Jeremy Allison [Thu, 22 Jan 2009 22:31:27 +0000 (14:31 -0800)]
Second part of the attemt to fix #4308 - Excel save operation corrupts file ACLs.
If the chown succeeds then the ACL set should also. Ensure this is the case
(refactor some of this code to make it simpler to read also).
Jeremy.
Jeremy Allison [Thu, 22 Jan 2009 18:58:38 +0000 (10:58 -0800)]
Another attempt to fix bug #4308 - Excel save operation corrupts file ACLs.
Simo is completely correct. We should be doing the chown *first*, and fail the
ACL set if this fails. The long standing assumption I made when writing the
initial POSIX ACL code was that Windows didn't control who could chown a file
in the same was as POSIX. In POSIX only root can do this whereas I wasn't sure
who could do this in Windows at the time (I didn't understand the privilege
model). So the assumption was that setting the ACL was more important (early
tests showed many failed ACL set's due to inability to chown). But now we have
privileges in smbd, and we must always fail an ACL set when we can't chown
first. The key that Simo noticed is that the CREATOR_OWNER bits in the ACL
incoming are relative to the *new* owner, not the old one. This is why the old
user owner disappears on ACL set - their access was set via the USER_OBJ in the
creator POSIX ACL and when the ownership changes they lose their access.
Patch is simple - just ensure we do the chown first before evaluating the
incoming ACL re-read the owners. We already have code to do this it just wasn't
rigorously being applied.
Jeremy.
s3:libsmb: handle the smb signing states the same in the krb5 and ntlmssp cases
SMB signing works the same regardless of the used auth mech.
We need to start with the temp signing ("BSRSPYL ")
and the session setup response with NT_STATUS_OK
is the first signed packet.
Now we set the krb5 session key if we got the NT_STATUS_OK
from the server and then recheck the packet.
All this is needed to make the fallback from krb5 to
ntlmssp possible. This commit also resets the cli->vuid
value to 0, if the krb5 auth didn't succeed. Otherwise
the server handles NTLMSSP packets as krb5 packets.
The restructuring of the SMB signing code is needed to
make sure the krb5 code only starts the signing engine
on success. Otherwise the NTLMSSP fallback could not initialize
the signing engine (again).
Jeremy Allison [Thu, 8 Jan 2009 18:56:36 +0000 (10:56 -0800)]
Fix race condition in alarm lock processing noticed by Richard Sharpe <realrichardsharpe@gmail.com>.
"It seems to me that if the lock is already held by another process when we
enter this code, there is a race between the timeout and the granting. If
the lock is subsequently granted, the process releasing the lock will signal
the wait variable (or whatever) and our process will be scheduled. However,
if the timeout occurs before we are scheduled, the timeout will be delivered
first.
We will have the lock but will forget we have the lock, and never release
it."
Jeremy.
Karolin Seeger [Wed, 17 Dec 2008 14:53:51 +0000 (15:53 +0100)]
s3/loadparm.c: Change default value for "ldap ssl".
LDAP_SSL_ON is not defined at all. That's why the actual default value
was "" for a long time. Set a more sensible default value without chnging the
default behaviour.
Michael Adam [Fri, 21 Nov 2008 23:42:54 +0000 (00:42 +0100)]
winbindd_ads: use the reconnect methods instead of the rpc methods directly
Some of the ads methods just point to the rpc methods.
This makes winbindd_ads use the reconnect methods instead of
calling the rpc methods directly in order to prevent
negative cache entries for e.g. name_to_sid, when the dc
has closed the connection without sending a reset.
Carsten Dumke [Thu, 13 Nov 2008 18:05:00 +0000 (12:05 -0600)]
net: Fix documentation of net rap printq info (bug #5892)
The man-page (see net(8)) and the usage-info (call "net help rap printq") of
'net rap printq' do contain an option "list" but in net_rap.c
the option is named "info".
Rename the option "list" in the documentation (man-pages + usage) to "info" to
match the code.
