]>
Commit | Line | Data |
---|---|---|
8e274523 LP |
1 | /*** |
2 | This file is part of systemd. | |
3 | ||
4ad49000 | 4 | Copyright 2013 Lennart Poettering |
8e274523 LP |
5 | |
6 | systemd is free software; you can redistribute it and/or modify it | |
5430f7f2 LP |
7 | under the terms of the GNU Lesser General Public License as published by |
8 | the Free Software Foundation; either version 2.1 of the License, or | |
8e274523 LP |
9 | (at your option) any later version. |
10 | ||
11 | systemd is distributed in the hope that it will be useful, but | |
12 | WITHOUT ANY WARRANTY; without even the implied warranty of | |
13 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
5430f7f2 | 14 | Lesser General Public License for more details. |
8e274523 | 15 | |
5430f7f2 | 16 | You should have received a copy of the GNU Lesser General Public License |
8e274523 LP |
17 | along with systemd; If not, see <http://www.gnu.org/licenses/>. |
18 | ***/ | |
19 | ||
c6c18be3 | 20 | #include <fcntl.h> |
e41969e3 | 21 | #include <fnmatch.h> |
8c6db833 | 22 | |
b5efdb8a | 23 | #include "alloc-util.h" |
906c06f6 | 24 | #include "bpf-firewall.h" |
03a7b521 | 25 | #include "cgroup-util.h" |
3ffd4af2 LP |
26 | #include "cgroup.h" |
27 | #include "fd-util.h" | |
0d39fa9c | 28 | #include "fileio.h" |
77601719 | 29 | #include "fs-util.h" |
6bedfcbb | 30 | #include "parse-util.h" |
9eb977db | 31 | #include "path-util.h" |
03a7b521 | 32 | #include "process-util.h" |
9444b1f2 | 33 | #include "special.h" |
906c06f6 | 34 | #include "stdio-util.h" |
8b43440b | 35 | #include "string-table.h" |
07630cea | 36 | #include "string-util.h" |
8e274523 | 37 | |
9a054909 LP |
38 | #define CGROUP_CPU_QUOTA_PERIOD_USEC ((usec_t) 100 * USEC_PER_MSEC) |
39 | ||
2b40998d | 40 | static void cgroup_compat_warn(void) { |
128fadc9 TH |
41 | static bool cgroup_compat_warned = false; |
42 | ||
43 | if (cgroup_compat_warned) | |
44 | return; | |
45 | ||
46 | log_warning("cgroup compatibility translation between legacy and unified hierarchy settings activated. See cgroup-compat debug messages for details."); | |
47 | cgroup_compat_warned = true; | |
48 | } | |
49 | ||
50 | #define log_cgroup_compat(unit, fmt, ...) do { \ | |
51 | cgroup_compat_warn(); \ | |
52 | log_unit_debug(unit, "cgroup-compat: " fmt, ##__VA_ARGS__); \ | |
2b40998d | 53 | } while (false) |
128fadc9 | 54 | |
4ad49000 LP |
55 | void cgroup_context_init(CGroupContext *c) { |
56 | assert(c); | |
57 | ||
58 | /* Initialize everything to the kernel defaults, assuming the | |
59 | * structure is preinitialized to 0 */ | |
60 | ||
66ebf6c0 TH |
61 | c->cpu_weight = CGROUP_WEIGHT_INVALID; |
62 | c->startup_cpu_weight = CGROUP_WEIGHT_INVALID; | |
63 | c->cpu_quota_per_sec_usec = USEC_INFINITY; | |
64 | ||
d53d9474 LP |
65 | c->cpu_shares = CGROUP_CPU_SHARES_INVALID; |
66 | c->startup_cpu_shares = CGROUP_CPU_SHARES_INVALID; | |
d53d9474 | 67 | |
da4d897e TH |
68 | c->memory_high = CGROUP_LIMIT_MAX; |
69 | c->memory_max = CGROUP_LIMIT_MAX; | |
96e131ea | 70 | c->memory_swap_max = CGROUP_LIMIT_MAX; |
da4d897e TH |
71 | |
72 | c->memory_limit = CGROUP_LIMIT_MAX; | |
b2f8b02e | 73 | |
13c31542 TH |
74 | c->io_weight = CGROUP_WEIGHT_INVALID; |
75 | c->startup_io_weight = CGROUP_WEIGHT_INVALID; | |
76 | ||
d53d9474 LP |
77 | c->blockio_weight = CGROUP_BLKIO_WEIGHT_INVALID; |
78 | c->startup_blockio_weight = CGROUP_BLKIO_WEIGHT_INVALID; | |
79 | ||
80 | c->tasks_max = (uint64_t) -1; | |
4ad49000 | 81 | } |
8e274523 | 82 | |
4ad49000 LP |
83 | void cgroup_context_free_device_allow(CGroupContext *c, CGroupDeviceAllow *a) { |
84 | assert(c); | |
85 | assert(a); | |
86 | ||
71fda00f | 87 | LIST_REMOVE(device_allow, c->device_allow, a); |
4ad49000 LP |
88 | free(a->path); |
89 | free(a); | |
90 | } | |
91 | ||
13c31542 TH |
92 | void cgroup_context_free_io_device_weight(CGroupContext *c, CGroupIODeviceWeight *w) { |
93 | assert(c); | |
94 | assert(w); | |
95 | ||
96 | LIST_REMOVE(device_weights, c->io_device_weights, w); | |
97 | free(w->path); | |
98 | free(w); | |
99 | } | |
100 | ||
101 | void cgroup_context_free_io_device_limit(CGroupContext *c, CGroupIODeviceLimit *l) { | |
102 | assert(c); | |
103 | assert(l); | |
104 | ||
105 | LIST_REMOVE(device_limits, c->io_device_limits, l); | |
106 | free(l->path); | |
107 | free(l); | |
108 | } | |
109 | ||
4ad49000 LP |
110 | void cgroup_context_free_blockio_device_weight(CGroupContext *c, CGroupBlockIODeviceWeight *w) { |
111 | assert(c); | |
112 | assert(w); | |
113 | ||
71fda00f | 114 | LIST_REMOVE(device_weights, c->blockio_device_weights, w); |
4ad49000 LP |
115 | free(w->path); |
116 | free(w); | |
117 | } | |
118 | ||
119 | void cgroup_context_free_blockio_device_bandwidth(CGroupContext *c, CGroupBlockIODeviceBandwidth *b) { | |
120 | assert(c); | |
8e274523 | 121 | assert(b); |
8e274523 | 122 | |
71fda00f | 123 | LIST_REMOVE(device_bandwidths, c->blockio_device_bandwidths, b); |
4ad49000 LP |
124 | free(b->path); |
125 | free(b); | |
126 | } | |
127 | ||
128 | void cgroup_context_done(CGroupContext *c) { | |
129 | assert(c); | |
130 | ||
13c31542 TH |
131 | while (c->io_device_weights) |
132 | cgroup_context_free_io_device_weight(c, c->io_device_weights); | |
133 | ||
134 | while (c->io_device_limits) | |
135 | cgroup_context_free_io_device_limit(c, c->io_device_limits); | |
136 | ||
4ad49000 LP |
137 | while (c->blockio_device_weights) |
138 | cgroup_context_free_blockio_device_weight(c, c->blockio_device_weights); | |
139 | ||
140 | while (c->blockio_device_bandwidths) | |
141 | cgroup_context_free_blockio_device_bandwidth(c, c->blockio_device_bandwidths); | |
142 | ||
143 | while (c->device_allow) | |
144 | cgroup_context_free_device_allow(c, c->device_allow); | |
6a48d82f DM |
145 | |
146 | c->ip_address_allow = ip_address_access_free_all(c->ip_address_allow); | |
147 | c->ip_address_deny = ip_address_access_free_all(c->ip_address_deny); | |
4ad49000 LP |
148 | } |
149 | ||
150 | void cgroup_context_dump(CGroupContext *c, FILE* f, const char *prefix) { | |
13c31542 TH |
151 | CGroupIODeviceLimit *il; |
152 | CGroupIODeviceWeight *iw; | |
4ad49000 LP |
153 | CGroupBlockIODeviceBandwidth *b; |
154 | CGroupBlockIODeviceWeight *w; | |
155 | CGroupDeviceAllow *a; | |
c21c9906 | 156 | IPAddressAccessItem *iaai; |
9a054909 | 157 | char u[FORMAT_TIMESPAN_MAX]; |
4ad49000 LP |
158 | |
159 | assert(c); | |
160 | assert(f); | |
161 | ||
162 | prefix = strempty(prefix); | |
163 | ||
164 | fprintf(f, | |
165 | "%sCPUAccounting=%s\n" | |
13c31542 | 166 | "%sIOAccounting=%s\n" |
4ad49000 LP |
167 | "%sBlockIOAccounting=%s\n" |
168 | "%sMemoryAccounting=%s\n" | |
d53d9474 | 169 | "%sTasksAccounting=%s\n" |
c21c9906 | 170 | "%sIPAccounting=%s\n" |
66ebf6c0 TH |
171 | "%sCPUWeight=%" PRIu64 "\n" |
172 | "%sStartupCPUWeight=%" PRIu64 "\n" | |
d53d9474 LP |
173 | "%sCPUShares=%" PRIu64 "\n" |
174 | "%sStartupCPUShares=%" PRIu64 "\n" | |
b2f8b02e | 175 | "%sCPUQuotaPerSecSec=%s\n" |
13c31542 TH |
176 | "%sIOWeight=%" PRIu64 "\n" |
177 | "%sStartupIOWeight=%" PRIu64 "\n" | |
d53d9474 LP |
178 | "%sBlockIOWeight=%" PRIu64 "\n" |
179 | "%sStartupBlockIOWeight=%" PRIu64 "\n" | |
da4d897e TH |
180 | "%sMemoryLow=%" PRIu64 "\n" |
181 | "%sMemoryHigh=%" PRIu64 "\n" | |
182 | "%sMemoryMax=%" PRIu64 "\n" | |
96e131ea | 183 | "%sMemorySwapMax=%" PRIu64 "\n" |
4ad49000 | 184 | "%sMemoryLimit=%" PRIu64 "\n" |
03a7b521 | 185 | "%sTasksMax=%" PRIu64 "\n" |
a931ad47 LP |
186 | "%sDevicePolicy=%s\n" |
187 | "%sDelegate=%s\n", | |
4ad49000 | 188 | prefix, yes_no(c->cpu_accounting), |
13c31542 | 189 | prefix, yes_no(c->io_accounting), |
4ad49000 LP |
190 | prefix, yes_no(c->blockio_accounting), |
191 | prefix, yes_no(c->memory_accounting), | |
d53d9474 | 192 | prefix, yes_no(c->tasks_accounting), |
c21c9906 | 193 | prefix, yes_no(c->ip_accounting), |
66ebf6c0 TH |
194 | prefix, c->cpu_weight, |
195 | prefix, c->startup_cpu_weight, | |
4ad49000 | 196 | prefix, c->cpu_shares, |
95ae05c0 | 197 | prefix, c->startup_cpu_shares, |
b1d6dcf5 | 198 | prefix, format_timespan(u, sizeof(u), c->cpu_quota_per_sec_usec, 1), |
13c31542 TH |
199 | prefix, c->io_weight, |
200 | prefix, c->startup_io_weight, | |
4ad49000 | 201 | prefix, c->blockio_weight, |
95ae05c0 | 202 | prefix, c->startup_blockio_weight, |
da4d897e TH |
203 | prefix, c->memory_low, |
204 | prefix, c->memory_high, | |
205 | prefix, c->memory_max, | |
96e131ea | 206 | prefix, c->memory_swap_max, |
4ad49000 | 207 | prefix, c->memory_limit, |
03a7b521 | 208 | prefix, c->tasks_max, |
a931ad47 LP |
209 | prefix, cgroup_device_policy_to_string(c->device_policy), |
210 | prefix, yes_no(c->delegate)); | |
4ad49000 | 211 | |
02638280 LP |
212 | if (c->delegate) { |
213 | _cleanup_free_ char *t = NULL; | |
214 | ||
215 | (void) cg_mask_to_string(c->delegate_controllers, &t); | |
216 | ||
47a78d41 | 217 | fprintf(f, "%sDelegateControllers=%s\n", |
02638280 LP |
218 | prefix, |
219 | strempty(t)); | |
220 | } | |
221 | ||
4ad49000 LP |
222 | LIST_FOREACH(device_allow, a, c->device_allow) |
223 | fprintf(f, | |
224 | "%sDeviceAllow=%s %s%s%s\n", | |
225 | prefix, | |
226 | a->path, | |
227 | a->r ? "r" : "", a->w ? "w" : "", a->m ? "m" : ""); | |
228 | ||
13c31542 TH |
229 | LIST_FOREACH(device_weights, iw, c->io_device_weights) |
230 | fprintf(f, | |
231 | "%sIODeviceWeight=%s %" PRIu64, | |
232 | prefix, | |
233 | iw->path, | |
234 | iw->weight); | |
235 | ||
236 | LIST_FOREACH(device_limits, il, c->io_device_limits) { | |
237 | char buf[FORMAT_BYTES_MAX]; | |
9be57249 TH |
238 | CGroupIOLimitType type; |
239 | ||
240 | for (type = 0; type < _CGROUP_IO_LIMIT_TYPE_MAX; type++) | |
241 | if (il->limits[type] != cgroup_io_limit_defaults[type]) | |
242 | fprintf(f, | |
243 | "%s%s=%s %s\n", | |
244 | prefix, | |
245 | cgroup_io_limit_type_to_string(type), | |
246 | il->path, | |
247 | format_bytes(buf, sizeof(buf), il->limits[type])); | |
13c31542 TH |
248 | } |
249 | ||
4ad49000 LP |
250 | LIST_FOREACH(device_weights, w, c->blockio_device_weights) |
251 | fprintf(f, | |
d53d9474 | 252 | "%sBlockIODeviceWeight=%s %" PRIu64, |
4ad49000 LP |
253 | prefix, |
254 | w->path, | |
255 | w->weight); | |
256 | ||
257 | LIST_FOREACH(device_bandwidths, b, c->blockio_device_bandwidths) { | |
258 | char buf[FORMAT_BYTES_MAX]; | |
259 | ||
979d0311 TH |
260 | if (b->rbps != CGROUP_LIMIT_MAX) |
261 | fprintf(f, | |
262 | "%sBlockIOReadBandwidth=%s %s\n", | |
263 | prefix, | |
264 | b->path, | |
265 | format_bytes(buf, sizeof(buf), b->rbps)); | |
266 | if (b->wbps != CGROUP_LIMIT_MAX) | |
267 | fprintf(f, | |
268 | "%sBlockIOWriteBandwidth=%s %s\n", | |
269 | prefix, | |
270 | b->path, | |
271 | format_bytes(buf, sizeof(buf), b->wbps)); | |
4ad49000 | 272 | } |
c21c9906 LP |
273 | |
274 | LIST_FOREACH(items, iaai, c->ip_address_allow) { | |
275 | _cleanup_free_ char *k = NULL; | |
276 | ||
277 | (void) in_addr_to_string(iaai->family, &iaai->address, &k); | |
278 | fprintf(f, "%sIPAddressAllow=%s/%u\n", prefix, strnull(k), iaai->prefixlen); | |
279 | } | |
280 | ||
281 | LIST_FOREACH(items, iaai, c->ip_address_deny) { | |
282 | _cleanup_free_ char *k = NULL; | |
283 | ||
284 | (void) in_addr_to_string(iaai->family, &iaai->address, &k); | |
285 | fprintf(f, "%sIPAddressDeny=%s/%u\n", prefix, strnull(k), iaai->prefixlen); | |
286 | } | |
4ad49000 LP |
287 | } |
288 | ||
13c31542 | 289 | static int lookup_block_device(const char *p, dev_t *dev) { |
4ad49000 LP |
290 | struct stat st; |
291 | int r; | |
292 | ||
293 | assert(p); | |
294 | assert(dev); | |
295 | ||
296 | r = stat(p, &st); | |
4a62c710 MS |
297 | if (r < 0) |
298 | return log_warning_errno(errno, "Couldn't stat device %s: %m", p); | |
8e274523 | 299 | |
4ad49000 LP |
300 | if (S_ISBLK(st.st_mode)) |
301 | *dev = st.st_rdev; | |
302 | else if (major(st.st_dev) != 0) { | |
303 | /* If this is not a device node then find the block | |
304 | * device this file is stored on */ | |
305 | *dev = st.st_dev; | |
306 | ||
307 | /* If this is a partition, try to get the originating | |
308 | * block device */ | |
309 | block_get_whole_disk(*dev, dev); | |
310 | } else { | |
311 | log_warning("%s is not a block device and file system block device cannot be determined or is not local.", p); | |
312 | return -ENODEV; | |
