]>
Commit | Line | Data |
---|---|---|
53e1b683 | 1 | /* SPDX-License-Identifier: LGPL-2.1+ */ |
8e274523 | 2 | |
c6c18be3 | 3 | #include <fcntl.h> |
8c6db833 | 4 | |
afcfaa69 LP |
5 | #include "sd-messages.h" |
6 | ||
b5efdb8a | 7 | #include "alloc-util.h" |
18c528e9 | 8 | #include "blockdev-util.h" |
d8b4d14d | 9 | #include "bpf-devices.h" |
906c06f6 | 10 | #include "bpf-firewall.h" |
45c2e068 | 11 | #include "btrfs-util.h" |
6592b975 | 12 | #include "bus-error.h" |
fdb3deca | 13 | #include "cgroup-setup.h" |
03a7b521 | 14 | #include "cgroup-util.h" |
3ffd4af2 LP |
15 | #include "cgroup.h" |
16 | #include "fd-util.h" | |
0d39fa9c | 17 | #include "fileio.h" |
77601719 | 18 | #include "fs-util.h" |
6bedfcbb | 19 | #include "parse-util.h" |
9eb977db | 20 | #include "path-util.h" |
03a7b521 | 21 | #include "process-util.h" |
c36a69f4 | 22 | #include "procfs-util.h" |
9444b1f2 | 23 | #include "special.h" |
74c48bf5 | 24 | #include "stat-util.h" |
906c06f6 | 25 | #include "stdio-util.h" |
8b43440b | 26 | #include "string-table.h" |
07630cea | 27 | #include "string-util.h" |
cc6271f1 | 28 | #include "virt.h" |
8e274523 | 29 | |
10f28641 | 30 | #define CGROUP_CPU_QUOTA_DEFAULT_PERIOD_USEC ((usec_t) 100 * USEC_PER_MSEC) |
9a054909 | 31 | |
39b9fefb LP |
32 | /* Returns the log level to use when cgroup attribute writes fail. When an attribute is missing or we have access |
33 | * problems we downgrade to LOG_DEBUG. This is supposed to be nice to container managers and kernels which want to mask | |
34 | * out specific attributes from us. */ | |
35 | #define LOG_LEVEL_CGROUP_WRITE(r) (IN_SET(abs(r), ENOENT, EROFS, EACCES, EPERM) ? LOG_DEBUG : LOG_WARNING) | |
36 | ||
611c4f8a | 37 | bool manager_owns_host_root_cgroup(Manager *m) { |
cc6271f1 LP |
38 | assert(m); |
39 | ||
40 | /* Returns true if we are managing the root cgroup. Note that it isn't sufficient to just check whether the | |
41 | * group root path equals "/" since that will also be the case if CLONE_NEWCGROUP is in the mix. Since there's | |
42 | * appears to be no nice way to detect whether we are in a CLONE_NEWCGROUP namespace we instead just check if | |
43 | * we run in any kind of container virtualization. */ | |
44 | ||
28cfdc5a LP |
45 | if (MANAGER_IS_USER(m)) |
46 | return false; | |
47 | ||
cc6271f1 LP |
48 | if (detect_container() > 0) |
49 | return false; | |
50 | ||
57ea45e1 | 51 | return empty_or_root(m->cgroup_root); |
cc6271f1 LP |
52 | } |
53 | ||
611c4f8a | 54 | bool unit_has_host_root_cgroup(Unit *u) { |
f3725e64 LP |
55 | assert(u); |
56 | ||
cc6271f1 LP |
57 | /* Returns whether this unit manages the root cgroup. This will return true if this unit is the root slice and |
58 | * the manager manages the root cgroup. */ | |
f3725e64 | 59 | |
611c4f8a | 60 | if (!manager_owns_host_root_cgroup(u->manager)) |
f3725e64 LP |
61 | return false; |
62 | ||
cc6271f1 | 63 | return unit_has_name(u, SPECIAL_ROOT_SLICE); |
f3725e64 LP |
64 | } |
65 | ||
293d32df LP |
66 | static int set_attribute_and_warn(Unit *u, const char *controller, const char *attribute, const char *value) { |
67 | int r; | |
68 | ||
69 | r = cg_set_attribute(controller, u->cgroup_path, attribute, value); | |
70 | if (r < 0) | |
71 | log_unit_full(u, LOG_LEVEL_CGROUP_WRITE(r), r, "Failed to set '%s' attribute on '%s' to '%.*s': %m", | |
72 | strna(attribute), isempty(u->cgroup_path) ? "/" : u->cgroup_path, (int) strcspn(value, NEWLINE), value); | |
73 | ||
74 | return r; | |
75 | } | |
76 | ||
2b40998d | 77 | static void cgroup_compat_warn(void) { |
128fadc9 TH |
78 | static bool cgroup_compat_warned = false; |
79 | ||
80 | if (cgroup_compat_warned) | |
81 | return; | |
82 | ||
cc6271f1 LP |
83 | log_warning("cgroup compatibility translation between legacy and unified hierarchy settings activated. " |
84 | "See cgroup-compat debug messages for details."); | |
85 | ||
128fadc9 TH |
86 | cgroup_compat_warned = true; |
87 | } | |
88 | ||
89 | #define log_cgroup_compat(unit, fmt, ...) do { \ | |
90 | cgroup_compat_warn(); \ | |
91 | log_unit_debug(unit, "cgroup-compat: " fmt, ##__VA_ARGS__); \ | |
2b40998d | 92 | } while (false) |
128fadc9 | 93 | |
4ad49000 LP |
94 | void cgroup_context_init(CGroupContext *c) { |
95 | assert(c); | |
96 | ||
de8a711a | 97 | /* Initialize everything to the kernel defaults. */ |
4ad49000 | 98 | |
de8a711a LP |
99 | *c = (CGroupContext) { |
100 | .cpu_weight = CGROUP_WEIGHT_INVALID, | |
101 | .startup_cpu_weight = CGROUP_WEIGHT_INVALID, | |
102 | .cpu_quota_per_sec_usec = USEC_INFINITY, | |
10f28641 | 103 | .cpu_quota_period_usec = USEC_INFINITY, |
66ebf6c0 | 104 | |
de8a711a LP |
105 | .cpu_shares = CGROUP_CPU_SHARES_INVALID, |
106 | .startup_cpu_shares = CGROUP_CPU_SHARES_INVALID, | |
d53d9474 | 107 | |
de8a711a LP |
108 | .memory_high = CGROUP_LIMIT_MAX, |
109 | .memory_max = CGROUP_LIMIT_MAX, | |
110 | .memory_swap_max = CGROUP_LIMIT_MAX, | |
da4d897e | 111 | |
de8a711a | 112 | .memory_limit = CGROUP_LIMIT_MAX, |
b2f8b02e | 113 | |
de8a711a LP |
114 | .io_weight = CGROUP_WEIGHT_INVALID, |
115 | .startup_io_weight = CGROUP_WEIGHT_INVALID, | |
13c31542 | 116 | |
de8a711a LP |
117 | .blockio_weight = CGROUP_BLKIO_WEIGHT_INVALID, |
118 | .startup_blockio_weight = CGROUP_BLKIO_WEIGHT_INVALID, | |
d53d9474 | 119 | |
de8a711a LP |
120 | .tasks_max = CGROUP_LIMIT_MAX, |
121 | }; | |
4ad49000 | 122 | } |
8e274523 | 123 | |
4ad49000 LP |
124 | void cgroup_context_free_device_allow(CGroupContext *c, CGroupDeviceAllow *a) { |
125 | assert(c); | |
126 | assert(a); | |
127 | ||
71fda00f | 128 | LIST_REMOVE(device_allow, c->device_allow, a); |
4ad49000 LP |
129 | free(a->path); |
130 | free(a); | |
131 | } | |
132 | ||
13c31542 TH |
133 | void cgroup_context_free_io_device_weight(CGroupContext *c, CGroupIODeviceWeight *w) { |
134 | assert(c); | |
135 | assert(w); | |
136 | ||
137 | LIST_REMOVE(device_weights, c->io_device_weights, w); | |
138 | free(w->path); | |
139 | free(w); | |
140 | } | |
141 | ||
6ae4283c TH |
142 | void cgroup_context_free_io_device_latency(CGroupContext *c, CGroupIODeviceLatency *l) { |
143 | assert(c); | |
144 | assert(l); | |
145 | ||
146 | LIST_REMOVE(device_latencies, c->io_device_latencies, l); | |
147 | free(l->path); | |
148 | free(l); | |
149 | } | |
150 | ||
13c31542 TH |
151 | void cgroup_context_free_io_device_limit(CGroupContext *c, CGroupIODeviceLimit *l) { |
152 | assert(c); | |
153 | assert(l); | |
154 | ||
155 | LIST_REMOVE(device_limits, c->io_device_limits, l); | |
156 | free(l->path); | |
157 | free(l); | |
158 | } | |
159 | ||
4ad49000 LP |
160 | void cgroup_context_free_blockio_device_weight(CGroupContext *c, CGroupBlockIODeviceWeight *w) { |
161 | assert(c); | |
162 | assert(w); | |
163 | ||
71fda00f | 164 | LIST_REMOVE(device_weights, c->blockio_device_weights, w); |
4ad49000 LP |
165 | free(w->path); |
166 | free(w); | |
167 | } | |
168 | ||
169 | void cgroup_context_free_blockio_device_bandwidth(CGroupContext *c, CGroupBlockIODeviceBandwidth *b) { | |
170 | assert(c); | |
8e274523 | 171 | assert(b); |
8e274523 | 172 | |
71fda00f | 173 | LIST_REMOVE(device_bandwidths, c->blockio_device_bandwidths, b); |
4ad49000 LP |
174 | free(b->path); |
175 | free(b); | |
176 | } | |
177 | ||
178 | void cgroup_context_done(CGroupContext *c) { | |
179 | assert(c); | |
180 | ||
13c31542 TH |
181 | while (c->io_device_weights) |
182 | cgroup_context_free_io_device_weight(c, c->io_device_weights); | |
183 | ||
6ae4283c TH |
184 | while (c->io_device_latencies) |
185 | cgroup_context_free_io_device_latency(c, c->io_device_latencies); | |
186 | ||
13c31542 TH |
187 | while (c->io_device_limits) |
188 | cgroup_context_free_io_device_limit(c, c->io_device_limits); | |
189 | ||
4ad49000 LP |
190 | while (c->blockio_device_weights) |
191 | cgroup_context_free_blockio_device_weight(c, c->blockio_device_weights); | |
192 | ||
193 | while (c->blockio_device_bandwidths) | |
194 | cgroup_context_free_blockio_device_bandwidth(c, c->blockio_device_bandwidths); | |
195 | ||
196 | while (c->device_allow) | |
197 | cgroup_context_free_device_allow(c, c->device_allow); | |
6a48d82f DM |
198 | |
199 | c->ip_address_allow = ip_address_access_free_all(c->ip_address_allow); | |
200 | c->ip_address_deny = ip_address_access_free_all(c->ip_address_deny); | |
fab34748 KL |
201 | |
202 | c->ip_filters_ingress = strv_free(c->ip_filters_ingress); | |
203 | c->ip_filters_egress = strv_free(c->ip_filters_egress); | |
047f5d63 PH |
204 | |
205 | cpu_set_reset(&c->cpuset_cpus); | |
206 | cpu_set_reset(&c->cpuset_mems); | |
4ad49000 LP |
207 | } |
208 | ||
74b5fb27 CD |
209 | static int unit_get_kernel_memory_limit(Unit *u, const char *file, uint64_t *ret) { |
210 | _cleanup_free_ char *raw_kval = NULL; | |
211 | uint64_t kval; | |
212 | int r; | |
213 | ||
214 | assert(u); | |
215 | ||
216 | if (!u->cgroup_realized) | |
217 | return -EOWNERDEAD; | |
218 | ||
219 | r = cg_get_attribute("memory", u->cgroup_path, file, &raw_kval); | |
220 | if (r < 0) | |
221 | return r; | |
222 | ||
223 | if (streq(raw_kval, "max")) { | |
224 | *ret = CGROUP_LIMIT_MAX; | |
225 | return 0; | |
226 | } | |
227 | ||
228 | r = safe_atou64(raw_kval, &kval); | |
229 | if (r < 0) | |
230 | return r; | |
231 | ||
232 | *ret = kval; | |
233 | ||
234 | return 0; | |
235 | } | |
236 | ||
237 | static int unit_compare_memory_limit(Unit *u, const char *property_name, uint64_t *ret_unit_value, uint64_t *ret_kernel_value) { | |
238 | CGroupContext *c; | |
239 | CGroupMask m; | |
240 | const char *file; | |
241 | uint64_t unit_value; | |
242 | int r; | |
243 | ||
244 | /* Compare kernel memcg configuration against our internal systemd state. Unsupported (and will | |
245 | * return -ENODATA) on cgroup v1. | |
246 | * | |
247 | * Returns: | |
248 | * | |
249 | * <0: On error. | |
250 | * 0: If the kernel memory setting doesn't match our configuration. | |
251 | * >0: If the kernel memory setting matches our configuration. | |
252 | * | |
253 | * The following values are only guaranteed to be populated on return >=0: | |
254 | * | |
255 | * - ret_unit_value will contain our internal expected value for the unit, page-aligned. | |
256 | * - ret_kernel_value will contain the actual value presented by the kernel. */ | |
257 | ||
258 | assert(u); | |
259 | ||
260 | r = cg_all_unified(); | |
261 | if (r < 0) | |
262 | return log_debug_errno(r, "Failed to determine cgroup hierarchy version: %m"); | |
263 | ||
264 | /* Unsupported on v1. | |
265 | * | |
266 | * We don't return ENOENT, since that could actually mask a genuine problem where somebody else has | |
267 | * silently masked the controller. */ | |
268 | if (r == 0) | |
269 | return -ENODATA; | |
270 | ||
271 | /* The root slice doesn't have any controller files, so we can't compare anything. */ | |
272 | if (unit_has_name(u, SPECIAL_ROOT_SLICE)) | |
273 | return -ENODATA; | |
274 | ||
275 | /* It's possible to have MemoryFoo set without systemd wanting to have the memory controller enabled, | |
276 | * for example, in the case of DisableControllers= or cgroup_disable on the kernel command line. To | |
277 | * avoid specious errors in these scenarios, check that we even expect the memory controller to be | |
278 | * enabled at all. */ | |
279 | m = unit_get_target_mask(u); | |
280 | if (!FLAGS_SET(m, CGROUP_MASK_MEMORY)) | |
281 | return -ENODATA; | |
282 | ||
283 | c = unit_get_cgroup_context(u); | |
284 | assert(c); | |
285 | ||
286 | if (streq(property_name, "MemoryLow")) { | |
287 | unit_value = unit_get_ancestor_memory_low(u); | |
288 | file = "memory.low"; | |
289 | } else if (streq(property_name, "MemoryMin")) { | |
290 | unit_value = unit_get_ancestor_memory_min(u); | |
291 | file = "memory.min"; | |
292 | } else if (streq(property_name, "MemoryHigh")) { | |
293 | unit_value = c->memory_high; | |
294 | file = "memory.high"; | |
295 | } else if (streq(property_name, "MemoryMax")) { | |
296 | unit_value = c->memory_max; | |
297 | file = "memory.max"; | |
298 | } else if (streq(property_name, "MemorySwapMax")) { | |
299 | unit_value = c->memory_swap_max; | |
300 | file = "memory.swap.max"; | |
301 | } else | |
302 | return -EINVAL; | |
303 | ||
304 | r = unit_get_kernel_memory_limit(u, file, ret_kernel_value); | |
305 | if (r < 0) | |
306 | return log_unit_debug_errno(u, r, "Failed to parse %s: %m", file); | |
307 | ||
308 | /* It's intended (soon) in a future kernel to not expose cgroup memory limits rounded to page | |
309 | * boundaries, but instead separate the user-exposed limit, which is whatever userspace told us, from | |
310 | * our internal page-counting. To support those future kernels, just check the value itself first | |
311 | * without any page-alignment. */ | |
312 | if (*ret_kernel_value == unit_value) { | |
313 | *ret_unit_value = unit_value; | |
314 | return 1; | |
315 | } | |
316 | ||
317 | /* The current kernel behaviour, by comparison, is that even if you write a particular number of | |
318 | * bytes into a cgroup memory file, it always returns that number page-aligned down (since the kernel | |
319 | * internally stores cgroup limits in pages). As such, so long as it aligns properly, everything is | |
320 | * cricket. */ | |
321 | if (unit_value != CGROUP_LIMIT_MAX) | |
322 | unit_value = PAGE_ALIGN_DOWN(unit_value); | |
323 | ||
324 | *ret_unit_value = unit_value; | |
325 | ||
326 | return *ret_kernel_value == *ret_unit_value; | |
327 | } | |
328 | ||
bc0623df CD |
329 | #define FORMAT_CGROUP_DIFF_MAX 128 |
330 | ||
331 | static char *format_cgroup_memory_limit_comparison(char *buf, size_t l, Unit *u, const char *property_name) { | |
332 | uint64_t kval, sval; | |
333 | int r; | |
334 | ||
335 | assert(u); | |
336 | assert(buf); | |
337 | assert(l > 0); | |
338 | ||
339 | r = unit_compare_memory_limit(u, property_name, &sval, &kval); | |
340 | ||
341 | /* memory.swap.max is special in that it relies on CONFIG_MEMCG_SWAP (and the default swapaccount=1). | |
342 | * In the absence of reliably being able to detect whether memcg swap support is available or not, | |
343 | * only complain if the error is not ENOENT. */ | |
344 | if (r > 0 || IN_SET(r, -ENODATA, -EOWNERDEAD) || | |
345 | (r == -ENOENT && streq(property_name, "MemorySwapMax"))) { | |
346 | buf[0] = 0; | |
347 | return buf; | |
348 | } | |
349 | ||
350 | if (r < 0) { | |
351 | snprintf(buf, l, " (error getting kernel value: %s)", strerror_safe(r)); | |
352 | return buf; | |
353 | } | |
354 | ||
355 | snprintf(buf, l, " (different value in kernel: %" PRIu64 ")", kval); | |
356 | ||
357 | return buf; | |
358 | } | |
359 | ||
360 | void cgroup_context_dump(Unit *u, FILE* f, const char *prefix) { | |
85c3b278 | 361 | _cleanup_free_ char *disable_controllers_str = NULL, *cpuset_cpus = NULL, *cpuset_mems = NULL; |
13c31542 TH |
362 | CGroupIODeviceLimit *il; |
363 | CGroupIODeviceWeight *iw; | |
6ae4283c | 364 | CGroupIODeviceLatency *l; |
4ad49000 LP |
365 | CGroupBlockIODeviceBandwidth *b; |
366 | CGroupBlockIODeviceWeight *w; | |
367 | CGroupDeviceAllow *a; | |
bc0623df | 368 | CGroupContext *c; |
c21c9906 | 369 | IPAddressAccessItem *iaai; |
fab34748 | 370 | char **path; |
bc0623df | 371 | char q[FORMAT_TIMESPAN_MAX]; |
10f28641 | 372 | char v[FORMAT_TIMESPAN_MAX]; |
4ad49000 | 373 | |
bc0623df CD |
374 | char cda[FORMAT_CGROUP_DIFF_MAX]; |
375 | char cdb[FORMAT_CGROUP_DIFF_MAX]; | |
376 | char cdc[FORMAT_CGROUP_DIFF_MAX]; | |
377 | char cdd[FORMAT_CGROUP_DIFF_MAX]; | |
378 | char cde[FORMAT_CGROUP_DIFF_MAX]; | |
379 | ||
380 | assert(u); | |
4ad49000 LP |
381 | assert(f); |
382 | ||
bc0623df CD |
383 | c = unit_get_cgroup_context(u); |
384 | assert(c); | |
385 | ||
4ad49000 LP |
386 | prefix = strempty(prefix); |
387 | ||
25cc30c4 AZ |
388 | (void) cg_mask_to_string(c->disable_controllers, &disable_controllers_str); |
389 | ||
047f5d63 PH |
390 | cpuset_cpus = cpu_set_to_range_string(&c->cpuset_cpus); |
391 | cpuset_mems = cpu_set_to_range_string(&c->cpuset_mems); | |
392 | ||
4ad49000 | 393 | fprintf(f, |
6dfb9282 CD |
394 | "%sCPUAccounting: %s\n" |
395 | "%sIOAccounting: %s\n" | |
396 | "%sBlockIOAccounting: %s\n" | |
397 | "%sMemoryAccounting: %s\n" | |
398 | "%sTasksAccounting: %s\n" | |
399 | "%sIPAccounting: %s\n" | |
400 | "%sCPUWeight: %" PRIu64 "\n" | |
401 | "%sStartupCPUWeight: %" PRIu64 "\n" | |
402 | "%sCPUShares: %" PRIu64 "\n" | |
403 | "%sStartupCPUShares: %" PRIu64 "\n" | |
404 | "%sCPUQuotaPerSecSec: %s\n" | |
405 | "%sCPUQuotaPeriodSec: %s\n" | |
406 | "%sAllowedCPUs: %s\n" | |
407 | "%sAllowedMemoryNodes: %s\n" | |
408 | "%sIOWeight: %" PRIu64 "\n" | |
409 | "%sStartupIOWeight: %" PRIu64 "\n" | |
410 | "%sBlockIOWeight: %" PRIu64 "\n" | |
411 | "%sStartupBlockIOWeight: %" PRIu64 "\n" | |
412 | "%sDefaultMemoryMin: %" PRIu64 "\n" | |
413 | "%sDefaultMemoryLow: %" PRIu64 "\n" | |
bc0623df CD |
414 | "%sMemoryMin: %" PRIu64 "%s\n" |
415 | "%sMemoryLow: %" PRIu64 "%s\n" | |
416 | "%sMemoryHigh: %" PRIu64 "%s\n" | |
417 | "%sMemoryMax: %" PRIu64 "%s\n" | |
418 | "%sMemorySwapMax: %" PRIu64 "%s\n" | |
6dfb9282 CD |
419 | "%sMemoryLimit: %" PRIu64 "\n" |
420 | "%sTasksMax: %" PRIu64 "\n" | |
421 | "%sDevicePolicy: %s\n" | |
422 | "%sDisableControllers: %s\n" | |
423 | "%sDelegate: %s\n", | |
4ad49000 | 424 | prefix, yes_no(c->cpu_accounting), |
13c31542 | 425 | prefix, yes_no(c->io_accounting), |
4ad49000 LP |
426 | prefix, yes_no(c->blockio_accounting), |
427 | prefix, yes_no(c->memory_accounting), | |
d53d9474 | 428 | prefix, yes_no(c->tasks_accounting), |
c21c9906 | 429 | prefix, yes_no(c->ip_accounting), |
66ebf6c0 TH |
430 | prefix, c->cpu_weight, |
431 | prefix, c->startup_cpu_weight, | |
4ad49000 | 432 | prefix, c->cpu_shares, |
95ae05c0 | 433 | prefix, c->startup_cpu_shares, |
bc0623df | 434 | prefix, format_timespan(q, sizeof(q), c->cpu_quota_per_sec_usec, 1), |
10f28641 | 435 | prefix, format_timespan(v, sizeof(v), c->cpu_quota_period_usec, 1), |
85c3b278 LP |
436 | prefix, strempty(cpuset_cpus), |
437 | prefix, strempty(cpuset_mems), | |
13c31542 TH |
438 | prefix, c->io_weight, |
439 | prefix, c->startup_io_weight, | |
4ad49000 | 440 | prefix, c->blockio_weight, |
95ae05c0 | 441 | prefix, c->startup_blockio_weight, |
7ad5439e | 442 | prefix, c->default_memory_min, |
c52db42b | 443 | prefix, c->default_memory_low, |
bc0623df CD |
444 | prefix, c->memory_min, format_cgroup_memory_limit_comparison(cda, sizeof(cda), u, "MemoryMin"), |
445 | prefix, c->memory_low, format_cgroup_memory_limit_comparison(cdb, sizeof(cdb), u, "MemoryLow"), | |
446 | prefix, c->memory_high, format_cgroup_memory_limit_comparison(cdc, sizeof(cdc), u, "MemoryHigh"), | |
447 | prefix, c->memory_max, format_cgroup_memory_limit_comparison(cdd, sizeof(cdd), u, "MemoryMax"), | |
448 | prefix, c->memory_swap_max, format_cgroup_memory_limit_comparison(cde, sizeof(cde), u, "MemorySwapMax"), | |
