]>
Commit | Line | Data |
---|---|---|
db9ecf05 | 1 | /* SPDX-License-Identifier: LGPL-2.1-or-later */ |
8e274523 | 2 | |
c6c18be3 | 3 | #include <fcntl.h> |
8c6db833 | 4 | |
afcfaa69 LP |
5 | #include "sd-messages.h" |
6 | ||
a4817536 | 7 | #include "af-list.h" |
b5efdb8a | 8 | #include "alloc-util.h" |
18c528e9 | 9 | #include "blockdev-util.h" |
d8b4d14d | 10 | #include "bpf-devices.h" |
906c06f6 | 11 | #include "bpf-firewall.h" |
506ea51b | 12 | #include "bpf-foreign.h" |
cd09a5f3 | 13 | #include "bpf-socket-bind.h" |
45c2e068 | 14 | #include "btrfs-util.h" |
6592b975 | 15 | #include "bus-error.h" |
78fa2f91 | 16 | #include "bus-locator.h" |
fdb3deca | 17 | #include "cgroup-setup.h" |
03a7b521 | 18 | #include "cgroup-util.h" |
3ffd4af2 | 19 | #include "cgroup.h" |
7176f06c | 20 | #include "devnum-util.h" |
3ffd4af2 | 21 | #include "fd-util.h" |
0d39fa9c | 22 | #include "fileio.h" |
84ebe6f0 | 23 | #include "in-addr-prefix-util.h" |
9e5fd717 | 24 | #include "inotify-util.h" |
d9e45bc3 | 25 | #include "io-util.h" |
5587ce7f | 26 | #include "ip-protocol-list.h" |
3a0f06c4 | 27 | #include "limits-util.h" |
d9e45bc3 | 28 | #include "nulstr-util.h" |
6bedfcbb | 29 | #include "parse-util.h" |
9eb977db | 30 | #include "path-util.h" |
1ead0b2a | 31 | #include "percent-util.h" |
03a7b521 | 32 | #include "process-util.h" |
c36a69f4 | 33 | #include "procfs-util.h" |
6f50d4f7 | 34 | #include "restrict-ifaces.h" |
9444b1f2 | 35 | #include "special.h" |
906c06f6 | 36 | #include "stdio-util.h" |
8b43440b | 37 | #include "string-table.h" |
07630cea | 38 | #include "string-util.h" |
cc6271f1 | 39 | #include "virt.h" |
8e274523 | 40 | |
b1994387 ILG |
41 | #if BPF_FRAMEWORK |
42 | #include "bpf-dlopen.h" | |
43 | #include "bpf-link.h" | |
44 | #include "bpf/restrict_fs/restrict-fs-skel.h" | |
45 | #endif | |
46 | ||
10f28641 | 47 | #define CGROUP_CPU_QUOTA_DEFAULT_PERIOD_USEC ((usec_t) 100 * USEC_PER_MSEC) |
9a054909 | 48 | |
39b9fefb LP |
49 | /* Returns the log level to use when cgroup attribute writes fail. When an attribute is missing or we have access |
50 | * problems we downgrade to LOG_DEBUG. This is supposed to be nice to container managers and kernels which want to mask | |
51 | * out specific attributes from us. */ | |
52 | #define LOG_LEVEL_CGROUP_WRITE(r) (IN_SET(abs(r), ENOENT, EROFS, EACCES, EPERM) ? LOG_DEBUG : LOG_WARNING) | |
53 | ||
3a0f06c4 ZJS |
54 | uint64_t tasks_max_resolve(const TasksMax *tasks_max) { |
55 | if (tasks_max->scale == 0) | |
56 | return tasks_max->value; | |
57 | ||
58 | return system_tasks_max_scale(tasks_max->value, tasks_max->scale); | |
59 | } | |
60 | ||
611c4f8a | 61 | bool manager_owns_host_root_cgroup(Manager *m) { |
cc6271f1 LP |
62 | assert(m); |
63 | ||
64 | /* Returns true if we are managing the root cgroup. Note that it isn't sufficient to just check whether the | |
65 | * group root path equals "/" since that will also be the case if CLONE_NEWCGROUP is in the mix. Since there's | |
66 | * appears to be no nice way to detect whether we are in a CLONE_NEWCGROUP namespace we instead just check if | |
67 | * we run in any kind of container virtualization. */ | |
68 | ||
28cfdc5a LP |
69 | if (MANAGER_IS_USER(m)) |
70 | return false; | |
71 | ||
cc6271f1 LP |
72 | if (detect_container() > 0) |
73 | return false; | |
74 | ||
57ea45e1 | 75 | return empty_or_root(m->cgroup_root); |
cc6271f1 LP |
76 | } |
77 | ||
9dfb6a3a PM |
78 | bool unit_has_startup_cgroup_constraints(Unit *u) { |
79 | assert(u); | |
80 | ||
81 | /* Returns true if this unit has any directives which apply during | |
82 | * startup/shutdown phases. */ | |
83 | ||
84 | CGroupContext *c; | |
85 | ||
86 | c = unit_get_cgroup_context(u); | |
87 | if (!c) | |
88 | return false; | |
89 | ||
90 | return c->startup_cpu_shares != CGROUP_CPU_SHARES_INVALID || | |
91 | c->startup_io_weight != CGROUP_WEIGHT_INVALID || | |
92 | c->startup_blockio_weight != CGROUP_BLKIO_WEIGHT_INVALID || | |
93 | c->startup_cpuset_cpus.set || | |
53fda560 LB |
94 | c->startup_cpuset_mems.set || |
95 | c->startup_memory_high_set || | |
96 | c->startup_memory_max_set || | |
97 | c->startup_memory_swap_max_set|| | |
98 | c->startup_memory_zswap_max_set || | |
99 | c->startup_memory_low_set; | |
9dfb6a3a PM |
100 | } |
101 | ||
611c4f8a | 102 | bool unit_has_host_root_cgroup(Unit *u) { |
f3725e64 LP |
103 | assert(u); |
104 | ||
cc6271f1 LP |
105 | /* Returns whether this unit manages the root cgroup. This will return true if this unit is the root slice and |
106 | * the manager manages the root cgroup. */ | |
f3725e64 | 107 | |
611c4f8a | 108 | if (!manager_owns_host_root_cgroup(u->manager)) |
f3725e64 LP |
109 | return false; |
110 | ||
cc6271f1 | 111 | return unit_has_name(u, SPECIAL_ROOT_SLICE); |
f3725e64 LP |
112 | } |
113 | ||
293d32df LP |
114 | static int set_attribute_and_warn(Unit *u, const char *controller, const char *attribute, const char *value) { |
115 | int r; | |
116 | ||
117 | r = cg_set_attribute(controller, u->cgroup_path, attribute, value); | |
118 | if (r < 0) | |
8ed6f81b | 119 | log_unit_full_errno(u, LOG_LEVEL_CGROUP_WRITE(r), r, "Failed to set '%s' attribute on '%s' to '%.*s': %m", |
6178e2f8 | 120 | strna(attribute), empty_to_root(u->cgroup_path), (int) strcspn(value, NEWLINE), value); |
293d32df LP |
121 | |
122 | return r; | |
123 | } | |
124 | ||
2b40998d | 125 | static void cgroup_compat_warn(void) { |
128fadc9 TH |
126 | static bool cgroup_compat_warned = false; |
127 | ||
128 | if (cgroup_compat_warned) | |
129 | return; | |
130 | ||
cc6271f1 LP |
131 | log_warning("cgroup compatibility translation between legacy and unified hierarchy settings activated. " |
132 | "See cgroup-compat debug messages for details."); | |
133 | ||
128fadc9 TH |
134 | cgroup_compat_warned = true; |
135 | } | |
136 | ||
137 | #define log_cgroup_compat(unit, fmt, ...) do { \ | |
138 | cgroup_compat_warn(); \ | |
139 | log_unit_debug(unit, "cgroup-compat: " fmt, ##__VA_ARGS__); \ | |
2b40998d | 140 | } while (false) |
128fadc9 | 141 | |
4ad49000 LP |
142 | void cgroup_context_init(CGroupContext *c) { |
143 | assert(c); | |
144 | ||
de8a711a | 145 | /* Initialize everything to the kernel defaults. */ |
4ad49000 | 146 | |
de8a711a LP |
147 | *c = (CGroupContext) { |
148 | .cpu_weight = CGROUP_WEIGHT_INVALID, | |
149 | .startup_cpu_weight = CGROUP_WEIGHT_INVALID, | |
150 | .cpu_quota_per_sec_usec = USEC_INFINITY, | |
10f28641 | 151 | .cpu_quota_period_usec = USEC_INFINITY, |
66ebf6c0 | 152 | |
de8a711a LP |
153 | .cpu_shares = CGROUP_CPU_SHARES_INVALID, |
154 | .startup_cpu_shares = CGROUP_CPU_SHARES_INVALID, | |
d53d9474 | 155 | |
de8a711a | 156 | .memory_high = CGROUP_LIMIT_MAX, |
53fda560 | 157 | .startup_memory_high = CGROUP_LIMIT_MAX, |
de8a711a | 158 | .memory_max = CGROUP_LIMIT_MAX, |
53fda560 | 159 | .startup_memory_max = CGROUP_LIMIT_MAX, |
de8a711a | 160 | .memory_swap_max = CGROUP_LIMIT_MAX, |
53fda560 | 161 | .startup_memory_swap_max = CGROUP_LIMIT_MAX, |
d7fe0a67 | 162 | .memory_zswap_max = CGROUP_LIMIT_MAX, |
53fda560 | 163 | .startup_memory_zswap_max = CGROUP_LIMIT_MAX, |
da4d897e | 164 | |
de8a711a | 165 | .memory_limit = CGROUP_LIMIT_MAX, |
b2f8b02e | 166 | |
de8a711a LP |
167 | .io_weight = CGROUP_WEIGHT_INVALID, |
168 | .startup_io_weight = CGROUP_WEIGHT_INVALID, | |
13c31542 | 169 | |
de8a711a LP |
170 | .blockio_weight = CGROUP_BLKIO_WEIGHT_INVALID, |
171 | .startup_blockio_weight = CGROUP_BLKIO_WEIGHT_INVALID, | |
d53d9474 | 172 | |
3a0f06c4 | 173 | .tasks_max = TASKS_MAX_UNSET, |
4d824a4e AZ |
174 | |
175 | .moom_swap = MANAGED_OOM_AUTO, | |
176 | .moom_mem_pressure = MANAGED_OOM_AUTO, | |
4e806bfa | 177 | .moom_preference = MANAGED_OOM_PREFERENCE_NONE, |
6bb00842 LP |
178 | |
179 | .memory_pressure_watch = _CGROUP_PRESSURE_WATCH_INVALID, | |
180 | .memory_pressure_threshold_usec = USEC_INFINITY, | |
de8a711a | 181 | }; |
4ad49000 | 182 | } |
8e274523 | 183 | |
4ad49000 LP |
184 | void cgroup_context_free_device_allow(CGroupContext *c, CGroupDeviceAllow *a) { |
185 | assert(c); | |
186 | assert(a); | |
187 | ||
71fda00f | 188 | LIST_REMOVE(device_allow, c->device_allow, a); |
4ad49000 LP |
189 | free(a->path); |
190 | free(a); | |
191 | } | |
192 | ||
13c31542 TH |
193 | void cgroup_context_free_io_device_weight(CGroupContext *c, CGroupIODeviceWeight *w) { |
194 | assert(c); | |
195 | assert(w); | |
196 | ||
197 | LIST_REMOVE(device_weights, c->io_device_weights, w); | |
198 | free(w->path); | |
199 | free(w); | |
200 | } | |
201 | ||
6ae4283c TH |
202 | void cgroup_context_free_io_device_latency(CGroupContext *c, CGroupIODeviceLatency *l) { |
203 | assert(c); | |
204 | assert(l); | |
205 | ||
206 | LIST_REMOVE(device_latencies, c->io_device_latencies, l); | |
207 | free(l->path); | |
208 | free(l); | |
209 | } | |
210 | ||
13c31542 TH |
211 | void cgroup_context_free_io_device_limit(CGroupContext *c, CGroupIODeviceLimit *l) { |
212 | assert(c); | |
213 | assert(l); | |
214 | ||
215 | LIST_REMOVE(device_limits, c->io_device_limits, l); | |
216 | free(l->path); | |
217 | free(l); | |
218 | } | |
219 | ||
4ad49000 LP |
220 | void cgroup_context_free_blockio_device_weight(CGroupContext *c, CGroupBlockIODeviceWeight *w) { |
221 | assert(c); | |
222 | assert(w); | |
223 | ||
71fda00f | 224 | LIST_REMOVE(device_weights, c->blockio_device_weights, w); |
4ad49000 LP |
225 | free(w->path); |
226 | free(w); | |
227 | } | |
228 | ||
229 | void cgroup_context_free_blockio_device_bandwidth(CGroupContext *c, CGroupBlockIODeviceBandwidth *b) { | |
230 | assert(c); | |
8e274523 | 231 | assert(b); |
8e274523 | 232 | |
71fda00f | 233 | LIST_REMOVE(device_bandwidths, c->blockio_device_bandwidths, b); |
4ad49000 LP |
234 | free(b->path); |
235 | free(b); | |
236 | } | |
237 | ||
b894ef1b JK |
238 | void cgroup_context_remove_bpf_foreign_program(CGroupContext *c, CGroupBPFForeignProgram *p) { |
239 | assert(c); | |
240 | assert(p); | |
241 | ||
242 | LIST_REMOVE(programs, c->bpf_foreign_programs, p); | |
243 | free(p->bpffs_path); | |
244 | free(p); | |
245 | } | |
246 | ||
b18e9fc1 | 247 | void cgroup_context_remove_socket_bind(CGroupSocketBindItem **head) { |
b18e9fc1 JK |
248 | assert(head); |
249 | ||
250 | while (*head) { | |
a67abc49 | 251 | CGroupSocketBindItem *h = *head; |
b18e9fc1 JK |
252 | LIST_REMOVE(socket_bind_items, *head, h); |
253 | free(h); | |
254 | } | |
255 | } | |
256 | ||
4ad49000 LP |
257 | void cgroup_context_done(CGroupContext *c) { |
258 | assert(c); | |
259 | ||
13c31542 TH |
260 | while (c->io_device_weights) |
261 | cgroup_context_free_io_device_weight(c, c->io_device_weights); | |
262 | ||
6ae4283c TH |
263 | while (c->io_device_latencies) |
264 | cgroup_context_free_io_device_latency(c, c->io_device_latencies); | |
265 | ||
13c31542 TH |
266 | while (c->io_device_limits) |
267 | cgroup_context_free_io_device_limit(c, c->io_device_limits); | |
268 | ||
4ad49000 LP |
269 | while (c->blockio_device_weights) |
270 | cgroup_context_free_blockio_device_weight(c, c->blockio_device_weights); | |
271 | ||
272 | while (c->blockio_device_bandwidths) | |
273 | cgroup_context_free_blockio_device_bandwidth(c, c->blockio_device_bandwidths); | |
274 | ||
275 | while (c->device_allow) | |
276 | cgroup_context_free_device_allow(c, c->device_allow); | |
6a48d82f | 277 | |
b18e9fc1 JK |
278 | cgroup_context_remove_socket_bind(&c->socket_bind_allow); |
279 | cgroup_context_remove_socket_bind(&c->socket_bind_deny); | |
280 | ||
84ebe6f0 YW |
281 | c->ip_address_allow = set_free(c->ip_address_allow); |
282 | c->ip_address_deny = set_free(c->ip_address_deny); | |
fab34748 KL |
283 | |
284 | c->ip_filters_ingress = strv_free(c->ip_filters_ingress); | |
285 | c->ip_filters_egress = strv_free(c->ip_filters_egress); | |
047f5d63 | 286 | |
b894ef1b JK |
287 | while (c->bpf_foreign_programs) |
288 | cgroup_context_remove_bpf_foreign_program(c, c->bpf_foreign_programs); | |
289 | ||
6f50d4f7 MV |
290 | c->restrict_network_interfaces = set_free(c->restrict_network_interfaces); |
291 | ||
047f5d63 | 292 | cpu_set_reset(&c->cpuset_cpus); |
31d3a520 | 293 | cpu_set_reset(&c->startup_cpuset_cpus); |
047f5d63 | 294 | cpu_set_reset(&c->cpuset_mems); |
31d3a520 | 295 | cpu_set_reset(&c->startup_cpuset_mems); |
a8b993dc LP |
296 | |
297 | c->delegate_subgroup = mfree(c->delegate_subgroup); | |
4ad49000 LP |
298 | } |
299 | ||
74b5fb27 | 300 | static int unit_get_kernel_memory_limit(Unit *u, const char *file, uint64_t *ret) { |
74b5fb27 CD |
301 | assert(u); |
302 | ||
303 | if (!u->cgroup_realized) | |
304 | return -EOWNERDEAD; | |
305 | ||
613328c3 | 306 | return cg_get_attribute_as_uint64("memory", u->cgroup_path, file, ret); |
74b5fb27 CD |
307 | } |
308 | ||
309 | static int unit_compare_memory_limit(Unit *u, const char *property_name, uint64_t *ret_unit_value, uint64_t *ret_kernel_value) { | |
310 | CGroupContext *c; | |
311 | CGroupMask m; | |
312 | const char *file; | |
313 | uint64_t unit_value; | |
314 | int r; | |
315 | ||
316 | /* Compare kernel memcg configuration against our internal systemd state. Unsupported (and will | |
317 | * return -ENODATA) on cgroup v1. | |
318 | * | |
319 | * Returns: | |
320 | * | |
321 | * <0: On error. | |
322 | * 0: If the kernel memory setting doesn't match our configuration. | |
323 | * >0: If the kernel memory setting matches our configuration. | |
324 | * | |
325 | * The following values are only guaranteed to be populated on return >=0: | |
326 | * | |
327 | * - ret_unit_value will contain our internal expected value for the unit, page-aligned. | |
328 | * - ret_kernel_value will contain the actual value presented by the kernel. */ | |
329 | ||
330 | assert(u); | |
331 | ||
332 | r = cg_all_unified(); | |
333 | if (r < 0) | |
334 | return log_debug_errno(r, "Failed to determine cgroup hierarchy version: %m"); | |
335 | ||
336 | /* Unsupported on v1. | |
337 | * | |
338 | * We don't return ENOENT, since that could actually mask a genuine problem where somebody else has | |
339 | * silently masked the controller. */ | |
340 | if (r == 0) | |
341 | return -ENODATA; | |
342 | ||
343 | /* The root slice doesn't have any controller files, so we can't compare anything. */ | |
344 | if (unit_has_name(u, SPECIAL_ROOT_SLICE)) | |
345 | return -ENODATA; | |
346 | ||
347 | /* It's possible to have MemoryFoo set without systemd wanting to have the memory controller enabled, | |
348 | * for example, in the case of DisableControllers= or cgroup_disable on the kernel command line. To | |
349 | * avoid specious errors in these scenarios, check that we even expect the memory controller to be | |
350 | * enabled at all. */ | |
351 | m = unit_get_target_mask(u); | |
352 | if (!FLAGS_SET(m, CGROUP_MASK_MEMORY)) | |
353 | return -ENODATA; | |
354 | ||
806a9362 | 355 | assert_se(c = unit_get_cgroup_context(u)); |
74b5fb27 | 356 | |
53fda560 LB |
357 | bool startup = u->manager && IN_SET(manager_state(u->manager), MANAGER_STARTING, MANAGER_INITIALIZING, MANAGER_STOPPING); |
358 | ||
74b5fb27 CD |
359 | if (streq(property_name, "MemoryLow")) { |
360 | unit_value = unit_get_ancestor_memory_low(u); | |
361 | file = "memory.low"; | |
53fda560 LB |
362 | } else if (startup && streq(property_name, "StartupMemoryLow")) { |
363 | unit_value = unit_get_ancestor_startup_memory_low(u); | |
364 | file = "memory.low"; | |
74b5fb27 CD |
365 | } else if (streq(property_name, "MemoryMin")) { |
366 | unit_value = unit_get_ancestor_memory_min(u); | |
367 | file = "memory.min"; | |
368 | } else if (streq(property_name, "MemoryHigh")) { | |
369 | unit_value = c->memory_high; | |
370 | file = "memory.high"; | |
53fda560 LB |
371 | } else if (startup && streq(property_name, "StartupMemoryHigh")) { |
372 | unit_value = c->startup_memory_high; | |
373 | file = "memory.high"; | |
74b5fb27 CD |
374 | } else if (streq(property_name, "MemoryMax")) { |
375 | unit_value = c->memory_max; | |
376 | file = "memory.max"; | |
53fda560 LB |
377 | } else if (startup && streq(property_name, "StartupMemoryMax")) { |
378 | unit_value = c->startup_memory_max; | |
379 | file = "memory.max"; | |
74b5fb27 CD |
380 | } else if (streq(property_name, "MemorySwapMax")) { |
381 | unit_value = c->memory_swap_max; | |
382 | file = "memory.swap.max"; | |
53fda560 LB |
383 | } else if (startup && streq(property_name, "StartupMemorySwapMax")) { |
384 | unit_value = c->startup_memory_swap_max; | |
385 | file = "memory.swap.max"; | |
d7fe0a67 PV |
386 | } else if (streq(property_name, "MemoryZSwapMax")) { |
387 | unit_value = c->memory_zswap_max; | |
388 | file = "memory.zswap.max"; | |
53fda560 LB |
389 | } else if (startup && streq(property_name, "StartupMemoryZSwapMax")) { |
390 | unit_value = c->startup_memory_zswap_max; | |
391 | file = "memory.zswap.max"; | |
74b5fb27 CD |
392 | } else |
393 | return -EINVAL; | |
394 | ||
395 | r = unit_get_kernel_memory_limit(u, file, ret_kernel_value); | |
396 | if (r < 0) | |
397 | return log_unit_debug_errno(u, r, "Failed to parse %s: %m", file); | |
398 | ||
399 | /* It's intended (soon) in a future kernel to not expose cgroup memory limits rounded to page | |
400 | * boundaries, but instead separate the user-exposed limit, which is whatever userspace told us, from | |
401 | * our internal page-counting. To support those future kernels, just check the value itself first | |
402 | * without any page-alignment. */ | |
403 | if (*ret_kernel_value == unit_value) { | |
404 | *ret_unit_value = unit_value; | |
405 | return 1; | |
406 | } | |
407 | ||
408 | /* The current kernel behaviour, by comparison, is that even if you write a particular number of | |
409 | * bytes into a cgroup memory file, it always returns that number page-aligned down (since the kernel | |
410 | * internally stores cgroup limits in pages). As such, so long as it aligns properly, everything is | |
411 | * cricket. */ | |
412 | if (unit_value != CGROUP_LIMIT_MAX) | |
413 | unit_value = PAGE_ALIGN_DOWN(unit_value); | |
414 | ||
415 | *ret_unit_value = unit_value; | |
416 | ||
417 | return *ret_kernel_value == *ret_unit_value; | |
418 | } | |
419 | ||
bc0623df CD |
420 | #define FORMAT_CGROUP_DIFF_MAX 128 |
421 | ||
422 | static char *format_cgroup_memory_limit_comparison(char *buf, size_t l, Unit *u, const char *property_name) { | |
423 | uint64_t kval, sval; | |
424 | int r; | |
425 | ||
426 | assert(u); | |
427 | assert(buf); | |
428 | assert(l > 0); | |
429 | ||
430 | r = unit_compare_memory_limit(u, property_name, &sval, &kval); | |
431 | ||
432 | /* memory.swap.max is special in that it relies on CONFIG_MEMCG_SWAP (and the default swapaccount=1). | |
433 | * In the absence of reliably being able to detect whether memcg swap support is available or not, | |
d7fe0a67 PV |
434 | * only complain if the error is not ENOENT. This is similarly the case for memory.zswap.max relying |
435 | * on CONFIG_ZSWAP. */ | |
bc0623df | 436 | if (r > 0 || IN_SET(r, -ENODATA, -EOWNERDEAD) || |
53fda560 LB |
437 | (r == -ENOENT && STR_IN_SET(property_name, |
438 | "MemorySwapMax", | |
439 | "StartupMemorySwapMax", | |
440 | "MemoryZSwapMax", | |
441 | "StartupMemoryZSwapMax"))) | |
bc0623df | 442 | buf[0] = 0; |
38553034 ZJS |
443 | else if (r < 0) { |
444 | errno = -r; | |
445 | (void) snprintf(buf, l, " (error getting kernel value: %m)"); | |
446 | } else | |
447 | (void) snprintf(buf, l, " (different value in kernel: %" PRIu64 ")", kval); | |
bc0623df CD |
448 | |
449 | return buf; | |
450 | } | |
451 | ||
452 | void cgroup_context_dump(Unit *u, FILE* f, const char *prefix) { | |
7b3693e4 | 453 | _cleanup_free_ char *disable_controllers_str = NULL, *delegate_controllers_str = NULL, *cpuset_cpus = NULL, *cpuset_mems = NULL, *startup_cpuset_cpus = NULL, *startup_cpuset_mems = NULL; |
bc0623df | 454 | CGroupContext *c; |
84ebe6f0 | 455 | struct in_addr_prefix *iaai; |
4ad49000 | 456 | |
bc0623df CD |
457 | char cda[FORMAT_CGROUP_DIFF_MAX]; |
458 | char cdb[FORMAT_CGROUP_DIFF_MAX]; | |
459 | char cdc[FORMAT_CGROUP_DIFF_MAX]; | |
460 | char cdd[FORMAT_CGROUP_DIFF_MAX]; | |
461 | char cde[FORMAT_CGROUP_DIFF_MAX]; | |
53fda560 LB |
462 | char cdf[FORMAT_CGROUP_DIFF_MAX]; |
463 | char cdg[FORMAT_CGROUP_DIFF_MAX]; | |
464 | char cdh[FORMAT_CGROUP_DIFF_MAX]; | |
465 | char cdi[FORMAT_CGROUP_DIFF_MAX]; | |
466 | char cdj[FORMAT_CGROUP_DIFF_MAX]; | |
467 | char cdk[FORMAT_CGROUP_DIFF_MAX]; | |
bc0623df CD |
468 | |
469 | assert(u); | |
4ad49000 LP |
470 | assert(f); |
471 | ||
806a9362 | 472 | assert_se(c = unit_get_cgroup_context(u)); |
bc0623df | 473 | |
4ad49000 LP |
474 | prefix = strempty(prefix); |
475 | ||
25cc30c4 | 476 | (void) cg_mask_to_string(c->disable_controllers, &disable_controllers_str); |
7b3693e4 | 477 | (void) cg_mask_to_string(c->delegate_controllers, &delegate_controllers_str); |
25cc30c4 | 478 | |
af2b151b ZJS |
479 | /* "Delegate=" means "yes, but no controllers". Show this as "(none)". */ |
480 | const char *delegate_str = delegate_controllers_str ?: c->delegate ? "(none)" : "no"; | |
481 | ||
047f5d63 | 482 | cpuset_cpus = cpu_set_to_range_string(&c->cpuset_cpus); |
31d3a520 | 483 | startup_cpuset_cpus = cpu_set_to_range_string(&c->startup_cpuset_cpus); |
047f5d63 | 484 | cpuset_mems = cpu_set_to_range_string(&c->cpuset_mems); |
31d3a520 | 485 | startup_cpuset_mems = cpu_set_to_range_string(&c->startup_cpuset_mems); |
047f5d63 | 486 | |
4ad49000 | 487 | fprintf(f, |
6dfb9282 CD |
488 | "%sCPUAccounting: %s\n" |
489 | "%sIOAccounting: %s\n" | |
490 | "%sBlockIOAccounting: %s\n" | |
491 | "%sMemoryAccounting: %s\n" | |
492 | "%sTasksAccounting: %s\n" | |
493 | "%sIPAccounting: %s\n" | |
494 | "%sCPUWeight: %" PRIu64 "\n" | |
495 | "%sStartupCPUWeight: %" PRIu64 "\n" | |
496 | "%sCPUShares: %" PRIu64 "\n" | |
497 | "%sStartupCPUShares: %" PRIu64 "\n" | |
498 | "%sCPUQuotaPerSecSec: %s\n" | |
499 | "%sCPUQuotaPeriodSec: %s\n" | |
500 | "%sAllowedCPUs: %s\n" | |
31d3a520 | 501 | "%sStartupAllowedCPUs: %s\n" |
6dfb9282 | 502 | "%sAllowedMemoryNodes: %s\n" |
31d3a520 | 503 | "%sStartupAllowedMemoryNodes: %s\n" |
6dfb9282 CD |
504 | "%sIOWeight: %" PRIu64 "\n" |
505 | "%sStartupIOWeight: %" PRIu64 "\n" | |
506 | "%sBlockIOWeight: %" PRIu64 "\n" | |
507 | "%sStartupBlockIOWeight: %" PRIu64 "\n" | |
508 | "%sDefaultMemoryMin: %" PRIu64 "\n" | |
509 | "%sDefaultMemoryLow: %" PRIu64 "\n" | |
bc0623df CD |
510 | "%sMemoryMin: %" PRIu64 "%s\n" |
511 | "%sMemoryLow: %" PRIu64 "%s\n" | |
53fda560 | 512 | "%sStartupMemoryLow: %" PRIu64 "%s\n" |
bc0623df | 513 | "%sMemoryHigh: %" PRIu64 "%s\n" |
53fda560 | 514 | "%sStartupMemoryHigh: %" PRIu64 "%s\n" |
bc0623df | 515 | "%sMemoryMax: %" PRIu64 "%s\n" |
53fda560 | 516 | "%sStartupMemoryMax: %" PRIu64 "%s\n" |
bc0623df | 517 | "%sMemorySwapMax: %" PRIu64 "%s\n" |
53fda560 | 518 | "%sStartupMemorySwapMax: %" PRIu64 "%s\n" |
d7fe0a67 | 519 | "%sMemoryZSwapMax: %" PRIu64 "%s\n" |
53fda560 | 520 | "%sStartupMemoryZSwapMax: %" PRIu64 "%s\n" |
6dfb9282 CD |
521 | "%sMemoryLimit: %" PRIu64 "\n" |
522 | "%sTasksMax: %" PRIu64 "\n" | |
523 | "%sDevicePolicy: %s\n" | |
524 | "%sDisableControllers: %s\n" | |
4d824a4e AZ |
525 | "%sDelegate: %s\n" |
526 | "%sManagedOOMSwap: %s\n" | |
527 | "%sManagedOOMMemoryPressure: %s\n" | |
d9d3f05d | 528 | "%sManagedOOMMemoryPressureLimit: " PERMYRIAD_AS_PERCENT_FORMAT_STR "\n" |
6bb00842 LP |
529 | "%sManagedOOMPreference: %s\n" |
530 | "%sMemoryPressureWatch: %s\n", | |
4ad49000 | 531 | prefix, yes_no(c->cpu_accounting), |
13c31542 | 532 | prefix, yes_no(c->io_accounting), |
4ad49000 LP |
533 | prefix, yes_no(c->blockio_accounting), |
534 | prefix, yes_no(c->memory_accounting), | |
d53d9474 | 535 | prefix, yes_no(c->tasks_accounting), |
c21c9906 | 536 | prefix, yes_no(c->ip_accounting), |
66ebf6c0 TH |
537 | prefix, c->cpu_weight, |
538 | prefix, c->startup_cpu_weight, | |
4ad49000 | 539 | prefix, c->cpu_shares, |
95ae05c0 | 540 | prefix, c->startup_cpu_shares, |
5291f26d ZJS |
541 | prefix, FORMAT_TIMESPAN(c->cpu_quota_per_sec_usec, 1), |
542 | prefix, FORMAT_TIMESPAN(c->cpu_quota_period_usec, 1), | |
