unionfs-2.6.39.y.git
13 years agoUnionfs: release 2.1.1. unionfs-2.1.1
Erez_Zadok [Sun, 19 Aug 2007 19:54:06 +0000 (15:54 -0400)]
Unionfs: release 2.1.1.

Signed-off-by: Erez Zadok <ezk@cs.sunysb.edu>
13 years agoUnionfs: minor usage.txt documentation fix
Erez_Zadok [Sun, 19 Aug 2007 19:22:43 +0000 (15:22 -0400)]
Unionfs: minor usage.txt documentation fix

Signed-off-by: Erez Zadok <ezk@cs.sunysb.edu>
13 years agoUnionfs: imported fixes from korg branch
Erez_Zadok [Sat, 18 Aug 2007 20:02:02 +0000 (16:02 -0400)]
Unionfs: imported fixes from korg branch

Export release_open_intent.
SElinux: xattr fixes (CAP_FOWNER).
unionfs_xattr_kfree inline function.
alloc_lower_nd -> init_lower_nd.
free_lower_nd -> release_lower_nd.
unionfs_purge_extras -> unionfs_postcopyup_release.
unionfs_inherit_mnt -> unionfs_postcopyup_setmnt.
minor code/copyright cleanups.
MS_SILENT remount fix.
simplify unionfs_mntget/put.

Signed-off-by: Erez Zadok <ezk@cs.sunysb.edu>
13 years agoUnionfs: back-port stuff from Jeff's korg branch to per-kernel branches
Erez_Zadok [Sat, 11 Aug 2007 18:01:08 +0000 (14:01 -0400)]
Unionfs: back-port stuff from Jeff's korg branch to per-kernel branches

Just a bunch of code cleanups: no functionality changed.

Signed-off-by: Erez Zadok <ezk@cs.sunysb.edu>
Signed-off-by: Josef 'Jeff' Sipek <jsipek@cs.sunysb.edu>
13 years agoRevert "bugfix: add MODULE_LICENSE to mtdsuper module so it loads"
Erez_Zadok [Sat, 11 Aug 2007 02:21:05 +0000 (22:21 -0400)]
Revert "bugfix: add MODULE_LICENSE to mtdsuper module so it loads"

This reverts commit 924185024a661a3a0fb772fde4ce178e587ed018.

13 years agoUnionfs/eCryptfs: remove lookup_one_len_nd and LOOKUP_ONE code
Erez_Zadok [Thu, 9 Aug 2007 21:04:19 +0000 (17:04 -0400)]
Unionfs/eCryptfs: remove lookup_one_len_nd and LOOKUP_ONE code

Signed-off-by: Josef 'Jeff' Sipek <jsipek@cs.sunysb.edu>
Signed-off-by: Erez Zadok <ezk@cs.sunysb.edu>
13 years agoUnionfs: support incremental versions
Erez_Zadok [Tue, 7 Aug 2007 22:51:20 +0000 (18:51 -0400)]
Unionfs: support incremental versions

Also: release Unionfs 2.1 (since code has gotten much more stable since 2.0)

Unionfs will now report its precise version number and the kernel it was
meant for, upon module loading.  This is useful for maintenance purposes:
users can report the precise release number of Unionfs, and it also helps
prevent running older versions when newer ones are intended, or mixing
versions among supported kernels.

Signed-off-by: Erez Zadok <ezk@cs.sunysb.edu>
13 years agoUnionfs: check for immutable files before read-only branches
Erez_Zadok [Wed, 1 Aug 2007 14:29:42 +0000 (10:29 -0400)]
Unionfs: check for immutable files before read-only branches

Immutable files should never be allowed to be copied-up on write, even if
they're on read-only file systems or branches.

Signed-off-by: Erez Zadok <ezk@cs.sunysb.edu>
13 years agoUnionfs: mmap fixes to unionfs_writepage
Erez_Zadok [Tue, 31 Jul 2007 23:25:55 +0000 (19:25 -0400)]
Unionfs: mmap fixes to unionfs_writepage

Don't call unlock_page on lower_page unless lower ->writepage failed.

Signed-off-by: Erez Zadok <ezk@cs.sunysb.edu>
13 years agoUnionfs: rewrite cleanup_file more cleanly
Erez_Zadok [Tue, 31 Jul 2007 22:12:25 +0000 (18:12 -0400)]
Unionfs: rewrite cleanup_file more cleanly

Signed-off-by: Erez Zadok <ezk@cs.sunysb.edu>
13 years agoUnionfs: improved printk upon copyup
Erez_Zadok [Tue, 31 Jul 2007 07:58:50 +0000 (03:58 -0400)]
Unionfs: improved printk upon copyup

Signed-off-by: Erez Zadok <ezk@cs.sunysb.edu>
13 years agoUnionfs: xattr copyup fixes
Erez_Zadok [Tue, 31 Jul 2007 07:53:47 +0000 (03:53 -0400)]
Unionfs: xattr copyup fixes

Rewrote xattr copyup code more cleanly; documented it better; eliminate one
possible leak in error path; and ignore another impossible copyup-time error
which caused fanout invariant violations under memory-pressure conditions.

Don't use vmalloc when allocating xattr buffers, as the VFS no longer does
so (just use kmalloc).  Eliminate unionfs_xattr_free which is now just plain
kfree.

Signed-off-by: Erez Zadok <ezk@cs.sunysb.edu>
13 years agoUnionfs: fixes to cache-coherency checking code
Erez_Zadok [Tue, 31 Jul 2007 07:33:24 +0000 (03:33 -0400)]
Unionfs: fixes to cache-coherency checking code

Under memory pressure, in_newer_lower could be called on uninitialized
dentries or inodes.  So return 0 safely in that case (rather than oops).
This is OK because is_newer_lower will be called again on the same object(s)
and cache-coherency will be validated and maintained then.

