]>
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" |
084c7007 | 10 | #include "bpf-devices.h" |
6592b975 | 11 | #include "bus-error.h" |
03a7b521 | 12 | #include "cgroup-util.h" |
3ffd4af2 LP |
13 | #include "cgroup.h" |
14 | #include "fd-util.h" | |
0d39fa9c | 15 | #include "fileio.h" |
77601719 | 16 | #include "fs-util.h" |
6bedfcbb | 17 | #include "parse-util.h" |
9eb977db | 18 | #include "path-util.h" |
03a7b521 | 19 | #include "process-util.h" |
c36a69f4 | 20 | #include "procfs-util.h" |
9444b1f2 | 21 | #include "special.h" |
74c48bf5 | 22 | #include "stat-util.h" |
906c06f6 | 23 | #include "stdio-util.h" |
8b43440b | 24 | #include "string-table.h" |
07630cea | 25 | #include "string-util.h" |
cc6271f1 | 26 | #include "virt.h" |
8e274523 | 27 | |
9a054909 LP |
28 | #define CGROUP_CPU_QUOTA_PERIOD_USEC ((usec_t) 100 * USEC_PER_MSEC) |
29 | ||
39b9fefb LP |
30 | /* Returns the log level to use when cgroup attribute writes fail. When an attribute is missing or we have access |
31 | * problems we downgrade to LOG_DEBUG. This is supposed to be nice to container managers and kernels which want to mask | |
32 | * out specific attributes from us. */ | |
33 | #define LOG_LEVEL_CGROUP_WRITE(r) (IN_SET(abs(r), ENOENT, EROFS, EACCES, EPERM) ? LOG_DEBUG : LOG_WARNING) | |
34 | ||
611c4f8a | 35 | bool manager_owns_host_root_cgroup(Manager *m) { |
cc6271f1 LP |
36 | assert(m); |
37 | ||
38 | /* Returns true if we are managing the root cgroup. Note that it isn't sufficient to just check whether the | |
39 | * group root path equals "/" since that will also be the case if CLONE_NEWCGROUP is in the mix. Since there's | |
40 | * appears to be no nice way to detect whether we are in a CLONE_NEWCGROUP namespace we instead just check if | |
41 | * we run in any kind of container virtualization. */ | |
42 | ||
28cfdc5a LP |
43 | if (MANAGER_IS_USER(m)) |
44 | return false; | |
45 | ||
cc6271f1 LP |
46 | if (detect_container() > 0) |
47 | return false; | |
48 | ||
57ea45e1 | 49 | return empty_or_root(m->cgroup_root); |
cc6271f1 LP |
50 | } |
51 | ||
611c4f8a | 52 | bool unit_has_host_root_cgroup(Unit *u) { |
f3725e64 LP |
53 | assert(u); |
54 | ||
cc6271f1 LP |
55 | /* Returns whether this unit manages the root cgroup. This will return true if this unit is the root slice and |
56 | * the manager manages the root cgroup. */ | |
f3725e64 | 57 | |
611c4f8a | 58 | if (!manager_owns_host_root_cgroup(u->manager)) |
f3725e64 LP |
59 | return false; |
60 | ||
cc6271f1 | 61 | return unit_has_name(u, SPECIAL_ROOT_SLICE); |
f3725e64 LP |
62 | } |
63 | ||
293d32df LP |
64 | static int set_attribute_and_warn(Unit *u, const char *controller, const char *attribute, const char *value) { |
65 | int r; | |
66 | ||
67 | r = cg_set_attribute(controller, u->cgroup_path, attribute, value); | |
68 | if (r < 0) | |
69 | log_unit_full(u, LOG_LEVEL_CGROUP_WRITE(r), r, "Failed to set '%s' attribute on '%s' to '%.*s': %m", | |
70 | strna(attribute), isempty(u->cgroup_path) ? "/" : u->cgroup_path, (int) strcspn(value, NEWLINE), value); | |
71 | ||
72 | return r; | |
73 | } | |
74 | ||
2b40998d | 75 | static void cgroup_compat_warn(void) { |
128fadc9 TH |
76 | static bool cgroup_compat_warned = false; |
77 | ||
78 | if (cgroup_compat_warned) | |
79 | return; | |
80 | ||
cc6271f1 LP |
81 | log_warning("cgroup compatibility translation between legacy and unified hierarchy settings activated. " |
82 | "See cgroup-compat debug messages for details."); | |
83 | ||
128fadc9 TH |
84 | cgroup_compat_warned = true; |
85 | } | |
86 | ||
87 | #define log_cgroup_compat(unit, fmt, ...) do { \ | |
88 | cgroup_compat_warn(); \ | |
89 | log_unit_debug(unit, "cgroup-compat: " fmt, ##__VA_ARGS__); \ | |
2b40998d | 90 | } while (false) |
128fadc9 | 91 | |
4ad49000 LP |
92 | void cgroup_context_init(CGroupContext *c) { |
93 | assert(c); | |
94 | ||
de8a711a | 95 | /* Initialize everything to the kernel defaults. */ |
4ad49000 | 96 | |
de8a711a LP |
97 | *c = (CGroupContext) { |
98 | .cpu_weight = CGROUP_WEIGHT_INVALID, | |
99 | .startup_cpu_weight = CGROUP_WEIGHT_INVALID, | |
100 | .cpu_quota_per_sec_usec = USEC_INFINITY, | |
66ebf6c0 | 101 | |
de8a711a LP |
102 | .cpu_shares = CGROUP_CPU_SHARES_INVALID, |
103 | .startup_cpu_shares = CGROUP_CPU_SHARES_INVALID, | |
d53d9474 | 104 | |
de8a711a LP |
105 | .memory_high = CGROUP_LIMIT_MAX, |
106 | .memory_max = CGROUP_LIMIT_MAX, | |
107 | .memory_swap_max = CGROUP_LIMIT_MAX, | |
da4d897e | 108 | |
de8a711a | 109 | .memory_limit = CGROUP_LIMIT_MAX, |
b2f8b02e | 110 | |
de8a711a LP |
111 | .io_weight = CGROUP_WEIGHT_INVALID, |
112 | .startup_io_weight = CGROUP_WEIGHT_INVALID, | |
13c31542 | 113 | |
de8a711a LP |
114 | .blockio_weight = CGROUP_BLKIO_WEIGHT_INVALID, |
115 | .startup_blockio_weight = CGROUP_BLKIO_WEIGHT_INVALID, | |
d53d9474 | 116 | |
de8a711a LP |
117 | .tasks_max = CGROUP_LIMIT_MAX, |
118 | }; | |
4ad49000 | 119 | } |
8e274523 | 120 | |
4ad49000 LP |
121 | void cgroup_context_free_device_allow(CGroupContext *c, CGroupDeviceAllow *a) { |
122 | assert(c); | |
123 | assert(a); | |
124 | ||
71fda00f | 125 | LIST_REMOVE(device_allow, c->device_allow, a); |
4ad49000 LP |
126 | free(a->path); |
127 | free(a); | |
128 | } | |
129 | ||
13c31542 TH |
130 | void cgroup_context_free_io_device_weight(CGroupContext *c, CGroupIODeviceWeight *w) { |
131 | assert(c); | |
132 | assert(w); | |
133 | ||
134 | LIST_REMOVE(device_weights, c->io_device_weights, w); | |
135 | free(w->path); | |
136 | free(w); | |
137 | } | |
138 | ||
6ae4283c TH |
139 | void cgroup_context_free_io_device_latency(CGroupContext *c, CGroupIODeviceLatency *l) { |
140 | assert(c); | |
141 | assert(l); | |
142 | ||
143 | LIST_REMOVE(device_latencies, c->io_device_latencies, l); | |
144 | free(l->path); | |
145 | free(l); | |
146 | } | |
147 | ||
13c31542 TH |
148 | void cgroup_context_free_io_device_limit(CGroupContext *c, CGroupIODeviceLimit *l) { |
149 | assert(c); | |
150 | assert(l); | |
151 | ||
152 | LIST_REMOVE(device_limits, c->io_device_limits, l); | |
153 | free(l->path); | |
154 | free(l); | |
155 | } | |
156 | ||
4ad49000 LP |
157 | void cgroup_context_free_blockio_device_weight(CGroupContext *c, CGroupBlockIODeviceWeight *w) { |
158 | assert(c); | |
159 | assert(w); | |
160 | ||
71fda00f | 161 | LIST_REMOVE(device_weights, c->blockio_device_weights, w); |
4ad49000 LP |
162 | free(w->path); |
163 | free(w); | |
164 | } | |
165 | ||
166 | void cgroup_context_free_blockio_device_bandwidth(CGroupContext *c, CGroupBlockIODeviceBandwidth *b) { | |
167 | assert(c); | |
8e274523 | 168 | assert(b); |
8e274523 | 169 | |
71fda00f | 170 | LIST_REMOVE(device_bandwidths, c->blockio_device_bandwidths, b); |
4ad49000 LP |
171 | free(b->path); |
172 | free(b); | |
173 | } | |
174 | ||
175 | void cgroup_context_done(CGroupContext *c) { | |
176 | assert(c); | |
177 | ||
13c31542 TH |
178 | while (c->io_device_weights) |
179 | cgroup_context_free_io_device_weight(c, c->io_device_weights); | |
180 | ||
6ae4283c TH |
181 | while (c->io_device_latencies) |
182 | cgroup_context_free_io_device_latency(c, c->io_device_latencies); | |
183 | ||
13c31542 TH |
184 | while (c->io_device_limits) |
185 | cgroup_context_free_io_device_limit(c, c->io_device_limits); | |
186 | ||
4ad49000 LP |
187 | while (c->blockio_device_weights) |
188 | cgroup_context_free_blockio_device_weight(c, c->blockio_device_weights); | |
189 | ||
190 | while (c->blockio_device_bandwidths) | |
191 | cgroup_context_free_blockio_device_bandwidth(c, c->blockio_device_bandwidths); | |
192 | ||
193 | while (c->device_allow) | |
194 | cgroup_context_free_device_allow(c, c->device_allow); | |
6a48d82f DM |
195 | |
196 | c->ip_address_allow = ip_address_access_free_all(c->ip_address_allow); | |
197 | c->ip_address_deny = ip_address_access_free_all(c->ip_address_deny); | |
4ad49000 LP |
198 | } |
199 | ||
200 | void cgroup_context_dump(CGroupContext *c, FILE* f, const char *prefix) { | |
13c31542 TH |
201 | CGroupIODeviceLimit *il; |
202 | CGroupIODeviceWeight *iw; | |
6ae4283c | 203 | CGroupIODeviceLatency *l; |
4ad49000 LP |
204 | CGroupBlockIODeviceBandwidth *b; |
205 | CGroupBlockIODeviceWeight *w; | |
206 | CGroupDeviceAllow *a; | |
c21c9906 | 207 | IPAddressAccessItem *iaai; |
9a054909 | 208 | char u[FORMAT_TIMESPAN_MAX]; |
4ad49000 LP |
209 | |
210 | assert(c); | |
211 | assert(f); | |
212 | ||
213 | prefix = strempty(prefix); | |
214 | ||
215 | fprintf(f, | |
216 | "%sCPUAccounting=%s\n" | |
13c31542 | 217 | "%sIOAccounting=%s\n" |
4ad49000 LP |
218 | "%sBlockIOAccounting=%s\n" |
219 | "%sMemoryAccounting=%s\n" | |
d53d9474 | 220 | "%sTasksAccounting=%s\n" |
c21c9906 | 221 | "%sIPAccounting=%s\n" |
66ebf6c0 TH |
222 | "%sCPUWeight=%" PRIu64 "\n" |
223 | "%sStartupCPUWeight=%" PRIu64 "\n" | |
d53d9474 LP |
224 | "%sCPUShares=%" PRIu64 "\n" |
225 | "%sStartupCPUShares=%" PRIu64 "\n" | |
b2f8b02e | 226 | "%sCPUQuotaPerSecSec=%s\n" |
13c31542 TH |
227 | "%sIOWeight=%" PRIu64 "\n" |
228 | "%sStartupIOWeight=%" PRIu64 "\n" | |
d53d9474 LP |
229 | "%sBlockIOWeight=%" PRIu64 "\n" |
230 | "%sStartupBlockIOWeight=%" PRIu64 "\n" | |
48422635 | 231 | "%sMemoryMin=%" PRIu64 "\n" |
da4d897e TH |
232 | "%sMemoryLow=%" PRIu64 "\n" |
233 | "%sMemoryHigh=%" PRIu64 "\n" | |
234 | "%sMemoryMax=%" PRIu64 "\n" | |
96e131ea | 235 | "%sMemorySwapMax=%" PRIu64 "\n" |
4ad49000 | 236 | "%sMemoryLimit=%" PRIu64 "\n" |
03a7b521 | 237 | "%sTasksMax=%" PRIu64 "\n" |
a931ad47 LP |
238 | "%sDevicePolicy=%s\n" |
239 | "%sDelegate=%s\n", | |
4ad49000 | 240 | prefix, yes_no(c->cpu_accounting), |
13c31542 | 241 | prefix, yes_no(c->io_accounting), |
4ad49000 LP |
242 | prefix, yes_no(c->blockio_accounting), |
243 | prefix, yes_no(c->memory_accounting), | |
d53d9474 | 244 | prefix, yes_no(c->tasks_accounting), |
c21c9906 | 245 | prefix, yes_no(c->ip_accounting), |
66ebf6c0 TH |
246 | prefix, c->cpu_weight, |
247 | prefix, c->startup_cpu_weight, | |
4ad49000 | 248 | prefix, c->cpu_shares, |
95ae05c0 | 249 | prefix, c->startup_cpu_shares, |
b1d6dcf5 | 250 | prefix, format_timespan(u, sizeof(u), c->cpu_quota_per_sec_usec, 1), |
13c31542 TH |
251 | prefix, c->io_weight, |
252 | prefix, c->startup_io_weight, | |
4ad49000 | 253 | prefix, c->blockio_weight, |
95ae05c0 | 254 | prefix, c->startup_blockio_weight, |
48422635 | 255 | prefix, c->memory_min, |
da4d897e TH |
256 | prefix, c->memory_low, |
257 | prefix, c->memory_high, | |
258 | prefix, c->memory_max, | |
96e131ea | 259 | prefix, c->memory_swap_max, |
4ad49000 | 260 | prefix, c->memory_limit, |
03a7b521 | 261 | prefix, c->tasks_max, |
a931ad47 LP |
262 | prefix, cgroup_device_policy_to_string(c->device_policy), |
263 | prefix, yes_no(c->delegate)); | |
4ad49000 | 264 | |
02638280 LP |
265 | if (c->delegate) { |
266 | _cleanup_free_ char *t = NULL; | |
267 | ||
268 | (void) cg_mask_to_string(c->delegate_controllers, &t); | |
269 | ||
47a78d41 | 270 | fprintf(f, "%sDelegateControllers=%s\n", |
02638280 LP |
271 | prefix, |
272 | strempty(t)); | |
273 | } | |
274 | ||
4ad49000 LP |
275 | LIST_FOREACH(device_allow, a, c->device_allow) |
276 | fprintf(f, | |
277 | "%sDeviceAllow=%s %s%s%s\n", | |
278 | prefix, | |
279 | a->path, | |
280 | a->r ? "r" : "", a->w ? "w" : "", a->m ? "m" : ""); | |
281 | ||
13c31542 TH |
282 | LIST_FOREACH(device_weights, iw, c->io_device_weights) |
283 | fprintf(f, | |
6ae4283c | 284 | "%sIODeviceWeight=%s %" PRIu64 "\n", |
13c31542 TH |
285 | prefix, |
286 | iw->path, | |
287 | iw->weight); | |
288 | ||
6ae4283c TH |
289 | LIST_FOREACH(device_latencies, l, c->io_device_latencies) |
290 | fprintf(f, | |
291 | "%sIODeviceLatencyTargetSec=%s %s\n", | |
292 | prefix, | |
293 | l->path, | |
294 | format_timespan(u, sizeof(u), l->target_usec, 1)); | |
295 | ||
13c31542 TH |
296 | LIST_FOREACH(device_limits, il, c->io_device_limits) { |
297 | char buf[FORMAT_BYTES_MAX]; | |
9be57249 TH |
298 | CGroupIOLimitType type; |
299 | ||
300 | for (type = 0; type < _CGROUP_IO_LIMIT_TYPE_MAX; type++) | |
301 | if (il->limits[type] != cgroup_io_limit_defaults[type]) | |
302 | fprintf(f, | |
303 | "%s%s=%s %s\n", | |
304 | prefix, | |
305 | cgroup_io_limit_type_to_string(type), | |
306 | il->path, | |
307 | format_bytes(buf, sizeof(buf), il->limits[type])); | |
13c31542 TH |
308 | } |
309 | ||
4ad49000 LP |
310 | LIST_FOREACH(device_weights, w, c->blockio_device_weights) |
311 | fprintf(f, | |
d53d9474 | 312 | "%sBlockIODeviceWeight=%s %" PRIu64, |
4ad49000 LP |
313 | prefix, |
314 | w->path, | |
315 | w->weight); | |
316 | ||
317 | LIST_FOREACH(device_bandwidths, b, c->blockio_device_bandwidths) { | |
318 | char buf[FORMAT_BYTES_MAX]; | |
319 | ||
979d0311 TH |
320 | if (b->rbps != CGROUP_LIMIT_MAX) |
321 | fprintf(f, | |
322 | "%sBlockIOReadBandwidth=%s %s\n", | |
323 | prefix, | |
324 | b->path, | |
325 | format_bytes(buf, sizeof(buf), b->rbps)); | |
326 | if (b->wbps != CGROUP_LIMIT_MAX) | |
327 | fprintf(f, | |
328 | "%sBlockIOWriteBandwidth=%s %s\n", | |
329 | prefix, | |
330 | b->path, | |
331 | format_bytes(buf, sizeof(buf), b->wbps)); | |
4ad49000 | 332 | } |
c21c9906 LP |
333 | |
334 | LIST_FOREACH(items, iaai, c->ip_address_allow) { | |
335 | _cleanup_free_ char *k = NULL; | |
336 | ||
337 | (void) in_addr_to_string(iaai->family, &iaai->address, &k); | |
338 | fprintf(f, "%sIPAddressAllow=%s/%u\n", prefix, strnull(k), iaai->prefixlen); | |
339 | } | |
340 | ||
341 | LIST_FOREACH(items, iaai, c->ip_address_deny) { | |
342 | _cleanup_free_ char *k = NULL; | |
343 | ||
344 | (void) in_addr_to_string(iaai->family, &iaai->address, &k); | |
345 | fprintf(f, "%sIPAddressDeny=%s/%u\n", prefix, strnull(k), iaai->prefixlen); | |
346 | } | |
4ad49000 LP |
347 | } |
348 | ||
fd870bac YW |
349 | int cgroup_add_device_allow(CGroupContext *c, const char *dev, const char *mode) { |
350 | _cleanup_free_ CGroupDeviceAllow *a = NULL; | |
351 | _cleanup_free_ char *d = NULL; | |
352 | ||
353 | assert(c); | |
354 | assert(dev); | |
355 | assert(isempty(mode) || in_charset(mode, "rwm")); | |
356 | ||
357 | a = new(CGroupDeviceAllow, 1); | |
358 | if (!a) | |
359 | return -ENOMEM; | |
360 | ||
361 | d = strdup(dev); | |
362 | if (!d) | |
363 | return -ENOMEM; | |
364 | ||
365 | *a = (CGroupDeviceAllow) { | |
366 | .path = TAKE_PTR(d), | |
490c5a37 LP |
367 | .r = isempty(mode) || strchr(mode, 'r'), |
368 | .w = isempty(mode) || strchr(mode, 'w'), | |
369 | .m = isempty(mode) || strchr(mode, 'm'), | |
fd870bac YW |
370 | }; |
371 | ||
372 | LIST_PREPEND(device_allow, c->device_allow, a); | |
373 | TAKE_PTR(a); | |
374 | ||
375 | return 0; | |
376 | } | |
377 | ||
0d2d6fbf CD |
378 | static void cgroup_xattr_apply(Unit *u) { |
379 | char ids[SD_ID128_STRING_MAX]; | |
380 | int r; | |
381 | ||
382 | assert(u); | |
383 | ||
384 | if (!MANAGER_IS_SYSTEM(u->manager)) | |
385 | return; | |
386 | ||
387 | if (sd_id128_is_null(u->invocation_id)) | |
388 | return; | |
389 | ||
390 | r = cg_set_xattr(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path, | |
391 | "trusted.invocation_id", | |
392 | sd_id128_to_string(u->invocation_id, ids), 32, | |
393 | 0); | |
394 | if (r < 0) | |
395 | log_unit_debug_errno(u, r, "Failed to set invocation ID on control group %s, ignoring: %m", u->cgroup_path); | |
396 | } | |
397 | ||
45c2e068 | 398 | static int lookup_block_device(const char *p, dev_t *ret) { |
d5aecba6 | 399 | struct stat st = {}; |
45c2e068 | 400 | int r; |
4ad49000 LP |
401 | |
402 | assert(p); | |
45c2e068 | 403 | assert(ret); |
4ad49000 | 404 | |
d5aecba6 LP |
405 | r = device_path_parse_major_minor(p, &st.st_mode, &st.st_rdev); |
406 | if (r == -ENODEV) { /* not a parsable device node, need to go to disk */ | |
407 | if (stat(p, &st) < 0) | |
408 | return log_warning_errno(errno, "Couldn't stat device '%s': %m", p); | |
409 | } else if (r < 0) | |
410 | return log_warning_errno(r, "Failed to parse major/minor from path '%s': %m", p); | |
411 | ||
412 | if (S_ISCHR(st.st_mode)) { | |
