Access fields in m_quotainfo directly instead of hiding them behind the
XFS_QI_* macros.  Add local variables for the quotainfo pointer in places
where we have lots of them.
Signed-off-by: NChristoph Hellwig <hch@lst.de>
Reviewed-by: NDave Chinner <david@fromorbit.com>

We need to wait for all pending direct I/O requests before taking care of
metadata in fsync and write_inode.
Signed-off-by: NChristoph Hellwig <hch@lst.de>
Reviewed-by: NDave Chinner <david@fromorbit.com>

fs/xfs/linux-2.6/xfs_trace.c: xfs_attr_sf.h is included more than once.
Signed-off-by: NAndrea Gelmini <andrea.gelmini@gelma.net>
Signed-off-by: NAlex Elder <aelder@sgi.com>

Odds and ends in "xfs_log_recover.c".  This patch just contains some
minor things that didn't seem to warrant their own individual
patches:
- In xlog_bread_noalign(), drop an assertion that a pointer is
  non-null (the crash will tell us it was a bad pointer).
- Add a more descriptive header comment for xlog_find_verify_cycle().
- Make a few additions to the comments in xlog_find_head().  Also
  rearrange some expressions in a few spots to produce the same
  result, but in a way that seems more clear what's being computed.

(Updated in response to Dave's review comments.  Note I did not
split this patch like I said I would.)
Signed-off-by: NAlex Elder <aelder@sgi.com>
Reviewed-by: NChristoph Hellwig <hch@lst.de>

In xlog_find_cycle_start() use a local variable for some repeated
operations rather than constantly accessing the memory location
whose address is passed in.

(This version drops an assertion that a pointer is non-null.)
Signed-off-by: NAlex Elder <aelder@sgi.com>
Reviewed-by: NChristoph Hellwig <hch@lst.de>

Rename a label used in xlog_find_head() that I thought was poorly
chosen.  Also combine two adjacent labels xlog_find_tail() into a
single label, and give it a more generic name.

(Now using Dave's suggested "validate_head" name for first label.)
Signed-off-by: NAlex Elder <aelder@sgi.com>
Reviewed-by: NChristoph Hellwig <hch@lst.de>

xfs_bwrite is used with the intention of synchronously writing out
buffers, but currently it does not actually clear the async flag if
that's left from previous writes but instead implements async
behaviour if it finds it.  Remove the code handling asynchronous
writes as we've got rid of those entirely outside of the log and
delwri buffers, and make sure that we clear the async and read flags
before writing the buffer.
Signed-off-by: NChristoph Hellwig <hch@lst.de>
Signed-off-by: NAlex Elder <aelder@sgi.com>

All modifications to the superblock are done transactional through
xfs_trans_log_buf, so there is no reason to initiate periodic
asynchronous writeback.  This only removes the superblock from the
delwri list and will lead to sub-optimal I/O scheduling.

Cut down xfs_sync_fsdata now that it's only used for synchronous
superblock writes and move the log coverage checks into the two
callers.
Signed-off-by: NChristoph Hellwig <hch@lst.de>
Signed-off-by: NAlex Elder <aelder@sgi.com>

The transaction ID that is written to the log for a transaction is
currently set by taking the lower 32 bits of the memory address of
the ticket structure.  This is not guaranteed to be unique as
tickets comes from a slab and slots can be reallocated immediately
after being freed. As a result, there is no guarantee of uniqueness
in the ticket ID value.

Fix this by assigning a random number to the ticket ID field so that
it is extremely unlikely that duplicates will occur and remove the
possibility of transactions being mixed up during recovery due to
duplicate IDs.
Signed-off-by: NDave Chinner <dchinner@redhat.com>
Reviewed-by: NChristoph Hellwig <hch@lst.de>

There are a number of places where a log sector size of 1 uses
special case code.  The round_up() and round_down() macros
produce the correct result even when the log sector size is 1, and
this eliminates the need for treating this as a special case.
Signed-off-by: NAlex Elder <aelder@sgi.com>
Reviewed-by: NChristoph Hellwig <hch@lst.de>

Define a function that encapsulates checking the validity of a log
block count.

(Updated from previous version--no longer includes error reporting in the
encapsulated validation function.)
Signed-off-by: NAlex Elder <aelder@sgi.com>
Reviewed-by: NDave Chinner <dchinner@redhat.com>
Signed-off-by: NDave Chinner <dchinner@redhat.com>

XLOG_SECTOR_ROUNDUP_BBCOUNT() and XLOG_SECTOR_ROUNDDOWN_BLKNO()
are now fairly simple macro translations.  Just get rid of them in
favor of the round_up() and round_down() macro calls they represent.

