]>
Commit | Line | Data |
---|---|---|
53e1b683 | 1 | /* SPDX-License-Identifier: LGPL-2.1+ */ |
8e274523 LP |
2 | /*** |
3 | This file is part of systemd. | |
4 | ||
4ad49000 | 5 | Copyright 2013 Lennart Poettering |
8e274523 LP |
6 | ***/ |
7 | ||
c6c18be3 | 8 | #include <fcntl.h> |
e41969e3 | 9 | #include <fnmatch.h> |
8c6db833 | 10 | |
b5efdb8a | 11 | #include "alloc-util.h" |
18c528e9 | 12 | #include "blockdev-util.h" |
906c06f6 | 13 | #include "bpf-firewall.h" |
6592b975 | 14 | #include "bus-error.h" |
03a7b521 | 15 | #include "cgroup-util.h" |
3ffd4af2 LP |
16 | #include "cgroup.h" |
17 | #include "fd-util.h" | |
0d39fa9c | 18 | #include "fileio.h" |
77601719 | 19 | #include "fs-util.h" |
6bedfcbb | 20 | #include "parse-util.h" |
9eb977db | 21 | #include "path-util.h" |
03a7b521 | 22 | #include "process-util.h" |
c36a69f4 | 23 | #include "procfs-util.h" |
9444b1f2 | 24 | #include "special.h" |
906c06f6 | 25 | #include "stdio-util.h" |
8b43440b | 26 | #include "string-table.h" |
07630cea | 27 | #include "string-util.h" |
cc6271f1 | 28 | #include "virt.h" |
8e274523 | 29 | |
9a054909 LP |
30 | #define CGROUP_CPU_QUOTA_PERIOD_USEC ((usec_t) 100 * USEC_PER_MSEC) |
31 | ||
cc6271f1 LP |
32 | bool manager_owns_root_cgroup(Manager *m) { |
33 | assert(m); | |
34 | ||
35 | /* Returns true if we are managing the root cgroup. Note that it isn't sufficient to just check whether the | |
36 | * group root path equals "/" since that will also be the case if CLONE_NEWCGROUP is in the mix. Since there's | |
37 | * appears to be no nice way to detect whether we are in a CLONE_NEWCGROUP namespace we instead just check if | |
38 | * we run in any kind of container virtualization. */ | |
39 | ||
40 | if (detect_container() > 0) | |
41 | return false; | |
42 | ||
57ea45e1 | 43 | return empty_or_root(m->cgroup_root); |
cc6271f1 LP |
44 | } |
45 | ||
f3725e64 LP |
46 | bool unit_has_root_cgroup(Unit *u) { |
47 | assert(u); | |
48 | ||
cc6271f1 LP |
49 | /* Returns whether this unit manages the root cgroup. This will return true if this unit is the root slice and |
50 | * the manager manages the root cgroup. */ | |
f3725e64 | 51 | |
cc6271f1 | 52 | if (!manager_owns_root_cgroup(u->manager)) |
f3725e64 LP |
53 | return false; |
54 | ||
cc6271f1 | 55 | return unit_has_name(u, SPECIAL_ROOT_SLICE); |
f3725e64 LP |
56 | } |
57 | ||
2b40998d | 58 | static void cgroup_compat_warn(void) { |
128fadc9 TH |
59 | static bool cgroup_compat_warned = false; |
60 | ||
61 | if (cgroup_compat_warned) | |
62 | return; | |
63 | ||
cc6271f1 LP |
64 | log_warning("cgroup compatibility translation between legacy and unified hierarchy settings activated. " |
65 | "See cgroup-compat debug messages for details."); | |
66 | ||
128fadc9 TH |
67 | cgroup_compat_warned = true; |
68 | } | |
69 | ||
70 | #define log_cgroup_compat(unit, fmt, ...) do { \ | |
71 | cgroup_compat_warn(); \ | |
72 | log_unit_debug(unit, "cgroup-compat: " fmt, ##__VA_ARGS__); \ | |
2b40998d | 73 | } while (false) |
128fadc9 | 74 | |
4ad49000 LP |
75 | void cgroup_context_init(CGroupContext *c) { |
76 | assert(c); | |
77 | ||
78 | /* Initialize everything to the kernel defaults, assuming the | |
79 | * structure is preinitialized to 0 */ | |
80 | ||
66ebf6c0 TH |
81 | c->cpu_weight = CGROUP_WEIGHT_INVALID; |
82 | c->startup_cpu_weight = CGROUP_WEIGHT_INVALID; | |
83 | c->cpu_quota_per_sec_usec = USEC_INFINITY; | |
84 | ||
d53d9474 LP |
85 | c->cpu_shares = CGROUP_CPU_SHARES_INVALID; |
86 | c->startup_cpu_shares = CGROUP_CPU_SHARES_INVALID; | |
d53d9474 | 87 | |
da4d897e TH |
88 | c->memory_high = CGROUP_LIMIT_MAX; |
89 | c->memory_max = CGROUP_LIMIT_MAX; | |
96e131ea | 90 | c->memory_swap_max = CGROUP_LIMIT_MAX; |
da4d897e TH |
91 | |
92 | c->memory_limit = CGROUP_LIMIT_MAX; | |
b2f8b02e | 93 | |
13c31542 TH |
94 | c->io_weight = CGROUP_WEIGHT_INVALID; |
95 | c->startup_io_weight = CGROUP_WEIGHT_INVALID; | |
96 | ||
d53d9474 LP |
97 | c->blockio_weight = CGROUP_BLKIO_WEIGHT_INVALID; |
98 | c->startup_blockio_weight = CGROUP_BLKIO_WEIGHT_INVALID; | |
99 | ||
100 | c->tasks_max = (uint64_t) -1; | |
4ad49000 | 101 | } |
8e274523 | 102 | |
4ad49000 LP |
103 | void cgroup_context_free_device_allow(CGroupContext *c, CGroupDeviceAllow *a) { |
104 | assert(c); | |
105 | assert(a); | |
106 | ||
71fda00f | 107 | LIST_REMOVE(device_allow, c->device_allow, a); |
4ad49000 LP |
108 | free(a->path); |
109 | free(a); | |
110 | } | |
111 | ||
13c31542 TH |
112 | void cgroup_context_free_io_device_weight(CGroupContext *c, CGroupIODeviceWeight *w) { |
113 | assert(c); | |
114 | assert(w); | |
115 | ||
116 | LIST_REMOVE(device_weights, c->io_device_weights, w); | |
117 | free(w->path); | |
118 | free(w); | |
119 | } | |
120 | ||
121 | void cgroup_context_free_io_device_limit(CGroupContext *c, CGroupIODeviceLimit *l) { | |
122 | assert(c); | |
123 | assert(l); | |
124 | ||
125 | LIST_REMOVE(device_limits, c->io_device_limits, l); | |
126 | free(l->path); | |
127 | free(l); | |
128 | } | |
129 | ||
4ad49000 LP |
130 | void cgroup_context_free_blockio_device_weight(CGroupContext *c, CGroupBlockIODeviceWeight *w) { |
131 | assert(c); | |
132 | assert(w); | |
133 | ||
71fda00f | 134 | LIST_REMOVE(device_weights, c->blockio_device_weights, w); |
4ad49000 LP |
135 | free(w->path); |
136 | free(w); | |
137 | } | |
138 | ||
139 | void cgroup_context_free_blockio_device_bandwidth(CGroupContext *c, CGroupBlockIODeviceBandwidth *b) { | |
140 | assert(c); | |
8e274523 | 141 | assert(b); |
8e274523 | 142 | |
71fda00f | 143 | LIST_REMOVE(device_bandwidths, c->blockio_device_bandwidths, b); |
4ad49000 LP |
144 | free(b->path); |
145 | free(b); | |
146 | } | |
147 | ||
148 | void cgroup_context_done(CGroupContext *c) { | |
149 | assert(c); | |
150 | ||
13c31542 TH |
151 | while (c->io_device_weights) |
152 | cgroup_context_free_io_device_weight(c, c->io_device_weights); | |
153 | ||
154 | while (c->io_device_limits) | |
155 | cgroup_context_free_io_device_limit(c, c->io_device_limits); | |
156 | ||
4ad49000 LP |
157 | while (c->blockio_device_weights) |
158 | cgroup_context_free_blockio_device_weight(c, c->blockio_device_weights); | |
159 | ||
160 | while (c->blockio_device_bandwidths) | |
161 | cgroup_context_free_blockio_device_bandwidth(c, c->blockio_device_bandwidths); | |
162 | ||
163 | while (c->device_allow) | |
164 | cgroup_context_free_device_allow(c, c->device_allow); | |
6a48d82f DM |
165 | |
166 | c->ip_address_allow = ip_address_access_free_all(c->ip_address_allow); | |
167 | c->ip_address_deny = ip_address_access_free_all(c->ip_address_deny); | |
4ad49000 LP |
168 | } |
169 | ||
170 | void cgroup_context_dump(CGroupContext *c, FILE* f, const char *prefix) { | |
13c31542 TH |
171 | CGroupIODeviceLimit *il; |
172 | CGroupIODeviceWeight *iw; | |
4ad49000 LP |
173 | CGroupBlockIODeviceBandwidth *b; |
174 | CGroupBlockIODeviceWeight *w; | |
175 | CGroupDeviceAllow *a; | |
c21c9906 | 176 | IPAddressAccessItem *iaai; |
9a054909 | 177 | char u[FORMAT_TIMESPAN_MAX]; |
4ad49000 LP |
178 | |
179 | assert(c); | |
180 | assert(f); | |
181 | ||
182 | prefix = strempty(prefix); | |
183 | ||
184 | fprintf(f, | |
185 | "%sCPUAccounting=%s\n" | |
13c31542 | 186 | "%sIOAccounting=%s\n" |
4ad49000 LP |
187 | "%sBlockIOAccounting=%s\n" |
188 | "%sMemoryAccounting=%s\n" | |
d53d9474 | 189 | "%sTasksAccounting=%s\n" |
c21c9906 | 190 | "%sIPAccounting=%s\n" |
66ebf6c0 TH |
191 | "%sCPUWeight=%" PRIu64 "\n" |
192 | "%sStartupCPUWeight=%" PRIu64 "\n" | |
d53d9474 LP |
193 | "%sCPUShares=%" PRIu64 "\n" |
194 | "%sStartupCPUShares=%" PRIu64 "\n" | |
b2f8b02e | 195 | "%sCPUQuotaPerSecSec=%s\n" |
13c31542 TH |
196 | "%sIOWeight=%" PRIu64 "\n" |
197 | "%sStartupIOWeight=%" PRIu64 "\n" | |
d53d9474 LP |
198 | "%sBlockIOWeight=%" PRIu64 "\n" |
199 | "%sStartupBlockIOWeight=%" PRIu64 "\n" | |
da4d897e TH |
200 | "%sMemoryLow=%" PRIu64 "\n" |
201 | "%sMemoryHigh=%" PRIu64 "\n" | |
202 | "%sMemoryMax=%" PRIu64 "\n" | |
96e131ea | 203 | "%sMemorySwapMax=%" PRIu64 "\n" |
4ad49000 | 204 | "%sMemoryLimit=%" PRIu64 "\n" |
03a7b521 | 205 | "%sTasksMax=%" PRIu64 "\n" |
a931ad47 LP |
206 | "%sDevicePolicy=%s\n" |
207 | "%sDelegate=%s\n", | |
4ad49000 | 208 | prefix, yes_no(c->cpu_accounting), |
13c31542 | 209 | prefix, yes_no(c->io_accounting), |
4ad49000 LP |
210 | prefix, yes_no(c->blockio_accounting), |
211 | prefix, yes_no(c->memory_accounting), | |
d53d9474 | 212 | prefix, yes_no(c->tasks_accounting), |
c21c9906 | 213 | prefix, yes_no(c->ip_accounting), |
66ebf6c0 TH |
214 | prefix, c->cpu_weight, |
215 | prefix, c->startup_cpu_weight, | |
4ad49000 | 216 | prefix, c->cpu_shares, |
95ae05c0 | 217 | prefix, c->startup_cpu_shares, |
b1d6dcf5 | 218 | prefix, format_timespan(u, sizeof(u), c->cpu_quota_per_sec_usec, 1), |
13c31542 TH |
219 | prefix, c->io_weight, |
220 | prefix, c->startup_io_weight, | |
4ad49000 | 221 | prefix, c->blockio_weight, |
95ae05c0 | 222 | prefix, c->startup_blockio_weight, |
da4d897e TH |
223 | prefix, c->memory_low, |
224 | prefix, c->memory_high, | |
225 | prefix, c->memory_max, | |
96e131ea | 226 | prefix, c->memory_swap_max, |
4ad49000 | 227 | prefix, c->memory_limit, |
03a7b521 | 228 | prefix, c->tasks_max, |
a931ad47 LP |
229 | prefix, cgroup_device_policy_to_string(c->device_policy), |
230 | prefix, yes_no(c->delegate)); | |
4ad49000 | 231 | |
02638280 LP |
232 | if (c->delegate) { |
233 | _cleanup_free_ char *t = NULL; | |
234 | ||
235 | (void) cg_mask_to_string(c->delegate_controllers, &t); | |
236 | ||
47a78d41 | 237 | fprintf(f, "%sDelegateControllers=%s\n", |
02638280 LP |
238 | prefix, |
239 | strempty(t)); | |
240 | } | |
241 | ||
4ad49000 LP |
242 | LIST_FOREACH(device_allow, a, c->device_allow) |
243 | fprintf(f, | |
244 | "%sDeviceAllow=%s %s%s%s\n", | |
245 | prefix, | |
246 | a->path, | |
247 | a->r ? "r" : "", a->w ? "w" : "", a->m ? "m" : ""); | |
248 | ||
13c31542 TH |
249 | LIST_FOREACH(device_weights, iw, c->io_device_weights) |
250 | fprintf(f, | |
251 | "%sIODeviceWeight=%s %" PRIu64, | |
252 | prefix, | |
253 | iw->path, | |
254 | iw->weight); | |
255 | ||
256 | LIST_FOREACH(device_limits, il, c->io_device_limits) { | |
257 | char buf[FORMAT_BYTES_MAX]; | |
9be57249 TH |
258 | CGroupIOLimitType type; |
259 | ||
260 | for (type = 0; type < _CGROUP_IO_LIMIT_TYPE_MAX; type++) | |
261 | if (il->limits[type] != cgroup_io_limit_defaults[type]) | |
262 | fprintf(f, | |
263 | "%s%s=%s %s\n", | |
264 | prefix, | |
265 | cgroup_io_limit_type_to_string(type), | |
266 | il->path, | |
267 | format_bytes(buf, sizeof(buf), il->limits[type])); | |
13c31542 TH |
268 | } |
269 | ||
4ad49000 LP |
270 | LIST_FOREACH(device_weights, w, c->blockio_device_weights) |
271 | fprintf(f, | |
d53d9474 | 272 | "%sBlockIODeviceWeight=%s %" PRIu64, |
4ad49000 LP |
273 | prefix, |
274 | w->path, | |
275 | w->weight); | |
276 | ||
277 | LIST_FOREACH(device_bandwidths, b, c->blockio_device_bandwidths) { | |
278 | char buf[FORMAT_BYTES_MAX]; | |
279 | ||
979d0311 TH |
280 | if (b->rbps != CGROUP_LIMIT_MAX) |
281 | fprintf(f, | |
282 | "%sBlockIOReadBandwidth=%s %s\n", | |
283 | prefix, | |
284 | b->path, | |
285 | format_bytes(buf, sizeof(buf), b->rbps)); | |
286 | if (b->wbps != CGROUP_LIMIT_MAX) | |
287 | fprintf(f, | |
288 | "%sBlockIOWriteBandwidth=%s %s\n", | |
289 | prefix, | |
290 | b->path, | |
291 | format_bytes(buf, sizeof(buf), b->wbps)); | |
4ad49000 | 292 | } |
c21c9906 LP |
293 | |
294 | LIST_FOREACH(items, iaai, c->ip_address_allow) { | |
295 | _cleanup_free_ char *k = NULL; | |
296 | ||
297 | (void) in_addr_to_string(iaai->family, &iaai->address, &k); | |
298 | fprintf(f, "%sIPAddressAllow=%s/%u\n", prefix, strnull(k), iaai->prefixlen); | |
299 | } | |
300 | ||
301 | LIST_FOREACH(items, iaai, c->ip_address_deny) { | |
302 | _cleanup_free_ char *k = NULL; | |
303 | ||
304 | (void) in_addr_to_string(iaai->family, &iaai->address, &k); | |
305 | fprintf(f, "%sIPAddressDeny=%s/%u\n", prefix, strnull(k), iaai->prefixlen); | |
306 | } | |
4ad49000 LP |
307 | } |
308 | ||
13c31542 | 309 | static int lookup_block_device(const char *p, dev_t *dev) { |
4ad49000 | 310 | struct stat st; |
4ad49000 LP |
311 | |
312 | assert(p); | |
313 | assert(dev); | |
314 | ||
b1c05b98 | 315 | if (stat(p, &st) < 0) |
4a62c710 | 316 | return log_warning_errno(errno, "Couldn't stat device %s: %m", p); |
8e274523 | 317 | |
4ad49000 LP |
318 | if (S_ISBLK(st.st_mode)) |
319 | *dev = st.st_rdev; | |
320 | else if (major(st.st_dev) != 0) { | |
321 | /* If this is not a device node then find the block | |
322 | * device this file is stored on */ | |
323 | *dev = st.st_dev; | |
324 | ||
325 | /* If this is a partition, try to get the originating | |
326 | * block device */ | |
18c528e9 | 327 | (void) block_get_whole_disk(*dev, dev); |
4ad49000 LP |
328 | } else { |
329 | log_warning("%s is not a block device and file system block device cannot be determined or is not local.", p); | |
330 | return -ENODEV; | |
331 | } | |
8e274523 | 332 | |
8e274523 | 333 | return 0; |
8e274523 LP |
334 | } |
335 | ||
4ad49000 LP |
336 | static int whitelist_device(const char *path, const char *node, const char *acc) { |
337 | char buf[2+DECIMAL_STR_MAX(dev_t)*2+2+4]; | |
338 | struct stat st; | |
b200489b | 339 | bool ignore_notfound; |
8c6db833 | 340 | int r; |
8e274523 | 341 | |
4ad49000 LP |
342 | assert(path); |
343 | assert(acc); | |
8e274523 | 344 | |
b200489b DR |
345 | if (node[0] == '-') { |
346 | /* Non-existent paths starting with "-" must be silently ignored */ | |
347 | node++; | |
348 | ignore_notfound = true; | |
