]>
Commit | Line | Data |
---|---|---|
53e1b683 | 1 | /* SPDX-License-Identifier: LGPL-2.1+ */ |
8e274523 | 2 | |
c6c18be3 | 3 | #include <fcntl.h> |
e41969e3 | 4 | #include <fnmatch.h> |
8c6db833 | 5 | |
b5efdb8a | 6 | #include "alloc-util.h" |
18c528e9 | 7 | #include "blockdev-util.h" |
906c06f6 | 8 | #include "bpf-firewall.h" |
45c2e068 | 9 | #include "btrfs-util.h" |
6592b975 | 10 | #include "bus-error.h" |
03a7b521 | 11 | #include "cgroup-util.h" |
3ffd4af2 LP |
12 | #include "cgroup.h" |
13 | #include "fd-util.h" | |
0d39fa9c | 14 | #include "fileio.h" |
77601719 | 15 | #include "fs-util.h" |
6bedfcbb | 16 | #include "parse-util.h" |
9eb977db | 17 | #include "path-util.h" |
03a7b521 | 18 | #include "process-util.h" |
c36a69f4 | 19 | #include "procfs-util.h" |
9444b1f2 | 20 | #include "special.h" |
906c06f6 | 21 | #include "stdio-util.h" |
8b43440b | 22 | #include "string-table.h" |
07630cea | 23 | #include "string-util.h" |
cc6271f1 | 24 | #include "virt.h" |
8e274523 | 25 | |
9a054909 LP |
26 | #define CGROUP_CPU_QUOTA_PERIOD_USEC ((usec_t) 100 * USEC_PER_MSEC) |
27 | ||
cc6271f1 LP |
28 | bool manager_owns_root_cgroup(Manager *m) { |
29 | assert(m); | |
30 | ||
31 | /* Returns true if we are managing the root cgroup. Note that it isn't sufficient to just check whether the | |
32 | * group root path equals "/" since that will also be the case if CLONE_NEWCGROUP is in the mix. Since there's | |
33 | * appears to be no nice way to detect whether we are in a CLONE_NEWCGROUP namespace we instead just check if | |
34 | * we run in any kind of container virtualization. */ | |
35 | ||
36 | if (detect_container() > 0) | |
37 | return false; | |
38 | ||
57ea45e1 | 39 | return empty_or_root(m->cgroup_root); |
cc6271f1 LP |
40 | } |
41 | ||
f3725e64 LP |
42 | bool unit_has_root_cgroup(Unit *u) { |
43 | assert(u); | |
44 | ||
cc6271f1 LP |
45 | /* Returns whether this unit manages the root cgroup. This will return true if this unit is the root slice and |
46 | * the manager manages the root cgroup. */ | |
f3725e64 | 47 | |
cc6271f1 | 48 | if (!manager_owns_root_cgroup(u->manager)) |
f3725e64 LP |
49 | return false; |
50 | ||
cc6271f1 | 51 | return unit_has_name(u, SPECIAL_ROOT_SLICE); |
f3725e64 LP |
52 | } |
53 | ||
2b40998d | 54 | static void cgroup_compat_warn(void) { |
128fadc9 TH |
55 | static bool cgroup_compat_warned = false; |
56 | ||
57 | if (cgroup_compat_warned) | |
58 | return; | |
59 | ||
cc6271f1 LP |
60 | log_warning("cgroup compatibility translation between legacy and unified hierarchy settings activated. " |
61 | "See cgroup-compat debug messages for details."); | |
62 | ||
128fadc9 TH |
63 | cgroup_compat_warned = true; |
64 | } | |
65 | ||
66 | #define log_cgroup_compat(unit, fmt, ...) do { \ | |
67 | cgroup_compat_warn(); \ | |
68 | log_unit_debug(unit, "cgroup-compat: " fmt, ##__VA_ARGS__); \ | |
2b40998d | 69 | } while (false) |
128fadc9 | 70 | |
4ad49000 LP |
71 | void cgroup_context_init(CGroupContext *c) { |
72 | assert(c); | |
73 | ||
74 | /* Initialize everything to the kernel defaults, assuming the | |
75 | * structure is preinitialized to 0 */ | |
76 | ||
66ebf6c0 TH |
77 | c->cpu_weight = CGROUP_WEIGHT_INVALID; |
78 | c->startup_cpu_weight = CGROUP_WEIGHT_INVALID; | |
79 | c->cpu_quota_per_sec_usec = USEC_INFINITY; | |
80 | ||
d53d9474 LP |
81 | c->cpu_shares = CGROUP_CPU_SHARES_INVALID; |
82 | c->startup_cpu_shares = CGROUP_CPU_SHARES_INVALID; | |
d53d9474 | 83 | |
da4d897e TH |
84 | c->memory_high = CGROUP_LIMIT_MAX; |
85 | c->memory_max = CGROUP_LIMIT_MAX; | |
96e131ea | 86 | c->memory_swap_max = CGROUP_LIMIT_MAX; |
da4d897e TH |
87 | |
88 | c->memory_limit = CGROUP_LIMIT_MAX; | |
b2f8b02e | 89 | |
13c31542 TH |
90 | c->io_weight = CGROUP_WEIGHT_INVALID; |
91 | c->startup_io_weight = CGROUP_WEIGHT_INVALID; | |
92 | ||
d53d9474 LP |
93 | c->blockio_weight = CGROUP_BLKIO_WEIGHT_INVALID; |
94 | c->startup_blockio_weight = CGROUP_BLKIO_WEIGHT_INVALID; | |
95 | ||
96 | c->tasks_max = (uint64_t) -1; | |
4ad49000 | 97 | } |
8e274523 | 98 | |
4ad49000 LP |
99 | void cgroup_context_free_device_allow(CGroupContext *c, CGroupDeviceAllow *a) { |
100 | assert(c); | |
101 | assert(a); | |
102 | ||
71fda00f | 103 | LIST_REMOVE(device_allow, c->device_allow, a); |
4ad49000 LP |
104 | free(a->path); |
105 | free(a); | |
106 | } | |
107 | ||
13c31542 TH |
108 | void cgroup_context_free_io_device_weight(CGroupContext *c, CGroupIODeviceWeight *w) { |
109 | assert(c); | |
110 | assert(w); | |
111 | ||
112 | LIST_REMOVE(device_weights, c->io_device_weights, w); | |
113 | free(w->path); | |
114 | free(w); | |
115 | } | |
116 | ||
117 | void cgroup_context_free_io_device_limit(CGroupContext *c, CGroupIODeviceLimit *l) { | |
118 | assert(c); | |
119 | assert(l); | |
120 | ||
121 | LIST_REMOVE(device_limits, c->io_device_limits, l); | |
122 | free(l->path); | |
123 | free(l); | |
124 | } | |
125 | ||
4ad49000 LP |
126 | void cgroup_context_free_blockio_device_weight(CGroupContext *c, CGroupBlockIODeviceWeight *w) { |
127 | assert(c); | |
128 | assert(w); | |
129 | ||
71fda00f | 130 | LIST_REMOVE(device_weights, c->blockio_device_weights, w); |
4ad49000 LP |
131 | free(w->path); |
132 | free(w); | |
133 | } | |
134 | ||
135 | void cgroup_context_free_blockio_device_bandwidth(CGroupContext *c, CGroupBlockIODeviceBandwidth *b) { | |
136 | assert(c); | |
8e274523 | 137 | assert(b); |
8e274523 | 138 | |
71fda00f | 139 | LIST_REMOVE(device_bandwidths, c->blockio_device_bandwidths, b); |
4ad49000 LP |
140 | free(b->path); |
141 | free(b); | |
142 | } | |
143 | ||
144 | void cgroup_context_done(CGroupContext *c) { | |
145 | assert(c); | |
146 | ||
13c31542 TH |
147 | while (c->io_device_weights) |
148 | cgroup_context_free_io_device_weight(c, c->io_device_weights); | |
149 | ||
150 | while (c->io_device_limits) | |
151 | cgroup_context_free_io_device_limit(c, c->io_device_limits); | |
152 | ||
4ad49000 LP |
153 | while (c->blockio_device_weights) |
154 | cgroup_context_free_blockio_device_weight(c, c->blockio_device_weights); | |
155 | ||
156 | while (c->blockio_device_bandwidths) | |
157 | cgroup_context_free_blockio_device_bandwidth(c, c->blockio_device_bandwidths); | |
158 | ||
159 | while (c->device_allow) | |
160 | cgroup_context_free_device_allow(c, c->device_allow); | |
6a48d82f DM |
161 | |
162 | c->ip_address_allow = ip_address_access_free_all(c->ip_address_allow); | |
163 | c->ip_address_deny = ip_address_access_free_all(c->ip_address_deny); | |
4ad49000 LP |
164 | } |
165 | ||
166 | void cgroup_context_dump(CGroupContext *c, FILE* f, const char *prefix) { | |
13c31542 TH |
167 | CGroupIODeviceLimit *il; |
168 | CGroupIODeviceWeight *iw; | |
4ad49000 LP |
169 | CGroupBlockIODeviceBandwidth *b; |
170 | CGroupBlockIODeviceWeight *w; | |
171 | CGroupDeviceAllow *a; | |
c21c9906 | 172 | IPAddressAccessItem *iaai; |
9a054909 | 173 | char u[FORMAT_TIMESPAN_MAX]; |
4ad49000 LP |
174 | |
175 | assert(c); | |
176 | assert(f); | |
177 | ||
178 | prefix = strempty(prefix); | |
179 | ||
180 | fprintf(f, | |
181 | "%sCPUAccounting=%s\n" | |
13c31542 | 182 | "%sIOAccounting=%s\n" |
4ad49000 LP |
183 | "%sBlockIOAccounting=%s\n" |
184 | "%sMemoryAccounting=%s\n" | |
d53d9474 | 185 | "%sTasksAccounting=%s\n" |
c21c9906 | 186 | "%sIPAccounting=%s\n" |
66ebf6c0 TH |
187 | "%sCPUWeight=%" PRIu64 "\n" |
188 | "%sStartupCPUWeight=%" PRIu64 "\n" | |
d53d9474 LP |
189 | "%sCPUShares=%" PRIu64 "\n" |
190 | "%sStartupCPUShares=%" PRIu64 "\n" | |
b2f8b02e | 191 | "%sCPUQuotaPerSecSec=%s\n" |
13c31542 TH |
192 | "%sIOWeight=%" PRIu64 "\n" |
193 | "%sStartupIOWeight=%" PRIu64 "\n" | |
d53d9474 LP |
194 | "%sBlockIOWeight=%" PRIu64 "\n" |
195 | "%sStartupBlockIOWeight=%" PRIu64 "\n" | |
48422635 | 196 | "%sMemoryMin=%" PRIu64 "\n" |
da4d897e TH |
197 | "%sMemoryLow=%" PRIu64 "\n" |
198 | "%sMemoryHigh=%" PRIu64 "\n" | |
199 | "%sMemoryMax=%" PRIu64 "\n" | |
96e131ea | 200 | "%sMemorySwapMax=%" PRIu64 "\n" |
4ad49000 | 201 | "%sMemoryLimit=%" PRIu64 "\n" |
03a7b521 | 202 | "%sTasksMax=%" PRIu64 "\n" |
a931ad47 LP |
203 | "%sDevicePolicy=%s\n" |
204 | "%sDelegate=%s\n", | |
4ad49000 | 205 | prefix, yes_no(c->cpu_accounting), |
13c31542 | 206 | prefix, yes_no(c->io_accounting), |
4ad49000 LP |
207 | prefix, yes_no(c->blockio_accounting), |
208 | prefix, yes_no(c->memory_accounting), | |
d53d9474 | 209 | prefix, yes_no(c->tasks_accounting), |
c21c9906 | 210 | prefix, yes_no(c->ip_accounting), |
66ebf6c0 TH |
211 | prefix, c->cpu_weight, |
212 | prefix, c->startup_cpu_weight, | |
4ad49000 | 213 | prefix, c->cpu_shares, |
95ae05c0 | 214 | prefix, c->startup_cpu_shares, |
b1d6dcf5 | 215 | prefix, format_timespan(u, sizeof(u), c->cpu_quota_per_sec_usec, 1), |
13c31542 TH |
216 | prefix, c->io_weight, |
217 | prefix, c->startup_io_weight, | |
4ad49000 | 218 | prefix, c->blockio_weight, |
95ae05c0 | 219 | prefix, c->startup_blockio_weight, |
48422635 | 220 | prefix, c->memory_min, |
da4d897e TH |
221 | prefix, c->memory_low, |
222 | prefix, c->memory_high, | |
223 | prefix, c->memory_max, | |
96e131ea | 224 | prefix, c->memory_swap_max, |
4ad49000 | 225 | prefix, c->memory_limit, |
03a7b521 | 226 | prefix, c->tasks_max, |
a931ad47 LP |
227 | prefix, cgroup_device_policy_to_string(c->device_policy), |
228 | prefix, yes_no(c->delegate)); | |
4ad49000 | 229 | |
02638280 LP |
230 | if (c->delegate) { |
231 | _cleanup_free_ char *t = NULL; | |
232 | ||
233 | (void) cg_mask_to_string(c->delegate_controllers, &t); | |
234 | ||
47a78d41 | 235 | fprintf(f, "%sDelegateControllers=%s\n", |
02638280 LP |
236 | prefix, |
237 | strempty(t)); | |
238 | } | |
239 | ||
4ad49000 LP |
240 | LIST_FOREACH(device_allow, a, c->device_allow) |
241 | fprintf(f, | |
242 | "%sDeviceAllow=%s %s%s%s\n", | |
243 | prefix, | |
244 | a->path, | |
245 | a->r ? "r" : "", a->w ? "w" : "", a->m ? "m" : ""); | |
246 | ||
13c31542 TH |
247 | LIST_FOREACH(device_weights, iw, c->io_device_weights) |
248 | fprintf(f, | |
249 | "%sIODeviceWeight=%s %" PRIu64, | |
250 | prefix, | |
251 | iw->path, | |
252 | iw->weight); | |
253 | ||
254 | LIST_FOREACH(device_limits, il, c->io_device_limits) { | |
255 | char buf[FORMAT_BYTES_MAX]; | |
9be57249 TH |
256 | CGroupIOLimitType type; |
257 | ||
258 | for (type = 0; type < _CGROUP_IO_LIMIT_TYPE_MAX; type++) | |
259 | if (il->limits[type] != cgroup_io_limit_defaults[type]) | |
260 | fprintf(f, | |
261 | "%s%s=%s %s\n", | |
262 | prefix, | |
263 | cgroup_io_limit_type_to_string(type), | |
264 | il->path, | |
265 | format_bytes(buf, sizeof(buf), il->limits[type])); | |
13c31542 TH |
266 | } |
267 | ||
4ad49000 LP |
268 | LIST_FOREACH(device_weights, w, c->blockio_device_weights) |
269 | fprintf(f, | |
d53d9474 | 270 | "%sBlockIODeviceWeight=%s %" PRIu64, |
4ad49000 LP |
271 | prefix, |
272 | w->path, | |
273 | w->weight); | |
274 | ||
275 | LIST_FOREACH(device_bandwidths, b, c->blockio_device_bandwidths) { | |
276 | char buf[FORMAT_BYTES_MAX]; | |
277 | ||
979d0311 TH |
278 | if (b->rbps != CGROUP_LIMIT_MAX) |
279 | fprintf(f, | |
280 | "%sBlockIOReadBandwidth=%s %s\n", | |
281 | prefix, | |
282 | b->path, | |
283 | format_bytes(buf, sizeof(buf), b->rbps)); | |
284 | if (b->wbps != CGROUP_LIMIT_MAX) | |
285 | fprintf(f, | |
286 | "%sBlockIOWriteBandwidth=%s %s\n", | |
287 | prefix, | |
288 | b->path, | |
289 | format_bytes(buf, sizeof(buf), b->wbps)); | |
4ad49000 | 290 | } |
c21c9906 LP |
291 | |
292 | LIST_FOREACH(items, iaai, c->ip_address_allow) { | |
293 | _cleanup_free_ char *k = NULL; | |
294 | ||
295 | (void) in_addr_to_string(iaai->family, &iaai->address, &k); | |
296 | fprintf(f, "%sIPAddressAllow=%s/%u\n", prefix, strnull(k), iaai->prefixlen); | |
297 | } | |
298 | ||
299 | LIST_FOREACH(items, iaai, c->ip_address_deny) { | |
300 | _cleanup_free_ char *k = NULL; | |
301 | ||
302 | (void) in_addr_to_string(iaai->family, &iaai->address, &k); | |
303 | fprintf(f, "%sIPAddressDeny=%s/%u\n", prefix, strnull(k), iaai->prefixlen); | |
304 | } | |
4ad49000 LP |
305 | } |
306 | ||
45c2e068 | 307 | static int lookup_block_device(const char *p, dev_t *ret) { |
4ad49000 | 308 | struct stat st; |
45c2e068 | 309 | int r; |
4ad49000 LP |
310 | |
311 | assert(p); | |
45c2e068 | 312 | assert(ret); |
4ad49000 | 313 | |
b1c05b98 | 314 | if (stat(p, &st) < 0) |
45c2e068 | 315 | return log_warning_errno(errno, "Couldn't stat device '%s': %m", p); |
8e274523 | 316 | |
4ad49000 | 317 | if (S_ISBLK(st.st_mode)) |
45c2e068 LP |
318 | *ret = st.st_rdev; |
319 | else if (major(st.st_dev) != 0) | |
320 | *ret = st.st_dev; /* If this is not a device node then use the block device this file is stored on */ | |
321 | else { | |
322 | /* If this is btrfs, getting the backing block device is a bit harder */ | |
323 | r = btrfs_get_block_device(p, ret); | |
324 | if (r < 0 && r != -ENOTTY) | |
325 | return log_warning_errno(r, "Failed to determine block device backing btrfs file system '%s': %m", p); | |
326 | if (r == -ENOTTY) { | |
327 | log_warning("'%s' is not a block device node, and file system block device cannot be determined or is not local.", p); | |
328 | return -ENODEV; | |
329 | } | |
4ad49000 | 330 | } |
8e274523 | 331 | |
45c2e068 LP |
332 | /* If this is a LUKS device, try to get the originating block device */ |
333 | (void) block_get_originating(*ret, ret); | |
334 | ||
335 | /* If this is a partition, try to get the originating block device */ | |
336 | (void) block_get_whole_disk(*ret, ret); | |
8e274523 | 337 | return 0; |
8e274523 LP |
338 | } |
339 | ||
4ad49000 LP |
340 | static int whitelist_device(const char *path, const char *node, const char *acc) { |