413 | log_warning("Device node '%s' is a character device, but block device needed.", p); | |
414 | return -ENOTBLK; | |
415 | } else if (S_ISBLK(st.st_mode)) | |
45c2e068 LP |
416 | *ret = st.st_rdev; |
417 | else if (major(st.st_dev) != 0) | |
418 | *ret = st.st_dev; /* If this is not a device node then use the block device this file is stored on */ | |
419 | else { | |
420 | /* If this is btrfs, getting the backing block device is a bit harder */ | |
421 | r = btrfs_get_block_device(p, ret); | |
422 | if (r < 0 && r != -ENOTTY) | |
423 | return log_warning_errno(r, "Failed to determine block device backing btrfs file system '%s': %m", p); | |
424 | if (r == -ENOTTY) { | |
425 | log_warning("'%s' is not a block device node, and file system block device cannot be determined or is not local.", p); | |
426 | return -ENODEV; | |
427 | } | |
4ad49000 | 428 | } |
8e274523 | 429 | |
45c2e068 LP |
430 | /* If this is a LUKS device, try to get the originating block device */ |
431 | (void) block_get_originating(*ret, ret); | |
432 | ||
433 | /* If this is a partition, try to get the originating block device */ | |
434 | (void) block_get_whole_disk(*ret, ret); | |
8e274523 | 435 | return 0; |
8e274523 LP |
436 | } |
437 | ||
084c7007 | 438 | static int whitelist_device(BPFProgram *prog, const char *path, const char *node, const char *acc) { |
846b3bd6 | 439 | struct stat st = {}; |
8c6db833 | 440 | int r; |
8e274523 | 441 | |
4ad49000 LP |
442 | assert(path); |
443 | assert(acc); | |
8e274523 | 444 | |
74c48bf5 LP |
445 | /* Some special handling for /dev/block/%u:%u, /dev/char/%u:%u, /run/systemd/inaccessible/chr and |
446 | * /run/systemd/inaccessible/blk paths. Instead of stat()ing these we parse out the major/minor directly. This | |
447 | * means clients can use these path without the device node actually around */ | |
846b3bd6 | 448 | r = device_path_parse_major_minor(node, &st.st_mode, &st.st_rdev); |
74c48bf5 LP |
449 | if (r < 0) { |
450 | if (r != -ENODEV) | |
451 | return log_warning_errno(r, "Couldn't parse major/minor from device path '%s': %m", node); | |
b200489b | 452 | |
74c48bf5 LP |
453 | if (stat(node, &st) < 0) |
454 | return log_warning_errno(errno, "Couldn't stat device %s: %m", node); | |
e7330dfe | 455 | |
74c48bf5 LP |
456 | if (!S_ISCHR(st.st_mode) && !S_ISBLK(st.st_mode)) { |
457 | log_warning("%s is not a device.", node); | |
458 | return -ENODEV; | |
459 | } | |
4ad49000 LP |
460 | } |
461 | ||
084c7007 RG |
462 | if (cg_all_unified() > 0) { |
463 | if (!prog) | |
464 | return 0; | |
4ad49000 | 465 | |
b9839ac9 LP |
466 | return cgroup_bpf_whitelist_device(prog, S_ISCHR(st.st_mode) ? BPF_DEVCG_DEV_CHAR : BPF_DEVCG_DEV_BLOCK, |
467 | major(st.st_rdev), minor(st.st_rdev), acc); | |
468 | ||
084c7007 RG |
469 | } else { |
470 | char buf[2+DECIMAL_STR_MAX(dev_t)*2+2+4]; | |
471 | ||
472 | sprintf(buf, | |
473 | "%c %u:%u %s", | |
474 | S_ISCHR(st.st_mode) ? 'c' : 'b', | |
475 | major(st.st_rdev), minor(st.st_rdev), | |
476 | acc); | |
477 | ||
8c838407 LP |
478 | /* Changing the devices list of a populated cgroup might result in EINVAL, hence ignore EINVAL here. */ |
479 | ||
084c7007 RG |
480 | r = cg_set_attribute("devices", path, "devices.allow", buf); |
481 | if (r < 0) | |
2c74e12b | 482 | return log_full_errno(IN_SET(r, -ENOENT, -EROFS, -EINVAL, -EACCES, -EPERM) ? LOG_DEBUG : LOG_WARNING, |
b9839ac9 | 483 | r, "Failed to set devices.allow on %s: %m", path); |
4ad49000 | 484 | |
b9839ac9 LP |
485 | return 0; |
486 | } | |
8e274523 LP |
487 | } |
488 | ||
084c7007 | 489 | static int whitelist_major(BPFProgram *prog, const char *path, const char *name, char type, const char *acc) { |
90060676 | 490 | _cleanup_fclose_ FILE *f = NULL; |
8e8b5d2e | 491 | char buf[2+DECIMAL_STR_MAX(unsigned)+3+4]; |
90060676 | 492 | bool good = false; |
8e8b5d2e | 493 | unsigned maj; |
90060676 LP |
494 | int r; |
495 | ||
496 | assert(path); | |
497 | assert(acc); | |
4c701096 | 498 | assert(IN_SET(type, 'b', 'c')); |
90060676 | 499 | |
8e8b5d2e LP |
500 | if (streq(name, "*")) { |
501 | /* If the name is a wildcard, then apply this list to all devices of this type */ | |
502 | ||
503 | if (cg_all_unified() > 0) { | |
504 | if (!prog) | |
505 | return 0; | |
506 | ||
507 | (void) cgroup_bpf_whitelist_class(prog, type == 'c' ? BPF_DEVCG_DEV_CHAR : BPF_DEVCG_DEV_BLOCK, acc); | |
508 | } else { | |
509 | xsprintf(buf, "%c *:* %s", type, acc); | |
510 | ||
511 | r = cg_set_attribute("devices", path, "devices.allow", buf); | |
512 | if (r < 0) | |
513 | log_full_errno(IN_SET(r, -ENOENT, -EROFS, -EINVAL, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, | |
514 | "Failed to set devices.allow on %s: %m", path); | |
515 | return 0; | |
516 | } | |
517 | } | |
518 | ||
519 | if (safe_atou(name, &maj) >= 0 && DEVICE_MAJOR_VALID(maj)) { | |
520 | /* The name is numeric and suitable as major. In that case, let's take is major, and create the entry | |
521 | * directly */ | |
522 | ||
523 | if (cg_all_unified() > 0) { | |
524 | if (!prog) | |
525 | return 0; | |
526 | ||
527 | (void) cgroup_bpf_whitelist_major(prog, | |
528 | type == 'c' ? BPF_DEVCG_DEV_CHAR : BPF_DEVCG_DEV_BLOCK, | |
529 | maj, acc); | |
530 | } else { | |
531 | xsprintf(buf, "%c %u:* %s", type, maj, acc); | |
532 | ||
533 | r = cg_set_attribute("devices", path, "devices.allow", buf); | |
534 | if (r < 0) | |
535 | log_full_errno(IN_SET(r, -ENOENT, -EROFS, -EINVAL, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, | |
536 | "Failed to set devices.allow on %s: %m", path); | |
537 | } | |
538 | ||
539 | return 0; | |
540 | } | |
541 | ||
90060676 | 542 | f = fopen("/proc/devices", "re"); |
4a62c710 MS |
543 | if (!f) |
544 | return log_warning_errno(errno, "Cannot open /proc/devices to resolve %s (%c): %m", name, type); | |
90060676 | 545 | |
c66e60a8 LP |
546 | for (;;) { |
547 | _cleanup_free_ char *line = NULL; | |
8e8b5d2e | 548 | char *w, *p; |
90060676 | 549 | |
c66e60a8 LP |
550 | r = read_line(f, LONG_LINE_MAX, &line); |
551 | if (r < 0) | |
552 | return log_warning_errno(r, "Failed to read /proc/devices: %m"); | |
553 | if (r == 0) | |
554 | break; | |
90060676 LP |
555 | |
556 | if (type == 'c' && streq(line, "Character devices:")) { | |
557 | good = true; | |
558 | continue; | |
559 | } | |
560 | ||
561 | if (type == 'b' && streq(line, "Block devices:")) { | |
562 | good = true; | |
563 | continue; | |
564 | } | |
565 | ||
566 | if (isempty(line)) { | |
567 | good = false; | |
568 | continue; | |
569 | } | |
570 | ||
571 | if (!good) | |
572 | continue; | |
573 | ||
574 | p = strstrip(line); | |
575 | ||
576 | w = strpbrk(p, WHITESPACE); | |
577 | if (!w) | |
578 | continue; | |
579 | *w = 0; | |
580 | ||
581 | r = safe_atou(p, &maj); | |
582 | if (r < 0) | |
583 | continue; | |
584 | if (maj <= 0) | |
585 | continue; | |
586 | ||
587 | w++; | |
588 | w += strspn(w, WHITESPACE); | |
e41969e3 LP |
589 | |
590 | if (fnmatch(name, w, 0) != 0) | |
90060676 LP |
591 | continue; |
592 | ||
084c7007 RG |
593 | if (cg_all_unified() > 0) { |
594 | if (!prog) | |
595 | continue; | |
90060676 | 596 | |
913c898c LP |
597 | (void) cgroup_bpf_whitelist_major(prog, |
598 | type == 'c' ? BPF_DEVCG_DEV_CHAR : BPF_DEVCG_DEV_BLOCK, | |
599 | maj, acc); | |
084c7007 | 600 | } else { |
084c7007 RG |
601 | sprintf(buf, |
602 | "%c %u:* %s", | |
603 | type, | |
604 | maj, | |
605 | acc); | |
606 | ||
8c838407 LP |
607 | /* Changing the devices list of a populated cgroup might result in EINVAL, hence ignore EINVAL |
608 | * here. */ | |
609 | ||
084c7007 RG |
610 | r = cg_set_attribute("devices", path, "devices.allow", buf); |
611 | if (r < 0) | |
2c74e12b | 612 | log_full_errno(IN_SET(r, -ENOENT, -EROFS, -EINVAL, -EACCES, -EPERM) ? LOG_DEBUG : LOG_WARNING, |
084c7007 RG |
613 | r, "Failed to set devices.allow on %s: %m", path); |
614 | } | |
90060676 LP |
615 | } |
616 | ||
617 | return 0; | |
90060676 LP |
618 | } |
619 | ||
66ebf6c0 TH |
620 | static bool cgroup_context_has_cpu_weight(CGroupContext *c) { |
621 | return c->cpu_weight != CGROUP_WEIGHT_INVALID || | |
622 | c->startup_cpu_weight != CGROUP_WEIGHT_INVALID; | |
623 | } | |
624 | ||
625 | static bool cgroup_context_has_cpu_shares(CGroupContext *c) { | |
626 | return c->cpu_shares != CGROUP_CPU_SHARES_INVALID || | |
627 | c->startup_cpu_shares != CGROUP_CPU_SHARES_INVALID; | |
628 | } | |
629 | ||
630 | static uint64_t cgroup_context_cpu_weight(CGroupContext *c, ManagerState state) { | |
631 | if (IN_SET(state, MANAGER_STARTING, MANAGER_INITIALIZING) && | |
632 | c->startup_cpu_weight != CGROUP_WEIGHT_INVALID) | |
633 | return c->startup_cpu_weight; | |
634 | else if (c->cpu_weight != CGROUP_WEIGHT_INVALID) | |
635 | return c->cpu_weight; | |
636 | else | |
637 | return CGROUP_WEIGHT_DEFAULT; | |
638 | } | |
639 | ||
640 | static uint64_t cgroup_context_cpu_shares(CGroupContext *c, ManagerState state) { | |
641 | if (IN_SET(state, MANAGER_STARTING, MANAGER_INITIALIZING) && | |
642 | c->startup_cpu_shares != CGROUP_CPU_SHARES_INVALID) | |
643 | return c->startup_cpu_shares; | |
644 | else if (c->cpu_shares != CGROUP_CPU_SHARES_INVALID) | |
645 | return c->cpu_shares; | |
646 | else | |
647 | return CGROUP_CPU_SHARES_DEFAULT; | |
648 | } | |
649 | ||
52fecf20 LP |
650 | static void cgroup_apply_unified_cpu_weight(Unit *u, uint64_t weight) { |
651 | char buf[DECIMAL_STR_MAX(uint64_t) + 2]; | |
66ebf6c0 TH |
652 | |
653 | xsprintf(buf, "%" PRIu64 "\n", weight); | |
293d32df | 654 | (void) set_attribute_and_warn(u, "cpu", "cpu.weight", buf); |
52fecf20 LP |
655 | } |
656 | ||
657 | static void cgroup_apply_unified_cpu_quota(Unit *u, usec_t quota) { | |
658 | char buf[(DECIMAL_STR_MAX(usec_t) + 1) * 2 + 1]; | |
66ebf6c0 TH |
659 | |
660 | if (quota != USEC_INFINITY) | |
661 | xsprintf(buf, USEC_FMT " " USEC_FMT "\n", | |
662 | quota * CGROUP_CPU_QUOTA_PERIOD_USEC / USEC_PER_SEC, CGROUP_CPU_QUOTA_PERIOD_USEC); | |
663 | else | |
664 | xsprintf(buf, "max " USEC_FMT "\n", CGROUP_CPU_QUOTA_PERIOD_USEC); | |
293d32df | 665 | (void) set_attribute_and_warn(u, "cpu", "cpu.max", buf); |
66ebf6c0 TH |
666 | } |
667 | ||
52fecf20 LP |
668 | static void cgroup_apply_legacy_cpu_shares(Unit *u, uint64_t shares) { |
669 | char buf[DECIMAL_STR_MAX(uint64_t) + 2]; | |
66ebf6c0 TH |
670 | |
671 | xsprintf(buf, "%" PRIu64 "\n", shares); | |
293d32df | 672 | (void) set_attribute_and_warn(u, "cpu", "cpu.shares", buf); |
52fecf20 LP |
673 | } |
674 | ||
675 | static void cgroup_apply_legacy_cpu_quota(Unit *u, usec_t quota) { | |
676 | char buf[DECIMAL_STR_MAX(usec_t) + 2]; | |
66ebf6c0 TH |
677 | |
678 | xsprintf(buf, USEC_FMT "\n", CGROUP_CPU_QUOTA_PERIOD_USEC); | |
293d32df | 679 | (void) set_attribute_and_warn(u, "cpu", "cpu.cfs_period_us", buf); |
66ebf6c0 TH |
680 | |
681 | if (quota != USEC_INFINITY) { | |
682 | xsprintf(buf, USEC_FMT "\n", quota * CGROUP_CPU_QUOTA_PERIOD_USEC / USEC_PER_SEC); | |
293d32df | 683 | (void) set_attribute_and_warn(u, "cpu", "cpu.cfs_quota_us", buf); |
66ebf6c0 | 684 | } else |
589a5f7a | 685 | (void) set_attribute_and_warn(u, "cpu", "cpu.cfs_quota_us", "-1\n"); |
66ebf6c0 TH |
686 | } |
687 | ||
688 | static uint64_t cgroup_cpu_shares_to_weight(uint64_t shares) { | |
689 | return CLAMP(shares * CGROUP_WEIGHT_DEFAULT / CGROUP_CPU_SHARES_DEFAULT, | |
690 | CGROUP_WEIGHT_MIN, CGROUP_WEIGHT_MAX); | |
691 | } | |
692 | ||
693 | static uint64_t cgroup_cpu_weight_to_shares(uint64_t weight) { | |
694 | return CLAMP(weight * CGROUP_CPU_SHARES_DEFAULT / CGROUP_WEIGHT_DEFAULT, | |
695 | CGROUP_CPU_SHARES_MIN, CGROUP_CPU_SHARES_MAX); | |
696 | } | |
697 | ||
508c45da | 698 | static bool cgroup_context_has_io_config(CGroupContext *c) { |
538b4852 TH |
699 | return c->io_accounting || |
700 | c->io_weight != CGROUP_WEIGHT_INVALID || | |
701 | c->startup_io_weight != CGROUP_WEIGHT_INVALID || | |
702 | c->io_device_weights || | |
6ae4283c | 703 | c->io_device_latencies || |
538b4852 TH |
704 | c->io_device_limits; |
705 | } | |
706 | ||
508c45da | 707 | static bool cgroup_context_has_blockio_config(CGroupContext *c) { |
538b4852 TH |
708 | return c->blockio_accounting || |
709 | c->blockio_weight != CGROUP_BLKIO_WEIGHT_INVALID || | |
710 | c->startup_blockio_weight != CGROUP_BLKIO_WEIGHT_INVALID || | |
711 | c->blockio_device_weights || | |
712 | c->blockio_device_bandwidths; | |
713 | } | |
714 | ||
508c45da | 715 | static uint64_t cgroup_context_io_weight(CGroupContext *c, ManagerState state) { |
64faf04c TH |
716 | if (IN_SET(state, MANAGER_STARTING, MANAGER_INITIALIZING) && |
717 | c->startup_io_weight != CGROUP_WEIGHT_INVALID) | |
718 | return c->startup_io_weight; | |
719 | else if (c->io_weight != CGROUP_WEIGHT_INVALID) | |
720 | return c->io_weight; | |
721 | else | |
722 | return CGROUP_WEIGHT_DEFAULT; | |
723 | } | |
724 | ||
508c45da | 725 | static uint64_t cgroup_context_blkio_weight(CGroupContext *c, ManagerState state) { |
64faf04c TH |
726 | if (IN_SET(state, MANAGER_STARTING, MANAGER_INITIALIZING) && |
727 | c->startup_blockio_weight != CGROUP_BLKIO_WEIGHT_INVALID) | |
728 | return c->startup_blockio_weight; | |
729 | else if (c->blockio_weight != CGROUP_BLKIO_WEIGHT_INVALID) | |
730 | return c->blockio_weight; | |
731 | else | |
732 | return CGROUP_BLKIO_WEIGHT_DEFAULT; | |
733 | } | |
734 | ||
508c45da | 735 | static uint64_t cgroup_weight_blkio_to_io(uint64_t blkio_weight) { |
538b4852 TH |
736 | return CLAMP(blkio_weight * CGROUP_WEIGHT_DEFAULT / CGROUP_BLKIO_WEIGHT_DEFAULT, |
737 | CGROUP_WEIGHT_MIN, CGROUP_WEIGHT_MAX); | |
738 | } | |
739 | ||
508c45da | 740 | static uint64_t cgroup_weight_io_to_blkio(uint64_t io_weight) { |
538b4852 TH |
741 | return CLAMP(io_weight * CGROUP_BLKIO_WEIGHT_DEFAULT / CGROUP_WEIGHT_DEFAULT, |
742 | CGROUP_BLKIO_WEIGHT_MIN, CGROUP_BLKIO_WEIGHT_MAX); | |
743 | } | |
744 | ||
f29ff115 | 745 | static void cgroup_apply_io_device_weight(Unit *u, const char *dev_path, uint64_t io_weight) { |
64faf04c TH |
746 | char buf[DECIMAL_STR_MAX(dev_t)*2+2+DECIMAL_STR_MAX(uint64_t)+1]; |
747 | dev_t dev; | |
748 | int r; | |
749 | ||
750 | r = lookup_block_device(dev_path, &dev); | |
751 | if (r < 0) | |
752 | return; | |
753 | ||
754 | xsprintf(buf, "%u:%u %" PRIu64 "\n", major(dev), minor(dev), io_weight); | |
293d32df | 755 | (void) set_attribute_and_warn(u, "io", "io.weight", buf); |
64faf04c TH |
756 | } |
757 | ||
f29ff115 | 758 | static void cgroup_apply_blkio_device_weight(Unit *u, const char *dev_path, uint64_t blkio_weight) { |
64faf04c TH |
759 | char buf[DECIMAL_STR_MAX(dev_t)*2+2+DECIMAL_STR_MAX(uint64_t)+1]; |
760 | dev_t dev; | |
761 | int r; | |
762 | ||
763 | r = lookup_block_device(dev_path, &dev); | |
764 | if (r < 0) | |
765 | return; | |
766 | ||
767 | xsprintf(buf, "%u:%u %" PRIu64 "\n", major(dev), minor(dev), blkio_weight); | |
293d32df | 768 | (void) set_attribute_and_warn(u, "blkio", "blkio.weight_device", buf); |
64faf04c TH |
769 | } |
770 | ||
6ae4283c TH |
771 | static void cgroup_apply_io_device_latency(Unit *u, const char *dev_path, usec_t target) { |
772 | char buf[DECIMAL_STR_MAX(dev_t)*2+2+7+DECIMAL_STR_MAX(uint64_t)+1]; | |
773 | dev_t dev; | |
774 | int r; | |
775 | ||
776 | r = lookup_block_device(dev_path, &dev); | |
777 | if (r < 0) | |
778 | return; | |
779 | ||
780 | if (target != USEC_INFINITY) | |
781 | xsprintf(buf, "%u:%u target=%" PRIu64 "\n", major(dev), minor(dev), target); | |
782 | else | |
783 | xsprintf(buf, "%u:%u target=max\n", major(dev), minor(dev)); | |
784 | ||
293d32df | 785 | (void) set_attribute_and_warn(u, "io", "io.latency", buf); |
6ae4283c TH |
786 | } |
787 | ||
17ae2780 | 788 | static void cgroup_apply_io_device_limit(Unit *u, const char *dev_path, uint64_t *limits) { |
64faf04c TH |
789 | char limit_bufs[_CGROUP_IO_LIMIT_TYPE_MAX][DECIMAL_STR_MAX(uint64_t)]; |
790 | char buf[DECIMAL_STR_MAX(dev_t)*2+2+(6+DECIMAL_STR_MAX(uint64_t)+1)*4]; | |
791 | CGroupIOLimitType type; | |
792 | dev_t dev; | |
64faf04c TH |
793 | int r; |
794 | ||
