Acked-by: NSerge Hallyn <serge.hallyn@canonical.com>
Signed-off-by: NEric W. Biederman <ebiederm@xmission.com>

The conversion of all of the users is not done yet there are too many to change
in one go and leave the code reviewable. For now I change just the header and
a few trivial users and rely on CONFIG_UIDGID_STRICT_TYPE_CHECKS not being set
to ensure that the code will still compile during the transition.

Helper functions i_uid_read, i_uid_write, i_gid_read, i_gid_write are added
so that in most cases filesystems can avoid the complexities of multiple user
namespaces and can concentrate on moving their raw numeric values into and
out of the vfs data structures.
Acked-by: NSerge Hallyn <serge.hallyn@canonical.com>
Signed-off-by: NEric W. Biederman <ebiederm@xmission.com>

- Use uid_eq when comparing kuids
  Use gid_eq when comparing kgids
- Use make_kuid(user_ns, 0) to talk about the user_namespace root uid
Acked-by: NSerge Hallyn <serge.hallyn@canonical.com>
Signed-off-by: NEric W. Biederman <ebiederm@xmission.com>

cred.h and a few trivial users of struct cred are changed.  The rest of the users
of struct cred are left for other patches as there are too many changes to make
in one go and leave the change reviewable.  If the user namespace is disabled and
CONFIG_UIDGID_STRICT_TYPE_CHECKS are disabled the code will contiue to compile
and behave correctly.
Acked-by: NSerge Hallyn <serge.hallyn@canonical.com>
Signed-off-by: NEric W. Biederman <ebiederm@xmission.com>

As a first step to converting struct cred to be all kuid_t and kgid_t
values convert the group values stored in group_info to always be
kgid_t values.   Unless user namespaces are used this change should
have no effect.
Acked-by: NSerge Hallyn <serge.hallyn@canonical.com>
Signed-off-by: NEric W. Biederman <ebiederm@xmission.com>

- Convert the old uid mapping functions into compatibility wrappers
- Add a uid/gid mapping layer from user space uid and gids to kernel
  internal uids and gids that is extent based for simplicty and speed.
  * Working with number space after mapping uids/gids into their kernel
    internal version adds only mapping complexity over what we have today,
    leaving the kernel code easy to understand and test.
- Add proc files /proc/self/uid_map /proc/self/gid_map
  These files display the mapping and allow a mapping to be added
  if a mapping does not exist.
- Allow entering the user namespace without a uid or gid mapping.
  Since we are starting with an existing user our uids and gids
  still have global mappings so are still valid and useful they just don't
  have local mappings.  The requirement for things to work are global uid
  and gid so it is odd but perfectly fine not to have a local uid
  and gid mapping.
  Not requiring global uid and gid mappings greatly simplifies
  the logic of setting up the uid and gid mappings by allowing
  the mappings to be set after the namespace is created which makes the
  slight weirdness worth it.
- Make the mappings in the initial user namespace to the global
  uid/gid space explicit.  Today it is an identity mapping
  but in the future we may want to twist this for debugging, similar
  to what we do with jiffies.
- Document the memory ordering requirements of setting the uid and
  gid mappings.  We only allow the mappings to be set once
  and there are no pointers involved so the requirments are
  trivial but a little atypical.

Performance:

In this scheme for the permission checks the performance is expected to
stay the same as the actuall machine instructions should remain the same.

The worst case I could think of is ls -l on a large directory where
all of the stat results need to be translated with from kuids and
kgids to uids and gids.  So I benchmarked that case on my laptop
with a dual core hyperthread Intel i5-2520M cpu with 3M of cpu cache.

My benchmark consisted of going to single user mode where nothing else
was running. On an ext4 filesystem opening 1,000,000 files and looping
through all of the files 1000 times and calling fstat on the
individuals files.  This was to ensure I was benchmarking stat times
where the inodes were in the kernels cache, but the inode values were
not in the processors cache.  My results:

v3.4-rc1:         ~= 156ns (unmodified v3.4-rc1 with user namespace support disabled)
v3.4-rc1-userns-: ~= 155ns (v3.4-rc1 with my user namespace patches and user namespace support disabled)
v3.4-rc1-userns+: ~= 164ns (v3.4-rc1 with my user namespace patches and user namespace support enabled)

All of the configurations ran in roughly 120ns when I performed tests
that ran in the cpu cache.

