| 1 | // SPDX-License-Identifier: GPL-2.0-only | 
|---|
| 2 | /* | 
|---|
| 3 | *  kernel/sched/syscalls.c | 
|---|
| 4 | * | 
|---|
| 5 | *  Core kernel scheduler syscalls related code | 
|---|
| 6 | * | 
|---|
| 7 | *  Copyright (C) 1991-2002  Linus Torvalds | 
|---|
| 8 | *  Copyright (C) 1998-2024  Ingo Molnar, Red Hat | 
|---|
| 9 | */ | 
|---|
| 10 | #include <linux/sched.h> | 
|---|
| 11 | #include <linux/cpuset.h> | 
|---|
| 12 | #include <linux/sched/debug.h> | 
|---|
| 13 |  | 
|---|
| 14 | #include <uapi/linux/sched/types.h> | 
|---|
| 15 |  | 
|---|
| 16 | #include "sched.h" | 
|---|
| 17 | #include "autogroup.h" | 
|---|
| 18 |  | 
|---|
| 19 | static inline int __normal_prio(int policy, int rt_prio, int nice) | 
|---|
| 20 | { | 
|---|
| 21 | int prio; | 
|---|
| 22 |  | 
|---|
| 23 | if (dl_policy(policy)) | 
|---|
| 24 | prio = MAX_DL_PRIO - 1; | 
|---|
| 25 | else if (rt_policy(policy)) | 
|---|
| 26 | prio = MAX_RT_PRIO - 1 - rt_prio; | 
|---|
| 27 | else | 
|---|
| 28 | prio = NICE_TO_PRIO(nice); | 
|---|
| 29 |  | 
|---|
| 30 | return prio; | 
|---|
| 31 | } | 
|---|
| 32 |  | 
|---|
| 33 | /* | 
|---|
| 34 | * Calculate the expected normal priority: i.e. priority | 
|---|
| 35 | * without taking RT-inheritance into account. Might be | 
|---|
| 36 | * boosted by interactivity modifiers. Changes upon fork, | 
|---|
| 37 | * setprio syscalls, and whenever the interactivity | 
|---|
| 38 | * estimator recalculates. | 
|---|
| 39 | */ | 
|---|
| 40 | static inline int normal_prio(struct task_struct *p) | 
|---|
| 41 | { | 
|---|
| 42 | return __normal_prio(policy: p->policy, rt_prio: p->rt_priority, PRIO_TO_NICE(p->static_prio)); | 
|---|
| 43 | } | 
|---|
| 44 |  | 
|---|
| 45 | /* | 
|---|
| 46 | * Calculate the current priority, i.e. the priority | 
|---|
| 47 | * taken into account by the scheduler. This value might | 
|---|
| 48 | * be boosted by RT tasks, or might be boosted by | 
|---|
| 49 | * interactivity modifiers. Will be RT if the task got | 
|---|
| 50 | * RT-boosted. If not then it returns p->normal_prio. | 
|---|
| 51 | */ | 
|---|
| 52 | static int effective_prio(struct task_struct *p) | 
|---|
| 53 | { | 
|---|
| 54 | p->normal_prio = normal_prio(p); | 
|---|
| 55 | /* | 
|---|
| 56 | * If we are RT tasks or we were boosted to RT priority, | 
|---|
| 57 | * keep the priority unchanged. Otherwise, update priority | 
|---|
| 58 | * to the normal priority: | 
|---|
| 59 | */ | 
|---|
| 60 | if (!rt_or_dl_prio(prio: p->prio)) | 
|---|
| 61 | return p->normal_prio; | 
|---|
| 62 | return p->prio; | 
|---|
| 63 | } | 
|---|
| 64 |  | 
|---|
| 65 | void set_user_nice(struct task_struct *p, long nice) | 
|---|
| 66 | { | 
|---|
| 67 | bool queued, running; | 
|---|
| 68 | struct rq *rq; | 
|---|
| 69 | int old_prio; | 
|---|
| 70 |  | 
|---|
| 71 | if (task_nice(p) == nice || nice < MIN_NICE || nice > MAX_NICE) | 
|---|
| 72 | return; | 
|---|
| 73 | /* | 
|---|
| 74 | * We have to be careful, if called from sys_setpriority(), | 
|---|
| 75 | * the task might be in the middle of scheduling on another CPU. | 
|---|
| 76 | */ | 
|---|
| 77 | CLASS(task_rq_lock, rq_guard)(l: p); | 
|---|
| 78 | rq = rq_guard.rq; | 
|---|
| 79 |  | 
|---|
| 80 | update_rq_clock(rq); | 
|---|
| 81 |  | 
|---|
| 82 | /* | 
|---|
| 83 | * The RT priorities are set via sched_setscheduler(), but we still | 
|---|
| 84 | * allow the 'normal' nice value to be set - but as expected | 
|---|
| 85 | * it won't have any effect on scheduling until the task is | 
|---|
| 86 | * SCHED_DEADLINE, SCHED_FIFO or SCHED_RR: | 
|---|
| 87 | */ | 
|---|
| 88 | if (task_has_dl_policy(p) || task_has_rt_policy(p)) { | 
|---|
| 89 | p->static_prio = NICE_TO_PRIO(nice); | 
|---|
| 90 | return; | 
|---|
| 91 | } | 
|---|
| 92 |  | 
|---|
| 93 | queued = task_on_rq_queued(p); | 
|---|
| 94 | running = task_current_donor(rq, p); | 
|---|
| 95 | if (queued) | 
|---|
| 96 | dequeue_task(rq, p, DEQUEUE_SAVE | DEQUEUE_NOCLOCK); | 
|---|
| 97 | if (running) | 
|---|
| 98 | put_prev_task(rq, prev: p); | 
|---|
| 99 |  | 
|---|
| 100 | p->static_prio = NICE_TO_PRIO(nice); | 
|---|
| 101 | set_load_weight(p, update_load: true); | 
|---|
| 102 | old_prio = p->prio; | 
|---|
| 103 | p->prio = effective_prio(p); | 
|---|
| 104 |  | 
|---|
| 105 | if (queued) | 
|---|
| 106 | enqueue_task(rq, p, ENQUEUE_RESTORE | ENQUEUE_NOCLOCK); | 
|---|
| 107 | if (running) | 
|---|
| 108 | set_next_task(rq, next: p); | 
|---|
| 109 |  | 
|---|
| 110 | /* | 
|---|
| 111 | * If the task increased its priority or is running and | 
|---|
| 112 | * lowered its priority, then reschedule its CPU: | 
|---|
| 113 | */ | 
|---|
| 114 | p->sched_class->prio_changed(rq, p, old_prio); | 
|---|
| 115 | } | 
|---|
| 116 | EXPORT_SYMBOL(set_user_nice); | 
|---|
| 117 |  | 
|---|
| 118 | /* | 
|---|
| 119 | * is_nice_reduction - check if nice value is an actual reduction | 
|---|
| 120 | * | 
|---|
| 121 | * Similar to can_nice() but does not perform a capability check. | 
|---|
| 122 | * | 
|---|
| 123 | * @p: task | 
|---|
| 124 | * @nice: nice value | 
|---|
| 125 | */ | 
|---|
| 126 | static bool is_nice_reduction(const struct task_struct *p, const int nice) | 
|---|
| 127 | { | 
|---|
| 128 | /* Convert nice value [19,-20] to rlimit style value [1,40]: */ | 
|---|
| 129 | int nice_rlim = nice_to_rlimit(nice); | 
|---|
| 130 |  | 
|---|
| 131 | return (nice_rlim <= task_rlimit(task: p, RLIMIT_NICE)); | 
|---|
| 132 | } | 
|---|
| 133 |  | 
|---|
| 134 | /* | 
|---|
| 135 | * can_nice - check if a task can reduce its nice value | 
|---|
| 136 | * @p: task | 
|---|
| 137 | * @nice: nice value | 
|---|
| 138 | */ | 
|---|
| 139 | int can_nice(const struct task_struct *p, const int nice) | 
|---|
| 140 | { | 
|---|
| 141 | return is_nice_reduction(p, nice) || capable(CAP_SYS_NICE); | 
|---|
| 142 | } | 
|---|
| 143 |  | 
|---|
| 144 | #ifdef __ARCH_WANT_SYS_NICE | 
|---|
| 145 |  | 
|---|
| 146 | /* | 
|---|
| 147 | * sys_nice - change the priority of the current process. | 
|---|
| 148 | * @increment: priority increment | 
|---|
| 149 | * | 
|---|
| 150 | * sys_setpriority is a more generic, but much slower function that | 
|---|
| 151 | * does similar things. | 
|---|
| 152 | */ | 
|---|
| 153 | SYSCALL_DEFINE1(nice, int, increment) | 
|---|
| 154 | { | 
|---|
| 155 | long nice, retval; | 
|---|
| 156 |  | 
|---|
| 157 | /* | 
|---|
| 158 | * Setpriority might change our priority at the same moment. | 
|---|
| 159 | * We don't have to worry. Conceptually one call occurs first | 
|---|
| 160 | * and we have a single winner. | 
|---|
| 161 | */ | 
|---|
| 162 | increment = clamp(increment, -NICE_WIDTH, NICE_WIDTH); | 
|---|
| 163 | nice = task_nice(current) + increment; | 
|---|
| 164 |  | 
|---|
| 165 | nice = clamp_val(nice, MIN_NICE, MAX_NICE); | 
|---|
| 166 | if (increment < 0 && !can_nice(current, nice)) | 
|---|
| 167 | return -EPERM; | 
|---|
| 168 |  | 
|---|
| 169 | retval = security_task_setnice(current, nice); | 
|---|
| 170 | if (retval) | 
|---|
| 171 | return retval; | 
|---|
| 172 |  | 
|---|
| 173 | set_user_nice(current, nice); | 
|---|
| 174 | return 0; | 
|---|
| 175 | } | 
|---|
| 176 |  | 
|---|
| 177 | #endif /* __ARCH_WANT_SYS_NICE */ | 
|---|
| 178 |  | 
|---|
| 179 | /** | 
|---|
| 180 | * task_prio - return the priority value of a given task. | 
|---|
| 181 | * @p: the task in question. | 
|---|
| 182 | * | 
|---|
| 183 | * Return: The priority value as seen by users in /proc. | 
|---|
| 184 | * | 
|---|
| 185 | * sched policy         return value   kernel prio    user prio/nice | 
|---|
| 186 | * | 
|---|
| 187 | * normal, batch, idle     [0 ... 39]  [100 ... 139]          0/[-20 ... 19] | 
|---|
| 188 | * fifo, rr             [-2 ... -100]     [98 ... 0]  [1 ... 99] | 
|---|
| 189 | * deadline                     -101             -1           0 | 
|---|
| 190 | */ | 
|---|
| 191 | int task_prio(const struct task_struct *p) | 
|---|
| 192 | { | 
|---|
| 193 | return p->prio - MAX_RT_PRIO; | 
|---|
| 194 | } | 
|---|
| 195 |  | 
|---|
| 196 | /** | 
|---|
| 197 | * idle_cpu - is a given CPU idle currently? | 
|---|
| 198 | * @cpu: the processor in question. | 