85c3b278 | 543 | prefix, strempty(cpuset_cpus), |
31d3a520 | 544 | prefix, strempty(startup_cpuset_cpus), |
85c3b278 | 545 | prefix, strempty(cpuset_mems), |
31d3a520 | 546 | prefix, strempty(startup_cpuset_mems), |
13c31542 TH |
547 | prefix, c->io_weight, |
548 | prefix, c->startup_io_weight, | |
4ad49000 | 549 | prefix, c->blockio_weight, |
95ae05c0 | 550 | prefix, c->startup_blockio_weight, |
7ad5439e | 551 | prefix, c->default_memory_min, |
c52db42b | 552 | prefix, c->default_memory_low, |
bc0623df CD |
553 | prefix, c->memory_min, format_cgroup_memory_limit_comparison(cda, sizeof(cda), u, "MemoryMin"), |
554 | prefix, c->memory_low, format_cgroup_memory_limit_comparison(cdb, sizeof(cdb), u, "MemoryLow"), | |
53fda560 LB |
555 | prefix, c->startup_memory_low, format_cgroup_memory_limit_comparison(cdc, sizeof(cdc), u, "StartupMemoryLow"), |
556 | prefix, c->memory_high, format_cgroup_memory_limit_comparison(cdd, sizeof(cdd), u, "MemoryHigh"), | |
557 | prefix, c->startup_memory_high, format_cgroup_memory_limit_comparison(cde, sizeof(cde), u, "StartupMemoryHigh"), | |
558 | prefix, c->memory_max, format_cgroup_memory_limit_comparison(cdf, sizeof(cdf), u, "MemoryMax"), | |
559 | prefix, c->startup_memory_max, format_cgroup_memory_limit_comparison(cdg, sizeof(cdg), u, "StartupMemoryMax"), | |
560 | prefix, c->memory_swap_max, format_cgroup_memory_limit_comparison(cdh, sizeof(cdh), u, "MemorySwapMax"), | |
561 | prefix, c->startup_memory_swap_max, format_cgroup_memory_limit_comparison(cdi, sizeof(cdi), u, "StartupMemorySwapMax"), | |
562 | prefix, c->memory_zswap_max, format_cgroup_memory_limit_comparison(cdj, sizeof(cdj), u, "MemoryZSwapMax"), | |
563 | prefix, c->startup_memory_zswap_max, format_cgroup_memory_limit_comparison(cdk, sizeof(cdk), u, "StartupMemoryZSwapMax"), | |
4ad49000 | 564 | prefix, c->memory_limit, |
3a0f06c4 | 565 | prefix, tasks_max_resolve(&c->tasks_max), |
a931ad47 | 566 | prefix, cgroup_device_policy_to_string(c->device_policy), |
f4c43a81 | 567 | prefix, strempty(disable_controllers_str), |
af2b151b | 568 | prefix, delegate_str, |
4d824a4e AZ |
569 | prefix, managed_oom_mode_to_string(c->moom_swap), |
570 | prefix, managed_oom_mode_to_string(c->moom_mem_pressure), | |
d9d3f05d | 571 | prefix, PERMYRIAD_AS_PERCENT_FORMAT_VAL(UINT32_SCALE_TO_PERMYRIAD(c->moom_mem_pressure_limit)), |
6bb00842 LP |
572 | prefix, managed_oom_preference_to_string(c->moom_preference), |
573 | prefix, cgroup_pressure_watch_to_string(c->memory_pressure_watch)); | |
574 | ||
a8b993dc LP |
575 | if (c->delegate_subgroup) |
576 | fprintf(f, "%sDelegateSubgroup: %s\n", | |
577 | prefix, c->delegate_subgroup); | |
578 | ||
6bb00842 LP |
579 | if (c->memory_pressure_threshold_usec != USEC_INFINITY) |
580 | fprintf(f, "%sMemoryPressureThresholdSec: %s\n", | |
581 | prefix, FORMAT_TIMESPAN(c->memory_pressure_threshold_usec, 1)); | |
4ad49000 LP |
582 | |
583 | LIST_FOREACH(device_allow, a, c->device_allow) | |
584 | fprintf(f, | |
6dfb9282 | 585 | "%sDeviceAllow: %s %s%s%s\n", |
4ad49000 LP |
586 | prefix, |
587 | a->path, | |
588 | a->r ? "r" : "", a->w ? "w" : "", a->m ? "m" : ""); | |
589 | ||
13c31542 TH |
590 | LIST_FOREACH(device_weights, iw, c->io_device_weights) |
591 | fprintf(f, | |
6dfb9282 | 592 | "%sIODeviceWeight: %s %" PRIu64 "\n", |
13c31542 TH |
593 | prefix, |
594 | iw->path, | |
595 | iw->weight); | |
596 | ||
6ae4283c TH |
597 | LIST_FOREACH(device_latencies, l, c->io_device_latencies) |
598 | fprintf(f, | |
6dfb9282 | 599 | "%sIODeviceLatencyTargetSec: %s %s\n", |
6ae4283c TH |
600 | prefix, |
601 | l->path, | |
5291f26d | 602 | FORMAT_TIMESPAN(l->target_usec, 1)); |
6ae4283c | 603 | |
2b59bf51 | 604 | LIST_FOREACH(device_limits, il, c->io_device_limits) |
e8616626 | 605 | for (CGroupIOLimitType type = 0; type < _CGROUP_IO_LIMIT_TYPE_MAX; type++) |
9be57249 TH |
606 | if (il->limits[type] != cgroup_io_limit_defaults[type]) |
607 | fprintf(f, | |
6dfb9282 | 608 | "%s%s: %s %s\n", |
9be57249 TH |
609 | prefix, |
610 | cgroup_io_limit_type_to_string(type), | |
611 | il->path, | |
2b59bf51 | 612 | FORMAT_BYTES(il->limits[type])); |
13c31542 | 613 | |
4ad49000 LP |
614 | LIST_FOREACH(device_weights, w, c->blockio_device_weights) |
615 | fprintf(f, | |
6dfb9282 | 616 | "%sBlockIODeviceWeight: %s %" PRIu64, |
4ad49000 LP |
617 | prefix, |
618 | w->path, | |
619 | w->weight); | |
620 | ||
621 | LIST_FOREACH(device_bandwidths, b, c->blockio_device_bandwidths) { | |
979d0311 TH |
622 | if (b->rbps != CGROUP_LIMIT_MAX) |
623 | fprintf(f, | |
6dfb9282 | 624 | "%sBlockIOReadBandwidth: %s %s\n", |
979d0311 TH |
625 | prefix, |
626 | b->path, | |
2b59bf51 | 627 | FORMAT_BYTES(b->rbps)); |
979d0311 TH |
628 | if (b->wbps != CGROUP_LIMIT_MAX) |
629 | fprintf(f, | |
6dfb9282 | 630 | "%sBlockIOWriteBandwidth: %s %s\n", |
979d0311 TH |
631 | prefix, |
632 | b->path, | |
2b59bf51 | 633 | FORMAT_BYTES(b->wbps)); |
4ad49000 | 634 | } |
c21c9906 | 635 | |
c71384a9 ZJS |
636 | SET_FOREACH(iaai, c->ip_address_allow) |
637 | fprintf(f, "%sIPAddressAllow: %s\n", prefix, | |
638 | IN_ADDR_PREFIX_TO_STRING(iaai->family, &iaai->address, iaai->prefixlen)); | |
639 | SET_FOREACH(iaai, c->ip_address_deny) | |
640 | fprintf(f, "%sIPAddressDeny: %s\n", prefix, | |
641 | IN_ADDR_PREFIX_TO_STRING(iaai->family, &iaai->address, iaai->prefixlen)); | |
fab34748 KL |
642 | |
643 | STRV_FOREACH(path, c->ip_filters_ingress) | |
6dfb9282 | 644 | fprintf(f, "%sIPIngressFilterPath: %s\n", prefix, *path); |
fab34748 | 645 | STRV_FOREACH(path, c->ip_filters_egress) |
6dfb9282 | 646 | fprintf(f, "%sIPEgressFilterPath: %s\n", prefix, *path); |
b894ef1b JK |
647 | |
648 | LIST_FOREACH(programs, p, c->bpf_foreign_programs) | |
649 | fprintf(f, "%sBPFProgram: %s:%s", | |
650 | prefix, bpf_cgroup_attach_type_to_string(p->attach_type), p->bpffs_path); | |
b18e9fc1 JK |
651 | |
652 | if (c->socket_bind_allow) { | |
653 | fprintf(f, "%sSocketBindAllow:", prefix); | |
654 | LIST_FOREACH(socket_bind_items, bi, c->socket_bind_allow) | |
655 | cgroup_context_dump_socket_bind_item(bi, f); | |
656 | fputc('\n', f); | |
657 | } | |
658 | ||
659 | if (c->socket_bind_deny) { | |
660 | fprintf(f, "%sSocketBindDeny:", prefix); | |
661 | LIST_FOREACH(socket_bind_items, bi, c->socket_bind_deny) | |
662 | cgroup_context_dump_socket_bind_item(bi, f); | |
663 | fputc('\n', f); | |
664 | } | |
6f50d4f7 MV |
665 | |
666 | if (c->restrict_network_interfaces) { | |
667 | char *iface; | |
668 | SET_FOREACH(iface, c->restrict_network_interfaces) | |
669 | fprintf(f, "%sRestrictNetworkInterfaces: %s\n", prefix, iface); | |
670 | } | |
b18e9fc1 JK |
671 | } |
672 | ||
673 | void cgroup_context_dump_socket_bind_item(const CGroupSocketBindItem *item, FILE *f) { | |
5587ce7f | 674 | const char *family, *colon1, *protocol = "", *colon2 = ""; |
a4817536 LP |
675 | |
676 | family = strempty(af_to_ipv4_ipv6(item->address_family)); | |
5587ce7f JK |
677 | colon1 = isempty(family) ? "" : ":"; |
678 | ||
679 | if (item->ip_protocol != 0) { | |
680 | protocol = ip_protocol_to_tcp_udp(item->ip_protocol); | |
681 | colon2 = ":"; | |
682 | } | |
b18e9fc1 JK |
683 | |
684 | if (item->nr_ports == 0) | |
5587ce7f | 685 | fprintf(f, " %s%s%s%sany", family, colon1, protocol, colon2); |
b18e9fc1 | 686 | else if (item->nr_ports == 1) |
5587ce7f | 687 | fprintf(f, " %s%s%s%s%" PRIu16, family, colon1, protocol, colon2, item->port_min); |
b18e9fc1 JK |
688 | else { |
689 | uint16_t port_max = item->port_min + item->nr_ports - 1; | |
5587ce7f JK |
690 | fprintf(f, " %s%s%s%s%" PRIu16 "-%" PRIu16, family, colon1, protocol, colon2, |
691 | item->port_min, port_max); | |
b18e9fc1 | 692 | } |
4ad49000 LP |
693 | } |
694 | ||
fd870bac YW |
695 | int cgroup_add_device_allow(CGroupContext *c, const char *dev, const char *mode) { |
696 | _cleanup_free_ CGroupDeviceAllow *a = NULL; | |
697 | _cleanup_free_ char *d = NULL; | |
698 | ||
699 | assert(c); | |
700 | assert(dev); | |
701 | assert(isempty(mode) || in_charset(mode, "rwm")); | |
702 | ||
703 | a = new(CGroupDeviceAllow, 1); | |
704 | if (!a) | |
705 | return -ENOMEM; | |
706 | ||
707 | d = strdup(dev); | |
708 | if (!d) | |
709 | return -ENOMEM; | |
710 | ||
711 | *a = (CGroupDeviceAllow) { | |
712 | .path = TAKE_PTR(d), | |
490c5a37 LP |
713 | .r = isempty(mode) || strchr(mode, 'r'), |
714 | .w = isempty(mode) || strchr(mode, 'w'), | |
715 | .m = isempty(mode) || strchr(mode, 'm'), | |
fd870bac YW |
716 | }; |
717 | ||
718 | LIST_PREPEND(device_allow, c->device_allow, a); | |
719 | TAKE_PTR(a); | |
720 | ||
721 | return 0; | |
722 | } | |
723 | ||
b894ef1b JK |
724 | int cgroup_add_bpf_foreign_program(CGroupContext *c, uint32_t attach_type, const char *bpffs_path) { |
725 | CGroupBPFForeignProgram *p; | |
726 | _cleanup_free_ char *d = NULL; | |
727 | ||
728 | assert(c); | |
729 | assert(bpffs_path); | |
730 | ||
731 | if (!path_is_normalized(bpffs_path) || !path_is_absolute(bpffs_path)) | |
732 | return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "Path is not normalized: %m"); | |
733 | ||
734 | d = strdup(bpffs_path); | |
735 | if (!d) | |
736 | return log_oom(); | |
737 | ||
738 | p = new(CGroupBPFForeignProgram, 1); | |
739 | if (!p) | |
740 | return log_oom(); | |
741 | ||
742 | *p = (CGroupBPFForeignProgram) { | |
743 | .attach_type = attach_type, | |
744 | .bpffs_path = TAKE_PTR(d), | |
745 | }; | |
746 | ||
747 | LIST_PREPEND(programs, c->bpf_foreign_programs, TAKE_PTR(p)); | |
748 | ||
749 | return 0; | |
750 | } | |
751 | ||
6264b85e CD |
752 | #define UNIT_DEFINE_ANCESTOR_MEMORY_LOOKUP(entry) \ |
753 | uint64_t unit_get_ancestor_##entry(Unit *u) { \ | |
754 | CGroupContext *c; \ | |
755 | \ | |
756 | /* 1. Is entry set in this unit? If so, use that. \ | |
757 | * 2. Is the default for this entry set in any \ | |
758 | * ancestor? If so, use that. \ | |
759 | * 3. Otherwise, return CGROUP_LIMIT_MIN. */ \ | |
760 | \ | |
761 | assert(u); \ | |
762 | \ | |
763 | c = unit_get_cgroup_context(u); \ | |
c5322608 | 764 | if (c && c->entry##_set) \ |
6264b85e CD |
765 | return c->entry; \ |
766 | \ | |
12f64221 | 767 | while ((u = UNIT_GET_SLICE(u))) { \ |
6264b85e | 768 | c = unit_get_cgroup_context(u); \ |
c5322608 | 769 | if (c && c->default_##entry##_set) \ |
6264b85e CD |
770 | return c->default_##entry; \ |
771 | } \ | |
772 | \ | |
773 | /* We've reached the root, but nobody had default for \ | |
774 | * this entry set, so set it to the kernel default. */ \ | |
775 | return CGROUP_LIMIT_MIN; \ | |
c52db42b CD |
776 | } |
777 | ||
6264b85e | 778 | UNIT_DEFINE_ANCESTOR_MEMORY_LOOKUP(memory_low); |
53fda560 | 779 | UNIT_DEFINE_ANCESTOR_MEMORY_LOOKUP(startup_memory_low); |
7ad5439e | 780 | UNIT_DEFINE_ANCESTOR_MEMORY_LOOKUP(memory_min); |
6264b85e | 781 | |
1fa3b6c2 LP |
782 | static void unit_set_xattr_graceful(Unit *u, const char *cgroup_path, const char *name, const void *data, size_t size) { |
783 | int r; | |
784 | ||
785 | assert(u); | |
786 | assert(name); | |
787 | ||
788 | if (!cgroup_path) { | |
789 | if (!u->cgroup_path) | |
790 | return; | |
791 | ||
792 | cgroup_path = u->cgroup_path; | |
793 | } | |
794 | ||
795 | r = cg_set_xattr(SYSTEMD_CGROUP_CONTROLLER, cgroup_path, name, data, size, 0); | |
796 | if (r < 0) | |
797 | log_unit_debug_errno(u, r, "Failed to set '%s' xattr on control group %s, ignoring: %m", name, empty_to_root(cgroup_path)); | |
798 | } | |
799 | ||
800 | static void unit_remove_xattr_graceful(Unit *u, const char *cgroup_path, const char *name) { | |
801 | int r; | |
802 | ||
803 | assert(u); | |
804 | assert(name); | |
805 | ||
806 | if (!cgroup_path) { | |
807 | if (!u->cgroup_path) | |
808 | return; | |
809 | ||
810 | cgroup_path = u->cgroup_path; | |
811 | } | |
812 | ||
813 | r = cg_remove_xattr(SYSTEMD_CGROUP_CONTROLLER, cgroup_path, name); | |
00675c36 | 814 | if (r < 0 && !ERRNO_IS_XATTR_ABSENT(r)) |
1fa3b6c2 LP |
815 | log_unit_debug_errno(u, r, "Failed to remove '%s' xattr flag on control group %s, ignoring: %m", name, empty_to_root(cgroup_path)); |
816 | } | |
817 | ||
4e806bfa AZ |
818 | void cgroup_oomd_xattr_apply(Unit *u, const char *cgroup_path) { |
819 | CGroupContext *c; | |
4e806bfa AZ |
820 | |
821 | assert(u); | |
822 | ||
823 | c = unit_get_cgroup_context(u); | |
824 | if (!c) | |
825 | return; | |
826 | ||
1fa3b6c2 LP |
827 | if (c->moom_preference == MANAGED_OOM_PREFERENCE_OMIT) |
828 | unit_set_xattr_graceful(u, cgroup_path, "user.oomd_omit", "1", 1); | |
4e806bfa | 829 | |
1fa3b6c2 LP |
830 | if (c->moom_preference == MANAGED_OOM_PREFERENCE_AVOID) |
831 | unit_set_xattr_graceful(u, cgroup_path, "user.oomd_avoid", "1", 1); | |
4e806bfa | 832 | |
1fa3b6c2 LP |
833 | if (c->moom_preference != MANAGED_OOM_PREFERENCE_AVOID) |
834 | unit_remove_xattr_graceful(u, cgroup_path, "user.oomd_avoid"); | |
4e806bfa | 835 | |
1fa3b6c2 LP |
836 | if (c->moom_preference != MANAGED_OOM_PREFERENCE_OMIT) |
837 | unit_remove_xattr_graceful(u, cgroup_path, "user.oomd_omit"); | |
4e806bfa AZ |
838 | } |
839 | ||
523ea123 QD |
840 | int cgroup_log_xattr_apply(Unit *u, const char *cgroup_path) { |
841 | ExecContext *c; | |
842 | size_t len, allowed_patterns_len, denied_patterns_len; | |
843 | _cleanup_free_ char *patterns = NULL, *allowed_patterns = NULL, *denied_patterns = NULL; | |
48d85160 | 844 | char *last; |
523ea123 QD |
845 | int r; |
846 | ||
847 | assert(u); | |
848 | ||
849 | c = unit_get_exec_context(u); | |
850 | if (!c) | |
851 | /* Some unit types have a cgroup context but no exec context, so we do not log | |
852 | * any error here to avoid confusion. */ | |
853 | return 0; | |
854 | ||
855 | if (set_isempty(c->log_filter_allowed_patterns) && set_isempty(c->log_filter_denied_patterns)) { | |
856 | unit_remove_xattr_graceful(u, cgroup_path, "user.journald_log_filter_patterns"); | |
857 | return 0; | |
858 | } | |
859 | ||
860 | r = set_make_nulstr(c->log_filter_allowed_patterns, &allowed_patterns, &allowed_patterns_len); | |
861 | if (r < 0) | |
862 | return log_debug_errno(r, "Failed to make nulstr from set: %m"); | |
863 | ||
864 | r = set_make_nulstr(c->log_filter_denied_patterns, &denied_patterns, &denied_patterns_len); | |
865 | if (r < 0) | |
866 | return log_debug_errno(r, "Failed to make nulstr from set: %m"); | |
867 | ||
868 | /* Use nul character separated strings without trailing nul */ | |
869 | allowed_patterns_len = LESS_BY(allowed_patterns_len, 1u); | |
870 | denied_patterns_len = LESS_BY(denied_patterns_len, 1u); | |
871 | ||
872 | len = allowed_patterns_len + 1 + denied_patterns_len; | |
873 | patterns = new(char, len); | |
874 | if (!patterns) | |
875 | return log_oom_debug(); | |
876 | ||
48d85160 QD |
877 | last = mempcpy_safe(patterns, allowed_patterns, allowed_patterns_len); |
878 | *(last++) = '\xff'; | |
879 | memcpy_safe(last, denied_patterns, denied_patterns_len); | |
523ea123 QD |
880 | |
881 | unit_set_xattr_graceful(u, cgroup_path, "user.journald_log_filter_patterns", patterns, len); | |
882 | ||
883 | return 0; | |
884 | } | |
885 | ||
0d2d6fbf | 886 | static void cgroup_xattr_apply(Unit *u) { |
d9bc1c36 | 887 | bool b; |
0d2d6fbf CD |
888 | |
889 | assert(u); | |
890 | ||
b1cd8ed4 NR |
891 | /* The 'user.*' xattrs can be set from a user manager. */ |
892 | cgroup_oomd_xattr_apply(u, u->cgroup_path); | |
523ea123 | 893 | cgroup_log_xattr_apply(u, u->cgroup_path); |
b1cd8ed4 | 894 | |
0d2d6fbf CD |
895 | if (!MANAGER_IS_SYSTEM(u->manager)) |
896 | return; | |
897 | ||
1fa3b6c2 LP |
898 | b = !sd_id128_is_null(u->invocation_id); |
899 | FOREACH_STRING(xn, "trusted.invocation_id", "user.invocation_id") { | |
900 | if (b) | |
901 | unit_set_xattr_graceful(u, NULL, xn, SD_ID128_TO_STRING(u->invocation_id), 32); | |
902 | else | |
903 | unit_remove_xattr_graceful(u, NULL, xn); | |
3288ea8f | 904 | } |
0d2d6fbf | 905 | |
d9bc1c36 LP |
906 | /* Indicate on the cgroup whether delegation is on, via an xattr. This is best-effort, as old kernels |
907 | * didn't support xattrs on cgroups at all. Later they got support for setting 'trusted.*' xattrs, | |
908 | * and even later 'user.*' xattrs. We started setting this field when 'trusted.*' was added, and | |
909 | * given this is now pretty much API, let's continue to support that. But also set 'user.*' as well, | |
910 | * since it is readable by any user, not just CAP_SYS_ADMIN. This hence comes with slightly weaker | |
911 | * security (as users who got delegated cgroups could turn it off if they like), but this shouldn't | |
912 | * be a big problem given this communicates delegation state to clients, but the manager never reads | |
913 | * it. */ | |
914 | b = unit_cgroup_delegate(u); | |
915 | FOREACH_STRING(xn, "trusted.delegate", "user.delegate") { | |
1fa3b6c2 LP |
916 | if (b) |
917 | unit_set_xattr_graceful(u, NULL, xn, "1", 1); | |
918 | else | |
919 | unit_remove_xattr_graceful(u, NULL, xn); | |
3288ea8f | 920 | } |
0d2d6fbf CD |
921 | } |
922 | ||
45c2e068 | 923 | static int lookup_block_device(const char *p, dev_t *ret) { |
f5855697 YS |
924 | dev_t rdev, dev = 0; |
925 | mode_t mode; | |
45c2e068 | 926 | int r; |
4ad49000 LP |
927 | |
928 | assert(p); | |
45c2e068 | 929 | assert(ret); |
4ad49000 | 930 | |
f5855697 | 931 | r = device_path_parse_major_minor(p, &mode, &rdev); |
d5aecba6 | 932 | if (r == -ENODEV) { /* not a parsable device node, need to go to disk */ |
f5855697 | 933 | struct stat st; |
57f1030b | 934 | |
d5aecba6 LP |
935 | if (stat(p, &st) < 0) |
936 | return log_warning_errno(errno, "Couldn't stat device '%s': %m", p); | |
57f1030b | 937 | |
f5855697 | 938 | mode = st.st_mode; |
a0d6590c LP |
939 | rdev = st.st_rdev; |
940 | dev = st.st_dev; | |
d5aecba6 LP |
941 | } else if (r < 0) |
942 | return log_warning_errno(r, "Failed to parse major/minor from path '%s': %m", p); | |
943 | ||
57f1030b LP |
944 | if (S_ISCHR(mode)) |
945 | return log_warning_errno(SYNTHETIC_ERRNO(ENOTBLK), | |
946 | "Device node '%s' is a character device, but block device needed.", p); | |
947 | if (S_ISBLK(mode)) | |
f5855697 YS |
948 | *ret = rdev; |
949 | else if (major(dev) != 0) | |
950 | *ret = dev; /* If this is not a device node then use the block device this file is stored on */ | |
45c2e068 LP |
951 | else { |
952 | /* If this is btrfs, getting the backing block device is a bit harder */ | |
953 | r = btrfs_get_block_device(p, ret); | |
57f1030b LP |
954 | if (r == -ENOTTY) |
955 | return log_warning_errno(SYNTHETIC_ERRNO(ENODEV), | |
956 | "'%s' is not a block device node, and file system block device cannot be determined or is not local.", p); | |
957 | if (r < 0) | |
45c2e068 | 958 | return log_warning_errno(r, "Failed to determine block device backing btrfs file system '%s': %m", p); |
4ad49000 | 959 | } |
8e274523 | 960 | |
b7cf4b4e BB |
961 | /* If this is a LUKS/DM device, recursively try to get the originating block device */ |
962 | while (block_get_originating(*ret, ret) > 0); | |
45c2e068 LP |
963 | |
964 | /* If this is a partition, try to get the originating block device */ | |
965 | (void) block_get_whole_disk(*ret, ret); | |
8e274523 | 966 | return 0; |
8e274523 LP |
967 | } |
968 | ||
66ebf6c0 TH |
969 | static bool cgroup_context_has_cpu_weight(CGroupContext *c) { |
970 | return c->cpu_weight != CGROUP_WEIGHT_INVALID || | |
971 | c->startup_cpu_weight != CGROUP_WEIGHT_INVALID; | |
972 | } | |
973 | ||
974 | static bool cgroup_context_has_cpu_shares(CGroupContext *c) { | |
975 | return c->cpu_shares != CGROUP_CPU_SHARES_INVALID || | |
976 | c->startup_cpu_shares != CGROUP_CPU_SHARES_INVALID; | |
977 | } | |
978 | ||
31d3a520 PM |
979 | static bool cgroup_context_has_allowed_cpus(CGroupContext *c) { |
980 | return c->cpuset_cpus.set || c->startup_cpuset_cpus.set; | |
981 | } | |
982 | ||
983 | static bool cgroup_context_has_allowed_mems(CGroupContext *c) { | |
984 | return c->cpuset_mems.set || c->startup_cpuset_mems.set; | |
985 | } | |
986 | ||
66ebf6c0 | 987 | static uint64_t cgroup_context_cpu_weight(CGroupContext *c, ManagerState state) { |
9dfb6a3a | 988 | if (IN_SET(state, MANAGER_STARTING, MANAGER_INITIALIZING, MANAGER_STOPPING) && |
66ebf6c0 TH |
989 | c->startup_cpu_weight != CGROUP_WEIGHT_INVALID) |
990 | return c->startup_cpu_weight; | |
991 | else if (c->cpu_weight != CGROUP_WEIGHT_INVALID) | |
992 | return c->cpu_weight; | |
993 | else | |
994 | return CGROUP_WEIGHT_DEFAULT; | |
995 | } | |
996 | ||
997 | static uint64_t cgroup_context_cpu_shares(CGroupContext *c, ManagerState state) { | |
9dfb6a3a | 998 | if (IN_SET(state, MANAGER_STARTING, MANAGER_INITIALIZING, MANAGER_STOPPING) && |
66ebf6c0 TH |
999 | c->startup_cpu_shares != CGROUP_CPU_SHARES_INVALID) |
1000 | return c->startup_cpu_shares; | |
1001 | else if (c->cpu_shares != CGROUP_CPU_SHARES_INVALID) | |
1002 | return c->cpu_shares; | |
1003 | else | |
1004 | return CGROUP_CPU_SHARES_DEFAULT; | |
1005 | } | |
1006 | ||
31d3a520 | 1007 | static CPUSet *cgroup_context_allowed_cpus(CGroupContext *c, ManagerState state) { |
9dfb6a3a | 1008 | if (IN_SET(state, MANAGER_STARTING, MANAGER_INITIALIZING, MANAGER_STOPPING) && |
31d3a520 PM |
1009 | c->startup_cpuset_cpus.set) |
1010 | return &c->startup_cpuset_cpus; | |
1011 | else | |
1012 | return &c->cpuset_cpus; | |
1013 | } | |
1014 | ||
1015 | static CPUSet *cgroup_context_allowed_mems(CGroupContext *c, ManagerState state) { | |
9dfb6a3a | 1016 | if (IN_SET(state, MANAGER_STARTING, MANAGER_INITIALIZING, MANAGER_STOPPING) && |
31d3a520 PM |
1017 | c->startup_cpuset_mems.set) |
1018 | return &c->startup_cpuset_mems; | |
1019 | else | |
1020 | return &c->cpuset_mems; | |
1021 | } | |
1022 | ||
10f28641 FB |
1023 | usec_t cgroup_cpu_adjust_period(usec_t period, usec_t quota, usec_t resolution, usec_t max_period) { |
1024 | /* kernel uses a minimum resolution of 1ms, so both period and (quota * period) | |
1025 | * need to be higher than that boundary. quota is specified in USecPerSec. | |
1026 | * Additionally, period must be at most max_period. */ | |
1027 | assert(quota > 0); | |
1028 | ||
1029 | return MIN(MAX3(period, resolution, resolution * USEC_PER_SEC / quota), max_period); | |
1030 | } | |
1031 | ||
1032 | static usec_t cgroup_cpu_adjust_period_and_log(Unit *u, usec_t period, usec_t quota) { | |
1033 | usec_t new_period; | |
1034 | ||
1035 | if (quota == USEC_INFINITY) | |
1036 | /* Always use default period for infinity quota. */ | |
1037 | return CGROUP_CPU_QUOTA_DEFAULT_PERIOD_USEC; | |
1038 | ||
1039 | if (period == USEC_INFINITY) | |
1040 | /* Default period was requested. */ | |
1041 | period = CGROUP_CPU_QUOTA_DEFAULT_PERIOD_USEC; | |
1042 | ||
1043 | /* Clamp to interval [1ms, 1s] */ | |
1044 | new_period = cgroup_cpu_adjust_period(period, quota, USEC_PER_MSEC, USEC_PER_SEC); | |
1045 | ||
1046 | if (new_period != period) { | |
8ed6f81b | 1047 | log_unit_full(u, u->warned_clamping_cpu_quota_period ? LOG_DEBUG : LOG_WARNING, |
10f28641 | 1048 | "Clamping CPU interval for cpu.max: period is now %s", |
5291f26d | 1049 | FORMAT_TIMESPAN(new_period, 1)); |
527ede0c | 1050 | u->warned_clamping_cpu_quota_period = true; |
10f28641 FB |
1051 | } |
1052 | ||
1053 | return new_period; | |
1054 | } | |
1055 | ||
52fecf20 LP |
1056 | static void cgroup_apply_unified_cpu_weight(Unit *u, uint64_t weight) { |
1057 | char buf[DECIMAL_STR_MAX(uint64_t) + 2]; | |
66ebf6c0 | 1058 | |
c8340822 | 1059 | if (weight == CGROUP_WEIGHT_IDLE) |
1060 | return; | |
66ebf6c0 | 1061 | xsprintf(buf, "%" PRIu64 "\n", weight); |
293d32df | 1062 | (void) set_attribute_and_warn(u, "cpu", "cpu.weight", buf); |
52fecf20 LP |
1063 | } |
1064 | ||
c8340822 | 1065 | static void cgroup_apply_unified_cpu_idle(Unit *u, uint64_t weight) { |
1066 | int r; | |
1067 | bool is_idle; | |
1068 | const char *idle_val; | |
1069 | ||
1070 | is_idle = weight == CGROUP_WEIGHT_IDLE; | |
1071 | idle_val = one_zero(is_idle); | |
1072 | r = cg_set_attribute("cpu", u->cgroup_path, "cpu.idle", idle_val); | |
1073 | if (r < 0 && (r != -ENOENT || is_idle)) | |
