Signed-off-by: Stefan Metzmacher <metze@samba.org> Reviewed-by: Jeremy Allison <jra@samba.org>
Autobuild-User(master): Jeremy Allison <jra@samba.org>
Autobuild-Date(master): Fri Aug 19 19:39:18 UTC 2022 on sn-devel-184
s3:smbd: inline fsp_lease_type_is_exclusive() logic into contend_level2_oplocks_begin_default
SMB2_LEASE_WRITE is the indication for an exclusive lease,
the fact that a SMB2_LEASE_WRITE can't exists without
SMB2_LEASE_READ is not important here.
Volker Lendecke [Mon, 31 Aug 2020 14:14:14 +0000 (16:14 +0200)]
vfs_gpfs: Prevent mangling of GPFS timestamps after 2106
gpfs_set_times as of August 2020 stores 32-bit unsigned tv_sec. We
should not silently garble time stamps but reject the attempt to set
an out-of-range timestamp.
s3:smbd: only clear LEASE_READ if there's no read lease is left
If contend_level2_oplocks_begin_default() skips break it's
own lease, we should not clear SHARE_MODE_LEASE_READ
in share_mode_data->flags.
Otherwise that lease won't see any lease break notifications
for writes from other clients (file handles not using the same lease
key).
So we need to count the number existing read leases (including
the one with the same lease key) in order to know it's
safe to clear SMB2_LEASE_READ/SHARE_MODE_LEASE_READ.
Otherwise the next run (likely from another client)
will get the wrong result from file_has_read_lease().
Signed-off-by: Stefan Metzmacher <metze@samba.org> Reviewed-by: Ralph Boehme <slow@samba.org>
Autobuild-User(master): Jeremy Allison <jra@samba.org>
Autobuild-Date(master): Thu Aug 18 19:41:33 UTC 2022 on sn-devel-184
Jeremy Allison [Wed, 17 Aug 2022 18:43:47 +0000 (11:43 -0700)]
s3: smbd: Plumb close_type parameter through close_file_in_loop(), file_close_conn()
Allows close_file_in_loop() to differentiate between SHUTDOWN_CLOSE
(previously it only used this close type) and ERROR_CLOSE - called
on error from smbXsrv_tcon_disconnect() in the error path. In that
case we want to close the fd, but not run any delete-on-close actions.
Signed-off-by: Jeremy Allison <jra@samba.org> Reivewed-by: Noel Power <npower@samba.org>
Autobuild-User(master): Noel Power <npower@samba.org>
Autobuild-Date(master): Thu Aug 18 14:10:18 UTC 2022 on sn-devel-184
Jeremy Allison [Wed, 17 Aug 2022 18:35:29 +0000 (11:35 -0700)]
s3: smbd: Add "enum file_close_type close_type" parameter to close_cnum().
Not yet used, but needed so we can differentiate between
SHUTDOWN_CLOSE and ERROR_CLOSE in smbXsrv_tcon_disconnect()
if we fail to chdir. In that case we want to close the fd,
but not run any delete-on-close actions.
Signed-off-by: Andreas Schneider <asn@samba.org> Reviewed-by: Pavel Filipenský <pfilipensky@samba.org>
Autobuild-User(master): Pavel Filipensky <pfilipensky@samba.org>
Autobuild-Date(master): Wed Aug 17 11:03:54 UTC 2022 on sn-devel-184
Jeremy Allison [Fri, 22 Jul 2022 15:28:03 +0000 (16:28 +0100)]
s3/smbd: Use after free when iterating smbd_server_connection->connections
Change conn_free() to just use a destructor. We now
catch any other places where we may have forgetten to
call conn_free() - it's implicit on talloc_free(conn).
Based on code from Noel Power <noel.power@suse.com>.
Signed-off-by: Jeremy Allison <jra@samba.org> Reviewed-by: Noel Power <npower@samba.org>
Autobuild-User(master): Noel Power <npower@samba.org>
Autobuild-Date(master): Wed Aug 17 09:54:06 UTC 2022 on sn-devel-184
If chdir_current_service(tcon->compat) fails, we return status without ever having
called close_cnum(tcon->compat, vuid), leaving the conn pointer left in the linked
list sconn->connections.
The caller frees tcon and (by ownership) tcon->compat, still leaving the
freed tcon->compat pointer on the sconn->connections linked list.
When deadtime_fn() fires and walks the sconn->connections list it
indirects this freed pointer. We must call close_cnum() on error also.
