1 | /* SPDX-License-Identifier: GPL-2.0 */ |
2 | #ifndef _FUTEX_H |
3 | #define _FUTEX_H |
4 | |
5 | #include <linux/futex.h> |
6 | #include <linux/rtmutex.h> |
7 | #include <linux/sched/wake_q.h> |
8 | #include <linux/compat.h> |
9 | |
10 | #ifdef CONFIG_PREEMPT_RT |
11 | #include <linux/rcuwait.h> |
12 | #endif |
13 | |
14 | #include <asm/futex.h> |
15 | |
16 | /* |
17 | * Futex flags used to encode options to functions and preserve them across |
18 | * restarts. |
19 | */ |
20 | #define FLAGS_SIZE_8 0x0000 |
21 | #define FLAGS_SIZE_16 0x0001 |
22 | #define FLAGS_SIZE_32 0x0002 |
23 | #define FLAGS_SIZE_64 0x0003 |
24 | |
25 | #define FLAGS_SIZE_MASK 0x0003 |
26 | |
27 | #ifdef CONFIG_MMU |
28 | # define FLAGS_SHARED 0x0010 |
29 | #else |
30 | /* |
31 | * NOMMU does not have per process address space. Let the compiler optimize |
32 | * code away. |
33 | */ |
34 | # define FLAGS_SHARED 0x0000 |
35 | #endif |
36 | #define FLAGS_CLOCKRT 0x0020 |
37 | #define FLAGS_HAS_TIMEOUT 0x0040 |
38 | #define FLAGS_NUMA 0x0080 |
39 | #define FLAGS_STRICT 0x0100 |
40 | |
41 | /* FUTEX_ to FLAGS_ */ |
42 | static inline unsigned int futex_to_flags(unsigned int op) |
43 | { |
44 | unsigned int flags = FLAGS_SIZE_32; |
45 | |
46 | if (!(op & FUTEX_PRIVATE_FLAG)) |
47 | flags |= FLAGS_SHARED; |
48 | |
49 | if (op & FUTEX_CLOCK_REALTIME) |
50 | flags |= FLAGS_CLOCKRT; |
51 | |
52 | return flags; |
53 | } |
54 | |
55 | #define FUTEX2_VALID_MASK (FUTEX2_SIZE_MASK | FUTEX2_PRIVATE) |
56 | |
57 | /* FUTEX2_ to FLAGS_ */ |
58 | static inline unsigned int futex2_to_flags(unsigned int flags2) |
59 | { |
60 | unsigned int flags = flags2 & FUTEX2_SIZE_MASK; |
61 | |
62 | if (!(flags2 & FUTEX2_PRIVATE)) |
63 | flags |= FLAGS_SHARED; |
64 | |
65 | if (flags2 & FUTEX2_NUMA) |
66 | flags |= FLAGS_NUMA; |
67 | |
68 | return flags; |
69 | } |
70 | |
71 | static inline unsigned int futex_size(unsigned int flags) |
72 | { |
73 | return 1 << (flags & FLAGS_SIZE_MASK); |
74 | } |
75 | |
76 | static inline bool futex_flags_valid(unsigned int flags) |
77 | { |
78 | /* Only 64bit futexes for 64bit code */ |
79 | if (!IS_ENABLED(CONFIG_64BIT) || in_compat_syscall()) { |
80 | if ((flags & FLAGS_SIZE_MASK) == FLAGS_SIZE_64) |
81 | return false; |
82 | } |
83 | |
84 | /* Only 32bit futexes are implemented -- for now */ |
85 | if ((flags & FLAGS_SIZE_MASK) != FLAGS_SIZE_32) |
86 | return false; |
87 | |
88 | return true; |
89 | } |
90 | |
91 | static inline bool futex_validate_input(unsigned int flags, u64 val) |
92 | { |
93 | int bits = 8 * futex_size(flags); |
94 | |
95 | if (bits < 64 && (val >> bits)) |
96 | return false; |
97 | |
98 | return true; |
99 | } |
100 | |
101 | #ifdef CONFIG_FAIL_FUTEX |
102 | extern bool should_fail_futex(bool fshared); |
103 | #else |
104 | static inline bool should_fail_futex(bool fshared) |
105 | { |
106 | return false; |
107 | } |
108 | #endif |
109 | |
110 | /* |
111 | * Hash buckets are shared by all the futex_keys that hash to the same |
112 | * location. Each key may have multiple futex_q structures, one for each task |
113 | * waiting on a futex. |
114 | */ |
115 | struct futex_hash_bucket { |
116 | atomic_t waiters; |
117 | spinlock_t lock; |
118 | struct plist_head chain; |
119 | } ____cacheline_aligned_in_smp; |
120 | |
121 | /* |
