sfrench/cifs-2.6.git
5 years agoNFS: Don't hold the group lock when calling nfs_release_request()
Trond Myklebust [Sat, 9 Sep 2017 19:31:28 +0000 (15:31 -0400)]
NFS: Don't hold the group lock when calling nfs_release_request()

That can deadlock if this is the last reference since
nfs_page_group_destroy() calls nfs_page_group_sync_on_bit().
Note that even if the page was removed from the subpage list,
the req->wb_head could still be pointing to the old head.

Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
5 years agoNFS: Remove pnfs_generic_transfer_commit_list()
Trond Myklebust [Sat, 9 Sep 2017 16:40:38 +0000 (12:40 -0400)]
NFS: Remove pnfs_generic_transfer_commit_list()

It's pretty much a duplicate of nfs_scan_commit_list() that also
clears the PG_COMMIT_TO_DS flag.

Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
5 years agoNFS: nfs_lock_and_join_requests and nfs_scan_commit_list can deadlock
Trond Myklebust [Sat, 9 Sep 2017 16:28:01 +0000 (12:28 -0400)]
NFS: nfs_lock_and_join_requests and nfs_scan_commit_list can deadlock

Since the commit list is not ordered, it is possible for nfs_scan_commit_list
to hold a request that nfs_lock_and_join_requests() is waiting for, while
at the same time trying to grab a request that nfs_lock_and_join_requests
already holds.

Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
5 years agoNFS: Fix 2 use after free issues in the I/O code
Trond Myklebust [Sat, 9 Sep 2017 01:28:11 +0000 (21:28 -0400)]
NFS: Fix 2 use after free issues in the I/O code

The writeback code wants to send a commit after processing the pages,
which is why we want to delay releasing the struct path until after
that's done.

Also, the layout code expects that we do not free the inode before
we've put the layout segments in pnfs_writehdr_free() and
pnfs_readhdr_free()

Fixes: 919e3bd9a875 ("NFS: Ensure we commit after writeback is complete")
Fixes: 4714fb51fd03 ("nfs: remove pgio_header refcount, related cleanup")
Cc: stable@vger.kernel.org
Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
5 years agoNFS: Sync the correct byte range during synchronous writes
tarangg@amazon.com [Thu, 7 Sep 2017 13:29:23 +0000 (09:29 -0400)]
NFS: Sync the correct byte range during synchronous writes

Since commit 18290650b1c8 ("NFS: Move buffered I/O locking into
nfs_file_write()") nfs_file_write() has not flushed the correct byte
range during synchronous writes.  generic_write_sync() expects that
iocb->ki_pos points to the right edge of the range rather than the
left edge.

To replicate the problem, open a file with O_DSYNC, have the client
write at increasing offsets, and then print the successful offsets.
Block port 2049 partway through that sequence, and observe that the
client application indicates successful writes in advance of what the
server received.

Fixes: 18290650b1c8 ("NFS: Move buffered I/O locking into nfs_file_write()")
Signed-off-by: Jacob Strauss <jsstraus@amazon.com>
Signed-off-by: Tarang Gupta <tarangg@amazon.com>
Tested-by: Tarang Gupta <tarangg@amazon.com>
Cc: stable@vger.kernel.org # v4.8+
Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
5 years agolockd: Delete an error message for a failed memory allocation in reclaimer()
Markus Elfring [Wed, 16 Aug 2017 20:35:24 +0000 (22:35 +0200)]
lockd: Delete an error message for a failed memory allocation in reclaimer()

Omit an extra message for a memory allocation failure in this function.

This issue was detected by using the Coccinelle software.

Signed-off-by: Markus Elfring <elfring@users.sourceforge.net>
Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
5 years agoNFS: remove jiffies field from access cache
NeilBrown [Wed, 16 Aug 2017 02:47:27 +0000 (12:47 +1000)]
NFS: remove jiffies field from access cache

This field hasn't been used since commit 57b691819ee2 ("NFS: Cache
access checks more aggressively").

Signed-off-by: NeilBrown <neilb@suse.com>
Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
5 years agoNFS: flush data when locking a file to ensure cache coherence for mmap.
NeilBrown [Fri, 18 Aug 2017 07:12:52 +0000 (17:12 +1000)]
NFS: flush data when locking a file to ensure cache coherence for mmap.

When a byte range lock (or flock) is taken out on an NFS file, the
validity of the cached data is checked and the inode is marked
NFS_INODE_INVALID_DATA.  However the cached data isn't flushed from
the page cache.

This is sufficient for future read() requests or mmap() requests as
they call nfs_revalidate_mapping() which performs the flush if
necessary.

However an existing mapping is not affected.  Accessing data through
that mapping will continue to return old data even though the inode is
marked NFS_INODE_INVALID_DATA.

This can easily be confirmed using the 'nfs' tool in
  git://github.com/okirch/twopence-nfs.git
and running

   nfs coherence FILENAME
on one client, and
   nfs coherence -r FILENAME
on another client.

It appears that prior to Linux 2.6.0 this worked correctly.

However commit:

http://git.kernel.org/cgit/linux/kernel/git/history/history.git/commit/?id=ca9268fe3ddd075714005adecd4afbd7f9ab87d0

removed the call to inode_invalidate_pages() from nfs_zap_caches().  I
haven't tested this code, but inspection suggests that prior to this
commit, file locking would invalidate all inode pages.

This patch adds a call to nfs_revalidate_mapping() after a
successful SETLK so that invalid data is flushed.  With this patch the
above test passes.  To minimize impact (and possibly avoid a GETATTR
call) this only happens if the mapping might be mapped into
userspace.

Cc: Olaf Kirch <okir@suse.com>
Signed-off-by: NeilBrown <neilb@suse.com>
Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
5 years agoSUNRPC: remove some dead code.
NeilBrown [Fri, 18 Aug 2017 07:12:52 +0000 (17:12 +1000)]
SUNRPC: remove some dead code.

RPC_TASK_NO_RETRANS_TIMEOUT is set when cl_noretranstimeo
is set, which happens when  RPC_CLNT_CREATE_NO_RETRANS_TIMEOUT is set,
which happens when NFS_CS_NO_RETRANS_TIMEOUT is set.

This flag means "don't resend on a timeout, only resend if the
connection gets broken for some reason".

cl_discrtry is set when RPC_CLNT_CREATE_DISCRTRY is set, which
happens when NFS_CS_DISCRTRY is set.

This flag means "always disconnect before resending".

NFS_CS_NO_RETRANS_TIMEOUT and NFS_CS_DISCRTRY are both only set
in nfs4_init_client(), and it always sets both.

So we will never have a situation where only one of the flags is set.
So this code, which tests if timeout retransmits are allowed, and
disconnection is required, will never run.

So it makes sense to remove this code as it cannot be tested and
could confuse people reading the code (like me).

(alternately we could leave it there with a comment saying
 it is never actually used).

Signed-off-by: NeilBrown <neilb@suse.com>
Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
5 years agoNFS: don't expect errors from mempool_alloc().
NeilBrown [Fri, 18 Aug 2017 07:12:51 +0000 (17:12 +1000)]
NFS: don't expect errors from mempool_alloc().

Commit fbe77c30e9ab ("NFS: move rw_mode to nfs_pageio_header")
reintroduced some pointless code that commit 518662e0fcb9 ("NFS: fix
usage of mempools.") had recently removed.

Remove it again.

Cc: Benjamin Coddington <bcodding@redhat.com>
Signed-off-by: NeilBrown <neilb@suse.com>
Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
5 years agoxprtrdma: Use xprt_pin_rqst in rpcrdma_reply_handler
Chuck Lever [Wed, 23 Aug 2017 21:05:58 +0000 (17:05 -0400)]
xprtrdma: Use xprt_pin_rqst in rpcrdma_reply_handler

Adopt the use of xprt_pin_rqst to eliminate contention between
Call-side users of rb_lock and the use of rb_lock in
rpcrdma_reply_handler.

This replaces the mechanism introduced in 431af645cf66 ("xprtrdma:
Fix client lock-up after application signal fires").

Use recv_lock to quickly find the completing rqst, pin it, then
drop the lock. At that point invalidation and pull-up of the Reply
XDR can be done. Both are often expensive operations.

Finally, take recv_lock again to signal completion to the RPC
layer. It also protects adjustment of "cwnd".

This greatly reduces the amount of time a lock is held by the
reply handler. Comparing lock_stat results shows a marked decrease
in contention on rb_lock and recv_lock.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
[trond.myklebust@primarydata.com: Remove call to rpcrdma_buffer_put() from
   the "out_norqst:" path in rpcrdma_reply_handler.]
Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
5 years agoMerge tag 'nfs-rdma-for-4.14-1' of git://git.linux-nfs.org/projects/anna/linux-nfs...
Trond Myklebust [Tue, 5 Sep 2017 19:16:04 +0000 (15:16 -0400)]
Merge tag 'nfs-rdma-for-4.14-1' of git://git.linux-nfs.org/projects/anna/linux-nfs into linux-next

NFS-over-RDMA client updates for Linux 4.14

Bugfixes and cleanups:
- Constify rpc_xprt_ops
- Harden RPC call encoding and decoding
- Clean up rpc call decoding to use xdr_streams
- Remove unused variables from various structures
- Refactor code to remove imul instructions
- Rearrange rx_stats structure for better cacheline sharing

5 years agoxprtrdma: Re-arrange struct rx_stats
Chuck Lever [Tue, 22 Aug 2017 15:19:35 +0000 (11:19 -0400)]
xprtrdma: Re-arrange struct rx_stats

To reduce false cacheline sharing, separate counters that are likely
to be accessed in the Call path from those accessed in the Reply
path.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
5 years agoMerge branch 'bugfixes'
Trond Myklebust [Sun, 20 Aug 2017 17:04:12 +0000 (13:04 -0400)]
Merge branch 'bugfixes'

5 years agoNFS: Fix NFSv2 security settings
Chuck Lever [Thu, 10 Aug 2017 20:41:31 +0000 (16:41 -0400)]
NFS: Fix NFSv2 security settings

For a while now any NFSv2 mount where sec= is specified uses
AUTH_NULL. If sec= is not specified, the mount uses AUTH_UNIX.
Commit e68fd7c8071d ("mount: use sec= that was specified on the
command line") attempted to address a very similar problem with
NFSv3, and should have fixed this too, but it has a bug.

