]>
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 || | |
94 | c->startup_cpuset_mems.set; | |
95 | } | |
96 | ||
611c4f8a | 97 | bool unit_has_host_root_cgroup(Unit *u) { |
f3725e64 LP |
98 | assert(u); |
99 | ||
cc6271f1 LP |
100 | /* Returns whether this unit manages the root cgroup. This will return true if this unit is the root slice and |
101 | * the manager manages the root cgroup. */ | |
f3725e64 | 102 | |
611c4f8a | 103 | if (!manager_owns_host_root_cgroup(u->manager)) |
f3725e64 LP |
104 | return false; |
105 | ||
cc6271f1 | 106 | return unit_has_name(u, SPECIAL_ROOT_SLICE); |
f3725e64 LP |
107 | } |
108 | ||
293d32df LP |
109 | static int set_attribute_and_warn(Unit *u, const char *controller, const char *attribute, const char *value) { |
110 | int r; | |
111 | ||
112 | r = cg_set_attribute(controller, u->cgroup_path, attribute, value); | |
113 | if (r < 0) | |
8ed6f81b | 114 | log_unit_full_errno(u, LOG_LEVEL_CGROUP_WRITE(r), r, "Failed to set '%s' attribute on '%s' to '%.*s': %m", |
6178e2f8 | 115 | strna(attribute), empty_to_root(u->cgroup_path), (int) strcspn(value, NEWLINE), value); |
293d32df LP |
116 | |
117 | return r; | |
118 | } | |
119 | ||
2b40998d | 120 | static void cgroup_compat_warn(void) { |
128fadc9 TH |
121 | static bool cgroup_compat_warned = false; |
122 | ||
123 | if (cgroup_compat_warned) | |
124 | return; | |
125 | ||
cc6271f1 LP |
126 | log_warning("cgroup compatibility translation between legacy and unified hierarchy settings activated. " |
127 | "See cgroup-compat debug messages for details."); | |
128 | ||
128fadc9 TH |
129 | cgroup_compat_warned = true; |
130 | } | |
131 | ||
132 | #define log_cgroup_compat(unit, fmt, ...) do { \ | |
133 | cgroup_compat_warn(); \ | |
134 | log_unit_debug(unit, "cgroup-compat: " fmt, ##__VA_ARGS__); \ | |
2b40998d | 135 | } while (false) |
128fadc9 | 136 | |
4ad49000 LP |
137 | void cgroup_context_init(CGroupContext *c) { |
138 | assert(c); | |
139 | ||
de8a711a | 140 | /* Initialize everything to the kernel defaults. */ |
4ad49000 | 141 | |
de8a711a LP |
142 | *c = (CGroupContext) { |
143 | .cpu_weight = CGROUP_WEIGHT_INVALID, | |
144 | .startup_cpu_weight = CGROUP_WEIGHT_INVALID, | |
145 | .cpu_quota_per_sec_usec = USEC_INFINITY, | |
10f28641 | 146 | .cpu_quota_period_usec = USEC_INFINITY, |
66ebf6c0 | 147 | |
de8a711a LP |
148 | .cpu_shares = CGROUP_CPU_SHARES_INVALID, |
149 | .startup_cpu_shares = CGROUP_CPU_SHARES_INVALID, | |
d53d9474 | 150 | |
de8a711a LP |
151 | .memory_high = CGROUP_LIMIT_MAX, |
152 | .memory_max = CGROUP_LIMIT_MAX, | |
153 | .memory_swap_max = CGROUP_LIMIT_MAX, | |
da4d897e | 154 | |
de8a711a | 155 | .memory_limit = CGROUP_LIMIT_MAX, |
b2f8b02e | 156 | |
de8a711a LP |
157 | .io_weight = CGROUP_WEIGHT_INVALID, |
158 | .startup_io_weight = CGROUP_WEIGHT_INVALID, | |
13c31542 | 159 | |
de8a711a LP |
160 | .blockio_weight = CGROUP_BLKIO_WEIGHT_INVALID, |
161 | .startup_blockio_weight = CGROUP_BLKIO_WEIGHT_INVALID, | |
d53d9474 | 162 | |
3a0f06c4 | 163 | .tasks_max = TASKS_MAX_UNSET, |
4d824a4e AZ |
164 | |
165 | .moom_swap = MANAGED_OOM_AUTO, | |
166 | .moom_mem_pressure = MANAGED_OOM_AUTO, | |
4e806bfa | 167 | .moom_preference = MANAGED_OOM_PREFERENCE_NONE, |
de8a711a | 168 | }; |
4ad49000 | 169 | } |
8e274523 | 170 | |
4ad49000 LP |
171 | void cgroup_context_free_device_allow(CGroupContext *c, CGroupDeviceAllow *a) { |
172 | assert(c); | |
173 | assert(a); | |
174 | ||
71fda00f | 175 | LIST_REMOVE(device_allow, c->device_allow, a); |
4ad49000 LP |
176 | free(a->path); |
177 | free(a); | |
178 | } | |
179 | ||
13c31542 TH |
180 | void cgroup_context_free_io_device_weight(CGroupContext *c, CGroupIODeviceWeight *w) { |
181 | assert(c); | |
182 | assert(w); | |
183 | ||
184 | LIST_REMOVE(device_weights, c->io_device_weights, w); | |
185 | free(w->path); | |
186 | free(w); | |
187 | } | |
188 | ||
6ae4283c TH |
189 | void cgroup_context_free_io_device_latency(CGroupContext *c, CGroupIODeviceLatency *l) { |
190 | assert(c); | |
191 | assert(l); | |
192 | ||
193 | LIST_REMOVE(device_latencies, c->io_device_latencies, l); | |
194 | free(l->path); | |
195 | free(l); | |
196 | } | |
197 | ||
13c31542 TH |
198 | void cgroup_context_free_io_device_limit(CGroupContext *c, CGroupIODeviceLimit *l) { |
199 | assert(c); | |
200 | assert(l); | |
201 | ||
202 | LIST_REMOVE(device_limits, c->io_device_limits, l); | |
203 | free(l->path); | |
204 | free(l); | |
205 | } | |
206 | ||
4ad49000 LP |
207 | void cgroup_context_free_blockio_device_weight(CGroupContext *c, CGroupBlockIODeviceWeight *w) { |
208 | assert(c); | |
209 | assert(w); | |
210 | ||
71fda00f | 211 | LIST_REMOVE(device_weights, c->blockio_device_weights, w); |
4ad49000 LP |
212 | free(w->path); |
213 | free(w); | |
214 | } | |
215 | ||
216 | void cgroup_context_free_blockio_device_bandwidth(CGroupContext *c, CGroupBlockIODeviceBandwidth *b) { | |
217 | assert(c); | |
8e274523 | 218 | assert(b); |
8e274523 | 219 | |
71fda00f | 220 | LIST_REMOVE(device_bandwidths, c->blockio_device_bandwidths, b); |
4ad49000 LP |
221 | free(b->path); |
222 | free(b); | |
223 | } | |
224 | ||
b894ef1b JK |
225 | void cgroup_context_remove_bpf_foreign_program(CGroupContext *c, CGroupBPFForeignProgram *p) { |
226 | assert(c); | |
227 | assert(p); | |
228 | ||
229 | LIST_REMOVE(programs, c->bpf_foreign_programs, p); | |
230 | free(p->bpffs_path); | |
231 | free(p); | |
232 | } | |
233 | ||
b18e9fc1 | 234 | void cgroup_context_remove_socket_bind(CGroupSocketBindItem **head) { |
b18e9fc1 JK |
235 | assert(head); |
236 | ||
237 | while (*head) { | |
a67abc49 | 238 | CGroupSocketBindItem *h = *head; |
b18e9fc1 JK |
239 | LIST_REMOVE(socket_bind_items, *head, h); |
240 | free(h); | |
241 | } | |
242 | } | |
243 | ||
4ad49000 LP |
244 | void cgroup_context_done(CGroupContext *c) { |
245 | assert(c); | |
246 | ||
13c31542 TH |
247 | while (c->io_device_weights) |
248 | cgroup_context_free_io_device_weight(c, c->io_device_weights); | |
249 | ||
6ae4283c TH |
250 | while (c->io_device_latencies) |
251 | cgroup_context_free_io_device_latency(c, c->io_device_latencies); | |
252 | ||
13c31542 TH |
253 | while (c->io_device_limits) |
254 | cgroup_context_free_io_device_limit(c, c->io_device_limits); | |
255 | ||
4ad49000 LP |
256 | while (c->blockio_device_weights) |
257 | cgroup_context_free_blockio_device_weight(c, c->blockio_device_weights); | |
258 | ||
259 | while (c->blockio_device_bandwidths) | |
260 | cgroup_context_free_blockio_device_bandwidth(c, c->blockio_device_bandwidths); | |
261 | ||
262 | while (c->device_allow) | |
263 | cgroup_context_free_device_allow(c, c->device_allow); | |
6a48d82f | 264 | |
b18e9fc1 JK |
265 | cgroup_context_remove_socket_bind(&c->socket_bind_allow); |
266 | cgroup_context_remove_socket_bind(&c->socket_bind_deny); | |
267 | ||
84ebe6f0 YW |
268 | c->ip_address_allow = set_free(c->ip_address_allow); |
269 | c->ip_address_deny = set_free(c->ip_address_deny); | |
fab34748 KL |
270 | |
271 | c->ip_filters_ingress = strv_free(c->ip_filters_ingress); | |
272 | c->ip_filters_egress = strv_free(c->ip_filters_egress); | |
047f5d63 | 273 | |
b894ef1b JK |
274 | while (c->bpf_foreign_programs) |
275 | cgroup_context_remove_bpf_foreign_program(c, c->bpf_foreign_programs); | |
276 | ||
6f50d4f7 MV |
277 | c->restrict_network_interfaces = set_free(c->restrict_network_interfaces); |
278 | ||
047f5d63 | 279 | cpu_set_reset(&c->cpuset_cpus); |
31d3a520 | 280 | cpu_set_reset(&c->startup_cpuset_cpus); |
047f5d63 | 281 | cpu_set_reset(&c->cpuset_mems); |
31d3a520 | 282 | cpu_set_reset(&c->startup_cpuset_mems); |
4ad49000 LP |
283 | } |
284 | ||
74b5fb27 | 285 | static int unit_get_kernel_memory_limit(Unit *u, const char *file, uint64_t *ret) { |
74b5fb27 CD |
286 | assert(u); |
287 | ||
288 | if (!u->cgroup_realized) | |
289 | return -EOWNERDEAD; | |
290 | ||
613328c3 | 291 | return cg_get_attribute_as_uint64("memory", u->cgroup_path, file, ret); |
74b5fb27 CD |
292 | } |
293 | ||
294 | static int unit_compare_memory_limit(Unit *u, const char *property_name, uint64_t *ret_unit_value, uint64_t *ret_kernel_value) { | |
295 | CGroupContext *c; | |
296 | CGroupMask m; | |
297 | const char *file; | |
298 | uint64_t unit_value; | |
299 | int r; | |
300 | ||
301 | /* Compare kernel memcg configuration against our internal systemd state. Unsupported (and will | |
302 | * return -ENODATA) on cgroup v1. | |
303 | * | |
304 | * Returns: | |
305 | * | |
306 | * <0: On error. | |
307 | * 0: If the kernel memory setting doesn't match our configuration. | |
308 | * >0: If the kernel memory setting matches our configuration. | |
309 | * | |
310 | * The following values are only guaranteed to be populated on return >=0: | |
311 | * | |
312 | * - ret_unit_value will contain our internal expected value for the unit, page-aligned. | |
313 | * - ret_kernel_value will contain the actual value presented by the kernel. */ | |
314 | ||
315 | assert(u); | |
316 | ||
317 | r = cg_all_unified(); | |
318 | if (r < 0) | |
319 | return log_debug_errno(r, "Failed to determine cgroup hierarchy version: %m"); | |
320 | ||
321 | /* Unsupported on v1. | |
322 | * | |
323 | * We don't return ENOENT, since that could actually mask a genuine problem where somebody else has | |
324 | * silently masked the controller. */ | |
325 | if (r == 0) | |
326 | return -ENODATA; | |
327 | ||
328 | /* The root slice doesn't have any controller files, so we can't compare anything. */ | |
329 | if (unit_has_name(u, SPECIAL_ROOT_SLICE)) | |
330 | return -ENODATA; | |
331 | ||
332 | /* It's possible to have MemoryFoo set without systemd wanting to have the memory controller enabled, | |
333 | * for example, in the case of DisableControllers= or cgroup_disable on the kernel command line. To | |
334 | * avoid specious errors in these scenarios, check that we even expect the memory controller to be | |
335 | * enabled at all. */ | |
336 | m = unit_get_target_mask(u); | |
337 | if (!FLAGS_SET(m, CGROUP_MASK_MEMORY)) | |
338 | return -ENODATA; | |
339 | ||
806a9362 | 340 | assert_se(c = unit_get_cgroup_context(u)); |
74b5fb27 CD |
341 | |
342 | if (streq(property_name, "MemoryLow")) { | |
343 | unit_value = unit_get_ancestor_memory_low(u); | |
344 | file = "memory.low"; | |
345 | } else if (streq(property_name, "MemoryMin")) { | |
346 | unit_value = unit_get_ancestor_memory_min(u); | |
347 | file = "memory.min"; | |
348 | } else if (streq(property_name, "MemoryHigh")) { | |
349 | unit_value = c->memory_high; | |
350 | file = "memory.high"; | |
351 | } else if (streq(property_name, "MemoryMax")) { | |
352 | unit_value = c->memory_max; | |
353 | file = "memory.max"; | |
354 | } else if (streq(property_name, "MemorySwapMax")) { | |
355 | unit_value = c->memory_swap_max; | |
356 | file = "memory.swap.max"; | |
357 | } else | |
358 | return -EINVAL; | |
359 | ||
360 | r = unit_get_kernel_memory_limit(u, file, ret_kernel_value); | |
361 | if (r < 0) | |
362 | return log_unit_debug_errno(u, r, "Failed to parse %s: %m", file); | |
363 | ||
364 | /* It's intended (soon) in a future kernel to not expose cgroup memory limits rounded to page | |
365 | * boundaries, but instead separate the user-exposed limit, which is whatever userspace told us, from | |
366 | * our internal page-counting. To support those future kernels, just check the value itself first | |
367 | * without any page-alignment. */ | |
368 | if (*ret_kernel_value == unit_value) { | |
369 | *ret_unit_value = unit_value; | |
370 | return 1; | |
371 | } | |
372 | ||
373 | /* The current kernel behaviour, by comparison, is that even if you write a particular number of | |
374 | * bytes into a cgroup memory file, it always returns that number page-aligned down (since the kernel | |
375 | * internally stores cgroup limits in pages). As such, so long as it aligns properly, everything is | |
376 | * cricket. */ | |
377 | if (unit_value != CGROUP_LIMIT_MAX) | |
378 | unit_value = PAGE_ALIGN_DOWN(unit_value); | |
379 | ||
380 | *ret_unit_value = unit_value; | |
381 | ||
382 | return *ret_kernel_value == *ret_unit_value; | |
383 | } | |
384 | ||
bc0623df CD |
385 | #define FORMAT_CGROUP_DIFF_MAX 128 |
386 | ||
387 | static char *format_cgroup_memory_limit_comparison(char *buf, size_t l, Unit *u, const char *property_name) { | |
388 | uint64_t kval, sval; | |
389 | int r; | |
390 | ||
391 | assert(u); | |
392 | assert(buf); | |
393 | assert(l > 0); | |
394 | ||
395 | r = unit_compare_memory_limit(u, property_name, &sval, &kval); | |
396 | ||
397 | /* memory.swap.max is special in that it relies on CONFIG_MEMCG_SWAP (and the default swapaccount=1). | |
398 | * In the absence of reliably being able to detect whether memcg swap support is available or not, | |
399 | * only complain if the error is not ENOENT. */ | |
400 | if (r > 0 || IN_SET(r, -ENODATA, -EOWNERDEAD) || | |
401 | (r == -ENOENT && streq(property_name, "MemorySwapMax"))) { | |
402 | buf[0] = 0; | |
403 | return buf; | |
404 | } | |
405 | ||
406 | if (r < 0) { | |
121ed16c | 407 | (void) snprintf(buf, l, " (error getting kernel value: %s)", strerror_safe(r)); |
bc0623df CD |
408 | return buf; |
409 | } | |
410 | ||
121ed16c | 411 | (void) snprintf(buf, l, " (different value in kernel: %" PRIu64 ")", kval); |
bc0623df CD |
412 | |
413 | return buf; | |
414 | } | |
415 | ||
416 | void cgroup_context_dump(Unit *u, FILE* f, const char *prefix) { | |
31d3a520 | 417 | _cleanup_free_ char *disable_controllers_str = NULL, *cpuset_cpus = NULL, *cpuset_mems = NULL, *startup_cpuset_cpus = NULL, *startup_cpuset_mems = NULL; |
bc0623df | 418 | CGroupContext *c; |
84ebe6f0 | 419 | struct in_addr_prefix *iaai; |
4ad49000 | 420 | |
bc0623df CD |
421 | char cda[FORMAT_CGROUP_DIFF_MAX]; |
422 | char cdb[FORMAT_CGROUP_DIFF_MAX]; | |
423 | char cdc[FORMAT_CGROUP_DIFF_MAX]; | |
424 | char cdd[FORMAT_CGROUP_DIFF_MAX]; | |
425 | char cde[FORMAT_CGROUP_DIFF_MAX]; | |
426 | ||
427 | assert(u); | |
4ad49000 LP |
428 | assert(f); |
429 | ||
806a9362 | 430 | assert_se(c = unit_get_cgroup_context(u)); |
bc0623df | 431 | |
4ad49000 LP |
432 | prefix = strempty(prefix); |
433 | ||
25cc30c4 AZ |
434 | (void) cg_mask_to_string(c->disable_controllers, &disable_controllers_str); |
435 | ||
047f5d63 | 436 | cpuset_cpus = cpu_set_to_range_string(&c->cpuset_cpus); |
31d3a520 | 437 | startup_cpuset_cpus = cpu_set_to_range_string(&c->startup_cpuset_cpus); |
047f5d63 | 438 | cpuset_mems = cpu_set_to_range_string(&c->cpuset_mems); |
31d3a520 | 439 | startup_cpuset_mems = cpu_set_to_range_string(&c->startup_cpuset_mems); |
047f5d63 | 440 | |
4ad49000 | 441 | fprintf(f, |
6dfb9282 CD |
442 | "%sCPUAccounting: %s\n" |
443 | "%sIOAccounting: %s\n" | |
444 | "%sBlockIOAccounting: %s\n" | |
445 | "%sMemoryAccounting: %s\n" | |
446 | "%sTasksAccounting: %s\n" | |
447 | "%sIPAccounting: %s\n" | |
448 | "%sCPUWeight: %" PRIu64 "\n" | |
449 | "%sStartupCPUWeight: %" PRIu64 "\n" | |
450 | "%sCPUShares: %" PRIu64 "\n" | |
451 | "%sStartupCPUShares: %" PRIu64 "\n" | |
452 | "%sCPUQuotaPerSecSec: %s\n" | |
453 | "%sCPUQuotaPeriodSec: %s\n" | |
454 | "%sAllowedCPUs: %s\n" | |
31d3a520 | 455 | "%sStartupAllowedCPUs: %s\n" |
6dfb9282 | 456 | "%sAllowedMemoryNodes: %s\n" |
31d3a520 | 457 | "%sStartupAllowedMemoryNodes: %s\n" |
6dfb9282 CD |
458 | "%sIOWeight: %" PRIu64 "\n" |
459 | "%sStartupIOWeight: %" PRIu64 "\n" | |
460 | "%sBlockIOWeight: %" PRIu64 "\n" | |
461 | "%sStartupBlockIOWeight: %" PRIu64 "\n" | |
462 | "%sDefaultMemoryMin: %" PRIu64 "\n" | |
463 | "%sDefaultMemoryLow: %" PRIu64 "\n" | |
bc0623df CD |
464 | "%sMemoryMin: %" PRIu64 "%s\n" |
465 | "%sMemoryLow: %" PRIu64 "%s\n" | |
466 | "%sMemoryHigh: %" PRIu64 "%s\n" | |
467 | "%sMemoryMax: %" PRIu64 "%s\n" | |
468 | "%sMemorySwapMax: %" PRIu64 "%s\n" | |
6dfb9282 CD |
469 | "%sMemoryLimit: %" PRIu64 "\n" |
470 | "%sTasksMax: %" PRIu64 "\n" | |
471 | "%sDevicePolicy: %s\n" | |
472 | "%sDisableControllers: %s\n" | |
4d824a4e AZ |
473 | "%sDelegate: %s\n" |
474 | "%sManagedOOMSwap: %s\n" | |
475 | "%sManagedOOMMemoryPressure: %s\n" | |
d9d3f05d | 476 | "%sManagedOOMMemoryPressureLimit: " PERMYRIAD_AS_PERCENT_FORMAT_STR "\n" |
0f6bb1ed | 477 | "%sManagedOOMPreference: %s\n", |
4ad49000 | 478 | prefix, yes_no(c->cpu_accounting), |
13c31542 | 479 | prefix, yes_no(c->io_accounting), |
4ad49000 LP |
480 | prefix, yes_no(c->blockio_accounting), |
481 | prefix, yes_no(c->memory_accounting), | |
d53d9474 | 482 | prefix, yes_no(c->tasks_accounting), |
c21c9906 | 483 | prefix, yes_no(c->ip_accounting), |
66ebf6c0 TH |
484 | prefix, c->cpu_weight, |
485 | prefix, c->startup_cpu_weight, | |
4ad49000 | 486 | prefix, c->cpu_shares, |
95ae05c0 | 487 | prefix, c->startup_cpu_shares, |
5291f26d ZJS |
488 | prefix, FORMAT_TIMESPAN(c->cpu_quota_per_sec_usec, 1), |
489 | prefix, FORMAT_TIMESPAN(c->cpu_quota_period_usec, 1), | |
85c3b278 | 490 | prefix, strempty(cpuset_cpus), |
31d3a520 | 491 | prefix, strempty(startup_cpuset_cpus), |
85c3b278 | 492 | prefix, strempty(cpuset_mems), |
31d3a520 | 493 | prefix, strempty(startup_cpuset_mems), |
13c31542 TH |
494 | prefix, c->io_weight, |
495 | prefix, c->startup_io_weight, | |
4ad49000 | 496 | prefix, c->blockio_weight, |
95ae05c0 | 497 | prefix, c->startup_blockio_weight, |
7ad5439e | 498 | prefix, c->default_memory_min, |
c52db42b | 499 | prefix, c->default_memory_low, |
bc0623df CD |
500 | prefix, c->memory_min, format_cgroup_memory_limit_comparison(cda, sizeof(cda), u, "MemoryMin"), |
501 | prefix, c->memory_low, format_cgroup_memory_limit_comparison(cdb, sizeof(cdb), u, "MemoryLow"), | |
502 | prefix, c->memory_high, format_cgroup_memory_limit_comparison(cdc, sizeof(cdc), u, "MemoryHigh"), | |
503 | prefix, c->memory_max, format_cgroup_memory_limit_comparison(cdd, sizeof(cdd), u, "MemoryMax"), | |
504 | prefix, c->memory_swap_max, format_cgroup_memory_limit_comparison(cde, sizeof(cde), u, "MemorySwapMax"), | |
4ad49000 | 505 | prefix, c->memory_limit, |
3a0f06c4 | 506 | prefix, tasks_max_resolve(&c->tasks_max), |
a931ad47 | 507 | prefix, cgroup_device_policy_to_string(c->device_policy), |
f4c43a81 | 508 | prefix, strempty(disable_controllers_str), |
4d824a4e AZ |
509 | prefix, yes_no(c->delegate), |
510 | prefix, managed_oom_mode_to_string(c->moom_swap), | |
511 | prefix, managed_oom_mode_to_string(c->moom_mem_pressure), | |
d9d3f05d | 512 | prefix, PERMYRIAD_AS_PERCENT_FORMAT_VAL(UINT32_SCALE_TO_PERMYRIAD(c->moom_mem_pressure_limit)), |
4e806bfa | 513 | prefix, managed_oom_preference_to_string(c->moom_preference)); |
4ad49000 | 514 | |
02638280 LP |
515 | if (c->delegate) { |
516 | _cleanup_free_ char *t = NULL; | |
517 | ||
518 | (void) cg_mask_to_string(c->delegate_controllers, &t); | |
519 | ||
6dfb9282 | 520 | fprintf(f, "%sDelegateControllers: %s\n", |
02638280 LP |
521 | prefix, |
522 | strempty(t)); | |
523 | } | |
524 | ||
4ad49000 LP |
525 | LIST_FOREACH(device_allow, a, c->device_allow) |
526 | fprintf(f, | |
6dfb9282 | 527 | "%sDeviceAllow: %s %s%s%s\n", |
4ad49000 LP |
528 | prefix, |
529 | a->path, | |
530 | a->r ? "r" : "", a->w ? "w" : "", a->m ? "m" : ""); | |
531 | ||
13c31542 TH |
532 | LIST_FOREACH(device_weights, iw, c->io_device_weights) |
533 | fprintf(f, | |
6dfb9282 | 534 | "%sIODeviceWeight: %s %" PRIu64 "\n", |
13c31542 TH |
535 | prefix, |
536 | iw->path, | |
537 | iw->weight); | |
538 | ||
6ae4283c TH |
539 | LIST_FOREACH(device_latencies, l, c->io_device_latencies) |
540 | fprintf(f, | |
6dfb9282 | 541 | "%sIODeviceLatencyTargetSec: %s %s\n", |
6ae4283c TH |
542 | prefix, |
543 | l->path, | |
5291f26d | 544 | FORMAT_TIMESPAN(l->target_usec, 1)); |
6ae4283c | 545 | |
2b59bf51 | 546 | LIST_FOREACH(device_limits, il, c->io_device_limits) |
e8616626 | 547 | for (CGroupIOLimitType type = 0; type < _CGROUP_IO_LIMIT_TYPE_MAX; type++) |
9be57249 TH |
548 | if (il->limits[type] != cgroup_io_limit_defaults[type]) |
549 | fprintf(f, | |
6dfb9282 | 550 | "%s%s: %s %s\n", |
9be57249 TH |
551 | prefix, |
552 | cgroup_io_limit_type_to_string(type), | |
553 | il->path, | |
2b59bf51 | 554 | FORMAT_BYTES(il->limits[type])); |
13c31542 | 555 | |
4ad49000 LP |
556 | LIST_FOREACH(device_weights, w, c->blockio_device_weights) |
557 | fprintf(f, | |
6dfb9282 | 558 | "%sBlockIODeviceWeight: %s %" PRIu64, |
4ad49000 LP |
559 | prefix, |
560 | w->path, | |
561 | w->weight); | |
562 | ||
563 | LIST_FOREACH(device_bandwidths, b, c->blockio_device_bandwidths) { | |
979d0311 TH |
564 | if (b->rbps != CGROUP_LIMIT_MAX) |
565 | fprintf(f, | |
6dfb9282 | 566 | "%sBlockIOReadBandwidth: %s %s\n", |
979d0311 TH |
567 | prefix, |
568 | b->path, | |
2b59bf51 | 569 | FORMAT_BYTES(b->rbps)); |
979d0311 TH |
570 | if (b->wbps != CGROUP_LIMIT_MAX) |
571 | fprintf(f, | |
6dfb9282 | 572 | "%sBlockIOWriteBandwidth: %s %s\n", |
979d0311 TH |
573 | prefix, |
574 | b->path, | |
2b59bf51 | 575 | FORMAT_BYTES(b->wbps)); |
4ad49000 | 576 | } |
c21c9906 | 577 | |
c71384a9 ZJS |
578 | SET_FOREACH(iaai, c->ip_address_allow) |
579 | fprintf(f, "%sIPAddressAllow: %s\n", prefix, | |
580 | IN_ADDR_PREFIX_TO_STRING(iaai->family, &iaai->address, iaai->prefixlen)); | |
581 | SET_FOREACH(iaai, c->ip_address_deny) | |
582 | fprintf(f, "%sIPAddressDeny: %s\n", prefix, | |
583 | IN_ADDR_PREFIX_TO_STRING(iaai->family, &iaai->address, iaai->prefixlen)); | |
fab34748 KL |
584 | |
585 | STRV_FOREACH(path, c->ip_filters_ingress) | |
6dfb9282 | 586 | fprintf(f, "%sIPIngressFilterPath: %s\n", prefix, *path); |
fab34748 | 587 | STRV_FOREACH(path, c->ip_filters_egress) |
6dfb9282 | 588 | fprintf(f, "%sIPEgressFilterPath: %s\n", prefix, *path); |
b894ef1b JK |
589 | |
590 | LIST_FOREACH(programs, p, c->bpf_foreign_programs) | |
591 | fprintf(f, "%sBPFProgram: %s:%s", | |
592 | prefix, bpf_cgroup_attach_type_to_string(p->attach_type), p->bpffs_path); | |
b18e9fc1 JK |
593 | |
594 | if (c->socket_bind_allow) { | |
595 | fprintf(f, "%sSocketBindAllow:", prefix); | |
596 | LIST_FOREACH(socket_bind_items, bi, c->socket_bind_allow) | |
597 | cgroup_context_dump_socket_bind_item(bi, f); | |
598 | fputc('\n', f); | |
599 | } | |
600 | ||
601 | if (c->socket_bind_deny) { | |
602 | fprintf(f, "%sSocketBindDeny:", prefix); | |