|---|
| 199 | * | 
|---|
| 200 | * Return: 1 if the CPU is currently idle. 0 otherwise. | 
|---|
| 201 | */ | 
|---|
| 202 | int idle_cpu(int cpu) | 
|---|
| 203 | { | 
|---|
| 204 | struct rq *rq = cpu_rq(cpu); | 
|---|
| 205 |  | 
|---|
| 206 | if (rq->curr != rq->idle) | 
|---|
| 207 | return 0; | 
|---|
| 208 |  | 
|---|
| 209 | if (rq->nr_running) | 
|---|
| 210 | return 0; | 
|---|
| 211 |  | 
|---|
| 212 | if (rq->ttwu_pending) | 
|---|
| 213 | return 0; | 
|---|
| 214 |  | 
|---|
| 215 | return 1; | 
|---|
| 216 | } | 
|---|
| 217 |  | 
|---|
| 218 | /** | 
|---|
| 219 | * available_idle_cpu - is a given CPU idle for enqueuing work. | 
|---|
| 220 | * @cpu: the CPU in question. | 
|---|
| 221 | * | 
|---|
| 222 | * Return: 1 if the CPU is currently idle. 0 otherwise. | 
|---|
| 223 | */ | 
|---|
| 224 | int available_idle_cpu(int cpu) | 
|---|
| 225 | { | 
|---|
| 226 | if (!idle_cpu(cpu)) | 
|---|
| 227 | return 0; | 
|---|
| 228 |  | 
|---|
| 229 | if (vcpu_is_preempted(cpu)) | 
|---|
| 230 | return 0; | 
|---|
| 231 |  | 
|---|
| 232 | return 1; | 
|---|
| 233 | } | 
|---|
| 234 |  | 
|---|
| 235 | /** | 
|---|
| 236 | * idle_task - return the idle task for a given CPU. | 
|---|
| 237 | * @cpu: the processor in question. | 
|---|
| 238 | * | 
|---|
| 239 | * Return: The idle task for the CPU @cpu. | 
|---|
| 240 | */ | 
|---|
| 241 | struct task_struct *idle_task(int cpu) | 
|---|
| 242 | { | 
|---|
| 243 | return cpu_rq(cpu)->idle; | 
|---|
| 244 | } | 
|---|
| 245 |  | 
|---|
| 246 | #ifdef CONFIG_SCHED_CORE | 
|---|
| 247 | int sched_core_idle_cpu(int cpu) | 
|---|
| 248 | { | 
|---|
| 249 | struct rq *rq = cpu_rq(cpu); | 
|---|
| 250 |  | 
|---|
| 251 | if (sched_core_enabled(rq) && rq->curr == rq->idle) | 
|---|
| 252 | return 1; | 
|---|
| 253 |  | 
|---|
| 254 | return idle_cpu(cpu); | 
|---|
| 255 | } | 
|---|
| 256 | #endif /* CONFIG_SCHED_CORE */ | 
|---|
| 257 |  | 
|---|
| 258 | /** | 
|---|
| 259 | * find_process_by_pid - find a process with a matching PID value. | 
|---|
| 260 | * @pid: the pid in question. | 
|---|
| 261 | * | 
|---|
| 262 | * The task of @pid, if found. %NULL otherwise. | 
|---|
| 263 | */ | 
|---|
| 264 | static struct task_struct *find_process_by_pid(pid_t pid) | 
|---|
| 265 | { | 
|---|
| 266 | return pid ? find_task_by_vpid(nr: pid) : current; | 
|---|
| 267 | } | 
|---|
| 268 |  | 
|---|
| 269 | static struct task_struct *find_get_task(pid_t pid) | 
|---|
| 270 | { | 
|---|
| 271 | struct task_struct *p; | 
|---|
| 272 | guard(rcu)(); | 
|---|
| 273 |  | 
|---|
| 274 | p = find_process_by_pid(pid); | 
|---|
| 275 | if (likely(p)) | 
|---|
| 276 | get_task_struct(t: p); | 
|---|
| 277 |  | 
|---|
| 278 | return p; | 
|---|
| 279 | } | 
|---|
| 280 |  | 
|---|
| 281 | DEFINE_CLASS(find_get_task, struct task_struct *, if (_T) put_task_struct(_T), | 
|---|
| 282 | find_get_task(pid), pid_t pid) | 
|---|
| 283 |  | 
|---|
| 284 | /* | 
|---|
| 285 | * sched_setparam() passes in -1 for its policy, to let the functions | 
|---|
| 286 | * it calls know not to change it. | 
|---|
| 287 | */ | 
|---|
| 288 | #define SETPARAM_POLICY	-1 | 
|---|
| 289 |  | 
|---|
| 290 | static void __setscheduler_params(struct task_struct *p, | 
|---|
| 291 | const struct sched_attr *attr) | 
|---|
| 292 | { | 
|---|
| 293 | int policy = attr->sched_policy; | 
|---|
| 294 |  | 
|---|
| 295 | if (policy == SETPARAM_POLICY) | 
|---|
| 296 | policy = p->policy; | 
|---|
| 297 |  | 
|---|
| 298 | p->policy = policy; | 
|---|
| 299 |  | 
|---|
| 300 | if (dl_policy(policy)) | 
|---|
| 301 | __setparam_dl(p, attr); | 
|---|
| 302 | else if (fair_policy(policy)) | 
|---|
| 303 | __setparam_fair(p, attr); | 
|---|
| 304 |  | 
|---|
| 305 | /* rt-policy tasks do not have a timerslack */ | 
|---|
| 306 | if (rt_or_dl_task_policy(tsk: p)) { | 
|---|
| 307 | p->timer_slack_ns = 0; | 
|---|
| 308 | } else if (p->timer_slack_ns == 0) { | 
|---|
| 309 | /* when switching back to non-rt policy, restore timerslack */ | 
|---|
| 310 | p->timer_slack_ns = p->default_timer_slack_ns; | 
|---|
| 311 | } | 
|---|
| 312 |  | 
|---|
| 313 | /* | 
|---|
| 314 | * __sched_setscheduler() ensures attr->sched_priority == 0 when | 
|---|
| 315 | * !rt_policy. Always setting this ensures that things like | 
|---|
| 316 | * getparam()/getattr() don't report silly values for !rt tasks. | 
|---|
| 317 | */ | 
|---|
| 318 | p->rt_priority = attr->sched_priority; | 
|---|
| 319 | p->normal_prio = normal_prio(p); | 
|---|
| 320 | set_load_weight(p, update_load: true); | 
|---|
| 321 | } | 
|---|
| 322 |  | 
|---|
| 323 | /* | 
|---|
| 324 | * Check the target process has a UID that matches the current process's: | 
|---|
| 325 | */ | 
|---|
| 326 | static bool check_same_owner(struct task_struct *p) | 
|---|
| 327 | { | 
|---|
| 328 | const struct cred *cred = current_cred(), *pcred; | 
|---|
| 329 | guard(rcu)(); | 
|---|
| 330 |  | 
|---|
| 331 | pcred = __task_cred(p); | 
|---|
| 332 | return (uid_eq(left: cred->euid, right: pcred->euid) || | 
|---|
| 333 | uid_eq(left: cred->euid, right: pcred->uid)); | 
|---|
| 334 | } | 
|---|
| 335 |  | 
|---|
| 336 | #ifdef CONFIG_UCLAMP_TASK | 
|---|
| 337 |  | 
|---|
| 338 | static int uclamp_validate(struct task_struct *p, | 
|---|
| 339 | const struct sched_attr *attr) | 
|---|
| 340 | { | 
|---|
| 341 | int util_min = p->uclamp_req[UCLAMP_MIN].value; | 
|---|
| 342 | int util_max = p->uclamp_req[UCLAMP_MAX].value; | 
|---|
| 343 |  | 
|---|
| 344 | if (attr->sched_flags & SCHED_FLAG_UTIL_CLAMP_MIN) { | 
|---|
| 345 | util_min = attr->sched_util_min; | 
|---|
| 346 |  | 
|---|
| 347 | if (util_min + 1 > SCHED_CAPACITY_SCALE + 1) | 
|---|
| 348 | return -EINVAL; | 
|---|
| 349 | } | 
|---|
| 350 |  | 
|---|
| 351 | if (attr->sched_flags & SCHED_FLAG_UTIL_CLAMP_MAX) { | 
|---|
| 352 | util_max = attr->sched_util_max; | 
|---|
| 353 |  | 
|---|
| 354 | if (util_max + 1 > SCHED_CAPACITY_SCALE + 1) | 
|---|
| 355 | return -EINVAL; | 
|---|
| 356 | } | 
|---|
| 357 |  | 
|---|
| 358 | if (util_min != -1 && util_max != -1 && util_min > util_max) | 
|---|
| 359 | return -EINVAL; | 
|---|
| 360 |  | 
|---|
| 361 | /* | 
|---|
| 362 | * We have valid uclamp attributes; make sure uclamp is enabled. | 
|---|
| 363 | * | 
|---|
| 364 | * We need to do that here, because enabling static branches is a | 
|---|
| 365 | * blocking operation which obviously cannot be done while holding | 
|---|
| 366 | * scheduler locks. | 
|---|
| 367 | */ | 
|---|
| 368 | sched_uclamp_enable(); | 
|---|
| 369 |  | 
|---|
| 370 | return 0; | 
|---|
| 371 | } | 
|---|
| 372 |  | 
|---|
| 373 | static bool uclamp_reset(const struct sched_attr *attr, | 
|---|
| 374 | enum uclamp_id clamp_id, | 
|---|
| 375 | struct uclamp_se *uc_se) | 
|---|
| 376 | { | 
|---|
| 377 | /* Reset on sched class change for a non user-defined clamp value. */ | 
|---|
| 378 | if (likely(!(attr->sched_flags & SCHED_FLAG_UTIL_CLAMP)) && | 
|---|
| 379 | !uc_se->user_defined) | 
|---|
| 380 | return true; | 
|---|
| 381 |  | 
|---|
| 382 | /* Reset on sched_util_{min,max} == -1. */ | 
|---|
| 383 | if (clamp_id == UCLAMP_MIN && | 
|---|
| 384 | attr->sched_flags & SCHED_FLAG_UTIL_CLAMP_MIN && | 
|---|
| 385 | attr->sched_util_min == -1) { | 
|---|
| 386 | return true; | 
|---|
| 387 | } | 
|---|
| 388 |  | 
|---|
| 389 | if (clamp_id == UCLAMP_MAX && | 
|---|
| 390 | attr->sched_flags & SCHED_FLAG_UTIL_CLAMP_MAX && | 
|---|
| 391 | attr->sched_util_max == -1) { | 
|---|
| 392 | return true; | 
|---|
| 393 | } | 
|---|
| 394 |  | 
|---|
| 395 | return false; | 
|---|
| 396 | } | 
|---|
| 397 |  | 
|---|
| 398 | static void __setscheduler_uclamp(struct task_struct *p, | 
|---|
| 399 | const struct sched_attr *attr) | 
|---|
| 400 | { | 
|---|
| 401 | enum uclamp_id clamp_id; | 
|---|
| 402 |  | 
|---|
| 403 | for_each_clamp_id(clamp_id) { | 
|---|
| 404 | struct uclamp_se *uc_se = &p->uclamp_req[clamp_id]; | 
|---|
| 405 | unsigned int value; | 
|---|
| 406 |  | 
|---|
| 407 | if (!uclamp_reset(attr, clamp_id, uc_se)) | 
|---|
| 408 | continue; | 
|---|
| 409 |  | 
|---|
| 410 | /* | 
|---|
