1.\" Copyright (c) 2007 Stephan Uphoff <ups@FreeBSD.org> 2.\" Copyright (c) 2006 Gleb Smirnoff <glebius@FreeBSD.org> 3.\" All rights reserved. 4.\" 5.\" Redistribution and use in source and binary forms, with or without 6.\" modification, are permitted provided that the following conditions 7.\" are met: 8.\" 1. Redistributions of source code must retain the above copyright 9.\" notice, this list of conditions and the following disclaimer. 10.\" 2. Redistributions in binary form must reproduce the above copyright 11.\" notice, this list of conditions and the following disclaimer in the 12.\" documentation and/or other materials provided with the distribution. 13.\" 14.\" THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 15.\" ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 16.\" IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 17.\" ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 18.\" FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 19.\" DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 20.\" OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 21.\" HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 22.\" LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 23.\" OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 24.\" SUCH DAMAGE. 25.\" 26.\" $FreeBSD$ 27.\" 28.\" Based on rwlock.9 man page 29.Dd December 13, 2014 30.Dt RMLOCK 9 31.Os 32.Sh NAME 33.Nm rmlock , 34.Nm rm_init , 35.Nm rm_init_flags , 36.Nm rm_destroy , 37.Nm rm_rlock , 38.Nm rm_try_rlock , 39.Nm rm_wlock , 40.Nm rm_runlock , 41.Nm rm_wunlock , 42.Nm rm_wowned , 43.Nm rm_sleep , 44.Nm rm_assert , 45.Nm RM_SYSINIT 46.Nd kernel reader/writer lock optimized for read-mostly access patterns 47.Sh SYNOPSIS 48.In sys/param.h 49.In sys/lock.h 50.In sys/rmlock.h 51.Ft void 52.Fn rm_init "struct rmlock *rm" "const char *name" 53.Ft void 54.Fn rm_init_flags "struct rmlock *rm" "const char *name" "int opts" 55.Ft void 56.Fn rm_destroy "struct rmlock *rm" 57.Ft void 58.Fn rm_rlock "struct rmlock *rm" "struct rm_priotracker* tracker" 59.Ft int 60.Fn rm_try_rlock "struct rmlock *rm" "struct rm_priotracker* tracker" 61.Ft void 62.Fn rm_wlock "struct rmlock *rm" 63.Ft void 64.Fn rm_runlock "struct rmlock *rm" "struct rm_priotracker* tracker" 65.Ft void 66.Fn rm_wunlock "struct rmlock *rm" 67.Ft int 68.Fn rm_wowned "const struct rmlock *rm" 69.Ft int 70.Fn rm_sleep "void *wchan" "struct rmlock *rm" "int priority" "const char *wmesg" "int timo" 71.Pp 72.Cd "options INVARIANTS" 73.Cd "options INVARIANT_SUPPORT" 74.Ft void 75.Fn rm_assert "struct rmlock *rm" "int what" 76.In sys/kernel.h 77.Fn RM_SYSINIT "name" "struct rmlock *rm" "const char *desc" "int opts" 78.Sh DESCRIPTION 79Read-mostly locks allow shared access to protected data by multiple threads, 80or exclusive access by a single thread. 81The threads with shared access are known as 82.Em readers 83since they only read the protected data. 84A thread with exclusive access is known as a 85.Em writer 86since it can modify protected data. 87.Pp 88Read-mostly locks are designed to be efficient for locks almost exclusively 89used as reader locks and as such should be used for protecting data that 90rarely changes. 91Acquiring an exclusive lock after the lock has been locked for shared access 92is an expensive operation. 93.Pp 94Normal read-mostly locks are similar to 95.Xr rwlock 9 96locks and follow the same lock ordering rules as 97.Xr rwlock 9 98locks. 99Read-mostly locks have full priority propagation like mutexes. 100Unlike 101.Xr rwlock 9 , 102read-mostly locks propagate priority to both readers and writers. 103This is implemented via the 104.Va rm_priotracker 105structure argument supplied to 106.Fn rm_rlock 107and 108.Fn rm_runlock . 109Readers can recurse if the lock is initialized with the 110.Dv RM_RECURSE 111option; 112however, writers are never allowed to recurse. 113.Pp 114Sleepable read-mostly locks are created by passing 115.Dv RM_SLEEPABLE 116to 117.Fn rm_init_flags . 118Unlike normal read-mostly locks, 119sleepable read-mostly locks follow the same lock ordering rules as 120.Xr sx 9 121locks. 122Sleepable read-mostly locks do not propagate priority to writers, 123but they do propagate priority to readers. 124Writers are permitted to sleep while holding a read-mostly lock, 125but readers are not. 126Unlike other sleepable locks such as 127.Xr sx 9 128locks, 129readers must use try operations on other sleepable locks to avoid sleeping. 130.Ss Macros and Functions 131.Bl -tag -width indent 132.It Fn rm_init "struct rmlock *rm" "const char *name" 133Initialize the read-mostly lock 134.Fa rm . 135The 136.Fa name 137description is used solely for debugging purposes. 138This function must be called before any other operations 139on the lock. 140.It Fn rm_init_flags "struct rmlock *rm" "const char *name" "int opts" 141Similar to 142.Fn rm_init , 143initialize the read-mostly lock 144.Fa rm 145with a set of optional flags. 