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 #include <linux/uaccess.h> 10 #include <linux/cleanup.h> 11 12 #ifdef CONFIG_PREEMPT_RT 13 #include <linux/rcuwait.h> 14 #endif 15 16 #include <asm/futex.h> 17 18 /* 19 * Futex flags used to encode options to functions and preserve them across 20 * restarts. 21 */ 22 #define FLAGS_SIZE_8 0x0000 23 #define FLAGS_SIZE_16 0x0001 24 #define FLAGS_SIZE_32 0x0002 25 #define FLAGS_SIZE_64 0x0003 26 27 #define FLAGS_SIZE_MASK 0x0003 28 29 #ifdef CONFIG_MMU 30 # define FLAGS_SHARED 0x0010 31 #else 32 /* 33 * NOMMU does not have per process address space. Let the compiler optimize 34 * code away. 35 */ 36 # define FLAGS_SHARED 0x0000 37 #endif 38 #define FLAGS_CLOCKRT 0x0020 39 #define FLAGS_HAS_TIMEOUT 0x0040 40 #define FLAGS_NUMA 0x0080 41 #define FLAGS_STRICT 0x0100 42 #define FLAGS_MPOL 0x0200 43 44 /* FUTEX_ to FLAGS_ */ 45 static inline unsigned int futex_to_flags(unsigned int op) 46 { 47 unsigned int flags = FLAGS_SIZE_32; 48 49 if (!(op & FUTEX_PRIVATE_FLAG)) 50 flags |= FLAGS_SHARED; 51 52 if (op & FUTEX_CLOCK_REALTIME) 53 flags |= FLAGS_CLOCKRT; 54 55 return flags; 56 } 57 58 #define FUTEX2_VALID_MASK (FUTEX2_SIZE_MASK | FUTEX2_NUMA | FUTEX2_MPOL | FUTEX2_PRIVATE) 59 60 /* FUTEX2_ to FLAGS_ */ 61 static inline unsigned int futex2_to_flags(unsigned int flags2) 62 { 63 unsigned int flags = flags2 & FUTEX2_SIZE_MASK; 64 65 if (!(flags2 & FUTEX2_PRIVATE)) 66 flags |= FLAGS_SHARED; 67 68 if (flags2 & FUTEX2_NUMA) 69 flags |= FLAGS_NUMA; 70 71 if (flags2 & FUTEX2_MPOL) 72 flags |= FLAGS_MPOL; 73 74 return flags; 75 } 76 77 static inline unsigned int futex_size(unsigned int flags) 78 { 79 return 1 << (flags & FLAGS_SIZE_MASK); 80 } 81 82 static inline bool futex_flags_valid(unsigned int flags) 83 { 84 /* Only 64bit futexes for 64bit code */ 85 if (!IS_ENABLED(CONFIG_64BIT) || in_compat_syscall()) { 86 if ((flags & FLAGS_SIZE_MASK) == FLAGS_SIZE_64) 87 return false; 88 } 89 90 /* Only 32bit futexes are implemented -- for now */ 91 if ((flags & FLAGS_SIZE_MASK) != FLAGS_SIZE_32) 92 return false; 93 94 /* 95 * Must be able to represent both FUTEX_NO_NODE and every valid nodeid 96 * in a futex word. 97 */ 98 if (flags & FLAGS_NUMA) { 99 int bits = 8 * futex_size(flags); 100 u64 max = ~0ULL; 101 102 max >>= 64 - bits; 103 if (nr_node_ids >= max) 104 return false; 105 } 106 107 return true; 108 } 109 110 static inline bool futex_validate_input(unsigned int flags, u64 val) 111 { 112 int bits = 8 * futex_size(flags); 113 114 if (bits < 64 && (val >> bits)) 115 return false; 116 117 return true; 118 } 119 120 #ifdef CONFIG_FAIL_FUTEX 121 extern bool should_fail_futex(bool fshared); 122 #else 123 static inline bool should_fail_futex(bool fshared) 124 { 125 return false; 126 } 127 #endif 128 129 /* 130 * Hash buckets are shared by all the futex_keys that hash to the same 131 * location. Each key may have multiple futex_q structures, one for each task 132 * waiting on a futex. 133 */ 134 struct futex_hash_bucket { 135 atomic_t waiters; 136 spinlock_t lock; 137 struct plist_head chain; 138 struct futex_private_hash *priv; 139 } ____cacheline_aligned_in_smp; 140 141 /* 142 * Priority Inheritance state: 143 */ 144 struct futex_pi_state { 145 /* 146 * list of 'owned' pi_state instances - these have to be 147 * cleaned up in do_exit() if the task exits prematurely: 148 */ 149 struct list_head list; 150 151 /* 152 * The PI object: 153 */ 154 struct rt_mutex_base pi_mutex; 155 156 struct task_struct *owner; 157 refcount_t refcount; 158 159 union futex_key key; 160 } __randomize_layout; 161 162 struct futex_q; 163 typedef void (futex_wake_fn)(struct wake_q_head *wake_q, struct futex_q *q); 164 165 /** 166 * struct futex_q - The hashed futex queue entry, one per waiting task 167 * @list: