]>
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 | |
10f28641 | 28 | #define CGROUP_CPU_QUOTA_DEFAULT_PERIOD_USEC ((usec_t) 100 * USEC_PER_MSEC) |
9a054909 | 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, | |
10f28641 | 101 | .cpu_quota_period_usec = USEC_INFINITY, |
66ebf6c0 | 102 | |
de8a711a LP |
103 | .cpu_shares = CGROUP_CPU_SHARES_INVALID, |
104 | .startup_cpu_shares = CGROUP_CPU_SHARES_INVALID, | |
d53d9474 | 105 | |
de8a711a LP |
106 | .memory_high = CGROUP_LIMIT_MAX, |
107 | .memory_max = CGROUP_LIMIT_MAX, | |
108 | .memory_swap_max = CGROUP_LIMIT_MAX, | |
da4d897e | 109 | |
de8a711a | 110 | .memory_limit = CGROUP_LIMIT_MAX, |
b2f8b02e | 111 | |
de8a711a LP |
112 | .io_weight = CGROUP_WEIGHT_INVALID, |
113 | .startup_io_weight = CGROUP_WEIGHT_INVALID, | |
13c31542 | 114 | |
de8a711a LP |
115 | .blockio_weight = CGROUP_BLKIO_WEIGHT_INVALID, |
116 | .startup_blockio_weight = CGROUP_BLKIO_WEIGHT_INVALID, | |
d53d9474 | 117 | |
de8a711a LP |
118 | .tasks_max = CGROUP_LIMIT_MAX, |
119 | }; | |
4ad49000 | 120 | } |
8e274523 | 121 | |
4ad49000 LP |
122 | void cgroup_context_free_device_allow(CGroupContext *c, CGroupDeviceAllow *a) { |
123 | assert(c); | |
124 | assert(a); | |
125 | ||
71fda00f | 126 | LIST_REMOVE(device_allow, c->device_allow, a); |
4ad49000 LP |
127 | free(a->path); |
128 | free(a); | |
129 | } | |
130 | ||
13c31542 TH |
131 | void cgroup_context_free_io_device_weight(CGroupContext *c, CGroupIODeviceWeight *w) { |
132 | assert(c); | |
133 | assert(w); | |
134 | ||
135 | LIST_REMOVE(device_weights, c->io_device_weights, w); | |
136 | free(w->path); | |
137 | free(w); | |
138 | } | |
139 | ||
6ae4283c TH |
140 | void cgroup_context_free_io_device_latency(CGroupContext *c, CGroupIODeviceLatency *l) { |
141 | assert(c); | |
142 | assert(l); | |
143 | ||
144 | LIST_REMOVE(device_latencies, c->io_device_latencies, l); | |
145 | free(l->path); | |
146 | free(l); | |
147 | } | |
148 | ||
13c31542 TH |
149 | void cgroup_context_free_io_device_limit(CGroupContext *c, CGroupIODeviceLimit *l) { |
150 | assert(c); | |
151 | assert(l); | |
152 | ||
153 | LIST_REMOVE(device_limits, c->io_device_limits, l); | |
154 | free(l->path); | |
155 | free(l); | |
156 | } | |
157 | ||
4ad49000 LP |
158 | void cgroup_context_free_blockio_device_weight(CGroupContext *c, CGroupBlockIODeviceWeight *w) { |
159 | assert(c); | |
160 | assert(w); | |
161 | ||
71fda00f | 162 | LIST_REMOVE(device_weights, c->blockio_device_weights, w); |
4ad49000 LP |
163 | free(w->path); |
164 | free(w); | |
165 | } | |
166 | ||
167 | void cgroup_context_free_blockio_device_bandwidth(CGroupContext *c, CGroupBlockIODeviceBandwidth *b) { | |
168 | assert(c); | |
8e274523 | 169 | assert(b); |
8e274523 | 170 | |
71fda00f | 171 | LIST_REMOVE(device_bandwidths, c->blockio_device_bandwidths, b); |
4ad49000 LP |
172 | free(b->path); |
173 | free(b); | |
174 | } | |
175 | ||
176 | void cgroup_context_done(CGroupContext *c) { | |
177 | assert(c); | |
178 | ||
13c31542 TH |
179 | while (c->io_device_weights) |
180 | cgroup_context_free_io_device_weight(c, c->io_device_weights); | |
181 | ||
6ae4283c TH |
182 | while (c->io_device_latencies) |
183 | cgroup_context_free_io_device_latency(c, c->io_device_latencies); | |
184 | ||
13c31542 TH |
185 | while (c->io_device_limits) |
186 | cgroup_context_free_io_device_limit(c, c->io_device_limits); | |
187 | ||
4ad49000 LP |
188 | while (c->blockio_device_weights) |
189 | cgroup_context_free_blockio_device_weight(c, c->blockio_device_weights); | |
190 | ||
191 | while (c->blockio_device_bandwidths) | |
192 | cgroup_context_free_blockio_device_bandwidth(c, c->blockio_device_bandwidths); | |
193 | ||
194 | while (c->device_allow) | |
195 | cgroup_context_free_device_allow(c, c->device_allow); | |
6a48d82f DM |
196 | |
197 | c->ip_address_allow = ip_address_access_free_all(c->ip_address_allow); | |
198 | c->ip_address_deny = ip_address_access_free_all(c->ip_address_deny); | |
4ad49000 LP |
199 | } |
200 | ||
201 | void cgroup_context_dump(CGroupContext *c, FILE* f, const char *prefix) { | |
13c31542 TH |
202 | CGroupIODeviceLimit *il; |
203 | CGroupIODeviceWeight *iw; | |
6ae4283c | 204 | CGroupIODeviceLatency *l; |
4ad49000 LP |
205 | CGroupBlockIODeviceBandwidth *b; |
206 | CGroupBlockIODeviceWeight *w; | |
207 | CGroupDeviceAllow *a; | |
c21c9906 | 208 | IPAddressAccessItem *iaai; |
9a054909 | 209 | char u[FORMAT_TIMESPAN_MAX]; |
10f28641 | 210 | char v[FORMAT_TIMESPAN_MAX]; |
4ad49000 LP |
211 | |
212 | assert(c); | |
213 | assert(f); | |
214 | ||
215 | prefix = strempty(prefix); | |
216 | ||
217 | fprintf(f, | |
218 | "%sCPUAccounting=%s\n" | |
13c31542 | 219 | "%sIOAccounting=%s\n" |
4ad49000 LP |
220 | "%sBlockIOAccounting=%s\n" |
221 | "%sMemoryAccounting=%s\n" | |
d53d9474 | 222 | "%sTasksAccounting=%s\n" |
c21c9906 | 223 | "%sIPAccounting=%s\n" |
66ebf6c0 TH |
224 | "%sCPUWeight=%" PRIu64 "\n" |
225 | "%sStartupCPUWeight=%" PRIu64 "\n" | |
d53d9474 LP |
226 | "%sCPUShares=%" PRIu64 "\n" |
227 | "%sStartupCPUShares=%" PRIu64 "\n" | |
b2f8b02e | 228 | "%sCPUQuotaPerSecSec=%s\n" |
10f28641 | 229 | "%sCPUQuotaPeriodSec=%s\n" |
13c31542 TH |
230 | "%sIOWeight=%" PRIu64 "\n" |
231 | "%sStartupIOWeight=%" PRIu64 "\n" | |
d53d9474 LP |
232 | "%sBlockIOWeight=%" PRIu64 "\n" |
233 | "%sStartupBlockIOWeight=%" PRIu64 "\n" | |
48422635 | 234 | "%sMemoryMin=%" PRIu64 "\n" |
da4d897e TH |
235 | "%sMemoryLow=%" PRIu64 "\n" |
236 | "%sMemoryHigh=%" PRIu64 "\n" | |
237 | "%sMemoryMax=%" PRIu64 "\n" | |
96e131ea | 238 | "%sMemorySwapMax=%" PRIu64 "\n" |
4ad49000 | 239 | "%sMemoryLimit=%" PRIu64 "\n" |
03a7b521 | 240 | "%sTasksMax=%" PRIu64 "\n" |
a931ad47 LP |
241 | "%sDevicePolicy=%s\n" |
242 | "%sDelegate=%s\n", | |
4ad49000 | 243 | prefix, yes_no(c->cpu_accounting), |
13c31542 | 244 | prefix, yes_no(c->io_accounting), |
4ad49000 LP |
245 | prefix, yes_no(c->blockio_accounting), |
246 | prefix, yes_no(c->memory_accounting), | |
d53d9474 | 247 | prefix, yes_no(c->tasks_accounting), |
c21c9906 | 248 | prefix, yes_no(c->ip_accounting), |
66ebf6c0 TH |
249 | prefix, c->cpu_weight, |
250 | prefix, c->startup_cpu_weight, | |
4ad49000 | 251 | prefix, c->cpu_shares, |
95ae05c0 | 252 | prefix, c->startup_cpu_shares, |
b1d6dcf5 | 253 | prefix, format_timespan(u, sizeof(u), c->cpu_quota_per_sec_usec, 1), |
10f28641 | 254 | prefix, format_timespan(v, sizeof(v), c->cpu_quota_period_usec, 1), |
13c31542 TH |
255 | prefix, c->io_weight, |
256 | prefix, c->startup_io_weight, | |
4ad49000 | 257 | prefix, c->blockio_weight, |
95ae05c0 | 258 | prefix, c->startup_blockio_weight, |
48422635 | 259 | prefix, c->memory_min, |
da4d897e TH |
260 | prefix, c->memory_low, |
261 | prefix, c->memory_high, | |
262 | prefix, c->memory_max, | |
96e131ea | 263 | prefix, c->memory_swap_max, |
4ad49000 | 264 | prefix, c->memory_limit, |
03a7b521 | 265 | prefix, c->tasks_max, |
a931ad47 LP |
266 | prefix, cgroup_device_policy_to_string(c->device_policy), |
267 | prefix, yes_no(c->delegate)); | |
4ad49000 | 268 | |
02638280 LP |
269 | if (c->delegate) { |
270 | _cleanup_free_ char *t = NULL; | |
271 | ||
272 | (void) cg_mask_to_string(c->delegate_controllers, &t); | |
273 | ||
47a78d41 | 274 | fprintf(f, "%sDelegateControllers=%s\n", |
02638280 LP |
275 | prefix, |
276 | strempty(t)); | |
277 | } | |
278 | ||
4ad49000 LP |
279 | LIST_FOREACH(device_allow, a, c->device_allow) |
280 | fprintf(f, | |
281 | "%sDeviceAllow=%s %s%s%s\n", | |
282 | prefix, | |
283 | a->path, | |
284 | a->r ? "r" : "", a->w ? "w" : "", a->m ? "m" : ""); | |
285 | ||
13c31542 TH |
286 | LIST_FOREACH(device_weights, iw, c->io_device_weights) |
287 | fprintf(f, | |
6ae4283c | 288 | "%sIODeviceWeight=%s %" PRIu64 "\n", |
13c31542 TH |
289 | prefix, |
290 | iw->path, | |
291 | iw->weight); | |
292 | ||
6ae4283c TH |
293 | LIST_FOREACH(device_latencies, l, c->io_device_latencies) |
294 | fprintf(f, | |
295 | "%sIODeviceLatencyTargetSec=%s %s\n", | |
296 | prefix, | |
297 | l->path, | |
298 | format_timespan(u, sizeof(u), l->target_usec, 1)); | |
299 | ||
13c31542 TH |
300 | LIST_FOREACH(device_limits, il, c->io_device_limits) { |
301 | char buf[FORMAT_BYTES_MAX]; | |
9be57249 TH |
302 | CGroupIOLimitType type; |
303 | ||
304 | for (type = 0; type < _CGROUP_IO_LIMIT_TYPE_MAX; type++) | |
305 | if (il->limits[type] != cgroup_io_limit_defaults[type]) | |
306 | fprintf(f, | |
307 | "%s%s=%s %s\n", | |
308 | prefix, | |
309 | cgroup_io_limit_type_to_string(type), | |
310 | il->path, | |
311 | format_bytes(buf, sizeof(buf), il->limits[type])); | |
13c31542 TH |
312 | } |
313 | ||
4ad49000 LP |
314 | LIST_FOREACH(device_weights, w, c->blockio_device_weights) |
315 | fprintf(f, | |
d53d9474 | 316 | "%sBlockIODeviceWeight=%s %" PRIu64, |
4ad49000 LP |
317 | prefix, |
318 | w->path, | |
319 | w->weight); | |
320 | ||
321 | LIST_FOREACH(device_bandwidths, b, c->blockio_device_bandwidths) { | |
322 | char buf[FORMAT_BYTES_MAX]; | |
323 | ||
979d0311 TH |
324 | if (b->rbps != CGROUP_LIMIT_MAX) |
325 | fprintf(f, | |
326 | "%sBlockIOReadBandwidth=%s %s\n", | |
327 | prefix, | |
328 | b->path, | |
329 | format_bytes(buf, sizeof(buf), b->rbps)); | |
330 | if (b->wbps != CGROUP_LIMIT_MAX) | |
331 | fprintf(f, | |
332 | "%sBlockIOWriteBandwidth=%s %s\n", | |
333 | prefix, | |
334 | b->path, | |
335 | format_bytes(buf, sizeof(buf), b->wbps)); | |
4ad49000 | 336 | } |
c21c9906 LP |
337 | |
338 | LIST_FOREACH(items, iaai, c->ip_address_allow) { | |
339 | _cleanup_free_ char *k = NULL; | |
340 | ||
341 | (void) in_addr_to_string(iaai->family, &iaai->address, &k); | |
342 | fprintf(f, "%sIPAddressAllow=%s/%u\n", prefix, strnull(k), iaai->prefixlen); | |
343 | } | |
344 | ||
345 | LIST_FOREACH(items, iaai, c->ip_address_deny) { | |
346 | _cleanup_free_ char *k = NULL; | |
347 | ||
348 | (void) in_addr_to_string(iaai->family, &iaai->address, &k); | |
349 | fprintf(f, "%sIPAddressDeny=%s/%u\n", prefix, strnull(k), iaai->prefixlen); | |
350 | } | |
4ad49000 LP |
351 | } |
352 | ||
fd870bac YW |
353 | int cgroup_add_device_allow(CGroupContext *c, const char *dev, const char *mode) { |
354 | _cleanup_free_ CGroupDeviceAllow *a = NULL; | |
355 | _cleanup_free_ char *d = NULL; | |
356 | ||
357 | assert(c); | |
358 | assert(dev); | |
359 | assert(isempty(mode) || in_charset(mode, "rwm")); | |
360 | ||
361 | a = new(CGroupDeviceAllow, 1); | |
362 | if (!a) | |
363 | return -ENOMEM; | |
364 | ||
365 | d = strdup(dev); | |
366 | if (!d) | |
367 | return -ENOMEM; | |
368 | ||
369 | *a = (CGroupDeviceAllow) { | |
370 | .path = TAKE_PTR(d), | |
490c5a37 LP |
371 | .r = isempty(mode) || strchr(mode, 'r'), |
372 | .w = isempty(mode) || strchr(mode, 'w'), | |
373 | .m = isempty(mode) || strchr(mode, 'm'), | |
fd870bac YW |
374 | }; |
375 | ||
376 | LIST_PREPEND(device_allow, c->device_allow, a); | |
377 | TAKE_PTR(a); | |
378 | ||
379 | return 0; | |
380 | } | |
381 | ||
0d2d6fbf CD |
382 | static void cgroup_xattr_apply(Unit *u) { |
383 | char ids[SD_ID128_STRING_MAX]; | |
384 | int r; | |
385 | ||
386 | assert(u); | |
387 | ||
388 | if (!MANAGER_IS_SYSTEM(u->manager)) | |
389 | return; | |
390 | ||
391 | if (sd_id128_is_null(u->invocation_id)) | |
392 | return; | |
393 | ||
394 | r = cg_set_xattr(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path, | |
395 | "trusted.invocation_id", | |
396 | sd_id128_to_string(u->invocation_id, ids), 32, | |
397 | 0); | |
398 | if (r < 0) | |
399 | log_unit_debug_errno(u, r, "Failed to set invocation ID on control group %s, ignoring: %m", u->cgroup_path); | |
400 | } | |
401 | ||
45c2e068 | 402 | static int lookup_block_device(const char *p, dev_t *ret) { |
f5855697 YS |
403 | dev_t rdev, dev = 0; |
404 | mode_t mode; | |
45c2e068 | 405 | int r; |
4ad49000 LP |
406 | |
407 | assert(p); | |
45c2e068 | 408 | assert(ret); |
4ad49000 | 409 | |
f5855697 | 410 | r = device_path_parse_major_minor(p, &mode, &rdev); |
d5aecba6 | 411 | if (r == -ENODEV) { /* not a parsable device node, need to go to disk */ |
f5855697 | 412 | struct stat st; |
d5aecba6 LP |
413 | if (stat(p, &st) < 0) |
414 | return log_warning_errno(errno, "Couldn't stat device '%s': %m", p); | |
f5855697 YS |
415 | rdev = (dev_t)st.st_rdev; |
416 | dev = (dev_t)st.st_dev; | |
417 | mode = st.st_mode; | |
d5aecba6 LP |
418 | } else if (r < 0) |
419 | return log_warning_errno(r, "Failed to parse major/minor from path '%s': %m", p); | |
420 | ||
f5855697 | 421 | if (S_ISCHR(mode)) { |
d5aecba6 LP |
422 | log_warning("Device node '%s' is a character device, but block device needed.", p); |
423 | return -ENOTBLK; | |
f5855697 YS |
424 | } else if (S_ISBLK(mode)) |
425 | *ret = rdev; | |
426 | else if (major(dev) != 0) | |
427 | *ret = dev; /* If this is not a device node then use the block device this file is stored on */ | |
45c2e068 LP |
428 | else { |
429 | /* If this is btrfs, getting the backing block device is a bit harder */ | |
430 | r = btrfs_get_block_device(p, ret); | |
431 | if (r < 0 && r != -ENOTTY) | |
432 | return log_warning_errno(r, "Failed to determine block device backing btrfs file system '%s': %m", p); | |
433 | if (r == -ENOTTY) { | |
434 | log_warning("'%s' is not a block device node, and file system block device cannot be determined or is not local.", p); | |
435 | return -ENODEV; | |
436 | } | |
4ad49000 | 437 | } |
8e274523 | 438 | |
45c2e068 LP |
439 | /* If this is a LUKS device, try to get the originating block device */ |
440 | (void) block_get_originating(*ret, ret); | |
441 | ||
442 | /* If this is a partition, try to get the originating block device */ | |
443 | (void) block_get_whole_disk(*ret, ret); | |
8e274523 | 444 | return 0; |
8e274523 LP |
445 | } |
446 | ||
084c7007 | 447 | static int whitelist_device(BPFProgram *prog, const char *path, const char *node, const char *acc) { |
f5855697 YS |
448 | dev_t rdev; |
449 | mode_t mode; | |
8c6db833 | 450 | int r; |
8e274523 | 451 | |
4ad49000 LP |
452 | assert(path); |
453 | assert(acc); | |
8e274523 | 454 | |
74c48bf5 LP |
455 | /* Some special handling for /dev/block/%u:%u, /dev/char/%u:%u, /run/systemd/inaccessible/chr and |
456 | * /run/systemd/inaccessible/blk paths. Instead of stat()ing these we parse out the major/minor directly. This | |
457 | * means clients can use these path without the device node actually around */ | |
f5855697 | 458 | r = device_path_parse_major_minor(node, &mode, &rdev); |
74c48bf5 LP |
459 | if (r < 0) { |
460 | if (r != -ENODEV) | |
461 | return log_warning_errno(r, "Couldn't parse major/minor from device path '%s': %m", node); | |
b200489b | 462 | |
f5855697 | 463 | struct stat st; |
74c48bf5 LP |
464 | if (stat(node, &st) < 0) |
465 | return log_warning_errno(errno, "Couldn't stat device %s: %m", node); | |
e7330dfe | 466 | |
74c48bf5 LP |
467 | if (!S_ISCHR(st.st_mode) && !S_ISBLK(st.st_mode)) { |
468 | log_warning("%s is not a device.", node); | |
469 | return -ENODEV; | |
470 | } | |
f5855697 YS |
471 | rdev = (dev_t) st.st_rdev; |
472 | mode = st.st_mode; | |
4ad49000 LP |
473 | } |
474 | ||
084c7007 RG |
475 | if (cg_all_unified() > 0) { |
476 | if (!prog) | |
477 | return 0; | |
4ad49000 | 478 | |
f5855697 YS |
479 | return cgroup_bpf_whitelist_device(prog, S_ISCHR(mode) ? BPF_DEVCG_DEV_CHAR : BPF_DEVCG_DEV_BLOCK, |
480 | major(rdev), minor(rdev), acc); | |
b9839ac9 | 481 | |
084c7007 RG |
482 | } else { |
483 | char buf[2+DECIMAL_STR_MAX(dev_t)*2+2+4]; | |
484 | ||
485 | sprintf(buf, | |
486 | "%c %u:%u %s", | |
f5855697 YS |
487 | S_ISCHR(mode) ? 'c' : 'b', |
488 | major(rdev), minor(rdev), | |
084c7007 RG |
489 | acc); |
490 | ||
8c838407 LP |
491 | /* Changing the devices list of a populated cgroup might result in EINVAL, hence ignore EINVAL here. */ |
492 | ||
084c7007 RG |
