Put dentry and inode fields into top of data structure. This allows RCU path
traversal to perform an RCU dentry lookup in a path walk by touching only the
first 56 bytes of the dentry.
We also fit in 8 bytes of inline name in the first 64 bytes, so for short
names, only 64 bytes needs to be touched to perform the lookup. We should
get rid of the hash->prev pointer from the first 64 bytes, and fit 16 bytes
of name in there, which will take care of 81% rather than 32% of the kernel
tree.
inode is also rearranged so that RCU lookup will only touch a single cacheline
in the inode, plus one in the i_ops structure.
This is important for directory component lookups in RCU path walking. In the
kernel source, directory names average is around 6 chars, so this works.
When we reach the last element of the lookup, we need to lock it and take its
refcount which requires another cacheline access.
Align dentry and inode operations structs, so members will be at predictable
offsets and we can group common operations into head of structure.
Signed-off-by: Nick Piggin <npiggin@kernel.dk>
Pseudo filesystems that don't put inode on RCU list or reachable by
rcu-walk dentries do not need to RCU free their inodes.
Signed-off-by: Nick Piggin <npiggin@kernel.dk>
RCU free the struct inode. This will allow:
- Subsequent store-free path walking patch. The inode must be consulted for
permissions when walking, so an RCU inode reference is a must.
- sb_inode_list_lock to be moved inside i_lock because sb list walkers who want
to take i_lock no longer need to take sb_inode_list_lock to walk the list in
the first place. This will simplify and optimize locking.
- Could remove some nested trylock loops in dcache code
- Could potentially simplify things a bit in VM land. Do not need to take the
page lock to follow page->mapping.
The downsides of this is the performance cost of using RCU. In a simple
creat/unlink microbenchmark, performance drops by about 10% due to inability to
reuse cache-hot slab objects. As iterations increase and RCU freeing starts
kicking over, this increases to about 20%.
In cases where inode lifetimes are longer (ie. many inodes may be allocated
during the average life span of a single inode), a lot of this cache reuse is
not applicable, so the regression caused by this patch is smaller.
The cache-hot regression could largely be avoided by using SLAB_DESTROY_BY_RCU,
however this adds some complexity to list walking and store-free path walking,
so I prefer to implement this at a later date, if it is shown to be a win in
real situations. I haven't found a regression in any non-micro benchmark so I
doubt it will be a problem.
Signed-off-by: Nick Piggin <npiggin@kernel.dk>
NFS needs to be able to release objects that are stored in the page
cache once the page itself is no longer visible from the page cache.
This patch adds a callback to the address space operations that allows
filesystems to perform page cleanups once the page has been removed
from the page cache.
Original patch by: Linus Torvalds <torvalds@linux-foundation.org>
[trondmy: cover the cases of invalidate_inode_pages2() and
truncate_inode_pages()]
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
dpkg uses fiemap but didn't particularly need to include stdint.h so far.
Since 367a51a339 ("fs: Add FITRIM ioctl"), build of linux/fs.h failed in
dpkg with:
In file included from ../../src/filesdb.c:27:0:
/usr/include/linux/fs.h:37:2: error: expected specifier-qualifier-list before 'uint64_t'
Use exportable type __u64 to avoid the dependency on stdint.h.
b31d42a5af ("Fix compile brekage with !CONFIG_BLOCK") fixed only the
kernel build by including linux/types.h, but this also fixed "make
headers_check", so don't revert it.
Signed-off-by: Loïc Minier <loic.minier@linaro.org>
Tested-by: Arnd Bergmann <arnd.bergmann@linaro.org>
Cc: Lukas Czerner <lczerner@redhat.com>
Cc: Dmitry Monakhov <dmonakhov@openvz.org>
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>
There was concern that FITRIM ioctl is not common enough to be included
in core vfs ioctl, as Christoph Hellwig pointed out there's no real point
in dispatching this out to a separate vector instead of just through
->ioctl.
So this commit removes ioctl_fstrim() from vfs ioctl and trim_fs
from super_operation structure.
Signed-off-by: Lukas Czerner <lczerner@redhat.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
This one was only used for a nasty hack in nfsd, which has recently
been removed.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
We modified setlease to require the caller to allocate the new lease in
the case of creating a new lease, but forgot to fix up the filesystem
methods.