The MNTv1 MNT procedure does not return a list of security flavors,
so our client makes up a list containing just AUTH_NULL. This should
enable nfs_verify_authflavors() to assign the sec= specified flavor,
but instead, it incorrectly sets it to AUTH_NULL.

I expect this would also be a problem for any NFSv3 server whose
MNTv3 MNT procedure returned a security flavor list containing only
AUTH_NULL.

Fixes: e68fd7c8071d ("mount: use sec= that was specified on ... ")
BugLink: https://bugzilla.linux-nfs.org/show_bug.cgi?id=310
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
5 years agoNFSv4.1: don't use machine credentials for CLOSE when using 'sec=sys'
NeilBrown [Fri, 18 Aug 2017 07:12:52 +0000 (17:12 +1000)]
NFSv4.1: don't use machine credentials for CLOSE when using 'sec=sys'

An NFSv4.1 client might close a file after the user who opened it has
logged off.  In this case the user's credentials may no longer be
valid, if they are e.g. kerberos credentials that have expired.

NFSv4.1 has a mechanism to allow the client to use machine credentials
to close a file.  However due to a short-coming in the RFC, a CLOSE
with those credentials may not be possible if the file in question
isn't exported to the same security flavor - the required PUTFH must
be rejected when this is the case.

Specifically if a server and client support kerberos in general and
have used it to form a machine credential, but the file is only
exported to "sec=sys", a PUTFH with the machine credentials will fail,
so CLOSE is not possible.

As RPC_AUTH_UNIX (used by sec=sys) credentials can never expire, there
is no value in using the machine credential in place of them.
So in that case, just use the users credentials for CLOSE etc, as you would
in NFSv4.0

Signed-off-by: Neil Brown <neilb@suse.com>
Signed-off-by: NeilBrown <neilb@suse.com>
Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
5 years agoSUNRPC: ECONNREFUSED should cause a rebind.
NeilBrown [Fri, 18 Aug 2017 07:12:51 +0000 (17:12 +1000)]
SUNRPC: ECONNREFUSED should cause a rebind.

If you
 - mount and NFSv3 filesystem
 - do some file locking which requires the server
   to make a GRANT call back
 - unmount
 - mount again and do the same locking

then the second attempt at locking suffers a 30 second delay.
Unmounting and remounting causes lockd to stop and restart,
which causes it to bind to a new port.
The server still thinks the old port is valid and gets ECONNREFUSED
when trying to contact it.
ECONNREFUSED should be seen as a hard error that is not worth
retrying.  Rebinding is the only reasonable response.

This patch forces a rebind if that makes sense.

Signed-off-by: NeilBrown <neilb@suse.com>
Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
5 years agoNFS: Remove unused parameter gfp_flags from nfs_pageio_init()
Trond Myklebust [Sun, 20 Aug 2017 15:33:25 +0000 (11:33 -0400)]
NFS: Remove unused parameter gfp_flags from nfs_pageio_init()

Now that the mirror allocation has been moved, the parameter can go.
Also remove the redundant symbol export.

Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
5 years agoNFSv4: Fix up mirror allocation
Trond Myklebust [Sat, 19 Aug 2017 14:10:34 +0000 (10:10 -0400)]
NFSv4: Fix up mirror allocation

There are a number of callers of nfs_pageio_complete() that want to
continue using the nfs_pageio_descriptor without needing to call
nfs_pageio_init() again. Examples include nfs_pageio_resend() and
nfs_pageio_cond_complete().

The problem is that nfs_pageio_complete() also calls
nfs_pageio_cleanup_mirroring(), which frees up the array of mirrors.
This can lead to writeback errors, in the next call to
nfs_pageio_setup_mirroring().

Fix by simply moving the allocation of the mirrors to
nfs_pageio_setup_mirroring().

Link: https://bugzilla.kernel.org/show_bug.cgi?id=196709
Reported-by: JianhongYin <yin-jianhong@163.com>
Cc: stable@vger.kernel.org # 4.0+
Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
5 years agoMerge branch 'writeback'
Trond Myklebust [Fri, 18 Aug 2017 18:51:10 +0000 (14:51 -0400)]
Merge branch 'writeback'

5 years agoSUNRPC: Add a separate spinlock to protect the RPC request receive list
Trond Myklebust [Wed, 16 Aug 2017 19:30:35 +0000 (15:30 -0400)]
SUNRPC: Add a separate spinlock to protect the RPC request receive list

This further reduces contention with the transport_lock, and allows us
to convert to using a non-bh-safe spinlock, since the list is now never
accessed from a bh context.

Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
5 years agoSUNRPC: Cleanup xs_tcp_read_common()
Trond Myklebust [Sun, 13 Aug 2017 06:13:45 +0000 (02:13 -0400)]
SUNRPC: Cleanup xs_tcp_read_common()

Simplify the code to avoid a full copy of the struct xdr_skb_reader.

Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
5 years agoSUNRPC: Don't loop forever in xs_tcp_data_receive()
Trond Myklebust [Sat, 12 Aug 2017 19:11:49 +0000 (15:11 -0400)]
SUNRPC: Don't loop forever in xs_tcp_data_receive()

Ensure that we don't hog the workqueue thread by requeuing the job
every 64 loops.

Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
5 years agoSUNRPC: Don't hold the transport lock when receiving backchannel data
Trond Myklebust [Wed, 16 Aug 2017 16:09:44 +0000 (12:09 -0400)]
SUNRPC: Don't hold the transport lock when receiving backchannel data

The backchannel request has no associated task, so it is going nowhere
until we call xprt_complete_bc_request().

Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
5 years agoSUNRPC: Don't hold the transport lock across socket copy operations
Trond Myklebust [Sun, 13 Aug 2017 14:03:59 +0000 (10:03 -0400)]
SUNRPC: Don't hold the transport lock across socket copy operations

Instead add a mechanism to ensure that the request doesn't disappear
from underneath us while copying from the socket. We do this by
preventing xprt_release() from freeing the XDR buffers until the
flag RPC_TASK_MSG_RECV has been cleared from the request.

Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
Reviewed-by: Chuck Lever <chuck.lever@oracle.com>
5 years agoxprtrdma: Remove imul instructions from chunk list encoders
Chuck Lever [Mon, 14 Aug 2017 19:38:30 +0000 (15:38 -0400)]
xprtrdma: Remove imul instructions from chunk list encoders

Re-arrange the pointer arithmetic in the chunk list encoders to
eliminate several more integer multiplication instructions during
Transport Header encoding.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
5 years agoxprtrdma: Remove imul instructions from rpcrdma_convert_iovs()
Chuck Lever [Mon, 14 Aug 2017 19:38:22 +0000 (15:38 -0400)]
xprtrdma: Remove imul instructions from rpcrdma_convert_iovs()

Re-arrange the pointer arithmetic in rpcrdma_convert_iovs() to
eliminate several integer multiplication instructions during
Transport Header encoding.

Also, array overflow does not occur outside development
environments, so replace overflow checking with one spot check
at the end. This reduces the number of conditional branches in
the common case.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
5 years agoNFS: Wait for requests that are locked on the commit list
Trond Myklebust [Tue, 1 Aug 2017 21:29:29 +0000 (17:29 -0400)]
NFS: Wait for requests that are locked on the commit list

If a request is on the commit list, but is locked, we will currently skip
it, which can lead to livelocking when the commit count doesn't reduce
to zero.

Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
5 years agoNFSv4/pnfs: Replace pnfs_put_lseg_locked() with pnfs_put_lseg()
Trond Myklebust [Tue, 1 Aug 2017 21:07:02 +0000 (17:07 -0400)]
NFSv4/pnfs: Replace pnfs_put_lseg_locked() with pnfs_put_lseg()

Now that we no longer hold the inode->i_lock when manipulating the
commit lists, it is safe to call pnfs_put_lseg() again.

Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
5 years agoNFS: Switch to using mapping->private_lock for page writeback lookups.
Trond Myklebust [Tue, 1 Aug 2017 16:34:44 +0000 (12:34 -0400)]
NFS: Switch to using mapping->private_lock for page writeback lookups.

Switch from using the inode->i_lock for this to avoid contention with
other metadata manipulation.

Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
5 years agoNFS: Use an atomic_long_t to count the number of commits
Trond Myklebust [Tue, 1 Aug 2017 21:04:12 +0000 (17:04 -0400)]
NFS: Use an atomic_long_t to count the number of commits

Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
5 years agoNFS: Use an atomic_long_t to count the number of requests
Trond Myklebust [Tue, 1 Aug 2017 19:39:46 +0000 (15:39 -0400)]
NFS: Use an atomic_long_t to count the number of requests

Rather than forcing us to take the inode->i_lock just in order to bump
the number.

Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
5 years agoNFSv4: Use a mutex to protect the per-inode commit lists
Trond Myklebust [Tue, 1 Aug 2017 15:53:49 +0000 (11:53 -0400)]
NFSv4: Use a mutex to protect the per-inode commit lists

The commit lists can get very large, so using the inode->i_lock can
end up affecting general metadata performance.

Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
5 years agoNFS: Refactor nfs_page_find_head_request()
Trond Myklebust [Tue, 1 Aug 2017 16:26:53 +0000 (12:26 -0400)]
NFS: Refactor nfs_page_find_head_request()

Split out the 2 cases so that we can treat the locking differently.
The issue is that the locking in the pageswapcache cache is highly
linked to the commit list locking.

Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
5 years agoNFSv4: Convert nfs_lock_and_join_requests() to use nfs_page_find_head_request()
Trond Myklebust [Tue, 1 Aug 2017 16:06:30 +0000 (12:06 -0400)]
NFSv4: Convert nfs_lock_and_join_requests() to use nfs_page_find_head_request()

Hide the locking from nfs_lock_and_join_requests() so that we can
separate out the requirements for swapcache pages.

Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
5 years agoNFS: Fix up nfs_page_group_covers_page()
Trond Myklebust [Mon, 17 Jul 2017 20:58:07 +0000 (16:58 -0400)]
NFS: Fix up nfs_page_group_covers_page()

Fix up the test in nfs_page_group_covers_page(). The simplest implementation
is to check that we have a set of intersecting or contiguous subrequests
that connect page offset 0 to nfs_page_length(req->wb_page).

Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
5 years agoNFS: Remove unused parameter from nfs_page_group_lock()
Trond Myklebust [Mon, 17 Jul 2017 14:54:14 +0000 (10:54 -0400)]
NFS: Remove unused parameter from nfs_page_group_lock()

nfs_page_group_lock() is now always called with the 'nonblock'
parameter set to 'false'.

Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
5 years agoNFS: Remove unuse function nfs_page_group_lock_wait()
Trond Myklebust [Mon, 17 Jul 2017 14:51:02 +0000 (10:51 -0400)]
NFS: Remove unuse function nfs_page_group_lock_wait()

Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
5 years agoNFS: Remove nfs_page_group_clear_bits()
Trond Myklebust [Wed, 19 Jul 2017 17:50:07 +0000 (13:50 -0400)]
NFS: Remove nfs_page_group_clear_bits()

At this point, we only expect ever to potentially see PG_REMOVE and
PG_TEARDOWN being set on the subrequests.

Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
5 years agoNFS: Fix nfs_page_group_destroy() and nfs_lock_and_join_requests() race cases
Trond Myklebust [Wed, 19 Jul 2017 14:06:36 +0000 (10:06 -0400)]
NFS: Fix nfs_page_group_destroy() and nfs_lock_and_join_requests() race cases

Since nfs_page_group_destroy() does not take any locks on the requests
to be freed, we need to ensure that we don't inadvertently free the
request in nfs_destroy_unlinked_subrequests() while the last reference
is being released elsewhere.

Do this by:

1) Taking a reference to the request unless it is already being freed
2) Checking (under the page group lock) if PG_TEARDOWN is already set before
   freeing an unreferenced request in nfs_destroy_unlinked_subrequests()

Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
5 years agoNFS: Further optimise nfs_lock_and_join_requests()
Trond Myklebust [Wed, 19 Jul 2017 12:23:10 +0000 (08:23 -0400)]
NFS: Further optimise nfs_lock_and_join_requests()

When locking the entire group in order to remove subrequests,
the locks are always taken in order, and with the page group
lock being taken after the page head is locked. The intention
is that:

1) The lock on the group head guarantees that requests may not
   be removed from the group (although new entries could be appended
   if we're not holding the group lock).
2) It is safe to drop and retake the page group lock while iterating
   through the list, in particular when waiting for a subrequest lock.

Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
5 years agoNFS: Reduce inode->i_lock contention in nfs_lock_and_join_requests()
Trond Myklebust [Mon, 17 Jul 2017 14:34:21 +0000 (10:34 -0400)]
NFS: Reduce inode->i_lock contention in nfs_lock_and_join_requests()

We should no longer need the inode->i_lock, now that we've
straightened out the request locking. The locking schema is now:

1) Lock page head request
2) Lock the page group
3) Lock the subrequests one by one

Note that there is a subtle race with nfs_inode_remove_request() due
to the fact that the latter does not lock the page head, when removing
it from the struct page. Only the last subrequest is locked, hence
we need to re-check that the PagePrivate(page) is still set after
we've locked all the subrequests.

Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
5 years agoNFS: Remove page group limit in nfs_flush_incompatible()
Trond Myklebust [Tue, 18 Jul 2017 00:00:46 +0000 (20:00 -0400)]
NFS: Remove page group limit in nfs_flush_incompatible()

nfs_try_to_update_request() should be able to cope now.

Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
5 years agoNFS: Teach nfs_try_to_update_request() to deal with request page_groups
Trond Myklebust [Mon, 17 Jul 2017 23:50:23 +0000 (19:50 -0400)]
NFS: Teach nfs_try_to_update_request() to deal with request page_groups

Simplify the code, and avoid some flushes to disk.

Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
5 years agoNFS: Fix the inode request accounting when pages have subrequests
Trond Myklebust [Tue, 18 Jul 2017 19:22:12 +0000 (15:22 -0400)]
NFS: Fix the inode request accounting when pages have subrequests

Both nfs_destroy_unlinked_subrequests() and nfs_lock_and_join_requests()
manipulate the inode flags adjusting the NFS_I(inode)->nrequests.

Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
5 years agoNFS: Don't unlock writebacks before declaring PG_WB_END
Trond Myklebust [Tue, 18 Jul 2017 23:18:49 +0000 (19:18 -0400)]
NFS: Don't unlock writebacks before declaring PG_WB_END

We don't want nfs_lock_and_join_requests() to start fiddling with
the request before the call to nfs_page_group_sync_on_bit().

Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
5 years agoNFS: Don't check request offset and size without holding a lock
Trond Myklebust [Mon, 17 Jul 2017 15:11:49 +0000 (11:11 -0400)]
NFS: Don't check request offset and size without holding a lock

Request offsets and sizes are not guaranteed to be stable unless you
are holding the request locked.

Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
5 years agoNFS: Fix an ABBA issue in nfs_lock_and_join_requests()
Trond Myklebust [Mon, 17 Jul 2017 14:29:32 +0000 (10:29 -0400)]
NFS: Fix an ABBA issue in nfs_lock_and_join_requests()

All other callers of nfs_page_group_lock() appear to already hold the
page lock on the head page, so doing it in the opposite order here
is inefficient, although not deadlock prone since we roll back all
locks on contention.

Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
5 years agoNFS: Fix a reference and lock leak in nfs_lock_and_join_requests()
Trond Myklebust [Mon, 17 Jul 2017 14:11:11 +0000 (10:11 -0400)]
NFS: Fix a reference and lock leak in nfs_lock_and_join_requests()

Yes, this is a situation that should never happen (hence the WARN_ON)
but we should still ensure that we free up the locks and references to
the faulty pages.

Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
5 years agoNFS: Ensure we always dereference the page head last
Trond Myklebust [Tue, 18 Jul 2017 23:31:10 +0000 (19:31 -0400)]
NFS: Ensure we always dereference the page head last

This fixes a race with nfs_page_group_sync_on_bit() whereby the
call to wake_up_bit() in nfs_page_group_unlock() could occur after
the page header had been freed.

Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
5 years agoNFS: Reduce lock contention in nfs_try_to_update_request()
Trond Myklebust [Mon, 17 Jul 2017 13:37:12 +0000 (09:37 -0400)]
NFS: Reduce lock contention in nfs_try_to_update_request()

Micro-optimisation to move the lockless check into the for(;;) loop.

Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
5 years agoNFS: Reduce lock contention in nfs_page_find_head_request()
Trond Myklebust [Mon, 17 Jul 2017 13:30:13 +0000 (09:30 -0400)]
NFS: Reduce lock contention in nfs_page_find_head_request()

Add a lockless check for whether or not the page might be carrying
an existing writeback before we grab the inode->i_lock.

Reported-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
5 years agoNFS: Simplify page writeback
Trond Myklebust [Sun, 9 Jul 2017 17:45:27 +0000 (13:45 -0400)]
NFS: Simplify page writeback

We don't expect the page header lock to ever be held across I/O, so
it should always be safe to wait for it, even if we're doing nonblocking
writebacks.

Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
5 years agoMerge branch 'open_state'
Trond Myklebust [Tue, 15 Aug 2017 15:54:13 +0000 (11:54 -0400)]
Merge branch 'open_state'

5 years agoNFSv4: Use the nfs4_state being recovered in _nfs4_opendata_to_nfs4_state()
Trond Myklebust [Tue, 8 Aug 2017 14:38:07 +0000 (10:38 -0400)]
NFSv4: Use the nfs4_state being recovered in _nfs4_opendata_to_nfs4_state()

If we're recovering a nfs4_state, then we should try to use that instead
of looking up a new stateid. Only do that if the inodes match, though.

Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
5 years agoNFSv4: Use correct inode in _nfs4_opendata_to_nfs4_state()
Trond Myklebust [Tue, 8 Aug 2017 13:06:18 +0000 (09:06 -0400)]
NFSv4: Use correct inode in _nfs4_opendata_to_nfs4_state()

When doing open by filehandle we don't really want to lookup a new inode,
but rather update the one we've got. Add a helper which does this for us.

Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
5 years agoLinux 4.13-rc5 v4.13-rc5
Linus Torvalds [Sun, 13 Aug 2017 23:01:32 +0000 (16:01 -0700)]
Linux 4.13-rc5

5 years agoMerge branch 'upstream' of git://git.linux-mips.org/pub/scm/ralf/upstream-linus
Linus Torvalds [Sun, 13 Aug 2017 22:34:28 +0000 (15:34 -0700)]
Merge branch 'upstream' of git://git.linux-mips.org/ralf/upstream-linus

Pull MIPS fixes from Ralf Baechle:
 "Another round of MIPS fixes:

   - compressed boot: Ignore a generated .c file

   - VDSO: Fix a register clobber list

   - DECstation: Fix an int-handler.S CPU_DADDI_WORKAROUNDS regression

   - Octeon: Fix recent cleanups that cleaned away a bit too much thus
     breaking the arch side of the EDAC and USB drivers.

   - uasm: Fix duplicate const in "const struct foo const bar[]" which
     GCC 7.1 no longer accepts.

   - Fix race on setting and getting cpu_online_mask

   - Fix preemption issue. To do so cleanly introduce macro to get the
     size of L3 cache line.

   - Revert include cleanup that sometimes results in build error

   - MicroMIPS uses bit 0 of the PC to indicate microMIPS mode. Make
     sure this bit is set for kernel entry as well.

   - Prevent configuring the kernel for both microMIPS and MT. There are
     no such CPUs currently and thus the combination is unsupported and
     results in build errors.