1074 | log_unit_full_errno(u, LOG_LEVEL_CGROUP_WRITE(r), r, "Failed to set '%s' attribute on '%s' to '%s': %m", | |
1075 | "cpu.idle", empty_to_root(u->cgroup_path), idle_val); | |
1076 | } | |
1077 | ||
10f28641 | 1078 | static void cgroup_apply_unified_cpu_quota(Unit *u, usec_t quota, usec_t period) { |
52fecf20 | 1079 | char buf[(DECIMAL_STR_MAX(usec_t) + 1) * 2 + 1]; |
66ebf6c0 | 1080 | |
10f28641 | 1081 | period = cgroup_cpu_adjust_period_and_log(u, period, quota); |
66ebf6c0 TH |
1082 | if (quota != USEC_INFINITY) |
1083 | xsprintf(buf, USEC_FMT " " USEC_FMT "\n", | |
10f28641 | 1084 | MAX(quota * period / USEC_PER_SEC, USEC_PER_MSEC), period); |
66ebf6c0 | 1085 | else |
10f28641 | 1086 | xsprintf(buf, "max " USEC_FMT "\n", period); |
293d32df | 1087 | (void) set_attribute_and_warn(u, "cpu", "cpu.max", buf); |
66ebf6c0 TH |
1088 | } |
1089 | ||
52fecf20 LP |
1090 | static void cgroup_apply_legacy_cpu_shares(Unit *u, uint64_t shares) { |
1091 | char buf[DECIMAL_STR_MAX(uint64_t) + 2]; | |
66ebf6c0 TH |
1092 | |
1093 | xsprintf(buf, "%" PRIu64 "\n", shares); | |
293d32df | 1094 | (void) set_attribute_and_warn(u, "cpu", "cpu.shares", buf); |
52fecf20 LP |
1095 | } |
1096 | ||
10f28641 | 1097 | static void cgroup_apply_legacy_cpu_quota(Unit *u, usec_t quota, usec_t period) { |
52fecf20 | 1098 | char buf[DECIMAL_STR_MAX(usec_t) + 2]; |
66ebf6c0 | 1099 | |
10f28641 FB |
1100 | period = cgroup_cpu_adjust_period_and_log(u, period, quota); |
1101 | ||
1102 | xsprintf(buf, USEC_FMT "\n", period); | |
293d32df | 1103 | (void) set_attribute_and_warn(u, "cpu", "cpu.cfs_period_us", buf); |
66ebf6c0 TH |
1104 | |
1105 | if (quota != USEC_INFINITY) { | |
10f28641 | 1106 | xsprintf(buf, USEC_FMT "\n", MAX(quota * period / USEC_PER_SEC, USEC_PER_MSEC)); |
293d32df | 1107 | (void) set_attribute_and_warn(u, "cpu", "cpu.cfs_quota_us", buf); |
66ebf6c0 | 1108 | } else |
589a5f7a | 1109 | (void) set_attribute_and_warn(u, "cpu", "cpu.cfs_quota_us", "-1\n"); |
66ebf6c0 TH |
1110 | } |
1111 | ||
1112 | static uint64_t cgroup_cpu_shares_to_weight(uint64_t shares) { | |
1113 | return CLAMP(shares * CGROUP_WEIGHT_DEFAULT / CGROUP_CPU_SHARES_DEFAULT, | |
1114 | CGROUP_WEIGHT_MIN, CGROUP_WEIGHT_MAX); | |
1115 | } | |
1116 | ||
1117 | static uint64_t cgroup_cpu_weight_to_shares(uint64_t weight) { | |
c8340822 | 1118 | /* we don't support idle in cgroupv1 */ |
1119 | if (weight == CGROUP_WEIGHT_IDLE) | |
1120 | return CGROUP_CPU_SHARES_MIN; | |
1121 | ||
66ebf6c0 TH |
1122 | return CLAMP(weight * CGROUP_CPU_SHARES_DEFAULT / CGROUP_WEIGHT_DEFAULT, |
1123 | CGROUP_CPU_SHARES_MIN, CGROUP_CPU_SHARES_MAX); | |
1124 | } | |
1125 | ||
2cea199e | 1126 | static void cgroup_apply_unified_cpuset(Unit *u, const CPUSet *cpus, const char *name) { |
047f5d63 PH |
1127 | _cleanup_free_ char *buf = NULL; |
1128 | ||
2cea199e | 1129 | buf = cpu_set_to_range_string(cpus); |
c259ac9a LP |
1130 | if (!buf) { |
1131 | log_oom(); | |
1132 | return; | |
1133 | } | |
047f5d63 PH |
1134 | |
1135 | (void) set_attribute_and_warn(u, "cpuset", name, buf); | |
1136 | } | |
1137 | ||
508c45da | 1138 | static bool cgroup_context_has_io_config(CGroupContext *c) { |
538b4852 TH |
1139 | return c->io_accounting || |
1140 | c->io_weight != CGROUP_WEIGHT_INVALID || | |
1141 | c->startup_io_weight != CGROUP_WEIGHT_INVALID || | |
1142 | c->io_device_weights || | |
6ae4283c | 1143 | c->io_device_latencies || |
538b4852 TH |
1144 | c->io_device_limits; |
1145 | } | |
1146 | ||
508c45da | 1147 | static bool cgroup_context_has_blockio_config(CGroupContext *c) { |
538b4852 TH |
1148 | return c->blockio_accounting || |
1149 | c->blockio_weight != CGROUP_BLKIO_WEIGHT_INVALID || | |
1150 | c->startup_blockio_weight != CGROUP_BLKIO_WEIGHT_INVALID || | |
1151 | c->blockio_device_weights || | |
1152 | c->blockio_device_bandwidths; | |
1153 | } | |
1154 | ||
508c45da | 1155 | static uint64_t cgroup_context_io_weight(CGroupContext *c, ManagerState state) { |
9dfb6a3a | 1156 | if (IN_SET(state, MANAGER_STARTING, MANAGER_INITIALIZING, MANAGER_STOPPING) && |
64faf04c TH |
1157 | c->startup_io_weight != CGROUP_WEIGHT_INVALID) |
1158 | return c->startup_io_weight; | |
d38655d7 | 1159 | if (c->io_weight != CGROUP_WEIGHT_INVALID) |
64faf04c | 1160 | return c->io_weight; |
d38655d7 | 1161 | return CGROUP_WEIGHT_DEFAULT; |
64faf04c TH |
1162 | } |
1163 | ||
508c45da | 1164 | static uint64_t cgroup_context_blkio_weight(CGroupContext *c, ManagerState state) { |
9dfb6a3a | 1165 | if (IN_SET(state, MANAGER_STARTING, MANAGER_INITIALIZING, MANAGER_STOPPING) && |
64faf04c TH |
1166 | c->startup_blockio_weight != CGROUP_BLKIO_WEIGHT_INVALID) |
1167 | return c->startup_blockio_weight; | |
d38655d7 | 1168 | if (c->blockio_weight != CGROUP_BLKIO_WEIGHT_INVALID) |
64faf04c | 1169 | return c->blockio_weight; |
d38655d7 | 1170 | return CGROUP_BLKIO_WEIGHT_DEFAULT; |
64faf04c TH |
1171 | } |
1172 | ||
508c45da | 1173 | static uint64_t cgroup_weight_blkio_to_io(uint64_t blkio_weight) { |
538b4852 TH |
1174 | return CLAMP(blkio_weight * CGROUP_WEIGHT_DEFAULT / CGROUP_BLKIO_WEIGHT_DEFAULT, |
1175 | CGROUP_WEIGHT_MIN, CGROUP_WEIGHT_MAX); | |
1176 | } | |
1177 | ||
508c45da | 1178 | static uint64_t cgroup_weight_io_to_blkio(uint64_t io_weight) { |
538b4852 TH |
1179 | return CLAMP(io_weight * CGROUP_BLKIO_WEIGHT_DEFAULT / CGROUP_WEIGHT_DEFAULT, |
1180 | CGROUP_BLKIO_WEIGHT_MIN, CGROUP_BLKIO_WEIGHT_MAX); | |
1181 | } | |
1182 | ||
3e6eafdd | 1183 | static int set_bfq_weight(Unit *u, const char *controller, dev_t dev, uint64_t io_weight) { |
1cf4a685 MK |
1184 | static const char * const prop_names[] = { |
1185 | "IOWeight", | |
1186 | "BlockIOWeight", | |
1187 | "IODeviceWeight", | |
1188 | "BlockIODeviceWeight", | |
1189 | }; | |
8d75f60e | 1190 | static bool warned = false; |
9f0c0c4e | 1191 | char buf[DECIMAL_STR_MAX(dev_t)*2+2+DECIMAL_STR_MAX(uint64_t)+STRLEN("\n")]; |
bec17e80 MK |
1192 | const char *p; |
1193 | uint64_t bfq_weight; | |
8d75f60e | 1194 | int r; |
bec17e80 MK |
1195 | |
1196 | /* FIXME: drop this function when distro kernels properly support BFQ through "io.weight" | |
1197 | * See also: https://github.com/systemd/systemd/pull/13335 and | |
1198 | * https://github.com/torvalds/linux/commit/65752aef0a407e1ef17ec78a7fc31ba4e0b360f9. */ | |
1199 | p = strjoina(controller, ".bfq.weight"); | |
1200 | /* Adjust to kernel range is 1..1000, the default is 100. */ | |
1201 | bfq_weight = BFQ_WEIGHT(io_weight); | |
1202 | ||
9f0c0c4e | 1203 | if (major(dev) > 0) |
ec61371f | 1204 | xsprintf(buf, DEVNUM_FORMAT_STR " %" PRIu64 "\n", DEVNUM_FORMAT_VAL(dev), bfq_weight); |
9f0c0c4e MK |
1205 | else |
1206 | xsprintf(buf, "%" PRIu64 "\n", bfq_weight); | |
bec17e80 | 1207 | |
8d75f60e MK |
1208 | r = cg_set_attribute(controller, u->cgroup_path, p, buf); |
1209 | ||
1210 | /* FIXME: drop this when kernels prior | |
1211 | * 795fe54c2a82 ("bfq: Add per-device weight") v5.4 | |
1212 | * are not interesting anymore. Old kernels will fail with EINVAL, while new kernels won't return | |
1213 | * EINVAL on properly formatted input by us. Treat EINVAL accordingly. */ | |
3e6eafdd MK |
1214 | if (r == -EINVAL && major(dev) > 0) { |
1215 | if (!warned) { | |
1216 | log_unit_warning(u, "Kernel version does not accept per-device setting in %s.", p); | |
1217 | warned = true; | |
1218 | } | |
1219 | r = -EOPNOTSUPP; /* mask as unconfigured device */ | |
1220 | } else if (r >= 0 && io_weight != bfq_weight) | |
1cf4a685 MK |
1221 | log_unit_debug(u, "%s=%" PRIu64 " scaled to %s=%" PRIu64, |
1222 | prop_names[2*(major(dev) > 0) + streq(controller, "blkio")], | |
bec17e80 | 1223 | io_weight, p, bfq_weight); |
3e6eafdd | 1224 | return r; |
bec17e80 MK |
1225 | } |
1226 | ||
f29ff115 | 1227 | static void cgroup_apply_io_device_weight(Unit *u, const char *dev_path, uint64_t io_weight) { |
64faf04c TH |
1228 | char buf[DECIMAL_STR_MAX(dev_t)*2+2+DECIMAL_STR_MAX(uint64_t)+1]; |
1229 | dev_t dev; | |
3e6eafdd | 1230 | int r, r1, r2; |
64faf04c | 1231 | |
3e6eafdd | 1232 | if (lookup_block_device(dev_path, &dev) < 0) |
64faf04c TH |
1233 | return; |
1234 | ||
3e6eafdd | 1235 | r1 = set_bfq_weight(u, "io", dev, io_weight); |
9f0c0c4e | 1236 | |
ec61371f | 1237 | xsprintf(buf, DEVNUM_FORMAT_STR " %" PRIu64 "\n", DEVNUM_FORMAT_VAL(dev), io_weight); |
3e6eafdd MK |
1238 | r2 = cg_set_attribute("io", u->cgroup_path, "io.weight", buf); |
1239 | ||
1240 | /* Look at the configured device, when both fail, prefer io.weight errno. */ | |
1241 | r = r2 == -EOPNOTSUPP ? r1 : r2; | |
1242 | ||
1243 | if (r < 0) | |
1244 | log_unit_full_errno(u, LOG_LEVEL_CGROUP_WRITE(r), | |
1245 | r, "Failed to set 'io[.bfq].weight' attribute on '%s' to '%.*s': %m", | |
1246 | empty_to_root(u->cgroup_path), (int) strcspn(buf, NEWLINE), buf); | |
64faf04c TH |
1247 | } |
1248 | ||
f29ff115 | 1249 | static void cgroup_apply_blkio_device_weight(Unit *u, const char *dev_path, uint64_t blkio_weight) { |
64faf04c TH |
1250 | char buf[DECIMAL_STR_MAX(dev_t)*2+2+DECIMAL_STR_MAX(uint64_t)+1]; |
1251 | dev_t dev; | |
1252 | int r; | |
1253 | ||
1254 | r = lookup_block_device(dev_path, &dev); | |
1255 | if (r < 0) | |
1256 | return; | |
1257 | ||
ec61371f | 1258 | xsprintf(buf, DEVNUM_FORMAT_STR " %" PRIu64 "\n", DEVNUM_FORMAT_VAL(dev), blkio_weight); |
293d32df | 1259 | (void) set_attribute_and_warn(u, "blkio", "blkio.weight_device", buf); |
64faf04c TH |
1260 | } |
1261 | ||
6ae4283c TH |
1262 | static void cgroup_apply_io_device_latency(Unit *u, const char *dev_path, usec_t target) { |
1263 | char buf[DECIMAL_STR_MAX(dev_t)*2+2+7+DECIMAL_STR_MAX(uint64_t)+1]; | |
1264 | dev_t dev; | |
1265 | int r; | |
1266 | ||
1267 | r = lookup_block_device(dev_path, &dev); | |
1268 | if (r < 0) | |
1269 | return; | |
1270 | ||
1271 | if (target != USEC_INFINITY) | |
ec61371f | 1272 | xsprintf(buf, DEVNUM_FORMAT_STR " target=%" PRIu64 "\n", DEVNUM_FORMAT_VAL(dev), target); |
6ae4283c | 1273 | else |
ec61371f | 1274 | xsprintf(buf, DEVNUM_FORMAT_STR " target=max\n", DEVNUM_FORMAT_VAL(dev)); |
6ae4283c | 1275 | |
293d32df | 1276 | (void) set_attribute_and_warn(u, "io", "io.latency", buf); |
6ae4283c TH |
1277 | } |
1278 | ||
17ae2780 | 1279 | static void cgroup_apply_io_device_limit(Unit *u, const char *dev_path, uint64_t *limits) { |
4c1f9343 ZJS |
1280 | char limit_bufs[_CGROUP_IO_LIMIT_TYPE_MAX][DECIMAL_STR_MAX(uint64_t)], |
1281 | buf[DECIMAL_STR_MAX(dev_t)*2+2+(6+DECIMAL_STR_MAX(uint64_t)+1)*4]; | |
64faf04c | 1282 | dev_t dev; |
64faf04c | 1283 | |
4c1f9343 | 1284 | if (lookup_block_device(dev_path, &dev) < 0) |
17ae2780 | 1285 | return; |
64faf04c | 1286 | |
4c1f9343 | 1287 | for (CGroupIOLimitType type = 0; type < _CGROUP_IO_LIMIT_TYPE_MAX; type++) |
17ae2780 | 1288 | if (limits[type] != cgroup_io_limit_defaults[type]) |
64faf04c | 1289 | xsprintf(limit_bufs[type], "%" PRIu64, limits[type]); |
17ae2780 | 1290 | else |
64faf04c | 1291 | xsprintf(limit_bufs[type], "%s", limits[type] == CGROUP_LIMIT_MAX ? "max" : "0"); |
64faf04c | 1292 | |
ec61371f | 1293 | xsprintf(buf, DEVNUM_FORMAT_STR " rbps=%s wbps=%s riops=%s wiops=%s\n", DEVNUM_FORMAT_VAL(dev), |
64faf04c TH |
1294 | limit_bufs[CGROUP_IO_RBPS_MAX], limit_bufs[CGROUP_IO_WBPS_MAX], |
1295 | limit_bufs[CGROUP_IO_RIOPS_MAX], limit_bufs[CGROUP_IO_WIOPS_MAX]); | |
293d32df | 1296 | (void) set_attribute_and_warn(u, "io", "io.max", buf); |
64faf04c TH |
1297 | } |
1298 | ||
17ae2780 | 1299 | static void cgroup_apply_blkio_device_limit(Unit *u, const char *dev_path, uint64_t rbps, uint64_t wbps) { |
64faf04c TH |
1300 | char buf[DECIMAL_STR_MAX(dev_t)*2+2+DECIMAL_STR_MAX(uint64_t)+1]; |
1301 | dev_t dev; | |
64faf04c | 1302 | |
4c1f9343 | 1303 | if (lookup_block_device(dev_path, &dev) < 0) |
17ae2780 | 1304 | return; |
64faf04c | 1305 | |
ec61371f | 1306 | sprintf(buf, DEVNUM_FORMAT_STR " %" PRIu64 "\n", DEVNUM_FORMAT_VAL(dev), rbps); |
293d32df | 1307 | (void) set_attribute_and_warn(u, "blkio", "blkio.throttle.read_bps_device", buf); |
64faf04c | 1308 | |
ec61371f | 1309 | sprintf(buf, DEVNUM_FORMAT_STR " %" PRIu64 "\n", DEVNUM_FORMAT_VAL(dev), wbps); |
293d32df | 1310 | (void) set_attribute_and_warn(u, "blkio", "blkio.throttle.write_bps_device", buf); |
64faf04c TH |
1311 | } |
1312 | ||
c52db42b CD |
1313 | static bool unit_has_unified_memory_config(Unit *u) { |
1314 | CGroupContext *c; | |
1315 | ||
1316 | assert(u); | |
1317 | ||
806a9362 | 1318 | assert_se(c = unit_get_cgroup_context(u)); |
c52db42b | 1319 | |
53fda560 LB |
1320 | return unit_get_ancestor_memory_min(u) > 0 || |
1321 | unit_get_ancestor_memory_low(u) > 0 || unit_get_ancestor_startup_memory_low(u) > 0 || | |
1322 | c->memory_high != CGROUP_LIMIT_MAX || c->startup_memory_high_set || | |
1323 | c->memory_max != CGROUP_LIMIT_MAX || c->startup_memory_max_set || | |
1324 | c->memory_swap_max != CGROUP_LIMIT_MAX || c->startup_memory_swap_max_set || | |
1325 | c->memory_zswap_max != CGROUP_LIMIT_MAX || c->startup_memory_zswap_max_set; | |
da4d897e TH |
1326 | } |
1327 | ||
f29ff115 | 1328 | static void cgroup_apply_unified_memory_limit(Unit *u, const char *file, uint64_t v) { |
589a5f7a | 1329 | char buf[DECIMAL_STR_MAX(uint64_t) + 1] = "max\n"; |
da4d897e TH |
1330 | |
1331 | if (v != CGROUP_LIMIT_MAX) | |
1332 | xsprintf(buf, "%" PRIu64 "\n", v); | |
1333 | ||
293d32df | 1334 | (void) set_attribute_and_warn(u, "memory", file, buf); |
da4d897e TH |
1335 | } |
1336 | ||
0f2d84d2 | 1337 | static void cgroup_apply_firewall(Unit *u) { |
0f2d84d2 LP |
1338 | assert(u); |
1339 | ||
acf7f253 | 1340 | /* Best-effort: let's apply IP firewalling and/or accounting if that's enabled */ |
906c06f6 | 1341 | |
acf7f253 | 1342 | if (bpf_firewall_compile(u) < 0) |
906c06f6 DM |
1343 | return; |
1344 | ||
fab34748 | 1345 | (void) bpf_firewall_load_custom(u); |
906c06f6 | 1346 | (void) bpf_firewall_install(u); |
906c06f6 DM |
1347 | } |
1348 | ||
a8e5eb17 JK |
1349 | static void cgroup_apply_socket_bind(Unit *u) { |
1350 | assert(u); | |
1351 | ||
cd09a5f3 | 1352 | (void) bpf_socket_bind_install(u); |
a8e5eb17 JK |
1353 | } |
1354 | ||
6f50d4f7 MV |
1355 | static void cgroup_apply_restrict_network_interfaces(Unit *u) { |
1356 | assert(u); | |
1357 | ||
1358 | (void) restrict_network_interfaces_install(u); | |
1359 | } | |
1360 | ||
8b139557 | 1361 | static int cgroup_apply_devices(Unit *u) { |
76dc1725 | 1362 | _cleanup_(bpf_program_freep) BPFProgram *prog = NULL; |
8b139557 ZJS |
1363 | const char *path; |
1364 | CGroupContext *c; | |
45669ae2 | 1365 | CGroupDevicePolicy policy; |
8b139557 ZJS |
1366 | int r; |
1367 | ||
1368 | assert_se(c = unit_get_cgroup_context(u)); | |
1369 | assert_se(path = u->cgroup_path); | |
1370 | ||
45669ae2 ZJS |
1371 | policy = c->device_policy; |
1372 | ||
8b139557 | 1373 | if (cg_all_unified() > 0) { |
45669ae2 | 1374 | r = bpf_devices_cgroup_init(&prog, policy, c->device_allow); |
8b139557 ZJS |
1375 | if (r < 0) |
1376 | return log_unit_warning_errno(u, r, "Failed to initialize device control bpf program: %m"); | |
1377 | ||
1378 | } else { | |
1379 | /* Changing the devices list of a populated cgroup might result in EINVAL, hence ignore | |
1380 | * EINVAL here. */ | |
1381 | ||
45669ae2 | 1382 | if (c->device_allow || policy != CGROUP_DEVICE_POLICY_AUTO) |
8b139557 ZJS |
1383 | r = cg_set_attribute("devices", path, "devices.deny", "a"); |
1384 | else | |
1385 | r = cg_set_attribute("devices", path, "devices.allow", "a"); | |
1386 | if (r < 0) | |
8ed6f81b YW |
1387 | log_unit_full_errno(u, IN_SET(r, -ENOENT, -EROFS, -EINVAL, -EACCES, -EPERM) ? LOG_DEBUG : LOG_WARNING, r, |
1388 | "Failed to reset devices.allow/devices.deny: %m"); | |
8b139557 ZJS |
1389 | } |
1390 | ||
6b000af4 | 1391 | bool allow_list_static = policy == CGROUP_DEVICE_POLICY_CLOSED || |
45669ae2 | 1392 | (policy == CGROUP_DEVICE_POLICY_AUTO && c->device_allow); |
6b000af4 LP |
1393 | if (allow_list_static) |
1394 | (void) bpf_devices_allow_list_static(prog, path); | |
8b139557 | 1395 | |
6b000af4 | 1396 | bool any = allow_list_static; |
8b139557 ZJS |
1397 | LIST_FOREACH(device_allow, a, c->device_allow) { |
1398 | char acc[4], *val; | |
1399 | unsigned k = 0; | |
1400 | ||
1401 | if (a->r) | |
1402 | acc[k++] = 'r'; | |
1403 | if (a->w) | |
1404 | acc[k++] = 'w'; | |
1405 | if (a->m) | |
1406 | acc[k++] = 'm'; | |
8b139557 ZJS |
1407 | if (k == 0) |
1408 | continue; | |
8b139557 ZJS |
1409 | acc[k++] = 0; |
1410 | ||
1411 | if (path_startswith(a->path, "/dev/")) | |
6b000af4 | 1412 | r = bpf_devices_allow_list_device(prog, path, a->path, acc); |
8b139557 | 1413 | else if ((val = startswith(a->path, "block-"))) |
6b000af4 | 1414 | r = bpf_devices_allow_list_major(prog, path, val, 'b', acc); |
8b139557 | 1415 | else if ((val = startswith(a->path, "char-"))) |
6b000af4 | 1416 | r = bpf_devices_allow_list_major(prog, path, val, 'c', acc); |
45669ae2 | 1417 | else { |
8b139557 | 1418 | log_unit_debug(u, "Ignoring device '%s' while writing cgroup attribute.", a->path); |
45669ae2 ZJS |
1419 | continue; |
1420 | } | |
1421 | ||
1422 | if (r >= 0) | |
1423 | any = true; | |
1424 | } | |
1425 | ||
1426 | if (prog && !any) { | |
1427 | log_unit_warning_errno(u, SYNTHETIC_ERRNO(ENODEV), "No devices matched by device filter."); | |
1428 | ||
1429 | /* The kernel verifier would reject a program we would build with the normal intro and outro | |
6b000af4 | 1430 | but no allow-listing rules (outro would contain an unreachable instruction for successful |
45669ae2 ZJS |
1431 | return). */ |
1432 | policy = CGROUP_DEVICE_POLICY_STRICT; | |
8b139557 ZJS |
1433 | } |
1434 | ||
76dc1725 | 1435 | r = bpf_devices_apply_policy(&prog, policy, any, path, &u->bpf_device_control_installed); |
8b139557 ZJS |
1436 | if (r < 0) { |
1437 | static bool warned = false; | |
1438 | ||
1439 | log_full_errno(warned ? LOG_DEBUG : LOG_WARNING, r, | |
1440 | "Unit %s configures device ACL, but the local system doesn't seem to support the BPF-based device controller.\n" | |
1441 | "Proceeding WITHOUT applying ACL (all devices will be accessible)!\n" | |
1442 | "(This warning is only shown for the first loaded unit using device ACL.)", u->id); | |
1443 | ||
1444 | warned = true; | |
1445 | } | |
1446 | return r; | |
1447 | } | |
1448 | ||
17283ce7 YW |
1449 | static void set_io_weight(Unit *u, uint64_t weight) { |
1450 | char buf[STRLEN("default \n")+DECIMAL_STR_MAX(uint64_t)]; | |
17283ce7 YW |
1451 | |
1452 | assert(u); | |
29eb0eef | 1453 | |
3e6eafdd | 1454 | (void) set_bfq_weight(u, "io", makedev(0, 0), weight); |
29eb0eef | 1455 | |
29eb0eef | 1456 | xsprintf(buf, "default %" PRIu64 "\n", weight); |
17283ce7 YW |
1457 | (void) set_attribute_and_warn(u, "io", "io.weight", buf); |
1458 | } | |
1459 | ||
1460 | static void set_blkio_weight(Unit *u, uint64_t weight) { | |
1461 | char buf[STRLEN("\n")+DECIMAL_STR_MAX(uint64_t)]; | |
17283ce7 YW |
1462 | |
1463 | assert(u); | |
29eb0eef | 1464 | |
3e6eafdd | 1465 | (void) set_bfq_weight(u, "blkio", makedev(0, 0), weight); |
17283ce7 YW |
1466 | |
1467 | xsprintf(buf, "%" PRIu64 "\n", weight); | |
1468 | (void) set_attribute_and_warn(u, "blkio", "blkio.weight", buf); | |
29eb0eef ZJS |
1469 | } |
1470 | ||
506ea51b JK |
1471 | static void cgroup_apply_bpf_foreign_program(Unit *u) { |
1472 | assert(u); | |
1473 | ||
1474 | (void) bpf_foreign_install(u); | |
1475 | } | |
1476 | ||
906c06f6 DM |
1477 | static void cgroup_context_apply( |
1478 | Unit *u, | |
1479 | CGroupMask apply_mask, | |
906c06f6 DM |
1480 | ManagerState state) { |
1481 | ||
f29ff115 TH |
1482 | const char *path; |
1483 | CGroupContext *c; | |
52fecf20 | 1484 | bool is_host_root, is_local_root; |
4ad49000 LP |
1485 | int r; |
1486 | ||
f29ff115 TH |
1487 | assert(u); |
1488 | ||
906c06f6 | 1489 | /* Nothing to do? Exit early! */ |
17f14955 | 1490 | if (apply_mask == 0) |
4ad49000 | 1491 | return; |
8e274523 | 1492 | |
52fecf20 LP |
1493 | /* Some cgroup attributes are not supported on the host root cgroup, hence silently ignore them here. And other |
1494 | * attributes should only be managed for cgroups further down the tree. */ | |
1495 | is_local_root = unit_has_name(u, SPECIAL_ROOT_SLICE); | |
1496 | is_host_root = unit_has_host_root_cgroup(u); | |
f3725e64 LP |
1497 | |
1498 | assert_se(c = unit_get_cgroup_context(u)); | |
1499 | assert_se(path = u->cgroup_path); | |
1500 | ||
52fecf20 | 1501 | if (is_local_root) /* Make sure we don't try to display messages with an empty path. */ |
6da13913 | 1502 | path = "/"; |
01efdf13 | 1503 | |
be2c0327 LP |
1504 | /* We generally ignore errors caused by read-only mounted cgroup trees (assuming we are running in a container |
1505 | * then), and missing cgroups, i.e. EROFS and ENOENT. */ | |
714e2e1d | 1506 | |
be2c0327 LP |
1507 | /* In fully unified mode these attributes don't exist on the host cgroup root. On legacy the weights exist, but |
1508 | * setting the weight makes very little sense on the host root cgroup, as there are no other cgroups at this | |
1509 | * level. The quota exists there too, but any attempt to write to it is refused with EINVAL. Inside of | |
4e1dfa45 | 1510 | * containers we want to leave control of these to the container manager (and if cgroup v2 delegation is used |
be2c0327 LP |
1511 | * we couldn't even write to them if we wanted to). */ |
1512 | if ((apply_mask & CGROUP_MASK_CPU) && !is_local_root) { | |
8e274523 | 1513 | |
b4cccbc1 | 1514 | if (cg_all_unified() > 0) { |
be2c0327 | 1515 | uint64_t weight; |
b2f8b02e | 1516 | |
be2c0327 LP |
1517 | if (cgroup_context_has_cpu_weight(c)) |
1518 | weight = cgroup_context_cpu_weight(c, state); | |
1519 | else if (cgroup_context_has_cpu_shares(c)) { | |
1520 | uint64_t shares; | |
66ebf6c0 | 1521 | |
be2c0327 LP |
1522 | shares = cgroup_context_cpu_shares(c, state); |
1523 | weight = cgroup_cpu_shares_to_weight(shares); | |
66ebf6c0 | 1524 | |
be2c0327 LP |
1525 | log_cgroup_compat(u, "Applying [Startup]CPUShares=%" PRIu64 " as [Startup]CPUWeight=%" PRIu64 " on %s", |
1526 | shares, weight, path); | |
1527 | } else | |
1528 | weight = CGROUP_WEIGHT_DEFAULT; | |
66ebf6c0 | 1529 | |
c8340822 | 1530 | cgroup_apply_unified_cpu_idle(u, weight); |
be2c0327 | 1531 | cgroup_apply_unified_cpu_weight(u, weight); |
10f28641 | 1532 | cgroup_apply_unified_cpu_quota(u, c->cpu_quota_per_sec_usec, c->cpu_quota_period_usec); |
66ebf6c0 | 1533 | |
52fecf20 | 1534 | } else { |
be2c0327 | 1535 | uint64_t shares; |
52fecf20 | 1536 | |
be2c0327 LP |
1537 | if (cgroup_context_has_cpu_weight(c)) { |
1538 | uint64_t weight; | |
52fecf20 | 1539 | |
be2c0327 LP |
1540 | weight = cgroup_context_cpu_weight(c, state); |
1541 | shares = cgroup_cpu_weight_to_shares(weight); | |
52fecf20 | 1542 | |
be2c0327 LP |
1543 | log_cgroup_compat(u, "Applying [Startup]CPUWeight=%" PRIu64 " as [Startup]CPUShares=%" PRIu64 " on %s", |
1544 | weight, shares, path); | |
1545 | } else if (cgroup_context_has_cpu_shares(c)) | |
1546 | shares = cgroup_context_cpu_shares(c, state); | |
1547 | else | |
1548 | shares = CGROUP_CPU_SHARES_DEFAULT; | |
66ebf6c0 | 1549 | |
be2c0327 | 1550 | cgroup_apply_legacy_cpu_shares(u, shares); |
10f28641 | 1551 | cgroup_apply_legacy_cpu_quota(u, c->cpu_quota_per_sec_usec, c->cpu_quota_period_usec); |
66ebf6c0 | 1552 | } |
4ad49000 LP |
1553 | } |
1554 | ||
047f5d63 | 1555 | if ((apply_mask & CGROUP_MASK_CPUSET) && !is_local_root) { |
31d3a520 PM |
1556 | cgroup_apply_unified_cpuset(u, cgroup_context_allowed_cpus(c, state), "cpuset.cpus"); |
1557 | cgroup_apply_unified_cpuset(u, cgroup_context_allowed_mems(c, state), "cpuset.mems"); | |
047f5d63 PH |
1558 | } |
1559 | ||
4e1dfa45 | 1560 | /* The 'io' controller attributes are not exported on the host's root cgroup (being a pure cgroup v2 |
52fecf20 LP |
1561 | * controller), and in case of containers we want to leave control of these attributes to the container manager |
1562 | * (and we couldn't access that stuff anyway, even if we tried if proper delegation is used). */ | |
1563 | if ((apply_mask & CGROUP_MASK_IO) && !is_local_root) { | |
52fecf20 LP |
