linux/kernel/kthread.c
<<
>>
Prefs
   1// SPDX-License-Identifier: GPL-2.0-only
   2/* Kernel thread helper functions.
   3 *   Copyright (C) 2004 IBM Corporation, Rusty Russell.
   4 *
   5 * Creation is done via kthreadd, so that we get a clean environment
   6 * even if we're invoked from userspace (think modprobe, hotplug cpu,
   7 * etc.).
   8 */
   9#include <uapi/linux/sched/types.h>
  10#include <linux/sched.h>
  11#include <linux/sched/task.h>
  12#include <linux/kthread.h>
  13#include <linux/completion.h>
  14#include <linux/err.h>
  15#include <linux/cgroup.h>
  16#include <linux/cpuset.h>
  17#include <linux/unistd.h>
  18#include <linux/file.h>
  19#include <linux/export.h>
  20#include <linux/mutex.h>
  21#include <linux/slab.h>
  22#include <linux/freezer.h>
  23#include <linux/ptrace.h>
  24#include <linux/uaccess.h>
  25#include <linux/numa.h>
  26#include <trace/events/sched.h>
  27
  28static DEFINE_SPINLOCK(kthread_create_lock);
  29static LIST_HEAD(kthread_create_list);
  30struct task_struct *kthreadd_task;
  31
  32struct kthread_create_info
  33{
  34        /* Information passed to kthread() from kthreadd. */
  35        int (*threadfn)(void *data);
  36        void *data;
  37        int node;
  38
  39        /* Result passed back to kthread_create() from kthreadd. */
  40        struct task_struct *result;
  41        struct completion *done;
  42
  43        struct list_head list;
  44};
  45
  46struct kthread {
  47        unsigned long flags;
  48        unsigned int cpu;
  49        void *data;
  50        struct completion parked;
  51        struct completion exited;
  52#ifdef CONFIG_BLK_CGROUP
  53        struct cgroup_subsys_state *blkcg_css;
  54#endif
  55};
  56
  57enum KTHREAD_BITS {
  58        KTHREAD_IS_PER_CPU = 0,
  59        KTHREAD_SHOULD_STOP,
  60        KTHREAD_SHOULD_PARK,
  61};
  62
  63static inline void set_kthread_struct(void *kthread)
  64{
  65        /*
  66         * We abuse ->set_child_tid to avoid the new member and because it
  67         * can't be wrongly copied by copy_process(). We also rely on fact
  68         * that the caller can't exec, so PF_KTHREAD can't be cleared.
  69         */
  70        current->set_child_tid = (__force void __user *)kthread;
  71}
  72
  73static inline struct kthread *to_kthread(struct task_struct *k)
  74{
  75        WARN_ON(!(k->flags & PF_KTHREAD));
  76        return (__force void *)k->set_child_tid;
  77}
  78
  79void free_kthread_struct(struct task_struct *k)
  80{
  81        struct kthread *kthread;
  82
  83        /*
  84         * Can be NULL if this kthread was created by kernel_thread()
  85         * or if kmalloc() in kthread() failed.
  86         */
  87        kthread = to_kthread(k);
  88#ifdef CONFIG_BLK_CGROUP
  89        WARN_ON_ONCE(kthread && kthread->blkcg_css);
  90#endif
  91        kfree(kthread);
  92}
  93
  94/**
  95 * kthread_should_stop - should this kthread return now?
  96 *
  97 * When someone calls kthread_stop() on your kthread, it will be woken
  98 * and this will return true.  You should then return, and your return
  99 * value will be passed through to kthread_stop().
 100 */
 101bool kthread_should_stop(void)
 102{
 103        return test_bit(KTHREAD_SHOULD_STOP, &to_kthread(current)->flags);
 104}
 105EXPORT_SYMBOL(kthread_should_stop);
 106
 107bool __kthread_should_park(struct task_struct *k)
 108{
 109        return test_bit(KTHREAD_SHOULD_PARK, &to_kthread(k)->flags);
 110}
 111EXPORT_SYMBOL_GPL(__kthread_should_park);
 112
 113/**
 114 * kthread_should_park - should this kthread park now?
 115 *
 116 * When someone calls kthread_park() on your kthread, it will be woken
 117 * and this will return true.  You should then do the necessary
 118 * cleanup and call kthread_parkme()
 119 *
 120 * Similar to kthread_should_stop(), but this keeps the thread alive
 121 * and in a park position. kthread_unpark() "restarts" the thread and
 122 * calls the thread function again.
 123 */
 124bool kthread_should_park(void)
 125{
 126        return __kthread_should_park(current);
 127}
 128EXPORT_SYMBOL_GPL(kthread_should_park);
 129
 130/**
 131 * kthread_freezable_should_stop - should this freezable kthread return now?
 132 * @was_frozen: optional out parameter, indicates whether %current was frozen
 133 *
 134 * kthread_should_stop() for freezable kthreads, which will enter
 135 * refrigerator if necessary.  This function is safe from kthread_stop() /
 136 * freezer deadlock and freezable kthreads should use this function instead
 137 * of calling try_to_freeze() directly.
 138 */
 139bool kthread_freezable_should_stop(bool *was_frozen)
 140{
 141        bool frozen = false;
 142
 143        might_sleep();
 144
 145        if (unlikely(freezing(current)))
 146                frozen = __refrigerator(true);
 147
 148        if (was_frozen)
 149                *was_frozen = frozen;
 150
 151        return kthread_should_stop();
 152}
 153EXPORT_SYMBOL_GPL(kthread_freezable_should_stop);
 154
 155/**
 156 * kthread_data - return data value specified on kthread creation
 157 * @task: kthread task in question
 158 *
 159 * Return the data value specified when kthread @task was created.
 160 * The caller is responsible for ensuring the validity of @task when
 161 * calling this function.
