xref: /linux-6.15/include/linux/futex.h (revision 4dbed85a)
1 #ifndef _LINUX_FUTEX_H
2 #define _LINUX_FUTEX_H
3 
4 #include <linux/sched.h>
5 
6 union ktime;
7 
8 /* Second argument to futex syscall */
9 
10 
11 #define FUTEX_WAIT		0
12 #define FUTEX_WAKE		1
13 #define FUTEX_FD		2
14 #define FUTEX_REQUEUE		3
15 #define FUTEX_CMP_REQUEUE	4
16 #define FUTEX_WAKE_OP		5
17 #define FUTEX_LOCK_PI		6
18 #define FUTEX_UNLOCK_PI		7
19 #define FUTEX_TRYLOCK_PI	8
20 
21 #define FUTEX_PRIVATE_FLAG	128
22 #define FUTEX_CMD_MASK		~FUTEX_PRIVATE_FLAG
23 
24 #define FUTEX_WAIT_PRIVATE	(FUTEX_WAIT | FUTEX_PRIVATE_FLAG)
25 #define FUTEX_WAKE_PRIVATE	(FUTEX_WAKE | FUTEX_PRIVATE_FLAG)
26 #define FUTEX_REQUEUE_PRIVATE	(FUTEX_REQUEUE | FUTEX_PRIVATE_FLAG)
27 #define FUTEX_CMP_REQUEUE_PRIVATE (FUTEX_CMP_REQUEUE | FUTEX_PRIVATE_FLAG)
28 #define FUTEX_WAKE_OP_PRIVATE	(FUTEX_WAKE_OP | FUTEX_PRIVATE_FLAG)
29 #define FUTEX_LOCK_PI_PRIVATE	(FUTEX_LOCK_PI | FUTEX_PRIVATE_FLAG)
30 #define FUTEX_UNLOCK_PI_PRIVATE	(FUTEX_UNLOCK_PI | FUTEX_PRIVATE_FLAG)
31 #define FUTEX_TRYLOCK_PI_PRIVATE (FUTEX_TRYLOCK_PI | FUTEX_PRIVATE_FLAG)
32 
33 /*
34  * Support for robust futexes: the kernel cleans up held futexes at
35  * thread exit time.
36  */
37 
38 /*
39  * Per-lock list entry - embedded in user-space locks, somewhere close
40  * to the futex field. (Note: user-space uses a double-linked list to
41  * achieve O(1) list add and remove, but the kernel only needs to know
42  * about the forward link)
43  *
44  * NOTE: this structure is part of the syscall ABI, and must not be
45  * changed.
46  */
47 struct robust_list {
48 	struct robust_list __user *next;
49 };
50 
51 /*
52  * Per-thread list head:
53  *
54  * NOTE: this structure is part of the syscall ABI, and must only be
55  * changed if the change is first communicated with the glibc folks.
56  * (When an incompatible change is done, we'll increase the structure
57  *  size, which glibc will detect)
58  */
59 struct robust_list_head {
60 	/*
61 	 * The head of the list. Points back to itself if empty:
62 	 */
63 	struct robust_list list;
64 
65 	/*
66 	 * This relative offset is set by user-space, it gives the kernel
67 	 * the relative position of the futex field to examine. This way
68 	 * we keep userspace flexible, to freely shape its data-structure,
69 	 * without hardcoding any particular offset into the kernel:
70 	 */
71 	long futex_offset;
72 
73 	/*
74 	 * The death of the thread may race with userspace setting
75 	 * up a lock's links. So to handle this race, userspace first
76 	 * sets this field to the address of the to-be-taken lock,
77 	 * then does the lock acquire, and then adds itself to the
78 	 * list, and then clears this field. Hence the kernel will
79 	 * always have full knowledge of all locks that the thread
80 	 * _might_ have taken. We check the owner TID in any case,
81 	 * so only truly owned locks will be handled.
82 	 */
83 	struct robust_list __user *list_op_pending;
84 };
85 
86 /*
87  * Are there any waiters for this robust futex:
88  */
89 #define FUTEX_WAITERS		0x80000000
90 
91 /*
92  * The kernel signals via this bit that a thread holding a futex
93  * has exited without unlocking the futex. The kernel also does
94  * a FUTEX_WAKE on such futexes, after setting the bit, to wake
95  * up any possible waiters:
96  */
97 #define FUTEX_OWNER_DIED	0x40000000
98 
99 /*
100  * The rest of the robust-futex field is for the TID:
101  */
102 #define FUTEX_TID_MASK		0x3fffffff
103 
104 /*
105  * This limit protects against a deliberately circular list.
