]>
Commit | Line | Data |
---|---|---|
53e1b683 | 1 | /* SPDX-License-Identifier: LGPL-2.1+ */ |
618234a5 | 2 | |
11c3a366 TA |
3 | #include <errno.h> |
4 | #include <stddef.h> | |
a832893f | 5 | #include <stdio.h> |
11c3a366 TA |
6 | #include <syslog.h> |
7 | ||
b5efdb8a | 8 | #include "alloc-util.h" |
6bedfcbb | 9 | #include "cpu-set-util.h" |
b070c7c0 MS |
10 | #include "dirent-util.h" |
11 | #include "errno-util.h" | |
84ac7bea | 12 | #include "extract-word.h" |
b070c7c0 | 13 | #include "fd-util.h" |
11c3a366 TA |
14 | #include "log.h" |
15 | #include "macro.h" | |
0985c7c4 | 16 | #include "memory-util.h" |
b070c7c0 | 17 | #include "missing_syscall.h" |
93cc7779 | 18 | #include "parse-util.h" |
b070c7c0 | 19 | #include "stat-util.h" |
b11d6a7b | 20 | #include "string-util.h" |
b070c7c0 MS |
21 | #include "string-table.h" |
22 | #include "strv.h" | |
23 | #include "util.h" | |
618234a5 | 24 | |
0985c7c4 | 25 | char* cpu_set_to_string(const CPUSet *a) { |
a832893f ZJS |
26 | _cleanup_free_ char *str = NULL; |
27 | size_t allocated = 0, len = 0; | |
28 | int i, r; | |
29 | ||
0985c7c4 ZJS |
30 | for (i = 0; (size_t) i < a->allocated * 8; i++) { |
31 | if (!CPU_ISSET_S(i, a->allocated, a->set)) | |
a832893f ZJS |
32 | continue; |
33 | ||
34 | if (!GREEDY_REALLOC(str, allocated, len + 1 + DECIMAL_STR_MAX(int))) | |
35 | return NULL; | |
36 | ||
37 | r = sprintf(str + len, len > 0 ? " %d" : "%d", i); | |
38 | assert_se(r > 0); | |
39 | len += r; | |
40 | } | |
41 | ||
42 | return TAKE_PTR(str) ?: strdup(""); | |
43 | } | |
44 | ||
71b28519 MS |
45 | char *cpu_set_to_range_string(const CPUSet *set) { |
46 | unsigned range_start = 0, range_end; | |
47 | _cleanup_free_ char *str = NULL; | |
48 | size_t allocated = 0, len = 0; | |
49 | bool in_range = false; | |
50 | int r; | |
51 | ||
52 | for (unsigned i = 0; i < set->allocated * 8; i++) | |
53 | if (CPU_ISSET_S(i, set->allocated, set->set)) { | |
54 | if (in_range) | |
55 | range_end++; | |
56 | else { | |
57 | range_start = range_end = i; | |
58 | in_range = true; | |
59 | } | |
60 | } else if (in_range) { | |
61 | in_range = false; | |
62 | ||
63 | if (!GREEDY_REALLOC(str, allocated, len + 2 + 2 * DECIMAL_STR_MAX(unsigned))) | |
64 | return NULL; | |
65 | ||
71923237 | 66 | if (range_end > range_start) |
1f57a176 ZJS |
67 | r = sprintf(str + len, len > 0 ? " %d-%d" : "%d-%d", range_start, range_end); |
68 | else | |
69 | r = sprintf(str + len, len > 0 ? " %d" : "%d", range_start); | |
71b28519 MS |
70 | assert_se(r > 0); |
71 | len += r; | |
72 | } | |
73 | ||
74 | if (in_range) { | |
75 | if (!GREEDY_REALLOC(str, allocated, len + 2 + 2 * DECIMAL_STR_MAX(int))) | |
76 | return NULL; | |
77 | ||
71923237 | 78 | if (range_end > range_start) |
1f57a176 ZJS |
79 | r = sprintf(str + len, len > 0 ? " %d-%d" : "%d-%d", range_start, range_end); |
80 | else | |
81 | r = sprintf(str + len, len > 0 ? " %d" : "%d", range_start); | |
71b28519 MS |
82 | assert_se(r > 0); |
83 | } | |
84 | ||
85 | return TAKE_PTR(str) ?: strdup(""); | |
86 | } | |
87 | ||