146The 147.Fa opts 148arguments contains one or more of the following flags: 149.Bl -tag -width ".Dv RM_NOWITNESS" 150.It Dv RM_NOWITNESS 151Instruct 152.Xr witness 4 153to ignore this lock. 154.It Dv RM_RECURSE 155Allow threads to recursively acquire shared locks for 156.Fa rm . 157.It Dv RM_SLEEPABLE 158Create a sleepable read-mostly lock. 159.It Dv RM_NEW 160If the kernel has been compiled with 161.Cd "option INVARIANTS" , 162.Fn rm_init_flags 163will assert that the 164.Fa rm 165has not been initialized multiple times without intervening calls to 166.Fn rm_destroy 167unless this option is specified. 168.El 169.It Fn rm_rlock "struct rmlock *rm" "struct rm_priotracker* tracker" 170Lock 171.Fa rm 172as a reader using 173.Fa tracker 174to track read owners of a lock for priority propagation. 175This data structure is only used internally by 176.Nm 177and must persist until 178.Fn rm_runlock 179has been called. 180This data structure can be allocated on the stack since 181readers cannot sleep. 182If any thread holds this lock exclusively, the current thread blocks, 183and its priority is propagated to the exclusive holder. 184If the lock was initialized with the 185.Dv RM_RECURSE 186option the 187.Fn rm_rlock 188function can be called when the current thread has already acquired reader 189access on 190.Fa rm . 191.It Fn rm_try_rlock "struct rmlock *rm" "struct rm_priotracker* tracker" 192Try to lock 193.Fa rm 194as a reader. 195.Fn rm_try_rlock 196will return 0 if the lock cannot be acquired immediately; 197otherwise, 198the lock will be acquired and a non-zero value will be returned. 199Note that 200.Fn rm_try_rlock 201may fail even while the lock is not currently held by a writer. 202If the lock was initialized with the 203.Dv RM_RECURSE 204option, 205.Fn rm_try_rlock 206will succeed if the current thread has already acquired reader access. 207.It Fn rm_wlock "struct rmlock *rm" 208Lock 209.Fa rm 210as a writer. 211If there are any shared owners of the lock, the current thread blocks. 212The 213.Fn rm_wlock 214function cannot be called recursively. 215.It Fn rm_runlock "struct rmlock *rm" "struct rm_priotracker* tracker" 216This function releases a shared lock previously acquired by 217.Fn rm_rlock . 218The 219.Fa tracker 220argument must match the 221.Fa tracker 222argument used for acquiring the shared lock 223.It Fn rm_wunlock "struct rmlock *rm" 224This function releases an exclusive lock previously acquired by 225.Fn rm_wlock . 226.It Fn rm_destroy "struct rmlock *rm" 227This functions destroys a lock previously initialized with 228.Fn rm_init . 229The 230.Fa rm 231lock must be unlocked. 232.It Fn rm_wowned "const struct rmlock *rm" 233This function returns a non-zero value if the current thread owns an 234exclusive lock on 235.Fa rm . 236.It Fn rm_sleep "void *wchan" "struct rmlock *rm" "int priority" "const char *wmesg" "int timo" 237This function atomically releases 238.Fa rm 239while waiting for an event. 240The 241.Fa rm 242lock must be exclusively locked. 243For more details on the parameters to this function, 244see 245.Xr sleep 9 . 246.It Fn rm_assert "struct rmlock *rm" "int what" 247This function asserts that the 248.Fa rm 249lock is in the state specified by 250.Fa what . 251If the assertions are not true and the kernel is compiled with 252.Cd "options INVARIANTS" 253and 254.Cd "options INVARIANT_SUPPORT" , 255the kernel will panic. 256Currently the following base assertions are supported: 257.Bl -tag -width ".Dv RA_UNLOCKED" 258.It Dv RA_LOCKED 259Assert that current thread holds either a shared or exclusive lock 260of 261.Fa rm . 262.It Dv RA_RLOCKED 263Assert that current thread holds a shared lock of 264.Fa rm . 265.It Dv RA_WLOCKED 266Assert that current thread holds an exclusive lock of 267.Fa rm . 268.It Dv RA_UNLOCKED 269Assert that current thread holds neither a shared nor exclusive lock of 270.Fa rm . 271.El 272.Pp 273In addition, one of the following optional flags may be specified with 274.Dv RA_LOCKED , 275.Dv RA_RLOCKED , 276or 277.Dv RA_WLOCKED : 278.Bl -tag -width ".Dv RA_NOTRECURSED" 279.It Dv RA_RECURSED 280Assert that the current thread holds a recursive lock of 281.Fa rm . 282.It Dv RA_NOTRECURSED 283Assert that the current thread does not hold a recursive lock of 284.Fa rm . 285.El 286.El 287.Sh SEE ALSO 288.Xr locking 9 , 289.Xr mutex 9 , 290.Xr panic 9 , 291.Xr rwlock 9 , 292.Xr sema 9 , 293.Xr sleep 9 , 294.Xr sx 9 295.Sh HISTORY 296These 297functions appeared in 298.Fx 7.0 . 299.Sh AUTHORS 300.An -nosplit 301The 302.Nm 303facility was written by 304.An "Stephan Uphoff" . 305This manual page was written by 306.An "Gleb Smirnoff" 307for rwlock and modified to reflect rmlock by 308.An "Stephan Uphoff" . 309.Sh BUGS 310The 311.Nm 312implementation is currently not optimized for single processor systems. 313.Pp 314.Fn rm_try_rlock 315can fail transiently even when there is no writer, while another reader 316updates the state on the local CPU. 317.Pp 318The 319.Nm 320implementation uses a single per CPU list shared by all 321rmlocks in the system. 322If rmlocks become popular, hashing to multiple per CPU queues may 323be needed to speed up the writer lock process. 324