Release of the EFI either occurs based on the reference count or the
extent count. The extent count used is either the count tracked in
the EFI or EFD, depending on the particular situation. In either
case, the count is initialized to the final value and thus always
matches the current efi_next_extent value once the EFI is completely
constructed.  For example, the EFI extent count is increased as the
extents are logged in xfs_bmap_finish() and the full free list is
always completely processed. Therefore, the count is guaranteed to
be complete once the EFI transaction is committed. The EFD uses the
efd_nextents counter to release the EFI. This counter is initialized
to the count of the EFI when the EFD is created. Thus the EFD, as
currently used, has no concept of partial EFI release based on
extent count.

Given that the EFI extent count is always released in whole, use of
the extent count for reference counting is unnecessary. Remove this
level of the API and release the EFI based on the core reference
count. The efi_next_extent counter remains because it is still used
to track the slot to log the next extent to free.
Signed-off-by: NBrian Foster <bfoster@redhat.com>
Reviewed-by: NDave Chinner <dchinner@redhat.com>
Signed-off-by: NDave Chinner <david@fromorbit.com>

Currently XFS calls file_remove_privs() without holding i_mutex. This is
wrong because that function can end up messing with file permissions and
file capabilities stored in xattrs for which we need i_mutex held.

Fix the problem by grabbing iolock exclusively when we will need to
change anything in permissions / xattrs.
Reviewed-by: NDave Chinner <dchinner@redhat.com>
Signed-off-by: NJan Kara <jack@suse.cz>
Signed-off-by: NAl Viro <viro@zeniv.linux.org.uk>

file_remove_suid() is a misnomer since it removes also file capabilities
stored in xattrs and sets S_NOSEC flag. Also should_remove_suid() tells
something else than whether file_remove_suid() call is necessary which
leads to bugs.
Signed-off-by: NJan Kara <jack@suse.cz>
Signed-off-by: NAl Viro <viro@zeniv.linux.org.uk>

The xfs_attr3_root_inactive() call from xfs_attr_inactive() assumes that
attribute blocks exist to invalidate. It is possible to have an
attribute fork without extents, however. Consider the case where the
attribute fork is created towards the beginning of xfs_attr_set() but
some part of the subsequent attribute set fails.

If an inode in such a state hits xfs_attr_inactive(), it eventually
calls xfs_dabuf_map() and possibly xfs_bmapi_read(). The former emits a
filesystem corruption warning, returns an error that bubbles back up to
xfs_attr_inactive(), and leads to destruction of the in-core attribute
fork without an on-disk reset. If the inode happens to make it back
through xfs_inactive() in this state (e.g., via a concurrent bulkstat
that cycles the inode from the reclaim state and releases it), i_afp
might not exist when xfs_bmapi_read() is called and causes a NULL
dereference panic.

A '-p 2' fsstress run to ENOSPC on a relatively small fs (1GB)
reproduces these problems. The behavior is a regression caused by:

6dfe5a04 xfs: xfs_attr_inactive leaves inconsistent attr fork state behind

... which removed logic that avoided the attribute extent truncate when
no extents exist. Restore this logic to ensure the attribute fork is
destroyed and reset correctly if it exists without any allocated
extents.

cc: stable@vger.kernel.org # 3.12 to 4.0.x
Signed-off-by: NBrian Foster <bfoster@redhat.com>
Reviewed-by: NDave Chinner <dchinner@redhat.com>
Signed-off-by: NDave Chinner <david@fromorbit.com>

We no longer calculate the minimum freelist size from the on-disk
AGF, so we don't need the macros used for this. That means the
nested macros can be cleaned up, and turn this into an actual
function so the logic is clear and concise. This will make it much
easier to add support for the rmap btree when the time comes.