603 | LIST_FOREACH(socket_bind_items, bi, c->socket_bind_deny) | |
604 | cgroup_context_dump_socket_bind_item(bi, f); | |
605 | fputc('\n', f); | |
606 | } | |
6f50d4f7 MV |
607 | |
608 | if (c->restrict_network_interfaces) { | |
609 | char *iface; | |
610 | SET_FOREACH(iface, c->restrict_network_interfaces) | |
611 | fprintf(f, "%sRestrictNetworkInterfaces: %s\n", prefix, iface); | |
612 | } | |
b18e9fc1 JK |
613 | } |
614 | ||
615 | void cgroup_context_dump_socket_bind_item(const CGroupSocketBindItem *item, FILE *f) { | |
5587ce7f | 616 | const char *family, *colon1, *protocol = "", *colon2 = ""; |
a4817536 LP |
617 | |
618 | family = strempty(af_to_ipv4_ipv6(item->address_family)); | |
5587ce7f JK |
619 | colon1 = isempty(family) ? "" : ":"; |
620 | ||
621 | if (item->ip_protocol != 0) { | |
622 | protocol = ip_protocol_to_tcp_udp(item->ip_protocol); | |
623 | colon2 = ":"; | |
624 | } | |
b18e9fc1 JK |
625 | |
626 | if (item->nr_ports == 0) | |
5587ce7f | 627 | fprintf(f, " %s%s%s%sany", family, colon1, protocol, colon2); |
b18e9fc1 | 628 | else if (item->nr_ports == 1) |
5587ce7f | 629 | fprintf(f, " %s%s%s%s%" PRIu16, family, colon1, protocol, colon2, item->port_min); |
b18e9fc1 JK |
630 | else { |
631 | uint16_t port_max = item->port_min + item->nr_ports - 1; | |
5587ce7f JK |
632 | fprintf(f, " %s%s%s%s%" PRIu16 "-%" PRIu16, family, colon1, protocol, colon2, |
633 | item->port_min, port_max); | |
b18e9fc1 | 634 | } |
4ad49000 LP |
635 | } |
636 | ||
fd870bac YW |
637 | int cgroup_add_device_allow(CGroupContext *c, const char *dev, const char *mode) { |
638 | _cleanup_free_ CGroupDeviceAllow *a = NULL; | |
639 | _cleanup_free_ char *d = NULL; | |
640 | ||
641 | assert(c); | |
642 | assert(dev); | |
643 | assert(isempty(mode) || in_charset(mode, "rwm")); | |
644 | ||
645 | a = new(CGroupDeviceAllow, 1); | |
646 | if (!a) | |
647 | return -ENOMEM; | |
648 | ||
649 | d = strdup(dev); | |
650 | if (!d) | |
651 | return -ENOMEM; | |
652 | ||
653 | *a = (CGroupDeviceAllow) { | |
654 | .path = TAKE_PTR(d), | |
490c5a37 LP |
655 | .r = isempty(mode) || strchr(mode, 'r'), |
656 | .w = isempty(mode) || strchr(mode, 'w'), | |
657 | .m = isempty(mode) || strchr(mode, 'm'), | |
fd870bac YW |
658 | }; |
659 | ||
660 | LIST_PREPEND(device_allow, c->device_allow, a); | |
661 | TAKE_PTR(a); | |
662 | ||
663 | return 0; | |
664 | } | |
665 | ||
b894ef1b JK |
666 | int cgroup_add_bpf_foreign_program(CGroupContext *c, uint32_t attach_type, const char *bpffs_path) { |
667 | CGroupBPFForeignProgram *p; | |
668 | _cleanup_free_ char *d = NULL; | |
669 | ||
670 | assert(c); | |
671 | assert(bpffs_path); | |
672 | ||
673 | if (!path_is_normalized(bpffs_path) || !path_is_absolute(bpffs_path)) | |
674 | return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "Path is not normalized: %m"); | |
675 | ||
676 | d = strdup(bpffs_path); | |
677 | if (!d) | |
678 | return log_oom(); | |
679 | ||
680 | p = new(CGroupBPFForeignProgram, 1); | |
681 | if (!p) | |
682 | return log_oom(); | |
683 | ||
684 | *p = (CGroupBPFForeignProgram) { | |
685 | .attach_type = attach_type, | |
686 | .bpffs_path = TAKE_PTR(d), | |
687 | }; | |
688 | ||
689 | LIST_PREPEND(programs, c->bpf_foreign_programs, TAKE_PTR(p)); | |
690 | ||
691 | return 0; | |
692 | } | |
693 | ||
6264b85e CD |
694 | #define UNIT_DEFINE_ANCESTOR_MEMORY_LOOKUP(entry) \ |
695 | uint64_t unit_get_ancestor_##entry(Unit *u) { \ | |
696 | CGroupContext *c; \ | |
697 | \ | |
698 | /* 1. Is entry set in this unit? If so, use that. \ | |
699 | * 2. Is the default for this entry set in any \ | |
700 | * ancestor? If so, use that. \ | |
701 | * 3. Otherwise, return CGROUP_LIMIT_MIN. */ \ | |
702 | \ | |
703 | assert(u); \ | |
704 | \ | |
705 | c = unit_get_cgroup_context(u); \ | |
c5322608 | 706 | if (c && c->entry##_set) \ |
6264b85e CD |
707 | return c->entry; \ |
708 | \ | |
12f64221 | 709 | while ((u = UNIT_GET_SLICE(u))) { \ |
6264b85e | 710 | c = unit_get_cgroup_context(u); \ |
c5322608 | 711 | if (c && c->default_##entry##_set) \ |
6264b85e CD |
712 | return c->default_##entry; \ |
713 | } \ | |
714 | \ | |
715 | /* We've reached the root, but nobody had default for \ | |
716 | * this entry set, so set it to the kernel default. */ \ | |
717 | return CGROUP_LIMIT_MIN; \ | |
c52db42b CD |
718 | } |
719 | ||
6264b85e | 720 | UNIT_DEFINE_ANCESTOR_MEMORY_LOOKUP(memory_low); |
7ad5439e | 721 | UNIT_DEFINE_ANCESTOR_MEMORY_LOOKUP(memory_min); |
6264b85e | 722 | |
1fa3b6c2 LP |
723 | static void unit_set_xattr_graceful(Unit *u, const char *cgroup_path, const char *name, const void *data, size_t size) { |
724 | int r; | |
725 | ||
726 | assert(u); | |
727 | assert(name); | |
728 | ||
729 | if (!cgroup_path) { | |
730 | if (!u->cgroup_path) | |
731 | return; | |
732 | ||
733 | cgroup_path = u->cgroup_path; | |
734 | } | |
735 | ||
736 | r = cg_set_xattr(SYSTEMD_CGROUP_CONTROLLER, cgroup_path, name, data, size, 0); | |
737 | if (r < 0) | |
738 | log_unit_debug_errno(u, r, "Failed to set '%s' xattr on control group %s, ignoring: %m", name, empty_to_root(cgroup_path)); | |
739 | } | |
740 | ||
741 | static void unit_remove_xattr_graceful(Unit *u, const char *cgroup_path, const char *name) { | |
742 | int r; | |
743 | ||
744 | assert(u); | |
745 | assert(name); | |
746 | ||
747 | if (!cgroup_path) { | |
748 | if (!u->cgroup_path) | |
749 | return; | |
750 | ||
751 | cgroup_path = u->cgroup_path; | |
752 | } | |
753 | ||
754 | r = cg_remove_xattr(SYSTEMD_CGROUP_CONTROLLER, cgroup_path, name); | |
755 | if (r < 0 && r != -ENODATA) | |
756 | log_unit_debug_errno(u, r, "Failed to remove '%s' xattr flag on control group %s, ignoring: %m", name, empty_to_root(cgroup_path)); | |
757 | } | |
758 | ||
4e806bfa AZ |
759 | void cgroup_oomd_xattr_apply(Unit *u, const char *cgroup_path) { |
760 | CGroupContext *c; | |
4e806bfa AZ |
761 | |
762 | assert(u); | |
763 | ||
764 | c = unit_get_cgroup_context(u); | |
765 | if (!c) | |
766 | return; | |
767 | ||
1fa3b6c2 LP |
768 | if (c->moom_preference == MANAGED_OOM_PREFERENCE_OMIT) |
769 | unit_set_xattr_graceful(u, cgroup_path, "user.oomd_omit", "1", 1); | |
4e806bfa | 770 | |
1fa3b6c2 LP |
771 | if (c->moom_preference == MANAGED_OOM_PREFERENCE_AVOID) |
772 | unit_set_xattr_graceful(u, cgroup_path, "user.oomd_avoid", "1", 1); | |
4e806bfa | 773 | |
1fa3b6c2 LP |
774 | if (c->moom_preference != MANAGED_OOM_PREFERENCE_AVOID) |
775 | unit_remove_xattr_graceful(u, cgroup_path, "user.oomd_avoid"); | |
4e806bfa | 776 | |
1fa3b6c2 LP |
777 | if (c->moom_preference != MANAGED_OOM_PREFERENCE_OMIT) |
778 | unit_remove_xattr_graceful(u, cgroup_path, "user.oomd_omit"); | |
4e806bfa AZ |
779 | } |
780 | ||
0d2d6fbf | 781 | static void cgroup_xattr_apply(Unit *u) { |
d9bc1c36 | 782 | bool b; |
0d2d6fbf CD |
783 | |
784 | assert(u); | |
785 | ||
786 | if (!MANAGER_IS_SYSTEM(u->manager)) | |
787 | return; | |
788 | ||
1fa3b6c2 LP |
789 | b = !sd_id128_is_null(u->invocation_id); |
790 | FOREACH_STRING(xn, "trusted.invocation_id", "user.invocation_id") { | |
791 | if (b) | |
792 | unit_set_xattr_graceful(u, NULL, xn, SD_ID128_TO_STRING(u->invocation_id), 32); | |
793 | else | |
794 | unit_remove_xattr_graceful(u, NULL, xn); | |
3288ea8f | 795 | } |
0d2d6fbf | 796 | |
d9bc1c36 LP |
797 | /* Indicate on the cgroup whether delegation is on, via an xattr. This is best-effort, as old kernels |
798 | * didn't support xattrs on cgroups at all. Later they got support for setting 'trusted.*' xattrs, | |
799 | * and even later 'user.*' xattrs. We started setting this field when 'trusted.*' was added, and | |
800 | * given this is now pretty much API, let's continue to support that. But also set 'user.*' as well, | |
801 | * since it is readable by any user, not just CAP_SYS_ADMIN. This hence comes with slightly weaker | |
802 | * security (as users who got delegated cgroups could turn it off if they like), but this shouldn't | |
803 | * be a big problem given this communicates delegation state to clients, but the manager never reads | |
804 | * it. */ | |
805 | b = unit_cgroup_delegate(u); | |
806 | FOREACH_STRING(xn, "trusted.delegate", "user.delegate") { | |
1fa3b6c2 LP |
807 | if (b) |
808 | unit_set_xattr_graceful(u, NULL, xn, "1", 1); | |
809 | else | |
810 | unit_remove_xattr_graceful(u, NULL, xn); | |
3288ea8f | 811 | } |
4e806bfa AZ |
812 | |
813 | cgroup_oomd_xattr_apply(u, u->cgroup_path); | |
0d2d6fbf CD |
814 | } |
815 | ||
45c2e068 | 816 | static int lookup_block_device(const char *p, dev_t *ret) { |
f5855697 YS |
817 | dev_t rdev, dev = 0; |
818 | mode_t mode; | |
45c2e068 | 819 | int r; |
4ad49000 LP |
820 | |
821 | assert(p); | |
45c2e068 | 822 | assert(ret); |
4ad49000 | 823 | |
f5855697 | 824 | r = device_path_parse_major_minor(p, &mode, &rdev); |
d5aecba6 | 825 | if (r == -ENODEV) { /* not a parsable device node, need to go to disk */ |
f5855697 | 826 | struct stat st; |
57f1030b | 827 | |
d5aecba6 LP |
828 | if (stat(p, &st) < 0) |
829 | return log_warning_errno(errno, "Couldn't stat device '%s': %m", p); | |
57f1030b | 830 | |
f5855697 | 831 | mode = st.st_mode; |
a0d6590c LP |
832 | rdev = st.st_rdev; |
833 | dev = st.st_dev; | |
d5aecba6 LP |
834 | } else if (r < 0) |
835 | return log_warning_errno(r, "Failed to parse major/minor from path '%s': %m", p); | |
836 | ||
57f1030b LP |
837 | if (S_ISCHR(mode)) |
838 | return log_warning_errno(SYNTHETIC_ERRNO(ENOTBLK), | |
839 | "Device node '%s' is a character device, but block device needed.", p); | |
840 | if (S_ISBLK(mode)) | |
f5855697 YS |
841 | *ret = rdev; |
842 | else if (major(dev) != 0) | |
843 | *ret = dev; /* If this is not a device node then use the block device this file is stored on */ | |
45c2e068 LP |
844 | else { |
845 | /* If this is btrfs, getting the backing block device is a bit harder */ | |
846 | r = btrfs_get_block_device(p, ret); | |
57f1030b LP |
847 | if (r == -ENOTTY) |
848 | return log_warning_errno(SYNTHETIC_ERRNO(ENODEV), | |
849 | "'%s' is not a block device node, and file system block device cannot be determined or is not local.", p); | |
850 | if (r < 0) | |
45c2e068 | 851 | return log_warning_errno(r, "Failed to determine block device backing btrfs file system '%s': %m", p); |
4ad49000 | 852 | } |
8e274523 | 853 | |
b7cf4b4e BB |
854 | /* If this is a LUKS/DM device, recursively try to get the originating block device */ |
855 | while (block_get_originating(*ret, ret) > 0); | |
45c2e068 LP |
856 | |
857 | /* If this is a partition, try to get the originating block device */ | |
858 | (void) block_get_whole_disk(*ret, ret); | |
8e274523 | 859 | return 0; |
8e274523 LP |
860 | } |
861 | ||
66ebf6c0 TH |
862 | static bool cgroup_context_has_cpu_weight(CGroupContext *c) { |
863 | return c->cpu_weight != CGROUP_WEIGHT_INVALID || | |
864 | c->startup_cpu_weight != CGROUP_WEIGHT_INVALID; | |
865 | } | |
866 | ||
867 | static bool cgroup_context_has_cpu_shares(CGroupContext *c) { | |
868 | return c->cpu_shares != CGROUP_CPU_SHARES_INVALID || | |
869 | c->startup_cpu_shares != CGROUP_CPU_SHARES_INVALID; | |
870 | } | |
871 | ||
31d3a520 PM |
872 | static bool cgroup_context_has_allowed_cpus(CGroupContext *c) { |
873 | return c->cpuset_cpus.set || c->startup_cpuset_cpus.set; | |
874 | } | |
875 | ||
876 | static bool cgroup_context_has_allowed_mems(CGroupContext *c) { | |
877 | return c->cpuset_mems.set || c->startup_cpuset_mems.set; | |
878 | } | |
879 | ||
66ebf6c0 | 880 | static uint64_t cgroup_context_cpu_weight(CGroupContext *c, ManagerState state) { |
9dfb6a3a | 881 | if (IN_SET(state, MANAGER_STARTING, MANAGER_INITIALIZING, MANAGER_STOPPING) && |
66ebf6c0 TH |
882 | c->startup_cpu_weight != CGROUP_WEIGHT_INVALID) |
883 | return c->startup_cpu_weight; | |
884 | else if (c->cpu_weight != CGROUP_WEIGHT_INVALID) | |
885 | return c->cpu_weight; | |
886 | else | |
887 | return CGROUP_WEIGHT_DEFAULT; | |
888 | } | |
889 | ||
890 | static uint64_t cgroup_context_cpu_shares(CGroupContext *c, ManagerState state) { | |
9dfb6a3a | 891 | if (IN_SET(state, MANAGER_STARTING, MANAGER_INITIALIZING, MANAGER_STOPPING) && |
66ebf6c0 TH |
892 | c->startup_cpu_shares != CGROUP_CPU_SHARES_INVALID) |
893 | return c->startup_cpu_shares; | |
894 | else if (c->cpu_shares != CGROUP_CPU_SHARES_INVALID) | |
895 | return c->cpu_shares; | |
896 | else | |
897 | return CGROUP_CPU_SHARES_DEFAULT; | |
898 | } | |
899 | ||
31d3a520 | 900 | static CPUSet *cgroup_context_allowed_cpus(CGroupContext *c, ManagerState state) { |
9dfb6a3a | 901 | if (IN_SET(state, MANAGER_STARTING, MANAGER_INITIALIZING, MANAGER_STOPPING) && |
31d3a520 PM |
902 | c->startup_cpuset_cpus.set) |
903 | return &c->startup_cpuset_cpus; | |
904 | else | |
905 | return &c->cpuset_cpus; | |
906 | } | |
907 | ||
908 | static CPUSet *cgroup_context_allowed_mems(CGroupContext *c, ManagerState state) { | |
9dfb6a3a | 909 | if (IN_SET(state, MANAGER_STARTING, MANAGER_INITIALIZING, MANAGER_STOPPING) && |
31d3a520 PM |
910 | c->startup_cpuset_mems.set) |
911 | return &c->startup_cpuset_mems; | |
912 | else | |
913 | return &c->cpuset_mems; | |
914 | } | |
915 | ||
10f28641 FB |
916 | usec_t cgroup_cpu_adjust_period(usec_t period, usec_t quota, usec_t resolution, usec_t max_period) { |
917 | /* kernel uses a minimum resolution of 1ms, so both period and (quota * period) | |
918 | * need to be higher than that boundary. quota is specified in USecPerSec. | |
919 | * Additionally, period must be at most max_period. */ | |
920 | assert(quota > 0); | |
921 | ||
922 | return MIN(MAX3(period, resolution, resolution * USEC_PER_SEC / quota), max_period); | |
923 | } | |
924 | ||
925 | static usec_t cgroup_cpu_adjust_period_and_log(Unit *u, usec_t period, usec_t quota) { | |
926 | usec_t new_period; | |
927 | ||
928 | if (quota == USEC_INFINITY) | |
929 | /* Always use default period for infinity quota. */ | |
930 | return CGROUP_CPU_QUOTA_DEFAULT_PERIOD_USEC; | |
931 | ||
932 | if (period == USEC_INFINITY) | |
933 | /* Default period was requested. */ | |
934 | period = CGROUP_CPU_QUOTA_DEFAULT_PERIOD_USEC; | |
935 | ||
936 | /* Clamp to interval [1ms, 1s] */ | |
937 | new_period = cgroup_cpu_adjust_period(period, quota, USEC_PER_MSEC, USEC_PER_SEC); | |
938 | ||
939 | if (new_period != period) { | |
8ed6f81b | 940 | log_unit_full(u, u->warned_clamping_cpu_quota_period ? LOG_DEBUG : LOG_WARNING, |
10f28641 | 941 | "Clamping CPU interval for cpu.max: period is now %s", |
5291f26d | 942 | FORMAT_TIMESPAN(new_period, 1)); |
527ede0c | 943 | u->warned_clamping_cpu_quota_period = true; |
10f28641 FB |
944 | } |
945 | ||
946 | return new_period; | |
947 | } | |
948 | ||
52fecf20 LP |
949 | static void cgroup_apply_unified_cpu_weight(Unit *u, uint64_t weight) { |
950 | char buf[DECIMAL_STR_MAX(uint64_t) + 2]; | |
66ebf6c0 | 951 | |
c8340822 | 952 | if (weight == CGROUP_WEIGHT_IDLE) |
953 | return; | |
66ebf6c0 | 954 | xsprintf(buf, "%" PRIu64 "\n", weight); |
293d32df | 955 | (void) set_attribute_and_warn(u, "cpu", "cpu.weight", buf); |
52fecf20 LP |
956 | } |
957 | ||
c8340822 | 958 | static void cgroup_apply_unified_cpu_idle(Unit *u, uint64_t weight) { |
959 | int r; | |
960 | bool is_idle; | |
961 | const char *idle_val; | |
962 | ||
963 | is_idle = weight == CGROUP_WEIGHT_IDLE; | |
964 | idle_val = one_zero(is_idle); | |
965 | r = cg_set_attribute("cpu", u->cgroup_path, "cpu.idle", idle_val); | |
966 | if (r < 0 && (r != -ENOENT || is_idle)) | |
967 | log_unit_full_errno(u, LOG_LEVEL_CGROUP_WRITE(r), r, "Failed to set '%s' attribute on '%s' to '%s': %m", | |
968 | "cpu.idle", empty_to_root(u->cgroup_path), idle_val); | |
969 | } | |
970 | ||
10f28641 | 971 | static void cgroup_apply_unified_cpu_quota(Unit *u, usec_t quota, usec_t period) { |
52fecf20 | 972 | char buf[(DECIMAL_STR_MAX(usec_t) + 1) * 2 + 1]; |
66ebf6c0 | 973 | |
10f28641 | 974 | period = cgroup_cpu_adjust_period_and_log(u, period, quota); |
66ebf6c0 TH |
975 | if (quota != USEC_INFINITY) |
976 | xsprintf(buf, USEC_FMT " " USEC_FMT "\n", | |
10f28641 | 977 | MAX(quota * period / USEC_PER_SEC, USEC_PER_MSEC), period); |
66ebf6c0 | 978 | else |
10f28641 | 979 | xsprintf(buf, "max " USEC_FMT "\n", period); |
293d32df | 980 | (void) set_attribute_and_warn(u, "cpu", "cpu.max", buf); |
66ebf6c0 TH |
981 | } |
982 | ||
52fecf20 LP |
983 | static void cgroup_apply_legacy_cpu_shares(Unit *u, uint64_t shares) { |
984 | char buf[DECIMAL_STR_MAX(uint64_t) + 2]; | |
66ebf6c0 TH |
985 | |
986 | xsprintf(buf, "%" PRIu64 "\n", shares); | |
293d32df | 987 | (void) set_attribute_and_warn(u, "cpu", "cpu.shares", buf); |
52fecf20 LP |
988 | } |
989 | ||
10f28641 | 990 | static void cgroup_apply_legacy_cpu_quota(Unit *u, usec_t quota, usec_t period) { |
52fecf20 | 991 | char buf[DECIMAL_STR_MAX(usec_t) + 2]; |
66ebf6c0 | 992 | |
10f28641 FB |
993 | period = cgroup_cpu_adjust_period_and_log(u, period, quota); |
994 | ||
995 | xsprintf(buf, USEC_FMT "\n", period); | |
293d32df | 996 | (void) set_attribute_and_warn(u, "cpu", "cpu.cfs_period_us", buf); |
66ebf6c0 TH |
997 | |
998 | if (quota != USEC_INFINITY) { | |
10f28641 | 999 | xsprintf(buf, USEC_FMT "\n", MAX(quota * period / USEC_PER_SEC, USEC_PER_MSEC)); |
293d32df | 1000 | (void) set_attribute_and_warn(u, "cpu", "cpu.cfs_quota_us", buf); |
66ebf6c0 | 1001 | } else |
589a5f7a | 1002 | (void) set_attribute_and_warn(u, "cpu", "cpu.cfs_quota_us", "-1\n"); |
66ebf6c0 TH |
1003 | } |
1004 | ||
1005 | static uint64_t cgroup_cpu_shares_to_weight(uint64_t shares) { | |
1006 | return CLAMP(shares * CGROUP_WEIGHT_DEFAULT / CGROUP_CPU_SHARES_DEFAULT, | |
1007 | CGROUP_WEIGHT_MIN, CGROUP_WEIGHT_MAX); | |
1008 | } | |
1009 | ||
1010 | static uint64_t cgroup_cpu_weight_to_shares(uint64_t weight) { | |
c8340822 | 1011 | /* we don't support idle in cgroupv1 */ |
1012 | if (weight == CGROUP_WEIGHT_IDLE) | |
1013 | return CGROUP_CPU_SHARES_MIN; | |
1014 | ||
66ebf6c0 TH |
1015 | return CLAMP(weight * CGROUP_CPU_SHARES_DEFAULT / CGROUP_WEIGHT_DEFAULT, |
1016 | CGROUP_CPU_SHARES_MIN, CGROUP_CPU_SHARES_MAX); | |
1017 | } | |
1018 | ||
2cea199e | 1019 | static void cgroup_apply_unified_cpuset(Unit *u, const CPUSet *cpus, const char *name) { |
047f5d63 PH |
1020 | _cleanup_free_ char *buf = NULL; |
1021 | ||
2cea199e | 1022 | buf = cpu_set_to_range_string(cpus); |
c259ac9a LP |
1023 | if (!buf) { |
1024 | log_oom(); | |
1025 | return; | |
1026 | } | |
047f5d63 PH |
1027 | |
1028 | (void) set_attribute_and_warn(u, "cpuset", name, buf); | |
1029 | } | |
1030 | ||
508c45da | 1031 | static bool cgroup_context_has_io_config(CGroupContext *c) { |
538b4852 TH |
1032 | return c->io_accounting || |
1033 | c->io_weight != CGROUP_WEIGHT_INVALID || | |
1034 | c->startup_io_weight != CGROUP_WEIGHT_INVALID || | |
1035 | c->io_device_weights || | |
6ae4283c | 1036 | c->io_device_latencies || |
538b4852 TH |
1037 | c->io_device_limits; |
1038 | } | |
1039 | ||
508c45da | 1040 | static bool cgroup_context_has_blockio_config(CGroupContext *c) { |
538b4852 TH |
1041 | return c->blockio_accounting || |
1042 | c->blockio_weight != CGROUP_BLKIO_WEIGHT_INVALID || | |
1043 | c->startup_blockio_weight != CGROUP_BLKIO_WEIGHT_INVALID || | |
1044 | c->blockio_device_weights || | |
1045 | c->blockio_device_bandwidths; | |
1046 | } | |
1047 | ||
508c45da | 1048 | static uint64_t cgroup_context_io_weight(CGroupContext *c, ManagerState state) { |
9dfb6a3a | 1049 | if (IN_SET(state, MANAGER_STARTING, MANAGER_INITIALIZING, MANAGER_STOPPING) && |
64faf04c TH |
1050 | c->startup_io_weight != CGROUP_WEIGHT_INVALID) |
1051 | return c->startup_io_weight; | |
d38655d7 | 1052 | if (c->io_weight != CGROUP_WEIGHT_INVALID) |
64faf04c | 1053 | return c->io_weight; |
d38655d7 | 1054 | return CGROUP_WEIGHT_DEFAULT; |
64faf04c TH |
1055 | } |
1056 | ||
508c45da | 1057 | static uint64_t cgroup_context_blkio_weight(CGroupContext *c, ManagerState state) { |
9dfb6a3a | 1058 | if (IN_SET(state, MANAGER_STARTING, MANAGER_INITIALIZING, MANAGER_STOPPING) && |
64faf04c TH |
1059 | c->startup_blockio_weight != CGROUP_BLKIO_WEIGHT_INVALID) |
1060 | return c->startup_blockio_weight; | |
d38655d7 | 1061 | if (c->blockio_weight != CGROUP_BLKIO_WEIGHT_INVALID) |
64faf04c | 1062 | return c->blockio_weight; |
d38655d7 | 1063 | return CGROUP_BLKIO_WEIGHT_DEFAULT; |
64faf04c TH |
1064 | } |
1065 | ||
508c45da | 1066 | static uint64_t cgroup_weight_blkio_to_io(uint64_t blkio_weight) { |
538b4852 TH |
1067 | return CLAMP(blkio_weight * CGROUP_WEIGHT_DEFAULT / CGROUP_BLKIO_WEIGHT_DEFAULT, |
1068 | CGROUP_WEIGHT_MIN, CGROUP_WEIGHT_MAX); | |
1069 | } | |
1070 | ||
508c45da | 1071 | static uint64_t cgroup_weight_io_to_blkio(uint64_t io_weight) { |
538b4852 TH |
1072 | return CLAMP(io_weight * CGROUP_BLKIO_WEIGHT_DEFAULT / CGROUP_WEIGHT_DEFAULT, |
1073 | CGROUP_BLKIO_WEIGHT_MIN, CGROUP_BLKIO_WEIGHT_MAX); | |
1074 | } | |
1075 | ||
3e6eafdd | 1076 | static int set_bfq_weight(Unit *u, const char *controller, dev_t dev, uint64_t io_weight) { |
1cf4a685 MK |
1077 | static const char * const prop_names[] = { |
1078 | "IOWeight", | |
1079 | "BlockIOWeight", | |
1080 | "IODeviceWeight", | |
1081 | "BlockIODeviceWeight", | |
1082 | }; | |
8d75f60e | 1083 | static bool warned = false; |
9f0c0c4e | 1084 | char buf[DECIMAL_STR_MAX(dev_t)*2+2+DECIMAL_STR_MAX(uint64_t)+STRLEN("\n")]; |
bec17e80 MK |
1085 | const char *p; |
1086 | uint64_t bfq_weight; | |
8d75f60e | 1087 | int r; |
bec17e80 MK |
1088 | |
1089 | /* FIXME: drop this function when distro kernels properly support BFQ through "io.weight" | |
1090 | * See also: https://github.com/systemd/systemd/pull/13335 and | |
1091 | * https://github.com/torvalds/linux/commit/65752aef0a407e1ef17ec78a7fc31ba4e0b360f9. */ | |
1092 | p = strjoina(controller, ".bfq.weight"); | |
1093 | /* Adjust to kernel range is 1..1000, the default is 100. */ | |
1094 | bfq_weight = BFQ_WEIGHT(io_weight); | |
1095 | ||
9f0c0c4e | 1096 | if (major(dev) > 0) |
ec61371f | 1097 | xsprintf(buf, DEVNUM_FORMAT_STR " %" PRIu64 "\n", DEVNUM_FORMAT_VAL(dev), bfq_weight); |
9f0c0c4e MK |
1098 | else |
1099 | xsprintf(buf, "%" PRIu64 "\n", bfq_weight); | |
bec17e80 | 1100 | |
8d75f60e MK |
