Under memory pressure, the system may evict dentries from cache.  When the
FAT driver receives a NFS request involving an evicted dentry, it is
unable to reconnect it to the filesystem root.  This causes the request to
fail, often with ENOENT.

This is partially due to ineffectiveness of the current FAT NFS
implementation, and partially due to an unimplemented fh_to_parent method.
 The latter can cause file accesses to fail on shares exported with
subtree_check.

This patch set provides the FAT driver with the ability to
reconnect dentries.  NFS file handle generation and lookups are simplified
and made congruent with ext2.

Testing has involved a memory-starved virtual machine running 3.5-rc5 that
exports a ~2 GB vfat filesystem containing a kernel tree (~770 MB, ~40000
files, 9 levels).  Both 'cp -r' and 'ls -lR' operations were performed
from a client, some overlapping, some consecutive.  Exports with
'subtree_check' and 'no_subtree_check' have been tested.

Note that while this patch set improves FAT's NFS support, it does not
eliminate ESTALE errors completely.

The following should be considered for NFS clients who are sensitive to ESTALE:

* Mounting with lookupcache=none
  Unfortunately this can degrade performance severely, particularly for deep
  filesystems.

* Incorporating VFS patches to retry ESTALE failures on the client-side,
  such as https://lkml.org/lkml/2012/6/29/381

* Handling ESTALE errors in client application code

This patch:

Move NFS-related code into its own C file.  No functional changes.
Signed-off-by: NSteven J. Magnani <steve@digidescorp.com>
Acked-by: NOGAWA Hirofumi <hirofumi@mail.parknet.co.jp>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

There's no reason to call rcu_barrier() on every
deactivate_locked_super().  We only need to make sure that all delayed rcu
free inodes are flushed before we destroy related cache.

Removing rcu_barrier() from deactivate_locked_super() affects some fast
paths.  E.g.  on my machine exit_group() of a last process in IPC
namespace takes 0.07538s.  rcu_barrier() takes 0.05188s of that time.
Signed-off-by: NKirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NAl Viro <viro@zeniv.linux.org.uk>

Acked-by: NOGAWA Hirofumi <hirofumi@mail.parknet.co.jp>
Acked-by: NSerge Hallyn <serge.hallyn@canonical.com>
Signed-off-by: NEric W. Biederman <ebiederm@xmission.com>

Simplify code by providing accessor functions for the directory entry
start cluster fields.
Signed-off-by: NSteven J. Magnani <steve@digidescorp.com>
Acked-by: NOGAWA Hirofumi <hirofumi@mail.parknet.co.jp>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

fat_encode_fh() can fetch an invalid i_pos value on systems where 64-bit
accesses are not atomic.  Make it use the same accessor as the rest of the
FAT code.
Signed-off-by: NSteven J. Magnani <steve@digidescorp.com>
Acked-by: NOGAWA Hirofumi <hirofumi@mail.parknet.co.jp>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

Currently FAT file-system maps the VFS "superblock" abstraction to the
FSINFO block.  The FSINFO block contains non-essential data about the
amount of free clusters and the next free cluster.  FAT file-system can
always find out this information by scanning the FAT table, but having it
in the FSINFO block may speed things up sometimes.  So FAT file-system
relies on the VFS superblock write-out services to make sure the FSINFO
block is written out to the media from time to time.

The whole "superblock write-out" VFS infrastructure is served by the
'sync_supers()' kernel thread, which wakes up every 5 (by default) seconds
and writes out all dirty superblock using the '->write_super()' call-back.
 But the problem with this thread is that it wastes power by waking up the
system every 5 seconds no matter what.  So we want to kill it completely
and thus, we need to make file-systems to stop using the '->write_super'
VFS service, and then remove it together with the kernel thread.

This patch switches the FAT FSINFO block management from
'->write_super()'/'->s_dirt' to 'fsinfo_inode'/'->write_inode'.  Now,
instead of setting the 's_dirt' flag, we just mark the special
'fsinfo_inode' inode as dirty and let VFS invoke the '->write_inode'
call-back when needed, where we write-out the FSINFO block.

This patch also makes sure we do not mark the 'fsinfo_inode' inode as
dirty if we are not FAT32 (FAT16 and FAT12 do not have the FSINFO block)
or if we are in R/O mode.

As a bonus, we can also remove the '->sync_fs()' and '->write_super()' FAT
call-back function because they become unneeded.
Signed-off-by: NArtem Bityutskiy <artem.bityutskiy@linux.intel.com>
Cc: OGAWA Hirofumi <hirofumi@mail.parknet.co.jp>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

This is patchset makes fatfs stop using the VFS '->write_super()' method
for writing out the FSINFO block.

