GitLab

Detail:
  Xlate.c: Directory names containing characters requiring translation (eg. "AcornC/C++", "A B/C") would be translated twice, leading to a 'Directory not found' error when trying to open a file within. Submission from Colin Granville.
  TCPIP.h: Order the headers so that 'netdb.h' comes after its dependents. Submission for TCP/IP bounty.
Admin:
  Test case test/Contentious run, passed OK.

Version 2.62. Tagged as 'LanManFS-2_62'

Make SMB_RxWords a bit bigger per static anaylsis
  https://www.riscosopen.org/forum/forums/4/topics/9503
when LONGNAMES is undefined, since NT LM 0.12 dialect is allowed at the same time as short names.

A change to name translation in LanManFS-2_60 broke all forms of nested contentious name resolution, ie. where a leafname containing a contentious character was itself in a directory containing one.
The cited problem names ('1 2/3' and '1 /s') can both be resolved at one level only, no 2nd pass is required. The resolution function walks along the elements of the path one at a time to reach the leaf (ie. no element is processed twice). A test case has been created to allow some regression testing when making future changes in this area.

When building tboxlibint
  Copy h.* <Lib$Dir>.tboxlibint.h.*
when the directory is empty FileSwitch checks the destination ending in ...\h\* which the short names translation would error with ENOWILDCARD, but the long names translation attempts to resolve it by finding the trailing '\' and appending a '*' (ie. not doing anything), this returns no results and is passed back unfaulted. The special check in fsentry_file() for ENOWILDCARD for the two ReadCatalogueInformation thus doesn't trigger, and an attempt to FileGetAttribs() is made on the still wildcarded path. This is rejected by the server when in SMB_GetAttribsX2() does a TRANSACT2_QUERYPATHINFORMATION, which expects a non wildcarded object (on Windows servers at least), propagating an error back up to the user.
This stopped a ROM build from completing the exports phase via LanManFS on a Windows share.

Const-ify a few things.

Changed files
  Xlate.c: Revert change from revision 1.21, return ENOWILDCARD when appropriate.
  SMC.c: Increase MAX_WCT buffer size to 17.
  UK/CmdHelp: Add some missing chevrons round the command syntax for required parameters.
  test/Contentious: Test case for nested contentious character resolution code.

Version 2.61. Tagged as 'LanManFS-2_61'

Detail:
The translate routine that converts contentious characters in  a RiscOS
file name to equivalent DOS ones was being invoked a second time. Normally
this wouldn't be critical, but a / in riscos was converted to the . in DOS,
correctly , then converted back again. This error mode with more than one
contentious character in (particularly) a directory name had recently
come to light.
Admin:
Tested by building a RiscOS rom image on a pc based LanManFS mount without
errors.

Version 2.60. Tagged as 'LanManFS-2_60'

Detail:
The translate routine that converts contentious characters in a RiscOS
file name to equivalent DOS ones was being invoked a second time. Normally
this wouldn't be critical, but a / in riscos was converted to the . in DOS,
correctly , then converted back again. This error mode with more than one
contentious character in (particularly) a directory name had recently
come to light.
Admin:
Tested by building a RiscOS rom image on a pc based LanManFS mount without
errors.

Version 2.60. Tagged as 'LanManFS-2_60'

Fix for truncating writes to servers that declare MaxBufferSize >= 64k smb.c, lines 2726 onwards: when chopping up the next buffer, make sure the size sent still fits in the SMB_COM_WRITE count field (16 bits). For completeness, make sure the estimate of bytes remaining is similarly saturated too.

smb.c, lines 2711-2713: cifs-v1-spec-02 section 5.27 says the server always expects to receive up to 65535 bytes, so this test should not be subject to the negotiated MaxBufferSize, and is expected to often exceed it. Currently we use 32k, if this is ever increased we really ought to take into account MaxRawSize that the server tells us.