Also, in spots in xlog_get_bp() and xlog_write_log_records(),
round_up() was being called with value 1, which just evaluates
to the macro's second argument; so just use that instead.
In the latter case, make use of that value, as long as it's
already been computed.
Signed-off-by: NAlex Elder <aelder@sgi.com>
Reviewed-by: NChristoph Hellwig <hch@lst.de>
Signed-off-by: NDave Chinner <dchinner@redhat.com>

XLOG_SECTOR_ROUNDUP_BBCOUNT() is defined in "fs/xfs/xfs_log_recover.c"
in an overly-complicated way.  It is basically roundup(), but that
is not at all clear from its definition.  (Actually, there is
another macro round_up() that applies for power-of-two-based masks
which I'll be using here.)

The operands in XLOG_SECTOR_ROUNDUP_BBCOUNT() are basically the
block number (bbs) and the log sector basic block mask
(log->l_sectbb_mask).  I'll call them B and M for this discussion.

The macro computes is value this way:
	M && (B & M) ? (B + M + 1) & ~M : B

Put another way, we can break it into 3 cases:
	1)  ! M          -> B			# 0 mask, no effect
	2)  ! (B & M)    -> B			# sector aligned
	3)  M && (B & M) -> (B + M + 1) & ~M	# round up otherwise

The round_up() macro is cleverly defined using a value, v, and a
power-of-2, p, and the result is the nearest multiple of p greater
than or equal to v.  Its value is computed something like this:
	((v - 1) | (p - 1)) + 1
Let's consider using this in the context of the 3 cases above.

When p = 2^0 = 1, the result boils down to ((v - 1) | 0) + 1, so it
just translates any value v to itself.  That handles case (1) above.

When p = 2^n, n > 0, we know that (p - 1) will be a mask with all n
bits 0..n-1 set.  The condition in this case occurs when none of
those mask bits is set in the value v provided.  If that is the
case, subtracting 1 from v will have 1's in all those lower bits (at
least).  Therefore, OR-ing the mask with that decremented value has
no effect, so adding the 1 back again will just translate the v to
itself.  This handles case (2).

Otherwise, the value v is greater than some multiple of p, and
decrementing it will produce a result greater than or equal to that
multiple.  OR-ing in the mask will produce a value 1 less than the
next multiple of p, so finally adding 1 back will result in the
desired rounded-up value.  This handles case (3).

Hopefully this is convincing.

While I was at it, I converted XLOG_SECTOR_ROUNDDOWN_BLKNO() to use
the round_down() macro.
Signed-off-by: NAlex Elder <aelder@sgi.com>
Reviewed-by: NChristoph Hellwig <hch@lst.de>
Signed-off-by: NDave Chinner <dchinner@redhat.com>

This fixes a bug in two places that I found by inspection.  In
xlog_find_verify_cycle() and xlog_write_log_records(), the code
attempts to allocate a buffer to hold as many blocks as possible.
It gives up if the number of blocks to be allocated gets too small.
Right now it uses log->l_sectbb_log as that lower bound, but I'm
sure it's supposed to be the actual log sector size instead.  That
is, the lower bound should be (1 << log->l_sectbb_log).

Also define a simple macro xlog_sectbb(log) to represent the number
of basic blocks in a sector for the given log.

(No change from original submission; I have implemented Christoph's
suggestion about storing l_sectsize rather than l_sectbb_log in
a new, separate patch in this series.)
Signed-off-by: NAlex Elder <aelder@sgi.com>
Reviewed-by: NDave Chinner <dchinner@redhat.com>
Signed-off-by: NDave Chinner <dchinner@redhat.com>

Change the tag and file name arguments to xfs_error_report() and
xfs_corruption_error() to use a const qualifier.
Signed-off-by: NAlex Elder <aelder@sgi.com>
Reviewed-by: NChristoph Hellwig <hch@lst.de>
Signed-off-by: NDave Chinner <dchinner@redhat.com>

The xfs_dqmarker structure does not need to exist anymore. Move the
remaining flags field out of it and remove the structure altogether.
Signed-off-by: NChristoph Hellwig <hch@lst.de>
Reviewed-by: NDave Chinner <david@fromorbit.com>

Convert the dquot free list on the filesystem to use listhead
infrastructure rather than the roll-your-own in the quota code.
Signed-off-by: NDave Chinner <dchinner@redhat.com>
Reviewed-by: NChristoph Hellwig <hch@lst.de>