This also gets rid of the XFS_AG_MAXLEVELS macro used by these
freelist macros as it is simply a wrapper around a single variable.
Signed-off-by: NDave Chinner <dchinner@redhat.com>
Reviewed-by: NBrian Foster <bfoster@redhat.com>
Signed-off-by: NDave Chinner <david@fromorbit.com>

The error handling is currently an inconsistent mess as every error
condition handles return values and releasing buffers individually.
Clean this up by using gotos and a sane error label stack.
Signed-off-by: NDave Chinner <dchinner@redhat.com>
Reviewed-by: NBrian Foster <bfoster@redhat.com>
Signed-off-by: NDave Chinner <david@fromorbit.com>

The longest extent length checks in xfs_alloc_fix_freelist() are now
essentially identical. Factor them out into a helper function, so we
know they are checking exactly the same thing before and after we
lock the AGF.
Signed-off-by: NDave Chinner <dchinner@redhat.com>
Reviewed-by: NBrian Foster <bfoster@redhat.com>
Signed-off-by: NDave Chinner <david@fromorbit.com>

At the moment, xfs_alloc_fix_freelist() uses a mix of per-ag based
access and agf buffer  based access to freelist and space usage
information. However, once the AGF buffer is locked inside this
function, it is guaranteed that both the in-memory and on-disk
values are identical. xfs_alloc_fix_freelist() doesn't modify the
values in the structures directly, so it is a read-only user of the
infomration, and hence can use the per-ag structure exclusively for
determining what it should do.

This opens up an avenue for cleaning up a lot of duplicated logic
whose only difference is the structure it gets the data from, and in
doing so removes a lot of needless byte swapping overhead when
fixing up the free list.
Signed-off-by: NDave Chinner <dchinner@redhat.com>
Reviewed-by: NBrian Foster <bfoster@redhat.com>
Signed-off-by: NDave Chinner <david@fromorbit.com>

Just use char pointers directly instead of the confusing typedef to a
pointer type.
Signed-off-by: NChristoph Hellwig <hch@lst.de>
Reviewed-by: NBrian Foster <bfoster@redhat.com>
Signed-off-by: NDave Chinner <david@fromorbit.com>

Compared to char pointers this saves us a lot of casting effort.  Also
add another local variable to make the code easier to read.
Signed-off-by: NChristoph Hellwig <hch@lst.de>
Reviewed-by: NBrian Foster <bfoster@redhat.com>
Signed-off-by: NDave Chinner <david@fromorbit.com>

This avoids all kinds of unessecary casts in an envrionment like Linux where
we can assume that pointer arithmetics are support on void pointers.
Signed-off-by: NChristoph Hellwig <hch@lst.de>
Reviewed-by: NBrian Foster <bfoster@redhat.com>
Signed-off-by: NDave Chinner <david@fromorbit.com>

We can simply use a void pointer to pass a long return addresses in the
debugging helpers.
Signed-off-by: NChristoph Hellwig <hch@lst.de>
Reviewed-by: NBrian Foster <bfoster@redhat.com>
Signed-off-by: NDave Chinner <david@fromorbit.com>

Replace uses of __psint_t with the proper uintptr_t and ptrdiff_t types,
and remove the defintions of __psint_t and __psunsigned_t.
Signed-off-by: NChristoph Hellwig <hch@lst.de>
Reviewed-by: NBrian Foster <bfoster@redhat.com>
Signed-off-by: NDave Chinner <david@fromorbit.com>

If we create a CRC filesystem, mount it, and create a symlink with
a path long enough that it can't live in the inode, we get a very
strange result upon remount:

# ls -l mnt
total 4
lrwxrwxrwx. 1 root root 929 Jun 15 16:58 link -> XSLM

XSLM is the V5 symlink block header magic (which happens to be
followed by a NUL, so the string looks terminated).

xfs_readlink_bmap() advanced cur_chunk by the size of the header
for CRC filesystems, but never actually used that pointer; it
kept reading from bp->b_addr, which is the start of the block,
rather than the start of the symlink data after the header.

Looks like this problem goes back to v3.10.

Fixing this gets us reading the proper link target, again.

Cc: stable@vger.kernel.org
Signed-off-by: NEric Sandeen <sandeen@redhat.com>
Reviewed-by: NDave Chinner <dchinner@redhat.com>
Signed-off-by: NDave Chinner <david@fromorbit.com>

Instead of the confusing flags argument pass a boolean flag to indicate if
we want to release or regrant a log reservation.

