This patch implements two closely related changes: First it embeds
a bio the ioend structure so that we don't have to allocate one
separately. Second it uses the block layer bio chaining mechanism
to chain additional bios off this first one if needed instead of
manually accounting for multiple bio completions in the ioend
structure. Together this removes a memory allocation per ioend and
greatly simplifies the ioend setup and I/O completion path.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Brian Foster <bfoster@redhat.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
Completion of an ioend requires us to walk the bufferhead list to
end writback on all the bufferheads. This, in turn, is needed so
that we can end writeback on all the pages we just did IO on.
To remove our dependency on bufferheads in writeback, we need to
turn this around the other way - we need to walk the pages we've
just completed IO on, and then walk the buffers attached to the
pages and complete their IO. In doing this, we remove the
requirement for the ioend to track bufferheads directly.
To enable IO completion to walk all the pages we've submitted IO on,
we need to keep the bios that we used for IO around until the ioend
has been completed. We can do this simply by chaining the bios to
the ioend at completion time, and then walking their pages directly
just before destroying the ioend.
Signed-off-by: Dave Chinner <dchinner@redhat.com>
[hch: changed the xfs_finish_page_writeback calling convention]
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Brian Foster <bfoster@redhat.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
Currently adding a buffer to the ioend and then building a bio from
the buffer list are two separate operations. We don't build the bios
and submit them until the ioend is submitted, and this places a
fixed dependency on bufferhead chaining in the ioend.
The first step to removing the bufferhead chaining in the ioend is
on the IO submission side. We can build the bio directly as we add
the buffers to the ioend chain, thereby removing the need for a
latter "buffer-to-bio" submission loop. This allows us to submit
bios on large ioends as soon as we cannot add more data to the bio.
These bios then get captured by the active plug, and hence will be
dispatched as soon as either the plug overflows or we schedule away
from the writeback context. This will reduce submission latency for
large IOs, but will also allow more timely request queue based
writeback blocking when the device becomes congested.
Signed-off-by: Dave Chinner <dchinner@redhat.com>
[hch: various small updates]
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Brian Foster <bfoster@redhat.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
Change summary:
o error propagation for direct IO failures fixes for both XFS and ext4
o new quota interfaces and XFS implementation for iterating all the quota IDs
in the filesystem
o locking fixes for real-time device extent allocation
o reduction of duplicate information in the xfs and vfs inode, saving roughly
100 bytes of memory per cached inode.
o buffer flag cleanup
o rework of the writepage code to use the generic write clustering mechanisms
o several fixes for inode flag based DAX enablement
o rework of remount option parsing
o compile time verification of on-disk format structure sizes
o delayed allocation reservation overrun fixes
o lots of little error handling fixes
o small memory leak fixes
o enable xfsaild freezing again
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1
iQIcBAABAgAGBQJW71DQAAoJEK3oKUf0dfodyiwP/0Tou9f1huzLC0kd7kmEoKKC
BWQmtJGEdo0iSpJNZhg/EJmjvRtbBiOB9CRcEyG8d71kqZ+MKW7t/4JjNvNG34aE
vHjhwMBVVqkw/q6azi2LiEDsVcOe5bXxUrXNZi18/09OAl4pHm+X8VERLnnC5y+i
QIHAOdB5R+36cXcceJm1HR6jTZedbNdQkT/ndhm5S60FGhvVI29cs9NwYwoi5aif
O55r6krSWBj6U/X6MsLvr+lNb6+1Sd1hyE8dGTE7lOUX/crFIysaDPEuQmWvDjsO
M1ulVfzKoBJHcyvpbdHwdBEyiBjzvETcrgndMRoWOjZiOLqNtWYsgIEiC+Nlidwd
+T4XhkJJJg5UUQ4r6Hs85SQn/THanzR5KoN5nbTsFtFkCKw1DRkUSNuh2mXP2xVG
JcNDCjDvvHG76EfQ1otlYf7ru79Ck+hjVs+szaEVPpOzAwz8yOtD+L7I8f73gQ6a
ayP8W2oZQpYvQRv+smgvt+HwQA4fNJk9ZseY3QD5+z5snJz7JEhZogqW+ngFYkNQ
dtA5Y7gpTkKfo3mKO0XmE5+3fcSXhGHGYQzmUgJFlgWTK7+E8fuDhn6D66wFcZSq
QhyRk9J7Xb7ZWuP5PlOkxb9DLd4hnuyie2bYw/0hVtOatjE/Em4gRJ3Oq3ZANwZx
OeMGj4Uyb3/MKAJwy3Gq
=ZoiX
-----END PGP SIGNATURE-----
Merge tag 'xfs-for-linus-4.6-rc1' of git://git.kernel.org/pub/scm/linux/kernel/git/dgc/linux-xfs
Pull xfs updates from Dave Chinner:
"There's quite a lot in this request, and there's some cross-over with
ext4, dax and quota code due to the nature of the changes being made.
