platform/kernel/linux-starfive.git
12 years agoext4: serialize fallocate with ext4_convert_unwritten_extents
Dmitry Monakhov [Fri, 5 Oct 2012 15:32:02 +0000 (11:32 -0400)]
ext4: serialize fallocate with ext4_convert_unwritten_extents

Fallocate should wait for pended ext4_convert_unwritten_extents()
otherwise following race may happen:

ftruncate( ,12288);
fallocate( ,0, 4096)
io_sibmit( ,0, 4096); /* Write to fallocated area, split extent if needed */
fallocate( ,0, 8192); /* Grow extent and broke assumption about extent */

Later kwork completion will do:
 ->ext4_convert_unwritten_extents (0, 4096)
   ->ext4_map_blocks(handle, inode, &map, EXT4_GET_BLOCKS_IO_CONVERT_EXT);
    ->ext4_ext_map_blocks() /* Will find new extent:  ex = [0,2] !!!!!! */
      ->ext4_ext_handle_uninitialized_extents()
        ->ext4_convert_unwritten_extents_endio()
        /* convert [0,2] extent to initialized, but only[0,1] was written */

Signed-off-by: Dmitry Monakhov <dmonakhov@openvz.org>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
12 years agoext4: fix ext4_flush_completed_IO wait semantics
Dmitry Monakhov [Fri, 5 Oct 2012 15:31:55 +0000 (11:31 -0400)]
ext4: fix ext4_flush_completed_IO wait semantics

BUG #1) All places where we call ext4_flush_completed_IO are broken
    because buffered io and DIO/AIO goes through three stages
    1) submitted io,
    2) completed io (in i_completed_io_list) conversion pended
    3) finished  io (conversion done)
    And by calling ext4_flush_completed_IO we will flush only
    requests which were in (2) stage, which is wrong because:
     1) punch_hole and truncate _must_ wait for all outstanding unwritten io
      regardless to it's state.
     2) fsync and nolock_dio_read should also wait because there is
        a time window between end_page_writeback() and ext4_add_complete_io()
        As result integrity fsync is broken in case of buffered write
        to fallocated region:
        fsync                                      blkdev_completion
 ->filemap_write_and_wait_range
                                                   ->ext4_end_bio
                                                     ->end_page_writeback
          <-- filemap_write_and_wait_range return
 ->ext4_flush_completed_IO
     sees empty i_completed_io_list but pended
     conversion still exist
                                                     ->ext4_add_complete_io

BUG #2) Race window becomes wider due to the 'ext4: completed_io
locking cleanup V4' patch series

This patch make following changes:
1) ext4_flush_completed_io() now first try to flush completed io and when
   wait for any outstanding unwritten io via ext4_unwritten_wait()
2) Rename function to more appropriate name.
3) Assert that all callers of ext4_flush_unwritten_io should hold i_mutex to
   prevent endless wait

Signed-off-by: Dmitry Monakhov <dmonakhov@openvz.org>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Reviewed-by: Jan Kara <jack@suse.cz>
12 years agoext4: fix mtime update in nodelalloc mode
Theodore Ts'o [Mon, 1 Oct 2012 03:04:56 +0000 (23:04 -0400)]
ext4: fix mtime update in nodelalloc mode

Commits 5e8830dc85d0 and 41c4d25f78c0 introduced a regression into
v3.6-rc1 for ext4 in nodealloc mode, such that mtime updates would not
take place for files modified via mmap if the page was already in the
page cache.  This would also affect ext3 file systems mounted using
the ext4 file system driver.

The problem was that ext4_page_mkwrite() had a shortcut which would
avoid calling __block_page_mkwrite() under some circumstances, and the
above two commit transferred the responsibility of calling
file_update_time() to __block_page_mkwrite --- which woudln't get
called in some circumstances.

Since __block_page_mkwrite() only has three callers,
block_page_mkwrite(), ext4_page_mkwrite, and nilfs_page_mkwrite(), the
best way to solve this is to move the responsibility for calling
file_update_time() to its caller.

This problem was found via xfstests #215 with a file system mounted
with -o nodelalloc.

Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Reviewed-by: Jan Kara <jack@suse.cz>
Cc: KONISHI Ryusuke <konishi.ryusuke@lab.ntt.co.jp>
Cc: stable@vger.kernel.org
12 years agoext4: fix ext_remove_space for punch_hole case
Dmitry Monakhov [Mon, 1 Oct 2012 03:03:50 +0000 (23:03 -0400)]
ext4: fix ext_remove_space for punch_hole case

Inode is allowed to have empty leaf only if it this is blockless inode.

Signed-off-by: Dmitry Monakhov <dmonakhov@openvz.org>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
12 years agoext4: punch_hole should wait for DIO writers
Dmitry Monakhov [Mon, 1 Oct 2012 03:03:42 +0000 (23:03 -0400)]
ext4: punch_hole should wait for DIO writers

punch_hole is the place where we have to wait for all existing writers
(writeback, aio, dio), but currently we simply flush pended end_io request
which is not sufficient. Other issue is that punch_hole performed w/o i_mutex
held which obviously result in dangerous data corruption due to
write-after-free.

This patch performs following changes:
- Guard punch_hole with i_mutex
- Recheck inode flags under i_mutex
- Block all new dio readers in order to prevent information leak caused by
  read-after-free pattern.
- punch_hole now wait for all writers in flight
  NOTE: XXX write-after-free race is still possible because new dirty pages
  may appear due to mmap(), and currently there is no easy way to stop
  writeback while punch_hole is in progress.

[ Fixed error return from ext4_ext_punch_hole() to make sure that we
  release i_mutex before returning EPERM or ETXTBUSY -- Ted ]

Signed-off-by: Dmitry Monakhov <dmonakhov@openvz.org>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
12 years agoext4: serialize truncate with owerwrite DIO workers
Dmitry Monakhov [Sat, 29 Sep 2012 04:58:26 +0000 (00:58 -0400)]
ext4: serialize truncate with owerwrite DIO workers

Jan Kara have spotted interesting issue:
There are  potential data corruption issue with  direct IO overwrites
racing with truncate:
 Like:
  dio write                      truncate_task
  ->ext4_ext_direct_IO
   ->overwrite == 1
    ->down_read(&EXT4_I(inode)->i_data_sem);
    ->mutex_unlock(&inode->i_mutex);
                               ->ext4_setattr()
                                ->inode_dio_wait()
                                ->truncate_setsize()
                                ->ext4_truncate()
                                 ->down_write(&EXT4_I(inode)->i_data_sem);
    ->__blockdev_direct_IO
     ->ext4_get_block
     ->submit_io()
    ->up_read(&EXT4_I(inode)->i_data_sem);
                                 # truncate data blocks, allocate them to
                                 # other inode - bad stuff happens because
                                 # dio is still in flight.

In order to serialize with truncate dio worker should grab extra i_dio_count
reference before drop i_mutex.

Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Dmitry Monakhov <dmonakhov@openvz.org>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
12 years agoext4: endless truncate due to nonlocked dio readers
Dmitry Monakhov [Sat, 29 Sep 2012 04:56:15 +0000 (00:56 -0400)]
ext4: endless truncate due to nonlocked dio readers

If we have enough aggressive DIO readers, truncate and other dio
waiters will wait forever inside inode_dio_wait(). It is reasonable
to disable nonlock DIO read optimization during truncate.

Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Dmitry Monakhov <dmonakhov@openvz.org>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
12 years agoext4: serialize unlocked dio reads with truncate
Dmitry Monakhov [Sat, 29 Sep 2012 04:55:23 +0000 (00:55 -0400)]
ext4: serialize unlocked dio reads with truncate

Current serialization will works only for DIO which holds
i_mutex, but nonlocked DIO following race is possible:

dio_nolock_read_task            truncate_task
->ext4_setattr()
 ->inode_dio_wait()
->ext4_ext_direct_IO
  ->ext4_ind_direct_IO
    ->__blockdev_direct_IO
      ->ext4_get_block
 ->truncate_setsize()
 ->ext4_truncate()
 #alloc truncated blocks
 #to other inode
      ->submit_io()
     #INFORMATION LEAK

In order to serialize with unlocked DIO reads we have to
rearrange wait sequence
1) update i_size first
2) if i_size about to be reduced wait for outstanding DIO requests
3) and only after that truncate inode blocks

Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Dmitry Monakhov <dmonakhov@openvz.org>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
12 years agoext4: serialize dio nonlocked reads with defrag workers
Dmitry Monakhov [Sat, 29 Sep 2012 04:41:21 +0000 (00:41 -0400)]
ext4: serialize dio nonlocked reads with defrag workers

Inode's block defrag and ext4_change_inode_journal_flag() may
affect nonlocked DIO reads result, so proper synchronization
required.

- Add missed inode_dio_wait() calls where appropriate
- Check inode state under extra i_dio_count reference.

Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Dmitry Monakhov <dmonakhov@openvz.org>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
12 years agoext4: completed_io locking cleanup
Dmitry Monakhov [Sat, 29 Sep 2012 04:14:55 +0000 (00:14 -0400)]
ext4: completed_io locking cleanup

Current unwritten extent conversion state-machine is very fuzzy.
- For unknown reason it performs conversion under i_mutex. What for?
  My diagnosis:
  We already protect extent tree with i_data_sem, truncate and punch_hole
  should wait for DIO, so the only data we have to protect is end_io->flags
  modification, but only flush_completed_IO and end_io_work modified this
  flags and we can serialize them via i_completed_io_lock.

  Currently all these games with mutex_trylock result in the following deadlock
   truncate:                          kworker:
    ext4_setattr                       ext4_end_io_work
    mutex_lock(i_mutex)
    inode_dio_wait(inode)  ->BLOCK
                             DEADLOCK<- mutex_trylock()
                                        inode_dio_done()
  #TEST_CASE1_BEGIN
  MNT=/mnt_scrach
  unlink $MNT/file
  fallocate -l $((1024*1024*1024)) $MNT/file
  aio-stress -I 100000 -O -s 100m -n -t 1 -c 10 -o 2 -o 3 $MNT/file
  sleep 2
  truncate -s 0 $MNT/file
  #TEST_CASE1_END

Or use 286's xfstests https://github.com/dmonakhov/xfstests/blob/devel/286

This patch makes state machine simple and clean:

(1) xxx_end_io schedule final extent conversion simply by calling
    ext4_add_complete_io(), which append it to ei->i_completed_io_list
    NOTE1: because of (2A) work should be queued only if
    ->i_completed_io_list was empty, otherwise the work is scheduled already.