493 | r = cg_set_attribute("devices", path, "devices.allow", buf); |
494 | if (r < 0) | |
2c74e12b | 495 | return log_full_errno(IN_SET(r, -ENOENT, -EROFS, -EINVAL, -EACCES, -EPERM) ? LOG_DEBUG : LOG_WARNING, |
b9839ac9 | 496 | r, "Failed to set devices.allow on %s: %m", path); |
4ad49000 | 497 | |
b9839ac9 LP |
498 | return 0; |
499 | } | |
8e274523 LP |
500 | } |
501 | ||
084c7007 | 502 | static int whitelist_major(BPFProgram *prog, const char *path, const char *name, char type, const char *acc) { |
90060676 | 503 | _cleanup_fclose_ FILE *f = NULL; |
8e8b5d2e | 504 | char buf[2+DECIMAL_STR_MAX(unsigned)+3+4]; |
90060676 | 505 | bool good = false; |
8e8b5d2e | 506 | unsigned maj; |
90060676 LP |
507 | int r; |
508 | ||
509 | assert(path); | |
510 | assert(acc); | |
4c701096 | 511 | assert(IN_SET(type, 'b', 'c')); |
90060676 | 512 | |
8e8b5d2e LP |
513 | if (streq(name, "*")) { |
514 | /* If the name is a wildcard, then apply this list to all devices of this type */ | |
515 | ||
516 | if (cg_all_unified() > 0) { | |
517 | if (!prog) | |
518 | return 0; | |
519 | ||
520 | (void) cgroup_bpf_whitelist_class(prog, type == 'c' ? BPF_DEVCG_DEV_CHAR : BPF_DEVCG_DEV_BLOCK, acc); | |
521 | } else { | |
522 | xsprintf(buf, "%c *:* %s", type, acc); | |
523 | ||
524 | r = cg_set_attribute("devices", path, "devices.allow", buf); | |
525 | if (r < 0) | |
526 | log_full_errno(IN_SET(r, -ENOENT, -EROFS, -EINVAL, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, | |
527 | "Failed to set devices.allow on %s: %m", path); | |
528 | return 0; | |
529 | } | |
530 | } | |
531 | ||
532 | if (safe_atou(name, &maj) >= 0 && DEVICE_MAJOR_VALID(maj)) { | |
533 | /* The name is numeric and suitable as major. In that case, let's take is major, and create the entry | |
534 | * directly */ | |
535 | ||
536 | if (cg_all_unified() > 0) { | |
537 | if (!prog) | |
538 | return 0; | |
539 | ||
540 | (void) cgroup_bpf_whitelist_major(prog, | |
541 | type == 'c' ? BPF_DEVCG_DEV_CHAR : BPF_DEVCG_DEV_BLOCK, | |
542 | maj, acc); | |
543 | } else { | |
544 | xsprintf(buf, "%c %u:* %s", type, maj, acc); | |
545 | ||
546 | r = cg_set_attribute("devices", path, "devices.allow", buf); | |
547 | if (r < 0) | |
548 | log_full_errno(IN_SET(r, -ENOENT, -EROFS, -EINVAL, -EACCES) ? LOG_DEBUG : LOG_WARNING, r, | |
549 | "Failed to set devices.allow on %s: %m", path); | |
550 | } | |
551 | ||
552 | return 0; | |
553 | } | |
554 | ||
90060676 | 555 | f = fopen("/proc/devices", "re"); |
4a62c710 MS |
556 | if (!f) |
557 | return log_warning_errno(errno, "Cannot open /proc/devices to resolve %s (%c): %m", name, type); | |
90060676 | 558 | |
c66e60a8 LP |
559 | for (;;) { |
560 | _cleanup_free_ char *line = NULL; | |
8e8b5d2e | 561 | char *w, *p; |
90060676 | 562 | |
c66e60a8 LP |
563 | r = read_line(f, LONG_LINE_MAX, &line); |
564 | if (r < 0) | |
565 | return log_warning_errno(r, "Failed to read /proc/devices: %m"); | |
566 | if (r == 0) | |
567 | break; | |
90060676 LP |
568 | |
569 | if (type == 'c' && streq(line, "Character devices:")) { | |
570 | good = true; | |
571 | continue; | |
572 | } | |
573 | ||
574 | if (type == 'b' && streq(line, "Block devices:")) { | |
575 | good = true; | |
576 | continue; | |
577 | } | |
578 | ||
579 | if (isempty(line)) { | |
580 | good = false; | |
581 | continue; | |
582 | } | |
583 | ||
584 | if (!good) | |
585 | continue; | |
586 | ||
587 | p = strstrip(line); | |
588 | ||
589 | w = strpbrk(p, WHITESPACE); | |
590 | if (!w) | |
591 | continue; | |
592 | *w = 0; | |
593 | ||
594 | r = safe_atou(p, &maj); | |
595 | if (r < 0) | |
596 | continue; | |
597 | if (maj <= 0) | |
598 | continue; | |
599 | ||
600 | w++; | |
601 | w += strspn(w, WHITESPACE); | |
e41969e3 LP |
602 | |
603 | if (fnmatch(name, w, 0) != 0) | |
90060676 LP |
604 | continue; |
605 | ||
084c7007 RG |
606 | if (cg_all_unified() > 0) { |
607 | if (!prog) | |
608 | continue; | |
90060676 | 609 | |
913c898c LP |
610 | (void) cgroup_bpf_whitelist_major(prog, |
611 | type == 'c' ? BPF_DEVCG_DEV_CHAR : BPF_DEVCG_DEV_BLOCK, | |
612 | maj, acc); | |
084c7007 | 613 | } else { |
084c7007 RG |
614 | sprintf(buf, |
615 | "%c %u:* %s", | |
616 | type, | |
617 | maj, | |
618 | acc); | |
619 | ||
8c838407 LP |
620 | /* Changing the devices list of a populated cgroup might result in EINVAL, hence ignore EINVAL |
621 | * here. */ | |
622 | ||
084c7007 RG |
623 | r = cg_set_attribute("devices", path, "devices.allow", buf); |
624 | if (r < 0) | |
2c74e12b | 625 | log_full_errno(IN_SET(r, -ENOENT, -EROFS, -EINVAL, -EACCES, -EPERM) ? LOG_DEBUG : LOG_WARNING, |
084c7007 RG |
626 | r, "Failed to set devices.allow on %s: %m", path); |
627 | } | |
90060676 LP |
628 | } |
629 | ||
630 | return 0; | |
90060676 LP |
631 | } |
632 | ||
66ebf6c0 TH |
633 | static bool cgroup_context_has_cpu_weight(CGroupContext *c) { |
634 | return c->cpu_weight != CGROUP_WEIGHT_INVALID || | |
635 | c->startup_cpu_weight != CGROUP_WEIGHT_INVALID; | |
636 | } | |
637 | ||
638 | static bool cgroup_context_has_cpu_shares(CGroupContext *c) { | |
639 | return c->cpu_shares != CGROUP_CPU_SHARES_INVALID || | |
640 | c->startup_cpu_shares != CGROUP_CPU_SHARES_INVALID; | |
641 | } | |
642 | ||
643 | static uint64_t cgroup_context_cpu_weight(CGroupContext *c, ManagerState state) { | |
644 | if (IN_SET(state, MANAGER_STARTING, MANAGER_INITIALIZING) && | |
645 | c->startup_cpu_weight != CGROUP_WEIGHT_INVALID) | |
646 | return c->startup_cpu_weight; | |
647 | else if (c->cpu_weight != CGROUP_WEIGHT_INVALID) | |
648 | return c->cpu_weight; | |
649 | else | |
650 | return CGROUP_WEIGHT_DEFAULT; | |
651 | } | |
652 | ||
653 | static uint64_t cgroup_context_cpu_shares(CGroupContext *c, ManagerState state) { | |
654 | if (IN_SET(state, MANAGER_STARTING, MANAGER_INITIALIZING) && | |
655 | c->startup_cpu_shares != CGROUP_CPU_SHARES_INVALID) | |
656 | return c->startup_cpu_shares; | |
657 | else if (c->cpu_shares != CGROUP_CPU_SHARES_INVALID) | |
658 | return c->cpu_shares; | |
659 | else | |
660 | return CGROUP_CPU_SHARES_DEFAULT; | |
661 | } | |
662 | ||
10f28641 FB |
663 | usec_t cgroup_cpu_adjust_period(usec_t period, usec_t quota, usec_t resolution, usec_t max_period) { |
664 | /* kernel uses a minimum resolution of 1ms, so both period and (quota * period) | |
665 | * need to be higher than that boundary. quota is specified in USecPerSec. | |
666 | * Additionally, period must be at most max_period. */ | |
667 | assert(quota > 0); | |
668 | ||
669 | return MIN(MAX3(period, resolution, resolution * USEC_PER_SEC / quota), max_period); | |
670 | } | |
671 | ||
672 | static usec_t cgroup_cpu_adjust_period_and_log(Unit *u, usec_t period, usec_t quota) { | |
673 | usec_t new_period; | |
674 | ||
675 | if (quota == USEC_INFINITY) | |
676 | /* Always use default period for infinity quota. */ | |
677 | return CGROUP_CPU_QUOTA_DEFAULT_PERIOD_USEC; | |
678 | ||
679 | if (period == USEC_INFINITY) | |
680 | /* Default period was requested. */ | |
681 | period = CGROUP_CPU_QUOTA_DEFAULT_PERIOD_USEC; | |
682 | ||
683 | /* Clamp to interval [1ms, 1s] */ | |
684 | new_period = cgroup_cpu_adjust_period(period, quota, USEC_PER_MSEC, USEC_PER_SEC); | |
685 | ||
686 | if (new_period != period) { | |
687 | char v[FORMAT_TIMESPAN_MAX]; | |
527ede0c | 688 | log_unit_full(u, u->warned_clamping_cpu_quota_period ? LOG_DEBUG : LOG_WARNING, 0, |
10f28641 FB |
689 | "Clamping CPU interval for cpu.max: period is now %s", |
690 | format_timespan(v, sizeof(v), new_period, 1)); | |
527ede0c | 691 | u->warned_clamping_cpu_quota_period = true; |
10f28641 FB |
692 | } |
693 | ||
694 | return new_period; | |
695 | } | |
696 | ||
52fecf20 LP |
697 | static void cgroup_apply_unified_cpu_weight(Unit *u, uint64_t weight) { |
698 | char buf[DECIMAL_STR_MAX(uint64_t) + 2]; | |
66ebf6c0 TH |
699 | |
700 | xsprintf(buf, "%" PRIu64 "\n", weight); | |
293d32df | 701 | (void) set_attribute_and_warn(u, "cpu", "cpu.weight", buf); |
52fecf20 LP |
702 | } |
703 | ||
10f28641 | 704 | static void cgroup_apply_unified_cpu_quota(Unit *u, usec_t quota, usec_t period) { |
52fecf20 | 705 | char buf[(DECIMAL_STR_MAX(usec_t) + 1) * 2 + 1]; |
66ebf6c0 | 706 | |
10f28641 | 707 | period = cgroup_cpu_adjust_period_and_log(u, period, quota); |
66ebf6c0 TH |
708 | if (quota != USEC_INFINITY) |
709 | xsprintf(buf, USEC_FMT " " USEC_FMT "\n", | |
10f28641 | 710 | MAX(quota * period / USEC_PER_SEC, USEC_PER_MSEC), period); |
66ebf6c0 | 711 | else |
10f28641 | 712 | xsprintf(buf, "max " USEC_FMT "\n", period); |
293d32df | 713 | (void) set_attribute_and_warn(u, "cpu", "cpu.max", buf); |
66ebf6c0 TH |
714 | } |
715 | ||
52fecf20 LP |
716 | static void cgroup_apply_legacy_cpu_shares(Unit *u, uint64_t shares) { |
717 | char buf[DECIMAL_STR_MAX(uint64_t) + 2]; | |
66ebf6c0 TH |
718 | |
719 | xsprintf(buf, "%" PRIu64 "\n", shares); | |
293d32df | 720 | (void) set_attribute_and_warn(u, "cpu", "cpu.shares", buf); |
52fecf20 LP |
721 | } |
722 | ||
10f28641 | 723 | static void cgroup_apply_legacy_cpu_quota(Unit *u, usec_t quota, usec_t period) { |
52fecf20 | 724 | char buf[DECIMAL_STR_MAX(usec_t) + 2]; |
66ebf6c0 | 725 | |
10f28641 FB |
726 | period = cgroup_cpu_adjust_period_and_log(u, period, quota); |
727 | ||
728 | xsprintf(buf, USEC_FMT "\n", period); | |
293d32df | 729 | (void) set_attribute_and_warn(u, "cpu", "cpu.cfs_period_us", buf); |
66ebf6c0 TH |
730 | |
731 | if (quota != USEC_INFINITY) { | |
10f28641 | 732 | xsprintf(buf, USEC_FMT "\n", MAX(quota * period / USEC_PER_SEC, USEC_PER_MSEC)); |
293d32df | 733 | (void) set_attribute_and_warn(u, "cpu", "cpu.cfs_quota_us", buf); |
66ebf6c0 | 734 | } else |
589a5f7a | 735 | (void) set_attribute_and_warn(u, "cpu", "cpu.cfs_quota_us", "-1\n"); |
66ebf6c0 TH |
736 | } |
737 | ||
738 | static uint64_t cgroup_cpu_shares_to_weight(uint64_t shares) { | |
739 | return CLAMP(shares * CGROUP_WEIGHT_DEFAULT / CGROUP_CPU_SHARES_DEFAULT, | |
740 | CGROUP_WEIGHT_MIN, CGROUP_WEIGHT_MAX); | |
741 | } | |
742 | ||
743 | static uint64_t cgroup_cpu_weight_to_shares(uint64_t weight) { | |
744 | return CLAMP(weight * CGROUP_CPU_SHARES_DEFAULT / CGROUP_WEIGHT_DEFAULT, | |
745 | CGROUP_CPU_SHARES_MIN, CGROUP_CPU_SHARES_MAX); | |
746 | } | |
747 | ||
508c45da | 748 | static bool cgroup_context_has_io_config(CGroupContext *c) { |
538b4852 TH |
749 | return c->io_accounting || |
750 | c->io_weight != CGROUP_WEIGHT_INVALID || | |
751 | c->startup_io_weight != CGROUP_WEIGHT_INVALID || | |
752 | c->io_device_weights || | |
6ae4283c | 753 | c->io_device_latencies || |
538b4852 TH |
754 | c->io_device_limits; |
755 | } | |
756 | ||
508c45da | 757 | static bool cgroup_context_has_blockio_config(CGroupContext *c) { |
538b4852 TH |
758 | return c->blockio_accounting || |
759 | c->blockio_weight != CGROUP_BLKIO_WEIGHT_INVALID || | |
760 | c->startup_blockio_weight != CGROUP_BLKIO_WEIGHT_INVALID || | |
761 | c->blockio_device_weights || | |
762 | c->blockio_device_bandwidths; | |
763 | } | |
764 | ||
508c45da | 765 | static uint64_t cgroup_context_io_weight(CGroupContext *c, ManagerState state) { |
64faf04c TH |
766 | if (IN_SET(state, MANAGER_STARTING, MANAGER_INITIALIZING) && |
767 | c->startup_io_weight != CGROUP_WEIGHT_INVALID) | |
768 | return c->startup_io_weight; | |
769 | else if (c->io_weight != CGROUP_WEIGHT_INVALID) | |
770 | return c->io_weight; | |
771 | else | |
772 | return CGROUP_WEIGHT_DEFAULT; | |
773 | } | |
774 | ||
508c45da | 775 | static uint64_t cgroup_context_blkio_weight(CGroupContext *c, ManagerState state) { |
64faf04c TH |
776 | if (IN_SET(state, MANAGER_STARTING, MANAGER_INITIALIZING) && |
777 | c->startup_blockio_weight != CGROUP_BLKIO_WEIGHT_INVALID) | |
778 | return c->startup_blockio_weight; | |
779 | else if (c->blockio_weight != CGROUP_BLKIO_WEIGHT_INVALID) | |
780 | return c->blockio_weight; | |
781 | else | |
782 | return CGROUP_BLKIO_WEIGHT_DEFAULT; | |
783 | } | |
784 | ||
508c45da | 785 | static uint64_t cgroup_weight_blkio_to_io(uint64_t blkio_weight) { |
538b4852 TH |
786 | return CLAMP(blkio_weight * CGROUP_WEIGHT_DEFAULT / CGROUP_BLKIO_WEIGHT_DEFAULT, |
787 | CGROUP_WEIGHT_MIN, CGROUP_WEIGHT_MAX); | |
788 | } | |
789 | ||
508c45da | 790 | static uint64_t cgroup_weight_io_to_blkio(uint64_t io_weight) { |
538b4852 TH |
791 | return CLAMP(io_weight * CGROUP_BLKIO_WEIGHT_DEFAULT / CGROUP_WEIGHT_DEFAULT, |
792 | CGROUP_BLKIO_WEIGHT_MIN, CGROUP_BLKIO_WEIGHT_MAX); | |
793 | } | |
794 | ||
f29ff115 | 795 | static void cgroup_apply_io_device_weight(Unit *u, const char *dev_path, uint64_t io_weight) { |
64faf04c TH |
796 | char buf[DECIMAL_STR_MAX(dev_t)*2+2+DECIMAL_STR_MAX(uint64_t)+1]; |
797 | dev_t dev; | |
798 | int r; | |
799 | ||
800 | r = lookup_block_device(dev_path, &dev); | |
801 | if (r < 0) | |
802 | return; | |
803 | ||
804 | xsprintf(buf, "%u:%u %" PRIu64 "\n", major(dev), minor(dev), io_weight); | |
293d32df | 805 | (void) set_attribute_and_warn(u, "io", "io.weight", buf); |
64faf04c TH |
806 | } |
807 | ||
f29ff115 | 808 | static void cgroup_apply_blkio_device_weight(Unit *u, const char *dev_path, uint64_t blkio_weight) { |
64faf04c TH |
809 | char buf[DECIMAL_STR_MAX(dev_t)*2+2+DECIMAL_STR_MAX(uint64_t)+1]; |
810 | dev_t dev; | |
811 | int r; | |
812 | ||
813 | r = lookup_block_device(dev_path, &dev); | |
814 | if (r < 0) | |
815 | return; | |
816 | ||
817 | xsprintf(buf, "%u:%u %" PRIu64 "\n", major(dev), minor(dev), blkio_weight); | |
293d32df | 818 | (void) set_attribute_and_warn(u, "blkio", "blkio.weight_device", buf); |
64faf04c TH |
819 | } |
820 | ||
6ae4283c TH |
821 | static void cgroup_apply_io_device_latency(Unit *u, const char *dev_path, usec_t target) { |
822 | char buf[DECIMAL_STR_MAX(dev_t)*2+2+7+DECIMAL_STR_MAX(uint64_t)+1]; | |
823 | dev_t dev; | |
824 | int r; | |
825 | ||
826 | r = lookup_block_device(dev_path, &dev); | |
827 | if (r < 0) | |
828 | return; | |
829 | ||
830 | if (target != USEC_INFINITY) | |
831 | xsprintf(buf, "%u:%u target=%" PRIu64 "\n", major(dev), minor(dev), target); | |
832 | else | |
833 | xsprintf(buf, "%u:%u target=max\n", major(dev), minor(dev)); | |
834 | ||
293d32df | 835 | (void) set_attribute_and_warn(u, "io", "io.latency", buf); |
6ae4283c TH |
836 | } |
837 | ||
17ae2780 | 838 | static void cgroup_apply_io_device_limit(Unit *u, const char *dev_path, uint64_t *limits) { |
64faf04c TH |
839 | char limit_bufs[_CGROUP_IO_LIMIT_TYPE_MAX][DECIMAL_STR_MAX(uint64_t)]; |
840 | char buf[DECIMAL_STR_MAX(dev_t)*2+2+(6+DECIMAL_STR_MAX(uint64_t)+1)*4]; | |
841 | CGroupIOLimitType type; | |
842 | dev_t dev; | |
64faf04c TH |
843 | int r; |
844 | ||
845 | r = lookup_block_device(dev_path, &dev); | |
846 | if (r < 0) | |
17ae2780 | 847 | return; |
64faf04c | 848 | |
17ae2780 LP |
849 | for (type = 0; type < _CGROUP_IO_LIMIT_TYPE_MAX; type++) |
850 | if (limits[type] != cgroup_io_limit_defaults[type]) | |
64faf04c | 851 | xsprintf(limit_bufs[type], "%" PRIu64, limits[type]); |
17ae2780 | 852 | else |
64faf04c | 853 | xsprintf(limit_bufs[type], "%s", limits[type] == CGROUP_LIMIT_MAX ? "max" : "0"); |
64faf04c TH |
854 | |
855 | xsprintf(buf, "%u:%u rbps=%s wbps=%s riops=%s wiops=%s\n", major(dev), minor(dev), | |
856 | limit_bufs[CGROUP_IO_RBPS_MAX], limit_bufs[CGROUP_IO_WBPS_MAX], | |
857 | limit_bufs[CGROUP_IO_RIOPS_MAX], limit_bufs[CGROUP_IO_WIOPS_MAX]); | |
293d32df | 858 | (void) set_attribute_and_warn(u, "io", "io.max", buf); |
64faf04c TH |
859 | } |
860 | ||
17ae2780 | 861 | static void cgroup_apply_blkio_device_limit(Unit *u, const char *dev_path, uint64_t rbps, uint64_t wbps) { |
64faf04c TH |
862 | char buf[DECIMAL_STR_MAX(dev_t)*2+2+DECIMAL_STR_MAX(uint64_t)+1]; |
863 | dev_t dev; | |
64faf04c TH |
864 | int r; |
865 | ||
866 | r = lookup_block_device(dev_path, &dev); | |