1564 | bool has_io, has_blockio; |
1565 | uint64_t weight; | |
13c31542 | 1566 | |
52fecf20 LP |
1567 | has_io = cgroup_context_has_io_config(c); |
1568 | has_blockio = cgroup_context_has_blockio_config(c); | |
13c31542 | 1569 | |
52fecf20 LP |
1570 | if (has_io) |
1571 | weight = cgroup_context_io_weight(c, state); | |
1572 | else if (has_blockio) { | |
1573 | uint64_t blkio_weight; | |
128fadc9 | 1574 | |
52fecf20 LP |
1575 | blkio_weight = cgroup_context_blkio_weight(c, state); |
1576 | weight = cgroup_weight_blkio_to_io(blkio_weight); | |
128fadc9 | 1577 | |
67e2ea15 | 1578 | log_cgroup_compat(u, "Applying [Startup]BlockIOWeight=%" PRIu64 " as [Startup]IOWeight=%" PRIu64, |
52fecf20 LP |
1579 | blkio_weight, weight); |
1580 | } else | |
1581 | weight = CGROUP_WEIGHT_DEFAULT; | |
13c31542 | 1582 | |
17283ce7 | 1583 | set_io_weight(u, weight); |
2dbc45ae | 1584 | |
52fecf20 | 1585 | if (has_io) { |
52fecf20 LP |
1586 | LIST_FOREACH(device_weights, w, c->io_device_weights) |
1587 | cgroup_apply_io_device_weight(u, w->path, w->weight); | |
128fadc9 | 1588 | |
52fecf20 LP |
1589 | LIST_FOREACH(device_limits, limit, c->io_device_limits) |
1590 | cgroup_apply_io_device_limit(u, limit->path, limit->limits); | |
6ae4283c | 1591 | |
52fecf20 LP |
1592 | LIST_FOREACH(device_latencies, latency, c->io_device_latencies) |
1593 | cgroup_apply_io_device_latency(u, latency->path, latency->target_usec); | |
6ae4283c | 1594 | |
52fecf20 | 1595 | } else if (has_blockio) { |
52fecf20 LP |
1596 | LIST_FOREACH(device_weights, w, c->blockio_device_weights) { |
1597 | weight = cgroup_weight_blkio_to_io(w->weight); | |
17ae2780 | 1598 | |
67e2ea15 | 1599 | log_cgroup_compat(u, "Applying BlockIODeviceWeight=%" PRIu64 " as IODeviceWeight=%" PRIu64 " for %s", |
52fecf20 | 1600 | w->weight, weight, w->path); |
538b4852 | 1601 | |
52fecf20 LP |
1602 | cgroup_apply_io_device_weight(u, w->path, weight); |
1603 | } | |
538b4852 | 1604 | |
17ae2780 | 1605 | LIST_FOREACH(device_bandwidths, b, c->blockio_device_bandwidths) { |
538b4852 | 1606 | uint64_t limits[_CGROUP_IO_LIMIT_TYPE_MAX]; |
538b4852 | 1607 | |
e8616626 | 1608 | for (CGroupIOLimitType type = 0; type < _CGROUP_IO_LIMIT_TYPE_MAX; type++) |
538b4852 TH |
1609 | limits[type] = cgroup_io_limit_defaults[type]; |
1610 | ||
1611 | limits[CGROUP_IO_RBPS_MAX] = b->rbps; | |
1612 | limits[CGROUP_IO_WBPS_MAX] = b->wbps; | |
1613 | ||
67e2ea15 | 1614 | log_cgroup_compat(u, "Applying BlockIO{Read|Write}Bandwidth=%" PRIu64 " %" PRIu64 " as IO{Read|Write}BandwidthMax= for %s", |
128fadc9 TH |
1615 | b->rbps, b->wbps, b->path); |
1616 | ||
17ae2780 | 1617 | cgroup_apply_io_device_limit(u, b->path, limits); |
538b4852 | 1618 | } |
13c31542 TH |
1619 | } |
1620 | } | |
1621 | ||
906c06f6 | 1622 | if (apply_mask & CGROUP_MASK_BLKIO) { |
52fecf20 | 1623 | bool has_io, has_blockio; |
4ad49000 | 1624 | |
52fecf20 LP |
1625 | has_io = cgroup_context_has_io_config(c); |
1626 | has_blockio = cgroup_context_has_blockio_config(c); | |
1627 | ||
1628 | /* Applying a 'weight' never makes sense for the host root cgroup, and for containers this should be | |
1629 | * left to our container manager, too. */ | |
1630 | if (!is_local_root) { | |
64faf04c | 1631 | uint64_t weight; |
64faf04c | 1632 | |
7d862ab8 | 1633 | if (has_io) { |
52fecf20 | 1634 | uint64_t io_weight; |
128fadc9 | 1635 | |
52fecf20 | 1636 | io_weight = cgroup_context_io_weight(c, state); |
538b4852 | 1637 | weight = cgroup_weight_io_to_blkio(cgroup_context_io_weight(c, state)); |
128fadc9 | 1638 | |
67e2ea15 | 1639 | log_cgroup_compat(u, "Applying [Startup]IOWeight=%" PRIu64 " as [Startup]BlockIOWeight=%" PRIu64, |
128fadc9 | 1640 | io_weight, weight); |
7d862ab8 TH |
1641 | } else if (has_blockio) |
1642 | weight = cgroup_context_blkio_weight(c, state); | |
1643 | else | |
538b4852 | 1644 | weight = CGROUP_BLKIO_WEIGHT_DEFAULT; |
64faf04c | 1645 | |
17283ce7 | 1646 | set_blkio_weight(u, weight); |
35e7a62c | 1647 | |
03677889 | 1648 | if (has_io) |
128fadc9 TH |
1649 | LIST_FOREACH(device_weights, w, c->io_device_weights) { |
1650 | weight = cgroup_weight_io_to_blkio(w->weight); | |
1651 | ||
67e2ea15 | 1652 | log_cgroup_compat(u, "Applying IODeviceWeight=%" PRIu64 " as BlockIODeviceWeight=%" PRIu64 " for %s", |
128fadc9 TH |
1653 | w->weight, weight, w->path); |
1654 | ||
1655 | cgroup_apply_blkio_device_weight(u, w->path, weight); | |
1656 | } | |
03677889 | 1657 | else if (has_blockio) |
7d862ab8 TH |
1658 | LIST_FOREACH(device_weights, w, c->blockio_device_weights) |
1659 | cgroup_apply_blkio_device_weight(u, w->path, w->weight); | |
4ad49000 LP |
1660 | } |
1661 | ||
5238e957 | 1662 | /* The bandwidth limits are something that make sense to be applied to the host's root but not container |
52fecf20 LP |
1663 | * roots, as there we want the container manager to handle it */ |
1664 | if (is_host_root || !is_local_root) { | |
03677889 | 1665 | if (has_io) |
52fecf20 | 1666 | LIST_FOREACH(device_limits, l, c->io_device_limits) { |
67e2ea15 | 1667 | log_cgroup_compat(u, "Applying IO{Read|Write}Bandwidth=%" PRIu64 " %" PRIu64 " as BlockIO{Read|Write}BandwidthMax= for %s", |
52fecf20 | 1668 | l->limits[CGROUP_IO_RBPS_MAX], l->limits[CGROUP_IO_WBPS_MAX], l->path); |
128fadc9 | 1669 | |
52fecf20 LP |
1670 | cgroup_apply_blkio_device_limit(u, l->path, l->limits[CGROUP_IO_RBPS_MAX], l->limits[CGROUP_IO_WBPS_MAX]); |
1671 | } | |
03677889 | 1672 | else if (has_blockio) |
52fecf20 LP |
1673 | LIST_FOREACH(device_bandwidths, b, c->blockio_device_bandwidths) |
1674 | cgroup_apply_blkio_device_limit(u, b->path, b->rbps, b->wbps); | |
d686d8a9 | 1675 | } |
8e274523 LP |
1676 | } |
1677 | ||
be2c0327 LP |
1678 | /* In unified mode 'memory' attributes do not exist on the root cgroup. In legacy mode 'memory.limit_in_bytes' |
1679 | * exists on the root cgroup, but any writes to it are refused with EINVAL. And if we run in a container we | |
4e1dfa45 | 1680 | * want to leave control to the container manager (and if proper cgroup v2 delegation is used we couldn't even |
be2c0327 LP |
1681 | * write to this if we wanted to.) */ |
1682 | if ((apply_mask & CGROUP_MASK_MEMORY) && !is_local_root) { | |
efdb0237 | 1683 | |
52fecf20 | 1684 | if (cg_all_unified() > 0) { |
53fda560 | 1685 | uint64_t max, swap_max = CGROUP_LIMIT_MAX, zswap_max = CGROUP_LIMIT_MAX, high = CGROUP_LIMIT_MAX; |
be2c0327 | 1686 | |
c52db42b | 1687 | if (unit_has_unified_memory_config(u)) { |
53fda560 LB |
1688 | bool startup = IN_SET(state, MANAGER_STARTING, MANAGER_INITIALIZING, MANAGER_STOPPING); |
1689 | ||
1690 | high = startup && c->startup_memory_high_set ? c->startup_memory_high : c->memory_high; | |
1691 | max = startup && c->startup_memory_max_set ? c->startup_memory_max : c->memory_max; | |
1692 | swap_max = startup && c->startup_memory_swap_max_set ? c->startup_memory_swap_max : c->memory_swap_max; | |
1693 | zswap_max = startup && c->startup_memory_zswap_max_set ? c->startup_memory_zswap_max : c->memory_zswap_max; | |
be2c0327 LP |
1694 | } else { |
1695 | max = c->memory_limit; | |
efdb0237 | 1696 | |
be2c0327 LP |
1697 | if (max != CGROUP_LIMIT_MAX) |
1698 | log_cgroup_compat(u, "Applying MemoryLimit=%" PRIu64 " as MemoryMax=", max); | |
128fadc9 | 1699 | } |
da4d897e | 1700 | |
64fe532e | 1701 | cgroup_apply_unified_memory_limit(u, "memory.min", unit_get_ancestor_memory_min(u)); |
c52db42b | 1702 | cgroup_apply_unified_memory_limit(u, "memory.low", unit_get_ancestor_memory_low(u)); |
53fda560 | 1703 | cgroup_apply_unified_memory_limit(u, "memory.high", high); |
be2c0327 LP |
1704 | cgroup_apply_unified_memory_limit(u, "memory.max", max); |
1705 | cgroup_apply_unified_memory_limit(u, "memory.swap.max", swap_max); | |
d7fe0a67 | 1706 | cgroup_apply_unified_memory_limit(u, "memory.zswap.max", zswap_max); |
128fadc9 | 1707 | |
afcfaa69 LP |
1708 | (void) set_attribute_and_warn(u, "memory", "memory.oom.group", one_zero(c->memory_oom_group)); |
1709 | ||
be2c0327 LP |
1710 | } else { |
1711 | char buf[DECIMAL_STR_MAX(uint64_t) + 1]; | |
1712 | uint64_t val; | |
52fecf20 | 1713 | |
c52db42b | 1714 | if (unit_has_unified_memory_config(u)) { |
be2c0327 | 1715 | val = c->memory_max; |
b7a41491 MK |
1716 | if (val != CGROUP_LIMIT_MAX) |
1717 | log_cgroup_compat(u, "Applying MemoryMax=%" PRIu64 " as MemoryLimit=", val); | |
be2c0327 LP |
1718 | } else |
1719 | val = c->memory_limit; | |
78a4ee59 | 1720 | |
be2c0327 LP |
1721 | if (val == CGROUP_LIMIT_MAX) |
1722 | strncpy(buf, "-1\n", sizeof(buf)); | |
1723 | else | |
1724 | xsprintf(buf, "%" PRIu64 "\n", val); | |
1725 | ||
1726 | (void) set_attribute_and_warn(u, "memory", "memory.limit_in_bytes", buf); | |
da4d897e | 1727 | } |
4ad49000 | 1728 | } |
8e274523 | 1729 | |
4e1dfa45 | 1730 | /* On cgroup v2 we can apply BPF everywhere. On cgroup v1 we apply it everywhere except for the root of |
52fecf20 LP |
1731 | * containers, where we leave this to the manager */ |
1732 | if ((apply_mask & (CGROUP_MASK_DEVICES | CGROUP_MASK_BPF_DEVICES)) && | |
8b139557 ZJS |
1733 | (is_host_root || cg_all_unified() > 0 || !is_local_root)) |
1734 | (void) cgroup_apply_devices(u); | |
03a7b521 | 1735 | |
00b5974f LP |
1736 | if (apply_mask & CGROUP_MASK_PIDS) { |
1737 | ||
52fecf20 | 1738 | if (is_host_root) { |
00b5974f LP |
1739 | /* So, the "pids" controller does not expose anything on the root cgroup, in order not to |
1740 | * replicate knobs exposed elsewhere needlessly. We abstract this away here however, and when | |
1741 | * the knobs of the root cgroup are modified propagate this to the relevant sysctls. There's a | |
1742 | * non-obvious asymmetry however: unlike the cgroup properties we don't really want to take | |
1743 | * exclusive ownership of the sysctls, but we still want to honour things if the user sets | |
1744 | * limits. Hence we employ sort of a one-way strategy: when the user sets a bounded limit | |
1745 | * through us it counts. When the user afterwards unsets it again (i.e. sets it to unbounded) | |
1746 | * it also counts. But if the user never set a limit through us (i.e. we are the default of | |
1747 | * "unbounded") we leave things unmodified. For this we manage a global boolean that we turn on | |
1748 | * the first time we set a limit. Note that this boolean is flushed out on manager reload, | |
5238e957 | 1749 | * which is desirable so that there's an official way to release control of the sysctl from |
00b5974f LP |
1750 | * systemd: set the limit to unbounded and reload. */ |
1751 | ||
3a0f06c4 | 1752 | if (tasks_max_isset(&c->tasks_max)) { |
00b5974f | 1753 | u->manager->sysctl_pid_max_changed = true; |
3a0f06c4 | 1754 | r = procfs_tasks_set_limit(tasks_max_resolve(&c->tasks_max)); |
00b5974f LP |
1755 | } else if (u->manager->sysctl_pid_max_changed) |
1756 | r = procfs_tasks_set_limit(TASKS_MAX); | |
1757 | else | |
1758 | r = 0; | |
00b5974f | 1759 | if (r < 0) |
8ed6f81b YW |
1760 | log_unit_full_errno(u, LOG_LEVEL_CGROUP_WRITE(r), r, |
1761 | "Failed to write to tasks limit sysctls: %m"); | |
52fecf20 | 1762 | } |
03a7b521 | 1763 | |
52fecf20 LP |
1764 | /* The attribute itself is not available on the host root cgroup, and in the container case we want to |
1765 | * leave it for the container manager. */ | |
1766 | if (!is_local_root) { | |
3a0f06c4 ZJS |
1767 | if (tasks_max_isset(&c->tasks_max)) { |
1768 | char buf[DECIMAL_STR_MAX(uint64_t) + 1]; | |
03a7b521 | 1769 | |
3a0f06c4 | 1770 | xsprintf(buf, "%" PRIu64 "\n", tasks_max_resolve(&c->tasks_max)); |
293d32df | 1771 | (void) set_attribute_and_warn(u, "pids", "pids.max", buf); |
00b5974f | 1772 | } else |
589a5f7a | 1773 | (void) set_attribute_and_warn(u, "pids", "pids.max", "max\n"); |
00b5974f | 1774 | } |
03a7b521 | 1775 | } |
906c06f6 | 1776 | |
17f14955 | 1777 | if (apply_mask & CGROUP_MASK_BPF_FIREWALL) |
0f2d84d2 | 1778 | cgroup_apply_firewall(u); |
506ea51b JK |
1779 | |
1780 | if (apply_mask & CGROUP_MASK_BPF_FOREIGN) | |
1781 | cgroup_apply_bpf_foreign_program(u); | |
a8e5eb17 JK |
1782 | |
1783 | if (apply_mask & CGROUP_MASK_BPF_SOCKET_BIND) | |
1784 | cgroup_apply_socket_bind(u); | |
6f50d4f7 MV |
1785 | |
1786 | if (apply_mask & CGROUP_MASK_BPF_RESTRICT_NETWORK_INTERFACES) | |
1787 | cgroup_apply_restrict_network_interfaces(u); | |
fb385181 LP |
1788 | } |
1789 | ||
16492445 LP |
1790 | static bool unit_get_needs_bpf_firewall(Unit *u) { |
1791 | CGroupContext *c; | |
16492445 LP |
1792 | assert(u); |
1793 | ||
1794 | c = unit_get_cgroup_context(u); | |
1795 | if (!c) | |
1796 | return false; | |
1797 | ||
1798 | if (c->ip_accounting || | |
84ebe6f0 YW |
1799 | !set_isempty(c->ip_address_allow) || |
1800 | !set_isempty(c->ip_address_deny) || | |
fab34748 KL |
1801 | c->ip_filters_ingress || |
1802 | c->ip_filters_egress) | |
16492445 LP |
1803 | return true; |
1804 | ||
1805 | /* If any parent slice has an IP access list defined, it applies too */ | |
e8616626 | 1806 | for (Unit *p = UNIT_GET_SLICE(u); p; p = UNIT_GET_SLICE(p)) { |
16492445 LP |
1807 | c = unit_get_cgroup_context(p); |
1808 | if (!c) | |
1809 | return false; | |
1810 | ||
84ebe6f0 YW |
1811 | if (!set_isempty(c->ip_address_allow) || |
1812 | !set_isempty(c->ip_address_deny)) | |
16492445 LP |
1813 | return true; |
1814 | } | |
1815 | ||
1816 | return false; | |
1817 | } | |
1818 | ||
506ea51b JK |
1819 | static bool unit_get_needs_bpf_foreign_program(Unit *u) { |
1820 | CGroupContext *c; | |
1821 | assert(u); | |
1822 | ||
1823 | c = unit_get_cgroup_context(u); | |
1824 | if (!c) | |
1825 | return false; | |
1826 | ||
64903d18 | 1827 | return !!c->bpf_foreign_programs; |
506ea51b JK |
1828 | } |
1829 | ||
a8e5eb17 JK |
1830 | static bool unit_get_needs_socket_bind(Unit *u) { |
1831 | CGroupContext *c; | |
1832 | assert(u); | |
1833 | ||
1834 | c = unit_get_cgroup_context(u); | |
1835 | if (!c) | |
1836 | return false; | |
1837 | ||
11ab01e4 | 1838 | return c->socket_bind_allow || c->socket_bind_deny; |
a8e5eb17 JK |
1839 | } |
1840 | ||
6f50d4f7 MV |
1841 | static bool unit_get_needs_restrict_network_interfaces(Unit *u) { |
1842 | CGroupContext *c; | |
1843 | assert(u); | |
1844 | ||
1845 | c = unit_get_cgroup_context(u); | |
1846 | if (!c) | |
1847 | return false; | |
1848 | ||
1849 | return !set_isempty(c->restrict_network_interfaces); | |
1850 | } | |
1851 | ||
c52db42b | 1852 | static CGroupMask unit_get_cgroup_mask(Unit *u) { |
efdb0237 | 1853 | CGroupMask mask = 0; |
c52db42b CD |
1854 | CGroupContext *c; |
1855 | ||
1856 | assert(u); | |
1857 | ||
806a9362 | 1858 | assert_se(c = unit_get_cgroup_context(u)); |
c710d3b4 | 1859 | |
fae9bc29 | 1860 | /* Figure out which controllers we need, based on the cgroup context object */ |
8e274523 | 1861 | |
fae9bc29 | 1862 | if (c->cpu_accounting) |
f98c2585 | 1863 | mask |= get_cpu_accounting_mask(); |
fae9bc29 LP |
1864 | |
1865 | if (cgroup_context_has_cpu_weight(c) || | |
66ebf6c0 | 1866 | cgroup_context_has_cpu_shares(c) || |
3a43da28 | 1867 | c->cpu_quota_per_sec_usec != USEC_INFINITY) |
fae9bc29 | 1868 | mask |= CGROUP_MASK_CPU; |
ecedd90f | 1869 | |
31d3a520 | 1870 | if (cgroup_context_has_allowed_cpus(c) || cgroup_context_has_allowed_mems(c)) |
047f5d63 PH |
1871 | mask |= CGROUP_MASK_CPUSET; |
1872 | ||
538b4852 TH |
1873 | if (cgroup_context_has_io_config(c) || cgroup_context_has_blockio_config(c)) |
1874 | mask |= CGROUP_MASK_IO | CGROUP_MASK_BLKIO; | |
ecedd90f | 1875 | |
4ad49000 | 1876 | if (c->memory_accounting || |
da4d897e | 1877 | c->memory_limit != CGROUP_LIMIT_MAX || |
c52db42b | 1878 | unit_has_unified_memory_config(u)) |
efdb0237 | 1879 | mask |= CGROUP_MASK_MEMORY; |
8e274523 | 1880 | |
a931ad47 | 1881 | if (c->device_allow || |
084870f9 | 1882 | c->device_policy != CGROUP_DEVICE_POLICY_AUTO) |
084c7007 | 1883 | mask |= CGROUP_MASK_DEVICES | CGROUP_MASK_BPF_DEVICES; |
4ad49000 | 1884 | |
03a7b521 | 1885 | if (c->tasks_accounting || |
3a0f06c4 | 1886 | tasks_max_isset(&c->tasks_max)) |
03a7b521 LP |
1887 | mask |= CGROUP_MASK_PIDS; |
1888 | ||
fae9bc29 | 1889 | return CGROUP_MASK_EXTEND_JOINED(mask); |
8e274523 LP |
1890 | } |
1891 | ||
53aea74a | 1892 | static CGroupMask unit_get_bpf_mask(Unit *u) { |
17f14955 RG |
1893 | CGroupMask mask = 0; |
1894 | ||
fae9bc29 LP |
1895 | /* Figure out which controllers we need, based on the cgroup context, possibly taking into account children |
1896 | * too. */ | |
1897 | ||
17f14955 RG |
1898 | if (unit_get_needs_bpf_firewall(u)) |
1899 | mask |= CGROUP_MASK_BPF_FIREWALL; | |
1900 | ||
506ea51b JK |
1901 | if (unit_get_needs_bpf_foreign_program(u)) |
1902 | mask |= CGROUP_MASK_BPF_FOREIGN; | |
1903 | ||
a8e5eb17 JK |
1904 | if (unit_get_needs_socket_bind(u)) |
1905 | mask |= CGROUP_MASK_BPF_SOCKET_BIND; | |
1906 | ||
6f50d4f7 MV |
1907 | if (unit_get_needs_restrict_network_interfaces(u)) |
1908 | mask |= CGROUP_MASK_BPF_RESTRICT_NETWORK_INTERFACES; | |
1909 | ||
17f14955 RG |
1910 | return mask; |
1911 | } | |
1912 | ||
efdb0237 | 1913 | CGroupMask unit_get_own_mask(Unit *u) { |
4ad49000 | 1914 | CGroupContext *c; |
8e274523 | 1915 | |
442ce775 LP |
1916 | /* Returns the mask of controllers the unit needs for itself. If a unit is not properly loaded, return an empty |
1917 | * mask, as we shouldn't reflect it in the cgroup hierarchy then. */ | |
1918 | ||
1919 | if (u->load_state != UNIT_LOADED) | |
1920 | return 0; | |
efdb0237 | 1921 | |
4ad49000 LP |
1922 | c = unit_get_cgroup_context(u); |
1923 | if (!c) | |
1924 | return 0; | |
8e274523 | 1925 | |
12b975e0 | 1926 | return unit_get_cgroup_mask(u) | unit_get_bpf_mask(u) | unit_get_delegate_mask(u); |
02638280 LP |
1927 | } |
1928 | ||
1929 | CGroupMask unit_get_delegate_mask(Unit *u) { | |
1930 | CGroupContext *c; | |
1931 | ||
1932 | /* If delegation is turned on, then turn on selected controllers, unless we are on the legacy hierarchy and the | |
1933 | * process we fork into is known to drop privileges, and hence shouldn't get access to the controllers. | |
19af675e | 1934 | * |
02638280 | 1935 | * Note that on the unified hierarchy it is safe to delegate controllers to unprivileged services. */ |
a931ad47 | 1936 | |
1d9cc876 | 1937 | if (!unit_cgroup_delegate(u)) |
02638280 LP |
1938 | return 0; |
1939 | ||
1940 | if (cg_all_unified() <= 0) { | |
a931ad47 LP |
1941 | ExecContext *e; |
1942 | ||
1943 | e = unit_get_exec_context(u); | |
02638280 LP |
1944 | if (e && !exec_context_maintains_privileges(e)) |
1945 | return 0; | |
a931ad47 LP |
1946 | } |
1947 | ||
1d9cc876 | 1948 | assert_se(c = unit_get_cgroup_context(u)); |
fae9bc29 | 1949 | return CGROUP_MASK_EXTEND_JOINED(c->delegate_controllers); |
8e274523 LP |
1950 | } |
1951 | ||
d9ef5944 MK |
1952 | static CGroupMask unit_get_subtree_mask(Unit *u) { |
1953 | ||
1954 | /* Returns the mask of this subtree, meaning of the group | |
1955 | * itself and its children. */ | |
1956 | ||
1957 | return unit_get_own_mask(u) | unit_get_members_mask(u); | |
1958 | } | |
1959 | ||
efdb0237 | 1960 | CGroupMask unit_get_members_mask(Unit *u) { |
4ad49000 | 1961 | assert(u); |
bc432dc7 | 1962 | |
02638280 | 1963 | /* Returns the mask of controllers all of the unit's children require, merged */ |
efdb0237 | 1964 | |
bc432dc7 | 1965 | if (u->cgroup_members_mask_valid) |
26a17ca2 | 1966 | return u->cgroup_members_mask; /* Use cached value if possible */ |
bc432dc7 | 1967 | |
64e844e5 | 1968 | u->cgroup_members_mask = 0; |
bc432dc7 LP |
1969 | |
1970 | if (u->type == UNIT_SLICE) { | |
1971 | Unit *member; | |
bc432dc7 | 1972 | |
d219a2b0 | 1973 | UNIT_FOREACH_DEPENDENCY(member, u, UNIT_ATOM_SLICE_OF) |
15ed3c3a | 1974 | u->cgroup_members_mask |= unit_get_subtree_mask(member); /* note that this calls ourselves again, for the children */ |
bc432dc7 LP |
1975 | } |
1976 | ||
1977 | u->cgroup_members_mask_valid = true; | |
6414b7c9 | 1978 | return u->cgroup_members_mask; |
246aa6dd LP |
1979 | } |
1980 | ||
efdb0237 | 1981 | CGroupMask unit_get_siblings_mask(Unit *u) { |
12f64221 | 1982 | Unit *slice; |
4ad49000 | 1983 | assert(u); |
246aa6dd | 1984 | |
efdb0237 LP |
1985 | /* Returns the mask of controllers all of the unit's siblings |
1986 | * require, i.e. the members mask of the unit's parent slice | |
1987 | * if there is one. */ | |
1988 | ||
12f64221 LP |
1989 | slice = UNIT_GET_SLICE(u); |
1990 | if (slice) | |
1991 | return unit_get_members_mask(slice); | |
4ad49000 | 1992 | |
64e844e5 | 1993 | return unit_get_subtree_mask(u); /* we are the top-level slice */ |
246aa6dd LP |
1994 | } |
1995 | ||
d9ef5944 | 1996 | static CGroupMask unit_get_disable_mask(Unit *u) { |
4f6f62e4 CD |
1997 | CGroupContext *c; |
1998 | ||
1999 | c = unit_get_cgroup_context(u); | |
2000 | if (!c) | |
2001 | return 0; | |
2002 | ||
2003 | return c->disable_controllers; | |
2004 | } | |
2005 | ||
2006 | CGroupMask unit_get_ancestor_disable_mask(Unit *u) { | |
2007 | CGroupMask mask; | |
12f64221 | 2008 | Unit *slice; |
4f6f62e4 CD |
2009 | |
2010 | assert(u); | |
2011 | mask = unit_get_disable_mask(u); | |
2012 | ||
2013 | /* Returns the mask of controllers which are marked as forcibly | |
2014 | * disabled in any ancestor unit or the unit in question. */ | |
2015 | ||
12f64221 LP |
2016 | slice = UNIT_GET_SLICE(u); |
2017 | if (slice) | |
2018 | mask |= unit_get_ancestor_disable_mask(slice); | |
4f6f62e4 CD |
2019 | |
2020 | return mask; | |
2021 | } | |
2022 | ||
efdb0237 | 2023 | CGroupMask unit_get_target_mask(Unit *u) { |
a437c5e4 | 2024 | CGroupMask own_mask, mask; |
efdb0237 | 2025 | |
a437c5e4 LP |
2026 | /* This returns the cgroup mask of all controllers to enable for a specific cgroup, i.e. everything |
2027 | * it needs itself, plus all that its children need, plus all that its siblings need. This is | |
2028 | * primarily useful on the legacy cgroup hierarchy, where we need to duplicate each cgroup in each | |
efdb0237 | 2029 | * hierarchy that shall be enabled for it. */ |
6414b7c9 | 2030 | |
a437c5e4 | 2031 | own_mask = unit_get_own_mask(u); |
84d2744b | 2032 | |
a437c5e4 | 2033 | if (own_mask & CGROUP_MASK_BPF_FIREWALL & ~u->manager->cgroup_supported) |
84d2744b ZJS |
2034 | emit_bpf_firewall_warning(u); |
2035 | ||
a437c5e4 LP |
2036 | mask = own_mask | unit_get_members_mask(u) | unit_get_siblings_mask(u); |
2037 | ||
efdb0237 | 2038 | mask &= u->manager->cgroup_supported; |
c72703e2 | 2039 | mask &= ~unit_get_ancestor_disable_mask(u); |
efdb0237 LP |
2040 | |
2041 | return mask; | |
2042 | } | |
2043 | ||
2044 | CGroupMask unit_get_enable_mask(Unit *u) { | |
2045 | CGroupMask mask; | |
2046 | ||
2047 | /* This returns the cgroup mask of all controllers to enable | |
2048 | * for the children of a specific cgroup. This is primarily | |
2049 | * useful for the unified cgroup hierarchy, where each cgroup | |
2050 | * controls which controllers are enabled for its children. */ | |
2051 | ||
2052 | mask = unit_get_members_mask(u); | |
6414b7c9 | 2053 | mask &= u->manager->cgroup_supported; |
c72703e2 | 2054 | mask &= ~unit_get_ancestor_disable_mask(u); |
6414b7c9 DS |
2055 | |
2056 | return mask; | |
2057 | } | |
2058 | ||
5af88058 | 2059 | void unit_invalidate_cgroup_members_masks(Unit *u) { |
12f64221 LP |
2060 | Unit *slice; |
2061 | ||
bc432dc7 LP |
2062 | assert(u); |
2063 | ||
5af88058 LP |
2064 | /* Recurse invalidate the member masks cache all the way up the tree */ |
2065 | u->cgroup_members_mask_valid = false; | |
bc432dc7 | 2066 | |
12f64221 LP |
2067 | slice = UNIT_GET_SLICE(u); |
2068 | if (slice) | |
2069 | unit_invalidate_cgroup_members_masks(slice); | |
6414b7c9 DS |
2070 | } |
2071 | ||
6592b975 | 2072 | const char *unit_get_realized_cgroup_path(Unit *u, CGroupMask mask) { |
03b90d4b | 2073 | |
6592b975 | 2074 | /* Returns the realized cgroup path of the specified unit where all specified controllers are available. */ |
03b90d4b LP |
2075 | |
2076 | while (u) { | |
6592b975 | 2077 | |
03b90d4b LP |
2078 | if (u->cgroup_path && |
2079 | u->cgroup_realized && | |
d94a24ca | 2080 | FLAGS_SET(u->cgroup_realized_mask, mask)) |
03b90d4b LP |
2081 | return u->cgroup_path; |
