This simple implementation just checks for no ACLs on the inode, and
if so, then the rcu-walk may proceed, otherwise fail it.
Signed-off-by: Nick Piggin <npiggin@kernel.dk>
Reduce some branches and memory accesses in dcache lookup by adding dentry
flags to indicate common d_ops are set, rather than having to check them.
This saves a pointer memory access (dentry->d_op) in common path lookup
situations, and saves another pointer load and branch in cases where we
have d_op but not the particular operation.
Patched with:
git grep -E '[.>]([[:space:]])*d_op([[:space:]])*=' | xargs sed -e 's/\([^\t ]*\)->d_op = \(.*\);/d_set_d_op(\1, \2);/' -e 's/\([^\t ]*\)\.d_op = \(.*\);/d_set_d_op(\&\1, \2);/' -i
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>
Change d_delete from a dentry deletion notification to a dentry caching
advise, more like ->drop_inode. Require it to be constant and idempotent,
and not take d_lock. This is how all existing filesystems use the callback
anyway.
This makes fine grained dentry locking of dput and dentry lru scanning
much simpler.
Signed-off-by: Nick Piggin <npiggin@kernel.dk>
The extent allocator has code that allows us to fill
allocations from any available block group, even if it doesn't
match the raid level we've requested.
This was put in because adding a new drive to a filesystem
made with the default mkfs options actually upgrades the metadata from
single spindle dup to full RAID1.
But, the code also allows us to allocate from a raid0 chunk when we
really want a raid1 or raid10 chunk. This can cause big trouble because
mkfs creates a small (4MB) raid0 chunk for data and metadata which then
goes unused for raid1/raid10 installs.
The allocator will happily wander in and allocate from that chunk when
things get tight, which is not correct.
The fix here is to make sure that we provide duplication when the
caller has asked for it. It does all the dups to be any raid level,
which preserves the dup->raid1 upgrade abilities.
Signed-off-by: Chris Mason <chris.mason@oracle.com>
When we mount in RAID degraded mode without adding a new device to
replace the failed one, we can end up using the wrong RAID flags for
allocations.
This results in strange combinations of block groups (raid1 in a raid10
filesystem) and corruptions when we try to allocate blocks from single
spindle chunks on drives that are actually missing.
The first device has two small 4MB chunks in it that mkfs creates and
these are usually unused in a raid1 or raid10 setup. But, in -o degraded,
the allocator will fall back to these because the mask of desired raid groups
isn't correct.
The fix here is to count the missing devices as we build up the list
of devices in the system. This count is used when picking the
raid level to make sure we continue using the same levels that were
in place before we lost a drive.
Signed-off-by: Chris Mason <chris.mason@oracle.com>
If we just get a plain IO error when we read tree roots, the code
wasn't properly sending that error up the chain. This allowed mounts to
continue when they should failed, and allowed operations
on partially setup root structs. The end result was usually oopsen
on spinlocks that hadn't been spun up correctly.
Signed-off-by: Chris Mason <chris.mason@oracle.com>
... regarding an unused function when !MIGRATION, and regarding a
printk() format string vs argument mismatch.
Signed-off-by: Jan Beulich <jbeulich@novell.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
If we had reserved some bytes in struct btrfs_ioctl_vol_args, we
wouldn't have to create a new structure for async snapshot creation.
Here we convert async snapshot ioctl to use a more generic ABI, as
we'll add more ioctls for snapshots/subvolumes in the future, readonly
snapshots for example.
Signed-off-by: Li Zefan <lizf@cn.fujitsu.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
This problem is found in meego testing:
http://bugs.meego.com/show_bug.cgi?id=6672
A file in btrfs is mmaped and the mmaped buffer is passed to pwrite to write to the same page
of the same file. In btrfs_file_aio_write(), the pages is locked by prepare_pages(). So when
btrfs_copy_from_user() is called, page fault happens and the same page needs to be locked again
in filemap_fault(). The fix is to move iov_iter_fault_in_readable() before prepage_pages() to make page
fault happen before pages are locked. And also disable page fault in critical region in
btrfs_copy_from_user().
Reviewed-by: Yan, Zheng<zheng.z.yan@intel.com>
Signed-off-by: Zhong, Xin <xin.zhong@intel.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
We should drop dentry before deactivating the superblock, otherwise
we can hit this bug:
BUG: Dentry f349a690{i=100,n=/} still in use (1) [unmount of btrfs loop1]
...