313 | } | |
8e274523 | 314 | |
8e274523 | 315 | return 0; |
8e274523 LP |
316 | } |
317 | ||
4ad49000 LP |
318 | static int whitelist_device(const char *path, const char *node, const char *acc) { |
319 | char buf[2+DECIMAL_STR_MAX(dev_t)*2+2+4]; | |
320 | struct stat st; | |
b200489b | 321 | bool ignore_notfound; |
8c6db833 | 322 | int r; |
8e274523 | 323 | |
4ad49000 LP |
324 | assert(path); |
325 | assert(acc); | |
8e274523 | 326 | |
b200489b DR |
327 | if (node[0] == '-') { |
328 | /* Non-existent paths starting with "-" must be silently ignored */ | |
329 | node++; | |
330 | ignore_notfound = true; | |
331 | } else | |
332 | ignore_notfound = false; | |
333 | ||
4ad49000 | 334 | if (stat(node, &st) < 0) { |
b200489b | 335 | if (errno == ENOENT && ignore_notfound) |
e7330dfe DP |
336 | return 0; |
337 | ||
338 | return log_warning_errno(errno, "Couldn't stat device %s: %m", node); | |
4ad49000 LP |
339 | } |
340 | ||
341 | if (!S_ISCHR(st.st_mode) && !S_ISBLK(st.st_mode)) { | |
342 | log_warning("%s is not a device.", node); | |
343 | return -ENODEV; | |
344 | } | |
345 | ||
346 | sprintf(buf, | |
347 | "%c %u:%u %s", | |
348 | S_ISCHR(st.st_mode) ? 'c' : 'b', | |
349 | major(st.st_rdev), minor(st.st_rdev), | |
350 | acc); | |
351 | ||
352 | r = cg_set_attribute("devices", path, "devices.allow", buf); | |
1aeab12b | 353 | if (r < 0) |
077ba06e | 354 | log_full_errno(IN_SET(r, -ENOENT, -EROFS, -EINVAL, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, |
714e2e1d | 355 | "Failed to set devices.allow on %s: %m", path); |
4ad49000 LP |
356 | |
357 | return r; | |
8e274523 LP |
358 | } |
359 | ||
90060676 LP |
360 | static int whitelist_major(const char *path, const char *name, char type, const char *acc) { |
361 | _cleanup_fclose_ FILE *f = NULL; | |
362 | char line[LINE_MAX]; | |
363 | bool good = false; | |
364 | int r; | |
365 | ||
366 | assert(path); | |
367 | assert(acc); | |
4c701096 | 368 | assert(IN_SET(type, 'b', 'c')); |
90060676 LP |
369 | |
370 | f = fopen("/proc/devices", "re"); | |
4a62c710 MS |
371 | if (!f) |
372 | return log_warning_errno(errno, "Cannot open /proc/devices to resolve %s (%c): %m", name, type); | |
90060676 LP |
373 | |
374 | FOREACH_LINE(line, f, goto fail) { | |
375 | char buf[2+DECIMAL_STR_MAX(unsigned)+3+4], *p, *w; | |
376 | unsigned maj; | |
377 | ||
378 | truncate_nl(line); | |
379 | ||
380 | if (type == 'c' && streq(line, "Character devices:")) { | |
381 | good = true; | |
382 | continue; | |
383 | } | |
384 | ||
385 | if (type == 'b' && streq(line, "Block devices:")) { | |
386 | good = true; | |
387 | continue; | |
388 | } | |
389 | ||
390 | if (isempty(line)) { | |
391 | good = false; | |
392 | continue; | |
393 | } | |
394 | ||
395 | if (!good) | |
396 | continue; | |
397 | ||
398 | p = strstrip(line); | |
399 | ||
400 | w = strpbrk(p, WHITESPACE); | |
401 | if (!w) | |
402 | continue; | |
403 | *w = 0; | |
404 | ||
405 | r = safe_atou(p, &maj); | |
406 | if (r < 0) | |
407 | continue; | |
408 | if (maj <= 0) | |
409 | continue; | |
410 | ||
411 | w++; | |
412 | w += strspn(w, WHITESPACE); | |
e41969e3 LP |
413 | |
414 | if (fnmatch(name, w, 0) != 0) | |
90060676 LP |
415 | continue; |
416 | ||
417 | sprintf(buf, | |
418 | "%c %u:* %s", | |
419 | type, | |
420 | maj, | |
421 | acc); | |
422 | ||
423 | r = cg_set_attribute("devices", path, "devices.allow", buf); | |
1aeab12b | 424 | if (r < 0) |
077ba06e | 425 | log_full_errno(IN_SET(r, -ENOENT, -EROFS, -EINVAL, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, |
714e2e1d | 426 | "Failed to set devices.allow on %s: %m", path); |
90060676 LP |
427 | } |
428 | ||
429 | return 0; | |
430 | ||
431 | fail: | |
25f027c5 | 432 | return log_warning_errno(errno, "Failed to read /proc/devices: %m"); |
90060676 LP |
433 | } |
434 | ||
66ebf6c0 TH |
435 | static bool cgroup_context_has_cpu_weight(CGroupContext *c) { |
436 | return c->cpu_weight != CGROUP_WEIGHT_INVALID || | |
437 | c->startup_cpu_weight != CGROUP_WEIGHT_INVALID; | |
438 | } | |
439 | ||
440 | static bool cgroup_context_has_cpu_shares(CGroupContext *c) { | |
441 | return c->cpu_shares != CGROUP_CPU_SHARES_INVALID || | |
442 | c->startup_cpu_shares != CGROUP_CPU_SHARES_INVALID; | |
443 | } | |
444 | ||
445 | static uint64_t cgroup_context_cpu_weight(CGroupContext *c, ManagerState state) { | |
446 | if (IN_SET(state, MANAGER_STARTING, MANAGER_INITIALIZING) && | |
447 | c->startup_cpu_weight != CGROUP_WEIGHT_INVALID) | |
448 | return c->startup_cpu_weight; | |
449 | else if (c->cpu_weight != CGROUP_WEIGHT_INVALID) | |
450 | return c->cpu_weight; | |
451 | else | |
452 | return CGROUP_WEIGHT_DEFAULT; | |
453 | } | |
454 | ||
455 | static uint64_t cgroup_context_cpu_shares(CGroupContext *c, ManagerState state) { | |
456 | if (IN_SET(state, MANAGER_STARTING, MANAGER_INITIALIZING) && | |
457 | c->startup_cpu_shares != CGROUP_CPU_SHARES_INVALID) | |
458 | return c->startup_cpu_shares; | |
459 | else if (c->cpu_shares != CGROUP_CPU_SHARES_INVALID) | |
460 | return c->cpu_shares; | |
461 | else | |
462 | return CGROUP_CPU_SHARES_DEFAULT; | |
463 | } | |
464 | ||
465 | static void cgroup_apply_unified_cpu_config(Unit *u, uint64_t weight, uint64_t quota) { | |
466 | char buf[MAX(DECIMAL_STR_MAX(uint64_t) + 1, (DECIMAL_STR_MAX(usec_t) + 1) * 2)]; | |
467 | int r; | |
468 | ||
469 | xsprintf(buf, "%" PRIu64 "\n", weight); | |
470 | r = cg_set_attribute("cpu", u->cgroup_path, "cpu.weight", buf); | |
471 | if (r < 0) | |
472 | log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, | |
473 | "Failed to set cpu.weight: %m"); | |
474 | ||
475 | if (quota != USEC_INFINITY) | |
476 | xsprintf(buf, USEC_FMT " " USEC_FMT "\n", | |
477 | quota * CGROUP_CPU_QUOTA_PERIOD_USEC / USEC_PER_SEC, CGROUP_CPU_QUOTA_PERIOD_USEC); | |
478 | else | |
479 | xsprintf(buf, "max " USEC_FMT "\n", CGROUP_CPU_QUOTA_PERIOD_USEC); | |
480 | ||
481 | r = cg_set_attribute("cpu", u->cgroup_path, "cpu.max", buf); | |
482 | ||
483 | if (r < 0) | |
484 | log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, | |
485 | "Failed to set cpu.max: %m"); | |
486 | } | |
487 | ||
488 | static void cgroup_apply_legacy_cpu_config(Unit *u, uint64_t shares, uint64_t quota) { | |
489 | char buf[MAX(DECIMAL_STR_MAX(uint64_t), DECIMAL_STR_MAX(usec_t)) + 1]; | |
490 | int r; | |
491 | ||
492 | xsprintf(buf, "%" PRIu64 "\n", shares); | |
493 | r = cg_set_attribute("cpu", u->cgroup_path, "cpu.shares", buf); | |
494 | if (r < 0) | |
495 | log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, | |
496 | "Failed to set cpu.shares: %m"); | |
497 | ||
498 | xsprintf(buf, USEC_FMT "\n", CGROUP_CPU_QUOTA_PERIOD_USEC); | |
499 | r = cg_set_attribute("cpu", u->cgroup_path, "cpu.cfs_period_us", buf); | |
500 | if (r < 0) | |
501 | log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, | |
502 | "Failed to set cpu.cfs_period_us: %m"); | |
503 | ||
504 | if (quota != USEC_INFINITY) { | |
505 | xsprintf(buf, USEC_FMT "\n", quota * CGROUP_CPU_QUOTA_PERIOD_USEC / USEC_PER_SEC); | |
506 | r = cg_set_attribute("cpu", u->cgroup_path, "cpu.cfs_quota_us", buf); | |
507 | } else | |
508 | r = cg_set_attribute("cpu", u->cgroup_path, "cpu.cfs_quota_us", "-1"); | |
509 | if (r < 0) | |
510 | log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, | |
511 | "Failed to set cpu.cfs_quota_us: %m"); | |
512 | } | |
513 | ||
514 | static uint64_t cgroup_cpu_shares_to_weight(uint64_t shares) { | |
515 | return CLAMP(shares * CGROUP_WEIGHT_DEFAULT / CGROUP_CPU_SHARES_DEFAULT, | |
516 | CGROUP_WEIGHT_MIN, CGROUP_WEIGHT_MAX); | |
517 | } | |
518 | ||
519 | static uint64_t cgroup_cpu_weight_to_shares(uint64_t weight) { | |
520 | return CLAMP(weight * CGROUP_CPU_SHARES_DEFAULT / CGROUP_WEIGHT_DEFAULT, | |
521 | CGROUP_CPU_SHARES_MIN, CGROUP_CPU_SHARES_MAX); | |
522 | } | |
523 | ||
508c45da | 524 | static bool cgroup_context_has_io_config(CGroupContext *c) { |
538b4852 TH |
525 | return c->io_accounting || |
526 | c->io_weight != CGROUP_WEIGHT_INVALID || | |
527 | c->startup_io_weight != CGROUP_WEIGHT_INVALID || | |
528 | c->io_device_weights || | |
529 | c->io_device_limits; | |
530 | } | |
531 | ||
508c45da | 532 | static bool cgroup_context_has_blockio_config(CGroupContext *c) { |
538b4852 TH |
533 | return c->blockio_accounting || |
534 | c->blockio_weight != CGROUP_BLKIO_WEIGHT_INVALID || | |
535 | c->startup_blockio_weight != CGROUP_BLKIO_WEIGHT_INVALID || | |
536 | c->blockio_device_weights || | |
537 | c->blockio_device_bandwidths; | |
538 | } | |
539 | ||
508c45da | 540 | static uint64_t cgroup_context_io_weight(CGroupContext *c, ManagerState state) { |
64faf04c TH |
541 | if (IN_SET(state, MANAGER_STARTING, MANAGER_INITIALIZING) && |
542 | c->startup_io_weight != CGROUP_WEIGHT_INVALID) | |
543 | return c->startup_io_weight; | |
544 | else if (c->io_weight != CGROUP_WEIGHT_INVALID) | |
545 | return c->io_weight; | |
546 | else | |
547 | return CGROUP_WEIGHT_DEFAULT; | |
548 | } | |
549 | ||
508c45da | 550 | static uint64_t cgroup_context_blkio_weight(CGroupContext *c, ManagerState state) { |
64faf04c TH |
551 | if (IN_SET(state, MANAGER_STARTING, MANAGER_INITIALIZING) && |
552 | c->startup_blockio_weight != CGROUP_BLKIO_WEIGHT_INVALID) | |
553 | return c->startup_blockio_weight; | |
554 | else if (c->blockio_weight != CGROUP_BLKIO_WEIGHT_INVALID) | |
555 | return c->blockio_weight; | |
556 | else | |
557 | return CGROUP_BLKIO_WEIGHT_DEFAULT; | |
558 | } | |
559 | ||
508c45da | 560 | static uint64_t cgroup_weight_blkio_to_io(uint64_t blkio_weight) { |
538b4852 TH |
561 | return CLAMP(blkio_weight * CGROUP_WEIGHT_DEFAULT / CGROUP_BLKIO_WEIGHT_DEFAULT, |
562 | CGROUP_WEIGHT_MIN, CGROUP_WEIGHT_MAX); | |
563 | } | |
564 | ||
508c45da | 565 | static uint64_t cgroup_weight_io_to_blkio(uint64_t io_weight) { |
538b4852 TH |
566 | return CLAMP(io_weight * CGROUP_BLKIO_WEIGHT_DEFAULT / CGROUP_WEIGHT_DEFAULT, |
567 | CGROUP_BLKIO_WEIGHT_MIN, CGROUP_BLKIO_WEIGHT_MAX); | |
568 | } | |
569 | ||
f29ff115 | 570 | static void cgroup_apply_io_device_weight(Unit *u, const char *dev_path, uint64_t io_weight) { |
64faf04c TH |
571 | char buf[DECIMAL_STR_MAX(dev_t)*2+2+DECIMAL_STR_MAX(uint64_t)+1]; |
572 | dev_t dev; | |
573 | int r; | |
574 | ||
575 | r = lookup_block_device(dev_path, &dev); | |
576 | if (r < 0) | |
577 | return; | |
578 | ||
579 | xsprintf(buf, "%u:%u %" PRIu64 "\n", major(dev), minor(dev), io_weight); | |
f29ff115 | 580 | r = cg_set_attribute("io", u->cgroup_path, "io.weight", buf); |
64faf04c | 581 | if (r < 0) |
f29ff115 TH |
582 | log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, |
583 | "Failed to set io.weight: %m"); | |
64faf04c TH |
584 | } |
585 | ||
f29ff115 | 586 | static void cgroup_apply_blkio_device_weight(Unit *u, const char *dev_path, uint64_t blkio_weight) { |
64faf04c TH |
587 | char buf[DECIMAL_STR_MAX(dev_t)*2+2+DECIMAL_STR_MAX(uint64_t)+1]; |
588 | dev_t dev; | |
589 | int r; | |
590 | ||
591 | r = lookup_block_device(dev_path, &dev); | |
592 | if (r < 0) | |
593 | return; | |
594 | ||
595 | xsprintf(buf, "%u:%u %" PRIu64 "\n", major(dev), minor(dev), blkio_weight); | |
f29ff115 | 596 | r = cg_set_attribute("blkio", u->cgroup_path, "blkio.weight_device", buf); |
64faf04c | 597 | if (r < 0) |
f29ff115 TH |
598 | log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, |
599 | "Failed to set blkio.weight_device: %m"); | |
64faf04c TH |
600 | } |
601 | ||
f29ff115 | 602 | static unsigned cgroup_apply_io_device_limit(Unit *u, const char *dev_path, uint64_t *limits) { |
64faf04c TH |
603 | char limit_bufs[_CGROUP_IO_LIMIT_TYPE_MAX][DECIMAL_STR_MAX(uint64_t)]; |
604 | char buf[DECIMAL_STR_MAX(dev_t)*2+2+(6+DECIMAL_STR_MAX(uint64_t)+1)*4]; | |
605 | CGroupIOLimitType type; | |
606 | dev_t dev; | |
607 | unsigned n = 0; | |
608 | int r; | |
609 | ||
610 | r = lookup_block_device(dev_path, &dev); | |
611 | if (r < 0) | |
612 | return 0; | |
613 | ||
614 | for (type = 0; type < _CGROUP_IO_LIMIT_TYPE_MAX; type++) { | |
615 | if (limits[type] != cgroup_io_limit_defaults[type]) { | |
616 | xsprintf(limit_bufs[type], "%" PRIu64, limits[type]); | |
617 | n++; | |
618 | } else { | |
619 | xsprintf(limit_bufs[type], "%s", limits[type] == CGROUP_LIMIT_MAX ? "max" : "0"); | |
620 | } | |
621 | } | |
622 | ||