Cc: Steven Whitehouse <swhiteho@redhat.com>
Cc: Steve French <sfrench@samba.org>
Cc: Trond Myklebust <Trond.Myklebust@netapp.com>
Signed-off-by: J. Bruce Fields <bfields@redhat.com>
Acked-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
We used to protect against overflow, but rather than return an error, do
what read/write does, namely to limit the total size to MAX_RW_COUNT.
This is not only more consistent, but it also means that any broken
low-level read/write routine that still keeps counts in 'int' can't
break.
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
eventual replacement for ->get_sb() - does *not* get vfsmount,
return ERR_PTR(error) or root of subtree to be mounted.
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Today's git tree fails to build on !CONFIG_BLOCK, due to upstream commit
367a51a339 ("fs: Add FITRIM ioctl"):
include/linux/fs.h:36: error: expected specifier-qualifier-list before ‘uint64_t’
include/linux/fs.h:36: error: expected specifier-qualifier-list before ‘uint64_t’
include/linux/fs.h:36: error: expected specifier-qualifier-list before ‘uint64_t’
The commit adds uint64_t type usage to fs.h, but linux/types.h is not included
explicitly - it's only included implicitly via linux/blk_types.h, and there only if
CONFIG_BLOCK is enabled.
Add the explicit #include to fix this.
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Adds an filesystem independent ioctl to allow implementation of file
system batched discard support. I takes fstrim_range structure as an
argument. fstrim_range is definec in the include/fs.h and its
definition is as follows.
struct fstrim_range {
start;
len;
minlen;
}
start - first Byte to trim
len - number of Bytes to trim from start
minlen - minimum extent length to trim, free extents shorter than this
number of Bytes will be ignored. This will be rounded up to fs
block size.
It is also possible to specify NULL as an argument. In this case the
arguments will set itself as follows:
start = 0;
len = ULLONG_MAX;
minlen = 0;
So it will trim the whole file system at one run.
After the FITRIM is done, the number of actually discarded Bytes is stored
in fstrim_range.len to give the user better insight on how much storage
space has been really released for wear-leveling.
Signed-off-by: Lukas Czerner <lczerner@redhat.com>
Reviewed-by: Dmitry Monakhov <dmonakhov@openvz.org>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
You currently cannot use "fasync_helper()" in an atomic environment to
insert a new fasync entry, because it will need to allocate the new
"struct fasync_struct".
Yet fcntl_setlease() wants to call this under lock_flocks(), which is in
the process of being converted from the BKL to a spinlock.
In order to fix this, this abstracts out the actual fasync list
insertion and the fasync allocations into functions of their own, and
teaches fs/locks.c to pre-allocate the fasync_struct entry. That way
the actual list insertion can happen while holding the required
spinlock.
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
[bfields@redhat.com: rebase on top of my changes to Arnd's patch]
Tested-by: J. Bruce Fields <bfields@redhat.com>
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
As suggested by Christoph Hellwig, this moves allocation
of new file locks out of generic_setlease into the
callers, nfs4_open_delegation and fcntl_setlease in order
to allow GFP_KERNEL allocations when lock_flocks has
become a spinlock.
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Acked-by: J. Bruce Fields <bfields@redhat.com>
Robin Holt tried to boot a 16TB system and found af_unix was overflowing
a 32bit value :
<quote>
We were seeing a failure which prevented boot. The kernel was incapable
of creating either a named pipe or unix domain socket. This comes down
to a common kernel function called unix_create1() which does:
atomic_inc(&unix_nr_socks);
if (atomic_read(&unix_nr_socks) > 2 * get_max_files())
goto out;
The function get_max_files() is a simple return of files_stat.max_files.
files_stat.max_files is a signed integer and is computed in
fs/file_table.c's files_init().
n = (mempages * (PAGE_SIZE / 1024)) / 10;
files_stat.max_files = n;
In our case, mempages (total_ram_pages) is approx 3,758,096,384
(0xe0000000). That leaves max_files at approximately 1,503,238,553.
This causes 2 * get_max_files() to integer overflow.