priority-sorted list of tasks waiting on this futex 168 * @task: the task waiting on the futex 169 * @lock_ptr: the hash bucket lock 170 * @wake: the wake handler for this queue 171 * @wake_data: data associated with the wake handler 172 * @key: the key the futex is hashed on 173 * @pi_state: optional priority inheritance state 174 * @rt_waiter: rt_waiter storage for use with requeue_pi 175 * @requeue_pi_key: the requeue_pi target futex key 176 * @bitset: bitset for the optional bitmasked wakeup 177 * @requeue_state: State field for futex_requeue_pi() 178 * @requeue_wait: RCU wait for futex_requeue_pi() (RT only) 179 * 180 * We use this hashed waitqueue, instead of a normal wait_queue_entry_t, so 181 * we can wake only the relevant ones (hashed queues may be shared). 182 * 183 * A futex_q has a woken state, just like tasks have TASK_RUNNING. 184 * It is considered woken when plist_node_empty(&q->list) || q->lock_ptr == 0. 185 * The order of wakeup is always to make the first condition true, then 186 * the second. 187 * 188 * PI futexes are typically woken before they are removed from the hash list via 189 * the rt_mutex code. See futex_unqueue_pi(). 190 */ 191 struct futex_q { 192 struct plist_node list; 193 194 struct task_struct *task; 195 spinlock_t *lock_ptr; 196 futex_wake_fn *wake; 197 void *wake_data; 198 union futex_key key; 199 struct futex_pi_state *pi_state; 200 struct rt_mutex_waiter *rt_waiter; 201 union futex_key *requeue_pi_key; 202 u32 bitset; 203 atomic_t requeue_state; 204 bool drop_hb_ref; 205 #ifdef CONFIG_PREEMPT_RT 206 struct rcuwait requeue_wait; 207 #endif 208 } __randomize_layout; 209 210 extern const struct futex_q futex_q_init; 211 212 enum futex_access { 213 FUTEX_READ, 214 FUTEX_WRITE 215 }; 216 217 extern int get_futex_key(u32 __user *uaddr, unsigned int flags, union futex_key *key, 218 enum futex_access rw); 219 extern void futex_q_lockptr_lock(struct futex_q *q); 220 extern struct hrtimer_sleeper * 221 futex_setup_timer(ktime_t *time, struct hrtimer_sleeper *timeout, 222 int flags, u64 range_ns); 223 224 extern struct futex_hash_bucket *futex_hash(union futex_key *key); 225 #ifdef CONFIG_FUTEX_PRIVATE_HASH 226 extern void futex_hash_get(struct futex_hash_bucket *hb); 227 extern void futex_hash_put(struct futex_hash_bucket *hb); 228 229 extern struct futex_private_hash *futex_private_hash(void); 230 extern bool futex_private_hash_get(struct futex_private_hash *fph); 231 extern void futex_private_hash_put(struct futex_private_hash *fph); 232 233 #else /* !CONFIG_FUTEX_PRIVATE_HASH */ 234 static inline void futex_hash_get(struct futex_hash_bucket *hb) { } 235 static inline void futex_hash_put(struct futex_hash_bucket *hb) { } 236 static inline struct futex_private_hash *futex_private_hash(void) { return NULL; } 237 static inline bool futex_private_hash_get(void) { return false; } 238 static inline void futex_private_hash_put(struct futex_private_hash *fph) { } 239 #endif 240 241 DEFINE_CLASS(hb, struct futex_hash_bucket *, 242 if (_T) futex_hash_put(_T), 243 futex_hash(key), union futex_key *key); 244 245 DEFINE_CLASS(private_hash, struct futex_private_hash *, 246 if (_T) futex_private_hash_put(_T), 247 futex_private_hash(), void); 248 249 /** 250 * futex_match - Check whether two futex keys are equal 251 * @key1: Pointer to key1 252 * @key2: Pointer to key2 253 * 254 * Return 1 if two futex_keys are equal, 0 otherwise. 