122 | * Priority Inheritance state: |
123 | */ |
124 | struct futex_pi_state { |
125 | /* |
126 | * list of 'owned' pi_state instances - these have to be |
127 | * cleaned up in do_exit() if the task exits prematurely: |
128 | */ |
129 | struct list_head list; |
130 | |
131 | /* |
132 | * The PI object: |
133 | */ |
134 | struct rt_mutex_base pi_mutex; |
135 | |
136 | struct task_struct *owner; |
137 | refcount_t refcount; |
138 | |
139 | union futex_key key; |
140 | } __randomize_layout; |
141 | |
142 | struct futex_q; |
143 | typedef void (futex_wake_fn)(struct wake_q_head *wake_q, struct futex_q *q); |
144 | |
145 | /** |
146 | * struct futex_q - The hashed futex queue entry, one per waiting task |
147 | * @list: priority-sorted list of tasks waiting on this futex |
148 | * @task: the task waiting on the futex |
149 | * @lock_ptr: the hash bucket lock |
150 | * @wake: the wake handler for this queue |
151 | * @wake_data: data associated with the wake handler |
152 | * @key: the key the futex is hashed on |
153 | * @pi_state: optional priority inheritance state |
154 | * @rt_waiter: rt_waiter storage for use with requeue_pi |
155 | * @requeue_pi_key: the requeue_pi target futex key |
156 | * @bitset: bitset for the optional bitmasked wakeup |
157 | * @requeue_state: State field for futex_requeue_pi() |
158 | * @requeue_wait: RCU wait for futex_requeue_pi() (RT only) |
159 | * |
160 | * We use this hashed waitqueue, instead of a normal wait_queue_entry_t, so |
161 | * we can wake only the relevant ones (hashed queues may be shared). |
162 | * |
163 | * A futex_q has a woken state, just like tasks have TASK_RUNNING. |
164 | * It is considered woken when plist_node_empty(&q->list) || q->lock_ptr == 0. |
165 | * The order of wakeup is always to make the first condition true, then |
166 | * the second. |
167 | * |
168 | * PI futexes are typically woken before they are removed from the hash list via |
169 | * the rt_mutex code. See futex_unqueue_pi(). |
170 | */ |
171 | struct futex_q { |
172 | struct plist_node list; |
173 | |
174 | struct task_struct *task; |
175 | spinlock_t *lock_ptr; |
176 | futex_wake_fn *wake; |
177 | void *wake_data; |
178 | union futex_key key; |
179 | struct futex_pi_state *pi_state; |
180 | struct rt_mutex_waiter *rt_waiter; |
181 | union futex_key *requeue_pi_key; |
182 | u32 bitset; |
183 | atomic_t requeue_state; |
184 | #ifdef CONFIG_PREEMPT_RT |
185 | struct rcuwait requeue_wait; |
186 | #endif |
187 | } __randomize_layout; |
188 | |
189 | extern const struct futex_q futex_q_init; |
190 | |
191 | enum futex_access { |
192 | FUTEX_READ, |
193 | FUTEX_WRITE |
194 | }; |
195 | |
196 | extern int get_futex_key(u32 __user *uaddr, unsigned int flags, union futex_key *key, |
197 | enum futex_access rw); |
198 | |
199 | extern struct hrtimer_sleeper * |
200 | futex_setup_timer(ktime_t *time, struct hrtimer_sleeper *timeout, |
201 | int flags, u64 range_ns); |
202 | |
203 | extern struct futex_hash_bucket *futex_hash(union futex_key *key); |
204 | |
205 | /** |
206 | * futex_match - Check whether two futex keys are equal |
207 | * @key1: Pointer to key1 |
208 | * @key2: Pointer to key2 |
209 | * |
210 | * Return 1 if two futex_keys are equal, 0 otherwise. |
211 | */ |
