]>
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 | |
be2c0327 LP |
878 | /* We generally ignore errors caused by read-only mounted cgroup trees (assuming we are running in a container |
879 | * then), and missing cgroups, i.e. EROFS and ENOENT. */ | |
714e2e1d | 880 | |
be2c0327 LP |
881 | /* In fully unified mode these attributes don't exist on the host cgroup root. On legacy the weights exist, but |
882 | * setting the weight makes very little sense on the host root cgroup, as there are no other cgroups at this | |
883 | * level. The quota exists there too, but any attempt to write to it is refused with EINVAL. Inside of | |
884 | * containers we want to leave control of these to the container manager (and if cgroupsv2 delegation is used | |
885 | * we couldn't even write to them if we wanted to). */ | |
886 | if ((apply_mask & CGROUP_MASK_CPU) && !is_local_root) { | |
8e274523 | 887 | |
b4cccbc1 | 888 | if (cg_all_unified() > 0) { |
be2c0327 | 889 | uint64_t weight; |
b2f8b02e | 890 | |
be2c0327 LP |
891 | if (cgroup_context_has_cpu_weight(c)) |
892 | weight = cgroup_context_cpu_weight(c, state); | |
893 | else if (cgroup_context_has_cpu_shares(c)) { | |
894 | uint64_t shares; | |
66ebf6c0 | 895 | |
be2c0327 LP |
896 | shares = cgroup_context_cpu_shares(c, state); |
897 | weight = cgroup_cpu_shares_to_weight(shares); | |
66ebf6c0 | 898 | |
be2c0327 LP |
899 | log_cgroup_compat(u, "Applying [Startup]CPUShares=%" PRIu64 " as [Startup]CPUWeight=%" PRIu64 " on %s", |
900 | shares, weight, path); | |
901 | } else | |
902 | weight = CGROUP_WEIGHT_DEFAULT; | |
66ebf6c0 | 903 | |
be2c0327 LP |
904 | cgroup_apply_unified_cpu_weight(u, weight); |
905 | cgroup_apply_unified_cpu_quota(u, c->cpu_quota_per_sec_usec); | |
66ebf6c0 | 906 | |
52fecf20 | 907 | } else { |
be2c0327 | 908 | uint64_t shares; |
52fecf20 | 909 | |
be2c0327 LP |
910 | if (cgroup_context_has_cpu_weight(c)) { |
911 | uint64_t weight; | |
52fecf20 | 912 | |
be2c0327 LP |
913 | weight = cgroup_context_cpu_weight(c, state); |
914 | shares = cgroup_cpu_weight_to_shares(weight); | |
52fecf20 | 915 | |
be2c0327 LP |
916 | log_cgroup_compat(u, "Applying [Startup]CPUWeight=%" PRIu64 " as [Startup]CPUShares=%" PRIu64 " on %s", |
917 | weight, shares, path); | |
918 | } else if (cgroup_context_has_cpu_shares(c)) | |
919 | shares = cgroup_context_cpu_shares(c, state); | |
920 | else | |
921 | shares = CGROUP_CPU_SHARES_DEFAULT; | |
66ebf6c0 | 922 | |
be2c0327 LP |
923 | cgroup_apply_legacy_cpu_shares(u, shares); |
924 | cgroup_apply_legacy_cpu_quota(u, c->cpu_quota_per_sec_usec); | |
66ebf6c0 | 925 | } |
4ad49000 LP |
926 | } |
927 | ||
52fecf20 LP |
928 | /* The 'io' controller attributes are not exported on the host's root cgroup (being a pure cgroupsv2 |
929 | * controller), and in case of containers we want to leave control of these attributes to the container manager | |
930 | * (and we couldn't access that stuff anyway, even if we tried if proper delegation is used). */ | |
931 | if ((apply_mask & CGROUP_MASK_IO) && !is_local_root) { | |
932 | char buf[8+DECIMAL_STR_MAX(uint64_t)+1]; | |
933 | bool has_io, has_blockio; | |
934 | uint64_t weight; | |
13c31542 | 935 | |
52fecf20 LP |
936 | has_io = cgroup_context_has_io_config(c); |
937 | has_blockio = cgroup_context_has_blockio_config(c); | |
13c31542 | 938 | |
52fecf20 LP |
939 | if (has_io) |
940 | weight = cgroup_context_io_weight(c, state); | |
941 | else if (has_blockio) { | |
942 | uint64_t blkio_weight; | |
128fadc9 | 943 | |
52fecf20 LP |
944 | blkio_weight = cgroup_context_blkio_weight(c, state); |
945 | weight = cgroup_weight_blkio_to_io(blkio_weight); | |
128fadc9 | 946 | |
67e2ea15 | 947 | log_cgroup_compat(u, "Applying [Startup]BlockIOWeight=%" PRIu64 " as [Startup]IOWeight=%" PRIu64, |
52fecf20 LP |
948 | blkio_weight, weight); |
949 | } else | |
950 | weight = CGROUP_WEIGHT_DEFAULT; | |
13c31542 | 951 | |
52fecf20 LP |
952 | xsprintf(buf, "default %" PRIu64 "\n", weight); |
953 | (void) set_attribute_and_warn(u, "io", "io.weight", buf); | |
538b4852 | 954 | |
52fecf20 LP |
955 | if (has_io) { |
956 | CGroupIODeviceLatency *latency; | |
957 | CGroupIODeviceLimit *limit; | |
958 | CGroupIODeviceWeight *w; | |
128fadc9 | 959 | |
52fecf20 LP |
960 | LIST_FOREACH(device_weights, w, c->io_device_weights) |
961 | cgroup_apply_io_device_weight(u, w->path, w->weight); | |
128fadc9 | 962 | |
52fecf20 LP |
963 | LIST_FOREACH(device_limits, limit, c->io_device_limits) |
964 | cgroup_apply_io_device_limit(u, limit->path, limit->limits); | |
6ae4283c | 965 | |
52fecf20 LP |
966 | LIST_FOREACH(device_latencies, latency, c->io_device_latencies) |
967 | cgroup_apply_io_device_latency(u, latency->path, latency->target_usec); | |
6ae4283c | 968 | |
52fecf20 LP |
969 | } else if (has_blockio) { |
970 | CGroupBlockIODeviceWeight *w; | |
971 | CGroupBlockIODeviceBandwidth *b; | |
13c31542 | 972 | |
52fecf20 LP |
973 | LIST_FOREACH(device_weights, w, c->blockio_device_weights) { |
974 | weight = cgroup_weight_blkio_to_io(w->weight); | |
17ae2780 | 975 | |
67e2ea15 | 976 | log_cgroup_compat(u, "Applying BlockIODeviceWeight=%" PRIu64 " as IODeviceWeight=%" PRIu64 " for %s", |
52fecf20 | 977 | w->weight, weight, w->path); |
538b4852 | 978 | |
52fecf20 LP |
979 | cgroup_apply_io_device_weight(u, w->path, weight); |
980 | } | |
538b4852 | 981 | |
17ae2780 | 982 | LIST_FOREACH(device_bandwidths, b, c->blockio_device_bandwidths) { |
538b4852 TH |
983 | uint64_t limits[_CGROUP_IO_LIMIT_TYPE_MAX]; |
984 | CGroupIOLimitType type; | |
985 | ||
986 | for (type = 0; type < _CGROUP_IO_LIMIT_TYPE_MAX; type++) | |
987 | limits[type] = cgroup_io_limit_defaults[type]; | |
988 | ||
989 | limits[CGROUP_IO_RBPS_MAX] = b->rbps; | |
990 | limits[CGROUP_IO_WBPS_MAX] = b->wbps; | |
991 | ||
67e2ea15 | 992 | log_cgroup_compat(u, "Applying BlockIO{Read|Write}Bandwidth=%" PRIu64 " %" PRIu64 " as IO{Read|Write}BandwidthMax= for %s", |
128fadc9 TH |
993 | b->rbps, b->wbps, b->path); |
994 | ||
17ae2780 | 995 | cgroup_apply_io_device_limit(u, b->path, limits); |
538b4852 | 996 | } |
13c31542 TH |
997 | } |
998 | } | |
999 | ||
906c06f6 | 1000 | if (apply_mask & CGROUP_MASK_BLKIO) { |
52fecf20 | 1001 | bool has_io, has_blockio; |
4ad49000 | 1002 | |
52fecf20 LP |
1003 | has_io = cgroup_context_has_io_config(c); |
1004 | has_blockio = cgroup_context_has_blockio_config(c); | |
1005 | ||
1006 | /* Applying a 'weight' never makes sense for the host root cgroup, and for containers this should be | |
1007 | * left to our container manager, too. */ | |
1008 | if (!is_local_root) { | |
64faf04c TH |
1009 | char buf[DECIMAL_STR_MAX(uint64_t)+1]; |
1010 | uint64_t weight; | |
64faf04c | 1011 | |
7d862ab8 | 1012 | if (has_io) { |
52fecf20 | 1013 | uint64_t io_weight; |
128fadc9 | 1014 | |
52fecf20 | 1015 | io_weight = cgroup_context_io_weight(c, state); |
538b4852 | 1016 | weight = cgroup_weight_io_to_blkio(cgroup_context_io_weight(c, state)); |
128fadc9 | 1017 | |
67e2ea15 | 1018 | log_cgroup_compat(u, "Applying [Startup]IOWeight=%" PRIu64 " as [Startup]BlockIOWeight=%" PRIu64, |
128fadc9 | 1019 | io_weight, weight); |
7d862ab8 TH |
1020 | } else if (has_blockio) |
1021 | weight = cgroup_context_blkio_weight(c, state); | |
1022 | else | |
538b4852 | 1023 | weight = CGROUP_BLKIO_WEIGHT_DEFAULT; |
64faf04c TH |
1024 | |
1025 | xsprintf(buf, "%" PRIu64 "\n", weight); | |
293d32df | 1026 | (void) set_attribute_and_warn(u, "blkio", "blkio.weight", buf); |
4ad49000 | 1027 | |
7d862ab8 | 1028 | if (has_io) { |
538b4852 TH |
1029 | CGroupIODeviceWeight *w; |
1030 | ||
128fadc9 TH |
1031 | LIST_FOREACH(device_weights, w, c->io_device_weights) { |
1032 | weight = cgroup_weight_io_to_blkio(w->weight); | |
1033 | ||
67e2ea15 | 1034 | log_cgroup_compat(u, "Applying IODeviceWeight=%" PRIu64 " as BlockIODeviceWeight=%" PRIu64 " for %s", |
128fadc9 TH |
1035 | w->weight, weight, w->path); |
1036 | ||
1037 | cgroup_apply_blkio_device_weight(u, w->path, weight); | |
1038 | } | |
7d862ab8 TH |
1039 | } else if (has_blockio) { |
1040 | CGroupBlockIODeviceWeight *w; | |
1041 | ||
7d862ab8 TH |
1042 | LIST_FOREACH(device_weights, w, c->blockio_device_weights) |
1043 | cgroup_apply_blkio_device_weight(u, w->path, w->weight); | |
538b4852 | 1044 | } |
4ad49000 LP |
1045 | } |
1046 | ||
52fecf20 LP |
1047 | /* The bandwith limits are something that make sense to be applied to the host's root but not container |
1048 | * roots, as there we want the container manager to handle it */ | |
1049 | if (is_host_root || !is_local_root) { | |
1050 | if (has_io) { | |
1051 | CGroupIODeviceLimit *l; | |
538b4852 | 1052 | |
52fecf20 | 1053 | LIST_FOREACH(device_limits, l, c->io_device_limits) { |
67e2ea15 | 1054 | log_cgroup_compat(u, "Applying IO{Read|Write}Bandwidth=%" PRIu64 " %" PRIu64 " as BlockIO{Read|Write}BandwidthMax= for %s", |
52fecf20 | 1055 | l->limits[CGROUP_IO_RBPS_MAX], l->limits[CGROUP_IO_WBPS_MAX], l->path); |
128fadc9 | 1056 | |
52fecf20 LP |
1057 | cgroup_apply_blkio_device_limit(u, l->path, l->limits[CGROUP_IO_RBPS_MAX], l->limits[CGROUP_IO_WBPS_MAX]); |
1058 | } | |
1059 | } else if (has_blockio) { | |
1060 | CGroupBlockIODeviceBandwidth *b; | |
7d862ab8 | 1061 | |
52fecf20 LP |
1062 | LIST_FOREACH(device_bandwidths, b, c->blockio_device_bandwidths) |
1063 | cgroup_apply_blkio_device_limit(u, b->path, b->rbps, b->wbps); | |
1064 | } | |
d686d8a9 | 1065 | } |
8e274523 LP |
1066 | } |
1067 | ||
be2c0327 LP |
1068 | /* In unified mode 'memory' attributes do not exist on the root cgroup. In legacy mode 'memory.limit_in_bytes' |
1069 | * exists on the root cgroup, but any writes to it are refused with EINVAL. And if we run in a container we | |
1070 | * want to leave control to the container manager (and if proper cgroupsv2 delegation is used we couldn't even | |
1071 | * write to this if we wanted to.) */ | |
1072 | if ((apply_mask & CGROUP_MASK_MEMORY) && !is_local_root) { | |
efdb0237 | 1073 | |
52fecf20 | 1074 | if (cg_all_unified() > 0) { |
be2c0327 LP |
1075 | uint64_t max, swap_max = CGROUP_LIMIT_MAX; |
1076 | ||
1077 | if (cgroup_context_has_unified_memory_config(c)) { | |
1078 | max = c->memory_max; | |
1079 | swap_max = c->memory_swap_max; | |
1080 | } else { | |
1081 | max = c->memory_limit; | |
efdb0237 | 1082 | |
be2c0327 LP |
1083 | if (max != CGROUP_LIMIT_MAX) |
1084 | log_cgroup_compat(u, "Applying MemoryLimit=%" PRIu64 " as MemoryMax=", max); | |
128fadc9 | 1085 | } |
da4d897e | 1086 | |
be2c0327 LP |
1087 | cgroup_apply_unified_memory_limit(u, "memory.min", c->memory_min); |
1088 | cgroup_apply_unified_memory_limit(u, "memory.low", c->memory_low); | |
1089 | cgroup_apply_unified_memory_limit(u, "memory.high", c->memory_high); | |
1090 | cgroup_apply_unified_memory_limit(u, "memory.max", max); | |
1091 | cgroup_apply_unified_memory_limit(u, "memory.swap.max", swap_max); | |
128fadc9 | 1092 | |
be2c0327 LP |
1093 | } else { |
1094 | char buf[DECIMAL_STR_MAX(uint64_t) + 1]; | |
1095 | uint64_t val; | |
52fecf20 | 1096 | |
be2c0327 LP |
1097 | if (cgroup_context_has_unified_memory_config(c)) { |
1098 | val = c->memory_max; | |
1099 | log_cgroup_compat(u, "Applying MemoryMax=%" PRIi64 " as MemoryLimit=", val); | |
1100 | } else | |
1101 | val = c->memory_limit; | |
78a4ee59 | 1102 | |
be2c0327 LP |
1103 | if (val == CGROUP_LIMIT_MAX) |
1104 | strncpy(buf, "-1\n", sizeof(buf)); | |
1105 | else | |
1106 | xsprintf(buf, "%" PRIu64 "\n", val); | |
1107 | ||