Valgrind trace from Noel Power <noel.power@suse.com> is:
==6432== Invalid read of size 8
==6432== at 0x52CED3A: conn_lastused_update (conn_idle.c:38)
==6432== by 0x52CEDB1: conn_idle_all (conn_idle.c:54)
==6432== by 0x5329971: deadtime_fn (smb2_process.c:1566)
==6432== by 0x5DA2339: smbd_idle_event_handler (util_event.c:45)
==6432== by 0x685F2F8: tevent_common_invoke_timer_handler (tevent_timed.c:376)
==6432== Address 0x19074b88 is 232 bytes inside a block of size 328 free'd
==6432== at 0x4C3451B: free (in /usr/lib64/valgrind/vgpreload_memcheck-amd64-linux.so)
==6432== by 0x5B38521: _tc_free_internal (talloc.c:1222)
==6432== by 0x5B39463: _tc_free_children_internal (talloc.c:1669)
==6432== by 0x5B38404: _tc_free_internal (talloc.c:1184)
==6432== by 0x5B39463: _tc_free_children_internal (talloc.c:1669)
==6432== by 0x5B38404: _tc_free_internal (talloc.c:1184)
==6432== by 0x5B39463: _tc_free_children_internal (talloc.c:1669)
==6432== by 0x5B38404: _tc_free_internal (talloc.c:1184)
==6432== by 0x5B39463: _tc_free_children_internal (talloc.c:1669)
==6432== by 0x5B38404: _tc_free_internal (talloc.c:1184)
==6432== by 0x5B385C5: _talloc_free_internal (talloc.c:1248)
==6432== by 0x5B3988D: _talloc_free (talloc.c:1792)
==6432== by 0x5349B22: smbd_smb2_flush_send_queue (smb2_server.c:4828)
==6432== Block was alloc'd at
==6432== at 0x4C332EF: malloc (in /usr/lib64/valgrind/vgpreload_memcheck-amd64-linux.so)
==6432== by 0x5B378D9: __talloc_with_prefix (talloc.c:783)
==6432== by 0x5B37A73: __talloc (talloc.c:825)
==6432== by 0x5B37E0C: _talloc_named_const (talloc.c:982)
==6432== by 0x5B3A8ED: _talloc_zero (talloc.c:2421)
==6432== by 0x539873A: conn_new (conn.c:70)
==6432== by 0x532D692: make_connection_smb2 (smb2_service.c:909)
==6432== by 0x5352B5E: smbd_smb2_tree_connect (smb2_tcon.c:344)
https://bugzilla.samba.org/show_bug.cgi?id=15128
Signed-off-by: Jeremy Allison <jra@samba.org> Reviewed-by: Noel Power <npower@samba.org>
Signed-off-by: Christian Ambach <ambi@samba.org> Reviewed-by: Andreas Schneider <asn@samba.org>
Autobuild-User(master): Andreas Schneider <asn@cryptomilk.org>
Autobuild-Date(master): Wed Aug 17 07:14:21 UTC 2022 on sn-devel-184
s3:vfs.h: change SMB_VFS_INTERFACE_VERSION to 48 for 4.18
Signed-off-by: Stefan Metzmacher <metze@samba.org> Reviewed-by: Ralph Boehme <slow@samba.org>
Autobuild-User(master): Stefan Metzmacher <metze@samba.org>
Autobuild-Date(master): Tue Aug 16 11:51:36 UTC 2022 on sn-devel-184
Jeremy Allison [Thu, 11 Aug 2022 17:03:58 +0000 (10:03 -0700)]
s3: smbd: Add IS_VETO_PATH checks to openat_pathref_fsp_case_insensitive().
Returns NT_STATUS_OBJECT_NAME_NOT_FOUND for final component.
Note we have to call the check before each call to
openat_pathref_fsp(), as each call may be using a
different filesystem name. The first name is the
one passed into openat_pathref_fsp_case_insensitive()
by the caller, the second one is a name retrieved from
get_real_filename_cache_key(), and the third one is the name
retrieved from get_real_filename_at(). The last two
calls may have demangled the client given name into
a veto'ed path on the filesystem.
Signed-off-by: Jeremy Allison <jra@samba.org> Reviewed-by: Stefan Metzmacher <metze@samba.org>
Autobuild-User(master): Stefan Metzmacher <metze@samba.org>
Autobuild-Date(master): Tue Aug 16 08:26:54 UTC 2022 on sn-devel-184
Jeremy Allison [Thu, 11 Aug 2022 16:55:56 +0000 (09:55 -0700)]
s3: smbd: Add IS_VETO_PATH check to openat_pathref_dirfsp_nosymlink().
Returns NT_STATUS_OBJECT_PATH_NOT_FOUND for directory component.
Note IS_VETO_PATH only looks at the last component, so we must
do it during the directory walk on each component.
Note, we also have to check after a call to get_real_filename_at()
as it may have demangled the client sent name into a filesystem
name that matches the "veto files" parameter.
