Quoting Linus (3 years ago, FUSE inclusion discussions):
"User-space filesystems are hard to get right. I'd claim that they
are almost impossible, unless you limit them somehow (shared
writable mappings are the nastiest part - if you don't have those,
you can reasonably limit your problems by limiting the number of
dirty pages you accept through normal "write()" calls)."
Instead of attempting the impossible, I've just waited for the dirty page
accounting infrastructure to materialize (thanks to Peter Zijlstra and
others). This nicely solved the biggest problem: limiting the number of pages
used for write caching.
Some small details remained, however, which this largish patch attempts to
address. It provides a page writeback implementation for fuse, which is
completely safe against VM related deadlocks. Performance may not be very
good for certain usage patterns, but generally it should be acceptable.
It has been tested extensively with fsx-linux and bash-shared-mapping.
Fuse page writeback design
--------------------------
fuse_writepage() allocates a new temporary page with GFP_NOFS|__GFP_HIGHMEM.
It copies the contents of the original page, and queues a WRITE request to the
userspace filesystem using this temp page.
The writeback is finished instantly from the MM's point of view: the page is
removed from the radix trees, and the PageDirty and PageWriteback flags are
cleared.
For the duration of the actual write, the NR_WRITEBACK_TEMP counter is
incremented. The per-bdi writeback count is not decremented until the actual
write completes.
On dirtying the page, fuse waits for a previous write to finish before
proceeding. This makes sure, there can only be one temporary page used at a
time for one cached page.
This approach is wasteful in both memory and CPU bandwidth, so why is this
complication needed?
The basic problem is that there can be no guarantee about the time in which
the userspace filesystem will complete a write. It may be buggy or even
malicious, and fail to complete WRITE requests. We don't want unrelated parts
of the system to grind to a halt in such cases.
Also a filesystem may need additional resources (particularly memory) to
complete a WRITE request. There's a great danger of a deadlock if that
allocation may wait for the writepage to finish.
Currently there are several cases where the kernel can block on page
writeback:
- allocation order is larger than PAGE_ALLOC_COSTLY_ORDER
- page migration
- throttle_vm_writeout (through NR_WRITEBACK)
- sync(2)
Of course in some cases (fsync, msync) we explicitly want to allow blocking.
So for these cases new code has to be added to fuse, since the VM is not
tracking writeback pages for us any more.
As an extra safetly measure, the maximum dirty ratio allocated to a single
fuse filesystem is set to 1% by default. This way one (or several) buggy or
malicious fuse filesystems cannot slow down the rest of the system by hogging
dirty memory.
With appropriate privileges, this limit can be raised through
'/sys/class/bdi/<bdi>/max_ratio'.
Signed-off-by: Miklos Szeredi <mszeredi@suse.cz>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Register FUSE's backing_dev_info under sysfs with the name "fuse-MAJOR:MINOR"
Make the fuse control filesystem use s_dev instead of a fuse specific ID.
This makes it easier to match directories under /sys/fs/fuse/connections/ with
directories under /sys/class/bdi, and with actual mounts.
Signed-off-by: Miklos Szeredi <mszeredi@suse.cz>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
I added a nasty local variable shadowing bug to fuse in 2.6.24, with the
result, that the 'default_permissions' mount option is basically ignored.
How did this happen?
- old err declaration in inner scope
- new err getting declared in outer scope
- 'return err' from inner scope getting removed
- old declaration not being noticed
-Wshadow would have saved us, but it doesn't seem practical for
the kernel :(
More testing would have also saved us :((
Signed-off-by: Miklos Szeredi <mszeredi@suse.cz>
Cc: <stable@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Stop the FUSE filesystem from using read_inode(), which it doesn't use anyway.
Signed-off-by: David Howells <dhowells@redhat.com>
Cc: Miklos Szeredi <miklos@szeredi.hu>
Acked-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Libfuse basically creates a new thread for each new request. This is fine for
synchronous requests, which are naturally limited. However background
requests (especially writepage) can cause a thread creation storm.
