096 | |
b2f8b02e | 1097 | if (c->cpu_accounting || |
66ebf6c0 TH |
1098 | cgroup_context_has_cpu_weight(c) || |
1099 | cgroup_context_has_cpu_shares(c) || | |
3a43da28 | 1100 | c->cpu_quota_per_sec_usec != USEC_INFINITY) |
efdb0237 | 1101 | mask |= CGROUP_MASK_CPUACCT | CGROUP_MASK_CPU; |
ecedd90f | 1102 | |
538b4852 TH |
1103 | if (cgroup_context_has_io_config(c) || cgroup_context_has_blockio_config(c)) |
1104 | mask |= CGROUP_MASK_IO | CGROUP_MASK_BLKIO; | |
ecedd90f | 1105 | |
4ad49000 | 1106 | if (c->memory_accounting || |
da4d897e TH |
1107 | c->memory_limit != CGROUP_LIMIT_MAX || |
1108 | cgroup_context_has_unified_memory_config(c)) | |
efdb0237 | 1109 | mask |= CGROUP_MASK_MEMORY; |
8e274523 | 1110 | |
a931ad47 LP |
1111 | if (c->device_allow || |
1112 | c->device_policy != CGROUP_AUTO) | |
3905f127 | 1113 | mask |= CGROUP_MASK_DEVICES; |
4ad49000 | 1114 | |
03a7b521 | 1115 | if (c->tasks_accounting || |
8793fa25 | 1116 | c->tasks_max != CGROUP_LIMIT_MAX) |
03a7b521 LP |
1117 | mask |= CGROUP_MASK_PIDS; |
1118 | ||
4ad49000 | 1119 | return mask; |
8e274523 LP |
1120 | } |
1121 | ||
efdb0237 | 1122 | CGroupMask unit_get_own_mask(Unit *u) { |
4ad49000 | 1123 | CGroupContext *c; |
8e274523 | 1124 | |
efdb0237 LP |
1125 | /* Returns the mask of controllers the unit needs for itself */ |
1126 | ||
4ad49000 LP |
1127 | c = unit_get_cgroup_context(u); |
1128 | if (!c) | |
1129 | return 0; | |
8e274523 | 1130 | |
64e844e5 | 1131 | return cgroup_context_get_mask(c) | unit_get_delegate_mask(u); |
02638280 LP |
1132 | } |
1133 | ||
1134 | CGroupMask unit_get_delegate_mask(Unit *u) { | |
1135 | CGroupContext *c; | |
1136 | ||
1137 | /* If delegation is turned on, then turn on selected controllers, unless we are on the legacy hierarchy and the | |
1138 | * process we fork into is known to drop privileges, and hence shouldn't get access to the controllers. | |
19af675e | 1139 | * |
02638280 | 1140 | * Note that on the unified hierarchy it is safe to delegate controllers to unprivileged services. */ |
a931ad47 | 1141 | |
1d9cc876 | 1142 | if (!unit_cgroup_delegate(u)) |
02638280 LP |
1143 | return 0; |
1144 | ||
1145 | if (cg_all_unified() <= 0) { | |
a931ad47 LP |
1146 | ExecContext *e; |
1147 | ||
1148 | e = unit_get_exec_context(u); | |
02638280 LP |
1149 | if (e && !exec_context_maintains_privileges(e)) |
1150 | return 0; | |
a931ad47 LP |
1151 | } |
1152 | ||
1d9cc876 | 1153 | assert_se(c = unit_get_cgroup_context(u)); |
02638280 | 1154 | return c->delegate_controllers; |
8e274523 LP |
1155 | } |
1156 | ||
efdb0237 | 1157 | CGroupMask unit_get_members_mask(Unit *u) { |
4ad49000 | 1158 | assert(u); |
bc432dc7 | 1159 | |
02638280 | 1160 | /* Returns the mask of controllers all of the unit's children require, merged */ |
efdb0237 | 1161 | |
bc432dc7 LP |
1162 | if (u->cgroup_members_mask_valid) |
1163 | return u->cgroup_members_mask; | |
1164 | ||
64e844e5 | 1165 | u->cgroup_members_mask = 0; |
bc432dc7 LP |
1166 | |
1167 | if (u->type == UNIT_SLICE) { | |
eef85c4a | 1168 | void *v; |
bc432dc7 LP |
1169 | Unit *member; |
1170 | Iterator i; | |
1171 | ||
eef85c4a | 1172 | HASHMAP_FOREACH_KEY(v, member, u->dependencies[UNIT_BEFORE], i) { |
bc432dc7 LP |
1173 | |
1174 | if (member == u) | |
1175 | continue; | |
1176 | ||
d4fdc205 | 1177 | if (UNIT_DEREF(member->slice) != u) |
bc432dc7 LP |
1178 | continue; |
1179 | ||
31604970 | 1180 | u->cgroup_members_mask |= unit_get_subtree_mask(member); /* note that this calls ourselves again, for the children */ |
bc432dc7 LP |
1181 | } |
1182 | } | |
1183 | ||
1184 | u->cgroup_members_mask_valid = true; | |
6414b7c9 | 1185 | return u->cgroup_members_mask; |
246aa6dd LP |
1186 | } |
1187 | ||
efdb0237 | 1188 | CGroupMask unit_get_siblings_mask(Unit *u) { |
4ad49000 | 1189 | assert(u); |
246aa6dd | 1190 | |
efdb0237 LP |
1191 | /* Returns the mask of controllers all of the unit's siblings |
1192 | * require, i.e. the members mask of the unit's parent slice | |
1193 | * if there is one. */ | |
1194 | ||
bc432dc7 | 1195 | if (UNIT_ISSET(u->slice)) |
637f421e | 1196 | return unit_get_members_mask(UNIT_DEREF(u->slice)); |
4ad49000 | 1197 | |
64e844e5 | 1198 | return unit_get_subtree_mask(u); /* we are the top-level slice */ |
246aa6dd LP |
1199 | } |
1200 | ||
efdb0237 LP |
1201 | CGroupMask unit_get_subtree_mask(Unit *u) { |
1202 | ||
1203 | /* Returns the mask of this subtree, meaning of the group | |
1204 | * itself and its children. */ | |
1205 | ||
1206 | return unit_get_own_mask(u) | unit_get_members_mask(u); | |
1207 | } | |
1208 | ||
1209 | CGroupMask unit_get_target_mask(Unit *u) { | |
1210 | CGroupMask mask; | |
1211 | ||
1212 | /* This returns the cgroup mask of all controllers to enable | |
1213 | * for a specific cgroup, i.e. everything it needs itself, | |
1214 | * plus all that its children need, plus all that its siblings | |
1215 | * need. This is primarily useful on the legacy cgroup | |
1216 | * hierarchy, where we need to duplicate each cgroup in each | |
1217 | * hierarchy that shall be enabled for it. */ | |
6414b7c9 | 1218 | |
efdb0237 LP |
1219 | mask = unit_get_own_mask(u) | unit_get_members_mask(u) | unit_get_siblings_mask(u); |
1220 | mask &= u->manager->cgroup_supported; | |
1221 | ||
1222 | return mask; | |
1223 | } | |
1224 | ||
1225 | CGroupMask unit_get_enable_mask(Unit *u) { | |
1226 | CGroupMask mask; | |
1227 | ||
1228 | /* This returns the cgroup mask of all controllers to enable | |
1229 | * for the children of a specific cgroup. This is primarily | |
1230 | * useful for the unified cgroup hierarchy, where each cgroup | |
1231 | * controls which controllers are enabled for its children. */ | |
1232 | ||
1233 | mask = unit_get_members_mask(u); | |
6414b7c9 DS |
1234 | mask &= u->manager->cgroup_supported; |
1235 | ||
1236 | return mask; | |
1237 | } | |
1238 | ||
906c06f6 DM |
1239 | bool unit_get_needs_bpf(Unit *u) { |
1240 | CGroupContext *c; | |
1241 | Unit *p; | |
1242 | assert(u); | |
1243 | ||
906c06f6 DM |
1244 | c = unit_get_cgroup_context(u); |
1245 | if (!c) | |
1246 | return false; | |
1247 | ||
1248 | if (c->ip_accounting || | |
1249 | c->ip_address_allow || | |
1250 | c->ip_address_deny) | |
1251 | return true; | |
1252 | ||
1253 | /* If any parent slice has an IP access list defined, it applies too */ | |
1254 | for (p = UNIT_DEREF(u->slice); p; p = UNIT_DEREF(p->slice)) { | |
1255 | c = unit_get_cgroup_context(p); | |
1256 | if (!c) | |
1257 | return false; | |
1258 | ||
1259 | if (c->ip_address_allow || | |
1260 | c->ip_address_deny) | |
1261 | return true; | |
1262 | } | |
1263 | ||
1264 | return false; | |
1265 | } | |
1266 | ||
6414b7c9 DS |
1267 | /* Recurse from a unit up through its containing slices, propagating |
1268 | * mask bits upward. A unit is also member of itself. */ | |
bc432dc7 | 1269 | void unit_update_cgroup_members_masks(Unit *u) { |
efdb0237 | 1270 | CGroupMask m; |
bc432dc7 LP |
1271 | bool more; |
1272 | ||
1273 | assert(u); | |
1274 | ||
1275 | /* Calculate subtree mask */ | |
efdb0237 | 1276 | m = unit_get_subtree_mask(u); |
bc432dc7 LP |
1277 | |
1278 | /* See if anything changed from the previous invocation. If | |
1279 | * not, we're done. */ | |
1280 | if (u->cgroup_subtree_mask_valid && m == u->cgroup_subtree_mask) | |
1281 | return; | |
1282 | ||
1283 | more = | |
1284 | u->cgroup_subtree_mask_valid && | |
1285 | ((m & ~u->cgroup_subtree_mask) != 0) && | |
1286 | ((~m & u->cgroup_subtree_mask) == 0); | |
1287 | ||
1288 | u->cgroup_subtree_mask = m; | |
1289 | u->cgroup_subtree_mask_valid = true; | |
1290 | ||
6414b7c9 DS |
1291 | if (UNIT_ISSET(u->slice)) { |
1292 | Unit *s = UNIT_DEREF(u->slice); | |
bc432dc7 LP |
1293 | |
1294 | if (more) | |
1295 | /* There's more set now than before. We | |
1296 | * propagate the new mask to the parent's mask | |
1297 | * (not caring if it actually was valid or | |
1298 | * not). */ | |
1299 | ||
1300 | s->cgroup_members_mask |= m; | |
1301 | ||
1302 | else | |
1303 | /* There's less set now than before (or we | |