</quote>
Fix is to let /proc/sys/fs/file-nr & /proc/sys/fs/file-max use long
integers, and change af_unix to use an atomic_long_t instead of atomic_t.
get_max_files() is changed to return an unsigned long. get_nr_files() is
changed to return a long.
unix_nr_socks is changed from atomic_t to atomic_long_t, while not
strictly needed to address Robin problem.
Before patch (on a 64bit kernel) :
# echo 2147483648 >/proc/sys/fs/file-max
# cat /proc/sys/fs/file-max
-18446744071562067968
After patch:
# echo 2147483648 >/proc/sys/fs/file-max
# cat /proc/sys/fs/file-max
2147483648
# cat /proc/sys/fs/file-nr
704 0 2147483648
Reported-by: Robin Holt <holt@sgi.com>
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Acked-by: David Miller <davem@davemloft.net>
Reviewed-by: Robin Holt <holt@sgi.com>
Tested-by: Robin Holt <holt@sgi.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Christoph Hellwig <hch@lst.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Currently for every removed inode IMA must take a global lock and search
the IMA rbtree looking for an associated integrity structure. Instead
we explicitly mark an inode when we add an integrity structure so we
only have to take the global lock and do the removal if it exists.
Signed-off-by: Eric Paris <eparis@redhat.com>
Acked-by: Mimi Zohar <zohar@linux.vnet.ibm.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
IMA currently allocated an inode integrity structure for every inode in
core. This stucture is about 120 bytes long. Most files however
(especially on a system which doesn't make use of IMA) will never need
any of this space. The problem is that if IMA is enabled we need to
know information about the number of readers and the number of writers
for every inode on the box. At the moment we collect that information
in the per inode iint structure and waste the rest of the space. This
patch moves those counters into the struct inode so we can eventually
stop allocating an IMA integrity structure except when absolutely
needed.
This patch does the minimum needed to move the location of the data.
Further cleanups, especially the location of counter updates, may still
be possible.
Signed-off-by: Eric Paris <eparis@redhat.com>
Acked-by: Mimi Zohar <zohar@linux.vnet.ibm.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
The use of the same inode list structure (inode->i_list) for two
different list constructs with different lifecycles and purposes
makes it impossible to separate the locking of the different
operations. Therefore, to enable the separation of the locking of
the writeback and reclaim lists, split the inode->i_list into two
separate lists dedicated to their specific tracking functions.
Signed-off-by: Nick Piggin <npiggin@suse.de>
Signed-off-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
The nr_dentry stat is a globally touched cacheline and atomic operation
twice over the lifetime of a dentry. It is used for the benfit of userspace
only. Turn it into a per-cpu counter and always decrement it in d_free instead
of doing various batching operations to reduce lock hold times in the callers.
Based on an earlier patch from Nick Piggin <npiggin@suse.de>.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Instead of always assigning an increasing inode number in new_inode
move the call to assign it into those callers that actually need it.
For now callers that need it is estimated conservatively, that is
the call is added to all filesystems that do not assign an i_ino
by themselves. For a few more filesystems we can avoid assigning
any inode number given that they aren't user visible, and for others
it could be done lazily when an inode number is actually needed,
but that's left for later patches.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Split up inode_add_to_list/__inode_add_to_list. Locking for the two
lists will be split soon so these helpers really don't buy us much
anymore.
The __ prefixes for the sb list helpers will go away soon, but until
inode_lock is gone we'll need them to distinguish between the locked
and unlocked variants.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Convert the inode LRU to use lazy updates to reduce lock and
cacheline traffic. We avoid moving inodes around in the LRU list
during iget/iput operations so these frequent operations don't need
to access the LRUs. Instead, we defer the refcount checks to
reclaim-time and use a per-inode state flag, I_REFERENCED, to tell
reclaim that iget has touched the inode in the past. This means that
only reclaim should be touching the LRU with any frequency, hence
significantly reducing lock acquisitions and the amount contention
on LRU updates.
This also removes the inode_in_use list, which means we now only
have one list for tracking the inode LRU status. This makes it much
simpler to split out the LRU list operations under it's own lock.
Signed-off-by: Nick Piggin <npiggin@suse.de>
Signed-off-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
The number of inodes allocated does not need to be tied to the
addition or removal of an inode to/from a list. If we are not tied
to a list lock, we could update the counters when inodes are
initialised or destroyed, but to do that we need to convert the
counters to be per-cpu (i.e. independent of a lock). This means that
we have the freedom to change the list/locking implementation
without needing to care about the counters.