Also ensure that xfs_log_done always drop the reference on the log ticket,
to both simplify the code and make the logic in xfs_trans_roll easier
to understand.
Signed-off-by: NChristoph Hellwig <hch@lst.de>
Reviewed-by: NDave Chinner <dchinner@redhat.com>
Signed-off-by: NDave Chinner <david@fromorbit.com>

The flags argument to xfs_trans_commit is not useful for most callers, as
a commit of a transaction without a permanent log reservation must pass
0 here, and all callers for a transaction with a permanent log reservation
except for xfs_trans_roll must pass XFS_TRANS_RELEASE_LOG_RES.  So remove
the flags argument from the public xfs_trans_commit interfaces, and
introduce low-level __xfs_trans_commit variant just for xfs_trans_roll
that regrants a log reservation instead of releasing it.
Signed-off-by: NChristoph Hellwig <hch@lst.de>
Reviewed-by: NDave Chinner <dchinner@redhat.com>
Signed-off-by: NDave Chinner <david@fromorbit.com>

xfs_trans_cancel takes two flags arguments: XFS_TRANS_RELEASE_LOG_RES and
XFS_TRANS_ABORT.  Both of them are a direct product of the transaction
state, and can be deducted:

 - any dirty transaction needs XFS_TRANS_ABORT to be properly canceled,
   and XFS_TRANS_ABORT is a noop for a transaction that is not dirty.
 - any transaction with a permanent log reservation needs
   XFS_TRANS_RELEASE_LOG_RES to be properly canceled, and passing
   XFS_TRANS_RELEASE_LOG_RES for a transaction without a permanent
   log reservation is invalid.

So just remove the flags argument and do the right thing.
Signed-off-by: NChristoph Hellwig <hch@lst.de>
Reviewed-by: NDave Chinner <dchinner@redhat.com>
Signed-off-by: NDave Chinner <david@fromorbit.com>

The flags value always was 0 or XFS_TRANS_ABORT.  Switch to a bool
parameter to allow further cleanups.
Signed-off-by: NChristoph Hellwig <hch@lst.de>
Reviewed-by: NDave Chinner <dchinner@redhat.com>
Signed-off-by: NDave Chinner <david@fromorbit.com>

We have three remaining callers of xfs_trans_dup:

 - xfs_itruncate_extents which open codes xfs_trans_roll
 - xfs_bmap_finish doesn't have an xfs_inode argument and thus leaves
   attaching them to it's callers, but otherwise is identical to
   xfs_trans_roll
 - xfs_dir_ialloc looks at the log reservations in the old xfs_trans
   structure instead of the log reservation parameters, but otherwise
   is identical to xfs_trans_roll.

By allowing a NULL xfs_inode argument to xfs_trans_roll we can switch
these three remaining users over to xfs_trans_roll and mark xfs_trans_dup
static.
Signed-off-by: NChristoph Hellwig <hch@lst.de>
Reviewed-by: NDave Chinner <dchinner@redhat.com>
Signed-off-by: NDave Chinner <david@fromorbit.com>

The inode allocator enables random sparse inode chunk allocations in
DEBUG mode to facilitate testing. Sparse inode allocations are not
always possible, however, depending on the fs geometry. For example,
there is no possibility for a sparse inode allocation on filesystems
where the block size is large enough to fit one or more inode chunks
within a single block.

Fix up the DEBUG mode sparse inode allocation logic to trigger random
sparse allocations only when the geometry of the fs allows it.
Signed-off-by: NBrian Foster <bfoster@redhat.com>
Reviewed-by: NDave Chinner <dchinner@redhat.com>
Signed-off-by: NDave Chinner <david@fromorbit.com>

The kbuild test robot reports the following compilation failure with a
32-bit kernel configuration:

	fs/built-in.o: In function `xfs_ifree_cluster':
	>> xfs_inode.c:(.text+0x17ac84): undefined reference to `__umoddi3'

This is due to the use of the modulus operator on a 64-bit variable in
the ASSERT() added as part of the following commit:

	xfs: skip unallocated regions of inode chunks in xfs_ifree_cluster()

This ASSERT() simply checks that the offset of the inode in a sparse
cluster is appropriately aligned. Since the maximum inode record offset
is 63 (for a 64 inode record) and the calculated offset here should be
something less than that, just use a 32-bit variable to store the offset
and call the do_mod() helper.
Reported-by: Nkbuild test robot <fengguang.wu@intel.com>
Signed-off-by: NBrian Foster <bfoster@redhat.com>
Reviewed-by: NDave Chinner <dchinner@redhat.com>
Signed-off-by: NDave Chinner <david@fromorbit.com>