341 | char buf[2+DECIMAL_STR_MAX(dev_t)*2+2+4]; | |
342 | struct stat st; | |
b200489b | 343 | bool ignore_notfound; |
8c6db833 | 344 | int r; |
8e274523 | 345 | |
4ad49000 LP |
346 | assert(path); |
347 | assert(acc); | |
8e274523 | 348 | |
b200489b DR |
349 | if (node[0] == '-') { |
350 | /* Non-existent paths starting with "-" must be silently ignored */ | |
351 | node++; | |
352 | ignore_notfound = true; | |
353 | } else | |
354 | ignore_notfound = false; | |
355 | ||
4ad49000 | 356 | if (stat(node, &st) < 0) { |
b200489b | 357 | if (errno == ENOENT && ignore_notfound) |
e7330dfe DP |
358 | return 0; |
359 | ||
360 | return log_warning_errno(errno, "Couldn't stat device %s: %m", node); | |
4ad49000 LP |
361 | } |
362 | ||
363 | if (!S_ISCHR(st.st_mode) && !S_ISBLK(st.st_mode)) { | |
364 | log_warning("%s is not a device.", node); | |
365 | return -ENODEV; | |
366 | } | |
367 | ||
368 | sprintf(buf, | |
369 | "%c %u:%u %s", | |
370 | S_ISCHR(st.st_mode) ? 'c' : 'b', | |
371 | major(st.st_rdev), minor(st.st_rdev), | |
372 | acc); | |
373 | ||
374 | r = cg_set_attribute("devices", path, "devices.allow", buf); | |
1aeab12b | 375 | if (r < 0) |
077ba06e | 376 | log_full_errno(IN_SET(r, -ENOENT, -EROFS, -EINVAL, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, |
714e2e1d | 377 | "Failed to set devices.allow on %s: %m", path); |
4ad49000 LP |
378 | |
379 | return r; | |
8e274523 LP |
380 | } |
381 | ||
90060676 LP |
382 | static int whitelist_major(const char *path, const char *name, char type, const char *acc) { |
383 | _cleanup_fclose_ FILE *f = NULL; | |
384 | char line[LINE_MAX]; | |
385 | bool good = false; | |
386 | int r; | |
387 | ||
388 | assert(path); | |
389 | assert(acc); | |
4c701096 | 390 | assert(IN_SET(type, 'b', 'c')); |
90060676 LP |
391 | |
392 | f = fopen("/proc/devices", "re"); | |
4a62c710 MS |
393 | if (!f) |
394 | return log_warning_errno(errno, "Cannot open /proc/devices to resolve %s (%c): %m", name, type); | |
90060676 LP |
395 | |
396 | FOREACH_LINE(line, f, goto fail) { | |
397 | char buf[2+DECIMAL_STR_MAX(unsigned)+3+4], *p, *w; | |
398 | unsigned maj; | |
399 | ||
400 | truncate_nl(line); | |
401 | ||
402 | if (type == 'c' && streq(line, "Character devices:")) { | |
403 | good = true; | |
404 | continue; | |
405 | } | |
406 | ||
407 | if (type == 'b' && streq(line, "Block devices:")) { | |
408 | good = true; | |
409 | continue; | |
410 | } | |
411 | ||
412 | if (isempty(line)) { | |
413 | good = false; | |
414 | continue; | |
415 | } | |
416 | ||
417 | if (!good) | |
418 | continue; | |
419 | ||
420 | p = strstrip(line); | |
421 | ||
422 | w = strpbrk(p, WHITESPACE); | |
423 | if (!w) | |
424 | continue; | |
425 | *w = 0; | |
426 | ||
427 | r = safe_atou(p, &maj); | |
428 | if (r < 0) | |
429 | continue; | |
430 | if (maj <= 0) | |
431 | continue; | |
432 | ||
433 | w++; | |
434 | w += strspn(w, WHITESPACE); | |
e41969e3 LP |
435 | |
436 | if (fnmatch(name, w, 0) != 0) | |
90060676 LP |
437 | continue; |
438 | ||
439 | sprintf(buf, | |
440 | "%c %u:* %s", | |
441 | type, | |
442 | maj, | |
443 | acc); | |
444 | ||
445 | r = cg_set_attribute("devices", path, "devices.allow", buf); | |
1aeab12b | 446 | if (r < 0) |
077ba06e | 447 | log_full_errno(IN_SET(r, -ENOENT, -EROFS, -EINVAL, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, |
714e2e1d | 448 | "Failed to set devices.allow on %s: %m", path); |
90060676 LP |
449 | } |
450 | ||
451 | return 0; | |
452 | ||
453 | fail: | |
25f027c5 | 454 | return log_warning_errno(errno, "Failed to read /proc/devices: %m"); |
90060676 LP |
455 | } |
456 | ||
66ebf6c0 TH |
457 | static bool cgroup_context_has_cpu_weight(CGroupContext *c) { |
458 | return c->cpu_weight != CGROUP_WEIGHT_INVALID || | |
459 | c->startup_cpu_weight != CGROUP_WEIGHT_INVALID; | |
460 | } | |
461 | ||
462 | static bool cgroup_context_has_cpu_shares(CGroupContext *c) { | |
463 | return c->cpu_shares != CGROUP_CPU_SHARES_INVALID || | |
464 | c->startup_cpu_shares != CGROUP_CPU_SHARES_INVALID; | |
465 | } | |
466 | ||
467 | static uint64_t cgroup_context_cpu_weight(CGroupContext *c, ManagerState state) { | |
468 | if (IN_SET(state, MANAGER_STARTING, MANAGER_INITIALIZING) && | |
469 | c->startup_cpu_weight != CGROUP_WEIGHT_INVALID) | |
470 | return c->startup_cpu_weight; | |
471 | else if (c->cpu_weight != CGROUP_WEIGHT_INVALID) | |
472 | return c->cpu_weight; | |
473 | else | |
474 | return CGROUP_WEIGHT_DEFAULT; | |
475 | } | |
476 | ||
477 | static uint64_t cgroup_context_cpu_shares(CGroupContext *c, ManagerState state) { | |
478 | if (IN_SET(state, MANAGER_STARTING, MANAGER_INITIALIZING) && | |
479 | c->startup_cpu_shares != CGROUP_CPU_SHARES_INVALID) | |
480 | return c->startup_cpu_shares; | |
481 | else if (c->cpu_shares != CGROUP_CPU_SHARES_INVALID) | |
482 | return c->cpu_shares; | |
483 | else | |
484 | return CGROUP_CPU_SHARES_DEFAULT; | |
485 | } | |
486 | ||
487 | static void cgroup_apply_unified_cpu_config(Unit *u, uint64_t weight, uint64_t quota) { | |
488 | char buf[MAX(DECIMAL_STR_MAX(uint64_t) + 1, (DECIMAL_STR_MAX(usec_t) + 1) * 2)]; | |
489 | int r; | |
490 | ||
491 | xsprintf(buf, "%" PRIu64 "\n", weight); | |
492 | r = cg_set_attribute("cpu", u->cgroup_path, "cpu.weight", buf); | |
493 | if (r < 0) | |
494 | log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, | |
495 | "Failed to set cpu.weight: %m"); | |
496 | ||
497 | if (quota != USEC_INFINITY) | |
498 | xsprintf(buf, USEC_FMT " " USEC_FMT "\n", | |
499 | quota * CGROUP_CPU_QUOTA_PERIOD_USEC / USEC_PER_SEC, CGROUP_CPU_QUOTA_PERIOD_USEC); | |
500 | else | |
501 | xsprintf(buf, "max " USEC_FMT "\n", CGROUP_CPU_QUOTA_PERIOD_USEC); | |
502 | ||
503 | r = cg_set_attribute("cpu", u->cgroup_path, "cpu.max", buf); | |
504 | ||
505 | if (r < 0) | |
506 | log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, | |
507 | "Failed to set cpu.max: %m"); | |
508 | } | |
509 | ||
510 | static void cgroup_apply_legacy_cpu_config(Unit *u, uint64_t shares, uint64_t quota) { | |
511 | char buf[MAX(DECIMAL_STR_MAX(uint64_t), DECIMAL_STR_MAX(usec_t)) + 1]; | |
512 | int r; | |
513 | ||
514 | xsprintf(buf, "%" PRIu64 "\n", shares); | |
515 | r = cg_set_attribute("cpu", u->cgroup_path, "cpu.shares", buf); | |
516 | if (r < 0) | |
517 | log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, | |
518 | "Failed to set cpu.shares: %m"); | |
519 | ||
520 | xsprintf(buf, USEC_FMT "\n", CGROUP_CPU_QUOTA_PERIOD_USEC); | |
521 | r = cg_set_attribute("cpu", u->cgroup_path, "cpu.cfs_period_us", buf); | |
522 | if (r < 0) | |
523 | log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, | |
524 | "Failed to set cpu.cfs_period_us: %m"); | |
525 | ||
526 | if (quota != USEC_INFINITY) { | |
527 | xsprintf(buf, USEC_FMT "\n", quota * CGROUP_CPU_QUOTA_PERIOD_USEC / USEC_PER_SEC); | |
528 | r = cg_set_attribute("cpu", u->cgroup_path, "cpu.cfs_quota_us", buf); | |
529 | } else | |
530 | r = cg_set_attribute("cpu", u->cgroup_path, "cpu.cfs_quota_us", "-1"); | |
531 | if (r < 0) | |
532 | log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, | |
533 | "Failed to set cpu.cfs_quota_us: %m"); | |
534 | } | |
535 | ||
536 | static uint64_t cgroup_cpu_shares_to_weight(uint64_t shares) { | |
537 | return CLAMP(shares * CGROUP_WEIGHT_DEFAULT / CGROUP_CPU_SHARES_DEFAULT, | |
538 | CGROUP_WEIGHT_MIN, CGROUP_WEIGHT_MAX); | |
539 | } | |
540 | ||
541 | static uint64_t cgroup_cpu_weight_to_shares(uint64_t weight) { | |
542 | return CLAMP(weight * CGROUP_CPU_SHARES_DEFAULT / CGROUP_WEIGHT_DEFAULT, | |
543 | CGROUP_CPU_SHARES_MIN, CGROUP_CPU_SHARES_MAX); | |
544 | } | |
545 | ||
508c45da | 546 | static bool cgroup_context_has_io_config(CGroupContext *c) { |
538b4852 TH |
547 | return c->io_accounting || |
548 | c->io_weight != CGROUP_WEIGHT_INVALID || | |
549 | c->startup_io_weight != CGROUP_WEIGHT_INVALID || | |
550 | c->io_device_weights || | |
551 | c->io_device_limits; | |
552 | } | |
553 | ||
508c45da | 554 | static bool cgroup_context_has_blockio_config(CGroupContext *c) { |
538b4852 TH |
555 | return c->blockio_accounting || |
556 | c->blockio_weight != CGROUP_BLKIO_WEIGHT_INVALID || | |
557 | c->startup_blockio_weight != CGROUP_BLKIO_WEIGHT_INVALID || | |
558 | c->blockio_device_weights || | |
559 | c->blockio_device_bandwidths; | |
560 | } | |
561 | ||
508c45da | 562 | static uint64_t cgroup_context_io_weight(CGroupContext *c, ManagerState state) { |
64faf04c TH |
563 | if (IN_SET(state, MANAGER_STARTING, MANAGER_INITIALIZING) && |
564 | c->startup_io_weight != CGROUP_WEIGHT_INVALID) | |
565 | return c->startup_io_weight; | |
566 | else if (c->io_weight != CGROUP_WEIGHT_INVALID) | |
567 | return c->io_weight; | |
568 | else | |
569 | return CGROUP_WEIGHT_DEFAULT; | |
570 | } | |
571 | ||
508c45da | 572 | static uint64_t cgroup_context_blkio_weight(CGroupContext *c, ManagerState state) { |
64faf04c TH |
573 | if (IN_SET(state, MANAGER_STARTING, MANAGER_INITIALIZING) && |
574 | c->startup_blockio_weight != CGROUP_BLKIO_WEIGHT_INVALID) | |
575 | return c->startup_blockio_weight; | |
576 | else if (c->blockio_weight != CGROUP_BLKIO_WEIGHT_INVALID) | |
577 | return c->blockio_weight; | |
578 | else | |
579 | return CGROUP_BLKIO_WEIGHT_DEFAULT; | |
580 | } | |
581 | ||
508c45da | 582 | static uint64_t cgroup_weight_blkio_to_io(uint64_t blkio_weight) { |
538b4852 TH |
583 | return CLAMP(blkio_weight * CGROUP_WEIGHT_DEFAULT / CGROUP_BLKIO_WEIGHT_DEFAULT, |
584 | CGROUP_WEIGHT_MIN, CGROUP_WEIGHT_MAX); | |
585 | } | |
586 | ||
508c45da | 587 | static uint64_t cgroup_weight_io_to_blkio(uint64_t io_weight) { |
538b4852 TH |
588 | return CLAMP(io_weight * CGROUP_BLKIO_WEIGHT_DEFAULT / CGROUP_WEIGHT_DEFAULT, |
589 | CGROUP_BLKIO_WEIGHT_MIN, CGROUP_BLKIO_WEIGHT_MAX); | |
590 | } | |
591 | ||
f29ff115 | 592 | static void cgroup_apply_io_device_weight(Unit *u, const char *dev_path, uint64_t io_weight) { |
64faf04c TH |
593 | char buf[DECIMAL_STR_MAX(dev_t)*2+2+DECIMAL_STR_MAX(uint64_t)+1]; |
594 | dev_t dev; | |
595 | int r; | |
596 | ||
597 | r = lookup_block_device(dev_path, &dev); | |
598 | if (r < 0) | |
599 | return; | |
600 | ||
601 | xsprintf(buf, "%u:%u %" PRIu64 "\n", major(dev), minor(dev), io_weight); | |
f29ff115 | 602 | r = cg_set_attribute("io", u->cgroup_path, "io.weight", buf); |
64faf04c | 603 | if (r < 0) |
f29ff115 TH |
604 | log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, |
605 | "Failed to set io.weight: %m"); | |
64faf04c TH |
606 | } |
607 | ||
f29ff115 | 608 | static void cgroup_apply_blkio_device_weight(Unit *u, const char *dev_path, uint64_t blkio_weight) { |
64faf04c TH |
609 | char buf[DECIMAL_STR_MAX(dev_t)*2+2+DECIMAL_STR_MAX(uint64_t)+1]; |
610 | dev_t dev; | |
611 | int r; | |
612 | ||
613 | r = lookup_block_device(dev_path, &dev); | |
614 | if (r < 0) | |
615 | return; | |
616 | ||
617 | xsprintf(buf, "%u:%u %" PRIu64 "\n", major(dev), minor(dev), blkio_weight); | |
f29ff115 | 618 | r = cg_set_attribute("blkio", u->cgroup_path, "blkio.weight_device", buf); |
64faf04c | 619 | if (r < 0) |
f29ff115 TH |
620 | log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, |
621 | "Failed to set blkio.weight_device: %m"); | |
64faf04c TH |
622 | } |
623 | ||
17ae2780 | 624 | static void cgroup_apply_io_device_limit(Unit *u, const char *dev_path, uint64_t *limits) { |
64faf04c TH |
625 | char limit_bufs[_CGROUP_IO_LIMIT_TYPE_MAX][DECIMAL_STR_MAX(uint64_t)]; |
626 | char buf[DECIMAL_STR_MAX(dev_t)*2+2+(6+DECIMAL_STR_MAX(uint64_t)+1)*4]; | |
627 | CGroupIOLimitType type; | |
628 | dev_t dev; | |
64faf04c TH |
629 | int r; |
630 | ||
631 | r = lookup_block_device(dev_path, &dev); | |
632 | if (r < 0) | |
17ae2780 | 633 | return; |
64faf04c | 634 | |
17ae2780 LP |
635 | for (type = 0; type < _CGROUP_IO_LIMIT_TYPE_MAX; type++) |
636 | if (limits[type] != cgroup_io_limit_defaults[type]) | |
64faf04c | 637 | xsprintf(limit_bufs[type], "%" PRIu64, limits[type]); |
17ae2780 | 638 | else |
64faf04c | 639 | xsprintf(limit_bufs[type], "%s", limits[type] == CGROUP_LIMIT_MAX ? "max" : "0"); |
64faf04c TH |
640 | |
641 | xsprintf(buf, "%u:%u rbps=%s wbps=%s riops=%s wiops=%s\n", major(dev), minor(dev), | |
642 | limit_bufs[CGROUP_IO_RBPS_MAX], limit_bufs[CGROUP_IO_WBPS_MAX], | |
643 | limit_bufs[CGROUP_IO_RIOPS_MAX], limit_bufs[CGROUP_IO_WIOPS_MAX]); | |
f29ff115 | 644 | r = cg_set_attribute("io", u->cgroup_path, "io.max", buf); |
64faf04c | 645 | if (r < 0) |
f29ff115 TH |
646 | log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, |
647 | "Failed to set io.max: %m"); | |
64faf04c TH |
648 | } |
649 | ||
17ae2780 | 650 | static void cgroup_apply_blkio_device_limit(Unit *u, const char *dev_path, uint64_t rbps, uint64_t wbps) { |
64faf04c TH |
651 | char buf[DECIMAL_STR_MAX(dev_t)*2+2+DECIMAL_STR_MAX(uint64_t)+1]; |
652 | dev_t dev; | |
64faf04c TH |
653 | int r; |
654 | ||
655 | r = lookup_block_device(dev_path, &dev); | |
656 | if (r < 0) | |
17ae2780 | 657 | return; |
64faf04c | 658 | |
64faf04c | 659 | sprintf(buf, "%u:%u %" PRIu64 "\n", major(dev), minor(dev), rbps); |
f29ff115 | 660 | r = cg_set_attribute("blkio", u->cgroup_path, "blkio.throttle.read_bps_device", buf); |
64faf04c | 661 | if (r < 0) |
f29ff115 TH |
662 | log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, |
663 | "Failed to set blkio.throttle.read_bps_device: %m"); | |
64faf04c | 664 | |
64faf04c | 665 | sprintf(buf, "%u:%u %" PRIu64 "\n", major(dev), minor(dev), wbps); |
f29ff115 | 666 | r = cg_set_attribute("blkio", u->cgroup_path, "blkio.throttle.write_bps_device", buf); |
64faf04c | 667 | if (r < 0) |
f29ff115 TH |
668 | log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, |
