Ensure pages are uptodate after returning from read_cache_page, which allows
us to cut out most of the filesystem-internal PageUptodate calls.
I didn't have a great look down the call chains, but this appears to fixes 7
possible use-before uptodate in hfs, 2 in hfsplus, 1 in jfs, a few in
ecryptfs, 1 in jffs2, and a possible cleared data overwritten with readpage in
block2mtd. All depending on whether the filler is async and/or can return
with a !uptodate page.
Signed-off-by: Nick Piggin <npiggin@suse.de>
Cc: Hugh Dickins <hugh@veritas.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Try running this script in an NFS mounted directory (Client relatively
recent - 2.6.18 has the problem as does 2.6.20).
------------------------------------------------------
#!/bin/bash
#
# This script will produce the following errormessage from tar:
#
# tar: newdir/innerdir/innerfile: file changed as we read it
# create dirs
rm -rf nfstest
mkdir -p nfstest/dir/innerdir
# create files (should not be empty)
echo "Hello World!" >nfstest/dir/file
echo "Hello World!" >nfstest/dir/innerdir/innerfile
# problem only happens if we sleep before chmod
sleep 1
# change file modes
chmod -R a+r nfstest
# rename dir
mv nfstest/dir nfstest/newdir
# tar it
tar -cf nfstest/nfstest.tar -C nfstest newdir
# restore old dir name
mv nfstest/newdir nfstest/dir
--------------------------------------------------------
What happens:
The 'chmod -R' does a readdir_plus in each directory and the results
get cached in the page cache. It then updates the ctime on each file
by one second. When this happens, the post-op attributes are used to
update the ctime stored on the client to match the value in the kernel.
The 'mv' calls shrink_dcache_parent on the directory tree which
flushes all the dentries (so a new lookup will be required) but
doesn't flush the inodes or pagecache.
The 'tar' does a readdir on each directory, but (in the case of
'innerdir' at least) satisfies it from the pagecache and uses the
READDIRPLUS data to update all the inodes. In the case of
'innerdir/innerfile', the ctime is out of date.
'tar' then calls 'lstat' on innerdir/innerfile getting an old ctime.
It then opens the file (triggering a GETATTR), reads the content, and
then calls fstat to see if anything has changed. It finds that ctime
has changed and so complains.
The problem seems to be that the cache readdirplus info is kept around
for too long.
My patch below discards pagecache data for directories when
dentry_iput is called on them. This effectively removes the symptom
which convinces me that I correctly understand the problem. However
I'm not convinced that is a proper solution, as there could easily be
other races that trigger the same problem without being affected by
this 'fix'.
One possibility would be to require that readdirplus pagecache data be
only used *once* to instantiate an inode. Somehow it should then be
invalidated so that if the dentry subsequently disappears, it will
cause a new request to the server to fill in the stat data.
Another possibility is to compare the cache_change_attribute on the
inode with something similar for the readdirplus info and reject the
info from readdirplus if it is too old.
I haven't tried to implement these and would value other opinions
before I do.
Thanks,
NeilBrown
Signed-off-by: Neil Brown <neilb@suse.de>
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
Don't use uninitialsed value for fattr->time_start in readdirplus results.
The 'fattr' structure filled in by nfs3_decode_direct does not get a
value for ->time_start set.
Thus if an entry is for an inode that we already have in cache,
when nfs_readdir_lookup calls nfs_fhget, it will call nfs_refresh_inode
and may update the inode with out-of-date information.
Directories are read a page at a time, so each page could have a
different timestamp that "should" be used to set the time_start for
the fattr for info in that page. However storing the timestamp per
page is awkward. (We could stick in the first 4 bytes and only read 4092
bytes, but that is a bigger code change than I am interested it).
This patch ignores the readdir_plus attributes if a readdir finds the
information already in cache, and otherwise sets ->time_start to the time
the readdir request was sent to the server.
It might be nice to store - in the directory inode - the time stamp for
the earliest readdir request that is still in the page cache, so that we
don't ignore attribute data that we don't have to. This patch doesn't do
that.