 162 */
 163void *kthread_data(struct task_struct *task)
 164{
 165        return to_kthread(task)->data;
 166}
 167
 168/**
 169 * kthread_probe_data - speculative version of kthread_data()
 170 * @task: possible kthread task in question
 171 *
 172 * @task could be a kthread task.  Return the data value specified when it
 173 * was created if accessible.  If @task isn't a kthread task or its data is
 174 * inaccessible for any reason, %NULL is returned.  This function requires
 175 * that @task itself is safe to dereference.
 176 */
 177void *kthread_probe_data(struct task_struct *task)
 178{
 179        struct kthread *kthread = to_kthread(task);
 180        void *data = NULL;
 181
 182        probe_kernel_read(&data, &kthread->data, sizeof(data));
 183        return data;
 184}
 185
 186static void __kthread_parkme(struct kthread *self)
 187{
 188        for (;;) {
 189                /*
 190                 * TASK_PARKED is a special state; we must serialize against
 191                 * possible pending wakeups to avoid store-store collisions on
 192                 * task->state.
 193                 *
 194                 * Such a collision might possibly result in the task state
 195                 * changin from TASK_PARKED and us failing the
 196                 * wait_task_inactive() in kthread_park().
 197                 */
 198                set_special_state(TASK_PARKED);
 199                if (!test_bit(KTHREAD_SHOULD_PARK, &self->flags))
 200                        break;
 201
 202                /*
 203                 * Thread is going to call schedule(), do not preempt it,
 204                 * or the caller of kthread_park() may spend more time in
 205                 * wait_task_inactive().
 206                 */
 207                preempt_disable();
 208                complete(&self->parked);
 209                schedule_preempt_disabled();
 210                preempt_enable();
 211        }
 212        __set_current_state(TASK_RUNNING);
 213}
 214
 215void kthread_parkme(void)
 216{
 217        __kthread_parkme(to_kthread(current));
 218}
 219EXPORT_SYMBOL_GPL(kthread_parkme);
 220
 221static int kthread(void *_create)
 222{
 223        /* Copy data: it's on kthread's stack */
 224        struct kthread_create_info *create = _create;
 225        int (*threadfn)(void *data) = create->threadfn;
 226        void *data = create->data;
 227        struct completion *done;
 228        struct kthread *self;
 229        int ret;
 230
 231        self = kzalloc(sizeof(*self), GFP_KERNEL);
 232        set_kthread_struct(self);
 233
 234        /* If user was SIGKILLed, I release the structure. */
 235        done = xchg(&create->done, NULL);
 236        if (!done) {
 237                kfree(create);
 238                do_exit(-EINTR);
 239        }
 240
 241        if (!self) {
 242                create->result = ERR_PTR(-ENOMEM);
 243                complete(done);
 244                do_exit(-ENOMEM);
 245        }
 246
 247        self->data = data;
 248        init_completion(&self->exited);
 249        init_completion(&self->parked);
 250        current->vfork_done = &self->exited;
 251
 252        /* OK, tell user we're spawned, wait for stop or wakeup */
 253        __set_current_state(TASK_UNINTERRUPTIBLE);
 254        create->result = current;
 255        /*
 256         * Thread is going to call schedule(), do not preempt it,
 257         * or the creator may spend more time in wait_task_inactive().
 258         */
 259        preempt_disable();
 260        complete(done);
 261        schedule_preempt_disabled();
 262        preempt_enable();
 263
 264        ret = -EINTR;
 265        if (!test_bit(KTHREAD_SHOULD_STOP, &self->flags)) {
 266                cgroup_kthread_ready();
 267                __kthread_parkme(self);
 268                ret = threadfn(data);
 269        }
 270        do_exit(ret);
 271}
 272
 273/* called from do_fork() to get node information for about to be created task */
 274int tsk_fork_get_node(struct task_struct *tsk)
 275{
 276#ifdef CONFIG_NUMA
 277        if (tsk == kthreadd_task)
 278                return tsk->pref_node_fork;
 279#endif
 280        return NUMA_NO_NODE;
 281}
 282
 283static void create_kthread(struct kthread_create_info *create)
 284{
 285        int pid;
 286
 287#ifdef CONFIG_NUMA
 288        current->pref_node_fork = create->node;
 289#endif
 290        /* We want our own signal handler (we take no signals by default). */
 291        pid = kernel_thread(kthread, create, CLONE_FS | CLONE_FILES | SIGCHLD);
 292        if (pid < 0) {
 293                /* If user was SIGKILLed, I release the structure. */
 294                struct completion *done = xchg(&create->done, NULL);
 295
 296                if (!done) {
 297                        kfree(create);
 298                        return;
 299                }
 300                create->result = ERR_PTR(pid);
 301                complete(done);
 302        }
 303}
 304
 305static __printf(4, 0)
 306struct task_struct *__kthread_create_on_node(int (*threadfn)(void *data),
 307                                                    void *data, int node,
 308                                                    const char namefmt[],
 309                                                    va_list args)
 310{
 311        DECLARE_COMPLETION_ONSTACK(done);
 312        struct task_struct *task;
 313        struct kthread_create_info *create = kmalloc(sizeof(*create),
 314                                                     GFP_KERNEL);
 315
 316        if (!create)
 317                return ERR_PTR(-ENOMEM);
 318        create->threadfn = threadfn;
 319        create->data = data;
 320        create->node = node;
 321        create->done = &done;
 322
 323        spin_lock(&kthread_create_lock);
 324        list_add_tail(&create->list, &kthread_create_list);
 325        spin_unlock(&kthread_create_lock);
 326
 327        wake_up_process(kthreadd_task);
 328        /*
 329         * Wait for completion in killable state, for I might be chosen by
 330         * the OOM killer while kthreadd is trying to allocate memory for
 331         * new kernel thread.
