Check for SYS/GL during library init. Reason is that

[AROS.git] / workbench / network / smbfs / documentation / history.doc

smbfs 1.19 (27.11.2000)

- Fixed ACTION_FINDOUTPUT implementation; the memory to hold the
  name of the file to be created was released before the file
  was created. This could cause garbage to be written to the
  output file.

- Fixed ACTION_COPY_DIR implementation; the name of the lock to
  be duplicated did not get duplicated.

- Fixed ACTION_EXAMINE_OBJECT and ACTION_EXAMINE_NEXT
  implementations; protection bits are now reported 'properly'
  with respect to read/write access.


smbfs 1.20 (27.11.2000)

- ACTION_EXAMINE_NEXT now filters out SMB names that include
  '/' or ':' characters.


smbfs 1.21 (28.11.2000)

- Modified ACTION_RENAME_DISK to remove the volume node before
  changing it. It also sends the appropriate diskremove/diskinserted
  input events now.

- Reorganized and rewrote the code to reduce dependencies on
  the compiler runtime library.

- Unified error reporting code.

- SMBFS can now be launched from Workbench using the same parameters
  you would use when starting it from Shell. Error reporting
  has been adapted to show an error requester and to pool error
  messages rather than printing them to the console.


smbfs 1.22 (30.11.2000)

- Fixed the test that looks for file/lock access mode collisions.

- No longer changes the case of the password; there is now an
  option to make the change.

- Fixed ACTION_SAME_LOCK which was comparing the same lock with
  itself.

- Added support for ACTION_EXAMINE_ALL, ACTION_EXAMINE_ALL_END
  and ACTION_WRITE_PROTECT packet types.

- No longer allows for files and locks to refer to "." and "..".
  The use of the backslash character '\' is also disallowed in
  Amiga file/directory names.


smbfs 1.23 (30.11.2000)

- In the ExNext()/ExAll() handling functions local error variables
  were shadowing each other. Fixed.

- The error reporting requester now also lists the name of the
  service to connect to. This should help in telling different
  invocations of the program apart.

- The size of the directory cache is now a configurable option.

- Added a new option which forces all name comparisons to be
  done in a case-sensitive fashion.

- Reading/writing data to/from a file in chunks larger than
  65535 bytes could fail and cause the file system to hang.


smbfs 1.24 (1.12.2000)

- Added an option to allow hidden files to be omitted from
  directory listings.

- Improved upon the error reporting in the code that directly
  calls the TCP/IP API functions.

- Reorganized the AmigaDOS packet handling switch..case
  list.

- Added support for ACTION_MORE_CACHE packet type. The
  cache in this context refers to the directory scanning
  cache.

- Improved upon the host lookup error reporting code.

- Added support for file name translation such as via the
  CrossDOS translation table files.


smbfs 1.25 (2.12.2000)

- SMBFS now displays some configuration information as soon as
  it has successfully connected to the file server and is ready
  for action. This output can be suppressed with the new
  "QUIET" command line parameter.

- When removing a directory, the error code returned is examined
  more closely to see whether the directory was not yet empty.

- When setting the protection bits of a file or directory, the
  file system now treats either 'protected from deletion' and
  'protected from writing' as indications to mark a file as
  write protected. Previously, a file had to be both protected
  from deletion and from writing to be treated as write protected.

- Cleaned up the program's data structures and eliminated all
  static data that would hold state information. The entire
  client side is now completely reentrant.


smbfs 1.26 (3.12.2000)

- File and directory names are no longer duplicated in the
  SMB abstraction layer. The strings supplied by the Amiga
  file system layer are referenced instead.

- Changed the way the ACTION_MORE_CACHE implementation returns
  its result code to be compatible with current practices.


smbfs 1.27 (4.12.2000)

- The changes I made to the file read command in order to allow it
  to read more than 65535 bytes at a time did not allow read
  attempts at the end of the file to come out properly. Fixed.