(2) ext4_flush_completed_IO is responsible for handling all pending
    end_io from ei->i_completed_io_list
    Flushing sequence consists of following stages:
    A) LOCKED: Atomically drain completed_io_list to local_list
    B) Perform extents conversion
    C) LOCKED: move converted io's to to_free list for final deletion
             This logic depends on context which we was called from.
    D) Final end_io context destruction
    NOTE1: i_mutex is no longer required because end_io->flags modification
    is protected by ei->ext4_complete_io_lock

Full list of changes:
- Move all completion end_io related routines to page-io.c in order to improve
  logic locality
- Move open coded logic from various xx_end_xx routines to ext4_add_complete_io()
- remove EXT4_IO_END_FSYNC
- Improve SMP scalability by removing useless i_mutex which does not
  protect io->flags anymore.
- Reduce lock contention on i_completed_io_lock by optimizing list walk.
- Rename ext4_end_io_nolock to end4_end_io and make it static
- Check flush completion status to ext4_ext_punch_hole(). Because it is
  not good idea to punch blocks from corrupted inode.

Changes since V3 (in request to Jan's comments):
  Fall back to active flush_completed_IO() approach in order to prevent
  performance issues with nolocked DIO reads.
Changes since V2:
  Fix use-after-free caused by race truncate vs end_io_work

Signed-off-by: Dmitry Monakhov <dmonakhov@openvz.org>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
12 years agoext4: fix unwritten counter leakage
Dmitry Monakhov [Sat, 29 Sep 2012 03:36:25 +0000 (23:36 -0400)]
ext4: fix unwritten counter leakage

ext4_set_io_unwritten_flag() will increment i_unwritten counter, so
once we mark end_io with EXT4_END_IO_UNWRITTEN we have to revert it back
on error path.

 - add missed error checks to prevent counter leakage
 - ext4_end_io_nolock() will clear EXT4_END_IO_UNWRITTEN flag to signal
   that conversion finished.
 - add BUG_ON to ext4_free_end_io() to prevent similar leakage in future.

Visible effect of this bug is that unaligned aio_stress may deadlock

Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Dmitry Monakhov <dmonakhov@openvz.org>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
12 years agoext4: give i_aiodio_unwritten a more appropriate name
Dmitry Monakhov [Sat, 29 Sep 2012 03:24:52 +0000 (23:24 -0400)]
ext4: give i_aiodio_unwritten a more appropriate name

AIO/DIO prefix is wrong because it account unwritten extents which
also may be scheduled from buffered write endio

Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Dmitry Monakhov <dmonakhov@openvz.org>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
12 years agoext4: ext4_inode_info diet
Dmitry Monakhov [Sat, 29 Sep 2012 03:21:09 +0000 (23:21 -0400)]
ext4: ext4_inode_info diet

Generic inode has unused i_private pointer which may be used as cur_aio_dio
storage.

TODO: If cur_aio_dio will be passed as an argument to get_block_t this allow
      to have concurent AIO_DIO requests.

Reviewed-by: Zheng Liu <wenqing.lz@taobao.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Dmitry Monakhov <dmonakhov@openvz.org>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
12 years agoext4: convert to use leXX_add_cpu()
Wei Yongjun [Thu, 27 Sep 2012 13:37:53 +0000 (09:37 -0400)]
ext4: convert to use leXX_add_cpu()

Convert cpu_to_leXX(leXX_to_cpu(E1) + E2) to use leXX_add_cpu().

dpatch engine is used to auto generate this patch.
(https://github.com/weiyj/dpatch)

Signed-off-by: Wei Yongjun <yongjun_wei@trendmicro.com.cn>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
12 years agoext4: ext4_bread usage audit
Carlos Maiolino [Thu, 27 Sep 2012 13:31:33 +0000 (09:31 -0400)]
ext4: ext4_bread usage audit

When ext4_bread() returns NULL and err is set to zero, this means
there is no phyical block mapped to the specified logical block
number.  (Previous to commit 90b0a97323, err was uninitialized in this
case, which caused other problems.)

The directory handling routines use ext4_bread() in many places, the
fact that ext4_bread() now returns NULL with err set to zero could
cause problems since a number of these functions will simply return
the value of err if the result of ext4_bread() was the NULL pointer,
causing the caller of the function to think that the function was
successful.

Since directories should never contain holes, this case can only
happen if the file system is corrupted.  This commit audits all of the
callers of ext4_bread(), and makes sure they do the right thing if a
hole in a directory is found by ext4_bread().

Some ext4_bread() callers did not need any changes either because they
already had its own hole detector paths.

Signed-off-by: Carlos Maiolino <cmaiolino@redhat.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
12 years agofs: reserve fallocate flag codepoint
Theodore Ts'o [Thu, 27 Sep 2012 13:29:33 +0000 (09:29 -0400)]
fs: reserve fallocate flag codepoint

As discussed at the Plumber's Conference, reserve the bit 0x04 in
fallocate() to prevent collisions with a commonly used out-of-tree
patch which implements the no-hide-stale feature.

Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
12 years agoext4: remove redundant offset check in mext_check_arguments()
Wang Sheng-Hui [Thu, 27 Sep 2012 12:00:01 +0000 (08:00 -0400)]
ext4: remove redundant offset check in mext_check_arguments()

In the check code above, if orig_start != donor_start, we would
return -EINVAL. So here, orig_start should be equal with donor_start.
Remove the redundant check here.

Signed-off-by: Wang Sheng-Hui <shhuiw@gmail.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
12 years agoext4: don't clear orphan list on ro mount with errors
Eric Sandeen [Thu, 27 Sep 2012 03:30:12 +0000 (23:30 -0400)]
ext4: don't clear orphan list on ro mount with errors

If the file system contains errors and it is being mounted read-only,
don't clear the orphan list.  We should minimize changes to the file
system if it is mounted read-only.

Signed-off-by: Eric Sandeen <sandeen@redhat.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
12 years agojbd2: fix assertion failure in commit code due to lacking transaction credits
Jan Kara [Thu, 27 Sep 2012 03:11:13 +0000 (23:11 -0400)]
jbd2: fix assertion failure in commit code due to lacking transaction credits

ext4 users of data=journal mode with blocksize < pagesize were
occasionally hitting assertion failure in
jbd2_journal_commit_transaction() checking whether the transaction has
at least as many credits reserved as buffers attached.  The core of the
problem is that when a file gets truncated, buffers that still need
checkpointing or that are attached to the committing transaction are
left with buffer_mapped set. When this happens to buffers beyond i_size
attached to a page stradding i_size, subsequent write extending the file
will see these buffers and as they are mapped (but underlying blocks
were freed) things go awry from here.

The assertion failure just coincidentally (and in this case luckily as
we would start corrupting filesystem) triggers due to journal_head not
being properly cleaned up as well.

We fix the problem by unmapping buffers if possible (in lots of cases we
just need a buffer attached to a transaction as a place holder but it
must not be written out anyway).  And in one case, we just have to bite
the bullet and wait for transaction commit to finish.

CC: Josef Bacik <jbacik@fusionio.com>
Signed-off-by: Jan Kara <jack@suse.cz>
12 years agoext4: release donor reference when EXT4_IOC_MOVE_EXT ioctl fails
Djalal Harouni [Thu, 27 Sep 2012 02:58:50 +0000 (22:58 -0400)]
ext4: release donor reference when EXT4_IOC_MOVE_EXT ioctl fails

When the EXT4_IOC_MOVE_EXT ioctl() fails on bigalloc file systems, we
should jump to the 'mext_out' label to release the donor file reference.

Signed-off-by: Djalal Harouni <tixxdz@opendz.org>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
12 years agoext4: enable FITRIM ioctl on bigalloc file system
Lukas Czerner [Thu, 27 Sep 2012 02:21:21 +0000 (22:21 -0400)]
ext4: enable FITRIM ioctl on bigalloc file system

With a minor tweaks regarding minimum extent size to discard and
discarded bytes reporting the FITRIM can be enabled on bigalloc file
system and it works without any problem.

This patch fixes minlen handling and discarded bytes reporting to
take into consideration bigalloc enabled file systems and finally
removes the restriction and allow FITRIM to be used on file system with
bigalloc feature enabled.

Reviewed-by: Carlos Maiolino <cmaiolino@redhat.com>
Signed-off-by: Lukas Czerner <lczerner@redhat.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
12 years agoext4: fix fdatasync() for files with only i_size changes
Jan Kara [Thu, 27 Sep 2012 01:52:20 +0000 (21:52 -0400)]
ext4: fix fdatasync() for files with only i_size changes

Code tracking when transaction needs to be committed on fdatasync(2) forgets
to handle a situation when only inode's i_size is changed. Thus in such
situations fdatasync(2) doesn't force transaction with new i_size to disk
and that can result in wrong i_size after a crash.

Fix the issue by updating inode's i_datasync_tid whenever its size is
updated.

CC: <stable@vger.kernel.org> # >= 2.6.32
Reported-by: Kristian Nielsen <knielsen@knielsen-hq.org>
Signed-off-by: Jan Kara <jack@suse.cz>
12 years agoext4: always set i_op in ext4_mknod()
Bernd Schubert [Thu, 27 Sep 2012 01:24:57 +0000 (21:24 -0400)]
ext4: always set i_op in ext4_mknod()

ext4_special_inode_operations have their own ifdef CONFIG_EXT4_FS_XATTR
to mask those methods. And ext4_iget also always sets it, so there is
an inconsistency.

Signed-off-by: Bernd Schubert <bernd.schubert@itwm.fraunhofer.de>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Cc: stable@vger.kernel.org
12 years agoext4: remove unused function ext4_ext_check_cache
Lukas Czerner [Thu, 27 Sep 2012 01:09:06 +0000 (21:09 -0400)]
ext4: remove unused function ext4_ext_check_cache

Remove unused function ext4_ext_check_cache() and merge the code back to
the ext4_ext_in_cache().

