]>
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" |
03a7b521 | 27 | #include "cgroup-util.h" |
3ffd4af2 LP |
28 | #include "cgroup.h" |
29 | #include "fd-util.h" | |
0d39fa9c | 30 | #include "fileio.h" |
77601719 | 31 | #include "fs-util.h" |
6bedfcbb | 32 | #include "parse-util.h" |
9eb977db | 33 | #include "path-util.h" |
03a7b521 | 34 | #include "process-util.h" |
9444b1f2 | 35 | #include "special.h" |
906c06f6 | 36 | #include "stdio-util.h" |
8b43440b | 37 | #include "string-table.h" |
07630cea | 38 | #include "string-util.h" |
8e274523 | 39 | |
9a054909 LP |
40 | #define CGROUP_CPU_QUOTA_PERIOD_USEC ((usec_t) 100 * USEC_PER_MSEC) |
41 | ||
2b40998d | 42 | static void cgroup_compat_warn(void) { |
128fadc9 TH |
43 | static bool cgroup_compat_warned = false; |
44 | ||
45 | if (cgroup_compat_warned) | |
46 | return; | |
47 | ||
48 | log_warning("cgroup compatibility translation between legacy and unified hierarchy settings activated. See cgroup-compat debug messages for details."); | |
49 | cgroup_compat_warned = true; | |
50 | } | |
51 | ||
52 | #define log_cgroup_compat(unit, fmt, ...) do { \ | |
53 | cgroup_compat_warn(); \ | |
54 | log_unit_debug(unit, "cgroup-compat: " fmt, ##__VA_ARGS__); \ | |
2b40998d | 55 | } while (false) |
128fadc9 | 56 | |
4ad49000 LP |
57 | void cgroup_context_init(CGroupContext *c) { |
58 | assert(c); | |
59 | ||
60 | /* Initialize everything to the kernel defaults, assuming the | |
61 | * structure is preinitialized to 0 */ | |
62 | ||
66ebf6c0 TH |
63 | c->cpu_weight = CGROUP_WEIGHT_INVALID; |
64 | c->startup_cpu_weight = CGROUP_WEIGHT_INVALID; | |
65 | c->cpu_quota_per_sec_usec = USEC_INFINITY; | |
66 | ||
d53d9474 LP |
67 | c->cpu_shares = CGROUP_CPU_SHARES_INVALID; |
68 | c->startup_cpu_shares = CGROUP_CPU_SHARES_INVALID; | |
d53d9474 | 69 | |
da4d897e TH |
70 | c->memory_high = CGROUP_LIMIT_MAX; |
71 | c->memory_max = CGROUP_LIMIT_MAX; | |
96e131ea | 72 | c->memory_swap_max = CGROUP_LIMIT_MAX; |
da4d897e TH |
73 | |
74 | c->memory_limit = CGROUP_LIMIT_MAX; | |
b2f8b02e | 75 | |
13c31542 TH |
76 | c->io_weight = CGROUP_WEIGHT_INVALID; |
77 | c->startup_io_weight = CGROUP_WEIGHT_INVALID; | |
78 | ||
d53d9474 LP |
79 | c->blockio_weight = CGROUP_BLKIO_WEIGHT_INVALID; |
80 | c->startup_blockio_weight = CGROUP_BLKIO_WEIGHT_INVALID; | |
81 | ||
82 | c->tasks_max = (uint64_t) -1; | |
4ad49000 | 83 | } |
8e274523 | 84 | |
4ad49000 LP |
85 | void cgroup_context_free_device_allow(CGroupContext *c, CGroupDeviceAllow *a) { |
86 | assert(c); | |
87 | assert(a); | |
88 | ||
71fda00f | 89 | LIST_REMOVE(device_allow, c->device_allow, a); |
4ad49000 LP |
90 | free(a->path); |
91 | free(a); | |
92 | } | |
93 | ||
13c31542 TH |
94 | void cgroup_context_free_io_device_weight(CGroupContext *c, CGroupIODeviceWeight *w) { |
95 | assert(c); | |
96 | assert(w); | |
97 | ||
98 | LIST_REMOVE(device_weights, c->io_device_weights, w); | |
99 | free(w->path); | |
100 | free(w); | |
101 | } | |
102 | ||
103 | void cgroup_context_free_io_device_limit(CGroupContext *c, CGroupIODeviceLimit *l) { | |
104 | assert(c); | |
105 | assert(l); | |
106 | ||
107 | LIST_REMOVE(device_limits, c->io_device_limits, l); | |
108 | free(l->path); | |
109 | free(l); | |
110 | } | |
111 | ||
4ad49000 LP |
112 | void cgroup_context_free_blockio_device_weight(CGroupContext *c, CGroupBlockIODeviceWeight *w) { |
113 | assert(c); | |
114 | assert(w); | |
115 | ||
71fda00f | 116 | LIST_REMOVE(device_weights, c->blockio_device_weights, w); |
4ad49000 LP |
117 | free(w->path); |
118 | free(w); | |
119 | } | |
120 | ||
121 | void cgroup_context_free_blockio_device_bandwidth(CGroupContext *c, CGroupBlockIODeviceBandwidth *b) { | |
122 | assert(c); | |
8e274523 | 123 | assert(b); |
8e274523 | 124 | |
71fda00f | 125 | LIST_REMOVE(device_bandwidths, c->blockio_device_bandwidths, b); |
4ad49000 LP |
126 | free(b->path); |
127 | free(b); | |
128 | } | |
129 | ||
130 | void cgroup_context_done(CGroupContext *c) { | |
131 | assert(c); | |
132 | ||
13c31542 TH |
133 | while (c->io_device_weights) |
134 | cgroup_context_free_io_device_weight(c, c->io_device_weights); | |
135 | ||
136 | while (c->io_device_limits) | |
137 | cgroup_context_free_io_device_limit(c, c->io_device_limits); | |
138 | ||
4ad49000 LP |
139 | while (c->blockio_device_weights) |
140 | cgroup_context_free_blockio_device_weight(c, c->blockio_device_weights); | |
141 | ||
142 | while (c->blockio_device_bandwidths) | |
143 | cgroup_context_free_blockio_device_bandwidth(c, c->blockio_device_bandwidths); | |
144 | ||
145 | while (c->device_allow) | |
146 | cgroup_context_free_device_allow(c, c->device_allow); | |
6a48d82f DM |
147 | |
148 | c->ip_address_allow = ip_address_access_free_all(c->ip_address_allow); | |
149 | c->ip_address_deny = ip_address_access_free_all(c->ip_address_deny); | |
4ad49000 LP |
150 | } |
151 | ||
152 | void cgroup_context_dump(CGroupContext *c, FILE* f, const char *prefix) { | |
13c31542 TH |
153 | CGroupIODeviceLimit *il; |
154 | CGroupIODeviceWeight *iw; | |
4ad49000 LP |
155 | CGroupBlockIODeviceBandwidth *b; |
156 | CGroupBlockIODeviceWeight *w; | |
157 | CGroupDeviceAllow *a; | |
c21c9906 | 158 | IPAddressAccessItem *iaai; |
9a054909 | 159 | char u[FORMAT_TIMESPAN_MAX]; |
4ad49000 LP |
160 | |
161 | assert(c); | |
162 | assert(f); | |
163 | ||
164 | prefix = strempty(prefix); | |
165 | ||
166 | fprintf(f, | |
167 | "%sCPUAccounting=%s\n" | |
13c31542 | 168 | "%sIOAccounting=%s\n" |
4ad49000 LP |
169 | "%sBlockIOAccounting=%s\n" |
170 | "%sMemoryAccounting=%s\n" | |
d53d9474 | 171 | "%sTasksAccounting=%s\n" |
c21c9906 | 172 | "%sIPAccounting=%s\n" |
66ebf6c0 TH |
173 | "%sCPUWeight=%" PRIu64 "\n" |
174 | "%sStartupCPUWeight=%" PRIu64 "\n" | |
d53d9474 LP |
175 | "%sCPUShares=%" PRIu64 "\n" |
176 | "%sStartupCPUShares=%" PRIu64 "\n" | |
b2f8b02e | 177 | "%sCPUQuotaPerSecSec=%s\n" |
13c31542 TH |
178 | "%sIOWeight=%" PRIu64 "\n" |
179 | "%sStartupIOWeight=%" PRIu64 "\n" | |
d53d9474 LP |
180 | "%sBlockIOWeight=%" PRIu64 "\n" |
181 | "%sStartupBlockIOWeight=%" PRIu64 "\n" | |
da4d897e TH |
182 | "%sMemoryLow=%" PRIu64 "\n" |
183 | "%sMemoryHigh=%" PRIu64 "\n" | |
184 | "%sMemoryMax=%" PRIu64 "\n" | |
96e131ea | 185 | "%sMemorySwapMax=%" PRIu64 "\n" |
4ad49000 | 186 | "%sMemoryLimit=%" PRIu64 "\n" |
03a7b521 | 187 | "%sTasksMax=%" PRIu64 "\n" |
a931ad47 LP |
188 | "%sDevicePolicy=%s\n" |
189 | "%sDelegate=%s\n", | |
4ad49000 | 190 | prefix, yes_no(c->cpu_accounting), |
13c31542 | 191 | prefix, yes_no(c->io_accounting), |
4ad49000 LP |
192 | prefix, yes_no(c->blockio_accounting), |
193 | prefix, yes_no(c->memory_accounting), | |
d53d9474 | 194 | prefix, yes_no(c->tasks_accounting), |
c21c9906 | 195 | prefix, yes_no(c->ip_accounting), |
66ebf6c0 TH |
196 | prefix, c->cpu_weight, |
197 | prefix, c->startup_cpu_weight, | |
4ad49000 | 198 | prefix, c->cpu_shares, |
95ae05c0 | 199 | prefix, c->startup_cpu_shares, |
b1d6dcf5 | 200 | prefix, format_timespan(u, sizeof(u), c->cpu_quota_per_sec_usec, 1), |
13c31542 TH |
201 | prefix, c->io_weight, |
202 | prefix, c->startup_io_weight, | |
4ad49000 | 203 | prefix, c->blockio_weight, |
95ae05c0 | 204 | prefix, c->startup_blockio_weight, |
da4d897e TH |
205 | prefix, c->memory_low, |
206 | prefix, c->memory_high, | |
207 | prefix, c->memory_max, | |
96e131ea | 208 | prefix, c->memory_swap_max, |
4ad49000 | 209 | prefix, c->memory_limit, |
03a7b521 | 210 | prefix, c->tasks_max, |
a931ad47 LP |
211 | prefix, cgroup_device_policy_to_string(c->device_policy), |
212 | prefix, yes_no(c->delegate)); | |
4ad49000 | 213 | |
02638280 LP |
214 | if (c->delegate) { |
215 | _cleanup_free_ char *t = NULL; | |
216 | ||
217 | (void) cg_mask_to_string(c->delegate_controllers, &t); | |
218 | ||
47a78d41 | 219 | fprintf(f, "%sDelegateControllers=%s\n", |
02638280 LP |
220 | prefix, |
221 | strempty(t)); | |
222 | } | |
223 | ||
4ad49000 LP |
224 | LIST_FOREACH(device_allow, a, c->device_allow) |
225 | fprintf(f, | |
226 | "%sDeviceAllow=%s %s%s%s\n", | |
227 | prefix, | |
228 | a->path, | |
229 | a->r ? "r" : "", a->w ? "w" : "", a->m ? "m" : ""); | |
230 | ||
13c31542 TH |
231 | LIST_FOREACH(device_weights, iw, c->io_device_weights) |
232 | fprintf(f, | |
233 | "%sIODeviceWeight=%s %" PRIu64, | |
234 | prefix, | |
235 | iw->path, | |
236 | iw->weight); | |
237 | ||
238 | LIST_FOREACH(device_limits, il, c->io_device_limits) { | |
239 | char buf[FORMAT_BYTES_MAX]; | |
9be57249 TH |
240 | CGroupIOLimitType type; |
241 | ||
242 | for (type = 0; type < _CGROUP_IO_LIMIT_TYPE_MAX; type++) | |
243 | if (il->limits[type] != cgroup_io_limit_defaults[type]) | |
244 | fprintf(f, | |
245 | "%s%s=%s %s\n", | |
246 | prefix, | |
247 | cgroup_io_limit_type_to_string(type), | |
248 | il->path, | |
249 | format_bytes(buf, sizeof(buf), il->limits[type])); | |
13c31542 TH |
250 | } |
251 | ||
4ad49000 LP |
252 | LIST_FOREACH(device_weights, w, c->blockio_device_weights) |
253 | fprintf(f, | |
d53d9474 | 254 | "%sBlockIODeviceWeight=%s %" PRIu64, |
4ad49000 LP |
255 | prefix, |
256 | w->path, | |
257 | w->weight); | |
258 | ||
259 | LIST_FOREACH(device_bandwidths, b, c->blockio_device_bandwidths) { | |
260 | char buf[FORMAT_BYTES_MAX]; | |
261 | ||
979d0311 TH |
262 | if (b->rbps != CGROUP_LIMIT_MAX) |
263 | fprintf(f, | |
264 | "%sBlockIOReadBandwidth=%s %s\n", | |
265 | prefix, | |
266 | b->path, | |
267 | format_bytes(buf, sizeof(buf), b->rbps)); | |
268 | if (b->wbps != CGROUP_LIMIT_MAX) | |
269 | fprintf(f, | |
270 | "%sBlockIOWriteBandwidth=%s %s\n", | |
271 | prefix, | |
272 | b->path, | |
273 | format_bytes(buf, sizeof(buf), b->wbps)); | |
4ad49000 | 274 | } |
c21c9906 LP |
275 | |
276 | LIST_FOREACH(items, iaai, c->ip_address_allow) { | |
277 | _cleanup_free_ char *k = NULL; | |
278 | ||
279 | (void) in_addr_to_string(iaai->family, &iaai->address, &k); | |
280 | fprintf(f, "%sIPAddressAllow=%s/%u\n", prefix, strnull(k), iaai->prefixlen); | |
281 | } | |
282 | ||
283 | LIST_FOREACH(items, iaai, c->ip_address_deny) { | |
284 | _cleanup_free_ char *k = NULL; | |
285 | ||
286 | (void) in_addr_to_string(iaai->family, &iaai->address, &k); | |
287 | fprintf(f, "%sIPAddressDeny=%s/%u\n", prefix, strnull(k), iaai->prefixlen); | |
288 | } | |
4ad49000 LP |
289 | } |
290 | ||
13c31542 | 291 | static int lookup_block_device(const char *p, dev_t *dev) { |
4ad49000 LP |
292 | struct stat st; |
293 | int r; | |
294 | ||
295 | assert(p); | |
296 | assert(dev); | |
297 | ||
298 | r = stat(p, &st); | |
4a62c710 MS |
299 | if (r < 0) |
300 | return log_warning_errno(errno, "Couldn't stat device %s: %m", p); | |
8e274523 | 301 | |
4ad49000 LP |
302 | if (S_ISBLK(st.st_mode)) |
303 | *dev = st.st_rdev; | |
304 | else if (major(st.st_dev) != 0) { | |
305 | /* If this is not a device node then find the block | |
306 | * device this file is stored on */ | |
307 | *dev = st.st_dev; | |
308 | ||
309 | /* If this is a partition, try to get the originating | |
310 | * block device */ | |
18c528e9 | 311 | (void) block_get_whole_disk(*dev, dev); |
4ad49000 LP |
312 | } else { |
313 | log_warning("%s is not a block device and file system block device cannot be determined or is not local.", p); | |
314 | return -ENODEV; | |
315 | } | |
8e274523 | 316 | |
8e274523 | 317 | return 0; |
8e274523 LP |
318 | } |
319 | ||
4ad49000 LP |
320 | static int whitelist_device(const char *path, const char *node, const char *acc) { |
321 | char buf[2+DECIMAL_STR_MAX(dev_t)*2+2+4]; | |
322 | struct stat st; | |
b200489b | 323 | bool ignore_notfound; |
8c6db833 | 324 | int r; |
8e274523 | 325 | |
4ad49000 LP |
326 | assert(path); |
327 | assert(acc); | |
8e274523 | 328 | |
b200489b DR |
329 | if (node[0] == '-') { |
330 | /* Non-existent paths starting with "-" must be silently ignored */ | |
331 | node++; | |
332 | ignore_notfound = true; | |
333 | } else | |
334 | ignore_notfound = false; | |
335 | ||
4ad49000 | 336 | if (stat(node, &st) < 0) { |
b200489b | 337 | if (errno == ENOENT && ignore_notfound) |
e7330dfe DP |
338 | return 0; |
339 | ||
340 | return log_warning_errno(errno, "Couldn't stat device %s: %m", node); | |
4ad49000 LP |
341 | } |
342 | ||
343 | if (!S_ISCHR(st.st_mode) && !S_ISBLK(st.st_mode)) { | |
344 | log_warning("%s is not a device.", node); | |
345 | return -ENODEV; | |
346 | } | |
347 | ||
348 | sprintf(buf, | |
349 | "%c %u:%u %s", | |
350 | S_ISCHR(st.st_mode) ? 'c' : 'b', | |
