Skip to content

K
Kernel

openeuler / Kernel
大约 1 年 前同步成功

通知 5
Star 0
Fork 0
K
Kernel

体验新版 GitCode，发现更多精彩内容 >>

切换分支/标签
  1. 22 11月, 2013 1 次提交
    • J
      configfs: fix race between dentry put and lookup · 76ae281f
      由 Junxiao Bi 提交于 
      A race window in configfs, it starts from one dentry is UNHASHED and end
before configfs_d_iput is called.  In this window, if a lookup happen,
since the original dentry was UNHASHED, so a new dentry will be
allocated, and then in configfs_attach_attr(), sd->s_dentry will be
updated to the new dentry.  Then in configfs_d_iput(),
BUG_ON(sd->s_dentry != dentry) will be triggered and system panic.

sys_open:                     sys_close:
 ...                           fput
                                dput
                                 dentry_kill
                                  __d_drop <--- dentry unhashed here,
                                           but sd->dentry still point
                                           to this dentry.

 lookup_real
  configfs_lookup
   configfs_attach_attr---> update sd->s_dentry
                            to new allocated dentry here.

                                   d_kill
                                     configfs_d_iput <--- BUG_ON(sd->s_dentry != dentry)
                                                     triggered here.

To fix it, change configfs_d_iput to not update sd->s_dentry if
sd->s_count > 2, that means there are another dentry is using the sd
beside the one that is going to be put.  Use configfs_dirent_lock in
configfs_attach_attr to sync with configfs_d_iput.

With the following steps, you can reproduce the bug.

1. enable ocfs2, this will mount configfs at /sys/kernel/config and
   fill configure in it.

2. run the following script.
	while [ 1 ]; do cat /sys/kernel/config/cluster/$your_cluster_name/idle_timeout_ms > /dev/null; done &
	while [ 1 ]; do cat /sys/kernel/config/cluster/$your_cluster_name/idle_timeout_ms > /dev/null; done &
Signed-off-by: NJunxiao Bi <junxiao.bi@oracle.com>
Cc: Joel Becker <jlbec@evilplan.org>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: <stable@vger.kernel.org>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>
      76ae281f
  3. 21 11月, 2013 16 次提交
  5. 20 11月, 2013 11 次提交
    • P
      Squashfs: Check stream is not NULL in decompressor_multi.c · ed4f381e
      由 Phillip Lougher 提交于 
      Fix static checker complaint that stream is not checked in
squashfs_decompressor_destroy().
Reported-by: NDan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: NPhillip Lougher <phillip@squashfs.org.uk>
Reviewed-by: NMinchan Kim <minchan@kernel.org>
      ed4f381e
    • P
      Squashfs: Directly decompress into the page cache for file data · 0d455c12
      由 Phillip Lougher 提交于 
      This introduces an implementation of squashfs_readpage_block()
that directly decompresses into the page cache.

This uses the previously added page handler abstraction to push
down the necessary kmap_atomic/kunmap_atomic operations on the
page cache buffers into the decompressors.  This enables
direct copying into the page cache without using the slow
kmap/kunmap calls.

The code detects when multiple threads are racing in
squashfs_readpage() to decompress the same block, and avoids
this regression by falling back to using an intermediate
buffer.

This patch enhances the performance of Squashfs significantly
when multiple processes are accessing the filesystem simultaneously
because it not only reduces memcopying, but it more importantly
eliminates the lock contention on the intermediate buffer.

Using single-thread decompression.

        dd if=file1 of=/dev/null bs=4096 &
        dd if=file2 of=/dev/null bs=4096 &
        dd if=file3 of=/dev/null bs=4096 &
        dd if=file4 of=/dev/null bs=4096

Before:

629145600 bytes (629 MB) copied, 45.8046 s, 13.7 MB/s

After:

629145600 bytes (629 MB) copied, 9.29414 s, 67.7 MB/s
Signed-off-by: NPhillip Lougher <phillip@squashfs.org.uk>
Reviewed-by: NMinchan Kim <minchan@kernel.org>
      0d455c12
    • P
      Squashfs: Restructure squashfs_readpage() · 5f55dbc0
      由 Phillip Lougher 提交于 
      Restructure squashfs_readpage() splitting it into separate
functions for datablocks, fragments and sparse blocks.

Move the memcpying (from squashfs cache entry) implementation of
squashfs_readpage_block into file_cache.c

This allows different implementations to be supported.
Signed-off-by: NPhillip Lougher <phillip@squashfs.org.uk>
Reviewed-by: NMinchan Kim <minchan@kernel.org>
      5f55dbc0
    • P
      Squashfs: Generalise paging handling in the decompressors · 846b730e
      由 Phillip Lougher 提交于 
      Further generalise the decompressors by adding a page handler
abstraction.  This adds helpers to allow the decompressors
to access and process the output buffers in an implementation
independant manner.

This allows different types of output buffer to be passed
to the decompressors, with the implementation specific
aspects handled at decompression time, but without the
knowledge being held in the decompressor wrapper code.

This will allow the decompressors to handle Squashfs
cache buffers, and page cache pages.