Signed-off-by: Erez Zadok <ezk@cs.sunysb.edu>
13 years agoUnionfs: mmap fixes to unionfs_writepage
Erez_Zadok [Tue, 31 Jul 2007 07:29:50 +0000 (03:29 -0400)]
Unionfs: mmap fixes to unionfs_writepage

This patch fixes hangs when calling sync(2) on memory-pressured systems.

Call find_lock_page instead of grab_cache_page.  We used to call
grab_cache_page(), but that was unnecessary as it would have tried to create
a new lower page if it didn't exist, leading to deadlocks (esp. under
memory-pressure conditions, when it is really a bad idea to *consume* more
memory).  Instead, we assume the lower page exists, and if we can find it,
then we ->writepage on it; if we can't find it, then it couldn't have
disappeared unless the kernel already flushed it, in which case we're still
OK.  This is especially correct if wbc->sync_mode is WB_SYNC_NONE (as per
Documentation/filesystems/vfs.txt).  If we can't flush our page because we
can't find a lower page, then at least we re-mark our page as dirty, and
return AOP_WRITEPAGE_ACTIVATE as the VFS expects us to.  (Note, if in the
future it'd turn out that we have to find a lower page no matter what, then
we'd have to resort to RAIF's page pointer flipping trick.)

Signed-off-by: Erez Zadok <ezk@cs.sunysb.edu>
13 years agoUnionfs: bugfix when mounting readonly exported NFS volumes (was: nfsro)
Erez_Zadok [Thu, 26 Jul 2007 04:04:58 +0000 (00:04 -0400)]
Unionfs: bugfix when mounting readonly exported NFS volumes (was: nfsro)

When stacking on top of readonly exported NFS volumes, which are mounted
locally read-write, attempting to create a new file yields the proper EROFS.
But attempting to modify an existing file returns an EACCES, which
interferes with unionfs's copyup policy: only EROFS triggers a copyup,
whereas EACCES does not (and shouldn't -- that'd be a security hole).  The
old unionfs 1.x attempted to workaround this EACCES condition by supporting
a special unionfs mount option called 'nfsro'; support for this option was
left in the latest unionfs 2.0, but the mount option was not made available
until we could properly investigate this issue with the latest NFS code.

This patch removes all remnants of this 'nfsro' support.  It is no longer
needed.  Instead, users can use the existing per-branch 'ro' unionfs mount
option, which would properly return the appropriate status conditions back
from unionfs_permission.  These return conditions result in a copyup if and
only if needed, even for readonly exported NFS volumes.  In effect, unionfs
per-branch 'ro' option now simulates a true readonly localhost mount.

Signed-off-by: Erez Zadok <ezk@cs.sunysb.edu>
13 years ago[PATCH] unionfs: make functions static
Adrian Bunk [Mon, 2 Jul 2007 18:27:57 +0000 (14:27 -0400)]
[PATCH] unionfs: make functions static

This patch makes some needlessly global functions static.

Signed-off-by: Adrian Bunk <bunk@stusta.de>
Signed-off-by: Josef 'Jeff' Sipek <jsipek@cs.sunysb.edu>
13 years agoUnionfs: pass nameidata intent information to lower level file systems
Erez_Zadok [Mon, 23 Jul 2007 07:13:01 +0000 (03:13 -0400)]
Unionfs: pass nameidata intent information to lower level file systems

As of 2.6.23-rc1, nfs2 and nfs3, like nfs4 before them, begin relying on the
struct nameidata and especially the intent information, which is passed to
vfs_create() and others.  So, as of now, unionfs properly creates and passes
that intent data to the lower level file system.  Currently supported are
LOOKUP_CREATE open intents.  Others can be supported in the future
incrementally.

Signed-off-by: Erez Zadok <ezk@cs.sunysb.edu>
13 years agoUnionfs: kmem_cache_create doesn't take a dtor argument any longer
Erez_Zadok [Mon, 23 Jul 2007 01:53:49 +0000 (21:53 -0400)]
Unionfs: kmem_cache_create doesn't take a dtor argument any longer

Porting unionfs to 2.6.23-rc1

Signed-off-by: Erez Zadok <ezk@cs.sunysb.edu>
13 years agoUnionfs: ensure cache coherency in unionfs_fsync/fasync
Erez_Zadok [Sun, 22 Jul 2007 05:10:45 +0000 (01:10 -0400)]
Unionfs: ensure cache coherency in unionfs_fsync/fasync

Signed-off-by: Erez Zadok <ezk@cs.sunysb.edu>
13 years agoUnionfs: implement native unionfs_fsync/unionfs_fasync methods
Erez_Zadok [Sun, 22 Jul 2007 03:56:16 +0000 (23:56 -0400)]
Unionfs: implement native unionfs_fsync/unionfs_fasync methods

Unionfs is not a block-based file system, but it has to work with both
block-based file systems and non-block-based ones (i.e., when
CONFIG_BLOCK=n).  We used to define our ->fsync method to file_fsync, but
that's wrong because file_fsync partially depends on CONFIG_BLOCK=y.  And we
didn't define an fasync method: now we have both.  Moreover, at best,
file_fsync would have caused unionfs to sync up one lower branch---but as a
fanout file system, we need to sync up all valid lower branches.

Signed-off-by: Erez Zadok <ezk@cs.sunysb.edu>
Conflicts:

fs/unionfs/file.c

13 years agoUnionfs: small documentation fix to usage.txt
Erez_Zadok [Sun, 22 Jul 2007 00:22:38 +0000 (20:22 -0400)]
Unionfs: small documentation fix to usage.txt

Reported by Daniel Farrugia.

Signed-off-by: Erez Zadok <ezk@cs.sunysb.edu>
13 years agoUnionfs: mmap fixes to ->writepage/readpage/sync_page
Erez_Zadok [Sat, 21 Jul 2007 06:44:32 +0000 (02:44 -0400)]
Unionfs: mmap fixes to ->writepage/readpage/sync_page

unionfs_writepage: handle true errors differently from
AOP_WRITEPAGE_ACTIVATE conditions returned by lower file systems (such as
tmpfs).

unionfs_readpage: call flush_dcache_page as required.

unionfs_sync_page: don't call grab_cache_page to get the lower page, because
that function does too much and could lead to deadlocks.  Instead, call the
lighter-weight find_lock_page.

