linux-2.6
15 years agoMerge branch 'next' of git://git.kernel.org/pub/scm/linux/kernel/git/davej/cpufreq
Linus Torvalds [Tue, 6 Jan 2009 02:33:38 +0000 (18:33 -0800)] 
Merge branch 'next' of git://git./linux/kernel/git/davej/cpufreq

* 'next' of git://git.kernel.org/pub/scm/linux/kernel/git/davej/cpufreq:
  [CPUFREQ] Fix on resume, now preserves user policy min/max.
  [CPUFREQ] Add Celeron Core support to p4-clockmod.
  [CPUFREQ] add to speedstep-lib additional fsb values for core processors
  [CPUFREQ] Disable sysfs ui for p4-clockmod.
  [CPUFREQ] p4-clockmod: reduce noise
  [CPUFREQ] clean up speedstep-centrino and reduce cpumask_t usage

15 years agoMerge branch 'upstream-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/mfashe...
Linus Torvalds [Tue, 6 Jan 2009 02:32:43 +0000 (18:32 -0800)] 
Merge branch 'upstream-linus' of git://git./linux/kernel/git/mfasheh/ocfs2

* 'upstream-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/mfasheh/ocfs2: (138 commits)
  ocfs2: Access the right buffer_head in ocfs2_merge_rec_left.
  ocfs2: use min_t in ocfs2_quota_read()
  ocfs2: remove unneeded lvb casts
  ocfs2: Add xattr support checking in init_security
  ocfs2: alloc xattr bucket in ocfs2_xattr_set_handle
  ocfs2: calculate and reserve credits for xattr value in mknod
  ocfs2/xattr: fix credits calculation during index create
  ocfs2/xattr: Always updating ctime during xattr set.
  ocfs2/xattr: Remove extend_trans call and add its credits from the beginning
  ocfs2/dlm: Fix race during lockres mastery
  ocfs2/dlm: Fix race in adding/removing lockres' to/from the tracking list
  ocfs2/dlm: Hold off sending lockres drop ref message while lockres is migrating
  ocfs2/dlm: Clean up errors in dlm_proxy_ast_handler()
  ocfs2/dlm: Fix a race between migrate request and exit domain
  ocfs2: One more hamming code optimization.
  ocfs2: Another hamming code optimization.
  ocfs2: Don't hand-code xor in ocfs2_hamming_encode().
  ocfs2: Enable metadata checksums.
  ocfs2: Validate superblock with checksum and ecc.
  ocfs2: Checksum and ECC for directory blocks.
  ...

15 years agoMerge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs-2.6
Linus Torvalds [Tue, 6 Jan 2009 02:32:06 +0000 (18:32 -0800)] 
Merge branch 'for-linus' of git://git./linux/kernel/git/viro/vfs-2.6

* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs-2.6:
  inotify: fix type errors in interfaces
  fix breakage in reiserfs_new_inode()
  fix the treatment of jfs special inodes
  vfs: remove duplicate code in get_fs_type()
  add a vfs_fsync helper
  sys_execve and sys_uselib do not call into fsnotify
  zero i_uid/i_gid on inode allocation
  inode->i_op is never NULL
  ntfs: don't NULL i_op
  isofs check for NULL ->i_op in root directory is dead code
  affs: do not zero ->i_op
  kill suid bit only for regular files
  vfs: lseek(fd, 0, SEEK_CUR) race condition

15 years agomm lockless pagecache barrier fix
Nick Piggin [Tue, 6 Jan 2009 02:05:50 +0000 (03:05 +0100)] 
mm lockless pagecache barrier fix

An XFS workload showed up a bug in the lockless pagecache patch. Basically it
would go into an "infinite" loop, although it would sometimes be able to break
out of the loop! The reason is a missing compiler barrier in the "increment
reference count unless it was zero" case of the lockless pagecache protocol in
the gang lookup functions.

This would cause the compiler to use a cached value of struct page pointer to
retry the operation with, rather than reload it. So the page might have been
removed from pagecache and freed (refcount==0) but the lookup would not correctly
notice the page is no longer in pagecache, and keep attempting to increment the
refcount and failing, until the page gets reallocated for something else. This
isn't a data corruption because the condition will be detected if the page has
been reallocated. However it can result in a lockup.

Linus points out that ACCESS_ONCE is also required in that pointer load, even
if it's absence is not causing a bug on our particular build. The most general
way to solve this is just to put an rcu_dereference in radix_tree_deref_slot.

Assembly of find_get_pages,
before:
.L220:
        movq    (%rbx), %rax    #* ivtmp.1162, tmp82
        movq    (%rax), %rdi    #, prephitmp.1149
.L218:
        testb   $1, %dil        #, prephitmp.1149
        jne     .L217   #,
        testq   %rdi, %rdi      # prephitmp.1149
        je      .L203   #,
        cmpq    $-1, %rdi       #, prephitmp.1149
        je      .L217   #,
        movl    8(%rdi), %esi   # <variable>._count.counter, c
        testl   %esi, %esi      # c
        je      .L218   #,

after:
.L212:
        movq    (%rbx), %rax    #* ivtmp.1109, tmp81
        movq    (%rax), %rdi    #, ret
        testb   $1, %dil        #, ret
        jne     .L211   #,
        testq   %rdi, %rdi      # ret
        je      .L197   #,
        cmpq    $-1, %rdi       #, ret
        je      .L211   #,
        movl    8(%rdi), %esi   # <variable>._count.counter, c
        testl   %esi, %esi      # c
        je      .L212   #,

(notice the obvious infinite loop in the first example, if page->count remains 0)

Signed-off-by: Nick Piggin <npiggin@suse.de>
Reviewed-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
15 years agoi2o: Update my address
Alan Cox [Mon, 5 Jan 2009 14:04:40 +0000 (14:04 +0000)] 
i2o: Update my address

Signed-off-by: Alan Cox <alan@redhat.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
15 years agomm: update my address
Alan Cox [Mon, 5 Jan 2009 14:06:29 +0000 (14:06 +0000)] 
mm: update my address

Signed-off-by: Alan Cox <alan@redhat.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
15 years agoX86_DEBUGCTLMSR won't work on uml
Al Viro [Mon, 5 Jan 2009 17:19:02 +0000 (17:19 +0000)] 
X86_DEBUGCTLMSR won't work on uml

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
15 years agouml got broken by commit 30742d5c2277c325fb0e9d2d817d55a19995fe8f
Al Viro [Mon, 5 Jan 2009 17:18:52 +0000 (17:18 +0000)] 
uml got broken by commit 30742d5c2277c325fb0e9d2d817d55a19995fe8f

... if you revert a commit, revert the fixups elsewhere that had been
triggered by it.  Such as 8c56250f48347750c82ab18d98d647dcf99ca674
(lockdep, UML: fix compilation when CONFIG_TRACE_IRQFLAGS_SUPPORT is not set).

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
15 years agoget rid of the last symlink in uml build
Al Viro [Mon, 5 Jan 2009 17:18:42 +0000 (17:18 +0000)] 
get rid of the last symlink in uml build

We need to make asm-offsets.h contents visible for objects built
with userland headers.  Instead of creating a symlink, just have the
file with equivalent include (relative to location of header) created
once.  That kills the last symlink used in arch/um builds.

Additionally, both generated headers can become dependencies of
archprepare now, killing the misuse of prepare.

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
15 years agoinotify: fix type errors in interfaces
Michael Kerrisk [Mon, 5 Jan 2009 12:19:16 +0000 (07:19 -0500)] 
inotify: fix type errors in interfaces

The problems lie in the types used for some inotify interfaces, both at the kernel level and at the glibc level. This mail addresses the kernel problem. I will follow up with some suggestions for glibc changes.