Tested with a handful of files of power-of-2 sizes and odd sizes. Inspection in Wireshark looks OK. Variation of a diagnosis by Will Ling https://www.riscosopen.org/forum/forums/4/topics/8966

Version 2.59. Tagged as 'LanManFS-2_59'

The args to LMTransport are mandatory, so should be in <> brackets rather than [].

Version 2.58. Tagged as 'LanManFS-2_58'

For *OPT1,2 the FS is supposed to report the leafname. LanManFS uses strrchr() to find the dot but hadn't incremented past it when returning the pointer.

Version 2.57. Tagged as 'LanManFS-2_57'

Detail:
  When the number of files read from the server didn't fit in the OS_GBPB buffer, continuation of the enumeration would fail.
  Resume key *and* filename used when resuming a directory enumeration to the same directory.
Admin:
  Submission from Colin Granville.

Version 2.56. Tagged as 'LanManFS-2_56'

Detail:
  c/LanMan - Don't pass silly error numbers to MessageTrans_ErrorLookup - the kernel will now fault them.
Admin:
  Tested on Raspberry Pi 3
  Fixes "bad error pointer" message when e.g. failing to log on to a share:
  https://www.riscosopen.org/forum/forums/4/topics/5264


Version 2.55. Tagged as 'LanManFS-2_55'

The PRM entry for Free reason 1 is very vague, it requires the length of the drive name including a terminator, not strlen() of the drive name. Free uses this directly to allocate RMA to copy into (though for drive names that aren't a multiple of 8 we're saved by OS_Module rounding up claims).
Also use Free SWI values from headers rather than defining them again locally.

Version 2.54. Tagged as 'LanManFS-2_54'

One too many %s's in the code, one too few in the messages file. This resulted in diagnostics messages (eg. at the bottom of *ListFS) to show a %s that hadn't got substituted.

Version 2.53. Tagged as 'LanManFS-2_53'

SMB.c:
Line 875 don't free chain pB that early it is (and was being) used later on.
Fixes to correct which SMB_RxWords the Sesskey and bloblen are retrieved from, for each of the dialects supported.
Call the new auth code when the protocol flags require it.
Don't uppercase passwords all the time, the server may require case sensitivity.
md5c.c/md4c.c:
Verbatim copies of the RSA reference implementations of MD4 and MD5 from the respective RFC's.
Auth.c:
Implementation of the hash/mash/reply that Microsoft requires. Define AUTHTEST of this file to build a simple application that verifies the algorithms.

Tested against a Win7 install, inspecting in Wireshark that the passwords are no longer sent in the clear.
Implements ticket #332.
Submission from Colin Granville.

Version 2.52. Tagged as 'LanManFS-2_52'

RES_PATH needs the component name too.

Version 2.51. Tagged as 'LanManFS-2_51'

RES_PATH needs the component name too.

Version 2.51. Tagged as 'LanManFS-2_51'

Take note of, and use, the MaxBufferSize that the server tells us during negotiation rather than a fixed value.
If the server doesn't say, use 1k as the default as before.
Submission from Colin Granville.

Version 2.50. Tagged as 'LanManFS-2_50'

Can't zero the t2flags having just set hServer to NULL.

Version 2.49. Tagged as 'LanManFS-2_49'

Careful analysis of packet captures by Colin Granville reveals that LanManFS waits for at least 1cs when no timeout was requested, so add a quick exit for that case. Timeouts of >= 1cs still delay for at least the requested time, and up to 0.999cs longer as before.
This reduces the turn around time which is most beneficial for interfaces faster than 10baseT.

Version 2.48. Tagged as 'LanManFS-2_48'

Also, rationalise the two Omni_RegisterClient calls to both use _swix since they're virtually the same function repeated twice.

Version 2.47. Tagged as 'LanManFS-2_47'

Due to an inability to count to 10 accurately, all the error tokens were wrong, leading to some inappropriate error messages.
Fixes problem "Message token E13 not found" when the SMB server gives a DOS error back.