Signed-off-by: Erez Zadok <ezk@cs.sunysb.edu>
13 years agoUnionfs: branch-management bugfix to unionfs_file_revalidate
Erez_Zadok [Sat, 21 Jul 2007 06:38:14 +0000 (02:38 -0400)]
Unionfs: branch-management bugfix to unionfs_file_revalidate

If we re-opened the file on a different branch than the original one, and
only if this was due to a new branch inserted, then update the mnt counts of
the old and new branches accordingly.

Signed-off-by: Erez Zadok <ezk@cs.sunysb.edu>
13 years agoUnionfs: cleanup: break line longer than 80 chars
Erez_Zadok [Tue, 17 Jul 2007 07:18:25 +0000 (03:18 -0400)]
Unionfs: cleanup: break line longer than 80 chars

Signed-off-by: Erez Zadok <ezk@cs.sunysb.edu>
13 years agoUnionfs: fix minor typo in revalidate_chain comment
Erez_Zadok [Tue, 17 Jul 2007 06:58:15 +0000 (02:58 -0400)]
Unionfs: fix minor typo in revalidate_chain comment

Signed-off-by: Erez Zadok <ezk@cs.sunysb.edu>
13 years agoUnionfs: bugfix: don't call branchget() until dentry_open succeeded
Erez_Zadok [Tue, 17 Jul 2007 03:56:26 +0000 (23:56 -0400)]
Unionfs: bugfix: don't call branchget() until dentry_open succeeded

If dentry_open fails, we incorrectly have incremented our branch counts
which would prevent unionfs from being unmounted (EBUSY).

Signed-off-by: Erez Zadok <ezk@cs.sunysb.edu>
13 years agoUnionfs: branch-management bug-fix to lower mnt counts
Erez_Zadok [Tue, 17 Jul 2007 03:54:11 +0000 (23:54 -0400)]
Unionfs: branch-management bug-fix to lower mnt counts

When we revalidate opened files, the files may point to a different set of
files on the lower branches: this could happen especially if we insert a new
leftmost branch which happens to have duplicate file names of already-open
files in lower-priority branches.  In this case, update the mnt counts of
the old and new branches which hold the file accordingly.  Without this fix,
the mntcnt of one branch is going to be one too high; and another will be
one too low.

Signed-off-by: Erez Zadok <ezk@cs.sunysb.edu>
13 years agoUnionfs: use igrab instead of atomic_inc inode refcnt
Erez_Zadok [Tue, 17 Jul 2007 03:45:46 +0000 (23:45 -0400)]
Unionfs: use igrab instead of atomic_inc inode refcnt

In branch management code, use igrab() which is better than directly
incrementing the lower inode reference count.

Signed-off-by: Erez Zadok <ezk@cs.sunysb.edu>
13 years agoUnionfs: cache-coherency: update times of root inode after branch management
Erez_Zadok [Tue, 17 Jul 2007 03:42:55 +0000 (23:42 -0400)]
Unionfs: cache-coherency: update times of root inode after branch management

Signed-off-by: Erez Zadok <ezk@cs.sunysb.edu>
13 years agoUnionfs: debugging: don't oops if sb->s_root is NULL
Erez_Zadok [Tue, 17 Jul 2007 03:40:07 +0000 (23:40 -0400)]
Unionfs: debugging: don't oops if sb->s_root is NULL

If we're calling show_branch_counts very early during mount or late during
unmount, don't oops if the s_root is already NULL.

Signed-off-by: Erez Zadok <ezk@cs.sunysb.edu>
13 years agoUnionfs: debugging: show inode reference counts for all branches
Erez_Zadok [Tue, 17 Jul 2007 03:38:30 +0000 (23:38 -0400)]
Unionfs: debugging: show inode reference counts for all branches

Another useful utility function for debugging, to show the inode refcnt's
of all lower inodes.

Signed-off-by: Erez Zadok <ezk@cs.sunysb.edu>
13 years agoUnionfs: fix small typo in comment
Erez_Zadok [Tue, 17 Jul 2007 03:35:19 +0000 (23:35 -0400)]
Unionfs: fix small typo in comment

Signed-off-by: Erez Zadok <ezk@cs.sunysb.edu>
13 years agoUnionfs: bugfix -- set copied up open file mode to read and write
Erez_Zadok [Sat, 14 Jul 2007 07:25:02 +0000 (03:25 -0400)]
Unionfs: bugfix -- set copied up open file mode to read and write

When we copyup a file, we used to set the new file's open mode to write
only.  But we need read permission too.  We need write permission because
we're creating a new file; and we need read permission because the original
file had to have been readable.  If we don't set the new files permission to
BOTH read and write, we get -EBADF when trying calling vfs_read/vfs_write on
a copied up file.  (This fix is somewhat related to unionfs_readpage fixes
that use vfs_read instead of calling the lower readpage.)

Signed-off-by: Erez Zadok <ezk@cs.sunysb.edu>
13 years agoUnionfs: rewrite do_unionfs_readpage to use vfs_read (bugfix)
Erez_Zadok [Sat, 14 Jul 2007 07:36:15 +0000 (03:36 -0400)]
Unionfs: rewrite do_unionfs_readpage to use vfs_read (bugfix)

In do_unionfs_readpage, we used to call the lower file system's ->readpage.
However, some file systems (e.g., tmpfs) don't implement ->readpage, causing
a NULL pointer dereference under certain conditions, especially under severe
memory pressure.  This patch reimplements do_unionfs_readpage using
vfs_read, which makes the code simpler and more reliable, as we depend on
the VFS to do most of the hard work (even if this implementation might be a
bit slower).

This fix also makes sense because it makes the mmap code in unionfs more
symmetric with unionfs_commit_write --- which uses vfs_write().