Reviewed-by: Carlos Maiolino <cmaiolino@redhat.com>
Signed-off-by: Lukas Czerner <lczerner@redhat.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
12 years agoext4: use kmem_cache_zalloc instead of kmem_cache_alloc/memset
Wei Yongjun [Thu, 27 Sep 2012 00:43:37 +0000 (20:43 -0400)]
ext4: use kmem_cache_zalloc instead of kmem_cache_alloc/memset

Using kmem_cache_zalloc() instead of kmem_cache_alloc() and memset().

spatch with a semantic match is used to found this problem.
(http://coccinelle.lip6.fr/)

Signed-off-by: Wei Yongjun <yongjun_wei@trendmicro.com.cn>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
12 years agoext4: reimplement uninit extent optimization for move_extent_per_page()
Dmitry Monakhov [Wed, 26 Sep 2012 16:54:52 +0000 (12:54 -0400)]
ext4: reimplement uninit extent optimization for move_extent_per_page()

Uninitialized extent may became initialized(parallel writeback task)
at any moment after we drop i_data_sem, so we have to recheck extent's
state after we hold page's lock and i_data_sem.

If we about to change page's mapping we must hold page's lock in order to
serialize other users.

Signed-off-by: Dmitry Monakhov <dmonakhov@openvz.org>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
12 years agoext4: clean up online defrag bugs in move_extent_per_page()
Dmitry Monakhov [Wed, 26 Sep 2012 16:52:07 +0000 (12:52 -0400)]
ext4: clean up online defrag bugs in move_extent_per_page()

Non-full list of bugs:
1) uninitialized extent optimization does not hold page's lock,
   and simply replace brunches after that writeback code goes
   crazy because block mapping changed under it's feets
   kernel BUG at fs/ext4/inode.c:1434!  ( 288'th xfstress)

2) uninitialized extent may became initialized right after we
   drop i_data_sem, so extent state must be rechecked

3) Locked pages goes uptodate via following sequence:
   ->readpage(page); lock_page(page); use_that_page(page)
   But after readpage() one may invalidate it because it is
   uptodate and unlocked (reclaimer does that)
   As result kernel bug at include/linux/buffer_head.c:133!

4) We call write_begin() with already opened stansaction which
   result in following deadlock:
->move_extent_per_page()
  ->ext4_journal_start()-> hold journal transaction
  ->write_begin()
    ->ext4_da_write_begin()
      ->ext4_nonda_switch()
        ->writeback_inodes_sb_if_idle()  --> will wait for journal_stop()

5) try_to_release_page() may fail and it does fail if one of page's bh was
   pinned by journal

6) If we about to change page's mapping we MUST hold it's lock during entire
   remapping procedure, this is true for both pages(original and donor one)

Fixes:

- Avoid (1) and (2) simply by temproraly drop uninitialized extent handling
  optimization, this will be reimplemented later.

- Fix (3) by manually forcing page to uptodate state w/o dropping it's lock

- Fix (4) by rearranging existing locking:
  from: journal_start(); ->write_begin
  to: write_begin(); journal_extend()
- Fix (5) simply by checking retvalue
- Fix (6) by locking both (original and donor one) pages during extent swap
  with help of mext_page_double_lock()

Signed-off-by: Dmitry Monakhov <dmonakhov@openvz.org>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
12 years agoext4: online defrag is not supported for journaled files
Dmitry Monakhov [Wed, 26 Sep 2012 16:32:54 +0000 (12:32 -0400)]
ext4: online defrag is not supported for journaled files

Proper block swap for inodes with full journaling enabled is
truly non obvious task. In order to be on a safe side let's
explicitly disable it for now.

Signed-off-by: Dmitry Monakhov <dmonakhov@openvz.org>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Cc: stable@vger.kernel.org
12 years agoext4: move_extent code cleanup
Dmitry Monakhov [Wed, 26 Sep 2012 16:32:19 +0000 (12:32 -0400)]
ext4: move_extent code cleanup

- Remove usless checks, because it is too late to check that inode != NULL
  at the moment it was referenced several times.
- Double lock routines looks very ugly and locking ordering relays on
  order of i_ino, but other kernel code rely on order of pointers.
  Let's make them simple and clean.
- check that inodes belongs to the same SB as soon as possible.

Signed-off-by: Dmitry Monakhov <dmonakhov@openvz.org>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Cc: stable@vger.kernel.org
12 years agoext4: don't call update_backups() multiple times for the same bg
Tao Ma [Wed, 26 Sep 2012 04:08:57 +0000 (00:08 -0400)]
ext4: don't call update_backups() multiple times for the same bg

When performing an online resize, we add a bunch of groups at one time
in ext4_flex_group_add, so in most cases a lot of group descriptors
will be in the same group block. But in the end of this function,
update_backups will be called for every group descriptor and the same
block will be copied and journalled again and again.  It is really a
waste.

Fix things so we only update a particular bg descriptor block once and
skip subsequent updates of the same block.

Signed-off-by: Tao Ma <boyu.mt@taobao.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
12 years agoext4: fix double unlock buffer mess during fs-resize
Dmitry Monakhov [Wed, 26 Sep 2012 03:19:25 +0000 (23:19 -0400)]
ext4: fix double unlock buffer mess during fs-resize

bh_submit_read() is responsible for unlock bh on endio.  In addition,
we need to use bh_uptodate_or_lock() to avoid races.

Signed-off-by: Dmitry Monakhov <dmonakhov@openvz.org>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
12 years agoext4: check free inode count before allocating an inode
Yongqiang Yang [Mon, 24 Sep 2012 03:16:03 +0000 (23:16 -0400)]
ext4: check free inode count before allocating an inode

Recently, I ecountered some corrupted filesystems in which some
groups' free inode counts were 65535, it seemed that free inode
count was overflow.  This patch teaches ext4 to check free inode
count before allocaing an inode.

Signed-off-by: Yongqiang Yang <xiaoqiangnk@gmail.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
12 years agoext4: check free block counters in ext4_mb_find_by_goal
Yongqiang Yang [Mon, 24 Sep 2012 03:10:51 +0000 (23:10 -0400)]
ext4: check free block counters in ext4_mb_find_by_goal

Free block counters should be checked before doing allocation.

Signed-off-by: Yongqiang Yang <xiaoqiangnk@gmail.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
12 years agoext4: fix crash when accessing /proc/mounts concurrently
Herton Ronaldo Krzesinski [Mon, 24 Sep 2012 02:49:12 +0000 (22:49 -0400)]
ext4: fix crash when accessing /proc/mounts concurrently

The crash was caused by a variable being erronously declared static in
token2str().

In addition to /proc/mounts, the problem can also be easily replicated
by accessing /proc/fs/ext4/<partition>/options in parallel:

$ cat /proc/fs/ext4/<partition>/options > options.txt

... and then running the following command in two different terminals:

$ while diff /proc/fs/ext4/<partition>/options options.txt; do true; done

This is also the cause of the following a crash while running xfstests
#234, as reported in the following bug reports:

https://bugs.launchpad.net/bugs/1053019
https://bugzilla.kernel.org/show_bug.cgi?id=47731

Signed-off-by: Herton Ronaldo Krzesinski <herton.krzesinski@canonical.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Cc: Brad Figg <brad.figg@canonical.com>
Cc: stable@vger.kernel.org
12 years agoext4: remove erroneous ext4_superblock_csum_set() in update_backups()
Tao Ma [Thu, 20 Sep 2012 15:35:38 +0000 (11:35 -0400)]
ext4: remove erroneous ext4_superblock_csum_set() in update_backups()

The update_backups() function is used to backup all the metadata
blocks, so we should not take it for granted that 'data' is pointed to
a super block and use ext4_superblock_csum_set to calculate the
checksum there.  In case where the data is a group descriptor block,
it will corrupt the last group descriptor, and then e2fsck will
complain about it it.

As all the metadata checksums should already be OK when we do the
backup, remove the wrong ext4_superblock_csum_set and it should be
just fine.

Reported-by: "Theodore Ts'o" <tytso@mit.edu>
Signed-off-by: Tao Ma <boyu.mt@taobao.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Cc: stable@vger.kernel.org
12 years agoext4: fix potential deadlock in ext4_nonda_switch()
Theodore Ts'o [Thu, 20 Sep 2012 02:42:36 +0000 (22:42 -0400)]
ext4: fix potential deadlock in ext4_nonda_switch()

In ext4_nonda_switch(), if the file system is getting full we used to
call writeback_inodes_sb_if_idle().  The problem is that we can be
holding i_mutex already, and this causes a potential deadlock when
writeback_inodes_sb_if_idle() when it tries to take s_umount.  (See
lockdep output below).

As it turns out we don't need need to hold s_umount; the fact that we
are in the middle of the write(2) system call will keep the superblock
pinned.  Unfortunately writeback_inodes_sb() checks to make sure
s_umount is taken, and the VFS uses a different mechanism for making
sure the file system doesn't get unmounted out from under us.  The
simplest way of dealing with this is to just simply grab s_umount
using a trylock, and skip kicking the writeback flusher thread in the
very unlikely case that we can't take a read lock on s_umount without
blocking.

Also, we now check the cirteria for kicking the writeback thread
before we decide to whether to fall back to non-delayed writeback, so
if there are any outstanding delayed allocation writes, we try to get
them resolved as soon as possible.

   [ INFO: possible circular locking dependency detected ]
   3.6.0-rc1-00042-gce894ca #367 Not tainted
   -------------------------------------------------------
   dd/8298 is trying to acquire lock:
    (&type->s_umount_key#18){++++..}, at: [<c02277d4>] writeback_inodes_sb_if_idle+0x28/0x46

   but task is already holding lock:
    (&sb->s_type->i_mutex_key#8){+.+...}, at: [<c01ddcce>] generic_file_aio_write+0x5f/0xd3

   which lock already depends on the new lock.