Steps to reproduce the bug:
# mount /dev/loop1 /mnt
# mkdir save
# btrfs subvolume snapshot /mnt save/snap1
# umount /mnt
# mount -o subvol=save/snap1 /dev/loop1 /mnt
(crash)
Reported-by: Michael Niederle <mniederle@gmx.at>
Signed-off-by: Li Zefan <lizf@cn.fujitsu.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
We were incorrectly taking the async path even for the sync ioctls by
passing in &transid unconditionally.
There's ample room for further cleanup here, but this keeps the fix simple.
Signed-off-by: Sage Weil <sage@newdream.net>
Reviewed-by: Li Zefan <lizf@cn.fujitsu.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
"start + num_bytes >= actual_end" can happen when compressed page writeback races
with file truncation. In that case we need unlock and release pages past the end
of file.
Signed-off-by: Yan, Zheng <zheng.z.yan@intel.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
Not being able to delete an orphan item isn't a horrible thing. The worst that
happens is the next time around we try and do the orphan cleanup and we can't
find the referenced object and just delete the item and move on.
Signed-off-by: Josef Bacik <josef@redhat.com>
If the orphan item doesn't exist, we return 1, which doesn't make any sense to
the callers. Instead return -ENOENT if we didn't find the item. Thanks,
Signed-off-by: Josef Bacik <josef@redhat.com>
Since the fast caching uses normal tree locking, we can possibly deadlock if we
get to the caching via a btrfs_search_slot() on the tree_root. So just check to
see if the root we are on is the tree root, and just don't do the fast caching.
Reported-by: Sage Weil <sage@newdream.net>
Signed-off-by: Josef Bacik <josef@redhat.com>
Currently if the space cache inode generation number doesn't match the
generation number in the space cache header we will just fail to load the space
cache, but we won't mark the space cache as an error, so we'll keep getting that
error each time somebody tries to cache that block group until we actually clear
the thing. Fix this by marking the space cache as having an error so we only
get the message once. This patch also makes it so that we don't try and setup
space cache for a block group that isn't cached, since we won't be able to write
it out anyway. None of these problems are actual problems, they are just
annoying and sub-optimal. Thanks,
Signed-off-by: Josef Bacik <josef@redhat.com>
This fixes a bug where we use dip after we have freed it. Instead just use the
file_offset that was passed to the function. Thanks,
Signed-off-by: Josef Bacik <josef@redhat.com>
The new DIO bio splitting code has problems when the bio
spans more than one ordered extent. This will happen as the
generic DIO code merges our get_blocks calls together into
a bigger single bio.
This fixes things by walking forward in the ordered extent
code finding all the overlapping ordered extents and completing them
all at once.
Signed-off-by: Chris Mason <chris.mason@oracle.com>
There is a problem with how we use sget, it searches through the list of supers
attached to the fs_type looking for a super with the same fs_devices as what
we're trying to mount. This depends on sb->s_fs_info being filled, but we don't
fill that in until we get to btrfs_fill_super, so we could hit supers on the
fs_type super list that have a null s_fs_info. In order to fix that we need to
go ahead and setup a blank root with a blank fs_info to hold fs_devices, that
way our test will work out right and then we can set s_fs_info in
btrfs_set_super, and then open_ctree will simply use our pre-allocated root and
fs_info when setting everything up. Thanks,
Signed-off-by: Josef Bacik <josef@redhat.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
There are two big problems currently with FIEMAP
1) We return extents for holes. This isn't supposed to happen, we just don't
return extents for holes and then userspace interprets the lack of an extent as
a hole.
2) We sometimes don't set FIEMAP_EXTENT_LAST properly. This is because we wait
to see a EXTENT_FLAG_VACANCY flag on the em, but this won't happen if say we ask
fiemap to map up to the last extent in a file, and there is nothing but holes up
to the i_size. To fix this we need to lookup the last extent in this file and
save the logical offset, so if we happen to try and map that extent we can be
sure to set FIEMAP_EXTENT_LAST.
With this patch we now pass xfstest 225, which we never have before.
Signed-off-by: Josef Bacik <josef@redhat.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
When mounting a btrfs file system btrfs_test_super() may attempt to
use sb->s_fs_info, the btrfs root, of a super block that is going away
and that has had the btrfs root set to NULL in its ->put_super(). But
if the super block is going away it cannot be an existing super block
so we can return false in this case.
Signed-off-by: Ian Kent <raven@themaw.net>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
Currently we fail xfstest 236 because we're not updating the inode ctime on
link. This is a simple fix, and makes it so we pass 236 now.