1108 | (void) set_attribute_and_warn(u, "memory", "memory.limit_in_bytes", buf); | |
da4d897e | 1109 | } |
4ad49000 | 1110 | } |
8e274523 | 1111 | |
be2c0327 | 1112 | /* On cgroupsv2 we can apply BPF everywhere. On cgroupsv1 we apply it everywhere except for the root of |
52fecf20 LP |
1113 | * containers, where we leave this to the manager */ |
1114 | if ((apply_mask & (CGROUP_MASK_DEVICES | CGROUP_MASK_BPF_DEVICES)) && | |
1115 | (is_host_root || cg_all_unified() > 0 || !is_local_root)) { | |
084c7007 | 1116 | _cleanup_(bpf_program_unrefp) BPFProgram *prog = NULL; |
4ad49000 | 1117 | CGroupDeviceAllow *a; |
8e274523 | 1118 | |
084c7007 RG |
1119 | if (cg_all_unified() > 0) { |
1120 | r = cgroup_init_device_bpf(&prog, c->device_policy, c->device_allow); | |
1121 | if (r < 0) | |
1122 | log_unit_warning_errno(u, r, "Failed to initialize device control bpf program: %m"); | |
1123 | } else { | |
8c838407 | 1124 | /* Changing the devices list of a populated cgroup might result in EINVAL, hence ignore EINVAL |
084c7007 | 1125 | * here. */ |
714e2e1d | 1126 | |
084c7007 RG |
1127 | if (c->device_allow || c->device_policy != CGROUP_AUTO) |
1128 | r = cg_set_attribute("devices", path, "devices.deny", "a"); | |
1129 | else | |
1130 | r = cg_set_attribute("devices", path, "devices.allow", "a"); | |
1131 | if (r < 0) | |
2c74e12b LP |
1132 | log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EINVAL, -EACCES, -EPERM) ? LOG_DEBUG : LOG_WARNING, r, |
1133 | "Failed to reset devices.allow/devices.deny: %m"); | |
084c7007 | 1134 | } |
fb385181 | 1135 | |
4ad49000 LP |
1136 | if (c->device_policy == CGROUP_CLOSED || |
1137 | (c->device_policy == CGROUP_AUTO && c->device_allow)) { | |
1138 | static const char auto_devices[] = | |
7d711efb LP |
1139 | "/dev/null\0" "rwm\0" |
1140 | "/dev/zero\0" "rwm\0" | |
1141 | "/dev/full\0" "rwm\0" | |
1142 | "/dev/random\0" "rwm\0" | |
1143 | "/dev/urandom\0" "rwm\0" | |
1144 | "/dev/tty\0" "rwm\0" | |
5a7f87a9 | 1145 | "/dev/ptmx\0" "rwm\0" |
0d9e7991 | 1146 | /* Allow /run/systemd/inaccessible/{chr,blk} devices for mapping InaccessiblePaths */ |
74c48bf5 LP |
1147 | "/run/systemd/inaccessible/chr\0" "rwm\0" |
1148 | "/run/systemd/inaccessible/blk\0" "rwm\0"; | |
4ad49000 LP |
1149 | |
1150 | const char *x, *y; | |
1151 | ||
1152 | NULSTR_FOREACH_PAIR(x, y, auto_devices) | |
913c898c | 1153 | (void) whitelist_device(prog, path, x, y); |
7d711efb | 1154 | |
5a7f87a9 | 1155 | /* PTS (/dev/pts) devices may not be duplicated, but accessed */ |
913c898c | 1156 | (void) whitelist_major(prog, path, "pts", 'c', "rw"); |
4ad49000 LP |
1157 | } |
1158 | ||
1159 | LIST_FOREACH(device_allow, a, c->device_allow) { | |
fb4650aa | 1160 | char acc[4], *val; |
4ad49000 LP |
1161 | unsigned k = 0; |
1162 | ||
1163 | if (a->r) | |
1164 | acc[k++] = 'r'; | |
1165 | if (a->w) | |
1166 | acc[k++] = 'w'; | |
1167 | if (a->m) | |
1168 | acc[k++] = 'm'; | |
fb385181 | 1169 | |
4ad49000 LP |
1170 | if (k == 0) |
1171 | continue; | |
fb385181 | 1172 | |
4ad49000 | 1173 | acc[k++] = 0; |
90060676 | 1174 | |
27458ed6 | 1175 | if (path_startswith(a->path, "/dev/")) |
913c898c | 1176 | (void) whitelist_device(prog, path, a->path, acc); |
fb4650aa | 1177 | else if ((val = startswith(a->path, "block-"))) |
913c898c | 1178 | (void) whitelist_major(prog, path, val, 'b', acc); |
fb4650aa | 1179 | else if ((val = startswith(a->path, "char-"))) |
913c898c | 1180 | (void) whitelist_major(prog, path, val, 'c', acc); |
90060676 | 1181 | else |
8e8b5d2e | 1182 | log_unit_debug(u, "Ignoring device '%s' while writing cgroup attribute.", a->path); |
4ad49000 | 1183 | } |
084c7007 RG |
1184 | |
1185 | r = cgroup_apply_device_bpf(u, prog, c->device_policy, c->device_allow); | |
1186 | if (r < 0) { | |
1187 | static bool warned = false; | |
1188 | ||
1189 | log_full_errno(warned ? LOG_DEBUG : LOG_WARNING, r, | |
1190 | "Unit %s configures device ACL, but the local system doesn't seem to support the BPF-based device controller.\n" | |
1191 | "Proceeding WITHOUT applying ACL (all devices will be accessible)!\n" | |
1192 | "(This warning is only shown for the first loaded unit using device ACL.)", u->id); | |
1193 | ||
1194 | warned = true; | |
1195 | } | |
4ad49000 | 1196 | } |
03a7b521 | 1197 | |
00b5974f LP |
1198 | if (apply_mask & CGROUP_MASK_PIDS) { |
1199 | ||
52fecf20 | 1200 | if (is_host_root) { |
00b5974f LP |
1201 | /* So, the "pids" controller does not expose anything on the root cgroup, in order not to |
1202 | * replicate knobs exposed elsewhere needlessly. We abstract this away here however, and when | |
1203 | * the knobs of the root cgroup are modified propagate this to the relevant sysctls. There's a | |
1204 | * non-obvious asymmetry however: unlike the cgroup properties we don't really want to take | |
1205 | * exclusive ownership of the sysctls, but we still want to honour things if the user sets | |
1206 | * limits. Hence we employ sort of a one-way strategy: when the user sets a bounded limit | |
1207 | * through us it counts. When the user afterwards unsets it again (i.e. sets it to unbounded) | |
1208 | * it also counts. But if the user never set a limit through us (i.e. we are the default of | |
1209 | * "unbounded") we leave things unmodified. For this we manage a global boolean that we turn on | |
1210 | * the first time we set a limit. Note that this boolean is flushed out on manager reload, | |
1211 | * which is desirable so that there's an offical way to release control of the sysctl from | |
1212 | * systemd: set the limit to unbounded and reload. */ | |
1213 | ||
1214 | if (c->tasks_max != CGROUP_LIMIT_MAX) { | |
1215 | u->manager->sysctl_pid_max_changed = true; | |
1216 | r = procfs_tasks_set_limit(c->tasks_max); | |
1217 | } else if (u->manager->sysctl_pid_max_changed) | |
1218 | r = procfs_tasks_set_limit(TASKS_MAX); | |
1219 | else | |
1220 | r = 0; | |
00b5974f | 1221 | if (r < 0) |
39b9fefb | 1222 | log_unit_full(u, LOG_LEVEL_CGROUP_WRITE(r), r, |
00b5974f | 1223 | "Failed to write to tasks limit sysctls: %m"); |
52fecf20 | 1224 | } |
03a7b521 | 1225 | |
52fecf20 LP |
1226 | /* The attribute itself is not available on the host root cgroup, and in the container case we want to |
1227 | * leave it for the container manager. */ | |
1228 | if (!is_local_root) { | |
00b5974f LP |
1229 | if (c->tasks_max != CGROUP_LIMIT_MAX) { |
1230 | char buf[DECIMAL_STR_MAX(uint64_t) + 2]; | |
03a7b521 | 1231 | |
00b5974f | 1232 | sprintf(buf, "%" PRIu64 "\n", c->tasks_max); |
293d32df | 1233 | (void) set_attribute_and_warn(u, "pids", "pids.max", buf); |
00b5974f | 1234 | } else |
589a5f7a | 1235 | (void) set_attribute_and_warn(u, "pids", "pids.max", "max\n"); |
00b5974f | 1236 | } |
03a7b521 | 1237 | } |
906c06f6 | 1238 | |
17f14955 | 1239 | if (apply_mask & CGROUP_MASK_BPF_FIREWALL) |
0f2d84d2 | 1240 | cgroup_apply_firewall(u); |
fb385181 LP |
1241 | } |
1242 | ||
16492445 LP |
1243 | static bool unit_get_needs_bpf_firewall(Unit *u) { |
1244 | CGroupContext *c; | |
1245 | Unit *p; | |
1246 | assert(u); | |
1247 | ||
1248 | c = unit_get_cgroup_context(u); | |
1249 | if (!c) | |
1250 | return false; | |
1251 | ||
1252 | if (c->ip_accounting || | |
1253 | c->ip_address_allow || | |
1254 | c->ip_address_deny) | |
1255 | return true; | |
1256 | ||
1257 | /* If any parent slice has an IP access list defined, it applies too */ | |
1258 | for (p = UNIT_DEREF(u->slice); p; p = UNIT_DEREF(p->slice)) { | |
1259 | c = unit_get_cgroup_context(p); | |
1260 | if (!c) | |
1261 | return false; | |
1262 | ||
1263 | if (c->ip_address_allow || | |
1264 | c->ip_address_deny) | |
1265 | return true; | |
1266 | } | |
1267 | ||
1268 | return false; | |
1269 | } | |
1270 | ||
53aea74a | 1271 | static CGroupMask cgroup_context_get_mask(CGroupContext *c) { |
efdb0237 | 1272 | CGroupMask mask = 0; |
8e274523 | 1273 | |
fae9bc29 | 1274 | /* Figure out which controllers we need, based on the cgroup context object */ |
8e274523 | 1275 | |
fae9bc29 | 1276 | if (c->cpu_accounting) |
f98c2585 | 1277 | mask |= get_cpu_accounting_mask(); |
fae9bc29 LP |
1278 | |
1279 | if (cgroup_context_has_cpu_weight(c) || | |
66ebf6c0 | 1280 | cgroup_context_has_cpu_shares(c) || |
3a43da28 | 1281 | c->cpu_quota_per_sec_usec != USEC_INFINITY) |
fae9bc29 | 1282 | mask |= CGROUP_MASK_CPU; |
ecedd90f | 1283 | |
538b4852 TH |
1284 | if (cgroup_context_has_io_config(c) || cgroup_context_has_blockio_config(c)) |
1285 | mask |= CGROUP_MASK_IO | CGROUP_MASK_BLKIO; | |
ecedd90f | 1286 | |
4ad49000 | 1287 | if (c->memory_accounting || |
da4d897e TH |
1288 | c->memory_limit != CGROUP_LIMIT_MAX || |
1289 | cgroup_context_has_unified_memory_config(c)) | |
efdb0237 | 1290 | mask |= CGROUP_MASK_MEMORY; |
8e274523 | 1291 | |
a931ad47 LP |
1292 | if (c->device_allow || |
1293 | c->device_policy != CGROUP_AUTO) | |
084c7007 | 1294 | mask |= CGROUP_MASK_DEVICES | CGROUP_MASK_BPF_DEVICES; |
4ad49000 | 1295 | |
03a7b521 | 1296 | if (c->tasks_accounting || |
8793fa25 | 1297 | c->tasks_max != CGROUP_LIMIT_MAX) |
03a7b521 LP |
1298 | mask |= CGROUP_MASK_PIDS; |
1299 | ||
fae9bc29 | 1300 | return CGROUP_MASK_EXTEND_JOINED(mask); |
8e274523 LP |
1301 | } |
1302 | ||
53aea74a | 1303 | static CGroupMask unit_get_bpf_mask(Unit *u) { |
17f14955 RG |
1304 | CGroupMask mask = 0; |
1305 | ||
fae9bc29 LP |
1306 | /* Figure out which controllers we need, based on the cgroup context, possibly taking into account children |
1307 | * too. */ | |
1308 | ||
17f14955 RG |
1309 | if (unit_get_needs_bpf_firewall(u)) |
1310 | mask |= CGROUP_MASK_BPF_FIREWALL; | |
1311 | ||
1312 | return mask; | |
1313 | } | |
1314 | ||
efdb0237 | 1315 | CGroupMask unit_get_own_mask(Unit *u) { |
4ad49000 | 1316 | CGroupContext *c; |
8e274523 | 1317 | |
442ce775 LP |
1318 | /* Returns the mask of controllers the unit needs for itself. If a unit is not properly loaded, return an empty |
1319 | * mask, as we shouldn't reflect it in the cgroup hierarchy then. */ | |
1320 | ||
1321 | if (u->load_state != UNIT_LOADED) | |
1322 | return 0; | |
efdb0237 | 1323 | |
4ad49000 LP |
1324 | c = unit_get_cgroup_context(u); |
1325 | if (!c) | |
1326 | return 0; | |
8e274523 | 1327 | |
c72703e2 | 1328 | return (cgroup_context_get_mask(c) | unit_get_bpf_mask(u) | unit_get_delegate_mask(u)) & ~unit_get_ancestor_disable_mask(u); |
02638280 LP |
1329 | } |
1330 | ||
1331 | CGroupMask unit_get_delegate_mask(Unit *u) { | |
1332 | CGroupContext *c; | |
1333 | ||
1334 | /* If delegation is turned on, then turn on selected controllers, unless we are on the legacy hierarchy and the | |
1335 | * process we fork into is known to drop privileges, and hence shouldn't get access to the controllers. | |
19af675e | 1336 | * |
02638280 | 1337 | * Note that on the unified hierarchy it is safe to delegate controllers to unprivileged services. */ |
a931ad47 | 1338 | |
1d9cc876 | 1339 | if (!unit_cgroup_delegate(u)) |
02638280 LP |
1340 | return 0; |
1341 | ||
1342 | if (cg_all_unified() <= 0) { | |
a931ad47 LP |
1343 | ExecContext *e; |
1344 | ||
1345 | e = unit_get_exec_context(u); | |
02638280 LP |
1346 | if (e && !exec_context_maintains_privileges(e)) |
1347 | return 0; | |
a931ad47 LP |
1348 | } |
1349 | ||
1d9cc876 | 1350 | assert_se(c = unit_get_cgroup_context(u)); |
fae9bc29 | 1351 | return CGROUP_MASK_EXTEND_JOINED(c->delegate_controllers); |
8e274523 LP |
1352 | } |
1353 | ||
efdb0237 | 1354 | CGroupMask unit_get_members_mask(Unit *u) { |
4ad49000 | 1355 | assert(u); |