For the sys_inotify_rm_watch() interface, the type of the 'wd' argument is
currently 'u32', it should be '__s32' .  That is Robert's suggestion, and
is consistent with the other declarations of watch descriptors in the
kernel source, in particular, the inotify_event structure in
include/linux/inotify.h:

struct inotify_event {
        __s32           wd;             /* watch descriptor */
        __u32           mask;           /* watch mask */
        __u32           cookie;         /* cookie to synchronize two events */
        __u32           len;            /* length (including nulls) of name */
        char            name[0];        /* stub for possible name */
};

The patch makes the changes needed for inotify_rm_watch().

Signed-off-by: Michael Kerrisk <mtk.manpages@googlemail.com>
Cc: Robert Love <rlove@google.com>
Cc: Vegard Nossum <vegard.nossum@gmail.com>
Cc: Ulrich Drepper <drepper@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
15 years agofix breakage in reiserfs_new_inode()
Al Viro [Fri, 2 Jan 2009 13:16:51 +0000 (08:16 -0500)] 
fix breakage in reiserfs_new_inode()

now that we use ih.key earlier, we need to do all its setup early enough

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
15 years agofix the treatment of jfs special inodes
Al Viro [Mon, 29 Dec 2008 12:40:31 +0000 (07:40 -0500)] 
fix the treatment of jfs special inodes

We used to put them on a single list, without any locking.  Racy.

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
15 years agovfs: remove duplicate code in get_fs_type()
Li Zefan [Thu, 25 Dec 2008 05:32:15 +0000 (13:32 +0800)] 
vfs: remove duplicate code in get_fs_type()

save 14 bytes:

   text    data     bss     dec     hex filename
   1354      32       4    1390     56e fs/filesystems.o.before
   text    data     bss     dec     hex filename
   1340      32       4    1376     560 fs/filesystems.o

Signed-off-by: Li Zefan <lizf@cn.fujitsu.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
15 years agoadd a vfs_fsync helper
Christoph Hellwig [Mon, 22 Dec 2008 20:11:15 +0000 (21:11 +0100)] 
add a vfs_fsync helper

Fsync currently has a fdatawrite/fdatawait pair around the method call,
and a mutex_lock/unlock of the inode mutex.  All callers of fsync have
to duplicate this, but we have a few and most of them don't quite get
it right.  This patch adds a new vfs_fsync that takes care of this.
It's a little more complicated as usual as ->fsync might get a NULL file
pointer and just a dentry from nfsd, but otherwise gets afile and we
want to take the mapping and file operations from it when it is there.

Notes on the fsync callers:

 - ecryptfs wasn't calling filemap_fdatawrite / filemap_fdatawait on the
    lower file
 - coda wasn't calling filemap_fdatawrite / filemap_fdatawait on the host
file, and returning 0 when ->fsync was missing
 - shm wasn't calling either filemap_fdatawrite / filemap_fdatawait nor
   taking i_mutex.  Now given that shared memory doesn't have disk
   backing not doing anything in fsync seems fine and I left it out of
   the vfs_fsync conversion for now, but in that case we might just
   not pass it through to the lower file at all but just call the no-op
   simple_sync_file directly.

[and now actually export vfs_fsync]

Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
15 years agosys_execve and sys_uselib do not call into fsnotify
Eric Paris [Wed, 17 Dec 2008 18:53:20 +0000 (13:53 -0500)] 
sys_execve and sys_uselib do not call into fsnotify

sys_execve and sys_uselib do not call into fsnotify so inotify does not get
open events for these types of syscalls.  This patch simply makes the
requisite fsnotify calls.

Signed-off-by: Eric Paris <eparis@redhat.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
15 years agozero i_uid/i_gid on inode allocation
Al Viro [Tue, 9 Dec 2008 14:34:39 +0000 (09:34 -0500)] 
zero i_uid/i_gid on inode allocation

... and don't bother in callers.  Don't bother with zeroing i_blocks,
while we are at it - it's already been zeroed.

i_mode is not worth the effort; it has no common default value.

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
15 years agoinode->i_op is never NULL
Al Viro [Thu, 4 Dec 2008 15:06:33 +0000 (10:06 -0500)] 
inode->i_op is never NULL

We used to have rather schizophrenic set of checks for NULL ->i_op even
though it had been eliminated years ago.  You'd need to go out of your
way to set it to NULL explicitly _and_ a bunch of code would die on
such inodes anyway.  After killing two remaining places that still
did that bogosity, all that crap can go away.

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
15 years agontfs: don't NULL i_op
Al Viro [Thu, 4 Dec 2008 14:59:23 +0000 (09:59 -0500)] 
ntfs: don't NULL i_op

it's already set to empty table (and no, ntfs doesn't have any explicit
checks for NULL ->i_op or NULL ->i_fop)

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
15 years agoisofs check for NULL ->i_op in root directory is dead code
Al Viro [Thu, 4 Dec 2008 14:57:25 +0000 (09:57 -0500)] 
isofs check for NULL ->i_op in root directory is dead code

for one thing it never happens, for another we check that inode
is a directory right after that place anyway (and we'd already
checked that reading it from disk has not failed).

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
15 years agoaffs: do not zero ->i_op
Al Viro [Thu, 4 Dec 2008 14:50:55 +0000 (09:50 -0500)] 
affs: do not zero ->i_op

it is already set to empty table and should never be NULL

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
15 years agokill suid bit only for regular files
Dmitri Monakhov [Mon, 1 Dec 2008 22:34:56 +0000 (14:34 -0800)] 
kill suid bit only for regular files

We don't have to do it because it is useless for non regular files.
In fact block device may trigger this path without dentry->d_inode->i_mutex.

(akpm: concerns were expressed (by me) about S_ISDIR inodes)

Signed-off-by: Dmitri Monakhov <dmonakhov@openvz.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
15 years agovfs: lseek(fd, 0, SEEK_CUR) race condition
Alain Knaff [Tue, 11 Nov 2008 01:08:08 +0000 (17:08 -0800)] 
vfs: lseek(fd, 0, SEEK_CUR) race condition

This patch fixes a race condition in lseek. While it is expected that
unpredictable behaviour may result while repositioning the offset of a
file descriptor concurrently with reading/writing to the same file
descriptor, this should not happen when merely *reading* the file
descriptor's offset.

Unfortunately, the only portable way in Unix to read a file
descriptor's offset is lseek(fd, 0, SEEK_CUR); however executing this
concurrently with read/write may mess up the position.

[with fixes from akpm]

Signed-off-by: Alain Knaff <alain@knaff.lu>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
15 years agoocfs2: Access the right buffer_head in ocfs2_merge_rec_left.
Tao Ma [Mon, 5 Jan 2009 06:45:24 +0000 (14:45 +0800)] 
ocfs2: Access the right buffer_head in ocfs2_merge_rec_left.

In commit "ocfs2: Use metadata-specific ocfs2_journal_access_*()
functions", the wrong buffer_head is accessed. So change it
to the right buffer_head.

Signed-off-by: Tao Ma <tao.ma@oracle.com>
Acked-by: Joel Becker <joel.becker@oracle.com>
Signed-off-by: Mark Fasheh <mfasheh@suse.com>
15 years agoocfs2: use min_t in ocfs2_quota_read()
Mark Fasheh [Thu, 25 Dec 2008 00:33:08 +0000 (16:33 -0800)] 
ocfs2: use min_t in ocfs2_quota_read()

This is preferred to min().

Signed-off-by: Mark Fasheh <mfasheh@suse.com>
15 years agoocfs2: remove unneeded lvb casts
Mark Fasheh [Thu, 25 Dec 2008 00:03:48 +0000 (16:03 -0800)] 
ocfs2: remove unneeded lvb casts

dlmglue.c has lots of code which casts the return value of ocfs2_dlm_lvb().
This is pointless however, as ocfs2_dlm_lvb() returns void *.

Signed-off-by: Mark Fasheh <mfasheh@suse.com>
15 years agoocfs2: Add xattr support checking in init_security
Tiger Yang [Wed, 17 Dec 2008 02:22:56 +0000 (10:22 +0800)] 
ocfs2: Add xattr support checking in init_security

We must check whether ocfs2 volume support xattr in init_security,
if not support xattr and security is enable, would cause failure of mknod.