867 | if (r < 0) | |
17ae2780 | 868 | return; |
64faf04c | 869 | |
64faf04c | 870 | sprintf(buf, "%u:%u %" PRIu64 "\n", major(dev), minor(dev), rbps); |
293d32df | 871 | (void) set_attribute_and_warn(u, "blkio", "blkio.throttle.read_bps_device", buf); |
64faf04c | 872 | |
64faf04c | 873 | sprintf(buf, "%u:%u %" PRIu64 "\n", major(dev), minor(dev), wbps); |
293d32df | 874 | (void) set_attribute_and_warn(u, "blkio", "blkio.throttle.write_bps_device", buf); |
64faf04c TH |
875 | } |
876 | ||
da4d897e | 877 | static bool cgroup_context_has_unified_memory_config(CGroupContext *c) { |
48422635 | 878 | 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 |
879 | } |
880 | ||
f29ff115 | 881 | static void cgroup_apply_unified_memory_limit(Unit *u, const char *file, uint64_t v) { |
589a5f7a | 882 | char buf[DECIMAL_STR_MAX(uint64_t) + 1] = "max\n"; |
da4d897e TH |
883 | |
884 | if (v != CGROUP_LIMIT_MAX) | |
885 | xsprintf(buf, "%" PRIu64 "\n", v); | |
886 | ||
293d32df | 887 | (void) set_attribute_and_warn(u, "memory", file, buf); |
da4d897e TH |
888 | } |
889 | ||
0f2d84d2 | 890 | static void cgroup_apply_firewall(Unit *u) { |
0f2d84d2 LP |
891 | assert(u); |
892 | ||
acf7f253 | 893 | /* Best-effort: let's apply IP firewalling and/or accounting if that's enabled */ |
906c06f6 | 894 | |
acf7f253 | 895 | if (bpf_firewall_compile(u) < 0) |
906c06f6 DM |
896 | return; |
897 | ||
898 | (void) bpf_firewall_install(u); | |
906c06f6 DM |
899 | } |
900 | ||
901 | static void cgroup_context_apply( | |
902 | Unit *u, | |
903 | CGroupMask apply_mask, | |
906c06f6 DM |
904 | ManagerState state) { |
905 | ||
f29ff115 TH |
906 | const char *path; |
907 | CGroupContext *c; | |
52fecf20 | 908 | bool is_host_root, is_local_root; |
4ad49000 LP |
909 | int r; |
910 | ||
f29ff115 TH |
911 | assert(u); |
912 | ||
906c06f6 | 913 | /* Nothing to do? Exit early! */ |
17f14955 | 914 | if (apply_mask == 0) |
4ad49000 | 915 | return; |
8e274523 | 916 | |
52fecf20 LP |
917 | /* Some cgroup attributes are not supported on the host root cgroup, hence silently ignore them here. And other |
918 | * attributes should only be managed for cgroups further down the tree. */ | |
919 | is_local_root = unit_has_name(u, SPECIAL_ROOT_SLICE); | |
920 | is_host_root = unit_has_host_root_cgroup(u); | |
f3725e64 LP |
921 | |
922 | assert_se(c = unit_get_cgroup_context(u)); | |
923 | assert_se(path = u->cgroup_path); | |
924 | ||
52fecf20 | 925 | if (is_local_root) /* Make sure we don't try to display messages with an empty path. */ |
6da13913 | 926 | path = "/"; |
01efdf13 | 927 | |
be2c0327 LP |
928 | /* We generally ignore errors caused by read-only mounted cgroup trees (assuming we are running in a container |
929 | * then), and missing cgroups, i.e. EROFS and ENOENT. */ | |
714e2e1d | 930 | |
be2c0327 LP |
931 | /* In fully unified mode these attributes don't exist on the host cgroup root. On legacy the weights exist, but |
932 | * setting the weight makes very little sense on the host root cgroup, as there are no other cgroups at this | |
933 | * level. The quota exists there too, but any attempt to write to it is refused with EINVAL. Inside of | |
4e1dfa45 | 934 | * containers we want to leave control of these to the container manager (and if cgroup v2 delegation is used |
be2c0327 LP |
935 | * we couldn't even write to them if we wanted to). */ |
936 | if ((apply_mask & CGROUP_MASK_CPU) && !is_local_root) { | |
8e274523 | 937 | |
b4cccbc1 | 938 | if (cg_all_unified() > 0) { |
be2c0327 | 939 | uint64_t weight; |
b2f8b02e | 940 | |
be2c0327 LP |
941 | if (cgroup_context_has_cpu_weight(c)) |
942 | weight = cgroup_context_cpu_weight(c, state); | |
943 | else if (cgroup_context_has_cpu_shares(c)) { | |
944 | uint64_t shares; | |
66ebf6c0 | 945 | |
be2c0327 LP |
946 | shares = cgroup_context_cpu_shares(c, state); |
947 | weight = cgroup_cpu_shares_to_weight(shares); | |
66ebf6c0 | 948 | |
be2c0327 LP |
949 | log_cgroup_compat(u, "Applying [Startup]CPUShares=%" PRIu64 " as [Startup]CPUWeight=%" PRIu64 " on %s", |
950 | shares, weight, path); | |
951 | } else | |
952 | weight = CGROUP_WEIGHT_DEFAULT; | |
66ebf6c0 | 953 | |
be2c0327 | 954 | cgroup_apply_unified_cpu_weight(u, weight); |
10f28641 | 955 | cgroup_apply_unified_cpu_quota(u, c->cpu_quota_per_sec_usec, c->cpu_quota_period_usec); |
66ebf6c0 | 956 | |
52fecf20 | 957 | } else { |
be2c0327 | 958 | uint64_t shares; |
52fecf20 | 959 | |
be2c0327 LP |
960 | if (cgroup_context_has_cpu_weight(c)) { |
961 | uint64_t weight; | |
52fecf20 | 962 | |
be2c0327 LP |
963 | weight = cgroup_context_cpu_weight(c, state); |
964 | shares = cgroup_cpu_weight_to_shares(weight); | |
52fecf20 | 965 | |
be2c0327 LP |
966 | log_cgroup_compat(u, "Applying [Startup]CPUWeight=%" PRIu64 " as [Startup]CPUShares=%" PRIu64 " on %s", |
967 | weight, shares, path); | |
968 | } else if (cgroup_context_has_cpu_shares(c)) | |
969 | shares = cgroup_context_cpu_shares(c, state); | |
970 | else | |
971 | shares = CGROUP_CPU_SHARES_DEFAULT; | |
66ebf6c0 | 972 | |
be2c0327 | 973 | cgroup_apply_legacy_cpu_shares(u, shares); |
10f28641 | 974 | cgroup_apply_legacy_cpu_quota(u, c->cpu_quota_per_sec_usec, c->cpu_quota_period_usec); |
66ebf6c0 | 975 | } |
4ad49000 LP |
976 | } |
977 | ||
4e1dfa45 | 978 | /* The 'io' controller attributes are not exported on the host's root cgroup (being a pure cgroup v2 |
52fecf20 LP |
979 | * controller), and in case of containers we want to leave control of these attributes to the container manager |
980 | * (and we couldn't access that stuff anyway, even if we tried if proper delegation is used). */ | |
981 | if ((apply_mask & CGROUP_MASK_IO) && !is_local_root) { | |
982 | char buf[8+DECIMAL_STR_MAX(uint64_t)+1]; | |
983 | bool has_io, has_blockio; | |
984 | uint64_t weight; | |
13c31542 | 985 | |
52fecf20 LP |
986 | has_io = cgroup_context_has_io_config(c); |
987 | has_blockio = cgroup_context_has_blockio_config(c); | |
13c31542 | 988 | |
52fecf20 LP |
989 | if (has_io) |
990 | weight = cgroup_context_io_weight(c, state); | |
991 | else if (has_blockio) { | |
992 | uint64_t blkio_weight; | |
128fadc9 | 993 | |
52fecf20 LP |
994 | blkio_weight = cgroup_context_blkio_weight(c, state); |
995 | weight = cgroup_weight_blkio_to_io(blkio_weight); | |
128fadc9 | 996 | |
67e2ea15 | 997 | log_cgroup_compat(u, "Applying [Startup]BlockIOWeight=%" PRIu64 " as [Startup]IOWeight=%" PRIu64, |
52fecf20 LP |
998 | blkio_weight, weight); |
999 | } else | |
1000 | weight = CGROUP_WEIGHT_DEFAULT; | |
13c31542 | 1001 | |
52fecf20 LP |
1002 | xsprintf(buf, "default %" PRIu64 "\n", weight); |
1003 | (void) set_attribute_and_warn(u, "io", "io.weight", buf); | |
538b4852 | 1004 | |
52fecf20 LP |
1005 | if (has_io) { |
1006 | CGroupIODeviceLatency *latency; | |
1007 | CGroupIODeviceLimit *limit; | |
1008 | CGroupIODeviceWeight *w; | |
128fadc9 | 1009 | |
52fecf20 LP |
1010 | LIST_FOREACH(device_weights, w, c->io_device_weights) |
1011 | cgroup_apply_io_device_weight(u, w->path, w->weight); | |
128fadc9 | 1012 | |
52fecf20 LP |
1013 | LIST_FOREACH(device_limits, limit, c->io_device_limits) |
1014 | cgroup_apply_io_device_limit(u, limit->path, limit->limits); | |
6ae4283c | 1015 | |
52fecf20 LP |
1016 | LIST_FOREACH(device_latencies, latency, c->io_device_latencies) |
1017 | cgroup_apply_io_device_latency(u, latency->path, latency->target_usec); | |
6ae4283c | 1018 | |
52fecf20 LP |
1019 | } else if (has_blockio) { |
1020 | CGroupBlockIODeviceWeight *w; | |
1021 | CGroupBlockIODeviceBandwidth *b; | |
13c31542 | 1022 | |
52fecf20 LP |
1023 | LIST_FOREACH(device_weights, w, c->blockio_device_weights) { |
1024 | weight = cgroup_weight_blkio_to_io(w->weight); | |
17ae2780 | 1025 | |
67e2ea15 | 1026 | log_cgroup_compat(u, "Applying BlockIODeviceWeight=%" PRIu64 " as IODeviceWeight=%" PRIu64 " for %s", |
52fecf20 | 1027 | w->weight, weight, w->path); |
538b4852 | 1028 | |
52fecf20 LP |
1029 | cgroup_apply_io_device_weight(u, w->path, weight); |
1030 | } | |
538b4852 | 1031 | |
17ae2780 | 1032 | LIST_FOREACH(device_bandwidths, b, c->blockio_device_bandwidths) { |
538b4852 TH |
1033 | uint64_t limits[_CGROUP_IO_LIMIT_TYPE_MAX]; |
1034 | CGroupIOLimitType type; | |
1035 | ||
1036 | for (type = 0; type < _CGROUP_IO_LIMIT_TYPE_MAX; type++) | |
1037 | limits[type] = cgroup_io_limit_defaults[type]; | |
1038 | ||
1039 | limits[CGROUP_IO_RBPS_MAX] = b->rbps; | |
1040 | limits[CGROUP_IO_WBPS_MAX] = b->wbps; | |
1041 | ||
67e2ea15 | 1042 | log_cgroup_compat(u, "Applying BlockIO{Read|Write}Bandwidth=%" PRIu64 " %" PRIu64 " as IO{Read|Write}BandwidthMax= for %s", |
128fadc9 TH |
1043 | b->rbps, b->wbps, b->path); |
1044 | ||
17ae2780 | 1045 | cgroup_apply_io_device_limit(u, b->path, limits); |
538b4852 | 1046 | } |
13c31542 TH |
1047 | } |
1048 | } | |
1049 | ||
906c06f6 | 1050 | if (apply_mask & CGROUP_MASK_BLKIO) { |
52fecf20 | 1051 | bool has_io, has_blockio; |
4ad49000 | 1052 | |
52fecf20 LP |
1053 | has_io = cgroup_context_has_io_config(c); |
1054 | has_blockio = cgroup_context_has_blockio_config(c); | |
1055 | ||
1056 | /* Applying a 'weight' never makes sense for the host root cgroup, and for containers this should be | |
1057 | * left to our container manager, too. */ | |
1058 | if (!is_local_root) { | |
64faf04c TH |
1059 | char buf[DECIMAL_STR_MAX(uint64_t)+1]; |
1060 | uint64_t weight; | |
64faf04c | 1061 | |
7d862ab8 | 1062 | if (has_io) { |
52fecf20 | 1063 | uint64_t io_weight; |
128fadc9 | 1064 | |
52fecf20 | 1065 | io_weight = cgroup_context_io_weight(c, state); |
538b4852 | 1066 | weight = cgroup_weight_io_to_blkio(cgroup_context_io_weight(c, state)); |
128fadc9 | 1067 | |
67e2ea15 | 1068 | log_cgroup_compat(u, "Applying [Startup]IOWeight=%" PRIu64 " as [Startup]BlockIOWeight=%" PRIu64, |
128fadc9 | 1069 | io_weight, weight); |
7d862ab8 TH |
1070 | } else if (has_blockio) |
1071 | weight = cgroup_context_blkio_weight(c, state); | |
1072 | else | |
538b4852 | 1073 | weight = CGROUP_BLKIO_WEIGHT_DEFAULT; |
64faf04c TH |
1074 | |
1075 | xsprintf(buf, "%" PRIu64 "\n", weight); | |
293d32df | 1076 | (void) set_attribute_and_warn(u, "blkio", "blkio.weight", buf); |
4ad49000 | 1077 | |
7d862ab8 | 1078 | if (has_io) { |
538b4852 TH |
1079 | CGroupIODeviceWeight *w; |
1080 | ||
128fadc9 TH |
1081 | LIST_FOREACH(device_weights, w, c->io_device_weights) { |
1082 | weight = cgroup_weight_io_to_blkio(w->weight); | |
1083 | ||
67e2ea15 | 1084 | log_cgroup_compat(u, "Applying IODeviceWeight=%" PRIu64 " as BlockIODeviceWeight=%" PRIu64 " for %s", |
128fadc9 TH |
1085 | w->weight, weight, w->path); |
1086 | ||
1087 | cgroup_apply_blkio_device_weight(u, w->path, weight); | |
1088 | } | |
7d862ab8 TH |
1089 | } else if (has_blockio) { |
1090 | CGroupBlockIODeviceWeight *w; | |
1091 | ||
7d862ab8 TH |
1092 | LIST_FOREACH(device_weights, w, c->blockio_device_weights) |
1093 | cgroup_apply_blkio_device_weight(u, w->path, w->weight); | |
538b4852 | 1094 | } |
4ad49000 LP |
1095 | } |
1096 | ||
52fecf20 LP |
1097 | /* The bandwith limits are something that make sense to be applied to the host's root but not container |
1098 | * roots, as there we want the container manager to handle it */ | |
1099 | if (is_host_root || !is_local_root) { | |
1100 | if (has_io) { | |
1101 | CGroupIODeviceLimit *l; | |
538b4852 | 1102 | |
52fecf20 | 1103 | LIST_FOREACH(device_limits, l, c->io_device_limits) { |
67e2ea15 | 1104 | log_cgroup_compat(u, "Applying IO{Read|Write}Bandwidth=%" PRIu64 " %" PRIu64 " as BlockIO{Read|Write}BandwidthMax= for %s", |
52fecf20 | 1105 | l->limits[CGROUP_IO_RBPS_MAX], l->limits[CGROUP_IO_WBPS_MAX], l->path); |
128fadc9 | 1106 | |
52fecf20 LP |
1107 | cgroup_apply_blkio_device_limit(u, l->path, l->limits[CGROUP_IO_RBPS_MAX], l->limits[CGROUP_IO_WBPS_MAX]); |
1108 | } | |
1109 | } else if (has_blockio) { | |
1110 | CGroupBlockIODeviceBandwidth *b; | |
7d862ab8 | 1111 | |
52fecf20 LP |
1112 | LIST_FOREACH(device_bandwidths, b, c->blockio_device_bandwidths) |
1113 | cgroup_apply_blkio_device_limit(u, b->path, b->rbps, b->wbps); | |
1114 | } | |
d686d8a9 | 1115 | } |
8e274523 LP |
1116 | } |
1117 | ||
be2c0327 LP |
1118 | /* In unified mode 'memory' attributes do not exist on the root cgroup. In legacy mode 'memory.limit_in_bytes' |
1119 | * exists on the root cgroup, but any writes to it are refused with EINVAL. And if we run in a container we | |
4e1dfa45 | 1120 | * want to leave control to the container manager (and if proper cgroup v2 delegation is used we couldn't even |
be2c0327 LP |
1121 | * write to this if we wanted to.) */ |
1122 | if ((apply_mask & CGROUP_MASK_MEMORY) && !is_local_root) { | |
efdb0237 | 1123 | |
52fecf20 | 1124 | if (cg_all_unified() > 0) { |
be2c0327 LP |
1125 | uint64_t max, swap_max = CGROUP_LIMIT_MAX; |
1126 | ||
1127 | if (cgroup_context_has_unified_memory_config(c)) { | |
1128 | max = c->memory_max; | |
1129 | swap_max = c->memory_swap_max; | |
1130 | } else { | |
1131 | max = c->memory_limit; | |
efdb0237 | 1132 | |
be2c0327 LP |
1133 | if (max != CGROUP_LIMIT_MAX) |
1134 | log_cgroup_compat(u, "Applying MemoryLimit=%" PRIu64 " as MemoryMax=", max); | |
128fadc9 | 1135 | } |
da4d897e | 1136 | |
be2c0327 LP |
1137 | cgroup_apply_unified_memory_limit(u, "memory.min", c->memory_min); |
1138 | cgroup_apply_unified_memory_limit(u, "memory.low", c->memory_low); | |
1139 | cgroup_apply_unified_memory_limit(u, "memory.high", c->memory_high); | |
1140 | cgroup_apply_unified_memory_limit(u, "memory.max", max); | |
1141 | cgroup_apply_unified_memory_limit(u, "memory.swap.max", swap_max); | |
128fadc9 | 1142 | |
be2c0327 LP |
1143 | } else { |
1144 | char buf[DECIMAL_STR_MAX(uint64_t) + 1]; | |
1145 | uint64_t val; | |
52fecf20 | 1146 | |
be2c0327 LP |
1147 | if (cgroup_context_has_unified_memory_config(c)) { |
1148 | val = c->memory_max; | |
1149 | log_cgroup_compat(u, "Applying MemoryMax=%" PRIi64 " as MemoryLimit=", val); | |
1150 | } else | |
1151 | val = c->memory_limit; | |
78a4ee59 | 1152 | |
be2c0327 LP |
1153 | if (val == CGROUP_LIMIT_MAX) |
1154 | strncpy(buf, "-1\n", sizeof(buf)); | |
1155 | else | |
1156 | xsprintf(buf, "%" PRIu64 "\n", val); | |
1157 | ||
1158 | (void) set_attribute_and_warn(u, "memory", "memory.limit_in_bytes", buf); | |
da4d897e | 1159 | } |
4ad49000 | 1160 | } |
8e274523 | 1161 | |
4e1dfa45 | 1162 | /* On cgroup v2 we can apply BPF everywhere. On cgroup v1 we apply it everywhere except for the root of |
52fecf20 LP |
1163 | * containers, where we leave this to the manager */ |
1164 | if ((apply_mask & (CGROUP_MASK_DEVICES | CGROUP_MASK_BPF_DEVICES)) && | |
1165 | (is_host_root || cg_all_unified() > 0 || !is_local_root)) { | |
084c7007 | 1166 | _cleanup_(bpf_program_unrefp) BPFProgram *prog = NULL; |
4ad49000 | 1167 | CGroupDeviceAllow *a; |
8e274523 | 1168 | |
084c7007 RG |
1169 | if (cg_all_unified() > 0) { |
1170 | r = cgroup_init_device_bpf(&prog, c->device_policy, c->device_allow); | |
1171 | if (r < 0) | |
1172 | log_unit_warning_errno(u, r, "Failed to initialize device control bpf program: %m"); | |
1173 | } else { | |
8c838407 | 1174 | /* Changing the devices list of a populated cgroup might result in EINVAL, hence ignore EINVAL |
084c7007 | 1175 | * here. */ |
714e2e1d | 1176 | |
084c7007 RG |
1177 | if (c->device_allow || c->device_policy != CGROUP_AUTO) |
1178 | r = cg_set_attribute("devices", path, "devices.deny", "a"); | |
1179 | else | |
1180 | r = cg_set_attribute("devices", path, "devices.allow", "a"); | |
1181 | if (r < 0) | |
2c74e12b LP |
1182 | log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EINVAL, -EACCES, -EPERM) ? LOG_DEBUG : LOG_WARNING, r, |
1183 | "Failed to reset devices.allow/devices.deny: %m"); | |
084c7007 | 1184 | } |
fb385181 | 1185 | |
4ad49000 LP |