1304 | * don't know), we need to recalculate | |
1305 | * everything, so let's invalidate the | |
1306 | * parent's members mask */ | |
1307 | ||
1308 | s->cgroup_members_mask_valid = false; | |
1309 | ||
1310 | /* And now make sure that this change also hits our | |
1311 | * grandparents */ | |
1312 | unit_update_cgroup_members_masks(s); | |
6414b7c9 DS |
1313 | } |
1314 | } | |
1315 | ||
6592b975 | 1316 | const char *unit_get_realized_cgroup_path(Unit *u, CGroupMask mask) { |
03b90d4b | 1317 | |
6592b975 | 1318 | /* Returns the realized cgroup path of the specified unit where all specified controllers are available. */ |
03b90d4b LP |
1319 | |
1320 | while (u) { | |
6592b975 | 1321 | |
03b90d4b LP |
1322 | if (u->cgroup_path && |
1323 | u->cgroup_realized && | |
d94a24ca | 1324 | FLAGS_SET(u->cgroup_realized_mask, mask)) |
03b90d4b LP |
1325 | return u->cgroup_path; |
1326 | ||
1327 | u = UNIT_DEREF(u->slice); | |
1328 | } | |
1329 | ||
1330 | return NULL; | |
1331 | } | |
1332 | ||
6592b975 LP |
1333 | static const char *migrate_callback(CGroupMask mask, void *userdata) { |
1334 | return unit_get_realized_cgroup_path(userdata, mask); | |
1335 | } | |
1336 | ||
efdb0237 LP |
1337 | char *unit_default_cgroup_path(Unit *u) { |
1338 | _cleanup_free_ char *escaped = NULL, *slice = NULL; | |
1339 | int r; | |
1340 | ||
1341 | assert(u); | |
1342 | ||
1343 | if (unit_has_name(u, SPECIAL_ROOT_SLICE)) | |
1344 | return strdup(u->manager->cgroup_root); | |
1345 | ||
1346 | if (UNIT_ISSET(u->slice) && !unit_has_name(UNIT_DEREF(u->slice), SPECIAL_ROOT_SLICE)) { | |
1347 | r = cg_slice_to_path(UNIT_DEREF(u->slice)->id, &slice); | |
1348 | if (r < 0) | |
1349 | return NULL; | |
1350 | } | |
1351 | ||
1352 | escaped = cg_escape(u->id); | |
1353 | if (!escaped) | |
1354 | return NULL; | |
1355 | ||
1356 | if (slice) | |
605405c6 ZJS |
1357 | return strjoin(u->manager->cgroup_root, "/", slice, "/", |
1358 | escaped); | |
efdb0237 | 1359 | else |
605405c6 | 1360 | return strjoin(u->manager->cgroup_root, "/", escaped); |
efdb0237 LP |
1361 | } |
1362 | ||
1363 | int unit_set_cgroup_path(Unit *u, const char *path) { | |
1364 | _cleanup_free_ char *p = NULL; | |
1365 | int r; | |
1366 | ||
1367 | assert(u); | |
1368 | ||
1369 | if (path) { | |
1370 | p = strdup(path); | |
1371 | if (!p) | |
1372 | return -ENOMEM; | |
1373 | } else | |
1374 | p = NULL; | |
1375 | ||
1376 | if (streq_ptr(u->cgroup_path, p)) | |
1377 | return 0; | |
1378 | ||
1379 | if (p) { | |
1380 | r = hashmap_put(u->manager->cgroup_unit, p, u); | |
1381 | if (r < 0) | |
1382 | return r; | |
1383 | } | |
1384 | ||
1385 | unit_release_cgroup(u); | |
1386 | ||
ae2a15bc | 1387 | u->cgroup_path = TAKE_PTR(p); |
efdb0237 LP |
1388 | |
1389 | return 1; | |
1390 | } | |
1391 | ||
1392 | int unit_watch_cgroup(Unit *u) { | |
ab2c3861 | 1393 | _cleanup_free_ char *events = NULL; |
efdb0237 LP |
1394 | int r; |
1395 | ||
1396 | assert(u); | |
1397 | ||
1398 | if (!u->cgroup_path) | |
1399 | return 0; | |
1400 | ||
1401 | if (u->cgroup_inotify_wd >= 0) | |
1402 | return 0; | |
1403 | ||
1404 | /* Only applies to the unified hierarchy */ | |
c22800e4 | 1405 | r = cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER); |
b4cccbc1 LP |
1406 | if (r < 0) |
1407 | return log_error_errno(r, "Failed to determine whether the name=systemd hierarchy is unified: %m"); | |
1408 | if (r == 0) | |
efdb0237 LP |
1409 | return 0; |
1410 | ||
1411 | /* Don't watch the root slice, it's pointless. */ | |
1412 | if (unit_has_name(u, SPECIAL_ROOT_SLICE)) | |
1413 | return 0; | |
1414 | ||
1415 | r = hashmap_ensure_allocated(&u->manager->cgroup_inotify_wd_unit, &trivial_hash_ops); | |
1416 | if (r < 0) | |
1417 | return log_oom(); | |
1418 | ||
ab2c3861 | 1419 | r = cg_get_path(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path, "cgroup.events", &events); |
efdb0237 LP |
1420 | if (r < 0) |
1421 | return log_oom(); | |
1422 | ||
ab2c3861 | 1423 | u->cgroup_inotify_wd = inotify_add_watch(u->manager->cgroup_inotify_fd, events, IN_MODIFY); |
efdb0237 LP |
1424 | if (u->cgroup_inotify_wd < 0) { |
1425 | ||
1426 | if (errno == ENOENT) /* If the directory is already | |
1427 | * gone we don't need to track | |
1428 | * it, so this is not an error */ | |
1429 | return 0; | |
1430 | ||
1431 | return log_unit_error_errno(u, errno, "Failed to add inotify watch descriptor for control group %s: %m", u->cgroup_path); | |
1432 | } | |
1433 | ||
1434 | r = hashmap_put(u->manager->cgroup_inotify_wd_unit, INT_TO_PTR(u->cgroup_inotify_wd), u); | |
1435 | if (r < 0) | |
1436 | return log_unit_error_errno(u, r, "Failed to add inotify watch descriptor to hash map: %m"); | |
1437 | ||
1438 | return 0; | |
1439 | } | |
1440 | ||
a4634b21 LP |
1441 | int unit_pick_cgroup_path(Unit *u) { |
1442 | _cleanup_free_ char *path = NULL; | |
1443 | int r; | |
1444 | ||
1445 | assert(u); | |
1446 | ||
1447 | if (u->cgroup_path) | |
1448 | return 0; | |
1449 | ||
1450 | if (!UNIT_HAS_CGROUP_CONTEXT(u)) | |
1451 | return -EINVAL; | |
1452 | ||
1453 | path = unit_default_cgroup_path(u); | |
1454 | if (!path) | |
1455 | return log_oom(); | |
1456 | ||
1457 | r = unit_set_cgroup_path(u, path); | |
1458 | if (r == -EEXIST) | |
1459 | return log_unit_error_errno(u, r, "Control group %s exists already.", path); | |
1460 | if (r < 0) | |
1461 | return log_unit_error_errno(u, r, "Failed to set unit's control group path to %s: %m", path); | |
1462 | ||
1463 | return 0; | |
1464 | } | |
1465 | ||
efdb0237 LP |
1466 | static int unit_create_cgroup( |
1467 | Unit *u, | |
1468 | CGroupMask target_mask, | |
906c06f6 DM |
1469 | CGroupMask enable_mask, |
1470 | bool needs_bpf) { | |
efdb0237 | 1471 | |
0cd385d3 | 1472 | CGroupContext *c; |
bc432dc7 | 1473 | int r; |
65be7e06 | 1474 | bool created; |
64747e2d | 1475 | |
4ad49000 | 1476 | assert(u); |
64747e2d | 1477 | |
0cd385d3 LP |
1478 | c = unit_get_cgroup_context(u); |
1479 | if (!c) | |
1480 | return 0; | |
1481 | ||
a4634b21 LP |
1482 | /* Figure out our cgroup path */ |
1483 | r = unit_pick_cgroup_path(u); | |
1484 | if (r < 0) | |
1485 | return r; | |
b58b8e11 | 1486 | |
03b90d4b | 1487 | /* First, create our own group */ |
efdb0237 | 1488 | r = cg_create_everywhere(u->manager->cgroup_supported, target_mask, u->cgroup_path); |
23bbb0de | 1489 | if (r < 0) |
efdb0237 | 1490 | return log_unit_error_errno(u, r, "Failed to create cgroup %s: %m", u->cgroup_path); |
65be7e06 | 1491 | created = !!r; |
efdb0237 LP |
1492 | |
1493 | /* Start watching it */ | |
1494 | (void) unit_watch_cgroup(u); | |
1495 | ||
65be7e06 ZJS |
1496 | /* Preserve enabled controllers in delegated units, adjust others. */ |
1497 | if (created || !unit_cgroup_delegate(u)) { | |
1498 | ||
1499 | /* Enable all controllers we need */ | |
1500 | r = cg_enable_everywhere(u->manager->cgroup_supported, enable_mask, u->cgroup_path); | |
1501 | if (r < 0) | |
1502 | log_unit_warning_errno(u, r, "Failed to enable controllers on cgroup %s, ignoring: %m", | |
1503 | u->cgroup_path); | |
1504 | } | |
03b90d4b LP |
1505 | |
1506 | /* Keep track that this is now realized */ | |
4ad49000 | 1507 | u->cgroup_realized = true; |
efdb0237 | 1508 | u->cgroup_realized_mask = target_mask; |
ccf78df1 | 1509 | u->cgroup_enabled_mask = enable_mask; |
906c06f6 | 1510 | u->cgroup_bpf_state = needs_bpf ? UNIT_CGROUP_BPF_ON : UNIT_CGROUP_BPF_OFF; |
4ad49000 | 1511 | |
1d9cc876 | 1512 | if (u->type != UNIT_SLICE && !unit_cgroup_delegate(u)) { |
0cd385d3 LP |
1513 | |
1514 | /* Then, possibly move things over, but not if | |
1515 | * subgroups may contain processes, which is the case | |
1516 | * for slice and delegation units. */ | |
1517 | r = cg_migrate_everywhere(u->manager->cgroup_supported, u->cgroup_path, u->cgroup_path, migrate_callback, u); | |
1518 | if (r < 0) | |
efdb0237 | 1519 | log_unit_warning_errno(u, r, "Failed to migrate cgroup from to %s, ignoring: %m", u->cgroup_path); |
0cd385d3 | 1520 | } |
03b90d4b | 1521 | |
64747e2d LP |
1522 | return 0; |