795 | r = lookup_block_device(dev_path, &dev); | |
796 | if (r < 0) | |
17ae2780 | 797 | return; |
64faf04c | 798 | |
17ae2780 LP |
799 | for (type = 0; type < _CGROUP_IO_LIMIT_TYPE_MAX; type++) |
800 | if (limits[type] != cgroup_io_limit_defaults[type]) | |
64faf04c | 801 | xsprintf(limit_bufs[type], "%" PRIu64, limits[type]); |
17ae2780 | 802 | else |
64faf04c | 803 | xsprintf(limit_bufs[type], "%s", limits[type] == CGROUP_LIMIT_MAX ? "max" : "0"); |
64faf04c TH |
804 | |
805 | xsprintf(buf, "%u:%u rbps=%s wbps=%s riops=%s wiops=%s\n", major(dev), minor(dev), | |
806 | limit_bufs[CGROUP_IO_RBPS_MAX], limit_bufs[CGROUP_IO_WBPS_MAX], | |
807 | limit_bufs[CGROUP_IO_RIOPS_MAX], limit_bufs[CGROUP_IO_WIOPS_MAX]); | |
293d32df | 808 | (void) set_attribute_and_warn(u, "io", "io.max", buf); |
64faf04c TH |
809 | } |
810 | ||
17ae2780 | 811 | static void cgroup_apply_blkio_device_limit(Unit *u, const char *dev_path, uint64_t rbps, uint64_t wbps) { |
64faf04c TH |
812 | char buf[DECIMAL_STR_MAX(dev_t)*2+2+DECIMAL_STR_MAX(uint64_t)+1]; |
813 | dev_t dev; | |
64faf04c TH |
814 | int r; |
815 | ||
816 | r = lookup_block_device(dev_path, &dev); | |
817 | if (r < 0) | |
17ae2780 | 818 | return; |
64faf04c | 819 | |
64faf04c | 820 | sprintf(buf, "%u:%u %" PRIu64 "\n", major(dev), minor(dev), rbps); |
293d32df | 821 | (void) set_attribute_and_warn(u, "blkio", "blkio.throttle.read_bps_device", buf); |
64faf04c | 822 | |
64faf04c | 823 | sprintf(buf, "%u:%u %" PRIu64 "\n", major(dev), minor(dev), wbps); |
293d32df | 824 | (void) set_attribute_and_warn(u, "blkio", "blkio.throttle.write_bps_device", buf); |
64faf04c TH |
825 | } |
826 | ||
da4d897e | 827 | static bool cgroup_context_has_unified_memory_config(CGroupContext *c) { |
48422635 | 828 | 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 |
829 | } |
830 | ||
f29ff115 | 831 | static void cgroup_apply_unified_memory_limit(Unit *u, const char *file, uint64_t v) { |
589a5f7a | 832 | char buf[DECIMAL_STR_MAX(uint64_t) + 1] = "max\n"; |
da4d897e TH |
833 | |
834 | if (v != CGROUP_LIMIT_MAX) | |
835 | xsprintf(buf, "%" PRIu64 "\n", v); | |
836 | ||
293d32df | 837 | (void) set_attribute_and_warn(u, "memory", file, buf); |
da4d897e TH |
838 | } |
839 | ||
0f2d84d2 | 840 | static void cgroup_apply_firewall(Unit *u) { |
0f2d84d2 LP |
841 | assert(u); |
842 | ||
acf7f253 | 843 | /* Best-effort: let's apply IP firewalling and/or accounting if that's enabled */ |
906c06f6 | 844 | |
acf7f253 | 845 | if (bpf_firewall_compile(u) < 0) |
906c06f6 DM |
846 | return; |
847 | ||
848 | (void) bpf_firewall_install(u); | |
906c06f6 DM |
849 | } |
850 | ||
851 | static void cgroup_context_apply( | |
852 | Unit *u, | |
853 | CGroupMask apply_mask, | |
906c06f6 DM |
854 | ManagerState state) { |
855 | ||
f29ff115 TH |
856 | const char *path; |
857 | CGroupContext *c; | |
52fecf20 | 858 | bool is_host_root, is_local_root; |
4ad49000 LP |
859 | int r; |
860 | ||
f29ff115 TH |
861 | assert(u); |
862 | ||
906c06f6 | 863 | /* Nothing to do? Exit early! */ |
17f14955 | 864 | if (apply_mask == 0) |
4ad49000 | 865 | return; |
8e274523 | 866 | |
52fecf20 LP |
867 | /* Some cgroup attributes are not supported on the host root cgroup, hence silently ignore them here. And other |
868 | * attributes should only be managed for cgroups further down the tree. */ | |
869 | is_local_root = unit_has_name(u, SPECIAL_ROOT_SLICE); | |
870 | is_host_root = unit_has_host_root_cgroup(u); | |
f3725e64 LP |
871 | |
872 | assert_se(c = unit_get_cgroup_context(u)); | |
873 | assert_se(path = u->cgroup_path); | |
874 | ||
52fecf20 | 875 | if (is_local_root) /* Make sure we don't try to display messages with an empty path. */ |
6da13913 | 876 | path = "/"; |
01efdf13 | 877 | |
714e2e1d LP |
878 | /* We generally ignore errors caused by read-only mounted |
879 | * cgroup trees (assuming we are running in a container then), | |
880 | * and missing cgroups, i.e. EROFS and ENOENT. */ | |
881 | ||
52fecf20 | 882 | if (apply_mask & CGROUP_MASK_CPU) { |
906c06f6 DM |
883 | bool has_weight, has_shares; |
884 | ||
885 | has_weight = cgroup_context_has_cpu_weight(c); | |
886 | has_shares = cgroup_context_has_cpu_shares(c); | |
8e274523 | 887 | |
b4cccbc1 | 888 | if (cg_all_unified() > 0) { |
b2f8b02e | 889 | |
52fecf20 LP |
890 | /* In fully unified mode these attributes don't exist on the host cgroup root, and inside of |
891 | * containers we want to leave control of these to the container manager (and if delegation is | |
892 | * used we couldn't even write to them if we wanted to). */ | |
893 | if (!is_local_root) { | |
894 | uint64_t weight; | |
66ebf6c0 | 895 | |
52fecf20 LP |
896 | if (has_weight) |
897 | weight = cgroup_context_cpu_weight(c, state); | |
898 | else if (has_shares) { | |
899 | uint64_t shares; | |
66ebf6c0 | 900 | |
52fecf20 LP |
901 | shares = cgroup_context_cpu_shares(c, state); |
902 | weight = cgroup_cpu_shares_to_weight(shares); | |
66ebf6c0 | 903 | |
52fecf20 LP |
904 | log_cgroup_compat(u, "Applying [Startup]CPUShares %" PRIu64 " as [Startup]CPUWeight %" PRIu64 " on %s", |
905 | shares, weight, path); | |
906 | } else | |
907 | weight = CGROUP_WEIGHT_DEFAULT; | |
66ebf6c0 | 908 | |
52fecf20 LP |
909 | cgroup_apply_unified_cpu_weight(u, weight); |
910 | cgroup_apply_unified_cpu_quota(u, c->cpu_quota_per_sec_usec); | |
911 | } | |
66ebf6c0 | 912 | |
52fecf20 LP |
913 | } else { |
914 | /* Setting the weight makes very little sense on the host root cgroup, as there are no other | |
915 | * cgroups at this level. And for containers we want to leave management of this to the | |
916 | * container manager */ | |
917 | if (!is_local_root) { | |
918 | uint64_t shares; | |
919 | ||
920 | if (has_weight) { | |
921 | uint64_t weight; | |
922 | ||
923 | weight = cgroup_context_cpu_weight(c, state); | |
924 | shares = cgroup_cpu_weight_to_shares(weight); | |
925 | ||
926 | log_cgroup_compat(u, "Applying [Startup]CPUWeight %" PRIu64 " as [Startup]CPUShares %" PRIu64 " on %s", | |
927 | weight, shares, path); | |
928 | } else if (has_shares) | |
929 | shares = cgroup_context_cpu_shares(c, state); | |
930 | else | |
931 | shares = CGROUP_CPU_SHARES_DEFAULT; | |
932 | ||
933 | cgroup_apply_legacy_cpu_shares(u, shares); | |
934 | } | |
66ebf6c0 | 935 | |
52fecf20 LP |
936 | /* The "cpu" quota attribute is available on the host root, hence manage it there. But in |
937 | * containers let's leave this to the container manager. */ | |
938 | if (is_host_root || !is_local_root) | |
939 | cgroup_apply_legacy_cpu_quota(u, c->cpu_quota_per_sec_usec); | |
66ebf6c0 | 940 | } |
4ad49000 LP |
941 | } |
942 | ||
52fecf20 LP |
943 | /* The 'io' controller attributes are not exported on the host's root cgroup (being a pure cgroupsv2 |
944 | * controller), and in case of containers we want to leave control of these attributes to the container manager | |
945 | * (and we couldn't access that stuff anyway, even if we tried if proper delegation is used). */ | |
946 | if ((apply_mask & CGROUP_MASK_IO) && !is_local_root) { | |
947 | char buf[8+DECIMAL_STR_MAX(uint64_t)+1]; | |
948 | bool has_io, has_blockio; | |
949 | uint64_t weight; | |
13c31542 | 950 | |
52fecf20 LP |
951 | has_io = cgroup_context_has_io_config(c); |
952 | has_blockio = cgroup_context_has_blockio_config(c); | |
13c31542 | 953 | |
52fecf20 LP |
954 | if (has_io) |
955 | weight = cgroup_context_io_weight(c, state); | |
956 | else if (has_blockio) { | |
957 | uint64_t blkio_weight; | |
128fadc9 | 958 | |
52fecf20 LP |
959 | blkio_weight = cgroup_context_blkio_weight(c, state); |
960 | weight = cgroup_weight_blkio_to_io(blkio_weight); | |
128fadc9 | 961 | |
52fecf20 LP |
962 | log_cgroup_compat(u, "Applying [Startup]BlockIOWeight %" PRIu64 " as [Startup]IOWeight %" PRIu64, |
963 | blkio_weight, weight); | |
964 | } else | |
965 | weight = CGROUP_WEIGHT_DEFAULT; | |
13c31542 | 966 | |
52fecf20 LP |
967 | xsprintf(buf, "default %" PRIu64 "\n", weight); |
968 | (void) set_attribute_and_warn(u, "io", "io.weight", buf); | |
538b4852 | 969 | |
52fecf20 LP |
970 | if (has_io) { |
971 | CGroupIODeviceLatency *latency; | |
972 | CGroupIODeviceLimit *limit; | |
973 | CGroupIODeviceWeight *w; | |
128fadc9 | 974 | |
52fecf20 LP |
975 | LIST_FOREACH(device_weights, w, c->io_device_weights) |
976 | cgroup_apply_io_device_weight(u, w->path, w->weight); | |
128fadc9 | 977 | |
52fecf20 LP |
978 | LIST_FOREACH(device_limits, limit, c->io_device_limits) |
979 | cgroup_apply_io_device_limit(u, limit->path, limit->limits); | |
6ae4283c | 980 | |
52fecf20 LP |
981 | LIST_FOREACH(device_latencies, latency, c->io_device_latencies) |
982 | cgroup_apply_io_device_latency(u, latency->path, latency->target_usec); | |
6ae4283c | 983 | |
52fecf20 LP |
984 | } else if (has_blockio) { |
985 | CGroupBlockIODeviceWeight *w; | |
986 | CGroupBlockIODeviceBandwidth *b; | |
13c31542 | 987 | |
52fecf20 LP |
988 | LIST_FOREACH(device_weights, w, c->blockio_device_weights) { |
989 | weight = cgroup_weight_blkio_to_io(w->weight); | |
17ae2780 | 990 | |
52fecf20 LP |
991 | log_cgroup_compat(u, "Applying BlockIODeviceWeight %" PRIu64 " as IODeviceWeight %" PRIu64 " for %s", |
992 | w->weight, weight, w->path); | |
538b4852 | 993 | |
52fecf20 LP |
994 | cgroup_apply_io_device_weight(u, w->path, weight); |
995 | } | |
538b4852 | 996 | |
17ae2780 | 997 | LIST_FOREACH(device_bandwidths, b, c->blockio_device_bandwidths) { |
538b4852 TH |
998 | uint64_t limits[_CGROUP_IO_LIMIT_TYPE_MAX]; |
999 | CGroupIOLimitType type; | |
1000 | ||
1001 | for (type = 0; type < _CGROUP_IO_LIMIT_TYPE_MAX; type++) | |
1002 | limits[type] = cgroup_io_limit_defaults[type]; | |
1003 | ||
1004 | limits[CGROUP_IO_RBPS_MAX] = b->rbps; | |
1005 | limits[CGROUP_IO_WBPS_MAX] = b->wbps; | |
1006 | ||
128fadc9 TH |
1007 | log_cgroup_compat(u, "Applying BlockIO{Read|Write}Bandwidth %" PRIu64 " %" PRIu64 " as IO{Read|Write}BandwidthMax for %s", |
1008 | b->rbps, b->wbps, b->path); | |
1009 | ||
17ae2780 | 1010 | cgroup_apply_io_device_limit(u, b->path, limits); |
538b4852 | 1011 | } |
13c31542 TH |
1012 | } |
1013 | } | |
1014 | ||
906c06f6 | 1015 | if (apply_mask & CGROUP_MASK_BLKIO) { |
52fecf20 | 1016 | bool has_io, has_blockio; |
4ad49000 | 1017 | |
52fecf20 LP |
1018 | has_io = cgroup_context_has_io_config(c); |
1019 | has_blockio = cgroup_context_has_blockio_config(c); | |
1020 | ||
1021 | /* Applying a 'weight' never makes sense for the host root cgroup, and for containers this should be | |
1022 | * left to our container manager, too. */ | |
1023 | if (!is_local_root) { | |
64faf04c TH |
1024 | char buf[DECIMAL_STR_MAX(uint64_t)+1]; |
1025 | uint64_t weight; | |
64faf04c | 1026 | |
7d862ab8 | 1027 | if (has_io) { |
52fecf20 | 1028 | uint64_t io_weight; |
128fadc9 | 1029 | |
52fecf20 | 1030 | io_weight = cgroup_context_io_weight(c, state); |
538b4852 | 1031 | weight = cgroup_weight_io_to_blkio(cgroup_context_io_weight(c, state)); |
128fadc9 TH |
1032 | |
1033 | log_cgroup_compat(u, "Applying [Startup]IOWeight %" PRIu64 " as [Startup]BlockIOWeight %" PRIu64, | |
1034 | io_weight, weight); | |
7d862ab8 TH |
1035 | } else if (has_blockio) |
1036 | weight = cgroup_context_blkio_weight(c, state); | |
1037 | else | |
538b4852 | 1038 | weight = CGROUP_BLKIO_WEIGHT_DEFAULT; |
64faf04c TH |
1039 | |
1040 | xsprintf(buf, "%" PRIu64 "\n", weight); | |
293d32df | 1041 | (void) set_attribute_and_warn(u, "blkio", "blkio.weight", buf); |
4ad49000 | 1042 | |
7d862ab8 | 1043 | if (has_io) { |
538b4852 TH |
1044 | CGroupIODeviceWeight *w; |
1045 | ||
128fadc9 TH |
1046 | LIST_FOREACH(device_weights, w, c->io_device_weights) { |
1047 | weight = cgroup_weight_io_to_blkio(w->weight); | |
1048 | ||
1049 | log_cgroup_compat(u, "Applying IODeviceWeight %" PRIu64 " as BlockIODeviceWeight %" PRIu64 " for %s", | |
1050 | w->weight, weight, w->path); | |
1051 | ||
1052 | cgroup_apply_blkio_device_weight(u, w->path, weight); | |
1053 | } | |
7d862ab8 TH |
1054 | } else if (has_blockio) { |
1055 | CGroupBlockIODeviceWeight *w; | |
1056 | ||
7d862ab8 TH |
1057 | LIST_FOREACH(device_weights, w, c->blockio_device_weights) |
1058 | cgroup_apply_blkio_device_weight(u, w->path, w->weight); | |
538b4852 | 1059 | } |
4ad49000 LP |
1060 | } |
1061 | ||
52fecf20 LP |
1062 | /* The bandwith limits are something that make sense to be applied to the host's root but not container |
1063 | * roots, as there we want the container manager to handle it */ | |
1064 | if (is_host_root || !is_local_root) { | |
1065 | if (has_io) { | |
1066 | CGroupIODeviceLimit *l; | |
538b4852 | 1067 | |
52fecf20 LP |
1068 | LIST_FOREACH(device_limits, l, c->io_device_limits) { |
1069 | log_cgroup_compat(u, "Applying IO{Read|Write}Bandwidth %" PRIu64 " %" PRIu64 " as BlockIO{Read|Write}BandwidthMax for %s", | |
1070 | l->limits[CGROUP_IO_RBPS_MAX], l->limits[CGROUP_IO_WBPS_MAX], l->path); | |
128fadc9 | 1071 | |
52fecf20 LP |
1072 | cgroup_apply_blkio_device_limit(u, l->path, l->limits[CGROUP_IO_RBPS_MAX], l->limits[CGROUP_IO_WBPS_MAX]); |
1073 | } | |
1074 | } else if (has_blockio) { | |
1075 | CGroupBlockIODeviceBandwidth *b; | |
7d862ab8 | 1076 | |
52fecf20 LP |
1077 | LIST_FOREACH(device_bandwidths, b, c->blockio_device_bandwidths) |
1078 | cgroup_apply_blkio_device_limit(u, b->path, b->rbps, b->wbps); | |
1079 | } | |
d686d8a9 | 1080 | } |
8e274523 LP |
1081 | } |
1082 | ||
52fecf20 | 1083 | if (apply_mask & CGROUP_MASK_MEMORY) { |
efdb0237 | 1084 | |
52fecf20 LP |
1085 | if (cg_all_unified() > 0) { |
1086 | /* In unified mode 'memory' attributes do not exist on the root cgroup. And if we run in a | |
1087 | * container we want to leave control to the container manager (and if proper delegation is | |
1088 | * used we couldn't even write to this if we wanted to. */ | |
1089 | if (!is_local_root) { | |
1090 | uint64_t max, swap_max = CGROUP_LIMIT_MAX; | |
1091 | ||
1092 | if (cgroup_context_has_unified_memory_config(c)) { | |
1093 | max = c->memory_max; | |
1094 | swap_max = c->memory_swap_max; | |
1095 | } else { | |
1096 | max = c->memory_limit; | |
1097 | ||
1098 | if (max != CGROUP_LIMIT_MAX) | |
1099 | log_cgroup_compat(u, "Applying MemoryLimit=%" PRIu64 " as MemoryMax=", max); | |
1100 | } | |
efdb0237 | 1101 | |
52fecf20 LP |
1102 | cgroup_apply_unified_memory_limit(u, "memory.min", c->memory_min); |
1103 | cgroup_apply_unified_memory_limit(u, "memory.low", c->memory_low); | |
1104 | cgroup_apply_unified_memory_limit(u, "memory.high", c->memory_high); | |
1105 | cgroup_apply_unified_memory_limit(u, "memory.max", max); | |
1106 | cgroup_apply_unified_memory_limit(u, "memory.swap.max", swap_max); | |
128fadc9 | 1107 | } |
efdb0237 | 1108 | } else { |
da4d897e | 1109 | |
52fecf20 LP |
1110 | /* In legacy mode 'memory' exists on the host root, but in container mode we want to leave it |
1111 | * to the container manager around us */ | |
1112 | if (is_host_root || !is_local_root) { | |
1113 | char buf[DECIMAL_STR_MAX(uint64_t) + 1]; | |
1114 | uint64_t val; | |
128fadc9 | 1115 | |
52fecf20 LP |
1116 | if (cgroup_context_has_unified_memory_config(c)) { |
1117 | val = c->memory_max; | |
1118 | log_cgroup_compat(u, "Applying MemoryMax=%" PRIi64 " as MemoryLimit=", val); | |
1119 | } else | |
1120 | val = c->memory_limit; | |
1121 | ||
1122 | if (val == CGROUP_LIMIT_MAX) | |
1123 | strncpy(buf, "-1\n", sizeof(buf)); | |
1124 | else | |
1125 | xsprintf(buf, "%" PRIu64 "\n", val); | |
78a4ee59 | 1126 | |
52fecf20 LP |
1127 | (void) set_attribute_and_warn(u, "memory", "memory.limit_in_bytes", buf); |
1128 | } | |
da4d897e | 1129 | } |
4ad49000 | 1130 | } |
8e274523 | 1131 | |
52fecf20 LP |
1132 | /* On cgroupsv2 we can apply BPF everywhre. On cgroupsv1 we apply it everywhere except for the root of |
1133 | * containers, where we leave this to the manager */ | |
1134 | if ((apply_mask & (CGROUP_MASK_DEVICES | CGROUP_MASK_BPF_DEVICES)) && | |
1135 | (is_host_root || cg_all_unified() > 0 || !is_local_root)) { | |
084c7007 | 1136 | _cleanup_(bpf_program_unrefp) BPFProgram *prog = NULL; |