The final goal is to get rid of the 'sync_supers()' kernel thread.  This
kernel thread wakes up every 5 seconds (by default) and calls
'->write_super()' for all mounted file-systems.  And the bad thing is that
this is done even if all the superblocks are clean.  Moreover, some
file-systems do not even need this end they do not register the
'->write_super()' method at all (e.g., btrfs).

So 'sync_supers()' most often just generates useless wake-ups and wastes
power.  I am trying to make all file-systems independent of
'->write_super()' and plan to remove 'sync_supers()' and '->write_super'
completely once there are no more users.

The '->write_supers()' method is mostly used by baroque file-systems like
hfs, udf, etc.  Modern file-systems like btrfs and xfs do not use it.
This justifies removing this stuff from VFS completely and make every FS
self-manage own superblock.

Tested with xfstests.

This patch:

Preparation for further changes.  It introduces a special inode
('fsinfo_inode') in FAT file-system which we'll later use for managing the
FSINFO block.  Note, this there is already one special inode ('fat_inode')
which is used for managing the FAT tables.

Introduce new 'MSDOS_FSINFO_INO' constant for this special inode.  It is
safe to do because FAT file-system does not store inode numbers on the
media but generates them run-time.

I've also cleaned up the comment to existing 'MSDOS_ROOT_INO' constant,
while on it.
Signed-off-by: NArtem Bityutskiy <artem.bityutskiy@linux.intel.com>
Cc: OGAWA Hirofumi <hirofumi@mail.parknet.co.jp>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

pass inode + parent's inode or NULL instead of dentry + bool saying
whether we want the parent or not.

NOTE: that needs ceph fix folded in.
Signed-off-by: NAl Viro <viro@zeniv.linux.org.uk>

After we moved inode_sync_wait() from end_writeback() it doesn't make sense
to call the function end_writeback() anymore. Rename it to clear_inode()
which well says what the function really does - set I_CLEAR flag.
Signed-off-by: NJan Kara <jack@suse.cz>
Signed-off-by: NFengguang Wu <fengguang.wu@intel.com>

Signed-off-by: NAl Viro <viro@zeniv.linux.org.uk>

Signed-off-by: NAl Viro <viro@zeniv.linux.org.uk>

Seeing that just about every destructor got that INIT_LIST_HEAD() copied into
it, there is no point whatsoever keeping this INIT_LIST_HEAD in inode_init_once();
the cost of taking it into inode_init_always() will be negligible for pipes
and sockets and negative for everything else.  Not to mention the removal of
boilerplate code from ->destroy_inode() instances...
Signed-off-by: NAl Viro <viro@zeniv.linux.org.uk>

Signed-off-by: NGeert Uytterhoeven <geert@linux-m68k.org>
Signed-off-by: NJiri Kosina <jkosina@suse.cz>

Replace remaining direct i_nlink updates with a new set_nlink()
updater function.
Signed-off-by: NMiklos Szeredi <mszeredi@suse.cz>
Tested-by: NToshiyuki Okajima <toshi.okajima@jp.fujitsu.com>
Signed-off-by: NChristoph Hellwig <hch@lst.de>

FAT16 support maximum 4GB vol/file size with 64KB cluster size.

Win NT/XP/7 increased the maximum cluster size to 64KB, and file/vol
size increased 4GB also.  Although increasing, the file size of linux
FAT is still limited at 2GB.

I found that it is limited by sb->maxbytes(0x7fffffff) when partition
is formatted by FAT16.  sb->s_maxbytes in fill_super should be set to
0xffffffff like fat32.
Signed-off-by: NNamjae Jeon <linkinjeon@gmail.com>
Signed-off-by: NOGAWA Hirofumi <hirofumi@mail.parknet.co.jp>

The fat_msg function already formats the given message and appends
a newline to it - we don't need to do this in the passed message
string as well, or will end up with a blank line printed in the
kernel log ring buffer.

Also change the loglevel from error to warning.
Signed-off-by: NMihai Moldovan <ionic@ionic.de>
Signed-off-by: NOGAWA Hirofumi <hirofumi@mail.parknet.co.jp>

Simple filesystems always pass inode->i_sb_bdev as the block device
argument, and never need a end_io handler.  Let's simply things for
them and for my grepping activity by dropping these arguments.  The
only thing not falling into that scheme is ext4, which passes and
end_io handler without needing special flags (yet), but given how
messy the direct I/O code there is use of __blockdev_direct_IO
in one instead of two out of three cases isn't going to make a large
difference anyway.
Signed-off-by: NChristoph Hellwig <hch@lst.de>
Signed-off-by: NAl Viro <viro@zeniv.linux.org.uk>

Add a new rw_semaphore to protect bmap against truncate.  Previous
i_alloc_sem was abused for this, but it's going away in this series.