Signed-off-by: Tiger Yang <tiger.yang@oracle.com>
Signed-off-by: Mark Fasheh <mfasheh@suse.com>
15 years agoocfs2: alloc xattr bucket in ocfs2_xattr_set_handle
Tiger Yang [Tue, 9 Dec 2008 08:43:08 +0000 (16:43 +0800)] 
ocfs2: alloc xattr bucket in ocfs2_xattr_set_handle

In extreme situation, may need xattr bucket for setting
security entry and acl entries during mknod. This only
happens when block size is too small.

Signed-off-by: Tiger Yang <tiger.yang@oracle.com>
Signed-off-by: Mark Fasheh <mfasheh@suse.com>
15 years agoocfs2: calculate and reserve credits for xattr value in mknod
Tiger Yang [Tue, 9 Dec 2008 08:42:51 +0000 (16:42 +0800)] 
ocfs2: calculate and reserve credits for xattr value in mknod

We extend the credits for xattr's large value in set_value_outside
before, this can give rise to a credits issue when we set one security
entry and two acl entries duing mknod. As we remove extend_trans form
set_value_outside, we must calculate and reserve the credits for
xattr's large value in mknod.

Signed-off-by: Tiger Yang <tiger.yang@oracle.com>
Signed-off-by: Mark Fasheh <mfasheh@suse.com>
15 years agoocfs2/xattr: fix credits calculation during index create
Tao Ma [Thu, 4 Dec 2008 22:20:56 +0000 (06:20 +0800)] 
ocfs2/xattr: fix credits calculation during index create

When creating a xattr index block, the old calculation forget
to add credits for the meta change of the alloc file. So add
more credits and more comments to explain it.

Signed-off-by: Tao Ma <tao.ma@oracle.com>
Signed-off-by: Mark Fasheh <mfasheh@suse.com>
15 years agoocfs2/xattr: Always updating ctime during xattr set.
Tao Ma [Thu, 4 Dec 2008 22:20:55 +0000 (06:20 +0800)] 
ocfs2/xattr: Always updating ctime during xattr set.

In xattr set, we should always update ctime if the operation goes
sucessfully. The old one mistakenly put it in ocfs2_xattr_set_entry
which is only called when we set xattr in inode or xattr block. The
side benefit is that it resolve the bug 1052 since in that scenario,
ocfs2_calc_xattr_set_need only calc out the xattr set credits while
ocfs2_xattr_set_entry update the inode also which isn't concerned with
the process of xattr set.

Signed-off-by: Tao Ma <tao.ma@oracle.com>
Signed-off-by: Mark Fasheh <mfasheh@suse.com>
15 years agoocfs2/xattr: Remove extend_trans call and add its credits from the beginning
Tao Ma [Thu, 4 Dec 2008 22:20:54 +0000 (06:20 +0800)] 
ocfs2/xattr: Remove extend_trans call and add its credits from the beginning

Actually, when setting a new xattr value, we know it from the very
beginning, and it isn't like the extension of bucket in which case
we can't figure it out. So remove ocfs2_extend_trans in that function
and calculate it before the transaction. It also relieve acl operation
from the worry about the side effect of ocfs2_extend_trans.

Signed-off-by: Tao Ma <tao.ma@oracle.com>
Signed-off-by: Mark Fasheh <mfasheh@suse.com>
15 years agoocfs2/dlm: Fix race during lockres mastery
Sunil Mushran [Tue, 16 Dec 2008 23:49:23 +0000 (15:49 -0800)] 
ocfs2/dlm: Fix race during lockres mastery

dlm_get_lock_resource() is supposed to return a lock resource with a proper
master. If multiple concurrent threads attempt to lookup the lockres for the
same lockid while the lock mastery in underway, one or more threads are likely
to return a lockres without a proper master.

This patch makes the threads wait in dlm_get_lock_resource() while the mastery
is underway, ensuring all threads return the lockres with a proper master.

This issue is known to be limited to users using the flock() syscall. For all
other fs operations, the ocfs2 dlmglue layer serializes the dlm op for each
lockid.

Users encountering this bug will see flock() return EINVAL and dmesg have the
following error:
ERROR: Dlm error "DLM_BADARGS" while calling dlmlock on resource <LOCKID>: bad api args

Reported-by: Coly Li <coyli@suse.de>
Signed-off-by: Sunil Mushran <sunil.mushran@oracle.com>
Signed-off-by: Mark Fasheh <mfasheh@suse.com>
15 years agoocfs2/dlm: Fix race in adding/removing lockres' to/from the tracking list
Sunil Mushran [Tue, 16 Dec 2008 23:49:22 +0000 (15:49 -0800)] 
ocfs2/dlm: Fix race in adding/removing lockres' to/from the tracking list

This patch adds a new lock, dlm->tracking_lock, to protect adding/removing
lockres' to/from the dlm->tracking_list. We were previously using dlm->spinlock
for the same, but that proved inadequate as we could be freeing a lockres from
a context that did not hold that lock. As the new lock only protects this list,
we can explicitly take it when removing the lockres from the tracking list.

This bug was exposed when testing multiple processes concurrently flock() the
same file.

Signed-off-by: Sunil Mushran <sunil.mushran@oracle.com>
Signed-off-by: Mark Fasheh <mfasheh@suse.com>
15 years agoocfs2/dlm: Hold off sending lockres drop ref message while lockres is migrating
Sunil Mushran [Tue, 16 Dec 2008 23:49:21 +0000 (15:49 -0800)] 
ocfs2/dlm: Hold off sending lockres drop ref message while lockres is migrating

During lockres purge, o2dlm sends a drop reference message to the lockres
master. This patch delays the message if the lockres is being migrated.

Fixes oss bugzilla#1012
http://oss.oracle.com/bugzilla/show_bug.cgi?id=1012

Signed-off-by: Sunil Mushran <sunil.mushran@oracle.com>
Signed-off-by: Mark Fasheh <mfasheh@suse.com>
15 years agoocfs2/dlm: Clean up errors in dlm_proxy_ast_handler()
Sunil Mushran [Tue, 16 Dec 2008 23:49:20 +0000 (15:49 -0800)] 
ocfs2/dlm: Clean up errors in dlm_proxy_ast_handler()

Patch cleans printed errors in dlm_proxy_ast_handler(). The errors now includes
the node number that sent the (b)ast. Also it reduces the number of endian swaps
of the cookie.

Signed-off-by: Sunil Mushran <sunil.mushran@oracle.com>
Signed-off-by: Mark Fasheh <mfasheh@suse.com>
15 years agoocfs2/dlm: Fix a race between migrate request and exit domain
Sunil Mushran [Tue, 16 Dec 2008 23:49:19 +0000 (15:49 -0800)] 
ocfs2/dlm: Fix a race between migrate request and exit domain

Patch address a racing migrate request message and an exit domain message.
Instead of blocking exit domains for the duration of the migrate, we ignore
failure to deliver that message. This is because an exiting domain should
not have any active locks and thus has no role to play in the migration.

Signed-off-by: Sunil Mushran <sunil.mushran@oracle.com>
Signed-off-by: Mark Fasheh <mfasheh@suse.com>
15 years agoocfs2: One more hamming code optimization.
Joel Becker [Tue, 16 Dec 2008 21:54:40 +0000 (13:54 -0800)] 
ocfs2: One more hamming code optimization.

The previous optimization used a fast find-highest-bit-set operation to
give us a good starting point in calc_code_bit().  This version lets the
caller cache the previous code buffer bit offset.  Thus, the next call
always starts where the last one left off.

This reduces the calculation another 39%, for a total 80% reduction from
the original, naive implementation.  At least, on my machine.  This also
brings the parity calculation to within an order of magnitude of the
crc32 calculation.

Signed-off-by: Joel Becker <joel.becker@oracle.com>
Signed-off-by: Mark Fasheh <mfasheh@suse.com>
15 years agoocfs2: Another hamming code optimization.
Joel Becker [Tue, 16 Dec 2008 02:24:33 +0000 (18:24 -0800)] 
ocfs2: Another hamming code optimization.

