It is incorrect to test inode dirty bits without participating in the inode
writeback protocol. Inode writeback sets I_SYNC and clears I_DIRTY_?, then
writes out the particular bits, then clears I_SYNC when it is done. BTW. it
may not completely write all pages out, so I_DIRTY_PAGES would get set
again.
This is a standard pattern used throughout the kernel's writeback caches
(I_SYNC ~= I_WRITEBACK, if that makes it clearer).
And so it is not possible to determine an inode's dirty status just by
checking I_DIRTY bits. Especially not for the purpose of data integrity
syncs.
Missing the check for these bits means that fsync can complete while
writeback to the inode is underway. Inode writeback functions get this
right, so call into them rather than try to shortcut things by testing
dirty state improperly.
Signed-off-by: Nick Piggin <npiggin@kernel.dk>
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
I stumbled on some of these prints in log files so, might
just submit the fixes.
* All i_ino prints in exofs should be hex
* All OSD_ERR prints should end with a "\n"
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
* 'bugfixes' of git://git.linux-nfs.org/projects/trondmy/nfs-2.6:
NFS: NFSROOT should default to "proto=udp"
nfs4: remove duplicated #include
NFSv4: nfs4_state_mark_reclaim_nograce() should be static
NFSv4: Fix the setlk error handler
NFSv4.1: Fix the handling of the SEQUENCE status bits
NFSv4/4.1: Fix nfs4_schedule_state_recovery abuses
NFSv4.1 reclaim complete must wait for completion
NFSv4: remove duplicate clientid in struct nfs_client
NFSv4.1: Retry CREATE_SESSION on NFS4ERR_DELAY
sunrpc: Propagate errors from xs_bind() through xs_create_sock()
(try3-resend) Fix nfs_compat_user_ino64 so it doesn't cause problems if bit 31 or 63 are set in fileid
nfs: fix compilation warning
nfs: add kmalloc return value check in decode_and_add_ds
SUNRPC: Remove resource leak in svc_rdma_send_error()
nfs: close NFSv4 COMMIT vs. CLOSE race
SUNRPC: Close a race in __rpc_wait_for_completion_task()
The kernel automatically evaluates partition tables of storage devices.
The code for evaluating OSF partitions contains a bug that leaks data
from kernel heap memory to userspace for certain corrupted OSF
partitions.
In more detail:
for (i = 0 ; i < le16_to_cpu(label->d_npartitions); i++, partition++) {
iterates from 0 to d_npartitions - 1, where d_npartitions is read from
the partition table without validation and partition is a pointer to an
array of at most 8 d_partitions.
Add the proper and obvious validation.
Signed-off-by: Timo Warns <warns@pre-sense.de>
Cc: stable@kernel.org
[ Changed the patch trivially to not repeat the whole le16_to_cpu()
thing, and to use an explicit constant for the magic value '8' ]
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Fix for a dumb preadv()/pwritev() compat bug - unlike the native
variants, the compat_... ones forget to check FMODE_P{READ,WRITE}, so
e.g. on pipe the native preadv() will fail with -ESPIPE and compat one
will act as readv() and succeed.
Not critical, but it's a clear bug with trivial fix, so IMO it's OK for
-final.
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
* git://git.kernel.org/pub/scm/linux/kernel/git/mason/btrfs-unstable:
Btrfs: break out of shrink_delalloc earlier
btrfs: fix not enough reserved space
btrfs: fix dip leak
Btrfs: make sure not to return overlapping extents to fiemap
Btrfs: deal with short returns from copy_from_user
Btrfs: fix regressions in copy_from_user handling
Josef had changed shrink_delalloc to exit after three shrink
attempts, which wasn't quite enough because new writers could
race in and steal free space.
But it also fixed deadlocks and stalls as we tried to recover
delalloc reservations. The code was tweaked to loop 1024
times, and would reset the counter any time a small amount
of progress was made. This was too drastic, and with a
lot of writers we can end up stuck in shrink_delalloc forever.
The shrink_delalloc loop is fairly complex because the caller is looping
too, and the caller will go ahead and force a transaction commit to make
sure we reclaim space.
This reworks things to exit shrink_delalloc when we've forced some
writeback and the delalloc reservations have gone down. This means
the writeback has not just started but has also finished at
least some of the metadata changes required to reclaim delalloc
space.
