Skip to content

K
Kernel

openeuler / Kernel
1 年多 前同步成功

通知 8
Star 0
Fork 0
K
Kernel
切换分支/标签
  1. 22 5月, 2018 3 次提交
    • D
      xfs, dax: introduce xfs_break_dax_layouts() · d6dc57e2
      由 Dan Williams 提交于 
      xfs_break_dax_layouts(), similar to xfs_break_leased_layouts(), scans
for busy / pinned dax pages and waits for those pages to go idle before
any potential extent unmap operation.

dax_layout_busy_page() handles synchronizing against new page-busy
events (get_user_pages). It invalidates all mappings to trigger the
get_user_pages slow path which will eventually block on the xfs inode
lock held in XFS_MMAPLOCK_EXCL mode. If dax_layout_busy_page() finds a
busy page it returns it for xfs to wait for the page-idle event that
will fire when the page reference count reaches 1 (recall ZONE_DEVICE
pages are idle at count 1, see generic_dax_pagefree()).

While waiting, the XFS_MMAPLOCK_EXCL lock is dropped in order to not
deadlock the process that might be trying to elevate the page count of
more pages before arranging for any of them to go idle. I.e. the typical
case of submitting I/O is that iov_iter_get_pages() elevates the
reference count of all pages in the I/O before starting I/O on the first
page. The process of elevating the reference count of all pages involved
in an I/O may cause faults that need to take XFS_MMAPLOCK_EXCL.

Although XFS_MMAPLOCK_EXCL is dropped while waiting, XFS_IOLOCK_EXCL is
held while sleeping. We need this to prevent starvation of the truncate
path as continuous submission of direct-I/O could starve the truncate
path indefinitely if the lock is dropped.

Cc: Dave Chinner <david@fromorbit.com>
Cc: Ross Zwisler <ross.zwisler@linux.intel.com>
Reported-by: NJan Kara <jack@suse.cz>
Reviewed-by: NJan Kara <jack@suse.cz>
Reviewed-by: NChristoph Hellwig <hch@lst.de>
Acked-by: NDarrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: NDan Williams <dan.j.williams@intel.com>
      d6dc57e2
    • D
      xfs: prepare xfs_break_layouts() for another layout type · 69eb5fa1
      由 Dan Williams 提交于 
      When xfs is operating as the back-end of a pNFS block server, it
prevents collisions between local and remote operations by requiring a
lease to be held for remotely accessed blocks. Local filesystem
operations break those leases before writing or mutating the extent map
of the file.

A similar mechanism is needed to prevent operations on pinned dax
mappings, like device-DMA, from colliding with extent unmap operations.

BREAK_WRITE and BREAK_UNMAP are introduced as two distinct levels of
layout breaking.

Layouts are broken in the BREAK_WRITE case to ensure that layout-holders
do not collide with local writes. Additionally, layouts are broken in
the BREAK_UNMAP case to make sure the layout-holder has a consistent
view of the file's extent map. While BREAK_WRITE breaks can be satisfied
be recalling FL_LAYOUT leases, BREAK_UNMAP breaks additionally require
waiting for busy dax-pages to go idle while holding XFS_MMAPLOCK_EXCL.

After this refactoring xfs_break_layouts() becomes the entry point for
coordinating both types of breaks. Finally, xfs_break_leased_layouts()
becomes just the BREAK_WRITE handler.

Note that the unlock tracking is needed in a follow on change. That will
coordinate retrying either break handler until both successfully test
for a lease break while maintaining the lock state.

Cc: Ross Zwisler <ross.zwisler@linux.intel.com>
Cc: "Darrick J. Wong" <darrick.wong@oracle.com>
Reported-by: NDave Chinner <david@fromorbit.com>
Reported-by: NChristoph Hellwig <hch@lst.de>
Reviewed-by: NChristoph Hellwig <hch@lst.de>
Signed-off-by: NDan Williams <dan.j.williams@intel.com>
      69eb5fa1
    • D
      xfs: prepare xfs_break_layouts() to be called with XFS_MMAPLOCK_EXCL · c63a8eae
      由 Dan Williams 提交于 
      In preparation for adding coordination between extent unmap operations
and busy dax-pages, update xfs_break_layouts() to permit it to be called
with the mmap lock held. This lock scheme will be required for
coordinating the break of 'dax layouts' (non-idle dax (ZONE_DEVICE)
pages mapped into the file's address space). Breaking dax layouts will
be added to xfs_break_layouts() in a future patch, for now this preps
the unmap call sites to take and hold XFS_MMAPLOCK_EXCL over the call to
xfs_break_layouts().