Signed-off-by: Josef Bacik <josef@redhat.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
We have been failing xfstest 228 forever, because we don't check to make sure
the new inode size is acceptable as far as RLIMIT is concerned. Just check to
make sure it's ok to create a inode with this new size and error out if not.
With this patch we now pass 228.
Signed-off-by: Josef Bacik <josef@redhat.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
There is a typo in __btrfs_prealloc_file_range() where we set the i_size to
actual_len/cur_offset, and then just set it to cur_offset again, and do the same
with btrfs_ordered_update_i_size(). This fixes it back to keeping i_size in a
local variable and then updating i_size properly. Tested this with
xfs_io -F -f -c "falloc 0 1" -c "pwrite 0 1" foo
stat'ing foo gives us a size of 1 instead of 4096 like it was. Thanks,
Signed-off-by: Josef Bacik <josef@redhat.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
Everybody who calls btrfs_add_nondir just passes in the dentry of the new file
and then dereference dentry->d_parent->d_inode, but everybody who calls
btrfs_add_nondir() are already passed the parent's inode. So instead of
dereferencing dentry->d_parent, just make btrfs_add_nondir take the dir inode as
an argument and pass that along so we don't have to worry about d_parent.
Thanks,
Signed-off-by: Josef Bacik <josef@redhat.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
Since we walk up the path logging all of the parts of the inode's path, we need
to hold i_mutex to make sure that the inode is not renamed while we're logging
everything. btrfs_log_dentry_safe does dget_parent and all of that jazz, but we
may get unexpected results if the rename changes the inode's location while
we're higher up the path logging those dentries, so do this for safety reasons.
Thanks,
Signed-off-by: Josef Bacik <josef@redhat.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
There are lots of places where we do dentry->d_parent->d_inode without holding
the dentry->d_lock. This could cause problems with rename. So instead we need
to use dget_parent() and hold the reference to the parent as long as we are
going to use it's inode and then dput it at the end.
Signed-off-by: Josef Bacik <josef@redhat.com>
Cc: raven@themaw.net
Signed-off-by: Chris Mason <chris.mason@oracle.com>
When creating new inodes we don't setup inode->i_generation. So if we generate
an fh with a newly created inode we save the generation of 0, but if we flush
the inode to disk and have to read it back when getting the inode on the server
we'll have the right i_generation, so gens wont match and we get ESTALE. This
patch properly sets inode->i_generation when we create the new inode and now I'm
no longer getting ESTALE. Thanks,
Signed-off-by: Josef Bacik <josef@redhat.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
People kept reporting NFS issues, specifically getting ESTALE alot. I figured
out how to reproduce the problem
SERVER
mkfs.btrfs /dev/sda1
mount /dev/sda1 /mnt/btrfs-test
<add /mnt/btrfs-test to /etc/exports>
btrfs subvol create /mnt/btrfs-test/foo
service nfs start
CLIENT
mount server:/mnt/btrfs /mnt/test
cd /mnt/test/foo
ls
SERVER
echo 3 > /proc/sys/vm/drop_caches
CLIENT
ls <-- get an ESTALE here
This is because the standard way to lookup a name in nfsd is to use readdir, and
what it does is do a readdir on the parent directory looking for the inode of
the child. So in this case the parent being / and the child being foo. Well
subvols all have the same inode number, so doing a readdir of / looking for
inode 256 will return '.', which obviously doesn't match foo. So instead we
need to have our own .get_name so that we can find the right name.
Our .get_name will either lookup the inode backref or the root backref,
whichever we're looking for, and return the name we find. Running the above
reproducer with this patch results in everything acting the way its supposed to.
Thanks,
Signed-off-by: Josef Bacik <josef@redhat.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
Set src_offset = 0, src_length = 20K, dest_offset = 20K. And the
original filesize of the dest file 'file2' is 30K:
# ls -l /mnt/file2
-rw-r--r-- 1 root root 30720 Nov 18 16:42 /mnt/file2
Now clone file1 to file2, the dest file should be 40K, but it
still shows 30K:
# ls -l /mnt/file2
-rw-r--r-- 1 root root 30720 Nov 18 16:42 /mnt/file2
Signed-off-by: Li Zefan <lizf@cn.fujitsu.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
We've done the check for src_offset and src_length, and We should
also check dest_offset, otherwise we'll corrupt the destination
file:
(After cloning file1 to file2 with unaligned dest_offset)
# cat /mnt/file2
cat: /mnt/file2: Input/output error
Signed-off-by: Li Zefan <lizf@cn.fujitsu.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
When I added the clear_cache option I screwed up and took the break out of
the space_cache case statement, so whenever you mount with space_cache you also
get clear_cache, which does you no good if you say set space_cache in fstab so
it always gets set. This patch adds the break back in properly.