Signed-off-by: Erez Zadok <ezk@cs.sunysb.edu>
13 years agoUnionfs: fix minor typo in a comment
Erez_Zadok [Fri, 13 Jul 2007 06:40:31 +0000 (02:40 -0400)]
Unionfs: fix minor typo in a comment

Signed-off-by: Erez Zadok <ezk@cs.sunysb.edu>
13 years agoUnionfs: use the splice_read file method instead of sendfile
Erez_Zadok [Wed, 11 Jul 2007 20:13:07 +0000 (16:13 -0400)]
Unionfs: use the splice_read file method instead of sendfile

In 2.6.23 (prior to rc1), sendfile was obsoleted in favor of splice_read.

Signed-off-by: Erez Zadok <ezk@cs.sunysb.edu>
13 years agounionfs: better handling when copying up permissions
Erez_Zadok [Tue, 10 Jul 2007 00:19:08 +0000 (20:19 -0400)]
unionfs: better handling when copying up permissions

When we copyup a file, directory, or symlink, we may be copying up from one
file system type to another.  The destination file system may not support
all of the features of the source file system, and the differences in
support may be minor.  For example jffs2 doesn't allow one to chmod a
symlink (and it returns a -EOPNOTSUPP).  So we ignore such harmless errors,
rather than propagating them up, which results in copyup errors and errors
returned back to users.

Signed-off-by: Erez Zadok <ezk@cs.sunysb.edu>
13 years agoUnionfs: better error handling in rename code when copyups are involved
Erez_Zadok [Tue, 10 Jul 2007 00:14:41 +0000 (20:14 -0400)]
Unionfs: better error handling in rename code when copyups are involved

First, rewrite code slightly and document it better to explain why we appear
to ignore copyup errors (because we try the next branch to the left).

Second, change a BUG_ON to a printk(KERN_ERR), because a mild failure to
copyup a file should not cause an oops.  For example, some file systems
don't support UIDs/GIDs (e.g., VFAT) and others don't allow you to chmod a
symlink (e.g., jffs2), possibly resulting in mild copyup failures; that
shouldn't be considered so severe as to cause an oops.

Signed-off-by: Erez Zadok <ezk@cs.sunysb.edu>
13 years agoUnionfs: bugfix when renaming symlinks on readonly branches
Erez_Zadok [Sun, 8 Jul 2007 01:31:28 +0000 (21:31 -0400)]
Unionfs: bugfix when renaming symlinks on readonly branches

Signed-off-by: Erez Zadok <ezk@cs.sunysb.edu>
13 years agoUnionfs: prevent deadlock with branch-management code.
Erez_Zadok [Tue, 3 Jul 2007 13:14:39 +0000 (09:14 -0400)]
Unionfs: prevent deadlock with branch-management code.

Don't grab the superblock read-lock in unionfs_permission, which prevents a
deadlock with the branch-management "add branch" code (which grabbed the
write lock).  It is safe to not grab the read lock here, because even with
branch management taking place, there is no chance that unionfs_permission,
or anything it calls, will use stale branch information.

Signed-off-by: Erez Zadok <ezk@cs.sunysb.edu>
13 years agoUnionfs: rename file->f_dentry references to file->f_path.dentry
Erez_Zadok [Fri, 29 Jun 2007 06:33:14 +0000 (02:33 -0400)]
Unionfs: rename file->f_dentry references to file->f_path.dentry

Signed-off-by: Erez Zadok <ezk@cs.sunysb.edu>
13 years agoUnionfs: improved comment above unionfs_follow_link
Erez_Zadok [Fri, 29 Jun 2007 06:02:42 +0000 (02:02 -0400)]
Unionfs: improved comment above unionfs_follow_link

Signed-off-by: Erez Zadok <ezk@cs.sunysb.edu>
13 years agoUnionfs: Change the semantics of sb info's rwsem
Erez_Zadok [Fri, 29 Jun 2007 06:00:24 +0000 (02:00 -0400)]
Unionfs: Change the semantics of sb info's rwsem

This rw semaphore is used to make sure that a branch management
operation...

1) will not begin before all currently in-flight operations complete

2) any new operations do not execute until the currently running branch
management operation completes

Reworked the patch a bit, added comments, and fixed some bugs, from the
version originally committed into the master branch.

TODO: rename the functions unionfs_{read,write}_{,un}lock() to something
more descriptive.

Signed-off-by: Josef 'Jeff' Sipek <jsipek@cs.sunysb.edu>
Signed-off-by: Erez Zadok <ezk@cs.sunysb.edu>
13 years agoUnionfs: Remove superfluous check for NULL pointer
Erez_Zadok [Fri, 29 Jun 2007 04:45:47 +0000 (00:45 -0400)]
Unionfs: Remove superfluous check for NULL pointer

Since we use containers and the struct inode is _inside_ the
unionfs_inode_info structure, UNIONFS_I will always (given a non-NULL inode
pointer), return a valid non-NULL pointer.

Signed-off-by: Josef 'Jeff' Sipek <jsipek@cs.sunysb.edu>
Conflicts:

fs/unionfs/super.c

13 years agoUnionfs: Change free_dentry_private_info to take a struct dentry
Josef 'Jeff' Sipek [Fri, 8 Jun 2007 14:58:22 +0000 (10:58 -0400)]
Unionfs: Change free_dentry_private_info to take a struct dentry

This makes it more symmetric with new_dentry_private_info.

Signed-off-by: Josef 'Jeff' Sipek <jsipek@cs.sunysb.edu>
13 years agoUnionfs: Cleanup new_dentry_private_data
Erez_Zadok [Fri, 29 Jun 2007 04:36:08 +0000 (00:36 -0400)]
Unionfs: Cleanup new_dentry_private_data

Signed-off-by: Josef 'Jeff' Sipek <jsipek@cs.sunysb.edu>
Conflicts:

fs/unionfs/lookup.c

13 years agoUnionfs: document cache-coherency design and implementation
Erez_Zadok [Mon, 18 Jun 2007 02:39:38 +0000 (22:39 -0400)]
Unionfs: document cache-coherency design and implementation

Document our cache-coherency design, implementation; also mention alternate
designs and a couple of limitations which might be overcome by special
kernel support at the VFS/MM levels.  Now that cache-coherency is working,
remove the now-obsolete older discussion of temporary workarounds.

