home
***
CD-ROM
|
disk
|
FTP
|
other
***
search
/
AmigActive 13
/
AACD13.ISO
/
AACD
/
Online
/
fsp
/
ChangeLog
next >
Wrap
Text File
|
1992-04-08
|
9KB
|
191 lines
Version 1.0 -- Dec 91. Original source created.
Version 2.0 -- Dec 91. Added man pages. Server bug fix. In 1.0, files
in a directory that is protected from deletion can still
be written over. In 2.0, it is prohibited as well.
Version 2.1 -- Jan 1, 92.
1) The maximum length of a name in pre-2.1 server is
both limited by the 1k-byte udp packet size and, in
some cases more severely, by the OS it is running in.
On some systems, the OS imposes a limite of 14 chars.
For those systems, the server can now be made to hash
long names into short ones. The effect is transparent
to the clients except when two long names gets hashed
into the same value. In that case, they will refer
to the same file during downloading, and they will
displace each others during uploading.
2) Since directories, unlike files, have undisputable
ownership, removal of a directory owned by a client
should depends only on the ownership of the directory
itself, and not on that of its parent directory. In
pre-2.1 server, the client must have delete permission
to the parent directory before it is allowed to delete
the subdirectory. In 2.1, the permission is not needed.
Version 2.2 -- Feb 15, 92.
1) The fput and fgetcmd utilities will now accept a list of
files from its standward input. See their man pages for
this expanded usage. This is useful if you don't want
file names to show on 'ps' output.
2) The retransmit timer for client code has been changed
to use an adaptive algorithm. In pre-2.2 versions of
FSP, clients will retransmit if it has not received a
response from the server after a delay of 3 seconds.
In version-2.2, two separate delays are used. The
"busy" delay is used when waiting for a reply for the
first try and the second try. Any time a second try
is necessary, the delay is increased by 50%. Every
first try brings the delay 12.5% closer to the initial
delay setting (3 seconds by default, settable by the
FSP_DELAY variable). For the third try and beyond,
the "idle" delay is used. For each try, the delay is
doubled.
3) Makes available FSP request code space 0x80 - 0xff for
future user extensions. Each such extension will begin
with a code byte in that range, and followed by a control
block, which is likely to be different from the existing
one. The basic operations of FSP will not require the
use of such extensions.
Note: 2.2 does not involve changes to the server itself.
The server is been checked out and added to by a
few people on the net. I will wait until the
modifications stabilize before I consolidate them.
Version 2.3 -- Mar 9, 92
1) The bsd_src/glob.c file has a bug in blkfree function
in that it tries to free something that is actually
on the stack of the glob function. The last call to
free has been commented out to fix the bug.
2) The fspd has a bug whoes only visible manifestation
is the that protection can't be changed. The fix has
required a moderate amount of code change in fspd.
3) .FSP_NO_DEL and .FSP_NO_ADD mechanism has been replaced
by .FSP_OK_DEL and .FSP_OK_ADD mechanism. Thus by
default, a directory is no delete and no add.
4) fprocmd is modified to relax option checking, allowing
arbitrary option characters be passed to the fsp server.
5) A fls bug is fixed so that a "fls /" in a directory other
than the root directory will not cause an error.
6) A -DDIRENT switch is added in Makefile to help support
those machines that does not have /usr/include/dirent.h.
Version 2.3a -- March 20, 1992
1) In a few places, non-critical errors of the form:
if(errno = EINTR) ....
was made. It obviously should have been == instead.
The problem is fixed.
2) In the previous versions there is a potential problem
such that if the first reply packet is dropped or is
messed up, making a retransmit necessary, fspd will
make the client utility wait for a 1 minute timeout.
A potential loop hole for this happening has been
closed by modifying server_lib.c where the packet key
is checked.
Version 2.4 -- March 27, 1992
1) fgrab/fgrabcmd client utility is added. It is like
fget except it also deletes the named files in such a
way that if there are multiple fgrabs for the same
file, only one will succeed. One bug/mis-feature of
fgrab is that when fgrab is terminated abnormally, the
file is not restored. This command is useful for things
like sharing bone files for modified versions of nethack.
2) FSP now allows multiple client programs to run at the
same time if one of the three multiplexing mechanisms
are chosen during compile time. It is likely that at
least one of them will work for your system. The
multiplexing is done at the client side, and does not
allow the client machine to issue request messages at
a higher rate compared to the previous versions. The
FSP_LOCALPORT variable is no longer needed. However,
if used, it will force serialization of client programs,
which might be desireable if none of the multiplexing
mechanisms are suitable for your machine. In order to
accommodate this change, the server code has been
changed to store temporary upload files with a different
name from the previous version. The names now have to
identify sender's port as well as sender's internet id.
Change is necessary to keep it under 14-character limit
of some OS.
3) Progress report mechanism enabled by FSP_TRACE variable
now try to update the kbyte display more promptly.
4) fcat/fcatcmd will now turn off FSP_TRACE during its
file transfer if its stdout is a tty.
Version 2.5 -- April 8, 1992
1) A bug introduced during the last release --- fspd will
not timeout and exit when running under inetd mode.
This bug is now fixed. It was introduced because it
is more convienent at one time for it not to exit when
it is still being debugged. The patch to stop it from
exiting was not removed before release time.
2) A very rarely seen bug in some OS is circumvented. UDP
packet's headers contain a checksum field. If the
checksum option is disabled, the OS will set the
checksum field to 0 before sending, and will not check
the checksum field when message is received. If the
option is enabled, the OS will stuff the checksum into
the checksum field. When a packet is received, the
checksum field is checked. If the checksum field is 0,
then the OS assume that the sender does not have option
enabled, and therefore the checking is skipped.
Problem arises when a packet's computed checksum is
zero. Some OS will stuff 0xffff into the checksum
field before sending such a message to distinguish it
from the case for which checksum is disabled. Other
OS couldn't recognize 0xffff as an indication that the
actual checksum could be either 0 or 0xffff, and
subsequently reject the message. The client can thus
get stuck forever in retransmitting a request that
happens to have 0 for the computed checksum, but
0xffff for the checksum field.
To get around the problem, the sequence number field
of the request message (which has meaning only to the
client software), is partitioned such that 2 of the
bits reflect the retry sequence, and the remaining 14
bits reflect the request sequence. This will ensure
that if one request got ignored because it happens to
have a checksum of 0, the next retry will result in
a packet that does not have a checksum of 0. The only
file affected here is client_lib.c.
3) In response to those who have severely limited UDP
packet sizes, the environment variable FSP_BUF_SIZE
is supplied to set the data size of file and directory
information transfers. The default is 1024. You cannot
set it to be bigger than 1024, but you can set it to
be smaller if you find that your system cannot pass
UDP packets that includes 1024 bytes of data. This
requires changes both in the server and the client
code. The first parameter field of get-file and
get-directory requests contains the path of the file
or directory. The second parameter field was unused,
but it is now used to hold the number of bytes to read.
This change is backward compatible in the sense that
if the second field is empty (old client code), the
new server code will default to 1024. Changes to the
client code is confined to client_util.c.