351 | major(st.st_rdev), minor(st.st_rdev), | |
352 | acc); | |
353 | ||
354 | r = cg_set_attribute("devices", path, "devices.allow", buf); | |
1aeab12b | 355 | if (r < 0) |
077ba06e | 356 | log_full_errno(IN_SET(r, -ENOENT, -EROFS, -EINVAL, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, |
714e2e1d | 357 | "Failed to set devices.allow on %s: %m", path); |
4ad49000 LP |
358 | |
359 | return r; | |
8e274523 LP |
360 | } |
361 | ||
90060676 LP |
362 | static int whitelist_major(const char *path, const char *name, char type, const char *acc) { |
363 | _cleanup_fclose_ FILE *f = NULL; | |
364 | char line[LINE_MAX]; | |
365 | bool good = false; | |
366 | int r; | |
367 | ||
368 | assert(path); | |
369 | assert(acc); | |
4c701096 | 370 | assert(IN_SET(type, 'b', 'c')); |
90060676 LP |
371 | |
372 | f = fopen("/proc/devices", "re"); | |
4a62c710 MS |
373 | if (!f) |
374 | return log_warning_errno(errno, "Cannot open /proc/devices to resolve %s (%c): %m", name, type); | |
90060676 LP |
375 | |
376 | FOREACH_LINE(line, f, goto fail) { | |
377 | char buf[2+DECIMAL_STR_MAX(unsigned)+3+4], *p, *w; | |
378 | unsigned maj; | |
379 | ||
380 | truncate_nl(line); | |
381 | ||
382 | if (type == 'c' && streq(line, "Character devices:")) { | |
383 | good = true; | |
384 | continue; | |
385 | } | |
386 | ||
387 | if (type == 'b' && streq(line, "Block devices:")) { | |
388 | good = true; | |
389 | continue; | |
390 | } | |
391 | ||
392 | if (isempty(line)) { | |
393 | good = false; | |
394 | continue; | |
395 | } | |
396 | ||
397 | if (!good) | |
398 | continue; | |
399 | ||
400 | p = strstrip(line); | |
401 | ||
402 | w = strpbrk(p, WHITESPACE); | |
403 | if (!w) | |
404 | continue; | |
405 | *w = 0; | |
406 | ||
407 | r = safe_atou(p, &maj); | |
408 | if (r < 0) | |
409 | continue; | |
410 | if (maj <= 0) | |
411 | continue; | |
412 | ||
413 | w++; | |
414 | w += strspn(w, WHITESPACE); | |
e41969e3 LP |
415 | |
416 | if (fnmatch(name, w, 0) != 0) | |
90060676 LP |
417 | continue; |
418 | ||
419 | sprintf(buf, | |
420 | "%c %u:* %s", | |
421 | type, | |
422 | maj, | |
423 | acc); | |
424 | ||
425 | r = cg_set_attribute("devices", path, "devices.allow", buf); | |
1aeab12b | 426 | if (r < 0) |
077ba06e | 427 | log_full_errno(IN_SET(r, -ENOENT, -EROFS, -EINVAL, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, |
714e2e1d | 428 | "Failed to set devices.allow on %s: %m", path); |
90060676 LP |
429 | } |
430 | ||
431 | return 0; | |
432 | ||
433 | fail: | |
25f027c5 | 434 | return log_warning_errno(errno, "Failed to read /proc/devices: %m"); |
90060676 LP |
435 | } |
436 | ||
66ebf6c0 TH |
437 | static bool cgroup_context_has_cpu_weight(CGroupContext *c) { |
438 | return c->cpu_weight != CGROUP_WEIGHT_INVALID || | |
439 | c->startup_cpu_weight != CGROUP_WEIGHT_INVALID; | |
440 | } | |
441 | ||
442 | static bool cgroup_context_has_cpu_shares(CGroupContext *c) { | |
443 | return c->cpu_shares != CGROUP_CPU_SHARES_INVALID || | |
444 | c->startup_cpu_shares != CGROUP_CPU_SHARES_INVALID; | |
445 | } | |
446 | ||
447 | static uint64_t cgroup_context_cpu_weight(CGroupContext *c, ManagerState state) { | |
448 | if (IN_SET(state, MANAGER_STARTING, MANAGER_INITIALIZING) && | |
449 | c->startup_cpu_weight != CGROUP_WEIGHT_INVALID) | |
450 | return c->startup_cpu_weight; | |
451 | else if (c->cpu_weight != CGROUP_WEIGHT_INVALID) | |
452 | return c->cpu_weight; | |
453 | else | |
454 | return CGROUP_WEIGHT_DEFAULT; | |
455 | } | |
456 | ||
457 | static uint64_t cgroup_context_cpu_shares(CGroupContext *c, ManagerState state) { | |
458 | if (IN_SET(state, MANAGER_STARTING, MANAGER_INITIALIZING) && | |
459 | c->startup_cpu_shares != CGROUP_CPU_SHARES_INVALID) | |
460 | return c->startup_cpu_shares; | |
461 | else if (c->cpu_shares != CGROUP_CPU_SHARES_INVALID) | |
462 | return c->cpu_shares; | |
463 | else | |
464 | return CGROUP_CPU_SHARES_DEFAULT; | |
465 | } | |
466 | ||
467 | static void cgroup_apply_unified_cpu_config(Unit *u, uint64_t weight, uint64_t quota) { | |
468 | char buf[MAX(DECIMAL_STR_MAX(uint64_t) + 1, (DECIMAL_STR_MAX(usec_t) + 1) * 2)]; | |
469 | int r; | |
470 | ||
471 | xsprintf(buf, "%" PRIu64 "\n", weight); | |
472 | r = cg_set_attribute("cpu", u->cgroup_path, "cpu.weight", buf); | |
473 | if (r < 0) | |
474 | log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, | |
475 | "Failed to set cpu.weight: %m"); | |
476 | ||
477 | if (quota != USEC_INFINITY) | |
478 | xsprintf(buf, USEC_FMT " " USEC_FMT "\n", | |
479 | quota * CGROUP_CPU_QUOTA_PERIOD_USEC / USEC_PER_SEC, CGROUP_CPU_QUOTA_PERIOD_USEC); | |
480 | else | |
481 | xsprintf(buf, "max " USEC_FMT "\n", CGROUP_CPU_QUOTA_PERIOD_USEC); | |
482 | ||
483 | r = cg_set_attribute("cpu", u->cgroup_path, "cpu.max", buf); | |
484 | ||
485 | if (r < 0) | |
486 | log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, | |
487 | "Failed to set cpu.max: %m"); | |
488 | } | |
489 | ||
490 | static void cgroup_apply_legacy_cpu_config(Unit *u, uint64_t shares, uint64_t quota) { | |
491 | char buf[MAX(DECIMAL_STR_MAX(uint64_t), DECIMAL_STR_MAX(usec_t)) + 1]; | |
492 | int r; | |
493 | ||
494 | xsprintf(buf, "%" PRIu64 "\n", shares); | |
495 | r = cg_set_attribute("cpu", u->cgroup_path, "cpu.shares", buf); | |
496 | if (r < 0) | |
497 | log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, | |
498 | "Failed to set cpu.shares: %m"); | |
499 | ||
500 | xsprintf(buf, USEC_FMT "\n", CGROUP_CPU_QUOTA_PERIOD_USEC); | |
501 | r = cg_set_attribute("cpu", u->cgroup_path, "cpu.cfs_period_us", buf); | |
502 | if (r < 0) | |
503 | log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, | |
504 | "Failed to set cpu.cfs_period_us: %m"); | |
505 | ||
506 | if (quota != USEC_INFINITY) { | |
507 | xsprintf(buf, USEC_FMT "\n", quota * CGROUP_CPU_QUOTA_PERIOD_USEC / USEC_PER_SEC); | |
508 | r = cg_set_attribute("cpu", u->cgroup_path, "cpu.cfs_quota_us", buf); | |
509 | } else | |
510 | r = cg_set_attribute("cpu", u->cgroup_path, "cpu.cfs_quota_us", "-1"); | |
511 | if (r < 0) | |
512 | log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, | |
513 | "Failed to set cpu.cfs_quota_us: %m"); | |
514 | } | |
515 | ||
516 | static uint64_t cgroup_cpu_shares_to_weight(uint64_t shares) { | |
517 | return CLAMP(shares * CGROUP_WEIGHT_DEFAULT / CGROUP_CPU_SHARES_DEFAULT, | |
518 | CGROUP_WEIGHT_MIN, CGROUP_WEIGHT_MAX); | |
519 | } | |
520 | ||
521 | static uint64_t cgroup_cpu_weight_to_shares(uint64_t weight) { | |
522 | return CLAMP(weight * CGROUP_CPU_SHARES_DEFAULT / CGROUP_WEIGHT_DEFAULT, | |
523 | CGROUP_CPU_SHARES_MIN, CGROUP_CPU_SHARES_MAX); | |
524 | } | |
525 | ||
508c45da | 526 | static bool cgroup_context_has_io_config(CGroupContext *c) { |
538b4852 TH |
527 | return c->io_accounting || |
528 | c->io_weight != CGROUP_WEIGHT_INVALID || | |
529 | c->startup_io_weight != CGROUP_WEIGHT_INVALID || | |
530 | c->io_device_weights || | |
531 | c->io_device_limits; | |
532 | } | |
533 | ||
508c45da | 534 | static bool cgroup_context_has_blockio_config(CGroupContext *c) { |
538b4852 TH |
535 | return c->blockio_accounting || |
536 | c->blockio_weight != CGROUP_BLKIO_WEIGHT_INVALID || | |
537 | c->startup_blockio_weight != CGROUP_BLKIO_WEIGHT_INVALID || | |
538 | c->blockio_device_weights || | |
539 | c->blockio_device_bandwidths; | |
540 | } | |
541 | ||
508c45da | 542 | static uint64_t cgroup_context_io_weight(CGroupContext *c, ManagerState state) { |
64faf04c TH |
543 | if (IN_SET(state, MANAGER_STARTING, MANAGER_INITIALIZING) && |
544 | c->startup_io_weight != CGROUP_WEIGHT_INVALID) | |
545 | return c->startup_io_weight; | |
546 | else if (c->io_weight != CGROUP_WEIGHT_INVALID) | |
547 | return c->io_weight; | |
548 | else | |
549 | return CGROUP_WEIGHT_DEFAULT; | |
550 | } | |
551 | ||
508c45da | 552 | static uint64_t cgroup_context_blkio_weight(CGroupContext *c, ManagerState state) { |
64faf04c TH |
553 | if (IN_SET(state, MANAGER_STARTING, MANAGER_INITIALIZING) && |
554 | c->startup_blockio_weight != CGROUP_BLKIO_WEIGHT_INVALID) | |
555 | return c->startup_blockio_weight; | |
556 | else if (c->blockio_weight != CGROUP_BLKIO_WEIGHT_INVALID) | |
557 | return c->blockio_weight; | |
558 | else | |
559 | return CGROUP_BLKIO_WEIGHT_DEFAULT; | |
560 | } | |
561 | ||
508c45da | 562 | static uint64_t cgroup_weight_blkio_to_io(uint64_t blkio_weight) { |
538b4852 TH |
563 | return CLAMP(blkio_weight * CGROUP_WEIGHT_DEFAULT / CGROUP_BLKIO_WEIGHT_DEFAULT, |
564 | CGROUP_WEIGHT_MIN, CGROUP_WEIGHT_MAX); | |
565 | } | |
566 | ||
508c45da | 567 | static uint64_t cgroup_weight_io_to_blkio(uint64_t io_weight) { |
538b4852 TH |
568 | return CLAMP(io_weight * CGROUP_BLKIO_WEIGHT_DEFAULT / CGROUP_WEIGHT_DEFAULT, |
569 | CGROUP_BLKIO_WEIGHT_MIN, CGROUP_BLKIO_WEIGHT_MAX); | |
570 | } | |
571 | ||
f29ff115 | 572 | static void cgroup_apply_io_device_weight(Unit *u, const char *dev_path, uint64_t io_weight) { |
64faf04c TH |
573 | char buf[DECIMAL_STR_MAX(dev_t)*2+2+DECIMAL_STR_MAX(uint64_t)+1]; |
574 | dev_t dev; | |
575 | int r; | |
576 | ||
577 | r = lookup_block_device(dev_path, &dev); | |
578 | if (r < 0) | |
579 | return; | |
580 | ||
581 | xsprintf(buf, "%u:%u %" PRIu64 "\n", major(dev), minor(dev), io_weight); | |
f29ff115 | 582 | r = cg_set_attribute("io", u->cgroup_path, "io.weight", buf); |
64faf04c | 583 | if (r < 0) |
f29ff115 TH |
584 | log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, |
585 | "Failed to set io.weight: %m"); | |
64faf04c TH |
586 | } |
587 | ||
f29ff115 | 588 | static void cgroup_apply_blkio_device_weight(Unit *u, const char *dev_path, uint64_t blkio_weight) { |
64faf04c TH |
589 | char buf[DECIMAL_STR_MAX(dev_t)*2+2+DECIMAL_STR_MAX(uint64_t)+1]; |
590 | dev_t dev; | |
591 | int r; | |
592 | ||
593 | r = lookup_block_device(dev_path, &dev); | |
594 | if (r < 0) | |
595 | return; | |
596 | ||
597 | xsprintf(buf, "%u:%u %" PRIu64 "\n", major(dev), minor(dev), blkio_weight); | |
f29ff115 | 598 | r = cg_set_attribute("blkio", u->cgroup_path, "blkio.weight_device", buf); |
64faf04c | 599 | if (r < 0) |
f29ff115 TH |
600 | log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, |
601 | "Failed to set blkio.weight_device: %m"); | |
64faf04c TH |
602 | } |
603 | ||
f29ff115 | 604 | static unsigned cgroup_apply_io_device_limit(Unit *u, const char *dev_path, uint64_t *limits) { |
64faf04c TH |
605 | char limit_bufs[_CGROUP_IO_LIMIT_TYPE_MAX][DECIMAL_STR_MAX(uint64_t)]; |
606 | char buf[DECIMAL_STR_MAX(dev_t)*2+2+(6+DECIMAL_STR_MAX(uint64_t)+1)*4]; | |
607 | CGroupIOLimitType type; | |
608 | dev_t dev; | |
609 | unsigned n = 0; | |
610 | int r; | |
611 | ||
612 | r = lookup_block_device(dev_path, &dev); | |
613 | if (r < 0) | |
614 | return 0; | |
615 | ||
616 | for (type = 0; type < _CGROUP_IO_LIMIT_TYPE_MAX; type++) { | |
617 | if (limits[type] != cgroup_io_limit_defaults[type]) { | |
618 | xsprintf(limit_bufs[type], "%" PRIu64, limits[type]); | |
619 | n++; | |
620 | } else { | |
621 | xsprintf(limit_bufs[type], "%s", limits[type] == CGROUP_LIMIT_MAX ? "max" : "0"); | |
622 | } | |
623 | } | |
624 | ||
625 | xsprintf(buf, "%u:%u rbps=%s wbps=%s riops=%s wiops=%s\n", major(dev), minor(dev), | |
626 | limit_bufs[CGROUP_IO_RBPS_MAX], limit_bufs[CGROUP_IO_WBPS_MAX], | |
627 | limit_bufs[CGROUP_IO_RIOPS_MAX], limit_bufs[CGROUP_IO_WIOPS_MAX]); | |
f29ff115 | 628 | r = cg_set_attribute("io", u->cgroup_path, "io.max", buf); |
64faf04c | 629 | if (r < 0) |
f29ff115 TH |
630 | log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, |
631 | "Failed to set io.max: %m"); | |
64faf04c TH |
632 | return n; |
633 | } | |
634 | ||
f29ff115 | 635 | static unsigned cgroup_apply_blkio_device_limit(Unit *u, const char *dev_path, uint64_t rbps, uint64_t wbps) { |
64faf04c TH |
636 | char buf[DECIMAL_STR_MAX(dev_t)*2+2+DECIMAL_STR_MAX(uint64_t)+1]; |
637 | dev_t dev; | |
638 | unsigned n = 0; | |
639 | int r; | |
640 | ||
641 | r = lookup_block_device(dev_path, &dev); | |
642 | if (r < 0) | |
643 | return 0; | |
644 | ||
645 | if (rbps != CGROUP_LIMIT_MAX) | |
646 | n++; | |
647 | sprintf(buf, "%u:%u %" PRIu64 "\n", major(dev), minor(dev), rbps); | |
f29ff115 | 648 | r = cg_set_attribute("blkio", u->cgroup_path, "blkio.throttle.read_bps_device", buf); |
64faf04c | 649 | if (r < 0) |
f29ff115 TH |
650 | log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, |
651 | "Failed to set blkio.throttle.read_bps_device: %m"); | |