669 | "Failed to set blkio.throttle.write_bps_device: %m"); | |
64faf04c TH |
670 | } |
671 | ||
da4d897e | 672 | static bool cgroup_context_has_unified_memory_config(CGroupContext *c) { |
48422635 | 673 | return c->memory_min > 0 || 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 |
674 | } |
675 | ||
f29ff115 | 676 | static void cgroup_apply_unified_memory_limit(Unit *u, const char *file, uint64_t v) { |
da4d897e TH |
677 | char buf[DECIMAL_STR_MAX(uint64_t) + 1] = "max"; |
678 | int r; | |
679 | ||
680 | if (v != CGROUP_LIMIT_MAX) | |
681 | xsprintf(buf, "%" PRIu64 "\n", v); | |
682 | ||
f29ff115 | 683 | r = cg_set_attribute("memory", u->cgroup_path, file, buf); |
da4d897e | 684 | if (r < 0) |
f29ff115 TH |
685 | log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, |
686 | "Failed to set %s: %m", file); | |
da4d897e TH |
687 | } |
688 | ||
0f2d84d2 | 689 | static void cgroup_apply_firewall(Unit *u) { |
0f2d84d2 LP |
690 | assert(u); |
691 | ||
acf7f253 | 692 | /* Best-effort: let's apply IP firewalling and/or accounting if that's enabled */ |
906c06f6 | 693 | |
acf7f253 | 694 | if (bpf_firewall_compile(u) < 0) |
906c06f6 DM |
695 | return; |
696 | ||
697 | (void) bpf_firewall_install(u); | |
906c06f6 DM |
698 | } |
699 | ||
700 | static void cgroup_context_apply( | |
701 | Unit *u, | |
702 | CGroupMask apply_mask, | |
703 | bool apply_bpf, | |
704 | ManagerState state) { | |
705 | ||
f29ff115 TH |
706 | const char *path; |
707 | CGroupContext *c; | |
01efdf13 | 708 | bool is_root; |
4ad49000 LP |
709 | int r; |
710 | ||
f29ff115 TH |
711 | assert(u); |
712 | ||
906c06f6 DM |
713 | /* Nothing to do? Exit early! */ |
714 | if (apply_mask == 0 && !apply_bpf) | |
4ad49000 | 715 | return; |
8e274523 | 716 | |
f3725e64 LP |
717 | /* Some cgroup attributes are not supported on the root cgroup, hence silently ignore */ |
718 | is_root = unit_has_root_cgroup(u); | |
719 | ||
720 | assert_se(c = unit_get_cgroup_context(u)); | |
721 | assert_se(path = u->cgroup_path); | |
722 | ||
723 | if (is_root) /* Make sure we don't try to display messages with an empty path. */ | |
6da13913 | 724 | path = "/"; |
01efdf13 | 725 | |
714e2e1d LP |
726 | /* We generally ignore errors caused by read-only mounted |
727 | * cgroup trees (assuming we are running in a container then), | |
728 | * and missing cgroups, i.e. EROFS and ENOENT. */ | |
729 | ||
906c06f6 DM |
730 | if ((apply_mask & CGROUP_MASK_CPU) && !is_root) { |
731 | bool has_weight, has_shares; | |
732 | ||
733 | has_weight = cgroup_context_has_cpu_weight(c); | |
734 | has_shares = cgroup_context_has_cpu_shares(c); | |
8e274523 | 735 | |
b4cccbc1 | 736 | if (cg_all_unified() > 0) { |
66ebf6c0 | 737 | uint64_t weight; |
b2f8b02e | 738 | |
66ebf6c0 TH |
739 | if (has_weight) |
740 | weight = cgroup_context_cpu_weight(c, state); | |
741 | else if (has_shares) { | |
742 | uint64_t shares = cgroup_context_cpu_shares(c, state); | |
b2f8b02e | 743 | |
66ebf6c0 TH |
744 | weight = cgroup_cpu_shares_to_weight(shares); |
745 | ||
746 | log_cgroup_compat(u, "Applying [Startup]CpuShares %" PRIu64 " as [Startup]CpuWeight %" PRIu64 " on %s", | |
747 | shares, weight, path); | |
748 | } else | |
749 | weight = CGROUP_WEIGHT_DEFAULT; | |
750 | ||
751 | cgroup_apply_unified_cpu_config(u, weight, c->cpu_quota_per_sec_usec); | |
752 | } else { | |
753 | uint64_t shares; | |
754 | ||
7d862ab8 | 755 | if (has_weight) { |
66ebf6c0 TH |
756 | uint64_t weight = cgroup_context_cpu_weight(c, state); |
757 | ||
758 | shares = cgroup_cpu_weight_to_shares(weight); | |
759 | ||
760 | log_cgroup_compat(u, "Applying [Startup]CpuWeight %" PRIu64 " as [Startup]CpuShares %" PRIu64 " on %s", | |
761 | weight, shares, path); | |
7d862ab8 TH |
762 | } else if (has_shares) |
763 | shares = cgroup_context_cpu_shares(c, state); | |
764 | else | |
66ebf6c0 TH |
765 | shares = CGROUP_CPU_SHARES_DEFAULT; |
766 | ||
767 | cgroup_apply_legacy_cpu_config(u, shares, c->cpu_quota_per_sec_usec); | |
768 | } | |
4ad49000 LP |
769 | } |
770 | ||
906c06f6 | 771 | if (apply_mask & CGROUP_MASK_IO) { |
538b4852 TH |
772 | bool has_io = cgroup_context_has_io_config(c); |
773 | bool has_blockio = cgroup_context_has_blockio_config(c); | |
13c31542 TH |
774 | |
775 | if (!is_root) { | |
64faf04c TH |
776 | char buf[8+DECIMAL_STR_MAX(uint64_t)+1]; |
777 | uint64_t weight; | |
13c31542 | 778 | |
538b4852 TH |
779 | if (has_io) |
780 | weight = cgroup_context_io_weight(c, state); | |
128fadc9 TH |
781 | else if (has_blockio) { |
782 | uint64_t blkio_weight = cgroup_context_blkio_weight(c, state); | |
783 | ||
784 | weight = cgroup_weight_blkio_to_io(blkio_weight); | |
785 | ||
786 | log_cgroup_compat(u, "Applying [Startup]BlockIOWeight %" PRIu64 " as [Startup]IOWeight %" PRIu64, | |
787 | blkio_weight, weight); | |
788 | } else | |
538b4852 | 789 | weight = CGROUP_WEIGHT_DEFAULT; |
13c31542 TH |
790 | |
791 | xsprintf(buf, "default %" PRIu64 "\n", weight); | |
792 | r = cg_set_attribute("io", path, "io.weight", buf); | |
793 | if (r < 0) | |
f29ff115 TH |
794 | log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, |
795 | "Failed to set io.weight: %m"); | |
13c31542 | 796 | |
538b4852 TH |
797 | if (has_io) { |
798 | CGroupIODeviceWeight *w; | |
799 | ||
800 | /* FIXME: no way to reset this list */ | |
801 | LIST_FOREACH(device_weights, w, c->io_device_weights) | |
f29ff115 | 802 | cgroup_apply_io_device_weight(u, w->path, w->weight); |
538b4852 TH |
803 | } else if (has_blockio) { |
804 | CGroupBlockIODeviceWeight *w; | |
805 | ||
806 | /* FIXME: no way to reset this list */ | |
128fadc9 TH |
807 | LIST_FOREACH(device_weights, w, c->blockio_device_weights) { |
808 | weight = cgroup_weight_blkio_to_io(w->weight); | |
809 | ||
810 | log_cgroup_compat(u, "Applying BlockIODeviceWeight %" PRIu64 " as IODeviceWeight %" PRIu64 " for %s", | |
811 | w->weight, weight, w->path); | |
812 | ||
813 | cgroup_apply_io_device_weight(u, w->path, weight); | |
814 | } | |
538b4852 | 815 | } |
13c31542 TH |
816 | } |
817 | ||
64faf04c | 818 | /* Apply limits and free ones without config. */ |
538b4852 | 819 | if (has_io) { |
17ae2780 LP |
820 | CGroupIODeviceLimit *l; |
821 | ||
822 | LIST_FOREACH(device_limits, l, c->io_device_limits) | |
823 | cgroup_apply_io_device_limit(u, l->path, l->limits); | |
538b4852 | 824 | |
538b4852 | 825 | } else if (has_blockio) { |
17ae2780 | 826 | CGroupBlockIODeviceBandwidth *b; |
538b4852 | 827 | |
17ae2780 | 828 | LIST_FOREACH(device_bandwidths, b, c->blockio_device_bandwidths) { |
538b4852 TH |
829 | uint64_t limits[_CGROUP_IO_LIMIT_TYPE_MAX]; |
830 | CGroupIOLimitType type; | |
831 | ||
832 | for (type = 0; type < _CGROUP_IO_LIMIT_TYPE_MAX; type++) | |
833 | limits[type] = cgroup_io_limit_defaults[type]; | |
834 | ||
835 | limits[CGROUP_IO_RBPS_MAX] = b->rbps; | |
836 | limits[CGROUP_IO_WBPS_MAX] = b->wbps; | |
837 | ||
128fadc9 TH |
838 | log_cgroup_compat(u, "Applying BlockIO{Read|Write}Bandwidth %" PRIu64 " %" PRIu64 " as IO{Read|Write}BandwidthMax for %s", |
839 | b->rbps, b->wbps, b->path); | |
840 | ||
17ae2780 | 841 | cgroup_apply_io_device_limit(u, b->path, limits); |
538b4852 | 842 | } |
13c31542 TH |
843 | } |
844 | } | |
845 | ||
906c06f6 | 846 | if (apply_mask & CGROUP_MASK_BLKIO) { |
538b4852 TH |
847 | bool has_io = cgroup_context_has_io_config(c); |
848 | bool has_blockio = cgroup_context_has_blockio_config(c); | |
4ad49000 | 849 | |
01efdf13 | 850 | if (!is_root) { |
64faf04c TH |
851 | char buf[DECIMAL_STR_MAX(uint64_t)+1]; |
852 | uint64_t weight; | |
64faf04c | 853 | |
7d862ab8 | 854 | if (has_io) { |
128fadc9 TH |
855 | uint64_t io_weight = cgroup_context_io_weight(c, state); |
856 | ||
538b4852 | 857 | weight = cgroup_weight_io_to_blkio(cgroup_context_io_weight(c, state)); |
128fadc9 TH |
858 | |
859 | log_cgroup_compat(u, "Applying [Startup]IOWeight %" PRIu64 " as [Startup]BlockIOWeight %" PRIu64, | |
860 | io_weight, weight); | |
7d862ab8 TH |
861 | } else if (has_blockio) |
862 | weight = cgroup_context_blkio_weight(c, state); | |
863 | else | |
538b4852 | 864 | weight = CGROUP_BLKIO_WEIGHT_DEFAULT; |
64faf04c TH |
865 | |
866 | xsprintf(buf, "%" PRIu64 "\n", weight); | |
01efdf13 | 867 | r = cg_set_attribute("blkio", path, "blkio.weight", buf); |
1aeab12b | 868 | if (r < 0) |
f29ff115 TH |
869 | log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, |
870 | "Failed to set blkio.weight: %m"); | |
4ad49000 | 871 | |
7d862ab8 | 872 | if (has_io) { |
538b4852 TH |
873 | CGroupIODeviceWeight *w; |
874 | ||
875 | /* FIXME: no way to reset this list */ | |
128fadc9 TH |
876 | LIST_FOREACH(device_weights, w, c->io_device_weights) { |
877 | weight = cgroup_weight_io_to_blkio(w->weight); | |
878 | ||
879 | log_cgroup_compat(u, "Applying IODeviceWeight %" PRIu64 " as BlockIODeviceWeight %" PRIu64 " for %s", | |
880 | w->weight, weight, w->path); | |
881 | ||
882 | cgroup_apply_blkio_device_weight(u, w->path, weight); | |
883 | } | |
7d862ab8 TH |
884 | } else if (has_blockio) { |
885 | CGroupBlockIODeviceWeight *w; | |
886 | ||
887 | /* FIXME: no way to reset this list */ | |
888 | LIST_FOREACH(device_weights, w, c->blockio_device_weights) | |
889 | cgroup_apply_blkio_device_weight(u, w->path, w->weight); | |
538b4852 | 890 | } |
4ad49000 LP |
891 | } |
892 | ||
64faf04c | 893 | /* Apply limits and free ones without config. */ |
7d862ab8 | 894 | if (has_io) { |
17ae2780 | 895 | CGroupIODeviceLimit *l; |
538b4852 | 896 | |
17ae2780 | 897 | LIST_FOREACH(device_limits, l, c->io_device_limits) { |
128fadc9 TH |
898 | log_cgroup_compat(u, "Applying IO{Read|Write}Bandwidth %" PRIu64 " %" PRIu64 " as BlockIO{Read|Write}BandwidthMax for %s", |
899 | l->limits[CGROUP_IO_RBPS_MAX], l->limits[CGROUP_IO_WBPS_MAX], l->path); | |
900 | ||
17ae2780 | 901 | cgroup_apply_blkio_device_limit(u, l->path, l->limits[CGROUP_IO_RBPS_MAX], l->limits[CGROUP_IO_WBPS_MAX]); |
538b4852 | 902 | } |
7d862ab8 | 903 | } else if (has_blockio) { |
17ae2780 | 904 | CGroupBlockIODeviceBandwidth *b; |
7d862ab8 | 905 | |
17ae2780 LP |
906 | LIST_FOREACH(device_bandwidths, b, c->blockio_device_bandwidths) |
907 | cgroup_apply_blkio_device_limit(u, b->path, b->rbps, b->wbps); | |
d686d8a9 | 908 | } |
8e274523 LP |
909 | } |
910 | ||
906c06f6 | 911 | if ((apply_mask & CGROUP_MASK_MEMORY) && !is_root) { |
b4cccbc1 LP |
912 | if (cg_all_unified() > 0) { |
913 | uint64_t max, swap_max = CGROUP_LIMIT_MAX; | |
efdb0237 | 914 | |
96e131ea | 915 | if (cgroup_context_has_unified_memory_config(c)) { |
da4d897e | 916 | max = c->memory_max; |
96e131ea WC |
917 | swap_max = c->memory_swap_max; |
918 | } else { | |
da4d897e | 919 | max = c->memory_limit; |
efdb0237 | 920 | |
128fadc9 TH |
921 | if (max != CGROUP_LIMIT_MAX) |
922 | log_cgroup_compat(u, "Applying MemoryLimit %" PRIu64 " as MemoryMax", max); | |
923 | } | |
924 | ||
48422635 | 925 | cgroup_apply_unified_memory_limit(u, "memory.min", c->memory_min); |
f29ff115 TH |
926 | cgroup_apply_unified_memory_limit(u, "memory.low", c->memory_low); |
927 | cgroup_apply_unified_memory_limit(u, "memory.high", c->memory_high); | |
928 | cgroup_apply_unified_memory_limit(u, "memory.max", max); | |
96e131ea | 929 | cgroup_apply_unified_memory_limit(u, "memory.swap.max", swap_max); |
efdb0237 | 930 | } else { |
da4d897e | 931 | char buf[DECIMAL_STR_MAX(uint64_t) + 1]; |
7d862ab8 | 932 | uint64_t val; |
da4d897e | 933 | |
7d862ab8 | 934 | if (cgroup_context_has_unified_memory_config(c)) { |
78a4ee59 | 935 | val = c->memory_max; |
7d862ab8 TH |
936 | log_cgroup_compat(u, "Applying MemoryMax %" PRIi64 " as MemoryLimit", val); |
937 | } else | |
938 | val = c->memory_limit; | |
128fadc9 | 939 | |
78a4ee59 DM |
940 | if (val == CGROUP_LIMIT_MAX) |
941 | strncpy(buf, "-1\n", sizeof(buf)); | |
942 | else | |
943 | xsprintf(buf, "%" PRIu64 "\n", val); | |
944 | ||
da4d897e TH |
945 | r = cg_set_attribute("memory", path, "memory.limit_in_bytes", buf); |
946 | if (r < 0) | |
f29ff115 TH |
947 | log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, |
948 | "Failed to set memory.limit_in_bytes: %m"); | |
da4d897e | 949 | } |
4ad49000 | 950 | } |
8e274523 | 951 | |
906c06f6 | 952 | if ((apply_mask & CGROUP_MASK_DEVICES) && !is_root) { |
4ad49000 | 953 | CGroupDeviceAllow *a; |
8e274523 | 954 | |
714e2e1d LP |
955 | /* Changing the devices list of a populated cgroup |
956 | * might result in EINVAL, hence ignore EINVAL | |
957 | * here. */ | |
958 | ||
4ad49000 LP |
959 | if (c->device_allow || c->device_policy != CGROUP_AUTO) |
960 | r = cg_set_attribute("devices", path, "devices.deny", "a"); | |
961 | else | |
962 | r = cg_set_attribute("devices", path, "devices.allow", "a"); | |
1aeab12b | 963 | if (r < 0) |
f29ff115 TH |
964 | log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EINVAL, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, |
965 | "Failed to reset devices.list: %m"); | |
fb385181 | 966 | |
4ad49000 LP |
967 | if (c->device_policy == CGROUP_CLOSED || |
968 | (c->device_policy == CGROUP_AUTO && c->device_allow)) { | |
969 | static const char auto_devices[] = | |
7d711efb LP |
970 | "/dev/null\0" "rwm\0" |
971 | "/dev/zero\0" "rwm\0" | |
972 | "/dev/full\0" "rwm\0" | |
973 | "/dev/random\0" "rwm\0" | |
974 | "/dev/urandom\0" "rwm\0" | |
975 | "/dev/tty\0" "rwm\0" | |
5a7f87a9 | 976 | "/dev/ptmx\0" "rwm\0" |
0d9e7991 | 977 | /* Allow /run/systemd/inaccessible/{chr,blk} devices for mapping InaccessiblePaths */ |
e7330dfe DP |
978 | "-/run/systemd/inaccessible/chr\0" "rwm\0" |
979 | "-/run/systemd/inaccessible/blk\0" "rwm\0"; | |
4ad49000 LP |
980 | |
981 | const char *x, *y; | |
982 | ||
983 | NULSTR_FOREACH_PAIR(x, y, auto_devices) | |
