Add rwpidforward mount option that switches on a mode when we forward
pid of a process who opened a file to any read and write operation.

This can prevent applications like WINE from failing on read or write
operation on a previously locked file region from the same netfd from
another process if we use mandatory brlock style.

It is actual for WINE because during a run of WINE program two processes
work on the same netfd - share the same file struct between several VFS
fds:
1) WINE-server does open and lock;
2) WINE-application does read and write.
Signed-off-by: NPavel Shilovsky <piastry@etersoft.ru>
Signed-off-by: NSteve French <sfrench@us.ibm.com>

Add cifs_match_super to use in sget to share superblock between mounts
that have the same //server/sharename, credentials and mount options.
It helps us to improve performance on work with future SMB2.1 leases.
Reviewed-by: NJeff Layton <jlayton@redhat.com>
Signed-off-by: NPavel Shilovsky <piastry@etersoft.ru>
Signed-off-by: NSteve French <sfrench@us.ibm.com>

Now we point superblock to a server share root and set a root dentry
appropriately. This let us share superblock between mounts like
//server/sharename/foo/bar and //server/sharename/foo further.
Reviewed-by: NJeff Layton <jlayton@redhat.com>
Signed-off-by: NPavel Shilovsky <piastry@etersoft.ru>
Signed-off-by: NSteve French <sfrench@us.ibm.com>

and simplify error handling code.
Signed-off-by: NPavel Shilovsky <piastry@etersoft.ru>
Signed-off-by: NSteve French <sfrench@us.ibm.com>

    When mandatory encryption is configured in samba server on a
    share (smb.conf parameter "smb encrypt = mandatory") the
    server will hang up the tcp session when we try to send
    the first frame after the tree connect if it is not a
    QueryFSUnixInfo, this causes cifs mount to hang (it must
    be killed with ctl-c).  Move the QueryFSUnixInfo call
    earlier in the mount sequence, and check whether the SetFSUnixInfo
    fails due to mandatory encryption so we can return a sensible
    error (EACCES) on mount.

    In a future patch (for 2.6.40) we will support mandatory
    encryption.

CC: Stable <stable@kernel.org>
Reviewed-by: NPavel Shilovsky <piastry@etersoft.ru>
Signed-off-by: NSteve French <sfrench@us.ibm.com>

We need it to make them work with mandatory locking style because
we can fail in a situation like when kernel need to flush dirty pages
and there is a lock held by a process who opened file.
Signed-off-by: NPavel Shilovsky <piastry@etersoft.ru>
Signed-off-by: NSteve French <sfrench@us.ibm.com>

Add the ability for CIFS to do an asynchronous write. The kernel will
set the frame up as it would for a "normal" SMBWrite2 request, and use
cifs_call_async to send it. The mid callback will then be configured to
handle the result.
Reviewed-by: NPavel Shilovsky <piastry@etersoft.ru>
Signed-off-by: NJeff Layton <jlayton@redhat.com>
Signed-off-by: NSteve French <sfrench@us.ibm.com>

Now that we can handle larger wsizes in writepages, fix up the
negotiation of the wsize to allow for that. find_get_pages only seems to
give out a max of 256 pages at a time, so that gives us a reasonable
default of 1M for the wsize.

If the server however does not support large writes via POSIX
extensions, then we cap the wsize to (128k - PAGE_CACHE_SIZE). That
gives us a size that goes up to the max frame size specified in RFC1001.

Finally, if CAP_LARGE_WRITE_AND_X isn't set, then further cap it to the
largest size allowed by the protocol (USHRT_MAX).
Signed-off-by: NJeff Layton <jlayton@redhat.com>
Reviewed-and-Tested-by: NPavel Shilovsky <piastry@etersoft.ru>
Signed-off-by: NSteve French <sfrench@us.ibm.com>

Have cifs_writepages issue asynchronous writes instead of waiting on
each write call to complete before issuing another. This also allows us
to return more quickly from writepages. It can just send out all of the
I/Os and not wait around for the replies.

In the WB_SYNC_ALL case, if the write completes with a retryable error,
then the completion workqueue job will resend the write.