 332         */
 333        if (unlikely(wait_for_completion_killable(&done))) {
 334                /*
 335                 * If I was SIGKILLed before kthreadd (or new kernel thread)
 336                 * calls complete(), leave the cleanup of this structure to
 337                 * that thread.
 338                 */
 339                if (xchg(&create->done, NULL))
 340                        return ERR_PTR(-EINTR);
 341                /*
 342                 * kthreadd (or new kernel thread) will call complete()
 343                 * shortly.
 344                 */
 345                wait_for_completion(&done);
 346        }
 347        task = create->result;
 348        if (!IS_ERR(task)) {
 349                static const struct sched_param param = { .sched_priority = 0 };
 350                char name[TASK_COMM_LEN];
 351
 352                /*
 353                 * task is already visible to other tasks, so updating
 354                 * COMM must be protected.
 355                 */
 356                vsnprintf(name, sizeof(name), namefmt, args);
 357                set_task_comm(task, name);
 358                /*
 359                 * root may have changed our (kthreadd's) priority or CPU mask.
 360                 * The kernel thread should not inherit these properties.
 361                 */
 362                sched_setscheduler_nocheck(task, SCHED_NORMAL, &param);
 363                set_cpus_allowed_ptr(task, cpu_all_mask);
 364        }
 365        kfree(create);
 366        return task;
 367}
 368
 369/**
 370 * kthread_create_on_node - create a kthread.
 371 * @threadfn: the function to run until signal_pending(current).
 372 * @data: data ptr for @threadfn.
 373 * @node: task and thread structures for the thread are allocated on this node
 374 * @namefmt: printf-style name for the thread.
 375 *
 376 * Description: This helper function creates and names a kernel
 377 * thread.  The thread will be stopped: use wake_up_process() to start
 378 * it.  See also kthread_run().  The new thread has SCHED_NORMAL policy and
 379 * is affine to all CPUs.
 380 *
 381 * If thread is going to be bound on a particular cpu, give its node
 382 * in @node, to get NUMA affinity for kthread stack, or else give NUMA_NO_NODE.
 383 * When woken, the thread will run @threadfn() with @data as its
 384 * argument. @threadfn() can either call do_exit() directly if it is a
 385 * standalone thread for which no one will call kthread_stop(), or
 386 * return when 'kthread_should_stop()' is true (which means
 387 * kthread_stop() has been called).  The return value should be zero
 388 * or a negative error number; it will be passed to kthread_stop().
 389 *
 390 * Returns a task_struct or ERR_PTR(-ENOMEM) or ERR_PTR(-EINTR).
 391 */
 392struct task_struct *kthread_create_on_node(int (*threadfn)(void *data),
 393                                           void *data, int node,
 394                                           const char namefmt[],
 395                                           ...)
 396{
 397        struct task_struct *task;
 398        va_list args;
 399
 400        va_start(args, namefmt);
 401        task = __kthread_create_on_node(threadfn, data, node, namefmt, args);
 402        va_end(args);
 403
 404        return task;
 405}
 406EXPORT_SYMBOL(kthread_create_on_node);
 407
 408static void __kthread_bind_mask(struct task_struct *p, const struct cpumask *mask, long state)
 409{
 410        unsigned long flags;
 411
 412        if (!wait_task_inactive(p, state)) {
 413                WARN_ON(1);
 414                return;
 415        }
 416
 417        /* It's safe because the task is inactive. */
 418        raw_spin_lock_irqsave(&p->pi_lock, flags);
 419        do_set_cpus_allowed(p, mask);
 420        p->flags |= PF_NO_SETAFFINITY;
 421        raw_spin_unlock_irqrestore(&p->pi_lock, flags);
 422}
 423
 424static void __kthread_bind(struct task_struct *p, unsigned int cpu, long state)
 425{
 426        __kthread_bind_mask(p, cpumask_of(cpu), state);
 427}
 428
 429void kthread_bind_mask(struct task_struct *p, const struct cpumask *mask)
 430{
 431        __kthread_bind_mask(p, mask, TASK_UNINTERRUPTIBLE);
 432}
 433
 434/**
 435 * kthread_bind - bind a just-created kthread to a cpu.
 436 * @p: thread created by kthread_create().
 437 * @cpu: cpu (might not be online, must be possible) for @k to run on.
 438 *
 439 * Description: This function is equivalent to set_cpus_allowed(),
 440 * except that @cpu doesn't need to be online, and the thread must be
 441 * stopped (i.e., just returned from kthread_create()).
 442 */
 443void kthread_bind(struct task_struct *p, unsigned int cpu)
 444{
 445        __kthread_bind(p, cpu, TASK_UNINTERRUPTIBLE);
 446}
 447EXPORT_SYMBOL(kthread_bind);
 448
 449/**
 450 * kthread_create_on_cpu - Create a cpu bound kthread
 451 * @threadfn: the function to run until signal_pending(current).
 452 * @data: data ptr for @threadfn.
 453 * @cpu: The cpu on which the thread should be bound,
 454 * @namefmt: printf-style name for the thread. Format is restricted
 455 *           to "name.*%u". Code fills in cpu number.
 456 *
 457 * Description: This helper function creates and names a kernel thread
 458 * The thread will be woken and put into park mode.
 459 */
 460struct task_struct *kthread_create_on_cpu(int (*threadfn)(void *data),
 461                                          void *data, unsigned int cpu,
 462                                          const char *namefmt)
 463{
 464        struct task_struct *p;
 465
 466        p = kthread_create_on_node(threadfn, data, cpu_to_node(cpu), namefmt,
 467                                   cpu);
 468        if (IS_ERR(p))
 469                return p;
 470        kthread_bind(p, cpu);
 471        /* CPU hotplug need to bind once again when unparking the thread. */
 472        set_bit(KTHREAD_IS_PER_CPU, &to_kthread(p)->flags);
 473        to_kthread(p)->cpu = cpu;
 474        return p;
 475}
 476
 477/**
 478 * kthread_unpark - unpark a thread created by kthread_create().