4ad49000 | 1137 | CGroupDeviceAllow *a; |
8e274523 | 1138 | |
084c7007 RG |
1139 | if (cg_all_unified() > 0) { |
1140 | r = cgroup_init_device_bpf(&prog, c->device_policy, c->device_allow); | |
1141 | if (r < 0) | |
1142 | log_unit_warning_errno(u, r, "Failed to initialize device control bpf program: %m"); | |
1143 | } else { | |
8c838407 | 1144 | /* Changing the devices list of a populated cgroup might result in EINVAL, hence ignore EINVAL |
084c7007 | 1145 | * here. */ |
714e2e1d | 1146 | |
084c7007 RG |
1147 | if (c->device_allow || c->device_policy != CGROUP_AUTO) |
1148 | r = cg_set_attribute("devices", path, "devices.deny", "a"); | |
1149 | else | |
1150 | r = cg_set_attribute("devices", path, "devices.allow", "a"); | |
1151 | if (r < 0) | |
2c74e12b LP |
1152 | log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EINVAL, -EACCES, -EPERM) ? LOG_DEBUG : LOG_WARNING, r, |
1153 | "Failed to reset devices.allow/devices.deny: %m"); | |
084c7007 | 1154 | } |
fb385181 | 1155 | |
4ad49000 LP |
1156 | if (c->device_policy == CGROUP_CLOSED || |
1157 | (c->device_policy == CGROUP_AUTO && c->device_allow)) { | |
1158 | static const char auto_devices[] = | |
7d711efb LP |
1159 | "/dev/null\0" "rwm\0" |
1160 | "/dev/zero\0" "rwm\0" | |
1161 | "/dev/full\0" "rwm\0" | |
1162 | "/dev/random\0" "rwm\0" | |
1163 | "/dev/urandom\0" "rwm\0" | |
1164 | "/dev/tty\0" "rwm\0" | |
5a7f87a9 | 1165 | "/dev/ptmx\0" "rwm\0" |
0d9e7991 | 1166 | /* Allow /run/systemd/inaccessible/{chr,blk} devices for mapping InaccessiblePaths */ |
74c48bf5 LP |
1167 | "/run/systemd/inaccessible/chr\0" "rwm\0" |
1168 | "/run/systemd/inaccessible/blk\0" "rwm\0"; | |
4ad49000 LP |
1169 | |
1170 | const char *x, *y; | |
1171 | ||
1172 | NULSTR_FOREACH_PAIR(x, y, auto_devices) | |
913c898c | 1173 | (void) whitelist_device(prog, path, x, y); |
7d711efb | 1174 | |
5a7f87a9 | 1175 | /* PTS (/dev/pts) devices may not be duplicated, but accessed */ |
913c898c | 1176 | (void) whitelist_major(prog, path, "pts", 'c', "rw"); |
4ad49000 LP |
1177 | } |
1178 | ||
1179 | LIST_FOREACH(device_allow, a, c->device_allow) { | |
fb4650aa | 1180 | char acc[4], *val; |
4ad49000 LP |
1181 | unsigned k = 0; |
1182 | ||
1183 | if (a->r) | |
1184 | acc[k++] = 'r'; | |
1185 | if (a->w) | |
1186 | acc[k++] = 'w'; | |
1187 | if (a->m) | |
1188 | acc[k++] = 'm'; | |
fb385181 | 1189 | |
4ad49000 LP |
1190 | if (k == 0) |
1191 | continue; | |
fb385181 | 1192 | |
4ad49000 | 1193 | acc[k++] = 0; |
90060676 | 1194 | |
27458ed6 | 1195 | if (path_startswith(a->path, "/dev/")) |
913c898c | 1196 | (void) whitelist_device(prog, path, a->path, acc); |
fb4650aa | 1197 | else if ((val = startswith(a->path, "block-"))) |
913c898c | 1198 | (void) whitelist_major(prog, path, val, 'b', acc); |
fb4650aa | 1199 | else if ((val = startswith(a->path, "char-"))) |
913c898c | 1200 | (void) whitelist_major(prog, path, val, 'c', acc); |
90060676 | 1201 | else |
8e8b5d2e | 1202 | log_unit_debug(u, "Ignoring device '%s' while writing cgroup attribute.", a->path); |
4ad49000 | 1203 | } |
084c7007 RG |
1204 | |
1205 | r = cgroup_apply_device_bpf(u, prog, c->device_policy, c->device_allow); | |
1206 | if (r < 0) { | |
1207 | static bool warned = false; | |
1208 | ||
1209 | log_full_errno(warned ? LOG_DEBUG : LOG_WARNING, r, | |
1210 | "Unit %s configures device ACL, but the local system doesn't seem to support the BPF-based device controller.\n" | |
1211 | "Proceeding WITHOUT applying ACL (all devices will be accessible)!\n" | |
1212 | "(This warning is only shown for the first loaded unit using device ACL.)", u->id); | |
1213 | ||
1214 | warned = true; | |
1215 | } | |
4ad49000 | 1216 | } |
03a7b521 | 1217 | |
00b5974f LP |
1218 | if (apply_mask & CGROUP_MASK_PIDS) { |
1219 | ||
52fecf20 | 1220 | if (is_host_root) { |
00b5974f LP |
1221 | /* So, the "pids" controller does not expose anything on the root cgroup, in order not to |
1222 | * replicate knobs exposed elsewhere needlessly. We abstract this away here however, and when | |
1223 | * the knobs of the root cgroup are modified propagate this to the relevant sysctls. There's a | |
1224 | * non-obvious asymmetry however: unlike the cgroup properties we don't really want to take | |
1225 | * exclusive ownership of the sysctls, but we still want to honour things if the user sets | |
1226 | * limits. Hence we employ sort of a one-way strategy: when the user sets a bounded limit | |
1227 | * through us it counts. When the user afterwards unsets it again (i.e. sets it to unbounded) | |
1228 | * it also counts. But if the user never set a limit through us (i.e. we are the default of | |
1229 | * "unbounded") we leave things unmodified. For this we manage a global boolean that we turn on | |
1230 | * the first time we set a limit. Note that this boolean is flushed out on manager reload, | |
1231 | * which is desirable so that there's an offical way to release control of the sysctl from | |
1232 | * systemd: set the limit to unbounded and reload. */ | |
1233 | ||
1234 | if (c->tasks_max != CGROUP_LIMIT_MAX) { | |
1235 | u->manager->sysctl_pid_max_changed = true; | |
1236 | r = procfs_tasks_set_limit(c->tasks_max); | |
1237 | } else if (u->manager->sysctl_pid_max_changed) | |
1238 | r = procfs_tasks_set_limit(TASKS_MAX); | |
1239 | else | |
1240 | r = 0; | |
03a7b521 | 1241 | |
00b5974f | 1242 | if (r < 0) |
39b9fefb | 1243 | log_unit_full(u, LOG_LEVEL_CGROUP_WRITE(r), r, |
00b5974f | 1244 | "Failed to write to tasks limit sysctls: %m"); |
52fecf20 | 1245 | } |
03a7b521 | 1246 | |
52fecf20 LP |
1247 | /* The attribute itself is not available on the host root cgroup, and in the container case we want to |
1248 | * leave it for the container manager. */ | |
1249 | if (!is_local_root) { | |
00b5974f LP |
1250 | if (c->tasks_max != CGROUP_LIMIT_MAX) { |
1251 | char buf[DECIMAL_STR_MAX(uint64_t) + 2]; | |
03a7b521 | 1252 | |
00b5974f | 1253 | sprintf(buf, "%" PRIu64 "\n", c->tasks_max); |
293d32df | 1254 | (void) set_attribute_and_warn(u, "pids", "pids.max", buf); |
00b5974f | 1255 | } else |
589a5f7a | 1256 | (void) set_attribute_and_warn(u, "pids", "pids.max", "max\n"); |
00b5974f | 1257 | } |
03a7b521 | 1258 | } |
906c06f6 | 1259 | |
17f14955 | 1260 | if (apply_mask & CGROUP_MASK_BPF_FIREWALL) |
0f2d84d2 | 1261 | cgroup_apply_firewall(u); |
fb385181 LP |
1262 | } |
1263 | ||
16492445 LP |
1264 | static bool unit_get_needs_bpf_firewall(Unit *u) { |
1265 | CGroupContext *c; | |
1266 | Unit *p; | |
1267 | assert(u); | |
1268 | ||
1269 | c = unit_get_cgroup_context(u); | |
1270 | if (!c) | |
1271 | return false; | |
1272 | ||
1273 | if (c->ip_accounting || | |
1274 | c->ip_address_allow || | |
1275 | c->ip_address_deny) | |
1276 | return true; | |
1277 | ||
1278 | /* If any parent slice has an IP access list defined, it applies too */ | |
1279 | for (p = UNIT_DEREF(u->slice); p; p = UNIT_DEREF(p->slice)) { | |
1280 | c = unit_get_cgroup_context(p); | |
1281 | if (!c) | |
1282 | return false; | |
1283 | ||
1284 | if (c->ip_address_allow || | |
1285 | c->ip_address_deny) | |
1286 | return true; | |
1287 | } | |
1288 | ||
1289 | return false; | |
1290 | } | |
1291 | ||
53aea74a | 1292 | static CGroupMask cgroup_context_get_mask(CGroupContext *c) { |
efdb0237 | 1293 | CGroupMask mask = 0; |
8e274523 | 1294 | |
fae9bc29 | 1295 | /* Figure out which controllers we need, based on the cgroup context object */ |
8e274523 | 1296 | |
fae9bc29 | 1297 | if (c->cpu_accounting) |
f98c2585 | 1298 | mask |= get_cpu_accounting_mask(); |
fae9bc29 LP |
1299 | |
1300 | if (cgroup_context_has_cpu_weight(c) || | |
66ebf6c0 | 1301 | cgroup_context_has_cpu_shares(c) || |
3a43da28 | 1302 | c->cpu_quota_per_sec_usec != USEC_INFINITY) |
fae9bc29 | 1303 | mask |= CGROUP_MASK_CPU; |
ecedd90f | 1304 | |
538b4852 TH |
1305 | if (cgroup_context_has_io_config(c) || cgroup_context_has_blockio_config(c)) |
1306 | mask |= CGROUP_MASK_IO | CGROUP_MASK_BLKIO; | |
ecedd90f | 1307 | |
4ad49000 | 1308 | if (c->memory_accounting || |
da4d897e TH |
1309 | c->memory_limit != CGROUP_LIMIT_MAX || |
1310 | cgroup_context_has_unified_memory_config(c)) | |
efdb0237 | 1311 | mask |= CGROUP_MASK_MEMORY; |
8e274523 | 1312 | |
a931ad47 LP |
1313 | if (c->device_allow || |
1314 | c->device_policy != CGROUP_AUTO) | |
084c7007 | 1315 | mask |= CGROUP_MASK_DEVICES | CGROUP_MASK_BPF_DEVICES; |
4ad49000 | 1316 | |
03a7b521 | 1317 | if (c->tasks_accounting || |
8793fa25 | 1318 | c->tasks_max != CGROUP_LIMIT_MAX) |
03a7b521 LP |
1319 | mask |= CGROUP_MASK_PIDS; |
1320 | ||
fae9bc29 | 1321 | return CGROUP_MASK_EXTEND_JOINED(mask); |
8e274523 LP |
1322 | } |
1323 | ||
53aea74a | 1324 | static CGroupMask unit_get_bpf_mask(Unit *u) { |
17f14955 RG |
1325 | CGroupMask mask = 0; |
1326 | ||
fae9bc29 LP |
1327 | /* Figure out which controllers we need, based on the cgroup context, possibly taking into account children |
1328 | * too. */ | |
1329 | ||
17f14955 RG |
1330 | if (unit_get_needs_bpf_firewall(u)) |
1331 | mask |= CGROUP_MASK_BPF_FIREWALL; | |
1332 | ||
1333 | return mask; | |
1334 | } | |
1335 | ||
efdb0237 | 1336 | CGroupMask unit_get_own_mask(Unit *u) { |
4ad49000 | 1337 | CGroupContext *c; |
8e274523 | 1338 | |
442ce775 LP |
1339 | /* Returns the mask of controllers the unit needs for itself. If a unit is not properly loaded, return an empty |
1340 | * mask, as we shouldn't reflect it in the cgroup hierarchy then. */ | |
1341 | ||
1342 | if (u->load_state != UNIT_LOADED) | |
1343 | return 0; | |
efdb0237 | 1344 | |
4ad49000 LP |
1345 | c = unit_get_cgroup_context(u); |
1346 | if (!c) | |
1347 | return 0; | |
8e274523 | 1348 | |
17f14955 | 1349 | return cgroup_context_get_mask(c) | unit_get_bpf_mask(u) | unit_get_delegate_mask(u); |
02638280 LP |
1350 | } |
1351 | ||
1352 | CGroupMask unit_get_delegate_mask(Unit *u) { | |
1353 | CGroupContext *c; | |
1354 | ||
1355 | /* If delegation is turned on, then turn on selected controllers, unless we are on the legacy hierarchy and the | |
1356 | * process we fork into is known to drop privileges, and hence shouldn't get access to the controllers. | |
19af675e | 1357 | * |
02638280 | 1358 | * Note that on the unified hierarchy it is safe to delegate controllers to unprivileged services. */ |
a931ad47 | 1359 | |
1d9cc876 | 1360 | if (!unit_cgroup_delegate(u)) |
02638280 LP |
1361 | return 0; |
1362 | ||
1363 | if (cg_all_unified() <= 0) { | |
a931ad47 LP |
1364 | ExecContext *e; |
1365 | ||
1366 | e = unit_get_exec_context(u); | |
02638280 LP |
1367 | if (e && !exec_context_maintains_privileges(e)) |
1368 | return 0; | |
a931ad47 LP |
1369 | } |
1370 | ||
1d9cc876 | 1371 | assert_se(c = unit_get_cgroup_context(u)); |
fae9bc29 | 1372 | return CGROUP_MASK_EXTEND_JOINED(c->delegate_controllers); |
8e274523 LP |
1373 | } |
1374 | ||
efdb0237 | 1375 | CGroupMask unit_get_members_mask(Unit *u) { |
4ad49000 | 1376 | assert(u); |
bc432dc7 | 1377 | |
02638280 | 1378 | /* Returns the mask of controllers all of the unit's children require, merged */ |
efdb0237 | 1379 | |
bc432dc7 | 1380 | if (u->cgroup_members_mask_valid) |
26a17ca2 | 1381 | return u->cgroup_members_mask; /* Use cached value if possible */ |
bc432dc7 | 1382 | |
64e844e5 | 1383 | u->cgroup_members_mask = 0; |
bc432dc7 LP |
1384 | |
1385 | if (u->type == UNIT_SLICE) { | |
eef85c4a | 1386 | void *v; |
bc432dc7 LP |
1387 | Unit *member; |
1388 | Iterator i; | |
1389 | ||
eef85c4a | 1390 | HASHMAP_FOREACH_KEY(v, member, u->dependencies[UNIT_BEFORE], i) { |
bc432dc7 LP |
1391 | |
1392 | if (member == u) | |
1393 | continue; | |
1394 | ||
d4fdc205 | 1395 | if (UNIT_DEREF(member->slice) != u) |
bc432dc7 LP |
1396 | continue; |
1397 | ||
31604970 | 1398 | u->cgroup_members_mask |= unit_get_subtree_mask(member); /* note that this calls ourselves again, for the children */ |
bc432dc7 LP |
1399 | } |
1400 | } | |
1401 | ||
1402 | u->cgroup_members_mask_valid = true; | |
6414b7c9 | 1403 | return u->cgroup_members_mask; |
246aa6dd LP |
1404 | } |
1405 | ||
efdb0237 | 1406 | CGroupMask unit_get_siblings_mask(Unit *u) { |
4ad49000 | 1407 | assert(u); |
246aa6dd | 1408 | |
efdb0237 LP |
1409 | /* Returns the mask of controllers all of the unit's siblings |
1410 | * require, i.e. the members mask of the unit's parent slice | |
1411 | * if there is one. */ | |
1412 | ||
bc432dc7 | 1413 | if (UNIT_ISSET(u->slice)) |
637f421e | 1414 | return unit_get_members_mask(UNIT_DEREF(u->slice)); |
4ad49000 | 1415 | |
64e844e5 | 1416 | return unit_get_subtree_mask(u); /* we are the top-level slice */ |
246aa6dd LP |
1417 | } |
1418 | ||
4f6f62e4 CD |
1419 | CGroupMask unit_get_disable_mask(Unit *u) { |
1420 | CGroupContext *c; | |
1421 | ||
1422 | c = unit_get_cgroup_context(u); | |
1423 | if (!c) | |
1424 | return 0; | |
1425 | ||
1426 | return c->disable_controllers; | |
1427 | } | |
1428 | ||
1429 | CGroupMask unit_get_ancestor_disable_mask(Unit *u) { | |
1430 | CGroupMask mask; | |
1431 | ||
1432 | assert(u); | |
1433 | mask = unit_get_disable_mask(u); | |
1434 | ||
1435 | /* Returns the mask of controllers which are marked as forcibly | |
1436 | * disabled in any ancestor unit or the unit in question. */ | |
1437 | ||
1438 | if (UNIT_ISSET(u->slice)) | |
1439 | mask |= unit_get_ancestor_disable_mask(UNIT_DEREF(u->slice)); | |
1440 | ||
1441 | return mask; | |
1442 | } | |
1443 | ||
efdb0237 LP |
1444 | CGroupMask unit_get_subtree_mask(Unit *u) { |
1445 | ||
1446 | /* Returns the mask of this subtree, meaning of the group | |
1447 | * itself and its children. */ | |
1448 | ||
1449 | return unit_get_own_mask(u) | unit_get_members_mask(u); | |
1450 | } | |
1451 | ||
1452 | CGroupMask unit_get_target_mask(Unit *u) { | |
1453 | CGroupMask mask; | |
1454 | ||
1455 | /* This returns the cgroup mask of all controllers to enable | |
1456 | * for a specific cgroup, i.e. everything it needs itself, | |
1457 | * plus all that its children need, plus all that its siblings | |
1458 | * need. This is primarily useful on the legacy cgroup | |
1459 | * hierarchy, where we need to duplicate each cgroup in each | |
1460 | * hierarchy that shall be enabled for it. */ | |
6414b7c9 | 1461 | |
efdb0237 LP |
1462 | mask = unit_get_own_mask(u) | unit_get_members_mask(u) | unit_get_siblings_mask(u); |
1463 | mask &= u->manager->cgroup_supported; | |
1464 | ||
1465 | return mask; | |
1466 | } | |
1467 | ||
1468 | CGroupMask unit_get_enable_mask(Unit *u) { | |
1469 | CGroupMask mask; | |
1470 | ||
1471 | /* This returns the cgroup mask of all controllers to enable | |
1472 | * for the children of a specific cgroup. This is primarily | |
1473 | * useful for the unified cgroup hierarchy, where each cgroup | |
1474 | * controls which controllers are enabled for its children. */ | |
1475 | ||
1476 | mask = unit_get_members_mask(u); | |
6414b7c9 DS |
1477 | mask &= u->manager->cgroup_supported; |
1478 | ||
1479 | return mask; | |
1480 | } | |
1481 | ||
5af88058 | 1482 | void unit_invalidate_cgroup_members_masks(Unit *u) { |
bc432dc7 LP |
1483 | assert(u); |
1484 | ||
5af88058 LP |
1485 | /* Recurse invalidate the member masks cache all the way up the tree */ |
1486 | u->cgroup_members_mask_valid = false; | |
bc432dc7 | 1487 | |
5af88058 LP |
1488 | if (UNIT_ISSET(u->slice)) |
1489 | unit_invalidate_cgroup_members_masks(UNIT_DEREF(u->slice)); | |
6414b7c9 DS |
1490 | } |
1491 | ||
6592b975 | 1492 | const char *unit_get_realized_cgroup_path(Unit *u, CGroupMask mask) { |
03b90d4b | 1493 | |
6592b975 | 1494 | /* Returns the realized cgroup path of the specified unit where all specified controllers are available. */ |
03b90d4b LP |
1495 | |
1496 | while (u) { | |
6592b975 | 1497 | |
03b90d4b LP |
1498 | if (u->cgroup_path && |
1499 | u->cgroup_realized && | |
d94a24ca | 1500 | FLAGS_SET(u->cgroup_realized_mask, mask)) |
03b90d4b LP |
1501 | return u->cgroup_path; |
1502 | ||
1503 | u = UNIT_DEREF(u->slice); | |
1504 | } | |
1505 | ||
1506 | return NULL; | |
1507 | } | |
1508 | ||