  This has been sitting in linux-next for a few days and has survived
  automated testing by Imagination's test farm. No known regressions
  pending except a number of issues that crept up due to lots of people
  switching to GCC 7.1"

* 'upstream' of git://git.linux-mips.org/pub/scm/ralf/upstream-linus:
  MIPS: Set ISA bit in entry-y for microMIPS kernels
  MIPS: Prevent building MT support for microMIPS kernels
  MIPS: PCI: Fix smp_processor_id() in preemptible
  MIPS: Introduce cpu_tcache_line_size
  MIPS: DEC: Fix an int-handler.S CPU_DADDI_WORKAROUNDS regression
  MIPS: VDSO: Fix clobber lists in fallback code paths
  Revert "MIPS: Don't unnecessarily include kmalloc.h into <asm/cache.h>."
  MIPS: OCTEON: Fix USB platform code breakage.
  MIPS: Octeon: Fix broken EDAC driver.
  MIPS: gitignore: ignore generated .c files
  MIPS: Fix race on setting and getting cpu_online_mask
  MIPS: mm: remove duplicate "const" qualifier on insn_table

5 years agoMerge tag 'driver-core-4.13-rc5' of git://git.kernel.org/pub/scm/linux/kernel/git...
Linus Torvalds [Sun, 13 Aug 2017 19:44:18 +0000 (12:44 -0700)]
Merge tag 'driver-core-4.13-rc5' of git://git./linux/kernel/git/gregkh/driver-core

Pull driver core fixes from Greg KH:
 "Here are three firmware core fixes for 4.13-rc5.

  All three of these fix reported issues and have been floating around
  for a few weeks. They have been in linux-next with no reported
  problems"

* tag 'driver-core-4.13-rc5' of git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/driver-core:
  firmware: avoid invalid fallback aborts by using killable wait
  firmware: fix batched requests - send wake up on failure on direct lookups
  firmware: fix batched requests - wake all waiters

5 years agoMerge tag 'char-misc-4.13-rc5' of git://git.kernel.org/pub/scm/linux/kernel/git/gregk...
Linus Torvalds [Sun, 13 Aug 2017 19:41:58 +0000 (12:41 -0700)]
Merge tag 'char-misc-4.13-rc5' of git://git./linux/kernel/git/gregkh/char-misc

Pull char/misc fixes from Greg KH:
 "Here are two patches for 4.13-rc5.

  One is a fix for a reported thunderbolt issue, and the other a fix for
  an MEI driver issue. Both have been in linux-next with no reported
  issues"

* tag 'char-misc-4.13-rc5' of git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/char-misc:
  thunderbolt: Do not enumerate more ports from DROM than the controller has
  mei: exclude device from suspend direct complete optimization

5 years agoMerge tag 'tty-4.13-rc5' of git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/tty
Linus Torvalds [Sun, 13 Aug 2017 19:33:35 +0000 (12:33 -0700)]
Merge tag 'tty-4.13-rc5' of git://git./linux/kernel/git/gregkh/tty

Pull tty/serial fixes from Greg KH:
 "Here are two tty serial driver fixes for 4.13-rc5. One is a revert of
  a -rc1 patch that turned out to not be a good idea, and the other is a
  fix for the pl011 serial driver.

  Both have been in linux-next with no reported issues"

* tag 'tty-4.13-rc5' of git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/tty:
  Revert "serial: Delete dead code for CIR serial ports"
  tty: pl011: fix initialization order of QDF2400 E44

5 years agoMerge tag 'staging-4.13-rc5' of git://git.kernel.org/pub/scm/linux/kernel/git/gregkh...
Linus Torvalds [Sun, 13 Aug 2017 19:30:17 +0000 (12:30 -0700)]
Merge tag 'staging-4.13-rc5' of git://git./linux/kernel/git/gregkh/staging

Pull staging/iio fixes from Greg KH:
 "Here are some Staging and IIO driver fixes for 4.13-rc5.

  Nothing major, just a number of small fixes for reported issues. All
  of these have been in linux-next for a while now with no reported
  issues. Full details are in the shortlog"

* tag 'staging-4.13-rc5' of git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/staging:
  staging: comedi: comedi_fops: do not call blocking ops when !TASK_RUNNING
  iio: aspeed-adc: wait for initial sequence.
  iio: accel: bmc150: Always restore device to normal mode after suspend-resume
  staging:iio:resolver:ad2s1210 fix negative IIO_ANGL_VEL read
  iio: adc: axp288: Fix the GPADC pin reading often wrongly returning 0
  iio: adc: vf610_adc: Fix VALT selection value for REFSEL bits
  iio: accel: st_accel: add SPI-3wire support
  iio: adc: Revert "axp288: Drop bogus AXP288_ADC_TS_PIN_CTRL register modifications"
  iio: adc: sun4i-gpadc-iio: fix unbalanced irq enable/disable
  iio: pressure: st_pressure_core: disable multiread by default for LPS22HB
  iio: light: tsl2563: use correct event code

5 years agoMerge tag 'usb-4.13-rc5' of git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/usb
Linus Torvalds [Sun, 13 Aug 2017 19:27:42 +0000 (12:27 -0700)]
Merge tag 'usb-4.13-rc5' of git://git./linux/kernel/git/gregkh/usb

Pull USB fixes from Greg KH:
 "Here are a number of small USB driver fixes and new device ids for
  4.13-rc5. There is the usual gadget driver fixes, some new quirks for
  "messy" hardware, and some new device ids.

  All have been in linux-next with no reported issues"

* tag 'usb-4.13-rc5' of git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/usb:
  USB: serial: pl2303: add new ATEN device id
  usb: quirks: Add no-lpm quirk for Moshi USB to Ethernet Adapter
  USB: Check for dropped connection before switching to full speed
  usb:xhci:Add quirk for Certain failing HP keyboard on reset after resume
  usb: renesas_usbhs: gadget: fix unused-but-set-variable warning
  usb: renesas_usbhs: Fix UGCTRL2 value for R-Car Gen3
  usb: phy: phy-msm-usb: Fix usage of devm_regulator_bulk_get()
  usb: gadget: udc: renesas_usb3: Fix usb_gadget_giveback_request() calling
  usb: dwc3: gadget: Correct ISOC DATA PIDs for short packets
  USB: serial: option: add D-Link DWM-222 device ID
  usb: musb: fix tx fifo flush handling again
  usb: core: unlink urbs from the tail of the endpoint's urb_list
  usb-storage: fix deadlock involving host lock and scsi_done
  uas: Add US_FL_IGNORE_RESIDUE for Initio Corporation INIC-3069
  USB: hcd: Mark secondary HCD as dead if the primary one died
  USB: serial: cp210x: add support for Qivicon USB ZigBee dongle

5 years agoMerge tag 'for-linus-20170812' of git://git.infradead.org/linux-mtd
Linus Torvalds [Sat, 12 Aug 2017 23:19:43 +0000 (16:19 -0700)]
Merge tag 'for-linus-20170812' of git://git.infradead.org/linux-mtd

Pull another MTD fix from Brian Norris:
 "An mtdblock regression occurred in -rc1 (all writes were broken!), in
  the process of some block subsystem refactoring. Noticed and fixed
  last week, but I'm a little slow on the uptake"

* tag 'for-linus-20170812' of git://git.infradead.org/linux-mtd:
  mtd: blkdevs: Fix mtd block write failure

5 years agomtd: blkdevs: Fix mtd block write failure
Abhishek Sahu [Wed, 2 Aug 2017 12:33:05 +0000 (18:03 +0530)]
mtd: blkdevs: Fix mtd block write failure

All the MTD block write requests are failing with
following error messages

    mkfs.ext4  /dev/mtdblock0

    print_req_error: I/O error, dev mtdblock0, sector 0
    Buffer I/O error on dev mtdblock0, logical block 0,
    lost async page write

The control is going to default case after block write request
because of missing return.

Fixes: commit 2a842acab109 ("block: introduce new block status code type")
Signed-off-by: Abhishek Sahu <absahu@codeaurora.org>
Signed-off-by: Brian Norris <computersforpeace@gmail.com>
5 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/nab/target-pending
Linus Torvalds [Sat, 12 Aug 2017 19:08:59 +0000 (12:08 -0700)]
Merge git://git./linux/kernel/git/nab/target-pending

Pull SCSI target fixes from Nicholas Bellinger:
 "The highlights include:

   - Fix iscsi-target payload memory leak during
     ISCSI_FLAG_TEXT_CONTINUE (Varun Prakash)

   - Fix tcm_qla2xxx incorrect use of tcm_qla2xxx_free_cmd during ABORT
     (Pascal de Bruijn + Himanshu Madhani + nab)

   - Fix iscsi-target long-standing issue with parallel delete of a
     single network portal across multiple target instances (Gary Guo +
     nab)

   - Fix target dynamic se_node GPF during uncached shutdown regression
     (Justin Maggard + nab)"

* git://git.kernel.org/pub/scm/linux/kernel/git/nab/target-pending:
  target: Fix node_acl demo-mode + uncached dynamic shutdown regression
  iscsi-target: Fix iscsi_np reset hung task during parallel delete
  qla2xxx: Fix incorrect tcm_qla2xxx_free_cmd use during TMR ABORT (v2)
  cxgbit: fix sg_nents calculation
  iscsi-target: fix invalid flags in text response
  iscsi-target: fix memory leak in iscsit_setup_text_cmd()
  cxgbit: add missing __kfree_skb()
  tcmu: free old string on reconfig
  tcmu: Fix possible to/from address overflow when doing the memcpy

5 years agoMerge tag 'for-linus-4.13b-rc5-tag' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Sat, 12 Aug 2017 16:01:36 +0000 (09:01 -0700)]
Merge tag 'for-linus-4.13b-rc5-tag' of git://git./linux/kernel/git/xen/tip

Pull xen fixes from Juergen Gross:
 "Some fixes for Xen:

   - a fix for a regression introduced in 4.13 for a Xen HVM-guest
     configured with KASLR

   - a fix for a possible deadlock in the xenbus driver when booting the
     system

   - a fix for lost interrupts in Xen guests"

* tag 'for-linus-4.13b-rc5-tag' of git://git.kernel.org/pub/scm/linux/kernel/git/xen/tip:
  xen/events: Fix interrupt lost during irq_disable and irq_enable
  xen: avoid deadlock in xenbus
  xen: fix hvm guest with kaslr enabled
  xen: split up xen_hvm_init_shared_info()
  x86: provide an init_mem_mapping hypervisor hook

5 years agoMerge tag 'nfs-for-4.13-5' of git://git.linux-nfs.org/projects/anna/linux-nfs
Linus Torvalds [Fri, 11 Aug 2017 20:54:09 +0000 (13:54 -0700)]
Merge tag 'nfs-for-4.13-5' of git://git.linux-nfs.org/projects/anna/linux-nfs

Pull NFS client fixes from Anna Schumaker:
 "A few more NFS client bugfixes from me for rc5.