106  * (Not worth introducing an rlimit for it)
107  */
108 #define ROBUST_LIST_LIMIT	2048
109 
110 #ifdef __KERNEL__
111 long do_futex(u32 __user *uaddr, int op, u32 val, union ktime *timeout,
112 	      u32 __user *uaddr2, u32 val2, u32 val3);
113 
114 extern int
115 handle_futex_death(u32 __user *uaddr, struct task_struct *curr, int pi);
116 
117 /*
118  * Futexes are matched on equal values of this key.
119  * The key type depends on whether it's a shared or private mapping.
120  * Don't rearrange members without looking at hash_futex().
121  *
122  * offset is aligned to a multiple of sizeof(u32) (== 4) by definition.
123  * We use the two low order bits of offset to tell what is the kind of key :
124  *  00 : Private process futex (PTHREAD_PROCESS_PRIVATE)
125  *       (no reference on an inode or mm)
126  *  01 : Shared futex (PTHREAD_PROCESS_SHARED)
127  *	mapped on a file (reference on the underlying inode)
128  *  10 : Shared futex (PTHREAD_PROCESS_SHARED)
129  *       (but private mapping on an mm, and reference taken on it)
130 */
131 
132 #define FUT_OFF_INODE    1 /* We set bit 0 if key has a reference on inode */
133 #define FUT_OFF_MMSHARED 2 /* We set bit 1 if key has a reference on mm */
134 
135 union futex_key {
136 	struct {
137 		unsigned long pgoff;
138 		struct inode *inode;
139 		int offset;
140 	} shared;
141 	struct {
142 		unsigned long address;
143 		struct mm_struct *mm;
144 		int offset;
145 	} private;
146 	struct {
147 		unsigned long word;
148 		void *ptr;
149 		int offset;
150 	} both;
151 };
152 
153 #ifdef CONFIG_FUTEX
154 extern void exit_robust_list(struct task_struct *curr);
155 extern void exit_pi_state_list(struct task_struct *curr);
156 #else
157 static inline void exit_robust_list(struct task_struct *curr)
158 {
159 }
160 static inline void exit_pi_state_list(struct task_struct *curr)
161 {
162 }
163 #endif
164 #endif /* __KERNEL__ */
165 
166 #define FUTEX_OP_SET		0	/* *(int *)UADDR2 = OPARG; */
167 #define FUTEX_OP_ADD		1	/* *(int *)UADDR2 += OPARG; */
168 #define FUTEX_OP_OR		2	/* *(int *)UADDR2 |= OPARG; */
169 #define FUTEX_OP_ANDN		3	/* *(int *)UADDR2 &= ~OPARG; */
170 #define FUTEX_OP_XOR		4	/* *(int *)UADDR2 ^= OPARG; */
171 
172 #define FUTEX_OP_OPARG_SHIFT	8	/* Use (1 << OPARG) instead of OPARG.  */
173 
174 #define FUTEX_OP_CMP_EQ		0	/* if (oldval == CMPARG) wake */
175 #define FUTEX_OP_CMP_NE		1	/* if (oldval != CMPARG) wake */
176 #define FUTEX_OP_CMP_LT		2	/* if (oldval < CMPARG) wake */
177 #define FUTEX_OP_CMP_LE		3	/* if (oldval <= CMPARG) wake */
178 #define FUTEX_OP_CMP_GT		4	/* if (oldval > CMPARG) wake */
179 #define FUTEX_OP_CMP_GE		5	/* if (oldval >= CMPARG) wake */
180 
181 /* FUTEX_WAKE_OP will perform atomically
182    int oldval = *(int *)UADDR2;
183    *(int *)UADDR2 = oldval OP OPARG;
184    if (oldval CMP CMPARG)
185      wake UADDR2;  */
186 
187 #define FUTEX_OP(op, oparg, cmp, cmparg) \
188   (((op & 0xf) << 28) | ((cmp & 0xf) << 24)		\
189    | ((oparg & 0xfff) << 12) | (cmparg & 0xfff))
190 
191 #endif
192