]>
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 | ||
906c06f6 DM |
680 | static void cgroup_apply_firewall(Unit *u, CGroupContext *c) { |
681 | int r; | |
682 | ||
683 | if (u->type == UNIT_SLICE) /* Skip this for slice units, they are inner cgroup nodes, and since bpf/cgroup is | |
684 | * not recursive we don't ever touch the bpf on them */ | |
685 | return; | |
686 | ||
687 | r = bpf_firewall_compile(u); | |
688 | if (r < 0) | |
689 | return; | |
690 | ||
691 | (void) bpf_firewall_install(u); | |
692 | return; | |
693 | } | |
694 | ||
695 | static void cgroup_context_apply( | |
696 | Unit *u, | |
697 | CGroupMask apply_mask, | |
698 | bool apply_bpf, | |
699 | ManagerState state) { | |
700 | ||
f29ff115 TH |
701 | const char *path; |
702 | CGroupContext *c; | |
01efdf13 | 703 | bool is_root; |
4ad49000 LP |
704 | int r; |
705 | ||
f29ff115 TH |
706 | assert(u); |
707 | ||
708 | c = unit_get_cgroup_context(u); | |
709 | path = u->cgroup_path; | |
710 | ||
4ad49000 LP |
711 | assert(c); |
712 | assert(path); | |
8e274523 | 713 | |
906c06f6 DM |
714 | /* Nothing to do? Exit early! */ |
715 | if (apply_mask == 0 && !apply_bpf) | |
4ad49000 | 716 | return; |
8e274523 | 717 | |
71c26873 | 718 | /* Some cgroup attributes are not supported on the root cgroup, |
01efdf13 LP |
719 | * hence silently ignore */ |
720 | is_root = isempty(path) || path_equal(path, "/"); | |
6da13913 ZJS |
721 | if (is_root) |
722 | /* Make sure we don't try to display messages with an empty path. */ | |
723 | path = "/"; | |
01efdf13 | 724 | |
714e2e1d LP |
725 | /* We generally ignore errors caused by read-only mounted |
726 | * cgroup trees (assuming we are running in a container then), | |
727 | * and missing cgroups, i.e. EROFS and ENOENT. */ | |
728 | ||
906c06f6 DM |
729 | if ((apply_mask & CGROUP_MASK_CPU) && !is_root) { |
730 | bool has_weight, has_shares; | |
731 | ||
732 | has_weight = cgroup_context_has_cpu_weight(c); | |
733 | has_shares = cgroup_context_has_cpu_shares(c); | |
8e274523 | 734 | |
b4cccbc1 | 735 | if (cg_all_unified() > 0) { |
66ebf6c0 | 736 | uint64_t weight; |
b2f8b02e | 737 | |
66ebf6c0 TH |
738 | if (has_weight) |
739 | weight = cgroup_context_cpu_weight(c, state); | |
740 | else if (has_shares) { | |
741 | uint64_t shares = cgroup_context_cpu_shares(c, state); | |
b2f8b02e | 742 | |
66ebf6c0 TH |
743 | weight = cgroup_cpu_shares_to_weight(shares); |
744 | ||
745 | log_cgroup_compat(u, "Applying [Startup]CpuShares %" PRIu64 " as [Startup]CpuWeight %" PRIu64 " on %s", | |
746 | shares, weight, path); | |
747 | } else | |
748 | weight = CGROUP_WEIGHT_DEFAULT; | |
749 | ||
750 | cgroup_apply_unified_cpu_config(u, weight, c->cpu_quota_per_sec_usec); | |
751 | } else { | |
752 | uint64_t shares; | |
753 | ||
7d862ab8 | 754 | if (has_weight) { |
66ebf6c0 TH |
755 | uint64_t weight = cgroup_context_cpu_weight(c, state); |
756 | ||
757 | shares = cgroup_cpu_weight_to_shares(weight); | |
758 | ||
759 | log_cgroup_compat(u, "Applying [Startup]CpuWeight %" PRIu64 " as [Startup]CpuShares %" PRIu64 " on %s", | |
760 | weight, shares, path); | |
7d862ab8 TH |
761 | } else if (has_shares) |
762 | shares = cgroup_context_cpu_shares(c, state); | |
763 | else | |
66ebf6c0 TH |
764 | shares = CGROUP_CPU_SHARES_DEFAULT; |
765 | ||
766 | cgroup_apply_legacy_cpu_config(u, shares, c->cpu_quota_per_sec_usec); | |
767 | } | |
4ad49000 LP |
768 | } |
769 | ||
906c06f6 | 770 | if (apply_mask & CGROUP_MASK_IO) { |
538b4852 TH |
771 | bool has_io = cgroup_context_has_io_config(c); |
772 | bool has_blockio = cgroup_context_has_blockio_config(c); | |
13c31542 TH |
773 | |
774 | if (!is_root) { | |
64faf04c TH |
775 | char buf[8+DECIMAL_STR_MAX(uint64_t)+1]; |
776 | uint64_t weight; | |
13c31542 | 777 | |
538b4852 TH |
778 | if (has_io) |
779 | weight = cgroup_context_io_weight(c, state); | |
128fadc9 TH |
780 | else if (has_blockio) { |
781 | uint64_t blkio_weight = cgroup_context_blkio_weight(c, state); | |
782 | ||
783 | weight = cgroup_weight_blkio_to_io(blkio_weight); | |
784 | ||
785 | log_cgroup_compat(u, "Applying [Startup]BlockIOWeight %" PRIu64 " as [Startup]IOWeight %" PRIu64, | |
786 | blkio_weight, weight); | |
787 | } else | |
538b4852 | 788 | weight = CGROUP_WEIGHT_DEFAULT; |
13c31542 TH |
789 | |
790 | xsprintf(buf, "default %" PRIu64 "\n", weight); | |
791 | r = cg_set_attribute("io", path, "io.weight", buf); | |
792 | if (r < 0) | |
f29ff115 TH |
793 | log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, |
794 | "Failed to set io.weight: %m"); | |
13c31542 | 795 | |
538b4852 TH |
796 | if (has_io) { |
797 | CGroupIODeviceWeight *w; | |
798 | ||
799 | /* FIXME: no way to reset this list */ | |
800 | LIST_FOREACH(device_weights, w, c->io_device_weights) | |
f29ff115 | 801 | cgroup_apply_io_device_weight(u, w->path, w->weight); |
538b4852 TH |
802 | } else if (has_blockio) { |
803 | CGroupBlockIODeviceWeight *w; | |
804 | ||
805 | /* FIXME: no way to reset this list */ | |
128fadc9 TH |
806 | LIST_FOREACH(device_weights, w, c->blockio_device_weights) { |
807 | weight = cgroup_weight_blkio_to_io(w->weight); | |
808 | ||
809 | log_cgroup_compat(u, "Applying BlockIODeviceWeight %" PRIu64 " as IODeviceWeight %" PRIu64 " for %s", | |
810 | w->weight, weight, w->path); | |
811 | ||
812 | cgroup_apply_io_device_weight(u, w->path, weight); | |
813 | } | |
538b4852 | 814 | } |
13c31542 TH |
815 | } |
816 | ||
64faf04c | 817 | /* Apply limits and free ones without config. */ |
538b4852 TH |
818 | if (has_io) { |
819 | CGroupIODeviceLimit *l, *next; | |
820 | ||
821 | LIST_FOREACH_SAFE(device_limits, l, next, c->io_device_limits) { | |
f29ff115 | 822 | if (!cgroup_apply_io_device_limit(u, l->path, l->limits)) |
538b4852 TH |
823 | cgroup_context_free_io_device_limit(c, l); |
824 | } | |
825 | } else if (has_blockio) { | |
826 | CGroupBlockIODeviceBandwidth *b, *next; | |
827 | ||
828 | LIST_FOREACH_SAFE(device_bandwidths, b, next, c->blockio_device_bandwidths) { | |
829 | uint64_t limits[_CGROUP_IO_LIMIT_TYPE_MAX]; | |
830 | CGroupIOLimitType type; | |
831 | ||
832 | for (type = 0; type < _CGROUP_IO_LIMIT_TYPE_MAX; type++) | |
833 | limits[type] = cgroup_io_limit_defaults[type]; | |
834 | ||
835 | limits[CGROUP_IO_RBPS_MAX] = b->rbps; | |
836 | limits[CGROUP_IO_WBPS_MAX] = b->wbps; | |
837 | ||
128fadc9 TH |
838 | log_cgroup_compat(u, "Applying BlockIO{Read|Write}Bandwidth %" PRIu64 " %" PRIu64 " as IO{Read|Write}BandwidthMax for %s", |
839 | b->rbps, b->wbps, b->path); | |
840 | ||
f29ff115 | 841 | if (!cgroup_apply_io_device_limit(u, b->path, limits)) |
538b4852 TH |
842 | cgroup_context_free_blockio_device_bandwidth(c, b); |
843 | } | |
13c31542 TH |
844 | } |
845 | } | |
846 | ||
906c06f6 | 847 | if (apply_mask & CGROUP_MASK_BLKIO) { |
538b4852 TH |
848 | bool has_io = cgroup_context_has_io_config(c); |
849 | bool has_blockio = cgroup_context_has_blockio_config(c); | |
4ad49000 | 850 | |
01efdf13 | 851 | if (!is_root) { |
64faf04c TH |
852 | char buf[DECIMAL_STR_MAX(uint64_t)+1]; |
853 | uint64_t weight; | |
64faf04c | 854 | |
7d862ab8 | 855 | if (has_io) { |
128fadc9 TH |
856 | uint64_t io_weight = cgroup_context_io_weight(c, state); |
857 | ||
538b4852 | 858 | weight = cgroup_weight_io_to_blkio(cgroup_context_io_weight(c, state)); |
128fadc9 TH |
859 | |
860 | log_cgroup_compat(u, "Applying [Startup]IOWeight %" PRIu64 " as [Startup]BlockIOWeight %" PRIu64, | |
861 | io_weight, weight); | |
7d862ab8 TH |
862 | } else if (has_blockio) |
863 | weight = cgroup_context_blkio_weight(c, state); | |
864 | else | |
538b4852 | 865 | weight = CGROUP_BLKIO_WEIGHT_DEFAULT; |
64faf04c TH |
866 | |
867 | xsprintf(buf, "%" PRIu64 "\n", weight); | |
01efdf13 | 868 | r = cg_set_attribute("blkio", path, "blkio.weight", buf); |
1aeab12b | 869 | if (r < 0) |
f29ff115 TH |
870 | log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, |
871 | "Failed to set blkio.weight: %m"); | |
4ad49000 | 872 | |
7d862ab8 | 873 | if (has_io) { |
538b4852 TH |
874 | CGroupIODeviceWeight *w; |
875 | ||
876 | /* FIXME: no way to reset this list */ | |
128fadc9 TH |
877 | LIST_FOREACH(device_weights, w, c->io_device_weights) { |
878 | weight = cgroup_weight_io_to_blkio(w->weight); | |
879 | ||
880 | log_cgroup_compat(u, "Applying IODeviceWeight %" PRIu64 " as BlockIODeviceWeight %" PRIu64 " for %s", | |
881 | w->weight, weight, w->path); | |
882 | ||
883 | cgroup_apply_blkio_device_weight(u, w->path, weight); | |
884 | } | |
7d862ab8 TH |
885 | } else if (has_blockio) { |
886 | CGroupBlockIODeviceWeight *w; | |
887 | ||
888 | /* FIXME: no way to reset this list */ | |
889 | LIST_FOREACH(device_weights, w, c->blockio_device_weights) | |
890 | cgroup_apply_blkio_device_weight(u, w->path, w->weight); | |
538b4852 | 891 | } |
4ad49000 LP |
892 | } |
893 | ||
64faf04c | 894 | /* Apply limits and free ones without config. */ |
7d862ab8 | 895 | if (has_io) { |