   2 locks held by dd/8298:
    #0:  (sb_writers#2){.+.+.+}, at: [<c01ddcc5>] generic_file_aio_write+0x56/0xd3
    #1:  (&sb->s_type->i_mutex_key#8){+.+...}, at: [<c01ddcce>] generic_file_aio_write+0x5f/0xd3

   stack backtrace:
   Pid: 8298, comm: dd Not tainted 3.6.0-rc1-00042-gce894ca #367
   Call Trace:
    [<c015b79c>] ? console_unlock+0x345/0x372
    [<c06d62a1>] print_circular_bug+0x190/0x19d
    [<c019906c>] __lock_acquire+0x86d/0xb6c
    [<c01999db>] ? mark_held_locks+0x5c/0x7b
    [<c0199724>] lock_acquire+0x66/0xb9
    [<c02277d4>] ? writeback_inodes_sb_if_idle+0x28/0x46
    [<c06db935>] down_read+0x28/0x58
    [<c02277d4>] ? writeback_inodes_sb_if_idle+0x28/0x46
    [<c02277d4>] writeback_inodes_sb_if_idle+0x28/0x46
    [<c026f3b2>] ext4_nonda_switch+0xe1/0xf4
    [<c0271ece>] ext4_da_write_begin+0x27/0x193
    [<c01dcdb0>] generic_file_buffered_write+0xc8/0x1bb
    [<c01ddc47>] __generic_file_aio_write+0x1dd/0x205
    [<c01ddce7>] generic_file_aio_write+0x78/0xd3
    [<c026d336>] ext4_file_write+0x480/0x4a6
    [<c0198c1d>] ? __lock_acquire+0x41e/0xb6c
    [<c0180944>] ? sched_clock_cpu+0x11a/0x13e
    [<c01967e9>] ? trace_hardirqs_off+0xb/0xd
    [<c018099f>] ? local_clock+0x37/0x4e
    [<c0209f2c>] do_sync_write+0x67/0x9d
    [<c0209ec5>] ? wait_on_retry_sync_kiocb+0x44/0x44
    [<c020a7b9>] vfs_write+0x7b/0xe6
    [<c020a9a6>] sys_write+0x3b/0x64
    [<c06dd4bd>] syscall_call+0x7/0xb

Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Cc: stable@vger.kernel.org
12 years agoext4: speed up truncate/unlink by not using bforget() unless needed
Andrey Sidorov [Wed, 19 Sep 2012 18:14:53 +0000 (14:14 -0400)]
ext4: speed up truncate/unlink by not using bforget() unless needed

Do not iterate over data blocks scanning for bh's to forget as they're
never exist. This improves time taken by unlink / truncate syscall.
Tested by continuously truncating file that is being written by dd.
Another test is rm -rf of linux tree while tar unpacks it. With
ordered data mode condition unlikely(!tbh) was always met in
ext4_free_blocks. With journal data mode tbh was found only few times,
so optimisation is also possible.

Unlinking fallocated 60G file after doing sync && echo 3 >
/proc/sys/vm/drop_caches && time rm --help

X86 before (linux 3.6-rc4):
# time rm -f test1
real    0m2.710s
user    0m0.000s
sys     0m1.530s

X86 after:
# time rm -f test1
real    0m0.644s
user    0m0.003s
sys     0m0.060s

MIPS before (linux 2.6.37):
# time rm -f test1
real    0m 4.93s
user    0m 0.00s
sys     0m 4.61s

MIPS after:
# time rm -f test1
real    0m 0.16s
user    0m 0.00s
sys     0m 0.06s

Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Signed-off-by: Andrey Sidorov <qrxd43@motorola.com>
12 years agoext4: fix online resizing when the # of block groups is constant
Theodore Ts'o [Wed, 19 Sep 2012 04:55:56 +0000 (00:55 -0400)]
ext4: fix online resizing when the # of block groups is constant

Commit 1c6bd7173d66b3 introduced a regression where an online resize
operation which did not change the number of block groups would fail,
i.e:

mke2fs -t /dev/vdc 60000
mount /dev/vdc
resize2fs /dev/vdc 60001

This was due to a bug in the logic regarding when to try converting
the filesystem to use meta_bg.

Also fix up a number of other minor issues with the online resizing
code: (a) Fix a sparse warning; (b) only check to make sure the device
is large enough once, instead of multiple times through the resize
loop.

Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
12 years agoext4: make orphan functions be no-op in no-journal mode
Anatol Pomozov [Tue, 18 Sep 2012 17:38:59 +0000 (13:38 -0400)]
ext4: make orphan functions be no-op in no-journal mode

Instead of checking whether the handle is valid, we check if journal
is enabled. This avoids taking the s_orphan_lock mutex in all cases
when there is no journal in use, including the error paths where
ext4_orphan_del() is called with a handle set to NULL.

Signed-off-by: Anatol Pomozov <anatol.pomozov@gmail.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
12 years agoext4: re-enable -o discard functionality in no-journal mode
Theodore Ts'o [Tue, 18 Sep 2012 17:33:44 +0000 (13:33 -0400)]
ext4: re-enable -o discard functionality in no-journal mode

This is a revert of commit b56ff9d397ce, which removed the call to
ext4_issue_discard() to fix a BUG reported because
ext4_issue_discard() was being called from inside a block group
spinlock.  As it turns out this bug had already been fixed by Lukas
Czerner in commit 53fdcf992d61 by the simple expedient of moving when
we call ext4_issue_discard() outside the spinlock.

So it should be safe to re-enable this functionality, which I tested
by putting an BUG_ON(in_atomic) just after the restored callsite to
ext4_issue_discard().

Addresses-Google-Bug: #6750518

Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Cc: Anatol Pomozov <anatol.pomozov@gmail.com>
12 years agoext4: fix possible non-initialized variable in htree_dirblock_to_tree()
Carlos Maiolino [Tue, 18 Sep 2012 03:39:12 +0000 (23:39 -0400)]
ext4: fix possible non-initialized variable in htree_dirblock_to_tree()

htree_dirblock_to_tree() declares a non-initialized 'err' variable,
which is passed as a reference to another functions expecting them to
set this variable with their error codes.

It's passed to ext4_bread(), which then passes it to ext4_getblk(). If
ext4_map_blocks() returns 0 due to a lookup failure, leaving the
ext4_getblk() buffer_head uninitialized, it will make ext4_getblk()
return to ext4_bread() without initialize the 'err' variable, and
ext4_bread() will return to htree_dirblock_to_tree() with this variable
still uninitialized.  htree_dirblock_to_tree() will pass this variable
with garbage back to ext4_htree_fill_tree(), which expects a number of
directory entries added to the rb-tree. which, in case, might return a
fake non-zero value due the garbage left in the 'err' variable, leading
the kernel to an Oops in ext4_dx_readdir(), once this is expecting a
filled rb-tree node, when in turn it will have a NULL-ed one, causing an
invalid page request when trying to get a fname struct from this NULL-ed
rb-tree node in this line:

fname = rb_entry(info->curr_node, struct fname, rb_hash);

The patch itself initializes the err variable in
htree_dirblock_to_tree() to avoid usage mistakes by the called
functions, and also fix ext4_getblk() to return a initialized 'err'
variable when ext4_map_blocks() fails a lookup.

Signed-off-by: Carlos Maiolino <cmaiolino@redhat.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
12 years agoext4: do not enable delalloc by default for ext2
Theodore Ts'o [Tue, 18 Sep 2012 02:54:36 +0000 (22:54 -0400)]
ext4: do not enable delalloc by default for ext2

Signed-off-by: Brian Foster <bfoster@redhat.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
12 years agoext4: advertise the fact that the kernel supports meta_bg resizing
Theodore Ts'o [Thu, 13 Sep 2012 16:11:40 +0000 (12:11 -0400)]
ext4: advertise the fact that the kernel supports meta_bg resizing

Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
12 years agoext4: log a resize update to the console every 10 seconds
Theodore Ts'o [Thu, 13 Sep 2012 14:24:21 +0000 (10:24 -0400)]
ext4: log a resize update to the console every 10 seconds

For very long online resizes, a periodic update to the console log is
helpful for debugging and for progress reporting.

Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
12 years agoext4: convert file system to meta_bg if needed during resizing
Theodore Ts'o [Thu, 13 Sep 2012 14:19:24 +0000 (10:19 -0400)]
ext4: convert file system to meta_bg if needed during resizing

If we have run out of reserved gdt blocks, then clear the resize_inode
feature and enable the meta_bg feature, so that we can continue
resizing the file system seamlessly.

Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
12 years agoext4: set bg_itable_unused when resizing
Theodore Ts'o [Wed, 12 Sep 2012 18:32:42 +0000 (14:32 -0400)]
ext4: set bg_itable_unused when resizing

Set bg_itable_unused for file systems that have uninit_bg enabled.
This will speed up the first e2fsck run after the file system is
resized.

Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
12 years agoext4: add online resizing support for meta_bg and 64-bit file systems
Yongqiang Yang [Wed, 5 Sep 2012 05:33:50 +0000 (01:33 -0400)]
ext4: add online resizing support for meta_bg and 64-bit file systems

This patch adds support for resizing file systems with the meta_bg and
64bit features.

[ Added a fix by tytso to fix a divide by zero when resizing a
  filesystem from 14 TB to 18TB.  Also fixed overhead accounting for
  meta_bg file systems.]

Signed-off-by: Yongqiang Yang <xiaoqiangnk@gmail.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
12 years agoext4: grow the s_group_info array as needed
Theodore Ts'o [Wed, 5 Sep 2012 05:31:50 +0000 (01:31 -0400)]
ext4: grow the s_group_info array as needed

Previously we allocated the s_group_info array with enough space for
any future possible growth of the file system via online resize.  This
is unfortunate because it wastes memory, and it doesn't work for the
meta_bg scheme, since there is no limit based on the number of
reserved gdt blocks.  So add the code to grow the s_group_info array
as needed.

Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
12 years agoext4: grow the s_flex_groups array as needed when resizing
Theodore Ts'o [Wed, 5 Sep 2012 05:29:50 +0000 (01:29 -0400)]
ext4: grow the s_flex_groups array as needed when resizing

Previously, we allocated the s_flex_groups array to the maximum size
that the file system could be resized.  There was two problems with
this approach.  First, it wasted memory in the common case where the
file system was not resized.  Secondly, once we start allowing online
resizing using the meta_bg scheme, there is no maximum size that the
file system can be resized.  So instead, we need to grow the
s_flex_groups at inline resize time.

Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
12 years agoext4: avoid duplicate writes of the backup bg descriptor blocks
Yongqiang Yang [Wed, 5 Sep 2012 05:27:50 +0000 (01:27 -0400)]
ext4: avoid duplicate writes of the backup bg descriptor blocks

The resize code was needlessly writing the backup block group
descriptor blocks multiple times (once per block group) during an
online resize.