Convert the dquot hash list on the filesystem to use listhead
infrastructure rather than the roll-your-own in the quota code.
Signed-off-by: NDave Chinner <dchinner@redhat.com>
Reviewed-by: NChristoph Hellwig <hch@lst.de>

The dquot shaker and the free-list reclaim code use exactly the same
algorithm but the code is duplicated and slightly different in each
case. Make the shaker code use the single dquot reclaim code to
remove the code duplication.
Signed-off-by: NDave Chinner <dchinner@redhat.com>
Reviewed-by: NChristoph Hellwig <hch@lst.de>

Convert the dquot list on the filesytesm to use listhead
infrastructure rather than the roll-your-own in the quota code.
Signed-off-by: NDave Chinner <dchinner@redhat.com>
Reviewed-by: NChristoph Hellwig <hch@lst.de>

Currently there is no tracing in log recovery, so it is difficult to
determine what is going on when something goes wrong.

Add tracing for log item recovery to provide visibility into the log
recovery process. The tracing added shows regions being extracted
from the log transactions and added to the transaction hash forming
recovery items, followed by the reordering, cancelling and finally
recovery of the items.
Signed-off-by: NDave Chinner <dchinner@redhat.com>
Reviewed-by: NChristoph Hellwig <hch@lst.de>

Replace the awkward xlog_write_adv_cnt with an inline helper that makes
it more obvious that it's modifying it's paramters, and replace the use
of an integer type for "ptr" with a real void pointer.  Also move
xlog_write_adv_cnt to xfs_log_priv.h as it will be used outside of
xfs_log.c in the delayed logging series.
Signed-off-by: NChristoph Hellwig <hch@lst.de>
Reviewed-by: NDave Chinner <dchinner@redhat.com>

The current log IO vector structure is a flat array and not
extensible. To make it possible to keep separate log IO vectors for
individual log items, we need a method of chaining log IO vectors
together.

Introduce a new log vector type that can be used to wrap the
existing log IO vectors on use that internally to the log. This
means that the existing external interface (xfs_log_write) does not
change and hence no changes to the transaction commit code are
required.
Signed-off-by: NChristoph Hellwig <hch@lst.de>
Signed-off-by: NDave Chinner <dchinner@redhat.com>
Reviewed-by: NDave Chinner <dchinner@redhat.com>

Reindent xlog_write to normal one tab indents and move all variable
declarations into the closest enclosing block.

Split from a bigger patch by Dave Chinner.
Signed-off-by: NChristoph Hellwig <hch@lst.de>
Reviewed-by: NDave Chinner <dchinner@redhat.com>

xlog_write is a mess that takes a lot of effort to understand. It is
a mass of nested loops with 4 space indents to get it to fit in 80 columns
and lots of funky variables that aren't obvious what they mean or do.

Break it down into understandable chunks.
Signed-off-by: NChristoph Hellwig <hch@lst.de>
Signed-off-by: NDave Chinner <dchinner@redhat.com>
Reviewed-by: NDave Chinner <dchinner@redhat.com>

When allocation a ticket for a transaction, the ticket is initialised with the
worst case log space usage based on the number of bytes the transaction may
consume. Part of this calculation is the number of log headers required for the
iclog space used up by the transaction.

This calculation makes an undocumented assumption that if the transaction uses
the log header space reservation on an iclog, then it consumes either the
entire iclog or it completes. That is - the transaction that is first in an
iclog is the transaction that the log header reservation is accounted to. If
the transaction is larger than the iclog, then it will use the entire iclog
itself. Document this assumption.

Further, the current calculation uses the rule that we can fit iclog_size bytes
of transaction data into an iclog. This is in correct - the amount of space
available in an iclog for transaction data is the size of the iclog minus the
space used for log record headers. This means that the calculation is out by
512 bytes per 32k of log space the transaction can consume. This is rarely an
issue because maximally sized transactions are extremely uncommon, and for 4k
block size filesystems maximal transaction reservations are about 400kb. Hence
the error in this case is less than the size of an iclog, so that makes it even
harder to hit.

However, anyone using larger directory blocks (16k directory blocks push the
maximum transaction size to approx. 900k on a 4k block size filesystem) or
larger block size (e.g. 64k blocks push transactions to the 3-4MB size) could
see the error grow to more than an iclog and at this point the transaction is
guaranteed to get a reservation underrun and shutdown the filesystem.