If we've got this wrong, we're returning ENOSPC too early, which
is a big improvement over the current behavior of hanging the machine.
Test 224 in xfstests hammers on this nicely, and with 1000 writers
trying to fill a 1GB drive we get our first ENOSPC at 93% full. The
other writers are able to continue until we get 100%.
This is a worst case test for btrfs because the 1000 writers are doing
small IO, and the small FS size means we don't have a lot of room
for metadata chunks.
Signed-off-by: Chris Mason <chris.mason@oracle.com>
There have been a number of recent reports that NFSROOT is no longer
working with default mount options, but fails only with certain NICs.
Brian Downing <bdowning@lavos.net> bisected to commit 56463e50 "NFS:
Use super.c for NFSROOT mount option parsing". Among other things,
this commit changes the default mount options for NFSROOT to use TCP
instead of UDP as the underlying transport.
TCP seems less able to deal with NICs that are slow to initialize.
The system logs that have accompanied reports of problems all show
that NFSROOT attempts to establish a TCP connection before the NIC is
fully initialized, and thus the TCP connection attempt fails.
When a TCP connection attempt fails during a mount operation, the
NFS stack needs to fail the operation. Usually user space knows how
and when to retry it. The network layer does not report a distinct
error code for this particular failure mode. Thus, there isn't a
clean way for the RPC client to see that it needs to retry in this
case, but not in others.
Because NFSROOT is used in some environments where it is not possible
to update the kernel command line to specify "udp", the proper thing
to do is change NFSROOT to use UDP by default, as it did before commit
56463e50.
To make it easier to see how to change default mount options for
NFSROOT and to distinguish default settings from mandatory settings,
I've adjusted a couple of areas to document the specifics.
root_nfs_cat() is also modified to deal with commas properly when
concatenating strings containing mount option lists. This keeps
root_nfs_cat() call sites simpler, now that we may be concatenating
multiple mount option strings.
Tested-by: Brian Downing <bdowning@lavos.net>
Tested-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Cc: <stable@kernel.org> # 2.6.37
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
There are no more external users of nfs4_state_mark_reclaim_nograce() or
nfs4_state_mark_reclaim_reboot(), so mark them as static.
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
We want SEQUENCE status bits to be handled by the state manager in order
to avoid threading issues.
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
nfs4_schedule_state_recovery() should only be used when we need to force
the state manager to check the lease. If we just want to start the
state manager in order to handle a state recovery situation, we should be
using nfs4_schedule_state_manager().
This patch fixes the abuses of nfs4_schedule_state_recovery() by replacing
its use with a set of helper functions that do the right thing.
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
Fix bug where we currently retry the EXCHANGEID call again, eventhough
we already have a valid clientid. Instead, delay and retry the CREATE_SESSION
call.
Signed-off-by: Ricardo Labiaga <Ricardo.Labiaga@netapp.com>
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
The problem was use of an int32, which when converted to a uint64
is sign extended resulting in a fileid that doesn't fit in 32 bits
even though the intent of the function is to fit the fileid into
32 bits.
Signed-off-by: Frank Filz <ffilzlnx@us.ibm.com>
Reviewed-by: Jeff Layton <jlayton@redhat.com>
[Trond: Added an include for compat.h]
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
add kmalloc return value check in decode_and_add_ds
Signed-off-by: Stanislav Fomichev <kernel@fomichev.me>
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
I've been adding in more artificial delays in the NFSv4 commit and close
codepaths to uncover races. The kernel I'm testing has the patch to
close the race in __rpc_wait_for_completion_task that's in Trond's
cthon2011 branch. The reproducer I've been using does this in a loop:
mkdir("DIR");
fd = open("DIR/FILE", O_WRONLY|O_CREAT|O_EXCL, 0644);
write(fd, "abcdefg", 7);
close(fd);
unlink("DIR/FILE");
rmdir("DIR");
The above reproducer shouldn't result in any silly-renaming. However,
when I add a "msleep(100)" just after the nfs_commit_clear_lock call in
nfs_commit_release, I can almost always force one to occur. If I can
force it to occur with that, then it can happen without that delay
given the right timing.
nfs_commit_inode waits for the NFS_INO_COMMIT bit to clear when called
with FLUSH_SYNC set. nfs_commit_rpcsetup on the other hand does not wait
for the task to complete before putting its reference to it, so the last
reference get put in rpc_release task and gets queued to a workqueue.