212 | static inline int futex_match(union futex_key *key1, union futex_key *key2) |
213 | { |
214 | return (key1 && key2 |
215 | && key1->both.word == key2->both.word |
216 | && key1->both.ptr == key2->both.ptr |
217 | && key1->both.offset == key2->both.offset); |
218 | } |
219 | |
220 | extern int futex_wait_setup(u32 __user *uaddr, u32 val, unsigned int flags, |
221 | struct futex_q *q, struct futex_hash_bucket **hb); |
222 | extern void futex_wait_queue(struct futex_hash_bucket *hb, struct futex_q *q, |
223 | struct hrtimer_sleeper *timeout); |
224 | extern bool __futex_wake_mark(struct futex_q *q); |
225 | extern void futex_wake_mark(struct wake_q_head *wake_q, struct futex_q *q); |
226 | |
227 | extern int fault_in_user_writeable(u32 __user *uaddr); |
228 | extern int futex_cmpxchg_value_locked(u32 *curval, u32 __user *uaddr, u32 uval, u32 newval); |
229 | extern int futex_get_value_locked(u32 *dest, u32 __user *from); |
230 | extern struct futex_q *futex_top_waiter(struct futex_hash_bucket *hb, union futex_key *key); |
231 | |
232 | extern void __futex_unqueue(struct futex_q *q); |
233 | extern void __futex_queue(struct futex_q *q, struct futex_hash_bucket *hb); |
234 | extern int futex_unqueue(struct futex_q *q); |
235 | |
236 | /** |
237 | * futex_queue() - Enqueue the futex_q on the futex_hash_bucket |
238 | * @q: The futex_q to enqueue |
239 | * @hb: The destination hash bucket |
240 | * |
241 | * The hb->lock must be held by the caller, and is released here. A call to |
242 | * futex_queue() is typically paired with exactly one call to futex_unqueue(). The |
243 | * exceptions involve the PI related operations, which may use futex_unqueue_pi() |
244 | * or nothing if the unqueue is done as part of the wake process and the unqueue |
245 | * state is implicit in the state of woken task (see futex_wait_requeue_pi() for |
246 | * an example). |
247 | */ |
248 | static inline void futex_queue(struct futex_q *q, struct futex_hash_bucket *hb) |
249 | __releases(&hb->lock) |
250 | { |
251 | __futex_queue(q, hb); |
252 | spin_unlock(lock: &hb->lock); |
253 | } |
254 | |
255 | extern void futex_unqueue_pi(struct futex_q *q); |
256 | |
257 | extern void wait_for_owner_exiting(int ret, struct task_struct *exiting); |
258 | |
259 | /* |
260 | * Reflects a new waiter being added to the waitqueue. |
261 | */ |
262 | static inline void futex_hb_waiters_inc(struct futex_hash_bucket *hb) |
263 | { |
264 | #ifdef CONFIG_SMP |
265 | atomic_inc(v: &hb->waiters); |
266 | /* |
267 | * Full barrier (A), see the ordering comment above. |
268 | */ |
269 | smp_mb__after_atomic(); |
270 | #endif |
271 | } |
272 | |
273 | /* |
274 | * Reflects a waiter being removed from the waitqueue by wakeup |
275 | * paths. |
276 | */ |
277 | static inline void futex_hb_waiters_dec(struct futex_hash_bucket *hb) |
278 | { |
279 | #ifdef CONFIG_SMP |
280 | atomic_dec(v: &hb->waiters); |
281 | #endif |
282 | } |
283 | |
284 | static inline int futex_hb_waiters_pending(struct futex_hash_bucket *hb) |
285 | { |
286 | #ifdef CONFIG_SMP |
287 | /* |
288 | * Full barrier (B), see the ordering comment above. |
289 | */ |
290 | smp_mb(); |
291 | return atomic_read(v: &hb->waiters); |
292 | #else |
293 | return 1; |
294 | #endif |
295 | } |
296 | |
297 | extern struct futex_hash_bucket *futex_q_lock(struct futex_q *q); |