To avoid this, limit the number of background requests available to userspace.
This is done by introducing another queue for background requests, and a
counter for the number of "active" requests, which are currently available for
userspace.
Signed-off-by: Miklos Szeredi <mszeredi@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Move the fields 'dentry' and 'vfsmount' into the request specific union, since
these are only used for the RELEASE request.
Signed-off-by: Miklos Szeredi <mszeredi@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Invalidate attributes on create, since st_ctime is updated. Reported by
Szabolcs Szakacsits.
Signed-off-by: Miklos Szeredi <mszeredi@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
There is no need for kobject_unregister() anymore, thanks to Kay's
kobject cleanup changes, so replace all instances of it with
kobject_put().
Cc: Kay Sievers <kay.sievers@vrfy.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
This also renames fs_subsys to fs_kobj to catch all current users with a
build error instead of a build warning which can easily be missed.
Cc: Kay Sievers <kay.sievers@vrfy.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
We don't need a kset here, a simple kobject will do just fine, so
dynamically create the kobject and use it.
Cc: Kay Sievers <kay.sievers@vrfy.org>
Cc: Miklos Szeredi <miklos@szeredi.hu>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
We don't need a "default" ktype for a kset. We should set this
explicitly every time for each kset. This change is needed so that we
can make ksets dynamic, and cleans up one of the odd, undocumented
assumption that the kset/kobject/ktype model has.
This patch is based on a lot of help from Kay Sievers.
Nasty bug in the block code was found by Dave Young
<hidave.darkstar@gmail.com>
Cc: Kay Sievers <kay.sievers@vrfy.org>
Cc: Dave Young <hidave.darkstar@gmail.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
Invalidate attributes on rename, since some filesystems may update
st_ctime. Reported by Szabolcs Szakacsits
Signed-off-by: Miklos Szeredi <mszeredi@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
I found problems accessing (executing) previously existing files, until
I did chmod on them (or setattr).
If the fi->attr_version is not initialized, then it could be
larger than fc->attr_version until a setattr is executed, and as a
result the inode attributes would never be set.
Signed-off-by: Miklos Szeredi <mszeredi@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
FUSE_FILE_OPS is meant to signal that the kernel will send the open file to to
the userspace filesystem for operations on open files, so that sillyrenaming
unlinked files becomes unnecessary.
However this needs VFS changes, which won't make it into 2.6.24.
Signed-off-by: Miklos Szeredi <mszeredi@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Some open flags (O_APPEND, O_DIRECT) can be changed with fcntl(F_SETFL, ...)
after open, but fuse currently only sends the flags to userspace in open.
To make it possible to correcly handle changing flags, send the
current value to userspace in each read and write.
Signed-off-by: Miklos Szeredi <mszeredi@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Currently reading a fuse file will stop at cached i_size and return
EOF, even though the file might have grown since the attributes were
last updated.
So detect if trying to read past EOF, and refresh the attributes
before continuing with the read.
Thanks to mpb for the report.
Signed-off-by: Miklos Szeredi <mszeredi@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
There are cases when the filesystem will be passed the buffer from a single
read or write call, namely:
1) in 'direct-io' mode (not O_DIRECT), read/write requests don't go
through the page cache, but go directly to the userspace fs
2) currently buffered writes are done with single page requests, but
if Nick's ->perform_write() patch goes it, it will be possible to
do larger write requests. But only if the original write() was
also bigger than a page.
In these cases the filesystem might want to give a hint to the app
about the optimal I/O size.
Allow the userspace filesystem to supply a blksize value to be returned by
stat() and friends. If the field is zero, it defaults to the old
PAGE_CACHE_SIZE value.
Signed-off-by: Miklos Szeredi <mszeredi@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
For mandatory locking the userspace filesystem needs to know the lock
ownership for read, write and truncate operations.
This patch adds the necessary fields to the protocol.
Signed-off-by: Miklos Szeredi <mszeredi@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
This patch adds a new helper function fuse_write_fill() which makes it
possible to send WRITE requests asynchronously.