This also changes the page locking semantics a little bit. Instead of
holding the page lock until the response is received, release it after
doing the send. This will reduce contention for the page lock and should
prevent processes that have the file mmap'ed from being blocked
unnecessarily.
Signed-off-by: NJeff Layton <jlayton@redhat.com>
Reviewed-and-Tested-by: NPavel Shilovsky <piastry@etersoft.ru>
Signed-off-by: NSteve French <sfrench@us.ibm.com>

Fix double kfree() calls on the same pointers and cleanup mount code.
Reviewed-and-Tested-by: NJeff Layton <jlayton@samba.org>
Signed-off-by: NPavel Shilovsky <piastry@etersoft.ru>
Signed-off-by: NSteve French <sfrench@us.ibm.com>

Minor revision to the last version of this patch -- the only difference
is the fix to the cFYI statement in cifs_reconnect.

Holding the spinlock while we call this function means that it can't
sleep, which really limits what it can do. Taking it out from under
the spinlock also means less contention for this global lock.

Change the semantics such that the Global_MidLock is not held when
the callback is called. To do this requires that we take extra care
not to have sync_mid_result remove the mid from the list when the
mid is in a state where that has already happened. This prevents
list corruption when the mid is sitting on a private list for
reconnect or when cifsd is coming down.
Reviewed-by: NShirish Pargaonkar <shirishpargaonkar@gmail.com>
Reviewed-by: NPavel Shilovsky <piastry@etersoft.ru>
Signed-off-by: NJeff Layton <jlayton@redhat.com>
Signed-off-by: NSteve French <sfrench@us.ibm.com>

Reorganize code to get mount option at first and when get a superblock.
This lets us use shared superblock model further for equal mounts.
Signed-off-by: NPavel Shilovsky <piastry@etersoft.ru>
Signed-off-by: NSteve French <sfrench@us.ibm.com>

Use separate functions for comparison between existing structure
and what we are requesting for to make server, session and tcon
search code easier to use on next superblock match call.
Reviewed-by: NJeff Layton <jlayton@samba.org>
Signed-off-by: NPavel Shilovsky <piastry@etersoft.ru>
Signed-off-by: NSteve French <sfrench@us.ibm.com>

There's no SMB2 support in the CIFS filesystem driver, so there's no need to
have a config and mount option for it.
Signed-off-by: NChristoph Hellwig <hch@lst.de>
Signed-off-by: NSteve French <sfrench@us.ibm.com>

The current code always ignores the max_pending limit. Have it instead
only optionally ignore the pending limit. For CIFSSMBEcho, we need to
ignore it to make sure they always can go out. For async reads, writes
and potentially other calls, we need to respect it.
Signed-off-by: NJeff Layton <jlayton@redhat.com>
Signed-off-by: NSteve French <sfrench@us.ibm.com>

We'll need this for async writes, so convert the call to take a kvec
array. CIFSSMBEcho is changed to put a kvec on the stack and pass
in the SMB buffer using that.
Reviewed-by: NPavel Shilovsky <piastry@etersoft.ru>
Signed-off-by: NJeff Layton <jlayton@redhat.com>
Signed-off-by: NSteve French <sfrench@us.ibm.com>

Further consolidate the SendReceive code by moving the checks run over
the packet into a separate function that all the SendReceive variants
can call.

We can also eliminate the check for a receive_len that's too big or too
small. cifs_demultiplex_thread already checks that and disconnects the
socket if that occurs, while setting the midStatus to MALFORMED. It'll
never call this code if that's the case.

Finally do a little cleanup. Use "goto out" on errors so that the flow
of code in the normal case is more evident. Also switch the logErr
variable in map_smb_to_linux_error to a bool.
Reviewed-by: NPavel Shilovsky <piastry@etersoft.ru>
Signed-off-by: NJeff Layton <jlayton@redhat.com>
Signed-off-by: NSteve French <sfrench@us.ibm.com>

Fixes this build error on s390 and probably other archs as well:

  fs/inode.c: In function 'new_inode':
  fs/inode.c:894:2: error: implicit declaration of function 'spin_lock_prefetch'
Signed-off-by: NHeiko Carstens <heiko.carstens@de.ibm.com>
[ Happens on architectures that don't define their own prefetch
  functions in <asm/processor.h>, and instead rely on the default
  ones in <linux/prefetch.h>   - Linus]
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

This solves a serious VFS-level bug in nested_symlink (which was
rewritten from do_follow_link), and follows the order of depth tests
that existed before.