Add initial DAX support to XFS. To do this we need a new mount
option to turn DAX on filesystem, and we need to propagate this into
the inode flags whenever an inode is instantiated so that the
per-inode checks throughout the code Do The Right Thing.
Signed-off-by: NDave Chinner <dchinner@redhat.com>
Reviewed-by: NBrian Foster <bfoster@redhat.com>
Signed-off-by: NDave Chinner <david@fromorbit.com>

DAX does not do buffered IO (can't buffer direct access!) and hence
all read/write IO is vectored through the direct IO path.  Hence we
need to add the DAX IO path callouts to the direct IO
infrastructure.
Signed-off-by: NDave Chinner <dchinner@redhat.com>
Reviewed-by: NBrian Foster <bfoster@redhat.com>
Signed-off-by: NDave Chinner <david@fromorbit.com>

When we truncate a DAX file, we need to call through the DAX page
truncation path rather than through block_truncate_page() so that
mappings and block zeroing are all handled correctly. Otherwise,
truncate does not need to change.
Signed-off-by: NDave Chinner <dchinner@redhat.com>
Reviewed-by: NBrian Foster <bfoster@redhat.com>
Signed-off-by: NDave Chinner <david@fromorbit.com>

Add initial support for DAX block zeroing operations to XFS. DAX
cannot use buffered IO through the page cache for zeroing, nor do we
need to issue IO for uncached block zeroing. In both cases, we can
simply call out to the dax block zeroing function.
Signed-off-by: NDave Chinner <dchinner@redhat.com>
Reviewed-by: NBrian Foster <bfoster@redhat.com>
Signed-off-by: NDave Chinner <david@fromorbit.com>

Add the initial support for DAX file operations to XFS. This
includes the necessary block allocation and mmap page fault hooks
for DAX to function.

Note that there are changes to the splice interfaces to ensure that
for DAX splice avoids direct page cache manipulations and instead
takes the DAX IO paths for read/write operations.
Signed-off-by: NDave Chinner <dchinner@redhat.com>
Reviewed-by: NBrian Foster <bfoster@redhat.com>
Signed-off-by: NDave Chinner <david@fromorbit.com>

Lock ordering for the new mmap lock needs to be:

mmap_sem
  sb_start_pagefault
    i_mmap_lock
      page lock
        <fault processsing>

Right now xfs_vm_page_mkwrite gets this the wrong way around,
While technically it cannot deadlock due to the current freeze
ordering, it's still a landmine that might explode if we change
anything in future. Hence we need to nest the locks correctly.
Signed-off-by: NDave Chinner <dchinner@redhat.com>
Reviewed-by: NJan Kara <jack@suse.cz>
Reviewed-by: NBrian Foster <bfoster@redhat.com>
Signed-off-by: NDave Chinner <david@fromorbit.com>

With the planned cgroup writeback support, backing-dev related
declarations will be more widely used across block and cgroup;
unfortunately, including backing-dev.h from include/linux/blkdev.h
makes cyclic include dependency quite likely.

This patch separates out backing-dev-defs.h which only has the
essential definitions and updates blkdev.h to include it.  c files
which need access to more backing-dev details now include
backing-dev.h directly.  This takes backing-dev.h off the common
include dependency chain making it a lot easier to use it across block
and cgroup.

v2: fs/fat build failure fixed.
Signed-off-by: NTejun Heo <tj@kernel.org>
Reviewed-by: NJan Kara <jack@suse.cz>
Cc: Jens Axboe <axboe@kernel.dk>
Signed-off-by: NJens Axboe <axboe@fb.com>

When modifying PG_Dirty on cached file pages, update the new
MEM_CGROUP_STAT_DIRTY counter.  This is done in the same places where
global NR_FILE_DIRTY is managed.  The new memcg stat is visible in the
per memcg memory.stat cgroupfs file.  The most recent past attempt at
this was http://thread.gmane.org/gmane.linux.kernel.cgroups/8632

The new accounting supports future efforts to add per cgroup dirty
page throttling and writeback.  It also helps an administrator break
down a container's memory usage and provides evidence to understand
memcg oom kills (the new dirty count is included in memcg oom kill
messages).