As for the rest of the XFS changes, there are lots of little things
all over the place, which add up to a lot of changes in the end.
The major changes are that we've reduced the size of the struct
xfs_inode by ~100 bytes (gives an inode cache footprint reduction of
>10%), the writepage code now only does a single set of mapping tree
lockups so uses less CPU, delayed allocation reservations won't
overrun under random write loads anymore, and we added compile time
verification for on-disk structure sizes so we find out when a commit
or platform/compiler change breaks the on disk structure as early as
possible.
Change summary:
- error propagation for direct IO failures fixes for both XFS and
ext4
- new quota interfaces and XFS implementation for iterating all the
quota IDs in the filesystem
- locking fixes for real-time device extent allocation
- reduction of duplicate information in the xfs and vfs inode, saving
roughly 100 bytes of memory per cached inode.
- buffer flag cleanup
- rework of the writepage code to use the generic write clustering
mechanisms
- several fixes for inode flag based DAX enablement
- rework of remount option parsing
- compile time verification of on-disk format structure sizes
- delayed allocation reservation overrun fixes
- lots of little error handling fixes
- small memory leak fixes
- enable xfsaild freezing again"
* tag 'xfs-for-linus-4.6-rc1' of git://git.kernel.org/pub/scm/linux/kernel/git/dgc/linux-xfs: (66 commits)
xfs: always set rvalp in xfs_dir2_node_trim_free
xfs: ensure committed is initialized in xfs_trans_roll
xfs: borrow indirect blocks from freed extent when available
xfs: refactor delalloc indlen reservation split into helper
xfs: update freeblocks counter after extent deletion
xfs: debug mode forced buffered write failure
xfs: remove impossible condition
xfs: check sizes of XFS on-disk structures at compile time
xfs: ioends require logically contiguous file offsets
xfs: use named array initializers for log item dumping
xfs: fix computation of inode btree maxlevels
xfs: reinitialise per-AG structures if geometry changes during recovery
xfs: remove xfs_trans_get_block_res
xfs: fix up inode32/64 (re)mount handling
xfs: fix format specifier , should be %llx and not %llu
xfs: sanitize remount options
xfs: convert mount option parsing to tokens
xfs: fix two memory leaks in xfs_attr_list.c error paths
xfs: XFS_DIFLAG2_DAX limited by PAGE_SIZE
xfs: dynamically switch modes when XFS_DIFLAG2_DAX is set/cleared
...
dax_clear_blocks() needs a valid struct block_device and previously it
was using inode->i_sb->s_bdev in all cases. This is correct for normal
inodes on mounted ext2, ext4 and XFS filesystems, but is incorrect for
DAX raw block devices and for XFS real-time devices.
Instead, rename dax_clear_blocks() to dax_clear_sectors(), and change
its arguments to take a bdev and a sector instead of an inode and a
block. This better reflects what the function does, and it allows the
filesystem and raw block device code to pass in an appropriate struct
block_device.
Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Suggested-by: Dan Williams <dan.j.williams@intel.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Cc: Theodore Ts'o <tytso@mit.edu>
Cc: Al Viro <viro@ftp.linux.org.uk>
Cc: Dave Chinner <david@fromorbit.com>
Cc: Jens Axboe <axboe@fb.com>
Cc: Matthew Wilcox <matthew.r.wilcox@intel.com>
Cc: Ross Zwisler <ross.zwisler@linux.intel.com>
Cc: Theodore Ts'o <tytso@mit.edu>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Currently we can build a long ioend chain during ->writepages that
gets attached to the writepage context. IO submission only then
occurs when we finish all the writepage processing. This means we
can have many ioends allocated and pending, and this violates the
mempool guarantees that we need to give about forwards progress.
i.e. we really should only have one ioend being built at a time,
otherwise we may drain the mempool trying to allocate a new ioend
and that blocks submission, completion and freeing of ioends that
are already in progress.
To prevent this situation from happening, we need to submit ioends
for IO as soon as they are ready for dispatch rather than queuing
them for later submission. This means the ioends have bios built
immediately and they get queued on any plug that is current active.
Hence if we schedule away from writeback, the ioends that have been
built will make forwards progress due to the plug flushing on
context switch. This will also prevent context switches from
creating unnecessary IO submission latency.
We can't completely avoid having nested IO allocation - when we have
a block size smaller than a page size, we still need to hold the
ioend submission until after we have marked the current page dirty.
Hence we may need multiple ioends to be held while the current page
is completely mapped and made ready for IO dispatch. We cannot avoid
this problem - the current code already has this ioend chaining
within a page so we can mostly ignore that it occurs.
Signed-off-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Dave Chinner <david@fromorbit.com>
xfs_vm_writepages() calls generic_writepages to writeback a range of
a file, but then xfs_vm_writepage() clusters pages itself as it does
not have any context it can pass between->writepage calls from
__write_cache_pages().
Introduce a writeback context for xfs_vm_writepages() and call
__write_cache_pages directly with our own writepage callback so that
we can pass that context to each writepage invocation. This
encapsulates the current mapping, whether it is valid or not, the
current ioend and it's IO type and the ioend chain being built.
This requires us to move the ioend submission up to the level where
the writepage context is declared. This does mean we do not submit
IO until we packaged the entire writeback range, but with the block
plugging in the writepages call this is the way IO is submitted,
anyway.
It also means that we need to handle discontiguous page ranges. If
the pages sent down by write_cache_pages to the writepage callback
are discontiguous, we need to detect this and put each discontiguous
page range into individual ioends. This is needed to ensure that the
ioend accurately represents the range of the file that it covers so
that file size updates during IO completion set the size correctly.
Failure to take into account the discontiguous ranges results in
files being too small when writeback patterns are non-sequential.
Signed-off-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Dave Chinner <david@fromorbit.com>
For DAX, we are now doing block zeroing during allocation. This
means we no longer need a special DAX fault IO completion callback
to do unwritten extent conversion. Because mmap never extends the
file size (it SEGVs the process) we don't need a callback to update
the file size, either. Hence we can remove the completion callbacks
from the __dax_fault and __dax_mkwrite calls.
Signed-off-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Brian Foster <bfoster@redhat.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
Both direct IO and DAX pass an offset and count into get_blocks that
will overflow a s64 variable when an IO goes into the last supported
block in a file (i.e. at offset 2^63 - 1FSB bytes). This can be seen
from the tracing:
xfs_get_blocks_alloc: [...] offset 0x7ffffffffffff000 count 4096
xfs_gbmap_direct: [...] offset 0x7ffffffffffff000 count 4096
xfs_gbmap_direct_none:[...] offset 0x7ffffffffffff000 count 4096
0x7ffffffffffff000 + 4096 = 0x8000000000000000, and hence that
overflows the s64 offset and we fail to detect the need for a
filesize update and an ioend is not allocated.