1101 | r = cg_set_attribute(controller, u->cgroup_path, p, buf); |
1102 | ||
1103 | /* FIXME: drop this when kernels prior | |
1104 | * 795fe54c2a82 ("bfq: Add per-device weight") v5.4 | |
1105 | * are not interesting anymore. Old kernels will fail with EINVAL, while new kernels won't return | |
1106 | * EINVAL on properly formatted input by us. Treat EINVAL accordingly. */ | |
3e6eafdd MK |
1107 | if (r == -EINVAL && major(dev) > 0) { |
1108 | if (!warned) { | |
1109 | log_unit_warning(u, "Kernel version does not accept per-device setting in %s.", p); | |
1110 | warned = true; | |
1111 | } | |
1112 | r = -EOPNOTSUPP; /* mask as unconfigured device */ | |
1113 | } else if (r >= 0 && io_weight != bfq_weight) | |
1cf4a685 MK |
1114 | log_unit_debug(u, "%s=%" PRIu64 " scaled to %s=%" PRIu64, |
1115 | prop_names[2*(major(dev) > 0) + streq(controller, "blkio")], | |
bec17e80 | 1116 | io_weight, p, bfq_weight); |
3e6eafdd | 1117 | return r; |
bec17e80 MK |
1118 | } |
1119 | ||
f29ff115 | 1120 | static void cgroup_apply_io_device_weight(Unit *u, const char *dev_path, uint64_t io_weight) { |
64faf04c TH |
1121 | char buf[DECIMAL_STR_MAX(dev_t)*2+2+DECIMAL_STR_MAX(uint64_t)+1]; |
1122 | dev_t dev; | |
3e6eafdd | 1123 | int r, r1, r2; |
64faf04c | 1124 | |
3e6eafdd | 1125 | if (lookup_block_device(dev_path, &dev) < 0) |
64faf04c TH |
1126 | return; |
1127 | ||
3e6eafdd | 1128 | r1 = set_bfq_weight(u, "io", dev, io_weight); |
9f0c0c4e | 1129 | |
ec61371f | 1130 | xsprintf(buf, DEVNUM_FORMAT_STR " %" PRIu64 "\n", DEVNUM_FORMAT_VAL(dev), io_weight); |
3e6eafdd MK |
1131 | r2 = cg_set_attribute("io", u->cgroup_path, "io.weight", buf); |
1132 | ||
1133 | /* Look at the configured device, when both fail, prefer io.weight errno. */ | |
1134 | r = r2 == -EOPNOTSUPP ? r1 : r2; | |
1135 | ||
1136 | if (r < 0) | |
1137 | log_unit_full_errno(u, LOG_LEVEL_CGROUP_WRITE(r), | |
1138 | r, "Failed to set 'io[.bfq].weight' attribute on '%s' to '%.*s': %m", | |
1139 | empty_to_root(u->cgroup_path), (int) strcspn(buf, NEWLINE), buf); | |
64faf04c TH |
1140 | } |
1141 | ||
f29ff115 | 1142 | static void cgroup_apply_blkio_device_weight(Unit *u, const char *dev_path, uint64_t blkio_weight) { |
64faf04c TH |
1143 | char buf[DECIMAL_STR_MAX(dev_t)*2+2+DECIMAL_STR_MAX(uint64_t)+1]; |
1144 | dev_t dev; | |
1145 | int r; | |
1146 | ||
1147 | r = lookup_block_device(dev_path, &dev); | |
1148 | if (r < 0) | |
1149 | return; | |
1150 | ||
ec61371f | 1151 | xsprintf(buf, DEVNUM_FORMAT_STR " %" PRIu64 "\n", DEVNUM_FORMAT_VAL(dev), blkio_weight); |
293d32df | 1152 | (void) set_attribute_and_warn(u, "blkio", "blkio.weight_device", buf); |
64faf04c TH |
1153 | } |
1154 | ||
6ae4283c TH |
1155 | static void cgroup_apply_io_device_latency(Unit *u, const char *dev_path, usec_t target) { |
1156 | char buf[DECIMAL_STR_MAX(dev_t)*2+2+7+DECIMAL_STR_MAX(uint64_t)+1]; | |
1157 | dev_t dev; | |
1158 | int r; | |
1159 | ||
1160 | r = lookup_block_device(dev_path, &dev); | |
1161 | if (r < 0) | |
1162 | return; | |
1163 | ||
1164 | if (target != USEC_INFINITY) | |
ec61371f | 1165 | xsprintf(buf, DEVNUM_FORMAT_STR " target=%" PRIu64 "\n", DEVNUM_FORMAT_VAL(dev), target); |
6ae4283c | 1166 | else |
ec61371f | 1167 | xsprintf(buf, DEVNUM_FORMAT_STR " target=max\n", DEVNUM_FORMAT_VAL(dev)); |
6ae4283c | 1168 | |
293d32df | 1169 | (void) set_attribute_and_warn(u, "io", "io.latency", buf); |
6ae4283c TH |
1170 | } |
1171 | ||
17ae2780 | 1172 | static void cgroup_apply_io_device_limit(Unit *u, const char *dev_path, uint64_t *limits) { |
4c1f9343 ZJS |
1173 | char limit_bufs[_CGROUP_IO_LIMIT_TYPE_MAX][DECIMAL_STR_MAX(uint64_t)], |
1174 | buf[DECIMAL_STR_MAX(dev_t)*2+2+(6+DECIMAL_STR_MAX(uint64_t)+1)*4]; | |
64faf04c | 1175 | dev_t dev; |
64faf04c | 1176 | |
4c1f9343 | 1177 | if (lookup_block_device(dev_path, &dev) < 0) |
17ae2780 | 1178 | return; |
64faf04c | 1179 | |
4c1f9343 | 1180 | for (CGroupIOLimitType type = 0; type < _CGROUP_IO_LIMIT_TYPE_MAX; type++) |
17ae2780 | 1181 | if (limits[type] != cgroup_io_limit_defaults[type]) |
64faf04c | 1182 | xsprintf(limit_bufs[type], "%" PRIu64, limits[type]); |
17ae2780 | 1183 | else |
64faf04c | 1184 | xsprintf(limit_bufs[type], "%s", limits[type] == CGROUP_LIMIT_MAX ? "max" : "0"); |
64faf04c | 1185 | |
ec61371f | 1186 | xsprintf(buf, DEVNUM_FORMAT_STR " rbps=%s wbps=%s riops=%s wiops=%s\n", DEVNUM_FORMAT_VAL(dev), |
64faf04c TH |
1187 | limit_bufs[CGROUP_IO_RBPS_MAX], limit_bufs[CGROUP_IO_WBPS_MAX], |
1188 | limit_bufs[CGROUP_IO_RIOPS_MAX], limit_bufs[CGROUP_IO_WIOPS_MAX]); | |
293d32df | 1189 | (void) set_attribute_and_warn(u, "io", "io.max", buf); |
64faf04c TH |
1190 | } |
1191 | ||
17ae2780 | 1192 | static void cgroup_apply_blkio_device_limit(Unit *u, const char *dev_path, uint64_t rbps, uint64_t wbps) { |
64faf04c TH |
1193 | char buf[DECIMAL_STR_MAX(dev_t)*2+2+DECIMAL_STR_MAX(uint64_t)+1]; |
1194 | dev_t dev; | |
64faf04c | 1195 | |
4c1f9343 | 1196 | if (lookup_block_device(dev_path, &dev) < 0) |
17ae2780 | 1197 | return; |
64faf04c | 1198 | |
ec61371f | 1199 | sprintf(buf, DEVNUM_FORMAT_STR " %" PRIu64 "\n", DEVNUM_FORMAT_VAL(dev), rbps); |
293d32df | 1200 | (void) set_attribute_and_warn(u, "blkio", "blkio.throttle.read_bps_device", buf); |
64faf04c | 1201 | |
ec61371f | 1202 | sprintf(buf, DEVNUM_FORMAT_STR " %" PRIu64 "\n", DEVNUM_FORMAT_VAL(dev), wbps); |
293d32df | 1203 | (void) set_attribute_and_warn(u, "blkio", "blkio.throttle.write_bps_device", buf); |
64faf04c TH |
1204 | } |
1205 | ||
c52db42b CD |
1206 | static bool unit_has_unified_memory_config(Unit *u) { |
1207 | CGroupContext *c; | |
1208 | ||
1209 | assert(u); | |
1210 | ||
806a9362 | 1211 | assert_se(c = unit_get_cgroup_context(u)); |
c52db42b | 1212 | |
7c9d2b79 | 1213 | return unit_get_ancestor_memory_min(u) > 0 || unit_get_ancestor_memory_low(u) > 0 || |
c52db42b CD |
1214 | c->memory_high != CGROUP_LIMIT_MAX || c->memory_max != CGROUP_LIMIT_MAX || |
1215 | c->memory_swap_max != CGROUP_LIMIT_MAX; | |
da4d897e TH |
1216 | } |
1217 | ||
f29ff115 | 1218 | static void cgroup_apply_unified_memory_limit(Unit *u, const char *file, uint64_t v) { |
589a5f7a | 1219 | char buf[DECIMAL_STR_MAX(uint64_t) + 1] = "max\n"; |
da4d897e TH |
1220 | |
1221 | if (v != CGROUP_LIMIT_MAX) | |
1222 | xsprintf(buf, "%" PRIu64 "\n", v); | |
1223 | ||
293d32df | 1224 | (void) set_attribute_and_warn(u, "memory", file, buf); |
da4d897e TH |
1225 | } |
1226 | ||
0f2d84d2 | 1227 | static void cgroup_apply_firewall(Unit *u) { |
0f2d84d2 LP |
1228 | assert(u); |
1229 | ||
acf7f253 | 1230 | /* Best-effort: let's apply IP firewalling and/or accounting if that's enabled */ |
906c06f6 | 1231 | |
acf7f253 | 1232 | if (bpf_firewall_compile(u) < 0) |
906c06f6 DM |
1233 | return; |
1234 | ||
fab34748 | 1235 | (void) bpf_firewall_load_custom(u); |
906c06f6 | 1236 | (void) bpf_firewall_install(u); |
906c06f6 DM |
1237 | } |
1238 | ||
a8e5eb17 JK |
1239 | static void cgroup_apply_socket_bind(Unit *u) { |
1240 | assert(u); | |
1241 | ||
cd09a5f3 | 1242 | (void) bpf_socket_bind_install(u); |
a8e5eb17 JK |
1243 | } |
1244 | ||
6f50d4f7 MV |
1245 | static void cgroup_apply_restrict_network_interfaces(Unit *u) { |
1246 | assert(u); | |
1247 | ||
1248 | (void) restrict_network_interfaces_install(u); | |
1249 | } | |
1250 | ||
8b139557 | 1251 | static int cgroup_apply_devices(Unit *u) { |
76dc1725 | 1252 | _cleanup_(bpf_program_freep) BPFProgram *prog = NULL; |
8b139557 ZJS |
1253 | const char *path; |
1254 | CGroupContext *c; | |
45669ae2 | 1255 | CGroupDevicePolicy policy; |
8b139557 ZJS |
1256 | int r; |
1257 | ||
1258 | assert_se(c = unit_get_cgroup_context(u)); | |
1259 | assert_se(path = u->cgroup_path); | |
1260 | ||
45669ae2 ZJS |
1261 | policy = c->device_policy; |
1262 | ||
8b139557 | 1263 | if (cg_all_unified() > 0) { |
45669ae2 | 1264 | r = bpf_devices_cgroup_init(&prog, policy, c->device_allow); |
8b139557 ZJS |
1265 | if (r < 0) |
1266 | return log_unit_warning_errno(u, r, "Failed to initialize device control bpf program: %m"); | |
1267 | ||
1268 | } else { | |
1269 | /* Changing the devices list of a populated cgroup might result in EINVAL, hence ignore | |
1270 | * EINVAL here. */ | |
1271 | ||
45669ae2 | 1272 | if (c->device_allow || policy != CGROUP_DEVICE_POLICY_AUTO) |
8b139557 ZJS |
1273 | r = cg_set_attribute("devices", path, "devices.deny", "a"); |
1274 | else | |
1275 | r = cg_set_attribute("devices", path, "devices.allow", "a"); | |
1276 | if (r < 0) | |
8ed6f81b YW |
1277 | log_unit_full_errno(u, IN_SET(r, -ENOENT, -EROFS, -EINVAL, -EACCES, -EPERM) ? LOG_DEBUG : LOG_WARNING, r, |
1278 | "Failed to reset devices.allow/devices.deny: %m"); | |
8b139557 ZJS |
1279 | } |
1280 | ||
6b000af4 | 1281 | bool allow_list_static = policy == CGROUP_DEVICE_POLICY_CLOSED || |
45669ae2 | 1282 | (policy == CGROUP_DEVICE_POLICY_AUTO && c->device_allow); |
6b000af4 LP |
1283 | if (allow_list_static) |
1284 | (void) bpf_devices_allow_list_static(prog, path); | |
8b139557 | 1285 | |
6b000af4 | 1286 | bool any = allow_list_static; |
8b139557 ZJS |
1287 | LIST_FOREACH(device_allow, a, c->device_allow) { |
1288 | char acc[4], *val; | |
1289 | unsigned k = 0; | |
1290 | ||
1291 | if (a->r) | |
1292 | acc[k++] = 'r'; | |
1293 | if (a->w) | |
1294 | acc[k++] = 'w'; | |
1295 | if (a->m) | |
1296 | acc[k++] = 'm'; | |
8b139557 ZJS |
1297 | if (k == 0) |
1298 | continue; | |
8b139557 ZJS |
1299 | acc[k++] = 0; |
1300 | ||
1301 | if (path_startswith(a->path, "/dev/")) | |
6b000af4 | 1302 | r = bpf_devices_allow_list_device(prog, path, a->path, acc); |
8b139557 | 1303 | else if ((val = startswith(a->path, "block-"))) |
6b000af4 | 1304 | r = bpf_devices_allow_list_major(prog, path, val, 'b', acc); |
8b139557 | 1305 | else if ((val = startswith(a->path, "char-"))) |
6b000af4 | 1306 | r = bpf_devices_allow_list_major(prog, path, val, 'c', acc); |
45669ae2 | 1307 | else { |
8b139557 | 1308 | log_unit_debug(u, "Ignoring device '%s' while writing cgroup attribute.", a->path); |
45669ae2 ZJS |
1309 | continue; |
1310 | } | |
1311 | ||
1312 | if (r >= 0) | |
1313 | any = true; | |
1314 | } | |
1315 | ||
1316 | if (prog && !any) { | |
1317 | log_unit_warning_errno(u, SYNTHETIC_ERRNO(ENODEV), "No devices matched by device filter."); | |
1318 | ||
1319 | /* The kernel verifier would reject a program we would build with the normal intro and outro | |
6b000af4 | 1320 | but no allow-listing rules (outro would contain an unreachable instruction for successful |
45669ae2 ZJS |
1321 | return). */ |
1322 | policy = CGROUP_DEVICE_POLICY_STRICT; | |
8b139557 ZJS |
1323 | } |
1324 | ||
76dc1725 | 1325 | r = bpf_devices_apply_policy(&prog, policy, any, path, &u->bpf_device_control_installed); |
8b139557 ZJS |
1326 | if (r < 0) { |
1327 | static bool warned = false; | |
1328 | ||
1329 | log_full_errno(warned ? LOG_DEBUG : LOG_WARNING, r, | |
1330 | "Unit %s configures device ACL, but the local system doesn't seem to support the BPF-based device controller.\n" | |
1331 | "Proceeding WITHOUT applying ACL (all devices will be accessible)!\n" | |
1332 | "(This warning is only shown for the first loaded unit using device ACL.)", u->id); | |
1333 | ||
1334 | warned = true; | |
1335 | } | |
1336 | return r; | |
1337 | } | |
1338 | ||
17283ce7 YW |
1339 | static void set_io_weight(Unit *u, uint64_t weight) { |
1340 | char buf[STRLEN("default \n")+DECIMAL_STR_MAX(uint64_t)]; | |
17283ce7 YW |
1341 | |
1342 | assert(u); | |
29eb0eef | 1343 | |
3e6eafdd | 1344 | (void) set_bfq_weight(u, "io", makedev(0, 0), weight); |
29eb0eef | 1345 | |
29eb0eef | 1346 | xsprintf(buf, "default %" PRIu64 "\n", weight); |
17283ce7 YW |
1347 | (void) set_attribute_and_warn(u, "io", "io.weight", buf); |
1348 | } | |
1349 | ||
1350 | static void set_blkio_weight(Unit *u, uint64_t weight) { | |
1351 | char buf[STRLEN("\n")+DECIMAL_STR_MAX(uint64_t)]; | |
17283ce7 YW |
1352 | |
1353 | assert(u); | |
29eb0eef | 1354 | |
3e6eafdd | 1355 | (void) set_bfq_weight(u, "blkio", makedev(0, 0), weight); |
17283ce7 YW |
1356 | |
1357 | xsprintf(buf, "%" PRIu64 "\n", weight); | |
1358 | (void) set_attribute_and_warn(u, "blkio", "blkio.weight", buf); | |
29eb0eef ZJS |
1359 | } |
1360 | ||
506ea51b JK |
1361 | static void cgroup_apply_bpf_foreign_program(Unit *u) { |
1362 | assert(u); | |
1363 | ||
1364 | (void) bpf_foreign_install(u); | |
1365 | } | |
1366 | ||
906c06f6 DM |
1367 | static void cgroup_context_apply( |
1368 | Unit *u, | |
1369 | CGroupMask apply_mask, | |
906c06f6 DM |
1370 | ManagerState state) { |
1371 | ||
f29ff115 TH |
1372 | const char *path; |
1373 | CGroupContext *c; | |
52fecf20 | 1374 | bool is_host_root, is_local_root; |
4ad49000 LP |
1375 | int r; |
1376 | ||
f29ff115 TH |
1377 | assert(u); |
1378 | ||
906c06f6 | 1379 | /* Nothing to do? Exit early! */ |
17f14955 | 1380 | if (apply_mask == 0) |
4ad49000 | 1381 | return; |
8e274523 | 1382 | |
52fecf20 LP |
1383 | /* Some cgroup attributes are not supported on the host root cgroup, hence silently ignore them here. And other |
1384 | * attributes should only be managed for cgroups further down the tree. */ | |
1385 | is_local_root = unit_has_name(u, SPECIAL_ROOT_SLICE); | |
1386 | is_host_root = unit_has_host_root_cgroup(u); | |
f3725e64 LP |
1387 | |
1388 | assert_se(c = unit_get_cgroup_context(u)); | |
1389 | assert_se(path = u->cgroup_path); | |
1390 | ||
52fecf20 | 1391 | if (is_local_root) /* Make sure we don't try to display messages with an empty path. */ |
6da13913 | 1392 | path = "/"; |
01efdf13 | 1393 | |
be2c0327 LP |
1394 | /* We generally ignore errors caused by read-only mounted cgroup trees (assuming we are running in a container |
1395 | * then), and missing cgroups, i.e. EROFS and ENOENT. */ | |
714e2e1d | 1396 | |
be2c0327 LP |
1397 | /* In fully unified mode these attributes don't exist on the host cgroup root. On legacy the weights exist, but |
1398 | * setting the weight makes very little sense on the host root cgroup, as there are no other cgroups at this | |
1399 | * level. The quota exists there too, but any attempt to write to it is refused with EINVAL. Inside of | |
4e1dfa45 | 1400 | * containers we want to leave control of these to the container manager (and if cgroup v2 delegation is used |
be2c0327 LP |
1401 | * we couldn't even write to them if we wanted to). */ |
1402 | if ((apply_mask & CGROUP_MASK_CPU) && !is_local_root) { | |
8e274523 | 1403 | |
b4cccbc1 | 1404 | if (cg_all_unified() > 0) { |
be2c0327 | 1405 | uint64_t weight; |
b2f8b02e | 1406 | |
be2c0327 LP |
1407 | if (cgroup_context_has_cpu_weight(c)) |
1408 | weight = cgroup_context_cpu_weight(c, state); | |
1409 | else if (cgroup_context_has_cpu_shares(c)) { | |
1410 | uint64_t shares; | |
66ebf6c0 | 1411 | |
be2c0327 LP |
1412 | shares = cgroup_context_cpu_shares(c, state); |
1413 | weight = cgroup_cpu_shares_to_weight(shares); | |
66ebf6c0 | 1414 | |
be2c0327 LP |
1415 | log_cgroup_compat(u, "Applying [Startup]CPUShares=%" PRIu64 " as [Startup]CPUWeight=%" PRIu64 " on %s", |
1416 | shares, weight, path); | |
1417 | } else | |
1418 | weight = CGROUP_WEIGHT_DEFAULT; | |
66ebf6c0 | 1419 | |
c8340822 | 1420 | cgroup_apply_unified_cpu_idle(u, weight); |
be2c0327 | 1421 | cgroup_apply_unified_cpu_weight(u, weight); |
10f28641 | 1422 | cgroup_apply_unified_cpu_quota(u, c->cpu_quota_per_sec_usec, c->cpu_quota_period_usec); |
66ebf6c0 | 1423 | |
52fecf20 | 1424 | } else { |
be2c0327 | 1425 | uint64_t shares; |
52fecf20 | 1426 | |
be2c0327 LP |
1427 | if (cgroup_context_has_cpu_weight(c)) { |
1428 | uint64_t weight; | |
52fecf20 | 1429 | |
be2c0327 LP |
1430 | weight = cgroup_context_cpu_weight(c, state); |
1431 | shares = cgroup_cpu_weight_to_shares(weight); | |
52fecf20 | 1432 | |
be2c0327 LP |
1433 | log_cgroup_compat(u, "Applying [Startup]CPUWeight=%" PRIu64 " as [Startup]CPUShares=%" PRIu64 " on %s", |
1434 | weight, shares, path); | |
1435 | } else if (cgroup_context_has_cpu_shares(c)) | |
1436 | shares = cgroup_context_cpu_shares(c, state); | |
1437 | else | |
1438 | shares = CGROUP_CPU_SHARES_DEFAULT; | |
66ebf6c0 | 1439 | |
be2c0327 | 1440 | cgroup_apply_legacy_cpu_shares(u, shares); |
10f28641 | 1441 | cgroup_apply_legacy_cpu_quota(u, c->cpu_quota_per_sec_usec, c->cpu_quota_period_usec); |
66ebf6c0 | 1442 | } |
4ad49000 LP |
1443 | } |
1444 | ||
047f5d63 | 1445 | if ((apply_mask & CGROUP_MASK_CPUSET) && !is_local_root) { |
31d3a520 PM |
1446 | cgroup_apply_unified_cpuset(u, cgroup_context_allowed_cpus(c, state), "cpuset.cpus"); |
1447 | cgroup_apply_unified_cpuset(u, cgroup_context_allowed_mems(c, state), "cpuset.mems"); | |
047f5d63 PH |
1448 | } |
1449 | ||
4e1dfa45 | 1450 | /* The 'io' controller attributes are not exported on the host's root cgroup (being a pure cgroup v2 |
52fecf20 LP |
1451 | * controller), and in case of containers we want to leave control of these attributes to the container manager |
1452 | * (and we couldn't access that stuff anyway, even if we tried if proper delegation is used). */ | |
1453 | if ((apply_mask & CGROUP_MASK_IO) && !is_local_root) { | |
52fecf20 LP |
1454 | bool has_io, has_blockio; |
1455 | uint64_t weight; | |
13c31542 | 1456 | |
52fecf20 LP |
1457 | has_io = cgroup_context_has_io_config(c); |
1458 | has_blockio = cgroup_context_has_blockio_config(c); | |
13c31542 | 1459 | |
52fecf20 LP |
1460 | if (has_io) |
1461 | weight = cgroup_context_io_weight(c, state); | |
1462 | else if (has_blockio) { | |
1463 | uint64_t blkio_weight; | |
128fadc9 | 1464 | |
52fecf20 LP |
1465 | blkio_weight = cgroup_context_blkio_weight(c, state); |
1466 | weight = cgroup_weight_blkio_to_io(blkio_weight); | |
128fadc9 | 1467 | |
67e2ea15 | 1468 | log_cgroup_compat(u, "Applying [Startup]BlockIOWeight=%" PRIu64 " as [Startup]IOWeight=%" PRIu64, |
52fecf20 LP |
1469 | blkio_weight, weight); |
1470 | } else | |
1471 | weight = CGROUP_WEIGHT_DEFAULT; | |
13c31542 | 1472 | |
17283ce7 | 1473 | set_io_weight(u, weight); |
2dbc45ae | 1474 | |
52fecf20 | 1475 | if (has_io) { |
52fecf20 LP |
1476 | LIST_FOREACH(device_weights, w, c->io_device_weights) |
1477 | cgroup_apply_io_device_weight(u, w->path, w->weight); | |
128fadc9 | 1478 | |
52fecf20 LP |
1479 | LIST_FOREACH(device_limits, limit, c->io_device_limits) |
1480 | cgroup_apply_io_device_limit(u, limit->path, limit->limits); | |
6ae4283c | 1481 | |
52fecf20 LP |
1482 | LIST_FOREACH(device_latencies, latency, c->io_device_latencies) |
1483 | cgroup_apply_io_device_latency(u, latency->path, latency->target_usec); | |
6ae4283c | 1484 | |
52fecf20 | 1485 | } else if (has_blockio) { |
52fecf20 LP |
1486 | LIST_FOREACH(device_weights, w, c->blockio_device_weights) { |
1487 | weight = cgroup_weight_blkio_to_io(w->weight); | |
17ae2780 | 1488 | |
67e2ea15 | 1489 | log_cgroup_compat(u, "Applying BlockIODeviceWeight=%" PRIu64 " as IODeviceWeight=%" PRIu64 " for %s", |
52fecf20 | 1490 | w->weight, weight, w->path); |
538b4852 | 1491 | |
52fecf20 LP |
1492 | cgroup_apply_io_device_weight(u, w->path, weight); |
1493 | } | |
538b4852 | 1494 | |
17ae2780 | 1495 | LIST_FOREACH(device_bandwidths, b, c->blockio_device_bandwidths) { |
538b4852 | 1496 | uint64_t limits[_CGROUP_IO_LIMIT_TYPE_MAX]; |
538b4852 | 1497 | |
e8616626 | 1498 | for (CGroupIOLimitType type = 0; type < _CGROUP_IO_LIMIT_TYPE_MAX; type++) |
538b4852 TH |
1499 | limits[type] = cgroup_io_limit_defaults[type]; |
1500 | ||
1501 | limits[CGROUP_IO_RBPS_MAX] = b->rbps; | |
1502 | limits[CGROUP_IO_WBPS_MAX] = b->wbps; | |
1503 | ||
67e2ea15 | 1504 | log_cgroup_compat(u, "Applying BlockIO{Read|Write}Bandwidth=%" PRIu64 " %" PRIu64 " as IO{Read|Write}BandwidthMax= for %s", |
128fadc9 TH |
1505 | b->rbps, b->wbps, b->path); |
1506 | ||
17ae2780 | 1507 | cgroup_apply_io_device_limit(u, b->path, limits); |
538b4852 | 1508 | } |
13c31542 TH |
1509 | } |
1510 | } | |
1511 | ||
906c06f6 | 1512 | if (apply_mask & CGROUP_MASK_BLKIO) { |
52fecf20 | 1513 | bool has_io, has_blockio; |
4ad49000 | 1514 | |
52fecf20 LP |
1515 | has_io = cgroup_context_has_io_config(c); |
1516 | has_blockio = cgroup_context_has_blockio_config(c); | |
1517 | ||
1518 | /* Applying a 'weight' never makes sense for the host root cgroup, and for containers this should be | |
1519 | * left to our container manager, too. */ | |
1520 | if (!is_local_root) { | |
64faf04c | 1521 | uint64_t weight; |
64faf04c | 1522 | |
7d862ab8 | 1523 | if (has_io) { |
52fecf20 | 1524 | uint64_t io_weight; |
128fadc9 | 1525 | |
52fecf20 | 1526 | io_weight = cgroup_context_io_weight(c, state); |
538b4852 | 1527 | weight = cgroup_weight_io_to_blkio(cgroup_context_io_weight(c, state)); |
128fadc9 | 1528 | |
67e2ea15 | 1529 | log_cgroup_compat(u, "Applying [Startup]IOWeight=%" PRIu64 " as [Startup]BlockIOWeight=%" PRIu64, |
128fadc9 | 1530 | io_weight, weight); |
7d862ab8 TH |
1531 | } else if (has_blockio) |
1532 | weight = cgroup_context_blkio_weight(c, state); | |
1533 | else | |
538b4852 | 1534 | weight = CGROUP_BLKIO_WEIGHT_DEFAULT; |
64faf04c | 1535 | |
17283ce7 | 1536 | set_blkio_weight(u, weight); |
35e7a62c | 1537 | |
03677889 | 1538 | if (has_io) |
128fadc9 TH |
1539 | LIST_FOREACH(device_weights, w, c->io_device_weights) { |
1540 | weight = cgroup_weight_io_to_blkio(w->weight); | |
1541 | ||
67e2ea15 | 1542 | log_cgroup_compat(u, "Applying IODeviceWeight=%" PRIu64 " as BlockIODeviceWeight=%" PRIu64 " for %s", |
128fadc9 TH |
1543 | w->weight, weight, w->path); |
1544 | ||
1545 | cgroup_apply_blkio_device_weight(u, w->path, weight); | |
1546 | } | |
03677889 | 1547 | else if (has_blockio) |
7d862ab8 TH |
1548 | LIST_FOREACH(device_weights, w, c->blockio_device_weights) |
1549 | cgroup_apply_blkio_device_weight(u, w->path, w->weight); | |
4ad49000 LP |
1550 | } |
1551 | ||
5238e957 | 1552 | /* The bandwidth limits are something that make sense to be applied to the host's root but not container |
52fecf20 LP |
1553 | * roots, as there we want the container manager to handle it */ |
1554 | if (is_host_root || !is_local_root) { | |
03677889 | 1555 | if (has_io) |
52fecf20 | 1556 | LIST_FOREACH(device_limits, l, c->io_device_limits) { |
67e2ea15 | 1557 | log_cgroup_compat(u, "Applying IO{Read|Write}Bandwidth=%" PRIu64 " %" PRIu64 " as BlockIO{Read|Write}BandwidthMax= for %s", |