 479 * @k:          thread created by kthread_create().
 480 *
 481 * Sets kthread_should_park() for @k to return false, wakes it, and
 482 * waits for it to return. If the thread is marked percpu then its
 483 * bound to the cpu again.
 484 */
 485void kthread_unpark(struct task_struct *k)
 486{
 487        struct kthread *kthread = to_kthread(k);
 488
 489        /*
 490         * Newly created kthread was parked when the CPU was offline.
 491         * The binding was lost and we need to set it again.
 492         */
 493        if (test_bit(KTHREAD_IS_PER_CPU, &kthread->flags))
 494                __kthread_bind(k, kthread->cpu, TASK_PARKED);
 495
 496        clear_bit(KTHREAD_SHOULD_PARK, &kthread->flags);
 497        /*
 498         * __kthread_parkme() will either see !SHOULD_PARK or get the wakeup.
 499         */
 500        wake_up_state(k, TASK_PARKED);
 501}
 502EXPORT_SYMBOL_GPL(kthread_unpark);
 503
 504/**
 505 * kthread_park - park a thread created by kthread_create().
 506 * @k: thread created by kthread_create().
 507 *
 508 * Sets kthread_should_park() for @k to return true, wakes it, and
 509 * waits for it to return. This can also be called after kthread_create()
 510 * instead of calling wake_up_process(): the thread will park without
 511 * calling threadfn().
 512 *
 513 * Returns 0 if the thread is parked, -ENOSYS if the thread exited.
 514 * If called by the kthread itself just the park bit is set.
 515 */
 516int kthread_park(struct task_struct *k)
 517{
 518        struct kthread *kthread = to_kthread(k);
 519
 520        if (WARN_ON(k->flags & PF_EXITING))
 521                return -ENOSYS;
 522
 523        if (WARN_ON_ONCE(test_bit(KTHREAD_SHOULD_PARK, &kthread->flags)))
 524                return -EBUSY;
 525
 526        set_bit(KTHREAD_SHOULD_PARK, &kthread->flags);
 527        if (k != current) {
 528                wake_up_process(k);
 529                /*
 530                 * Wait for __kthread_parkme() to complete(), this means we
 531                 * _will_ have TASK_PARKED and are about to call schedule().
 532                 */
 533                wait_for_completion(&kthread->parked);
 534                /*
 535                 * Now wait for that schedule() to complete and the task to
 536                 * get scheduled out.
 537                 */
 538                WARN_ON_ONCE(!wait_task_inactive(k, TASK_PARKED));
 539        }
 540
 541        return 0;
 542}
 543EXPORT_SYMBOL_GPL(kthread_park);
 544
 545/**
 546 * kthread_stop - stop a thread created by kthread_create().
 547 * @k: thread created by kthread_create().
 548 *
 549 * Sets kthread_should_stop() for @k to return true, wakes it, and
 550 * waits for it to exit. This can also be called after kthread_create()
 551 * instead of calling wake_up_process(): the thread will exit without
 552 * calling threadfn().
 553 *
 554 * If threadfn() may call do_exit() itself, the caller must ensure
 555 * task_struct can't go away.
 556 *
 557 * Returns the result of threadfn(), or %-EINTR if wake_up_process()
 558 * was never called.
 559 */
 560int kthread_stop(struct task_struct *k)
 561{
 562        struct kthread *kthread;
 563        int ret;
 564
 565        trace_sched_kthread_stop(k);
 566
 567        get_task_struct(k);
 568        kthread = to_kthread(k);
 569        set_bit(KTHREAD_SHOULD_STOP, &kthread->flags);
 570        kthread_unpark(k);
 571        wake_up_process(k);
 572        wait_for_completion(&kthread->exited);
 573        ret = k->exit_code;
 574        put_task_struct(k);
 575
 576        trace_sched_kthread_stop_ret(ret);
 577        return ret;
 578}
 579EXPORT_SYMBOL(kthread_stop);
 580
 581int kthreadd(void *unused)
 582{
 583        struct task_struct *tsk = current;
 584
 585        /* Setup a clean context for our children to inherit. */
 586        set_task_comm(tsk, "kthreadd");
 587        ignore_signals(tsk);
 588        set_cpus_allowed_ptr(tsk, cpu_all_mask);
 589        set_mems_allowed(node_states[N_MEMORY]);
 590
 591        current->flags |= PF_NOFREEZE;
 592        cgroup_init_kthreadd();
 593
 594        for (;;) {
 595                set_current_state(TASK_INTERRUPTIBLE);
 596                if (list_empty(&kthread_create_list))
 597                        schedule();
 598                __set_current_state(TASK_RUNNING);
 599
 600                spin_lock(&kthread_create_lock);
 601                while (!list_empty(&kthread_create_list)) {
 602                        struct kthread_create_info *create;
 603
 604                        create = list_entry(kthread_create_list.next,
 605                                            struct kthread_create_info, list);
 606                        list_del_init(&create->list);
 607                        spin_unlock(&kthread_create_lock);
 608
 609                        create_kthread(create);
 610
 611                        spin_lock(&kthread_create_lock);
 612                }
 613                spin_unlock(&kthread_create_lock);
 614        }
 615
 616        return 0;
 617}
 618
 619void __kthread_init_worker(struct kthread_worker *worker,
 620                                const char *name,
 621                                struct lock_class_key *key)
 622{
 623        memset(worker, 0, sizeof(struct kthread_worker));
 624        raw_spin_lock_init(&worker->lock);
 625        lockdep_set_class_and_name(&worker->lock, key, name);
 626        INIT_LIST_HEAD(&worker->work_list);
 627        INIT_LIST_HEAD(&worker->delayed_work_list);
 628}
 629EXPORT_SYMBOL_GPL(__kthread_init_worker);
 630
 631/**
 632 * kthread_worker_fn - kthread function to process kthread_worker
 633 * @worker_ptr: pointer to initialized kthread_worker
 634 *
 635 * This function implements the main cycle of kthread worker. It processes
 636 * work_list until it is stopped with kthread_stop(). It sleeps when the queue
 637 * is empty.