6592b975 LP |
1509 | static const char *migrate_callback(CGroupMask mask, void *userdata) { |
1510 | return unit_get_realized_cgroup_path(userdata, mask); | |
1511 | } | |
1512 | ||
efdb0237 LP |
1513 | char *unit_default_cgroup_path(Unit *u) { |
1514 | _cleanup_free_ char *escaped = NULL, *slice = NULL; | |
1515 | int r; | |
1516 | ||
1517 | assert(u); | |
1518 | ||
1519 | if (unit_has_name(u, SPECIAL_ROOT_SLICE)) | |
1520 | return strdup(u->manager->cgroup_root); | |
1521 | ||
1522 | if (UNIT_ISSET(u->slice) && !unit_has_name(UNIT_DEREF(u->slice), SPECIAL_ROOT_SLICE)) { | |
1523 | r = cg_slice_to_path(UNIT_DEREF(u->slice)->id, &slice); | |
1524 | if (r < 0) | |
1525 | return NULL; | |
1526 | } | |
1527 | ||
1528 | escaped = cg_escape(u->id); | |
1529 | if (!escaped) | |
1530 | return NULL; | |
1531 | ||
1532 | if (slice) | |
605405c6 ZJS |
1533 | return strjoin(u->manager->cgroup_root, "/", slice, "/", |
1534 | escaped); | |
efdb0237 | 1535 | else |
605405c6 | 1536 | return strjoin(u->manager->cgroup_root, "/", escaped); |
efdb0237 LP |
1537 | } |
1538 | ||
1539 | int unit_set_cgroup_path(Unit *u, const char *path) { | |
1540 | _cleanup_free_ char *p = NULL; | |
1541 | int r; | |
1542 | ||
1543 | assert(u); | |
1544 | ||
1545 | if (path) { | |
1546 | p = strdup(path); | |
1547 | if (!p) | |
1548 | return -ENOMEM; | |
1549 | } else | |
1550 | p = NULL; | |
1551 | ||
1552 | if (streq_ptr(u->cgroup_path, p)) | |
1553 | return 0; | |
1554 | ||
1555 | if (p) { | |
1556 | r = hashmap_put(u->manager->cgroup_unit, p, u); | |
1557 | if (r < 0) | |
1558 | return r; | |
1559 | } | |
1560 | ||
1561 | unit_release_cgroup(u); | |
1562 | ||
ae2a15bc | 1563 | u->cgroup_path = TAKE_PTR(p); |
efdb0237 LP |
1564 | |
1565 | return 1; | |
1566 | } | |
1567 | ||
1568 | int unit_watch_cgroup(Unit *u) { | |
ab2c3861 | 1569 | _cleanup_free_ char *events = NULL; |
efdb0237 LP |
1570 | int r; |
1571 | ||
1572 | assert(u); | |
1573 | ||
1574 | if (!u->cgroup_path) | |
1575 | return 0; | |
1576 | ||
1577 | if (u->cgroup_inotify_wd >= 0) | |
1578 | return 0; | |
1579 | ||
1580 | /* Only applies to the unified hierarchy */ | |
c22800e4 | 1581 | r = cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER); |
b4cccbc1 LP |
1582 | if (r < 0) |
1583 | return log_error_errno(r, "Failed to determine whether the name=systemd hierarchy is unified: %m"); | |
1584 | if (r == 0) | |
efdb0237 LP |
1585 | return 0; |
1586 | ||
1587 | /* Don't watch the root slice, it's pointless. */ | |
1588 | if (unit_has_name(u, SPECIAL_ROOT_SLICE)) | |
1589 | return 0; | |
1590 | ||
1591 | r = hashmap_ensure_allocated(&u->manager->cgroup_inotify_wd_unit, &trivial_hash_ops); | |
1592 | if (r < 0) | |
1593 | return log_oom(); | |
1594 | ||
ab2c3861 | 1595 | r = cg_get_path(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path, "cgroup.events", &events); |
efdb0237 LP |
1596 | if (r < 0) |
1597 | return log_oom(); | |
1598 | ||
ab2c3861 | 1599 | u->cgroup_inotify_wd = inotify_add_watch(u->manager->cgroup_inotify_fd, events, IN_MODIFY); |
efdb0237 LP |
1600 | if (u->cgroup_inotify_wd < 0) { |
1601 | ||
1602 | if (errno == ENOENT) /* If the directory is already | |
1603 | * gone we don't need to track | |
1604 | * it, so this is not an error */ | |
1605 | return 0; | |
1606 | ||
1607 | return log_unit_error_errno(u, errno, "Failed to add inotify watch descriptor for control group %s: %m", u->cgroup_path); | |
1608 | } | |
1609 | ||
1610 | r = hashmap_put(u->manager->cgroup_inotify_wd_unit, INT_TO_PTR(u->cgroup_inotify_wd), u); | |
1611 | if (r < 0) | |
1612 | return log_unit_error_errno(u, r, "Failed to add inotify watch descriptor to hash map: %m"); | |
1613 | ||
1614 | return 0; | |
1615 | } | |
1616 | ||
a4634b21 LP |
1617 | int unit_pick_cgroup_path(Unit *u) { |
1618 | _cleanup_free_ char *path = NULL; | |
1619 | int r; | |
1620 | ||
1621 | assert(u); | |
1622 | ||
1623 | if (u->cgroup_path) | |
1624 | return 0; | |
1625 | ||
1626 | if (!UNIT_HAS_CGROUP_CONTEXT(u)) | |
1627 | return -EINVAL; | |
1628 | ||
1629 | path = unit_default_cgroup_path(u); | |
1630 | if (!path) | |
1631 | return log_oom(); | |
1632 | ||
1633 | r = unit_set_cgroup_path(u, path); | |
1634 | if (r == -EEXIST) | |
1635 | return log_unit_error_errno(u, r, "Control group %s exists already.", path); | |
1636 | if (r < 0) | |
1637 | return log_unit_error_errno(u, r, "Failed to set unit's control group path to %s: %m", path); | |
1638 | ||
1639 | return 0; | |
1640 | } | |
1641 | ||
efdb0237 LP |
1642 | static int unit_create_cgroup( |
1643 | Unit *u, | |
1644 | CGroupMask target_mask, | |
0d2d6fbf CD |
1645 | CGroupMask enable_mask, |
1646 | ManagerState state) { | |
efdb0237 | 1647 | |
65be7e06 | 1648 | bool created; |
27adcc97 | 1649 | int r; |
64747e2d | 1650 | |
4ad49000 | 1651 | assert(u); |
64747e2d | 1652 | |
27c4ed79 | 1653 | if (!UNIT_HAS_CGROUP_CONTEXT(u)) |
0cd385d3 LP |
1654 | return 0; |
1655 | ||
a4634b21 LP |
1656 | /* Figure out our cgroup path */ |
1657 | r = unit_pick_cgroup_path(u); | |
1658 | if (r < 0) | |
1659 | return r; | |
b58b8e11 | 1660 | |
03b90d4b | 1661 | /* First, create our own group */ |
efdb0237 | 1662 | r = cg_create_everywhere(u->manager->cgroup_supported, target_mask, u->cgroup_path); |
23bbb0de | 1663 | if (r < 0) |
efdb0237 | 1664 | return log_unit_error_errno(u, r, "Failed to create cgroup %s: %m", u->cgroup_path); |
490c5a37 | 1665 | created = r; |
efdb0237 LP |
1666 | |
1667 | /* Start watching it */ | |
1668 | (void) unit_watch_cgroup(u); | |
1669 | ||
65be7e06 | 1670 | /* Preserve enabled controllers in delegated units, adjust others. */ |
1fd3a10c | 1671 | if (created || !u->cgroup_realized || !unit_cgroup_delegate(u)) { |
27adcc97 | 1672 | CGroupMask result_mask = 0; |
65be7e06 ZJS |
1673 | |
1674 | /* Enable all controllers we need */ | |
27adcc97 | 1675 | r = cg_enable_everywhere(u->manager->cgroup_supported, enable_mask, u->cgroup_path, &result_mask); |
65be7e06 | 1676 | if (r < 0) |
27adcc97 LP |
1677 | log_unit_warning_errno(u, r, "Failed to enable/disable controllers on cgroup %s, ignoring: %m", u->cgroup_path); |
1678 | ||
1679 | /* If we just turned off a controller, this might release the controller for our parent too, let's | |
1680 | * enqueue the parent for re-realization in that case again. */ | |
1681 | if (UNIT_ISSET(u->slice)) { | |
1682 | CGroupMask turned_off; | |
1683 | ||
1684 | turned_off = (u->cgroup_realized ? u->cgroup_enabled_mask & ~result_mask : 0); | |
1685 | if (turned_off != 0) { | |
1686 | Unit *parent; | |
1687 | ||
1688 | /* Force the parent to propagate the enable mask to the kernel again, by invalidating | |
1689 | * the controller we just turned off. */ | |
1690 | ||
1691 | for (parent = UNIT_DEREF(u->slice); parent; parent = UNIT_DEREF(parent->slice)) | |
1692 | unit_invalidate_cgroup(parent, turned_off); | |
1693 | } | |
1694 | } | |
1695 | ||
1696 | /* Remember what's actually enabled now */ | |
1697 | u->cgroup_enabled_mask = result_mask; | |
65be7e06 | 1698 | } |
03b90d4b LP |
1699 | |
1700 | /* Keep track that this is now realized */ | |
4ad49000 | 1701 | u->cgroup_realized = true; |
efdb0237 | 1702 | u->cgroup_realized_mask = target_mask; |
4ad49000 | 1703 | |
1d9cc876 | 1704 | if (u->type != UNIT_SLICE && !unit_cgroup_delegate(u)) { |
0cd385d3 LP |
1705 | |
1706 | /* Then, possibly move things over, but not if | |
1707 | * subgroups may contain processes, which is the case | |
1708 | * for slice and delegation units. */ | |
1709 | r = cg_migrate_everywhere(u->manager->cgroup_supported, u->cgroup_path, u->cgroup_path, migrate_callback, u); | |
1710 | if (r < 0) | |
efdb0237 | 1711 | log_unit_warning_errno(u, r, "Failed to migrate cgroup from to %s, ignoring: %m", u->cgroup_path); |
0cd385d3 | 1712 | } |
03b90d4b | 1713 | |
0d2d6fbf CD |
1714 | /* Set attributes */ |
1715 | cgroup_context_apply(u, target_mask, state); | |
1716 | cgroup_xattr_apply(u); | |
1717 | ||
64747e2d LP |
1718 | return 0; |
1719 | } | |
1720 | ||
6592b975 LP |
1721 | static int unit_attach_pid_to_cgroup_via_bus(Unit *u, pid_t pid, const char *suffix_path) { |
1722 | _cleanup_(sd_bus_error_free) sd_bus_error error = SD_BUS_ERROR_NULL; | |
1723 | char *pp; | |
7b3fd631 | 1724 | int r; |
6592b975 | 1725 | |
7b3fd631 LP |
1726 | assert(u); |
1727 | ||
6592b975 LP |
1728 | if (MANAGER_IS_SYSTEM(u->manager)) |
1729 | return -EINVAL; | |
1730 | ||
1731 | if (!u->manager->system_bus) | |
1732 | return -EIO; | |
1733 | ||
1734 | if (!u->cgroup_path) | |
1735 | return -EINVAL; | |
1736 | ||
1737 | /* Determine this unit's cgroup path relative to our cgroup root */ | |
1738 | pp = path_startswith(u->cgroup_path, u->manager->cgroup_root); | |
1739 | if (!pp) | |
1740 | return -EINVAL; | |
1741 | ||
1742 | pp = strjoina("/", pp, suffix_path); | |
858d36c1 | 1743 | path_simplify(pp, false); |
6592b975 LP |
1744 | |
1745 | r = sd_bus_call_method(u->manager->system_bus, | |
1746 | "org.freedesktop.systemd1", | |
1747 | "/org/freedesktop/systemd1", | |
1748 | "org.freedesktop.systemd1.Manager", | |
1749 | "AttachProcessesToUnit", | |
1750 | &error, NULL, | |
1751 | "ssau", | |
1752 | NULL /* empty unit name means client's unit, i.e. us */, pp, 1, (uint32_t) pid); | |
7b3fd631 | 1753 | if (r < 0) |
6592b975 LP |
1754 | return log_unit_debug_errno(u, r, "Failed to attach unit process " PID_FMT " via the bus: %s", pid, bus_error_message(&error, r)); |
1755 | ||
1756 | return 0; | |
1757 | } | |
1758 | ||
1759 | int unit_attach_pids_to_cgroup(Unit *u, Set *pids, const char *suffix_path) { | |
1760 | CGroupMask delegated_mask; | |
1761 | const char *p; | |
1762 | Iterator i; | |
1763 | void *pidp; | |
1764 | int r, q; | |
1765 | ||
1766 | assert(u); | |
1767 | ||
1768 | if (!UNIT_HAS_CGROUP_CONTEXT(u)) | |
1769 | return -EINVAL; | |
1770 | ||
1771 | if (set_isempty(pids)) | |
1772 | return 0; | |
7b3fd631 | 1773 | |
6592b975 | 1774 | r = unit_realize_cgroup(u); |
7b3fd631 LP |
1775 | if (r < 0) |
1776 | return r; | |
1777 | ||
6592b975 LP |
1778 | if (isempty(suffix_path)) |
1779 | p = u->cgroup_path; | |
1780 | else | |
1781 | p = strjoina(u->cgroup_path, "/", suffix_path); | |
1782 | ||
1783 | delegated_mask = unit_get_delegate_mask(u); | |
1784 | ||
1785 | r = 0; | |
1786 | SET_FOREACH(pidp, pids, i) { | |
1787 | pid_t pid = PTR_TO_PID(pidp); | |
1788 | CGroupController c; | |
1789 | ||
1790 | /* First, attach the PID to the main cgroup hierarchy */ | |
1791 | q = cg_attach(SYSTEMD_CGROUP_CONTROLLER, p, pid); | |
1792 | if (q < 0) { | |
1793 | log_unit_debug_errno(u, q, "Couldn't move process " PID_FMT " to requested cgroup '%s': %m", pid, p); | |
1794 | ||
1795 | if (MANAGER_IS_USER(u->manager) && IN_SET(q, -EPERM, -EACCES)) { | |
1796 | int z; | |
1797 | ||
1798 | /* If we are in a user instance, and we can't move the process ourselves due to | |
1799 | * permission problems, let's ask the system instance about it instead. Since it's more | |
1800 | * privileged it might be able to move the process across the leaves of a subtree who's | |
1801 | * top node is not owned by us. */ | |
1802 | ||
1803 | z = unit_attach_pid_to_cgroup_via_bus(u, pid, suffix_path); | |
1804 | if (z < 0) | |
1805 | log_unit_debug_errno(u, z, "Couldn't move process " PID_FMT " to requested cgroup '%s' via the system bus either: %m", pid, p); | |
1806 | else | |
1807 | continue; /* When the bus thing worked via the bus we are fully done for this PID. */ | |
1808 | } | |
1809 | ||
1810 | if (r >= 0) | |
1811 | r = q; /* Remember first error */ | |
1812 | ||
1813 | continue; | |
1814 | } | |
1815 | ||
1816 | q = cg_all_unified(); | |
1817 | if (q < 0) | |
1818 | return q; | |
1819 | if (q > 0) | |
1820 | continue; | |
1821 | ||
1822 | /* In the legacy hierarchy, attach the process to the request cgroup if possible, and if not to the | |
1823 | * innermost realized one */ | |
1824 | ||
1825 | for (c = 0; c < _CGROUP_CONTROLLER_MAX; c++) { | |
1826 | CGroupMask bit = CGROUP_CONTROLLER_TO_MASK(c); | |
1827 | const char *realized; | |
1828 | ||
1829 | if (!(u->manager->cgroup_supported & bit)) | |
1830 | continue; | |
1831 | ||
1832 | /* If this controller is delegated and realized, honour the caller's request for the cgroup suffix. */ | |
1833 | if (delegated_mask & u->cgroup_realized_mask & bit) { | |
1834 | q = cg_attach(cgroup_controller_to_string(c), p, pid); | |
1835 | if (q >= 0) | |
1836 | continue; /* Success! */ | |
1837 | ||
1838 | 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", | |
1839 | pid, p, cgroup_controller_to_string(c)); | |
1840 | } | |
1841 | ||
1842 | /* So this controller is either not delegate or realized, or something else weird happened. In | |
1843 | * that case let's attach the PID at least to the closest cgroup up the tree that is | |
1844 | * realized. */ | |
1845 | realized = unit_get_realized_cgroup_path(u, bit); | |
1846 | if (!realized) | |
1847 | continue; /* Not even realized in the root slice? Then let's not bother */ | |
1848 | ||
1849 | q = cg_attach(cgroup_controller_to_string(c), realized, pid); | |
1850 | if (q < 0) | |
1851 | log_unit_debug_errno(u, q, "Failed to attach PID " PID_FMT " to realized cgroup %s in controller %s, ignoring: %m", | |
1852 | pid, realized, cgroup_controller_to_string(c)); | |
1853 | } | |
1854 | } | |
1855 | ||
1856 | return r; | |
7b3fd631 LP |
1857 | } |
1858 | ||
906c06f6 DM |
1859 | static bool unit_has_mask_realized( |
1860 | Unit *u, | |
1861 | CGroupMask target_mask, | |
17f14955 | 1862 | CGroupMask enable_mask) { |
906c06f6 | 1863 | |
bc432dc7 LP |
1864 | assert(u); |
1865 | ||
d5095dcd LP |
1866 | /* Returns true if this unit is fully realized. We check four things: |
1867 | * | |
1868 | * 1. Whether the cgroup was created at all | |
1869 | * 2. Whether the cgroup was created in all the hierarchies we need it to be created in (in case of cgroupsv1) | |
1870 | * 3. Whether the cgroup has all the right controllers enabled (in case of cgroupsv2) | |
1871 | * 4. Whether the invalidation mask is currently zero | |
1872 | * | |
1873 | * If you wonder why we mask the target realization and enable mask with CGROUP_MASK_V1/CGROUP_MASK_V2: note | |
1874 | * that there are three sets of bitmasks: CGROUP_MASK_V1 (for real cgroupv1 controllers), CGROUP_MASK_V2 (for | |
1875 | * real cgroupv2 controllers) and CGROUP_MASK_BPF (for BPF-based pseudo-controllers). Now, cgroup_realized_mask | |
1876 | * is only matters for cgroupsv1 controllers, and cgroup_enabled_mask only used for cgroupsv2, and if they | |
1877 | * differ in the others, we don't really care. (After all, the cgroup_enabled_mask tracks with controllers are | |
1878 | * enabled through cgroup.subtree_control, and since the BPF pseudo-controllers don't show up there, they | |
1879 | * simply don't matter. */ | |
1880 | ||
906c06f6 | 1881 | return u->cgroup_realized && |
d5095dcd LP |
1882 | ((u->cgroup_realized_mask ^ target_mask) & CGROUP_MASK_V1) == 0 && |
1883 | ((u->cgroup_enabled_mask ^ enable_mask) & CGROUP_MASK_V2) == 0 && | |
17f14955 | 1884 | u->cgroup_invalidated_mask == 0; |
6414b7c9 DS |
1885 | } |
1886 | ||
4f6f62e4 CD |
1887 | static bool unit_has_mask_disables_realized( |
1888 | Unit *u, | |
1889 | CGroupMask target_mask, | |
1890 | CGroupMask enable_mask) { | |
1891 | ||
1892 | assert(u); | |
1893 | ||
1894 | /* Returns true if all controllers which should be disabled are indeed disabled. | |
1895 | * | |
1896 | * Unlike unit_has_mask_realized, we don't care what was enabled, only that anything we want to remove is | |
1897 | * already removed. */ | |
1898 | ||
1899 | return !u->cgroup_realized || | |
1900 | (FLAGS_SET(u->cgroup_realized_mask, target_mask & CGROUP_MASK_V1) && | |
1901 | FLAGS_SET(u->cgroup_enabled_mask, enable_mask & CGROUP_MASK_V2)); | |
1902 | } | |
1903 | ||
a57669d2 CD |
1904 | static bool unit_has_mask_enables_realized( |
1905 | Unit *u, | |
1906 | CGroupMask target_mask, | |
1907 | CGroupMask enable_mask) { | |
1908 | ||
1909 | assert(u); | |
