Hello to the samba developers, this bug has been tested on svn build Version 3.0.14pre1-SVN-build-6016. Copying 1 (or more files) from a samba share to a local OS/2 drive is failing, when the shell 4os2.exe is used inside an OS/2 text commandline window. The file does not contain any extended attributes. Two SMB shares have been "mounted": net use k: \\linux200\testgk samba3 share net use w: \\winxp1\samba windowsXP-ProSP2 share The same file "client.c" is copied from samba and from winxp to a local drive: [f:\samba3 2]copy k:client.c . K:\samba3\client.c => F:\samba3\client.c The parameter is incorrect. "F:\samba3\client.c" 0 files copied [f:\samba3 0]copy w:client.c . W:\test\client.c => F:\samba3\client.c 1 file copied I have sniffed both copy operations and will attach them. Best wishes. Guenter
Created attachment 1105 [details] failing samba3 copy case
Created attachment 1106 [details] working winxp copy case
Created attachment 1109 [details] Running FOX Pro from Samba drive Windows'98 Here is attempt to run FoxPro 2.6 from Samba 3.0.14pre1-SVN-build-6031. I've run fox.exe. fox.exe detects 386+ processor and attempts to find and run protected-mode foxprox.exe. But it can not do this on 3.0.12 and 3.0.14. On 3.0.10 everything was OK.
Hi Jeremy, i had time to dig a bit deeper. The shell 4os2.exe also tries to copy the extended attributes (EAs). That is failing during QUERY_FILE_INFO -> Info query all EAs. To shorten the sniffs, i now used a very short textfile (containing only "huhu"). I also tried another os2 box - a warpserver 5.0. The following is displayed: [<server01>-f:\tmp\samba3]copy y:gk\short.txt . Y:\gk\short.txt => F:\tmp\samba3\short.txt SYS0255: The extended-attribute list size is incorrect. "F:\tmp\samba3\short.txt" 0 files copied You'll see the fix, when you compare the winxp sniff. I attach 2 new sniffs, the os2 sniff also contains the NegProt stuff. Please have a closer look to the protocol index, which is returned in the negprot response packet. The index is 3, but should be 4! (also ethereal's text info is a bit misleading). Have a look into smb_server/negprot.c from samba4. The supported_protocols[] array should be tuned in samba3. We had the same problem with negprot in samba4.... Further on, DosErrors should be flagged in FLAGS2 (not NT_STATUS). Also the FLAGS2_EXTENDED_ATTRIBUTES bit is (often) not returned the right way. (smb.conf settings are ignored most the time, too). That's it for now. Best wishes. Guenter
Created attachment 1117 [details] NEW - failing samba3 ethereal sniff with NegProt
Created attachment 1118 [details] NEW - same ethereal sniff, when copy winxp to os2 (with NegProt)
Created attachment 1122 [details] Proposed patch. Ok, here is the fix for selecting the wrong negprot value plus the wrong EA size in the info query. I've applied these to the latest SVN code. The NT status codes has already been fixed in SVN. I think these should fix this problem. Please check. Thanks, Jeremy.
I think the patches now in the tree should fix this one. Please re-open if not. Jeremy.
Hello Jeremy, i tested again with svn build Version 3.0.14pre1-SVN-build-6016. The NegProt stuff is okay now. :-) But the copy bug is still there. When you look at the WinXP sniff, you'll see, that for QUERY_FILE_INFO -> Info query all EAs, _even_ when no EAs can be found, the "EA List Length" parameter must be returned as 4 (and not 0)! (btw - the file "short.txt" has no EAs)! The following 1 addition is needed to get that right: in smb/trans2.c -> static int call_trans2qfilepathinfo() ... case SMB_INFO_QUERY_ALL_EAS: { /* We have data_size bytes to put EA's into. */ size_t total_ea_len = 0; DEBUG(10,("call_trans2qfilepathinfo: SMB_INFO_QUERY_ALL_EAS\n")); ea_ctx = talloc_init("ea_ctx"); if (!ea_ctx) { return ERROR_NT(NT_STATUS_NO_MEMORY); } ea_list = get_ea_list_from_file(ea_ctx, conn, fsp, fname, &total_ea_len); if (!ea_list || (total_ea_len > data_size)) { talloc_destroy(ea_ctx); + SIVAL(pdata,0,4); /* EA List Length must be set to 4 */ data_size = 4; break; } data_size = fill_ea_buffer(ea_ctx, pdata, data_size, conn, ea_list); talloc_destroy(ea_ctx); break; } ..... -------------------------- I made that change locally and rebuild. All tested stuff is ok then. I checked that for both EA settings in smb.conf: ea support = yes ea support = no and for files with and without EAs. ea_list = get_ea_list_from_file() returns NULL, when "ea support = no", so that case is also caught here. I guess, the change you made to fill_ea_buffer() must also be corrected to SIVAL(pdata,0,4); but in this case that function is not called anyway. Best wishes. Guenter
Yes, you're correct ! Thanks. Applied. Jeremy.
sorry for the same, cleaning up the database to prevent unecessary reopens of bugs.