 638 *
 639 * The works are not allowed to keep any locks, disable preemption or interrupts
 640 * when they finish. There is defined a safe point for freezing when one work
 641 * finishes and before a new one is started.
 642 *
 643 * Also the works must not be handled by more than one worker at the same time,
 644 * see also kthread_queue_work().
 645 */
 646int kthread_worker_fn(void *worker_ptr)
 647{
 648        struct kthread_worker *worker = worker_ptr;
 649        struct kthread_work *work;
 650
 651        /*
 652         * FIXME: Update the check and remove the assignment when all kthread
 653         * worker users are created using kthread_create_worker*() functions.
 654         */
 655        WARN_ON(worker->task && worker->task != current);
 656        worker->task = current;
 657
 658        if (worker->flags & KTW_FREEZABLE)
 659                set_freezable();
 660
 661repeat:
 662        set_current_state(TASK_INTERRUPTIBLE);  /* mb paired w/ kthread_stop */
 663
 664        if (kthread_should_stop()) {
 665                __set_current_state(TASK_RUNNING);
 666                raw_spin_lock_irq(&worker->lock);
 667                worker->task = NULL;
 668                raw_spin_unlock_irq(&worker->lock);
 669                return 0;
 670        }
 671
 672        work = NULL;
 673        raw_spin_lock_irq(&worker->lock);
 674        if (!list_empty(&worker->work_list)) {
 675                work = list_first_entry(&worker->work_list,
 676                                        struct kthread_work, node);
 677                list_del_init(&work->node);
 678        }
 679        worker->current_work = work;
 680        raw_spin_unlock_irq(&worker->lock);
 681
 682        if (work) {
 683                __set_current_state(TASK_RUNNING);
 684                work->func(work);
 685        } else if (!freezing(current))
 686                schedule();
 687
 688        try_to_freeze();
 689        cond_resched();
 690        goto repeat;
 691}
 692EXPORT_SYMBOL_GPL(kthread_worker_fn);
 693
 694static __printf(3, 0) struct kthread_worker *
 695__kthread_create_worker(int cpu, unsigned int flags,
 696                        const char namefmt[], va_list args)
 697{
 698        struct kthread_worker *worker;
 699        struct task_struct *task;
 700        int node = NUMA_NO_NODE;
 701
 702        worker = kzalloc(sizeof(*worker), GFP_KERNEL);
 703        if (!worker)
 704                return ERR_PTR(-ENOMEM);
 705
 706        kthread_init_worker(worker);
 707
 708        if (cpu >= 0)
 709                node = cpu_to_node(cpu);
 710
 711        task = __kthread_create_on_node(kthread_worker_fn, worker,
 712                                                node, namefmt, args);
 713        if (IS_ERR(task))
 714                goto fail_task;
 715
 716        if (cpu >= 0)
 717                kthread_bind(task, cpu);
 718
 719        worker->flags = flags;
 720        worker->task = task;
 721        wake_up_process(task);
 722        return worker;
 723
 724fail_task:
 725        kfree(worker);
 726        return ERR_CAST(task);
 727}
 728
 729/**
 730 * kthread_create_worker - create a kthread worker
 731 * @flags: flags modifying the default behavior of the worker
 732 * @namefmt: printf-style name for the kthread worker (task).
 733 *
 734 * Returns a pointer to the allocated worker on success, ERR_PTR(-ENOMEM)
 735 * when the needed structures could not get allocated, and ERR_PTR(-EINTR)
 736 * when the worker was SIGKILLed.
 737 */
 738struct kthread_worker *
 739kthread_create_worker(unsigned int flags, const char namefmt[], ...)
 740{
 741        struct kthread_worker *worker;
 742        va_list args;
 743
 744        va_start(args, namefmt);
 745        worker = __kthread_create_worker(-1, flags, namefmt, args);
 746        va_end(args);
 747
 748        return worker;
 749}
 750EXPORT_SYMBOL(kthread_create_worker);
 751
 752/**
 753 * kthread_create_worker_on_cpu - create a kthread worker and bind it
 754 *      it to a given CPU and the associated NUMA node.
 755 * @cpu: CPU number
 756 * @flags: flags modifying the default behavior of the worker
 757 * @namefmt: printf-style name for the kthread worker (task).
 758 *
 759 * Use a valid CPU number if you want to bind the kthread worker
 760 * to the given CPU and the associated NUMA node.
 761 *
 762 * A good practice is to add the cpu number also into the worker name.
 763 * For example, use kthread_create_worker_on_cpu(cpu, "helper/%d", cpu).
 764 *
 765 * Returns a pointer to the allocated worker on success, ERR_PTR(-ENOMEM)
 766 * when the needed structures could not get allocated, and ERR_PTR(-EINTR)
 767 * when the worker was SIGKILLed.
 768 */
 769struct kthread_worker *
 770kthread_create_worker_on_cpu(int cpu, unsigned int flags,
 771                             const char namefmt[], ...)
 772{
 773        struct kthread_worker *worker;
 774        va_list args;
 775
 776        va_start(args, namefmt);
 777        worker = __kthread_create_worker(cpu, flags, namefmt, args);
 778        va_end(args);
 779
 780        return worker;
 781}
 782EXPORT_SYMBOL(kthread_create_worker_on_cpu);
 783
 784/*
 785 * Returns true when the work could not be queued at the moment.
 786 * It happens when it is already pending in a worker list
 787 * or when it is being cancelled.