538b4852 TH |
896 | CGroupIODeviceLimit *l, *next; |
897 | ||
898 | LIST_FOREACH_SAFE(device_limits, l, next, c->io_device_limits) { | |
128fadc9 TH |
899 | log_cgroup_compat(u, "Applying IO{Read|Write}Bandwidth %" PRIu64 " %" PRIu64 " as BlockIO{Read|Write}BandwidthMax for %s", |
900 | l->limits[CGROUP_IO_RBPS_MAX], l->limits[CGROUP_IO_WBPS_MAX], l->path); | |
901 | ||
f29ff115 | 902 | if (!cgroup_apply_blkio_device_limit(u, l->path, l->limits[CGROUP_IO_RBPS_MAX], l->limits[CGROUP_IO_WBPS_MAX])) |
538b4852 TH |
903 | cgroup_context_free_io_device_limit(c, l); |
904 | } | |
7d862ab8 TH |
905 | } else if (has_blockio) { |
906 | CGroupBlockIODeviceBandwidth *b, *next; | |
907 | ||
908 | LIST_FOREACH_SAFE(device_bandwidths, b, next, c->blockio_device_bandwidths) | |
909 | if (!cgroup_apply_blkio_device_limit(u, b->path, b->rbps, b->wbps)) | |
910 | cgroup_context_free_blockio_device_bandwidth(c, b); | |
d686d8a9 | 911 | } |
8e274523 LP |
912 | } |
913 | ||
906c06f6 | 914 | if ((apply_mask & CGROUP_MASK_MEMORY) && !is_root) { |
b4cccbc1 LP |
915 | if (cg_all_unified() > 0) { |
916 | uint64_t max, swap_max = CGROUP_LIMIT_MAX; | |
efdb0237 | 917 | |
96e131ea | 918 | if (cgroup_context_has_unified_memory_config(c)) { |
da4d897e | 919 | max = c->memory_max; |
96e131ea WC |
920 | swap_max = c->memory_swap_max; |
921 | } else { | |
da4d897e | 922 | max = c->memory_limit; |
efdb0237 | 923 | |
128fadc9 TH |
924 | if (max != CGROUP_LIMIT_MAX) |
925 | log_cgroup_compat(u, "Applying MemoryLimit %" PRIu64 " as MemoryMax", max); | |
926 | } | |
927 | ||
f29ff115 TH |
928 | cgroup_apply_unified_memory_limit(u, "memory.low", c->memory_low); |
929 | cgroup_apply_unified_memory_limit(u, "memory.high", c->memory_high); | |
930 | cgroup_apply_unified_memory_limit(u, "memory.max", max); | |
96e131ea | 931 | cgroup_apply_unified_memory_limit(u, "memory.swap.max", swap_max); |
efdb0237 | 932 | } else { |
da4d897e | 933 | char buf[DECIMAL_STR_MAX(uint64_t) + 1]; |
7d862ab8 | 934 | uint64_t val; |
da4d897e | 935 | |
7d862ab8 | 936 | if (cgroup_context_has_unified_memory_config(c)) { |
78a4ee59 | 937 | val = c->memory_max; |
7d862ab8 TH |
938 | log_cgroup_compat(u, "Applying MemoryMax %" PRIi64 " as MemoryLimit", val); |
939 | } else | |
940 | val = c->memory_limit; | |
128fadc9 | 941 | |
78a4ee59 DM |
942 | if (val == CGROUP_LIMIT_MAX) |
943 | strncpy(buf, "-1\n", sizeof(buf)); | |
944 | else | |
945 | xsprintf(buf, "%" PRIu64 "\n", val); | |
946 | ||
da4d897e TH |
947 | r = cg_set_attribute("memory", path, "memory.limit_in_bytes", buf); |
948 | if (r < 0) | |
f29ff115 TH |
949 | log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, |
950 | "Failed to set memory.limit_in_bytes: %m"); | |
da4d897e | 951 | } |
4ad49000 | 952 | } |
8e274523 | 953 | |
906c06f6 | 954 | if ((apply_mask & CGROUP_MASK_DEVICES) && !is_root) { |
4ad49000 | 955 | CGroupDeviceAllow *a; |
8e274523 | 956 | |
714e2e1d LP |
957 | /* Changing the devices list of a populated cgroup |
958 | * might result in EINVAL, hence ignore EINVAL | |
959 | * here. */ | |
960 | ||
4ad49000 LP |
961 | if (c->device_allow || c->device_policy != CGROUP_AUTO) |
962 | r = cg_set_attribute("devices", path, "devices.deny", "a"); | |
963 | else | |
964 | r = cg_set_attribute("devices", path, "devices.allow", "a"); | |
1aeab12b | 965 | if (r < 0) |
f29ff115 TH |
966 | log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EINVAL, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, |
967 | "Failed to reset devices.list: %m"); | |
fb385181 | 968 | |
4ad49000 LP |
969 | if (c->device_policy == CGROUP_CLOSED || |
970 | (c->device_policy == CGROUP_AUTO && c->device_allow)) { | |
971 | static const char auto_devices[] = | |
7d711efb LP |
972 | "/dev/null\0" "rwm\0" |
973 | "/dev/zero\0" "rwm\0" | |
974 | "/dev/full\0" "rwm\0" | |
975 | "/dev/random\0" "rwm\0" | |
976 | "/dev/urandom\0" "rwm\0" | |
977 | "/dev/tty\0" "rwm\0" | |
0d9e7991 AP |
978 | "/dev/pts/ptmx\0" "rw\0" /* /dev/pts/ptmx may not be duplicated, but accessed */ |
979 | /* Allow /run/systemd/inaccessible/{chr,blk} devices for mapping InaccessiblePaths */ | |
e7330dfe DP |
980 | "-/run/systemd/inaccessible/chr\0" "rwm\0" |
981 | "-/run/systemd/inaccessible/blk\0" "rwm\0"; | |
4ad49000 LP |
982 | |
983 | const char *x, *y; | |
984 | ||
985 | NULSTR_FOREACH_PAIR(x, y, auto_devices) | |
986 | whitelist_device(path, x, y); | |
7d711efb LP |
987 | |
988 | whitelist_major(path, "pts", 'c', "rw"); | |
4ad49000 LP |
989 | } |
990 | ||
991 | LIST_FOREACH(device_allow, a, c->device_allow) { | |
fb4650aa | 992 | char acc[4], *val; |
4ad49000 LP |
993 | unsigned k = 0; |
994 | ||
995 | if (a->r) | |
996 | acc[k++] = 'r'; | |
997 | if (a->w) | |
998 | acc[k++] = 'w'; | |
999 | if (a->m) | |
1000 | acc[k++] = 'm'; | |
fb385181 | 1001 | |
4ad49000 LP |
1002 | if (k == 0) |
1003 | continue; | |
fb385181 | 1004 | |
4ad49000 | 1005 | acc[k++] = 0; |
90060676 | 1006 | |
27458ed6 | 1007 | if (path_startswith(a->path, "/dev/")) |
90060676 | 1008 | whitelist_device(path, a->path, acc); |
fb4650aa ZJS |
1009 | else if ((val = startswith(a->path, "block-"))) |
1010 | whitelist_major(path, val, 'b', acc); | |
1011 | else if ((val = startswith(a->path, "char-"))) | |
1012 | whitelist_major(path, val, 'c', acc); | |
90060676 | 1013 | else |
f29ff115 | 1014 | log_unit_debug(u, "Ignoring device %s while writing cgroup attribute.", a->path); |
4ad49000 LP |
1015 | } |
1016 | } | |
03a7b521 | 1017 | |
906c06f6 | 1018 | if ((apply_mask & CGROUP_MASK_PIDS) && !is_root) { |
03a7b521 | 1019 | |
f5058264 | 1020 | if (c->tasks_max != CGROUP_LIMIT_MAX) { |
03a7b521 LP |
1021 | char buf[DECIMAL_STR_MAX(uint64_t) + 2]; |
1022 | ||
1023 | sprintf(buf, "%" PRIu64 "\n", c->tasks_max); | |
1024 | r = cg_set_attribute("pids", path, "pids.max", buf); | |
1025 | } else | |
1026 | r = cg_set_attribute("pids", path, "pids.max", "max"); | |
1027 | ||
1028 | if (r < 0) | |
f29ff115 TH |
1029 | log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, |
1030 | "Failed to set pids.max: %m"); | |
03a7b521 | 1031 | } |
906c06f6 DM |
1032 | |
1033 | if (apply_bpf) | |
1034 | cgroup_apply_firewall(u, c); | |
fb385181 LP |
1035 | } |
1036 | ||
efdb0237 LP |
1037 | CGroupMask cgroup_context_get_mask(CGroupContext *c) { |
1038 | CGroupMask mask = 0; | |
8e274523 | 1039 | |
4ad49000 | 1040 | /* Figure out which controllers we need */ |
8e274523 | 1041 | |
b2f8b02e | 1042 | if (c->cpu_accounting || |
66ebf6c0 TH |
1043 | cgroup_context_has_cpu_weight(c) || |
1044 | cgroup_context_has_cpu_shares(c) || | |
3a43da28 | 1045 | c->cpu_quota_per_sec_usec != USEC_INFINITY) |
efdb0237 | 1046 | mask |= CGROUP_MASK_CPUACCT | CGROUP_MASK_CPU; |
ecedd90f | 1047 | |
538b4852 TH |
1048 | if (cgroup_context_has_io_config(c) || cgroup_context_has_blockio_config(c)) |
1049 | mask |= CGROUP_MASK_IO | CGROUP_MASK_BLKIO; | |
ecedd90f | 1050 | |
4ad49000 | 1051 | if (c->memory_accounting || |
da4d897e TH |
1052 | c->memory_limit != CGROUP_LIMIT_MAX || |
1053 | cgroup_context_has_unified_memory_config(c)) | |
efdb0237 | 1054 | mask |= CGROUP_MASK_MEMORY; |
8e274523 | 1055 | |
a931ad47 LP |
1056 | if (c->device_allow || |
1057 | c->device_policy != CGROUP_AUTO) | |
3905f127 | 1058 | mask |= CGROUP_MASK_DEVICES; |
4ad49000 | 1059 | |
03a7b521 LP |
1060 | if (c->tasks_accounting || |
1061 | c->tasks_max != (uint64_t) -1) | |
1062 | mask |= CGROUP_MASK_PIDS; | |
1063 | ||
4ad49000 | 1064 | return mask; |
8e274523 LP |
1065 | } |
1066 | ||
efdb0237 | 1067 | CGroupMask unit_get_own_mask(Unit *u) { |
4ad49000 | 1068 | CGroupContext *c; |
8e274523 | 1069 | |
efdb0237 LP |
1070 | /* Returns the mask of controllers the unit needs for itself */ |
1071 | ||
4ad49000 LP |
1072 | c = unit_get_cgroup_context(u); |
1073 | if (!c) | |
1074 | return 0; | |
8e274523 | 1075 | |
02638280 LP |
1076 | return cgroup_context_get_mask(c); |
1077 | } | |
1078 | ||
1079 | CGroupMask unit_get_delegate_mask(Unit *u) { | |
1080 | CGroupContext *c; | |
1081 | ||
1082 | /* If delegation is turned on, then turn on selected controllers, unless we are on the legacy hierarchy and the | |
1083 | * process we fork into is known to drop privileges, and hence shouldn't get access to the controllers. | |
19af675e | 1084 | * |
02638280 | 1085 | * Note that on the unified hierarchy it is safe to delegate controllers to unprivileged services. */ |
a931ad47 | 1086 | |
02638280 LP |
1087 | if (u->type == UNIT_SLICE) |
1088 | return 0; | |
1089 | ||
1090 | c = unit_get_cgroup_context(u); | |
1091 | if (!c) | |
1092 | return 0; | |
1093 | ||
1094 | if (!c->delegate) | |
1095 | return 0; | |
1096 | ||
1097 | if (cg_all_unified() <= 0) { | |
a931ad47 LP |
1098 | ExecContext *e; |
1099 | ||
1100 | e = unit_get_exec_context(u); | |
02638280 LP |
1101 | if (e && !exec_context_maintains_privileges(e)) |
1102 | return 0; | |
a931ad47 LP |
1103 | } |