bc432dc7 | 1356 | |
02638280 | 1357 | /* Returns the mask of controllers all of the unit's children require, merged */ |
efdb0237 | 1358 | |
bc432dc7 | 1359 | if (u->cgroup_members_mask_valid) |
26a17ca2 | 1360 | return u->cgroup_members_mask; /* Use cached value if possible */ |
bc432dc7 | 1361 | |
64e844e5 | 1362 | u->cgroup_members_mask = 0; |
bc432dc7 LP |
1363 | |
1364 | if (u->type == UNIT_SLICE) { | |
eef85c4a | 1365 | void *v; |
bc432dc7 LP |
1366 | Unit *member; |
1367 | Iterator i; | |
1368 | ||
eef85c4a | 1369 | HASHMAP_FOREACH_KEY(v, member, u->dependencies[UNIT_BEFORE], i) { |
cb5e3bc3 CD |
1370 | if (UNIT_DEREF(member->slice) == u) |
1371 | u->cgroup_members_mask |= unit_get_subtree_mask(member); /* note that this calls ourselves again, for the children */ | |
bc432dc7 LP |
1372 | } |
1373 | } | |
1374 | ||
1375 | u->cgroup_members_mask_valid = true; | |
6414b7c9 | 1376 | return u->cgroup_members_mask; |
246aa6dd LP |
1377 | } |
1378 | ||
efdb0237 | 1379 | CGroupMask unit_get_siblings_mask(Unit *u) { |
4ad49000 | 1380 | assert(u); |
246aa6dd | 1381 | |
efdb0237 LP |
1382 | /* Returns the mask of controllers all of the unit's siblings |
1383 | * require, i.e. the members mask of the unit's parent slice | |
1384 | * if there is one. */ | |
1385 | ||
bc432dc7 | 1386 | if (UNIT_ISSET(u->slice)) |
637f421e | 1387 | return unit_get_members_mask(UNIT_DEREF(u->slice)); |
4ad49000 | 1388 | |
64e844e5 | 1389 | return unit_get_subtree_mask(u); /* we are the top-level slice */ |
246aa6dd LP |
1390 | } |
1391 | ||
4f6f62e4 CD |
1392 | CGroupMask unit_get_disable_mask(Unit *u) { |
1393 | CGroupContext *c; | |
1394 | ||
1395 | c = unit_get_cgroup_context(u); | |
1396 | if (!c) | |
1397 | return 0; | |
1398 | ||
1399 | return c->disable_controllers; | |
1400 | } | |
1401 | ||
1402 | CGroupMask unit_get_ancestor_disable_mask(Unit *u) { | |
1403 | CGroupMask mask; | |
1404 | ||
1405 | assert(u); | |
1406 | mask = unit_get_disable_mask(u); | |
1407 | ||
1408 | /* Returns the mask of controllers which are marked as forcibly | |
1409 | * disabled in any ancestor unit or the unit in question. */ | |
1410 | ||
1411 | if (UNIT_ISSET(u->slice)) | |
1412 | mask |= unit_get_ancestor_disable_mask(UNIT_DEREF(u->slice)); | |
1413 | ||
1414 | return mask; | |
1415 | } | |
1416 | ||
efdb0237 LP |
1417 | CGroupMask unit_get_subtree_mask(Unit *u) { |
1418 | ||
1419 | /* Returns the mask of this subtree, meaning of the group | |
1420 | * itself and its children. */ | |
1421 | ||
1422 | return unit_get_own_mask(u) | unit_get_members_mask(u); | |
1423 | } | |
1424 | ||
1425 | CGroupMask unit_get_target_mask(Unit *u) { | |
1426 | CGroupMask mask; | |
1427 | ||
1428 | /* This returns the cgroup mask of all controllers to enable | |
1429 | * for a specific cgroup, i.e. everything it needs itself, | |
1430 | * plus all that its children need, plus all that its siblings | |
1431 | * need. This is primarily useful on the legacy cgroup | |
1432 | * hierarchy, where we need to duplicate each cgroup in each | |
1433 | * hierarchy that shall be enabled for it. */ | |
6414b7c9 | 1434 | |
efdb0237 LP |
1435 | mask = unit_get_own_mask(u) | unit_get_members_mask(u) | unit_get_siblings_mask(u); |
1436 | mask &= u->manager->cgroup_supported; | |
c72703e2 | 1437 | mask &= ~unit_get_ancestor_disable_mask(u); |
efdb0237 LP |
1438 | |
1439 | return mask; | |
1440 | } | |
1441 | ||
1442 | CGroupMask unit_get_enable_mask(Unit *u) { | |
1443 | CGroupMask mask; | |
1444 | ||
1445 | /* This returns the cgroup mask of all controllers to enable | |
1446 | * for the children of a specific cgroup. This is primarily | |
1447 | * useful for the unified cgroup hierarchy, where each cgroup | |
1448 | * controls which controllers are enabled for its children. */ | |
1449 | ||
1450 | mask = unit_get_members_mask(u); | |
6414b7c9 | 1451 | mask &= u->manager->cgroup_supported; |
c72703e2 | 1452 | mask &= ~unit_get_ancestor_disable_mask(u); |
6414b7c9 DS |
1453 | |
1454 | return mask; | |
1455 | } | |
1456 | ||
5af88058 | 1457 | void unit_invalidate_cgroup_members_masks(Unit *u) { |
bc432dc7 LP |
1458 | assert(u); |
1459 | ||
5af88058 LP |
1460 | /* Recurse invalidate the member masks cache all the way up the tree */ |
1461 | u->cgroup_members_mask_valid = false; | |
bc432dc7 | 1462 | |
5af88058 LP |
1463 | if (UNIT_ISSET(u->slice)) |
1464 | unit_invalidate_cgroup_members_masks(UNIT_DEREF(u->slice)); | |
6414b7c9 DS |
1465 | } |
1466 | ||
6592b975 | 1467 | const char *unit_get_realized_cgroup_path(Unit *u, CGroupMask mask) { |
03b90d4b | 1468 | |
6592b975 | 1469 | /* Returns the realized cgroup path of the specified unit where all specified controllers are available. */ |
03b90d4b LP |
1470 | |
1471 | while (u) { | |
6592b975 | 1472 | |
03b90d4b LP |
1473 | if (u->cgroup_path && |
1474 | u->cgroup_realized && | |
d94a24ca | 1475 | FLAGS_SET(u->cgroup_realized_mask, mask)) |
03b90d4b LP |
1476 | return u->cgroup_path; |
1477 | ||
1478 | u = UNIT_DEREF(u->slice); | |
1479 | } | |
1480 | ||
1481 | return NULL; | |
1482 | } | |
1483 | ||
6592b975 LP |
1484 | static const char *migrate_callback(CGroupMask mask, void *userdata) { |
1485 | return unit_get_realized_cgroup_path(userdata, mask); | |
1486 | } | |
1487 | ||
efdb0237 LP |
1488 | char *unit_default_cgroup_path(Unit *u) { |
1489 | _cleanup_free_ char *escaped = NULL, *slice = NULL; | |
1490 | int r; | |
1491 | ||
1492 | assert(u); | |
1493 | ||
1494 | if (unit_has_name(u, SPECIAL_ROOT_SLICE)) | |
1495 | return strdup(u->manager->cgroup_root); | |
1496 | ||
1497 | if (UNIT_ISSET(u->slice) && !unit_has_name(UNIT_DEREF(u->slice), SPECIAL_ROOT_SLICE)) { | |
1498 | r = cg_slice_to_path(UNIT_DEREF(u->slice)->id, &slice); | |
1499 | if (r < 0) | |
1500 | return NULL; | |
1501 | } | |
1502 | ||
1503 | escaped = cg_escape(u->id); | |
1504 | if (!escaped) | |
1505 | return NULL; | |
1506 | ||
1507 | if (slice) | |
605405c6 ZJS |
1508 | return strjoin(u->manager->cgroup_root, "/", slice, "/", |
1509 | escaped); | |
efdb0237 | 1510 | else |
605405c6 | 1511 | return strjoin(u->manager->cgroup_root, "/", escaped); |
efdb0237 LP |
1512 | } |
1513 | ||
1514 | int unit_set_cgroup_path(Unit *u, const char *path) { | |
1515 | _cleanup_free_ char *p = NULL; | |
1516 | int r; | |
1517 | ||
1518 | assert(u); | |
1519 | ||
1520 | if (path) { | |
1521 | p = strdup(path); | |
1522 | if (!p) | |
1523 | return -ENOMEM; | |
1524 | } else | |
1525 | p = NULL; | |
1526 | ||
1527 | if (streq_ptr(u->cgroup_path, p)) | |
1528 | return 0; | |
1529 | ||
1530 | if (p) { | |
1531 | r = hashmap_put(u->manager->cgroup_unit, p, u); | |
1532 | if (r < 0) | |
1533 | return r; | |
1534 | } | |
1535 | ||
1536 | unit_release_cgroup(u); | |
1537 | ||
ae2a15bc | 1538 | u->cgroup_path = TAKE_PTR(p); |
efdb0237 LP |
1539 | |
1540 | return 1; | |
1541 | } | |
1542 | ||
1543 | int unit_watch_cgroup(Unit *u) { | |
ab2c3861 | 1544 | _cleanup_free_ char *events = NULL; |
efdb0237 LP |
1545 | int r; |
1546 | ||
1547 | assert(u); | |
1548 | ||
1549 | if (!u->cgroup_path) | |
1550 | return 0; | |
1551 | ||
1552 | if (u->cgroup_inotify_wd >= 0) | |
1553 | return 0; | |
1554 | ||
1555 | /* Only applies to the unified hierarchy */ | |
c22800e4 | 1556 | r = cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER); |
b4cccbc1 LP |
1557 | if (r < 0) |
1558 | return log_error_errno(r, "Failed to determine whether the name=systemd hierarchy is unified: %m"); | |
1559 | if (r == 0) | |
efdb0237 LP |
1560 | return 0; |
1561 | ||
1562 | /* Don't watch the root slice, it's pointless. */ | |
1563 | if (unit_has_name(u, SPECIAL_ROOT_SLICE)) | |
1564 | return 0; | |
1565 | ||
1566 | r = hashmap_ensure_allocated(&u->manager->cgroup_inotify_wd_unit, &trivial_hash_ops); | |
1567 | if (r < 0) | |
1568 | return log_oom(); | |
1569 | ||
ab2c3861 | 1570 | r = cg_get_path(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path, "cgroup.events", &events); |
efdb0237 LP |
1571 | if (r < 0) |
1572 | return log_oom(); | |
1573 | ||
ab2c3861 | 1574 | u->cgroup_inotify_wd = inotify_add_watch(u->manager->cgroup_inotify_fd, events, IN_MODIFY); |
efdb0237 LP |
1575 | if (u->cgroup_inotify_wd < 0) { |
1576 | ||
1577 | if (errno == ENOENT) /* If the directory is already | |
1578 | * gone we don't need to track | |
1579 | * it, so this is not an error */ | |
1580 | return 0; | |
1581 | ||
1582 | return log_unit_error_errno(u, errno, "Failed to add inotify watch descriptor for control group %s: %m", u->cgroup_path); | |
1583 | } | |
1584 | ||
1585 | r = hashmap_put(u->manager->cgroup_inotify_wd_unit, INT_TO_PTR(u->cgroup_inotify_wd), u); | |
1586 | if (r < 0) | |
1587 | return log_unit_error_errno(u, r, "Failed to add inotify watch descriptor to hash map: %m"); | |
1588 | ||
1589 | return 0; | |
1590 | } | |
1591 | ||
a4634b21 LP |
1592 | int unit_pick_cgroup_path(Unit *u) { |
1593 | _cleanup_free_ char *path = NULL; | |
1594 | int r; | |
1595 | ||
1596 | assert(u); | |
1597 | ||
1598 | if (u->cgroup_path) | |
1599 | return 0; | |
1600 | ||
1601 | if (!UNIT_HAS_CGROUP_CONTEXT(u)) | |
1602 | return -EINVAL; | |
1603 | ||
1604 | path = unit_default_cgroup_path(u); | |
1605 | if (!path) | |
1606 | return log_oom(); | |
1607 | ||
1608 | r = unit_set_cgroup_path(u, path); | |
1609 | if (r == -EEXIST) | |
1610 | return log_unit_error_errno(u, r, "Control group %s exists already.", path); | |
1611 | if (r < 0) | |
1612 | return log_unit_error_errno(u, r, "Failed to set unit's control group path to %s: %m", path); | |
1613 | ||
1614 | return 0; | |
1615 | } | |
1616 | ||
efdb0237 LP |
1617 | static int unit_create_cgroup( |
1618 | Unit *u, | |
1619 | CGroupMask target_mask, | |
0d2d6fbf CD |
1620 | CGroupMask enable_mask, |
1621 | ManagerState state) { | |
efdb0237 | 1622 | |
65be7e06 | 1623 | bool created; |
27adcc97 | 1624 | int r; |
64747e2d | 1625 | |
4ad49000 | 1626 | assert(u); |
64747e2d | 1627 | |
27c4ed79 | 1628 | if (!UNIT_HAS_CGROUP_CONTEXT(u)) |
0cd385d3 LP |
1629 | return 0; |
1630 | ||
a4634b21 LP |
1631 | /* Figure out our cgroup path */ |
1632 | r = unit_pick_cgroup_path(u); | |
1633 | if (r < 0) | |
1634 | return r; | |
b58b8e11 | 1635 | |
03b90d4b | 1636 | /* First, create our own group */ |
efdb0237 | 1637 | r = cg_create_everywhere(u->manager->cgroup_supported, target_mask, u->cgroup_path); |
23bbb0de | 1638 | if (r < 0) |
efdb0237 | 1639 | return log_unit_error_errno(u, r, "Failed to create cgroup %s: %m", u->cgroup_path); |
490c5a37 | 1640 | created = r; |
efdb0237 LP |
1641 | |
1642 | /* Start watching it */ | |
1643 | (void) unit_watch_cgroup(u); | |
1644 | ||
65be7e06 | 1645 | /* Preserve enabled controllers in delegated units, adjust others. */ |
1fd3a10c | 1646 | if (created || !u->cgroup_realized || !unit_cgroup_delegate(u)) { |
27adcc97 | 1647 | CGroupMask result_mask = 0; |
65be7e06 ZJS |
1648 | |
1649 | /* Enable all controllers we need */ | |
27adcc97 | 1650 | r = cg_enable_everywhere(u->manager->cgroup_supported, enable_mask, u->cgroup_path, &result_mask); |
65be7e06 | 1651 | if (r < 0) |
27adcc97 LP |
1652 | log_unit_warning_errno(u, r, "Failed to enable/disable controllers on cgroup %s, ignoring: %m", u->cgroup_path); |
1653 | ||
1654 | /* If we just turned off a controller, this might release the controller for our parent too, let's | |
1655 | * enqueue the parent for re-realization in that case again. */ | |
1656 | if (UNIT_ISSET(u->slice)) { | |
1657 | CGroupMask turned_off; | |