Signed-off-by: Yongqiang Yang <xiaoqiangnk@gmail.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Cc: stable@vger.kernel.org
12 years agoext4: don't copy non-existent gdt blocks when resizing
Yongqiang Yang [Wed, 5 Sep 2012 05:25:50 +0000 (01:25 -0400)]
ext4: don't copy non-existent gdt blocks when resizing

The resize code was copying blocks at the beginning of each block
group in order to copy the superblock and block group descriptor table
(gdt) blocks.  This was, unfortunately, being done even for block
groups that did not have super blocks or gdt blocks.  This is a
complete waste of perfectly good I/O bandwidth, to skip writing those
blocks for sparse bg's.

Signed-off-by: Yongqiang Yang <xiaoqiangnk@gmail.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Cc: stable@vger.kernel.org
12 years agoext4: report the original old blocks count in a debug message when resizing
Yongqiang Yang [Wed, 5 Sep 2012 05:23:50 +0000 (01:23 -0400)]
ext4: report the original old blocks count in a debug message when resizing

Avoid changing o_blocks_count, since it is used later when reporting
old blocks count in debug mode.

Signed-off-by: Yongqiang Yang <xiaoqiangnk@gmail.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
12 years agoext4: ignore last group w/o enough space when resizing instead of BUG'ing
Yongqiang Yang [Wed, 5 Sep 2012 05:21:50 +0000 (01:21 -0400)]
ext4: ignore last group w/o enough space when resizing instead of BUG'ing

If the last group does not have enough space for group tables, ignore
it instead of calling BUG_ON().

Reported-by: Daniel Drake <dsd@laptop.org>
Signed-off-by: Yongqiang Yang <xiaoqiangnk@gmail.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Cc: stable@vger.kernel.org
12 years agoext4: remove duplicated declarations in inode.c
Zheng Liu [Sun, 19 Aug 2012 22:07:40 +0000 (18:07 -0400)]
ext4: remove duplicated declarations in inode.c

In patch cb20d5188366f04d96d2e07b1240cc92170ade40, ext4_set_bh_endio
and ext4_end_io_buffer_write are declared at the beginning of inode.c,
and again later on in the middle of the file.  Remove the second set
of duplicated function declarations.

Signed-off-by: Zheng Liu <wenqing.lz@taobao.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
12 years agoext4: fix trivial typo in comment
Wang Sheng-Hui [Sun, 19 Aug 2012 02:38:07 +0000 (22:38 -0400)]
ext4: fix trivial typo in comment

Signed-off-by: Wang Sheng-Hui <shhuiw@gmail.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
12 years agoext4: no need to add inode to orphan list during hole punch
Ashish Sangwan [Sun, 19 Aug 2012 02:29:46 +0000 (22:29 -0400)]
ext4: no need to add inode to orphan list during hole punch

While performing punch hole for an inode, i_disksize is not changed.
So, there is no need to add the inode to orphan list.

Signed-off-by: Ashish Sangwan <ashish.sangwan2@gmail.com>
Signed-off-by: Namjae Jeon <linkinjeon@gmail.com>
Acked-by: Zheng Liu <wenqing.lz@taobao.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
12 years agojbd2: don't write superblock when if its empty
Eric Sandeen [Sun, 19 Aug 2012 02:29:40 +0000 (22:29 -0400)]
jbd2: don't write superblock when if its empty

This sequence:

# truncate --size=1g fsfile
# mkfs.ext4 -F fsfile
# mount -o loop,ro fsfile /mnt
# umount /mnt
# dmesg | tail

results in an IO error when unmounting the RO filesystem:

[  318.020828] Buffer I/O error on device loop1, logical block 196608
[  318.027024] lost page write due to I/O error on loop1
[  318.032088] JBD2: Error -5 detected when updating journal superblock for loop1-8.

This was a regression introduced by commit 24bcc89c7e7c: "jbd2: split
updating of journal superblock and marking journal empty".

Signed-off-by: Eric Sandeen <sandeen@redhat.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Cc: stable@vger.kernel.org
12 years agoext4: replace plain integer with NULL in super.c
Sachin Kamat [Sun, 19 Aug 2012 02:29:18 +0000 (22:29 -0400)]
ext4: replace plain integer with NULL in super.c

Fixes the following sparse warning:
fs/ext4/super.c:1672:45: warning: Using plain integer as NULL pointer

Signed-off-by: Sachin Kamat <sachin.kamat@linaro.org>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
12 years agoext4: drop lock_super()/unlock_super()
Theodore Ts'o [Fri, 17 Aug 2012 23:08:42 +0000 (19:08 -0400)]
ext4: drop lock_super()/unlock_super()

We don't need lock_super()/unlock_super() any more, since the places
where it is used, we are protected by the s_umount r/w semaphore.

Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Cc: Marco Stornelli <marco.stornelli@gmail.com>
12 years agoext4: return an error if kset_create_and_add fails in ext4_init_fs()
Theodore Ts'o [Fri, 17 Aug 2012 14:04:17 +0000 (10:04 -0400)]
ext4: return an error if kset_create_and_add fails in ext4_init_fs()

In the very unlikely case that kset_create_and_add() fails when the
ext4.ko module is being loaded (or during kernel startup) set err so
that it's clear that the module load failed.

https://bugzilla.kernel.org/show_bug.cgi?id=27912

Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
12 years agoext4: remove unused function argument 'order' in mb_find_extent()
Robin Dong [Fri, 17 Aug 2012 14:02:17 +0000 (10:02 -0400)]
ext4: remove unused function argument 'order' in mb_find_extent()

All the routines call mb_find_extent are setting argument 'order' to 0
just like:

mb_find_extent(e4b, 0, ex.fe_start, ex.fe_len, &ex);

therefore the useless argument should be removed.

Signed-off-by: Robin Dong <sanbai@taobao.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
12 years agoext4: remove unused macro MB_DEFAULT_MAX_GROUPS_TO_SCAN
Robin Dong [Fri, 17 Aug 2012 14:00:17 +0000 (10:00 -0400)]
ext4: remove unused macro MB_DEFAULT_MAX_GROUPS_TO_SCAN

Signed-off-by: Robin Dong <sanbai@taobao.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
12 years agoext4: check return value of blkdev_issue_flush()
Theodore Ts'o [Fri, 17 Aug 2012 13:58:17 +0000 (09:58 -0400)]
ext4: check return value of blkdev_issue_flush()

blkdev_issue_flush() can fail; make sure the error gets properly
propagated.

This is a port of the equivalent ext3 patch from commit 44f4f729e7a1.

Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
12 years agojbd2: check return value of blkdev_issue_flush()
Theodore Ts'o [Fri, 17 Aug 2012 13:56:17 +0000 (09:56 -0400)]
jbd2: check return value of blkdev_issue_flush()

blkdev_issue_flush() can fail; make sure the error gets properly
propagated.

This is a port of the equivalent jbd patch from commit 349ecd6a3c0e.

Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
12 years agoext4: make the zero-out chunk size tunable
Zheng Liu [Fri, 17 Aug 2012 13:54:17 +0000 (09:54 -0400)]
ext4: make the zero-out chunk size tunable

Currently in ext4 the length of zero-out chunk is set to 7 file system
blocks.  But if an inode has uninitailized extents from using
fallocate to preallocate space, and the workload issues many random
writes, this can cause a fragmented extent tree that will
unnecessarily grow the extent tree.

So create a new sysfs tunable, extent_max_zeroout_kb, which controls
the maximum size where blocks will be zeroed out instead of creating a
new uninitialized extent.  The default of this has been sent to 32kb.

CC: Zach Brown <zab@zabbo.net>
CC: Andreas Dilger <adilger@dilger.ca>
Signed-off-by: Zheng Liu <wenqing.lz@taobao.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
12 years agoext4: add missing space to trace message
Anatol Pomozov [Fri, 17 Aug 2012 13:52:17 +0000 (09:52 -0400)]
ext4: add missing space to trace message

Signed-off-by: Anatol Pomozov <anatol.pomozov@gmail.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
12 years agoext4: realign trace events structs to make it smaller
Anatol Pomozov [Fri, 17 Aug 2012 13:50:17 +0000 (09:50 -0400)]
ext4: realign trace events structs to make it smaller

Most hardware architectures require that data (including struct fields)
have to be aligned in memory. To make it happen compiler inserts padding
between struct fields if they are not aligned correctly.

Reorder fields to remove paddings and make structures denser. Making data
smaller saves some memory that is very important for trace events.
Tracing buffer has limited size and making objects smaller we can put more
of them without overflowing the tracing buffer.

To find data struct holes I used 'pahole -H 1 -E -I vmlinux.o' from
'dwarves' package.

Signed-off-by: Anatol Pomozov <anatol.pomozov@gmail.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
12 years agoext4: add max_dir_size_kb mount option
Theodore Ts'o [Fri, 17 Aug 2012 13:48:17 +0000 (09:48 -0400)]
ext4: add max_dir_size_kb mount option

Very large directories can cause significant performance problems, or
perhaps even invoke the OOM killer, if the process is running in a
highly constrained memory environment (whether it is VM's with a small
amount of memory or in a small memory cgroup).

So it is useful, in cloud server/data center environments, to be able
to set a filesystem-wide cap on the maximum size of a directory, to
ensure that directories never get larger than a sane size.  We do this
via a new mount option, max_dir_size_kb.  If there is an attempt to
grow the directory larger than max_dir_size_kb, the system call will
return ENOSPC instead.

Google-Bug-Id: 6863013

Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
12 years agoext4: don't load the block bitmap for block groups which have no space
Theodore Ts'o [Fri, 17 Aug 2012 13:46:17 +0000 (09:46 -0400)]
ext4: don't load the block bitmap for block groups which have no space

Add a short circuit check to ext4_mb_group_group() so that we don't
bother to load the block bitmap for a block group which does not have
any space available.  (Or which does not have enough space until we
are in desperation mode, i.e., when cr == 3.)

Resolves-bug: https://bugzilla.kernel.org/show_bug.cgi?id=45741
Reported-by: mirek@me.com
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
12 years agoext4: collapse a single extent tree block into the inode if possible
Theodore Ts'o [Fri, 17 Aug 2012 13:44:17 +0000 (09:44 -0400)]
ext4: collapse a single extent tree block into the inode if possible

If an inode has more than 4 extents, but then later some of the
extents are merged together, we can optimize the file system by moving
the extents up into the inode, and discarding the extent tree block.
This is important, because if there are a large number of inodes with
an external extent tree blocks where the contents could fit in the
inode, this can significantly increase the fsck time of the file
system.