In the calc_code_bit() function, we must find all powers of two beneath
the code bit number, *after* it's shifted by those powers of two.  This
requires a loop to see where it ends up.

We can optimize it by starting at its most significant bit.  This shaves
32% off the time, for a total of 67.6% shaved off of the original, naive
implementation.

Signed-off-by: Joel Becker <joel.becker@oracle.com>
Signed-off-by: Mark Fasheh <mfasheh@suse.com>
15 years agoocfs2: Don't hand-code xor in ocfs2_hamming_encode().
Joel Becker [Tue, 16 Dec 2008 01:13:48 +0000 (17:13 -0800)] 
ocfs2: Don't hand-code xor in ocfs2_hamming_encode().

When I wrote ocfs2_hamming_encode(), I was following documentation of
the algorithm and didn't have quite the (possibly still imperfect) grasp
of it I do now.  As part of this, I literally hand-coded xor.  I would
test a bit, and then add that bit via xor to the parity word.

I can, of course, just do a single xor of the parity word and the source
word (the code buffer bit offset).  This cuts CPU usage by 53% on a
mostly populated buffer (an inode containing utmp.h inline).

Joel

Signed-off-by: Joel Becker <joel.becker@oracle.com>
Signed-off-by: Mark Fasheh <mfasheh@suse.com>
15 years agoocfs2: Enable metadata checksums.
Joel Becker [Fri, 17 Oct 2008 00:53:29 +0000 (17:53 -0700)] 
ocfs2: Enable metadata checksums.

Add OCFS2_FEATURE_INCOMPAT_META_ECC to the list of supported features.

Signed-off-by: Joel Becker <joel.becker@oracle.com>
Signed-off-by: Mark Fasheh <mfasheh@suse.com>
15 years agoocfs2: Validate superblock with checksum and ecc.
Joel Becker [Thu, 11 Dec 2008 23:04:14 +0000 (15:04 -0800)] 
ocfs2: Validate superblock with checksum and ecc.

The superblock is read via a raw call.  Validate it after we find it
from its signature.

Signed-off-by: Joel Becker <joel.becker@oracle.com>
Signed-off-by: Mark Fasheh <mfasheh@suse.com>
15 years agoocfs2: Checksum and ECC for directory blocks.
Joel Becker [Thu, 11 Dec 2008 01:58:22 +0000 (17:58 -0800)] 
ocfs2: Checksum and ECC for directory blocks.

Use the db_check field of ocfs2_dir_block_trailer to crc/ecc the
dirblocks.

Signed-off-by: Joel Becker <joel.becker@oracle.com>
Signed-off-by: Mark Fasheh <mfasheh@suse.com>
15 years agoocfs2: Add directory block trailers.
Mark Fasheh [Thu, 11 Dec 2008 01:36:25 +0000 (17:36 -0800)] 
ocfs2: Add directory block trailers.

Future ocfs2 features metaecc and indexed directories need to store a
little bit of data in each dirblock.  For compatibility, we place this
in a trailer at the end of the dirblock.  The trailer plays itself as an
empty dirent, so that if the features are turned off, it can be reused
without requiring a tunefs scan.

This code adds the trailer and validates it when the block is read in.

[ Mark is the original author, but I reinserted this code before his
  dir index work.  -- Joel ]

Signed-off-by: Joel Becker <joel.becker@oracle.com>
Signed-off-by: Mark Fasheh <mfasheh@suse.com>
15 years agoocfs2: Use proper journal_access function in xattr.c
Joel Becker [Wed, 10 Dec 2008 00:11:49 +0000 (16:11 -0800)] 
ocfs2: Use proper journal_access function in xattr.c

Change the rest of the naked ocfs2_journal_access() calls in
fs/ocfs2/xattr.c to use the appropriate ocfs2_journal_access_*() call
for their metadata type.

Signed-off-by: Joel Becker <joel.becker@oracle.com>
Signed-off-by: Mark Fasheh <mfasheh@suse.com>
15 years agoocfs2: Pass value buf to ocfs2_remove_value_outside().
Joel Becker [Wed, 10 Dec 2008 00:24:43 +0000 (16:24 -0800)] 
ocfs2: Pass value buf to ocfs2_remove_value_outside().

ocfs2_remove_value_outside() needs to know the type of buffer it is
looking at.  Pass in an ocfs2_xattr_value_buf.

Signed-off-by: Joel Becker <joel.becker@oracle.com>
Signed-off-by: Mark Fasheh <mfasheh@suse.com>
15 years agoocfs2: Use ocfs2_xattr_value_buf in ocfs2_xattr_set_entry().
Joel Becker [Tue, 9 Dec 2008 23:58:35 +0000 (15:58 -0800)] 
ocfs2: Use ocfs2_xattr_value_buf in ocfs2_xattr_set_entry().

ocfs2_xattr_set_entry is the function that knows what type of block it
is setting into.  This is what we wanted from ocfs2_xattr_value_buf.
Plus, moving the value buf up into ocfs2_xattr_set_entry() allows us to
pass it into ocfs2_xattr_set_value_outside() and ocfs2_xattr_cleanup().

Signed-off-by: Joel Becker <joel.becker@oracle.com>
Signed-off-by: Mark Fasheh <mfasheh@suse.com>
15 years agoocfs2: Pass value buf to ocfs2_xattr_update_entry().
Joel Becker [Tue, 9 Dec 2008 23:46:15 +0000 (15:46 -0800)] 
ocfs2: Pass value buf to ocfs2_xattr_update_entry().

ocfs2_xattr_update_entry() updates the entry portion of an xattr buffer.
This can be part of multiple metadata block types, so pass the buffer in
via an ocfs2_xattr_value_buf.

Signed-off-by: Joel Becker <joel.becker@oracle.com>
Signed-off-by: Mark Fasheh <mfasheh@suse.com>
15 years agoocfs2: Pass ocfs2_xattr_value_buf into ocfs2_xattr_value_truncate().
Joel Becker [Tue, 9 Dec 2008 23:01:04 +0000 (15:01 -0800)] 
ocfs2: Pass ocfs2_xattr_value_buf into ocfs2_xattr_value_truncate().

The callers of ocfs2_xattr_value_truncate() now pass in
ocfs2_xattr_value_bufs.  These callers are the ones that calculated the
xv location, so they are the right starting point.

Signed-off-by: Joel Becker <joel.becker@oracle.com>
Signed-off-by: Mark Fasheh <mfasheh@suse.com>
15 years agoocfs2: Pull ocfs2_xattr_value_buf up into ocfs2_xattr_value_truncate().
Joel Becker [Tue, 9 Dec 2008 22:36:50 +0000 (14:36 -0800)] 
ocfs2: Pull ocfs2_xattr_value_buf up into ocfs2_xattr_value_truncate().

Place an ocfs2_xattr_value_buf in ocfs2_xattr_value_truncate() and pass
it down to ocfs2_xattr_shrink_size().  We can also pass it into
ocfs2_xattr_extend_allocation(), replacing its ocfs2_xattr_value_buf.

Signed-off-by: Joel Becker <joel.becker@oracle.com>
Signed-off-by: Mark Fasheh <mfasheh@suse.com>
15 years agoocfs2: Pull ocfs2_xattr_value_buf up from __ocfs2_remove_xattr_range().
Joel Becker [Tue, 9 Dec 2008 22:30:41 +0000 (14:30 -0800)] 
ocfs2: Pull ocfs2_xattr_value_buf up from __ocfs2_remove_xattr_range().

Place an ocfs2_xattr_value_buf in __ocfs2_xattr_shrink_size() and pass
it down to __ocfs2_remove_xattr_range().

Signed-off-by: Joel Becker <joel.becker@oracle.com>
Signed-off-by: Mark Fasheh <mfasheh@suse.com>
15 years agoocfs2: Create ocfs2_xattr_value_buf.
Joel Becker [Tue, 9 Dec 2008 22:24:33 +0000 (14:24 -0800)] 
ocfs2: Create ocfs2_xattr_value_buf.