1523 | } | |
1524 | ||
6592b975 LP |
1525 | static int unit_attach_pid_to_cgroup_via_bus(Unit *u, pid_t pid, const char *suffix_path) { |
1526 | _cleanup_(sd_bus_error_free) sd_bus_error error = SD_BUS_ERROR_NULL; | |
1527 | char *pp; | |
7b3fd631 | 1528 | int r; |
6592b975 | 1529 | |
7b3fd631 LP |
1530 | assert(u); |
1531 | ||
6592b975 LP |
1532 | if (MANAGER_IS_SYSTEM(u->manager)) |
1533 | return -EINVAL; | |
1534 | ||
1535 | if (!u->manager->system_bus) | |
1536 | return -EIO; | |
1537 | ||
1538 | if (!u->cgroup_path) | |
1539 | return -EINVAL; | |
1540 | ||
1541 | /* Determine this unit's cgroup path relative to our cgroup root */ | |
1542 | pp = path_startswith(u->cgroup_path, u->manager->cgroup_root); | |
1543 | if (!pp) | |
1544 | return -EINVAL; | |
1545 | ||
1546 | pp = strjoina("/", pp, suffix_path); | |
858d36c1 | 1547 | path_simplify(pp, false); |
6592b975 LP |
1548 | |
1549 | r = sd_bus_call_method(u->manager->system_bus, | |
1550 | "org.freedesktop.systemd1", | |
1551 | "/org/freedesktop/systemd1", | |
1552 | "org.freedesktop.systemd1.Manager", | |
1553 | "AttachProcessesToUnit", | |
1554 | &error, NULL, | |
1555 | "ssau", | |
1556 | NULL /* empty unit name means client's unit, i.e. us */, pp, 1, (uint32_t) pid); | |
7b3fd631 | 1557 | if (r < 0) |
6592b975 LP |
1558 | return log_unit_debug_errno(u, r, "Failed to attach unit process " PID_FMT " via the bus: %s", pid, bus_error_message(&error, r)); |
1559 | ||
1560 | return 0; | |
1561 | } | |
1562 | ||
1563 | int unit_attach_pids_to_cgroup(Unit *u, Set *pids, const char *suffix_path) { | |
1564 | CGroupMask delegated_mask; | |
1565 | const char *p; | |
1566 | Iterator i; | |
1567 | void *pidp; | |
1568 | int r, q; | |
1569 | ||
1570 | assert(u); | |
1571 | ||
1572 | if (!UNIT_HAS_CGROUP_CONTEXT(u)) | |
1573 | return -EINVAL; | |
1574 | ||
1575 | if (set_isempty(pids)) | |
1576 | return 0; | |
7b3fd631 | 1577 | |
6592b975 | 1578 | r = unit_realize_cgroup(u); |
7b3fd631 LP |
1579 | if (r < 0) |
1580 | return r; | |
1581 | ||
6592b975 LP |
1582 | if (isempty(suffix_path)) |
1583 | p = u->cgroup_path; | |
1584 | else | |
1585 | p = strjoina(u->cgroup_path, "/", suffix_path); | |
1586 | ||
1587 | delegated_mask = unit_get_delegate_mask(u); | |
1588 | ||
1589 | r = 0; | |
1590 | SET_FOREACH(pidp, pids, i) { | |
1591 | pid_t pid = PTR_TO_PID(pidp); | |
1592 | CGroupController c; | |
1593 | ||
1594 | /* First, attach the PID to the main cgroup hierarchy */ | |
1595 | q = cg_attach(SYSTEMD_CGROUP_CONTROLLER, p, pid); | |
1596 | if (q < 0) { | |
1597 | log_unit_debug_errno(u, q, "Couldn't move process " PID_FMT " to requested cgroup '%s': %m", pid, p); | |
1598 | ||
1599 | if (MANAGER_IS_USER(u->manager) && IN_SET(q, -EPERM, -EACCES)) { | |
1600 | int z; | |
1601 | ||
1602 | /* If we are in a user instance, and we can't move the process ourselves due to | |
1603 | * permission problems, let's ask the system instance about it instead. Since it's more | |
1604 | * privileged it might be able to move the process across the leaves of a subtree who's | |
1605 | * top node is not owned by us. */ | |
1606 | ||
1607 | z = unit_attach_pid_to_cgroup_via_bus(u, pid, suffix_path); | |
1608 | if (z < 0) | |
1609 | log_unit_debug_errno(u, z, "Couldn't move process " PID_FMT " to requested cgroup '%s' via the system bus either: %m", pid, p); | |
1610 | else | |
1611 | continue; /* When the bus thing worked via the bus we are fully done for this PID. */ | |
1612 | } | |
1613 | ||
1614 | if (r >= 0) | |
1615 | r = q; /* Remember first error */ | |
1616 | ||
1617 | continue; | |
1618 | } | |
1619 | ||
1620 | q = cg_all_unified(); | |
1621 | if (q < 0) | |
1622 | return q; | |
1623 | if (q > 0) | |
1624 | continue; | |
1625 | ||
1626 | /* In the legacy hierarchy, attach the process to the request cgroup if possible, and if not to the | |
1627 | * innermost realized one */ | |
1628 | ||
1629 | for (c = 0; c < _CGROUP_CONTROLLER_MAX; c++) { | |
1630 | CGroupMask bit = CGROUP_CONTROLLER_TO_MASK(c); | |
1631 | const char *realized; | |
1632 | ||
1633 | if (!(u->manager->cgroup_supported & bit)) | |
1634 | continue; | |
1635 | ||
1636 | /* If this controller is delegated and realized, honour the caller's request for the cgroup suffix. */ | |
1637 | if (delegated_mask & u->cgroup_realized_mask & bit) { | |
1638 | q = cg_attach(cgroup_controller_to_string(c), p, pid); | |
1639 | if (q >= 0) | |
1640 | continue; /* Success! */ | |
1641 | ||
1642 | 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", | |
1643 | pid, p, cgroup_controller_to_string(c)); | |
1644 | } | |
1645 | ||
1646 | /* So this controller is either not delegate or realized, or something else weird happened. In | |
1647 | * that case let's attach the PID at least to the closest cgroup up the tree that is | |
1648 | * realized. */ | |
1649 | realized = unit_get_realized_cgroup_path(u, bit); | |
1650 | if (!realized) | |
1651 | continue; /* Not even realized in the root slice? Then let's not bother */ | |
1652 | ||
1653 | q = cg_attach(cgroup_controller_to_string(c), realized, pid); | |
1654 | if (q < 0) | |
1655 | log_unit_debug_errno(u, q, "Failed to attach PID " PID_FMT " to realized cgroup %s in controller %s, ignoring: %m", | |
1656 | pid, realized, cgroup_controller_to_string(c)); | |
1657 | } | |
1658 | } | |
1659 | ||
1660 | return r; | |
7b3fd631 LP |
1661 | } |
1662 | ||
4b58153d LP |
1663 | static void cgroup_xattr_apply(Unit *u) { |
1664 | char ids[SD_ID128_STRING_MAX]; | |
1665 | int r; | |
1666 | ||
1667 | assert(u); | |
1668 | ||
1669 | if (!MANAGER_IS_SYSTEM(u->manager)) | |
1670 | return; | |
1671 | ||
1672 | if (sd_id128_is_null(u->invocation_id)) | |
1673 | return; | |
1674 | ||
1675 | r = cg_set_xattr(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path, | |
1676 | "trusted.invocation_id", | |
1677 | sd_id128_to_string(u->invocation_id, ids), 32, | |
1678 | 0); | |
1679 | if (r < 0) | |
0fb84499 | 1680 | log_unit_debug_errno(u, r, "Failed to set invocation ID on control group %s, ignoring: %m", u->cgroup_path); |
4b58153d LP |
1681 | } |
1682 | ||
906c06f6 DM |
1683 | static bool unit_has_mask_realized( |
1684 | Unit *u, | |
1685 | CGroupMask target_mask, | |
1686 | CGroupMask enable_mask, | |
1687 | bool needs_bpf) { | |
1688 | ||
bc432dc7 LP |
1689 | assert(u); |
1690 | ||
906c06f6 DM |
1691 | return u->cgroup_realized && |
1692 | u->cgroup_realized_mask == target_mask && | |
1693 | u->cgroup_enabled_mask == enable_mask && | |
1694 | ((needs_bpf && u->cgroup_bpf_state == UNIT_CGROUP_BPF_ON) || | |
1695 | (!needs_bpf && u->cgroup_bpf_state == UNIT_CGROUP_BPF_OFF)); | |
6414b7c9 DS |
1696 | } |
1697 | ||
2aa57a65 LP |
1698 | static void unit_add_to_cgroup_realize_queue(Unit *u) { |
1699 | assert(u); | |
1700 | ||
1701 | if (u->in_cgroup_realize_queue) | |
1702 | return; | |
1703 | ||
1704 | LIST_PREPEND(cgroup_realize_queue, u->manager->cgroup_realize_queue, u); | |
1705 | u->in_cgroup_realize_queue = true; | |
1706 | } | |
1707 | ||
1708 | static void unit_remove_from_cgroup_realize_queue(Unit *u) { | |
1709 | assert(u); | |
1710 | ||
1711 | if (!u->in_cgroup_realize_queue) | |
1712 | return; | |
1713 | ||
1714 | LIST_REMOVE(cgroup_realize_queue, u->manager->cgroup_realize_queue, u); | |
1715 | u->in_cgroup_realize_queue = false; | |
1716 | } | |
1717 | ||
6414b7c9 DS |
1718 | /* Check if necessary controllers and attributes for a unit are in place. |
1719 | * | |
1720 | * If so, do nothing. | |
1721 | * If not, create paths, move processes over, and set attributes. | |
1722 | * | |
1723 | * Returns 0 on success and < 0 on failure. */ | |
db785129 | 1724 | static int unit_realize_cgroup_now(Unit *u, ManagerState state) { |
efdb0237 | 1725 | CGroupMask target_mask, enable_mask; |
906c06f6 | 1726 | bool needs_bpf, apply_bpf; |
6414b7c9 | 1727 | int r; |
64747e2d | 1728 | |
4ad49000 | 1729 | assert(u); |
64747e2d | 1730 | |
2aa57a65 | 1731 | unit_remove_from_cgroup_realize_queue(u); |
64747e2d | 1732 | |
efdb0237 | 1733 | target_mask = unit_get_target_mask(u); |
