1.\" Copyright (c) 2006 Gleb Smirnoff <glebius@FreeBSD.org> 2.\" All rights reserved. 3.\" 4.\" Redistribution and use in source and binary forms, with or without 5.\" modification, are permitted provided that the following conditions 6.\" are met: 7.\" 1. Redistributions of source code must retain the above copyright 8.\" notice, this list of conditions and the following disclaimer. 9.\" 2. Redistributions in binary form must reproduce the above copyright 10.\" notice, this list of conditions and the following disclaimer in the 11.\" documentation and/or other materials provided with the distribution. 12.\" 13.\" THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 14.\" ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 15.\" IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 16.\" ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 17.\" FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 18.\" DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 19.\" OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 20.\" HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 21.\" LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 22.\" OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 23.\" SUCH DAMAGE. 24.\" 25.\" $FreeBSD$ 26.\" 27.Dd November 16, 2011 28.Dt RWLOCK 9 29.Os 30.Sh NAME 31.Nm rwlock , 32.Nm rw_init , 33.Nm rw_init_flags, 34.Nm rw_destroy , 35.Nm rw_rlock , 36.Nm rw_wlock , 37.Nm rw_runlock , 38.Nm rw_wunlock , 39.Nm rw_unlock , 40.Nm rw_try_rlock , 41.Nm rw_try_upgrade , 42.Nm rw_try_wlock , 43.Nm rw_downgrade , 44.Nm rw_sleep , 45.Nm rw_initialized , 46.Nm rw_wowned , 47.Nm rw_assert , 48.Nm RW_SYSINIT 49.Nd kernel reader/writer lock 50.Sh SYNOPSIS 51.In sys/param.h 52.In sys/lock.h 53.In sys/rwlock.h 54.Ft void 55.Fn rw_init "struct rwlock *rw" "const char *name" 56.Ft void 57.Fn rw_init_flags "struct rwlock *rw" "const char *name" "int opts" 58.Ft void 59.Fn rw_destroy "struct rwlock *rw" 60.Ft void 61.Fn rw_rlock "struct rwlock *rw" 62.Ft void 63.Fn rw_wlock "struct rwlock *rw" 64.Ft int 65.Fn rw_try_rlock "struct rwlock *rw" 66.Ft int 67.Fn rw_try_wlock "struct rwlock *rw" 68.Ft void 69.Fn rw_runlock "struct rwlock *rw" 70.Ft void 71.Fn rw_wunlock "struct rwlock *rw" 72.Ft void 73.Fn rw_unlock "struct rwlock *rw" 74.Ft int 75.Fn rw_try_upgrade "struct rwlock *rw" 76.Ft void 77.Fn rw_downgrade "struct rwlock *rw" 78.Ft int 79.Fn rw_sleep "void *chan" "struct rwlock *rw" "int priority" "const char *wmesg" "int timo" 80.Ft int 81.Fn rw_initialized "const struct rwlock *rw" 82.Ft int 83.Fn rw_wowned "const struct rwlock *rw" 84.Pp 85.Cd "options INVARIANTS" 86.Cd "options INVARIANT_SUPPORT" 87.Ft void 88.Fn rw_assert "const struct rwlock *rw" "int what" 89.In sys/kernel.h 90.Fn RW_SYSINIT "name" "struct rwlock *rw" "const char *desc" 91.Sh DESCRIPTION 92Reader/writer locks allow shared access to protected data by multiple threads, 93or exclusive access by a single thread. 94The threads with shared access are known as 95.Em readers 96since they only read the protected data. 97A thread with exclusive access is known as a 98.Em writer 99since it can modify protected data. 100.Pp 101Although reader/writer locks look very similar to 102.Xr sx 9 103locks, their usage pattern is different. 104Reader/writer locks can be treated as mutexes (see 105.Xr mutex 9 ) 106with shared/exclusive semantics. 107Unlike 108.Xr sx 9 , 109an 110.Nm 111can be locked while holding a non-spin mutex, and an 112.Nm 113cannot be held while sleeping. 114The 115.Nm 116locks have priority propagation like mutexes, but priority 117can be propagated only to writers. 118This limitation comes from the fact that readers 119are anonymous. 120Another important property is that readers can always recurse, 121and exclusive locks can be made recursive selectively. 122.Ss Macros and Functions 123.Bl -tag -width indent 124.It Fn rw_init "struct rwlock *rw" "const char *name" 125Initialize structure located at 126.Fa rw 127as reader/writer lock, described by name 128.Fa name . 129The description is used solely for debugging purposes. 130This function must be called before any other operations 131on the lock. 132.It Fn rw_init_flags "struct rwlock *rw" "const char *name" "int opts" 133Initialize the rw lock just like the 134.Fn rw_init 135function, but specifying a set of optional flags to alter the 136behaviour of 137.Fa rw , 138through the 139.Fa opts 140argument. 141It contains one or more of the following flags: 142.Bl -tag -width ".Dv RW_NOPROFILE" 143.It Dv RW_DUPOK 144Witness should not log messages about duplicate locks being acquired. 