1910 | ||
1911 | /* Returns true if all controllers which should be enabled are indeed enabled. | |
1912 | * | |
1913 | * Unlike unit_has_mask_realized, we don't care about the controllers that are not present, only that anything | |
1914 | * we want to add is already added. */ | |
1915 | ||
1916 | return u->cgroup_realized && | |
1917 | ((u->cgroup_realized_mask | target_mask) & CGROUP_MASK_V1) == u->cgroup_realized_mask && | |
1918 | ((u->cgroup_enabled_mask | enable_mask) & CGROUP_MASK_V2) == u->cgroup_enabled_mask; | |
1919 | } | |
1920 | ||
27adcc97 | 1921 | void unit_add_to_cgroup_realize_queue(Unit *u) { |
2aa57a65 LP |
1922 | assert(u); |
1923 | ||
1924 | if (u->in_cgroup_realize_queue) | |
1925 | return; | |
1926 | ||
1927 | LIST_PREPEND(cgroup_realize_queue, u->manager->cgroup_realize_queue, u); | |
1928 | u->in_cgroup_realize_queue = true; | |
1929 | } | |
1930 | ||
1931 | static void unit_remove_from_cgroup_realize_queue(Unit *u) { | |
1932 | assert(u); | |
1933 | ||
1934 | if (!u->in_cgroup_realize_queue) | |
1935 | return; | |
1936 | ||
1937 | LIST_REMOVE(cgroup_realize_queue, u->manager->cgroup_realize_queue, u); | |
1938 | u->in_cgroup_realize_queue = false; | |
1939 | } | |
1940 | ||
a57669d2 CD |
1941 | /* Controllers can only be enabled breadth-first, from the root of the |
1942 | * hierarchy downwards to the unit in question. */ | |
1943 | static int unit_realize_cgroup_now_enable(Unit *u, ManagerState state) { | |
1944 | CGroupMask target_mask, enable_mask, new_target_mask, new_enable_mask; | |
1945 | int r; | |
1946 | ||
1947 | assert(u); | |
1948 | ||
1949 | /* First go deal with this unit's parent, or we won't be able to enable | |
1950 | * any new controllers at this layer. */ | |
1951 | if (UNIT_ISSET(u->slice)) { | |
1952 | r = unit_realize_cgroup_now_enable(UNIT_DEREF(u->slice), state); | |
1953 | if (r < 0) | |
1954 | return r; | |
1955 | } | |
1956 | ||
1957 | target_mask = unit_get_target_mask(u); | |
1958 | enable_mask = unit_get_enable_mask(u); | |
1959 | ||
1960 | /* We can only enable in this direction, don't try to disable anything. | |
1961 | */ | |
1962 | if (unit_has_mask_enables_realized(u, target_mask, enable_mask)) | |
1963 | return 0; | |
1964 | ||
1965 | new_target_mask = u->cgroup_realized_mask | target_mask; | |
1966 | new_enable_mask = u->cgroup_enabled_mask | enable_mask; | |
1967 | ||
1968 | r = unit_create_cgroup(u, new_target_mask, new_enable_mask, state); | |
1969 | if (r < 0) | |
1970 | return r; | |
1971 | ||
1972 | return 0; | |
1973 | } | |
1974 | ||
4f6f62e4 CD |
1975 | /* Controllers can only be disabled depth-first, from the leaves of the |
1976 | * hierarchy upwards to the unit in question. */ | |
1977 | static int unit_realize_cgroup_now_disable(Unit *u, ManagerState state) { | |
1978 | Iterator i; | |
1979 | Unit *m; | |
1980 | void *v; | |
1981 | ||
1982 | assert(u); | |
1983 | ||
1984 | if (u->type != UNIT_SLICE) | |
1985 | return 0; | |
1986 | ||
1987 | HASHMAP_FOREACH_KEY(v, m, u->dependencies[UNIT_BEFORE], i) { | |
1988 | CGroupMask target_mask, enable_mask, new_target_mask, new_enable_mask; | |
1989 | int r; | |
1990 | ||
1991 | if (UNIT_DEREF(m->slice) != u) | |
1992 | continue; | |
1993 | ||
1994 | /* The cgroup for this unit might not actually be fully | |
1995 | * realised yet, in which case it isn't holding any controllers | |
1996 | * open anyway. */ | |
1997 | if (!m->cgroup_path) | |
1998 | continue; | |
1999 | ||
2000 | /* We must disable those below us first in order to release the | |
2001 | * controller. */ | |
2002 | if (m->type == UNIT_SLICE) | |
2003 | (void) unit_realize_cgroup_now_disable(m, state); | |
2004 | ||
2005 | target_mask = unit_get_target_mask(m); | |
2006 | enable_mask = unit_get_enable_mask(m); | |
2007 | ||
2008 | /* We can only disable in this direction, don't try to enable | |
2009 | * anything. */ | |
2010 | if (unit_has_mask_disables_realized(m, target_mask, enable_mask)) | |
2011 | continue; | |
2012 | ||
2013 | new_target_mask = m->cgroup_realized_mask & target_mask; | |
2014 | new_enable_mask = m->cgroup_enabled_mask & enable_mask; | |
2015 | ||
2016 | r = unit_create_cgroup(m, new_target_mask, new_enable_mask, state); | |
2017 | if (r < 0) | |
2018 | return r; | |
2019 | } | |
2020 | ||
2021 | return 0; | |
2022 | } | |
a57669d2 | 2023 | |
6414b7c9 DS |
2024 | /* Check if necessary controllers and attributes for a unit are in place. |
2025 | * | |
a57669d2 CD |
2026 | * - If so, do nothing. |
2027 | * - If not, create paths, move processes over, and set attributes. | |
2028 | * | |
2029 | * Controllers can only be *enabled* in a breadth-first way, and *disabled* in | |
2030 | * a depth-first way. As such the process looks like this: | |
2031 | * | |
2032 | * Suppose we have a cgroup hierarchy which looks like this: | |
2033 | * | |
2034 | * root | |
2035 | * / \ | |
2036 | * / \ | |
2037 | * / \ | |
2038 | * a b | |
2039 | * / \ / \ | |
2040 | * / \ / \ | |
2041 | * c d e f | |
2042 | * / \ / \ / \ / \ | |
2043 | * h i j k l m n o | |
2044 | * | |
2045 | * 1. We want to realise cgroup "d" now. | |
2046 | * 2. cgroup "a" has DisableController=cpu in the associated unit. | |
2047 | * 3. cgroup "k" just started requesting the memory controller. | |
2048 | * | |
2049 | * To make this work we must do the following in order: | |
2050 | * | |
2051 | * 1. Disable CPU controller in k, j | |
2052 | * 2. Disable CPU controller in d | |
2053 | * 3. Enable memory controller in root | |
2054 | * 4. Enable memory controller in a | |
2055 | * 5. Enable memory controller in d | |
2056 | * 6. Enable memory controller in k | |
2057 | * | |
2058 | * Notice that we need to touch j in one direction, but not the other. We also | |
2059 | * don't go beyond d when disabling -- it's up to "a" to get realized if it | |
2060 | * wants to disable further. The basic rules are therefore: | |
2061 | * | |
2062 | * - If you're disabling something, you need to realise all of the cgroups from | |
2063 | * your recursive descendants to the root. This starts from the leaves. | |
2064 | * - If you're enabling something, you need to realise from the root cgroup | |
2065 | * downwards, but you don't need to iterate your recursive descendants. | |
6414b7c9 DS |
2066 | * |
2067 | * Returns 0 on success and < 0 on failure. */ | |
db785129 | 2068 | static int unit_realize_cgroup_now(Unit *u, ManagerState state) { |
efdb0237 | 2069 | CGroupMask target_mask, enable_mask; |
6414b7c9 | 2070 | int r; |
64747e2d | 2071 | |
4ad49000 | 2072 | assert(u); |
64747e2d | 2073 | |
2aa57a65 | 2074 | unit_remove_from_cgroup_realize_queue(u); |
64747e2d | 2075 | |
efdb0237 | 2076 | target_mask = unit_get_target_mask(u); |
ccf78df1 TH |
2077 | enable_mask = unit_get_enable_mask(u); |
2078 | ||
17f14955 | 2079 | if (unit_has_mask_realized(u, target_mask, enable_mask)) |
0a1eb06d | 2080 | return 0; |
64747e2d | 2081 | |
4f6f62e4 CD |
2082 | /* Disable controllers below us, if there are any */ |
2083 | r = unit_realize_cgroup_now_disable(u, state); | |
2084 | if (r < 0) | |
2085 | return r; | |
2086 | ||
2087 | /* Enable controllers above us, if there are any */ | |
6414b7c9 | 2088 | if (UNIT_ISSET(u->slice)) { |
a57669d2 | 2089 | r = unit_realize_cgroup_now_enable(UNIT_DEREF(u->slice), state); |
6414b7c9 DS |
2090 | if (r < 0) |
2091 | return r; | |
2092 | } | |
4ad49000 | 2093 | |
0d2d6fbf CD |
2094 | /* Now actually deal with the cgroup we were trying to realise and set attributes */ |
2095 | r = unit_create_cgroup(u, target_mask, enable_mask, state); | |
6414b7c9 DS |
2096 | if (r < 0) |
2097 | return r; | |
2098 | ||
c2baf11c LP |
2099 | /* Now, reset the invalidation mask */ |
2100 | u->cgroup_invalidated_mask = 0; | |
6414b7c9 | 2101 | return 0; |
64747e2d LP |
2102 | } |
2103 | ||
91a6073e | 2104 | unsigned manager_dispatch_cgroup_realize_queue(Manager *m) { |
db785129 | 2105 | ManagerState state; |
4ad49000 | 2106 | unsigned n = 0; |
db785129 | 2107 | Unit *i; |
6414b7c9 | 2108 | int r; |
ecedd90f | 2109 | |
91a6073e LP |
2110 | assert(m); |
2111 | ||
db785129 LP |
2112 | state = manager_state(m); |
2113 | ||
91a6073e LP |
2114 | while ((i = m->cgroup_realize_queue)) { |
2115 | assert(i->in_cgroup_realize_queue); | |
ecedd90f | 2116 | |
2aa57a65 LP |
2117 | if (UNIT_IS_INACTIVE_OR_FAILED(unit_active_state(i))) { |
2118 | /* Maybe things changed, and the unit is not actually active anymore? */ | |
2119 | unit_remove_from_cgroup_realize_queue(i); | |
2120 | continue; | |
2121 | } | |
2122 | ||
db785129 | 2123 | r = unit_realize_cgroup_now(i, state); |
6414b7c9 | 2124 | if (r < 0) |
efdb0237 | 2125 | log_warning_errno(r, "Failed to realize cgroups for queued unit %s, ignoring: %m", i->id); |
0a1eb06d | 2126 | |
4ad49000 LP |
2127 | n++; |
2128 | } | |
ecedd90f | 2129 | |
4ad49000 | 2130 | return n; |
8e274523 LP |
2131 | } |
2132 | ||
91a6073e | 2133 | static void unit_add_siblings_to_cgroup_realize_queue(Unit *u) { |
4ad49000 | 2134 | Unit *slice; |
ca949c9d | 2135 | |
4ad49000 LP |
2136 | /* This adds the siblings of the specified unit and the |
2137 | * siblings of all parent units to the cgroup queue. (But | |
2138 | * neither the specified unit itself nor the parents.) */ | |
2139 | ||
2140 | while ((slice = UNIT_DEREF(u->slice))) { | |
2141 | Iterator i; | |
2142 | Unit *m; | |
eef85c4a | 2143 | void *v; |
8f53a7b8 | 2144 | |
eef85c4a | 2145 | HASHMAP_FOREACH_KEY(v, m, u->dependencies[UNIT_BEFORE], i) { |
4ad49000 LP |
2146 | if (m == u) |
2147 | continue; | |
8e274523 | 2148 | |
6414b7c9 DS |
2149 | /* Skip units that have a dependency on the slice |
2150 | * but aren't actually in it. */ | |
4ad49000 | 2151 | if (UNIT_DEREF(m->slice) != slice) |
50159e6a | 2152 | continue; |
8e274523 | 2153 | |
6414b7c9 DS |
2154 | /* No point in doing cgroup application for units |
2155 | * without active processes. */ | |
2156 | if (UNIT_IS_INACTIVE_OR_FAILED(unit_active_state(m))) | |
2157 | continue; | |
2158 | ||
2159 | /* If the unit doesn't need any new controllers | |
2160 | * and has current ones realized, it doesn't need | |
2161 | * any changes. */ | |
906c06f6 DM |
2162 | if (unit_has_mask_realized(m, |
2163 | unit_get_target_mask(m), | |
17f14955 | 2164 | unit_get_enable_mask(m))) |
6414b7c9 DS |
2165 | continue; |
2166 | ||
91a6073e | 2167 | unit_add_to_cgroup_realize_queue(m); |
50159e6a LP |
2168 | } |
2169 | ||
4ad49000 | 2170 | u = slice; |
8e274523 | 2171 | } |
4ad49000 LP |
2172 | } |
2173 | ||
0a1eb06d | 2174 | int unit_realize_cgroup(Unit *u) { |
4ad49000 LP |
2175 | assert(u); |
2176 | ||
35b7ff80 | 2177 | if (!UNIT_HAS_CGROUP_CONTEXT(u)) |
0a1eb06d | 2178 | return 0; |
8e274523 | 2179 | |
4ad49000 LP |
2180 | /* So, here's the deal: when realizing the cgroups for this |
2181 | * unit, we need to first create all parents, but there's more | |
2182 | * actually: for the weight-based controllers we also need to | |
2183 | * make sure that all our siblings (i.e. units that are in the | |
73e231ab | 2184 | * same slice as we are) have cgroups, too. Otherwise, things |
4ad49000 LP |
2185 | * would become very uneven as each of their processes would |
2186 | * get as much resources as all our group together. This call | |
2187 | * will synchronously create the parent cgroups, but will | |
2188 | * defer work on the siblings to the next event loop | |
2189 | * iteration. */ | |
ca949c9d | 2190 | |
4ad49000 | 2191 | /* Add all sibling slices to the cgroup queue. */ |
91a6073e | 2192 | unit_add_siblings_to_cgroup_realize_queue(u); |
4ad49000 | 2193 | |
6414b7c9 | 2194 | /* And realize this one now (and apply the values) */ |
db785129 | 2195 | return unit_realize_cgroup_now(u, manager_state(u->manager)); |
8e274523 LP |
2196 | } |
2197 | ||
efdb0237 LP |
2198 | void unit_release_cgroup(Unit *u) { |
2199 | assert(u); | |
2200 | ||
8a0d5388 LP |
2201 | /* Forgets all cgroup details for this cgroup — but does *not* destroy the cgroup. This is hence OK to call |
2202 | * when we close down everything for reexecution, where we really want to leave the cgroup in place. */ | |
efdb0237 LP |
2203 | |
2204 | if (u->cgroup_path) { | |
2205 | (void) hashmap_remove(u->manager->cgroup_unit, u->cgroup_path); | |
2206 | u->cgroup_path = mfree(u->cgroup_path); | |
2207 | } | |
2208 | ||
2209 | if (u->cgroup_inotify_wd >= 0) { | |
2210 | if (inotify_rm_watch(u->manager->cgroup_inotify_fd, u->cgroup_inotify_wd) < 0) | |
5e1ee764 | 2211 | log_unit_debug_errno(u, errno, "Failed to remove cgroup inotify watch %i for %s, ignoring: %m", u->cgroup_inotify_wd, u->id); |
efdb0237 LP |
2212 | |
2213 | (void) hashmap_remove(u->manager->cgroup_inotify_wd_unit, INT_TO_PTR(u->cgroup_inotify_wd)); | |
2214 | u->cgroup_inotify_wd = -1; | |
2215 | } | |
2216 | } | |
2217 | ||
2218 | void unit_prune_cgroup(Unit *u) { | |
8e274523 | 2219 | int r; |
efdb0237 | 2220 | bool is_root_slice; |
8e274523 | 2221 | |
4ad49000 | 2222 | assert(u); |
8e274523 | 2223 | |
efdb0237 LP |
2224 | /* Removes the cgroup, if empty and possible, and stops watching it. */ |
2225 | ||
4ad49000 LP |
2226 | if (!u->cgroup_path) |
2227 | return; | |
8e274523 | 2228 | |
fe700f46 LP |
2229 | (void) unit_get_cpu_usage(u, NULL); /* Cache the last CPU usage value before we destroy the cgroup */ |
2230 | ||
efdb0237 LP |
2231 | is_root_slice = unit_has_name(u, SPECIAL_ROOT_SLICE); |
2232 | ||
2233 | r = cg_trim_everywhere(u->manager->cgroup_supported, u->cgroup_path, !is_root_slice); | |
dab5bf85 | 2234 | if (r < 0) { |
f29ff115 | 2235 | log_unit_debug_errno(u, r, "Failed to destroy cgroup %s, ignoring: %m", u->cgroup_path); |
dab5bf85 RL |
2236 | return; |
2237 | } | |
8e274523 | 2238 | |
efdb0237 LP |
2239 | if (is_root_slice) |
2240 | return; | |
2241 | ||
2242 | unit_release_cgroup(u); | |
0a1eb06d | 2243 | |
4ad49000 | 2244 | u->cgroup_realized = false; |
bc432dc7 | 2245 | u->cgroup_realized_mask = 0; |
ccf78df1 | 2246 | u->cgroup_enabled_mask = 0; |
084c7007 RG |
2247 | |
2248 | u->bpf_device_control_installed = bpf_program_unref(u->bpf_device_control_installed); | |
8e274523 LP |
2249 | } |
2250 | ||
efdb0237 | 2251 | int unit_search_main_pid(Unit *u, pid_t *ret) { |
4ad49000 LP |
2252 | _cleanup_fclose_ FILE *f = NULL; |
2253 | pid_t pid = 0, npid, mypid; | |
efdb0237 | 2254 | int r; |
4ad49000 LP |
2255 | |
2256 | assert(u); | |
efdb0237 | 2257 | assert(ret); |
4ad49000 LP |
2258 | |
2259 | if (!u->cgroup_path) | |
efdb0237 | 2260 | return -ENXIO; |
4ad49000 | 2261 | |
efdb0237 LP |
2262 | r = cg_enumerate_processes(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path, &f); |
2263 | if (r < 0) | |
2264 | return r; | |
4ad49000 | 2265 | |
df0ff127 | 2266 | mypid = getpid_cached(); |
4ad49000 LP |
2267 | while (cg_read_pid(f, &npid) > 0) { |
2268 | pid_t ppid; | |
2269 | ||
2270 | if (npid == pid) | |
2271 | continue; | |
8e274523 | 2272 | |
4ad49000 | 2273 | /* Ignore processes that aren't our kids */ |
6bc73acb | 2274 | if (get_process_ppid(npid, &ppid) >= 0 && ppid != mypid) |
4ad49000 | 2275 | continue; |
8e274523 | 2276 | |
efdb0237 | 2277 | if (pid != 0) |
4ad49000 LP |
2278 | /* Dang, there's more than one daemonized PID |
2279 | in this group, so we don't know what process | |
2280 | is the main process. */ | |
efdb0237 LP |
2281 | |
2282 | return -ENODATA; | |
8e274523 | 2283 | |
4ad49000 | 2284 | pid = npid; |
8e274523 LP |
2285 | } |
2286 | ||
efdb0237 LP |
2287 | *ret = pid; |
2288 | return 0; | |
2289 | } | |
2290 | ||
2291 | static int unit_watch_pids_in_path(Unit *u, const char *path) { | |
b3c5bad3 | 2292 | _cleanup_closedir_ DIR *d = NULL; |
efdb0237 LP |
2293 | _cleanup_fclose_ FILE *f = NULL; |
2294 | int ret = 0, r; | |
2295 | ||
2296 | assert(u); | |
2297 | assert(path); | |
2298 | ||
2299 | r = cg_enumerate_processes(SYSTEMD_CGROUP_CONTROLLER, path, &f); | |
2300 | if (r < 0) | |
2301 | ret = r; | |
2302 | else { | |
2303 | pid_t pid; | |
2304 | ||
2305 | while ((r = cg_read_pid(f, &pid)) > 0) { | |