In this situation, the last open context reference may be put by the
COMMIT release instead of the close() syscall. The close() syscall
returns too quickly and the unlink runs while the d_count is still
high since the COMMIT release hasn't put its dentry reference yet.
Fix this by having rpc_commit_rpcsetup wait for the RPC call to complete
before putting the task reference when FLUSH_SYNC is set. With this, the
last reference is put by the process that's initiating the FLUSH_SYNC
commit and the race is closed.
Signed-off-by: Jeff Layton <jlayton@redhat.com>
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
Although they run as rpciod background tasks, under normal operation
(i.e. no SIGKILL), functions like nfs_sillyrename(), nfs4_proc_unlck()
and nfs4_do_close() want to be fully synchronous. This means that when we
exit, we want all references to the rpc_task to be gone, and we want
any dentry references etc. held by that task to be released.
For this reason these functions call __rpc_wait_for_completion_task(),
followed by rpc_put_task() in the expectation that the latter will be
releasing the last reference to the rpc_task, and thus ensuring that the
callback_ops->rpc_release() has been called synchronously.
This patch fixes a race which exists due to the fact that
rpciod calls rpc_complete_task() (in order to wake up the callers of
__rpc_wait_for_completion_task()) and then subsequently calls
rpc_put_task() without ensuring that these two steps are done atomically.
In order to avoid adding new spin locks, the patch uses the existing
waitqueue spin lock to order the rpc_task reference count releases between
the waiting process and rpciod.
The common case where nobody is waiting for completion is optimised for by
checking if the RPC_TASK_ASYNC flag is cleared and/or if the rpc_task
reference count is 1: in those cases we drop trying to grab the spin lock,
and immediately free up the rpc_task.
Those few processes that need to put the rpc_task from inside an
asynchronous context and that do not care about ordering are given a new
helper: rpc_put_task_async().
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
btrfs_link() will insert 3 items(inode ref, dir name item and dir index item)
into the b+ tree and update 2 items(its inode, and parent's inode) in the b+
tree. So we should reserve space for these 5 items, not 3 items.
Reported-by: Tsutomu Itoh <t-itoh@jp.fujitsu.com>
Signed-off-by: Miao Xie <miaox@cn.fujitsu.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
The btrfs DIO code leaks dip structs when dip->csums allocation
fails; bio->bi_end_io isn't set at the point where the free_ordered
branch is consequently taken, thus bio_endio doesn't call the function
which would free it in the normal case. Fix.
Signed-off-by: Daniel J Blueman <daniel.blueman@gmail.com>
Acked-by: Miao Xie <miaox@cn.fujitsu.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
Without this patch, inodes are not promptly freed on last close of an
unlinked file by an nfs client:
client$ mount -tnfs4 server:/export/ /mnt/
client$ tail -f /mnt/FOO
...
server$ df -i /export
server$ rm /export/FOO
(^C the tail -f)
server$ df -i /export
server$ echo 2 >/proc/sys/vm/drop_caches
server$ df -i /export
the df's will show that the inode is not freed on the filesystem until
the last step, when it could have been freed after killing the client's
tail -f. On-disk data won't be deallocated either, leading to possible
spurious ENOSPC.
This occurs because when the client does the close, it arrives in a
compound with a putfh and a close, processed like:
- putfh: look up the filehandle. The only alias found for the
inode will be DCACHE_UNHASHED alias referenced by the filp
this, so it creates a new DCACHE_DISCONECTED dentry and
returns that instead.
- close: closes the existing filp, which is destroyed
immediately by dput() since it's DCACHE_UNHASHED.
- end of the compound: release the reference
to the current filehandle, and dput() the new
DCACHE_DISCONECTED dentry, which gets put on the
unused list instead of being destroyed immediately.
Nick Piggin suggested fixing this by allowing d_obtain_alias to return
the unhashed dentry that is referenced by the filp, instead of making it
create a new dentry.
Leave __d_find_alias() alone to avoid changing behavior of other
callers.
Also nfsd doesn't need all the checks of __d_find_alias(); any dentry,
hashed or unhashed, disconnected or not, should work.
