2 Todo/Status List for Linux-NFS
5 o denotes draft implementation, possibly commented out
7 + denotes done and tested
8 ------------------------------------------------------------------
12 * Server-side AUTH_DES authentication
16 * stat() calls don't check whether the cached attrs are stil valid
17 (this is a problem in the VFS).
18 - NFS_ROOT stuff needs fixing.
22 * Avoid recursion in low memory situations where
23 kmalloc may call try_to_swap_out etc ad inf.
24 * Don't do async I/O on swap files.
26 For special-casing related to NFS swap I/O, flag swap file
27 semantics in inode->i_flags. In swapfile.c, change functions
28 to call readpage/writepage if available, otherwise proceed
32 * Disable page cache invalidation/flushing for locked file
34 - Directory caching (we now have page-sized dircache entries
35 which could easily be organized into a linked list). These
36 dircache pages come along as a linked list that can be copied
37 almost 1-to-1 into a dirent struct. If this is put into the
38 VFS, other remote fs's will also benefit.
40 [Note: I just increased the readdir cache to hold more than
41 one directory. With this, the exclusive lock on readdir goes
42 away, too. With a larger cache, it may also be worth to think
43 about directory readahead...]
44 * Better lookup caching?
45 * When a read lock is present, don't time out attr cache or
46 page cache for that region. Likewise, if a write lock is present,
47 be lazy on write-back.
49 - BUG: Invalidate readdir cache after remove/rename/unlink
50 * Automatic `mounting' when the server crosses mount points
51 transparently (some IRIX machines seem to do this when
53 * NFSv3 support. This requires careful design to maximize
54 code sharing between NFSv2 and NFSv3.
55 * More robust rename handling (see comment before nfs_rename).
56 * Add Miquel's O_EXCL hack for file creation.
57 * Performance improvement: When a complete reply is received, and
58 the (async) task is woken up, don't put it on rpciod's scheduling
59 queue, but add it to a `fast scheduler queue.' The fast scheduler
60 could be a special handler that's registered on the tq_scheduler task
61 queue. This queue is fired by the kernel scheduler as soon as
62 the other bottom halves have been run.
64 Note that implementing this for sync tasks is even trickier than
65 for async tasks, because you have to make sure you do the right
66 thing in rpc_sleep_on().
67 * writeback of writable mmaps. Dirty pages are not subject to
68 writeback scheduling. Also, msync should make sure pages are
69 written with O_SYNC on.
74 * uid/gid mapping, and rpc.ugidd support
75 - Don't read/write a file that might have mandatory locks.
76 * Implement secure/kerberos export options (take care of lockd
77 fopen() calls--most clients seem to use NULL creds for lockd).
78 - there's a bug in readdir wrt large directories. Try mounting
79 the linux source tree and do an ls on include/linux...
80 * Support for UNIX socket creation.
81 * Someone should look over the error return codes. I tend to
82 mix up EPERM and EACCES.
84 - Refuse to look up inodes in procfs (security issues).
85 o Delayed writes (delay syncing of file data when nfsd handles
86 several write requests for the same file concurrently).
87 (Draft - see nfsd_write in fs/nfsd/write.c. Needs benchmarking).
88 * Faster read operations (single copy): mmap the file region
89 to be read into VM, and pass the VMA to the xdr routines
90 which pass the region's VM address into sock->ops->writemsg.
91 This copies the file data directly from the page cache into
93 Release the vma region after encoding.
94 * Faster write operations (single copy, with IPv6 net layout):
95 Get the unfragmented UDP datagram, pull the header and
96 do normal processing. Then mmap the file, copy the write data,
98 - Clear setuid/setgid bit after write().
103 * Server should run on privileged port.
104 * Testing reclaim support.
105 * HP lockd accepts our GRANT_MSG callback and passes on the grant
106 to the blocking process, but doesn't reply with a GRANT_RES.
107 It's not clear to me why it would do this.
108 * Unregister hosts (SM_UNMON) with rpc.statd when appropriate.
112 * Unregister service from portmapper upon exit/SIGTERM
116 * If available, use version 3 of the mount protocol and
117 obtain pathconf data (fill in data->bsize).
121 - Manpages need to be written