52fecf20 | 1558 | l->limits[CGROUP_IO_RBPS_MAX], l->limits[CGROUP_IO_WBPS_MAX], l->path); |
128fadc9 | 1559 | |
52fecf20 LP |
1560 | cgroup_apply_blkio_device_limit(u, l->path, l->limits[CGROUP_IO_RBPS_MAX], l->limits[CGROUP_IO_WBPS_MAX]); |
1561 | } | |
03677889 | 1562 | else if (has_blockio) |
52fecf20 LP |
1563 | LIST_FOREACH(device_bandwidths, b, c->blockio_device_bandwidths) |
1564 | cgroup_apply_blkio_device_limit(u, b->path, b->rbps, b->wbps); | |
d686d8a9 | 1565 | } |
8e274523 LP |
1566 | } |
1567 | ||
be2c0327 LP |
1568 | /* In unified mode 'memory' attributes do not exist on the root cgroup. In legacy mode 'memory.limit_in_bytes' |
1569 | * exists on the root cgroup, but any writes to it are refused with EINVAL. And if we run in a container we | |
4e1dfa45 | 1570 | * want to leave control to the container manager (and if proper cgroup v2 delegation is used we couldn't even |
be2c0327 LP |
1571 | * write to this if we wanted to.) */ |
1572 | if ((apply_mask & CGROUP_MASK_MEMORY) && !is_local_root) { | |
efdb0237 | 1573 | |
52fecf20 | 1574 | if (cg_all_unified() > 0) { |
be2c0327 LP |
1575 | uint64_t max, swap_max = CGROUP_LIMIT_MAX; |
1576 | ||
c52db42b | 1577 | if (unit_has_unified_memory_config(u)) { |
be2c0327 LP |
1578 | max = c->memory_max; |
1579 | swap_max = c->memory_swap_max; | |
1580 | } else { | |
1581 | max = c->memory_limit; | |
efdb0237 | 1582 | |
be2c0327 LP |
1583 | if (max != CGROUP_LIMIT_MAX) |
1584 | log_cgroup_compat(u, "Applying MemoryLimit=%" PRIu64 " as MemoryMax=", max); | |
128fadc9 | 1585 | } |
da4d897e | 1586 | |
64fe532e | 1587 | cgroup_apply_unified_memory_limit(u, "memory.min", unit_get_ancestor_memory_min(u)); |
c52db42b | 1588 | cgroup_apply_unified_memory_limit(u, "memory.low", unit_get_ancestor_memory_low(u)); |
be2c0327 LP |
1589 | cgroup_apply_unified_memory_limit(u, "memory.high", c->memory_high); |
1590 | cgroup_apply_unified_memory_limit(u, "memory.max", max); | |
1591 | cgroup_apply_unified_memory_limit(u, "memory.swap.max", swap_max); | |
128fadc9 | 1592 | |
afcfaa69 LP |
1593 | (void) set_attribute_and_warn(u, "memory", "memory.oom.group", one_zero(c->memory_oom_group)); |
1594 | ||
be2c0327 LP |
1595 | } else { |
1596 | char buf[DECIMAL_STR_MAX(uint64_t) + 1]; | |
1597 | uint64_t val; | |
52fecf20 | 1598 | |
c52db42b | 1599 | if (unit_has_unified_memory_config(u)) { |
be2c0327 LP |
1600 | val = c->memory_max; |
1601 | log_cgroup_compat(u, "Applying MemoryMax=%" PRIi64 " as MemoryLimit=", val); | |
1602 | } else | |
1603 | val = c->memory_limit; | |
78a4ee59 | 1604 | |
be2c0327 LP |
1605 | if (val == CGROUP_LIMIT_MAX) |
1606 | strncpy(buf, "-1\n", sizeof(buf)); | |
1607 | else | |
1608 | xsprintf(buf, "%" PRIu64 "\n", val); | |
1609 | ||
1610 | (void) set_attribute_and_warn(u, "memory", "memory.limit_in_bytes", buf); | |
da4d897e | 1611 | } |
4ad49000 | 1612 | } |
8e274523 | 1613 | |
4e1dfa45 | 1614 | /* On cgroup v2 we can apply BPF everywhere. On cgroup v1 we apply it everywhere except for the root of |
52fecf20 LP |
1615 | * containers, where we leave this to the manager */ |
1616 | if ((apply_mask & (CGROUP_MASK_DEVICES | CGROUP_MASK_BPF_DEVICES)) && | |
8b139557 ZJS |
1617 | (is_host_root || cg_all_unified() > 0 || !is_local_root)) |
1618 | (void) cgroup_apply_devices(u); | |
03a7b521 | 1619 | |
00b5974f LP |
1620 | if (apply_mask & CGROUP_MASK_PIDS) { |
1621 | ||
52fecf20 | 1622 | if (is_host_root) { |
00b5974f LP |
1623 | /* So, the "pids" controller does not expose anything on the root cgroup, in order not to |
1624 | * replicate knobs exposed elsewhere needlessly. We abstract this away here however, and when | |
1625 | * the knobs of the root cgroup are modified propagate this to the relevant sysctls. There's a | |
1626 | * non-obvious asymmetry however: unlike the cgroup properties we don't really want to take | |
1627 | * exclusive ownership of the sysctls, but we still want to honour things if the user sets | |
1628 | * limits. Hence we employ sort of a one-way strategy: when the user sets a bounded limit | |
1629 | * through us it counts. When the user afterwards unsets it again (i.e. sets it to unbounded) | |
1630 | * it also counts. But if the user never set a limit through us (i.e. we are the default of | |
1631 | * "unbounded") we leave things unmodified. For this we manage a global boolean that we turn on | |
1632 | * the first time we set a limit. Note that this boolean is flushed out on manager reload, | |
5238e957 | 1633 | * which is desirable so that there's an official way to release control of the sysctl from |
00b5974f LP |
1634 | * systemd: set the limit to unbounded and reload. */ |
1635 | ||
3a0f06c4 | 1636 | if (tasks_max_isset(&c->tasks_max)) { |
00b5974f | 1637 | u->manager->sysctl_pid_max_changed = true; |
3a0f06c4 | 1638 | r = procfs_tasks_set_limit(tasks_max_resolve(&c->tasks_max)); |
00b5974f LP |
1639 | } else if (u->manager->sysctl_pid_max_changed) |
1640 | r = procfs_tasks_set_limit(TASKS_MAX); | |
1641 | else | |
1642 | r = 0; | |
00b5974f | 1643 | if (r < 0) |
8ed6f81b YW |
1644 | log_unit_full_errno(u, LOG_LEVEL_CGROUP_WRITE(r), r, |
1645 | "Failed to write to tasks limit sysctls: %m"); | |
52fecf20 | 1646 | } |
03a7b521 | 1647 | |
52fecf20 LP |
1648 | /* The attribute itself is not available on the host root cgroup, and in the container case we want to |
1649 | * leave it for the container manager. */ | |
1650 | if (!is_local_root) { | |
3a0f06c4 ZJS |
1651 | if (tasks_max_isset(&c->tasks_max)) { |
1652 | char buf[DECIMAL_STR_MAX(uint64_t) + 1]; | |
03a7b521 | 1653 | |
3a0f06c4 | 1654 | xsprintf(buf, "%" PRIu64 "\n", tasks_max_resolve(&c->tasks_max)); |
293d32df | 1655 | (void) set_attribute_and_warn(u, "pids", "pids.max", buf); |
00b5974f | 1656 | } else |
589a5f7a | 1657 | (void) set_attribute_and_warn(u, "pids", "pids.max", "max\n"); |
00b5974f | 1658 | } |
03a7b521 | 1659 | } |
906c06f6 | 1660 | |
17f14955 | 1661 | if (apply_mask & CGROUP_MASK_BPF_FIREWALL) |
0f2d84d2 | 1662 | cgroup_apply_firewall(u); |
506ea51b JK |
1663 | |
1664 | if (apply_mask & CGROUP_MASK_BPF_FOREIGN) | |
1665 | cgroup_apply_bpf_foreign_program(u); | |
a8e5eb17 JK |
1666 | |
1667 | if (apply_mask & CGROUP_MASK_BPF_SOCKET_BIND) | |
1668 | cgroup_apply_socket_bind(u); | |
6f50d4f7 MV |
1669 | |
1670 | if (apply_mask & CGROUP_MASK_BPF_RESTRICT_NETWORK_INTERFACES) | |
1671 | cgroup_apply_restrict_network_interfaces(u); | |
fb385181 LP |
1672 | } |
1673 | ||
16492445 LP |
1674 | static bool unit_get_needs_bpf_firewall(Unit *u) { |
1675 | CGroupContext *c; | |
16492445 LP |
1676 | assert(u); |
1677 | ||
1678 | c = unit_get_cgroup_context(u); | |
1679 | if (!c) | |
1680 | return false; | |
1681 | ||
1682 | if (c->ip_accounting || | |
84ebe6f0 YW |
1683 | !set_isempty(c->ip_address_allow) || |
1684 | !set_isempty(c->ip_address_deny) || | |
fab34748 KL |
1685 | c->ip_filters_ingress || |
1686 | c->ip_filters_egress) | |
16492445 LP |
1687 | return true; |
1688 | ||
1689 | /* If any parent slice has an IP access list defined, it applies too */ | |
e8616626 | 1690 | for (Unit *p = UNIT_GET_SLICE(u); p; p = UNIT_GET_SLICE(p)) { |
16492445 LP |
1691 | c = unit_get_cgroup_context(p); |
1692 | if (!c) | |
1693 | return false; | |
1694 | ||
84ebe6f0 YW |
1695 | if (!set_isempty(c->ip_address_allow) || |
1696 | !set_isempty(c->ip_address_deny)) | |
16492445 LP |
1697 | return true; |
1698 | } | |
1699 | ||
1700 | return false; | |
1701 | } | |
1702 | ||
506ea51b JK |
1703 | static bool unit_get_needs_bpf_foreign_program(Unit *u) { |
1704 | CGroupContext *c; | |
1705 | assert(u); | |
1706 | ||
1707 | c = unit_get_cgroup_context(u); | |
1708 | if (!c) | |
1709 | return false; | |
1710 | ||
64903d18 | 1711 | return !!c->bpf_foreign_programs; |
506ea51b JK |
1712 | } |
1713 | ||
a8e5eb17 JK |
1714 | static bool unit_get_needs_socket_bind(Unit *u) { |
1715 | CGroupContext *c; | |
1716 | assert(u); | |
1717 | ||
1718 | c = unit_get_cgroup_context(u); | |
1719 | if (!c) | |
1720 | return false; | |
1721 | ||
11ab01e4 | 1722 | return c->socket_bind_allow || c->socket_bind_deny; |
a8e5eb17 JK |
1723 | } |
1724 | ||
6f50d4f7 MV |
1725 | static bool unit_get_needs_restrict_network_interfaces(Unit *u) { |
1726 | CGroupContext *c; | |
1727 | assert(u); | |
1728 | ||
1729 | c = unit_get_cgroup_context(u); | |
1730 | if (!c) | |
1731 | return false; | |
1732 | ||
1733 | return !set_isempty(c->restrict_network_interfaces); | |
1734 | } | |
1735 | ||
c52db42b | 1736 | static CGroupMask unit_get_cgroup_mask(Unit *u) { |
efdb0237 | 1737 | CGroupMask mask = 0; |
c52db42b CD |
1738 | CGroupContext *c; |
1739 | ||
1740 | assert(u); | |
1741 | ||
806a9362 | 1742 | assert_se(c = unit_get_cgroup_context(u)); |
c710d3b4 | 1743 | |
fae9bc29 | 1744 | /* Figure out which controllers we need, based on the cgroup context object */ |
8e274523 | 1745 | |
fae9bc29 | 1746 | if (c->cpu_accounting) |
f98c2585 | 1747 | mask |= get_cpu_accounting_mask(); |
fae9bc29 LP |
1748 | |
1749 | if (cgroup_context_has_cpu_weight(c) || | |
66ebf6c0 | 1750 | cgroup_context_has_cpu_shares(c) || |
3a43da28 | 1751 | c->cpu_quota_per_sec_usec != USEC_INFINITY) |
fae9bc29 | 1752 | mask |= CGROUP_MASK_CPU; |
ecedd90f | 1753 | |
31d3a520 | 1754 | if (cgroup_context_has_allowed_cpus(c) || cgroup_context_has_allowed_mems(c)) |
047f5d63 PH |
1755 | mask |= CGROUP_MASK_CPUSET; |
1756 | ||
538b4852 TH |
1757 | if (cgroup_context_has_io_config(c) || cgroup_context_has_blockio_config(c)) |
1758 | mask |= CGROUP_MASK_IO | CGROUP_MASK_BLKIO; | |
ecedd90f | 1759 | |
4ad49000 | 1760 | if (c->memory_accounting || |
da4d897e | 1761 | c->memory_limit != CGROUP_LIMIT_MAX || |
c52db42b | 1762 | unit_has_unified_memory_config(u)) |
efdb0237 | 1763 | mask |= CGROUP_MASK_MEMORY; |
8e274523 | 1764 | |
a931ad47 | 1765 | if (c->device_allow || |
084870f9 | 1766 | c->device_policy != CGROUP_DEVICE_POLICY_AUTO) |
084c7007 | 1767 | mask |= CGROUP_MASK_DEVICES | CGROUP_MASK_BPF_DEVICES; |
4ad49000 | 1768 | |
03a7b521 | 1769 | if (c->tasks_accounting || |
3a0f06c4 | 1770 | tasks_max_isset(&c->tasks_max)) |
03a7b521 LP |
1771 | mask |= CGROUP_MASK_PIDS; |
1772 | ||
fae9bc29 | 1773 | return CGROUP_MASK_EXTEND_JOINED(mask); |
8e274523 LP |
1774 | } |
1775 | ||
53aea74a | 1776 | static CGroupMask unit_get_bpf_mask(Unit *u) { |
17f14955 RG |
1777 | CGroupMask mask = 0; |
1778 | ||
fae9bc29 LP |
1779 | /* Figure out which controllers we need, based on the cgroup context, possibly taking into account children |
1780 | * too. */ | |
1781 | ||
17f14955 RG |
1782 | if (unit_get_needs_bpf_firewall(u)) |
1783 | mask |= CGROUP_MASK_BPF_FIREWALL; | |
1784 | ||
506ea51b JK |
1785 | if (unit_get_needs_bpf_foreign_program(u)) |
1786 | mask |= CGROUP_MASK_BPF_FOREIGN; | |
1787 | ||
a8e5eb17 JK |
1788 | if (unit_get_needs_socket_bind(u)) |
1789 | mask |= CGROUP_MASK_BPF_SOCKET_BIND; | |
1790 | ||
6f50d4f7 MV |
1791 | if (unit_get_needs_restrict_network_interfaces(u)) |
1792 | mask |= CGROUP_MASK_BPF_RESTRICT_NETWORK_INTERFACES; | |
1793 | ||
17f14955 RG |
1794 | return mask; |
1795 | } | |
1796 | ||
efdb0237 | 1797 | CGroupMask unit_get_own_mask(Unit *u) { |
4ad49000 | 1798 | CGroupContext *c; |
8e274523 | 1799 | |
442ce775 LP |
1800 | /* Returns the mask of controllers the unit needs for itself. If a unit is not properly loaded, return an empty |
1801 | * mask, as we shouldn't reflect it in the cgroup hierarchy then. */ | |
1802 | ||
1803 | if (u->load_state != UNIT_LOADED) | |
1804 | return 0; | |
efdb0237 | 1805 | |
4ad49000 LP |
1806 | c = unit_get_cgroup_context(u); |
1807 | if (!c) | |
1808 | return 0; | |
8e274523 | 1809 | |
12b975e0 | 1810 | return unit_get_cgroup_mask(u) | unit_get_bpf_mask(u) | unit_get_delegate_mask(u); |
02638280 LP |
1811 | } |
1812 | ||
1813 | CGroupMask unit_get_delegate_mask(Unit *u) { | |
1814 | CGroupContext *c; | |
1815 | ||
1816 | /* If delegation is turned on, then turn on selected controllers, unless we are on the legacy hierarchy and the | |
1817 | * process we fork into is known to drop privileges, and hence shouldn't get access to the controllers. | |
19af675e | 1818 | * |
02638280 | 1819 | * Note that on the unified hierarchy it is safe to delegate controllers to unprivileged services. */ |
a931ad47 | 1820 | |
1d9cc876 | 1821 | if (!unit_cgroup_delegate(u)) |
02638280 LP |
1822 | return 0; |
1823 | ||
1824 | if (cg_all_unified() <= 0) { | |
a931ad47 LP |
1825 | ExecContext *e; |
1826 | ||
1827 | e = unit_get_exec_context(u); | |
02638280 LP |
1828 | if (e && !exec_context_maintains_privileges(e)) |
1829 | return 0; | |
a931ad47 LP |
1830 | } |
1831 | ||
1d9cc876 | 1832 | assert_se(c = unit_get_cgroup_context(u)); |
fae9bc29 | 1833 | return CGROUP_MASK_EXTEND_JOINED(c->delegate_controllers); |
8e274523 LP |
1834 | } |
1835 | ||
d9ef5944 MK |
1836 | static CGroupMask unit_get_subtree_mask(Unit *u) { |
1837 | ||
1838 | /* Returns the mask of this subtree, meaning of the group | |
1839 | * itself and its children. */ | |
1840 | ||
1841 | return unit_get_own_mask(u) | unit_get_members_mask(u); | |
1842 | } | |
1843 | ||
efdb0237 | 1844 | CGroupMask unit_get_members_mask(Unit *u) { |
4ad49000 | 1845 | assert(u); |
bc432dc7 | 1846 | |
02638280 | 1847 | /* Returns the mask of controllers all of the unit's children require, merged */ |
efdb0237 | 1848 | |
bc432dc7 | 1849 | if (u->cgroup_members_mask_valid) |
26a17ca2 | 1850 | return u->cgroup_members_mask; /* Use cached value if possible */ |
bc432dc7 | 1851 | |
64e844e5 | 1852 | u->cgroup_members_mask = 0; |
bc432dc7 LP |
1853 | |
1854 | if (u->type == UNIT_SLICE) { | |
1855 | Unit *member; | |
bc432dc7 | 1856 | |
d219a2b0 | 1857 | UNIT_FOREACH_DEPENDENCY(member, u, UNIT_ATOM_SLICE_OF) |
15ed3c3a | 1858 | u->cgroup_members_mask |= unit_get_subtree_mask(member); /* note that this calls ourselves again, for the children */ |
bc432dc7 LP |
1859 | } |
1860 | ||
1861 | u->cgroup_members_mask_valid = true; | |
6414b7c9 | 1862 | return u->cgroup_members_mask; |
246aa6dd LP |
1863 | } |
1864 | ||
efdb0237 | 1865 | CGroupMask unit_get_siblings_mask(Unit *u) { |
12f64221 | 1866 | Unit *slice; |
4ad49000 | 1867 | assert(u); |
246aa6dd | 1868 | |
efdb0237 LP |
1869 | /* Returns the mask of controllers all of the unit's siblings |
1870 | * require, i.e. the members mask of the unit's parent slice | |
1871 | * if there is one. */ | |
1872 | ||
12f64221 LP |
1873 | slice = UNIT_GET_SLICE(u); |
1874 | if (slice) | |
1875 | return unit_get_members_mask(slice); | |
4ad49000 | 1876 | |
64e844e5 | 1877 | return unit_get_subtree_mask(u); /* we are the top-level slice */ |
246aa6dd LP |
1878 | } |
1879 | ||
d9ef5944 | 1880 | static CGroupMask unit_get_disable_mask(Unit *u) { |
4f6f62e4 CD |
1881 | CGroupContext *c; |
1882 | ||
1883 | c = unit_get_cgroup_context(u); | |
1884 | if (!c) | |
1885 | return 0; | |
1886 | ||
1887 | return c->disable_controllers; | |
1888 | } | |
1889 | ||
1890 | CGroupMask unit_get_ancestor_disable_mask(Unit *u) { | |
1891 | CGroupMask mask; | |
12f64221 | 1892 | Unit *slice; |
4f6f62e4 CD |
1893 | |
1894 | assert(u); | |
1895 | mask = unit_get_disable_mask(u); | |
1896 | ||
1897 | /* Returns the mask of controllers which are marked as forcibly | |
1898 | * disabled in any ancestor unit or the unit in question. */ | |
1899 | ||
12f64221 LP |
1900 | slice = UNIT_GET_SLICE(u); |
1901 | if (slice) | |
1902 | mask |= unit_get_ancestor_disable_mask(slice); | |
4f6f62e4 CD |
1903 | |
1904 | return mask; | |
1905 | } | |
1906 | ||
efdb0237 | 1907 | CGroupMask unit_get_target_mask(Unit *u) { |
a437c5e4 | 1908 | CGroupMask own_mask, mask; |
efdb0237 | 1909 | |
a437c5e4 LP |
1910 | /* This returns the cgroup mask of all controllers to enable for a specific cgroup, i.e. everything |
1911 | * it needs itself, plus all that its children need, plus all that its siblings need. This is | |
1912 | * primarily useful on the legacy cgroup hierarchy, where we need to duplicate each cgroup in each | |
efdb0237 | 1913 | * hierarchy that shall be enabled for it. */ |
6414b7c9 | 1914 | |
a437c5e4 | 1915 | own_mask = unit_get_own_mask(u); |
84d2744b | 1916 | |
a437c5e4 | 1917 | if (own_mask & CGROUP_MASK_BPF_FIREWALL & ~u->manager->cgroup_supported) |
84d2744b ZJS |
1918 | emit_bpf_firewall_warning(u); |
1919 | ||
a437c5e4 LP |
1920 | mask = own_mask | unit_get_members_mask(u) | unit_get_siblings_mask(u); |
1921 | ||
efdb0237 | 1922 | mask &= u->manager->cgroup_supported; |
c72703e2 | 1923 | mask &= ~unit_get_ancestor_disable_mask(u); |
efdb0237 LP |
1924 | |
1925 | return mask; | |
1926 | } | |
1927 | ||
1928 | CGroupMask unit_get_enable_mask(Unit *u) { | |
1929 | CGroupMask mask; | |
1930 | ||
1931 | /* This returns the cgroup mask of all controllers to enable | |
1932 | * for the children of a specific cgroup. This is primarily | |
1933 | * useful for the unified cgroup hierarchy, where each cgroup | |
1934 | * controls which controllers are enabled for its children. */ | |
1935 | ||
1936 | mask = unit_get_members_mask(u); | |
6414b7c9 | 1937 | mask &= u->manager->cgroup_supported; |
c72703e2 | 1938 | mask &= ~unit_get_ancestor_disable_mask(u); |
6414b7c9 DS |
1939 | |
1940 | return mask; | |
1941 | } | |
1942 | ||
5af88058 | 1943 | void unit_invalidate_cgroup_members_masks(Unit *u) { |
12f64221 LP |
1944 | Unit *slice; |
1945 | ||
bc432dc7 LP |
1946 | assert(u); |
1947 | ||
5af88058 LP |
1948 | /* Recurse invalidate the member masks cache all the way up the tree */ |
1949 | u->cgroup_members_mask_valid = false; | |
bc432dc7 | 1950 | |
12f64221 LP |
1951 | slice = UNIT_GET_SLICE(u); |
1952 | if (slice) | |
1953 | unit_invalidate_cgroup_members_masks(slice); | |
6414b7c9 DS |
1954 | } |
1955 | ||
6592b975 | 1956 | const char *unit_get_realized_cgroup_path(Unit *u, CGroupMask mask) { |
03b90d4b | 1957 | |
6592b975 | 1958 | /* Returns the realized cgroup path of the specified unit where all specified controllers are available. */ |
03b90d4b LP |
1959 | |
1960 | while (u) { | |
6592b975 | 1961 | |
03b90d4b LP |
1962 | if (u->cgroup_path && |
1963 | u->cgroup_realized && | |
d94a24ca | 1964 | FLAGS_SET(u->cgroup_realized_mask, mask)) |
03b90d4b LP |
1965 | return u->cgroup_path; |
1966 | ||
12f64221 | 1967 | u = UNIT_GET_SLICE(u); |
03b90d4b LP |
1968 | } |
1969 | ||
1970 | return NULL; | |
1971 | } | |
1972 | ||
6592b975 | 1973 | static const char *migrate_callback(CGroupMask mask, void *userdata) { |
7b639614 MK |
1974 | /* If not realized at all, migrate to root (""). |
1975 | * It may happen if we're upgrading from older version that didn't clean up. | |
1976 | */ | |
1977 | return strempty(unit_get_realized_cgroup_path(userdata, mask)); | |
6592b975 LP |
1978 | } |
1979 | ||
303ee601 | 1980 | char *unit_default_cgroup_path(const Unit *u) { |
12f64221 LP |
1981 | _cleanup_free_ char *escaped = NULL, *slice_path = NULL; |
1982 | Unit *slice; | |
efdb0237 LP |
1983 | int r; |
1984 | ||
1985 | assert(u); | |
1986 | ||
1987 | if (unit_has_name(u, SPECIAL_ROOT_SLICE)) | |
1988 | return strdup(u->manager->cgroup_root); | |
1989 | ||
12f64221 LP |
1990 | slice = UNIT_GET_SLICE(u); |
1991 | if (slice && !unit_has_name(slice, SPECIAL_ROOT_SLICE)) { | |
1992 | r = cg_slice_to_path(slice->id, &slice_path); | |
efdb0237 LP |
1993 | if (r < 0) |
1994 | return NULL; | |
1995 | } | |
1996 | ||
1997 | escaped = cg_escape(u->id); | |
1998 | if (!escaped) | |
1999 | return NULL; | |
2000 | ||
12f64221 | 2001 | return path_join(empty_to_root(u->manager->cgroup_root), slice_path, escaped); |
efdb0237 LP |
2002 | } |
2003 | ||
2004 | int unit_set_cgroup_path(Unit *u, const char *path) { | |
2005 | _cleanup_free_ char *p = NULL; | |
2006 | int r; | |
2007 | ||
2008 | assert(u); | |
2009 | ||
5210387e LP |
2010 | if (streq_ptr(u->cgroup_path, path)) |
2011 | return 0; | |
2012 | ||
efdb0237 LP |
2013 | if (path) { |
2014 | p = strdup(path); | |
2015 | if (!p) | |
2016 | return -ENOMEM; | |
5210387e | 2017 | } |
efdb0237 LP |
2018 | |
2019 | if (p) { | |
2020 | r = hashmap_put(u->manager->cgroup_unit, p, u); | |
2021 | if (r < 0) | |
2022 | return r; | |
2023 | } | |
2024 | ||
2025 | unit_release_cgroup(u); | |
ae2a15bc | 2026 | u->cgroup_path = TAKE_PTR(p); |
efdb0237 LP |
2027 | |
2028 | return 1; | |
2029 | } | |
2030 | ||
2031 | int unit_watch_cgroup(Unit *u) { | |
ab2c3861 | 2032 | _cleanup_free_ char *events = NULL; |
efdb0237 LP |
2033 | int r; |
2034 | ||
2035 | assert(u); | |
2036 | ||
0bb814c2 LP |
2037 | /* Watches the "cgroups.events" attribute of this unit's cgroup for "empty" events, but only if |
2038 | * cgroupv2 is available. */ | |
2039 | ||
efdb0237 LP |
2040 | if (!u->cgroup_path) |
2041 | return 0; | |
2042 | ||
0bb814c2 | 2043 | if (u->cgroup_control_inotify_wd >= 0) |
efdb0237 LP |
2044 | return 0; |
2045 | ||
2046 | /* Only applies to the unified hierarchy */ | |
c22800e4 | 2047 | r = cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER); |
b4cccbc1 LP |
2048 | if (r < 0) |
2049 | return log_error_errno(r, "Failed to determine whether the name=systemd hierarchy is unified: %m"); | |
2050 | if (r == 0) | |
efdb0237 LP |
2051 | return 0; |
2052 | ||
0bb814c2 | 2053 | /* No point in watch the top-level slice, it's never going to run empty. */ |
efdb0237 LP |
2054 | if (unit_has_name(u, SPECIAL_ROOT_SLICE)) |
2055 | return 0; | |
2056 | ||
0bb814c2 | 2057 | r = hashmap_ensure_allocated(&u->manager->cgroup_control_inotify_wd_unit, &trivial_hash_ops); |
efdb0237 LP |
2058 | if (r < 0) |
2059 | return log_oom(); | |
2060 | ||
ab2c3861 | 2061 | r = cg_get_path(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path, "cgroup.events", &events); |
efdb0237 LP |
2062 | if (r < 0) |
2063 | return log_oom(); | |
2064 | ||
0bb814c2 LP |
2065 | u->cgroup_control_inotify_wd = inotify_add_watch(u->manager->cgroup_inotify_fd, events, IN_MODIFY); |
2066 | if (u->cgroup_control_inotify_wd < 0) { | |
efdb0237 | 2067 | |
0bb814c2 LP |
2068 | if (errno == ENOENT) /* If the directory is already gone we don't need to track it, so this |
2069 | * is not an error */ | |
efdb0237 LP |
2070 | return 0; |
2071 | ||