349 | } else | |
350 | ignore_notfound = false; | |
351 | ||
4ad49000 | 352 | if (stat(node, &st) < 0) { |
b200489b | 353 | if (errno == ENOENT && ignore_notfound) |
e7330dfe DP |
354 | return 0; |
355 | ||
356 | return log_warning_errno(errno, "Couldn't stat device %s: %m", node); | |
4ad49000 LP |
357 | } |
358 | ||
359 | if (!S_ISCHR(st.st_mode) && !S_ISBLK(st.st_mode)) { | |
360 | log_warning("%s is not a device.", node); | |
361 | return -ENODEV; | |
362 | } | |
363 | ||
364 | sprintf(buf, | |
365 | "%c %u:%u %s", | |
366 | S_ISCHR(st.st_mode) ? 'c' : 'b', | |
367 | major(st.st_rdev), minor(st.st_rdev), | |
368 | acc); | |
369 | ||
370 | r = cg_set_attribute("devices", path, "devices.allow", buf); | |
1aeab12b | 371 | if (r < 0) |
077ba06e | 372 | log_full_errno(IN_SET(r, -ENOENT, -EROFS, -EINVAL, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, |
714e2e1d | 373 | "Failed to set devices.allow on %s: %m", path); |
4ad49000 LP |
374 | |
375 | return r; | |
8e274523 LP |
376 | } |
377 | ||
90060676 LP |
378 | static int whitelist_major(const char *path, const char *name, char type, const char *acc) { |
379 | _cleanup_fclose_ FILE *f = NULL; | |
380 | char line[LINE_MAX]; | |
381 | bool good = false; | |
382 | int r; | |
383 | ||
384 | assert(path); | |
385 | assert(acc); | |
4c701096 | 386 | assert(IN_SET(type, 'b', 'c')); |
90060676 LP |
387 | |
388 | f = fopen("/proc/devices", "re"); | |
4a62c710 MS |
389 | if (!f) |
390 | return log_warning_errno(errno, "Cannot open /proc/devices to resolve %s (%c): %m", name, type); | |
90060676 LP |
391 | |
392 | FOREACH_LINE(line, f, goto fail) { | |
393 | char buf[2+DECIMAL_STR_MAX(unsigned)+3+4], *p, *w; | |
394 | unsigned maj; | |
395 | ||
396 | truncate_nl(line); | |
397 | ||
398 | if (type == 'c' && streq(line, "Character devices:")) { | |
399 | good = true; | |
400 | continue; | |
401 | } | |
402 | ||
403 | if (type == 'b' && streq(line, "Block devices:")) { | |
404 | good = true; | |
405 | continue; | |
406 | } | |
407 | ||
408 | if (isempty(line)) { | |
409 | good = false; | |
410 | continue; | |
411 | } | |
412 | ||
413 | if (!good) | |
414 | continue; | |
415 | ||
416 | p = strstrip(line); | |
417 | ||
418 | w = strpbrk(p, WHITESPACE); | |
419 | if (!w) | |
420 | continue; | |
421 | *w = 0; | |
422 | ||
423 | r = safe_atou(p, &maj); | |
424 | if (r < 0) | |
425 | continue; | |
426 | if (maj <= 0) | |
427 | continue; | |
428 | ||
429 | w++; | |
430 | w += strspn(w, WHITESPACE); | |
e41969e3 LP |
431 | |
432 | if (fnmatch(name, w, 0) != 0) | |
90060676 LP |
433 | continue; |
434 | ||
435 | sprintf(buf, | |
436 | "%c %u:* %s", | |
437 | type, | |
438 | maj, | |
439 | acc); | |
440 | ||
441 | r = cg_set_attribute("devices", path, "devices.allow", buf); | |
1aeab12b | 442 | if (r < 0) |
077ba06e | 443 | log_full_errno(IN_SET(r, -ENOENT, -EROFS, -EINVAL, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, |
714e2e1d | 444 | "Failed to set devices.allow on %s: %m", path); |
90060676 LP |
445 | } |
446 | ||
447 | return 0; | |
448 | ||
449 | fail: | |
25f027c5 | 450 | return log_warning_errno(errno, "Failed to read /proc/devices: %m"); |
90060676 LP |
451 | } |
452 | ||
66ebf6c0 TH |
453 | static bool cgroup_context_has_cpu_weight(CGroupContext *c) { |
454 | return c->cpu_weight != CGROUP_WEIGHT_INVALID || | |
455 | c->startup_cpu_weight != CGROUP_WEIGHT_INVALID; | |
456 | } | |
457 | ||
458 | static bool cgroup_context_has_cpu_shares(CGroupContext *c) { | |
459 | return c->cpu_shares != CGROUP_CPU_SHARES_INVALID || | |
460 | c->startup_cpu_shares != CGROUP_CPU_SHARES_INVALID; | |
461 | } | |
462 | ||
463 | static uint64_t cgroup_context_cpu_weight(CGroupContext *c, ManagerState state) { | |
464 | if (IN_SET(state, MANAGER_STARTING, MANAGER_INITIALIZING) && | |
465 | c->startup_cpu_weight != CGROUP_WEIGHT_INVALID) | |
466 | return c->startup_cpu_weight; | |
467 | else if (c->cpu_weight != CGROUP_WEIGHT_INVALID) | |
468 | return c->cpu_weight; | |
469 | else | |
470 | return CGROUP_WEIGHT_DEFAULT; | |
471 | } | |
472 | ||
473 | static uint64_t cgroup_context_cpu_shares(CGroupContext *c, ManagerState state) { | |
474 | if (IN_SET(state, MANAGER_STARTING, MANAGER_INITIALIZING) && | |
475 | c->startup_cpu_shares != CGROUP_CPU_SHARES_INVALID) | |
476 | return c->startup_cpu_shares; | |
477 | else if (c->cpu_shares != CGROUP_CPU_SHARES_INVALID) | |
478 | return c->cpu_shares; | |
479 | else | |
480 | return CGROUP_CPU_SHARES_DEFAULT; | |
481 | } | |
482 | ||
483 | static void cgroup_apply_unified_cpu_config(Unit *u, uint64_t weight, uint64_t quota) { | |
484 | char buf[MAX(DECIMAL_STR_MAX(uint64_t) + 1, (DECIMAL_STR_MAX(usec_t) + 1) * 2)]; | |
485 | int r; | |
486 | ||
487 | xsprintf(buf, "%" PRIu64 "\n", weight); | |
488 | r = cg_set_attribute("cpu", u->cgroup_path, "cpu.weight", buf); | |
489 | if (r < 0) | |
490 | log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, | |
491 | "Failed to set cpu.weight: %m"); | |
492 | ||
493 | if (quota != USEC_INFINITY) | |
494 | xsprintf(buf, USEC_FMT " " USEC_FMT "\n", | |
495 | quota * CGROUP_CPU_QUOTA_PERIOD_USEC / USEC_PER_SEC, CGROUP_CPU_QUOTA_PERIOD_USEC); | |
496 | else | |
497 | xsprintf(buf, "max " USEC_FMT "\n", CGROUP_CPU_QUOTA_PERIOD_USEC); | |
498 | ||
499 | r = cg_set_attribute("cpu", u->cgroup_path, "cpu.max", buf); | |
500 | ||
501 | if (r < 0) | |
502 | log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, | |
503 | "Failed to set cpu.max: %m"); | |
504 | } | |
505 | ||
506 | static void cgroup_apply_legacy_cpu_config(Unit *u, uint64_t shares, uint64_t quota) { | |
507 | char buf[MAX(DECIMAL_STR_MAX(uint64_t), DECIMAL_STR_MAX(usec_t)) + 1]; | |
508 | int r; | |
509 | ||
510 | xsprintf(buf, "%" PRIu64 "\n", shares); | |
511 | r = cg_set_attribute("cpu", u->cgroup_path, "cpu.shares", buf); | |
512 | if (r < 0) | |
513 | log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, | |
514 | "Failed to set cpu.shares: %m"); | |
515 | ||
516 | xsprintf(buf, USEC_FMT "\n", CGROUP_CPU_QUOTA_PERIOD_USEC); | |
517 | r = cg_set_attribute("cpu", u->cgroup_path, "cpu.cfs_period_us", buf); | |
518 | if (r < 0) | |
519 | log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, | |
520 | "Failed to set cpu.cfs_period_us: %m"); | |
521 | ||
522 | if (quota != USEC_INFINITY) { | |
523 | xsprintf(buf, USEC_FMT "\n", quota * CGROUP_CPU_QUOTA_PERIOD_USEC / USEC_PER_SEC); | |
524 | r = cg_set_attribute("cpu", u->cgroup_path, "cpu.cfs_quota_us", buf); | |
525 | } else | |
526 | r = cg_set_attribute("cpu", u->cgroup_path, "cpu.cfs_quota_us", "-1"); | |
527 | if (r < 0) | |
528 | log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, | |
529 | "Failed to set cpu.cfs_quota_us: %m"); | |
530 | } | |
531 | ||
532 | static uint64_t cgroup_cpu_shares_to_weight(uint64_t shares) { | |
533 | return CLAMP(shares * CGROUP_WEIGHT_DEFAULT / CGROUP_CPU_SHARES_DEFAULT, | |
534 | CGROUP_WEIGHT_MIN, CGROUP_WEIGHT_MAX); | |
535 | } | |
536 | ||
537 | static uint64_t cgroup_cpu_weight_to_shares(uint64_t weight) { | |
538 | return CLAMP(weight * CGROUP_CPU_SHARES_DEFAULT / CGROUP_WEIGHT_DEFAULT, | |
539 | CGROUP_CPU_SHARES_MIN, CGROUP_CPU_SHARES_MAX); | |
540 | } | |
541 | ||
508c45da | 542 | static bool cgroup_context_has_io_config(CGroupContext *c) { |
538b4852 TH |
543 | return c->io_accounting || |
544 | c->io_weight != CGROUP_WEIGHT_INVALID || | |
545 | c->startup_io_weight != CGROUP_WEIGHT_INVALID || | |
546 | c->io_device_weights || | |
547 | c->io_device_limits; | |
548 | } | |
549 | ||
508c45da | 550 | static bool cgroup_context_has_blockio_config(CGroupContext *c) { |
538b4852 TH |
551 | return c->blockio_accounting || |
552 | c->blockio_weight != CGROUP_BLKIO_WEIGHT_INVALID || | |
553 | c->startup_blockio_weight != CGROUP_BLKIO_WEIGHT_INVALID || | |
554 | c->blockio_device_weights || | |
555 | c->blockio_device_bandwidths; | |
556 | } | |
557 | ||
508c45da | 558 | static uint64_t cgroup_context_io_weight(CGroupContext *c, ManagerState state) { |
64faf04c TH |
559 | if (IN_SET(state, MANAGER_STARTING, MANAGER_INITIALIZING) && |
560 | c->startup_io_weight != CGROUP_WEIGHT_INVALID) | |
561 | return c->startup_io_weight; | |
562 | else if (c->io_weight != CGROUP_WEIGHT_INVALID) | |
563 | return c->io_weight; | |
564 | else | |
565 | return CGROUP_WEIGHT_DEFAULT; | |
566 | } | |
567 | ||
508c45da | 568 | static uint64_t cgroup_context_blkio_weight(CGroupContext *c, ManagerState state) { |
64faf04c TH |
569 | if (IN_SET(state, MANAGER_STARTING, MANAGER_INITIALIZING) && |
570 | c->startup_blockio_weight != CGROUP_BLKIO_WEIGHT_INVALID) | |
571 | return c->startup_blockio_weight; | |
572 | else if (c->blockio_weight != CGROUP_BLKIO_WEIGHT_INVALID) | |
573 | return c->blockio_weight; | |
574 | else | |
575 | return CGROUP_BLKIO_WEIGHT_DEFAULT; | |
576 | } | |
577 | ||
508c45da | 578 | static uint64_t cgroup_weight_blkio_to_io(uint64_t blkio_weight) { |
538b4852 TH |
579 | return CLAMP(blkio_weight * CGROUP_WEIGHT_DEFAULT / CGROUP_BLKIO_WEIGHT_DEFAULT, |
580 | CGROUP_WEIGHT_MIN, CGROUP_WEIGHT_MAX); | |
581 | } | |
582 | ||
508c45da | 583 | static uint64_t cgroup_weight_io_to_blkio(uint64_t io_weight) { |
538b4852 TH |
584 | return CLAMP(io_weight * CGROUP_BLKIO_WEIGHT_DEFAULT / CGROUP_WEIGHT_DEFAULT, |
585 | CGROUP_BLKIO_WEIGHT_MIN, CGROUP_BLKIO_WEIGHT_MAX); | |
586 | } | |
587 | ||
f29ff115 | 588 | static void cgroup_apply_io_device_weight(Unit *u, const char *dev_path, uint64_t io_weight) { |
64faf04c TH |
589 | char buf[DECIMAL_STR_MAX(dev_t)*2+2+DECIMAL_STR_MAX(uint64_t)+1]; |
590 | dev_t dev; | |
591 | int r; | |
592 | ||
593 | r = lookup_block_device(dev_path, &dev); | |
594 | if (r < 0) | |
595 | return; | |
596 | ||
597 | xsprintf(buf, "%u:%u %" PRIu64 "\n", major(dev), minor(dev), io_weight); | |
f29ff115 | 598 | r = cg_set_attribute("io", u->cgroup_path, "io.weight", buf); |
64faf04c | 599 | if (r < 0) |
f29ff115 TH |
600 | log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, |
601 | "Failed to set io.weight: %m"); | |
64faf04c TH |
602 | } |
603 | ||
f29ff115 | 604 | static void cgroup_apply_blkio_device_weight(Unit *u, const char *dev_path, uint64_t blkio_weight) { |
64faf04c TH |
605 | char buf[DECIMAL_STR_MAX(dev_t)*2+2+DECIMAL_STR_MAX(uint64_t)+1]; |
606 | dev_t dev; | |
607 | int r; | |
608 | ||
609 | r = lookup_block_device(dev_path, &dev); | |
610 | if (r < 0) | |
611 | return; | |
612 | ||
613 | xsprintf(buf, "%u:%u %" PRIu64 "\n", major(dev), minor(dev), blkio_weight); | |
f29ff115 | 614 | r = cg_set_attribute("blkio", u->cgroup_path, "blkio.weight_device", buf); |
64faf04c | 615 | if (r < 0) |
f29ff115 TH |
616 | log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, |
617 | "Failed to set blkio.weight_device: %m"); | |
64faf04c TH |
618 | } |
619 | ||
f29ff115 | 620 | static unsigned cgroup_apply_io_device_limit(Unit *u, const char *dev_path, uint64_t *limits) { |
64faf04c TH |
621 | char limit_bufs[_CGROUP_IO_LIMIT_TYPE_MAX][DECIMAL_STR_MAX(uint64_t)]; |
622 | char buf[DECIMAL_STR_MAX(dev_t)*2+2+(6+DECIMAL_STR_MAX(uint64_t)+1)*4]; | |
623 | CGroupIOLimitType type; | |
624 | dev_t dev; | |
625 | unsigned n = 0; | |
626 | int r; | |
627 | ||
628 | r = lookup_block_device(dev_path, &dev); | |
629 | if (r < 0) | |
630 | return 0; | |
631 | ||
632 | for (type = 0; type < _CGROUP_IO_LIMIT_TYPE_MAX; type++) { | |
633 | if (limits[type] != cgroup_io_limit_defaults[type]) { | |
634 | xsprintf(limit_bufs[type], "%" PRIu64, limits[type]); | |
635 | n++; | |
636 | } else { | |
637 | xsprintf(limit_bufs[type], "%s", limits[type] == CGROUP_LIMIT_MAX ? "max" : "0"); | |
638 | } | |
639 | } | |
640 | ||
641 | xsprintf(buf, "%u:%u rbps=%s wbps=%s riops=%s wiops=%s\n", major(dev), minor(dev), | |
642 | limit_bufs[CGROUP_IO_RBPS_MAX], limit_bufs[CGROUP_IO_WBPS_MAX], | |
643 | limit_bufs[CGROUP_IO_RIOPS_MAX], limit_bufs[CGROUP_IO_WIOPS_MAX]); | |
f29ff115 | 644 | r = cg_set_attribute("io", u->cgroup_path, "io.max", buf); |
64faf04c | 645 | if (r < 0) |
f29ff115 TH |
646 | log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, |
647 | "Failed to set io.max: %m"); | |
64faf04c TH |
648 | return n; |
649 | } | |
650 | ||
f29ff115 | 651 | static unsigned cgroup_apply_blkio_device_limit(Unit *u, const char *dev_path, uint64_t rbps, uint64_t wbps) { |
64faf04c TH |
652 | char buf[DECIMAL_STR_MAX(dev_t)*2+2+DECIMAL_STR_MAX(uint64_t)+1]; |
653 | dev_t dev; | |
654 | unsigned n = 0; | |
655 | int r; | |
656 | ||
657 | r = lookup_block_device(dev_path, &dev); | |
658 | if (r < 0) | |
659 | return 0; | |
660 | ||
661 | if (rbps != CGROUP_LIMIT_MAX) | |
662 | n++; | |
663 | sprintf(buf, "%u:%u %" PRIu64 "\n", major(dev), minor(dev), rbps); | |
f29ff115 | 664 | r = cg_set_attribute("blkio", u->cgroup_path, "blkio.throttle.read_bps_device", buf); |
64faf04c | 665 | if (r < 0) |
f29ff115 TH |
666 | log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, |
667 | "Failed to set blkio.throttle.read_bps_device: %m"); | |
64faf04c TH |
668 | |
669 | if (wbps != CGROUP_LIMIT_MAX) | |
670 | n++; | |
671 | sprintf(buf, "%u:%u %" PRIu64 "\n", major(dev), minor(dev), wbps); | |
f29ff115 | 672 | r = cg_set_attribute("blkio", u->cgroup_path, "blkio.throttle.write_bps_device", buf); |
64faf04c | 673 | if (r < 0) |
f29ff115 TH |
674 | log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, |
675 | "Failed to set blkio.throttle.write_bps_device: %m"); | |
64faf04c TH |
676 | |
677 | return n; | |
678 | } | |
679 | ||
da4d897e | 680 | static bool cgroup_context_has_unified_memory_config(CGroupContext *c) { |