 788 */
 789static inline bool queuing_blocked(struct kthread_worker *worker,
 790                                   struct kthread_work *work)
 791{
 792        lockdep_assert_held(&worker->lock);
 793
 794        return !list_empty(&work->node) || work->canceling;
 795}
 796
 797static void kthread_insert_work_sanity_check(struct kthread_worker *worker,
 798                                             struct kthread_work *work)
 799{
 800        lockdep_assert_held(&worker->lock);
 801        WARN_ON_ONCE(!list_empty(&work->node));
 802        /* Do not use a work with >1 worker, see kthread_queue_work() */
 803        WARN_ON_ONCE(work->worker && work->worker != worker);
 804}
 805
 806/* insert @work before @pos in @worker */
 807static void kthread_insert_work(struct kthread_worker *worker,
 808                                struct kthread_work *work,
 809                                struct list_head *pos)
 810{
 811        kthread_insert_work_sanity_check(worker, work);
 812
 813        list_add_tail(&work->node, pos);
 814        work->worker = worker;
 815        if (!worker->current_work && likely(worker->task))
 816                wake_up_process(worker->task);
 817}
 818
 819/**
 820 * kthread_queue_work - queue a kthread_work
 821 * @worker: target kthread_worker
 822 * @work: kthread_work to queue
 823 *
 824 * Queue @work to work processor @task for async execution.  @task
 825 * must have been created with kthread_worker_create().  Returns %true
 826 * if @work was successfully queued, %false if it was already pending.
 827 *
 828 * Reinitialize the work if it needs to be used by another worker.
 829 * For example, when the worker was stopped and started again.
 830 */
 831bool kthread_queue_work(struct kthread_worker *worker,
 832                        struct kthread_work *work)
 833{
 834        bool ret = false;
 835        unsigned long flags;
 836
 837        raw_spin_lock_irqsave(&worker->lock, flags);
 838        if (!queuing_blocked(worker, work)) {
 839                kthread_insert_work(worker, work, &worker->work_list);
 840                ret = true;
 841        }
 842        raw_spin_unlock_irqrestore(&worker->lock, flags);
 843        return ret;
 844}
 845EXPORT_SYMBOL_GPL(kthread_queue_work);
 846
 847/**
 848 * kthread_delayed_work_timer_fn - callback that queues the associated kthread
 849 *      delayed work when the timer expires.
 850 * @t: pointer to the expired timer
 851 *
 852 * The format of the function is defined by struct timer_list.
 853 * It should have been called from irqsafe timer with irq already off.
 854 */
 855void kthread_delayed_work_timer_fn(struct timer_list *t)
 856{
 857        struct kthread_delayed_work *dwork = from_timer(dwork, t, timer);
 858        struct kthread_work *work = &dwork->work;
 859        struct kthread_worker *worker = work->worker;
 860        unsigned long flags;
 861
 862        /*
 863         * This might happen when a pending work is reinitialized.
 864         * It means that it is used a wrong way.
 865         */
 866        if (WARN_ON_ONCE(!worker))
 867                return;
 868
 869        raw_spin_lock_irqsave(&worker->lock, flags);
 870        /* Work must not be used with >1 worker, see kthread_queue_work(). */
 871        WARN_ON_ONCE(work->worker != worker);
 872
 873        /* Move the work from worker->delayed_work_list. */
 874        WARN_ON_ONCE(list_empty(&work->node));
 875        list_del_init(&work->node);
 876        kthread_insert_work(worker, work, &worker->work_list);
 877
 878        raw_spin_unlock_irqrestore(&worker->lock, flags);
 879}
 880EXPORT_SYMBOL(kthread_delayed_work_timer_fn);
 881
 882static void __kthread_queue_delayed_work(struct kthread_worker *worker,
 883                                         struct kthread_delayed_work *dwork,
 884                                         unsigned long delay)
 885{
 886        struct timer_list *timer = &dwork->timer;
 887        struct kthread_work *work = &dwork->work;
 888
 889        WARN_ON_ONCE(timer->function != kthread_delayed_work_timer_fn);
 890
 891        /*
 892         * If @delay is 0, queue @dwork->work immediately.  This is for
 893         * both optimization and correctness.  The earliest @timer can
 894         * expire is on the closest next tick and delayed_work users depend
 895         * on that there's no such delay when @delay is 0.
 896         */
 897        if (!delay) {
 898                kthread_insert_work(worker, work, &worker->work_list);
 899                return;
 900        }
 901
 902        /* Be paranoid and try to detect possible races already now. */
 903        kthread_insert_work_sanity_check(worker, work);
 904
 905        list_add(&work->node, &worker->delayed_work_list);
 906        work->worker = worker;
 907        timer->expires = jiffies + delay;
 908        add_timer(timer);
 909}
 910
 911/**
 912 * kthread_queue_delayed_work - queue the associated kthread work
 913 *      after a delay.
 914 * @worker: target kthread_worker
 915 * @dwork: kthread_delayed_work to queue
 916 * @delay: number of jiffies to wait before queuing
 917 *
 918 * If the work has not been pending it starts a timer that will queue
 919 * the work after the given @delay. If @delay is zero, it queues the
 920 * work immediately.
 921 *
 922 * Return: %false if the @work has already been pending. It means that
 923 * either the timer was running or the work was queued. It returns %true
 924 * otherwise.
