SYNC(1) General Commands Manual Fully-rendered PDF

sync — flush kernel caches

sync [-d|-f] [file]…

Instruct the kernel to commit cached writes to files to their backing storage. If none specified, flush all caches.

, --data
Don't flush metadata.
, --file-system
Flush entire filesystem containing each file. Only availabe under Linux.

fdatasync(2), fsync(2), sync(2), syncfs(2)

Compatible with Version 4 AT&T UNIX; arguments are a GNU extension.

sync(II) appears in the second edition of the UNIX Programmer's Manual as

sync ........................ assure synchronisation
in the index and
on its page, described as causing "the super block for all file systems to be written out", and to only be necessary on systems incorporating "hardware protection facilities", as those may "delay this writing for a long time". It also notes that it "should be used by programs which examine a file system", as those read the underlying block devices directly.

Version 4 AT&T UNIX broadens the description to causing "all information in core memory that should be on disk to be written out", listing "modified super blocks" and i-nodes, and "delayed block I/O". as well as noting it to "be mandatory before a boot", with the new sync(VIII) and update(VIII) in SEE ALSO.

sync(VIII) ("update the super block") is described simply as

executes the sync system primitive
. whereas update(VIII) ("periodically update the super block")
executes the sync primitive every 30 seconds. This insures that the file system is fairly up to date in case of a crash
. nevertheless, its BUGS section states, that
There is a system bug which, it is suspected, may be aggravated by this program. Until further notice, update should not be run
.

Version 5 AT&T UNIX elucidates the synchronisation/insurance relationship in sync(VIII) and expands update(VIII) BUGS:

With update running, if the CPU is halted just as the sync is executed, a file system can be damaged. This is partially due to DEC hardware that writes zeros when NPR requests fail. A fix would be to have sync temporarily increment the system time by at least 30 seconds to trigger the execution of update. This would give 30 seconds grace to halt the CPU.

Version 7 AT&T UNIX sync(2) notes in its BUGS section that

The writing, although scheduled, is not necessarily complete upon return from sync
. sync(VIII) moves to sync(1M) (though sync(2) refers to sync(1)) with no further changes.

update(8) disappears from Programmer's Workbench (PWB/UNIX). Further, AT&T System III UNIX moves the Version 7 AT&T UNIX sync(2) BUGS line into the description.

4.1BSD moves sync(1M) back to sync(8), despite the executable remaining squarely in the global /bin.

The BSD sees a steady evolution of update(8), and, with 4.4BSD, shortening the BUGS to

It is possible on some systems that a sync occuring simultaneously with a crash may cause file system damage. See fsck(8).
(with no useful notes in fsck(8)) and being mentioned in sync(2) as a requirement for a healthy system.

This is not necessarily entirely unlike the user-space flushing daemons (bdflush(8)/bdflushd/updated/bdflush(2)), as found in Linux 1.1.3 through 2.5.12, though with a different scope (bdflush(2) is in some configurations an entire kernel-side flushing daemon) and for a different reason (Linux being single-threaded).

November 23, 2022 voreutils 5a9f9f29