- Integrated password encryption code lifted from Samba. Hope
  it works as it should, because if it doesn't, I wouldn't know
  how to fix it...


smbfs 1.28 (4.12.2000)

- Oh dear, more fixes to the password encryption code and the
  connection setup to follow it. This time I managed to test the
  code briefly with Samba and password encryption enabled.
  Amazingly, it seems to work this time!


smbfs 1.29 (6.12.2000)

- Writes larger than 65535 bytes could stall and slow down the
  file system. Fixed. [Thore Böckelmann]

- The ACTION_MORE_CACHE implementation never returned the total
  number of cache entries allocated. Fixed. [Thore Böckelmann]

- Plain text passwords didn't seem to work any more after I made
  the changes to allow for encrypted passwords to be used. Fixed.


smbfs 1.30 (6.12.2000)

- Changed the structure of the share names by which the file system
  refers to the data on the server; they all start with a backslash
  now. This seems to help NT and doesn't break the rest.

- For file shares exported using 'share level security', no passwords
  will be transmitted.


smbfs 1.31 (7.12.2000)

- The ACTION_EXAMINE and ACTION_EXAMINE_FH implementations now verify
  that the name to return will fit into the FileInfoBlock's limited
  buffer space.

- Deleting files and directories from a directory currently being
  scanned via ACTION_EXAMINE_NEXT or ACTION_EXAMINE_ALL will now cause
  the directory scanner to restart. This is ugly, I know, but at least
  it won't cause the scanning process to hickup or lose entries
  directory entries during its course of duty. The flip side is that
  the scanner will probably return some entries twice. There must
  be a better way to implement this, but for now, this is it.

- Tried to make device and volume node removal safer.

- SMBFS now always adds a device node to go along with the volume node.
  It's safer that way because for every file system you either have a
  single device node in the system (and no volume node) or a volume
  node *and* a device node. If the device node is missing, there's
  typically something wrong and some application software cannot handle
  this case properly. SMBFS will attempt to add a new, unique device
  node every time it is launched.

- When specifying the name of the file system device to add, SMBFS
  now complains if that name already exists.


smbfs 1.32 (8.12.2000)

- File and directory modification dates were always off by six hours and
  one second. I don't know how that happens, it doesn't seem to be a bug
  in the date conversion routines. I worked around the problem by adding
  and subtracting an adjustment value.


smbfs 1.33 (10.12.2000)

- The ACTION_PARENT_DIR implementation did not return a ZERO lock for
  the root directory. Fixed. [Rudolph Riedel]

- Replaced the doubly-linked Exec lists with skip lists which
  are better suited for the repeated lookup operations that
  are so common in the file system. This time the list code
  seems to work properly.

- Added -- largely untested -- time and date decoding routines for
  directory entries returned by NT.


smbfs 1.34 (10.12.2000)

- Had to take out the skip list code again since it did not seem to
  work reliably.

- Further tweaking to the NT date and time decoding code.


smbfs 1.35 (10.12.2000)

- Small modification to the ACTION_EXAMINE_NEXT/ACTION_EXAMINE_ALL code
  which tries to work around a strange effect caused by Amiga Samba
  returning the contents of an empty drawer.


smbfs 1.36 (11.12.2000)

- More fixes to the NT date and time conversion code and the
  strange six hour fixup that seems to be necessary for the
  'regular' time conversion to work.


smbfs 1.37 (11.12.2000)

- Replaced the entire 'regular' date and time conversion code. The
  strange six hour fixup is no longer necessary.


smbfs 1.38 (11.12.2000)

- Ditched the packet size kludges in "sock.c" and "proc.c" I had put
  in there. It seems that only the 'smba_write()'/'smba_read()' code
  could trigger it, and that code was fixed quite a while ago.

- Migrated some code changes from the Samba source over to the
  directory scanner.

- Maximum transmission size per packet is now the same as the
  number the server allowed us to use rather than a fixed number
  imposed by VM page sizes.

- Tried to clean up more of the code.

- Discovered that the 'smb_valid_packet()' routine never could have
  worked and fixed it so that it now finally does what it should do.

- Better error propagation for some routines in "proc.c".

- More streamlined error testing code.


smbfs 1.39 (11.12.2000)

- Almost fixed several bugs in 'smb_proc_readdir_long()'.

- 'smb_proc_readdir_long()' does not work correctly. It only seems to
  work, but will fail if the scanner is to restart before the entire
  directory is read. Restored the old behaviour :(


smbfs 1.40 (13.12.2000)

- Took out some more unused and unnecessary code.