| 411 | * RT by default have a 100% boost value that could be modified | 
|---|
| 412 | * at runtime. | 
|---|
| 413 | */ | 
|---|
| 414 | if (unlikely(rt_task(p) && clamp_id == UCLAMP_MIN)) | 
|---|
| 415 | value = sysctl_sched_uclamp_util_min_rt_default; | 
|---|
| 416 | else | 
|---|
| 417 | value = uclamp_none(clamp_id); | 
|---|
| 418 |  | 
|---|
| 419 | uclamp_se_set(uc_se, value, false); | 
|---|
| 420 |  | 
|---|
| 421 | } | 
|---|
| 422 |  | 
|---|
| 423 | if (likely(!(attr->sched_flags & SCHED_FLAG_UTIL_CLAMP))) | 
|---|
| 424 | return; | 
|---|
| 425 |  | 
|---|
| 426 | if (attr->sched_flags & SCHED_FLAG_UTIL_CLAMP_MIN && | 
|---|
| 427 | attr->sched_util_min != -1) { | 
|---|
| 428 | uclamp_se_set(&p->uclamp_req[UCLAMP_MIN], | 
|---|
| 429 | attr->sched_util_min, true); | 
|---|
| 430 | } | 
|---|
| 431 |  | 
|---|
| 432 | if (attr->sched_flags & SCHED_FLAG_UTIL_CLAMP_MAX && | 
|---|
| 433 | attr->sched_util_max != -1) { | 
|---|
| 434 | uclamp_se_set(&p->uclamp_req[UCLAMP_MAX], | 
|---|
| 435 | attr->sched_util_max, true); | 
|---|
| 436 | } | 
|---|
| 437 | } | 
|---|
| 438 |  | 
|---|
| 439 | #else /* !CONFIG_UCLAMP_TASK: */ | 
|---|
| 440 |  | 
|---|
| 441 | static inline int uclamp_validate(struct task_struct *p, | 
|---|
| 442 | const struct sched_attr *attr) | 
|---|
| 443 | { | 
|---|
| 444 | return -EOPNOTSUPP; | 
|---|
| 445 | } | 
|---|
| 446 | static void __setscheduler_uclamp(struct task_struct *p, | 
|---|
| 447 | const struct sched_attr *attr) { } | 
|---|
| 448 | #endif /* !CONFIG_UCLAMP_TASK */ | 
|---|
| 449 |  | 
|---|
| 450 | /* | 
|---|
| 451 | * Allow unprivileged RT tasks to decrease priority. | 
|---|
| 452 | * Only issue a capable test if needed and only once to avoid an audit | 
|---|
| 453 | * event on permitted non-privileged operations: | 
|---|
| 454 | */ | 
|---|
| 455 | static int user_check_sched_setscheduler(struct task_struct *p, | 
|---|
| 456 | const struct sched_attr *attr, | 
|---|
| 457 | int policy, int reset_on_fork) | 
|---|
| 458 | { | 
|---|
| 459 | if (fair_policy(policy)) { | 
|---|
| 460 | if (attr->sched_nice < task_nice(p) && | 
|---|
| 461 | !is_nice_reduction(p, nice: attr->sched_nice)) | 
|---|
| 462 | goto req_priv; | 
|---|
| 463 | } | 
|---|
| 464 |  | 
|---|
| 465 | if (rt_policy(policy)) { | 
|---|
| 466 | unsigned long rlim_rtprio = task_rlimit(task: p, RLIMIT_RTPRIO); | 
|---|
| 467 |  | 
|---|
| 468 | /* Can't set/change the rt policy: */ | 
|---|
| 469 | if (policy != p->policy && !rlim_rtprio) | 
|---|
| 470 | goto req_priv; | 
|---|
| 471 |  | 
|---|
| 472 | /* Can't increase priority: */ | 
|---|
| 473 | if (attr->sched_priority > p->rt_priority && | 
|---|
| 474 | attr->sched_priority > rlim_rtprio) | 
|---|
| 475 | goto req_priv; | 
|---|
| 476 | } | 
|---|
| 477 |  | 
|---|
| 478 | /* | 
|---|
| 479 | * Can't set/change SCHED_DEADLINE policy at all for now | 
|---|
| 480 | * (safest behavior); in the future we would like to allow | 
|---|
| 481 | * unprivileged DL tasks to increase their relative deadline | 
|---|
| 482 | * or reduce their runtime (both ways reducing utilization) | 
|---|
| 483 | */ | 
|---|
| 484 | if (dl_policy(policy)) | 
|---|
| 485 | goto req_priv; | 
|---|
| 486 |  | 
|---|
| 487 | /* | 
|---|
| 488 | * Treat SCHED_IDLE as nice 20. Only allow a switch to | 
|---|
| 489 | * SCHED_NORMAL if the RLIMIT_NICE would normally permit it. | 
|---|
| 490 | */ | 
|---|
| 491 | if (task_has_idle_policy(p) && !idle_policy(policy)) { | 
|---|
| 492 | if (!is_nice_reduction(p, nice: task_nice(p))) | 
|---|
| 493 | goto req_priv; | 
|---|
| 494 | } | 
|---|
| 495 |  | 
|---|
| 496 | /* Can't change other user's priorities: */ | 
|---|
| 497 | if (!check_same_owner(p)) | 
|---|
| 498 | goto req_priv; | 
|---|
| 499 |  | 
|---|
| 500 | /* Normal users shall not reset the sched_reset_on_fork flag: */ | 
|---|
| 501 | if (p->sched_reset_on_fork && !reset_on_fork) | 
|---|
| 502 | goto req_priv; | 
|---|
| 503 |  | 
|---|
| 504 | return 0; | 
|---|
| 505 |  | 
|---|
| 506 | req_priv: | 
|---|
| 507 | if (!capable(CAP_SYS_NICE)) | 
|---|
| 508 | return -EPERM; | 
|---|
| 509 |  | 
|---|
| 510 | return 0; | 
|---|
| 511 | } | 
|---|
| 512 |  | 
|---|
| 513 | int __sched_setscheduler(struct task_struct *p, | 
|---|
| 514 | const struct sched_attr *attr, | 
|---|
| 515 | bool user, bool pi) | 
|---|
| 516 | { | 
|---|
| 517 | int oldpolicy = -1, policy = attr->sched_policy; | 
|---|
| 518 | int retval, oldprio, newprio, queued, running; | 
|---|
| 519 | const struct sched_class *prev_class, *next_class; | 
|---|
| 520 | struct balance_callback *head; | 
|---|
| 521 | struct rq_flags rf; | 
|---|
| 522 | int reset_on_fork; | 
|---|
| 523 | int queue_flags = DEQUEUE_SAVE | DEQUEUE_MOVE | DEQUEUE_NOCLOCK; | 
|---|
| 524 | struct rq *rq; | 
|---|
| 525 | bool cpuset_locked = false; | 
|---|
| 526 |  | 
|---|
| 527 | /* The pi code expects interrupts enabled */ | 
|---|
| 528 | BUG_ON(pi && in_interrupt()); | 
|---|
| 529 | recheck: | 
|---|
| 530 | /* Double check policy once rq lock held: */ | 
|---|
| 531 | if (policy < 0) { | 
|---|
| 532 | reset_on_fork = p->sched_reset_on_fork; | 
|---|
| 533 | policy = oldpolicy = p->policy; | 
|---|
| 534 | } else { | 
|---|
| 535 | reset_on_fork = !!(attr->sched_flags & SCHED_FLAG_RESET_ON_FORK); | 
|---|
| 536 |  | 
|---|
| 537 | if (!valid_policy(policy)) | 
|---|
| 538 | return -EINVAL; | 
|---|
| 539 | } | 
|---|
| 540 |  | 
|---|
| 541 | if (attr->sched_flags & ~(SCHED_FLAG_ALL | SCHED_FLAG_SUGOV)) | 
|---|
| 542 | return -EINVAL; | 
|---|
| 543 |  | 
|---|
| 544 | /* | 
|---|
| 545 | * Valid priorities for SCHED_FIFO and SCHED_RR are | 
|---|
| 546 | * 1..MAX_RT_PRIO-1, valid priority for SCHED_NORMAL, | 
|---|
| 547 | * SCHED_BATCH and SCHED_IDLE is 0. | 
|---|
| 548 | */ | 
|---|
| 549 | if (attr->sched_priority > MAX_RT_PRIO-1) | 
|---|
| 550 | return -EINVAL; | 
|---|
| 551 | if ((dl_policy(policy) && !__checkparam_dl(attr)) || | 
|---|
| 552 | (rt_policy(policy) != (attr->sched_priority != 0))) | 
|---|
| 553 | return -EINVAL; | 
|---|
| 554 |  | 
|---|
| 555 | if (user) { | 
|---|
| 556 | retval = user_check_sched_setscheduler(p, attr, policy, reset_on_fork); | 
|---|
| 557 | if (retval) | 
|---|
| 558 | return retval; | 
|---|
| 559 |  | 
|---|
| 560 | if (attr->sched_flags & SCHED_FLAG_SUGOV) | 
|---|
| 561 | return -EINVAL; | 
|---|
| 562 |  | 
|---|
| 563 | retval = security_task_setscheduler(p); | 
|---|
| 564 | if (retval) | 
|---|
| 565 | return retval; | 
|---|
| 566 | } | 
|---|
| 567 |  | 
|---|
| 568 | /* Update task specific "requested" clamps */ | 
|---|
| 569 | if (attr->sched_flags & SCHED_FLAG_UTIL_CLAMP) { | 
|---|
| 570 | retval = uclamp_validate(p, attr); | 
|---|
| 571 | if (retval) | 
|---|
| 572 | return retval; | 
|---|
| 573 | } | 
|---|
| 574 |  | 
|---|
| 575 | /* | 
|---|
| 576 | * SCHED_DEADLINE bandwidth accounting relies on stable cpusets | 
|---|
| 577 | * information. | 
|---|
| 578 | */ | 
|---|
| 579 | if (dl_policy(policy) || dl_policy(policy: p->policy)) { | 
|---|
| 580 | cpuset_locked = true; | 
|---|
| 581 | cpuset_lock(); | 
|---|
| 582 | } | 
|---|
| 583 |  | 
|---|
| 584 | /* | 
|---|
| 585 | * Make sure no PI-waiters arrive (or leave) while we are | 
|---|
| 586 | * changing the priority of the task: | 
|---|
| 587 | * | 
|---|
| 588 | * To be able to change p->policy safely, the appropriate | 
|---|
| 589 | * runqueue lock must be held. | 
|---|
| 590 | */ | 
|---|
| 591 | rq = task_rq_lock(p, rf: &rf); | 
|---|
| 592 | update_rq_clock(rq); | 
|---|
| 593 |  | 
|---|
| 594 | /* | 
|---|
| 595 | * Changing the policy of the stop threads its a very bad idea: | 
|---|
| 596 | */ | 
|---|
| 597 | if (p == rq->stop) { | 
|---|
| 598 | retval = -EINVAL; | 
|---|
| 599 | goto unlock; | 
|---|
| 600 | } | 
|---|
| 601 |  | 
|---|
| 602 | retval = scx_check_setscheduler(p, policy); | 
|---|
| 603 | if (retval) | 
|---|
| 604 | goto unlock; | 
|---|
| 605 |  | 
|---|
| 606 | /* | 
|---|
| 607 | * If not changing anything there's no need to proceed further, | 
|---|
| 608 | * but store a possible modification of reset_on_fork. | 
|---|
| 609 | */ | 
|---|