A new flag for WRITE requests is also added which indicates that this a write
from the page cache, and not a "normal" file write.
This patch is in preparation for writable mmap support.
Signed-off-by: Miklos Szeredi <mszeredi@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Each WRITE request must carry a valid file descriptor. When a page is written
back from a memory mapping, the file through which the page was dirtied is not
available, so a new mechananism is needed to find a suitable file in
->writepage(s).
A list of fuse_files is added to fuse_inode. The file is removed from the
list in fuse_release().
This patch is in preparation for writable mmap support.
Signed-off-by: Miklos Szeredi <mszeredi@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
It is trivial to add support for flock(2) semantics to the existing protocol,
by setting the lock owner field to the file pointer, and passing a new
FUSE_LK_FLOCK flag with the locking request.
Signed-off-by: Miklos Szeredi <mszeredi@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
This patch allows fuse filesystems to implement open(..., O_TRUNC) as a single
request, instead of separate truncate and open requests.
Signed-off-by: Miklos Szeredi <mszeredi@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Add two new flags for setattr: FATTR_ATIME_NOW and FATTR_MTIME_NOW. These
mean, that atime or mtime should be changed to the current time.
Also it is now possible to update atime or mtime individually, not just
together.
Signed-off-by: Miklos Szeredi <mszeredi@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Clean up supplying open file to the setattr operation. In addition to being a
cleanup it prepares for the changes in the way the open file is passed to the
setattr method.
Signed-off-by: Miklos Szeredi <mszeredi@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Add necessary protocol changes for supplying a file handle with the getattr
operation. Step the API version to 7.9.
This patch doesn't actually supply the file handle, because that needs some
kind of VFS support, which we haven't yet been able to agree upon.
[akpm@linux-foundation.org: coding-style fixes]
Signed-off-by: Miklos Szeredi <mszeredi@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Getattr and lookup operations can be running in parallel to attribute changing
operations, such as write and setattr.
This means, that if for example getattr was slower than a write, the cached
size attribute could be set to a stale value.
To prevent this race, introduce a per-filesystem attribute version counter.
This counter is incremented whenever cached attributes are modified, and the
incremented value stored in the inode.
Before storing new attributes in the cache, getattr and lookup check, using
the version number, whether the attributes have been modified during the
request's lifetime. If so, the returned attributes are not cached, because
they might be stale.
Thanks to Jakub Bogusz for the bug report and test program.
[akpm@linux-foundation.org: coding-style fixes]
Signed-off-by: Miklos Szeredi <mszeredi@suse.cz>
Cc: Jakub Bogusz <jakub.bogusz@gemius.pl>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
The following operation didn't check if sending the request was allowed:
setattr
listxattr
statfs
Some other operations don't explicitly do the check, but VFS calls
->permission() which checks this.
Signed-off-by: Miklos Szeredi <mszeredi@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Define a new function fuse_refresh_attributes() that conditionally refreshes
the attributes based on the validity timeout.
In fuse_permission() only refresh the attributes for checking the execute bits
if necessary.
Signed-off-by: Miklos Szeredi <mszeredi@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Don't return -ENOENT for a read() on the fuse device when the request was
aborted. Instead return -ENODEV, meaning the filesystem has been
force-umounted or aborted.
Previously ENOENT meant that the request was interrupted, but now the
'aborted' flag is not set in case of interrupts.
Signed-off-by: Miklos Szeredi <mszeredi@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Don't set 'aborted' flag on a request if it's interrupted. We have to wait
for the answer anyway, and this would only a very little time while copying
the reply.
This means, that write() on the fuse device will not return -ENOENT during
normal operation, only if the filesystem is aborted by a forced umount or
through the fusectl interface.
This could simplify userspace code somewhat when backward compatibility with
earlier kernel versions is not required.
Signed-off-by: Miklos Szeredi <mszeredi@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Move dput/mntput pair from request_end() to fuse_release_end(), because
there's no other place they are used.