Note that we can't simply use i_mutex, given that the swapon code
calls ->bmap under it.
Signed-off-by: NChristoph Hellwig <hch@lst.de>
Signed-off-by: NAl Viro <viro@zeniv.linux.org.uk>

pathconf(, _PC_NAME_MAX) is too small for long Unicode filename on fat.

255 as max filename size on fat is Unicode UTF-16 characters.
it's not byte size.

https://bugzilla.kernel.org/show_bug.cgi?id=16469

To fix it, this returns "len * NLS_MAX_CHARSET_SIZE" instead.
Reported-by: NTakumi Asaki <takumi.asaki@nokia.com>
Signed-off-by: NOGAWA Hirofumi <hirofumi@mail.parknet.co.jp>

Replace all printk with fat_msg()
Signed-off-by: NAlexey Fisher <bug-track@fisher-privat.net>
Signed-off-by: NOGAWA Hirofumi <hirofumi@mail.parknet.co.jp>

Signed-off-by: NOGAWA Hirofumi <hirofumi@mail.parknet.co.jp>

The exportfs encode handle function should return the minimum required
handle size. This helps user to find out the handle size by passing 0
handle size in the first step and then redoing to the call again with
the returned handle size value.
Acked-by: NSerge Hallyn <serue@us.ibm.com>
Signed-off-by: NAneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Signed-off-by: NAl Viro <viro@zeniv.linux.org.uk>

Code has been converted over to the new explicit on-stack plugging,
and delay users have been converted to use the new API for that.
So lets kill off the old plugging along with aops->sync_page().
Signed-off-by: NJens Axboe <jaxboe@fusionio.com>

don't bother with lock_super() in fat_fill_super() callers, while
we are at it - there won't be any concurrency anyway.
Signed-off-by: NAl Viro <viro@zeniv.linux.org.uk>

Reduce some branches and memory accesses in dcache lookup by adding dentry
flags to indicate common d_ops are set, rather than having to check them.
This saves a pointer memory access (dentry->d_op) in common path lookup
situations, and saves another pointer load and branch in cases where we
have d_op but not the particular operation.

Patched with:

git grep -E '[.>]([[:space:]])*d_op([[:space:]])*=' | xargs sed -e 's/\([^\t ]*\)->d_op = \(.*\);/d_set_d_op(\1, \2);/' -e 's/\([^\t ]*\)\.d_op = \(.*\);/d_set_d_op(\&\1, \2);/' -i
Signed-off-by: NNick Piggin <npiggin@kernel.dk>

RCU free the struct inode. This will allow:

- Subsequent store-free path walking patch. The inode must be consulted for
  permissions when walking, so an RCU inode reference is a must.
- sb_inode_list_lock to be moved inside i_lock because sb list walkers who want
  to take i_lock no longer need to take sb_inode_list_lock to walk the list in
  the first place. This will simplify and optimize locking.
- Could remove some nested trylock loops in dcache code
- Could potentially simplify things a bit in VM land. Do not need to take the
  page lock to follow page->mapping.

The downsides of this is the performance cost of using RCU. In a simple
creat/unlink microbenchmark, performance drops by about 10% due to inability to
reuse cache-hot slab objects. As iterations increase and RCU freeing starts
kicking over, this increases to about 20%.

In cases where inode lifetimes are longer (ie. many inodes may be allocated
during the average life span of a single inode), a lot of this cache reuse is
not applicable, so the regression caused by this patch is smaller.

The cache-hot regression could largely be avoided by using SLAB_DESTROY_BY_RCU,
however this adds some complexity to list walking and store-free path walking,
so I prefer to implement this at a later date, if it is shown to be a win in
real situations. I haven't found a regression in any non-micro benchmark so I
doubt it will be a problem.
Signed-off-by: NNick Piggin <npiggin@kernel.dk>

The lock_kernel in fat_put_super is not needed because
it only protects the super block itself and we know that
no other thread can reach it because we are about to
kfree the object.