64faf04c TH |
652 | |
653 | if (wbps != CGROUP_LIMIT_MAX) | |
654 | n++; | |
655 | sprintf(buf, "%u:%u %" PRIu64 "\n", major(dev), minor(dev), wbps); | |
f29ff115 | 656 | r = cg_set_attribute("blkio", u->cgroup_path, "blkio.throttle.write_bps_device", buf); |
64faf04c | 657 | if (r < 0) |
f29ff115 TH |
658 | log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, |
659 | "Failed to set blkio.throttle.write_bps_device: %m"); | |
64faf04c TH |
660 | |
661 | return n; | |
662 | } | |
663 | ||
da4d897e | 664 | static bool cgroup_context_has_unified_memory_config(CGroupContext *c) { |
96e131ea | 665 | 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 |
666 | } |
667 | ||
f29ff115 | 668 | static void cgroup_apply_unified_memory_limit(Unit *u, const char *file, uint64_t v) { |
da4d897e TH |
669 | char buf[DECIMAL_STR_MAX(uint64_t) + 1] = "max"; |
670 | int r; | |
671 | ||
672 | if (v != CGROUP_LIMIT_MAX) | |
673 | xsprintf(buf, "%" PRIu64 "\n", v); | |
674 | ||
f29ff115 | 675 | r = cg_set_attribute("memory", u->cgroup_path, file, buf); |
da4d897e | 676 | if (r < 0) |
f29ff115 TH |
677 | log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, |
678 | "Failed to set %s: %m", file); | |
da4d897e TH |
679 | } |
680 | ||
0f2d84d2 | 681 | static void cgroup_apply_firewall(Unit *u) { |
906c06f6 DM |
682 | int r; |
683 | ||
0f2d84d2 LP |
684 | assert(u); |
685 | ||
906c06f6 DM |
686 | if (u->type == UNIT_SLICE) /* Skip this for slice units, they are inner cgroup nodes, and since bpf/cgroup is |
687 | * not recursive we don't ever touch the bpf on them */ | |
688 | return; | |
689 | ||
690 | r = bpf_firewall_compile(u); | |
691 | if (r < 0) | |
692 | return; | |
693 | ||
694 | (void) bpf_firewall_install(u); | |
695 | return; | |
696 | } | |
697 | ||
698 | static void cgroup_context_apply( | |
699 | Unit *u, | |
700 | CGroupMask apply_mask, | |
701 | bool apply_bpf, | |
702 | ManagerState state) { | |
703 | ||
f29ff115 TH |
704 | const char *path; |
705 | CGroupContext *c; | |
01efdf13 | 706 | bool is_root; |
4ad49000 LP |
707 | int r; |
708 | ||
f29ff115 TH |
709 | assert(u); |
710 | ||
711 | c = unit_get_cgroup_context(u); | |
712 | path = u->cgroup_path; | |
713 | ||
4ad49000 LP |
714 | assert(c); |
715 | assert(path); | |
8e274523 | 716 | |
906c06f6 DM |
717 | /* Nothing to do? Exit early! */ |
718 | if (apply_mask == 0 && !apply_bpf) | |
4ad49000 | 719 | return; |
8e274523 | 720 | |
71c26873 | 721 | /* Some cgroup attributes are not supported on the root cgroup, |
01efdf13 LP |
722 | * hence silently ignore */ |
723 | is_root = isempty(path) || path_equal(path, "/"); | |
6da13913 ZJS |
724 | if (is_root) |
725 | /* Make sure we don't try to display messages with an empty path. */ | |
726 | path = "/"; | |
01efdf13 | 727 | |
714e2e1d LP |
728 | /* We generally ignore errors caused by read-only mounted |
729 | * cgroup trees (assuming we are running in a container then), | |
730 | * and missing cgroups, i.e. EROFS and ENOENT. */ | |
731 | ||
906c06f6 DM |
732 | if ((apply_mask & CGROUP_MASK_CPU) && !is_root) { |
733 | bool has_weight, has_shares; | |
734 | ||
735 | has_weight = cgroup_context_has_cpu_weight(c); | |
736 | has_shares = cgroup_context_has_cpu_shares(c); | |
8e274523 | 737 | |
b4cccbc1 | 738 | if (cg_all_unified() > 0) { |
66ebf6c0 | 739 | uint64_t weight; |
b2f8b02e | 740 | |
66ebf6c0 TH |
741 | if (has_weight) |
742 | weight = cgroup_context_cpu_weight(c, state); | |
743 | else if (has_shares) { | |
744 | uint64_t shares = cgroup_context_cpu_shares(c, state); | |
b2f8b02e | 745 | |
66ebf6c0 TH |
746 | weight = cgroup_cpu_shares_to_weight(shares); |
747 | ||
748 | log_cgroup_compat(u, "Applying [Startup]CpuShares %" PRIu64 " as [Startup]CpuWeight %" PRIu64 " on %s", | |
749 | shares, weight, path); | |
750 | } else | |
751 | weight = CGROUP_WEIGHT_DEFAULT; | |
752 | ||
753 | cgroup_apply_unified_cpu_config(u, weight, c->cpu_quota_per_sec_usec); | |
754 | } else { | |
755 | uint64_t shares; | |
756 | ||
7d862ab8 | 757 | if (has_weight) { |
66ebf6c0 TH |
758 | uint64_t weight = cgroup_context_cpu_weight(c, state); |
759 | ||
760 | shares = cgroup_cpu_weight_to_shares(weight); | |
761 | ||
762 | log_cgroup_compat(u, "Applying [Startup]CpuWeight %" PRIu64 " as [Startup]CpuShares %" PRIu64 " on %s", | |
763 | weight, shares, path); | |
7d862ab8 TH |
764 | } else if (has_shares) |
765 | shares = cgroup_context_cpu_shares(c, state); | |
766 | else | |
66ebf6c0 TH |
767 | shares = CGROUP_CPU_SHARES_DEFAULT; |
768 | ||
769 | cgroup_apply_legacy_cpu_config(u, shares, c->cpu_quota_per_sec_usec); | |
770 | } | |
4ad49000 LP |
771 | } |
772 | ||
906c06f6 | 773 | if (apply_mask & CGROUP_MASK_IO) { |
538b4852 TH |
774 | bool has_io = cgroup_context_has_io_config(c); |
775 | bool has_blockio = cgroup_context_has_blockio_config(c); | |
13c31542 TH |
776 | |
777 | if (!is_root) { | |
64faf04c TH |
778 | char buf[8+DECIMAL_STR_MAX(uint64_t)+1]; |
779 | uint64_t weight; | |
13c31542 | 780 | |
538b4852 TH |
781 | if (has_io) |
782 | weight = cgroup_context_io_weight(c, state); | |
128fadc9 TH |
783 | else if (has_blockio) { |
784 | uint64_t blkio_weight = cgroup_context_blkio_weight(c, state); | |
785 | ||
786 | weight = cgroup_weight_blkio_to_io(blkio_weight); | |
787 | ||
788 | log_cgroup_compat(u, "Applying [Startup]BlockIOWeight %" PRIu64 " as [Startup]IOWeight %" PRIu64, | |
789 | blkio_weight, weight); | |
790 | } else | |
538b4852 | 791 | weight = CGROUP_WEIGHT_DEFAULT; |
13c31542 TH |
792 | |
793 | xsprintf(buf, "default %" PRIu64 "\n", weight); | |
794 | r = cg_set_attribute("io", path, "io.weight", buf); | |
795 | if (r < 0) | |
f29ff115 TH |
796 | log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, |
797 | "Failed to set io.weight: %m"); | |
13c31542 | 798 | |
538b4852 TH |
799 | if (has_io) { |
800 | CGroupIODeviceWeight *w; | |
801 | ||
802 | /* FIXME: no way to reset this list */ | |
803 | LIST_FOREACH(device_weights, w, c->io_device_weights) | |
f29ff115 | 804 | cgroup_apply_io_device_weight(u, w->path, w->weight); |
538b4852 TH |
805 | } else if (has_blockio) { |
806 | CGroupBlockIODeviceWeight *w; | |
807 | ||
808 | /* FIXME: no way to reset this list */ | |
128fadc9 TH |
809 | LIST_FOREACH(device_weights, w, c->blockio_device_weights) { |
810 | weight = cgroup_weight_blkio_to_io(w->weight); | |
811 | ||
812 | log_cgroup_compat(u, "Applying BlockIODeviceWeight %" PRIu64 " as IODeviceWeight %" PRIu64 " for %s", | |
813 | w->weight, weight, w->path); | |
814 | ||
815 | cgroup_apply_io_device_weight(u, w->path, weight); | |
816 | } | |
538b4852 | 817 | } |
13c31542 TH |
818 | } |
819 | ||
64faf04c | 820 | /* Apply limits and free ones without config. */ |
538b4852 TH |
821 | if (has_io) { |
822 | CGroupIODeviceLimit *l, *next; | |
823 | ||
824 | LIST_FOREACH_SAFE(device_limits, l, next, c->io_device_limits) { | |
f29ff115 | 825 | if (!cgroup_apply_io_device_limit(u, l->path, l->limits)) |
538b4852 TH |
826 | cgroup_context_free_io_device_limit(c, l); |
827 | } | |
828 | } else if (has_blockio) { | |
829 | CGroupBlockIODeviceBandwidth *b, *next; | |
830 | ||
831 | LIST_FOREACH_SAFE(device_bandwidths, b, next, c->blockio_device_bandwidths) { | |
832 | uint64_t limits[_CGROUP_IO_LIMIT_TYPE_MAX]; | |
833 | CGroupIOLimitType type; | |
834 | ||
835 | for (type = 0; type < _CGROUP_IO_LIMIT_TYPE_MAX; type++) | |
836 | limits[type] = cgroup_io_limit_defaults[type]; | |
837 | ||
838 | limits[CGROUP_IO_RBPS_MAX] = b->rbps; | |
839 | limits[CGROUP_IO_WBPS_MAX] = b->wbps; | |
840 | ||
128fadc9 TH |
841 | log_cgroup_compat(u, "Applying BlockIO{Read|Write}Bandwidth %" PRIu64 " %" PRIu64 " as IO{Read|Write}BandwidthMax for %s", |
842 | b->rbps, b->wbps, b->path); | |
843 | ||
f29ff115 | 844 | if (!cgroup_apply_io_device_limit(u, b->path, limits)) |
538b4852 TH |
845 | cgroup_context_free_blockio_device_bandwidth(c, b); |
846 | } | |
13c31542 TH |
847 | } |
848 | } | |
849 | ||
906c06f6 | 850 | if (apply_mask & CGROUP_MASK_BLKIO) { |
538b4852 TH |
851 | bool has_io = cgroup_context_has_io_config(c); |
852 | bool has_blockio = cgroup_context_has_blockio_config(c); | |
4ad49000 | 853 | |
01efdf13 | 854 | if (!is_root) { |
64faf04c TH |
855 | char buf[DECIMAL_STR_MAX(uint64_t)+1]; |
856 | uint64_t weight; | |
64faf04c | 857 | |
7d862ab8 | 858 | if (has_io) { |
128fadc9 TH |
859 | uint64_t io_weight = cgroup_context_io_weight(c, state); |
860 | ||
538b4852 | 861 | weight = cgroup_weight_io_to_blkio(cgroup_context_io_weight(c, state)); |
128fadc9 TH |
862 | |
863 | log_cgroup_compat(u, "Applying [Startup]IOWeight %" PRIu64 " as [Startup]BlockIOWeight %" PRIu64, | |
864 | io_weight, weight); | |
7d862ab8 TH |
865 | } else if (has_blockio) |
866 | weight = cgroup_context_blkio_weight(c, state); | |
867 | else | |
538b4852 | 868 | weight = CGROUP_BLKIO_WEIGHT_DEFAULT; |
64faf04c TH |
869 | |
870 | xsprintf(buf, "%" PRIu64 "\n", weight); | |
01efdf13 | 871 | r = cg_set_attribute("blkio", path, "blkio.weight", buf); |
1aeab12b | 872 | if (r < 0) |
f29ff115 TH |
873 | log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, |
874 | "Failed to set blkio.weight: %m"); | |
4ad49000 | 875 | |
7d862ab8 | 876 | if (has_io) { |
538b4852 TH |
877 | CGroupIODeviceWeight *w; |
878 | ||
879 | /* FIXME: no way to reset this list */ | |
128fadc9 TH |
880 | LIST_FOREACH(device_weights, w, c->io_device_weights) { |
881 | weight = cgroup_weight_io_to_blkio(w->weight); | |
882 | ||
883 | log_cgroup_compat(u, "Applying IODeviceWeight %" PRIu64 " as BlockIODeviceWeight %" PRIu64 " for %s", | |
884 | w->weight, weight, w->path); | |
885 | ||
886 | cgroup_apply_blkio_device_weight(u, w->path, weight); | |
887 | } | |
7d862ab8 TH |
888 | } else if (has_blockio) { |
889 | CGroupBlockIODeviceWeight *w; | |
890 | ||
891 | /* FIXME: no way to reset this list */ | |
892 | LIST_FOREACH(device_weights, w, c->blockio_device_weights) | |
893 | cgroup_apply_blkio_device_weight(u, w->path, w->weight); | |
538b4852 | 894 | } |
4ad49000 LP |
895 | } |
896 | ||
64faf04c | 897 | /* Apply limits and free ones without config. */ |
7d862ab8 | 898 | if (has_io) { |
538b4852 TH |
899 | CGroupIODeviceLimit *l, *next; |
900 | ||
901 | LIST_FOREACH_SAFE(device_limits, l, next, c->io_device_limits) { | |
128fadc9 TH |
902 | log_cgroup_compat(u, "Applying IO{Read|Write}Bandwidth %" PRIu64 " %" PRIu64 " as BlockIO{Read|Write}BandwidthMax for %s", |
903 | l->limits[CGROUP_IO_RBPS_MAX], l->limits[CGROUP_IO_WBPS_MAX], l->path); | |
904 | ||
f29ff115 | 905 | if (!cgroup_apply_blkio_device_limit(u, l->path, l->limits[CGROUP_IO_RBPS_MAX], l->limits[CGROUP_IO_WBPS_MAX])) |
538b4852 TH |
906 | cgroup_context_free_io_device_limit(c, l); |
907 | } | |
7d862ab8 TH |
908 | } else if (has_blockio) { |
909 | CGroupBlockIODeviceBandwidth *b, *next; | |
910 | ||
911 | LIST_FOREACH_SAFE(device_bandwidths, b, next, c->blockio_device_bandwidths) | |
912 | if (!cgroup_apply_blkio_device_limit(u, b->path, b->rbps, b->wbps)) | |
913 | cgroup_context_free_blockio_device_bandwidth(c, b); | |
d686d8a9 | 914 | } |
8e274523 LP |
915 | } |
916 | ||
906c06f6 | 917 | if ((apply_mask & CGROUP_MASK_MEMORY) && !is_root) { |
b4cccbc1 LP |
918 | if (cg_all_unified() > 0) { |
919 | uint64_t max, swap_max = CGROUP_LIMIT_MAX; | |
efdb0237 | 920 | |
96e131ea | 921 | if (cgroup_context_has_unified_memory_config(c)) { |
da4d897e | 922 | max = c->memory_max; |
96e131ea WC |
923 | swap_max = c->memory_swap_max; |
924 | } else { | |
da4d897e | 925 | max = c->memory_limit; |
efdb0237 | 926 | |
128fadc9 TH |
927 | if (max != CGROUP_LIMIT_MAX) |
928 | log_cgroup_compat(u, "Applying MemoryLimit %" PRIu64 " as MemoryMax", max); | |
929 | } | |
930 | ||
f29ff115 TH |
931 | cgroup_apply_unified_memory_limit(u, "memory.low", c->memory_low); |
932 | cgroup_apply_unified_memory_limit(u, "memory.high", c->memory_high); | |
933 | cgroup_apply_unified_memory_limit(u, "memory.max", max); | |
96e131ea | 934 | cgroup_apply_unified_memory_limit(u, "memory.swap.max", swap_max); |
efdb0237 | 935 | } else { |
da4d897e | 936 | char buf[DECIMAL_STR_MAX(uint64_t) + 1]; |
7d862ab8 | 937 | uint64_t val; |
da4d897e | 938 | |
7d862ab8 | 939 | if (cgroup_context_has_unified_memory_config(c)) { |
78a4ee59 | 940 | val = c->memory_max; |
7d862ab8 TH |
941 | log_cgroup_compat(u, "Applying MemoryMax %" PRIi64 " as MemoryLimit", val); |