Based on a patch originally from Eric Dumazet.
[AV: cleaned up a bit, fixed build breakage on weird configs
Signed-off-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Now, rw_verify_area() checsk f_pos is negative or not. And if negative,
returns -EINVAL.
But, some special files as /dev/(k)mem and /proc/<pid>/mem etc.. has
negative offsets. And we can't do any access via read/write to the
file(device).
So introduce FMODE_UNSIGNED_OFFSET to allow negative file offsets.
Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Al Viro <viro@ZenIV.linux.org.uk>
Cc: Heiko Carstens <heiko.carstens@de.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Andrew,
Could you please review this patch, you probably are the right guy to
take it, because it crosses fs and net trees.
Note : /proc/sys/fs/file-nr is a read-only file, so this patch doesnt
depend on previous patch (sysctl: fix min/max handling in
__do_proc_doulongvec_minmax())
Thanks !
[PATCH V4] fs: allow for more than 2^31 files
Robin Holt tried to boot a 16TB system and found af_unix was overflowing
a 32bit value :
<quote>
We were seeing a failure which prevented boot. The kernel was incapable
of creating either a named pipe or unix domain socket. This comes down
to a common kernel function called unix_create1() which does:
atomic_inc(&unix_nr_socks);
if (atomic_read(&unix_nr_socks) > 2 * get_max_files())
goto out;
The function get_max_files() is a simple return of files_stat.max_files.
files_stat.max_files is a signed integer and is computed in
fs/file_table.c's files_init().
n = (mempages * (PAGE_SIZE / 1024)) / 10;
files_stat.max_files = n;
In our case, mempages (total_ram_pages) is approx 3,758,096,384
(0xe0000000). That leaves max_files at approximately 1,503,238,553.
This causes 2 * get_max_files() to integer overflow.
</quote>
Fix is to let /proc/sys/fs/file-nr & /proc/sys/fs/file-max use long
integers, and change af_unix to use an atomic_long_t instead of
atomic_t.
get_max_files() is changed to return an unsigned long.
get_nr_files() is changed to return a long.
unix_nr_socks is changed from atomic_t to atomic_long_t, while not
strictly needed to address Robin problem.
Before patch (on a 64bit kernel) :
# echo 2147483648 >/proc/sys/fs/file-max
# cat /proc/sys/fs/file-max
-18446744071562067968
After patch:
# echo 2147483648 >/proc/sys/fs/file-max
# cat /proc/sys/fs/file-max
2147483648
# cat /proc/sys/fs/file-nr
704 0 2147483648
Reported-by: Robin Holt <holt@sgi.com>
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Acked-by: David Miller <davem@davemloft.net>
Reviewed-by: Robin Holt <holt@sgi.com>
Tested-by: Robin Holt <holt@sgi.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Hugetlbfs used to need it, but after the destroy_inode and evict_inode
changes it's not required anymore.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Add a new helper to write out the inode using the writeback code,
that is including the correct dirty bit and list manipulation. A few
of filesystems already opencode this, and a lot of others should be
using it instead of using write_inode_now which also writes out the
data.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
This prepares the removal of the big kernel lock from the
file locking code. We still use the BKL as long as fs/lockd
uses it and ceph might sleep, but we can flip the definition
to a private spinlock as soon as that's done.
All users outside of fs/lockd get converted to use
lock_flocks() instead of lock_kernel() where appropriate.
Based on an earlier patch to use a spinlock from Matthew
Wilcox, who has attempted this a few times before, the
earliest patch from over 10 years ago turned it into
a semaphore, which ended up being slower than the BKL
and was subsequently reverted.
Someone should do some serious performance testing when
this becomes a spinlock, since this has caused problems
before. Using a spinlock should be at least as good
as the BKL in theory, but who knows...
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Acked-by: Matthew Wilcox <willy@linux.intel.com>
Cc: Christoph Hellwig <hch@lst.de>
Cc: Trond Myklebust <trond.myklebust@fys.uio.no>
Cc: "J. Bruce Fields" <bfields@fieldses.org>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Miklos Szeredi <mszeredi@suse.cz>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: John Kacur <jkacur@redhat.com>
Cc: Sage Weil <sage@newdream.net>
Cc: linux-kernel@vger.kernel.org
Cc: linux-fsdevel@vger.kernel.org
The lock structs are currently protected by the BKL, but are accessed by
code in fs/locks.c and misc file system and DLM code. These stubs will
allow all users to switch to the new interface before the implementation
is changed to a spinlock.
