Michael Adam [Wed, 25 Feb 2009 08:53:16 +0000 (09:53 +0100)]
s3:registry: hash the list of subkeys in the regsubkey_ctr
This removes many loops over all the arrays
(from regsubkey_ctr_key_exists) and thus
reduces "net conf drop" from 1m55 to 48seconds
and "net conf import" from 1m55 to 58 seconds
for 2000 shares on my box.
This fix is reverted, because the speedup is going to move
further down into reg_objects.c. The unsorted list of subkey names
is going to be indexed: This O(n^2) search bites us in more places.
This re-establishes the abstraction of reg_objects.c.
Michael Adam [Wed, 4 Mar 2009 20:46:32 +0000 (21:46 +0100)]
s3:net conf: reduce memory usage of "net conf import".
"net conf import" was wrapped in one big transaction.
This lead to MAX_TALLOC_SIZE being exceeded at roughly
1500 shares. This patch resolves that problem by
limiting the top level transactions in "net conf import"
to 100 shares.
Michael Adam [Tue, 24 Feb 2009 10:25:44 +0000 (11:25 +0100)]
s3:libsmbconf: add transactions to the libsmbconf api
This is useful for wrapping higher level aggregate operations
in transactions. The text backend implementations just return
WERR_OK, the registry backend implementatoins use the
regdb_transaction_start|commit|cancel routines just added.
Michael Adam [Fri, 27 Feb 2009 14:28:52 +0000 (15:28 +0100)]
s3: Fix a memleak in dbwrap_rbt.
The SMB_MALLOC'ed rbt node data was not free'd on talloc free of
the db context. This is a quick fix using talloc instead of malloc
for allocation of the node data.
Since malloc was originally used for performance reasons, one
might want to reverse to malloc and create a talloc destructor
that walks the tree and frees all the node data if this talloc
approach proves to be too slow..
Michael Adam [Tue, 21 Oct 2008 21:20:57 +0000 (23:20 +0200)]
libsmbconf: fall back to file backend when no valid backend was found
Interpret the source string as a file name when it contains
a ':' sign but the initial part is not a known backend.
This might occur even implicitly when "%T" is used in an
include file name (even though this is not realistic..).
Jeremy Allison [Thu, 19 Mar 2009 02:59:04 +0000 (19:59 -0700)]
Fix bug #6196 - Unable to serve files with colons to Linux CIFS/VFS client
Looks like the pathname parsing for POSIX paths got
broken when the code for doing Windows streams parsing got added.
Jeremy.
Jeremy Allison [Wed, 18 Mar 2009 21:31:56 +0000 (14:31 -0700)]
Fix bug #6195 - Migrating from 3.0.x to 3.3.x can fail to update passdb.tdb correctly.
This is a really nasty one to fix as in order to successfully update the
passdb.tdb we must do the equivalent of a tdbbackup to move to the new hash
values before we do the upgrade.
Jeremy.
s3:signing: the seqnum should only be decremented by 1 for ntcancel requests
[MS-SMB] 3.3.5.1 Receiving Any Message says that the seqnum
is incremented by only for ntcancel requests for any other
request it's by incremented by 2, even if it doesn't expect
a response.
Andrew Tridgell [Tue, 10 Mar 2009 05:45:45 +0000 (16:45 +1100)]
fixed a bug in message handling for code the change notify code
The change notify code registered a separate message handler for each
tree connect. This registration uses the global messaging context.
The messaging code would consider a 2nd registration for the same
messaging type as being an 'update' of the handler, rather than a new
handler. It also would only call the first handler in the linked list
for a given message type when dispatching messages.
This patch changes the messaging code to allow for multiple
registrations of the same message type, and allow for multiple calls
to different messaging handler for one incoming message.
This fixes the problem with the test_notify_tcon() test that I
recently committed to the S4 smbtorture
Jeremy Allison [Fri, 6 Mar 2009 01:17:52 +0000 (17:17 -0800)]
Fix bug #6161 - smbclient corrupts source path in tar mode
This was my fault. I broke the smbclient tar argument processing
in creating the string for chdir when removing pstrings.
Jeremy.
Volker Lendecke [Thu, 5 Mar 2009 21:20:55 +0000 (22:20 +0100)]
Complete the fix for bug 6100
According to [MS-RPCE].pdf, section 2.2.2.11:
----
A client or a server that (during composing of a PDU) has allocated more space
for the authentication token than the security provider fills in SHOULD fill in
the rest of the allocated space with zero octets. These zero octets are still
considered to belong to the authentication token part of the PDU.<36>
----
RPC implementations are allowed to send padding bytes at the end of an auth
footer. Windows 7 makes use of this.
Jeremy Allison [Wed, 4 Mar 2009 00:05:47 +0000 (16:05 -0800)]
Fix bug #6155 - "force group" is no longer working as expected.
We need to store the "force group" uid separately from the
conn->server_info token as we need to apply it separately also.
Volker PLEASE CHECK !
Jeremy.
Dan Sledz [Tue, 3 Mar 2009 00:42:37 +0000 (16:42 -0800)]
It appears that the first time we see a uid/gid that winbind can't map,
we end up returning the null sid instead of falling back to the legacy
code. Next time through the code we'll hit the negative cache and do
the right thing, but we still fail the first time.
If we fail the winbind id to sid mapping, call the legacy version. This
catches the case where we don't have a negative cache entry for the mapping.
This is better than returning the NULL sid to the caller.
projects / thirdparty / samba.git / log
