]>
Commit | Line | Data |
---|---|---|
53e1b683 | 1 | /* SPDX-License-Identifier: LGPL-2.1+ */ |
8e274523 LP |
2 | /*** |
3 | This file is part of systemd. | |
4 | ||
4ad49000 | 5 | Copyright 2013 Lennart Poettering |
8e274523 LP |
6 | |
7 | systemd is free software; you can redistribute it and/or modify it | |
5430f7f2 LP |
8 | under the terms of the GNU Lesser General Public License as published by |
9 | the Free Software Foundation; either version 2.1 of the License, or | |
8e274523 LP |
10 | (at your option) any later version. |
11 | ||
12 | systemd is distributed in the hope that it will be useful, but | |
13 | WITHOUT ANY WARRANTY; without even the implied warranty of | |
14 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
5430f7f2 | 15 | Lesser General Public License for more details. |
8e274523 | 16 | |
5430f7f2 | 17 | You should have received a copy of the GNU Lesser General Public License |
8e274523 LP |
18 | along with systemd; If not, see <http://www.gnu.org/licenses/>. |
19 | ***/ | |
20 | ||
c6c18be3 | 21 | #include <fcntl.h> |
e41969e3 | 22 | #include <fnmatch.h> |
8c6db833 | 23 | |
b5efdb8a | 24 | #include "alloc-util.h" |
18c528e9 | 25 | #include "blockdev-util.h" |
906c06f6 | 26 | #include "bpf-firewall.h" |
6592b975 | 27 | #include "bus-error.h" |
03a7b521 | 28 | #include "cgroup-util.h" |
3ffd4af2 LP |
29 | #include "cgroup.h" |
30 | #include "fd-util.h" | |
0d39fa9c | 31 | #include "fileio.h" |
77601719 | 32 | #include "fs-util.h" |
6bedfcbb | 33 | #include "parse-util.h" |
9eb977db | 34 | #include "path-util.h" |
03a7b521 | 35 | #include "process-util.h" |
c36a69f4 | 36 | #include "procfs-util.h" |
9444b1f2 | 37 | #include "special.h" |
906c06f6 | 38 | #include "stdio-util.h" |
8b43440b | 39 | #include "string-table.h" |
07630cea | 40 | #include "string-util.h" |
8e274523 | 41 | |
9a054909 LP |
42 | #define CGROUP_CPU_QUOTA_PERIOD_USEC ((usec_t) 100 * USEC_PER_MSEC) |
43 | ||
f3725e64 LP |
44 | bool unit_has_root_cgroup(Unit *u) { |
45 | assert(u); | |
46 | ||
47 | /* Returns whether this unit manages the root cgroup. Note that this is different from being named "-.slice", | |
48 | * as inside of containers the root slice won't be identical to the root cgroup. */ | |
49 | ||
50 | if (!u->cgroup_path) | |
51 | return false; | |
52 | ||
53 | return isempty(u->cgroup_path) || path_equal(u->cgroup_path, "/"); | |
54 | } | |
55 | ||
2b40998d | 56 | static void cgroup_compat_warn(void) { |
128fadc9 TH |
57 | static bool cgroup_compat_warned = false; |
58 | ||
59 | if (cgroup_compat_warned) | |
60 | return; | |
61 | ||
62 | log_warning("cgroup compatibility translation between legacy and unified hierarchy settings activated. See cgroup-compat debug messages for details."); | |
63 | cgroup_compat_warned = true; | |
64 | } | |
65 | ||
66 | #define log_cgroup_compat(unit, fmt, ...) do { \ | |
67 | cgroup_compat_warn(); \ | |
68 | log_unit_debug(unit, "cgroup-compat: " fmt, ##__VA_ARGS__); \ | |
2b40998d | 69 | } while (false) |
128fadc9 | 70 | |
4ad49000 LP |
71 | void cgroup_context_init(CGroupContext *c) { |
72 | assert(c); | |
73 | ||
74 | /* Initialize everything to the kernel defaults, assuming the | |
75 | * structure is preinitialized to 0 */ | |
76 | ||
66ebf6c0 TH |
77 | c->cpu_weight = CGROUP_WEIGHT_INVALID; |
78 | c->startup_cpu_weight = CGROUP_WEIGHT_INVALID; | |
79 | c->cpu_quota_per_sec_usec = USEC_INFINITY; | |
80 | ||
d53d9474 LP |
81 | c->cpu_shares = CGROUP_CPU_SHARES_INVALID; |
82 | c->startup_cpu_shares = CGROUP_CPU_SHARES_INVALID; | |
d53d9474 | 83 | |
da4d897e TH |
84 | c->memory_high = CGROUP_LIMIT_MAX; |
85 | c->memory_max = CGROUP_LIMIT_MAX; | |
96e131ea | 86 | c->memory_swap_max = CGROUP_LIMIT_MAX; |
da4d897e TH |
87 | |
88 | c->memory_limit = CGROUP_LIMIT_MAX; | |
b2f8b02e | 89 | |
13c31542 TH |
90 | c->io_weight = CGROUP_WEIGHT_INVALID; |
91 | c->startup_io_weight = CGROUP_WEIGHT_INVALID; | |
92 | ||
d53d9474 LP |
93 | c->blockio_weight = CGROUP_BLKIO_WEIGHT_INVALID; |
94 | c->startup_blockio_weight = CGROUP_BLKIO_WEIGHT_INVALID; | |
95 | ||
96 | c->tasks_max = (uint64_t) -1; | |
4ad49000 | 97 | } |
8e274523 | 98 | |
4ad49000 LP |
99 | void cgroup_context_free_device_allow(CGroupContext *c, CGroupDeviceAllow *a) { |
100 | assert(c); | |
101 | assert(a); | |
102 | ||
71fda00f | 103 | LIST_REMOVE(device_allow, c->device_allow, a); |
4ad49000 LP |
104 | free(a->path); |
105 | free(a); | |
106 | } | |
107 | ||
13c31542 TH |
108 | void cgroup_context_free_io_device_weight(CGroupContext *c, CGroupIODeviceWeight *w) { |
109 | assert(c); | |
110 | assert(w); | |
111 | ||
112 | LIST_REMOVE(device_weights, c->io_device_weights, w); | |
113 | free(w->path); | |
114 | free(w); | |
115 | } | |
116 | ||
117 | void cgroup_context_free_io_device_limit(CGroupContext *c, CGroupIODeviceLimit *l) { | |
118 | assert(c); | |
119 | assert(l); | |
120 | ||
121 | LIST_REMOVE(device_limits, c->io_device_limits, l); | |
122 | free(l->path); | |
123 | free(l); | |
124 | } | |
125 | ||
4ad49000 LP |
126 | void cgroup_context_free_blockio_device_weight(CGroupContext *c, CGroupBlockIODeviceWeight *w) { |
127 | assert(c); | |
128 | assert(w); | |
129 | ||
71fda00f | 130 | LIST_REMOVE(device_weights, c->blockio_device_weights, w); |
4ad49000 LP |
131 | free(w->path); |
132 | free(w); | |
133 | } | |
134 | ||
135 | void cgroup_context_free_blockio_device_bandwidth(CGroupContext *c, CGroupBlockIODeviceBandwidth *b) { | |
136 | assert(c); | |
8e274523 | 137 | assert(b); |
8e274523 | 138 | |
71fda00f | 139 | LIST_REMOVE(device_bandwidths, c->blockio_device_bandwidths, b); |
4ad49000 LP |
140 | free(b->path); |
141 | free(b); | |
142 | } | |
143 | ||
144 | void cgroup_context_done(CGroupContext *c) { | |
145 | assert(c); | |
146 | ||
13c31542 TH |
147 | while (c->io_device_weights) |
148 | cgroup_context_free_io_device_weight(c, c->io_device_weights); | |
149 | ||
150 | while (c->io_device_limits) | |
151 | cgroup_context_free_io_device_limit(c, c->io_device_limits); | |
152 | ||
4ad49000 LP |
153 | while (c->blockio_device_weights) |
154 | cgroup_context_free_blockio_device_weight(c, c->blockio_device_weights); | |
155 | ||
156 | while (c->blockio_device_bandwidths) | |
157 | cgroup_context_free_blockio_device_bandwidth(c, c->blockio_device_bandwidths); | |
158 | ||
159 | while (c->device_allow) | |
160 | cgroup_context_free_device_allow(c, c->device_allow); | |
6a48d82f DM |
161 | |
162 | c->ip_address_allow = ip_address_access_free_all(c->ip_address_allow); | |
163 | c->ip_address_deny = ip_address_access_free_all(c->ip_address_deny); | |
4ad49000 LP |
164 | } |
165 | ||
166 | void cgroup_context_dump(CGroupContext *c, FILE* f, const char *prefix) { | |
13c31542 TH |
167 | CGroupIODeviceLimit *il; |
168 | CGroupIODeviceWeight *iw; | |
4ad49000 LP |
169 | CGroupBlockIODeviceBandwidth *b; |
170 | CGroupBlockIODeviceWeight *w; | |
171 | CGroupDeviceAllow *a; | |
c21c9906 | 172 | IPAddressAccessItem *iaai; |
9a054909 | 173 | char u[FORMAT_TIMESPAN_MAX]; |
4ad49000 LP |
174 | |
175 | assert(c); | |
176 | assert(f); | |
177 | ||
178 | prefix = strempty(prefix); | |
179 | ||
180 | fprintf(f, | |
181 | "%sCPUAccounting=%s\n" | |
13c31542 | 182 | "%sIOAccounting=%s\n" |
4ad49000 LP |
183 | "%sBlockIOAccounting=%s\n" |
184 | "%sMemoryAccounting=%s\n" | |
d53d9474 | 185 | "%sTasksAccounting=%s\n" |
c21c9906 | 186 | "%sIPAccounting=%s\n" |
66ebf6c0 TH |
187 | "%sCPUWeight=%" PRIu64 "\n" |
188 | "%sStartupCPUWeight=%" PRIu64 "\n" | |
d53d9474 LP |
189 | "%sCPUShares=%" PRIu64 "\n" |
190 | "%sStartupCPUShares=%" PRIu64 "\n" | |
b2f8b02e | 191 | "%sCPUQuotaPerSecSec=%s\n" |
13c31542 TH |
192 | "%sIOWeight=%" PRIu64 "\n" |
193 | "%sStartupIOWeight=%" PRIu64 "\n" | |
d53d9474 LP |
194 | "%sBlockIOWeight=%" PRIu64 "\n" |
195 | "%sStartupBlockIOWeight=%" PRIu64 "\n" | |
da4d897e TH |
196 | "%sMemoryLow=%" PRIu64 "\n" |
197 | "%sMemoryHigh=%" PRIu64 "\n" | |
198 | "%sMemoryMax=%" PRIu64 "\n" | |
96e131ea | 199 | "%sMemorySwapMax=%" PRIu64 "\n" |
4ad49000 | 200 | "%sMemoryLimit=%" PRIu64 "\n" |
03a7b521 | 201 | "%sTasksMax=%" PRIu64 "\n" |
a931ad47 LP |
202 | "%sDevicePolicy=%s\n" |
203 | "%sDelegate=%s\n", | |
4ad49000 | 204 | prefix, yes_no(c->cpu_accounting), |
13c31542 | 205 | prefix, yes_no(c->io_accounting), |
4ad49000 LP |
206 | prefix, yes_no(c->blockio_accounting), |
207 | prefix, yes_no(c->memory_accounting), | |
d53d9474 | 208 | prefix, yes_no(c->tasks_accounting), |
c21c9906 | 209 | prefix, yes_no(c->ip_accounting), |
66ebf6c0 TH |
210 | prefix, c->cpu_weight, |
211 | prefix, c->startup_cpu_weight, | |
4ad49000 | 212 | prefix, c->cpu_shares, |
95ae05c0 | 213 | prefix, c->startup_cpu_shares, |
b1d6dcf5 | 214 | prefix, format_timespan(u, sizeof(u), c->cpu_quota_per_sec_usec, 1), |
13c31542 TH |
215 | prefix, c->io_weight, |
216 | prefix, c->startup_io_weight, | |
4ad49000 | 217 | prefix, c->blockio_weight, |
95ae05c0 | 218 | prefix, c->startup_blockio_weight, |
da4d897e TH |
219 | prefix, c->memory_low, |
220 | prefix, c->memory_high, | |
221 | prefix, c->memory_max, | |
96e131ea | 222 | prefix, c->memory_swap_max, |
4ad49000 | 223 | prefix, c->memory_limit, |
03a7b521 | 224 | prefix, c->tasks_max, |
a931ad47 LP |
225 | prefix, cgroup_device_policy_to_string(c->device_policy), |
226 | prefix, yes_no(c->delegate)); | |
4ad49000 | 227 | |
02638280 LP |
228 | if (c->delegate) { |
229 | _cleanup_free_ char *t = NULL; | |
230 | ||
231 | (void) cg_mask_to_string(c->delegate_controllers, &t); | |
232 | ||
47a78d41 | 233 | fprintf(f, "%sDelegateControllers=%s\n", |
02638280 LP |
234 | prefix, |
235 | strempty(t)); | |
236 | } | |
237 | ||
4ad49000 LP |
238 | LIST_FOREACH(device_allow, a, c->device_allow) |
239 | fprintf(f, | |
240 | "%sDeviceAllow=%s %s%s%s\n", | |
241 | prefix, | |
242 | a->path, | |
243 | a->r ? "r" : "", a->w ? "w" : "", a->m ? "m" : ""); | |
244 | ||
13c31542 TH |
245 | LIST_FOREACH(device_weights, iw, c->io_device_weights) |
246 | fprintf(f, | |
247 | "%sIODeviceWeight=%s %" PRIu64, | |
248 | prefix, | |
249 | iw->path, | |
250 | iw->weight); | |
251 | ||
252 | LIST_FOREACH(device_limits, il, c->io_device_limits) { | |
253 | char buf[FORMAT_BYTES_MAX]; | |
9be57249 TH |
254 | CGroupIOLimitType type; |
255 | ||
256 | for (type = 0; type < _CGROUP_IO_LIMIT_TYPE_MAX; type++) | |
257 | if (il->limits[type] != cgroup_io_limit_defaults[type]) | |
258 | fprintf(f, | |
259 | "%s%s=%s %s\n", | |
260 | prefix, | |
261 | cgroup_io_limit_type_to_string(type), | |
262 | il->path, | |
263 | format_bytes(buf, sizeof(buf), il->limits[type])); | |
13c31542 TH |
264 | } |
265 | ||
4ad49000 LP |
266 | LIST_FOREACH(device_weights, w, c->blockio_device_weights) |
267 | fprintf(f, | |
d53d9474 | 268 | "%sBlockIODeviceWeight=%s %" PRIu64, |
4ad49000 LP |
269 | prefix, |
270 | w->path, | |
271 | w->weight); | |
272 | ||
273 | LIST_FOREACH(device_bandwidths, b, c->blockio_device_bandwidths) { | |
274 | char buf[FORMAT_BYTES_MAX]; | |
275 | ||
979d0311 TH |
276 | if (b->rbps != CGROUP_LIMIT_MAX) |
277 | fprintf(f, | |
278 | "%sBlockIOReadBandwidth=%s %s\n", | |
279 | prefix, | |
280 | b->path, | |
281 | format_bytes(buf, sizeof(buf), b->rbps)); | |
282 | if (b->wbps != CGROUP_LIMIT_MAX) | |
283 | fprintf(f, | |
284 | "%sBlockIOWriteBandwidth=%s %s\n", | |
285 | prefix, | |
286 | b->path, | |
287 | format_bytes(buf, sizeof(buf), b->wbps)); | |
4ad49000 | 288 | } |
c21c9906 LP |
289 | |
290 | LIST_FOREACH(items, iaai, c->ip_address_allow) { | |
291 | _cleanup_free_ char *k = NULL; | |
292 | ||
293 | (void) in_addr_to_string(iaai->family, &iaai->address, &k); | |
294 | fprintf(f, "%sIPAddressAllow=%s/%u\n", prefix, strnull(k), iaai->prefixlen); | |
295 | } | |
296 | ||
297 | LIST_FOREACH(items, iaai, c->ip_address_deny) { | |
298 | _cleanup_free_ char *k = NULL; | |
299 | ||
300 | (void) in_addr_to_string(iaai->family, &iaai->address, &k); | |
301 | fprintf(f, "%sIPAddressDeny=%s/%u\n", prefix, strnull(k), iaai->prefixlen); | |
302 | } | |
4ad49000 LP |
303 | } |
304 | ||
13c31542 | 305 | static int lookup_block_device(const char *p, dev_t *dev) { |
4ad49000 LP |
306 | struct stat st; |
307 | int r; | |
308 | ||
309 | assert(p); | |
310 | assert(dev); | |
311 | ||
312 | r = stat(p, &st); | |
4a62c710 MS |
313 | if (r < 0) |
314 | return log_warning_errno(errno, "Couldn't stat device %s: %m", p); | |
8e274523 | 315 | |
4ad49000 LP |
316 | if (S_ISBLK(st.st_mode)) |
317 | *dev = st.st_rdev; | |
318 | else if (major(st.st_dev) != 0) { | |
319 | /* If this is not a device node then find the block | |
320 | * device this file is stored on */ | |
321 | *dev = st.st_dev; | |
322 | ||
323 | /* If this is a partition, try to get the originating | |
324 | * block device */ | |
18c528e9 | 325 | (void) block_get_whole_disk(*dev, dev); |
4ad49000 LP |
326 | } else { |
327 | log_warning("%s is not a block device and file system block device cannot be determined or is not local.", p); | |
328 | return -ENODEV; | |
329 | } | |
8e274523 | 330 | |
8e274523 | 331 | return 0; |
8e274523 LP |
332 | } |
333 | ||
4ad49000 LP |
334 | static int whitelist_device(const char *path, const char *node, const char *acc) { |
335 | char buf[2+DECIMAL_STR_MAX(dev_t)*2+2+4]; | |
336 | struct stat st; | |
b200489b | 337 | bool ignore_notfound; |
8c6db833 | 338 | int r; |
8e274523 | 339 | |
4ad49000 LP |
340 | assert(path); |
341 | assert(acc); | |
8e274523 | 342 | |
b200489b DR |
343 | if (node[0] == '-') { |
344 | /* Non-existent paths starting with "-" must be silently ignored */ | |
345 | node++; | |