Acked-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Sage Weil <sage@newdream.net>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
As part of adding support for OCFS2 to mount huge volumes, we need to
check that the sector_t and page cache of the system are capable of
addressing the entire volume.
An identical check already appears in ext3 and ext4. This patch moves
the addressability check into its own function in fs/libfs.c and
modifies ext3 and ext4 to invoke it.
[Edited to -EINVAL instead of BUG_ON() for bad blocksize_bits -- Joel]
Signed-off-by: Patrick LoPresti <lopresti@gmail.com>
Cc: linux-ext4@vger.kernel.org
Acked-by: Andreas Dilger <adilger@dilger.ca>
Signed-off-by: Joel Becker <joel.becker@oracle.com>
Remove support for barriers on discards, which is unused now. Also
remove the DISCARD_NOBARRIER I/O type in favour of just setting the
rw flags up locally in blkdev_issue_discard.
tj: Also remove DISCARD_SECURE and use REQ_SECURE directly.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Acked-by: Mike Snitzer <snitzer@redhat.com>
Signed-off-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Jens Axboe <jaxboe@fusionio.com>
Now that the backend conversion is complete, export sequenced
FLUSH/FUA capability through REQ_FLUSH/FUA flags. REQ_FLUSH means the
device cache should be flushed before executing the request. REQ_FUA
means that the data in the request should be on non-volatile media on
completion.
Block layer will choose the correct way of implementing the semantics
and execute it. The request may be passed to the device directly if
the device can handle it; otherwise, it will be sequenced using one or
more proxy requests. Devices will never see REQ_FLUSH and/or FUA
which it doesn't support.
Also, unlike the original REQ_HARDBARRIER, REQ_FLUSH/FUA requests are
never failed with -EOPNOTSUPP. If the underlying device doesn't
support FLUSH/FUA, the block layer simply make those noop. IOW, it no
longer distinguishes between writeback cache which doesn't support
cache flush and writethrough/no cache. Devices which have WB cache
w/o flush are very difficult to come by these days and there's nothing
much we can do anyway, so it doesn't make sense to require everyone to
implement -EOPNOTSUPP handling. This will simplify filesystems and
block drivers as they can drop -EOPNOTSUPP retry logic for barriers.
* QUEUE_ORDERED_* are removed and QUEUE_FSEQ_* are moved into
blk-flush.c.
* REQ_FLUSH w/o data can also be directly passed to drivers without
sequencing but some drivers assume that zero length requests don't
have rq->bio which isn't true for these requests requiring the use
of proxy requests.
* REQ_COMMON_MASK now includes REQ_FLUSH | REQ_FUA so that they are
copied from bio to request.
* WRITE_BARRIER is marked deprecated and WRITE_FLUSH, WRITE_FUA and
WRITE_FLUSH_FUA are added.
Signed-off-by: Tejun Heo <tj@kernel.org>
Cc: Christoph Hellwig <hch@infradead.org>
Signed-off-by: Jens Axboe <jaxboe@fusionio.com>
This adds annotations for RCU operations in core kernel components
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Cc: Al Viro <viro@ZenIV.linux.org.uk>
Cc: Jens Axboe <jens.axboe@oracle.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Josh Triplett <josh@joshtriplett.org>
fs: scale files_lock
Improve scalability of files_lock by adding per-cpu, per-sb files lists,
protected with an lglock. The lglock provides fast access to the per-cpu lists
to add and remove files. It also provides a snapshot of all the per-cpu lists
(although this is very slow).
One difficulty with this approach is that a file can be removed from the list
by another CPU. We must track which per-cpu list the file is on with a new
variale in the file struct (packed into a hole on 64-bit archs). Scalability
could suffer if files are frequently removed from different cpu's list.
However loads with frequent removal of files imply short interval between
adding and removing the files, and the scheduler attempts to avoid moving
processes too far away. Also, even in the case of cross-CPU removal, the
hardware has much more opportunity to parallelise cacheline transfers with N
cachelines than with 1.