2306 | r = unit_watch_pid(u, pid); | |
2307 | if (r < 0 && ret >= 0) | |
2308 | ret = r; | |
2309 | } | |
2310 | ||
2311 | if (r < 0 && ret >= 0) | |
2312 | ret = r; | |
2313 | } | |
2314 | ||
2315 | r = cg_enumerate_subgroups(SYSTEMD_CGROUP_CONTROLLER, path, &d); | |
2316 | if (r < 0) { | |
2317 | if (ret >= 0) | |
2318 | ret = r; | |
2319 | } else { | |
2320 | char *fn; | |
2321 | ||
2322 | while ((r = cg_read_subgroup(d, &fn)) > 0) { | |
2323 | _cleanup_free_ char *p = NULL; | |
2324 | ||
605405c6 | 2325 | p = strjoin(path, "/", fn); |
efdb0237 LP |
2326 | free(fn); |
2327 | ||
2328 | if (!p) | |
2329 | return -ENOMEM; | |
2330 | ||
2331 | r = unit_watch_pids_in_path(u, p); | |
2332 | if (r < 0 && ret >= 0) | |
2333 | ret = r; | |
2334 | } | |
2335 | ||
2336 | if (r < 0 && ret >= 0) | |
2337 | ret = r; | |
2338 | } | |
2339 | ||
2340 | return ret; | |
2341 | } | |
2342 | ||
11aef522 LP |
2343 | int unit_synthesize_cgroup_empty_event(Unit *u) { |
2344 | int r; | |
2345 | ||
2346 | assert(u); | |
2347 | ||
2348 | /* Enqueue a synthetic cgroup empty event if this unit doesn't watch any PIDs anymore. This is compatibility | |
2349 | * support for non-unified systems where notifications aren't reliable, and hence need to take whatever we can | |
2350 | * get as notification source as soon as we stopped having any useful PIDs to watch for. */ | |
2351 | ||
2352 | if (!u->cgroup_path) | |
2353 | return -ENOENT; | |
2354 | ||
2355 | r = cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER); | |
2356 | if (r < 0) | |
2357 | return r; | |
2358 | if (r > 0) /* On unified we have reliable notifications, and don't need this */ | |
2359 | return 0; | |
2360 | ||
2361 | if (!set_isempty(u->pids)) | |
2362 | return 0; | |
2363 | ||
2364 | unit_add_to_cgroup_empty_queue(u); | |
2365 | return 0; | |
2366 | } | |
2367 | ||
efdb0237 | 2368 | int unit_watch_all_pids(Unit *u) { |
b4cccbc1 LP |
2369 | int r; |
2370 | ||
efdb0237 LP |
2371 | assert(u); |
2372 | ||
2373 | /* Adds all PIDs from our cgroup to the set of PIDs we | |
2374 | * watch. This is a fallback logic for cases where we do not | |
2375 | * get reliable cgroup empty notifications: we try to use | |
2376 | * SIGCHLD as replacement. */ | |
2377 | ||
2378 | if (!u->cgroup_path) | |
2379 | return -ENOENT; | |
2380 | ||
c22800e4 | 2381 | r = cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER); |
b4cccbc1 LP |
2382 | if (r < 0) |
2383 | return r; | |
2384 | if (r > 0) /* On unified we can use proper notifications */ | |
efdb0237 LP |
2385 | return 0; |
2386 | ||
2387 | return unit_watch_pids_in_path(u, u->cgroup_path); | |
2388 | } | |
2389 | ||
09e24654 LP |
2390 | static int on_cgroup_empty_event(sd_event_source *s, void *userdata) { |
2391 | Manager *m = userdata; | |
2392 | Unit *u; | |
efdb0237 LP |
2393 | int r; |
2394 | ||
09e24654 LP |
2395 | assert(s); |
2396 | assert(m); | |
efdb0237 | 2397 | |
09e24654 LP |
2398 | u = m->cgroup_empty_queue; |
2399 | if (!u) | |
efdb0237 LP |
2400 | return 0; |
2401 | ||
09e24654 LP |
2402 | assert(u->in_cgroup_empty_queue); |
2403 | u->in_cgroup_empty_queue = false; | |
2404 | LIST_REMOVE(cgroup_empty_queue, m->cgroup_empty_queue, u); | |
2405 | ||
2406 | if (m->cgroup_empty_queue) { | |
2407 | /* More stuff queued, let's make sure we remain enabled */ | |
2408 | r = sd_event_source_set_enabled(s, SD_EVENT_ONESHOT); | |
2409 | if (r < 0) | |
19a691a9 | 2410 | log_debug_errno(r, "Failed to reenable cgroup empty event source, ignoring: %m"); |
09e24654 | 2411 | } |
efdb0237 LP |
2412 | |
2413 | unit_add_to_gc_queue(u); | |
2414 | ||
2415 | if (UNIT_VTABLE(u)->notify_cgroup_empty) | |
2416 | UNIT_VTABLE(u)->notify_cgroup_empty(u); | |
2417 | ||
2418 | return 0; | |
2419 | } | |
2420 | ||
09e24654 LP |
2421 | void unit_add_to_cgroup_empty_queue(Unit *u) { |
2422 | int r; | |
2423 | ||
2424 | assert(u); | |
2425 | ||
2426 | /* Note that there are four different ways how cgroup empty events reach us: | |
2427 | * | |
2428 | * 1. On the unified hierarchy we get an inotify event on the cgroup | |
2429 | * | |
2430 | * 2. On the legacy hierarchy, when running in system mode, we get a datagram on the cgroup agent socket | |
2431 | * | |
2432 | * 3. On the legacy hierarchy, when running in user mode, we get a D-Bus signal on the system bus | |
2433 | * | |
2434 | * 4. On the legacy hierarchy, in service units we start watching all processes of the cgroup for SIGCHLD as | |
2435 | * soon as we get one SIGCHLD, to deal with unreliable cgroup notifications. | |
2436 | * | |
2437 | * Regardless which way we got the notification, we'll verify it here, and then add it to a separate | |
2438 | * queue. This queue will be dispatched at a lower priority than the SIGCHLD handler, so that we always use | |
2439 | * SIGCHLD if we can get it first, and only use the cgroup empty notifications if there's no SIGCHLD pending | |
2440 | * (which might happen if the cgroup doesn't contain processes that are our own child, which is typically the | |
2441 | * case for scope units). */ | |
2442 | ||
2443 | if (u->in_cgroup_empty_queue) | |
2444 | return; | |
2445 | ||
2446 | /* Let's verify that the cgroup is really empty */ | |
2447 | if (!u->cgroup_path) | |
2448 | return; | |
2449 | r = cg_is_empty_recursive(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path); | |
2450 | if (r < 0) { | |
2451 | log_unit_debug_errno(u, r, "Failed to determine whether cgroup %s is empty: %m", u->cgroup_path); | |
2452 | return; | |
2453 | } | |
2454 | if (r == 0) | |
2455 | return; | |
2456 | ||
2457 | LIST_PREPEND(cgroup_empty_queue, u->manager->cgroup_empty_queue, u); | |
2458 | u->in_cgroup_empty_queue = true; | |
2459 | ||
2460 | /* Trigger the defer event */ | |
2461 | r = sd_event_source_set_enabled(u->manager->cgroup_empty_event_source, SD_EVENT_ONESHOT); | |
2462 | if (r < 0) | |
2463 | log_debug_errno(r, "Failed to enable cgroup empty event source: %m"); | |
2464 | } | |
2465 | ||
efdb0237 LP |
2466 | static int on_cgroup_inotify_event(sd_event_source *s, int fd, uint32_t revents, void *userdata) { |
2467 | Manager *m = userdata; | |
2468 | ||
2469 | assert(s); | |
2470 | assert(fd >= 0); | |
2471 | assert(m); | |
2472 | ||
2473 | for (;;) { | |
2474 | union inotify_event_buffer buffer; | |
2475 | struct inotify_event *e; | |
2476 | ssize_t l; | |
2477 | ||
2478 | l = read(fd, &buffer, sizeof(buffer)); | |
2479 | if (l < 0) { | |
47249640 | 2480 | if (IN_SET(errno, EINTR, EAGAIN)) |
efdb0237 LP |
2481 | return 0; |
2482 | ||
2483 | return log_error_errno(errno, "Failed to read control group inotify events: %m"); | |
2484 | } | |
2485 | ||
2486 | FOREACH_INOTIFY_EVENT(e, buffer, l) { | |
2487 | Unit *u; | |
2488 | ||
2489 | if (e->wd < 0) | |
2490 | /* Queue overflow has no watch descriptor */ | |
2491 | continue; | |
2492 | ||
2493 | if (e->mask & IN_IGNORED) | |
2494 | /* The watch was just removed */ | |
2495 | continue; | |
2496 | ||
2497 | u = hashmap_get(m->cgroup_inotify_wd_unit, INT_TO_PTR(e->wd)); | |
2498 | if (!u) /* Not that inotify might deliver | |
2499 | * events for a watch even after it | |
2500 | * was removed, because it was queued | |
2501 | * before the removal. Let's ignore | |
2502 | * this here safely. */ | |
2503 | continue; | |
2504 | ||
09e24654 | 2505 | unit_add_to_cgroup_empty_queue(u); |
efdb0237 LP |
2506 | } |
2507 | } | |
8e274523 LP |
2508 | } |
2509 | ||
17f14955 RG |
2510 | static int cg_bpf_mask_supported(CGroupMask *ret) { |
2511 | CGroupMask mask = 0; | |
2512 | int r; | |
2513 | ||
2514 | /* BPF-based firewall */ | |
2515 | r = bpf_firewall_supported(); | |
2516 | if (r > 0) | |
2517 | mask |= CGROUP_MASK_BPF_FIREWALL; | |
2518 | ||
084c7007 RG |
2519 | /* BPF-based device access control */ |
2520 | r = bpf_devices_supported(); | |
2521 | if (r > 0) | |
2522 | mask |= CGROUP_MASK_BPF_DEVICES; | |
2523 | ||
17f14955 RG |
2524 | *ret = mask; |
2525 | return 0; | |
2526 | } | |
2527 | ||
8e274523 | 2528 | int manager_setup_cgroup(Manager *m) { |
9444b1f2 | 2529 | _cleanup_free_ char *path = NULL; |
10bd3e2e | 2530 | const char *scope_path; |
efdb0237 | 2531 | CGroupController c; |
b4cccbc1 | 2532 | int r, all_unified; |
17f14955 | 2533 | CGroupMask mask; |
efdb0237 | 2534 | char *e; |
8e274523 LP |
2535 | |
2536 | assert(m); | |
2537 | ||
35d2e7ec | 2538 | /* 1. Determine hierarchy */ |
efdb0237 | 2539 | m->cgroup_root = mfree(m->cgroup_root); |
9444b1f2 | 2540 | r = cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER, 0, &m->cgroup_root); |
23bbb0de MS |
2541 | if (r < 0) |
2542 | return log_error_errno(r, "Cannot determine cgroup we are running in: %m"); | |
8e274523 | 2543 | |
efdb0237 LP |
2544 | /* Chop off the init scope, if we are already located in it */ |
2545 | e = endswith(m->cgroup_root, "/" SPECIAL_INIT_SCOPE); | |
0d8c31ff | 2546 | |
efdb0237 LP |
2547 | /* LEGACY: Also chop off the system slice if we are in |
2548 | * it. This is to support live upgrades from older systemd | |
2549 | * versions where PID 1 was moved there. Also see | |
2550 | * cg_get_root_path(). */ | |
463d0d15 | 2551 | if (!e && MANAGER_IS_SYSTEM(m)) { |
9444b1f2 | 2552 | e = endswith(m->cgroup_root, "/" SPECIAL_SYSTEM_SLICE); |
15c60e99 | 2553 | if (!e) |
efdb0237 | 2554 | e = endswith(m->cgroup_root, "/system"); /* even more legacy */ |
0baf24dd | 2555 | } |
efdb0237 LP |
2556 | if (e) |
2557 | *e = 0; | |
7ccfb64a | 2558 | |
7546145e LP |
2559 | /* And make sure to store away the root value without trailing slash, even for the root dir, so that we can |
2560 | * easily prepend it everywhere. */ | |
2561 | delete_trailing_chars(m->cgroup_root, "/"); | |
8e274523 | 2562 | |
35d2e7ec | 2563 | /* 2. Show data */ |
9444b1f2 | 2564 | r = cg_get_path(SYSTEMD_CGROUP_CONTROLLER, m->cgroup_root, NULL, &path); |
23bbb0de MS |
2565 | if (r < 0) |
2566 | return log_error_errno(r, "Cannot find cgroup mount point: %m"); | |
8e274523 | 2567 | |
415fc41c TH |
2568 | r = cg_unified_flush(); |
2569 | if (r < 0) | |
2570 | return log_error_errno(r, "Couldn't determine if we are running in the unified hierarchy: %m"); | |
5da38d07 | 2571 | |
b4cccbc1 | 2572 | all_unified = cg_all_unified(); |
d4c819ed ZJS |
2573 | if (all_unified < 0) |
2574 | return log_error_errno(all_unified, "Couldn't determine whether we are in all unified mode: %m"); | |
2575 | if (all_unified > 0) | |
efdb0237 | 2576 | log_debug("Unified cgroup hierarchy is located at %s.", path); |
b4cccbc1 | 2577 | else { |
c22800e4 | 2578 | r = cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER); |
b4cccbc1 LP |
2579 | if (r < 0) |
2580 | return log_error_errno(r, "Failed to determine whether systemd's own controller is in unified mode: %m"); | |
2581 | if (r > 0) | |
2582 | log_debug("Unified cgroup hierarchy is located at %s. Controllers are on legacy hierarchies.", path); | |
2583 | else | |
2584 | log_debug("Using cgroup controller " SYSTEMD_CGROUP_CONTROLLER_LEGACY ". File system hierarchy is at %s.", path); | |
2585 | } | |
efdb0237 | 2586 | |
09e24654 LP |
2587 | /* 3. Allocate cgroup empty defer event source */ |
2588 | m->cgroup_empty_event_source = sd_event_source_unref(m->cgroup_empty_event_source); | |
2589 | r = sd_event_add_defer(m->event, &m->cgroup_empty_event_source, on_cgroup_empty_event, m); | |
2590 | if (r < 0) | |
2591 | return log_error_errno(r, "Failed to create cgroup empty event source: %m"); | |
2592 | ||
2593 | r = sd_event_source_set_priority(m->cgroup_empty_event_source, SD_EVENT_PRIORITY_NORMAL-5); | |
2594 | if (r < 0) | |
2595 | return log_error_errno(r, "Failed to set priority of cgroup empty event source: %m"); | |
2596 | ||
2597 | r = sd_event_source_set_enabled(m->cgroup_empty_event_source, SD_EVENT_OFF); | |
2598 | if (r < 0) | |
2599 | return log_error_errno(r, "Failed to disable cgroup empty event source: %m"); | |
2600 | ||
2601 | (void) sd_event_source_set_description(m->cgroup_empty_event_source, "cgroup-empty"); | |
2602 | ||
2603 | /* 4. Install notifier inotify object, or agent */ | |
10bd3e2e | 2604 | if (cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER) > 0) { |
c6c18be3 | 2605 | |
09e24654 | 2606 | /* In the unified hierarchy we can get cgroup empty notifications via inotify. */ |
efdb0237 | 2607 | |
10bd3e2e LP |
2608 | m->cgroup_inotify_event_source = sd_event_source_unref(m->cgroup_inotify_event_source); |
2609 | safe_close(m->cgroup_inotify_fd); | |
efdb0237 | 2610 | |
10bd3e2e LP |
2611 | m->cgroup_inotify_fd = inotify_init1(IN_NONBLOCK|IN_CLOEXEC); |
2612 | if (m->cgroup_inotify_fd < 0) | |
2613 | return log_error_errno(errno, "Failed to create control group inotify object: %m"); | |
efdb0237 | 2614 | |
10bd3e2e LP |
2615 | r = sd_event_add_io(m->event, &m->cgroup_inotify_event_source, m->cgroup_inotify_fd, EPOLLIN, on_cgroup_inotify_event, m); |
2616 | if (r < 0) | |
2617 | return log_error_errno(r, "Failed to watch control group inotify object: %m"); | |
efdb0237 | 2618 | |
10bd3e2e LP |
2619 | /* Process cgroup empty notifications early, but after service notifications and SIGCHLD. Also |
2620 | * see handling of cgroup agent notifications, for the classic cgroup hierarchy support. */ | |
09e24654 | 2621 | r = sd_event_source_set_priority(m->cgroup_inotify_event_source, SD_EVENT_PRIORITY_NORMAL-4); |
10bd3e2e LP |
2622 | if (r < 0) |
2623 | return log_error_errno(r, "Failed to set priority of inotify event source: %m"); | |
efdb0237 | 2624 | |
10bd3e2e | 2625 | (void) sd_event_source_set_description(m->cgroup_inotify_event_source, "cgroup-inotify"); |
efdb0237 | 2626 | |
611c4f8a | 2627 | } else if (MANAGER_IS_SYSTEM(m) && manager_owns_host_root_cgroup(m) && !MANAGER_IS_TEST_RUN(m)) { |
efdb0237 | 2628 | |
10bd3e2e LP |
2629 | /* On the legacy hierarchy we only get notifications via cgroup agents. (Which isn't really reliable, |
2630 | * since it does not generate events when control groups with children run empty. */ | |
8e274523 | 2631 | |
10bd3e2e | 2632 | r = cg_install_release_agent(SYSTEMD_CGROUP_CONTROLLER, SYSTEMD_CGROUP_AGENT_PATH); |
23bbb0de | 2633 | if (r < 0) |
10bd3e2e LP |
2634 | log_warning_errno(r, "Failed to install release agent, ignoring: %m"); |
2635 | else if (r > 0) | |
2636 | log_debug("Installed release agent."); | |
2637 | else if (r == 0) | |
2638 | log_debug("Release agent already installed."); | |
2639 | } | |
efdb0237 | 2640 | |
09e24654 | 2641 | /* 5. Make sure we are in the special "init.scope" unit in the root slice. */ |
10bd3e2e LP |
2642 | scope_path = strjoina(m->cgroup_root, "/" SPECIAL_INIT_SCOPE); |
2643 | r = cg_create_and_attach(SYSTEMD_CGROUP_CONTROLLER, scope_path, 0); | |
aa77e234 MS |
2644 | if (r >= 0) { |
2645 | /* Also, move all other userspace processes remaining in the root cgroup into that scope. */ | |
2646 | r = cg_migrate(SYSTEMD_CGROUP_CONTROLLER, m->cgroup_root, SYSTEMD_CGROUP_CONTROLLER, scope_path, 0); | |
2647 | if (r < 0) | |
2648 | log_warning_errno(r, "Couldn't move remaining userspace processes, ignoring: %m"); | |
c6c18be3 | 2649 | |
aa77e234 MS |
2650 | /* 6. And pin it, so that it cannot be unmounted */ |
2651 | safe_close(m->pin_cgroupfs_fd); | |
2652 | m->pin_cgroupfs_fd = open(path, O_RDONLY|O_CLOEXEC|O_DIRECTORY|O_NOCTTY|O_NONBLOCK); | |
2653 | if (m->pin_cgroupfs_fd < 0) | |
2654 | return log_error_errno(errno, "Failed to open pin file: %m"); | |
0d8c31ff | 2655 | |
638cece4 | 2656 | } else if (!MANAGER_IS_TEST_RUN(m)) |
aa77e234 | 2657 | return log_error_errno(r, "Failed to create %s control group: %m", scope_path); |
10bd3e2e | 2658 | |
09e24654 | 2659 | /* 7. Always enable hierarchical support if it exists... */ |
638cece4 | 2660 | if (!all_unified && !MANAGER_IS_TEST_RUN(m)) |
10bd3e2e | 2661 | (void) cg_set_attribute("memory", "/", "memory.use_hierarchy", "1"); |
c6c18be3 | 2662 | |
17f14955 | 2663 | /* 8. Figure out which controllers are supported */ |
efdb0237 LP |
2664 | r = cg_mask_supported(&m->cgroup_supported); |
2665 | if (r < 0) | |