The bug triggers a BUG_ON in fs/namei.c:1381, when running racer with
symlink and rename ops.
Signed-off-by: NErez Zadok <ezk@cs.sunysb.edu>
Acked-by: NMiklos Szeredi <mszeredi@suse.cz>
Cc: stable@kernel.org
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

As Ben Hutchings discovered [1], the patch for CVE-2011-1017 (buffer
overflow in ldm_frag_add) is not sufficient.  The original patch in
commit c340b1d6 ("fs/partitions/ldm.c: fix oops caused by corrupted
partition table") does not consider that, for subsequent fragments,
previously allocated memory is used.

[1] http://lkml.org/lkml/2011/5/6/407Reported-by: NBen Hutchings <ben@decadent.org.uk>
Signed-off-by: NTimo Warns <warns@pre-sense.de>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

Commit e66eed65 ("list: remove prefetching from regular list
iterators") removed the include of prefetch.h from list.h, which
uncovered several cases that had apparently relied on that rather
obscure header file dependency.

So this fixes things up a bit, using

   grep -L linux/prefetch.h $(git grep -l '[^a-z_]prefetchw*(' -- '*.[ch]')
   grep -L 'prefetchw*(' $(git grep -l 'linux/prefetch.h' -- '*.[ch]')

to guide us in finding files that either need <linux/prefetch.h>
inclusion, or have it despite not needing it.

There are more of them around (mostly network drivers), but this gets
many core ones.
Reported-by: NStephen Rothwell <sfr@canb.auug.org.au>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

Fix to earlier "Simplify invalidate part (try #6)" patch
That patch caused problems with connectathon test 5.
Reviewed-by: NJeff Layton <jlayton@samba.org>
Signed-off-by: NPavel Shilovsky <piastry@etersoft.ru>
Signed-off-by: NSteve French <sfrench@us.ibm.com>

Move extern for cifsConvertToUCS to different header to prevent following warning:

CHECK   fs/cifs/cifs_unicode.c
fs/cifs/cifs_unicode.c:267:1: warning: symbol 'cifsConvertToUCS' was not declared. Should it be static?
Signed-off-by: NSteve French <sfrench@us.ibm.com>
Signed-off-by: NPavel Shilovsky <piastry@etersoft.ru>
Signed-off-by: NSteve French <sfrench@us.ibm.com>

Signed-off-by: NSteve French <sfrench@us.ibm.com>

Change idmap key name from cifs.cifs_idmap to cifs.idmap.
Removed unused structure wksidarr and function match_sid().
Handle errors correctly in function init_cifs().
Signed-off-by: NShirish Pargaonkar <shirishpargaonkar@gmail.com>
Reviewed-by: NJeff Layton <jlayton@redhat.com>
Signed-off-by: NSteve French <sfrench@us.ibm.com>

Previously mount options were copied and updated in the cifs_sb_info
struct only when CONFIG_CIFS_DFS_UPCALL was enabled.  Making this
information generally available allows us to remove a number of ifdefs,
extra function params, and temporary variables.
Reviewed-by: NJeff Layton <jlayton@redhat.com>
Signed-off-by: NSean Finney <seanius@seanius.net>
Signed-off-by: NSteve French <sfrench@us.ibm.com>

A relatively minor nit, but also clarified the "consensus" from the
preceding comments that it is in fact better to try for the kstrdup
early and cleanup while cleaning up is still a simple thing to do.
Reviewed-By: NSteve French <smfrench@gmail.com>
Signed-off-by: NSean Finney <seanius@seanius.net>
Signed-off-by: NPavel Shilovsky <piastry@etersoft.ru>
Signed-off-by: NSteve French <sfrench@us.ibm.com>