942 | } else | |
943 | val = c->memory_limit; | |
128fadc9 | 944 | |
78a4ee59 DM |
945 | if (val == CGROUP_LIMIT_MAX) |
946 | strncpy(buf, "-1\n", sizeof(buf)); | |
947 | else | |
948 | xsprintf(buf, "%" PRIu64 "\n", val); | |
949 | ||
da4d897e TH |
950 | r = cg_set_attribute("memory", path, "memory.limit_in_bytes", buf); |
951 | if (r < 0) | |
f29ff115 TH |
952 | log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, |
953 | "Failed to set memory.limit_in_bytes: %m"); | |
da4d897e | 954 | } |
4ad49000 | 955 | } |
8e274523 | 956 | |
906c06f6 | 957 | if ((apply_mask & CGROUP_MASK_DEVICES) && !is_root) { |
4ad49000 | 958 | CGroupDeviceAllow *a; |
8e274523 | 959 | |
714e2e1d LP |
960 | /* Changing the devices list of a populated cgroup |
961 | * might result in EINVAL, hence ignore EINVAL | |
962 | * here. */ | |
963 | ||
4ad49000 LP |
964 | if (c->device_allow || c->device_policy != CGROUP_AUTO) |
965 | r = cg_set_attribute("devices", path, "devices.deny", "a"); | |
966 | else | |
967 | r = cg_set_attribute("devices", path, "devices.allow", "a"); | |
1aeab12b | 968 | if (r < 0) |
f29ff115 TH |
969 | log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EINVAL, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, |
970 | "Failed to reset devices.list: %m"); | |
fb385181 | 971 | |
4ad49000 LP |
972 | if (c->device_policy == CGROUP_CLOSED || |
973 | (c->device_policy == CGROUP_AUTO && c->device_allow)) { | |
974 | static const char auto_devices[] = | |
7d711efb LP |
975 | "/dev/null\0" "rwm\0" |
976 | "/dev/zero\0" "rwm\0" | |
977 | "/dev/full\0" "rwm\0" | |
978 | "/dev/random\0" "rwm\0" | |
979 | "/dev/urandom\0" "rwm\0" | |
980 | "/dev/tty\0" "rwm\0" | |
5a7f87a9 | 981 | "/dev/ptmx\0" "rwm\0" |
0d9e7991 | 982 | /* Allow /run/systemd/inaccessible/{chr,blk} devices for mapping InaccessiblePaths */ |
e7330dfe DP |
983 | "-/run/systemd/inaccessible/chr\0" "rwm\0" |
984 | "-/run/systemd/inaccessible/blk\0" "rwm\0"; | |
4ad49000 LP |
985 | |
986 | const char *x, *y; | |
987 | ||
988 | NULSTR_FOREACH_PAIR(x, y, auto_devices) | |
989 | whitelist_device(path, x, y); | |
7d711efb | 990 | |
5a7f87a9 | 991 | /* PTS (/dev/pts) devices may not be duplicated, but accessed */ |
7d711efb | 992 | whitelist_major(path, "pts", 'c', "rw"); |
4ad49000 LP |
993 | } |
994 | ||
995 | LIST_FOREACH(device_allow, a, c->device_allow) { | |
fb4650aa | 996 | char acc[4], *val; |
4ad49000 LP |
997 | unsigned k = 0; |
998 | ||
999 | if (a->r) | |
1000 | acc[k++] = 'r'; | |
1001 | if (a->w) | |
1002 | acc[k++] = 'w'; | |
1003 | if (a->m) | |
1004 | acc[k++] = 'm'; | |
fb385181 | 1005 | |
4ad49000 LP |
1006 | if (k == 0) |
1007 | continue; | |
fb385181 | 1008 | |
4ad49000 | 1009 | acc[k++] = 0; |
90060676 | 1010 | |
27458ed6 | 1011 | if (path_startswith(a->path, "/dev/")) |
90060676 | 1012 | whitelist_device(path, a->path, acc); |
fb4650aa ZJS |
1013 | else if ((val = startswith(a->path, "block-"))) |
1014 | whitelist_major(path, val, 'b', acc); | |
1015 | else if ((val = startswith(a->path, "char-"))) | |
1016 | whitelist_major(path, val, 'c', acc); | |
90060676 | 1017 | else |
f29ff115 | 1018 | log_unit_debug(u, "Ignoring device %s while writing cgroup attribute.", a->path); |
4ad49000 LP |
1019 | } |
1020 | } | |
03a7b521 | 1021 | |
906c06f6 | 1022 | if ((apply_mask & CGROUP_MASK_PIDS) && !is_root) { |
03a7b521 | 1023 | |
f5058264 | 1024 | if (c->tasks_max != CGROUP_LIMIT_MAX) { |
03a7b521 LP |
1025 | char buf[DECIMAL_STR_MAX(uint64_t) + 2]; |
1026 | ||
1027 | sprintf(buf, "%" PRIu64 "\n", c->tasks_max); | |
1028 | r = cg_set_attribute("pids", path, "pids.max", buf); | |
1029 | } else | |
1030 | r = cg_set_attribute("pids", path, "pids.max", "max"); | |
1031 | ||
1032 | if (r < 0) | |
f29ff115 TH |
1033 | log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, |
1034 | "Failed to set pids.max: %m"); | |
03a7b521 | 1035 | } |
906c06f6 DM |
1036 | |
1037 | if (apply_bpf) | |
0f2d84d2 | 1038 | cgroup_apply_firewall(u); |
fb385181 LP |
1039 | } |
1040 | ||
efdb0237 LP |
1041 | CGroupMask cgroup_context_get_mask(CGroupContext *c) { |
1042 | CGroupMask mask = 0; | |
8e274523 | 1043 | |
4ad49000 | 1044 | /* Figure out which controllers we need */ |
8e274523 | 1045 | |
b2f8b02e | 1046 | if (c->cpu_accounting || |
66ebf6c0 TH |
1047 | cgroup_context_has_cpu_weight(c) || |
1048 | cgroup_context_has_cpu_shares(c) || | |
3a43da28 | 1049 | c->cpu_quota_per_sec_usec != USEC_INFINITY) |
efdb0237 | 1050 | mask |= CGROUP_MASK_CPUACCT | CGROUP_MASK_CPU; |
ecedd90f | 1051 | |
538b4852 TH |
1052 | if (cgroup_context_has_io_config(c) || cgroup_context_has_blockio_config(c)) |
1053 | mask |= CGROUP_MASK_IO | CGROUP_MASK_BLKIO; | |
ecedd90f | 1054 | |
4ad49000 | 1055 | if (c->memory_accounting || |
da4d897e TH |
1056 | c->memory_limit != CGROUP_LIMIT_MAX || |
1057 | cgroup_context_has_unified_memory_config(c)) | |
efdb0237 | 1058 | mask |= CGROUP_MASK_MEMORY; |
8e274523 | 1059 | |
a931ad47 LP |
1060 | if (c->device_allow || |
1061 | c->device_policy != CGROUP_AUTO) | |
3905f127 | 1062 | mask |= CGROUP_MASK_DEVICES; |
4ad49000 | 1063 | |
03a7b521 LP |
1064 | if (c->tasks_accounting || |
1065 | c->tasks_max != (uint64_t) -1) | |
1066 | mask |= CGROUP_MASK_PIDS; | |
1067 | ||
4ad49000 | 1068 | return mask; |
8e274523 LP |
1069 | } |
1070 | ||
efdb0237 | 1071 | CGroupMask unit_get_own_mask(Unit *u) { |
4ad49000 | 1072 | CGroupContext *c; |
8e274523 | 1073 | |
efdb0237 LP |
1074 | /* Returns the mask of controllers the unit needs for itself */ |
1075 | ||
4ad49000 LP |
1076 | c = unit_get_cgroup_context(u); |
1077 | if (!c) | |
1078 | return 0; | |
8e274523 | 1079 | |
64e844e5 | 1080 | return cgroup_context_get_mask(c) | unit_get_delegate_mask(u); |
02638280 LP |
1081 | } |
1082 | ||
1083 | CGroupMask unit_get_delegate_mask(Unit *u) { | |
1084 | CGroupContext *c; | |
1085 | ||
1086 | /* If delegation is turned on, then turn on selected controllers, unless we are on the legacy hierarchy and the | |
1087 | * process we fork into is known to drop privileges, and hence shouldn't get access to the controllers. | |
19af675e | 1088 | * |
02638280 | 1089 | * Note that on the unified hierarchy it is safe to delegate controllers to unprivileged services. */ |
a931ad47 | 1090 | |
02638280 LP |
1091 | if (u->type == UNIT_SLICE) |
1092 | return 0; | |
1093 | ||
1094 | c = unit_get_cgroup_context(u); | |
1095 | if (!c) | |
1096 | return 0; | |
1097 | ||
1098 | if (!c->delegate) | |
1099 | return 0; | |
1100 | ||
1101 | if (cg_all_unified() <= 0) { | |
a931ad47 LP |
1102 | ExecContext *e; |
1103 | ||
1104 | e = unit_get_exec_context(u); | |
02638280 LP |
1105 | if (e && !exec_context_maintains_privileges(e)) |
1106 | return 0; | |
a931ad47 LP |
1107 | } |
1108 | ||
02638280 | 1109 | return c->delegate_controllers; |
8e274523 LP |
1110 | } |
1111 | ||
efdb0237 | 1112 | CGroupMask unit_get_members_mask(Unit *u) { |
4ad49000 | 1113 | assert(u); |
bc432dc7 | 1114 | |
02638280 | 1115 | /* Returns the mask of controllers all of the unit's children require, merged */ |
efdb0237 | 1116 | |
bc432dc7 LP |
1117 | if (u->cgroup_members_mask_valid) |
1118 | return u->cgroup_members_mask; | |
1119 | ||
64e844e5 | 1120 | u->cgroup_members_mask = 0; |
bc432dc7 LP |
1121 | |
1122 | if (u->type == UNIT_SLICE) { | |
eef85c4a | 1123 | void *v; |
bc432dc7 LP |
1124 | Unit *member; |
1125 | Iterator i; | |
1126 | ||
eef85c4a | 1127 | HASHMAP_FOREACH_KEY(v, member, u->dependencies[UNIT_BEFORE], i) { |
bc432dc7 LP |
1128 | |
1129 | if (member == u) | |
1130 | continue; | |
1131 | ||
d4fdc205 | 1132 | if (UNIT_DEREF(member->slice) != u) |
bc432dc7 LP |
1133 | continue; |
1134 | ||
31604970 | 1135 | u->cgroup_members_mask |= unit_get_subtree_mask(member); /* note that this calls ourselves again, for the children */ |
bc432dc7 LP |
1136 | } |
1137 | } | |
1138 | ||
1139 | u->cgroup_members_mask_valid = true; | |
6414b7c9 | 1140 | return u->cgroup_members_mask; |
246aa6dd LP |
1141 | } |
1142 | ||
efdb0237 | 1143 | CGroupMask unit_get_siblings_mask(Unit *u) { |
4ad49000 | 1144 | assert(u); |
246aa6dd | 1145 | |
efdb0237 LP |
1146 | /* Returns the mask of controllers all of the unit's siblings |
1147 | * require, i.e. the members mask of the unit's parent slice | |
1148 | * if there is one. */ | |
1149 | ||
bc432dc7 | 1150 | if (UNIT_ISSET(u->slice)) |
637f421e | 1151 | return unit_get_members_mask(UNIT_DEREF(u->slice)); |
4ad49000 | 1152 | |
64e844e5 | 1153 | return unit_get_subtree_mask(u); /* we are the top-level slice */ |
246aa6dd LP |
1154 | } |
1155 | ||
efdb0237 LP |
1156 | CGroupMask unit_get_subtree_mask(Unit *u) { |
1157 | ||
1158 | /* Returns the mask of this subtree, meaning of the group | |
1159 | * itself and its children. */ | |
1160 | ||
1161 | return unit_get_own_mask(u) | unit_get_members_mask(u); | |
1162 | } | |
1163 | ||
1164 | CGroupMask unit_get_target_mask(Unit *u) { | |
1165 | CGroupMask mask; | |
1166 | ||
1167 | /* This returns the cgroup mask of all controllers to enable | |
1168 | * for a specific cgroup, i.e. everything it needs itself, | |
1169 | * plus all that its children need, plus all that its siblings | |
1170 | * need. This is primarily useful on the legacy cgroup | |
1171 | * hierarchy, where we need to duplicate each cgroup in each | |
1172 | * hierarchy that shall be enabled for it. */ | |
6414b7c9 | 1173 | |
efdb0237 LP |
1174 | mask = unit_get_own_mask(u) | unit_get_members_mask(u) | unit_get_siblings_mask(u); |
1175 | mask &= u->manager->cgroup_supported; | |
1176 | ||
1177 | return mask; | |
1178 | } | |
1179 | ||
1180 | CGroupMask unit_get_enable_mask(Unit *u) { | |
1181 | CGroupMask mask; | |
1182 | ||
1183 | /* This returns the cgroup mask of all controllers to enable | |
1184 | * for the children of a specific cgroup. This is primarily | |
1185 | * useful for the unified cgroup hierarchy, where each cgroup | |
1186 | * controls which controllers are enabled for its children. */ | |
1187 | ||
1188 | mask = unit_get_members_mask(u); | |
6414b7c9 DS |
1189 | mask &= u->manager->cgroup_supported; |
1190 | ||
1191 | return mask; | |
1192 | } | |
1193 | ||
906c06f6 DM |
1194 | bool unit_get_needs_bpf(Unit *u) { |
1195 | CGroupContext *c; | |
1196 | Unit *p; | |
1197 | assert(u); | |
1198 | ||
1199 | /* We never attach BPF to slice units, as they are inner cgroup nodes and cgroup/BPF is not recursive at the | |
1200 | * moment. */ | |
1201 | if (u->type == UNIT_SLICE) | |
1202 | return false; | |
1203 | ||
1204 | c = unit_get_cgroup_context(u); | |
1205 | if (!c) | |
1206 | return false; | |
1207 | ||
1208 | if (c->ip_accounting || | |
1209 | c->ip_address_allow || | |
1210 | c->ip_address_deny) | |
1211 | return true; | |
1212 | ||
1213 | /* If any parent slice has an IP access list defined, it applies too */ | |
1214 | for (p = UNIT_DEREF(u->slice); p; p = UNIT_DEREF(p->slice)) { | |
1215 | c = unit_get_cgroup_context(p); | |
1216 | if (!c) | |
1217 | return false; | |
1218 | ||
1219 | if (c->ip_address_allow || | |
1220 | c->ip_address_deny) | |
1221 | return true; | |
1222 | } | |
1223 | ||
1224 | return false; | |
1225 | } | |
1226 | ||
6414b7c9 DS |
1227 | /* Recurse from a unit up through its containing slices, propagating |
1228 | * mask bits upward. A unit is also member of itself. */ | |
bc432dc7 | 1229 | void unit_update_cgroup_members_masks(Unit *u) { |
efdb0237 | 1230 | CGroupMask m; |
bc432dc7 LP |
1231 | bool more; |
1232 | ||
1233 | assert(u); | |
1234 | ||
1235 | /* Calculate subtree mask */ | |
efdb0237 | 1236 | m = unit_get_subtree_mask(u); |
bc432dc7 LP |
1237 | |
1238 | /* See if anything changed from the previous invocation. If | |
1239 | * not, we're done. */ | |
1240 | if (u->cgroup_subtree_mask_valid && m == u->cgroup_subtree_mask) | |
1241 | return; | |
1242 | ||
1243 | more = | |
1244 | u->cgroup_subtree_mask_valid && | |
1245 | ((m & ~u->cgroup_subtree_mask) != 0) && | |
1246 | ((~m & u->cgroup_subtree_mask) == 0); | |
1247 | ||
1248 | u->cgroup_subtree_mask = m; | |
1249 | u->cgroup_subtree_mask_valid = true; | |
1250 | ||
6414b7c9 DS |
1251 | if (UNIT_ISSET(u->slice)) { |
1252 | Unit *s = UNIT_DEREF(u->slice); | |
bc432dc7 LP |
1253 | |
1254 | if (more) | |
1255 | /* There's more set now than before. We | |
1256 | * propagate the new mask to the parent's mask | |
1257 | * (not caring if it actually was valid or | |
1258 | * not). */ | |
