]>
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 | systemd is free software; you can redistribute it and/or modify it | |
5430f7f2 LP |
8 | under the terms of the GNU Lesser General Public License as published by |
9 | the Free Software Foundation; either version 2.1 of the License, or | |
8e274523 LP |
10 | (at your option) any later version. |
11 | ||
12 | systemd is distributed in the hope that it will be useful, but | |
13 | WITHOUT ANY WARRANTY; without even the implied warranty of | |
14 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
5430f7f2 | 15 | Lesser General Public License for more details. |
8e274523 | 16 | |
5430f7f2 | 17 | You should have received a copy of the GNU Lesser General Public License |
8e274523 LP |
18 | along with systemd; If not, see <http://www.gnu.org/licenses/>. |
19 | ***/ | |
20 | ||
c6c18be3 | 21 | #include <fcntl.h> |
e41969e3 | 22 | #include <fnmatch.h> |
8c6db833 | 23 | |
b5efdb8a | 24 | #include "alloc-util.h" |
906c06f6 | 25 | #include "bpf-firewall.h" |
03a7b521 | 26 | #include "cgroup-util.h" |
3ffd4af2 LP |
27 | #include "cgroup.h" |
28 | #include "fd-util.h" | |
0d39fa9c | 29 | #include "fileio.h" |
77601719 | 30 | #include "fs-util.h" |
6bedfcbb | 31 | #include "parse-util.h" |
9eb977db | 32 | #include "path-util.h" |
03a7b521 | 33 | #include "process-util.h" |
9444b1f2 | 34 | #include "special.h" |
906c06f6 | 35 | #include "stdio-util.h" |
8b43440b | 36 | #include "string-table.h" |
07630cea | 37 | #include "string-util.h" |
8e274523 | 38 | |
9a054909 LP |
39 | #define CGROUP_CPU_QUOTA_PERIOD_USEC ((usec_t) 100 * USEC_PER_MSEC) |
40 | ||
2b40998d | 41 | static void cgroup_compat_warn(void) { |
128fadc9 TH |
42 | static bool cgroup_compat_warned = false; |
43 | ||
44 | if (cgroup_compat_warned) | |
45 | return; | |
46 | ||
47 | log_warning("cgroup compatibility translation between legacy and unified hierarchy settings activated. See cgroup-compat debug messages for details."); | |
48 | cgroup_compat_warned = true; | |
49 | } | |
50 | ||
51 | #define log_cgroup_compat(unit, fmt, ...) do { \ | |
52 | cgroup_compat_warn(); \ | |
53 | log_unit_debug(unit, "cgroup-compat: " fmt, ##__VA_ARGS__); \ | |
2b40998d | 54 | } while (false) |
128fadc9 | 55 | |
4ad49000 LP |
56 | void cgroup_context_init(CGroupContext *c) { |
57 | assert(c); | |
58 | ||
59 | /* Initialize everything to the kernel defaults, assuming the | |
60 | * structure is preinitialized to 0 */ | |
61 | ||
66ebf6c0 TH |
62 | c->cpu_weight = CGROUP_WEIGHT_INVALID; |
63 | c->startup_cpu_weight = CGROUP_WEIGHT_INVALID; | |
64 | c->cpu_quota_per_sec_usec = USEC_INFINITY; | |
65 | ||
d53d9474 LP |
66 | c->cpu_shares = CGROUP_CPU_SHARES_INVALID; |
67 | c->startup_cpu_shares = CGROUP_CPU_SHARES_INVALID; | |
d53d9474 | 68 | |
da4d897e TH |
69 | c->memory_high = CGROUP_LIMIT_MAX; |
70 | c->memory_max = CGROUP_LIMIT_MAX; | |
96e131ea | 71 | c->memory_swap_max = CGROUP_LIMIT_MAX; |
da4d897e TH |
72 | |
73 | c->memory_limit = CGROUP_LIMIT_MAX; | |
b2f8b02e | 74 | |
13c31542 TH |
75 | c->io_weight = CGROUP_WEIGHT_INVALID; |
76 | c->startup_io_weight = CGROUP_WEIGHT_INVALID; | |
77 | ||
d53d9474 LP |
78 | c->blockio_weight = CGROUP_BLKIO_WEIGHT_INVALID; |
79 | c->startup_blockio_weight = CGROUP_BLKIO_WEIGHT_INVALID; | |
80 | ||
81 | c->tasks_max = (uint64_t) -1; | |
4ad49000 | 82 | } |
8e274523 | 83 | |
4ad49000 LP |
84 | void cgroup_context_free_device_allow(CGroupContext *c, CGroupDeviceAllow *a) { |
85 | assert(c); | |
86 | assert(a); | |
87 | ||
71fda00f | 88 | LIST_REMOVE(device_allow, c->device_allow, a); |
4ad49000 LP |
89 | free(a->path); |
90 | free(a); | |
91 | } | |
92 | ||
13c31542 TH |
93 | void cgroup_context_free_io_device_weight(CGroupContext *c, CGroupIODeviceWeight *w) { |
94 | assert(c); | |
95 | assert(w); | |
96 | ||
97 | LIST_REMOVE(device_weights, c->io_device_weights, w); | |
98 | free(w->path); | |
99 | free(w); | |
100 | } | |
101 | ||
102 | void cgroup_context_free_io_device_limit(CGroupContext *c, CGroupIODeviceLimit *l) { | |
103 | assert(c); | |
104 | assert(l); | |
105 | ||
106 | LIST_REMOVE(device_limits, c->io_device_limits, l); | |
107 | free(l->path); | |
108 | free(l); | |
109 | } | |
110 | ||
4ad49000 LP |
111 | void cgroup_context_free_blockio_device_weight(CGroupContext *c, CGroupBlockIODeviceWeight *w) { |
112 | assert(c); | |
113 | assert(w); | |
114 | ||
71fda00f | 115 | LIST_REMOVE(device_weights, c->blockio_device_weights, w); |
4ad49000 LP |
116 | free(w->path); |
117 | free(w); | |
118 | } | |
119 | ||
120 | void cgroup_context_free_blockio_device_bandwidth(CGroupContext *c, CGroupBlockIODeviceBandwidth *b) { | |
121 | assert(c); | |
8e274523 | 122 | assert(b); |
8e274523 | 123 | |
71fda00f | 124 | LIST_REMOVE(device_bandwidths, c->blockio_device_bandwidths, b); |
4ad49000 LP |
125 | free(b->path); |
126 | free(b); | |
127 | } | |
128 | ||
129 | void cgroup_context_done(CGroupContext *c) { | |
130 | assert(c); | |
131 | ||
13c31542 TH |
132 | while (c->io_device_weights) |
133 | cgroup_context_free_io_device_weight(c, c->io_device_weights); | |
134 | ||
135 | while (c->io_device_limits) | |
136 | cgroup_context_free_io_device_limit(c, c->io_device_limits); | |
137 | ||
4ad49000 LP |
138 | while (c->blockio_device_weights) |
139 | cgroup_context_free_blockio_device_weight(c, c->blockio_device_weights); | |
140 | ||
141 | while (c->blockio_device_bandwidths) | |
142 | cgroup_context_free_blockio_device_bandwidth(c, c->blockio_device_bandwidths); | |
143 | ||
144 | while (c->device_allow) | |
145 | cgroup_context_free_device_allow(c, c->device_allow); | |
6a48d82f DM |
146 | |
147 | c->ip_address_allow = ip_address_access_free_all(c->ip_address_allow); | |
148 | c->ip_address_deny = ip_address_access_free_all(c->ip_address_deny); | |
4ad49000 LP |
149 | } |
150 | ||
151 | void cgroup_context_dump(CGroupContext *c, FILE* f, const char *prefix) { | |
13c31542 TH |
152 | CGroupIODeviceLimit *il; |
153 | CGroupIODeviceWeight *iw; | |
4ad49000 LP |
154 | CGroupBlockIODeviceBandwidth *b; |
155 | CGroupBlockIODeviceWeight *w; | |
156 | CGroupDeviceAllow *a; | |
c21c9906 | 157 | IPAddressAccessItem *iaai; |
9a054909 | 158 | char u[FORMAT_TIMESPAN_MAX]; |
4ad49000 LP |
159 | |
160 | assert(c); | |
161 | assert(f); | |
162 | ||
163 | prefix = strempty(prefix); | |
164 | ||
165 | fprintf(f, | |
166 | "%sCPUAccounting=%s\n" | |
13c31542 | 167 | "%sIOAccounting=%s\n" |
4ad49000 LP |
168 | "%sBlockIOAccounting=%s\n" |
169 | "%sMemoryAccounting=%s\n" | |
d53d9474 | 170 | "%sTasksAccounting=%s\n" |
c21c9906 | 171 | "%sIPAccounting=%s\n" |
66ebf6c0 TH |
172 | "%sCPUWeight=%" PRIu64 "\n" |
173 | "%sStartupCPUWeight=%" PRIu64 "\n" | |
d53d9474 LP |
174 | "%sCPUShares=%" PRIu64 "\n" |
175 | "%sStartupCPUShares=%" PRIu64 "\n" | |
b2f8b02e | 176 | "%sCPUQuotaPerSecSec=%s\n" |
13c31542 TH |
177 | "%sIOWeight=%" PRIu64 "\n" |
178 | "%sStartupIOWeight=%" PRIu64 "\n" | |
d53d9474 LP |
179 | "%sBlockIOWeight=%" PRIu64 "\n" |
180 | "%sStartupBlockIOWeight=%" PRIu64 "\n" | |
da4d897e TH |
181 | "%sMemoryLow=%" PRIu64 "\n" |
182 | "%sMemoryHigh=%" PRIu64 "\n" | |
183 | "%sMemoryMax=%" PRIu64 "\n" | |
96e131ea | 184 | "%sMemorySwapMax=%" PRIu64 "\n" |
4ad49000 | 185 | "%sMemoryLimit=%" PRIu64 "\n" |
03a7b521 | 186 | "%sTasksMax=%" PRIu64 "\n" |
a931ad47 LP |
187 | "%sDevicePolicy=%s\n" |
188 | "%sDelegate=%s\n", | |
4ad49000 | 189 | prefix, yes_no(c->cpu_accounting), |
13c31542 | 190 | prefix, yes_no(c->io_accounting), |
4ad49000 LP |
191 | prefix, yes_no(c->blockio_accounting), |
192 | prefix, yes_no(c->memory_accounting), | |
d53d9474 | 193 | prefix, yes_no(c->tasks_accounting), |
c21c9906 | 194 | prefix, yes_no(c->ip_accounting), |
66ebf6c0 TH |
195 | prefix, c->cpu_weight, |
196 | prefix, c->startup_cpu_weight, | |
4ad49000 | 197 | prefix, c->cpu_shares, |
95ae05c0 | 198 | prefix, c->startup_cpu_shares, |
b1d6dcf5 | 199 | prefix, format_timespan(u, sizeof(u), c->cpu_quota_per_sec_usec, 1), |
13c31542 TH |
200 | prefix, c->io_weight, |
201 | prefix, c->startup_io_weight, | |
4ad49000 | 202 | prefix, c->blockio_weight, |
95ae05c0 | 203 | prefix, c->startup_blockio_weight, |
da4d897e TH |
204 | prefix, c->memory_low, |
205 | prefix, c->memory_high, | |
206 | prefix, c->memory_max, | |
96e131ea | 207 | prefix, c->memory_swap_max, |
4ad49000 | 208 | prefix, c->memory_limit, |
03a7b521 | 209 | prefix, c->tasks_max, |
a931ad47 LP |
210 | prefix, cgroup_device_policy_to_string(c->device_policy), |
211 | prefix, yes_no(c->delegate)); | |
4ad49000 | 212 | |
02638280 LP |
213 | if (c->delegate) { |
214 | _cleanup_free_ char *t = NULL; | |
215 | ||
216 | (void) cg_mask_to_string(c->delegate_controllers, &t); | |
217 | ||
47a78d41 | 218 | fprintf(f, "%sDelegateControllers=%s\n", |
02638280 LP |
219 | prefix, |
220 | strempty(t)); | |
221 | } | |
222 | ||
4ad49000 LP |
223 | LIST_FOREACH(device_allow, a, c->device_allow) |
224 | fprintf(f, | |
225 | "%sDeviceAllow=%s %s%s%s\n", | |
226 | prefix, | |
227 | a->path, | |
228 | a->r ? "r" : "", a->w ? "w" : "", a->m ? "m" : ""); | |
229 | ||
13c31542 TH |
230 | LIST_FOREACH(device_weights, iw, c->io_device_weights) |
231 | fprintf(f, | |
232 | "%sIODeviceWeight=%s %" PRIu64, | |
233 | prefix, | |
234 | iw->path, | |
235 | iw->weight); | |
236 | ||
237 | LIST_FOREACH(device_limits, il, c->io_device_limits) { | |
238 | char buf[FORMAT_BYTES_MAX]; | |
9be57249 TH |
239 | CGroupIOLimitType type; |
240 | ||
241 | for (type = 0; type < _CGROUP_IO_LIMIT_TYPE_MAX; type++) | |
242 | if (il->limits[type] != cgroup_io_limit_defaults[type]) | |
243 | fprintf(f, | |
244 | "%s%s=%s %s\n", | |
245 | prefix, | |
246 | cgroup_io_limit_type_to_string(type), | |
247 | il->path, | |
248 | format_bytes(buf, sizeof(buf), il->limits[type])); | |
13c31542 TH |
249 | } |
250 | ||
4ad49000 LP |
251 | LIST_FOREACH(device_weights, w, c->blockio_device_weights) |
252 | fprintf(f, | |
d53d9474 | 253 | "%sBlockIODeviceWeight=%s %" PRIu64, |
4ad49000 LP |
254 | prefix, |
255 | w->path, | |
256 | w->weight); | |
257 | ||
258 | LIST_FOREACH(device_bandwidths, b, c->blockio_device_bandwidths) { | |
259 | char buf[FORMAT_BYTES_MAX]; | |
260 | ||
979d0311 TH |
261 | if (b->rbps != CGROUP_LIMIT_MAX) |
262 | fprintf(f, | |
263 | "%sBlockIOReadBandwidth=%s %s\n", | |
264 | prefix, | |
265 | b->path, | |
266 | format_bytes(buf, sizeof(buf), b->rbps)); | |
267 | if (b->wbps != CGROUP_LIMIT_MAX) | |
268 | fprintf(f, | |
269 | "%sBlockIOWriteBandwidth=%s %s\n", | |
270 | prefix, | |
271 | b->path, | |
272 | format_bytes(buf, sizeof(buf), b->wbps)); | |
4ad49000 | 273 | } |
c21c9906 LP |
274 | |
275 | LIST_FOREACH(items, iaai, c->ip_address_allow) { | |
276 | _cleanup_free_ char *k = NULL; | |
277 | ||
278 | (void) in_addr_to_string(iaai->family, &iaai->address, &k); | |
279 | fprintf(f, "%sIPAddressAllow=%s/%u\n", prefix, strnull(k), iaai->prefixlen); | |
280 | } | |
281 | ||
282 | LIST_FOREACH(items, iaai, c->ip_address_deny) { | |
283 | _cleanup_free_ char *k = NULL; | |
284 | ||
285 | (void) in_addr_to_string(iaai->family, &iaai->address, &k); | |
286 | fprintf(f, "%sIPAddressDeny=%s/%u\n", prefix, strnull(k), iaai->prefixlen); | |
287 | } | |
4ad49000 LP |
288 | } |
289 | ||
13c31542 | 290 | static int lookup_block_device(const char *p, dev_t *dev) { |
4ad49000 LP |
291 | struct stat st; |
292 | int r; | |
293 | ||
294 | assert(p); | |
295 | assert(dev); | |
296 | ||
297 | r = stat(p, &st); | |
4a62c710 MS |
298 | if (r < 0) |
299 | return log_warning_errno(errno, "Couldn't stat device %s: %m", p); | |
8e274523 | 300 | |
4ad49000 LP |
301 | if (S_ISBLK(st.st_mode)) |
302 | *dev = st.st_rdev; | |
303 | else if (major(st.st_dev) != 0) { | |
304 | /* If this is not a device node then find the block | |
305 | * device this file is stored on */ | |
306 | *dev = st.st_dev; | |
307 | ||
308 | /* If this is a partition, try to get the originating | |
309 | * block device */ | |
310 | block_get_whole_disk(*dev, dev); | |
311 | } else { | |
312 | log_warning("%s is not a block device and file system block device cannot be determined or is not local.", p); | |
313 | return -ENODEV; | |
314 | } | |
8e274523 | 315 | |
8e274523 | 316 | return 0; |
8e274523 LP |
317 | } |
318 | ||
4ad49000 LP |
319 | static int whitelist_device(const char *path, const char *node, const char *acc) { |