1186 | if (c->device_policy == CGROUP_CLOSED || |
1187 | (c->device_policy == CGROUP_AUTO && c->device_allow)) { | |
1188 | static const char auto_devices[] = | |
7d711efb LP |
1189 | "/dev/null\0" "rwm\0" |
1190 | "/dev/zero\0" "rwm\0" | |
1191 | "/dev/full\0" "rwm\0" | |
1192 | "/dev/random\0" "rwm\0" | |
1193 | "/dev/urandom\0" "rwm\0" | |
1194 | "/dev/tty\0" "rwm\0" | |
5a7f87a9 | 1195 | "/dev/ptmx\0" "rwm\0" |
0d9e7991 | 1196 | /* Allow /run/systemd/inaccessible/{chr,blk} devices for mapping InaccessiblePaths */ |
74c48bf5 LP |
1197 | "/run/systemd/inaccessible/chr\0" "rwm\0" |
1198 | "/run/systemd/inaccessible/blk\0" "rwm\0"; | |
4ad49000 LP |
1199 | |
1200 | const char *x, *y; | |
1201 | ||
1202 | NULSTR_FOREACH_PAIR(x, y, auto_devices) | |
913c898c | 1203 | (void) whitelist_device(prog, path, x, y); |
7d711efb | 1204 | |
5a7f87a9 | 1205 | /* PTS (/dev/pts) devices may not be duplicated, but accessed */ |
913c898c | 1206 | (void) whitelist_major(prog, path, "pts", 'c', "rw"); |
4ad49000 LP |
1207 | } |
1208 | ||
1209 | LIST_FOREACH(device_allow, a, c->device_allow) { | |
fb4650aa | 1210 | char acc[4], *val; |
4ad49000 LP |
1211 | unsigned k = 0; |
1212 | ||
1213 | if (a->r) | |
1214 | acc[k++] = 'r'; | |
1215 | if (a->w) | |
1216 | acc[k++] = 'w'; | |
1217 | if (a->m) | |
1218 | acc[k++] = 'm'; | |
fb385181 | 1219 | |
4ad49000 LP |
1220 | if (k == 0) |
1221 | continue; | |
fb385181 | 1222 | |
4ad49000 | 1223 | acc[k++] = 0; |
90060676 | 1224 | |
27458ed6 | 1225 | if (path_startswith(a->path, "/dev/")) |
913c898c | 1226 | (void) whitelist_device(prog, path, a->path, acc); |
fb4650aa | 1227 | else if ((val = startswith(a->path, "block-"))) |
913c898c | 1228 | (void) whitelist_major(prog, path, val, 'b', acc); |
fb4650aa | 1229 | else if ((val = startswith(a->path, "char-"))) |
913c898c | 1230 | (void) whitelist_major(prog, path, val, 'c', acc); |
90060676 | 1231 | else |
8e8b5d2e | 1232 | log_unit_debug(u, "Ignoring device '%s' while writing cgroup attribute.", a->path); |
4ad49000 | 1233 | } |
084c7007 RG |
1234 | |
1235 | r = cgroup_apply_device_bpf(u, prog, c->device_policy, c->device_allow); | |
1236 | if (r < 0) { | |
1237 | static bool warned = false; | |
1238 | ||
1239 | log_full_errno(warned ? LOG_DEBUG : LOG_WARNING, r, | |
1240 | "Unit %s configures device ACL, but the local system doesn't seem to support the BPF-based device controller.\n" | |
1241 | "Proceeding WITHOUT applying ACL (all devices will be accessible)!\n" | |
1242 | "(This warning is only shown for the first loaded unit using device ACL.)", u->id); | |
1243 | ||
1244 | warned = true; | |
1245 | } | |
4ad49000 | 1246 | } |
03a7b521 | 1247 | |
00b5974f LP |
1248 | if (apply_mask & CGROUP_MASK_PIDS) { |
1249 | ||
52fecf20 | 1250 | if (is_host_root) { |
00b5974f LP |
1251 | /* So, the "pids" controller does not expose anything on the root cgroup, in order not to |
1252 | * replicate knobs exposed elsewhere needlessly. We abstract this away here however, and when | |
1253 | * the knobs of the root cgroup are modified propagate this to the relevant sysctls. There's a | |
1254 | * non-obvious asymmetry however: unlike the cgroup properties we don't really want to take | |
1255 | * exclusive ownership of the sysctls, but we still want to honour things if the user sets | |
1256 | * limits. Hence we employ sort of a one-way strategy: when the user sets a bounded limit | |
1257 | * through us it counts. When the user afterwards unsets it again (i.e. sets it to unbounded) | |
1258 | * it also counts. But if the user never set a limit through us (i.e. we are the default of | |
1259 | * "unbounded") we leave things unmodified. For this we manage a global boolean that we turn on | |
1260 | * the first time we set a limit. Note that this boolean is flushed out on manager reload, | |
1261 | * which is desirable so that there's an offical way to release control of the sysctl from | |
1262 | * systemd: set the limit to unbounded and reload. */ | |
1263 | ||
1264 | if (c->tasks_max != CGROUP_LIMIT_MAX) { | |
1265 | u->manager->sysctl_pid_max_changed = true; | |
1266 | r = procfs_tasks_set_limit(c->tasks_max); | |
1267 | } else if (u->manager->sysctl_pid_max_changed) | |
1268 | r = procfs_tasks_set_limit(TASKS_MAX); | |
1269 | else | |
1270 | r = 0; | |
00b5974f | 1271 | if (r < 0) |
39b9fefb | 1272 | log_unit_full(u, LOG_LEVEL_CGROUP_WRITE(r), r, |
00b5974f | 1273 | "Failed to write to tasks limit sysctls: %m"); |
52fecf20 | 1274 | } |
03a7b521 | 1275 | |
52fecf20 LP |
1276 | /* The attribute itself is not available on the host root cgroup, and in the container case we want to |
1277 | * leave it for the container manager. */ | |
1278 | if (!is_local_root) { | |
00b5974f LP |
1279 | if (c->tasks_max != CGROUP_LIMIT_MAX) { |
1280 | char buf[DECIMAL_STR_MAX(uint64_t) + 2]; | |
03a7b521 | 1281 | |
00b5974f | 1282 | sprintf(buf, "%" PRIu64 "\n", c->tasks_max); |
293d32df | 1283 | (void) set_attribute_and_warn(u, "pids", "pids.max", buf); |
00b5974f | 1284 | } else |
589a5f7a | 1285 | (void) set_attribute_and_warn(u, "pids", "pids.max", "max\n"); |
00b5974f | 1286 | } |
03a7b521 | 1287 | } |
906c06f6 | 1288 | |
17f14955 | 1289 | if (apply_mask & CGROUP_MASK_BPF_FIREWALL) |
0f2d84d2 | 1290 | cgroup_apply_firewall(u); |
fb385181 LP |
1291 | } |
1292 | ||
16492445 LP |
1293 | static bool unit_get_needs_bpf_firewall(Unit *u) { |
1294 | CGroupContext *c; | |
1295 | Unit *p; | |
1296 | assert(u); | |
1297 | ||
1298 | c = unit_get_cgroup_context(u); | |
1299 | if (!c) | |
1300 | return false; | |
1301 | ||
1302 | if (c->ip_accounting || | |
1303 | c->ip_address_allow || | |
1304 | c->ip_address_deny) | |
1305 | return true; | |
1306 | ||
1307 | /* If any parent slice has an IP access list defined, it applies too */ | |
1308 | for (p = UNIT_DEREF(u->slice); p; p = UNIT_DEREF(p->slice)) { | |
1309 | c = unit_get_cgroup_context(p); | |
1310 | if (!c) | |
1311 | return false; | |
1312 | ||
1313 | if (c->ip_address_allow || | |
1314 | c->ip_address_deny) | |
1315 | return true; | |
1316 | } | |
1317 | ||
1318 | return false; | |
1319 | } | |
1320 | ||
53aea74a | 1321 | static CGroupMask cgroup_context_get_mask(CGroupContext *c) { |
efdb0237 | 1322 | CGroupMask mask = 0; |
8e274523 | 1323 | |
fae9bc29 | 1324 | /* Figure out which controllers we need, based on the cgroup context object */ |
8e274523 | 1325 | |
fae9bc29 | 1326 | if (c->cpu_accounting) |
f98c2585 | 1327 | mask |= get_cpu_accounting_mask(); |
fae9bc29 LP |
1328 | |
1329 | if (cgroup_context_has_cpu_weight(c) || | |
66ebf6c0 | 1330 | cgroup_context_has_cpu_shares(c) || |
3a43da28 | 1331 | c->cpu_quota_per_sec_usec != USEC_INFINITY) |
fae9bc29 | 1332 | mask |= CGROUP_MASK_CPU; |
ecedd90f | 1333 | |
538b4852 TH |
1334 | if (cgroup_context_has_io_config(c) || cgroup_context_has_blockio_config(c)) |
1335 | mask |= CGROUP_MASK_IO | CGROUP_MASK_BLKIO; | |
ecedd90f | 1336 | |
4ad49000 | 1337 | if (c->memory_accounting || |
da4d897e TH |
1338 | c->memory_limit != CGROUP_LIMIT_MAX || |
1339 | cgroup_context_has_unified_memory_config(c)) | |
efdb0237 | 1340 | mask |= CGROUP_MASK_MEMORY; |
8e274523 | 1341 | |
a931ad47 LP |
1342 | if (c->device_allow || |
1343 | c->device_policy != CGROUP_AUTO) | |
084c7007 | 1344 | mask |= CGROUP_MASK_DEVICES | CGROUP_MASK_BPF_DEVICES; |
4ad49000 | 1345 | |
03a7b521 | 1346 | if (c->tasks_accounting || |
8793fa25 | 1347 | c->tasks_max != CGROUP_LIMIT_MAX) |
03a7b521 LP |
1348 | mask |= CGROUP_MASK_PIDS; |
1349 | ||
fae9bc29 | 1350 | return CGROUP_MASK_EXTEND_JOINED(mask); |
8e274523 LP |
1351 | } |
1352 | ||
53aea74a | 1353 | static CGroupMask unit_get_bpf_mask(Unit *u) { |
17f14955 RG |
1354 | CGroupMask mask = 0; |
1355 | ||
fae9bc29 LP |
1356 | /* Figure out which controllers we need, based on the cgroup context, possibly taking into account children |
1357 | * too. */ | |
1358 | ||
17f14955 RG |
1359 | if (unit_get_needs_bpf_firewall(u)) |
1360 | mask |= CGROUP_MASK_BPF_FIREWALL; | |
1361 | ||
1362 | return mask; | |
1363 | } | |
1364 | ||
efdb0237 | 1365 | CGroupMask unit_get_own_mask(Unit *u) { |
4ad49000 | 1366 | CGroupContext *c; |
8e274523 | 1367 | |
442ce775 LP |
1368 | /* Returns the mask of controllers the unit needs for itself. If a unit is not properly loaded, return an empty |
1369 | * mask, as we shouldn't reflect it in the cgroup hierarchy then. */ | |
1370 | ||
1371 | if (u->load_state != UNIT_LOADED) | |
1372 | return 0; | |
efdb0237 | 1373 | |
4ad49000 LP |
1374 | c = unit_get_cgroup_context(u); |
1375 | if (!c) | |
1376 | return 0; | |
8e274523 | 1377 | |
c72703e2 | 1378 | return (cgroup_context_get_mask(c) | unit_get_bpf_mask(u) | unit_get_delegate_mask(u)) & ~unit_get_ancestor_disable_mask(u); |
02638280 LP |
1379 | } |
1380 | ||
1381 | CGroupMask unit_get_delegate_mask(Unit *u) { | |
1382 | CGroupContext *c; | |
1383 | ||
1384 | /* If delegation is turned on, then turn on selected controllers, unless we are on the legacy hierarchy and the | |
1385 | * process we fork into is known to drop privileges, and hence shouldn't get access to the controllers. | |
19af675e | 1386 | * |
02638280 | 1387 | * Note that on the unified hierarchy it is safe to delegate controllers to unprivileged services. */ |
a931ad47 | 1388 | |
1d9cc876 | 1389 | if (!unit_cgroup_delegate(u)) |
02638280 LP |
1390 | return 0; |
1391 | ||
1392 | if (cg_all_unified() <= 0) { | |
a931ad47 LP |
1393 | ExecContext *e; |
1394 | ||
1395 | e = unit_get_exec_context(u); | |
02638280 LP |
1396 | if (e && !exec_context_maintains_privileges(e)) |
1397 | return 0; | |
a931ad47 LP |
1398 | } |
1399 | ||
1d9cc876 | 1400 | assert_se(c = unit_get_cgroup_context(u)); |
fae9bc29 | 1401 | return CGROUP_MASK_EXTEND_JOINED(c->delegate_controllers); |
8e274523 LP |
1402 | } |
1403 | ||
efdb0237 | 1404 | CGroupMask unit_get_members_mask(Unit *u) { |
4ad49000 | 1405 | assert(u); |
bc432dc7 | 1406 | |
02638280 | 1407 | /* Returns the mask of controllers all of the unit's children require, merged */ |
efdb0237 | 1408 | |
bc432dc7 | 1409 | if (u->cgroup_members_mask_valid) |
26a17ca2 | 1410 | return u->cgroup_members_mask; /* Use cached value if possible */ |
bc432dc7 | 1411 | |
64e844e5 | 1412 | u->cgroup_members_mask = 0; |
bc432dc7 LP |
1413 | |
1414 | if (u->type == UNIT_SLICE) { | |
eef85c4a | 1415 | void *v; |
bc432dc7 LP |
1416 | Unit *member; |
1417 | Iterator i; | |
1418 | ||
eef85c4a | 1419 | HASHMAP_FOREACH_KEY(v, member, u->dependencies[UNIT_BEFORE], i) { |
cb5e3bc3 CD |
1420 | if (UNIT_DEREF(member->slice) == u) |
1421 | u->cgroup_members_mask |= unit_get_subtree_mask(member); /* note that this calls ourselves again, for the children */ | |
bc432dc7 LP |
1422 | } |
1423 | } | |
1424 | ||
1425 | u->cgroup_members_mask_valid = true; | |
6414b7c9 | 1426 | return u->cgroup_members_mask; |
246aa6dd LP |
1427 | } |
1428 | ||
efdb0237 | 1429 | CGroupMask unit_get_siblings_mask(Unit *u) { |
4ad49000 | 1430 | assert(u); |
246aa6dd | 1431 | |
efdb0237 LP |
1432 | /* Returns the mask of controllers all of the unit's siblings |
1433 | * require, i.e. the members mask of the unit's parent slice | |
1434 | * if there is one. */ | |
1435 | ||
bc432dc7 | 1436 | if (UNIT_ISSET(u->slice)) |
637f421e | 1437 | return unit_get_members_mask(UNIT_DEREF(u->slice)); |
4ad49000 | 1438 | |
64e844e5 | 1439 | return unit_get_subtree_mask(u); /* we are the top-level slice */ |
246aa6dd LP |
1440 | } |
1441 | ||
4f6f62e4 CD |
1442 | CGroupMask unit_get_disable_mask(Unit *u) { |
1443 | CGroupContext *c; | |
1444 | ||
1445 | c = unit_get_cgroup_context(u); | |
1446 | if (!c) | |
1447 | return 0; | |
1448 | ||
1449 | return c->disable_controllers; | |
1450 | } | |
1451 | ||
1452 | CGroupMask unit_get_ancestor_disable_mask(Unit *u) { | |
1453 | CGroupMask mask; | |
1454 | ||
1455 | assert(u); | |
1456 | mask = unit_get_disable_mask(u); | |
1457 | ||
1458 | /* Returns the mask of controllers which are marked as forcibly | |
1459 | * disabled in any ancestor unit or the unit in question. */ | |
1460 | ||
1461 | if (UNIT_ISSET(u->slice)) | |
1462 | mask |= unit_get_ancestor_disable_mask(UNIT_DEREF(u->slice)); | |
1463 | ||
1464 | return mask; | |
1465 | } | |
1466 | ||
efdb0237 LP |
1467 | CGroupMask unit_get_subtree_mask(Unit *u) { |
1468 | ||
1469 | /* Returns the mask of this subtree, meaning of the group | |
1470 | * itself and its children. */ | |
1471 | ||
1472 | return unit_get_own_mask(u) | unit_get_members_mask(u); | |
1473 | } | |
1474 | ||
1475 | CGroupMask unit_get_target_mask(Unit *u) { | |
1476 | CGroupMask mask; | |
1477 | ||
1478 | /* This returns the cgroup mask of all controllers to enable | |
1479 | * for a specific cgroup, i.e. everything it needs itself, | |
1480 | * plus all that its children need, plus all that its siblings | |
1481 | * need. This is primarily useful on the legacy cgroup | |
1482 | * hierarchy, where we need to duplicate each cgroup in each | |
1483 | * hierarchy that shall be enabled for it. */ | |
6414b7c9 | 1484 | |
efdb0237 LP |
1485 | mask = unit_get_own_mask(u) | unit_get_members_mask(u) | unit_get_siblings_mask(u); |
1486 | mask &= u->manager->cgroup_supported; | |
c72703e2 | 1487 | mask &= ~unit_get_ancestor_disable_mask(u); |
efdb0237 LP |
1488 | |
1489 | return mask; | |
1490 | } | |
1491 | ||
1492 | CGroupMask unit_get_enable_mask(Unit *u) { | |
1493 | CGroupMask mask; | |
1494 | ||
1495 | /* This returns the cgroup mask of all controllers to enable | |
1496 | * for the children of a specific cgroup. This is primarily | |
1497 | * useful for the unified cgroup hierarchy, where each cgroup | |
1498 | * controls which controllers are enabled for its children. */ | |
1499 | ||
1500 | mask = unit_get_members_mask(u); | |
6414b7c9 | 1501 | mask &= u->manager->cgroup_supported; |
c72703e2 | 1502 | mask &= ~unit_get_ancestor_disable_mask(u); |
6414b7c9 DS |
1503 | |
1504 | return mask; | |
1505 | } | |
1506 | ||
5af88058 | 1507 | void unit_invalidate_cgroup_members_masks(Unit *u) { |
bc432dc7 LP |
1508 | assert(u); |
1509 | ||
5af88058 LP |
1510 | /* Recurse invalidate the member masks cache all the way up the tree */ |
1511 | u->cgroup_members_mask_valid = false; | |
bc432dc7 | 1512 | |
5af88058 LP |
1513 | if (UNIT_ISSET(u->slice)) |
1514 | unit_invalidate_cgroup_members_masks(UNIT_DEREF(u->slice)); | |
6414b7c9 DS |
1515 | } |
1516 | ||
6592b975 | 1517 | const char *unit_get_realized_cgroup_path(Unit *u, CGroupMask mask) { |
03b90d4b | 1518 | |
6592b975 | 1519 | /* Returns the realized cgroup path of the specified unit where all specified controllers are available. */ |
03b90d4b LP |
1520 | |
1521 | while (u) { | |
6592b975 | 1522 | |
03b90d4b LP |
1523 | if (u->cgroup_path && |
1524 | u->cgroup_realized && | |
d94a24ca | 1525 | FLAGS_SET(u->cgroup_realized_mask, mask)) |
03b90d4b LP |
1526 | return u->cgroup_path; |
1527 | ||
1528 | u = UNIT_DEREF(u->slice); | |
1529 | } | |
1530 | ||
1531 | return NULL; | |
1532 | } | |
1533 | ||
6592b975 LP |
1534 | static const char *migrate_callback(CGroupMask mask, void *userdata) { |
1535 | return unit_get_realized_cgroup_path(userdata, mask); | |
1536 | } | |
1537 | ||
303ee601 | 1538 | char *unit_default_cgroup_path(const Unit *u) { |
efdb0237 LP |
1539 | _cleanup_free_ char *escaped = NULL, *slice = NULL; |
1540 | int r; | |
1541 | ||
1542 | assert(u); | |
1543 | ||
1544 | if (unit_has_name(u, SPECIAL_ROOT_SLICE)) | |
1545 | return strdup(u->manager->cgroup_root); | |
1546 | ||
1547 | if (UNIT_ISSET(u->slice) && !unit_has_name(UNIT_DEREF(u->slice), SPECIAL_ROOT_SLICE)) { | |
1548 | r = cg_slice_to_path(UNIT_DEREF(u->slice)->id, &slice); | |
1549 | if (r < 0) | |
1550 | return NULL; | |
1551 | } | |
1552 | ||
1553 | escaped = cg_escape(u->id); | |
1554 | if (!escaped) | |