1104 | ||
02638280 | 1105 | return c->delegate_controllers; |
8e274523 LP |
1106 | } |
1107 | ||
efdb0237 | 1108 | CGroupMask unit_get_members_mask(Unit *u) { |
4ad49000 | 1109 | assert(u); |
bc432dc7 | 1110 | |
02638280 | 1111 | /* Returns the mask of controllers all of the unit's children require, merged */ |
efdb0237 | 1112 | |
bc432dc7 LP |
1113 | if (u->cgroup_members_mask_valid) |
1114 | return u->cgroup_members_mask; | |
1115 | ||
02638280 | 1116 | u->cgroup_members_mask = unit_get_delegate_mask(u); |
bc432dc7 LP |
1117 | |
1118 | if (u->type == UNIT_SLICE) { | |
eef85c4a | 1119 | void *v; |
bc432dc7 LP |
1120 | Unit *member; |
1121 | Iterator i; | |
1122 | ||
eef85c4a | 1123 | HASHMAP_FOREACH_KEY(v, member, u->dependencies[UNIT_BEFORE], i) { |
bc432dc7 LP |
1124 | |
1125 | if (member == u) | |
1126 | continue; | |
1127 | ||
d4fdc205 | 1128 | if (UNIT_DEREF(member->slice) != u) |
bc432dc7 LP |
1129 | continue; |
1130 | ||
31604970 | 1131 | u->cgroup_members_mask |= unit_get_subtree_mask(member); /* note that this calls ourselves again, for the children */ |
bc432dc7 LP |
1132 | } |
1133 | } | |
1134 | ||
1135 | u->cgroup_members_mask_valid = true; | |
6414b7c9 | 1136 | return u->cgroup_members_mask; |
246aa6dd LP |
1137 | } |
1138 | ||
efdb0237 | 1139 | CGroupMask unit_get_siblings_mask(Unit *u) { |
4ad49000 | 1140 | assert(u); |
246aa6dd | 1141 | |
efdb0237 LP |
1142 | /* Returns the mask of controllers all of the unit's siblings |
1143 | * require, i.e. the members mask of the unit's parent slice | |
1144 | * if there is one. */ | |
1145 | ||
bc432dc7 | 1146 | if (UNIT_ISSET(u->slice)) |
637f421e | 1147 | return unit_get_members_mask(UNIT_DEREF(u->slice)); |
4ad49000 | 1148 | |
31604970 | 1149 | return unit_get_subtree_mask(u); |
246aa6dd LP |
1150 | } |
1151 | ||
efdb0237 LP |
1152 | CGroupMask unit_get_subtree_mask(Unit *u) { |
1153 | ||
1154 | /* Returns the mask of this subtree, meaning of the group | |
1155 | * itself and its children. */ | |
1156 | ||
1157 | return unit_get_own_mask(u) | unit_get_members_mask(u); | |
1158 | } | |
1159 | ||
1160 | CGroupMask unit_get_target_mask(Unit *u) { | |
1161 | CGroupMask mask; | |
1162 | ||
1163 | /* This returns the cgroup mask of all controllers to enable | |
1164 | * for a specific cgroup, i.e. everything it needs itself, | |
1165 | * plus all that its children need, plus all that its siblings | |
1166 | * need. This is primarily useful on the legacy cgroup | |
1167 | * hierarchy, where we need to duplicate each cgroup in each | |
1168 | * hierarchy that shall be enabled for it. */ | |
6414b7c9 | 1169 | |
efdb0237 LP |
1170 | mask = unit_get_own_mask(u) | unit_get_members_mask(u) | unit_get_siblings_mask(u); |
1171 | mask &= u->manager->cgroup_supported; | |
1172 | ||
1173 | return mask; | |
1174 | } | |
1175 | ||
1176 | CGroupMask unit_get_enable_mask(Unit *u) { | |
1177 | CGroupMask mask; | |
1178 | ||
1179 | /* This returns the cgroup mask of all controllers to enable | |
1180 | * for the children of a specific cgroup. This is primarily | |
1181 | * useful for the unified cgroup hierarchy, where each cgroup | |
1182 | * controls which controllers are enabled for its children. */ | |
1183 | ||
1184 | mask = unit_get_members_mask(u); | |
6414b7c9 DS |
1185 | mask &= u->manager->cgroup_supported; |
1186 | ||
1187 | return mask; | |
1188 | } | |
1189 | ||
906c06f6 DM |
1190 | bool unit_get_needs_bpf(Unit *u) { |
1191 | CGroupContext *c; | |
1192 | Unit *p; | |
1193 | assert(u); | |
1194 | ||
1195 | /* We never attach BPF to slice units, as they are inner cgroup nodes and cgroup/BPF is not recursive at the | |
1196 | * moment. */ | |
1197 | if (u->type == UNIT_SLICE) | |
1198 | return false; | |
1199 | ||
1200 | c = unit_get_cgroup_context(u); | |
1201 | if (!c) | |
1202 | return false; | |
1203 | ||
1204 | if (c->ip_accounting || | |
1205 | c->ip_address_allow || | |
1206 | c->ip_address_deny) | |
1207 | return true; | |
1208 | ||
1209 | /* If any parent slice has an IP access list defined, it applies too */ | |
1210 | for (p = UNIT_DEREF(u->slice); p; p = UNIT_DEREF(p->slice)) { | |
1211 | c = unit_get_cgroup_context(p); | |
1212 | if (!c) | |
1213 | return false; | |
1214 | ||
1215 | if (c->ip_address_allow || | |
1216 | c->ip_address_deny) | |
1217 | return true; | |
1218 | } | |
1219 | ||
1220 | return false; | |
1221 | } | |
1222 | ||
6414b7c9 DS |
1223 | /* Recurse from a unit up through its containing slices, propagating |
1224 | * mask bits upward. A unit is also member of itself. */ | |
bc432dc7 | 1225 | void unit_update_cgroup_members_masks(Unit *u) { |
efdb0237 | 1226 | CGroupMask m; |
bc432dc7 LP |
1227 | bool more; |
1228 | ||
1229 | assert(u); | |
1230 | ||
1231 | /* Calculate subtree mask */ | |
efdb0237 | 1232 | m = unit_get_subtree_mask(u); |
bc432dc7 LP |
1233 | |
1234 | /* See if anything changed from the previous invocation. If | |
1235 | * not, we're done. */ | |
1236 | if (u->cgroup_subtree_mask_valid && m == u->cgroup_subtree_mask) | |
1237 | return; | |
1238 | ||
1239 | more = | |
1240 | u->cgroup_subtree_mask_valid && | |
1241 | ((m & ~u->cgroup_subtree_mask) != 0) && | |
1242 | ((~m & u->cgroup_subtree_mask) == 0); | |
1243 | ||
1244 | u->cgroup_subtree_mask = m; | |
1245 | u->cgroup_subtree_mask_valid = true; | |
1246 | ||
6414b7c9 DS |
1247 | if (UNIT_ISSET(u->slice)) { |
1248 | Unit *s = UNIT_DEREF(u->slice); | |
bc432dc7 LP |
1249 | |
1250 | if (more) | |
1251 | /* There's more set now than before. We | |
1252 | * propagate the new mask to the parent's mask | |
1253 | * (not caring if it actually was valid or | |
1254 | * not). */ | |
1255 | ||
1256 | s->cgroup_members_mask |= m; | |
1257 | ||
1258 | else | |
1259 | /* There's less set now than before (or we | |
1260 | * don't know), we need to recalculate | |
1261 | * everything, so let's invalidate the | |
1262 | * parent's members mask */ | |
1263 | ||
1264 | s->cgroup_members_mask_valid = false; | |
1265 | ||
1266 | /* And now make sure that this change also hits our | |
1267 | * grandparents */ | |
1268 | unit_update_cgroup_members_masks(s); | |
6414b7c9 DS |
1269 | } |
1270 | } | |
1271 | ||
efdb0237 | 1272 | static const char *migrate_callback(CGroupMask mask, void *userdata) { |
03b90d4b LP |
1273 | Unit *u = userdata; |
1274 | ||
1275 | assert(mask != 0); | |
1276 | assert(u); | |
1277 | ||
1278 | while (u) { | |
1279 | if (u->cgroup_path && | |
1280 | u->cgroup_realized && | |
1281 | (u->cgroup_realized_mask & mask) == mask) | |
1282 | return u->cgroup_path; | |
1283 | ||
1284 | u = UNIT_DEREF(u->slice); | |
1285 | } | |
1286 | ||
1287 | return NULL; | |
1288 | } | |
1289 | ||
efdb0237 LP |
1290 | char *unit_default_cgroup_path(Unit *u) { |
1291 | _cleanup_free_ char *escaped = NULL, *slice = NULL; | |
1292 | int r; | |
1293 | ||
1294 | assert(u); | |
1295 | ||
1296 | if (unit_has_name(u, SPECIAL_ROOT_SLICE)) | |
1297 | return strdup(u->manager->cgroup_root); | |
1298 | ||
1299 | if (UNIT_ISSET(u->slice) && !unit_has_name(UNIT_DEREF(u->slice), SPECIAL_ROOT_SLICE)) { | |
1300 | r = cg_slice_to_path(UNIT_DEREF(u->slice)->id, &slice); | |
1301 | if (r < 0) | |
1302 | return NULL; | |
1303 | } | |
1304 | ||
1305 | escaped = cg_escape(u->id); | |
1306 | if (!escaped) | |
1307 | return NULL; | |
1308 | ||
1309 | if (slice) | |
605405c6 ZJS |
1310 | return strjoin(u->manager->cgroup_root, "/", slice, "/", |
1311 | escaped); | |
efdb0237 | 1312 | else |
605405c6 | 1313 | return strjoin(u->manager->cgroup_root, "/", escaped); |
efdb0237 LP |
1314 | } |
1315 | ||
1316 | int unit_set_cgroup_path(Unit *u, const char *path) { | |
1317 | _cleanup_free_ char *p = NULL; | |
1318 | int r; | |
1319 | ||
1320 | assert(u); | |
1321 | ||
1322 | if (path) { | |
1323 | p = strdup(path); | |
1324 | if (!p) | |
1325 | return -ENOMEM; | |
1326 | } else | |
1327 | p = NULL; | |
1328 | ||
1329 | if (streq_ptr(u->cgroup_path, p)) | |
1330 | return 0; | |
1331 | ||
1332 | if (p) { | |
1333 | r = hashmap_put(u->manager->cgroup_unit, p, u); | |
1334 | if (r < 0) | |
1335 | return r; | |
1336 | } | |
1337 | ||
1338 | unit_release_cgroup(u); | |
1339 | ||
1340 | u->cgroup_path = p; | |
1341 | p = NULL; | |
1342 | ||
1343 | return 1; | |
1344 | } | |
1345 | ||
1346 | int unit_watch_cgroup(Unit *u) { | |
ab2c3861 | 1347 | _cleanup_free_ char *events = NULL; |
efdb0237 LP |
1348 | int r; |
1349 | ||
1350 | assert(u); | |
1351 | ||
1352 | if (!u->cgroup_path) | |
1353 | return 0; | |
1354 | ||
1355 | if (u->cgroup_inotify_wd >= 0) | |
1356 | return 0; | |
1357 | ||
1358 | /* Only applies to the unified hierarchy */ | |
c22800e4 | 1359 | r = cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER); |
b4cccbc1 LP |
1360 | if (r < 0) |
1361 | return log_error_errno(r, "Failed to determine whether the name=systemd hierarchy is unified: %m"); | |
1362 | if (r == 0) | |
efdb0237 LP |