Version 2.46. Tagged as 'LanManFS-2_46'

The table was ambiguously translating () to () and also [] to (). Now, we error names containing ,[] in the same manner as DOSFS.
The substitutions for ^% have also been changed to match DOSFS so copying from one to the other doesn't give two representations if viewed on a PC.
Marked both tables as const to save 512B of RMA.
This change is slightly academic since most of the time the long filename resolution is in use.

Tested against WinXP SP3 with a build with long filename support turned off.

Version 2.45. Tagged as 'LanManFS-2_45'

CoreFn.c:
The size returned by OS_File 5 might not match the size shown in the Filer viewer (or via *EX) because the last cached version was being returned. This would manifest itself as an error from !Edit if the file on the server had grown since last queried, when loading the file the amount of memory reserved was for the smaller (stale cache copy size).
The name cache is now always flushed on exit from the FileSwitch functions, this doesn't alter performance because the size was already being queried across the network, it was just that the result was overridden by the cached copy.
LanMan.c:
Be more robust and only capture the default driver name when it starts up.
This could be a problem if LanManFS starts after the ethernet driver, and the value of Inet$EtherType is unset. When the driver dies LanManFS would have taken that name as the default - leading to "Bad parameter" when trying to connect to a share.

Tested on IOMD with WinXP SP3.

Version 2.44. Tagged as 'LanManFS-2_44'

Version 2.43. Tagged as 'LanManFS-2_43'

Whose bright idea was it to use 0 for both? Fixes problem reported
  http://www.riscosopen.org/forum/forums/4/topics/2122

Version 2.42. Tagged as 'LanManFS-2_42'

Missed in earlier internationalisation but spotted in the binary - the stats report function is only called from a debug command, but the linker wasn't spotting that and removing the table of stats names.
Translate the descriptive text in debug_err in RPC.c.
Retagged as LanManFS-2_41.

With LanManFS in ROM it's not actually possible to enable verbose mode, since it relies on command arguments being passed to it on initialisation. Since RMREINIT doesn't allow that you can't ever pass '-v'. Even the RAM loaded case isn't that useful, since the verbose text is output via printf() which doesn't appear in the desktop (with OmniClient).
Swap the few uses of verbose mode to route via DebugLib instead, controlled via the 'DEBUG' predefine.
Also
  Makefile ammended so debug version has a rule to make the resources area.
Built, but not tested.

Version 2.41. Tagged as 'LanManFS-2_41'

Unused LMFiler put in the attic.
As LanMan has historically put its resources into ThirdPart/OmniClient this component has no ROM resources phase - it still registers the resources at run time.

Version 2.40. Tagged as 'LanManFS-2_40'

Fixed up a couple of trivial warnings with the stricter checking.
Removed OmniS_FastMultiply64 and rewrote the function in C, this allows the compiler to spot and use UMULL on architectures supporting that instruction.
Tested in a RAM build against a WinXP SP3 server with a drive > 4GB.

Version 2.39. Tagged as 'LanManFS-2_39'

Internally LanManFS was representing file offsets and data lengths as signed numbers, while in general this doesn't matter (they are passed through opaquely as just numbers and not manipulated) in a couple of places they were being used as while () loop exit conditions.
Function WriteZeroes (used when extending a file via OS_Args) therefore never performed the call to SMB_Write when beyond 2G because the loop condition was never met.
Additionally, it was literally transferring buffers of zeros across the network which is unnecessary as the server is documented as doing a zero fill itself, this represents at least a x40 speed up (on a slow 1MB/s drive write speed at the server end) or more.
Changed offset & lengths from int to uint.
Added the earlier CIFS spec (the later copy is missing section 5 for some reason) to docs.

Version 2.38. Tagged as 'LanManFS-2_38'

When a file ended in ',xxx' (unfinished DEADDEAD) or ',lxa' (load/execution addresses) LanManFS was correctly identifying these internally but at the last moment OR'd the result unconditionally with 0xFFF00000 when calculating the date stamp/access attributes.
Now, an unstamped file will propagate the load & exec addresses properly.
Tested with WinXP SP3 and a variety of hand constructed examples, plus copying a large file with the filer showed the incomplete file icon until done.