1555 | return NULL; | |
1556 | ||
1557 | if (slice) | |
605405c6 ZJS |
1558 | return strjoin(u->manager->cgroup_root, "/", slice, "/", |
1559 | escaped); | |
efdb0237 | 1560 | else |
605405c6 | 1561 | return strjoin(u->manager->cgroup_root, "/", escaped); |
efdb0237 LP |
1562 | } |
1563 | ||
1564 | int unit_set_cgroup_path(Unit *u, const char *path) { | |
1565 | _cleanup_free_ char *p = NULL; | |
1566 | int r; | |
1567 | ||
1568 | assert(u); | |
1569 | ||
1570 | if (path) { | |
1571 | p = strdup(path); | |
1572 | if (!p) | |
1573 | return -ENOMEM; | |
1574 | } else | |
1575 | p = NULL; | |
1576 | ||
1577 | if (streq_ptr(u->cgroup_path, p)) | |
1578 | return 0; | |
1579 | ||
1580 | if (p) { | |
1581 | r = hashmap_put(u->manager->cgroup_unit, p, u); | |
1582 | if (r < 0) | |
1583 | return r; | |
1584 | } | |
1585 | ||
1586 | unit_release_cgroup(u); | |
1587 | ||
ae2a15bc | 1588 | u->cgroup_path = TAKE_PTR(p); |
efdb0237 LP |
1589 | |
1590 | return 1; | |
1591 | } | |
1592 | ||
1593 | int unit_watch_cgroup(Unit *u) { | |
ab2c3861 | 1594 | _cleanup_free_ char *events = NULL; |
efdb0237 LP |
1595 | int r; |
1596 | ||
1597 | assert(u); | |
1598 | ||
1599 | if (!u->cgroup_path) | |
1600 | return 0; | |
1601 | ||
1602 | if (u->cgroup_inotify_wd >= 0) | |
1603 | return 0; | |
1604 | ||
1605 | /* Only applies to the unified hierarchy */ | |
c22800e4 | 1606 | r = cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER); |
b4cccbc1 LP |
1607 | if (r < 0) |
1608 | return log_error_errno(r, "Failed to determine whether the name=systemd hierarchy is unified: %m"); | |
1609 | if (r == 0) | |
efdb0237 LP |
1610 | return 0; |
1611 | ||
1612 | /* Don't watch the root slice, it's pointless. */ | |
1613 | if (unit_has_name(u, SPECIAL_ROOT_SLICE)) | |
1614 | return 0; | |
1615 | ||
1616 | r = hashmap_ensure_allocated(&u->manager->cgroup_inotify_wd_unit, &trivial_hash_ops); | |
1617 | if (r < 0) | |
1618 | return log_oom(); | |
1619 | ||
ab2c3861 | 1620 | r = cg_get_path(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path, "cgroup.events", &events); |
efdb0237 LP |
1621 | if (r < 0) |
1622 | return log_oom(); | |
1623 | ||
ab2c3861 | 1624 | u->cgroup_inotify_wd = inotify_add_watch(u->manager->cgroup_inotify_fd, events, IN_MODIFY); |
efdb0237 LP |
1625 | if (u->cgroup_inotify_wd < 0) { |
1626 | ||
1627 | if (errno == ENOENT) /* If the directory is already | |
1628 | * gone we don't need to track | |
1629 | * it, so this is not an error */ | |
1630 | return 0; | |
1631 | ||
1632 | return log_unit_error_errno(u, errno, "Failed to add inotify watch descriptor for control group %s: %m", u->cgroup_path); | |
1633 | } | |
1634 | ||
1635 | r = hashmap_put(u->manager->cgroup_inotify_wd_unit, INT_TO_PTR(u->cgroup_inotify_wd), u); | |
1636 | if (r < 0) | |
1637 | return log_unit_error_errno(u, r, "Failed to add inotify watch descriptor to hash map: %m"); | |
1638 | ||
1639 | return 0; | |
1640 | } | |
1641 | ||
a4634b21 LP |
1642 | int unit_pick_cgroup_path(Unit *u) { |
1643 | _cleanup_free_ char *path = NULL; | |
1644 | int r; | |
1645 | ||
1646 | assert(u); | |
1647 | ||
1648 | if (u->cgroup_path) | |
1649 | return 0; | |
1650 | ||
1651 | if (!UNIT_HAS_CGROUP_CONTEXT(u)) | |
1652 | return -EINVAL; | |
1653 | ||
1654 | path = unit_default_cgroup_path(u); | |
1655 | if (!path) | |
1656 | return log_oom(); | |
1657 | ||
1658 | r = unit_set_cgroup_path(u, path); | |
1659 | if (r == -EEXIST) | |
1660 | return log_unit_error_errno(u, r, "Control group %s exists already.", path); | |
1661 | if (r < 0) | |
1662 | return log_unit_error_errno(u, r, "Failed to set unit's control group path to %s: %m", path); | |
1663 | ||
1664 | return 0; | |
1665 | } | |
1666 | ||
efdb0237 LP |
1667 | static int unit_create_cgroup( |
1668 | Unit *u, | |
1669 | CGroupMask target_mask, | |
0d2d6fbf CD |
1670 | CGroupMask enable_mask, |
1671 | ManagerState state) { | |
efdb0237 | 1672 | |
65be7e06 | 1673 | bool created; |
27adcc97 | 1674 | int r; |
64747e2d | 1675 | |
4ad49000 | 1676 | assert(u); |
64747e2d | 1677 | |
27c4ed79 | 1678 | if (!UNIT_HAS_CGROUP_CONTEXT(u)) |
0cd385d3 LP |
1679 | return 0; |
1680 | ||
a4634b21 LP |
1681 | /* Figure out our cgroup path */ |
1682 | r = unit_pick_cgroup_path(u); | |
1683 | if (r < 0) | |
1684 | return r; | |
b58b8e11 | 1685 | |
03b90d4b | 1686 | /* First, create our own group */ |
efdb0237 | 1687 | r = cg_create_everywhere(u->manager->cgroup_supported, target_mask, u->cgroup_path); |
23bbb0de | 1688 | if (r < 0) |
efdb0237 | 1689 | return log_unit_error_errno(u, r, "Failed to create cgroup %s: %m", u->cgroup_path); |
490c5a37 | 1690 | created = r; |
efdb0237 LP |
1691 | |
1692 | /* Start watching it */ | |
1693 | (void) unit_watch_cgroup(u); | |
1694 | ||
65be7e06 | 1695 | /* Preserve enabled controllers in delegated units, adjust others. */ |
1fd3a10c | 1696 | if (created || !u->cgroup_realized || !unit_cgroup_delegate(u)) { |
27adcc97 | 1697 | CGroupMask result_mask = 0; |
65be7e06 ZJS |
1698 | |
1699 | /* Enable all controllers we need */ | |
27adcc97 | 1700 | r = cg_enable_everywhere(u->manager->cgroup_supported, enable_mask, u->cgroup_path, &result_mask); |
65be7e06 | 1701 | if (r < 0) |
27adcc97 LP |
1702 | log_unit_warning_errno(u, r, "Failed to enable/disable controllers on cgroup %s, ignoring: %m", u->cgroup_path); |
1703 | ||
1704 | /* If we just turned off a controller, this might release the controller for our parent too, let's | |
1705 | * enqueue the parent for re-realization in that case again. */ | |
1706 | if (UNIT_ISSET(u->slice)) { | |
1707 | CGroupMask turned_off; | |
1708 | ||
1709 | turned_off = (u->cgroup_realized ? u->cgroup_enabled_mask & ~result_mask : 0); | |
1710 | if (turned_off != 0) { | |
1711 | Unit *parent; | |
1712 | ||
1713 | /* Force the parent to propagate the enable mask to the kernel again, by invalidating | |
1714 | * the controller we just turned off. */ | |
1715 | ||
1716 | for (parent = UNIT_DEREF(u->slice); parent; parent = UNIT_DEREF(parent->slice)) | |
1717 | unit_invalidate_cgroup(parent, turned_off); | |
1718 | } | |
1719 | } | |
1720 | ||
1721 | /* Remember what's actually enabled now */ | |
1722 | u->cgroup_enabled_mask = result_mask; | |
65be7e06 | 1723 | } |
03b90d4b LP |
1724 | |
1725 | /* Keep track that this is now realized */ | |
4ad49000 | 1726 | u->cgroup_realized = true; |
efdb0237 | 1727 | u->cgroup_realized_mask = target_mask; |
4ad49000 | 1728 | |
1d9cc876 | 1729 | if (u->type != UNIT_SLICE && !unit_cgroup_delegate(u)) { |
0cd385d3 LP |
1730 | |
1731 | /* Then, possibly move things over, but not if | |
1732 | * subgroups may contain processes, which is the case | |
1733 | * for slice and delegation units. */ | |
1734 | r = cg_migrate_everywhere(u->manager->cgroup_supported, u->cgroup_path, u->cgroup_path, migrate_callback, u); | |
1735 | if (r < 0) | |
efdb0237 | 1736 | log_unit_warning_errno(u, r, "Failed to migrate cgroup from to %s, ignoring: %m", u->cgroup_path); |
0cd385d3 | 1737 | } |
03b90d4b | 1738 | |
0d2d6fbf CD |
1739 | /* Set attributes */ |
1740 | cgroup_context_apply(u, target_mask, state); | |
1741 | cgroup_xattr_apply(u); | |
1742 | ||
64747e2d LP |
1743 | return 0; |
1744 | } | |
1745 | ||
6592b975 LP |
1746 | static int unit_attach_pid_to_cgroup_via_bus(Unit *u, pid_t pid, const char *suffix_path) { |
1747 | _cleanup_(sd_bus_error_free) sd_bus_error error = SD_BUS_ERROR_NULL; | |
1748 | char *pp; | |
7b3fd631 | 1749 | int r; |
6592b975 | 1750 | |
7b3fd631 LP |
1751 | assert(u); |
1752 | ||
6592b975 LP |
1753 | if (MANAGER_IS_SYSTEM(u->manager)) |
1754 | return -EINVAL; | |
1755 | ||
1756 | if (!u->manager->system_bus) | |
1757 | return -EIO; | |
1758 | ||
1759 | if (!u->cgroup_path) | |
1760 | return -EINVAL; | |
1761 | ||
1762 | /* Determine this unit's cgroup path relative to our cgroup root */ | |
1763 | pp = path_startswith(u->cgroup_path, u->manager->cgroup_root); | |
1764 | if (!pp) | |
1765 | return -EINVAL; | |
1766 | ||
1767 | pp = strjoina("/", pp, suffix_path); | |
858d36c1 | 1768 | path_simplify(pp, false); |
6592b975 LP |
1769 | |
1770 | r = sd_bus_call_method(u->manager->system_bus, | |
1771 | "org.freedesktop.systemd1", | |
1772 | "/org/freedesktop/systemd1", | |
1773 | "org.freedesktop.systemd1.Manager", | |
1774 | "AttachProcessesToUnit", | |
1775 | &error, NULL, | |
1776 | "ssau", | |
1777 | NULL /* empty unit name means client's unit, i.e. us */, pp, 1, (uint32_t) pid); | |
7b3fd631 | 1778 | if (r < 0) |
6592b975 LP |
1779 | return log_unit_debug_errno(u, r, "Failed to attach unit process " PID_FMT " via the bus: %s", pid, bus_error_message(&error, r)); |
1780 | ||
1781 | return 0; | |
1782 | } | |
1783 | ||
1784 | int unit_attach_pids_to_cgroup(Unit *u, Set *pids, const char *suffix_path) { | |
1785 | CGroupMask delegated_mask; | |
1786 | const char *p; | |
1787 | Iterator i; | |
1788 | void *pidp; | |
1789 | int r, q; | |
1790 | ||
1791 | assert(u); | |
1792 | ||
1793 | if (!UNIT_HAS_CGROUP_CONTEXT(u)) | |
1794 | return -EINVAL; | |
1795 | ||
1796 | if (set_isempty(pids)) | |
1797 | return 0; | |
7b3fd631 | 1798 | |
6592b975 | 1799 | r = unit_realize_cgroup(u); |
7b3fd631 LP |
1800 | if (r < 0) |
1801 | return r; | |
1802 | ||
6592b975 LP |
1803 | if (isempty(suffix_path)) |
1804 | p = u->cgroup_path; | |
1805 | else | |
1806 | p = strjoina(u->cgroup_path, "/", suffix_path); | |
1807 | ||
1808 | delegated_mask = unit_get_delegate_mask(u); | |
1809 | ||
1810 | r = 0; | |
1811 | SET_FOREACH(pidp, pids, i) { | |
1812 | pid_t pid = PTR_TO_PID(pidp); | |
1813 | CGroupController c; | |
1814 | ||
1815 | /* First, attach the PID to the main cgroup hierarchy */ | |
1816 | q = cg_attach(SYSTEMD_CGROUP_CONTROLLER, p, pid); | |
1817 | if (q < 0) { | |
1818 | log_unit_debug_errno(u, q, "Couldn't move process " PID_FMT " to requested cgroup '%s': %m", pid, p); | |
1819 | ||
1820 | if (MANAGER_IS_USER(u->manager) && IN_SET(q, -EPERM, -EACCES)) { | |
1821 | int z; | |
1822 | ||
1823 | /* If we are in a user instance, and we can't move the process ourselves due to | |
1824 | * permission problems, let's ask the system instance about it instead. Since it's more | |
1825 | * privileged it might be able to move the process across the leaves of a subtree who's | |
1826 | * top node is not owned by us. */ | |
1827 | ||
1828 | z = unit_attach_pid_to_cgroup_via_bus(u, pid, suffix_path); | |
1829 | if (z < 0) | |
1830 | log_unit_debug_errno(u, z, "Couldn't move process " PID_FMT " to requested cgroup '%s' via the system bus either: %m", pid, p); | |
1831 | else | |
1832 | continue; /* When the bus thing worked via the bus we are fully done for this PID. */ | |
1833 | } | |
1834 | ||
1835 | if (r >= 0) | |
1836 | r = q; /* Remember first error */ | |
1837 | ||
1838 | continue; | |
1839 | } | |
1840 | ||
1841 | q = cg_all_unified(); | |
1842 | if (q < 0) | |
1843 | return q; | |
1844 | if (q > 0) | |
1845 | continue; | |
1846 | ||
1847 | /* In the legacy hierarchy, attach the process to the request cgroup if possible, and if not to the | |
1848 | * innermost realized one */ | |
1849 | ||
1850 | for (c = 0; c < _CGROUP_CONTROLLER_MAX; c++) { | |
1851 | CGroupMask bit = CGROUP_CONTROLLER_TO_MASK(c); | |
1852 | const char *realized; | |
1853 | ||
1854 | if (!(u->manager->cgroup_supported & bit)) | |
1855 | continue; | |
1856 | ||
1857 | /* If this controller is delegated and realized, honour the caller's request for the cgroup suffix. */ | |
1858 | if (delegated_mask & u->cgroup_realized_mask & bit) { | |
1859 | q = cg_attach(cgroup_controller_to_string(c), p, pid); | |
1860 | if (q >= 0) | |
1861 | continue; /* Success! */ | |
1862 | ||
1863 | 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", | |
1864 | pid, p, cgroup_controller_to_string(c)); | |
1865 | } | |
1866 | ||
1867 | /* So this controller is either not delegate or realized, or something else weird happened. In | |
1868 | * that case let's attach the PID at least to the closest cgroup up the tree that is | |
1869 | * realized. */ | |
1870 | realized = unit_get_realized_cgroup_path(u, bit); | |
1871 | if (!realized) | |
1872 | continue; /* Not even realized in the root slice? Then let's not bother */ | |
1873 | ||
1874 | q = cg_attach(cgroup_controller_to_string(c), realized, pid); | |
1875 | if (q < 0) | |
1876 | log_unit_debug_errno(u, q, "Failed to attach PID " PID_FMT " to realized cgroup %s in controller %s, ignoring: %m", | |
1877 | pid, realized, cgroup_controller_to_string(c)); | |
1878 | } | |
1879 | } | |
1880 | ||
1881 | return r; | |
7b3fd631 LP |
1882 | } |
1883 | ||
906c06f6 DM |
1884 | static bool unit_has_mask_realized( |
1885 | Unit *u, | |
1886 | CGroupMask target_mask, | |
17f14955 | 1887 | CGroupMask enable_mask) { |
906c06f6 | 1888 | |
bc432dc7 LP |
1889 | assert(u); |
1890 | ||
d5095dcd LP |
1891 | /* Returns true if this unit is fully realized. We check four things: |
1892 | * | |
1893 | * 1. Whether the cgroup was created at all | |
4e1dfa45 CD |
1894 | * 2. Whether the cgroup was created in all the hierarchies we need it to be created in (in case of cgroup v1) |
1895 | * 3. Whether the cgroup has all the right controllers enabled (in case of cgroup v2) | |
d5095dcd LP |
1896 | * 4. Whether the invalidation mask is currently zero |
1897 | * | |
1898 | * If you wonder why we mask the target realization and enable mask with CGROUP_MASK_V1/CGROUP_MASK_V2: note | |
4e1dfa45 CD |
1899 | * that there are three sets of bitmasks: CGROUP_MASK_V1 (for real cgroup v1 controllers), CGROUP_MASK_V2 (for |
1900 | * real cgroup v2 controllers) and CGROUP_MASK_BPF (for BPF-based pseudo-controllers). Now, cgroup_realized_mask | |
1901 | * is only matters for cgroup v1 controllers, and cgroup_enabled_mask only used for cgroup v2, and if they | |
d5095dcd LP |
1902 | * differ in the others, we don't really care. (After all, the cgroup_enabled_mask tracks with controllers are |
1903 | * enabled through cgroup.subtree_control, and since the BPF pseudo-controllers don't show up there, they | |
1904 | * simply don't matter. */ | |
1905 | ||
906c06f6 | 1906 | return u->cgroup_realized && |
d5095dcd LP |
1907 | ((u->cgroup_realized_mask ^ target_mask) & CGROUP_MASK_V1) == 0 && |
1908 | ((u->cgroup_enabled_mask ^ enable_mask) & CGROUP_MASK_V2) == 0 && | |
17f14955 | 1909 | u->cgroup_invalidated_mask == 0; |
6414b7c9 DS |
1910 | } |
1911 | ||
4f6f62e4 CD |
1912 | static bool unit_has_mask_disables_realized( |
1913 | Unit *u, | |
1914 | CGroupMask target_mask, | |
1915 | CGroupMask enable_mask) { | |
1916 | ||
1917 | assert(u); | |
1918 | ||
1919 | /* Returns true if all controllers which should be disabled are indeed disabled. | |
1920 | * | |
1921 | * Unlike unit_has_mask_realized, we don't care what was enabled, only that anything we want to remove is | |
1922 | * already removed. */ | |
1923 | ||
1924 | return !u->cgroup_realized || | |
1925 | (FLAGS_SET(u->cgroup_realized_mask, target_mask & CGROUP_MASK_V1) && | |
1926 | FLAGS_SET(u->cgroup_enabled_mask, enable_mask & CGROUP_MASK_V2)); | |
1927 | } | |
1928 | ||
a57669d2 CD |
1929 | static bool unit_has_mask_enables_realized( |
1930 | Unit *u, | |
1931 | CGroupMask target_mask, | |
1932 | CGroupMask enable_mask) { | |
1933 | ||
1934 | assert(u); | |
1935 | ||
1936 | /* Returns true if all controllers which should be enabled are indeed enabled. | |
1937 | * | |
1938 | * Unlike unit_has_mask_realized, we don't care about the controllers that are not present, only that anything | |
1939 | * we want to add is already added. */ | |
1940 | ||
1941 | return u->cgroup_realized && | |
c72703e2 CD |
1942 | ((u->cgroup_realized_mask | target_mask) & CGROUP_MASK_V1) == (u->cgroup_realized_mask & CGROUP_MASK_V1) && |