167a776d | 88 | int cpu_set_realloc(CPUSet *cpu_set, unsigned ncpus) { |
0985c7c4 ZJS |
89 | size_t need; |
90 | ||
91 | assert(cpu_set); | |
92 | ||
93 | need = CPU_ALLOC_SIZE(ncpus); | |
94 | if (need > cpu_set->allocated) { | |
95 | cpu_set_t *t; | |
96 | ||
97 | t = realloc(cpu_set->set, need); | |
98 | if (!t) | |
99 | return -ENOMEM; | |
100 | ||
101 | memzero((uint8_t*) t + cpu_set->allocated, need - cpu_set->allocated); | |
102 | ||
103 | cpu_set->set = t; | |
104 | cpu_set->allocated = need; | |
105 | } | |
106 | ||
107 | return 0; | |
108 | } | |
109 | ||
110 | static int cpu_set_add(CPUSet *cpu_set, unsigned cpu) { | |
111 | int r; | |
112 | ||
113 | if (cpu >= 8192) | |
114 | /* As of kernel 5.1, CONFIG_NR_CPUS can be set to 8192 on PowerPC */ | |
115 | return -ERANGE; | |
116 | ||
117 | r = cpu_set_realloc(cpu_set, cpu + 1); | |
118 | if (r < 0) | |
119 | return r; | |
120 | ||
121 | CPU_SET_S(cpu, cpu_set->allocated, cpu_set->set); | |
122 | return 0; | |
123 | } | |
124 | ||
125 | int cpu_set_add_all(CPUSet *a, const CPUSet *b) { | |
126 | int r; | |
127 | ||
128 | /* Do this backwards, so if we fail, we fail before changing anything. */ | |
129 | for (unsigned cpu_p1 = b->allocated * 8; cpu_p1 > 0; cpu_p1--) | |
130 | if (CPU_ISSET_S(cpu_p1 - 1, b->allocated, b->set)) { | |
131 | r = cpu_set_add(a, cpu_p1 - 1); | |
132 | if (r < 0) | |
133 | return r; | |
134 | } | |
135 | ||
136 | return 0; | |
137 | } | |
138 | ||
139 | int parse_cpu_set_full( | |
618234a5 | 140 | const char *rvalue, |
0985c7c4 | 141 | CPUSet *cpu_set, |
6d8a29b2 | 142 | bool warn, |
618234a5 LP |
143 | const char *unit, |
144 | const char *filename, | |
145 | unsigned line, | |
146 | const char *lvalue) { | |
147 | ||
0985c7c4 | 148 | _cleanup_(cpu_set_reset) CPUSet c = {}; |
6d8a29b2 | 149 | const char *p = rvalue; |
618234a5 | 150 | |
0985c7c4 | 151 | assert(p); |
618234a5 LP |
152 | |
153 | for (;;) { | |
154 | _cleanup_free_ char *word = NULL; | |
0985c7c4 | 155 | unsigned cpu_lower, cpu_upper; |
618234a5 LP |
156 | int r; |
157 | ||
4ec85141 | 158 | r = extract_first_word(&p, &word, WHITESPACE ",", EXTRACT_UNQUOTE); |
6d8a29b2 YW |
159 | if (r == -ENOMEM) |
160 | return warn ? log_oom() : -ENOMEM; | |
a26662ce | 161 | if (r < 0) |
6d8a29b2 | 162 | return warn ? log_syntax(unit, LOG_ERR, filename, line, r, "Invalid value for %s: %s", lvalue, rvalue) : r; |
618234a5 LP |
163 | if (r == 0) |
164 | break; | |
165 | ||
a26662ce FB |
166 | r = parse_range(word, &cpu_lower, &cpu_upper); |
167 | if (r < 0) | |
6d8a29b2 | 168 | return warn ? log_syntax(unit, LOG_ERR, filename, line, r, "Failed to parse CPU affinity '%s'", word) : r; |
032cf8e4 | 169 | |
6d8a29b2 YW |
170 | if (cpu_lower > cpu_upper) { |
171 | if (warn) | |
0985c7c4 ZJS |
172 | log_syntax(unit, LOG_WARNING, filename, line, 0, "Range '%s' is invalid, %u > %u, ignoring.", |
173 | word, cpu_lower, cpu_upper); | |
174 | ||
175 | /* Make sure something is allocated, to distinguish this from the empty case */ | |
176 | r = cpu_set_realloc(&c, 1); | |
177 | if (r < 0) | |
178 | return r; | |
032cf8e4 YW |
179 | } |