4ad49000 | 449 | prefix, c->memory_limit, |
03a7b521 | 450 | prefix, c->tasks_max, |
a931ad47 | 451 | prefix, cgroup_device_policy_to_string(c->device_policy), |
f4c43a81 | 452 | prefix, strempty(disable_controllers_str), |
a931ad47 | 453 | prefix, yes_no(c->delegate)); |
4ad49000 | 454 | |
02638280 LP |
455 | if (c->delegate) { |
456 | _cleanup_free_ char *t = NULL; | |
457 | ||
458 | (void) cg_mask_to_string(c->delegate_controllers, &t); | |
459 | ||
6dfb9282 | 460 | fprintf(f, "%sDelegateControllers: %s\n", |
02638280 LP |
461 | prefix, |
462 | strempty(t)); | |
463 | } | |
464 | ||
4ad49000 LP |
465 | LIST_FOREACH(device_allow, a, c->device_allow) |
466 | fprintf(f, | |
6dfb9282 | 467 | "%sDeviceAllow: %s %s%s%s\n", |
4ad49000 LP |
468 | prefix, |
469 | a->path, | |
470 | a->r ? "r" : "", a->w ? "w" : "", a->m ? "m" : ""); | |
471 | ||
13c31542 TH |
472 | LIST_FOREACH(device_weights, iw, c->io_device_weights) |
473 | fprintf(f, | |
6dfb9282 | 474 | "%sIODeviceWeight: %s %" PRIu64 "\n", |
13c31542 TH |
475 | prefix, |
476 | iw->path, | |
477 | iw->weight); | |
478 | ||
6ae4283c TH |
479 | LIST_FOREACH(device_latencies, l, c->io_device_latencies) |
480 | fprintf(f, | |
6dfb9282 | 481 | "%sIODeviceLatencyTargetSec: %s %s\n", |
6ae4283c TH |
482 | prefix, |
483 | l->path, | |
bc0623df | 484 | format_timespan(q, sizeof(q), l->target_usec, 1)); |
6ae4283c | 485 | |
13c31542 TH |
486 | LIST_FOREACH(device_limits, il, c->io_device_limits) { |
487 | char buf[FORMAT_BYTES_MAX]; | |
9be57249 TH |
488 | CGroupIOLimitType type; |
489 | ||
490 | for (type = 0; type < _CGROUP_IO_LIMIT_TYPE_MAX; type++) | |
491 | if (il->limits[type] != cgroup_io_limit_defaults[type]) | |
492 | fprintf(f, | |
6dfb9282 | 493 | "%s%s: %s %s\n", |
9be57249 TH |
494 | prefix, |
495 | cgroup_io_limit_type_to_string(type), | |
496 | il->path, | |
497 | format_bytes(buf, sizeof(buf), il->limits[type])); | |
13c31542 TH |
498 | } |
499 | ||
4ad49000 LP |
500 | LIST_FOREACH(device_weights, w, c->blockio_device_weights) |
501 | fprintf(f, | |
6dfb9282 | 502 | "%sBlockIODeviceWeight: %s %" PRIu64, |
4ad49000 LP |
503 | prefix, |
504 | w->path, | |
505 | w->weight); | |
506 | ||
507 | LIST_FOREACH(device_bandwidths, b, c->blockio_device_bandwidths) { | |
508 | char buf[FORMAT_BYTES_MAX]; | |
509 | ||
979d0311 TH |
510 | if (b->rbps != CGROUP_LIMIT_MAX) |
511 | fprintf(f, | |
6dfb9282 | 512 | "%sBlockIOReadBandwidth: %s %s\n", |
979d0311 TH |
513 | prefix, |
514 | b->path, | |
515 | format_bytes(buf, sizeof(buf), b->rbps)); | |
516 | if (b->wbps != CGROUP_LIMIT_MAX) | |
517 | fprintf(f, | |
6dfb9282 | 518 | "%sBlockIOWriteBandwidth: %s %s\n", |
979d0311 TH |
519 | prefix, |
520 | b->path, | |
521 | format_bytes(buf, sizeof(buf), b->wbps)); | |
4ad49000 | 522 | } |
c21c9906 LP |
523 | |
524 | LIST_FOREACH(items, iaai, c->ip_address_allow) { | |
525 | _cleanup_free_ char *k = NULL; | |
526 | ||
527 | (void) in_addr_to_string(iaai->family, &iaai->address, &k); | |
6dfb9282 | 528 | fprintf(f, "%sIPAddressAllow: %s/%u\n", prefix, strnull(k), iaai->prefixlen); |
c21c9906 LP |
529 | } |
530 | ||
531 | LIST_FOREACH(items, iaai, c->ip_address_deny) { | |
532 | _cleanup_free_ char *k = NULL; | |
533 | ||
534 | (void) in_addr_to_string(iaai->family, &iaai->address, &k); | |
6dfb9282 | 535 | fprintf(f, "%sIPAddressDeny: %s/%u\n", prefix, strnull(k), iaai->prefixlen); |
c21c9906 | 536 | } |
fab34748 KL |
537 | |
538 | STRV_FOREACH(path, c->ip_filters_ingress) | |
6dfb9282 | 539 | fprintf(f, "%sIPIngressFilterPath: %s\n", prefix, *path); |
fab34748 KL |
540 | |
541 | STRV_FOREACH(path, c->ip_filters_egress) | |
6dfb9282 | 542 | fprintf(f, "%sIPEgressFilterPath: %s\n", prefix, *path); |
4ad49000 LP |
543 | } |
544 | ||
fd870bac YW |
545 | int cgroup_add_device_allow(CGroupContext *c, const char *dev, const char *mode) { |
546 | _cleanup_free_ CGroupDeviceAllow *a = NULL; | |
547 | _cleanup_free_ char *d = NULL; | |
548 | ||
549 | assert(c); | |
550 | assert(dev); | |
551 | assert(isempty(mode) || in_charset(mode, "rwm")); | |
552 | ||
553 | a = new(CGroupDeviceAllow, 1); | |
554 | if (!a) | |
555 | return -ENOMEM; | |
556 | ||
557 | d = strdup(dev); | |
558 | if (!d) | |
559 | return -ENOMEM; | |
560 | ||
561 | *a = (CGroupDeviceAllow) { | |
562 | .path = TAKE_PTR(d), | |
490c5a37 LP |
563 | .r = isempty(mode) || strchr(mode, 'r'), |
564 | .w = isempty(mode) || strchr(mode, 'w'), | |
565 | .m = isempty(mode) || strchr(mode, 'm'), | |
fd870bac YW |
566 | }; |
567 | ||
568 | LIST_PREPEND(device_allow, c->device_allow, a); | |
569 | TAKE_PTR(a); | |
570 | ||
571 | return 0; | |
572 | } | |
573 | ||
6264b85e CD |
574 | #define UNIT_DEFINE_ANCESTOR_MEMORY_LOOKUP(entry) \ |
575 | uint64_t unit_get_ancestor_##entry(Unit *u) { \ | |
576 | CGroupContext *c; \ | |
577 | \ | |
578 | /* 1. Is entry set in this unit? If so, use that. \ | |
579 | * 2. Is the default for this entry set in any \ | |
580 | * ancestor? If so, use that. \ | |
581 | * 3. Otherwise, return CGROUP_LIMIT_MIN. */ \ | |
582 | \ | |
583 | assert(u); \ | |
584 | \ | |
585 | c = unit_get_cgroup_context(u); \ | |
c5322608 | 586 | if (c && c->entry##_set) \ |
6264b85e CD |
587 | return c->entry; \ |
588 | \ | |
c5322608 | 589 | while ((u = UNIT_DEREF(u->slice))) { \ |
6264b85e | 590 | c = unit_get_cgroup_context(u); \ |
c5322608 | 591 | if (c && c->default_##entry##_set) \ |
6264b85e CD |
592 | return c->default_##entry; \ |
593 | } \ | |
594 | \ | |
595 | /* We've reached the root, but nobody had default for \ | |
596 | * this entry set, so set it to the kernel default. */ \ | |
597 | return CGROUP_LIMIT_MIN; \ | |
c52db42b CD |
598 | } |
599 | ||
6264b85e | 600 | UNIT_DEFINE_ANCESTOR_MEMORY_LOOKUP(memory_low); |
7ad5439e | 601 | UNIT_DEFINE_ANCESTOR_MEMORY_LOOKUP(memory_min); |
6264b85e | 602 | |
0d2d6fbf CD |
603 | static void cgroup_xattr_apply(Unit *u) { |
604 | char ids[SD_ID128_STRING_MAX]; | |
605 | int r; | |
606 | ||
607 | assert(u); | |
608 | ||
609 | if (!MANAGER_IS_SYSTEM(u->manager)) | |
610 | return; | |
611 | ||
612 | if (sd_id128_is_null(u->invocation_id)) | |
613 | return; | |
614 | ||
615 | r = cg_set_xattr(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path, | |
616 | "trusted.invocation_id", | |
617 | sd_id128_to_string(u->invocation_id, ids), 32, | |
618 | 0); | |
619 | if (r < 0) | |
620 | log_unit_debug_errno(u, r, "Failed to set invocation ID on control group %s, ignoring: %m", u->cgroup_path); | |
621 | } | |
622 | ||
45c2e068 | 623 | static int lookup_block_device(const char *p, dev_t *ret) { |
f5855697 YS |
624 | dev_t rdev, dev = 0; |
625 | mode_t mode; | |
45c2e068 | 626 | int r; |
4ad49000 LP |
627 | |
628 | assert(p); | |
45c2e068 | 629 | assert(ret); |
4ad49000 | 630 | |
f5855697 | 631 | r = device_path_parse_major_minor(p, &mode, &rdev); |
d5aecba6 | 632 | if (r == -ENODEV) { /* not a parsable device node, need to go to disk */ |
f5855697 | 633 | struct stat st; |
d5aecba6 LP |
634 | if (stat(p, &st) < 0) |
635 | return log_warning_errno(errno, "Couldn't stat device '%s': %m", p); | |
f5855697 YS |
636 | rdev = (dev_t)st.st_rdev; |
637 | dev = (dev_t)st.st_dev; | |
638 | mode = st.st_mode; | |
d5aecba6 LP |
639 | } else if (r < 0) |
640 | return log_warning_errno(r, "Failed to parse major/minor from path '%s': %m", p); | |
641 | ||
f5855697 | 642 | if (S_ISCHR(mode)) { |
d5aecba6 LP |
643 | log_warning("Device node '%s' is a character device, but block device needed.", p); |
644 | return -ENOTBLK; | |
f5855697 YS |
645 | } else if (S_ISBLK(mode)) |
646 | *ret = rdev; | |
647 | else if (major(dev) != 0) | |
648 | *ret = dev; /* If this is not a device node then use the block device this file is stored on */ | |
45c2e068 LP |
649 | else { |
650 | /* If this is btrfs, getting the backing block device is a bit harder */ | |
651 | r = btrfs_get_block_device(p, ret); | |
652 | if (r < 0 && r != -ENOTTY) | |
653 | return log_warning_errno(r, "Failed to determine block device backing btrfs file system '%s': %m", p); | |
654 | if (r == -ENOTTY) { | |
655 | log_warning("'%s' is not a block device node, and file system block device cannot be determined or is not local.", p); | |
656 | return -ENODEV; | |
657 | } | |
4ad49000 | 658 | } |
8e274523 | 659 | |
45c2e068 LP |
660 | /* If this is a LUKS device, try to get the originating block device */ |
661 | (void) block_get_originating(*ret, ret); | |
662 | ||
663 | /* If this is a partition, try to get the originating block device */ | |
664 | (void) block_get_whole_disk(*ret, ret); | |
8e274523 | 665 | return 0; |
8e274523 LP |
666 | } |
667 | ||
66ebf6c0 TH |
668 | static bool cgroup_context_has_cpu_weight(CGroupContext *c) { |
669 | return c->cpu_weight != CGROUP_WEIGHT_INVALID || | |
670 | c->startup_cpu_weight != CGROUP_WEIGHT_INVALID; | |
671 | } | |
672 | ||
673 | static bool cgroup_context_has_cpu_shares(CGroupContext *c) { | |
674 | return c->cpu_shares != CGROUP_CPU_SHARES_INVALID || | |
675 | c->startup_cpu_shares != CGROUP_CPU_SHARES_INVALID; | |
676 | } | |
677 | ||
678 | static uint64_t cgroup_context_cpu_weight(CGroupContext *c, ManagerState state) { | |
679 | if (IN_SET(state, MANAGER_STARTING, MANAGER_INITIALIZING) && | |
680 | c->startup_cpu_weight != CGROUP_WEIGHT_INVALID) | |
681 | return c->startup_cpu_weight; | |
682 | else if (c->cpu_weight != CGROUP_WEIGHT_INVALID) | |
683 | return c->cpu_weight; | |
684 | else | |
685 | return CGROUP_WEIGHT_DEFAULT; | |
686 | } | |
687 | ||
688 | static uint64_t cgroup_context_cpu_shares(CGroupContext *c, ManagerState state) { | |
689 | if (IN_SET(state, MANAGER_STARTING, MANAGER_INITIALIZING) && | |
690 | c->startup_cpu_shares != CGROUP_CPU_SHARES_INVALID) | |
691 | return c->startup_cpu_shares; | |
692 | else if (c->cpu_shares != CGROUP_CPU_SHARES_INVALID) | |
693 | return c->cpu_shares; | |
694 | else | |
695 | return CGROUP_CPU_SHARES_DEFAULT; | |
696 | } | |
697 | ||
10f28641 FB |
698 | usec_t cgroup_cpu_adjust_period(usec_t period, usec_t quota, usec_t resolution, usec_t max_period) { |
699 | /* kernel uses a minimum resolution of 1ms, so both period and (quota * period) | |
700 | * need to be higher than that boundary. quota is specified in USecPerSec. | |
701 | * Additionally, period must be at most max_period. */ | |
702 | assert(quota > 0); | |
703 | ||
704 | return MIN(MAX3(period, resolution, resolution * USEC_PER_SEC / quota), max_period); | |
705 | } | |
706 | ||
707 | static usec_t cgroup_cpu_adjust_period_and_log(Unit *u, usec_t period, usec_t quota) { | |
708 | usec_t new_period; | |
709 | ||
710 | if (quota == USEC_INFINITY) | |
711 | /* Always use default period for infinity quota. */ | |
712 | return CGROUP_CPU_QUOTA_DEFAULT_PERIOD_USEC; | |
713 | ||
714 | if (period == USEC_INFINITY) | |
715 | /* Default period was requested. */ | |
716 | period = CGROUP_CPU_QUOTA_DEFAULT_PERIOD_USEC; | |
717 | ||
718 | /* Clamp to interval [1ms, 1s] */ | |
719 | new_period = cgroup_cpu_adjust_period(period, quota, USEC_PER_MSEC, USEC_PER_SEC); | |
720 | ||
721 | if (new_period != period) { | |
722 | char v[FORMAT_TIMESPAN_MAX]; | |
527ede0c | 723 | log_unit_full(u, u->warned_clamping_cpu_quota_period ? LOG_DEBUG : LOG_WARNING, 0, |
10f28641 FB |
724 | "Clamping CPU interval for cpu.max: period is now %s", |
725 | format_timespan(v, sizeof(v), new_period, 1)); | |
527ede0c | 726 | u->warned_clamping_cpu_quota_period = true; |
10f28641 FB |
727 | } |
728 | ||
729 | return new_period; | |
730 | } | |
731 | ||
52fecf20 LP |
732 | static void cgroup_apply_unified_cpu_weight(Unit *u, uint64_t weight) { |
733 | char buf[DECIMAL_STR_MAX(uint64_t) + 2]; | |
66ebf6c0 TH |
734 | |
735 | xsprintf(buf, "%" PRIu64 "\n", weight); | |
293d32df | 736 | (void) set_attribute_and_warn(u, "cpu", "cpu.weight", buf); |
52fecf20 LP |
737 | } |
738 | ||
10f28641 | 739 | static void cgroup_apply_unified_cpu_quota(Unit *u, usec_t quota, usec_t period) { |
52fecf20 | 740 | char buf[(DECIMAL_STR_MAX(usec_t) + 1) * 2 + 1]; |
66ebf6c0 | 741 | |
10f28641 | 742 | period = cgroup_cpu_adjust_period_and_log(u, period, quota); |
66ebf6c0 TH |
743 | if (quota != USEC_INFINITY) |
744 | xsprintf(buf, USEC_FMT " " USEC_FMT "\n", | |
10f28641 | 745 | MAX(quota * period / USEC_PER_SEC, USEC_PER_MSEC), period); |
66ebf6c0 | 746 | else |
10f28641 | 747 | xsprintf(buf, "max " USEC_FMT "\n", period); |
293d32df | 748 | (void) set_attribute_and_warn(u, "cpu", "cpu.max", buf); |
66ebf6c0 TH |
749 | } |
750 | ||
52fecf20 LP |
751 | static void cgroup_apply_legacy_cpu_shares(Unit *u, uint64_t shares) { |
752 | char buf[DECIMAL_STR_MAX(uint64_t) + 2]; | |
66ebf6c0 TH |
753 | |
754 | xsprintf(buf, "%" PRIu64 "\n", shares); | |
293d32df | 755 | (void) set_attribute_and_warn(u, "cpu", "cpu.shares", buf); |
52fecf20 LP |
756 | } |
757 | ||
10f28641 | 758 | static void cgroup_apply_legacy_cpu_quota(Unit *u, usec_t quota, usec_t period) { |
52fecf20 | 759 | char buf[DECIMAL_STR_MAX(usec_t) + 2]; |
66ebf6c0 | 760 | |
10f28641 FB |
761 | period = cgroup_cpu_adjust_period_and_log(u, period, quota); |
762 | ||
763 | xsprintf(buf, USEC_FMT "\n", period); | |
293d32df | 764 | (void) set_attribute_and_warn(u, "cpu", "cpu.cfs_period_us", buf); |
66ebf6c0 TH |
765 | |
766 | if (quota != USEC_INFINITY) { | |
10f28641 | 767 | xsprintf(buf, USEC_FMT "\n", MAX(quota * period / USEC_PER_SEC, USEC_PER_MSEC)); |
293d32df | 768 | (void) set_attribute_and_warn(u, "cpu", "cpu.cfs_quota_us", buf); |
66ebf6c0 | 769 | } else |
589a5f7a | 770 | (void) set_attribute_and_warn(u, "cpu", "cpu.cfs_quota_us", "-1\n"); |
66ebf6c0 TH |
771 | } |
772 | ||
773 | static uint64_t cgroup_cpu_shares_to_weight(uint64_t shares) { | |
774 | return CLAMP(shares * CGROUP_WEIGHT_DEFAULT / CGROUP_CPU_SHARES_DEFAULT, | |
775 | CGROUP_WEIGHT_MIN, CGROUP_WEIGHT_MAX); | |
776 | } | |
777 | ||
778 | static uint64_t cgroup_cpu_weight_to_shares(uint64_t weight) { | |
779 | return CLAMP(weight * CGROUP_CPU_SHARES_DEFAULT / CGROUP_WEIGHT_DEFAULT, | |
780 | CGROUP_CPU_SHARES_MIN, CGROUP_CPU_SHARES_MAX); | |
781 | } | |
782 | ||
2cea199e | 783 | static void cgroup_apply_unified_cpuset(Unit *u, const CPUSet *cpus, const char *name) { |
047f5d63 PH |
784 | _cleanup_free_ char *buf = NULL; |
785 | ||
2cea199e | 786 | buf = cpu_set_to_range_string(cpus); |
c259ac9a LP |
787 | if (!buf) { |
788 | log_oom(); | |
789 | return; | |
790 | } | |
047f5d63 PH |
791 | |
792 | (void) set_attribute_and_warn(u, "cpuset", name, buf); | |
793 | } | |
794 | ||
508c45da | 795 | static bool cgroup_context_has_io_config(CGroupContext *c) { |
538b4852 TH |
796 | return c->io_accounting || |
797 | c->io_weight != CGROUP_WEIGHT_INVALID || | |
798 | c->startup_io_weight != CGROUP_WEIGHT_INVALID || | |
799 | c->io_device_weights || | |
6ae4283c | 800 | c->io_device_latencies || |
538b4852 TH |
801 | c->io_device_limits; |
802 | } | |
803 | ||
508c45da | 804 | static bool cgroup_context_has_blockio_config(CGroupContext *c) { |
538b4852 TH |
805 | return c->blockio_accounting || |
806 | c->blockio_weight != CGROUP_BLKIO_WEIGHT_INVALID || | |
807 | c->startup_blockio_weight != CGROUP_BLKIO_WEIGHT_INVALID || | |
808 | c->blockio_device_weights || | |
809 | c->blockio_device_bandwidths; | |
810 | } | |
811 | ||
508c45da | 812 | static uint64_t cgroup_context_io_weight(CGroupContext *c, ManagerState state) { |
64faf04c TH |
813 | if (IN_SET(state, MANAGER_STARTING, MANAGER_INITIALIZING) && |
814 | c->startup_io_weight != CGROUP_WEIGHT_INVALID) | |
815 | return c->startup_io_weight; | |
816 | else if (c->io_weight != CGROUP_WEIGHT_INVALID) | |
817 | return c->io_weight; | |
818 | else | |
819 | return CGROUP_WEIGHT_DEFAULT; | |
820 | } | |
821 | ||
508c45da | 822 | static uint64_t cgroup_context_blkio_weight(CGroupContext *c, ManagerState state) { |
64faf04c TH |
823 | if (IN_SET(state, MANAGER_STARTING, MANAGER_INITIALIZING) && |
824 | c->startup_blockio_weight != CGROUP_BLKIO_WEIGHT_INVALID) | |
825 | return c->startup_blockio_weight; | |
826 | else if (c->blockio_weight != CGROUP_BLKIO_WEIGHT_INVALID) | |
827 | return c->blockio_weight; | |
828 | else | |
829 | return CGROUP_BLKIO_WEIGHT_DEFAULT; | |
830 | } | |
831 | ||
508c45da | 832 | static uint64_t cgroup_weight_blkio_to_io(uint64_t blkio_weight) { |
538b4852 TH |
833 | return CLAMP(blkio_weight * CGROUP_WEIGHT_DEFAULT / CGROUP_BLKIO_WEIGHT_DEFAULT, |
834 | CGROUP_WEIGHT_MIN, CGROUP_WEIGHT_MAX); | |
835 | } | |
836 | ||
508c45da | 837 | static uint64_t cgroup_weight_io_to_blkio(uint64_t io_weight) { |
538b4852 TH |
838 | return CLAMP(io_weight * CGROUP_BLKIO_WEIGHT_DEFAULT / CGROUP_WEIGHT_DEFAULT, |
839 | CGROUP_BLKIO_WEIGHT_MIN, CGROUP_BLKIO_WEIGHT_MAX); | |
840 | } | |
841 | ||
f29ff115 | 842 | static void cgroup_apply_io_device_weight(Unit *u, const char *dev_path, uint64_t io_weight) { |
64faf04c TH |
843 | char buf[DECIMAL_STR_MAX(dev_t)*2+2+DECIMAL_STR_MAX(uint64_t)+1]; |
844 | dev_t dev; | |
845 | int r; | |
846 | ||
847 | r = lookup_block_device(dev_path, &dev); | |
848 | if (r < 0) | |
849 | return; | |
850 | ||
851 | xsprintf(buf, "%u:%u %" PRIu64 "\n", major(dev), minor(dev), io_weight); | |
293d32df | 852 | (void) set_attribute_and_warn(u, "io", "io.weight", buf); |
64faf04c TH |
853 | } |
854 | ||
f29ff115 | 855 | static void cgroup_apply_blkio_device_weight(Unit *u, const char *dev_path, uint64_t blkio_weight) { |
64faf04c TH |
856 | char buf[DECIMAL_STR_MAX(dev_t)*2+2+DECIMAL_STR_MAX(uint64_t)+1]; |
857 | dev_t dev; | |
858 | int r; | |
859 | ||
860 | r = lookup_block_device(dev_path, &dev); | |
861 | if (r < 0) | |
862 | return; | |
863 | ||
864 | xsprintf(buf, "%u:%u %" PRIu64 "\n", major(dev), minor(dev), blkio_weight); | |
293d32df | 865 | (void) set_attribute_and_warn(u, "blkio", "blkio.weight_device", buf); |
64faf04c TH |
866 | } |
867 | ||
6ae4283c TH |
868 | static void cgroup_apply_io_device_latency(Unit *u, const char *dev_path, usec_t target) { |
869 | char buf[DECIMAL_STR_MAX(dev_t)*2+2+7+DECIMAL_STR_MAX(uint64_t)+1]; | |
870 | dev_t dev; | |
871 | int r; | |
872 | ||
873 | r = lookup_block_device(dev_path, &dev); | |
874 | if (r < 0) | |
875 | return; | |
876 | ||
877 | if (target != USEC_INFINITY) | |
878 | xsprintf(buf, "%u:%u target=%" PRIu64 "\n", major(dev), minor(dev), target); | |
879 | else | |
880 | xsprintf(buf, "%u:%u target=max\n", major(dev), minor(dev)); | |
881 | ||
293d32df | 882 | (void) set_attribute_and_warn(u, "io", "io.latency", buf); |
6ae4283c TH |
883 | } |
884 | ||
17ae2780 | 885 | static void cgroup_apply_io_device_limit(Unit *u, const char *dev_path, uint64_t *limits) { |
64faf04c TH |
886 | char limit_bufs[_CGROUP_IO_LIMIT_TYPE_MAX][DECIMAL_STR_MAX(uint64_t)]; |
887 | char buf[DECIMAL_STR_MAX(dev_t)*2+2+(6+DECIMAL_STR_MAX(uint64_t)+1)*4]; | |
888 | CGroupIOLimitType type; | |
889 | dev_t dev; | |