Jeff Layton [Thu, 6 Nov 2008 20:20:55 +0000 (15:20 -0500)]
mount.cifs: use lock/unlock_mtab scheme from util-linux-ng mount prog
The util-linux-ng sources have a good, but rather complex scheme for
locking the mtab before updating it. Mount helpers need to follow the
same scheme. Advisory locking only works if everyone is using the same
locking scheme.
Copy the routines we need from util-linux-ng into a separate source file
and then have mount.cifs and umount.cifs link in this object.
The long term goal is to have these routines in a separate helper
library (libmount). Mount helpers can then dynamically link in that lib.
Until that happens, this should serve as a suitable stopgap solution.
saf_join_store() should be called after a successful
domain join, the affinity to the dc used at join time
has a larger ttl, to avoid problems with delayed replication.
metze
Signed-off-by: Stefan Metzmacher <metze@samba.org> Signed-off-by: Günther Deschner <gd@samba.org>
Michael Adam [Mon, 27 Oct 2008 13:28:44 +0000 (14:28 +0100)]
winbind: fix smbd hanging on Solaris when winbindd closes socket.
On some versions of Solaris, we observed a strange effect of close(2)
on a socket: After the server (here winbindd) called close, the client fd
was not marked as readable for select. And a write call to the fd did
not produce an error EPIPE but just returned as if successful.
So while winbindd had called remove_client(), the corresponding smbd
still thought that it was connected, but failed to retrieve answers
for its queries.
This patch works around the problem by forcing the client fd to
the readable state: Just write one byte into the socket before
closing.
Jeremy Allison [Fri, 17 Oct 2008 04:01:56 +0000 (21:01 -0700)]
Cope with bad trans2mkdir requests from System i QNTC IBM SMB client.
If total_data == 4 Windows doesn't care what values
are placed in that field, it just ignores them.
The System i QNTC IBM SMB client puts bad values here,
so ignore them.
Jeremy.
Jeff Layton [Thu, 9 Oct 2008 14:51:33 +0000 (10:51 -0400)]
mount.cifs: make return codes match the return codes for /bin/mount (try #3)
The manpage for /bin/mount specifies that the return code should be a
positive integer (actually, it's a bitfield). Clean up the return
codes from mount.cifs to make them match the expected return values
from /bin/mount. This necessary for proper integration with autofs.
This is the third attempt at this patch. The changes here are minor,
just changing some return's from main() into exit() calls for
consistency's sake.
Jeff Layton [Thu, 9 Oct 2008 14:47:45 +0000 (10:47 -0400)]
mount.cifs: have uppercase_string return success on NULL pointer
We currently don't attempt to uppercase the device portion of the mount
string if there isn't a prefixpath. Fix that by making uppercase_string
return success without doing anything on a NULL pointer.
Jeremy Allison [Wed, 8 Oct 2008 18:37:43 +0000 (11:37 -0700)]
Fix bug #5814 - Winbindd dumping core in a strange manner while doing "rescan_trusted_domain".
From analysis by hargagan <shargagan@novell.com> :
"The winbindd_child_died() is also getting called from process_loop() in case of
SIGCHLD signal. In this case it doesn't make the timeout_handler to NULL for
the first request. It then initiate a new request using
schedule_async_request() which installs a new timeout handler for the same
request. In such a case, for a badly unresponsive system both the timeout
handler can be called. For the first call the "private_data" will be cleared
and for another call the timeout handler will be detecting the double free. So,
for such a case as well, the winbindd_child_died() should make the
timeout_handler to NULL."
Jeremy.
Jeremy Allison [Wed, 1 Oct 2008 20:23:36 +0000 (13:23 -0700)]
Fix use of DLIST_REMOVE as spotted by Constantine Vetoshev <gepardcv@gmail.com>.
This API is unusual in that if used to remove a non-list head it nulls out
the next and prev pointers. This is what you want for debugging (don't want
an entry removed from the list to be still virtually linked into it) but
means there is no consistent idiom for use as the next and prev pointers
get trashed on removal from the list, meaning you must save them yourself.
You can use it one way when deleting everything via the head pointer, as
this preserves the next pointer, but you *must* use it another way when not
deleting everything via the head pointer. Fix all known uses of this (the main
one is in conn_free_internal() and would not free all the private data entries
for vfs modules. The other changes in web/statuspage.c and winbindd_util.c
are not strictly neccessary, as the head pointer is being used, but I've done
them for consistency. Long term we must revisit this as this API is too hard
to use correctly.
