Brian Nitz wrote: > Darren Kenny wrote: >> >> o File Monitoring >> + In OpenSolaris, there is a kernel API called File >> Event Mechanism. Unfortunately this still doesn't have >> a userland API that could be used by applications or >> libraries. Linux and AIX both have inotify to do this, >> but OpenSolaris doesn't. This would be an interesting >> project to get done and have libfam modified to >> support OpenSolaris using this mechanism rather than >> polling > Shouldn't it be possible to emulate linux and/or AIX's file event > mechanism via libdtrace? > _______________________________________________ > desktop-discuss mailing list > desktop-discuss at opensolaris.org
It isn't actually a question of mechanism, but what user semantic one would like to provide. If you watch a file for writes, does that trip when the file is unlinked and replaced with another like editors do when they re-write a file? What about mmapped files? Should we map writes via an mmapped buffer to a write via an FD? Personally, I'm leaning in the mtime/atime direction; if a file's mtime changes, we should detect that as a write; if a files atime changes, we should consider that as a read. This way programs that monitor via stat today will "just work", - Bart -- Bart Smaalders Solaris Kernel Performance barts at cyber.eng.sun.com http://blogs.sun.com/barts