346 | ignore_notfound = true; | |
347 | } else | |
348 | ignore_notfound = false; | |
349 | ||
4ad49000 | 350 | if (stat(node, &st) < 0) { |
b200489b | 351 | if (errno == ENOENT && ignore_notfound) |
e7330dfe DP |
352 | return 0; |
353 | ||
354 | return log_warning_errno(errno, "Couldn't stat device %s: %m", node); | |
4ad49000 LP |
355 | } |
356 | ||
357 | if (!S_ISCHR(st.st_mode) && !S_ISBLK(st.st_mode)) { | |
358 | log_warning("%s is not a device.", node); | |
359 | return -ENODEV; | |
360 | } | |
361 | ||
362 | sprintf(buf, | |
363 | "%c %u:%u %s", | |
364 | S_ISCHR(st.st_mode) ? 'c' : 'b', | |
365 | major(st.st_rdev), minor(st.st_rdev), | |
366 | acc); | |
367 | ||
368 | r = cg_set_attribute("devices", path, "devices.allow", buf); | |
1aeab12b | 369 | if (r < 0) |
077ba06e | 370 | log_full_errno(IN_SET(r, -ENOENT, -EROFS, -EINVAL, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, |
714e2e1d | 371 | "Failed to set devices.allow on %s: %m", path); |
4ad49000 LP |
372 | |
373 | return r; | |
8e274523 LP |
374 | } |
375 | ||
90060676 LP |
376 | static int whitelist_major(const char *path, const char *name, char type, const char *acc) { |
377 | _cleanup_fclose_ FILE *f = NULL; | |
378 | char line[LINE_MAX]; | |
379 | bool good = false; | |
380 | int r; | |
381 | ||
382 | assert(path); | |
383 | assert(acc); | |
4c701096 | 384 | assert(IN_SET(type, 'b', 'c')); |
90060676 LP |
385 | |
386 | f = fopen("/proc/devices", "re"); | |
4a62c710 MS |
387 | if (!f) |
388 | return log_warning_errno(errno, "Cannot open /proc/devices to resolve %s (%c): %m", name, type); | |
90060676 LP |
389 | |
390 | FOREACH_LINE(line, f, goto fail) { | |
391 | char buf[2+DECIMAL_STR_MAX(unsigned)+3+4], *p, *w; | |
392 | unsigned maj; | |
393 | ||
394 | truncate_nl(line); | |
395 | ||
396 | if (type == 'c' && streq(line, "Character devices:")) { | |
397 | good = true; | |
398 | continue; | |
399 | } | |
400 | ||
401 | if (type == 'b' && streq(line, "Block devices:")) { | |
402 | good = true; | |
403 | continue; | |
404 | } | |
405 | ||
406 | if (isempty(line)) { | |
407 | good = false; | |
408 | continue; | |
409 | } | |
410 | ||
411 | if (!good) | |
412 | continue; | |
413 | ||
414 | p = strstrip(line); | |
415 | ||
416 | w = strpbrk(p, WHITESPACE); | |
417 | if (!w) | |
418 | continue; | |
419 | *w = 0; | |
420 | ||
421 | r = safe_atou(p, &maj); | |
422 | if (r < 0) | |
423 | continue; | |
424 | if (maj <= 0) | |
425 | continue; | |
426 | ||
427 | w++; | |
428 | w += strspn(w, WHITESPACE); | |
e41969e3 LP |
429 | |
430 | if (fnmatch(name, w, 0) != 0) | |
90060676 LP |
431 | continue; |
432 | ||
433 | sprintf(buf, | |
434 | "%c %u:* %s", | |
435 | type, | |
436 | maj, | |
437 | acc); | |
438 | ||
439 | r = cg_set_attribute("devices", path, "devices.allow", buf); | |
1aeab12b | 440 | if (r < 0) |
077ba06e | 441 | log_full_errno(IN_SET(r, -ENOENT, -EROFS, -EINVAL, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, |
714e2e1d | 442 | "Failed to set devices.allow on %s: %m", path); |
90060676 LP |
443 | } |
444 | ||
445 | return 0; | |
446 | ||
447 | fail: | |
25f027c5 | 448 | return log_warning_errno(errno, "Failed to read /proc/devices: %m"); |
90060676 LP |
449 | } |
450 | ||
66ebf6c0 TH |
451 | static bool cgroup_context_has_cpu_weight(CGroupContext *c) { |
452 | return c->cpu_weight != CGROUP_WEIGHT_INVALID || | |
453 | c->startup_cpu_weight != CGROUP_WEIGHT_INVALID; | |
454 | } | |
455 | ||
456 | static bool cgroup_context_has_cpu_shares(CGroupContext *c) { | |
457 | return c->cpu_shares != CGROUP_CPU_SHARES_INVALID || | |
458 | c->startup_cpu_shares != CGROUP_CPU_SHARES_INVALID; | |
459 | } | |
460 | ||
461 | static uint64_t cgroup_context_cpu_weight(CGroupContext *c, ManagerState state) { | |
462 | if (IN_SET(state, MANAGER_STARTING, MANAGER_INITIALIZING) && | |
463 | c->startup_cpu_weight != CGROUP_WEIGHT_INVALID) | |
464 | return c->startup_cpu_weight; | |
465 | else if (c->cpu_weight != CGROUP_WEIGHT_INVALID) | |
466 | return c->cpu_weight; | |
467 | else | |
468 | return CGROUP_WEIGHT_DEFAULT; | |
469 | } | |
470 | ||
471 | static uint64_t cgroup_context_cpu_shares(CGroupContext *c, ManagerState state) { | |
472 | if (IN_SET(state, MANAGER_STARTING, MANAGER_INITIALIZING) && | |
473 | c->startup_cpu_shares != CGROUP_CPU_SHARES_INVALID) | |
474 | return c->startup_cpu_shares; | |
475 | else if (c->cpu_shares != CGROUP_CPU_SHARES_INVALID) | |
476 | return c->cpu_shares; | |
477 | else | |
478 | return CGROUP_CPU_SHARES_DEFAULT; | |
479 | } | |
480 | ||
481 | static void cgroup_apply_unified_cpu_config(Unit *u, uint64_t weight, uint64_t quota) { | |
482 | char buf[MAX(DECIMAL_STR_MAX(uint64_t) + 1, (DECIMAL_STR_MAX(usec_t) + 1) * 2)]; | |
483 | int r; | |
484 | ||
485 | xsprintf(buf, "%" PRIu64 "\n", weight); | |
486 | r = cg_set_attribute("cpu", u->cgroup_path, "cpu.weight", buf); | |
487 | if (r < 0) | |
488 | log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, | |
489 | "Failed to set cpu.weight: %m"); | |
490 | ||
491 | if (quota != USEC_INFINITY) | |
492 | xsprintf(buf, USEC_FMT " " USEC_FMT "\n", | |
493 | quota * CGROUP_CPU_QUOTA_PERIOD_USEC / USEC_PER_SEC, CGROUP_CPU_QUOTA_PERIOD_USEC); | |
494 | else | |
495 | xsprintf(buf, "max " USEC_FMT "\n", CGROUP_CPU_QUOTA_PERIOD_USEC); | |
496 | ||
497 | r = cg_set_attribute("cpu", u->cgroup_path, "cpu.max", buf); | |
498 | ||
499 | if (r < 0) | |
500 | log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, | |
501 | "Failed to set cpu.max: %m"); | |
502 | } | |
503 | ||
504 | static void cgroup_apply_legacy_cpu_config(Unit *u, uint64_t shares, uint64_t quota) { | |
505 | char buf[MAX(DECIMAL_STR_MAX(uint64_t), DECIMAL_STR_MAX(usec_t)) + 1]; | |
506 | int r; | |
507 | ||
508 | xsprintf(buf, "%" PRIu64 "\n", shares); | |
509 | r = cg_set_attribute("cpu", u->cgroup_path, "cpu.shares", buf); | |
510 | if (r < 0) | |
511 | log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, | |
512 | "Failed to set cpu.shares: %m"); | |
513 | ||
514 | xsprintf(buf, USEC_FMT "\n", CGROUP_CPU_QUOTA_PERIOD_USEC); | |
515 | r = cg_set_attribute("cpu", u->cgroup_path, "cpu.cfs_period_us", buf); | |
516 | if (r < 0) | |
517 | log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, | |
518 | "Failed to set cpu.cfs_period_us: %m"); | |
519 | ||
520 | if (quota != USEC_INFINITY) { | |
521 | xsprintf(buf, USEC_FMT "\n", quota * CGROUP_CPU_QUOTA_PERIOD_USEC / USEC_PER_SEC); | |
522 | r = cg_set_attribute("cpu", u->cgroup_path, "cpu.cfs_quota_us", buf); | |
523 | } else | |
524 | r = cg_set_attribute("cpu", u->cgroup_path, "cpu.cfs_quota_us", "-1"); | |
525 | if (r < 0) | |
526 | log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, | |
527 | "Failed to set cpu.cfs_quota_us: %m"); | |
528 | } | |
529 | ||
530 | static uint64_t cgroup_cpu_shares_to_weight(uint64_t shares) { | |
531 | return CLAMP(shares * CGROUP_WEIGHT_DEFAULT / CGROUP_CPU_SHARES_DEFAULT, | |
532 | CGROUP_WEIGHT_MIN, CGROUP_WEIGHT_MAX); | |
533 | } | |
534 | ||
535 | static uint64_t cgroup_cpu_weight_to_shares(uint64_t weight) { | |
536 | return CLAMP(weight * CGROUP_CPU_SHARES_DEFAULT / CGROUP_WEIGHT_DEFAULT, | |
537 | CGROUP_CPU_SHARES_MIN, CGROUP_CPU_SHARES_MAX); | |
538 | } | |
539 | ||
508c45da | 540 | static bool cgroup_context_has_io_config(CGroupContext *c) { |
538b4852 TH |
541 | return c->io_accounting || |
542 | c->io_weight != CGROUP_WEIGHT_INVALID || | |
543 | c->startup_io_weight != CGROUP_WEIGHT_INVALID || | |
544 | c->io_device_weights || | |
545 | c->io_device_limits; | |
546 | } | |
547 | ||
508c45da | 548 | static bool cgroup_context_has_blockio_config(CGroupContext *c) { |
538b4852 TH |
549 | return c->blockio_accounting || |
550 | c->blockio_weight != CGROUP_BLKIO_WEIGHT_INVALID || | |
551 | c->startup_blockio_weight != CGROUP_BLKIO_WEIGHT_INVALID || | |
552 | c->blockio_device_weights || | |
553 | c->blockio_device_bandwidths; | |
554 | } | |
555 | ||
508c45da | 556 | static uint64_t cgroup_context_io_weight(CGroupContext *c, ManagerState state) { |
64faf04c TH |
557 | if (IN_SET(state, MANAGER_STARTING, MANAGER_INITIALIZING) && |
558 | c->startup_io_weight != CGROUP_WEIGHT_INVALID) | |
559 | return c->startup_io_weight; | |
560 | else if (c->io_weight != CGROUP_WEIGHT_INVALID) | |
561 | return c->io_weight; | |
562 | else | |
563 | return CGROUP_WEIGHT_DEFAULT; | |
564 | } | |
565 | ||
508c45da | 566 | static uint64_t cgroup_context_blkio_weight(CGroupContext *c, ManagerState state) { |
64faf04c TH |
567 | if (IN_SET(state, MANAGER_STARTING, MANAGER_INITIALIZING) && |
568 | c->startup_blockio_weight != CGROUP_BLKIO_WEIGHT_INVALID) | |
569 | return c->startup_blockio_weight; | |
570 | else if (c->blockio_weight != CGROUP_BLKIO_WEIGHT_INVALID) | |
571 | return c->blockio_weight; | |
572 | else | |
573 | return CGROUP_BLKIO_WEIGHT_DEFAULT; | |
574 | } | |
575 | ||
508c45da | 576 | static uint64_t cgroup_weight_blkio_to_io(uint64_t blkio_weight) { |
538b4852 TH |
577 | return CLAMP(blkio_weight * CGROUP_WEIGHT_DEFAULT / CGROUP_BLKIO_WEIGHT_DEFAULT, |
578 | CGROUP_WEIGHT_MIN, CGROUP_WEIGHT_MAX); | |
579 | } | |
580 | ||
508c45da | 581 | static uint64_t cgroup_weight_io_to_blkio(uint64_t io_weight) { |
538b4852 TH |
582 | return CLAMP(io_weight * CGROUP_BLKIO_WEIGHT_DEFAULT / CGROUP_WEIGHT_DEFAULT, |
583 | CGROUP_BLKIO_WEIGHT_MIN, CGROUP_BLKIO_WEIGHT_MAX); | |
584 | } | |
585 | ||
f29ff115 | 586 | static void cgroup_apply_io_device_weight(Unit *u, const char *dev_path, uint64_t io_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), io_weight); | |
f29ff115 | 596 | r = cg_set_attribute("io", u->cgroup_path, "io.weight", 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 io.weight: %m"); | |
64faf04c TH |
600 | } |
601 | ||
f29ff115 | 602 | static void cgroup_apply_blkio_device_weight(Unit *u, const char *dev_path, uint64_t blkio_weight) { |
64faf04c TH |
603 | char buf[DECIMAL_STR_MAX(dev_t)*2+2+DECIMAL_STR_MAX(uint64_t)+1]; |
604 | dev_t dev; | |
605 | int r; | |
606 | ||
607 | r = lookup_block_device(dev_path, &dev); | |
608 | if (r < 0) | |
609 | return; | |
610 | ||
611 | xsprintf(buf, "%u:%u %" PRIu64 "\n", major(dev), minor(dev), blkio_weight); | |
f29ff115 | 612 | r = cg_set_attribute("blkio", u->cgroup_path, "blkio.weight_device", buf); |
64faf04c | 613 | if (r < 0) |
f29ff115 TH |
614 | log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, |
615 | "Failed to set blkio.weight_device: %m"); | |
64faf04c TH |
616 | } |
617 | ||
f29ff115 | 618 | static unsigned cgroup_apply_io_device_limit(Unit *u, const char *dev_path, uint64_t *limits) { |
64faf04c TH |
619 | char limit_bufs[_CGROUP_IO_LIMIT_TYPE_MAX][DECIMAL_STR_MAX(uint64_t)]; |
620 | char buf[DECIMAL_STR_MAX(dev_t)*2+2+(6+DECIMAL_STR_MAX(uint64_t)+1)*4]; | |
621 | CGroupIOLimitType type; | |
622 | dev_t dev; | |
623 | unsigned n = 0; | |
624 | int r; | |
625 | ||
626 | r = lookup_block_device(dev_path, &dev); | |
627 | if (r < 0) | |
628 | return 0; | |
629 | ||
630 | for (type = 0; type < _CGROUP_IO_LIMIT_TYPE_MAX; type++) { | |
631 | if (limits[type] != cgroup_io_limit_defaults[type]) { | |
632 | xsprintf(limit_bufs[type], "%" PRIu64, limits[type]); | |
633 | n++; | |
634 | } else { | |
635 | xsprintf(limit_bufs[type], "%s", limits[type] == CGROUP_LIMIT_MAX ? "max" : "0"); | |
636 | } | |
637 | } | |
638 | ||
639 | xsprintf(buf, "%u:%u rbps=%s wbps=%s riops=%s wiops=%s\n", major(dev), minor(dev), | |
640 | limit_bufs[CGROUP_IO_RBPS_MAX], limit_bufs[CGROUP_IO_WBPS_MAX], | |
641 | limit_bufs[CGROUP_IO_RIOPS_MAX], limit_bufs[CGROUP_IO_WIOPS_MAX]); | |
f29ff115 | 642 | r = cg_set_attribute("io", u->cgroup_path, "io.max", buf); |
64faf04c | 643 | if (r < 0) |
f29ff115 TH |
644 | log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, |
645 | "Failed to set io.max: %m"); | |
64faf04c TH |
646 | return n; |
647 | } | |
648 | ||
f29ff115 | 649 | static unsigned cgroup_apply_blkio_device_limit(Unit *u, const char *dev_path, uint64_t rbps, uint64_t wbps) { |
64faf04c TH |
650 | char buf[DECIMAL_STR_MAX(dev_t)*2+2+DECIMAL_STR_MAX(uint64_t)+1]; |
651 | dev_t dev; | |
652 | unsigned n = 0; | |
653 | int r; | |
654 | ||
655 | r = lookup_block_device(dev_path, &dev); | |
656 | if (r < 0) | |
657 | return 0; | |
658 | ||
659 | if (rbps != CGROUP_LIMIT_MAX) | |
660 | n++; | |
661 | sprintf(buf, "%u:%u %" PRIu64 "\n", major(dev), minor(dev), rbps); | |
f29ff115 | 662 | r = cg_set_attribute("blkio", u->cgroup_path, "blkio.throttle.read_bps_device", buf); |
64faf04c | 663 | if (r < 0) |
f29ff115 TH |
664 | log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, |
665 | "Failed to set blkio.throttle.read_bps_device: %m"); | |
64faf04c TH |
666 | |
667 | if (wbps != CGROUP_LIMIT_MAX) | |
668 | n++; | |
669 | sprintf(buf, "%u:%u %" PRIu64 "\n", major(dev), minor(dev), wbps); | |
f29ff115 | 670 | r = cg_set_attribute("blkio", u->cgroup_path, "blkio.throttle.write_bps_device", buf); |
64faf04c | 671 | if (r < 0) |
f29ff115 TH |
672 | log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, |
673 | "Failed to set blkio.throttle.write_bps_device: %m"); | |
64faf04c TH |
674 | |
675 | return n; | |
676 | } | |
677 | ||
da4d897e | 678 | static bool cgroup_context_has_unified_memory_config(CGroupContext *c) { |
96e131ea | 679 | 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 |
680 | } |
681 | ||
f29ff115 | 682 | static void cgroup_apply_unified_memory_limit(Unit *u, const char *file, uint64_t v) { |
da4d897e TH |
683 | char buf[DECIMAL_STR_MAX(uint64_t) + 1] = "max"; |
684 | int r; | |
685 | ||