 925 */
 926bool kthread_queue_delayed_work(struct kthread_worker *worker,
 927                                struct kthread_delayed_work *dwork,
 928                                unsigned long delay)
 929{
 930        struct kthread_work *work = &dwork->work;
 931        unsigned long flags;
 932        bool ret = false;
 933
 934        raw_spin_lock_irqsave(&worker->lock, flags);
 935
 936        if (!queuing_blocked(worker, work)) {
 937                __kthread_queue_delayed_work(worker, dwork, delay);
 938                ret = true;
 939        }
 940
 941        raw_spin_unlock_irqrestore(&worker->lock, flags);
 942        return ret;
 943}
 944EXPORT_SYMBOL_GPL(kthread_queue_delayed_work);
 945
 946struct kthread_flush_work {
 947        struct kthread_work     work;
 948        struct completion       done;
 949};
 950
 951static void kthread_flush_work_fn(struct kthread_work *work)
 952{
 953        struct kthread_flush_work *fwork =
 954                container_of(work, struct kthread_flush_work, work);
 955        complete(&fwork->done);
 956}
 957
 958/**
 959 * kthread_flush_work - flush a kthread_work
 960 * @work: work to flush
 961 *
 962 * If @work is queued or executing, wait for it to finish execution.
 963 */
 964void kthread_flush_work(struct kthread_work *work)
 965{
 966        struct kthread_flush_work fwork = {
 967                KTHREAD_WORK_INIT(fwork.work, kthread_flush_work_fn),
 968                COMPLETION_INITIALIZER_ONSTACK(fwork.done),
 969        };
 970        struct kthread_worker *worker;
 971        bool noop = false;
 972
 973        worker = work->worker;
 974        if (!worker)
 975                return;
 976
 977        raw_spin_lock_irq(&worker->lock);
 978        /* Work must not be used with >1 worker, see kthread_queue_work(). */
 979        WARN_ON_ONCE(work->worker != worker);
 980
 981        if (!list_empty(&work->node))
 982                kthread_insert_work(worker, &fwork.work, work->node.next);
 983        else if (worker->current_work == work)
 984                kthread_insert_work(worker, &fwork.work,
 985                                    worker->work_list.next);
 986        else
 987                noop = true;
 988
 989        raw_spin_unlock_irq(&worker->lock);
 990
 991        if (!noop)
 992                wait_for_completion(&fwork.done);
 993}
 994EXPORT_SYMBOL_GPL(kthread_flush_work);
 995
 996/*
 997 * This function removes the work from the worker queue. Also it makes sure
 998 * that it won't get queued later via the delayed work's timer.
 999 *
1000 * The work might still be in use when this function finishes. See the
1001 * current_work proceed by the worker.
1002 *
1003 * Return: %true if @work was pending and successfully canceled,
1004 *      %false if @work was not pending
1005 */
1006static bool __kthread_cancel_work(struct kthread_work *work, bool is_dwork,
1007                                  unsigned long *flags)
1008{
1009        /* Try to cancel the timer if exists. */
1010        if (is_dwork) {
1011                struct kthread_delayed_work *dwork =
1012                        container_of(work, struct kthread_delayed_work, work);
1013                struct kthread_worker *worker = work->worker;
1014
1015                /*
1016                 * del_timer_sync() must be called to make sure that the timer
1017                 * callback is not running. The lock must be temporary released
1018                 * to avoid a deadlock with the callback. In the meantime,
1019                 * any queuing is blocked by setting the canceling counter.
1020                 */
1021                work->canceling++;
1022                raw_spin_unlock_irqrestore(&worker->lock, *flags);
1023                del_timer_sync(&dwork->timer);
1024                raw_spin_lock_irqsave(&worker->lock, *flags);
1025                work->canceling--;
1026        }
1027
1028        /*
1029         * Try to remove the work from a worker list. It might either
1030         * be from worker->work_list or from worker->delayed_work_list.
1031         */
1032        if (!list_empty(&work->node)) {
1033                list_del_init(&work->node);
1034                return true;
1035        }
1036
1037        return false;
1038}
1039
1040/**
1041 * kthread_mod_delayed_work - modify delay of or queue a kthread delayed work
1042 * @worker: kthread worker to use
1043 * @dwork: kthread delayed work to queue
1044 * @delay: number of jiffies to wait before queuing
1045 *
1046 * If @dwork is idle, equivalent to kthread_queue_delayed_work(). Otherwise,
1047 * modify @dwork's timer so that it expires after @delay. If @delay is zero,
1048 * @work is guaranteed to be queued immediately.
1049 *
1050 * Return: %true if @dwork was pending and its timer was modified,
1051 * %false otherwise.
1052 *
1053 * A special case is when the work is being canceled in parallel.
1054 * It might be caused either by the real kthread_cancel_delayed_work_sync()
1055 * or yet another kthread_mod_delayed_work() call. We let the other command
1056 * win and return %false here. The caller is supposed to synchronize these
1057 * operations a reasonable way.
1058 *
1059 * This function is safe to call from any context including IRQ handler.
1060 * See __kthread_cancel_work() and kthread_delayed_work_timer_fn()
1061 * for details.
1062 */
1063bool kthread_mod_delayed_work(struct kthread_worker *worker,
1064                              struct kthread_delayed_work *dwork,
1065                              unsigned long delay)
1066{
1067        struct kthread_work *work = &dwork->work;
1068        unsigned long flags;
1069        int ret = false;
1070
1071        raw_spin_lock_irqsave(&worker->lock, flags);
1072
1073        /* Do not bother with canceling when never queued. */
1074        if (!work->worker)
1075                goto fast_queue;
1076
1077        /* Work must not be used with >1 worker, see kthread_queue_work() */
1078        WARN_ON_ONCE(work->worker != worker);
1079
1080        /* Do not fight with another command that is canceling this work. */
1081        if (work->canceling)
1082                goto out;
1083
1084        ret = __kthread_cancel_work(work, true, &flags);
1085fast_queue:
1086        __kthread_queue_delayed_work(worker, dwork, delay);
1087out:
1088        raw_spin_unlock_irqrestore(&worker->lock, flags);
1089        return ret;
1090}
1091EXPORT_SYMBOL_GPL(kthread_mod_delayed_work);
1092
1093static bool __kthread_cancel_work_sync(struct kthread_work *work, bool is_dwork)
1094{
1095        struct kthread_worker *worker = work->worker;
1096        unsigned long flags;
1097        int ret = false;
1098
1099        if (!worker)
1100                goto out;
1101
1102        raw_spin_lock_irqsave(&worker->lock, flags);
1103        /* Work must not be used with >1 worker, see kthread_queue_work(). */
1104        WARN_ON_ONCE(work->worker != worker);
1105
1106        ret = __kthread_cancel_work(work, is_dwork, &flags);
1107
1108        if (worker->current_work != work)
1109                goto out_fast;
1110
1111        /*
1112         * The work is in progress and we need to wait with the lock released.