1259 | ||
1260 | s->cgroup_members_mask |= m; | |
1261 | ||
1262 | else | |
1263 | /* There's less set now than before (or we | |
1264 | * don't know), we need to recalculate | |
1265 | * everything, so let's invalidate the | |
1266 | * parent's members mask */ | |
1267 | ||
1268 | s->cgroup_members_mask_valid = false; | |
1269 | ||
1270 | /* And now make sure that this change also hits our | |
1271 | * grandparents */ | |
1272 | unit_update_cgroup_members_masks(s); | |
6414b7c9 DS |
1273 | } |
1274 | } | |
1275 | ||
efdb0237 | 1276 | static const char *migrate_callback(CGroupMask mask, void *userdata) { |
03b90d4b LP |
1277 | Unit *u = userdata; |
1278 | ||
1279 | assert(mask != 0); | |
1280 | assert(u); | |
1281 | ||
1282 | while (u) { | |
1283 | if (u->cgroup_path && | |
1284 | u->cgroup_realized && | |
1285 | (u->cgroup_realized_mask & mask) == mask) | |
1286 | return u->cgroup_path; | |
1287 | ||
1288 | u = UNIT_DEREF(u->slice); | |
1289 | } | |
1290 | ||
1291 | return NULL; | |
1292 | } | |
1293 | ||
efdb0237 LP |
1294 | char *unit_default_cgroup_path(Unit *u) { |
1295 | _cleanup_free_ char *escaped = NULL, *slice = NULL; | |
1296 | int r; | |
1297 | ||
1298 | assert(u); | |
1299 | ||
1300 | if (unit_has_name(u, SPECIAL_ROOT_SLICE)) | |
1301 | return strdup(u->manager->cgroup_root); | |
1302 | ||
1303 | if (UNIT_ISSET(u->slice) && !unit_has_name(UNIT_DEREF(u->slice), SPECIAL_ROOT_SLICE)) { | |
1304 | r = cg_slice_to_path(UNIT_DEREF(u->slice)->id, &slice); | |
1305 | if (r < 0) | |
1306 | return NULL; | |
1307 | } | |
1308 | ||
1309 | escaped = cg_escape(u->id); | |
1310 | if (!escaped) | |
1311 | return NULL; | |
1312 | ||
1313 | if (slice) | |
605405c6 ZJS |
1314 | return strjoin(u->manager->cgroup_root, "/", slice, "/", |
1315 | escaped); | |
efdb0237 | 1316 | else |
605405c6 | 1317 | return strjoin(u->manager->cgroup_root, "/", escaped); |
efdb0237 LP |
1318 | } |
1319 | ||
1320 | int unit_set_cgroup_path(Unit *u, const char *path) { | |
1321 | _cleanup_free_ char *p = NULL; | |
1322 | int r; | |
1323 | ||
1324 | assert(u); | |
1325 | ||
1326 | if (path) { | |
1327 | p = strdup(path); | |
1328 | if (!p) | |
1329 | return -ENOMEM; | |
1330 | } else | |
1331 | p = NULL; | |
1332 | ||
1333 | if (streq_ptr(u->cgroup_path, p)) | |
1334 | return 0; | |
1335 | ||
1336 | if (p) { | |
1337 | r = hashmap_put(u->manager->cgroup_unit, p, u); | |
1338 | if (r < 0) | |
1339 | return r; | |
1340 | } | |
1341 | ||
1342 | unit_release_cgroup(u); | |
1343 | ||
1344 | u->cgroup_path = p; | |
1345 | p = NULL; | |
1346 | ||
1347 | return 1; | |
1348 | } | |
1349 | ||
1350 | int unit_watch_cgroup(Unit *u) { | |
ab2c3861 | 1351 | _cleanup_free_ char *events = NULL; |
efdb0237 LP |
1352 | int r; |
1353 | ||
1354 | assert(u); | |
1355 | ||
1356 | if (!u->cgroup_path) | |
1357 | return 0; | |
1358 | ||
1359 | if (u->cgroup_inotify_wd >= 0) | |
1360 | return 0; | |
1361 | ||
1362 | /* Only applies to the unified hierarchy */ | |
c22800e4 | 1363 | r = cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER); |
b4cccbc1 LP |
1364 | if (r < 0) |
1365 | return log_error_errno(r, "Failed to determine whether the name=systemd hierarchy is unified: %m"); | |
1366 | if (r == 0) | |
efdb0237 LP |
1367 | return 0; |
1368 | ||
1369 | /* Don't watch the root slice, it's pointless. */ | |
1370 | if (unit_has_name(u, SPECIAL_ROOT_SLICE)) | |
1371 | return 0; | |
1372 | ||
1373 | r = hashmap_ensure_allocated(&u->manager->cgroup_inotify_wd_unit, &trivial_hash_ops); | |
1374 | if (r < 0) | |
1375 | return log_oom(); | |
1376 | ||
ab2c3861 | 1377 | r = cg_get_path(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path, "cgroup.events", &events); |
efdb0237 LP |
1378 | if (r < 0) |
1379 | return log_oom(); | |
1380 | ||
ab2c3861 | 1381 | u->cgroup_inotify_wd = inotify_add_watch(u->manager->cgroup_inotify_fd, events, IN_MODIFY); |
efdb0237 LP |
1382 | if (u->cgroup_inotify_wd < 0) { |
1383 | ||
1384 | if (errno == ENOENT) /* If the directory is already | |
1385 | * gone we don't need to track | |
1386 | * it, so this is not an error */ | |
1387 | return 0; | |
1388 | ||
1389 | return log_unit_error_errno(u, errno, "Failed to add inotify watch descriptor for control group %s: %m", u->cgroup_path); | |
1390 | } | |
1391 | ||
1392 | r = hashmap_put(u->manager->cgroup_inotify_wd_unit, INT_TO_PTR(u->cgroup_inotify_wd), u); | |
1393 | if (r < 0) | |
1394 | return log_unit_error_errno(u, r, "Failed to add inotify watch descriptor to hash map: %m"); | |
1395 | ||
1396 | return 0; | |
1397 | } | |
1398 | ||
a4634b21 LP |
1399 | int unit_pick_cgroup_path(Unit *u) { |
1400 | _cleanup_free_ char *path = NULL; | |
1401 | int r; | |
1402 | ||
1403 | assert(u); | |
1404 | ||
1405 | if (u->cgroup_path) | |
1406 | return 0; | |
1407 | ||
1408 | if (!UNIT_HAS_CGROUP_CONTEXT(u)) | |
1409 | return -EINVAL; | |
1410 | ||
1411 | path = unit_default_cgroup_path(u); | |
1412 | if (!path) | |
1413 | return log_oom(); | |
1414 | ||
1415 | r = unit_set_cgroup_path(u, path); | |
1416 | if (r == -EEXIST) | |
1417 | return log_unit_error_errno(u, r, "Control group %s exists already.", path); | |
1418 | if (r < 0) | |
1419 | return log_unit_error_errno(u, r, "Failed to set unit's control group path to %s: %m", path); | |
1420 | ||
1421 | return 0; | |
1422 | } | |
1423 | ||
efdb0237 LP |
1424 | static int unit_create_cgroup( |
1425 | Unit *u, | |
1426 | CGroupMask target_mask, | |
906c06f6 DM |
1427 | CGroupMask enable_mask, |
1428 | bool needs_bpf) { | |
efdb0237 | 1429 | |
0cd385d3 | 1430 | CGroupContext *c; |
bc432dc7 | 1431 | int r; |
64747e2d | 1432 | |
4ad49000 | 1433 | assert(u); |
64747e2d | 1434 | |
0cd385d3 LP |
1435 | c = unit_get_cgroup_context(u); |
1436 | if (!c) | |
1437 | return 0; | |
1438 | ||
a4634b21 LP |
1439 | /* Figure out our cgroup path */ |
1440 | r = unit_pick_cgroup_path(u); | |
1441 | if (r < 0) | |
1442 | return r; | |
b58b8e11 | 1443 | |
03b90d4b | 1444 | /* First, create our own group */ |
efdb0237 | 1445 | r = cg_create_everywhere(u->manager->cgroup_supported, target_mask, u->cgroup_path); |
23bbb0de | 1446 | if (r < 0) |
efdb0237 LP |
1447 | return log_unit_error_errno(u, r, "Failed to create cgroup %s: %m", u->cgroup_path); |
1448 | ||
1449 | /* Start watching it */ | |
1450 | (void) unit_watch_cgroup(u); | |
1451 | ||
1452 | /* Enable all controllers we need */ | |
1453 | r = cg_enable_everywhere(u->manager->cgroup_supported, enable_mask, u->cgroup_path); | |
1454 | if (r < 0) | |
1455 | log_unit_warning_errno(u, r, "Failed to enable controllers on cgroup %s, ignoring: %m", u->cgroup_path); | |
03b90d4b LP |
1456 | |
1457 | /* Keep track that this is now realized */ | |
4ad49000 | 1458 | u->cgroup_realized = true; |
efdb0237 | 1459 | u->cgroup_realized_mask = target_mask; |
ccf78df1 | 1460 | u->cgroup_enabled_mask = enable_mask; |
906c06f6 | 1461 | u->cgroup_bpf_state = needs_bpf ? UNIT_CGROUP_BPF_ON : UNIT_CGROUP_BPF_OFF; |
4ad49000 | 1462 | |
0cd385d3 LP |
1463 | if (u->type != UNIT_SLICE && !c->delegate) { |
1464 | ||
1465 | /* Then, possibly move things over, but not if | |
1466 | * subgroups may contain processes, which is the case | |
1467 | * for slice and delegation units. */ | |
1468 | r = cg_migrate_everywhere(u->manager->cgroup_supported, u->cgroup_path, u->cgroup_path, migrate_callback, u); | |
1469 | if (r < 0) | |
efdb0237 | 1470 | log_unit_warning_errno(u, r, "Failed to migrate cgroup from to %s, ignoring: %m", u->cgroup_path); |
0cd385d3 | 1471 | } |
03b90d4b | 1472 | |
64747e2d LP |
1473 | return 0; |
1474 | } | |
1475 | ||
7b3fd631 LP |
1476 | int unit_attach_pids_to_cgroup(Unit *u) { |
1477 | int r; | |
1478 | assert(u); | |
1479 | ||
1480 | r = unit_realize_cgroup(u); | |
1481 | if (r < 0) | |
1482 | return r; | |
1483 | ||
1484 | r = cg_attach_many_everywhere(u->manager->cgroup_supported, u->cgroup_path, u->pids, migrate_callback, u); | |
1485 | if (r < 0) | |
1486 | return r; | |
1487 | ||
1488 | return 0; | |
1489 | } | |
1490 | ||
4b58153d LP |
1491 | static void cgroup_xattr_apply(Unit *u) { |
1492 | char ids[SD_ID128_STRING_MAX]; | |
1493 | int r; | |
1494 | ||
1495 | assert(u); | |
1496 | ||
1497 | if (!MANAGER_IS_SYSTEM(u->manager)) | |
1498 | return; | |
1499 | ||
1500 | if (sd_id128_is_null(u->invocation_id)) | |
1501 | return; | |
1502 | ||
1503 | r = cg_set_xattr(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path, | |
1504 | "trusted.invocation_id", | |
1505 | sd_id128_to_string(u->invocation_id, ids), 32, | |
1506 | 0); | |
1507 | if (r < 0) | |
0fb84499 | 1508 | log_unit_debug_errno(u, r, "Failed to set invocation ID on control group %s, ignoring: %m", u->cgroup_path); |
4b58153d LP |
1509 | } |
1510 | ||
906c06f6 DM |
1511 | static bool unit_has_mask_realized( |
1512 | Unit *u, | |
1513 | CGroupMask target_mask, | |
1514 | CGroupMask enable_mask, | |
1515 | bool needs_bpf) { | |
1516 | ||
bc432dc7 LP |
1517 | assert(u); |
1518 | ||
906c06f6 DM |
1519 | return u->cgroup_realized && |
1520 | u->cgroup_realized_mask == target_mask && | |
1521 | u->cgroup_enabled_mask == enable_mask && | |
1522 | ((needs_bpf && u->cgroup_bpf_state == UNIT_CGROUP_BPF_ON) || | |
1523 | (!needs_bpf && u->cgroup_bpf_state == UNIT_CGROUP_BPF_OFF)); | |
6414b7c9 DS |
1524 | } |
1525 | ||
2aa57a65 LP |
1526 | static void unit_add_to_cgroup_realize_queue(Unit *u) { |
1527 | assert(u); | |
1528 | ||
1529 | if (u->in_cgroup_realize_queue) | |
1530 | return; | |
1531 | ||
1532 | LIST_PREPEND(cgroup_realize_queue, u->manager->cgroup_realize_queue, u); | |
1533 | u->in_cgroup_realize_queue = true; | |
1534 | } | |
1535 | ||
1536 | static void unit_remove_from_cgroup_realize_queue(Unit *u) { | |
1537 | assert(u); | |
1538 | ||
1539 | if (!u->in_cgroup_realize_queue) | |
1540 | return; | |
1541 | ||
1542 | LIST_REMOVE(cgroup_realize_queue, u->manager->cgroup_realize_queue, u); | |
1543 | u->in_cgroup_realize_queue = false; | |
1544 | } | |
1545 | ||
1546 | ||
6414b7c9 DS |
1547 | /* Check if necessary controllers and attributes for a unit are in place. |
1548 | * | |
1549 | * If so, do nothing. | |
1550 | * If not, create paths, move processes over, and set attributes. | |
1551 | * | |
1552 | * Returns 0 on success and < 0 on failure. */ | |
db785129 | 1553 | static int unit_realize_cgroup_now(Unit *u, ManagerState state) { |
efdb0237 | 1554 | CGroupMask target_mask, enable_mask; |
906c06f6 | 1555 | bool needs_bpf, apply_bpf; |
6414b7c9 | 1556 | int r; |
64747e2d | 1557 | |
4ad49000 | 1558 | assert(u); |
64747e2d | 1559 | |
2aa57a65 | 1560 | unit_remove_from_cgroup_realize_queue(u); |
64747e2d | 1561 | |
efdb0237 | 1562 | target_mask = unit_get_target_mask(u); |
ccf78df1 | 1563 | enable_mask = unit_get_enable_mask(u); |
906c06f6 | 1564 | needs_bpf = unit_get_needs_bpf(u); |
ccf78df1 | 1565 | |
906c06f6 | 1566 | if (unit_has_mask_realized(u, target_mask, enable_mask, needs_bpf)) |
0a1eb06d | 1567 | return 0; |
64747e2d | 1568 | |
906c06f6 DM |
1569 | /* Make sure we apply the BPF filters either when one is configured, or if none is configured but previously |
1570 | * the state was anything but off. This way, if a unit with a BPF filter applied is reconfigured to lose it | |
1571 | * this will trickle down properly to cgroupfs. */ | |
1572 | apply_bpf = needs_bpf || u->cgroup_bpf_state != UNIT_CGROUP_BPF_OFF; | |
1573 | ||
4ad49000 | 1574 | /* First, realize parents */ |
6414b7c9 | 1575 | if (UNIT_ISSET(u->slice)) { |
db785129 | 1576 | r = unit_realize_cgroup_now(UNIT_DEREF(u->slice), state); |
6414b7c9 DS |
1577 | if (r < 0) |
1578 | return r; | |
1579 | } | |
4ad49000 LP |
1580 | |
1581 | /* And then do the real work */ | |
906c06f6 | 1582 | r = unit_create_cgroup(u, target_mask, enable_mask, needs_bpf); |
6414b7c9 DS |
1583 | if (r < 0) |
1584 | return r; | |
1585 | ||
1586 | /* Finally, apply the necessary attributes. */ | |
906c06f6 | 1587 | cgroup_context_apply(u, target_mask, apply_bpf, state); |
4b58153d | 1588 | cgroup_xattr_apply(u); |
6414b7c9 DS |
1589 | |
1590 | return 0; | |
64747e2d LP |
1591 | } |
1592 | ||
91a6073e | 1593 | unsigned manager_dispatch_cgroup_realize_queue(Manager *m) { |
db785129 | 1594 | ManagerState state; |
4ad49000 | 1595 | unsigned n = 0; |
db785129 | 1596 | Unit *i; |
6414b7c9 | 1597 | int r; |
ecedd90f | 1598 | |
91a6073e LP |
1599 | assert(m); |
1600 | ||
db785129 LP |
1601 | state = manager_state(m); |