2666 | return log_error_errno(r, "Failed to determine supported controllers: %m"); | |
17f14955 RG |
2667 | |
2668 | /* 9. Figure out which bpf-based pseudo-controllers are supported */ | |
2669 | r = cg_bpf_mask_supported(&mask); | |
2670 | if (r < 0) | |
2671 | return log_error_errno(r, "Failed to determine supported bpf-based pseudo-controllers: %m"); | |
2672 | m->cgroup_supported |= mask; | |
2673 | ||
2674 | /* 10. Log which controllers are supported */ | |
efdb0237 | 2675 | for (c = 0; c < _CGROUP_CONTROLLER_MAX; c++) |
eee0a1e4 | 2676 | log_debug("Controller '%s' supported: %s", cgroup_controller_to_string(c), yes_no(m->cgroup_supported & CGROUP_CONTROLLER_TO_MASK(c))); |
9156e799 | 2677 | |
a32360f1 | 2678 | return 0; |
8e274523 LP |
2679 | } |
2680 | ||
c6c18be3 | 2681 | void manager_shutdown_cgroup(Manager *m, bool delete) { |
8e274523 LP |
2682 | assert(m); |
2683 | ||
9444b1f2 LP |
2684 | /* We can't really delete the group, since we are in it. But |
2685 | * let's trim it. */ | |
f6c63f6f | 2686 | if (delete && m->cgroup_root && m->test_run_flags != MANAGER_TEST_RUN_MINIMAL) |
efdb0237 LP |
2687 | (void) cg_trim(SYSTEMD_CGROUP_CONTROLLER, m->cgroup_root, false); |
2688 | ||
09e24654 LP |
2689 | m->cgroup_empty_event_source = sd_event_source_unref(m->cgroup_empty_event_source); |
2690 | ||
efdb0237 LP |
2691 | m->cgroup_inotify_wd_unit = hashmap_free(m->cgroup_inotify_wd_unit); |
2692 | ||
2693 | m->cgroup_inotify_event_source = sd_event_source_unref(m->cgroup_inotify_event_source); | |
2694 | m->cgroup_inotify_fd = safe_close(m->cgroup_inotify_fd); | |
8e274523 | 2695 | |
03e334a1 | 2696 | m->pin_cgroupfs_fd = safe_close(m->pin_cgroupfs_fd); |
c6c18be3 | 2697 | |
efdb0237 | 2698 | m->cgroup_root = mfree(m->cgroup_root); |
8e274523 LP |
2699 | } |
2700 | ||
4ad49000 | 2701 | Unit* manager_get_unit_by_cgroup(Manager *m, const char *cgroup) { |
acb14d31 | 2702 | char *p; |
4ad49000 | 2703 | Unit *u; |
acb14d31 LP |
2704 | |
2705 | assert(m); | |
2706 | assert(cgroup); | |
acb14d31 | 2707 | |
4ad49000 LP |
2708 | u = hashmap_get(m->cgroup_unit, cgroup); |
2709 | if (u) | |
2710 | return u; | |
acb14d31 | 2711 | |
8e70580b | 2712 | p = strdupa(cgroup); |
acb14d31 LP |
2713 | for (;;) { |
2714 | char *e; | |
2715 | ||
2716 | e = strrchr(p, '/'); | |
efdb0237 LP |
2717 | if (!e || e == p) |
2718 | return hashmap_get(m->cgroup_unit, SPECIAL_ROOT_SLICE); | |
acb14d31 LP |
2719 | |
2720 | *e = 0; | |
2721 | ||
4ad49000 LP |
2722 | u = hashmap_get(m->cgroup_unit, p); |
2723 | if (u) | |
2724 | return u; | |
acb14d31 LP |
2725 | } |
2726 | } | |
2727 | ||
b3ac818b | 2728 | Unit *manager_get_unit_by_pid_cgroup(Manager *m, pid_t pid) { |
4ad49000 | 2729 | _cleanup_free_ char *cgroup = NULL; |
8e274523 | 2730 | |
8c47c732 LP |
2731 | assert(m); |
2732 | ||
62a76913 | 2733 | if (!pid_is_valid(pid)) |
b3ac818b LP |
2734 | return NULL; |
2735 | ||
62a76913 | 2736 | if (cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER, pid, &cgroup) < 0) |
b3ac818b LP |
2737 | return NULL; |
2738 | ||
2739 | return manager_get_unit_by_cgroup(m, cgroup); | |
2740 | } | |
2741 | ||
2742 | Unit *manager_get_unit_by_pid(Manager *m, pid_t pid) { | |
62a76913 | 2743 | Unit *u, **array; |
b3ac818b LP |
2744 | |
2745 | assert(m); | |
2746 | ||
62a76913 LP |
2747 | /* Note that a process might be owned by multiple units, we return only one here, which is good enough for most |
2748 | * cases, though not strictly correct. We prefer the one reported by cgroup membership, as that's the most | |
2749 | * relevant one as children of the process will be assigned to that one, too, before all else. */ | |
2750 | ||
2751 | if (!pid_is_valid(pid)) | |
8c47c732 LP |
2752 | return NULL; |
2753 | ||
2ca9d979 | 2754 | if (pid == getpid_cached()) |
efdb0237 LP |
2755 | return hashmap_get(m->units, SPECIAL_INIT_SCOPE); |
2756 | ||
62a76913 | 2757 | u = manager_get_unit_by_pid_cgroup(m, pid); |
5fe8876b LP |
2758 | if (u) |
2759 | return u; | |
2760 | ||
62a76913 | 2761 | u = hashmap_get(m->watch_pids, PID_TO_PTR(pid)); |
5fe8876b LP |
2762 | if (u) |
2763 | return u; | |
2764 | ||
62a76913 LP |
2765 | array = hashmap_get(m->watch_pids, PID_TO_PTR(-pid)); |
2766 | if (array) | |
2767 | return array[0]; | |
2768 | ||
2769 | return NULL; | |
6dde1f33 | 2770 | } |
4fbf50b3 | 2771 | |
4ad49000 LP |
2772 | int manager_notify_cgroup_empty(Manager *m, const char *cgroup) { |
2773 | Unit *u; | |
4fbf50b3 | 2774 | |
4ad49000 LP |
2775 | assert(m); |
2776 | assert(cgroup); | |
4fbf50b3 | 2777 | |
09e24654 LP |
2778 | /* Called on the legacy hierarchy whenever we get an explicit cgroup notification from the cgroup agent process |
2779 | * or from the --system instance */ | |
2780 | ||
d8fdc620 LP |
2781 | log_debug("Got cgroup empty notification for: %s", cgroup); |
2782 | ||
4ad49000 | 2783 | u = manager_get_unit_by_cgroup(m, cgroup); |
5ad096b3 LP |
2784 | if (!u) |
2785 | return 0; | |
b56c28c3 | 2786 | |
09e24654 LP |
2787 | unit_add_to_cgroup_empty_queue(u); |
2788 | return 1; | |
5ad096b3 LP |
2789 | } |
2790 | ||
2791 | int unit_get_memory_current(Unit *u, uint64_t *ret) { | |
2792 | _cleanup_free_ char *v = NULL; | |
2793 | int r; | |
2794 | ||
2795 | assert(u); | |
2796 | assert(ret); | |
2797 | ||
2e4025c0 | 2798 | if (!UNIT_CGROUP_BOOL(u, memory_accounting)) |
cf3b4be1 LP |
2799 | return -ENODATA; |
2800 | ||
5ad096b3 LP |
2801 | if (!u->cgroup_path) |
2802 | return -ENODATA; | |
2803 | ||
1f73aa00 | 2804 | /* The root cgroup doesn't expose this information, let's get it from /proc instead */ |
611c4f8a | 2805 | if (unit_has_host_root_cgroup(u)) |
1f73aa00 LP |
2806 | return procfs_memory_get_current(ret); |
2807 | ||
efdb0237 | 2808 | if ((u->cgroup_realized_mask & CGROUP_MASK_MEMORY) == 0) |
5ad096b3 LP |
2809 | return -ENODATA; |
2810 | ||
b4cccbc1 LP |
2811 | r = cg_all_unified(); |
2812 | if (r < 0) | |
2813 | return r; | |
2814 | if (r > 0) | |
efdb0237 | 2815 | r = cg_get_attribute("memory", u->cgroup_path, "memory.current", &v); |
b4cccbc1 LP |
2816 | else |
2817 | r = cg_get_attribute("memory", u->cgroup_path, "memory.usage_in_bytes", &v); | |
5ad096b3 LP |
2818 | if (r == -ENOENT) |
2819 | return -ENODATA; | |
2820 | if (r < 0) | |
2821 | return r; | |
2822 | ||
2823 | return safe_atou64(v, ret); | |
2824 | } | |
2825 | ||
03a7b521 LP |
2826 | int unit_get_tasks_current(Unit *u, uint64_t *ret) { |
2827 | _cleanup_free_ char *v = NULL; | |
2828 | int r; | |
2829 | ||
2830 | assert(u); | |
2831 | assert(ret); | |
2832 | ||
2e4025c0 | 2833 | if (!UNIT_CGROUP_BOOL(u, tasks_accounting)) |
cf3b4be1 LP |
2834 | return -ENODATA; |
2835 | ||
03a7b521 LP |
2836 | if (!u->cgroup_path) |
2837 | return -ENODATA; | |
2838 | ||
c36a69f4 | 2839 | /* The root cgroup doesn't expose this information, let's get it from /proc instead */ |
611c4f8a | 2840 | if (unit_has_host_root_cgroup(u)) |
c36a69f4 LP |
2841 | return procfs_tasks_get_current(ret); |
2842 | ||
1f73aa00 LP |
2843 | if ((u->cgroup_realized_mask & CGROUP_MASK_PIDS) == 0) |
2844 | return -ENODATA; | |
2845 | ||
03a7b521 LP |
2846 | r = cg_get_attribute("pids", u->cgroup_path, "pids.current", &v); |
2847 | if (r == -ENOENT) | |
2848 | return -ENODATA; | |
2849 | if (r < 0) | |
2850 | return r; | |
2851 | ||
2852 | return safe_atou64(v, ret); | |
2853 | } | |
2854 | ||
5ad096b3 LP |
2855 | static int unit_get_cpu_usage_raw(Unit *u, nsec_t *ret) { |
2856 | _cleanup_free_ char *v = NULL; | |
2857 | uint64_t ns; | |
2858 | int r; | |
2859 | ||
2860 | assert(u); | |
2861 | assert(ret); | |
2862 | ||
2863 | if (!u->cgroup_path) | |
2864 | return -ENODATA; | |
2865 | ||
1f73aa00 | 2866 | /* The root cgroup doesn't expose this information, let's get it from /proc instead */ |
611c4f8a | 2867 | if (unit_has_host_root_cgroup(u)) |
1f73aa00 LP |
2868 | return procfs_cpu_get_usage(ret); |
2869 | ||
b4cccbc1 LP |
2870 | r = cg_all_unified(); |
2871 | if (r < 0) | |
2872 | return r; | |
f98c2585 CD |
2873 | |
2874 | /* Requisite controllers for CPU accounting are not enabled */ | |
2875 | if ((get_cpu_accounting_mask() & ~u->cgroup_realized_mask) != 0) | |
2876 | return -ENODATA; | |
2877 | ||
b4cccbc1 | 2878 | if (r > 0) { |
66ebf6c0 TH |
2879 | _cleanup_free_ char *val = NULL; |
2880 | uint64_t us; | |
5ad096b3 | 2881 | |
b734a4ff | 2882 | r = cg_get_keyed_attribute("cpu", u->cgroup_path, "cpu.stat", STRV_MAKE("usage_usec"), &val); |
66ebf6c0 TH |
2883 | if (r < 0) |
2884 | return r; | |
b734a4ff LP |
2885 | if (IN_SET(r, -ENOENT, -ENXIO)) |
2886 | return -ENODATA; | |
66ebf6c0 TH |
2887 | |
2888 | r = safe_atou64(val, &us); | |
2889 | if (r < 0) | |
2890 | return r; | |
2891 | ||
2892 | ns = us * NSEC_PER_USEC; | |
2893 | } else { | |
66ebf6c0 TH |
2894 | r = cg_get_attribute("cpuacct", u->cgroup_path, "cpuacct.usage", &v); |
2895 | if (r == -ENOENT) | |
2896 | return -ENODATA; | |
2897 | if (r < 0) | |
2898 | return r; | |
2899 | ||
2900 | r = safe_atou64(v, &ns); | |
2901 | if (r < 0) | |
2902 | return r; | |
2903 | } | |
5ad096b3 LP |
2904 | |
2905 | *ret = ns; | |
2906 | return 0; | |
2907 | } | |
2908 | ||
2909 | int unit_get_cpu_usage(Unit *u, nsec_t *ret) { | |
2910 | nsec_t ns; | |
2911 | int r; | |
2912 | ||
fe700f46 LP |
2913 | assert(u); |
2914 | ||
2915 | /* Retrieve the current CPU usage counter. This will subtract the CPU counter taken when the unit was | |
2916 | * started. If the cgroup has been removed already, returns the last cached value. To cache the value, simply | |
2917 | * call this function with a NULL return value. */ | |
2918 | ||
2e4025c0 | 2919 | if (!UNIT_CGROUP_BOOL(u, cpu_accounting)) |
cf3b4be1 LP |
2920 | return -ENODATA; |
2921 | ||
5ad096b3 | 2922 | r = unit_get_cpu_usage_raw(u, &ns); |
fe700f46 LP |
2923 | if (r == -ENODATA && u->cpu_usage_last != NSEC_INFINITY) { |
2924 | /* If we can't get the CPU usage anymore (because the cgroup was already removed, for example), use our | |
2925 | * cached value. */ | |
2926 | ||
2927 | if (ret) | |
2928 | *ret = u->cpu_usage_last; | |
2929 | return 0; | |
2930 | } | |
5ad096b3 LP |
2931 | if (r < 0) |
2932 | return r; | |
2933 | ||
66ebf6c0 TH |
2934 | if (ns > u->cpu_usage_base) |
2935 | ns -= u->cpu_usage_base; | |
5ad096b3 LP |
2936 | else |
2937 | ns = 0; | |
2938 | ||
fe700f46 LP |
2939 | u->cpu_usage_last = ns; |
2940 | if (ret) | |
2941 | *ret = ns; | |
2942 | ||
5ad096b3 LP |
2943 | return 0; |
2944 | } | |
2945 | ||
906c06f6 DM |
2946 | int unit_get_ip_accounting( |
2947 | Unit *u, | |
2948 | CGroupIPAccountingMetric metric, | |
2949 | uint64_t *ret) { | |
2950 | ||
6b659ed8 | 2951 | uint64_t value; |
906c06f6 DM |
2952 | int fd, r; |
2953 | ||
2954 | assert(u); | |
2955 | assert(metric >= 0); | |
2956 | assert(metric < _CGROUP_IP_ACCOUNTING_METRIC_MAX); | |
2957 | assert(ret); | |
2958 | ||
2e4025c0 | 2959 | if (!UNIT_CGROUP_BOOL(u, ip_accounting)) |
cf3b4be1 LP |
2960 | return -ENODATA; |
2961 | ||
906c06f6 DM |
2962 | fd = IN_SET(metric, CGROUP_IP_INGRESS_BYTES, CGROUP_IP_INGRESS_PACKETS) ? |
2963 | u->ip_accounting_ingress_map_fd : | |
2964 | u->ip_accounting_egress_map_fd; | |
906c06f6 DM |
2965 | if (fd < 0) |
2966 | return -ENODATA; | |
2967 | ||
2968 | if (IN_SET(metric, CGROUP_IP_INGRESS_BYTES, CGROUP_IP_EGRESS_BYTES)) | |
6b659ed8 | 2969 | r = bpf_firewall_read_accounting(fd, &value, NULL); |
906c06f6 | 2970 | else |
6b659ed8 LP |
2971 | r = bpf_firewall_read_accounting(fd, NULL, &value); |
2972 | if (r < 0) | |
2973 | return r; | |
2974 | ||
2975 | /* Add in additional metrics from a previous runtime. Note that when reexecing/reloading the daemon we compile | |
2976 | * all BPF programs and maps anew, but serialize the old counters. When deserializing we store them in the | |
2977 | * ip_accounting_extra[] field, and add them in here transparently. */ | |
2978 | ||
2979 | *ret = value + u->ip_accounting_extra[metric]; | |
906c06f6 DM |
2980 | |
2981 | return r; | |
2982 | } | |
2983 | ||
2984 | int unit_reset_cpu_accounting(Unit *u) { | |
5ad096b3 LP |
2985 | nsec_t ns; |
2986 | int r; | |
2987 | ||
2988 | assert(u); | |
2989 | ||
fe700f46 LP |
2990 | u->cpu_usage_last = NSEC_INFINITY; |
2991 | ||
5ad096b3 LP |
2992 | r = unit_get_cpu_usage_raw(u, &ns); |
2993 | if (r < 0) { | |
66ebf6c0 | 2994 | u->cpu_usage_base = 0; |
5ad096b3 | 2995 | return r; |
b56c28c3 | 2996 | } |
2633eb83 | 2997 | |
66ebf6c0 | 2998 | u->cpu_usage_base = ns; |
4ad49000 | 2999 | return 0; |
4fbf50b3 LP |
3000 | } |
3001 | ||
906c06f6 DM |
3002 | int unit_reset_ip_accounting(Unit *u) { |
3003 | int r = 0, q = 0; | |
3004 | ||
3005 | assert(u); | |
3006 | ||
3007 | if (u->ip_accounting_ingress_map_fd >= 0) | |
3008 | r = bpf_firewall_reset_accounting(u->ip_accounting_ingress_map_fd); | |
3009 | ||
3010 | if (u->ip_accounting_egress_map_fd >= 0) | |
3011 | q = bpf_firewall_reset_accounting(u->ip_accounting_egress_map_fd); | |
3012 | ||
6b659ed8 LP |
3013 | zero(u->ip_accounting_extra); |
3014 | ||
906c06f6 DM |
3015 | return r < 0 ? r : q; |
3016 | } | |
3017 | ||
e7ab4d1a LP |
3018 | void unit_invalidate_cgroup(Unit *u, CGroupMask m) { |
3019 | assert(u); | |
3020 | ||
3021 | if (!UNIT_HAS_CGROUP_CONTEXT(u)) | |
3022 | return; | |
3023 | ||
3024 | if (m == 0) | |
3025 | return; | |
3026 | ||
538b4852 TH |
3027 | /* always invalidate compat pairs together */ |
3028 | if (m & (CGROUP_MASK_IO | CGROUP_MASK_BLKIO)) | |
3029 | m |= CGROUP_MASK_IO | CGROUP_MASK_BLKIO; | |
3030 | ||
7cce4fb7 LP |
3031 | if (m & (CGROUP_MASK_CPU | CGROUP_MASK_CPUACCT)) |
3032 | m |= CGROUP_MASK_CPU | CGROUP_MASK_CPUACCT; | |
3033 | ||
e00068e7 | 3034 | if (FLAGS_SET(u->cgroup_invalidated_mask, m)) /* NOP? */ |
e7ab4d1a LP |
3035 | return; |
3036 | ||
e00068e7 | 3037 | u->cgroup_invalidated_mask |= m; |
91a6073e | 3038 | unit_add_to_cgroup_realize_queue(u); |
e7ab4d1a LP |
3039 | } |
3040 | ||
906c06f6 DM |
3041 | void unit_invalidate_cgroup_bpf(Unit *u) { |
3042 | assert(u); | |
3043 | ||
3044 | if (!UNIT_HAS_CGROUP_CONTEXT(u)) | |
3045 | return; | |
3046 | ||
17f14955 | 3047 | if (u->cgroup_invalidated_mask & CGROUP_MASK_BPF_FIREWALL) /* NOP? */ |
906c06f6 DM |
3048 | return; |
3049 | ||
17f14955 | 3050 | u->cgroup_invalidated_mask |= CGROUP_MASK_BPF_FIREWALL; |
91a6073e | 3051 | unit_add_to_cgroup_realize_queue(u); |
906c06f6 DM |
3052 | |
3053 | /* If we are a slice unit, we also need to put compile a new BPF program for all our children, as the IP access | |
3054 | * list of our children includes our own. */ | |
3055 | if (u->type == UNIT_SLICE) { | |
3056 | Unit *member; | |
3057 | Iterator i; | |
eef85c4a | 3058 | void *v; |
906c06f6 | 3059 | |
eef85c4a | 3060 | HASHMAP_FOREACH_KEY(v, member, u->dependencies[UNIT_BEFORE], i) { |
906c06f6 DM |
3061 | if (member == u) |
3062 | continue; | |
3063 | ||
3064 | if (UNIT_DEREF(member->slice) != u) | |
3065 | continue; | |
3066 | ||
3067 | unit_invalidate_cgroup_bpf(member); | |
3068 | } | |
3069 | } | |
3070 | } | |
3071 | ||
1d9cc876 LP |
3072 | bool unit_cgroup_delegate(Unit *u) { |
3073 | CGroupContext *c; | |
3074 | ||
3075 | assert(u); | |
3076 | ||
3077 | if (!UNIT_VTABLE(u)->can_delegate) | |
3078 | return false; | |
3079 | ||
3080 | c = unit_get_cgroup_context(u); | |
3081 | if (!c) | |
3082 | return false; | |
3083 | ||
3084 | return c->delegate; | |
3085 | } | |
3086 | ||
e7ab4d1a LP |
3087 | void manager_invalidate_startup_units(Manager *m) { |
3088 | Iterator i; | |
3089 | Unit *u; | |
3090 | ||
3091 | assert(m); | |
3092 | ||
3093 | SET_FOREACH(u, m->startup_units, i) | |
13c31542 | 3094 | unit_invalidate_cgroup(u, CGROUP_MASK_CPU|CGROUP_MASK_IO|CGROUP_MASK_BLKIO); |
e7ab4d1a LP |
3095 | } |
3096 | ||
4ad49000 LP |
3097 | static const char* const cgroup_device_policy_table[_CGROUP_DEVICE_POLICY_MAX] = { |
3098 | [CGROUP_AUTO] = "auto", | |
3099 | [CGROUP_CLOSED] = "closed", | |
3100 | [CGROUP_STRICT] = "strict", | |
3101 | }; | |
4fbf50b3 | 3102 | |
4ad49000 | 3103 | DEFINE_STRING_TABLE_LOOKUP(cgroup_device_policy, CGroupDevicePolicy); |