ccf78df1 | 1734 | enable_mask = unit_get_enable_mask(u); |
906c06f6 | 1735 | needs_bpf = unit_get_needs_bpf(u); |
ccf78df1 | 1736 | |
906c06f6 | 1737 | if (unit_has_mask_realized(u, target_mask, enable_mask, needs_bpf)) |
0a1eb06d | 1738 | return 0; |
64747e2d | 1739 | |
906c06f6 DM |
1740 | /* Make sure we apply the BPF filters either when one is configured, or if none is configured but previously |
1741 | * the state was anything but off. This way, if a unit with a BPF filter applied is reconfigured to lose it | |
1742 | * this will trickle down properly to cgroupfs. */ | |
1743 | apply_bpf = needs_bpf || u->cgroup_bpf_state != UNIT_CGROUP_BPF_OFF; | |
1744 | ||
4ad49000 | 1745 | /* First, realize parents */ |
6414b7c9 | 1746 | if (UNIT_ISSET(u->slice)) { |
db785129 | 1747 | r = unit_realize_cgroup_now(UNIT_DEREF(u->slice), state); |
6414b7c9 DS |
1748 | if (r < 0) |
1749 | return r; | |
1750 | } | |
4ad49000 LP |
1751 | |
1752 | /* And then do the real work */ | |
906c06f6 | 1753 | r = unit_create_cgroup(u, target_mask, enable_mask, needs_bpf); |
6414b7c9 DS |
1754 | if (r < 0) |
1755 | return r; | |
1756 | ||
1757 | /* Finally, apply the necessary attributes. */ | |
906c06f6 | 1758 | cgroup_context_apply(u, target_mask, apply_bpf, state); |
4b58153d | 1759 | cgroup_xattr_apply(u); |
6414b7c9 DS |
1760 | |
1761 | return 0; | |
64747e2d LP |
1762 | } |
1763 | ||
91a6073e | 1764 | unsigned manager_dispatch_cgroup_realize_queue(Manager *m) { |
db785129 | 1765 | ManagerState state; |
4ad49000 | 1766 | unsigned n = 0; |
db785129 | 1767 | Unit *i; |
6414b7c9 | 1768 | int r; |
ecedd90f | 1769 | |
91a6073e LP |
1770 | assert(m); |
1771 | ||
db785129 LP |
1772 | state = manager_state(m); |
1773 | ||
91a6073e LP |
1774 | while ((i = m->cgroup_realize_queue)) { |
1775 | assert(i->in_cgroup_realize_queue); | |
ecedd90f | 1776 | |
2aa57a65 LP |
1777 | if (UNIT_IS_INACTIVE_OR_FAILED(unit_active_state(i))) { |
1778 | /* Maybe things changed, and the unit is not actually active anymore? */ | |
1779 | unit_remove_from_cgroup_realize_queue(i); | |
1780 | continue; | |
1781 | } | |
1782 | ||
db785129 | 1783 | r = unit_realize_cgroup_now(i, state); |
6414b7c9 | 1784 | if (r < 0) |
efdb0237 | 1785 | log_warning_errno(r, "Failed to realize cgroups for queued unit %s, ignoring: %m", i->id); |
0a1eb06d | 1786 | |
4ad49000 LP |
1787 | n++; |
1788 | } | |
ecedd90f | 1789 | |
4ad49000 | 1790 | return n; |
8e274523 LP |
1791 | } |
1792 | ||
91a6073e | 1793 | static void unit_add_siblings_to_cgroup_realize_queue(Unit *u) { |
4ad49000 | 1794 | Unit *slice; |
ca949c9d | 1795 | |
4ad49000 LP |
1796 | /* This adds the siblings of the specified unit and the |
1797 | * siblings of all parent units to the cgroup queue. (But | |
1798 | * neither the specified unit itself nor the parents.) */ | |
1799 | ||
1800 | while ((slice = UNIT_DEREF(u->slice))) { | |
1801 | Iterator i; | |
1802 | Unit *m; | |
eef85c4a | 1803 | void *v; |
8f53a7b8 | 1804 | |
eef85c4a | 1805 | HASHMAP_FOREACH_KEY(v, m, u->dependencies[UNIT_BEFORE], i) { |
4ad49000 LP |
1806 | if (m == u) |
1807 | continue; | |
8e274523 | 1808 | |
6414b7c9 DS |
1809 | /* Skip units that have a dependency on the slice |
1810 | * but aren't actually in it. */ | |
4ad49000 | 1811 | if (UNIT_DEREF(m->slice) != slice) |
50159e6a | 1812 | continue; |
8e274523 | 1813 | |
6414b7c9 DS |
1814 | /* No point in doing cgroup application for units |
1815 | * without active processes. */ | |
1816 | if (UNIT_IS_INACTIVE_OR_FAILED(unit_active_state(m))) | |
1817 | continue; | |
1818 | ||
1819 | /* If the unit doesn't need any new controllers | |
1820 | * and has current ones realized, it doesn't need | |
1821 | * any changes. */ | |
906c06f6 DM |
1822 | if (unit_has_mask_realized(m, |
1823 | unit_get_target_mask(m), | |
1824 | unit_get_enable_mask(m), | |
1825 | unit_get_needs_bpf(m))) | |
6414b7c9 DS |
1826 | continue; |
1827 | ||
91a6073e | 1828 | unit_add_to_cgroup_realize_queue(m); |
50159e6a LP |
1829 | } |
1830 | ||
4ad49000 | 1831 | u = slice; |
8e274523 | 1832 | } |
4ad49000 LP |
1833 | } |
1834 | ||
0a1eb06d | 1835 | int unit_realize_cgroup(Unit *u) { |
4ad49000 LP |
1836 | assert(u); |
1837 | ||
35b7ff80 | 1838 | if (!UNIT_HAS_CGROUP_CONTEXT(u)) |
0a1eb06d | 1839 | return 0; |
8e274523 | 1840 | |
4ad49000 LP |
1841 | /* So, here's the deal: when realizing the cgroups for this |
1842 | * unit, we need to first create all parents, but there's more | |
1843 | * actually: for the weight-based controllers we also need to | |
1844 | * make sure that all our siblings (i.e. units that are in the | |
73e231ab | 1845 | * same slice as we are) have cgroups, too. Otherwise, things |
4ad49000 LP |
1846 | * would become very uneven as each of their processes would |
1847 | * get as much resources as all our group together. This call | |
1848 | * will synchronously create the parent cgroups, but will | |
1849 | * defer work on the siblings to the next event loop | |
1850 | * iteration. */ | |
ca949c9d | 1851 | |
4ad49000 | 1852 | /* Add all sibling slices to the cgroup queue. */ |
91a6073e | 1853 | unit_add_siblings_to_cgroup_realize_queue(u); |
4ad49000 | 1854 | |
6414b7c9 | 1855 | /* And realize this one now (and apply the values) */ |
db785129 | 1856 | return unit_realize_cgroup_now(u, manager_state(u->manager)); |
8e274523 LP |
1857 | } |
1858 | ||
efdb0237 LP |
1859 | void unit_release_cgroup(Unit *u) { |
1860 | assert(u); | |
1861 | ||
1862 | /* Forgets all cgroup details for this cgroup */ | |
1863 | ||
1864 | if (u->cgroup_path) { | |
1865 | (void) hashmap_remove(u->manager->cgroup_unit, u->cgroup_path); | |
1866 | u->cgroup_path = mfree(u->cgroup_path); | |
1867 | } | |
1868 | ||
1869 | if (u->cgroup_inotify_wd >= 0) { | |
1870 | if (inotify_rm_watch(u->manager->cgroup_inotify_fd, u->cgroup_inotify_wd) < 0) | |
1871 | log_unit_debug_errno(u, errno, "Failed to remove cgroup inotify watch %i for %s, ignoring", u->cgroup_inotify_wd, u->id); | |
1872 | ||
1873 | (void) hashmap_remove(u->manager->cgroup_inotify_wd_unit, INT_TO_PTR(u->cgroup_inotify_wd)); | |
1874 | u->cgroup_inotify_wd = -1; | |
1875 | } | |
1876 | } | |
1877 | ||
1878 | void unit_prune_cgroup(Unit *u) { | |
8e274523 | 1879 | int r; |
efdb0237 | 1880 | bool is_root_slice; |
8e274523 | 1881 | |
4ad49000 | 1882 | assert(u); |
8e274523 | 1883 | |
efdb0237 LP |
1884 | /* Removes the cgroup, if empty and possible, and stops watching it. */ |
1885 | ||
4ad49000 LP |
1886 | if (!u->cgroup_path) |
1887 | return; | |
8e274523 | 1888 | |
fe700f46 LP |
1889 | (void) unit_get_cpu_usage(u, NULL); /* Cache the last CPU usage value before we destroy the cgroup */ |
1890 | ||
efdb0237 LP |
1891 | is_root_slice = unit_has_name(u, SPECIAL_ROOT_SLICE); |
1892 | ||
1893 | r = cg_trim_everywhere(u->manager->cgroup_supported, u->cgroup_path, !is_root_slice); | |
dab5bf85 | 1894 | if (r < 0) { |
f29ff115 | 1895 | log_unit_debug_errno(u, r, "Failed to destroy cgroup %s, ignoring: %m", u->cgroup_path); |
dab5bf85 RL |
1896 | return; |
1897 | } | |
8e274523 | 1898 | |
efdb0237 LP |
1899 | if (is_root_slice) |
1900 | return; | |
1901 | ||
1902 | unit_release_cgroup(u); | |
0a1eb06d | 1903 | |
4ad49000 | 1904 | u->cgroup_realized = false; |
bc432dc7 | 1905 | u->cgroup_realized_mask = 0; |
ccf78df1 | 1906 | u->cgroup_enabled_mask = 0; |
8e274523 LP |
1907 | } |
1908 | ||
efdb0237 | 1909 | int unit_search_main_pid(Unit *u, pid_t *ret) { |
4ad49000 LP |
1910 | _cleanup_fclose_ FILE *f = NULL; |
1911 | pid_t pid = 0, npid, mypid; | |
efdb0237 | 1912 | int r; |
4ad49000 LP |
1913 | |
1914 | assert(u); | |
efdb0237 | 1915 | assert(ret); |
4ad49000 LP |
1916 | |
1917 | if (!u->cgroup_path) | |
efdb0237 | 1918 | return -ENXIO; |
4ad49000 | 1919 | |
efdb0237 LP |
1920 | r = cg_enumerate_processes(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path, &f); |
1921 | if (r < 0) | |
1922 | return r; | |
4ad49000 | 1923 | |
df0ff127 | 1924 | mypid = getpid_cached(); |
4ad49000 LP |