With CONFIG_DFS_UPCALL enabled, maintain the submount options in
cifs_sb->mountdata, simplifying the code just a bit as well as making
corner-case allocation problems less likely.
Reviewed-by: NJeff Layton <jlayton@redhat.com>
Signed-off-by: NSean Finney <seanius@seanius.net>
Signed-off-by: NPavel Shilovsky <piastry@etersoft.ru>
Signed-off-by: NSteve French <sfrench@us.ibm.com>

To keep strings passed to cifs_parse_mount_options re-usable (which is
needed to clean up the DFS referral handling), tokenize a copy of the
mount options instead.  If values are needed from this tokenized string,
they too must be duplicated (previously, some options were copied and
others duplicated).

Since we are not on the critical path and any cleanup is relatively easy,
the extra memory usage shouldn't be a problem (and it is a bit simpler
than trying to implement something smarter).
Reviewed-by: NJeff Layton <jlayton@redhat.com>
Signed-off-by: NSean Finney <seanius@seanius.net>
Signed-off-by: NPavel Shilovsky <piastry@etersoft.ru>
Signed-off-by: NSteve French <sfrench@us.ibm.com>

Windows 2008 CIFS servers do not always return PATH_NOT_COVERED when
attempting to access a DFS share.  Therefore, when checking for remote
shares, unconditionally ask for a DFS referral for the UNC (w/out prepath)
before continuing with previous behavior of attempting to access the UNC +
prepath and checking for PATH_NOT_COVERED.

Bugzilla: https://bugzilla.kernel.org/show_bug.cgi?id=31092Reviewed-by: NJeff Layton <jlayton@redhat.com>
Signed-off-by: NSean Finney <seanius@seanius.net>
Signed-off-by: NPavel Shilovsky <piastry@etersoft.ru>
Signed-off-by: NSteve French <sfrench@us.ibm.com>

The logic behind the expansion of DFS referrals is now extracted from
cifs_mount into a new static function, expand_dfs_referral.  This will
reduce duplicate code in upcoming commits.
Reviewed-by: NJeff Layton <jlayton@redhat.com>
Signed-off-by: NSean Finney <seanius@seanius.net>
Signed-off-by: NPavel Shilovsky <piastry@etersoft.ru>
Signed-off-by: NSteve French <sfrench@us.ibm.com>

It's a bad idea to have macro functions that reference variables more
than once, as the arguments could have side effects. Turn BCC() into
a static inlined function instead.

While we're at it, make it return a void * to discourage anyone from
dereferencing it as-is.
Reported-and-acked-by: NDavid Howells <dhowells@redhat.com>
Signed-off-by: NJeff Layton <jlayton@redhat.com>
Signed-off-by: NSteve French <sfrench@us.ibm.com>
Signed-off-by: NPavel Shilovsky <piastry@etersoft.ru>
Signed-off-by: NSteve French <sfrench@us.ibm.com>

This is the same patch as originally posted, just with some merge
conflicts fixed up...

Currently, the ByteCount is usually converted to host-endian on receive.
This is confusing however, as we need to keep two sets of routines for
accessing it, and keep track of when to use each routine. Munging
received packets like this also limits when the signature can be
calulated.

Simplify the code by keeping the received ByteCount in little-endian
format. This allows us to eliminate a set of routines for accessing it
and we can now drop the *_le suffixes from the accessor functions since
that's now implied.

While we're at it, switch all of the places that read the ByteCount
directly to use the get_bcc inline which should also clean up some
unaligned accesses.
Signed-off-by: NJeff Layton <jlayton@redhat.com>
Signed-off-by: NSteve French <sfrench@us.ibm.com>

fs/cifs/cifsacl.c: In function ‘id_rb_search’:
fs/cifs/cifsacl.c:215:19: warning: variable ‘linkto’ set but not used
[-Wunused-but-set-variable]
fs/cifs/cifsacl.c:214:18: warning: variable ‘parent’ set but not used
[-Wunused-but-set-variable]
Reviewed-by: NShirish Pargaonkar <shirishpargaonkar@gmail.com>
Signed-off-by: NJeff Layton <jlayton@redhat.com>
Signed-off-by: NSteve French <sfrench@us.ibm.com>