  Dros has a stable fix for flexfiles to prevent leaking the
  nfs4_ff_ds_version arrays when freeing a layout, Trond fixed a
  potential recovery loop situation with the TEST_STATEID operation, and
  Christoph fixed up the pNFS blocklayout Kconfig options to prevent
  unsafe use with kernels that don't have large block device support.
  Summary:

  Stable fix:
   - fix leaking nfs4_ff_ds_version array

  Other fixes:
   - improve TEST_STATEID OLD_STATEID handling to prevent recovery loop

   - require 64-bit sector_t for pNFS blocklayout to prevent 32-bit
     compile errors"

* tag 'nfs-for-4.13-5' of git://git.linux-nfs.org/projects/anna/linux-nfs:
  pnfs/blocklayout: require 64-bit sector_t
  NFSv4: Ignore NFS4ERR_OLD_STATEID in nfs41_check_open_stateid()
  nfs/flexfiles: fix leak of nfs4_ff_ds_version arrays

5 years agoMerge branch 'for-linus' of git://git.kernel.dk/linux-block
Linus Torvalds [Fri, 11 Aug 2017 19:26:49 +0000 (12:26 -0700)]
Merge branch 'for-linus' of git://git.kernel.dk/linux-block

Pull block fixes from Jens Axboe:
 "A set of fixes that should go into this series. This contains:

   - Fix from Bart for blk-mq requeue queue running, preventing a
     continued loop of run/restart.

   - Fix for a bio/blk-integrity issue, in two parts. One from
     Christoph, fixing where verification happens, and one from Milan,
     for a NULL profile.

   - NVMe pull request, most of the changes being for nvme-fc, but also
     a few trivial core/pci fixes"

* 'for-linus' of git://git.kernel.dk/linux-block:
  nvme: fix directive command numd calculation
  nvme: fix nvme reset command timeout handling
  nvme-pci: fix CMB sysfs file removal in reset path
  lpfc: support nvmet_fc defer_rcv callback
  nvmet_fc: add defer_req callback for deferment of cmd buffer return
  nvme: strip trailing 0-bytes in wwid_show
  block: Make blk_mq_delay_kick_requeue_list() rerun the queue at a quiet time
  bio-integrity: only verify integrity on the lowest stacked driver
  bio-integrity: Fix regression if profile verify_fn is NULL

5 years agoMerge tag 'mmc-v4.13-rc4' of git://git.kernel.org/pub/scm/linux/kernel/git/ulfh/mmc
Linus Torvalds [Fri, 11 Aug 2017 18:56:54 +0000 (11:56 -0700)]
Merge tag 'mmc-v4.13-rc4' of git://git./linux/kernel/git/ulfh/mmc

Pull MMC fixes from Ulf Hansson:
 "MMC core:

   - fix lockdep splat when removing mmc_block module

   - fix the logic for setting eMMC HS400ES signal voltage

  MMC host:

   - omap_hsmmc: add CMD23 capability to fix -EIO errors"

* tag 'mmc-v4.13-rc4' of git://git.kernel.org/pub/scm/linux/kernel/git/ulfh/mmc:
  mmc: block: fix lockdep splat when removing mmc_block module
  mmc: mmc: correct the logic for setting HS400ES signal voltage
  mmc: host: omap_hsmmc: Add CMD23 capability to omap_hsmmc driver

5 years agoMerge tag 'fbdev-v4.13-rc5' of git://github.com/bzolnier/linux
Linus Torvalds [Fri, 11 Aug 2017 18:44:18 +0000 (11:44 -0700)]
Merge tag 'fbdev-v4.13-rc5' of git://github.com/bzolnier/linux

Pull fbdev fixes from Bartlomiej Zolnierkiewicz:

 - allow user to disable write combined mapping in efifb driver (Dave
   Airlie)

 - fix use after free bugs on driver removal in imxfb driver (Dan
   Carpenter)

 - fix unused variable warning in omapfb driver (Arnd Bergmann)

* tag 'fbdev-v4.13-rc5' of git://github.com/bzolnier/linux:
  efifb: allow user to disable write combined mapping.
  fbdev: omapfb: remove unused variable
  video: fbdev: imxfb: use after free in imxfb_remove()

5 years agoMerge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/mszeredi...
Linus Torvalds [Fri, 11 Aug 2017 18:20:48 +0000 (11:20 -0700)]
Merge branch 'for-linus' of git://git./linux/kernel/git/mszeredi/fuse

Pull fuse fixes from Miklos Szeredi:
 "Fix a few bugs in fuse"

* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/mszeredi/fuse:
  fuse: set mapping error in writepage_locked when it fails
  fuse: Dont call set_page_dirty_lock() for ITER_BVEC pages for async_dio
  fuse: initialize the flock flag in fuse_file on allocation

5 years agoMerge tag 'iommu-fixes-v4.13-rc4' of git://git.kernel.org/pub/scm/linux/kernel/git...
Linus Torvalds [Fri, 11 Aug 2017 18:15:51 +0000 (11:15 -0700)]
Merge tag 'iommu-fixes-v4.13-rc4' of git://git./linux/kernel/git/joro/iommu

Pull IOMMU fix from Joerg Roedel:
 "Fix a NULL-pointer dereference in arm_smmu_add_device"

* tag 'iommu-fixes-v4.13-rc4' of git://git.kernel.org/pub/scm/linux/kernel/git/joro/iommu:
  iommu/arm-smmu: fix null-pointer dereference in arm_smmu_add_device

5 years agopnfs/blocklayout: require 64-bit sector_t
Christoph Hellwig [Sat, 5 Aug 2017 08:59:14 +0000 (10:59 +0200)]
pnfs/blocklayout: require 64-bit sector_t

The blocklayout code does not compile cleanly for a 32-bit sector_t,
and also has no reliable checks for devices sizes, which makes it
unsafe to use with a kernel that doesn't support large block devices.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reported-by: Arnd Bergmann <arnd@arndb.de>
Fixes: 5c83746a0cf2 ("pnfs/blocklayout: in-kernel GETDEVICEINFO XDR parsing")
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
5 years agoxprtrdma: Clean up rpcrdma_bc_marshal_reply()
Chuck Lever [Thu, 10 Aug 2017 16:47:44 +0000 (12:47 -0400)]
xprtrdma: Clean up rpcrdma_bc_marshal_reply()

Same changes as in rpcrdma_marshal_req(). This removes
C-structure style encoding from the backchannel.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
5 years agoxprtrdma: Harden chunk list encoding against send buffer overflow
Chuck Lever [Thu, 10 Aug 2017 16:47:36 +0000 (12:47 -0400)]
xprtrdma: Harden chunk list encoding against send buffer overflow

While marshaling chunk lists which are variable-length XDR objects,
check for XDR buffer overflow at every step. Measurements show no
significant changes in CPU utilization.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
5 years agoxprtrdma: Set up an xdr_stream in rpcrdma_marshal_req()
Chuck Lever [Thu, 10 Aug 2017 16:47:28 +0000 (12:47 -0400)]
xprtrdma: Set up an xdr_stream in rpcrdma_marshal_req()

Initialize an xdr_stream at the top of rpcrdma_marshal_req(), and
use it to encode the fixed transport header fields. This xdr_stream
will be used to encode the chunk lists in a subsequent patch.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
5 years agoxprtrdma: Remove rpclen from rpcrdma_marshal_req
Chuck Lever [Thu, 10 Aug 2017 16:47:20 +0000 (12:47 -0400)]
xprtrdma: Remove rpclen from rpcrdma_marshal_req

Clean up: Remove a variable whose result is no longer used.
Commit 655fec6987be ("xprtrdma: Use gathered Send for large inline
messages") should have removed it.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
5 years agoxprtrdma: Clean up rpcrdma_marshal_req() synopsis
Chuck Lever [Thu, 10 Aug 2017 16:47:12 +0000 (12:47 -0400)]
xprtrdma: Clean up rpcrdma_marshal_req() synopsis

Clean up: The caller already has rpcrdma_xprt, so pass that directly
instead. And provide a documenting comment for this critical
function.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>
5 years agoMerge tag 'powerpc-4.13-6' of git://git.kernel.org/pub/scm/linux/kernel/git/powerpc...
Linus Torvalds [Fri, 11 Aug 2017 15:56:01 +0000 (08:56 -0700)]
Merge tag 'powerpc-4.13-6' of git://git./linux/kernel/git/powerpc/linux

Pull powerpc fixes from Michael Ellerman:
 "All fixes for code that went in this cycle.

   - a revert of an optimisation to the syscall exit path, which could
     lead to an oops on either older machines or machines with > 1TB of
     memory

   - disable some deep idle states if the firmware configuration for
     them fails

   - re-enable HARD/SOFT lockup detectors in defconfigs after a Kconfig
     change

   - six fairly small patches fixing bugs in our new watchdog code

  Thanks to: Gautham R Shenoy, Nicholas Piggin"

* tag 'powerpc-4.13-6' of git://git.kernel.org/pub/scm/linux/kernel/git/powerpc/linux:
  powerpc/watchdog: add locking around init/exit functions
  powerpc/watchdog: Fix marking of stuck CPUs
  powerpc/watchdog: Fix final-check recovered case
  powerpc/watchdog: Moderate touch_nmi_watchdog overhead
  powerpc/watchdog: Improve watchdog lock primitive
  powerpc: NMI IPI improve lock primitive
  powerpc/configs: Re-enable HARD/SOFT lockup detectors
  powerpc/powernv/idle: Disable LOSE_FULL_CONTEXT states when stop-api fails
  Revert "powerpc/64: Avoid restore_math call if possible in syscall exit"

5 years agoiommu/arm-smmu: fix null-pointer dereference in arm_smmu_add_device
Artem Savkov [Tue, 8 Aug 2017 10:26:02 +0000 (12:26 +0200)]
iommu/arm-smmu: fix null-pointer dereference in arm_smmu_add_device

Commit c54451a "iommu/arm-smmu: Fix the error path in arm_smmu_add_device"
removed fwspec assignment in legacy_binding path as redundant which is
wrong. It needs to be updated after fwspec initialisation in
arm_smmu_register_legacy_master() as it is dereferenced later. Without
this there is a NULL-pointer dereference panic during boot on some hosts.