Google-Bug-Id: 6801242

Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
12 years agoext4: fix kernel BUG on large-scale rm -rf commands
Theodore Ts'o [Fri, 17 Aug 2012 12:54:52 +0000 (08:54 -0400)]
ext4: fix kernel BUG on large-scale rm -rf commands

Commit 968dee7722: "ext4: fix hole punch failure when depth is greater
than 0" introduced a regression in v3.5.1/v3.6-rc1 which caused kernel
crashes when users ran run "rm -rf" on large directory hierarchy on
ext4 filesystems on RAID devices:

    BUG: unable to handle kernel NULL pointer dereference at 0000000000000028

    Process rm (pid: 18229, threadinfo ffff8801276bc000, task ffff880123631710)
    Call Trace:
     [<ffffffff81236483>] ? __ext4_handle_dirty_metadata+0x83/0x110
     [<ffffffff812353d3>] ext4_ext_truncate+0x193/0x1d0
     [<ffffffff8120a8cf>] ? ext4_mark_inode_dirty+0x7f/0x1f0
     [<ffffffff81207e05>] ext4_truncate+0xf5/0x100
     [<ffffffff8120cd51>] ext4_evict_inode+0x461/0x490
     [<ffffffff811a1312>] evict+0xa2/0x1a0
     [<ffffffff811a1513>] iput+0x103/0x1f0
     [<ffffffff81196d84>] do_unlinkat+0x154/0x1c0
     [<ffffffff8118cc3a>] ? sys_newfstatat+0x2a/0x40
     [<ffffffff81197b0b>] sys_unlinkat+0x1b/0x50
     [<ffffffff816135e9>] system_call_fastpath+0x16/0x1b
    Code: 8b 4d 20 0f b7 41 02 48 8d 04 40 48 8d 04 81 49 89 45 18 0f b7 49 02 48 83 c1 01 49 89 4d 00 e9 ae f8 ff ff 0f 1f 00 49 8b 45 28 <48> 8b 40 28 49 89 45 20 e9 85 f8 ff ff 0f 1f 80 00 00 00

    RIP  [<ffffffff81233164>] ext4_ext_remove_space+0xa34/0xdf0

This could be reproduced as follows:

The problem in commit 968dee7722 was that caused the variable 'i' to
be left uninitialized if the truncate required more space than was
available in the journal.  This resulted in the function
ext4_ext_truncate_extend_restart() returning -EAGAIN, which caused
ext4_ext_remove_space() to restart the truncate operation after
starting a new jbd2 handle.

Reported-by: Maciej Żenczykowski <maze@google.com>
Reported-by: Marti Raudsepp <marti@juffo.org>
Tested-by: Fengguang Wu <fengguang.wu@intel.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Cc: stable@vger.kernel.org
12 years agoext4: fix long mount times on very big file systems
Theodore Ts'o [Thu, 16 Aug 2012 15:59:04 +0000 (11:59 -0400)]
ext4: fix long mount times on very big file systems

Commit 8aeb00ff85a: "ext4: fix overhead calculation used by
ext4_statfs()" introduced a O(n**2) calculation which makes very large
file systems take forever to mount.  Fix this with an optimization for
non-bigalloc file systems.  (For bigalloc file systems the overhead
needs to be set in the the superblock.)

Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Cc: stable@vger.kernel.org
12 years agoext4: don't call ext4_error while block group is locked
Theodore Ts'o [Fri, 10 Aug 2012 17:57:52 +0000 (13:57 -0400)]
ext4: don't call ext4_error while block group is locked

While in ext4_validate_block_bitmap(), if an block allocation bitmap
is found to be invalid, we call ext4_error() while the block group is
still locked.  This causes ext4_commit_super() to call a function
which might sleep while in an atomic context.

There's no need to keep the block group locked at this point, so hoist
the ext4_error() call up to ext4_validate_block_bitmap() and release
the block group spinlock before calling ext4_error().

The reported stack trace can be found at:

http://article.gmane.org/gmane.comp.file-systems.ext4/33731

Reported-by: Dave Jones <davej@redhat.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Cc: stable@vger.kernel.org
12 years agoext4: avoid kmemcheck complaint from reading uninitialized memory
Theodore Ts'o [Mon, 6 Aug 2012 03:28:16 +0000 (23:28 -0400)]
ext4: avoid kmemcheck complaint from reading uninitialized memory

Commit 03179fe923 introduced a kmemcheck complaint in
ext4_da_get_block_prep() because we save and restore
ei->i_da_metadata_calc_last_lblock even though it is left
uninitialized in the case where i_da_metadata_calc_len is zero.

This doesn't hurt anything, but silencing the kmemcheck complaint
makes it easier for people to find real bugs.

Addresses https://bugzilla.kernel.org/show_bug.cgi?id=45631
(which is marked as a regression).

Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Cc: stable@vger.kernel.org
12 years agoext4: make sure the journal sb is written in ext4_clear_journal_err()
Theodore Ts'o [Sun, 5 Aug 2012 23:04:57 +0000 (19:04 -0400)]
ext4: make sure the journal sb is written in ext4_clear_journal_err()

After we transfer set the EXT4_ERROR_FS bit in the file system
superblock, it's not enough to call jbd2_journal_clear_err() to clear
the error indication from journal superblock --- we need to call
jbd2_journal_update_sb_errno() as well.  Otherwise, when the root file
system is mounted read-only, the journal is replayed, and the error
indicator is transferred to the superblock --- but the s_errno field
in the jbd2 superblock is left set (since although we cleared it in
memory, we never flushed it out to disk).

This can end up confusing e2fsck.  We should make e2fsck more robust
in this case, but the kernel shouldn't be leaving things in this
confused state, either.

Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Cc: stable@kernel.org
12 years agoLinux 3.6-rc1
Linus Torvalds [Thu, 2 Aug 2012 23:38:10 +0000 (16:38 -0700)]
Linux 3.6-rc1

12 years agoMerge branch 'for-linus-3.6' of git://dev.laptop.org/users/dilinger/linux-olpc
Linus Torvalds [Thu, 2 Aug 2012 18:52:39 +0000 (11:52 -0700)]
Merge branch 'for-linus-3.6' of git://dev.laptop.org/users/dilinger/linux-olpc

Pull OLPC platform updates from Andres Salomon:
 "These move the OLPC Embedded Controller driver out of
  arch/x86/platform and into drivers/platform/olpc.

  OLPC machines are now ARM-based (which means lots of x86 and ARM
  changes), but are typically pretty self-contained..  so it makes more
  sense to go through a separate OLPC tree after getting the appropriate
  review/ACKs."

* 'for-linus-3.6' of git://dev.laptop.org/users/dilinger/linux-olpc:
  x86: OLPC: move s/r-related EC cmds to EC driver
  Platform: OLPC: move global variables into priv struct
  Platform: OLPC: move debugfs support from x86 EC driver
  x86: OLPC: switch over to using new EC driver on x86
  Platform: OLPC: add a suspended flag to the EC driver
  Platform: OLPC: turn EC driver into a platform_driver
  Platform: OLPC: allow EC cmd to be overridden, and create a workqueue to call it
  drivers: OLPC: update various drivers to include olpc-ec.h
  Platform: OLPC: add a stub to drivers/platform/ for the OLPC EC driver

12 years agoMerge tag 'dt2' of git://git.kernel.org/pub/scm/linux/kernel/git/arm/arm-soc
Linus Torvalds [Thu, 2 Aug 2012 18:50:24 +0000 (11:50 -0700)]
Merge tag 'dt2' of git://git./linux/kernel/git/arm/arm-soc

Pull arm-soc Marvell Orion device-tree updates from Olof Johansson:
 "This contains a set of device-tree conversions for Marvell Orion
  platforms that were staged early but took a few tries to get the
  branch into a format where it was suitable for us to pick up.

  Given that most people working on these platforms are hobbyists with
  limited time, we were a bit more flexible with merging it even though
  it came in late."

* tag 'dt2' of git://git.kernel.org/pub/scm/linux/kernel/git/arm/arm-soc: (21 commits)
  ARM: Kirkwood: Replace mrvl with marvell
  ARM: Kirkwood: Describe GoFlex Net LEDs and SATA in DT.
  ARM: Kirkwood: Describe Dreamplug LEDs in DT.
  ARM: Kirkwood: Describe iConnects LEDs in DT.
  ARM: Kirkwood: Describe iConnects temperature sensor in DT.
  ARM: Kirkwood: Describe IB62x0 LEDs in DT.
  ARM: Kirkwood: Describe IB62x0 gpio-keys in DT.
  ARM: Kirkwood: Describe DNS32? gpio-keys in DT.
  ARM: Kirkwood: Move common portions into a kirkwood-dnskw.dtsi
  ARM: Kirkwood: Replace DNS-320/DNS-325 leds with dt bindings
  ARM: Kirkwood: Describe DNS325 temperature sensor in DT.
  ARM: Kirkwood: Use DT to configure SATA device.
  ARM: kirkwood: use devicetree for SPI on dreamplug
  ARM: kirkwood: Add LS-XHL and LS-CHLv2 support
  ARM: Kirkwood: Initial DTS support for Kirkwood GoFlex Net
  ARM: Kirkwood: Add basic device tree support for QNAP TS219.
  ATA: sata_mv: Add device tree support
  ARM: Orion: DTify the watchdog timer.
  ARM: Orion: Add arch support needed for I2C via DT.
  ARM: kirkwood: use devicetree for orion-spi
  ...

Conflicts:
drivers/watchdog/orion_wdt.c

12 years agoMerge tag 'pm2' of git://git.kernel.org/pub/scm/linux/kernel/git/arm/arm-soc
Linus Torvalds [Thu, 2 Aug 2012 18:48:54 +0000 (11:48 -0700)]
Merge tag 'pm2' of git://git./linux/kernel/git/arm/arm-soc

Pull arm-soc cpuidle enablement for OMAP from Olof Johansson:
 "Coupled cpuidle was meant to merge for 3.5 through Len Brown's tree,
  but didn't go in because the pull request ended up rejected.  So it
  just got merged, and we got this staged branch that enables the
  coupled cpuidle code on OMAP.

  With a stable git workflow from the other maintainer we could have
  staged this earlier, but that wasn't the case so we have had to merge
  it late.