1943 | ((u->cgroup_enabled_mask | enable_mask) & CGROUP_MASK_V2) == (u->cgroup_enabled_mask & CGROUP_MASK_V2); | |
a57669d2 CD |
1944 | } |
1945 | ||
27adcc97 | 1946 | void unit_add_to_cgroup_realize_queue(Unit *u) { |
2aa57a65 LP |
1947 | assert(u); |
1948 | ||
1949 | if (u->in_cgroup_realize_queue) | |
1950 | return; | |
1951 | ||
1952 | LIST_PREPEND(cgroup_realize_queue, u->manager->cgroup_realize_queue, u); | |
1953 | u->in_cgroup_realize_queue = true; | |
1954 | } | |
1955 | ||
1956 | static void unit_remove_from_cgroup_realize_queue(Unit *u) { | |
1957 | assert(u); | |
1958 | ||
1959 | if (!u->in_cgroup_realize_queue) | |
1960 | return; | |
1961 | ||
1962 | LIST_REMOVE(cgroup_realize_queue, u->manager->cgroup_realize_queue, u); | |
1963 | u->in_cgroup_realize_queue = false; | |
1964 | } | |
1965 | ||
a57669d2 CD |
1966 | /* Controllers can only be enabled breadth-first, from the root of the |
1967 | * hierarchy downwards to the unit in question. */ | |
1968 | static int unit_realize_cgroup_now_enable(Unit *u, ManagerState state) { | |
1969 | CGroupMask target_mask, enable_mask, new_target_mask, new_enable_mask; | |
1970 | int r; | |
1971 | ||
1972 | assert(u); | |
1973 | ||
1974 | /* First go deal with this unit's parent, or we won't be able to enable | |
1975 | * any new controllers at this layer. */ | |
1976 | if (UNIT_ISSET(u->slice)) { | |
1977 | r = unit_realize_cgroup_now_enable(UNIT_DEREF(u->slice), state); | |
1978 | if (r < 0) | |
1979 | return r; | |
1980 | } | |
1981 | ||
1982 | target_mask = unit_get_target_mask(u); | |
1983 | enable_mask = unit_get_enable_mask(u); | |
1984 | ||
1985 | /* We can only enable in this direction, don't try to disable anything. | |
1986 | */ | |
1987 | if (unit_has_mask_enables_realized(u, target_mask, enable_mask)) | |
1988 | return 0; | |
1989 | ||
1990 | new_target_mask = u->cgroup_realized_mask | target_mask; | |
1991 | new_enable_mask = u->cgroup_enabled_mask | enable_mask; | |
1992 | ||
c72703e2 | 1993 | return unit_create_cgroup(u, new_target_mask, new_enable_mask, state); |
a57669d2 CD |
1994 | } |
1995 | ||
4f6f62e4 CD |
1996 | /* Controllers can only be disabled depth-first, from the leaves of the |
1997 | * hierarchy upwards to the unit in question. */ | |
1998 | static int unit_realize_cgroup_now_disable(Unit *u, ManagerState state) { | |
1999 | Iterator i; | |
2000 | Unit *m; | |
2001 | void *v; | |
2002 | ||
2003 | assert(u); | |
2004 | ||
2005 | if (u->type != UNIT_SLICE) | |
2006 | return 0; | |
2007 | ||
2008 | HASHMAP_FOREACH_KEY(v, m, u->dependencies[UNIT_BEFORE], i) { | |
2009 | CGroupMask target_mask, enable_mask, new_target_mask, new_enable_mask; | |
2010 | int r; | |
2011 | ||
2012 | if (UNIT_DEREF(m->slice) != u) | |
2013 | continue; | |
2014 | ||
2015 | /* The cgroup for this unit might not actually be fully | |
2016 | * realised yet, in which case it isn't holding any controllers | |
2017 | * open anyway. */ | |
2018 | if (!m->cgroup_path) | |
2019 | continue; | |
2020 | ||
2021 | /* We must disable those below us first in order to release the | |
2022 | * controller. */ | |
2023 | if (m->type == UNIT_SLICE) | |
2024 | (void) unit_realize_cgroup_now_disable(m, state); | |
2025 | ||
2026 | target_mask = unit_get_target_mask(m); | |
2027 | enable_mask = unit_get_enable_mask(m); | |
2028 | ||
2029 | /* We can only disable in this direction, don't try to enable | |
2030 | * anything. */ | |
2031 | if (unit_has_mask_disables_realized(m, target_mask, enable_mask)) | |
2032 | continue; | |
2033 | ||
2034 | new_target_mask = m->cgroup_realized_mask & target_mask; | |
2035 | new_enable_mask = m->cgroup_enabled_mask & enable_mask; | |
2036 | ||
2037 | r = unit_create_cgroup(m, new_target_mask, new_enable_mask, state); | |
2038 | if (r < 0) | |
2039 | return r; | |
2040 | } | |
2041 | ||
2042 | return 0; | |
2043 | } | |
a57669d2 | 2044 | |
6414b7c9 DS |
2045 | /* Check if necessary controllers and attributes for a unit are in place. |
2046 | * | |
a57669d2 CD |
2047 | * - If so, do nothing. |
2048 | * - If not, create paths, move processes over, and set attributes. | |
2049 | * | |
2050 | * Controllers can only be *enabled* in a breadth-first way, and *disabled* in | |
2051 | * a depth-first way. As such the process looks like this: | |
2052 | * | |
2053 | * Suppose we have a cgroup hierarchy which looks like this: | |
2054 | * | |
2055 | * root | |
2056 | * / \ | |
2057 | * / \ | |
2058 | * / \ | |
2059 | * a b | |
2060 | * / \ / \ | |
2061 | * / \ / \ | |
2062 | * c d e f | |
2063 | * / \ / \ / \ / \ | |
2064 | * h i j k l m n o | |
2065 | * | |
2066 | * 1. We want to realise cgroup "d" now. | |
c72703e2 | 2067 | * 2. cgroup "a" has DisableControllers=cpu in the associated unit. |
a57669d2 CD |
2068 | * 3. cgroup "k" just started requesting the memory controller. |
2069 | * | |
2070 | * To make this work we must do the following in order: | |
2071 | * | |
2072 | * 1. Disable CPU controller in k, j | |
2073 | * 2. Disable CPU controller in d | |
2074 | * 3. Enable memory controller in root | |
2075 | * 4. Enable memory controller in a | |
2076 | * 5. Enable memory controller in d | |
2077 | * 6. Enable memory controller in k | |
2078 | * | |
2079 | * Notice that we need to touch j in one direction, but not the other. We also | |
2080 | * don't go beyond d when disabling -- it's up to "a" to get realized if it | |
2081 | * wants to disable further. The basic rules are therefore: | |
2082 | * | |
2083 | * - If you're disabling something, you need to realise all of the cgroups from | |
2084 | * your recursive descendants to the root. This starts from the leaves. | |
2085 | * - If you're enabling something, you need to realise from the root cgroup | |
2086 | * downwards, but you don't need to iterate your recursive descendants. | |
6414b7c9 DS |
2087 | * |
2088 | * Returns 0 on success and < 0 on failure. */ | |
db785129 | 2089 | static int unit_realize_cgroup_now(Unit *u, ManagerState state) { |
efdb0237 | 2090 | CGroupMask target_mask, enable_mask; |
6414b7c9 | 2091 | int r; |
64747e2d | 2092 | |
4ad49000 | 2093 | assert(u); |
64747e2d | 2094 | |
2aa57a65 | 2095 | unit_remove_from_cgroup_realize_queue(u); |
64747e2d | 2096 | |
efdb0237 | 2097 | target_mask = unit_get_target_mask(u); |
ccf78df1 TH |
2098 | enable_mask = unit_get_enable_mask(u); |
2099 | ||
17f14955 | 2100 | if (unit_has_mask_realized(u, target_mask, enable_mask)) |
0a1eb06d | 2101 | return 0; |
64747e2d | 2102 | |
4f6f62e4 CD |
2103 | /* Disable controllers below us, if there are any */ |
2104 | r = unit_realize_cgroup_now_disable(u, state); | |
2105 | if (r < 0) | |
2106 | return r; | |
2107 | ||
2108 | /* Enable controllers above us, if there are any */ | |
6414b7c9 | 2109 | if (UNIT_ISSET(u->slice)) { |
a57669d2 | 2110 | r = unit_realize_cgroup_now_enable(UNIT_DEREF(u->slice), state); |
6414b7c9 DS |
2111 | if (r < 0) |
2112 | return r; | |
2113 | } | |
4ad49000 | 2114 | |
0d2d6fbf CD |
2115 | /* Now actually deal with the cgroup we were trying to realise and set attributes */ |
2116 | r = unit_create_cgroup(u, target_mask, enable_mask, state); | |
6414b7c9 DS |
2117 | if (r < 0) |
2118 | return r; | |
2119 | ||
c2baf11c LP |
2120 | /* Now, reset the invalidation mask */ |
2121 | u->cgroup_invalidated_mask = 0; | |
6414b7c9 | 2122 | return 0; |
64747e2d LP |
2123 | } |
2124 | ||
91a6073e | 2125 | unsigned manager_dispatch_cgroup_realize_queue(Manager *m) { |
db785129 | 2126 | ManagerState state; |
4ad49000 | 2127 | unsigned n = 0; |
db785129 | 2128 | Unit *i; |
6414b7c9 | 2129 | int r; |
ecedd90f | 2130 | |
91a6073e LP |
2131 | assert(m); |
2132 | ||
db785129 LP |
2133 | state = manager_state(m); |
2134 | ||
91a6073e LP |
2135 | while ((i = m->cgroup_realize_queue)) { |
2136 | assert(i->in_cgroup_realize_queue); | |
ecedd90f | 2137 | |
2aa57a65 LP |
2138 | if (UNIT_IS_INACTIVE_OR_FAILED(unit_active_state(i))) { |
2139 | /* Maybe things changed, and the unit is not actually active anymore? */ | |
2140 | unit_remove_from_cgroup_realize_queue(i); | |
2141 | continue; | |
2142 | } | |
2143 | ||
db785129 | 2144 | r = unit_realize_cgroup_now(i, state); |
6414b7c9 | 2145 | if (r < 0) |
efdb0237 | 2146 | log_warning_errno(r, "Failed to realize cgroups for queued unit %s, ignoring: %m", i->id); |
0a1eb06d | 2147 | |
4ad49000 LP |
2148 | n++; |
2149 | } | |
ecedd90f | 2150 | |
4ad49000 | 2151 | return n; |
8e274523 LP |
2152 | } |
2153 | ||
91a6073e | 2154 | static void unit_add_siblings_to_cgroup_realize_queue(Unit *u) { |
4ad49000 | 2155 | Unit *slice; |
ca949c9d | 2156 | |
4ad49000 LP |
2157 | /* This adds the siblings of the specified unit and the |
2158 | * siblings of all parent units to the cgroup queue. (But | |
2159 | * neither the specified unit itself nor the parents.) */ | |
2160 | ||
2161 | while ((slice = UNIT_DEREF(u->slice))) { | |
2162 | Iterator i; | |
2163 | Unit *m; | |
eef85c4a | 2164 | void *v; |
8f53a7b8 | 2165 | |
eef85c4a | 2166 | HASHMAP_FOREACH_KEY(v, m, u->dependencies[UNIT_BEFORE], i) { |
6414b7c9 DS |
2167 | /* Skip units that have a dependency on the slice |
2168 | * but aren't actually in it. */ | |
4ad49000 | 2169 | if (UNIT_DEREF(m->slice) != slice) |
50159e6a | 2170 | continue; |
8e274523 | 2171 | |
6414b7c9 DS |
2172 | /* No point in doing cgroup application for units |
2173 | * without active processes. */ | |
2174 | if (UNIT_IS_INACTIVE_OR_FAILED(unit_active_state(m))) | |
2175 | continue; | |
2176 | ||
2177 | /* If the unit doesn't need any new controllers | |
2178 | * and has current ones realized, it doesn't need | |
2179 | * any changes. */ | |
906c06f6 DM |
2180 | if (unit_has_mask_realized(m, |
2181 | unit_get_target_mask(m), | |
17f14955 | 2182 | unit_get_enable_mask(m))) |
6414b7c9 DS |
2183 | continue; |
2184 | ||
91a6073e | 2185 | unit_add_to_cgroup_realize_queue(m); |
50159e6a LP |
2186 | } |
2187 | ||
4ad49000 | 2188 | u = slice; |
8e274523 | 2189 | } |
4ad49000 LP |
2190 | } |
2191 | ||
0a1eb06d | 2192 | int unit_realize_cgroup(Unit *u) { |
4ad49000 LP |
2193 | assert(u); |
2194 | ||
35b7ff80 | 2195 | if (!UNIT_HAS_CGROUP_CONTEXT(u)) |
0a1eb06d | 2196 | return 0; |
8e274523 | 2197 | |
4ad49000 LP |
2198 | /* So, here's the deal: when realizing the cgroups for this |
2199 | * unit, we need to first create all parents, but there's more | |
2200 | * actually: for the weight-based controllers we also need to | |
2201 | * make sure that all our siblings (i.e. units that are in the | |
73e231ab | 2202 | * same slice as we are) have cgroups, too. Otherwise, things |
4ad49000 LP |
2203 | * would become very uneven as each of their processes would |
2204 | * get as much resources as all our group together. This call | |
2205 | * will synchronously create the parent cgroups, but will | |
2206 | * defer work on the siblings to the next event loop | |
2207 | * iteration. */ | |
ca949c9d | 2208 | |
4ad49000 | 2209 | /* Add all sibling slices to the cgroup queue. */ |
91a6073e | 2210 | unit_add_siblings_to_cgroup_realize_queue(u); |
4ad49000 | 2211 | |
6414b7c9 | 2212 | /* And realize this one now (and apply the values) */ |
db785129 | 2213 | return unit_realize_cgroup_now(u, manager_state(u->manager)); |
8e274523 LP |
2214 | } |
2215 | ||
efdb0237 LP |
2216 | void unit_release_cgroup(Unit *u) { |
2217 | assert(u); | |
2218 | ||
8a0d5388 LP |
2219 | /* Forgets all cgroup details for this cgroup — but does *not* destroy the cgroup. This is hence OK to call |
2220 | * when we close down everything for reexecution, where we really want to leave the cgroup in place. */ | |
efdb0237 LP |
2221 | |
2222 | if (u->cgroup_path) { | |
2223 | (void) hashmap_remove(u->manager->cgroup_unit, u->cgroup_path); | |
2224 | u->cgroup_path = mfree(u->cgroup_path); | |
2225 | } | |
2226 | ||
2227 | if (u->cgroup_inotify_wd >= 0) { | |
2228 | if (inotify_rm_watch(u->manager->cgroup_inotify_fd, u->cgroup_inotify_wd) < 0) | |
5e1ee764 | 2229 | 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 |
2230 | |
2231 | (void) hashmap_remove(u->manager->cgroup_inotify_wd_unit, INT_TO_PTR(u->cgroup_inotify_wd)); | |
2232 | u->cgroup_inotify_wd = -1; | |
2233 | } | |
2234 | } | |
2235 | ||
2236 | void unit_prune_cgroup(Unit *u) { | |
8e274523 | 2237 | int r; |
efdb0237 | 2238 | bool is_root_slice; |
8e274523 | 2239 | |
4ad49000 | 2240 | assert(u); |
8e274523 | 2241 | |
efdb0237 LP |
2242 | /* Removes the cgroup, if empty and possible, and stops watching it. */ |
2243 | ||
4ad49000 LP |
2244 | if (!u->cgroup_path) |
2245 | return; | |
8e274523 | 2246 | |
fe700f46 LP |
2247 | (void) unit_get_cpu_usage(u, NULL); /* Cache the last CPU usage value before we destroy the cgroup */ |
2248 | ||
efdb0237 LP |
2249 | is_root_slice = unit_has_name(u, SPECIAL_ROOT_SLICE); |
2250 | ||
2251 | r = cg_trim_everywhere(u->manager->cgroup_supported, u->cgroup_path, !is_root_slice); | |
dab5bf85 | 2252 | if (r < 0) { |
f29ff115 | 2253 | log_unit_debug_errno(u, r, "Failed to destroy cgroup %s, ignoring: %m", u->cgroup_path); |
dab5bf85 RL |
2254 | return; |
2255 | } | |
8e274523 | 2256 | |
efdb0237 LP |
2257 | if (is_root_slice) |
2258 | return; | |
2259 | ||
2260 | unit_release_cgroup(u); | |
0a1eb06d | 2261 | |
4ad49000 | 2262 | u->cgroup_realized = false; |
bc432dc7 | 2263 | u->cgroup_realized_mask = 0; |
ccf78df1 | 2264 | u->cgroup_enabled_mask = 0; |
084c7007 RG |
2265 | |
2266 | u->bpf_device_control_installed = bpf_program_unref(u->bpf_device_control_installed); | |
8e274523 LP |
2267 | } |
2268 | ||
efdb0237 | 2269 | int unit_search_main_pid(Unit *u, pid_t *ret) { |
4ad49000 LP |
2270 | _cleanup_fclose_ FILE *f = NULL; |
2271 | pid_t pid = 0, npid, mypid; | |
efdb0237 | 2272 | int r; |
4ad49000 LP |
2273 | |
2274 | assert(u); | |
efdb0237 | 2275 | assert(ret); |
4ad49000 LP |
2276 | |
2277 | if (!u->cgroup_path) | |
efdb0237 | 2278 | return -ENXIO; |
4ad49000 | 2279 | |
efdb0237 LP |
2280 | r = cg_enumerate_processes(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path, &f); |
2281 | if (r < 0) | |
2282 | return r; | |
4ad49000 | 2283 | |
df0ff127 | 2284 | mypid = getpid_cached(); |
4ad49000 LP |
2285 | while (cg_read_pid(f, &npid) > 0) { |
2286 | pid_t ppid; | |
2287 | ||
2288 | if (npid == pid) | |
2289 | continue; | |
8e274523 | 2290 | |
4ad49000 | 2291 | /* Ignore processes that aren't our kids */ |
6bc73acb | 2292 | if (get_process_ppid(npid, &ppid) >= 0 && ppid != mypid) |
4ad49000 | 2293 | continue; |
8e274523 | 2294 | |
efdb0237 | 2295 | if (pid != 0) |
4ad49000 LP |
2296 | /* Dang, there's more than one daemonized PID |
2297 | in this group, so we don't know what process | |
2298 | is the main process. */ | |
efdb0237 LP |
2299 | |
2300 | return -ENODATA; | |
8e274523 | 2301 | |
4ad49000 | 2302 | pid = npid; |
8e274523 LP |
2303 | } |
2304 | ||
efdb0237 LP |
2305 | *ret = pid; |
2306 | return 0; | |
2307 | } | |
2308 | ||
2309 | static int unit_watch_pids_in_path(Unit *u, const char *path) { | |
b3c5bad3 | 2310 | _cleanup_closedir_ DIR *d = NULL; |
efdb0237 LP |
2311 | _cleanup_fclose_ FILE *f = NULL; |
2312 | int ret = 0, r; | |
2313 | ||
2314 | assert(u); | |
2315 | assert(path); | |
2316 | ||
2317 | r = cg_enumerate_processes(SYSTEMD_CGROUP_CONTROLLER, path, &f); | |
2318 | if (r < 0) | |
2319 | ret = r; | |
2320 | else { | |
2321 | pid_t pid; | |
2322 | ||
2323 | while ((r = cg_read_pid(f, &pid)) > 0) { | |
2324 | r = unit_watch_pid(u, pid); | |
2325 | if (r < 0 && ret >= 0) | |
2326 | ret = r; | |
2327 | } | |
2328 | ||
2329 | if (r < 0 && ret >= 0) | |
2330 | ret = r; | |
2331 | } | |
2332 | ||
2333 | r = cg_enumerate_subgroups(SYSTEMD_CGROUP_CONTROLLER, path, &d); | |
2334 | if (r < 0) { | |
2335 | if (ret >= 0) | |
2336 | ret = r; | |
2337 | } else { | |
2338 | char *fn; | |
2339 | ||