623 | xsprintf(buf, "%u:%u rbps=%s wbps=%s riops=%s wiops=%s\n", major(dev), minor(dev), | |
624 | limit_bufs[CGROUP_IO_RBPS_MAX], limit_bufs[CGROUP_IO_WBPS_MAX], | |
625 | limit_bufs[CGROUP_IO_RIOPS_MAX], limit_bufs[CGROUP_IO_WIOPS_MAX]); | |
f29ff115 | 626 | r = cg_set_attribute("io", u->cgroup_path, "io.max", buf); |
64faf04c | 627 | if (r < 0) |
f29ff115 TH |
628 | log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, |
629 | "Failed to set io.max: %m"); | |
64faf04c TH |
630 | return n; |
631 | } | |
632 | ||
f29ff115 | 633 | static unsigned cgroup_apply_blkio_device_limit(Unit *u, const char *dev_path, uint64_t rbps, uint64_t wbps) { |
64faf04c TH |
634 | char buf[DECIMAL_STR_MAX(dev_t)*2+2+DECIMAL_STR_MAX(uint64_t)+1]; |
635 | dev_t dev; | |
636 | unsigned n = 0; | |
637 | int r; | |
638 | ||
639 | r = lookup_block_device(dev_path, &dev); | |
640 | if (r < 0) | |
641 | return 0; | |
642 | ||
643 | if (rbps != CGROUP_LIMIT_MAX) | |
644 | n++; | |
645 | sprintf(buf, "%u:%u %" PRIu64 "\n", major(dev), minor(dev), rbps); | |
f29ff115 | 646 | r = cg_set_attribute("blkio", u->cgroup_path, "blkio.throttle.read_bps_device", buf); |
64faf04c | 647 | if (r < 0) |
f29ff115 TH |
648 | log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, |
649 | "Failed to set blkio.throttle.read_bps_device: %m"); | |
64faf04c TH |
650 | |
651 | if (wbps != CGROUP_LIMIT_MAX) | |
652 | n++; | |
653 | sprintf(buf, "%u:%u %" PRIu64 "\n", major(dev), minor(dev), wbps); | |
f29ff115 | 654 | r = cg_set_attribute("blkio", u->cgroup_path, "blkio.throttle.write_bps_device", buf); |
64faf04c | 655 | if (r < 0) |
f29ff115 TH |
656 | log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, |
657 | "Failed to set blkio.throttle.write_bps_device: %m"); | |
64faf04c TH |
658 | |
659 | return n; | |
660 | } | |
661 | ||
da4d897e | 662 | static bool cgroup_context_has_unified_memory_config(CGroupContext *c) { |
96e131ea | 663 | 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 |
664 | } |
665 | ||
f29ff115 | 666 | static void cgroup_apply_unified_memory_limit(Unit *u, const char *file, uint64_t v) { |
da4d897e TH |
667 | char buf[DECIMAL_STR_MAX(uint64_t) + 1] = "max"; |
668 | int r; | |
669 | ||
670 | if (v != CGROUP_LIMIT_MAX) | |
671 | xsprintf(buf, "%" PRIu64 "\n", v); | |
672 | ||
f29ff115 | 673 | r = cg_set_attribute("memory", u->cgroup_path, file, buf); |
da4d897e | 674 | if (r < 0) |
f29ff115 TH |
675 | log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, |
676 | "Failed to set %s: %m", file); | |
da4d897e TH |
677 | } |
678 | ||
906c06f6 DM |
679 | static void cgroup_apply_firewall(Unit *u, CGroupContext *c) { |
680 | int r; | |
681 | ||
682 | if (u->type == UNIT_SLICE) /* Skip this for slice units, they are inner cgroup nodes, and since bpf/cgroup is | |
683 | * not recursive we don't ever touch the bpf on them */ | |
684 | return; | |
685 | ||
686 | r = bpf_firewall_compile(u); | |
687 | if (r < 0) | |
688 | return; | |
689 | ||
690 | (void) bpf_firewall_install(u); | |
691 | return; | |
692 | } | |
693 | ||
694 | static void cgroup_context_apply( | |
695 | Unit *u, | |
696 | CGroupMask apply_mask, | |
697 | bool apply_bpf, | |
698 | ManagerState state) { | |
699 | ||
f29ff115 TH |
700 | const char *path; |
701 | CGroupContext *c; | |
01efdf13 | 702 | bool is_root; |
4ad49000 LP |
703 | int r; |
704 | ||
f29ff115 TH |
705 | assert(u); |
706 | ||
707 | c = unit_get_cgroup_context(u); | |
708 | path = u->cgroup_path; | |
709 | ||
4ad49000 LP |
710 | assert(c); |
711 | assert(path); | |
8e274523 | 712 | |
906c06f6 DM |
713 | /* Nothing to do? Exit early! */ |
714 | if (apply_mask == 0 && !apply_bpf) | |
4ad49000 | 715 | return; |
8e274523 | 716 | |
71c26873 | 717 | /* Some cgroup attributes are not supported on the root cgroup, |
01efdf13 LP |
718 | * hence silently ignore */ |
719 | is_root = isempty(path) || path_equal(path, "/"); | |
6da13913 ZJS |
720 | if (is_root) |
721 | /* Make sure we don't try to display messages with an empty path. */ | |
722 | path = "/"; | |
01efdf13 | 723 | |
714e2e1d LP |
724 | /* We generally ignore errors caused by read-only mounted |
725 | * cgroup trees (assuming we are running in a container then), | |
726 | * and missing cgroups, i.e. EROFS and ENOENT. */ | |
727 | ||
906c06f6 DM |
728 | if ((apply_mask & CGROUP_MASK_CPU) && !is_root) { |
729 | bool has_weight, has_shares; | |
730 | ||
731 | has_weight = cgroup_context_has_cpu_weight(c); | |
732 | has_shares = cgroup_context_has_cpu_shares(c); | |
8e274523 | 733 | |
b4cccbc1 | 734 | if (cg_all_unified() > 0) { |
66ebf6c0 | 735 | uint64_t weight; |
b2f8b02e | 736 | |
66ebf6c0 TH |
737 | if (has_weight) |
738 | weight = cgroup_context_cpu_weight(c, state); | |
739 | else if (has_shares) { | |
740 | uint64_t shares = cgroup_context_cpu_shares(c, state); | |
b2f8b02e | 741 | |
66ebf6c0 TH |
742 | weight = cgroup_cpu_shares_to_weight(shares); |
743 | ||
744 | log_cgroup_compat(u, "Applying [Startup]CpuShares %" PRIu64 " as [Startup]CpuWeight %" PRIu64 " on %s", | |
745 | shares, weight, path); | |
746 | } else | |
747 | weight = CGROUP_WEIGHT_DEFAULT; | |
748 | ||
749 | cgroup_apply_unified_cpu_config(u, weight, c->cpu_quota_per_sec_usec); | |
750 | } else { | |
751 | uint64_t shares; | |
752 | ||
7d862ab8 | 753 | if (has_weight) { |
66ebf6c0 TH |
754 | uint64_t weight = cgroup_context_cpu_weight(c, state); |
755 | ||
756 | shares = cgroup_cpu_weight_to_shares(weight); | |
757 | ||
758 | log_cgroup_compat(u, "Applying [Startup]CpuWeight %" PRIu64 " as [Startup]CpuShares %" PRIu64 " on %s", | |
759 | weight, shares, path); | |
7d862ab8 TH |
760 | } else if (has_shares) |
761 | shares = cgroup_context_cpu_shares(c, state); | |
762 | else | |
66ebf6c0 TH |
763 | shares = CGROUP_CPU_SHARES_DEFAULT; |
764 | ||
765 | cgroup_apply_legacy_cpu_config(u, shares, c->cpu_quota_per_sec_usec); | |
766 | } | |
4ad49000 LP |
767 | } |
768 | ||
906c06f6 | 769 | if (apply_mask & CGROUP_MASK_IO) { |
538b4852 TH |
770 | bool has_io = cgroup_context_has_io_config(c); |
771 | bool has_blockio = cgroup_context_has_blockio_config(c); | |
13c31542 TH |
772 | |
773 | if (!is_root) { | |
64faf04c TH |
774 | char buf[8+DECIMAL_STR_MAX(uint64_t)+1]; |
775 | uint64_t weight; | |
13c31542 | 776 | |
538b4852 TH |
777 | if (has_io) |
778 | weight = cgroup_context_io_weight(c, state); | |
128fadc9 TH |
779 | else if (has_blockio) { |
780 | uint64_t blkio_weight = cgroup_context_blkio_weight(c, state); | |
781 | ||
782 | weight = cgroup_weight_blkio_to_io(blkio_weight); | |
783 | ||
784 | log_cgroup_compat(u, "Applying [Startup]BlockIOWeight %" PRIu64 " as [Startup]IOWeight %" PRIu64, | |
785 | blkio_weight, weight); | |
786 | } else | |
538b4852 | 787 | weight = CGROUP_WEIGHT_DEFAULT; |
13c31542 TH |
788 | |
789 | xsprintf(buf, "default %" PRIu64 "\n", weight); | |
790 | r = cg_set_attribute("io", path, "io.weight", buf); | |
791 | if (r < 0) | |
f29ff115 TH |
792 | log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, |
793 | "Failed to set io.weight: %m"); | |
13c31542 | 794 | |
538b4852 TH |
795 | if (has_io) { |
796 | CGroupIODeviceWeight *w; | |
797 | ||
798 | /* FIXME: no way to reset this list */ | |
799 | LIST_FOREACH(device_weights, w, c->io_device_weights) | |
f29ff115 | 800 | cgroup_apply_io_device_weight(u, w->path, w->weight); |
538b4852 TH |
801 | } else if (has_blockio) { |
802 | CGroupBlockIODeviceWeight *w; | |
803 | ||
804 | /* FIXME: no way to reset this list */ | |
128fadc9 TH |
805 | LIST_FOREACH(device_weights, w, c->blockio_device_weights) { |
806 | weight = cgroup_weight_blkio_to_io(w->weight); | |
807 | ||
808 | log_cgroup_compat(u, "Applying BlockIODeviceWeight %" PRIu64 " as IODeviceWeight %" PRIu64 " for %s", | |
809 | w->weight, weight, w->path); | |
810 | ||
811 | cgroup_apply_io_device_weight(u, w->path, weight); | |
812 | } | |
538b4852 | 813 | } |
13c31542 TH |
814 | } |
815 | ||
64faf04c | 816 | /* Apply limits and free ones without config. */ |
538b4852 TH |
817 | if (has_io) { |
818 | CGroupIODeviceLimit *l, *next; | |
819 | ||
820 | LIST_FOREACH_SAFE(device_limits, l, next, c->io_device_limits) { | |
f29ff115 | 821 | if (!cgroup_apply_io_device_limit(u, l->path, l->limits)) |
538b4852 TH |
822 | cgroup_context_free_io_device_limit(c, l); |
823 | } | |
824 | } else if (has_blockio) { | |
825 | CGroupBlockIODeviceBandwidth *b, *next; | |
826 | ||
827 | LIST_FOREACH_SAFE(device_bandwidths, b, next, c->blockio_device_bandwidths) { | |
828 | uint64_t limits[_CGROUP_IO_LIMIT_TYPE_MAX]; | |
829 | CGroupIOLimitType type; | |
830 | ||
831 | for (type = 0; type < _CGROUP_IO_LIMIT_TYPE_MAX; type++) | |
832 | limits[type] = cgroup_io_limit_defaults[type]; | |
833 | ||
834 | limits[CGROUP_IO_RBPS_MAX] = b->rbps; | |
835 | limits[CGROUP_IO_WBPS_MAX] = b->wbps; | |
836 | ||
128fadc9 TH |
837 | log_cgroup_compat(u, "Applying BlockIO{Read|Write}Bandwidth %" PRIu64 " %" PRIu64 " as IO{Read|Write}BandwidthMax for %s", |
838 | b->rbps, b->wbps, b->path); | |
839 | ||
f29ff115 | 840 | if (!cgroup_apply_io_device_limit(u, b->path, limits)) |
538b4852 TH |
841 | cgroup_context_free_blockio_device_bandwidth(c, b); |
842 | } | |
13c31542 TH |
843 | } |
844 | } | |
845 | ||
906c06f6 | 846 | if (apply_mask & CGROUP_MASK_BLKIO) { |
538b4852 TH |
847 | bool has_io = cgroup_context_has_io_config(c); |
848 | bool has_blockio = cgroup_context_has_blockio_config(c); | |
4ad49000 | 849 | |
01efdf13 | 850 | if (!is_root) { |
64faf04c TH |
851 | char buf[DECIMAL_STR_MAX(uint64_t)+1]; |
852 | uint64_t weight; | |
64faf04c | 853 | |
7d862ab8 | 854 | if (has_io) { |
128fadc9 TH |
855 | uint64_t io_weight = cgroup_context_io_weight(c, state); |
856 | ||
538b4852 | 857 | weight = cgroup_weight_io_to_blkio(cgroup_context_io_weight(c, state)); |
128fadc9 TH |
858 | |
859 | log_cgroup_compat(u, "Applying [Startup]IOWeight %" PRIu64 " as [Startup]BlockIOWeight %" PRIu64, | |
860 | io_weight, weight); | |
7d862ab8 TH |
861 | } else if (has_blockio) |
862 | weight = cgroup_context_blkio_weight(c, state); | |
863 | else | |
538b4852 | 864 | weight = CGROUP_BLKIO_WEIGHT_DEFAULT; |
64faf04c TH |
865 | |
866 | xsprintf(buf, "%" PRIu64 "\n", weight); | |
01efdf13 | 867 | r = cg_set_attribute("blkio", path, "blkio.weight", buf); |
1aeab12b | 868 | if (r < 0) |
f29ff115 TH |
869 | log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, |
870 | "Failed to set blkio.weight: %m"); | |
4ad49000 | 871 | |
7d862ab8 | 872 | if (has_io) { |
538b4852 TH |
873 | CGroupIODeviceWeight *w; |
874 | ||
875 | /* FIXME: no way to reset this list */ | |
128fadc9 TH |
876 | LIST_FOREACH(device_weights, w, c->io_device_weights) { |
877 | weight = cgroup_weight_io_to_blkio(w->weight); | |
878 | ||
879 | log_cgroup_compat(u, "Applying IODeviceWeight %" PRIu64 " as BlockIODeviceWeight %" PRIu64 " for %s", | |
880 | w->weight, weight, w->path); | |
881 | ||
882 | cgroup_apply_blkio_device_weight(u, w->path, weight); | |
883 | } | |
7d862ab8 TH |
884 | } else if (has_blockio) { |
885 | CGroupBlockIODeviceWeight *w; | |
886 | ||
887 | /* FIXME: no way to reset this list */ | |
888 | LIST_FOREACH(device_weights, w, c->blockio_device_weights) | |
889 | cgroup_apply_blkio_device_weight(u, w->path, w->weight); | |
538b4852 | 890 | } |
4ad49000 LP |
891 | } |
892 | ||
64faf04c | 893 | /* Apply limits and free ones without config. */ |
7d862ab8 | 894 | if (has_io) { |
538b4852 TH |
895 | CGroupIODeviceLimit *l, *next; |
896 | ||
897 | LIST_FOREACH_SAFE(device_limits, l, next, c->io_device_limits) { | |
128fadc9 TH |
898 | log_cgroup_compat(u, "Applying IO{Read|Write}Bandwidth %" PRIu64 " %" PRIu64 " as BlockIO{Read|Write}BandwidthMax for %s", |
899 | l->limits[CGROUP_IO_RBPS_MAX], l->limits[CGROUP_IO_WBPS_MAX], l->path); | |
900 | ||
f29ff115 | 901 | if (!cgroup_apply_blkio_device_limit(u, l->path, l->limits[CGROUP_IO_RBPS_MAX], l->limits[CGROUP_IO_WBPS_MAX])) |
538b4852 TH |
902 | cgroup_context_free_io_device_limit(c, l); |
903 | } | |
7d862ab8 TH |
904 | } else if (has_blockio) { |
905 | CGroupBlockIODeviceBandwidth *b, *next; | |
906 | ||
907 | LIST_FOREACH_SAFE(device_bandwidths, b, next, c->blockio_device_bandwidths) | |
908 | if (!cgroup_apply_blkio_device_limit(u, b->path, b->rbps, b->wbps)) | |