So in summary the performance impact is:
1ns improvement in the worst case with user namespace support compiled out.
8ns aka 5% slowdown in the worst case with user namespace support compiled in.
Acked-by: NSerge Hallyn <serge.hallyn@canonical.com>
Signed-off-by: NEric W. Biederman <ebiederm@xmission.com>

Modify alloc_uid to take a kuid and make the user hash table global.
Stop holding a reference to the user namespace in struct user_struct.

This simplifies the code and makes the per user accounting not
care about which user namespace a uid happens to appear in.
Acked-by: NSerge Hallyn <serge.hallyn@canonical.com>
Signed-off-by: NEric W. Biederman <ebiederm@xmission.com>

This represents a change in strategy of how to handle user namespaces.
Instead of tagging everything explicitly with a user namespace and bulking
up all of the comparisons of uids and gids in the kernel,  all uids and gids
in use will have a mapping to a flat kuid and kgid spaces respectively.  This
allows much more of the existing logic to be preserved and in general
allows for faster code.

In this new and improved world we allow someone to utiliize capabilities
over an inode if the inodes owner mapps into the capabilities holders user
namespace and the user has capabilities in their user namespace.  Which
is simple and efficient.

Moving the fs uid comparisons to be comparisons in a flat kuid space
follows in later patches, something that is only significant if you
are using user namespaces.
Acked-by: NSerge Hallyn <serge.hallyn@canonical.com>
Signed-off-by: NEric W. Biederman <ebiederm@xmission.com>

Optimize performance and prepare for the removal of the user_ns reference
from user_struct.  Remove the slow long walk through cred->user->user_ns and
instead go straight to cred->user_ns.
Acked-by: NSerge Hallyn <serge.hallyn@canonical.com>
Signed-off-by: NEric W. Biederman <ebiederm@xmission.com>

Safely making device nodes in a container is solvable but simply
having the capability in a user namespace is not sufficient to make
this work.
Acked-by: NSerge Hallyn <serge.hallyn@canonical.com>
Signed-off-by: NEric W. Biederman <ebiederm@xmission.com>

64252c75 "vfs: remove dget() from
dentry_unhash()" changed the implementation but not the comment.

Cc: Sage Weil <sage@newdream.net>
Signed-off-by: NJ. Bruce Fields <bfields@redhat.com>
Signed-off-by: NAl Viro <viro@zeniv.linux.org.uk>

Split __lookup_hash into two component functions:

 lookup_dcache - tries cached lookup, returns whether real lookup is needed
 lookup_real - calls i_op->lookup

This eliminates code duplication between d_alloc_and_lookup() and
d_inode_lookup().
Signed-off-by: NMiklos Szeredi <mszeredi@suse.cz>
Signed-off-by: NAl Viro <viro@zeniv.linux.org.uk>

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

now we have __lookup_hash() open-coded if !dentry case;
just call the damn thing instead...
Signed-off-by: NAl Viro <viro@zeniv.linux.org.uk>

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

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

Reorder if-else cases for starters...
Signed-off-by: NAl Viro <viro@zeniv.linux.org.uk>

Everything arriving into if (!dentry) will have need_reval = 1.
Indeed, the only way to get there with need_reval reset to 0 would
be via
	if (unlikely(d_need_lookup(dentry)))
		goto unlazy;
	if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) {
		status = d_revalidate(dentry, nd);
	if (unlikely(status <= 0)) {
		if (status != -ECHILD)
			need_reval = 0;
		goto unlazy;
...
unlazy:
	/* no assignments to dentry */
	if (dentry && unlikely(d_need_lookup(dentry))) {
		dput(dentry);
		dentry = NULL;
	}
and if d_need_lookup() had already been false the first time around, it
will remain false on the second call as well.
Signed-off-by: NAl Viro <viro@zeniv.linux.org.uk>

d_lookup() *will* fail after successful d_invalidate(), if we are
holding i_mutex all along.  IOW, we don't need to jump back to
l: - we know what path will be taken there and can do that (i.e.
d_alloc_and_lookup()) directly.
Signed-off-by: NAl Viro <viro@zeniv.linux.org.uk>

keep holding ->i_mutex over revalidation parts
Signed-off-by: NAl Viro <viro@zeniv.linux.org.uk>

Duplicate the revalidation-related parts into if (!dentry) branch.
Next step will be to pull them under i_mutex.