1925 | while (cg_read_pid(f, &npid) > 0) { |
1926 | pid_t ppid; | |
1927 | ||
1928 | if (npid == pid) | |
1929 | continue; | |
8e274523 | 1930 | |
4ad49000 | 1931 | /* Ignore processes that aren't our kids */ |
6bc73acb | 1932 | if (get_process_ppid(npid, &ppid) >= 0 && ppid != mypid) |
4ad49000 | 1933 | continue; |
8e274523 | 1934 | |
efdb0237 | 1935 | if (pid != 0) |
4ad49000 LP |
1936 | /* Dang, there's more than one daemonized PID |
1937 | in this group, so we don't know what process | |
1938 | is the main process. */ | |
efdb0237 LP |
1939 | |
1940 | return -ENODATA; | |
8e274523 | 1941 | |
4ad49000 | 1942 | pid = npid; |
8e274523 LP |
1943 | } |
1944 | ||
efdb0237 LP |
1945 | *ret = pid; |
1946 | return 0; | |
1947 | } | |
1948 | ||
1949 | static int unit_watch_pids_in_path(Unit *u, const char *path) { | |
b3c5bad3 | 1950 | _cleanup_closedir_ DIR *d = NULL; |
efdb0237 LP |
1951 | _cleanup_fclose_ FILE *f = NULL; |
1952 | int ret = 0, r; | |
1953 | ||
1954 | assert(u); | |
1955 | assert(path); | |
1956 | ||
1957 | r = cg_enumerate_processes(SYSTEMD_CGROUP_CONTROLLER, path, &f); | |
1958 | if (r < 0) | |
1959 | ret = r; | |
1960 | else { | |
1961 | pid_t pid; | |
1962 | ||
1963 | while ((r = cg_read_pid(f, &pid)) > 0) { | |
1964 | r = unit_watch_pid(u, pid); | |
1965 | if (r < 0 && ret >= 0) | |
1966 | ret = r; | |
1967 | } | |
1968 | ||
1969 | if (r < 0 && ret >= 0) | |
1970 | ret = r; | |
1971 | } | |
1972 | ||
1973 | r = cg_enumerate_subgroups(SYSTEMD_CGROUP_CONTROLLER, path, &d); | |
1974 | if (r < 0) { | |
1975 | if (ret >= 0) | |
1976 | ret = r; | |
1977 | } else { | |
1978 | char *fn; | |
1979 | ||
1980 | while ((r = cg_read_subgroup(d, &fn)) > 0) { | |
1981 | _cleanup_free_ char *p = NULL; | |
1982 | ||
605405c6 | 1983 | p = strjoin(path, "/", fn); |
efdb0237 LP |
1984 | free(fn); |
1985 | ||
1986 | if (!p) | |
1987 | return -ENOMEM; | |
1988 | ||
1989 | r = unit_watch_pids_in_path(u, p); | |
1990 | if (r < 0 && ret >= 0) | |
1991 | ret = r; | |
1992 | } | |
1993 | ||
1994 | if (r < 0 && ret >= 0) | |
1995 | ret = r; | |
1996 | } | |
1997 | ||
1998 | return ret; | |
1999 | } | |
2000 | ||
11aef522 LP |
2001 | int unit_synthesize_cgroup_empty_event(Unit *u) { |
2002 | int r; | |
2003 | ||
2004 | assert(u); | |
2005 | ||
2006 | /* Enqueue a synthetic cgroup empty event if this unit doesn't watch any PIDs anymore. This is compatibility | |
2007 | * support for non-unified systems where notifications aren't reliable, and hence need to take whatever we can | |
2008 | * get as notification source as soon as we stopped having any useful PIDs to watch for. */ | |
2009 | ||
2010 | if (!u->cgroup_path) | |
2011 | return -ENOENT; | |
2012 | ||
2013 | r = cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER); | |
2014 | if (r < 0) | |
2015 | return r; | |
2016 | if (r > 0) /* On unified we have reliable notifications, and don't need this */ | |
2017 | return 0; | |
2018 | ||
2019 | if (!set_isempty(u->pids)) | |
2020 | return 0; | |
2021 | ||
2022 | unit_add_to_cgroup_empty_queue(u); | |
2023 | return 0; | |
2024 | } | |
2025 | ||
efdb0237 | 2026 | int unit_watch_all_pids(Unit *u) { |
b4cccbc1 LP |
2027 | int r; |
2028 | ||
efdb0237 LP |
2029 | assert(u); |
2030 | ||
2031 | /* Adds all PIDs from our cgroup to the set of PIDs we | |
2032 | * watch. This is a fallback logic for cases where we do not | |
2033 | * get reliable cgroup empty notifications: we try to use | |
2034 | * SIGCHLD as replacement. */ | |
2035 | ||
2036 | if (!u->cgroup_path) | |
2037 | return -ENOENT; | |
2038 | ||
c22800e4 | 2039 | r = cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER); |
b4cccbc1 LP |
2040 | if (r < 0) |
2041 | return r; | |
2042 | if (r > 0) /* On unified we can use proper notifications */ | |
efdb0237 LP |
2043 | return 0; |
2044 | ||
2045 | return unit_watch_pids_in_path(u, u->cgroup_path); | |
2046 | } | |
2047 | ||
09e24654 LP |
2048 | static int on_cgroup_empty_event(sd_event_source *s, void *userdata) { |
2049 | Manager *m = userdata; | |
2050 | Unit *u; | |
efdb0237 LP |
2051 | int r; |
2052 | ||
09e24654 LP |
2053 | assert(s); |
2054 | assert(m); | |
efdb0237 | 2055 | |
09e24654 LP |
2056 | u = m->cgroup_empty_queue; |
2057 | if (!u) | |
efdb0237 LP |
2058 | return 0; |
2059 | ||
09e24654 LP |
2060 | assert(u->in_cgroup_empty_queue); |
2061 | u->in_cgroup_empty_queue = false; | |
2062 | LIST_REMOVE(cgroup_empty_queue, m->cgroup_empty_queue, u); | |
2063 | ||
2064 | if (m->cgroup_empty_queue) { | |
2065 | /* More stuff queued, let's make sure we remain enabled */ | |
2066 | r = sd_event_source_set_enabled(s, SD_EVENT_ONESHOT); | |
2067 | if (r < 0) | |
19a691a9 | 2068 | log_debug_errno(r, "Failed to reenable cgroup empty event source, ignoring: %m"); |
09e24654 | 2069 | } |
efdb0237 LP |
2070 | |
2071 | unit_add_to_gc_queue(u); | |
2072 | ||
2073 | if (UNIT_VTABLE(u)->notify_cgroup_empty) | |
2074 | UNIT_VTABLE(u)->notify_cgroup_empty(u); | |
2075 | ||
2076 | return 0; | |
2077 | } | |
2078 | ||
09e24654 LP |
2079 | void unit_add_to_cgroup_empty_queue(Unit *u) { |
2080 | int r; | |
2081 | ||
2082 | assert(u); | |
2083 | ||
2084 | /* Note that there are four different ways how cgroup empty events reach us: | |
2085 | * | |
2086 | * 1. On the unified hierarchy we get an inotify event on the cgroup | |
2087 | * | |
2088 | * 2. On the legacy hierarchy, when running in system mode, we get a datagram on the cgroup agent socket | |
2089 | * | |
2090 | * 3. On the legacy hierarchy, when running in user mode, we get a D-Bus signal on the system bus | |
2091 | * | |
2092 | * 4. On the legacy hierarchy, in service units we start watching all processes of the cgroup for SIGCHLD as | |
2093 | * soon as we get one SIGCHLD, to deal with unreliable cgroup notifications. | |
2094 | * | |
2095 | * Regardless which way we got the notification, we'll verify it here, and then add it to a separate | |
2096 | * queue. This queue will be dispatched at a lower priority than the SIGCHLD handler, so that we always use | |
2097 | * SIGCHLD if we can get it first, and only use the cgroup empty notifications if there's no SIGCHLD pending | |
2098 | * (which might happen if the cgroup doesn't contain processes that are our own child, which is typically the | |
2099 | * case for scope units). */ | |
2100 | ||
2101 | if (u->in_cgroup_empty_queue) | |
2102 | return; | |
2103 | ||
2104 | /* Let's verify that the cgroup is really empty */ | |
2105 | if (!u->cgroup_path) | |
2106 | return; | |
2107 | r = cg_is_empty_recursive(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path); | |
2108 | if (r < 0) { | |
2109 | log_unit_debug_errno(u, r, "Failed to determine whether cgroup %s is empty: %m", u->cgroup_path); | |
2110 | return; | |
2111 | } | |
2112 | if (r == 0) | |
2113 | return; | |
2114 | ||
2115 | LIST_PREPEND(cgroup_empty_queue, u->manager->cgroup_empty_queue, u); | |
2116 | u->in_cgroup_empty_queue = true; | |
2117 | ||
2118 | /* Trigger the defer event */ | |
2119 | r = sd_event_source_set_enabled(u->manager->cgroup_empty_event_source, SD_EVENT_ONESHOT); | |
2120 | if (r < 0) | |
2121 | log_debug_errno(r, "Failed to enable cgroup empty event source: %m"); | |
2122 | } | |
2123 | ||
efdb0237 LP |
2124 | static int on_cgroup_inotify_event(sd_event_source *s, int fd, uint32_t revents, void *userdata) { |
2125 | Manager *m = userdata; | |
2126 | ||
2127 | assert(s); | |
2128 | assert(fd >= 0); | |
2129 | assert(m); | |
2130 | ||
2131 | for (;;) { | |
2132 | union inotify_event_buffer buffer; | |
2133 | struct inotify_event *e; | |
2134 | ssize_t l; | |
2135 | ||
2136 | l = read(fd, &buffer, sizeof(buffer)); | |
2137 | if (l < 0) { | |
47249640 | 2138 | if (IN_SET(errno, EINTR, EAGAIN)) |
efdb0237 LP |
2139 | return 0; |
2140 | ||
2141 | return log_error_errno(errno, "Failed to read control group inotify events: %m"); | |
2142 | } | |
2143 | ||
2144 | FOREACH_INOTIFY_EVENT(e, buffer, l) { | |
2145 | Unit *u; | |
2146 | ||
2147 | if (e->wd < 0) | |
2148 | /* Queue overflow has no watch descriptor */ | |