320 | char buf[2+DECIMAL_STR_MAX(dev_t)*2+2+4]; | |
321 | struct stat st; | |
b200489b | 322 | bool ignore_notfound; |
8c6db833 | 323 | int r; |
8e274523 | 324 | |
4ad49000 LP |
325 | assert(path); |
326 | assert(acc); | |
8e274523 | 327 | |
b200489b DR |
328 | if (node[0] == '-') { |
329 | /* Non-existent paths starting with "-" must be silently ignored */ | |
330 | node++; | |
331 | ignore_notfound = true; | |
332 | } else | |
333 | ignore_notfound = false; | |
334 | ||
4ad49000 | 335 | if (stat(node, &st) < 0) { |
b200489b | 336 | if (errno == ENOENT && ignore_notfound) |
e7330dfe DP |
337 | return 0; |
338 | ||
339 | return log_warning_errno(errno, "Couldn't stat device %s: %m", node); | |
4ad49000 LP |
340 | } |
341 | ||
342 | if (!S_ISCHR(st.st_mode) && !S_ISBLK(st.st_mode)) { | |
343 | log_warning("%s is not a device.", node); | |
344 | return -ENODEV; | |
345 | } | |
346 | ||
347 | sprintf(buf, | |
348 | "%c %u:%u %s", | |
349 | S_ISCHR(st.st_mode) ? 'c' : 'b', | |
350 | major(st.st_rdev), minor(st.st_rdev), | |
351 | acc); | |
352 | ||
353 | r = cg_set_attribute("devices", path, "devices.allow", buf); | |
1aeab12b | 354 | if (r < 0) |
077ba06e | 355 | log_full_errno(IN_SET(r, -ENOENT, -EROFS, -EINVAL, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, |
714e2e1d | 356 | "Failed to set devices.allow on %s: %m", path); |
4ad49000 LP |
357 | |
358 | return r; | |
8e274523 LP |
359 | } |
360 | ||
90060676 LP |
361 | static int whitelist_major(const char *path, const char *name, char type, const char *acc) { |
362 | _cleanup_fclose_ FILE *f = NULL; | |
363 | char line[LINE_MAX]; | |
364 | bool good = false; | |
365 | int r; | |
366 | ||
367 | assert(path); | |
368 | assert(acc); | |
4c701096 | 369 | assert(IN_SET(type, 'b', 'c')); |
90060676 LP |
370 | |
371 | f = fopen("/proc/devices", "re"); | |
4a62c710 MS |
372 | if (!f) |
373 | return log_warning_errno(errno, "Cannot open /proc/devices to resolve %s (%c): %m", name, type); | |
90060676 LP |
374 | |
375 | FOREACH_LINE(line, f, goto fail) { | |
376 | char buf[2+DECIMAL_STR_MAX(unsigned)+3+4], *p, *w; | |
377 | unsigned maj; | |
378 | ||
379 | truncate_nl(line); | |
380 | ||
381 | if (type == 'c' && streq(line, "Character devices:")) { | |
382 | good = true; | |
383 | continue; | |
384 | } | |
385 | ||
386 | if (type == 'b' && streq(line, "Block devices:")) { | |
387 | good = true; | |
388 | continue; | |
389 | } | |
390 | ||
391 | if (isempty(line)) { | |
392 | good = false; | |
393 | continue; | |
394 | } | |
395 | ||
396 | if (!good) | |
397 | continue; | |
398 | ||
399 | p = strstrip(line); | |
400 | ||
401 | w = strpbrk(p, WHITESPACE); | |
402 | if (!w) | |
403 | continue; | |
404 | *w = 0; | |
405 | ||
406 | r = safe_atou(p, &maj); | |
407 | if (r < 0) | |
408 | continue; | |
409 | if (maj <= 0) | |
410 | continue; | |
411 | ||
412 | w++; | |
413 | w += strspn(w, WHITESPACE); | |
e41969e3 LP |
414 | |
415 | if (fnmatch(name, w, 0) != 0) | |
90060676 LP |
416 | continue; |
417 | ||
418 | sprintf(buf, | |
419 | "%c %u:* %s", | |
420 | type, | |
421 | maj, | |
422 | acc); | |
423 | ||
424 | r = cg_set_attribute("devices", path, "devices.allow", buf); | |
1aeab12b | 425 | if (r < 0) |
077ba06e | 426 | log_full_errno(IN_SET(r, -ENOENT, -EROFS, -EINVAL, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, |
714e2e1d | 427 | "Failed to set devices.allow on %s: %m", path); |
90060676 LP |
428 | } |
429 | ||
430 | return 0; | |
431 | ||
432 | fail: | |
25f027c5 | 433 | return log_warning_errno(errno, "Failed to read /proc/devices: %m"); |
90060676 LP |
434 | } |
435 | ||
66ebf6c0 TH |
436 | static bool cgroup_context_has_cpu_weight(CGroupContext *c) { |
437 | return c->cpu_weight != CGROUP_WEIGHT_INVALID || | |
438 | c->startup_cpu_weight != CGROUP_WEIGHT_INVALID; | |
439 | } | |
440 | ||
441 | static bool cgroup_context_has_cpu_shares(CGroupContext *c) { | |
442 | return c->cpu_shares != CGROUP_CPU_SHARES_INVALID || | |
443 | c->startup_cpu_shares != CGROUP_CPU_SHARES_INVALID; | |
444 | } | |
445 | ||
446 | static uint64_t cgroup_context_cpu_weight(CGroupContext *c, ManagerState state) { | |
447 | if (IN_SET(state, MANAGER_STARTING, MANAGER_INITIALIZING) && | |
448 | c->startup_cpu_weight != CGROUP_WEIGHT_INVALID) | |
449 | return c->startup_cpu_weight; | |
450 | else if (c->cpu_weight != CGROUP_WEIGHT_INVALID) | |
451 | return c->cpu_weight; | |
452 | else | |
453 | return CGROUP_WEIGHT_DEFAULT; | |
454 | } | |
455 | ||
456 | static uint64_t cgroup_context_cpu_shares(CGroupContext *c, ManagerState state) { | |
457 | if (IN_SET(state, MANAGER_STARTING, MANAGER_INITIALIZING) && | |
458 | c->startup_cpu_shares != CGROUP_CPU_SHARES_INVALID) | |
459 | return c->startup_cpu_shares; | |
460 | else if (c->cpu_shares != CGROUP_CPU_SHARES_INVALID) | |
461 | return c->cpu_shares; | |
462 | else | |
463 | return CGROUP_CPU_SHARES_DEFAULT; | |
464 | } | |
465 | ||
466 | static void cgroup_apply_unified_cpu_config(Unit *u, uint64_t weight, uint64_t quota) { | |
467 | char buf[MAX(DECIMAL_STR_MAX(uint64_t) + 1, (DECIMAL_STR_MAX(usec_t) + 1) * 2)]; | |
468 | int r; | |
469 | ||
470 | xsprintf(buf, "%" PRIu64 "\n", weight); | |
471 | r = cg_set_attribute("cpu", u->cgroup_path, "cpu.weight", buf); | |
472 | if (r < 0) | |
473 | log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, | |
474 | "Failed to set cpu.weight: %m"); | |
475 | ||
476 | if (quota != USEC_INFINITY) | |
477 | xsprintf(buf, USEC_FMT " " USEC_FMT "\n", | |
478 | quota * CGROUP_CPU_QUOTA_PERIOD_USEC / USEC_PER_SEC, CGROUP_CPU_QUOTA_PERIOD_USEC); | |
479 | else | |
480 | xsprintf(buf, "max " USEC_FMT "\n", CGROUP_CPU_QUOTA_PERIOD_USEC); | |
481 | ||
482 | r = cg_set_attribute("cpu", u->cgroup_path, "cpu.max", buf); | |
483 | ||
484 | if (r < 0) | |
485 | log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, | |
486 | "Failed to set cpu.max: %m"); | |
487 | } | |
488 | ||
489 | static void cgroup_apply_legacy_cpu_config(Unit *u, uint64_t shares, uint64_t quota) { | |
490 | char buf[MAX(DECIMAL_STR_MAX(uint64_t), DECIMAL_STR_MAX(usec_t)) + 1]; | |
491 | int r; | |
492 | ||
493 | xsprintf(buf, "%" PRIu64 "\n", shares); | |
494 | r = cg_set_attribute("cpu", u->cgroup_path, "cpu.shares", buf); | |
495 | if (r < 0) | |
496 | log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, | |
497 | "Failed to set cpu.shares: %m"); | |
498 | ||
499 | xsprintf(buf, USEC_FMT "\n", CGROUP_CPU_QUOTA_PERIOD_USEC); | |
500 | r = cg_set_attribute("cpu", u->cgroup_path, "cpu.cfs_period_us", buf); | |
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.cfs_period_us: %m"); | |
504 | ||
505 | if (quota != USEC_INFINITY) { | |
506 | xsprintf(buf, USEC_FMT "\n", quota * CGROUP_CPU_QUOTA_PERIOD_USEC / USEC_PER_SEC); | |
507 | r = cg_set_attribute("cpu", u->cgroup_path, "cpu.cfs_quota_us", buf); | |
508 | } else | |
509 | r = cg_set_attribute("cpu", u->cgroup_path, "cpu.cfs_quota_us", "-1"); | |
510 | if (r < 0) | |
511 | log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, | |
512 | "Failed to set cpu.cfs_quota_us: %m"); | |
513 | } | |
514 | ||
515 | static uint64_t cgroup_cpu_shares_to_weight(uint64_t shares) { | |
516 | return CLAMP(shares * CGROUP_WEIGHT_DEFAULT / CGROUP_CPU_SHARES_DEFAULT, | |
517 | CGROUP_WEIGHT_MIN, CGROUP_WEIGHT_MAX); | |
518 | } | |
519 | ||
520 | static uint64_t cgroup_cpu_weight_to_shares(uint64_t weight) { | |
521 | return CLAMP(weight * CGROUP_CPU_SHARES_DEFAULT / CGROUP_WEIGHT_DEFAULT, | |
522 | CGROUP_CPU_SHARES_MIN, CGROUP_CPU_SHARES_MAX); | |
523 | } | |
524 | ||
508c45da | 525 | static bool cgroup_context_has_io_config(CGroupContext *c) { |
538b4852 TH |
526 | return c->io_accounting || |
527 | c->io_weight != CGROUP_WEIGHT_INVALID || | |
528 | c->startup_io_weight != CGROUP_WEIGHT_INVALID || | |
529 | c->io_device_weights || | |
530 | c->io_device_limits; | |
531 | } | |
532 | ||
508c45da | 533 | static bool cgroup_context_has_blockio_config(CGroupContext *c) { |
538b4852 TH |
534 | return c->blockio_accounting || |
535 | c->blockio_weight != CGROUP_BLKIO_WEIGHT_INVALID || | |
536 | c->startup_blockio_weight != CGROUP_BLKIO_WEIGHT_INVALID || | |
537 | c->blockio_device_weights || | |
538 | c->blockio_device_bandwidths; | |
539 | } | |
540 | ||
508c45da | 541 | static uint64_t cgroup_context_io_weight(CGroupContext *c, ManagerState state) { |
64faf04c TH |
542 | if (IN_SET(state, MANAGER_STARTING, MANAGER_INITIALIZING) && |
543 | c->startup_io_weight != CGROUP_WEIGHT_INVALID) | |
544 | return c->startup_io_weight; | |
545 | else if (c->io_weight != CGROUP_WEIGHT_INVALID) | |
546 | return c->io_weight; | |
547 | else | |
548 | return CGROUP_WEIGHT_DEFAULT; | |
549 | } | |
550 | ||
508c45da | 551 | static uint64_t cgroup_context_blkio_weight(CGroupContext *c, ManagerState state) { |
64faf04c TH |
552 | if (IN_SET(state, MANAGER_STARTING, MANAGER_INITIALIZING) && |
553 | c->startup_blockio_weight != CGROUP_BLKIO_WEIGHT_INVALID) | |
554 | return c->startup_blockio_weight; | |
555 | else if (c->blockio_weight != CGROUP_BLKIO_WEIGHT_INVALID) | |
556 | return c->blockio_weight; | |
557 | else | |
558 | return CGROUP_BLKIO_WEIGHT_DEFAULT; | |
559 | } | |
560 | ||
508c45da | 561 | static uint64_t cgroup_weight_blkio_to_io(uint64_t blkio_weight) { |
538b4852 TH |
562 | return CLAMP(blkio_weight * CGROUP_WEIGHT_DEFAULT / CGROUP_BLKIO_WEIGHT_DEFAULT, |
563 | CGROUP_WEIGHT_MIN, CGROUP_WEIGHT_MAX); | |
564 | } | |
565 | ||
508c45da | 566 | static uint64_t cgroup_weight_io_to_blkio(uint64_t io_weight) { |
538b4852 TH |
567 | return CLAMP(io_weight * CGROUP_BLKIO_WEIGHT_DEFAULT / CGROUP_WEIGHT_DEFAULT, |
568 | CGROUP_BLKIO_WEIGHT_MIN, CGROUP_BLKIO_WEIGHT_MAX); | |
569 | } | |
570 | ||
f29ff115 | 571 | static void cgroup_apply_io_device_weight(Unit *u, const char *dev_path, uint64_t io_weight) { |
64faf04c TH |
572 | char buf[DECIMAL_STR_MAX(dev_t)*2+2+DECIMAL_STR_MAX(uint64_t)+1]; |
573 | dev_t dev; | |
574 | int r; | |
575 | ||
576 | r = lookup_block_device(dev_path, &dev); | |
577 | if (r < 0) | |
578 | return; | |
579 | ||
580 | xsprintf(buf, "%u:%u %" PRIu64 "\n", major(dev), minor(dev), io_weight); | |
f29ff115 | 581 | r = cg_set_attribute("io", u->cgroup_path, "io.weight", buf); |
64faf04c | 582 | if (r < 0) |
f29ff115 TH |
583 | log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, |
584 | "Failed to set io.weight: %m"); | |
64faf04c TH |
585 | } |
586 | ||
f29ff115 | 587 | static void cgroup_apply_blkio_device_weight(Unit *u, const char *dev_path, uint64_t blkio_weight) { |
64faf04c TH |
588 | char buf[DECIMAL_STR_MAX(dev_t)*2+2+DECIMAL_STR_MAX(uint64_t)+1]; |
589 | dev_t dev; | |
590 | int r; | |
591 | ||
592 | r = lookup_block_device(dev_path, &dev); | |
593 | if (r < 0) | |
594 | return; | |
595 | ||
596 | xsprintf(buf, "%u:%u %" PRIu64 "\n", major(dev), minor(dev), blkio_weight); | |
f29ff115 | 597 | r = cg_set_attribute("blkio", u->cgroup_path, "blkio.weight_device", buf); |
64faf04c | 598 | if (r < 0) |
f29ff115 TH |
599 | log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, |
600 | "Failed to set blkio.weight_device: %m"); | |
64faf04c TH |
601 | } |
602 | ||
f29ff115 | 603 | static unsigned cgroup_apply_io_device_limit(Unit *u, const char *dev_path, uint64_t *limits) { |
64faf04c TH |
604 | char limit_bufs[_CGROUP_IO_LIMIT_TYPE_MAX][DECIMAL_STR_MAX(uint64_t)]; |
605 | char buf[DECIMAL_STR_MAX(dev_t)*2+2+(6+DECIMAL_STR_MAX(uint64_t)+1)*4]; | |
606 | CGroupIOLimitType type; | |
607 | dev_t dev; | |
608 | unsigned n = 0; | |
609 | int r; | |
610 | ||
611 | r = lookup_block_device(dev_path, &dev); | |
612 | if (r < 0) | |
613 | return 0; | |
614 | ||
615 | for (type = 0; type < _CGROUP_IO_LIMIT_TYPE_MAX; type++) { | |
616 | if (limits[type] != cgroup_io_limit_defaults[type]) { | |
617 | xsprintf(limit_bufs[type], "%" PRIu64, limits[type]); | |
618 | n++; | |
619 | } else { | |
620 | xsprintf(limit_bufs[type], "%s", limits[type] == CGROUP_LIMIT_MAX ? "max" : "0"); | |
621 | } | |
622 | } | |
623 | ||
624 | xsprintf(buf, "%u:%u rbps=%s wbps=%s riops=%s wiops=%s\n", major(dev), minor(dev), | |
625 | limit_bufs[CGROUP_IO_RBPS_MAX], limit_bufs[CGROUP_IO_WBPS_MAX], | |
626 | limit_bufs[CGROUP_IO_RIOPS_MAX], limit_bufs[CGROUP_IO_WIOPS_MAX]); | |
f29ff115 | 627 | r = cg_set_attribute("io", u->cgroup_path, "io.max", buf); |