1658 | ||
1659 | turned_off = (u->cgroup_realized ? u->cgroup_enabled_mask & ~result_mask : 0); | |
1660 | if (turned_off != 0) { | |
1661 | Unit *parent; | |
1662 | ||
1663 | /* Force the parent to propagate the enable mask to the kernel again, by invalidating | |
1664 | * the controller we just turned off. */ | |
1665 | ||
1666 | for (parent = UNIT_DEREF(u->slice); parent; parent = UNIT_DEREF(parent->slice)) | |
1667 | unit_invalidate_cgroup(parent, turned_off); | |
1668 | } | |
1669 | } | |
1670 | ||
1671 | /* Remember what's actually enabled now */ | |
1672 | u->cgroup_enabled_mask = result_mask; | |
65be7e06 | 1673 | } |
03b90d4b LP |
1674 | |
1675 | /* Keep track that this is now realized */ | |
4ad49000 | 1676 | u->cgroup_realized = true; |
efdb0237 | 1677 | u->cgroup_realized_mask = target_mask; |
4ad49000 | 1678 | |
1d9cc876 | 1679 | if (u->type != UNIT_SLICE && !unit_cgroup_delegate(u)) { |
0cd385d3 LP |
1680 | |
1681 | /* Then, possibly move things over, but not if | |
1682 | * subgroups may contain processes, which is the case | |
1683 | * for slice and delegation units. */ | |
1684 | r = cg_migrate_everywhere(u->manager->cgroup_supported, u->cgroup_path, u->cgroup_path, migrate_callback, u); | |
1685 | if (r < 0) | |
efdb0237 | 1686 | log_unit_warning_errno(u, r, "Failed to migrate cgroup from to %s, ignoring: %m", u->cgroup_path); |
0cd385d3 | 1687 | } |
03b90d4b | 1688 | |
0d2d6fbf CD |
1689 | /* Set attributes */ |
1690 | cgroup_context_apply(u, target_mask, state); | |
1691 | cgroup_xattr_apply(u); | |
1692 | ||
64747e2d LP |
1693 | return 0; |
1694 | } | |
1695 | ||
6592b975 LP |
1696 | static int unit_attach_pid_to_cgroup_via_bus(Unit *u, pid_t pid, const char *suffix_path) { |
1697 | _cleanup_(sd_bus_error_free) sd_bus_error error = SD_BUS_ERROR_NULL; | |
1698 | char *pp; | |
7b3fd631 | 1699 | int r; |
6592b975 | 1700 | |
7b3fd631 LP |
1701 | assert(u); |
1702 | ||
6592b975 LP |
1703 | if (MANAGER_IS_SYSTEM(u->manager)) |
1704 | return -EINVAL; | |
1705 | ||
1706 | if (!u->manager->system_bus) | |
1707 | return -EIO; | |
1708 | ||
1709 | if (!u->cgroup_path) | |
1710 | return -EINVAL; | |
1711 | ||
1712 | /* Determine this unit's cgroup path relative to our cgroup root */ | |
1713 | pp = path_startswith(u->cgroup_path, u->manager->cgroup_root); | |
1714 | if (!pp) | |
1715 | return -EINVAL; | |
1716 | ||
1717 | pp = strjoina("/", pp, suffix_path); | |
858d36c1 | 1718 | path_simplify(pp, false); |
6592b975 LP |
1719 | |
1720 | r = sd_bus_call_method(u->manager->system_bus, | |
1721 | "org.freedesktop.systemd1", | |
1722 | "/org/freedesktop/systemd1", | |
1723 | "org.freedesktop.systemd1.Manager", | |
1724 | "AttachProcessesToUnit", | |
1725 | &error, NULL, | |
1726 | "ssau", | |
1727 | NULL /* empty unit name means client's unit, i.e. us */, pp, 1, (uint32_t) pid); | |
7b3fd631 | 1728 | if (r < 0) |
6592b975 LP |
1729 | return log_unit_debug_errno(u, r, "Failed to attach unit process " PID_FMT " via the bus: %s", pid, bus_error_message(&error, r)); |
1730 | ||
1731 | return 0; | |
1732 | } | |
1733 | ||
1734 | int unit_attach_pids_to_cgroup(Unit *u, Set *pids, const char *suffix_path) { | |
1735 | CGroupMask delegated_mask; | |
1736 | const char *p; | |
1737 | Iterator i; | |
1738 | void *pidp; | |
1739 | int r, q; | |
1740 | ||
1741 | assert(u); | |
1742 | ||
1743 | if (!UNIT_HAS_CGROUP_CONTEXT(u)) | |
1744 | return -EINVAL; | |
1745 | ||
1746 | if (set_isempty(pids)) | |
1747 | return 0; | |
7b3fd631 | 1748 | |
6592b975 | 1749 | r = unit_realize_cgroup(u); |
7b3fd631 LP |
1750 | if (r < 0) |
1751 | return r; | |
1752 | ||
6592b975 LP |
1753 | if (isempty(suffix_path)) |
1754 | p = u->cgroup_path; | |
1755 | else | |
1756 | p = strjoina(u->cgroup_path, "/", suffix_path); | |
1757 | ||
1758 | delegated_mask = unit_get_delegate_mask(u); | |
1759 | ||
1760 | r = 0; | |
1761 | SET_FOREACH(pidp, pids, i) { | |
1762 | pid_t pid = PTR_TO_PID(pidp); | |
1763 | CGroupController c; | |
1764 | ||
1765 | /* First, attach the PID to the main cgroup hierarchy */ | |
1766 | q = cg_attach(SYSTEMD_CGROUP_CONTROLLER, p, pid); | |
1767 | if (q < 0) { | |
1768 | log_unit_debug_errno(u, q, "Couldn't move process " PID_FMT " to requested cgroup '%s': %m", pid, p); | |
1769 | ||
1770 | if (MANAGER_IS_USER(u->manager) && IN_SET(q, -EPERM, -EACCES)) { | |
1771 | int z; | |
1772 | ||
1773 | /* If we are in a user instance, and we can't move the process ourselves due to | |
1774 | * permission problems, let's ask the system instance about it instead. Since it's more | |
1775 | * privileged it might be able to move the process across the leaves of a subtree who's | |
1776 | * top node is not owned by us. */ | |
1777 | ||
1778 | z = unit_attach_pid_to_cgroup_via_bus(u, pid, suffix_path); | |
1779 | if (z < 0) | |
1780 | log_unit_debug_errno(u, z, "Couldn't move process " PID_FMT " to requested cgroup '%s' via the system bus either: %m", pid, p); | |
1781 | else | |
1782 | continue; /* When the bus thing worked via the bus we are fully done for this PID. */ | |
1783 | } | |
1784 | ||
1785 | if (r >= 0) | |
1786 | r = q; /* Remember first error */ | |
1787 | ||
1788 | continue; | |
1789 | } | |
1790 | ||
1791 | q = cg_all_unified(); | |
1792 | if (q < 0) | |
1793 | return q; | |
1794 | if (q > 0) | |
1795 | continue; | |
1796 | ||
1797 | /* In the legacy hierarchy, attach the process to the request cgroup if possible, and if not to the | |
1798 | * innermost realized one */ | |
1799 | ||
1800 | for (c = 0; c < _CGROUP_CONTROLLER_MAX; c++) { | |
1801 | CGroupMask bit = CGROUP_CONTROLLER_TO_MASK(c); | |
1802 | const char *realized; | |
1803 | ||
1804 | if (!(u->manager->cgroup_supported & bit)) | |
1805 | continue; | |
1806 | ||
1807 | /* If this controller is delegated and realized, honour the caller's request for the cgroup suffix. */ | |
1808 | if (delegated_mask & u->cgroup_realized_mask & bit) { | |
1809 | q = cg_attach(cgroup_controller_to_string(c), p, pid); | |
1810 | if (q >= 0) | |
1811 | continue; /* Success! */ | |
1812 | ||
1813 | 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", | |
1814 | pid, p, cgroup_controller_to_string(c)); | |
1815 | } | |
1816 | ||
1817 | /* So this controller is either not delegate or realized, or something else weird happened. In | |
1818 | * that case let's attach the PID at least to the closest cgroup up the tree that is | |
1819 | * realized. */ | |
1820 | realized = unit_get_realized_cgroup_path(u, bit); | |
1821 | if (!realized) | |
1822 | continue; /* Not even realized in the root slice? Then let's not bother */ | |
1823 | ||
1824 | q = cg_attach(cgroup_controller_to_string(c), realized, pid); | |
1825 | if (q < 0) | |
1826 | log_unit_debug_errno(u, q, "Failed to attach PID " PID_FMT " to realized cgroup %s in controller %s, ignoring: %m", | |
1827 | pid, realized, cgroup_controller_to_string(c)); | |
1828 | } | |
1829 | } | |
1830 | ||
1831 | return r; | |
7b3fd631 LP |
1832 | } |
1833 | ||
906c06f6 DM |
1834 | static bool unit_has_mask_realized( |
1835 | Unit *u, | |
1836 | CGroupMask target_mask, | |
17f14955 | 1837 | CGroupMask enable_mask) { |
906c06f6 | 1838 | |
bc432dc7 LP |
1839 | assert(u); |
1840 | ||
d5095dcd LP |
1841 | /* Returns true if this unit is fully realized. We check four things: |
1842 | * | |
1843 | * 1. Whether the cgroup was created at all | |
1844 | * 2. Whether the cgroup was created in all the hierarchies we need it to be created in (in case of cgroupsv1) | |
1845 | * 3. Whether the cgroup has all the right controllers enabled (in case of cgroupsv2) | |
1846 | * 4. Whether the invalidation mask is currently zero | |
1847 | * | |
1848 | * If you wonder why we mask the target realization and enable mask with CGROUP_MASK_V1/CGROUP_MASK_V2: note | |
1849 | * that there are three sets of bitmasks: CGROUP_MASK_V1 (for real cgroupv1 controllers), CGROUP_MASK_V2 (for | |
1850 | * real cgroupv2 controllers) and CGROUP_MASK_BPF (for BPF-based pseudo-controllers). Now, cgroup_realized_mask | |
1851 | * is only matters for cgroupsv1 controllers, and cgroup_enabled_mask only used for cgroupsv2, and if they | |
1852 | * differ in the others, we don't really care. (After all, the cgroup_enabled_mask tracks with controllers are | |
1853 | * enabled through cgroup.subtree_control, and since the BPF pseudo-controllers don't show up there, they | |
1854 | * simply don't matter. */ | |
1855 | ||
906c06f6 | 1856 | return u->cgroup_realized && |
d5095dcd LP |
1857 | ((u->cgroup_realized_mask ^ target_mask) & CGROUP_MASK_V1) == 0 && |
1858 | ((u->cgroup_enabled_mask ^ enable_mask) & CGROUP_MASK_V2) == 0 && | |
17f14955 | 1859 | u->cgroup_invalidated_mask == 0; |
6414b7c9 DS |
1860 | } |
1861 | ||
4f6f62e4 CD |
1862 | static bool unit_has_mask_disables_realized( |
1863 | Unit *u, | |
1864 | CGroupMask target_mask, | |
1865 | CGroupMask enable_mask) { | |
1866 | ||
1867 | assert(u); | |
1868 | ||
1869 | /* Returns true if all controllers which should be disabled are indeed disabled. | |
1870 | * | |
1871 | * Unlike unit_has_mask_realized, we don't care what was enabled, only that anything we want to remove is | |
1872 | * already removed. */ | |
1873 | ||
1874 | return !u->cgroup_realized || | |
1875 | (FLAGS_SET(u->cgroup_realized_mask, target_mask & CGROUP_MASK_V1) && | |
1876 | FLAGS_SET(u->cgroup_enabled_mask, enable_mask & CGROUP_MASK_V2)); | |
1877 | } | |
1878 | ||
a57669d2 CD |
1879 | static bool unit_has_mask_enables_realized( |
1880 | Unit *u, | |
1881 | CGroupMask target_mask, | |
1882 | CGroupMask enable_mask) { | |
1883 | ||
1884 | assert(u); | |
1885 | ||
1886 | /* Returns true if all controllers which should be enabled are indeed enabled. | |
1887 | * | |
1888 | * Unlike unit_has_mask_realized, we don't care about the controllers that are not present, only that anything | |
1889 | * we want to add is already added. */ | |
1890 | ||
1891 | return u->cgroup_realized && | |
c72703e2 CD |
1892 | ((u->cgroup_realized_mask | target_mask) & CGROUP_MASK_V1) == (u->cgroup_realized_mask & CGROUP_MASK_V1) && |
1893 | ((u->cgroup_enabled_mask | enable_mask) & CGROUP_MASK_V2) == (u->cgroup_enabled_mask & CGROUP_MASK_V2); | |
a57669d2 CD |
1894 | } |
1895 | ||
27adcc97 | 1896 | void unit_add_to_cgroup_realize_queue(Unit *u) { |
2aa57a65 LP |
1897 | assert(u); |
1898 | ||
1899 | if (u->in_cgroup_realize_queue) | |
1900 | return; | |
1901 | ||
1902 | LIST_PREPEND(cgroup_realize_queue, u->manager->cgroup_realize_queue, u); | |
1903 | u->in_cgroup_realize_queue = true; | |
1904 | } | |
1905 | ||
1906 | static void unit_remove_from_cgroup_realize_queue(Unit *u) { | |
1907 | assert(u); | |
1908 | ||
1909 | if (!u->in_cgroup_realize_queue) | |
1910 | return; | |
1911 | ||
1912 | LIST_REMOVE(cgroup_realize_queue, u->manager->cgroup_realize_queue, u); | |
1913 | u->in_cgroup_realize_queue = false; | |
1914 | } | |
1915 | ||
a57669d2 CD |
1916 | /* Controllers can only be enabled breadth-first, from the root of the |
1917 | * hierarchy downwards to the unit in question. */ | |
1918 | static int unit_realize_cgroup_now_enable(Unit *u, ManagerState state) { | |
1919 | CGroupMask target_mask, enable_mask, new_target_mask, new_enable_mask; | |
1920 | int r; | |
1921 | ||
1922 | assert(u); | |
1923 | ||
1924 | /* First go deal with this unit's parent, or we won't be able to enable | |
1925 | * any new controllers at this layer. */ | |
1926 | if (UNIT_ISSET(u->slice)) { | |