2340 | while ((r = cg_read_subgroup(d, &fn)) > 0) { | |
2341 | _cleanup_free_ char *p = NULL; | |
2342 | ||
605405c6 | 2343 | p = strjoin(path, "/", fn); |
efdb0237 LP |
2344 | free(fn); |
2345 | ||
2346 | if (!p) | |
2347 | return -ENOMEM; | |
2348 | ||
2349 | r = unit_watch_pids_in_path(u, p); | |
2350 | if (r < 0 && ret >= 0) | |
2351 | ret = r; | |
2352 | } | |
2353 | ||
2354 | if (r < 0 && ret >= 0) | |
2355 | ret = r; | |
2356 | } | |
2357 | ||
2358 | return ret; | |
2359 | } | |
2360 | ||
11aef522 LP |
2361 | int unit_synthesize_cgroup_empty_event(Unit *u) { |
2362 | int r; | |
2363 | ||
2364 | assert(u); | |
2365 | ||
2366 | /* Enqueue a synthetic cgroup empty event if this unit doesn't watch any PIDs anymore. This is compatibility | |
2367 | * support for non-unified systems where notifications aren't reliable, and hence need to take whatever we can | |
2368 | * get as notification source as soon as we stopped having any useful PIDs to watch for. */ | |
2369 | ||
2370 | if (!u->cgroup_path) | |
2371 | return -ENOENT; | |
2372 | ||
2373 | r = cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER); | |
2374 | if (r < 0) | |
2375 | return r; | |
2376 | if (r > 0) /* On unified we have reliable notifications, and don't need this */ | |
2377 | return 0; | |
2378 | ||
2379 | if (!set_isempty(u->pids)) | |
2380 | return 0; | |
2381 | ||
2382 | unit_add_to_cgroup_empty_queue(u); | |
2383 | return 0; | |
2384 | } | |
2385 | ||
efdb0237 | 2386 | int unit_watch_all_pids(Unit *u) { |
b4cccbc1 LP |
2387 | int r; |
2388 | ||
efdb0237 LP |
2389 | assert(u); |
2390 | ||
2391 | /* Adds all PIDs from our cgroup to the set of PIDs we | |
2392 | * watch. This is a fallback logic for cases where we do not | |
2393 | * get reliable cgroup empty notifications: we try to use | |
2394 | * SIGCHLD as replacement. */ | |
2395 | ||
2396 | if (!u->cgroup_path) | |
2397 | return -ENOENT; | |
2398 | ||
c22800e4 | 2399 | r = cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER); |
b4cccbc1 LP |
2400 | if (r < 0) |
2401 | return r; | |
2402 | if (r > 0) /* On unified we can use proper notifications */ | |
efdb0237 LP |
2403 | return 0; |
2404 | ||
2405 | return unit_watch_pids_in_path(u, u->cgroup_path); | |
2406 | } | |
2407 | ||
09e24654 LP |
2408 | static int on_cgroup_empty_event(sd_event_source *s, void *userdata) { |
2409 | Manager *m = userdata; | |
2410 | Unit *u; | |
efdb0237 LP |
2411 | int r; |
2412 | ||
09e24654 LP |
2413 | assert(s); |
2414 | assert(m); | |
efdb0237 | 2415 | |
09e24654 LP |
2416 | u = m->cgroup_empty_queue; |
2417 | if (!u) | |
efdb0237 LP |
2418 | return 0; |
2419 | ||
09e24654 LP |
2420 | assert(u->in_cgroup_empty_queue); |
2421 | u->in_cgroup_empty_queue = false; | |
2422 | LIST_REMOVE(cgroup_empty_queue, m->cgroup_empty_queue, u); | |
2423 | ||
2424 | if (m->cgroup_empty_queue) { | |
2425 | /* More stuff queued, let's make sure we remain enabled */ | |
2426 | r = sd_event_source_set_enabled(s, SD_EVENT_ONESHOT); | |
2427 | if (r < 0) | |
19a691a9 | 2428 | log_debug_errno(r, "Failed to reenable cgroup empty event source, ignoring: %m"); |
09e24654 | 2429 | } |
efdb0237 LP |
2430 | |
2431 | unit_add_to_gc_queue(u); | |
2432 | ||
2433 | if (UNIT_VTABLE(u)->notify_cgroup_empty) | |
2434 | UNIT_VTABLE(u)->notify_cgroup_empty(u); | |
2435 | ||
2436 | return 0; | |
2437 | } | |
2438 | ||
09e24654 LP |
2439 | void unit_add_to_cgroup_empty_queue(Unit *u) { |
2440 | int r; | |
2441 | ||
2442 | assert(u); | |
2443 | ||
2444 | /* Note that there are four different ways how cgroup empty events reach us: | |
2445 | * | |
2446 | * 1. On the unified hierarchy we get an inotify event on the cgroup | |
2447 | * | |
2448 | * 2. On the legacy hierarchy, when running in system mode, we get a datagram on the cgroup agent socket | |
2449 | * | |
2450 | * 3. On the legacy hierarchy, when running in user mode, we get a D-Bus signal on the system bus | |
2451 | * | |
2452 | * 4. On the legacy hierarchy, in service units we start watching all processes of the cgroup for SIGCHLD as | |
2453 | * soon as we get one SIGCHLD, to deal with unreliable cgroup notifications. | |
2454 | * | |
2455 | * Regardless which way we got the notification, we'll verify it here, and then add it to a separate | |
2456 | * queue. This queue will be dispatched at a lower priority than the SIGCHLD handler, so that we always use | |
2457 | * SIGCHLD if we can get it first, and only use the cgroup empty notifications if there's no SIGCHLD pending | |
2458 | * (which might happen if the cgroup doesn't contain processes that are our own child, which is typically the | |
2459 | * case for scope units). */ | |
2460 | ||
2461 | if (u->in_cgroup_empty_queue) | |
2462 | return; | |
2463 | ||
2464 | /* Let's verify that the cgroup is really empty */ | |
2465 | if (!u->cgroup_path) | |
2466 | return; | |
2467 | r = cg_is_empty_recursive(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path); | |
2468 | if (r < 0) { | |
2469 | log_unit_debug_errno(u, r, "Failed to determine whether cgroup %s is empty: %m", u->cgroup_path); | |
2470 | return; | |
2471 | } | |
2472 | if (r == 0) | |
2473 | return; | |
2474 | ||
2475 | LIST_PREPEND(cgroup_empty_queue, u->manager->cgroup_empty_queue, u); | |
2476 | u->in_cgroup_empty_queue = true; | |
2477 | ||
2478 | /* Trigger the defer event */ | |
2479 | r = sd_event_source_set_enabled(u->manager->cgroup_empty_event_source, SD_EVENT_ONESHOT); | |
2480 | if (r < 0) | |
2481 | log_debug_errno(r, "Failed to enable cgroup empty event source: %m"); | |
2482 | } | |
2483 | ||
efdb0237 LP |
2484 | static int on_cgroup_inotify_event(sd_event_source *s, int fd, uint32_t revents, void *userdata) { |
2485 | Manager *m = userdata; | |
2486 | ||
2487 | assert(s); | |
2488 | assert(fd >= 0); | |
2489 | assert(m); | |
2490 | ||
2491 | for (;;) { | |
2492 | union inotify_event_buffer buffer; | |
2493 | struct inotify_event *e; | |
2494 | ssize_t l; | |
2495 | ||
2496 | l = read(fd, &buffer, sizeof(buffer)); | |
2497 | if (l < 0) { | |
47249640 | 2498 | if (IN_SET(errno, EINTR, EAGAIN)) |
efdb0237 LP |
2499 | return 0; |
2500 | ||
2501 | return log_error_errno(errno, "Failed to read control group inotify events: %m"); | |
2502 | } | |
2503 | ||
2504 | FOREACH_INOTIFY_EVENT(e, buffer, l) { | |
2505 | Unit *u; | |
2506 | ||
2507 | if (e->wd < 0) | |
2508 | /* Queue overflow has no watch descriptor */ | |
2509 | continue; | |
2510 | ||
2511 | if (e->mask & IN_IGNORED) | |
2512 | /* The watch was just removed */ | |
2513 | continue; | |
2514 | ||
2515 | u = hashmap_get(m->cgroup_inotify_wd_unit, INT_TO_PTR(e->wd)); | |
2516 | if (!u) /* Not that inotify might deliver | |
2517 | * events for a watch even after it | |
2518 | * was removed, because it was queued | |
2519 | * before the removal. Let's ignore | |
2520 | * this here safely. */ | |
2521 | continue; | |
2522 | ||
09e24654 | 2523 | unit_add_to_cgroup_empty_queue(u); |
efdb0237 LP |
2524 | } |
2525 | } | |
8e274523 LP |
2526 | } |
2527 | ||
17f14955 RG |
2528 | static int cg_bpf_mask_supported(CGroupMask *ret) { |
2529 | CGroupMask mask = 0; | |
2530 | int r; | |
2531 | ||
2532 | /* BPF-based firewall */ | |
2533 | r = bpf_firewall_supported(); | |
2534 | if (r > 0) | |
2535 | mask |= CGROUP_MASK_BPF_FIREWALL; | |
2536 | ||
084c7007 RG |
2537 | /* BPF-based device access control */ |
2538 | r = bpf_devices_supported(); | |
2539 | if (r > 0) | |
2540 | mask |= CGROUP_MASK_BPF_DEVICES; | |
2541 | ||
17f14955 RG |
2542 | *ret = mask; |
2543 | return 0; | |
2544 | } | |
2545 | ||
8e274523 | 2546 | int manager_setup_cgroup(Manager *m) { |
9444b1f2 | 2547 | _cleanup_free_ char *path = NULL; |
10bd3e2e | 2548 | const char *scope_path; |
efdb0237 | 2549 | CGroupController c; |
b4cccbc1 | 2550 | int r, all_unified; |
17f14955 | 2551 | CGroupMask mask; |
efdb0237 | 2552 | char *e; |
8e274523 LP |
2553 | |
2554 | assert(m); | |
2555 | ||
35d2e7ec | 2556 | /* 1. Determine hierarchy */ |
efdb0237 | 2557 | m->cgroup_root = mfree(m->cgroup_root); |
9444b1f2 | 2558 | r = cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER, 0, &m->cgroup_root); |
23bbb0de MS |
2559 | if (r < 0) |
2560 | return log_error_errno(r, "Cannot determine cgroup we are running in: %m"); | |
8e274523 | 2561 | |
efdb0237 LP |
2562 | /* Chop off the init scope, if we are already located in it */ |
2563 | e = endswith(m->cgroup_root, "/" SPECIAL_INIT_SCOPE); | |
0d8c31ff | 2564 | |
efdb0237 LP |
2565 | /* LEGACY: Also chop off the system slice if we are in |
2566 | * it. This is to support live upgrades from older systemd | |
2567 | * versions where PID 1 was moved there. Also see | |
2568 | * cg_get_root_path(). */ | |
463d0d15 | 2569 | if (!e && MANAGER_IS_SYSTEM(m)) { |
9444b1f2 | 2570 | e = endswith(m->cgroup_root, "/" SPECIAL_SYSTEM_SLICE); |
15c60e99 | 2571 | if (!e) |
efdb0237 | 2572 | e = endswith(m->cgroup_root, "/system"); /* even more legacy */ |
0baf24dd | 2573 | } |
efdb0237 LP |
2574 | if (e) |
2575 | *e = 0; | |
7ccfb64a | 2576 | |
7546145e LP |
2577 | /* And make sure to store away the root value without trailing slash, even for the root dir, so that we can |
2578 | * easily prepend it everywhere. */ | |
2579 | delete_trailing_chars(m->cgroup_root, "/"); | |
8e274523 | 2580 | |
35d2e7ec | 2581 | /* 2. Show data */ |
9444b1f2 | 2582 | r = cg_get_path(SYSTEMD_CGROUP_CONTROLLER, m->cgroup_root, NULL, &path); |
23bbb0de MS |
2583 | if (r < 0) |
2584 | return log_error_errno(r, "Cannot find cgroup mount point: %m"); | |
8e274523 | 2585 | |
415fc41c TH |
2586 | r = cg_unified_flush(); |
2587 | if (r < 0) | |
2588 | return log_error_errno(r, "Couldn't determine if we are running in the unified hierarchy: %m"); | |
5da38d07 | 2589 | |
b4cccbc1 | 2590 | all_unified = cg_all_unified(); |
d4c819ed ZJS |
2591 | if (all_unified < 0) |
2592 | return log_error_errno(all_unified, "Couldn't determine whether we are in all unified mode: %m"); | |
2593 | if (all_unified > 0) | |
efdb0237 | 2594 | log_debug("Unified cgroup hierarchy is located at %s.", path); |
b4cccbc1 | 2595 | else { |
c22800e4 | 2596 | r = cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER); |
b4cccbc1 LP |
2597 | if (r < 0) |
2598 | return log_error_errno(r, "Failed to determine whether systemd's own controller is in unified mode: %m"); | |
2599 | if (r > 0) | |
2600 | log_debug("Unified cgroup hierarchy is located at %s. Controllers are on legacy hierarchies.", path); | |
2601 | else | |
2602 | log_debug("Using cgroup controller " SYSTEMD_CGROUP_CONTROLLER_LEGACY ". File system hierarchy is at %s.", path); | |
2603 | } | |
efdb0237 | 2604 | |
09e24654 LP |
2605 | /* 3. Allocate cgroup empty defer event source */ |
2606 | m->cgroup_empty_event_source = sd_event_source_unref(m->cgroup_empty_event_source); | |
2607 | r = sd_event_add_defer(m->event, &m->cgroup_empty_event_source, on_cgroup_empty_event, m); | |
2608 | if (r < 0) | |
2609 | return log_error_errno(r, "Failed to create cgroup empty event source: %m"); | |
2610 | ||
2611 | r = sd_event_source_set_priority(m->cgroup_empty_event_source, SD_EVENT_PRIORITY_NORMAL-5); | |
2612 | if (r < 0) | |
2613 | return log_error_errno(r, "Failed to set priority of cgroup empty event source: %m"); | |
2614 | ||
2615 | r = sd_event_source_set_enabled(m->cgroup_empty_event_source, SD_EVENT_OFF); | |
2616 | if (r < 0) | |
2617 | return log_error_errno(r, "Failed to disable cgroup empty event source: %m"); | |
2618 | ||
2619 | (void) sd_event_source_set_description(m->cgroup_empty_event_source, "cgroup-empty"); | |
2620 | ||
2621 | /* 4. Install notifier inotify object, or agent */ | |
10bd3e2e | 2622 | if (cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER) > 0) { |
c6c18be3 | 2623 | |
09e24654 | 2624 | /* In the unified hierarchy we can get cgroup empty notifications via inotify. */ |
efdb0237 | 2625 | |
10bd3e2e LP |
2626 | m->cgroup_inotify_event_source = sd_event_source_unref(m->cgroup_inotify_event_source); |
2627 | safe_close(m->cgroup_inotify_fd); | |
efdb0237 | 2628 | |
10bd3e2e LP |
2629 | m->cgroup_inotify_fd = inotify_init1(IN_NONBLOCK|IN_CLOEXEC); |
2630 | if (m->cgroup_inotify_fd < 0) | |
2631 | return log_error_errno(errno, "Failed to create control group inotify object: %m"); | |
efdb0237 | 2632 | |
10bd3e2e LP |
2633 | r = sd_event_add_io(m->event, &m->cgroup_inotify_event_source, m->cgroup_inotify_fd, EPOLLIN, on_cgroup_inotify_event, m); |
2634 | if (r < 0) | |
2635 | return log_error_errno(r, "Failed to watch control group inotify object: %m"); | |
efdb0237 | 2636 | |
10bd3e2e LP |
2637 | /* Process cgroup empty notifications early, but after service notifications and SIGCHLD. Also |
2638 | * see handling of cgroup agent notifications, for the classic cgroup hierarchy support. */ | |
09e24654 | 2639 | r = sd_event_source_set_priority(m->cgroup_inotify_event_source, SD_EVENT_PRIORITY_NORMAL-4); |
10bd3e2e LP |
2640 | if (r < 0) |
2641 | return log_error_errno(r, "Failed to set priority of inotify event source: %m"); | |
efdb0237 | 2642 | |
10bd3e2e | 2643 | (void) sd_event_source_set_description(m->cgroup_inotify_event_source, "cgroup-inotify"); |
efdb0237 | 2644 | |
611c4f8a | 2645 | } else if (MANAGER_IS_SYSTEM(m) && manager_owns_host_root_cgroup(m) && !MANAGER_IS_TEST_RUN(m)) { |
efdb0237 | 2646 | |
10bd3e2e LP |
2647 | /* On the legacy hierarchy we only get notifications via cgroup agents. (Which isn't really reliable, |
2648 | * since it does not generate events when control groups with children run empty. */ | |
8e274523 | 2649 | |
10bd3e2e | 2650 | r = cg_install_release_agent(SYSTEMD_CGROUP_CONTROLLER, SYSTEMD_CGROUP_AGENT_PATH); |
23bbb0de | 2651 | if (r < 0) |
10bd3e2e LP |
2652 | log_warning_errno(r, "Failed to install release agent, ignoring: %m"); |
2653 | else if (r > 0) | |
2654 | log_debug("Installed release agent."); | |
2655 | else if (r == 0) | |
2656 | log_debug("Release agent already installed."); | |
2657 | } | |
efdb0237 | 2658 | |
09e24654 | 2659 | /* 5. Make sure we are in the special "init.scope" unit in the root slice. */ |
10bd3e2e LP |
2660 | scope_path = strjoina(m->cgroup_root, "/" SPECIAL_INIT_SCOPE); |
2661 | r = cg_create_and_attach(SYSTEMD_CGROUP_CONTROLLER, scope_path, 0); | |
aa77e234 MS |
2662 | if (r >= 0) { |
2663 | /* Also, move all other userspace processes remaining in the root cgroup into that scope. */ | |
2664 | r = cg_migrate(SYSTEMD_CGROUP_CONTROLLER, m->cgroup_root, SYSTEMD_CGROUP_CONTROLLER, scope_path, 0); | |
2665 | if (r < 0) | |
2666 | log_warning_errno(r, "Couldn't move remaining userspace processes, ignoring: %m"); | |
c6c18be3 | 2667 | |
aa77e234 MS |
2668 | /* 6. And pin it, so that it cannot be unmounted */ |
2669 | safe_close(m->pin_cgroupfs_fd); | |
2670 | m->pin_cgroupfs_fd = open(path, O_RDONLY|O_CLOEXEC|O_DIRECTORY|O_NOCTTY|O_NONBLOCK); | |
2671 | if (m->pin_cgroupfs_fd < 0) | |
2672 | return log_error_errno(errno, "Failed to open pin file: %m"); | |
0d8c31ff | 2673 | |
638cece4 | 2674 | } else if (!MANAGER_IS_TEST_RUN(m)) |
aa77e234 | 2675 | return log_error_errno(r, "Failed to create %s control group: %m", scope_path); |
10bd3e2e | 2676 | |
09e24654 | 2677 | /* 7. Always enable hierarchical support if it exists... */ |
638cece4 | 2678 | if (!all_unified && !MANAGER_IS_TEST_RUN(m)) |
10bd3e2e | 2679 | (void) cg_set_attribute("memory", "/", "memory.use_hierarchy", "1"); |
c6c18be3 | 2680 | |
17f14955 | 2681 | /* 8. Figure out which controllers are supported */ |
efdb0237 LP |
2682 | r = cg_mask_supported(&m->cgroup_supported); |
2683 | if (r < 0) | |
2684 | return log_error_errno(r, "Failed to determine supported controllers: %m"); | |
17f14955 RG |
2685 | |
2686 | /* 9. Figure out which bpf-based pseudo-controllers are supported */ | |
