This problem was found when running 'tests/tst-zfs-mount.so' multiple times.
At the first time, all tests succeed, however, a subsequent run would
fail at the test: 'mkdir /foo/bar', the error message reported
that the target file already exists.

The test basically creates a directory /foo/bar, rename it to /foo/bar2,
then remove /foo/bar2. How could /foo/bar still be there?

Quite simple. Our shutdown function calls unmount_rootfs() which will
attempt to unmount zfs with the flag MNT_FOURCE, however, it's not being
passed to zfs_unmount(), neither unmount_rootfs() tests itself the
return status (which was always getting failures previously).
So OSv is really being shutdown while there is remaining data waiting to
be synced with the backing store. As a result, inconsitency.

This problem was fixed by passing the flag to VFS_UNMOUNT which will now
unmount the fs properly on sudden shutdowns.

Signed-off-by: Raphael S. Carvalho <raphaelsc@cloudius-systems.com>
Signed-off-by: Pekka Enberg <penberg@cloudius-systems.com>

Start using spaces instead of tabs and surround all single-line
control statements with curly braces.

Signed-off-by: Raphael S. Carvalho <raphaelsc@cloudius-systems.com>
Signed-off-by: Pekka Enberg <penberg@cloudius-systems.com>

It will be useful to take better and safer VFS decisions in the future.
For example, avoiding code that uses the absolute path to determine something.

Signed-off-by: Raphael S. Carvalho <raphaelsc@cloudius-systems.com>
Signed-off-by: Pekka Enberg <penberg@cloudius-systems.com>

newmp->m_covered must be released if not NULL.
Found this problem while dumping dcache content.

Signed-off-by: Raphael S. Carvalho <raphaelsc@cloudius-systems.com>
Signed-off-by: Pekka Enberg <penberg@cloudius-systems.com>

Currently, vflush is used in the unmount process to release remaining
dentries. vflush in turn calls vevict that is releasing dentries that
it doesn't own.
This behavior is not correct neither good to the future of VFS.

So Avi suggested switching to a different approach. We could only
release those dentries owned by the mountpoint when unmounting it as
there wouldn't be anything else in the dcache (given its functionality).

The problem was fixed by doing the following steps:
 - Drop vflush calls in sys_umount2, make vevict an empty function,
and remove vevict.

 - Created the function release_mp_dentries to release dentries of a mount
point which will be called by VFS_UNMOUNT. It cannot be called before
VFS_UNMOUNT as failures must be considered, neither after as the mount point
would be considered busy.
Don't respect this "rule", and that previously seen ZFS replay transaction
error would happen.

NOTE: vflush is currently duplicated in zfs unmount cases to address the problem
above. This patch fixes this duplication as well.

Signed-off-by: Raphael S. Carvalho <raphaelsc@cloudius-systems.com>
Signed-off-by: Pekka Enberg <penberg@cloudius-systems.com>

Instead of the old C-style file-operation function types and fo_*()
functions, since recently we have methods of the "file" class. All our
filesystem code is now C++, and can use these methods directly.

So this patch drops the old types and functions, and uses the class methods
instead.

Signed-off-by: Nadav Har'El <nyh@cloudius-systems.com>
Signed-off-by: Avi Kivity <avi@cloudius-systems.com>

Each implementation of "struct file" needs to implement 8 different file
operations. Most special file implementations, such as pipe, socketpair,
epoll and timerfd, don't support many of these operations. We had in
unsupported.h functions that can be reused for the unsupported operation,
but this resulted in a lot of ugly boiler-plate code.

Instead, this patch switches to a cleaner, more C++-like, method:
It defines a new "file" subclass, called "special_file", which implements
all file operations except close(), with a default implementation identical
to the old unsupported.h implementations.

The files of pipe(), socketpair(), timerfd() and epoll_create() now inherit
from special_file, and only override the file operations they really want
to implement.

Signed-off-by: Nadav Har'El <nyh@cloudius-systems.com>
Signed-off-by: Avi Kivity <avi@cloudius-systems.com>

Signed-off-by: Asias He <asias@cloudius-systems.com>
Signed-off-by: Pekka Enberg <penberg@cloudius-systems.com>

If iov->iov_len == 0, it will loop forever.
uiomove() will take care of the zero iov len case.