255 */ 256 static inline int futex_match(union futex_key *key1, union futex_key *key2) 257 { 258 return (key1 && key2 259 && key1->both.word == key2->both.word 260 && key1->both.ptr == key2->both.ptr 261 && key1->both.offset == key2->both.offset); 262 } 263 264 extern int futex_wait_setup(u32 __user *uaddr, u32 val, unsigned int flags, 265 struct futex_q *q, union futex_key *key2, 266 struct task_struct *task); 267 extern void futex_do_wait(struct futex_q *q, struct hrtimer_sleeper *timeout); 268 extern bool __futex_wake_mark(struct futex_q *q); 269 extern void futex_wake_mark(struct wake_q_head *wake_q, struct futex_q *q); 270 271 extern int fault_in_user_writeable(u32 __user *uaddr); 272 extern struct futex_q *futex_top_waiter(struct futex_hash_bucket *hb, union futex_key *key); 273 274 static inline int futex_cmpxchg_value_locked(u32 *curval, u32 __user *uaddr, u32 uval, u32 newval) 275 { 276 int ret; 277 278 pagefault_disable(); 279 ret = futex_atomic_cmpxchg_inatomic(curval, uaddr, uval, newval); 280 pagefault_enable(); 281 282 return ret; 283 } 284 285 /* 286 * This does a plain atomic user space read, and the user pointer has 287 * already been verified earlier by get_futex_key() to be both aligned 288 * and actually in user space, just like futex_atomic_cmpxchg_inatomic(). 289 * 290 * We still want to avoid any speculation, and while __get_user() is 291 * the traditional model for this, it's actually slower than doing 292 * this manually these days. 293 * 294 * We could just have a per-architecture special function for it, 295 * the same way we do futex_atomic_cmpxchg_inatomic(), but rather 296 * than force everybody to do that, write it out long-hand using 297 * the low-level user-access infrastructure. 298 * 299 * This looks a bit overkill, but generally just results in a couple 300 * of instructions. 301 */ 302 static __always_inline int futex_get_value(u32 *dest, u32 __user *from) 303 { 304 u32 val; 305 306 if (can_do_masked_user_access()) 307 from = masked_user_access_begin(from); 308 else if (!user_read_access_begin(from, sizeof(*from))) 309 return -EFAULT; 310 unsafe_get_user(val, from, Efault); 311 user_read_access_end(); 312 *dest = val; 313 return 0; 314 Efault: 315 user_read_access_end(); 316 return -EFAULT; 317 } 318 319 static __always_inline int futex_put_value(u32 val, u32 __user *to) 320 { 321 if (can_do_masked_user_access()) 322 to = masked_user_access_begin(to); 323 else if (!user_read_access_begin(to, sizeof(*to))) 324 return -EFAULT; 325 unsafe_put_user(val, to, Efault); 326 user_read_access_end(); 327 return 0; 328 Efault: 329 user_read_access_end(); 330 return -EFAULT; 331 } 332 333 static inline int futex_get_value_locked(u32 *dest, u32 __user *from) 334 { 335 int ret; 336 337 pagefault_disable(); 338 ret = futex_get_value(dest, from); 339 pagefault_enable(); 340 341 return ret; 342 } 343 344 extern void __futex_unqueue(struct futex_q *q); 345 extern void __futex_queue(struct futex_q *q, struct futex_hash_bucket *hb, 346 struct task_struct *task); 347 extern int futex_unqueue(struct futex_q *q); 348 349 /** 350 * futex_queue() - Enqueue the futex_q on the futex_hash_bucket 351 * @q: The futex_q to enqueue 352 * @hb: The destination hash bucket 353 * @task: Task queueing this futex 354 * 355 * The hb->lock must be held by the caller, and is released here. A call to 356 * futex_queue() is typically paired with exactly one call to futex_unqueue(). The 357 * exceptions involve the PI related operations, which may use futex_unqueue_pi() 358 * or nothing if the unqueue is done as part of the wake process and the unqueue 359 * state is implicit in the state of woken task (see futex_wait_requeue_pi() for 360 * an example). 361 * 362 * Note that @task may be NULL, for async usage of futexes. 363 */ 364 static inline void futex_queue(struct futex_q *q, struct futex_hash_bucket *hb, 365 struct task_struct *task) 366 __releases(&hb->lock) 367 { 368 __futex_queue(q, hb, task); 369 spin_unlock(&hb->lock); 370 } 371 372 extern void futex_unqueue_pi(struct futex_q *q); 373 374 extern void wait_for_owner_exiting(int ret, struct task_struct *exiting); 375 376 /* 377 * Reflects a new waiter being added to the waitqueue. 