Signed-off-by: Erez Zadok <ezk@cs.sunysb.edu>
13 years agoUnionfs: cache-coherency calls to maintain the time invariants
Erez_Zadok [Mon, 18 Jun 2007 02:39:05 +0000 (22:39 -0400)]
Unionfs: cache-coherency calls to maintain the time invariants

This patch represents several types of related changes.  First, we invoke
functions to synchronize the upper and lower times as and when needed.  Many
of these were bug fixes which were discovered during the development of the
cache-coherency code.  That is, Unionfs itself wasn't maintaining
appropriate
times in some places, which if not fixed would have been detected by the
invariant-checking code as a false positive (incorrectly considered as if a
user modified the lower objects directly).

Second, we do not call invariant-validation functions (unionfs_check_file,
unionfs_check_dentry, etc.) until *after* we've revalidated them.  Otherwise
we produced false positives.

Third, we pass a flag "willwrite" to __unionfs_d_revalidate_chain to tell it
to purge data pages if the inode lower times appear to be newer.

See Documentation/filesystems/unionfs/concepts.txt under the "Cache
Coherency" section for more details of this design and implementation.

Signed-off-by: Erez Zadok <ezk@cs.sunysb.edu>
13 years agoUnionfs: core cache-coherency code
Erez_Zadok [Mon, 18 Jun 2007 02:37:54 +0000 (22:37 -0400)]
Unionfs: core cache-coherency code

This represents the core of the cache-coherency code, the code which
maintains time-based invariants; the lower objects should never be newer
than the upper Unionfs objects.  If they are newer, then it means that a
user has modified the lower Unionfs branches directly.  If so, then we have
to revalidate our objects by rebuilding them, and possibly discard any stale
data or meta-data.  See Documentation/filesystems/unionfs/concepts.txt under
the "Cache Coherency" section for more details of this design and
implementation.

Signed-off-by: Erez Zadok <ezk@cs.sunysb.edu>
13 years agoUnionfs: introduce two functions to maintain time invariants
Erez_Zadok [Mon, 18 Jun 2007 02:36:55 +0000 (22:36 -0400)]
Unionfs: introduce two functions to maintain time invariants

A new function unionfs_copy_attr_times to maintain the inode time invariants
as per the Unionfs design (i.e., lower times should never be newer than our
times), as documented in Documentation/filesystems/unionfs/concepts.txt.
The second function, unionfs_copy_attr_all, is the unionfs-specific function
to copy all attributes: it uses the special Unionfs handling of hard-links
(via unionfs_get_nlinks) and it uses the special a/c/mtime handling (via
unionfs_copy_attr_times).  This second function is the special
fanout-version of fsstack_copy_attr_all, which was simplified for
linear-stack stackable file systems.

Signed-off-by: Erez Zadok <ezk@cs.sunysb.edu>
13 years agoUnionfs: prototype and location change for unionfs_get_nlinks
Erez_Zadok [Mon, 18 Jun 2007 02:36:03 +0000 (22:36 -0400)]
Unionfs: prototype and location change for unionfs_get_nlinks

Change prototype of unionfs_get_nlinks to take a const inode.  Also move
extern for unionfs_get_nlinks so it can be refereed to in other Unionfs
header files.

Signed-off-by: Erez Zadok <ezk@cs.sunysb.edu>
13 years agoUnionfs: m/c/atime invariant checking debugging code
Erez_Zadok [Mon, 18 Jun 2007 02:33:56 +0000 (22:33 -0400)]
Unionfs: m/c/atime invariant checking debugging code

Add checks to verify that lower inode times are not newer than our own inode
times.  Introduce two useful utility debugging functions to print inode
times.

Signed-off-by: Erez Zadok <ezk@cs.sunysb.edu>
13 years agoUnionfs: bugfix in debugging infrastructure
Erez_Zadok [Mon, 18 Jun 2007 02:29:40 +0000 (22:29 -0400)]
Unionfs: bugfix in debugging infrastructure

Off-by-one bug fixed in range checking for lower directories.
Ensure that we PRINT_CALLER before an additional debug message.

Signed-off-by: Erez Zadok <ezk@cs.sunysb.edu>
13 years agoCache coherency: resync unionfs data/meta-data when lower files change
Erez_Zadok [Mon, 11 Jun 2007 16:48:29 +0000 (12:48 -0400)]
Cache coherency: resync unionfs data/meta-data when lower files change

Whenever we revalidate a file or dentry, we check to see if any of the lower
inodes have changed (new mtime/ctime).  If so, we revalidate the upper
unionfs objects.  This method "works" in that as long as a user process will
have caused unionfs to be called, directly or indirectly, even to just do
->d_revalidate, then we will have purged the current unionfs data and the
process will see the new data.  For example, a process that continually
re-reads the same file's data will see the NEW data as soon as the lower
file had changed, upon the next read(2) syscall.  This also works for
meta-data changes which change the ctime (chmod, chown, chgrp, etc).

However, this doesn't work when the process re-reads the file's data via
mmap and the data was already read before via mmap: once we respond to
->readpage(s), then the kernel maps the page into the process's address
space and there doesn't appear to be a way to force the kernel to invalidate
those pages/mappings, and force the process to re-issue ->readpage.  Note:
only pages that have already been readpage'ed are not updated; any other
pages which unionfs's ->readpage would be called on, WILL get the updated
data.  If there's a way to invalidate active mappings and force a
->readpage, let us know please (invalidate_inode_pages2 doesn't do the
trick).

Signed-off-by: Erez Zadok <ezk@cs.sunysb.edu>
13 years agorevert changes to new_dentry_private_data and document it
Erez_Zadok [Mon, 11 Jun 2007 14:52:21 +0000 (10:52 -0400)]
revert changes to new_dentry_private_data and document it

Revert bad changes to this function, but this time explain better its precise
semantics.  It can take a dentry whose private unionfs 'info' node exists or
not, and it should allocate and un/lock it only if it's NULL.

