cachepc-linux

Fork of AMDESE/linux with modifications for CachePC side-channel attack
git clone https://git.sinitax.com/sinitax/cachepc-linux
Log | Files | Refs | README | LICENSE | sfeed.txt

pnfs.rst (3021B)


      1==========================
      2Reference counting in pnfs
      3==========================
      4
      5The are several inter-related caches.  We have layouts which can
      6reference multiple devices, each of which can reference multiple data servers.
      7Each data server can be referenced by multiple devices.  Each device
      8can be referenced by multiple layouts. To keep all of this straight,
      9we need to reference count.
     10
     11
     12struct pnfs_layout_hdr
     13======================
     14
     15The on-the-wire command LAYOUTGET corresponds to struct
     16pnfs_layout_segment, usually referred to by the variable name lseg.
     17Each nfs_inode may hold a pointer to a cache of these layout
     18segments in nfsi->layout, of type struct pnfs_layout_hdr.
     19
     20We reference the header for the inode pointing to it, across each
     21outstanding RPC call that references it (LAYOUTGET, LAYOUTRETURN,
     22LAYOUTCOMMIT), and for each lseg held within.
     23
     24Each header is also (when non-empty) put on a list associated with
     25struct nfs_client (cl_layouts).  Being put on this list does not bump
     26the reference count, as the layout is kept around by the lseg that
     27keeps it in the list.
     28
     29deviceid_cache
     30==============
     31
     32lsegs reference device ids, which are resolved per nfs_client and
     33layout driver type.  The device ids are held in a RCU cache (struct
     34nfs4_deviceid_cache).  The cache itself is referenced across each
     35mount.  The entries (struct nfs4_deviceid) themselves are held across
     36the lifetime of each lseg referencing them.
     37
     38RCU is used because the deviceid is basically a write once, read many
     39data structure.  The hlist size of 32 buckets needs better
     40justification, but seems reasonable given that we can have multiple
     41deviceid's per filesystem, and multiple filesystems per nfs_client.
     42
     43The hash code is copied from the nfsd code base.  A discussion of
     44hashing and variations of this algorithm can be found `here.
     45<http://groups.google.com/group/comp.lang.c/browse_thread/thread/9522965e2b8d3809>`_
     46
     47data server cache
     48=================
     49
     50file driver devices refer to data servers, which are kept in a module
     51level cache.  Its reference is held over the lifetime of the deviceid
     52pointing to it.
     53
     54lseg
     55====
     56
     57lseg maintains an extra reference corresponding to the NFS_LSEG_VALID
     58bit which holds it in the pnfs_layout_hdr's list.  When the final lseg
     59is removed from the pnfs_layout_hdr's list, the NFS_LAYOUT_DESTROYED
     60bit is set, preventing any new lsegs from being added.
     61
     62layout drivers
     63==============
     64
     65PNFS utilizes what is called layout drivers. The STD defines 4 basic
     66layout types: "files", "objects", "blocks", and "flexfiles". For each
     67of these types there is a layout-driver with a common function-vectors
     68table which are called by the nfs-client pnfs-core to implement the
     69different layout types.
     70
     71Files-layout-driver code is in: fs/nfs/filelayout/.. directory
     72Blocks-layout-driver code is in: fs/nfs/blocklayout/.. directory
     73Flexfiles-layout-driver code is in: fs/nfs/flexfilelayout/.. directory
     74
     75blocks-layout setup
     76===================
     77
     78TODO: Document the setup needs of the blocks layout driver