| 610 | if (unlikely(policy == p->policy)) { | 
|---|
| 611 | if (fair_policy(policy) && | 
|---|
| 612 | (attr->sched_nice != task_nice(p) || | 
|---|
| 613 | (attr->sched_runtime != p->se.slice))) | 
|---|
| 614 | goto change; | 
|---|
| 615 | if (rt_policy(policy) && attr->sched_priority != p->rt_priority) | 
|---|
| 616 | goto change; | 
|---|
| 617 | if (dl_policy(policy) && dl_param_changed(p, attr)) | 
|---|
| 618 | goto change; | 
|---|
| 619 | if (attr->sched_flags & SCHED_FLAG_UTIL_CLAMP) | 
|---|
| 620 | goto change; | 
|---|
| 621 |  | 
|---|
| 622 | p->sched_reset_on_fork = reset_on_fork; | 
|---|
| 623 | retval = 0; | 
|---|
| 624 | goto unlock; | 
|---|
| 625 | } | 
|---|
| 626 | change: | 
|---|
| 627 |  | 
|---|
| 628 | if (user) { | 
|---|
| 629 | #ifdef CONFIG_RT_GROUP_SCHED | 
|---|
| 630 | /* | 
|---|
| 631 | * Do not allow real-time tasks into groups that have no runtime | 
|---|
| 632 | * assigned. | 
|---|
| 633 | */ | 
|---|
| 634 | if (rt_group_sched_enabled() && | 
|---|
| 635 | rt_bandwidth_enabled() && rt_policy(policy) && | 
|---|
| 636 | task_group(p)->rt_bandwidth.rt_runtime == 0 && | 
|---|
| 637 | !task_group_is_autogroup(task_group(p))) { | 
|---|
| 638 | retval = -EPERM; | 
|---|
| 639 | goto unlock; | 
|---|
| 640 | } | 
|---|
| 641 | #endif /* CONFIG_RT_GROUP_SCHED */ | 
|---|
| 642 | if (dl_bandwidth_enabled() && dl_policy(policy) && | 
|---|
| 643 | !(attr->sched_flags & SCHED_FLAG_SUGOV)) { | 
|---|
| 644 | cpumask_t *span = rq->rd->span; | 
|---|
| 645 |  | 
|---|
| 646 | /* | 
|---|
| 647 | * Don't allow tasks with an affinity mask smaller than | 
|---|
| 648 | * the entire root_domain to become SCHED_DEADLINE. We | 
|---|
| 649 | * will also fail if there's no bandwidth available. | 
|---|
| 650 | */ | 
|---|
| 651 | if (!cpumask_subset(src1p: span, src2p: p->cpus_ptr) || | 
|---|
| 652 | rq->rd->dl_bw.bw == 0) { | 
|---|
| 653 | retval = -EPERM; | 
|---|
| 654 | goto unlock; | 
|---|
| 655 | } | 
|---|
| 656 | } | 
|---|
| 657 | } | 
|---|
| 658 |  | 
|---|
| 659 | /* Re-check policy now with rq lock held: */ | 
|---|
| 660 | if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) { | 
|---|
| 661 | policy = oldpolicy = -1; | 
|---|
| 662 | task_rq_unlock(rq, p, rf: &rf); | 
|---|
| 663 | if (cpuset_locked) | 
|---|
| 664 | cpuset_unlock(); | 
|---|
| 665 | goto recheck; | 
|---|
| 666 | } | 
|---|
| 667 |  | 
|---|
| 668 | /* | 
|---|
| 669 | * If setscheduling to SCHED_DEADLINE (or changing the parameters | 
|---|
| 670 | * of a SCHED_DEADLINE task) we need to check if enough bandwidth | 
|---|
| 671 | * is available. | 
|---|
| 672 | */ | 
|---|
| 673 | if ((dl_policy(policy) || dl_task(p)) && sched_dl_overflow(p, policy, attr)) { | 
|---|
| 674 | retval = -EBUSY; | 
|---|
| 675 | goto unlock; | 
|---|
| 676 | } | 
|---|
| 677 |  | 
|---|
| 678 | p->sched_reset_on_fork = reset_on_fork; | 
|---|
| 679 | oldprio = p->prio; | 
|---|
| 680 |  | 
|---|
| 681 | newprio = __normal_prio(policy, rt_prio: attr->sched_priority, nice: attr->sched_nice); | 
|---|
| 682 | if (pi) { | 
|---|
| 683 | /* | 
|---|
| 684 | * Take priority boosted tasks into account. If the new | 
|---|
| 685 | * effective priority is unchanged, we just store the new | 
|---|
| 686 | * normal parameters and do not touch the scheduler class and | 
|---|
| 687 | * the runqueue. This will be done when the task deboost | 
|---|
| 688 | * itself. | 
|---|
| 689 | */ | 
|---|
| 690 | newprio = rt_effective_prio(p, prio: newprio); | 
|---|
| 691 | if (newprio == oldprio) | 
|---|
| 692 | queue_flags &= ~DEQUEUE_MOVE; | 
|---|
| 693 | } | 
|---|
| 694 |  | 
|---|
| 695 | prev_class = p->sched_class; | 
|---|
| 696 | next_class = __setscheduler_class(policy, prio: newprio); | 
|---|
| 697 |  | 
|---|
| 698 | if (prev_class != next_class && p->se.sched_delayed) | 
|---|
| 699 | dequeue_task(rq, p, DEQUEUE_SLEEP | DEQUEUE_DELAYED | DEQUEUE_NOCLOCK); | 
|---|
| 700 |  | 
|---|
| 701 | queued = task_on_rq_queued(p); | 
|---|
| 702 | running = task_current_donor(rq, p); | 
|---|
| 703 | if (queued) | 
|---|
| 704 | dequeue_task(rq, p, flags: queue_flags); | 
|---|
| 705 | if (running) | 
|---|
| 706 | put_prev_task(rq, prev: p); | 
|---|
| 707 |  | 
|---|
| 708 | if (!(attr->sched_flags & SCHED_FLAG_KEEP_PARAMS)) { | 
|---|
| 709 | __setscheduler_params(p, attr); | 
|---|
| 710 | p->sched_class = next_class; | 
|---|
| 711 | p->prio = newprio; | 
|---|
| 712 | } | 
|---|
| 713 | __setscheduler_uclamp(p, attr); | 
|---|
| 714 | check_class_changing(rq, p, prev_class); | 
|---|
| 715 |  | 
|---|
| 716 | if (queued) { | 
|---|
| 717 | /* | 
|---|
| 718 | * We enqueue to tail when the priority of a task is | 
|---|
| 719 | * increased (user space view). | 
|---|
| 720 | */ | 
|---|
| 721 | if (oldprio < p->prio) | 
|---|
| 722 | queue_flags |= ENQUEUE_HEAD; | 
|---|
| 723 |  | 
|---|
| 724 | enqueue_task(rq, p, flags: queue_flags); | 
|---|
| 725 | } | 
|---|
| 726 | if (running) | 
|---|
| 727 | set_next_task(rq, next: p); | 
|---|
| 728 |  | 
|---|
| 729 | check_class_changed(rq, p, prev_class, oldprio); | 
|---|
| 730 |  | 
|---|
| 731 | /* Avoid rq from going away on us: */ | 
|---|
| 732 | preempt_disable(); | 
|---|
| 733 | head = splice_balance_callbacks(rq); | 
|---|
| 734 | task_rq_unlock(rq, p, rf: &rf); | 
|---|
| 735 |  | 
|---|
| 736 | if (pi) { | 
|---|
| 737 | if (cpuset_locked) | 
|---|
| 738 | cpuset_unlock(); | 
|---|
| 739 | rt_mutex_adjust_pi(p); | 
|---|
| 740 | } | 
|---|
| 741 |  | 
|---|
| 742 | /* Run balance callbacks after we've adjusted the PI chain: */ | 
|---|
| 743 | balance_callbacks(rq, head); | 
|---|
| 744 | preempt_enable(); | 
|---|
| 745 |  | 
|---|
| 746 | return 0; | 
|---|
| 747 |  | 
|---|
| 748 | unlock: | 
|---|
| 749 | task_rq_unlock(rq, p, rf: &rf); | 
|---|
| 750 | if (cpuset_locked) | 
|---|
| 751 | cpuset_unlock(); | 
|---|
| 752 | return retval; | 
|---|
| 753 | } | 
|---|
| 754 |  | 
|---|
| 755 | static int _sched_setscheduler(struct task_struct *p, int policy, | 
|---|
| 756 | const struct sched_param *param, bool check) | 
|---|
| 757 | { | 
|---|
| 758 | struct sched_attr attr = { | 
|---|
| 759 | .sched_policy   = policy, | 
|---|
| 760 | .sched_priority = param->sched_priority, | 
|---|
| 761 | .sched_nice	= PRIO_TO_NICE(p->static_prio), | 
|---|
| 762 | }; | 
|---|
| 763 |  | 
|---|
| 764 | if (p->se.custom_slice) | 
|---|
| 765 | attr.sched_runtime = p->se.slice; | 
|---|
| 766 |  | 
|---|
| 767 | /* Fixup the legacy SCHED_RESET_ON_FORK hack. */ | 
|---|
| 768 | if ((policy != SETPARAM_POLICY) && (policy & SCHED_RESET_ON_FORK)) { | 
|---|
| 769 | attr.sched_flags |= SCHED_FLAG_RESET_ON_FORK; | 
|---|
| 770 | policy &= ~SCHED_RESET_ON_FORK; | 
|---|
| 771 | attr.sched_policy = policy; | 
|---|
| 772 | } | 
|---|
| 773 |  | 
|---|
| 774 | return __sched_setscheduler(p, attr: &attr, user: check, pi: true); | 
|---|
| 775 | } | 
|---|
| 776 | /** | 
|---|
| 777 | * sched_setscheduler - change the scheduling policy and/or RT priority of a thread. | 
|---|
| 778 | * @p: the task in question. | 
|---|
| 779 | * @policy: new policy. | 
|---|
| 780 | * @param: structure containing the new RT priority. | 
|---|
| 781 | * | 
|---|
| 782 | * Use sched_set_fifo(), read its comment. | 
|---|
| 783 | * | 
|---|
| 784 | * Return: 0 on success. An error code otherwise. | 
|---|
| 785 | * | 
|---|
| 786 | * NOTE that the task may be already dead. | 
|---|
| 787 | */ | 
|---|
| 788 | int sched_setscheduler(struct task_struct *p, int policy, | 
|---|
| 789 | const struct sched_param *param) | 
|---|
| 790 | { | 
|---|
| 791 | return _sched_setscheduler(p, policy, param, check: true); | 
|---|
| 792 | } | 
|---|
| 793 |  | 
|---|
| 794 | int sched_setattr(struct task_struct *p, const struct sched_attr *attr) | 
|---|
| 795 | { | 
|---|
| 796 | return __sched_setscheduler(p, attr, user: true, pi: true); | 
|---|
| 797 | } | 
|---|
| 798 |  | 
|---|
| 799 | int sched_setattr_nocheck(struct task_struct *p, const struct sched_attr *attr) | 
|---|
| 800 | { | 
|---|
| 801 | return __sched_setscheduler(p, attr, user: false, pi: true); | 
|---|
| 802 | } | 
|---|
| 803 | EXPORT_SYMBOL_GPL(sched_setattr_nocheck); | 