Version 2.37. Tagged as 'LanManFS-2_37'

Version 2.36. Tagged as 'LanManFS-2_36'

Retagged.

These two displays take different code paths, and *INFO goes via a check to workaround a bug in Win98 servers. Unfortunately, the check no longer works because Microsoft don't return the '.' and '..' entries in SMB Trans2 FIND_FIRST, and the logic previously always set the "need to frig the dates" flag. Now, the flag is only set if something comes back and after doing a test swap the results are also correct.
Turned on the 'auto close Trans2 FIND_FIRST' flag in the request block to save having to do a second transaction to close it.
Swapped the LDMFD/STMFD for Push and Pull in the assembler.
Tested against WinXP SP3, dates are no longer bonkers.

Version 2.35. Tagged as 'LanManFS-2_35'

Detail:
 Added an extra check after the Xlate path resolution code so that
 the case of the path given to resolve is honoured. This now means that
 the strcmp() done in SMB_Rename fails and the rename is actually sent to
 the server. Before "*rename test Test" would have the path resolved and
 since 'test' already existed both names would be the same, in this example
 something like "*rename A:\test A:\test" would result.
Admin:
 Made the non LONGNAMES version compile again (not tested).
 Removed local definitions of FSControl reason codes, uses Interface header
 now.
 Made the case of #includes match CVS to allow future compilation on a case
 sensitive system.
 Deleted unused "Version" header.

Version 2.34. Tagged as 'LanManFS-2_34'

Detail:
  When the underlying FS is NTFS, eg. WinXP Pro, the requested
  resume keys for a directory search are always zero - this is passed
  back via OSGBPB and on the next call it triggers a new search. So
  we get stuck in an infinite loop.
  Now checks for a resume key stuck at zero, and makes a fake one static
  to the search context then sets the continuation bit in the SMB_FIND_NEXT
  call because the server returning the duff key remembers the position.
  Dir_CallbackFn could return with "taken" undefined. Move a few lines
  higher.
  Missing "#else" added to CHECK_ARMBOOT_EXISTS so if this is disabled
  no further attempt is made to find !ArmBoot.
  Couple of typos corrected.
  Spelling of "disk" changed to "disc" in an error message.
Admin:
  Tested with XP Pro SP3 with NTFS, directories which previously hung
  the filer now open correctly. Checked with XP Home SP3 with FAT32 to
  verify behaviour unchanged from 2.32.
  Note, with large directories the initial path translation triggers a
  dir search too which takes ages compared with the actual number of
  entries returned. This could be cached to make counting far faster.

Version 2.33. Tagged as 'LanManFS-2_33'

Detail:
  Too many to list here: see doc/RPSChanges.
Admin:
  Received from Rob Sprowson approx 2009-02-08. Builds, but not tested by ROOL.

Version 2.32. Tagged as 'LanManFS-2_32'

Detail:
  Security fix MS09-001 (detailed in Microsoft Knowledge base article KB958687)
  fixes some vulnerabilities in the SMB protocol which LanManFS uses. This
  highlighted an error in SMB_WriteRaw which reported zero data bytes but then
  supplied an offset to the non existant bytes: this is now faulted by Windows.
  As there are no data bytes, the offset should also be zero.
Admin:
  Fix received from Rob Sprowson 2009-01-19.

Version 2.31. Tagged as 'LanManFS-2_31'

Version 2.30. Tagged as 'LanManFS-2_30'

Detail:
  > Default filetype now text
  >
  > Default type given to any file that has neither a dos extent, nor a
  > ',xxx' riscos extent at the server
  >
  > Ensure that files given the default type never have a ',xxx' riscos
  > extent appended.
  >
  > Add environment variable 'LMDefaultType' which checked for a default
  > type on module initialisation .. thus setting this to (e.g.) dos, then
  > rmreinitting LanManFS will change the default type to (e.g.) dos.
Admin:
  Not tested here yet.

Version 2.29. Tagged as 'LanManFS-2_29'