Signed-off-by: Neil Brown <neilb@suse.de>
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
It looks like nfs_setattr() and nfs_rename() also need to test whether the
target is a regular file before calling nfs_wb_all()...
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Many struct inode_operations in the kernel can be "const". Marking them const
moves these to the .rodata section, which avoids false sharing with potential
dirty data. In addition it'll catch accidental writes at compile time to
these shared resources.
Signed-off-by: Arjan van de Ven <arjan@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
nfs_lookup_revalidate and friends are not serialised, so it is currently
quite possible for the dentry to be revalidated, and then have the
updated verifier replaced with an older value by another process.
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
If the fileid of the cached dentry fails to match that returned by
the readdir call, then we should also d_drop. Try to take into account the
fact that on NFSv4, readdir may return the "mounted_on_fileid" by looking
for submounts.
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
Make sure that nfs_readdir_lookup() handles negative dentries correctly.
If d_lookup() returns a negative dentry, then we need to d_drop() that
since readdir shows that it should be positive.
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
When a file is being scheduled for deletion by means of the sillyrename
mechanism, it makes sense to start out writeback of the dirty data as
soon as possible in order to ensure that the delete can occur. Examples of
cases where this is an issue include "rm -rf", which will busy-wait until
the file is closed, and the sillyrename completes.
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
Prevent the call to invalidate_inode_pages2() from racing with file writes
by taking the inode->i_mutex across the page cache flush and invalidate.
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Change all the uses of f_{dentry,vfsmnt} to f_path.{dentry,mnt} in the nfs
client code.
Signed-off-by: Josef "Jeff" Sipek <jsipek@cs.sunysb.edu>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
If someone has renamed a directory on the server, triggering the d_move
code in d_materialise_unique(), then we need to invalidate the cached
directory information in the source parent directory.
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
Cc: Miklos Szeredi <miklos@szeredi.hu>
Cc: Maneesh Soni <maneesh@in.ibm.com>
Cc: Dipankar Sarma <dipankar@in.ibm.com>
Cc: Neil Brown <neilb@cse.unsw.edu.au>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Christoph Hellwig <hch@lst.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
If the caller tries to instantiate a directory using an inode that already
has a dentry alias, then we attempt to rename the existing dentry instead
of instantiating a new one. Fail with an ELOOP error if the rename would
affect one of our parent directories.
This behaviour is needed in order to avoid issues such as
http://bugzilla.kernel.org/show_bug.cgi?id=7178
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
Cc: Miklos Szeredi <miklos@szeredi.hu>
Cc: Maneesh Soni <maneesh@in.ibm.com>
Cc: Dipankar Sarma <dipankar@in.ibm.com>
Cc: Neil Brown <neilb@cse.unsw.edu.au>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Christoph Hellwig <hch@lst.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
on-the-wire data is big-endian
[in large part pulled from Alexey's patch]
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Acked-by: Trond Myklebust <trond.myklebust@fys.uio.no>
Acked-by: Neil Brown <neilb@suse.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
The original code confused a zero return code from pagevec_add() as success.
Test plan:
None.
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
If invalidate_inode_pages2() fails, then it should in principle just be
because the current process was signalled. In that case, we just want to
ensure that the inode's page cache remains marked as invalid.
Also add a helper to allow the O_DIRECT code to simply mark the page cache as
invalid once it is finished writing, instead of calling
invalidate_inode_pages2() itself.
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Some filesystems, instead of simply decrementing i_nlink, simply zero it
during an unlink operation. We need to catch these in addition to the
decrement operations.
Signed-off-by: Dave Hansen <haveblue@us.ibm.com>
Acked-by: Christoph Hellwig <hch@lst.de>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
When a filesystem decrements i_nlink to zero, it means that a write must be
performed in order to drop the inode from the filesystem.
We're shortly going to have keep filesystems from being remounted r/o between
the time that this i_nlink decrement and that write occurs.
So, add a little helper function to do the decrements. We'll tie into it in a
bit to note when i_nlink hits zero.