2082 | ||
12f64221 | 2083 | u = UNIT_GET_SLICE(u); |
03b90d4b LP |
2084 | } |
2085 | ||
2086 | return NULL; | |
2087 | } | |
2088 | ||
6592b975 | 2089 | static const char *migrate_callback(CGroupMask mask, void *userdata) { |
7b639614 MK |
2090 | /* If not realized at all, migrate to root (""). |
2091 | * It may happen if we're upgrading from older version that didn't clean up. | |
2092 | */ | |
2093 | return strempty(unit_get_realized_cgroup_path(userdata, mask)); | |
6592b975 LP |
2094 | } |
2095 | ||
1a56b0c0 LP |
2096 | int unit_default_cgroup_path(const Unit *u, char **ret) { |
2097 | _cleanup_free_ char *p = NULL; | |
efdb0237 LP |
2098 | int r; |
2099 | ||
2100 | assert(u); | |
1a56b0c0 | 2101 | assert(ret); |
efdb0237 LP |
2102 | |
2103 | if (unit_has_name(u, SPECIAL_ROOT_SLICE)) | |
1a56b0c0 LP |
2104 | p = strdup(u->manager->cgroup_root); |
2105 | else { | |
2106 | _cleanup_free_ char *escaped = NULL, *slice_path = NULL; | |
2107 | Unit *slice; | |
efdb0237 | 2108 | |
1a56b0c0 LP |
2109 | slice = UNIT_GET_SLICE(u); |
2110 | if (slice && !unit_has_name(slice, SPECIAL_ROOT_SLICE)) { | |
2111 | r = cg_slice_to_path(slice->id, &slice_path); | |
2112 | if (r < 0) | |
2113 | return r; | |
2114 | } | |
2115 | ||
2116 | r = cg_escape(u->id, &escaped); | |
efdb0237 | 2117 | if (r < 0) |
1a56b0c0 | 2118 | return r; |
efdb0237 | 2119 | |
1a56b0c0 LP |
2120 | p = path_join(empty_to_root(u->manager->cgroup_root), slice_path, escaped); |
2121 | } | |
2122 | if (!p) | |
2123 | return -ENOMEM; | |
efdb0237 | 2124 | |
1a56b0c0 LP |
2125 | *ret = TAKE_PTR(p); |
2126 | return 0; | |
efdb0237 LP |
2127 | } |
2128 | ||
2129 | int unit_set_cgroup_path(Unit *u, const char *path) { | |
2130 | _cleanup_free_ char *p = NULL; | |
2131 | int r; | |
2132 | ||
2133 | assert(u); | |
2134 | ||
5210387e LP |
2135 | if (streq_ptr(u->cgroup_path, path)) |
2136 | return 0; | |
2137 | ||
efdb0237 LP |
2138 | if (path) { |
2139 | p = strdup(path); | |
2140 | if (!p) | |
2141 | return -ENOMEM; | |
5210387e | 2142 | } |
efdb0237 LP |
2143 | |
2144 | if (p) { | |
2145 | r = hashmap_put(u->manager->cgroup_unit, p, u); | |
2146 | if (r < 0) | |
2147 | return r; | |
2148 | } | |
2149 | ||
2150 | unit_release_cgroup(u); | |
ae2a15bc | 2151 | u->cgroup_path = TAKE_PTR(p); |
efdb0237 LP |
2152 | |
2153 | return 1; | |
2154 | } | |
2155 | ||
2156 | int unit_watch_cgroup(Unit *u) { | |
ab2c3861 | 2157 | _cleanup_free_ char *events = NULL; |
efdb0237 LP |
2158 | int r; |
2159 | ||
2160 | assert(u); | |
2161 | ||
0bb814c2 LP |
2162 | /* Watches the "cgroups.events" attribute of this unit's cgroup for "empty" events, but only if |
2163 | * cgroupv2 is available. */ | |
2164 | ||
efdb0237 LP |
2165 | if (!u->cgroup_path) |
2166 | return 0; | |
2167 | ||
0bb814c2 | 2168 | if (u->cgroup_control_inotify_wd >= 0) |
efdb0237 LP |
2169 | return 0; |
2170 | ||
2171 | /* Only applies to the unified hierarchy */ | |
c22800e4 | 2172 | r = cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER); |
b4cccbc1 LP |
2173 | if (r < 0) |
2174 | return log_error_errno(r, "Failed to determine whether the name=systemd hierarchy is unified: %m"); | |
2175 | if (r == 0) | |
efdb0237 LP |
2176 | return 0; |
2177 | ||
0bb814c2 | 2178 | /* No point in watch the top-level slice, it's never going to run empty. */ |
efdb0237 LP |
2179 | if (unit_has_name(u, SPECIAL_ROOT_SLICE)) |
2180 | return 0; | |
2181 | ||
0bb814c2 | 2182 | r = hashmap_ensure_allocated(&u->manager->cgroup_control_inotify_wd_unit, &trivial_hash_ops); |
efdb0237 LP |
2183 | if (r < 0) |
2184 | return log_oom(); | |
2185 | ||
ab2c3861 | 2186 | r = cg_get_path(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path, "cgroup.events", &events); |
efdb0237 LP |
2187 | if (r < 0) |
2188 | return log_oom(); | |
2189 | ||
0bb814c2 LP |
2190 | u->cgroup_control_inotify_wd = inotify_add_watch(u->manager->cgroup_inotify_fd, events, IN_MODIFY); |
2191 | if (u->cgroup_control_inotify_wd < 0) { | |
efdb0237 | 2192 | |
0bb814c2 LP |
2193 | if (errno == ENOENT) /* If the directory is already gone we don't need to track it, so this |
2194 | * is not an error */ | |
efdb0237 LP |
2195 | return 0; |
2196 | ||
6178e2f8 | 2197 | return log_unit_error_errno(u, errno, "Failed to add control inotify watch descriptor for control group %s: %m", empty_to_root(u->cgroup_path)); |
efdb0237 LP |
2198 | } |
2199 | ||
0bb814c2 | 2200 | r = hashmap_put(u->manager->cgroup_control_inotify_wd_unit, INT_TO_PTR(u->cgroup_control_inotify_wd), u); |
efdb0237 | 2201 | if (r < 0) |
6178e2f8 | 2202 | return log_unit_error_errno(u, r, "Failed to add control inotify watch descriptor for control group %s to hash map: %m", empty_to_root(u->cgroup_path)); |
efdb0237 LP |
2203 | |
2204 | return 0; | |
2205 | } | |
2206 | ||
afcfaa69 LP |
2207 | int unit_watch_cgroup_memory(Unit *u) { |
2208 | _cleanup_free_ char *events = NULL; | |
2209 | CGroupContext *c; | |
2210 | int r; | |
2211 | ||
2212 | assert(u); | |
2213 | ||
2214 | /* Watches the "memory.events" attribute of this unit's cgroup for "oom_kill" events, but only if | |
2215 | * cgroupv2 is available. */ | |
2216 | ||
2217 | if (!u->cgroup_path) | |
2218 | return 0; | |
2219 | ||
2220 | c = unit_get_cgroup_context(u); | |
2221 | if (!c) | |
2222 | return 0; | |
2223 | ||
2224 | /* The "memory.events" attribute is only available if the memory controller is on. Let's hence tie | |
2225 | * this to memory accounting, in a way watching for OOM kills is a form of memory accounting after | |
2226 | * all. */ | |
2227 | if (!c->memory_accounting) | |
2228 | return 0; | |
2229 | ||
2230 | /* Don't watch inner nodes, as the kernel doesn't report oom_kill events recursively currently, and | |
2231 | * we also don't want to generate a log message for each parent cgroup of a process. */ | |
2232 | if (u->type == UNIT_SLICE) | |
2233 | return 0; | |
2234 | ||
2235 | if (u->cgroup_memory_inotify_wd >= 0) | |
2236 | return 0; | |
2237 | ||
2238 | /* Only applies to the unified hierarchy */ | |
2239 | r = cg_all_unified(); | |
2240 | if (r < 0) | |
2241 | return log_error_errno(r, "Failed to determine whether the memory controller is unified: %m"); | |
2242 | if (r == 0) | |
2243 | return 0; | |
2244 | ||
2245 | r = hashmap_ensure_allocated(&u->manager->cgroup_memory_inotify_wd_unit, &trivial_hash_ops); | |
2246 | if (r < 0) | |
2247 | return log_oom(); | |
2248 | ||
2249 | r = cg_get_path(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path, "memory.events", &events); | |
2250 | if (r < 0) | |
2251 | return log_oom(); | |
2252 | ||
2253 | u->cgroup_memory_inotify_wd = inotify_add_watch(u->manager->cgroup_inotify_fd, events, IN_MODIFY); | |
2254 | if (u->cgroup_memory_inotify_wd < 0) { | |
2255 | ||
2256 | if (errno == ENOENT) /* If the directory is already gone we don't need to track it, so this | |
2257 | * is not an error */ | |
2258 | return 0; | |
2259 | ||
6178e2f8 | 2260 | return log_unit_error_errno(u, errno, "Failed to add memory inotify watch descriptor for control group %s: %m", empty_to_root(u->cgroup_path)); |
afcfaa69 LP |
2261 | } |
2262 | ||
2263 | r = hashmap_put(u->manager->cgroup_memory_inotify_wd_unit, INT_TO_PTR(u->cgroup_memory_inotify_wd), u); | |
2264 | if (r < 0) | |
6178e2f8 | 2265 | return log_unit_error_errno(u, r, "Failed to add memory inotify watch descriptor for control group %s to hash map: %m", empty_to_root(u->cgroup_path)); |
afcfaa69 LP |
2266 | |
2267 | return 0; | |
2268 | } | |
2269 | ||
a4634b21 LP |
2270 | int unit_pick_cgroup_path(Unit *u) { |
2271 | _cleanup_free_ char *path = NULL; | |
2272 | int r; | |
2273 | ||
2274 | assert(u); | |
2275 | ||
2276 | if (u->cgroup_path) | |
2277 | return 0; | |
2278 | ||
2279 | if (!UNIT_HAS_CGROUP_CONTEXT(u)) | |
2280 | return -EINVAL; | |
2281 | ||
1a56b0c0 LP |
2282 | r = unit_default_cgroup_path(u, &path); |
2283 | if (r < 0) | |
2284 | return log_unit_error_errno(u, r, "Failed to generate default cgroup path: %m"); | |
a4634b21 LP |
2285 | |
2286 | r = unit_set_cgroup_path(u, path); | |
2287 | if (r == -EEXIST) | |
6178e2f8 | 2288 | return log_unit_error_errno(u, r, "Control group %s exists already.", empty_to_root(path)); |
a4634b21 | 2289 | if (r < 0) |
6178e2f8 | 2290 | return log_unit_error_errno(u, r, "Failed to set unit's control group path to %s: %m", empty_to_root(path)); |
a4634b21 LP |
2291 | |
2292 | return 0; | |
2293 | } | |
2294 | ||
7b639614 | 2295 | static int unit_update_cgroup( |
efdb0237 LP |
2296 | Unit *u, |
2297 | CGroupMask target_mask, | |
0d2d6fbf CD |
2298 | CGroupMask enable_mask, |
2299 | ManagerState state) { | |
efdb0237 | 2300 | |
7b639614 MK |
2301 | bool created, is_root_slice; |
2302 | CGroupMask migrate_mask = 0; | |
184b4f78 | 2303 | _cleanup_free_ char *cgroup_full_path = NULL; |
27adcc97 | 2304 | int r; |
64747e2d | 2305 | |
4ad49000 | 2306 | assert(u); |
64747e2d | 2307 | |
27c4ed79 | 2308 | if (!UNIT_HAS_CGROUP_CONTEXT(u)) |
0cd385d3 LP |
2309 | return 0; |
2310 | ||
a4634b21 LP |
2311 | /* Figure out our cgroup path */ |
2312 | r = unit_pick_cgroup_path(u); | |
2313 | if (r < 0) | |
2314 | return r; | |
b58b8e11 | 2315 | |
03b90d4b | 2316 | /* First, create our own group */ |
efdb0237 | 2317 | r = cg_create_everywhere(u->manager->cgroup_supported, target_mask, u->cgroup_path); |
23bbb0de | 2318 | if (r < 0) |
6178e2f8 | 2319 | return log_unit_error_errno(u, r, "Failed to create cgroup %s: %m", empty_to_root(u->cgroup_path)); |
490c5a37 | 2320 | created = r; |
efdb0237 | 2321 | |
184b4f78 | 2322 | if (cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER) > 0) { |
1b420223 LP |
2323 | uint64_t cgroup_id = 0; |
2324 | ||
184b4f78 ILG |
2325 | r = cg_get_path(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path, NULL, &cgroup_full_path); |
2326 | if (r == 0) { | |
2327 | r = cg_path_get_cgroupid(cgroup_full_path, &cgroup_id); | |
2328 | if (r < 0) | |
1b420223 LP |
2329 | log_unit_full_errno(u, ERRNO_IS_NOT_SUPPORTED(r) ? LOG_DEBUG : LOG_WARNING, r, |
2330 | "Failed to get cgroup ID of cgroup %s, ignoring: %m", cgroup_full_path); | |
184b4f78 ILG |
2331 | } else |
2332 | log_unit_warning_errno(u, r, "Failed to get full cgroup path on cgroup %s, ignoring: %m", empty_to_root(u->cgroup_path)); | |
2333 | ||
2334 | u->cgroup_id = cgroup_id; | |
2335 | } | |
2336 | ||
efdb0237 LP |
2337 | /* Start watching it */ |
2338 | (void) unit_watch_cgroup(u); | |
afcfaa69 | 2339 | (void) unit_watch_cgroup_memory(u); |
efdb0237 | 2340 | |
7b639614 MK |
2341 | /* For v2 we preserve enabled controllers in delegated units, adjust others, |
2342 | * for v1 we figure out which controller hierarchies need migration. */ | |
1fd3a10c | 2343 | if (created || !u->cgroup_realized || !unit_cgroup_delegate(u)) { |
27adcc97 | 2344 | CGroupMask result_mask = 0; |
65be7e06 ZJS |
2345 | |
2346 | /* Enable all controllers we need */ | |
27adcc97 | 2347 | r = cg_enable_everywhere(u->manager->cgroup_supported, enable_mask, u->cgroup_path, &result_mask); |
65be7e06 | 2348 | if (r < 0) |
6178e2f8 | 2349 | log_unit_warning_errno(u, r, "Failed to enable/disable controllers on cgroup %s, ignoring: %m", empty_to_root(u->cgroup_path)); |
27adcc97 | 2350 | |
27adcc97 LP |
2351 | /* Remember what's actually enabled now */ |
2352 | u->cgroup_enabled_mask = result_mask; | |
7b639614 MK |
2353 | |
2354 | migrate_mask = u->cgroup_realized_mask ^ target_mask; | |
65be7e06 | 2355 | } |
03b90d4b LP |
2356 | |
2357 | /* Keep track that this is now realized */ | |
4ad49000 | 2358 | u->cgroup_realized = true; |
efdb0237 | 2359 | u->cgroup_realized_mask = target_mask; |
4ad49000 | 2360 | |
7b639614 MK |
2361 | /* Migrate processes in controller hierarchies both downwards (enabling) and upwards (disabling). |
2362 | * | |
2363 | * Unnecessary controller cgroups are trimmed (after emptied by upward migration). | |
2364 | * We perform migration also with whole slices for cases when users don't care about leave | |
2365 | * granularity. Since delegated_mask is subset of target mask, we won't trim slice subtree containing | |
2366 | * delegated units. | |
2367 | */ | |
2368 | if (cg_all_unified() == 0) { | |
2369 | r = cg_migrate_v1_controllers(u->manager->cgroup_supported, migrate_mask, u->cgroup_path, migrate_callback, u); | |
2370 | if (r < 0) | |
6178e2f8 | 2371 | log_unit_warning_errno(u, r, "Failed to migrate controller cgroups from %s, ignoring: %m", empty_to_root(u->cgroup_path)); |
0cd385d3 | 2372 | |
7b639614 MK |
2373 | is_root_slice = unit_has_name(u, SPECIAL_ROOT_SLICE); |
2374 | r = cg_trim_v1_controllers(u->manager->cgroup_supported, ~target_mask, u->cgroup_path, !is_root_slice); | |
0cd385d3 | 2375 | if (r < 0) |
6178e2f8 | 2376 | log_unit_warning_errno(u, r, "Failed to delete controller cgroups %s, ignoring: %m", empty_to_root(u->cgroup_path)); |
0cd385d3 | 2377 | } |
03b90d4b | 2378 | |
0d2d6fbf CD |
2379 | /* Set attributes */ |
2380 | cgroup_context_apply(u, target_mask, state); | |
2381 | cgroup_xattr_apply(u); | |
2382 | ||
29e6b0c1 LP |
2383 | /* For most units we expect that memory monitoring is set up before the unit is started and we won't |
2384 | * touch it after. For PID 1 this is different though, because we couldn't possibly do that given | |
2385 | * that PID 1 runs before init.scope is even set up. Hence, whenever init.scope is realized, let's | |
2386 | * try to open the memory pressure interface anew. */ | |
2387 | if (unit_has_name(u, SPECIAL_INIT_SCOPE)) | |
2388 | (void) manager_setup_memory_pressure_event_source(u->manager); | |
2389 | ||
64747e2d LP |
2390 | return 0; |
2391 | } | |
2392 | ||
6592b975 LP |
2393 | static int unit_attach_pid_to_cgroup_via_bus(Unit *u, pid_t pid, const char *suffix_path) { |
2394 | _cleanup_(sd_bus_error_free) sd_bus_error error = SD_BUS_ERROR_NULL; | |
2395 | char *pp; | |
7b3fd631 | 2396 | int r; |
6592b975 | 2397 | |
7b3fd631 LP |
2398 | assert(u); |
2399 | ||
6592b975 LP |
2400 | if (MANAGER_IS_SYSTEM(u->manager)) |
2401 | return -EINVAL; | |
2402 | ||
2403 | if (!u->manager->system_bus) | |
2404 | return -EIO; | |
2405 | ||
2406 | if (!u->cgroup_path) | |
2407 | return -EINVAL; | |
2408 | ||
2409 | /* Determine this unit's cgroup path relative to our cgroup root */ | |
2410 | pp = path_startswith(u->cgroup_path, u->manager->cgroup_root); | |
2411 | if (!pp) | |
2412 | return -EINVAL; | |
2413 | ||
2414 | pp = strjoina("/", pp, suffix_path); | |
4ff361cc | 2415 | path_simplify(pp); |
6592b975 | 2416 | |
78fa2f91 | 2417 | r = bus_call_method(u->manager->system_bus, |
2418 | bus_systemd_mgr, | |
2419 | "AttachProcessesToUnit", | |
2420 | &error, NULL, | |
2421 | "ssau", | |
2422 | NULL /* empty unit name means client's unit, i.e. us */, pp, 1, (uint32_t) pid); | |
7b3fd631 | 2423 | if (r < 0) |
6592b975 LP |
2424 | return log_unit_debug_errno(u, r, "Failed to attach unit process " PID_FMT " via the bus: %s", pid, bus_error_message(&error, r)); |
2425 | ||
2426 | return 0; | |
2427 | } | |
2428 | ||
2429 | int unit_attach_pids_to_cgroup(Unit *u, Set *pids, const char *suffix_path) { | |
8e7e4a73 | 2430 | _cleanup_free_ char *joined = NULL; |
6592b975 LP |
2431 | CGroupMask delegated_mask; |
2432 | const char *p; | |
6592b975 | 2433 | void *pidp; |
db4229d1 | 2434 | int ret, r; |
6592b975 LP |
2435 | |
2436 | assert(u); | |
2437 | ||
2438 | if (!UNIT_HAS_CGROUP_CONTEXT(u)) | |
2439 | return -EINVAL; | |
2440 | ||
2441 | if (set_isempty(pids)) | |
2442 | return 0; | |
7b3fd631 | 2443 | |
fab34748 KL |
2444 | /* Load any custom firewall BPF programs here once to test if they are existing and actually loadable. |
2445 | * Fail here early since later errors in the call chain unit_realize_cgroup to cgroup_context_apply are ignored. */ | |
2446 | r = bpf_firewall_load_custom(u); | |
2447 | if (r < 0) | |
2448 | return r; | |
2449 | ||
6592b975 | 2450 | r = unit_realize_cgroup(u); |
7b3fd631 LP |
2451 | if (r < 0) |
2452 | return r; | |
2453 | ||
6592b975 LP |
2454 | if (isempty(suffix_path)) |
2455 | p = u->cgroup_path; | |
8e7e4a73 LP |
2456 | else { |
2457 | joined = path_join(u->cgroup_path, suffix_path); | |
2458 | if (!joined) | |
2459 | return -ENOMEM; | |
2460 | ||
2461 | p = joined; | |
2462 | } | |
6592b975 LP |
2463 | |
2464 | delegated_mask = unit_get_delegate_mask(u); | |
2465 | ||
db4229d1 | 2466 | ret = 0; |
90e74a66 | 2467 | SET_FOREACH(pidp, pids) { |
6592b975 | 2468 | pid_t pid = PTR_TO_PID(pidp); |
6592b975 LP |
2469 | |
2470 | /* First, attach the PID to the main cgroup hierarchy */ | |
db4229d1 LP |
2471 | r = cg_attach(SYSTEMD_CGROUP_CONTROLLER, p, pid); |
2472 | if (r < 0) { | |
2473 | bool again = MANAGER_IS_USER(u->manager) && ERRNO_IS_PRIVILEGE(r); | |
6592b975 | 2474 | |
db4229d1 | 2475 | log_unit_full_errno(u, again ? LOG_DEBUG : LOG_INFO, r, |
7a2ba407 | 2476 | "Couldn't move process "PID_FMT" to%s requested cgroup '%s': %m", |
6178e2f8 | 2477 | pid, again ? " directly" : "", empty_to_root(p)); |
7a2ba407 ZJS |
2478 | |
2479 | if (again) { | |
6592b975 LP |
2480 | int z; |
2481 | ||
7a2ba407 ZJS |
2482 | /* If we are in a user instance, and we can't move the process ourselves due |
2483 | * to permission problems, let's ask the system instance about it instead. | |
2484 | * Since it's more privileged it might be able to move the process across the | |
2485 | * leaves of a subtree whose top node is not owned by us. */ | |
6592b975 LP |
2486 | |
2487 | z = unit_attach_pid_to_cgroup_via_bus(u, pid, suffix_path); | |
2488 | if (z < 0) | |
6178e2f8 | 2489 | log_unit_info_errno(u, z, "Couldn't move process "PID_FMT" to requested cgroup '%s' (directly or via the system bus): %m", pid, empty_to_root(p)); |
c65417a0 JW |
2490 | else { |
2491 | if (ret >= 0) | |
2492 | ret++; /* Count successful additions */ | |
6592b975 | 2493 | continue; /* When the bus thing worked via the bus we are fully done for this PID. */ |
c65417a0 | 2494 | } |
6592b975 LP |
2495 | } |
2496 | ||
db4229d1 LP |
2497 | if (ret >= 0) |
2498 | ret = r; /* Remember first error */ | |
6592b975 LP |
2499 | |
2500 | continue; | |
8d3e4ac7 LP |
2501 | } else if (ret >= 0) |
2502 | ret++; /* Count successful additions */ | |
6592b975 | 2503 | |
db4229d1 LP |
2504 | r = cg_all_unified(); |
2505 | if (r < 0) | |
2506 | return r; | |
2507 | if (r > 0) | |
6592b975 LP |
2508 | continue; |
2509 | ||
2510 | /* In the legacy hierarchy, attach the process to the request cgroup if possible, and if not to the | |
2511 | * innermost realized one */ | |
2512 | ||
e8616626 | 2513 | for (CGroupController c = 0; c < _CGROUP_CONTROLLER_MAX; c++) { |
6592b975 LP |
2514 | CGroupMask bit = CGROUP_CONTROLLER_TO_MASK(c); |
2515 | const char *realized; | |
2516 | ||
2517 | if (!(u->manager->cgroup_supported & bit)) | |
2518 | continue; | |
2519 | ||
2520 | /* If this controller is delegated and realized, honour the caller's request for the cgroup suffix. */ | |
2521 | if (delegated_mask & u->cgroup_realized_mask & bit) { | |
db4229d1 LP |
2522 | r = cg_attach(cgroup_controller_to_string(c), p, pid); |
2523 | if (r >= 0) | |
6592b975 LP |
2524 | continue; /* Success! */ |
2525 | ||
db4229d1 | 2526 | log_unit_debug_errno(u, r, "Failed to attach PID " PID_FMT " to requested cgroup %s in controller %s, falling back to unit's cgroup: %m", |
6178e2f8 | 2527 | pid, empty_to_root(p), cgroup_controller_to_string(c)); |
6592b975 LP |
2528 | } |
2529 | ||
2530 | /* So this controller is either not delegate or realized, or something else weird happened. In | |
2531 | * that case let's attach the PID at least to the closest cgroup up the tree that is | |
2532 | * realized. */ | |
2533 | realized = unit_get_realized_cgroup_path(u, bit); | |
2534 | if (!realized) | |
2535 | continue; /* Not even realized in the root slice? Then let's not bother */ | |
2536 | ||
db4229d1 LP |
2537 | r = cg_attach(cgroup_controller_to_string(c), realized, pid); |
2538 | if (r < 0) | |
2539 | log_unit_debug_errno(u, r, "Failed to attach PID " PID_FMT " to realized cgroup %s in controller %s, ignoring: %m", | |
6592b975 LP |
2540 | pid, realized, cgroup_controller_to_string(c)); |
2541 | } | |
2542 | } | |
2543 | ||
db4229d1 | 2544 | return ret; |
7b3fd631 LP |
2545 | } |
2546 | ||
906c06f6 DM |
2547 | static bool unit_has_mask_realized( |
2548 | Unit *u, | |
2549 | CGroupMask target_mask, | |
17f14955 | 2550 | CGroupMask enable_mask) { |
906c06f6 | 2551 | |
bc432dc7 LP |
2552 | assert(u); |
2553 | ||
d5095dcd LP |
2554 | /* Returns true if this unit is fully realized. We check four things: |
2555 | * | |
2556 | * 1. Whether the cgroup was created at all | |
4e1dfa45 CD |
2557 | * 2. Whether the cgroup was created in all the hierarchies we need it to be created in (in case of cgroup v1) |
2558 | * 3. Whether the cgroup has all the right controllers enabled (in case of cgroup v2) | |
d5095dcd LP |
2559 | * 4. Whether the invalidation mask is currently zero |
2560 | * | |
2561 | * If you wonder why we mask the target realization and enable mask with CGROUP_MASK_V1/CGROUP_MASK_V2: note | |
4e1dfa45 CD |
2562 | * that there are three sets of bitmasks: CGROUP_MASK_V1 (for real cgroup v1 controllers), CGROUP_MASK_V2 (for |
2563 | * real cgroup v2 controllers) and CGROUP_MASK_BPF (for BPF-based pseudo-controllers). Now, cgroup_realized_mask | |
2564 | * is only matters for cgroup v1 controllers, and cgroup_enabled_mask only used for cgroup v2, and if they | |
d5095dcd LP |
2565 | * differ in the others, we don't really care. (After all, the cgroup_enabled_mask tracks with controllers are |
2566 | * enabled through cgroup.subtree_control, and since the BPF pseudo-controllers don't show up there, they | |
2567 | * simply don't matter. */ | |
2568 | ||
906c06f6 | 2569 | return u->cgroup_realized && |
d5095dcd LP |
2570 | ((u->cgroup_realized_mask ^ target_mask) & CGROUP_MASK_V1) == 0 && |
2571 | ((u->cgroup_enabled_mask ^ enable_mask) & CGROUP_MASK_V2) == 0 && | |
17f14955 | 2572 | u->cgroup_invalidated_mask == 0; |
6414b7c9 DS |
2573 | } |
2574 | ||
4f6f62e4 CD |
2575 | static bool unit_has_mask_disables_realized( |
2576 | Unit *u, | |
2577 | CGroupMask target_mask, | |
2578 | CGroupMask enable_mask) { | |
2579 | ||
2580 | assert(u); | |
2581 | ||
2582 | /* Returns true if all controllers which should be disabled are indeed disabled. | |
2583 | * | |
2584 | * Unlike unit_has_mask_realized, we don't care what was enabled, only that anything we want to remove is | |
2585 | * already removed. */ | |
2586 | ||
2587 | return !u->cgroup_realized || | |
2588 | (FLAGS_SET(u->cgroup_realized_mask, target_mask & CGROUP_MASK_V1) && | |
2589 | FLAGS_SET(u->cgroup_enabled_mask, enable_mask & CGROUP_MASK_V2)); | |
2590 | } | |
2591 | ||
a57669d2 CD |
2592 | static bool unit_has_mask_enables_realized( |
2593 | Unit *u, | |
2594 | CGroupMask target_mask, | |
2595 | CGroupMask enable_mask) { | |
2596 | ||
2597 | assert(u); | |
2598 | ||
2599 | /* Returns true if all controllers which should be enabled are indeed enabled. | |
2600 | * | |
2601 | * Unlike unit_has_mask_realized, we don't care about the controllers that are not present, only that anything | |
2602 | * we want to add is already added. */ | |
2603 | ||
2604 | return u->cgroup_realized && | |
c72703e2 CD |
2605 | ((u->cgroup_realized_mask | target_mask) & CGROUP_MASK_V1) == (u->cgroup_realized_mask & CGROUP_MASK_V1) && |
2606 | ((u->cgroup_enabled_mask | enable_mask) & CGROUP_MASK_V2) == (u->cgroup_enabled_mask & CGROUP_MASK_V2); | |
a57669d2 CD |
2607 | } |
2608 | ||
020b2e41 | 2609 | void unit_add_to_cgroup_realize_queue(Unit *u) { |
2aa57a65 LP |
2610 | assert(u); |
2611 | ||
2612 | if (u->in_cgroup_realize_queue) | |
2613 | return; | |
2614 | ||
a479c21e | 2615 | LIST_APPEND(cgroup_realize_queue, u->manager->cgroup_realize_queue, u); |
2aa57a65 LP |
2616 | u->in_cgroup_realize_queue = true; |
2617 | } | |
2618 | ||
2619 | static void unit_remove_from_cgroup_realize_queue(Unit *u) { | |
2620 | assert(u); | |
2621 | ||
2622 | if (!u->in_cgroup_realize_queue) | |
2623 | return; | |
