xref: /linux-6.15/include/linux/futex.h (revision d289bf7b)
1 #ifndef _LINUX_FUTEX_H
2 #define _LINUX_FUTEX_H
3 
4 #include <linux/compiler.h>
5 #include <linux/types.h>
6 
7 struct inode;
8 struct mm_struct;
9 struct task_struct;
10 union ktime;
11 
12 /* Second argument to futex syscall */
13 
14 
15 #define FUTEX_WAIT		0
16 #define FUTEX_WAKE		1
17 #define FUTEX_FD		2
18 #define FUTEX_REQUEUE		3
19 #define FUTEX_CMP_REQUEUE	4
20 #define FUTEX_WAKE_OP		5
21 #define FUTEX_LOCK_PI		6
22 #define FUTEX_UNLOCK_PI		7
23 #define FUTEX_TRYLOCK_PI	8
24 #define FUTEX_WAIT_BITSET	9
25 #define FUTEX_WAKE_BITSET	10
26 
27 #define FUTEX_PRIVATE_FLAG	128
28 #define FUTEX_CMD_MASK		~FUTEX_PRIVATE_FLAG
29 
30 #define FUTEX_WAIT_PRIVATE	(FUTEX_WAIT | FUTEX_PRIVATE_FLAG)
31 #define FUTEX_WAKE_PRIVATE	(FUTEX_WAKE | FUTEX_PRIVATE_FLAG)
32 #define FUTEX_REQUEUE_PRIVATE	(FUTEX_REQUEUE | FUTEX_PRIVATE_FLAG)
33 #define FUTEX_CMP_REQUEUE_PRIVATE (FUTEX_CMP_REQUEUE | FUTEX_PRIVATE_FLAG)
34 #define FUTEX_WAKE_OP_PRIVATE	(FUTEX_WAKE_OP | FUTEX_PRIVATE_FLAG)
35 #define FUTEX_LOCK_PI_PRIVATE	(FUTEX_LOCK_PI | FUTEX_PRIVATE_FLAG)
36 #define FUTEX_UNLOCK_PI_PRIVATE	(FUTEX_UNLOCK_PI | FUTEX_PRIVATE_FLAG)
37 #define FUTEX_TRYLOCK_PI_PRIVATE (FUTEX_TRYLOCK_PI | FUTEX_PRIVATE_FLAG)
38 #define FUTEX_WAIT_BITSET_PRIVATE	(FUTEX_WAIT_BITS | FUTEX_PRIVATE_FLAG)
39 #define FUTEX_WAKE_BITSET_PRIVATE	(FUTEX_WAKE_BITS | FUTEX_PRIVATE_FLAG)
40 
41 /*
42  * Support for robust futexes: the kernel cleans up held futexes at
43  * thread exit time.
44  */
45 
46 /*
47  * Per-lock list entry - embedded in user-space locks, somewhere close
48  * to the futex field. (Note: user-space uses a double-linked list to
49  * achieve O(1) list add and remove, but the kernel only needs to know
50  * about the forward link)
51  *
52  * NOTE: this structure is part of the syscall ABI, and must not be
53  * changed.
54  */
55 struct robust_list {
56 	struct robust_list __user *next;
57 };
58 
59 /*
60  * Per-thread list head:
61  *
62  * NOTE: this structure is part of the syscall ABI, and must only be
63  * changed if the change is first communicated with the glibc folks.
64  * (When an incompatible change is done, we'll increase the structure
65  *  size, which glibc will detect)
66  */
67 struct robust_list_head {
68 	/*
69 	 * The head of the list. Points back to itself if empty:
70 	 */
71 	struct robust_list list;
72 
73 	/*
74 	 * This relative offset is set by user-space, it gives the kernel
75 	 * the relative position of the futex field to examine. This way
76 	 * we keep userspace flexible, to freely shape its data-structure,
77 	 * without hardcoding any particular offset into the kernel:
78 	 */
79 	long futex_offset;
80 
81 	/*
82 	 * The death of the thread may race with userspace setting
83 	 * up a lock's links. So to handle this race, userspace first
84 	 * sets this field to the address of the to-be-taken lock,
85 	 * then does the lock acquire, and then adds itself to the
86 	 * list, and then clears this field. Hence the kernel will
87 	 * always have full knowledge of all locks that the thread
88 	 * _might_ have taken. We check the owner TID in any case,
89 	 * so only truly owned locks will be handled.