Signed-off-by: Dave Hansen <haveblue@us.ibm.com>
Acked-by: Christoph Hellwig <hch@lst.de>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Some file systems want to manually d_move() the dentries involved in a
rename. We can do this by making use of the FS_ODD_RENAME flag if we just
have nfs_rename() unconditionally do the d_move(). While there, we rename
the flag to be more descriptive.
OCFS2 uses this to protect that part of the rename operation with a cluster
lock.
Signed-off-by: Mark Fasheh <mark.fasheh@oracle.com>
Cc: Trond Myklebust <trond.myklebust@fys.uio.no>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Christoph Hellwig <hch@lst.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
If the open intents tell us that a given lookup is going to result in a,
exclusive create, we currently optimize away the lookup call itself. The
reason is that the lookup would not be atomic with the create RPC call, so
why do it in the first place?
A problem occurs, however, if the VFS aborts the exclusive create operation
after the lookup, but before the call to create the file/directory: in this
case we will end up with a hashed negative dentry in the dcache that has
never been looked up.
Fix this by only actually hashing the dentry once the create operation has
been successfully completed.
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
Now that we have a copy of the symlink path in the page cache, we can pass
a struct page down to the XDR routines instead of a string buffer.
Test plan:
Connectathon, all NFS versions.
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
Currently the NFS client does not cache symlinks it creates. They get
cached only when the NFS client reads them back from the server.
Copy the symlink into the page cache before sending it.
Test plan:
Connectathon, all NFS versions.
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
If the LOOKUP or GETATTR in nfs_instantiate fail, nfs_instantiate will do a
d_drop before returning. But some callers already do a d_drop in the case
of an error return. Make certain we do only one d_drop in all error paths.
This issue was introduced because over time, the symlink proc API diverged
slightly from the create/mkdir/mknod proc API. To prevent other coding
mistakes of this type, change the symlink proc API to be more like
create/mkdir/mknod and move the nfs_instantiate call into the symlink proc
routines so it is used in exactly the same way for create, mkdir, mknod,
and symlink.
Test plan:
Connectathon, all versions of NFS.
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
In the early days of NFS, there was no duplicate reply cache on the server.
Thus retransmitted non-idempotent requests often found that the request had
already completed on the server. To avoid passing an unanticipated return
code to unsuspecting applications, NFS clients would often shunt error
codes that implied the request had been retried but already completed.
Thanks to NFS over TCP, duplicate reply caches on the server, and network
performance and reliability improvements, it is safe to remove such checks.
Test plan:
None.
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
The attached patch makes NFS share superblocks between mounts from the same
server and FSID over the same protocol.
It does this by creating each superblock with a false root and returning the
real root dentry in the vfsmount presented by get_sb(). The root dentry set
starts off as an anonymous dentry if we don't already have the dentry for its
inode, otherwise it simply returns the dentry we already have.
We may thus end up with several trees of dentries in the superblock, and if at
some later point one of anonymous tree roots is discovered by normal filesystem
activity to be located in another tree within the superblock, the anonymous
root is named and materialises attached to the second tree at the appropriate
point.
Why do it this way? Why not pass an extra argument to the mount() syscall to
indicate the subpath and then pathwalk from the server root to the desired
directory? You can't guarantee this will work for two reasons:
(1) The root and intervening nodes may not be accessible to the client.
With NFS2 and NFS3, for instance, mountd is called on the server to get
the filehandle for the tip of a path. mountd won't give us handles for
anything we don't have permission to access, and so we can't set up NFS
inodes for such nodes, and so can't easily set up dentries (we'd have to
have ghost inodes or something).
With this patch we don't actually create dentries until we get handles
from the server that we can use to set up their inodes, and we don't
actually bind them into the tree until we know for sure where they go.
(2) Inaccessible symbolic links.
