On Thu, Apr 09, 2020 at 09:59:27AM -0700, Darrick J. Wong wrote: [snip] > And today: > > 1. There exists an in-kernel access mode flag S_DAX that is set when > file accesses go directly to persistent memory, bypassing the page > cache. Applications must call statx to discover the current S_DAX > state (STATX_ATTR_DAX). > > 2. There exists an advisory file inode flag FS_XFLAG_DAX that is > inherited from the parent directory FS_XFLAG_DAX inode flag at file > creation time. This advisory flag can be set or cleared at any > time, but doing so does not immediately affect the S_DAX state. > > Unless overridden by mount options (see (3)), if FS_XFLAG_DAX is set > and the fs is on pmem then it will enable S_DAX at inode load time; > if FS_XFLAG_DAX is not set, it will not enable S_DAX. > > 3. There exists a dax= mount option. > > "-o dax=never" means "never set S_DAX, ignore FS_XFLAG_DAX." > > "-o dax=always" means "always set S_DAX (at least on pmem), > and ignore FS_XFLAG_DAX." > > "-o dax" is an alias for "dax=always". > > "-o dax=inode" means "follow FS_XFLAG_DAX" and is the default. > > 4. There exists an advisory directory inode flag FS_XFLAG_DAX that can > be set or cleared at any time. The flag state is copied into any > files or subdirectories when they are created within that directory. > > 5. Programs that require a specific file access mode (DAX or not DAX) > can do one of the following: > > (a) Create files in directories that the FS_XFLAG_DAX flag set as > needed; or > > (b) Have the administrator set an override via mount option; or > > (c) Set or clear the file's FS_XFLAG_DAX flag as needed. Programs > must then cause the kernel to evict the inode from memory. This > can be done by: > > i> Closing the file and re-opening the file and using statx to > see if the fs has changed the S_DAX flag; and > > ii> If the file still does not have the desired S_DAX access > mode, either unmount and remount the filesystem, or close > the file and use drop_caches. > > 6. It's not unreasonable that users who want to squeeze every last bit > of performance out of the particular rough and tumble bits of their > storage also be exposed to the difficulties of what happens when the > operating system can't totally virtualize those hardware > capabilities. Your high performance sports car is not a Toyota > minivan, as it were. > > Given our overnight discussions, I don't think it'll be difficult to > hoist XFS_IDONTCACHE to the VFS so that 5.c.i is enough to change the > S_DAX state if nobody else is using the file. Agreed! One note on implementation, I plan to get XFS_IDONTCACHE tested with XFS and leave hoisting it to VFS for the series which enables ext4 as that is when we would need such hoisting. Thanks everyone for another round of discussions! :-D Ira