64faf04c TH |
890 | int r; |
891 | ||
892 | r = lookup_block_device(dev_path, &dev); | |
893 | if (r < 0) | |
17ae2780 | 894 | return; |
64faf04c | 895 | |
17ae2780 LP |
896 | for (type = 0; type < _CGROUP_IO_LIMIT_TYPE_MAX; type++) |
897 | if (limits[type] != cgroup_io_limit_defaults[type]) | |
64faf04c | 898 | xsprintf(limit_bufs[type], "%" PRIu64, limits[type]); |
17ae2780 | 899 | else |
64faf04c | 900 | xsprintf(limit_bufs[type], "%s", limits[type] == CGROUP_LIMIT_MAX ? "max" : "0"); |
64faf04c TH |
901 | |
902 | xsprintf(buf, "%u:%u rbps=%s wbps=%s riops=%s wiops=%s\n", major(dev), minor(dev), | |
903 | limit_bufs[CGROUP_IO_RBPS_MAX], limit_bufs[CGROUP_IO_WBPS_MAX], | |
904 | limit_bufs[CGROUP_IO_RIOPS_MAX], limit_bufs[CGROUP_IO_WIOPS_MAX]); | |
293d32df | 905 | (void) set_attribute_and_warn(u, "io", "io.max", buf); |
64faf04c TH |
906 | } |
907 | ||
17ae2780 | 908 | static void cgroup_apply_blkio_device_limit(Unit *u, const char *dev_path, uint64_t rbps, uint64_t wbps) { |
64faf04c TH |
909 | char buf[DECIMAL_STR_MAX(dev_t)*2+2+DECIMAL_STR_MAX(uint64_t)+1]; |
910 | dev_t dev; | |
64faf04c TH |
911 | int r; |
912 | ||
913 | r = lookup_block_device(dev_path, &dev); | |
914 | if (r < 0) | |
17ae2780 | 915 | return; |
64faf04c | 916 | |
64faf04c | 917 | sprintf(buf, "%u:%u %" PRIu64 "\n", major(dev), minor(dev), rbps); |
293d32df | 918 | (void) set_attribute_and_warn(u, "blkio", "blkio.throttle.read_bps_device", buf); |
64faf04c | 919 | |
64faf04c | 920 | sprintf(buf, "%u:%u %" PRIu64 "\n", major(dev), minor(dev), wbps); |
293d32df | 921 | (void) set_attribute_and_warn(u, "blkio", "blkio.throttle.write_bps_device", buf); |
64faf04c TH |
922 | } |
923 | ||
c52db42b CD |
924 | static bool unit_has_unified_memory_config(Unit *u) { |
925 | CGroupContext *c; | |
926 | ||
927 | assert(u); | |
928 | ||
929 | c = unit_get_cgroup_context(u); | |
930 | assert(c); | |
931 | ||
7c9d2b79 | 932 | return unit_get_ancestor_memory_min(u) > 0 || unit_get_ancestor_memory_low(u) > 0 || |
c52db42b CD |
933 | c->memory_high != CGROUP_LIMIT_MAX || c->memory_max != CGROUP_LIMIT_MAX || |
934 | c->memory_swap_max != CGROUP_LIMIT_MAX; | |
da4d897e TH |
935 | } |
936 | ||
f29ff115 | 937 | static void cgroup_apply_unified_memory_limit(Unit *u, const char *file, uint64_t v) { |
589a5f7a | 938 | char buf[DECIMAL_STR_MAX(uint64_t) + 1] = "max\n"; |
da4d897e TH |
939 | |
940 | if (v != CGROUP_LIMIT_MAX) | |
941 | xsprintf(buf, "%" PRIu64 "\n", v); | |
942 | ||
293d32df | 943 | (void) set_attribute_and_warn(u, "memory", file, buf); |
da4d897e TH |
944 | } |
945 | ||
0f2d84d2 | 946 | static void cgroup_apply_firewall(Unit *u) { |
0f2d84d2 LP |
947 | assert(u); |
948 | ||
acf7f253 | 949 | /* Best-effort: let's apply IP firewalling and/or accounting if that's enabled */ |
906c06f6 | 950 | |
acf7f253 | 951 | if (bpf_firewall_compile(u) < 0) |
906c06f6 DM |
952 | return; |
953 | ||
fab34748 | 954 | (void) bpf_firewall_load_custom(u); |
906c06f6 | 955 | (void) bpf_firewall_install(u); |
906c06f6 DM |
956 | } |
957 | ||
8b139557 ZJS |
958 | static int cgroup_apply_devices(Unit *u) { |
959 | _cleanup_(bpf_program_unrefp) BPFProgram *prog = NULL; | |
960 | const char *path; | |
961 | CGroupContext *c; | |
962 | CGroupDeviceAllow *a; | |
45669ae2 | 963 | CGroupDevicePolicy policy; |
8b139557 ZJS |
964 | int r; |
965 | ||
966 | assert_se(c = unit_get_cgroup_context(u)); | |
967 | assert_se(path = u->cgroup_path); | |
968 | ||
45669ae2 ZJS |
969 | policy = c->device_policy; |
970 | ||
8b139557 | 971 | if (cg_all_unified() > 0) { |
45669ae2 | 972 | r = bpf_devices_cgroup_init(&prog, policy, c->device_allow); |
8b139557 ZJS |
973 | if (r < 0) |
974 | return log_unit_warning_errno(u, r, "Failed to initialize device control bpf program: %m"); | |
975 | ||
976 | } else { | |
977 | /* Changing the devices list of a populated cgroup might result in EINVAL, hence ignore | |
978 | * EINVAL here. */ | |
979 | ||
45669ae2 | 980 | if (c->device_allow || policy != CGROUP_DEVICE_POLICY_AUTO) |
8b139557 ZJS |
981 | r = cg_set_attribute("devices", path, "devices.deny", "a"); |
982 | else | |
983 | r = cg_set_attribute("devices", path, "devices.allow", "a"); | |
984 | if (r < 0) | |
985 | log_unit_full(u, IN_SET(r, -ENOENT, -EROFS, -EINVAL, -EACCES, -EPERM) ? LOG_DEBUG : LOG_WARNING, r, | |
986 | "Failed to reset devices.allow/devices.deny: %m"); | |
987 | } | |
988 | ||
45669ae2 ZJS |
989 | bool whitelist_static = policy == CGROUP_DEVICE_POLICY_CLOSED || |
990 | (policy == CGROUP_DEVICE_POLICY_AUTO && c->device_allow); | |
991 | if (whitelist_static) | |
8b139557 ZJS |
992 | (void) bpf_devices_whitelist_static(prog, path); |
993 | ||
45669ae2 | 994 | bool any = whitelist_static; |
8b139557 ZJS |
995 | LIST_FOREACH(device_allow, a, c->device_allow) { |
996 | char acc[4], *val; | |
997 | unsigned k = 0; | |
998 | ||
999 | if (a->r) | |
1000 | acc[k++] = 'r'; | |
1001 | if (a->w) | |
1002 | acc[k++] = 'w'; | |
1003 | if (a->m) | |
1004 | acc[k++] = 'm'; | |
8b139557 ZJS |
1005 | if (k == 0) |
1006 | continue; | |
8b139557 ZJS |
1007 | acc[k++] = 0; |
1008 | ||
1009 | if (path_startswith(a->path, "/dev/")) | |
45669ae2 | 1010 | r = bpf_devices_whitelist_device(prog, path, a->path, acc); |
8b139557 | 1011 | else if ((val = startswith(a->path, "block-"))) |
45669ae2 | 1012 | r = bpf_devices_whitelist_major(prog, path, val, 'b', acc); |
8b139557 | 1013 | else if ((val = startswith(a->path, "char-"))) |
45669ae2 ZJS |
1014 | r = bpf_devices_whitelist_major(prog, path, val, 'c', acc); |
1015 | else { | |
8b139557 | 1016 | log_unit_debug(u, "Ignoring device '%s' while writing cgroup attribute.", a->path); |
45669ae2 ZJS |
1017 | continue; |
1018 | } | |
1019 | ||
1020 | if (r >= 0) | |
1021 | any = true; | |
1022 | } | |
1023 | ||
1024 | if (prog && !any) { | |
1025 | log_unit_warning_errno(u, SYNTHETIC_ERRNO(ENODEV), "No devices matched by device filter."); | |
1026 | ||
1027 | /* The kernel verifier would reject a program we would build with the normal intro and outro | |
1028 | but no whitelisting rules (outro would contain an unreachable instruction for successful | |
1029 | return). */ | |
1030 | policy = CGROUP_DEVICE_POLICY_STRICT; | |
8b139557 ZJS |
1031 | } |
1032 | ||
45669ae2 | 1033 | r = bpf_devices_apply_policy(prog, policy, any, path, &u->bpf_device_control_installed); |
8b139557 ZJS |
1034 | if (r < 0) { |
1035 | static bool warned = false; | |
1036 | ||
1037 | log_full_errno(warned ? LOG_DEBUG : LOG_WARNING, r, | |
1038 | "Unit %s configures device ACL, but the local system doesn't seem to support the BPF-based device controller.\n" | |
1039 | "Proceeding WITHOUT applying ACL (all devices will be accessible)!\n" | |
1040 | "(This warning is only shown for the first loaded unit using device ACL.)", u->id); | |
1041 | ||
1042 | warned = true; | |
1043 | } | |
1044 | return r; | |
1045 | } | |
1046 | ||
906c06f6 DM |
1047 | static void cgroup_context_apply( |
1048 | Unit *u, | |
1049 | CGroupMask apply_mask, | |
906c06f6 DM |
1050 | ManagerState state) { |
1051 | ||
f29ff115 TH |
1052 | const char *path; |
1053 | CGroupContext *c; | |
52fecf20 | 1054 | bool is_host_root, is_local_root; |
4ad49000 LP |
1055 | int r; |
1056 | ||
f29ff115 TH |
1057 | assert(u); |
1058 | ||
906c06f6 | 1059 | /* Nothing to do? Exit early! */ |
17f14955 | 1060 | if (apply_mask == 0) |
4ad49000 | 1061 | return; |
8e274523 | 1062 | |
52fecf20 LP |
1063 | /* Some cgroup attributes are not supported on the host root cgroup, hence silently ignore them here. And other |
1064 | * attributes should only be managed for cgroups further down the tree. */ | |
1065 | is_local_root = unit_has_name(u, SPECIAL_ROOT_SLICE); | |
1066 | is_host_root = unit_has_host_root_cgroup(u); | |
f3725e64 LP |
1067 | |
1068 | assert_se(c = unit_get_cgroup_context(u)); | |
1069 | assert_se(path = u->cgroup_path); | |
1070 | ||
52fecf20 | 1071 | if (is_local_root) /* Make sure we don't try to display messages with an empty path. */ |
6da13913 | 1072 | path = "/"; |
01efdf13 | 1073 | |
be2c0327 LP |
1074 | /* We generally ignore errors caused by read-only mounted cgroup trees (assuming we are running in a container |
1075 | * then), and missing cgroups, i.e. EROFS and ENOENT. */ | |
714e2e1d | 1076 | |
be2c0327 LP |
1077 | /* In fully unified mode these attributes don't exist on the host cgroup root. On legacy the weights exist, but |
1078 | * setting the weight makes very little sense on the host root cgroup, as there are no other cgroups at this | |
1079 | * level. The quota exists there too, but any attempt to write to it is refused with EINVAL. Inside of | |
4e1dfa45 | 1080 | * containers we want to leave control of these to the container manager (and if cgroup v2 delegation is used |
be2c0327 LP |
1081 | * we couldn't even write to them if we wanted to). */ |
1082 | if ((apply_mask & CGROUP_MASK_CPU) && !is_local_root) { | |
8e274523 | 1083 | |
b4cccbc1 | 1084 | if (cg_all_unified() > 0) { |
be2c0327 | 1085 | uint64_t weight; |
b2f8b02e | 1086 | |
be2c0327 LP |
1087 | if (cgroup_context_has_cpu_weight(c)) |
1088 | weight = cgroup_context_cpu_weight(c, state); | |
1089 | else if (cgroup_context_has_cpu_shares(c)) { | |
1090 | uint64_t shares; | |
66ebf6c0 | 1091 | |
be2c0327 LP |
1092 | shares = cgroup_context_cpu_shares(c, state); |
1093 | weight = cgroup_cpu_shares_to_weight(shares); | |
66ebf6c0 | 1094 | |
be2c0327 LP |
1095 | log_cgroup_compat(u, "Applying [Startup]CPUShares=%" PRIu64 " as [Startup]CPUWeight=%" PRIu64 " on %s", |
1096 | shares, weight, path); | |
1097 | } else | |
1098 | weight = CGROUP_WEIGHT_DEFAULT; | |
66ebf6c0 | 1099 | |
be2c0327 | 1100 | cgroup_apply_unified_cpu_weight(u, weight); |
10f28641 | 1101 | cgroup_apply_unified_cpu_quota(u, c->cpu_quota_per_sec_usec, c->cpu_quota_period_usec); |
66ebf6c0 | 1102 | |
52fecf20 | 1103 | } else { |
be2c0327 | 1104 | uint64_t shares; |
52fecf20 | 1105 | |
be2c0327 LP |
1106 | if (cgroup_context_has_cpu_weight(c)) { |
1107 | uint64_t weight; | |
52fecf20 | 1108 | |
be2c0327 LP |
1109 | weight = cgroup_context_cpu_weight(c, state); |
1110 | shares = cgroup_cpu_weight_to_shares(weight); | |
52fecf20 | 1111 | |
be2c0327 LP |
1112 | log_cgroup_compat(u, "Applying [Startup]CPUWeight=%" PRIu64 " as [Startup]CPUShares=%" PRIu64 " on %s", |
1113 | weight, shares, path); | |
1114 | } else if (cgroup_context_has_cpu_shares(c)) | |
1115 | shares = cgroup_context_cpu_shares(c, state); | |
1116 | else | |
1117 | shares = CGROUP_CPU_SHARES_DEFAULT; | |
66ebf6c0 | 1118 | |
be2c0327 | 1119 | cgroup_apply_legacy_cpu_shares(u, shares); |
10f28641 | 1120 | cgroup_apply_legacy_cpu_quota(u, c->cpu_quota_per_sec_usec, c->cpu_quota_period_usec); |
66ebf6c0 | 1121 | } |
4ad49000 LP |
1122 | } |
1123 | ||
047f5d63 | 1124 | if ((apply_mask & CGROUP_MASK_CPUSET) && !is_local_root) { |
2cea199e ZJS |
1125 | cgroup_apply_unified_cpuset(u, &c->cpuset_cpus, "cpuset.cpus"); |
1126 | cgroup_apply_unified_cpuset(u, &c->cpuset_mems, "cpuset.mems"); | |
047f5d63 PH |
1127 | } |
1128 | ||
4e1dfa45 | 1129 | /* The 'io' controller attributes are not exported on the host's root cgroup (being a pure cgroup v2 |
52fecf20 LP |
1130 | * controller), and in case of containers we want to leave control of these attributes to the container manager |
1131 | * (and we couldn't access that stuff anyway, even if we tried if proper delegation is used). */ | |
1132 | if ((apply_mask & CGROUP_MASK_IO) && !is_local_root) { | |
1133 | char buf[8+DECIMAL_STR_MAX(uint64_t)+1]; | |
1134 | bool has_io, has_blockio; | |
1135 | uint64_t weight; | |
13c31542 | 1136 | |
52fecf20 LP |
1137 | has_io = cgroup_context_has_io_config(c); |
1138 | has_blockio = cgroup_context_has_blockio_config(c); | |
13c31542 | 1139 | |
52fecf20 LP |
1140 | if (has_io) |
1141 | weight = cgroup_context_io_weight(c, state); | |
1142 | else if (has_blockio) { | |
1143 | uint64_t blkio_weight; | |
128fadc9 | 1144 | |
52fecf20 LP |
1145 | blkio_weight = cgroup_context_blkio_weight(c, state); |
1146 | weight = cgroup_weight_blkio_to_io(blkio_weight); | |
128fadc9 | 1147 | |
67e2ea15 | 1148 | log_cgroup_compat(u, "Applying [Startup]BlockIOWeight=%" PRIu64 " as [Startup]IOWeight=%" PRIu64, |
52fecf20 LP |
1149 | blkio_weight, weight); |
1150 | } else | |
1151 | weight = CGROUP_WEIGHT_DEFAULT; | |
13c31542 | 1152 | |
52fecf20 LP |
1153 | xsprintf(buf, "default %" PRIu64 "\n", weight); |
1154 | (void) set_attribute_and_warn(u, "io", "io.weight", buf); | |
538b4852 | 1155 | |
2dbc45ae KK |
1156 | /* FIXME: drop this when distro kernels properly support BFQ through "io.weight" |
1157 | * See also: https://github.com/systemd/systemd/pull/13335 */ | |
1158 | xsprintf(buf, "%" PRIu64 "\n", weight); | |
1159 | (void) set_attribute_and_warn(u, "io", "io.bfq.weight", buf); | |
1160 | ||
52fecf20 LP |
1161 | if (has_io) { |
1162 | CGroupIODeviceLatency *latency; | |
1163 | CGroupIODeviceLimit *limit; | |
1164 | CGroupIODeviceWeight *w; | |
128fadc9 | 1165 | |
52fecf20 LP |
1166 | LIST_FOREACH(device_weights, w, c->io_device_weights) |
1167 | cgroup_apply_io_device_weight(u, w->path, w->weight); | |
128fadc9 | 1168 | |
52fecf20 LP |
1169 | LIST_FOREACH(device_limits, limit, c->io_device_limits) |
1170 | cgroup_apply_io_device_limit(u, limit->path, limit->limits); | |
6ae4283c | 1171 | |
52fecf20 LP |
1172 | LIST_FOREACH(device_latencies, latency, c->io_device_latencies) |
1173 | cgroup_apply_io_device_latency(u, latency->path, latency->target_usec); | |
6ae4283c | 1174 | |
52fecf20 LP |
1175 | } else if (has_blockio) { |
1176 | CGroupBlockIODeviceWeight *w; | |
1177 | CGroupBlockIODeviceBandwidth *b; | |
13c31542 | 1178 | |
52fecf20 LP |
1179 | LIST_FOREACH(device_weights, w, c->blockio_device_weights) { |
1180 | weight = cgroup_weight_blkio_to_io(w->weight); | |
17ae2780 | 1181 | |
67e2ea15 | 1182 | log_cgroup_compat(u, "Applying BlockIODeviceWeight=%" PRIu64 " as IODeviceWeight=%" PRIu64 " for %s", |
52fecf20 | 1183 | w->weight, weight, w->path); |
538b4852 | 1184 | |
52fecf20 LP |
1185 | cgroup_apply_io_device_weight(u, w->path, weight); |
1186 | } | |
538b4852 | 1187 | |
17ae2780 | 1188 | LIST_FOREACH(device_bandwidths, b, c->blockio_device_bandwidths) { |
538b4852 TH |
1189 | uint64_t limits[_CGROUP_IO_LIMIT_TYPE_MAX]; |
1190 | CGroupIOLimitType type; | |
1191 | ||
1192 | for (type = 0; type < _CGROUP_IO_LIMIT_TYPE_MAX; type++) | |
1193 | limits[type] = cgroup_io_limit_defaults[type]; | |
1194 | ||
1195 | limits[CGROUP_IO_RBPS_MAX] = b->rbps; | |
1196 | limits[CGROUP_IO_WBPS_MAX] = b->wbps; | |
1197 | ||
67e2ea15 | 1198 | log_cgroup_compat(u, "Applying BlockIO{Read|Write}Bandwidth=%" PRIu64 " %" PRIu64 " as IO{Read|Write}BandwidthMax= for %s", |
128fadc9 TH |
1199 | b->rbps, b->wbps, b->path); |
1200 | ||
17ae2780 | 1201 | cgroup_apply_io_device_limit(u, b->path, limits); |
538b4852 | 1202 | } |
13c31542 TH |
1203 | } |
1204 | } | |
1205 | ||
906c06f6 | 1206 | if (apply_mask & CGROUP_MASK_BLKIO) { |
52fecf20 | 1207 | bool has_io, has_blockio; |
4ad49000 | 1208 | |
52fecf20 LP |
1209 | has_io = cgroup_context_has_io_config(c); |
1210 | has_blockio = cgroup_context_has_blockio_config(c); | |
1211 | ||
1212 | /* Applying a 'weight' never makes sense for the host root cgroup, and for containers this should be | |
1213 | * left to our container manager, too. */ | |
1214 | if (!is_local_root) { | |
64faf04c TH |
1215 | char buf[DECIMAL_STR_MAX(uint64_t)+1]; |
1216 | uint64_t weight; | |
64faf04c | 1217 | |
7d862ab8 | 1218 | if (has_io) { |
52fecf20 | 1219 | uint64_t io_weight; |
128fadc9 | 1220 | |
52fecf20 | 1221 | io_weight = cgroup_context_io_weight(c, state); |
538b4852 | 1222 | weight = cgroup_weight_io_to_blkio(cgroup_context_io_weight(c, state)); |
128fadc9 | 1223 | |
67e2ea15 | 1224 | log_cgroup_compat(u, "Applying [Startup]IOWeight=%" PRIu64 " as [Startup]BlockIOWeight=%" PRIu64, |
128fadc9 | 1225 | io_weight, weight); |
7d862ab8 TH |
1226 | } else if (has_blockio) |
1227 | weight = cgroup_context_blkio_weight(c, state); | |
1228 | else | |
538b4852 | 1229 | weight = CGROUP_BLKIO_WEIGHT_DEFAULT; |
64faf04c TH |
1230 | |
1231 | xsprintf(buf, "%" PRIu64 "\n", weight); | |
293d32df | 1232 | (void) set_attribute_and_warn(u, "blkio", "blkio.weight", buf); |
4ad49000 | 1233 | |
7d862ab8 | 1234 | if (has_io) { |
538b4852 TH |
1235 | CGroupIODeviceWeight *w; |
1236 | ||
128fadc9 TH |
1237 | LIST_FOREACH(device_weights, w, c->io_device_weights) { |
1238 | weight = cgroup_weight_io_to_blkio(w->weight); | |
1239 | ||
67e2ea15 | 1240 | log_cgroup_compat(u, "Applying IODeviceWeight=%" PRIu64 " as BlockIODeviceWeight=%" PRIu64 " for %s", |
128fadc9 TH |
1241 | w->weight, weight, w->path); |
1242 | ||
1243 | cgroup_apply_blkio_device_weight(u, w->path, weight); | |
1244 | } | |
7d862ab8 TH |
1245 | } else if (has_blockio) { |
1246 | CGroupBlockIODeviceWeight *w; | |
1247 | ||
7d862ab8 TH |
1248 | LIST_FOREACH(device_weights, w, c->blockio_device_weights) |
1249 | cgroup_apply_blkio_device_weight(u, w->path, w->weight); | |
538b4852 | 1250 | } |
4ad49000 LP |
1251 | } |
1252 | ||
5238e957 | 1253 | /* The bandwidth limits are something that make sense to be applied to the host's root but not container |
52fecf20 LP |
1254 | * roots, as there we want the container manager to handle it */ |
1255 | if (is_host_root || !is_local_root) { | |
1256 | if (has_io) { | |
1257 | CGroupIODeviceLimit *l; | |
538b4852 | 1258 | |
52fecf20 | 1259 | LIST_FOREACH(device_limits, l, c->io_device_limits) { |
67e2ea15 | 1260 | log_cgroup_compat(u, "Applying IO{Read|Write}Bandwidth=%" PRIu64 " %" PRIu64 " as BlockIO{Read|Write}BandwidthMax= for %s", |
52fecf20 | 1261 | l->limits[CGROUP_IO_RBPS_MAX], l->limits[CGROUP_IO_WBPS_MAX], l->path); |
128fadc9 | 1262 | |
52fecf20 LP |
1263 | cgroup_apply_blkio_device_limit(u, l->path, l->limits[CGROUP_IO_RBPS_MAX], l->limits[CGROUP_IO_WBPS_MAX]); |
1264 | } | |
1265 | } else if (has_blockio) { | |
1266 | CGroupBlockIODeviceBandwidth *b; | |
7d862ab8 | 1267 | |
52fecf20 LP |
1268 | LIST_FOREACH(device_bandwidths, b, c->blockio_device_bandwidths) |
1269 | cgroup_apply_blkio_device_limit(u, b->path, b->rbps, b->wbps); | |
1270 | } | |
d686d8a9 | 1271 | } |
8e274523 LP |
1272 | } |
1273 | ||
be2c0327 LP |
1274 | /* In unified mode 'memory' attributes do not exist on the root cgroup. In legacy mode 'memory.limit_in_bytes' |
1275 | * exists on the root cgroup, but any writes to it are refused with EINVAL. And if we run in a container we | |
4e1dfa45 | 1276 | * want to leave control to the container manager (and if proper cgroup v2 delegation is used we couldn't even |
be2c0327 LP |
1277 | * write to this if we wanted to.) */ |
1278 | if ((apply_mask & CGROUP_MASK_MEMORY) && !is_local_root) { | |
efdb0237 | 1279 | |
52fecf20 | 1280 | if (cg_all_unified() > 0) { |