686 | if (v != CGROUP_LIMIT_MAX) | |
687 | xsprintf(buf, "%" PRIu64 "\n", v); | |
688 | ||
f29ff115 | 689 | r = cg_set_attribute("memory", u->cgroup_path, file, buf); |
da4d897e | 690 | if (r < 0) |
f29ff115 TH |
691 | log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, |
692 | "Failed to set %s: %m", file); | |
da4d897e TH |
693 | } |
694 | ||
0f2d84d2 | 695 | static void cgroup_apply_firewall(Unit *u) { |
906c06f6 DM |
696 | int r; |
697 | ||
0f2d84d2 LP |
698 | assert(u); |
699 | ||
906c06f6 DM |
700 | if (u->type == UNIT_SLICE) /* Skip this for slice units, they are inner cgroup nodes, and since bpf/cgroup is |
701 | * not recursive we don't ever touch the bpf on them */ | |
702 | return; | |
703 | ||
704 | r = bpf_firewall_compile(u); | |
705 | if (r < 0) | |
706 | return; | |
707 | ||
708 | (void) bpf_firewall_install(u); | |
709 | return; | |
710 | } | |
711 | ||
712 | static void cgroup_context_apply( | |
713 | Unit *u, | |
714 | CGroupMask apply_mask, | |
715 | bool apply_bpf, | |
716 | ManagerState state) { | |
717 | ||
f29ff115 TH |
718 | const char *path; |
719 | CGroupContext *c; | |
01efdf13 | 720 | bool is_root; |
4ad49000 LP |
721 | int r; |
722 | ||
f29ff115 TH |
723 | assert(u); |
724 | ||
906c06f6 DM |
725 | /* Nothing to do? Exit early! */ |
726 | if (apply_mask == 0 && !apply_bpf) | |
4ad49000 | 727 | return; |
8e274523 | 728 | |
f3725e64 LP |
729 | /* Some cgroup attributes are not supported on the root cgroup, hence silently ignore */ |
730 | is_root = unit_has_root_cgroup(u); | |
731 | ||
732 | assert_se(c = unit_get_cgroup_context(u)); | |
733 | assert_se(path = u->cgroup_path); | |
734 | ||
735 | if (is_root) /* Make sure we don't try to display messages with an empty path. */ | |
6da13913 | 736 | path = "/"; |
01efdf13 | 737 | |
714e2e1d LP |
738 | /* We generally ignore errors caused by read-only mounted |
739 | * cgroup trees (assuming we are running in a container then), | |
740 | * and missing cgroups, i.e. EROFS and ENOENT. */ | |
741 | ||
906c06f6 DM |
742 | if ((apply_mask & CGROUP_MASK_CPU) && !is_root) { |
743 | bool has_weight, has_shares; | |
744 | ||
745 | has_weight = cgroup_context_has_cpu_weight(c); | |
746 | has_shares = cgroup_context_has_cpu_shares(c); | |
8e274523 | 747 | |
b4cccbc1 | 748 | if (cg_all_unified() > 0) { |
66ebf6c0 | 749 | uint64_t weight; |
b2f8b02e | 750 | |
66ebf6c0 TH |
751 | if (has_weight) |
752 | weight = cgroup_context_cpu_weight(c, state); | |
753 | else if (has_shares) { | |
754 | uint64_t shares = cgroup_context_cpu_shares(c, state); | |
b2f8b02e | 755 | |
66ebf6c0 TH |
756 | weight = cgroup_cpu_shares_to_weight(shares); |
757 | ||
758 | log_cgroup_compat(u, "Applying [Startup]CpuShares %" PRIu64 " as [Startup]CpuWeight %" PRIu64 " on %s", | |
759 | shares, weight, path); | |
760 | } else | |
761 | weight = CGROUP_WEIGHT_DEFAULT; | |
762 | ||
763 | cgroup_apply_unified_cpu_config(u, weight, c->cpu_quota_per_sec_usec); | |
764 | } else { | |
765 | uint64_t shares; | |
766 | ||
7d862ab8 | 767 | if (has_weight) { |
66ebf6c0 TH |
768 | uint64_t weight = cgroup_context_cpu_weight(c, state); |
769 | ||
770 | shares = cgroup_cpu_weight_to_shares(weight); | |
771 | ||
772 | log_cgroup_compat(u, "Applying [Startup]CpuWeight %" PRIu64 " as [Startup]CpuShares %" PRIu64 " on %s", | |
773 | weight, shares, path); | |
7d862ab8 TH |
774 | } else if (has_shares) |
775 | shares = cgroup_context_cpu_shares(c, state); | |
776 | else | |
66ebf6c0 TH |
777 | shares = CGROUP_CPU_SHARES_DEFAULT; |
778 | ||
779 | cgroup_apply_legacy_cpu_config(u, shares, c->cpu_quota_per_sec_usec); | |
780 | } | |
4ad49000 LP |
781 | } |
782 | ||
906c06f6 | 783 | if (apply_mask & CGROUP_MASK_IO) { |
538b4852 TH |
784 | bool has_io = cgroup_context_has_io_config(c); |
785 | bool has_blockio = cgroup_context_has_blockio_config(c); | |
13c31542 TH |
786 | |
787 | if (!is_root) { | |
64faf04c TH |
788 | char buf[8+DECIMAL_STR_MAX(uint64_t)+1]; |
789 | uint64_t weight; | |
13c31542 | 790 | |
538b4852 TH |
791 | if (has_io) |
792 | weight = cgroup_context_io_weight(c, state); | |
128fadc9 TH |
793 | else if (has_blockio) { |
794 | uint64_t blkio_weight = cgroup_context_blkio_weight(c, state); | |
795 | ||
796 | weight = cgroup_weight_blkio_to_io(blkio_weight); | |
797 | ||
798 | log_cgroup_compat(u, "Applying [Startup]BlockIOWeight %" PRIu64 " as [Startup]IOWeight %" PRIu64, | |
799 | blkio_weight, weight); | |
800 | } else | |
538b4852 | 801 | weight = CGROUP_WEIGHT_DEFAULT; |
13c31542 TH |
802 | |
803 | xsprintf(buf, "default %" PRIu64 "\n", weight); | |
804 | r = cg_set_attribute("io", path, "io.weight", buf); | |
805 | if (r < 0) | |
f29ff115 TH |
806 | log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, |
807 | "Failed to set io.weight: %m"); | |
13c31542 | 808 | |
538b4852 TH |
809 | if (has_io) { |
810 | CGroupIODeviceWeight *w; | |
811 | ||
812 | /* FIXME: no way to reset this list */ | |
813 | LIST_FOREACH(device_weights, w, c->io_device_weights) | |
f29ff115 | 814 | cgroup_apply_io_device_weight(u, w->path, w->weight); |
538b4852 TH |
815 | } else if (has_blockio) { |
816 | CGroupBlockIODeviceWeight *w; | |
817 | ||
818 | /* FIXME: no way to reset this list */ | |
128fadc9 TH |
819 | LIST_FOREACH(device_weights, w, c->blockio_device_weights) { |
820 | weight = cgroup_weight_blkio_to_io(w->weight); | |
821 | ||
822 | log_cgroup_compat(u, "Applying BlockIODeviceWeight %" PRIu64 " as IODeviceWeight %" PRIu64 " for %s", | |
823 | w->weight, weight, w->path); | |
824 | ||
825 | cgroup_apply_io_device_weight(u, w->path, weight); | |
826 | } | |
538b4852 | 827 | } |
13c31542 TH |
828 | } |
829 | ||
64faf04c | 830 | /* Apply limits and free ones without config. */ |
538b4852 TH |
831 | if (has_io) { |
832 | CGroupIODeviceLimit *l, *next; | |
833 | ||
834 | LIST_FOREACH_SAFE(device_limits, l, next, c->io_device_limits) { | |
f29ff115 | 835 | if (!cgroup_apply_io_device_limit(u, l->path, l->limits)) |
538b4852 TH |
836 | cgroup_context_free_io_device_limit(c, l); |
837 | } | |
838 | } else if (has_blockio) { | |
839 | CGroupBlockIODeviceBandwidth *b, *next; | |
840 | ||
841 | LIST_FOREACH_SAFE(device_bandwidths, b, next, c->blockio_device_bandwidths) { | |
842 | uint64_t limits[_CGROUP_IO_LIMIT_TYPE_MAX]; | |
843 | CGroupIOLimitType type; | |
844 | ||
845 | for (type = 0; type < _CGROUP_IO_LIMIT_TYPE_MAX; type++) | |
846 | limits[type] = cgroup_io_limit_defaults[type]; | |
847 | ||
848 | limits[CGROUP_IO_RBPS_MAX] = b->rbps; | |
849 | limits[CGROUP_IO_WBPS_MAX] = b->wbps; | |
850 | ||
128fadc9 TH |
851 | log_cgroup_compat(u, "Applying BlockIO{Read|Write}Bandwidth %" PRIu64 " %" PRIu64 " as IO{Read|Write}BandwidthMax for %s", |
852 | b->rbps, b->wbps, b->path); | |
853 | ||
f29ff115 | 854 | if (!cgroup_apply_io_device_limit(u, b->path, limits)) |
538b4852 TH |
855 | cgroup_context_free_blockio_device_bandwidth(c, b); |
856 | } | |
13c31542 TH |
857 | } |
858 | } | |
859 | ||
906c06f6 | 860 | if (apply_mask & CGROUP_MASK_BLKIO) { |
538b4852 TH |
861 | bool has_io = cgroup_context_has_io_config(c); |
862 | bool has_blockio = cgroup_context_has_blockio_config(c); | |
4ad49000 | 863 | |
01efdf13 | 864 | if (!is_root) { |
64faf04c TH |
865 | char buf[DECIMAL_STR_MAX(uint64_t)+1]; |
866 | uint64_t weight; | |
64faf04c | 867 | |
7d862ab8 | 868 | if (has_io) { |
128fadc9 TH |
869 | uint64_t io_weight = cgroup_context_io_weight(c, state); |
870 | ||
538b4852 | 871 | weight = cgroup_weight_io_to_blkio(cgroup_context_io_weight(c, state)); |
128fadc9 TH |
872 | |
873 | log_cgroup_compat(u, "Applying [Startup]IOWeight %" PRIu64 " as [Startup]BlockIOWeight %" PRIu64, | |
874 | io_weight, weight); | |
7d862ab8 TH |
875 | } else if (has_blockio) |
876 | weight = cgroup_context_blkio_weight(c, state); | |
877 | else | |
538b4852 | 878 | weight = CGROUP_BLKIO_WEIGHT_DEFAULT; |
64faf04c TH |
879 | |
880 | xsprintf(buf, "%" PRIu64 "\n", weight); | |
01efdf13 | 881 | r = cg_set_attribute("blkio", path, "blkio.weight", buf); |
1aeab12b | 882 | if (r < 0) |
f29ff115 TH |
883 | log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, |
884 | "Failed to set blkio.weight: %m"); | |
4ad49000 | 885 | |
7d862ab8 | 886 | if (has_io) { |
538b4852 TH |
887 | CGroupIODeviceWeight *w; |
888 | ||
889 | /* FIXME: no way to reset this list */ | |
128fadc9 TH |
890 | LIST_FOREACH(device_weights, w, c->io_device_weights) { |
891 | weight = cgroup_weight_io_to_blkio(w->weight); | |
892 | ||
893 | log_cgroup_compat(u, "Applying IODeviceWeight %" PRIu64 " as BlockIODeviceWeight %" PRIu64 " for %s", | |
894 | w->weight, weight, w->path); | |
895 | ||
896 | cgroup_apply_blkio_device_weight(u, w->path, weight); | |
897 | } | |
7d862ab8 TH |
898 | } else if (has_blockio) { |
899 | CGroupBlockIODeviceWeight *w; | |
900 | ||
901 | /* FIXME: no way to reset this list */ | |
902 | LIST_FOREACH(device_weights, w, c->blockio_device_weights) | |
903 | cgroup_apply_blkio_device_weight(u, w->path, w->weight); | |
538b4852 | 904 | } |
4ad49000 LP |
905 | } |
906 | ||
64faf04c | 907 | /* Apply limits and free ones without config. */ |
7d862ab8 | 908 | if (has_io) { |
538b4852 TH |
909 | CGroupIODeviceLimit *l, *next; |
910 | ||
911 | LIST_FOREACH_SAFE(device_limits, l, next, c->io_device_limits) { | |
128fadc9 TH |
912 | log_cgroup_compat(u, "Applying IO{Read|Write}Bandwidth %" PRIu64 " %" PRIu64 " as BlockIO{Read|Write}BandwidthMax for %s", |
913 | l->limits[CGROUP_IO_RBPS_MAX], l->limits[CGROUP_IO_WBPS_MAX], l->path); | |
914 | ||
f29ff115 | 915 | if (!cgroup_apply_blkio_device_limit(u, l->path, l->limits[CGROUP_IO_RBPS_MAX], l->limits[CGROUP_IO_WBPS_MAX])) |
538b4852 TH |
916 | cgroup_context_free_io_device_limit(c, l); |
917 | } | |
7d862ab8 TH |
918 | } else if (has_blockio) { |
919 | CGroupBlockIODeviceBandwidth *b, *next; | |
920 | ||
921 | LIST_FOREACH_SAFE(device_bandwidths, b, next, c->blockio_device_bandwidths) | |
922 | if (!cgroup_apply_blkio_device_limit(u, b->path, b->rbps, b->wbps)) | |
923 | cgroup_context_free_blockio_device_bandwidth(c, b); | |
d686d8a9 | 924 | } |
8e274523 LP |
925 | } |
926 | ||
906c06f6 | 927 | if ((apply_mask & CGROUP_MASK_MEMORY) && !is_root) { |
b4cccbc1 LP |
928 | if (cg_all_unified() > 0) { |
929 | uint64_t max, swap_max = CGROUP_LIMIT_MAX; | |
efdb0237 | 930 | |
96e131ea | 931 | if (cgroup_context_has_unified_memory_config(c)) { |
da4d897e | 932 | max = c->memory_max; |
96e131ea WC |
933 | swap_max = c->memory_swap_max; |
934 | } else { | |
da4d897e | 935 | max = c->memory_limit; |
efdb0237 | 936 | |
128fadc9 TH |
937 | if (max != CGROUP_LIMIT_MAX) |
938 | log_cgroup_compat(u, "Applying MemoryLimit %" PRIu64 " as MemoryMax", max); | |
939 | } | |
940 | ||
f29ff115 TH |
941 | cgroup_apply_unified_memory_limit(u, "memory.low", c->memory_low); |
942 | cgroup_apply_unified_memory_limit(u, "memory.high", c->memory_high); | |
943 | cgroup_apply_unified_memory_limit(u, "memory.max", max); | |
96e131ea | 944 | cgroup_apply_unified_memory_limit(u, "memory.swap.max", swap_max); |
efdb0237 | 945 | } else { |
da4d897e | 946 | char buf[DECIMAL_STR_MAX(uint64_t) + 1]; |
7d862ab8 | 947 | uint64_t val; |
da4d897e | 948 | |
7d862ab8 | 949 | if (cgroup_context_has_unified_memory_config(c)) { |
78a4ee59 | 950 | val = c->memory_max; |
7d862ab8 TH |
951 | log_cgroup_compat(u, "Applying MemoryMax %" PRIi64 " as MemoryLimit", val); |
952 | } else | |
953 | val = c->memory_limit; | |
128fadc9 | 954 | |
78a4ee59 DM |
955 | if (val == CGROUP_LIMIT_MAX) |
956 | strncpy(buf, "-1\n", sizeof(buf)); | |
957 | else | |
958 | xsprintf(buf, "%" PRIu64 "\n", val); | |
959 | ||
da4d897e TH |
960 | r = cg_set_attribute("memory", path, "memory.limit_in_bytes", buf); |
961 | if (r < 0) | |
f29ff115 TH |
962 | log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, |
963 | "Failed to set memory.limit_in_bytes: %m"); | |
da4d897e | 964 | } |
4ad49000 | 965 | } |
8e274523 | 966 | |
906c06f6 | 967 | if ((apply_mask & CGROUP_MASK_DEVICES) && !is_root) { |
4ad49000 | 968 | CGroupDeviceAllow *a; |
8e274523 | 969 | |
714e2e1d LP |
970 | /* Changing the devices list of a populated cgroup |
971 | * might result in EINVAL, hence ignore EINVAL | |
972 | * here. */ | |
973 | ||
4ad49000 LP |
974 | if (c->device_allow || c->device_policy != CGROUP_AUTO) |
975 | r = cg_set_attribute("devices", path, "devices.deny", "a"); | |
976 | else | |
977 | r = cg_set_attribute("devices", path, "devices.allow", "a"); | |
1aeab12b | 978 | if (r < 0) |
f29ff115 TH |
979 | log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EINVAL, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, |
980 | "Failed to reset devices.list: %m"); | |
fb385181 | 981 | |
4ad49000 LP |
982 | if (c->device_policy == CGROUP_CLOSED || |
983 | (c->device_policy == CGROUP_AUTO && c->device_allow)) { | |
984 | static const char auto_devices[] = | |
7d711efb LP |
985 | "/dev/null\0" "rwm\0" |
986 | "/dev/zero\0" "rwm\0" | |
987 | "/dev/full\0" "rwm\0" | |
988 | "/dev/random\0" "rwm\0" | |
989 | "/dev/urandom\0" "rwm\0" | |
990 | "/dev/tty\0" "rwm\0" | |
5a7f87a9 | 991 | "/dev/ptmx\0" "rwm\0" |
0d9e7991 | 992 | /* Allow /run/systemd/inaccessible/{chr,blk} devices for mapping InaccessiblePaths */ |
e7330dfe DP |
993 | "-/run/systemd/inaccessible/chr\0" "rwm\0" |
994 | "-/run/systemd/inaccessible/blk\0" "rwm\0"; | |
4ad49000 LP |
995 | |
996 | const char *x, *y; | |
997 | ||
998 | NULSTR_FOREACH_PAIR(x, y, auto_devices) | |
999 | whitelist_device(path, x, y); | |
7d711efb | 1000 | |
5a7f87a9 | 1001 | /* PTS (/dev/pts) devices may not be duplicated, but accessed */ |