909 | cgroup_context_free_blockio_device_bandwidth(c, b); | |
d686d8a9 | 910 | } |
8e274523 LP |
911 | } |
912 | ||
906c06f6 | 913 | if ((apply_mask & CGROUP_MASK_MEMORY) && !is_root) { |
b4cccbc1 LP |
914 | if (cg_all_unified() > 0) { |
915 | uint64_t max, swap_max = CGROUP_LIMIT_MAX; | |
efdb0237 | 916 | |
96e131ea | 917 | if (cgroup_context_has_unified_memory_config(c)) { |
da4d897e | 918 | max = c->memory_max; |
96e131ea WC |
919 | swap_max = c->memory_swap_max; |
920 | } else { | |
da4d897e | 921 | max = c->memory_limit; |
efdb0237 | 922 | |
128fadc9 TH |
923 | if (max != CGROUP_LIMIT_MAX) |
924 | log_cgroup_compat(u, "Applying MemoryLimit %" PRIu64 " as MemoryMax", max); | |
925 | } | |
926 | ||
f29ff115 TH |
927 | cgroup_apply_unified_memory_limit(u, "memory.low", c->memory_low); |
928 | cgroup_apply_unified_memory_limit(u, "memory.high", c->memory_high); | |
929 | cgroup_apply_unified_memory_limit(u, "memory.max", max); | |
96e131ea | 930 | cgroup_apply_unified_memory_limit(u, "memory.swap.max", swap_max); |
efdb0237 | 931 | } else { |
da4d897e | 932 | char buf[DECIMAL_STR_MAX(uint64_t) + 1]; |
7d862ab8 | 933 | uint64_t val; |
da4d897e | 934 | |
7d862ab8 | 935 | if (cgroup_context_has_unified_memory_config(c)) { |
78a4ee59 | 936 | val = c->memory_max; |
7d862ab8 TH |
937 | log_cgroup_compat(u, "Applying MemoryMax %" PRIi64 " as MemoryLimit", val); |
938 | } else | |
939 | val = c->memory_limit; | |
128fadc9 | 940 | |
78a4ee59 DM |
941 | if (val == CGROUP_LIMIT_MAX) |
942 | strncpy(buf, "-1\n", sizeof(buf)); | |
943 | else | |
944 | xsprintf(buf, "%" PRIu64 "\n", val); | |
945 | ||
da4d897e TH |
946 | r = cg_set_attribute("memory", path, "memory.limit_in_bytes", buf); |
947 | if (r < 0) | |
f29ff115 TH |
948 | log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, |
949 | "Failed to set memory.limit_in_bytes: %m"); | |
da4d897e | 950 | } |
4ad49000 | 951 | } |
8e274523 | 952 | |
906c06f6 | 953 | if ((apply_mask & CGROUP_MASK_DEVICES) && !is_root) { |
4ad49000 | 954 | CGroupDeviceAllow *a; |
8e274523 | 955 | |
714e2e1d LP |
956 | /* Changing the devices list of a populated cgroup |
957 | * might result in EINVAL, hence ignore EINVAL | |
958 | * here. */ | |
959 | ||
4ad49000 LP |
960 | if (c->device_allow || c->device_policy != CGROUP_AUTO) |
961 | r = cg_set_attribute("devices", path, "devices.deny", "a"); | |
962 | else | |
963 | r = cg_set_attribute("devices", path, "devices.allow", "a"); | |
1aeab12b | 964 | if (r < 0) |
f29ff115 TH |
965 | log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EINVAL, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, |
966 | "Failed to reset devices.list: %m"); | |
fb385181 | 967 | |
4ad49000 LP |
968 | if (c->device_policy == CGROUP_CLOSED || |
969 | (c->device_policy == CGROUP_AUTO && c->device_allow)) { | |
970 | static const char auto_devices[] = | |
7d711efb LP |
971 | "/dev/null\0" "rwm\0" |
972 | "/dev/zero\0" "rwm\0" | |
973 | "/dev/full\0" "rwm\0" | |
974 | "/dev/random\0" "rwm\0" | |
975 | "/dev/urandom\0" "rwm\0" | |
976 | "/dev/tty\0" "rwm\0" | |
0d9e7991 AP |
977 | "/dev/pts/ptmx\0" "rw\0" /* /dev/pts/ptmx may not be duplicated, but accessed */ |
978 | /* Allow /run/systemd/inaccessible/{chr,blk} devices for mapping InaccessiblePaths */ | |
e7330dfe DP |
979 | "-/run/systemd/inaccessible/chr\0" "rwm\0" |
980 | "-/run/systemd/inaccessible/blk\0" "rwm\0"; | |
4ad49000 LP |
981 | |
982 | const char *x, *y; | |
983 | ||
984 | NULSTR_FOREACH_PAIR(x, y, auto_devices) | |
985 | whitelist_device(path, x, y); | |
7d711efb LP |
986 | |
987 | whitelist_major(path, "pts", 'c', "rw"); | |
4ad49000 LP |
988 | } |
989 | ||
990 | LIST_FOREACH(device_allow, a, c->device_allow) { | |
fb4650aa | 991 | char acc[4], *val; |
4ad49000 LP |
992 | unsigned k = 0; |
993 | ||
994 | if (a->r) | |
995 | acc[k++] = 'r'; | |
996 | if (a->w) | |
997 | acc[k++] = 'w'; | |
998 | if (a->m) | |
999 | acc[k++] = 'm'; | |
fb385181 | 1000 | |
4ad49000 LP |
1001 | if (k == 0) |
1002 | continue; | |
fb385181 | 1003 | |
4ad49000 | 1004 | acc[k++] = 0; |
90060676 | 1005 | |
27458ed6 | 1006 | if (path_startswith(a->path, "/dev/")) |
90060676 | 1007 | whitelist_device(path, a->path, acc); |
fb4650aa ZJS |
1008 | else if ((val = startswith(a->path, "block-"))) |
1009 | whitelist_major(path, val, 'b', acc); | |
1010 | else if ((val = startswith(a->path, "char-"))) | |
1011 | whitelist_major(path, val, 'c', acc); | |
90060676 | 1012 | else |
f29ff115 | 1013 | log_unit_debug(u, "Ignoring device %s while writing cgroup attribute.", a->path); |
4ad49000 LP |
1014 | } |
1015 | } | |
03a7b521 | 1016 | |
906c06f6 | 1017 | if ((apply_mask & CGROUP_MASK_PIDS) && !is_root) { |
03a7b521 | 1018 | |
f5058264 | 1019 | if (c->tasks_max != CGROUP_LIMIT_MAX) { |
03a7b521 LP |
1020 | char buf[DECIMAL_STR_MAX(uint64_t) + 2]; |
1021 | ||
1022 | sprintf(buf, "%" PRIu64 "\n", c->tasks_max); | |
1023 | r = cg_set_attribute("pids", path, "pids.max", buf); | |
1024 | } else | |
1025 | r = cg_set_attribute("pids", path, "pids.max", "max"); | |
1026 | ||
1027 | if (r < 0) | |
f29ff115 TH |
1028 | log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, |
1029 | "Failed to set pids.max: %m"); | |
03a7b521 | 1030 | } |
906c06f6 DM |
1031 | |
1032 | if (apply_bpf) | |
1033 | cgroup_apply_firewall(u, c); | |
fb385181 LP |
1034 | } |
1035 | ||
efdb0237 LP |
1036 | CGroupMask cgroup_context_get_mask(CGroupContext *c) { |
1037 | CGroupMask mask = 0; | |
8e274523 | 1038 | |
4ad49000 | 1039 | /* Figure out which controllers we need */ |
8e274523 | 1040 | |
b2f8b02e | 1041 | if (c->cpu_accounting || |
66ebf6c0 TH |
1042 | cgroup_context_has_cpu_weight(c) || |
1043 | cgroup_context_has_cpu_shares(c) || | |
3a43da28 | 1044 | c->cpu_quota_per_sec_usec != USEC_INFINITY) |
efdb0237 | 1045 | mask |= CGROUP_MASK_CPUACCT | CGROUP_MASK_CPU; |
ecedd90f | 1046 | |
538b4852 TH |
1047 | if (cgroup_context_has_io_config(c) || cgroup_context_has_blockio_config(c)) |
1048 | mask |= CGROUP_MASK_IO | CGROUP_MASK_BLKIO; | |
ecedd90f | 1049 | |
4ad49000 | 1050 | if (c->memory_accounting || |
da4d897e TH |
1051 | c->memory_limit != CGROUP_LIMIT_MAX || |
1052 | cgroup_context_has_unified_memory_config(c)) | |
efdb0237 | 1053 | mask |= CGROUP_MASK_MEMORY; |
8e274523 | 1054 | |
a931ad47 LP |
1055 | if (c->device_allow || |
1056 | c->device_policy != CGROUP_AUTO) | |
3905f127 | 1057 | mask |= CGROUP_MASK_DEVICES; |
4ad49000 | 1058 | |
03a7b521 LP |
1059 | if (c->tasks_accounting || |
1060 | c->tasks_max != (uint64_t) -1) | |
1061 | mask |= CGROUP_MASK_PIDS; | |
1062 | ||
4ad49000 | 1063 | return mask; |
8e274523 LP |
1064 | } |
1065 | ||
efdb0237 | 1066 | CGroupMask unit_get_own_mask(Unit *u) { |
4ad49000 | 1067 | CGroupContext *c; |
8e274523 | 1068 | |
efdb0237 LP |
1069 | /* Returns the mask of controllers the unit needs for itself */ |
1070 | ||
4ad49000 LP |
1071 | c = unit_get_cgroup_context(u); |
1072 | if (!c) | |
1073 | return 0; | |
8e274523 | 1074 | |
02638280 LP |
1075 | return cgroup_context_get_mask(c); |
1076 | } | |
1077 | ||
1078 | CGroupMask unit_get_delegate_mask(Unit *u) { | |
1079 | CGroupContext *c; | |
1080 | ||
1081 | /* If delegation is turned on, then turn on selected controllers, unless we are on the legacy hierarchy and the | |
1082 | * process we fork into is known to drop privileges, and hence shouldn't get access to the controllers. | |
19af675e | 1083 | * |
02638280 | 1084 | * Note that on the unified hierarchy it is safe to delegate controllers to unprivileged services. */ |
a931ad47 | 1085 | |
02638280 LP |
1086 | if (u->type == UNIT_SLICE) |
1087 | return 0; | |
1088 | ||
1089 | c = unit_get_cgroup_context(u); | |
1090 | if (!c) | |
1091 | return 0; | |
1092 | ||
1093 | if (!c->delegate) | |
1094 | return 0; | |
1095 | ||
1096 | if (cg_all_unified() <= 0) { | |
a931ad47 LP |
1097 | ExecContext *e; |
1098 | ||
1099 | e = unit_get_exec_context(u); | |
02638280 LP |
1100 | if (e && !exec_context_maintains_privileges(e)) |
1101 | return 0; | |
a931ad47 LP |
1102 | } |
1103 | ||
02638280 | 1104 | return c->delegate_controllers; |
8e274523 LP |
1105 | } |
1106 | ||
efdb0237 | 1107 | CGroupMask unit_get_members_mask(Unit *u) { |
4ad49000 | 1108 | assert(u); |
bc432dc7 | 1109 | |
02638280 | 1110 | /* Returns the mask of controllers all of the unit's children require, merged */ |
efdb0237 | 1111 | |
bc432dc7 LP |
1112 | if (u->cgroup_members_mask_valid) |
1113 | return u->cgroup_members_mask; | |
1114 | ||
02638280 | 1115 | u->cgroup_members_mask = unit_get_delegate_mask(u); |
bc432dc7 LP |
1116 | |
1117 | if (u->type == UNIT_SLICE) { | |
eef85c4a | 1118 | void *v; |
bc432dc7 LP |
1119 | Unit *member; |
1120 | Iterator i; | |
1121 | ||
eef85c4a | 1122 | HASHMAP_FOREACH_KEY(v, member, u->dependencies[UNIT_BEFORE], i) { |
bc432dc7 LP |
1123 | |
1124 | if (member == u) | |
1125 | continue; | |
1126 | ||
d4fdc205 | 1127 | if (UNIT_DEREF(member->slice) != u) |
bc432dc7 LP |
1128 | continue; |
1129 | ||
31604970 | 1130 | u->cgroup_members_mask |= unit_get_subtree_mask(member); /* note that this calls ourselves again, for the children */ |
bc432dc7 LP |
1131 | } |
1132 | } | |
1133 | ||
1134 | u->cgroup_members_mask_valid = true; | |
6414b7c9 | 1135 | return u->cgroup_members_mask; |
246aa6dd LP |
1136 | } |
1137 | ||
efdb0237 | 1138 | CGroupMask unit_get_siblings_mask(Unit *u) { |
4ad49000 | 1139 | assert(u); |
246aa6dd | 1140 | |
efdb0237 LP |
1141 | /* Returns the mask of controllers all of the unit's siblings |
1142 | * require, i.e. the members mask of the unit's parent slice | |
1143 | * if there is one. */ | |
1144 | ||
bc432dc7 | 1145 | if (UNIT_ISSET(u->slice)) |
637f421e | 1146 | return unit_get_members_mask(UNIT_DEREF(u->slice)); |
4ad49000 | 1147 | |
31604970 | 1148 | return unit_get_subtree_mask(u); |
246aa6dd LP |
1149 | } |
1150 | ||
efdb0237 LP |
1151 | CGroupMask unit_get_subtree_mask(Unit *u) { |
1152 | ||
1153 | /* Returns the mask of this subtree, meaning of the group | |
1154 | * itself and its children. */ | |
1155 | ||
1156 | return unit_get_own_mask(u) | unit_get_members_mask(u); | |
1157 | } | |
1158 | ||
1159 | CGroupMask unit_get_target_mask(Unit *u) { | |
1160 | CGroupMask mask; | |
1161 | ||
1162 | /* This returns the cgroup mask of all controllers to enable | |
1163 | * for a specific cgroup, i.e. everything it needs itself, | |
1164 | * plus all that its children need, plus all that its siblings | |
1165 | * need. This is primarily useful on the legacy cgroup | |
1166 | * hierarchy, where we need to duplicate each cgroup in each | |
1167 | * hierarchy that shall be enabled for it. */ | |
6414b7c9 | 1168 | |
efdb0237 LP |
1169 | mask = unit_get_own_mask(u) | unit_get_members_mask(u) | unit_get_siblings_mask(u); |
1170 | mask &= u->manager->cgroup_supported; | |
1171 | ||
1172 | return mask; | |
1173 | } | |
1174 | ||
1175 | CGroupMask unit_get_enable_mask(Unit *u) { | |
1176 | CGroupMask mask; | |
1177 | ||
1178 | /* This returns the cgroup mask of all controllers to enable | |
1179 | * for the children of a specific cgroup. This is primarily | |
1180 | * useful for the unified cgroup hierarchy, where each cgroup | |
1181 | * controls which controllers are enabled for its children. */ | |
1182 | ||
1183 | mask = unit_get_members_mask(u); | |
6414b7c9 DS |
1184 | mask &= u->manager->cgroup_supported; |
1185 | ||
1186 | return mask; | |
1187 | } | |
1188 | ||
906c06f6 DM |
1189 | bool unit_get_needs_bpf(Unit *u) { |
1190 | CGroupContext *c; | |
1191 | Unit *p; | |
1192 | assert(u); | |
1193 | ||
1194 | /* We never attach BPF to slice units, as they are inner cgroup nodes and cgroup/BPF is not recursive at the | |
1195 | * moment. */ | |
1196 | if (u->type == UNIT_SLICE) | |
1197 | return false; | |
1198 | ||
1199 | c = unit_get_cgroup_context(u); | |
1200 | if (!c) | |
1201 | return false; | |
1202 | ||
1203 | if (c->ip_accounting || | |
1204 | c->ip_address_allow || | |
1205 | c->ip_address_deny) | |
1206 | return true; | |
1207 | ||
1208 | /* If any parent slice has an IP access list defined, it applies too */ | |
1209 | for (p = UNIT_DEREF(u->slice); p; p = UNIT_DEREF(p->slice)) { | |
1210 | c = unit_get_cgroup_context(p); | |
1211 | if (!c) | |
1212 | return false; | |
1213 | ||
1214 | if (c->ip_address_allow || | |