96e131ea | 681 | return 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 |
682 | } |
683 | ||
f29ff115 | 684 | static void cgroup_apply_unified_memory_limit(Unit *u, const char *file, uint64_t v) { |
da4d897e TH |
685 | char buf[DECIMAL_STR_MAX(uint64_t) + 1] = "max"; |
686 | int r; | |
687 | ||
688 | if (v != CGROUP_LIMIT_MAX) | |
689 | xsprintf(buf, "%" PRIu64 "\n", v); | |
690 | ||
f29ff115 | 691 | r = cg_set_attribute("memory", u->cgroup_path, file, buf); |
da4d897e | 692 | if (r < 0) |
f29ff115 TH |
693 | log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, |
694 | "Failed to set %s: %m", file); | |
da4d897e TH |
695 | } |
696 | ||
0f2d84d2 | 697 | static void cgroup_apply_firewall(Unit *u) { |
0f2d84d2 LP |
698 | assert(u); |
699 | ||
acf7f253 | 700 | /* Best-effort: let's apply IP firewalling and/or accounting if that's enabled */ |
906c06f6 | 701 | |
acf7f253 | 702 | if (bpf_firewall_compile(u) < 0) |
906c06f6 DM |
703 | return; |
704 | ||
705 | (void) bpf_firewall_install(u); | |
906c06f6 DM |
706 | } |
707 | ||
708 | static void cgroup_context_apply( | |
709 | Unit *u, | |
710 | CGroupMask apply_mask, | |
711 | bool apply_bpf, | |
712 | ManagerState state) { | |
713 | ||
f29ff115 TH |
714 | const char *path; |
715 | CGroupContext *c; | |
01efdf13 | 716 | bool is_root; |
4ad49000 LP |
717 | int r; |
718 | ||
f29ff115 TH |
719 | assert(u); |
720 | ||
906c06f6 DM |
721 | /* Nothing to do? Exit early! */ |
722 | if (apply_mask == 0 && !apply_bpf) | |
4ad49000 | 723 | return; |
8e274523 | 724 | |
f3725e64 LP |
725 | /* Some cgroup attributes are not supported on the root cgroup, hence silently ignore */ |
726 | is_root = unit_has_root_cgroup(u); | |
727 | ||
728 | assert_se(c = unit_get_cgroup_context(u)); | |
729 | assert_se(path = u->cgroup_path); | |
730 | ||
731 | if (is_root) /* Make sure we don't try to display messages with an empty path. */ | |
6da13913 | 732 | path = "/"; |
01efdf13 | 733 | |
714e2e1d LP |
734 | /* We generally ignore errors caused by read-only mounted |
735 | * cgroup trees (assuming we are running in a container then), | |
736 | * and missing cgroups, i.e. EROFS and ENOENT. */ | |
737 | ||
906c06f6 DM |
738 | if ((apply_mask & CGROUP_MASK_CPU) && !is_root) { |
739 | bool has_weight, has_shares; | |
740 | ||
741 | has_weight = cgroup_context_has_cpu_weight(c); | |
742 | has_shares = cgroup_context_has_cpu_shares(c); | |
8e274523 | 743 | |
b4cccbc1 | 744 | if (cg_all_unified() > 0) { |
66ebf6c0 | 745 | uint64_t weight; |
b2f8b02e | 746 | |
66ebf6c0 TH |
747 | if (has_weight) |
748 | weight = cgroup_context_cpu_weight(c, state); | |
749 | else if (has_shares) { | |
750 | uint64_t shares = cgroup_context_cpu_shares(c, state); | |
b2f8b02e | 751 | |
66ebf6c0 TH |
752 | weight = cgroup_cpu_shares_to_weight(shares); |
753 | ||
754 | log_cgroup_compat(u, "Applying [Startup]CpuShares %" PRIu64 " as [Startup]CpuWeight %" PRIu64 " on %s", | |
755 | shares, weight, path); | |
756 | } else | |
757 | weight = CGROUP_WEIGHT_DEFAULT; | |
758 | ||
759 | cgroup_apply_unified_cpu_config(u, weight, c->cpu_quota_per_sec_usec); | |
760 | } else { | |
761 | uint64_t shares; | |
762 | ||
7d862ab8 | 763 | if (has_weight) { |
66ebf6c0 TH |
764 | uint64_t weight = cgroup_context_cpu_weight(c, state); |
765 | ||
766 | shares = cgroup_cpu_weight_to_shares(weight); | |
767 | ||
768 | log_cgroup_compat(u, "Applying [Startup]CpuWeight %" PRIu64 " as [Startup]CpuShares %" PRIu64 " on %s", | |
769 | weight, shares, path); | |
7d862ab8 TH |
770 | } else if (has_shares) |
771 | shares = cgroup_context_cpu_shares(c, state); | |
772 | else | |
66ebf6c0 TH |
773 | shares = CGROUP_CPU_SHARES_DEFAULT; |
774 | ||
775 | cgroup_apply_legacy_cpu_config(u, shares, c->cpu_quota_per_sec_usec); | |
776 | } | |
4ad49000 LP |
777 | } |
778 | ||
906c06f6 | 779 | if (apply_mask & CGROUP_MASK_IO) { |
538b4852 TH |
780 | bool has_io = cgroup_context_has_io_config(c); |
781 | bool has_blockio = cgroup_context_has_blockio_config(c); | |
13c31542 TH |
782 | |
783 | if (!is_root) { | |
64faf04c TH |
784 | char buf[8+DECIMAL_STR_MAX(uint64_t)+1]; |
785 | uint64_t weight; | |
13c31542 | 786 | |
538b4852 TH |
787 | if (has_io) |
788 | weight = cgroup_context_io_weight(c, state); | |
128fadc9 TH |
789 | else if (has_blockio) { |
790 | uint64_t blkio_weight = cgroup_context_blkio_weight(c, state); | |
791 | ||
792 | weight = cgroup_weight_blkio_to_io(blkio_weight); | |
793 | ||
794 | log_cgroup_compat(u, "Applying [Startup]BlockIOWeight %" PRIu64 " as [Startup]IOWeight %" PRIu64, | |
795 | blkio_weight, weight); | |
796 | } else | |
538b4852 | 797 | weight = CGROUP_WEIGHT_DEFAULT; |
13c31542 TH |
798 | |
799 | xsprintf(buf, "default %" PRIu64 "\n", weight); | |
800 | r = cg_set_attribute("io", path, "io.weight", buf); | |
801 | if (r < 0) | |
f29ff115 TH |
802 | log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, |
803 | "Failed to set io.weight: %m"); | |
13c31542 | 804 | |
538b4852 TH |
805 | if (has_io) { |
806 | CGroupIODeviceWeight *w; | |
807 | ||
808 | /* FIXME: no way to reset this list */ | |
809 | LIST_FOREACH(device_weights, w, c->io_device_weights) | |
f29ff115 | 810 | cgroup_apply_io_device_weight(u, w->path, w->weight); |
538b4852 TH |
811 | } else if (has_blockio) { |
812 | CGroupBlockIODeviceWeight *w; | |
813 | ||
814 | /* FIXME: no way to reset this list */ | |
128fadc9 TH |
815 | LIST_FOREACH(device_weights, w, c->blockio_device_weights) { |
816 | weight = cgroup_weight_blkio_to_io(w->weight); | |
817 | ||
818 | log_cgroup_compat(u, "Applying BlockIODeviceWeight %" PRIu64 " as IODeviceWeight %" PRIu64 " for %s", | |
819 | w->weight, weight, w->path); | |
820 | ||
821 | cgroup_apply_io_device_weight(u, w->path, weight); | |
822 | } | |
538b4852 | 823 | } |
13c31542 TH |
824 | } |
825 | ||
64faf04c | 826 | /* Apply limits and free ones without config. */ |
538b4852 TH |
827 | if (has_io) { |
828 | CGroupIODeviceLimit *l, *next; | |
829 | ||
830 | LIST_FOREACH_SAFE(device_limits, l, next, c->io_device_limits) { | |
f29ff115 | 831 | if (!cgroup_apply_io_device_limit(u, l->path, l->limits)) |
538b4852 TH |
832 | cgroup_context_free_io_device_limit(c, l); |
833 | } | |
834 | } else if (has_blockio) { | |
835 | CGroupBlockIODeviceBandwidth *b, *next; | |
836 | ||
837 | LIST_FOREACH_SAFE(device_bandwidths, b, next, c->blockio_device_bandwidths) { | |
838 | uint64_t limits[_CGROUP_IO_LIMIT_TYPE_MAX]; | |
839 | CGroupIOLimitType type; | |
840 | ||
841 | for (type = 0; type < _CGROUP_IO_LIMIT_TYPE_MAX; type++) | |
842 | limits[type] = cgroup_io_limit_defaults[type]; | |
843 | ||
844 | limits[CGROUP_IO_RBPS_MAX] = b->rbps; | |
845 | limits[CGROUP_IO_WBPS_MAX] = b->wbps; | |
846 | ||
128fadc9 TH |
847 | log_cgroup_compat(u, "Applying BlockIO{Read|Write}Bandwidth %" PRIu64 " %" PRIu64 " as IO{Read|Write}BandwidthMax for %s", |
848 | b->rbps, b->wbps, b->path); | |
849 | ||
f29ff115 | 850 | if (!cgroup_apply_io_device_limit(u, b->path, limits)) |
538b4852 TH |
851 | cgroup_context_free_blockio_device_bandwidth(c, b); |
852 | } | |
13c31542 TH |
853 | } |
854 | } | |
855 | ||
906c06f6 | 856 | if (apply_mask & CGROUP_MASK_BLKIO) { |
538b4852 TH |
857 | bool has_io = cgroup_context_has_io_config(c); |
858 | bool has_blockio = cgroup_context_has_blockio_config(c); | |
4ad49000 | 859 | |
01efdf13 | 860 | if (!is_root) { |
64faf04c TH |
861 | char buf[DECIMAL_STR_MAX(uint64_t)+1]; |
862 | uint64_t weight; | |
64faf04c | 863 | |
7d862ab8 | 864 | if (has_io) { |
128fadc9 TH |
865 | uint64_t io_weight = cgroup_context_io_weight(c, state); |
866 | ||
538b4852 | 867 | weight = cgroup_weight_io_to_blkio(cgroup_context_io_weight(c, state)); |
128fadc9 TH |
868 | |
869 | log_cgroup_compat(u, "Applying [Startup]IOWeight %" PRIu64 " as [Startup]BlockIOWeight %" PRIu64, | |
870 | io_weight, weight); | |
7d862ab8 TH |
871 | } else if (has_blockio) |
872 | weight = cgroup_context_blkio_weight(c, state); | |
873 | else | |
538b4852 | 874 | weight = CGROUP_BLKIO_WEIGHT_DEFAULT; |
64faf04c TH |
875 | |
876 | xsprintf(buf, "%" PRIu64 "\n", weight); | |
01efdf13 | 877 | r = cg_set_attribute("blkio", path, "blkio.weight", buf); |
1aeab12b | 878 | if (r < 0) |
f29ff115 TH |
879 | log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, |
880 | "Failed to set blkio.weight: %m"); | |
4ad49000 | 881 | |
7d862ab8 | 882 | if (has_io) { |
538b4852 TH |
883 | CGroupIODeviceWeight *w; |
884 | ||
885 | /* FIXME: no way to reset this list */ | |
128fadc9 TH |
886 | LIST_FOREACH(device_weights, w, c->io_device_weights) { |
887 | weight = cgroup_weight_io_to_blkio(w->weight); | |
888 | ||
889 | log_cgroup_compat(u, "Applying IODeviceWeight %" PRIu64 " as BlockIODeviceWeight %" PRIu64 " for %s", | |
890 | w->weight, weight, w->path); | |
891 | ||
892 | cgroup_apply_blkio_device_weight(u, w->path, weight); | |
893 | } | |
7d862ab8 TH |
894 | } else if (has_blockio) { |
895 | CGroupBlockIODeviceWeight *w; | |
896 | ||
897 | /* FIXME: no way to reset this list */ | |
898 | LIST_FOREACH(device_weights, w, c->blockio_device_weights) | |
899 | cgroup_apply_blkio_device_weight(u, w->path, w->weight); | |
538b4852 | 900 | } |
4ad49000 LP |
901 | } |
902 | ||
64faf04c | 903 | /* Apply limits and free ones without config. */ |
7d862ab8 | 904 | if (has_io) { |
538b4852 TH |
905 | CGroupIODeviceLimit *l, *next; |
906 | ||
907 | LIST_FOREACH_SAFE(device_limits, l, next, c->io_device_limits) { | |
128fadc9 TH |
908 | log_cgroup_compat(u, "Applying IO{Read|Write}Bandwidth %" PRIu64 " %" PRIu64 " as BlockIO{Read|Write}BandwidthMax for %s", |
909 | l->limits[CGROUP_IO_RBPS_MAX], l->limits[CGROUP_IO_WBPS_MAX], l->path); | |
910 | ||
f29ff115 | 911 | if (!cgroup_apply_blkio_device_limit(u, l->path, l->limits[CGROUP_IO_RBPS_MAX], l->limits[CGROUP_IO_WBPS_MAX])) |
538b4852 TH |
912 | cgroup_context_free_io_device_limit(c, l); |
913 | } | |
7d862ab8 TH |
914 | } else if (has_blockio) { |
915 | CGroupBlockIODeviceBandwidth *b, *next; | |
916 | ||
917 | LIST_FOREACH_SAFE(device_bandwidths, b, next, c->blockio_device_bandwidths) | |
918 | if (!cgroup_apply_blkio_device_limit(u, b->path, b->rbps, b->wbps)) | |
919 | cgroup_context_free_blockio_device_bandwidth(c, b); | |
d686d8a9 | 920 | } |
8e274523 LP |
921 | } |
922 | ||
906c06f6 | 923 | if ((apply_mask & CGROUP_MASK_MEMORY) && !is_root) { |
b4cccbc1 LP |
924 | if (cg_all_unified() > 0) { |
925 | uint64_t max, swap_max = CGROUP_LIMIT_MAX; | |
efdb0237 | 926 | |
96e131ea | 927 | if (cgroup_context_has_unified_memory_config(c)) { |
da4d897e | 928 | max = c->memory_max; |
96e131ea WC |
929 | swap_max = c->memory_swap_max; |
930 | } else { | |
da4d897e | 931 | max = c->memory_limit; |
efdb0237 | 932 | |
128fadc9 TH |
933 | if (max != CGROUP_LIMIT_MAX) |
934 | log_cgroup_compat(u, "Applying MemoryLimit %" PRIu64 " as MemoryMax", max); | |
935 | } | |
936 | ||
f29ff115 TH |
937 | cgroup_apply_unified_memory_limit(u, "memory.low", c->memory_low); |
938 | cgroup_apply_unified_memory_limit(u, "memory.high", c->memory_high); | |
939 | cgroup_apply_unified_memory_limit(u, "memory.max", max); | |
96e131ea | 940 | cgroup_apply_unified_memory_limit(u, "memory.swap.max", swap_max); |
efdb0237 | 941 | } else { |
da4d897e | 942 | char buf[DECIMAL_STR_MAX(uint64_t) + 1]; |
7d862ab8 | 943 | uint64_t val; |
da4d897e | 944 | |
7d862ab8 | 945 | if (cgroup_context_has_unified_memory_config(c)) { |
78a4ee59 | 946 | val = c->memory_max; |
7d862ab8 TH |
947 | log_cgroup_compat(u, "Applying MemoryMax %" PRIi64 " as MemoryLimit", val); |
948 | } else | |
949 | val = c->memory_limit; | |
128fadc9 | 950 | |
78a4ee59 DM |
951 | if (val == CGROUP_LIMIT_MAX) |
952 | strncpy(buf, "-1\n", sizeof(buf)); | |
953 | else | |
954 | xsprintf(buf, "%" PRIu64 "\n", val); | |
955 | ||
da4d897e TH |
956 | r = cg_set_attribute("memory", path, "memory.limit_in_bytes", buf); |
957 | if (r < 0) | |
f29ff115 TH |
958 | log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, |
959 | "Failed to set memory.limit_in_bytes: %m"); | |
da4d897e | 960 | } |
4ad49000 | 961 | } |
8e274523 | 962 | |
906c06f6 | 963 | if ((apply_mask & CGROUP_MASK_DEVICES) && !is_root) { |
4ad49000 | 964 | CGroupDeviceAllow *a; |
8e274523 | 965 | |
714e2e1d LP |
966 | /* Changing the devices list of a populated cgroup |
967 | * might result in EINVAL, hence ignore EINVAL | |
968 | * here. */ | |
969 | ||
4ad49000 LP |
970 | if (c->device_allow || c->device_policy != CGROUP_AUTO) |
971 | r = cg_set_attribute("devices", path, "devices.deny", "a"); | |
972 | else | |
973 | r = cg_set_attribute("devices", path, "devices.allow", "a"); | |
1aeab12b | 974 | if (r < 0) |
f29ff115 TH |
975 | log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EINVAL, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, |
976 | "Failed to reset devices.list: %m"); | |
fb385181 | 977 | |
4ad49000 LP |
978 | if (c->device_policy == CGROUP_CLOSED || |
979 | (c->device_policy == CGROUP_AUTO && c->device_allow)) { | |
980 | static const char auto_devices[] = | |
7d711efb LP |
981 | "/dev/null\0" "rwm\0" |
982 | "/dev/zero\0" "rwm\0" | |
983 | "/dev/full\0" "rwm\0" | |
984 | "/dev/random\0" "rwm\0" | |
985 | "/dev/urandom\0" "rwm\0" | |
986 | "/dev/tty\0" "rwm\0" | |
5a7f87a9 | 987 | "/dev/ptmx\0" "rwm\0" |
0d9e7991 | 988 | /* Allow /run/systemd/inaccessible/{chr,blk} devices for mapping InaccessiblePaths */ |
e7330dfe DP |
989 | "-/run/systemd/inaccessible/chr\0" "rwm\0" |
990 | "-/run/systemd/inaccessible/blk\0" "rwm\0"; | |
4ad49000 LP |
991 | |
992 | const char *x, *y; | |
993 | ||
994 | NULSTR_FOREACH_PAIR(x, y, auto_devices) | |