7d711efb | 1002 | whitelist_major(path, "pts", 'c', "rw"); |
4ad49000 LP |
1003 | } |
1004 | ||
1005 | LIST_FOREACH(device_allow, a, c->device_allow) { | |
fb4650aa | 1006 | char acc[4], *val; |
4ad49000 LP |
1007 | unsigned k = 0; |
1008 | ||
1009 | if (a->r) | |
1010 | acc[k++] = 'r'; | |
1011 | if (a->w) | |
1012 | acc[k++] = 'w'; | |
1013 | if (a->m) | |
1014 | acc[k++] = 'm'; | |
fb385181 | 1015 | |
4ad49000 LP |
1016 | if (k == 0) |
1017 | continue; | |
fb385181 | 1018 | |
4ad49000 | 1019 | acc[k++] = 0; |
90060676 | 1020 | |
27458ed6 | 1021 | if (path_startswith(a->path, "/dev/")) |
90060676 | 1022 | whitelist_device(path, a->path, acc); |
fb4650aa ZJS |
1023 | else if ((val = startswith(a->path, "block-"))) |
1024 | whitelist_major(path, val, 'b', acc); | |
1025 | else if ((val = startswith(a->path, "char-"))) | |
1026 | whitelist_major(path, val, 'c', acc); | |
90060676 | 1027 | else |
f29ff115 | 1028 | log_unit_debug(u, "Ignoring device %s while writing cgroup attribute.", a->path); |
4ad49000 LP |
1029 | } |
1030 | } | |
03a7b521 | 1031 | |
00b5974f LP |
1032 | if (apply_mask & CGROUP_MASK_PIDS) { |
1033 | ||
1034 | if (is_root) { | |
1035 | /* So, the "pids" controller does not expose anything on the root cgroup, in order not to | |
1036 | * replicate knobs exposed elsewhere needlessly. We abstract this away here however, and when | |
1037 | * the knobs of the root cgroup are modified propagate this to the relevant sysctls. There's a | |
1038 | * non-obvious asymmetry however: unlike the cgroup properties we don't really want to take | |
1039 | * exclusive ownership of the sysctls, but we still want to honour things if the user sets | |
1040 | * limits. Hence we employ sort of a one-way strategy: when the user sets a bounded limit | |
1041 | * through us it counts. When the user afterwards unsets it again (i.e. sets it to unbounded) | |
1042 | * it also counts. But if the user never set a limit through us (i.e. we are the default of | |
1043 | * "unbounded") we leave things unmodified. For this we manage a global boolean that we turn on | |
1044 | * the first time we set a limit. Note that this boolean is flushed out on manager reload, | |
1045 | * which is desirable so that there's an offical way to release control of the sysctl from | |
1046 | * systemd: set the limit to unbounded and reload. */ | |
1047 | ||
1048 | if (c->tasks_max != CGROUP_LIMIT_MAX) { | |
1049 | u->manager->sysctl_pid_max_changed = true; | |
1050 | r = procfs_tasks_set_limit(c->tasks_max); | |
1051 | } else if (u->manager->sysctl_pid_max_changed) | |
1052 | r = procfs_tasks_set_limit(TASKS_MAX); | |
1053 | else | |
1054 | r = 0; | |
03a7b521 | 1055 | |
00b5974f LP |
1056 | if (r < 0) |
1057 | log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, | |
1058 | "Failed to write to tasks limit sysctls: %m"); | |
03a7b521 | 1059 | |
00b5974f LP |
1060 | } else { |
1061 | if (c->tasks_max != CGROUP_LIMIT_MAX) { | |
1062 | char buf[DECIMAL_STR_MAX(uint64_t) + 2]; | |
03a7b521 | 1063 | |
00b5974f LP |
1064 | sprintf(buf, "%" PRIu64 "\n", c->tasks_max); |
1065 | r = cg_set_attribute("pids", path, "pids.max", buf); | |
1066 | } else | |
1067 | r = cg_set_attribute("pids", path, "pids.max", "max"); | |
1068 | if (r < 0) | |
1069 | log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, | |
1070 | "Failed to set pids.max: %m"); | |
1071 | } | |
03a7b521 | 1072 | } |
906c06f6 DM |
1073 | |
1074 | if (apply_bpf) | |
0f2d84d2 | 1075 | cgroup_apply_firewall(u); |
fb385181 LP |
1076 | } |
1077 | ||
efdb0237 LP |
1078 | CGroupMask cgroup_context_get_mask(CGroupContext *c) { |
1079 | CGroupMask mask = 0; | |
8e274523 | 1080 | |
4ad49000 | 1081 | /* Figure out which controllers we need */ |
8e274523 | 1082 | |
b2f8b02e | 1083 | if (c->cpu_accounting || |
66ebf6c0 TH |
1084 | cgroup_context_has_cpu_weight(c) || |
1085 | cgroup_context_has_cpu_shares(c) || | |
3a43da28 | 1086 | c->cpu_quota_per_sec_usec != USEC_INFINITY) |
efdb0237 | 1087 | mask |= CGROUP_MASK_CPUACCT | CGROUP_MASK_CPU; |
ecedd90f | 1088 | |
538b4852 TH |
1089 | if (cgroup_context_has_io_config(c) || cgroup_context_has_blockio_config(c)) |
1090 | mask |= CGROUP_MASK_IO | CGROUP_MASK_BLKIO; | |
ecedd90f | 1091 | |
4ad49000 | 1092 | if (c->memory_accounting || |
da4d897e TH |
1093 | c->memory_limit != CGROUP_LIMIT_MAX || |
1094 | cgroup_context_has_unified_memory_config(c)) | |
efdb0237 | 1095 | mask |= CGROUP_MASK_MEMORY; |
8e274523 | 1096 | |
a931ad47 LP |
1097 | if (c->device_allow || |
1098 | c->device_policy != CGROUP_AUTO) | |
3905f127 | 1099 | mask |= CGROUP_MASK_DEVICES; |
4ad49000 | 1100 | |
03a7b521 | 1101 | if (c->tasks_accounting || |
8793fa25 | 1102 | c->tasks_max != CGROUP_LIMIT_MAX) |
03a7b521 LP |
1103 | mask |= CGROUP_MASK_PIDS; |
1104 | ||
4ad49000 | 1105 | return mask; |
8e274523 LP |
1106 | } |
1107 | ||
efdb0237 | 1108 | CGroupMask unit_get_own_mask(Unit *u) { |
4ad49000 | 1109 | CGroupContext *c; |
8e274523 | 1110 | |
efdb0237 LP |
1111 | /* Returns the mask of controllers the unit needs for itself */ |
1112 | ||
4ad49000 LP |
1113 | c = unit_get_cgroup_context(u); |
1114 | if (!c) | |
1115 | return 0; | |
8e274523 | 1116 | |
64e844e5 | 1117 | return cgroup_context_get_mask(c) | unit_get_delegate_mask(u); |
02638280 LP |
1118 | } |
1119 | ||
1120 | CGroupMask unit_get_delegate_mask(Unit *u) { | |
1121 | CGroupContext *c; | |
1122 | ||
1123 | /* If delegation is turned on, then turn on selected controllers, unless we are on the legacy hierarchy and the | |
1124 | * process we fork into is known to drop privileges, and hence shouldn't get access to the controllers. | |
19af675e | 1125 | * |
02638280 | 1126 | * Note that on the unified hierarchy it is safe to delegate controllers to unprivileged services. */ |
a931ad47 | 1127 | |
1d9cc876 | 1128 | if (!unit_cgroup_delegate(u)) |
02638280 LP |
1129 | return 0; |
1130 | ||
1131 | if (cg_all_unified() <= 0) { | |
a931ad47 LP |
1132 | ExecContext *e; |
1133 | ||
1134 | e = unit_get_exec_context(u); | |
02638280 LP |
1135 | if (e && !exec_context_maintains_privileges(e)) |
1136 | return 0; | |
a931ad47 LP |
1137 | } |
1138 | ||
1d9cc876 | 1139 | assert_se(c = unit_get_cgroup_context(u)); |
02638280 | 1140 | return c->delegate_controllers; |
8e274523 LP |
1141 | } |
1142 | ||
efdb0237 | 1143 | CGroupMask unit_get_members_mask(Unit *u) { |
4ad49000 | 1144 | assert(u); |
bc432dc7 | 1145 | |
02638280 | 1146 | /* Returns the mask of controllers all of the unit's children require, merged */ |
efdb0237 | 1147 | |
bc432dc7 LP |
1148 | if (u->cgroup_members_mask_valid) |
1149 | return u->cgroup_members_mask; | |
1150 | ||
64e844e5 | 1151 | u->cgroup_members_mask = 0; |
bc432dc7 LP |
1152 | |
1153 | if (u->type == UNIT_SLICE) { | |
eef85c4a | 1154 | void *v; |
bc432dc7 LP |
1155 | Unit *member; |
1156 | Iterator i; | |
1157 | ||
eef85c4a | 1158 | HASHMAP_FOREACH_KEY(v, member, u->dependencies[UNIT_BEFORE], i) { |
bc432dc7 LP |
1159 | |
1160 | if (member == u) | |
1161 | continue; | |
1162 | ||
d4fdc205 | 1163 | if (UNIT_DEREF(member->slice) != u) |
bc432dc7 LP |
1164 | continue; |
1165 | ||
31604970 | 1166 | u->cgroup_members_mask |= unit_get_subtree_mask(member); /* note that this calls ourselves again, for the children */ |
bc432dc7 LP |
1167 | } |
1168 | } | |
1169 | ||
1170 | u->cgroup_members_mask_valid = true; | |
6414b7c9 | 1171 | return u->cgroup_members_mask; |
246aa6dd LP |
1172 | } |
1173 | ||
efdb0237 | 1174 | CGroupMask unit_get_siblings_mask(Unit *u) { |
4ad49000 | 1175 | assert(u); |
246aa6dd | 1176 | |
efdb0237 LP |
1177 | /* Returns the mask of controllers all of the unit's siblings |
1178 | * require, i.e. the members mask of the unit's parent slice | |
1179 | * if there is one. */ | |
1180 | ||
bc432dc7 | 1181 | if (UNIT_ISSET(u->slice)) |
637f421e | 1182 | return unit_get_members_mask(UNIT_DEREF(u->slice)); |
4ad49000 | 1183 | |
64e844e5 | 1184 | return unit_get_subtree_mask(u); /* we are the top-level slice */ |
246aa6dd LP |
1185 | } |
1186 | ||
efdb0237 LP |
1187 | CGroupMask unit_get_subtree_mask(Unit *u) { |
1188 | ||
1189 | /* Returns the mask of this subtree, meaning of the group | |
1190 | * itself and its children. */ | |
1191 | ||
1192 | return unit_get_own_mask(u) | unit_get_members_mask(u); | |
1193 | } | |
1194 | ||
1195 | CGroupMask unit_get_target_mask(Unit *u) { | |
1196 | CGroupMask mask; | |
1197 | ||
1198 | /* This returns the cgroup mask of all controllers to enable | |
1199 | * for a specific cgroup, i.e. everything it needs itself, | |
1200 | * plus all that its children need, plus all that its siblings | |
1201 | * need. This is primarily useful on the legacy cgroup | |
1202 | * hierarchy, where we need to duplicate each cgroup in each | |
1203 | * hierarchy that shall be enabled for it. */ | |
6414b7c9 | 1204 | |
efdb0237 LP |
1205 | mask = unit_get_own_mask(u) | unit_get_members_mask(u) | unit_get_siblings_mask(u); |
1206 | mask &= u->manager->cgroup_supported; | |
1207 | ||
1208 | return mask; | |
1209 | } | |
1210 | ||
1211 | CGroupMask unit_get_enable_mask(Unit *u) { | |
1212 | CGroupMask mask; | |
1213 | ||
1214 | /* This returns the cgroup mask of all controllers to enable | |
1215 | * for the children of a specific cgroup. This is primarily | |
1216 | * useful for the unified cgroup hierarchy, where each cgroup | |
1217 | * controls which controllers are enabled for its children. */ | |
1218 | ||
1219 | mask = unit_get_members_mask(u); | |
6414b7c9 DS |
1220 | mask &= u->manager->cgroup_supported; |
1221 | ||
1222 | return mask; | |
1223 | } | |
1224 | ||
906c06f6 DM |
1225 | bool unit_get_needs_bpf(Unit *u) { |
1226 | CGroupContext *c; | |
1227 | Unit *p; | |
1228 | assert(u); | |
1229 | ||
1230 | /* We never attach BPF to slice units, as they are inner cgroup nodes and cgroup/BPF is not recursive at the | |
1231 | * moment. */ | |
1232 | if (u->type == UNIT_SLICE) | |
1233 | return false; | |
1234 | ||
1235 | c = unit_get_cgroup_context(u); | |
1236 | if (!c) | |
1237 | return false; | |
1238 | ||
1239 | if (c->ip_accounting || | |
1240 | c->ip_address_allow || | |
1241 | c->ip_address_deny) | |
1242 | return true; | |
1243 | ||
1244 | /* If any parent slice has an IP access list defined, it applies too */ | |
1245 | for (p = UNIT_DEREF(u->slice); p; p = UNIT_DEREF(p->slice)) { | |
1246 | c = unit_get_cgroup_context(p); | |
1247 | if (!c) | |
1248 | return false; | |
1249 | ||
1250 | if (c->ip_address_allow || | |
1251 | c->ip_address_deny) | |
1252 | return true; | |
1253 | } | |
1254 | ||
1255 | return false; | |
1256 | } | |
1257 | ||
6414b7c9 DS |
1258 | /* Recurse from a unit up through its containing slices, propagating |
1259 | * mask bits upward. A unit is also member of itself. */ | |
bc432dc7 | 1260 | void unit_update_cgroup_members_masks(Unit *u) { |
efdb0237 | 1261 | CGroupMask m; |
bc432dc7 LP |
1262 | bool more; |
1263 | ||
1264 | assert(u); | |
1265 | ||
1266 | /* Calculate subtree mask */ | |
efdb0237 | 1267 | m = unit_get_subtree_mask(u); |
bc432dc7 LP |
1268 | |
1269 | /* See if anything changed from the previous invocation. If | |
1270 | * not, we're done. */ | |
1271 | if (u->cgroup_subtree_mask_valid && m == u->cgroup_subtree_mask) | |
1272 | return; | |
1273 | ||
1274 | more = | |
1275 | u->cgroup_subtree_mask_valid && | |
1276 | ((m & ~u->cgroup_subtree_mask) != 0) && | |
1277 | ((~m & u->cgroup_subtree_mask) == 0); | |
1278 | ||
1279 | u->cgroup_subtree_mask = m; | |
1280 | u->cgroup_subtree_mask_valid = true; | |
1281 | ||
6414b7c9 DS |
1282 | if (UNIT_ISSET(u->slice)) { |
1283 | Unit *s = UNIT_DEREF(u->slice); | |
bc432dc7 LP |
1284 | |
1285 | if (more) | |
1286 | /* There's more set now than before. We | |
1287 | * propagate the new mask to the parent's mask | |
1288 | * (not caring if it actually was valid or | |
1289 | * not). */ | |
1290 | ||
1291 | s->cgroup_members_mask |= m; | |
1292 | ||
1293 | else | |
1294 | /* There's less set now than before (or we | |
1295 | * don't know), we need to recalculate | |
1296 | * everything, so let's invalidate the | |
1297 | * parent's members mask */ | |
1298 | ||
1299 | s->cgroup_members_mask_valid = false; | |
1300 | ||
1301 | /* And now make sure that this change also hits our | |
1302 | * grandparents */ | |
1303 | unit_update_cgroup_members_masks(s); | |
6414b7c9 DS |
1304 | } |
1305 | } | |
1306 | ||
6592b975 | 1307 | const char *unit_get_realized_cgroup_path(Unit *u, CGroupMask mask) { |
03b90d4b | 1308 | |
6592b975 | 1309 | /* Returns the realized cgroup path of the specified unit where all specified controllers are available. */ |
03b90d4b LP |
1310 | |
1311 | while (u) { | |
6592b975 | 1312 | |
03b90d4b LP |
1313 | if (u->cgroup_path && |
1314 | u->cgroup_realized && | |
1315 | (u->cgroup_realized_mask & mask) == mask) | |
1316 | return u->cgroup_path; | |
1317 | ||
1318 | u = UNIT_DEREF(u->slice); | |
1319 | } | |
1320 | ||
1321 | return NULL; | |
1322 | } | |
1323 | ||
6592b975 LP |
1324 | static const char *migrate_callback(CGroupMask mask, void *userdata) { |
1325 | return unit_get_realized_cgroup_path(userdata, mask); | |
1326 | } | |
1327 | ||
efdb0237 LP |
1328 | char *unit_default_cgroup_path(Unit *u) { |
1329 | _cleanup_free_ char *escaped = NULL, *slice = NULL; | |
1330 | int r; | |
1331 | ||
1332 | assert(u); | |
1333 | ||
1334 | if (unit_has_name(u, SPECIAL_ROOT_SLICE)) | |
1335 | return strdup(u->manager->cgroup_root); | |
1336 | ||
1337 | if (UNIT_ISSET(u->slice) && !unit_has_name(UNIT_DEREF(u->slice), SPECIAL_ROOT_SLICE)) { | |