64faf04c | 628 | if (r < 0) |
f29ff115 TH |
629 | log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, |
630 | "Failed to set io.max: %m"); | |
64faf04c TH |
631 | return n; |
632 | } | |
633 | ||
f29ff115 | 634 | static unsigned cgroup_apply_blkio_device_limit(Unit *u, const char *dev_path, uint64_t rbps, uint64_t wbps) { |
64faf04c TH |
635 | char buf[DECIMAL_STR_MAX(dev_t)*2+2+DECIMAL_STR_MAX(uint64_t)+1]; |
636 | dev_t dev; | |
637 | unsigned n = 0; | |
638 | int r; | |
639 | ||
640 | r = lookup_block_device(dev_path, &dev); | |
641 | if (r < 0) | |
642 | return 0; | |
643 | ||
644 | if (rbps != CGROUP_LIMIT_MAX) | |
645 | n++; | |
646 | sprintf(buf, "%u:%u %" PRIu64 "\n", major(dev), minor(dev), rbps); | |
f29ff115 | 647 | r = cg_set_attribute("blkio", u->cgroup_path, "blkio.throttle.read_bps_device", buf); |
64faf04c | 648 | if (r < 0) |
f29ff115 TH |
649 | log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, |
650 | "Failed to set blkio.throttle.read_bps_device: %m"); | |
64faf04c TH |
651 | |
652 | if (wbps != CGROUP_LIMIT_MAX) | |
653 | n++; | |
654 | sprintf(buf, "%u:%u %" PRIu64 "\n", major(dev), minor(dev), wbps); | |
f29ff115 | 655 | r = cg_set_attribute("blkio", u->cgroup_path, "blkio.throttle.write_bps_device", buf); |
64faf04c | 656 | if (r < 0) |
f29ff115 TH |
657 | log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, |
658 | "Failed to set blkio.throttle.write_bps_device: %m"); | |
64faf04c TH |
659 | |
660 | return n; | |
661 | } | |
662 | ||
da4d897e | 663 | static bool cgroup_context_has_unified_memory_config(CGroupContext *c) { |
96e131ea | 664 | 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 |
665 | } |
666 | ||
f29ff115 | 667 | static void cgroup_apply_unified_memory_limit(Unit *u, const char *file, uint64_t v) { |
da4d897e TH |
668 | char buf[DECIMAL_STR_MAX(uint64_t) + 1] = "max"; |
669 | int r; | |
670 | ||
671 | if (v != CGROUP_LIMIT_MAX) | |
672 | xsprintf(buf, "%" PRIu64 "\n", v); | |
673 | ||
f29ff115 | 674 | r = cg_set_attribute("memory", u->cgroup_path, file, buf); |
da4d897e | 675 | if (r < 0) |
f29ff115 TH |
676 | log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, |
677 | "Failed to set %s: %m", file); | |
da4d897e TH |
678 | } |
679 | ||
0f2d84d2 | 680 | static void cgroup_apply_firewall(Unit *u) { |
906c06f6 DM |
681 | int r; |
682 | ||
0f2d84d2 LP |
683 | assert(u); |
684 | ||
906c06f6 DM |
685 | if (u->type == UNIT_SLICE) /* Skip this for slice units, they are inner cgroup nodes, and since bpf/cgroup is |
686 | * not recursive we don't ever touch the bpf on them */ | |
687 | return; | |
688 | ||
689 | r = bpf_firewall_compile(u); | |
690 | if (r < 0) | |
691 | return; | |
692 | ||
693 | (void) bpf_firewall_install(u); | |
694 | return; | |
695 | } | |
696 | ||
697 | static void cgroup_context_apply( | |
698 | Unit *u, | |
699 | CGroupMask apply_mask, | |
700 | bool apply_bpf, | |
701 | ManagerState state) { | |
702 | ||
f29ff115 TH |
703 | const char *path; |
704 | CGroupContext *c; | |
01efdf13 | 705 | bool is_root; |
4ad49000 LP |
706 | int r; |
707 | ||
f29ff115 TH |
708 | assert(u); |
709 | ||
710 | c = unit_get_cgroup_context(u); | |
711 | path = u->cgroup_path; | |
712 | ||
4ad49000 LP |
713 | assert(c); |
714 | assert(path); | |
8e274523 | 715 | |
906c06f6 DM |
716 | /* Nothing to do? Exit early! */ |
717 | if (apply_mask == 0 && !apply_bpf) | |
4ad49000 | 718 | return; |
8e274523 | 719 | |
71c26873 | 720 | /* Some cgroup attributes are not supported on the root cgroup, |
01efdf13 LP |
721 | * hence silently ignore */ |
722 | is_root = isempty(path) || path_equal(path, "/"); | |
6da13913 ZJS |
723 | if (is_root) |
724 | /* Make sure we don't try to display messages with an empty path. */ | |
725 | path = "/"; | |
01efdf13 | 726 | |
714e2e1d LP |
727 | /* We generally ignore errors caused by read-only mounted |
728 | * cgroup trees (assuming we are running in a container then), | |
729 | * and missing cgroups, i.e. EROFS and ENOENT. */ | |
730 | ||
906c06f6 DM |
731 | if ((apply_mask & CGROUP_MASK_CPU) && !is_root) { |
732 | bool has_weight, has_shares; | |
733 | ||
734 | has_weight = cgroup_context_has_cpu_weight(c); | |
735 | has_shares = cgroup_context_has_cpu_shares(c); | |
8e274523 | 736 | |
b4cccbc1 | 737 | if (cg_all_unified() > 0) { |
66ebf6c0 | 738 | uint64_t weight; |
b2f8b02e | 739 | |
66ebf6c0 TH |
740 | if (has_weight) |
741 | weight = cgroup_context_cpu_weight(c, state); | |
742 | else if (has_shares) { | |
743 | uint64_t shares = cgroup_context_cpu_shares(c, state); | |
b2f8b02e | 744 | |
66ebf6c0 TH |
745 | weight = cgroup_cpu_shares_to_weight(shares); |
746 | ||
747 | log_cgroup_compat(u, "Applying [Startup]CpuShares %" PRIu64 " as [Startup]CpuWeight %" PRIu64 " on %s", | |
748 | shares, weight, path); | |
749 | } else | |
750 | weight = CGROUP_WEIGHT_DEFAULT; | |
751 | ||
752 | cgroup_apply_unified_cpu_config(u, weight, c->cpu_quota_per_sec_usec); | |
753 | } else { | |
754 | uint64_t shares; | |
755 | ||
7d862ab8 | 756 | if (has_weight) { |
66ebf6c0 TH |
757 | uint64_t weight = cgroup_context_cpu_weight(c, state); |
758 | ||
759 | shares = cgroup_cpu_weight_to_shares(weight); | |
760 | ||
761 | log_cgroup_compat(u, "Applying [Startup]CpuWeight %" PRIu64 " as [Startup]CpuShares %" PRIu64 " on %s", | |
762 | weight, shares, path); | |
7d862ab8 TH |
763 | } else if (has_shares) |
764 | shares = cgroup_context_cpu_shares(c, state); | |
765 | else | |
66ebf6c0 TH |
766 | shares = CGROUP_CPU_SHARES_DEFAULT; |
767 | ||
768 | cgroup_apply_legacy_cpu_config(u, shares, c->cpu_quota_per_sec_usec); | |
769 | } | |
4ad49000 LP |
770 | } |
771 | ||
906c06f6 | 772 | if (apply_mask & CGROUP_MASK_IO) { |
538b4852 TH |
773 | bool has_io = cgroup_context_has_io_config(c); |
774 | bool has_blockio = cgroup_context_has_blockio_config(c); | |
13c31542 TH |
775 | |
776 | if (!is_root) { | |
64faf04c TH |
777 | char buf[8+DECIMAL_STR_MAX(uint64_t)+1]; |
778 | uint64_t weight; | |
13c31542 | 779 | |
538b4852 TH |
780 | if (has_io) |
781 | weight = cgroup_context_io_weight(c, state); | |
128fadc9 TH |
782 | else if (has_blockio) { |
783 | uint64_t blkio_weight = cgroup_context_blkio_weight(c, state); | |
784 | ||
785 | weight = cgroup_weight_blkio_to_io(blkio_weight); | |
786 | ||
787 | log_cgroup_compat(u, "Applying [Startup]BlockIOWeight %" PRIu64 " as [Startup]IOWeight %" PRIu64, | |
788 | blkio_weight, weight); | |
789 | } else | |
538b4852 | 790 | weight = CGROUP_WEIGHT_DEFAULT; |
13c31542 TH |
791 | |
792 | xsprintf(buf, "default %" PRIu64 "\n", weight); | |
793 | r = cg_set_attribute("io", path, "io.weight", buf); | |
794 | if (r < 0) | |
f29ff115 TH |
795 | log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, |
796 | "Failed to set io.weight: %m"); | |
13c31542 | 797 | |
538b4852 TH |
798 | if (has_io) { |
799 | CGroupIODeviceWeight *w; | |
800 | ||
801 | /* FIXME: no way to reset this list */ | |
802 | LIST_FOREACH(device_weights, w, c->io_device_weights) | |
f29ff115 | 803 | cgroup_apply_io_device_weight(u, w->path, w->weight); |
538b4852 TH |
804 | } else if (has_blockio) { |
805 | CGroupBlockIODeviceWeight *w; | |
806 | ||
807 | /* FIXME: no way to reset this list */ | |
128fadc9 TH |
808 | LIST_FOREACH(device_weights, w, c->blockio_device_weights) { |
809 | weight = cgroup_weight_blkio_to_io(w->weight); | |
810 | ||
811 | log_cgroup_compat(u, "Applying BlockIODeviceWeight %" PRIu64 " as IODeviceWeight %" PRIu64 " for %s", | |
812 | w->weight, weight, w->path); | |
813 | ||
814 | cgroup_apply_io_device_weight(u, w->path, weight); | |
815 | } | |
538b4852 | 816 | } |
13c31542 TH |
817 | } |
818 | ||
64faf04c | 819 | /* Apply limits and free ones without config. */ |
538b4852 TH |
820 | if (has_io) { |
821 | CGroupIODeviceLimit *l, *next; | |
822 | ||
823 | LIST_FOREACH_SAFE(device_limits, l, next, c->io_device_limits) { | |
f29ff115 | 824 | if (!cgroup_apply_io_device_limit(u, l->path, l->limits)) |
538b4852 TH |
825 | cgroup_context_free_io_device_limit(c, l); |
826 | } | |
827 | } else if (has_blockio) { | |
828 | CGroupBlockIODeviceBandwidth *b, *next; | |
829 | ||
830 | LIST_FOREACH_SAFE(device_bandwidths, b, next, c->blockio_device_bandwidths) { | |
831 | uint64_t limits[_CGROUP_IO_LIMIT_TYPE_MAX]; | |
832 | CGroupIOLimitType type; | |
833 | ||
834 | for (type = 0; type < _CGROUP_IO_LIMIT_TYPE_MAX; type++) | |
835 | limits[type] = cgroup_io_limit_defaults[type]; | |
836 | ||
837 | limits[CGROUP_IO_RBPS_MAX] = b->rbps; | |
838 | limits[CGROUP_IO_WBPS_MAX] = b->wbps; | |
839 | ||
128fadc9 TH |
840 | log_cgroup_compat(u, "Applying BlockIO{Read|Write}Bandwidth %" PRIu64 " %" PRIu64 " as IO{Read|Write}BandwidthMax for %s", |
841 | b->rbps, b->wbps, b->path); | |
842 | ||
f29ff115 | 843 | if (!cgroup_apply_io_device_limit(u, b->path, limits)) |
538b4852 TH |
844 | cgroup_context_free_blockio_device_bandwidth(c, b); |
845 | } | |
13c31542 TH |
846 | } |
847 | } | |
848 | ||
906c06f6 | 849 | if (apply_mask & CGROUP_MASK_BLKIO) { |
538b4852 TH |
850 | bool has_io = cgroup_context_has_io_config(c); |
851 | bool has_blockio = cgroup_context_has_blockio_config(c); | |
4ad49000 | 852 | |
01efdf13 | 853 | if (!is_root) { |
64faf04c TH |
854 | char buf[DECIMAL_STR_MAX(uint64_t)+1]; |
855 | uint64_t weight; | |
64faf04c | 856 | |
7d862ab8 | 857 | if (has_io) { |
128fadc9 TH |
858 | uint64_t io_weight = cgroup_context_io_weight(c, state); |
859 | ||
538b4852 | 860 | weight = cgroup_weight_io_to_blkio(cgroup_context_io_weight(c, state)); |
128fadc9 TH |
861 | |
862 | log_cgroup_compat(u, "Applying [Startup]IOWeight %" PRIu64 " as [Startup]BlockIOWeight %" PRIu64, | |
863 | io_weight, weight); | |
7d862ab8 TH |
864 | } else if (has_blockio) |
865 | weight = cgroup_context_blkio_weight(c, state); | |
866 | else | |
538b4852 | 867 | weight = CGROUP_BLKIO_WEIGHT_DEFAULT; |
64faf04c TH |
868 | |
869 | xsprintf(buf, "%" PRIu64 "\n", weight); | |
01efdf13 | 870 | r = cg_set_attribute("blkio", path, "blkio.weight", buf); |
1aeab12b | 871 | if (r < 0) |
f29ff115 TH |
872 | log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, |
873 | "Failed to set blkio.weight: %m"); | |
4ad49000 | 874 | |
7d862ab8 | 875 | if (has_io) { |
538b4852 TH |
876 | CGroupIODeviceWeight *w; |
877 | ||
878 | /* FIXME: no way to reset this list */ | |
128fadc9 TH |
879 | LIST_FOREACH(device_weights, w, c->io_device_weights) { |
880 | weight = cgroup_weight_io_to_blkio(w->weight); | |
881 | ||
882 | log_cgroup_compat(u, "Applying IODeviceWeight %" PRIu64 " as BlockIODeviceWeight %" PRIu64 " for %s", | |
883 | w->weight, weight, w->path); | |
884 | ||
885 | cgroup_apply_blkio_device_weight(u, w->path, weight); | |
886 | } | |
7d862ab8 TH |
887 | } else if (has_blockio) { |
888 | CGroupBlockIODeviceWeight *w; | |
889 | ||
890 | /* FIXME: no way to reset this list */ | |
891 | LIST_FOREACH(device_weights, w, c->blockio_device_weights) | |
892 | cgroup_apply_blkio_device_weight(u, w->path, w->weight); | |
538b4852 | 893 | } |
4ad49000 LP |
894 | } |
895 | ||
64faf04c | 896 | /* Apply limits and free ones without config. */ |
7d862ab8 | 897 | if (has_io) { |
538b4852 TH |
898 | CGroupIODeviceLimit *l, *next; |
899 | ||
900 | LIST_FOREACH_SAFE(device_limits, l, next, c->io_device_limits) { | |
128fadc9 TH |
901 | log_cgroup_compat(u, "Applying IO{Read|Write}Bandwidth %" PRIu64 " %" PRIu64 " as BlockIO{Read|Write}BandwidthMax for %s", |
902 | l->limits[CGROUP_IO_RBPS_MAX], l->limits[CGROUP_IO_WBPS_MAX], l->path); | |
903 | ||
f29ff115 | 904 | if (!cgroup_apply_blkio_device_limit(u, l->path, l->limits[CGROUP_IO_RBPS_MAX], l->limits[CGROUP_IO_WBPS_MAX])) |
538b4852 TH |
905 | cgroup_context_free_io_device_limit(c, l); |
906 | } | |
7d862ab8 TH |
907 | } else if (has_blockio) { |
908 | CGroupBlockIODeviceBandwidth *b, *next; | |
909 | ||
910 | LIST_FOREACH_SAFE(device_bandwidths, b, next, c->blockio_device_bandwidths) | |
911 | if (!cgroup_apply_blkio_device_limit(u, b->path, b->rbps, b->wbps)) | |
912 | cgroup_context_free_blockio_device_bandwidth(c, b); | |
d686d8a9 | 913 | } |
8e274523 LP |
914 | } |
915 | ||
906c06f6 | 916 | if ((apply_mask & CGROUP_MASK_MEMORY) && !is_root) { |
b4cccbc1 LP |
917 | if (cg_all_unified() > 0) { |
918 | uint64_t max, swap_max = CGROUP_LIMIT_MAX; | |
efdb0237 | 919 | |