1363 | return 0; |
1364 | ||
1365 | /* Don't watch the root slice, it's pointless. */ | |
1366 | if (unit_has_name(u, SPECIAL_ROOT_SLICE)) | |
1367 | return 0; | |
1368 | ||
1369 | r = hashmap_ensure_allocated(&u->manager->cgroup_inotify_wd_unit, &trivial_hash_ops); | |
1370 | if (r < 0) | |
1371 | return log_oom(); | |
1372 | ||
ab2c3861 | 1373 | r = cg_get_path(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path, "cgroup.events", &events); |
efdb0237 LP |
1374 | if (r < 0) |
1375 | return log_oom(); | |
1376 | ||
ab2c3861 | 1377 | u->cgroup_inotify_wd = inotify_add_watch(u->manager->cgroup_inotify_fd, events, IN_MODIFY); |
efdb0237 LP |
1378 | if (u->cgroup_inotify_wd < 0) { |
1379 | ||
1380 | if (errno == ENOENT) /* If the directory is already | |
1381 | * gone we don't need to track | |
1382 | * it, so this is not an error */ | |
1383 | return 0; | |
1384 | ||
1385 | return log_unit_error_errno(u, errno, "Failed to add inotify watch descriptor for control group %s: %m", u->cgroup_path); | |
1386 | } | |
1387 | ||
1388 | r = hashmap_put(u->manager->cgroup_inotify_wd_unit, INT_TO_PTR(u->cgroup_inotify_wd), u); | |
1389 | if (r < 0) | |
1390 | return log_unit_error_errno(u, r, "Failed to add inotify watch descriptor to hash map: %m"); | |
1391 | ||
1392 | return 0; | |
1393 | } | |
1394 | ||
1395 | static int unit_create_cgroup( | |
1396 | Unit *u, | |
1397 | CGroupMask target_mask, | |
906c06f6 DM |
1398 | CGroupMask enable_mask, |
1399 | bool needs_bpf) { | |
efdb0237 | 1400 | |
0cd385d3 | 1401 | CGroupContext *c; |
bc432dc7 | 1402 | int r; |
64747e2d | 1403 | |
4ad49000 | 1404 | assert(u); |
64747e2d | 1405 | |
0cd385d3 LP |
1406 | c = unit_get_cgroup_context(u); |
1407 | if (!c) | |
1408 | return 0; | |
1409 | ||
7b3fd631 LP |
1410 | if (!u->cgroup_path) { |
1411 | _cleanup_free_ char *path = NULL; | |
64747e2d | 1412 | |
7b3fd631 LP |
1413 | path = unit_default_cgroup_path(u); |
1414 | if (!path) | |
1415 | return log_oom(); | |
1416 | ||
efdb0237 LP |
1417 | r = unit_set_cgroup_path(u, path); |
1418 | if (r == -EEXIST) | |
1419 | return log_unit_error_errno(u, r, "Control group %s exists already.", path); | |
1420 | if (r < 0) | |
1421 | return log_unit_error_errno(u, r, "Failed to set unit's control group path to %s: %m", path); | |
b58b8e11 HH |
1422 | } |
1423 | ||
03b90d4b | 1424 | /* First, create our own group */ |
efdb0237 | 1425 | r = cg_create_everywhere(u->manager->cgroup_supported, target_mask, u->cgroup_path); |
23bbb0de | 1426 | if (r < 0) |
efdb0237 LP |
1427 | return log_unit_error_errno(u, r, "Failed to create cgroup %s: %m", u->cgroup_path); |
1428 | ||
1429 | /* Start watching it */ | |
1430 | (void) unit_watch_cgroup(u); | |
1431 | ||
1432 | /* Enable all controllers we need */ | |
1433 | r = cg_enable_everywhere(u->manager->cgroup_supported, enable_mask, u->cgroup_path); | |
1434 | if (r < 0) | |
1435 | log_unit_warning_errno(u, r, "Failed to enable controllers on cgroup %s, ignoring: %m", u->cgroup_path); | |
03b90d4b LP |
1436 | |
1437 | /* Keep track that this is now realized */ | |
4ad49000 | 1438 | u->cgroup_realized = true; |
efdb0237 | 1439 | u->cgroup_realized_mask = target_mask; |
ccf78df1 | 1440 | u->cgroup_enabled_mask = enable_mask; |
906c06f6 | 1441 | u->cgroup_bpf_state = needs_bpf ? UNIT_CGROUP_BPF_ON : UNIT_CGROUP_BPF_OFF; |
4ad49000 | 1442 | |
0cd385d3 LP |
1443 | if (u->type != UNIT_SLICE && !c->delegate) { |
1444 | ||
1445 | /* Then, possibly move things over, but not if | |
1446 | * subgroups may contain processes, which is the case | |
1447 | * for slice and delegation units. */ | |
1448 | r = cg_migrate_everywhere(u->manager->cgroup_supported, u->cgroup_path, u->cgroup_path, migrate_callback, u); | |
1449 | if (r < 0) | |
efdb0237 | 1450 | log_unit_warning_errno(u, r, "Failed to migrate cgroup from to %s, ignoring: %m", u->cgroup_path); |
0cd385d3 | 1451 | } |
03b90d4b | 1452 | |
64747e2d LP |
1453 | return 0; |
1454 | } | |
1455 | ||
7b3fd631 LP |
1456 | int unit_attach_pids_to_cgroup(Unit *u) { |
1457 | int r; | |
1458 | assert(u); | |
1459 | ||
1460 | r = unit_realize_cgroup(u); | |
1461 | if (r < 0) | |
1462 | return r; | |
1463 | ||
1464 | r = cg_attach_many_everywhere(u->manager->cgroup_supported, u->cgroup_path, u->pids, migrate_callback, u); | |
1465 | if (r < 0) | |
1466 | return r; | |
1467 | ||
1468 | return 0; | |
1469 | } | |
1470 | ||
4b58153d LP |
1471 | static void cgroup_xattr_apply(Unit *u) { |
1472 | char ids[SD_ID128_STRING_MAX]; | |
1473 | int r; | |
1474 | ||
1475 | assert(u); | |
1476 | ||
1477 | if (!MANAGER_IS_SYSTEM(u->manager)) | |
1478 | return; | |
1479 | ||
1480 | if (sd_id128_is_null(u->invocation_id)) | |
1481 | return; | |
1482 | ||
1483 | r = cg_set_xattr(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path, | |
1484 | "trusted.invocation_id", | |
1485 | sd_id128_to_string(u->invocation_id, ids), 32, | |
1486 | 0); | |
1487 | if (r < 0) | |
1488 | log_unit_warning_errno(u, r, "Failed to set invocation ID on control group %s, ignoring: %m", u->cgroup_path); | |
1489 | } | |
1490 | ||
906c06f6 DM |
1491 | static bool unit_has_mask_realized( |
1492 | Unit *u, | |
1493 | CGroupMask target_mask, | |
1494 | CGroupMask enable_mask, | |
1495 | bool needs_bpf) { | |
1496 | ||
bc432dc7 LP |
1497 | assert(u); |
1498 | ||
906c06f6 DM |
1499 | return u->cgroup_realized && |
1500 | u->cgroup_realized_mask == target_mask && | |
1501 | u->cgroup_enabled_mask == enable_mask && | |
1502 | ((needs_bpf && u->cgroup_bpf_state == UNIT_CGROUP_BPF_ON) || | |
1503 | (!needs_bpf && u->cgroup_bpf_state == UNIT_CGROUP_BPF_OFF)); | |
6414b7c9 DS |
1504 | } |
1505 | ||
1506 | /* Check if necessary controllers and attributes for a unit are in place. | |
1507 | * | |
1508 | * If so, do nothing. | |
1509 | * If not, create paths, move processes over, and set attributes. | |
1510 | * | |
1511 | * Returns 0 on success and < 0 on failure. */ | |
db785129 | 1512 | static int unit_realize_cgroup_now(Unit *u, ManagerState state) { |
efdb0237 | 1513 | CGroupMask target_mask, enable_mask; |
906c06f6 | 1514 | bool needs_bpf, apply_bpf; |
6414b7c9 | 1515 | int r; |
64747e2d | 1516 | |
4ad49000 | 1517 | assert(u); |
64747e2d | 1518 | |
91a6073e LP |
1519 | if (u->in_cgroup_realize_queue) { |
1520 | LIST_REMOVE(cgroup_realize_queue, u->manager->cgroup_realize_queue, u); | |
1521 | u->in_cgroup_realize_queue = false; | |
4ad49000 | 1522 | } |
64747e2d | 1523 | |
efdb0237 | 1524 | target_mask = unit_get_target_mask(u); |
ccf78df1 | 1525 | enable_mask = unit_get_enable_mask(u); |
906c06f6 | 1526 | needs_bpf = unit_get_needs_bpf(u); |
ccf78df1 | 1527 | |
906c06f6 | 1528 | if (unit_has_mask_realized(u, target_mask, enable_mask, needs_bpf)) |
0a1eb06d | 1529 | return 0; |
64747e2d | 1530 | |
906c06f6 DM |
1531 | /* Make sure we apply the BPF filters either when one is configured, or if none is configured but previously |
1532 | * the state was anything but off. This way, if a unit with a BPF filter applied is reconfigured to lose it | |
1533 | * this will trickle down properly to cgroupfs. */ | |
1534 | apply_bpf = needs_bpf || u->cgroup_bpf_state != UNIT_CGROUP_BPF_OFF; | |
1535 | ||
4ad49000 | 1536 | /* First, realize parents */ |
6414b7c9 | 1537 | if (UNIT_ISSET(u->slice)) { |
db785129 | 1538 | r = unit_realize_cgroup_now(UNIT_DEREF(u->slice), state); |
6414b7c9 DS |
1539 | if (r < 0) |
1540 | return r; | |
1541 | } | |
4ad49000 LP |
1542 | |
1543 | /* And then do the real work */ | |
906c06f6 | 1544 | r = unit_create_cgroup(u, target_mask, enable_mask, needs_bpf); |
6414b7c9 DS |
1545 | if (r < 0) |
1546 | return r; | |
1547 | ||
1548 | /* Finally, apply the necessary attributes. */ | |
906c06f6 | 1549 | cgroup_context_apply(u, target_mask, apply_bpf, state); |
4b58153d | 1550 | cgroup_xattr_apply(u); |
6414b7c9 DS |
1551 | |
1552 | return 0; | |
64747e2d LP |
1553 | } |
1554 | ||
91a6073e | 1555 | static void unit_add_to_cgroup_realize_queue(Unit *u) { |
58d83430 | 1556 | assert(u); |
ecedd90f | 1557 | |
91a6073e | 1558 | if (u->in_cgroup_realize_queue) |
4ad49000 | 1559 | return; |
8e274523 | 1560 | |
91a6073e LP |
1561 | LIST_PREPEND(cgroup_realize_queue, u->manager->cgroup_realize_queue, u); |
1562 | u->in_cgroup_realize_queue = true; | |
4ad49000 | 1563 | } |
8c6db833 | 1564 | |
91a6073e | 1565 | unsigned manager_dispatch_cgroup_realize_queue(Manager *m) { |
db785129 | 1566 | ManagerState state; |
4ad49000 | 1567 | unsigned n = 0; |
db785129 | 1568 | Unit *i; |
6414b7c9 | 1569 | int r; |
ecedd90f | 1570 | |
91a6073e LP |
1571 | assert(m); |
1572 | ||
db785129 LP |
1573 | state = manager_state(m); |
1574 | ||
91a6073e LP |
1575 | while ((i = m->cgroup_realize_queue)) { |
1576 | assert(i->in_cgroup_realize_queue); | |
ecedd90f | 1577 | |
db785129 | 1578 | r = unit_realize_cgroup_now(i, state); |
6414b7c9 | 1579 | if (r < 0) |