Jeremy Allison [Thu, 11 Aug 2022 16:51:11 +0000 (09:51 -0700)]
s3: tests: Add samba3.blackbox.test_veto_files.
Shows we currently don't look at smb.conf veto files parameter
when opening a file or directory. Checks multi-component paths.
Also checks veto files that might be hidden behind a mangled
name.
When O_PATH is specified in flags, flag bits other than O_CLOEXEC,
O_DIRECTORY, and O_NOFOLLOW are ignored.
In preparation to use openat2(), which gives an error instead of
ignoring flags, we better remove unexpected flags, callers typically
pass O_RDONLY and O_NONBLOCK.
Signed-off-by: Stefan Metzmacher <metze@samba.org> Reviewed-by: Volker Lendecke <vl@samba.org>
s3:smbd: let openat_pathref_dirfsp_nosymlink() do a verification loop against . and .. first
I guess we should catch NT_STATUS_OBJECT_NAME_INVALID first,
currently the check is already done in check_path_syntax*,
but we may remove it in future.
But the most important reason for this is the
openat2(RESOLVE_NO_SYMLINK) optimization, which will
be introduced in the following commits.
Review with: git show -w
Signed-off-by: Stefan Metzmacher <metze@samba.org> Reviewed-by: Volker Lendecke <vl@samba.org>
Volker Lendecke [Fri, 3 Jun 2022 14:45:41 +0000 (16:45 +0200)]
vfs: define VFS_OPEN_HOW_RESOLVE_NO_SYMLINKS
This will allow us to make use of openat2(RESOLVE_NO_SYMLINKS) soon.
The caller should check if connection_struct.open_how_resolve contains
VFS_OPEN_HOW_RESOLVE_NO_SYMLINKS before using it, this avoids waisting
cpu time. But even then the caller must be prepared to handle -1/ENOSYS.
Pair-Programmed-With: Stefan Metzmacher <metze@samba.org>
Signed-off-by: Volker Lendecke <vl@samba.org> Signed-off-by: Stefan Metzmacher <metze@samba.org>
vfs_io_uring: hide a possible definition of struct open_how in liburing/compat.h
liburing.h will include liburing/compat.h, which either includes
linux/openat2.h or defines struct open_how itself.
This will help with the following changes, which will provide
openat2() via libreplace's system/filesys.h, either including
linux/openat2.h or defining open_how ourself.
Signed-off-by: Stefan Metzmacher <metze@samba.org> Reviewed-by: Volker Lendecke <vl@samba.org>
Signed-off-by: Andreas Schneider <asn@samba.org> Reviewed-by: Jeremy Allison <jra@samba.org>
Autobuild-User(master): Jeremy Allison <jra@samba.org>
Autobuild-Date(master): Fri Aug 12 21:50:23 UTC 2022 on sn-devel-184
Signed-off-by: Jeremy Allison <jra@samba.org> Reviewed-by: Ralph Boehme <slow@samba.org>
Autobuild-User(master): Jeremy Allison <jra@samba.org>
Autobuild-Date(master): Fri Aug 12 19:18:25 UTC 2022 on sn-devel-184
Jeremy Allison [Mon, 8 Aug 2022 20:18:56 +0000 (13:18 -0700)]
s3: smbd: In filename_convert_dirfsp_nosymlink(), cope with an MS-DFS link as the terminal component.
If the terminal component was an MSDFS link, openat_pathref_fsp_case_insensitive() will
return NT_STATUS_OBJECT_NAME_NOT_FOUND with a VALID_STAT of a symlink.
If this is the case, check if we actually found a terminal MS-DFS link
at the end of the pathname and return NT_STATUS_PATH_NOT_COVERED.
Jeremy Allison [Mon, 8 Aug 2022 18:16:17 +0000 (11:16 -0700)]
s3: smbd: Add dfs_filename_convert(). Simple wrapper around parse_dfs_path().
Not yet used.
This is what we will use to replace dfs_redirect() in the filename
conversion code. Keep as a wrapper for now as we might want to
add some error checking around the 'hostname' and 'service'
returns.
Jeremy Allison [Fri, 5 Aug 2022 19:16:44 +0000 (12:16 -0700)]
s3: smbd: Add new function check_path_syntax_smb2_msdfs() for SMB2 MSDFS paths.
#ifdef'ed out as static and not yet used.
We can't just call check_path_syntax() on these as
they are of the form hostname\share[\extrapath]
(where [\extrapath] is optional).
hostname here can be an IPv6 ':' separated address,
which check_path_syntax() fails on due to the streamname
processing.
NB. This also has to cope with out existing (broken)
libsmbclient libraries that sometimes set the DFS
flag and then send a local pathname. Cope by just
calling the normal check_path_syntax() on the
whole pathname in that case.