Cc: "Darrick J. Wong" <darrick.wong@oracle.com>
Cc: Ross Zwisler <ross.zwisler@linux.intel.com>
Cc: Dave Chinner <david@fromorbit.com>
Suggested-by: NChristoph Hellwig <hch@lst.de>
Reviewed-by: NChristoph Hellwig <hch@lst.de>
Reviewed-by: N"Darrick J. Wong" <darrick.wong@oracle.com>
Signed-off-by: NDan Williams <dan.j.williams@intel.com>
      c63a8eae
  3. 16 3月, 2018 1 次提交
  5. 15 3月, 2018 1 次提交
  7. 08 1月, 2018 1 次提交
  9. 03 11月, 2017 3 次提交
  11. 27 10月, 2017 1 次提交
  13. 24 10月, 2017 1 次提交
  15. 12 10月, 2017 1 次提交
  17. 27 9月, 2017 1 次提交
  19. 26 9月, 2017 2 次提交
  21. 07 9月, 2017 1 次提交
    • R
      dax: use common 4k zero page for dax mmap reads · 91d25ba8
      由 Ross Zwisler 提交于 
      When servicing mmap() reads from file holes the current DAX code
allocates a page cache page of all zeroes and places the struct page
pointer in the mapping->page_tree radix tree.

This has three major drawbacks:

1) It consumes memory unnecessarily. For every 4k page that is read via
   a DAX mmap() over a hole, we allocate a new page cache page. This
   means that if you read 1GiB worth of pages, you end up using 1GiB of
   zeroed memory. This is easily visible by looking at the overall
   memory consumption of the system or by looking at /proc/[pid]/smaps:

	7f62e72b3000-7f63272b3000 rw-s 00000000 103:00 12   /root/dax/data
	Size:            1048576 kB
	Rss:             1048576 kB
	Pss:             1048576 kB
	Shared_Clean:          0 kB
	Shared_Dirty:          0 kB
	Private_Clean:   1048576 kB
	Private_Dirty:         0 kB
	Referenced:      1048576 kB
	Anonymous:             0 kB
	LazyFree:              0 kB
	AnonHugePages:         0 kB
	ShmemPmdMapped:        0 kB
	Shared_Hugetlb:        0 kB
	Private_Hugetlb:       0 kB
	Swap:                  0 kB
	SwapPss:               0 kB
	KernelPageSize:        4 kB
	MMUPageSize:           4 kB
	Locked:                0 kB

2) It is slower than using a common zero page because each page fault
   has more work to do. Instead of just inserting a common zero page we
   have to allocate a page cache page, zero it, and then insert it. Here
   are the average latencies of dax_load_hole() as measured by ftrace on
   a random test box:

    Old method, using zeroed page cache pages:	3.4 us
    New method, using the common 4k zero page:	0.8 us

   This was the average latency over 1 GiB of sequential reads done by
   this simple fio script:

     [global]
     size=1G
     filename=/root/dax/data
     fallocate=none
     [io]
     rw=read
     ioengine=mmap

3) The fact that we had to check for both DAX exceptional entries and
   for page cache pages in the radix tree made the DAX code more
   complex.

Solve these issues by following the lead of the DAX PMD code and using a
common 4k zero page instead.  As with the PMD code we will now insert a
DAX exceptional entry into the radix tree instead of a struct page
pointer which allows us to remove all the special casing in the DAX
code.

Note that we do still pretty aggressively check for regular pages in the
DAX radix tree, especially where we take action based on the bits set in
the page.  If we ever find a regular page in our radix tree now that
most likely means that someone besides DAX is inserting pages (which has
happened lots of times in the past), and we want to find that out early
and fail loudly.

This solution also removes the extra memory consumption.  Here is that
same /proc/[pid]/smaps after 1GiB of reading from a hole with the new
code:

	7f2054a74000-7f2094a74000 rw-s 00000000 103:00 12   /root/dax/data
	Size:            1048576 kB
	Rss:                   0 kB
	Pss:                   0 kB
	Shared_Clean:          0 kB
	Shared_Dirty:          0 kB
	Private_Clean:         0 kB
	Private_Dirty:         0 kB
	Referenced:            0 kB
	Anonymous:             0 kB
	LazyFree:              0 kB
	AnonHugePages:         0 kB
	ShmemPmdMapped:        0 kB
	Shared_Hugetlb:        0 kB
	Private_Hugetlb:       0 kB
	Swap:                  0 kB
	SwapPss:               0 kB
	KernelPageSize:        4 kB
	MMUPageSize:           4 kB
	Locked:                0 kB

Overall system memory consumption is similarly improved.

Another major change is that we remove dax_pfn_mkwrite() from our fault
flow, and instead rely on the page fault itself to make the PTE dirty
and writeable.  The following description from the patch adding the
vm_insert_mixed_mkwrite() call explains this a little more:

   "To be able to use the common 4k zero page in DAX we need to have our
    PTE fault path look more like our PMD fault path where a PTE entry
    can be marked as dirty and writeable as it is first inserted rather
    than waiting for a follow-up dax_pfn_mkwrite() =>
    finish_mkwrite_fault() call.

