IF YOU WOULD LIKE TO GET AN ACCOUNT, please write an
email to Administrator. User accounts are meant only to access repo
and report issues and/or generate pull requests.
This is a purpose-specific Git hosting for
BaseALT
projects. Thank you for your understanding!
Только зарегистрированные пользователи имеют доступ к сервису!
Для получения аккаунта, обратитесь к администратору.
On systems with /dev/urandom, this avoids a change to secrets.tdb for every fork().
For other systems, we now only re-seed after a fork, and on startup.
No need to do it per-operation. This removes the 'need_reseed'
parameter from generate_random_buffer().
Andrew Bartlett
- NTLM2 support in the server
- KEY_EXCH support in the server
- variable length session keys.
In detail:
- NTLM2 is an extension of NTLMv1, that is compatible with existing
domain controllers (unlike NTLMv2, which requires a DC upgrade).
* This is known as 'NTLMv2 session security' *
(This is not yet implemented on the RPC pipes however, so there may
well still be issues for PDC setups, particuarly around password
changes. We do not fully understand the sign/seal implications of
NTLM2 on RPC pipes.)
This requires modifications to our authentication subsystem, as we
must handle the 'challege' input into the challenge-response algorithm
being changed. This also needs to be turned off for
'security=server', which does not support this.
- KEY_EXCH is another 'security' mechanism, whereby the session key
actually used by the server is sent by the client, rather than being
the shared-secret directly or indirectly.
- As both these methods change the session key, the auth subsystem
needed to be changed, to 'override' session keys provided by the
backend.
- There has also been a major overhaul of the NTLMSSP subsystem, to merge the 'client' and 'server' functions, so they both operate on a single structure. This should help the SPNEGO implementation.
- The 'names blob' in NTLMSSP is always in unicode - never in ascii.
Don't make an ascii version ever.
- The other big change is to allow variable length session keys. We
have always assumed that session keys are 16 bytes long - and padded
to this length if shorter. However, Kerberos session keys are 8 bytes
long, when the krb5 login uses DES.
* This fix allows SMB signging on machines not yet running MIT KRB5 1.3.1. *
- Add better DEBUG() messages to ntlm_auth, warning administrators of
misconfigurations that prevent access to the privileged pipe. This
should help reduce some of the 'it just doesn't work' issues.
- Fix data_blob_talloc() to behave the same way data_blob() does when
passed a NULL data pointer. (just allocate)
REMEMBER to make clean after this commit - I have changed plenty of data structures...
- The 'not implmented' checks are now done by all auth modules
- the ntdomain/trustdomain/winbind modules are more presise as to
what domain names they can and cannot handle
- The become_root() calls are now around the winbind pipe opening only,
not the entire auth call
- The unix username is kept seperate from the NT username, removing the
need for 'clean off the domain\' in parse_net.c
- All sid->uid translations are now validated with getpwuid() to put a very
basic stop to logins with 'half deleted' accounts.
Andrew Bartlett
to handle domain trusts. Jeremy and I talked about this
and it's going in as working code. It keeps winbind clean
and solves the trust problem with minimal changes.
To summarize, there are 2 basic cases where the deadlock would
occur. (1) lookuping up secondary groups for a user, and
(2) get[gr|pw]nam() calls that fall through the NSS layer because
they don't exist anywhere.
o To handle case #1, we bypass winbindd in sys_getgrouplist() unless
the username includes the 'winbind separator'.
o Case #2 is handled by adding checks in winbindd to return failure
if we are a DC and the domain matches our own.
This code has been tested using basic share connections, domain
logons, and with pam_winbind (both with and without 'winbind
use default domain'). The 'trustdomain' auth module should work
as well if an admin wants to manually create UNIX users for
acounts in the trusted domains.
Other misc fixes:
* we need to fix check_ntlm_password() to be able to determine
if an auth module is authoritative over a user (NT_STATUS_WRONG_PASSWORD,
etc...). I worked around my specific situation, but this needs to be
fixed. the winbindd auth module was causing delays.
* fix named server mutex deadlock between trust domain auth module
and winbindd looking up a uid
* make sure SAM_ACCOUNT gets stored in the server_info struct for the
_net_sam_logon() reply.
Configuration details:
The recommended method for supporting trusts is to use winbind.
The gets us around some of the server mutex issues as well.
* set 'files winbind' for passwd: and group: in /etc/nsswitch.conf
* create domain trusts like normal
* join winbind on the pdc to the Samba domain using 'net rpc join'
* add normal parameters to smb.conf for winbind
* set 'auth method = guest sam winbind'
* start smbd, nmbd, & winbindd
Problems that remain:
* join a Windows 2k/XP box to a Samba domain.