984 | whitelist_device(path, x, y); | |
7d711efb | 985 | |
5a7f87a9 | 986 | /* PTS (/dev/pts) devices may not be duplicated, but accessed */ |
7d711efb | 987 | whitelist_major(path, "pts", 'c', "rw"); |
4ad49000 LP |
988 | } |
989 | ||
990 | LIST_FOREACH(device_allow, a, c->device_allow) { | |
fb4650aa | 991 | char acc[4], *val; |
4ad49000 LP |
992 | unsigned k = 0; |
993 | ||
994 | if (a->r) | |
995 | acc[k++] = 'r'; | |
996 | if (a->w) | |
997 | acc[k++] = 'w'; | |
998 | if (a->m) | |
999 | acc[k++] = 'm'; | |
fb385181 | 1000 | |
4ad49000 LP |
1001 | if (k == 0) |
1002 | continue; | |
fb385181 | 1003 | |
4ad49000 | 1004 | acc[k++] = 0; |
90060676 | 1005 | |
27458ed6 | 1006 | if (path_startswith(a->path, "/dev/")) |
90060676 | 1007 | whitelist_device(path, a->path, acc); |
fb4650aa ZJS |
1008 | else if ((val = startswith(a->path, "block-"))) |
1009 | whitelist_major(path, val, 'b', acc); | |
1010 | else if ((val = startswith(a->path, "char-"))) | |
1011 | whitelist_major(path, val, 'c', acc); | |
90060676 | 1012 | else |
f29ff115 | 1013 | log_unit_debug(u, "Ignoring device %s while writing cgroup attribute.", a->path); |
4ad49000 LP |
1014 | } |
1015 | } | |
03a7b521 | 1016 | |
00b5974f LP |
1017 | if (apply_mask & CGROUP_MASK_PIDS) { |
1018 | ||
1019 | if (is_root) { | |
1020 | /* So, the "pids" controller does not expose anything on the root cgroup, in order not to | |
1021 | * replicate knobs exposed elsewhere needlessly. We abstract this away here however, and when | |
1022 | * the knobs of the root cgroup are modified propagate this to the relevant sysctls. There's a | |
1023 | * non-obvious asymmetry however: unlike the cgroup properties we don't really want to take | |
1024 | * exclusive ownership of the sysctls, but we still want to honour things if the user sets | |
1025 | * limits. Hence we employ sort of a one-way strategy: when the user sets a bounded limit | |
1026 | * through us it counts. When the user afterwards unsets it again (i.e. sets it to unbounded) | |
1027 | * it also counts. But if the user never set a limit through us (i.e. we are the default of | |
1028 | * "unbounded") we leave things unmodified. For this we manage a global boolean that we turn on | |
1029 | * the first time we set a limit. Note that this boolean is flushed out on manager reload, | |
1030 | * which is desirable so that there's an offical way to release control of the sysctl from | |
1031 | * systemd: set the limit to unbounded and reload. */ | |
1032 | ||
1033 | if (c->tasks_max != CGROUP_LIMIT_MAX) { | |
1034 | u->manager->sysctl_pid_max_changed = true; | |
1035 | r = procfs_tasks_set_limit(c->tasks_max); | |
1036 | } else if (u->manager->sysctl_pid_max_changed) | |
1037 | r = procfs_tasks_set_limit(TASKS_MAX); | |
1038 | else | |
1039 | r = 0; | |
03a7b521 | 1040 | |
00b5974f LP |
1041 | if (r < 0) |
1042 | log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, | |
1043 | "Failed to write to tasks limit sysctls: %m"); | |
03a7b521 | 1044 | |
00b5974f LP |
1045 | } else { |
1046 | if (c->tasks_max != CGROUP_LIMIT_MAX) { | |
1047 | char buf[DECIMAL_STR_MAX(uint64_t) + 2]; | |
03a7b521 | 1048 | |
00b5974f LP |
1049 | sprintf(buf, "%" PRIu64 "\n", c->tasks_max); |
1050 | r = cg_set_attribute("pids", path, "pids.max", buf); | |
1051 | } else | |
1052 | r = cg_set_attribute("pids", path, "pids.max", "max"); | |
1053 | if (r < 0) | |
1054 | log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, | |
1055 | "Failed to set pids.max: %m"); | |
1056 | } | |
03a7b521 | 1057 | } |
906c06f6 DM |
1058 | |
1059 | if (apply_bpf) | |
0f2d84d2 | 1060 | cgroup_apply_firewall(u); |
fb385181 LP |
1061 | } |
1062 | ||
efdb0237 LP |
1063 | CGroupMask cgroup_context_get_mask(CGroupContext *c) { |
1064 | CGroupMask mask = 0; | |
8e274523 | 1065 | |
4ad49000 | 1066 | /* Figure out which controllers we need */ |
8e274523 | 1067 | |
b2f8b02e | 1068 | if (c->cpu_accounting || |
66ebf6c0 TH |
1069 | cgroup_context_has_cpu_weight(c) || |
1070 | cgroup_context_has_cpu_shares(c) || | |
3a43da28 | 1071 | c->cpu_quota_per_sec_usec != USEC_INFINITY) |
efdb0237 | 1072 | mask |= CGROUP_MASK_CPUACCT | CGROUP_MASK_CPU; |
ecedd90f | 1073 | |
538b4852 TH |
1074 | if (cgroup_context_has_io_config(c) || cgroup_context_has_blockio_config(c)) |
1075 | mask |= CGROUP_MASK_IO | CGROUP_MASK_BLKIO; | |
ecedd90f | 1076 | |
4ad49000 | 1077 | if (c->memory_accounting || |
da4d897e TH |
1078 | c->memory_limit != CGROUP_LIMIT_MAX || |
1079 | cgroup_context_has_unified_memory_config(c)) | |
efdb0237 | 1080 | mask |= CGROUP_MASK_MEMORY; |
8e274523 | 1081 | |
a931ad47 LP |
1082 | if (c->device_allow || |
1083 | c->device_policy != CGROUP_AUTO) | |
3905f127 | 1084 | mask |= CGROUP_MASK_DEVICES; |
4ad49000 | 1085 | |
03a7b521 | 1086 | if (c->tasks_accounting || |
8793fa25 | 1087 | c->tasks_max != CGROUP_LIMIT_MAX) |
03a7b521 LP |
1088 | mask |= CGROUP_MASK_PIDS; |
1089 | ||
4ad49000 | 1090 | return mask; |
8e274523 LP |
1091 | } |
1092 | ||
efdb0237 | 1093 | CGroupMask unit_get_own_mask(Unit *u) { |
4ad49000 | 1094 | CGroupContext *c; |
8e274523 | 1095 | |
efdb0237 LP |
1096 | /* Returns the mask of controllers the unit needs for itself */ |
1097 | ||
4ad49000 LP |
1098 | c = unit_get_cgroup_context(u); |
1099 | if (!c) | |
1100 | return 0; | |
8e274523 | 1101 | |
64e844e5 | 1102 | return cgroup_context_get_mask(c) | unit_get_delegate_mask(u); |
02638280 LP |
1103 | } |
1104 | ||
1105 | CGroupMask unit_get_delegate_mask(Unit *u) { | |
1106 | CGroupContext *c; | |
1107 | ||
1108 | /* If delegation is turned on, then turn on selected controllers, unless we are on the legacy hierarchy and the | |
1109 | * process we fork into is known to drop privileges, and hence shouldn't get access to the controllers. | |
19af675e | 1110 | * |
02638280 | 1111 | * Note that on the unified hierarchy it is safe to delegate controllers to unprivileged services. */ |
a931ad47 | 1112 | |
1d9cc876 | 1113 | if (!unit_cgroup_delegate(u)) |
02638280 LP |
1114 | return 0; |
1115 | ||
1116 | if (cg_all_unified() <= 0) { | |
a931ad47 LP |
1117 | ExecContext *e; |
1118 | ||
1119 | e = unit_get_exec_context(u); | |
02638280 LP |
1120 | if (e && !exec_context_maintains_privileges(e)) |
1121 | return 0; | |
a931ad47 LP |
1122 | } |
1123 | ||
1d9cc876 | 1124 | assert_se(c = unit_get_cgroup_context(u)); |
02638280 | 1125 | return c->delegate_controllers; |
8e274523 LP |
1126 | } |
1127 | ||
efdb0237 | 1128 | CGroupMask unit_get_members_mask(Unit *u) { |
4ad49000 | 1129 | assert(u); |
bc432dc7 | 1130 | |
02638280 | 1131 | /* Returns the mask of controllers all of the unit's children require, merged */ |
efdb0237 | 1132 | |
bc432dc7 LP |
1133 | if (u->cgroup_members_mask_valid) |
1134 | return u->cgroup_members_mask; | |
1135 | ||
64e844e5 | 1136 | u->cgroup_members_mask = 0; |
bc432dc7 LP |
1137 | |
1138 | if (u->type == UNIT_SLICE) { | |
eef85c4a | 1139 | void *v; |
bc432dc7 LP |
1140 | Unit *member; |
1141 | Iterator i; | |
1142 | ||
eef85c4a | 1143 | HASHMAP_FOREACH_KEY(v, member, u->dependencies[UNIT_BEFORE], i) { |
bc432dc7 LP |
1144 | |
1145 | if (member == u) | |
1146 | continue; | |
1147 | ||
d4fdc205 | 1148 | if (UNIT_DEREF(member->slice) != u) |
bc432dc7 LP |
1149 | continue; |
1150 | ||
31604970 | 1151 | u->cgroup_members_mask |= unit_get_subtree_mask(member); /* note that this calls ourselves again, for the children */ |
bc432dc7 LP |
1152 | } |
1153 | } | |
1154 | ||
1155 | u->cgroup_members_mask_valid = true; | |
6414b7c9 | 1156 | return u->cgroup_members_mask; |
246aa6dd LP |
1157 | } |
1158 | ||
efdb0237 | 1159 | CGroupMask unit_get_siblings_mask(Unit *u) { |
4ad49000 | 1160 | assert(u); |
246aa6dd | 1161 | |
efdb0237 LP |
1162 | /* Returns the mask of controllers all of the unit's siblings |
1163 | * require, i.e. the members mask of the unit's parent slice | |
1164 | * if there is one. */ | |
1165 | ||
bc432dc7 | 1166 | if (UNIT_ISSET(u->slice)) |
637f421e | 1167 | return unit_get_members_mask(UNIT_DEREF(u->slice)); |
4ad49000 | 1168 | |
64e844e5 | 1169 | return unit_get_subtree_mask(u); /* we are the top-level slice */ |
246aa6dd LP |
1170 | } |
1171 | ||
efdb0237 LP |
1172 | CGroupMask unit_get_subtree_mask(Unit *u) { |
1173 | ||
1174 | /* Returns the mask of this subtree, meaning of the group | |
1175 | * itself and its children. */ | |
1176 | ||
1177 | return unit_get_own_mask(u) | unit_get_members_mask(u); | |
1178 | } | |
1179 | ||
1180 | CGroupMask unit_get_target_mask(Unit *u) { | |
1181 | CGroupMask mask; | |
1182 | ||
1183 | /* This returns the cgroup mask of all controllers to enable | |
1184 | * for a specific cgroup, i.e. everything it needs itself, | |
1185 | * plus all that its children need, plus all that its siblings | |
1186 | * need. This is primarily useful on the legacy cgroup | |
1187 | * hierarchy, where we need to duplicate each cgroup in each | |
1188 | * hierarchy that shall be enabled for it. */ | |
6414b7c9 | 1189 | |
efdb0237 LP |
1190 | mask = unit_get_own_mask(u) | unit_get_members_mask(u) | unit_get_siblings_mask(u); |
1191 | mask &= u->manager->cgroup_supported; | |
1192 | ||
1193 | return mask; | |
1194 | } | |
1195 | ||
1196 | CGroupMask unit_get_enable_mask(Unit *u) { | |
1197 | CGroupMask mask; | |
1198 | ||
1199 | /* This returns the cgroup mask of all controllers to enable | |
1200 | * for the children of a specific cgroup. This is primarily | |
1201 | * useful for the unified cgroup hierarchy, where each cgroup | |
1202 | * controls which controllers are enabled for its children. */ | |
1203 | ||
1204 | mask = unit_get_members_mask(u); | |
6414b7c9 DS |
1205 | mask &= u->manager->cgroup_supported; |
1206 | ||
1207 | return mask; | |
1208 | } | |
1209 | ||
906c06f6 DM |
1210 | bool unit_get_needs_bpf(Unit *u) { |
1211 | CGroupContext *c; | |
1212 | Unit *p; | |
1213 | assert(u); | |
1214 | ||
906c06f6 DM |
1215 | c = unit_get_cgroup_context(u); |
1216 | if (!c) | |
1217 | return false; | |
1218 | ||
1219 | if (c->ip_accounting || | |
1220 | c->ip_address_allow || | |
1221 | c->ip_address_deny) | |
1222 | return true; | |
1223 | ||
1224 | /* If any parent slice has an IP access list defined, it applies too */ | |
1225 | for (p = UNIT_DEREF(u->slice); p; p = UNIT_DEREF(p->slice)) { | |
1226 | c = unit_get_cgroup_context(p); | |
1227 | if (!c) | |
1228 | return false; | |
1229 | ||
1230 | if (c->ip_address_allow || | |
1231 | c->ip_address_deny) | |
1232 | return true; | |
1233 | } | |
1234 | ||
1235 | return false; | |
1236 | } | |
1237 | ||
6414b7c9 DS |
1238 | /* Recurse from a unit up through its containing slices, propagating |
1239 | * mask bits upward. A unit is also member of itself. */ | |
bc432dc7 | 1240 | void unit_update_cgroup_members_masks(Unit *u) { |
efdb0237 | 1241 | CGroupMask m; |
bc432dc7 LP |
1242 | bool more; |
1243 | ||
1244 | assert(u); | |
1245 | ||
1246 | /* Calculate subtree mask */ | |
efdb0237 | 1247 | m = unit_get_subtree_mask(u); |
bc432dc7 LP |
1248 | |
1249 | /* See if anything changed from the previous invocation. If | |
1250 | * not, we're done. */ | |
1251 | if (u->cgroup_subtree_mask_valid && m == u->cgroup_subtree_mask) | |
1252 | return; | |
1253 | ||
1254 | more = | |
1255 | u->cgroup_subtree_mask_valid && | |
1256 | ((m & ~u->cgroup_subtree_mask) != 0) && | |
1257 | ((~m & u->cgroup_subtree_mask) == 0); | |
1258 | ||
1259 | u->cgroup_subtree_mask = m; | |
1260 | u->cgroup_subtree_mask_valid = true; | |
1261 | ||
6414b7c9 DS |
1262 | if (UNIT_ISSET(u->slice)) { |
1263 | Unit *s = UNIT_DEREF(u->slice); | |
bc432dc7 LP |
1264 | |
1265 | if (more) | |
1266 | /* There's more set now than before. We | |
1267 | * propagate the new mask to the parent's mask | |
1268 | * (not caring if it actually was valid or | |
1269 | * not). */ | |
1270 | ||
1271 | s->cgroup_members_mask |= m; | |
1272 | ||
1273 | else | |
1274 | /* There's less set now than before (or we | |
1275 | * don't know), we need to recalculate | |
1276 | * everything, so let's invalidate the | |
1277 | * parent's members mask */ | |
1278 | ||
1279 | s->cgroup_members_mask_valid = false; | |
1280 | ||
1281 | /* And now make sure that this change also hits our | |
1282 | * grandparents */ | |
1283 | unit_update_cgroup_members_masks(s); | |
6414b7c9 DS |
1284 | } |
1285 | } | |
1286 | ||
6592b975 | 1287 | const char *unit_get_realized_cgroup_path(Unit *u, CGroupMask mask) { |
03b90d4b | 1288 | |
6592b975 | 1289 | /* Returns the realized cgroup path of the specified unit where all specified controllers are available. */ |
03b90d4b LP |
1290 | |
1291 | while (u) { | |
6592b975 | 1292 | |
03b90d4b LP |
1293 | if (u->cgroup_path && |
1294 | u->cgroup_realized && | |
d94a24ca | 1295 | FLAGS_SET(u->cgroup_realized_mask, mask)) |
03b90d4b LP |
1296 | return u->cgroup_path; |
1297 | ||
1298 | u = UNIT_DEREF(u->slice); | |
1299 | } | |
1300 | ||
1301 | return NULL; | |
1302 | } | |
1303 | ||
6592b975 LP |
1304 | static const char *migrate_callback(CGroupMask mask, void *userdata) { |
1305 | return unit_get_realized_cgroup_path(userdata, mask); | |
1306 | } | |
1307 | ||
efdb0237 LP |
1308 | char *unit_default_cgroup_path(Unit *u) { |
1309 | _cleanup_free_ char *escaped = NULL, *slice = NULL; | |
1310 | int r; | |
1311 | ||
1312 | assert(u); | |
1313 | ||
1314 | if (unit_has_name(u, SPECIAL_ROOT_SLICE)) | |
1315 | return strdup(u->manager->cgroup_root); | |
1316 | ||
1317 | if (UNIT_ISSET(u->slice) && !unit_has_name(UNIT_DEREF(u->slice), SPECIAL_ROOT_SLICE)) { | |
1318 | r = cg_slice_to_path(UNIT_DEREF(u->slice)->id, &slice); | |
1319 | if (r < 0) | |
1320 | return NULL; | |