efdb0237 | 1580 | log_warning_errno(r, "Failed to realize cgroups for queued unit %s, ignoring: %m", i->id); |
0a1eb06d | 1581 | |
4ad49000 LP |
1582 | n++; |
1583 | } | |
ecedd90f | 1584 | |
4ad49000 | 1585 | return n; |
8e274523 LP |
1586 | } |
1587 | ||
91a6073e | 1588 | static void unit_add_siblings_to_cgroup_realize_queue(Unit *u) { |
4ad49000 | 1589 | Unit *slice; |
ca949c9d | 1590 | |
4ad49000 LP |
1591 | /* This adds the siblings of the specified unit and the |
1592 | * siblings of all parent units to the cgroup queue. (But | |
1593 | * neither the specified unit itself nor the parents.) */ | |
1594 | ||
1595 | while ((slice = UNIT_DEREF(u->slice))) { | |
1596 | Iterator i; | |
1597 | Unit *m; | |
eef85c4a | 1598 | void *v; |
8f53a7b8 | 1599 | |
eef85c4a | 1600 | HASHMAP_FOREACH_KEY(v, m, u->dependencies[UNIT_BEFORE], i) { |
4ad49000 LP |
1601 | if (m == u) |
1602 | continue; | |
8e274523 | 1603 | |
6414b7c9 DS |
1604 | /* Skip units that have a dependency on the slice |
1605 | * but aren't actually in it. */ | |
4ad49000 | 1606 | if (UNIT_DEREF(m->slice) != slice) |
50159e6a | 1607 | continue; |
8e274523 | 1608 | |
6414b7c9 DS |
1609 | /* No point in doing cgroup application for units |
1610 | * without active processes. */ | |
1611 | if (UNIT_IS_INACTIVE_OR_FAILED(unit_active_state(m))) | |
1612 | continue; | |
1613 | ||
1614 | /* If the unit doesn't need any new controllers | |
1615 | * and has current ones realized, it doesn't need | |
1616 | * any changes. */ | |
906c06f6 DM |
1617 | if (unit_has_mask_realized(m, |
1618 | unit_get_target_mask(m), | |
1619 | unit_get_enable_mask(m), | |
1620 | unit_get_needs_bpf(m))) | |
6414b7c9 DS |
1621 | continue; |
1622 | ||
91a6073e | 1623 | unit_add_to_cgroup_realize_queue(m); |
50159e6a LP |
1624 | } |
1625 | ||
4ad49000 | 1626 | u = slice; |
8e274523 | 1627 | } |
4ad49000 LP |
1628 | } |
1629 | ||
0a1eb06d | 1630 | int unit_realize_cgroup(Unit *u) { |
4ad49000 LP |
1631 | assert(u); |
1632 | ||
35b7ff80 | 1633 | if (!UNIT_HAS_CGROUP_CONTEXT(u)) |
0a1eb06d | 1634 | return 0; |
8e274523 | 1635 | |
4ad49000 LP |
1636 | /* So, here's the deal: when realizing the cgroups for this |
1637 | * unit, we need to first create all parents, but there's more | |
1638 | * actually: for the weight-based controllers we also need to | |
1639 | * make sure that all our siblings (i.e. units that are in the | |
73e231ab | 1640 | * same slice as we are) have cgroups, too. Otherwise, things |
4ad49000 LP |
1641 | * would become very uneven as each of their processes would |
1642 | * get as much resources as all our group together. This call | |
1643 | * will synchronously create the parent cgroups, but will | |
1644 | * defer work on the siblings to the next event loop | |
1645 | * iteration. */ | |
ca949c9d | 1646 | |
4ad49000 | 1647 | /* Add all sibling slices to the cgroup queue. */ |
91a6073e | 1648 | unit_add_siblings_to_cgroup_realize_queue(u); |
4ad49000 | 1649 | |
6414b7c9 | 1650 | /* And realize this one now (and apply the values) */ |
db785129 | 1651 | return unit_realize_cgroup_now(u, manager_state(u->manager)); |
8e274523 LP |
1652 | } |
1653 | ||
efdb0237 LP |
1654 | void unit_release_cgroup(Unit *u) { |
1655 | assert(u); | |
1656 | ||
1657 | /* Forgets all cgroup details for this cgroup */ | |
1658 | ||
1659 | if (u->cgroup_path) { | |
1660 | (void) hashmap_remove(u->manager->cgroup_unit, u->cgroup_path); | |
1661 | u->cgroup_path = mfree(u->cgroup_path); | |
1662 | } | |
1663 | ||
1664 | if (u->cgroup_inotify_wd >= 0) { | |
1665 | if (inotify_rm_watch(u->manager->cgroup_inotify_fd, u->cgroup_inotify_wd) < 0) | |
1666 | log_unit_debug_errno(u, errno, "Failed to remove cgroup inotify watch %i for %s, ignoring", u->cgroup_inotify_wd, u->id); | |
1667 | ||
1668 | (void) hashmap_remove(u->manager->cgroup_inotify_wd_unit, INT_TO_PTR(u->cgroup_inotify_wd)); | |
1669 | u->cgroup_inotify_wd = -1; | |
1670 | } | |
1671 | } | |
1672 | ||
1673 | void unit_prune_cgroup(Unit *u) { | |
8e274523 | 1674 | int r; |
efdb0237 | 1675 | bool is_root_slice; |
8e274523 | 1676 | |
4ad49000 | 1677 | assert(u); |
8e274523 | 1678 | |
efdb0237 LP |
1679 | /* Removes the cgroup, if empty and possible, and stops watching it. */ |
1680 | ||
4ad49000 LP |
1681 | if (!u->cgroup_path) |
1682 | return; | |
8e274523 | 1683 | |
fe700f46 LP |
1684 | (void) unit_get_cpu_usage(u, NULL); /* Cache the last CPU usage value before we destroy the cgroup */ |
1685 | ||
efdb0237 LP |
1686 | is_root_slice = unit_has_name(u, SPECIAL_ROOT_SLICE); |
1687 | ||
1688 | r = cg_trim_everywhere(u->manager->cgroup_supported, u->cgroup_path, !is_root_slice); | |
dab5bf85 | 1689 | if (r < 0) { |
f29ff115 | 1690 | log_unit_debug_errno(u, r, "Failed to destroy cgroup %s, ignoring: %m", u->cgroup_path); |
dab5bf85 RL |
1691 | return; |
1692 | } | |
8e274523 | 1693 | |
efdb0237 LP |
1694 | if (is_root_slice) |
1695 | return; | |
1696 | ||
1697 | unit_release_cgroup(u); | |
0a1eb06d | 1698 | |
4ad49000 | 1699 | u->cgroup_realized = false; |
bc432dc7 | 1700 | u->cgroup_realized_mask = 0; |
ccf78df1 | 1701 | u->cgroup_enabled_mask = 0; |
8e274523 LP |
1702 | } |
1703 | ||
efdb0237 | 1704 | int unit_search_main_pid(Unit *u, pid_t *ret) { |
4ad49000 LP |
1705 | _cleanup_fclose_ FILE *f = NULL; |
1706 | pid_t pid = 0, npid, mypid; | |
efdb0237 | 1707 | int r; |
4ad49000 LP |
1708 | |
1709 | assert(u); | |
efdb0237 | 1710 | assert(ret); |
4ad49000 LP |
1711 | |
1712 | if (!u->cgroup_path) | |
efdb0237 | 1713 | return -ENXIO; |
4ad49000 | 1714 | |
efdb0237 LP |
1715 | r = cg_enumerate_processes(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path, &f); |
1716 | if (r < 0) | |
1717 | return r; | |
4ad49000 | 1718 | |
df0ff127 | 1719 | mypid = getpid_cached(); |
4ad49000 LP |
1720 | while (cg_read_pid(f, &npid) > 0) { |
1721 | pid_t ppid; | |
1722 | ||
1723 | if (npid == pid) | |
1724 | continue; | |
8e274523 | 1725 | |
4ad49000 | 1726 | /* Ignore processes that aren't our kids */ |
6bc73acb | 1727 | if (get_process_ppid(npid, &ppid) >= 0 && ppid != mypid) |
4ad49000 | 1728 | continue; |
8e274523 | 1729 | |
efdb0237 | 1730 | if (pid != 0) |
4ad49000 LP |
1731 | /* Dang, there's more than one daemonized PID |
1732 | in this group, so we don't know what process | |
1733 | is the main process. */ | |
efdb0237 LP |
1734 | |
1735 | return -ENODATA; | |
8e274523 | 1736 | |
4ad49000 | 1737 | pid = npid; |
8e274523 LP |
1738 | } |
1739 | ||
efdb0237 LP |
1740 | *ret = pid; |
1741 | return 0; | |
1742 | } | |
1743 | ||
1744 | static int unit_watch_pids_in_path(Unit *u, const char *path) { | |
b3c5bad3 | 1745 | _cleanup_closedir_ DIR *d = NULL; |
efdb0237 LP |
1746 | _cleanup_fclose_ FILE *f = NULL; |
1747 | int ret = 0, r; | |
1748 | ||
1749 | assert(u); | |
1750 | assert(path); | |
1751 | ||
1752 | r = cg_enumerate_processes(SYSTEMD_CGROUP_CONTROLLER, path, &f); | |
1753 | if (r < 0) | |
1754 | ret = r; | |
1755 | else { | |
1756 | pid_t pid; | |
1757 | ||
1758 | while ((r = cg_read_pid(f, &pid)) > 0) { | |
1759 | r = unit_watch_pid(u, pid); | |
1760 | if (r < 0 && ret >= 0) | |
1761 | ret = r; | |
1762 | } | |
1763 | ||
1764 | if (r < 0 && ret >= 0) | |
1765 | ret = r; | |
1766 | } | |
1767 | ||
1768 | r = cg_enumerate_subgroups(SYSTEMD_CGROUP_CONTROLLER, path, &d); | |
1769 | if (r < 0) { | |
1770 | if (ret >= 0) | |
1771 | ret = r; | |
1772 | } else { | |
1773 | char *fn; | |
1774 | ||
1775 | while ((r = cg_read_subgroup(d, &fn)) > 0) { | |
1776 | _cleanup_free_ char *p = NULL; | |
1777 | ||
605405c6 | 1778 | p = strjoin(path, "/", fn); |
efdb0237 LP |
1779 | free(fn); |
1780 | ||
1781 | if (!p) | |
1782 | return -ENOMEM; | |
1783 | ||
1784 | r = unit_watch_pids_in_path(u, p); | |
1785 | if (r < 0 && ret >= 0) | |
1786 | ret = r; | |
1787 | } | |
1788 | ||
1789 | if (r < 0 && ret >= 0) | |
1790 | ret = r; | |
1791 | } | |
1792 | ||
1793 | return ret; | |
1794 | } | |
1795 | ||
1796 | int unit_watch_all_pids(Unit *u) { | |
b4cccbc1 LP |
1797 | int r; |
1798 | ||
efdb0237 LP |
1799 | assert(u); |
1800 | ||
1801 | /* Adds all PIDs from our cgroup to the set of PIDs we | |
1802 | * watch. This is a fallback logic for cases where we do not | |
1803 | * get reliable cgroup empty notifications: we try to use | |
1804 | * SIGCHLD as replacement. */ | |
1805 | ||
1806 | if (!u->cgroup_path) | |
1807 | return -ENOENT; | |
1808 | ||
c22800e4 | 1809 | r = cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER); |
b4cccbc1 LP |
1810 | if (r < 0) |
1811 | return r; | |
1812 | if (r > 0) /* On unified we can use proper notifications */ | |
efdb0237 LP |
1813 | return 0; |
1814 | ||
1815 | return unit_watch_pids_in_path(u, u->cgroup_path); | |
1816 | } | |
1817 | ||
09e24654 LP |
1818 | static int on_cgroup_empty_event(sd_event_source *s, void *userdata) { |
1819 | Manager *m = userdata; | |
1820 | Unit *u; | |
efdb0237 LP |
1821 | int r; |
1822 | ||
09e24654 LP |