1215 | c->ip_address_deny) | |
1216 | return true; | |
1217 | } | |
1218 | ||
1219 | return false; | |
1220 | } | |
1221 | ||
6414b7c9 DS |
1222 | /* Recurse from a unit up through its containing slices, propagating |
1223 | * mask bits upward. A unit is also member of itself. */ | |
bc432dc7 | 1224 | void unit_update_cgroup_members_masks(Unit *u) { |
efdb0237 | 1225 | CGroupMask m; |
bc432dc7 LP |
1226 | bool more; |
1227 | ||
1228 | assert(u); | |
1229 | ||
1230 | /* Calculate subtree mask */ | |
efdb0237 | 1231 | m = unit_get_subtree_mask(u); |
bc432dc7 LP |
1232 | |
1233 | /* See if anything changed from the previous invocation. If | |
1234 | * not, we're done. */ | |
1235 | if (u->cgroup_subtree_mask_valid && m == u->cgroup_subtree_mask) | |
1236 | return; | |
1237 | ||
1238 | more = | |
1239 | u->cgroup_subtree_mask_valid && | |
1240 | ((m & ~u->cgroup_subtree_mask) != 0) && | |
1241 | ((~m & u->cgroup_subtree_mask) == 0); | |
1242 | ||
1243 | u->cgroup_subtree_mask = m; | |
1244 | u->cgroup_subtree_mask_valid = true; | |
1245 | ||
6414b7c9 DS |
1246 | if (UNIT_ISSET(u->slice)) { |
1247 | Unit *s = UNIT_DEREF(u->slice); | |
bc432dc7 LP |
1248 | |
1249 | if (more) | |
1250 | /* There's more set now than before. We | |
1251 | * propagate the new mask to the parent's mask | |
1252 | * (not caring if it actually was valid or | |
1253 | * not). */ | |
1254 | ||
1255 | s->cgroup_members_mask |= m; | |
1256 | ||
1257 | else | |
1258 | /* There's less set now than before (or we | |
1259 | * don't know), we need to recalculate | |
1260 | * everything, so let's invalidate the | |
1261 | * parent's members mask */ | |
1262 | ||
1263 | s->cgroup_members_mask_valid = false; | |
1264 | ||
1265 | /* And now make sure that this change also hits our | |
1266 | * grandparents */ | |
1267 | unit_update_cgroup_members_masks(s); | |
6414b7c9 DS |
1268 | } |
1269 | } | |
1270 | ||
efdb0237 | 1271 | static const char *migrate_callback(CGroupMask mask, void *userdata) { |
03b90d4b LP |
1272 | Unit *u = userdata; |
1273 | ||
1274 | assert(mask != 0); | |
1275 | assert(u); | |
1276 | ||
1277 | while (u) { | |
1278 | if (u->cgroup_path && | |
1279 | u->cgroup_realized && | |
1280 | (u->cgroup_realized_mask & mask) == mask) | |
1281 | return u->cgroup_path; | |
1282 | ||
1283 | u = UNIT_DEREF(u->slice); | |
1284 | } | |
1285 | ||
1286 | return NULL; | |
1287 | } | |
1288 | ||
efdb0237 LP |
1289 | char *unit_default_cgroup_path(Unit *u) { |
1290 | _cleanup_free_ char *escaped = NULL, *slice = NULL; | |
1291 | int r; | |
1292 | ||
1293 | assert(u); | |
1294 | ||
1295 | if (unit_has_name(u, SPECIAL_ROOT_SLICE)) | |
1296 | return strdup(u->manager->cgroup_root); | |
1297 | ||
1298 | if (UNIT_ISSET(u->slice) && !unit_has_name(UNIT_DEREF(u->slice), SPECIAL_ROOT_SLICE)) { | |
1299 | r = cg_slice_to_path(UNIT_DEREF(u->slice)->id, &slice); | |
1300 | if (r < 0) | |
1301 | return NULL; | |
1302 | } | |
1303 | ||
1304 | escaped = cg_escape(u->id); | |
1305 | if (!escaped) | |
1306 | return NULL; | |
1307 | ||
1308 | if (slice) | |
605405c6 ZJS |
1309 | return strjoin(u->manager->cgroup_root, "/", slice, "/", |
1310 | escaped); | |
efdb0237 | 1311 | else |
605405c6 | 1312 | return strjoin(u->manager->cgroup_root, "/", escaped); |
efdb0237 LP |
1313 | } |
1314 | ||
1315 | int unit_set_cgroup_path(Unit *u, const char *path) { | |
1316 | _cleanup_free_ char *p = NULL; | |
1317 | int r; | |
1318 | ||
1319 | assert(u); | |
1320 | ||
1321 | if (path) { | |
1322 | p = strdup(path); | |
1323 | if (!p) | |
1324 | return -ENOMEM; | |
1325 | } else | |
1326 | p = NULL; | |
1327 | ||
1328 | if (streq_ptr(u->cgroup_path, p)) | |
1329 | return 0; | |
1330 | ||
1331 | if (p) { | |
1332 | r = hashmap_put(u->manager->cgroup_unit, p, u); | |
1333 | if (r < 0) | |
1334 | return r; | |
1335 | } | |
1336 | ||
1337 | unit_release_cgroup(u); | |
1338 | ||
1339 | u->cgroup_path = p; | |
1340 | p = NULL; | |
1341 | ||
1342 | return 1; | |
1343 | } | |
1344 | ||
1345 | int unit_watch_cgroup(Unit *u) { | |
ab2c3861 | 1346 | _cleanup_free_ char *events = NULL; |
efdb0237 LP |
1347 | int r; |
1348 | ||
1349 | assert(u); | |
1350 | ||
1351 | if (!u->cgroup_path) | |
1352 | return 0; | |
1353 | ||
1354 | if (u->cgroup_inotify_wd >= 0) | |
1355 | return 0; | |
1356 | ||
1357 | /* Only applies to the unified hierarchy */ | |
c22800e4 | 1358 | r = cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER); |
b4cccbc1 LP |
1359 | if (r < 0) |
1360 | return log_error_errno(r, "Failed to determine whether the name=systemd hierarchy is unified: %m"); | |
1361 | if (r == 0) | |
efdb0237 LP |
1362 | return 0; |
1363 | ||
1364 | /* Don't watch the root slice, it's pointless. */ | |
1365 | if (unit_has_name(u, SPECIAL_ROOT_SLICE)) | |
1366 | return 0; | |
1367 | ||
1368 | r = hashmap_ensure_allocated(&u->manager->cgroup_inotify_wd_unit, &trivial_hash_ops); | |
1369 | if (r < 0) | |
1370 | return log_oom(); | |
1371 | ||
ab2c3861 | 1372 | r = cg_get_path(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path, "cgroup.events", &events); |
efdb0237 LP |
1373 | if (r < 0) |
1374 | return log_oom(); | |
1375 | ||
ab2c3861 | 1376 | u->cgroup_inotify_wd = inotify_add_watch(u->manager->cgroup_inotify_fd, events, IN_MODIFY); |
efdb0237 LP |
1377 | if (u->cgroup_inotify_wd < 0) { |
1378 | ||
1379 | if (errno == ENOENT) /* If the directory is already | |
1380 | * gone we don't need to track | |
1381 | * it, so this is not an error */ | |
1382 | return 0; | |
1383 | ||
1384 | return log_unit_error_errno(u, errno, "Failed to add inotify watch descriptor for control group %s: %m", u->cgroup_path); | |
1385 | } | |
1386 | ||
1387 | r = hashmap_put(u->manager->cgroup_inotify_wd_unit, INT_TO_PTR(u->cgroup_inotify_wd), u); | |
1388 | if (r < 0) | |
1389 | return log_unit_error_errno(u, r, "Failed to add inotify watch descriptor to hash map: %m"); | |
1390 | ||
1391 | return 0; | |
1392 | } | |
1393 | ||
1394 | static int unit_create_cgroup( | |
1395 | Unit *u, | |
1396 | CGroupMask target_mask, | |
906c06f6 DM |
1397 | CGroupMask enable_mask, |
1398 | bool needs_bpf) { | |
efdb0237 | 1399 | |
0cd385d3 | 1400 | CGroupContext *c; |
bc432dc7 | 1401 | int r; |
64747e2d | 1402 | |
4ad49000 | 1403 | assert(u); |
64747e2d | 1404 | |
0cd385d3 LP |
1405 | c = unit_get_cgroup_context(u); |
1406 | if (!c) | |
1407 | return 0; | |
1408 | ||
7b3fd631 LP |
1409 | if (!u->cgroup_path) { |
1410 | _cleanup_free_ char *path = NULL; | |
64747e2d | 1411 | |
7b3fd631 LP |
1412 | path = unit_default_cgroup_path(u); |
1413 | if (!path) | |
1414 | return log_oom(); | |
1415 | ||
efdb0237 LP |
1416 | r = unit_set_cgroup_path(u, path); |
1417 | if (r == -EEXIST) | |
1418 | return log_unit_error_errno(u, r, "Control group %s exists already.", path); | |
1419 | if (r < 0) | |
1420 | return log_unit_error_errno(u, r, "Failed to set unit's control group path to %s: %m", path); | |
b58b8e11 HH |
1421 | } |
1422 | ||
03b90d4b | 1423 | /* First, create our own group */ |
efdb0237 | 1424 | r = cg_create_everywhere(u->manager->cgroup_supported, target_mask, u->cgroup_path); |
23bbb0de | 1425 | if (r < 0) |
efdb0237 LP |
1426 | return log_unit_error_errno(u, r, "Failed to create cgroup %s: %m", u->cgroup_path); |
1427 | ||
1428 | /* Start watching it */ | |
1429 | (void) unit_watch_cgroup(u); | |
1430 | ||
1431 | /* Enable all controllers we need */ | |
1432 | r = cg_enable_everywhere(u->manager->cgroup_supported, enable_mask, u->cgroup_path); | |
1433 | if (r < 0) | |
1434 | log_unit_warning_errno(u, r, "Failed to enable controllers on cgroup %s, ignoring: %m", u->cgroup_path); | |
03b90d4b LP |
1435 | |
1436 | /* Keep track that this is now realized */ | |
4ad49000 | 1437 | u->cgroup_realized = true; |
efdb0237 | 1438 | u->cgroup_realized_mask = target_mask; |
ccf78df1 | 1439 | u->cgroup_enabled_mask = enable_mask; |
906c06f6 | 1440 | u->cgroup_bpf_state = needs_bpf ? UNIT_CGROUP_BPF_ON : UNIT_CGROUP_BPF_OFF; |
4ad49000 | 1441 | |
0cd385d3 LP |
1442 | if (u->type != UNIT_SLICE && !c->delegate) { |
1443 | ||
1444 | /* Then, possibly move things over, but not if | |
1445 | * subgroups may contain processes, which is the case | |
1446 | * for slice and delegation units. */ | |
1447 | r = cg_migrate_everywhere(u->manager->cgroup_supported, u->cgroup_path, u->cgroup_path, migrate_callback, u); | |
1448 | if (r < 0) | |
efdb0237 | 1449 | log_unit_warning_errno(u, r, "Failed to migrate cgroup from to %s, ignoring: %m", u->cgroup_path); |
0cd385d3 | 1450 | } |
03b90d4b | 1451 | |
64747e2d LP |
1452 | return 0; |
1453 | } | |
1454 | ||
7b3fd631 LP |
1455 | int unit_attach_pids_to_cgroup(Unit *u) { |
1456 | int r; | |
1457 | assert(u); | |
1458 | ||
1459 | r = unit_realize_cgroup(u); | |
1460 | if (r < 0) | |
1461 | return r; | |
1462 | ||
1463 | r = cg_attach_many_everywhere(u->manager->cgroup_supported, u->cgroup_path, u->pids, migrate_callback, u); | |
1464 | if (r < 0) | |
1465 | return r; | |
1466 | ||
1467 | return 0; | |
1468 | } | |
1469 | ||
4b58153d LP |
1470 | static void cgroup_xattr_apply(Unit *u) { |
1471 | char ids[SD_ID128_STRING_MAX]; | |
1472 | int r; | |
1473 | ||
1474 | assert(u); | |
1475 | ||
1476 | if (!MANAGER_IS_SYSTEM(u->manager)) | |
1477 | return; | |
1478 | ||
1479 | if (sd_id128_is_null(u->invocation_id)) | |
1480 | return; | |
1481 | ||
1482 | r = cg_set_xattr(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path, | |
1483 | "trusted.invocation_id", | |
1484 | sd_id128_to_string(u->invocation_id, ids), 32, | |
1485 | 0); | |
1486 | if (r < 0) | |
1487 | log_unit_warning_errno(u, r, "Failed to set invocation ID on control group %s, ignoring: %m", u->cgroup_path); | |
1488 | } | |
1489 | ||
906c06f6 DM |
1490 | static bool unit_has_mask_realized( |
1491 | Unit *u, | |
1492 | CGroupMask target_mask, | |
1493 | CGroupMask enable_mask, | |
1494 | bool needs_bpf) { | |
1495 | ||
bc432dc7 LP |
1496 | assert(u); |
1497 | ||
906c06f6 DM |
1498 | return u->cgroup_realized && |
1499 | u->cgroup_realized_mask == target_mask && | |
1500 | u->cgroup_enabled_mask == enable_mask && | |
1501 | ((needs_bpf && u->cgroup_bpf_state == UNIT_CGROUP_BPF_ON) || | |
1502 | (!needs_bpf && u->cgroup_bpf_state == UNIT_CGROUP_BPF_OFF)); | |
6414b7c9 DS |
1503 | } |
1504 | ||
1505 | /* Check if necessary controllers and attributes for a unit are in place. | |
1506 | * | |
1507 | * If so, do nothing. | |
1508 | * If not, create paths, move processes over, and set attributes. | |
1509 | * | |
1510 | * Returns 0 on success and < 0 on failure. */ | |
db785129 | 1511 | static int unit_realize_cgroup_now(Unit *u, ManagerState state) { |
efdb0237 | 1512 | CGroupMask target_mask, enable_mask; |
906c06f6 | 1513 | bool needs_bpf, apply_bpf; |
6414b7c9 | 1514 | int r; |
64747e2d | 1515 | |
4ad49000 | 1516 | assert(u); |
64747e2d | 1517 | |
91a6073e LP |
1518 | if (u->in_cgroup_realize_queue) { |
1519 | LIST_REMOVE(cgroup_realize_queue, u->manager->cgroup_realize_queue, u); | |
1520 | u->in_cgroup_realize_queue = false; | |
4ad49000 | 1521 | } |
64747e2d | 1522 | |
efdb0237 | 1523 | target_mask = unit_get_target_mask(u); |
ccf78df1 | 1524 | enable_mask = unit_get_enable_mask(u); |
906c06f6 | 1525 | needs_bpf = unit_get_needs_bpf(u); |
ccf78df1 | 1526 | |
906c06f6 | 1527 | if (unit_has_mask_realized(u, target_mask, enable_mask, needs_bpf)) |
0a1eb06d | 1528 | return 0; |
64747e2d | 1529 | |
906c06f6 DM |
1530 | /* Make sure we apply the BPF filters either when one is configured, or if none is configured but previously |
1531 | * the state was anything but off. This way, if a unit with a BPF filter applied is reconfigured to lose it | |
1532 | * this will trickle down properly to cgroupfs. */ | |
1533 | apply_bpf = needs_bpf || u->cgroup_bpf_state != UNIT_CGROUP_BPF_OFF; | |
1534 | ||
4ad49000 | 1535 | /* First, realize parents */ |
6414b7c9 | 1536 | if (UNIT_ISSET(u->slice)) { |
db785129 | 1537 | r = unit_realize_cgroup_now(UNIT_DEREF(u->slice), state); |
6414b7c9 DS |
1538 | if (r < 0) |
1539 | return r; | |
1540 | } | |
4ad49000 LP |
1541 | |
1542 | /* And then do the real work */ | |
906c06f6 | 1543 | r = unit_create_cgroup(u, target_mask, enable_mask, needs_bpf); |
6414b7c9 DS |
1544 | if (r < 0) |
1545 | return r; | |
1546 | ||