  The alternative is to hold it off until 3.7 but given that the code is
  well-isolated to OMAP and they are eager to see it go in, I didn't
  push back hard in that direction."

* tag 'pm2' of git://git.kernel.org/pub/scm/linux/kernel/git/arm/arm-soc:
  ARM: OMAP4: CPUidle: Open broadcast clock-event device.
  ARM: OMAP4: CPUidle: add synchronization for coupled idle states
  ARM: OMAP4: CPUidle: Use coupled cpuidle states to implement SMP cpuidle.
  ARM: OMAP: timer: allow gp timer clock-event to be used on both cpus

12 years agoMerge tag 'fixes-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/arm...
Linus Torvalds [Thu, 2 Aug 2012 18:48:20 +0000 (11:48 -0700)]
Merge tag 'fixes-for-linus' of git://git./linux/kernel/git/arm/arm-soc

Pull ARM SoC fixes from Olof Johansson:
 "A few fixes for merge window fallout, and a bugfix for timer resume on
  PRIMA2."

* tag 'fixes-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/arm/arm-soc:
  ARM: mmp: add missing irqs.h
  arm: mvebu: fix typo in .dtsi comment for Armada XP SoCs
  ARM: PRIMA2: delete redundant codes to restore LATCHED when timer resumes
  ARM: mxc: Include missing irqs.h header

12 years agoMerge tag 'sh-for-linus' of git://github.com/pmundt/linux-sh
Linus Torvalds [Thu, 2 Aug 2012 18:45:42 +0000 (11:45 -0700)]
Merge tag 'sh-for-linus' of git://github.com/pmundt/linux-sh

Pull SuperH fixes from Paul Mundt.

* tag 'sh-for-linus' of git://github.com/pmundt/linux-sh: (24 commits)
  sh: explicitly include sh_dma.h in setup-sh7722.c
  sh: ecovec: care CN5 VBUS if USB host mode
  sh: sh7724: fixup renesas_usbhs clock settings
  sh: intc: initial irqdomain support.
  sh: pfc: Fix up init ordering mess.
  serial: sh-sci: fix compilation breakage, when DMA is enabled
  dmaengine: shdma: restore partial transfer calculation
  sh: modify the sh_dmae_slave_config for RSPI in setup-sh7757
  sh: Fix up recursive fault in oops with unset TTB.
  sh: pfc: Build fix for pinctrl_remove_gpio_range() changes.
  sh: select the fixed regulator driver on several boards
  sh: ecovec: switch MMC power control to regulators
  sh: add fixed voltage regulators to se7724
  sh: add fixed voltage regulators to sdk7786
  sh: add fixed voltage regulators to rsk
  sh: add fixed voltage regulators to migor
  sh: add fixed voltage regulators to kfr2r09
  sh: add fixed voltage regulators to ap325rxa
  sh: add fixed voltage regulators to sh7757lcr
  sh: add fixed voltage regulators to sh2007
  ...

12 years agoMerge tag 'md-3.6' of git://neil.brown.name/md
Linus Torvalds [Thu, 2 Aug 2012 18:34:40 +0000 (11:34 -0700)]
Merge tag 'md-3.6' of git://neil.brown.name/md

Pull additional md update from NeilBrown:
 "This contains a few patches that depend on plugging changes in the
  block layer so needed to wait for those.

  It also contains a Kconfig fix for the new RAID10 support in dm-raid."

* tag 'md-3.6' of git://neil.brown.name/md:
  md/dm-raid: DM_RAID should select MD_RAID10
  md/raid1: submit IO from originating thread instead of md thread.
  raid5: raid5d handle stripe in batch way
  raid5: make_request use batch stripe release

12 years agoMerge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/sage/ceph...
Linus Torvalds [Thu, 2 Aug 2012 17:57:31 +0000 (10:57 -0700)]
Merge branch 'for-linus' of git://git./linux/kernel/git/sage/ceph-client

Pull two ceph fixes from Sage Weil:
 "The first patch fixes up the old crufty open intent code to use the
  atomic_open stuff properly, and the second fixes a possible null deref
  and memory leak with the crypto keys."

* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/sage/ceph-client:
  libceph: fix crypto key null deref, memory leak
  ceph: simplify+fix atomic_open

12 years agoMerge tag 'ecryptfs-3.6-rc1-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git...
Linus Torvalds [Thu, 2 Aug 2012 17:56:34 +0000 (10:56 -0700)]
Merge tag 'ecryptfs-3.6-rc1-fixes' of git://git./linux/kernel/git/tyhicks/ecryptfs

Pull ecryptfs fixes from Tyler Hicks:
 - Fixes a bug when the lower filesystem mount options include 'acl',
   but the eCryptfs mount options do not
 - Cleanups in the messaging code
 - Better handling of empty files in the lower filesystem to improve
   usability.  Failed file creations are now cleaned up and empty lower
   files are converted into eCryptfs during open().
 - The write-through cache changes are being reverted due to bugs that
   are not easy to fix.  Stability outweighs the performance
   enhancements here.
 - Improvement to the mount code to catch unsupported ciphers specified
   in the mount options

* tag 'ecryptfs-3.6-rc1-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/tyhicks/ecryptfs:
  eCryptfs: check for eCryptfs cipher support at mount
  eCryptfs: Revert to a writethrough cache model
  eCryptfs: Initialize empty lower files when opening them
  eCryptfs: Unlink lower inode when ecryptfs_create() fails
  eCryptfs: Make all miscdev functions use daemon ptr in file private_data
  eCryptfs: Remove unused messaging declarations and function
  eCryptfs: Copy up POSIX ACL and read-only flags from lower mount

12 years agoMerge branch 'for-next' of git://git.samba.org/sfrench/cifs-2.6
Linus Torvalds [Thu, 2 Aug 2012 17:54:11 +0000 (10:54 -0700)]
Merge branch 'for-next' of git://git.samba.org/sfrench/cifs-2.6

Pull CIFS update from Steve French:
 "Adds SMB2 rmdir/mkdir capability to the SMB2/SMB2.1 support in cifs.

  I am holding up a few more days on merging the remainder of the
  SMB2/SMB2.1 enablement although it is nearing review completion, in
  order to address some review comments from Jeff Layton on a few of the
  subsequent SMB2 patches, and also to debug an unrelated cifs problem
  that Pavel discovered."

* 'for-next' of git://git.samba.org/sfrench/cifs-2.6:
  CIFS: Add SMB2 support for rmdir
  CIFS: Move rmdir code to ops struct
  CIFS: Add SMB2 support for mkdir operation
  CIFS: Separate protocol specific part from mkdir
  CIFS: Simplify cifs_mkdir call

12 years agomm: remove node_start_pfn checking in new WARN_ON for now
Linus Torvalds [Thu, 2 Aug 2012 17:37:03 +0000 (10:37 -0700)]
mm: remove node_start_pfn checking in new WARN_ON for now

Borislav Petkov reports that the new warning added in commit
88fdf75d1bb5 ("mm: warn if pg_data_t isn't initialized with zero")
triggers for him, and it is the node_start_pfn field that has already
been initialized once.

The call trace looks like this:

  x86_64_start_kernel ->
    x86_64_start_reservations ->
    start_kernel ->
    setup_arch ->
    paging_init ->
    zone_sizes_init ->
    free_area_init_nodes ->
    free_area_init_node

and (with the warning replaced by debug output), Borislav sees

  On node 0 totalpages: 4193848
    DMA zone: 64 pages used for memmap
    DMA zone: 6 pages reserved
    DMA zone: 3890 pages, LIFO batch:0
    DMA32 zone: 16320 pages used for memmap
    DMA32 zone: 798464 pages, LIFO batch:31
    Normal zone: 52736 pages used for memmap
    Normal zone: 3322368 pages, LIFO batch:31
  free_area_init_node: pgdat->node_start_pfn: 4423680      <----
  On node 1 totalpages: 4194304
    Normal zone: 65536 pages used for memmap
    Normal zone: 4128768 pages, LIFO batch:31
  free_area_init_node: pgdat->node_start_pfn: 8617984      <----
  On node 2 totalpages: 4194304
    Normal zone: 65536 pages used for memmap
    Normal zone: 4128768 pages, LIFO batch:31
  free_area_init_node: pgdat->node_start_pfn: 12812288     <----
  On node 3 totalpages: 4194304
    Normal zone: 65536 pages used for memmap
    Normal zone: 4128768 pages, LIFO batch:31

so remove the bogus warning for now to avoid annoying people.  Minchan
Kim is looking at it.

Reported-by: Borislav Petkov <bp@amd64.org>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
12 years agoARM: mmp: add missing irqs.h
Haojian Zhuang [Mon, 30 Jul 2012 14:20:34 +0000 (22:20 +0800)]
ARM: mmp: add missing irqs.h

arch/arm/mach-mmp/gplugd.c:195:13: error: ‘MMP_NR_IRQS’ undeclared here
(not in a function)
make[1]: *** [arch/arm/mach-mmp/gplugd.o] Error 1

Include <mach/irqs.h> to fix this issue.

Signed-off-by: Haojian Zhuang <haojian.zhuang@gmail.com>
Signed-off-by: Olof Johansson <olof@lixom.net>
12 years agoarm: mvebu: fix typo in .dtsi comment for Armada XP SoCs
Thomas Petazzoni [Thu, 2 Aug 2012 15:13:47 +0000 (17:13 +0200)]
arm: mvebu: fix typo in .dtsi comment for Armada XP SoCs

The comment was wrongly referring to Armada 370 while the file is
related to Armada XP.

Signed-off-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
Signed-off-by: Olof Johansson <olof@lixom.net>
12 years agoARM: PRIMA2: delete redundant codes to restore LATCHED when timer resumes
Barry Song [Mon, 30 Jul 2012 05:29:30 +0000 (13:29 +0800)]
ARM: PRIMA2: delete redundant codes to restore LATCHED when timer resumes

The only way to write LATCHED registers to write LATCH_BIT to LATCH register,
that will latch COUNTER into LATCHED.e.g.
writel_relaxed(SIRFSOC_TIMER_LATCH_BIT, sirfsoc_timer_base +
SIRFSOC_TIMER_LATCH);

Writing values to LATCHED registers directly is useless at all.