1338 | r = cg_slice_to_path(UNIT_DEREF(u->slice)->id, &slice); | |
1339 | if (r < 0) | |
1340 | return NULL; | |
1341 | } | |
1342 | ||
1343 | escaped = cg_escape(u->id); | |
1344 | if (!escaped) | |
1345 | return NULL; | |
1346 | ||
1347 | if (slice) | |
605405c6 ZJS |
1348 | return strjoin(u->manager->cgroup_root, "/", slice, "/", |
1349 | escaped); | |
efdb0237 | 1350 | else |
605405c6 | 1351 | return strjoin(u->manager->cgroup_root, "/", escaped); |
efdb0237 LP |
1352 | } |
1353 | ||
1354 | int unit_set_cgroup_path(Unit *u, const char *path) { | |
1355 | _cleanup_free_ char *p = NULL; | |
1356 | int r; | |
1357 | ||
1358 | assert(u); | |
1359 | ||
1360 | if (path) { | |
1361 | p = strdup(path); | |
1362 | if (!p) | |
1363 | return -ENOMEM; | |
1364 | } else | |
1365 | p = NULL; | |
1366 | ||
1367 | if (streq_ptr(u->cgroup_path, p)) | |
1368 | return 0; | |
1369 | ||
1370 | if (p) { | |
1371 | r = hashmap_put(u->manager->cgroup_unit, p, u); | |
1372 | if (r < 0) | |
1373 | return r; | |
1374 | } | |
1375 | ||
1376 | unit_release_cgroup(u); | |
1377 | ||
1378 | u->cgroup_path = p; | |
1379 | p = NULL; | |
1380 | ||
1381 | return 1; | |
1382 | } | |
1383 | ||
1384 | int unit_watch_cgroup(Unit *u) { | |
ab2c3861 | 1385 | _cleanup_free_ char *events = NULL; |
efdb0237 LP |
1386 | int r; |
1387 | ||
1388 | assert(u); | |
1389 | ||
1390 | if (!u->cgroup_path) | |
1391 | return 0; | |
1392 | ||
1393 | if (u->cgroup_inotify_wd >= 0) | |
1394 | return 0; | |
1395 | ||
1396 | /* Only applies to the unified hierarchy */ | |
c22800e4 | 1397 | r = cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER); |
b4cccbc1 LP |
1398 | if (r < 0) |
1399 | return log_error_errno(r, "Failed to determine whether the name=systemd hierarchy is unified: %m"); | |
1400 | if (r == 0) | |
efdb0237 LP |
1401 | return 0; |
1402 | ||
1403 | /* Don't watch the root slice, it's pointless. */ | |
1404 | if (unit_has_name(u, SPECIAL_ROOT_SLICE)) | |
1405 | return 0; | |
1406 | ||
1407 | r = hashmap_ensure_allocated(&u->manager->cgroup_inotify_wd_unit, &trivial_hash_ops); | |
1408 | if (r < 0) | |
1409 | return log_oom(); | |
1410 | ||
ab2c3861 | 1411 | r = cg_get_path(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path, "cgroup.events", &events); |
efdb0237 LP |
1412 | if (r < 0) |
1413 | return log_oom(); | |
1414 | ||
ab2c3861 | 1415 | u->cgroup_inotify_wd = inotify_add_watch(u->manager->cgroup_inotify_fd, events, IN_MODIFY); |
efdb0237 LP |
1416 | if (u->cgroup_inotify_wd < 0) { |
1417 | ||
1418 | if (errno == ENOENT) /* If the directory is already | |
1419 | * gone we don't need to track | |
1420 | * it, so this is not an error */ | |
1421 | return 0; | |
1422 | ||
1423 | return log_unit_error_errno(u, errno, "Failed to add inotify watch descriptor for control group %s: %m", u->cgroup_path); | |
1424 | } | |
1425 | ||
1426 | r = hashmap_put(u->manager->cgroup_inotify_wd_unit, INT_TO_PTR(u->cgroup_inotify_wd), u); | |
1427 | if (r < 0) | |
1428 | return log_unit_error_errno(u, r, "Failed to add inotify watch descriptor to hash map: %m"); | |
1429 | ||
1430 | return 0; | |
1431 | } | |
1432 | ||
a4634b21 LP |
1433 | int unit_pick_cgroup_path(Unit *u) { |
1434 | _cleanup_free_ char *path = NULL; | |
1435 | int r; | |
1436 | ||
1437 | assert(u); | |
1438 | ||
1439 | if (u->cgroup_path) | |
1440 | return 0; | |
1441 | ||
1442 | if (!UNIT_HAS_CGROUP_CONTEXT(u)) | |
1443 | return -EINVAL; | |
1444 | ||
1445 | path = unit_default_cgroup_path(u); | |
1446 | if (!path) | |
1447 | return log_oom(); | |
1448 | ||
1449 | r = unit_set_cgroup_path(u, path); | |
1450 | if (r == -EEXIST) | |
1451 | return log_unit_error_errno(u, r, "Control group %s exists already.", path); | |
1452 | if (r < 0) | |
1453 | return log_unit_error_errno(u, r, "Failed to set unit's control group path to %s: %m", path); | |
1454 | ||
1455 | return 0; | |
1456 | } | |
1457 | ||
efdb0237 LP |
1458 | static int unit_create_cgroup( |
1459 | Unit *u, | |
1460 | CGroupMask target_mask, | |
906c06f6 DM |
1461 | CGroupMask enable_mask, |
1462 | bool needs_bpf) { | |
efdb0237 | 1463 | |
0cd385d3 | 1464 | CGroupContext *c; |
bc432dc7 | 1465 | int r; |
64747e2d | 1466 | |
4ad49000 | 1467 | assert(u); |
64747e2d | 1468 | |
0cd385d3 LP |
1469 | c = unit_get_cgroup_context(u); |
1470 | if (!c) | |
1471 | return 0; | |
1472 | ||
a4634b21 LP |
1473 | /* Figure out our cgroup path */ |
1474 | r = unit_pick_cgroup_path(u); | |
1475 | if (r < 0) | |
1476 | return r; | |
b58b8e11 | 1477 | |
03b90d4b | 1478 | /* First, create our own group */ |
efdb0237 | 1479 | r = cg_create_everywhere(u->manager->cgroup_supported, target_mask, u->cgroup_path); |
23bbb0de | 1480 | if (r < 0) |
efdb0237 LP |
1481 | return log_unit_error_errno(u, r, "Failed to create cgroup %s: %m", u->cgroup_path); |
1482 | ||
1483 | /* Start watching it */ | |
1484 | (void) unit_watch_cgroup(u); | |
1485 | ||
1486 | /* Enable all controllers we need */ | |
1487 | r = cg_enable_everywhere(u->manager->cgroup_supported, enable_mask, u->cgroup_path); | |
1488 | if (r < 0) | |
1489 | log_unit_warning_errno(u, r, "Failed to enable controllers on cgroup %s, ignoring: %m", u->cgroup_path); | |
03b90d4b LP |
1490 | |
1491 | /* Keep track that this is now realized */ | |
4ad49000 | 1492 | u->cgroup_realized = true; |
efdb0237 | 1493 | u->cgroup_realized_mask = target_mask; |
ccf78df1 | 1494 | u->cgroup_enabled_mask = enable_mask; |
906c06f6 | 1495 | u->cgroup_bpf_state = needs_bpf ? UNIT_CGROUP_BPF_ON : UNIT_CGROUP_BPF_OFF; |
4ad49000 | 1496 | |
1d9cc876 | 1497 | if (u->type != UNIT_SLICE && !unit_cgroup_delegate(u)) { |
0cd385d3 LP |
1498 | |
1499 | /* Then, possibly move things over, but not if | |
1500 | * subgroups may contain processes, which is the case | |
1501 | * for slice and delegation units. */ | |
1502 | r = cg_migrate_everywhere(u->manager->cgroup_supported, u->cgroup_path, u->cgroup_path, migrate_callback, u); | |
1503 | if (r < 0) | |
efdb0237 | 1504 | log_unit_warning_errno(u, r, "Failed to migrate cgroup from to %s, ignoring: %m", u->cgroup_path); |
0cd385d3 | 1505 | } |
03b90d4b | 1506 | |
64747e2d LP |
1507 | return 0; |
1508 | } | |
1509 | ||
6592b975 LP |
1510 | static int unit_attach_pid_to_cgroup_via_bus(Unit *u, pid_t pid, const char *suffix_path) { |
1511 | _cleanup_(sd_bus_error_free) sd_bus_error error = SD_BUS_ERROR_NULL; | |
1512 | char *pp; | |
7b3fd631 | 1513 | int r; |
6592b975 | 1514 | |
7b3fd631 LP |
1515 | assert(u); |
1516 | ||
6592b975 LP |
1517 | if (MANAGER_IS_SYSTEM(u->manager)) |
1518 | return -EINVAL; | |
1519 | ||
1520 | if (!u->manager->system_bus) | |
1521 | return -EIO; | |
1522 | ||
1523 | if (!u->cgroup_path) | |
1524 | return -EINVAL; | |
1525 | ||
1526 | /* Determine this unit's cgroup path relative to our cgroup root */ | |
1527 | pp = path_startswith(u->cgroup_path, u->manager->cgroup_root); | |
1528 | if (!pp) | |
1529 | return -EINVAL; | |
1530 | ||
1531 | pp = strjoina("/", pp, suffix_path); | |
1532 | path_kill_slashes(pp); | |
1533 | ||
1534 | r = sd_bus_call_method(u->manager->system_bus, | |
1535 | "org.freedesktop.systemd1", | |
1536 | "/org/freedesktop/systemd1", | |
1537 | "org.freedesktop.systemd1.Manager", | |
1538 | "AttachProcessesToUnit", | |
1539 | &error, NULL, | |
1540 | "ssau", | |
1541 | NULL /* empty unit name means client's unit, i.e. us */, pp, 1, (uint32_t) pid); | |
7b3fd631 | 1542 | if (r < 0) |
6592b975 LP |
1543 | return log_unit_debug_errno(u, r, "Failed to attach unit process " PID_FMT " via the bus: %s", pid, bus_error_message(&error, r)); |
1544 | ||
1545 | return 0; | |
1546 | } | |
1547 | ||
1548 | int unit_attach_pids_to_cgroup(Unit *u, Set *pids, const char *suffix_path) { | |
1549 | CGroupMask delegated_mask; | |
1550 | const char *p; | |
1551 | Iterator i; | |
1552 | void *pidp; | |
1553 | int r, q; | |
1554 | ||
1555 | assert(u); | |
1556 | ||
1557 | if (!UNIT_HAS_CGROUP_CONTEXT(u)) | |
1558 | return -EINVAL; | |
1559 | ||
1560 | if (set_isempty(pids)) | |
1561 | return 0; | |
7b3fd631 | 1562 | |
6592b975 | 1563 | r = unit_realize_cgroup(u); |
7b3fd631 LP |
1564 | if (r < 0) |
1565 | return r; | |
1566 | ||
6592b975 LP |
1567 | if (isempty(suffix_path)) |
1568 | p = u->cgroup_path; | |
1569 | else | |
1570 | p = strjoina(u->cgroup_path, "/", suffix_path); | |
1571 | ||
1572 | delegated_mask = unit_get_delegate_mask(u); | |
1573 | ||
1574 | r = 0; | |
1575 | SET_FOREACH(pidp, pids, i) { | |
1576 | pid_t pid = PTR_TO_PID(pidp); | |
1577 | CGroupController c; | |
1578 | ||
1579 | /* First, attach the PID to the main cgroup hierarchy */ | |
1580 | q = cg_attach(SYSTEMD_CGROUP_CONTROLLER, p, pid); | |
1581 | if (q < 0) { | |
1582 | log_unit_debug_errno(u, q, "Couldn't move process " PID_FMT " to requested cgroup '%s': %m", pid, p); | |
1583 | ||
1584 | if (MANAGER_IS_USER(u->manager) && IN_SET(q, -EPERM, -EACCES)) { | |
1585 | int z; | |
1586 | ||
1587 | /* If we are in a user instance, and we can't move the process ourselves due to | |
1588 | * permission problems, let's ask the system instance about it instead. Since it's more | |
1589 | * privileged it might be able to move the process across the leaves of a subtree who's | |
1590 | * top node is not owned by us. */ | |
1591 | ||
1592 | z = unit_attach_pid_to_cgroup_via_bus(u, pid, suffix_path); | |
1593 | if (z < 0) | |
1594 | log_unit_debug_errno(u, z, "Couldn't move process " PID_FMT " to requested cgroup '%s' via the system bus either: %m", pid, p); | |
1595 | else | |
1596 | continue; /* When the bus thing worked via the bus we are fully done for this PID. */ | |
1597 | } | |
1598 | ||
1599 | if (r >= 0) | |
1600 | r = q; /* Remember first error */ | |
1601 | ||
1602 | continue; | |
1603 | } | |
1604 | ||
1605 | q = cg_all_unified(); | |
1606 | if (q < 0) | |
1607 | return q; | |
1608 | if (q > 0) | |
1609 | continue; | |
1610 | ||
1611 | /* In the legacy hierarchy, attach the process to the request cgroup if possible, and if not to the | |
1612 | * innermost realized one */ | |
1613 | ||
1614 | for (c = 0; c < _CGROUP_CONTROLLER_MAX; c++) { | |
1615 | CGroupMask bit = CGROUP_CONTROLLER_TO_MASK(c); | |
1616 | const char *realized; | |
1617 | ||
1618 | if (!(u->manager->cgroup_supported & bit)) | |
1619 | continue; | |
1620 | ||
1621 | /* If this controller is delegated and realized, honour the caller's request for the cgroup suffix. */ | |
1622 | if (delegated_mask & u->cgroup_realized_mask & bit) { | |
1623 | q = cg_attach(cgroup_controller_to_string(c), p, pid); | |
1624 | if (q >= 0) | |
1625 | continue; /* Success! */ | |
1626 | ||
1627 | log_unit_debug_errno(u, q, "Failed to attach PID " PID_FMT " to requested cgroup %s in controller %s, falling back to unit's cgroup: %m", | |
1628 | pid, p, cgroup_controller_to_string(c)); | |
1629 | } | |
1630 | ||
1631 | /* So this controller is either not delegate or realized, or something else weird happened. In | |
1632 | * that case let's attach the PID at least to the closest cgroup up the tree that is | |
1633 | * realized. */ | |
1634 | realized = unit_get_realized_cgroup_path(u, bit); | |
1635 | if (!realized) | |
1636 | continue; /* Not even realized in the root slice? Then let's not bother */ | |
1637 | ||
1638 | q = cg_attach(cgroup_controller_to_string(c), realized, pid); | |
1639 | if (q < 0) | |
1640 | log_unit_debug_errno(u, q, "Failed to attach PID " PID_FMT " to realized cgroup %s in controller %s, ignoring: %m", | |
1641 | pid, realized, cgroup_controller_to_string(c)); | |
1642 | } | |
1643 | } | |
1644 | ||
1645 | return r; | |
7b3fd631 LP |
1646 | } |
1647 | ||
4b58153d LP |
1648 | static void cgroup_xattr_apply(Unit *u) { |
1649 | char ids[SD_ID128_STRING_MAX]; | |
1650 | int r; | |
1651 | ||
1652 | assert(u); | |
1653 | ||
1654 | if (!MANAGER_IS_SYSTEM(u->manager)) | |
1655 | return; | |
1656 | ||
1657 | if (sd_id128_is_null(u->invocation_id)) | |
1658 | return; | |
1659 | ||
1660 | r = cg_set_xattr(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path, | |
1661 | "trusted.invocation_id", | |
1662 | sd_id128_to_string(u->invocation_id, ids), 32, | |
1663 | 0); | |
1664 | if (r < 0) | |
0fb84499 | 1665 | log_unit_debug_errno(u, r, "Failed to set invocation ID on control group %s, ignoring: %m", u->cgroup_path); |
4b58153d LP |
1666 | } |
1667 | ||
906c06f6 DM |
1668 | static bool unit_has_mask_realized( |
1669 | Unit *u, | |
1670 | CGroupMask target_mask, | |
1671 | CGroupMask enable_mask, | |
1672 | bool needs_bpf) { | |
1673 | ||
bc432dc7 LP |
1674 | assert(u); |
1675 | ||
906c06f6 DM |
1676 | return u->cgroup_realized && |
1677 | u->cgroup_realized_mask == target_mask && | |
1678 | u->cgroup_enabled_mask == enable_mask && | |
1679 | ((needs_bpf && u->cgroup_bpf_state == UNIT_CGROUP_BPF_ON) || | |
1680 | (!needs_bpf && u->cgroup_bpf_state == UNIT_CGROUP_BPF_OFF)); | |
6414b7c9 DS |
1681 | } |
1682 | ||
2aa57a65 LP |
1683 | static void unit_add_to_cgroup_realize_queue(Unit *u) { |
1684 | assert(u); | |
1685 | ||
1686 | if (u->in_cgroup_realize_queue) | |
1687 | return; | |
1688 | ||
1689 | LIST_PREPEND(cgroup_realize_queue, u->manager->cgroup_realize_queue, u); | |
1690 | u->in_cgroup_realize_queue = true; | |
1691 | } | |
1692 | ||
1693 | static void unit_remove_from_cgroup_realize_queue(Unit *u) { | |
1694 | assert(u); | |
1695 | ||
1696 | if (!u->in_cgroup_realize_queue) | |
1697 | return; | |
1698 | ||
1699 | LIST_REMOVE(cgroup_realize_queue, u->manager->cgroup_realize_queue, u); | |
1700 | u->in_cgroup_realize_queue = false; | |
1701 | } | |