1823 | assert(s); |
1824 | assert(m); | |
efdb0237 | 1825 | |
09e24654 LP |
1826 | u = m->cgroup_empty_queue; |
1827 | if (!u) | |
efdb0237 LP |
1828 | return 0; |
1829 | ||
09e24654 LP |
1830 | assert(u->in_cgroup_empty_queue); |
1831 | u->in_cgroup_empty_queue = false; | |
1832 | LIST_REMOVE(cgroup_empty_queue, m->cgroup_empty_queue, u); | |
1833 | ||
1834 | if (m->cgroup_empty_queue) { | |
1835 | /* More stuff queued, let's make sure we remain enabled */ | |
1836 | r = sd_event_source_set_enabled(s, SD_EVENT_ONESHOT); | |
1837 | if (r < 0) | |
1838 | log_debug_errno(r, "Failed to reenable cgroup empty event source: %m"); | |
1839 | } | |
efdb0237 LP |
1840 | |
1841 | unit_add_to_gc_queue(u); | |
1842 | ||
1843 | if (UNIT_VTABLE(u)->notify_cgroup_empty) | |
1844 | UNIT_VTABLE(u)->notify_cgroup_empty(u); | |
1845 | ||
1846 | return 0; | |
1847 | } | |
1848 | ||
09e24654 LP |
1849 | void unit_add_to_cgroup_empty_queue(Unit *u) { |
1850 | int r; | |
1851 | ||
1852 | assert(u); | |
1853 | ||
1854 | /* Note that there are four different ways how cgroup empty events reach us: | |
1855 | * | |
1856 | * 1. On the unified hierarchy we get an inotify event on the cgroup | |
1857 | * | |
1858 | * 2. On the legacy hierarchy, when running in system mode, we get a datagram on the cgroup agent socket | |
1859 | * | |
1860 | * 3. On the legacy hierarchy, when running in user mode, we get a D-Bus signal on the system bus | |
1861 | * | |
1862 | * 4. On the legacy hierarchy, in service units we start watching all processes of the cgroup for SIGCHLD as | |
1863 | * soon as we get one SIGCHLD, to deal with unreliable cgroup notifications. | |
1864 | * | |
1865 | * Regardless which way we got the notification, we'll verify it here, and then add it to a separate | |
1866 | * queue. This queue will be dispatched at a lower priority than the SIGCHLD handler, so that we always use | |
1867 | * SIGCHLD if we can get it first, and only use the cgroup empty notifications if there's no SIGCHLD pending | |
1868 | * (which might happen if the cgroup doesn't contain processes that are our own child, which is typically the | |
1869 | * case for scope units). */ | |
1870 | ||
1871 | if (u->in_cgroup_empty_queue) | |
1872 | return; | |
1873 | ||
1874 | /* Let's verify that the cgroup is really empty */ | |
1875 | if (!u->cgroup_path) | |
1876 | return; | |
1877 | r = cg_is_empty_recursive(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path); | |
1878 | if (r < 0) { | |
1879 | log_unit_debug_errno(u, r, "Failed to determine whether cgroup %s is empty: %m", u->cgroup_path); | |
1880 | return; | |
1881 | } | |
1882 | if (r == 0) | |
1883 | return; | |
1884 | ||
1885 | LIST_PREPEND(cgroup_empty_queue, u->manager->cgroup_empty_queue, u); | |
1886 | u->in_cgroup_empty_queue = true; | |
1887 | ||
1888 | /* Trigger the defer event */ | |
1889 | r = sd_event_source_set_enabled(u->manager->cgroup_empty_event_source, SD_EVENT_ONESHOT); | |
1890 | if (r < 0) | |
1891 | log_debug_errno(r, "Failed to enable cgroup empty event source: %m"); | |
1892 | } | |
1893 | ||
efdb0237 LP |
1894 | static int on_cgroup_inotify_event(sd_event_source *s, int fd, uint32_t revents, void *userdata) { |
1895 | Manager *m = userdata; | |
1896 | ||
1897 | assert(s); | |
1898 | assert(fd >= 0); | |
1899 | assert(m); | |
1900 | ||
1901 | for (;;) { | |
1902 | union inotify_event_buffer buffer; | |
1903 | struct inotify_event *e; | |
1904 | ssize_t l; | |
1905 | ||
1906 | l = read(fd, &buffer, sizeof(buffer)); | |
1907 | if (l < 0) { | |
47249640 | 1908 | if (IN_SET(errno, EINTR, EAGAIN)) |
efdb0237 LP |
1909 | return 0; |
1910 | ||
1911 | return log_error_errno(errno, "Failed to read control group inotify events: %m"); | |
1912 | } | |
1913 | ||
1914 | FOREACH_INOTIFY_EVENT(e, buffer, l) { | |
1915 | Unit *u; | |
1916 | ||
1917 | if (e->wd < 0) | |
1918 | /* Queue overflow has no watch descriptor */ | |
1919 | continue; | |
1920 | ||
1921 | if (e->mask & IN_IGNORED) | |
1922 | /* The watch was just removed */ | |
1923 | continue; | |
1924 | ||
1925 | u = hashmap_get(m->cgroup_inotify_wd_unit, INT_TO_PTR(e->wd)); | |
1926 | if (!u) /* Not that inotify might deliver | |
1927 | * events for a watch even after it | |
1928 | * was removed, because it was queued | |
1929 | * before the removal. Let's ignore | |
1930 | * this here safely. */ | |
1931 | continue; | |
1932 | ||
09e24654 | 1933 | unit_add_to_cgroup_empty_queue(u); |
efdb0237 LP |
1934 | } |
1935 | } | |
8e274523 LP |
1936 | } |
1937 | ||
8e274523 | 1938 | int manager_setup_cgroup(Manager *m) { |
9444b1f2 | 1939 | _cleanup_free_ char *path = NULL; |
10bd3e2e | 1940 | const char *scope_path; |
efdb0237 | 1941 | CGroupController c; |
b4cccbc1 | 1942 | int r, all_unified; |
efdb0237 | 1943 | char *e; |
8e274523 LP |
1944 | |
1945 | assert(m); | |
1946 | ||
35d2e7ec | 1947 | /* 1. Determine hierarchy */ |
efdb0237 | 1948 | m->cgroup_root = mfree(m->cgroup_root); |
9444b1f2 | 1949 | r = cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER, 0, &m->cgroup_root); |
23bbb0de MS |
1950 | if (r < 0) |
1951 | return log_error_errno(r, "Cannot determine cgroup we are running in: %m"); | |
8e274523 | 1952 | |
efdb0237 LP |
1953 | /* Chop off the init scope, if we are already located in it */ |
1954 | e = endswith(m->cgroup_root, "/" SPECIAL_INIT_SCOPE); | |
0d8c31ff | 1955 | |
efdb0237 LP |
1956 | /* LEGACY: Also chop off the system slice if we are in |
1957 | * it. This is to support live upgrades from older systemd | |
1958 | * versions where PID 1 was moved there. Also see | |
1959 | * cg_get_root_path(). */ | |
463d0d15 | 1960 | if (!e && MANAGER_IS_SYSTEM(m)) { |
9444b1f2 | 1961 | e = endswith(m->cgroup_root, "/" SPECIAL_SYSTEM_SLICE); |
15c60e99 | 1962 | if (!e) |
efdb0237 | 1963 | e = endswith(m->cgroup_root, "/system"); /* even more legacy */ |
0baf24dd | 1964 | } |
efdb0237 LP |
1965 | if (e) |
1966 | *e = 0; | |
7ccfb64a | 1967 | |
7546145e LP |
1968 | /* And make sure to store away the root value without trailing slash, even for the root dir, so that we can |
1969 | * easily prepend it everywhere. */ | |
1970 | delete_trailing_chars(m->cgroup_root, "/"); | |
8e274523 | 1971 | |
35d2e7ec | 1972 | /* 2. Show data */ |
9444b1f2 | 1973 | r = cg_get_path(SYSTEMD_CGROUP_CONTROLLER, m->cgroup_root, NULL, &path); |
23bbb0de MS |
1974 | if (r < 0) |
1975 | return log_error_errno(r, "Cannot find cgroup mount point: %m"); | |
8e274523 | 1976 | |
415fc41c TH |
1977 | r = cg_unified_flush(); |
1978 | if (r < 0) | |
1979 | return log_error_errno(r, "Couldn't determine if we are running in the unified hierarchy: %m"); | |
5da38d07 | 1980 | |
b4cccbc1 | 1981 | all_unified = cg_all_unified(); |
d4c819ed ZJS |
1982 | if (all_unified < 0) |
1983 | return log_error_errno(all_unified, "Couldn't determine whether we are in all unified mode: %m"); | |
1984 | if (all_unified > 0) | |
efdb0237 | 1985 | log_debug("Unified cgroup hierarchy is located at %s.", path); |
b4cccbc1 | 1986 | else { |
c22800e4 | 1987 | r = cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER); |
b4cccbc1 LP |
1988 | if (r < 0) |
1989 | return log_error_errno(r, "Failed to determine whether systemd's own controller is in unified mode: %m"); | |
1990 | if (r > 0) | |
1991 | log_debug("Unified cgroup hierarchy is located at %s. Controllers are on legacy hierarchies.", path); | |
1992 | else | |
1993 | log_debug("Using cgroup controller " SYSTEMD_CGROUP_CONTROLLER_LEGACY ". File system hierarchy is at %s.", path); | |
1994 | } | |
efdb0237 | 1995 | |
09e24654 LP |
1996 | /* 3. Allocate cgroup empty defer event source */ |
1997 | m->cgroup_empty_event_source = sd_event_source_unref(m->cgroup_empty_event_source); | |
1998 | r = sd_event_add_defer(m->event, &m->cgroup_empty_event_source, on_cgroup_empty_event, m); | |
1999 | if (r < 0) | |
2000 | return log_error_errno(r, "Failed to create cgroup empty event source: %m"); | |
2001 | ||
2002 | r = sd_event_source_set_priority(m->cgroup_empty_event_source, SD_EVENT_PRIORITY_NORMAL-5); | |
2003 | if (r < 0) | |
2004 | return log_error_errno(r, "Failed to set priority of cgroup empty event source: %m"); | |
2005 | ||
2006 | r = sd_event_source_set_enabled(m->cgroup_empty_event_source, SD_EVENT_OFF); | |
2007 | if (r < 0) | |
2008 | return log_error_errno(r, "Failed to disable cgroup empty event source: %m"); | |
2009 | ||
2010 | (void) sd_event_source_set_description(m->cgroup_empty_event_source, "cgroup-empty"); | |
2011 | ||
2012 | /* 4. Install notifier inotify object, or agent */ | |
10bd3e2e | 2013 | if (cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER) > 0) { |
c6c18be3 | 2014 | |
09e24654 | 2015 | /* In the unified hierarchy we can get cgroup empty notifications via inotify. */ |
efdb0237 | 2016 | |
10bd3e2e LP |
2017 | m->cgroup_inotify_event_source = sd_event_source_unref(m->cgroup_inotify_event_source); |