Jeremy Allison [Sat, 6 Aug 2022 02:27:33 +0000 (19:27 -0700)]
s3: smbd: Fix cosmetic bug logging pathnames from Linux kernel clients using SMB1 DFS calls.
The Linux kernel SMB1 client has a bug - it sends
DFS pathnames as:
\\server\share\path
instead of:
\server\share\path
Causing us to mis-parse server,share,remaining_path here
and jump into 'goto local_path' at 'share\path' instead
of 'path'.
This doesn't cause an error as the limits on share names
are similar to those on pathnames.
parse_dfs_path() which we call before filename parsing
copes with this by calling trim_char on the leading '\'
characters before processing.
Do the same here so logging of pathnames looks better.
How did I find this ? Lots and lots of manual
testing with the Linux kernel client to make
sure all the recent changes haven't broken Linux
SMB1/2/3 DFS :-).
Sometimes the bottleneck is the smbtorture process.
In order to bring the smbd process to 100% cpu, you can use
'--option="libsmb:client_guid=6112f7d3-9528-4a2a-8861-0ca129aae6c4"'
and run multiple instances of the test at the same time,
which both talk to the same smbd process.
This is a very useful test to show how many requests are possible
at the raw SMB2 layer.
Signed-off-by: Stefan Metzmacher <metze@samba.org> Reviewed-by: Jeremy Allison <jra@samba.org>
Autobuild-User(master): Jeremy Allison <jra@samba.org>
Autobuild-Date(master): Thu Aug 11 19:23:37 UTC 2022 on sn-devel-184
This improves 'time smbtorture3 //foo/bar -U% local-g-lock-ping-pong -o 50000000'
from ~1.400.000 to ~3.400.000 operations per second any a testsystem.
As we also use TDB_VOLATILE for locking.tdb, this is a much more
realistic test now.
Signed-off-by: Stefan Metzmacher <metze@samba.org> Reviewed-by: Jeremy Allison <jra@samba.org>
s3:passdb: Consolidate error checking in fetch_ldap_pw()
Signed-off-by: Pavel Filipenský <pfilipensky@samba.org> Reviewed-by: Andreas Schneider <asn@samba.org>
Autobuild-User(master): Andreas Schneider <asn@cryptomilk.org>
Autobuild-Date(master): Thu Aug 11 06:34:56 UTC 2022 on sn-devel-184
Ralph Boehme [Wed, 27 Jul 2022 16:40:21 +0000 (18:40 +0200)]
vfs_default: assert all passed in fsp's and names are non-stream type
Enforce fsp is a non-stream one in as many VFS operations as possible in
vfs_default. We really need an assert here instead of returning an error, as
otherwise he can have very hard to diagnose bugs.
Ralph Boehme [Fri, 29 Jul 2022 12:56:41 +0000 (14:56 +0200)]
smbd: ignore request to set the SPARSE attribute on streams
As per MS-FSA 2.1.1.5 this is a per stream attribute, but our backends don't
support it in a consistent way, therefor just pretend success and ignore the
request.
Ralph Boehme [Wed, 27 Jul 2022 11:37:32 +0000 (13:37 +0200)]
CI: add a test trying to delete a stream on a pathref ("stat open") handle
When using vfs_streams_xattr, for a pathref handle of a stream the system fd
will be a fake fd created by pipe() in vfs_fake_fd().
For the following callchain we wrongly pass a stream fsp to
SMB_VFS_FGET_NT_ACL():
SMB_VFS_CREATE_FILE(..., "file:stream", ...)
=> open_file():
if (open_fd):
-> taking the else branch:
-> smbd_check_access_rights_fsp(stream_fsp)
-> SMB_VFS_FGET_NT_ACL(stream_fsp)
This is obviously wrong and can lead to strange permission errors when using
vfs_acl_xattr:
in vfs_acl_xattr we will try to read the stored ACL by calling
fgetxattr(fake-fd) which of course faild with EBADF. Now unfortunately the
vfs_acl_xattr code ignores the specific error and handles this as if there was
no ACL stored and subsequently runs the code to synthesize a default ACL
according to the setting of "acl:default acl style".
As the correct access check for streams has already been carried out by calling
check_base_file_access() from create_file_unixpath(), the above problem is not
a security issue: it can only lead to "decreased" permissions resulting in
unexpected ACCESS_DENIED errors.
The fix is obviously going to be calling
smbd_check_access_rights_fsp(stream_fsp->base_fsp).
This test verifies that deleting a file works when the stored NT ACL grants
DELETE_FILE while the basic POSIX permissions (used in the acl_xattr fallback
code) do not.
Allows passing on "user." xattr to the backend. This can be useful for testing
specific aspects of operation on streams when "streams_xattr" is configured as
stream filesystem backend.
projects / thirdparty / samba.git / log