- Rewrote the time and date conversion routines.

- Modified the directory reading code so that it no longer relies
  upon the last directory entry index which doesn't seem to be
  correct most of the time.

- Directory scanning errors are now flagged as such.


smbfs 1.41 (16.12.2000)

- In the SMB abstraction layer, the directory cache is now invalidated
  only if the respective action that could have invalidated it succeeds.

- Now clears the directory cache only if the respective action would
  have affected the directory being cached.

- In debug mode [Ctrl]+F will display a list of open files and allocated
  locks.

- Added very silly support for ACTION_SET_COMMENT packet.

- Found an interesting bug fix for 'smb_proc_readdir_long()' in the
  Linux 2.4.0 kernel which seemed worth integrating into the file system.


smbfs 1.42 (19.12.2000)

- Integrated Chris Hertel's broadcast name query code. This means that
  the name of the host whose file shares you want to access need no
  longer be in the hosts database. However, you still cannot refer to
  a host solely by its IP address.


smbfs 1.43 (21.12.2000)

- The integrated broadcase name query code would conflict with the
  DNS lookup. Fixed. [Thore Böckelmann]

- The NetBIOS name query code now retries four times to get a response
  out of the network.

- Put the mysterious 6 hour time offset fix back in.


smbfs 1.44 (29.12.2000)

- Rewrote the host lookup error reporting code to use the appropriate
  API function.

- Discovered that all the functionality required by SMBFS was supported
  by the original free AmiTCP 3.0 release. Changed the "bsdsocket.library"
  open parameters.

- The ACTION_EXAMINE_OBJECT implementation consistently got the name of
  the root directory wrong. Fixed.


smbfs 1.45 (7.1.2001)

- Discovered why file and directory creation dates were wrong by about
  six hours. Fixed.

- The order of the parameters passed to the 'ExAll()' filter hook was
  wrong. Fixed [Thorsteinn Sveinsson and Matt Sealey].


smbfs 1.46 (1.2.2001)

- Removed the workarounds for the mysterious 'off by six hours' bug.
  The time delta value added was consistently wrong.


smbfs 1.47 (18.2.2001)

- The name of the domain/workgroup, the name of the user and the
  password to be used can now be preset using environment variables.

- Changed the command line argument template to work better with the
  newly introduced environment variable settings.


smbfs 1.48 (3.3.2001)

- Modified the command template so that the final parameter can no
  longer cause all other parameters to be swallowed. Also updated
  the documentation.


smbfs 1.49 (3.3.2001)

- Handling of ACTION_FINDUPDATE needed improving. It should have
  created a file that didn't already exist. It now does [David Gerber].


smbfs 1.50 (4.3.2001)

- ACTION_FINDUPDATE is now associated with a shared lock. It used
  to be an exclusive lock [David Gerber].

- The test to follow ACTION_FINDUPDATE that determines whether the
  file to open already exists now also tries to 'stat()' the file
  in question to find out whether it exists.


smbfs 1.51 (13.3.2001)

- Whoops. The new code to pull data from environment variables didn't
  actually check whether the work group name could be set up properly
  and thus didn't complain about it either. It now notifies you if
  no work group name was provided and none could be found.


smbfs 1.52 (22.6.2001)

- Added another option which allows you to preset the time zone offset
  to use when translating between the local time and the SMB server
  time. Using the default locale doesn't work for that purpose all the
  time since there are so many misconfigured servers out there.

- The ACTION_DIE packet didn't have the desired effect if the file
  system didn't have a reason to wait for any resources to be
  released again. It just wouldn't quit. Now it does.

- Before it exits, the file system no longer posts a disk change
  event if it was launched from Workbench. I found that the steps
  taken so far could end up deadlocking Workbench.


smbfs 1.53 (26.6.2001)

- Changed the file name parser; it should now be possible to tie assigns
  to files and directories on an smb volume. Since the updated file name
  translation code also is a bit smarter about when to use which
  file name separator character, accesses to the media should be a
  bit faster, too.