* create a 2-way trust between the Samba domain
and an NT domain
* logon to the windows client as a user from theh trusted
domain
* try to browse server in the trusted domain (or other
workstations). an NT client seems to work ok, but 2k
and XP either prompt for passwords or fail with errors.
apparanently this never got tested since no one has ever been
able to logon as a trusted user to a Samba domain from a Windows
client.
I'm still not convinced that sharing the option name with the administrative
code is the best idea, but anyway...
Tested by vl, bug #41.
Andrew Bartlett
- better error codes than NT_STATUS_UNSUCCESSFUL for domain logon errors
- make auth_winbind load the ntdomain module if winbind isn't there.
- use new trusted domains cache to determine if the domain is valid.
Andrew Bartlett
- NTLMSSP over SPENGO (sesssion-setup-and-x) cleanup and code refactor.
- also consequential changes to the NTLMSSP and SPNEGO parsing functions
- and the client code that uses the same functions
- Add ntlm_auth, a NTLMSSP authentication interface for use by applications
like Squid and Apache.
- also consquential changes to use common code for base64 encode/decode.
- Winbind changes to support ntlm_auth (I don't want this program to need
to read smb.conf, instead getting all it's details over the pipe).
- nmbd changes for fstrcat() instead of fstrcpy().
Andrew Bartlett
This commit builds on the auth subsystem to give Samba support for trusting NT4
domains. It is off by default, but is enabled by adding 'trustdomain' to the
'auth methods' smb.conf paramater.
Tested against NT4 only - there are still some issues with the join code for
Win2k servers (spnego stuff).
The main work TODO involves enumerating the trusted domains (including the RPC
calls to match), and getting winbind to run on the PDC correctly.
Similarly, work remains on getting NT4 to trust Samba domains.
Andrew Bartlett
The auth_authsupplied_info typedef is now just a plain struct - auth_context,
but it has been modified to contain the function pointers to the rest
of the auth subsystem's components.
(Who needs non-static functions anyway?)
In working all this mess out, I fixed a number of memory leaks and moved the
entire auth subsystem over to talloc().
Note that the TALLOC_CTX attached to the auth_context can be rather long-lived,
it is provided for things that are intended to live as long. (The
global_negprot_auth_context lasts the whole life of the smbd).
I've also adjusted a few things in auth_domain.c, mainly passing the domain as
a paramater to a few functions instead of looking up lp_workgroup(). I'm
hopign to make this entire thing a bit more trusted domains (as PDC) freindly
in the near future.
Other than that, I moved a bit of the code around, hence the rather messy diff.
Andrew Bartlett
- Move rpc_client/cli_trust.c to smbd/change_trust_pw.c
- It hasn't been used by anything else since smbpasswd lost its -j
- Add a TALLOC_CTX to the auth subsytem. These are only valid for the length
of the calls to the individual modules, if you want a longer context hide it
in your private data.
Similarly, all returns (like the server_info) should still be malloced.
- Move the 'ntdomain' module (security=domain in oldspeak) over to use the new
libsmb domain logon code. Also rework much of the code to use some better
helper functions for the connection - getting us much better error returns
(the new code is NTSTATUS).
The only remaining thing to do is to figure out if tpot's 0xdead 0xbeef for
the LUID feilds is sufficient, or if we should do random LUIDs as per the old
code.
Similarly, I'll move winbind over to this when I get a chance.
This leaves the SPOOLSS code and some cli_pipe code as the only stuff still in
rpc_client, at least as far as smbd is concerned.
While I've given this a basic rundown, any testing is as always appriciated.
Andrew Bartlett
probably will never actually be genearted, but I like the style in any case.
Also fix a segfault in 'net rpc' when the login failed and a small memory leak
on failure in the auth_info.c code.
Andrew Bartlett
the method used for checking if a domain is a trusted domain is very
crude, we should really call a backend fn of some sort. For now I'm
using winbindd to do the dirty work.
We now default encrypt passwords = yes
We now check plaintext passwords (however aquired) with the 'sam' backend
rather than unix, if encrypt passwords = yes.
(this kills off the 'local' backed. The sam backend may be renamed in its
place)
The new 'samstrict' wrapper backend checks that the user's domain is one of
our netbios aliases - this ensures that we don't get fallback crazies with
security = domain.
Similarly, the code in the 'ntdomain' and 'smbserver' backends now checks
that the user was not local before contacting the DC.
The default ordering has changed, we now check the local stuff first - but
becouse of the changes above, we will really only ever contact one
auth source.
Andrew Bartlett
subystem.
The particular aim is to modularized the interface - so that we
can have arbitrary password back-ends.
This code adds one such back-end, a 'winbind' module to authenticate
against the winbind_auth_crap functionality. While fully-functional
this code is mainly useful as a demonstration, because we don't get
back the info3 as we would for direct ntdomain authentication.
This commit introduced the new 'auth methods' parameter, in the
spirit of the 'auth order' discussed on the lists. It is renamed
because not all the methods may be consulted, even if previous
methods fail - they may not have a suitable challenge for example.