1321 | } | |
1322 | ||
1323 | escaped = cg_escape(u->id); | |
1324 | if (!escaped) | |
1325 | return NULL; | |
1326 | ||
1327 | if (slice) | |
605405c6 ZJS |
1328 | return strjoin(u->manager->cgroup_root, "/", slice, "/", |
1329 | escaped); | |
efdb0237 | 1330 | else |
605405c6 | 1331 | return strjoin(u->manager->cgroup_root, "/", escaped); |
efdb0237 LP |
1332 | } |
1333 | ||
1334 | int unit_set_cgroup_path(Unit *u, const char *path) { | |
1335 | _cleanup_free_ char *p = NULL; | |
1336 | int r; | |
1337 | ||
1338 | assert(u); | |
1339 | ||
1340 | if (path) { | |
1341 | p = strdup(path); | |
1342 | if (!p) | |
1343 | return -ENOMEM; | |
1344 | } else | |
1345 | p = NULL; | |
1346 | ||
1347 | if (streq_ptr(u->cgroup_path, p)) | |
1348 | return 0; | |
1349 | ||
1350 | if (p) { | |
1351 | r = hashmap_put(u->manager->cgroup_unit, p, u); | |
1352 | if (r < 0) | |
1353 | return r; | |
1354 | } | |
1355 | ||
1356 | unit_release_cgroup(u); | |
1357 | ||
ae2a15bc | 1358 | u->cgroup_path = TAKE_PTR(p); |
efdb0237 LP |
1359 | |
1360 | return 1; | |
1361 | } | |
1362 | ||
1363 | int unit_watch_cgroup(Unit *u) { | |
ab2c3861 | 1364 | _cleanup_free_ char *events = NULL; |
efdb0237 LP |
1365 | int r; |
1366 | ||
1367 | assert(u); | |
1368 | ||
1369 | if (!u->cgroup_path) | |
1370 | return 0; | |
1371 | ||
1372 | if (u->cgroup_inotify_wd >= 0) | |
1373 | return 0; | |
1374 | ||
1375 | /* Only applies to the unified hierarchy */ | |
c22800e4 | 1376 | r = cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER); |
b4cccbc1 LP |
1377 | if (r < 0) |
1378 | return log_error_errno(r, "Failed to determine whether the name=systemd hierarchy is unified: %m"); | |
1379 | if (r == 0) | |
efdb0237 LP |
1380 | return 0; |
1381 | ||
1382 | /* Don't watch the root slice, it's pointless. */ | |
1383 | if (unit_has_name(u, SPECIAL_ROOT_SLICE)) | |
1384 | return 0; | |
1385 | ||
1386 | r = hashmap_ensure_allocated(&u->manager->cgroup_inotify_wd_unit, &trivial_hash_ops); | |
1387 | if (r < 0) | |
1388 | return log_oom(); | |
1389 | ||
ab2c3861 | 1390 | r = cg_get_path(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path, "cgroup.events", &events); |
efdb0237 LP |
1391 | if (r < 0) |
1392 | return log_oom(); | |
1393 | ||
ab2c3861 | 1394 | u->cgroup_inotify_wd = inotify_add_watch(u->manager->cgroup_inotify_fd, events, IN_MODIFY); |
efdb0237 LP |
1395 | if (u->cgroup_inotify_wd < 0) { |
1396 | ||
1397 | if (errno == ENOENT) /* If the directory is already | |
1398 | * gone we don't need to track | |
1399 | * it, so this is not an error */ | |
1400 | return 0; | |
1401 | ||
1402 | return log_unit_error_errno(u, errno, "Failed to add inotify watch descriptor for control group %s: %m", u->cgroup_path); | |
1403 | } | |
1404 | ||
1405 | r = hashmap_put(u->manager->cgroup_inotify_wd_unit, INT_TO_PTR(u->cgroup_inotify_wd), u); | |
1406 | if (r < 0) | |
1407 | return log_unit_error_errno(u, r, "Failed to add inotify watch descriptor to hash map: %m"); | |
1408 | ||
1409 | return 0; | |
1410 | } | |
1411 | ||
a4634b21 LP |
1412 | int unit_pick_cgroup_path(Unit *u) { |
1413 | _cleanup_free_ char *path = NULL; | |
1414 | int r; | |
1415 | ||
1416 | assert(u); | |
1417 | ||
1418 | if (u->cgroup_path) | |
1419 | return 0; | |
1420 | ||
1421 | if (!UNIT_HAS_CGROUP_CONTEXT(u)) | |
1422 | return -EINVAL; | |
1423 | ||
1424 | path = unit_default_cgroup_path(u); | |
1425 | if (!path) | |
1426 | return log_oom(); | |
1427 | ||
1428 | r = unit_set_cgroup_path(u, path); | |
1429 | if (r == -EEXIST) | |
1430 | return log_unit_error_errno(u, r, "Control group %s exists already.", path); | |
1431 | if (r < 0) | |
1432 | return log_unit_error_errno(u, r, "Failed to set unit's control group path to %s: %m", path); | |
1433 | ||
1434 | return 0; | |
1435 | } | |
1436 | ||
efdb0237 LP |
1437 | static int unit_create_cgroup( |
1438 | Unit *u, | |
1439 | CGroupMask target_mask, | |
906c06f6 DM |
1440 | CGroupMask enable_mask, |
1441 | bool needs_bpf) { | |
efdb0237 | 1442 | |
0cd385d3 | 1443 | CGroupContext *c; |
bc432dc7 | 1444 | int r; |
65be7e06 | 1445 | bool created; |
64747e2d | 1446 | |
4ad49000 | 1447 | assert(u); |
64747e2d | 1448 | |
0cd385d3 LP |
1449 | c = unit_get_cgroup_context(u); |
1450 | if (!c) | |
1451 | return 0; | |
1452 | ||
a4634b21 LP |
1453 | /* Figure out our cgroup path */ |
1454 | r = unit_pick_cgroup_path(u); | |
1455 | if (r < 0) | |
1456 | return r; | |
b58b8e11 | 1457 | |
03b90d4b | 1458 | /* First, create our own group */ |
efdb0237 | 1459 | r = cg_create_everywhere(u->manager->cgroup_supported, target_mask, u->cgroup_path); |
23bbb0de | 1460 | if (r < 0) |
efdb0237 | 1461 | return log_unit_error_errno(u, r, "Failed to create cgroup %s: %m", u->cgroup_path); |
65be7e06 | 1462 | created = !!r; |
efdb0237 LP |
1463 | |
1464 | /* Start watching it */ | |
1465 | (void) unit_watch_cgroup(u); | |
1466 | ||
65be7e06 ZJS |
1467 | /* Preserve enabled controllers in delegated units, adjust others. */ |
1468 | if (created || !unit_cgroup_delegate(u)) { | |
1469 | ||
1470 | /* Enable all controllers we need */ | |
1471 | r = cg_enable_everywhere(u->manager->cgroup_supported, enable_mask, u->cgroup_path); | |
1472 | if (r < 0) | |
1473 | log_unit_warning_errno(u, r, "Failed to enable controllers on cgroup %s, ignoring: %m", | |
1474 | u->cgroup_path); | |
1475 | } | |
03b90d4b LP |
1476 | |
1477 | /* Keep track that this is now realized */ | |
4ad49000 | 1478 | u->cgroup_realized = true; |
efdb0237 | 1479 | u->cgroup_realized_mask = target_mask; |
ccf78df1 | 1480 | u->cgroup_enabled_mask = enable_mask; |
906c06f6 | 1481 | u->cgroup_bpf_state = needs_bpf ? UNIT_CGROUP_BPF_ON : UNIT_CGROUP_BPF_OFF; |
4ad49000 | 1482 | |
1d9cc876 | 1483 | if (u->type != UNIT_SLICE && !unit_cgroup_delegate(u)) { |
0cd385d3 LP |
1484 | |
1485 | /* Then, possibly move things over, but not if | |
1486 | * subgroups may contain processes, which is the case | |
1487 | * for slice and delegation units. */ | |
1488 | r = cg_migrate_everywhere(u->manager->cgroup_supported, u->cgroup_path, u->cgroup_path, migrate_callback, u); | |
1489 | if (r < 0) | |
efdb0237 | 1490 | log_unit_warning_errno(u, r, "Failed to migrate cgroup from to %s, ignoring: %m", u->cgroup_path); |
0cd385d3 | 1491 | } |
03b90d4b | 1492 | |
64747e2d LP |
1493 | return 0; |
1494 | } | |
1495 | ||
6592b975 LP |
1496 | static int unit_attach_pid_to_cgroup_via_bus(Unit *u, pid_t pid, const char *suffix_path) { |
1497 | _cleanup_(sd_bus_error_free) sd_bus_error error = SD_BUS_ERROR_NULL; | |
1498 | char *pp; | |
7b3fd631 | 1499 | int r; |
6592b975 | 1500 | |
7b3fd631 LP |
1501 | assert(u); |
1502 | ||
6592b975 LP |
1503 | if (MANAGER_IS_SYSTEM(u->manager)) |
1504 | return -EINVAL; | |
1505 | ||
1506 | if (!u->manager->system_bus) | |
1507 | return -EIO; | |
1508 | ||
1509 | if (!u->cgroup_path) | |
1510 | return -EINVAL; | |
1511 | ||
1512 | /* Determine this unit's cgroup path relative to our cgroup root */ | |
1513 | pp = path_startswith(u->cgroup_path, u->manager->cgroup_root); | |
1514 | if (!pp) | |
1515 | return -EINVAL; | |
1516 | ||
1517 | pp = strjoina("/", pp, suffix_path); | |
858d36c1 | 1518 | path_simplify(pp, false); |
6592b975 LP |
1519 | |
1520 | r = sd_bus_call_method(u->manager->system_bus, | |
1521 | "org.freedesktop.systemd1", | |
1522 | "/org/freedesktop/systemd1", | |
1523 | "org.freedesktop.systemd1.Manager", | |
1524 | "AttachProcessesToUnit", | |
1525 | &error, NULL, | |
1526 | "ssau", | |
1527 | NULL /* empty unit name means client's unit, i.e. us */, pp, 1, (uint32_t) pid); | |
7b3fd631 | 1528 | if (r < 0) |
6592b975 LP |
1529 | return log_unit_debug_errno(u, r, "Failed to attach unit process " PID_FMT " via the bus: %s", pid, bus_error_message(&error, r)); |
1530 | ||
1531 | return 0; | |
1532 | } | |
1533 | ||
1534 | int unit_attach_pids_to_cgroup(Unit *u, Set *pids, const char *suffix_path) { | |
1535 | CGroupMask delegated_mask; | |
1536 | const char *p; | |
1537 | Iterator i; | |
1538 | void *pidp; | |
1539 | int r, q; | |
1540 | ||
1541 | assert(u); | |
1542 | ||
1543 | if (!UNIT_HAS_CGROUP_CONTEXT(u)) | |
1544 | return -EINVAL; | |
1545 | ||
1546 | if (set_isempty(pids)) | |
1547 | return 0; | |
7b3fd631 | 1548 | |
6592b975 | 1549 | r = unit_realize_cgroup(u); |
7b3fd631 LP |
1550 | if (r < 0) |
1551 | return r; | |
1552 | ||
6592b975 LP |
1553 | if (isempty(suffix_path)) |
1554 | p = u->cgroup_path; | |
1555 | else | |
1556 | p = strjoina(u->cgroup_path, "/", suffix_path); | |
1557 | ||
1558 | delegated_mask = unit_get_delegate_mask(u); | |
1559 | ||
1560 | r = 0; | |
1561 | SET_FOREACH(pidp, pids, i) { | |
1562 | pid_t pid = PTR_TO_PID(pidp); | |
1563 | CGroupController c; | |
1564 | ||
1565 | /* First, attach the PID to the main cgroup hierarchy */ | |
1566 | q = cg_attach(SYSTEMD_CGROUP_CONTROLLER, p, pid); | |
1567 | if (q < 0) { | |
1568 | log_unit_debug_errno(u, q, "Couldn't move process " PID_FMT " to requested cgroup '%s': %m", pid, p); | |
1569 | ||
1570 | if (MANAGER_IS_USER(u->manager) && IN_SET(q, -EPERM, -EACCES)) { | |
1571 | int z; | |
1572 | ||
1573 | /* If we are in a user instance, and we can't move the process ourselves due to | |
1574 | * permission problems, let's ask the system instance about it instead. Since it's more | |
1575 | * privileged it might be able to move the process across the leaves of a subtree who's | |
1576 | * top node is not owned by us. */ | |
1577 | ||
1578 | z = unit_attach_pid_to_cgroup_via_bus(u, pid, suffix_path); | |
1579 | if (z < 0) | |
1580 | log_unit_debug_errno(u, z, "Couldn't move process " PID_FMT " to requested cgroup '%s' via the system bus either: %m", pid, p); | |
1581 | else | |
1582 | continue; /* When the bus thing worked via the bus we are fully done for this PID. */ | |
1583 | } | |
1584 | ||
1585 | if (r >= 0) | |
1586 | r = q; /* Remember first error */ | |
1587 | ||
1588 | continue; | |
1589 | } | |
1590 | ||
1591 | q = cg_all_unified(); | |
1592 | if (q < 0) | |
1593 | return q; | |
1594 | if (q > 0) | |
1595 | continue; | |
1596 | ||
1597 | /* In the legacy hierarchy, attach the process to the request cgroup if possible, and if not to the | |
1598 | * innermost realized one */ | |
1599 | ||
1600 | for (c = 0; c < _CGROUP_CONTROLLER_MAX; c++) { | |
1601 | CGroupMask bit = CGROUP_CONTROLLER_TO_MASK(c); | |
1602 | const char *realized; | |
1603 | ||
1604 | if (!(u->manager->cgroup_supported & bit)) | |
1605 | continue; | |
1606 | ||
1607 | /* If this controller is delegated and realized, honour the caller's request for the cgroup suffix. */ | |
1608 | if (delegated_mask & u->cgroup_realized_mask & bit) { | |
1609 | q = cg_attach(cgroup_controller_to_string(c), p, pid); | |
1610 | if (q >= 0) | |
1611 | continue; /* Success! */ | |
1612 | ||
1613 | log_unit_debug_errno(u, q, "Failed to attach PID " PID_FMT " to requested cgroup %s in controller %s, falling back to unit's cgroup: %m", | |
1614 | pid, p, cgroup_controller_to_string(c)); | |
1615 | } | |
1616 | ||
1617 | /* So this controller is either not delegate or realized, or something else weird happened. In | |
1618 | * that case let's attach the PID at least to the closest cgroup up the tree that is | |
1619 | * realized. */ | |
1620 | realized = unit_get_realized_cgroup_path(u, bit); | |
1621 | if (!realized) | |
1622 | continue; /* Not even realized in the root slice? Then let's not bother */ | |
1623 | ||
1624 | q = cg_attach(cgroup_controller_to_string(c), realized, pid); | |
1625 | if (q < 0) | |
1626 | log_unit_debug_errno(u, q, "Failed to attach PID " PID_FMT " to realized cgroup %s in controller %s, ignoring: %m", | |
1627 | pid, realized, cgroup_controller_to_string(c)); | |
1628 | } | |
1629 | } | |
1630 | ||
1631 | return r; | |
7b3fd631 LP |
1632 | } |
1633 | ||
4b58153d LP |
1634 | static void cgroup_xattr_apply(Unit *u) { |
1635 | char ids[SD_ID128_STRING_MAX]; | |
1636 | int r; | |
1637 | ||
1638 | assert(u); | |
1639 | ||
1640 | if (!MANAGER_IS_SYSTEM(u->manager)) | |
1641 | return; | |
1642 | ||
1643 | if (sd_id128_is_null(u->invocation_id)) | |
1644 | return; | |
1645 | ||
1646 | r = cg_set_xattr(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path, | |
1647 | "trusted.invocation_id", | |
1648 | sd_id128_to_string(u->invocation_id, ids), 32, | |
1649 | 0); | |
1650 | if (r < 0) | |
0fb84499 | 1651 | log_unit_debug_errno(u, r, "Failed to set invocation ID on control group %s, ignoring: %m", u->cgroup_path); |
4b58153d LP |
1652 | } |
1653 | ||
906c06f6 DM |
1654 | static bool unit_has_mask_realized( |
1655 | Unit *u, | |
1656 | CGroupMask target_mask, | |
1657 | CGroupMask enable_mask, | |
1658 | bool needs_bpf) { | |
1659 | ||
bc432dc7 LP |
1660 | assert(u); |
1661 | ||
906c06f6 DM |
1662 | return u->cgroup_realized && |
1663 | u->cgroup_realized_mask == target_mask && | |
1664 | u->cgroup_enabled_mask == enable_mask && | |
1665 | ((needs_bpf && u->cgroup_bpf_state == UNIT_CGROUP_BPF_ON) || | |
1666 | (!needs_bpf && u->cgroup_bpf_state == UNIT_CGROUP_BPF_OFF)); | |
6414b7c9 DS |
1667 | } |
1668 | ||
2aa57a65 LP |
1669 | static void unit_add_to_cgroup_realize_queue(Unit *u) { |
1670 | assert(u); | |
1671 | ||
1672 | if (u->in_cgroup_realize_queue) | |
1673 | return; | |
1674 | ||
1675 | LIST_PREPEND(cgroup_realize_queue, u->manager->cgroup_realize_queue, u); | |
1676 | u->in_cgroup_realize_queue = true; | |
1677 | } | |
1678 | ||
1679 | static void unit_remove_from_cgroup_realize_queue(Unit *u) { | |
1680 | assert(u); | |
1681 | ||
1682 | if (!u->in_cgroup_realize_queue) | |
1683 | return; | |
1684 | ||
1685 | LIST_REMOVE(cgroup_realize_queue, u->manager->cgroup_realize_queue, u); | |