96e131ea | 920 | if (cgroup_context_has_unified_memory_config(c)) { |
da4d897e | 921 | max = c->memory_max; |
96e131ea WC |
922 | swap_max = c->memory_swap_max; |
923 | } else { | |
da4d897e | 924 | max = c->memory_limit; |
efdb0237 | 925 | |
128fadc9 TH |
926 | if (max != CGROUP_LIMIT_MAX) |
927 | log_cgroup_compat(u, "Applying MemoryLimit %" PRIu64 " as MemoryMax", max); | |
928 | } | |
929 | ||
f29ff115 TH |
930 | cgroup_apply_unified_memory_limit(u, "memory.low", c->memory_low); |
931 | cgroup_apply_unified_memory_limit(u, "memory.high", c->memory_high); | |
932 | cgroup_apply_unified_memory_limit(u, "memory.max", max); | |
96e131ea | 933 | cgroup_apply_unified_memory_limit(u, "memory.swap.max", swap_max); |
efdb0237 | 934 | } else { |
da4d897e | 935 | char buf[DECIMAL_STR_MAX(uint64_t) + 1]; |
7d862ab8 | 936 | uint64_t val; |
da4d897e | 937 | |
7d862ab8 | 938 | if (cgroup_context_has_unified_memory_config(c)) { |
78a4ee59 | 939 | val = c->memory_max; |
7d862ab8 TH |
940 | log_cgroup_compat(u, "Applying MemoryMax %" PRIi64 " as MemoryLimit", val); |
941 | } else | |
942 | val = c->memory_limit; | |
128fadc9 | 943 | |
78a4ee59 DM |
944 | if (val == CGROUP_LIMIT_MAX) |
945 | strncpy(buf, "-1\n", sizeof(buf)); | |
946 | else | |
947 | xsprintf(buf, "%" PRIu64 "\n", val); | |
948 | ||
da4d897e TH |
949 | r = cg_set_attribute("memory", path, "memory.limit_in_bytes", buf); |
950 | if (r < 0) | |
f29ff115 TH |
951 | log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, |
952 | "Failed to set memory.limit_in_bytes: %m"); | |
da4d897e | 953 | } |
4ad49000 | 954 | } |
8e274523 | 955 | |
906c06f6 | 956 | if ((apply_mask & CGROUP_MASK_DEVICES) && !is_root) { |
4ad49000 | 957 | CGroupDeviceAllow *a; |
8e274523 | 958 | |
714e2e1d LP |
959 | /* Changing the devices list of a populated cgroup |
960 | * might result in EINVAL, hence ignore EINVAL | |
961 | * here. */ | |
962 | ||
4ad49000 LP |
963 | if (c->device_allow || c->device_policy != CGROUP_AUTO) |
964 | r = cg_set_attribute("devices", path, "devices.deny", "a"); | |
965 | else | |
966 | r = cg_set_attribute("devices", path, "devices.allow", "a"); | |
1aeab12b | 967 | if (r < 0) |
f29ff115 TH |
968 | log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EINVAL, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, |
969 | "Failed to reset devices.list: %m"); | |
fb385181 | 970 | |
4ad49000 LP |
971 | if (c->device_policy == CGROUP_CLOSED || |
972 | (c->device_policy == CGROUP_AUTO && c->device_allow)) { | |
973 | static const char auto_devices[] = | |
7d711efb LP |
974 | "/dev/null\0" "rwm\0" |
975 | "/dev/zero\0" "rwm\0" | |
976 | "/dev/full\0" "rwm\0" | |
977 | "/dev/random\0" "rwm\0" | |
978 | "/dev/urandom\0" "rwm\0" | |
979 | "/dev/tty\0" "rwm\0" | |
0d9e7991 AP |
980 | "/dev/pts/ptmx\0" "rw\0" /* /dev/pts/ptmx may not be duplicated, but accessed */ |
981 | /* Allow /run/systemd/inaccessible/{chr,blk} devices for mapping InaccessiblePaths */ | |
e7330dfe DP |
982 | "-/run/systemd/inaccessible/chr\0" "rwm\0" |
983 | "-/run/systemd/inaccessible/blk\0" "rwm\0"; | |
4ad49000 LP |
984 | |
985 | const char *x, *y; | |
986 | ||
987 | NULSTR_FOREACH_PAIR(x, y, auto_devices) | |
988 | whitelist_device(path, x, y); | |
7d711efb LP |
989 | |
990 | whitelist_major(path, "pts", 'c', "rw"); | |
4ad49000 LP |
991 | } |
992 | ||
993 | LIST_FOREACH(device_allow, a, c->device_allow) { | |
fb4650aa | 994 | char acc[4], *val; |
4ad49000 LP |
995 | unsigned k = 0; |
996 | ||
997 | if (a->r) | |
998 | acc[k++] = 'r'; | |
999 | if (a->w) | |
1000 | acc[k++] = 'w'; | |
1001 | if (a->m) | |
1002 | acc[k++] = 'm'; | |
fb385181 | 1003 | |
4ad49000 LP |
1004 | if (k == 0) |
1005 | continue; | |
fb385181 | 1006 | |
4ad49000 | 1007 | acc[k++] = 0; |
90060676 | 1008 | |
27458ed6 | 1009 | if (path_startswith(a->path, "/dev/")) |
90060676 | 1010 | whitelist_device(path, a->path, acc); |
fb4650aa ZJS |
1011 | else if ((val = startswith(a->path, "block-"))) |
1012 | whitelist_major(path, val, 'b', acc); | |
1013 | else if ((val = startswith(a->path, "char-"))) | |
1014 | whitelist_major(path, val, 'c', acc); | |
90060676 | 1015 | else |
f29ff115 | 1016 | log_unit_debug(u, "Ignoring device %s while writing cgroup attribute.", a->path); |
4ad49000 LP |
1017 | } |
1018 | } | |
03a7b521 | 1019 | |
906c06f6 | 1020 | if ((apply_mask & CGROUP_MASK_PIDS) && !is_root) { |
03a7b521 | 1021 | |
f5058264 | 1022 | if (c->tasks_max != CGROUP_LIMIT_MAX) { |
03a7b521 LP |
1023 | char buf[DECIMAL_STR_MAX(uint64_t) + 2]; |
1024 | ||
1025 | sprintf(buf, "%" PRIu64 "\n", c->tasks_max); | |
1026 | r = cg_set_attribute("pids", path, "pids.max", buf); | |
1027 | } else | |
1028 | r = cg_set_attribute("pids", path, "pids.max", "max"); | |
1029 | ||
1030 | if (r < 0) | |
f29ff115 TH |
1031 | log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, |
1032 | "Failed to set pids.max: %m"); | |
03a7b521 | 1033 | } |
906c06f6 DM |
1034 | |
1035 | if (apply_bpf) | |
0f2d84d2 | 1036 | cgroup_apply_firewall(u); |
fb385181 LP |
1037 | } |
1038 | ||
efdb0237 LP |
1039 | CGroupMask cgroup_context_get_mask(CGroupContext *c) { |
1040 | CGroupMask mask = 0; | |
8e274523 | 1041 | |
4ad49000 | 1042 | /* Figure out which controllers we need */ |
8e274523 | 1043 | |
b2f8b02e | 1044 | if (c->cpu_accounting || |
66ebf6c0 TH |
1045 | cgroup_context_has_cpu_weight(c) || |
1046 | cgroup_context_has_cpu_shares(c) || | |
3a43da28 | 1047 | c->cpu_quota_per_sec_usec != USEC_INFINITY) |
efdb0237 | 1048 | mask |= CGROUP_MASK_CPUACCT | CGROUP_MASK_CPU; |
ecedd90f | 1049 | |
538b4852 TH |
1050 | if (cgroup_context_has_io_config(c) || cgroup_context_has_blockio_config(c)) |
1051 | mask |= CGROUP_MASK_IO | CGROUP_MASK_BLKIO; | |
ecedd90f | 1052 | |
4ad49000 | 1053 | if (c->memory_accounting || |
da4d897e TH |
1054 | c->memory_limit != CGROUP_LIMIT_MAX || |
1055 | cgroup_context_has_unified_memory_config(c)) | |
efdb0237 | 1056 | mask |= CGROUP_MASK_MEMORY; |
8e274523 | 1057 | |
a931ad47 LP |
1058 | if (c->device_allow || |
1059 | c->device_policy != CGROUP_AUTO) | |
3905f127 | 1060 | mask |= CGROUP_MASK_DEVICES; |
4ad49000 | 1061 | |
03a7b521 LP |
1062 | if (c->tasks_accounting || |
1063 | c->tasks_max != (uint64_t) -1) | |
1064 | mask |= CGROUP_MASK_PIDS; | |
1065 | ||
4ad49000 | 1066 | return mask; |
8e274523 LP |
1067 | } |
1068 | ||
efdb0237 | 1069 | CGroupMask unit_get_own_mask(Unit *u) { |
4ad49000 | 1070 | CGroupContext *c; |
8e274523 | 1071 | |
efdb0237 LP |
1072 | /* Returns the mask of controllers the unit needs for itself */ |
1073 | ||
4ad49000 LP |
1074 | c = unit_get_cgroup_context(u); |
1075 | if (!c) | |
1076 | return 0; | |
8e274523 | 1077 | |
64e844e5 | 1078 | return cgroup_context_get_mask(c) | unit_get_delegate_mask(u); |
02638280 LP |
1079 | } |
1080 | ||
1081 | CGroupMask unit_get_delegate_mask(Unit *u) { | |
1082 | CGroupContext *c; | |
1083 | ||
1084 | /* If delegation is turned on, then turn on selected controllers, unless we are on the legacy hierarchy and the | |
1085 | * process we fork into is known to drop privileges, and hence shouldn't get access to the controllers. | |
19af675e | 1086 | * |
02638280 | 1087 | * Note that on the unified hierarchy it is safe to delegate controllers to unprivileged services. */ |
a931ad47 | 1088 | |
02638280 LP |
1089 | if (u->type == UNIT_SLICE) |
1090 | return 0; | |
1091 | ||
1092 | c = unit_get_cgroup_context(u); | |
1093 | if (!c) | |
1094 | return 0; | |
1095 | ||
1096 | if (!c->delegate) | |
1097 | return 0; | |
1098 | ||
1099 | if (cg_all_unified() <= 0) { | |
a931ad47 LP |
1100 | ExecContext *e; |
1101 | ||
1102 | e = unit_get_exec_context(u); | |
02638280 LP |
1103 | if (e && !exec_context_maintains_privileges(e)) |
1104 | return 0; | |
a931ad47 LP |
1105 | } |
1106 | ||
02638280 | 1107 | return c->delegate_controllers; |
8e274523 LP |
1108 | } |
1109 | ||
efdb0237 | 1110 | CGroupMask unit_get_members_mask(Unit *u) { |
4ad49000 | 1111 | assert(u); |
bc432dc7 | 1112 | |
02638280 | 1113 | /* Returns the mask of controllers all of the unit's children require, merged */ |
efdb0237 | 1114 | |
bc432dc7 LP |
1115 | if (u->cgroup_members_mask_valid) |
1116 | return u->cgroup_members_mask; | |
1117 | ||
64e844e5 | 1118 | u->cgroup_members_mask = 0; |
bc432dc7 LP |
1119 | |
1120 | if (u->type == UNIT_SLICE) { | |
eef85c4a | 1121 | void *v; |
bc432dc7 LP |
1122 | Unit *member; |
1123 | Iterator i; | |
1124 | ||
eef85c4a | 1125 | HASHMAP_FOREACH_KEY(v, member, u->dependencies[UNIT_BEFORE], i) { |
bc432dc7 LP |
1126 | |
1127 | if (member == u) | |
1128 | continue; | |
1129 | ||
d4fdc205 | 1130 | if (UNIT_DEREF(member->slice) != u) |
bc432dc7 LP |
1131 | continue; |
1132 | ||
31604970 | 1133 | u->cgroup_members_mask |= unit_get_subtree_mask(member); /* note that this calls ourselves again, for the children */ |
bc432dc7 LP |
1134 | } |
1135 | } | |
1136 | ||
1137 | u->cgroup_members_mask_valid = true; | |
6414b7c9 | 1138 | return u->cgroup_members_mask; |
246aa6dd LP |
1139 | } |
1140 | ||
efdb0237 | 1141 | CGroupMask unit_get_siblings_mask(Unit *u) { |
4ad49000 | 1142 | assert(u); |
246aa6dd | 1143 | |
efdb0237 LP |
1144 | /* Returns the mask of controllers all of the unit's siblings |
1145 | * require, i.e. the members mask of the unit's parent slice | |
1146 | * if there is one. */ | |
1147 | ||
bc432dc7 | 1148 | if (UNIT_ISSET(u->slice)) |
637f421e | 1149 | return unit_get_members_mask(UNIT_DEREF(u->slice)); |
4ad49000 | 1150 | |
64e844e5 | 1151 | return unit_get_subtree_mask(u); /* we are the top-level slice */ |
246aa6dd LP |
1152 | } |
1153 | ||
efdb0237 LP |
1154 | CGroupMask unit_get_subtree_mask(Unit *u) { |
1155 | ||
1156 | /* Returns the mask of this subtree, meaning of the group | |
1157 | * itself and its children. */ | |
1158 | ||
1159 | return unit_get_own_mask(u) | unit_get_members_mask(u); | |
1160 | } | |
1161 | ||
1162 | CGroupMask unit_get_target_mask(Unit *u) { | |
1163 | CGroupMask mask; | |
1164 | ||
1165 | /* This returns the cgroup mask of all controllers to enable | |
1166 | * for a specific cgroup, i.e. everything it needs itself, | |
1167 | * plus all that its children need, plus all that its siblings | |
1168 | * need. This is primarily useful on the legacy cgroup | |
1169 | * hierarchy, where we need to duplicate each cgroup in each | |
1170 | * hierarchy that shall be enabled for it. */ | |
6414b7c9 | 1171 | |
efdb0237 LP |
1172 | mask = unit_get_own_mask(u) | unit_get_members_mask(u) | unit_get_siblings_mask(u); |
1173 | mask &= u->manager->cgroup_supported; | |
1174 | ||
1175 | return mask; | |
1176 | } | |
1177 | ||
1178 | CGroupMask unit_get_enable_mask(Unit *u) { | |
1179 | CGroupMask mask; | |
1180 | ||
1181 | /* This returns the cgroup mask of all controllers to enable | |
1182 | * for the children of a specific cgroup. This is primarily | |
1183 | * useful for the unified cgroup hierarchy, where each cgroup | |
1184 | * controls which controllers are enabled for its children. */ | |
1185 | ||
1186 | mask = unit_get_members_mask(u); | |
6414b7c9 DS |
1187 | mask &= u->manager->cgroup_supported; |
1188 | ||
1189 | return mask; | |
1190 | } | |
1191 | ||
906c06f6 DM |
1192 | bool unit_get_needs_bpf(Unit *u) { |
1193 | CGroupContext *c; | |
1194 | Unit *p; | |
1195 | assert(u); | |
1196 | ||
1197 | /* We never attach BPF to slice units, as they are inner cgroup nodes and cgroup/BPF is not recursive at the | |
1198 | * moment. */ | |
1199 | if (u->type == UNIT_SLICE) | |
1200 | return false; | |
1201 | ||
1202 | c = unit_get_cgroup_context(u); | |
1203 | if (!c) | |
1204 | return false; | |
1205 | ||
1206 | if (c->ip_accounting || | |
1207 | c->ip_address_allow || | |
1208 | c->ip_address_deny) | |
1209 | return true; | |
1210 | ||
1211 | /* If any parent slice has an IP access list defined, it applies too */ | |
1212 | for (p = UNIT_DEREF(u->slice); p; p = UNIT_DEREF(p->slice)) { | |
1213 | c = unit_get_cgroup_context(p); | |
1214 | if (!c) | |
1215 | return false; | |
1216 | ||
1217 | if (c->ip_address_allow || | |
1218 | c->ip_address_deny) | |
1219 | return true; | |
1220 | } | |
1221 | ||
1222 | return false; | |
1223 | } | |
1224 | ||
6414b7c9 DS |
1225 | /* Recurse from a unit up through its containing slices, propagating |
1226 | * mask bits upward. A unit is also member of itself. */ | |