Signed-off-by: Artem Savkov <asavkov@redhat.com>
Reviewed-by: Robin Murphy <robin.murphy@arm.com>
Acked-by: Will Deacon <will.deacon@arm.com>
Signed-off-by: Joerg Roedel <jroedel@suse.de>
5 years agoxen/events: Fix interrupt lost during irq_disable and irq_enable
Liu Shuo [Sat, 29 Jul 2017 16:59:57 +0000 (00:59 +0800)]
xen/events: Fix interrupt lost during irq_disable and irq_enable

Here is a device has xen-pirq-MSI interrupt. Dom0 might lost interrupt
during driver irq_disable/irq_enable. Here is the scenario,
 1. irq_disable -> disable_dynirq -> mask_evtchn(irq channel)
 2. dev interrupt raised by HW and Xen mark its evtchn as pending
 3. irq_enable -> startup_pirq -> eoi_pirq ->
    clear_evtchn(channel of irq) -> clear pending status
 4. consume_one_event process the irq event without pending bit assert
    which result in interrupt lost once
 5. No HW interrupt raising anymore.

Now use enable_dynirq for enable_pirq of xen_pirq_chip to remove
eoi_pirq when irq_enable.

Signed-off-by: Liu Shuo <shuo.a.liu@intel.com>
Reviewed-by: Boris Ostrovsky <boris.ostrovsky@oracle.com>
Signed-off-by: Juergen Gross <jgross@suse.com>
5 years agoxen: avoid deadlock in xenbus
Juergen Gross [Fri, 28 Jul 2017 14:53:55 +0000 (16:53 +0200)]
xen: avoid deadlock in xenbus

When starting the xenwatch thread a theoretical deadlock situation is
possible:

xs_init() contains:

    task = kthread_run(xenwatch_thread, NULL, "xenwatch");
    if (IS_ERR(task))
        return PTR_ERR(task);
    xenwatch_pid = task->pid;

And xenwatch_thread() does:

    mutex_lock(&xenwatch_mutex);
    ...
    event->handle->callback();
    ...
    mutex_unlock(&xenwatch_mutex);

The callback could call unregister_xenbus_watch() which does:

    ...
    if (current->pid != xenwatch_pid)
        mutex_lock(&xenwatch_mutex);
    ...

In case a watch is firing before xenwatch_pid could be set and the
callback of that watch unregisters a watch, then a self-deadlock would
occur.

Avoid this by setting xenwatch_pid in xenwatch_thread().

Signed-off-by: Juergen Gross <jgross@suse.com>
Reviewed-by: Boris Ostrovsky <boris.ostrovsky@oracle.com>
Signed-off-by: Juergen Gross <jgross@suse.com>
5 years agoMerge branch 'nvme-4.13' of git://git.infradead.org/nvme into for-linus
Jens Axboe [Fri, 11 Aug 2017 14:07:19 +0000 (08:07 -0600)]
Merge branch 'nvme-4.13' of git://git.infradead.org/nvme into for-linus

Pull NVMe fixes from Christoph:

"A few more small fixes - the fc/lpfc update is the biggest by far."

5 years agoxen: fix hvm guest with kaslr enabled
Juergen Gross [Fri, 28 Jul 2017 10:23:14 +0000 (12:23 +0200)]
xen: fix hvm guest with kaslr enabled

A Xen HVM guest running with KASLR enabled will die rather soon today
because the shared info page mapping is using va() too early. This was
introduced by commit a5d5f328b0e2baa5ee7c119fd66324eb79eeeb66 ("xen:
allocate page for shared info page from low memory").

In order to fix this use early_memremap() to get a temporary virtual
address for shared info until va() can be used safely.

Signed-off-by: Juergen Gross <jgross@suse.com>
Reviewed-by: Boris Ostrovsky <boris.ostrovsky@oracle.com>
Acked-by: Ingo Molnar <mingo@kernel.org>
Signed-off-by: Juergen Gross <jgross@suse.com>
5 years agoxen: split up xen_hvm_init_shared_info()
Juergen Gross [Fri, 28 Jul 2017 10:23:13 +0000 (12:23 +0200)]
xen: split up xen_hvm_init_shared_info()

Instead of calling xen_hvm_init_shared_info() on boot and resume split
it up into a boot time function searching for the pfn to use and a
mapping function doing the hypervisor mapping call.

Signed-off-by: Juergen Gross <jgross@suse.com>
Reviewed-by: Boris Ostrovsky <boris.ostrovsky@oracle.com>
Acked-by: Ingo Molnar <mingo@kernel.org>
Signed-off-by: Juergen Gross <jgross@suse.com>
5 years agox86: provide an init_mem_mapping hypervisor hook
Juergen Gross [Fri, 28 Jul 2017 10:23:12 +0000 (12:23 +0200)]
x86: provide an init_mem_mapping hypervisor hook

Provide a hook in hypervisor_x86 called after setting up initial
memory mapping.

This is needed e.g. by Xen HVM guests to map the hypervisor shared
info page.

Signed-off-by: Juergen Gross <jgross@suse.com>
Reviewed-by: Boris Ostrovsky <boris.ostrovsky@oracle.com>
Acked-by: Ingo Molnar <mingo@kernel.org>
Signed-off-by: Juergen Gross <jgross@suse.com>
5 years agofuse: set mapping error in writepage_locked when it fails
Jeff Layton [Thu, 25 May 2017 10:57:50 +0000 (06:57 -0400)]
fuse: set mapping error in writepage_locked when it fails

This ensures that we see errors on fsync when writeback fails.

Signed-off-by: Jeff Layton <jlayton@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Miklos Szeredi <mszeredi@redhat.com>
5 years agoMerge tag 'drm-fixes-for-v4.13-rc5' of git://people.freedesktop.org/~airlied/linux
Linus Torvalds [Fri, 11 Aug 2017 05:33:47 +0000 (22:33 -0700)]
Merge tag 'drm-fixes-for-v4.13-rc5' of git://people.freedesktop.org/~airlied/linux

Pull drm fixes from Dave Airlie:
 "Nothing too earth shattering here, it just seems like lots of little
  things all over the place.

  msm has probably the larger amount of changes, but they all seem fine,
  otherwise, some rockchip, i915, etnaviv and exynos fixes, along with
  one nouveau regression fix for some older GPUs"

* tag 'drm-fixes-for-v4.13-rc5' of git://people.freedesktop.org/~airlied/linux: (35 commits)
  drm/nouveau/disp/nv04: avoid creation of output paths
  drm: make DRM_STM default n
  drm/exynos: forbid creating framebuffers from too small GEM buffers
  drm/etnaviv: Fix off-by-one error in reloc checking
  drm/i915: fix backlight invert for non-zero minimum brightness
  drm/i915/shrinker: Wrap need_resched() inside preempt-disable
  drm/i915/perf: fix flex eu registers programming
  drm/i915: Fix out-of-bounds array access in bdw_load_gamma_lut
  drm/i915/gvt: Change the max length of mmio_reg_rw from 4 to 8
  drm/i915/gvt: Initialize MMIO Block with HW state
  drm/rockchip: vop: report error when check resource error
  drm/rockchip: vop: round_up pitches to word align
  drm/rockchip: vop: fix NV12 video display error
  drm/rockchip: vop: fix iommu page fault when resume
  drm/i915/gvt: clean workload queue if error happened
  drm/i915/gvt: change resetting to resetting_eng
  drm/msm: gpu: don't abuse dma_alloc for non-DMA allocations
  drm/msm: gpu: call qcom_mdt interfaces only for ARCH_QCOM
  drm/msm/adreno: Prevent unclocked access when retrieving timestamps
  drm/msm: Remove __user from __u64 data types
  ...

5 years agoMerge branch 'akpm' (patches from Andrew)
Linus Torvalds [Thu, 10 Aug 2017 23:20:52 +0000 (16:20 -0700)]
Merge branch 'akpm' (patches from Andrew)

Merge misc fixes from Andrew Morton:
 "21 fixes"

* emailed patches from Andrew Morton <akpm@linux-foundation.org>: (21 commits)
  userfaultfd: replace ENOSPC with ESRCH in case mm has gone during copy/zeropage
  zram: rework copy of compressor name in comp_algorithm_store()
  rmap: do not call mmu_notifier_invalidate_page() under ptl
  mm: fix list corruptions on shmem shrinklist
  mm/balloon_compaction.c: don't zero ballooned pages
  MAINTAINERS: copy virtio on balloon_compaction.c
  mm: fix KSM data corruption
  mm: fix MADV_[FREE|DONTNEED] TLB flush miss problem
  mm: make tlb_flush_pending global
  mm: refactor TLB gathering API
  Revert "mm: numa: defer TLB flush for THP migration as long as possible"
  mm: migrate: fix barriers around tlb_flush_pending
  mm: migrate: prevent racy access to tlb_flush_pending
  fault-inject: fix wrong should_fail() decision in task context
  test_kmod: fix small memory leak on filesystem tests
  test_kmod: fix the lock in register_test_dev_kmod()
  test_kmod: fix bug which allows negative values on two config options
  test_kmod: fix spelling mistake: "EMTPY" -> "EMPTY"
  userfaultfd: hugetlbfs: remove superfluous page unlock in VM_SHARED case
  mm: ratelimit PFNs busy info message
  ...