|---|
| 804 |  | 
|---|
| 805 | /** | 
|---|
| 806 | * sched_setscheduler_nocheck - change the scheduling policy and/or RT priority of a thread from kernel-space. | 
|---|
| 807 | * @p: the task in question. | 
|---|
| 808 | * @policy: new policy. | 
|---|
| 809 | * @param: structure containing the new RT priority. | 
|---|
| 810 | * | 
|---|
| 811 | * Just like sched_setscheduler, only don't bother checking if the | 
|---|
| 812 | * current context has permission.  For example, this is needed in | 
|---|
| 813 | * stop_machine(): we create temporary high priority worker threads, | 
|---|
| 814 | * but our caller might not have that capability. | 
|---|
| 815 | * | 
|---|
| 816 | * Return: 0 on success. An error code otherwise. | 
|---|
| 817 | */ | 
|---|
| 818 | int sched_setscheduler_nocheck(struct task_struct *p, int policy, | 
|---|
| 819 | const struct sched_param *param) | 
|---|
| 820 | { | 
|---|
| 821 | return _sched_setscheduler(p, policy, param, check: false); | 
|---|
| 822 | } | 
|---|
| 823 |  | 
|---|
| 824 | /* | 
|---|
| 825 | * SCHED_FIFO is a broken scheduler model; that is, it is fundamentally | 
|---|
| 826 | * incapable of resource management, which is the one thing an OS really should | 
|---|
| 827 | * be doing. | 
|---|
| 828 | * | 
|---|
| 829 | * This is of course the reason it is limited to privileged users only. | 
|---|
| 830 | * | 
|---|
| 831 | * Worse still; it is fundamentally impossible to compose static priority | 
|---|
| 832 | * workloads. You cannot take two correctly working static prio workloads | 
|---|
| 833 | * and smash them together and still expect them to work. | 
|---|
| 834 | * | 
|---|
| 835 | * For this reason 'all' FIFO tasks the kernel creates are basically at: | 
|---|
| 836 | * | 
|---|
| 837 | *   MAX_RT_PRIO / 2 | 
|---|
| 838 | * | 
|---|
| 839 | * The administrator _MUST_ configure the system, the kernel simply doesn't | 
|---|
| 840 | * know enough information to make a sensible choice. | 
|---|
| 841 | */ | 
|---|
| 842 | void sched_set_fifo(struct task_struct *p) | 
|---|
| 843 | { | 
|---|
| 844 | struct sched_param sp = { .sched_priority = MAX_RT_PRIO / 2 }; | 
|---|
| 845 | WARN_ON_ONCE(sched_setscheduler_nocheck(p, SCHED_FIFO, &sp) != 0); | 
|---|
| 846 | } | 
|---|
| 847 | EXPORT_SYMBOL_GPL(sched_set_fifo); | 
|---|
| 848 |  | 
|---|
| 849 | /* | 
|---|
| 850 | * For when you don't much care about FIFO, but want to be above SCHED_NORMAL. | 
|---|
| 851 | */ | 
|---|
| 852 | void sched_set_fifo_low(struct task_struct *p) | 
|---|
| 853 | { | 
|---|
| 854 | struct sched_param sp = { .sched_priority = 1 }; | 
|---|
| 855 | WARN_ON_ONCE(sched_setscheduler_nocheck(p, SCHED_FIFO, &sp) != 0); | 
|---|
| 856 | } | 
|---|
| 857 | EXPORT_SYMBOL_GPL(sched_set_fifo_low); | 
|---|
| 858 |  | 
|---|
| 859 | void sched_set_normal(struct task_struct *p, int nice) | 
|---|
| 860 | { | 
|---|
| 861 | struct sched_attr attr = { | 
|---|
| 862 | .sched_policy = SCHED_NORMAL, | 
|---|
| 863 | .sched_nice = nice, | 
|---|
| 864 | }; | 
|---|
| 865 | WARN_ON_ONCE(sched_setattr_nocheck(p, &attr) != 0); | 
|---|
| 866 | } | 
|---|
| 867 | EXPORT_SYMBOL_GPL(sched_set_normal); | 
|---|
| 868 |  | 
|---|
| 869 | static int | 
|---|
| 870 | do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | 
|---|
| 871 | { | 
|---|
| 872 | struct sched_param lparam; | 
|---|
| 873 |  | 
|---|
| 874 | if (unlikely(!param || pid < 0)) | 
|---|
| 875 | return -EINVAL; | 
|---|
| 876 | if (copy_from_user(to: &lparam, from: param, n: sizeof(struct sched_param))) | 
|---|
| 877 | return -EFAULT; | 
|---|
| 878 |  | 
|---|
| 879 | CLASS(find_get_task, p)(pid); | 
|---|
| 880 | if (!p) | 
|---|
| 881 | return -ESRCH; | 
|---|
| 882 |  | 
|---|
| 883 | return sched_setscheduler(p, policy, param: &lparam); | 
|---|
| 884 | } | 
|---|
| 885 |  | 
|---|
| 886 | /* | 
|---|
| 887 | * Mimics kernel/events/core.c perf_copy_attr(). | 
|---|
| 888 | */ | 
|---|
| 889 | static int sched_copy_attr(struct sched_attr __user *uattr, struct sched_attr *attr) | 
|---|
| 890 | { | 
|---|
| 891 | u32 size; | 
|---|
| 892 | int ret; | 
|---|
| 893 |  | 
|---|
| 894 | /* Zero the full structure, so that a short copy will be nice: */ | 
|---|
| 895 | memset(s: attr, c: 0, n: sizeof(*attr)); | 
|---|
| 896 |  | 
|---|
| 897 | ret = get_user(size, &uattr->size); | 
|---|
| 898 | if (ret) | 
|---|
| 899 | return ret; | 
|---|
| 900 |  | 
|---|
| 901 | /* ABI compatibility quirk: */ | 
|---|
| 902 | if (!size) | 
|---|
| 903 | size = SCHED_ATTR_SIZE_VER0; | 
|---|
| 904 | if (size < SCHED_ATTR_SIZE_VER0 || size > PAGE_SIZE) | 
|---|
| 905 | goto err_size; | 
|---|
| 906 |  | 
|---|
| 907 | ret = copy_struct_from_user(dst: attr, ksize: sizeof(*attr), src: uattr, usize: size); | 
|---|
| 908 | if (ret) { | 
|---|
| 909 | if (ret == -E2BIG) | 
|---|
| 910 | goto err_size; | 
|---|
| 911 | return ret; | 
|---|
| 912 | } | 
|---|
| 913 |  | 
|---|
| 914 | if ((attr->sched_flags & SCHED_FLAG_UTIL_CLAMP) && | 
|---|
| 915 | size < SCHED_ATTR_SIZE_VER1) | 
|---|
| 916 | return -EINVAL; | 
|---|
| 917 |  | 
|---|
| 918 | /* | 
|---|
| 919 | * XXX: Do we want to be lenient like existing syscalls; or do we want | 
|---|
| 920 | * to be strict and return an error on out-of-bounds values? | 
|---|
| 921 | */ | 
|---|
| 922 | attr->sched_nice = clamp(attr->sched_nice, MIN_NICE, MAX_NICE); | 
|---|
| 923 |  | 
|---|
| 924 | return 0; | 
|---|
| 925 |  | 
|---|
| 926 | err_size: | 
|---|
| 927 | put_user(sizeof(*attr), &uattr->size); | 
|---|
| 928 | return -E2BIG; | 
|---|
| 929 | } | 
|---|
| 930 |  | 
|---|
| 931 | static void get_params(struct task_struct *p, struct sched_attr *attr) | 
|---|
| 932 | { | 
|---|
| 933 | if (task_has_dl_policy(p)) { | 
|---|
| 934 | __getparam_dl(p, attr); | 
|---|
| 935 | } else if (task_has_rt_policy(p)) { | 
|---|
| 936 | attr->sched_priority = p->rt_priority; | 
|---|
| 937 | } else { | 
|---|
| 938 | attr->sched_nice = task_nice(p); | 
|---|
| 939 | attr->sched_runtime = p->se.slice; | 
|---|
| 940 | } | 
|---|
| 941 | } | 
|---|
| 942 |  | 
|---|
| 943 | /** | 
|---|
| 944 | * sys_sched_setscheduler - set/change the scheduler policy and RT priority | 
|---|
| 945 | * @pid: the pid in question. | 
|---|
| 946 | * @policy: new policy. | 
|---|
| 947 | * @param: structure containing the new RT priority. | 
|---|
| 948 | * | 
|---|
| 949 | * Return: 0 on success. An error code otherwise. | 
|---|
| 950 | */ | 
|---|
| 951 | SYSCALL_DEFINE3(sched_setscheduler, pid_t, pid, int, policy, struct sched_param __user *, param) | 
|---|
| 952 | { | 
|---|
| 953 | if (policy < 0) | 
|---|
| 954 | return -EINVAL; | 
|---|
| 955 |  | 
|---|
| 956 | return do_sched_setscheduler(pid, policy, param); | 
|---|
| 957 | } | 
|---|
| 958 |  | 
|---|
| 959 | /** | 
|---|
| 960 | * sys_sched_setparam - set/change the RT priority of a thread | 
|---|
| 961 | * @pid: the pid in question. | 
|---|
| 962 | * @param: structure containing the new RT priority. | 
|---|
| 963 | * | 
|---|
| 964 | * Return: 0 on success. An error code otherwise. | 
|---|
| 965 | */ | 
|---|
| 966 | SYSCALL_DEFINE2(sched_setparam, pid_t, pid, struct sched_param __user *, param) | 
|---|
| 967 | { | 
|---|
| 968 | return do_sched_setscheduler(pid, SETPARAM_POLICY, param); | 
|---|
| 969 | } | 
|---|
| 970 |  | 
|---|
| 971 | /** | 
|---|
| 972 | * sys_sched_setattr - same as above, but with extended sched_attr | 
|---|
| 973 | * @pid: the pid in question. | 
|---|
| 974 | * @uattr: structure containing the extended parameters. | 
|---|
| 975 | * @flags: for future extension. | 
|---|
| 976 | */ | 
|---|
| 977 | SYSCALL_DEFINE3(sched_setattr, pid_t, pid, struct sched_attr __user *, uattr, | 
|---|
| 978 | unsigned int, flags) | 
|---|
| 979 | { | 
|---|
| 980 | struct sched_attr attr; | 
|---|
| 981 | int retval; | 
|---|
| 982 |  | 
|---|
| 983 | if (unlikely(!uattr || pid < 0 || flags)) | 
|---|
| 984 | return -EINVAL; | 
|---|
| 985 |  | 
|---|
| 986 | retval = sched_copy_attr(uattr, attr: &attr); | 
|---|
| 987 | if (retval) | 
|---|
| 988 | return retval; | 
|---|
| 989 |  | 
|---|
| 990 | if ((int)attr.sched_policy < 0) | 