bc432dc7 | 1227 | void unit_update_cgroup_members_masks(Unit *u) { |
efdb0237 | 1228 | CGroupMask m; |
bc432dc7 LP |
1229 | bool more; |
1230 | ||
1231 | assert(u); | |
1232 | ||
1233 | /* Calculate subtree mask */ | |
efdb0237 | 1234 | m = unit_get_subtree_mask(u); |
bc432dc7 LP |
1235 | |
1236 | /* See if anything changed from the previous invocation. If | |
1237 | * not, we're done. */ | |
1238 | if (u->cgroup_subtree_mask_valid && m == u->cgroup_subtree_mask) | |
1239 | return; | |
1240 | ||
1241 | more = | |
1242 | u->cgroup_subtree_mask_valid && | |
1243 | ((m & ~u->cgroup_subtree_mask) != 0) && | |
1244 | ((~m & u->cgroup_subtree_mask) == 0); | |
1245 | ||
1246 | u->cgroup_subtree_mask = m; | |
1247 | u->cgroup_subtree_mask_valid = true; | |
1248 | ||
6414b7c9 DS |
1249 | if (UNIT_ISSET(u->slice)) { |
1250 | Unit *s = UNIT_DEREF(u->slice); | |
bc432dc7 LP |
1251 | |
1252 | if (more) | |
1253 | /* There's more set now than before. We | |
1254 | * propagate the new mask to the parent's mask | |
1255 | * (not caring if it actually was valid or | |
1256 | * not). */ | |
1257 | ||
1258 | s->cgroup_members_mask |= m; | |
1259 | ||
1260 | else | |
1261 | /* There's less set now than before (or we | |
1262 | * don't know), we need to recalculate | |
1263 | * everything, so let's invalidate the | |
1264 | * parent's members mask */ | |
1265 | ||
1266 | s->cgroup_members_mask_valid = false; | |
1267 | ||
1268 | /* And now make sure that this change also hits our | |
1269 | * grandparents */ | |
1270 | unit_update_cgroup_members_masks(s); | |
6414b7c9 DS |
1271 | } |
1272 | } | |
1273 | ||
efdb0237 | 1274 | static const char *migrate_callback(CGroupMask mask, void *userdata) { |
03b90d4b LP |
1275 | Unit *u = userdata; |
1276 | ||
1277 | assert(mask != 0); | |
1278 | assert(u); | |
1279 | ||
1280 | while (u) { | |
1281 | if (u->cgroup_path && | |
1282 | u->cgroup_realized && | |
1283 | (u->cgroup_realized_mask & mask) == mask) | |
1284 | return u->cgroup_path; | |
1285 | ||
1286 | u = UNIT_DEREF(u->slice); | |
1287 | } | |
1288 | ||
1289 | return NULL; | |
1290 | } | |
1291 | ||
efdb0237 LP |
1292 | char *unit_default_cgroup_path(Unit *u) { |
1293 | _cleanup_free_ char *escaped = NULL, *slice = NULL; | |
1294 | int r; | |
1295 | ||
1296 | assert(u); | |
1297 | ||
1298 | if (unit_has_name(u, SPECIAL_ROOT_SLICE)) | |
1299 | return strdup(u->manager->cgroup_root); | |
1300 | ||
1301 | if (UNIT_ISSET(u->slice) && !unit_has_name(UNIT_DEREF(u->slice), SPECIAL_ROOT_SLICE)) { | |
1302 | r = cg_slice_to_path(UNIT_DEREF(u->slice)->id, &slice); | |
1303 | if (r < 0) | |
1304 | return NULL; | |
1305 | } | |
1306 | ||
1307 | escaped = cg_escape(u->id); | |
1308 | if (!escaped) | |
1309 | return NULL; | |
1310 | ||
1311 | if (slice) | |
605405c6 ZJS |
1312 | return strjoin(u->manager->cgroup_root, "/", slice, "/", |
1313 | escaped); | |
efdb0237 | 1314 | else |
605405c6 | 1315 | return strjoin(u->manager->cgroup_root, "/", escaped); |
efdb0237 LP |
1316 | } |
1317 | ||
1318 | int unit_set_cgroup_path(Unit *u, const char *path) { | |
1319 | _cleanup_free_ char *p = NULL; | |
1320 | int r; | |
1321 | ||
1322 | assert(u); | |
1323 | ||
1324 | if (path) { | |
1325 | p = strdup(path); | |
1326 | if (!p) | |
1327 | return -ENOMEM; | |
1328 | } else | |
1329 | p = NULL; | |
1330 | ||
1331 | if (streq_ptr(u->cgroup_path, p)) | |
1332 | return 0; | |
1333 | ||
1334 | if (p) { | |
1335 | r = hashmap_put(u->manager->cgroup_unit, p, u); | |
1336 | if (r < 0) | |
1337 | return r; | |
1338 | } | |
1339 | ||
1340 | unit_release_cgroup(u); | |
1341 | ||
1342 | u->cgroup_path = p; | |
1343 | p = NULL; | |
1344 | ||
1345 | return 1; | |
1346 | } | |
1347 | ||
1348 | int unit_watch_cgroup(Unit *u) { | |
ab2c3861 | 1349 | _cleanup_free_ char *events = NULL; |
efdb0237 LP |
1350 | int r; |
1351 | ||
1352 | assert(u); | |
1353 | ||
1354 | if (!u->cgroup_path) | |
1355 | return 0; | |
1356 | ||
1357 | if (u->cgroup_inotify_wd >= 0) | |
1358 | return 0; | |
1359 | ||
1360 | /* Only applies to the unified hierarchy */ | |
c22800e4 | 1361 | r = cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER); |
b4cccbc1 LP |
1362 | if (r < 0) |
1363 | return log_error_errno(r, "Failed to determine whether the name=systemd hierarchy is unified: %m"); | |
1364 | if (r == 0) | |
efdb0237 LP |
1365 | return 0; |
1366 | ||
1367 | /* Don't watch the root slice, it's pointless. */ | |
1368 | if (unit_has_name(u, SPECIAL_ROOT_SLICE)) | |
1369 | return 0; | |
1370 | ||
1371 | r = hashmap_ensure_allocated(&u->manager->cgroup_inotify_wd_unit, &trivial_hash_ops); | |
1372 | if (r < 0) | |
1373 | return log_oom(); | |
1374 | ||
ab2c3861 | 1375 | r = cg_get_path(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path, "cgroup.events", &events); |
efdb0237 LP |
1376 | if (r < 0) |
1377 | return log_oom(); | |
1378 | ||
ab2c3861 | 1379 | u->cgroup_inotify_wd = inotify_add_watch(u->manager->cgroup_inotify_fd, events, IN_MODIFY); |
efdb0237 LP |
1380 | if (u->cgroup_inotify_wd < 0) { |
1381 | ||
1382 | if (errno == ENOENT) /* If the directory is already | |
1383 | * gone we don't need to track | |
1384 | * it, so this is not an error */ | |
1385 | return 0; | |
1386 | ||
1387 | return log_unit_error_errno(u, errno, "Failed to add inotify watch descriptor for control group %s: %m", u->cgroup_path); | |
1388 | } | |
1389 | ||
1390 | r = hashmap_put(u->manager->cgroup_inotify_wd_unit, INT_TO_PTR(u->cgroup_inotify_wd), u); | |
1391 | if (r < 0) | |
1392 | return log_unit_error_errno(u, r, "Failed to add inotify watch descriptor to hash map: %m"); | |
1393 | ||
1394 | return 0; | |
1395 | } | |
1396 | ||
1397 | static int unit_create_cgroup( | |
1398 | Unit *u, | |
1399 | CGroupMask target_mask, | |
906c06f6 DM |
1400 | CGroupMask enable_mask, |
1401 | bool needs_bpf) { | |
efdb0237 | 1402 | |
0cd385d3 | 1403 | CGroupContext *c; |
bc432dc7 | 1404 | int r; |
64747e2d | 1405 | |
4ad49000 | 1406 | assert(u); |
64747e2d | 1407 | |
0cd385d3 LP |
1408 | c = unit_get_cgroup_context(u); |
1409 | if (!c) | |
1410 | return 0; | |
1411 | ||
7b3fd631 LP |
1412 | if (!u->cgroup_path) { |
1413 | _cleanup_free_ char *path = NULL; | |
64747e2d | 1414 | |
7b3fd631 LP |
1415 | path = unit_default_cgroup_path(u); |
1416 | if (!path) | |
1417 | return log_oom(); | |
1418 | ||
efdb0237 LP |
1419 | r = unit_set_cgroup_path(u, path); |
1420 | if (r == -EEXIST) | |
1421 | return log_unit_error_errno(u, r, "Control group %s exists already.", path); | |
1422 | if (r < 0) | |
1423 | return log_unit_error_errno(u, r, "Failed to set unit's control group path to %s: %m", path); | |
b58b8e11 HH |
1424 | } |
1425 | ||
03b90d4b | 1426 | /* First, create our own group */ |
efdb0237 | 1427 | r = cg_create_everywhere(u->manager->cgroup_supported, target_mask, u->cgroup_path); |
23bbb0de | 1428 | if (r < 0) |
efdb0237 LP |
1429 | return log_unit_error_errno(u, r, "Failed to create cgroup %s: %m", u->cgroup_path); |
1430 | ||
1431 | /* Start watching it */ | |
1432 | (void) unit_watch_cgroup(u); | |
1433 | ||
1434 | /* Enable all controllers we need */ | |
1435 | r = cg_enable_everywhere(u->manager->cgroup_supported, enable_mask, u->cgroup_path); | |
1436 | if (r < 0) | |
1437 | log_unit_warning_errno(u, r, "Failed to enable controllers on cgroup %s, ignoring: %m", u->cgroup_path); | |
03b90d4b LP |
1438 | |
1439 | /* Keep track that this is now realized */ | |
4ad49000 | 1440 | u->cgroup_realized = true; |
efdb0237 | 1441 | u->cgroup_realized_mask = target_mask; |
ccf78df1 | 1442 | u->cgroup_enabled_mask = enable_mask; |
906c06f6 | 1443 | u->cgroup_bpf_state = needs_bpf ? UNIT_CGROUP_BPF_ON : UNIT_CGROUP_BPF_OFF; |
4ad49000 | 1444 | |
0cd385d3 LP |
1445 | if (u->type != UNIT_SLICE && !c->delegate) { |
1446 | ||
1447 | /* Then, possibly move things over, but not if | |
1448 | * subgroups may contain processes, which is the case | |
1449 | * for slice and delegation units. */ | |
1450 | r = cg_migrate_everywhere(u->manager->cgroup_supported, u->cgroup_path, u->cgroup_path, migrate_callback, u); | |
1451 | if (r < 0) | |
efdb0237 | 1452 | log_unit_warning_errno(u, r, "Failed to migrate cgroup from to %s, ignoring: %m", u->cgroup_path); |
0cd385d3 | 1453 | } |
03b90d4b | 1454 | |
64747e2d LP |
1455 | return 0; |
1456 | } | |
1457 | ||
7b3fd631 LP |
1458 | int unit_attach_pids_to_cgroup(Unit *u) { |
1459 | int r; | |
1460 | assert(u); | |
1461 | ||
1462 | r = unit_realize_cgroup(u); | |
1463 | if (r < 0) | |
1464 | return r; | |
1465 | ||
1466 | r = cg_attach_many_everywhere(u->manager->cgroup_supported, u->cgroup_path, u->pids, migrate_callback, u); | |
1467 | if (r < 0) | |
1468 | return r; | |
1469 | ||
1470 | return 0; | |
1471 | } | |
1472 | ||
4b58153d LP |
1473 | static void cgroup_xattr_apply(Unit *u) { |
1474 | char ids[SD_ID128_STRING_MAX]; | |
1475 | int r; | |
1476 | ||
1477 | assert(u); | |
1478 | ||
1479 | if (!MANAGER_IS_SYSTEM(u->manager)) | |
1480 | return; | |
1481 | ||
1482 | if (sd_id128_is_null(u->invocation_id)) | |
1483 | return; | |
1484 | ||
1485 | r = cg_set_xattr(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path, | |
1486 | "trusted.invocation_id", | |
1487 | sd_id128_to_string(u->invocation_id, ids), 32, | |
1488 | 0); | |
1489 | if (r < 0) | |
0fb84499 | 1490 | log_unit_debug_errno(u, r, "Failed to set invocation ID on control group %s, ignoring: %m", u->cgroup_path); |
4b58153d LP |
1491 | } |
1492 | ||
906c06f6 DM |
1493 | static bool unit_has_mask_realized( |
1494 | Unit *u, | |
1495 | CGroupMask target_mask, | |
1496 | CGroupMask enable_mask, | |
1497 | bool needs_bpf) { | |
1498 | ||
bc432dc7 LP |
1499 | assert(u); |
1500 | ||
906c06f6 DM |
1501 | return u->cgroup_realized && |
1502 | u->cgroup_realized_mask == target_mask && | |
1503 | u->cgroup_enabled_mask == enable_mask && | |
1504 | ((needs_bpf && u->cgroup_bpf_state == UNIT_CGROUP_BPF_ON) || | |
1505 | (!needs_bpf && u->cgroup_bpf_state == UNIT_CGROUP_BPF_OFF)); | |
6414b7c9 DS |
1506 | } |
1507 | ||
2aa57a65 LP |
1508 | static void unit_add_to_cgroup_realize_queue(Unit *u) { |
1509 | assert(u); | |
1510 | ||
1511 | if (u->in_cgroup_realize_queue) | |
1512 | return; | |
1513 | ||
1514 | LIST_PREPEND(cgroup_realize_queue, u->manager->cgroup_realize_queue, u); | |
1515 | u->in_cgroup_realize_queue = true; | |
1516 | } | |
1517 | ||
1518 | static void unit_remove_from_cgroup_realize_queue(Unit *u) { | |
1519 | assert(u); | |
1520 | ||
1521 | if (!u->in_cgroup_realize_queue) | |
1522 | return; | |
1523 | ||
1524 | LIST_REMOVE(cgroup_realize_queue, u->manager->cgroup_realize_queue, u); | |
1525 | u->in_cgroup_realize_queue = false; | |
1526 | } | |
1527 | ||
1528 | ||
6414b7c9 DS |
1529 | /* Check if necessary controllers and attributes for a unit are in place. |
1530 | * | |
1531 | * If so, do nothing. | |
1532 | * If not, create paths, move processes over, and set attributes. | |
1533 | * | |
1534 | * Returns 0 on success and < 0 on failure. */ | |
db785129 | 1535 | static int unit_realize_cgroup_now(Unit *u, ManagerState state) { |
efdb0237 | 1536 | CGroupMask target_mask, enable_mask; |
906c06f6 | 1537 | bool needs_bpf, apply_bpf; |
6414b7c9 | 1538 | int r; |
64747e2d | 1539 | |
4ad49000 | 1540 | assert(u); |
64747e2d | 1541 | |
2aa57a65 | 1542 | unit_remove_from_cgroup_realize_queue(u); |
64747e2d | 1543 | |
efdb0237 | 1544 | target_mask = unit_get_target_mask(u); |
ccf78df1 | 1545 | enable_mask = unit_get_enable_mask(u); |
906c06f6 | 1546 | needs_bpf = unit_get_needs_bpf(u); |
ccf78df1 | 1547 | |
906c06f6 | 1548 | if (unit_has_mask_realized(u, target_mask, enable_mask, needs_bpf)) |
0a1eb06d | 1549 | return 0; |
64747e2d | 1550 | |
906c06f6 DM |
1551 | /* Make sure we apply the BPF filters either when one is configured, or if none is configured but previously |
1552 | * the state was anything but off. This way, if a unit with a BPF filter applied is reconfigured to lose it | |
1553 | * this will trickle down properly to cgroupfs. */ | |
1554 | apply_bpf = needs_bpf || u->cgroup_bpf_state != UNIT_CGROUP_BPF_OFF; | |
1555 | ||
4ad49000 | 1556 | /* First, realize parents */ |
6414b7c9 | 1557 | if (UNIT_ISSET(u->slice)) { |
db785129 | 1558 | r = unit_realize_cgroup_now(UNIT_DEREF(u->slice), state); |
6414b7c9 DS |
1559 | if (r < 0) |
1560 | return r; | |