995 | whitelist_device(path, x, y); | |
7d711efb | 996 | |
5a7f87a9 | 997 | /* PTS (/dev/pts) devices may not be duplicated, but accessed */ |
7d711efb | 998 | whitelist_major(path, "pts", 'c', "rw"); |
4ad49000 LP |
999 | } |
1000 | ||
1001 | LIST_FOREACH(device_allow, a, c->device_allow) { | |
fb4650aa | 1002 | char acc[4], *val; |
4ad49000 LP |
1003 | unsigned k = 0; |
1004 | ||
1005 | if (a->r) | |
1006 | acc[k++] = 'r'; | |
1007 | if (a->w) | |
1008 | acc[k++] = 'w'; | |
1009 | if (a->m) | |
1010 | acc[k++] = 'm'; | |
fb385181 | 1011 | |
4ad49000 LP |
1012 | if (k == 0) |
1013 | continue; | |
fb385181 | 1014 | |
4ad49000 | 1015 | acc[k++] = 0; |
90060676 | 1016 | |
27458ed6 | 1017 | if (path_startswith(a->path, "/dev/")) |
90060676 | 1018 | whitelist_device(path, a->path, acc); |
fb4650aa ZJS |
1019 | else if ((val = startswith(a->path, "block-"))) |
1020 | whitelist_major(path, val, 'b', acc); | |
1021 | else if ((val = startswith(a->path, "char-"))) | |
1022 | whitelist_major(path, val, 'c', acc); | |
90060676 | 1023 | else |
f29ff115 | 1024 | log_unit_debug(u, "Ignoring device %s while writing cgroup attribute.", a->path); |
4ad49000 LP |
1025 | } |
1026 | } | |
03a7b521 | 1027 | |
00b5974f LP |
1028 | if (apply_mask & CGROUP_MASK_PIDS) { |
1029 | ||
1030 | if (is_root) { | |
1031 | /* So, the "pids" controller does not expose anything on the root cgroup, in order not to | |
1032 | * replicate knobs exposed elsewhere needlessly. We abstract this away here however, and when | |
1033 | * the knobs of the root cgroup are modified propagate this to the relevant sysctls. There's a | |
1034 | * non-obvious asymmetry however: unlike the cgroup properties we don't really want to take | |
1035 | * exclusive ownership of the sysctls, but we still want to honour things if the user sets | |
1036 | * limits. Hence we employ sort of a one-way strategy: when the user sets a bounded limit | |
1037 | * through us it counts. When the user afterwards unsets it again (i.e. sets it to unbounded) | |
1038 | * it also counts. But if the user never set a limit through us (i.e. we are the default of | |
1039 | * "unbounded") we leave things unmodified. For this we manage a global boolean that we turn on | |
1040 | * the first time we set a limit. Note that this boolean is flushed out on manager reload, | |
1041 | * which is desirable so that there's an offical way to release control of the sysctl from | |
1042 | * systemd: set the limit to unbounded and reload. */ | |
1043 | ||
1044 | if (c->tasks_max != CGROUP_LIMIT_MAX) { | |
1045 | u->manager->sysctl_pid_max_changed = true; | |
1046 | r = procfs_tasks_set_limit(c->tasks_max); | |
1047 | } else if (u->manager->sysctl_pid_max_changed) | |
1048 | r = procfs_tasks_set_limit(TASKS_MAX); | |
1049 | else | |
1050 | r = 0; | |
03a7b521 | 1051 | |
00b5974f LP |
1052 | if (r < 0) |
1053 | log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, | |
1054 | "Failed to write to tasks limit sysctls: %m"); | |
03a7b521 | 1055 | |
00b5974f LP |
1056 | } else { |
1057 | if (c->tasks_max != CGROUP_LIMIT_MAX) { | |
1058 | char buf[DECIMAL_STR_MAX(uint64_t) + 2]; | |
03a7b521 | 1059 | |
00b5974f LP |
1060 | sprintf(buf, "%" PRIu64 "\n", c->tasks_max); |
1061 | r = cg_set_attribute("pids", path, "pids.max", buf); | |
1062 | } else | |
1063 | r = cg_set_attribute("pids", path, "pids.max", "max"); | |
1064 | if (r < 0) | |
1065 | log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, | |
1066 | "Failed to set pids.max: %m"); | |
1067 | } | |
03a7b521 | 1068 | } |
906c06f6 DM |
1069 | |
1070 | if (apply_bpf) | |
0f2d84d2 | 1071 | cgroup_apply_firewall(u); |
fb385181 LP |
1072 | } |
1073 | ||
efdb0237 LP |
1074 | CGroupMask cgroup_context_get_mask(CGroupContext *c) { |
1075 | CGroupMask mask = 0; | |
8e274523 | 1076 | |
4ad49000 | 1077 | /* Figure out which controllers we need */ |
8e274523 | 1078 | |
b2f8b02e | 1079 | if (c->cpu_accounting || |
66ebf6c0 TH |
1080 | cgroup_context_has_cpu_weight(c) || |
1081 | cgroup_context_has_cpu_shares(c) || | |
3a43da28 | 1082 | c->cpu_quota_per_sec_usec != USEC_INFINITY) |
efdb0237 | 1083 | mask |= CGROUP_MASK_CPUACCT | CGROUP_MASK_CPU; |
ecedd90f | 1084 | |
538b4852 TH |
1085 | if (cgroup_context_has_io_config(c) || cgroup_context_has_blockio_config(c)) |
1086 | mask |= CGROUP_MASK_IO | CGROUP_MASK_BLKIO; | |
ecedd90f | 1087 | |
4ad49000 | 1088 | if (c->memory_accounting || |
da4d897e TH |
1089 | c->memory_limit != CGROUP_LIMIT_MAX || |
1090 | cgroup_context_has_unified_memory_config(c)) | |
efdb0237 | 1091 | mask |= CGROUP_MASK_MEMORY; |
8e274523 | 1092 | |
a931ad47 LP |
1093 | if (c->device_allow || |
1094 | c->device_policy != CGROUP_AUTO) | |
3905f127 | 1095 | mask |= CGROUP_MASK_DEVICES; |
4ad49000 | 1096 | |
03a7b521 | 1097 | if (c->tasks_accounting || |
8793fa25 | 1098 | c->tasks_max != CGROUP_LIMIT_MAX) |
03a7b521 LP |
1099 | mask |= CGROUP_MASK_PIDS; |
1100 | ||
4ad49000 | 1101 | return mask; |
8e274523 LP |
1102 | } |
1103 | ||
efdb0237 | 1104 | CGroupMask unit_get_own_mask(Unit *u) { |
4ad49000 | 1105 | CGroupContext *c; |
8e274523 | 1106 | |
efdb0237 LP |
1107 | /* Returns the mask of controllers the unit needs for itself */ |
1108 | ||
4ad49000 LP |
1109 | c = unit_get_cgroup_context(u); |
1110 | if (!c) | |
1111 | return 0; | |
8e274523 | 1112 | |
64e844e5 | 1113 | return cgroup_context_get_mask(c) | unit_get_delegate_mask(u); |
02638280 LP |
1114 | } |
1115 | ||
1116 | CGroupMask unit_get_delegate_mask(Unit *u) { | |
1117 | CGroupContext *c; | |
1118 | ||
1119 | /* If delegation is turned on, then turn on selected controllers, unless we are on the legacy hierarchy and the | |
1120 | * process we fork into is known to drop privileges, and hence shouldn't get access to the controllers. | |
19af675e | 1121 | * |
02638280 | 1122 | * Note that on the unified hierarchy it is safe to delegate controllers to unprivileged services. */ |
a931ad47 | 1123 | |
1d9cc876 | 1124 | if (!unit_cgroup_delegate(u)) |
02638280 LP |
1125 | return 0; |
1126 | ||
1127 | if (cg_all_unified() <= 0) { | |
a931ad47 LP |
1128 | ExecContext *e; |
1129 | ||
1130 | e = unit_get_exec_context(u); | |
02638280 LP |
1131 | if (e && !exec_context_maintains_privileges(e)) |
1132 | return 0; | |
a931ad47 LP |
1133 | } |
1134 | ||
1d9cc876 | 1135 | assert_se(c = unit_get_cgroup_context(u)); |
02638280 | 1136 | return c->delegate_controllers; |
8e274523 LP |
1137 | } |
1138 | ||
efdb0237 | 1139 | CGroupMask unit_get_members_mask(Unit *u) { |
4ad49000 | 1140 | assert(u); |
bc432dc7 | 1141 | |
02638280 | 1142 | /* Returns the mask of controllers all of the unit's children require, merged */ |
efdb0237 | 1143 | |
bc432dc7 LP |
1144 | if (u->cgroup_members_mask_valid) |
1145 | return u->cgroup_members_mask; | |
1146 | ||
64e844e5 | 1147 | u->cgroup_members_mask = 0; |
bc432dc7 LP |
1148 | |
1149 | if (u->type == UNIT_SLICE) { | |
eef85c4a | 1150 | void *v; |
bc432dc7 LP |
1151 | Unit *member; |
1152 | Iterator i; | |
1153 | ||
eef85c4a | 1154 | HASHMAP_FOREACH_KEY(v, member, u->dependencies[UNIT_BEFORE], i) { |
bc432dc7 LP |
1155 | |
1156 | if (member == u) | |
1157 | continue; | |
1158 | ||
d4fdc205 | 1159 | if (UNIT_DEREF(member->slice) != u) |
bc432dc7 LP |
1160 | continue; |
1161 | ||
31604970 | 1162 | u->cgroup_members_mask |= unit_get_subtree_mask(member); /* note that this calls ourselves again, for the children */ |
bc432dc7 LP |
1163 | } |
1164 | } | |
1165 | ||
1166 | u->cgroup_members_mask_valid = true; | |
6414b7c9 | 1167 | return u->cgroup_members_mask; |
246aa6dd LP |
1168 | } |
1169 | ||
efdb0237 | 1170 | CGroupMask unit_get_siblings_mask(Unit *u) { |
4ad49000 | 1171 | assert(u); |
246aa6dd | 1172 | |
efdb0237 LP |
1173 | /* Returns the mask of controllers all of the unit's siblings |
1174 | * require, i.e. the members mask of the unit's parent slice | |
1175 | * if there is one. */ | |
1176 | ||
bc432dc7 | 1177 | if (UNIT_ISSET(u->slice)) |
637f421e | 1178 | return unit_get_members_mask(UNIT_DEREF(u->slice)); |
4ad49000 | 1179 | |
64e844e5 | 1180 | return unit_get_subtree_mask(u); /* we are the top-level slice */ |
246aa6dd LP |
1181 | } |
1182 | ||
efdb0237 LP |
1183 | CGroupMask unit_get_subtree_mask(Unit *u) { |
1184 | ||
1185 | /* Returns the mask of this subtree, meaning of the group | |
1186 | * itself and its children. */ | |
1187 | ||
1188 | return unit_get_own_mask(u) | unit_get_members_mask(u); | |
1189 | } | |
1190 | ||
1191 | CGroupMask unit_get_target_mask(Unit *u) { | |
1192 | CGroupMask mask; | |
1193 | ||
1194 | /* This returns the cgroup mask of all controllers to enable | |
1195 | * for a specific cgroup, i.e. everything it needs itself, | |
1196 | * plus all that its children need, plus all that its siblings | |
1197 | * need. This is primarily useful on the legacy cgroup | |
1198 | * hierarchy, where we need to duplicate each cgroup in each | |
1199 | * hierarchy that shall be enabled for it. */ | |
6414b7c9 | 1200 | |
efdb0237 LP |
1201 | mask = unit_get_own_mask(u) | unit_get_members_mask(u) | unit_get_siblings_mask(u); |
1202 | mask &= u->manager->cgroup_supported; | |
1203 | ||
1204 | return mask; | |
1205 | } | |
1206 | ||
1207 | CGroupMask unit_get_enable_mask(Unit *u) { | |
1208 | CGroupMask mask; | |
1209 | ||
1210 | /* This returns the cgroup mask of all controllers to enable | |
1211 | * for the children of a specific cgroup. This is primarily | |
1212 | * useful for the unified cgroup hierarchy, where each cgroup | |
1213 | * controls which controllers are enabled for its children. */ | |
1214 | ||
1215 | mask = unit_get_members_mask(u); | |
6414b7c9 DS |
1216 | mask &= u->manager->cgroup_supported; |
1217 | ||
1218 | return mask; | |
1219 | } | |
1220 | ||
906c06f6 DM |
1221 | bool unit_get_needs_bpf(Unit *u) { |
1222 | CGroupContext *c; | |
1223 | Unit *p; | |
1224 | assert(u); | |
1225 | ||
906c06f6 DM |
1226 | c = unit_get_cgroup_context(u); |
1227 | if (!c) | |
1228 | return false; | |
1229 | ||
1230 | if (c->ip_accounting || | |
1231 | c->ip_address_allow || | |
1232 | c->ip_address_deny) | |
1233 | return true; | |
1234 | ||
1235 | /* If any parent slice has an IP access list defined, it applies too */ | |
1236 | for (p = UNIT_DEREF(u->slice); p; p = UNIT_DEREF(p->slice)) { | |
1237 | c = unit_get_cgroup_context(p); | |
1238 | if (!c) | |
1239 | return false; | |
1240 | ||
1241 | if (c->ip_address_allow || | |
1242 | c->ip_address_deny) | |
1243 | return true; | |
1244 | } | |
1245 | ||
1246 | return false; | |
1247 | } | |
1248 | ||
6414b7c9 DS |
1249 | /* Recurse from a unit up through its containing slices, propagating |
1250 | * mask bits upward. A unit is also member of itself. */ | |
bc432dc7 | 1251 | void unit_update_cgroup_members_masks(Unit *u) { |
efdb0237 | 1252 | CGroupMask m; |
bc432dc7 LP |
1253 | bool more; |
1254 | ||
1255 | assert(u); | |
1256 | ||
1257 | /* Calculate subtree mask */ | |
efdb0237 | 1258 | m = unit_get_subtree_mask(u); |
bc432dc7 LP |
1259 | |
1260 | /* See if anything changed from the previous invocation. If | |
1261 | * not, we're done. */ | |
1262 | if (u->cgroup_subtree_mask_valid && m == u->cgroup_subtree_mask) | |
1263 | return; | |
1264 | ||
1265 | more = | |
1266 | u->cgroup_subtree_mask_valid && | |
1267 | ((m & ~u->cgroup_subtree_mask) != 0) && | |
1268 | ((~m & u->cgroup_subtree_mask) == 0); | |
1269 | ||
1270 | u->cgroup_subtree_mask = m; | |
1271 | u->cgroup_subtree_mask_valid = true; | |
1272 | ||
6414b7c9 DS |
1273 | if (UNIT_ISSET(u->slice)) { |
1274 | Unit *s = UNIT_DEREF(u->slice); | |
bc432dc7 LP |
1275 | |
1276 | if (more) | |
1277 | /* There's more set now than before. We | |
1278 | * propagate the new mask to the parent's mask | |
1279 | * (not caring if it actually was valid or | |
1280 | * not). */ | |
1281 | ||
1282 | s->cgroup_members_mask |= m; | |
1283 | ||
1284 | else | |
1285 | /* There's less set now than before (or we | |
1286 | * don't know), we need to recalculate | |
1287 | * everything, so let's invalidate the | |
1288 | * parent's members mask */ | |
1289 | ||
1290 | s->cgroup_members_mask_valid = false; | |
1291 | ||
1292 | /* And now make sure that this change also hits our | |
1293 | * grandparents */ | |
1294 | unit_update_cgroup_members_masks(s); | |
6414b7c9 DS |
1295 | } |
1296 | } | |
1297 | ||
6592b975 | 1298 | const char *unit_get_realized_cgroup_path(Unit *u, CGroupMask mask) { |
03b90d4b | 1299 | |
6592b975 | 1300 | /* Returns the realized cgroup path of the specified unit where all specified controllers are available. */ |
03b90d4b LP |
1301 | |
1302 | while (u) { | |
6592b975 | 1303 | |
03b90d4b LP |
1304 | if (u->cgroup_path && |
1305 | u->cgroup_realized && | |
d94a24ca | 1306 | FLAGS_SET(u->cgroup_realized_mask, mask)) |
03b90d4b LP |
1307 | return u->cgroup_path; |
1308 | ||
1309 | u = UNIT_DEREF(u->slice); | |
1310 | } | |
1311 | ||
1312 | return NULL; | |
1313 | } | |
1314 | ||
6592b975 LP |
1315 | static const char *migrate_callback(CGroupMask mask, void *userdata) { |
1316 | return unit_get_realized_cgroup_path(userdata, mask); | |
1317 | } | |
1318 | ||
efdb0237 LP |
1319 | char *unit_default_cgroup_path(Unit *u) { |
1320 | _cleanup_free_ char *escaped = NULL, *slice = NULL; | |
1321 | int r; | |
1322 | ||
1323 | assert(u); | |
1324 | ||
1325 | if (unit_has_name(u, SPECIAL_ROOT_SLICE)) | |
1326 | return strdup(u->manager->cgroup_root); | |
1327 | ||
1328 | if (UNIT_ISSET(u->slice) && !unit_has_name(UNIT_DEREF(u->slice), SPECIAL_ROOT_SLICE)) { | |