2149 | continue; | |
2150 | ||
2151 | if (e->mask & IN_IGNORED) | |
2152 | /* The watch was just removed */ | |
2153 | continue; | |
2154 | ||
2155 | u = hashmap_get(m->cgroup_inotify_wd_unit, INT_TO_PTR(e->wd)); | |
2156 | if (!u) /* Not that inotify might deliver | |
2157 | * events for a watch even after it | |
2158 | * was removed, because it was queued | |
2159 | * before the removal. Let's ignore | |
2160 | * this here safely. */ | |
2161 | continue; | |
2162 | ||
09e24654 | 2163 | unit_add_to_cgroup_empty_queue(u); |
efdb0237 LP |
2164 | } |
2165 | } | |
8e274523 LP |
2166 | } |
2167 | ||
8e274523 | 2168 | int manager_setup_cgroup(Manager *m) { |
9444b1f2 | 2169 | _cleanup_free_ char *path = NULL; |
10bd3e2e | 2170 | const char *scope_path; |
efdb0237 | 2171 | CGroupController c; |
b4cccbc1 | 2172 | int r, all_unified; |
efdb0237 | 2173 | char *e; |
8e274523 LP |
2174 | |
2175 | assert(m); | |
2176 | ||
35d2e7ec | 2177 | /* 1. Determine hierarchy */ |
efdb0237 | 2178 | m->cgroup_root = mfree(m->cgroup_root); |
9444b1f2 | 2179 | r = cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER, 0, &m->cgroup_root); |
23bbb0de MS |
2180 | if (r < 0) |
2181 | return log_error_errno(r, "Cannot determine cgroup we are running in: %m"); | |
8e274523 | 2182 | |
efdb0237 LP |
2183 | /* Chop off the init scope, if we are already located in it */ |
2184 | e = endswith(m->cgroup_root, "/" SPECIAL_INIT_SCOPE); | |
0d8c31ff | 2185 | |
efdb0237 LP |
2186 | /* LEGACY: Also chop off the system slice if we are in |
2187 | * it. This is to support live upgrades from older systemd | |
2188 | * versions where PID 1 was moved there. Also see | |
2189 | * cg_get_root_path(). */ | |
463d0d15 | 2190 | if (!e && MANAGER_IS_SYSTEM(m)) { |
9444b1f2 | 2191 | e = endswith(m->cgroup_root, "/" SPECIAL_SYSTEM_SLICE); |
15c60e99 | 2192 | if (!e) |
efdb0237 | 2193 | e = endswith(m->cgroup_root, "/system"); /* even more legacy */ |
0baf24dd | 2194 | } |
efdb0237 LP |
2195 | if (e) |
2196 | *e = 0; | |
7ccfb64a | 2197 | |
7546145e LP |
2198 | /* And make sure to store away the root value without trailing slash, even for the root dir, so that we can |
2199 | * easily prepend it everywhere. */ | |
2200 | delete_trailing_chars(m->cgroup_root, "/"); | |
8e274523 | 2201 | |
35d2e7ec | 2202 | /* 2. Show data */ |
9444b1f2 | 2203 | r = cg_get_path(SYSTEMD_CGROUP_CONTROLLER, m->cgroup_root, NULL, &path); |
23bbb0de MS |
2204 | if (r < 0) |
2205 | return log_error_errno(r, "Cannot find cgroup mount point: %m"); | |
8e274523 | 2206 | |
415fc41c TH |
2207 | r = cg_unified_flush(); |
2208 | if (r < 0) | |
2209 | return log_error_errno(r, "Couldn't determine if we are running in the unified hierarchy: %m"); | |
5da38d07 | 2210 | |
b4cccbc1 | 2211 | all_unified = cg_all_unified(); |
d4c819ed ZJS |
2212 | if (all_unified < 0) |
2213 | return log_error_errno(all_unified, "Couldn't determine whether we are in all unified mode: %m"); | |
2214 | if (all_unified > 0) | |
efdb0237 | 2215 | log_debug("Unified cgroup hierarchy is located at %s.", path); |
b4cccbc1 | 2216 | else { |
c22800e4 | 2217 | r = cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER); |
b4cccbc1 LP |
2218 | if (r < 0) |
2219 | return log_error_errno(r, "Failed to determine whether systemd's own controller is in unified mode: %m"); | |
2220 | if (r > 0) | |
2221 | log_debug("Unified cgroup hierarchy is located at %s. Controllers are on legacy hierarchies.", path); | |
2222 | else | |
2223 | log_debug("Using cgroup controller " SYSTEMD_CGROUP_CONTROLLER_LEGACY ". File system hierarchy is at %s.", path); | |
2224 | } | |
efdb0237 | 2225 | |
09e24654 LP |
2226 | /* 3. Allocate cgroup empty defer event source */ |
2227 | m->cgroup_empty_event_source = sd_event_source_unref(m->cgroup_empty_event_source); | |
2228 | r = sd_event_add_defer(m->event, &m->cgroup_empty_event_source, on_cgroup_empty_event, m); | |
2229 | if (r < 0) | |
2230 | return log_error_errno(r, "Failed to create cgroup empty event source: %m"); | |
2231 | ||
2232 | r = sd_event_source_set_priority(m->cgroup_empty_event_source, SD_EVENT_PRIORITY_NORMAL-5); | |
2233 | if (r < 0) | |
2234 | return log_error_errno(r, "Failed to set priority of cgroup empty event source: %m"); | |
2235 | ||
2236 | r = sd_event_source_set_enabled(m->cgroup_empty_event_source, SD_EVENT_OFF); | |
2237 | if (r < 0) | |
2238 | return log_error_errno(r, "Failed to disable cgroup empty event source: %m"); | |
2239 | ||
2240 | (void) sd_event_source_set_description(m->cgroup_empty_event_source, "cgroup-empty"); | |
2241 | ||
2242 | /* 4. Install notifier inotify object, or agent */ | |
10bd3e2e | 2243 | if (cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER) > 0) { |
c6c18be3 | 2244 | |
09e24654 | 2245 | /* In the unified hierarchy we can get cgroup empty notifications via inotify. */ |
efdb0237 | 2246 | |
10bd3e2e LP |
2247 | m->cgroup_inotify_event_source = sd_event_source_unref(m->cgroup_inotify_event_source); |
2248 | safe_close(m->cgroup_inotify_fd); | |
efdb0237 | 2249 | |
10bd3e2e LP |
2250 | m->cgroup_inotify_fd = inotify_init1(IN_NONBLOCK|IN_CLOEXEC); |
2251 | if (m->cgroup_inotify_fd < 0) | |
2252 | return log_error_errno(errno, "Failed to create control group inotify object: %m"); | |
efdb0237 | 2253 | |
10bd3e2e LP |
2254 | r = sd_event_add_io(m->event, &m->cgroup_inotify_event_source, m->cgroup_inotify_fd, EPOLLIN, on_cgroup_inotify_event, m); |
2255 | if (r < 0) | |
2256 | return log_error_errno(r, "Failed to watch control group inotify object: %m"); | |
efdb0237 | 2257 | |
10bd3e2e LP |
2258 | /* Process cgroup empty notifications early, but after service notifications and SIGCHLD. Also |
2259 | * see handling of cgroup agent notifications, for the classic cgroup hierarchy support. */ | |
09e24654 | 2260 | r = sd_event_source_set_priority(m->cgroup_inotify_event_source, SD_EVENT_PRIORITY_NORMAL-4); |
10bd3e2e LP |
2261 | if (r < 0) |
2262 | return log_error_errno(r, "Failed to set priority of inotify event source: %m"); | |
efdb0237 | 2263 | |
10bd3e2e | 2264 | (void) sd_event_source_set_description(m->cgroup_inotify_event_source, "cgroup-inotify"); |
efdb0237 | 2265 | |
10bd3e2e | 2266 | } else if (MANAGER_IS_SYSTEM(m) && m->test_run_flags == 0) { |
efdb0237 | 2267 | |
10bd3e2e LP |
2268 | /* On the legacy hierarchy we only get notifications via cgroup agents. (Which isn't really reliable, |
2269 | * since it does not generate events when control groups with children run empty. */ | |
8e274523 | 2270 | |
10bd3e2e | 2271 | r = cg_install_release_agent(SYSTEMD_CGROUP_CONTROLLER, SYSTEMD_CGROUP_AGENT_PATH); |
23bbb0de | 2272 | if (r < 0) |
10bd3e2e LP |
2273 | log_warning_errno(r, "Failed to install release agent, ignoring: %m"); |
2274 | else if (r > 0) | |
2275 | log_debug("Installed release agent."); | |
2276 | else if (r == 0) | |
2277 | log_debug("Release agent already installed."); | |
2278 | } | |
efdb0237 | 2279 | |
09e24654 | 2280 | /* 5. Make sure we are in the special "init.scope" unit in the root slice. */ |
10bd3e2e LP |
2281 | scope_path = strjoina(m->cgroup_root, "/" SPECIAL_INIT_SCOPE); |
2282 | r = cg_create_and_attach(SYSTEMD_CGROUP_CONTROLLER, scope_path, 0); | |
aa77e234 MS |
2283 | if (r >= 0) { |
2284 | /* Also, move all other userspace processes remaining in the root cgroup into that scope. */ | |
2285 | r = cg_migrate(SYSTEMD_CGROUP_CONTROLLER, m->cgroup_root, SYSTEMD_CGROUP_CONTROLLER, scope_path, 0); | |
2286 | if (r < 0) | |
2287 | log_warning_errno(r, "Couldn't move remaining userspace processes, ignoring: %m"); | |
c6c18be3 | 2288 | |
aa77e234 MS |
2289 | /* 6. And pin it, so that it cannot be unmounted */ |
2290 | safe_close(m->pin_cgroupfs_fd); | |
2291 | m->pin_cgroupfs_fd = open(path, O_RDONLY|O_CLOEXEC|O_DIRECTORY|O_NOCTTY|O_NONBLOCK); | |
2292 | if (m->pin_cgroupfs_fd < 0) | |
2293 | return log_error_errno(errno, "Failed to open pin file: %m"); | |
0d8c31ff | 2294 | |
b4dec49f | 2295 | } else if (!m->test_run_flags) |