2624 | ||
2625 | LIST_REMOVE(cgroup_realize_queue, u->manager->cgroup_realize_queue, u); | |
2626 | u->in_cgroup_realize_queue = false; | |
2627 | } | |
2628 | ||
a57669d2 CD |
2629 | /* Controllers can only be enabled breadth-first, from the root of the |
2630 | * hierarchy downwards to the unit in question. */ | |
2631 | static int unit_realize_cgroup_now_enable(Unit *u, ManagerState state) { | |
2632 | CGroupMask target_mask, enable_mask, new_target_mask, new_enable_mask; | |
12f64221 | 2633 | Unit *slice; |
a57669d2 CD |
2634 | int r; |
2635 | ||
2636 | assert(u); | |
2637 | ||
2638 | /* First go deal with this unit's parent, or we won't be able to enable | |
2639 | * any new controllers at this layer. */ | |
12f64221 LP |
2640 | slice = UNIT_GET_SLICE(u); |
2641 | if (slice) { | |
2642 | r = unit_realize_cgroup_now_enable(slice, state); | |
a57669d2 CD |
2643 | if (r < 0) |
2644 | return r; | |
2645 | } | |
2646 | ||
2647 | target_mask = unit_get_target_mask(u); | |
2648 | enable_mask = unit_get_enable_mask(u); | |
2649 | ||
2650 | /* We can only enable in this direction, don't try to disable anything. | |
2651 | */ | |
2652 | if (unit_has_mask_enables_realized(u, target_mask, enable_mask)) | |
2653 | return 0; | |
2654 | ||
2655 | new_target_mask = u->cgroup_realized_mask | target_mask; | |
2656 | new_enable_mask = u->cgroup_enabled_mask | enable_mask; | |
2657 | ||
7b639614 | 2658 | return unit_update_cgroup(u, new_target_mask, new_enable_mask, state); |
a57669d2 CD |
2659 | } |
2660 | ||
4f6f62e4 CD |
2661 | /* Controllers can only be disabled depth-first, from the leaves of the |
2662 | * hierarchy upwards to the unit in question. */ | |
2663 | static int unit_realize_cgroup_now_disable(Unit *u, ManagerState state) { | |
4f6f62e4 | 2664 | Unit *m; |
4f6f62e4 CD |
2665 | |
2666 | assert(u); | |
2667 | ||
2668 | if (u->type != UNIT_SLICE) | |
2669 | return 0; | |
2670 | ||
d219a2b0 | 2671 | UNIT_FOREACH_DEPENDENCY(m, u, UNIT_ATOM_SLICE_OF) { |
4f6f62e4 CD |
2672 | CGroupMask target_mask, enable_mask, new_target_mask, new_enable_mask; |
2673 | int r; | |
2674 | ||
defe63b0 LP |
2675 | /* The cgroup for this unit might not actually be fully realised yet, in which case it isn't |
2676 | * holding any controllers open anyway. */ | |
d9ef5944 | 2677 | if (!m->cgroup_realized) |
4f6f62e4 CD |
2678 | continue; |
2679 | ||
defe63b0 | 2680 | /* We must disable those below us first in order to release the controller. */ |
4f6f62e4 CD |
2681 | if (m->type == UNIT_SLICE) |
2682 | (void) unit_realize_cgroup_now_disable(m, state); | |
2683 | ||
2684 | target_mask = unit_get_target_mask(m); | |
2685 | enable_mask = unit_get_enable_mask(m); | |
2686 | ||
defe63b0 | 2687 | /* We can only disable in this direction, don't try to enable anything. */ |
4f6f62e4 CD |
2688 | if (unit_has_mask_disables_realized(m, target_mask, enable_mask)) |
2689 | continue; | |
2690 | ||
2691 | new_target_mask = m->cgroup_realized_mask & target_mask; | |
2692 | new_enable_mask = m->cgroup_enabled_mask & enable_mask; | |
2693 | ||
7b639614 | 2694 | r = unit_update_cgroup(m, new_target_mask, new_enable_mask, state); |
4f6f62e4 CD |
2695 | if (r < 0) |
2696 | return r; | |
2697 | } | |
2698 | ||
2699 | return 0; | |
2700 | } | |
a57669d2 | 2701 | |
6414b7c9 DS |
2702 | /* Check if necessary controllers and attributes for a unit are in place. |
2703 | * | |
a57669d2 CD |
2704 | * - If so, do nothing. |
2705 | * - If not, create paths, move processes over, and set attributes. | |
2706 | * | |
2707 | * Controllers can only be *enabled* in a breadth-first way, and *disabled* in | |
2708 | * a depth-first way. As such the process looks like this: | |
2709 | * | |
2710 | * Suppose we have a cgroup hierarchy which looks like this: | |
2711 | * | |
2712 | * root | |
2713 | * / \ | |
2714 | * / \ | |
2715 | * / \ | |
2716 | * a b | |
2717 | * / \ / \ | |
2718 | * / \ / \ | |
2719 | * c d e f | |
2720 | * / \ / \ / \ / \ | |
2721 | * h i j k l m n o | |
2722 | * | |
2723 | * 1. We want to realise cgroup "d" now. | |
c72703e2 | 2724 | * 2. cgroup "a" has DisableControllers=cpu in the associated unit. |
a57669d2 CD |
2725 | * 3. cgroup "k" just started requesting the memory controller. |
2726 | * | |
2727 | * To make this work we must do the following in order: | |
2728 | * | |
2729 | * 1. Disable CPU controller in k, j | |
2730 | * 2. Disable CPU controller in d | |
2731 | * 3. Enable memory controller in root | |
2732 | * 4. Enable memory controller in a | |
2733 | * 5. Enable memory controller in d | |
2734 | * 6. Enable memory controller in k | |
2735 | * | |
2736 | * Notice that we need to touch j in one direction, but not the other. We also | |
2737 | * don't go beyond d when disabling -- it's up to "a" to get realized if it | |
2738 | * wants to disable further. The basic rules are therefore: | |
2739 | * | |
2740 | * - If you're disabling something, you need to realise all of the cgroups from | |
2741 | * your recursive descendants to the root. This starts from the leaves. | |
2742 | * - If you're enabling something, you need to realise from the root cgroup | |
2743 | * downwards, but you don't need to iterate your recursive descendants. | |
6414b7c9 DS |
2744 | * |
2745 | * Returns 0 on success and < 0 on failure. */ | |
db785129 | 2746 | static int unit_realize_cgroup_now(Unit *u, ManagerState state) { |
efdb0237 | 2747 | CGroupMask target_mask, enable_mask; |
12f64221 | 2748 | Unit *slice; |
6414b7c9 | 2749 | int r; |
64747e2d | 2750 | |
4ad49000 | 2751 | assert(u); |
64747e2d | 2752 | |
2aa57a65 | 2753 | unit_remove_from_cgroup_realize_queue(u); |
64747e2d | 2754 | |
efdb0237 | 2755 | target_mask = unit_get_target_mask(u); |
ccf78df1 TH |
2756 | enable_mask = unit_get_enable_mask(u); |
2757 | ||
17f14955 | 2758 | if (unit_has_mask_realized(u, target_mask, enable_mask)) |
0a1eb06d | 2759 | return 0; |
64747e2d | 2760 | |
4f6f62e4 CD |
2761 | /* Disable controllers below us, if there are any */ |
2762 | r = unit_realize_cgroup_now_disable(u, state); | |
2763 | if (r < 0) | |
2764 | return r; | |
2765 | ||
2766 | /* Enable controllers above us, if there are any */ | |
12f64221 LP |
2767 | slice = UNIT_GET_SLICE(u); |
2768 | if (slice) { | |
2769 | r = unit_realize_cgroup_now_enable(slice, state); | |
6414b7c9 DS |
2770 | if (r < 0) |
2771 | return r; | |
2772 | } | |
4ad49000 | 2773 | |
0d2d6fbf | 2774 | /* Now actually deal with the cgroup we were trying to realise and set attributes */ |
7b639614 | 2775 | r = unit_update_cgroup(u, target_mask, enable_mask, state); |
6414b7c9 DS |
2776 | if (r < 0) |
2777 | return r; | |
2778 | ||
c2baf11c LP |
2779 | /* Now, reset the invalidation mask */ |
2780 | u->cgroup_invalidated_mask = 0; | |
6414b7c9 | 2781 | return 0; |
64747e2d LP |
2782 | } |
2783 | ||
91a6073e | 2784 | unsigned manager_dispatch_cgroup_realize_queue(Manager *m) { |
db785129 | 2785 | ManagerState state; |
4ad49000 | 2786 | unsigned n = 0; |
db785129 | 2787 | Unit *i; |
6414b7c9 | 2788 | int r; |
ecedd90f | 2789 | |
91a6073e LP |
2790 | assert(m); |
2791 | ||
db785129 LP |
2792 | state = manager_state(m); |
2793 | ||
91a6073e LP |
2794 | while ((i = m->cgroup_realize_queue)) { |
2795 | assert(i->in_cgroup_realize_queue); | |
ecedd90f | 2796 | |
2aa57a65 LP |
2797 | if (UNIT_IS_INACTIVE_OR_FAILED(unit_active_state(i))) { |
2798 | /* Maybe things changed, and the unit is not actually active anymore? */ | |
2799 | unit_remove_from_cgroup_realize_queue(i); | |
2800 | continue; | |
2801 | } | |
2802 | ||
db785129 | 2803 | r = unit_realize_cgroup_now(i, state); |
6414b7c9 | 2804 | if (r < 0) |
efdb0237 | 2805 | log_warning_errno(r, "Failed to realize cgroups for queued unit %s, ignoring: %m", i->id); |
0a1eb06d | 2806 | |
4ad49000 LP |
2807 | n++; |
2808 | } | |
ecedd90f | 2809 | |
4ad49000 | 2810 | return n; |
8e274523 LP |
2811 | } |
2812 | ||
4c591f39 MK |
2813 | void unit_add_family_to_cgroup_realize_queue(Unit *u) { |
2814 | assert(u); | |
2815 | assert(u->type == UNIT_SLICE); | |
ca949c9d | 2816 | |
4c591f39 MK |
2817 | /* Family of a unit for is defined as (immediate) children of the unit and immediate children of all |
2818 | * its ancestors. | |
2819 | * | |
2820 | * Ideally we would enqueue ancestor path only (bottom up). However, on cgroup-v1 scheduling becomes | |
2821 | * very weird if two units that own processes reside in the same slice, but one is realized in the | |
2822 | * "cpu" hierarchy and one is not (for example because one has CPUWeight= set and the other does | |
2823 | * not), because that means individual processes need to be scheduled against whole cgroups. Let's | |
2824 | * avoid this asymmetry by always ensuring that siblings of a unit are always realized in their v1 | |
2825 | * controller hierarchies too (if unit requires the controller to be realized). | |
e1e98911 | 2826 | * |
4c591f39 MK |
2827 | * The function must invalidate cgroup_members_mask of all ancestors in order to calculate up to date |
2828 | * masks. */ | |
2829 | ||
2830 | do { | |
4ad49000 | 2831 | Unit *m; |
8f53a7b8 | 2832 | |
4c591f39 MK |
2833 | /* Children of u likely changed when we're called */ |
2834 | u->cgroup_members_mask_valid = false; | |
f23ba94d | 2835 | |
d219a2b0 | 2836 | UNIT_FOREACH_DEPENDENCY(m, u, UNIT_ATOM_SLICE_OF) { |
8e274523 | 2837 | |
65f6b6bd | 2838 | /* No point in doing cgroup application for units without active processes. */ |
6414b7c9 DS |
2839 | if (UNIT_IS_INACTIVE_OR_FAILED(unit_active_state(m))) |
2840 | continue; | |
2841 | ||
e1e98911 LP |
2842 | /* We only enqueue siblings if they were realized once at least, in the main |
2843 | * hierarchy. */ | |
2844 | if (!m->cgroup_realized) | |
2845 | continue; | |
2846 | ||
defe63b0 LP |
2847 | /* If the unit doesn't need any new controllers and has current ones |
2848 | * realized, it doesn't need any changes. */ | |
906c06f6 DM |
2849 | if (unit_has_mask_realized(m, |
2850 | unit_get_target_mask(m), | |
17f14955 | 2851 | unit_get_enable_mask(m))) |
6414b7c9 DS |
2852 | continue; |
2853 | ||
91a6073e | 2854 | unit_add_to_cgroup_realize_queue(m); |
50159e6a LP |
2855 | } |
2856 | ||
4c591f39 MK |
2857 | /* Parent comes after children */ |
2858 | unit_add_to_cgroup_realize_queue(u); | |
12f64221 LP |
2859 | |
2860 | u = UNIT_GET_SLICE(u); | |
2861 | } while (u); | |
4ad49000 LP |
2862 | } |
2863 | ||
0a1eb06d | 2864 | int unit_realize_cgroup(Unit *u) { |
12f64221 LP |
2865 | Unit *slice; |
2866 | ||
4ad49000 LP |
2867 | assert(u); |
2868 | ||
35b7ff80 | 2869 | if (!UNIT_HAS_CGROUP_CONTEXT(u)) |
0a1eb06d | 2870 | return 0; |
8e274523 | 2871 | |
4c591f39 MK |
2872 | /* So, here's the deal: when realizing the cgroups for this unit, we need to first create all |
2873 | * parents, but there's more actually: for the weight-based controllers we also need to make sure | |
2874 | * that all our siblings (i.e. units that are in the same slice as we are) have cgroups, too. On the | |
2875 | * other hand, when a controller is removed from realized set, it may become unnecessary in siblings | |
2876 | * and ancestors and they should be (de)realized too. | |
2877 | * | |
2878 | * This call will defer work on the siblings and derealized ancestors to the next event loop | |
2879 | * iteration and synchronously creates the parent cgroups (unit_realize_cgroup_now). */ | |
ca949c9d | 2880 | |
12f64221 LP |
2881 | slice = UNIT_GET_SLICE(u); |
2882 | if (slice) | |
2883 | unit_add_family_to_cgroup_realize_queue(slice); | |
4ad49000 | 2884 | |
6414b7c9 | 2885 | /* And realize this one now (and apply the values) */ |
db785129 | 2886 | return unit_realize_cgroup_now(u, manager_state(u->manager)); |
8e274523 LP |
2887 | } |
2888 | ||
efdb0237 LP |
2889 | void unit_release_cgroup(Unit *u) { |
2890 | assert(u); | |
2891 | ||
8a0d5388 LP |
2892 | /* Forgets all cgroup details for this cgroup — but does *not* destroy the cgroup. This is hence OK to call |
2893 | * when we close down everything for reexecution, where we really want to leave the cgroup in place. */ | |
efdb0237 LP |
2894 | |
2895 | if (u->cgroup_path) { | |
2896 | (void) hashmap_remove(u->manager->cgroup_unit, u->cgroup_path); | |
2897 | u->cgroup_path = mfree(u->cgroup_path); | |
2898 | } | |
2899 | ||
0bb814c2 LP |
2900 | if (u->cgroup_control_inotify_wd >= 0) { |
2901 | if (inotify_rm_watch(u->manager->cgroup_inotify_fd, u->cgroup_control_inotify_wd) < 0) | |
2902 | log_unit_debug_errno(u, errno, "Failed to remove cgroup control inotify watch %i for %s, ignoring: %m", u->cgroup_control_inotify_wd, u->id); | |
efdb0237 | 2903 | |
0bb814c2 LP |
2904 | (void) hashmap_remove(u->manager->cgroup_control_inotify_wd_unit, INT_TO_PTR(u->cgroup_control_inotify_wd)); |
2905 | u->cgroup_control_inotify_wd = -1; | |
efdb0237 | 2906 | } |
afcfaa69 LP |
2907 | |
2908 | if (u->cgroup_memory_inotify_wd >= 0) { | |
2909 | if (inotify_rm_watch(u->manager->cgroup_inotify_fd, u->cgroup_memory_inotify_wd) < 0) | |
2910 | log_unit_debug_errno(u, errno, "Failed to remove cgroup memory inotify watch %i for %s, ignoring: %m", u->cgroup_memory_inotify_wd, u->id); | |
2911 | ||
2912 | (void) hashmap_remove(u->manager->cgroup_memory_inotify_wd_unit, INT_TO_PTR(u->cgroup_memory_inotify_wd)); | |
2913 | u->cgroup_memory_inotify_wd = -1; | |
2914 | } | |
efdb0237 LP |
2915 | } |
2916 | ||
e08dabfe AZ |
2917 | bool unit_maybe_release_cgroup(Unit *u) { |
2918 | int r; | |
2919 | ||
2920 | assert(u); | |
2921 | ||
2922 | if (!u->cgroup_path) | |
2923 | return true; | |
2924 | ||
2925 | /* Don't release the cgroup if there are still processes under it. If we get notified later when all the | |
2926 | * processes exit (e.g. the processes were in D-state and exited after the unit was marked as failed) | |
2927 | * we need the cgroup paths to continue to be tracked by the manager so they can be looked up and cleaned | |
2928 | * up later. */ | |
2929 | r = cg_is_empty_recursive(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path); | |
2930 | if (r < 0) | |
2931 | log_unit_debug_errno(u, r, "Error checking if the cgroup is recursively empty, ignoring: %m"); | |
2932 | else if (r == 1) { | |
2933 | unit_release_cgroup(u); | |
2934 | return true; | |
2935 | } | |
2936 | ||
2937 | return false; | |
2938 | } | |
2939 | ||
efdb0237 | 2940 | void unit_prune_cgroup(Unit *u) { |
8e274523 | 2941 | int r; |
efdb0237 | 2942 | bool is_root_slice; |
8e274523 | 2943 | |
4ad49000 | 2944 | assert(u); |
8e274523 | 2945 | |
efdb0237 LP |
2946 | /* Removes the cgroup, if empty and possible, and stops watching it. */ |
2947 | ||
4ad49000 LP |
2948 | if (!u->cgroup_path) |
2949 | return; | |
8e274523 | 2950 | |
fe700f46 LP |
2951 | (void) unit_get_cpu_usage(u, NULL); /* Cache the last CPU usage value before we destroy the cgroup */ |
2952 | ||
b1994387 ILG |
2953 | #if BPF_FRAMEWORK |
2954 | (void) lsm_bpf_cleanup(u); /* Remove cgroup from the global LSM BPF map */ | |
2955 | #endif | |
2956 | ||
efdb0237 LP |
2957 | is_root_slice = unit_has_name(u, SPECIAL_ROOT_SLICE); |
2958 | ||
2959 | r = cg_trim_everywhere(u->manager->cgroup_supported, u->cgroup_path, !is_root_slice); | |
0219b352 DB |
2960 | if (r < 0) |
2961 | /* One reason we could have failed here is, that the cgroup still contains a process. | |
2962 | * However, if the cgroup becomes removable at a later time, it might be removed when | |
2963 | * the containing slice is stopped. So even if we failed now, this unit shouldn't assume | |
2964 | * that the cgroup is still realized the next time it is started. Do not return early | |
2965 | * on error, continue cleanup. */ | |
6178e2f8 | 2966 | log_unit_full_errno(u, r == -EBUSY ? LOG_DEBUG : LOG_WARNING, r, "Failed to destroy cgroup %s, ignoring: %m", empty_to_root(u->cgroup_path)); |
8e274523 | 2967 | |
efdb0237 LP |
2968 | if (is_root_slice) |
2969 | return; | |
2970 | ||
e08dabfe AZ |
2971 | if (!unit_maybe_release_cgroup(u)) /* Returns true if the cgroup was released */ |
2972 | return; | |
0a1eb06d | 2973 | |
4ad49000 | 2974 | u->cgroup_realized = false; |
bc432dc7 | 2975 | u->cgroup_realized_mask = 0; |
ccf78df1 | 2976 | u->cgroup_enabled_mask = 0; |
084c7007 | 2977 | |
76dc1725 | 2978 | u->bpf_device_control_installed = bpf_program_free(u->bpf_device_control_installed); |
8e274523 LP |
2979 | } |
2980 | ||
efdb0237 | 2981 | int unit_search_main_pid(Unit *u, pid_t *ret) { |
4ad49000 | 2982 | _cleanup_fclose_ FILE *f = NULL; |
4d051546 | 2983 | pid_t pid = 0, npid; |
efdb0237 | 2984 | int r; |
4ad49000 LP |
2985 | |
2986 | assert(u); | |
efdb0237 | 2987 | assert(ret); |
4ad49000 LP |
2988 | |
2989 | if (!u->cgroup_path) | |
efdb0237 | 2990 | return -ENXIO; |
4ad49000 | 2991 | |
efdb0237 LP |
2992 | r = cg_enumerate_processes(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path, &f); |
2993 | if (r < 0) | |
2994 | return r; | |
4ad49000 | 2995 | |
4ad49000 | 2996 | while (cg_read_pid(f, &npid) > 0) { |
4ad49000 LP |
2997 | |
2998 | if (npid == pid) | |
2999 | continue; | |
8e274523 | 3000 | |
4d051546 | 3001 | if (pid_is_my_child(npid) == 0) |
4ad49000 | 3002 | continue; |
8e274523 | 3003 | |
efdb0237 | 3004 | if (pid != 0) |
4ad49000 LP |
3005 | /* Dang, there's more than one daemonized PID |
3006 | in this group, so we don't know what process | |
3007 | is the main process. */ | |
efdb0237 LP |
3008 | |
3009 | return -ENODATA; | |
8e274523 | 3010 | |
4ad49000 | 3011 | pid = npid; |
8e274523 LP |
3012 | } |
3013 | ||
efdb0237 LP |
3014 | *ret = pid; |
3015 | return 0; | |
3016 | } | |
3017 | ||
3018 | static int unit_watch_pids_in_path(Unit *u, const char *path) { | |
b3c5bad3 | 3019 | _cleanup_closedir_ DIR *d = NULL; |
efdb0237 LP |
3020 | _cleanup_fclose_ FILE *f = NULL; |
3021 | int ret = 0, r; | |
3022 | ||
3023 | assert(u); | |
3024 | assert(path); | |
3025 | ||
3026 | r = cg_enumerate_processes(SYSTEMD_CGROUP_CONTROLLER, path, &f); | |
3027 | if (r < 0) | |
3028 | ret = r; | |
3029 | else { | |
3030 | pid_t pid; | |
3031 | ||
3032 | while ((r = cg_read_pid(f, &pid)) > 0) { | |
f75f613d | 3033 | r = unit_watch_pid(u, pid, false); |
efdb0237 LP |
3034 | if (r < 0 && ret >= 0) |
3035 | ret = r; | |
3036 | } | |
3037 | ||
3038 | if (r < 0 && ret >= 0) | |
3039 | ret = r; | |
3040 | } | |
3041 | ||
3042 | r = cg_enumerate_subgroups(SYSTEMD_CGROUP_CONTROLLER, path, &d); | |
3043 | if (r < 0) { | |
3044 | if (ret >= 0) | |
3045 | ret = r; | |
3046 | } else { | |
3047 | char *fn; | |
3048 | ||
3049 | while ((r = cg_read_subgroup(d, &fn)) > 0) { | |
3050 | _cleanup_free_ char *p = NULL; | |
3051 | ||
95b21cff | 3052 | p = path_join(empty_to_root(path), fn); |
efdb0237 LP |
3053 | free(fn); |
3054 | ||
3055 | if (!p) | |
3056 | return -ENOMEM; | |
3057 | ||
3058 | r = unit_watch_pids_in_path(u, p); | |
3059 | if (r < 0 && ret >= 0) | |
3060 | ret = r; | |
3061 | } | |
3062 | ||
3063 | if (r < 0 && ret >= 0) | |
3064 | ret = r; | |
3065 | } | |
3066 | ||
3067 | return ret; | |
3068 | } | |
3069 | ||
11aef522 LP |
3070 | int unit_synthesize_cgroup_empty_event(Unit *u) { |
3071 | int r; | |
3072 | ||
3073 | assert(u); | |
3074 | ||
3075 | /* Enqueue a synthetic cgroup empty event if this unit doesn't watch any PIDs anymore. This is compatibility | |
3076 | * support for non-unified systems where notifications aren't reliable, and hence need to take whatever we can | |
3077 | * get as notification source as soon as we stopped having any useful PIDs to watch for. */ | |
3078 | ||
3079 | if (!u->cgroup_path) | |
3080 | return -ENOENT; | |
3081 | ||
3082 | r = cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER); | |
3083 | if (r < 0) | |
3084 | return r; | |
3085 | if (r > 0) /* On unified we have reliable notifications, and don't need this */ | |
3086 | return 0; | |
3087 | ||
3088 | if (!set_isempty(u->pids)) | |
3089 | return 0; | |
3090 | ||
3091 | unit_add_to_cgroup_empty_queue(u); | |
3092 | return 0; | |
3093 | } | |
3094 | ||
efdb0237 | 3095 | int unit_watch_all_pids(Unit *u) { |
b4cccbc1 LP |
3096 | int r; |
3097 | ||
efdb0237 LP |
3098 | assert(u); |
3099 | ||
3100 | /* Adds all PIDs from our cgroup to the set of PIDs we | |
3101 | * watch. This is a fallback logic for cases where we do not | |
3102 | * get reliable cgroup empty notifications: we try to use | |
3103 | * SIGCHLD as replacement. */ | |
3104 | ||
3105 | if (!u->cgroup_path) | |
3106 | return -ENOENT; | |
3107 | ||
c22800e4 | 3108 | r = cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER); |
b4cccbc1 LP |
3109 | if (r < 0) |
3110 | return r; | |
3111 | if (r > 0) /* On unified we can use proper notifications */ | |
efdb0237 LP |
3112 | return 0; |
3113 | ||
3114 | return unit_watch_pids_in_path(u, u->cgroup_path); | |
3115 | } | |
3116 | ||
09e24654 | 3117 | static int on_cgroup_empty_event(sd_event_source *s, void *userdata) { |
99534007 | 3118 | Manager *m = ASSERT_PTR(userdata); |
09e24654 | 3119 | Unit *u; |
efdb0237 LP |
3120 | int r; |
3121 | ||
09e24654 | 3122 | assert(s); |
efdb0237 | 3123 | |
09e24654 LP |
3124 | u = m->cgroup_empty_queue; |
3125 | if (!u) | |
efdb0237 LP |
3126 | return 0; |
3127 | ||
09e24654 LP |
3128 | assert(u->in_cgroup_empty_queue); |
3129 | u->in_cgroup_empty_queue = false; | |
3130 | LIST_REMOVE(cgroup_empty_queue, m->cgroup_empty_queue, u); | |
3131 | ||
3132 | if (m->cgroup_empty_queue) { | |
3133 | /* More stuff queued, let's make sure we remain enabled */ | |
3134 | r = sd_event_source_set_enabled(s, SD_EVENT_ONESHOT); | |
3135 | if (r < 0) | |
19a691a9 | 3136 | log_debug_errno(r, "Failed to reenable cgroup empty event source, ignoring: %m"); |
09e24654 | 3137 | } |
efdb0237 | 3138 | |
f7829525 NK |
3139 | /* Update state based on OOM kills before we notify about cgroup empty event */ |
3140 | (void) unit_check_oom(u); | |
3141 | (void) unit_check_oomd_kill(u); | |
3142 | ||
efdb0237 LP |
3143 | unit_add_to_gc_queue(u); |
3144 | ||
380dd177 RP |
3145 | if (IN_SET(unit_active_state(u), UNIT_INACTIVE, UNIT_FAILED)) |
3146 | unit_prune_cgroup(u); | |
3147 | else if (UNIT_VTABLE(u)->notify_cgroup_empty) | |
efdb0237 LP |
3148 | UNIT_VTABLE(u)->notify_cgroup_empty(u); |
3149 | ||
3150 | return 0; | |
3151 | } | |
3152 | ||
09e24654 LP |
3153 | void unit_add_to_cgroup_empty_queue(Unit *u) { |
3154 | int r; | |
3155 | ||
3156 | assert(u); | |
3157 | ||
3158 | /* Note that there are four different ways how cgroup empty events reach us: | |
3159 | * | |
3160 | * 1. On the unified hierarchy we get an inotify event on the cgroup | |
3161 | * | |
3162 | * 2. On the legacy hierarchy, when running in system mode, we get a datagram on the cgroup agent socket | |
3163 | * | |
3164 | * 3. On the legacy hierarchy, when running in user mode, we get a D-Bus signal on the system bus | |
3165 | * | |
3166 | * 4. On the legacy hierarchy, in service units we start watching all processes of the cgroup for SIGCHLD as | |
3167 | * soon as we get one SIGCHLD, to deal with unreliable cgroup notifications. | |
3168 | * | |
3169 | * Regardless which way we got the notification, we'll verify it here, and then add it to a separate | |
3170 | * queue. This queue will be dispatched at a lower priority than the SIGCHLD handler, so that we always use | |
3171 | * SIGCHLD if we can get it first, and only use the cgroup empty notifications if there's no SIGCHLD pending | |
3172 | * (which might happen if the cgroup doesn't contain processes that are our own child, which is typically the | |
3173 | * case for scope units). */ | |
3174 | ||
3175 | if (u->in_cgroup_empty_queue) | |
3176 | return; | |
3177 | ||
3178 | /* Let's verify that the cgroup is really empty */ | |
3179 | if (!u->cgroup_path) | |
3180 | return; | |
e1e98911 | 3181 | |
09e24654 LP |
3182 | r = cg_is_empty_recursive(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path); |
3183 | if (r < 0) { | |
6178e2f8 | 3184 | log_unit_debug_errno(u, r, "Failed to determine whether cgroup %s is empty: %m", empty_to_root(u->cgroup_path)); |
09e24654 LP |
3185 | return; |
3186 | } | |
3187 | if (r == 0) | |