|---|
| 991 | return -EINVAL; | 
|---|
| 992 | if (attr.sched_flags & SCHED_FLAG_KEEP_POLICY) | 
|---|
| 993 | attr.sched_policy = SETPARAM_POLICY; | 
|---|
| 994 |  | 
|---|
| 995 | CLASS(find_get_task, p)(pid); | 
|---|
| 996 | if (!p) | 
|---|
| 997 | return -ESRCH; | 
|---|
| 998 |  | 
|---|
| 999 | if (attr.sched_flags & SCHED_FLAG_KEEP_PARAMS) | 
|---|
| 1000 | get_params(p, attr: &attr); | 
|---|
| 1001 |  | 
|---|
| 1002 | return sched_setattr(p, attr: &attr); | 
|---|
| 1003 | } | 
|---|
| 1004 |  | 
|---|
| 1005 | /** | 
|---|
| 1006 | * sys_sched_getscheduler - get the policy (scheduling class) of a thread | 
|---|
| 1007 | * @pid: the pid in question. | 
|---|
| 1008 | * | 
|---|
| 1009 | * Return: On success, the policy of the thread. Otherwise, a negative error | 
|---|
| 1010 | * code. | 
|---|
| 1011 | */ | 
|---|
| 1012 | SYSCALL_DEFINE1(sched_getscheduler, pid_t, pid) | 
|---|
| 1013 | { | 
|---|
| 1014 | struct task_struct *p; | 
|---|
| 1015 | int retval; | 
|---|
| 1016 |  | 
|---|
| 1017 | if (pid < 0) | 
|---|
| 1018 | return -EINVAL; | 
|---|
| 1019 |  | 
|---|
| 1020 | guard(rcu)(); | 
|---|
| 1021 | p = find_process_by_pid(pid); | 
|---|
| 1022 | if (!p) | 
|---|
| 1023 | return -ESRCH; | 
|---|
| 1024 |  | 
|---|
| 1025 | retval = security_task_getscheduler(p); | 
|---|
| 1026 | if (!retval) { | 
|---|
| 1027 | retval = p->policy; | 
|---|
| 1028 | if (p->sched_reset_on_fork) | 
|---|
| 1029 | retval |= SCHED_RESET_ON_FORK; | 
|---|
| 1030 | } | 
|---|
| 1031 | return retval; | 
|---|
| 1032 | } | 
|---|
| 1033 |  | 
|---|
| 1034 | /** | 
|---|
| 1035 | * sys_sched_getparam - get the RT priority of a thread | 
|---|
| 1036 | * @pid: the pid in question. | 
|---|
| 1037 | * @param: structure containing the RT priority. | 
|---|
| 1038 | * | 
|---|
| 1039 | * Return: On success, 0 and the RT priority is in @param. Otherwise, an error | 
|---|
| 1040 | * code. | 
|---|
| 1041 | */ | 
|---|
| 1042 | SYSCALL_DEFINE2(sched_getparam, pid_t, pid, struct sched_param __user *, param) | 
|---|
| 1043 | { | 
|---|
| 1044 | struct sched_param lp = { .sched_priority = 0 }; | 
|---|
| 1045 | struct task_struct *p; | 
|---|
| 1046 | int retval; | 
|---|
| 1047 |  | 
|---|
| 1048 | if (unlikely(!param || pid < 0)) | 
|---|
| 1049 | return -EINVAL; | 
|---|
| 1050 |  | 
|---|
| 1051 | scoped_guard (rcu) { | 
|---|
| 1052 | p = find_process_by_pid(pid); | 
|---|
| 1053 | if (!p) | 
|---|
| 1054 | return -ESRCH; | 
|---|
| 1055 |  | 
|---|
| 1056 | retval = security_task_getscheduler(p); | 
|---|
| 1057 | if (retval) | 
|---|
| 1058 | return retval; | 
|---|
| 1059 |  | 
|---|
| 1060 | if (task_has_rt_policy(p)) | 
|---|
| 1061 | lp.sched_priority = p->rt_priority; | 
|---|
| 1062 | } | 
|---|
| 1063 |  | 
|---|
| 1064 | /* | 
|---|
| 1065 | * This one might sleep, we cannot do it with a spinlock held ... | 
|---|
| 1066 | */ | 
|---|
| 1067 | return copy_to_user(to: param, from: &lp, n: sizeof(*param)) ? -EFAULT : 0; | 
|---|
| 1068 | } | 
|---|
| 1069 |  | 
|---|
| 1070 | /** | 
|---|
| 1071 | * sys_sched_getattr - similar to sched_getparam, but with sched_attr | 
|---|
| 1072 | * @pid: the pid in question. | 
|---|
| 1073 | * @uattr: structure containing the extended parameters. | 
|---|
| 1074 | * @usize: sizeof(attr) for fwd/bwd comp. | 
|---|
| 1075 | * @flags: for future extension. | 
|---|
| 1076 | */ | 
|---|
| 1077 | SYSCALL_DEFINE4(sched_getattr, pid_t, pid, struct sched_attr __user *, uattr, | 
|---|
| 1078 | unsigned int, usize, unsigned int, flags) | 
|---|
| 1079 | { | 
|---|
| 1080 | struct sched_attr kattr = { }; | 
|---|
| 1081 | struct task_struct *p; | 
|---|
| 1082 | int retval; | 
|---|
| 1083 |  | 
|---|
| 1084 | if (unlikely(!uattr || pid < 0 || usize > PAGE_SIZE || | 
|---|
| 1085 | usize < SCHED_ATTR_SIZE_VER0 || flags)) | 
|---|
| 1086 | return -EINVAL; | 
|---|
| 1087 |  | 
|---|
| 1088 | scoped_guard (rcu) { | 
|---|
| 1089 | p = find_process_by_pid(pid); | 
|---|
| 1090 | if (!p) | 
|---|
| 1091 | return -ESRCH; | 
|---|
| 1092 |  | 
|---|
| 1093 | retval = security_task_getscheduler(p); | 
|---|
| 1094 | if (retval) | 
|---|
| 1095 | return retval; | 
|---|
| 1096 |  | 
|---|
| 1097 | kattr.sched_policy = p->policy; | 
|---|
| 1098 | if (p->sched_reset_on_fork) | 
|---|
| 1099 | kattr.sched_flags |= SCHED_FLAG_RESET_ON_FORK; | 
|---|
| 1100 | get_params(p, attr: &kattr); | 
|---|
| 1101 | kattr.sched_flags &= SCHED_FLAG_ALL; | 
|---|
| 1102 |  | 
|---|
| 1103 | #ifdef CONFIG_UCLAMP_TASK | 
|---|
| 1104 | /* | 
|---|
| 1105 | * This could race with another potential updater, but this is fine | 
|---|
| 1106 | * because it'll correctly read the old or the new value. We don't need | 
|---|
| 1107 | * to guarantee who wins the race as long as it doesn't return garbage. | 
|---|
| 1108 | */ | 
|---|
| 1109 | kattr.sched_util_min = p->uclamp_req[UCLAMP_MIN].value; | 
|---|
| 1110 | kattr.sched_util_max = p->uclamp_req[UCLAMP_MAX].value; | 
|---|
| 1111 | #endif | 
|---|
| 1112 | } | 
|---|
| 1113 |  | 
|---|
| 1114 | kattr.size = min(usize, sizeof(kattr)); | 
|---|
| 1115 | return copy_struct_to_user(dst: uattr, usize, src: &kattr, ksize: sizeof(kattr), NULL); | 
|---|
| 1116 | } | 
|---|
| 1117 |  | 
|---|
| 1118 | int dl_task_check_affinity(struct task_struct *p, const struct cpumask *mask) | 
|---|
| 1119 | { | 
|---|
| 1120 | /* | 
|---|
| 1121 | * If the task isn't a deadline task or admission control is | 
|---|
| 1122 | * disabled then we don't care about affinity changes. | 
|---|
| 1123 | */ | 
|---|
| 1124 | if (!task_has_dl_policy(p) || !dl_bandwidth_enabled()) | 
|---|
| 1125 | return 0; | 
|---|
| 1126 |  | 
|---|
| 1127 | /* | 
|---|
| 1128 | * The special/sugov task isn't part of regular bandwidth/admission | 
|---|
| 1129 | * control so let userspace change affinities. | 
|---|
| 1130 | */ | 
|---|
| 1131 | if (dl_entity_is_special(dl_se: &p->dl)) | 
|---|
| 1132 | return 0; | 
|---|
| 1133 |  | 
|---|
| 1134 | /* | 
|---|
| 1135 | * Since bandwidth control happens on root_domain basis, | 
|---|
| 1136 | * if admission test is enabled, we only admit -deadline | 
|---|
| 1137 | * tasks allowed to run on all the CPUs in the task's | 
|---|
| 1138 | * root_domain. | 
|---|
| 1139 | */ | 
|---|
| 1140 | guard(rcu)(); | 
|---|
| 1141 | if (!cpumask_subset(task_rq(p)->rd->span, src2p: mask)) | 
|---|
| 1142 | return -EBUSY; | 
|---|
| 1143 |  | 
|---|
| 1144 | return 0; | 
|---|
| 1145 | } | 
|---|
| 1146 |  | 
|---|
| 1147 | int __sched_setaffinity(struct task_struct *p, struct affinity_context *ctx) | 
|---|
| 1148 | { | 
|---|
| 1149 | int retval; | 
|---|
| 1150 | cpumask_var_t cpus_allowed, new_mask; | 
|---|
| 1151 |  | 
|---|
| 1152 | if (!alloc_cpumask_var(mask: &cpus_allowed, GFP_KERNEL)) | 
|---|
| 1153 | return -ENOMEM; | 
|---|
| 1154 |  | 
|---|
| 1155 | if (!alloc_cpumask_var(mask: &new_mask, GFP_KERNEL)) { | 
|---|
| 1156 | retval = -ENOMEM; | 
|---|
| 1157 | goto out_free_cpus_allowed; | 
|---|
| 1158 | } | 
|---|
| 1159 |  | 
|---|
| 1160 | cpuset_cpus_allowed(p, mask: cpus_allowed); | 
|---|
| 1161 | cpumask_and(dstp: new_mask, src1p: ctx->new_mask, src2p: cpus_allowed); | 
|---|
| 1162 |  | 
|---|
| 1163 | ctx->new_mask = new_mask; | 
|---|
| 1164 | ctx->flags |= SCA_CHECK; | 
|---|
| 1165 |  | 
|---|
| 1166 | retval = dl_task_check_affinity(p, mask: new_mask); | 
|---|
| 1167 | if (retval) | 
|---|
| 1168 | goto out_free_new_mask; | 
|---|
| 1169 |  | 
|---|
| 1170 | retval = __set_cpus_allowed_ptr(p, ctx); | 
|---|
| 1171 | if (retval) | 
|---|
| 1172 | goto out_free_new_mask; | 
|---|
| 1173 |  | 
|---|
| 1174 | cpuset_cpus_allowed(p, mask: cpus_allowed); | 
|---|
| 1175 | if (!cpumask_subset(src1p: new_mask, src2p: cpus_allowed)) { | 
|---|
| 1176 | /* | 
|---|
| 1177 | * We must have raced with a concurrent cpuset update. | 
|---|
| 1178 | * Just reset the cpumask to the cpuset's cpus_allowed. | 
|---|
| 1179 | */ | 
|---|
| 1180 | cpumask_copy(dstp: new_mask, srcp: cpus_allowed); | 
|---|
| 1181 |  | 
|---|
| 1182 | /* | 
|---|
| 1183 | * If SCA_USER is set, a 2nd call to __set_cpus_allowed_ptr() | 
|---|