1561 | } | |
4ad49000 LP |
1562 | |
1563 | /* And then do the real work */ | |
906c06f6 | 1564 | r = unit_create_cgroup(u, target_mask, enable_mask, needs_bpf); |
6414b7c9 DS |
1565 | if (r < 0) |
1566 | return r; | |
1567 | ||
1568 | /* Finally, apply the necessary attributes. */ | |
906c06f6 | 1569 | cgroup_context_apply(u, target_mask, apply_bpf, state); |
4b58153d | 1570 | cgroup_xattr_apply(u); |
6414b7c9 DS |
1571 | |
1572 | return 0; | |
64747e2d LP |
1573 | } |
1574 | ||
91a6073e | 1575 | unsigned manager_dispatch_cgroup_realize_queue(Manager *m) { |
db785129 | 1576 | ManagerState state; |
4ad49000 | 1577 | unsigned n = 0; |
db785129 | 1578 | Unit *i; |
6414b7c9 | 1579 | int r; |
ecedd90f | 1580 | |
91a6073e LP |
1581 | assert(m); |
1582 | ||
db785129 LP |
1583 | state = manager_state(m); |
1584 | ||
91a6073e LP |
1585 | while ((i = m->cgroup_realize_queue)) { |
1586 | assert(i->in_cgroup_realize_queue); | |
ecedd90f | 1587 | |
2aa57a65 LP |
1588 | if (UNIT_IS_INACTIVE_OR_FAILED(unit_active_state(i))) { |
1589 | /* Maybe things changed, and the unit is not actually active anymore? */ | |
1590 | unit_remove_from_cgroup_realize_queue(i); | |
1591 | continue; | |
1592 | } | |
1593 | ||
db785129 | 1594 | r = unit_realize_cgroup_now(i, state); |
6414b7c9 | 1595 | if (r < 0) |
efdb0237 | 1596 | log_warning_errno(r, "Failed to realize cgroups for queued unit %s, ignoring: %m", i->id); |
0a1eb06d | 1597 | |
4ad49000 LP |
1598 | n++; |
1599 | } | |
ecedd90f | 1600 | |
4ad49000 | 1601 | return n; |
8e274523 LP |
1602 | } |
1603 | ||
91a6073e | 1604 | static void unit_add_siblings_to_cgroup_realize_queue(Unit *u) { |
4ad49000 | 1605 | Unit *slice; |
ca949c9d | 1606 | |
4ad49000 LP |
1607 | /* This adds the siblings of the specified unit and the |
1608 | * siblings of all parent units to the cgroup queue. (But | |
1609 | * neither the specified unit itself nor the parents.) */ | |
1610 | ||
1611 | while ((slice = UNIT_DEREF(u->slice))) { | |
1612 | Iterator i; | |
1613 | Unit *m; | |
eef85c4a | 1614 | void *v; |
8f53a7b8 | 1615 | |
eef85c4a | 1616 | HASHMAP_FOREACH_KEY(v, m, u->dependencies[UNIT_BEFORE], i) { |
4ad49000 LP |
1617 | if (m == u) |
1618 | continue; | |
8e274523 | 1619 | |
6414b7c9 DS |
1620 | /* Skip units that have a dependency on the slice |
1621 | * but aren't actually in it. */ | |
4ad49000 | 1622 | if (UNIT_DEREF(m->slice) != slice) |
50159e6a | 1623 | continue; |
8e274523 | 1624 | |
6414b7c9 DS |
1625 | /* No point in doing cgroup application for units |
1626 | * without active processes. */ | |
1627 | if (UNIT_IS_INACTIVE_OR_FAILED(unit_active_state(m))) | |
1628 | continue; | |
1629 | ||
1630 | /* If the unit doesn't need any new controllers | |
1631 | * and has current ones realized, it doesn't need | |
1632 | * any changes. */ | |
906c06f6 DM |
1633 | if (unit_has_mask_realized(m, |
1634 | unit_get_target_mask(m), | |
1635 | unit_get_enable_mask(m), | |
1636 | unit_get_needs_bpf(m))) | |
6414b7c9 DS |
1637 | continue; |
1638 | ||
91a6073e | 1639 | unit_add_to_cgroup_realize_queue(m); |
50159e6a LP |
1640 | } |
1641 | ||
4ad49000 | 1642 | u = slice; |
8e274523 | 1643 | } |
4ad49000 LP |
1644 | } |
1645 | ||
0a1eb06d | 1646 | int unit_realize_cgroup(Unit *u) { |
4ad49000 LP |
1647 | assert(u); |
1648 | ||
35b7ff80 | 1649 | if (!UNIT_HAS_CGROUP_CONTEXT(u)) |
0a1eb06d | 1650 | return 0; |
8e274523 | 1651 | |
4ad49000 LP |
1652 | /* So, here's the deal: when realizing the cgroups for this |
1653 | * unit, we need to first create all parents, but there's more | |
1654 | * actually: for the weight-based controllers we also need to | |
1655 | * make sure that all our siblings (i.e. units that are in the | |
73e231ab | 1656 | * same slice as we are) have cgroups, too. Otherwise, things |
4ad49000 LP |
1657 | * would become very uneven as each of their processes would |
1658 | * get as much resources as all our group together. This call | |
1659 | * will synchronously create the parent cgroups, but will | |
1660 | * defer work on the siblings to the next event loop | |
1661 | * iteration. */ | |
ca949c9d | 1662 | |
4ad49000 | 1663 | /* Add all sibling slices to the cgroup queue. */ |
91a6073e | 1664 | unit_add_siblings_to_cgroup_realize_queue(u); |
4ad49000 | 1665 | |
6414b7c9 | 1666 | /* And realize this one now (and apply the values) */ |
db785129 | 1667 | return unit_realize_cgroup_now(u, manager_state(u->manager)); |
8e274523 LP |
1668 | } |
1669 | ||
efdb0237 LP |
1670 | void unit_release_cgroup(Unit *u) { |
1671 | assert(u); | |
1672 | ||
1673 | /* Forgets all cgroup details for this cgroup */ | |
1674 | ||
1675 | if (u->cgroup_path) { | |
1676 | (void) hashmap_remove(u->manager->cgroup_unit, u->cgroup_path); | |
1677 | u->cgroup_path = mfree(u->cgroup_path); | |
1678 | } | |
1679 | ||
1680 | if (u->cgroup_inotify_wd >= 0) { | |
1681 | if (inotify_rm_watch(u->manager->cgroup_inotify_fd, u->cgroup_inotify_wd) < 0) | |
1682 | log_unit_debug_errno(u, errno, "Failed to remove cgroup inotify watch %i for %s, ignoring", u->cgroup_inotify_wd, u->id); | |
1683 | ||
1684 | (void) hashmap_remove(u->manager->cgroup_inotify_wd_unit, INT_TO_PTR(u->cgroup_inotify_wd)); | |
1685 | u->cgroup_inotify_wd = -1; | |
1686 | } | |
1687 | } | |
1688 | ||
1689 | void unit_prune_cgroup(Unit *u) { | |
8e274523 | 1690 | int r; |
efdb0237 | 1691 | bool is_root_slice; |
8e274523 | 1692 | |
4ad49000 | 1693 | assert(u); |
8e274523 | 1694 | |
efdb0237 LP |
1695 | /* Removes the cgroup, if empty and possible, and stops watching it. */ |
1696 | ||
4ad49000 LP |
1697 | if (!u->cgroup_path) |
1698 | return; | |
8e274523 | 1699 | |
fe700f46 LP |
1700 | (void) unit_get_cpu_usage(u, NULL); /* Cache the last CPU usage value before we destroy the cgroup */ |
1701 | ||
efdb0237 LP |
1702 | is_root_slice = unit_has_name(u, SPECIAL_ROOT_SLICE); |
1703 | ||
1704 | r = cg_trim_everywhere(u->manager->cgroup_supported, u->cgroup_path, !is_root_slice); | |
dab5bf85 | 1705 | if (r < 0) { |
f29ff115 | 1706 | log_unit_debug_errno(u, r, "Failed to destroy cgroup %s, ignoring: %m", u->cgroup_path); |
dab5bf85 RL |
1707 | return; |
1708 | } | |
8e274523 | 1709 | |
efdb0237 LP |
1710 | if (is_root_slice) |
1711 | return; | |
1712 | ||
1713 | unit_release_cgroup(u); | |
0a1eb06d | 1714 | |
4ad49000 | 1715 | u->cgroup_realized = false; |
bc432dc7 | 1716 | u->cgroup_realized_mask = 0; |
ccf78df1 | 1717 | u->cgroup_enabled_mask = 0; |
8e274523 LP |
1718 | } |
1719 | ||
efdb0237 | 1720 | int unit_search_main_pid(Unit *u, pid_t *ret) { |
4ad49000 LP |
1721 | _cleanup_fclose_ FILE *f = NULL; |
1722 | pid_t pid = 0, npid, mypid; | |
efdb0237 | 1723 | int r; |
4ad49000 LP |
1724 | |
1725 | assert(u); | |
efdb0237 | 1726 | assert(ret); |
4ad49000 LP |
1727 | |
1728 | if (!u->cgroup_path) | |
efdb0237 | 1729 | return -ENXIO; |
4ad49000 | 1730 | |
efdb0237 LP |
1731 | r = cg_enumerate_processes(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path, &f); |
1732 | if (r < 0) | |
1733 | return r; | |
4ad49000 | 1734 | |
df0ff127 | 1735 | mypid = getpid_cached(); |
4ad49000 LP |
1736 | while (cg_read_pid(f, &npid) > 0) { |
1737 | pid_t ppid; | |
1738 | ||
1739 | if (npid == pid) | |
1740 | continue; | |
8e274523 | 1741 | |
4ad49000 | 1742 | /* Ignore processes that aren't our kids */ |
6bc73acb | 1743 | if (get_process_ppid(npid, &ppid) >= 0 && ppid != mypid) |
4ad49000 | 1744 | continue; |
8e274523 | 1745 | |
efdb0237 | 1746 | if (pid != 0) |
4ad49000 LP |
1747 | /* Dang, there's more than one daemonized PID |
1748 | in this group, so we don't know what process | |
1749 | is the main process. */ | |
efdb0237 LP |
1750 | |
1751 | return -ENODATA; | |
8e274523 | 1752 | |
4ad49000 | 1753 | pid = npid; |
8e274523 LP |
1754 | } |
1755 | ||
efdb0237 LP |
1756 | *ret = pid; |
1757 | return 0; | |
1758 | } | |
1759 | ||
1760 | static int unit_watch_pids_in_path(Unit *u, const char *path) { | |
b3c5bad3 | 1761 | _cleanup_closedir_ DIR *d = NULL; |
efdb0237 LP |
1762 | _cleanup_fclose_ FILE *f = NULL; |
1763 | int ret = 0, r; | |
1764 | ||
1765 | assert(u); | |
1766 | assert(path); | |
1767 | ||
1768 | r = cg_enumerate_processes(SYSTEMD_CGROUP_CONTROLLER, path, &f); | |
1769 | if (r < 0) | |
1770 | ret = r; | |
1771 | else { | |
1772 | pid_t pid; | |
1773 | ||
1774 | while ((r = cg_read_pid(f, &pid)) > 0) { | |
1775 | r = unit_watch_pid(u, pid); | |
1776 | if (r < 0 && ret >= 0) | |
1777 | ret = r; | |
1778 | } | |
1779 | ||
1780 | if (r < 0 && ret >= 0) | |
1781 | ret = r; | |
1782 | } | |
1783 | ||
1784 | r = cg_enumerate_subgroups(SYSTEMD_CGROUP_CONTROLLER, path, &d); | |
1785 | if (r < 0) { | |
1786 | if (ret >= 0) | |
1787 | ret = r; | |
1788 | } else { | |
1789 | char *fn; | |
1790 | ||
1791 | while ((r = cg_read_subgroup(d, &fn)) > 0) { | |
1792 | _cleanup_free_ char *p = NULL; | |
1793 | ||
605405c6 | 1794 | p = strjoin(path, "/", fn); |
efdb0237 LP |
1795 | free(fn); |
1796 | ||
1797 | if (!p) | |
1798 | return -ENOMEM; | |
1799 | ||
1800 | r = unit_watch_pids_in_path(u, p); | |
1801 | if (r < 0 && ret >= 0) | |
1802 | ret = r; | |
1803 | } | |
1804 | ||
1805 | if (r < 0 && ret >= 0) | |
1806 | ret = r; | |
1807 | } | |
1808 | ||
1809 | return ret; | |
1810 | } | |
1811 | ||
1812 | int unit_watch_all_pids(Unit *u) { | |
b4cccbc1 LP |
1813 | int r; |
1814 | ||
efdb0237 LP |
1815 | assert(u); |
1816 | ||
1817 | /* Adds all PIDs from our cgroup to the set of PIDs we | |
1818 | * watch. This is a fallback logic for cases where we do not | |
1819 | * get reliable cgroup empty notifications: we try to use | |
1820 | * SIGCHLD as replacement. */ | |
1821 | ||
1822 | if (!u->cgroup_path) | |
1823 | return -ENOENT; | |
1824 | ||
c22800e4 | 1825 | r = cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER); |
b4cccbc1 LP |
1826 | if (r < 0) |
1827 | return r; | |
1828 | if (r > 0) /* On unified we can use proper notifications */ | |
efdb0237 LP |
1829 | return 0; |
1830 | ||
1831 | return unit_watch_pids_in_path(u, u->cgroup_path); | |
1832 | } | |
1833 | ||
09e24654 LP |
1834 | static int on_cgroup_empty_event(sd_event_source *s, void *userdata) { |
1835 | Manager *m = userdata; | |
1836 | Unit *u; | |
efdb0237 LP |
1837 | int r; |
1838 | ||
09e24654 LP |
1839 | assert(s); |
1840 | assert(m); | |
efdb0237 | 1841 | |
09e24654 LP |
1842 | u = m->cgroup_empty_queue; |
1843 | if (!u) | |
efdb0237 LP |
1844 | return 0; |
1845 | ||
09e24654 LP |
1846 | assert(u->in_cgroup_empty_queue); |
1847 | u->in_cgroup_empty_queue = false; | |
1848 | LIST_REMOVE(cgroup_empty_queue, m->cgroup_empty_queue, u); | |
1849 | ||
1850 | if (m->cgroup_empty_queue) { | |
1851 | /* More stuff queued, let's make sure we remain enabled */ | |
1852 | r = sd_event_source_set_enabled(s, SD_EVENT_ONESHOT); | |
1853 | if (r < 0) | |
1854 | log_debug_errno(r, "Failed to reenable cgroup empty event source: %m"); | |
1855 | } | |
efdb0237 LP |
1856 | |
1857 | unit_add_to_gc_queue(u); | |
1858 | ||
1859 | if (UNIT_VTABLE(u)->notify_cgroup_empty) | |
1860 | UNIT_VTABLE(u)->notify_cgroup_empty(u); | |
1861 | ||
1862 | return 0; | |
1863 | } | |
1864 | ||
09e24654 LP |
1865 | void unit_add_to_cgroup_empty_queue(Unit *u) { |
1866 | int r; | |
1867 | ||
1868 | assert(u); | |
1869 | ||
1870 | /* Note that there are four different ways how cgroup empty events reach us: | |
1871 | * | |
1872 | * 1. On the unified hierarchy we get an inotify event on the cgroup | |
1873 | * | |
1874 | * 2. On the legacy hierarchy, when running in system mode, we get a datagram on the cgroup agent socket | |
1875 | * | |
1876 | * 3. On the legacy hierarchy, when running in user mode, we get a D-Bus signal on the system bus | |
1877 | * | |
1878 | * 4. On the legacy hierarchy, in service units we start watching all processes of the cgroup for SIGCHLD as | |
1879 | * soon as we get one SIGCHLD, to deal with unreliable cgroup notifications. | |
1880 | * | |
1881 | * Regardless which way we got the notification, we'll verify it here, and then add it to a separate | |
1882 | * queue. This queue will be dispatched at a lower priority than the SIGCHLD handler, so that we always use | |
1883 | * SIGCHLD if we can get it first, and only use the cgroup empty notifications if there's no SIGCHLD pending | |
1884 | * (which might happen if the cgroup doesn't contain processes that are our own child, which is typically the | |