1686 | u->in_cgroup_realize_queue = false; | |
1687 | } | |
1688 | ||
6414b7c9 DS |
1689 | /* Check if necessary controllers and attributes for a unit are in place. |
1690 | * | |
1691 | * If so, do nothing. | |
1692 | * If not, create paths, move processes over, and set attributes. | |
1693 | * | |
1694 | * Returns 0 on success and < 0 on failure. */ | |
db785129 | 1695 | static int unit_realize_cgroup_now(Unit *u, ManagerState state) { |
efdb0237 | 1696 | CGroupMask target_mask, enable_mask; |
906c06f6 | 1697 | bool needs_bpf, apply_bpf; |
6414b7c9 | 1698 | int r; |
64747e2d | 1699 | |
4ad49000 | 1700 | assert(u); |
64747e2d | 1701 | |
2aa57a65 | 1702 | unit_remove_from_cgroup_realize_queue(u); |
64747e2d | 1703 | |
efdb0237 | 1704 | target_mask = unit_get_target_mask(u); |
ccf78df1 | 1705 | enable_mask = unit_get_enable_mask(u); |
906c06f6 | 1706 | needs_bpf = unit_get_needs_bpf(u); |
ccf78df1 | 1707 | |
906c06f6 | 1708 | if (unit_has_mask_realized(u, target_mask, enable_mask, needs_bpf)) |
0a1eb06d | 1709 | return 0; |
64747e2d | 1710 | |
906c06f6 DM |
1711 | /* Make sure we apply the BPF filters either when one is configured, or if none is configured but previously |
1712 | * the state was anything but off. This way, if a unit with a BPF filter applied is reconfigured to lose it | |
1713 | * this will trickle down properly to cgroupfs. */ | |
1714 | apply_bpf = needs_bpf || u->cgroup_bpf_state != UNIT_CGROUP_BPF_OFF; | |
1715 | ||
4ad49000 | 1716 | /* First, realize parents */ |
6414b7c9 | 1717 | if (UNIT_ISSET(u->slice)) { |
db785129 | 1718 | r = unit_realize_cgroup_now(UNIT_DEREF(u->slice), state); |
6414b7c9 DS |
1719 | if (r < 0) |
1720 | return r; | |
1721 | } | |
4ad49000 LP |
1722 | |
1723 | /* And then do the real work */ | |
906c06f6 | 1724 | r = unit_create_cgroup(u, target_mask, enable_mask, needs_bpf); |
6414b7c9 DS |
1725 | if (r < 0) |
1726 | return r; | |
1727 | ||
1728 | /* Finally, apply the necessary attributes. */ | |
906c06f6 | 1729 | cgroup_context_apply(u, target_mask, apply_bpf, state); |
4b58153d | 1730 | cgroup_xattr_apply(u); |
6414b7c9 DS |
1731 | |
1732 | return 0; | |
64747e2d LP |
1733 | } |
1734 | ||
91a6073e | 1735 | unsigned manager_dispatch_cgroup_realize_queue(Manager *m) { |
db785129 | 1736 | ManagerState state; |
4ad49000 | 1737 | unsigned n = 0; |
db785129 | 1738 | Unit *i; |
6414b7c9 | 1739 | int r; |
ecedd90f | 1740 | |
91a6073e LP |
1741 | assert(m); |
1742 | ||
db785129 LP |
1743 | state = manager_state(m); |
1744 | ||
91a6073e LP |
1745 | while ((i = m->cgroup_realize_queue)) { |
1746 | assert(i->in_cgroup_realize_queue); | |
ecedd90f | 1747 | |
2aa57a65 LP |
1748 | if (UNIT_IS_INACTIVE_OR_FAILED(unit_active_state(i))) { |
1749 | /* Maybe things changed, and the unit is not actually active anymore? */ | |
1750 | unit_remove_from_cgroup_realize_queue(i); | |
1751 | continue; | |
1752 | } | |
1753 | ||
db785129 | 1754 | r = unit_realize_cgroup_now(i, state); |
6414b7c9 | 1755 | if (r < 0) |
efdb0237 | 1756 | log_warning_errno(r, "Failed to realize cgroups for queued unit %s, ignoring: %m", i->id); |
0a1eb06d | 1757 | |
4ad49000 LP |
1758 | n++; |
1759 | } | |
ecedd90f | 1760 | |
4ad49000 | 1761 | return n; |
8e274523 LP |
1762 | } |
1763 | ||
91a6073e | 1764 | static void unit_add_siblings_to_cgroup_realize_queue(Unit *u) { |
4ad49000 | 1765 | Unit *slice; |
ca949c9d | 1766 | |
4ad49000 LP |
1767 | /* This adds the siblings of the specified unit and the |
1768 | * siblings of all parent units to the cgroup queue. (But | |
1769 | * neither the specified unit itself nor the parents.) */ | |
1770 | ||
1771 | while ((slice = UNIT_DEREF(u->slice))) { | |
1772 | Iterator i; | |
1773 | Unit *m; | |
eef85c4a | 1774 | void *v; |
8f53a7b8 | 1775 | |
eef85c4a | 1776 | HASHMAP_FOREACH_KEY(v, m, u->dependencies[UNIT_BEFORE], i) { |
4ad49000 LP |
1777 | if (m == u) |
1778 | continue; | |
8e274523 | 1779 | |
6414b7c9 DS |
1780 | /* Skip units that have a dependency on the slice |
1781 | * but aren't actually in it. */ | |
4ad49000 | 1782 | if (UNIT_DEREF(m->slice) != slice) |
50159e6a | 1783 | continue; |
8e274523 | 1784 | |
6414b7c9 DS |
1785 | /* No point in doing cgroup application for units |
1786 | * without active processes. */ | |
1787 | if (UNIT_IS_INACTIVE_OR_FAILED(unit_active_state(m))) | |
1788 | continue; | |
1789 | ||
1790 | /* If the unit doesn't need any new controllers | |
1791 | * and has current ones realized, it doesn't need | |
1792 | * any changes. */ | |
906c06f6 DM |
1793 | if (unit_has_mask_realized(m, |
1794 | unit_get_target_mask(m), | |
1795 | unit_get_enable_mask(m), | |
1796 | unit_get_needs_bpf(m))) | |
6414b7c9 DS |
1797 | continue; |
1798 | ||
91a6073e | 1799 | unit_add_to_cgroup_realize_queue(m); |
50159e6a LP |
1800 | } |
1801 | ||
4ad49000 | 1802 | u = slice; |
8e274523 | 1803 | } |
4ad49000 LP |
1804 | } |
1805 | ||
0a1eb06d | 1806 | int unit_realize_cgroup(Unit *u) { |
4ad49000 LP |
1807 | assert(u); |
1808 | ||
35b7ff80 | 1809 | if (!UNIT_HAS_CGROUP_CONTEXT(u)) |
0a1eb06d | 1810 | return 0; |
8e274523 | 1811 | |
4ad49000 LP |
1812 | /* So, here's the deal: when realizing the cgroups for this |
1813 | * unit, we need to first create all parents, but there's more | |
1814 | * actually: for the weight-based controllers we also need to | |
1815 | * make sure that all our siblings (i.e. units that are in the | |
73e231ab | 1816 | * same slice as we are) have cgroups, too. Otherwise, things |
4ad49000 LP |
1817 | * would become very uneven as each of their processes would |
1818 | * get as much resources as all our group together. This call | |
1819 | * will synchronously create the parent cgroups, but will | |
1820 | * defer work on the siblings to the next event loop | |
1821 | * iteration. */ | |
ca949c9d | 1822 | |
4ad49000 | 1823 | /* Add all sibling slices to the cgroup queue. */ |
91a6073e | 1824 | unit_add_siblings_to_cgroup_realize_queue(u); |
4ad49000 | 1825 | |
6414b7c9 | 1826 | /* And realize this one now (and apply the values) */ |
db785129 | 1827 | return unit_realize_cgroup_now(u, manager_state(u->manager)); |
8e274523 LP |
1828 | } |
1829 | ||
efdb0237 LP |
1830 | void unit_release_cgroup(Unit *u) { |
1831 | assert(u); | |
1832 | ||
1833 | /* Forgets all cgroup details for this cgroup */ | |
1834 | ||
1835 | if (u->cgroup_path) { | |
1836 | (void) hashmap_remove(u->manager->cgroup_unit, u->cgroup_path); | |
1837 | u->cgroup_path = mfree(u->cgroup_path); | |
1838 | } | |
1839 | ||
1840 | if (u->cgroup_inotify_wd >= 0) { | |
1841 | if (inotify_rm_watch(u->manager->cgroup_inotify_fd, u->cgroup_inotify_wd) < 0) | |
1842 | log_unit_debug_errno(u, errno, "Failed to remove cgroup inotify watch %i for %s, ignoring", u->cgroup_inotify_wd, u->id); | |
1843 | ||
1844 | (void) hashmap_remove(u->manager->cgroup_inotify_wd_unit, INT_TO_PTR(u->cgroup_inotify_wd)); | |
1845 | u->cgroup_inotify_wd = -1; | |
1846 | } | |
1847 | } | |
1848 | ||
1849 | void unit_prune_cgroup(Unit *u) { | |
8e274523 | 1850 | int r; |
efdb0237 | 1851 | bool is_root_slice; |
8e274523 | 1852 | |
4ad49000 | 1853 | assert(u); |
8e274523 | 1854 | |
efdb0237 LP |
1855 | /* Removes the cgroup, if empty and possible, and stops watching it. */ |
1856 | ||
4ad49000 LP |
1857 | if (!u->cgroup_path) |
1858 | return; | |
8e274523 | 1859 | |
fe700f46 LP |
1860 | (void) unit_get_cpu_usage(u, NULL); /* Cache the last CPU usage value before we destroy the cgroup */ |
1861 | ||
efdb0237 LP |
1862 | is_root_slice = unit_has_name(u, SPECIAL_ROOT_SLICE); |
1863 | ||
1864 | r = cg_trim_everywhere(u->manager->cgroup_supported, u->cgroup_path, !is_root_slice); | |
dab5bf85 | 1865 | if (r < 0) { |
f29ff115 | 1866 | log_unit_debug_errno(u, r, "Failed to destroy cgroup %s, ignoring: %m", u->cgroup_path); |
dab5bf85 RL |
1867 | return; |
1868 | } | |
8e274523 | 1869 | |
efdb0237 LP |
1870 | if (is_root_slice) |
1871 | return; | |
1872 | ||
1873 | unit_release_cgroup(u); | |
0a1eb06d | 1874 | |
4ad49000 | 1875 | u->cgroup_realized = false; |
bc432dc7 | 1876 | u->cgroup_realized_mask = 0; |
ccf78df1 | 1877 | u->cgroup_enabled_mask = 0; |
8e274523 LP |
1878 | } |
1879 | ||
efdb0237 | 1880 | int unit_search_main_pid(Unit *u, pid_t *ret) { |
4ad49000 LP |
1881 | _cleanup_fclose_ FILE *f = NULL; |
1882 | pid_t pid = 0, npid, mypid; | |
efdb0237 | 1883 | int r; |
4ad49000 LP |
1884 | |
1885 | assert(u); | |
efdb0237 | 1886 | assert(ret); |
4ad49000 LP |
1887 | |
1888 | if (!u->cgroup_path) | |
efdb0237 | 1889 | return -ENXIO; |
4ad49000 | 1890 | |
efdb0237 LP |
1891 | r = cg_enumerate_processes(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path, &f); |
1892 | if (r < 0) | |
1893 | return r; | |
4ad49000 | 1894 | |
df0ff127 | 1895 | mypid = getpid_cached(); |
4ad49000 LP |
1896 | while (cg_read_pid(f, &npid) > 0) { |
1897 | pid_t ppid; | |
1898 | ||
1899 | if (npid == pid) | |
1900 | continue; | |
8e274523 | 1901 | |
4ad49000 | 1902 | /* Ignore processes that aren't our kids */ |
6bc73acb | 1903 | if (get_process_ppid(npid, &ppid) >= 0 && ppid != mypid) |
4ad49000 | 1904 | continue; |
8e274523 | 1905 | |
efdb0237 | 1906 | if (pid != 0) |
4ad49000 LP |
1907 | /* Dang, there's more than one daemonized PID |
1908 | in this group, so we don't know what process | |
1909 | is the main process. */ | |
efdb0237 LP |
1910 | |
1911 | return -ENODATA; | |
8e274523 | 1912 | |
4ad49000 | 1913 | pid = npid; |
8e274523 LP |
1914 | } |
1915 | ||
efdb0237 LP |
1916 | *ret = pid; |
1917 | return 0; | |
1918 | } | |
1919 | ||
1920 | static int unit_watch_pids_in_path(Unit *u, const char *path) { | |
b3c5bad3 | 1921 | _cleanup_closedir_ DIR *d = NULL; |
efdb0237 LP |
1922 | _cleanup_fclose_ FILE *f = NULL; |
1923 | int ret = 0, r; | |
1924 | ||
1925 | assert(u); | |
1926 | assert(path); | |
1927 | ||
1928 | r = cg_enumerate_processes(SYSTEMD_CGROUP_CONTROLLER, path, &f); | |
1929 | if (r < 0) | |
1930 | ret = r; | |
1931 | else { | |
1932 | pid_t pid; | |
1933 | ||
1934 | while ((r = cg_read_pid(f, &pid)) > 0) { | |
1935 | r = unit_watch_pid(u, pid); | |
1936 | if (r < 0 && ret >= 0) | |
1937 | ret = r; | |
1938 | } | |
1939 | ||
1940 | if (r < 0 && ret >= 0) | |
1941 | ret = r; | |
1942 | } | |
1943 | ||
1944 | r = cg_enumerate_subgroups(SYSTEMD_CGROUP_CONTROLLER, path, &d); | |
1945 | if (r < 0) { | |
1946 | if (ret >= 0) | |
1947 | ret = r; | |
1948 | } else { | |
1949 | char *fn; | |
1950 | ||
1951 | while ((r = cg_read_subgroup(d, &fn)) > 0) { | |
1952 | _cleanup_free_ char *p = NULL; | |
1953 | ||
605405c6 | 1954 | p = strjoin(path, "/", fn); |
efdb0237 LP |
1955 | free(fn); |
1956 | ||
1957 | if (!p) | |
1958 | return -ENOMEM; | |
1959 | ||
1960 | r = unit_watch_pids_in_path(u, p); | |
1961 | if (r < 0 && ret >= 0) | |
1962 | ret = r; | |
1963 | } | |
1964 | ||
1965 | if (r < 0 && ret >= 0) | |
1966 | ret = r; | |
1967 | } | |
1968 | ||
1969 | return ret; | |
1970 | } | |
1971 | ||
11aef522 LP |
1972 | int unit_synthesize_cgroup_empty_event(Unit *u) { |
1973 | int r; | |
1974 | ||
1975 | assert(u); | |
1976 | ||
1977 | /* Enqueue a synthetic cgroup empty event if this unit doesn't watch any PIDs anymore. This is compatibility | |
1978 | * support for non-unified systems where notifications aren't reliable, and hence need to take whatever we can | |
1979 | * get as notification source as soon as we stopped having any useful PIDs to watch for. */ | |
1980 | ||
1981 | if (!u->cgroup_path) | |
1982 | return -ENOENT; | |
1983 | ||
1984 | r = cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER); | |
1985 | if (r < 0) | |
1986 | return r; | |
1987 | if (r > 0) /* On unified we have reliable notifications, and don't need this */ | |
1988 | return 0; | |
1989 | ||
1990 | if (!set_isempty(u->pids)) | |
1991 | return 0; | |
1992 | ||
1993 | unit_add_to_cgroup_empty_queue(u); | |
1994 | return 0; | |
1995 | } | |
1996 | ||
efdb0237 | 1997 | int unit_watch_all_pids(Unit *u) { |
b4cccbc1 LP |
1998 | int r; |
1999 | ||
efdb0237 LP |
2000 | assert(u); |
2001 | ||
2002 | /* Adds all PIDs from our cgroup to the set of PIDs we | |
2003 | * watch. This is a fallback logic for cases where we do not | |
2004 | * get reliable cgroup empty notifications: we try to use | |
2005 | * SIGCHLD as replacement. */ | |
2006 | ||
2007 | if (!u->cgroup_path) | |
2008 | return -ENOENT; | |
2009 | ||
c22800e4 | 2010 | r = cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER); |
b4cccbc1 LP |
2011 | if (r < 0) |
2012 | return r; | |
2013 | if (r > 0) /* On unified we can use proper notifications */ | |
efdb0237 LP |
2014 | return 0; |
2015 | ||
2016 | return unit_watch_pids_in_path(u, u->cgroup_path); | |
2017 | } | |
2018 | ||
09e24654 LP |
2019 | static int on_cgroup_empty_event(sd_event_source *s, void *userdata) { |
2020 | Manager *m = userdata; | |
2021 | Unit *u; | |
efdb0237 LP |
2022 | int r; |
2023 | ||
09e24654 LP |
2024 | assert(s); |
2025 | assert(m); | |
efdb0237 | 2026 | |
09e24654 LP |
2027 | u = m->cgroup_empty_queue; |
2028 | if (!u) | |
efdb0237 LP |
2029 | return 0; |
2030 | ||
09e24654 LP |
2031 | assert(u->in_cgroup_empty_queue); |
2032 | u->in_cgroup_empty_queue = false; | |
2033 | LIST_REMOVE(cgroup_empty_queue, m->cgroup_empty_queue, u); | |
2034 | ||
2035 | if (m->cgroup_empty_queue) { | |
2036 | /* More stuff queued, let's make sure we remain enabled */ | |
2037 | r = sd_event_source_set_enabled(s, SD_EVENT_ONESHOT); | |
2038 | if (r < 0) | |
19a691a9 | 2039 | log_debug_errno(r, "Failed to reenable cgroup empty event source, ignoring: %m"); |
09e24654 | 2040 | } |
efdb0237 LP |
2041 | |
2042 | unit_add_to_gc_queue(u); | |