Signed-off-by: Barry Song <Baohua.Song@csr.com>
Signed-off-by: Olof Johansson <olof@lixom.net>
12 years agolibceph: fix crypto key null deref, memory leak
Sylvain Munaut [Thu, 2 Aug 2012 16:12:59 +0000 (09:12 -0700)]
libceph: fix crypto key null deref, memory leak

Avoid crashing if the crypto key payload was NULL, as when it was not correctly
allocated and initialized.  Also, avoid leaking it.

Signed-off-by: Sylvain Munaut <tnt@246tNt.com>
Signed-off-by: Sage Weil <sage@inktank.com>
Reviewed-by: Alex Elder <elder@inktank.com>
12 years agoceph: simplify+fix atomic_open
Sage Weil [Tue, 31 Jul 2012 18:27:36 +0000 (11:27 -0700)]
ceph: simplify+fix atomic_open

The initial ->atomic_open op was carried over from the old intent code,
which was incomplete and didn't really work.  Replace it with a fresh
method.  In particular:

 * always attempt to do an atomic open+lookup, both for the create case
   and for lookups of existing files.
 * fix symlink handling by returning 1 to the VFS so that we can follow
   the link to its destination. This fixes a longstanding ceph bug (#2392).

Signed-off-by: Sage Weil <sage@inktank.com>
12 years agosh: explicitly include sh_dma.h in setup-sh7722.c
Guennadi Liakhovetski [Wed, 1 Aug 2012 08:44:57 +0000 (10:44 +0200)]
sh: explicitly include sh_dma.h in setup-sh7722.c

setup-sh7722.c defines several objects, whose types are defined in
sh_dma.h, so, it has to be included explicitly.

Signed-off-by: Guennadi Liakhovetski <g.liakhovetski@gmx.de>
Signed-off-by: Paul Mundt <lethal@linux-sh.org>
12 years agoMerge branch 'upstream' of git://git.linux-mips.org/pub/scm/ralf/upstream-linus
Linus Torvalds [Wed, 1 Aug 2012 23:47:15 +0000 (16:47 -0700)]
Merge branch 'upstream' of git://git.linux-mips.org/ralf/upstream-linus

Pull MIPS updates from Ralf Baechle:
 "The lion share of this pull request are fixes for clk-related breakage
  caused by other changes during this merge window.  For some platforms
  the fix was as simple as selecting HAVE_CLK, for others like the
  Loongson 2 significant restructuring was required.

  The remainder are changes required to get the Lantiq code to work
  again."

* 'upstream' of git://git.linux-mips.org/pub/scm/ralf/upstream-linus:
  MIPS: Loongson 2: Sort out clock managment.
  MIPS: Loongson 1: more clk support and add select HAVE_CLK
  MIPS: txx9: Fix redefinition of clk_* by adding select HAVE_CLK
  MIPS: BCM63xx: Fix redefinition of clk_* by adding select HAVE_CLK
  MIPS: AR7: Fix redefinition of clk_* by adding select HAVE_CLK
  MIPS: Lantiq: Platform specific CLK fixup
  MIPS: Lantiq: Add device_tree_init function
  MIPS: Lantiq: Fix interface clock and PCI control register offset

12 years agoMerge branch 'for-linus-3.6-rc1' of git://git.kernel.org/pub/scm/linux/kernel/git...
Linus Torvalds [Wed, 1 Aug 2012 23:45:02 +0000 (16:45 -0700)]
Merge branch 'for-linus-3.6-rc1' of git://git./linux/kernel/git/rw/uml

Pull UML fixes from Richard Weinberger:
 "This patch set contains mostly fixes and cleanups.  The UML tty driver
  uses now tty_port and is no longer broken like hell  :-)"

* 'for-linus-3.6-rc1' of git://git.kernel.org/pub/scm/linux/kernel/git/rw/uml:
  um: Add arch/x86/um to MAINTAINERS
  um: pass siginfo to guest process
  um: fix ubd_file_size for read-only files
  um: pull interrupt_end() into userspace()
  um: split syscall_trace(), pass pt_regs to it
  um: switch UPT_SET_RETURN_VALUE and regs_return_value to pt_regs
  um: set BLK_CGROUP=y in defconfig
  um: remove count_lock
  um: fully use tty_port
  um: Remove dead code
  um: remove line_ioctl()
  TTY: um/line, use tty from tty_port
  TTY: um/line, add tty_port

12 years agoMerge branch 'dmaengine' of git://git.linaro.org/people/rmk/linux-arm
Linus Torvalds [Wed, 1 Aug 2012 23:41:07 +0000 (16:41 -0700)]
Merge branch 'dmaengine' of git://git.linaro.org/people/rmk/linux-arm

Pull ARM DMA engine updates from Russell King:
 "This looks scary at first glance, but what it is is:
   - a rework of the sa11x0 DMA engine driver merged during the previous
     cycle, to extract a common set of helper functions for DMA engine
     implementations.
   - conversion of amba-pl08x.c to use these helper functions.
   - addition of OMAP DMA engine driver (using these helper functions),
     and conversion of some of the OMAP DMA users to use DMA engine.

  Nothing in the helper functions is ARM specific, so I hope that other
  implementations can consolidate some of their code by making use of
  these helpers.

  This has been sitting in linux-next most of the merge cycle, and has
  been tested by several OMAP folk.  I've tested it on sa11x0 platforms,
  and given it my best shot on my broken platforms which have the
  amba-pl08x controller.

  The last point is the addition to feature-removal-schedule.txt, which
  will have a merge conflict.  Between myself and TI, we're planning to
  remove the old TI DMA implementation next year."

Fix up trivial add/add conflicts in Documentation/feature-removal-schedule.txt
and drivers/dma/{Kconfig,Makefile}

* 'dmaengine' of git://git.linaro.org/people/rmk/linux-arm: (53 commits)
  ARM: 7481/1: OMAP2+: omap2plus_defconfig: enable OMAP DMA engine
  ARM: 7464/1: mmc: omap_hsmmc: ensure probe returns error if DMA channel request fails
  Add feature removal of old OMAP private DMA implementation
  mtd: omap2: remove private DMA API implementation
  mtd: omap2: add DMA engine support
  spi: omap2-mcspi: remove private DMA API implementation
  spi: omap2-mcspi: add DMA engine support
  ARM: omap: remove mmc platform data dma_mask and initialization
  mmc: omap: remove private DMA API implementation
  mmc: omap: add DMA engine support
  mmc: omap_hsmmc: remove private DMA API implementation
  mmc: omap_hsmmc: add DMA engine support
  dmaengine: omap: add support for cyclic DMA
  dmaengine: omap: add support for setting fi
  dmaengine: omap: add support for returning residue in tx_state method
  dmaengine: add OMAP DMA engine driver
  dmaengine: sa11x0-dma: add cyclic DMA support
  dmaengine: sa11x0-dma: fix DMA residue support
  dmaengine: PL08x: ensure all descriptors are freed when channel is released
  dmaengine: PL08x: get rid of write only pool_ctr and free_txd locking
  ...

12 years agoMerge branch 'audit' of git://git.linaro.org/people/rmk/linux-arm
Linus Torvalds [Wed, 1 Aug 2012 23:35:37 +0000 (16:35 -0700)]
Merge branch 'audit' of git://git.linaro.org/people/rmk/linux-arm

Pull ARM audit/signal updates from Russell King:
 "ARM audit/signal handling updates from Al and Will.  This improves on
  the work Viro did last merge window, and sorts out some of the issues
  found with that work."

* 'audit' of git://git.linaro.org/people/rmk/linux-arm:
  ARM: 7475/1: sys_trace: allow all syscall arguments to be updated via ptrace
  ARM: 7474/1: get rid of TIF_SYSCALL_RESTARTSYS
  ARM: 7473/1: deal with handlerless restarts without leaving the kernel
  ARM: 7472/1: pull all work_pending logics into C function
  ARM: 7471/1: Revert "7442/1: Revert "remove unused restart trampoline""
  ARM: 7470/1: Revert "7443/1: Revert "new way of handling ERESTART_RESTARTBLOCK""

12 years agoMerge branch 'fixes' of git://git.linaro.org/people/rmk/linux-arm
Linus Torvalds [Wed, 1 Aug 2012 23:30:45 +0000 (16:30 -0700)]
Merge branch 'fixes' of git://git.linaro.org/people/rmk/linux-arm

Pull ARM fixes from Russell King:
 "This fixes various issues found during July"

* 'fixes' of git://git.linaro.org/people/rmk/linux-arm:
  ARM: 7479/1: mm: avoid NULL dereference when flushing gate_vma with VIVT caches
  ARM: Fix undefined instruction exception handling
  ARM: 7480/1: only call smp_send_stop() on SMP
  ARM: 7478/1: errata: extend workaround for erratum #720789
  ARM: 7477/1: vfp: Always save VFP state in vfp_pm_suspend on UP
  ARM: 7476/1: vfp: only clear vfp state for current cpu in vfp_pm_suspend
  ARM: 7468/1: ftrace: Trace function entry before updating index
  ARM: 7467/1: mutex: use generic xchg-based implementation for ARMv6+
  ARM: 7466/1: disable interrupt before spinning endlessly
  ARM: 7465/1: Handle >4GB memory sizes in device tree and mem=size@start option

12 years agoum: Add arch/x86/um to MAINTAINERS
Richard Weinberger [Wed, 1 Aug 2012 23:00:47 +0000 (01:00 +0200)]
um: Add arch/x86/um to MAINTAINERS

Signed-off-by: Richard Weinberger <richard@nod.at>
12 years agoum: pass siginfo to guest process
Martin Pärtel [Wed, 1 Aug 2012 22:49:17 +0000 (00:49 +0200)]
um: pass siginfo to guest process

UML guest processes now get correct siginfo_t for SIGTRAP, SIGFPE,
SIGILL and SIGBUS. Specifically, si_addr and si_code are now correct
where previously they were si_addr = NULL and si_code = 128.

Signed-off-by: Martin Pärtel <martin.partel@gmail.com>
Signed-off-by: Richard Weinberger <richard@nod.at>
12 years agoum: fix ubd_file_size for read-only files
Martin Pärtel [Wed, 1 Aug 2012 22:44:22 +0000 (00:44 +0200)]
um: fix ubd_file_size for read-only files

Made ubd_file_size not request write access. Fixes use of read-only images.

Signed-off-by: Martin Pärtel <martin.partel@gmail.com>
Signed-off-by: Richard Weinberger <richard@nod.at>