3188 | return; | |
3189 | ||
3190 | LIST_PREPEND(cgroup_empty_queue, u->manager->cgroup_empty_queue, u); | |
3191 | u->in_cgroup_empty_queue = true; | |
3192 | ||
3193 | /* Trigger the defer event */ | |
3194 | r = sd_event_source_set_enabled(u->manager->cgroup_empty_event_source, SD_EVENT_ONESHOT); | |
3195 | if (r < 0) | |
3196 | log_debug_errno(r, "Failed to enable cgroup empty event source: %m"); | |
3197 | } | |
3198 | ||
d9e45bc3 MS |
3199 | static void unit_remove_from_cgroup_empty_queue(Unit *u) { |
3200 | assert(u); | |
3201 | ||
3202 | if (!u->in_cgroup_empty_queue) | |
3203 | return; | |
3204 | ||
3205 | LIST_REMOVE(cgroup_empty_queue, u->manager->cgroup_empty_queue, u); | |
3206 | u->in_cgroup_empty_queue = false; | |
3207 | } | |
3208 | ||
fe8d22fb AZ |
3209 | int unit_check_oomd_kill(Unit *u) { |
3210 | _cleanup_free_ char *value = NULL; | |
3211 | bool increased; | |
3212 | uint64_t n = 0; | |
3213 | int r; | |
3214 | ||
3215 | if (!u->cgroup_path) | |
3216 | return 0; | |
3217 | ||
3218 | r = cg_all_unified(); | |
3219 | if (r < 0) | |
3220 | return log_unit_debug_errno(u, r, "Couldn't determine whether we are in all unified mode: %m"); | |
3221 | else if (r == 0) | |
3222 | return 0; | |
3223 | ||
38c41427 | 3224 | r = cg_get_xattr_malloc(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path, "user.oomd_ooms", &value); |
00675c36 | 3225 | if (r < 0 && !ERRNO_IS_XATTR_ABSENT(r)) |
fe8d22fb AZ |
3226 | return r; |
3227 | ||
3228 | if (!isempty(value)) { | |
3229 | r = safe_atou64(value, &n); | |
3230 | if (r < 0) | |
3231 | return r; | |
3232 | } | |
3233 | ||
3234 | increased = n > u->managed_oom_kill_last; | |
3235 | u->managed_oom_kill_last = n; | |
3236 | ||
3237 | if (!increased) | |
3238 | return 0; | |
3239 | ||
38c41427 NK |
3240 | n = 0; |
3241 | value = mfree(value); | |
3242 | r = cg_get_xattr_malloc(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path, "user.oomd_kill", &value); | |
3243 | if (r >= 0 && !isempty(value)) | |
3244 | (void) safe_atou64(value, &n); | |
3245 | ||
fe8d22fb | 3246 | if (n > 0) |
c2503e35 RH |
3247 | log_unit_struct(u, LOG_NOTICE, |
3248 | "MESSAGE_ID=" SD_MESSAGE_UNIT_OOMD_KILL_STR, | |
3249 | LOG_UNIT_INVOCATION_ID(u), | |
38c41427 NK |
3250 | LOG_UNIT_MESSAGE(u, "systemd-oomd killed %"PRIu64" process(es) in this unit.", n), |
3251 | "N_PROCESSES=%" PRIu64, n); | |
3252 | else | |
3253 | log_unit_struct(u, LOG_NOTICE, | |
3254 | "MESSAGE_ID=" SD_MESSAGE_UNIT_OOMD_KILL_STR, | |
3255 | LOG_UNIT_INVOCATION_ID(u), | |
3256 | LOG_UNIT_MESSAGE(u, "systemd-oomd killed some process(es) in this unit.")); | |
3257 | ||
3258 | unit_notify_cgroup_oom(u, /* ManagedOOM= */ true); | |
fe8d22fb AZ |
3259 | |
3260 | return 1; | |
3261 | } | |
3262 | ||
2ba6ae6b | 3263 | int unit_check_oom(Unit *u) { |
afcfaa69 LP |
3264 | _cleanup_free_ char *oom_kill = NULL; |
3265 | bool increased; | |
3266 | uint64_t c; | |
3267 | int r; | |
3268 | ||
3269 | if (!u->cgroup_path) | |
3270 | return 0; | |
3271 | ||
3272 | r = cg_get_keyed_attribute("memory", u->cgroup_path, "memory.events", STRV_MAKE("oom_kill"), &oom_kill); | |
fc594dee LP |
3273 | if (IN_SET(r, -ENOENT, -ENXIO)) /* Handle gracefully if cgroup or oom_kill attribute don't exist */ |
3274 | c = 0; | |
3275 | else if (r < 0) | |
afcfaa69 | 3276 | return log_unit_debug_errno(u, r, "Failed to read oom_kill field of memory.events cgroup attribute: %m"); |
fc594dee LP |
3277 | else { |
3278 | r = safe_atou64(oom_kill, &c); | |
3279 | if (r < 0) | |
3280 | return log_unit_debug_errno(u, r, "Failed to parse oom_kill field: %m"); | |
3281 | } | |
afcfaa69 LP |
3282 | |
3283 | increased = c > u->oom_kill_last; | |
3284 | u->oom_kill_last = c; | |
3285 | ||
3286 | if (!increased) | |
3287 | return 0; | |
3288 | ||
c2503e35 RH |
3289 | log_unit_struct(u, LOG_NOTICE, |
3290 | "MESSAGE_ID=" SD_MESSAGE_UNIT_OUT_OF_MEMORY_STR, | |
3291 | LOG_UNIT_INVOCATION_ID(u), | |
3292 | LOG_UNIT_MESSAGE(u, "A process of this unit has been killed by the OOM killer.")); | |
afcfaa69 | 3293 | |
38c41427 | 3294 | unit_notify_cgroup_oom(u, /* ManagedOOM= */ false); |
afcfaa69 LP |
3295 | |
3296 | return 1; | |
3297 | } | |
3298 | ||
3299 | static int on_cgroup_oom_event(sd_event_source *s, void *userdata) { | |
99534007 | 3300 | Manager *m = ASSERT_PTR(userdata); |
afcfaa69 LP |
3301 | Unit *u; |
3302 | int r; | |
3303 | ||
3304 | assert(s); | |
afcfaa69 LP |
3305 | |
3306 | u = m->cgroup_oom_queue; | |
3307 | if (!u) | |
3308 | return 0; | |
3309 | ||
3310 | assert(u->in_cgroup_oom_queue); | |
3311 | u->in_cgroup_oom_queue = false; | |
3312 | LIST_REMOVE(cgroup_oom_queue, m->cgroup_oom_queue, u); | |
3313 | ||
3314 | if (m->cgroup_oom_queue) { | |
3315 | /* More stuff queued, let's make sure we remain enabled */ | |
3316 | r = sd_event_source_set_enabled(s, SD_EVENT_ONESHOT); | |
3317 | if (r < 0) | |
3318 | log_debug_errno(r, "Failed to reenable cgroup oom event source, ignoring: %m"); | |
3319 | } | |
3320 | ||
3321 | (void) unit_check_oom(u); | |
3322 | return 0; | |
3323 | } | |
3324 | ||
3325 | static void unit_add_to_cgroup_oom_queue(Unit *u) { | |
3326 | int r; | |
3327 | ||
3328 | assert(u); | |
3329 | ||
3330 | if (u->in_cgroup_oom_queue) | |
3331 | return; | |
3332 | if (!u->cgroup_path) | |
3333 | return; | |
3334 | ||
3335 | LIST_PREPEND(cgroup_oom_queue, u->manager->cgroup_oom_queue, u); | |
3336 | u->in_cgroup_oom_queue = true; | |
3337 | ||
3338 | /* Trigger the defer event */ | |
3339 | if (!u->manager->cgroup_oom_event_source) { | |
3340 | _cleanup_(sd_event_source_unrefp) sd_event_source *s = NULL; | |
3341 | ||
3342 | r = sd_event_add_defer(u->manager->event, &s, on_cgroup_oom_event, u->manager); | |
3343 | if (r < 0) { | |
3344 | log_error_errno(r, "Failed to create cgroup oom event source: %m"); | |
3345 | return; | |
3346 | } | |
3347 | ||
3348 | r = sd_event_source_set_priority(s, SD_EVENT_PRIORITY_NORMAL-8); | |
3349 | if (r < 0) { | |
3350 | log_error_errno(r, "Failed to set priority of cgroup oom event source: %m"); | |
3351 | return; | |
3352 | } | |
3353 | ||
3354 | (void) sd_event_source_set_description(s, "cgroup-oom"); | |
3355 | u->manager->cgroup_oom_event_source = TAKE_PTR(s); | |
3356 | } | |
3357 | ||
3358 | r = sd_event_source_set_enabled(u->manager->cgroup_oom_event_source, SD_EVENT_ONESHOT); | |
3359 | if (r < 0) | |
3360 | log_error_errno(r, "Failed to enable cgroup oom event source: %m"); | |
3361 | } | |
3362 | ||
d9e45bc3 MS |
3363 | static int unit_check_cgroup_events(Unit *u) { |
3364 | char *values[2] = {}; | |
3365 | int r; | |
3366 | ||
3367 | assert(u); | |
3368 | ||
869f52f2 DS |
3369 | if (!u->cgroup_path) |
3370 | return 0; | |
3371 | ||
d9e45bc3 MS |
3372 | r = cg_get_keyed_attribute_graceful(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path, "cgroup.events", |
3373 | STRV_MAKE("populated", "frozen"), values); | |
3374 | if (r < 0) | |
3375 | return r; | |
3376 | ||
3377 | /* The cgroup.events notifications can be merged together so act as we saw the given state for the | |
3378 | * first time. The functions we call to handle given state are idempotent, which makes them | |
3379 | * effectively remember the previous state. */ | |
3380 | if (values[0]) { | |
3381 | if (streq(values[0], "1")) | |
3382 | unit_remove_from_cgroup_empty_queue(u); | |
3383 | else | |
3384 | unit_add_to_cgroup_empty_queue(u); | |
3385 | } | |
3386 | ||
3387 | /* Disregard freezer state changes due to operations not initiated by us */ | |
3388 | if (values[1] && IN_SET(u->freezer_state, FREEZER_FREEZING, FREEZER_THAWING)) { | |
3389 | if (streq(values[1], "0")) | |
3390 | unit_thawed(u); | |
3391 | else | |
3392 | unit_frozen(u); | |
3393 | } | |
3394 | ||
3395 | free(values[0]); | |
3396 | free(values[1]); | |
3397 | ||
3398 | return 0; | |
3399 | } | |
3400 | ||
efdb0237 | 3401 | static int on_cgroup_inotify_event(sd_event_source *s, int fd, uint32_t revents, void *userdata) { |
99534007 | 3402 | Manager *m = ASSERT_PTR(userdata); |
efdb0237 LP |
3403 | |
3404 | assert(s); | |
3405 | assert(fd >= 0); | |
efdb0237 LP |
3406 | |
3407 | for (;;) { | |
3408 | union inotify_event_buffer buffer; | |
efdb0237 LP |
3409 | ssize_t l; |
3410 | ||
3411 | l = read(fd, &buffer, sizeof(buffer)); | |
3412 | if (l < 0) { | |
8add30a0 | 3413 | if (ERRNO_IS_TRANSIENT(errno)) |
efdb0237 LP |
3414 | return 0; |
3415 | ||
3416 | return log_error_errno(errno, "Failed to read control group inotify events: %m"); | |
3417 | } | |
3418 | ||
00adc340 | 3419 | FOREACH_INOTIFY_EVENT_WARN(e, buffer, l) { |
efdb0237 LP |
3420 | Unit *u; |
3421 | ||
3422 | if (e->wd < 0) | |
3423 | /* Queue overflow has no watch descriptor */ | |
3424 | continue; | |
3425 | ||
3426 | if (e->mask & IN_IGNORED) | |
3427 | /* The watch was just removed */ | |
3428 | continue; | |
3429 | ||
afcfaa69 LP |
3430 | /* Note that inotify might deliver events for a watch even after it was removed, |
3431 | * because it was queued before the removal. Let's ignore this here safely. */ | |
3432 | ||
0bb814c2 | 3433 | u = hashmap_get(m->cgroup_control_inotify_wd_unit, INT_TO_PTR(e->wd)); |
afcfaa69 | 3434 | if (u) |
d9e45bc3 | 3435 | unit_check_cgroup_events(u); |
efdb0237 | 3436 | |
afcfaa69 LP |
3437 | u = hashmap_get(m->cgroup_memory_inotify_wd_unit, INT_TO_PTR(e->wd)); |
3438 | if (u) | |
3439 | unit_add_to_cgroup_oom_queue(u); | |
efdb0237 LP |
3440 | } |
3441 | } | |
8e274523 LP |
3442 | } |
3443 | ||
17f14955 RG |
3444 | static int cg_bpf_mask_supported(CGroupMask *ret) { |
3445 | CGroupMask mask = 0; | |
3446 | int r; | |
3447 | ||
3448 | /* BPF-based firewall */ | |
3449 | r = bpf_firewall_supported(); | |
ad13559e YW |
3450 | if (r < 0) |
3451 | return r; | |
17f14955 RG |
3452 | if (r > 0) |
3453 | mask |= CGROUP_MASK_BPF_FIREWALL; | |
3454 | ||
084c7007 RG |
3455 | /* BPF-based device access control */ |
3456 | r = bpf_devices_supported(); | |
ad13559e YW |
3457 | if (r < 0) |
3458 | return r; | |
084c7007 RG |
3459 | if (r > 0) |
3460 | mask |= CGROUP_MASK_BPF_DEVICES; | |
3461 | ||
506ea51b JK |
3462 | /* BPF pinned prog */ |
3463 | r = bpf_foreign_supported(); | |
ad13559e YW |
3464 | if (r < 0) |
3465 | return r; | |
506ea51b JK |
3466 | if (r > 0) |
3467 | mask |= CGROUP_MASK_BPF_FOREIGN; | |
3468 | ||
a8e5eb17 | 3469 | /* BPF-based bind{4|6} hooks */ |
cd09a5f3 | 3470 | r = bpf_socket_bind_supported(); |
ad13559e YW |
3471 | if (r < 0) |
3472 | return r; | |
a8e5eb17 JK |
3473 | if (r > 0) |
3474 | mask |= CGROUP_MASK_BPF_SOCKET_BIND; | |
3475 | ||
6f50d4f7 MV |
3476 | /* BPF-based cgroup_skb/{egress|ingress} hooks */ |
3477 | r = restrict_network_interfaces_supported(); | |
ad13559e YW |
3478 | if (r < 0) |
3479 | return r; | |
6f50d4f7 MV |
3480 | if (r > 0) |
3481 | mask |= CGROUP_MASK_BPF_RESTRICT_NETWORK_INTERFACES; | |
3482 | ||
17f14955 RG |
3483 | *ret = mask; |
3484 | return 0; | |
3485 | } | |
3486 | ||
8e274523 | 3487 | int manager_setup_cgroup(Manager *m) { |
9444b1f2 | 3488 | _cleanup_free_ char *path = NULL; |
10bd3e2e | 3489 | const char *scope_path; |
b4cccbc1 | 3490 | int r, all_unified; |
17f14955 | 3491 | CGroupMask mask; |
efdb0237 | 3492 | char *e; |
8e274523 LP |
3493 | |
3494 | assert(m); | |
3495 | ||
35d2e7ec | 3496 | /* 1. Determine hierarchy */ |
efdb0237 | 3497 | m->cgroup_root = mfree(m->cgroup_root); |
9444b1f2 | 3498 | r = cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER, 0, &m->cgroup_root); |
23bbb0de MS |
3499 | if (r < 0) |
3500 | return log_error_errno(r, "Cannot determine cgroup we are running in: %m"); | |
8e274523 | 3501 | |
efdb0237 LP |
3502 | /* Chop off the init scope, if we are already located in it */ |
3503 | e = endswith(m->cgroup_root, "/" SPECIAL_INIT_SCOPE); | |
0d8c31ff | 3504 | |
efdb0237 LP |
3505 | /* LEGACY: Also chop off the system slice if we are in |
3506 | * it. This is to support live upgrades from older systemd | |
3507 | * versions where PID 1 was moved there. Also see | |
3508 | * cg_get_root_path(). */ | |
463d0d15 | 3509 | if (!e && MANAGER_IS_SYSTEM(m)) { |
9444b1f2 | 3510 | e = endswith(m->cgroup_root, "/" SPECIAL_SYSTEM_SLICE); |
15c60e99 | 3511 | if (!e) |
efdb0237 | 3512 | e = endswith(m->cgroup_root, "/system"); /* even more legacy */ |
0baf24dd | 3513 | } |
efdb0237 LP |
3514 | if (e) |
3515 | *e = 0; | |
7ccfb64a | 3516 | |
7546145e LP |
3517 | /* And make sure to store away the root value without trailing slash, even for the root dir, so that we can |
3518 | * easily prepend it everywhere. */ | |
3519 | delete_trailing_chars(m->cgroup_root, "/"); | |
8e274523 | 3520 | |
35d2e7ec | 3521 | /* 2. Show data */ |
9444b1f2 | 3522 | r = cg_get_path(SYSTEMD_CGROUP_CONTROLLER, m->cgroup_root, NULL, &path); |
23bbb0de MS |
3523 | if (r < 0) |
3524 | return log_error_errno(r, "Cannot find cgroup mount point: %m"); | |
8e274523 | 3525 | |
d4d99bc6 | 3526 | r = cg_unified(); |
415fc41c TH |
3527 | if (r < 0) |
3528 | return log_error_errno(r, "Couldn't determine if we are running in the unified hierarchy: %m"); | |
5da38d07 | 3529 | |
b4cccbc1 | 3530 | all_unified = cg_all_unified(); |
d4c819ed ZJS |
3531 | if (all_unified < 0) |
3532 | return log_error_errno(all_unified, "Couldn't determine whether we are in all unified mode: %m"); | |
3533 | if (all_unified > 0) | |
efdb0237 | 3534 | log_debug("Unified cgroup hierarchy is located at %s.", path); |
b4cccbc1 | 3535 | else { |
c22800e4 | 3536 | r = cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER); |
b4cccbc1 LP |
3537 | if (r < 0) |
3538 | return log_error_errno(r, "Failed to determine whether systemd's own controller is in unified mode: %m"); | |
3539 | if (r > 0) | |
3540 | log_debug("Unified cgroup hierarchy is located at %s. Controllers are on legacy hierarchies.", path); | |
3541 | else | |
3542 | log_debug("Using cgroup controller " SYSTEMD_CGROUP_CONTROLLER_LEGACY ". File system hierarchy is at %s.", path); | |
3543 | } | |
efdb0237 | 3544 | |
09e24654 | 3545 | /* 3. Allocate cgroup empty defer event source */ |
5dcadb4c | 3546 | m->cgroup_empty_event_source = sd_event_source_disable_unref(m->cgroup_empty_event_source); |
09e24654 LP |
3547 | r = sd_event_add_defer(m->event, &m->cgroup_empty_event_source, on_cgroup_empty_event, m); |
3548 | if (r < 0) | |
3549 | return log_error_errno(r, "Failed to create cgroup empty event source: %m"); | |
3550 | ||
cbe83389 LP |
3551 | /* Schedule cgroup empty checks early, but after having processed service notification messages or |
3552 | * SIGCHLD signals, so that a cgroup running empty is always just the last safety net of | |
3553 | * notification, and we collected the metadata the notification and SIGCHLD stuff offers first. */ | |
09e24654 LP |
3554 | r = sd_event_source_set_priority(m->cgroup_empty_event_source, SD_EVENT_PRIORITY_NORMAL-5); |
3555 | if (r < 0) | |
3556 | return log_error_errno(r, "Failed to set priority of cgroup empty event source: %m"); | |
3557 | ||
3558 | r = sd_event_source_set_enabled(m->cgroup_empty_event_source, SD_EVENT_OFF); | |
3559 | if (r < 0) | |
3560 | return log_error_errno(r, "Failed to disable cgroup empty event source: %m"); | |
3561 | ||
3562 | (void) sd_event_source_set_description(m->cgroup_empty_event_source, "cgroup-empty"); | |
3563 | ||
3564 | /* 4. Install notifier inotify object, or agent */ | |
10bd3e2e | 3565 | if (cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER) > 0) { |
c6c18be3 | 3566 | |
09e24654 | 3567 | /* In the unified hierarchy we can get cgroup empty notifications via inotify. */ |
efdb0237 | 3568 | |
5dcadb4c | 3569 | m->cgroup_inotify_event_source = sd_event_source_disable_unref(m->cgroup_inotify_event_source); |
10bd3e2e | 3570 | safe_close(m->cgroup_inotify_fd); |
efdb0237 | 3571 | |
10bd3e2e LP |
3572 | m->cgroup_inotify_fd = inotify_init1(IN_NONBLOCK|IN_CLOEXEC); |
3573 | if (m->cgroup_inotify_fd < 0) | |
3574 | return log_error_errno(errno, "Failed to create control group inotify object: %m"); | |
efdb0237 | 3575 | |
10bd3e2e LP |
3576 | r = sd_event_add_io(m->event, &m->cgroup_inotify_event_source, m->cgroup_inotify_fd, EPOLLIN, on_cgroup_inotify_event, m); |
3577 | if (r < 0) | |
3578 | return log_error_errno(r, "Failed to watch control group inotify object: %m"); | |
efdb0237 | 3579 | |
cbe83389 LP |
3580 | /* Process cgroup empty notifications early. Note that when this event is dispatched it'll |
3581 | * just add the unit to a cgroup empty queue, hence let's run earlier than that. Also see | |
3582 | * handling of cgroup agent notifications, for the classic cgroup hierarchy support. */ | |
3583 | r = sd_event_source_set_priority(m->cgroup_inotify_event_source, SD_EVENT_PRIORITY_NORMAL-9); | |
10bd3e2e LP |
3584 | if (r < 0) |
3585 | return log_error_errno(r, "Failed to set priority of inotify event source: %m"); | |
efdb0237 | 3586 | |
10bd3e2e | 3587 | (void) sd_event_source_set_description(m->cgroup_inotify_event_source, "cgroup-inotify"); |
efdb0237 | 3588 | |
611c4f8a | 3589 | } else if (MANAGER_IS_SYSTEM(m) && manager_owns_host_root_cgroup(m) && !MANAGER_IS_TEST_RUN(m)) { |
efdb0237 | 3590 | |
10bd3e2e LP |
3591 | /* On the legacy hierarchy we only get notifications via cgroup agents. (Which isn't really reliable, |
3592 | * since it does not generate events when control groups with children run empty. */ | |
8e274523 | 3593 | |
ce906769 | 3594 | r = cg_install_release_agent(SYSTEMD_CGROUP_CONTROLLER, SYSTEMD_CGROUPS_AGENT_PATH); |
23bbb0de | 3595 | if (r < 0) |
10bd3e2e LP |
3596 | log_warning_errno(r, "Failed to install release agent, ignoring: %m"); |
3597 | else if (r > 0) | |
3598 | log_debug("Installed release agent."); | |
3599 | else if (r == 0) | |
3600 | log_debug("Release agent already installed."); | |
3601 | } | |
efdb0237 | 3602 | |
09e24654 | 3603 | /* 5. Make sure we are in the special "init.scope" unit in the root slice. */ |
10bd3e2e LP |
3604 | scope_path = strjoina(m->cgroup_root, "/" SPECIAL_INIT_SCOPE); |
3605 | r = cg_create_and_attach(SYSTEMD_CGROUP_CONTROLLER, scope_path, 0); | |
aa77e234 MS |
3606 | if (r >= 0) { |
3607 | /* Also, move all other userspace processes remaining in the root cgroup into that scope. */ | |
3608 | r = cg_migrate(SYSTEMD_CGROUP_CONTROLLER, m->cgroup_root, SYSTEMD_CGROUP_CONTROLLER, scope_path, 0); | |
3609 | if (r < 0) | |
3610 | log_warning_errno(r, "Couldn't move remaining userspace processes, ignoring: %m"); | |
c6c18be3 | 3611 | |
aa77e234 MS |
3612 | /* 6. And pin it, so that it cannot be unmounted */ |
3613 | safe_close(m->pin_cgroupfs_fd); | |
3614 | m->pin_cgroupfs_fd = open(path, O_RDONLY|O_CLOEXEC|O_DIRECTORY|O_NOCTTY|O_NONBLOCK); | |
3615 | if (m->pin_cgroupfs_fd < 0) | |
3616 | return log_error_errno(errno, "Failed to open pin file: %m"); | |
0d8c31ff | 3617 | |
638cece4 | 3618 | } else if (!MANAGER_IS_TEST_RUN(m)) |
aa77e234 | 3619 | return log_error_errno(r, "Failed to create %s control group: %m", scope_path); |
10bd3e2e | 3620 | |
09e24654 | 3621 | /* 7. Always enable hierarchical support if it exists... */ |
638cece4 | 3622 | if (!all_unified && !MANAGER_IS_TEST_RUN(m)) |
10bd3e2e | 3623 | (void) cg_set_attribute("memory", "/", "memory.use_hierarchy", "1"); |
c6c18be3 | 3624 | |
17f14955 | 3625 | /* 8. Figure out which controllers are supported */ |
0fa7b500 | 3626 | r = cg_mask_supported_subtree(m->cgroup_root, &m->cgroup_supported); |
efdb0237 LP |
3627 | if (r < 0) |
3628 | return log_error_errno(r, "Failed to determine supported controllers: %m"); | |
17f14955 RG |
3629 | |
3630 | /* 9. Figure out which bpf-based pseudo-controllers are supported */ | |
3631 | r = cg_bpf_mask_supported(&mask); | |
3632 | if (r < 0) | |
3633 | return log_error_errno(r, "Failed to determine supported bpf-based pseudo-controllers: %m"); | |
3634 | m->cgroup_supported |= mask; | |
3635 | ||
3636 | /* 10. Log which controllers are supported */ | |
e8616626 ZJS |
3637 | for (CGroupController c = 0; c < _CGROUP_CONTROLLER_MAX; c++) |
3638 | log_debug("Controller '%s' supported: %s", cgroup_controller_to_string(c), | |
3639 | yes_no(m->cgroup_supported & CGROUP_CONTROLLER_TO_MASK(c))); | |
9156e799 | 3640 | |
a32360f1 | 3641 | return 0; |
8e274523 LP |
3642 | } |
3643 | ||
c6c18be3 | 3644 | void manager_shutdown_cgroup(Manager *m, bool delete) { |
8e274523 LP |
3645 | assert(m); |
3646 | ||
9444b1f2 LP |
3647 | /* We can't really delete the group, since we are in it. But |
3648 | * let's trim it. */ | |
5dd2f5ff | 3649 | if (delete && m->cgroup_root && !FLAGS_SET(m->test_run_flags, MANAGER_TEST_RUN_MINIMAL)) |
efdb0237 LP |
3650 | (void) cg_trim(SYSTEMD_CGROUP_CONTROLLER, m->cgroup_root, false); |
3651 | ||
5dcadb4c | 3652 | m->cgroup_empty_event_source = sd_event_source_disable_unref(m->cgroup_empty_event_source); |
09e24654 | 3653 | |
0bb814c2 | 3654 | m->cgroup_control_inotify_wd_unit = hashmap_free(m->cgroup_control_inotify_wd_unit); |
afcfaa69 | 3655 | m->cgroup_memory_inotify_wd_unit = hashmap_free(m->cgroup_memory_inotify_wd_unit); |
efdb0237 | 3656 | |
5dcadb4c | 3657 | m->cgroup_inotify_event_source = sd_event_source_disable_unref(m->cgroup_inotify_event_source); |
efdb0237 | 3658 | m->cgroup_inotify_fd = safe_close(m->cgroup_inotify_fd); |
8e274523 | 3659 | |
03e334a1 | 3660 | m->pin_cgroupfs_fd = safe_close(m->pin_cgroupfs_fd); |
c6c18be3 | 3661 | |
efdb0237 | 3662 | m->cgroup_root = mfree(m->cgroup_root); |
8e274523 LP |
3663 | } |
3664 | ||
4ad49000 | 3665 | Unit* manager_get_unit_by_cgroup(Manager *m, const char *cgroup) { |
acb14d31 | 3666 | char *p; |
4ad49000 | 3667 | Unit *u; |
acb14d31 LP |
3668 | |
3669 | assert(m); | |
3670 | assert(cgroup); | |
acb14d31 | 3671 | |
4ad49000 LP |
3672 | u = hashmap_get(m->cgroup_unit, cgroup); |
3673 | if (u) | |
3674 | return u; | |
acb14d31 | 3675 | |
2f82562b | 3676 | p = strdupa_safe(cgroup); |
acb14d31 LP |
3677 | for (;;) { |
3678 | char *e; | |
3679 | ||
3680 | e = strrchr(p, '/'); | |
efdb0237 LP |
3681 | if (!e || e == p) |
3682 | return hashmap_get(m->cgroup_unit, SPECIAL_ROOT_SLICE); | |
acb14d31 LP |
3683 | |
3684 | *e = 0; | |
3685 | ||
4ad49000 LP |
3686 | u = hashmap_get(m->cgroup_unit, p); |
3687 | if (u) | |
3688 | return u; | |
acb14d31 LP |
3689 | } |
3690 | } | |
3691 | ||
b3ac818b | 3692 | Unit *manager_get_unit_by_pid_cgroup(Manager *m, pid_t pid) { |
4ad49000 | 3693 | _cleanup_free_ char *cgroup = NULL; |
8e274523 | 3694 | |
8c47c732 LP |
3695 | assert(m); |
3696 | ||
62a76913 | 3697 | if (!pid_is_valid(pid)) |
b3ac818b LP |
3698 | return NULL; |
3699 | ||
62a76913 | 3700 | if (cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER, pid, &cgroup) < 0) |
b3ac818b LP |
3701 | return NULL; |
3702 | ||
3703 | return manager_get_unit_by_cgroup(m, cgroup); | |
3704 | } | |
3705 | ||
3706 | Unit *manager_get_unit_by_pid(Manager *m, pid_t pid) { | |
62a76913 | 3707 | Unit *u, **array; |
b3ac818b LP |
3708 | |
3709 | assert(m); | |
3710 | ||
62a76913 LP |
3711 | /* Note that a process might be owned by multiple units, we return only one here, which is good enough for most |
3712 | * cases, though not strictly correct. We prefer the one reported by cgroup membership, as that's the most | |
3713 | * relevant one as children of the process will be assigned to that one, too, before all else. */ | |
3714 | ||
3715 | if (!pid_is_valid(pid)) | |
8c47c732 LP |
3716 | return NULL; |
3717 | ||
2ca9d979 | 3718 | if (pid == getpid_cached()) |
efdb0237 LP |
3719 | return hashmap_get(m->units, SPECIAL_INIT_SCOPE); |
3720 | ||
62a76913 | 3721 | u = manager_get_unit_by_pid_cgroup(m, pid); |
5fe8876b LP |
3722 | if (u) |
3723 | return u; | |
3724 | ||
62a76913 | 3725 | u = hashmap_get(m->watch_pids, PID_TO_PTR(pid)); |
5fe8876b LP |
3726 | if (u) |
3727 | return u; | |
3728 | ||
62a76913 LP |
3729 | array = hashmap_get(m->watch_pids, PID_TO_PTR(-pid)); |
3730 | if (array) | |
3731 | return array[0]; | |
3732 | ||
3733 | return NULL; | |
6dde1f33 | 3734 | } |
4fbf50b3 | 3735 | |
4ad49000 LP |
3736 | int manager_notify_cgroup_empty(Manager *m, const char *cgroup) { |