- When translating path names, trailing '/' characters are now properly
  parsed, i.e. Lock("/",...) should always bring up the parent directory
  [Matt Sealey].

- Duplicating a ZERO lock now works as expected.


smbfs 1.54 (27.6.2001)

- Directory and file creation didn't work that properly after applying
  the changes I made to the parser. Fixed [Matt Sealey].


smbfs 1.55 (3.3.2002)
 
- At least samba 2.2.2-12 on linux can send total_data of 0 bytes. This
  caused sock.c/smb_receive_trans2() malloc() (==AllocVecPooled) to fail
  on smba_readdir(). Fixed [Harry Sintonen].

- Added quickfix against crash if smb_abstraction.c/smb_smba_readdir()
  fails. This is by no means a real fix, but at least it doesn't crash
  [Harry Sintonen].

- There is definitely something fishy in the smbfs restart/retry scheme,
  it just doesn't work properly (all state data has to be, and is,
  reset when restarts occur, but still it tries to use the old data as if
  nothing happened). Someone should really take a closer look at this
  [Harry Sintonen].


smbfs 1.56 (3.9.2002)

- Rewrote the Setup() code which allocates and initializes the device and
  volume nodes. Lost all Forbid()..Permit() pairs on the way, making the
  code more robust and less weird.

- In proc.c/smb_proc_readdir_long() now specifically watches for the
  smb_trans2_request() function return empty data and parameter values.
  This is now treated as an attempt to read from an empty directory.
  Previously, there was a slight change that the information was
  taken seriously even if NULL was returned.


smbfs 1.57 (3.9.2002)

- Ran the code through GCC and rewrote the code to get rid of the compiler
  warnings that appeared to have something more serious about them. This
  includes, for example, comparisons between signed and unsigned values.


smbfs 1.58 (5.9.2002)

- Rebuilt using my own TCP/IP stack SDK. Rebuilding the code with the
  Miami SDK should still be possible, though.

- Changed the way smb_proc_readdir_long() aborts the job if no valid
  data is read.


smbfs 1.59 (10.9.2003)  [Harry Sintonen <sintonen -at- iki -dot- fi>]

- Fixed ACTION_DELETE_OBJECT to restart directory scanning, it
  searched for wrong object. Fixes the problem where every other
  file was skipped when deleting directory.


smbfs 1.60 (11.2.2004) [Peter Riede <Noster-Riede -at- T-Online -dot- de>]

- Fixed ACTION_SEEK, it doesn't works if followed by ACTION_WRITE.
  Fixes the problem by using the SMBlseek packet instead of just
  remembering the position in the local FileNode structure

- Fixed ACTION_SETFILESIZE, has returned an errorcode of 37 instead
  of 0L on success.

- Added support for the ACTION_LOCK_RECORD and ACTION_UNLOCK_RECORD
  packets.


smbfs 1.61 (27.4.2004)

- Reworked the ACTION_SEEK implementation. Turns out that the SMB
  lseek packet only seems to work for the Samba server but produces
  unpredictable results with Microsoft flavoured servers. The new
  code, which helps to avoid trouble with overlapping write access
  and seeks, now performs much like the old one except that it
  "notifies" the SMB server of the new file position to be used.
  Tested both with Samba and Windows XP: works as it should.


smbfs 1.62 (8.5.2004)

- Added sanity checks to the record locking code. For example,
  locking a record of length 0 must be considered illegal. Also,
  since the protocol deals with 32 bit signed integers for
  file position and record length, parameters that are "too large"
  and come out as negative values are now rejected.


smbfs 1.63 (9.5.2004)

- Added more sanity checks to the record locking code, this time
  to avoid integer overflows.

- Added support for the SMB 'archive' and 'system' file attributes,
  which are mapped to the Amiga 'archived' and 'pure' protection
  bits.

- Cleaned up the header files.


smbfs 1.64 (13.5.2004)

- ExamineFH() did not fill in the file protection bits. Fixed.

- Ported to AmigaOS4.


smbfs 1.65 (16.5.2004)

- The port numbers of the NetBIOS session and name services are no
  longer hard-coded. They are now looked up (netbios-ns/udp and
  netbios-ssn/tcp), and if that fails, the hard-coded default port
  numbers are used instead.