The ability to move page accounting between memcg
(memory.move_charge_at_immigrate) makes this accounting more
complicated than the global counter.  The existing
mem_cgroup_{begin,end}_page_stat() lock is used to serialize move
accounting with stat updates.
Typical update operation:
	memcg = mem_cgroup_begin_page_stat(page)
	if (TestSetPageDirty()) {
		[...]
		mem_cgroup_update_page_stat(memcg)
	}
	mem_cgroup_end_page_stat(memcg)

Summary of mem_cgroup_end_page_stat() overhead:
- Without CONFIG_MEMCG it's a no-op
- With CONFIG_MEMCG and no inter memcg task movement, it's just
  rcu_read_lock()
- With CONFIG_MEMCG and inter memcg  task movement, it's
  rcu_read_lock() + spin_lock_irqsave()

A memcg parameter is added to several routines because their callers
now grab mem_cgroup_begin_page_stat() which returns the memcg later
needed by for mem_cgroup_update_page_stat().

Because mem_cgroup_begin_page_stat() may disable interrupts, some
adjustments are needed:
- move __mark_inode_dirty() from __set_page_dirty() to its caller.
  __mark_inode_dirty() locking does not want interrupts disabled.
- use spin_lock_irqsave(tree_lock) rather than spin_lock_irq() in
  __delete_from_page_cache(), replace_page_cache_page(),
  invalidate_complete_page2(), and __remove_mapping().

   text    data     bss      dec    hex filename
8925147 1774832 1785856 12485835 be84cb vmlinux-!CONFIG_MEMCG-before
8925339 1774832 1785856 12486027 be858b vmlinux-!CONFIG_MEMCG-after
                            +192 text bytes
8965977 1784992 1785856 12536825 bf4bf9 vmlinux-CONFIG_MEMCG-before
8966750 1784992 1785856 12537598 bf4efe vmlinux-CONFIG_MEMCG-after
                            +773 text bytes

Performance tests run on v4.0-rc1-36-g4f671fe2.  Lower is better for
all metrics, they're all wall clock or cycle counts.  The read and write
fault benchmarks just measure fault time, they do not include I/O time.

* CONFIG_MEMCG not set:
                            baseline                              patched
  kbuild                 1m25.030000(+-0.088% 3 samples)       1m25.426667(+-0.120% 3 samples)
  dd write 100 MiB          0.859211561 +-15.10%                  0.874162885 +-15.03%
  dd write 200 MiB          1.670653105 +-17.87%                  1.669384764 +-11.99%
  dd write 1000 MiB         8.434691190 +-14.15%                  8.474733215 +-14.77%
  read fault cycles       254.0(+-0.000% 10 samples)            253.0(+-0.000% 10 samples)
  write fault cycles     2021.2(+-3.070% 10 samples)           1984.5(+-1.036% 10 samples)

* CONFIG_MEMCG=y root_memcg:
                            baseline                              patched
  kbuild                 1m25.716667(+-0.105% 3 samples)       1m25.686667(+-0.153% 3 samples)
  dd write 100 MiB          0.855650830 +-14.90%                  0.887557919 +-14.90%
  dd write 200 MiB          1.688322953 +-12.72%                  1.667682724 +-13.33%
  dd write 1000 MiB         8.418601605 +-14.30%                  8.673532299 +-15.00%
  read fault cycles       266.0(+-0.000% 10 samples)            266.0(+-0.000% 10 samples)
  write fault cycles     2051.7(+-1.349% 10 samples)           2049.6(+-1.686% 10 samples)

* CONFIG_MEMCG=y non-root_memcg:
                            baseline                              patched
  kbuild                 1m26.120000(+-0.273% 3 samples)       1m25.763333(+-0.127% 3 samples)
  dd write 100 MiB          0.861723964 +-15.25%                  0.818129350 +-14.82%
  dd write 200 MiB          1.669887569 +-13.30%                  1.698645885 +-13.27%
  dd write 1000 MiB         8.383191730 +-14.65%                  8.351742280 +-14.52%
  read fault cycles       265.7(+-0.172% 10 samples)            267.0(+-0.000% 10 samples)
  write fault cycles     2070.6(+-1.512% 10 samples)           2084.4(+-2.148% 10 samples)

As expected anon page faults are not affected by this patch.

tj: Updated to apply on top of the recent cancel_dirty_page() changes.
Signed-off-by: NSha Zhengju <handai.szj@gmail.com>
Signed-off-by: NGreg Thelen <gthelen@google.com>
Signed-off-by: NTejun Heo <tj@kernel.org>
Signed-off-by: NJens Axboe <axboe@fb.com>

Fixed two missing spaces.
Signed-off-by: NNan Jia <jiananmail@gmail.com>
Reviewed-by: NDave Chinner <dchinner@redhat.com>
Signed-off-by: NDave Chinner <david@fromorbit.com>

The commit:

a9273ca5 xfs: convert attr to use unsigned names

added these (unsigned char *) casts, but then the _SIZE macros
return "7" - size of a pointer minus one - not the length of
the string.  This is harmless in the kernel, because the _SIZE
macros are not used, but as we sync up with userspace, this will
matter.