6178e2f8 | 2072 | 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 |
2073 | } |
2074 | ||
0bb814c2 | 2075 | r = hashmap_put(u->manager->cgroup_control_inotify_wd_unit, INT_TO_PTR(u->cgroup_control_inotify_wd), u); |
efdb0237 | 2076 | if (r < 0) |
6178e2f8 | 2077 | 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 |
2078 | |
2079 | return 0; | |
2080 | } | |
2081 | ||
afcfaa69 LP |
2082 | int unit_watch_cgroup_memory(Unit *u) { |
2083 | _cleanup_free_ char *events = NULL; | |
2084 | CGroupContext *c; | |
2085 | int r; | |
2086 | ||
2087 | assert(u); | |
2088 | ||
2089 | /* Watches the "memory.events" attribute of this unit's cgroup for "oom_kill" events, but only if | |
2090 | * cgroupv2 is available. */ | |
2091 | ||
2092 | if (!u->cgroup_path) | |
2093 | return 0; | |
2094 | ||
2095 | c = unit_get_cgroup_context(u); | |
2096 | if (!c) | |
2097 | return 0; | |
2098 | ||
2099 | /* The "memory.events" attribute is only available if the memory controller is on. Let's hence tie | |
2100 | * this to memory accounting, in a way watching for OOM kills is a form of memory accounting after | |
2101 | * all. */ | |
2102 | if (!c->memory_accounting) | |
2103 | return 0; | |
2104 | ||
2105 | /* Don't watch inner nodes, as the kernel doesn't report oom_kill events recursively currently, and | |
2106 | * we also don't want to generate a log message for each parent cgroup of a process. */ | |
2107 | if (u->type == UNIT_SLICE) | |
2108 | return 0; | |
2109 | ||
2110 | if (u->cgroup_memory_inotify_wd >= 0) | |
2111 | return 0; | |
2112 | ||
2113 | /* Only applies to the unified hierarchy */ | |
2114 | r = cg_all_unified(); | |
2115 | if (r < 0) | |
2116 | return log_error_errno(r, "Failed to determine whether the memory controller is unified: %m"); | |
2117 | if (r == 0) | |
2118 | return 0; | |
2119 | ||
2120 | r = hashmap_ensure_allocated(&u->manager->cgroup_memory_inotify_wd_unit, &trivial_hash_ops); | |
2121 | if (r < 0) | |
2122 | return log_oom(); | |
2123 | ||
2124 | r = cg_get_path(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path, "memory.events", &events); | |
2125 | if (r < 0) | |
2126 | return log_oom(); | |
2127 | ||
2128 | u->cgroup_memory_inotify_wd = inotify_add_watch(u->manager->cgroup_inotify_fd, events, IN_MODIFY); | |
2129 | if (u->cgroup_memory_inotify_wd < 0) { | |
2130 | ||
2131 | if (errno == ENOENT) /* If the directory is already gone we don't need to track it, so this | |
2132 | * is not an error */ | |
2133 | return 0; | |
2134 | ||
6178e2f8 | 2135 | 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 |
2136 | } |
2137 | ||
2138 | r = hashmap_put(u->manager->cgroup_memory_inotify_wd_unit, INT_TO_PTR(u->cgroup_memory_inotify_wd), u); | |
2139 | if (r < 0) | |
6178e2f8 | 2140 | 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 |
2141 | |
2142 | return 0; | |
2143 | } | |
2144 | ||
a4634b21 LP |
2145 | int unit_pick_cgroup_path(Unit *u) { |
2146 | _cleanup_free_ char *path = NULL; | |
2147 | int r; | |
2148 | ||
2149 | assert(u); | |
2150 | ||
2151 | if (u->cgroup_path) | |
2152 | return 0; | |
2153 | ||
2154 | if (!UNIT_HAS_CGROUP_CONTEXT(u)) | |
2155 | return -EINVAL; | |
2156 | ||
2157 | path = unit_default_cgroup_path(u); | |
2158 | if (!path) | |
2159 | return log_oom(); | |
2160 | ||
2161 | r = unit_set_cgroup_path(u, path); | |
2162 | if (r == -EEXIST) | |
6178e2f8 | 2163 | return log_unit_error_errno(u, r, "Control group %s exists already.", empty_to_root(path)); |
a4634b21 | 2164 | if (r < 0) |
6178e2f8 | 2165 | return log_unit_error_errno(u, r, "Failed to set unit's control group path to %s: %m", empty_to_root(path)); |
a4634b21 LP |
2166 | |
2167 | return 0; | |
2168 | } | |
2169 | ||
7b639614 | 2170 | static int unit_update_cgroup( |
efdb0237 LP |
2171 | Unit *u, |
2172 | CGroupMask target_mask, | |
0d2d6fbf CD |
2173 | CGroupMask enable_mask, |
2174 | ManagerState state) { | |
efdb0237 | 2175 | |
7b639614 MK |
2176 | bool created, is_root_slice; |
2177 | CGroupMask migrate_mask = 0; | |
184b4f78 | 2178 | _cleanup_free_ char *cgroup_full_path = NULL; |
27adcc97 | 2179 | int r; |
64747e2d | 2180 | |
4ad49000 | 2181 | assert(u); |
64747e2d | 2182 | |
27c4ed79 | 2183 | if (!UNIT_HAS_CGROUP_CONTEXT(u)) |
0cd385d3 LP |
2184 | return 0; |
2185 | ||
a4634b21 LP |
2186 | /* Figure out our cgroup path */ |
2187 | r = unit_pick_cgroup_path(u); | |
2188 | if (r < 0) | |
2189 | return r; | |
b58b8e11 | 2190 | |
03b90d4b | 2191 | /* First, create our own group */ |
efdb0237 | 2192 | r = cg_create_everywhere(u->manager->cgroup_supported, target_mask, u->cgroup_path); |
23bbb0de | 2193 | if (r < 0) |
6178e2f8 | 2194 | return log_unit_error_errno(u, r, "Failed to create cgroup %s: %m", empty_to_root(u->cgroup_path)); |
490c5a37 | 2195 | created = r; |
efdb0237 | 2196 | |
184b4f78 | 2197 | if (cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER) > 0) { |
1b420223 LP |
2198 | uint64_t cgroup_id = 0; |
2199 | ||
184b4f78 ILG |
2200 | r = cg_get_path(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path, NULL, &cgroup_full_path); |
2201 | if (r == 0) { | |
2202 | r = cg_path_get_cgroupid(cgroup_full_path, &cgroup_id); | |
2203 | if (r < 0) | |
1b420223 LP |
2204 | log_unit_full_errno(u, ERRNO_IS_NOT_SUPPORTED(r) ? LOG_DEBUG : LOG_WARNING, r, |
2205 | "Failed to get cgroup ID of cgroup %s, ignoring: %m", cgroup_full_path); | |
184b4f78 ILG |
2206 | } else |
2207 | log_unit_warning_errno(u, r, "Failed to get full cgroup path on cgroup %s, ignoring: %m", empty_to_root(u->cgroup_path)); | |
2208 | ||
2209 | u->cgroup_id = cgroup_id; | |
2210 | } | |
2211 | ||
efdb0237 LP |
2212 | /* Start watching it */ |
2213 | (void) unit_watch_cgroup(u); | |
afcfaa69 | 2214 | (void) unit_watch_cgroup_memory(u); |
efdb0237 | 2215 | |
7b639614 MK |
2216 | /* For v2 we preserve enabled controllers in delegated units, adjust others, |
2217 | * for v1 we figure out which controller hierarchies need migration. */ | |
1fd3a10c | 2218 | if (created || !u->cgroup_realized || !unit_cgroup_delegate(u)) { |
27adcc97 | 2219 | CGroupMask result_mask = 0; |
65be7e06 ZJS |
2220 | |
2221 | /* Enable all controllers we need */ | |
27adcc97 | 2222 | r = cg_enable_everywhere(u->manager->cgroup_supported, enable_mask, u->cgroup_path, &result_mask); |
65be7e06 | 2223 | if (r < 0) |
6178e2f8 | 2224 | log_unit_warning_errno(u, r, "Failed to enable/disable controllers on cgroup %s, ignoring: %m", empty_to_root(u->cgroup_path)); |
27adcc97 | 2225 | |
27adcc97 LP |
2226 | /* Remember what's actually enabled now */ |
2227 | u->cgroup_enabled_mask = result_mask; | |
7b639614 MK |
2228 | |
2229 | migrate_mask = u->cgroup_realized_mask ^ target_mask; | |
65be7e06 | 2230 | } |
03b90d4b LP |
2231 | |
2232 | /* Keep track that this is now realized */ | |
4ad49000 | 2233 | u->cgroup_realized = true; |
efdb0237 | 2234 | u->cgroup_realized_mask = target_mask; |
4ad49000 | 2235 | |
7b639614 MK |
2236 | /* Migrate processes in controller hierarchies both downwards (enabling) and upwards (disabling). |
2237 | * | |
2238 | * Unnecessary controller cgroups are trimmed (after emptied by upward migration). | |
2239 | * We perform migration also with whole slices for cases when users don't care about leave | |
2240 | * granularity. Since delegated_mask is subset of target mask, we won't trim slice subtree containing | |
2241 | * delegated units. | |
2242 | */ | |
2243 | if (cg_all_unified() == 0) { | |
2244 | r = cg_migrate_v1_controllers(u->manager->cgroup_supported, migrate_mask, u->cgroup_path, migrate_callback, u); | |
2245 | if (r < 0) | |
6178e2f8 | 2246 | log_unit_warning_errno(u, r, "Failed to migrate controller cgroups from %s, ignoring: %m", empty_to_root(u->cgroup_path)); |
0cd385d3 | 2247 | |
7b639614 MK |
2248 | is_root_slice = unit_has_name(u, SPECIAL_ROOT_SLICE); |
2249 | r = cg_trim_v1_controllers(u->manager->cgroup_supported, ~target_mask, u->cgroup_path, !is_root_slice); | |
0cd385d3 | 2250 | if (r < 0) |
6178e2f8 | 2251 | log_unit_warning_errno(u, r, "Failed to delete controller cgroups %s, ignoring: %m", empty_to_root(u->cgroup_path)); |
0cd385d3 | 2252 | } |
03b90d4b | 2253 | |
0d2d6fbf CD |
2254 | /* Set attributes */ |
2255 | cgroup_context_apply(u, target_mask, state); | |
2256 | cgroup_xattr_apply(u); | |
2257 | ||
64747e2d LP |
2258 | return 0; |
2259 | } | |
2260 | ||
6592b975 LP |
2261 | static int unit_attach_pid_to_cgroup_via_bus(Unit *u, pid_t pid, const char *suffix_path) { |
2262 | _cleanup_(sd_bus_error_free) sd_bus_error error = SD_BUS_ERROR_NULL; | |
2263 | char *pp; | |
7b3fd631 | 2264 | int r; |
6592b975 | 2265 | |
7b3fd631 LP |
2266 | assert(u); |
2267 | ||
6592b975 LP |
2268 | if (MANAGER_IS_SYSTEM(u->manager)) |
2269 | return -EINVAL; | |
2270 | ||
2271 | if (!u->manager->system_bus) | |
2272 | return -EIO; | |
2273 | ||
2274 | if (!u->cgroup_path) | |
2275 | return -EINVAL; | |
2276 | ||
2277 | /* Determine this unit's cgroup path relative to our cgroup root */ | |
2278 | pp = path_startswith(u->cgroup_path, u->manager->cgroup_root); | |
2279 | if (!pp) | |
2280 | return -EINVAL; | |
2281 | ||
2282 | pp = strjoina("/", pp, suffix_path); | |
4ff361cc | 2283 | path_simplify(pp); |
6592b975 | 2284 | |
78fa2f91 | 2285 | r = bus_call_method(u->manager->system_bus, |
2286 | bus_systemd_mgr, | |
2287 | "AttachProcessesToUnit", | |
2288 | &error, NULL, | |
2289 | "ssau", | |
2290 | NULL /* empty unit name means client's unit, i.e. us */, pp, 1, (uint32_t) pid); | |
7b3fd631 | 2291 | if (r < 0) |
6592b975 LP |
2292 | return log_unit_debug_errno(u, r, "Failed to attach unit process " PID_FMT " via the bus: %s", pid, bus_error_message(&error, r)); |
2293 | ||
2294 | return 0; | |
2295 | } | |
2296 | ||
2297 | int unit_attach_pids_to_cgroup(Unit *u, Set *pids, const char *suffix_path) { | |
2298 | CGroupMask delegated_mask; | |
2299 | const char *p; | |
6592b975 | 2300 | void *pidp; |
db4229d1 | 2301 | int ret, r; |
6592b975 LP |
2302 | |
2303 | assert(u); | |
2304 | ||
2305 | if (!UNIT_HAS_CGROUP_CONTEXT(u)) | |
2306 | return -EINVAL; | |
2307 | ||
2308 | if (set_isempty(pids)) | |
2309 | return 0; | |
7b3fd631 | 2310 | |
fab34748 KL |
2311 | /* Load any custom firewall BPF programs here once to test if they are existing and actually loadable. |
2312 | * Fail here early since later errors in the call chain unit_realize_cgroup to cgroup_context_apply are ignored. */ | |
2313 | r = bpf_firewall_load_custom(u); | |
2314 | if (r < 0) | |
2315 | return r; | |
2316 | ||
6592b975 | 2317 | r = unit_realize_cgroup(u); |
7b3fd631 LP |
2318 | if (r < 0) |
2319 | return r; | |
2320 | ||
6592b975 LP |
2321 | if (isempty(suffix_path)) |
2322 | p = u->cgroup_path; | |
2323 | else | |
270384b2 | 2324 | p = prefix_roota(u->cgroup_path, suffix_path); |
6592b975 LP |
2325 | |
2326 | delegated_mask = unit_get_delegate_mask(u); | |
2327 | ||
db4229d1 | 2328 | ret = 0; |
90e74a66 | 2329 | SET_FOREACH(pidp, pids) { |
6592b975 | 2330 | pid_t pid = PTR_TO_PID(pidp); |
6592b975 LP |
2331 | |
2332 | /* First, attach the PID to the main cgroup hierarchy */ | |
db4229d1 LP |
2333 | r = cg_attach(SYSTEMD_CGROUP_CONTROLLER, p, pid); |
2334 | if (r < 0) { | |
2335 | bool again = MANAGER_IS_USER(u->manager) && ERRNO_IS_PRIVILEGE(r); | |
6592b975 | 2336 | |
db4229d1 | 2337 | log_unit_full_errno(u, again ? LOG_DEBUG : LOG_INFO, r, |
7a2ba407 | 2338 | "Couldn't move process "PID_FMT" to%s requested cgroup '%s': %m", |
6178e2f8 | 2339 | pid, again ? " directly" : "", empty_to_root(p)); |
7a2ba407 ZJS |
2340 | |
2341 | if (again) { | |
6592b975 LP |
2342 | int z; |
2343 | ||
7a2ba407 ZJS |
2344 | /* If we are in a user instance, and we can't move the process ourselves due |
2345 | * to permission problems, let's ask the system instance about it instead. | |
2346 | * Since it's more privileged it might be able to move the process across the | |
2347 | * leaves of a subtree whose top node is not owned by us. */ | |
6592b975 LP |
2348 | |
2349 | z = unit_attach_pid_to_cgroup_via_bus(u, pid, suffix_path); | |
2350 | if (z < 0) | |
6178e2f8 | 2351 | 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 |
2352 | else { |
2353 | if (ret >= 0) | |
2354 | ret++; /* Count successful additions */ | |
6592b975 | 2355 | continue; /* When the bus thing worked via the bus we are fully done for this PID. */ |
c65417a0 | 2356 | } |
6592b975 LP |
2357 | } |
2358 | ||
db4229d1 LP |
2359 | if (ret >= 0) |
2360 | ret = r; /* Remember first error */ | |
6592b975 LP |
2361 | |
2362 | continue; | |
8d3e4ac7 LP |
2363 | } else if (ret >= 0) |
2364 | ret++; /* Count successful additions */ | |
6592b975 | 2365 | |
db4229d1 LP |
2366 | r = cg_all_unified(); |
2367 | if (r < 0) | |
2368 | return r; | |
2369 | if (r > 0) | |
6592b975 LP |
2370 | continue; |
2371 | ||
2372 | /* In the legacy hierarchy, attach the process to the request cgroup if possible, and if not to the | |
2373 | * innermost realized one */ | |
2374 | ||
e8616626 | 2375 | for (CGroupController c = 0; c < _CGROUP_CONTROLLER_MAX; c++) { |
6592b975 LP |
2376 | CGroupMask bit = CGROUP_CONTROLLER_TO_MASK(c); |
2377 | const char *realized; | |
2378 | ||
2379 | if (!(u->manager->cgroup_supported & bit)) | |
2380 | continue; | |
2381 | ||
2382 | /* If this controller is delegated and realized, honour the caller's request for the cgroup suffix. */ | |
2383 | if (delegated_mask & u->cgroup_realized_mask & bit) { | |
db4229d1 LP |
2384 | r = cg_attach(cgroup_controller_to_string(c), p, pid); |
2385 | if (r >= 0) | |
6592b975 LP |
2386 | continue; /* Success! */ |
2387 | ||
db4229d1 | 2388 | 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 | 2389 | pid, empty_to_root(p), cgroup_controller_to_string(c)); |
6592b975 LP |
2390 | } |
2391 | ||
2392 | /* So this controller is either not delegate or realized, or something else weird happened. In | |
2393 | * that case let's attach the PID at least to the closest cgroup up the tree that is | |
2394 | * realized. */ | |
2395 | realized = unit_get_realized_cgroup_path(u, bit); | |
2396 | if (!realized) | |
2397 | continue; /* Not even realized in the root slice? Then let's not bother */ | |
2398 | ||
db4229d1 LP |
2399 | r = cg_attach(cgroup_controller_to_string(c), realized, pid); |
2400 | if (r < 0) | |
2401 | log_unit_debug_errno(u, r, "Failed to attach PID " PID_FMT " to realized cgroup %s in controller %s, ignoring: %m", | |
6592b975 LP |
2402 | pid, realized, cgroup_controller_to_string(c)); |
2403 | } | |
2404 | } | |
2405 | ||
db4229d1 | 2406 | return ret; |
7b3fd631 LP |
2407 | } |
2408 | ||
906c06f6 DM |
2409 | static bool unit_has_mask_realized( |
2410 | Unit *u, | |
2411 | CGroupMask target_mask, | |
17f14955 | 2412 | CGroupMask enable_mask) { |
906c06f6 | 2413 | |
bc432dc7 LP |
2414 | assert(u); |
2415 | ||
d5095dcd LP |
2416 | /* Returns true if this unit is fully realized. We check four things: |
2417 | * | |
2418 | * 1. Whether the cgroup was created at all | |
4e1dfa45 CD |
2419 | * 2. Whether the cgroup was created in all the hierarchies we need it to be created in (in case of cgroup v1) |
2420 | * 3. Whether the cgroup has all the right controllers enabled (in case of cgroup v2) | |
d5095dcd LP |
2421 | * 4. Whether the invalidation mask is currently zero |
2422 | * | |
2423 | * If you wonder why we mask the target realization and enable mask with CGROUP_MASK_V1/CGROUP_MASK_V2: note | |
4e1dfa45 CD |
2424 | * that there are three sets of bitmasks: CGROUP_MASK_V1 (for real cgroup v1 controllers), CGROUP_MASK_V2 (for |
2425 | * real cgroup v2 controllers) and CGROUP_MASK_BPF (for BPF-based pseudo-controllers). Now, cgroup_realized_mask | |
2426 | * is only matters for cgroup v1 controllers, and cgroup_enabled_mask only used for cgroup v2, and if they | |
d5095dcd LP |
2427 | * differ in the others, we don't really care. (After all, the cgroup_enabled_mask tracks with controllers are |
2428 | * enabled through cgroup.subtree_control, and since the BPF pseudo-controllers don't show up there, they | |
2429 | * simply don't matter. */ | |
2430 | ||
906c06f6 | 2431 | return u->cgroup_realized && |
d5095dcd LP |
2432 | ((u->cgroup_realized_mask ^ target_mask) & CGROUP_MASK_V1) == 0 && |
2433 | ((u->cgroup_enabled_mask ^ enable_mask) & CGROUP_MASK_V2) == 0 && | |
17f14955 | 2434 | u->cgroup_invalidated_mask == 0; |
6414b7c9 DS |
2435 | } |
2436 | ||
4f6f62e4 CD |
2437 | static bool unit_has_mask_disables_realized( |
2438 | Unit *u, | |
2439 | CGroupMask target_mask, | |
2440 | CGroupMask enable_mask) { | |
2441 | ||
2442 | assert(u); | |
2443 | ||
2444 | /* Returns true if all controllers which should be disabled are indeed disabled. | |
2445 | * | |
2446 | * Unlike unit_has_mask_realized, we don't care what was enabled, only that anything we want to remove is | |
2447 | * already removed. */ | |
2448 | ||
2449 | return !u->cgroup_realized || | |
2450 | (FLAGS_SET(u->cgroup_realized_mask, target_mask & CGROUP_MASK_V1) && | |
2451 | FLAGS_SET(u->cgroup_enabled_mask, enable_mask & CGROUP_MASK_V2)); | |
2452 | } | |
2453 | ||
a57669d2 CD |
2454 | static bool unit_has_mask_enables_realized( |
2455 | Unit *u, | |
2456 | CGroupMask target_mask, | |
2457 | CGroupMask enable_mask) { | |
2458 | ||
2459 | assert(u); | |
2460 | ||
2461 | /* Returns true if all controllers which should be enabled are indeed enabled. | |
2462 | * | |
2463 | * Unlike unit_has_mask_realized, we don't care about the controllers that are not present, only that anything | |
2464 | * we want to add is already added. */ | |
2465 | ||
2466 | return u->cgroup_realized && | |
c72703e2 CD |
2467 | ((u->cgroup_realized_mask | target_mask) & CGROUP_MASK_V1) == (u->cgroup_realized_mask & CGROUP_MASK_V1) && |
2468 | ((u->cgroup_enabled_mask | enable_mask) & CGROUP_MASK_V2) == (u->cgroup_enabled_mask & CGROUP_MASK_V2); | |
a57669d2 CD |
2469 | } |
2470 | ||
fb46fca7 | 2471 | static void unit_add_to_cgroup_realize_queue(Unit *u) { |
2aa57a65 LP |
2472 | assert(u); |
2473 | ||
2474 | if (u->in_cgroup_realize_queue) | |
2475 | return; | |
2476 | ||
a479c21e | 2477 | LIST_APPEND(cgroup_realize_queue, u->manager->cgroup_realize_queue, u); |
2aa57a65 LP |
2478 | u->in_cgroup_realize_queue = true; |
2479 | } | |
2480 | ||
2481 | static void unit_remove_from_cgroup_realize_queue(Unit *u) { | |
2482 | assert(u); | |
2483 | ||
2484 | if (!u->in_cgroup_realize_queue) | |
2485 | return; | |
2486 | ||
2487 | LIST_REMOVE(cgroup_realize_queue, u->manager->cgroup_realize_queue, u); | |
2488 | u->in_cgroup_realize_queue = false; | |
2489 | } | |
2490 | ||
a57669d2 CD |
2491 | /* Controllers can only be enabled breadth-first, from the root of the |
2492 | * hierarchy downwards to the unit in question. */ | |
2493 | static int unit_realize_cgroup_now_enable(Unit *u, ManagerState state) { | |
2494 | CGroupMask target_mask, enable_mask, new_target_mask, new_enable_mask; | |
12f64221 | 2495 | Unit *slice; |
a57669d2 CD |
2496 | int r; |
2497 | ||
2498 | assert(u); | |
2499 | ||
2500 | /* First go deal with this unit's parent, or we won't be able to enable | |
2501 | * any new controllers at this layer. */ | |
12f64221 LP |
2502 | slice = UNIT_GET_SLICE(u); |
2503 | if (slice) { | |
2504 | r = unit_realize_cgroup_now_enable(slice, state); | |
a57669d2 CD |
2505 | if (r < 0) |
2506 | return r; | |
2507 | } | |
2508 | ||
2509 | target_mask = unit_get_target_mask(u); | |
2510 | enable_mask = unit_get_enable_mask(u); | |
2511 | ||
2512 | /* We can only enable in this direction, don't try to disable anything. | |
2513 | */ | |
2514 | if (unit_has_mask_enables_realized(u, target_mask, enable_mask)) | |
2515 | return 0; | |
2516 | ||
2517 | new_target_mask = u->cgroup_realized_mask | target_mask; | |
2518 | new_enable_mask = u->cgroup_enabled_mask | enable_mask; | |
2519 | ||
7b639614 | 2520 | return unit_update_cgroup(u, new_target_mask, new_enable_mask, state); |
a57669d2 CD |
2521 | } |
2522 | ||
4f6f62e4 CD |
2523 | /* Controllers can only be disabled depth-first, from the leaves of the |
2524 | * hierarchy upwards to the unit in question. */ | |
2525 | static int unit_realize_cgroup_now_disable(Unit *u, ManagerState state) { | |
4f6f62e4 | 2526 | Unit *m; |
4f6f62e4 CD |
2527 | |
2528 | assert(u); | |
2529 | ||
2530 | if (u->type != UNIT_SLICE) | |
2531 | return 0; | |
2532 | ||
d219a2b0 | 2533 | UNIT_FOREACH_DEPENDENCY(m, u, UNIT_ATOM_SLICE_OF) { |
4f6f62e4 CD |
2534 | CGroupMask target_mask, enable_mask, new_target_mask, new_enable_mask; |
2535 | int r; | |
2536 | ||
defe63b0 LP |
2537 | /* The cgroup for this unit might not actually be fully realised yet, in which case it isn't |
2538 | * holding any controllers open anyway. */ | |
d9ef5944 | 2539 | if (!m->cgroup_realized) |
4f6f62e4 CD |
2540 | continue; |
2541 | ||
defe63b0 | 2542 | /* We must disable those below us first in order to release the controller. */ |
4f6f62e4 CD |
2543 | if (m->type == UNIT_SLICE) |
2544 | (void) unit_realize_cgroup_now_disable(m, state); | |
2545 | ||
2546 | target_mask = unit_get_target_mask(m); | |
2547 | enable_mask = unit_get_enable_mask(m); | |
2548 | ||
defe63b0 | 2549 | /* We can only disable in this direction, don't try to enable anything. */ |
4f6f62e4 CD |
2550 | if (unit_has_mask_disables_realized(m, target_mask, enable_mask)) |
2551 | continue; | |
2552 | ||
2553 | new_target_mask = m->cgroup_realized_mask & target_mask; | |
2554 | new_enable_mask = m->cgroup_enabled_mask & enable_mask; | |
2555 | ||
7b639614 | 2556 | r = unit_update_cgroup(m, new_target_mask, new_enable_mask, state); |
4f6f62e4 CD |
2557 | if (r < 0) |
2558 | return r; | |
2559 | } | |
2560 | ||
2561 | return 0; | |
2562 | } | |
a57669d2 | 2563 | |
6414b7c9 DS |
2564 | /* Check if necessary controllers and attributes for a unit are in place. |
2565 | * | |
a57669d2 CD |
2566 | * - If so, do nothing. |
2567 | * - If not, create paths, move processes over, and set attributes. | |
2568 | * | |
2569 | * Controllers can only be *enabled* in a breadth-first way, and *disabled* in | |
2570 | * a depth-first way. As such the process looks like this: | |
2571 | * | |
2572 | * Suppose we have a cgroup hierarchy which looks like this: | |
2573 | * | |