- Tried to reduce the risk of triggering buffer overflows by replacing
  calls to strcpy/strcat. This revealed weaknesses in a number of
  places, including the code that fills in the names of the directory
  entries.

- Instead of calling memcpy(), the code now invokes exec.library/CopyMem(),
  which should be a wee bit faster than what the 'C' runtime library can do.

- Merged with Peter Riede's changes: for small packets to be sent,
  SMBwrite is used instead of SMBwritebraw (which greatly enhances
  performance), the server max_xmit field is initialized differently
  if the session was opened with a SMBtconx packet and ACTION_SEEK
  no longer sets the error code to an undefined value in case of
  success.


smbfs 1.66 (10.6.2004)

- Added a paranoia check to the ExAll() implementation which should
  avoid trouble with extremely short data buffers.


smbfs 1.67 (27.5.2005)

- Replaced the long NT date conversion code with something hopefully
  much more robust. The results so far are both encouraging and
  irritating. Dates that previously came out as "unknown" are now
  processed, but there are differences between the dates shown in the
  directory listing and by listing the files by name. Go figure...

- Transplanted some more code from Samba to handle directory entry
  data conversion.


smbfs 1.68 (3.6.2005)

- It appears that Windows XP can produce directory listings with
  file/directory modification times set to 0, indicating that no
  such information is available, while the associated last file
  write access data is present. Previously, smbfs only reported
  the modification. If this data is unavailable and the date of
  the last write access is, the last write access will be
  reported instead.


smbfs 1.69 (13.6.2005)

- The time stamps used in "proc.c", as returned and submitted to
  the SMB file system on the other end of the wire are apparently
  all in Universally Coordinated Time already (or at least, this
  seems to be the case with Samba and Windows XP). Hence no conversion
  between local time and UTC is necessary, which would otherwise
  distort all date stamps converted. I modified the file system to
  leave all time stamps essentially unadjusted for local time in
  "proc.c". The local time zone adjustments are now performed only
  where the time in question is known to be Amiga-specific, such as
  the current system time or the date stamp to set for a file.

  The time conversion appears to be working correctly now, but it
  does ignore the effects of daylight savings time, which you might
  want to adjust for manually through the TIMEZONEOFFSET option
  (careful though: this overrides your current locale-defined time
  zone settings as far as smbfs is concerned).

  Unsolved problem: at least Windows XP seems to return different
  time stamps for directory listings and for indidivual files. As
  far as I can tell, the sets of time stamps returned for either
  doesn't match anywhere.


smbfs 1.70 (13.6.2005)

- Introduced a new option to account for daylight savings time
  when appropriate.

- Looks like some of the time stamps used in "proc.c" are
  transmitted in local time after all; brought back the
  conversion functions.


smbfs 1.71 (13.6.2005)

- The extended directory scanning entry conversion code now swaps
  the last modification and last write access date stamps. This
  matches the time of the last modification, as returned by the
  regular ACTION_EXAMINE_FH/ACTION_EXAMINE packets. Note that the
  time resolution is a little bit coarser because the modification
  time as expressed by the SMB getattr function cannot represent
  60 distinct seconds per minute but only 30. The net effect is
  that the resulting time stamps in a directory listing and by
  examining a file can differ by one second.

- The DST option's time offset was added to rather than subtracted
  from the local time. Fixed.


smbfs 1.72 (14.4.2009)

- In proc.c, smb_setup_header() initialized the SMB header length
  field with a number which was too large by four bytes. Consequently,
  what was later committed to the wire would have four trailing data
  bytes which could contain random values. This often didn't do much
  harm, but it seems that Samba 3.2.4 and Windows Vista don't like the
  looks of the trailing junk bytes.


smbfs 1.73 (17.4.2009)

- Modified smb_valid_packet() in proc.c to check whether the packet
  received is smaller than expected, not whether the length matches
  exactly. This seems to fix the protocol negotiation with Samba 3.2.5
  for now. Haven't checked Vista yet, though...


smbfs 1.74 (31.8.2009)

- Integrated Harry Sintonen's fix for the problem caused by large
  writes to files. The respective packet size was off by one byte.
  Thank you very much!