When an ocfs2 extended attribute is large enough to require its own
allocation tree, we root it with an ocfs2_xattr_value_root.  However,
these roots can be a part of inodes, xattr blocks, or xattr buckets.
Thus, they need a different journal access function for each container.

We wrap the bh, its journal access function, and the value root (xv) in
a structure called ocfs2_xattr_valu_buf.  This is a package that can
be passed around.  In this first pass, we simply pass it to the
extent tree code.

Signed-off-by: Joel Becker <joel.becker@oracle.com>
Signed-off-by: Mark Fasheh <mfasheh@suse.com>
15 years agoocfs2: Add ecc and checksums to ocfs2 xattr buckets.
Joel Becker [Fri, 5 Dec 2008 19:19:37 +0000 (11:19 -0800)] 
ocfs2: Add ecc and checksums to ocfs2 xattr buckets.

The xattr bucket can span multiple blocks on disk.  We have wrappers
for this structure in the code.  We use the new multi-block ecc calls to
calculate and validate the bucket.

Signed-off-by: Joel Becker <joel.becker@oracle.com>
Signed-off-by: Mark Fasheh <mfasheh@suse.com>
15 years agoocfs2: Use metadata-specific ocfs2_journal_access_*() functions.
Joel Becker [Sat, 18 Oct 2008 02:25:01 +0000 (19:25 -0700)] 
ocfs2: Use metadata-specific ocfs2_journal_access_*() functions.

The per-metadata-type ocfs2_journal_access_*() functions hook up jbd2
commit triggers and allow us to compute metadata ecc right before the
buffers are written out.  This commit provides ecc for inodes, extent
blocks, group descriptors, and quota blocks.  It is not safe to use
extened attributes and metaecc at the same time yet.

The ocfs2_extent_tree and ocfs2_path abstractions in alloc.c both hide
the type of block at their root.  Before, it didn't matter, but now the
root block must use the appropriate ocfs2_journal_access_*() function.
To keep this abstract, the structures now have a pointer to the matching
journal_access function and a wrapper call to call it.

A few places use naked ocfs2_write_block() calls instead of adding the
blocks to the journal.  We make sure to calculate their checksum and ecc
before the write.

Since we pass around the journal_access functions.  Let's typedef them
in ocfs2.h.

Signed-off-by: Joel Becker <joel.becker@oracle.com>
Signed-off-by: Mark Fasheh <mfasheh@suse.com>
15 years agoocfs2: Wrap up the common use cases of ocfs2_new_path().
Joel Becker [Sat, 18 Oct 2008 05:32:01 +0000 (22:32 -0700)] 
ocfs2: Wrap up the common use cases of ocfs2_new_path().

The majority of ocfs2_new_path() calls are:

ocfs2_new_path(path_root_bh(otherpath),
       path_root_el(otherpath));

Let's call that ocfs2_new_path_from_path().  The rest do similar things
from struct ocfs2_extent_tree.  Let's call those
ocfs2_new_path_from_et().  This will make the next change easier.

Signed-off-by: Joel Becker <joel.becker@oracle.com>
Signed-off-by: Mark Fasheh <mfasheh@suse.com>
15 years agoocfs2: Add journal_access functions with jbd2 triggers.
Joel Becker [Thu, 11 Sep 2008 22:53:07 +0000 (15:53 -0700)] 
ocfs2: Add journal_access functions with jbd2 triggers.

We create wrappers for ocfs2_journal_access() that are specific to the
type of metadata block.  This allows us to associate jbd2 commit
triggers with the block.  The triggers will compute metadata ecc in a
future commit.

Signed-off-by: Joel Becker <joel.becker@oracle.com>
Signed-off-by: Mark Fasheh <mfasheh@suse.com>
15 years agoocfs2: block read meta ecc.
Joel Becker [Fri, 17 Oct 2008 21:55:01 +0000 (14:55 -0700)] 
ocfs2: block read meta ecc.

Add block check calls to the read_block validate functions.  This is the
almost all of the read-side checking of metaecc.  xattr buckets are not checked
yet.   Writes are also unchecked, and so a read-write mount will quickly fail.

Signed-off-by: Joel Becker <joel.becker@oracle.com>
Signed-off-by: Mark Fasheh <mfasheh@suse.com>
15 years agoocfs2: Add a validation hook for quota block reads.
Joel Becker [Wed, 3 Dec 2008 01:44:05 +0000 (17:44 -0800)] 
ocfs2: Add a validation hook for quota block reads.

Add a currently-returns-success hook for quota block reads.  We'll be
adding checks to this.

Signed-off-by: Joel Becker <joel.becker@oracle.com>
Signed-off-by: Mark Fasheh <mfasheh@suse.com>
15 years agoocfs2: Add the underlying blockcheck code.
Joel Becker [Fri, 17 Oct 2008 00:54:25 +0000 (17:54 -0700)] 
ocfs2: Add the underlying blockcheck code.

This is the code that computes crc32 and ecc for ocfs2 metadata blocks.
There are high-level functions that check whether the filesystem has the
ecc feature, mid-level functions that work on a single block or array of
buffer_heads, and the low-level ecc hamming code that can handle
multiple buffers like crc32_le().

It's not hooked up to the filesystem yet.

Signed-off-by: Joel Becker <joel.becker@oracle.com>
Cc: Christoph Hellwig <hch@lst.de>
Signed-off-by: Mark Fasheh <mfasheh@suse.com>
15 years agoocfs2: Add the on-disk structures for metadata checksums.
Joel Becker [Fri, 17 Oct 2008 00:50:30 +0000 (17:50 -0700)] 
ocfs2: Add the on-disk structures for metadata checksums.

Define struct ocfs2_block_check, an 8-byte structure containing a 32bit
crc32_le and a 16bit hamming code ecc.  This will be used for metadata
checksums.  Add the structure to free spaces in the various metadata
structures.

Add the OCFS2_FEATURE_INCOMPAT_META_ECC bit.

Signed-off-by: Joel Becker <joel.becker@oracle.com>
Signed-off-by: Mark Fasheh <mfasheh@suse.com>
15 years agojbd2: Add buffer triggers
Joel Becker [Thu, 11 Sep 2008 22:35:47 +0000 (15:35 -0700)] 
jbd2: Add buffer triggers

Filesystems often to do compute intensive operation on some
metadata.  If this operation is repeated many times, it can be very
expensive.  It would be much nicer if the operation could be performed
once before a buffer goes to disk.

This adds triggers to jbd2 buffer heads.  Just before writing a metadata
buffer to the journal, jbd2 will optionally call a commit trigger associated
with the buffer.  If the journal is aborted, an abort trigger will be
called on any dirty buffers as they are dropped from pending
transactions.

ocfs2 will use this feature.

Initially I tried to come up with a more generic trigger that could be
used for non-buffer-related events like transaction completion.  It
doesn't tie nicely, because the information a buffer trigger needs
(specific to a journal_head) isn't the same as what a transaction
trigger needs (specific to a tranaction_t or perhaps journal_t).  So I
implemented a buffer set, with the understanding that
journal/transaction wide triggers should be implemented separately.

There is only one trigger set allowed per buffer.  I can't think of any
reason to attach more than one set.  Contrast this with a journal or
transaction in which multiple places may want to watch the entire
transaction separately.

The trigger sets are considered static allocation from the jbd2
perspective.  ocfs2 will just have one trigger set per block type,
setting the same set on every bh of the same type.

Signed-off-by: Joel Becker <joel.becker@oracle.com>
Cc: "Theodore Ts'o" <tytso@mit.edu>
Cc: <linux-ext4@vger.kernel.org>
Signed-off-by: Mark Fasheh <mfasheh@suse.com>
15 years agoocfs2/quota: Add QUOTA in mlog_attribute.
Tao Ma [Sun, 14 Dec 2008 22:03:41 +0000 (06:03 +0800)] 
ocfs2/quota: Add QUOTA in mlog_attribute.

