kernel/task_work.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 #include <linux/spinlock.h>
   3 #include <linux/task_work.h>
   4 #include <linux/tracehook.h>
   5
   6 static struct callback_head work_exited; /* all we need is ->next == NULL */
   7
   8 /**
   9  * task_work_add - ask the @task to execute @work->func()
  10  * @task: the task which should run the callback
  11  * @work: the callback to run
  12  * @notify: how to notify the targeted task
  13  *
  14  * Queue @work for task_work_run() below and notify the @task if @notify
  15  * is @TWA_RESUME or @TWA_SIGNAL. @TWA_SIGNAL works like signals, in that the
  16  * it will interrupt the targeted task and run the task_work. @TWA_RESUME
  17  * work is run only when the task exits the kernel and returns to user mode,
  18  * or before entering guest mode. Fails if the @task is exiting/exited and thus
  19  * it can't process this @work. Otherwise @work->func() will be called when the
  20  * @task goes through one of the aforementioned transitions, or exits.
  21  *
  22  * If the targeted task is exiting, then an error is returned and the work item
  23  * is not queued. It's up to the caller to arrange for an alternative mechanism
  24  * in that case.
  25  *
  26  * Note: there is no ordering guarantee on works queued here. The task_work
  27  * list is LIFO.
  28  *
  29  * RETURNS:
  30  * 0 if succeeds or -ESRCH.
  31  */
  32 int task_work_add(struct task_struct *task, struct callback_head *work,
  33                   enum task_work_notify_mode notify)
  34 {
  35         struct callback_head *head;
  36
  37         /* record the work call stack in order to print it in KASAN reports */
  38         kasan_record_aux_stack(work);
  39
  40         do {
  41                 head = READ_ONCE(task->task_works);
  42                 if (unlikely(head == &work_exited))
  43                         return -ESRCH;
  44                 work->next = head;
  45         } while (cmpxchg(&task->task_works, head, work) != head);
  46
  47         switch (notify) {
  48         case TWA_NONE:
  49                 break;
  50         case TWA_RESUME:
  51                 set_notify_resume(task);
  52                 break;
  53         case TWA_SIGNAL:
  54                 set_notify_signal(task);
  55                 break;
  56         default:
  57                 WARN_ON_ONCE(1);
  58                 break;
  59         }
  60
  61         return 0;
  62 }
  63 EXPORT_SYMBOL(task_work_add);
  64
  65 /**
  66  * task_work_cancel_match - cancel a pending work added by task_work_add()
  67  * @task: the task which should execute the work
  68  * @match: match function to call
  69  *
  70  * RETURNS:
  71  * The found work or NULL if not found.
  72  */
  73 struct callback_head *
  74 task_work_cancel_match(struct task_struct *task,
  75                        bool (*match)(struct callback_head *, void *data),
  76                        void *data)
  77 {
  78         struct callback_head **pprev = &task->task_works;
  79         struct callback_head *work;
  80         unsigned long flags;
  81
  82         if (likely(!task->task_works))
  83                 return NULL;
  84         /*
  85          * If cmpxchg() fails we continue without updating pprev.
  86          * Either we raced with task_work_add() which added the
  87          * new entry before this work, we will find it again. Or
  88          * we raced with task_work_run(), *pprev == NULL/exited.
  89          */
  90         raw_spin_lock_irqsave(&task->pi_lock, flags);
  91         while ((work = READ_ONCE(*pprev))) {
  92                 if (!match(work, data))
  93                         pprev = &work->next;
  94                 else if (cmpxchg(pprev, work, work->next) == work)
  95                         break;
  96         }
  97         raw_spin_unlock_irqrestore(&task->pi_lock, flags);
  98
  99         return work;
 100 }
 101
 102 static bool task_work_func_match(struct callback_head *cb, void *data)
 103 {
 104         return cb->func == data;
 105 }
 106
 107 /**
 108  * task_work_cancel - cancel a pending work added by task_work_add()
 109  * @task: the task which should execute the work
 110  * @func: identifies the work to remove
 111  *
 112  * Find the last queued pending work with ->func == @func and remove
 113  * it from queue.
 114  *
 115  * RETURNS:
 116  * The found work or NULL if not found.
 117  */
 118 struct callback_head *
 119 task_work_cancel(struct task_struct *task, task_work_func_t func)
 120 {
 121         return task_work_cancel_match(task, task_work_func_match, func);
 122 }
 123
 124 /**
 125  * task_work_run - execute the works added by task_work_add()
 126  *
 127  * Flush the pending works. Should be used by the core kernel code.
 128  * Called before the task returns to the user-mode or stops, or when
 129  * it exits. In the latter case task_work_add() can no longer add the
 130  * new work after task_work_run() returns.
 131  */
 132 void task_work_run(void)
 133 {
 134         struct task_struct *task = current;
 135         struct callback_head *work, *head, *next;
 136
 137         for (;;) {
 138                 /*
 139                  * work->func() can do task_work_add(), do not set
 140                  * work_exited unless the list is empty.
 141                  */
 142                 do {
 143                         head = NULL;
 144                         work = READ_ONCE(task->task_works);
 145                         if (!work) {
 146                                 if (task->flags & PF_EXITING)
 147                                         head = &work_exited;
 148                                 else
 149                                         break;
 150                         }
 151                 } while (cmpxchg(&task->task_works, work, head) != work);
 152
 153                 if (!work)
 154                         break;
 155                 /*
 156                  * Synchronize with task_work_cancel(). It can not remove
 157                  * the first entry == work, cmpxchg(task_works) must fail.
 158                  * But it can remove another entry from the ->next list.
 159                  */
 160                 raw_spin_lock_irq(&task->pi_lock);
 161                 raw_spin_unlock_irq(&task->pi_lock);
 162
 163                 do {
 164                         next = work->next;
 165                         work->func(work);
 166                         work = next;
 167                         cond_resched();
 168                 } while (work);
 169         }
 170 }
 171 EXPORT_SYMBOL_GPL(task_work_run);