.\" .\" Copyright (C) 2019 Mariusz Zaborski .\" .\" Redistribution and use in source and binary forms, with or without .\" modification, are permitted provided that the following conditions .\" are met: .\" 1. Redistributions of source code must retain the above copyright .\" notice(s), this list of conditions and the following disclaimer as .\" the first lines of this file unmodified other than the possible .\" addition of one or more copyright notices. .\" 2. Redistributions in binary form must reproduce the above copyright .\" notice(s), this list of conditions and the following disclaimer in the .\" documentation and/or other materials provided with the distribution. .\" .\" THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDER(S) ``AS IS'' AND ANY .\" EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED .\" WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE .\" DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER(S) BE LIABLE FOR ANY .\" DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES .\" (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR .\" SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER .\" CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT .\" LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY .\" OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH .\" DAMAGE. .\" .Dd July 29, 2019 .Dt SEQC 9 .Os .Sh NAME .Nm seqc_consistent , .Nm seqc_read , .Nm seqc_write_begin , .Nm seqc_write_end .Nd "lockless read algorithm" .Sh SYNOPSIS .In sys/seqc.h .Ft void .Fn seqc_write_begin "seqc_t *seqcp" .Ft void .Fn seqc_write_end "seqc_t *seqcp" .Ft seqc_t .Fn seqc_read "seqc_t *seqcp" .Ft seqc_t .Fn seqc_consistent "const seqc_t *seqcp" "seqc_t oldseqc" .Sh DESCRIPTION The .Nm seqc allows zero or more readers and zero or one writer to concurrently access an object, providing a consistent snapshot of the object for readers. No mutual exclusion between readers and writers is required, but readers may be starved indefinitely by writers. .Pp The functions .Fn seqc_write_begin and .Fn seqc_write_end are used to create a transaction for writer, and notify the readers that the object will be modified. .Pp The .Fn seqc_read function returns the current sequence number. If a writer has started a transaction, this function will spin until the transaction has ended. .Pp The .Fn seqc_consistent function compares the sequence number with a previously fetched value. The .Fa oldseqc variable should contain a sequence number from the beginning of read transaction. .Pp The reader at the end of a transaction checks if the sequence number has changed. If the sequence number didn't change the object wasn't modified, and fetched variables are valid. If the sequence number changed the object was modified and the fetch should be repeated. In case when sequence number is odd the object change is in progress and the reader will wait until the write will the sequence number will become even. .Sh EXAMPLES The following example for a writer changees the .Va var1 and .Va var2 variables in the .Va obj structure: .Bd -literal lock_exclusive(&obj->lock); seqc_write_begin(&obj->seqc); obj->var1 = 1; obj->var2 = 2; seqc_write_end(&obj->seqc); unlock_exclusive(&obj->lock); .Ed .Pp The following example for a reader reads the .Va var1 and .Va var2 variables from the .Va obj structure. In the case where the sequence number was changed it restarts the whole process. .Bd -literal int var1, var2; seqc_t seqc; for (;;) { seqc = seqc_read(&obj->seqc); var1 = obj->var1; var2 = obj->var2; if (seqc_consistent(&obj->seqc, seqc)) break; } .Ed .Sh AUTHORS The .Nm seqc functions was implemented by .An Mateusz Guzik Aq Mt mjg@FreeBSD.org . This manual page was written by .An Mariusz Zaborski Aq Mt oshogbo@FreeBSD.org . .Sh CAVEATS There is no guarantee of progress for readers. In case when there are a lot of writers the reader can be starved. This concern may be solved by returning error after a few attempts. .Pp Theoretically if reading takes a very long time, and when there are many writers the counter may overflow and wrap around to the same value. In that case the reader will not notice that the object was changed. Given that this needs 4 billion transactional writes across a single contended reader, it is unlikely to ever happen. This could be avoided by extending the interface to allow 64-bit counters.