2018 | safe_close(m->cgroup_inotify_fd); | |
efdb0237 | 2019 | |
10bd3e2e LP |
2020 | m->cgroup_inotify_fd = inotify_init1(IN_NONBLOCK|IN_CLOEXEC); |
2021 | if (m->cgroup_inotify_fd < 0) | |
2022 | return log_error_errno(errno, "Failed to create control group inotify object: %m"); | |
efdb0237 | 2023 | |
10bd3e2e LP |
2024 | r = sd_event_add_io(m->event, &m->cgroup_inotify_event_source, m->cgroup_inotify_fd, EPOLLIN, on_cgroup_inotify_event, m); |
2025 | if (r < 0) | |
2026 | return log_error_errno(r, "Failed to watch control group inotify object: %m"); | |
efdb0237 | 2027 | |
10bd3e2e LP |
2028 | /* Process cgroup empty notifications early, but after service notifications and SIGCHLD. Also |
2029 | * see handling of cgroup agent notifications, for the classic cgroup hierarchy support. */ | |
09e24654 | 2030 | r = sd_event_source_set_priority(m->cgroup_inotify_event_source, SD_EVENT_PRIORITY_NORMAL-4); |
10bd3e2e LP |
2031 | if (r < 0) |
2032 | return log_error_errno(r, "Failed to set priority of inotify event source: %m"); | |
efdb0237 | 2033 | |
10bd3e2e | 2034 | (void) sd_event_source_set_description(m->cgroup_inotify_event_source, "cgroup-inotify"); |
efdb0237 | 2035 | |
10bd3e2e | 2036 | } else if (MANAGER_IS_SYSTEM(m) && m->test_run_flags == 0) { |
efdb0237 | 2037 | |
10bd3e2e LP |
2038 | /* On the legacy hierarchy we only get notifications via cgroup agents. (Which isn't really reliable, |
2039 | * since it does not generate events when control groups with children run empty. */ | |
8e274523 | 2040 | |
10bd3e2e | 2041 | r = cg_install_release_agent(SYSTEMD_CGROUP_CONTROLLER, SYSTEMD_CGROUP_AGENT_PATH); |
23bbb0de | 2042 | if (r < 0) |
10bd3e2e LP |
2043 | log_warning_errno(r, "Failed to install release agent, ignoring: %m"); |
2044 | else if (r > 0) | |
2045 | log_debug("Installed release agent."); | |
2046 | else if (r == 0) | |
2047 | log_debug("Release agent already installed."); | |
2048 | } | |
efdb0237 | 2049 | |
09e24654 | 2050 | /* 5. Make sure we are in the special "init.scope" unit in the root slice. */ |
10bd3e2e LP |
2051 | scope_path = strjoina(m->cgroup_root, "/" SPECIAL_INIT_SCOPE); |
2052 | r = cg_create_and_attach(SYSTEMD_CGROUP_CONTROLLER, scope_path, 0); | |
2053 | if (r < 0) | |
2054 | return log_error_errno(r, "Failed to create %s control group: %m", scope_path); | |
c6c18be3 | 2055 | |
09e24654 | 2056 | /* Also, move all other userspace processes remaining in the root cgroup into that scope. */ |
10bd3e2e LP |
2057 | r = cg_migrate(SYSTEMD_CGROUP_CONTROLLER, m->cgroup_root, SYSTEMD_CGROUP_CONTROLLER, scope_path, 0); |
2058 | if (r < 0) | |
2059 | log_warning_errno(r, "Couldn't move remaining userspace processes, ignoring: %m"); | |
0d8c31ff | 2060 | |
09e24654 | 2061 | /* 6. And pin it, so that it cannot be unmounted */ |
10bd3e2e LP |
2062 | safe_close(m->pin_cgroupfs_fd); |
2063 | m->pin_cgroupfs_fd = open(path, O_RDONLY|O_CLOEXEC|O_DIRECTORY|O_NOCTTY|O_NONBLOCK); | |
2064 | if (m->pin_cgroupfs_fd < 0) | |
2065 | return log_error_errno(errno, "Failed to open pin file: %m"); | |
2066 | ||
09e24654 | 2067 | /* 7. Always enable hierarchical support if it exists... */ |
10bd3e2e LP |
2068 | if (!all_unified && m->test_run_flags == 0) |
2069 | (void) cg_set_attribute("memory", "/", "memory.use_hierarchy", "1"); | |
c6c18be3 | 2070 | |
09e24654 | 2071 | /* 8. Figure out which controllers are supported, and log about it */ |
efdb0237 LP |
2072 | r = cg_mask_supported(&m->cgroup_supported); |
2073 | if (r < 0) | |
2074 | return log_error_errno(r, "Failed to determine supported controllers: %m"); | |
efdb0237 | 2075 | for (c = 0; c < _CGROUP_CONTROLLER_MAX; c++) |
eee0a1e4 | 2076 | log_debug("Controller '%s' supported: %s", cgroup_controller_to_string(c), yes_no(m->cgroup_supported & CGROUP_CONTROLLER_TO_MASK(c))); |
9156e799 | 2077 | |
a32360f1 | 2078 | return 0; |
8e274523 LP |
2079 | } |
2080 | ||
c6c18be3 | 2081 | void manager_shutdown_cgroup(Manager *m, bool delete) { |
8e274523 LP |
2082 | assert(m); |
2083 | ||
9444b1f2 LP |
2084 | /* We can't really delete the group, since we are in it. But |
2085 | * let's trim it. */ | |
2086 | if (delete && m->cgroup_root) | |
efdb0237 LP |
2087 | (void) cg_trim(SYSTEMD_CGROUP_CONTROLLER, m->cgroup_root, false); |
2088 | ||
09e24654 LP |
2089 | m->cgroup_empty_event_source = sd_event_source_unref(m->cgroup_empty_event_source); |
2090 | ||
efdb0237 LP |
2091 | m->cgroup_inotify_wd_unit = hashmap_free(m->cgroup_inotify_wd_unit); |
2092 | ||
2093 | m->cgroup_inotify_event_source = sd_event_source_unref(m->cgroup_inotify_event_source); | |
2094 | m->cgroup_inotify_fd = safe_close(m->cgroup_inotify_fd); | |
8e274523 | 2095 | |
03e334a1 | 2096 | m->pin_cgroupfs_fd = safe_close(m->pin_cgroupfs_fd); |
c6c18be3 | 2097 | |
efdb0237 | 2098 | m->cgroup_root = mfree(m->cgroup_root); |
8e274523 LP |
2099 | } |
2100 | ||
4ad49000 | 2101 | Unit* manager_get_unit_by_cgroup(Manager *m, const char *cgroup) { |
acb14d31 | 2102 | char *p; |
4ad49000 | 2103 | Unit *u; |
acb14d31 LP |
2104 | |
2105 | assert(m); | |
2106 | assert(cgroup); | |
acb14d31 | 2107 | |
4ad49000 LP |
2108 | u = hashmap_get(m->cgroup_unit, cgroup); |
2109 | if (u) | |
2110 | return u; | |
acb14d31 | 2111 | |
8e70580b | 2112 | p = strdupa(cgroup); |
acb14d31 LP |
2113 | for (;;) { |
2114 | char *e; | |
2115 | ||
2116 | e = strrchr(p, '/'); | |
efdb0237 LP |
2117 | if (!e || e == p) |
2118 | return hashmap_get(m->cgroup_unit, SPECIAL_ROOT_SLICE); | |
acb14d31 LP |
2119 | |
2120 | *e = 0; | |
2121 | ||
4ad49000 LP |
2122 | u = hashmap_get(m->cgroup_unit, p); |
2123 | if (u) | |
2124 | return u; | |
acb14d31 LP |
2125 | } |
2126 | } | |
2127 | ||
b3ac818b | 2128 | Unit *manager_get_unit_by_pid_cgroup(Manager *m, pid_t pid) { |
4ad49000 | 2129 | _cleanup_free_ char *cgroup = NULL; |
acb14d31 | 2130 | int r; |
8e274523 | 2131 | |
8c47c732 LP |
2132 | assert(m); |
2133 | ||
b3ac818b LP |
2134 | if (pid <= 0) |
2135 | return NULL; | |
2136 | ||
2137 | r = cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER, pid, &cgroup); | |
2138 | if (r < 0) | |
2139 | return NULL; | |
2140 | ||
2141 | return manager_get_unit_by_cgroup(m, cgroup); | |
2142 | } | |
2143 | ||
2144 | Unit *manager_get_unit_by_pid(Manager *m, pid_t pid) { | |
2145 | Unit *u; | |
2146 | ||
2147 | assert(m); | |
2148 | ||
efdb0237 | 2149 | if (pid <= 0) |
8c47c732 LP |
2150 | return NULL; |
2151 | ||
efdb0237 LP |
2152 | if (pid == 1) |
2153 | return hashmap_get(m->units, SPECIAL_INIT_SCOPE); | |
2154 | ||
fea72cc0 | 2155 | u = hashmap_get(m->watch_pids1, PID_TO_PTR(pid)); |
5fe8876b LP |
2156 | if (u) |
2157 | return u; | |
2158 | ||
fea72cc0 | 2159 | u = hashmap_get(m->watch_pids2, PID_TO_PTR(pid)); |
5fe8876b LP |
2160 | if (u) |
2161 | return u; | |
2162 | ||
b3ac818b | 2163 | return manager_get_unit_by_pid_cgroup(m, pid); |
6dde1f33 | 2164 | } |
4fbf50b3 | 2165 | |
4ad49000 LP |
2166 | int manager_notify_cgroup_empty(Manager *m, const char *cgroup) { |
2167 | Unit *u; | |
4fbf50b3 | 2168 | |
4ad49000 LP |
2169 | assert(m); |
2170 | assert(cgroup); | |
4fbf50b3 | 2171 | |
09e24654 LP |
2172 | /* Called on the legacy hierarchy whenever we get an explicit cgroup notification from the cgroup agent process |
2173 | * or from the --system instance */ | |
2174 | ||
d8fdc620 LP |
2175 | log_debug("Got cgroup empty notification for: %s", cgroup); |
2176 | ||
4ad49000 | 2177 | u = manager_get_unit_by_cgroup(m, cgroup); |
5ad096b3 LP |
2178 | if (!u) |
2179 | return 0; | |
b56c28c3 | 2180 | |
09e24654 LP |
2181 | unit_add_to_cgroup_empty_queue(u); |
2182 | return 1; | |
5ad096b3 LP |
2183 | } |
2184 | ||
2185 | int unit_get_memory_current(Unit *u, uint64_t *ret) { | |
2186 | _cleanup_free_ char *v = NULL; | |
2187 | int r; | |
2188 | ||
2189 | assert(u); | |
2190 | assert(ret); | |
2191 | ||
2e4025c0 | 2192 | if (!UNIT_CGROUP_BOOL(u, memory_accounting)) |
cf3b4be1 LP |
2193 | return -ENODATA; |
2194 | ||
5ad096b3 LP |
2195 | if (!u->cgroup_path) |
2196 | return -ENODATA; | |
2197 | ||
efdb0237 | 2198 | if ((u->cgroup_realized_mask & CGROUP_MASK_MEMORY) == 0) |
5ad096b3 LP |
2199 | return -ENODATA; |
2200 | ||
b4cccbc1 LP |
2201 | r = cg_all_unified(); |
2202 | if (r < 0) | |
2203 | return r; | |
2204 | if (r > 0) | |
efdb0237 | 2205 | r = cg_get_attribute("memory", u->cgroup_path, "memory.current", &v); |
b4cccbc1 LP |
2206 | else |
2207 | r = cg_get_attribute("memory", u->cgroup_path, "memory.usage_in_bytes", &v); | |
5ad096b3 LP |
2208 | if (r == -ENOENT) |
2209 | return -ENODATA; | |
2210 | if (r < 0) | |
2211 | return r; | |
2212 | ||
2213 | return safe_atou64(v, ret); | |
2214 | } | |
2215 | ||
03a7b521 LP |
2216 | int unit_get_tasks_current(Unit *u, uint64_t *ret) { |
2217 | _cleanup_free_ char *v = NULL; | |
2218 | int r; | |
2219 | ||
2220 | assert(u); | |
2221 | assert(ret); | |
2222 | ||