aa77e234 | 2296 | return log_error_errno(r, "Failed to create %s control group: %m", scope_path); |
10bd3e2e | 2297 | |
09e24654 | 2298 | /* 7. Always enable hierarchical support if it exists... */ |
10bd3e2e LP |
2299 | if (!all_unified && m->test_run_flags == 0) |
2300 | (void) cg_set_attribute("memory", "/", "memory.use_hierarchy", "1"); | |
c6c18be3 | 2301 | |
09e24654 | 2302 | /* 8. Figure out which controllers are supported, and log about it */ |
efdb0237 LP |
2303 | r = cg_mask_supported(&m->cgroup_supported); |
2304 | if (r < 0) | |
2305 | return log_error_errno(r, "Failed to determine supported controllers: %m"); | |
efdb0237 | 2306 | for (c = 0; c < _CGROUP_CONTROLLER_MAX; c++) |
eee0a1e4 | 2307 | log_debug("Controller '%s' supported: %s", cgroup_controller_to_string(c), yes_no(m->cgroup_supported & CGROUP_CONTROLLER_TO_MASK(c))); |
9156e799 | 2308 | |
a32360f1 | 2309 | return 0; |
8e274523 LP |
2310 | } |
2311 | ||
c6c18be3 | 2312 | void manager_shutdown_cgroup(Manager *m, bool delete) { |
8e274523 LP |
2313 | assert(m); |
2314 | ||
9444b1f2 LP |
2315 | /* We can't really delete the group, since we are in it. But |
2316 | * let's trim it. */ | |
f6c63f6f | 2317 | if (delete && m->cgroup_root && m->test_run_flags != MANAGER_TEST_RUN_MINIMAL) |
efdb0237 LP |
2318 | (void) cg_trim(SYSTEMD_CGROUP_CONTROLLER, m->cgroup_root, false); |
2319 | ||
09e24654 LP |
2320 | m->cgroup_empty_event_source = sd_event_source_unref(m->cgroup_empty_event_source); |
2321 | ||
efdb0237 LP |
2322 | m->cgroup_inotify_wd_unit = hashmap_free(m->cgroup_inotify_wd_unit); |
2323 | ||
2324 | m->cgroup_inotify_event_source = sd_event_source_unref(m->cgroup_inotify_event_source); | |
2325 | m->cgroup_inotify_fd = safe_close(m->cgroup_inotify_fd); | |
8e274523 | 2326 | |
03e334a1 | 2327 | m->pin_cgroupfs_fd = safe_close(m->pin_cgroupfs_fd); |
c6c18be3 | 2328 | |
efdb0237 | 2329 | m->cgroup_root = mfree(m->cgroup_root); |
8e274523 LP |
2330 | } |
2331 | ||
4ad49000 | 2332 | Unit* manager_get_unit_by_cgroup(Manager *m, const char *cgroup) { |
acb14d31 | 2333 | char *p; |
4ad49000 | 2334 | Unit *u; |
acb14d31 LP |
2335 | |
2336 | assert(m); | |
2337 | assert(cgroup); | |
acb14d31 | 2338 | |
4ad49000 LP |
2339 | u = hashmap_get(m->cgroup_unit, cgroup); |
2340 | if (u) | |
2341 | return u; | |
acb14d31 | 2342 | |
8e70580b | 2343 | p = strdupa(cgroup); |
acb14d31 LP |
2344 | for (;;) { |
2345 | char *e; | |
2346 | ||
2347 | e = strrchr(p, '/'); | |
efdb0237 LP |
2348 | if (!e || e == p) |
2349 | return hashmap_get(m->cgroup_unit, SPECIAL_ROOT_SLICE); | |
acb14d31 LP |
2350 | |
2351 | *e = 0; | |
2352 | ||
4ad49000 LP |
2353 | u = hashmap_get(m->cgroup_unit, p); |
2354 | if (u) | |
2355 | return u; | |
acb14d31 LP |
2356 | } |
2357 | } | |
2358 | ||
b3ac818b | 2359 | Unit *manager_get_unit_by_pid_cgroup(Manager *m, pid_t pid) { |
4ad49000 | 2360 | _cleanup_free_ char *cgroup = NULL; |
8e274523 | 2361 | |
8c47c732 LP |
2362 | assert(m); |
2363 | ||
62a76913 | 2364 | if (!pid_is_valid(pid)) |
b3ac818b LP |
2365 | return NULL; |
2366 | ||
62a76913 | 2367 | if (cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER, pid, &cgroup) < 0) |
b3ac818b LP |
2368 | return NULL; |
2369 | ||
2370 | return manager_get_unit_by_cgroup(m, cgroup); | |
2371 | } | |
2372 | ||
2373 | Unit *manager_get_unit_by_pid(Manager *m, pid_t pid) { | |
62a76913 | 2374 | Unit *u, **array; |
b3ac818b LP |
2375 | |
2376 | assert(m); | |
2377 | ||
62a76913 LP |
2378 | /* Note that a process might be owned by multiple units, we return only one here, which is good enough for most |
2379 | * cases, though not strictly correct. We prefer the one reported by cgroup membership, as that's the most | |
2380 | * relevant one as children of the process will be assigned to that one, too, before all else. */ | |
2381 | ||
2382 | if (!pid_is_valid(pid)) | |
8c47c732 LP |
2383 | return NULL; |
2384 | ||
2ca9d979 | 2385 | if (pid == getpid_cached()) |
efdb0237 LP |
2386 | return hashmap_get(m->units, SPECIAL_INIT_SCOPE); |
2387 | ||
62a76913 | 2388 | u = manager_get_unit_by_pid_cgroup(m, pid); |
5fe8876b LP |
2389 | if (u) |
2390 | return u; | |
2391 | ||
62a76913 | 2392 | u = hashmap_get(m->watch_pids, PID_TO_PTR(pid)); |
5fe8876b LP |
2393 | if (u) |
2394 | return u; | |
2395 | ||
62a76913 LP |
2396 | array = hashmap_get(m->watch_pids, PID_TO_PTR(-pid)); |
2397 | if (array) | |
2398 | return array[0]; | |
2399 | ||
2400 | return NULL; | |
6dde1f33 | 2401 | } |
4fbf50b3 | 2402 | |
4ad49000 LP |
2403 | int manager_notify_cgroup_empty(Manager *m, const char *cgroup) { |
2404 | Unit *u; | |
4fbf50b3 | 2405 | |
4ad49000 LP |
2406 | assert(m); |
2407 | assert(cgroup); | |
4fbf50b3 | 2408 | |
09e24654 LP |
2409 | /* Called on the legacy hierarchy whenever we get an explicit cgroup notification from the cgroup agent process |
2410 | * or from the --system instance */ | |
2411 | ||
d8fdc620 LP |
2412 | log_debug("Got cgroup empty notification for: %s", cgroup); |
2413 | ||
4ad49000 | 2414 | u = manager_get_unit_by_cgroup(m, cgroup); |
5ad096b3 LP |
2415 | if (!u) |
2416 | return 0; | |
b56c28c3 | 2417 | |
09e24654 LP |
2418 | unit_add_to_cgroup_empty_queue(u); |
2419 | return 1; | |
5ad096b3 LP |
2420 | } |
2421 | ||
2422 | int unit_get_memory_current(Unit *u, uint64_t *ret) { | |
2423 | _cleanup_free_ char *v = NULL; | |
2424 | int r; | |
2425 | ||
2426 | assert(u); | |
2427 | assert(ret); | |
2428 | ||
2e4025c0 | 2429 | if (!UNIT_CGROUP_BOOL(u, memory_accounting)) |
cf3b4be1 LP |
2430 | return -ENODATA; |
2431 | ||
5ad096b3 LP |
2432 | if (!u->cgroup_path) |
2433 | return -ENODATA; | |
2434 | ||
1f73aa00 LP |
2435 | /* The root cgroup doesn't expose this information, let's get it from /proc instead */ |
2436 | if (unit_has_root_cgroup(u)) | |
2437 | return procfs_memory_get_current(ret); | |
2438 | ||
efdb0237 | 2439 | if ((u->cgroup_realized_mask & CGROUP_MASK_MEMORY) == 0) |
5ad096b3 LP |
2440 | return -ENODATA; |
2441 | ||
b4cccbc1 LP |
2442 | r = cg_all_unified(); |
2443 | if (r < 0) | |
2444 | return r; | |
2445 | if (r > 0) | |
efdb0237 | 2446 | r = cg_get_attribute("memory", u->cgroup_path, "memory.current", &v); |
b4cccbc1 LP |
2447 | else |
2448 | r = cg_get_attribute("memory", u->cgroup_path, "memory.usage_in_bytes", &v); | |
5ad096b3 LP |
2449 | if (r == -ENOENT) |
2450 | return -ENODATA; | |
2451 | if (r < 0) | |
2452 | return r; | |
2453 | ||
2454 | return safe_atou64(v, ret); | |
2455 | } | |
2456 | ||
03a7b521 LP |
2457 | int unit_get_tasks_current(Unit *u, uint64_t *ret) { |
2458 | _cleanup_free_ char *v = NULL; | |
2459 | int r; | |
2460 | ||
2461 | assert(u); | |
2462 | assert(ret); | |
2463 | ||
2e4025c0 | 2464 | if (!UNIT_CGROUP_BOOL(u, tasks_accounting)) |
cf3b4be1 LP |
2465 | return -ENODATA; |
2466 | ||
03a7b521 LP |
2467 | if (!u->cgroup_path) |
2468 | return -ENODATA; | |
2469 | ||
c36a69f4 LP |
2470 | /* The root cgroup doesn't expose this information, let's get it from /proc instead */ |
2471 | if (unit_has_root_cgroup(u)) | |
2472 | return procfs_tasks_get_current(ret); | |
2473 | ||
1f73aa00 LP |
2474 | if ((u->cgroup_realized_mask & CGROUP_MASK_PIDS) == 0) |
2475 | return -ENODATA; | |
2476 | ||
03a7b521 LP |
2477 | r = cg_get_attribute("pids", u->cgroup_path, "pids.current", &v); |
2478 | if (r == -ENOENT) | |
2479 | return -ENODATA; | |
2480 | if (r < 0) | |
2481 | return r; | |
2482 | ||
2483 | return safe_atou64(v, ret); | |
2484 | } | |
2485 | ||
5ad096b3 LP |
2486 | static int unit_get_cpu_usage_raw(Unit *u, nsec_t *ret) { |
2487 | _cleanup_free_ char *v = NULL; | |
2488 | uint64_t ns; | |
2489 | int r; | |
2490 | ||
2491 | assert(u); | |
2492 | assert(ret); | |
2493 | ||
2494 | if (!u->cgroup_path) | |
2495 | return -ENODATA; | |
2496 | ||
1f73aa00 LP |
2497 | /* The root cgroup doesn't expose this information, let's get it from /proc instead */ |
2498 | if (unit_has_root_cgroup(u)) | |
2499 | return procfs_cpu_get_usage(ret); | |
2500 | ||
b4cccbc1 LP |
2501 | r = cg_all_unified(); |
2502 | if (r < 0) | |