If we're asked to mount two exports from the server, eg:
mount warthog:/warthog/aaa/xxx /mmm
mount warthog:/warthog/bbb/yyy /nnn
We may not be able to access anything nearer the root than xxx and yyy,
but we may find out later that /mmm/www/yyy, say, is actually the same
directory as the one mounted on /nnn. What we might then find out, for
example, is that /warthog/bbb was actually a symbolic link to
/warthog/aaa/xxx/www, but we can't actually determine that by talking to
the server until /warthog is made available by NFS.
This would lead to having constructed an errneous dentry tree which we
can't easily fix. We can end up with a dentry marked as a directory when
it should actually be a symlink, or we could end up with an apparently
hardlinked directory.
With this patch we need not make assumptions about the type of a dentry
for which we can't retrieve information, nor need we assume we know its
place in the grand scheme of things until we actually see that place.
This patch reduces the possibility of aliasing in the inode and page caches for
inodes that may be accessed by more than one NFS export. It also reduces the
number of superblocks required for NFS where there are many NFS exports being
used from a server (home directory server + autofs for example).
This in turn makes it simpler to do local caching of network filesystems, as it
can then be guaranteed that there won't be links from multiple inodes in
separate superblocks to the same cache file.
Obviously, cache aliasing between different levels of NFS protocol could still
be a problem, but at least that gives us another key to use when indexing the
cache.
This patch makes the following changes:
(1) The server record construction/destruction has been abstracted out into
its own set of functions to make things easier to get right. These have
been moved into fs/nfs/client.c.
All the code in fs/nfs/client.c has to do with the management of
connections to servers, and doesn't touch superblocks in any way; the
remaining code in fs/nfs/super.c has to do with VFS superblock management.
(2) The sequence of events undertaken by NFS mount is now reordered:
(a) A volume representation (struct nfs_server) is allocated.
(b) A server representation (struct nfs_client) is acquired. This may be
allocated or shared, and is keyed on server address, port and NFS
version.
(c) If allocated, the client representation is initialised. The state
member variable of nfs_client is used to prevent a race during
initialisation from two mounts.
(d) For NFS4 a simple pathwalk is performed, walking from FH to FH to find
the root filehandle for the mount (fs/nfs/getroot.c). For NFS2/3 we
are given the root FH in advance.
(e) The volume FSID is probed for on the root FH.
(f) The volume representation is initialised from the FSINFO record
retrieved on the root FH.
(g) sget() is called to acquire a superblock. This may be allocated or
shared, keyed on client pointer and FSID.
(h) If allocated, the superblock is initialised.
(i) If the superblock is shared, then the new nfs_server record is
discarded.
(j) The root dentry for this mount is looked up from the root FH.
(k) The root dentry for this mount is assigned to the vfsmount.
(3) nfs_readdir_lookup() creates dentries for each of the entries readdir()
returns; this function now attaches disconnected trees from alternate
roots that happen to be discovered attached to a directory being read (in
the same way nfs_lookup() is made to do for lookup ops).
The new d_materialise_unique() function is now used to do this, thus
permitting the whole thing to be done under one set of locks, and thus
avoiding any race between mount and lookup operations on the same
directory.
(4) The client management code uses a new debug facility: NFSDBG_CLIENT which
is set by echoing 1024 to /proc/net/sunrpc/nfs_debug.
(5) Clone mounts are now called xdev mounts.
(6) Use the dentry passed to the statfs() op as the handle for retrieving fs
statistics rather than the root dentry of the superblock (which is now a
dummy).
Signed-Off-By: David Howells <dhowells@redhat.com>
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
Move the rpc_ops from the nfs_server struct to the nfs_client struct as they're
common to all server records of a particular NFS protocol version.
Signed-Off-By: David Howells <dhowells@redhat.com>
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
A pinned inode may in theory end up filling memory with cached ACCESS
calls. This patch ensures that the VM may shrink away the cache in these
particular cases.
The shrinker works by iterating through the list of inodes on the global
nfs_access_lru_list, and removing the least recently used access
cache entry until it is done (or until the entire cache is empty).
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
The current access cache only allows one entry at a time to be cached for each
inode. Add a per-inode red-black tree in order to allow more than one to
be cached at a time.