2574 | * root | |
2575 | * / \ | |
2576 | * / \ | |
2577 | * / \ | |
2578 | * a b | |
2579 | * / \ / \ | |
2580 | * / \ / \ | |
2581 | * c d e f | |
2582 | * / \ / \ / \ / \ | |
2583 | * h i j k l m n o | |
2584 | * | |
2585 | * 1. We want to realise cgroup "d" now. | |
c72703e2 | 2586 | * 2. cgroup "a" has DisableControllers=cpu in the associated unit. |
a57669d2 CD |
2587 | * 3. cgroup "k" just started requesting the memory controller. |
2588 | * | |
2589 | * To make this work we must do the following in order: | |
2590 | * | |
2591 | * 1. Disable CPU controller in k, j | |
2592 | * 2. Disable CPU controller in d | |
2593 | * 3. Enable memory controller in root | |
2594 | * 4. Enable memory controller in a | |
2595 | * 5. Enable memory controller in d | |
2596 | * 6. Enable memory controller in k | |
2597 | * | |
2598 | * Notice that we need to touch j in one direction, but not the other. We also | |
2599 | * don't go beyond d when disabling -- it's up to "a" to get realized if it | |
2600 | * wants to disable further. The basic rules are therefore: | |
2601 | * | |
2602 | * - If you're disabling something, you need to realise all of the cgroups from | |
2603 | * your recursive descendants to the root. This starts from the leaves. | |
2604 | * - If you're enabling something, you need to realise from the root cgroup | |
2605 | * downwards, but you don't need to iterate your recursive descendants. | |
6414b7c9 DS |
2606 | * |
2607 | * Returns 0 on success and < 0 on failure. */ | |
db785129 | 2608 | static int unit_realize_cgroup_now(Unit *u, ManagerState state) { |
efdb0237 | 2609 | CGroupMask target_mask, enable_mask; |
12f64221 | 2610 | Unit *slice; |
6414b7c9 | 2611 | int r; |
64747e2d | 2612 | |
4ad49000 | 2613 | assert(u); |
64747e2d | 2614 | |
2aa57a65 | 2615 | unit_remove_from_cgroup_realize_queue(u); |
64747e2d | 2616 | |
efdb0237 | 2617 | target_mask = unit_get_target_mask(u); |
ccf78df1 TH |
2618 | enable_mask = unit_get_enable_mask(u); |
2619 | ||
17f14955 | 2620 | if (unit_has_mask_realized(u, target_mask, enable_mask)) |
0a1eb06d | 2621 | return 0; |
64747e2d | 2622 | |
4f6f62e4 CD |
2623 | /* Disable controllers below us, if there are any */ |
2624 | r = unit_realize_cgroup_now_disable(u, state); | |
2625 | if (r < 0) | |
2626 | return r; | |
2627 | ||
2628 | /* Enable controllers above us, if there are any */ | |
12f64221 LP |
2629 | slice = UNIT_GET_SLICE(u); |
2630 | if (slice) { | |
2631 | r = unit_realize_cgroup_now_enable(slice, state); | |
6414b7c9 DS |
2632 | if (r < 0) |
2633 | return r; | |
2634 | } | |
4ad49000 | 2635 | |
0d2d6fbf | 2636 | /* Now actually deal with the cgroup we were trying to realise and set attributes */ |
7b639614 | 2637 | r = unit_update_cgroup(u, target_mask, enable_mask, state); |
6414b7c9 DS |
2638 | if (r < 0) |
2639 | return r; | |
2640 | ||
c2baf11c LP |
2641 | /* Now, reset the invalidation mask */ |
2642 | u->cgroup_invalidated_mask = 0; | |
6414b7c9 | 2643 | return 0; |
64747e2d LP |
2644 | } |
2645 | ||
91a6073e | 2646 | unsigned manager_dispatch_cgroup_realize_queue(Manager *m) { |
db785129 | 2647 | ManagerState state; |
4ad49000 | 2648 | unsigned n = 0; |
db785129 | 2649 | Unit *i; |
6414b7c9 | 2650 | int r; |
ecedd90f | 2651 | |
91a6073e LP |
2652 | assert(m); |
2653 | ||
db785129 LP |
2654 | state = manager_state(m); |
2655 | ||
91a6073e LP |
2656 | while ((i = m->cgroup_realize_queue)) { |
2657 | assert(i->in_cgroup_realize_queue); | |
ecedd90f | 2658 | |
2aa57a65 LP |
2659 | if (UNIT_IS_INACTIVE_OR_FAILED(unit_active_state(i))) { |
2660 | /* Maybe things changed, and the unit is not actually active anymore? */ | |
2661 | unit_remove_from_cgroup_realize_queue(i); | |
2662 | continue; | |
2663 | } | |
2664 | ||
db785129 | 2665 | r = unit_realize_cgroup_now(i, state); |
6414b7c9 | 2666 | if (r < 0) |
efdb0237 | 2667 | log_warning_errno(r, "Failed to realize cgroups for queued unit %s, ignoring: %m", i->id); |
0a1eb06d | 2668 | |
4ad49000 LP |
2669 | n++; |
2670 | } | |
ecedd90f | 2671 | |
4ad49000 | 2672 | return n; |
8e274523 LP |
2673 | } |
2674 | ||
4c591f39 MK |
2675 | void unit_add_family_to_cgroup_realize_queue(Unit *u) { |
2676 | assert(u); | |
2677 | assert(u->type == UNIT_SLICE); | |
ca949c9d | 2678 | |
4c591f39 MK |
2679 | /* Family of a unit for is defined as (immediate) children of the unit and immediate children of all |
2680 | * its ancestors. | |
2681 | * | |
2682 | * Ideally we would enqueue ancestor path only (bottom up). However, on cgroup-v1 scheduling becomes | |
2683 | * very weird if two units that own processes reside in the same slice, but one is realized in the | |
2684 | * "cpu" hierarchy and one is not (for example because one has CPUWeight= set and the other does | |
2685 | * not), because that means individual processes need to be scheduled against whole cgroups. Let's | |
2686 | * avoid this asymmetry by always ensuring that siblings of a unit are always realized in their v1 | |
2687 | * controller hierarchies too (if unit requires the controller to be realized). | |
e1e98911 | 2688 | * |
4c591f39 MK |
2689 | * The function must invalidate cgroup_members_mask of all ancestors in order to calculate up to date |
2690 | * masks. */ | |
2691 | ||
2692 | do { | |
4ad49000 | 2693 | Unit *m; |
8f53a7b8 | 2694 | |
4c591f39 MK |
2695 | /* Children of u likely changed when we're called */ |
2696 | u->cgroup_members_mask_valid = false; | |
f23ba94d | 2697 | |
d219a2b0 | 2698 | UNIT_FOREACH_DEPENDENCY(m, u, UNIT_ATOM_SLICE_OF) { |
8e274523 | 2699 | |
65f6b6bd | 2700 | /* No point in doing cgroup application for units without active processes. */ |
6414b7c9 DS |
2701 | if (UNIT_IS_INACTIVE_OR_FAILED(unit_active_state(m))) |
2702 | continue; | |
2703 | ||
e1e98911 LP |
2704 | /* We only enqueue siblings if they were realized once at least, in the main |
2705 | * hierarchy. */ | |
2706 | if (!m->cgroup_realized) | |
2707 | continue; | |
2708 | ||
defe63b0 LP |
2709 | /* If the unit doesn't need any new controllers and has current ones |
2710 | * realized, it doesn't need any changes. */ | |
906c06f6 DM |
2711 | if (unit_has_mask_realized(m, |
2712 | unit_get_target_mask(m), | |
17f14955 | 2713 | unit_get_enable_mask(m))) |
6414b7c9 DS |
2714 | continue; |
2715 | ||
91a6073e | 2716 | unit_add_to_cgroup_realize_queue(m); |
50159e6a LP |
2717 | } |
2718 | ||
4c591f39 MK |
2719 | /* Parent comes after children */ |
2720 | unit_add_to_cgroup_realize_queue(u); | |
12f64221 LP |
2721 | |
2722 | u = UNIT_GET_SLICE(u); | |
2723 | } while (u); | |
4ad49000 LP |
2724 | } |
2725 | ||
0a1eb06d | 2726 | int unit_realize_cgroup(Unit *u) { |
12f64221 LP |
2727 | Unit *slice; |
2728 | ||
4ad49000 LP |
2729 | assert(u); |
2730 | ||
35b7ff80 | 2731 | if (!UNIT_HAS_CGROUP_CONTEXT(u)) |
0a1eb06d | 2732 | return 0; |
8e274523 | 2733 | |
4c591f39 MK |
2734 | /* So, here's the deal: when realizing the cgroups for this unit, we need to first create all |
2735 | * parents, but there's more actually: for the weight-based controllers we also need to make sure | |
2736 | * that all our siblings (i.e. units that are in the same slice as we are) have cgroups, too. On the | |
2737 | * other hand, when a controller is removed from realized set, it may become unnecessary in siblings | |
2738 | * and ancestors and they should be (de)realized too. | |
2739 | * | |
2740 | * This call will defer work on the siblings and derealized ancestors to the next event loop | |
2741 | * iteration and synchronously creates the parent cgroups (unit_realize_cgroup_now). */ | |
ca949c9d | 2742 | |
12f64221 LP |
2743 | slice = UNIT_GET_SLICE(u); |
2744 | if (slice) | |
2745 | unit_add_family_to_cgroup_realize_queue(slice); | |
4ad49000 | 2746 | |
6414b7c9 | 2747 | /* And realize this one now (and apply the values) */ |
db785129 | 2748 | return unit_realize_cgroup_now(u, manager_state(u->manager)); |
8e274523 LP |
2749 | } |
2750 | ||
efdb0237 LP |
2751 | void unit_release_cgroup(Unit *u) { |
2752 | assert(u); | |
2753 | ||
8a0d5388 LP |
2754 | /* Forgets all cgroup details for this cgroup — but does *not* destroy the cgroup. This is hence OK to call |
2755 | * when we close down everything for reexecution, where we really want to leave the cgroup in place. */ | |
efdb0237 LP |
2756 | |
2757 | if (u->cgroup_path) { | |
2758 | (void) hashmap_remove(u->manager->cgroup_unit, u->cgroup_path); | |
2759 | u->cgroup_path = mfree(u->cgroup_path); | |
2760 | } | |
2761 | ||
0bb814c2 LP |
2762 | if (u->cgroup_control_inotify_wd >= 0) { |
2763 | if (inotify_rm_watch(u->manager->cgroup_inotify_fd, u->cgroup_control_inotify_wd) < 0) | |
2764 | 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 | 2765 | |
0bb814c2 LP |
2766 | (void) hashmap_remove(u->manager->cgroup_control_inotify_wd_unit, INT_TO_PTR(u->cgroup_control_inotify_wd)); |
2767 | u->cgroup_control_inotify_wd = -1; | |
efdb0237 | 2768 | } |
afcfaa69 LP |
2769 | |
2770 | if (u->cgroup_memory_inotify_wd >= 0) { | |
2771 | if (inotify_rm_watch(u->manager->cgroup_inotify_fd, u->cgroup_memory_inotify_wd) < 0) | |
2772 | 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); | |
2773 | ||
2774 | (void) hashmap_remove(u->manager->cgroup_memory_inotify_wd_unit, INT_TO_PTR(u->cgroup_memory_inotify_wd)); | |
2775 | u->cgroup_memory_inotify_wd = -1; | |
2776 | } | |
efdb0237 LP |
2777 | } |
2778 | ||
e08dabfe AZ |
2779 | bool unit_maybe_release_cgroup(Unit *u) { |
2780 | int r; | |
2781 | ||
2782 | assert(u); | |
2783 | ||
2784 | if (!u->cgroup_path) | |
2785 | return true; | |
2786 | ||
2787 | /* Don't release the cgroup if there are still processes under it. If we get notified later when all the | |
2788 | * processes exit (e.g. the processes were in D-state and exited after the unit was marked as failed) | |
2789 | * we need the cgroup paths to continue to be tracked by the manager so they can be looked up and cleaned | |
2790 | * up later. */ | |
2791 | r = cg_is_empty_recursive(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path); | |
2792 | if (r < 0) | |
2793 | log_unit_debug_errno(u, r, "Error checking if the cgroup is recursively empty, ignoring: %m"); | |
2794 | else if (r == 1) { | |
2795 | unit_release_cgroup(u); | |
2796 | return true; | |
2797 | } | |
2798 | ||
2799 | return false; | |
2800 | } | |
2801 | ||
efdb0237 | 2802 | void unit_prune_cgroup(Unit *u) { |
8e274523 | 2803 | int r; |
efdb0237 | 2804 | bool is_root_slice; |
8e274523 | 2805 | |
4ad49000 | 2806 | assert(u); |
8e274523 | 2807 | |
efdb0237 LP |
2808 | /* Removes the cgroup, if empty and possible, and stops watching it. */ |
2809 | ||
4ad49000 LP |
2810 | if (!u->cgroup_path) |
2811 | return; | |
8e274523 | 2812 | |
fe700f46 LP |
2813 | (void) unit_get_cpu_usage(u, NULL); /* Cache the last CPU usage value before we destroy the cgroup */ |
2814 | ||
b1994387 ILG |
2815 | #if BPF_FRAMEWORK |
2816 | (void) lsm_bpf_cleanup(u); /* Remove cgroup from the global LSM BPF map */ | |
2817 | #endif | |
2818 | ||
efdb0237 LP |
2819 | is_root_slice = unit_has_name(u, SPECIAL_ROOT_SLICE); |
2820 | ||
2821 | r = cg_trim_everywhere(u->manager->cgroup_supported, u->cgroup_path, !is_root_slice); | |
0219b352 DB |
2822 | if (r < 0) |
2823 | /* One reason we could have failed here is, that the cgroup still contains a process. | |
2824 | * However, if the cgroup becomes removable at a later time, it might be removed when | |
2825 | * the containing slice is stopped. So even if we failed now, this unit shouldn't assume | |
2826 | * that the cgroup is still realized the next time it is started. Do not return early | |
2827 | * on error, continue cleanup. */ | |
6178e2f8 | 2828 | 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 | 2829 | |
efdb0237 LP |
2830 | if (is_root_slice) |
2831 | return; | |
2832 | ||
e08dabfe AZ |
2833 | if (!unit_maybe_release_cgroup(u)) /* Returns true if the cgroup was released */ |
2834 | return; | |
0a1eb06d | 2835 | |
4ad49000 | 2836 | u->cgroup_realized = false; |
bc432dc7 | 2837 | u->cgroup_realized_mask = 0; |
ccf78df1 | 2838 | u->cgroup_enabled_mask = 0; |
084c7007 | 2839 | |
76dc1725 | 2840 | u->bpf_device_control_installed = bpf_program_free(u->bpf_device_control_installed); |
8e274523 LP |
2841 | } |
2842 | ||
efdb0237 | 2843 | int unit_search_main_pid(Unit *u, pid_t *ret) { |
4ad49000 | 2844 | _cleanup_fclose_ FILE *f = NULL; |
4d051546 | 2845 | pid_t pid = 0, npid; |
efdb0237 | 2846 | int r; |
4ad49000 LP |
2847 | |
2848 | assert(u); | |
efdb0237 | 2849 | assert(ret); |
4ad49000 LP |
2850 | |
2851 | if (!u->cgroup_path) | |
efdb0237 | 2852 | return -ENXIO; |
4ad49000 | 2853 | |
efdb0237 LP |
2854 | r = cg_enumerate_processes(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path, &f); |
2855 | if (r < 0) | |
2856 | return r; | |
4ad49000 | 2857 | |
4ad49000 | 2858 | while (cg_read_pid(f, &npid) > 0) { |
4ad49000 LP |
2859 | |
2860 | if (npid == pid) | |
2861 | continue; | |
8e274523 | 2862 | |
4d051546 | 2863 | if (pid_is_my_child(npid) == 0) |
4ad49000 | 2864 | continue; |
8e274523 | 2865 | |
efdb0237 | 2866 | if (pid != 0) |
4ad49000 LP |
2867 | /* Dang, there's more than one daemonized PID |
2868 | in this group, so we don't know what process | |
2869 | is the main process. */ | |
efdb0237 LP |
2870 | |
2871 | return -ENODATA; | |
8e274523 | 2872 | |
4ad49000 | 2873 | pid = npid; |
8e274523 LP |
2874 | } |
2875 | ||
efdb0237 LP |
2876 | *ret = pid; |
2877 | return 0; | |
2878 | } | |
2879 | ||
2880 | static int unit_watch_pids_in_path(Unit *u, const char *path) { | |
b3c5bad3 | 2881 | _cleanup_closedir_ DIR *d = NULL; |
efdb0237 LP |
2882 | _cleanup_fclose_ FILE *f = NULL; |
2883 | int ret = 0, r; | |
2884 | ||
2885 | assert(u); | |
2886 | assert(path); | |
2887 | ||
2888 | r = cg_enumerate_processes(SYSTEMD_CGROUP_CONTROLLER, path, &f); | |
2889 | if (r < 0) | |
2890 | ret = r; | |
2891 | else { | |
2892 | pid_t pid; | |
2893 | ||
2894 | while ((r = cg_read_pid(f, &pid)) > 0) { | |
f75f613d | 2895 | r = unit_watch_pid(u, pid, false); |
efdb0237 LP |
2896 | if (r < 0 && ret >= 0) |
2897 | ret = r; | |
2898 | } | |
2899 | ||
2900 | if (r < 0 && ret >= 0) | |
2901 | ret = r; | |
2902 | } | |
2903 | ||
2904 | r = cg_enumerate_subgroups(SYSTEMD_CGROUP_CONTROLLER, path, &d); | |
2905 | if (r < 0) { | |
2906 | if (ret >= 0) | |
2907 | ret = r; | |
2908 | } else { | |
2909 | char *fn; | |
2910 | ||
2911 | while ((r = cg_read_subgroup(d, &fn)) > 0) { | |
2912 | _cleanup_free_ char *p = NULL; | |
2913 | ||
95b21cff | 2914 | p = path_join(empty_to_root(path), fn); |
efdb0237 LP |
2915 | free(fn); |
2916 | ||
2917 | if (!p) | |
2918 | return -ENOMEM; | |
2919 | ||
2920 | r = unit_watch_pids_in_path(u, p); | |
2921 | if (r < 0 && ret >= 0) | |
2922 | ret = r; | |
2923 | } | |
2924 | ||
2925 | if (r < 0 && ret >= 0) | |
2926 | ret = r; | |
2927 | } | |
2928 | ||
2929 | return ret; | |
2930 | } | |
2931 | ||
11aef522 LP |
2932 | int unit_synthesize_cgroup_empty_event(Unit *u) { |
2933 | int r; | |
2934 | ||
2935 | assert(u); | |
2936 | ||
2937 | /* Enqueue a synthetic cgroup empty event if this unit doesn't watch any PIDs anymore. This is compatibility | |
2938 | * support for non-unified systems where notifications aren't reliable, and hence need to take whatever we can | |
2939 | * get as notification source as soon as we stopped having any useful PIDs to watch for. */ | |
2940 | ||
2941 | if (!u->cgroup_path) | |
2942 | return -ENOENT; | |
2943 | ||
2944 | r = cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER); | |
2945 | if (r < 0) | |
2946 | return r; | |
2947 | if (r > 0) /* On unified we have reliable notifications, and don't need this */ | |
2948 | return 0; | |
2949 | ||
2950 | if (!set_isempty(u->pids)) | |
2951 | return 0; | |
2952 | ||
2953 | unit_add_to_cgroup_empty_queue(u); | |
2954 | return 0; | |
2955 | } | |
2956 | ||
efdb0237 | 2957 | int unit_watch_all_pids(Unit *u) { |
b4cccbc1 LP |
2958 | int r; |
2959 | ||
efdb0237 LP |
2960 | assert(u); |
2961 | ||
2962 | /* Adds all PIDs from our cgroup to the set of PIDs we | |
2963 | * watch. This is a fallback logic for cases where we do not | |
2964 | * get reliable cgroup empty notifications: we try to use | |
2965 | * SIGCHLD as replacement. */ | |
2966 | ||
2967 | if (!u->cgroup_path) | |
2968 | return -ENOENT; | |
2969 | ||
c22800e4 | 2970 | r = cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER); |
b4cccbc1 LP |
2971 | if (r < 0) |
2972 | return r; | |
2973 | if (r > 0) /* On unified we can use proper notifications */ | |
efdb0237 LP |
2974 | return 0; |
2975 | ||
2976 | return unit_watch_pids_in_path(u, u->cgroup_path); | |
2977 | } | |
2978 | ||
09e24654 LP |
2979 | static int on_cgroup_empty_event(sd_event_source *s, void *userdata) { |
2980 | Manager *m = userdata; | |
2981 | Unit *u; | |
efdb0237 LP |
2982 | int r; |
2983 | ||
09e24654 LP |
2984 | assert(s); |
2985 | assert(m); | |
efdb0237 | 2986 | |
09e24654 LP |
2987 | u = m->cgroup_empty_queue; |
2988 | if (!u) | |
efdb0237 LP |
2989 | return 0; |
2990 | ||
09e24654 LP |
2991 | assert(u->in_cgroup_empty_queue); |
2992 | u->in_cgroup_empty_queue = false; | |
2993 | LIST_REMOVE(cgroup_empty_queue, m->cgroup_empty_queue, u); | |
2994 | ||
2995 | if (m->cgroup_empty_queue) { | |
2996 | /* More stuff queued, let's make sure we remain enabled */ | |
2997 | r = sd_event_source_set_enabled(s, SD_EVENT_ONESHOT); | |
2998 | if (r < 0) | |
19a691a9 | 2999 | log_debug_errno(r, "Failed to reenable cgroup empty event source, ignoring: %m"); |
09e24654 | 3000 | } |
efdb0237 | 3001 | |
f7829525 NK |
3002 | /* Update state based on OOM kills before we notify about cgroup empty event */ |
3003 | (void) unit_check_oom(u); | |
3004 | (void) unit_check_oomd_kill(u); | |
3005 | ||
efdb0237 LP |
3006 | unit_add_to_gc_queue(u); |
3007 | ||
3008 | if (UNIT_VTABLE(u)->notify_cgroup_empty) | |
3009 | UNIT_VTABLE(u)->notify_cgroup_empty(u); | |
3010 | ||
3011 | return 0; | |
3012 | } | |
3013 | ||
09e24654 LP |
3014 | void unit_add_to_cgroup_empty_queue(Unit *u) { |
3015 | int r; | |
3016 | ||
3017 | assert(u); | |
3018 | ||
3019 | /* Note that there are four different ways how cgroup empty events reach us: | |
3020 | * | |
3021 | * 1. On the unified hierarchy we get an inotify event on the cgroup | |
3022 | * | |
3023 | * 2. On the legacy hierarchy, when running in system mode, we get a datagram on the cgroup agent socket | |
3024 | * | |
3025 | * 3. On the legacy hierarchy, when running in user mode, we get a D-Bus signal on the system bus | |
3026 | * | |
3027 | * 4. On the legacy hierarchy, in service units we start watching all processes of the cgroup for SIGCHLD as | |
3028 | * soon as we get one SIGCHLD, to deal with unreliable cgroup notifications. | |
3029 | * | |
3030 | * Regardless which way we got the notification, we'll verify it here, and then add it to a separate | |
3031 | * queue. This queue will be dispatched at a lower priority than the SIGCHLD handler, so that we always use | |
3032 | * SIGCHLD if we can get it first, and only use the cgroup empty notifications if there's no SIGCHLD pending | |
3033 | * (which might happen if the cgroup doesn't contain processes that are our own child, which is typically the | |
3034 | * case for scope units). */ | |
3035 | ||
3036 | if (u->in_cgroup_empty_queue) | |
3037 | return; | |
3038 | ||
3039 | /* Let's verify that the cgroup is really empty */ | |
3040 | if (!u->cgroup_path) | |
3041 | return; | |
e1e98911 | 3042 | |
09e24654 LP |
3043 | r = cg_is_empty_recursive(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path); |
3044 | if (r < 0) { | |
6178e2f8 | 3045 | log_unit_debug_errno(u, r, "Failed to determine whether cgroup %s is empty: %m", empty_to_root(u->cgroup_path)); |
09e24654 LP |
3046 | return; |
3047 | } | |
3048 | if (r == 0) | |
3049 | return; | |
3050 | ||
3051 | LIST_PREPEND(cgroup_empty_queue, u->manager->cgroup_empty_queue, u); | |
3052 | u->in_cgroup_empty_queue = true; | |
3053 | ||
3054 | /* Trigger the defer event */ | |
3055 | r = sd_event_source_set_enabled(u->manager->cgroup_empty_event_source, SD_EVENT_ONESHOT); | |
3056 | if (r < 0) | |
3057 | log_debug_errno(r, "Failed to enable cgroup empty event source: %m"); | |
3058 | } | |
3059 | ||
d9e45bc3 MS |
3060 | static void unit_remove_from_cgroup_empty_queue(Unit *u) { |
3061 | assert(u); | |
3062 | ||
3063 | if (!u->in_cgroup_empty_queue) | |
3064 | return; | |
3065 | ||
3066 | LIST_REMOVE(cgroup_empty_queue, u->manager->cgroup_empty_queue, u); | |
3067 | u->in_cgroup_empty_queue = false; | |
3068 | } | |
3069 | ||
fe8d22fb AZ |
3070 | int unit_check_oomd_kill(Unit *u) { |
3071 | _cleanup_free_ char *value = NULL; | |
3072 | bool increased; | |
3073 | uint64_t n = 0; | |
3074 | int r; | |
3075 | ||
3076 | if (!u->cgroup_path) | |
3077 | return 0; | |
3078 | ||
3079 | r = cg_all_unified(); | |
3080 | if (r < 0) | |
3081 | return log_unit_debug_errno(u, r, "Couldn't determine whether we are in all unified mode: %m"); | |
3082 | else if (r == 0) | |
3083 | return 0; | |
3084 | ||
38c41427 | 3085 | r = cg_get_xattr_malloc(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path, "user.oomd_ooms", &value); |
fe8d22fb AZ |
3086 | if (r < 0 && r != -ENODATA) |
3087 | return r; | |
3088 | ||
3089 | if (!isempty(value)) { | |
3090 | r = safe_atou64(value, &n); | |
3091 | if (r < 0) | |
3092 | return r; | |
3093 | } | |
3094 | ||
3095 | increased = n > u->managed_oom_kill_last; | |
3096 | u->managed_oom_kill_last = n; | |
3097 | ||
3098 | if (!increased) | |
3099 | return 0; | |
3100 | ||
38c41427 NK |
3101 | n = 0; |
3102 | value = mfree(value); | |
3103 | r = cg_get_xattr_malloc(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path, "user.oomd_kill", &value); | |
3104 | if (r >= 0 && !isempty(value)) | |
3105 | (void) safe_atou64(value, &n); | |
3106 | ||
fe8d22fb | 3107 | if (n > 0) |
c2503e35 RH |
3108 | log_unit_struct(u, LOG_NOTICE, |
3109 | "MESSAGE_ID=" SD_MESSAGE_UNIT_OOMD_KILL_STR, | |
3110 | LOG_UNIT_INVOCATION_ID(u), | |
38c41427 NK |
3111 | LOG_UNIT_MESSAGE(u, "systemd-oomd killed %"PRIu64" process(es) in this unit.", n), |
3112 | "N_PROCESSES=%" PRIu64, n); | |