    Right now we can rely on having a dax_pfn_mkwrite() call because we
    can distinguish between these two cases in do_wp_page():

            case 1: 4k zero page => writable DAX storage
            case 2: read-only DAX storage => writeable DAX storage

    This distinction is made by via vm_normal_page(). vm_normal_page()
    returns false for the common 4k zero page, though, just as it does
    for DAX ptes. Instead of special casing the DAX + 4k zero page case
    we will simplify our DAX PTE page fault sequence so that it matches
    our DAX PMD sequence, and get rid of the dax_pfn_mkwrite() helper.
    We will instead use dax_iomap_fault() to handle write-protection
    faults.

    This means that insert_pfn() needs to follow the lead of
    insert_pfn_pmd() and allow us to pass in a 'mkwrite' flag. If
    'mkwrite' is set insert_pfn() will do the work that was previously
    done by wp_page_reuse() as part of the dax_pfn_mkwrite() call path"

Link: http://lkml.kernel.org/r/20170724170616.25810-4-ross.zwisler@linux.intel.comSigned-off-by: NRoss Zwisler <ross.zwisler@linux.intel.com>
Reviewed-by: NJan Kara <jack@suse.cz>
Cc: "Darrick J. Wong" <darrick.wong@oracle.com>
Cc: "Theodore Ts'o" <tytso@mit.edu>
Cc: Alexander Viro <viro@zeniv.linux.org.uk>
Cc: Andreas Dilger <adilger.kernel@dilger.ca>
Cc: Christoph Hellwig <hch@lst.de>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Dave Chinner <david@fromorbit.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: Matthew Wilcox <mawilcox@microsoft.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>
      91d25ba8
  23. 05 9月, 2017 1 次提交
  25. 02 9月, 2017 2 次提交
  27. 06 7月, 2017 1 次提交
  29. 03 7月, 2017 1 次提交
  31. 28 6月, 2017 1 次提交
  33. 21 6月, 2017 1 次提交
  35. 20 6月, 2017 1 次提交
  37. 26 5月, 2017 4 次提交
  39. 28 2月, 2017 1 次提交
  41. 25 2月, 2017 3 次提交
  43. 23 2月, 2017 2 次提交
  45. 07 2月, 2017 2 次提交
  47. 03 2月, 2017 1 次提交
  49. 31 1月, 2017 1 次提交
    • B
      xfs: sync eofblocks scans under iolock are livelock prone · c3155097
      由 Brian Foster 提交于 
      The xfs_eofblocks.eof_scan_owner field is an internal field to
facilitate invoking eofb scans from the kernel while under the iolock.
This is necessary because the eofb scan acquires the iolock of each
inode. Synchronous scans are invoked on certain buffered write failures
while under iolock. In such cases, the scan owner indicates that the
context for the scan already owns the particular iolock and prevents a
double lock deadlock.

eofblocks scans while under iolock are still livelock prone in the event
of multiple parallel scans, however. If multiple buffered writes to
different inodes fail and invoke eofblocks scans at the same time, each
scan avoids a deadlock with its own inode by virtue of the
eof_scan_owner field, but will never be able to acquire the iolock of
the inode from the parallel scan. Because the low free space scans are
invoked with SYNC_WAIT, the scan will not return until it has processed
every tagged inode and thus both scans will spin indefinitely on the
iolock being held across the opposite scan. This problem can be
reproduced reliably by generic/224 on systems with higher cpu counts
(x16).

To avoid this problem, simplify the semantics of eofblocks scans to
never invoke a scan while under iolock. This means that the buffered
write context must drop the iolock before the scan. It must reacquire
the lock before the write retry and also repeat the initial write
checks, as the original state might no longer be valid once the iolock
was dropped.
Signed-off-by: NBrian Foster <bfoster@redhat.com>
Reviewed-by: NChristoph Hellwig <hch@lst.de>
Reviewed-by: NDarrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: NDarrick J. Wong <darrick.wong@oracle.com>
      c3155097
  51. 10 12月, 2016 1 次提交
    • C
      fs: try to clone files first in vfs_copy_file_range · a76b5b04
      由 Christoph Hellwig 提交于 
      A clone is a perfectly fine implementation of a file copy, so most
file systems just implement the copy that way.  Instead of duplicating
this logic move it to the VFS.  Currently btrfs and XFS implement copies
the same way as clones and there is no behavior change for them, cifs
only implements clones and grow support for copy_file_range with this
patch.  NFS implements both, so this will allow copy_file_range to work
on servers that only implement CLONE and be lot more efficient on servers
that implements CLONE and COPY.
Signed-off-by: NChristoph Hellwig <hch@lst.de>
      a76b5b04
  53. 09 12月, 2016 1 次提交