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  */
task_work_add(struct task_struct * task,struct callback_head * work,enum task_work_notify_mode notify)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 	unsigned long flags;
37 
38 	do {
39 		head = READ_ONCE(task->task_works);
40 		if (unlikely(head == &work_exited))
41 			return -ESRCH;
42 		work->next = head;
43 	} while (cmpxchg(&task->task_works, head, work) != head);
44 
45 	switch (notify) {
46 	case TWA_NONE:
47 		break;
48 	case TWA_RESUME:
49 		set_notify_resume(task);
50 		break;
51 	case TWA_SIGNAL:
52 		/*
53 		 * Only grab the sighand lock if we don't already have some
54 		 * task_work pending. This pairs with the smp_store_mb()
55 		 * in get_signal(), see comment there.
56 		 */
57 		if (!(READ_ONCE(task->jobctl) & JOBCTL_TASK_WORK) &&
58 		    lock_task_sighand(task, &flags)) {
59 			task->jobctl |= JOBCTL_TASK_WORK;
60 			signal_wake_up(task, 0);
61 			unlock_task_sighand(task, &flags);
62 		}
63 		break;
64 	default:
65 		WARN_ON_ONCE(1);
66 		break;
67 	}
68 
69 	return 0;
70 }
71 
72 /**
73  * task_work_cancel - cancel a pending work added by task_work_add()
74  * @task: the task which should execute the work
75  * @func: identifies the work to remove
76  *
77  * Find the last queued pending work with ->func == @func and remove
78  * it from queue.
79  *
80  * RETURNS:
81  * The found work or NULL if not found.
82  */
83 struct callback_head *
task_work_cancel(struct task_struct * task,task_work_func_t func)84 task_work_cancel(struct task_struct *task, task_work_func_t func)
85 {
86 	struct callback_head **pprev = &task->task_works;
87 	struct callback_head *work;
88 	unsigned long flags;
89 
90 	if (likely(!task->task_works))
91 		return NULL;
92 	/*
93 	 * If cmpxchg() fails we continue without updating pprev.
94 	 * Either we raced with task_work_add() which added the
95 	 * new entry before this work, we will find it again. Or
96 	 * we raced with task_work_run(), *pprev == NULL/exited.
97 	 */
98 	raw_spin_lock_irqsave(&task->pi_lock, flags);
99 	while ((work = READ_ONCE(*pprev))) {
100 		if (work->func != func)
101 			pprev = &work->next;
102 		else if (cmpxchg(pprev, work, work->next) == work)
103 			break;
104 	}
105 	raw_spin_unlock_irqrestore(&task->pi_lock, flags);
106 
107 	return work;
108 }
109 
110 /**
111  * task_work_run - execute the works added by task_work_add()
112  *
113  * Flush the pending works. Should be used by the core kernel code.
114  * Called before the task returns to the user-mode or stops, or when
115  * it exits. In the latter case task_work_add() can no longer add the
116  * new work after task_work_run() returns.
117  */
task_work_run(void)118 void task_work_run(void)
119 {
120 	struct task_struct *task = current;
121 	struct callback_head *work, *head, *next;
122 
123 	for (;;) {
124 		/*
125 		 * work->func() can do task_work_add(), do not set
126 		 * work_exited unless the list is empty.
127 		 */
128 		do {
129 			head = NULL;
130 			work = READ_ONCE(task->task_works);
131 			if (!work) {
132 				if (task->flags & PF_EXITING)
133 					head = &work_exited;
134 				else
135 					break;
136 			}
137 		} while (cmpxchg(&task->task_works, work, head) != work);
138 
139 		if (!work)
140 			break;
141 		/*
142 		 * Synchronize with task_work_cancel(). It can not remove
143 		 * the first entry == work, cmpxchg(task_works) must fail.
144 		 * But it can remove another entry from the ->next list.
145 		 */
146 		raw_spin_lock_irq(&task->pi_lock);
147 		raw_spin_unlock_irq(&task->pi_lock);
148 
149 		do {
150 			next = work->next;
151 			work->func(work);
152 			work = next;
153 			cond_resched();
154 		} while (work);
155 	}
156 }
157