Signed-off-by: J. Bruce Fields <bfields@redhat.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
In the fallocate path the kernel doesn't check for the immutable/append
flag. It's possible to have a race condition in this scenario: an
application open a file in read/write and it does something, meanwhile
root set the immutable flag on the file, the application at that point
can call fallocate with success. In addition, we don't allow to do any
unreserve operation on an append only file but only the reserve one.
Signed-off-by: Marco Stornelli <marco.stornelli@gmail.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
* 'for-2.6.38' of git://linux-nfs.org/~bfields/linux:
nfsd: wrong index used in inner loop
nfsd4: fix bad pointer on failure to find delegation
NFSD: fix decode_cb_sequence4resok
* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs-2.6:
nd->inode is not set on the second attempt in path_walk()
unfuck proc_sysctl ->d_compare()
minimal fix for do_filp_open() race
We leave it at whatever it had been pointing to after the
first link_path_walk() had failed with -ESTALE. Things
do not work well after that...
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Index i was already used in the outer loop
Cc: stable@kernel.org
Signed-off-by: Roel Kluin <roel.kluin@gmail.com>
Signed-off-by: J. Bruce Fields <bfields@redhat.com>
The btrfs fiemap code was incorrectly returning duplicate or overlapping
extents in some cases. cp was blindly trusting this result and we would
end up with a destination file that was bigger than the original because
some bytes were copied twice.
The fix here adjusts our offsets to make sure we're always moving
forward in the fiemap results.
Signed-off-by: Chris Mason <chris.mason@oracle.com>
a) struct inode is not going to be freed under ->d_compare();
however, the thing PROC_I(inode)->sysctl points to just might.
Fortunately, it's enough to make freeing that sucker delayed,
provided that we don't step on its ->unregistering, clear
the pointer to it in PROC_I(inode) before dropping the reference
and check if it's NULL in ->d_compare().
b) I'm not sure that we *can* walk into NULL inode here (we recheck
dentry->seq between verifying that it's still hashed / fetching
dentry->d_inode and passing it to ->d_compare() and there's no
negative hashed dentries in /proc/sys/*), but if we can walk into
that, we really should not have ->d_compare() return 0 on it!
Said that, I really suspect that this check can be simply killed.
Nick?
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
In case of a nonempty list, the return on error here is obviously bogus;
it ends up being a pointer to the list head instead of to any valid
delegation on the list.
In particular, if nfsd4_delegreturn() hits this case, and you're quite unlucky,
then renew_client may oops, and it may take an embarassingly long time to
figure out why. Facepalm.
BUG: unable to handle kernel NULL pointer dereference at 0000000000000090
IP: [<ffffffff81292965>] nfsd4_delegreturn+0x125/0x200
...
Cc: stable@kernel.org
Signed-off-by: J. Bruce Fields <bfields@redhat.com>
When copy_from_user is only able to copy some of the bytes we requested,
we may end up creating a partially up to date page. To avoid garbage in
the page, we need to treat a partial copy as a zero length copy.
This makes the rest of the file_write code drop the page and
retry the whole copy instead of marking the partially up to
date page as dirty.
Signed-off-by: Chris Mason <chris.mason@oracle.com>
cc: stable@kernel.org
Commit 914ee295af fixed deadlocks in
btrfs_file_write where we would catch page faults on pages we had
locked.
But, there were a few problems:
1) The x86-32 iov_iter_copy_from_user_atomic code always fails to copy
data when the amount to copy is more than 4K and the offset to start
copying from is not page aligned. The result was btrfs_file_write
looping forever retrying the iov_iter_copy_from_user_atomic
We deal with this by changing btrfs_file_write to drop down to single
page copies when iov_iter_copy_from_user_atomic starts returning failure.
2) The btrfs_file_write code was leaking delalloc reservations when
iov_iter_copy_from_user_atomic returned zero. The looping above would
result in the entire filesystem running out of delalloc reservations and
constantly trying to flush things to disk.
3) btrfs_file_write will lock down page cache pages, make sure
any writeback is finished, do the copy_from_user and then release them.
Before the loop runs we check the first and last pages in the write to
see if they are only being partially modified. If the start or end of
the write isn't aligned, we make sure the corresponding pages are
up to date so that we don't introduce garbage into the file.
With the copy_from_user changes, we're allowing the VM to reclaim the
pages after a partial update from copy_from_user, but we're not
making sure the page cache page is up to date when we loop around to
resume the write.