1602 | ||
91a6073e LP |
1603 | while ((i = m->cgroup_realize_queue)) { |
1604 | assert(i->in_cgroup_realize_queue); | |
ecedd90f | 1605 | |
2aa57a65 LP |
1606 | if (UNIT_IS_INACTIVE_OR_FAILED(unit_active_state(i))) { |
1607 | /* Maybe things changed, and the unit is not actually active anymore? */ | |
1608 | unit_remove_from_cgroup_realize_queue(i); | |
1609 | continue; | |
1610 | } | |
1611 | ||
db785129 | 1612 | r = unit_realize_cgroup_now(i, state); |
6414b7c9 | 1613 | if (r < 0) |
efdb0237 | 1614 | log_warning_errno(r, "Failed to realize cgroups for queued unit %s, ignoring: %m", i->id); |
0a1eb06d | 1615 | |
4ad49000 LP |
1616 | n++; |
1617 | } | |
ecedd90f | 1618 | |
4ad49000 | 1619 | return n; |
8e274523 LP |
1620 | } |
1621 | ||
91a6073e | 1622 | static void unit_add_siblings_to_cgroup_realize_queue(Unit *u) { |
4ad49000 | 1623 | Unit *slice; |
ca949c9d | 1624 | |
4ad49000 LP |
1625 | /* This adds the siblings of the specified unit and the |
1626 | * siblings of all parent units to the cgroup queue. (But | |
1627 | * neither the specified unit itself nor the parents.) */ | |
1628 | ||
1629 | while ((slice = UNIT_DEREF(u->slice))) { | |
1630 | Iterator i; | |
1631 | Unit *m; | |
eef85c4a | 1632 | void *v; |
8f53a7b8 | 1633 | |
eef85c4a | 1634 | HASHMAP_FOREACH_KEY(v, m, u->dependencies[UNIT_BEFORE], i) { |
4ad49000 LP |
1635 | if (m == u) |
1636 | continue; | |
8e274523 | 1637 | |
6414b7c9 DS |
1638 | /* Skip units that have a dependency on the slice |
1639 | * but aren't actually in it. */ | |
4ad49000 | 1640 | if (UNIT_DEREF(m->slice) != slice) |
50159e6a | 1641 | continue; |
8e274523 | 1642 | |
6414b7c9 DS |
1643 | /* No point in doing cgroup application for units |
1644 | * without active processes. */ | |
1645 | if (UNIT_IS_INACTIVE_OR_FAILED(unit_active_state(m))) | |
1646 | continue; | |
1647 | ||
1648 | /* If the unit doesn't need any new controllers | |
1649 | * and has current ones realized, it doesn't need | |
1650 | * any changes. */ | |
906c06f6 DM |
1651 | if (unit_has_mask_realized(m, |
1652 | unit_get_target_mask(m), | |
1653 | unit_get_enable_mask(m), | |
1654 | unit_get_needs_bpf(m))) | |
6414b7c9 DS |
1655 | continue; |
1656 | ||
91a6073e | 1657 | unit_add_to_cgroup_realize_queue(m); |
50159e6a LP |
1658 | } |
1659 | ||
4ad49000 | 1660 | u = slice; |
8e274523 | 1661 | } |
4ad49000 LP |
1662 | } |
1663 | ||
0a1eb06d | 1664 | int unit_realize_cgroup(Unit *u) { |
4ad49000 LP |
1665 | assert(u); |
1666 | ||
35b7ff80 | 1667 | if (!UNIT_HAS_CGROUP_CONTEXT(u)) |
0a1eb06d | 1668 | return 0; |
8e274523 | 1669 | |
4ad49000 LP |
1670 | /* So, here's the deal: when realizing the cgroups for this |
1671 | * unit, we need to first create all parents, but there's more | |
1672 | * actually: for the weight-based controllers we also need to | |
1673 | * make sure that all our siblings (i.e. units that are in the | |
73e231ab | 1674 | * same slice as we are) have cgroups, too. Otherwise, things |
4ad49000 LP |
1675 | * would become very uneven as each of their processes would |
1676 | * get as much resources as all our group together. This call | |
1677 | * will synchronously create the parent cgroups, but will | |
1678 | * defer work on the siblings to the next event loop | |
1679 | * iteration. */ | |
ca949c9d | 1680 | |
4ad49000 | 1681 | /* Add all sibling slices to the cgroup queue. */ |
91a6073e | 1682 | unit_add_siblings_to_cgroup_realize_queue(u); |
4ad49000 | 1683 | |
6414b7c9 | 1684 | /* And realize this one now (and apply the values) */ |
db785129 | 1685 | return unit_realize_cgroup_now(u, manager_state(u->manager)); |
8e274523 LP |
1686 | } |
1687 | ||
efdb0237 LP |
1688 | void unit_release_cgroup(Unit *u) { |
1689 | assert(u); | |
1690 | ||
1691 | /* Forgets all cgroup details for this cgroup */ | |
1692 | ||
1693 | if (u->cgroup_path) { | |
1694 | (void) hashmap_remove(u->manager->cgroup_unit, u->cgroup_path); | |
1695 | u->cgroup_path = mfree(u->cgroup_path); | |
1696 | } | |
1697 | ||
1698 | if (u->cgroup_inotify_wd >= 0) { | |
1699 | if (inotify_rm_watch(u->manager->cgroup_inotify_fd, u->cgroup_inotify_wd) < 0) | |
1700 | log_unit_debug_errno(u, errno, "Failed to remove cgroup inotify watch %i for %s, ignoring", u->cgroup_inotify_wd, u->id); | |
1701 | ||
1702 | (void) hashmap_remove(u->manager->cgroup_inotify_wd_unit, INT_TO_PTR(u->cgroup_inotify_wd)); | |
1703 | u->cgroup_inotify_wd = -1; | |
1704 | } | |
1705 | } | |
1706 | ||
1707 | void unit_prune_cgroup(Unit *u) { | |
8e274523 | 1708 | int r; |
efdb0237 | 1709 | bool is_root_slice; |
8e274523 | 1710 | |
4ad49000 | 1711 | assert(u); |
8e274523 | 1712 | |
efdb0237 LP |
1713 | /* Removes the cgroup, if empty and possible, and stops watching it. */ |
1714 | ||
4ad49000 LP |
1715 | if (!u->cgroup_path) |
1716 | return; | |
8e274523 | 1717 | |
fe700f46 LP |
1718 | (void) unit_get_cpu_usage(u, NULL); /* Cache the last CPU usage value before we destroy the cgroup */ |
1719 | ||
efdb0237 LP |
1720 | is_root_slice = unit_has_name(u, SPECIAL_ROOT_SLICE); |
1721 | ||
1722 | r = cg_trim_everywhere(u->manager->cgroup_supported, u->cgroup_path, !is_root_slice); | |
dab5bf85 | 1723 | if (r < 0) { |
f29ff115 | 1724 | log_unit_debug_errno(u, r, "Failed to destroy cgroup %s, ignoring: %m", u->cgroup_path); |
dab5bf85 RL |
1725 | return; |
1726 | } | |
8e274523 | 1727 | |
efdb0237 LP |
1728 | if (is_root_slice) |
1729 | return; | |
1730 | ||
1731 | unit_release_cgroup(u); | |
0a1eb06d | 1732 | |
4ad49000 | 1733 | u->cgroup_realized = false; |
bc432dc7 | 1734 | u->cgroup_realized_mask = 0; |
ccf78df1 | 1735 | u->cgroup_enabled_mask = 0; |
8e274523 LP |
1736 | } |
1737 | ||
efdb0237 | 1738 | int unit_search_main_pid(Unit *u, pid_t *ret) { |
4ad49000 LP |
1739 | _cleanup_fclose_ FILE *f = NULL; |
1740 | pid_t pid = 0, npid, mypid; | |
efdb0237 | 1741 | int r; |
4ad49000 LP |
1742 | |
1743 | assert(u); | |
efdb0237 | 1744 | assert(ret); |
4ad49000 LP |
1745 | |
1746 | if (!u->cgroup_path) | |
efdb0237 | 1747 | return -ENXIO; |
4ad49000 | 1748 | |
efdb0237 LP |
1749 | r = cg_enumerate_processes(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path, &f); |
1750 | if (r < 0) | |
1751 | return r; | |
4ad49000 | 1752 | |
df0ff127 | 1753 | mypid = getpid_cached(); |
4ad49000 LP |
1754 | while (cg_read_pid(f, &npid) > 0) { |
1755 | pid_t ppid; | |
1756 | ||
1757 | if (npid == pid) | |
1758 | continue; | |
8e274523 | 1759 | |
4ad49000 | 1760 | /* Ignore processes that aren't our kids */ |
6bc73acb | 1761 | if (get_process_ppid(npid, &ppid) >= 0 && ppid != mypid) |
4ad49000 | 1762 | continue; |
8e274523 | 1763 | |
efdb0237 | 1764 | if (pid != 0) |
4ad49000 LP |
1765 | /* Dang, there's more than one daemonized PID |
1766 | in this group, so we don't know what process | |
1767 | is the main process. */ | |
efdb0237 LP |
1768 | |
1769 | return -ENODATA; | |
8e274523 | 1770 | |
4ad49000 | 1771 | pid = npid; |
8e274523 LP |
1772 | } |
1773 | ||
efdb0237 LP |
1774 | *ret = pid; |
1775 | return 0; | |
1776 | } | |
1777 | ||
1778 | static int unit_watch_pids_in_path(Unit *u, const char *path) { | |
b3c5bad3 | 1779 | _cleanup_closedir_ DIR *d = NULL; |
efdb0237 LP |
1780 | _cleanup_fclose_ FILE *f = NULL; |
1781 | int ret = 0, r; | |
1782 | ||
1783 | assert(u); | |
1784 | assert(path); | |
1785 | ||
1786 | r = cg_enumerate_processes(SYSTEMD_CGROUP_CONTROLLER, path, &f); | |
1787 | if (r < 0) | |
1788 | ret = r; | |
1789 | else { | |
1790 | pid_t pid; | |
1791 | ||
1792 | while ((r = cg_read_pid(f, &pid)) > 0) { | |
1793 | r = unit_watch_pid(u, pid); | |
1794 | if (r < 0 && ret >= 0) | |
1795 | ret = r; | |
1796 | } | |
1797 | ||
1798 | if (r < 0 && ret >= 0) | |
1799 | ret = r; | |
1800 | } | |
1801 | ||
1802 | r = cg_enumerate_subgroups(SYSTEMD_CGROUP_CONTROLLER, path, &d); | |
1803 | if (r < 0) { | |
1804 | if (ret >= 0) | |
1805 | ret = r; | |
1806 | } else { | |
1807 | char *fn; | |
1808 | ||
1809 | while ((r = cg_read_subgroup(d, &fn)) > 0) { | |
1810 | _cleanup_free_ char *p = NULL; | |
1811 | ||
605405c6 | 1812 | p = strjoin(path, "/", fn); |
efdb0237 LP |
1813 | free(fn); |
1814 | ||
1815 | if (!p) | |
1816 | return -ENOMEM; | |
1817 | ||
1818 | r = unit_watch_pids_in_path(u, p); | |
1819 | if (r < 0 && ret >= 0) | |
1820 | ret = r; | |
1821 | } | |
1822 | ||
1823 | if (r < 0 && ret >= 0) | |
1824 | ret = r; | |
1825 | } | |
1826 | ||
1827 | return ret; | |
1828 | } | |
1829 | ||
11aef522 LP |
1830 | int unit_synthesize_cgroup_empty_event(Unit *u) { |
1831 | int r; | |
1832 | ||
1833 | assert(u); | |
1834 | ||
1835 | /* Enqueue a synthetic cgroup empty event if this unit doesn't watch any PIDs anymore. This is compatibility | |
1836 | * support for non-unified systems where notifications aren't reliable, and hence need to take whatever we can | |
1837 | * get as notification source as soon as we stopped having any useful PIDs to watch for. */ | |
1838 | ||
1839 | if (!u->cgroup_path) | |
1840 | return -ENOENT; | |
1841 | ||
1842 | r = cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER); | |
1843 | if (r < 0) | |
1844 | return r; | |
1845 | if (r > 0) /* On unified we have reliable notifications, and don't need this */ | |
1846 | return 0; | |
1847 | ||
1848 | if (!set_isempty(u->pids)) | |
1849 | return 0; | |
1850 | ||
1851 | unit_add_to_cgroup_empty_queue(u); | |
1852 | return 0; | |
1853 | } | |
1854 | ||
efdb0237 | 1855 | int unit_watch_all_pids(Unit *u) { |
b4cccbc1 LP |
1856 | int r; |
1857 | ||
efdb0237 LP |
1858 | assert(u); |
1859 | ||
1860 | /* Adds all PIDs from our cgroup to the set of PIDs we | |
1861 | * watch. This is a fallback logic for cases where we do not | |
1862 | * get reliable cgroup empty notifications: we try to use | |
1863 | * SIGCHLD as replacement. */ | |
1864 | ||
1865 | if (!u->cgroup_path) | |
1866 | return -ENOENT; | |
1867 | ||
c22800e4 | 1868 | r = cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER); |
b4cccbc1 LP |
1869 | if (r < 0) |
1870 | return r; | |
1871 | if (r > 0) /* On unified we can use proper notifications */ | |
efdb0237 LP |
1872 | return 0; |
1873 | ||
1874 | return unit_watch_pids_in_path(u, u->cgroup_path); | |
1875 | } | |
1876 | ||
09e24654 LP |
1877 | static int on_cgroup_empty_event(sd_event_source *s, void *userdata) { |
1878 | Manager *m = userdata; | |
1879 | Unit *u; | |
efdb0237 LP |
1880 | int r; |
1881 | ||
09e24654 LP |
1882 | assert(s); |
1883 | assert(m); | |
efdb0237 | 1884 | |
09e24654 LP |
1885 | u = m->cgroup_empty_queue; |
1886 | if (!u) | |
efdb0237 LP |
1887 | return 0; |
1888 | ||
09e24654 LP |
1889 | assert(u->in_cgroup_empty_queue); |
1890 | u->in_cgroup_empty_queue = false; | |
1891 | LIST_REMOVE(cgroup_empty_queue, m->cgroup_empty_queue, u); | |
1892 | ||
1893 | if (m->cgroup_empty_queue) { | |
1894 | /* More stuff queued, let's make sure we remain enabled */ | |
1895 | r = sd_event_source_set_enabled(s, SD_EVENT_ONESHOT); | |
1896 | if (r < 0) | |
1897 | log_debug_errno(r, "Failed to reenable cgroup empty event source: %m"); | |
1898 | } | |
efdb0237 LP |
1899 | |
1900 | unit_add_to_gc_queue(u); | |
1901 | ||
1902 | if (UNIT_VTABLE(u)->notify_cgroup_empty) | |
1903 | UNIT_VTABLE(u)->notify_cgroup_empty(u); | |
1904 | ||
1905 | return 0; | |
1906 | } | |
1907 | ||
09e24654 LP |
1908 | void unit_add_to_cgroup_empty_queue(Unit *u) { |
1909 | int r; | |
1910 | ||
1911 | assert(u); | |
1912 | ||
1913 | /* Note that there are four different ways how cgroup empty events reach us: | |
1914 | * | |
1915 | * 1. On the unified hierarchy we get an inotify event on the cgroup | |
1916 | * | |
1917 | * 2. On the legacy hierarchy, when running in system mode, we get a datagram on the cgroup agent socket | |
1918 | * | |
1919 | * 3. On the legacy hierarchy, when running in user mode, we get a D-Bus signal on the system bus | |
1920 | * | |
1921 | * 4. On the legacy hierarchy, in service units we start watching all processes of the cgroup for SIGCHLD as | |
1922 | * soon as we get one SIGCHLD, to deal with unreliable cgroup notifications. | |
1923 | * | |
1924 | * Regardless which way we got the notification, we'll verify it here, and then add it to a separate | |
1925 | * queue. This queue will be dispatched at a lower priority than the SIGCHLD handler, so that we always use | |
1926 | * SIGCHLD if we can get it first, and only use the cgroup empty notifications if there's no SIGCHLD pending | |