1329 | r = cg_slice_to_path(UNIT_DEREF(u->slice)->id, &slice); | |
1330 | if (r < 0) | |
1331 | return NULL; | |
1332 | } | |
1333 | ||
1334 | escaped = cg_escape(u->id); | |
1335 | if (!escaped) | |
1336 | return NULL; | |
1337 | ||
1338 | if (slice) | |
605405c6 ZJS |
1339 | return strjoin(u->manager->cgroup_root, "/", slice, "/", |
1340 | escaped); | |
efdb0237 | 1341 | else |
605405c6 | 1342 | return strjoin(u->manager->cgroup_root, "/", escaped); |
efdb0237 LP |
1343 | } |
1344 | ||
1345 | int unit_set_cgroup_path(Unit *u, const char *path) { | |
1346 | _cleanup_free_ char *p = NULL; | |
1347 | int r; | |
1348 | ||
1349 | assert(u); | |
1350 | ||
1351 | if (path) { | |
1352 | p = strdup(path); | |
1353 | if (!p) | |
1354 | return -ENOMEM; | |
1355 | } else | |
1356 | p = NULL; | |
1357 | ||
1358 | if (streq_ptr(u->cgroup_path, p)) | |
1359 | return 0; | |
1360 | ||
1361 | if (p) { | |
1362 | r = hashmap_put(u->manager->cgroup_unit, p, u); | |
1363 | if (r < 0) | |
1364 | return r; | |
1365 | } | |
1366 | ||
1367 | unit_release_cgroup(u); | |
1368 | ||
ae2a15bc | 1369 | u->cgroup_path = TAKE_PTR(p); |
efdb0237 LP |
1370 | |
1371 | return 1; | |
1372 | } | |
1373 | ||
1374 | int unit_watch_cgroup(Unit *u) { | |
ab2c3861 | 1375 | _cleanup_free_ char *events = NULL; |
efdb0237 LP |
1376 | int r; |
1377 | ||
1378 | assert(u); | |
1379 | ||
1380 | if (!u->cgroup_path) | |
1381 | return 0; | |
1382 | ||
1383 | if (u->cgroup_inotify_wd >= 0) | |
1384 | return 0; | |
1385 | ||
1386 | /* Only applies to the unified hierarchy */ | |
c22800e4 | 1387 | r = cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER); |
b4cccbc1 LP |
1388 | if (r < 0) |
1389 | return log_error_errno(r, "Failed to determine whether the name=systemd hierarchy is unified: %m"); | |
1390 | if (r == 0) | |
efdb0237 LP |
1391 | return 0; |
1392 | ||
1393 | /* Don't watch the root slice, it's pointless. */ | |
1394 | if (unit_has_name(u, SPECIAL_ROOT_SLICE)) | |
1395 | return 0; | |
1396 | ||
1397 | r = hashmap_ensure_allocated(&u->manager->cgroup_inotify_wd_unit, &trivial_hash_ops); | |
1398 | if (r < 0) | |
1399 | return log_oom(); | |
1400 | ||
ab2c3861 | 1401 | r = cg_get_path(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path, "cgroup.events", &events); |
efdb0237 LP |
1402 | if (r < 0) |
1403 | return log_oom(); | |
1404 | ||
ab2c3861 | 1405 | u->cgroup_inotify_wd = inotify_add_watch(u->manager->cgroup_inotify_fd, events, IN_MODIFY); |
efdb0237 LP |
1406 | if (u->cgroup_inotify_wd < 0) { |
1407 | ||
1408 | if (errno == ENOENT) /* If the directory is already | |
1409 | * gone we don't need to track | |
1410 | * it, so this is not an error */ | |
1411 | return 0; | |
1412 | ||
1413 | return log_unit_error_errno(u, errno, "Failed to add inotify watch descriptor for control group %s: %m", u->cgroup_path); | |
1414 | } | |
1415 | ||
1416 | r = hashmap_put(u->manager->cgroup_inotify_wd_unit, INT_TO_PTR(u->cgroup_inotify_wd), u); | |
1417 | if (r < 0) | |
1418 | return log_unit_error_errno(u, r, "Failed to add inotify watch descriptor to hash map: %m"); | |
1419 | ||
1420 | return 0; | |
1421 | } | |
1422 | ||
a4634b21 LP |
1423 | int unit_pick_cgroup_path(Unit *u) { |
1424 | _cleanup_free_ char *path = NULL; | |
1425 | int r; | |
1426 | ||
1427 | assert(u); | |
1428 | ||
1429 | if (u->cgroup_path) | |
1430 | return 0; | |
1431 | ||
1432 | if (!UNIT_HAS_CGROUP_CONTEXT(u)) | |
1433 | return -EINVAL; | |
1434 | ||
1435 | path = unit_default_cgroup_path(u); | |
1436 | if (!path) | |
1437 | return log_oom(); | |
1438 | ||
1439 | r = unit_set_cgroup_path(u, path); | |
1440 | if (r == -EEXIST) | |
1441 | return log_unit_error_errno(u, r, "Control group %s exists already.", path); | |
1442 | if (r < 0) | |
1443 | return log_unit_error_errno(u, r, "Failed to set unit's control group path to %s: %m", path); | |
1444 | ||
1445 | return 0; | |
1446 | } | |
1447 | ||
efdb0237 LP |
1448 | static int unit_create_cgroup( |
1449 | Unit *u, | |
1450 | CGroupMask target_mask, | |
906c06f6 DM |
1451 | CGroupMask enable_mask, |
1452 | bool needs_bpf) { | |
efdb0237 | 1453 | |
0cd385d3 | 1454 | CGroupContext *c; |
bc432dc7 | 1455 | int r; |
64747e2d | 1456 | |
4ad49000 | 1457 | assert(u); |
64747e2d | 1458 | |
0cd385d3 LP |
1459 | c = unit_get_cgroup_context(u); |
1460 | if (!c) | |
1461 | return 0; | |
1462 | ||
a4634b21 LP |
1463 | /* Figure out our cgroup path */ |
1464 | r = unit_pick_cgroup_path(u); | |
1465 | if (r < 0) | |
1466 | return r; | |
b58b8e11 | 1467 | |
03b90d4b | 1468 | /* First, create our own group */ |
efdb0237 | 1469 | r = cg_create_everywhere(u->manager->cgroup_supported, target_mask, u->cgroup_path); |
23bbb0de | 1470 | if (r < 0) |
efdb0237 LP |
1471 | return log_unit_error_errno(u, r, "Failed to create cgroup %s: %m", u->cgroup_path); |
1472 | ||
1473 | /* Start watching it */ | |
1474 | (void) unit_watch_cgroup(u); | |
1475 | ||
1476 | /* Enable all controllers we need */ | |
1477 | r = cg_enable_everywhere(u->manager->cgroup_supported, enable_mask, u->cgroup_path); | |
1478 | if (r < 0) | |
1479 | log_unit_warning_errno(u, r, "Failed to enable controllers on cgroup %s, ignoring: %m", u->cgroup_path); | |
03b90d4b LP |
1480 | |
1481 | /* Keep track that this is now realized */ | |
4ad49000 | 1482 | u->cgroup_realized = true; |
efdb0237 | 1483 | u->cgroup_realized_mask = target_mask; |
ccf78df1 | 1484 | u->cgroup_enabled_mask = enable_mask; |
906c06f6 | 1485 | u->cgroup_bpf_state = needs_bpf ? UNIT_CGROUP_BPF_ON : UNIT_CGROUP_BPF_OFF; |
4ad49000 | 1486 | |
1d9cc876 | 1487 | if (u->type != UNIT_SLICE && !unit_cgroup_delegate(u)) { |
0cd385d3 LP |
1488 | |
1489 | /* Then, possibly move things over, but not if | |
1490 | * subgroups may contain processes, which is the case | |
1491 | * for slice and delegation units. */ | |
1492 | r = cg_migrate_everywhere(u->manager->cgroup_supported, u->cgroup_path, u->cgroup_path, migrate_callback, u); | |
1493 | if (r < 0) | |
efdb0237 | 1494 | log_unit_warning_errno(u, r, "Failed to migrate cgroup from to %s, ignoring: %m", u->cgroup_path); |
0cd385d3 | 1495 | } |
03b90d4b | 1496 | |
64747e2d LP |
1497 | return 0; |
1498 | } | |
1499 | ||
6592b975 LP |
1500 | static int unit_attach_pid_to_cgroup_via_bus(Unit *u, pid_t pid, const char *suffix_path) { |
1501 | _cleanup_(sd_bus_error_free) sd_bus_error error = SD_BUS_ERROR_NULL; | |
1502 | char *pp; | |
7b3fd631 | 1503 | int r; |
6592b975 | 1504 | |
7b3fd631 LP |
1505 | assert(u); |
1506 | ||
6592b975 LP |
1507 | if (MANAGER_IS_SYSTEM(u->manager)) |
1508 | return -EINVAL; | |
1509 | ||
1510 | if (!u->manager->system_bus) | |
1511 | return -EIO; | |
1512 | ||
1513 | if (!u->cgroup_path) | |
1514 | return -EINVAL; | |
1515 | ||
1516 | /* Determine this unit's cgroup path relative to our cgroup root */ | |
1517 | pp = path_startswith(u->cgroup_path, u->manager->cgroup_root); | |
1518 | if (!pp) | |
1519 | return -EINVAL; | |
1520 | ||
1521 | pp = strjoina("/", pp, suffix_path); | |
858d36c1 | 1522 | path_simplify(pp, false); |
6592b975 LP |
1523 | |
1524 | r = sd_bus_call_method(u->manager->system_bus, | |
1525 | "org.freedesktop.systemd1", | |
1526 | "/org/freedesktop/systemd1", | |
1527 | "org.freedesktop.systemd1.Manager", | |
1528 | "AttachProcessesToUnit", | |
1529 | &error, NULL, | |
1530 | "ssau", | |
1531 | NULL /* empty unit name means client's unit, i.e. us */, pp, 1, (uint32_t) pid); | |
7b3fd631 | 1532 | if (r < 0) |
6592b975 LP |
1533 | return log_unit_debug_errno(u, r, "Failed to attach unit process " PID_FMT " via the bus: %s", pid, bus_error_message(&error, r)); |
1534 | ||
1535 | return 0; | |
1536 | } | |
1537 | ||
1538 | int unit_attach_pids_to_cgroup(Unit *u, Set *pids, const char *suffix_path) { | |
1539 | CGroupMask delegated_mask; | |
1540 | const char *p; | |
1541 | Iterator i; | |
1542 | void *pidp; | |
1543 | int r, q; | |
1544 | ||
1545 | assert(u); | |
1546 | ||
1547 | if (!UNIT_HAS_CGROUP_CONTEXT(u)) | |
1548 | return -EINVAL; | |
1549 | ||
1550 | if (set_isempty(pids)) | |
1551 | return 0; | |
7b3fd631 | 1552 | |
6592b975 | 1553 | r = unit_realize_cgroup(u); |
7b3fd631 LP |
1554 | if (r < 0) |
1555 | return r; | |
1556 | ||
6592b975 LP |
1557 | if (isempty(suffix_path)) |
1558 | p = u->cgroup_path; | |
1559 | else | |
1560 | p = strjoina(u->cgroup_path, "/", suffix_path); | |
1561 | ||
1562 | delegated_mask = unit_get_delegate_mask(u); | |
1563 | ||
1564 | r = 0; | |
1565 | SET_FOREACH(pidp, pids, i) { | |
1566 | pid_t pid = PTR_TO_PID(pidp); | |
1567 | CGroupController c; | |
1568 | ||
1569 | /* First, attach the PID to the main cgroup hierarchy */ | |
1570 | q = cg_attach(SYSTEMD_CGROUP_CONTROLLER, p, pid); | |
1571 | if (q < 0) { | |
1572 | log_unit_debug_errno(u, q, "Couldn't move process " PID_FMT " to requested cgroup '%s': %m", pid, p); | |
1573 | ||
1574 | if (MANAGER_IS_USER(u->manager) && IN_SET(q, -EPERM, -EACCES)) { | |
1575 | int z; | |
1576 | ||
1577 | /* If we are in a user instance, and we can't move the process ourselves due to | |
1578 | * permission problems, let's ask the system instance about it instead. Since it's more | |
1579 | * privileged it might be able to move the process across the leaves of a subtree who's | |
1580 | * top node is not owned by us. */ | |
1581 | ||
1582 | z = unit_attach_pid_to_cgroup_via_bus(u, pid, suffix_path); | |
1583 | if (z < 0) | |
1584 | log_unit_debug_errno(u, z, "Couldn't move process " PID_FMT " to requested cgroup '%s' via the system bus either: %m", pid, p); | |
1585 | else | |
1586 | continue; /* When the bus thing worked via the bus we are fully done for this PID. */ | |
1587 | } | |
1588 | ||
1589 | if (r >= 0) | |
1590 | r = q; /* Remember first error */ | |
1591 | ||
1592 | continue; | |
1593 | } | |
1594 | ||
1595 | q = cg_all_unified(); | |
1596 | if (q < 0) | |
1597 | return q; | |
1598 | if (q > 0) | |
1599 | continue; | |
1600 | ||
1601 | /* In the legacy hierarchy, attach the process to the request cgroup if possible, and if not to the | |
1602 | * innermost realized one */ | |
1603 | ||
1604 | for (c = 0; c < _CGROUP_CONTROLLER_MAX; c++) { | |
1605 | CGroupMask bit = CGROUP_CONTROLLER_TO_MASK(c); | |
1606 | const char *realized; | |
1607 | ||
1608 | if (!(u->manager->cgroup_supported & bit)) | |
1609 | continue; | |
1610 | ||
1611 | /* If this controller is delegated and realized, honour the caller's request for the cgroup suffix. */ | |
1612 | if (delegated_mask & u->cgroup_realized_mask & bit) { | |
1613 | q = cg_attach(cgroup_controller_to_string(c), p, pid); | |
1614 | if (q >= 0) | |
1615 | continue; /* Success! */ | |
1616 | ||
1617 | 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", | |
1618 | pid, p, cgroup_controller_to_string(c)); | |
1619 | } | |
1620 | ||
1621 | /* So this controller is either not delegate or realized, or something else weird happened. In | |
1622 | * that case let's attach the PID at least to the closest cgroup up the tree that is | |
1623 | * realized. */ | |
1624 | realized = unit_get_realized_cgroup_path(u, bit); | |
1625 | if (!realized) | |
1626 | continue; /* Not even realized in the root slice? Then let's not bother */ | |
1627 | ||
1628 | q = cg_attach(cgroup_controller_to_string(c), realized, pid); | |
1629 | if (q < 0) | |
1630 | log_unit_debug_errno(u, q, "Failed to attach PID " PID_FMT " to realized cgroup %s in controller %s, ignoring: %m", | |
1631 | pid, realized, cgroup_controller_to_string(c)); | |
1632 | } | |
1633 | } | |
1634 | ||
1635 | return r; | |
7b3fd631 LP |
1636 | } |
1637 | ||
4b58153d LP |
1638 | static void cgroup_xattr_apply(Unit *u) { |
1639 | char ids[SD_ID128_STRING_MAX]; | |
1640 | int r; | |
1641 | ||
1642 | assert(u); | |
1643 | ||
1644 | if (!MANAGER_IS_SYSTEM(u->manager)) | |
1645 | return; | |
1646 | ||
1647 | if (sd_id128_is_null(u->invocation_id)) | |
1648 | return; | |
1649 | ||
1650 | r = cg_set_xattr(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path, | |
1651 | "trusted.invocation_id", | |
1652 | sd_id128_to_string(u->invocation_id, ids), 32, | |
1653 | 0); | |
1654 | if (r < 0) | |
0fb84499 | 1655 | log_unit_debug_errno(u, r, "Failed to set invocation ID on control group %s, ignoring: %m", u->cgroup_path); |
4b58153d LP |
1656 | } |
1657 | ||
906c06f6 DM |
1658 | static bool unit_has_mask_realized( |
1659 | Unit *u, | |
1660 | CGroupMask target_mask, | |
1661 | CGroupMask enable_mask, | |
1662 | bool needs_bpf) { | |
1663 | ||
bc432dc7 LP |
1664 | assert(u); |
1665 | ||
906c06f6 DM |
1666 | return u->cgroup_realized && |
1667 | u->cgroup_realized_mask == target_mask && | |
1668 | u->cgroup_enabled_mask == enable_mask && | |
1669 | ((needs_bpf && u->cgroup_bpf_state == UNIT_CGROUP_BPF_ON) || | |
1670 | (!needs_bpf && u->cgroup_bpf_state == UNIT_CGROUP_BPF_OFF)); | |
6414b7c9 DS |
1671 | } |
1672 | ||
2aa57a65 LP |
1673 | static void unit_add_to_cgroup_realize_queue(Unit *u) { |
1674 | assert(u); | |
1675 | ||
1676 | if (u->in_cgroup_realize_queue) | |
1677 | return; | |
1678 | ||
1679 | LIST_PREPEND(cgroup_realize_queue, u->manager->cgroup_realize_queue, u); | |
1680 | u->in_cgroup_realize_queue = true; | |
1681 | } | |
1682 | ||
1683 | static void unit_remove_from_cgroup_realize_queue(Unit *u) { | |
1684 | assert(u); | |
1685 | ||
1686 | if (!u->in_cgroup_realize_queue) | |
1687 | return; | |
1688 | ||
1689 | LIST_REMOVE(cgroup_realize_queue, u->manager->cgroup_realize_queue, u); | |
1690 | u->in_cgroup_realize_queue = false; | |
1691 | } | |
1692 | ||
6414b7c9 DS |
1693 | /* Check if necessary controllers and attributes for a unit are in place. |
1694 | * | |