A new mlog mask has to be added into mlog_attribute before it can
be really used in mlog. ML_QUOTA is only added in masklog.h, so
add it to the array to enable it.

Signed-off-by: Tao Ma <tao.ma@oracle.com>
Signed-off-by: Mark Fasheh <mfasheh@suse.com>
15 years agoocfs2: Pass xs->bucket into ocfs2_add_new_xattr_bucket().
Joel Becker [Wed, 26 Nov 2008 23:25:41 +0000 (15:25 -0800)] 
ocfs2: Pass xs->bucket into ocfs2_add_new_xattr_bucket().

Pass the actual target bucket for insert through to
ocfs2_add_new_xattr_bucket().  Now growing a bucket has no buffer_head
knowledge.

ocfs2_add_new_xattr_bucket() leavs xs->bucket in the proper state for
insert.  However, it doesn't update the rest of the search fields in xs,
so we still have to relse() and re-find.  That's OK, because everything
is cached.

Signed-off-by: Joel Becker <joel.becker@oracle.com>
Signed-off-by: Mark Fasheh <mfasheh@suse.com>
15 years agoocfs2: Move buckets up into ocfs2_add_new_xattr_bucket().
Joel Becker [Wed, 26 Nov 2008 23:08:44 +0000 (15:08 -0800)] 
ocfs2: Move buckets up into ocfs2_add_new_xattr_bucket().

Lift the buckets from ocfs2_add_new_xattr_cluster() up into
ocfs2_add_new_xattr_bucket().  Now ocfs2_add_new_xattr_cluster()
doesn't deal with buffer_heads.  In fact, we no longer have to play
get_bh() tricks at all.

Signed-off-by: Joel Becker <joel.becker@oracle.com>
Signed-off-by: Mark Fasheh <mfasheh@suse.com>
15 years agoocfs2: Move buckets up into ocfs2_add_new_xattr_cluster().
Joel Becker [Wed, 26 Nov 2008 22:43:31 +0000 (14:43 -0800)] 
ocfs2: Move buckets up into ocfs2_add_new_xattr_cluster().

Lift the buckets from ocfs2_adjust_xattr_cross_cluster() up into
ocfs2_add_new_xattr_cluster().  Now ocfs2_adjust_xattr_cross_cluster()
doesn't deal with buffer_heads.

Signed-off-by: Joel Becker <joel.becker@oracle.com>
Signed-off-by: Mark Fasheh <mfasheh@suse.com>
15 years agoocfs2: Pass buckets into ocfs2_mv_xattr_bucket_cross_cluster().
Joel Becker [Wed, 26 Nov 2008 22:25:21 +0000 (14:25 -0800)] 
ocfs2: Pass buckets into ocfs2_mv_xattr_bucket_cross_cluster().

Now that ocfs2_adjust_xattr_cross_cluster() has buckets, it can pass
them into ocfs2_mv_xattr_bucket_cross_cluster().  It no longer has to
care about buffer_heads.  The manipulation of first_bh and header_bh
moves up to ocfs2_adjust_xattr_cross_cluster().

Signed-off-by: Joel Becker <joel.becker@oracle.com>
Signed-off-by: Mark Fasheh <mfasheh@suse.com>
15 years agoocfs2: Start using buckets in ocfs2_adjust_xattr_cross_cluster().
Joel Becker [Wed, 26 Nov 2008 22:12:09 +0000 (14:12 -0800)] 
ocfs2: Start using buckets in ocfs2_adjust_xattr_cross_cluster().

We want to be passing around buckets instead of buffer_heads.  Let's get
them into ocfs2_adjust_xattr_cross_cluster.

Signed-off-by: Joel Becker <joel.becker@oracle.com>
Signed-off-by: Mark Fasheh <mfasheh@suse.com>
15 years agoocfs2: Use ocfs2_mv_xattr_buckets() in ocfs2_mv_xattr_bucket_cross_cluster().
Joel Becker [Wed, 26 Nov 2008 21:36:24 +0000 (13:36 -0800)] 
ocfs2: Use ocfs2_mv_xattr_buckets() in ocfs2_mv_xattr_bucket_cross_cluster().

Now that ocfs2_mv_xattr_buckets() can move a partial cluster's worth of
buckets, ocfs2_mv_xattr_bucket_cross_cluster() can use it.

Signed-off-by: Joel Becker <joel.becker@oracle.com>
Signed-off-by: Mark Fasheh <mfasheh@suse.com>
15 years agoocfs2: ocfs2_mv_xattr_buckets() can handle a partial cluster now.
Joel Becker [Wed, 26 Nov 2008 21:18:31 +0000 (13:18 -0800)] 
ocfs2: ocfs2_mv_xattr_buckets() can handle a partial cluster now.

If you look at ocfs2_mv_xattr_bucket_cross_cluster(), you'll notice that
two-thirds of the code is almost identical to ocfs2_mv_xattr_buckets().
The only difference is that ocfs2_mv_xattr_buckets() moves a whole
cluster's worth, while ocfs2_mv_xattr_bucket_cross_cluster() moves half
the cluster.

We change ocfs2_mv_xattr_buckets() to allow moving partial clusters.
The original caller of ocfs2_mv_xattr_buckets() still moves the whole
cluster's worth - it just passes a start_bucket of 0.

Signed-off-by: Joel Becker <joel.becker@oracle.com>
Signed-off-by: Mark Fasheh <mfasheh@suse.com>
15 years agoocfs2: Rename ocfs2_cp_xattr_cluster() to ocfs2_mv_xattr_buckets().
Joel Becker [Wed, 26 Nov 2008 21:02:18 +0000 (13:02 -0800)] 
ocfs2: Rename ocfs2_cp_xattr_cluster() to ocfs2_mv_xattr_buckets().

ocfs2_cp_xattr_cluster() takes the last cluster of an xattr extent,
copies its buckets to the front of a new extent, and then shrinks the bucket
count of the original extent.  So it's really moving the data, not
copying it.

While we're here, the function doesn't need a buffer_head for the old
extent, just the block number.

Signed-off-by: Joel Becker <joel.becker@oracle.com>
Signed-off-by: Mark Fasheh <mfasheh@suse.com>
15 years agoocfs2: Use ocfs2_cp_xattr_bucket() in ocfs2_mv_xattr_bucket_cross_cluster().
Joel Becker [Wed, 26 Nov 2008 03:58:16 +0000 (19:58 -0800)] 
ocfs2: Use ocfs2_cp_xattr_bucket() in ocfs2_mv_xattr_bucket_cross_cluster().

The buffer copy loop of ocfs2_mv_xattr_bucket_cross_cluster() actually
looks a lot like ocfs2_cp_xattr_bucket().  Let's just use that instead.
We also use bucket operations to update the buckets at the start of each
extent.

Signed-off-by: Joel Becker <joel.becker@oracle.com>
Signed-off-by: Mark Fasheh <mfasheh@suse.com>
15 years agoocfs2: Explain t_is_new in ocfs2_cp_xattr_cluster().
Joel Becker [Wed, 26 Nov 2008 03:00:15 +0000 (19:00 -0800)] 
ocfs2: Explain t_is_new in ocfs2_cp_xattr_cluster().

I was unsure of the JOURNAL_ACCESS parameters in
ocfs2_cp_xattr_cluster().  They're based on the function argument
't_is_new', but I couldn't quite figure out how t_is_new mapped to
allocation.  ocfs2_cp_xattr_cluster() actually overwrites the target,
regardless of t_is_new.

Well, I just figured it out.  So I'm adding a big fat comment for those
who come after me.  ocfs2_divide_xattr_cluster() has the same behavior.