1547 | /* Finally, apply the necessary attributes. */ | |
906c06f6 | 1548 | cgroup_context_apply(u, target_mask, apply_bpf, state); |
4b58153d | 1549 | cgroup_xattr_apply(u); |
6414b7c9 DS |
1550 | |
1551 | return 0; | |
64747e2d LP |
1552 | } |
1553 | ||
91a6073e | 1554 | static void unit_add_to_cgroup_realize_queue(Unit *u) { |
58d83430 | 1555 | assert(u); |
ecedd90f | 1556 | |
91a6073e | 1557 | if (u->in_cgroup_realize_queue) |
4ad49000 | 1558 | return; |
8e274523 | 1559 | |
91a6073e LP |
1560 | LIST_PREPEND(cgroup_realize_queue, u->manager->cgroup_realize_queue, u); |
1561 | u->in_cgroup_realize_queue = true; | |
4ad49000 | 1562 | } |
8c6db833 | 1563 | |
91a6073e | 1564 | unsigned manager_dispatch_cgroup_realize_queue(Manager *m) { |
db785129 | 1565 | ManagerState state; |
4ad49000 | 1566 | unsigned n = 0; |
db785129 | 1567 | Unit *i; |
6414b7c9 | 1568 | int r; |
ecedd90f | 1569 | |
91a6073e LP |
1570 | assert(m); |
1571 | ||
db785129 LP |
1572 | state = manager_state(m); |
1573 | ||
91a6073e LP |
1574 | while ((i = m->cgroup_realize_queue)) { |
1575 | assert(i->in_cgroup_realize_queue); | |
ecedd90f | 1576 | |
db785129 | 1577 | r = unit_realize_cgroup_now(i, state); |
6414b7c9 | 1578 | if (r < 0) |
efdb0237 | 1579 | log_warning_errno(r, "Failed to realize cgroups for queued unit %s, ignoring: %m", i->id); |
0a1eb06d | 1580 | |
4ad49000 LP |
1581 | n++; |
1582 | } | |
ecedd90f | 1583 | |
4ad49000 | 1584 | return n; |
8e274523 LP |
1585 | } |
1586 | ||
91a6073e | 1587 | static void unit_add_siblings_to_cgroup_realize_queue(Unit *u) { |
4ad49000 | 1588 | Unit *slice; |
ca949c9d | 1589 | |
4ad49000 LP |
1590 | /* This adds the siblings of the specified unit and the |
1591 | * siblings of all parent units to the cgroup queue. (But | |
1592 | * neither the specified unit itself nor the parents.) */ | |
1593 | ||
1594 | while ((slice = UNIT_DEREF(u->slice))) { | |
1595 | Iterator i; | |
1596 | Unit *m; | |
eef85c4a | 1597 | void *v; |
8f53a7b8 | 1598 | |
eef85c4a | 1599 | HASHMAP_FOREACH_KEY(v, m, u->dependencies[UNIT_BEFORE], i) { |
4ad49000 LP |
1600 | if (m == u) |
1601 | continue; | |
8e274523 | 1602 | |
6414b7c9 DS |
1603 | /* Skip units that have a dependency on the slice |
1604 | * but aren't actually in it. */ | |
4ad49000 | 1605 | if (UNIT_DEREF(m->slice) != slice) |
50159e6a | 1606 | continue; |
8e274523 | 1607 | |
6414b7c9 DS |
1608 | /* No point in doing cgroup application for units |
1609 | * without active processes. */ | |
1610 | if (UNIT_IS_INACTIVE_OR_FAILED(unit_active_state(m))) | |
1611 | continue; | |
1612 | ||
1613 | /* If the unit doesn't need any new controllers | |
1614 | * and has current ones realized, it doesn't need | |
1615 | * any changes. */ | |
906c06f6 DM |
1616 | if (unit_has_mask_realized(m, |
1617 | unit_get_target_mask(m), | |
1618 | unit_get_enable_mask(m), | |
1619 | unit_get_needs_bpf(m))) | |
6414b7c9 DS |
1620 | continue; |
1621 | ||
91a6073e | 1622 | unit_add_to_cgroup_realize_queue(m); |
50159e6a LP |
1623 | } |
1624 | ||
4ad49000 | 1625 | u = slice; |
8e274523 | 1626 | } |
4ad49000 LP |
1627 | } |
1628 | ||
0a1eb06d | 1629 | int unit_realize_cgroup(Unit *u) { |
4ad49000 LP |
1630 | assert(u); |
1631 | ||
35b7ff80 | 1632 | if (!UNIT_HAS_CGROUP_CONTEXT(u)) |
0a1eb06d | 1633 | return 0; |
8e274523 | 1634 | |
4ad49000 LP |
1635 | /* So, here's the deal: when realizing the cgroups for this |
1636 | * unit, we need to first create all parents, but there's more | |
1637 | * actually: for the weight-based controllers we also need to | |
1638 | * make sure that all our siblings (i.e. units that are in the | |
73e231ab | 1639 | * same slice as we are) have cgroups, too. Otherwise, things |
4ad49000 LP |
1640 | * would become very uneven as each of their processes would |
1641 | * get as much resources as all our group together. This call | |
1642 | * will synchronously create the parent cgroups, but will | |
1643 | * defer work on the siblings to the next event loop | |
1644 | * iteration. */ | |
ca949c9d | 1645 | |
4ad49000 | 1646 | /* Add all sibling slices to the cgroup queue. */ |
91a6073e | 1647 | unit_add_siblings_to_cgroup_realize_queue(u); |
4ad49000 | 1648 | |
6414b7c9 | 1649 | /* And realize this one now (and apply the values) */ |
db785129 | 1650 | return unit_realize_cgroup_now(u, manager_state(u->manager)); |
8e274523 LP |
1651 | } |
1652 | ||
efdb0237 LP |
1653 | void unit_release_cgroup(Unit *u) { |
1654 | assert(u); | |
1655 | ||
1656 | /* Forgets all cgroup details for this cgroup */ | |
1657 | ||
1658 | if (u->cgroup_path) { | |
1659 | (void) hashmap_remove(u->manager->cgroup_unit, u->cgroup_path); | |
1660 | u->cgroup_path = mfree(u->cgroup_path); | |
1661 | } | |
1662 | ||
1663 | if (u->cgroup_inotify_wd >= 0) { | |
1664 | if (inotify_rm_watch(u->manager->cgroup_inotify_fd, u->cgroup_inotify_wd) < 0) | |
1665 | log_unit_debug_errno(u, errno, "Failed to remove cgroup inotify watch %i for %s, ignoring", u->cgroup_inotify_wd, u->id); | |
1666 | ||
1667 | (void) hashmap_remove(u->manager->cgroup_inotify_wd_unit, INT_TO_PTR(u->cgroup_inotify_wd)); | |
1668 | u->cgroup_inotify_wd = -1; | |
1669 | } | |
1670 | } | |
1671 | ||
1672 | void unit_prune_cgroup(Unit *u) { | |
8e274523 | 1673 | int r; |
efdb0237 | 1674 | bool is_root_slice; |
8e274523 | 1675 | |
4ad49000 | 1676 | assert(u); |
8e274523 | 1677 | |
efdb0237 LP |
1678 | /* Removes the cgroup, if empty and possible, and stops watching it. */ |
1679 | ||
4ad49000 LP |
1680 | if (!u->cgroup_path) |
1681 | return; | |
8e274523 | 1682 | |
fe700f46 LP |
1683 | (void) unit_get_cpu_usage(u, NULL); /* Cache the last CPU usage value before we destroy the cgroup */ |
1684 | ||
efdb0237 LP |
1685 | is_root_slice = unit_has_name(u, SPECIAL_ROOT_SLICE); |
1686 | ||
1687 | r = cg_trim_everywhere(u->manager->cgroup_supported, u->cgroup_path, !is_root_slice); | |
dab5bf85 | 1688 | if (r < 0) { |
f29ff115 | 1689 | log_unit_debug_errno(u, r, "Failed to destroy cgroup %s, ignoring: %m", u->cgroup_path); |
dab5bf85 RL |
1690 | return; |
1691 | } | |
8e274523 | 1692 | |
efdb0237 LP |
1693 | if (is_root_slice) |
1694 | return; | |
1695 | ||
1696 | unit_release_cgroup(u); | |
0a1eb06d | 1697 | |
4ad49000 | 1698 | u->cgroup_realized = false; |
bc432dc7 | 1699 | u->cgroup_realized_mask = 0; |
ccf78df1 | 1700 | u->cgroup_enabled_mask = 0; |
8e274523 LP |
1701 | } |
1702 | ||
efdb0237 | 1703 | int unit_search_main_pid(Unit *u, pid_t *ret) { |
4ad49000 LP |
1704 | _cleanup_fclose_ FILE *f = NULL; |
1705 | pid_t pid = 0, npid, mypid; | |
efdb0237 | 1706 | int r; |
4ad49000 LP |
1707 | |
1708 | assert(u); | |
efdb0237 | 1709 | assert(ret); |
4ad49000 LP |
1710 | |
1711 | if (!u->cgroup_path) | |
efdb0237 | 1712 | return -ENXIO; |
4ad49000 | 1713 | |
efdb0237 LP |
1714 | r = cg_enumerate_processes(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path, &f); |
1715 | if (r < 0) | |
1716 | return r; | |
4ad49000 | 1717 | |
df0ff127 | 1718 | mypid = getpid_cached(); |
4ad49000 LP |
1719 | while (cg_read_pid(f, &npid) > 0) { |
1720 | pid_t ppid; | |
1721 | ||
1722 | if (npid == pid) | |
1723 | continue; | |
8e274523 | 1724 | |
4ad49000 | 1725 | /* Ignore processes that aren't our kids */ |
6bc73acb | 1726 | if (get_process_ppid(npid, &ppid) >= 0 && ppid != mypid) |
4ad49000 | 1727 | continue; |
8e274523 | 1728 | |
efdb0237 | 1729 | if (pid != 0) |
4ad49000 LP |
1730 | /* Dang, there's more than one daemonized PID |
1731 | in this group, so we don't know what process | |
1732 | is the main process. */ | |
efdb0237 LP |
1733 | |
1734 | return -ENODATA; | |
8e274523 | 1735 | |
4ad49000 | 1736 | pid = npid; |
8e274523 LP |
1737 | } |
1738 | ||
efdb0237 LP |
1739 | *ret = pid; |
1740 | return 0; | |
1741 | } | |
1742 | ||
1743 | static int unit_watch_pids_in_path(Unit *u, const char *path) { | |
b3c5bad3 | 1744 | _cleanup_closedir_ DIR *d = NULL; |
efdb0237 LP |
1745 | _cleanup_fclose_ FILE *f = NULL; |
1746 | int ret = 0, r; | |
1747 | ||
1748 | assert(u); | |
1749 | assert(path); | |
1750 | ||
1751 | r = cg_enumerate_processes(SYSTEMD_CGROUP_CONTROLLER, path, &f); | |
1752 | if (r < 0) | |
1753 | ret = r; | |
1754 | else { | |
1755 | pid_t pid; | |
1756 | ||
1757 | while ((r = cg_read_pid(f, &pid)) > 0) { | |
1758 | r = unit_watch_pid(u, pid); | |
1759 | if (r < 0 && ret >= 0) | |
1760 | ret = r; | |
1761 | } | |
1762 | ||
1763 | if (r < 0 && ret >= 0) | |
1764 | ret = r; | |
1765 | } | |
1766 | ||
1767 | r = cg_enumerate_subgroups(SYSTEMD_CGROUP_CONTROLLER, path, &d); | |
1768 | if (r < 0) { | |
1769 | if (ret >= 0) | |
1770 | ret = r; | |
1771 | } else { | |
1772 | char *fn; | |
1773 | ||
1774 | while ((r = cg_read_subgroup(d, &fn)) > 0) { | |
1775 | _cleanup_free_ char *p = NULL; | |
1776 | ||
605405c6 | 1777 | p = strjoin(path, "/", fn); |
efdb0237 LP |
1778 | free(fn); |
1779 | ||
1780 | if (!p) | |
1781 | return -ENOMEM; | |
1782 | ||
1783 | r = unit_watch_pids_in_path(u, p); | |
1784 | if (r < 0 && ret >= 0) | |
1785 | ret = r; | |
1786 | } | |
1787 | ||
1788 | if (r < 0 && ret >= 0) | |
1789 | ret = r; | |
1790 | } | |
1791 | ||
1792 | return ret; | |
1793 | } | |
1794 | ||
1795 | int unit_watch_all_pids(Unit *u) { | |
b4cccbc1 LP |
1796 | int r; |
1797 | ||
efdb0237 LP |
1798 | assert(u); |
1799 | ||
1800 | /* Adds all PIDs from our cgroup to the set of PIDs we | |
1801 | * watch. This is a fallback logic for cases where we do not | |
1802 | * get reliable cgroup empty notifications: we try to use | |
1803 | * SIGCHLD as replacement. */ | |
1804 | ||
1805 | if (!u->cgroup_path) | |
1806 | return -ENOENT; | |
1807 | ||
c22800e4 | 1808 | r = cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER); |
b4cccbc1 LP |
1809 | if (r < 0) |
1810 | return r; | |
1811 | if (r > 0) /* On unified we can use proper notifications */ | |
efdb0237 LP |
1812 | return 0; |
1813 | ||
1814 | return unit_watch_pids_in_path(u, u->cgroup_path); | |
1815 | } | |
1816 | ||
09e24654 LP |
1817 | static int on_cgroup_empty_event(sd_event_source *s, void *userdata) { |
1818 | Manager *m = userdata; | |
1819 | Unit *u; | |
efdb0237 LP |
1820 | int r; |
1821 | ||
09e24654 LP |
1822 | assert(s); |
1823 | assert(m); | |
efdb0237 | 1824 | |
09e24654 LP |
1825 | u = m->cgroup_empty_queue; |
1826 | if (!u) | |
efdb0237 LP |
1827 | return 0; |
1828 | ||
09e24654 LP |
1829 | assert(u->in_cgroup_empty_queue); |
1830 | u->in_cgroup_empty_queue = false; | |
1831 | LIST_REMOVE(cgroup_empty_queue, m->cgroup_empty_queue, u); | |
1832 | ||
1833 | if (m->cgroup_empty_queue) { | |
1834 | /* More stuff queued, let's make sure we remain enabled */ | |
1835 | r = sd_event_source_set_enabled(s, SD_EVENT_ONESHOT); | |
1836 | if (r < 0) | |
1837 | log_debug_errno(r, "Failed to reenable cgroup empty event source: %m"); | |
1838 | } | |
efdb0237 LP |
1839 | |
1840 | unit_add_to_gc_queue(u); | |
1841 | ||
1842 | if (UNIT_VTABLE(u)->notify_cgroup_empty) | |
1843 | UNIT_VTABLE(u)->notify_cgroup_empty(u); | |
1844 | ||
1845 | return 0; | |
1846 | } | |
1847 | ||
09e24654 LP |
1848 | void unit_add_to_cgroup_empty_queue(Unit *u) { |
1849 | int r; | |
1850 | ||
1851 | assert(u); | |
1852 | ||
1853 | /* Note that there are four different ways how cgroup empty events reach us: | |
1854 | * | |
1855 | * 1. On the unified hierarchy we get an inotify event on the cgroup | |
1856 | * | |
1857 | * 2. On the legacy hierarchy, when running in system mode, we get a datagram on the cgroup agent socket | |
1858 | * | |
1859 | * 3. On the legacy hierarchy, when running in user mode, we get a D-Bus signal on the system bus | |
1860 | * | |
1861 | * 4. On the legacy hierarchy, in service units we start watching all processes of the cgroup for SIGCHLD as | |
1862 | * soon as we get one SIGCHLD, to deal with unreliable cgroup notifications. | |
1863 | * | |
1864 | * Regardless which way we got the notification, we'll verify it here, and then add it to a separate | |
1865 | * queue. This queue will be dispatched at a lower priority than the SIGCHLD handler, so that we always use | |
1866 | * SIGCHLD if we can get it first, and only use the cgroup empty notifications if there's no SIGCHLD pending | |
1867 | * (which might happen if the cgroup doesn't contain processes that are our own child, which is typically the | |