2503 | return r; | |
2504 | if (r > 0) { | |
66ebf6c0 TH |
2505 | _cleanup_free_ char *val = NULL; |
2506 | uint64_t us; | |
5ad096b3 | 2507 | |
66ebf6c0 TH |
2508 | if ((u->cgroup_realized_mask & CGROUP_MASK_CPU) == 0) |
2509 | return -ENODATA; | |
5ad096b3 | 2510 | |
b734a4ff | 2511 | r = cg_get_keyed_attribute("cpu", u->cgroup_path, "cpu.stat", STRV_MAKE("usage_usec"), &val); |
66ebf6c0 TH |
2512 | if (r < 0) |
2513 | return r; | |
b734a4ff LP |
2514 | if (IN_SET(r, -ENOENT, -ENXIO)) |
2515 | return -ENODATA; | |
66ebf6c0 TH |
2516 | |
2517 | r = safe_atou64(val, &us); | |
2518 | if (r < 0) | |
2519 | return r; | |
2520 | ||
2521 | ns = us * NSEC_PER_USEC; | |
2522 | } else { | |
2523 | if ((u->cgroup_realized_mask & CGROUP_MASK_CPUACCT) == 0) | |
2524 | return -ENODATA; | |
2525 | ||
2526 | r = cg_get_attribute("cpuacct", u->cgroup_path, "cpuacct.usage", &v); | |
2527 | if (r == -ENOENT) | |
2528 | return -ENODATA; | |
2529 | if (r < 0) | |
2530 | return r; | |
2531 | ||
2532 | r = safe_atou64(v, &ns); | |
2533 | if (r < 0) | |
2534 | return r; | |
2535 | } | |
5ad096b3 LP |
2536 | |
2537 | *ret = ns; | |
2538 | return 0; | |
2539 | } | |
2540 | ||
2541 | int unit_get_cpu_usage(Unit *u, nsec_t *ret) { | |
2542 | nsec_t ns; | |
2543 | int r; | |
2544 | ||
fe700f46 LP |
2545 | assert(u); |
2546 | ||
2547 | /* Retrieve the current CPU usage counter. This will subtract the CPU counter taken when the unit was | |
2548 | * started. If the cgroup has been removed already, returns the last cached value. To cache the value, simply | |
2549 | * call this function with a NULL return value. */ | |
2550 | ||
2e4025c0 | 2551 | if (!UNIT_CGROUP_BOOL(u, cpu_accounting)) |
cf3b4be1 LP |
2552 | return -ENODATA; |
2553 | ||
5ad096b3 | 2554 | r = unit_get_cpu_usage_raw(u, &ns); |
fe700f46 LP |
2555 | if (r == -ENODATA && u->cpu_usage_last != NSEC_INFINITY) { |
2556 | /* If we can't get the CPU usage anymore (because the cgroup was already removed, for example), use our | |
2557 | * cached value. */ | |
2558 | ||
2559 | if (ret) | |
2560 | *ret = u->cpu_usage_last; | |
2561 | return 0; | |
2562 | } | |
5ad096b3 LP |
2563 | if (r < 0) |
2564 | return r; | |
2565 | ||
66ebf6c0 TH |
2566 | if (ns > u->cpu_usage_base) |
2567 | ns -= u->cpu_usage_base; | |
5ad096b3 LP |
2568 | else |
2569 | ns = 0; | |
2570 | ||
fe700f46 LP |
2571 | u->cpu_usage_last = ns; |
2572 | if (ret) | |
2573 | *ret = ns; | |
2574 | ||
5ad096b3 LP |
2575 | return 0; |
2576 | } | |
2577 | ||
906c06f6 DM |
2578 | int unit_get_ip_accounting( |
2579 | Unit *u, | |
2580 | CGroupIPAccountingMetric metric, | |
2581 | uint64_t *ret) { | |
2582 | ||
6b659ed8 | 2583 | uint64_t value; |
906c06f6 DM |
2584 | int fd, r; |
2585 | ||
2586 | assert(u); | |
2587 | assert(metric >= 0); | |
2588 | assert(metric < _CGROUP_IP_ACCOUNTING_METRIC_MAX); | |
2589 | assert(ret); | |
2590 | ||
2e4025c0 | 2591 | if (!UNIT_CGROUP_BOOL(u, ip_accounting)) |
cf3b4be1 LP |
2592 | return -ENODATA; |
2593 | ||
906c06f6 DM |
2594 | fd = IN_SET(metric, CGROUP_IP_INGRESS_BYTES, CGROUP_IP_INGRESS_PACKETS) ? |
2595 | u->ip_accounting_ingress_map_fd : | |
2596 | u->ip_accounting_egress_map_fd; | |
906c06f6 DM |
2597 | if (fd < 0) |
2598 | return -ENODATA; | |
2599 | ||
2600 | if (IN_SET(metric, CGROUP_IP_INGRESS_BYTES, CGROUP_IP_EGRESS_BYTES)) | |
6b659ed8 | 2601 | r = bpf_firewall_read_accounting(fd, &value, NULL); |
906c06f6 | 2602 | else |
6b659ed8 LP |
2603 | r = bpf_firewall_read_accounting(fd, NULL, &value); |
2604 | if (r < 0) | |
2605 | return r; | |
2606 | ||
2607 | /* Add in additional metrics from a previous runtime. Note that when reexecing/reloading the daemon we compile | |
2608 | * all BPF programs and maps anew, but serialize the old counters. When deserializing we store them in the | |
2609 | * ip_accounting_extra[] field, and add them in here transparently. */ | |
2610 | ||
2611 | *ret = value + u->ip_accounting_extra[metric]; | |
906c06f6 DM |
2612 | |
2613 | return r; | |
2614 | } | |
2615 | ||
2616 | int unit_reset_cpu_accounting(Unit *u) { | |
5ad096b3 LP |
2617 | nsec_t ns; |
2618 | int r; | |
2619 | ||
2620 | assert(u); | |
2621 | ||
fe700f46 LP |
2622 | u->cpu_usage_last = NSEC_INFINITY; |
2623 | ||
5ad096b3 LP |
2624 | r = unit_get_cpu_usage_raw(u, &ns); |
2625 | if (r < 0) { | |
66ebf6c0 | 2626 | u->cpu_usage_base = 0; |
5ad096b3 | 2627 | return r; |
b56c28c3 | 2628 | } |
2633eb83 | 2629 | |
66ebf6c0 | 2630 | u->cpu_usage_base = ns; |
4ad49000 | 2631 | return 0; |
4fbf50b3 LP |
2632 | } |
2633 | ||
906c06f6 DM |
2634 | int unit_reset_ip_accounting(Unit *u) { |
2635 | int r = 0, q = 0; | |
2636 | ||
2637 | assert(u); | |
2638 | ||
2639 | if (u->ip_accounting_ingress_map_fd >= 0) | |
2640 | r = bpf_firewall_reset_accounting(u->ip_accounting_ingress_map_fd); | |
2641 | ||
2642 | if (u->ip_accounting_egress_map_fd >= 0) | |
2643 | q = bpf_firewall_reset_accounting(u->ip_accounting_egress_map_fd); | |
2644 | ||
6b659ed8 LP |
2645 | zero(u->ip_accounting_extra); |
2646 | ||
906c06f6 DM |
2647 | return r < 0 ? r : q; |
2648 | } | |
2649 | ||
e7ab4d1a LP |
2650 | void unit_invalidate_cgroup(Unit *u, CGroupMask m) { |
2651 | assert(u); | |
2652 | ||
2653 | if (!UNIT_HAS_CGROUP_CONTEXT(u)) | |
2654 | return; | |
2655 | ||
2656 | if (m == 0) | |
2657 | return; | |
2658 | ||
538b4852 TH |
2659 | /* always invalidate compat pairs together */ |
2660 | if (m & (CGROUP_MASK_IO | CGROUP_MASK_BLKIO)) | |
2661 | m |= CGROUP_MASK_IO | CGROUP_MASK_BLKIO; | |
2662 | ||
7cce4fb7 LP |
2663 | if (m & (CGROUP_MASK_CPU | CGROUP_MASK_CPUACCT)) |
2664 | m |= CGROUP_MASK_CPU | CGROUP_MASK_CPUACCT; | |
2665 | ||
60c728ad | 2666 | if ((u->cgroup_realized_mask & m) == 0) /* NOP? */ |
e7ab4d1a LP |
2667 | return; |
2668 | ||
2669 | u->cgroup_realized_mask &= ~m; | |
91a6073e | 2670 | unit_add_to_cgroup_realize_queue(u); |
e7ab4d1a LP |
2671 | } |
2672 | ||
906c06f6 DM |
2673 | void unit_invalidate_cgroup_bpf(Unit *u) { |
2674 | assert(u); | |
2675 | ||
2676 | if (!UNIT_HAS_CGROUP_CONTEXT(u)) | |
2677 | return; | |
2678 | ||
60c728ad | 2679 | if (u->cgroup_bpf_state == UNIT_CGROUP_BPF_INVALIDATED) /* NOP? */ |
906c06f6 DM |
2680 | return; |
2681 | ||
2682 | u->cgroup_bpf_state = UNIT_CGROUP_BPF_INVALIDATED; | |
91a6073e | 2683 | unit_add_to_cgroup_realize_queue(u); |
906c06f6 DM |
2684 | |
2685 | /* If we are a slice unit, we also need to put compile a new BPF program for all our children, as the IP access | |
2686 | * list of our children includes our own. */ | |
2687 | if (u->type == UNIT_SLICE) { | |
2688 | Unit *member; | |
2689 | Iterator i; | |
eef85c4a | 2690 | void *v; |
906c06f6 | 2691 | |
eef85c4a | 2692 | HASHMAP_FOREACH_KEY(v, member, u->dependencies[UNIT_BEFORE], i) { |
906c06f6 DM |
2693 | if (member == u) |
2694 | continue; | |
2695 | ||
2696 | if (UNIT_DEREF(member->slice) != u) | |
2697 | continue; | |
2698 | ||
2699 | unit_invalidate_cgroup_bpf(member); | |
2700 | } | |
2701 | } | |
2702 | } | |
2703 | ||
1d9cc876 LP |
2704 | bool unit_cgroup_delegate(Unit *u) { |
2705 | CGroupContext *c; | |
2706 | ||
2707 | assert(u); | |
2708 | ||
2709 | if (!UNIT_VTABLE(u)->can_delegate) | |
2710 | return false; | |
2711 | ||
2712 | c = unit_get_cgroup_context(u); | |
2713 | if (!c) | |
2714 | return false; | |
2715 | ||
2716 | return c->delegate; | |
2717 | } | |
2718 | ||
e7ab4d1a LP |
2719 | void manager_invalidate_startup_units(Manager *m) { |
2720 | Iterator i; | |
2721 | Unit *u; | |
2722 | ||
2723 | assert(m); | |
2724 | ||
2725 | SET_FOREACH(u, m->startup_units, i) | |
13c31542 | 2726 | unit_invalidate_cgroup(u, CGROUP_MASK_CPU|CGROUP_MASK_IO|CGROUP_MASK_BLKIO); |
e7ab4d1a LP |
2727 | } |
2728 | ||
4ad49000 LP |
2729 | static const char* const cgroup_device_policy_table[_CGROUP_DEVICE_POLICY_MAX] = { |
2730 | [CGROUP_AUTO] = "auto", | |
2731 | [CGROUP_CLOSED] = "closed", | |
2732 | [CGROUP_STRICT] = "strict", | |
2733 | }; | |
4fbf50b3 | 2734 | |
4ad49000 | 2735 | DEFINE_STRING_TABLE_LOOKUP(cgroup_device_policy, CGroupDevicePolicy); |