180 | ||
0985c7c4 ZJS |
181 | for (unsigned cpu_p1 = MIN(cpu_upper, UINT_MAX-1) + 1; cpu_p1 > cpu_lower; cpu_p1--) { |
182 | r = cpu_set_add(&c, cpu_p1 - 1); | |
183 | if (r < 0) | |
184 | return warn ? log_syntax(unit, LOG_ERR, filename, line, r, | |
185 | "Cannot add CPU %u to set: %m", cpu_p1 - 1) : r; | |
186 | } | |
032cf8e4 YW |
187 | } |
188 | ||
0985c7c4 ZJS |
189 | /* On success, transfer ownership to the output variable */ |
190 | *cpu_set = c; | |
191 | c = (CPUSet) {}; | |
192 | ||
193 | return 0; | |
194 | } | |
195 | ||
196 | int parse_cpu_set_extend( | |
197 | const char *rvalue, | |
198 | CPUSet *old, | |
199 | bool warn, | |
200 | const char *unit, | |
201 | const char *filename, | |
202 | unsigned line, | |
203 | const char *lvalue) { | |
204 | ||
205 | _cleanup_(cpu_set_reset) CPUSet cpuset = {}; | |
206 | int r; | |
207 | ||
208 | r = parse_cpu_set_full(rvalue, &cpuset, true, unit, filename, line, lvalue); | |
209 | if (r < 0) | |
210 | return r; | |
211 | ||
212 | if (!cpuset.set) { | |
213 | /* An empty assignment resets the CPU list */ | |
214 | cpu_set_reset(old); | |
215 | return 0; | |
216 | } | |
217 | ||
218 | if (!old->set) { | |
219 | *old = cpuset; | |
220 | cpuset = (CPUSet) {}; | |
221 | return 0; | |
222 | } | |
032cf8e4 | 223 | |
0985c7c4 | 224 | return cpu_set_add_all(old, &cpuset); |
032cf8e4 | 225 | } |
f44b3035 ZJS |
226 | |
227 | int cpus_in_affinity_mask(void) { | |
228 | size_t n = 16; | |
229 | int r; | |
230 | ||
231 | for (;;) { | |
232 | cpu_set_t *c; | |
233 | ||
234 | c = CPU_ALLOC(n); | |
235 | if (!c) | |
236 | return -ENOMEM; | |
237 | ||
238 | if (sched_getaffinity(0, CPU_ALLOC_SIZE(n), c) >= 0) { | |
239 | int k; | |
240 | ||
241 | k = CPU_COUNT_S(CPU_ALLOC_SIZE(n), c); | |
242 | CPU_FREE(c); | |
243 | ||
244 | if (k <= 0) | |
245 | return -EINVAL; | |
246 | ||
247 | return k; | |
248 | } | |
249 | ||
250 | r = -errno; | |
251 | CPU_FREE(c); | |
252 | ||
253 | if (r != -EINVAL) | |
254 | return r; | |
255 | if (n > SIZE_MAX/2) | |
256 | return -ENOMEM; | |
257 | n *= 2; | |
258 | } | |
259 | } | |
c367f996 MS |
260 | |
261 | int cpu_set_to_dbus(const CPUSet *set, uint8_t **ret, size_t *allocated) { | |
262 | uint8_t *out; | |
263 | ||
264 | assert(set); | |
265 | assert(ret); | |
266 | ||
267 | out = new0(uint8_t, set->allocated); | |
268 | if (!out) | |
269 | return -ENOMEM; | |
270 | ||
271 | for (unsigned cpu = 0; cpu < set->allocated * 8; cpu++) | |
272 | if (CPU_ISSET_S(cpu, set->allocated, set->set)) | |
273 | out[cpu / 8] |= 1u << (cpu % 8); | |
274 | ||
275 | *ret = out; | |
276 | *allocated = set->allocated; | |
277 | return 0; | |
278 | } | |
279 | ||
280 | int cpu_set_from_dbus(const uint8_t *bits, size_t size, CPUSet *set) { | |
281 | _cleanup_(cpu_set_reset) CPUSet s = {}; | |
282 | int r; | |
283 | ||
284 | assert(bits); | |
285 | assert(set); | |
286 | ||
287 | for (unsigned cpu = size * 8; cpu > 0; cpu--) | |
288 | if (bits[(cpu - 1) / 8] & (1u << ((cpu - 1) % 8))) { | |
289 | r = cpu_set_add(&s, cpu - 1); | |
290 | if (r < 0) | |
291 | return r; | |