1868 | * case for scope units). */ | |
1869 | ||
1870 | if (u->in_cgroup_empty_queue) | |
1871 | return; | |
1872 | ||
1873 | /* Let's verify that the cgroup is really empty */ | |
1874 | if (!u->cgroup_path) | |
1875 | return; | |
1876 | r = cg_is_empty_recursive(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path); | |
1877 | if (r < 0) { | |
1878 | log_unit_debug_errno(u, r, "Failed to determine whether cgroup %s is empty: %m", u->cgroup_path); | |
1879 | return; | |
1880 | } | |
1881 | if (r == 0) | |
1882 | return; | |
1883 | ||
1884 | LIST_PREPEND(cgroup_empty_queue, u->manager->cgroup_empty_queue, u); | |
1885 | u->in_cgroup_empty_queue = true; | |
1886 | ||
1887 | /* Trigger the defer event */ | |
1888 | r = sd_event_source_set_enabled(u->manager->cgroup_empty_event_source, SD_EVENT_ONESHOT); | |
1889 | if (r < 0) | |
1890 | log_debug_errno(r, "Failed to enable cgroup empty event source: %m"); | |
1891 | } | |
1892 | ||
efdb0237 LP |
1893 | static int on_cgroup_inotify_event(sd_event_source *s, int fd, uint32_t revents, void *userdata) { |
1894 | Manager *m = userdata; | |
1895 | ||
1896 | assert(s); | |
1897 | assert(fd >= 0); | |
1898 | assert(m); | |
1899 | ||
1900 | for (;;) { | |
1901 | union inotify_event_buffer buffer; | |
1902 | struct inotify_event *e; | |
1903 | ssize_t l; | |
1904 | ||
1905 | l = read(fd, &buffer, sizeof(buffer)); | |
1906 | if (l < 0) { | |
47249640 | 1907 | if (IN_SET(errno, EINTR, EAGAIN)) |
efdb0237 LP |
1908 | return 0; |
1909 | ||
1910 | return log_error_errno(errno, "Failed to read control group inotify events: %m"); | |
1911 | } | |
1912 | ||
1913 | FOREACH_INOTIFY_EVENT(e, buffer, l) { | |
1914 | Unit *u; | |
1915 | ||
1916 | if (e->wd < 0) | |
1917 | /* Queue overflow has no watch descriptor */ | |
1918 | continue; | |
1919 | ||
1920 | if (e->mask & IN_IGNORED) | |
1921 | /* The watch was just removed */ | |
1922 | continue; | |
1923 | ||
1924 | u = hashmap_get(m->cgroup_inotify_wd_unit, INT_TO_PTR(e->wd)); | |
1925 | if (!u) /* Not that inotify might deliver | |
1926 | * events for a watch even after it | |
1927 | * was removed, because it was queued | |
1928 | * before the removal. Let's ignore | |
1929 | * this here safely. */ | |
1930 | continue; | |
1931 | ||
09e24654 | 1932 | unit_add_to_cgroup_empty_queue(u); |
efdb0237 LP |
1933 | } |
1934 | } | |
8e274523 LP |
1935 | } |
1936 | ||
8e274523 | 1937 | int manager_setup_cgroup(Manager *m) { |
9444b1f2 | 1938 | _cleanup_free_ char *path = NULL; |
10bd3e2e | 1939 | const char *scope_path; |
efdb0237 | 1940 | CGroupController c; |
b4cccbc1 | 1941 | int r, all_unified; |
efdb0237 | 1942 | char *e; |
8e274523 LP |
1943 | |
1944 | assert(m); | |
1945 | ||
35d2e7ec | 1946 | /* 1. Determine hierarchy */ |
efdb0237 | 1947 | m->cgroup_root = mfree(m->cgroup_root); |
9444b1f2 | 1948 | r = cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER, 0, &m->cgroup_root); |
23bbb0de MS |
1949 | if (r < 0) |
1950 | return log_error_errno(r, "Cannot determine cgroup we are running in: %m"); | |
8e274523 | 1951 | |
efdb0237 LP |
1952 | /* Chop off the init scope, if we are already located in it */ |
1953 | e = endswith(m->cgroup_root, "/" SPECIAL_INIT_SCOPE); | |
0d8c31ff | 1954 | |
efdb0237 LP |
1955 | /* LEGACY: Also chop off the system slice if we are in |
1956 | * it. This is to support live upgrades from older systemd | |
1957 | * versions where PID 1 was moved there. Also see | |
1958 | * cg_get_root_path(). */ | |
463d0d15 | 1959 | if (!e && MANAGER_IS_SYSTEM(m)) { |
9444b1f2 | 1960 | e = endswith(m->cgroup_root, "/" SPECIAL_SYSTEM_SLICE); |
15c60e99 | 1961 | if (!e) |
efdb0237 | 1962 | e = endswith(m->cgroup_root, "/system"); /* even more legacy */ |
0baf24dd | 1963 | } |
efdb0237 LP |
1964 | if (e) |
1965 | *e = 0; | |
7ccfb64a | 1966 | |
7546145e LP |
1967 | /* And make sure to store away the root value without trailing slash, even for the root dir, so that we can |
1968 | * easily prepend it everywhere. */ | |
1969 | delete_trailing_chars(m->cgroup_root, "/"); | |
8e274523 | 1970 | |
35d2e7ec | 1971 | /* 2. Show data */ |
9444b1f2 | 1972 | r = cg_get_path(SYSTEMD_CGROUP_CONTROLLER, m->cgroup_root, NULL, &path); |
23bbb0de MS |
1973 | if (r < 0) |
1974 | return log_error_errno(r, "Cannot find cgroup mount point: %m"); | |
8e274523 | 1975 | |
415fc41c TH |
1976 | r = cg_unified_flush(); |
1977 | if (r < 0) | |
1978 | return log_error_errno(r, "Couldn't determine if we are running in the unified hierarchy: %m"); | |
5da38d07 | 1979 | |
b4cccbc1 | 1980 | all_unified = cg_all_unified(); |
d4c819ed ZJS |
1981 | if (all_unified < 0) |
1982 | return log_error_errno(all_unified, "Couldn't determine whether we are in all unified mode: %m"); | |
1983 | if (all_unified > 0) | |
efdb0237 | 1984 | log_debug("Unified cgroup hierarchy is located at %s.", path); |
b4cccbc1 | 1985 | else { |
c22800e4 | 1986 | r = cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER); |
b4cccbc1 LP |
1987 | if (r < 0) |
1988 | return log_error_errno(r, "Failed to determine whether systemd's own controller is in unified mode: %m"); | |
1989 | if (r > 0) | |
1990 | log_debug("Unified cgroup hierarchy is located at %s. Controllers are on legacy hierarchies.", path); | |
1991 | else | |
1992 | log_debug("Using cgroup controller " SYSTEMD_CGROUP_CONTROLLER_LEGACY ". File system hierarchy is at %s.", path); | |
1993 | } | |
efdb0237 | 1994 | |
09e24654 LP |
1995 | /* 3. Allocate cgroup empty defer event source */ |
1996 | m->cgroup_empty_event_source = sd_event_source_unref(m->cgroup_empty_event_source); | |
1997 | r = sd_event_add_defer(m->event, &m->cgroup_empty_event_source, on_cgroup_empty_event, m); | |
1998 | if (r < 0) | |
1999 | return log_error_errno(r, "Failed to create cgroup empty event source: %m"); | |
2000 | ||
2001 | r = sd_event_source_set_priority(m->cgroup_empty_event_source, SD_EVENT_PRIORITY_NORMAL-5); | |
2002 | if (r < 0) | |
2003 | return log_error_errno(r, "Failed to set priority of cgroup empty event source: %m"); | |
2004 | ||
2005 | r = sd_event_source_set_enabled(m->cgroup_empty_event_source, SD_EVENT_OFF); | |
2006 | if (r < 0) | |
2007 | return log_error_errno(r, "Failed to disable cgroup empty event source: %m"); | |
2008 | ||
2009 | (void) sd_event_source_set_description(m->cgroup_empty_event_source, "cgroup-empty"); | |
2010 | ||
2011 | /* 4. Install notifier inotify object, or agent */ | |
10bd3e2e | 2012 | if (cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER) > 0) { |
c6c18be3 | 2013 | |
09e24654 | 2014 | /* In the unified hierarchy we can get cgroup empty notifications via inotify. */ |
efdb0237 | 2015 | |
10bd3e2e LP |
2016 | m->cgroup_inotify_event_source = sd_event_source_unref(m->cgroup_inotify_event_source); |
2017 | safe_close(m->cgroup_inotify_fd); | |
efdb0237 | 2018 | |
10bd3e2e LP |
2019 | m->cgroup_inotify_fd = inotify_init1(IN_NONBLOCK|IN_CLOEXEC); |
2020 | if (m->cgroup_inotify_fd < 0) | |
2021 | return log_error_errno(errno, "Failed to create control group inotify object: %m"); | |
efdb0237 | 2022 | |
10bd3e2e LP |
2023 | r = sd_event_add_io(m->event, &m->cgroup_inotify_event_source, m->cgroup_inotify_fd, EPOLLIN, on_cgroup_inotify_event, m); |
2024 | if (r < 0) | |
2025 | return log_error_errno(r, "Failed to watch control group inotify object: %m"); | |
efdb0237 | 2026 | |
10bd3e2e LP |
2027 | /* Process cgroup empty notifications early, but after service notifications and SIGCHLD. Also |
2028 | * see handling of cgroup agent notifications, for the classic cgroup hierarchy support. */ | |
09e24654 | 2029 | r = sd_event_source_set_priority(m->cgroup_inotify_event_source, SD_EVENT_PRIORITY_NORMAL-4); |
10bd3e2e LP |
2030 | if (r < 0) |
2031 | return log_error_errno(r, "Failed to set priority of inotify event source: %m"); | |
efdb0237 | 2032 | |
10bd3e2e | 2033 | (void) sd_event_source_set_description(m->cgroup_inotify_event_source, "cgroup-inotify"); |
efdb0237 | 2034 | |
10bd3e2e | 2035 | } else if (MANAGER_IS_SYSTEM(m) && m->test_run_flags == 0) { |
efdb0237 | 2036 | |
10bd3e2e LP |
2037 | /* On the legacy hierarchy we only get notifications via cgroup agents. (Which isn't really reliable, |
2038 | * since it does not generate events when control groups with children run empty. */ | |
8e274523 | 2039 | |
10bd3e2e | 2040 | r = cg_install_release_agent(SYSTEMD_CGROUP_CONTROLLER, SYSTEMD_CGROUP_AGENT_PATH); |
23bbb0de | 2041 | if (r < 0) |
10bd3e2e LP |
2042 | log_warning_errno(r, "Failed to install release agent, ignoring: %m"); |
2043 | else if (r > 0) | |
2044 | log_debug("Installed release agent."); | |
2045 | else if (r == 0) | |
2046 | log_debug("Release agent already installed."); | |
2047 | } | |
efdb0237 | 2048 | |
09e24654 | 2049 | /* 5. Make sure we are in the special "init.scope" unit in the root slice. */ |
10bd3e2e LP |
2050 | scope_path = strjoina(m->cgroup_root, "/" SPECIAL_INIT_SCOPE); |
2051 | r = cg_create_and_attach(SYSTEMD_CGROUP_CONTROLLER, scope_path, 0); | |
2052 | if (r < 0) | |
2053 | return log_error_errno(r, "Failed to create %s control group: %m", scope_path); | |
c6c18be3 | 2054 | |
09e24654 | 2055 | /* Also, move all other userspace processes remaining in the root cgroup into that scope. */ |
10bd3e2e LP |
2056 | r = cg_migrate(SYSTEMD_CGROUP_CONTROLLER, m->cgroup_root, SYSTEMD_CGROUP_CONTROLLER, scope_path, 0); |
2057 | if (r < 0) | |
2058 | log_warning_errno(r, "Couldn't move remaining userspace processes, ignoring: %m"); | |
0d8c31ff | 2059 | |
09e24654 | 2060 | /* 6. And pin it, so that it cannot be unmounted */ |
10bd3e2e LP |
2061 | safe_close(m->pin_cgroupfs_fd); |
2062 | m->pin_cgroupfs_fd = open(path, O_RDONLY|O_CLOEXEC|O_DIRECTORY|O_NOCTTY|O_NONBLOCK); | |
2063 | if (m->pin_cgroupfs_fd < 0) | |
2064 | return log_error_errno(errno, "Failed to open pin file: %m"); | |
2065 | ||
09e24654 | 2066 | /* 7. Always enable hierarchical support if it exists... */ |
10bd3e2e LP |
2067 | if (!all_unified && m->test_run_flags == 0) |
2068 | (void) cg_set_attribute("memory", "/", "memory.use_hierarchy", "1"); | |
c6c18be3 | 2069 | |
09e24654 | 2070 | /* 8. Figure out which controllers are supported, and log about it */ |
efdb0237 LP |
2071 | r = cg_mask_supported(&m->cgroup_supported); |
2072 | if (r < 0) | |
2073 | return log_error_errno(r, "Failed to determine supported controllers: %m"); | |
efdb0237 | 2074 | for (c = 0; c < _CGROUP_CONTROLLER_MAX; c++) |
eee0a1e4 | 2075 | log_debug("Controller '%s' supported: %s", cgroup_controller_to_string(c), yes_no(m->cgroup_supported & CGROUP_CONTROLLER_TO_MASK(c))); |
9156e799 | 2076 | |
a32360f1 | 2077 | return 0; |
8e274523 LP |
2078 | } |
2079 | ||
c6c18be3 | 2080 | void manager_shutdown_cgroup(Manager *m, bool delete) { |
8e274523 LP |
2081 | assert(m); |
2082 | ||
9444b1f2 LP |
2083 | /* We can't really delete the group, since we are in it. But |
2084 | * let's trim it. */ | |
2085 | if (delete && m->cgroup_root) | |
efdb0237 LP |
2086 | (void) cg_trim(SYSTEMD_CGROUP_CONTROLLER, m->cgroup_root, false); |
2087 | ||
09e24654 LP |
2088 | m->cgroup_empty_event_source = sd_event_source_unref(m->cgroup_empty_event_source); |
2089 | ||
efdb0237 LP |
2090 | m->cgroup_inotify_wd_unit = hashmap_free(m->cgroup_inotify_wd_unit); |
2091 | ||
2092 | m->cgroup_inotify_event_source = sd_event_source_unref(m->cgroup_inotify_event_source); | |
2093 | m->cgroup_inotify_fd = safe_close(m->cgroup_inotify_fd); | |
8e274523 | 2094 | |
03e334a1 | 2095 | m->pin_cgroupfs_fd = safe_close(m->pin_cgroupfs_fd); |
c6c18be3 | 2096 | |
efdb0237 | 2097 | m->cgroup_root = mfree(m->cgroup_root); |
8e274523 LP |
2098 | } |
2099 | ||
4ad49000 | 2100 | Unit* manager_get_unit_by_cgroup(Manager *m, const char *cgroup) { |
acb14d31 | 2101 | char *p; |
4ad49000 | 2102 | Unit *u; |
acb14d31 LP |
2103 | |
2104 | assert(m); | |
2105 | assert(cgroup); | |
acb14d31 | 2106 | |
4ad49000 LP |
2107 | u = hashmap_get(m->cgroup_unit, cgroup); |
2108 | if (u) | |
2109 | return u; | |
acb14d31 | 2110 | |
8e70580b | 2111 | p = strdupa(cgroup); |
acb14d31 LP |
2112 | for (;;) { |
2113 | char *e; | |
2114 | ||
2115 | e = strrchr(p, '/'); | |
efdb0237 LP |
2116 | if (!e || e == p) |
2117 | return hashmap_get(m->cgroup_unit, SPECIAL_ROOT_SLICE); | |
acb14d31 LP |
2118 | |