1885 | * case for scope units). */ | |
1886 | ||
1887 | if (u->in_cgroup_empty_queue) | |
1888 | return; | |
1889 | ||
1890 | /* Let's verify that the cgroup is really empty */ | |
1891 | if (!u->cgroup_path) | |
1892 | return; | |
1893 | r = cg_is_empty_recursive(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path); | |
1894 | if (r < 0) { | |
1895 | log_unit_debug_errno(u, r, "Failed to determine whether cgroup %s is empty: %m", u->cgroup_path); | |
1896 | return; | |
1897 | } | |
1898 | if (r == 0) | |
1899 | return; | |
1900 | ||
1901 | LIST_PREPEND(cgroup_empty_queue, u->manager->cgroup_empty_queue, u); | |
1902 | u->in_cgroup_empty_queue = true; | |
1903 | ||
1904 | /* Trigger the defer event */ | |
1905 | r = sd_event_source_set_enabled(u->manager->cgroup_empty_event_source, SD_EVENT_ONESHOT); | |
1906 | if (r < 0) | |
1907 | log_debug_errno(r, "Failed to enable cgroup empty event source: %m"); | |
1908 | } | |
1909 | ||
efdb0237 LP |
1910 | static int on_cgroup_inotify_event(sd_event_source *s, int fd, uint32_t revents, void *userdata) { |
1911 | Manager *m = userdata; | |
1912 | ||
1913 | assert(s); | |
1914 | assert(fd >= 0); | |
1915 | assert(m); | |
1916 | ||
1917 | for (;;) { | |
1918 | union inotify_event_buffer buffer; | |
1919 | struct inotify_event *e; | |
1920 | ssize_t l; | |
1921 | ||
1922 | l = read(fd, &buffer, sizeof(buffer)); | |
1923 | if (l < 0) { | |
47249640 | 1924 | if (IN_SET(errno, EINTR, EAGAIN)) |
efdb0237 LP |
1925 | return 0; |
1926 | ||
1927 | return log_error_errno(errno, "Failed to read control group inotify events: %m"); | |
1928 | } | |
1929 | ||
1930 | FOREACH_INOTIFY_EVENT(e, buffer, l) { | |
1931 | Unit *u; | |
1932 | ||
1933 | if (e->wd < 0) | |
1934 | /* Queue overflow has no watch descriptor */ | |
1935 | continue; | |
1936 | ||
1937 | if (e->mask & IN_IGNORED) | |
1938 | /* The watch was just removed */ | |
1939 | continue; | |
1940 | ||
1941 | u = hashmap_get(m->cgroup_inotify_wd_unit, INT_TO_PTR(e->wd)); | |
1942 | if (!u) /* Not that inotify might deliver | |
1943 | * events for a watch even after it | |
1944 | * was removed, because it was queued | |
1945 | * before the removal. Let's ignore | |
1946 | * this here safely. */ | |
1947 | continue; | |
1948 | ||
09e24654 | 1949 | unit_add_to_cgroup_empty_queue(u); |
efdb0237 LP |
1950 | } |
1951 | } | |
8e274523 LP |
1952 | } |
1953 | ||
8e274523 | 1954 | int manager_setup_cgroup(Manager *m) { |
9444b1f2 | 1955 | _cleanup_free_ char *path = NULL; |
10bd3e2e | 1956 | const char *scope_path; |
efdb0237 | 1957 | CGroupController c; |
b4cccbc1 | 1958 | int r, all_unified; |
efdb0237 | 1959 | char *e; |
8e274523 LP |
1960 | |
1961 | assert(m); | |
1962 | ||
35d2e7ec | 1963 | /* 1. Determine hierarchy */ |
efdb0237 | 1964 | m->cgroup_root = mfree(m->cgroup_root); |
9444b1f2 | 1965 | r = cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER, 0, &m->cgroup_root); |
23bbb0de MS |
1966 | if (r < 0) |
1967 | return log_error_errno(r, "Cannot determine cgroup we are running in: %m"); | |
8e274523 | 1968 | |
efdb0237 LP |
1969 | /* Chop off the init scope, if we are already located in it */ |
1970 | e = endswith(m->cgroup_root, "/" SPECIAL_INIT_SCOPE); | |
0d8c31ff | 1971 | |
efdb0237 LP |
1972 | /* LEGACY: Also chop off the system slice if we are in |
1973 | * it. This is to support live upgrades from older systemd | |
1974 | * versions where PID 1 was moved there. Also see | |
1975 | * cg_get_root_path(). */ | |
463d0d15 | 1976 | if (!e && MANAGER_IS_SYSTEM(m)) { |
9444b1f2 | 1977 | e = endswith(m->cgroup_root, "/" SPECIAL_SYSTEM_SLICE); |
15c60e99 | 1978 | if (!e) |
efdb0237 | 1979 | e = endswith(m->cgroup_root, "/system"); /* even more legacy */ |
0baf24dd | 1980 | } |
efdb0237 LP |
1981 | if (e) |
1982 | *e = 0; | |
7ccfb64a | 1983 | |
7546145e LP |
1984 | /* And make sure to store away the root value without trailing slash, even for the root dir, so that we can |
1985 | * easily prepend it everywhere. */ | |
1986 | delete_trailing_chars(m->cgroup_root, "/"); | |
8e274523 | 1987 | |
35d2e7ec | 1988 | /* 2. Show data */ |
9444b1f2 | 1989 | r = cg_get_path(SYSTEMD_CGROUP_CONTROLLER, m->cgroup_root, NULL, &path); |
23bbb0de MS |
1990 | if (r < 0) |
1991 | return log_error_errno(r, "Cannot find cgroup mount point: %m"); | |
8e274523 | 1992 | |
415fc41c TH |
1993 | r = cg_unified_flush(); |
1994 | if (r < 0) | |
1995 | return log_error_errno(r, "Couldn't determine if we are running in the unified hierarchy: %m"); | |
5da38d07 | 1996 | |
b4cccbc1 | 1997 | all_unified = cg_all_unified(); |
d4c819ed ZJS |
1998 | if (all_unified < 0) |
1999 | return log_error_errno(all_unified, "Couldn't determine whether we are in all unified mode: %m"); | |
2000 | if (all_unified > 0) | |
efdb0237 | 2001 | log_debug("Unified cgroup hierarchy is located at %s.", path); |
b4cccbc1 | 2002 | else { |
c22800e4 | 2003 | r = cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER); |
b4cccbc1 LP |
2004 | if (r < 0) |
2005 | return log_error_errno(r, "Failed to determine whether systemd's own controller is in unified mode: %m"); | |
2006 | if (r > 0) | |
2007 | log_debug("Unified cgroup hierarchy is located at %s. Controllers are on legacy hierarchies.", path); | |
2008 | else | |
2009 | log_debug("Using cgroup controller " SYSTEMD_CGROUP_CONTROLLER_LEGACY ". File system hierarchy is at %s.", path); | |
2010 | } | |
efdb0237 | 2011 | |
09e24654 LP |
2012 | /* 3. Allocate cgroup empty defer event source */ |
2013 | m->cgroup_empty_event_source = sd_event_source_unref(m->cgroup_empty_event_source); | |
2014 | r = sd_event_add_defer(m->event, &m->cgroup_empty_event_source, on_cgroup_empty_event, m); | |
2015 | if (r < 0) | |
2016 | return log_error_errno(r, "Failed to create cgroup empty event source: %m"); | |
2017 | ||
2018 | r = sd_event_source_set_priority(m->cgroup_empty_event_source, SD_EVENT_PRIORITY_NORMAL-5); | |
2019 | if (r < 0) | |
2020 | return log_error_errno(r, "Failed to set priority of cgroup empty event source: %m"); | |
2021 | ||
2022 | r = sd_event_source_set_enabled(m->cgroup_empty_event_source, SD_EVENT_OFF); | |
2023 | if (r < 0) | |
2024 | return log_error_errno(r, "Failed to disable cgroup empty event source: %m"); | |
2025 | ||
2026 | (void) sd_event_source_set_description(m->cgroup_empty_event_source, "cgroup-empty"); | |
2027 | ||
2028 | /* 4. Install notifier inotify object, or agent */ | |
10bd3e2e | 2029 | if (cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER) > 0) { |
c6c18be3 | 2030 | |
09e24654 | 2031 | /* In the unified hierarchy we can get cgroup empty notifications via inotify. */ |
efdb0237 | 2032 | |
10bd3e2e LP |
2033 | m->cgroup_inotify_event_source = sd_event_source_unref(m->cgroup_inotify_event_source); |
2034 | safe_close(m->cgroup_inotify_fd); | |
efdb0237 | 2035 | |
10bd3e2e LP |
2036 | m->cgroup_inotify_fd = inotify_init1(IN_NONBLOCK|IN_CLOEXEC); |
2037 | if (m->cgroup_inotify_fd < 0) | |
2038 | return log_error_errno(errno, "Failed to create control group inotify object: %m"); | |
efdb0237 | 2039 | |
10bd3e2e LP |
2040 | r = sd_event_add_io(m->event, &m->cgroup_inotify_event_source, m->cgroup_inotify_fd, EPOLLIN, on_cgroup_inotify_event, m); |
2041 | if (r < 0) | |
2042 | return log_error_errno(r, "Failed to watch control group inotify object: %m"); | |
efdb0237 | 2043 | |
10bd3e2e LP |
2044 | /* Process cgroup empty notifications early, but after service notifications and SIGCHLD. Also |
2045 | * see handling of cgroup agent notifications, for the classic cgroup hierarchy support. */ | |
09e24654 | 2046 | r = sd_event_source_set_priority(m->cgroup_inotify_event_source, SD_EVENT_PRIORITY_NORMAL-4); |
10bd3e2e LP |
2047 | if (r < 0) |
2048 | return log_error_errno(r, "Failed to set priority of inotify event source: %m"); | |
efdb0237 | 2049 | |
10bd3e2e | 2050 | (void) sd_event_source_set_description(m->cgroup_inotify_event_source, "cgroup-inotify"); |
efdb0237 | 2051 | |
10bd3e2e | 2052 | } else if (MANAGER_IS_SYSTEM(m) && m->test_run_flags == 0) { |
efdb0237 | 2053 | |
10bd3e2e LP |
2054 | /* On the legacy hierarchy we only get notifications via cgroup agents. (Which isn't really reliable, |
2055 | * since it does not generate events when control groups with children run empty. */ | |
8e274523 | 2056 | |
10bd3e2e | 2057 | r = cg_install_release_agent(SYSTEMD_CGROUP_CONTROLLER, SYSTEMD_CGROUP_AGENT_PATH); |
23bbb0de | 2058 | if (r < 0) |
10bd3e2e LP |
2059 | log_warning_errno(r, "Failed to install release agent, ignoring: %m"); |
2060 | else if (r > 0) | |
2061 | log_debug("Installed release agent."); | |
2062 | else if (r == 0) | |
2063 | log_debug("Release agent already installed."); | |
2064 | } | |
efdb0237 | 2065 | |
09e24654 | 2066 | /* 5. Make sure we are in the special "init.scope" unit in the root slice. */ |
10bd3e2e LP |
2067 | scope_path = strjoina(m->cgroup_root, "/" SPECIAL_INIT_SCOPE); |
2068 | r = cg_create_and_attach(SYSTEMD_CGROUP_CONTROLLER, scope_path, 0); | |
2069 | if (r < 0) | |
2070 | return log_error_errno(r, "Failed to create %s control group: %m", scope_path); | |
c6c18be3 | 2071 | |
09e24654 | 2072 | /* Also, move all other userspace processes remaining in the root cgroup into that scope. */ |
10bd3e2e LP |
2073 | r = cg_migrate(SYSTEMD_CGROUP_CONTROLLER, m->cgroup_root, SYSTEMD_CGROUP_CONTROLLER, scope_path, 0); |
2074 | if (r < 0) | |
2075 | log_warning_errno(r, "Couldn't move remaining userspace processes, ignoring: %m"); | |
0d8c31ff | 2076 | |
09e24654 | 2077 | /* 6. And pin it, so that it cannot be unmounted */ |
10bd3e2e LP |
2078 | safe_close(m->pin_cgroupfs_fd); |
2079 | m->pin_cgroupfs_fd = open(path, O_RDONLY|O_CLOEXEC|O_DIRECTORY|O_NOCTTY|O_NONBLOCK); | |
2080 | if (m->pin_cgroupfs_fd < 0) | |
2081 | return log_error_errno(errno, "Failed to open pin file: %m"); | |
2082 | ||
09e24654 | 2083 | /* 7. Always enable hierarchical support if it exists... */ |
10bd3e2e LP |
2084 | if (!all_unified && m->test_run_flags == 0) |
2085 | (void) cg_set_attribute("memory", "/", "memory.use_hierarchy", "1"); | |
c6c18be3 | 2086 | |
09e24654 | 2087 | /* 8. Figure out which controllers are supported, and log about it */ |
efdb0237 LP |
2088 | r = cg_mask_supported(&m->cgroup_supported); |
2089 | if (r < 0) | |
2090 | return log_error_errno(r, "Failed to determine supported controllers: %m"); | |
efdb0237 | 2091 | for (c = 0; c < _CGROUP_CONTROLLER_MAX; c++) |
eee0a1e4 | 2092 | log_debug("Controller '%s' supported: %s", cgroup_controller_to_string(c), yes_no(m->cgroup_supported & CGROUP_CONTROLLER_TO_MASK(c))); |
9156e799 | 2093 | |
a32360f1 | 2094 | return 0; |
8e274523 LP |
2095 | } |
2096 | ||
c6c18be3 | 2097 | void manager_shutdown_cgroup(Manager *m, bool delete) { |
8e274523 LP |
2098 | assert(m); |
2099 | ||
9444b1f2 LP |
2100 | /* We can't really delete the group, since we are in it. But |
2101 | * let's trim it. */ | |
2102 | if (delete && m->cgroup_root) | |
efdb0237 LP |
2103 | (void) cg_trim(SYSTEMD_CGROUP_CONTROLLER, m->cgroup_root, false); |
2104 | ||
09e24654 LP |
2105 | m->cgroup_empty_event_source = sd_event_source_unref(m->cgroup_empty_event_source); |
2106 | ||
efdb0237 LP |
2107 | m->cgroup_inotify_wd_unit = hashmap_free(m->cgroup_inotify_wd_unit); |
2108 | ||
2109 | m->cgroup_inotify_event_source = sd_event_source_unref(m->cgroup_inotify_event_source); | |
2110 | m->cgroup_inotify_fd = safe_close(m->cgroup_inotify_fd); | |
8e274523 | 2111 | |
03e334a1 | 2112 | m->pin_cgroupfs_fd = safe_close(m->pin_cgroupfs_fd); |
c6c18be3 | 2113 | |
efdb0237 | 2114 | m->cgroup_root = mfree(m->cgroup_root); |
8e274523 LP |
2115 | } |
2116 | ||
4ad49000 | 2117 | Unit* manager_get_unit_by_cgroup(Manager *m, const char *cgroup) { |
acb14d31 | 2118 | char *p; |
4ad49000 | 2119 | Unit *u; |
acb14d31 LP |
2120 | |
2121 | assert(m); | |
2122 | assert(cgroup); | |
acb14d31 | 2123 | |
4ad49000 LP |
2124 | u = hashmap_get(m->cgroup_unit, cgroup); |
2125 | if (u) | |
2126 | return u; | |
acb14d31 | 2127 | |
8e70580b | 2128 | p = strdupa(cgroup); |
acb14d31 LP |
2129 | for (;;) { |
2130 | char *e; | |
2131 | ||
2132 | e = strrchr(p, '/'); | |
efdb0237 LP |
2133 | if (!e || e == p) |
2134 | return hashmap_get(m->cgroup_unit, SPECIAL_ROOT_SLICE); | |
acb14d31 LP |
2135 | |
2136 | *e = 0; | |