1702 | ||
1703 | ||
6414b7c9 DS |
1704 | /* Check if necessary controllers and attributes for a unit are in place. |
1705 | * | |
1706 | * If so, do nothing. | |
1707 | * If not, create paths, move processes over, and set attributes. | |
1708 | * | |
1709 | * Returns 0 on success and < 0 on failure. */ | |
db785129 | 1710 | static int unit_realize_cgroup_now(Unit *u, ManagerState state) { |
efdb0237 | 1711 | CGroupMask target_mask, enable_mask; |
906c06f6 | 1712 | bool needs_bpf, apply_bpf; |
6414b7c9 | 1713 | int r; |
64747e2d | 1714 | |
4ad49000 | 1715 | assert(u); |
64747e2d | 1716 | |
2aa57a65 | 1717 | unit_remove_from_cgroup_realize_queue(u); |
64747e2d | 1718 | |
efdb0237 | 1719 | target_mask = unit_get_target_mask(u); |
ccf78df1 | 1720 | enable_mask = unit_get_enable_mask(u); |
906c06f6 | 1721 | needs_bpf = unit_get_needs_bpf(u); |
ccf78df1 | 1722 | |
906c06f6 | 1723 | if (unit_has_mask_realized(u, target_mask, enable_mask, needs_bpf)) |
0a1eb06d | 1724 | return 0; |
64747e2d | 1725 | |
906c06f6 DM |
1726 | /* Make sure we apply the BPF filters either when one is configured, or if none is configured but previously |
1727 | * the state was anything but off. This way, if a unit with a BPF filter applied is reconfigured to lose it | |
1728 | * this will trickle down properly to cgroupfs. */ | |
1729 | apply_bpf = needs_bpf || u->cgroup_bpf_state != UNIT_CGROUP_BPF_OFF; | |
1730 | ||
4ad49000 | 1731 | /* First, realize parents */ |
6414b7c9 | 1732 | if (UNIT_ISSET(u->slice)) { |
db785129 | 1733 | r = unit_realize_cgroup_now(UNIT_DEREF(u->slice), state); |
6414b7c9 DS |
1734 | if (r < 0) |
1735 | return r; | |
1736 | } | |
4ad49000 LP |
1737 | |
1738 | /* And then do the real work */ | |
906c06f6 | 1739 | r = unit_create_cgroup(u, target_mask, enable_mask, needs_bpf); |
6414b7c9 DS |
1740 | if (r < 0) |
1741 | return r; | |
1742 | ||
1743 | /* Finally, apply the necessary attributes. */ | |
906c06f6 | 1744 | cgroup_context_apply(u, target_mask, apply_bpf, state); |
4b58153d | 1745 | cgroup_xattr_apply(u); |
6414b7c9 DS |
1746 | |
1747 | return 0; | |
64747e2d LP |
1748 | } |
1749 | ||
91a6073e | 1750 | unsigned manager_dispatch_cgroup_realize_queue(Manager *m) { |
db785129 | 1751 | ManagerState state; |
4ad49000 | 1752 | unsigned n = 0; |
db785129 | 1753 | Unit *i; |
6414b7c9 | 1754 | int r; |
ecedd90f | 1755 | |
91a6073e LP |
1756 | assert(m); |
1757 | ||
db785129 LP |
1758 | state = manager_state(m); |
1759 | ||
91a6073e LP |
1760 | while ((i = m->cgroup_realize_queue)) { |
1761 | assert(i->in_cgroup_realize_queue); | |
ecedd90f | 1762 | |
2aa57a65 LP |
1763 | if (UNIT_IS_INACTIVE_OR_FAILED(unit_active_state(i))) { |
1764 | /* Maybe things changed, and the unit is not actually active anymore? */ | |
1765 | unit_remove_from_cgroup_realize_queue(i); | |
1766 | continue; | |
1767 | } | |
1768 | ||
db785129 | 1769 | r = unit_realize_cgroup_now(i, state); |
6414b7c9 | 1770 | if (r < 0) |
efdb0237 | 1771 | log_warning_errno(r, "Failed to realize cgroups for queued unit %s, ignoring: %m", i->id); |
0a1eb06d | 1772 | |
4ad49000 LP |
1773 | n++; |
1774 | } | |
ecedd90f | 1775 | |
4ad49000 | 1776 | return n; |
8e274523 LP |
1777 | } |
1778 | ||
91a6073e | 1779 | static void unit_add_siblings_to_cgroup_realize_queue(Unit *u) { |
4ad49000 | 1780 | Unit *slice; |
ca949c9d | 1781 | |
4ad49000 LP |
1782 | /* This adds the siblings of the specified unit and the |
1783 | * siblings of all parent units to the cgroup queue. (But | |
1784 | * neither the specified unit itself nor the parents.) */ | |
1785 | ||
1786 | while ((slice = UNIT_DEREF(u->slice))) { | |
1787 | Iterator i; | |
1788 | Unit *m; | |
eef85c4a | 1789 | void *v; |
8f53a7b8 | 1790 | |
eef85c4a | 1791 | HASHMAP_FOREACH_KEY(v, m, u->dependencies[UNIT_BEFORE], i) { |
4ad49000 LP |
1792 | if (m == u) |
1793 | continue; | |
8e274523 | 1794 | |
6414b7c9 DS |
1795 | /* Skip units that have a dependency on the slice |
1796 | * but aren't actually in it. */ | |
4ad49000 | 1797 | if (UNIT_DEREF(m->slice) != slice) |
50159e6a | 1798 | continue; |
8e274523 | 1799 | |
6414b7c9 DS |
1800 | /* No point in doing cgroup application for units |
1801 | * without active processes. */ | |
1802 | if (UNIT_IS_INACTIVE_OR_FAILED(unit_active_state(m))) | |
1803 | continue; | |
1804 | ||
1805 | /* If the unit doesn't need any new controllers | |
1806 | * and has current ones realized, it doesn't need | |
1807 | * any changes. */ | |
906c06f6 DM |
1808 | if (unit_has_mask_realized(m, |
1809 | unit_get_target_mask(m), | |
1810 | unit_get_enable_mask(m), | |
1811 | unit_get_needs_bpf(m))) | |
6414b7c9 DS |
1812 | continue; |
1813 | ||
91a6073e | 1814 | unit_add_to_cgroup_realize_queue(m); |
50159e6a LP |
1815 | } |
1816 | ||
4ad49000 | 1817 | u = slice; |
8e274523 | 1818 | } |
4ad49000 LP |
1819 | } |
1820 | ||
0a1eb06d | 1821 | int unit_realize_cgroup(Unit *u) { |
4ad49000 LP |
1822 | assert(u); |
1823 | ||
35b7ff80 | 1824 | if (!UNIT_HAS_CGROUP_CONTEXT(u)) |
0a1eb06d | 1825 | return 0; |
8e274523 | 1826 | |
4ad49000 LP |
1827 | /* So, here's the deal: when realizing the cgroups for this |
1828 | * unit, we need to first create all parents, but there's more | |
1829 | * actually: for the weight-based controllers we also need to | |
1830 | * make sure that all our siblings (i.e. units that are in the | |
73e231ab | 1831 | * same slice as we are) have cgroups, too. Otherwise, things |
4ad49000 LP |
1832 | * would become very uneven as each of their processes would |
1833 | * get as much resources as all our group together. This call | |
1834 | * will synchronously create the parent cgroups, but will | |
1835 | * defer work on the siblings to the next event loop | |
1836 | * iteration. */ | |
ca949c9d | 1837 | |
4ad49000 | 1838 | /* Add all sibling slices to the cgroup queue. */ |
91a6073e | 1839 | unit_add_siblings_to_cgroup_realize_queue(u); |
4ad49000 | 1840 | |
6414b7c9 | 1841 | /* And realize this one now (and apply the values) */ |
db785129 | 1842 | return unit_realize_cgroup_now(u, manager_state(u->manager)); |
8e274523 LP |
1843 | } |
1844 | ||
efdb0237 LP |
1845 | void unit_release_cgroup(Unit *u) { |
1846 | assert(u); | |
1847 | ||
1848 | /* Forgets all cgroup details for this cgroup */ | |
1849 | ||
1850 | if (u->cgroup_path) { | |
1851 | (void) hashmap_remove(u->manager->cgroup_unit, u->cgroup_path); | |
1852 | u->cgroup_path = mfree(u->cgroup_path); | |
1853 | } | |
1854 | ||
1855 | if (u->cgroup_inotify_wd >= 0) { | |
1856 | if (inotify_rm_watch(u->manager->cgroup_inotify_fd, u->cgroup_inotify_wd) < 0) | |
1857 | log_unit_debug_errno(u, errno, "Failed to remove cgroup inotify watch %i for %s, ignoring", u->cgroup_inotify_wd, u->id); | |
1858 | ||
1859 | (void) hashmap_remove(u->manager->cgroup_inotify_wd_unit, INT_TO_PTR(u->cgroup_inotify_wd)); | |
1860 | u->cgroup_inotify_wd = -1; | |
1861 | } | |
1862 | } | |
1863 | ||
1864 | void unit_prune_cgroup(Unit *u) { | |
8e274523 | 1865 | int r; |
efdb0237 | 1866 | bool is_root_slice; |
8e274523 | 1867 | |
4ad49000 | 1868 | assert(u); |
8e274523 | 1869 | |
efdb0237 LP |
1870 | /* Removes the cgroup, if empty and possible, and stops watching it. */ |
1871 | ||
4ad49000 LP |
1872 | if (!u->cgroup_path) |
1873 | return; | |
8e274523 | 1874 | |
fe700f46 LP |
1875 | (void) unit_get_cpu_usage(u, NULL); /* Cache the last CPU usage value before we destroy the cgroup */ |
1876 | ||
efdb0237 LP |
1877 | is_root_slice = unit_has_name(u, SPECIAL_ROOT_SLICE); |
1878 | ||
1879 | r = cg_trim_everywhere(u->manager->cgroup_supported, u->cgroup_path, !is_root_slice); | |
dab5bf85 | 1880 | if (r < 0) { |
f29ff115 | 1881 | log_unit_debug_errno(u, r, "Failed to destroy cgroup %s, ignoring: %m", u->cgroup_path); |
dab5bf85 RL |
1882 | return; |
1883 | } | |
8e274523 | 1884 | |
efdb0237 LP |
1885 | if (is_root_slice) |
1886 | return; | |
1887 | ||
1888 | unit_release_cgroup(u); | |
0a1eb06d | 1889 | |
4ad49000 | 1890 | u->cgroup_realized = false; |
bc432dc7 | 1891 | u->cgroup_realized_mask = 0; |
ccf78df1 | 1892 | u->cgroup_enabled_mask = 0; |
8e274523 LP |
1893 | } |
1894 | ||
efdb0237 | 1895 | int unit_search_main_pid(Unit *u, pid_t *ret) { |
4ad49000 LP |
1896 | _cleanup_fclose_ FILE *f = NULL; |
1897 | pid_t pid = 0, npid, mypid; | |
efdb0237 | 1898 | int r; |
4ad49000 LP |
1899 | |
1900 | assert(u); | |
efdb0237 | 1901 | assert(ret); |
4ad49000 LP |
1902 | |
1903 | if (!u->cgroup_path) | |
efdb0237 | 1904 | return -ENXIO; |
4ad49000 | 1905 | |
efdb0237 LP |
1906 | r = cg_enumerate_processes(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path, &f); |
1907 | if (r < 0) | |
1908 | return r; | |
4ad49000 | 1909 | |
df0ff127 | 1910 | mypid = getpid_cached(); |
4ad49000 LP |
1911 | while (cg_read_pid(f, &npid) > 0) { |
1912 | pid_t ppid; | |
1913 | ||
1914 | if (npid == pid) | |
1915 | continue; | |
8e274523 | 1916 | |
4ad49000 | 1917 | /* Ignore processes that aren't our kids */ |
6bc73acb | 1918 | if (get_process_ppid(npid, &ppid) >= 0 && ppid != mypid) |
4ad49000 | 1919 | continue; |
8e274523 | 1920 | |
efdb0237 | 1921 | if (pid != 0) |
4ad49000 LP |
1922 | /* Dang, there's more than one daemonized PID |
1923 | in this group, so we don't know what process | |
1924 | is the main process. */ | |
efdb0237 LP |
1925 | |
1926 | return -ENODATA; | |
8e274523 | 1927 | |
4ad49000 | 1928 | pid = npid; |
8e274523 LP |
1929 | } |
1930 | ||
efdb0237 LP |
1931 | *ret = pid; |
1932 | return 0; | |
1933 | } | |
1934 | ||
1935 | static int unit_watch_pids_in_path(Unit *u, const char *path) { | |
b3c5bad3 | 1936 | _cleanup_closedir_ DIR *d = NULL; |
efdb0237 LP |
1937 | _cleanup_fclose_ FILE *f = NULL; |
1938 | int ret = 0, r; | |
1939 | ||
1940 | assert(u); | |
1941 | assert(path); | |
1942 | ||
1943 | r = cg_enumerate_processes(SYSTEMD_CGROUP_CONTROLLER, path, &f); | |
1944 | if (r < 0) | |
1945 | ret = r; | |
1946 | else { | |
1947 | pid_t pid; | |
1948 | ||
1949 | while ((r = cg_read_pid(f, &pid)) > 0) { | |
1950 | r = unit_watch_pid(u, pid); | |
1951 | if (r < 0 && ret >= 0) | |
1952 | ret = r; | |
1953 | } | |
1954 | ||
1955 | if (r < 0 && ret >= 0) | |
1956 | ret = r; | |
1957 | } | |
1958 | ||
1959 | r = cg_enumerate_subgroups(SYSTEMD_CGROUP_CONTROLLER, path, &d); | |
1960 | if (r < 0) { | |
1961 | if (ret >= 0) | |
1962 | ret = r; | |
1963 | } else { | |
1964 | char *fn; | |
1965 | ||
1966 | while ((r = cg_read_subgroup(d, &fn)) > 0) { | |
1967 | _cleanup_free_ char *p = NULL; | |
1968 | ||
605405c6 | 1969 | p = strjoin(path, "/", fn); |
efdb0237 LP |
1970 | free(fn); |
1971 | ||
1972 | if (!p) | |
1973 | return -ENOMEM; | |
1974 | ||
1975 | r = unit_watch_pids_in_path(u, p); | |
1976 | if (r < 0 && ret >= 0) | |
1977 | ret = r; | |
1978 | } | |
1979 | ||
1980 | if (r < 0 && ret >= 0) | |
1981 | ret = r; | |
1982 | } | |
1983 | ||
1984 | return ret; | |
1985 | } | |
1986 | ||
11aef522 LP |
1987 | int unit_synthesize_cgroup_empty_event(Unit *u) { |
1988 | int r; | |
1989 | ||
1990 | assert(u); | |
1991 | ||
1992 | /* Enqueue a synthetic cgroup empty event if this unit doesn't watch any PIDs anymore. This is compatibility | |
1993 | * support for non-unified systems where notifications aren't reliable, and hence need to take whatever we can | |
1994 | * get as notification source as soon as we stopped having any useful PIDs to watch for. */ | |
1995 | ||
1996 | if (!u->cgroup_path) | |
1997 | return -ENOENT; | |
1998 | ||
1999 | r = cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER); | |
2000 | if (r < 0) | |
2001 | return r; | |
2002 | if (r > 0) /* On unified we have reliable notifications, and don't need this */ | |
2003 | return 0; | |
2004 | ||
2005 | if (!set_isempty(u->pids)) | |
2006 | return 0; | |
2007 | ||
2008 | unit_add_to_cgroup_empty_queue(u); | |
2009 | return 0; | |
2010 | } | |
2011 | ||
efdb0237 | 2012 | int unit_watch_all_pids(Unit *u) { |
b4cccbc1 LP |
2013 | int r; |
2014 | ||
efdb0237 LP |
2015 | assert(u); |
2016 | ||
2017 | /* Adds all PIDs from our cgroup to the set of PIDs we | |
2018 | * watch. This is a fallback logic for cases where we do not | |
2019 | * get reliable cgroup empty notifications: we try to use | |
2020 | * SIGCHLD as replacement. */ | |
2021 | ||
2022 | if (!u->cgroup_path) | |
2023 | return -ENOENT; | |
2024 | ||
c22800e4 | 2025 | r = cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER); |
b4cccbc1 LP |
2026 | if (r < 0) |
2027 | return r; | |
2028 | if (r > 0) /* On unified we can use proper notifications */ | |
efdb0237 LP |
2029 | return 0; |
2030 | ||
2031 | return unit_watch_pids_in_path(u, u->cgroup_path); | |
2032 | } | |
2033 | ||
09e24654 LP |
2034 | static int on_cgroup_empty_event(sd_event_source *s, void *userdata) { |
2035 | Manager *m = userdata; | |
2036 | Unit *u; | |
efdb0237 LP |
2037 | int r; |
2038 | ||
09e24654 LP |
2039 | assert(s); |
2040 | assert(m); | |
efdb0237 | 2041 | |
09e24654 LP |
2042 | u = m->cgroup_empty_queue; |
2043 | if (!u) | |
efdb0237 LP |
2044 | return 0; |
2045 | ||
09e24654 LP |
2046 | assert(u->in_cgroup_empty_queue); |
2047 | u->in_cgroup_empty_queue = false; | |
2048 | LIST_REMOVE(cgroup_empty_queue, m->cgroup_empty_queue, u); | |
2049 | ||
2050 | if (m->cgroup_empty_queue) { | |
2051 | /* More stuff queued, let's make sure we remain enabled */ | |
2052 | r = sd_event_source_set_enabled(s, SD_EVENT_ONESHOT); | |
2053 | if (r < 0) | |
2054 | log_debug_errno(r, "Failed to reenable cgroup empty event source: %m"); | |
2055 | } | |
efdb0237 LP |
2056 | |
2057 | unit_add_to_gc_queue(u); | |
2058 | ||