Signed-off-by: Asias He <asias@cloudius-systems.com>
Signed-off-by: Pekka Enberg <penberg@cloudius-systems.com>

Signed-off-by: Raphael S. Carvalho <raphaelsc@cloudius-systems.com>
Signed-off-by: Pekka Enberg <penberg@cloudius-systems.com>

Currently, namei() does vget() unconditionally if no dentry is found.
This is wrong because the path can be a hard link that points to a vnode
that's already in memory.

To fix the problem:

  - Use inode number as part of the hash in vget()

  - Use vn_lookup() in vget() to make sure we have one vnode in memory
    per inode number.

  - Push the vget() calls down to individual filesystems and make
    VOP_LOOKUP return an vnode

Changes since v2:
  - v1 dropped lock in vn_lookup, thus assert that vnode_lock is held.

Changes since v3:
  - Fix lock ordering issue in dentry_lookup. The lock respective to the parent
node must be acquired before dentry_lookup and released after the process is
done. Otherwise, a second thread looking up for the same dentry may take the
'NULL' path incorrectly.

Signed-off-by: Pekka Enberg <penberg@cloudius-systems.com>
Signed-off-by: Raphael S. Carvalho <raphaelsc@cloudius-systems.com>
Signed-off-by: Avi Kivity <avi@cloudius-systems.com>

This patch fixes the error codes in four error cases:

1. unlink() of a directory used to return EPERM (as in Posix), and now
   returns EISDIR (as in Linux).

2. rmdir() of a non-empty directory used to return EEXIST (as in Posix)
   and now returns ENOTEMPTY (as in Linux).

3. rmdir() of a regular file (non-directory) used to return EBADF
   and now returns ENOTDIR (as in Linux).

4. readdir() of a regular file (non-directory) used to return EBADF
   and now returns ENOTDIR (as in Linux).

This patch also adds a test, tst-remove.cc, for the various unlink() and
rmdir() success and failure modes.

Fixes #123.

Signed-off-by: Nadav Har'El <nyh@cloudius-systems.com>
Signed-off-by: Pekka Enberg <penberg@cloudius-systems.com>

I tried using a test which called mknod() (to create an empty regular file).
Despite us having an mknod() implementation, it didn't work, and failed on
lookup of the symbol __xmknod.

Turns out that in glibc, mknod() is source-only, and converted to the ABI
function which is __xmknod, whose first parameter is a version number
_MKNOD_VER_LINUX (0 on x86-64 Linux).

So this patch implements __xmknod, and now mknod() works.

Note we already had the same kind of trick for __xstat(), needed so that
stat() would work.

Signed-off-by: Nadav Har'El <nyh@cloudius-systems.com>
Signed-off-by: Pekka Enberg <penberg@cloudius-systems.com>

main.cc was still using tab characters instead of spaces as our coding
conventions dictate. Reindent it, using Eclipse's ctrl-I.
This patch doesn't change anything else.

Signed-off-by: Nadav Har'El <nyh@cloudius-systems.com>
Signed-off-by: Avi Kivity <avi@cloudius-systems.com>

This reverts commit e4aad1ba.

It causes tst-vfs.so to just hang.

Signed-off-by: Pekka Enberg <penberg@cloudius-systems.com>

Currently, namei() does vget() unconditionally if no dentry is found.
This is wrong because the path can be a hard link that points to a vnode
that's already in memory.

To fix the problem:

  - Use inode number as part of the hash in vget()

  - Use vn_lookup() in vget() to make sure we have one vnode in memory
    per inode number.

  - Push the vget() calls down to individual filesystems and make
    VOP_LOOKUP return an vnode

  - Drop lock in vn_lookup() and assert that vnode_lock is held.

Signed-off-by: Pekka Enberg <penberg@cloudius-systems.com>
Signed-off-by: Raphael S. Carvalho <raphaelsc@cloudius-systems.com>
Signed-off-by: Avi Kivity <avi@cloudius-systems.com>

Everyone is now overriding file's virtual functions; we can make them
pure virtual and remove fileops completely.