| 1184 | * will restore the previous user_cpus_ptr value. | 
|---|
| 1185 | * | 
|---|
| 1186 | * In the unlikely event a previous user_cpus_ptr exists, | 
|---|
| 1187 | * we need to further restrict the mask to what is allowed | 
|---|
| 1188 | * by that old user_cpus_ptr. | 
|---|
| 1189 | */ | 
|---|
| 1190 | if (unlikely((ctx->flags & SCA_USER) && ctx->user_mask)) { | 
|---|
| 1191 | bool empty = !cpumask_and(dstp: new_mask, src1p: new_mask, | 
|---|
| 1192 | src2p: ctx->user_mask); | 
|---|
| 1193 |  | 
|---|
| 1194 | if (empty) | 
|---|
| 1195 | cpumask_copy(dstp: new_mask, srcp: cpus_allowed); | 
|---|
| 1196 | } | 
|---|
| 1197 | __set_cpus_allowed_ptr(p, ctx); | 
|---|
| 1198 | retval = -EINVAL; | 
|---|
| 1199 | } | 
|---|
| 1200 |  | 
|---|
| 1201 | out_free_new_mask: | 
|---|
| 1202 | free_cpumask_var(mask: new_mask); | 
|---|
| 1203 | out_free_cpus_allowed: | 
|---|
| 1204 | free_cpumask_var(mask: cpus_allowed); | 
|---|
| 1205 | return retval; | 
|---|
| 1206 | } | 
|---|
| 1207 |  | 
|---|
| 1208 | long sched_setaffinity(pid_t pid, const struct cpumask *in_mask) | 
|---|
| 1209 | { | 
|---|
| 1210 | struct affinity_context ac; | 
|---|
| 1211 | struct cpumask *user_mask; | 
|---|
| 1212 | int retval; | 
|---|
| 1213 |  | 
|---|
| 1214 | CLASS(find_get_task, p)(pid); | 
|---|
| 1215 | if (!p) | 
|---|
| 1216 | return -ESRCH; | 
|---|
| 1217 |  | 
|---|
| 1218 | if (p->flags & PF_NO_SETAFFINITY) | 
|---|
| 1219 | return -EINVAL; | 
|---|
| 1220 |  | 
|---|
| 1221 | if (!check_same_owner(p)) { | 
|---|
| 1222 | guard(rcu)(); | 
|---|
| 1223 | if (!ns_capable(__task_cred(p)->user_ns, CAP_SYS_NICE)) | 
|---|
| 1224 | return -EPERM; | 
|---|
| 1225 | } | 
|---|
| 1226 |  | 
|---|
| 1227 | retval = security_task_setscheduler(p); | 
|---|
| 1228 | if (retval) | 
|---|
| 1229 | return retval; | 
|---|
| 1230 |  | 
|---|
| 1231 | /* | 
|---|
| 1232 | * With non-SMP configs, user_cpus_ptr/user_mask isn't used and | 
|---|
| 1233 | * alloc_user_cpus_ptr() returns NULL. | 
|---|
| 1234 | */ | 
|---|
| 1235 | user_mask = alloc_user_cpus_ptr(NUMA_NO_NODE); | 
|---|
| 1236 | if (user_mask) { | 
|---|
| 1237 | cpumask_copy(dstp: user_mask, srcp: in_mask); | 
|---|
| 1238 | } else { | 
|---|
| 1239 | return -ENOMEM; | 
|---|
| 1240 | } | 
|---|
| 1241 |  | 
|---|
| 1242 | ac = (struct affinity_context){ | 
|---|
| 1243 | .new_mask  = in_mask, | 
|---|
| 1244 | .user_mask = user_mask, | 
|---|
| 1245 | .flags     = SCA_USER, | 
|---|
| 1246 | }; | 
|---|
| 1247 |  | 
|---|
| 1248 | retval = __sched_setaffinity(p, ctx: &ac); | 
|---|
| 1249 | kfree(objp: ac.user_mask); | 
|---|
| 1250 |  | 
|---|
| 1251 | return retval; | 
|---|
| 1252 | } | 
|---|
| 1253 |  | 
|---|
| 1254 | static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len, | 
|---|
| 1255 | struct cpumask *new_mask) | 
|---|
| 1256 | { | 
|---|
| 1257 | if (len < cpumask_size()) | 
|---|
| 1258 | cpumask_clear(dstp: new_mask); | 
|---|
| 1259 | else if (len > cpumask_size()) | 
|---|
| 1260 | len = cpumask_size(); | 
|---|
| 1261 |  | 
|---|
| 1262 | return copy_from_user(to: new_mask, from: user_mask_ptr, n: len) ? -EFAULT : 0; | 
|---|
| 1263 | } | 
|---|
| 1264 |  | 
|---|
| 1265 | /** | 
|---|
| 1266 | * sys_sched_setaffinity - set the CPU affinity of a process | 
|---|
| 1267 | * @pid: pid of the process | 
|---|
| 1268 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | 
|---|
| 1269 | * @user_mask_ptr: user-space pointer to the new CPU mask | 
|---|
| 1270 | * | 
|---|
| 1271 | * Return: 0 on success. An error code otherwise. | 
|---|
| 1272 | */ | 
|---|
| 1273 | SYSCALL_DEFINE3(sched_setaffinity, pid_t, pid, unsigned int, len, | 
|---|
| 1274 | unsigned long __user *, user_mask_ptr) | 
|---|
| 1275 | { | 
|---|
| 1276 | cpumask_var_t new_mask; | 
|---|
| 1277 | int retval; | 
|---|
| 1278 |  | 
|---|
| 1279 | if (!alloc_cpumask_var(mask: &new_mask, GFP_KERNEL)) | 
|---|
| 1280 | return -ENOMEM; | 
|---|
| 1281 |  | 
|---|
| 1282 | retval = get_user_cpu_mask(user_mask_ptr, len, new_mask); | 
|---|
| 1283 | if (retval == 0) | 
|---|
| 1284 | retval = sched_setaffinity(pid, in_mask: new_mask); | 
|---|
| 1285 | free_cpumask_var(mask: new_mask); | 
|---|
| 1286 | return retval; | 
|---|
| 1287 | } | 
|---|
| 1288 |  | 
|---|
| 1289 | long sched_getaffinity(pid_t pid, struct cpumask *mask) | 
|---|
| 1290 | { | 
|---|
| 1291 | struct task_struct *p; | 
|---|
| 1292 | int retval; | 
|---|
| 1293 |  | 
|---|
| 1294 | guard(rcu)(); | 
|---|
| 1295 | p = find_process_by_pid(pid); | 
|---|
| 1296 | if (!p) | 
|---|
| 1297 | return -ESRCH; | 
|---|
| 1298 |  | 
|---|
| 1299 | retval = security_task_getscheduler(p); | 
|---|
| 1300 | if (retval) | 
|---|
| 1301 | return retval; | 
|---|
| 1302 |  | 
|---|
| 1303 | guard(raw_spinlock_irqsave)(l: &p->pi_lock); | 
|---|
| 1304 | cpumask_and(dstp: mask, src1p: &p->cpus_mask, cpu_active_mask); | 
|---|
| 1305 |  | 
|---|
| 1306 | return 0; | 
|---|
| 1307 | } | 
|---|
| 1308 |  | 
|---|
| 1309 | /** | 
|---|
| 1310 | * sys_sched_getaffinity - get the CPU affinity of a process | 
|---|
| 1311 | * @pid: pid of the process | 
|---|
| 1312 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | 
|---|
| 1313 | * @user_mask_ptr: user-space pointer to hold the current CPU mask | 
|---|
| 1314 | * | 
|---|
| 1315 | * Return: size of CPU mask copied to user_mask_ptr on success. An | 
|---|
| 1316 | * error code otherwise. | 
|---|
| 1317 | */ | 
|---|
| 1318 | SYSCALL_DEFINE3(sched_getaffinity, pid_t, pid, unsigned int, len, | 
|---|
| 1319 | unsigned long __user *, user_mask_ptr) | 
|---|
| 1320 | { | 
|---|
| 1321 | int ret; | 
|---|
| 1322 | cpumask_var_t mask; | 
|---|
| 1323 |  | 
|---|
| 1324 | if ((len * BITS_PER_BYTE) < nr_cpu_ids) | 
|---|
| 1325 | return -EINVAL; | 
|---|
| 1326 | if (len & (sizeof(unsigned long)-1)) | 
|---|
| 1327 | return -EINVAL; | 
|---|
| 1328 |  | 
|---|
| 1329 | if (!zalloc_cpumask_var(mask: &mask, GFP_KERNEL)) | 
|---|
| 1330 | return -ENOMEM; | 
|---|
| 1331 |  | 
|---|
| 1332 | ret = sched_getaffinity(pid, mask); | 
|---|
| 1333 | if (ret == 0) { | 
|---|
| 1334 | unsigned int retlen = min(len, cpumask_size()); | 
|---|
| 1335 |  | 
|---|
| 1336 | if (copy_to_user(to: user_mask_ptr, cpumask_bits(mask), n: retlen)) | 
|---|
| 1337 | ret = -EFAULT; | 
|---|
| 1338 | else | 
|---|
| 1339 | ret = retlen; | 
|---|
| 1340 | } | 
|---|
| 1341 | free_cpumask_var(mask); | 
|---|
| 1342 |  | 
|---|
| 1343 | return ret; | 
|---|
| 1344 | } | 
|---|
| 1345 |  | 
|---|
| 1346 | static void do_sched_yield(void) | 
|---|
| 1347 | { | 
|---|
| 1348 | struct rq_flags rf; | 
|---|
| 1349 | struct rq *rq; | 
|---|
| 1350 |  | 
|---|
| 1351 | rq = this_rq_lock_irq(rf: &rf); | 
|---|
| 1352 |  | 
|---|
| 1353 | schedstat_inc(rq->yld_count); | 
|---|
| 1354 | current->sched_class->yield_task(rq); | 
|---|
| 1355 |  | 
|---|
| 1356 | preempt_disable(); | 
|---|
| 1357 | rq_unlock_irq(rq, rf: &rf); | 
|---|
| 1358 | sched_preempt_enable_no_resched(); | 
|---|
| 1359 |  | 
|---|
| 1360 | schedule(); | 
|---|
| 1361 | } | 
|---|
| 1362 |  | 
|---|
| 1363 | /** | 
|---|
| 1364 | * sys_sched_yield - yield the current processor to other threads. | 
|---|
| 1365 | * | 
|---|
| 1366 | * This function yields the current CPU to other tasks. If there are no | 
|---|
| 1367 | * other threads running on this CPU then this function will return. | 
|---|
| 1368 | * | 
|---|
| 1369 | * Return: 0. | 
|---|
| 1370 | */ | 
|---|
| 1371 | SYSCALL_DEFINE0(sched_yield) | 
|---|
| 1372 | { | 
|---|
| 1373 | do_sched_yield(); | 
|---|
| 1374 | return 0; | 
|---|
| 1375 | } | 
|---|
| 1376 |  | 
|---|
| 1377 | /** | 
|---|
| 1378 | * yield - yield the current processor to other threads. | 
|---|
| 1379 | * | 
|---|
| 1380 | * Do not ever use this function, there's a 99% chance you're doing it wrong. | 
|---|
| 1381 | * | 
|---|
| 1382 | * The scheduler is at all times free to pick the calling task as the most | 
|---|
| 1383 | * eligible task to run, if removing the yield() call from your code breaks | 
|---|
| 1384 | * it, it's already broken. | 
|---|
| 1385 | * | 
|---|
| 1386 | * Typical broken usage is: | 
|---|
| 1387 | * | 