I don't think the cast is necessary; i.e. assigning the string
literal to an unsigned char *, or passing it to a function
expecting an unsigned char *, should be ok, right?
Signed-off-by: NEric Sandeen <sandeen@redhat.com>
Reviewed-by: NBrian Foster <bfoster@redhat.com>
Signed-off-by: NDave Chinner <david@fromorbit.com>

Al Viro reports that generic/231 fails frequently on XFS and bisected
the problem to the following commit:

	5d11fb4b xfs: rework zero range to prevent invalid i_size updates

... which is just the first commit that happens to cause fsx to
reproduce the problem. fsx reproduces via zero range calls. The
aforementioned commit overhauls zero range to use hole punch and
fallocate. As it turns out, the problem is reproducible on demand using
basic hole punch as follows:

$ mkfs.xfs -f -m crc=1,finobt=1 <dev>
$ mount <dev> /mnt -o uquota
$ xfs_io -f -c "falloc 0 50m" /mnt/file
$ for i in $(seq 1 20); do xfs_io -c "fpunch ${i}m 32k" /mnt/file; done
$ rm -f /mnt/file
$ repquota -us /mnt
...
User            used    soft    hard  grace    used  soft  hard  grace
----------------------------------------------------------------------
root      --     32K      0K      0K              3     0     0

A file is allocated with a single 50m extent. The extent count increases
via hole punches until the bmap converts to btree format. The file is
removed but quota reports 32k of space usage for the user. This
reservation is effectively leaked for the lifetime of the mount.

The reason this occurs is because the quota block reservation tracking
is confused when a transaction happens to free and allocate blocks at
the same time. Consider the following sequence of events:

- tp is allocated from xfs_free_file_space() and reserves several blocks
  for btree management. Blocks are reserved against the dquot and marked
  as such in the transaction (qtrx->qt_blk_res).
- 8 blocks are accounted free when the 32k range is punched out.
  xfs_trans_mod_dquot() is called with XFS_TRANS_DQ_BCOUNT and sets
  ->qt_bcount_delta to -8.
- Subsequently, a block is allocated against the same transaction by
  xfs_bmap_extents_to_btree() for btree conversion. A call to
  xfs_trans_mod_dquot() increases qt_blk_res_used to 1 and qt_bcount_delta
  to -7.
- The transaction is dup'd and committed by xfs_bmap_finish().
  xfs_trans_dup_dqinfo() sets the first transaction up such that it has a
  matching qt_blk_res and qt_blk_res_used of 1. The remaining unused
  reservation is transferred to the duplicate tp.

When the transactions are committed, the dquots are fixed up in
xfs_trans_apply_dquot_deltas() according to one of two methods:

1.) If the transaction holds a block reservation (->qt_blk_res != 0),
_only_ the unused portion reservation is unaccounted from the dquot.
Note that the tp duplication behavior of xfs_bmap_finish() makes it such
that qt_blk_res is typically 0 for tp's with unused reservation.
2.) Otherwise, the dquot is fixed up based on the block delta
(->qt_bcount_delta) created by the transaction.

Therefore, if a transaction has a negative qt_bcount_delta and positive
qt_blk_res_used, the former set of blocks that have been removed from
the file are never factored out of the in-core dquot reservation.
Instead, *_apply_dquot_deltas() sees 1 block used out of a 1 block
reservation and believes there is nothing to fix up. The on-disk
d_bcount is updated independently from qt_bcount_delta, and thus is
correct (and allows the quota usage to correct on remount).