298 | extern void futex_q_unlock(struct futex_hash_bucket *hb); |
299 | |
300 | |
301 | extern int futex_lock_pi_atomic(u32 __user *uaddr, struct futex_hash_bucket *hb, |
302 | union futex_key *key, |
303 | struct futex_pi_state **ps, |
304 | struct task_struct *task, |
305 | struct task_struct **exiting, |
306 | int set_waiters); |
307 | |
308 | extern int refill_pi_state_cache(void); |
309 | extern void get_pi_state(struct futex_pi_state *pi_state); |
310 | extern void put_pi_state(struct futex_pi_state *pi_state); |
311 | extern int fixup_pi_owner(u32 __user *uaddr, struct futex_q *q, int locked); |
312 | |
313 | /* |
314 | * Express the locking dependencies for lockdep: |
315 | */ |
316 | static inline void |
317 | double_lock_hb(struct futex_hash_bucket *hb1, struct futex_hash_bucket *hb2) |
318 | { |
319 | if (hb1 > hb2) |
320 | swap(hb1, hb2); |
321 | |
322 | spin_lock(lock: &hb1->lock); |
323 | if (hb1 != hb2) |
324 | spin_lock_nested(&hb2->lock, SINGLE_DEPTH_NESTING); |
325 | } |
326 | |
327 | static inline void |
328 | double_unlock_hb(struct futex_hash_bucket *hb1, struct futex_hash_bucket *hb2) |
329 | { |
330 | spin_unlock(lock: &hb1->lock); |
331 | if (hb1 != hb2) |
332 | spin_unlock(lock: &hb2->lock); |
333 | } |
334 | |
335 | /* syscalls */ |
336 | |
337 | extern int futex_wait_requeue_pi(u32 __user *uaddr, unsigned int flags, u32 |
338 | val, ktime_t *abs_time, u32 bitset, u32 __user |
339 | *uaddr2); |
340 | |
341 | extern int futex_requeue(u32 __user *uaddr1, unsigned int flags1, |
342 | u32 __user *uaddr2, unsigned int flags2, |
343 | int nr_wake, int nr_requeue, |
344 | u32 *cmpval, int requeue_pi); |
345 | |
346 | extern int __futex_wait(u32 __user *uaddr, unsigned int flags, u32 val, |
347 | struct hrtimer_sleeper *to, u32 bitset); |
348 | |
349 | extern int futex_wait(u32 __user *uaddr, unsigned int flags, u32 val, |
350 | ktime_t *abs_time, u32 bitset); |
351 | |
352 | /** |
353 | * struct futex_vector - Auxiliary struct for futex_waitv() |
354 | * @w: Userspace provided data |
355 | * @q: Kernel side data |
356 | * |
357 | * Struct used to build an array with all data need for futex_waitv() |
358 | */ |
359 | struct futex_vector { |
360 | struct futex_waitv w; |
361 | struct futex_q q; |
362 | }; |
363 | |
364 | extern int futex_parse_waitv(struct futex_vector *futexv, |
365 | struct futex_waitv __user *uwaitv, |
366 | unsigned int nr_futexes, futex_wake_fn *wake, |
367 | void *wake_data); |
368 | |
369 | extern int futex_wait_multiple_setup(struct futex_vector *vs, int count, |
370 | int *woken); |
371 | |
372 | extern int futex_unqueue_multiple(struct futex_vector *v, int count); |
373 | |
374 | extern int futex_wait_multiple(struct futex_vector *vs, unsigned int count, |
375 | struct hrtimer_sleeper *to); |
376 | |
377 | extern int futex_wake(u32 __user *uaddr, unsigned int flags, int nr_wake, u32 bitset); |
378 | |
379 | extern int futex_wake_op(u32 __user *uaddr1, unsigned int flags, |
380 | u32 __user *uaddr2, int nr_wake, int nr_wake2, int op); |
381 | |
382 | extern int futex_unlock_pi(u32 __user *uaddr, unsigned int flags); |
383 | |
384 | extern int futex_lock_pi(u32 __user *uaddr, unsigned int flags, ktime_t *time, int trylock); |
385 | |
386 | #endif /* _FUTEX_H */ |
387 | |