2687 | r = cg_bpf_mask_supported(&mask); | |
2688 | if (r < 0) | |
2689 | return log_error_errno(r, "Failed to determine supported bpf-based pseudo-controllers: %m"); | |
2690 | m->cgroup_supported |= mask; | |
2691 | ||
2692 | /* 10. Log which controllers are supported */ | |
efdb0237 | 2693 | for (c = 0; c < _CGROUP_CONTROLLER_MAX; c++) |
eee0a1e4 | 2694 | log_debug("Controller '%s' supported: %s", cgroup_controller_to_string(c), yes_no(m->cgroup_supported & CGROUP_CONTROLLER_TO_MASK(c))); |
9156e799 | 2695 | |
a32360f1 | 2696 | return 0; |
8e274523 LP |
2697 | } |
2698 | ||
c6c18be3 | 2699 | void manager_shutdown_cgroup(Manager *m, bool delete) { |
8e274523 LP |
2700 | assert(m); |
2701 | ||
9444b1f2 LP |
2702 | /* We can't really delete the group, since we are in it. But |
2703 | * let's trim it. */ | |
f6c63f6f | 2704 | if (delete && m->cgroup_root && m->test_run_flags != MANAGER_TEST_RUN_MINIMAL) |
efdb0237 LP |
2705 | (void) cg_trim(SYSTEMD_CGROUP_CONTROLLER, m->cgroup_root, false); |
2706 | ||
09e24654 LP |
2707 | m->cgroup_empty_event_source = sd_event_source_unref(m->cgroup_empty_event_source); |
2708 | ||
efdb0237 LP |
2709 | m->cgroup_inotify_wd_unit = hashmap_free(m->cgroup_inotify_wd_unit); |
2710 | ||
2711 | m->cgroup_inotify_event_source = sd_event_source_unref(m->cgroup_inotify_event_source); | |
2712 | m->cgroup_inotify_fd = safe_close(m->cgroup_inotify_fd); | |
8e274523 | 2713 | |
03e334a1 | 2714 | m->pin_cgroupfs_fd = safe_close(m->pin_cgroupfs_fd); |
c6c18be3 | 2715 | |
efdb0237 | 2716 | m->cgroup_root = mfree(m->cgroup_root); |
8e274523 LP |
2717 | } |
2718 | ||
4ad49000 | 2719 | Unit* manager_get_unit_by_cgroup(Manager *m, const char *cgroup) { |
acb14d31 | 2720 | char *p; |
4ad49000 | 2721 | Unit *u; |
acb14d31 LP |
2722 | |
2723 | assert(m); | |
2724 | assert(cgroup); | |
acb14d31 | 2725 | |
4ad49000 LP |
2726 | u = hashmap_get(m->cgroup_unit, cgroup); |
2727 | if (u) | |
2728 | return u; | |
acb14d31 | 2729 | |
8e70580b | 2730 | p = strdupa(cgroup); |
acb14d31 LP |
2731 | for (;;) { |
2732 | char *e; | |
2733 | ||
2734 | e = strrchr(p, '/'); | |
efdb0237 LP |
2735 | if (!e || e == p) |
2736 | return hashmap_get(m->cgroup_unit, SPECIAL_ROOT_SLICE); | |
acb14d31 LP |
2737 | |
2738 | *e = 0; | |
2739 | ||
4ad49000 LP |
2740 | u = hashmap_get(m->cgroup_unit, p); |
2741 | if (u) | |
2742 | return u; | |
acb14d31 LP |
2743 | } |
2744 | } | |
2745 | ||
b3ac818b | 2746 | Unit *manager_get_unit_by_pid_cgroup(Manager *m, pid_t pid) { |
4ad49000 | 2747 | _cleanup_free_ char *cgroup = NULL; |
8e274523 | 2748 | |
8c47c732 LP |
2749 | assert(m); |
2750 | ||
62a76913 | 2751 | if (!pid_is_valid(pid)) |
b3ac818b LP |
2752 | return NULL; |
2753 | ||
62a76913 | 2754 | if (cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER, pid, &cgroup) < 0) |
b3ac818b LP |
2755 | return NULL; |
2756 | ||
2757 | return manager_get_unit_by_cgroup(m, cgroup); | |
2758 | } | |
2759 | ||
2760 | Unit *manager_get_unit_by_pid(Manager *m, pid_t pid) { | |
62a76913 | 2761 | Unit *u, **array; |
b3ac818b LP |
2762 | |
2763 | assert(m); | |
2764 | ||
62a76913 LP |
2765 | /* Note that a process might be owned by multiple units, we return only one here, which is good enough for most |
2766 | * cases, though not strictly correct. We prefer the one reported by cgroup membership, as that's the most | |
2767 | * relevant one as children of the process will be assigned to that one, too, before all else. */ | |
2768 | ||
2769 | if (!pid_is_valid(pid)) | |
8c47c732 LP |
2770 | return NULL; |
2771 | ||
2ca9d979 | 2772 | if (pid == getpid_cached()) |
efdb0237 LP |
2773 | return hashmap_get(m->units, SPECIAL_INIT_SCOPE); |
2774 | ||
62a76913 | 2775 | u = manager_get_unit_by_pid_cgroup(m, pid); |
5fe8876b LP |
2776 | if (u) |
2777 | return u; | |
2778 | ||
62a76913 | 2779 | u = hashmap_get(m->watch_pids, PID_TO_PTR(pid)); |
5fe8876b LP |
2780 | if (u) |
2781 | return u; | |
2782 | ||
62a76913 LP |
2783 | array = hashmap_get(m->watch_pids, PID_TO_PTR(-pid)); |
2784 | if (array) | |
2785 | return array[0]; | |
2786 | ||
2787 | return NULL; | |
6dde1f33 | 2788 | } |
4fbf50b3 | 2789 | |
4ad49000 LP |
2790 | int manager_notify_cgroup_empty(Manager *m, const char *cgroup) { |
2791 | Unit *u; | |
4fbf50b3 | 2792 | |
4ad49000 LP |
2793 | assert(m); |
2794 | assert(cgroup); | |
4fbf50b3 | 2795 | |
09e24654 LP |
2796 | /* Called on the legacy hierarchy whenever we get an explicit cgroup notification from the cgroup agent process |
2797 | * or from the --system instance */ | |
2798 | ||
d8fdc620 LP |
2799 | log_debug("Got cgroup empty notification for: %s", cgroup); |
2800 | ||
4ad49000 | 2801 | u = manager_get_unit_by_cgroup(m, cgroup); |
5ad096b3 LP |
2802 | if (!u) |
2803 | return 0; | |
b56c28c3 | 2804 | |
09e24654 LP |
2805 | unit_add_to_cgroup_empty_queue(u); |
2806 | return 1; | |
5ad096b3 LP |
2807 | } |
2808 | ||
2809 | int unit_get_memory_current(Unit *u, uint64_t *ret) { | |
2810 | _cleanup_free_ char *v = NULL; | |
2811 | int r; | |
2812 | ||
2813 | assert(u); | |
2814 | assert(ret); | |
2815 | ||
2e4025c0 | 2816 | if (!UNIT_CGROUP_BOOL(u, memory_accounting)) |
cf3b4be1 LP |
2817 | return -ENODATA; |
2818 | ||
5ad096b3 LP |
2819 | if (!u->cgroup_path) |
2820 | return -ENODATA; | |
2821 | ||
1f73aa00 | 2822 | /* The root cgroup doesn't expose this information, let's get it from /proc instead */ |
611c4f8a | 2823 | if (unit_has_host_root_cgroup(u)) |
c482724a | 2824 | return procfs_memory_get_used(ret); |
1f73aa00 | 2825 | |
efdb0237 | 2826 | if ((u->cgroup_realized_mask & CGROUP_MASK_MEMORY) == 0) |
5ad096b3 LP |
2827 | return -ENODATA; |
2828 | ||
b4cccbc1 LP |
2829 | r = cg_all_unified(); |
2830 | if (r < 0) | |
2831 | return r; | |
2832 | if (r > 0) | |
efdb0237 | 2833 | r = cg_get_attribute("memory", u->cgroup_path, "memory.current", &v); |
b4cccbc1 LP |
2834 | else |
2835 | r = cg_get_attribute("memory", u->cgroup_path, "memory.usage_in_bytes", &v); | |
5ad096b3 LP |
2836 | if (r == -ENOENT) |
2837 | return -ENODATA; | |
2838 | if (r < 0) | |
2839 | return r; | |
2840 | ||
2841 | return safe_atou64(v, ret); | |
2842 | } | |
2843 | ||
03a7b521 LP |
2844 | int unit_get_tasks_current(Unit *u, uint64_t *ret) { |
2845 | _cleanup_free_ char *v = NULL; | |
2846 | int r; | |
2847 | ||
2848 | assert(u); | |
2849 | assert(ret); | |
2850 | ||
2e4025c0 | 2851 | if (!UNIT_CGROUP_BOOL(u, tasks_accounting)) |
cf3b4be1 LP |
2852 | return -ENODATA; |
2853 | ||
03a7b521 LP |
2854 | if (!u->cgroup_path) |
2855 | return -ENODATA; | |
2856 | ||
c36a69f4 | 2857 | /* The root cgroup doesn't expose this information, let's get it from /proc instead */ |
611c4f8a | 2858 | if (unit_has_host_root_cgroup(u)) |
c36a69f4 LP |
2859 | return procfs_tasks_get_current(ret); |
2860 | ||
1f73aa00 LP |
2861 | if ((u->cgroup_realized_mask & CGROUP_MASK_PIDS) == 0) |
2862 | return -ENODATA; | |
2863 | ||
03a7b521 LP |
2864 | r = cg_get_attribute("pids", u->cgroup_path, "pids.current", &v); |
2865 | if (r == -ENOENT) | |
2866 | return -ENODATA; | |
2867 | if (r < 0) | |
2868 | return r; | |
2869 | ||
2870 | return safe_atou64(v, ret); | |
2871 | } | |
2872 | ||
5ad096b3 LP |
2873 | static int unit_get_cpu_usage_raw(Unit *u, nsec_t *ret) { |
2874 | _cleanup_free_ char *v = NULL; | |
2875 | uint64_t ns; | |
2876 | int r; | |
2877 | ||
2878 | assert(u); | |
2879 | assert(ret); | |
2880 | ||
2881 | if (!u->cgroup_path) | |
2882 | return -ENODATA; | |
2883 | ||
1f73aa00 | 2884 | /* The root cgroup doesn't expose this information, let's get it from /proc instead */ |
611c4f8a | 2885 | if (unit_has_host_root_cgroup(u)) |
1f73aa00 LP |
2886 | return procfs_cpu_get_usage(ret); |
2887 | ||
f98c2585 CD |
2888 | /* Requisite controllers for CPU accounting are not enabled */ |
2889 | if ((get_cpu_accounting_mask() & ~u->cgroup_realized_mask) != 0) | |
2890 | return -ENODATA; | |
2891 | ||
92a99304 LP |
2892 | r = cg_all_unified(); |
2893 | if (r < 0) | |
2894 | return r; | |
b4cccbc1 | 2895 | if (r > 0) { |
66ebf6c0 TH |
2896 | _cleanup_free_ char *val = NULL; |
2897 | uint64_t us; | |
5ad096b3 | 2898 | |
b734a4ff | 2899 | r = cg_get_keyed_attribute("cpu", u->cgroup_path, "cpu.stat", STRV_MAKE("usage_usec"), &val); |
b734a4ff LP |
2900 | if (IN_SET(r, -ENOENT, -ENXIO)) |
2901 | return -ENODATA; | |
d742f4b5 LP |
2902 | if (r < 0) |
2903 | return r; | |
66ebf6c0 TH |
2904 | |
2905 | r = safe_atou64(val, &us); | |
2906 | if (r < 0) | |
2907 | return r; | |
2908 | ||
2909 | ns = us * NSEC_PER_USEC; | |
2910 | } else { | |
66ebf6c0 TH |
2911 | r = cg_get_attribute("cpuacct", u->cgroup_path, "cpuacct.usage", &v); |
2912 | if (r == -ENOENT) | |
2913 | return -ENODATA; | |
2914 | if (r < 0) | |
2915 | return r; | |
2916 | ||
2917 | r = safe_atou64(v, &ns); | |
2918 | if (r < 0) | |
2919 | return r; | |
2920 | } | |
5ad096b3 LP |
2921 | |
2922 | *ret = ns; | |
2923 | return 0; | |
2924 | } | |
2925 | ||
2926 | int unit_get_cpu_usage(Unit *u, nsec_t *ret) { | |
2927 | nsec_t ns; | |
2928 | int r; | |
2929 | ||
fe700f46 LP |
2930 | assert(u); |
2931 | ||
2932 | /* Retrieve the current CPU usage counter. This will subtract the CPU counter taken when the unit was | |
2933 | * started. If the cgroup has been removed already, returns the last cached value. To cache the value, simply | |
2934 | * call this function with a NULL return value. */ | |
2935 | ||
2e4025c0 | 2936 | if (!UNIT_CGROUP_BOOL(u, cpu_accounting)) |
cf3b4be1 LP |
2937 | return -ENODATA; |
2938 | ||
5ad096b3 | 2939 | r = unit_get_cpu_usage_raw(u, &ns); |
fe700f46 LP |
2940 | if (r == -ENODATA && u->cpu_usage_last != NSEC_INFINITY) { |
2941 | /* If we can't get the CPU usage anymore (because the cgroup was already removed, for example), use our | |
2942 | * cached value. */ | |
2943 | ||
2944 | if (ret) | |
2945 | *ret = u->cpu_usage_last; | |
2946 | return 0; | |
2947 | } | |
5ad096b3 LP |
2948 | if (r < 0) |
2949 | return r; | |
2950 | ||
66ebf6c0 TH |
2951 | if (ns > u->cpu_usage_base) |
2952 | ns -= u->cpu_usage_base; | |
5ad096b3 LP |
2953 | else |
2954 | ns = 0; | |
2955 | ||
fe700f46 LP |
2956 | u->cpu_usage_last = ns; |
2957 | if (ret) | |
2958 | *ret = ns; | |
2959 | ||
5ad096b3 LP |
2960 | return 0; |
2961 | } | |
2962 | ||
906c06f6 DM |
2963 | int unit_get_ip_accounting( |
2964 | Unit *u, | |
2965 | CGroupIPAccountingMetric metric, | |
2966 | uint64_t *ret) { | |
2967 | ||
6b659ed8 | 2968 | uint64_t value; |
906c06f6 DM |
2969 | int fd, r; |
2970 | ||
2971 | assert(u); | |
2972 | assert(metric >= 0); | |
2973 | assert(metric < _CGROUP_IP_ACCOUNTING_METRIC_MAX); | |
2974 | assert(ret); | |
2975 | ||
2e4025c0 | 2976 | if (!UNIT_CGROUP_BOOL(u, ip_accounting)) |
cf3b4be1 LP |
2977 | return -ENODATA; |
2978 | ||
906c06f6 DM |
2979 | fd = IN_SET(metric, CGROUP_IP_INGRESS_BYTES, CGROUP_IP_INGRESS_PACKETS) ? |
2980 | u->ip_accounting_ingress_map_fd : | |
2981 | u->ip_accounting_egress_map_fd; | |
906c06f6 DM |
2982 | if (fd < 0) |
2983 | return -ENODATA; | |
2984 | ||
2985 | if (IN_SET(metric, CGROUP_IP_INGRESS_BYTES, CGROUP_IP_EGRESS_BYTES)) | |
6b659ed8 | 2986 | r = bpf_firewall_read_accounting(fd, &value, NULL); |
906c06f6 | 2987 | else |
6b659ed8 LP |
2988 | r = bpf_firewall_read_accounting(fd, NULL, &value); |
2989 | if (r < 0) | |
2990 | return r; | |
2991 | ||
2992 | /* Add in additional metrics from a previous runtime. Note that when reexecing/reloading the daemon we compile | |
2993 | * all BPF programs and maps anew, but serialize the old counters. When deserializing we store them in the | |
2994 | * ip_accounting_extra[] field, and add them in here transparently. */ | |
2995 | ||
2996 | *ret = value + u->ip_accounting_extra[metric]; | |
906c06f6 DM |
2997 | |
2998 | return r; | |
2999 | } | |
3000 | ||
3001 | int unit_reset_cpu_accounting(Unit *u) { | |
5ad096b3 LP |
3002 | nsec_t ns; |
3003 | int r; | |
3004 | ||
3005 | assert(u); | |
3006 | ||
fe700f46 LP |
3007 | u->cpu_usage_last = NSEC_INFINITY; |
3008 | ||
5ad096b3 LP |
3009 | r = unit_get_cpu_usage_raw(u, &ns); |
3010 | if (r < 0) { | |
66ebf6c0 | 3011 | u->cpu_usage_base = 0; |
5ad096b3 | 3012 | return r; |
b56c28c3 | 3013 | } |
2633eb83 | 3014 | |
66ebf6c0 | 3015 | u->cpu_usage_base = ns; |
4ad49000 | 3016 | return 0; |
4fbf50b3 LP |
3017 | } |
3018 | ||
906c06f6 DM |
3019 | int unit_reset_ip_accounting(Unit *u) { |
3020 | int r = 0, q = 0; | |
3021 | ||
3022 | assert(u); | |
3023 | ||
3024 | if (u->ip_accounting_ingress_map_fd >= 0) | |
3025 | r = bpf_firewall_reset_accounting(u->ip_accounting_ingress_map_fd); | |
3026 | ||
3027 | if (u->ip_accounting_egress_map_fd >= 0) | |
3028 | q = bpf_firewall_reset_accounting(u->ip_accounting_egress_map_fd); | |
3029 | ||
6b659ed8 LP |
3030 | zero(u->ip_accounting_extra); |
3031 | ||
906c06f6 DM |
3032 | return r < 0 ? r : q; |
3033 | } | |
3034 | ||
e7ab4d1a LP |
3035 | void unit_invalidate_cgroup(Unit *u, CGroupMask m) { |
3036 | assert(u); | |
3037 | ||
3038 | if (!UNIT_HAS_CGROUP_CONTEXT(u)) | |
3039 | return; | |
3040 | ||
3041 | if (m == 0) | |
3042 | return; | |
3043 | ||
538b4852 TH |
3044 | /* always invalidate compat pairs together */ |
3045 | if (m & (CGROUP_MASK_IO | CGROUP_MASK_BLKIO)) | |
3046 | m |= CGROUP_MASK_IO | CGROUP_MASK_BLKIO; | |
3047 | ||
7cce4fb7 LP |
3048 | if (m & (CGROUP_MASK_CPU | CGROUP_MASK_CPUACCT)) |
3049 | m |= CGROUP_MASK_CPU | CGROUP_MASK_CPUACCT; | |
3050 | ||
e00068e7 | 3051 | if (FLAGS_SET(u->cgroup_invalidated_mask, m)) /* NOP? */ |
e7ab4d1a LP |
3052 | return; |
3053 | ||
e00068e7 | 3054 | u->cgroup_invalidated_mask |= m; |
91a6073e | 3055 | unit_add_to_cgroup_realize_queue(u); |
e7ab4d1a LP |
3056 | } |
3057 | ||
906c06f6 DM |
3058 | void unit_invalidate_cgroup_bpf(Unit *u) { |
3059 | assert(u); | |
3060 | ||
3061 | if (!UNIT_HAS_CGROUP_CONTEXT(u)) | |
3062 | return; | |
3063 | ||
17f14955 | 3064 | if (u->cgroup_invalidated_mask & CGROUP_MASK_BPF_FIREWALL) /* NOP? */ |
906c06f6 DM |
3065 | return; |
3066 | ||
17f14955 | 3067 | u->cgroup_invalidated_mask |= CGROUP_MASK_BPF_FIREWALL; |
91a6073e | 3068 | unit_add_to_cgroup_realize_queue(u); |
906c06f6 DM |
3069 | |
3070 | /* If we are a slice unit, we also need to put compile a new BPF program for all our children, as the IP access | |
3071 | * list of our children includes our own. */ | |
3072 | if (u->type == UNIT_SLICE) { | |
3073 | Unit *member; | |
3074 | Iterator i; | |
eef85c4a | 3075 | void *v; |
906c06f6 | 3076 | |
eef85c4a | 3077 | HASHMAP_FOREACH_KEY(v, member, u->dependencies[UNIT_BEFORE], i) { |
cb5e3bc3 CD |
3078 | if (UNIT_DEREF(member->slice) == u) |
3079 | unit_invalidate_cgroup_bpf(member); | |
906c06f6 DM |
3080 | } |
3081 | } | |
3082 | } | |
3083 | ||
1d9cc876 LP |
3084 | bool unit_cgroup_delegate(Unit *u) { |
3085 | CGroupContext *c; | |
3086 | ||
3087 | assert(u); | |
3088 | ||
3089 | if (!UNIT_VTABLE(u)->can_delegate) | |
3090 | return false; | |
3091 | ||
3092 | c = unit_get_cgroup_context(u); | |
3093 | if (!c) | |
3094 | return false; | |
3095 | ||
3096 | return c->delegate; | |
3097 | } | |
3098 | ||
e7ab4d1a LP |
3099 | void manager_invalidate_startup_units(Manager *m) { |
3100 | Iterator i; | |
3101 | Unit *u; | |
3102 | ||
3103 | assert(m); | |
3104 | ||
3105 | SET_FOREACH(u, m->startup_units, i) | |
13c31542 | 3106 | unit_invalidate_cgroup(u, CGROUP_MASK_CPU|CGROUP_MASK_IO|CGROUP_MASK_BLKIO); |
e7ab4d1a LP |
3107 | } |
3108 | ||
4ad49000 LP |
3109 | static const char* const cgroup_device_policy_table[_CGROUP_DEVICE_POLICY_MAX] = { |
3110 | [CGROUP_AUTO] = "auto", | |
3111 | [CGROUP_CLOSED] = "closed", | |
3112 | [CGROUP_STRICT] = "strict", | |
3113 | }; | |
4fbf50b3 | 3114 | |
4ad49000 | 3115 | DEFINE_STRING_TABLE_LOOKUP(cgroup_device_policy, CGroupDevicePolicy); |