This patch adds the abstraction and an implementation for
the caches.
Signed-off-by: NPhillip Lougher <phillip@squashfs.org.uk>
Reviewed-by: NMinchan Kim <minchan@kernel.org>
      846b730e
    • P
      Squashfs: add multi-threaded decompression using percpu variable · d208383d
      由 Phillip Lougher 提交于 
      Add a multi-threaded decompression implementation which uses
percpu variables.

Using percpu variables has advantages and disadvantages over
implementations which do not use percpu variables.

Advantages:
  * the nature of percpu variables ensures decompression is
    load-balanced across the multiple cores.
  * simplicity.

Disadvantages: it limits decompression to one thread per core.
Signed-off-by: NPhillip Lougher <phillip@squashfs.org.uk>
      d208383d
    • M
      squashfs: Enhance parallel I/O · cd59c2ec
      由 Minchan Kim 提交于 
      Now squashfs have used for only one stream buffer for decompression
so it hurts parallel read performance so this patch supports
multiple decompressor to enhance performance parallel I/O.

Four 1G file dd read on KVM machine which has 2 CPU and 4G memory.

dd if=test/test1.dat of=/dev/null &
dd if=test/test2.dat of=/dev/null &
dd if=test/test3.dat of=/dev/null &
dd if=test/test4.dat of=/dev/null &

old : 1m39s -> new : 9s

* From v1
  * Change comp_strm with decomp_strm - Phillip
  * Change/add comments - Phillip
Signed-off-by: NMinchan Kim <minchan@kernel.org>
Signed-off-by: NPhillip Lougher <phillip@squashfs.org.uk>
      cd59c2ec
    • P
      Squashfs: Refactor decompressor interface and code · 9508c6b9
      由 Phillip Lougher 提交于 
      The decompressor interface and code was written from
the point of view of single-threaded operation.  In doing
so it mixed a lot of single-threaded implementation specific
aspects into the decompressor code and elsewhere which makes it
difficult to seamlessly support multiple different decompressor
implementations.

This patch does the following:

1.  It removes compressor_options parsing from the decompressor
    init() function.  This allows the decompressor init() function
    to be dynamically called to instantiate multiple decompressors,
    without the compressor options needing to be read and parsed each
    time.

2.  It moves threading and all sleeping operations out of the
    decompressors.  In doing so, it makes the decompressors
    non-blocking wrappers which only deal with interfacing with
    the decompressor implementation.

3. It splits decompressor.[ch] into decompressor generic functions
   in decompressor.[ch], and moves the single threaded
   decompressor implementation into decompressor_single.c.

The result of this patch is Squashfs should now be able to
support multiple decompressors by adding new decompressor_xxx.c
files with specialised implementations of the functions in
decompressor_single.c
Signed-off-by: NPhillip Lougher <phillip@squashfs.org.uk>
Reviewed-by: NMinchan Kim <minchan@kernel.org>
      9508c6b9
    • J
      genetlink: make multicast groups const, prevent abuse · 2a94fe48
      由 Johannes Berg 提交于 
      Register generic netlink multicast groups as an array with
the family and give them contiguous group IDs. Then instead
of passing the global group ID to the various functions that
send messages, pass the ID relative to the family - for most
families that's just 0 because the only have one group.

This avoids the list_head and ID in each group, adding a new
field for the mcast group ID offset to the family.

At the same time, this allows us to prevent abusing groups
again like the quota and dropmon code did, since we can now
check that a family only uses a group it owns.
Signed-off-by: NJohannes Berg <johannes.berg@intel.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>
      2a94fe48
    • J
      genetlink: pass family to functions using groups · 68eb5503
      由 Johannes Berg 提交于 
      This doesn't really change anything, but prepares for the
next patch that will change the APIs to pass the group ID
within the family, rather than the global group ID.
Signed-off-by: NJohannes Berg <johannes.berg@intel.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>
      68eb5503
    • J
      quota/genetlink: use proper genetlink multicast APIs · 2ecf7536
      由 Johannes Berg 提交于 
      The quota code is abusing the genetlink API and is using
its family ID as the multicast group ID, which is invalid
and may belong to somebody else (and likely will.)

Make the quota code use the correct API, but since this
is already used as-is by userspace, reserve a family ID
for this code and also reserve that group ID to not break
userspace assumptions.
Acked-by: NJan Kara <jack@suse.cz>
Signed-off-by: NJohannes Berg <johannes.berg@intel.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>
      2ecf7536
    • J
      genetlink: only pass array to genl_register_family_with_ops() · c53ed742
      由 Johannes Berg 提交于 
      As suggested by David Miller, make genl_register_family_with_ops()
a macro and pass only the array, evaluating ARRAY_SIZE() in the
macro, this is a little safer.