2059 | if (UNIT_VTABLE(u)->notify_cgroup_empty) | |
2060 | UNIT_VTABLE(u)->notify_cgroup_empty(u); | |
2061 | ||
2062 | return 0; | |
2063 | } | |
2064 | ||
09e24654 LP |
2065 | void unit_add_to_cgroup_empty_queue(Unit *u) { |
2066 | int r; | |
2067 | ||
2068 | assert(u); | |
2069 | ||
2070 | /* Note that there are four different ways how cgroup empty events reach us: | |
2071 | * | |
2072 | * 1. On the unified hierarchy we get an inotify event on the cgroup | |
2073 | * | |
2074 | * 2. On the legacy hierarchy, when running in system mode, we get a datagram on the cgroup agent socket | |
2075 | * | |
2076 | * 3. On the legacy hierarchy, when running in user mode, we get a D-Bus signal on the system bus | |
2077 | * | |
2078 | * 4. On the legacy hierarchy, in service units we start watching all processes of the cgroup for SIGCHLD as | |
2079 | * soon as we get one SIGCHLD, to deal with unreliable cgroup notifications. | |
2080 | * | |
2081 | * Regardless which way we got the notification, we'll verify it here, and then add it to a separate | |
2082 | * queue. This queue will be dispatched at a lower priority than the SIGCHLD handler, so that we always use | |
2083 | * SIGCHLD if we can get it first, and only use the cgroup empty notifications if there's no SIGCHLD pending | |
2084 | * (which might happen if the cgroup doesn't contain processes that are our own child, which is typically the | |
2085 | * case for scope units). */ | |
2086 | ||
2087 | if (u->in_cgroup_empty_queue) | |
2088 | return; | |
2089 | ||
2090 | /* Let's verify that the cgroup is really empty */ | |
2091 | if (!u->cgroup_path) | |
2092 | return; | |
2093 | r = cg_is_empty_recursive(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path); | |
2094 | if (r < 0) { | |
2095 | log_unit_debug_errno(u, r, "Failed to determine whether cgroup %s is empty: %m", u->cgroup_path); | |
2096 | return; | |
2097 | } | |
2098 | if (r == 0) | |
2099 | return; | |
2100 | ||
2101 | LIST_PREPEND(cgroup_empty_queue, u->manager->cgroup_empty_queue, u); | |
2102 | u->in_cgroup_empty_queue = true; | |
2103 | ||
2104 | /* Trigger the defer event */ | |
2105 | r = sd_event_source_set_enabled(u->manager->cgroup_empty_event_source, SD_EVENT_ONESHOT); | |
2106 | if (r < 0) | |
2107 | log_debug_errno(r, "Failed to enable cgroup empty event source: %m"); | |
2108 | } | |
2109 | ||
efdb0237 LP |
2110 | static int on_cgroup_inotify_event(sd_event_source *s, int fd, uint32_t revents, void *userdata) { |
2111 | Manager *m = userdata; | |
2112 | ||
2113 | assert(s); | |
2114 | assert(fd >= 0); | |
2115 | assert(m); | |
2116 | ||
2117 | for (;;) { | |
2118 | union inotify_event_buffer buffer; | |
2119 | struct inotify_event *e; | |
2120 | ssize_t l; | |
2121 | ||
2122 | l = read(fd, &buffer, sizeof(buffer)); | |
2123 | if (l < 0) { | |
47249640 | 2124 | if (IN_SET(errno, EINTR, EAGAIN)) |
efdb0237 LP |
2125 | return 0; |
2126 | ||
2127 | return log_error_errno(errno, "Failed to read control group inotify events: %m"); | |
2128 | } | |
2129 | ||
2130 | FOREACH_INOTIFY_EVENT(e, buffer, l) { | |
2131 | Unit *u; | |
2132 | ||
2133 | if (e->wd < 0) | |
2134 | /* Queue overflow has no watch descriptor */ | |
2135 | continue; | |
2136 | ||
2137 | if (e->mask & IN_IGNORED) | |
2138 | /* The watch was just removed */ | |
2139 | continue; | |
2140 | ||
2141 | u = hashmap_get(m->cgroup_inotify_wd_unit, INT_TO_PTR(e->wd)); | |
2142 | if (!u) /* Not that inotify might deliver | |
2143 | * events for a watch even after it | |
2144 | * was removed, because it was queued | |
2145 | * before the removal. Let's ignore | |
2146 | * this here safely. */ | |
2147 | continue; | |
2148 | ||
09e24654 | 2149 | unit_add_to_cgroup_empty_queue(u); |
efdb0237 LP |
2150 | } |
2151 | } | |
8e274523 LP |
2152 | } |
2153 | ||
8e274523 | 2154 | int manager_setup_cgroup(Manager *m) { |
9444b1f2 | 2155 | _cleanup_free_ char *path = NULL; |
10bd3e2e | 2156 | const char *scope_path; |
efdb0237 | 2157 | CGroupController c; |
b4cccbc1 | 2158 | int r, all_unified; |
efdb0237 | 2159 | char *e; |
8e274523 LP |
2160 | |
2161 | assert(m); | |
2162 | ||
35d2e7ec | 2163 | /* 1. Determine hierarchy */ |
efdb0237 | 2164 | m->cgroup_root = mfree(m->cgroup_root); |
9444b1f2 | 2165 | r = cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER, 0, &m->cgroup_root); |
23bbb0de MS |
2166 | if (r < 0) |
2167 | return log_error_errno(r, "Cannot determine cgroup we are running in: %m"); | |
8e274523 | 2168 | |
efdb0237 LP |
2169 | /* Chop off the init scope, if we are already located in it */ |
2170 | e = endswith(m->cgroup_root, "/" SPECIAL_INIT_SCOPE); | |
0d8c31ff | 2171 | |
efdb0237 LP |
2172 | /* LEGACY: Also chop off the system slice if we are in |
2173 | * it. This is to support live upgrades from older systemd | |
2174 | * versions where PID 1 was moved there. Also see | |
2175 | * cg_get_root_path(). */ | |
463d0d15 | 2176 | if (!e && MANAGER_IS_SYSTEM(m)) { |
9444b1f2 | 2177 | e = endswith(m->cgroup_root, "/" SPECIAL_SYSTEM_SLICE); |
15c60e99 | 2178 | if (!e) |
efdb0237 | 2179 | e = endswith(m->cgroup_root, "/system"); /* even more legacy */ |
0baf24dd | 2180 | } |
efdb0237 LP |
2181 | if (e) |
2182 | *e = 0; | |
7ccfb64a | 2183 | |
7546145e LP |
2184 | /* And make sure to store away the root value without trailing slash, even for the root dir, so that we can |
2185 | * easily prepend it everywhere. */ | |
2186 | delete_trailing_chars(m->cgroup_root, "/"); | |
8e274523 | 2187 | |
35d2e7ec | 2188 | /* 2. Show data */ |
9444b1f2 | 2189 | r = cg_get_path(SYSTEMD_CGROUP_CONTROLLER, m->cgroup_root, NULL, &path); |
23bbb0de MS |
2190 | if (r < 0) |
2191 | return log_error_errno(r, "Cannot find cgroup mount point: %m"); | |
8e274523 | 2192 | |
415fc41c TH |
2193 | r = cg_unified_flush(); |
2194 | if (r < 0) | |
2195 | return log_error_errno(r, "Couldn't determine if we are running in the unified hierarchy: %m"); | |
5da38d07 | 2196 | |
b4cccbc1 | 2197 | all_unified = cg_all_unified(); |
d4c819ed ZJS |
2198 | if (all_unified < 0) |
2199 | return log_error_errno(all_unified, "Couldn't determine whether we are in all unified mode: %m"); | |
2200 | if (all_unified > 0) | |
efdb0237 | 2201 | log_debug("Unified cgroup hierarchy is located at %s.", path); |
b4cccbc1 | 2202 | else { |
c22800e4 | 2203 | r = cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER); |
b4cccbc1 LP |
2204 | if (r < 0) |
2205 | return log_error_errno(r, "Failed to determine whether systemd's own controller is in unified mode: %m"); | |
2206 | if (r > 0) | |
2207 | log_debug("Unified cgroup hierarchy is located at %s. Controllers are on legacy hierarchies.", path); | |
2208 | else | |
2209 | log_debug("Using cgroup controller " SYSTEMD_CGROUP_CONTROLLER_LEGACY ". File system hierarchy is at %s.", path); | |
2210 | } | |
efdb0237 | 2211 | |
09e24654 LP |
2212 | /* 3. Allocate cgroup empty defer event source */ |
2213 | m->cgroup_empty_event_source = sd_event_source_unref(m->cgroup_empty_event_source); | |
2214 | r = sd_event_add_defer(m->event, &m->cgroup_empty_event_source, on_cgroup_empty_event, m); | |
2215 | if (r < 0) | |
2216 | return log_error_errno(r, "Failed to create cgroup empty event source: %m"); | |
2217 | ||
2218 | r = sd_event_source_set_priority(m->cgroup_empty_event_source, SD_EVENT_PRIORITY_NORMAL-5); | |
2219 | if (r < 0) | |
2220 | return log_error_errno(r, "Failed to set priority of cgroup empty event source: %m"); | |
2221 | ||
2222 | r = sd_event_source_set_enabled(m->cgroup_empty_event_source, SD_EVENT_OFF); | |
2223 | if (r < 0) | |
2224 | return log_error_errno(r, "Failed to disable cgroup empty event source: %m"); | |
2225 | ||
2226 | (void) sd_event_source_set_description(m->cgroup_empty_event_source, "cgroup-empty"); | |
2227 | ||
2228 | /* 4. Install notifier inotify object, or agent */ | |
10bd3e2e | 2229 | if (cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER) > 0) { |
c6c18be3 | 2230 | |
09e24654 | 2231 | /* In the unified hierarchy we can get cgroup empty notifications via inotify. */ |
efdb0237 | 2232 | |
10bd3e2e LP |
2233 | m->cgroup_inotify_event_source = sd_event_source_unref(m->cgroup_inotify_event_source); |
2234 | safe_close(m->cgroup_inotify_fd); | |
efdb0237 | 2235 | |
10bd3e2e LP |
2236 | m->cgroup_inotify_fd = inotify_init1(IN_NONBLOCK|IN_CLOEXEC); |
2237 | if (m->cgroup_inotify_fd < 0) | |
2238 | return log_error_errno(errno, "Failed to create control group inotify object: %m"); | |
efdb0237 | 2239 | |
10bd3e2e LP |
2240 | r = sd_event_add_io(m->event, &m->cgroup_inotify_event_source, m->cgroup_inotify_fd, EPOLLIN, on_cgroup_inotify_event, m); |
2241 | if (r < 0) | |
2242 | return log_error_errno(r, "Failed to watch control group inotify object: %m"); | |
efdb0237 | 2243 | |
10bd3e2e LP |
2244 | /* Process cgroup empty notifications early, but after service notifications and SIGCHLD. Also |
2245 | * see handling of cgroup agent notifications, for the classic cgroup hierarchy support. */ | |
09e24654 | 2246 | r = sd_event_source_set_priority(m->cgroup_inotify_event_source, SD_EVENT_PRIORITY_NORMAL-4); |
10bd3e2e LP |
2247 | if (r < 0) |
2248 | return log_error_errno(r, "Failed to set priority of inotify event source: %m"); | |
efdb0237 | 2249 | |
10bd3e2e | 2250 | (void) sd_event_source_set_description(m->cgroup_inotify_event_source, "cgroup-inotify"); |
efdb0237 | 2251 | |
10bd3e2e | 2252 | } else if (MANAGER_IS_SYSTEM(m) && m->test_run_flags == 0) { |
efdb0237 | 2253 | |
10bd3e2e LP |
2254 | /* On the legacy hierarchy we only get notifications via cgroup agents. (Which isn't really reliable, |
2255 | * since it does not generate events when control groups with children run empty. */ | |
8e274523 | 2256 | |
10bd3e2e | 2257 | r = cg_install_release_agent(SYSTEMD_CGROUP_CONTROLLER, SYSTEMD_CGROUP_AGENT_PATH); |
23bbb0de | 2258 | if (r < 0) |
10bd3e2e LP |
2259 | log_warning_errno(r, "Failed to install release agent, ignoring: %m"); |
2260 | else if (r > 0) | |
2261 | log_debug("Installed release agent."); | |
2262 | else if (r == 0) | |
2263 | log_debug("Release agent already installed."); | |
2264 | } | |
efdb0237 | 2265 | |
09e24654 | 2266 | /* 5. Make sure we are in the special "init.scope" unit in the root slice. */ |
10bd3e2e LP |
2267 | scope_path = strjoina(m->cgroup_root, "/" SPECIAL_INIT_SCOPE); |
2268 | r = cg_create_and_attach(SYSTEMD_CGROUP_CONTROLLER, scope_path, 0); | |
2269 | if (r < 0) | |
2270 | return log_error_errno(r, "Failed to create %s control group: %m", scope_path); | |
c6c18be3 | 2271 | |
09e24654 | 2272 | /* Also, move all other userspace processes remaining in the root cgroup into that scope. */ |
10bd3e2e LP |
2273 | r = cg_migrate(SYSTEMD_CGROUP_CONTROLLER, m->cgroup_root, SYSTEMD_CGROUP_CONTROLLER, scope_path, 0); |
2274 | if (r < 0) | |
2275 | log_warning_errno(r, "Couldn't move remaining userspace processes, ignoring: %m"); | |
0d8c31ff | 2276 | |
09e24654 | 2277 | /* 6. And pin it, so that it cannot be unmounted */ |
10bd3e2e LP |
2278 | safe_close(m->pin_cgroupfs_fd); |
2279 | m->pin_cgroupfs_fd = open(path, O_RDONLY|O_CLOEXEC|O_DIRECTORY|O_NOCTTY|O_NONBLOCK); | |
2280 | if (m->pin_cgroupfs_fd < 0) | |
2281 | return log_error_errno(errno, "Failed to open pin file: %m"); | |
2282 | ||
09e24654 | 2283 | /* 7. Always enable hierarchical support if it exists... */ |
10bd3e2e LP |
2284 | if (!all_unified && m->test_run_flags == 0) |
2285 | (void) cg_set_attribute("memory", "/", "memory.use_hierarchy", "1"); | |
c6c18be3 | 2286 | |
09e24654 | 2287 | /* 8. Figure out which controllers are supported, and log about it */ |
efdb0237 LP |
2288 | r = cg_mask_supported(&m->cgroup_supported); |
2289 | if (r < 0) | |
2290 | return log_error_errno(r, "Failed to determine supported controllers: %m"); | |
efdb0237 | 2291 | for (c = 0; c < _CGROUP_CONTROLLER_MAX; c++) |
eee0a1e4 | 2292 | log_debug("Controller '%s' supported: %s", cgroup_controller_to_string(c), yes_no(m->cgroup_supported & CGROUP_CONTROLLER_TO_MASK(c))); |
9156e799 | 2293 | |
a32360f1 | 2294 | return 0; |
8e274523 LP |
2295 | } |
2296 | ||
c6c18be3 | 2297 | void manager_shutdown_cgroup(Manager *m, bool delete) { |
8e274523 LP |
2298 | assert(m); |
2299 | ||
9444b1f2 LP |
2300 | /* We can't really delete the group, since we are in it. But |
2301 | * let's trim it. */ | |
2302 | if (delete && m->cgroup_root) | |
efdb0237 LP |
2303 | (void) cg_trim(SYSTEMD_CGROUP_CONTROLLER, m->cgroup_root, false); |
2304 | ||
09e24654 LP |
2305 | m->cgroup_empty_event_source = sd_event_source_unref(m->cgroup_empty_event_source); |
2306 | ||
efdb0237 LP |
2307 | m->cgroup_inotify_wd_unit = hashmap_free(m->cgroup_inotify_wd_unit); |
2308 | ||
2309 | m->cgroup_inotify_event_source = sd_event_source_unref(m->cgroup_inotify_event_source); | |
2310 | m->cgroup_inotify_fd = safe_close(m->cgroup_inotify_fd); | |
8e274523 | 2311 | |
03e334a1 | 2312 | m->pin_cgroupfs_fd = safe_close(m->pin_cgroupfs_fd); |
c6c18be3 | 2313 | |
efdb0237 | 2314 | m->cgroup_root = mfree(m->cgroup_root); |
8e274523 LP |
2315 | } |
2316 | ||
4ad49000 | 2317 | Unit* manager_get_unit_by_cgroup(Manager *m, const char *cgroup) { |
acb14d31 | 2318 | char *p; |
4ad49000 | 2319 | Unit *u; |
acb14d31 LP |
2320 | |
2321 | assert(m); | |
2322 | assert(cgroup); | |
acb14d31 | 2323 | |
4ad49000 LP |
2324 | u = hashmap_get(m->cgroup_unit, cgroup); |
2325 | if (u) | |
2326 | return u; | |
acb14d31 | 2327 | |
8e70580b | 2328 | p = strdupa(cgroup); |
acb14d31 LP |
2329 | for (;;) { |
2330 | char *e; | |
2331 | ||
2332 | e = strrchr(p, '/'); | |