A worst-case test of 1 CPU allocating files subsequently being freed by N CPUs
degenerates to contending on a single lock, which is no worse than before. When
more than one CPU are allocating files, even if they are always freed by
different CPUs, there will be more parallelism than the single-lock case.
Testing results:
On a 2 socket, 8 core opteron, I measure the number of times the lock is taken
to remove the file, the number of times it is removed by the same CPU that
added it, and the number of times it is removed by the same node that added it.
Booting: locks= 25049 cpu-hits= 23174 (92.5%) node-hits= 23945 (95.6%)
kbuild -j16 locks=2281913 cpu-hits=2208126 (96.8%) node-hits=2252674 (98.7%)
dbench 64 locks=4306582 cpu-hits=4287247 (99.6%) node-hits=4299527 (99.8%)
So a file is removed from the same CPU it was added by over 90% of the time.
It remains within the same node 95% of the time.
Tim Chen ran some numbers for a 64 thread Nehalem system performing a compile.
throughput
2.6.34-rc2 24.5
+patch 24.9
us sys idle IO wait (in %)
2.6.34-rc2 51.25 28.25 17.25 3.25
+patch 53.75 18.5 19 8.75
So significantly less CPU time spent in kernel code, higher idle time and
slightly higher throughput.
Single threaded performance difference was within the noise of microbenchmarks.
That is not to say penalty does not exist, the code is larger and more memory
accesses required so it will be slightly slower.
Cc: linux-kernel@vger.kernel.org
Cc: Tim Chen <tim.c.chen@linux.intel.com>
Cc: Andi Kleen <ak@linux.intel.com>
Signed-off-by: Nick Piggin <npiggin@kernel.dk>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
tty: fix fu_list abuse
tty code abuses fu_list, which causes a bug in remount,ro handling.
If a tty device node is opened on a filesystem, then the last link to the inode
removed, the filesystem will be allowed to be remounted readonly. This is
because fs_may_remount_ro does not find the 0 link tty inode on the file sb
list (because the tty code incorrectly removed it to use for its own purpose).
This can result in a filesystem with errors after it is marked "clean".
Taking idea from Christoph's initial patch, allocate a tty private struct
at file->private_data and put our required list fields in there, linking
file and tty. This makes tty nodes behave the same way as other device nodes
and avoid meddling with the vfs, and avoids this bug.
The error handling is not trivial in the tty code, so for this bugfix, I take
the simple approach of using __GFP_NOFAIL and don't worry about memory errors.
This is not a problem because our allocator doesn't fail small allocs as a rule
anyway. So proper error handling is left as an exercise for tty hackers.
[ Arguably filesystem's device inode would ideally be divorced from the
driver's pseudo inode when it is opened, but in practice it's not clear whether
that will ever be worth implementing. ]
Cc: linux-kernel@vger.kernel.org
Cc: Christoph Hellwig <hch@infradead.org>
Cc: Alan Cox <alan@lxorguk.ukuu.org.uk>
Cc: Greg Kroah-Hartman <gregkh@suse.de>
Signed-off-by: Nick Piggin <npiggin@kernel.dk>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
fs: cleanup files_lock locking
Lock tty_files with a new spinlock, tty_files_lock; provide helpers to
manipulate the per-sb files list; unexport the files_lock spinlock.
Cc: linux-kernel@vger.kernel.org
Cc: Christoph Hellwig <hch@infradead.org>
Cc: Alan Cox <alan@lxorguk.ukuu.org.uk>
Acked-by: Andi Kleen <ak@linux.intel.com>
Acked-by: Greg Kroah-Hartman <gregkh@suse.de>
Signed-off-by: Nick Piggin <npiggin@kernel.dk>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
These flags aren't real I/O types, but tell ll_rw_block to always
lock the buffer instead of giving up on a failed trylock.
Instead add a new write_dirty_buffer helper that implements this semantic
and use it from the existing SWRITE* callers. Note that the ll_rw_block
code had a bug where it didn't promote WRITE_SYNC_PLUG properly, which
this patch fixes.
In the ufs code clean up the helper that used to call ll_rw_block
to mirror sync_dirty_buffer, which is the function it implements for
compound buffers.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Nick Piggin