3113 | else | |
3114 | log_unit_struct(u, LOG_NOTICE, | |
3115 | "MESSAGE_ID=" SD_MESSAGE_UNIT_OOMD_KILL_STR, | |
3116 | LOG_UNIT_INVOCATION_ID(u), | |
3117 | LOG_UNIT_MESSAGE(u, "systemd-oomd killed some process(es) in this unit.")); | |
3118 | ||
3119 | unit_notify_cgroup_oom(u, /* ManagedOOM= */ true); | |
fe8d22fb AZ |
3120 | |
3121 | return 1; | |
3122 | } | |
3123 | ||
2ba6ae6b | 3124 | int unit_check_oom(Unit *u) { |
afcfaa69 LP |
3125 | _cleanup_free_ char *oom_kill = NULL; |
3126 | bool increased; | |
3127 | uint64_t c; | |
3128 | int r; | |
3129 | ||
3130 | if (!u->cgroup_path) | |
3131 | return 0; | |
3132 | ||
3133 | r = cg_get_keyed_attribute("memory", u->cgroup_path, "memory.events", STRV_MAKE("oom_kill"), &oom_kill); | |
fc594dee LP |
3134 | if (IN_SET(r, -ENOENT, -ENXIO)) /* Handle gracefully if cgroup or oom_kill attribute don't exist */ |
3135 | c = 0; | |
3136 | else if (r < 0) | |
afcfaa69 | 3137 | return log_unit_debug_errno(u, r, "Failed to read oom_kill field of memory.events cgroup attribute: %m"); |
fc594dee LP |
3138 | else { |
3139 | r = safe_atou64(oom_kill, &c); | |
3140 | if (r < 0) | |
3141 | return log_unit_debug_errno(u, r, "Failed to parse oom_kill field: %m"); | |
3142 | } | |
afcfaa69 LP |
3143 | |
3144 | increased = c > u->oom_kill_last; | |
3145 | u->oom_kill_last = c; | |
3146 | ||
3147 | if (!increased) | |
3148 | return 0; | |
3149 | ||
c2503e35 RH |
3150 | log_unit_struct(u, LOG_NOTICE, |
3151 | "MESSAGE_ID=" SD_MESSAGE_UNIT_OUT_OF_MEMORY_STR, | |
3152 | LOG_UNIT_INVOCATION_ID(u), | |
3153 | LOG_UNIT_MESSAGE(u, "A process of this unit has been killed by the OOM killer.")); | |
afcfaa69 | 3154 | |
38c41427 | 3155 | unit_notify_cgroup_oom(u, /* ManagedOOM= */ false); |
afcfaa69 LP |
3156 | |
3157 | return 1; | |
3158 | } | |
3159 | ||
3160 | static int on_cgroup_oom_event(sd_event_source *s, void *userdata) { | |
3161 | Manager *m = userdata; | |
3162 | Unit *u; | |
3163 | int r; | |
3164 | ||
3165 | assert(s); | |
3166 | assert(m); | |
3167 | ||
3168 | u = m->cgroup_oom_queue; | |
3169 | if (!u) | |
3170 | return 0; | |
3171 | ||
3172 | assert(u->in_cgroup_oom_queue); | |
3173 | u->in_cgroup_oom_queue = false; | |
3174 | LIST_REMOVE(cgroup_oom_queue, m->cgroup_oom_queue, u); | |
3175 | ||
3176 | if (m->cgroup_oom_queue) { | |
3177 | /* More stuff queued, let's make sure we remain enabled */ | |
3178 | r = sd_event_source_set_enabled(s, SD_EVENT_ONESHOT); | |
3179 | if (r < 0) | |
3180 | log_debug_errno(r, "Failed to reenable cgroup oom event source, ignoring: %m"); | |
3181 | } | |
3182 | ||
3183 | (void) unit_check_oom(u); | |
3184 | return 0; | |
3185 | } | |
3186 | ||
3187 | static void unit_add_to_cgroup_oom_queue(Unit *u) { | |
3188 | int r; | |
3189 | ||
3190 | assert(u); | |
3191 | ||
3192 | if (u->in_cgroup_oom_queue) | |
3193 | return; | |
3194 | if (!u->cgroup_path) | |
3195 | return; | |
3196 | ||
3197 | LIST_PREPEND(cgroup_oom_queue, u->manager->cgroup_oom_queue, u); | |
3198 | u->in_cgroup_oom_queue = true; | |
3199 | ||
3200 | /* Trigger the defer event */ | |
3201 | if (!u->manager->cgroup_oom_event_source) { | |
3202 | _cleanup_(sd_event_source_unrefp) sd_event_source *s = NULL; | |
3203 | ||
3204 | r = sd_event_add_defer(u->manager->event, &s, on_cgroup_oom_event, u->manager); | |
3205 | if (r < 0) { | |
3206 | log_error_errno(r, "Failed to create cgroup oom event source: %m"); | |
3207 | return; | |
3208 | } | |
3209 | ||
3210 | r = sd_event_source_set_priority(s, SD_EVENT_PRIORITY_NORMAL-8); | |
3211 | if (r < 0) { | |
3212 | log_error_errno(r, "Failed to set priority of cgroup oom event source: %m"); | |
3213 | return; | |
3214 | } | |
3215 | ||
3216 | (void) sd_event_source_set_description(s, "cgroup-oom"); | |
3217 | u->manager->cgroup_oom_event_source = TAKE_PTR(s); | |
3218 | } | |
3219 | ||
3220 | r = sd_event_source_set_enabled(u->manager->cgroup_oom_event_source, SD_EVENT_ONESHOT); | |
3221 | if (r < 0) | |
3222 | log_error_errno(r, "Failed to enable cgroup oom event source: %m"); | |
3223 | } | |
3224 | ||
d9e45bc3 MS |
3225 | static int unit_check_cgroup_events(Unit *u) { |
3226 | char *values[2] = {}; | |
3227 | int r; | |
3228 | ||
3229 | assert(u); | |
3230 | ||
869f52f2 DS |
3231 | if (!u->cgroup_path) |
3232 | return 0; | |
3233 | ||
d9e45bc3 MS |
3234 | r = cg_get_keyed_attribute_graceful(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path, "cgroup.events", |
3235 | STRV_MAKE("populated", "frozen"), values); | |
3236 | if (r < 0) | |
3237 | return r; | |
3238 | ||
3239 | /* The cgroup.events notifications can be merged together so act as we saw the given state for the | |
3240 | * first time. The functions we call to handle given state are idempotent, which makes them | |
3241 | * effectively remember the previous state. */ | |
3242 | if (values[0]) { | |
3243 | if (streq(values[0], "1")) | |
3244 | unit_remove_from_cgroup_empty_queue(u); | |
3245 | else | |
3246 | unit_add_to_cgroup_empty_queue(u); | |
3247 | } | |
3248 | ||
3249 | /* Disregard freezer state changes due to operations not initiated by us */ | |
3250 | if (values[1] && IN_SET(u->freezer_state, FREEZER_FREEZING, FREEZER_THAWING)) { | |
3251 | if (streq(values[1], "0")) | |
3252 | unit_thawed(u); | |
3253 | else | |
3254 | unit_frozen(u); | |
3255 | } | |
3256 | ||
3257 | free(values[0]); | |
3258 | free(values[1]); | |
3259 | ||
3260 | return 0; | |
3261 | } | |
3262 | ||
efdb0237 LP |
3263 | static int on_cgroup_inotify_event(sd_event_source *s, int fd, uint32_t revents, void *userdata) { |
3264 | Manager *m = userdata; | |
3265 | ||
3266 | assert(s); | |
3267 | assert(fd >= 0); | |
3268 | assert(m); | |
3269 | ||
3270 | for (;;) { | |
3271 | union inotify_event_buffer buffer; | |
efdb0237 LP |
3272 | ssize_t l; |
3273 | ||
3274 | l = read(fd, &buffer, sizeof(buffer)); | |
3275 | if (l < 0) { | |
8add30a0 | 3276 | if (ERRNO_IS_TRANSIENT(errno)) |
efdb0237 LP |
3277 | return 0; |
3278 | ||
3279 | return log_error_errno(errno, "Failed to read control group inotify events: %m"); | |
3280 | } | |
3281 | ||
00adc340 | 3282 | FOREACH_INOTIFY_EVENT_WARN(e, buffer, l) { |
efdb0237 LP |
3283 | Unit *u; |
3284 | ||
3285 | if (e->wd < 0) | |
3286 | /* Queue overflow has no watch descriptor */ | |
3287 | continue; | |
3288 | ||
3289 | if (e->mask & IN_IGNORED) | |
3290 | /* The watch was just removed */ | |
3291 | continue; | |
3292 | ||
afcfaa69 LP |
3293 | /* Note that inotify might deliver events for a watch even after it was removed, |
3294 | * because it was queued before the removal. Let's ignore this here safely. */ | |
3295 | ||
0bb814c2 | 3296 | u = hashmap_get(m->cgroup_control_inotify_wd_unit, INT_TO_PTR(e->wd)); |
afcfaa69 | 3297 | if (u) |
d9e45bc3 | 3298 | unit_check_cgroup_events(u); |
efdb0237 | 3299 | |
afcfaa69 LP |
3300 | u = hashmap_get(m->cgroup_memory_inotify_wd_unit, INT_TO_PTR(e->wd)); |
3301 | if (u) | |
3302 | unit_add_to_cgroup_oom_queue(u); | |
efdb0237 LP |
3303 | } |
3304 | } | |
8e274523 LP |
3305 | } |
3306 | ||
17f14955 RG |
3307 | static int cg_bpf_mask_supported(CGroupMask *ret) { |
3308 | CGroupMask mask = 0; | |
3309 | int r; | |
3310 | ||
3311 | /* BPF-based firewall */ | |
3312 | r = bpf_firewall_supported(); | |
ad13559e YW |
3313 | if (r < 0) |
3314 | return r; | |
17f14955 RG |
3315 | if (r > 0) |
3316 | mask |= CGROUP_MASK_BPF_FIREWALL; | |
3317 | ||
084c7007 RG |
3318 | /* BPF-based device access control */ |
3319 | r = bpf_devices_supported(); | |
ad13559e YW |
3320 | if (r < 0) |
3321 | return r; | |
084c7007 RG |
3322 | if (r > 0) |
3323 | mask |= CGROUP_MASK_BPF_DEVICES; | |
3324 | ||
506ea51b JK |
3325 | /* BPF pinned prog */ |
3326 | r = bpf_foreign_supported(); | |
ad13559e YW |
3327 | if (r < 0) |
3328 | return r; | |
506ea51b JK |
3329 | if (r > 0) |
3330 | mask |= CGROUP_MASK_BPF_FOREIGN; | |
3331 | ||
a8e5eb17 | 3332 | /* BPF-based bind{4|6} hooks */ |
cd09a5f3 | 3333 | r = bpf_socket_bind_supported(); |
ad13559e YW |
3334 | if (r < 0) |
3335 | return r; | |
a8e5eb17 JK |
3336 | if (r > 0) |
3337 | mask |= CGROUP_MASK_BPF_SOCKET_BIND; | |
3338 | ||
6f50d4f7 MV |
3339 | /* BPF-based cgroup_skb/{egress|ingress} hooks */ |
3340 | r = restrict_network_interfaces_supported(); | |
ad13559e YW |
3341 | if (r < 0) |
3342 | return r; | |
6f50d4f7 MV |
3343 | if (r > 0) |
3344 | mask |= CGROUP_MASK_BPF_RESTRICT_NETWORK_INTERFACES; | |
3345 | ||
17f14955 RG |
3346 | *ret = mask; |
3347 | return 0; | |
3348 | } | |
3349 | ||
8e274523 | 3350 | int manager_setup_cgroup(Manager *m) { |
9444b1f2 | 3351 | _cleanup_free_ char *path = NULL; |
10bd3e2e | 3352 | const char *scope_path; |
b4cccbc1 | 3353 | int r, all_unified; |
17f14955 | 3354 | CGroupMask mask; |
efdb0237 | 3355 | char *e; |
8e274523 LP |
3356 | |
3357 | assert(m); | |
3358 | ||
35d2e7ec | 3359 | /* 1. Determine hierarchy */ |
efdb0237 | 3360 | m->cgroup_root = mfree(m->cgroup_root); |
9444b1f2 | 3361 | r = cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER, 0, &m->cgroup_root); |
23bbb0de MS |
3362 | if (r < 0) |
3363 | return log_error_errno(r, "Cannot determine cgroup we are running in: %m"); | |
8e274523 | 3364 | |
efdb0237 LP |
3365 | /* Chop off the init scope, if we are already located in it */ |
3366 | e = endswith(m->cgroup_root, "/" SPECIAL_INIT_SCOPE); | |
0d8c31ff | 3367 | |
efdb0237 LP |
3368 | /* LEGACY: Also chop off the system slice if we are in |
3369 | * it. This is to support live upgrades from older systemd | |
3370 | * versions where PID 1 was moved there. Also see | |
3371 | * cg_get_root_path(). */ | |
463d0d15 | 3372 | if (!e && MANAGER_IS_SYSTEM(m)) { |
9444b1f2 | 3373 | e = endswith(m->cgroup_root, "/" SPECIAL_SYSTEM_SLICE); |
15c60e99 | 3374 | if (!e) |
efdb0237 | 3375 | e = endswith(m->cgroup_root, "/system"); /* even more legacy */ |
0baf24dd | 3376 | } |
efdb0237 LP |
3377 | if (e) |
3378 | *e = 0; | |
7ccfb64a | 3379 | |
7546145e LP |
3380 | /* And make sure to store away the root value without trailing slash, even for the root dir, so that we can |
3381 | * easily prepend it everywhere. */ | |
3382 | delete_trailing_chars(m->cgroup_root, "/"); | |
8e274523 | 3383 | |
35d2e7ec | 3384 | /* 2. Show data */ |
9444b1f2 | 3385 | r = cg_get_path(SYSTEMD_CGROUP_CONTROLLER, m->cgroup_root, NULL, &path); |
23bbb0de MS |
3386 | if (r < 0) |
3387 | return log_error_errno(r, "Cannot find cgroup mount point: %m"); | |
8e274523 | 3388 | |
d4d99bc6 | 3389 | r = cg_unified(); |
415fc41c TH |
3390 | if (r < 0) |
3391 | return log_error_errno(r, "Couldn't determine if we are running in the unified hierarchy: %m"); | |
5da38d07 | 3392 | |
b4cccbc1 | 3393 | all_unified = cg_all_unified(); |
d4c819ed ZJS |
3394 | if (all_unified < 0) |
3395 | return log_error_errno(all_unified, "Couldn't determine whether we are in all unified mode: %m"); | |
3396 | if (all_unified > 0) | |
efdb0237 | 3397 | log_debug("Unified cgroup hierarchy is located at %s.", path); |
b4cccbc1 | 3398 | else { |
c22800e4 | 3399 | r = cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER); |
b4cccbc1 LP |
3400 | if (r < 0) |
3401 | return log_error_errno(r, "Failed to determine whether systemd's own controller is in unified mode: %m"); | |
3402 | if (r > 0) | |
3403 | log_debug("Unified cgroup hierarchy is located at %s. Controllers are on legacy hierarchies.", path); | |
3404 | else | |
3405 | log_debug("Using cgroup controller " SYSTEMD_CGROUP_CONTROLLER_LEGACY ". File system hierarchy is at %s.", path); | |
3406 | } | |
efdb0237 | 3407 | |
09e24654 | 3408 | /* 3. Allocate cgroup empty defer event source */ |
5dcadb4c | 3409 | m->cgroup_empty_event_source = sd_event_source_disable_unref(m->cgroup_empty_event_source); |
09e24654 LP |
3410 | r = sd_event_add_defer(m->event, &m->cgroup_empty_event_source, on_cgroup_empty_event, m); |
3411 | if (r < 0) | |
3412 | return log_error_errno(r, "Failed to create cgroup empty event source: %m"); | |
3413 | ||
cbe83389 LP |
3414 | /* Schedule cgroup empty checks early, but after having processed service notification messages or |
3415 | * SIGCHLD signals, so that a cgroup running empty is always just the last safety net of | |
3416 | * notification, and we collected the metadata the notification and SIGCHLD stuff offers first. */ | |
09e24654 LP |
3417 | r = sd_event_source_set_priority(m->cgroup_empty_event_source, SD_EVENT_PRIORITY_NORMAL-5); |
3418 | if (r < 0) | |
3419 | return log_error_errno(r, "Failed to set priority of cgroup empty event source: %m"); | |
3420 | ||
3421 | r = sd_event_source_set_enabled(m->cgroup_empty_event_source, SD_EVENT_OFF); | |
3422 | if (r < 0) | |
3423 | return log_error_errno(r, "Failed to disable cgroup empty event source: %m"); | |
3424 | ||
3425 | (void) sd_event_source_set_description(m->cgroup_empty_event_source, "cgroup-empty"); | |
3426 | ||
3427 | /* 4. Install notifier inotify object, or agent */ | |
10bd3e2e | 3428 | if (cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER) > 0) { |
c6c18be3 | 3429 | |
09e24654 | 3430 | /* In the unified hierarchy we can get cgroup empty notifications via inotify. */ |
efdb0237 | 3431 | |
5dcadb4c | 3432 | m->cgroup_inotify_event_source = sd_event_source_disable_unref(m->cgroup_inotify_event_source); |
10bd3e2e | 3433 | safe_close(m->cgroup_inotify_fd); |
efdb0237 | 3434 | |
10bd3e2e LP |
3435 | m->cgroup_inotify_fd = inotify_init1(IN_NONBLOCK|IN_CLOEXEC); |
3436 | if (m->cgroup_inotify_fd < 0) | |
3437 | return log_error_errno(errno, "Failed to create control group inotify object: %m"); | |
efdb0237 | 3438 | |
10bd3e2e LP |
3439 | r = sd_event_add_io(m->event, &m->cgroup_inotify_event_source, m->cgroup_inotify_fd, EPOLLIN, on_cgroup_inotify_event, m); |
3440 | if (r < 0) | |
3441 | return log_error_errno(r, "Failed to watch control group inotify object: %m"); | |
efdb0237 | 3442 | |
cbe83389 LP |
3443 | /* Process cgroup empty notifications early. Note that when this event is dispatched it'll |
3444 | * just add the unit to a cgroup empty queue, hence let's run earlier than that. Also see | |
3445 | * handling of cgroup agent notifications, for the classic cgroup hierarchy support. */ | |
3446 | r = sd_event_source_set_priority(m->cgroup_inotify_event_source, SD_EVENT_PRIORITY_NORMAL-9); | |
10bd3e2e LP |
3447 | if (r < 0) |
3448 | return log_error_errno(r, "Failed to set priority of inotify event source: %m"); | |
efdb0237 | 3449 | |
10bd3e2e | 3450 | (void) sd_event_source_set_description(m->cgroup_inotify_event_source, "cgroup-inotify"); |
efdb0237 | 3451 | |
611c4f8a | 3452 | } else if (MANAGER_IS_SYSTEM(m) && manager_owns_host_root_cgroup(m) && !MANAGER_IS_TEST_RUN(m)) { |
efdb0237 | 3453 | |
10bd3e2e LP |
3454 | /* On the legacy hierarchy we only get notifications via cgroup agents. (Which isn't really reliable, |
3455 | * since it does not generate events when control groups with children run empty. */ | |
8e274523 | 3456 | |
ce906769 | 3457 | r = cg_install_release_agent(SYSTEMD_CGROUP_CONTROLLER, SYSTEMD_CGROUPS_AGENT_PATH); |
23bbb0de | 3458 | if (r < 0) |
10bd3e2e LP |
3459 | log_warning_errno(r, "Failed to install release agent, ignoring: %m"); |
3460 | else if (r > 0) | |
3461 | log_debug("Installed release agent."); | |
3462 | else if (r == 0) | |
3463 | log_debug("Release agent already installed."); | |
3464 | } | |
efdb0237 | 3465 | |
09e24654 | 3466 | /* 5. Make sure we are in the special "init.scope" unit in the root slice. */ |
10bd3e2e LP |
3467 | scope_path = strjoina(m->cgroup_root, "/" SPECIAL_INIT_SCOPE); |
3468 | r = cg_create_and_attach(SYSTEMD_CGROUP_CONTROLLER, scope_path, 0); | |
aa77e234 MS |
3469 | if (r >= 0) { |
3470 | /* Also, move all other userspace processes remaining in the root cgroup into that scope. */ | |
3471 | r = cg_migrate(SYSTEMD_CGROUP_CONTROLLER, m->cgroup_root, SYSTEMD_CGROUP_CONTROLLER, scope_path, 0); | |
3472 | if (r < 0) | |
3473 | log_warning_errno(r, "Couldn't move remaining userspace processes, ignoring: %m"); | |
c6c18be3 | 3474 | |
aa77e234 MS |
3475 | /* 6. And pin it, so that it cannot be unmounted */ |
3476 | safe_close(m->pin_cgroupfs_fd); | |
3477 | m->pin_cgroupfs_fd = open(path, O_RDONLY|O_CLOEXEC|O_DIRECTORY|O_NOCTTY|O_NONBLOCK); | |
3478 | if (m->pin_cgroupfs_fd < 0) | |
3479 | return log_error_errno(errno, "Failed to open pin file: %m"); | |
0d8c31ff | 3480 | |
638cece4 | 3481 | } else if (!MANAGER_IS_TEST_RUN(m)) |
aa77e234 | 3482 | return log_error_errno(r, "Failed to create %s control group: %m", scope_path); |
10bd3e2e | 3483 | |
09e24654 | 3484 | /* 7. Always enable hierarchical support if it exists... */ |
638cece4 | 3485 | if (!all_unified && !MANAGER_IS_TEST_RUN(m)) |
10bd3e2e | 3486 | (void) cg_set_attribute("memory", "/", "memory.use_hierarchy", "1"); |
c6c18be3 | 3487 | |
17f14955 | 3488 | /* 8. Figure out which controllers are supported */ |
0fa7b500 | 3489 | r = cg_mask_supported_subtree(m->cgroup_root, &m->cgroup_supported); |
efdb0237 LP |
3490 | if (r < 0) |
3491 | return log_error_errno(r, "Failed to determine supported controllers: %m"); | |
17f14955 RG |
3492 | |
3493 | /* 9. Figure out which bpf-based pseudo-controllers are supported */ | |
3494 | r = cg_bpf_mask_supported(&mask); | |
3495 | if (r < 0) | |
3496 | return log_error_errno(r, "Failed to determine supported bpf-based pseudo-controllers: %m"); | |
3497 | m->cgroup_supported |= mask; | |
3498 | ||
3499 | /* 10. Log which controllers are supported */ | |
e8616626 ZJS |
3500 | for (CGroupController c = 0; c < _CGROUP_CONTROLLER_MAX; c++) |
3501 | log_debug("Controller '%s' supported: %s", cgroup_controller_to_string(c), | |
3502 | yes_no(m->cgroup_supported & CGROUP_CONTROLLER_TO_MASK(c))); | |
9156e799 | 3503 | |
a32360f1 | 3504 | return 0; |
8e274523 LP |
3505 | } |
3506 | ||
c6c18be3 | 3507 | void manager_shutdown_cgroup(Manager *m, bool delete) { |
8e274523 LP |
3508 | assert(m); |
3509 | ||
9444b1f2 LP |
3510 | /* We can't really delete the group, since we are in it. But |
3511 | * let's trim it. */ | |
5dd2f5ff | 3512 | if (delete && m->cgroup_root && !FLAGS_SET(m->test_run_flags, MANAGER_TEST_RUN_MINIMAL)) |
efdb0237 LP |
3513 | (void) cg_trim(SYSTEMD_CGROUP_CONTROLLER, m->cgroup_root, false); |
3514 | ||
5dcadb4c | 3515 | m->cgroup_empty_event_source = sd_event_source_disable_unref(m->cgroup_empty_event_source); |
09e24654 | 3516 | |
0bb814c2 | 3517 | m->cgroup_control_inotify_wd_unit = hashmap_free(m->cgroup_control_inotify_wd_unit); |
afcfaa69 | 3518 | m->cgroup_memory_inotify_wd_unit = hashmap_free(m->cgroup_memory_inotify_wd_unit); |
efdb0237 | 3519 | |
5dcadb4c | 3520 | m->cgroup_inotify_event_source = sd_event_source_disable_unref(m->cgroup_inotify_event_source); |
efdb0237 | 3521 | m->cgroup_inotify_fd = safe_close(m->cgroup_inotify_fd); |
8e274523 | 3522 | |
03e334a1 | 3523 | m->pin_cgroupfs_fd = safe_close(m->pin_cgroupfs_fd); |
c6c18be3 | 3524 | |
efdb0237 | 3525 | m->cgroup_root = mfree(m->cgroup_root); |
8e274523 LP |
3526 | } |
3527 | ||
4ad49000 | 3528 | Unit* manager_get_unit_by_cgroup(Manager *m, const char *cgroup) { |
acb14d31 | 3529 | char *p; |
4ad49000 | 3530 | Unit *u; |
acb14d31 LP |
3531 | |
3532 | assert(m); | |
3533 | assert(cgroup); | |
acb14d31 | 3534 | |
4ad49000 LP |
3535 | u = hashmap_get(m->cgroup_unit, cgroup); |
3536 | if (u) | |
3537 | return u; | |
acb14d31 | 3538 | |
2f82562b | 3539 | p = strdupa_safe(cgroup); |
acb14d31 LP |
3540 | for (;;) { |
3541 | char *e; | |
3542 | ||
3543 | e = strrchr(p, '/'); | |
efdb0237 LP |
3544 | if (!e || e == p) |
3545 | return hashmap_get(m->cgroup_unit, SPECIAL_ROOT_SLICE); | |
acb14d31 LP |
3546 | |
3547 | *e = 0; | |
3548 | ||
4ad49000 LP |
3549 | u = hashmap_get(m->cgroup_unit, p); |
3550 | if (u) | |
3551 | return u; | |
acb14d31 LP |
3552 | } |
3553 | } | |
3554 | ||
b3ac818b | 3555 | Unit *manager_get_unit_by_pid_cgroup(Manager *m, pid_t pid) { |
4ad49000 | 3556 | _cleanup_free_ char *cgroup = NULL; |
8e274523 | 3557 | |
8c47c732 LP |
3558 | assert(m); |
3559 | ||
62a76913 | 3560 | if (!pid_is_valid(pid)) |
b3ac818b LP |
3561 | return NULL; |
3562 | ||
62a76913 | 3563 | if (cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER, pid, &cgroup) < 0) |
b3ac818b LP |
3564 | return NULL; |
3565 | ||
3566 | return manager_get_unit_by_cgroup(m, cgroup); | |
3567 | } | |
3568 | ||
3569 | Unit *manager_get_unit_by_pid(Manager *m, pid_t pid) { | |
62a76913 | 3570 | Unit *u, **array; |
b3ac818b LP |
3571 | |
3572 | assert(m); | |
3573 | ||
62a76913 LP |
3574 | /* Note that a process might be owned by multiple units, we return only one here, which is good enough for most |
3575 | * cases, though not strictly correct. We prefer the one reported by cgroup membership, as that's the most | |
3576 | * relevant one as children of the process will be assigned to that one, too, before all else. */ | |
3577 | ||
3578 | if (!pid_is_valid(pid)) | |
8c47c732 LP |
3579 | return NULL; |
3580 | ||
2ca9d979 | 3581 | if (pid == getpid_cached()) |
efdb0237 LP |
3582 | return hashmap_get(m->units, SPECIAL_INIT_SCOPE); |
3583 | ||
62a76913 | 3584 | u = manager_get_unit_by_pid_cgroup(m, pid); |
5fe8876b LP |
3585 | if (u) |
3586 | return u; | |
3587 | ||
62a76913 | 3588 | u = hashmap_get(m->watch_pids, PID_TO_PTR(pid)); |
5fe8876b LP |
3589 | if (u) |
3590 | return u; | |
3591 | ||
62a76913 LP |
3592 | array = hashmap_get(m->watch_pids, PID_TO_PTR(-pid)); |
3593 | if (array) | |
3594 | return array[0]; | |
3595 | ||
3596 | return NULL; | |
6dde1f33 | 3597 | } |
4fbf50b3 | 3598 | |
4ad49000 LP |
3599 | int manager_notify_cgroup_empty(Manager *m, const char *cgroup) { |
3600 | Unit *u; | |
4fbf50b3 | 3601 | |
4ad49000 LP |
3602 | assert(m); |
3603 | assert(cgroup); | |
4fbf50b3 | 3604 | |
09e24654 LP |
3605 | /* Called on the legacy hierarchy whenever we get an explicit cgroup notification from the cgroup agent process |
3606 | * or from the --system instance */ | |
3607 | ||
d8fdc620 LP |
3608 | log_debug("Got cgroup empty notification for: %s", cgroup); |
3609 | ||
4ad49000 | 3610 | u = manager_get_unit_by_cgroup(m, cgroup); |
5ad096b3 LP |
3611 | if (!u) |
3612 | return 0; | |
b56c28c3 | 3613 | |
09e24654 LP |