be2c0327 LP |
1281 | uint64_t max, swap_max = CGROUP_LIMIT_MAX; |
1282 | ||
c52db42b | 1283 | if (unit_has_unified_memory_config(u)) { |
be2c0327 LP |
1284 | max = c->memory_max; |
1285 | swap_max = c->memory_swap_max; | |
1286 | } else { | |
1287 | max = c->memory_limit; | |
efdb0237 | 1288 | |
be2c0327 LP |
1289 | if (max != CGROUP_LIMIT_MAX) |
1290 | log_cgroup_compat(u, "Applying MemoryLimit=%" PRIu64 " as MemoryMax=", max); | |
128fadc9 | 1291 | } |
da4d897e | 1292 | |
64fe532e | 1293 | cgroup_apply_unified_memory_limit(u, "memory.min", unit_get_ancestor_memory_min(u)); |
c52db42b | 1294 | cgroup_apply_unified_memory_limit(u, "memory.low", unit_get_ancestor_memory_low(u)); |
be2c0327 LP |
1295 | cgroup_apply_unified_memory_limit(u, "memory.high", c->memory_high); |
1296 | cgroup_apply_unified_memory_limit(u, "memory.max", max); | |
1297 | cgroup_apply_unified_memory_limit(u, "memory.swap.max", swap_max); | |
128fadc9 | 1298 | |
afcfaa69 LP |
1299 | (void) set_attribute_and_warn(u, "memory", "memory.oom.group", one_zero(c->memory_oom_group)); |
1300 | ||
be2c0327 LP |
1301 | } else { |
1302 | char buf[DECIMAL_STR_MAX(uint64_t) + 1]; | |
1303 | uint64_t val; | |
52fecf20 | 1304 | |
c52db42b | 1305 | if (unit_has_unified_memory_config(u)) { |
be2c0327 LP |
1306 | val = c->memory_max; |
1307 | log_cgroup_compat(u, "Applying MemoryMax=%" PRIi64 " as MemoryLimit=", val); | |
1308 | } else | |
1309 | val = c->memory_limit; | |
78a4ee59 | 1310 | |
be2c0327 LP |
1311 | if (val == CGROUP_LIMIT_MAX) |
1312 | strncpy(buf, "-1\n", sizeof(buf)); | |
1313 | else | |
1314 | xsprintf(buf, "%" PRIu64 "\n", val); | |
1315 | ||
1316 | (void) set_attribute_and_warn(u, "memory", "memory.limit_in_bytes", buf); | |
da4d897e | 1317 | } |
4ad49000 | 1318 | } |
8e274523 | 1319 | |
4e1dfa45 | 1320 | /* On cgroup v2 we can apply BPF everywhere. On cgroup v1 we apply it everywhere except for the root of |
52fecf20 LP |
1321 | * containers, where we leave this to the manager */ |
1322 | if ((apply_mask & (CGROUP_MASK_DEVICES | CGROUP_MASK_BPF_DEVICES)) && | |
8b139557 ZJS |
1323 | (is_host_root || cg_all_unified() > 0 || !is_local_root)) |
1324 | (void) cgroup_apply_devices(u); | |
03a7b521 | 1325 | |
00b5974f LP |
1326 | if (apply_mask & CGROUP_MASK_PIDS) { |
1327 | ||
52fecf20 | 1328 | if (is_host_root) { |
00b5974f LP |
1329 | /* So, the "pids" controller does not expose anything on the root cgroup, in order not to |
1330 | * replicate knobs exposed elsewhere needlessly. We abstract this away here however, and when | |
1331 | * the knobs of the root cgroup are modified propagate this to the relevant sysctls. There's a | |
1332 | * non-obvious asymmetry however: unlike the cgroup properties we don't really want to take | |
1333 | * exclusive ownership of the sysctls, but we still want to honour things if the user sets | |
1334 | * limits. Hence we employ sort of a one-way strategy: when the user sets a bounded limit | |
1335 | * through us it counts. When the user afterwards unsets it again (i.e. sets it to unbounded) | |
1336 | * it also counts. But if the user never set a limit through us (i.e. we are the default of | |
1337 | * "unbounded") we leave things unmodified. For this we manage a global boolean that we turn on | |
1338 | * the first time we set a limit. Note that this boolean is flushed out on manager reload, | |
5238e957 | 1339 | * which is desirable so that there's an official way to release control of the sysctl from |
00b5974f LP |
1340 | * systemd: set the limit to unbounded and reload. */ |
1341 | ||
1342 | if (c->tasks_max != CGROUP_LIMIT_MAX) { | |
1343 | u->manager->sysctl_pid_max_changed = true; | |
1344 | r = procfs_tasks_set_limit(c->tasks_max); | |
1345 | } else if (u->manager->sysctl_pid_max_changed) | |
1346 | r = procfs_tasks_set_limit(TASKS_MAX); | |
1347 | else | |
1348 | r = 0; | |
00b5974f | 1349 | if (r < 0) |
39b9fefb | 1350 | log_unit_full(u, LOG_LEVEL_CGROUP_WRITE(r), r, |
00b5974f | 1351 | "Failed to write to tasks limit sysctls: %m"); |
52fecf20 | 1352 | } |
03a7b521 | 1353 | |
52fecf20 LP |
1354 | /* The attribute itself is not available on the host root cgroup, and in the container case we want to |
1355 | * leave it for the container manager. */ | |
1356 | if (!is_local_root) { | |
00b5974f LP |
1357 | if (c->tasks_max != CGROUP_LIMIT_MAX) { |
1358 | char buf[DECIMAL_STR_MAX(uint64_t) + 2]; | |
03a7b521 | 1359 | |
00b5974f | 1360 | sprintf(buf, "%" PRIu64 "\n", c->tasks_max); |
293d32df | 1361 | (void) set_attribute_and_warn(u, "pids", "pids.max", buf); |
00b5974f | 1362 | } else |
589a5f7a | 1363 | (void) set_attribute_and_warn(u, "pids", "pids.max", "max\n"); |
00b5974f | 1364 | } |
03a7b521 | 1365 | } |
906c06f6 | 1366 | |
17f14955 | 1367 | if (apply_mask & CGROUP_MASK_BPF_FIREWALL) |
0f2d84d2 | 1368 | cgroup_apply_firewall(u); |
fb385181 LP |
1369 | } |
1370 | ||
16492445 LP |
1371 | static bool unit_get_needs_bpf_firewall(Unit *u) { |
1372 | CGroupContext *c; | |
1373 | Unit *p; | |
1374 | assert(u); | |
1375 | ||
1376 | c = unit_get_cgroup_context(u); | |
1377 | if (!c) | |
1378 | return false; | |
1379 | ||
1380 | if (c->ip_accounting || | |
1381 | c->ip_address_allow || | |
fab34748 KL |
1382 | c->ip_address_deny || |
1383 | c->ip_filters_ingress || | |
1384 | c->ip_filters_egress) | |
16492445 LP |
1385 | return true; |
1386 | ||
1387 | /* If any parent slice has an IP access list defined, it applies too */ | |
1388 | for (p = UNIT_DEREF(u->slice); p; p = UNIT_DEREF(p->slice)) { | |
1389 | c = unit_get_cgroup_context(p); | |
1390 | if (!c) | |
1391 | return false; | |
1392 | ||
1393 | if (c->ip_address_allow || | |
1394 | c->ip_address_deny) | |
1395 | return true; | |
1396 | } | |
1397 | ||
1398 | return false; | |
1399 | } | |
1400 | ||
c52db42b | 1401 | static CGroupMask unit_get_cgroup_mask(Unit *u) { |
efdb0237 | 1402 | CGroupMask mask = 0; |
c52db42b CD |
1403 | CGroupContext *c; |
1404 | ||
1405 | assert(u); | |
1406 | ||
1407 | c = unit_get_cgroup_context(u); | |
8e274523 | 1408 | |
c710d3b4 CD |
1409 | assert(c); |
1410 | ||
fae9bc29 | 1411 | /* Figure out which controllers we need, based on the cgroup context object */ |
8e274523 | 1412 | |
fae9bc29 | 1413 | if (c->cpu_accounting) |
f98c2585 | 1414 | mask |= get_cpu_accounting_mask(); |
fae9bc29 LP |
1415 | |
1416 | if (cgroup_context_has_cpu_weight(c) || | |
66ebf6c0 | 1417 | cgroup_context_has_cpu_shares(c) || |
3a43da28 | 1418 | c->cpu_quota_per_sec_usec != USEC_INFINITY) |
fae9bc29 | 1419 | mask |= CGROUP_MASK_CPU; |
ecedd90f | 1420 | |
047f5d63 PH |
1421 | if (c->cpuset_cpus.set || c->cpuset_mems.set) |
1422 | mask |= CGROUP_MASK_CPUSET; | |
1423 | ||
538b4852 TH |
1424 | if (cgroup_context_has_io_config(c) || cgroup_context_has_blockio_config(c)) |
1425 | mask |= CGROUP_MASK_IO | CGROUP_MASK_BLKIO; | |
ecedd90f | 1426 | |
4ad49000 | 1427 | if (c->memory_accounting || |
da4d897e | 1428 | c->memory_limit != CGROUP_LIMIT_MAX || |
c52db42b | 1429 | unit_has_unified_memory_config(u)) |
efdb0237 | 1430 | mask |= CGROUP_MASK_MEMORY; |
8e274523 | 1431 | |
a931ad47 | 1432 | if (c->device_allow || |
084870f9 | 1433 | c->device_policy != CGROUP_DEVICE_POLICY_AUTO) |
084c7007 | 1434 | mask |= CGROUP_MASK_DEVICES | CGROUP_MASK_BPF_DEVICES; |
4ad49000 | 1435 | |
03a7b521 | 1436 | if (c->tasks_accounting || |
8793fa25 | 1437 | c->tasks_max != CGROUP_LIMIT_MAX) |
03a7b521 LP |
1438 | mask |= CGROUP_MASK_PIDS; |
1439 | ||
fae9bc29 | 1440 | return CGROUP_MASK_EXTEND_JOINED(mask); |
8e274523 LP |
1441 | } |
1442 | ||
53aea74a | 1443 | static CGroupMask unit_get_bpf_mask(Unit *u) { |
17f14955 RG |
1444 | CGroupMask mask = 0; |
1445 | ||
fae9bc29 LP |
1446 | /* Figure out which controllers we need, based on the cgroup context, possibly taking into account children |
1447 | * too. */ | |
1448 | ||
17f14955 RG |
1449 | if (unit_get_needs_bpf_firewall(u)) |
1450 | mask |= CGROUP_MASK_BPF_FIREWALL; | |
1451 | ||
1452 | return mask; | |
1453 | } | |
1454 | ||
efdb0237 | 1455 | CGroupMask unit_get_own_mask(Unit *u) { |
4ad49000 | 1456 | CGroupContext *c; |
8e274523 | 1457 | |
442ce775 LP |
1458 | /* Returns the mask of controllers the unit needs for itself. If a unit is not properly loaded, return an empty |
1459 | * mask, as we shouldn't reflect it in the cgroup hierarchy then. */ | |
1460 | ||
1461 | if (u->load_state != UNIT_LOADED) | |
1462 | return 0; | |
efdb0237 | 1463 | |
4ad49000 LP |
1464 | c = unit_get_cgroup_context(u); |
1465 | if (!c) | |
1466 | return 0; | |
8e274523 | 1467 | |
c52db42b | 1468 | return (unit_get_cgroup_mask(u) | unit_get_bpf_mask(u) | unit_get_delegate_mask(u)) & ~unit_get_ancestor_disable_mask(u); |
02638280 LP |
1469 | } |
1470 | ||
1471 | CGroupMask unit_get_delegate_mask(Unit *u) { | |
1472 | CGroupContext *c; | |
1473 | ||
1474 | /* If delegation is turned on, then turn on selected controllers, unless we are on the legacy hierarchy and the | |
1475 | * process we fork into is known to drop privileges, and hence shouldn't get access to the controllers. | |
19af675e | 1476 | * |
02638280 | 1477 | * Note that on the unified hierarchy it is safe to delegate controllers to unprivileged services. */ |
a931ad47 | 1478 | |
1d9cc876 | 1479 | if (!unit_cgroup_delegate(u)) |
02638280 LP |
1480 | return 0; |
1481 | ||
1482 | if (cg_all_unified() <= 0) { | |
a931ad47 LP |
1483 | ExecContext *e; |
1484 | ||
1485 | e = unit_get_exec_context(u); | |
02638280 LP |
1486 | if (e && !exec_context_maintains_privileges(e)) |
1487 | return 0; | |
a931ad47 LP |
1488 | } |
1489 | ||
1d9cc876 | 1490 | assert_se(c = unit_get_cgroup_context(u)); |
fae9bc29 | 1491 | return CGROUP_MASK_EXTEND_JOINED(c->delegate_controllers); |
8e274523 LP |
1492 | } |
1493 | ||
efdb0237 | 1494 | CGroupMask unit_get_members_mask(Unit *u) { |
4ad49000 | 1495 | assert(u); |
bc432dc7 | 1496 | |
02638280 | 1497 | /* Returns the mask of controllers all of the unit's children require, merged */ |
efdb0237 | 1498 | |
bc432dc7 | 1499 | if (u->cgroup_members_mask_valid) |
26a17ca2 | 1500 | return u->cgroup_members_mask; /* Use cached value if possible */ |
bc432dc7 | 1501 | |
64e844e5 | 1502 | u->cgroup_members_mask = 0; |
bc432dc7 LP |
1503 | |
1504 | if (u->type == UNIT_SLICE) { | |
eef85c4a | 1505 | void *v; |
bc432dc7 LP |
1506 | Unit *member; |
1507 | Iterator i; | |
1508 | ||
eef85c4a | 1509 | HASHMAP_FOREACH_KEY(v, member, u->dependencies[UNIT_BEFORE], i) { |
cb5e3bc3 CD |
1510 | if (UNIT_DEREF(member->slice) == u) |
1511 | u->cgroup_members_mask |= unit_get_subtree_mask(member); /* note that this calls ourselves again, for the children */ | |
bc432dc7 LP |
1512 | } |
1513 | } | |
1514 | ||
1515 | u->cgroup_members_mask_valid = true; | |
6414b7c9 | 1516 | return u->cgroup_members_mask; |
246aa6dd LP |
1517 | } |
1518 | ||
efdb0237 | 1519 | CGroupMask unit_get_siblings_mask(Unit *u) { |
4ad49000 | 1520 | assert(u); |
246aa6dd | 1521 | |
efdb0237 LP |
1522 | /* Returns the mask of controllers all of the unit's siblings |
1523 | * require, i.e. the members mask of the unit's parent slice | |
1524 | * if there is one. */ | |
1525 | ||
bc432dc7 | 1526 | if (UNIT_ISSET(u->slice)) |
637f421e | 1527 | return unit_get_members_mask(UNIT_DEREF(u->slice)); |
4ad49000 | 1528 | |
64e844e5 | 1529 | return unit_get_subtree_mask(u); /* we are the top-level slice */ |
246aa6dd LP |
1530 | } |
1531 | ||
4f6f62e4 CD |
1532 | CGroupMask unit_get_disable_mask(Unit *u) { |
1533 | CGroupContext *c; | |
1534 | ||
1535 | c = unit_get_cgroup_context(u); | |
1536 | if (!c) | |
1537 | return 0; | |
1538 | ||
1539 | return c->disable_controllers; | |
1540 | } | |
1541 | ||
1542 | CGroupMask unit_get_ancestor_disable_mask(Unit *u) { | |
1543 | CGroupMask mask; | |
1544 | ||
1545 | assert(u); | |
1546 | mask = unit_get_disable_mask(u); | |
1547 | ||
1548 | /* Returns the mask of controllers which are marked as forcibly | |
1549 | * disabled in any ancestor unit or the unit in question. */ | |
1550 | ||
1551 | if (UNIT_ISSET(u->slice)) | |
1552 | mask |= unit_get_ancestor_disable_mask(UNIT_DEREF(u->slice)); | |
1553 | ||
1554 | return mask; | |
1555 | } | |
1556 | ||
efdb0237 LP |
1557 | CGroupMask unit_get_subtree_mask(Unit *u) { |
1558 | ||
1559 | /* Returns the mask of this subtree, meaning of the group | |
1560 | * itself and its children. */ | |
1561 | ||
1562 | return unit_get_own_mask(u) | unit_get_members_mask(u); | |
1563 | } | |
1564 | ||
1565 | CGroupMask unit_get_target_mask(Unit *u) { | |
1566 | CGroupMask mask; | |
1567 | ||
1568 | /* This returns the cgroup mask of all controllers to enable | |
1569 | * for a specific cgroup, i.e. everything it needs itself, | |
1570 | * plus all that its children need, plus all that its siblings | |
1571 | * need. This is primarily useful on the legacy cgroup | |
1572 | * hierarchy, where we need to duplicate each cgroup in each | |
1573 | * hierarchy that shall be enabled for it. */ | |
6414b7c9 | 1574 | |
efdb0237 | 1575 | mask = unit_get_own_mask(u) | unit_get_members_mask(u) | unit_get_siblings_mask(u); |
84d2744b ZJS |
1576 | |
1577 | if (mask & CGROUP_MASK_BPF_FIREWALL & ~u->manager->cgroup_supported) | |
1578 | emit_bpf_firewall_warning(u); | |
1579 | ||
efdb0237 | 1580 | mask &= u->manager->cgroup_supported; |
c72703e2 | 1581 | mask &= ~unit_get_ancestor_disable_mask(u); |
efdb0237 LP |
1582 | |
1583 | return mask; | |
1584 | } | |
1585 | ||
1586 | CGroupMask unit_get_enable_mask(Unit *u) { | |
1587 | CGroupMask mask; | |
1588 | ||
1589 | /* This returns the cgroup mask of all controllers to enable | |
1590 | * for the children of a specific cgroup. This is primarily | |
1591 | * useful for the unified cgroup hierarchy, where each cgroup | |
1592 | * controls which controllers are enabled for its children. */ | |
1593 | ||
1594 | mask = unit_get_members_mask(u); | |
6414b7c9 | 1595 | mask &= u->manager->cgroup_supported; |
c72703e2 | 1596 | mask &= ~unit_get_ancestor_disable_mask(u); |
6414b7c9 DS |
1597 | |
1598 | return mask; | |
1599 | } | |
1600 | ||
5af88058 | 1601 | void unit_invalidate_cgroup_members_masks(Unit *u) { |
bc432dc7 LP |
1602 | assert(u); |
1603 | ||
5af88058 LP |
1604 | /* Recurse invalidate the member masks cache all the way up the tree */ |
1605 | u->cgroup_members_mask_valid = false; | |
bc432dc7 | 1606 | |
5af88058 LP |
1607 | if (UNIT_ISSET(u->slice)) |
1608 | unit_invalidate_cgroup_members_masks(UNIT_DEREF(u->slice)); | |
6414b7c9 DS |
1609 | } |
1610 | ||
6592b975 | 1611 | const char *unit_get_realized_cgroup_path(Unit *u, CGroupMask mask) { |
03b90d4b | 1612 | |
6592b975 | 1613 | /* Returns the realized cgroup path of the specified unit where all specified controllers are available. */ |
03b90d4b LP |
1614 | |
1615 | while (u) { | |
6592b975 | 1616 | |
03b90d4b LP |
1617 | if (u->cgroup_path && |
1618 | u->cgroup_realized && | |
d94a24ca | 1619 | FLAGS_SET(u->cgroup_realized_mask, mask)) |
03b90d4b LP |
1620 | return u->cgroup_path; |
1621 | ||
1622 | u = UNIT_DEREF(u->slice); | |
1623 | } | |
1624 | ||
1625 | return NULL; | |
1626 | } | |
1627 | ||
6592b975 LP |
1628 | static const char *migrate_callback(CGroupMask mask, void *userdata) { |
1629 | return unit_get_realized_cgroup_path(userdata, mask); | |
1630 | } | |
1631 | ||
303ee601 | 1632 | char *unit_default_cgroup_path(const Unit *u) { |
efdb0237 LP |
1633 | _cleanup_free_ char *escaped = NULL, *slice = NULL; |
1634 | int r; | |
1635 | ||
1636 | assert(u); | |
1637 | ||
1638 | if (unit_has_name(u, SPECIAL_ROOT_SLICE)) | |
1639 | return strdup(u->manager->cgroup_root); | |
1640 | ||
1641 | if (UNIT_ISSET(u->slice) && !unit_has_name(UNIT_DEREF(u->slice), SPECIAL_ROOT_SLICE)) { | |
1642 | r = cg_slice_to_path(UNIT_DEREF(u->slice)->id, &slice); | |
1643 | if (r < 0) | |
1644 | return NULL; | |
1645 | } | |
1646 | ||
1647 | escaped = cg_escape(u->id); | |
1648 | if (!escaped) | |
1649 | return NULL; | |
1650 | ||
657ee2d8 | 1651 | return path_join(empty_to_root(u->manager->cgroup_root), slice, escaped); |
efdb0237 LP |
1652 | } |
1653 | ||
1654 | int unit_set_cgroup_path(Unit *u, const char *path) { | |
1655 | _cleanup_free_ char *p = NULL; | |
1656 | int r; | |
1657 | ||
1658 | assert(u); | |
1659 | ||
5210387e LP |
1660 | if (streq_ptr(u->cgroup_path, path)) |
1661 | return 0; | |
1662 | ||
efdb0237 LP |
1663 | if (path) { |
1664 | p = strdup(path); | |
1665 | if (!p) | |
1666 | return -ENOMEM; | |
5210387e | 1667 | } |
efdb0237 LP |
1668 | |
1669 | if (p) { | |
1670 | r = hashmap_put(u->manager->cgroup_unit, p, u); | |
1671 | if (r < 0) | |
1672 | return r; | |
1673 | } | |
1674 | ||
1675 | unit_release_cgroup(u); | |
ae2a15bc | 1676 | u->cgroup_path = TAKE_PTR(p); |
efdb0237 LP |
1677 | |
1678 | return 1; | |
1679 | } | |
1680 | ||
1681 | int unit_watch_cgroup(Unit *u) { | |
ab2c3861 | 1682 | _cleanup_free_ char *events = NULL; |
efdb0237 LP |
1683 | int r; |
1684 | ||
1685 | assert(u); | |
1686 | ||
0bb814c2 LP |
1687 | /* Watches the "cgroups.events" attribute of this unit's cgroup for "empty" events, but only if |
1688 | * cgroupv2 is available. */ | |
1689 | ||
efdb0237 LP |
1690 | if (!u->cgroup_path) |
1691 | return 0; | |
1692 | ||
0bb814c2 | 1693 | if (u->cgroup_control_inotify_wd >= 0) |
efdb0237 LP |
1694 | return 0; |
1695 | ||
1696 | /* Only applies to the unified hierarchy */ | |
c22800e4 | 1697 | r = cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER); |
b4cccbc1 LP |
1698 | if (r < 0) |
1699 | return log_error_errno(r, "Failed to determine whether the name=systemd hierarchy is unified: %m"); | |
1700 | if (r == 0) | |
efdb0237 LP |
1701 | return 0; |
1702 | ||
0bb814c2 | 1703 | /* No point in watch the top-level slice, it's never going to run empty. */ |
efdb0237 LP |
1704 | if (unit_has_name(u, SPECIAL_ROOT_SLICE)) |
1705 | return 0; | |
1706 | ||
0bb814c2 | 1707 | r = hashmap_ensure_allocated(&u->manager->cgroup_control_inotify_wd_unit, &trivial_hash_ops); |
efdb0237 LP |
1708 | if (r < 0) |
1709 | return log_oom(); | |
1710 | ||
ab2c3861 | 1711 | r = cg_get_path(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path, "cgroup.events", &events); |
efdb0237 LP |
1712 | if (r < 0) |
1713 | return log_oom(); | |
1714 | ||
0bb814c2 LP |
1715 | u->cgroup_control_inotify_wd = inotify_add_watch(u->manager->cgroup_inotify_fd, events, IN_MODIFY); |
1716 | if (u->cgroup_control_inotify_wd < 0) { | |
efdb0237 | 1717 | |
0bb814c2 LP |
1718 | if (errno == ENOENT) /* If the directory is already gone we don't need to track it, so this |
1719 | * is not an error */ | |
efdb0237 LP |
1720 | return 0; |
1721 | ||
0bb814c2 | 1722 | return log_unit_error_errno(u, errno, "Failed to add control inotify watch descriptor for control group %s: %m", u->cgroup_path); |
efdb0237 LP |
1723 | } |
1724 | ||
