Home | History | Annotate | Line # | Download | only in lfs
TODO revision 1.5 
      1 #   $NetBSD: TODO,v 1.5 2001/07/13 20:30:22 perseant Exp $
      2 
      3 - Investigate alternate inode locking strategy: Inode locks are useful
      4   for locking against simultaneous changes to inode size (balloc,
      5   truncate, write) but because the assignment of disk blocks is also
      6   covered by the segment lock, we don't really need to pay attention to
      7   the inode lock when writing a segment, right?  If this is true, the
      8   locking problem in lfs_{bmapv,markv} goes away and lfs_reserve can go,
      9   too.
     10 
     11 - Fully working fsck_lfs.  (Really, need a general-purpose external
     12   partial-segment writer.)
     13 
     14 - Inode blocks are currently the same size as the fs block size; but all
     15   the ones I've seen are mostly empty, and this will be especially true
     16   if atime information is kept in the ifile instead of the inode.  Could
     17   we shrink the inode block size to DEV_BSIZE?  Or parametrize it at fs
     18   creation time?
     19 
     20 - Get rid of DEV_BSIZE, pay attention to the media block size at mount time.
     21 
     22 - More fs ops need to call lfs_imtime.  Which ones?  (Blackwell et al., 1995)
     23 
     24 - lfs_vunref_head exists so that vnodes loaded solely for cleaning can
     25   be put back on the *head* of the vnode free list.  Make sure we
     26   actually do this, since we now take IN_CLEANING off during segment write.
     27 
     28 - The cleaner could be enhanced to be controlled from other processes,
     29   and possibly perform additional tasks:
     30 
     31   - Backups.  At a minimum, turn the cleaner off and on to allow
     32 	effective live backups.  More aggressively, the cleaner itself could
     33 	be the backup agent, and dump_lfs would merely be a controller.
     34 
     35   - Cleaning time policies.  Be able to tweak the cleaner's thresholds
     36 	to allow more thorough cleaning during policy-determined idle
     37 	periods (regardless of actual idleness) or put off until later
     38 	during short, intensive write periods.
     39 
     40   - File coalescing and placement.  During periods we expect to be idle,
     41     coalesce fragmented files into one place on disk for better read
     42     performance.  Ideally, move files that have not been accessed in a
     43     while to the extremes of the disk, thereby shortening seek times for
     44     files that are accessed more frequently (though how the cleaner
     45     should communicate "please put this near the beginning or end of the
     46     disk" to the kernel is a very good question; flags to lfs_markv?).
     47 
     48   - Versioning.  When it cleans a segment it could write data for files
     49     that were less than n versions old to tape or elsewhere.  Perhaps it
     50     could even write them back onto the disk, although that requires
     51     more thought (and kernel mods).
     52 
     53 - Move lfs_countlocked() into vfs_bio.c, to replace count_locked_queue;
     54   perhaps keep the name, replace the function.  Could it count referenced
     55   vnodes as well, if it was in vfs_subr.c instead?
     56 
     57 - Why not delete the lfs_bmapv call, just mark everything dirty that
     58   isn't deleted/truncated?  Get some numbers about what percentage of
     59   the stuff that the cleaner thinks might be live is live.  If it's
     60   high, get rid of lfs_bmapv.
     61 
     62 - There is a nasty problem in that it may take *more* room to write the
     63   data to clean a segment than is returned by the new segment because of
     64   indirect blocks in segment 2 being dirtied by the data being copied
     65   into the log from segment 1.  The suggested solution at this point is
     66   to detect it when we have no space left on the filesystem, write the
     67   extra data into the last segment (leaving no clean ones), make it a
     68   checkpoint and shut down the file system for fixing by a utility
     69   reading the raw partition.  Argument is that this should never happen
     70   and is practically impossible to fix since the cleaner would have to
     71   theoretically build a model of the entire filesystem in memory to
     72   detect the condition occurring.  A file coalescing cleaner will help
     73   avoid the problem, and one that reads/writes from the raw disk could
     74   fix it.
     75 
     76 - Need to keep vnode v_numoutput up to date for pending writes?
     77 
     78 - If delete a file that's being executed, the version number isn't
     79   updated, and fsck_lfs has to figure this out; case is the same as if
     80   have an inode that no directory references, so the file should be
     81   reattached into lost+found.
     82 
     83 - Currently there's no notion of write error checking.
     84   + Failed data/inode writes should be rescheduled (kernel level bad blocking).
     85   + Failed superblock writes should cause selection of new superblock
     86   for checkpointing.
     87 
     88 - Future fantasies:
     89   - unrm, versioning
     90   - transactions
     91   - extended cleaner policies (hot/cold data, data placement)
     92 
     93 - Problem with the concept of multiple buffer headers referencing the segment:
     94   Positives:
     95     Don't lock down 1 segment per file system of physical memory.
     96     Don't copy from buffers to segment memory.
     97     Don't tie down the bus to transfer 1M.
     98     Works on controllers supporting less than large transfers.
     99     Disk can start writing immediately instead of waiting 1/2 rotation
    100         and the full transfer.
    101   Negatives:
    102     Have to do segment write then segment summary write, since the latter
    103     is what verifies that the segment is okay.  (Is there another way
    104     to do this?)
    105 
    106 - The algorithm for selecting the disk addresses of the super-blocks
    107   has to be available to the user program which checks the file system.
    108