2e4025c0 | 2223 | if (!UNIT_CGROUP_BOOL(u, tasks_accounting)) |
cf3b4be1 LP |
2224 | return -ENODATA; |
2225 | ||
03a7b521 LP |
2226 | if (!u->cgroup_path) |
2227 | return -ENODATA; | |
2228 | ||
2229 | if ((u->cgroup_realized_mask & CGROUP_MASK_PIDS) == 0) | |
2230 | return -ENODATA; | |
2231 | ||
2232 | r = cg_get_attribute("pids", u->cgroup_path, "pids.current", &v); | |
2233 | if (r == -ENOENT) | |
2234 | return -ENODATA; | |
2235 | if (r < 0) | |
2236 | return r; | |
2237 | ||
2238 | return safe_atou64(v, ret); | |
2239 | } | |
2240 | ||
5ad096b3 LP |
2241 | static int unit_get_cpu_usage_raw(Unit *u, nsec_t *ret) { |
2242 | _cleanup_free_ char *v = NULL; | |
2243 | uint64_t ns; | |
2244 | int r; | |
2245 | ||
2246 | assert(u); | |
2247 | assert(ret); | |
2248 | ||
2249 | if (!u->cgroup_path) | |
2250 | return -ENODATA; | |
2251 | ||
b4cccbc1 LP |
2252 | r = cg_all_unified(); |
2253 | if (r < 0) | |
2254 | return r; | |
2255 | if (r > 0) { | |
66ebf6c0 TH |
2256 | const char *keys[] = { "usage_usec", NULL }; |
2257 | _cleanup_free_ char *val = NULL; | |
2258 | uint64_t us; | |
5ad096b3 | 2259 | |
66ebf6c0 TH |
2260 | if ((u->cgroup_realized_mask & CGROUP_MASK_CPU) == 0) |
2261 | return -ENODATA; | |
5ad096b3 | 2262 | |
66ebf6c0 TH |
2263 | r = cg_get_keyed_attribute("cpu", u->cgroup_path, "cpu.stat", keys, &val); |
2264 | if (r < 0) | |
2265 | return r; | |
2266 | ||
2267 | r = safe_atou64(val, &us); | |
2268 | if (r < 0) | |
2269 | return r; | |
2270 | ||
2271 | ns = us * NSEC_PER_USEC; | |
2272 | } else { | |
2273 | if ((u->cgroup_realized_mask & CGROUP_MASK_CPUACCT) == 0) | |
2274 | return -ENODATA; | |
2275 | ||
2276 | r = cg_get_attribute("cpuacct", u->cgroup_path, "cpuacct.usage", &v); | |
2277 | if (r == -ENOENT) | |
2278 | return -ENODATA; | |
2279 | if (r < 0) | |
2280 | return r; | |
2281 | ||
2282 | r = safe_atou64(v, &ns); | |
2283 | if (r < 0) | |
2284 | return r; | |
2285 | } | |
5ad096b3 LP |
2286 | |
2287 | *ret = ns; | |
2288 | return 0; | |
2289 | } | |
2290 | ||
2291 | int unit_get_cpu_usage(Unit *u, nsec_t *ret) { | |
2292 | nsec_t ns; | |
2293 | int r; | |
2294 | ||
fe700f46 LP |
2295 | assert(u); |
2296 | ||
2297 | /* Retrieve the current CPU usage counter. This will subtract the CPU counter taken when the unit was | |
2298 | * started. If the cgroup has been removed already, returns the last cached value. To cache the value, simply | |
2299 | * call this function with a NULL return value. */ | |
2300 | ||
2e4025c0 | 2301 | if (!UNIT_CGROUP_BOOL(u, cpu_accounting)) |
cf3b4be1 LP |
2302 | return -ENODATA; |
2303 | ||
5ad096b3 | 2304 | r = unit_get_cpu_usage_raw(u, &ns); |
fe700f46 LP |
2305 | if (r == -ENODATA && u->cpu_usage_last != NSEC_INFINITY) { |
2306 | /* If we can't get the CPU usage anymore (because the cgroup was already removed, for example), use our | |
2307 | * cached value. */ | |
2308 | ||
2309 | if (ret) | |
2310 | *ret = u->cpu_usage_last; | |
2311 | return 0; | |
2312 | } | |
5ad096b3 LP |
2313 | if (r < 0) |
2314 | return r; | |
2315 | ||
66ebf6c0 TH |
2316 | if (ns > u->cpu_usage_base) |
2317 | ns -= u->cpu_usage_base; | |
5ad096b3 LP |
2318 | else |
2319 | ns = 0; | |
2320 | ||
fe700f46 LP |
2321 | u->cpu_usage_last = ns; |
2322 | if (ret) | |
2323 | *ret = ns; | |
2324 | ||
5ad096b3 LP |
2325 | return 0; |
2326 | } | |
2327 | ||
906c06f6 DM |
2328 | int unit_get_ip_accounting( |
2329 | Unit *u, | |
2330 | CGroupIPAccountingMetric metric, | |
2331 | uint64_t *ret) { | |
2332 | ||
6b659ed8 | 2333 | uint64_t value; |
906c06f6 DM |
2334 | int fd, r; |
2335 | ||
2336 | assert(u); | |
2337 | assert(metric >= 0); | |
2338 | assert(metric < _CGROUP_IP_ACCOUNTING_METRIC_MAX); | |
2339 | assert(ret); | |
2340 | ||
cf3b4be1 LP |
2341 | /* IP accounting is currently not recursive, and hence we refuse to return any data for slice nodes. Slices are |
2342 | * inner cgroup nodes and hence have no processes directly attached, hence their counters would be zero | |
2343 | * anyway. And if we block this now we can later open this up, if the kernel learns recursive BPF cgroup | |
2344 | * filters. */ | |
2345 | if (u->type == UNIT_SLICE) | |
2346 | return -ENODATA; | |
2347 | ||
2e4025c0 | 2348 | if (!UNIT_CGROUP_BOOL(u, ip_accounting)) |
cf3b4be1 LP |
2349 | return -ENODATA; |
2350 | ||
906c06f6 DM |
2351 | fd = IN_SET(metric, CGROUP_IP_INGRESS_BYTES, CGROUP_IP_INGRESS_PACKETS) ? |
2352 | u->ip_accounting_ingress_map_fd : | |
2353 | u->ip_accounting_egress_map_fd; | |
2354 | ||
2355 | if (fd < 0) | |
2356 | return -ENODATA; | |
2357 | ||
2358 | if (IN_SET(metric, CGROUP_IP_INGRESS_BYTES, CGROUP_IP_EGRESS_BYTES)) | |
6b659ed8 | 2359 | r = bpf_firewall_read_accounting(fd, &value, NULL); |
906c06f6 | 2360 | else |
6b659ed8 LP |
2361 | r = bpf_firewall_read_accounting(fd, NULL, &value); |
2362 | if (r < 0) | |
2363 | return r; | |
2364 | ||
2365 | /* Add in additional metrics from a previous runtime. Note that when reexecing/reloading the daemon we compile | |
2366 | * all BPF programs and maps anew, but serialize the old counters. When deserializing we store them in the | |
2367 | * ip_accounting_extra[] field, and add them in here transparently. */ | |
2368 | ||
2369 | *ret = value + u->ip_accounting_extra[metric]; | |
906c06f6 DM |
2370 | |
2371 | return r; | |
2372 | } | |
2373 | ||
2374 | int unit_reset_cpu_accounting(Unit *u) { | |
5ad096b3 LP |
2375 | nsec_t ns; |
2376 | int r; | |
2377 | ||
2378 | assert(u); | |
2379 | ||
fe700f46 LP |
2380 | u->cpu_usage_last = NSEC_INFINITY; |
2381 | ||
5ad096b3 LP |
2382 | r = unit_get_cpu_usage_raw(u, &ns); |
2383 | if (r < 0) { | |
66ebf6c0 | 2384 | u->cpu_usage_base = 0; |
5ad096b3 | 2385 | return r; |
b56c28c3 | 2386 | } |
2633eb83 | 2387 | |
66ebf6c0 | 2388 | u->cpu_usage_base = ns; |
4ad49000 | 2389 | return 0; |
4fbf50b3 LP |
2390 | } |
2391 | ||
906c06f6 DM |
2392 | int unit_reset_ip_accounting(Unit *u) { |
2393 | int r = 0, q = 0; | |
2394 | ||
2395 | assert(u); | |
2396 | ||
2397 | if (u->ip_accounting_ingress_map_fd >= 0) | |
2398 | r = bpf_firewall_reset_accounting(u->ip_accounting_ingress_map_fd); | |
2399 | ||
2400 | if (u->ip_accounting_egress_map_fd >= 0) | |
2401 | q = bpf_firewall_reset_accounting(u->ip_accounting_egress_map_fd); | |
2402 | ||
6b659ed8 LP |
2403 | zero(u->ip_accounting_extra); |
2404 | ||
906c06f6 DM |
2405 | return r < 0 ? r : q; |
2406 | } | |
2407 | ||
e7ab4d1a LP |
2408 | void unit_invalidate_cgroup(Unit *u, CGroupMask m) { |
2409 | assert(u); | |
2410 | ||
2411 | if (!UNIT_HAS_CGROUP_CONTEXT(u)) | |
2412 | return; | |
2413 | ||
2414 | if (m == 0) | |
2415 | return; | |
2416 | ||
538b4852 TH |
2417 | /* always invalidate compat pairs together */ |
2418 | if (m & (CGROUP_MASK_IO | CGROUP_MASK_BLKIO)) | |
2419 | m |= CGROUP_MASK_IO | CGROUP_MASK_BLKIO; | |
2420 | ||
7cce4fb7 LP |
2421 | if (m & (CGROUP_MASK_CPU | CGROUP_MASK_CPUACCT)) |
2422 | m |= CGROUP_MASK_CPU | CGROUP_MASK_CPUACCT; | |
2423 | ||
e7ab4d1a LP |
2424 | if ((u->cgroup_realized_mask & m) == 0) |
2425 | return; | |
2426 | ||
2427 | u->cgroup_realized_mask &= ~m; | |
91a6073e | 2428 | unit_add_to_cgroup_realize_queue(u); |
e7ab4d1a LP |
2429 | } |
2430 | ||
906c06f6 DM |
2431 | void unit_invalidate_cgroup_bpf(Unit *u) { |
2432 | assert(u); | |
2433 | ||
2434 | if (!UNIT_HAS_CGROUP_CONTEXT(u)) | |
2435 | return; | |
2436 | ||
2437 | if (u->cgroup_bpf_state == UNIT_CGROUP_BPF_INVALIDATED) | |
2438 | return; | |
2439 | ||
2440 | u->cgroup_bpf_state = UNIT_CGROUP_BPF_INVALIDATED; | |
91a6073e | 2441 | unit_add_to_cgroup_realize_queue(u); |
906c06f6 DM |
2442 | |
2443 | /* If we are a slice unit, we also need to put compile a new BPF program for all our children, as the IP access | |
2444 | * list of our children includes our own. */ | |
2445 | if (u->type == UNIT_SLICE) { | |
2446 | Unit *member; | |
2447 | Iterator i; | |
eef85c4a | 2448 | void *v; |
906c06f6 | 2449 | |
eef85c4a | 2450 | HASHMAP_FOREACH_KEY(v, member, u->dependencies[UNIT_BEFORE], i) { |
906c06f6 DM |
2451 | if (member == u) |
2452 | continue; | |
2453 | ||
2454 | if (UNIT_DEREF(member->slice) != u) | |
2455 | continue; | |
2456 | ||
2457 | unit_invalidate_cgroup_bpf(member); | |
2458 | } | |
2459 | } | |
2460 | } | |
2461 | ||
e7ab4d1a LP |
2462 | void manager_invalidate_startup_units(Manager *m) { |
2463 | Iterator i; | |
2464 | Unit *u; | |
2465 | ||
2466 | assert(m); | |
2467 | ||
2468 | SET_FOREACH(u, m->startup_units, i) | |
13c31542 | 2469 | unit_invalidate_cgroup(u, CGROUP_MASK_CPU|CGROUP_MASK_IO|CGROUP_MASK_BLKIO); |
e7ab4d1a LP |
2470 | } |
2471 | ||
4ad49000 LP |
2472 | static const char* const cgroup_device_policy_table[_CGROUP_DEVICE_POLICY_MAX] = { |
2473 | [CGROUP_AUTO] = "auto", | |
2474 | [CGROUP_CLOSED] = "closed", | |
2475 | [CGROUP_STRICT] = "strict", | |
2476 | }; | |
4fbf50b3 | 2477 | |
4ad49000 | 2478 | DEFINE_STRING_TABLE_LOOKUP(cgroup_device_policy, CGroupDevicePolicy); |