Signed-off-by: Avi Kivity <avi@cloudius-systems.com>

Unused.

Signed-off-by: Avi Kivity <avi@cloudius-systems.com>

Everyone switched to the nifty variadic type.

Signed-off-by: Avi Kivity <avi@cloudius-systems.com>

Signed-off-by: Avi Kivity <avi@cloudius-systems.com>

Derived file objects will be initialized by the class constructor, no need
for fo_init().

Signed-off-by: Avi Kivity <avi@cloudius-systems.com>

Once file::~file() is called, virtual functions no longer dispatch to the
derived type (which has since been destroyed) but to the base type, which
is uninteresting.

Move the call to close() from the destructor to fdrop().

Signed-off-by: Avi Kivity <avi@cloudius-systems.com>

The default is to dispatch directly to the corresponding member of f_ops,
but that can be overridden.

The fo_*() functions are redirected to dispatch via the virtual functions.

Signed-off-by: Avi Kivity <avi@cloudius-systems.com>

Signed-off-by: Avi Kivity <avi@cloudius-systems.com>

Make sure ZFS intention log is flushed to disk by unmounting rootfs at
poweroff.

Fixes #108.

Signed-off-by: Pekka Enberg <penberg@cloudius-systems.com>

Make sure the new root vnode is not evicted in vflush() and that we
don't keep reference to the old root dentry in ->m_covered.

Fixes negative dentry count when pivoted root is unmounted.

Signed-off-by: Pekka Enberg <penberg@cloudius-systems.com>

Add sys_umount2() and implement support for MNT_FORCE that will be used
to force rootfs unmount at poweroff.

Signed-off-by: Pekka Enberg <penberg@cloudius-systems.com>

This paves the way for moving f_ops into the vtable.

Signed-off-by: Avi Kivity <avi@cloudius-systems.com>

Signed-off-by: Avi Kivity <avi@cloudius-systems.com>

Simpler.

Signed-off-by: Avi Kivity <avi@cloudius-systems.com>

'new file' doesn't return nullptr, it throws (well except in osv where new
always succeeds).

Replace with new (nothrow) to get the expected behaviour.

Signed-off-by: Avi Kivity <avi@cloudius-systems.com>

They are called at the same time, so make falloc_noinstall() responsible
purely for allocation.

Signed-off-by: Avi Kivity <avi@cloudius-systems.com>

If we're going to have a vtable in there, the memset() will kill it.

Instead, add initializers for those members not already initialized by
make_file().

Signed-off-by: Avi Kivity <avi@cloudius-systems.com>

The only caller, soo_close() can only be called from a context where
no file references remain, so no further file API calls can be made.

Remove it.

Signed-off-by: Avi Kivity <avi@cloudius-systems.com>

Subsumed by make_file().

Signed-off-by: Avi Kivity <avi@cloudius-systems.com>
Signed-off-by: Pekka Enberg <penberg@cloudius-systems.com>

Signed-off-by: Avi Kivity <avi@cloudius-systems.com>
Signed-off-by: Pekka Enberg <penberg@cloudius-systems.com>

Makes it easy to avoid losing them.

Signed-off-by: Avi Kivity <avi@cloudius-systems.com>
Signed-off-by: Pekka Enberg <penberg@cloudius-systems.com>

Combines falloc_noinstall() and finit(), which are always used together.

Signed-off-by: Avi Kivity <avi@cloudius-systems.com>
Signed-off-by: Pekka Enberg <penberg@cloudius-systems.com>

falloc() is inherently racy in that it installs an uninitialized file
descriptor in a user accessible fd.  It is also hard to use correctly when
an error occurs.  Luckily, we don't use it anywhere, so we can just remove it.

Signed-off-by: Avi Kivity <avi@cloudius-systems.com>

Copying filerefs is expensive since the reference count is incremented,
avoid it.

Signed-off-by: Avi Kivity <avi@cloudius-systems.com>