1927 | r = unit_realize_cgroup_now_enable(UNIT_DEREF(u->slice), state); | |
1928 | if (r < 0) | |
1929 | return r; | |
1930 | } | |
1931 | ||
1932 | target_mask = unit_get_target_mask(u); | |
1933 | enable_mask = unit_get_enable_mask(u); | |
1934 | ||
1935 | /* We can only enable in this direction, don't try to disable anything. | |
1936 | */ | |
1937 | if (unit_has_mask_enables_realized(u, target_mask, enable_mask)) | |
1938 | return 0; | |
1939 | ||
1940 | new_target_mask = u->cgroup_realized_mask | target_mask; | |
1941 | new_enable_mask = u->cgroup_enabled_mask | enable_mask; | |
1942 | ||
c72703e2 | 1943 | return unit_create_cgroup(u, new_target_mask, new_enable_mask, state); |
a57669d2 CD |
1944 | } |
1945 | ||
4f6f62e4 CD |
1946 | /* Controllers can only be disabled depth-first, from the leaves of the |
1947 | * hierarchy upwards to the unit in question. */ | |
1948 | static int unit_realize_cgroup_now_disable(Unit *u, ManagerState state) { | |
1949 | Iterator i; | |
1950 | Unit *m; | |
1951 | void *v; | |
1952 | ||
1953 | assert(u); | |
1954 | ||
1955 | if (u->type != UNIT_SLICE) | |
1956 | return 0; | |
1957 | ||
1958 | HASHMAP_FOREACH_KEY(v, m, u->dependencies[UNIT_BEFORE], i) { | |
1959 | CGroupMask target_mask, enable_mask, new_target_mask, new_enable_mask; | |
1960 | int r; | |
1961 | ||
1962 | if (UNIT_DEREF(m->slice) != u) | |
1963 | continue; | |
1964 | ||
1965 | /* The cgroup for this unit might not actually be fully | |
1966 | * realised yet, in which case it isn't holding any controllers | |
1967 | * open anyway. */ | |
1968 | if (!m->cgroup_path) | |
1969 | continue; | |
1970 | ||
1971 | /* We must disable those below us first in order to release the | |
1972 | * controller. */ | |
1973 | if (m->type == UNIT_SLICE) | |
1974 | (void) unit_realize_cgroup_now_disable(m, state); | |
1975 | ||
1976 | target_mask = unit_get_target_mask(m); | |
1977 | enable_mask = unit_get_enable_mask(m); | |
1978 | ||
1979 | /* We can only disable in this direction, don't try to enable | |
1980 | * anything. */ | |
1981 | if (unit_has_mask_disables_realized(m, target_mask, enable_mask)) | |
1982 | continue; | |
1983 | ||
1984 | new_target_mask = m->cgroup_realized_mask & target_mask; | |
1985 | new_enable_mask = m->cgroup_enabled_mask & enable_mask; | |
1986 | ||
1987 | r = unit_create_cgroup(m, new_target_mask, new_enable_mask, state); | |
1988 | if (r < 0) | |
1989 | return r; | |
1990 | } | |
1991 | ||
1992 | return 0; | |
1993 | } | |
a57669d2 | 1994 | |
6414b7c9 DS |
1995 | /* Check if necessary controllers and attributes for a unit are in place. |
1996 | * | |
a57669d2 CD |
1997 | * - If so, do nothing. |
1998 | * - If not, create paths, move processes over, and set attributes. | |
1999 | * | |
2000 | * Controllers can only be *enabled* in a breadth-first way, and *disabled* in | |
2001 | * a depth-first way. As such the process looks like this: | |
2002 | * | |
2003 | * Suppose we have a cgroup hierarchy which looks like this: | |
2004 | * | |
2005 | * root | |
2006 | * / \ | |
2007 | * / \ | |
2008 | * / \ | |
2009 | * a b | |
2010 | * / \ / \ | |
2011 | * / \ / \ | |
2012 | * c d e f | |
2013 | * / \ / \ / \ / \ | |
2014 | * h i j k l m n o | |
2015 | * | |
2016 | * 1. We want to realise cgroup "d" now. | |
c72703e2 | 2017 | * 2. cgroup "a" has DisableControllers=cpu in the associated unit. |
a57669d2 CD |
2018 | * 3. cgroup "k" just started requesting the memory controller. |
2019 | * | |
2020 | * To make this work we must do the following in order: | |
2021 | * | |
2022 | * 1. Disable CPU controller in k, j | |
2023 | * 2. Disable CPU controller in d | |
2024 | * 3. Enable memory controller in root | |
2025 | * 4. Enable memory controller in a | |
2026 | * 5. Enable memory controller in d | |
2027 | * 6. Enable memory controller in k | |
2028 | * | |
2029 | * Notice that we need to touch j in one direction, but not the other. We also | |
2030 | * don't go beyond d when disabling -- it's up to "a" to get realized if it | |
2031 | * wants to disable further. The basic rules are therefore: | |
2032 | * | |
2033 | * - If you're disabling something, you need to realise all of the cgroups from | |
2034 | * your recursive descendants to the root. This starts from the leaves. | |
2035 | * - If you're enabling something, you need to realise from the root cgroup | |
2036 | * downwards, but you don't need to iterate your recursive descendants. | |
6414b7c9 DS |
2037 | * |
2038 | * Returns 0 on success and < 0 on failure. */ | |
db785129 | 2039 | static int unit_realize_cgroup_now(Unit *u, ManagerState state) { |
efdb0237 | 2040 | CGroupMask target_mask, enable_mask; |
6414b7c9 | 2041 | int r; |
64747e2d | 2042 | |
4ad49000 | 2043 | assert(u); |
64747e2d | 2044 | |
2aa57a65 | 2045 | unit_remove_from_cgroup_realize_queue(u); |
64747e2d | 2046 | |
efdb0237 | 2047 | target_mask = unit_get_target_mask(u); |
ccf78df1 TH |
2048 | enable_mask = unit_get_enable_mask(u); |
2049 | ||
17f14955 | 2050 | if (unit_has_mask_realized(u, target_mask, enable_mask)) |
0a1eb06d | 2051 | return 0; |
64747e2d | 2052 | |
4f6f62e4 CD |
2053 | /* Disable controllers below us, if there are any */ |
2054 | r = unit_realize_cgroup_now_disable(u, state); | |
2055 | if (r < 0) | |
2056 | return r; | |
2057 | ||
2058 | /* Enable controllers above us, if there are any */ | |
6414b7c9 | 2059 | if (UNIT_ISSET(u->slice)) { |
a57669d2 | 2060 | r = unit_realize_cgroup_now_enable(UNIT_DEREF(u->slice), state); |
6414b7c9 DS |
2061 | if (r < 0) |
2062 | return r; | |
2063 | } | |
4ad49000 | 2064 | |
0d2d6fbf CD |
2065 | /* Now actually deal with the cgroup we were trying to realise and set attributes */ |
2066 | r = unit_create_cgroup(u, target_mask, enable_mask, state); | |
6414b7c9 DS |
2067 | if (r < 0) |
2068 | return r; | |
2069 | ||
c2baf11c LP |
2070 | /* Now, reset the invalidation mask */ |
2071 | u->cgroup_invalidated_mask = 0; | |
6414b7c9 | 2072 | return 0; |
64747e2d LP |
2073 | } |
2074 | ||
91a6073e | 2075 | unsigned manager_dispatch_cgroup_realize_queue(Manager *m) { |
db785129 | 2076 | ManagerState state; |
4ad49000 | 2077 | unsigned n = 0; |
db785129 | 2078 | Unit *i; |
6414b7c9 | 2079 | int r; |
ecedd90f | 2080 | |
91a6073e LP |
2081 | assert(m); |
2082 | ||
db785129 LP |
2083 | state = manager_state(m); |
2084 | ||
91a6073e LP |
2085 | while ((i = m->cgroup_realize_queue)) { |
2086 | assert(i->in_cgroup_realize_queue); | |
ecedd90f | 2087 | |
2aa57a65 LP |
2088 | if (UNIT_IS_INACTIVE_OR_FAILED(unit_active_state(i))) { |
2089 | /* Maybe things changed, and the unit is not actually active anymore? */ | |
2090 | unit_remove_from_cgroup_realize_queue(i); | |
2091 | continue; | |
2092 | } | |
2093 | ||
db785129 | 2094 | r = unit_realize_cgroup_now(i, state); |
6414b7c9 | 2095 | if (r < 0) |
efdb0237 | 2096 | log_warning_errno(r, "Failed to realize cgroups for queued unit %s, ignoring: %m", i->id); |
0a1eb06d | 2097 | |
4ad49000 LP |
2098 | n++; |
2099 | } | |
ecedd90f | 2100 | |
4ad49000 | 2101 | return n; |
8e274523 LP |
2102 | } |
2103 | ||
91a6073e | 2104 | static void unit_add_siblings_to_cgroup_realize_queue(Unit *u) { |
4ad49000 | 2105 | Unit *slice; |
ca949c9d | 2106 | |
4ad49000 LP |
2107 | /* This adds the siblings of the specified unit and the |
2108 | * siblings of all parent units to the cgroup queue. (But | |
2109 | * neither the specified unit itself nor the parents.) */ | |
2110 | ||
2111 | while ((slice = UNIT_DEREF(u->slice))) { | |
2112 | Iterator i; | |
2113 | Unit *m; | |
eef85c4a | 2114 | void *v; |
8f53a7b8 | 2115 | |
eef85c4a | 2116 | HASHMAP_FOREACH_KEY(v, m, u->dependencies[UNIT_BEFORE], i) { |
6414b7c9 DS |
2117 | /* Skip units that have a dependency on the slice |
2118 | * but aren't actually in it. */ | |
4ad49000 | 2119 | if (UNIT_DEREF(m->slice) != slice) |
50159e6a | 2120 | continue; |
8e274523 | 2121 | |
6414b7c9 DS |
2122 | /* No point in doing cgroup application for units |
2123 | * without active processes. */ | |
2124 | if (UNIT_IS_INACTIVE_OR_FAILED(unit_active_state(m))) | |
2125 | continue; | |
2126 | ||
2127 | /* If the unit doesn't need any new controllers | |
2128 | * and has current ones realized, it doesn't need | |
2129 | * any changes. */ | |
906c06f6 DM |
2130 | if (unit_has_mask_realized(m, |
2131 | unit_get_target_mask(m), | |
17f14955 | 2132 | unit_get_enable_mask(m))) |
6414b7c9 DS |
2133 | continue; |
2134 | ||
91a6073e | 2135 | unit_add_to_cgroup_realize_queue(m); |
50159e6a LP |
2136 | } |
2137 | ||
4ad49000 | 2138 | u = slice; |
8e274523 | 2139 | } |
4ad49000 LP |
2140 | } |
2141 | ||
0a1eb06d | 2142 | int unit_realize_cgroup(Unit *u) { |
4ad49000 LP |
2143 | assert(u); |
2144 | ||
35b7ff80 | 2145 | if (!UNIT_HAS_CGROUP_CONTEXT(u)) |
0a1eb06d | 2146 | return 0; |
8e274523 | 2147 | |
4ad49000 LP |
2148 | /* So, here's the deal: when realizing the cgroups for this |
2149 | * unit, we need to first create all parents, but there's more | |
2150 | * actually: for the weight-based controllers we also need to | |
2151 | * make sure that all our siblings (i.e. units that are in the | |
73e231ab | 2152 | * same slice as we are) have cgroups, too. Otherwise, things |
4ad49000 LP |
2153 | * would become very uneven as each of their processes would |
2154 | * get as much resources as all our group together. This call | |
2155 | * will synchronously create the parent cgroups, but will | |
2156 | * defer work on the siblings to the next event loop | |
2157 | * iteration. */ | |
ca949c9d | 2158 | |
4ad49000 | 2159 | /* Add all sibling slices to the cgroup queue. */ |
91a6073e | 2160 | unit_add_siblings_to_cgroup_realize_queue(u); |
4ad49000 | 2161 | |
6414b7c9 | 2162 | /* And realize this one now (and apply the values) */ |
db785129 | 2163 | return unit_realize_cgroup_now(u, manager_state(u->manager)); |
8e274523 LP |
2164 | } |
2165 | ||
efdb0237 LP |
2166 | void unit_release_cgroup(Unit *u) { |
2167 | assert(u); | |
2168 | ||
8a0d5388 LP |
2169 | /* Forgets all cgroup details for this cgroup — but does *not* destroy the cgroup. This is hence OK to call |
2170 | * when we close down everything for reexecution, where we really want to leave the cgroup in place. */ | |
efdb0237 LP |
2171 | |
2172 | if (u->cgroup_path) { | |
2173 | (void) hashmap_remove(u->manager->cgroup_unit, u->cgroup_path); | |
2174 | u->cgroup_path = mfree(u->cgroup_path); | |
2175 | } | |
2176 | ||
2177 | if (u->cgroup_inotify_wd >= 0) { | |
2178 | if (inotify_rm_watch(u->manager->cgroup_inotify_fd, u->cgroup_inotify_wd) < 0) | |
5e1ee764 | 2179 | 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 |
2180 | |
2181 | (void) hashmap_remove(u->manager->cgroup_inotify_wd_unit, INT_TO_PTR(u->cgroup_inotify_wd)); | |
2182 | u->cgroup_inotify_wd = -1; | |
2183 | } | |
2184 | } | |
2185 | ||
2186 | void unit_prune_cgroup(Unit *u) { | |
8e274523 | 2187 | int r; |
efdb0237 | 2188 | bool is_root_slice; |
8e274523 | 2189 | |
4ad49000 | 2190 | assert(u); |
8e274523 | 2191 | |
efdb0237 LP |
2192 | /* Removes the cgroup, if empty and possible, and stops watching it. */ |
2193 | ||
4ad49000 LP |
2194 | if (!u->cgroup_path) |
2195 | return; | |
8e274523 | 2196 | |
fe700f46 LP |
2197 | (void) unit_get_cpu_usage(u, NULL); /* Cache the last CPU usage value before we destroy the cgroup */ |
2198 | ||
efdb0237 LP |
2199 | is_root_slice = unit_has_name(u, SPECIAL_ROOT_SLICE); |
2200 | ||
2201 | r = cg_trim_everywhere(u->manager->cgroup_supported, u->cgroup_path, !is_root_slice); | |