1695 | * If so, do nothing. | |
1696 | * If not, create paths, move processes over, and set attributes. | |
1697 | * | |
1698 | * Returns 0 on success and < 0 on failure. */ | |
db785129 | 1699 | static int unit_realize_cgroup_now(Unit *u, ManagerState state) { |
efdb0237 | 1700 | CGroupMask target_mask, enable_mask; |
906c06f6 | 1701 | bool needs_bpf, apply_bpf; |
6414b7c9 | 1702 | int r; |
64747e2d | 1703 | |
4ad49000 | 1704 | assert(u); |
64747e2d | 1705 | |
2aa57a65 | 1706 | unit_remove_from_cgroup_realize_queue(u); |
64747e2d | 1707 | |
efdb0237 | 1708 | target_mask = unit_get_target_mask(u); |
ccf78df1 | 1709 | enable_mask = unit_get_enable_mask(u); |
906c06f6 | 1710 | needs_bpf = unit_get_needs_bpf(u); |
ccf78df1 | 1711 | |
906c06f6 | 1712 | if (unit_has_mask_realized(u, target_mask, enable_mask, needs_bpf)) |
0a1eb06d | 1713 | return 0; |
64747e2d | 1714 | |
906c06f6 DM |
1715 | /* Make sure we apply the BPF filters either when one is configured, or if none is configured but previously |
1716 | * the state was anything but off. This way, if a unit with a BPF filter applied is reconfigured to lose it | |
1717 | * this will trickle down properly to cgroupfs. */ | |
1718 | apply_bpf = needs_bpf || u->cgroup_bpf_state != UNIT_CGROUP_BPF_OFF; | |
1719 | ||
4ad49000 | 1720 | /* First, realize parents */ |
6414b7c9 | 1721 | if (UNIT_ISSET(u->slice)) { |
db785129 | 1722 | r = unit_realize_cgroup_now(UNIT_DEREF(u->slice), state); |
6414b7c9 DS |
1723 | if (r < 0) |
1724 | return r; | |
1725 | } | |
4ad49000 LP |
1726 | |
1727 | /* And then do the real work */ | |
906c06f6 | 1728 | r = unit_create_cgroup(u, target_mask, enable_mask, needs_bpf); |
6414b7c9 DS |
1729 | if (r < 0) |
1730 | return r; | |
1731 | ||
1732 | /* Finally, apply the necessary attributes. */ | |
906c06f6 | 1733 | cgroup_context_apply(u, target_mask, apply_bpf, state); |
4b58153d | 1734 | cgroup_xattr_apply(u); |
6414b7c9 DS |
1735 | |
1736 | return 0; | |
64747e2d LP |
1737 | } |
1738 | ||
91a6073e | 1739 | unsigned manager_dispatch_cgroup_realize_queue(Manager *m) { |
db785129 | 1740 | ManagerState state; |
4ad49000 | 1741 | unsigned n = 0; |
db785129 | 1742 | Unit *i; |
6414b7c9 | 1743 | int r; |
ecedd90f | 1744 | |
91a6073e LP |
1745 | assert(m); |
1746 | ||
db785129 LP |
1747 | state = manager_state(m); |
1748 | ||
91a6073e LP |
1749 | while ((i = m->cgroup_realize_queue)) { |
1750 | assert(i->in_cgroup_realize_queue); | |
ecedd90f | 1751 | |
2aa57a65 LP |
1752 | if (UNIT_IS_INACTIVE_OR_FAILED(unit_active_state(i))) { |
1753 | /* Maybe things changed, and the unit is not actually active anymore? */ | |
1754 | unit_remove_from_cgroup_realize_queue(i); | |
1755 | continue; | |
1756 | } | |
1757 | ||
db785129 | 1758 | r = unit_realize_cgroup_now(i, state); |
6414b7c9 | 1759 | if (r < 0) |
efdb0237 | 1760 | log_warning_errno(r, "Failed to realize cgroups for queued unit %s, ignoring: %m", i->id); |
0a1eb06d | 1761 | |
4ad49000 LP |
1762 | n++; |
1763 | } | |
ecedd90f | 1764 | |
4ad49000 | 1765 | return n; |
8e274523 LP |
1766 | } |
1767 | ||
91a6073e | 1768 | static void unit_add_siblings_to_cgroup_realize_queue(Unit *u) { |
4ad49000 | 1769 | Unit *slice; |
ca949c9d | 1770 | |
4ad49000 LP |
1771 | /* This adds the siblings of the specified unit and the |
1772 | * siblings of all parent units to the cgroup queue. (But | |
1773 | * neither the specified unit itself nor the parents.) */ | |
1774 | ||
1775 | while ((slice = UNIT_DEREF(u->slice))) { | |
1776 | Iterator i; | |
1777 | Unit *m; | |
eef85c4a | 1778 | void *v; |
8f53a7b8 | 1779 | |
eef85c4a | 1780 | HASHMAP_FOREACH_KEY(v, m, u->dependencies[UNIT_BEFORE], i) { |
4ad49000 LP |
1781 | if (m == u) |
1782 | continue; | |
8e274523 | 1783 | |
6414b7c9 DS |
1784 | /* Skip units that have a dependency on the slice |
1785 | * but aren't actually in it. */ | |
4ad49000 | 1786 | if (UNIT_DEREF(m->slice) != slice) |
50159e6a | 1787 | continue; |
8e274523 | 1788 | |
6414b7c9 DS |
1789 | /* No point in doing cgroup application for units |
1790 | * without active processes. */ | |
1791 | if (UNIT_IS_INACTIVE_OR_FAILED(unit_active_state(m))) | |
1792 | continue; | |
1793 | ||
1794 | /* If the unit doesn't need any new controllers | |
1795 | * and has current ones realized, it doesn't need | |
1796 | * any changes. */ | |
906c06f6 DM |
1797 | if (unit_has_mask_realized(m, |
1798 | unit_get_target_mask(m), | |
1799 | unit_get_enable_mask(m), | |
1800 | unit_get_needs_bpf(m))) | |
6414b7c9 DS |
1801 | continue; |
1802 | ||
91a6073e | 1803 | unit_add_to_cgroup_realize_queue(m); |
50159e6a LP |
1804 | } |
1805 | ||
4ad49000 | 1806 | u = slice; |
8e274523 | 1807 | } |
4ad49000 LP |
1808 | } |
1809 | ||
0a1eb06d | 1810 | int unit_realize_cgroup(Unit *u) { |
4ad49000 LP |
1811 | assert(u); |
1812 | ||
35b7ff80 | 1813 | if (!UNIT_HAS_CGROUP_CONTEXT(u)) |
0a1eb06d | 1814 | return 0; |
8e274523 | 1815 | |
4ad49000 LP |
1816 | /* So, here's the deal: when realizing the cgroups for this |
1817 | * unit, we need to first create all parents, but there's more | |
1818 | * actually: for the weight-based controllers we also need to | |
1819 | * make sure that all our siblings (i.e. units that are in the | |
73e231ab | 1820 | * same slice as we are) have cgroups, too. Otherwise, things |
4ad49000 LP |
1821 | * would become very uneven as each of their processes would |
1822 | * get as much resources as all our group together. This call | |
1823 | * will synchronously create the parent cgroups, but will | |
1824 | * defer work on the siblings to the next event loop | |
1825 | * iteration. */ | |
ca949c9d | 1826 | |
4ad49000 | 1827 | /* Add all sibling slices to the cgroup queue. */ |
91a6073e | 1828 | unit_add_siblings_to_cgroup_realize_queue(u); |
4ad49000 | 1829 | |
6414b7c9 | 1830 | /* And realize this one now (and apply the values) */ |
db785129 | 1831 | return unit_realize_cgroup_now(u, manager_state(u->manager)); |
8e274523 LP |
1832 | } |
1833 | ||
efdb0237 LP |
1834 | void unit_release_cgroup(Unit *u) { |
1835 | assert(u); | |
1836 | ||
1837 | /* Forgets all cgroup details for this cgroup */ | |
1838 | ||
1839 | if (u->cgroup_path) { | |
1840 | (void) hashmap_remove(u->manager->cgroup_unit, u->cgroup_path); | |
1841 | u->cgroup_path = mfree(u->cgroup_path); | |
1842 | } | |
1843 | ||
1844 | if (u->cgroup_inotify_wd >= 0) { | |
1845 | if (inotify_rm_watch(u->manager->cgroup_inotify_fd, u->cgroup_inotify_wd) < 0) | |
1846 | log_unit_debug_errno(u, errno, "Failed to remove cgroup inotify watch %i for %s, ignoring", u->cgroup_inotify_wd, u->id); | |
1847 | ||
1848 | (void) hashmap_remove(u->manager->cgroup_inotify_wd_unit, INT_TO_PTR(u->cgroup_inotify_wd)); | |
1849 | u->cgroup_inotify_wd = -1; | |
1850 | } | |
1851 | } | |
1852 | ||
1853 | void unit_prune_cgroup(Unit *u) { | |
8e274523 | 1854 | int r; |
efdb0237 | 1855 | bool is_root_slice; |
8e274523 | 1856 | |
4ad49000 | 1857 | assert(u); |
8e274523 | 1858 | |
efdb0237 LP |
1859 | /* Removes the cgroup, if empty and possible, and stops watching it. */ |
1860 | ||
4ad49000 LP |
1861 | if (!u->cgroup_path) |
1862 | return; | |
8e274523 | 1863 | |
fe700f46 LP |
1864 | (void) unit_get_cpu_usage(u, NULL); /* Cache the last CPU usage value before we destroy the cgroup */ |
1865 | ||
efdb0237 LP |
1866 | is_root_slice = unit_has_name(u, SPECIAL_ROOT_SLICE); |
1867 | ||
1868 | r = cg_trim_everywhere(u->manager->cgroup_supported, u->cgroup_path, !is_root_slice); | |
dab5bf85 | 1869 | if (r < 0) { |
f29ff115 | 1870 | log_unit_debug_errno(u, r, "Failed to destroy cgroup %s, ignoring: %m", u->cgroup_path); |
dab5bf85 RL |
1871 | return; |
1872 | } | |
8e274523 | 1873 | |
efdb0237 LP |
1874 | if (is_root_slice) |
1875 | return; | |
1876 | ||
1877 | unit_release_cgroup(u); | |
0a1eb06d | 1878 | |
4ad49000 | 1879 | u->cgroup_realized = false; |
bc432dc7 | 1880 | u->cgroup_realized_mask = 0; |
ccf78df1 | 1881 | u->cgroup_enabled_mask = 0; |
8e274523 LP |
1882 | } |
1883 | ||
efdb0237 | 1884 | int unit_search_main_pid(Unit *u, pid_t *ret) { |
4ad49000 LP |
1885 | _cleanup_fclose_ FILE *f = NULL; |
1886 | pid_t pid = 0, npid, mypid; | |
efdb0237 | 1887 | int r; |
4ad49000 LP |
1888 | |
1889 | assert(u); | |
efdb0237 | 1890 | assert(ret); |
4ad49000 LP |
1891 | |
1892 | if (!u->cgroup_path) | |
efdb0237 | 1893 | return -ENXIO; |
4ad49000 | 1894 | |
efdb0237 LP |
1895 | r = cg_enumerate_processes(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path, &f); |
1896 | if (r < 0) | |
1897 | return r; | |
4ad49000 | 1898 | |
df0ff127 | 1899 | mypid = getpid_cached(); |
4ad49000 LP |
1900 | while (cg_read_pid(f, &npid) > 0) { |
1901 | pid_t ppid; | |
1902 | ||
1903 | if (npid == pid) | |
1904 | continue; | |
8e274523 | 1905 | |
4ad49000 | 1906 | /* Ignore processes that aren't our kids */ |
6bc73acb | 1907 | if (get_process_ppid(npid, &ppid) >= 0 && ppid != mypid) |
4ad49000 | 1908 | continue; |
8e274523 | 1909 | |
efdb0237 | 1910 | if (pid != 0) |
4ad49000 LP |
1911 | /* Dang, there's more than one daemonized PID |
1912 | in this group, so we don't know what process | |
1913 | is the main process. */ | |
efdb0237 LP |
1914 | |
1915 | return -ENODATA; | |
8e274523 | 1916 | |
4ad49000 | 1917 | pid = npid; |
8e274523 LP |
1918 | } |
1919 | ||
efdb0237 LP |
1920 | *ret = pid; |
1921 | return 0; | |
1922 | } | |
1923 | ||
1924 | static int unit_watch_pids_in_path(Unit *u, const char *path) { | |
b3c5bad3 | 1925 | _cleanup_closedir_ DIR *d = NULL; |
efdb0237 LP |
1926 | _cleanup_fclose_ FILE *f = NULL; |
1927 | int ret = 0, r; | |
1928 | ||
1929 | assert(u); | |
1930 | assert(path); | |
1931 | ||
1932 | r = cg_enumerate_processes(SYSTEMD_CGROUP_CONTROLLER, path, &f); | |
1933 | if (r < 0) | |
1934 | ret = r; | |
1935 | else { | |
1936 | pid_t pid; | |
1937 | ||
1938 | while ((r = cg_read_pid(f, &pid)) > 0) { | |
1939 | r = unit_watch_pid(u, pid); | |
1940 | if (r < 0 && ret >= 0) | |
1941 | ret = r; | |
1942 | } | |
1943 | ||
1944 | if (r < 0 && ret >= 0) | |
1945 | ret = r; | |
1946 | } | |
1947 | ||
1948 | r = cg_enumerate_subgroups(SYSTEMD_CGROUP_CONTROLLER, path, &d); | |
1949 | if (r < 0) { | |
1950 | if (ret >= 0) | |
1951 | ret = r; | |
1952 | } else { | |
1953 | char *fn; | |
1954 | ||
1955 | while ((r = cg_read_subgroup(d, &fn)) > 0) { | |
1956 | _cleanup_free_ char *p = NULL; | |
1957 | ||
605405c6 | 1958 | p = strjoin(path, "/", fn); |
efdb0237 LP |
1959 | free(fn); |
1960 | ||
1961 | if (!p) | |
1962 | return -ENOMEM; | |
1963 | ||
1964 | r = unit_watch_pids_in_path(u, p); | |
1965 | if (r < 0 && ret >= 0) | |
1966 | ret = r; | |
1967 | } | |
1968 | ||
1969 | if (r < 0 && ret >= 0) | |
1970 | ret = r; | |
1971 | } | |
1972 | ||
1973 | return ret; | |
1974 | } | |
1975 | ||
11aef522 LP |
1976 | int unit_synthesize_cgroup_empty_event(Unit *u) { |
1977 | int r; | |
1978 | ||
1979 | assert(u); | |
1980 | ||
1981 | /* Enqueue a synthetic cgroup empty event if this unit doesn't watch any PIDs anymore. This is compatibility | |
1982 | * support for non-unified systems where notifications aren't reliable, and hence need to take whatever we can | |
1983 | * get as notification source as soon as we stopped having any useful PIDs to watch for. */ | |
1984 | ||
1985 | if (!u->cgroup_path) | |
1986 | return -ENOENT; | |
1987 | ||
1988 | r = cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER); | |
1989 | if (r < 0) | |
1990 | return r; | |
1991 | if (r > 0) /* On unified we have reliable notifications, and don't need this */ | |
1992 | return 0; | |
1993 | ||
1994 | if (!set_isempty(u->pids)) | |
1995 | return 0; | |
1996 | ||
1997 | unit_add_to_cgroup_empty_queue(u); | |
1998 | return 0; | |
1999 | } | |
2000 | ||
efdb0237 | 2001 | int unit_watch_all_pids(Unit *u) { |
b4cccbc1 LP |
2002 | int r; |
2003 | ||
efdb0237 LP |
2004 | assert(u); |
2005 | ||
2006 | /* Adds all PIDs from our cgroup to the set of PIDs we | |
2007 | * watch. This is a fallback logic for cases where we do not | |
2008 | * get reliable cgroup empty notifications: we try to use | |
2009 | * SIGCHLD as replacement. */ | |
2010 | ||
2011 | if (!u->cgroup_path) | |
2012 | return -ENOENT; | |
2013 | ||
c22800e4 | 2014 | r = cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER); |
b4cccbc1 LP |
2015 | if (r < 0) |
2016 | return r; | |
2017 | if (r > 0) /* On unified we can use proper notifications */ | |
efdb0237 LP |
2018 | return 0; |
2019 | ||
2020 | return unit_watch_pids_in_path(u, u->cgroup_path); | |
2021 | } | |
2022 | ||
09e24654 LP |
2023 | static int on_cgroup_empty_event(sd_event_source *s, void *userdata) { |
2024 | Manager *m = userdata; | |
2025 | Unit *u; | |
efdb0237 LP |
2026 | int r; |
2027 | ||
09e24654 LP |
2028 | assert(s); |
2029 | assert(m); | |
efdb0237 | 2030 | |
09e24654 LP |
2031 | u = m->cgroup_empty_queue; |
2032 | if (!u) | |
efdb0237 LP |
2033 | return 0; |
2034 | ||
09e24654 LP |
2035 | assert(u->in_cgroup_empty_queue); |
2036 | u->in_cgroup_empty_queue = false; | |
2037 | LIST_REMOVE(cgroup_empty_queue, m->cgroup_empty_queue, u); | |
2038 | ||
2039 | if (m->cgroup_empty_queue) { | |
2040 | /* More stuff queued, let's make sure we remain enabled */ | |
2041 | r = sd_event_source_set_enabled(s, SD_EVENT_ONESHOT); | |
2042 | if (r < 0) | |
19a691a9 | 2043 | log_debug_errno(r, "Failed to reenable cgroup empty event source, ignoring: %m"); |
09e24654 | 2044 | } |
efdb0237 LP |
2045 | |
2046 | unit_add_to_gc_queue(u); | |
2047 | ||
2048 | if (UNIT_VTABLE(u)->notify_cgroup_empty) | |
2049 | UNIT_VTABLE(u)->notify_cgroup_empty(u); | |
2050 | ||