2119 | *e = 0; | |
2120 | ||
4ad49000 LP |
2121 | u = hashmap_get(m->cgroup_unit, p); |
2122 | if (u) | |
2123 | return u; | |
acb14d31 LP |
2124 | } |
2125 | } | |
2126 | ||
b3ac818b | 2127 | Unit *manager_get_unit_by_pid_cgroup(Manager *m, pid_t pid) { |
4ad49000 | 2128 | _cleanup_free_ char *cgroup = NULL; |
acb14d31 | 2129 | int r; |
8e274523 | 2130 | |
8c47c732 LP |
2131 | assert(m); |
2132 | ||
b3ac818b LP |
2133 | if (pid <= 0) |
2134 | return NULL; | |
2135 | ||
2136 | r = cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER, pid, &cgroup); | |
2137 | if (r < 0) | |
2138 | return NULL; | |
2139 | ||
2140 | return manager_get_unit_by_cgroup(m, cgroup); | |
2141 | } | |
2142 | ||
2143 | Unit *manager_get_unit_by_pid(Manager *m, pid_t pid) { | |
2144 | Unit *u; | |
2145 | ||
2146 | assert(m); | |
2147 | ||
efdb0237 | 2148 | if (pid <= 0) |
8c47c732 LP |
2149 | return NULL; |
2150 | ||
efdb0237 LP |
2151 | if (pid == 1) |
2152 | return hashmap_get(m->units, SPECIAL_INIT_SCOPE); | |
2153 | ||
fea72cc0 | 2154 | u = hashmap_get(m->watch_pids1, PID_TO_PTR(pid)); |
5fe8876b LP |
2155 | if (u) |
2156 | return u; | |
2157 | ||
fea72cc0 | 2158 | u = hashmap_get(m->watch_pids2, PID_TO_PTR(pid)); |
5fe8876b LP |
2159 | if (u) |
2160 | return u; | |
2161 | ||
b3ac818b | 2162 | return manager_get_unit_by_pid_cgroup(m, pid); |
6dde1f33 | 2163 | } |
4fbf50b3 | 2164 | |
4ad49000 LP |
2165 | int manager_notify_cgroup_empty(Manager *m, const char *cgroup) { |
2166 | Unit *u; | |
4fbf50b3 | 2167 | |
4ad49000 LP |
2168 | assert(m); |
2169 | assert(cgroup); | |
4fbf50b3 | 2170 | |
09e24654 LP |
2171 | /* Called on the legacy hierarchy whenever we get an explicit cgroup notification from the cgroup agent process |
2172 | * or from the --system instance */ | |
2173 | ||
d8fdc620 LP |
2174 | log_debug("Got cgroup empty notification for: %s", cgroup); |
2175 | ||
4ad49000 | 2176 | u = manager_get_unit_by_cgroup(m, cgroup); |
5ad096b3 LP |
2177 | if (!u) |
2178 | return 0; | |
b56c28c3 | 2179 | |
09e24654 LP |
2180 | unit_add_to_cgroup_empty_queue(u); |
2181 | return 1; | |
5ad096b3 LP |
2182 | } |
2183 | ||
2184 | int unit_get_memory_current(Unit *u, uint64_t *ret) { | |
2185 | _cleanup_free_ char *v = NULL; | |
2186 | int r; | |
2187 | ||
2188 | assert(u); | |
2189 | assert(ret); | |
2190 | ||
2e4025c0 | 2191 | if (!UNIT_CGROUP_BOOL(u, memory_accounting)) |
cf3b4be1 LP |
2192 | return -ENODATA; |
2193 | ||
5ad096b3 LP |
2194 | if (!u->cgroup_path) |
2195 | return -ENODATA; | |
2196 | ||
efdb0237 | 2197 | if ((u->cgroup_realized_mask & CGROUP_MASK_MEMORY) == 0) |
5ad096b3 LP |
2198 | return -ENODATA; |
2199 | ||
b4cccbc1 LP |
2200 | r = cg_all_unified(); |
2201 | if (r < 0) | |
2202 | return r; | |
2203 | if (r > 0) | |
efdb0237 | 2204 | r = cg_get_attribute("memory", u->cgroup_path, "memory.current", &v); |
b4cccbc1 LP |
2205 | else |
2206 | r = cg_get_attribute("memory", u->cgroup_path, "memory.usage_in_bytes", &v); | |
5ad096b3 LP |
2207 | if (r == -ENOENT) |
2208 | return -ENODATA; | |
2209 | if (r < 0) | |
2210 | return r; | |
2211 | ||
2212 | return safe_atou64(v, ret); | |
2213 | } | |
2214 | ||
03a7b521 LP |
2215 | int unit_get_tasks_current(Unit *u, uint64_t *ret) { |
2216 | _cleanup_free_ char *v = NULL; | |
2217 | int r; | |
2218 | ||
2219 | assert(u); | |
2220 | assert(ret); | |
2221 | ||
2e4025c0 | 2222 | if (!UNIT_CGROUP_BOOL(u, tasks_accounting)) |
cf3b4be1 LP |
2223 | return -ENODATA; |
2224 | ||
03a7b521 LP |
2225 | if (!u->cgroup_path) |
2226 | return -ENODATA; | |
2227 | ||
2228 | if ((u->cgroup_realized_mask & CGROUP_MASK_PIDS) == 0) | |
2229 | return -ENODATA; | |
2230 | ||
2231 | r = cg_get_attribute("pids", u->cgroup_path, "pids.current", &v); | |
2232 | if (r == -ENOENT) | |
2233 | return -ENODATA; | |
2234 | if (r < 0) | |
2235 | return r; | |
2236 | ||
2237 | return safe_atou64(v, ret); | |
2238 | } | |
2239 | ||
5ad096b3 LP |
2240 | static int unit_get_cpu_usage_raw(Unit *u, nsec_t *ret) { |
2241 | _cleanup_free_ char *v = NULL; | |
2242 | uint64_t ns; | |
2243 | int r; | |
2244 | ||
2245 | assert(u); | |
2246 | assert(ret); | |
2247 | ||
2248 | if (!u->cgroup_path) | |
2249 | return -ENODATA; | |
2250 | ||
b4cccbc1 LP |
2251 | r = cg_all_unified(); |
2252 | if (r < 0) | |
2253 | return r; | |
2254 | if (r > 0) { | |
66ebf6c0 TH |
2255 | const char *keys[] = { "usage_usec", NULL }; |
2256 | _cleanup_free_ char *val = NULL; | |
2257 | uint64_t us; | |
5ad096b3 | 2258 | |
66ebf6c0 TH |
2259 | if ((u->cgroup_realized_mask & CGROUP_MASK_CPU) == 0) |
2260 | return -ENODATA; | |
5ad096b3 | 2261 | |
66ebf6c0 TH |
2262 | r = cg_get_keyed_attribute("cpu", u->cgroup_path, "cpu.stat", keys, &val); |
2263 | if (r < 0) | |
2264 | return r; | |
2265 | ||
2266 | r = safe_atou64(val, &us); | |
2267 | if (r < 0) | |
2268 | return r; | |
2269 | ||
2270 | ns = us * NSEC_PER_USEC; | |
2271 | } else { | |
2272 | if ((u->cgroup_realized_mask & CGROUP_MASK_CPUACCT) == 0) | |
2273 | return -ENODATA; | |
2274 | ||
2275 | r = cg_get_attribute("cpuacct", u->cgroup_path, "cpuacct.usage", &v); | |
2276 | if (r == -ENOENT) | |
2277 | return -ENODATA; | |
2278 | if (r < 0) | |
2279 | return r; | |
2280 | ||
2281 | r = safe_atou64(v, &ns); | |
2282 | if (r < 0) | |
2283 | return r; | |
2284 | } | |
5ad096b3 LP |
2285 | |
2286 | *ret = ns; | |
2287 | return 0; | |
2288 | } | |
2289 | ||
2290 | int unit_get_cpu_usage(Unit *u, nsec_t *ret) { | |
2291 | nsec_t ns; | |
2292 | int r; | |
2293 | ||
fe700f46 LP |
2294 | assert(u); |
2295 | ||
2296 | /* Retrieve the current CPU usage counter. This will subtract the CPU counter taken when the unit was | |
2297 | * started. If the cgroup has been removed already, returns the last cached value. To cache the value, simply | |
2298 | * call this function with a NULL return value. */ | |
2299 | ||
2e4025c0 | 2300 | if (!UNIT_CGROUP_BOOL(u, cpu_accounting)) |
cf3b4be1 LP |
2301 | return -ENODATA; |
2302 | ||
5ad096b3 | 2303 | r = unit_get_cpu_usage_raw(u, &ns); |
fe700f46 LP |
2304 | if (r == -ENODATA && u->cpu_usage_last != NSEC_INFINITY) { |
2305 | /* If we can't get the CPU usage anymore (because the cgroup was already removed, for example), use our | |
2306 | * cached value. */ | |
2307 | ||
2308 | if (ret) | |
2309 | *ret = u->cpu_usage_last; | |
2310 | return 0; | |
2311 | } | |
5ad096b3 LP |
2312 | if (r < 0) |
2313 | return r; | |
2314 | ||
66ebf6c0 TH |
2315 | if (ns > u->cpu_usage_base) |
2316 | ns -= u->cpu_usage_base; | |
5ad096b3 LP |
2317 | else |
2318 | ns = 0; | |
2319 | ||
fe700f46 LP |
2320 | u->cpu_usage_last = ns; |
2321 | if (ret) | |
2322 | *ret = ns; | |
2323 | ||
5ad096b3 LP |
2324 | return 0; |
2325 | } | |
2326 | ||
906c06f6 DM |
2327 | int unit_get_ip_accounting( |
2328 | Unit *u, | |
2329 | CGroupIPAccountingMetric metric, | |
2330 | uint64_t *ret) { | |
2331 | ||
6b659ed8 | 2332 | uint64_t value; |
906c06f6 DM |
2333 | int fd, r; |
2334 | ||
2335 | assert(u); | |
2336 | assert(metric >= 0); | |
2337 | assert(metric < _CGROUP_IP_ACCOUNTING_METRIC_MAX); | |
2338 | assert(ret); | |
2339 | ||
cf3b4be1 LP |
2340 | /* IP accounting is currently not recursive, and hence we refuse to return any data for slice nodes. Slices are |
2341 | * inner cgroup nodes and hence have no processes directly attached, hence their counters would be zero | |
2342 | * anyway. And if we block this now we can later open this up, if the kernel learns recursive BPF cgroup | |
2343 | * filters. */ | |
2344 | if (u->type == UNIT_SLICE) | |
2345 | return -ENODATA; | |
2346 | ||
2e4025c0 | 2347 | if (!UNIT_CGROUP_BOOL(u, ip_accounting)) |
cf3b4be1 LP |
2348 | return -ENODATA; |
2349 | ||
906c06f6 DM |
2350 | fd = IN_SET(metric, CGROUP_IP_INGRESS_BYTES, CGROUP_IP_INGRESS_PACKETS) ? |
2351 | u->ip_accounting_ingress_map_fd : | |
2352 | u->ip_accounting_egress_map_fd; | |
2353 | ||
2354 | if (fd < 0) | |
2355 | return -ENODATA; | |
2356 | ||
2357 | if (IN_SET(metric, CGROUP_IP_INGRESS_BYTES, CGROUP_IP_EGRESS_BYTES)) | |
6b659ed8 | 2358 | r = bpf_firewall_read_accounting(fd, &value, NULL); |
906c06f6 | 2359 | else |
6b659ed8 LP |
2360 | r = bpf_firewall_read_accounting(fd, NULL, &value); |
2361 | if (r < 0) | |
2362 | return r; | |
2363 | ||
2364 | /* Add in additional metrics from a previous runtime. Note that when reexecing/reloading the daemon we compile | |
2365 | * all BPF programs and maps anew, but serialize the old counters. When deserializing we store them in the | |
2366 | * ip_accounting_extra[] field, and add them in here transparently. */ | |
2367 | ||
2368 | *ret = value + u->ip_accounting_extra[metric]; | |
906c06f6 DM |
2369 | |
2370 | return r; | |
2371 | } | |
2372 | ||
2373 | int unit_reset_cpu_accounting(Unit *u) { | |
5ad096b3 LP |
2374 | nsec_t ns; |
2375 | int r; | |
2376 | ||
2377 | assert(u); | |
2378 | ||
fe700f46 LP |
2379 | u->cpu_usage_last = NSEC_INFINITY; |
2380 | ||
5ad096b3 LP |
2381 | r = unit_get_cpu_usage_raw(u, &ns); |
2382 | if (r < 0) { | |
66ebf6c0 | 2383 | u->cpu_usage_base = 0; |
5ad096b3 | 2384 | return r; |
b56c28c3 | 2385 | } |
2633eb83 | 2386 | |
66ebf6c0 | 2387 | u->cpu_usage_base = ns; |
4ad49000 | 2388 | return 0; |
4fbf50b3 LP |
2389 | } |
2390 | ||
906c06f6 DM |
2391 | int unit_reset_ip_accounting(Unit *u) { |
2392 | int r = 0, q = 0; | |
2393 | ||
2394 | assert(u); | |
2395 | ||
2396 | if (u->ip_accounting_ingress_map_fd >= 0) | |
2397 | r = bpf_firewall_reset_accounting(u->ip_accounting_ingress_map_fd); | |
2398 | ||
2399 | if (u->ip_accounting_egress_map_fd >= 0) | |
2400 | q = bpf_firewall_reset_accounting(u->ip_accounting_egress_map_fd); | |
2401 | ||
6b659ed8 LP |
2402 | zero(u->ip_accounting_extra); |
2403 | ||
906c06f6 DM |
2404 | return r < 0 ? r : q; |
2405 | } | |
2406 | ||
e7ab4d1a LP |
2407 | void unit_invalidate_cgroup(Unit *u, CGroupMask m) { |
2408 | assert(u); | |
2409 | ||
2410 | if (!UNIT_HAS_CGROUP_CONTEXT(u)) | |
2411 | return; | |
2412 | ||
2413 | if (m == 0) | |
2414 | return; | |
2415 | ||
538b4852 TH |
2416 | /* always invalidate compat pairs together */ |
2417 | if (m & (CGROUP_MASK_IO | CGROUP_MASK_BLKIO)) | |
2418 | m |= CGROUP_MASK_IO | CGROUP_MASK_BLKIO; | |
2419 | ||
7cce4fb7 LP |
2420 | if (m & (CGROUP_MASK_CPU | CGROUP_MASK_CPUACCT)) |
2421 | m |= CGROUP_MASK_CPU | CGROUP_MASK_CPUACCT; | |
2422 | ||
e7ab4d1a LP |
2423 | if ((u->cgroup_realized_mask & m) == 0) |
2424 | return; | |
2425 | ||
2426 | u->cgroup_realized_mask &= ~m; | |
91a6073e | 2427 | unit_add_to_cgroup_realize_queue(u); |
e7ab4d1a LP |
2428 | } |
2429 | ||
906c06f6 DM |
2430 | void unit_invalidate_cgroup_bpf(Unit *u) { |
2431 | assert(u); | |
2432 | ||
2433 | if (!UNIT_HAS_CGROUP_CONTEXT(u)) | |
2434 | return; | |
2435 | ||
2436 | if (u->cgroup_bpf_state == UNIT_CGROUP_BPF_INVALIDATED) | |
2437 | return; | |
2438 | ||
2439 | u->cgroup_bpf_state = UNIT_CGROUP_BPF_INVALIDATED; | |
91a6073e | 2440 | unit_add_to_cgroup_realize_queue(u); |
906c06f6 DM |
2441 | |
2442 | /* If we are a slice unit, we also need to put compile a new BPF program for all our children, as the IP access | |
2443 | * list of our children includes our own. */ | |
2444 | if (u->type == UNIT_SLICE) { | |
2445 | Unit *member; | |
2446 | Iterator i; | |
eef85c4a | 2447 | void *v; |
906c06f6 | 2448 | |
eef85c4a | 2449 | HASHMAP_FOREACH_KEY(v, member, u->dependencies[UNIT_BEFORE], i) { |
906c06f6 DM |
2450 | if (member == u) |
2451 | continue; | |
2452 | ||
2453 | if (UNIT_DEREF(member->slice) != u) | |
2454 | continue; | |
2455 | ||
2456 | unit_invalidate_cgroup_bpf(member); | |
2457 | } | |
2458 | } | |
2459 | } | |
2460 | ||
e7ab4d1a LP |
2461 | void manager_invalidate_startup_units(Manager *m) { |
2462 | Iterator i; | |
2463 | Unit *u; | |
2464 | ||
2465 | assert(m); | |
2466 | ||
2467 | SET_FOREACH(u, m->startup_units, i) | |
13c31542 | 2468 | unit_invalidate_cgroup(u, CGROUP_MASK_CPU|CGROUP_MASK_IO|CGROUP_MASK_BLKIO); |
e7ab4d1a LP |
2469 | } |
2470 | ||
4ad49000 LP |
2471 | static const char* const cgroup_device_policy_table[_CGROUP_DEVICE_POLICY_MAX] = { |
2472 | [CGROUP_AUTO] = "auto", | |
2473 | [CGROUP_CLOSED] = "closed", | |
2474 | [CGROUP_STRICT] = "strict", | |
2475 | }; | |
4fbf50b3 | 2476 | |
4ad49000 | 2477 | DEFINE_STRING_TABLE_LOOKUP(cgroup_device_policy, CGroupDevicePolicy); |