3614 | unit_add_to_cgroup_empty_queue(u); |
3615 | return 1; | |
5ad096b3 LP |
3616 | } |
3617 | ||
93ff34e4 LB |
3618 | int unit_get_memory_available(Unit *u, uint64_t *ret) { |
3619 | uint64_t unit_current, available = UINT64_MAX; | |
3620 | CGroupContext *unit_context; | |
3621 | const char *memory_file; | |
3622 | int r; | |
3623 | ||
3624 | assert(u); | |
3625 | assert(ret); | |
3626 | ||
3627 | /* If data from cgroups can be accessed, try to find out how much more memory a unit can | |
3628 | * claim before hitting the configured cgroup limits (if any). Consider both MemoryHigh | |
3629 | * and MemoryMax, and also any slice the unit might be nested below. */ | |
3630 | ||
3631 | if (!UNIT_CGROUP_BOOL(u, memory_accounting)) | |
3632 | return -ENODATA; | |
3633 | ||
3634 | if (!u->cgroup_path) | |
3635 | return -ENODATA; | |
3636 | ||
3637 | /* The root cgroup doesn't expose this information */ | |
3638 | if (unit_has_host_root_cgroup(u)) | |
3639 | return -ENODATA; | |
3640 | ||
3641 | if ((u->cgroup_realized_mask & CGROUP_MASK_MEMORY) == 0) | |
3642 | return -ENODATA; | |
3643 | ||
3644 | r = cg_all_unified(); | |
3645 | if (r < 0) | |
3646 | return r; | |
3647 | memory_file = r > 0 ? "memory.current" : "memory.usage_in_bytes"; | |
3648 | ||
3649 | r = cg_get_attribute_as_uint64("memory", u->cgroup_path, memory_file, &unit_current); | |
3650 | if (r < 0) | |
3651 | return r; | |
3652 | ||
3653 | assert_se(unit_context = unit_get_cgroup_context(u)); | |
3654 | ||
3655 | if (unit_context->memory_max != UINT64_MAX || unit_context->memory_high != UINT64_MAX) | |
3656 | available = LESS_BY(MIN(unit_context->memory_max, unit_context->memory_high), unit_current); | |
3657 | ||
3658 | for (Unit *slice = UNIT_GET_SLICE(u); slice; slice = UNIT_GET_SLICE(slice)) { | |
3659 | uint64_t slice_current, slice_available = UINT64_MAX; | |
3660 | CGroupContext *slice_context; | |
3661 | ||
3662 | /* No point in continuing if we can't go any lower */ | |
3663 | if (available == 0) | |
3664 | break; | |
3665 | ||
3666 | if (!slice->cgroup_path) | |
3667 | continue; | |
3668 | ||
3669 | slice_context = unit_get_cgroup_context(slice); | |
3670 | if (!slice_context) | |
3671 | continue; | |
3672 | ||
3673 | if (slice_context->memory_max == UINT64_MAX && slice_context->memory_high == UINT64_MAX) | |
3674 | continue; | |
3675 | ||
3676 | r = cg_get_attribute_as_uint64("memory", slice->cgroup_path, memory_file, &slice_current); | |
3677 | if (r < 0) | |
3678 | continue; | |
3679 | ||
3680 | slice_available = LESS_BY(MIN(slice_context->memory_max, slice_context->memory_high), slice_current); | |
3681 | available = MIN(slice_available, available); | |
3682 | } | |
3683 | ||
3684 | *ret = available; | |
3685 | ||
3686 | return 0; | |
3687 | } | |
3688 | ||
5ad096b3 | 3689 | int unit_get_memory_current(Unit *u, uint64_t *ret) { |
5ad096b3 LP |
3690 | int r; |
3691 | ||
3692 | assert(u); | |
3693 | assert(ret); | |
3694 | ||
2e4025c0 | 3695 | if (!UNIT_CGROUP_BOOL(u, memory_accounting)) |
cf3b4be1 LP |
3696 | return -ENODATA; |
3697 | ||
5ad096b3 LP |
3698 | if (!u->cgroup_path) |
3699 | return -ENODATA; | |
3700 | ||
1f73aa00 | 3701 | /* The root cgroup doesn't expose this information, let's get it from /proc instead */ |
611c4f8a | 3702 | if (unit_has_host_root_cgroup(u)) |
c482724a | 3703 | return procfs_memory_get_used(ret); |
1f73aa00 | 3704 | |
efdb0237 | 3705 | if ((u->cgroup_realized_mask & CGROUP_MASK_MEMORY) == 0) |
5ad096b3 LP |
3706 | return -ENODATA; |
3707 | ||
b4cccbc1 LP |
3708 | r = cg_all_unified(); |
3709 | if (r < 0) | |
3710 | return r; | |
5ad096b3 | 3711 | |
613328c3 | 3712 | return cg_get_attribute_as_uint64("memory", u->cgroup_path, r > 0 ? "memory.current" : "memory.usage_in_bytes", ret); |
5ad096b3 LP |
3713 | } |
3714 | ||
03a7b521 | 3715 | int unit_get_tasks_current(Unit *u, uint64_t *ret) { |
03a7b521 LP |
3716 | assert(u); |
3717 | assert(ret); | |
3718 | ||
2e4025c0 | 3719 | if (!UNIT_CGROUP_BOOL(u, tasks_accounting)) |
cf3b4be1 LP |
3720 | return -ENODATA; |
3721 | ||
03a7b521 LP |
3722 | if (!u->cgroup_path) |
3723 | return -ENODATA; | |
3724 | ||
c36a69f4 | 3725 | /* The root cgroup doesn't expose this information, let's get it from /proc instead */ |
611c4f8a | 3726 | if (unit_has_host_root_cgroup(u)) |
c36a69f4 LP |
3727 | return procfs_tasks_get_current(ret); |
3728 | ||
1f73aa00 LP |
3729 | if ((u->cgroup_realized_mask & CGROUP_MASK_PIDS) == 0) |
3730 | return -ENODATA; | |
3731 | ||
613328c3 | 3732 | return cg_get_attribute_as_uint64("pids", u->cgroup_path, "pids.current", ret); |
03a7b521 LP |
3733 | } |
3734 | ||
5ad096b3 | 3735 | static int unit_get_cpu_usage_raw(Unit *u, nsec_t *ret) { |
5ad096b3 LP |
3736 | uint64_t ns; |
3737 | int r; | |
3738 | ||
3739 | assert(u); | |
3740 | assert(ret); | |
3741 | ||
3742 | if (!u->cgroup_path) | |
3743 | return -ENODATA; | |
3744 | ||
1f73aa00 | 3745 | /* The root cgroup doesn't expose this information, let's get it from /proc instead */ |
611c4f8a | 3746 | if (unit_has_host_root_cgroup(u)) |
1f73aa00 LP |
3747 | return procfs_cpu_get_usage(ret); |
3748 | ||
f98c2585 CD |
3749 | /* Requisite controllers for CPU accounting are not enabled */ |
3750 | if ((get_cpu_accounting_mask() & ~u->cgroup_realized_mask) != 0) | |
3751 | return -ENODATA; | |
3752 | ||
92a99304 LP |
3753 | r = cg_all_unified(); |
3754 | if (r < 0) | |
3755 | return r; | |
b4cccbc1 | 3756 | if (r > 0) { |
66ebf6c0 TH |
3757 | _cleanup_free_ char *val = NULL; |
3758 | uint64_t us; | |
5ad096b3 | 3759 | |
b734a4ff | 3760 | r = cg_get_keyed_attribute("cpu", u->cgroup_path, "cpu.stat", STRV_MAKE("usage_usec"), &val); |
b734a4ff LP |
3761 | if (IN_SET(r, -ENOENT, -ENXIO)) |
3762 | return -ENODATA; | |
d742f4b5 LP |
3763 | if (r < 0) |
3764 | return r; | |
66ebf6c0 TH |
3765 | |
3766 | r = safe_atou64(val, &us); | |
3767 | if (r < 0) | |
3768 | return r; | |
3769 | ||
3770 | ns = us * NSEC_PER_USEC; | |
613328c3 AZ |
3771 | } else |
3772 | return cg_get_attribute_as_uint64("cpuacct", u->cgroup_path, "cpuacct.usage", ret); | |
5ad096b3 LP |
3773 | |
3774 | *ret = ns; | |
3775 | return 0; | |
3776 | } | |
3777 | ||
3778 | int unit_get_cpu_usage(Unit *u, nsec_t *ret) { | |
3779 | nsec_t ns; | |
3780 | int r; | |
3781 | ||
fe700f46 LP |
3782 | assert(u); |
3783 | ||
3784 | /* Retrieve the current CPU usage counter. This will subtract the CPU counter taken when the unit was | |
3785 | * started. If the cgroup has been removed already, returns the last cached value. To cache the value, simply | |
3786 | * call this function with a NULL return value. */ | |
3787 | ||
2e4025c0 | 3788 | if (!UNIT_CGROUP_BOOL(u, cpu_accounting)) |
cf3b4be1 LP |
3789 | return -ENODATA; |
3790 | ||
5ad096b3 | 3791 | r = unit_get_cpu_usage_raw(u, &ns); |
fe700f46 LP |
3792 | if (r == -ENODATA && u->cpu_usage_last != NSEC_INFINITY) { |
3793 | /* If we can't get the CPU usage anymore (because the cgroup was already removed, for example), use our | |
3794 | * cached value. */ | |
3795 | ||
3796 | if (ret) | |
3797 | *ret = u->cpu_usage_last; | |
3798 | return 0; | |
3799 | } | |
5ad096b3 LP |
3800 | if (r < 0) |
3801 | return r; | |
3802 | ||
66ebf6c0 TH |
3803 | if (ns > u->cpu_usage_base) |
3804 | ns -= u->cpu_usage_base; | |
5ad096b3 LP |
3805 | else |
3806 | ns = 0; | |
3807 | ||
fe700f46 LP |
3808 | u->cpu_usage_last = ns; |
3809 | if (ret) | |
3810 | *ret = ns; | |
3811 | ||
5ad096b3 LP |
3812 | return 0; |
3813 | } | |
3814 | ||
906c06f6 DM |
3815 | int unit_get_ip_accounting( |
3816 | Unit *u, | |
3817 | CGroupIPAccountingMetric metric, | |
3818 | uint64_t *ret) { | |
3819 | ||
6b659ed8 | 3820 | uint64_t value; |
906c06f6 DM |
3821 | int fd, r; |
3822 | ||
3823 | assert(u); | |
3824 | assert(metric >= 0); | |
3825 | assert(metric < _CGROUP_IP_ACCOUNTING_METRIC_MAX); | |
3826 | assert(ret); | |
3827 | ||
2e4025c0 | 3828 | if (!UNIT_CGROUP_BOOL(u, ip_accounting)) |
cf3b4be1 LP |
3829 | return -ENODATA; |
3830 | ||
906c06f6 DM |
3831 | fd = IN_SET(metric, CGROUP_IP_INGRESS_BYTES, CGROUP_IP_INGRESS_PACKETS) ? |
3832 | u->ip_accounting_ingress_map_fd : | |
3833 | u->ip_accounting_egress_map_fd; | |
906c06f6 DM |
3834 | if (fd < 0) |
3835 | return -ENODATA; | |
3836 | ||
3837 | if (IN_SET(metric, CGROUP_IP_INGRESS_BYTES, CGROUP_IP_EGRESS_BYTES)) | |
6b659ed8 | 3838 | r = bpf_firewall_read_accounting(fd, &value, NULL); |
906c06f6 | 3839 | else |
6b659ed8 LP |
3840 | r = bpf_firewall_read_accounting(fd, NULL, &value); |
3841 | if (r < 0) | |
3842 | return r; | |
3843 | ||
3844 | /* Add in additional metrics from a previous runtime. Note that when reexecing/reloading the daemon we compile | |
3845 | * all BPF programs and maps anew, but serialize the old counters. When deserializing we store them in the | |
3846 | * ip_accounting_extra[] field, and add them in here transparently. */ | |
3847 | ||
3848 | *ret = value + u->ip_accounting_extra[metric]; | |
906c06f6 DM |
3849 | |
3850 | return r; | |
3851 | } | |
3852 | ||
fbe14fc9 LP |
3853 | static int unit_get_io_accounting_raw(Unit *u, uint64_t ret[static _CGROUP_IO_ACCOUNTING_METRIC_MAX]) { |
3854 | static const char *const field_names[_CGROUP_IO_ACCOUNTING_METRIC_MAX] = { | |
3855 | [CGROUP_IO_READ_BYTES] = "rbytes=", | |
3856 | [CGROUP_IO_WRITE_BYTES] = "wbytes=", | |
3857 | [CGROUP_IO_READ_OPERATIONS] = "rios=", | |
3858 | [CGROUP_IO_WRITE_OPERATIONS] = "wios=", | |
3859 | }; | |
3860 | uint64_t acc[_CGROUP_IO_ACCOUNTING_METRIC_MAX] = {}; | |
3861 | _cleanup_free_ char *path = NULL; | |
3862 | _cleanup_fclose_ FILE *f = NULL; | |
3863 | int r; | |
3864 | ||
3865 | assert(u); | |
3866 | ||
3867 | if (!u->cgroup_path) | |
3868 | return -ENODATA; | |
3869 | ||
3870 | if (unit_has_host_root_cgroup(u)) | |
3871 | return -ENODATA; /* TODO: return useful data for the top-level cgroup */ | |
3872 | ||
3873 | r = cg_all_unified(); | |
3874 | if (r < 0) | |
3875 | return r; | |
3876 | if (r == 0) /* TODO: support cgroupv1 */ | |
3877 | return -ENODATA; | |
3878 | ||
3879 | if (!FLAGS_SET(u->cgroup_realized_mask, CGROUP_MASK_IO)) | |
3880 | return -ENODATA; | |
3881 | ||
3882 | r = cg_get_path("io", u->cgroup_path, "io.stat", &path); | |
3883 | if (r < 0) | |
3884 | return r; | |
3885 | ||
3886 | f = fopen(path, "re"); | |
3887 | if (!f) | |
3888 | return -errno; | |
3889 | ||
3890 | for (;;) { | |
3891 | _cleanup_free_ char *line = NULL; | |
3892 | const char *p; | |
3893 | ||
3894 | r = read_line(f, LONG_LINE_MAX, &line); | |
3895 | if (r < 0) | |
3896 | return r; | |
3897 | if (r == 0) | |
3898 | break; | |
3899 | ||
3900 | p = line; | |
3901 | p += strcspn(p, WHITESPACE); /* Skip over device major/minor */ | |
3902 | p += strspn(p, WHITESPACE); /* Skip over following whitespace */ | |
3903 | ||
3904 | for (;;) { | |
3905 | _cleanup_free_ char *word = NULL; | |
3906 | ||
3907 | r = extract_first_word(&p, &word, NULL, EXTRACT_RETAIN_ESCAPE); | |
3908 | if (r < 0) | |
3909 | return r; | |
3910 | if (r == 0) | |
3911 | break; | |
3912 | ||
3913 | for (CGroupIOAccountingMetric i = 0; i < _CGROUP_IO_ACCOUNTING_METRIC_MAX; i++) { | |
3914 | const char *x; | |
3915 | ||
3916 | x = startswith(word, field_names[i]); | |
3917 | if (x) { | |
3918 | uint64_t w; | |
3919 | ||
3920 | r = safe_atou64(x, &w); | |
3921 | if (r < 0) | |
3922 | return r; | |
3923 | ||
3924 | /* Sum up the stats of all devices */ | |
3925 | acc[i] += w; | |
3926 | break; | |
3927 | } | |
3928 | } | |
3929 | } | |
3930 | } | |
3931 | ||
3932 | memcpy(ret, acc, sizeof(acc)); | |
3933 | return 0; | |
3934 | } | |
3935 | ||
3936 | int unit_get_io_accounting( | |
3937 | Unit *u, | |
3938 | CGroupIOAccountingMetric metric, | |
3939 | bool allow_cache, | |
3940 | uint64_t *ret) { | |
3941 | ||
3942 | uint64_t raw[_CGROUP_IO_ACCOUNTING_METRIC_MAX]; | |
3943 | int r; | |
3944 | ||
3945 | /* Retrieve an IO account parameter. This will subtract the counter when the unit was started. */ | |
3946 | ||
3947 | if (!UNIT_CGROUP_BOOL(u, io_accounting)) | |
3948 | return -ENODATA; | |
3949 | ||
3950 | if (allow_cache && u->io_accounting_last[metric] != UINT64_MAX) | |
3951 | goto done; | |
3952 | ||
3953 | r = unit_get_io_accounting_raw(u, raw); | |
3954 | if (r == -ENODATA && u->io_accounting_last[metric] != UINT64_MAX) | |
3955 | goto done; | |
3956 | if (r < 0) | |
3957 | return r; | |
3958 | ||
3959 | for (CGroupIOAccountingMetric i = 0; i < _CGROUP_IO_ACCOUNTING_METRIC_MAX; i++) { | |
3960 | /* Saturated subtraction */ | |
3961 | if (raw[i] > u->io_accounting_base[i]) | |
3962 | u->io_accounting_last[i] = raw[i] - u->io_accounting_base[i]; | |
3963 | else | |
3964 | u->io_accounting_last[i] = 0; | |
3965 | } | |
3966 | ||
3967 | done: | |
3968 | if (ret) | |
3969 | *ret = u->io_accounting_last[metric]; | |
3970 | ||
3971 | return 0; | |
3972 | } | |
3973 | ||
906c06f6 | 3974 | int unit_reset_cpu_accounting(Unit *u) { |
5ad096b3 LP |
3975 | int r; |
3976 | ||
3977 | assert(u); | |
3978 | ||
fe700f46 LP |
3979 | u->cpu_usage_last = NSEC_INFINITY; |
3980 | ||
0bbff7d6 | 3981 | r = unit_get_cpu_usage_raw(u, &u->cpu_usage_base); |
5ad096b3 | 3982 | if (r < 0) { |
66ebf6c0 | 3983 | u->cpu_usage_base = 0; |
5ad096b3 | 3984 | return r; |
b56c28c3 | 3985 | } |
2633eb83 | 3986 | |
4ad49000 | 3987 | return 0; |
4fbf50b3 LP |
3988 | } |
3989 | ||
906c06f6 DM |
3990 | int unit_reset_ip_accounting(Unit *u) { |
3991 | int r = 0, q = 0; | |
3992 | ||
3993 | assert(u); | |
3994 | ||
3995 | if (u->ip_accounting_ingress_map_fd >= 0) | |
3996 | r = bpf_firewall_reset_accounting(u->ip_accounting_ingress_map_fd); | |
3997 | ||
3998 | if (u->ip_accounting_egress_map_fd >= 0) | |
3999 | q = bpf_firewall_reset_accounting(u->ip_accounting_egress_map_fd); | |
4000 | ||
6b659ed8 LP |
4001 | zero(u->ip_accounting_extra); |
4002 | ||
906c06f6 DM |
4003 | return r < 0 ? r : q; |
4004 | } | |
4005 | ||
fbe14fc9 LP |
4006 | int unit_reset_io_accounting(Unit *u) { |
4007 | int r; | |
4008 | ||
4009 | assert(u); | |
4010 | ||
4011 | for (CGroupIOAccountingMetric i = 0; i < _CGROUP_IO_ACCOUNTING_METRIC_MAX; i++) | |
4012 | u->io_accounting_last[i] = UINT64_MAX; | |
4013 | ||
4014 | r = unit_get_io_accounting_raw(u, u->io_accounting_base); | |
4015 | if (r < 0) { | |
4016 | zero(u->io_accounting_base); | |
4017 | return r; | |
4018 | } | |
4019 | ||
4020 | return 0; | |
4021 | } | |
4022 | ||
9b2559a1 | 4023 | int unit_reset_accounting(Unit *u) { |
fbe14fc9 | 4024 | int r, q, v; |
9b2559a1 LP |
4025 | |
4026 | assert(u); | |
4027 | ||
4028 | r = unit_reset_cpu_accounting(u); | |
fbe14fc9 LP |
4029 | q = unit_reset_io_accounting(u); |
4030 | v = unit_reset_ip_accounting(u); | |
9b2559a1 | 4031 | |
fbe14fc9 | 4032 | return r < 0 ? r : q < 0 ? q : v; |
9b2559a1 LP |
4033 | } |
4034 | ||
e7ab4d1a LP |
4035 | void unit_invalidate_cgroup(Unit *u, CGroupMask m) { |
4036 | assert(u); | |
4037 | ||
4038 | if (!UNIT_HAS_CGROUP_CONTEXT(u)) | |
4039 | return; | |
4040 | ||
4041 | if (m == 0) | |
4042 | return; | |
4043 | ||
538b4852 TH |
4044 | /* always invalidate compat pairs together */ |
4045 | if (m & (CGROUP_MASK_IO | CGROUP_MASK_BLKIO)) | |
4046 | m |= CGROUP_MASK_IO | CGROUP_MASK_BLKIO; | |
4047 | ||
7cce4fb7 LP |
4048 | if (m & (CGROUP_MASK_CPU | CGROUP_MASK_CPUACCT)) |
4049 | m |= CGROUP_MASK_CPU | CGROUP_MASK_CPUACCT; | |
4050 | ||
e00068e7 | 4051 | if (FLAGS_SET(u->cgroup_invalidated_mask, m)) /* NOP? */ |
e7ab4d1a LP |
4052 | return; |
4053 | ||
e00068e7 | 4054 | u->cgroup_invalidated_mask |= m; |
91a6073e | 4055 | unit_add_to_cgroup_realize_queue(u); |
e7ab4d1a LP |
4056 | } |
4057 | ||
906c06f6 DM |
4058 | void unit_invalidate_cgroup_bpf(Unit *u) { |
4059 | assert(u); | |
4060 | ||
4061 | if (!UNIT_HAS_CGROUP_CONTEXT(u)) | |
4062 | return; | |
4063 | ||
17f14955 | 4064 | if (u->cgroup_invalidated_mask & CGROUP_MASK_BPF_FIREWALL) /* NOP? */ |
906c06f6 DM |
4065 | return; |
4066 | ||
17f14955 | 4067 | u->cgroup_invalidated_mask |= CGROUP_MASK_BPF_FIREWALL; |
91a6073e | 4068 | unit_add_to_cgroup_realize_queue(u); |
906c06f6 DM |
4069 | |
4070 | /* If we are a slice unit, we also need to put compile a new BPF program for all our children, as the IP access | |
4071 | * list of our children includes our own. */ | |
4072 | if (u->type == UNIT_SLICE) { | |
4073 | Unit *member; | |
906c06f6 | 4074 | |
d219a2b0 | 4075 | UNIT_FOREACH_DEPENDENCY(member, u, UNIT_ATOM_SLICE_OF) |
15ed3c3a | 4076 | unit_invalidate_cgroup_bpf(member); |
906c06f6 DM |
4077 | } |
4078 | } | |
4079 | ||
869f52f2 DS |
4080 | void unit_cgroup_catchup(Unit *u) { |
4081 | assert(u); | |
4082 | ||
4083 | if (!UNIT_HAS_CGROUP_CONTEXT(u)) | |
4084 | return; | |
4085 | ||
4086 | /* We dropped the inotify watch during reexec/reload, so we need to | |
4087 | * check these as they may have changed. | |
4088 | * Note that (currently) the kernel doesn't actually update cgroup | |
4089 | * file modification times, so we can't just serialize and then check | |
4090 | * the mtime for file(s) we are interested in. */ | |
4091 | (void) unit_check_cgroup_events(u); | |
4092 | unit_add_to_cgroup_oom_queue(u); | |
4093 | } | |
4094 | ||
1d9cc876 LP |
4095 | bool unit_cgroup_delegate(Unit *u) { |
4096 | CGroupContext *c; | |
4097 | ||
4098 | assert(u); | |
4099 | ||
4100 | if (!UNIT_VTABLE(u)->can_delegate) | |
4101 | return false; | |
4102 | ||
4103 | c = unit_get_cgroup_context(u); | |
4104 | if (!c) | |
4105 | return false; | |
4106 | ||
4107 | return c->delegate; | |
4108 | } | |
4109 | ||
e7ab4d1a | 4110 | void manager_invalidate_startup_units(Manager *m) { |
e7ab4d1a LP |
4111 | Unit *u; |
4112 | ||
4113 | assert(m); | |
4114 | ||
90e74a66 | 4115 | SET_FOREACH(u, m->startup_units) |
9dfb6a3a | 4116 | unit_invalidate_cgroup(u, CGROUP_MASK_CPU|CGROUP_MASK_IO|CGROUP_MASK_BLKIO|CGROUP_MASK_CPUSET); |
e7ab4d1a LP |
4117 | } |
4118 | ||
da8e1782 MO |
4119 | static int unit_get_nice(Unit *u) { |
4120 | ExecContext *ec; | |
4121 | ||
4122 | ec = unit_get_exec_context(u); | |
4123 | return ec ? ec->nice : 0; | |
4124 | } | |
4125 | ||
4126 | static uint64_t unit_get_cpu_weight(Unit *u) { | |
4127 | ManagerState state = manager_state(u->manager); | |
4128 | CGroupContext *cc; | |
4129 | ||
4130 | cc = unit_get_cgroup_context(u); | |
4131 | return cc ? cgroup_context_cpu_weight(cc, state) : CGROUP_WEIGHT_DEFAULT; | |
4132 | } | |
4133 | ||
4134 | int compare_job_priority(const void *a, const void *b) { | |
4135 | const Job *x = a, *y = b; | |
4136 | int nice_x, nice_y; | |
4137 | uint64_t weight_x, weight_y; | |
4138 | int ret; | |
4139 | ||
217b7b33 ZJS |
4140 | if ((ret = CMP(x->unit->type, y->unit->type)) != 0) |
4141 | return -ret; | |
4142 | ||
da8e1782 MO |
4143 | weight_x = unit_get_cpu_weight(x->unit); |
4144 | weight_y = unit_get_cpu_weight(y->unit); | |
4145 | ||
217b7b33 ZJS |
4146 | if ((ret = CMP(weight_x, weight_y)) != 0) |
4147 | return -ret; | |
da8e1782 MO |
4148 | |
4149 | nice_x = unit_get_nice(x->unit); | |
4150 | nice_y = unit_get_nice(y->unit); | |
4151 | ||
4152 | if ((ret = CMP(nice_x, nice_y)) != 0) | |
4153 | return ret; | |
4154 | ||
da8e1782 MO |
4155 | return strcmp(x->unit->id, y->unit->id); |
4156 | } | |
4157 | ||
d9e45bc3 MS |
4158 | int unit_cgroup_freezer_action(Unit *u, FreezerAction action) { |
4159 | _cleanup_free_ char *path = NULL; | |
4160 | FreezerState target, kernel = _FREEZER_STATE_INVALID; | |
4161 | int r; | |
4162 | ||
4163 | assert(u); | |
4164 | assert(IN_SET(action, FREEZER_FREEZE, FREEZER_THAW)); | |
4165 | ||
9a1e90ae MS |
4166 | if (!cg_freezer_supported()) |
4167 | return 0; | |
4168 | ||
d9e45bc3 MS |
4169 | if (!u->cgroup_realized) |
4170 | return -EBUSY; | |
4171 | ||
4172 | target = action == FREEZER_FREEZE ? FREEZER_FROZEN : FREEZER_RUNNING; | |
4173 | ||
4174 | r = unit_freezer_state_kernel(u, &kernel); | |
4175 | if (r < 0) | |
4176 | log_unit_debug_errno(u, r, "Failed to obtain cgroup freezer state: %m"); | |
4177 | ||
4178 | if (target == kernel) { | |
4179 | u->freezer_state = target; | |
4180 | return 0; | |
4181 | } | |
4182 | ||
4183 | r = cg_get_path(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path, "cgroup.freeze", &path); | |
4184 | if (r < 0) | |
4185 | return r; | |
4186 | ||
4187 | log_unit_debug(u, "%s unit.", action == FREEZER_FREEZE ? "Freezing" : "Thawing"); | |
4188 | ||
4189 | if (action == FREEZER_FREEZE) | |
4190 | u->freezer_state = FREEZER_FREEZING; | |
4191 | else | |
4192 | u->freezer_state = FREEZER_THAWING; | |
4193 | ||
4194 | r = write_string_file(path, one_zero(action == FREEZER_FREEZE), WRITE_STRING_FILE_DISABLE_BUFFER); | |
4195 | if (r < 0) | |
4196 | return r; | |
4197 | ||
d910f4c2 | 4198 | return 1; |
d9e45bc3 MS |
4199 | } |
4200 | ||
047f5d63 PH |
4201 | int unit_get_cpuset(Unit *u, CPUSet *cpus, const char *name) { |
4202 | _cleanup_free_ char *v = NULL; | |
4203 | int r; | |
4204 | ||
4205 | assert(u); | |
4206 | assert(cpus); | |
4207 | ||
4208 | if (!u->cgroup_path) | |
4209 | return -ENODATA; | |
4210 | ||
4211 | if ((u->cgroup_realized_mask & CGROUP_MASK_CPUSET) == 0) | |
4212 | return -ENODATA; | |
4213 | ||
4214 | r = cg_all_unified(); | |
4215 | if (r < 0) | |
4216 | return r; | |
4217 | if (r == 0) | |
4218 | return -ENODATA; | |
48fd01e5 LP |
4219 | |
4220 | r = cg_get_attribute("cpuset", u->cgroup_path, name, &v); | |
047f5d63 PH |
4221 | if (r == -ENOENT) |
4222 | return -ENODATA; | |
4223 | if (r < 0) | |
4224 | return r; | |
4225 | ||
4226 | return parse_cpu_set_full(v, cpus, false, NULL, NULL, 0, NULL); | |
4227 | } | |
4228 | ||
4e806bfa AZ |
4229 | static const char* const cgroup_device_policy_table[_CGROUP_DEVICE_POLICY_MAX] = { |
4230 | [CGROUP_DEVICE_POLICY_AUTO] = "auto", | |
4231 | [CGROUP_DEVICE_POLICY_CLOSED] = "closed", | |
4232 | [CGROUP_DEVICE_POLICY_STRICT] = "strict", | |
4233 | }; | |
4234 | ||
4ad49000 | 4235 | DEFINE_STRING_TABLE_LOOKUP(cgroup_device_policy, CGroupDevicePolicy); |
d9e45bc3 MS |
4236 | |
4237 | static const char* const freezer_action_table[_FREEZER_ACTION_MAX] = { | |
4238 | [FREEZER_FREEZE] = "freeze", | |
4239 | [FREEZER_THAW] = "thaw", | |
4240 | }; | |
4241 | ||
4242 | DEFINE_STRING_TABLE_LOOKUP(freezer_action, FreezerAction); |