Signed-off-by: Erez Zadok <ezk@cs.sunysb.edu>
13 years agocleanup: convert all instances of "hidden" to "lower".
Erez_Zadok [Sun, 10 Jun 2007 23:04:30 +0000 (19:04 -0400)]
cleanup: convert all instances of "hidden" to "lower".

The unionfs code contained many references to symbols whose name had the
string "hidden" in it, referring to, for example, a "hidden inode" below the
unionfs inode.  The term "hidden" was there for historical reasons and was a
misnomer, implying that the objects at unionfs's lower branches were somehow
hidden or unavailable.  This was not just incorrect, but confusing.  The
lower objects are just that: lower.  They are not hidden from users.  In
fact, users can 'cd' to lower branches and manipulate files directly there.
This long patch renames all such mis-named symbols, and re-indents the code
as needed to comply with coding standards.

Signed-off-by: Erez Zadok <ezk@cs.sunysb.edu>
13 years agoFix section marker in header file:
Erez_Zadok [Sun, 10 Jun 2007 21:55:23 +0000 (17:55 -0400)]
Fix section marker in header file:

WARNING: fs/unionfs/unionfs.o(.init.text+0x56): Section mismatch: reference to .exit.text:stop_sioq (between 'init_module' and 'init_sioq')

Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com>
Signed-off-by: Josef 'Jeff' Sipek <jsipek@cs.sunysb.edu>
Signed-off-by: Erez Zadok <ezk@cs.sunysb.edu>
13 years agominor indentation correction
Erez_Zadok [Sun, 10 Jun 2007 21:51:42 +0000 (17:51 -0400)]
minor indentation correction

Signed-off-by: Erez Zadok <ezk@cs.sunysb.edu>
13 years agobugfix: add MODULE_LICENSE to mtdsuper module so it loads
Erez_Zadok [Wed, 6 Jun 2007 05:20:15 +0000 (01:20 -0400)]
bugfix: add MODULE_LICENSE to mtdsuper module so it loads

This is a temp fix in rc4.  Also reported this bug to MTD maintainer.

13 years agobugfix: properly reset lower inode start/end range
Erez_Zadok [Wed, 6 Jun 2007 01:58:07 +0000 (21:58 -0400)]
bugfix: properly reset lower inode start/end range

Signed-off-by: Erez Zadok <ezk@cs.sunysb.edu>
13 years agocleanup: rewrite new_dentry_private data more simply
Erez_Zadok [Sun, 3 Jun 2007 06:27:37 +0000 (02:27 -0400)]
cleanup: rewrite new_dentry_private data more simply

Also remove unnecessary variables and statements.

Signed-off-by: Erez Zadok <ezk@cs.sunysb.edu>
13 years agobugfix: remove extra superblock rwsem unlock call
Erez_Zadok [Sun, 3 Jun 2007 06:26:36 +0000 (02:26 -0400)]
bugfix: remove extra superblock rwsem unlock call

This also makes the code more symmetric: the same code which locks also
unlocks the rwsem.

Signed-off-by: Erez Zadok <ezk@cs.sunysb.edu>
13 years agoremove unnecessary comment
Erez_Zadok [Fri, 1 Jun 2007 04:47:20 +0000 (00:47 -0400)]
remove unnecessary comment

Signed-off-by: Erez Zadok <ezk@cs.sunysb.edu>
13 years agoproperly format a multi line comment
Erez_Zadok [Fri, 1 Jun 2007 04:01:03 +0000 (00:01 -0400)]
properly format a multi line comment

Signed-off-by: Erez Zadok <ezk@cs.sunysb.edu>
13 years agoreformat all lines longer than 80 characters
Erez_Zadok [Fri, 1 Jun 2007 03:34:59 +0000 (23:34 -0400)]
reformat all lines longer than 80 characters

Signed-off-by: Erez Zadok <ezk@cs.sunysb.edu>
13 years agodebugging: indent code properly and cleanups
Erez_Zadok [Fri, 1 Jun 2007 03:13:14 +0000 (23:13 -0400)]
debugging: indent code properly and cleanups

Also make PRINT_CALLER look like a function.
Fix one small bug in __show_branch_counts.

Signed-off-by: Erez Zadok <ezk@cs.sunysb.edu>
13 years agocleanup: reformat all lines longer than 80 characters
Erez_Zadok [Fri, 1 Jun 2007 02:32:10 +0000 (22:32 -0400)]
cleanup: reformat all lines longer than 80 characters

Signed-off-by: Erez Zadok <ezk@cs.sunysb.edu>
13 years agospell check all strings and comments
Erez_Zadok [Thu, 31 May 2007 19:53:28 +0000 (15:53 -0400)]
spell check all strings and comments

Signed-off-by: Erez Zadok <ezk@cs.sunysb.edu>
13 years agocleanup: ensure proper comment on every "#endif"
Erez_Zadok [Thu, 31 May 2007 19:30:55 +0000 (15:30 -0400)]
cleanup: ensure proper comment on every "#endif"

Signed-off-by: Erez Zadok <ezk@cs.sunysb.edu>
13 years agoRevert "Unionfs: Cleanup locking in new_dentry_private_data"
Erez_Zadok [Thu, 31 May 2007 03:34:23 +0000 (23:34 -0400)]
Revert "Unionfs: Cleanup locking in new_dentry_private_data"

This reverts commit ad2d6a5f5e0501ecf704bf2c748e9818ee8e019b.

Conflicts:

fs/unionfs/lookup.c

13 years agobugfix: ensure dentry/inode/mnt validity after a successful ioctl
Erez_Zadok [Thu, 31 May 2007 00:14:07 +0000 (20:14 -0400)]
bugfix: ensure dentry/inode/mnt validity after a successful ioctl

We call unionfs_partial_lookup in our queryfile ioctl method, so we can find
all instances of a lower object to report back to a suer.  This can violate
the fanout invariants (e.g., a regular file should have only one lower
object active at a time).  So we have to re-establish the invariants on the
lower dentries, inodes, and mnts.