5 years agouserfaultfd: replace ENOSPC with ESRCH in case mm has gone during copy/zeropage
Mike Rapoport [Thu, 10 Aug 2017 22:24:32 +0000 (15:24 -0700)]
userfaultfd: replace ENOSPC with ESRCH in case mm has gone during copy/zeropage

When the process exit races with outstanding mcopy_atomic, it would be
better to return ESRCH error.  When such race occurs the process and
it's mm are going away and returning "no such process" to the uffd
monitor seems better fit than ENOSPC.

Link: http://lkml.kernel.org/r/1502111545-32305-1-git-send-email-rppt@linux.vnet.ibm.com
Signed-off-by: Mike Rapoport <rppt@linux.vnet.ibm.com>
Suggested-by: Michal Hocko <mhocko@suse.com>
Acked-by: Michal Hocko <mhocko@suse.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: "Dr. David Alan Gilbert" <dgilbert@redhat.com>
Cc: Pavel Emelyanov <xemul@virtuozzo.com>
Cc: Mike Kravetz <mike.kravetz@oracle.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
5 years agozram: rework copy of compressor name in comp_algorithm_store()
Matthias Kaehlcke [Thu, 10 Aug 2017 22:24:29 +0000 (15:24 -0700)]
zram: rework copy of compressor name in comp_algorithm_store()

comp_algorithm_store() passes the size of the source buffer to strlcpy()
instead of the destination buffer size.  Make it explicit that the two
buffers have the same size and use strcpy() instead of strlcpy().  The
latter can be done safely since the function ensures that the string in
the source buffer is terminated.

Link: http://lkml.kernel.org/r/20170803163350.45245-1-mka@chromium.org
Signed-off-by: Matthias Kaehlcke <mka@chromium.org>
Reviewed-by: Douglas Anderson <dianders@chromium.org>
Reviewed-by: Sergey Senozhatsky <sergey.senozhatsky@gmail.com>
Acked-by: Minchan Kim <minchan@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
5 years agormap: do not call mmu_notifier_invalidate_page() under ptl
Kirill A. Shutemov [Thu, 10 Aug 2017 22:24:27 +0000 (15:24 -0700)]
rmap: do not call mmu_notifier_invalidate_page() under ptl

MMU notifiers can sleep, but in page_mkclean_one() we call
mmu_notifier_invalidate_page() under page table lock.

Let's instead use mmu_notifier_invalidate_range() outside
page_vma_mapped_walk() loop.

[jglisse@redhat.com: try_to_unmap_one() do not call mmu_notifier under ptl]
Link: http://lkml.kernel.org/r/20170809204333.27485-1-jglisse@redhat.com
Link: http://lkml.kernel.org/r/20170804134928.l4klfcnqatni7vsc@black.fi.intel.com
Fixes: c7ab0d2fdc84 ("mm: convert try_to_unmap_one() to use page_vma_mapped_walk()")
Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Signed-off-by: Jérôme Glisse <jglisse@redhat.com>
Reported-by: axie <axie@amd.com>
Cc: Alex Deucher <alexander.deucher@amd.com>
Cc: "Writer, Tim" <Tim.Writer@amd.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
5 years agomm: fix list corruptions on shmem shrinklist
Cong Wang [Thu, 10 Aug 2017 22:24:24 +0000 (15:24 -0700)]
mm: fix list corruptions on shmem shrinklist

We saw many list corruption warnings on shmem shrinklist:

  WARNING: CPU: 18 PID: 177 at lib/list_debug.c:59 __list_del_entry+0x9e/0xc0
  list_del corruption. prev->next should be ffff9ae5694b82d8, but was ffff9ae5699ba960
  Modules linked in: intel_rapl sb_edac edac_core x86_pkg_temp_thermal coretemp iTCO_wdt iTCO_vendor_support crct10dif_pclmul crc32_pclmul ghash_clmulni_intel raid0 dcdbas shpchp wmi hed i2c_i801 ioatdma lpc_ich i2c_smbus acpi_cpufreq tcp_diag inet_diag sch_fq_codel ipmi_si ipmi_devintf ipmi_msghandler igb ptp crc32c_intel pps_core i2c_algo_bit i2c_core dca ipv6 crc_ccitt
  CPU: 18 PID: 177 Comm: kswapd1 Not tainted 4.9.34-t3.el7.twitter.x86_64 #1
  Hardware name: Dell Inc. PowerEdge C6220/0W6W6G, BIOS 2.2.3 11/07/2013
  Call Trace:
    dump_stack+0x4d/0x66
    __warn+0xcb/0xf0
    warn_slowpath_fmt+0x4f/0x60
    __list_del_entry+0x9e/0xc0
    shmem_unused_huge_shrink+0xfa/0x2e0
    shmem_unused_huge_scan+0x20/0x30
    super_cache_scan+0x193/0x1a0
    shrink_slab.part.41+0x1e3/0x3f0
    shrink_slab+0x29/0x30
    shrink_node+0xf9/0x2f0
    kswapd+0x2d8/0x6c0
    kthread+0xd7/0xf0
    ret_from_fork+0x22/0x30

  WARNING: CPU: 23 PID: 639 at lib/list_debug.c:33 __list_add+0x89/0xb0
  list_add corruption. prev->next should be next (ffff9ae5699ba960), but was ffff9ae5694b82d8. (prev=ffff9ae5694b82d8).
  Modules linked in: intel_rapl sb_edac edac_core x86_pkg_temp_thermal coretemp iTCO_wdt iTCO_vendor_support crct10dif_pclmul crc32_pclmul ghash_clmulni_intel raid0 dcdbas shpchp wmi hed i2c_i801 ioatdma lpc_ich i2c_smbus acpi_cpufreq tcp_diag inet_diag sch_fq_codel ipmi_si ipmi_devintf ipmi_msghandler igb ptp crc32c_intel pps_core i2c_algo_bit i2c_core dca ipv6 crc_ccitt
  CPU: 23 PID: 639 Comm: systemd-udevd Tainted: G        W       4.9.34-t3.el7.twitter.x86_64 #1
  Hardware name: Dell Inc. PowerEdge C6220/0W6W6G, BIOS 2.2.3 11/07/2013
  Call Trace:
    dump_stack+0x4d/0x66
    __warn+0xcb/0xf0
    warn_slowpath_fmt+0x4f/0x60
    __list_add+0x89/0xb0
    shmem_setattr+0x204/0x230
    notify_change+0x2ef/0x440
    do_truncate+0x5d/0x90
    path_openat+0x331/0x1190
    do_filp_open+0x7e/0xe0
    do_sys_open+0x123/0x200
    SyS_open+0x1e/0x20
    do_syscall_64+0x61/0x170
    entry_SYSCALL64_slow_path+0x25/0x25

The problem is that shmem_unused_huge_shrink() moves entries from the
global sbinfo->shrinklist to its local lists and then releases the
spinlock.  However, a parallel shmem_setattr() could access one of these
entries directly and add it back to the global shrinklist if it is
removed, with the spinlock held.

The logic itself looks solid since an entry could be either in a local
list or the global list, otherwise it is removed from one of them by
list_del_init().  So probably the race condition is that, one CPU is in
the middle of INIT_LIST_HEAD() but the other CPU calls list_empty()
which returns true too early then the following list_add_tail() sees a
corrupted entry.

list_empty_careful() is designed to fix this situation.

[akpm@linux-foundation.org: add comments]
Link: http://lkml.kernel.org/r/20170803054630.18775-1-xiyou.wangcong@gmail.com
Fixes: 779750d20b93 ("shmem: split huge pages beyond i_size under memory pressure")
Signed-off-by: Cong Wang <xiyou.wangcong@gmail.com>
Acked-by: Linus Torvalds <torvalds@linux-foundation.org>
Acked-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
5 years agomm/balloon_compaction.c: don't zero ballooned pages
Wei Wang [Thu, 10 Aug 2017 22:24:21 +0000 (15:24 -0700)]
mm/balloon_compaction.c: don't zero ballooned pages

Revert commit bb01b64cfab7 ("mm/balloon_compaction.c: enqueue zero page
to balloon device")'

Zeroing ballon pages is rather time consuming, especially when a lot of
pages are in flight. E.g. 7GB worth of ballooned memory takes 2.8s with
__GFP_ZERO while it takes ~491ms without it.

The original commit argued that zeroing will help ksmd to merge these
pages on the host but this argument is assuming that the host actually
marks balloon pages for ksm which is not universally true.  So we pay
performance penalty for something that even might not be used in the end
which is wrong.  The host can zero out pages on its own when there is a
need.

[mhocko@kernel.org: new changelog text]
Link: http://lkml.kernel.org/r/1501761557-9758-1-git-send-email-wei.w.wang@intel.com
Fixes: bb01b64cfab7 ("mm/balloon_compaction.c: enqueue zero page to balloon device")
Signed-off-by: Wei Wang <wei.w.wang@intel.com>
Acked-by: Michael S. Tsirkin <mst@redhat.com>
Acked-by: Michal Hocko <mhocko@suse.com>
Cc: zhenwei.pi <zhenwei.pi@youruncloud.com>
Cc: David Hildenbrand <david@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
5 years agoMAINTAINERS: copy virtio on balloon_compaction.c
Michael S. Tsirkin [Thu, 10 Aug 2017 22:24:18 +0000 (15:24 -0700)]
MAINTAINERS: copy virtio on balloon_compaction.c

Changes to mm/balloon_compaction.c can easily break virtio, and virtio
is the only user of that interface.  Add a line to MAINTAINERS so
whoever changes that file remembers to copy us.

Link: http://lkml.kernel.org/r/1501764010-24456-1-git-send-email-mst@redhat.com
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Acked-by: Rafael Aquini <aquini@redhat.com>
Acked-by: Wei Wang <wei.w.wang@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
5 years agomm: fix KSM data corruption
Minchan Kim [Thu, 10 Aug 2017 22:24:15 +0000 (15:24 -0700)]
mm: fix KSM data corruption

Nadav reported KSM can corrupt the user data by the TLB batching
race[1].  That means data user written can be lost.

Quote from Nadav Amit:
 "For this race we need 4 CPUs:

  CPU0: Caches a writable and dirty PTE entry, and uses the stale value
  for write later.

  CPU1: Runs madvise_free on the range that includes the PTE. It would
  clear the dirty-bit. It batches TLB flushes.