0bb814c2 | 1725 | r = hashmap_put(u->manager->cgroup_control_inotify_wd_unit, INT_TO_PTR(u->cgroup_control_inotify_wd), u); |
efdb0237 | 1726 | if (r < 0) |
0bb814c2 | 1727 | return log_unit_error_errno(u, r, "Failed to add control inotify watch descriptor to hash map: %m"); |
efdb0237 LP |
1728 | |
1729 | return 0; | |
1730 | } | |
1731 | ||
afcfaa69 LP |
1732 | int unit_watch_cgroup_memory(Unit *u) { |
1733 | _cleanup_free_ char *events = NULL; | |
1734 | CGroupContext *c; | |
1735 | int r; | |
1736 | ||
1737 | assert(u); | |
1738 | ||
1739 | /* Watches the "memory.events" attribute of this unit's cgroup for "oom_kill" events, but only if | |
1740 | * cgroupv2 is available. */ | |
1741 | ||
1742 | if (!u->cgroup_path) | |
1743 | return 0; | |
1744 | ||
1745 | c = unit_get_cgroup_context(u); | |
1746 | if (!c) | |
1747 | return 0; | |
1748 | ||
1749 | /* The "memory.events" attribute is only available if the memory controller is on. Let's hence tie | |
1750 | * this to memory accounting, in a way watching for OOM kills is a form of memory accounting after | |
1751 | * all. */ | |
1752 | if (!c->memory_accounting) | |
1753 | return 0; | |
1754 | ||
1755 | /* Don't watch inner nodes, as the kernel doesn't report oom_kill events recursively currently, and | |
1756 | * we also don't want to generate a log message for each parent cgroup of a process. */ | |
1757 | if (u->type == UNIT_SLICE) | |
1758 | return 0; | |
1759 | ||
1760 | if (u->cgroup_memory_inotify_wd >= 0) | |
1761 | return 0; | |
1762 | ||
1763 | /* Only applies to the unified hierarchy */ | |
1764 | r = cg_all_unified(); | |
1765 | if (r < 0) | |
1766 | return log_error_errno(r, "Failed to determine whether the memory controller is unified: %m"); | |
1767 | if (r == 0) | |
1768 | return 0; | |
1769 | ||
1770 | r = hashmap_ensure_allocated(&u->manager->cgroup_memory_inotify_wd_unit, &trivial_hash_ops); | |
1771 | if (r < 0) | |
1772 | return log_oom(); | |
1773 | ||
1774 | r = cg_get_path(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path, "memory.events", &events); | |
1775 | if (r < 0) | |
1776 | return log_oom(); | |
1777 | ||
1778 | u->cgroup_memory_inotify_wd = inotify_add_watch(u->manager->cgroup_inotify_fd, events, IN_MODIFY); | |
1779 | if (u->cgroup_memory_inotify_wd < 0) { | |
1780 | ||
1781 | if (errno == ENOENT) /* If the directory is already gone we don't need to track it, so this | |
1782 | * is not an error */ | |
1783 | return 0; | |
1784 | ||
1785 | return log_unit_error_errno(u, errno, "Failed to add memory inotify watch descriptor for control group %s: %m", u->cgroup_path); | |
1786 | } | |
1787 | ||
1788 | r = hashmap_put(u->manager->cgroup_memory_inotify_wd_unit, INT_TO_PTR(u->cgroup_memory_inotify_wd), u); | |
1789 | if (r < 0) | |
1790 | return log_unit_error_errno(u, r, "Failed to add memory inotify watch descriptor to hash map: %m"); | |
1791 | ||
1792 | return 0; | |
1793 | } | |
1794 | ||
a4634b21 LP |
1795 | int unit_pick_cgroup_path(Unit *u) { |
1796 | _cleanup_free_ char *path = NULL; | |
1797 | int r; | |
1798 | ||
1799 | assert(u); | |
1800 | ||
1801 | if (u->cgroup_path) | |
1802 | return 0; | |
1803 | ||
1804 | if (!UNIT_HAS_CGROUP_CONTEXT(u)) | |
1805 | return -EINVAL; | |
1806 | ||
1807 | path = unit_default_cgroup_path(u); | |
1808 | if (!path) | |
1809 | return log_oom(); | |
1810 | ||
1811 | r = unit_set_cgroup_path(u, path); | |
1812 | if (r == -EEXIST) | |
1813 | return log_unit_error_errno(u, r, "Control group %s exists already.", path); | |
1814 | if (r < 0) | |
1815 | return log_unit_error_errno(u, r, "Failed to set unit's control group path to %s: %m", path); | |
1816 | ||
1817 | return 0; | |
1818 | } | |
1819 | ||
efdb0237 LP |
1820 | static int unit_create_cgroup( |
1821 | Unit *u, | |
1822 | CGroupMask target_mask, | |
0d2d6fbf CD |
1823 | CGroupMask enable_mask, |
1824 | ManagerState state) { | |
efdb0237 | 1825 | |
65be7e06 | 1826 | bool created; |
27adcc97 | 1827 | int r; |
64747e2d | 1828 | |
4ad49000 | 1829 | assert(u); |
64747e2d | 1830 | |
27c4ed79 | 1831 | if (!UNIT_HAS_CGROUP_CONTEXT(u)) |
0cd385d3 LP |
1832 | return 0; |
1833 | ||
a4634b21 LP |
1834 | /* Figure out our cgroup path */ |
1835 | r = unit_pick_cgroup_path(u); | |
1836 | if (r < 0) | |
1837 | return r; | |
b58b8e11 | 1838 | |
03b90d4b | 1839 | /* First, create our own group */ |
efdb0237 | 1840 | r = cg_create_everywhere(u->manager->cgroup_supported, target_mask, u->cgroup_path); |
23bbb0de | 1841 | if (r < 0) |
efdb0237 | 1842 | return log_unit_error_errno(u, r, "Failed to create cgroup %s: %m", u->cgroup_path); |
490c5a37 | 1843 | created = r; |
efdb0237 LP |
1844 | |
1845 | /* Start watching it */ | |
1846 | (void) unit_watch_cgroup(u); | |
afcfaa69 | 1847 | (void) unit_watch_cgroup_memory(u); |
efdb0237 | 1848 | |
65be7e06 | 1849 | /* Preserve enabled controllers in delegated units, adjust others. */ |
1fd3a10c | 1850 | if (created || !u->cgroup_realized || !unit_cgroup_delegate(u)) { |
27adcc97 | 1851 | CGroupMask result_mask = 0; |
65be7e06 ZJS |
1852 | |
1853 | /* Enable all controllers we need */ | |
27adcc97 | 1854 | r = cg_enable_everywhere(u->manager->cgroup_supported, enable_mask, u->cgroup_path, &result_mask); |
65be7e06 | 1855 | if (r < 0) |
27adcc97 LP |
1856 | log_unit_warning_errno(u, r, "Failed to enable/disable controllers on cgroup %s, ignoring: %m", u->cgroup_path); |
1857 | ||
1858 | /* If we just turned off a controller, this might release the controller for our parent too, let's | |
1859 | * enqueue the parent for re-realization in that case again. */ | |
1860 | if (UNIT_ISSET(u->slice)) { | |
1861 | CGroupMask turned_off; | |
1862 | ||
1863 | turned_off = (u->cgroup_realized ? u->cgroup_enabled_mask & ~result_mask : 0); | |
1864 | if (turned_off != 0) { | |
1865 | Unit *parent; | |
1866 | ||
1867 | /* Force the parent to propagate the enable mask to the kernel again, by invalidating | |
1868 | * the controller we just turned off. */ | |
1869 | ||
1870 | for (parent = UNIT_DEREF(u->slice); parent; parent = UNIT_DEREF(parent->slice)) | |
1871 | unit_invalidate_cgroup(parent, turned_off); | |
1872 | } | |
1873 | } | |
1874 | ||
1875 | /* Remember what's actually enabled now */ | |
1876 | u->cgroup_enabled_mask = result_mask; | |
65be7e06 | 1877 | } |
03b90d4b LP |
1878 | |
1879 | /* Keep track that this is now realized */ | |
4ad49000 | 1880 | u->cgroup_realized = true; |
efdb0237 | 1881 | u->cgroup_realized_mask = target_mask; |
4ad49000 | 1882 | |
1d9cc876 | 1883 | if (u->type != UNIT_SLICE && !unit_cgroup_delegate(u)) { |
0cd385d3 LP |
1884 | |
1885 | /* Then, possibly move things over, but not if | |
1886 | * subgroups may contain processes, which is the case | |
1887 | * for slice and delegation units. */ | |
1888 | r = cg_migrate_everywhere(u->manager->cgroup_supported, u->cgroup_path, u->cgroup_path, migrate_callback, u); | |
1889 | if (r < 0) | |
efdb0237 | 1890 | log_unit_warning_errno(u, r, "Failed to migrate cgroup from to %s, ignoring: %m", u->cgroup_path); |
0cd385d3 | 1891 | } |
03b90d4b | 1892 | |
0d2d6fbf CD |
1893 | /* Set attributes */ |
1894 | cgroup_context_apply(u, target_mask, state); | |
1895 | cgroup_xattr_apply(u); | |
1896 | ||
64747e2d LP |
1897 | return 0; |
1898 | } | |
1899 | ||
6592b975 LP |
1900 | static int unit_attach_pid_to_cgroup_via_bus(Unit *u, pid_t pid, const char *suffix_path) { |
1901 | _cleanup_(sd_bus_error_free) sd_bus_error error = SD_BUS_ERROR_NULL; | |
1902 | char *pp; | |
7b3fd631 | 1903 | int r; |
6592b975 | 1904 | |
7b3fd631 LP |
1905 | assert(u); |
1906 | ||
6592b975 LP |
1907 | if (MANAGER_IS_SYSTEM(u->manager)) |
1908 | return -EINVAL; | |
1909 | ||
1910 | if (!u->manager->system_bus) | |
1911 | return -EIO; | |
1912 | ||
1913 | if (!u->cgroup_path) | |
1914 | return -EINVAL; | |
1915 | ||
1916 | /* Determine this unit's cgroup path relative to our cgroup root */ | |
1917 | pp = path_startswith(u->cgroup_path, u->manager->cgroup_root); | |
1918 | if (!pp) | |
1919 | return -EINVAL; | |
1920 | ||
1921 | pp = strjoina("/", pp, suffix_path); | |
858d36c1 | 1922 | path_simplify(pp, false); |
6592b975 LP |
1923 | |
1924 | r = sd_bus_call_method(u->manager->system_bus, | |
1925 | "org.freedesktop.systemd1", | |
1926 | "/org/freedesktop/systemd1", | |
1927 | "org.freedesktop.systemd1.Manager", | |
1928 | "AttachProcessesToUnit", | |
1929 | &error, NULL, | |
1930 | "ssau", | |
1931 | NULL /* empty unit name means client's unit, i.e. us */, pp, 1, (uint32_t) pid); | |
7b3fd631 | 1932 | if (r < 0) |
6592b975 LP |
1933 | return log_unit_debug_errno(u, r, "Failed to attach unit process " PID_FMT " via the bus: %s", pid, bus_error_message(&error, r)); |
1934 | ||
1935 | return 0; | |
1936 | } | |
1937 | ||
1938 | int unit_attach_pids_to_cgroup(Unit *u, Set *pids, const char *suffix_path) { | |
1939 | CGroupMask delegated_mask; | |
1940 | const char *p; | |
1941 | Iterator i; | |
1942 | void *pidp; | |
1943 | int r, q; | |
1944 | ||
1945 | assert(u); | |
1946 | ||
1947 | if (!UNIT_HAS_CGROUP_CONTEXT(u)) | |
1948 | return -EINVAL; | |
1949 | ||
1950 | if (set_isempty(pids)) | |
1951 | return 0; | |
7b3fd631 | 1952 | |
fab34748 KL |
1953 | /* Load any custom firewall BPF programs here once to test if they are existing and actually loadable. |
1954 | * Fail here early since later errors in the call chain unit_realize_cgroup to cgroup_context_apply are ignored. */ | |
1955 | r = bpf_firewall_load_custom(u); | |
1956 | if (r < 0) | |
1957 | return r; | |
1958 | ||
6592b975 | 1959 | r = unit_realize_cgroup(u); |
7b3fd631 LP |
1960 | if (r < 0) |
1961 | return r; | |
1962 | ||
6592b975 LP |
1963 | if (isempty(suffix_path)) |
1964 | p = u->cgroup_path; | |
1965 | else | |
270384b2 | 1966 | p = prefix_roota(u->cgroup_path, suffix_path); |
6592b975 LP |
1967 | |
1968 | delegated_mask = unit_get_delegate_mask(u); | |
1969 | ||
1970 | r = 0; | |
1971 | SET_FOREACH(pidp, pids, i) { | |
1972 | pid_t pid = PTR_TO_PID(pidp); | |
1973 | CGroupController c; | |
1974 | ||
1975 | /* First, attach the PID to the main cgroup hierarchy */ | |
1976 | q = cg_attach(SYSTEMD_CGROUP_CONTROLLER, p, pid); | |
1977 | if (q < 0) { | |
1978 | log_unit_debug_errno(u, q, "Couldn't move process " PID_FMT " to requested cgroup '%s': %m", pid, p); | |
1979 | ||
1980 | if (MANAGER_IS_USER(u->manager) && IN_SET(q, -EPERM, -EACCES)) { | |
1981 | int z; | |
1982 | ||
1983 | /* If we are in a user instance, and we can't move the process ourselves due to | |
1984 | * permission problems, let's ask the system instance about it instead. Since it's more | |
1985 | * privileged it might be able to move the process across the leaves of a subtree who's | |
1986 | * top node is not owned by us. */ | |
1987 | ||
1988 | z = unit_attach_pid_to_cgroup_via_bus(u, pid, suffix_path); | |
1989 | if (z < 0) | |
1990 | log_unit_debug_errno(u, z, "Couldn't move process " PID_FMT " to requested cgroup '%s' via the system bus either: %m", pid, p); | |
1991 | else | |
1992 | continue; /* When the bus thing worked via the bus we are fully done for this PID. */ | |
1993 | } | |
1994 | ||
1995 | if (r >= 0) | |
1996 | r = q; /* Remember first error */ | |
1997 | ||
1998 | continue; | |
1999 | } | |
2000 | ||
2001 | q = cg_all_unified(); | |
2002 | if (q < 0) | |
2003 | return q; | |
2004 | if (q > 0) | |
2005 | continue; | |
2006 | ||
2007 | /* In the legacy hierarchy, attach the process to the request cgroup if possible, and if not to the | |
2008 | * innermost realized one */ | |
2009 | ||
2010 | for (c = 0; c < _CGROUP_CONTROLLER_MAX; c++) { | |
2011 | CGroupMask bit = CGROUP_CONTROLLER_TO_MASK(c); | |
2012 | const char *realized; | |
2013 | ||
2014 | if (!(u->manager->cgroup_supported & bit)) | |
2015 | continue; | |
2016 | ||
2017 | /* If this controller is delegated and realized, honour the caller's request for the cgroup suffix. */ | |
2018 | if (delegated_mask & u->cgroup_realized_mask & bit) { | |
2019 | q = cg_attach(cgroup_controller_to_string(c), p, pid); | |
2020 | if (q >= 0) | |
2021 | continue; /* Success! */ | |
2022 | ||
2023 | log_unit_debug_errno(u, q, "Failed to attach PID " PID_FMT " to requested cgroup %s in controller %s, falling back to unit's cgroup: %m", | |
2024 | pid, p, cgroup_controller_to_string(c)); | |
2025 | } | |
2026 | ||
2027 | /* So this controller is either not delegate or realized, or something else weird happened. In | |
2028 | * that case let's attach the PID at least to the closest cgroup up the tree that is | |
2029 | * realized. */ | |
2030 | realized = unit_get_realized_cgroup_path(u, bit); | |
2031 | if (!realized) | |
2032 | continue; /* Not even realized in the root slice? Then let's not bother */ | |
2033 | ||
2034 | q = cg_attach(cgroup_controller_to_string(c), realized, pid); | |
2035 | if (q < 0) | |
2036 | log_unit_debug_errno(u, q, "Failed to attach PID " PID_FMT " to realized cgroup %s in controller %s, ignoring: %m", | |
2037 | pid, realized, cgroup_controller_to_string(c)); | |
2038 | } | |
2039 | } | |
2040 | ||
2041 | return r; | |
7b3fd631 LP |
2042 | } |
2043 | ||
906c06f6 DM |
2044 | static bool unit_has_mask_realized( |
2045 | Unit *u, | |
2046 | CGroupMask target_mask, | |
17f14955 | 2047 | CGroupMask enable_mask) { |
906c06f6 | 2048 | |
bc432dc7 LP |
2049 | assert(u); |
2050 | ||
d5095dcd LP |
2051 | /* Returns true if this unit is fully realized. We check four things: |
2052 | * | |
2053 | * 1. Whether the cgroup was created at all | |
4e1dfa45 CD |
2054 | * 2. Whether the cgroup was created in all the hierarchies we need it to be created in (in case of cgroup v1) |
2055 | * 3. Whether the cgroup has all the right controllers enabled (in case of cgroup v2) | |
d5095dcd LP |
2056 | * 4. Whether the invalidation mask is currently zero |
2057 | * | |
2058 | * If you wonder why we mask the target realization and enable mask with CGROUP_MASK_V1/CGROUP_MASK_V2: note | |
4e1dfa45 CD |
2059 | * that there are three sets of bitmasks: CGROUP_MASK_V1 (for real cgroup v1 controllers), CGROUP_MASK_V2 (for |
2060 | * real cgroup v2 controllers) and CGROUP_MASK_BPF (for BPF-based pseudo-controllers). Now, cgroup_realized_mask | |
2061 | * is only matters for cgroup v1 controllers, and cgroup_enabled_mask only used for cgroup v2, and if they | |
d5095dcd LP |
2062 | * differ in the others, we don't really care. (After all, the cgroup_enabled_mask tracks with controllers are |
2063 | * enabled through cgroup.subtree_control, and since the BPF pseudo-controllers don't show up there, they | |
2064 | * simply don't matter. */ | |
2065 | ||
906c06f6 | 2066 | return u->cgroup_realized && |
d5095dcd LP |
2067 | ((u->cgroup_realized_mask ^ target_mask) & CGROUP_MASK_V1) == 0 && |
2068 | ((u->cgroup_enabled_mask ^ enable_mask) & CGROUP_MASK_V2) == 0 && | |
17f14955 | 2069 | u->cgroup_invalidated_mask == 0; |
6414b7c9 DS |
2070 | } |
2071 | ||
4f6f62e4 CD |
2072 | static bool unit_has_mask_disables_realized( |
2073 | Unit *u, | |
2074 | CGroupMask target_mask, | |
2075 | CGroupMask enable_mask) { | |
2076 | ||
2077 | assert(u); | |
2078 | ||
2079 | /* Returns true if all controllers which should be disabled are indeed disabled. | |
2080 | * | |
2081 | * Unlike unit_has_mask_realized, we don't care what was enabled, only that anything we want to remove is | |
2082 | * already removed. */ | |
2083 | ||
2084 | return !u->cgroup_realized || | |
2085 | (FLAGS_SET(u->cgroup_realized_mask, target_mask & CGROUP_MASK_V1) && | |
2086 | FLAGS_SET(u->cgroup_enabled_mask, enable_mask & CGROUP_MASK_V2)); | |
2087 | } | |
2088 | ||
a57669d2 CD |
2089 | static bool unit_has_mask_enables_realized( |
2090 | Unit *u, | |
2091 | CGroupMask target_mask, | |
2092 | CGroupMask enable_mask) { | |
2093 | ||
2094 | assert(u); | |
2095 | ||
2096 | /* Returns true if all controllers which should be enabled are indeed enabled. | |
2097 | * | |
2098 | * Unlike unit_has_mask_realized, we don't care about the controllers that are not present, only that anything | |
2099 | * we want to add is already added. */ | |
2100 | ||
2101 | return u->cgroup_realized && | |
c72703e2 CD |
2102 | ((u->cgroup_realized_mask | target_mask) & CGROUP_MASK_V1) == (u->cgroup_realized_mask & CGROUP_MASK_V1) && |
2103 | ((u->cgroup_enabled_mask | enable_mask) & CGROUP_MASK_V2) == (u->cgroup_enabled_mask & CGROUP_MASK_V2); | |
a57669d2 CD |
2104 | } |
2105 | ||
27adcc97 | 2106 | void unit_add_to_cgroup_realize_queue(Unit *u) { |
2aa57a65 LP |
2107 | assert(u); |
2108 | ||
2109 | if (u->in_cgroup_realize_queue) | |
2110 | return; | |
2111 | ||
2112 | LIST_PREPEND(cgroup_realize_queue, u->manager->cgroup_realize_queue, u); | |
2113 | u->in_cgroup_realize_queue = true; | |
2114 | } | |
2115 | ||
2116 | static void unit_remove_from_cgroup_realize_queue(Unit *u) { | |
2117 | assert(u); | |
2118 | ||
2119 | if (!u->in_cgroup_realize_queue) | |
2120 | return; | |
2121 | ||
2122 | LIST_REMOVE(cgroup_realize_queue, u->manager->cgroup_realize_queue, u); | |
2123 | u->in_cgroup_realize_queue = false; | |
2124 | } | |
2125 | ||
a57669d2 CD |
2126 | /* Controllers can only be enabled breadth-first, from the root of the |
2127 | * hierarchy downwards to the unit in question. */ | |
2128 | static int unit_realize_cgroup_now_enable(Unit *u, ManagerState state) { | |
2129 | CGroupMask target_mask, enable_mask, new_target_mask, new_enable_mask; | |
2130 | int r; | |
2131 | ||
2132 | assert(u); | |
2133 | ||
2134 | /* First go deal with this unit's parent, or we won't be able to enable | |
2135 | * any new controllers at this layer. */ | |
2136 | if (UNIT_ISSET(u->slice)) { | |
2137 | r = unit_realize_cgroup_now_enable(UNIT_DEREF(u->slice), state); | |
2138 | if (r < 0) | |
2139 | return r; | |
2140 | } | |
2141 | ||
2142 | target_mask = unit_get_target_mask(u); | |
2143 | enable_mask = unit_get_enable_mask(u); | |
2144 | ||
2145 | /* We can only enable in this direction, don't try to disable anything. | |
2146 | */ | |
2147 | if (unit_has_mask_enables_realized(u, target_mask, enable_mask)) | |
2148 | return 0; | |
2149 | ||
2150 | new_target_mask = u->cgroup_realized_mask | target_mask; | |
2151 | new_enable_mask = u->cgroup_enabled_mask | enable_mask; | |
2152 | ||
c72703e2 | 2153 | return unit_create_cgroup(u, new_target_mask, new_enable_mask, state); |
a57669d2 CD |
2154 | } |
2155 | ||
4f6f62e4 CD |
2156 | /* Controllers can only be disabled depth-first, from the leaves of the |
2157 | * hierarchy upwards to the unit in question. */ | |
2158 | static int unit_realize_cgroup_now_disable(Unit *u, ManagerState state) { | |
2159 | Iterator i; | |
2160 | Unit *m; | |
2161 | void *v; | |
2162 | ||
2163 | assert(u); | |
2164 | ||
2165 | if (u->type != UNIT_SLICE) | |
2166 | return 0; | |
2167 | ||
2168 | HASHMAP_FOREACH_KEY(v, m, u->dependencies[UNIT_BEFORE], i) { | |
2169 | CGroupMask target_mask, enable_mask, new_target_mask, new_enable_mask; | |