|---|
| 1388 | * while (!event) | 
|---|
| 1389 | *	yield(); | 
|---|
| 1390 | * | 
|---|
| 1391 | * where one assumes that yield() will let 'the other' process run that will | 
|---|
| 1392 | * make event true. If the current task is a SCHED_FIFO task that will never | 
|---|
| 1393 | * happen. Never use yield() as a progress guarantee!! | 
|---|
| 1394 | * | 
|---|
| 1395 | * If you want to use yield() to wait for something, use wait_event(). | 
|---|
| 1396 | * If you want to use yield() to be 'nice' for others, use cond_resched(). | 
|---|
| 1397 | * If you still want to use yield(), do not! | 
|---|
| 1398 | */ | 
|---|
| 1399 | void __sched yield(void) | 
|---|
| 1400 | { | 
|---|
| 1401 | set_current_state(TASK_RUNNING); | 
|---|
| 1402 | do_sched_yield(); | 
|---|
| 1403 | } | 
|---|
| 1404 | EXPORT_SYMBOL(yield); | 
|---|
| 1405 |  | 
|---|
| 1406 | /** | 
|---|
| 1407 | * yield_to - yield the current processor to another thread in | 
|---|
| 1408 | * your thread group, or accelerate that thread toward the | 
|---|
| 1409 | * processor it's on. | 
|---|
| 1410 | * @p: target task | 
|---|
| 1411 | * @preempt: whether task preemption is allowed or not | 
|---|
| 1412 | * | 
|---|
| 1413 | * It's the caller's job to ensure that the target task struct | 
|---|
| 1414 | * can't go away on us before we can do any checks. | 
|---|
| 1415 | * | 
|---|
| 1416 | * Return: | 
|---|
| 1417 | *	true (>0) if we indeed boosted the target task. | 
|---|
| 1418 | *	false (0) if we failed to boost the target. | 
|---|
| 1419 | *	-ESRCH if there's no task to yield to. | 
|---|
| 1420 | */ | 
|---|
| 1421 | int __sched yield_to(struct task_struct *p, bool preempt) | 
|---|
| 1422 | { | 
|---|
| 1423 | struct task_struct *curr = current; | 
|---|
| 1424 | struct rq *rq, *p_rq; | 
|---|
| 1425 | int yielded = 0; | 
|---|
| 1426 |  | 
|---|
| 1427 | scoped_guard (raw_spinlock_irqsave, &p->pi_lock) { | 
|---|
| 1428 | rq = this_rq(); | 
|---|
| 1429 |  | 
|---|
| 1430 | again: | 
|---|
| 1431 | p_rq = task_rq(p); | 
|---|
| 1432 | /* | 
|---|
| 1433 | * If we're the only runnable task on the rq and target rq also | 
|---|
| 1434 | * has only one task, there's absolutely no point in yielding. | 
|---|
| 1435 | */ | 
|---|
| 1436 | if (rq->nr_running == 1 && p_rq->nr_running == 1) | 
|---|
| 1437 | return -ESRCH; | 
|---|
| 1438 |  | 
|---|
| 1439 | guard(double_rq_lock)(lock: rq, lock2: p_rq); | 
|---|
| 1440 | if (task_rq(p) != p_rq) | 
|---|
| 1441 | goto again; | 
|---|
| 1442 |  | 
|---|
| 1443 | if (!curr->sched_class->yield_to_task) | 
|---|
| 1444 | return 0; | 
|---|
| 1445 |  | 
|---|
| 1446 | if (curr->sched_class != p->sched_class) | 
|---|
| 1447 | return 0; | 
|---|
| 1448 |  | 
|---|
| 1449 | if (task_on_cpu(rq: p_rq, p) || !task_is_running(p)) | 
|---|
| 1450 | return 0; | 
|---|
| 1451 |  | 
|---|
| 1452 | yielded = curr->sched_class->yield_to_task(rq, p); | 
|---|
| 1453 | if (yielded) { | 
|---|
| 1454 | schedstat_inc(rq->yld_count); | 
|---|
| 1455 | /* | 
|---|
| 1456 | * Make p's CPU reschedule; pick_next_entity | 
|---|
| 1457 | * takes care of fairness. | 
|---|
| 1458 | */ | 
|---|
| 1459 | if (preempt && rq != p_rq) | 
|---|
| 1460 | resched_curr(rq: p_rq); | 
|---|
| 1461 | } | 
|---|
| 1462 | } | 
|---|
| 1463 |  | 
|---|
| 1464 | if (yielded) | 
|---|
| 1465 | schedule(); | 
|---|
| 1466 |  | 
|---|
| 1467 | return yielded; | 
|---|
| 1468 | } | 
|---|
| 1469 | EXPORT_SYMBOL_GPL(yield_to); | 
|---|
| 1470 |  | 
|---|
| 1471 | /** | 
|---|
| 1472 | * sys_sched_get_priority_max - return maximum RT priority. | 
|---|
| 1473 | * @policy: scheduling class. | 
|---|
| 1474 | * | 
|---|
| 1475 | * Return: On success, this syscall returns the maximum | 
|---|
| 1476 | * rt_priority that can be used by a given scheduling class. | 
|---|
| 1477 | * On failure, a negative error code is returned. | 
|---|
| 1478 | */ | 
|---|
| 1479 | SYSCALL_DEFINE1(sched_get_priority_max, int, policy) | 
|---|
| 1480 | { | 
|---|
| 1481 | int ret = -EINVAL; | 
|---|
| 1482 |  | 
|---|
| 1483 | switch (policy) { | 
|---|
| 1484 | case SCHED_FIFO: | 
|---|
| 1485 | case SCHED_RR: | 
|---|
| 1486 | ret = MAX_RT_PRIO-1; | 
|---|
| 1487 | break; | 
|---|
| 1488 | case SCHED_DEADLINE: | 
|---|
| 1489 | case SCHED_NORMAL: | 
|---|
| 1490 | case SCHED_BATCH: | 
|---|
| 1491 | case SCHED_IDLE: | 
|---|
| 1492 | case SCHED_EXT: | 
|---|
| 1493 | ret = 0; | 
|---|
| 1494 | break; | 
|---|
| 1495 | } | 
|---|
| 1496 | return ret; | 
|---|
| 1497 | } | 
|---|
| 1498 |  | 
|---|
| 1499 | /** | 
|---|
| 1500 | * sys_sched_get_priority_min - return minimum RT priority. | 
|---|
| 1501 | * @policy: scheduling class. | 
|---|
| 1502 | * | 
|---|
| 1503 | * Return: On success, this syscall returns the minimum | 
|---|
| 1504 | * rt_priority that can be used by a given scheduling class. | 
|---|
| 1505 | * On failure, a negative error code is returned. | 
|---|
| 1506 | */ | 
|---|
| 1507 | SYSCALL_DEFINE1(sched_get_priority_min, int, policy) | 
|---|
| 1508 | { | 
|---|
| 1509 | int ret = -EINVAL; | 
|---|
| 1510 |  | 
|---|
| 1511 | switch (policy) { | 
|---|
| 1512 | case SCHED_FIFO: | 
|---|
| 1513 | case SCHED_RR: | 
|---|
| 1514 | ret = 1; | 
|---|
| 1515 | break; | 
|---|
| 1516 | case SCHED_DEADLINE: | 
|---|
| 1517 | case SCHED_NORMAL: | 
|---|
| 1518 | case SCHED_BATCH: | 
|---|
| 1519 | case SCHED_IDLE: | 
|---|
| 1520 | case SCHED_EXT: | 
|---|
| 1521 | ret = 0; | 
|---|
| 1522 | } | 
|---|
| 1523 | return ret; | 
|---|
| 1524 | } | 
|---|
| 1525 |  | 
|---|
| 1526 | static int sched_rr_get_interval(pid_t pid, struct timespec64 *t) | 
|---|
| 1527 | { | 
|---|
| 1528 | unsigned int time_slice = 0; | 
|---|
| 1529 | int retval; | 
|---|
| 1530 |  | 
|---|
| 1531 | if (pid < 0) | 
|---|
| 1532 | return -EINVAL; | 
|---|
| 1533 |  | 
|---|
| 1534 | scoped_guard (rcu) { | 
|---|
| 1535 | struct task_struct *p = find_process_by_pid(pid); | 
|---|
| 1536 | if (!p) | 
|---|
| 1537 | return -ESRCH; | 
|---|
| 1538 |  | 
|---|
| 1539 | retval = security_task_getscheduler(p); | 
|---|
| 1540 | if (retval) | 
|---|
| 1541 | return retval; | 
|---|
| 1542 |  | 
|---|
| 1543 | scoped_guard (task_rq_lock, p) { | 
|---|
| 1544 | struct rq *rq = scope.rq; | 
|---|
| 1545 | if (p->sched_class->get_rr_interval) | 
|---|
| 1546 | time_slice = p->sched_class->get_rr_interval(rq, p); | 
|---|
| 1547 | } | 
|---|
| 1548 | } | 
|---|
| 1549 |  | 
|---|
| 1550 | jiffies_to_timespec64(jiffies: time_slice, value: t); | 
|---|
| 1551 | return 0; | 
|---|
| 1552 | } | 
|---|
| 1553 |  | 
|---|
| 1554 | /** | 
|---|
| 1555 | * sys_sched_rr_get_interval - return the default time-slice of a process. | 
|---|
| 1556 | * @pid: pid of the process. | 
|---|
| 1557 | * @interval: userspace pointer to the time-slice value. | 
|---|
| 1558 | * | 
|---|
| 1559 | * this syscall writes the default time-slice value of a given process | 
|---|
| 1560 | * into the user-space timespec buffer. A value of '0' means infinity. | 
|---|
| 1561 | * | 
|---|
| 1562 | * Return: On success, 0 and the time-slice is in @interval. Otherwise, | 
|---|
| 1563 | * an error code. | 
|---|
| 1564 | */ | 
|---|
| 1565 | SYSCALL_DEFINE2(sched_rr_get_interval, pid_t, pid, | 
|---|
| 1566 | struct __kernel_timespec __user *, interval) | 
|---|
| 1567 | { | 
|---|
| 1568 | struct timespec64 t; | 
|---|
| 1569 | int retval = sched_rr_get_interval(pid, t: &t); | 
|---|
| 1570 |  | 
|---|
| 1571 | if (retval == 0) | 
|---|
| 1572 | retval = put_timespec64(ts: &t, uts: interval); | 
|---|
| 1573 |  | 
|---|
| 1574 | return retval; | 
|---|
| 1575 | } | 
|---|
| 1576 |  | 
|---|
| 1577 | #ifdef CONFIG_COMPAT_32BIT_TIME | 
|---|
| 1578 | SYSCALL_DEFINE2(sched_rr_get_interval_time32, pid_t, pid, | 
|---|
| 1579 | struct old_timespec32 __user *, interval) | 
|---|
| 1580 | { | 
|---|
| 1581 | struct timespec64 t; | 
|---|
| 1582 | int retval = sched_rr_get_interval(pid, t: &t); | 
|---|
| 1583 |  | 
|---|
| 1584 | if (retval == 0) | 
|---|
| 1585 | retval = put_old_timespec32(&t, interval); | 
|---|
| 1586 | return retval; | 
|---|
| 1587 | } | 
|---|
| 1588 | #endif | 
|---|
| 1589 |  | 
|---|