Signed-off-by: Erez Zadok <ezk@cs.sunysb.edu>
13 years agobugfix: release superblock lock at end of ioctl method
Erez_Zadok [Thu, 31 May 2007 00:09:07 +0000 (20:09 -0400)]
bugfix: release superblock lock at end of ioctl method

Signed-off-by: Josef 'Jeff' Sipek <jsipek@cs.sunysb.edu>
Signed-off-by: Erez Zadok <ezk@cs.sunysb.edu>
13 years agobug fixes: revalidate dentries passed to all inode/super operations
Erez_Zadok [Wed, 30 May 2007 05:02:20 +0000 (01:02 -0400)]
bug fixes: revalidate dentries passed to all inode/super operations

Be sure to properly revalidate all dentry chains passed to all inode and
super_block operations.  Remove the older BUG_ON test is_valid_dentry().
This should help improve cache-coherency.

Signed-off-by: Erez Zadok <ezk@cs.sunysb.edu>
13 years agocode consistency: ensure that every #endif has a matching comment
Erez_Zadok [Wed, 30 May 2007 02:09:59 +0000 (22:09 -0400)]
code consistency: ensure that every #endif has a matching comment

Signed-off-by: Erez Zadok <ezk@cs.sunysb.edu>
13 years agoinvariants: don't complain if no lower dentry of a deleted dentry
Erez_Zadok [Tue, 29 May 2007 03:24:58 +0000 (23:24 -0400)]
invariants: don't complain if no lower dentry of a deleted dentry

Signed-off-by: Erez Zadok <ezk@cs.sunysb.edu>
13 years agoinvariants: don't complain about directories with some NULL lower objects
Erez_Zadok [Mon, 28 May 2007 23:47:35 +0000 (19:47 -0400)]
invariants: don't complain about directories with some NULL lower objects

Directories can have NULL lower objects in between start/end, but NOT if at
the start/end range.  We cannot verify that this dentry is a type=DIR,
because it may already be a negative dentry.  But if dbstart is greater than
dbend, we know that this couldn't have been a regular file: it had to have
been a directory.  In that case, don't complain about NULL lower objects in
in between start/end.

Signed-off-by: Erez Zadok <ezk@cs.sunysb.edu>
13 years agoRevert "invariants: don't complain about directories with some NULL lower objects"
Erez_Zadok [Mon, 28 May 2007 23:45:45 +0000 (19:45 -0400)]
Revert "invariants: don't complain about directories with some NULL lower objects"

This reverts commit 9d8ee83d478be30faf351af7628066ec6658efb5.

13 years agoinvariants: don't complain about directories with some NULL lower objects
Erez_Zadok [Mon, 28 May 2007 23:42:20 +0000 (19:42 -0400)]
invariants: don't complain about directories with some NULL lower objects

Directories can have NULL lower objects in between start/end, but NOT if at
the start/end range.  We cannot verify that this dentry is a type=DIR,
because it may already be a negative dentry.  But if dbstart is greater than
dbend, we know that this couldn't have been a regular file: it had to have
been a directory.  In that case, don't complain about NULL lower objects in
in between start/end.

Signed-off-by: Erez Zadok <ezk@cs.sunysb.edu>
13 years agocleanup: rewrite unionfs_partial_lookup more cleanly and add comments
Erez_Zadok [Mon, 28 May 2007 23:39:44 +0000 (19:39 -0400)]
cleanup: rewrite unionfs_partial_lookup more cleanly and add comments

Signed-off-by: Erez Zadok <ezk@cs.sunysb.edu>
13 years agobug fix: prevent null mntget warning messages in unionfs_lookup_backend
Erez_Zadok [Mon, 28 May 2007 23:38:14 +0000 (19:38 -0400)]
bug fix: prevent null mntget warning messages in unionfs_lookup_backend

Don't try to mntget a lower mnt at a point where the lower mnts don't yet
exist (the callers will have them later on).  Instead, get the lower mnt of
the sb->s_root.

Signed-off-by: Erez Zadok <ezk@cs.sunysb.edu>
13 years agodocumentation: expand comments to unionfs_lookup_backend
Erez_Zadok [Mon, 28 May 2007 23:36:02 +0000 (19:36 -0400)]
documentation: expand comments to unionfs_lookup_backend

Signed-off-by: Erez Zadok <ezk@cs.sunysb.edu>
13 years agocleanup: remove unnecessary call to inherit_mnt
Erez_Zadok [Mon, 28 May 2007 23:23:42 +0000 (19:23 -0400)]
cleanup: remove unnecessary call to inherit_mnt

Our unionfs_lookup doesn't need to call inherit_mnt: the lower mnts will be
correct here (but not if other methods such as unionfs_rename call our
lookup_backend helper function directly).

Signed-off-by: Erez Zadok <ezk@cs.sunysb.edu>
13 years agobug fix: don't warn if trying to mntput a "negative" union object
Erez_Zadok [Mon, 28 May 2007 23:21:21 +0000 (19:21 -0400)]
bug fix: don't warn if trying to mntput a "negative" union object

In unionfs_d_release, we may get negative dentries which don't have a lower
mnt.  In that case, don't print a warning message from unionfs_mntput about
trying to put a NULL mnt.

Signed-off-by: Erez Zadok <ezk@cs.sunysb.edu>
13 years agobug fix: get lower mnt from root dentry instead of lower dentry
Erez_Zadok [Mon, 28 May 2007 23:18:45 +0000 (19:18 -0400)]
bug fix: get lower mnt from root dentry instead of lower dentry

While copying-up a file, the union object hasn't been fully formed yet, so
we don't have all the lower dentries/mnts to use (the caller will have them
later on).  Therefore, we should use the sb->s_root's lower mnt here.  This
fixes a "NULL mnt" warning message that came from unionfs_mntget.