Signed-off-by: Josef Bacik <josef@redhat.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
'unused' calculated with wrong sign in reserve_metadata_bytes().
This might have lead to unwanted over-reservations.
Signed-off-by: Arne Jansen <sensille@gmx.net>
Reviewed-by: Josef Bacik <josef@redhat.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
extent_bio_alloc() and compressed_bio_alloc() are similar, cleanup
similar source code.
Signed-off-by: Miao Xie <miaox@cn.fujitsu.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
bio_endio() will free dip and dip->csums, so dip and dip->csums twice will
be freed twice. Fix it.
Signed-off-by: Miao Xie <miaox@cn.fujitsu.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
Migrate page will directly call the btrfs btree writepage function,
which isn't actually allowed.
Our writepage assumes that you have locked the extent_buffer and
flagged the block as written. Without doing these steps, we can
corrupt metadata blocks.
A later commit will remove the btree writepage function since
it is really only safely used internally by btrfs. We
use writepages for everything else.
Signed-off-by: Chris Mason <chris.mason@oracle.com>
During unlink we remove any references to the inode from
the tree log. It can return -ENOENT and other errors,
and this changes the unlink code to deal with it.
Signed-off-by: Chris Mason <chris.mason@oracle.com>
Add a mount option user_subvol_rm_allowed that allows users to delete a
(potentially non-empty!) subvol when they would otherwise we allowed to do
an rmdir(2). We duplicate the may_delete() checks from the core VFS code
to implement identical security checks (minus the directory size check).
We additionally require that the user has write+exec permission on the
subvol root inode.
Signed-off-by: Sage Weil <sage@newdream.net>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
There is no reason to force an immediate commit when deleting a snapshot.
Users have some expectation that space from a deleted snapshot be freed
immediately, but even if we do commit the reclaim is a background process.
If users _do_ want the deletion to be durable, they can call 'sync'.
Signed-off-by: Sage Weil <sage@newdream.net>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
Create a snap without waiting for it to commit to disk. The ioctl is
ordered such that subsequent operations will not be contained by the
created snapshot, and the commit is initiated, but the ioctl does not
wait for the snapshot to commit to disk.
We return the specific transid to userspace so that an application can wait
for this specific snapshot creation to commit via the WAIT_SYNC ioctl.
Signed-off-by: Sage Weil <sage@newdream.net>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
START_SYNC will start a sync/commit, but not wait for it to
complete. Any modification started after the ioctl returns is
guaranteed not to be included in the commit. If a non-NULL
pointer is passed, the transaction id will be returned to
userspace.
WAIT_SYNC will wait for any in-progress commit to complete. If a
transaction id is specified, the ioctl will block and then
return (success) when the specified transaction has committed.
If it has already committed when we call the ioctl, it returns
immediately. If the specified transaction doesn't exist, it
returns EINVAL.
If no transaction id is specified, WAIT_SYNC will wait for the
currently committing transaction to finish it's commit to disk.
If there is no currently committing transaction, it returns
success.
These ioctls are useful for applications which want to impose an
ordering on when fs modifications reach disk, but do not want to
wait for the full (slow) commit process to do so.
Picky callers can take the transid returned by START_SYNC and
feed it to WAIT_SYNC, and be certain to wait only as long as
necessary for the transaction _they_ started to reach disk.
Sloppy callers can START_SYNC and WAIT_SYNC without a transid,
and provided they didn't wait too long between the calls, they
will get the same result. However, if a second commit starts
before they call WAIT_SYNC, they may end up waiting longer for
it to commit as well. Even so, a START_SYNC+WAIT_SYNC still
guarantees that any operation completed before the START_SYNC
reaches disk.
Signed-off-by: Sage Weil <sage@newdream.net>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
Add support for an async transaction commit that is ordered such that any
subsequent operations will join the following transaction, but does not
wait until the current commit is fully on disk. This avoids much of the
latency associated with the btrfs_commit_transaction for callers concerned
with serialization and not safety.
The wait_for_unblock flag controls whether we wait for the 'middle' portion
of commit_transaction to complete, which is necessary if the caller expects
some of the modifications contained in the commit to be available (this is
the case for subvol/snapshot creation).
Signed-off-by: Sage Weil <sage@newdream.net>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
Nick Piggin