dab5bf85 | 2202 | if (r < 0) { |
f29ff115 | 2203 | log_unit_debug_errno(u, r, "Failed to destroy cgroup %s, ignoring: %m", u->cgroup_path); |
dab5bf85 RL |
2204 | return; |
2205 | } | |
8e274523 | 2206 | |
efdb0237 LP |
2207 | if (is_root_slice) |
2208 | return; | |
2209 | ||
2210 | unit_release_cgroup(u); | |
0a1eb06d | 2211 | |
4ad49000 | 2212 | u->cgroup_realized = false; |
bc432dc7 | 2213 | u->cgroup_realized_mask = 0; |
ccf78df1 | 2214 | u->cgroup_enabled_mask = 0; |
084c7007 RG |
2215 | |
2216 | u->bpf_device_control_installed = bpf_program_unref(u->bpf_device_control_installed); | |
8e274523 LP |
2217 | } |
2218 | ||
efdb0237 | 2219 | int unit_search_main_pid(Unit *u, pid_t *ret) { |
4ad49000 LP |
2220 | _cleanup_fclose_ FILE *f = NULL; |
2221 | pid_t pid = 0, npid, mypid; | |
efdb0237 | 2222 | int r; |
4ad49000 LP |
2223 | |
2224 | assert(u); | |
efdb0237 | 2225 | assert(ret); |
4ad49000 LP |
2226 | |
2227 | if (!u->cgroup_path) | |
efdb0237 | 2228 | return -ENXIO; |
4ad49000 | 2229 | |
efdb0237 LP |
2230 | r = cg_enumerate_processes(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path, &f); |
2231 | if (r < 0) | |
2232 | return r; | |
4ad49000 | 2233 | |
df0ff127 | 2234 | mypid = getpid_cached(); |
4ad49000 LP |
2235 | while (cg_read_pid(f, &npid) > 0) { |
2236 | pid_t ppid; | |
2237 | ||
2238 | if (npid == pid) | |
2239 | continue; | |
8e274523 | 2240 | |
4ad49000 | 2241 | /* Ignore processes that aren't our kids */ |
6bc73acb | 2242 | if (get_process_ppid(npid, &ppid) >= 0 && ppid != mypid) |
4ad49000 | 2243 | continue; |
8e274523 | 2244 | |
efdb0237 | 2245 | if (pid != 0) |
4ad49000 LP |
2246 | /* Dang, there's more than one daemonized PID |
2247 | in this group, so we don't know what process | |
2248 | is the main process. */ | |
efdb0237 LP |
2249 | |
2250 | return -ENODATA; | |
8e274523 | 2251 | |
4ad49000 | 2252 | pid = npid; |
8e274523 LP |
2253 | } |
2254 | ||
efdb0237 LP |
2255 | *ret = pid; |
2256 | return 0; | |
2257 | } | |
2258 | ||
2259 | static int unit_watch_pids_in_path(Unit *u, const char *path) { | |
b3c5bad3 | 2260 | _cleanup_closedir_ DIR *d = NULL; |
efdb0237 LP |
2261 | _cleanup_fclose_ FILE *f = NULL; |
2262 | int ret = 0, r; | |
2263 | ||
2264 | assert(u); | |
2265 | assert(path); | |
2266 | ||
2267 | r = cg_enumerate_processes(SYSTEMD_CGROUP_CONTROLLER, path, &f); | |
2268 | if (r < 0) | |
2269 | ret = r; | |
2270 | else { | |
2271 | pid_t pid; | |
2272 | ||
2273 | while ((r = cg_read_pid(f, &pid)) > 0) { | |
2274 | r = unit_watch_pid(u, pid); | |
2275 | if (r < 0 && ret >= 0) | |
2276 | ret = r; | |
2277 | } | |
2278 | ||
2279 | if (r < 0 && ret >= 0) | |
2280 | ret = r; | |
2281 | } | |
2282 | ||
2283 | r = cg_enumerate_subgroups(SYSTEMD_CGROUP_CONTROLLER, path, &d); | |
2284 | if (r < 0) { | |
2285 | if (ret >= 0) | |
2286 | ret = r; | |
2287 | } else { | |
2288 | char *fn; | |
2289 | ||
2290 | while ((r = cg_read_subgroup(d, &fn)) > 0) { | |
2291 | _cleanup_free_ char *p = NULL; | |
2292 | ||
605405c6 | 2293 | p = strjoin(path, "/", fn); |
efdb0237 LP |
2294 | free(fn); |
2295 | ||
2296 | if (!p) | |
2297 | return -ENOMEM; | |
2298 | ||
2299 | r = unit_watch_pids_in_path(u, p); | |
2300 | if (r < 0 && ret >= 0) | |
2301 | ret = r; | |
2302 | } | |
2303 | ||
2304 | if (r < 0 && ret >= 0) | |
2305 | ret = r; | |
2306 | } | |
2307 | ||
2308 | return ret; | |
2309 | } | |
2310 | ||
11aef522 LP |
2311 | int unit_synthesize_cgroup_empty_event(Unit *u) { |
2312 | int r; | |
2313 | ||
2314 | assert(u); | |
2315 | ||
2316 | /* Enqueue a synthetic cgroup empty event if this unit doesn't watch any PIDs anymore. This is compatibility | |
2317 | * support for non-unified systems where notifications aren't reliable, and hence need to take whatever we can | |
2318 | * get as notification source as soon as we stopped having any useful PIDs to watch for. */ | |
2319 | ||
2320 | if (!u->cgroup_path) | |
2321 | return -ENOENT; | |
2322 | ||
2323 | r = cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER); | |
2324 | if (r < 0) | |
2325 | return r; | |
2326 | if (r > 0) /* On unified we have reliable notifications, and don't need this */ | |
2327 | return 0; | |
2328 | ||
2329 | if (!set_isempty(u->pids)) | |
2330 | return 0; | |
2331 | ||
2332 | unit_add_to_cgroup_empty_queue(u); | |
2333 | return 0; | |
2334 | } | |
2335 | ||
efdb0237 | 2336 | int unit_watch_all_pids(Unit *u) { |
b4cccbc1 LP |
2337 | int r; |
2338 | ||
efdb0237 LP |
2339 | assert(u); |
2340 | ||
2341 | /* Adds all PIDs from our cgroup to the set of PIDs we | |
2342 | * watch. This is a fallback logic for cases where we do not | |
2343 | * get reliable cgroup empty notifications: we try to use | |
2344 | * SIGCHLD as replacement. */ | |
2345 | ||
2346 | if (!u->cgroup_path) | |
2347 | return -ENOENT; | |
2348 | ||
c22800e4 | 2349 | r = cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER); |
b4cccbc1 LP |
2350 | if (r < 0) |
2351 | return r; | |
2352 | if (r > 0) /* On unified we can use proper notifications */ | |
efdb0237 LP |
2353 | return 0; |
2354 | ||
2355 | return unit_watch_pids_in_path(u, u->cgroup_path); | |
2356 | } | |
2357 | ||
09e24654 LP |
2358 | static int on_cgroup_empty_event(sd_event_source *s, void *userdata) { |
2359 | Manager *m = userdata; | |
2360 | Unit *u; | |
efdb0237 LP |
2361 | int r; |
2362 | ||
09e24654 LP |
2363 | assert(s); |
2364 | assert(m); | |
efdb0237 | 2365 | |
09e24654 LP |
2366 | u = m->cgroup_empty_queue; |
2367 | if (!u) | |
efdb0237 LP |
2368 | return 0; |
2369 | ||
09e24654 LP |
2370 | assert(u->in_cgroup_empty_queue); |
2371 | u->in_cgroup_empty_queue = false; | |
2372 | LIST_REMOVE(cgroup_empty_queue, m->cgroup_empty_queue, u); | |
2373 | ||
2374 | if (m->cgroup_empty_queue) { | |
2375 | /* More stuff queued, let's make sure we remain enabled */ | |
2376 | r = sd_event_source_set_enabled(s, SD_EVENT_ONESHOT); | |
2377 | if (r < 0) | |
19a691a9 | 2378 | log_debug_errno(r, "Failed to reenable cgroup empty event source, ignoring: %m"); |
09e24654 | 2379 | } |
efdb0237 LP |
2380 | |
2381 | unit_add_to_gc_queue(u); | |
2382 | ||
2383 | if (UNIT_VTABLE(u)->notify_cgroup_empty) | |
2384 | UNIT_VTABLE(u)->notify_cgroup_empty(u); | |
2385 | ||
2386 | return 0; | |
2387 | } | |
2388 | ||
09e24654 LP |
2389 | void unit_add_to_cgroup_empty_queue(Unit *u) { |
2390 | int r; | |
2391 | ||
2392 | assert(u); | |
2393 | ||
2394 | /* Note that there are four different ways how cgroup empty events reach us: | |
2395 | * | |
2396 | * 1. On the unified hierarchy we get an inotify event on the cgroup | |
2397 | * | |
2398 | * 2. On the legacy hierarchy, when running in system mode, we get a datagram on the cgroup agent socket | |
2399 | * | |
2400 | * 3. On the legacy hierarchy, when running in user mode, we get a D-Bus signal on the system bus | |
2401 | * | |
2402 | * 4. On the legacy hierarchy, in service units we start watching all processes of the cgroup for SIGCHLD as | |
2403 | * soon as we get one SIGCHLD, to deal with unreliable cgroup notifications. | |
2404 | * | |
2405 | * Regardless which way we got the notification, we'll verify it here, and then add it to a separate | |
2406 | * queue. This queue will be dispatched at a lower priority than the SIGCHLD handler, so that we always use | |
2407 | * SIGCHLD if we can get it first, and only use the cgroup empty notifications if there's no SIGCHLD pending | |
2408 | * (which might happen if the cgroup doesn't contain processes that are our own child, which is typically the | |
2409 | * case for scope units). */ | |
2410 | ||
2411 | if (u->in_cgroup_empty_queue) | |
2412 | return; | |
2413 | ||
2414 | /* Let's verify that the cgroup is really empty */ | |
2415 | if (!u->cgroup_path) | |
2416 | return; | |
2417 | r = cg_is_empty_recursive(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path); | |
2418 | if (r < 0) { | |
2419 | log_unit_debug_errno(u, r, "Failed to determine whether cgroup %s is empty: %m", u->cgroup_path); | |
2420 | return; | |
2421 | } | |
2422 | if (r == 0) | |
2423 | return; | |
2424 | ||
2425 | LIST_PREPEND(cgroup_empty_queue, u->manager->cgroup_empty_queue, u); | |
2426 | u->in_cgroup_empty_queue = true; | |
2427 | ||
2428 | /* Trigger the defer event */ | |
2429 | r = sd_event_source_set_enabled(u->manager->cgroup_empty_event_source, SD_EVENT_ONESHOT); | |
2430 | if (r < 0) | |
2431 | log_debug_errno(r, "Failed to enable cgroup empty event source: %m"); | |
2432 | } | |
2433 | ||
efdb0237 LP |
2434 | static int on_cgroup_inotify_event(sd_event_source *s, int fd, uint32_t revents, void *userdata) { |
2435 | Manager *m = userdata; | |
2436 | ||
2437 | assert(s); | |
2438 | assert(fd >= 0); | |
2439 | assert(m); | |
2440 | ||
2441 | for (;;) { | |
2442 | union inotify_event_buffer buffer; | |
2443 | struct inotify_event *e; | |
2444 | ssize_t l; | |
2445 | ||
2446 | l = read(fd, &buffer, sizeof(buffer)); | |
2447 | if (l < 0) { | |
47249640 | 2448 | if (IN_SET(errno, EINTR, EAGAIN)) |
efdb0237 LP |
2449 | return 0; |
2450 | ||
2451 | return log_error_errno(errno, "Failed to read control group inotify events: %m"); | |
2452 | } | |
2453 | ||
2454 | FOREACH_INOTIFY_EVENT(e, buffer, l) { | |
2455 | Unit *u; | |
2456 | ||
2457 | if (e->wd < 0) | |
2458 | /* Queue overflow has no watch descriptor */ | |
2459 | continue; | |
2460 | ||
2461 | if (e->mask & IN_IGNORED) | |
2462 | /* The watch was just removed */ | |
2463 | continue; | |
2464 | ||
2465 | u = hashmap_get(m->cgroup_inotify_wd_unit, INT_TO_PTR(e->wd)); | |
2466 | if (!u) /* Not that inotify might deliver | |
2467 | * events for a watch even after it | |
2468 | * was removed, because it was queued | |
2469 | * before the removal. Let's ignore | |
2470 | * this here safely. */ | |
2471 | continue; | |
2472 | ||
09e24654 | 2473 | unit_add_to_cgroup_empty_queue(u); |
efdb0237 LP |
2474 | } |
2475 | } | |
8e274523 LP |
2476 | } |
2477 | ||
17f14955 RG |
2478 | static int cg_bpf_mask_supported(CGroupMask *ret) { |
2479 | CGroupMask mask = 0; | |
2480 | int r; | |
2481 | ||
2482 | /* BPF-based firewall */ | |
2483 | r = bpf_firewall_supported(); | |
2484 | if (r > 0) | |
2485 | mask |= CGROUP_MASK_BPF_FIREWALL; | |
2486 | ||
084c7007 RG |
2487 | /* BPF-based device access control */ |
2488 | r = bpf_devices_supported(); | |
2489 | if (r > 0) | |
2490 | mask |= CGROUP_MASK_BPF_DEVICES; | |
2491 | ||
17f14955 RG |
2492 | *ret = mask; |
2493 | return 0; | |
2494 | } | |
2495 | ||
8e274523 | 2496 | int manager_setup_cgroup(Manager *m) { |
9444b1f2 | 2497 | _cleanup_free_ char *path = NULL; |
10bd3e2e | 2498 | const char *scope_path; |
efdb0237 | 2499 | CGroupController c; |
b4cccbc1 | 2500 | int r, all_unified; |
17f14955 | 2501 | CGroupMask mask; |
efdb0237 | 2502 | char *e; |
8e274523 LP |
2503 | |
2504 | assert(m); | |
2505 | ||
35d2e7ec | 2506 | /* 1. Determine hierarchy */ |
efdb0237 | 2507 | m->cgroup_root = mfree(m->cgroup_root); |
9444b1f2 | 2508 | r = cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER, 0, &m->cgroup_root); |
23bbb0de MS |
2509 | if (r < 0) |
2510 | return log_error_errno(r, "Cannot determine cgroup we are running in: %m"); | |
8e274523 | 2511 | |
efdb0237 LP |
2512 | /* Chop off the init scope, if we are already located in it */ |
2513 | e = endswith(m->cgroup_root, "/" SPECIAL_INIT_SCOPE); | |
0d8c31ff | 2514 | |
efdb0237 LP |
2515 | /* LEGACY: Also chop off the system slice if we are in |