To deal with this situation, we effectively want the "used reservation"
part of the transaction to be consistent with any freed blocks with
respect to quota tracking. For example, if 8 blocks are freed, the
subsequent single block allocation does not need to consume the initial
reservation made by the tp. Instead, it simply borrows one from the
previously freed. One possible implementation of such borrowing is to
avoid the blks_res_used increment when bcount_delta is negative. This
alone is flawed logic in that it only handles the case where blocks are
freed before allocated, however.

Rather than add more complexity to manage synchronization between
bcount_delta and blks_res_used, kill the latter entirely. blk_res_used
is only updated in one place and always in sync with delta_bcount.
Therefore, the net block reservation consumption of the transaction is
always available from bcount_delta. Calculate the reservation
consumption on the fly where necessary based on whether the tp has a
reservation and results in a positive net block delta on the inode.
Reported-by: NAl Viro <viro@ZenIV.linux.org.uk>
Signed-off-by: NBrian Foster <bfoster@redhat.com>
Reviewed-by: NDave Chinner <dchinner@redhat.com>
Signed-off-by: NDave Chinner <david@fromorbit.com>

The fsync() requirements for crash consistency on XFS are to flush file
data and force any in-core inode updates to the log. We currently check
whether the inode is pinned to identify whether the log needs to be
forced, since a non-zero pin count generally represents an inode that
has transactions awaiting a flush to the on-disk log.

This is not sufficient in all cases, however. Reports of xfstests test
generic/311 failures on ppc64/s390x hosts have identified failures to
fsync outstanding inode modifications due to the inode not being pinned
at the time of the fsync. This occurs because certain bmap updates can
complete by logging bmapbt buffers but without ever dirtying (and thus
pinning) the core inode. The following is a specific incarnation of this
problem:

$ mount $dev /mnt -o noatime,nobarrier
$ for i in $(seq 0 2 31); do \
        xfs_io -f -c "falloc $((i * 32768)) 32k" -c fsync /mnt/file; \
	done
$ xfs_io -c "pwrite -S 0 80k 16k" -c fsync -c "pwrite 76k 4k" -c fsync /mnt/file; \
	hexdump /mnt/file; \
	./xfstests-dev/src/godown /mnt
...
0000000 0000 0000 0000 0000 0000 0000 0000 0000
*
0013000 cdcd cdcd cdcd cdcd cdcd cdcd cdcd cdcd
*
0014000 0000 0000 0000 0000 0000 0000 0000 0000
*
00f8000
$ umount /mnt; mount ...
$ hexdump /mnt/file
0000000 0000 0000 0000 0000 0000 0000 0000 0000
*
00f8000

In short, the unwritten extent conversion for the last write is lost
despite the fact that an fsync executed before the filesystem was
shutdown. Note that this is impossible to reproduce on v5 supers due to
unconditional time callbacks for di_changecount and highly difficult to
reproduce on CONFIG_HZ=1000 kernels due to those same callbacks
frequently updating cmtime prior to the bmap update. CONFIG_HZ=100
reduces timer granularity enough to increase the odds that time updates
are skipped and allows this to reproduce within a handful of attempts.

To deal with this problem, unconditionally log the core in the unwritten
extent conversion path. Fix up logflags after the extent conversion to
keep the extent update code consistent with the other extent update
helpers. This fixup is not necessary for the other (hole, delay) extent
helpers because they execute in the block allocation codepath, which
already logs the inode for other reasons (e.g., for di_nblocks).
Signed-off-by: NBrian Foster <bfoster@redhat.com>
Reviewed-by: NDave Chinner <dchinner@redhat.com>
Signed-off-by: NDave Chinner <david@fromorbit.com>

Enable mounting of filesystems with sparse inode support enabled. Add
the incompat. feature bit to the *_ALL mask.
Signed-off-by: NBrian Foster <bfoster@redhat.com>
Reviewed-by: NDave Chinner <dchinner@redhat.com>
Signed-off-by: NDave Chinner <david@fromorbit.com>

xfs_ifree_cluster() is called to mark all in-memory inodes and inode
buffers as stale. This occurs after we've removed the inobt records and
dropped any references of inobt data. xfs_ifree_cluster() uses the
starting inode number to walk the namespace of inodes expected for a
single chunk a cluster buffer at a time. The cluster buffer disk
addresses are calculated by decoding the sequential inode numbers
expected from the chunk.

The problem with this approach is that if the inode chunk being removed
is a sparse chunk, not all of the buffer addresses that are calculated
as part of this sequence may be inode clusters. Attempting to acquire
the buffer based on expected inode characterstics (i.e., cluster length)
can lead to errors and is generally incorrect.