  CPU2: Writes 4 to /proc/PID/clear_refs , clearing the PTEs soft-dirty.
  We care about the fact that it clears the PTE write-bit, and of
  course, batches TLB flushes.

  CPU3: Runs KSM. Our purpose is to pass the following test in
  write_protect_page():

if (pte_write(*pvmw.pte) || pte_dirty(*pvmw.pte) ||
    (pte_protnone(*pvmw.pte) && pte_savedwrite(*pvmw.pte)))

  Since it will avoid TLB flush. And we want to do it while the PTE is
  stale. Later, and before replacing the page, we would be able to
  change the page.

  Note that all the operations the CPU1-3 perform canhappen in parallel
  since they only acquire mmap_sem for read.

  We start with two identical pages. Everything below regards the same
  page/PTE.

  CPU0        CPU1        CPU2        CPU3
  ----        ----        ----        ----
  Write the same
  value on page

  [cache PTE as
   dirty in TLB]

              MADV_FREE
              pte_mkclean()

                          4 > clear_refs
                          pte_wrprotect()

                                      write_protect_page()
                                      [ success, no flush ]

                                      pages_indentical()
                                      [ ok ]

  Write to page
  different value

  [Ok, using stale
   PTE]

                                      replace_page()

  Later, CPU1, CPU2 and CPU3 would flush the TLB, but that is too late.
  CPU0 already wrote on the page, but KSM ignored this write, and it got
  lost"

In above scenario, MADV_FREE is fixed by changing TLB batching API
including [set|clear]_tlb_flush_pending.  Remained thing is soft-dirty
part.

This patch changes soft-dirty uses TLB batching API instead of
flush_tlb_mm and KSM checks pending TLB flush by using
mm_tlb_flush_pending so that it will flush TLB to avoid data lost if
there are other parallel threads pending TLB flush.

[1] http://lkml.kernel.org/r/BD3A0EBE-ECF4-41D4-87FA-C755EA9AB6BD@gmail.com

Link: http://lkml.kernel.org/r/20170802000818.4760-8-namit@vmware.com
Signed-off-by: Minchan Kim <minchan@kernel.org>
Signed-off-by: Nadav Amit <namit@vmware.com>
Reported-by: Nadav Amit <namit@vmware.com>
Tested-by: Nadav Amit <namit@vmware.com>
Reviewed-by: Andrea Arcangeli <aarcange@redhat.com>
Cc: Mel Gorman <mgorman@techsingularity.net>
Cc: Hugh Dickins <hughd@google.com>
Cc: "David S. Miller" <davem@davemloft.net>
Cc: Andy Lutomirski <luto@kernel.org>
Cc: Heiko Carstens <heiko.carstens@de.ibm.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Jeff Dike <jdike@addtoit.com>
Cc: Martin Schwidefsky <schwidefsky@de.ibm.com>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Nadav Amit <nadav.amit@gmail.com>
Cc: Rik van Riel <riel@redhat.com>
Cc: Russell King <linux@armlinux.org.uk>
Cc: Sergey Senozhatsky <sergey.senozhatsky@gmail.com>
Cc: Tony Luck <tony.luck@intel.com>
Cc: Yoshinori Sato <ysato@users.sourceforge.jp>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
5 years agomm: fix MADV_[FREE|DONTNEED] TLB flush miss problem
Minchan Kim [Thu, 10 Aug 2017 22:24:12 +0000 (15:24 -0700)]
mm: fix MADV_[FREE|DONTNEED] TLB flush miss problem

Nadav reported parallel MADV_DONTNEED on same range has a stale TLB
problem and Mel fixed it[1] and found same problem on MADV_FREE[2].

Quote from Mel Gorman:
 "The race in question is CPU 0 running madv_free and updating some PTEs
  while CPU 1 is also running madv_free and looking at the same PTEs.
  CPU 1 may have writable TLB entries for a page but fail the pte_dirty
  check (because CPU 0 has updated it already) and potentially fail to
  flush.

  Hence, when madv_free on CPU 1 returns, there are still potentially
  writable TLB entries and the underlying PTE is still present so that a
  subsequent write does not necessarily propagate the dirty bit to the
  underlying PTE any more. Reclaim at some unknown time at the future
  may then see that the PTE is still clean and discard the page even
  though a write has happened in the meantime. I think this is possible
  but I could have missed some protection in madv_free that prevents it
  happening."

This patch aims for solving both problems all at once and is ready for
other problem with KSM, MADV_FREE and soft-dirty story[3].

TLB batch API(tlb_[gather|finish]_mmu] uses [inc|dec]_tlb_flush_pending
and mmu_tlb_flush_pending so that when tlb_finish_mmu is called, we can
catch there are parallel threads going on.  In that case, forcefully,
flush TLB to prevent for user to access memory via stale TLB entry
although it fail to gather page table entry.

I confirmed this patch works with [4] test program Nadav gave so this
patch supersedes "mm: Always flush VMA ranges affected by zap_page_range
v2" in current mmotm.

NOTE:

This patch modifies arch-specific TLB gathering interface(x86, ia64,
s390, sh, um).  It seems most of architecture are straightforward but
s390 need to be careful because tlb_flush_mmu works only if
mm->context.flush_mm is set to non-zero which happens only a pte entry
really is cleared by ptep_get_and_clear and friends.  However, this
problem never changes the pte entries but need to flush to prevent
memory access from stale tlb.

[1] http://lkml.kernel.org/r/20170725101230.5v7gvnjmcnkzzql3@techsingularity.net
[2] http://lkml.kernel.org/r/20170725100722.2dxnmgypmwnrfawp@suse.de
[3] http://lkml.kernel.org/r/BD3A0EBE-ECF4-41D4-87FA-C755EA9AB6BD@gmail.com
[4] https://patchwork.kernel.org/patch/9861621/

[minchan@kernel.org: decrease tlb flush pending count in tlb_finish_mmu]
Link: http://lkml.kernel.org/r/20170808080821.GA31730@bbox
Link: http://lkml.kernel.org/r/20170802000818.4760-7-namit@vmware.com
Signed-off-by: Minchan Kim <minchan@kernel.org>
Signed-off-by: Nadav Amit <namit@vmware.com>
Reported-by: Nadav Amit <namit@vmware.com>
Reported-by: Mel Gorman <mgorman@techsingularity.net>
Acked-by: Mel Gorman <mgorman@techsingularity.net>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Russell King <linux@armlinux.org.uk>
Cc: Tony Luck <tony.luck@intel.com>
Cc: Martin Schwidefsky <schwidefsky@de.ibm.com>
Cc: "David S. Miller" <davem@davemloft.net>
Cc: Heiko Carstens <heiko.carstens@de.ibm.com>
Cc: Yoshinori Sato <ysato@users.sourceforge.jp>
Cc: Jeff Dike <jdike@addtoit.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Andy Lutomirski <luto@kernel.org>
Cc: Hugh Dickins <hughd@google.com>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Nadav Amit <nadav.amit@gmail.com>
Cc: Rik van Riel <riel@redhat.com>
Cc: Sergey Senozhatsky <sergey.senozhatsky@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
5 years agomm: make tlb_flush_pending global
Minchan Kim [Thu, 10 Aug 2017 22:24:09 +0000 (15:24 -0700)]
mm: make tlb_flush_pending global

Currently, tlb_flush_pending is used only for CONFIG_[NUMA_BALANCING|
COMPACTION] but upcoming patches to solve subtle TLB flush batching
problem will use it regardless of compaction/NUMA so this patch doesn't
remove the dependency.

[akpm@linux-foundation.org: remove more ifdefs from world's ugliest printk statement]
Link: http://lkml.kernel.org/r/20170802000818.4760-6-namit@vmware.com
Signed-off-by: Minchan Kim <minchan@kernel.org>
Signed-off-by: Nadav Amit <namit@vmware.com>
Acked-by: Mel Gorman <mgorman@techsingularity.net>
Cc: "David S. Miller" <davem@davemloft.net>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Andy Lutomirski <luto@kernel.org>
Cc: Heiko Carstens <heiko.carstens@de.ibm.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Jeff Dike <jdike@addtoit.com>
Cc: Martin Schwidefsky <schwidefsky@de.ibm.com>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Nadav Amit <nadav.amit@gmail.com>
Cc: Rik van Riel <riel@redhat.com>
Cc: Russell King <linux@armlinux.org.uk>
Cc: Sergey Senozhatsky <sergey.senozhatsky@gmail.com>
Cc: Tony Luck <tony.luck@intel.com>
Cc: Yoshinori Sato <ysato@users.sourceforge.jp>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
5 years agomm: refactor TLB gathering API
Minchan Kim [Thu, 10 Aug 2017 22:24:05 +0000 (15:24 -0700)]
mm: refactor TLB gathering API

This patch is a preparatory patch for solving race problems caused by
TLB batch.  For that, we will increase/decrease TLB flush pending count
of mm_struct whenever tlb_[gather|finish]_mmu is called.

Before making it simple, this patch separates architecture specific part
and rename it to arch_tlb_[gather|finish]_mmu and generic part just
calls it.

It shouldn't change any behavior.

Link: http://lkml.kernel.org/r/20170802000818.4760-5-namit@vmware.com
Signed-off-by: Minchan Kim <minchan@kernel.org>
Signed-off-by: Nadav Amit <namit@vmware.com>
Acked-by: Mel Gorman <mgorman@techsingularity.net>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Russell King <linux@armlinux.org.uk>
Cc: Tony Luck <tony.luck@intel.com>
Cc: Martin Schwidefsky <schwidefsky@de.ibm.com>
Cc: "David S. Miller" <davem@davemloft.net>
Cc: Heiko Carstens <heiko.carstens@de.ibm.com>
Cc: Yoshinori Sato <ysato@users.sourceforge.jp>
Cc: Jeff Dike <jdike@addtoit.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Andy Lutomirski <luto@kernel.org>
Cc: Hugh Dickins <hughd@google.com>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Nadav Amit <nadav.amit@gmail.com>
Cc: Rik van Riel <riel@redhat.com>
Cc: Sergey Senozhatsky <sergey.senozhatsky@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>