2516 | * it. This is to support live upgrades from older systemd | |
2517 | * versions where PID 1 was moved there. Also see | |
2518 | * cg_get_root_path(). */ | |
463d0d15 | 2519 | if (!e && MANAGER_IS_SYSTEM(m)) { |
9444b1f2 | 2520 | e = endswith(m->cgroup_root, "/" SPECIAL_SYSTEM_SLICE); |
15c60e99 | 2521 | if (!e) |
efdb0237 | 2522 | e = endswith(m->cgroup_root, "/system"); /* even more legacy */ |
0baf24dd | 2523 | } |
efdb0237 LP |
2524 | if (e) |
2525 | *e = 0; | |
7ccfb64a | 2526 | |
7546145e LP |
2527 | /* And make sure to store away the root value without trailing slash, even for the root dir, so that we can |
2528 | * easily prepend it everywhere. */ | |
2529 | delete_trailing_chars(m->cgroup_root, "/"); | |
8e274523 | 2530 | |
35d2e7ec | 2531 | /* 2. Show data */ |
9444b1f2 | 2532 | r = cg_get_path(SYSTEMD_CGROUP_CONTROLLER, m->cgroup_root, NULL, &path); |
23bbb0de MS |
2533 | if (r < 0) |
2534 | return log_error_errno(r, "Cannot find cgroup mount point: %m"); | |
8e274523 | 2535 | |
415fc41c TH |
2536 | r = cg_unified_flush(); |
2537 | if (r < 0) | |
2538 | return log_error_errno(r, "Couldn't determine if we are running in the unified hierarchy: %m"); | |
5da38d07 | 2539 | |
b4cccbc1 | 2540 | all_unified = cg_all_unified(); |
d4c819ed ZJS |
2541 | if (all_unified < 0) |
2542 | return log_error_errno(all_unified, "Couldn't determine whether we are in all unified mode: %m"); | |
2543 | if (all_unified > 0) | |
efdb0237 | 2544 | log_debug("Unified cgroup hierarchy is located at %s.", path); |
b4cccbc1 | 2545 | else { |
c22800e4 | 2546 | r = cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER); |
b4cccbc1 LP |
2547 | if (r < 0) |
2548 | return log_error_errno(r, "Failed to determine whether systemd's own controller is in unified mode: %m"); | |
2549 | if (r > 0) | |
2550 | log_debug("Unified cgroup hierarchy is located at %s. Controllers are on legacy hierarchies.", path); | |
2551 | else | |
2552 | log_debug("Using cgroup controller " SYSTEMD_CGROUP_CONTROLLER_LEGACY ". File system hierarchy is at %s.", path); | |
2553 | } | |
efdb0237 | 2554 | |
09e24654 LP |
2555 | /* 3. Allocate cgroup empty defer event source */ |
2556 | m->cgroup_empty_event_source = sd_event_source_unref(m->cgroup_empty_event_source); | |
2557 | r = sd_event_add_defer(m->event, &m->cgroup_empty_event_source, on_cgroup_empty_event, m); | |
2558 | if (r < 0) | |
2559 | return log_error_errno(r, "Failed to create cgroup empty event source: %m"); | |
2560 | ||
2561 | r = sd_event_source_set_priority(m->cgroup_empty_event_source, SD_EVENT_PRIORITY_NORMAL-5); | |
2562 | if (r < 0) | |
2563 | return log_error_errno(r, "Failed to set priority of cgroup empty event source: %m"); | |
2564 | ||
2565 | r = sd_event_source_set_enabled(m->cgroup_empty_event_source, SD_EVENT_OFF); | |
2566 | if (r < 0) | |
2567 | return log_error_errno(r, "Failed to disable cgroup empty event source: %m"); | |
2568 | ||
2569 | (void) sd_event_source_set_description(m->cgroup_empty_event_source, "cgroup-empty"); | |
2570 | ||
2571 | /* 4. Install notifier inotify object, or agent */ | |
10bd3e2e | 2572 | if (cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER) > 0) { |
c6c18be3 | 2573 | |
09e24654 | 2574 | /* In the unified hierarchy we can get cgroup empty notifications via inotify. */ |
efdb0237 | 2575 | |
10bd3e2e LP |
2576 | m->cgroup_inotify_event_source = sd_event_source_unref(m->cgroup_inotify_event_source); |
2577 | safe_close(m->cgroup_inotify_fd); | |
efdb0237 | 2578 | |
10bd3e2e LP |
2579 | m->cgroup_inotify_fd = inotify_init1(IN_NONBLOCK|IN_CLOEXEC); |
2580 | if (m->cgroup_inotify_fd < 0) | |
2581 | return log_error_errno(errno, "Failed to create control group inotify object: %m"); | |
efdb0237 | 2582 | |
10bd3e2e LP |
2583 | r = sd_event_add_io(m->event, &m->cgroup_inotify_event_source, m->cgroup_inotify_fd, EPOLLIN, on_cgroup_inotify_event, m); |
2584 | if (r < 0) | |
2585 | return log_error_errno(r, "Failed to watch control group inotify object: %m"); | |
efdb0237 | 2586 | |
10bd3e2e LP |
2587 | /* Process cgroup empty notifications early, but after service notifications and SIGCHLD. Also |
2588 | * see handling of cgroup agent notifications, for the classic cgroup hierarchy support. */ | |
09e24654 | 2589 | r = sd_event_source_set_priority(m->cgroup_inotify_event_source, SD_EVENT_PRIORITY_NORMAL-4); |
10bd3e2e LP |
2590 | if (r < 0) |
2591 | return log_error_errno(r, "Failed to set priority of inotify event source: %m"); | |
efdb0237 | 2592 | |
10bd3e2e | 2593 | (void) sd_event_source_set_description(m->cgroup_inotify_event_source, "cgroup-inotify"); |
efdb0237 | 2594 | |
611c4f8a | 2595 | } else if (MANAGER_IS_SYSTEM(m) && manager_owns_host_root_cgroup(m) && !MANAGER_IS_TEST_RUN(m)) { |
efdb0237 | 2596 | |
10bd3e2e LP |
2597 | /* On the legacy hierarchy we only get notifications via cgroup agents. (Which isn't really reliable, |
2598 | * since it does not generate events when control groups with children run empty. */ | |
8e274523 | 2599 | |
10bd3e2e | 2600 | r = cg_install_release_agent(SYSTEMD_CGROUP_CONTROLLER, SYSTEMD_CGROUP_AGENT_PATH); |
23bbb0de | 2601 | if (r < 0) |
10bd3e2e LP |
2602 | log_warning_errno(r, "Failed to install release agent, ignoring: %m"); |
2603 | else if (r > 0) | |
2604 | log_debug("Installed release agent."); | |
2605 | else if (r == 0) | |
2606 | log_debug("Release agent already installed."); | |
2607 | } | |
efdb0237 | 2608 | |
09e24654 | 2609 | /* 5. Make sure we are in the special "init.scope" unit in the root slice. */ |
10bd3e2e LP |
2610 | scope_path = strjoina(m->cgroup_root, "/" SPECIAL_INIT_SCOPE); |
2611 | r = cg_create_and_attach(SYSTEMD_CGROUP_CONTROLLER, scope_path, 0); | |
aa77e234 MS |
2612 | if (r >= 0) { |
2613 | /* Also, move all other userspace processes remaining in the root cgroup into that scope. */ | |
2614 | r = cg_migrate(SYSTEMD_CGROUP_CONTROLLER, m->cgroup_root, SYSTEMD_CGROUP_CONTROLLER, scope_path, 0); | |
2615 | if (r < 0) | |
2616 | log_warning_errno(r, "Couldn't move remaining userspace processes, ignoring: %m"); | |
c6c18be3 | 2617 | |
aa77e234 MS |
2618 | /* 6. And pin it, so that it cannot be unmounted */ |
2619 | safe_close(m->pin_cgroupfs_fd); | |
2620 | m->pin_cgroupfs_fd = open(path, O_RDONLY|O_CLOEXEC|O_DIRECTORY|O_NOCTTY|O_NONBLOCK); | |
2621 | if (m->pin_cgroupfs_fd < 0) | |
2622 | return log_error_errno(errno, "Failed to open pin file: %m"); | |
0d8c31ff | 2623 | |
638cece4 | 2624 | } else if (!MANAGER_IS_TEST_RUN(m)) |
aa77e234 | 2625 | return log_error_errno(r, "Failed to create %s control group: %m", scope_path); |
10bd3e2e | 2626 | |
09e24654 | 2627 | /* 7. Always enable hierarchical support if it exists... */ |
638cece4 | 2628 | if (!all_unified && !MANAGER_IS_TEST_RUN(m)) |
10bd3e2e | 2629 | (void) cg_set_attribute("memory", "/", "memory.use_hierarchy", "1"); |
c6c18be3 | 2630 | |
17f14955 | 2631 | /* 8. Figure out which controllers are supported */ |
efdb0237 LP |
2632 | r = cg_mask_supported(&m->cgroup_supported); |
2633 | if (r < 0) | |
2634 | return log_error_errno(r, "Failed to determine supported controllers: %m"); | |
17f14955 RG |
2635 | |
2636 | /* 9. Figure out which bpf-based pseudo-controllers are supported */ | |
2637 | r = cg_bpf_mask_supported(&mask); | |
2638 | if (r < 0) | |
2639 | return log_error_errno(r, "Failed to determine supported bpf-based pseudo-controllers: %m"); | |
2640 | m->cgroup_supported |= mask; | |
2641 | ||
2642 | /* 10. Log which controllers are supported */ | |
efdb0237 | 2643 | for (c = 0; c < _CGROUP_CONTROLLER_MAX; c++) |
eee0a1e4 | 2644 | log_debug("Controller '%s' supported: %s", cgroup_controller_to_string(c), yes_no(m->cgroup_supported & CGROUP_CONTROLLER_TO_MASK(c))); |
9156e799 | 2645 | |
a32360f1 | 2646 | return 0; |
8e274523 LP |
2647 | } |
2648 | ||
c6c18be3 | 2649 | void manager_shutdown_cgroup(Manager *m, bool delete) { |
8e274523 LP |
2650 | assert(m); |
2651 | ||
9444b1f2 LP |
2652 | /* We can't really delete the group, since we are in it. But |
2653 | * let's trim it. */ | |
f6c63f6f | 2654 | if (delete && m->cgroup_root && m->test_run_flags != MANAGER_TEST_RUN_MINIMAL) |
efdb0237 LP |
2655 | (void) cg_trim(SYSTEMD_CGROUP_CONTROLLER, m->cgroup_root, false); |
2656 | ||
09e24654 LP |
2657 | m->cgroup_empty_event_source = sd_event_source_unref(m->cgroup_empty_event_source); |
2658 | ||
efdb0237 LP |
2659 | m->cgroup_inotify_wd_unit = hashmap_free(m->cgroup_inotify_wd_unit); |
2660 | ||
2661 | m->cgroup_inotify_event_source = sd_event_source_unref(m->cgroup_inotify_event_source); | |
2662 | m->cgroup_inotify_fd = safe_close(m->cgroup_inotify_fd); | |
8e274523 | 2663 | |
03e334a1 | 2664 | m->pin_cgroupfs_fd = safe_close(m->pin_cgroupfs_fd); |
c6c18be3 | 2665 | |
efdb0237 | 2666 | m->cgroup_root = mfree(m->cgroup_root); |
8e274523 LP |
2667 | } |
2668 | ||
4ad49000 | 2669 | Unit* manager_get_unit_by_cgroup(Manager *m, const char *cgroup) { |
acb14d31 | 2670 | char *p; |
4ad49000 | 2671 | Unit *u; |
acb14d31 LP |
2672 | |
2673 | assert(m); | |
2674 | assert(cgroup); | |
acb14d31 | 2675 | |
4ad49000 LP |
2676 | u = hashmap_get(m->cgroup_unit, cgroup); |
2677 | if (u) | |
2678 | return u; | |
acb14d31 | 2679 | |
8e70580b | 2680 | p = strdupa(cgroup); |
acb14d31 LP |
2681 | for (;;) { |
2682 | char *e; | |
2683 | ||
2684 | e = strrchr(p, '/'); | |
efdb0237 LP |
2685 | if (!e || e == p) |
2686 | return hashmap_get(m->cgroup_unit, SPECIAL_ROOT_SLICE); | |
acb14d31 LP |
2687 | |
2688 | *e = 0; | |
2689 | ||
4ad49000 LP |
2690 | u = hashmap_get(m->cgroup_unit, p); |
2691 | if (u) | |
2692 | return u; | |
acb14d31 LP |
2693 | } |
2694 | } | |
2695 | ||
b3ac818b | 2696 | Unit *manager_get_unit_by_pid_cgroup(Manager *m, pid_t pid) { |
4ad49000 | 2697 | _cleanup_free_ char *cgroup = NULL; |
8e274523 | 2698 | |
8c47c732 LP |
2699 | assert(m); |
2700 | ||
62a76913 | 2701 | if (!pid_is_valid(pid)) |
b3ac818b LP |
2702 | return NULL; |
2703 | ||
62a76913 | 2704 | if (cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER, pid, &cgroup) < 0) |
b3ac818b LP |
2705 | return NULL; |
2706 | ||
2707 | return manager_get_unit_by_cgroup(m, cgroup); | |
2708 | } | |
2709 | ||
2710 | Unit *manager_get_unit_by_pid(Manager *m, pid_t pid) { | |
62a76913 | 2711 | Unit *u, **array; |
b3ac818b LP |
2712 | |
2713 | assert(m); | |
2714 | ||
62a76913 LP |
2715 | /* Note that a process might be owned by multiple units, we return only one here, which is good enough for most |
2716 | * cases, though not strictly correct. We prefer the one reported by cgroup membership, as that's the most | |
2717 | * relevant one as children of the process will be assigned to that one, too, before all else. */ | |
2718 | ||
2719 | if (!pid_is_valid(pid)) | |
8c47c732 LP |
2720 | return NULL; |
2721 | ||
2ca9d979 | 2722 | if (pid == getpid_cached()) |
efdb0237 LP |
2723 | return hashmap_get(m->units, SPECIAL_INIT_SCOPE); |
2724 | ||
62a76913 | 2725 | u = manager_get_unit_by_pid_cgroup(m, pid); |
5fe8876b LP |
2726 | if (u) |
2727 | return u; | |
2728 | ||
62a76913 | 2729 | u = hashmap_get(m->watch_pids, PID_TO_PTR(pid)); |
5fe8876b LP |
2730 | if (u) |
2731 | return u; | |
2732 | ||
62a76913 LP |
2733 | array = hashmap_get(m->watch_pids, PID_TO_PTR(-pid)); |
2734 | if (array) | |
2735 | return array[0]; | |
2736 | ||
2737 | return NULL; | |
6dde1f33 | 2738 | } |
4fbf50b3 | 2739 | |
4ad49000 LP |
2740 | int manager_notify_cgroup_empty(Manager *m, const char *cgroup) { |
2741 | Unit *u; | |