292 | } | |
293 | ||
294 | *set = s; | |
295 | s = (CPUSet) {}; | |
296 | return 0; | |
297 | } | |
b070c7c0 MS |
298 | |
299 | bool numa_policy_is_valid(const NUMAPolicy *policy) { | |
300 | assert(policy); | |
301 | ||
302 | if (!mpol_is_valid(numa_policy_get_type(policy))) | |
303 | return false; | |
304 | ||
305 | if (!policy->nodes.set && | |
306 | !IN_SET(numa_policy_get_type(policy), MPOL_DEFAULT, MPOL_LOCAL, MPOL_PREFERRED)) | |
307 | return false; | |
308 | ||
309 | if (policy->nodes.set && | |
310 | numa_policy_get_type(policy) == MPOL_PREFERRED && | |
311 | CPU_COUNT_S(policy->nodes.allocated, policy->nodes.set) != 1) | |
312 | return false; | |
313 | ||
314 | return true; | |
315 | } | |
316 | ||
317 | static int numa_policy_to_mempolicy(const NUMAPolicy *policy, unsigned long *ret_maxnode, unsigned long **ret_nodes) { | |
318 | unsigned node, bits = 0, ulong_bits; | |
319 | _cleanup_free_ unsigned long *out = NULL; | |
320 | ||
321 | assert(policy); | |
322 | assert(ret_maxnode); | |
323 | assert(ret_nodes); | |
324 | ||
325 | if (IN_SET(numa_policy_get_type(policy), MPOL_DEFAULT, MPOL_LOCAL) || | |
326 | (numa_policy_get_type(policy) == MPOL_PREFERRED && !policy->nodes.set)) { | |
327 | *ret_nodes = NULL; | |
328 | *ret_maxnode = 0; | |
329 | return 0; | |
330 | } | |
331 | ||
332 | bits = policy->nodes.allocated * 8; | |
333 | ulong_bits = sizeof(unsigned long) * 8; | |
334 | ||
335 | out = new0(unsigned long, DIV_ROUND_UP(policy->nodes.allocated, sizeof(unsigned long))); | |
336 | if (!out) | |
337 | return -ENOMEM; | |
338 | ||
339 | /* We don't make any assumptions about internal type libc is using to store NUMA node mask. | |
340 | Hence we need to convert the node mask to the representation expected by set_mempolicy() */ | |
341 | for (node = 0; node < bits; node++) | |
342 | if (CPU_ISSET_S(node, policy->nodes.allocated, policy->nodes.set)) | |
343 | out[node / ulong_bits] |= 1ul << (node % ulong_bits); | |
344 | ||
345 | *ret_nodes = TAKE_PTR(out); | |
346 | *ret_maxnode = bits + 1; | |
347 | return 0; | |
348 | } | |
349 | ||
350 | int apply_numa_policy(const NUMAPolicy *policy) { | |
351 | int r; | |
352 | _cleanup_free_ unsigned long *nodes = NULL; | |
353 | unsigned long maxnode; | |
354 | ||
355 | assert(policy); | |
356 | ||
357 | if (get_mempolicy(NULL, NULL, 0, 0, 0) < 0 && errno == ENOSYS) | |
358 | return -EOPNOTSUPP; | |
359 | ||
360 | if (!numa_policy_is_valid(policy)) | |
361 | return -EINVAL; | |
362 | ||
363 | r = numa_policy_to_mempolicy(policy, &maxnode, &nodes); | |
364 | if (r < 0) | |
365 | return r; | |
366 | ||
367 | r = set_mempolicy(numa_policy_get_type(policy), nodes, maxnode); | |
368 | if (r < 0) | |
369 | return -errno; | |
370 | ||
371 | return 0; | |
372 | } | |
373 | ||
374 | static const char* const mpol_table[] = { | |
375 | [MPOL_DEFAULT] = "default", | |
376 | [MPOL_PREFERRED] = "preferred", | |
377 | [MPOL_BIND] = "bind", | |
378 | [MPOL_INTERLEAVE] = "interleave", | |
379 | [MPOL_LOCAL] = "local", | |
380 | }; | |
381 | ||
382 | DEFINE_STRING_TABLE_LOOKUP(mpol, int); |