2170 | int r; | |
2171 | ||
2172 | if (UNIT_DEREF(m->slice) != u) | |
2173 | continue; | |
2174 | ||
2175 | /* The cgroup for this unit might not actually be fully | |
2176 | * realised yet, in which case it isn't holding any controllers | |
2177 | * open anyway. */ | |
2178 | if (!m->cgroup_path) | |
2179 | continue; | |
2180 | ||
2181 | /* We must disable those below us first in order to release the | |
2182 | * controller. */ | |
2183 | if (m->type == UNIT_SLICE) | |
2184 | (void) unit_realize_cgroup_now_disable(m, state); | |
2185 | ||
2186 | target_mask = unit_get_target_mask(m); | |
2187 | enable_mask = unit_get_enable_mask(m); | |
2188 | ||
2189 | /* We can only disable in this direction, don't try to enable | |
2190 | * anything. */ | |
2191 | if (unit_has_mask_disables_realized(m, target_mask, enable_mask)) | |
2192 | continue; | |
2193 | ||
2194 | new_target_mask = m->cgroup_realized_mask & target_mask; | |
2195 | new_enable_mask = m->cgroup_enabled_mask & enable_mask; | |
2196 | ||
2197 | r = unit_create_cgroup(m, new_target_mask, new_enable_mask, state); | |
2198 | if (r < 0) | |
2199 | return r; | |
2200 | } | |
2201 | ||
2202 | return 0; | |
2203 | } | |
a57669d2 | 2204 | |
6414b7c9 DS |
2205 | /* Check if necessary controllers and attributes for a unit are in place. |
2206 | * | |
a57669d2 CD |
2207 | * - If so, do nothing. |
2208 | * - If not, create paths, move processes over, and set attributes. | |
2209 | * | |
2210 | * Controllers can only be *enabled* in a breadth-first way, and *disabled* in | |
2211 | * a depth-first way. As such the process looks like this: | |
2212 | * | |
2213 | * Suppose we have a cgroup hierarchy which looks like this: | |
2214 | * | |
2215 | * root | |
2216 | * / \ | |
2217 | * / \ | |
2218 | * / \ | |
2219 | * a b | |
2220 | * / \ / \ | |
2221 | * / \ / \ | |
2222 | * c d e f | |
2223 | * / \ / \ / \ / \ | |
2224 | * h i j k l m n o | |
2225 | * | |
2226 | * 1. We want to realise cgroup "d" now. | |
c72703e2 | 2227 | * 2. cgroup "a" has DisableControllers=cpu in the associated unit. |
a57669d2 CD |
2228 | * 3. cgroup "k" just started requesting the memory controller. |
2229 | * | |
2230 | * To make this work we must do the following in order: | |
2231 | * | |
2232 | * 1. Disable CPU controller in k, j | |
2233 | * 2. Disable CPU controller in d | |
2234 | * 3. Enable memory controller in root | |
2235 | * 4. Enable memory controller in a | |
2236 | * 5. Enable memory controller in d | |
2237 | * 6. Enable memory controller in k | |
2238 | * | |
2239 | * Notice that we need to touch j in one direction, but not the other. We also | |
2240 | * don't go beyond d when disabling -- it's up to "a" to get realized if it | |
2241 | * wants to disable further. The basic rules are therefore: | |
2242 | * | |
2243 | * - If you're disabling something, you need to realise all of the cgroups from | |
2244 | * your recursive descendants to the root. This starts from the leaves. | |
2245 | * - If you're enabling something, you need to realise from the root cgroup | |
2246 | * downwards, but you don't need to iterate your recursive descendants. | |
6414b7c9 DS |
2247 | * |
2248 | * Returns 0 on success and < 0 on failure. */ | |
db785129 | 2249 | static int unit_realize_cgroup_now(Unit *u, ManagerState state) { |
efdb0237 | 2250 | CGroupMask target_mask, enable_mask; |
6414b7c9 | 2251 | int r; |
64747e2d | 2252 | |
4ad49000 | 2253 | assert(u); |
64747e2d | 2254 | |
2aa57a65 | 2255 | unit_remove_from_cgroup_realize_queue(u); |
64747e2d | 2256 | |
efdb0237 | 2257 | target_mask = unit_get_target_mask(u); |
ccf78df1 TH |
2258 | enable_mask = unit_get_enable_mask(u); |
2259 | ||
17f14955 | 2260 | if (unit_has_mask_realized(u, target_mask, enable_mask)) |
0a1eb06d | 2261 | return 0; |
64747e2d | 2262 | |
4f6f62e4 CD |
2263 | /* Disable controllers below us, if there are any */ |
2264 | r = unit_realize_cgroup_now_disable(u, state); | |
2265 | if (r < 0) | |
2266 | return r; | |
2267 | ||
2268 | /* Enable controllers above us, if there are any */ | |
6414b7c9 | 2269 | if (UNIT_ISSET(u->slice)) { |
a57669d2 | 2270 | r = unit_realize_cgroup_now_enable(UNIT_DEREF(u->slice), state); |
6414b7c9 DS |
2271 | if (r < 0) |
2272 | return r; | |
2273 | } | |
4ad49000 | 2274 | |
0d2d6fbf CD |
2275 | /* Now actually deal with the cgroup we were trying to realise and set attributes */ |
2276 | r = unit_create_cgroup(u, target_mask, enable_mask, state); | |
6414b7c9 DS |
2277 | if (r < 0) |
2278 | return r; | |
2279 | ||
c2baf11c LP |
2280 | /* Now, reset the invalidation mask */ |
2281 | u->cgroup_invalidated_mask = 0; | |
6414b7c9 | 2282 | return 0; |
64747e2d LP |
2283 | } |
2284 | ||
91a6073e | 2285 | unsigned manager_dispatch_cgroup_realize_queue(Manager *m) { |
db785129 | 2286 | ManagerState state; |
4ad49000 | 2287 | unsigned n = 0; |
db785129 | 2288 | Unit *i; |
6414b7c9 | 2289 | int r; |
ecedd90f | 2290 | |
91a6073e LP |
2291 | assert(m); |
2292 | ||
db785129 LP |
2293 | state = manager_state(m); |
2294 | ||
91a6073e LP |
2295 | while ((i = m->cgroup_realize_queue)) { |
2296 | assert(i->in_cgroup_realize_queue); | |
ecedd90f | 2297 | |
2aa57a65 LP |
2298 | if (UNIT_IS_INACTIVE_OR_FAILED(unit_active_state(i))) { |
2299 | /* Maybe things changed, and the unit is not actually active anymore? */ | |
2300 | unit_remove_from_cgroup_realize_queue(i); | |
2301 | continue; | |
2302 | } | |
2303 | ||
db785129 | 2304 | r = unit_realize_cgroup_now(i, state); |
6414b7c9 | 2305 | if (r < 0) |
efdb0237 | 2306 | log_warning_errno(r, "Failed to realize cgroups for queued unit %s, ignoring: %m", i->id); |
0a1eb06d | 2307 | |
4ad49000 LP |
2308 | n++; |
2309 | } | |
ecedd90f | 2310 | |
4ad49000 | 2311 | return n; |
8e274523 LP |
2312 | } |
2313 | ||
91a6073e | 2314 | static void unit_add_siblings_to_cgroup_realize_queue(Unit *u) { |
4ad49000 | 2315 | Unit *slice; |
ca949c9d | 2316 | |
4ad49000 LP |
2317 | /* This adds the siblings of the specified unit and the |
2318 | * siblings of all parent units to the cgroup queue. (But | |
2319 | * neither the specified unit itself nor the parents.) */ | |
2320 | ||
2321 | while ((slice = UNIT_DEREF(u->slice))) { | |
2322 | Iterator i; | |
2323 | Unit *m; | |
eef85c4a | 2324 | void *v; |
8f53a7b8 | 2325 | |
eef85c4a | 2326 | HASHMAP_FOREACH_KEY(v, m, u->dependencies[UNIT_BEFORE], i) { |
6414b7c9 DS |
2327 | /* Skip units that have a dependency on the slice |
2328 | * but aren't actually in it. */ | |
4ad49000 | 2329 | if (UNIT_DEREF(m->slice) != slice) |
50159e6a | 2330 | continue; |
8e274523 | 2331 | |
6414b7c9 DS |
2332 | /* No point in doing cgroup application for units |
2333 | * without active processes. */ | |
2334 | if (UNIT_IS_INACTIVE_OR_FAILED(unit_active_state(m))) | |
2335 | continue; | |
2336 | ||
2337 | /* If the unit doesn't need any new controllers | |
2338 | * and has current ones realized, it doesn't need | |
2339 | * any changes. */ | |
906c06f6 DM |
2340 | if (unit_has_mask_realized(m, |
2341 | unit_get_target_mask(m), | |
17f14955 | 2342 | unit_get_enable_mask(m))) |
6414b7c9 DS |
2343 | continue; |
2344 | ||
91a6073e | 2345 | unit_add_to_cgroup_realize_queue(m); |
50159e6a LP |
2346 | } |
2347 | ||
4ad49000 | 2348 | u = slice; |
8e274523 | 2349 | } |
4ad49000 LP |
2350 | } |
2351 | ||
0a1eb06d | 2352 | int unit_realize_cgroup(Unit *u) { |
4ad49000 LP |
2353 | assert(u); |
2354 | ||
35b7ff80 | 2355 | if (!UNIT_HAS_CGROUP_CONTEXT(u)) |
0a1eb06d | 2356 | return 0; |
8e274523 | 2357 | |
4ad49000 LP |
2358 | /* So, here's the deal: when realizing the cgroups for this |
2359 | * unit, we need to first create all parents, but there's more | |
2360 | * actually: for the weight-based controllers we also need to | |
2361 | * make sure that all our siblings (i.e. units that are in the | |
73e231ab | 2362 | * same slice as we are) have cgroups, too. Otherwise, things |
4ad49000 LP |
2363 | * would become very uneven as each of their processes would |
2364 | * get as much resources as all our group together. This call | |
2365 | * will synchronously create the parent cgroups, but will | |
2366 | * defer work on the siblings to the next event loop | |
2367 | * iteration. */ | |
ca949c9d | 2368 | |
4ad49000 | 2369 | /* Add all sibling slices to the cgroup queue. */ |
91a6073e | 2370 | unit_add_siblings_to_cgroup_realize_queue(u); |
4ad49000 | 2371 | |
6414b7c9 | 2372 | /* And realize this one now (and apply the values) */ |
db785129 | 2373 | return unit_realize_cgroup_now(u, manager_state(u->manager)); |
8e274523 LP |
2374 | } |
2375 | ||
efdb0237 LP |
2376 | void unit_release_cgroup(Unit *u) { |
2377 | assert(u); | |
2378 | ||
8a0d5388 LP |
2379 | /* Forgets all cgroup details for this cgroup — but does *not* destroy the cgroup. This is hence OK to call |
2380 | * when we close down everything for reexecution, where we really want to leave the cgroup in place. */ | |
efdb0237 LP |
2381 | |
2382 | if (u->cgroup_path) { | |
2383 | (void) hashmap_remove(u->manager->cgroup_unit, u->cgroup_path); | |
2384 | u->cgroup_path = mfree(u->cgroup_path); | |
2385 | } | |
2386 | ||
0bb814c2 LP |
2387 | if (u->cgroup_control_inotify_wd >= 0) { |
2388 | if (inotify_rm_watch(u->manager->cgroup_inotify_fd, u->cgroup_control_inotify_wd) < 0) | |
2389 | 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 | 2390 | |
0bb814c2 LP |
2391 | (void) hashmap_remove(u->manager->cgroup_control_inotify_wd_unit, INT_TO_PTR(u->cgroup_control_inotify_wd)); |
2392 | u->cgroup_control_inotify_wd = -1; | |
efdb0237 | 2393 | } |
afcfaa69 LP |
2394 | |
2395 | if (u->cgroup_memory_inotify_wd >= 0) { | |
2396 | if (inotify_rm_watch(u->manager->cgroup_inotify_fd, u->cgroup_memory_inotify_wd) < 0) | |
2397 | 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); | |
2398 | ||
2399 | (void) hashmap_remove(u->manager->cgroup_memory_inotify_wd_unit, INT_TO_PTR(u->cgroup_memory_inotify_wd)); | |
2400 | u->cgroup_memory_inotify_wd = -1; | |
2401 | } | |
efdb0237 LP |
2402 | } |
2403 | ||
2404 | void unit_prune_cgroup(Unit *u) { | |
8e274523 | 2405 | int r; |
efdb0237 | 2406 | bool is_root_slice; |
8e274523 | 2407 | |
4ad49000 | 2408 | assert(u); |
8e274523 | 2409 | |
efdb0237 LP |
2410 | /* Removes the cgroup, if empty and possible, and stops watching it. */ |
2411 | ||
4ad49000 LP |
2412 | if (!u->cgroup_path) |
2413 | return; | |
8e274523 | 2414 | |
fe700f46 LP |
2415 | (void) unit_get_cpu_usage(u, NULL); /* Cache the last CPU usage value before we destroy the cgroup */ |
2416 | ||
efdb0237 LP |
2417 | is_root_slice = unit_has_name(u, SPECIAL_ROOT_SLICE); |
2418 | ||
2419 | r = cg_trim_everywhere(u->manager->cgroup_supported, u->cgroup_path, !is_root_slice); | |
0219b352 DB |
2420 | if (r < 0) |
2421 | /* One reason we could have failed here is, that the cgroup still contains a process. | |
2422 | * However, if the cgroup becomes removable at a later time, it might be removed when | |
2423 | * the containing slice is stopped. So even if we failed now, this unit shouldn't assume | |
2424 | * that the cgroup is still realized the next time it is started. Do not return early | |
2425 | * on error, continue cleanup. */ | |
2426 | log_unit_full(u, r == -EBUSY ? LOG_DEBUG : LOG_WARNING, r, "Failed to destroy cgroup %s, ignoring: %m", u->cgroup_path); | |
8e274523 | 2427 | |
efdb0237 LP |
2428 | if (is_root_slice) |
2429 | return; | |
2430 | ||
2431 | unit_release_cgroup(u); | |
0a1eb06d | 2432 | |
4ad49000 | 2433 | u->cgroup_realized = false; |
bc432dc7 | 2434 | u->cgroup_realized_mask = 0; |
ccf78df1 | 2435 | u->cgroup_enabled_mask = 0; |
084c7007 RG |
2436 | |
2437 | u->bpf_device_control_installed = bpf_program_unref(u->bpf_device_control_installed); | |
8e274523 LP |
2438 | } |
2439 | ||
efdb0237 | 2440 | int unit_search_main_pid(Unit *u, pid_t *ret) { |
4ad49000 | 2441 | _cleanup_fclose_ FILE *f = NULL; |
4d051546 | 2442 | pid_t pid = 0, npid; |
efdb0237 | 2443 | int r; |
4ad49000 LP |
2444 | |
2445 | assert(u); | |
efdb0237 | 2446 | assert(ret); |
4ad49000 LP |
2447 | |
2448 | if (!u->cgroup_path) | |
efdb0237 | 2449 | return -ENXIO; |
4ad49000 | 2450 | |
efdb0237 LP |
2451 | r = cg_enumerate_processes(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path, &f); |
2452 | if (r < 0) | |
2453 | return r; | |
4ad49000 | 2454 | |
4ad49000 | 2455 | while (cg_read_pid(f, &npid) > 0) { |
4ad49000 LP |
2456 | |
2457 | if (npid == pid) | |
2458 | continue; | |
8e274523 | 2459 | |
4d051546 | 2460 | if (pid_is_my_child(npid) == 0) |
4ad49000 | 2461 | continue; |
8e274523 | 2462 | |
efdb0237 | 2463 | if (pid != 0) |
4ad49000 LP |
2464 | /* Dang, there's more than one daemonized PID |
2465 | in this group, so we don't know what process | |
2466 | is the main process. */ | |
efdb0237 LP |
2467 | |
2468 | return -ENODATA; | |
8e274523 | 2469 | |
4ad49000 | 2470 | pid = npid; |
8e274523 LP |
2471 | } |
2472 | ||
efdb0237 LP |
2473 | *ret = pid; |
2474 | return 0; | |
2475 | } | |
2476 | ||
2477 | static int unit_watch_pids_in_path(Unit *u, const char *path) { | |
b3c5bad3 | 2478 | _cleanup_closedir_ DIR *d = NULL; |
efdb0237 LP |
2479 | _cleanup_fclose_ FILE *f = NULL; |
2480 | int ret = 0, r; | |
2481 | ||
2482 | assert(u); | |
2483 | assert(path); | |
2484 | ||
2485 | r = cg_enumerate_processes(SYSTEMD_CGROUP_CONTROLLER, path, &f); | |
2486 | if (r < 0) | |
2487 | ret = r; | |
2488 | else { | |
2489 | pid_t pid; | |
2490 | ||
2491 | while ((r = cg_read_pid(f, &pid)) > 0) { | |
f75f613d | 2492 | r = unit_watch_pid(u, pid, false); |
efdb0237 LP |
2493 | if (r < 0 && ret >= 0) |
2494 | ret = r; | |
2495 | } | |
2496 | ||
2497 | if (r < 0 && ret >= 0) | |
2498 | ret = r; | |
2499 | } | |
2500 | ||
2501 | r = cg_enumerate_subgroups(SYSTEMD_CGROUP_CONTROLLER, path, &d); | |
2502 | if (r < 0) { | |
2503 | if (ret >= 0) | |
2504 | ret = r; | |
2505 | } else { | |
2506 | char *fn; | |
2507 | ||
2508 | while ((r = cg_read_subgroup(d, &fn)) > 0) { | |
2509 | _cleanup_free_ char *p = NULL; | |
2510 | ||
95b21cff | 2511 | p = path_join(empty_to_root(path), fn); |
efdb0237 LP |
2512 | free(fn); |
2513 | ||
2514 | if (!p) | |
2515 | return -ENOMEM; | |
2516 | ||
2517 | r = unit_watch_pids_in_path(u, p); | |
2518 | if (r < 0 && ret >= 0) | |
2519 | ret = r; | |
2520 | } | |
2521 | ||
2522 | if (r < 0 && ret >= 0) | |
2523 | ret = r; | |
2524 | } | |
2525 | ||
2526 | return ret; | |
2527 | } | |
2528 | ||
11aef522 LP |
2529 | int unit_synthesize_cgroup_empty_event(Unit *u) { |
2530 | int r; | |
2531 | ||
2532 | assert(u); | |
2533 | ||
2534 | /* Enqueue a synthetic cgroup empty event if this unit doesn't watch any PIDs anymore. This is compatibility | |
2535 | * support for non-unified systems where notifications aren't reliable, and hence need to take whatever we can | |
2536 | * get as notification source as soon as we stopped having any useful PIDs to watch for. */ | |
2537 | ||
2538 | if (!u->cgroup_path) | |
2539 | return -ENOENT; | |
2540 | ||
2541 | r = cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER); | |
2542 | if (r < 0) | |
2543 | return r; | |
2544 | if (r > 0) /* On unified we have reliable notifications, and don't need this */ | |
2545 | return 0; | |
2546 | ||
2547 | if (!set_isempty(u->pids)) | |
2548 | return 0; | |
2549 | ||
2550 | unit_add_to_cgroup_empty_queue(u); | |
2551 | return 0; | |
2552 | } | |
2553 | ||
efdb0237 | 2554 | int unit_watch_all_pids(Unit *u) { |
b4cccbc1 LP |
2555 | int r; |
2556 | ||
efdb0237 LP |
2557 | assert(u); |
2558 | ||
2559 | /* Adds all PIDs from our cgroup to the set of PIDs we | |
2560 | * watch. This is a fallback logic for cases where we do not | |
2561 | * get reliable cgroup empty notifications: we try to use | |
2562 | * SIGCHLD as replacement. */ | |
2563 | ||
2564 | if (!u->cgroup_path) | |
2565 | return -ENOENT; | |
2566 | ||
c22800e4 | 2567 | r = cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER); |
b4cccbc1 LP |
2568 | if (r < 0) |
2569 | return r; | |
2570 | if (r > 0) /* On unified we can use proper notifications */ | |
efdb0237 LP |
2571 | return 0; |
2572 | ||
2573 | return unit_watch_pids_in_path(u, u->cgroup_path); | |
2574 | } | |
2575 | ||
09e24654 LP |
2576 | static int on_cgroup_empty_event(sd_event_source *s, void *userdata) { |
2577 | Manager *m = userdata; | |
2578 | Unit *u; | |
efdb0237 LP |
2579 | int r; |
2580 | ||
09e24654 LP |
2581 | assert(s); |
2582 | assert(m); | |
efdb0237 | 2583 | |
09e24654 LP |
2584 | u = m->cgroup_empty_queue; |
2585 | if (!u) | |
efdb0237 LP |
2586 | return 0; |
2587 | ||
09e24654 LP |
2588 | assert(u->in_cgroup_empty_queue); |
2589 | u->in_cgroup_empty_queue = false; | |
2590 | LIST_REMOVE(cgroup_empty_queue, m->cgroup_empty_queue, u); | |
2591 | ||
2592 | if (m->cgroup_empty_queue) { | |
2593 | /* More stuff queued, let's make sure we remain enabled */ | |
2594 | r = sd_event_source_set_enabled(s, SD_EVENT_ONESHOT); | |
2595 | if (r < 0) | |
19a691a9 | 2596 | log_debug_errno(r, "Failed to reenable cgroup empty event source, ignoring: %m"); |
09e24654 | 2597 | } |
efdb0237 LP |
2598 | |
2599 | unit_add_to_gc_queue(u); | |
2600 | ||
2601 | if (UNIT_VTABLE(u)->notify_cgroup_empty) | |
2602 | UNIT_VTABLE(u)->notify_cgroup_empty(u); | |
2603 | ||
2604 | return 0; | |
2605 | } | |
2606 | ||
09e24654 LP |
2607 | void unit_add_to_cgroup_empty_queue(Unit *u) { |
2608 | int r; | |
2609 | ||
2610 | assert(u); | |
2611 | ||
2612 | /* Note that there are four different ways how cgroup empty events reach us: | |
2613 | * | |
2614 | * 1. On the unified hierarchy we get an inotify event on the cgroup | |
2615 | * | |
2616 | * 2. On the legacy hierarchy, when running in system mode, we get a datagram on the cgroup agent socket | |
2617 | * | |
2618 | * 3. On the legacy hierarchy, when running in user mode, we get a D-Bus signal on the system bus | |
2619 | * | |
2620 | * 4. On the legacy hierarchy, in service units we start watching all processes of the cgroup for SIGCHLD as | |
2621 | * soon as we get one SIGCHLD, to deal with unreliable cgroup notifications. | |
2622 | * | |
2623 | * Regardless which way we got the notification, we'll verify it here, and then add it to a separate | |
2624 | * queue. This queue will be dispatched at a lower priority than the SIGCHLD handler, so that we always use | |
2625 | * SIGCHLD if we can get it first, and only use the cgroup empty notifications if there's no SIGCHLD pending | |
2626 | * (which might happen if the cgroup doesn't contain processes that are our own child, which is typically the | |
2627 | * case for scope units). */ | |
2628 | ||
2629 | if (u->in_cgroup_empty_queue) | |
2630 | return; | |
2631 | ||
2632 | /* Let's verify that the cgroup is really empty */ | |
2633 | if (!u->cgroup_path) | |
2634 | return; | |
2635 | r = cg_is_empty_recursive(SYSTEMD_CGROUP_CONTROLLER, u->cgroup_path); | |