Signed-off-by: Miklos Szeredi <mszeredi@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
The VFS checks sticky bits on the parent directory even if the filesystem
defines it's own ->permission(). In some situations (sshfs, mountlo, etc) the
user does have permission to delete a file even if the attribute based
checking would not allow it.
So work around this by storing the permission bits separately and returning
them in stat(), but cutting the permission bits off from inode->i_mode.
This is slightly hackish, but it's probably not worth it to add new
infrastructure in VFS and a slight performance penalty for all filesystems,
just for the sake of fuse.
[Jan Engelhardt] cosmetic fixes
Signed-off-by: Miklos Szeredi <mszeredi@suse.cz>
Cc: Jan Engelhardt <jengelh@linux01.gwdg.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
fuse_permission() didn't refresh inode attributes before using them, even if
the validity has already expired.
Thanks to Junjiro Okajima for spotting this.
Also remove some old code to unconditionally refresh the attributes on the
root inode.
Signed-off-by: Miklos Szeredi <mszeredi@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Aufs seems to depend on a positive i_nlink value. So fill in a dummy but sane
value for the root inode at mount time.
The inode attributes are refreshed with the correct values at the first
opportunity.
Signed-off-by: Miklos Szeredi <mszeredi@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Other than truncate, there are two cases, when fuse tries to get rid
of cached pages:
a) in open, if KEEP_CACHE flag is not set
b) in getattr, if file size changed spontaneously
Until now invalidate_mapping_pages() were used, which didn't get rid
of mapped pages. This is wrong, and becomes more wrong as dirty pages
are introduced. So instead properly invalidate all pages with
invalidate_inode_pages2().
Signed-off-by: Miklos Szeredi <mszeredi@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Memory mappings were only truncated on an explicit truncate, but not when the
file size was changed externally.
Fix this by moving the truncation code from fuse_setattr to
fuse_change_attributes.
Yes, there are races between write and and external truncation, but we can't
really do anything about them.
Signed-off-by: Miklos Szeredi <mszeredi@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Make lifetime of 'struct fuse_file' independent from 'struct file' by adding a
reference counter and destructor.
This will enable asynchronous page writeback, where it cannot be guaranteed,
that the file is not released while a request with this file handle is being
served.
The actual RELEASE request is only sent when there are no more references to
the fuse_file.
Signed-off-by: Miklos Szeredi <mszeredi@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Use wake_up_all instead of wake_up in put_reserved_req(), otherwise it is
possible that the right task is not woken up.
Also create a separate reserved_req_waitq in addition to the blocked_waitq,
since they fulfill totally separate functions.
Signed-off-by: Miklos Szeredi <mszeredi@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Set the read and write congestion state if the request queue is close to
blocking, and clear it when it's not.
This prevents unnecessary blocking in readahead and (when writable mmaps are
allowed) writeback.
Signed-off-by: Miklos Szeredi <mszeredi@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Slab constructors currently have a flags parameter that is never used. And
the order of the arguments is opposite to other slab functions. The object
pointer is placed before the kmem_cache pointer.
Convert
ctor(void *object, struct kmem_cache *s, unsigned long flags)
to
ctor(struct kmem_cache *s, void *object)
throughout the kernel
[akpm@linux-foundation.org: coupla fixes]
Signed-off-by: Christoph Lameter <clameter@sgi.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
[mszeredi]
- don't send zero length write requests
- it is not legal for the filesystem to return with zero written bytes
Signed-off-by: Nick Piggin <npiggin@suse.de>
Signed-off-by: Miklos Szeredi <mszeredi@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Slab destructors were no longer supported after Christoph's
c59def9f22 change. They've been
BUGs for both slab and slub, and slob never supported them
either.
This rips out support for the dtor pointer from kmem_cache_create()
completely and fixes up every single callsite in the kernel (there were
about 224, not including the slab allocator definitions themselves,
or the documentation references).
Signed-off-by: Paul Mundt <lethal@linux-sh.org>
They can use generic_file_splice_read() instead. Since sys_sendfile() now
prefers that, there should be no change in behaviour.
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