2051 | return 0; | |
2052 | } | |
2053 | ||
09e24654 LP |
2054 | void unit_add_to_cgroup_empty_queue(Unit *u) { |
2055 | int r; | |
2056 | ||
2057 | assert(u); | |
2058 | ||
2059 | /* Note that there are four different ways how cgroup empty events reach us: | |
2060 | * | |
2061 | * 1. On the unified hierarchy we get an inotify event on the cgroup | |
2062 | * | |
2063 | * 2. On the legacy hierarchy, when running in system mode, we get a datagram on the cgroup agent socket | |
2064 | * | |
2065 | * 3. On the legacy hierarchy, when running in user mode, we get a D-Bus signal on the system bus | |
2066 | * | |
2067 | * 4. On the legacy hierarchy, in service units we start watching all processes of the cgroup for SIGCHLD as | |
2068 | * soon as we get one SIGCHLD, to deal with unreliable cgroup notifications. | |
2069 | * | |
2070 | * Regardless which way we got the notification, we'll verify it here, and then add it to a separate | |
2071 | * queue. This queue will be dispatched at a lower priority than the SIGCHLD handler, so that we always use | |
2072 | * SIGCHLD if we can get it first, and only use the cgroup empty notifications if there's no SIGCHLD pending | |
2073 | * (which might happen if the cgroup doesn't contain processes that are our own child, which is typically the | |
2074 | * case for scope units). */ | |
2075 | ||
2076 | if (u->in_cgroup_empty_queue) | |
2077 | return; | |
2078 | ||
2079 | /* Let's verify that the cgroup is really empty */ | |
2080 | if (!u->cgroup_path) | |
2081 | return; | |
2082 | r = cg_is_empty_recursive(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path); | |
2083 | if (r < 0) { | |
2084 | log_unit_debug_errno(u, r, "Failed to determine whether cgroup %s is empty: %m", u->cgroup_path); | |
2085 | return; | |
2086 | } | |
2087 | if (r == 0) | |
2088 | return; | |
2089 | ||
2090 | LIST_PREPEND(cgroup_empty_queue, u->manager->cgroup_empty_queue, u); | |
2091 | u->in_cgroup_empty_queue = true; | |
2092 | ||
2093 | /* Trigger the defer event */ | |
2094 | r = sd_event_source_set_enabled(u->manager->cgroup_empty_event_source, SD_EVENT_ONESHOT); | |
2095 | if (r < 0) | |
2096 | log_debug_errno(r, "Failed to enable cgroup empty event source: %m"); | |
2097 | } | |
2098 | ||
efdb0237 LP |
2099 | static int on_cgroup_inotify_event(sd_event_source *s, int fd, uint32_t revents, void *userdata) { |
2100 | Manager *m = userdata; | |
2101 | ||
2102 | assert(s); | |
2103 | assert(fd >= 0); | |
2104 | assert(m); | |
2105 | ||
2106 | for (;;) { | |
2107 | union inotify_event_buffer buffer; | |
2108 | struct inotify_event *e; | |
2109 | ssize_t l; | |
2110 | ||
2111 | l = read(fd, &buffer, sizeof(buffer)); | |
2112 | if (l < 0) { | |
47249640 | 2113 | if (IN_SET(errno, EINTR, EAGAIN)) |
efdb0237 LP |
2114 | return 0; |
2115 | ||
2116 | return log_error_errno(errno, "Failed to read control group inotify events: %m"); | |
2117 | } | |
2118 | ||
2119 | FOREACH_INOTIFY_EVENT(e, buffer, l) { | |
2120 | Unit *u; | |
2121 | ||
2122 | if (e->wd < 0) | |
2123 | /* Queue overflow has no watch descriptor */ | |
2124 | continue; | |
2125 | ||
2126 | if (e->mask & IN_IGNORED) | |
2127 | /* The watch was just removed */ | |
2128 | continue; | |
2129 | ||
2130 | u = hashmap_get(m->cgroup_inotify_wd_unit, INT_TO_PTR(e->wd)); | |
2131 | if (!u) /* Not that inotify might deliver | |
2132 | * events for a watch even after it | |
2133 | * was removed, because it was queued | |
2134 | * before the removal. Let's ignore | |
2135 | * this here safely. */ | |
2136 | continue; | |
2137 | ||
09e24654 | 2138 | unit_add_to_cgroup_empty_queue(u); |
efdb0237 LP |
2139 | } |
2140 | } | |
8e274523 LP |
2141 | } |
2142 | ||
8e274523 | 2143 | int manager_setup_cgroup(Manager *m) { |
9444b1f2 | 2144 | _cleanup_free_ char *path = NULL; |
10bd3e2e | 2145 | const char *scope_path; |
efdb0237 | 2146 | CGroupController c; |
b4cccbc1 | 2147 | int r, all_unified; |
efdb0237 | 2148 | char *e; |
8e274523 LP |
2149 | |
2150 | assert(m); | |
2151 | ||
35d2e7ec | 2152 | /* 1. Determine hierarchy */ |
efdb0237 | 2153 | m->cgroup_root = mfree(m->cgroup_root); |
9444b1f2 | 2154 | r = cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER, 0, &m->cgroup_root); |
23bbb0de MS |
2155 | if (r < 0) |
2156 | return log_error_errno(r, "Cannot determine cgroup we are running in: %m"); | |
8e274523 | 2157 | |
efdb0237 LP |
2158 | /* Chop off the init scope, if we are already located in it */ |
2159 | e = endswith(m->cgroup_root, "/" SPECIAL_INIT_SCOPE); | |
0d8c31ff | 2160 | |
efdb0237 LP |
2161 | /* LEGACY: Also chop off the system slice if we are in |
2162 | * it. This is to support live upgrades from older systemd | |
2163 | * versions where PID 1 was moved there. Also see | |
2164 | * cg_get_root_path(). */ | |
463d0d15 | 2165 | if (!e && MANAGER_IS_SYSTEM(m)) { |
9444b1f2 | 2166 | e = endswith(m->cgroup_root, "/" SPECIAL_SYSTEM_SLICE); |
15c60e99 | 2167 | if (!e) |
efdb0237 | 2168 | e = endswith(m->cgroup_root, "/system"); /* even more legacy */ |
0baf24dd | 2169 | } |
efdb0237 LP |
2170 | if (e) |
2171 | *e = 0; | |
7ccfb64a | 2172 | |
7546145e LP |
2173 | /* And make sure to store away the root value without trailing slash, even for the root dir, so that we can |
2174 | * easily prepend it everywhere. */ | |
2175 | delete_trailing_chars(m->cgroup_root, "/"); | |
8e274523 | 2176 | |
35d2e7ec | 2177 | /* 2. Show data */ |
9444b1f2 | 2178 | r = cg_get_path(SYSTEMD_CGROUP_CONTROLLER, m->cgroup_root, NULL, &path); |
23bbb0de MS |
2179 | if (r < 0) |
2180 | return log_error_errno(r, "Cannot find cgroup mount point: %m"); | |
8e274523 | 2181 | |
415fc41c TH |
2182 | r = cg_unified_flush(); |
2183 | if (r < 0) | |
2184 | return log_error_errno(r, "Couldn't determine if we are running in the unified hierarchy: %m"); | |
5da38d07 | 2185 | |
b4cccbc1 | 2186 | all_unified = cg_all_unified(); |
d4c819ed ZJS |
2187 | if (all_unified < 0) |
2188 | return log_error_errno(all_unified, "Couldn't determine whether we are in all unified mode: %m"); | |
2189 | if (all_unified > 0) | |
efdb0237 | 2190 | log_debug("Unified cgroup hierarchy is located at %s.", path); |
b4cccbc1 | 2191 | else { |
c22800e4 | 2192 | r = cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER); |
b4cccbc1 LP |
2193 | if (r < 0) |
2194 | return log_error_errno(r, "Failed to determine whether systemd's own controller is in unified mode: %m"); | |
2195 | if (r > 0) | |
2196 | log_debug("Unified cgroup hierarchy is located at %s. Controllers are on legacy hierarchies.", path); | |
2197 | else | |
2198 | log_debug("Using cgroup controller " SYSTEMD_CGROUP_CONTROLLER_LEGACY ". File system hierarchy is at %s.", path); | |
2199 | } | |
efdb0237 | 2200 | |
09e24654 LP |
2201 | /* 3. Allocate cgroup empty defer event source */ |
2202 | m->cgroup_empty_event_source = sd_event_source_unref(m->cgroup_empty_event_source); | |
2203 | r = sd_event_add_defer(m->event, &m->cgroup_empty_event_source, on_cgroup_empty_event, m); | |
2204 | if (r < 0) | |
2205 | return log_error_errno(r, "Failed to create cgroup empty event source: %m"); | |
2206 | ||
2207 | r = sd_event_source_set_priority(m->cgroup_empty_event_source, SD_EVENT_PRIORITY_NORMAL-5); | |
2208 | if (r < 0) | |
2209 | return log_error_errno(r, "Failed to set priority of cgroup empty event source: %m"); | |
2210 | ||
2211 | r = sd_event_source_set_enabled(m->cgroup_empty_event_source, SD_EVENT_OFF); | |
2212 | if (r < 0) | |
2213 | return log_error_errno(r, "Failed to disable cgroup empty event source: %m"); | |
2214 | ||
2215 | (void) sd_event_source_set_description(m->cgroup_empty_event_source, "cgroup-empty"); | |
2216 | ||
2217 | /* 4. Install notifier inotify object, or agent */ | |
10bd3e2e | 2218 | if (cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER) > 0) { |
c6c18be3 | 2219 | |
09e24654 | 2220 | /* In the unified hierarchy we can get cgroup empty notifications via inotify. */ |
efdb0237 | 2221 | |
10bd3e2e LP |
2222 | m->cgroup_inotify_event_source = sd_event_source_unref(m->cgroup_inotify_event_source); |
2223 | safe_close(m->cgroup_inotify_fd); | |
efdb0237 | 2224 | |
10bd3e2e LP |
2225 | m->cgroup_inotify_fd = inotify_init1(IN_NONBLOCK|IN_CLOEXEC); |
2226 | if (m->cgroup_inotify_fd < 0) | |
2227 | return log_error_errno(errno, "Failed to create control group inotify object: %m"); | |
efdb0237 | 2228 | |
10bd3e2e LP |
2229 | r = sd_event_add_io(m->event, &m->cgroup_inotify_event_source, m->cgroup_inotify_fd, EPOLLIN, on_cgroup_inotify_event, m); |
2230 | if (r < 0) | |
2231 | return log_error_errno(r, "Failed to watch control group inotify object: %m"); | |
efdb0237 | 2232 | |
10bd3e2e LP |
2233 | /* Process cgroup empty notifications early, but after service notifications and SIGCHLD. Also |
2234 | * see handling of cgroup agent notifications, for the classic cgroup hierarchy support. */ | |
09e24654 | 2235 | r = sd_event_source_set_priority(m->cgroup_inotify_event_source, SD_EVENT_PRIORITY_NORMAL-4); |
10bd3e2e LP |
2236 | if (r < 0) |
2237 | return log_error_errno(r, "Failed to set priority of inotify event source: %m"); | |
efdb0237 | 2238 | |
10bd3e2e | 2239 | (void) sd_event_source_set_description(m->cgroup_inotify_event_source, "cgroup-inotify"); |
efdb0237 | 2240 | |
10bd3e2e | 2241 | } else if (MANAGER_IS_SYSTEM(m) && m->test_run_flags == 0) { |
efdb0237 | 2242 | |
10bd3e2e LP |
2243 | /* On the legacy hierarchy we only get notifications via cgroup agents. (Which isn't really reliable, |
2244 | * since it does not generate events when control groups with children run empty. */ | |
8e274523 | 2245 | |
10bd3e2e | 2246 | r = cg_install_release_agent(SYSTEMD_CGROUP_CONTROLLER, SYSTEMD_CGROUP_AGENT_PATH); |
23bbb0de | 2247 | if (r < 0) |
10bd3e2e LP |
2248 | log_warning_errno(r, "Failed to install release agent, ignoring: %m"); |
2249 | else if (r > 0) | |
2250 | log_debug("Installed release agent."); | |
2251 | else if (r == 0) | |
2252 | log_debug("Release agent already installed."); | |
2253 | } | |
efdb0237 | 2254 | |
09e24654 | 2255 | /* 5. Make sure we are in the special "init.scope" unit in the root slice. */ |
10bd3e2e LP |
2256 | scope_path = strjoina(m->cgroup_root, "/" SPECIAL_INIT_SCOPE); |
2257 | r = cg_create_and_attach(SYSTEMD_CGROUP_CONTROLLER, scope_path, 0); | |
aa77e234 MS |
2258 | if (r >= 0) { |
2259 | /* Also, move all other userspace processes remaining in the root cgroup into that scope. */ | |
2260 | r = cg_migrate(SYSTEMD_CGROUP_CONTROLLER, m->cgroup_root, SYSTEMD_CGROUP_CONTROLLER, scope_path, 0); | |
2261 | if (r < 0) | |
2262 | log_warning_errno(r, "Couldn't move remaining userspace processes, ignoring: %m"); | |
c6c18be3 | 2263 | |
aa77e234 MS |
2264 | /* 6. And pin it, so that it cannot be unmounted */ |
2265 | safe_close(m->pin_cgroupfs_fd); | |
2266 | m->pin_cgroupfs_fd = open(path, O_RDONLY|O_CLOEXEC|O_DIRECTORY|O_NOCTTY|O_NONBLOCK); | |
2267 | if (m->pin_cgroupfs_fd < 0) | |
2268 | return log_error_errno(errno, "Failed to open pin file: %m"); | |
0d8c31ff | 2269 | |
aa77e234 MS |
2270 | } else if (r < 0 && !m->test_run_flags) |
2271 | return log_error_errno(r, "Failed to create %s control group: %m", scope_path); | |
10bd3e2e | 2272 | |
09e24654 | 2273 | /* 7. Always enable hierarchical support if it exists... */ |
10bd3e2e LP |
2274 | if (!all_unified && m->test_run_flags == 0) |
2275 | (void) cg_set_attribute("memory", "/", "memory.use_hierarchy", "1"); | |
c6c18be3 | 2276 | |
09e24654 | 2277 | /* 8. Figure out which controllers are supported, and log about it */ |
efdb0237 LP |
2278 | r = cg_mask_supported(&m->cgroup_supported); |
2279 | if (r < 0) | |
2280 | return log_error_errno(r, "Failed to determine supported controllers: %m"); | |
efdb0237 | 2281 | for (c = 0; c < _CGROUP_CONTROLLER_MAX; c++) |
eee0a1e4 | 2282 | log_debug("Controller '%s' supported: %s", cgroup_controller_to_string(c), yes_no(m->cgroup_supported & CGROUP_CONTROLLER_TO_MASK(c))); |
9156e799 | 2283 | |
a32360f1 | 2284 | return 0; |
8e274523 LP |
2285 | } |
2286 | ||
c6c18be3 | 2287 | void manager_shutdown_cgroup(Manager *m, bool delete) { |
8e274523 LP |
2288 | assert(m); |
2289 | ||
9444b1f2 LP |
2290 | /* We can't really delete the group, since we are in it. But |
2291 | * let's trim it. */ | |
f6c63f6f | 2292 | if (delete && m->cgroup_root && m->test_run_flags != MANAGER_TEST_RUN_MINIMAL) |
efdb0237 LP |
2293 | (void) cg_trim(SYSTEMD_CGROUP_CONTROLLER, m->cgroup_root, false); |
2294 | ||
09e24654 LP |
2295 | m->cgroup_empty_event_source = sd_event_source_unref(m->cgroup_empty_event_source); |
2296 | ||
efdb0237 LP |
2297 | m->cgroup_inotify_wd_unit = hashmap_free(m->cgroup_inotify_wd_unit); |
2298 | ||
2299 | m->cgroup_inotify_event_source = sd_event_source_unref(m->cgroup_inotify_event_source); | |
2300 | m->cgroup_inotify_fd = safe_close(m->cgroup_inotify_fd); | |
8e274523 | 2301 | |
03e334a1 | 2302 | m->pin_cgroupfs_fd = safe_close(m->pin_cgroupfs_fd); |
c6c18be3 | 2303 | |
efdb0237 | 2304 | m->cgroup_root = mfree(m->cgroup_root); |
8e274523 LP |
2305 | } |
2306 | ||
4ad49000 | 2307 | Unit* manager_get_unit_by_cgroup(Manager *m, const char *cgroup) { |
acb14d31 | 2308 | char *p; |
4ad49000 | 2309 | Unit *u; |
acb14d31 LP |
2310 | |
2311 | assert(m); | |
2312 | assert(cgroup); | |
acb14d31 | 2313 | |
4ad49000 LP |
2314 | u = hashmap_get(m->cgroup_unit, cgroup); |
2315 | if (u) | |
2316 | return u; | |
acb14d31 | 2317 | |
8e70580b | 2318 | p = strdupa(cgroup); |
acb14d31 LP |
2319 | for (;;) { |
2320 | char *e; | |
2321 | ||
2322 | e = strrchr(p, '/'); | |
efdb0237 LP |