This and the next 8 commits are more or less a splitup of patch
by Miklos; folks, when you are working with something that convoluted,
carve your patches up into easily reviewed steps, especially when
a lot of codepaths involved are rarely hit...
Signed-off-by: NAl Viro <viro@zeniv.linux.org.uk>

The only caller of __lookup_hash() that needs the exec permission check on
parent is lookup_one_len().

All lookup_hash() callers already checked permission in LOOKUP_PARENT walk.
Signed-off-by: NMiklos Szeredi <mszeredi@suse.cz>
Signed-off-by: NAl Viro <viro@zeniv.linux.org.uk>

__lookup_hash() calls ->lookup() if the dentry needs lookup and on success
revalidates the dentry (all under dir->i_mutex).

While this is harmless it doesn't make a lot of sense.
Signed-off-by: NMiklos Szeredi <mszeredi@suse.cz>
Signed-off-by: NAl Viro <viro@zeniv.linux.org.uk>

Doing revalidate on a dentry which has not yet been looked up makes no sense.

Move the d_need_lookup() check before d_revalidate().
Signed-off-by: NMiklos Szeredi <mszeredi@suse.cz>
Signed-off-by: NAl Viro <viro@zeniv.linux.org.uk>

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

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

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

move mode-dependent parts to callers, kill unused arguments
Signed-off-by: NAl Viro <viro@zeniv.linux.org.uk>

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

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

This reverts commit b43d17f3.

Dave Jones reports that it causes lockups on his laptop, and his debug
output showed a lot of processes hung waiting for page_writeback (or
more commonly - processes hung waiting for a lock that was held during
that writeback wait).

The page_writeback hint made Ted suggest that Dave look at this commit,
and Dave verified that reverting it makes his problems go away.

Ted says:
 "That commit fixes a race which is seen when you write into fallocated
  (and hence uninitialized) disk blocks under *very* heavy memory
  pressure.  Furthermore, although theoretically it could trigger under
  normal direct I/O writes, it only seems to trigger if you are issuing
  a huge number of AIO writes, such that a just-written page can get
  evicted from memory, and then read back into memory, before the
  workqueue has a chance to update the extent tree.

  This race has been around for a little over a year, and no one noticed
  until two months ago; it only happens under fairly exotic conditions,
  and in fact even after trying very hard to create a simple repro under
  lab conditions, we could only reproduce the problem and confirm the
  fix on production servers running MySQL on very fast PCIe-attached
  flash devices.

  Given that Dave was able to hit this problem pretty quickly, if we
  confirm that this commit is at fault, the only reasonable thing to do
  is to revert it IMO."
Reported-and-tested-by: NDave Jones <davej@redhat.com>
Acked-by: NTheodore Ts'o <tytso@mit.edu>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

Commit 025c5b24 ("thp: optimize away unnecessary page table
locking") moves spin_lock() into pmd_trans_huge_lock() in order to avoid
locking unless pmd is for thp.  So this spin_lock() is a bug.
Reported-by: NSasha Levin <levinsasha928@gmail.com>
Signed-off-by: NNaoya Horiguchi <n-horiguchi@ah.jp.nec.com>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

Dave Sterba had put in patches to look for mixed data/metadata groups
with metadata bigger than 4KB.  But these ended up in the wrong place
and it wasn't testing the feature flag correctly.

This updates the tests to make sure our sizes are matching
Signed-off-by: NChris Mason <chris.mason@oracle.com>

When we use autodefrag, we forget to update the index which indicates
the last page we've dirty.  And we'll set dirty flags on a same set of
pages again and again.
Signed-off-by: NLiu Bo <liubo2009@cn.fujitsu.com>
Signed-off-by: NChris Mason <chris.mason@oracle.com>

$ mkfs.btrfs /dev/sdb7
$ mount /dev/sdb7 /mnt/btrfs/ -oautodefrag
$ dd if=/dev/zero of=/mnt/btrfs/foobar bs=4k count=10 oflag=direct 2>/dev/null
$ filefrag -v /mnt/btrfs/foobar
Filesystem type is: 9123683e
File size of /mnt/btrfs/foobar is 40960 (10 blocks, blocksize 4096)
 ext logical physical expected length flags
   0       0     3072              10 eof
/mnt/btrfs/foobar: 1 extent found

Now we have a big real extent [0, 40960), but autodefrag will still defrag it.