1927 | * (which might happen if the cgroup doesn't contain processes that are our own child, which is typically the | |
1928 | * case for scope units). */ | |
1929 | ||
1930 | if (u->in_cgroup_empty_queue) | |
1931 | return; | |
1932 | ||
1933 | /* Let's verify that the cgroup is really empty */ | |
1934 | if (!u->cgroup_path) | |
1935 | return; | |
1936 | r = cg_is_empty_recursive(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path); | |
1937 | if (r < 0) { | |
1938 | log_unit_debug_errno(u, r, "Failed to determine whether cgroup %s is empty: %m", u->cgroup_path); | |
1939 | return; | |
1940 | } | |
1941 | if (r == 0) | |
1942 | return; | |
1943 | ||
1944 | LIST_PREPEND(cgroup_empty_queue, u->manager->cgroup_empty_queue, u); | |
1945 | u->in_cgroup_empty_queue = true; | |
1946 | ||
1947 | /* Trigger the defer event */ | |
1948 | r = sd_event_source_set_enabled(u->manager->cgroup_empty_event_source, SD_EVENT_ONESHOT); | |
1949 | if (r < 0) | |
1950 | log_debug_errno(r, "Failed to enable cgroup empty event source: %m"); | |
1951 | } | |
1952 | ||
efdb0237 LP |
1953 | static int on_cgroup_inotify_event(sd_event_source *s, int fd, uint32_t revents, void *userdata) { |
1954 | Manager *m = userdata; | |
1955 | ||
1956 | assert(s); | |
1957 | assert(fd >= 0); | |
1958 | assert(m); | |
1959 | ||
1960 | for (;;) { | |
1961 | union inotify_event_buffer buffer; | |
1962 | struct inotify_event *e; | |
1963 | ssize_t l; | |
1964 | ||
1965 | l = read(fd, &buffer, sizeof(buffer)); | |
1966 | if (l < 0) { | |
47249640 | 1967 | if (IN_SET(errno, EINTR, EAGAIN)) |
efdb0237 LP |
1968 | return 0; |
1969 | ||
1970 | return log_error_errno(errno, "Failed to read control group inotify events: %m"); | |
1971 | } | |
1972 | ||
1973 | FOREACH_INOTIFY_EVENT(e, buffer, l) { | |
1974 | Unit *u; | |
1975 | ||
1976 | if (e->wd < 0) | |
1977 | /* Queue overflow has no watch descriptor */ | |
1978 | continue; | |
1979 | ||
1980 | if (e->mask & IN_IGNORED) | |
1981 | /* The watch was just removed */ | |
1982 | continue; | |
1983 | ||
1984 | u = hashmap_get(m->cgroup_inotify_wd_unit, INT_TO_PTR(e->wd)); | |
1985 | if (!u) /* Not that inotify might deliver | |
1986 | * events for a watch even after it | |
1987 | * was removed, because it was queued | |
1988 | * before the removal. Let's ignore | |
1989 | * this here safely. */ | |
1990 | continue; | |
1991 | ||
09e24654 | 1992 | unit_add_to_cgroup_empty_queue(u); |
efdb0237 LP |
1993 | } |
1994 | } | |
8e274523 LP |
1995 | } |
1996 | ||
8e274523 | 1997 | int manager_setup_cgroup(Manager *m) { |
9444b1f2 | 1998 | _cleanup_free_ char *path = NULL; |
10bd3e2e | 1999 | const char *scope_path; |
efdb0237 | 2000 | CGroupController c; |
b4cccbc1 | 2001 | int r, all_unified; |
efdb0237 | 2002 | char *e; |
8e274523 LP |
2003 | |
2004 | assert(m); | |
2005 | ||
35d2e7ec | 2006 | /* 1. Determine hierarchy */ |
efdb0237 | 2007 | m->cgroup_root = mfree(m->cgroup_root); |
9444b1f2 | 2008 | r = cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER, 0, &m->cgroup_root); |
23bbb0de MS |
2009 | if (r < 0) |
2010 | return log_error_errno(r, "Cannot determine cgroup we are running in: %m"); | |
8e274523 | 2011 | |
efdb0237 LP |
2012 | /* Chop off the init scope, if we are already located in it */ |
2013 | e = endswith(m->cgroup_root, "/" SPECIAL_INIT_SCOPE); | |
0d8c31ff | 2014 | |
efdb0237 LP |
2015 | /* LEGACY: Also chop off the system slice if we are in |
2016 | * it. This is to support live upgrades from older systemd | |
2017 | * versions where PID 1 was moved there. Also see | |
2018 | * cg_get_root_path(). */ | |
463d0d15 | 2019 | if (!e && MANAGER_IS_SYSTEM(m)) { |
9444b1f2 | 2020 | e = endswith(m->cgroup_root, "/" SPECIAL_SYSTEM_SLICE); |
15c60e99 | 2021 | if (!e) |
efdb0237 | 2022 | e = endswith(m->cgroup_root, "/system"); /* even more legacy */ |
0baf24dd | 2023 | } |
efdb0237 LP |
2024 | if (e) |
2025 | *e = 0; | |
7ccfb64a | 2026 | |
7546145e LP |
2027 | /* And make sure to store away the root value without trailing slash, even for the root dir, so that we can |
2028 | * easily prepend it everywhere. */ | |
2029 | delete_trailing_chars(m->cgroup_root, "/"); | |
8e274523 | 2030 | |
35d2e7ec | 2031 | /* 2. Show data */ |
9444b1f2 | 2032 | r = cg_get_path(SYSTEMD_CGROUP_CONTROLLER, m->cgroup_root, NULL, &path); |
23bbb0de MS |
2033 | if (r < 0) |
2034 | return log_error_errno(r, "Cannot find cgroup mount point: %m"); | |
8e274523 | 2035 | |
415fc41c TH |
2036 | r = cg_unified_flush(); |
2037 | if (r < 0) | |
2038 | return log_error_errno(r, "Couldn't determine if we are running in the unified hierarchy: %m"); | |
5da38d07 | 2039 | |
b4cccbc1 | 2040 | all_unified = cg_all_unified(); |
d4c819ed ZJS |
2041 | if (all_unified < 0) |
2042 | return log_error_errno(all_unified, "Couldn't determine whether we are in all unified mode: %m"); | |
2043 | if (all_unified > 0) | |
efdb0237 | 2044 | log_debug("Unified cgroup hierarchy is located at %s.", path); |
b4cccbc1 | 2045 | else { |
c22800e4 | 2046 | r = cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER); |
b4cccbc1 LP |
2047 | if (r < 0) |
2048 | return log_error_errno(r, "Failed to determine whether systemd's own controller is in unified mode: %m"); | |
2049 | if (r > 0) | |
2050 | log_debug("Unified cgroup hierarchy is located at %s. Controllers are on legacy hierarchies.", path); | |
2051 | else | |
2052 | log_debug("Using cgroup controller " SYSTEMD_CGROUP_CONTROLLER_LEGACY ". File system hierarchy is at %s.", path); | |
2053 | } | |
efdb0237 | 2054 | |
09e24654 LP |
2055 | /* 3. Allocate cgroup empty defer event source */ |
2056 | m->cgroup_empty_event_source = sd_event_source_unref(m->cgroup_empty_event_source); | |
2057 | r = sd_event_add_defer(m->event, &m->cgroup_empty_event_source, on_cgroup_empty_event, m); | |
2058 | if (r < 0) | |
2059 | return log_error_errno(r, "Failed to create cgroup empty event source: %m"); | |
2060 | ||
2061 | r = sd_event_source_set_priority(m->cgroup_empty_event_source, SD_EVENT_PRIORITY_NORMAL-5); | |
2062 | if (r < 0) | |
2063 | return log_error_errno(r, "Failed to set priority of cgroup empty event source: %m"); | |
2064 | ||
2065 | r = sd_event_source_set_enabled(m->cgroup_empty_event_source, SD_EVENT_OFF); | |
2066 | if (r < 0) | |
2067 | return log_error_errno(r, "Failed to disable cgroup empty event source: %m"); | |
2068 | ||
2069 | (void) sd_event_source_set_description(m->cgroup_empty_event_source, "cgroup-empty"); | |
2070 | ||
2071 | /* 4. Install notifier inotify object, or agent */ | |
10bd3e2e | 2072 | if (cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER) > 0) { |
c6c18be3 | 2073 | |
09e24654 | 2074 | /* In the unified hierarchy we can get cgroup empty notifications via inotify. */ |
efdb0237 | 2075 | |
10bd3e2e LP |
2076 | m->cgroup_inotify_event_source = sd_event_source_unref(m->cgroup_inotify_event_source); |
2077 | safe_close(m->cgroup_inotify_fd); | |
efdb0237 | 2078 | |
10bd3e2e LP |
2079 | m->cgroup_inotify_fd = inotify_init1(IN_NONBLOCK|IN_CLOEXEC); |
2080 | if (m->cgroup_inotify_fd < 0) | |
2081 | return log_error_errno(errno, "Failed to create control group inotify object: %m"); | |
efdb0237 | 2082 | |
10bd3e2e LP |
2083 | r = sd_event_add_io(m->event, &m->cgroup_inotify_event_source, m->cgroup_inotify_fd, EPOLLIN, on_cgroup_inotify_event, m); |
2084 | if (r < 0) | |
2085 | return log_error_errno(r, "Failed to watch control group inotify object: %m"); | |
efdb0237 | 2086 | |
10bd3e2e LP |
2087 | /* Process cgroup empty notifications early, but after service notifications and SIGCHLD. Also |
2088 | * see handling of cgroup agent notifications, for the classic cgroup hierarchy support. */ | |
09e24654 | 2089 | r = sd_event_source_set_priority(m->cgroup_inotify_event_source, SD_EVENT_PRIORITY_NORMAL-4); |
10bd3e2e LP |
2090 | if (r < 0) |
2091 | return log_error_errno(r, "Failed to set priority of inotify event source: %m"); | |
efdb0237 | 2092 | |
10bd3e2e | 2093 | (void) sd_event_source_set_description(m->cgroup_inotify_event_source, "cgroup-inotify"); |
efdb0237 | 2094 | |
10bd3e2e | 2095 | } else if (MANAGER_IS_SYSTEM(m) && m->test_run_flags == 0) { |
efdb0237 | 2096 | |
10bd3e2e LP |
2097 | /* On the legacy hierarchy we only get notifications via cgroup agents. (Which isn't really reliable, |
2098 | * since it does not generate events when control groups with children run empty. */ | |
8e274523 | 2099 | |
10bd3e2e | 2100 | r = cg_install_release_agent(SYSTEMD_CGROUP_CONTROLLER, SYSTEMD_CGROUP_AGENT_PATH); |
23bbb0de | 2101 | if (r < 0) |
10bd3e2e LP |
2102 | log_warning_errno(r, "Failed to install release agent, ignoring: %m"); |
2103 | else if (r > 0) | |
2104 | log_debug("Installed release agent."); | |
2105 | else if (r == 0) | |
2106 | log_debug("Release agent already installed."); | |
2107 | } | |
efdb0237 | 2108 | |
09e24654 | 2109 | /* 5. Make sure we are in the special "init.scope" unit in the root slice. */ |
10bd3e2e LP |
2110 | scope_path = strjoina(m->cgroup_root, "/" SPECIAL_INIT_SCOPE); |
2111 | r = cg_create_and_attach(SYSTEMD_CGROUP_CONTROLLER, scope_path, 0); | |
2112 | if (r < 0) | |
2113 | return log_error_errno(r, "Failed to create %s control group: %m", scope_path); | |
c6c18be3 | 2114 | |
09e24654 | 2115 | /* Also, move all other userspace processes remaining in the root cgroup into that scope. */ |
10bd3e2e LP |
2116 | r = cg_migrate(SYSTEMD_CGROUP_CONTROLLER, m->cgroup_root, SYSTEMD_CGROUP_CONTROLLER, scope_path, 0); |
2117 | if (r < 0) | |
2118 | log_warning_errno(r, "Couldn't move remaining userspace processes, ignoring: %m"); | |
0d8c31ff | 2119 | |
09e24654 | 2120 | /* 6. And pin it, so that it cannot be unmounted */ |
10bd3e2e LP |
2121 | safe_close(m->pin_cgroupfs_fd); |
2122 | m->pin_cgroupfs_fd = open(path, O_RDONLY|O_CLOEXEC|O_DIRECTORY|O_NOCTTY|O_NONBLOCK); | |
2123 | if (m->pin_cgroupfs_fd < 0) | |
2124 | return log_error_errno(errno, "Failed to open pin file: %m"); | |
2125 | ||
09e24654 | 2126 | /* 7. Always enable hierarchical support if it exists... */ |
10bd3e2e LP |
2127 | if (!all_unified && m->test_run_flags == 0) |
2128 | (void) cg_set_attribute("memory", "/", "memory.use_hierarchy", "1"); | |
c6c18be3 | 2129 | |
09e24654 | 2130 | /* 8. Figure out which controllers are supported, and log about it */ |
efdb0237 LP |
2131 | r = cg_mask_supported(&m->cgroup_supported); |
2132 | if (r < 0) | |
2133 | return log_error_errno(r, "Failed to determine supported controllers: %m"); | |
efdb0237 | 2134 | for (c = 0; c < _CGROUP_CONTROLLER_MAX; c++) |
eee0a1e4 | 2135 | log_debug("Controller '%s' supported: %s", cgroup_controller_to_string(c), yes_no(m->cgroup_supported & CGROUP_CONTROLLER_TO_MASK(c))); |
9156e799 | 2136 | |
a32360f1 | 2137 | return 0; |
8e274523 LP |
2138 | } |
2139 | ||
c6c18be3 | 2140 | void manager_shutdown_cgroup(Manager *m, bool delete) { |
8e274523 LP |
2141 | assert(m); |
2142 | ||
9444b1f2 LP |
2143 | /* We can't really delete the group, since we are in it. But |
2144 | * let's trim it. */ | |
2145 | if (delete && m->cgroup_root) | |
efdb0237 LP |
2146 | (void) cg_trim(SYSTEMD_CGROUP_CONTROLLER, m->cgroup_root, false); |
2147 | ||
09e24654 LP |
2148 | m->cgroup_empty_event_source = sd_event_source_unref(m->cgroup_empty_event_source); |
2149 | ||
efdb0237 LP |
2150 | m->cgroup_inotify_wd_unit = hashmap_free(m->cgroup_inotify_wd_unit); |
2151 | ||
2152 | m->cgroup_inotify_event_source = sd_event_source_unref(m->cgroup_inotify_event_source); | |
2153 | m->cgroup_inotify_fd = safe_close(m->cgroup_inotify_fd); | |
8e274523 | 2154 | |
03e334a1 | 2155 | m->pin_cgroupfs_fd = safe_close(m->pin_cgroupfs_fd); |
c6c18be3 | 2156 | |
efdb0237 | 2157 | m->cgroup_root = mfree(m->cgroup_root); |
8e274523 LP |
2158 | } |
2159 | ||
4ad49000 | 2160 | Unit* manager_get_unit_by_cgroup(Manager *m, const char *cgroup) { |
acb14d31 | 2161 | char *p; |
4ad49000 | 2162 | Unit *u; |
acb14d31 LP |
2163 | |
2164 | assert(m); | |
2165 | assert(cgroup); | |
acb14d31 | 2166 | |
4ad49000 LP |
2167 | u = hashmap_get(m->cgroup_unit, cgroup); |
2168 | if (u) | |
2169 | return u; | |
acb14d31 | 2170 | |
8e70580b | 2171 | p = strdupa(cgroup); |
acb14d31 LP |
2172 | for (;;) { |
2173 | char *e; | |
2174 | ||
2175 | e = strrchr(p, '/'); | |
efdb0237 LP |
2176 | if (!e || e == p) |
2177 | return hashmap_get(m->cgroup_unit, SPECIAL_ROOT_SLICE); | |