1113         * In the meantime, block any queuing by setting the canceling counter.
1114         */
1115        work->canceling++;
1116        raw_spin_unlock_irqrestore(&worker->lock, flags);
1117        kthread_flush_work(work);
1118        raw_spin_lock_irqsave(&worker->lock, flags);
1119        work->canceling--;
1120
1121out_fast:
1122        raw_spin_unlock_irqrestore(&worker->lock, flags);
1123out:
1124        return ret;
1125}
1126
1127/**
1128 * kthread_cancel_work_sync - cancel a kthread work and wait for it to finish
1129 * @work: the kthread work to cancel
1130 *
1131 * Cancel @work and wait for its execution to finish.  This function
1132 * can be used even if the work re-queues itself. On return from this
1133 * function, @work is guaranteed to be not pending or executing on any CPU.
1134 *
1135 * kthread_cancel_work_sync(&delayed_work->work) must not be used for
1136 * delayed_work's. Use kthread_cancel_delayed_work_sync() instead.
1137 *
1138 * The caller must ensure that the worker on which @work was last
1139 * queued can't be destroyed before this function returns.
1140 *
1141 * Return: %true if @work was pending, %false otherwise.
1142 */
1143bool kthread_cancel_work_sync(struct kthread_work *work)
1144{
1145        return __kthread_cancel_work_sync(work, false);
1146}
1147EXPORT_SYMBOL_GPL(kthread_cancel_work_sync);
1148
1149/**
1150 * kthread_cancel_delayed_work_sync - cancel a kthread delayed work and
1151 *      wait for it to finish.
1152 * @dwork: the kthread delayed work to cancel
1153 *
1154 * This is kthread_cancel_work_sync() for delayed works.
1155 *
1156 * Return: %true if @dwork was pending, %false otherwise.
1157 */
1158bool kthread_cancel_delayed_work_sync(struct kthread_delayed_work *dwork)
1159{
1160        return __kthread_cancel_work_sync(&dwork->work, true);
1161}
1162EXPORT_SYMBOL_GPL(kthread_cancel_delayed_work_sync);
1163
1164/**
1165 * kthread_flush_worker - flush all current works on a kthread_worker
1166 * @worker: worker to flush
1167 *
1168 * Wait until all currently executing or pending works on @worker are
1169 * finished.
1170 */
1171void kthread_flush_worker(struct kthread_worker *worker)
1172{
1173        struct kthread_flush_work fwork = {
1174                KTHREAD_WORK_INIT(fwork.work, kthread_flush_work_fn),
1175                COMPLETION_INITIALIZER_ONSTACK(fwork.done),
1176        };
1177
1178        kthread_queue_work(worker, &fwork.work);
1179        wait_for_completion(&fwork.done);
1180}
1181EXPORT_SYMBOL_GPL(kthread_flush_worker);
1182
1183/**
1184 * kthread_destroy_worker - destroy a kthread worker
1185 * @worker: worker to be destroyed
1186 *
1187 * Flush and destroy @worker.  The simple flush is enough because the kthread
1188 * worker API is used only in trivial scenarios.  There are no multi-step state
1189 * machines needed.
1190 */
1191void kthread_destroy_worker(struct kthread_worker *worker)
1192{
1193        struct task_struct *task;
1194
1195        task = worker->task;
1196        if (WARN_ON(!task))
1197                return;
1198
1199        kthread_flush_worker(worker);
1200        kthread_stop(task);
1201        WARN_ON(!list_empty(&worker->work_list));
1202        kfree(worker);
1203}
1204EXPORT_SYMBOL(kthread_destroy_worker);
1205
1206#ifdef CONFIG_BLK_CGROUP
1207/**
1208 * kthread_associate_blkcg - associate blkcg to current kthread
1209 * @css: the cgroup info
1210 *
1211 * Current thread must be a kthread. The thread is running jobs on behalf of
1212 * other threads. In some cases, we expect the jobs attach cgroup info of
1213 * original threads instead of that of current thread. This function stores
1214 * original thread's cgroup info in current kthread context for later
1215 * retrieval.
1216 */
1217void kthread_associate_blkcg(struct cgroup_subsys_state *css)
1218{
1219        struct kthread *kthread;
1220
1221        if (!(current->flags & PF_KTHREAD))
1222                return;
1223        kthread = to_kthread(current);
1224        if (!kthread)
1225                return;
1226
1227        if (kthread->blkcg_css) {
1228                css_put(kthread->blkcg_css);
1229                kthread->blkcg_css = NULL;
1230        }
1231        if (css) {
1232                css_get(css);
1233                kthread->blkcg_css = css;
1234        }
1235}
1236EXPORT_SYMBOL(kthread_associate_blkcg);
1237
1238/**
1239 * kthread_blkcg - get associated blkcg css of current kthread
1240 *
1241 * Current thread must be a kthread.
1242 */
1243struct cgroup_subsys_state *kthread_blkcg(void)
1244{
1245        struct kthread *kthread;
1246
1247        if (current->flags & PF_KTHREAD) {
1248                kthread = to_kthread(current);
1249                if (kthread)
1250                        return kthread->blkcg_css;
1251        }
1252        return NULL;
1253}
1254EXPORT_SYMBOL(kthread_blkcg);
1255#endif
1256