2043 | ||
2044 | if (UNIT_VTABLE(u)->notify_cgroup_empty) | |
2045 | UNIT_VTABLE(u)->notify_cgroup_empty(u); | |
2046 | ||
2047 | return 0; | |
2048 | } | |
2049 | ||
09e24654 LP |
2050 | void unit_add_to_cgroup_empty_queue(Unit *u) { |
2051 | int r; | |
2052 | ||
2053 | assert(u); | |
2054 | ||
2055 | /* Note that there are four different ways how cgroup empty events reach us: | |
2056 | * | |
2057 | * 1. On the unified hierarchy we get an inotify event on the cgroup | |
2058 | * | |
2059 | * 2. On the legacy hierarchy, when running in system mode, we get a datagram on the cgroup agent socket | |
2060 | * | |
2061 | * 3. On the legacy hierarchy, when running in user mode, we get a D-Bus signal on the system bus | |
2062 | * | |
2063 | * 4. On the legacy hierarchy, in service units we start watching all processes of the cgroup for SIGCHLD as | |
2064 | * soon as we get one SIGCHLD, to deal with unreliable cgroup notifications. | |
2065 | * | |
2066 | * Regardless which way we got the notification, we'll verify it here, and then add it to a separate | |
2067 | * queue. This queue will be dispatched at a lower priority than the SIGCHLD handler, so that we always use | |
2068 | * SIGCHLD if we can get it first, and only use the cgroup empty notifications if there's no SIGCHLD pending | |
2069 | * (which might happen if the cgroup doesn't contain processes that are our own child, which is typically the | |
2070 | * case for scope units). */ | |
2071 | ||
2072 | if (u->in_cgroup_empty_queue) | |
2073 | return; | |
2074 | ||
2075 | /* Let's verify that the cgroup is really empty */ | |
2076 | if (!u->cgroup_path) | |
2077 | return; | |
2078 | r = cg_is_empty_recursive(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path); | |
2079 | if (r < 0) { | |
2080 | log_unit_debug_errno(u, r, "Failed to determine whether cgroup %s is empty: %m", u->cgroup_path); | |
2081 | return; | |
2082 | } | |
2083 | if (r == 0) | |
2084 | return; | |
2085 | ||
2086 | LIST_PREPEND(cgroup_empty_queue, u->manager->cgroup_empty_queue, u); | |
2087 | u->in_cgroup_empty_queue = true; | |
2088 | ||
2089 | /* Trigger the defer event */ | |
2090 | r = sd_event_source_set_enabled(u->manager->cgroup_empty_event_source, SD_EVENT_ONESHOT); | |
2091 | if (r < 0) | |
2092 | log_debug_errno(r, "Failed to enable cgroup empty event source: %m"); | |
2093 | } | |
2094 | ||
efdb0237 LP |
2095 | static int on_cgroup_inotify_event(sd_event_source *s, int fd, uint32_t revents, void *userdata) { |
2096 | Manager *m = userdata; | |
2097 | ||
2098 | assert(s); | |
2099 | assert(fd >= 0); | |
2100 | assert(m); | |
2101 | ||
2102 | for (;;) { | |
2103 | union inotify_event_buffer buffer; | |
2104 | struct inotify_event *e; | |
2105 | ssize_t l; | |
2106 | ||
2107 | l = read(fd, &buffer, sizeof(buffer)); | |
2108 | if (l < 0) { | |
47249640 | 2109 | if (IN_SET(errno, EINTR, EAGAIN)) |
efdb0237 LP |
2110 | return 0; |
2111 | ||
2112 | return log_error_errno(errno, "Failed to read control group inotify events: %m"); | |
2113 | } | |
2114 | ||
2115 | FOREACH_INOTIFY_EVENT(e, buffer, l) { | |
2116 | Unit *u; | |
2117 | ||
2118 | if (e->wd < 0) | |
2119 | /* Queue overflow has no watch descriptor */ | |
2120 | continue; | |
2121 | ||
2122 | if (e->mask & IN_IGNORED) | |
2123 | /* The watch was just removed */ | |
2124 | continue; | |
2125 | ||
2126 | u = hashmap_get(m->cgroup_inotify_wd_unit, INT_TO_PTR(e->wd)); | |
2127 | if (!u) /* Not that inotify might deliver | |
2128 | * events for a watch even after it | |
2129 | * was removed, because it was queued | |
2130 | * before the removal. Let's ignore | |
2131 | * this here safely. */ | |
2132 | continue; | |
2133 | ||
09e24654 | 2134 | unit_add_to_cgroup_empty_queue(u); |
efdb0237 LP |
2135 | } |
2136 | } | |
8e274523 LP |
2137 | } |
2138 | ||
8e274523 | 2139 | int manager_setup_cgroup(Manager *m) { |
9444b1f2 | 2140 | _cleanup_free_ char *path = NULL; |
10bd3e2e | 2141 | const char *scope_path; |
efdb0237 | 2142 | CGroupController c; |
b4cccbc1 | 2143 | int r, all_unified; |
efdb0237 | 2144 | char *e; |
8e274523 LP |
2145 | |
2146 | assert(m); | |
2147 | ||
35d2e7ec | 2148 | /* 1. Determine hierarchy */ |
efdb0237 | 2149 | m->cgroup_root = mfree(m->cgroup_root); |
9444b1f2 | 2150 | r = cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER, 0, &m->cgroup_root); |
23bbb0de MS |
2151 | if (r < 0) |
2152 | return log_error_errno(r, "Cannot determine cgroup we are running in: %m"); | |
8e274523 | 2153 | |
efdb0237 LP |
2154 | /* Chop off the init scope, if we are already located in it */ |
2155 | e = endswith(m->cgroup_root, "/" SPECIAL_INIT_SCOPE); | |
0d8c31ff | 2156 | |
efdb0237 LP |
2157 | /* LEGACY: Also chop off the system slice if we are in |
2158 | * it. This is to support live upgrades from older systemd | |
2159 | * versions where PID 1 was moved there. Also see | |
2160 | * cg_get_root_path(). */ | |
463d0d15 | 2161 | if (!e && MANAGER_IS_SYSTEM(m)) { |
9444b1f2 | 2162 | e = endswith(m->cgroup_root, "/" SPECIAL_SYSTEM_SLICE); |
15c60e99 | 2163 | if (!e) |
efdb0237 | 2164 | e = endswith(m->cgroup_root, "/system"); /* even more legacy */ |
0baf24dd | 2165 | } |
efdb0237 LP |
2166 | if (e) |
2167 | *e = 0; | |
7ccfb64a | 2168 | |
7546145e LP |
2169 | /* And make sure to store away the root value without trailing slash, even for the root dir, so that we can |
2170 | * easily prepend it everywhere. */ | |
2171 | delete_trailing_chars(m->cgroup_root, "/"); | |
8e274523 | 2172 | |
35d2e7ec | 2173 | /* 2. Show data */ |
9444b1f2 | 2174 | r = cg_get_path(SYSTEMD_CGROUP_CONTROLLER, m->cgroup_root, NULL, &path); |
23bbb0de MS |
2175 | if (r < 0) |
2176 | return log_error_errno(r, "Cannot find cgroup mount point: %m"); | |
8e274523 | 2177 | |
415fc41c TH |
2178 | r = cg_unified_flush(); |
2179 | if (r < 0) | |
2180 | return log_error_errno(r, "Couldn't determine if we are running in the unified hierarchy: %m"); | |
5da38d07 | 2181 | |
b4cccbc1 | 2182 | all_unified = cg_all_unified(); |
d4c819ed ZJS |
2183 | if (all_unified < 0) |
2184 | return log_error_errno(all_unified, "Couldn't determine whether we are in all unified mode: %m"); | |
2185 | if (all_unified > 0) | |
efdb0237 | 2186 | log_debug("Unified cgroup hierarchy is located at %s.", path); |
b4cccbc1 | 2187 | else { |
c22800e4 | 2188 | r = cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER); |
b4cccbc1 LP |
2189 | if (r < 0) |
2190 | return log_error_errno(r, "Failed to determine whether systemd's own controller is in unified mode: %m"); | |
2191 | if (r > 0) | |
2192 | log_debug("Unified cgroup hierarchy is located at %s. Controllers are on legacy hierarchies.", path); | |
2193 | else | |
2194 | log_debug("Using cgroup controller " SYSTEMD_CGROUP_CONTROLLER_LEGACY ". File system hierarchy is at %s.", path); | |
2195 | } | |
efdb0237 | 2196 | |
09e24654 LP |
2197 | /* 3. Allocate cgroup empty defer event source */ |
2198 | m->cgroup_empty_event_source = sd_event_source_unref(m->cgroup_empty_event_source); | |
2199 | r = sd_event_add_defer(m->event, &m->cgroup_empty_event_source, on_cgroup_empty_event, m); | |
2200 | if (r < 0) | |
2201 | return log_error_errno(r, "Failed to create cgroup empty event source: %m"); | |
2202 | ||
2203 | r = sd_event_source_set_priority(m->cgroup_empty_event_source, SD_EVENT_PRIORITY_NORMAL-5); | |
2204 | if (r < 0) | |
2205 | return log_error_errno(r, "Failed to set priority of cgroup empty event source: %m"); | |
2206 | ||
2207 | r = sd_event_source_set_enabled(m->cgroup_empty_event_source, SD_EVENT_OFF); | |
2208 | if (r < 0) | |
2209 | return log_error_errno(r, "Failed to disable cgroup empty event source: %m"); | |
2210 | ||
2211 | (void) sd_event_source_set_description(m->cgroup_empty_event_source, "cgroup-empty"); | |
2212 | ||
2213 | /* 4. Install notifier inotify object, or agent */ | |
10bd3e2e | 2214 | if (cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER) > 0) { |
c6c18be3 | 2215 | |
09e24654 | 2216 | /* In the unified hierarchy we can get cgroup empty notifications via inotify. */ |
efdb0237 | 2217 | |
10bd3e2e LP |
2218 | m->cgroup_inotify_event_source = sd_event_source_unref(m->cgroup_inotify_event_source); |
2219 | safe_close(m->cgroup_inotify_fd); | |
efdb0237 | 2220 | |
10bd3e2e LP |
2221 | m->cgroup_inotify_fd = inotify_init1(IN_NONBLOCK|IN_CLOEXEC); |
2222 | if (m->cgroup_inotify_fd < 0) | |
2223 | return log_error_errno(errno, "Failed to create control group inotify object: %m"); | |
efdb0237 | 2224 | |
10bd3e2e LP |
2225 | r = sd_event_add_io(m->event, &m->cgroup_inotify_event_source, m->cgroup_inotify_fd, EPOLLIN, on_cgroup_inotify_event, m); |
2226 | if (r < 0) | |
2227 | return log_error_errno(r, "Failed to watch control group inotify object: %m"); | |
efdb0237 | 2228 | |
10bd3e2e LP |
2229 | /* Process cgroup empty notifications early, but after service notifications and SIGCHLD. Also |
2230 | * see handling of cgroup agent notifications, for the classic cgroup hierarchy support. */ | |
09e24654 | 2231 | r = sd_event_source_set_priority(m->cgroup_inotify_event_source, SD_EVENT_PRIORITY_NORMAL-4); |
10bd3e2e LP |
2232 | if (r < 0) |
2233 | return log_error_errno(r, "Failed to set priority of inotify event source: %m"); | |
efdb0237 | 2234 | |
10bd3e2e | 2235 | (void) sd_event_source_set_description(m->cgroup_inotify_event_source, "cgroup-inotify"); |
efdb0237 | 2236 | |
10bd3e2e | 2237 | } else if (MANAGER_IS_SYSTEM(m) && m->test_run_flags == 0) { |
efdb0237 | 2238 | |
10bd3e2e LP |
2239 | /* On the legacy hierarchy we only get notifications via cgroup agents. (Which isn't really reliable, |
2240 | * since it does not generate events when control groups with children run empty. */ | |
8e274523 | 2241 | |
10bd3e2e | 2242 | r = cg_install_release_agent(SYSTEMD_CGROUP_CONTROLLER, SYSTEMD_CGROUP_AGENT_PATH); |
23bbb0de | 2243 | if (r < 0) |
10bd3e2e LP |
2244 | log_warning_errno(r, "Failed to install release agent, ignoring: %m"); |
2245 | else if (r > 0) | |
2246 | log_debug("Installed release agent."); | |
2247 | else if (r == 0) | |
2248 | log_debug("Release agent already installed."); | |
2249 | } | |
efdb0237 | 2250 | |
09e24654 | 2251 | /* 5. Make sure we are in the special "init.scope" unit in the root slice. */ |
10bd3e2e LP |
2252 | scope_path = strjoina(m->cgroup_root, "/" SPECIAL_INIT_SCOPE); |
2253 | r = cg_create_and_attach(SYSTEMD_CGROUP_CONTROLLER, scope_path, 0); | |
aa77e234 MS |
2254 | if (r >= 0) { |
2255 | /* Also, move all other userspace processes remaining in the root cgroup into that scope. */ | |
2256 | r = cg_migrate(SYSTEMD_CGROUP_CONTROLLER, m->cgroup_root, SYSTEMD_CGROUP_CONTROLLER, scope_path, 0); | |
2257 | if (r < 0) | |
2258 | log_warning_errno(r, "Couldn't move remaining userspace processes, ignoring: %m"); | |
c6c18be3 | 2259 | |
aa77e234 MS |
2260 | /* 6. And pin it, so that it cannot be unmounted */ |
2261 | safe_close(m->pin_cgroupfs_fd); | |
2262 | m->pin_cgroupfs_fd = open(path, O_RDONLY|O_CLOEXEC|O_DIRECTORY|O_NOCTTY|O_NONBLOCK); | |
2263 | if (m->pin_cgroupfs_fd < 0) | |
2264 | return log_error_errno(errno, "Failed to open pin file: %m"); | |
0d8c31ff | 2265 | |
b4dec49f | 2266 | } else if (!m->test_run_flags) |
aa77e234 | 2267 | return log_error_errno(r, "Failed to create %s control group: %m", scope_path); |
10bd3e2e | 2268 | |
09e24654 | 2269 | /* 7. Always enable hierarchical support if it exists... */ |
10bd3e2e LP |
2270 | if (!all_unified && m->test_run_flags == 0) |
2271 | (void) cg_set_attribute("memory", "/", "memory.use_hierarchy", "1"); | |
c6c18be3 | 2272 | |
09e24654 | 2273 | /* 8. Figure out which controllers are supported, and log about it */ |
efdb0237 LP |
2274 | r = cg_mask_supported(&m->cgroup_supported); |
2275 | if (r < 0) | |
2276 | return log_error_errno(r, "Failed to determine supported controllers: %m"); | |
efdb0237 | 2277 | for (c = 0; c < _CGROUP_CONTROLLER_MAX; c++) |
eee0a1e4 | 2278 | log_debug("Controller '%s' supported: %s", cgroup_controller_to_string(c), yes_no(m->cgroup_supported & CGROUP_CONTROLLER_TO_MASK(c))); |
9156e799 | 2279 | |
a32360f1 | 2280 | return 0; |
8e274523 LP |
2281 | } |
2282 | ||
c6c18be3 | 2283 | void manager_shutdown_cgroup(Manager *m, bool delete) { |
8e274523 LP |
2284 | assert(m); |
2285 | ||
9444b1f2 LP |
2286 | /* We can't really delete the group, since we are in it. But |
2287 | * let's trim it. */ | |
f6c63f6f | 2288 | if (delete && m->cgroup_root && m->test_run_flags != MANAGER_TEST_RUN_MINIMAL) |
efdb0237 LP |
2289 | (void) cg_trim(SYSTEMD_CGROUP_CONTROLLER, m->cgroup_root, false); |
2290 | ||
09e24654 LP |
2291 | m->cgroup_empty_event_source = sd_event_source_unref(m->cgroup_empty_event_source); |
2292 | ||
efdb0237 LP |
2293 | m->cgroup_inotify_wd_unit = hashmap_free(m->cgroup_inotify_wd_unit); |
2294 | ||
2295 | m->cgroup_inotify_event_source = sd_event_source_unref(m->cgroup_inotify_event_source); | |
2296 | m->cgroup_inotify_fd = safe_close(m->cgroup_inotify_fd); | |
8e274523 | 2297 | |
03e334a1 | 2298 | m->pin_cgroupfs_fd = safe_close(m->pin_cgroupfs_fd); |
c6c18be3 | 2299 | |
efdb0237 | 2300 | m->cgroup_root = mfree(m->cgroup_root); |
8e274523 LP |
2301 | } |
2302 | ||
4ad49000 | 2303 | Unit* manager_get_unit_by_cgroup(Manager *m, const char *cgroup) { |
acb14d31 | 2304 | char *p; |
4ad49000 | 2305 | Unit *u; |
acb14d31 LP |
2306 | |
2307 | assert(m); | |
2308 | assert(cgroup); | |
acb14d31 | 2309 | |
4ad49000 LP |
2310 | u = hashmap_get(m->cgroup_unit, cgroup); |
2311 | if (u) | |
2312 | return u; | |
acb14d31 | 2313 | |
8e70580b | 2314 | p = strdupa(cgroup); |
acb14d31 LP |
2315 | for (;;) { |
2316 | char *e; | |