acb14d31 LP |
2178 | |
2179 | *e = 0; | |
2180 | ||
4ad49000 LP |
2181 | u = hashmap_get(m->cgroup_unit, p); |
2182 | if (u) | |
2183 | return u; | |
acb14d31 LP |
2184 | } |
2185 | } | |
2186 | ||
b3ac818b | 2187 | Unit *manager_get_unit_by_pid_cgroup(Manager *m, pid_t pid) { |
4ad49000 | 2188 | _cleanup_free_ char *cgroup = NULL; |
acb14d31 | 2189 | int r; |
8e274523 | 2190 | |
8c47c732 LP |
2191 | assert(m); |
2192 | ||
b3ac818b LP |
2193 | if (pid <= 0) |
2194 | return NULL; | |
2195 | ||
2196 | r = cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER, pid, &cgroup); | |
2197 | if (r < 0) | |
2198 | return NULL; | |
2199 | ||
2200 | return manager_get_unit_by_cgroup(m, cgroup); | |
2201 | } | |
2202 | ||
2203 | Unit *manager_get_unit_by_pid(Manager *m, pid_t pid) { | |
2204 | Unit *u; | |
2205 | ||
2206 | assert(m); | |
2207 | ||
efdb0237 | 2208 | if (pid <= 0) |
8c47c732 LP |
2209 | return NULL; |
2210 | ||
2ca9d979 | 2211 | if (pid == getpid_cached()) |
efdb0237 LP |
2212 | return hashmap_get(m->units, SPECIAL_INIT_SCOPE); |
2213 | ||
fea72cc0 | 2214 | u = hashmap_get(m->watch_pids1, PID_TO_PTR(pid)); |
5fe8876b LP |
2215 | if (u) |
2216 | return u; | |
2217 | ||
fea72cc0 | 2218 | u = hashmap_get(m->watch_pids2, PID_TO_PTR(pid)); |
5fe8876b LP |
2219 | if (u) |
2220 | return u; | |
2221 | ||
b3ac818b | 2222 | return manager_get_unit_by_pid_cgroup(m, pid); |
6dde1f33 | 2223 | } |
4fbf50b3 | 2224 | |
4ad49000 LP |
2225 | int manager_notify_cgroup_empty(Manager *m, const char *cgroup) { |
2226 | Unit *u; | |
4fbf50b3 | 2227 | |
4ad49000 LP |
2228 | assert(m); |
2229 | assert(cgroup); | |
4fbf50b3 | 2230 | |
09e24654 LP |
2231 | /* Called on the legacy hierarchy whenever we get an explicit cgroup notification from the cgroup agent process |
2232 | * or from the --system instance */ | |
2233 | ||
d8fdc620 LP |
2234 | log_debug("Got cgroup empty notification for: %s", cgroup); |
2235 | ||
4ad49000 | 2236 | u = manager_get_unit_by_cgroup(m, cgroup); |
5ad096b3 LP |
2237 | if (!u) |
2238 | return 0; | |
b56c28c3 | 2239 | |
09e24654 LP |
2240 | unit_add_to_cgroup_empty_queue(u); |
2241 | return 1; | |
5ad096b3 LP |
2242 | } |
2243 | ||
2244 | int unit_get_memory_current(Unit *u, uint64_t *ret) { | |
2245 | _cleanup_free_ char *v = NULL; | |
2246 | int r; | |
2247 | ||
2248 | assert(u); | |
2249 | assert(ret); | |
2250 | ||
2e4025c0 | 2251 | if (!UNIT_CGROUP_BOOL(u, memory_accounting)) |
cf3b4be1 LP |
2252 | return -ENODATA; |
2253 | ||
5ad096b3 LP |
2254 | if (!u->cgroup_path) |
2255 | return -ENODATA; | |
2256 | ||
efdb0237 | 2257 | if ((u->cgroup_realized_mask & CGROUP_MASK_MEMORY) == 0) |
5ad096b3 LP |
2258 | return -ENODATA; |
2259 | ||
b4cccbc1 LP |
2260 | r = cg_all_unified(); |
2261 | if (r < 0) | |
2262 | return r; | |
2263 | if (r > 0) | |
efdb0237 | 2264 | r = cg_get_attribute("memory", u->cgroup_path, "memory.current", &v); |
b4cccbc1 LP |
2265 | else |
2266 | r = cg_get_attribute("memory", u->cgroup_path, "memory.usage_in_bytes", &v); | |
5ad096b3 LP |
2267 | if (r == -ENOENT) |
2268 | return -ENODATA; | |
2269 | if (r < 0) | |
2270 | return r; | |
2271 | ||
2272 | return safe_atou64(v, ret); | |
2273 | } | |
2274 | ||
03a7b521 LP |
2275 | int unit_get_tasks_current(Unit *u, uint64_t *ret) { |
2276 | _cleanup_free_ char *v = NULL; | |
2277 | int r; | |
2278 | ||
2279 | assert(u); | |
2280 | assert(ret); | |
2281 | ||
2e4025c0 | 2282 | if (!UNIT_CGROUP_BOOL(u, tasks_accounting)) |
cf3b4be1 LP |
2283 | return -ENODATA; |
2284 | ||
03a7b521 LP |
2285 | if (!u->cgroup_path) |
2286 | return -ENODATA; | |
2287 | ||
2288 | if ((u->cgroup_realized_mask & CGROUP_MASK_PIDS) == 0) | |
2289 | return -ENODATA; | |
2290 | ||
2291 | r = cg_get_attribute("pids", u->cgroup_path, "pids.current", &v); | |
2292 | if (r == -ENOENT) | |
2293 | return -ENODATA; | |
2294 | if (r < 0) | |
2295 | return r; | |
2296 | ||
2297 | return safe_atou64(v, ret); | |
2298 | } | |
2299 | ||
5ad096b3 LP |
2300 | static int unit_get_cpu_usage_raw(Unit *u, nsec_t *ret) { |
2301 | _cleanup_free_ char *v = NULL; | |
2302 | uint64_t ns; | |
2303 | int r; | |
2304 | ||
2305 | assert(u); | |
2306 | assert(ret); | |
2307 | ||
2308 | if (!u->cgroup_path) | |
2309 | return -ENODATA; | |
2310 | ||
b4cccbc1 LP |
2311 | r = cg_all_unified(); |
2312 | if (r < 0) | |
2313 | return r; | |
2314 | if (r > 0) { | |
66ebf6c0 TH |
2315 | const char *keys[] = { "usage_usec", NULL }; |
2316 | _cleanup_free_ char *val = NULL; | |
2317 | uint64_t us; | |
5ad096b3 | 2318 | |
66ebf6c0 TH |
2319 | if ((u->cgroup_realized_mask & CGROUP_MASK_CPU) == 0) |
2320 | return -ENODATA; | |
5ad096b3 | 2321 | |
66ebf6c0 TH |
2322 | r = cg_get_keyed_attribute("cpu", u->cgroup_path, "cpu.stat", keys, &val); |
2323 | if (r < 0) | |
2324 | return r; | |
2325 | ||
2326 | r = safe_atou64(val, &us); | |
2327 | if (r < 0) | |
2328 | return r; | |
2329 | ||
2330 | ns = us * NSEC_PER_USEC; | |
2331 | } else { | |
2332 | if ((u->cgroup_realized_mask & CGROUP_MASK_CPUACCT) == 0) | |
2333 | return -ENODATA; | |
2334 | ||
2335 | r = cg_get_attribute("cpuacct", u->cgroup_path, "cpuacct.usage", &v); | |
2336 | if (r == -ENOENT) | |
2337 | return -ENODATA; | |
2338 | if (r < 0) | |
2339 | return r; | |
2340 | ||
2341 | r = safe_atou64(v, &ns); | |
2342 | if (r < 0) | |
2343 | return r; | |
2344 | } | |
5ad096b3 LP |
2345 | |
2346 | *ret = ns; | |
2347 | return 0; | |
2348 | } | |
2349 | ||
2350 | int unit_get_cpu_usage(Unit *u, nsec_t *ret) { | |
2351 | nsec_t ns; | |
2352 | int r; | |
2353 | ||
fe700f46 LP |
2354 | assert(u); |
2355 | ||
2356 | /* Retrieve the current CPU usage counter. This will subtract the CPU counter taken when the unit was | |
2357 | * started. If the cgroup has been removed already, returns the last cached value. To cache the value, simply | |
2358 | * call this function with a NULL return value. */ | |
2359 | ||
2e4025c0 | 2360 | if (!UNIT_CGROUP_BOOL(u, cpu_accounting)) |
cf3b4be1 LP |
2361 | return -ENODATA; |
2362 | ||
5ad096b3 | 2363 | r = unit_get_cpu_usage_raw(u, &ns); |
fe700f46 LP |
2364 | if (r == -ENODATA && u->cpu_usage_last != NSEC_INFINITY) { |
2365 | /* If we can't get the CPU usage anymore (because the cgroup was already removed, for example), use our | |
2366 | * cached value. */ | |
2367 | ||
2368 | if (ret) | |
2369 | *ret = u->cpu_usage_last; | |
2370 | return 0; | |
2371 | } | |
5ad096b3 LP |
2372 | if (r < 0) |
2373 | return r; | |
2374 | ||
66ebf6c0 TH |
2375 | if (ns > u->cpu_usage_base) |
2376 | ns -= u->cpu_usage_base; | |
5ad096b3 LP |
2377 | else |
2378 | ns = 0; | |
2379 | ||
fe700f46 LP |
2380 | u->cpu_usage_last = ns; |
2381 | if (ret) | |
2382 | *ret = ns; | |
2383 | ||
5ad096b3 LP |
2384 | return 0; |
2385 | } | |
2386 | ||
906c06f6 DM |
2387 | int unit_get_ip_accounting( |
2388 | Unit *u, | |
2389 | CGroupIPAccountingMetric metric, | |
2390 | uint64_t *ret) { | |
2391 | ||
6b659ed8 | 2392 | uint64_t value; |
906c06f6 DM |
2393 | int fd, r; |
2394 | ||
2395 | assert(u); | |
2396 | assert(metric >= 0); | |
2397 | assert(metric < _CGROUP_IP_ACCOUNTING_METRIC_MAX); | |
2398 | assert(ret); | |
2399 | ||
cf3b4be1 LP |
2400 | /* IP accounting is currently not recursive, and hence we refuse to return any data for slice nodes. Slices are |
2401 | * inner cgroup nodes and hence have no processes directly attached, hence their counters would be zero | |
2402 | * anyway. And if we block this now we can later open this up, if the kernel learns recursive BPF cgroup | |
2403 | * filters. */ | |
2404 | if (u->type == UNIT_SLICE) | |
2405 | return -ENODATA; | |
2406 | ||
2e4025c0 | 2407 | if (!UNIT_CGROUP_BOOL(u, ip_accounting)) |
cf3b4be1 LP |
2408 | return -ENODATA; |
2409 | ||
906c06f6 DM |
2410 | fd = IN_SET(metric, CGROUP_IP_INGRESS_BYTES, CGROUP_IP_INGRESS_PACKETS) ? |
2411 | u->ip_accounting_ingress_map_fd : | |
2412 | u->ip_accounting_egress_map_fd; | |
906c06f6 DM |
2413 | if (fd < 0) |
2414 | return -ENODATA; | |
2415 | ||
2416 | if (IN_SET(metric, CGROUP_IP_INGRESS_BYTES, CGROUP_IP_EGRESS_BYTES)) | |
6b659ed8 | 2417 | r = bpf_firewall_read_accounting(fd, &value, NULL); |
906c06f6 | 2418 | else |
6b659ed8 LP |
2419 | r = bpf_firewall_read_accounting(fd, NULL, &value); |
2420 | if (r < 0) | |
2421 | return r; | |
2422 | ||
2423 | /* Add in additional metrics from a previous runtime. Note that when reexecing/reloading the daemon we compile | |
2424 | * all BPF programs and maps anew, but serialize the old counters. When deserializing we store them in the | |
2425 | * ip_accounting_extra[] field, and add them in here transparently. */ | |
2426 | ||
2427 | *ret = value + u->ip_accounting_extra[metric]; | |
906c06f6 DM |
2428 | |
2429 | return r; | |
2430 | } | |
2431 | ||
2432 | int unit_reset_cpu_accounting(Unit *u) { | |
5ad096b3 LP |
2433 | nsec_t ns; |
2434 | int r; | |
2435 | ||
2436 | assert(u); | |
2437 | ||
fe700f46 LP |
2438 | u->cpu_usage_last = NSEC_INFINITY; |
2439 | ||
5ad096b3 LP |
2440 | r = unit_get_cpu_usage_raw(u, &ns); |
2441 | if (r < 0) { | |
66ebf6c0 | 2442 | u->cpu_usage_base = 0; |
5ad096b3 | 2443 | return r; |
b56c28c3 | 2444 | } |
2633eb83 | 2445 | |
66ebf6c0 | 2446 | u->cpu_usage_base = ns; |
4ad49000 | 2447 | return 0; |
4fbf50b3 LP |
2448 | } |
2449 | ||
906c06f6 DM |
2450 | int unit_reset_ip_accounting(Unit *u) { |
2451 | int r = 0, q = 0; | |
2452 | ||
2453 | assert(u); | |
2454 | ||
2455 | if (u->ip_accounting_ingress_map_fd >= 0) | |
2456 | r = bpf_firewall_reset_accounting(u->ip_accounting_ingress_map_fd); | |
2457 | ||
2458 | if (u->ip_accounting_egress_map_fd >= 0) | |
2459 | q = bpf_firewall_reset_accounting(u->ip_accounting_egress_map_fd); | |
2460 | ||
6b659ed8 LP |
2461 | zero(u->ip_accounting_extra); |
2462 | ||
906c06f6 DM |
2463 | return r < 0 ? r : q; |
2464 | } | |
2465 | ||
e7ab4d1a LP |
2466 | void unit_invalidate_cgroup(Unit *u, CGroupMask m) { |
2467 | assert(u); | |
2468 | ||
2469 | if (!UNIT_HAS_CGROUP_CONTEXT(u)) | |
2470 | return; | |
2471 | ||
2472 | if (m == 0) | |
2473 | return; | |
2474 | ||
538b4852 TH |
2475 | /* always invalidate compat pairs together */ |
2476 | if (m & (CGROUP_MASK_IO | CGROUP_MASK_BLKIO)) | |
2477 | m |= CGROUP_MASK_IO | CGROUP_MASK_BLKIO; | |
2478 | ||
7cce4fb7 LP |
2479 | if (m & (CGROUP_MASK_CPU | CGROUP_MASK_CPUACCT)) |
2480 | m |= CGROUP_MASK_CPU | CGROUP_MASK_CPUACCT; | |
2481 | ||
60c728ad | 2482 | if ((u->cgroup_realized_mask & m) == 0) /* NOP? */ |
e7ab4d1a LP |
2483 | return; |
2484 | ||
2485 | u->cgroup_realized_mask &= ~m; | |
91a6073e | 2486 | unit_add_to_cgroup_realize_queue(u); |
e7ab4d1a LP |
2487 | } |
2488 | ||
906c06f6 DM |
2489 | void unit_invalidate_cgroup_bpf(Unit *u) { |
2490 | assert(u); | |
2491 | ||
2492 | if (!UNIT_HAS_CGROUP_CONTEXT(u)) | |
2493 | return; | |
2494 | ||
60c728ad | 2495 | if (u->cgroup_bpf_state == UNIT_CGROUP_BPF_INVALIDATED) /* NOP? */ |
906c06f6 DM |
2496 | return; |
2497 | ||
2498 | u->cgroup_bpf_state = UNIT_CGROUP_BPF_INVALIDATED; | |
91a6073e | 2499 | unit_add_to_cgroup_realize_queue(u); |
906c06f6 DM |
2500 | |
2501 | /* If we are a slice unit, we also need to put compile a new BPF program for all our children, as the IP access | |
2502 | * list of our children includes our own. */ | |
2503 | if (u->type == UNIT_SLICE) { | |
2504 | Unit *member; | |
2505 | Iterator i; | |
eef85c4a | 2506 | void *v; |
906c06f6 | 2507 | |
eef85c4a | 2508 | HASHMAP_FOREACH_KEY(v, member, u->dependencies[UNIT_BEFORE], i) { |
906c06f6 DM |
2509 | if (member == u) |
2510 | continue; | |
2511 | ||
2512 | if (UNIT_DEREF(member->slice) != u) | |
2513 | continue; | |
2514 | ||
2515 | unit_invalidate_cgroup_bpf(member); | |
2516 | } | |
2517 | } | |
2518 | } | |
2519 | ||
e7ab4d1a LP |
2520 | void manager_invalidate_startup_units(Manager *m) { |
2521 | Iterator i; | |
2522 | Unit *u; | |
2523 | ||
2524 | assert(m); | |
2525 | ||
2526 | SET_FOREACH(u, m->startup_units, i) | |
13c31542 | 2527 | unit_invalidate_cgroup(u, CGROUP_MASK_CPU|CGROUP_MASK_IO|CGROUP_MASK_BLKIO); |
e7ab4d1a LP |
2528 | } |
2529 | ||
4ad49000 LP |
2530 | static const char* const cgroup_device_policy_table[_CGROUP_DEVICE_POLICY_MAX] = { |
2531 | [CGROUP_AUTO] = "auto", | |
2532 | [CGROUP_CLOSED] = "closed", | |
2533 | [CGROUP_STRICT] = "strict", | |
2534 | }; | |
4fbf50b3 | 2535 | |
4ad49000 | 2536 | DEFINE_STRING_TABLE_LOOKUP(cgroup_device_policy, CGroupDevicePolicy); |