Also, we have a 'local' authentication method, for old-style
'unix if plaintext, sam if encrypted' authentication and a
'guest' module to handle guest logins in a single place.
While this current design is not ideal, I feel that it does
provide a better infrastructure than the current design, and can
be built upon.
The following parameters have changed:
- use rhosts =
This has been replaced by the 'rhosts' authentication method,
and can be specified like 'auth methods = guest rhosts'
- hosts equiv =
This needs both this parameter and an 'auth methods' entry
to be effective. (auth methods = guest hostsequiv ....)
- plaintext to smbpasswd =
This is replaced by specifying 'sam' rather than 'local'
in the auth methods.
The security = parameter is unchanged, and now provides defaults
for the 'auth methods' parameter.
The available auth methods are:
guest
rhosts
hostsequiv
sam (passdb direct hash access)
unix (PAM, crypt() etc)
local (the combination of the above, based on encryption)
smbserver (old security=server)
ntdomain (old security=domain)
winbind (use winbind to cache DC connections)
Assistance in testing, or the production of new and interesting
authentication modules is always appreciated.
Andrew Bartlett
This changes the way we process guest logons - we now treat them as normal
logons, but set the 'guest' flag. In particular this is needed becouse Win2k
will do an NTLMSSP login with username "", therefore missing our previous guest
connection code - this is getting a pain to do as a special case all over the
shop.
Tridge: We don't seem to be setting a guest bit for NTLMSSP, in either the
anonymous or authenticated case, can you take a look at this?
Also some cleanups in the check_password() code that should make some of the
debugs clearer.
Various other minor cleanups:
- change the session code to just take a vuser, rather than having to do a
vuid lookup on vuser.vuid
- Change some of the global_client_caps linking
- Better debug in authorise_login(): show the vuid.
Andrew Bartlett
code.
In particular this assists tpot in some of his work, becouse it provides the
connection between the authenticaion and the vuid generation.
Major Changes:
- Fully malloc'ed structures.
- Massive rework of the code so that all structures are made and destroyed
using malloc and free, rather than hanging around on the stack.
- SAM_ACCOUNT unix uids and gids are now pointers to the same, to allow them
to be declared 'invalid' without the chance that people might get ROOT by
default.
- kill off some of the "DOMAIN\user" lookups. These can be readded at a more
appropriate place (probably domain_client_validate.c) in the future. They
don't belong in session setups.
- Massive introduction of DATA_BLOB structures, particularly for passwords.
- Use NTLMSSP flags to tell the backend what its getting, rather than magic
lenghths.
- Fix winbind back up again, but tpot is redoing this soon anyway.
- Abstract much of the work in srv_netlog_nt back into auth helper functions.
This is a LARGE change, and any assistance is testing it is appriciated.
Domain logons are still broken (as far as I can tell) but other functionality
seems
intact.
Needs testing with a wide variety of MS clients.
Andrew Bartlett
To obtain the full group membership of a user (i.e nested groups on a
win2k native mode server) it is necessary to merge this list of groups
with the groups returned by winbindd when creating an nt access token.
This breaks winbindd linking while AB and I sync up our changes to the
authentication subsystem.
The big one is a global change to allow us to NULLify the free'ed pointer to a
former passdb object. This was done to allow idra's SAFE_FREE() macro to do
its magic, and to satisfy the input test in pdb_init_sam() for a NULL pointer
to start with.
This NULL pointer test was what was breaking the adding of accounts up until
now, and this code has been reworked to avoid duplicating work - I hope this
will avoid a similar mess-up in future.
Finally, I fixed a few nasty bugs where the pdb_ fuctions's return codes were
being ignored. Some of these functions malloc() and are permitted to fail.
Also, this caught a nasty bug where pdb_set_lanman_password(sam, NULL) acheived
precisely didilly-squat, just returning False. Now that we check the returns
this bug was spotted. This could allow different LM and NT passwords.
- the pdbedit code needs to start checking these too, but I havn't had a
chance to fix it.
I have also fixed up where some of the password changing code was using the
pdb_set functions to store *internal* data. I assume this is from a previous
lot of mass conversion work...
Most likally (and going on past experience) I have missed somthing, probably in
the LanMan password change code which I havn't yet been able to test, but this
lot is in much better shape than it was before.
If all this is too much to swallow (particularly for 2.2.2) then just adding a
sam_pass = NULL to the particular line of passdb.c should do the trick for the
ovbious bug.
Andrew Bartlett
they can have general effect.
Fixed up workstaion support in the rest of samba, so that we can do these
checks.
Pass through the workstation for cli_net_logon(), if supplied.
- the usersupplied_info now contains a smb_username (as it comes across on
the wire) and a unix_username (after being passed through mapping
functions)
- when doing security={server,domain} use the smb_username, otherwise use
the unix_username