2323 | if (!e || e == p) |
2324 | return hashmap_get(m->cgroup_unit, SPECIAL_ROOT_SLICE); | |
acb14d31 LP |
2325 | |
2326 | *e = 0; | |
2327 | ||
4ad49000 LP |
2328 | u = hashmap_get(m->cgroup_unit, p); |
2329 | if (u) | |
2330 | return u; | |
acb14d31 LP |
2331 | } |
2332 | } | |
2333 | ||
b3ac818b | 2334 | Unit *manager_get_unit_by_pid_cgroup(Manager *m, pid_t pid) { |
4ad49000 | 2335 | _cleanup_free_ char *cgroup = NULL; |
8e274523 | 2336 | |
8c47c732 LP |
2337 | assert(m); |
2338 | ||
62a76913 | 2339 | if (!pid_is_valid(pid)) |
b3ac818b LP |
2340 | return NULL; |
2341 | ||
62a76913 | 2342 | if (cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER, pid, &cgroup) < 0) |
b3ac818b LP |
2343 | return NULL; |
2344 | ||
2345 | return manager_get_unit_by_cgroup(m, cgroup); | |
2346 | } | |
2347 | ||
2348 | Unit *manager_get_unit_by_pid(Manager *m, pid_t pid) { | |
62a76913 | 2349 | Unit *u, **array; |
b3ac818b LP |
2350 | |
2351 | assert(m); | |
2352 | ||
62a76913 LP |
2353 | /* Note that a process might be owned by multiple units, we return only one here, which is good enough for most |
2354 | * cases, though not strictly correct. We prefer the one reported by cgroup membership, as that's the most | |
2355 | * relevant one as children of the process will be assigned to that one, too, before all else. */ | |
2356 | ||
2357 | if (!pid_is_valid(pid)) | |
8c47c732 LP |
2358 | return NULL; |
2359 | ||
2ca9d979 | 2360 | if (pid == getpid_cached()) |
efdb0237 LP |
2361 | return hashmap_get(m->units, SPECIAL_INIT_SCOPE); |
2362 | ||
62a76913 | 2363 | u = manager_get_unit_by_pid_cgroup(m, pid); |
5fe8876b LP |
2364 | if (u) |
2365 | return u; | |
2366 | ||
62a76913 | 2367 | u = hashmap_get(m->watch_pids, PID_TO_PTR(pid)); |
5fe8876b LP |
2368 | if (u) |
2369 | return u; | |
2370 | ||
62a76913 LP |
2371 | array = hashmap_get(m->watch_pids, PID_TO_PTR(-pid)); |
2372 | if (array) | |
2373 | return array[0]; | |
2374 | ||
2375 | return NULL; | |
6dde1f33 | 2376 | } |
4fbf50b3 | 2377 | |
4ad49000 LP |
2378 | int manager_notify_cgroup_empty(Manager *m, const char *cgroup) { |
2379 | Unit *u; | |
4fbf50b3 | 2380 | |
4ad49000 LP |
2381 | assert(m); |
2382 | assert(cgroup); | |
4fbf50b3 | 2383 | |
09e24654 LP |
2384 | /* Called on the legacy hierarchy whenever we get an explicit cgroup notification from the cgroup agent process |
2385 | * or from the --system instance */ | |
2386 | ||
d8fdc620 LP |
2387 | log_debug("Got cgroup empty notification for: %s", cgroup); |
2388 | ||
4ad49000 | 2389 | u = manager_get_unit_by_cgroup(m, cgroup); |
5ad096b3 LP |
2390 | if (!u) |
2391 | return 0; | |
b56c28c3 | 2392 | |
09e24654 LP |
2393 | unit_add_to_cgroup_empty_queue(u); |
2394 | return 1; | |
5ad096b3 LP |
2395 | } |
2396 | ||
2397 | int unit_get_memory_current(Unit *u, uint64_t *ret) { | |
2398 | _cleanup_free_ char *v = NULL; | |
2399 | int r; | |
2400 | ||
2401 | assert(u); | |
2402 | assert(ret); | |
2403 | ||
2e4025c0 | 2404 | if (!UNIT_CGROUP_BOOL(u, memory_accounting)) |
cf3b4be1 LP |
2405 | return -ENODATA; |
2406 | ||
5ad096b3 LP |
2407 | if (!u->cgroup_path) |
2408 | return -ENODATA; | |
2409 | ||
1f73aa00 LP |
2410 | /* The root cgroup doesn't expose this information, let's get it from /proc instead */ |
2411 | if (unit_has_root_cgroup(u)) | |
2412 | return procfs_memory_get_current(ret); | |
2413 | ||
efdb0237 | 2414 | if ((u->cgroup_realized_mask & CGROUP_MASK_MEMORY) == 0) |
5ad096b3 LP |
2415 | return -ENODATA; |
2416 | ||
b4cccbc1 LP |
2417 | r = cg_all_unified(); |
2418 | if (r < 0) | |
2419 | return r; | |
2420 | if (r > 0) | |
efdb0237 | 2421 | r = cg_get_attribute("memory", u->cgroup_path, "memory.current", &v); |
b4cccbc1 LP |
2422 | else |
2423 | r = cg_get_attribute("memory", u->cgroup_path, "memory.usage_in_bytes", &v); | |
5ad096b3 LP |
2424 | if (r == -ENOENT) |
2425 | return -ENODATA; | |
2426 | if (r < 0) | |
2427 | return r; | |
2428 | ||
2429 | return safe_atou64(v, ret); | |
2430 | } | |
2431 | ||
03a7b521 LP |
2432 | int unit_get_tasks_current(Unit *u, uint64_t *ret) { |
2433 | _cleanup_free_ char *v = NULL; | |
2434 | int r; | |
2435 | ||
2436 | assert(u); | |
2437 | assert(ret); | |
2438 | ||
2e4025c0 | 2439 | if (!UNIT_CGROUP_BOOL(u, tasks_accounting)) |
cf3b4be1 LP |
2440 | return -ENODATA; |
2441 | ||
03a7b521 LP |
2442 | if (!u->cgroup_path) |
2443 | return -ENODATA; | |
2444 | ||
c36a69f4 LP |
2445 | /* The root cgroup doesn't expose this information, let's get it from /proc instead */ |
2446 | if (unit_has_root_cgroup(u)) | |
2447 | return procfs_tasks_get_current(ret); | |
2448 | ||
1f73aa00 LP |
2449 | if ((u->cgroup_realized_mask & CGROUP_MASK_PIDS) == 0) |
2450 | return -ENODATA; | |
2451 | ||
03a7b521 LP |
2452 | r = cg_get_attribute("pids", u->cgroup_path, "pids.current", &v); |
2453 | if (r == -ENOENT) | |
2454 | return -ENODATA; | |
2455 | if (r < 0) | |
2456 | return r; | |
2457 | ||
2458 | return safe_atou64(v, ret); | |
2459 | } | |
2460 | ||
5ad096b3 LP |
2461 | static int unit_get_cpu_usage_raw(Unit *u, nsec_t *ret) { |
2462 | _cleanup_free_ char *v = NULL; | |
2463 | uint64_t ns; | |
2464 | int r; | |
2465 | ||
2466 | assert(u); | |
2467 | assert(ret); | |
2468 | ||
2469 | if (!u->cgroup_path) | |
2470 | return -ENODATA; | |
2471 | ||
1f73aa00 LP |
2472 | /* The root cgroup doesn't expose this information, let's get it from /proc instead */ |
2473 | if (unit_has_root_cgroup(u)) | |
2474 | return procfs_cpu_get_usage(ret); | |
2475 | ||
b4cccbc1 LP |
2476 | r = cg_all_unified(); |
2477 | if (r < 0) | |
2478 | return r; | |
2479 | if (r > 0) { | |
66ebf6c0 TH |
2480 | _cleanup_free_ char *val = NULL; |
2481 | uint64_t us; | |
5ad096b3 | 2482 | |
66ebf6c0 TH |
2483 | if ((u->cgroup_realized_mask & CGROUP_MASK_CPU) == 0) |
2484 | return -ENODATA; | |
5ad096b3 | 2485 | |
b734a4ff | 2486 | r = cg_get_keyed_attribute("cpu", u->cgroup_path, "cpu.stat", STRV_MAKE("usage_usec"), &val); |
66ebf6c0 TH |
2487 | if (r < 0) |
2488 | return r; | |
b734a4ff LP |
2489 | if (IN_SET(r, -ENOENT, -ENXIO)) |
2490 | return -ENODATA; | |
66ebf6c0 TH |
2491 | |
2492 | r = safe_atou64(val, &us); | |
2493 | if (r < 0) | |
2494 | return r; | |
2495 | ||
2496 | ns = us * NSEC_PER_USEC; | |
2497 | } else { | |
2498 | if ((u->cgroup_realized_mask & CGROUP_MASK_CPUACCT) == 0) | |
2499 | return -ENODATA; | |
2500 | ||
2501 | r = cg_get_attribute("cpuacct", u->cgroup_path, "cpuacct.usage", &v); | |
2502 | if (r == -ENOENT) | |
2503 | return -ENODATA; | |
2504 | if (r < 0) | |
2505 | return r; | |
2506 | ||
2507 | r = safe_atou64(v, &ns); | |
2508 | if (r < 0) | |
2509 | return r; | |
2510 | } | |
5ad096b3 LP |
2511 | |
2512 | *ret = ns; | |
2513 | return 0; | |
2514 | } | |
2515 | ||
2516 | int unit_get_cpu_usage(Unit *u, nsec_t *ret) { | |
2517 | nsec_t ns; | |
2518 | int r; | |
2519 | ||
fe700f46 LP |
2520 | assert(u); |
2521 | ||
2522 | /* Retrieve the current CPU usage counter. This will subtract the CPU counter taken when the unit was | |
2523 | * started. If the cgroup has been removed already, returns the last cached value. To cache the value, simply | |
2524 | * call this function with a NULL return value. */ | |
2525 | ||
2e4025c0 | 2526 | if (!UNIT_CGROUP_BOOL(u, cpu_accounting)) |
cf3b4be1 LP |
2527 | return -ENODATA; |
2528 | ||
5ad096b3 | 2529 | r = unit_get_cpu_usage_raw(u, &ns); |
fe700f46 LP |
2530 | if (r == -ENODATA && u->cpu_usage_last != NSEC_INFINITY) { |
2531 | /* If we can't get the CPU usage anymore (because the cgroup was already removed, for example), use our | |
2532 | * cached value. */ | |
2533 | ||
2534 | if (ret) | |
2535 | *ret = u->cpu_usage_last; | |
2536 | return 0; | |
2537 | } | |
5ad096b3 LP |
2538 | if (r < 0) |
2539 | return r; | |
2540 | ||
66ebf6c0 TH |
2541 | if (ns > u->cpu_usage_base) |
2542 | ns -= u->cpu_usage_base; | |
5ad096b3 LP |
2543 | else |
2544 | ns = 0; | |
2545 | ||
fe700f46 LP |
2546 | u->cpu_usage_last = ns; |
2547 | if (ret) | |
2548 | *ret = ns; | |
2549 | ||
5ad096b3 LP |
2550 | return 0; |
2551 | } | |
2552 | ||
906c06f6 DM |
2553 | int unit_get_ip_accounting( |
2554 | Unit *u, | |
2555 | CGroupIPAccountingMetric metric, | |
2556 | uint64_t *ret) { | |
2557 | ||
6b659ed8 | 2558 | uint64_t value; |
906c06f6 DM |
2559 | int fd, r; |
2560 | ||
2561 | assert(u); | |
2562 | assert(metric >= 0); | |
2563 | assert(metric < _CGROUP_IP_ACCOUNTING_METRIC_MAX); | |
2564 | assert(ret); | |
2565 | ||
2e4025c0 | 2566 | if (!UNIT_CGROUP_BOOL(u, ip_accounting)) |
cf3b4be1 LP |
2567 | return -ENODATA; |
2568 | ||
906c06f6 DM |
2569 | fd = IN_SET(metric, CGROUP_IP_INGRESS_BYTES, CGROUP_IP_INGRESS_PACKETS) ? |
2570 | u->ip_accounting_ingress_map_fd : | |
2571 | u->ip_accounting_egress_map_fd; | |
906c06f6 DM |
2572 | if (fd < 0) |
2573 | return -ENODATA; | |
2574 | ||
2575 | if (IN_SET(metric, CGROUP_IP_INGRESS_BYTES, CGROUP_IP_EGRESS_BYTES)) | |
6b659ed8 | 2576 | r = bpf_firewall_read_accounting(fd, &value, NULL); |
906c06f6 | 2577 | else |
6b659ed8 LP |
2578 | r = bpf_firewall_read_accounting(fd, NULL, &value); |
2579 | if (r < 0) | |
2580 | return r; | |
2581 | ||
2582 | /* Add in additional metrics from a previous runtime. Note that when reexecing/reloading the daemon we compile | |
2583 | * all BPF programs and maps anew, but serialize the old counters. When deserializing we store them in the | |
2584 | * ip_accounting_extra[] field, and add them in here transparently. */ | |
2585 | ||
2586 | *ret = value + u->ip_accounting_extra[metric]; | |
906c06f6 DM |
2587 | |
2588 | return r; | |
2589 | } | |
2590 | ||
2591 | int unit_reset_cpu_accounting(Unit *u) { | |
5ad096b3 LP |
2592 | nsec_t ns; |
2593 | int r; | |
2594 | ||
2595 | assert(u); | |
2596 | ||
fe700f46 LP |
2597 | u->cpu_usage_last = NSEC_INFINITY; |
2598 | ||
5ad096b3 LP |
2599 | r = unit_get_cpu_usage_raw(u, &ns); |
2600 | if (r < 0) { | |
66ebf6c0 | 2601 | u->cpu_usage_base = 0; |
5ad096b3 | 2602 | return r; |
b56c28c3 | 2603 | } |
2633eb83 | 2604 | |
66ebf6c0 | 2605 | u->cpu_usage_base = ns; |
4ad49000 | 2606 | return 0; |
4fbf50b3 LP |
2607 | } |
2608 | ||
906c06f6 DM |
2609 | int unit_reset_ip_accounting(Unit *u) { |
2610 | int r = 0, q = 0; | |
2611 | ||
2612 | assert(u); | |
2613 | ||
2614 | if (u->ip_accounting_ingress_map_fd >= 0) | |
2615 | r = bpf_firewall_reset_accounting(u->ip_accounting_ingress_map_fd); | |
2616 | ||
2617 | if (u->ip_accounting_egress_map_fd >= 0) | |
2618 | q = bpf_firewall_reset_accounting(u->ip_accounting_egress_map_fd); | |
2619 | ||
6b659ed8 LP |
2620 | zero(u->ip_accounting_extra); |
2621 | ||
906c06f6 DM |
2622 | return r < 0 ? r : q; |
2623 | } | |
2624 | ||
e7ab4d1a LP |
2625 | void unit_invalidate_cgroup(Unit *u, CGroupMask m) { |
2626 | assert(u); | |
2627 | ||
2628 | if (!UNIT_HAS_CGROUP_CONTEXT(u)) | |
2629 | return; | |
2630 | ||
2631 | if (m == 0) | |
2632 | return; | |
2633 | ||
538b4852 TH |
2634 | /* always invalidate compat pairs together */ |
2635 | if (m & (CGROUP_MASK_IO | CGROUP_MASK_BLKIO)) | |
2636 | m |= CGROUP_MASK_IO | CGROUP_MASK_BLKIO; | |
2637 | ||
7cce4fb7 LP |
2638 | if (m & (CGROUP_MASK_CPU | CGROUP_MASK_CPUACCT)) |
2639 | m |= CGROUP_MASK_CPU | CGROUP_MASK_CPUACCT; | |
2640 | ||
60c728ad | 2641 | if ((u->cgroup_realized_mask & m) == 0) /* NOP? */ |
e7ab4d1a LP |
2642 | return; |
2643 | ||
2644 | u->cgroup_realized_mask &= ~m; | |
91a6073e | 2645 | unit_add_to_cgroup_realize_queue(u); |
e7ab4d1a LP |
2646 | } |
2647 | ||
906c06f6 DM |
2648 | void unit_invalidate_cgroup_bpf(Unit *u) { |
2649 | assert(u); | |
2650 | ||
2651 | if (!UNIT_HAS_CGROUP_CONTEXT(u)) | |
2652 | return; | |
2653 | ||
60c728ad | 2654 | if (u->cgroup_bpf_state == UNIT_CGROUP_BPF_INVALIDATED) /* NOP? */ |
906c06f6 DM |
2655 | return; |
2656 | ||
2657 | u->cgroup_bpf_state = UNIT_CGROUP_BPF_INVALIDATED; | |
91a6073e | 2658 | unit_add_to_cgroup_realize_queue(u); |
906c06f6 DM |
2659 | |
2660 | /* If we are a slice unit, we also need to put compile a new BPF program for all our children, as the IP access | |
2661 | * list of our children includes our own. */ | |
2662 | if (u->type == UNIT_SLICE) { | |
2663 | Unit *member; | |
2664 | Iterator i; | |
eef85c4a | 2665 | void *v; |
906c06f6 | 2666 | |
eef85c4a | 2667 | HASHMAP_FOREACH_KEY(v, member, u->dependencies[UNIT_BEFORE], i) { |
906c06f6 DM |
2668 | if (member == u) |
2669 | continue; | |
2670 | ||
2671 | if (UNIT_DEREF(member->slice) != u) | |
2672 | continue; | |
2673 | ||
2674 | unit_invalidate_cgroup_bpf(member); | |
2675 | } | |
2676 | } | |
2677 | } | |
2678 | ||
1d9cc876 LP |
2679 | bool unit_cgroup_delegate(Unit *u) { |
2680 | CGroupContext *c; | |
2681 | ||
2682 | assert(u); | |
2683 | ||
2684 | if (!UNIT_VTABLE(u)->can_delegate) | |
2685 | return false; | |
2686 | ||
2687 | c = unit_get_cgroup_context(u); | |
2688 | if (!c) | |
2689 | return false; | |
2690 | ||
2691 | return c->delegate; | |
2692 | } | |
2693 | ||
e7ab4d1a LP |
2694 | void manager_invalidate_startup_units(Manager *m) { |
2695 | Iterator i; | |
2696 | Unit *u; | |
2697 | ||
2698 | assert(m); | |
2699 | ||
2700 | SET_FOREACH(u, m->startup_units, i) | |
13c31542 | 2701 | unit_invalidate_cgroup(u, CGROUP_MASK_CPU|CGROUP_MASK_IO|CGROUP_MASK_BLKIO); |
e7ab4d1a LP |
2702 | } |
2703 | ||
4ad49000 LP |
2704 | static const char* const cgroup_device_policy_table[_CGROUP_DEVICE_POLICY_MAX] = { |
2705 | [CGROUP_AUTO] = "auto", | |
2706 | [CGROUP_CLOSED] = "closed", | |
2707 | [CGROUP_STRICT] = "strict", | |
2708 | }; | |
4fbf50b3 | 2709 | |
4ad49000 | 2710 | DEFINE_STRING_TABLE_LOOKUP(cgroup_device_policy, CGroupDevicePolicy); |