efdb0237 LP |
2333 | if (!e || e == p) |
2334 | return hashmap_get(m->cgroup_unit, SPECIAL_ROOT_SLICE); | |
acb14d31 LP |
2335 | |
2336 | *e = 0; | |
2337 | ||
4ad49000 LP |
2338 | u = hashmap_get(m->cgroup_unit, p); |
2339 | if (u) | |
2340 | return u; | |
acb14d31 LP |
2341 | } |
2342 | } | |
2343 | ||
b3ac818b | 2344 | Unit *manager_get_unit_by_pid_cgroup(Manager *m, pid_t pid) { |
4ad49000 | 2345 | _cleanup_free_ char *cgroup = NULL; |
8e274523 | 2346 | |
8c47c732 LP |
2347 | assert(m); |
2348 | ||
62a76913 | 2349 | if (!pid_is_valid(pid)) |
b3ac818b LP |
2350 | return NULL; |
2351 | ||
62a76913 | 2352 | if (cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER, pid, &cgroup) < 0) |
b3ac818b LP |
2353 | return NULL; |
2354 | ||
2355 | return manager_get_unit_by_cgroup(m, cgroup); | |
2356 | } | |
2357 | ||
2358 | Unit *manager_get_unit_by_pid(Manager *m, pid_t pid) { | |
62a76913 | 2359 | Unit *u, **array; |
b3ac818b LP |
2360 | |
2361 | assert(m); | |
2362 | ||
62a76913 LP |
2363 | /* Note that a process might be owned by multiple units, we return only one here, which is good enough for most |
2364 | * cases, though not strictly correct. We prefer the one reported by cgroup membership, as that's the most | |
2365 | * relevant one as children of the process will be assigned to that one, too, before all else. */ | |
2366 | ||
2367 | if (!pid_is_valid(pid)) | |
8c47c732 LP |
2368 | return NULL; |
2369 | ||
2ca9d979 | 2370 | if (pid == getpid_cached()) |
efdb0237 LP |
2371 | return hashmap_get(m->units, SPECIAL_INIT_SCOPE); |
2372 | ||
62a76913 | 2373 | u = manager_get_unit_by_pid_cgroup(m, pid); |
5fe8876b LP |
2374 | if (u) |
2375 | return u; | |
2376 | ||
62a76913 | 2377 | u = hashmap_get(m->watch_pids, PID_TO_PTR(pid)); |
5fe8876b LP |
2378 | if (u) |
2379 | return u; | |
2380 | ||
62a76913 LP |
2381 | array = hashmap_get(m->watch_pids, PID_TO_PTR(-pid)); |
2382 | if (array) | |
2383 | return array[0]; | |
2384 | ||
2385 | return NULL; | |
6dde1f33 | 2386 | } |
4fbf50b3 | 2387 | |
4ad49000 LP |
2388 | int manager_notify_cgroup_empty(Manager *m, const char *cgroup) { |
2389 | Unit *u; | |
4fbf50b3 | 2390 | |
4ad49000 LP |
2391 | assert(m); |
2392 | assert(cgroup); | |
4fbf50b3 | 2393 | |
09e24654 LP |
2394 | /* Called on the legacy hierarchy whenever we get an explicit cgroup notification from the cgroup agent process |
2395 | * or from the --system instance */ | |
2396 | ||
d8fdc620 LP |
2397 | log_debug("Got cgroup empty notification for: %s", cgroup); |
2398 | ||
4ad49000 | 2399 | u = manager_get_unit_by_cgroup(m, cgroup); |
5ad096b3 LP |
2400 | if (!u) |
2401 | return 0; | |
b56c28c3 | 2402 | |
09e24654 LP |
2403 | unit_add_to_cgroup_empty_queue(u); |
2404 | return 1; | |
5ad096b3 LP |
2405 | } |
2406 | ||
2407 | int unit_get_memory_current(Unit *u, uint64_t *ret) { | |
2408 | _cleanup_free_ char *v = NULL; | |
2409 | int r; | |
2410 | ||
2411 | assert(u); | |
2412 | assert(ret); | |
2413 | ||
2e4025c0 | 2414 | if (!UNIT_CGROUP_BOOL(u, memory_accounting)) |
cf3b4be1 LP |
2415 | return -ENODATA; |
2416 | ||
5ad096b3 LP |
2417 | if (!u->cgroup_path) |
2418 | return -ENODATA; | |
2419 | ||
efdb0237 | 2420 | if ((u->cgroup_realized_mask & CGROUP_MASK_MEMORY) == 0) |
5ad096b3 LP |
2421 | return -ENODATA; |
2422 | ||
b4cccbc1 LP |
2423 | r = cg_all_unified(); |
2424 | if (r < 0) | |
2425 | return r; | |
2426 | if (r > 0) | |
efdb0237 | 2427 | r = cg_get_attribute("memory", u->cgroup_path, "memory.current", &v); |
b4cccbc1 LP |
2428 | else |
2429 | r = cg_get_attribute("memory", u->cgroup_path, "memory.usage_in_bytes", &v); | |
5ad096b3 LP |
2430 | if (r == -ENOENT) |
2431 | return -ENODATA; | |
2432 | if (r < 0) | |
2433 | return r; | |
2434 | ||
2435 | return safe_atou64(v, ret); | |
2436 | } | |
2437 | ||
03a7b521 LP |
2438 | int unit_get_tasks_current(Unit *u, uint64_t *ret) { |
2439 | _cleanup_free_ char *v = NULL; | |
2440 | int r; | |
2441 | ||
2442 | assert(u); | |
2443 | assert(ret); | |
2444 | ||
2e4025c0 | 2445 | if (!UNIT_CGROUP_BOOL(u, tasks_accounting)) |
cf3b4be1 LP |
2446 | return -ENODATA; |
2447 | ||
03a7b521 LP |
2448 | if (!u->cgroup_path) |
2449 | return -ENODATA; | |
2450 | ||
2451 | if ((u->cgroup_realized_mask & CGROUP_MASK_PIDS) == 0) | |
2452 | return -ENODATA; | |
2453 | ||
c36a69f4 LP |
2454 | /* The root cgroup doesn't expose this information, let's get it from /proc instead */ |
2455 | if (unit_has_root_cgroup(u)) | |
2456 | return procfs_tasks_get_current(ret); | |
2457 | ||
03a7b521 LP |
2458 | r = cg_get_attribute("pids", u->cgroup_path, "pids.current", &v); |
2459 | if (r == -ENOENT) | |
2460 | return -ENODATA; | |
2461 | if (r < 0) | |
2462 | return r; | |
2463 | ||
2464 | return safe_atou64(v, ret); | |
2465 | } | |
2466 | ||
5ad096b3 LP |
2467 | static int unit_get_cpu_usage_raw(Unit *u, nsec_t *ret) { |
2468 | _cleanup_free_ char *v = NULL; | |
2469 | uint64_t ns; | |
2470 | int r; | |
2471 | ||
2472 | assert(u); | |
2473 | assert(ret); | |
2474 | ||
2475 | if (!u->cgroup_path) | |
2476 | return -ENODATA; | |
2477 | ||
b4cccbc1 LP |
2478 | r = cg_all_unified(); |
2479 | if (r < 0) | |
2480 | return r; | |
2481 | if (r > 0) { | |
66ebf6c0 TH |
2482 | const char *keys[] = { "usage_usec", NULL }; |
2483 | _cleanup_free_ char *val = NULL; | |
2484 | uint64_t us; | |
5ad096b3 | 2485 | |
66ebf6c0 TH |
2486 | if ((u->cgroup_realized_mask & CGROUP_MASK_CPU) == 0) |
2487 | return -ENODATA; | |
5ad096b3 | 2488 | |
66ebf6c0 TH |
2489 | r = cg_get_keyed_attribute("cpu", u->cgroup_path, "cpu.stat", keys, &val); |
2490 | if (r < 0) | |
2491 | return r; | |
2492 | ||
2493 | r = safe_atou64(val, &us); | |
2494 | if (r < 0) | |
2495 | return r; | |
2496 | ||
2497 | ns = us * NSEC_PER_USEC; | |
2498 | } else { | |
2499 | if ((u->cgroup_realized_mask & CGROUP_MASK_CPUACCT) == 0) | |
2500 | return -ENODATA; | |
2501 | ||
2502 | r = cg_get_attribute("cpuacct", u->cgroup_path, "cpuacct.usage", &v); | |
2503 | if (r == -ENOENT) | |
2504 | return -ENODATA; | |
2505 | if (r < 0) | |
2506 | return r; | |
2507 | ||
2508 | r = safe_atou64(v, &ns); | |
2509 | if (r < 0) | |
2510 | return r; | |
2511 | } | |
5ad096b3 LP |
2512 | |
2513 | *ret = ns; | |
2514 | return 0; | |
2515 | } | |
2516 | ||
2517 | int unit_get_cpu_usage(Unit *u, nsec_t *ret) { | |
2518 | nsec_t ns; | |
2519 | int r; | |
2520 | ||
fe700f46 LP |
2521 | assert(u); |
2522 | ||
2523 | /* Retrieve the current CPU usage counter. This will subtract the CPU counter taken when the unit was | |
2524 | * started. If the cgroup has been removed already, returns the last cached value. To cache the value, simply | |
2525 | * call this function with a NULL return value. */ | |
2526 | ||
2e4025c0 | 2527 | if (!UNIT_CGROUP_BOOL(u, cpu_accounting)) |
cf3b4be1 LP |
2528 | return -ENODATA; |
2529 | ||
5ad096b3 | 2530 | r = unit_get_cpu_usage_raw(u, &ns); |
fe700f46 LP |
2531 | if (r == -ENODATA && u->cpu_usage_last != NSEC_INFINITY) { |
2532 | /* If we can't get the CPU usage anymore (because the cgroup was already removed, for example), use our | |
2533 | * cached value. */ | |
2534 | ||
2535 | if (ret) | |
2536 | *ret = u->cpu_usage_last; | |
2537 | return 0; | |
2538 | } | |
5ad096b3 LP |
2539 | if (r < 0) |
2540 | return r; | |
2541 | ||
66ebf6c0 TH |
2542 | if (ns > u->cpu_usage_base) |
2543 | ns -= u->cpu_usage_base; | |
5ad096b3 LP |
2544 | else |
2545 | ns = 0; | |
2546 | ||
fe700f46 LP |
2547 | u->cpu_usage_last = ns; |
2548 | if (ret) | |
2549 | *ret = ns; | |
2550 | ||
5ad096b3 LP |
2551 | return 0; |
2552 | } | |
2553 | ||
906c06f6 DM |
2554 | int unit_get_ip_accounting( |
2555 | Unit *u, | |
2556 | CGroupIPAccountingMetric metric, | |
2557 | uint64_t *ret) { | |
2558 | ||
6b659ed8 | 2559 | uint64_t value; |
906c06f6 DM |
2560 | int fd, r; |
2561 | ||
2562 | assert(u); | |
2563 | assert(metric >= 0); | |
2564 | assert(metric < _CGROUP_IP_ACCOUNTING_METRIC_MAX); | |
2565 | assert(ret); | |
2566 | ||
cf3b4be1 LP |
2567 | /* IP accounting is currently not recursive, and hence we refuse to return any data for slice nodes. Slices are |
2568 | * inner cgroup nodes and hence have no processes directly attached, hence their counters would be zero | |
2569 | * anyway. And if we block this now we can later open this up, if the kernel learns recursive BPF cgroup | |
2570 | * filters. */ | |
2571 | if (u->type == UNIT_SLICE) | |
2572 | return -ENODATA; | |
2573 | ||
2e4025c0 | 2574 | if (!UNIT_CGROUP_BOOL(u, ip_accounting)) |
cf3b4be1 LP |
2575 | return -ENODATA; |
2576 | ||
906c06f6 DM |
2577 | fd = IN_SET(metric, CGROUP_IP_INGRESS_BYTES, CGROUP_IP_INGRESS_PACKETS) ? |
2578 | u->ip_accounting_ingress_map_fd : | |
2579 | u->ip_accounting_egress_map_fd; | |
906c06f6 DM |
2580 | if (fd < 0) |
2581 | return -ENODATA; | |
2582 | ||
2583 | if (IN_SET(metric, CGROUP_IP_INGRESS_BYTES, CGROUP_IP_EGRESS_BYTES)) | |
6b659ed8 | 2584 | r = bpf_firewall_read_accounting(fd, &value, NULL); |
906c06f6 | 2585 | else |
6b659ed8 LP |
2586 | r = bpf_firewall_read_accounting(fd, NULL, &value); |
2587 | if (r < 0) | |
2588 | return r; | |
2589 | ||
2590 | /* Add in additional metrics from a previous runtime. Note that when reexecing/reloading the daemon we compile | |
2591 | * all BPF programs and maps anew, but serialize the old counters. When deserializing we store them in the | |
2592 | * ip_accounting_extra[] field, and add them in here transparently. */ | |
2593 | ||
2594 | *ret = value + u->ip_accounting_extra[metric]; | |
906c06f6 DM |
2595 | |
2596 | return r; | |
2597 | } | |
2598 | ||
2599 | int unit_reset_cpu_accounting(Unit *u) { | |
5ad096b3 LP |
2600 | nsec_t ns; |
2601 | int r; | |
2602 | ||
2603 | assert(u); | |
2604 | ||
fe700f46 LP |
2605 | u->cpu_usage_last = NSEC_INFINITY; |
2606 | ||
5ad096b3 LP |
2607 | r = unit_get_cpu_usage_raw(u, &ns); |
2608 | if (r < 0) { | |
66ebf6c0 | 2609 | u->cpu_usage_base = 0; |
5ad096b3 | 2610 | return r; |
b56c28c3 | 2611 | } |
2633eb83 | 2612 | |
66ebf6c0 | 2613 | u->cpu_usage_base = ns; |
4ad49000 | 2614 | return 0; |
4fbf50b3 LP |
2615 | } |
2616 | ||
906c06f6 DM |
2617 | int unit_reset_ip_accounting(Unit *u) { |
2618 | int r = 0, q = 0; | |
2619 | ||
2620 | assert(u); | |
2621 | ||
2622 | if (u->ip_accounting_ingress_map_fd >= 0) | |
2623 | r = bpf_firewall_reset_accounting(u->ip_accounting_ingress_map_fd); | |
2624 | ||
2625 | if (u->ip_accounting_egress_map_fd >= 0) | |
2626 | q = bpf_firewall_reset_accounting(u->ip_accounting_egress_map_fd); | |
2627 | ||
6b659ed8 LP |
2628 | zero(u->ip_accounting_extra); |
2629 | ||
906c06f6 DM |
2630 | return r < 0 ? r : q; |
2631 | } | |
2632 | ||
e7ab4d1a LP |
2633 | void unit_invalidate_cgroup(Unit *u, CGroupMask m) { |
2634 | assert(u); | |
2635 | ||
2636 | if (!UNIT_HAS_CGROUP_CONTEXT(u)) | |
2637 | return; | |
2638 | ||
2639 | if (m == 0) | |
2640 | return; | |
2641 | ||
538b4852 TH |
2642 | /* always invalidate compat pairs together */ |
2643 | if (m & (CGROUP_MASK_IO | CGROUP_MASK_BLKIO)) | |
2644 | m |= CGROUP_MASK_IO | CGROUP_MASK_BLKIO; | |
2645 | ||
7cce4fb7 LP |
2646 | if (m & (CGROUP_MASK_CPU | CGROUP_MASK_CPUACCT)) |
2647 | m |= CGROUP_MASK_CPU | CGROUP_MASK_CPUACCT; | |
2648 | ||
60c728ad | 2649 | if ((u->cgroup_realized_mask & m) == 0) /* NOP? */ |
e7ab4d1a LP |
2650 | return; |
2651 | ||
2652 | u->cgroup_realized_mask &= ~m; | |
91a6073e | 2653 | unit_add_to_cgroup_realize_queue(u); |
e7ab4d1a LP |
2654 | } |
2655 | ||
906c06f6 DM |
2656 | void unit_invalidate_cgroup_bpf(Unit *u) { |
2657 | assert(u); | |
2658 | ||
2659 | if (!UNIT_HAS_CGROUP_CONTEXT(u)) | |
2660 | return; | |
2661 | ||
60c728ad | 2662 | if (u->cgroup_bpf_state == UNIT_CGROUP_BPF_INVALIDATED) /* NOP? */ |
906c06f6 DM |
2663 | return; |
2664 | ||
2665 | u->cgroup_bpf_state = UNIT_CGROUP_BPF_INVALIDATED; | |
91a6073e | 2666 | unit_add_to_cgroup_realize_queue(u); |
906c06f6 DM |
2667 | |
2668 | /* If we are a slice unit, we also need to put compile a new BPF program for all our children, as the IP access | |
2669 | * list of our children includes our own. */ | |
2670 | if (u->type == UNIT_SLICE) { | |
2671 | Unit *member; | |
2672 | Iterator i; | |
eef85c4a | 2673 | void *v; |
906c06f6 | 2674 | |
eef85c4a | 2675 | HASHMAP_FOREACH_KEY(v, member, u->dependencies[UNIT_BEFORE], i) { |
906c06f6 DM |
2676 | if (member == u) |
2677 | continue; | |
2678 | ||
2679 | if (UNIT_DEREF(member->slice) != u) | |
2680 | continue; | |
2681 | ||
2682 | unit_invalidate_cgroup_bpf(member); | |
2683 | } | |
2684 | } | |
2685 | } | |
2686 | ||
1d9cc876 LP |
2687 | bool unit_cgroup_delegate(Unit *u) { |
2688 | CGroupContext *c; | |
2689 | ||
2690 | assert(u); | |
2691 | ||
2692 | if (!UNIT_VTABLE(u)->can_delegate) | |
2693 | return false; | |
2694 | ||
2695 | c = unit_get_cgroup_context(u); | |
2696 | if (!c) | |
2697 | return false; | |
2698 | ||
2699 | return c->delegate; | |
2700 | } | |
2701 | ||
e7ab4d1a LP |
2702 | void manager_invalidate_startup_units(Manager *m) { |
2703 | Iterator i; | |
2704 | Unit *u; | |
2705 | ||
2706 | assert(m); | |
2707 | ||
2708 | SET_FOREACH(u, m->startup_units, i) | |
13c31542 | 2709 | unit_invalidate_cgroup(u, CGROUP_MASK_CPU|CGROUP_MASK_IO|CGROUP_MASK_BLKIO); |
e7ab4d1a LP |
2710 | } |
2711 | ||
4ad49000 LP |
2712 | static const char* const cgroup_device_policy_table[_CGROUP_DEVICE_POLICY_MAX] = { |
2713 | [CGROUP_AUTO] = "auto", | |
2714 | [CGROUP_CLOSED] = "closed", | |
2715 | [CGROUP_STRICT] = "strict", | |
2716 | }; | |
4fbf50b3 | 2717 | |
4ad49000 | 2718 | DEFINE_STRING_TABLE_LOOKUP(cgroup_device_policy, CGroupDevicePolicy); |