90 	 */
91 	struct robust_list __user *list_op_pending;
92 };
93 
94 /*
95  * Are there any waiters for this robust futex:
96  */
97 #define FUTEX_WAITERS		0x80000000
98 
99 /*
100  * The kernel signals via this bit that a thread holding a futex
101  * has exited without unlocking the futex. The kernel also does
102  * a FUTEX_WAKE on such futexes, after setting the bit, to wake
103  * up any possible waiters:
104  */
105 #define FUTEX_OWNER_DIED	0x40000000
106 
107 /*
108  * The rest of the robust-futex field is for the TID:
109  */
110 #define FUTEX_TID_MASK		0x3fffffff
111 
112 /*
113  * This limit protects against a deliberately circular list.
114  * (Not worth introducing an rlimit for it)
115  */
116 #define ROBUST_LIST_LIMIT	2048
117 
118 /*
119  * bitset with all bits set for the FUTEX_xxx_BITSET OPs to request a
120  * match of any bit.
121  */
122 #define FUTEX_BITSET_MATCH_ANY	0xffffffff
123 
124 #ifdef __KERNEL__
125 long do_futex(u32 __user *uaddr, int op, u32 val, union ktime *timeout,
126 	      u32 __user *uaddr2, u32 val2, u32 val3);
127 
128 extern int
129 handle_futex_death(u32 __user *uaddr, struct task_struct *curr, int pi);
130 
131 /*
132  * Futexes are matched on equal values of this key.
133  * The key type depends on whether it's a shared or private mapping.
134  * Don't rearrange members without looking at hash_futex().
135  *
136  * offset is aligned to a multiple of sizeof(u32) (== 4) by definition.
137  * We use the two low order bits of offset to tell what is the kind of key :
138  *  00 : Private process futex (PTHREAD_PROCESS_PRIVATE)
139  *       (no reference on an inode or mm)
140  *  01 : Shared futex (PTHREAD_PROCESS_SHARED)
141  *	mapped on a file (reference on the underlying inode)
142  *  10 : Shared futex (PTHREAD_PROCESS_SHARED)
143  *       (but private mapping on an mm, and reference taken on it)
144 */
145 
146 #define FUT_OFF_INODE    1 /* We set bit 0 if key has a reference on inode */
147 #define FUT_OFF_MMSHARED 2 /* We set bit 1 if key has a reference on mm */
148 
149 union futex_key {
150 	struct {
151 		unsigned long pgoff;
152 		struct inode *inode;
153 		int offset;
154 	} shared;
155 	struct {
156 		unsigned long address;
157 		struct mm_struct *mm;
158 		int offset;
159 	} private;
160 	struct {
161 		unsigned long word;
162 		void *ptr;
163 		int offset;
164 	} both;
165 };
166 
167 #ifdef CONFIG_FUTEX
168 extern void exit_robust_list(struct task_struct *curr);
169 extern void exit_pi_state_list(struct task_struct *curr);
170 extern int futex_cmpxchg_enabled;
171 #else
172 static inline void exit_robust_list(struct task_struct *curr)
173 {
174 }
175 static inline void exit_pi_state_list(struct task_struct *curr)
176 {
177 }
178 #endif
179 #endif /* __KERNEL__ */
180 
181 #define FUTEX_OP_SET		0	/* *(int *)UADDR2 = OPARG; */
182 #define FUTEX_OP_ADD		1	/* *(int *)UADDR2 += OPARG; */
183 #define FUTEX_OP_OR		2	/* *(int *)UADDR2 |= OPARG; */
184 #define FUTEX_OP_ANDN		3	/* *(int *)UADDR2 &= ~OPARG; */
185 #define FUTEX_OP_XOR		4	/* *(int *)UADDR2 ^= OPARG; */
186 
187 #define FUTEX_OP_OPARG_SHIFT	8	/* Use (1 << OPARG) instead of OPARG.  */
188 
189 #define FUTEX_OP_CMP_EQ		0	/* if (oldval == CMPARG) wake */
190 #define FUTEX_OP_CMP_NE		1	/* if (oldval != CMPARG) wake */
191 #define FUTEX_OP_CMP_LT		2	/* if (oldval < CMPARG) wake */
192 #define FUTEX_OP_CMP_LE		3	/* if (oldval <= CMPARG) wake */
193 #define FUTEX_OP_CMP_GT		4	/* if (oldval > CMPARG) wake */
194 #define FUTEX_OP_CMP_GE		5	/* if (oldval >= CMPARG) wake */
195 
196 /* FUTEX_WAKE_OP will perform atomically
197    int oldval = *(int *)UADDR2;
198    *(int *)UADDR2 = oldval OP OPARG;
199    if (oldval CMP CMPARG)
200      wake UADDR2;  */
201 
202 #define FUTEX_OP(op, oparg, cmp, cmparg) \
203   (((op & 0xf) << 28) | ((cmp & 0xf) << 24)		\
204    | ((oparg & 0xfff) << 12) | (cmparg & 0xfff))
205 
206 #endif
207