Should significantly cut down the time spent in path traversal for shared
directories such as ${PATH}, /usr/share, etc.
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
In the case when compiling via a symlink tree, we want to ensure that the
close-to-open GETATTR call is applied only to the final file, and not to
the symlink.
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
Separate out the function of revalidating the inode metadata, and
revalidating the mapping. The former may be called by lookup(),
and only really needs to check that permissions, ctime, etc haven't changed
whereas the latter needs only done when we want to read data from the page
cache, and may need to sync and then invalidate the mapping.
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
Local variable res was initialized to 0 - no check needed here.
Signed-off-by: Carsten Otte <cotte@de.ibm.com>
Cc: Trond Myklebust <trond.myklebust@fys.uio.no>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
This is a conversion to make the various file_operations structs in fs/
const. Basically a regexp job, with a few manual fixups
The goal is both to increase correctness (harder to accidentally write to
shared datastructures) and reducing the false sharing of cachelines with
things that get dirty in .data (while .rodata is nicely read only and thus
cache clean)
Signed-off-by: Arjan van de Ven <arjan@infradead.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Currently it returns NULL, which usually gets interpreted as ENOMEM. In
fact it can mean a host of issues.
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
Reuse NFSDBG_DIRCACHE and NFSDBG_LOOKUPCACHE to provide additional
diagnostic messages that trace the operation of the NFS client's
directory behavior. A few new messages are now generated when NFSDBG_VFS
is active, as well, to trace normal VFS activity. This compromise
provides better trace debugging for those who use pre-built kernels,
without adding a lot of extra noise to the standard debug settings.
Test-plan:
Enable NFS trace debugging with flags 1, 2, or 4. You should be able to
see different types of trace messages with each flag setting.
Signed-off-by: Chuck Lever <cel@netapp.com>
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
Invoke the byte and event counter macros where we want to count bytes and
events.
Clean-up: fix a possible NULL dereference in nfs_lock, and simplify
nfs_file_open.
Test-plan:
fsx and iozone on UP and SMP systems, with and without pre-emption. Watch
for memory overwrite bugs, and performance loss (significantly more CPU
required per op).
Signed-off-by: Chuck Lever <cel@netapp.com>
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
This patch converts the inode semaphore to a mutex. I have tested it on
XFS and compiled as much as one can consider on an ia64. Anyway your
luck with it might be different.
Modified-by: Ingo Molnar <mingo@elte.hu>
(finished the conversion)
Signed-off-by: Jes Sorensen <jes@sgi.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
I missed this one... Any form of rename will result in a delegation
recall, so it is more efficient to return the one we hold before
trying the rename.
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
nfs_lookup() used to consult a lookup cache before trying an actual wire
lookup operation. The lookup cache would be invalid, of course, if the
parent directory's mtime had changed, so nfs_lookup performed an inode
revalidation on the parent.
Since nfs_lookup() doesn't use a cache anymore, the revalidation is no
longer necessary. There are cases where it will generate a lot of
unnecessary GETATTR traffic.
See http://bugzilla.linux-nfs.org/show_bug.cgi?id=9
Test-plan:
Use lndir and "rm -rf" and watch for excess GETATTR traffic or application
level errors.
Signed-off-by: Chuck Lever <cel@netapp.com>
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
If someone tries to rename a directory onto an empty directory, we
currently fail and return EBUSY.
This patch ensures that we try the rename if both source and target
are directories, and that we fail with a correct error of EISDIR if
the source is not a directory.
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
Make NFSv4 return the fully initialized file pointer with the
stateid that it created in the lookup w/intent.
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
Fixes a condition whereby the kernel is returning the non-POSIX error
EBADCOOKIE to userspace.
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
When doing a rename on top of an existing file that is not in use,
the inode of the overwritten file will remain in the icache.
The fix is to decrement i_nlink of the overwritten inode, like we
do for unlink, rmdir etc already.
Problem diagnosed by Olaf Kirch. This patch is a slight variation
on his fix.
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
Nick Piggin