This is *mostly* avoided for direct IO because such extending IOs
occur with full block allocation, and so the "IS_UNWRITTEN()" check
still evaluates as true and we get an ioend that way. However, doing
single sector extending IOs to this last block will expose the fact
that file size updates will not occur after the first allocating
direct IO as the overflow will then be exposed.
There is one further complexity: the DAX page fault path also
exposes the same issue in block allocation. However, page faults
cannot extend the file size, so in this case we want to allocate the
block but do not want to allocate an ioend to enable file size
update at IO completion. Hence we now need to distinguish between
the direct IO patch allocation and dax fault path allocation to
avoid leaking ioend structures.
Signed-off-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Brian Foster <bfoster@redhat.com>
Signed-off-by: Dave 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: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Brian Foster <bfoster@redhat.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
Back in the days when the direct I/O ->end_io callback could be called
from interrupt context for AIO we needed a structure to hand off to the
workqueue, and reused the ioend structure for this purpose. These days
->end_io is always called from user or workqueue context, which allows us
to avoid this memory allocation and simplify the code significantly.
[dchinner: removed now unused xfs_finish_ioend_sync() function after
Brian Foster did an initial review. ]
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
Add support to the core direct-io code to defer AIO completions to user
context using a workqueue. This replaces opencoded and less efficient
code in XFS and ext4 (we save a memory allocation for each direct IO)
and will be needed to properly support O_(D)SYNC for AIO.
The communication between the filesystem and the direct I/O code requires
a new buffer head flag, which is a bit ugly but not avoidable until the
direct I/O code stops abusing the buffer_head structure for communicating
with the filesystems.
Currently this creates a per-superblock unbound workqueue for these
completions, which is taken from an earlier patch by Jan Kara. I'm
not really convinced about this use and would prefer a "normal" global
workqueue with a high concurrency limit, but this needs further discussion.
JK: Fixed ext4 part, dynamic allocation of the workqueue.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Add a XFS_ prefix to IO_DIRECT,XFS_IO_DELALLOC, XFS_IO_UNWRITTEN and
XFS_IO_OVERWRITE. This to avoid namespace conflict with other modules.
Signed-off-by: Alain Renaud <arenaud@sgi.com>
Reviewed-by: Rich Johnston <rjohnston@sgi.com>
Signed-off-by: Ben Myers <bpm@sgi.com>
Do not use unlogged metadata updates and the VFS dirty bit for updating
the file size after writeback. In addition to causing various problems
with updates getting delayed for far too long this also drags in the
unscalable VFS dirty tracking, and is one of the few remaining unlogged
metadata updates.
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Mark Tinguely <tinguely@sgi.com>
Signed-off-by: Ben Myers <bpm@sgi.com>
The new concurrency managed workqueues are cheap enough that we can create
per-filesystem instead of global workqueues. This allows us to remove the
trylock or defer scheme on the ilock, which is not helpful once we have
outstanding log reservations until finishing a size update.
Also allow the default concurrency on this workqueues so that I/O completions
blocking on the ilock for one inode do not block process for another inode.
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Mark Tinguely <tinguely@sgi.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Ben Myers <bpm@sgi.com>
We now have an i_dio_count filed and surrounding infrastructure to wait
for direct I/O completion instead of i_icount, and we have never needed
to iocount waits for buffered I/O given that we only set the page uptodate
after finishing all required work. Thus remove i_iocount, and replace
the actually needed waits with calls to inode_dio_wait.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Alex Elder <aelder@sgi.com>
We really shouldn't complete AIO or DIO requests until we have finished
the unwritten extent conversion and size update. This means fsync never
has to pick up any ioends as all work has been completed when signalling
I/O completion.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Alex Elder <aelder@sgi.com>
Use the move from Linux 2.6 to Linux 3.x as an excuse to kill the
annoying subdirectories in the XFS source code. Besides the large
amount of file rename the only changes are to the Makefile, a few
files including headers with the subdirectory prefix, and the binary
sysctl compat code that includes a header under fs/xfs/ from
kernel/.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Alex Elder <aelder@sgi.com>
Christoph Hellwig