Signed-off-by: Joel Becker <joel.becker@oracle.com>
Signed-off-by: Mark Fasheh <mfasheh@suse.com>
15 years agoocfs2: Dirty the entire first bucket in ocfs2_cp_xattr_cluster().
Joel Becker [Wed, 26 Nov 2008 02:36:42 +0000 (18:36 -0800)] 
ocfs2: Dirty the entire first bucket in ocfs2_cp_xattr_cluster().

ocfs2_cp_xattr_cluster() takes the last bucket of a full extent and
copies it over to a new extent.  It then updates the headers of both
extents to reflect the new state.  It is passed the first bh of
the first bucket in order to update that first extent's bucket count.
It reads and dirties the first bh of the new extent for the same reason.

However, future code wants to always dirty the entire bucket when it
is changed.  So it is changed to read the entire bucket it is updating
for both extents.

Signed-off-by: Joel Becker <joel.becker@oracle.com>
Signed-off-by: Mark Fasheh <mfasheh@suse.com>
15 years agoocfs2: Dirty the entire first bucket in ocfs2_extend_xattr_bucket()
Joel Becker [Wed, 26 Nov 2008 01:06:40 +0000 (17:06 -0800)] 
ocfs2: Dirty the entire first bucket in ocfs2_extend_xattr_bucket()

ocfs2_extend_xattr_bucket() takes an extent of buckets and shifts some
of them down to make room for a new xattr.  It is passed the first bh of
the first bucket, because that is where we store the number of buckets
in the extent.

However, future code wants to always dirty the entire bucket when it
is changed.  So let's pass the entire bucket into this function, skip
any block reads (we have them), and add the access/dirty logic.  We also
can skip passing in the target bucket bh - we only need its block
number.

Signed-off-by: Joel Becker <joel.becker@oracle.com>
Signed-off-by: Mark Fasheh <mfasheh@suse.com>
15 years agoocfs2: Narrow the transaction for deleting xattrs from a bucket.
Tao Ma [Thu, 11 Dec 2008 00:54:11 +0000 (08:54 +0800)] 
ocfs2: Narrow the transaction for deleting xattrs from a bucket.

We move the transaction into the loop because in
ocfs2_remove_extent, we will double the credits in function
ocfs2_extend_rotate_transaction. So if we have a large loop
number, we will soon waste much the journal space.

Signed-off-by: Tao Ma <tao.ma@oracle.com>
Signed-off-by: Joel Becker <joel.becker@oracle.com>
Signed-off-by: Mark Fasheh <mfasheh@suse.com>
15 years agoocfs2: Dirty the entire bucket in ocfs2_bucket_value_truncate()
Joel Becker [Tue, 25 Nov 2008 03:32:13 +0000 (19:32 -0800)] 
ocfs2: Dirty the entire bucket in ocfs2_bucket_value_truncate()

ocfs2_bucket_value_truncate() currently takes the first bh of the
bucket, and magically plays around with the value bh - even though
the bucket structure in the calling function already has it.

In addition, future code wants to always dirty the entire bucket when it
is changed.  So let's pass the entire bucket into this function, skip
any block reads (we have them), and add the access/dirty logic.

ocfs2_xattr_update_value_size() is no longer necessary, as it only did
one thing other than journal access/dirty.

Signed-off-by: Joel Becker <joel.becker@oracle.com>
Signed-off-by: Mark Fasheh <mfasheh@suse.com>
15 years agoocfs2/quota: sparse fixes for quota
Tao Ma [Mon, 24 Nov 2008 23:21:36 +0000 (07:21 +0800)] 
ocfs2/quota: sparse fixes for quota

Fix 2 minor things in quota. They are both found by sparse check.
1. an endian bug in ocfs2_local_quota_add_chunk.
2. change olq_alloc_dquot to static.

Signed-off-by: Tao Ma <tao.ma@oracle.com>
Signed-off-by: Mark Fasheh <mfasheh@suse.com>
15 years agoocfs2: fix indendation in ocfs2_dquot_drop_slow
Tao Ma [Thu, 27 Nov 2008 00:20:19 +0000 (16:20 -0800)] 
ocfs2: fix indendation in ocfs2_dquot_drop_slow

Signed-off-by: Tao Ma <tao.ma@oracle.com>
Signed-off-by: Mark Fasheh <mfasheh@suse.com>
15 years agoext4: Add default allocation routines for quota structures
Jan Kara [Tue, 25 Nov 2008 14:31:35 +0000 (15:31 +0100)] 
ext4: Add default allocation routines for quota structures

Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Mark Fasheh <mfasheh@suse.com>
15 years agoext3: Add default allocation routines for quota structures
Jan Kara [Tue, 25 Nov 2008 14:31:34 +0000 (15:31 +0100)] 
ext3: Add default allocation routines for quota structures

Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Mark Fasheh <mfasheh@suse.com>
15 years agoreiserfs: Add default allocation routines for quota structures
Jan Kara [Tue, 25 Nov 2008 14:31:33 +0000 (15:31 +0100)] 
reiserfs: Add default allocation routines for quota structures

Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Mark Fasheh <mfasheh@suse.com>
15 years agoquota: Export dquot_alloc() and dquot_destroy() functions
Jan Kara [Tue, 25 Nov 2008 14:31:32 +0000 (15:31 +0100)] 
quota: Export dquot_alloc() and dquot_destroy() functions

These are default functions for creating and destroying quota structures
and they should be used from filesystems.

Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Mark Fasheh <mfasheh@suse.com>
15 years agoquota: Unexport dqblk_v1.h and dqblk_v2.h
Jan Kara [Tue, 25 Nov 2008 14:31:31 +0000 (15:31 +0100)] 
quota: Unexport dqblk_v1.h and dqblk_v2.h

Unexport header files dqblk_v[12].h since except for quota format ID they
don't contain information userspace should be interested in. Move ID
definitions to quota.h.

Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Mark Fasheh <mfasheh@suse.com>
15 years agoocfs2: Fix build warnings (64-bit types vs long long)
Jan Kara [Tue, 25 Nov 2008 14:31:30 +0000 (15:31 +0100)] 
ocfs2: Fix build warnings (64-bit types vs long long)

fs/ocfs2/quota_local.c: In function 'olq_set_dquot':
fs/ocfs2/quota_local.c:844: warning: format '%lld' expects type 'long long int', but argument 7 has type '__le64'
fs/ocfs2/quota_local.c:844: warning: format '%lld' expects type 'long long int', but argument 8 has type '__le64'
fs/ocfs2/quota_local.c:844: warning: format '%lld' expects type 'long long int', but argument 7 has type '__le64'
fs/ocfs2/quota_local.c:844: warning: format '%lld' expects type 'long long int', but argument 8 has type '__le64'
fs/ocfs2/quota_local.c:844: warning: format '%lld' expects type 'long long int', but argument 7 has type '__le64'
fs/ocfs2/quota_local.c:844: warning: format '%lld' expects type 'long long int', but argument 8 has type '__le64'
fs/ocfs2/quota_global.c: In function '__ocfs2_sync_dquot':
fs/ocfs2/quota_global.c:457: warning: format '%lld' expects type 'long long int', but argument 8 has type 's64'
fs/ocfs2/quota_global.c:457: warning: format '%lld' expects type 'long long int', but argument 10 has type 's64'
fs/ocfs2/quota_global.c:457: warning: format '%lld' expects type 'long long int', but argument 8 has type 's64'
fs/ocfs2/quota_global.c:457: warning: format '%lld' expects type 'long long int', but argument 10 has type 's64'
fs/ocfs2/quota_global.c:457: warning: format '%lld' expects type 'long long int', but argument 8 has type 's64'
fs/ocfs2/quota_global.c:457: warning: format '%lld' expects type 'long long int', but argument 10 has type 's64'

Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Mark Fasheh <mfasheh@suse.com>
15 years agoocfs2: Make ocfs2_get_quota_block() consistent with ocfs2_read_quota_block()
Jan Kara [Tue, 25 Nov 2008 14:31:29 +0000 (15:31 +0100)] 
ocfs2: Make ocfs2_get_quota_block() consistent with ocfs2_read_quota_block()

Make function return error status and not buffer pointer so that it's
consistent with ocfs2_read_quota_block().

Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Mark Fasheh <mfasheh@suse.com>
15 years agoocfs2: Fix oops when extending quota files
Jan Kara [Tue, 25 Nov 2008 14:31:28 +0000 (15:31 +0100)] 
ocfs2: Fix oops when extending quota files

We have to mark buffer as uptodate before calling ocfs2_journal_access() and
ocfs2_set_buffer_uptodate() does not do this for us.

Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Mark Fasheh <mfasheh@suse.com>
15 years agoocfs2: Fix ocfs2_read_quota_block() error handling.
Joel Becker [Tue, 25 Nov 2008 14:31:27 +0000 (15:31 +0100)] 
ocfs2: Fix ocfs2_read_quota_block() error handling.

ocfs2_bread() has become ocfs2_read_virt_blocks(), with a prototype to
match ocfs2_read_blocks().  The quota code, converting from
ocfs2_bread(), wraps the call to ocfs2_read_virt_blocks() in
ocfs2_read_quota_block().  Unfortunately, the prototype of
ocfs2_read_quota_block() matches the old prototype of ocfs2_bread().

The problem is that ocfs2_bread() returned the buffer head, and callers
assumed that a NULL pointer was indicative of error.  It wasn't.  This
is why ocfs2_bread() took an int*err argument as well.

The new prototype of ocfs2_read_virt_blocks() avoids this error handling
confusion.  Let's change ocfs2_read_quota_block() to match.

Signed-off-by: Joel Becker <joel.becker@oracle.com>
Acked-by: Jan Kara <jack@suse.cz>
Signed-off-by: Mark Fasheh <mfasheh@suse.com>
15 years agoocfs2: Add missing initialization
Jan Kara [Tue, 25 Nov 2008 14:31:26 +0000 (15:31 +0100)] 
ocfs2: Add missing initialization

Add missing variable initialization to ocfs2_dquot_drop_slow().

Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Mark Fasheh <mfasheh@suse.com>
15 years agoocfs2: Use BH_JBDPrivateStart instead of BH_Unshadow
Mark Fasheh [Wed, 19 Nov 2008 01:16:47 +0000 (17:16 -0800)] 
ocfs2: Use BH_JBDPrivateStart instead of BH_Unshadow

This is safer. We no longer have to worry about tracking changes to
jbd_state_bits.

Signed-off-by: Mark Fasheh <mfasheh@suse.com>
15 years agojbd2: Add BH_JBDPrivateStart
Mark Fasheh [Wed, 19 Nov 2008 01:15:24 +0000 (17:15 -0800)] 
jbd2: Add BH_JBDPrivateStart

Add this so that file systems using JBD2 can safely allocate unused b_state
bits.

In this case, we add it so that Ocfs2 can define a single bit for tracking
the validation state of a buffer.

Acked-by: "Theodore Ts'o" <tytso@mit.edu>
Signed-off-by: Mark Fasheh <mfasheh@suse.com>
15 years agoocfs2: Enable quota accounting on mount, disable on umount
Jan Kara [Thu, 21 Aug 2008 18:13:17 +0000 (20:13 +0200)] 
ocfs2: Enable quota accounting on mount, disable on umount

Enable quota usage tracking on mount and disable it on umount. Also
add support for quota on and quota off quotactls and usrquota and
grpquota mount options. Add quota features among supported ones.

Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Mark Fasheh <mfasheh@suse.com>
15 years agoocfs2: Implement quota recovery
Jan Kara [Mon, 20 Oct 2008 21:50:38 +0000 (23:50 +0200)] 
ocfs2: Implement quota recovery

Implement functions for recovery after a crash. Functions just
read local quota file and sync info to global quota file.

Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Mark Fasheh <mfasheh@suse.com>
15 years agoocfs2: Periodic quota syncing
Mark Fasheh [Mon, 20 Oct 2008 13:36:47 +0000 (15:36 +0200)] 
ocfs2: Periodic quota syncing

This patch creates a work queue for periodic syncing of locally cached quota
information to the global quota files. We constantly queue a delayed work
item, to get the periodic behavior.

Signed-off-by: Mark Fasheh <mfasheh@suse.com>
Acked-by: Jan Kara <jack@suse.cz>
15 years agoocfs2: Add quota calls for allocation and freeing of inodes and space
Jan Kara [Thu, 9 Oct 2008 17:38:40 +0000 (19:38 +0200)] 
ocfs2: Add quota calls for allocation and freeing of inodes and space

Add quota calls for allocation and freeing of inodes and space, also update
estimates on number of needed credits for a transaction. Move out inode
allocation from ocfs2_mknod_locked() because vfs_dq_init() must be called
outside of a transaction.

Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Mark Fasheh <mfasheh@suse.com>
15 years agoocfs2: Implementation of local and global quota file handling
Jan Kara [Mon, 25 Aug 2008 17:56:50 +0000 (19:56 +0200)] 
ocfs2: Implementation of local and global quota file handling

For each quota type each node has local quota file. In this file it stores
changes users have made to disk usage via this node. Once in a while this
information is synced to global file (and thus with other nodes) so that
limits enforcement at least aproximately works.

Global quota files contain all the information about usage and limits. It's
mostly handled by the generic VFS code (which implements a trie of structures
inside a quota file). We only have to provide functions to convert structures
from on-disk format to in-memory one. We also have to provide wrappers for
various quota functions starting transactions and acquiring necessary cluster
locks before the actual IO is really started.

Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Mark Fasheh <mfasheh@suse.com>
15 years agoocfs2: Mark system files as not subject to quota accounting
Jan Kara [Thu, 21 Aug 2008 16:22:30 +0000 (18:22 +0200)] 
ocfs2: Mark system files as not subject to quota accounting

Mark system files as not subject to quota accounting. This prevents
possible recursions into quota code and thus deadlocks.

Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Mark Fasheh <mfasheh@suse.com>
15 years agoocfs2: Assign feature bits and system inodes to quota feature and quota files
Jan Kara [Wed, 20 Aug 2008 13:43:36 +0000 (15:43 +0200)] 
ocfs2: Assign feature bits and system inodes to quota feature and quota files

Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Mark Fasheh <mfasheh@suse.com>
15 years agoocfs2: Support nested transactions
Jan Kara [Wed, 27 Aug 2008 20:30:28 +0000 (22:30 +0200)] 
ocfs2: Support nested transactions

OCFS2 can easily support nested transactions. We just have to
take care and not spoil statistics acquire semaphore unnecessarily.

Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Mark Fasheh <mfasheh@suse.com>
15 years agoquota: Implement function for scanning active dquots
Jan Kara [Mon, 20 Oct 2008 15:05:00 +0000 (17:05 +0200)] 
quota: Implement function for scanning active dquots

OCFS2 needs to scan all active dquots once in a while and sync quota
information among cluster nodes. Provide a helper function for it so
that it does not have to reimplement internally a list which VFS
already has. Moreover this function is probably going to be useful
for other clustered filesystems if they decide to use VFS quotas.

Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Mark Fasheh <mfasheh@suse.com>
15 years agoquota: Add helpers to allow ocfs2 specific quota initialization, freeing and recovery
Jan Kara [Fri, 10 Oct 2008 14:12:23 +0000 (16:12 +0200)] 
quota: Add helpers to allow ocfs2 specific quota initialization, freeing and recovery

OCFS2 needs to peek whether quota structure is already in memory so
that it can avoid expensive cluster locking in that case. Similarly
when freeing dquots, it checks whether it is the last quota structure
user or not. Finally, it needs to get reference to dquot structure for
specified id and quota type when recovering quota file after crash.

Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Mark Fasheh <mfasheh@suse.com>
15 years agoquota: Update version number
Jan Kara [Thu, 30 Oct 2008 08:17:52 +0000 (09:17 +0100)] 
quota: Update version number

Increase reported version number of quota support since quota core has changed
significantly. Also remove __DQUOT_NUM_VERSION__ since nobody uses it.

Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Mark Fasheh <mfasheh@suse.com>