$ sync
$ filefrag -v /mnt/btrfs/foobar
Filesystem type is: 9123683e
File size of /mnt/btrfs/foobar is 40960 (10 blocks, blocksize 4096)
 ext logical physical expected length flags
   0       0     3082              10 eof
/mnt/btrfs/foobar: 1 extent found

So if we already find a big real extent, we're ok about that, just skip it.
Signed-off-by: NLiu Bo <liubo2009@cn.fujitsu.com>
Signed-off-by: NChris Mason <chris.mason@oracle.com>

If our file's layout is as follows:
| hole | data1 | hole | data2 |

we do not need to defrag this file, because this file has holes and
cannot be merged into one extent.
Signed-off-by: NLiu Bo <liubo2009@cn.fujitsu.com>
Signed-off-by: NChris Mason <chris.mason@oracle.com>

$ mkfs.btrfs disk
$ mount disk /mnt -o autodefrag
$ dd if=/dev/zero of=/mnt/foobar bs=4k count=10 2>/dev/null && sync
$ for i in `seq 9 -2 0`; do dd if=/dev/zero of=/mnt/foobar bs=4k count=1 \
  seek=$i conv=notrunc 2> /dev/null; done && sync

then we'll get to defrag "foobar" again and again.
So does option "-o autodefrag,compress".

Reasons:
When the cleaner kthread gets to fetch inodes from the defrag tree and defrag
them, it will dirty pages and submit them, this will comes to another DATA COW
where the processing inode will be inserted to the defrag tree again.

This patch sets a rule for COW code, i.e. insert an inode when we're really
going to make some defragments.
Signed-off-by: NLiu Bo <liubo2009@cn.fujitsu.com>
Signed-off-by: NChris Mason <chris.mason@oracle.com>

commit 600a45e1
(Btrfs: fix deadlock on page lock when doing auto-defragment)
fixes the deadlock on page, but it also introduces another bug.

A page may have been truncated after unlock & lock.
So we need to find it again to get the right one.

And since we've held i_mutex lock, inode size remains unchanged and
we can drop isize overflow checks.
Signed-off-by: NLiu Bo <liubo2009@cn.fujitsu.com>
Signed-off-by: NMiao Xie <miaox@cn.fujitsu.com>
Signed-off-by: NChris Mason <chris.mason@oracle.com>

The bug is from running xfstests 209 with autodefrag.

The race is as follows:
       t1                       t2(autodefrag)
   direct IO
     invalidate pagecache
     dio(old data)             add_inode_defrag
     invalidate pagecache
   endio

   direct IO
     invalidate pagecache
                                run_defrag
                                  readpage(old data)
                                  set page dirty (old data)
     dio(new data, rewrite)
     invalidate pagecache (*)
     endio

t2(autodefrag) will get old data into pagecache via readpage and set
pagecache dirty.  Meanwhile, invalidate pagecache(*) will fail due to
dirty flags in pages.  So the old data may be flushed into disk by
flush thread, which will lead to data loss.

And so does the case of user defragment progs.

The patch fixes this race by holding i_mutex when we readpage and set page dirty.
Signed-off-by: NLiu Bo <liubo2009@cn.fujitsu.com>
Signed-off-by: NMiao Xie <miaox@cn.fujitsu.com>
Signed-off-by: NChris Mason <chris.mason@oracle.com>

This deadlock comes from xfstests 251.

We'll hold the chunk_mutex throughout the whole of a chunk allocation.
But if we find that we've used up system chunk space, we need to allocate a
new system chunk, but this will lead to a recursion of chunk allocation and end
up with a deadlock on chunk_mutex.
So instead we need to allocate the system chunk first if we find we're in ENOSPC.
Signed-off-by: NLiu Bo <liubo2009@cn.fujitsu.com>
Signed-off-by: NChris Mason <chris.mason@oracle.com>