2636 | if (r < 0) { | |
2637 | log_unit_debug_errno(u, r, "Failed to determine whether cgroup %s is empty: %m", u->cgroup_path); | |
2638 | return; | |
2639 | } | |
2640 | if (r == 0) | |
2641 | return; | |
2642 | ||
2643 | LIST_PREPEND(cgroup_empty_queue, u->manager->cgroup_empty_queue, u); | |
2644 | u->in_cgroup_empty_queue = true; | |
2645 | ||
2646 | /* Trigger the defer event */ | |
2647 | r = sd_event_source_set_enabled(u->manager->cgroup_empty_event_source, SD_EVENT_ONESHOT); | |
2648 | if (r < 0) | |
2649 | log_debug_errno(r, "Failed to enable cgroup empty event source: %m"); | |
2650 | } | |
2651 | ||
2ba6ae6b | 2652 | int unit_check_oom(Unit *u) { |
afcfaa69 LP |
2653 | _cleanup_free_ char *oom_kill = NULL; |
2654 | bool increased; | |
2655 | uint64_t c; | |
2656 | int r; | |
2657 | ||
2658 | if (!u->cgroup_path) | |
2659 | return 0; | |
2660 | ||
2661 | r = cg_get_keyed_attribute("memory", u->cgroup_path, "memory.events", STRV_MAKE("oom_kill"), &oom_kill); | |
2662 | if (r < 0) | |
2663 | return log_unit_debug_errno(u, r, "Failed to read oom_kill field of memory.events cgroup attribute: %m"); | |
2664 | ||
2665 | r = safe_atou64(oom_kill, &c); | |
2666 | if (r < 0) | |
2667 | return log_unit_debug_errno(u, r, "Failed to parse oom_kill field: %m"); | |
2668 | ||
2669 | increased = c > u->oom_kill_last; | |
2670 | u->oom_kill_last = c; | |
2671 | ||
2672 | if (!increased) | |
2673 | return 0; | |
2674 | ||
2675 | log_struct(LOG_NOTICE, | |
2676 | "MESSAGE_ID=" SD_MESSAGE_UNIT_OUT_OF_MEMORY_STR, | |
2677 | LOG_UNIT_ID(u), | |
2678 | LOG_UNIT_INVOCATION_ID(u), | |
2679 | LOG_UNIT_MESSAGE(u, "A process of this unit has been killed by the OOM killer.")); | |
2680 | ||
2681 | if (UNIT_VTABLE(u)->notify_cgroup_oom) | |
2682 | UNIT_VTABLE(u)->notify_cgroup_oom(u); | |
2683 | ||
2684 | return 1; | |
2685 | } | |
2686 | ||
2687 | static int on_cgroup_oom_event(sd_event_source *s, void *userdata) { | |
2688 | Manager *m = userdata; | |
2689 | Unit *u; | |
2690 | int r; | |
2691 | ||
2692 | assert(s); | |
2693 | assert(m); | |
2694 | ||
2695 | u = m->cgroup_oom_queue; | |
2696 | if (!u) | |
2697 | return 0; | |
2698 | ||
2699 | assert(u->in_cgroup_oom_queue); | |
2700 | u->in_cgroup_oom_queue = false; | |
2701 | LIST_REMOVE(cgroup_oom_queue, m->cgroup_oom_queue, u); | |
2702 | ||
2703 | if (m->cgroup_oom_queue) { | |
2704 | /* More stuff queued, let's make sure we remain enabled */ | |
2705 | r = sd_event_source_set_enabled(s, SD_EVENT_ONESHOT); | |
2706 | if (r < 0) | |
2707 | log_debug_errno(r, "Failed to reenable cgroup oom event source, ignoring: %m"); | |
2708 | } | |
2709 | ||
2710 | (void) unit_check_oom(u); | |
2711 | return 0; | |
2712 | } | |
2713 | ||
2714 | static void unit_add_to_cgroup_oom_queue(Unit *u) { | |
2715 | int r; | |
2716 | ||
2717 | assert(u); | |
2718 | ||
2719 | if (u->in_cgroup_oom_queue) | |
2720 | return; | |
2721 | if (!u->cgroup_path) | |
2722 | return; | |
2723 | ||
2724 | LIST_PREPEND(cgroup_oom_queue, u->manager->cgroup_oom_queue, u); | |
2725 | u->in_cgroup_oom_queue = true; | |
2726 | ||
2727 | /* Trigger the defer event */ | |
2728 | if (!u->manager->cgroup_oom_event_source) { | |
2729 | _cleanup_(sd_event_source_unrefp) sd_event_source *s = NULL; | |
2730 | ||
2731 | r = sd_event_add_defer(u->manager->event, &s, on_cgroup_oom_event, u->manager); | |
2732 | if (r < 0) { | |
2733 | log_error_errno(r, "Failed to create cgroup oom event source: %m"); | |
2734 | return; | |
2735 | } | |
2736 | ||
2737 | r = sd_event_source_set_priority(s, SD_EVENT_PRIORITY_NORMAL-8); | |
2738 | if (r < 0) { | |
2739 | log_error_errno(r, "Failed to set priority of cgroup oom event source: %m"); | |
2740 | return; | |
2741 | } | |
2742 | ||
2743 | (void) sd_event_source_set_description(s, "cgroup-oom"); | |
2744 | u->manager->cgroup_oom_event_source = TAKE_PTR(s); | |
2745 | } | |
2746 | ||
2747 | r = sd_event_source_set_enabled(u->manager->cgroup_oom_event_source, SD_EVENT_ONESHOT); | |
2748 | if (r < 0) | |
2749 | log_error_errno(r, "Failed to enable cgroup oom event source: %m"); | |
2750 | } | |
2751 | ||
efdb0237 LP |
2752 | static int on_cgroup_inotify_event(sd_event_source *s, int fd, uint32_t revents, void *userdata) { |
2753 | Manager *m = userdata; | |
2754 | ||
2755 | assert(s); | |
2756 | assert(fd >= 0); | |
2757 | assert(m); | |
2758 | ||
2759 | for (;;) { | |
2760 | union inotify_event_buffer buffer; | |
2761 | struct inotify_event *e; | |
2762 | ssize_t l; | |
2763 | ||
2764 | l = read(fd, &buffer, sizeof(buffer)); | |
2765 | if (l < 0) { | |
47249640 | 2766 | if (IN_SET(errno, EINTR, EAGAIN)) |
efdb0237 LP |
2767 | return 0; |
2768 | ||
2769 | return log_error_errno(errno, "Failed to read control group inotify events: %m"); | |
2770 | } | |
2771 | ||
2772 | FOREACH_INOTIFY_EVENT(e, buffer, l) { | |
2773 | Unit *u; | |
2774 | ||
2775 | if (e->wd < 0) | |
2776 | /* Queue overflow has no watch descriptor */ | |
2777 | continue; | |
2778 | ||
2779 | if (e->mask & IN_IGNORED) | |
2780 | /* The watch was just removed */ | |
2781 | continue; | |
2782 | ||
afcfaa69 LP |
2783 | /* Note that inotify might deliver events for a watch even after it was removed, |
2784 | * because it was queued before the removal. Let's ignore this here safely. */ | |
2785 | ||
0bb814c2 | 2786 | u = hashmap_get(m->cgroup_control_inotify_wd_unit, INT_TO_PTR(e->wd)); |
afcfaa69 LP |
2787 | if (u) |
2788 | unit_add_to_cgroup_empty_queue(u); | |
efdb0237 | 2789 | |
afcfaa69 LP |
2790 | u = hashmap_get(m->cgroup_memory_inotify_wd_unit, INT_TO_PTR(e->wd)); |
2791 | if (u) | |
2792 | unit_add_to_cgroup_oom_queue(u); | |
efdb0237 LP |
2793 | } |
2794 | } | |
8e274523 LP |
2795 | } |
2796 | ||
17f14955 RG |
2797 | static int cg_bpf_mask_supported(CGroupMask *ret) { |
2798 | CGroupMask mask = 0; | |
2799 | int r; | |
2800 | ||
2801 | /* BPF-based firewall */ | |
2802 | r = bpf_firewall_supported(); | |
2803 | if (r > 0) | |
2804 | mask |= CGROUP_MASK_BPF_FIREWALL; | |
2805 | ||
084c7007 RG |
2806 | /* BPF-based device access control */ |
2807 | r = bpf_devices_supported(); | |
2808 | if (r > 0) | |
2809 | mask |= CGROUP_MASK_BPF_DEVICES; | |
2810 | ||
17f14955 RG |
2811 | *ret = mask; |
2812 | return 0; | |
2813 | } | |
2814 | ||
8e274523 | 2815 | int manager_setup_cgroup(Manager *m) { |
9444b1f2 | 2816 | _cleanup_free_ char *path = NULL; |
10bd3e2e | 2817 | const char *scope_path; |
efdb0237 | 2818 | CGroupController c; |
b4cccbc1 | 2819 | int r, all_unified; |
17f14955 | 2820 | CGroupMask mask; |
efdb0237 | 2821 | char *e; |
8e274523 LP |
2822 | |
2823 | assert(m); | |
2824 | ||
35d2e7ec | 2825 | /* 1. Determine hierarchy */ |
efdb0237 | 2826 | m->cgroup_root = mfree(m->cgroup_root); |
9444b1f2 | 2827 | r = cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER, 0, &m->cgroup_root); |
23bbb0de MS |
2828 | if (r < 0) |
2829 | return log_error_errno(r, "Cannot determine cgroup we are running in: %m"); | |
8e274523 | 2830 | |
efdb0237 LP |
2831 | /* Chop off the init scope, if we are already located in it */ |
2832 | e = endswith(m->cgroup_root, "/" SPECIAL_INIT_SCOPE); | |
0d8c31ff | 2833 | |
efdb0237 LP |
2834 | /* LEGACY: Also chop off the system slice if we are in |
2835 | * it. This is to support live upgrades from older systemd | |
2836 | * versions where PID 1 was moved there. Also see | |
2837 | * cg_get_root_path(). */ | |
463d0d15 | 2838 | if (!e && MANAGER_IS_SYSTEM(m)) { |
9444b1f2 | 2839 | e = endswith(m->cgroup_root, "/" SPECIAL_SYSTEM_SLICE); |
15c60e99 | 2840 | if (!e) |
efdb0237 | 2841 | e = endswith(m->cgroup_root, "/system"); /* even more legacy */ |
0baf24dd | 2842 | } |
efdb0237 LP |
2843 | if (e) |
2844 | *e = 0; | |
7ccfb64a | 2845 | |
7546145e LP |
2846 | /* And make sure to store away the root value without trailing slash, even for the root dir, so that we can |
2847 | * easily prepend it everywhere. */ | |
2848 | delete_trailing_chars(m->cgroup_root, "/"); | |
8e274523 | 2849 | |
35d2e7ec | 2850 | /* 2. Show data */ |
9444b1f2 | 2851 | r = cg_get_path(SYSTEMD_CGROUP_CONTROLLER, m->cgroup_root, NULL, &path); |
23bbb0de MS |
2852 | if (r < 0) |
2853 | return log_error_errno(r, "Cannot find cgroup mount point: %m"); | |
8e274523 | 2854 | |
d4d99bc6 | 2855 | r = cg_unified(); |
415fc41c TH |
2856 | if (r < 0) |
2857 | return log_error_errno(r, "Couldn't determine if we are running in the unified hierarchy: %m"); | |
5da38d07 | 2858 | |
b4cccbc1 | 2859 | all_unified = cg_all_unified(); |
d4c819ed ZJS |
2860 | if (all_unified < 0) |
2861 | return log_error_errno(all_unified, "Couldn't determine whether we are in all unified mode: %m"); | |
2862 | if (all_unified > 0) | |
efdb0237 | 2863 | log_debug("Unified cgroup hierarchy is located at %s.", path); |
b4cccbc1 | 2864 | else { |
c22800e4 | 2865 | r = cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER); |
b4cccbc1 LP |
2866 | if (r < 0) |
2867 | return log_error_errno(r, "Failed to determine whether systemd's own controller is in unified mode: %m"); | |
2868 | if (r > 0) | |
2869 | log_debug("Unified cgroup hierarchy is located at %s. Controllers are on legacy hierarchies.", path); | |
2870 | else | |
2871 | log_debug("Using cgroup controller " SYSTEMD_CGROUP_CONTROLLER_LEGACY ". File system hierarchy is at %s.", path); | |
2872 | } | |
efdb0237 | 2873 | |
09e24654 LP |
2874 | /* 3. Allocate cgroup empty defer event source */ |
2875 | m->cgroup_empty_event_source = sd_event_source_unref(m->cgroup_empty_event_source); | |
2876 | r = sd_event_add_defer(m->event, &m->cgroup_empty_event_source, on_cgroup_empty_event, m); | |
2877 | if (r < 0) | |
2878 | return log_error_errno(r, "Failed to create cgroup empty event source: %m"); | |
2879 | ||
cbe83389 LP |
2880 | /* Schedule cgroup empty checks early, but after having processed service notification messages or |
2881 | * SIGCHLD signals, so that a cgroup running empty is always just the last safety net of | |
2882 | * notification, and we collected the metadata the notification and SIGCHLD stuff offers first. */ | |
09e24654 LP |
2883 | r = sd_event_source_set_priority(m->cgroup_empty_event_source, SD_EVENT_PRIORITY_NORMAL-5); |
2884 | if (r < 0) | |
2885 | return log_error_errno(r, "Failed to set priority of cgroup empty event source: %m"); | |
2886 | ||
2887 | r = sd_event_source_set_enabled(m->cgroup_empty_event_source, SD_EVENT_OFF); | |
2888 | if (r < 0) | |
2889 | return log_error_errno(r, "Failed to disable cgroup empty event source: %m"); | |
2890 | ||
2891 | (void) sd_event_source_set_description(m->cgroup_empty_event_source, "cgroup-empty"); | |
2892 | ||
2893 | /* 4. Install notifier inotify object, or agent */ | |
10bd3e2e | 2894 | if (cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER) > 0) { |
c6c18be3 | 2895 | |
09e24654 | 2896 | /* In the unified hierarchy we can get cgroup empty notifications via inotify. */ |
efdb0237 | 2897 | |
10bd3e2e LP |
2898 | m->cgroup_inotify_event_source = sd_event_source_unref(m->cgroup_inotify_event_source); |
2899 | safe_close(m->cgroup_inotify_fd); | |
efdb0237 | 2900 | |
10bd3e2e LP |
2901 | m->cgroup_inotify_fd = inotify_init1(IN_NONBLOCK|IN_CLOEXEC); |
2902 | if (m->cgroup_inotify_fd < 0) | |
2903 | return log_error_errno(errno, "Failed to create control group inotify object: %m"); | |
efdb0237 | 2904 | |
10bd3e2e LP |
2905 | r = sd_event_add_io(m->event, &m->cgroup_inotify_event_source, m->cgroup_inotify_fd, EPOLLIN, on_cgroup_inotify_event, m); |
2906 | if (r < 0) | |
2907 | return log_error_errno(r, "Failed to watch control group inotify object: %m"); | |
efdb0237 | 2908 | |
cbe83389 LP |
2909 | /* Process cgroup empty notifications early. Note that when this event is dispatched it'll |
2910 | * just add the unit to a cgroup empty queue, hence let's run earlier than that. Also see | |
2911 | * handling of cgroup agent notifications, for the classic cgroup hierarchy support. */ | |
2912 | r = sd_event_source_set_priority(m->cgroup_inotify_event_source, SD_EVENT_PRIORITY_NORMAL-9); | |
10bd3e2e LP |
2913 | if (r < 0) |
2914 | return log_error_errno(r, "Failed to set priority of inotify event source: %m"); | |
efdb0237 | 2915 | |
10bd3e2e | 2916 | (void) sd_event_source_set_description(m->cgroup_inotify_event_source, "cgroup-inotify"); |
efdb0237 | 2917 | |
611c4f8a | 2918 | } else if (MANAGER_IS_SYSTEM(m) && manager_owns_host_root_cgroup(m) && !MANAGER_IS_TEST_RUN(m)) { |
efdb0237 | 2919 | |
10bd3e2e LP |
2920 | /* On the legacy hierarchy we only get notifications via cgroup agents. (Which isn't really reliable, |
2921 | * since it does not generate events when control groups with children run empty. */ | |
8e274523 | 2922 | |
10bd3e2e | 2923 | r = cg_install_release_agent(SYSTEMD_CGROUP_CONTROLLER, SYSTEMD_CGROUP_AGENT_PATH); |
23bbb0de | 2924 | if (r < 0) |
10bd3e2e LP |
2925 | log_warning_errno(r, "Failed to install release agent, ignoring: %m"); |
2926 | else if (r > 0) | |
2927 | log_debug("Installed release agent."); | |
2928 | else if (r == 0) | |
2929 | log_debug("Release agent already installed."); | |
2930 | } | |
efdb0237 | 2931 | |
09e24654 | 2932 | /* 5. Make sure we are in the special "init.scope" unit in the root slice. */ |
10bd3e2e LP |
2933 | scope_path = strjoina(m->cgroup_root, "/" SPECIAL_INIT_SCOPE); |
2934 | r = cg_create_and_attach(SYSTEMD_CGROUP_CONTROLLER, scope_path, 0); | |
aa77e234 MS |
2935 | if (r >= 0) { |
2936 | /* Also, move all other userspace processes remaining in the root cgroup into that scope. */ | |
2937 | r = cg_migrate(SYSTEMD_CGROUP_CONTROLLER, m->cgroup_root, SYSTEMD_CGROUP_CONTROLLER, scope_path, 0); | |
2938 | if (r < 0) | |
2939 | log_warning_errno(r, "Couldn't move remaining userspace processes, ignoring: %m"); | |
c6c18be3 | 2940 | |
aa77e234 MS |
2941 | /* 6. And pin it, so that it cannot be unmounted */ |
2942 | safe_close(m->pin_cgroupfs_fd); | |
2943 | m->pin_cgroupfs_fd = open(path, O_RDONLY|O_CLOEXEC|O_DIRECTORY|O_NOCTTY|O_NONBLOCK); | |
2944 | if (m->pin_cgroupfs_fd < 0) | |
2945 | return log_error_errno(errno, "Failed to open pin file: %m"); | |
0d8c31ff | 2946 | |
638cece4 | 2947 | } else if (!MANAGER_IS_TEST_RUN(m)) |
aa77e234 | 2948 | return log_error_errno(r, "Failed to create %s control group: %m", scope_path); |
10bd3e2e | 2949 | |
09e24654 | 2950 | /* 7. Always enable hierarchical support if it exists... */ |
638cece4 | 2951 | if (!all_unified && !MANAGER_IS_TEST_RUN(m)) |
10bd3e2e | 2952 | (void) cg_set_attribute("memory", "/", "memory.use_hierarchy", "1"); |
c6c18be3 | 2953 | |
17f14955 | 2954 | /* 8. Figure out which controllers are supported */ |
efdb0237 LP |
2955 | r = cg_mask_supported(&m->cgroup_supported); |
2956 | if (r < 0) | |
2957 | return log_error_errno(r, "Failed to determine supported controllers: %m"); | |
17f14955 RG |
2958 | |
2959 | /* 9. Figure out which bpf-based pseudo-controllers are supported */ | |
2960 | r = cg_bpf_mask_supported(&mask); | |
2961 | if (r < 0) | |
2962 | return log_error_errno(r, "Failed to determine supported bpf-based pseudo-controllers: %m"); | |
2963 | m->cgroup_supported |= mask; | |
2964 | ||
2965 | /* 10. Log which controllers are supported */ | |
efdb0237 | 2966 | for (c = 0; c < _CGROUP_CONTROLLER_MAX; c++) |
eee0a1e4 | 2967 | log_debug("Controller '%s' supported: %s", cgroup_controller_to_string(c), yes_no(m->cgroup_supported & CGROUP_CONTROLLER_TO_MASK(c))); |
9156e799 | 2968 | |
a32360f1 | 2969 | return 0; |
8e274523 LP |
2970 | } |
2971 | ||
c6c18be3 | 2972 | void manager_shutdown_cgroup(Manager *m, bool delete) { |
8e274523 LP |
2973 | assert(m); |
2974 | ||
9444b1f2 LP |
2975 | /* We can't really delete the group, since we are in it. But |
2976 | * let's trim it. */ | |
f6c63f6f | 2977 | if (delete && m->cgroup_root && m->test_run_flags != MANAGER_TEST_RUN_MINIMAL) |
efdb0237 LP |
2978 | (void) cg_trim(SYSTEMD_CGROUP_CONTROLLER, m->cgroup_root, false); |
2979 | ||
09e24654 LP |
2980 | m->cgroup_empty_event_source = sd_event_source_unref(m->cgroup_empty_event_source); |
2981 | ||
0bb814c2 | 2982 | m->cgroup_control_inotify_wd_unit = hashmap_free(m->cgroup_control_inotify_wd_unit); |
afcfaa69 | 2983 | m->cgroup_memory_inotify_wd_unit = hashmap_free(m->cgroup_memory_inotify_wd_unit); |
efdb0237 LP |
2984 | |
2985 | m->cgroup_inotify_event_source = sd_event_source_unref(m->cgroup_inotify_event_source); | |
2986 | m->cgroup_inotify_fd = safe_close(m->cgroup_inotify_fd); | |
8e274523 | 2987 | |
03e334a1 | 2988 | m->pin_cgroupfs_fd = safe_close(m->pin_cgroupfs_fd); |
c6c18be3 | 2989 | |
efdb0237 | 2990 | m->cgroup_root = mfree(m->cgroup_root); |
8e274523 LP |
2991 | } |
2992 | ||
4ad49000 | 2993 | Unit* manager_get_unit_by_cgroup(Manager *m, const char *cgroup) { |
acb14d31 | 2994 | char *p; |
4ad49000 | 2995 | Unit *u; |
acb14d31 LP |
2996 | |
2997 | assert(m); | |
2998 | assert(cgroup); | |
acb14d31 | 2999 | |
4ad49000 LP |
3000 | u = hashmap_get(m->cgroup_unit, cgroup); |
3001 | if (u) | |
3002 | return u; | |
acb14d31 | 3003 | |
8e70580b | 3004 | p = strdupa(cgroup); |
acb14d31 LP |
3005 | for (;;) { |
3006 | char *e; | |
3007 | ||
3008 | e = strrchr(p, '/'); | |
efdb0237 LP |
3009 | if (!e || e == p) |
3010 | return hashmap_get(m->cgroup_unit, SPECIAL_ROOT_SLICE); | |
acb14d31 LP |
3011 | |
3012 | *e = 0; | |
3013 | ||
4ad49000 LP |
3014 | u = hashmap_get(m->cgroup_unit, p); |
3015 | if (u) | |
3016 | return u; | |
acb14d31 LP |
3017 | } |
3018 | } | |
3019 | ||
b3ac818b | 3020 | Unit *manager_get_unit_by_pid_cgroup(Manager *m, pid_t pid) { |
4ad49000 | 3021 | _cleanup_free_ char *cgroup = NULL; |
8e274523 | 3022 | |
8c47c732 LP |
3023 | assert(m); |
3024 | ||
62a76913 | 3025 | if (!pid_is_valid(pid)) |
b3ac818b LP |
3026 | return NULL; |
3027 | ||
62a76913 | 3028 | if (cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER, pid, &cgroup) < 0) |
b3ac818b LP |
3029 | return NULL; |
3030 | ||
3031 | return manager_get_unit_by_cgroup(m, cgroup); | |
3032 | } | |
3033 | ||
3034 | Unit *manager_get_unit_by_pid(Manager *m, pid_t pid) { | |
62a76913 | 3035 | Unit *u, **array; |
b3ac818b LP |
3036 | |
3037 | assert(m); | |
3038 | ||
62a76913 LP |
3039 | /* Note that a process might be owned by multiple units, we return only one here, which is good enough for most |
3040 | * cases, though not strictly correct. We prefer the one reported by cgroup membership, as that's the most | |
3041 | * relevant one as children of the process will be assigned to that one, too, before all else. */ | |
3042 | ||
3043 | if (!pid_is_valid(pid)) | |
8c47c732 LP |
3044 | return NULL; |
3045 | ||
2ca9d979 | 3046 | if (pid == getpid_cached()) |
efdb0237 LP |
3047 | return hashmap_get(m->units, SPECIAL_INIT_SCOPE); |
3048 | ||
62a76913 | 3049 | u = manager_get_unit_by_pid_cgroup(m, pid); |
5fe8876b LP |
3050 | if (u) |