Fix this by adjusting the calculation to calculate the correct number of iclogs
required and account for them all up front.
Signed-off-by: NDave Chinner <dchinner@redhat.com>
Reviewed-by: NChristoph Hellwig <hch@lst.de>

Now that the code has been factored, clean up all the remaining
style cruft, simplify the code and re-order functions so that it
doesn't need forward declarations.

Also move the remaining functions that require forward declarations
(xfs_trans_uncommit, xfs_trans_free) so that all the forward
declarations can be removed from the file.
Signed-off-by: NDave Chinner <dchinner@redhat.com>
Reviewed-by: NChristoph Hellwig <hch@lst.de>

The function header to xfs-trans_committed has long had this
comment:

 * THIS SHOULD BE REWRITTEN TO USE xfs_trans_next_item()

To prepare for different methods of committing items, convert the
code to use xfs_trans_next_item() and factor the code into smaller,
more digestible chunks.
Signed-off-by: NDave Chinner <dchinner@redhat.com>
Reviewed-by: NChristoph Hellwig <hch@lst.de>

> +shut_us_down:
> +	shutdown = XFS_FORCED_SHUTDOWN(mp) ? EIO : 0;
> +	if (!(tp->t_flags & XFS_TRANS_DIRTY) || shutdown) {
> +		xfs_trans_unreserve_and_mod_sb(tp);
> +		/*

This whole area in _xfs_trans_commit is still a complete mess.

So while touching this code, unravel this mess as well to make the
whole flow of the function simpler and clearer.
Signed-off-by: NChristoph Hellwig <hch@lst.de>
Reviewed-by: NDave Chinner <david@fromorbit.com>

Split the the part of xfs_trans_commit() that deals with writing the
transaction into the iclog into a separate function. This isolates the
physical commit process from the logical commit operation and makes
it easier to insert different transaction commit paths without affecting
the existing algorithm adversely.
Signed-off-by: NDave Chinner <dchinner@redhat.com>
Reviewed-by: NChristoph Hellwig <hch@lst.de>

xfs_bmap_add_attrfork() passes XFS_TRANS_PERM_LOG_RES to xfs_trans_commit()
to indicate that the commit should release the permanent log reservation
as part of the commit. This is wrong - the correct flag is
XFS_TRANS_RELEASE_LOG_RES - and it is only by the chance that both these
flags have the value of 0x4 that the code is doing the right thing.

Fix it by changing to use the correct flag.
Signed-off-by: NDave Chinner <dchinner@redhat.com>
Reviewed-by: NChristoph Hellwig <hch@lst.de>

The staleness of a object being unpinned can be directly derived
from the object itself - there is no need to extract it from the
object then pass it as a parameter into IOP_UNPIN().

This means we can kill the XFS_LID_BUF_STALE flag - it is set,
checked and cleared in the same places XFS_BLI_STALE flag in the
xfs_buf_log_item so it is now redundant and hence safe to remove.
Signed-off-by: NDave Chinner <dchinner@redhat.com>
Reviewed-by: NChristoph Hellwig <hch@lst.de>

We don't record pin counts in inode events right now, and this makes
it difficult to track down problems related to pinning inodes. Add
the pin count to the inode trace class and add trace events for
pinning and unpinning inodes.
Signed-off-by: NDave Chinner <dchinner@redhat.com>
Reviewed-by: NChristoph Hellwig <hch@lst.de>

Each log item type does manual initialisation of the log item.
Delayed logging introduces new fields that need initialisation, so
factor all the open coded initialisation into a common function
first.
Signed-off-by: NDave Chinner <dchinner@redhat.com>
Reviewed-by: NChristoph Hellwig <hch@lst.de>

This allows to see in `ps` and similar tools which kthreads are
allotted to which block device/filesystem, similar to what jbd2
does. As the process name is a fixed 16-char array, no extra
space is needed in tasks.