145.It Dv RW_NOPROFILE 146Do not profile this lock. 147.It Dv RW_NOWITNESS 148Instruct 149.Xr witness 4 150to ignore this lock. 151.It Dv RW_QUIET 152Do not log any operations for this lock via 153.Xr ktr 4 . 154.It Dv RW_RECURSE 155Allow threads to recursively acquire exclusive locks for 156.Fa rw . 157.El 158.It Fn rw_rlock "struct rwlock *rw" 159Lock 160.Fa rw 161as a reader. 162If any thread holds this lock exclusively, the current thread blocks, 163and its priority is propagated to the exclusive holder. 164The 165.Fn rw_rlock 166function can be called when the thread has already acquired reader 167access on 168.Fa rw . 169This is called 170.Dq "recursing on a lock" . 171.It Fn rw_wlock "struct rwlock *rw" 172Lock 173.Fa rw 174as a writer. 175If there are any shared owners of the lock, the current thread blocks. 176The 177.Fn rw_wlock 178function can be called recursively only if 179.Fa rw 180has been initialized with the 181.Dv RW_RECURSE 182option enabled. 183.It Fn rw_try_rlock "struct rwlock *rw" 184Try to lock 185.Fa rw 186as a reader. 187This function will return true if the operation succeeds, otherwise 0 188will be returned. 189.It Fn rw_try_wlock "struct rwlock *rw" 190Try to lock 191.Fa rw 192as a writer. 193This function will return true if the operation succeeds, otherwise 0 194will be returned. 195.It Fn rw_runlock "struct rwlock *rw" 196This function releases a shared lock previously acquired by 197.Fn rw_rlock . 198.It Fn rw_wunlock "struct rwlock *rw" 199This function releases an exclusive lock previously acquired by 200.Fn rw_wlock . 201.It Fn rw_unlock "struct rwlock *rw" 202This function releases a shared lock previously acquired by 203.Fn rw_rlock 204or an exclusive lock previously acquired by 205.Fn rw_wlock . 206.It Fn rw_try_upgrade "struct rwlock *rw" 207Attempt to upgrade a single shared lock to an exclusive lock. 208The current thread must hold a shared lock of 209.Fa rw . 210This will only succeed if the current thread holds the only shared lock on 211.Fa rw , 212and it only holds a single shared lock. 213If the attempt succeeds 214.Fn rw_try_upgrade 215will return a non-zero value, 216and the current thread will hold an exclusive lock. 217If the attempt fails 218.Fn rw_try_upgrade 219will return zero, 220and the current thread will still hold a shared lock. 221.It Fn rw_downgrade "struct rwlock *rw" 222Convert an exclusive lock into a single shared lock. 223The current thread must hold an exclusive lock of 224.Fa rw . 225.It Fn rw_sleep "void *chan" "struct rwlock *rw" "int priority" "const char *wmesg" "int timo" 226Atomically release 227.Fa rw 228while waiting for an event. 229For more details on the parameters to this function, 230see 231.Xr sleep 9 . 232.It Fn rw_initialized "const struct rwlock *rw" 233This function returns non-zero if 234.Fa rw 235has been initialized, and zero otherwise. 236.It Fn rw_destroy "struct rwlock *rw" 237This functions destroys a lock previously initialized with 238.Fn rw_init . 239The 240.Fa rw 241lock must be unlocked. 242.It Fn rw_wowned "const struct rwlock *rw" 243This function returns a non-zero value if the current thread owns an 244exclusive lock on 245.Fa rw . 246.It Fn rw_assert "const struct rwlock *rw" "int what" 247This function allows assertions specified in 248.Fa what 249to be made about 250.Fa rw . 251If the assertions are not true and the kernel is compiled 252with 253.Cd "options INVARIANTS" 254and 255.Cd "options INVARIANT_SUPPORT" , 256the kernel will panic. 257Currently the following assertions are supported: 258.Bl -tag -width ".Dv RA_UNLOCKED" 259.It Dv RA_LOCKED 260Assert that current thread holds either a shared or exclusive lock 261of 262.Fa rw . 263.It Dv RA_RLOCKED 264Assert that current thread holds a shared lock of 265.Fa rw . 266.It Dv RA_WLOCKED 267Assert that current thread holds an exclusive lock of 268.Fa rw . 269.It Dv RA_UNLOCKED 270Assert that current thread holds neither a shared nor exclusive lock of 271.Fa rw . 272.El 273.El 274.Sh SEE ALSO 275.Xr locking 9 , 276.Xr mutex 9 , 277.Xr panic 9 , 278.Xr sema 9 , 279.Xr sx 9 280.Sh HISTORY 281These 282functions appeared in 283.Fx 7.0 . 284.Sh AUTHORS 285.An -nosplit 286The 287.Nm 288facility was written by 289.An "John Baldwin" . 290This manual page was written by 291.An "Gleb Smirnoff" . 292.Sh BUGS 293If 294.Dv WITNESS 295is not included in the kernel, 296then it is impossible to assert that the current thread does or does not 297hold a read lock. 298In the 299.Pf non- Dv WITNESS 300case, the 301.Dv RA_LOCKED 302and 303.Dv RA_RLOCKED 304assertions merely check that some thread holds a read lock. 305.Pp 306Reader/writer is a bit of an awkward name. 307An 308.Nm 309can also be called a 310.Dq Robert Watson 311lock if desired. 312