2317 | ||
2318 | e = strrchr(p, '/'); | |
efdb0237 LP |
2319 | if (!e || e == p) |
2320 | return hashmap_get(m->cgroup_unit, SPECIAL_ROOT_SLICE); | |
acb14d31 LP |
2321 | |
2322 | *e = 0; | |
2323 | ||
4ad49000 LP |
2324 | u = hashmap_get(m->cgroup_unit, p); |
2325 | if (u) | |
2326 | return u; | |
acb14d31 LP |
2327 | } |
2328 | } | |
2329 | ||
b3ac818b | 2330 | Unit *manager_get_unit_by_pid_cgroup(Manager *m, pid_t pid) { |
4ad49000 | 2331 | _cleanup_free_ char *cgroup = NULL; |
8e274523 | 2332 | |
8c47c732 LP |
2333 | assert(m); |
2334 | ||
62a76913 | 2335 | if (!pid_is_valid(pid)) |
b3ac818b LP |
2336 | return NULL; |
2337 | ||
62a76913 | 2338 | if (cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER, pid, &cgroup) < 0) |
b3ac818b LP |
2339 | return NULL; |
2340 | ||
2341 | return manager_get_unit_by_cgroup(m, cgroup); | |
2342 | } | |
2343 | ||
2344 | Unit *manager_get_unit_by_pid(Manager *m, pid_t pid) { | |
62a76913 | 2345 | Unit *u, **array; |
b3ac818b LP |
2346 | |
2347 | assert(m); | |
2348 | ||
62a76913 LP |
2349 | /* Note that a process might be owned by multiple units, we return only one here, which is good enough for most |
2350 | * cases, though not strictly correct. We prefer the one reported by cgroup membership, as that's the most | |
2351 | * relevant one as children of the process will be assigned to that one, too, before all else. */ | |
2352 | ||
2353 | if (!pid_is_valid(pid)) | |
8c47c732 LP |
2354 | return NULL; |
2355 | ||
2ca9d979 | 2356 | if (pid == getpid_cached()) |
efdb0237 LP |
2357 | return hashmap_get(m->units, SPECIAL_INIT_SCOPE); |
2358 | ||
62a76913 | 2359 | u = manager_get_unit_by_pid_cgroup(m, pid); |
5fe8876b LP |
2360 | if (u) |
2361 | return u; | |
2362 | ||
62a76913 | 2363 | u = hashmap_get(m->watch_pids, PID_TO_PTR(pid)); |
5fe8876b LP |
2364 | if (u) |
2365 | return u; | |
2366 | ||
62a76913 LP |
2367 | array = hashmap_get(m->watch_pids, PID_TO_PTR(-pid)); |
2368 | if (array) | |
2369 | return array[0]; | |
2370 | ||
2371 | return NULL; | |
6dde1f33 | 2372 | } |
4fbf50b3 | 2373 | |
4ad49000 LP |
2374 | int manager_notify_cgroup_empty(Manager *m, const char *cgroup) { |
2375 | Unit *u; | |
4fbf50b3 | 2376 | |
4ad49000 LP |
2377 | assert(m); |
2378 | assert(cgroup); | |
4fbf50b3 | 2379 | |
09e24654 LP |
2380 | /* Called on the legacy hierarchy whenever we get an explicit cgroup notification from the cgroup agent process |
2381 | * or from the --system instance */ | |
2382 | ||
d8fdc620 LP |
2383 | log_debug("Got cgroup empty notification for: %s", cgroup); |
2384 | ||
4ad49000 | 2385 | u = manager_get_unit_by_cgroup(m, cgroup); |
5ad096b3 LP |
2386 | if (!u) |
2387 | return 0; | |
b56c28c3 | 2388 | |
09e24654 LP |
2389 | unit_add_to_cgroup_empty_queue(u); |
2390 | return 1; | |
5ad096b3 LP |
2391 | } |
2392 | ||
2393 | int unit_get_memory_current(Unit *u, uint64_t *ret) { | |
2394 | _cleanup_free_ char *v = NULL; | |
2395 | int r; | |
2396 | ||
2397 | assert(u); | |
2398 | assert(ret); | |
2399 | ||
2e4025c0 | 2400 | if (!UNIT_CGROUP_BOOL(u, memory_accounting)) |
cf3b4be1 LP |
2401 | return -ENODATA; |
2402 | ||
5ad096b3 LP |
2403 | if (!u->cgroup_path) |
2404 | return -ENODATA; | |
2405 | ||
1f73aa00 LP |
2406 | /* The root cgroup doesn't expose this information, let's get it from /proc instead */ |
2407 | if (unit_has_root_cgroup(u)) | |
2408 | return procfs_memory_get_current(ret); | |
2409 | ||
efdb0237 | 2410 | if ((u->cgroup_realized_mask & CGROUP_MASK_MEMORY) == 0) |
5ad096b3 LP |
2411 | return -ENODATA; |
2412 | ||
b4cccbc1 LP |
2413 | r = cg_all_unified(); |
2414 | if (r < 0) | |
2415 | return r; | |
2416 | if (r > 0) | |
efdb0237 | 2417 | r = cg_get_attribute("memory", u->cgroup_path, "memory.current", &v); |
b4cccbc1 LP |
2418 | else |
2419 | r = cg_get_attribute("memory", u->cgroup_path, "memory.usage_in_bytes", &v); | |
5ad096b3 LP |
2420 | if (r == -ENOENT) |
2421 | return -ENODATA; | |
2422 | if (r < 0) | |
2423 | return r; | |
2424 | ||
2425 | return safe_atou64(v, ret); | |
2426 | } | |
2427 | ||
03a7b521 LP |
2428 | int unit_get_tasks_current(Unit *u, uint64_t *ret) { |
2429 | _cleanup_free_ char *v = NULL; | |
2430 | int r; | |
2431 | ||
2432 | assert(u); | |
2433 | assert(ret); | |
2434 | ||
2e4025c0 | 2435 | if (!UNIT_CGROUP_BOOL(u, tasks_accounting)) |
cf3b4be1 LP |
2436 | return -ENODATA; |
2437 | ||
03a7b521 LP |
2438 | if (!u->cgroup_path) |
2439 | return -ENODATA; | |
2440 | ||
c36a69f4 LP |
2441 | /* The root cgroup doesn't expose this information, let's get it from /proc instead */ |
2442 | if (unit_has_root_cgroup(u)) | |
2443 | return procfs_tasks_get_current(ret); | |
2444 | ||
1f73aa00 LP |
2445 | if ((u->cgroup_realized_mask & CGROUP_MASK_PIDS) == 0) |
2446 | return -ENODATA; | |
2447 | ||
03a7b521 LP |
2448 | r = cg_get_attribute("pids", u->cgroup_path, "pids.current", &v); |
2449 | if (r == -ENOENT) | |
2450 | return -ENODATA; | |
2451 | if (r < 0) | |
2452 | return r; | |
2453 | ||
2454 | return safe_atou64(v, ret); | |
2455 | } | |
2456 | ||
5ad096b3 LP |
2457 | static int unit_get_cpu_usage_raw(Unit *u, nsec_t *ret) { |
2458 | _cleanup_free_ char *v = NULL; | |
2459 | uint64_t ns; | |
2460 | int r; | |
2461 | ||
2462 | assert(u); | |
2463 | assert(ret); | |
2464 | ||
2465 | if (!u->cgroup_path) | |
2466 | return -ENODATA; | |
2467 | ||
1f73aa00 LP |
2468 | /* The root cgroup doesn't expose this information, let's get it from /proc instead */ |
2469 | if (unit_has_root_cgroup(u)) | |
2470 | return procfs_cpu_get_usage(ret); | |
2471 | ||
b4cccbc1 LP |
2472 | r = cg_all_unified(); |
2473 | if (r < 0) | |
2474 | return r; | |
2475 | if (r > 0) { | |
66ebf6c0 TH |
2476 | _cleanup_free_ char *val = NULL; |
2477 | uint64_t us; | |
5ad096b3 | 2478 | |
66ebf6c0 TH |
2479 | if ((u->cgroup_realized_mask & CGROUP_MASK_CPU) == 0) |
2480 | return -ENODATA; | |
5ad096b3 | 2481 | |
b734a4ff | 2482 | r = cg_get_keyed_attribute("cpu", u->cgroup_path, "cpu.stat", STRV_MAKE("usage_usec"), &val); |
66ebf6c0 TH |
2483 | if (r < 0) |
2484 | return r; | |
b734a4ff LP |
2485 | if (IN_SET(r, -ENOENT, -ENXIO)) |
2486 | return -ENODATA; | |
66ebf6c0 TH |
2487 | |
2488 | r = safe_atou64(val, &us); | |
2489 | if (r < 0) | |
2490 | return r; | |
2491 | ||
2492 | ns = us * NSEC_PER_USEC; | |
2493 | } else { | |
2494 | if ((u->cgroup_realized_mask & CGROUP_MASK_CPUACCT) == 0) | |
2495 | return -ENODATA; | |
2496 | ||
2497 | r = cg_get_attribute("cpuacct", u->cgroup_path, "cpuacct.usage", &v); | |
2498 | if (r == -ENOENT) | |
2499 | return -ENODATA; | |
2500 | if (r < 0) | |
2501 | return r; | |
2502 | ||
2503 | r = safe_atou64(v, &ns); | |
2504 | if (r < 0) | |
2505 | return r; | |
2506 | } | |
5ad096b3 LP |
2507 | |
2508 | *ret = ns; | |
2509 | return 0; | |
2510 | } | |
2511 | ||
2512 | int unit_get_cpu_usage(Unit *u, nsec_t *ret) { | |
2513 | nsec_t ns; | |
2514 | int r; | |
2515 | ||
fe700f46 LP |
2516 | assert(u); |
2517 | ||
2518 | /* Retrieve the current CPU usage counter. This will subtract the CPU counter taken when the unit was | |
2519 | * started. If the cgroup has been removed already, returns the last cached value. To cache the value, simply | |
2520 | * call this function with a NULL return value. */ | |
2521 | ||
2e4025c0 | 2522 | if (!UNIT_CGROUP_BOOL(u, cpu_accounting)) |
cf3b4be1 LP |
2523 | return -ENODATA; |
2524 | ||
5ad096b3 | 2525 | r = unit_get_cpu_usage_raw(u, &ns); |
fe700f46 LP |
2526 | if (r == -ENODATA && u->cpu_usage_last != NSEC_INFINITY) { |
2527 | /* If we can't get the CPU usage anymore (because the cgroup was already removed, for example), use our | |
2528 | * cached value. */ | |
2529 | ||
2530 | if (ret) | |
2531 | *ret = u->cpu_usage_last; | |
2532 | return 0; | |
2533 | } | |
5ad096b3 LP |
2534 | if (r < 0) |
2535 | return r; | |
2536 | ||
66ebf6c0 TH |
2537 | if (ns > u->cpu_usage_base) |
2538 | ns -= u->cpu_usage_base; | |
5ad096b3 LP |
2539 | else |
2540 | ns = 0; | |
2541 | ||
fe700f46 LP |
2542 | u->cpu_usage_last = ns; |
2543 | if (ret) | |
2544 | *ret = ns; | |
2545 | ||
5ad096b3 LP |
2546 | return 0; |
2547 | } | |
2548 | ||
906c06f6 DM |
2549 | int unit_get_ip_accounting( |
2550 | Unit *u, | |
2551 | CGroupIPAccountingMetric metric, | |
2552 | uint64_t *ret) { | |
2553 | ||
6b659ed8 | 2554 | uint64_t value; |
906c06f6 DM |
2555 | int fd, r; |
2556 | ||
2557 | assert(u); | |
2558 | assert(metric >= 0); | |
2559 | assert(metric < _CGROUP_IP_ACCOUNTING_METRIC_MAX); | |
2560 | assert(ret); | |
2561 | ||
2e4025c0 | 2562 | if (!UNIT_CGROUP_BOOL(u, ip_accounting)) |
cf3b4be1 LP |
2563 | return -ENODATA; |
2564 | ||
906c06f6 DM |
2565 | fd = IN_SET(metric, CGROUP_IP_INGRESS_BYTES, CGROUP_IP_INGRESS_PACKETS) ? |
2566 | u->ip_accounting_ingress_map_fd : | |
2567 | u->ip_accounting_egress_map_fd; | |
906c06f6 DM |
2568 | if (fd < 0) |
2569 | return -ENODATA; | |
2570 | ||
2571 | if (IN_SET(metric, CGROUP_IP_INGRESS_BYTES, CGROUP_IP_EGRESS_BYTES)) | |
6b659ed8 | 2572 | r = bpf_firewall_read_accounting(fd, &value, NULL); |
906c06f6 | 2573 | else |
6b659ed8 LP |
2574 | r = bpf_firewall_read_accounting(fd, NULL, &value); |
2575 | if (r < 0) | |
2576 | return r; | |
2577 | ||
2578 | /* Add in additional metrics from a previous runtime. Note that when reexecing/reloading the daemon we compile | |
2579 | * all BPF programs and maps anew, but serialize the old counters. When deserializing we store them in the | |
2580 | * ip_accounting_extra[] field, and add them in here transparently. */ | |
2581 | ||
2582 | *ret = value + u->ip_accounting_extra[metric]; | |
906c06f6 DM |
2583 | |
2584 | return r; | |
2585 | } | |
2586 | ||
2587 | int unit_reset_cpu_accounting(Unit *u) { | |
5ad096b3 LP |
2588 | nsec_t ns; |
2589 | int r; | |
2590 | ||
2591 | assert(u); | |
2592 | ||
fe700f46 LP |
2593 | u->cpu_usage_last = NSEC_INFINITY; |
2594 | ||
5ad096b3 LP |
2595 | r = unit_get_cpu_usage_raw(u, &ns); |
2596 | if (r < 0) { | |
66ebf6c0 | 2597 | u->cpu_usage_base = 0; |
5ad096b3 | 2598 | return r; |
b56c28c3 | 2599 | } |
2633eb83 | 2600 | |
66ebf6c0 | 2601 | u->cpu_usage_base = ns; |
4ad49000 | 2602 | return 0; |
4fbf50b3 LP |
2603 | } |
2604 | ||
906c06f6 DM |
2605 | int unit_reset_ip_accounting(Unit *u) { |
2606 | int r = 0, q = 0; | |
2607 | ||
2608 | assert(u); | |
2609 | ||
2610 | if (u->ip_accounting_ingress_map_fd >= 0) | |
2611 | r = bpf_firewall_reset_accounting(u->ip_accounting_ingress_map_fd); | |
2612 | ||
2613 | if (u->ip_accounting_egress_map_fd >= 0) | |
2614 | q = bpf_firewall_reset_accounting(u->ip_accounting_egress_map_fd); | |
2615 | ||
6b659ed8 LP |
2616 | zero(u->ip_accounting_extra); |
2617 | ||
906c06f6 DM |
2618 | return r < 0 ? r : q; |
2619 | } | |
2620 | ||
e7ab4d1a LP |
2621 | void unit_invalidate_cgroup(Unit *u, CGroupMask m) { |
2622 | assert(u); | |
2623 | ||
2624 | if (!UNIT_HAS_CGROUP_CONTEXT(u)) | |
2625 | return; | |
2626 | ||
2627 | if (m == 0) | |
2628 | return; | |
2629 | ||
538b4852 TH |
2630 | /* always invalidate compat pairs together */ |
2631 | if (m & (CGROUP_MASK_IO | CGROUP_MASK_BLKIO)) | |
2632 | m |= CGROUP_MASK_IO | CGROUP_MASK_BLKIO; | |
2633 | ||
7cce4fb7 LP |
2634 | if (m & (CGROUP_MASK_CPU | CGROUP_MASK_CPUACCT)) |
2635 | m |= CGROUP_MASK_CPU | CGROUP_MASK_CPUACCT; | |
2636 | ||
60c728ad | 2637 | if ((u->cgroup_realized_mask & m) == 0) /* NOP? */ |
e7ab4d1a LP |
2638 | return; |
2639 | ||
2640 | u->cgroup_realized_mask &= ~m; | |
91a6073e | 2641 | unit_add_to_cgroup_realize_queue(u); |
e7ab4d1a LP |
2642 | } |
2643 | ||
906c06f6 DM |
2644 | void unit_invalidate_cgroup_bpf(Unit *u) { |
2645 | assert(u); | |
2646 | ||
2647 | if (!UNIT_HAS_CGROUP_CONTEXT(u)) | |
2648 | return; | |
2649 | ||
60c728ad | 2650 | if (u->cgroup_bpf_state == UNIT_CGROUP_BPF_INVALIDATED) /* NOP? */ |
906c06f6 DM |
2651 | return; |
2652 | ||
2653 | u->cgroup_bpf_state = UNIT_CGROUP_BPF_INVALIDATED; | |
91a6073e | 2654 | unit_add_to_cgroup_realize_queue(u); |
906c06f6 DM |
2655 | |
2656 | /* If we are a slice unit, we also need to put compile a new BPF program for all our children, as the IP access | |
2657 | * list of our children includes our own. */ | |
2658 | if (u->type == UNIT_SLICE) { | |
2659 | Unit *member; | |
2660 | Iterator i; | |
eef85c4a | 2661 | void *v; |
906c06f6 | 2662 | |
eef85c4a | 2663 | HASHMAP_FOREACH_KEY(v, member, u->dependencies[UNIT_BEFORE], i) { |
906c06f6 DM |
2664 | if (member == u) |
2665 | continue; | |
2666 | ||
2667 | if (UNIT_DEREF(member->slice) != u) | |
2668 | continue; | |
2669 | ||
2670 | unit_invalidate_cgroup_bpf(member); | |
2671 | } | |
2672 | } | |
2673 | } | |
2674 | ||
1d9cc876 LP |
2675 | bool unit_cgroup_delegate(Unit *u) { |
2676 | CGroupContext *c; | |
2677 | ||
2678 | assert(u); | |
2679 | ||
2680 | if (!UNIT_VTABLE(u)->can_delegate) | |
2681 | return false; | |
2682 | ||
2683 | c = unit_get_cgroup_context(u); | |
2684 | if (!c) | |
2685 | return false; | |
2686 | ||
2687 | return c->delegate; | |
2688 | } | |
2689 | ||
e7ab4d1a LP |
2690 | void manager_invalidate_startup_units(Manager *m) { |
2691 | Iterator i; | |
2692 | Unit *u; | |
2693 | ||
2694 | assert(m); | |
2695 | ||
2696 | SET_FOREACH(u, m->startup_units, i) | |
13c31542 | 2697 | unit_invalidate_cgroup(u, CGROUP_MASK_CPU|CGROUP_MASK_IO|CGROUP_MASK_BLKIO); |
e7ab4d1a LP |
2698 | } |
2699 | ||
4ad49000 LP |
2700 | static const char* const cgroup_device_policy_table[_CGROUP_DEVICE_POLICY_MAX] = { |
2701 | [CGROUP_AUTO] = "auto", | |
2702 | [CGROUP_CLOSED] = "closed", | |
2703 | [CGROUP_STRICT] = "strict", | |
2704 | }; | |
4fbf50b3 | 2705 | |
4ad49000 | 2706 | DEFINE_STRING_TABLE_LOOKUP(cgroup_device_policy, CGroupDevicePolicy); |