The openvswitch has some indirection, assing ops/n_ops directly in
that code. This might ultimately just assign the pointers in the
family initializations, saving the struct genl_family_and_ops and
code (once mcast groups are handled differently.)
Signed-off-by: NJohannes Berg <johannes.berg@intel.com>
Signed-off-by: NDavid S. Miller <davem@davemloft.net>
      c53ed742
  7. 19 11月, 2013 2 次提交
  9. 18 11月, 2013 3 次提交
    • D
      xfs: open code inc_inode_iversion when logging an inode · 2fe8c1c0
      由 Dave Chinner 提交于 
      Michael L Semon reported that generic/069 runtime increased on v5
superblocks by 100% compared to v4 superblocks. his perf-based
analysis pointed directly at the timestamp updates being done by the
write path in this workload. The append writers are doing 4-byte
writes, so there are lots of timestamp updates occurring.

The thing is, they aren't being triggered by timestamp changes -
they are being triggered by the inode change counter needing to be
updated. That is, every write(2) system call needs to bump the inode
version count, and it does that through the timestamp update
mechanism. Hence for v5 filesystems, test generic/069 is running 3
orders of magnitude more timestmap update transactions on v5
filesystems due to the fact it does a huge number of *4 byte*
write(2) calls.

This isn't a real world scenario we really need to address - anyone
doing such sequential IO should be using fwrite(3), not write(2).
i.e. fwrite(3) buffers the writes in userspace to minimise the
number of write(2) syscalls, and the problem goes away.

However, there is a small change we can make to improve the
situation - removing the expensive lock operation on the change
counter update.  All inode version counter changes in XFS occur
under the ip->i_ilock during a transaction, and therefore we
don't actually need the spin lock that provides exclusive access to
it through inc_inode_iversion().

Hence avoid the lock and just open code the increment ourselves when
logging the inode.
Reported-by: NMichael L. Semon <mlsemon35@gmail.com>
Signed-off-by: NDave Chinner <dchinner@redhat.com>
Reviewed-by: NChristoph Hellwig <hch@lst.de>
Signed-off-by: NBen Myers <bpm@sgi.com>
      2fe8c1c0
    • D
      xfs: increase inode cluster size for v5 filesystems · 8f80587b
      由 Dave Chinner 提交于 
      v5 filesystems use 512 byte inodes as a minimum, so read inodes in
clusters that are effectively half the size of a v4 filesystem with
256 byte inodes. For v5 fielsystems, scale the inode cluster size
with the size of the inode so that we keep a constant 32 inodes per
cluster ratio for all inode IO.

This only works if mkfs.xfs sets the inode alignment appropriately
for larger inode clusters, so this functionality is made conditional
on mkfs doing the right thing. xfs_repair needs to know about
the inode alignment changes, too.

Wall time:
	create	bulkstat	find+stat	ls -R	unlink
v4	237s	161s		173s		201s	299s
v5	235s	163s		205s		 31s	356s
patched	234s	160s		182s		 29s	317s

System time:
	create	bulkstat	find+stat	ls -R	unlink
v4	2601s	2490s		1653s		1656s	2960s
v5	2637s	2497s		1681s		  20s	3216s
patched	2613s	2451s		1658s		  20s	3007s

So, wall time same or down across the board, system time same or
down across the board, and cache hit rates all improve except for
the ls -R case which is a pure cold cache directory read workload
on v5 filesystems...

So, this patch removes most of the performance and CPU usage
differential between v4 and v5 filesystems on traversal related
workloads.

Note: while this patch is currently for v5 filesystems only, there
is no reason it can't be ported back to v4 filesystems.  This hasn't
been done here because bringing the code back to v4 requires
forwards and backwards kernel compatibility testing.  i.e. to
deterine if older kernels(*) do the right thing with larger inode
alignments but still only using 8k inode cluster sizes. None of this
testing and validation on v4 filesystems has been done, so for the
moment larger inode clusters is limited to v5 superblocks.

(*) a current default config v4 filesystem should mount just fine on
2.6.23 (when lazy-count support was introduced), and so if we change
the alignment emitted by mkfs without a feature bit then we have to
make sure it works properly on all kernels since 2.6.23. And if we
allow it to be changed when the lazy-count bit is not set, then it's
all kernels since v2 logs were introduced that need to be tested for
compatibility...
Signed-off-by: NDave Chinner <dchinner@redhat.com>
Reviewed-by: NChristoph Hellwig <hch@lst.de>
Reviewed-by: NEric Sandeen <sandeen@redhat.com>
Signed-off-by: NBen Myers <bpm@sgi.com>
      8f80587b
    • M
      xfs: fix unlock in xfs_bmap_add_attrfork · 9e3908e3
      由 Mark Tinguely 提交于 
      xfs_trans_ijoin() activates the inode in a transaction and
also can specify which lock to free when the transaction is
committed or canceled.

xfs_bmap_add_attrfork call locks and adds the lock to the
transaction but also manually removes the lock. Change the
routine to not add the lock to the transaction and manually
remove lock on completion.

While here, clean up the xfs_trans_cancel flags and goto names.
Signed-off-by: NMark Tinguely <tinguely@sgi.com>
Reviewed-by: NChristoph Hellwig <hch@lst.de>
Signed-off-by: NBen Myers <bpm@sgi.com>
      9e3908e3
  11. 16 11月, 2013 7 次提交