  PID TTY      STAT   TIME COMMAND
    2 ?        S      0:00 [kthreadd]
  197 ?        S      0:00  \_ [jbd2/sda2-8]
  198 ?        S      0:00  \_ [ext4-dio-unwrit]
  204 ?        S      0:00  \_ [flush-8:0]
 2647 ?        S      0:00  \_ [xfs_mru_cache]
 2648 ?        S      0:00  \_ [xfslogd/0]
 2649 ?        S      0:00  \_ [xfsdatad/0]
 2650 ?        S      0:00  \_ [xfsconvertd/0]
 2651 ?        S      0:00  \_ [xfsbufd/ram0]
 2652 ?        S      0:00  \_ [xfsaild/ram0]
 2653 ?        S      0:00  \_ [xfssyncd/ram0]
Signed-off-by: NJan Engelhardt <jengelh@medozas.de>
Reviewed-by: NDave Chinner <david@fromorbit.com>

The am_hreq.opcount field in the xfs_attrmulti_by_handle() interface
is not bounded correctly. The opcount is used to determine the size
of the buffer required. The size is bounded, but can overflow and so
the size checks may not be sufficient to catch invalid opcounts.
Fix it by catching opcount values that would cause overflows before
calculating the size.
Signed-off-by: NZhitong Wang <zhitong.wangzt@alibaba-inc.com>
Reviewed-by: NDave Chinner <david@fromorbit.com>

On low memory boxes or those with highmem, kernel can OOM before the
background reclaims inodes via xfssyncd. Add a shrinker to run inode
reclaim so that it inode reclaim is expedited when memory is low.

This is more complex than it needs to be because the VM folk don't
want a context added to the shrinker infrastructure. Hence we need
to add a global list of XFS mount structures so the shrinker can
traverse them.
Signed-off-by: NDave Chinner <dchinner@redhat.com>
Reviewed-by: NChristoph Hellwig <hch@lst.de>

A new xfsqa test (226) with a prototype xfs_fsr change to try to
handle dynamic fork offsets better triggers an assertion failure
where the inode data fork is in btree format, yet there is room in
the inode for it to be in extent format. The two inodes look like:

before: ino 0x101 (target), num_extents 11, Max in-fork extents 6, broot size 40, fork offset 96
before: ino 0x115 (temp),  num_extents 5, Max in-fork extents 3, broot size 40, fork offset 56
after: ino 0x101 (target), num_extents 5, Max in-fork extents 6, broot size 40, fork offset 96
after: ino 0x115 (temp), num_extents 11, Max in-fork extents 3, broot size 40, fork offset 56

Basically the target inode ends up with 5 extents in btree format,
but it had space for 6 extents in extent format, so ends up
incorrect. Notably here the broot size is the same, and that is
where the kernel code is going wrong - the btree root will fit, so
it lets the swap go ahead.

The check should not allow the swap to take place if the number of
extents while in btree format is less than the number of extents
that can fit in the inode in extent format. Adding that check will
prevent this swap and corruption from occurring.
Signed-off-by: NDave Chinner <dchinner@redhat.com>
Reviewed-by: NChristoph Hellwig <hch@lst.de>

Any inode reclaim flush that returns EAGAIN will result in the inode
reclaim being attempted again later. There is no need to issue a
warning into the logs about this situation.
Signed-off-by: NDave Chinner <dchinner@redhat.com>
Reviewed-by: NAlex Elder <aelder@sgi.com>
Signed-off-by: NAlex Elder <aelder@sgi.com>

Updates to the VFS layer removed an extra ->sync_fs call into the
filesystem during the sync process (from the quota code).
Unfortunately the sync code was unknowingly relying on this call to
make sure metadata buffers were flushed via a xfs_buftarg_flush()
call to move the tail of the log forward in memory before the final
transactions of the sync process were issued.

As a result, the old code would write a very recent log tail value
to the log by the end of the sync process, and so a subsequent crash
would leave nothing for log recovery to do. Hence in qa test 182,
log recovery only replayed a small handle for inode fsync
transactions in this case.

However, with the removal of the extra ->sync_fs call, the log tail
was now not moved forward with the inode fsync transactions near the
end of the sync procese the first (and only) buftarg flush occurred
after these transactions went to disk. The result is that log
recovery now sees a large number of transactions for metadata that
is already on disk.

This usually isn't a problem, but when the transactions include
inode chunk allocation, the inode create transactions and all
subsequent changes are replayed as we cannt rely on what is on disk
is valid. As a result, if the inode was written and contains
unlogged changes, the unlogged changes are lost, thereby violating
sync semantics.

The fix is to always issue a transaction after the buftarg flush
occurs is the log iѕ not idle or covered. This results in a dummy
transaction being written that contains the up-to-date log tail
value, which will be very recent. Indeed, it will be at least as
recent as the old code would have left on disk, so log recovery
will behave exactly as it used to in this situation.
Signed-off-by: NDave Chinner <dchinner@redhat.com>
Reviewed-by: NChristoph Hellwig <hch@lst.de>
Signed-off-by: NAlex Elder <aelder@sgi.com>