3051 | return u; | |
3052 | ||
62a76913 | 3053 | u = hashmap_get(m->watch_pids, PID_TO_PTR(pid)); |
5fe8876b LP |
3054 | if (u) |
3055 | return u; | |
3056 | ||
62a76913 LP |
3057 | array = hashmap_get(m->watch_pids, PID_TO_PTR(-pid)); |
3058 | if (array) | |
3059 | return array[0]; | |
3060 | ||
3061 | return NULL; | |
6dde1f33 | 3062 | } |
4fbf50b3 | 3063 | |
4ad49000 LP |
3064 | int manager_notify_cgroup_empty(Manager *m, const char *cgroup) { |
3065 | Unit *u; | |
4fbf50b3 | 3066 | |
4ad49000 LP |
3067 | assert(m); |
3068 | assert(cgroup); | |
4fbf50b3 | 3069 | |
09e24654 LP |
3070 | /* Called on the legacy hierarchy whenever we get an explicit cgroup notification from the cgroup agent process |
3071 | * or from the --system instance */ | |
3072 | ||
d8fdc620 LP |
3073 | log_debug("Got cgroup empty notification for: %s", cgroup); |
3074 | ||
4ad49000 | 3075 | u = manager_get_unit_by_cgroup(m, cgroup); |
5ad096b3 LP |
3076 | if (!u) |
3077 | return 0; | |
b56c28c3 | 3078 | |
09e24654 LP |
3079 | unit_add_to_cgroup_empty_queue(u); |
3080 | return 1; | |
5ad096b3 LP |
3081 | } |
3082 | ||
3083 | int unit_get_memory_current(Unit *u, uint64_t *ret) { | |
3084 | _cleanup_free_ char *v = NULL; | |
3085 | int r; | |
3086 | ||
3087 | assert(u); | |
3088 | assert(ret); | |
3089 | ||
2e4025c0 | 3090 | if (!UNIT_CGROUP_BOOL(u, memory_accounting)) |
cf3b4be1 LP |
3091 | return -ENODATA; |
3092 | ||
5ad096b3 LP |
3093 | if (!u->cgroup_path) |
3094 | return -ENODATA; | |
3095 | ||
1f73aa00 | 3096 | /* The root cgroup doesn't expose this information, let's get it from /proc instead */ |
611c4f8a | 3097 | if (unit_has_host_root_cgroup(u)) |
c482724a | 3098 | return procfs_memory_get_used(ret); |
1f73aa00 | 3099 | |
efdb0237 | 3100 | if ((u->cgroup_realized_mask & CGROUP_MASK_MEMORY) == 0) |
5ad096b3 LP |
3101 | return -ENODATA; |
3102 | ||
b4cccbc1 LP |
3103 | r = cg_all_unified(); |
3104 | if (r < 0) | |
3105 | return r; | |
3106 | if (r > 0) | |
efdb0237 | 3107 | r = cg_get_attribute("memory", u->cgroup_path, "memory.current", &v); |
b4cccbc1 LP |
3108 | else |
3109 | r = cg_get_attribute("memory", u->cgroup_path, "memory.usage_in_bytes", &v); | |
5ad096b3 LP |
3110 | if (r == -ENOENT) |
3111 | return -ENODATA; | |
3112 | if (r < 0) | |
3113 | return r; | |
3114 | ||
3115 | return safe_atou64(v, ret); | |
3116 | } | |
3117 | ||
03a7b521 LP |
3118 | int unit_get_tasks_current(Unit *u, uint64_t *ret) { |
3119 | _cleanup_free_ char *v = NULL; | |
3120 | int r; | |
3121 | ||
3122 | assert(u); | |
3123 | assert(ret); | |
3124 | ||
2e4025c0 | 3125 | if (!UNIT_CGROUP_BOOL(u, tasks_accounting)) |
cf3b4be1 LP |
3126 | return -ENODATA; |
3127 | ||
03a7b521 LP |
3128 | if (!u->cgroup_path) |
3129 | return -ENODATA; | |
3130 | ||
c36a69f4 | 3131 | /* The root cgroup doesn't expose this information, let's get it from /proc instead */ |
611c4f8a | 3132 | if (unit_has_host_root_cgroup(u)) |
c36a69f4 LP |
3133 | return procfs_tasks_get_current(ret); |
3134 | ||
1f73aa00 LP |
3135 | if ((u->cgroup_realized_mask & CGROUP_MASK_PIDS) == 0) |
3136 | return -ENODATA; | |
3137 | ||
03a7b521 LP |
3138 | r = cg_get_attribute("pids", u->cgroup_path, "pids.current", &v); |
3139 | if (r == -ENOENT) | |
3140 | return -ENODATA; | |
3141 | if (r < 0) | |
3142 | return r; | |
3143 | ||
3144 | return safe_atou64(v, ret); | |
3145 | } | |
3146 | ||
5ad096b3 LP |
3147 | static int unit_get_cpu_usage_raw(Unit *u, nsec_t *ret) { |
3148 | _cleanup_free_ char *v = NULL; | |
3149 | uint64_t ns; | |
3150 | int r; | |
3151 | ||
3152 | assert(u); | |
3153 | assert(ret); | |
3154 | ||
3155 | if (!u->cgroup_path) | |
3156 | return -ENODATA; | |
3157 | ||
1f73aa00 | 3158 | /* The root cgroup doesn't expose this information, let's get it from /proc instead */ |
611c4f8a | 3159 | if (unit_has_host_root_cgroup(u)) |
1f73aa00 LP |
3160 | return procfs_cpu_get_usage(ret); |
3161 | ||
f98c2585 CD |
3162 | /* Requisite controllers for CPU accounting are not enabled */ |
3163 | if ((get_cpu_accounting_mask() & ~u->cgroup_realized_mask) != 0) | |
3164 | return -ENODATA; | |
3165 | ||
92a99304 LP |
3166 | r = cg_all_unified(); |
3167 | if (r < 0) | |
3168 | return r; | |
b4cccbc1 | 3169 | if (r > 0) { |
66ebf6c0 TH |
3170 | _cleanup_free_ char *val = NULL; |
3171 | uint64_t us; | |
5ad096b3 | 3172 | |
b734a4ff | 3173 | r = cg_get_keyed_attribute("cpu", u->cgroup_path, "cpu.stat", STRV_MAKE("usage_usec"), &val); |
b734a4ff LP |
3174 | if (IN_SET(r, -ENOENT, -ENXIO)) |
3175 | return -ENODATA; | |
d742f4b5 LP |
3176 | if (r < 0) |
3177 | return r; | |
66ebf6c0 TH |
3178 | |
3179 | r = safe_atou64(val, &us); | |
3180 | if (r < 0) | |
3181 | return r; | |
3182 | ||
3183 | ns = us * NSEC_PER_USEC; | |
3184 | } else { | |
66ebf6c0 TH |
3185 | r = cg_get_attribute("cpuacct", u->cgroup_path, "cpuacct.usage", &v); |
3186 | if (r == -ENOENT) | |
3187 | return -ENODATA; | |
3188 | if (r < 0) | |
3189 | return r; | |
3190 | ||
3191 | r = safe_atou64(v, &ns); | |
3192 | if (r < 0) | |
3193 | return r; | |
3194 | } | |
5ad096b3 LP |
3195 | |
3196 | *ret = ns; | |
3197 | return 0; | |
3198 | } | |
3199 | ||
3200 | int unit_get_cpu_usage(Unit *u, nsec_t *ret) { | |
3201 | nsec_t ns; | |
3202 | int r; | |
3203 | ||
fe700f46 LP |
3204 | assert(u); |
3205 | ||
3206 | /* Retrieve the current CPU usage counter. This will subtract the CPU counter taken when the unit was | |
3207 | * started. If the cgroup has been removed already, returns the last cached value. To cache the value, simply | |
3208 | * call this function with a NULL return value. */ | |
3209 | ||
2e4025c0 | 3210 | if (!UNIT_CGROUP_BOOL(u, cpu_accounting)) |
cf3b4be1 LP |
3211 | return -ENODATA; |
3212 | ||
5ad096b3 | 3213 | r = unit_get_cpu_usage_raw(u, &ns); |
fe700f46 LP |
3214 | if (r == -ENODATA && u->cpu_usage_last != NSEC_INFINITY) { |
3215 | /* If we can't get the CPU usage anymore (because the cgroup was already removed, for example), use our | |
3216 | * cached value. */ | |
3217 | ||
3218 | if (ret) | |
3219 | *ret = u->cpu_usage_last; | |
3220 | return 0; | |
3221 | } | |
5ad096b3 LP |
3222 | if (r < 0) |
3223 | return r; | |
3224 | ||
66ebf6c0 TH |
3225 | if (ns > u->cpu_usage_base) |
3226 | ns -= u->cpu_usage_base; | |
5ad096b3 LP |
3227 | else |
3228 | ns = 0; | |
3229 | ||
fe700f46 LP |
3230 | u->cpu_usage_last = ns; |
3231 | if (ret) | |
3232 | *ret = ns; | |
3233 | ||
5ad096b3 LP |
3234 | return 0; |
3235 | } | |
3236 | ||
906c06f6 DM |
3237 | int unit_get_ip_accounting( |
3238 | Unit *u, | |
3239 | CGroupIPAccountingMetric metric, | |
3240 | uint64_t *ret) { | |
3241 | ||
6b659ed8 | 3242 | uint64_t value; |
906c06f6 DM |
3243 | int fd, r; |
3244 | ||
3245 | assert(u); | |
3246 | assert(metric >= 0); | |
3247 | assert(metric < _CGROUP_IP_ACCOUNTING_METRIC_MAX); | |
3248 | assert(ret); | |
3249 | ||
2e4025c0 | 3250 | if (!UNIT_CGROUP_BOOL(u, ip_accounting)) |
cf3b4be1 LP |
3251 | return -ENODATA; |
3252 | ||
906c06f6 DM |
3253 | fd = IN_SET(metric, CGROUP_IP_INGRESS_BYTES, CGROUP_IP_INGRESS_PACKETS) ? |
3254 | u->ip_accounting_ingress_map_fd : | |
3255 | u->ip_accounting_egress_map_fd; | |
906c06f6 DM |
3256 | if (fd < 0) |
3257 | return -ENODATA; | |
3258 | ||
3259 | if (IN_SET(metric, CGROUP_IP_INGRESS_BYTES, CGROUP_IP_EGRESS_BYTES)) | |
6b659ed8 | 3260 | r = bpf_firewall_read_accounting(fd, &value, NULL); |
906c06f6 | 3261 | else |
6b659ed8 LP |
3262 | r = bpf_firewall_read_accounting(fd, NULL, &value); |
3263 | if (r < 0) | |
3264 | return r; | |
3265 | ||
3266 | /* Add in additional metrics from a previous runtime. Note that when reexecing/reloading the daemon we compile | |
3267 | * all BPF programs and maps anew, but serialize the old counters. When deserializing we store them in the | |
3268 | * ip_accounting_extra[] field, and add them in here transparently. */ | |
3269 | ||
3270 | *ret = value + u->ip_accounting_extra[metric]; | |
906c06f6 DM |
3271 | |
3272 | return r; | |
3273 | } | |
3274 | ||
fbe14fc9 LP |
3275 | static int unit_get_io_accounting_raw(Unit *u, uint64_t ret[static _CGROUP_IO_ACCOUNTING_METRIC_MAX]) { |
3276 | static const char *const field_names[_CGROUP_IO_ACCOUNTING_METRIC_MAX] = { | |
3277 | [CGROUP_IO_READ_BYTES] = "rbytes=", | |
3278 | [CGROUP_IO_WRITE_BYTES] = "wbytes=", | |
3279 | [CGROUP_IO_READ_OPERATIONS] = "rios=", | |
3280 | [CGROUP_IO_WRITE_OPERATIONS] = "wios=", | |
3281 | }; | |
3282 | uint64_t acc[_CGROUP_IO_ACCOUNTING_METRIC_MAX] = {}; | |
3283 | _cleanup_free_ char *path = NULL; | |
3284 | _cleanup_fclose_ FILE *f = NULL; | |
3285 | int r; | |
3286 | ||
3287 | assert(u); | |
3288 | ||
3289 | if (!u->cgroup_path) | |
3290 | return -ENODATA; | |
3291 | ||
3292 | if (unit_has_host_root_cgroup(u)) | |
3293 | return -ENODATA; /* TODO: return useful data for the top-level cgroup */ | |
3294 | ||
3295 | r = cg_all_unified(); | |
3296 | if (r < 0) | |
3297 | return r; | |
3298 | if (r == 0) /* TODO: support cgroupv1 */ | |
3299 | return -ENODATA; | |
3300 | ||
3301 | if (!FLAGS_SET(u->cgroup_realized_mask, CGROUP_MASK_IO)) | |
3302 | return -ENODATA; | |
3303 | ||
3304 | r = cg_get_path("io", u->cgroup_path, "io.stat", &path); | |
3305 | if (r < 0) | |
3306 | return r; | |
3307 | ||
3308 | f = fopen(path, "re"); | |
3309 | if (!f) | |
3310 | return -errno; | |
3311 | ||
3312 | for (;;) { | |
3313 | _cleanup_free_ char *line = NULL; | |
3314 | const char *p; | |
3315 | ||
3316 | r = read_line(f, LONG_LINE_MAX, &line); | |
3317 | if (r < 0) | |
3318 | return r; | |
3319 | if (r == 0) | |
3320 | break; | |
3321 | ||
3322 | p = line; | |
3323 | p += strcspn(p, WHITESPACE); /* Skip over device major/minor */ | |
3324 | p += strspn(p, WHITESPACE); /* Skip over following whitespace */ | |
3325 | ||
3326 | for (;;) { | |
3327 | _cleanup_free_ char *word = NULL; | |
3328 | ||
3329 | r = extract_first_word(&p, &word, NULL, EXTRACT_RETAIN_ESCAPE); | |
3330 | if (r < 0) | |
3331 | return r; | |
3332 | if (r == 0) | |
3333 | break; | |
3334 | ||
3335 | for (CGroupIOAccountingMetric i = 0; i < _CGROUP_IO_ACCOUNTING_METRIC_MAX; i++) { | |
3336 | const char *x; | |
3337 | ||
3338 | x = startswith(word, field_names[i]); | |
3339 | if (x) { | |
3340 | uint64_t w; | |
3341 | ||
3342 | r = safe_atou64(x, &w); | |
3343 | if (r < 0) | |
3344 | return r; | |
3345 | ||
3346 | /* Sum up the stats of all devices */ | |
3347 | acc[i] += w; | |
3348 | break; | |
3349 | } | |
3350 | } | |
3351 | } | |
3352 | } | |
3353 | ||
3354 | memcpy(ret, acc, sizeof(acc)); | |
3355 | return 0; | |
3356 | } | |
3357 | ||
3358 | int unit_get_io_accounting( | |
3359 | Unit *u, | |
3360 | CGroupIOAccountingMetric metric, | |
3361 | bool allow_cache, | |
3362 | uint64_t *ret) { | |
3363 | ||
3364 | uint64_t raw[_CGROUP_IO_ACCOUNTING_METRIC_MAX]; | |
3365 | int r; | |
3366 | ||
3367 | /* Retrieve an IO account parameter. This will subtract the counter when the unit was started. */ | |
3368 | ||
3369 | if (!UNIT_CGROUP_BOOL(u, io_accounting)) | |
3370 | return -ENODATA; | |
3371 | ||
3372 | if (allow_cache && u->io_accounting_last[metric] != UINT64_MAX) | |
3373 | goto done; | |
3374 | ||
3375 | r = unit_get_io_accounting_raw(u, raw); | |
3376 | if (r == -ENODATA && u->io_accounting_last[metric] != UINT64_MAX) | |
3377 | goto done; | |
3378 | if (r < 0) | |
3379 | return r; | |
3380 | ||
3381 | for (CGroupIOAccountingMetric i = 0; i < _CGROUP_IO_ACCOUNTING_METRIC_MAX; i++) { | |
3382 | /* Saturated subtraction */ | |
3383 | if (raw[i] > u->io_accounting_base[i]) | |
3384 | u->io_accounting_last[i] = raw[i] - u->io_accounting_base[i]; | |
3385 | else | |
3386 | u->io_accounting_last[i] = 0; | |
3387 | } | |
3388 | ||
3389 | done: | |
3390 | if (ret) | |
3391 | *ret = u->io_accounting_last[metric]; | |
3392 | ||
3393 | return 0; | |
3394 | } | |
3395 | ||
906c06f6 | 3396 | int unit_reset_cpu_accounting(Unit *u) { |
5ad096b3 LP |
3397 | int r; |
3398 | ||
3399 | assert(u); | |
3400 | ||
fe700f46 LP |
3401 | u->cpu_usage_last = NSEC_INFINITY; |
3402 | ||
0bbff7d6 | 3403 | r = unit_get_cpu_usage_raw(u, &u->cpu_usage_base); |
5ad096b3 | 3404 | if (r < 0) { |
66ebf6c0 | 3405 | u->cpu_usage_base = 0; |
5ad096b3 | 3406 | return r; |
b56c28c3 | 3407 | } |
2633eb83 | 3408 | |
4ad49000 | 3409 | return 0; |
4fbf50b3 LP |
3410 | } |
3411 | ||
906c06f6 DM |
3412 | int unit_reset_ip_accounting(Unit *u) { |
3413 | int r = 0, q = 0; | |
3414 | ||
3415 | assert(u); | |
3416 | ||
3417 | if (u->ip_accounting_ingress_map_fd >= 0) | |
3418 | r = bpf_firewall_reset_accounting(u->ip_accounting_ingress_map_fd); | |
3419 | ||
3420 | if (u->ip_accounting_egress_map_fd >= 0) | |
3421 | q = bpf_firewall_reset_accounting(u->ip_accounting_egress_map_fd); | |
3422 | ||
6b659ed8 LP |
3423 | zero(u->ip_accounting_extra); |
3424 | ||
906c06f6 DM |
3425 | return r < 0 ? r : q; |
3426 | } | |
3427 | ||
fbe14fc9 LP |
3428 | int unit_reset_io_accounting(Unit *u) { |
3429 | int r; | |
3430 | ||
3431 | assert(u); | |
3432 | ||
3433 | for (CGroupIOAccountingMetric i = 0; i < _CGROUP_IO_ACCOUNTING_METRIC_MAX; i++) | |
3434 | u->io_accounting_last[i] = UINT64_MAX; | |
3435 | ||
3436 | r = unit_get_io_accounting_raw(u, u->io_accounting_base); | |
3437 | if (r < 0) { | |
3438 | zero(u->io_accounting_base); | |
3439 | return r; | |
3440 | } | |
3441 | ||
3442 | return 0; | |
3443 | } | |
3444 | ||
9b2559a1 | 3445 | int unit_reset_accounting(Unit *u) { |
fbe14fc9 | 3446 | int r, q, v; |
9b2559a1 LP |
3447 | |
3448 | assert(u); | |
3449 | ||
3450 | r = unit_reset_cpu_accounting(u); | |
fbe14fc9 LP |
3451 | q = unit_reset_io_accounting(u); |
3452 | v = unit_reset_ip_accounting(u); | |
9b2559a1 | 3453 | |
fbe14fc9 | 3454 | return r < 0 ? r : q < 0 ? q : v; |
9b2559a1 LP |
3455 | } |
3456 | ||
e7ab4d1a LP |
3457 | void unit_invalidate_cgroup(Unit *u, CGroupMask m) { |
3458 | assert(u); | |
3459 | ||
3460 | if (!UNIT_HAS_CGROUP_CONTEXT(u)) | |
3461 | return; | |
3462 | ||
3463 | if (m == 0) | |
3464 | return; | |
3465 | ||
538b4852 TH |
3466 | /* always invalidate compat pairs together */ |
3467 | if (m & (CGROUP_MASK_IO | CGROUP_MASK_BLKIO)) | |
3468 | m |= CGROUP_MASK_IO | CGROUP_MASK_BLKIO; | |
3469 | ||
7cce4fb7 LP |
3470 | if (m & (CGROUP_MASK_CPU | CGROUP_MASK_CPUACCT)) |
3471 | m |= CGROUP_MASK_CPU | CGROUP_MASK_CPUACCT; | |
3472 | ||
e00068e7 | 3473 | if (FLAGS_SET(u->cgroup_invalidated_mask, m)) /* NOP? */ |
e7ab4d1a LP |
3474 | return; |
3475 | ||
e00068e7 | 3476 | u->cgroup_invalidated_mask |= m; |
91a6073e | 3477 | unit_add_to_cgroup_realize_queue(u); |
e7ab4d1a LP |
3478 | } |
3479 | ||
906c06f6 DM |
3480 | void unit_invalidate_cgroup_bpf(Unit *u) { |
3481 | assert(u); | |
3482 | ||
3483 | if (!UNIT_HAS_CGROUP_CONTEXT(u)) | |
3484 | return; | |
3485 | ||
17f14955 | 3486 | if (u->cgroup_invalidated_mask & CGROUP_MASK_BPF_FIREWALL) /* NOP? */ |
906c06f6 DM |
3487 | return; |
3488 | ||
17f14955 | 3489 | u->cgroup_invalidated_mask |= CGROUP_MASK_BPF_FIREWALL; |
91a6073e | 3490 | unit_add_to_cgroup_realize_queue(u); |
906c06f6 DM |
3491 | |
3492 | /* If we are a slice unit, we also need to put compile a new BPF program for all our children, as the IP access | |
3493 | * list of our children includes our own. */ | |
3494 | if (u->type == UNIT_SLICE) { | |
3495 | Unit *member; | |
3496 | Iterator i; | |
eef85c4a | 3497 | void *v; |
906c06f6 | 3498 | |
eef85c4a | 3499 | HASHMAP_FOREACH_KEY(v, member, u->dependencies[UNIT_BEFORE], i) { |
cb5e3bc3 CD |
3500 | if (UNIT_DEREF(member->slice) == u) |
3501 | unit_invalidate_cgroup_bpf(member); | |
906c06f6 DM |
3502 | } |
3503 | } | |
3504 | } | |
3505 | ||
1d9cc876 LP |
3506 | bool unit_cgroup_delegate(Unit *u) { |
3507 | CGroupContext *c; | |
3508 | ||
3509 | assert(u); | |
3510 | ||
3511 | if (!UNIT_VTABLE(u)->can_delegate) | |
3512 | return false; | |
3513 | ||
3514 | c = unit_get_cgroup_context(u); | |
3515 | if (!c) | |
3516 | return false; | |
3517 | ||
3518 | return c->delegate; | |
3519 | } | |
3520 | ||
e7ab4d1a LP |
3521 | void manager_invalidate_startup_units(Manager *m) { |
3522 | Iterator i; | |
3523 | Unit *u; | |
3524 | ||
3525 | assert(m); | |
3526 | ||
3527 | SET_FOREACH(u, m->startup_units, i) | |
13c31542 | 3528 | unit_invalidate_cgroup(u, CGROUP_MASK_CPU|CGROUP_MASK_IO|CGROUP_MASK_BLKIO); |
e7ab4d1a LP |
3529 | } |
3530 | ||
da8e1782 MO |
3531 | static int unit_get_nice(Unit *u) { |
3532 | ExecContext *ec; | |
3533 | ||
3534 | ec = unit_get_exec_context(u); | |
3535 | return ec ? ec->nice : 0; | |
3536 | } | |
3537 | ||
3538 | static uint64_t unit_get_cpu_weight(Unit *u) { | |
3539 | ManagerState state = manager_state(u->manager); | |
3540 | CGroupContext *cc; | |
3541 | ||
3542 | cc = unit_get_cgroup_context(u); | |
3543 | return cc ? cgroup_context_cpu_weight(cc, state) : CGROUP_WEIGHT_DEFAULT; | |
3544 | } | |
3545 | ||
3546 | int compare_job_priority(const void *a, const void *b) { | |
3547 | const Job *x = a, *y = b; | |
3548 | int nice_x, nice_y; | |
3549 | uint64_t weight_x, weight_y; | |
3550 | int ret; | |
3551 | ||
217b7b33 ZJS |
3552 | if ((ret = CMP(x->unit->type, y->unit->type)) != 0) |
3553 | return -ret; | |
3554 | ||
da8e1782 MO |
3555 | weight_x = unit_get_cpu_weight(x->unit); |
3556 | weight_y = unit_get_cpu_weight(y->unit); | |
3557 | ||
217b7b33 ZJS |
3558 | if ((ret = CMP(weight_x, weight_y)) != 0) |
3559 | return -ret; | |
da8e1782 MO |
3560 | |
3561 | nice_x = unit_get_nice(x->unit); | |
3562 | nice_y = unit_get_nice(y->unit); | |
3563 | ||
3564 | if ((ret = CMP(nice_x, nice_y)) != 0) | |
3565 | return ret; | |
3566 | ||
da8e1782 MO |
3567 | return strcmp(x->unit->id, y->unit->id); |
3568 | } | |
3569 | ||
4ad49000 | 3570 | static const char* const cgroup_device_policy_table[_CGROUP_DEVICE_POLICY_MAX] = { |
084870f9 ZJS |
3571 | [CGROUP_DEVICE_POLICY_AUTO] = "auto", |
3572 | [CGROUP_DEVICE_POLICY_CLOSED] = "closed", | |
3573 | [CGROUP_DEVICE_POLICY_STRICT] = "strict", | |
4ad49000 | 3574 | }; |
4fbf50b3 | 3575 | |
047f5d63 PH |
3576 | int unit_get_cpuset(Unit *u, CPUSet *cpus, const char *name) { |
3577 | _cleanup_free_ char *v = NULL; | |
3578 | int r; | |
3579 | ||
3580 | assert(u); | |
3581 | assert(cpus); | |
3582 | ||
3583 | if (!u->cgroup_path) | |
3584 | return -ENODATA; | |
3585 | ||
3586 | if ((u->cgroup_realized_mask & CGROUP_MASK_CPUSET) == 0) | |
3587 | return -ENODATA; | |
3588 | ||
3589 | r = cg_all_unified(); | |
3590 | if (r < 0) | |
3591 | return r; | |
3592 | if (r == 0) | |
3593 | return -ENODATA; | |
3594 | if (r > 0) | |
3595 | r = cg_get_attribute("cpuset", u->cgroup_path, name, &v); | |
3596 | if (r == -ENOENT) | |
3597 | return -ENODATA; | |
3598 | if (r < 0) | |
3599 | return r; | |
3600 | ||
3601 | return parse_cpu_set_full(v, cpus, false, NULL, NULL, 0, NULL); | |
3602 | } | |
3603 | ||
4ad49000 | 3604 | DEFINE_STRING_TABLE_LOOKUP(cgroup_device_policy, CGroupDevicePolicy); |