4fbf50b3 | 2742 | |
4ad49000 LP |
2743 | assert(m); |
2744 | assert(cgroup); | |
4fbf50b3 | 2745 | |
09e24654 LP |
2746 | /* Called on the legacy hierarchy whenever we get an explicit cgroup notification from the cgroup agent process |
2747 | * or from the --system instance */ | |
2748 | ||
d8fdc620 LP |
2749 | log_debug("Got cgroup empty notification for: %s", cgroup); |
2750 | ||
4ad49000 | 2751 | u = manager_get_unit_by_cgroup(m, cgroup); |
5ad096b3 LP |
2752 | if (!u) |
2753 | return 0; | |
b56c28c3 | 2754 | |
09e24654 LP |
2755 | unit_add_to_cgroup_empty_queue(u); |
2756 | return 1; | |
5ad096b3 LP |
2757 | } |
2758 | ||
2759 | int unit_get_memory_current(Unit *u, uint64_t *ret) { | |
2760 | _cleanup_free_ char *v = NULL; | |
2761 | int r; | |
2762 | ||
2763 | assert(u); | |
2764 | assert(ret); | |
2765 | ||
2e4025c0 | 2766 | if (!UNIT_CGROUP_BOOL(u, memory_accounting)) |
cf3b4be1 LP |
2767 | return -ENODATA; |
2768 | ||
5ad096b3 LP |
2769 | if (!u->cgroup_path) |
2770 | return -ENODATA; | |
2771 | ||
1f73aa00 | 2772 | /* The root cgroup doesn't expose this information, let's get it from /proc instead */ |
611c4f8a | 2773 | if (unit_has_host_root_cgroup(u)) |
1f73aa00 LP |
2774 | return procfs_memory_get_current(ret); |
2775 | ||
efdb0237 | 2776 | if ((u->cgroup_realized_mask & CGROUP_MASK_MEMORY) == 0) |
5ad096b3 LP |
2777 | return -ENODATA; |
2778 | ||
b4cccbc1 LP |
2779 | r = cg_all_unified(); |
2780 | if (r < 0) | |
2781 | return r; | |
2782 | if (r > 0) | |
efdb0237 | 2783 | r = cg_get_attribute("memory", u->cgroup_path, "memory.current", &v); |
b4cccbc1 LP |
2784 | else |
2785 | r = cg_get_attribute("memory", u->cgroup_path, "memory.usage_in_bytes", &v); | |
5ad096b3 LP |
2786 | if (r == -ENOENT) |
2787 | return -ENODATA; | |
2788 | if (r < 0) | |
2789 | return r; | |
2790 | ||
2791 | return safe_atou64(v, ret); | |
2792 | } | |
2793 | ||
03a7b521 LP |
2794 | int unit_get_tasks_current(Unit *u, uint64_t *ret) { |
2795 | _cleanup_free_ char *v = NULL; | |
2796 | int r; | |
2797 | ||
2798 | assert(u); | |
2799 | assert(ret); | |
2800 | ||
2e4025c0 | 2801 | if (!UNIT_CGROUP_BOOL(u, tasks_accounting)) |
cf3b4be1 LP |
2802 | return -ENODATA; |
2803 | ||
03a7b521 LP |
2804 | if (!u->cgroup_path) |
2805 | return -ENODATA; | |
2806 | ||
c36a69f4 | 2807 | /* The root cgroup doesn't expose this information, let's get it from /proc instead */ |
611c4f8a | 2808 | if (unit_has_host_root_cgroup(u)) |
c36a69f4 LP |
2809 | return procfs_tasks_get_current(ret); |
2810 | ||
1f73aa00 LP |
2811 | if ((u->cgroup_realized_mask & CGROUP_MASK_PIDS) == 0) |
2812 | return -ENODATA; | |
2813 | ||
03a7b521 LP |
2814 | r = cg_get_attribute("pids", u->cgroup_path, "pids.current", &v); |
2815 | if (r == -ENOENT) | |
2816 | return -ENODATA; | |
2817 | if (r < 0) | |
2818 | return r; | |
2819 | ||
2820 | return safe_atou64(v, ret); | |
2821 | } | |
2822 | ||
5ad096b3 LP |
2823 | static int unit_get_cpu_usage_raw(Unit *u, nsec_t *ret) { |
2824 | _cleanup_free_ char *v = NULL; | |
2825 | uint64_t ns; | |
2826 | int r; | |
2827 | ||
2828 | assert(u); | |
2829 | assert(ret); | |
2830 | ||
2831 | if (!u->cgroup_path) | |
2832 | return -ENODATA; | |
2833 | ||
1f73aa00 | 2834 | /* The root cgroup doesn't expose this information, let's get it from /proc instead */ |
611c4f8a | 2835 | if (unit_has_host_root_cgroup(u)) |
1f73aa00 LP |
2836 | return procfs_cpu_get_usage(ret); |
2837 | ||
f98c2585 CD |
2838 | /* Requisite controllers for CPU accounting are not enabled */ |
2839 | if ((get_cpu_accounting_mask() & ~u->cgroup_realized_mask) != 0) | |
2840 | return -ENODATA; | |
2841 | ||
92a99304 LP |
2842 | r = cg_all_unified(); |
2843 | if (r < 0) | |
2844 | return r; | |
b4cccbc1 | 2845 | if (r > 0) { |
66ebf6c0 TH |
2846 | _cleanup_free_ char *val = NULL; |
2847 | uint64_t us; | |
5ad096b3 | 2848 | |
b734a4ff | 2849 | r = cg_get_keyed_attribute("cpu", u->cgroup_path, "cpu.stat", STRV_MAKE("usage_usec"), &val); |
b734a4ff LP |
2850 | if (IN_SET(r, -ENOENT, -ENXIO)) |
2851 | return -ENODATA; | |
d742f4b5 LP |
2852 | if (r < 0) |
2853 | return r; | |
66ebf6c0 TH |
2854 | |
2855 | r = safe_atou64(val, &us); | |
2856 | if (r < 0) | |
2857 | return r; | |
2858 | ||
2859 | ns = us * NSEC_PER_USEC; | |
2860 | } else { | |
66ebf6c0 TH |
2861 | r = cg_get_attribute("cpuacct", u->cgroup_path, "cpuacct.usage", &v); |
2862 | if (r == -ENOENT) | |
2863 | return -ENODATA; | |
2864 | if (r < 0) | |
2865 | return r; | |
2866 | ||
2867 | r = safe_atou64(v, &ns); | |
2868 | if (r < 0) | |
2869 | return r; | |
2870 | } | |
5ad096b3 LP |
2871 | |
2872 | *ret = ns; | |
2873 | return 0; | |
2874 | } | |
2875 | ||
2876 | int unit_get_cpu_usage(Unit *u, nsec_t *ret) { | |
2877 | nsec_t ns; | |
2878 | int r; | |
2879 | ||
fe700f46 LP |
2880 | assert(u); |
2881 | ||
2882 | /* Retrieve the current CPU usage counter. This will subtract the CPU counter taken when the unit was | |
2883 | * started. If the cgroup has been removed already, returns the last cached value. To cache the value, simply | |
2884 | * call this function with a NULL return value. */ | |
2885 | ||
2e4025c0 | 2886 | if (!UNIT_CGROUP_BOOL(u, cpu_accounting)) |
cf3b4be1 LP |
2887 | return -ENODATA; |
2888 | ||
5ad096b3 | 2889 | r = unit_get_cpu_usage_raw(u, &ns); |
fe700f46 LP |
2890 | if (r == -ENODATA && u->cpu_usage_last != NSEC_INFINITY) { |
2891 | /* If we can't get the CPU usage anymore (because the cgroup was already removed, for example), use our | |
2892 | * cached value. */ | |
2893 | ||
2894 | if (ret) | |
2895 | *ret = u->cpu_usage_last; | |
2896 | return 0; | |
2897 | } | |
5ad096b3 LP |
2898 | if (r < 0) |
2899 | return r; | |
2900 | ||
66ebf6c0 TH |
2901 | if (ns > u->cpu_usage_base) |
2902 | ns -= u->cpu_usage_base; | |
5ad096b3 LP |
2903 | else |
2904 | ns = 0; | |
2905 | ||
fe700f46 LP |
2906 | u->cpu_usage_last = ns; |
2907 | if (ret) | |
2908 | *ret = ns; | |
2909 | ||
5ad096b3 LP |
2910 | return 0; |
2911 | } | |
2912 | ||
906c06f6 DM |
2913 | int unit_get_ip_accounting( |
2914 | Unit *u, | |
2915 | CGroupIPAccountingMetric metric, | |
2916 | uint64_t *ret) { | |
2917 | ||
6b659ed8 | 2918 | uint64_t value; |
906c06f6 DM |
2919 | int fd, r; |
2920 | ||
2921 | assert(u); | |
2922 | assert(metric >= 0); | |
2923 | assert(metric < _CGROUP_IP_ACCOUNTING_METRIC_MAX); | |
2924 | assert(ret); | |
2925 | ||
2e4025c0 | 2926 | if (!UNIT_CGROUP_BOOL(u, ip_accounting)) |
cf3b4be1 LP |
2927 | return -ENODATA; |
2928 | ||
906c06f6 DM |
2929 | fd = IN_SET(metric, CGROUP_IP_INGRESS_BYTES, CGROUP_IP_INGRESS_PACKETS) ? |
2930 | u->ip_accounting_ingress_map_fd : | |
2931 | u->ip_accounting_egress_map_fd; | |
906c06f6 DM |
2932 | if (fd < 0) |
2933 | return -ENODATA; | |
2934 | ||
2935 | if (IN_SET(metric, CGROUP_IP_INGRESS_BYTES, CGROUP_IP_EGRESS_BYTES)) | |
6b659ed8 | 2936 | r = bpf_firewall_read_accounting(fd, &value, NULL); |
906c06f6 | 2937 | else |
6b659ed8 LP |
2938 | r = bpf_firewall_read_accounting(fd, NULL, &value); |
2939 | if (r < 0) | |
2940 | return r; | |
2941 | ||
2942 | /* Add in additional metrics from a previous runtime. Note that when reexecing/reloading the daemon we compile | |
2943 | * all BPF programs and maps anew, but serialize the old counters. When deserializing we store them in the | |
2944 | * ip_accounting_extra[] field, and add them in here transparently. */ | |
2945 | ||
2946 | *ret = value + u->ip_accounting_extra[metric]; | |
906c06f6 DM |
2947 | |
2948 | return r; | |
2949 | } | |
2950 | ||
2951 | int unit_reset_cpu_accounting(Unit *u) { | |
5ad096b3 LP |
2952 | nsec_t ns; |
2953 | int r; | |
2954 | ||
2955 | assert(u); | |
2956 | ||
fe700f46 LP |
2957 | u->cpu_usage_last = NSEC_INFINITY; |
2958 | ||
5ad096b3 LP |
2959 | r = unit_get_cpu_usage_raw(u, &ns); |
2960 | if (r < 0) { | |
66ebf6c0 | 2961 | u->cpu_usage_base = 0; |
5ad096b3 | 2962 | return r; |
b56c28c3 | 2963 | } |
2633eb83 | 2964 | |
66ebf6c0 | 2965 | u->cpu_usage_base = ns; |
4ad49000 | 2966 | return 0; |
4fbf50b3 LP |
2967 | } |
2968 | ||
906c06f6 DM |
2969 | int unit_reset_ip_accounting(Unit *u) { |
2970 | int r = 0, q = 0; | |
2971 | ||
2972 | assert(u); | |
2973 | ||
2974 | if (u->ip_accounting_ingress_map_fd >= 0) | |
2975 | r = bpf_firewall_reset_accounting(u->ip_accounting_ingress_map_fd); | |
2976 | ||
2977 | if (u->ip_accounting_egress_map_fd >= 0) | |
2978 | q = bpf_firewall_reset_accounting(u->ip_accounting_egress_map_fd); | |
2979 | ||
6b659ed8 LP |
2980 | zero(u->ip_accounting_extra); |
2981 | ||
906c06f6 DM |
2982 | return r < 0 ? r : q; |
2983 | } | |
2984 | ||
e7ab4d1a LP |
2985 | void unit_invalidate_cgroup(Unit *u, CGroupMask m) { |
2986 | assert(u); | |
2987 | ||
2988 | if (!UNIT_HAS_CGROUP_CONTEXT(u)) | |
2989 | return; | |
2990 | ||
2991 | if (m == 0) | |
2992 | return; | |
2993 | ||
538b4852 TH |
2994 | /* always invalidate compat pairs together */ |
2995 | if (m & (CGROUP_MASK_IO | CGROUP_MASK_BLKIO)) | |
2996 | m |= CGROUP_MASK_IO | CGROUP_MASK_BLKIO; | |
2997 | ||
7cce4fb7 LP |
2998 | if (m & (CGROUP_MASK_CPU | CGROUP_MASK_CPUACCT)) |
2999 | m |= CGROUP_MASK_CPU | CGROUP_MASK_CPUACCT; | |
3000 | ||
e00068e7 | 3001 | if (FLAGS_SET(u->cgroup_invalidated_mask, m)) /* NOP? */ |
e7ab4d1a LP |
3002 | return; |
3003 | ||
e00068e7 | 3004 | u->cgroup_invalidated_mask |= m; |
91a6073e | 3005 | unit_add_to_cgroup_realize_queue(u); |
e7ab4d1a LP |
3006 | } |
3007 | ||
906c06f6 DM |
3008 | void unit_invalidate_cgroup_bpf(Unit *u) { |
3009 | assert(u); | |
3010 | ||
3011 | if (!UNIT_HAS_CGROUP_CONTEXT(u)) | |
3012 | return; | |
3013 | ||
17f14955 | 3014 | if (u->cgroup_invalidated_mask & CGROUP_MASK_BPF_FIREWALL) /* NOP? */ |
906c06f6 DM |
3015 | return; |
3016 | ||
17f14955 | 3017 | u->cgroup_invalidated_mask |= CGROUP_MASK_BPF_FIREWALL; |
91a6073e | 3018 | unit_add_to_cgroup_realize_queue(u); |
906c06f6 DM |
3019 | |
3020 | /* If we are a slice unit, we also need to put compile a new BPF program for all our children, as the IP access | |
3021 | * list of our children includes our own. */ | |
3022 | if (u->type == UNIT_SLICE) { | |
3023 | Unit *member; | |
3024 | Iterator i; | |
eef85c4a | 3025 | void *v; |
906c06f6 | 3026 | |
eef85c4a | 3027 | HASHMAP_FOREACH_KEY(v, member, u->dependencies[UNIT_BEFORE], i) { |
cb5e3bc3 CD |
3028 | if (UNIT_DEREF(member->slice) == u) |
3029 | unit_invalidate_cgroup_bpf(member); | |
906c06f6 DM |
3030 | } |
3031 | } | |
3032 | } | |
3033 | ||
1d9cc876 LP |
3034 | bool unit_cgroup_delegate(Unit *u) { |
3035 | CGroupContext *c; | |
3036 | ||
3037 | assert(u); | |
3038 | ||
3039 | if (!UNIT_VTABLE(u)->can_delegate) | |
3040 | return false; | |
3041 | ||
3042 | c = unit_get_cgroup_context(u); | |
3043 | if (!c) | |
3044 | return false; | |
3045 | ||
3046 | return c->delegate; | |
3047 | } | |
3048 | ||
e7ab4d1a LP |
3049 | void manager_invalidate_startup_units(Manager *m) { |
3050 | Iterator i; | |
3051 | Unit *u; | |
3052 | ||
3053 | assert(m); | |
3054 | ||
3055 | SET_FOREACH(u, m->startup_units, i) | |
13c31542 | 3056 | unit_invalidate_cgroup(u, CGROUP_MASK_CPU|CGROUP_MASK_IO|CGROUP_MASK_BLKIO); |
e7ab4d1a LP |
3057 | } |
3058 | ||
4ad49000 LP |
3059 | static const char* const cgroup_device_policy_table[_CGROUP_DEVICE_POLICY_MAX] = { |
3060 | [CGROUP_AUTO] = "auto", | |
3061 | [CGROUP_CLOSED] = "closed", | |
3062 | [CGROUP_STRICT] = "strict", | |
3063 | }; | |
4fbf50b3 | 3064 | |
4ad49000 | 3065 | DEFINE_STRING_TABLE_LOOKUP(cgroup_device_policy, CGroupDevicePolicy); |