Simplify many places when we call cifs_revalidate/invalidate to make
it do what it exactly needs.
Reviewed-by: NJeff Layton <jlayton@samba.org>
Signed-off-by: NPavel Shilovsky <piastry@etersoft.ru>
Signed-off-by: NSteve French <sfrench@us.ibm.com>

Recently introduced strictcache mode brought a new code that can be
efficiently used by directio part. That's let us add vectored operations
and break unnecessary cifs_user_read and cifs_user_write.
Signed-off-by: NPavel Shilovsky <piastry@etersoft.ru>
Reviewed-by: NJeff Layton <jlayton@redhat.com>
Signed-off-by: NSteve French <sfrench@us.ibm.com>

       There is one big endian field in the cifs protocol, the RFC1001
       length, which cifs code (unlike in the smb2 code) had been handling as
       u32 until the last possible moment, when it was converted to be32 (its
       native form) before sending on the wire.   To remove the last sparse
       endian warning, and to make this consistent with the smb2
       implementation  (which always treats the fields in their
       native size and endianness), convert all uses of smb_buf_length to
       be32.

       This version incorporates Christoph's comment about
       using be32_add_cpu, and fixes a typo in the second
       version of the patch.
Signed-off-by: NSteve French <sfrench@us.ibm.com>
Signed-off-by: NPavel Shilovsky <piastry@etersoft.ru>
Signed-off-by: NSteve French <sfrench@us.ibm.com>

rb tree search and insertion routines.

A SID which needs to be mapped, is looked up in one of the rb trees
depending on whether SID is either owner or group SID.
If found in the tree, a (mapped) id from that node is assigned to
uid or gid as appropriate.  If unmapped, an upcall is attempted to
map the SID to an id.  If upcall is successful, node is marked as
mapped.  If upcall fails, node stays marked as unmapped and a mapping
is attempted again only after an arbitrary time period has passed.

To map a SID, which can be either a Owner SID or a Group SID, key
description starts with the string "os" or "gs" followed by SID converted
to a string. Without "os" or "gs", cifs.upcall does not know whether
SID needs to be mapped to either an uid or a gid.

Nodes in rb tree have fields to prevent multiple upcalls for
a SID.  Searching, adding, and removing nodes is done within global locks.
Whenever a node is either found or inserted in a tree, a reference
is taken on that node.
Shrinker routine prunes a node if it has expired but does not prune
an expired node if its refcount is not zero (i.e. sid/id of that node
is_being/will_be accessed).
Thus a node, if its SID needs to be mapped by making an upcall,
can safely stay and its fields accessed without shrinker pruning it.
A reference (refcount) is put on the node without holding the spinlock
but a reference is get on the node by holding the spinlock.

Every time an existing mapped node is accessed or mapping is attempted,
its timestamp is updated to prevent it from getting erased or a
to prevent multiple unnecessary repeat mapping retries respectively.

For now, cifs.upcall is only used to map a SID to an id (uid or gid) but
it would be used to obtain an SID for an id.
Signed-off-by: NShirish Pargaonkar <shirishpargaonkar@gmail.com>
Reviewed-by: NJeff Layton <jlayton@redhat.com>
Signed-off-by: NSteve French <sfrench@us.ibm.com>

Define (global) data structures to store ids, uids and gids, to which a
SID maps.  There are two separate trees, one for SID/uid and another one
for SID/gid.

A new type of key, cifs_idmap_key_type, is used.

Keys are instantiated and searched using credential of the root by
overriding and restoring the credentials of the caller requesting the key.

Id mapping functions are invoked under config option of cifs acl.
Signed-off-by: NShirish Pargaonkar <shirishpargaonkar@gmail.com>
Signed-off-by: NSteve French <sfrench@us.ibm.com>

Add this let us drop filemap_write_and_wait from cifs_invalidate_mapping
and simplify the code to properly process invalidate logic.
Reviewed-by: NJeff Layton <jlayton@redhat.com>
Signed-off-by: NPavel Shilovsky <piastry@etersoft.ru>
Signed-off-by: NSteve French <sfrench@us.ibm.com>