In the two fill_super functions, this converts the locking
to use lock_super like elsewhere in the fat code. This
is probably not needed either, but is consistent and puts
us on the safe side.
Signed-off-by: NArnd Bergmann <arnd@arndb.de>
Cc: OGAWA Hirofumi <hirofumi@mail.parknet.co.jp>
Cc: Jan Blunck <jblunck@infradead.org>

Signed-off-by: NAl Viro <viro@zeniv.linux.org.uk>

Move the call to vmtruncate to get rid of accessive blocks to the callers
in preparation of the new truncate sequence and rename the non-truncating
version to cont_write_begin.
Signed-off-by: NChristoph Hellwig <hch@lst.de>
Signed-off-by: NAl Viro <viro@zeniv.linux.org.uk>

Move the call to vmtruncate to get rid of accessive blocks to the callers
in prepearation of the new truncate calling sequence.  This was only done
for DIO_LOCKING filesystems, so the __blockdev_direct_IO_newtrunc variant
was not needed anyway.  Get rid of blockdev_direct_IO_no_locking and
its _newtrunc variant while at it as just opencoding the two additional
paramters is shorted than the name suffix.
Signed-off-by: NChristoph Hellwig <hch@lst.de>
Signed-off-by: NAl Viro <viro@zeniv.linux.org.uk>

Cc: OGAWA Hirofumi <hirofumi@mail.parknet.co.jp>
Cc: Christoph Hellwig <hch@lst.de>
Signed-off-by: NNick Piggin <npiggin@suse.de>
Signed-off-by: NAl Viro <viro@zeniv.linux.org.uk>

Signed-off-by: NOGAWA Hirofumi <hirofumi@mail.parknet.co.jp>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

Other users doesn't check NULL explicitly. So, these doesn't also
check to remove inconsistency.
Signed-off-by: NOGAWA Hirofumi <hirofumi@mail.parknet.co.jp>

This gives the filesystem more information about the writeback that
is happening.  Trond requested this for the NFS unstable write handling,
and other filesystems might benefit from this too by beeing able to
distinguish between the different callers in more detail.
Signed-off-by: NChristoph Hellwig <hch@lst.de>
Signed-off-by: NAl Viro <viro@zeniv.linux.org.uk>

I found that the length of a file name when created cannot exceed 255
characters, yet, pathconf(), via statfs(), returns the maximum as 260.
Signed-off-by: NKevin Dankwardt <k@kcomputing.com>
Signed-off-by: NOGAWA Hirofumi <hirofumi@mail.parknet.co.jp>

Currently shipping discard capable SSDs and arrays have rather sub-optimal
implementations of the command and can the use of it can cause massive
slowdowns.  Make issueing these commands option as it's already in btrfs
and gfs2.
Signed-off-by: NChristoph Hellwig <hch@lst.de>
[hirofumi@mail.parknet.co.jp: tweaks, and add "discard" to fat_show_options]
Signed-off-by: NOGAWA Hirofumi <hirofumi@mail.parknet.co.jp>

Most call sites of unload_nls() do:
	if (nls)
		unload_nls(nls);

Check the pointer inside unload_nls() like we do in kfree() and
simplify the call sites.
Signed-off-by: NThomas Gleixner <tglx@linutronix.de>
Cc: Steve French <sfrench@us.ibm.com>
Cc: OGAWA Hirofumi <hirofumi@mail.parknet.co.jp>
Cc: Roman Zippel <zippel@linux-m68k.org>
Cc: Dave Kleikamp <shaggy@linux.vnet.ibm.com>
Cc: Petr Vandrovec <vandrove@vc.cvut.cz>
Cc: Anton Altaparmakov <aia21@cantab.net>
Signed-off-by: NAl Viro <viro@zeniv.linux.org.uk>

If we didn't check sb->s_dirt, it will update the FSINFO
unconditionally. It will reduce the filetime of flash base device.

So, this checks sb->s_dirt. sb->s_dirt is racy, however FSINFO is just
hint. So even if there is race, and we hit it, it would not become big
problem.

And this also is as workaround of suspend problem.
Signed-off-by: NOGAWA Hirofumi <hirofumi@mail.parknet.co.jp>

Because, with "shortname=lower", copying one FAT filesystem tree to
another FAT filesystem tree using Linux results in semantically
different filesystems. (E.g.: Filenames which were once "all
uppercase" are now "all lowercase").

So, this changes the default of "shortname=lower" to "shortname=mixed".
Signed-off-by: NPaul Wise <pabs3@bonedaddy.net>
[change fat_show_options()]
Signed-off-by: NOGAWA Hirofumi <hirofumi@mail.parknet.co.jp>

(ce3b0f8d: New helper - current_umask())
is removing the opts->fs_dmask, probably it's a cut-and-paste
miss or something.
Signed-off-by: NOGAWA Hirofumi <hirofumi@mail.parknet.co.jp>