378 */ 379 static inline void futex_hb_waiters_inc(struct futex_hash_bucket *hb) 380 { 381 #ifdef CONFIG_SMP 382 atomic_inc(&hb->waiters); 383 /* 384 * Full barrier (A), see the ordering comment above. 385 */ 386 smp_mb__after_atomic(); 387 #endif 388 } 389 390 /* 391 * Reflects a waiter being removed from the waitqueue by wakeup 392 * paths. 393 */ 394 static inline void futex_hb_waiters_dec(struct futex_hash_bucket *hb) 395 { 396 #ifdef CONFIG_SMP 397 atomic_dec(&hb->waiters); 398 #endif 399 } 400 401 static inline int futex_hb_waiters_pending(struct futex_hash_bucket *hb) 402 { 403 #ifdef CONFIG_SMP 404 /* 405 * Full barrier (B), see the ordering comment above. 406 */ 407 smp_mb(); 408 return atomic_read(&hb->waiters); 409 #else 410 return 1; 411 #endif 412 } 413 414 extern void futex_q_lock(struct futex_q *q, struct futex_hash_bucket *hb); 415 extern void futex_q_unlock(struct futex_hash_bucket *hb); 416 417 418 extern int futex_lock_pi_atomic(u32 __user *uaddr, struct futex_hash_bucket *hb, 419 union futex_key *key, 420 struct futex_pi_state **ps, 421 struct task_struct *task, 422 struct task_struct **exiting, 423 int set_waiters); 424 425 extern int refill_pi_state_cache(void); 426 extern void get_pi_state(struct futex_pi_state *pi_state); 427 extern void put_pi_state(struct futex_pi_state *pi_state); 428 extern int fixup_pi_owner(u32 __user *uaddr, struct futex_q *q, int locked); 429 430 /* 431 * Express the locking dependencies for lockdep: 432 */ 433 static inline void 434 double_lock_hb(struct futex_hash_bucket *hb1, struct futex_hash_bucket *hb2) 435 { 436 if (hb1 > hb2) 437 swap(hb1, hb2); 438 439 spin_lock(&hb1->lock); 440 if (hb1 != hb2) 441 spin_lock_nested(&hb2->lock, SINGLE_DEPTH_NESTING); 442 } 443 444 static inline void 445 double_unlock_hb(struct futex_hash_bucket *hb1, struct futex_hash_bucket *hb2) 446 { 447 spin_unlock(&hb1->lock); 448 if (hb1 != hb2) 449 spin_unlock(&hb2->lock); 450 } 451 452 /* syscalls */ 453 454 extern int futex_wait_requeue_pi(u32 __user *uaddr, unsigned int flags, u32 455 val, ktime_t *abs_time, u32 bitset, u32 __user 456 *uaddr2); 457 458 extern int futex_requeue(u32 __user *uaddr1, unsigned int flags1, 459 u32 __user *uaddr2, unsigned int flags2, 460 int nr_wake, int nr_requeue, 461 u32 *cmpval, int requeue_pi); 462 463 extern int __futex_wait(u32 __user *uaddr, unsigned int flags, u32 val, 464 struct hrtimer_sleeper *to, u32 bitset); 465 466 extern int futex_wait(u32 __user *uaddr, unsigned int flags, u32 val, 467 ktime_t *abs_time, u32 bitset); 468 469 /** 470 * struct futex_vector - Auxiliary struct for futex_waitv() 471 * @w: Userspace provided data 472 * @q: Kernel side data 473 * 474 * Struct used to build an array with all data need for futex_waitv() 475 */ 476 struct futex_vector { 477 struct futex_waitv w; 478 struct futex_q q; 479 }; 480 481 extern int futex_parse_waitv(struct futex_vector *futexv, 482 struct futex_waitv __user *uwaitv, 483 unsigned int nr_futexes, futex_wake_fn *wake, 484 void *wake_data); 485 486 extern int futex_wait_multiple_setup(struct futex_vector *vs, int count, 487 int *woken); 488 489 extern int futex_unqueue_multiple(struct futex_vector *v, int count); 490 491 extern int futex_wait_multiple(struct futex_vector *vs, unsigned int count, 492 struct hrtimer_sleeper *to); 493 494 extern int futex_wake(u32 __user *uaddr, unsigned int flags, int nr_wake, u32 bitset); 495 496 extern int futex_wake_op(u32 __user *uaddr1, unsigned int flags, 497 u32 __user *uaddr2, int nr_wake, int nr_wake2, int op); 498 499 extern int futex_unlock_pi(u32 __user *uaddr, unsigned int flags); 500 501 extern int futex_lock_pi(u32 __user *uaddr, unsigned int flags, ktime_t *time, int trylock); 502 503 #endif /* _FUTEX_H */ 504