We deal with this by pushing the up to date checks down into the page
prep code. This fits better with how the rest of file_write works.
Signed-off-by: Chris Mason <chris.mason@oracle.com>
Reported-by: Mitch Harder <mitch.harder@sabayonlinux.org>
cc: stable@kernel.org
* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/sage/ceph-client:
ceph: no .snap inside of snapped namespace
libceph: fix msgr standby handling
libceph: fix msgr keepalive flag
libceph: fix msgr backoff
libceph: retry after authorization failure
libceph: fix handling of short returns from get_user_pages
ceph: do not clear I_COMPLETE from d_release
ceph: do not set I_COMPLETE
Revert "ceph: keep reference to parent inode on ceph_dentry"
The "bad_page()" page allocator sanity check was reported recently (call
chain as follows):
bad_page+0x69/0x91
free_hot_cold_page+0x81/0x144
skb_release_data+0x5f/0x98
__kfree_skb+0x11/0x1a
tcp_ack+0x6a3/0x1868
tcp_rcv_established+0x7a6/0x8b9
tcp_v4_do_rcv+0x2a/0x2fa
tcp_v4_rcv+0x9a2/0x9f6
do_timer+0x2df/0x52c
ip_local_deliver+0x19d/0x263
ip_rcv+0x539/0x57c
netif_receive_skb+0x470/0x49f
:virtio_net:virtnet_poll+0x46b/0x5c5
net_rx_action+0xac/0x1b3
__do_softirq+0x89/0x133
call_softirq+0x1c/0x28
do_softirq+0x2c/0x7d
do_IRQ+0xec/0xf5
default_idle+0x0/0x50
ret_from_intr+0x0/0xa
default_idle+0x29/0x50
cpu_idle+0x95/0xb8
start_kernel+0x220/0x225
_sinittext+0x22f/0x236
It occurs because an skb with a fraglist was freed from the tcp
retransmit queue when it was acked, but a page on that fraglist had
PG_Slab set (indicating it was allocated from the Slab allocator (which
means the free path above can't safely free it via put_page.
We tracked this back to an nfsv4 setacl operation, in which the nfs code
attempted to fill convert the passed in buffer to an array of pages in
__nfs4_proc_set_acl, which gets used by the skb->frags list in
xs_sendpages. __nfs4_proc_set_acl just converts each page in the buffer
to a page struct via virt_to_page, but the vfs allocates the buffer via
kmalloc, meaning the PG_slab bit is set. We can't create a buffer with
kmalloc and free it later in the tcp ack path with put_page, so we need
to either:
1) ensure that when we create the list of pages, no page struct has
PG_Slab set
or
2) not use a page list to send this data
Given that these buffers can be multiple pages and arbitrarily sized, I
think (1) is the right way to go. I've written the below patch to
allocate a page from the buddy allocator directly and copy the data over
to it. This ensures that we have a put_page free-able page for every
entry that winds up on an skb frag list, so it can be safely freed when
the frame is acked. We do a put page on each entry after the
rpc_call_sync call so as to drop our own reference count to the page,
leaving only the ref count taken by tcp_sendpages. This way the data
will be properly freed when the ack comes in
Successfully tested by myself to solve the above oops.
Note, as this is the result of a setacl operation that exceeded a page
of data, I think this amounts to a local DOS triggerable by an
uprivlidged user, so I'm CCing security on this as well.
Signed-off-by: Neil Horman <nhorman@tuxdriver.com>
CC: Trond Myklebust <Trond.Myklebust@netapp.com>
CC: security@kernel.org
CC: Jeff Layton <jlayton@redhat.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
failure exits on the no-O_CREAT side of do_filp_open() merge with
those of O_CREAT one; unfortunately, if do_path_lookup() returns
-ESTALE, we'll get out_filp:, notice that we are about to return
-ESTALE without having trying to create the sucker with LOOKUP_REVAL
and jump right into the O_CREAT side of code. And proceed to try
and create a file. Usually that'll fail with -ESTALE again, but
we can race and get that attempt of pathname resolution to succeed.
open() without O_CREAT really shouldn't end up creating files, races
or not. The real fix is to rearchitect the whole do_filp_open(),
but for now splitting the failure exits will do.
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