2137 | ||
4ad49000 LP |
2138 | u = hashmap_get(m->cgroup_unit, p); |
2139 | if (u) | |
2140 | return u; | |
acb14d31 LP |
2141 | } |
2142 | } | |
2143 | ||
b3ac818b | 2144 | Unit *manager_get_unit_by_pid_cgroup(Manager *m, pid_t pid) { |
4ad49000 | 2145 | _cleanup_free_ char *cgroup = NULL; |
acb14d31 | 2146 | int r; |
8e274523 | 2147 | |
8c47c732 LP |
2148 | assert(m); |
2149 | ||
b3ac818b LP |
2150 | if (pid <= 0) |
2151 | return NULL; | |
2152 | ||
2153 | r = cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER, pid, &cgroup); | |
2154 | if (r < 0) | |
2155 | return NULL; | |
2156 | ||
2157 | return manager_get_unit_by_cgroup(m, cgroup); | |
2158 | } | |
2159 | ||
2160 | Unit *manager_get_unit_by_pid(Manager *m, pid_t pid) { | |
2161 | Unit *u; | |
2162 | ||
2163 | assert(m); | |
2164 | ||
efdb0237 | 2165 | if (pid <= 0) |
8c47c732 LP |
2166 | return NULL; |
2167 | ||
efdb0237 LP |
2168 | if (pid == 1) |
2169 | return hashmap_get(m->units, SPECIAL_INIT_SCOPE); | |
2170 | ||
fea72cc0 | 2171 | u = hashmap_get(m->watch_pids1, PID_TO_PTR(pid)); |
5fe8876b LP |
2172 | if (u) |
2173 | return u; | |
2174 | ||
fea72cc0 | 2175 | u = hashmap_get(m->watch_pids2, PID_TO_PTR(pid)); |
5fe8876b LP |
2176 | if (u) |
2177 | return u; | |
2178 | ||
b3ac818b | 2179 | return manager_get_unit_by_pid_cgroup(m, pid); |
6dde1f33 | 2180 | } |
4fbf50b3 | 2181 | |
4ad49000 LP |
2182 | int manager_notify_cgroup_empty(Manager *m, const char *cgroup) { |
2183 | Unit *u; | |
4fbf50b3 | 2184 | |
4ad49000 LP |
2185 | assert(m); |
2186 | assert(cgroup); | |
4fbf50b3 | 2187 | |
09e24654 LP |
2188 | /* Called on the legacy hierarchy whenever we get an explicit cgroup notification from the cgroup agent process |
2189 | * or from the --system instance */ | |
2190 | ||
d8fdc620 LP |
2191 | log_debug("Got cgroup empty notification for: %s", cgroup); |
2192 | ||
4ad49000 | 2193 | u = manager_get_unit_by_cgroup(m, cgroup); |
5ad096b3 LP |
2194 | if (!u) |
2195 | return 0; | |
b56c28c3 | 2196 | |
09e24654 LP |
2197 | unit_add_to_cgroup_empty_queue(u); |
2198 | return 1; | |
5ad096b3 LP |
2199 | } |
2200 | ||
2201 | int unit_get_memory_current(Unit *u, uint64_t *ret) { | |
2202 | _cleanup_free_ char *v = NULL; | |
2203 | int r; | |
2204 | ||
2205 | assert(u); | |
2206 | assert(ret); | |
2207 | ||
2e4025c0 | 2208 | if (!UNIT_CGROUP_BOOL(u, memory_accounting)) |
cf3b4be1 LP |
2209 | return -ENODATA; |
2210 | ||
5ad096b3 LP |
2211 | if (!u->cgroup_path) |
2212 | return -ENODATA; | |
2213 | ||
efdb0237 | 2214 | if ((u->cgroup_realized_mask & CGROUP_MASK_MEMORY) == 0) |
5ad096b3 LP |
2215 | return -ENODATA; |
2216 | ||
b4cccbc1 LP |
2217 | r = cg_all_unified(); |
2218 | if (r < 0) | |
2219 | return r; | |
2220 | if (r > 0) | |
efdb0237 | 2221 | r = cg_get_attribute("memory", u->cgroup_path, "memory.current", &v); |
b4cccbc1 LP |
2222 | else |
2223 | r = cg_get_attribute("memory", u->cgroup_path, "memory.usage_in_bytes", &v); | |
5ad096b3 LP |
2224 | if (r == -ENOENT) |
2225 | return -ENODATA; | |
2226 | if (r < 0) | |
2227 | return r; | |
2228 | ||
2229 | return safe_atou64(v, ret); | |
2230 | } | |
2231 | ||
03a7b521 LP |
2232 | int unit_get_tasks_current(Unit *u, uint64_t *ret) { |
2233 | _cleanup_free_ char *v = NULL; | |
2234 | int r; | |
2235 | ||
2236 | assert(u); | |
2237 | assert(ret); | |
2238 | ||
2e4025c0 | 2239 | if (!UNIT_CGROUP_BOOL(u, tasks_accounting)) |
cf3b4be1 LP |
2240 | return -ENODATA; |
2241 | ||
03a7b521 LP |
2242 | if (!u->cgroup_path) |
2243 | return -ENODATA; | |
2244 | ||
2245 | if ((u->cgroup_realized_mask & CGROUP_MASK_PIDS) == 0) | |
2246 | return -ENODATA; | |
2247 | ||
2248 | r = cg_get_attribute("pids", u->cgroup_path, "pids.current", &v); | |
2249 | if (r == -ENOENT) | |
2250 | return -ENODATA; | |
2251 | if (r < 0) | |
2252 | return r; | |
2253 | ||
2254 | return safe_atou64(v, ret); | |
2255 | } | |
2256 | ||
5ad096b3 LP |
2257 | static int unit_get_cpu_usage_raw(Unit *u, nsec_t *ret) { |
2258 | _cleanup_free_ char *v = NULL; | |
2259 | uint64_t ns; | |
2260 | int r; | |
2261 | ||
2262 | assert(u); | |
2263 | assert(ret); | |
2264 | ||
2265 | if (!u->cgroup_path) | |
2266 | return -ENODATA; | |
2267 | ||
b4cccbc1 LP |
2268 | r = cg_all_unified(); |
2269 | if (r < 0) | |
2270 | return r; | |
2271 | if (r > 0) { | |
66ebf6c0 TH |
2272 | const char *keys[] = { "usage_usec", NULL }; |
2273 | _cleanup_free_ char *val = NULL; | |
2274 | uint64_t us; | |
5ad096b3 | 2275 | |
66ebf6c0 TH |
2276 | if ((u->cgroup_realized_mask & CGROUP_MASK_CPU) == 0) |
2277 | return -ENODATA; | |
5ad096b3 | 2278 | |
66ebf6c0 TH |
2279 | r = cg_get_keyed_attribute("cpu", u->cgroup_path, "cpu.stat", keys, &val); |
2280 | if (r < 0) | |
2281 | return r; | |
2282 | ||
2283 | r = safe_atou64(val, &us); | |
2284 | if (r < 0) | |
2285 | return r; | |
2286 | ||
2287 | ns = us * NSEC_PER_USEC; | |
2288 | } else { | |
2289 | if ((u->cgroup_realized_mask & CGROUP_MASK_CPUACCT) == 0) | |
2290 | return -ENODATA; | |
2291 | ||
2292 | r = cg_get_attribute("cpuacct", u->cgroup_path, "cpuacct.usage", &v); | |
2293 | if (r == -ENOENT) | |
2294 | return -ENODATA; | |
2295 | if (r < 0) | |
2296 | return r; | |
2297 | ||
2298 | r = safe_atou64(v, &ns); | |
2299 | if (r < 0) | |
2300 | return r; | |
2301 | } | |
5ad096b3 LP |
2302 | |
2303 | *ret = ns; | |
2304 | return 0; | |
2305 | } | |
2306 | ||
2307 | int unit_get_cpu_usage(Unit *u, nsec_t *ret) { | |
2308 | nsec_t ns; | |
2309 | int r; | |
2310 | ||
fe700f46 LP |
2311 | assert(u); |
2312 | ||
2313 | /* Retrieve the current CPU usage counter. This will subtract the CPU counter taken when the unit was | |
2314 | * started. If the cgroup has been removed already, returns the last cached value. To cache the value, simply | |
2315 | * call this function with a NULL return value. */ | |
2316 | ||
2e4025c0 | 2317 | if (!UNIT_CGROUP_BOOL(u, cpu_accounting)) |
cf3b4be1 LP |
2318 | return -ENODATA; |
2319 | ||
5ad096b3 | 2320 | r = unit_get_cpu_usage_raw(u, &ns); |
fe700f46 LP |
2321 | if (r == -ENODATA && u->cpu_usage_last != NSEC_INFINITY) { |
2322 | /* If we can't get the CPU usage anymore (because the cgroup was already removed, for example), use our | |
2323 | * cached value. */ | |
2324 | ||
2325 | if (ret) | |
2326 | *ret = u->cpu_usage_last; | |
2327 | return 0; | |
2328 | } | |
5ad096b3 LP |
2329 | if (r < 0) |
2330 | return r; | |
2331 | ||
66ebf6c0 TH |
2332 | if (ns > u->cpu_usage_base) |
2333 | ns -= u->cpu_usage_base; | |
5ad096b3 LP |
2334 | else |
2335 | ns = 0; | |
2336 | ||
fe700f46 LP |
2337 | u->cpu_usage_last = ns; |
2338 | if (ret) | |
2339 | *ret = ns; | |
2340 | ||
5ad096b3 LP |
2341 | return 0; |
2342 | } | |
2343 | ||
906c06f6 DM |
2344 | int unit_get_ip_accounting( |
2345 | Unit *u, | |
2346 | CGroupIPAccountingMetric metric, | |
2347 | uint64_t *ret) { | |
2348 | ||
6b659ed8 | 2349 | uint64_t value; |
906c06f6 DM |
2350 | int fd, r; |
2351 | ||
2352 | assert(u); | |
2353 | assert(metric >= 0); | |
2354 | assert(metric < _CGROUP_IP_ACCOUNTING_METRIC_MAX); | |
2355 | assert(ret); | |
2356 | ||
cf3b4be1 LP |
2357 | /* IP accounting is currently not recursive, and hence we refuse to return any data for slice nodes. Slices are |
2358 | * inner cgroup nodes and hence have no processes directly attached, hence their counters would be zero | |
2359 | * anyway. And if we block this now we can later open this up, if the kernel learns recursive BPF cgroup | |
2360 | * filters. */ | |
2361 | if (u->type == UNIT_SLICE) | |
2362 | return -ENODATA; | |
2363 | ||
2e4025c0 | 2364 | if (!UNIT_CGROUP_BOOL(u, ip_accounting)) |
cf3b4be1 LP |
2365 | return -ENODATA; |
2366 | ||
906c06f6 DM |
2367 | fd = IN_SET(metric, CGROUP_IP_INGRESS_BYTES, CGROUP_IP_INGRESS_PACKETS) ? |
2368 | u->ip_accounting_ingress_map_fd : | |
2369 | u->ip_accounting_egress_map_fd; | |
906c06f6 DM |
2370 | if (fd < 0) |
2371 | return -ENODATA; | |
2372 | ||
2373 | if (IN_SET(metric, CGROUP_IP_INGRESS_BYTES, CGROUP_IP_EGRESS_BYTES)) | |
6b659ed8 | 2374 | r = bpf_firewall_read_accounting(fd, &value, NULL); |
906c06f6 | 2375 | else |
6b659ed8 LP |
2376 | r = bpf_firewall_read_accounting(fd, NULL, &value); |
2377 | if (r < 0) | |
2378 | return r; | |
2379 | ||
2380 | /* Add in additional metrics from a previous runtime. Note that when reexecing/reloading the daemon we compile | |
2381 | * all BPF programs and maps anew, but serialize the old counters. When deserializing we store them in the | |
2382 | * ip_accounting_extra[] field, and add them in here transparently. */ | |
2383 | ||
2384 | *ret = value + u->ip_accounting_extra[metric]; | |
906c06f6 DM |
2385 | |
2386 | return r; | |
2387 | } | |
2388 | ||
2389 | int unit_reset_cpu_accounting(Unit *u) { | |
5ad096b3 LP |
2390 | nsec_t ns; |
2391 | int r; | |
2392 | ||
2393 | assert(u); | |
2394 | ||
fe700f46 LP |
2395 | u->cpu_usage_last = NSEC_INFINITY; |
2396 | ||
5ad096b3 LP |
2397 | r = unit_get_cpu_usage_raw(u, &ns); |
2398 | if (r < 0) { | |
66ebf6c0 | 2399 | u->cpu_usage_base = 0; |
5ad096b3 | 2400 | return r; |
b56c28c3 | 2401 | } |
2633eb83 | 2402 | |
66ebf6c0 | 2403 | u->cpu_usage_base = ns; |
4ad49000 | 2404 | return 0; |
4fbf50b3 LP |
2405 | } |
2406 | ||
906c06f6 DM |
2407 | int unit_reset_ip_accounting(Unit *u) { |
2408 | int r = 0, q = 0; | |
2409 | ||
2410 | assert(u); | |
2411 | ||
2412 | if (u->ip_accounting_ingress_map_fd >= 0) | |
2413 | r = bpf_firewall_reset_accounting(u->ip_accounting_ingress_map_fd); | |
2414 | ||
2415 | if (u->ip_accounting_egress_map_fd >= 0) | |
2416 | q = bpf_firewall_reset_accounting(u->ip_accounting_egress_map_fd); | |
2417 | ||
6b659ed8 LP |
2418 | zero(u->ip_accounting_extra); |
2419 | ||
906c06f6 DM |
2420 | return r < 0 ? r : q; |
2421 | } | |
2422 | ||
e7ab4d1a LP |
2423 | void unit_invalidate_cgroup(Unit *u, CGroupMask m) { |
2424 | assert(u); | |
2425 | ||
2426 | if (!UNIT_HAS_CGROUP_CONTEXT(u)) | |
2427 | return; | |
2428 | ||
2429 | if (m == 0) | |
2430 | return; | |
2431 | ||
538b4852 TH |
2432 | /* always invalidate compat pairs together */ |
2433 | if (m & (CGROUP_MASK_IO | CGROUP_MASK_BLKIO)) | |
2434 | m |= CGROUP_MASK_IO | CGROUP_MASK_BLKIO; | |
2435 | ||
7cce4fb7 LP |
2436 | if (m & (CGROUP_MASK_CPU | CGROUP_MASK_CPUACCT)) |
2437 | m |= CGROUP_MASK_CPU | CGROUP_MASK_CPUACCT; | |
2438 | ||
60c728ad | 2439 | if ((u->cgroup_realized_mask & m) == 0) /* NOP? */ |
e7ab4d1a LP |
2440 | return; |
2441 | ||
2442 | u->cgroup_realized_mask &= ~m; | |
91a6073e | 2443 | unit_add_to_cgroup_realize_queue(u); |
e7ab4d1a LP |
2444 | } |
2445 | ||
906c06f6 DM |
2446 | void unit_invalidate_cgroup_bpf(Unit *u) { |
2447 | assert(u); | |
2448 | ||
2449 | if (!UNIT_HAS_CGROUP_CONTEXT(u)) | |
2450 | return; | |
2451 | ||
60c728ad | 2452 | if (u->cgroup_bpf_state == UNIT_CGROUP_BPF_INVALIDATED) /* NOP? */ |
906c06f6 DM |
2453 | return; |
2454 | ||
2455 | u->cgroup_bpf_state = UNIT_CGROUP_BPF_INVALIDATED; | |
91a6073e | 2456 | unit_add_to_cgroup_realize_queue(u); |
906c06f6 DM |
2457 | |
2458 | /* If we are a slice unit, we also need to put compile a new BPF program for all our children, as the IP access | |
2459 | * list of our children includes our own. */ | |
2460 | if (u->type == UNIT_SLICE) { | |
2461 | Unit *member; | |
2462 | Iterator i; | |
eef85c4a | 2463 | void *v; |
906c06f6 | 2464 | |
eef85c4a | 2465 | HASHMAP_FOREACH_KEY(v, member, u->dependencies[UNIT_BEFORE], i) { |
906c06f6 DM |
2466 | if (member == u) |
2467 | continue; | |
2468 | ||
2469 | if (UNIT_DEREF(member->slice) != u) | |
2470 | continue; | |
2471 | ||
2472 | unit_invalidate_cgroup_bpf(member); | |
2473 | } | |
2474 | } | |
2475 | } | |
2476 | ||
e7ab4d1a LP |
2477 | void manager_invalidate_startup_units(Manager *m) { |
2478 | Iterator i; | |
2479 | Unit *u; | |
2480 | ||
2481 | assert(m); | |
2482 | ||
2483 | SET_FOREACH(u, m->startup_units, i) | |
13c31542 | 2484 | unit_invalidate_cgroup(u, CGROUP_MASK_CPU|CGROUP_MASK_IO|CGROUP_MASK_BLKIO); |
e7ab4d1a LP |
2485 | } |
2486 | ||
4ad49000 LP |
2487 | static const char* const cgroup_device_policy_table[_CGROUP_DEVICE_POLICY_MAX] = { |
2488 | [CGROUP_AUTO] = "auto", | |
2489 | [CGROUP_CLOSED] = "closed", | |
2490 | [CGROUP_STRICT] = "strict", | |
2491 | }; | |
4fbf50b3 | 2492 | |
4ad49000 | 2493 | DEFINE_STRING_TABLE_LOOKUP(cgroup_device_policy, CGroupDevicePolicy); |