Signed-off-by: Erez Zadok <ezk@cs.sunysb.edu>
13 years agoUnionfs: allow users to override Makefile options in fs/unionfs/local.mk
Erez_Zadok [Sat, 26 May 2007 02:36:13 +0000 (22:36 -0400)]
Unionfs: allow users to override Makefile options in fs/unionfs/local.mk

Signed-off-by: Erez Zadok <ezk@cs.sunysb.edu>
13 years agodocumentation: better comment in branch_id_to_idx
Erez_Zadok [Sat, 26 May 2007 02:01:03 +0000 (22:01 -0400)]
documentation: better comment in branch_id_to_idx

Signed-off-by: Erez Zadok <ezk@cs.sunysb.edu>
13 years agobug fix: catch bad use of dirs= options (extra ':' separators)
Erez_Zadok [Fri, 25 May 2007 21:01:00 +0000 (17:01 -0400)]
bug fix: catch bad use of dirs= options (extra ':' separators)

Signed-off-by: Erez Zadok <ezk@cs.sunysb.edu>
13 years agounionfs: merge find_new_branch_index and branch_id_to_idx into one function
Erez_Zadok [Fri, 25 May 2007 20:20:36 +0000 (16:20 -0400)]
unionfs: merge find_new_branch_index and branch_id_to_idx into one function

Useful code cleanup and consolidation between the ODF code and non-ODF code.

Signed-off-by: Erez Zadok <ezk@cs.sunysb.edu>
13 years agounionfs: minor cleanup and better commenting in new_dentry_private_data
Erez_Zadok [Fri, 25 May 2007 19:47:44 +0000 (15:47 -0400)]
unionfs: minor cleanup and better commenting in new_dentry_private_data

Signed-off-by: Erez Zadok <ezk@cs.sunysb.edu>
13 years agoUnionfs: Cleanup locking in new_dentry_private_data
Erez_Zadok [Fri, 25 May 2007 19:45:30 +0000 (15:45 -0400)]
Unionfs: Cleanup locking in new_dentry_private_data

Signed-off-by: Josef 'Jeff' Sipek <jsipek@cs.sunysb.edu>
Conflicts:

fs/unionfs/lookup.c

13 years agobugfix: prevent null-deref oops if lower f/s is NFS (mmap writes)
Erez_Zadok [Fri, 25 May 2007 19:37:38 +0000 (15:37 -0400)]
bugfix: prevent null-deref oops if lower f/s is NFS (mmap writes)

This is a workaround fora deficiency of the linux MM layer, which doesn't
allow clean coordination between upper and lower pages in stackable layers.
We prevent an oops, but the cost is that we're not able to implement
writepages cleanly, not can we call the lower file system's writepages.

Signed-off-by: Erez Zadok <ezk@cs.sunysb.edu>
13 years agocleanup: use krealloc instead of open-coding it with kmalloc/kfree/memcpy
Erez_Zadok [Thu, 24 May 2007 04:08:03 +0000 (00:08 -0400)]
cleanup: use krealloc instead of open-coding it with kmalloc/kfree/memcpy

13 years agobugfix: prevent a NULL ptr deref if branch is nfs and using mmap writes
Erez_Zadok [Thu, 24 May 2007 01:27:58 +0000 (21:27 -0400)]
bugfix: prevent a NULL ptr deref if branch is nfs and using mmap writes

By some strange condition, the VFS can pass our unionfs_writepage a
writeback_control structure which, if passed as is to the lower ->writepage
-- here nfs_writepage, triggers a NULL ptr deref in NFS.  This fix works
around this issue, and results in a successfully written mmaped file.

Signed-off-by: Erez Zadok <ezk@cs.sunysb.edu>
13 years agobugfix: handle lower file systems that do not support writeable mappings
Erez_Zadok [Thu, 24 May 2007 01:26:01 +0000 (21:26 -0400)]
bugfix: handle lower file systems that do not support writeable mappings

File systems such as jffs2 (tested in 2.6.22-rc22) do not have a
->writepage.  Make sure unionfs detects that conditions and returns EINVAL
when a user tries to mmap such a file with VM_WRITE.

Signed-off-by: Erez Zadok <ezk@cs.sunysb.edu>
13 years agodocumentation: comment to ODF developers to address
Erez Zadok [Wed, 23 May 2007 03:53:25 +0000 (23:53 -0400)]
documentation: comment to ODF developers to address

Signed-off-by: Erez Zadok <ezk@cs.sunysb.edu>
13 years agobug fix: set lower inodes correctly after branch management succeeds
Erez Zadok [Wed, 23 May 2007 03:52:12 +0000 (23:52 -0400)]
bug fix: set lower inodes correctly after branch management succeeds

Signed-off-by: Erez Zadok <ezk@cs.sunysb.edu>
13 years agobug fix: prevent self deadlock with remount code in pivot_root scenarios
Erez Zadok [Wed, 23 May 2007 03:50:44 +0000 (23:50 -0400)]
bug fix: prevent self deadlock with remount code in pivot_root scenarios

Signed-off-by: Erez Zadok <ezk@cs.sunysb.edu>
13 years agodebugging: test for freed (0x5a) inodes
Erez Zadok [Wed, 23 May 2007 03:49:35 +0000 (23:49 -0400)]
debugging: test for freed (0x5a) inodes

Signed-off-by: Erez Zadok <ezk@cs.sunysb.edu>
13 years agobugfix: statfs lower file system properly
Erez Zadok [Tue, 22 May 2007 22:04:14 +0000 (18:04 -0400)]
bugfix: statfs lower file system properly

Get the correct lower dentry to use to statfs the first branch (always), now
works with disk-based as well as network file systems.  Also document the
code better.

Signed-off-by: Erez Zadok <ezk@cs.sunysb.edu>
13 years agoAdd Erez Zadok to maintainers list
Erez_Zadok [Sun, 20 May 2007 01:11:22 +0000 (21:11 -0400)]
Add Erez Zadok to maintainers list