We already use a couple variables to carry requisite state from
xfs_difree() to xfs_ifree_cluster(). Rather than add a third, define a
new internal structure to carry the existing parameters through these
functions. Add an alloc field that represents the physical allocation
bitmap of inodes in the chunk being removed. Modify xfs_ifree_cluster()
to check each inode against the bitmap and skip the clusters that were
never allocated as real inodes on disk.
Signed-off-by: NBrian Foster <bfoster@redhat.com>
Reviewed-by: NDave Chinner <dchinner@redhat.com>
Signed-off-by: NDave Chinner <david@fromorbit.com>

An inode chunk is currently added to the transaction free list based on
a simple fsb conversion and hardcoded chunk length. The nature of sparse
chunks is such that the physical chunk of inodes on disk may consist of
one or more discontiguous parts. Blocks that reside in the holes of the
inode chunk are not inodes and could be allocated to any other use or
not allocated at all.

Refactor the existing xfs_bmap_add_free() call into the
xfs_difree_inode_chunk() helper. The new helper uses the existing
calculation if a chunk is not sparse. Otherwise, use the inobt record
holemask to free the contiguous regions of the chunk.
Signed-off-by: NBrian Foster <bfoster@redhat.com>
Reviewed-by: NDave Chinner <dchinner@redhat.com>
Signed-off-by: NDave Chinner <david@fromorbit.com>

Inode allocation from an existing record with free inodes traditionally
selects the first inode available according to the ir_free mask. With
sparse inode chunks, the ir_free mask could refer to an unallocated
region. We must mask the unallocated regions out of ir_free before using
it to select a free inode in the chunk.

Update the xfs_inobt_first_free_inode() helper to find the first free
inode available of the allocated regions of the inode chunk.
Signed-off-by: NBrian Foster <bfoster@redhat.com>
Reviewed-by: NDave Chinner <dchinner@redhat.com>
Signed-off-by: NDave Chinner <david@fromorbit.com>

Sparse inode allocations generally only occur when full inode chunk
allocation fails. This requires some level of filesystem space usage and
fragmentation.

For filesystems formatted with sparse inode chunks enabled, do random
sparse inode chunk allocs when compiled in DEBUG mode to increase test
coverage.
Signed-off-by: NBrian Foster <bfoster@redhat.com>
Reviewed-by: NDave Chinner <dchinner@redhat.com>
Signed-off-by: NDave Chinner <david@fromorbit.com>

xfs_ialloc_ag_alloc() makes several attempts to allocate a full inode
chunk. If all else fails, reduce the allocation to the sparse length and
alignment and attempt to allocate a sparse inode chunk.

If sparse chunk allocation succeeds, check whether an inobt record
already exists that can track the chunk. If so, inherit and update the
existing record. Otherwise, insert a new record for the sparse chunk.

Create helpers to align sparse chunk inode records and insert or update
existing records in the inode btrees. The xfs_inobt_insert_sprec()
helper implements the merge or update semantics required for sparse
inode records with respect to both the inobt and finobt. To update the
inobt, either insert a new record or merge with an existing record. To
update the finobt, use the updated inobt record to either insert or
replace an existing record.
Signed-off-by: NBrian Foster <bfoster@redhat.com>
Reviewed-by: NDave Chinner <dchinner@redhat.com>
Signed-off-by: NDave Chinner <david@fromorbit.com>

The inobt record holemask field is a condensed data type designed to fit
into the existing on-disk record and is zero based (allocated regions
are set to 0, sparse regions are set to 1) to provide backwards
compatibility. This makes the type somewhat complex for use in higher
level inode manipulations such as individual inode allocation, etc.

Rather than foist the complexity of dealing with this field to every bit
of logic that requires inode granular information, create a helper to
convert the holemask to an inode allocation bitmap. The inode allocation
bitmap is inode granularity similar to the inobt record free mask and
indicates which inodes of the chunk are physically allocated on disk,
irrespective of whether the inode is considered allocated or free by the
filesystem.
Signed-off-by: NBrian Foster <bfoster@redhat.com>
Reviewed-by: NDave Chinner <dchinner@redhat.com>
Signed-off-by: NDave Chinner <david@fromorbit.com>