3737 | Unit *u; | |
4fbf50b3 | 3738 | |
4ad49000 LP |
3739 | assert(m); |
3740 | assert(cgroup); | |
4fbf50b3 | 3741 | |
09e24654 LP |
3742 | /* Called on the legacy hierarchy whenever we get an explicit cgroup notification from the cgroup agent process |
3743 | * or from the --system instance */ | |
3744 | ||
d8fdc620 LP |
3745 | log_debug("Got cgroup empty notification for: %s", cgroup); |
3746 | ||
4ad49000 | 3747 | u = manager_get_unit_by_cgroup(m, cgroup); |
5ad096b3 LP |
3748 | if (!u) |
3749 | return 0; | |
b56c28c3 | 3750 | |
09e24654 LP |
3751 | unit_add_to_cgroup_empty_queue(u); |
3752 | return 1; | |
5ad096b3 LP |
3753 | } |
3754 | ||
93ff34e4 LB |
3755 | int unit_get_memory_available(Unit *u, uint64_t *ret) { |
3756 | uint64_t unit_current, available = UINT64_MAX; | |
3757 | CGroupContext *unit_context; | |
3758 | const char *memory_file; | |
3759 | int r; | |
3760 | ||
3761 | assert(u); | |
3762 | assert(ret); | |
3763 | ||
3764 | /* If data from cgroups can be accessed, try to find out how much more memory a unit can | |
3765 | * claim before hitting the configured cgroup limits (if any). Consider both MemoryHigh | |
3766 | * and MemoryMax, and also any slice the unit might be nested below. */ | |
3767 | ||
3768 | if (!UNIT_CGROUP_BOOL(u, memory_accounting)) | |
3769 | return -ENODATA; | |
3770 | ||
3771 | if (!u->cgroup_path) | |
3772 | return -ENODATA; | |
3773 | ||
3774 | /* The root cgroup doesn't expose this information */ | |
3775 | if (unit_has_host_root_cgroup(u)) | |
3776 | return -ENODATA; | |
3777 | ||
3778 | if ((u->cgroup_realized_mask & CGROUP_MASK_MEMORY) == 0) | |
3779 | return -ENODATA; | |
3780 | ||
3781 | r = cg_all_unified(); | |
3782 | if (r < 0) | |
3783 | return r; | |
3784 | memory_file = r > 0 ? "memory.current" : "memory.usage_in_bytes"; | |
3785 | ||
3786 | r = cg_get_attribute_as_uint64("memory", u->cgroup_path, memory_file, &unit_current); | |
3787 | if (r < 0) | |
3788 | return r; | |
3789 | ||
3790 | assert_se(unit_context = unit_get_cgroup_context(u)); | |
3791 | ||
3792 | if (unit_context->memory_max != UINT64_MAX || unit_context->memory_high != UINT64_MAX) | |
3793 | available = LESS_BY(MIN(unit_context->memory_max, unit_context->memory_high), unit_current); | |
3794 | ||
3795 | for (Unit *slice = UNIT_GET_SLICE(u); slice; slice = UNIT_GET_SLICE(slice)) { | |
3796 | uint64_t slice_current, slice_available = UINT64_MAX; | |
3797 | CGroupContext *slice_context; | |
3798 | ||
3799 | /* No point in continuing if we can't go any lower */ | |
3800 | if (available == 0) | |
3801 | break; | |
3802 | ||
3803 | if (!slice->cgroup_path) | |
3804 | continue; | |
3805 | ||
3806 | slice_context = unit_get_cgroup_context(slice); | |
3807 | if (!slice_context) | |
3808 | continue; | |
3809 | ||
3810 | if (slice_context->memory_max == UINT64_MAX && slice_context->memory_high == UINT64_MAX) | |
3811 | continue; | |
3812 | ||
3813 | r = cg_get_attribute_as_uint64("memory", slice->cgroup_path, memory_file, &slice_current); | |
3814 | if (r < 0) | |
3815 | continue; | |
3816 | ||
3817 | slice_available = LESS_BY(MIN(slice_context->memory_max, slice_context->memory_high), slice_current); | |
3818 | available = MIN(slice_available, available); | |
3819 | } | |
3820 | ||
3821 | *ret = available; | |
3822 | ||
3823 | return 0; | |
3824 | } | |
3825 | ||
5ad096b3 | 3826 | int unit_get_memory_current(Unit *u, uint64_t *ret) { |
5ad096b3 LP |
3827 | int r; |
3828 | ||
3829 | assert(u); | |
3830 | assert(ret); | |
3831 | ||
2e4025c0 | 3832 | if (!UNIT_CGROUP_BOOL(u, memory_accounting)) |
cf3b4be1 LP |
3833 | return -ENODATA; |
3834 | ||
5ad096b3 LP |
3835 | if (!u->cgroup_path) |
3836 | return -ENODATA; | |
3837 | ||
1f73aa00 | 3838 | /* The root cgroup doesn't expose this information, let's get it from /proc instead */ |
611c4f8a | 3839 | if (unit_has_host_root_cgroup(u)) |
c482724a | 3840 | return procfs_memory_get_used(ret); |
1f73aa00 | 3841 | |
efdb0237 | 3842 | if ((u->cgroup_realized_mask & CGROUP_MASK_MEMORY) == 0) |
5ad096b3 LP |
3843 | return -ENODATA; |
3844 | ||
b4cccbc1 LP |
3845 | r = cg_all_unified(); |
3846 | if (r < 0) | |
3847 | return r; | |
5ad096b3 | 3848 | |
613328c3 | 3849 | return cg_get_attribute_as_uint64("memory", u->cgroup_path, r > 0 ? "memory.current" : "memory.usage_in_bytes", ret); |
5ad096b3 LP |
3850 | } |
3851 | ||
03a7b521 | 3852 | int unit_get_tasks_current(Unit *u, uint64_t *ret) { |
03a7b521 LP |
3853 | assert(u); |
3854 | assert(ret); | |
3855 | ||
2e4025c0 | 3856 | if (!UNIT_CGROUP_BOOL(u, tasks_accounting)) |
cf3b4be1 LP |
3857 | return -ENODATA; |
3858 | ||
03a7b521 LP |
3859 | if (!u->cgroup_path) |
3860 | return -ENODATA; | |
3861 | ||
c36a69f4 | 3862 | /* The root cgroup doesn't expose this information, let's get it from /proc instead */ |
611c4f8a | 3863 | if (unit_has_host_root_cgroup(u)) |
c36a69f4 LP |
3864 | return procfs_tasks_get_current(ret); |
3865 | ||
1f73aa00 LP |
3866 | if ((u->cgroup_realized_mask & CGROUP_MASK_PIDS) == 0) |
3867 | return -ENODATA; | |
3868 | ||
613328c3 | 3869 | return cg_get_attribute_as_uint64("pids", u->cgroup_path, "pids.current", ret); |
03a7b521 LP |
3870 | } |
3871 | ||
5ad096b3 | 3872 | static int unit_get_cpu_usage_raw(Unit *u, nsec_t *ret) { |
5ad096b3 LP |
3873 | uint64_t ns; |
3874 | int r; | |
3875 | ||
3876 | assert(u); | |
3877 | assert(ret); | |
3878 | ||
3879 | if (!u->cgroup_path) | |
3880 | return -ENODATA; | |
3881 | ||
1f73aa00 | 3882 | /* The root cgroup doesn't expose this information, let's get it from /proc instead */ |
611c4f8a | 3883 | if (unit_has_host_root_cgroup(u)) |
1f73aa00 LP |
3884 | return procfs_cpu_get_usage(ret); |
3885 | ||
f98c2585 CD |
3886 | /* Requisite controllers for CPU accounting are not enabled */ |
3887 | if ((get_cpu_accounting_mask() & ~u->cgroup_realized_mask) != 0) | |
3888 | return -ENODATA; | |
3889 | ||
92a99304 LP |
3890 | r = cg_all_unified(); |
3891 | if (r < 0) | |
3892 | return r; | |
b4cccbc1 | 3893 | if (r > 0) { |
66ebf6c0 TH |
3894 | _cleanup_free_ char *val = NULL; |
3895 | uint64_t us; | |
5ad096b3 | 3896 | |
b734a4ff | 3897 | r = cg_get_keyed_attribute("cpu", u->cgroup_path, "cpu.stat", STRV_MAKE("usage_usec"), &val); |
b734a4ff LP |
3898 | if (IN_SET(r, -ENOENT, -ENXIO)) |
3899 | return -ENODATA; | |
d742f4b5 LP |
3900 | if (r < 0) |
3901 | return r; | |
66ebf6c0 TH |
3902 | |
3903 | r = safe_atou64(val, &us); | |
3904 | if (r < 0) | |
3905 | return r; | |
3906 | ||
3907 | ns = us * NSEC_PER_USEC; | |
613328c3 AZ |
3908 | } else |
3909 | return cg_get_attribute_as_uint64("cpuacct", u->cgroup_path, "cpuacct.usage", ret); | |
5ad096b3 LP |
3910 | |
3911 | *ret = ns; | |
3912 | return 0; | |
3913 | } | |
3914 | ||
3915 | int unit_get_cpu_usage(Unit *u, nsec_t *ret) { | |
3916 | nsec_t ns; | |
3917 | int r; | |
3918 | ||
fe700f46 LP |
3919 | assert(u); |
3920 | ||
3921 | /* Retrieve the current CPU usage counter. This will subtract the CPU counter taken when the unit was | |
3922 | * started. If the cgroup has been removed already, returns the last cached value. To cache the value, simply | |
3923 | * call this function with a NULL return value. */ | |
3924 | ||
2e4025c0 | 3925 | if (!UNIT_CGROUP_BOOL(u, cpu_accounting)) |
cf3b4be1 LP |
3926 | return -ENODATA; |
3927 | ||
5ad096b3 | 3928 | r = unit_get_cpu_usage_raw(u, &ns); |
fe700f46 LP |
3929 | if (r == -ENODATA && u->cpu_usage_last != NSEC_INFINITY) { |
3930 | /* If we can't get the CPU usage anymore (because the cgroup was already removed, for example), use our | |
3931 | * cached value. */ | |
3932 | ||
3933 | if (ret) | |
3934 | *ret = u->cpu_usage_last; | |
3935 | return 0; | |
3936 | } | |
5ad096b3 LP |
3937 | if (r < 0) |
3938 | return r; | |
3939 | ||
66ebf6c0 TH |
3940 | if (ns > u->cpu_usage_base) |
3941 | ns -= u->cpu_usage_base; | |
5ad096b3 LP |
3942 | else |
3943 | ns = 0; | |
3944 | ||
fe700f46 LP |
3945 | u->cpu_usage_last = ns; |
3946 | if (ret) | |
3947 | *ret = ns; | |
3948 | ||
5ad096b3 LP |
3949 | return 0; |
3950 | } | |
3951 | ||
906c06f6 DM |
3952 | int unit_get_ip_accounting( |
3953 | Unit *u, | |
3954 | CGroupIPAccountingMetric metric, | |
3955 | uint64_t *ret) { | |
3956 | ||
6b659ed8 | 3957 | uint64_t value; |
906c06f6 DM |
3958 | int fd, r; |
3959 | ||
3960 | assert(u); | |
3961 | assert(metric >= 0); | |
3962 | assert(metric < _CGROUP_IP_ACCOUNTING_METRIC_MAX); | |
3963 | assert(ret); | |
3964 | ||
2e4025c0 | 3965 | if (!UNIT_CGROUP_BOOL(u, ip_accounting)) |
cf3b4be1 LP |
3966 | return -ENODATA; |
3967 | ||
906c06f6 DM |
3968 | fd = IN_SET(metric, CGROUP_IP_INGRESS_BYTES, CGROUP_IP_INGRESS_PACKETS) ? |
3969 | u->ip_accounting_ingress_map_fd : | |
3970 | u->ip_accounting_egress_map_fd; | |
906c06f6 DM |
3971 | if (fd < 0) |
3972 | return -ENODATA; | |
3973 | ||
3974 | if (IN_SET(metric, CGROUP_IP_INGRESS_BYTES, CGROUP_IP_EGRESS_BYTES)) | |
6b659ed8 | 3975 | r = bpf_firewall_read_accounting(fd, &value, NULL); |
906c06f6 | 3976 | else |
6b659ed8 LP |
3977 | r = bpf_firewall_read_accounting(fd, NULL, &value); |
3978 | if (r < 0) | |
3979 | return r; | |
3980 | ||
3981 | /* Add in additional metrics from a previous runtime. Note that when reexecing/reloading the daemon we compile | |
3982 | * all BPF programs and maps anew, but serialize the old counters. When deserializing we store them in the | |
3983 | * ip_accounting_extra[] field, and add them in here transparently. */ | |
3984 | ||
3985 | *ret = value + u->ip_accounting_extra[metric]; | |
906c06f6 DM |
3986 | |
3987 | return r; | |
3988 | } | |
3989 | ||
fbe14fc9 LP |
3990 | static int unit_get_io_accounting_raw(Unit *u, uint64_t ret[static _CGROUP_IO_ACCOUNTING_METRIC_MAX]) { |
3991 | static const char *const field_names[_CGROUP_IO_ACCOUNTING_METRIC_MAX] = { | |
3992 | [CGROUP_IO_READ_BYTES] = "rbytes=", | |
3993 | [CGROUP_IO_WRITE_BYTES] = "wbytes=", | |
3994 | [CGROUP_IO_READ_OPERATIONS] = "rios=", | |
3995 | [CGROUP_IO_WRITE_OPERATIONS] = "wios=", | |
3996 | }; | |
3997 | uint64_t acc[_CGROUP_IO_ACCOUNTING_METRIC_MAX] = {}; | |
3998 | _cleanup_free_ char *path = NULL; | |
3999 | _cleanup_fclose_ FILE *f = NULL; | |
4000 | int r; | |
4001 | ||
4002 | assert(u); | |
4003 | ||
4004 | if (!u->cgroup_path) | |
4005 | return -ENODATA; | |
4006 | ||
4007 | if (unit_has_host_root_cgroup(u)) | |
4008 | return -ENODATA; /* TODO: return useful data for the top-level cgroup */ | |
4009 | ||
4010 | r = cg_all_unified(); | |
4011 | if (r < 0) | |
4012 | return r; | |
4013 | if (r == 0) /* TODO: support cgroupv1 */ | |
4014 | return -ENODATA; | |
4015 | ||
4016 | if (!FLAGS_SET(u->cgroup_realized_mask, CGROUP_MASK_IO)) | |
4017 | return -ENODATA; | |
4018 | ||
4019 | r = cg_get_path("io", u->cgroup_path, "io.stat", &path); | |
4020 | if (r < 0) | |
4021 | return r; | |
4022 | ||
4023 | f = fopen(path, "re"); | |
4024 | if (!f) | |
4025 | return -errno; | |
4026 | ||
4027 | for (;;) { | |
4028 | _cleanup_free_ char *line = NULL; | |
4029 | const char *p; | |
4030 | ||
4031 | r = read_line(f, LONG_LINE_MAX, &line); | |
4032 | if (r < 0) | |
4033 | return r; | |
4034 | if (r == 0) | |
4035 | break; | |
4036 | ||
4037 | p = line; | |
4038 | p += strcspn(p, WHITESPACE); /* Skip over device major/minor */ | |
4039 | p += strspn(p, WHITESPACE); /* Skip over following whitespace */ | |
4040 | ||
4041 | for (;;) { | |
4042 | _cleanup_free_ char *word = NULL; | |
4043 | ||
4044 | r = extract_first_word(&p, &word, NULL, EXTRACT_RETAIN_ESCAPE); | |
4045 | if (r < 0) | |
4046 | return r; | |
4047 | if (r == 0) | |
4048 | break; | |
4049 | ||
4050 | for (CGroupIOAccountingMetric i = 0; i < _CGROUP_IO_ACCOUNTING_METRIC_MAX; i++) { | |
4051 | const char *x; | |
4052 | ||
4053 | x = startswith(word, field_names[i]); | |
4054 | if (x) { | |
4055 | uint64_t w; | |
4056 | ||
4057 | r = safe_atou64(x, &w); | |
4058 | if (r < 0) | |
4059 | return r; | |
4060 | ||
4061 | /* Sum up the stats of all devices */ | |
4062 | acc[i] += w; | |
4063 | break; | |
4064 | } | |
4065 | } | |
4066 | } | |
4067 | } | |
4068 | ||
4069 | memcpy(ret, acc, sizeof(acc)); | |
4070 | return 0; | |
4071 | } | |
4072 | ||
4073 | int unit_get_io_accounting( | |
4074 | Unit *u, | |
4075 | CGroupIOAccountingMetric metric, | |
4076 | bool allow_cache, | |
4077 | uint64_t *ret) { | |
4078 | ||
4079 | uint64_t raw[_CGROUP_IO_ACCOUNTING_METRIC_MAX]; | |
4080 | int r; | |
4081 | ||
4082 | /* Retrieve an IO account parameter. This will subtract the counter when the unit was started. */ | |
4083 | ||
4084 | if (!UNIT_CGROUP_BOOL(u, io_accounting)) | |
4085 | return -ENODATA; | |
4086 | ||
4087 | if (allow_cache && u->io_accounting_last[metric] != UINT64_MAX) | |
4088 | goto done; | |
4089 | ||
4090 | r = unit_get_io_accounting_raw(u, raw); | |
4091 | if (r == -ENODATA && u->io_accounting_last[metric] != UINT64_MAX) | |
4092 | goto done; | |
4093 | if (r < 0) | |
4094 | return r; | |
4095 | ||
4096 | for (CGroupIOAccountingMetric i = 0; i < _CGROUP_IO_ACCOUNTING_METRIC_MAX; i++) { | |
4097 | /* Saturated subtraction */ | |
4098 | if (raw[i] > u->io_accounting_base[i]) | |
4099 | u->io_accounting_last[i] = raw[i] - u->io_accounting_base[i]; | |
4100 | else | |
4101 | u->io_accounting_last[i] = 0; | |
4102 | } | |
4103 | ||
4104 | done: | |
4105 | if (ret) | |
4106 | *ret = u->io_accounting_last[metric]; | |
4107 | ||
4108 | return 0; | |
4109 | } | |
4110 | ||
906c06f6 | 4111 | int unit_reset_cpu_accounting(Unit *u) { |
5ad096b3 LP |
4112 | int r; |
4113 | ||
4114 | assert(u); | |
4115 | ||
fe700f46 LP |
4116 | u->cpu_usage_last = NSEC_INFINITY; |
4117 | ||
0bbff7d6 | 4118 | r = unit_get_cpu_usage_raw(u, &u->cpu_usage_base); |
5ad096b3 | 4119 | if (r < 0) { |
66ebf6c0 | 4120 | u->cpu_usage_base = 0; |
5ad096b3 | 4121 | return r; |
b56c28c3 | 4122 | } |
2633eb83 | 4123 | |
4ad49000 | 4124 | return 0; |
4fbf50b3 LP |
4125 | } |
4126 | ||
906c06f6 DM |
4127 | int unit_reset_ip_accounting(Unit *u) { |
4128 | int r = 0, q = 0; | |
4129 | ||
4130 | assert(u); | |
4131 | ||
4132 | if (u->ip_accounting_ingress_map_fd >= 0) | |
4133 | r = bpf_firewall_reset_accounting(u->ip_accounting_ingress_map_fd); | |
4134 | ||
4135 | if (u->ip_accounting_egress_map_fd >= 0) | |
4136 | q = bpf_firewall_reset_accounting(u->ip_accounting_egress_map_fd); | |
4137 | ||
6b659ed8 LP |
4138 | zero(u->ip_accounting_extra); |
4139 | ||
906c06f6 DM |
4140 | return r < 0 ? r : q; |
4141 | } | |
4142 | ||
fbe14fc9 LP |
4143 | int unit_reset_io_accounting(Unit *u) { |
4144 | int r; | |
4145 | ||
4146 | assert(u); | |
4147 | ||
4148 | for (CGroupIOAccountingMetric i = 0; i < _CGROUP_IO_ACCOUNTING_METRIC_MAX; i++) | |
4149 | u->io_accounting_last[i] = UINT64_MAX; | |
4150 | ||
4151 | r = unit_get_io_accounting_raw(u, u->io_accounting_base); | |
4152 | if (r < 0) { | |
4153 | zero(u->io_accounting_base); | |
4154 | return r; | |
4155 | } | |
4156 | ||
4157 | return 0; | |
4158 | } | |
4159 | ||
9b2559a1 | 4160 | int unit_reset_accounting(Unit *u) { |
fbe14fc9 | 4161 | int r, q, v; |
9b2559a1 LP |
4162 | |
4163 | assert(u); | |
4164 | ||
4165 | r = unit_reset_cpu_accounting(u); | |
fbe14fc9 LP |
4166 | q = unit_reset_io_accounting(u); |
4167 | v = unit_reset_ip_accounting(u); | |
9b2559a1 | 4168 | |
fbe14fc9 | 4169 | return r < 0 ? r : q < 0 ? q : v; |
9b2559a1 LP |
4170 | } |
4171 | ||
e7ab4d1a LP |
4172 | void unit_invalidate_cgroup(Unit *u, CGroupMask m) { |
4173 | assert(u); | |
4174 | ||
4175 | if (!UNIT_HAS_CGROUP_CONTEXT(u)) | |
4176 | return; | |
4177 | ||
4178 | if (m == 0) | |
4179 | return; | |
4180 | ||
538b4852 TH |
4181 | /* always invalidate compat pairs together */ |
4182 | if (m & (CGROUP_MASK_IO | CGROUP_MASK_BLKIO)) | |
4183 | m |= CGROUP_MASK_IO | CGROUP_MASK_BLKIO; | |
4184 | ||
7cce4fb7 LP |
4185 | if (m & (CGROUP_MASK_CPU | CGROUP_MASK_CPUACCT)) |
4186 | m |= CGROUP_MASK_CPU | CGROUP_MASK_CPUACCT; | |
4187 | ||
e00068e7 | 4188 | if (FLAGS_SET(u->cgroup_invalidated_mask, m)) /* NOP? */ |
e7ab4d1a LP |
4189 | return; |
4190 | ||
e00068e7 | 4191 | u->cgroup_invalidated_mask |= m; |
91a6073e | 4192 | unit_add_to_cgroup_realize_queue(u); |
e7ab4d1a LP |
4193 | } |
4194 | ||
906c06f6 DM |
4195 | void unit_invalidate_cgroup_bpf(Unit *u) { |
4196 | assert(u); | |
4197 | ||
4198 | if (!UNIT_HAS_CGROUP_CONTEXT(u)) | |
4199 | return; | |
4200 | ||
17f14955 | 4201 | if (u->cgroup_invalidated_mask & CGROUP_MASK_BPF_FIREWALL) /* NOP? */ |
906c06f6 DM |
4202 | return; |
4203 | ||
17f14955 | 4204 | u->cgroup_invalidated_mask |= CGROUP_MASK_BPF_FIREWALL; |
91a6073e | 4205 | unit_add_to_cgroup_realize_queue(u); |
906c06f6 DM |
4206 | |
4207 | /* If we are a slice unit, we also need to put compile a new BPF program for all our children, as the IP access | |
4208 | * list of our children includes our own. */ | |
4209 | if (u->type == UNIT_SLICE) { | |
4210 | Unit *member; | |
906c06f6 | 4211 | |
d219a2b0 | 4212 | UNIT_FOREACH_DEPENDENCY(member, u, UNIT_ATOM_SLICE_OF) |
15ed3c3a | 4213 | unit_invalidate_cgroup_bpf(member); |
906c06f6 DM |
4214 | } |
4215 | } | |
4216 | ||
869f52f2 DS |
4217 | void unit_cgroup_catchup(Unit *u) { |
4218 | assert(u); | |
4219 | ||
4220 | if (!UNIT_HAS_CGROUP_CONTEXT(u)) | |
4221 | return; | |
4222 | ||
4223 | /* We dropped the inotify watch during reexec/reload, so we need to | |
4224 | * check these as they may have changed. | |
4225 | * Note that (currently) the kernel doesn't actually update cgroup | |
4226 | * file modification times, so we can't just serialize and then check | |
4227 | * the mtime for file(s) we are interested in. */ | |
4228 | (void) unit_check_cgroup_events(u); | |
4229 | unit_add_to_cgroup_oom_queue(u); | |
4230 | } | |
4231 | ||
1d9cc876 LP |
4232 | bool unit_cgroup_delegate(Unit *u) { |
4233 | CGroupContext *c; | |
4234 | ||
4235 | assert(u); | |
4236 | ||
4237 | if (!UNIT_VTABLE(u)->can_delegate) | |
4238 | return false; | |
4239 | ||
4240 | c = unit_get_cgroup_context(u); | |
4241 | if (!c) | |
4242 | return false; | |
4243 | ||
4244 | return c->delegate; | |
4245 | } | |
4246 | ||
e7ab4d1a | 4247 | void manager_invalidate_startup_units(Manager *m) { |
e7ab4d1a LP |
4248 | Unit *u; |
4249 | ||
4250 | assert(m); | |
4251 | ||
90e74a66 | 4252 | SET_FOREACH(u, m->startup_units) |
9dfb6a3a | 4253 | unit_invalidate_cgroup(u, CGROUP_MASK_CPU|CGROUP_MASK_IO|CGROUP_MASK_BLKIO|CGROUP_MASK_CPUSET); |
e7ab4d1a LP |
4254 | } |
4255 | ||
da8e1782 MO |
4256 | static int unit_get_nice(Unit *u) { |
4257 | ExecContext *ec; | |
4258 | ||
4259 | ec = unit_get_exec_context(u); | |
4260 | return ec ? ec->nice : 0; | |
4261 | } | |
4262 | ||
4263 | static uint64_t unit_get_cpu_weight(Unit *u) { | |
4264 | ManagerState state = manager_state(u->manager); | |
4265 | CGroupContext *cc; | |
4266 | ||
4267 | cc = unit_get_cgroup_context(u); | |
4268 | return cc ? cgroup_context_cpu_weight(cc, state) : CGROUP_WEIGHT_DEFAULT; | |
4269 | } | |
4270 | ||
4271 | int compare_job_priority(const void *a, const void *b) { | |
4272 | const Job *x = a, *y = b; | |
4273 | int nice_x, nice_y; | |
4274 | uint64_t weight_x, weight_y; | |
4275 | int ret; | |
4276 | ||
217b7b33 ZJS |
4277 | if ((ret = CMP(x->unit->type, y->unit->type)) != 0) |
4278 | return -ret; | |
4279 | ||
da8e1782 MO |
4280 | weight_x = unit_get_cpu_weight(x->unit); |
4281 | weight_y = unit_get_cpu_weight(y->unit); | |
4282 | ||
217b7b33 ZJS |
4283 | if ((ret = CMP(weight_x, weight_y)) != 0) |
4284 | return -ret; | |
da8e1782 MO |
4285 | |
4286 | nice_x = unit_get_nice(x->unit); | |
4287 | nice_y = unit_get_nice(y->unit); | |
4288 | ||
4289 | if ((ret = CMP(nice_x, nice_y)) != 0) | |
4290 | return ret; | |
4291 | ||
da8e1782 MO |
4292 | return strcmp(x->unit->id, y->unit->id); |
4293 | } | |
4294 | ||
d9e45bc3 MS |
4295 | int unit_cgroup_freezer_action(Unit *u, FreezerAction action) { |
4296 | _cleanup_free_ char *path = NULL; | |
4297 | FreezerState target, kernel = _FREEZER_STATE_INVALID; | |
7fcd2697 | 4298 | int r, ret; |
d9e45bc3 MS |
4299 | |
4300 | assert(u); | |
4301 | assert(IN_SET(action, FREEZER_FREEZE, FREEZER_THAW)); | |
4302 | ||
9a1e90ae MS |
4303 | if (!cg_freezer_supported()) |
4304 | return 0; | |
4305 | ||
a14137d9 | 4306 | /* Ignore all requests to thaw init.scope or -.slice and reject all requests to freeze them */ |
4307 | if (unit_has_name(u, SPECIAL_ROOT_SLICE) || unit_has_name(u, SPECIAL_INIT_SCOPE)) | |
4308 | return action == FREEZER_FREEZE ? -EPERM : 0; | |
4309 | ||
d9e45bc3 MS |
4310 | if (!u->cgroup_realized) |
4311 | return -EBUSY; | |
4312 | ||
a14137d9 | 4313 | if (action == FREEZER_THAW) { |
4314 | Unit *slice = UNIT_GET_SLICE(u); | |
4315 | ||
4316 | if (slice) { | |
4317 | r = unit_cgroup_freezer_action(slice, FREEZER_THAW); | |
4318 | if (r < 0) | |
4319 | return log_unit_error_errno(u, r, "Failed to thaw slice %s of unit: %m", slice->id); | |
4320 | } | |
4321 | } | |
4322 | ||
d9e45bc3 MS |
4323 | target = action == FREEZER_FREEZE ? FREEZER_FROZEN : FREEZER_RUNNING; |
4324 | ||
4325 | r = unit_freezer_state_kernel(u, &kernel); | |
4326 | if (r < 0) | |
4327 | log_unit_debug_errno(u, r, "Failed to obtain cgroup freezer state: %m"); | |
4328 | ||
4329 | if (target == kernel) { | |
4330 | u->freezer_state = target; | |
7fcd2697 | 4331 | if (action == FREEZER_FREEZE) |
4332 | return 0; | |
4333 | ret = 0; | |
4334 | } else | |
4335 | ret = 1; | |
d9e45bc3 MS |
4336 | |
4337 | r = cg_get_path(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path, "cgroup.freeze", &path); | |
4338 | if (r < 0) | |
4339 | return r; | |
4340 | ||
4341 | log_unit_debug(u, "%s unit.", action == FREEZER_FREEZE ? "Freezing" : "Thawing"); | |
4342 | ||
7fcd2697 | 4343 | if (target != kernel) { |
4344 | if (action == FREEZER_FREEZE) | |
4345 | u->freezer_state = FREEZER_FREEZING; | |
4346 | else | |
4347 | u->freezer_state = FREEZER_THAWING; | |
4348 | } | |
d9e45bc3 MS |
4349 | |
4350 | r = write_string_file(path, one_zero(action == FREEZER_FREEZE), WRITE_STRING_FILE_DISABLE_BUFFER); | |
4351 | if (r < 0) | |
4352 | return r; | |
4353 | ||
7fcd2697 | 4354 | return ret; |
d9e45bc3 MS |
4355 | } |
4356 | ||
047f5d63 PH |
4357 | int unit_get_cpuset(Unit *u, CPUSet *cpus, const char *name) { |
4358 | _cleanup_free_ char *v = NULL; | |
4359 | int r; | |
4360 | ||
4361 | assert(u); | |
4362 | assert(cpus); | |
4363 | ||
4364 | if (!u->cgroup_path) | |
4365 | return -ENODATA; | |
4366 | ||
4367 | if ((u->cgroup_realized_mask & CGROUP_MASK_CPUSET) == 0) | |
4368 | return -ENODATA; | |
4369 | ||
4370 | r = cg_all_unified(); | |
4371 | if (r < 0) | |
4372 | return r; | |
4373 | if (r == 0) | |
4374 | return -ENODATA; | |
48fd01e5 LP |
4375 | |
4376 | r = cg_get_attribute("cpuset", u->cgroup_path, name, &v); | |
047f5d63 PH |
4377 | if (r == -ENOENT) |
4378 | return -ENODATA; | |
4379 | if (r < 0) | |
4380 | return r; | |
4381 | ||
4382 | return parse_cpu_set_full(v, cpus, false, NULL, NULL, 0, NULL); | |
4383 | } | |
4384 | ||
4e806bfa AZ |
4385 | static const char* const cgroup_device_policy_table[_CGROUP_DEVICE_POLICY_MAX] = { |
4386 | [CGROUP_DEVICE_POLICY_AUTO] = "auto", | |
4387 | [CGROUP_DEVICE_POLICY_CLOSED] = "closed", | |
4388 | [CGROUP_DEVICE_POLICY_STRICT] = "strict", | |
4389 | }; | |
4390 | ||
4ad49000 | 4391 | DEFINE_STRING_TABLE_LOOKUP(cgroup_device_policy, CGroupDevicePolicy); |
d9e45bc3 MS |
4392 | |
4393 | static const char* const freezer_action_table[_FREEZER_ACTION_MAX] = { | |
4394 | [FREEZER_FREEZE] = "freeze", | |
4395 | [FREEZER_THAW] = "thaw", | |
4396 | }; | |
4397 | ||
4398 | DEFINE_STRING_TABLE_LOOKUP(freezer_action, FreezerAction); | |
6bb00842 LP |
4399 | |
4400 | static const char* const cgroup_pressure_watch_table[_CGROUP_PRESSURE_WATCH_MAX] = { | |
4401 | [CGROUP_PRESSURE_WATCH_OFF] = "off", | |
4402 | [CGROUP_PRESSURE_WATCH_AUTO] = "auto", | |
4403 | [CGROUP_PRESSURE_WATCH_ON] = "on", | |
4404 | [CGROUP_PRESSURE_WATCH_SKIP] = "skip", | |
4405 | }; | |
4406 | ||
4407 | DEFINE_STRING_TABLE_LOOKUP_WITH_BOOLEAN(cgroup_pressure_watch, CGroupPressureWatch, CGROUP_PRESSURE_WATCH_ON); |