Jeremy.
Jeremy Allison [Sat, 27 Sep 2008 01:41:05 +0000 (18:41 -0700)]
Fix for bug #5790 - samba returns STATUS_OBJECT_NAME_NOT_FOUND on set file disposition call.
This was my fault. I use a singleton cache (positive and negative) to speed up pathname based
qfileinfo/setfileinfo lookups for alternate fsp's open on the same path. I only invalidated the
negative cache on adding a new file fsp, as I incorrectly imagined the new fsp was put at the *end* of
the open files list. DLIST_ADD puts it at the start, meaning any subsequent open wasn't seen once the
cache was set. Doh !
Jeremy.
Jeff Layton [Fri, 12 Sep 2008 20:58:00 +0000 (16:58 -0400)]
mount.cifs: fix several problems when mounting subdirectories of shares (try 2)
This patch is the second patch to attempt to fix up some of the problems
with mounting subdirectories of shares. The earlier patch didn't handle
this correctly when POSIX extensions were enabled. This one does.
This is a bit of a confusing area since the different components of
a service string have different rules:
1) hostname: no '/' (slash) or '\' (backslash) is allowed to be
embedded within the string
2) sharename: same rules as hostname
3) prefixpath: '\' *is* allowed to be embedded in a path component,
iff POSIX extensions are enabled. Otherwise, neither
character is allowed.
The idea here is to allow either character to act as a delimiter when we
know that the character can't be anything but a delimiter (namely
everywhere up to the start of the prefixpath). The patch will convert
any '\' unconditionally to '/' in the UNC portion of the string.
However, inside the prefixpath, we can't make assumptions about what
constitutes a delimiter because POSIX allows for embedded '\'
characters. So there we don't attempt to do any conversion, and pass the
prefixpath to the kernel as is. Once the kernel determines whether POSIX
extensions are enabled, it can then convert the path if needed and it's
able to do so. A patch to handle this has already been committed to the
cifs-2.6 git tree.
This patch also fixes an annoyance. When you mount a subdir of a share,
mount.cifs munges the device string so that you can't tell what the
prefixpath is. So if I mount:
//server/share/p1/p2/p3
..then /proc/mounts and mtab will show only:
//server/share
Finally, it also tries to apply some consistent rules to the uppercasing
of strings.
Jeff Layton [Fri, 12 Sep 2008 20:57:59 +0000 (16:57 -0400)]
cifs.upcall: bump SPNEGO msg version number and don't reject old versions
When we added the ability for the kernel to send sec=mskrb5 to the
upcall, we subtly broke old cifs.upcall versions that don't understand
it. Bump the spnego message version to 2 to make this clear. Also,
change cifs.upcall to not reject requests with a version that's lower
than the current one, and to send the reply with the same version that
the request sent. The idea is to try and keep cifs.upcall backward
compatible with old kernels.
Jeff Layton [Fri, 12 Sep 2008 20:57:59 +0000 (16:57 -0400)]
cifs.upcall: handle MSKRB5 OID properly
When the kernel sends the upcall a sec=mskrb5 parameter, that means
the the MSKRB5 OID is preferred by the server. This patch fixes the
upcall to use that OID in place of the "normal" krb5 OID when it
gets a sec=mskrb5 parameter.
Signed-off-by: Jeff Layton <jlayton@redhat.com> Acked-by: Steve French <smfrench@gmail.com>
Jeff Layton [Fri, 12 Sep 2008 20:57:59 +0000 (16:57 -0400)]
cifs.upcall: negatively instantiate keys on error
When a request-key upcall exits without instantiating a key, the kernel
will negatively instantiate the key with a 60s timeout. Older kernels,
however seem to also link that key into the session keyring. This
behavior can interefere with subsequent mount attempts until the
key times out. The next request_key() call will get this negative key
even if the upcall would have worked the second time.
Fix this by having cifs.upcall negatively instantiate the key itself
with a 1s timeout and don't attach it to the session keyring.
projects / thirdparty / samba.git / log
