This patch addresses the fact that there are drivers, specifically tun,
that will call into the network page fragment allocators with buffer sizes
that are not cache aligned. Doing this could result in data alignment
and DMA performance issues as these fragment pools are also shared with the
skb allocator and any other devices that will use napi_alloc_frags or
netdev_alloc_frags.

Fixes: ffde7328 ("net: Split netdev_alloc_frag into __alloc_page_frag and add __napi_alloc_frag")
Reported-by: Jann Horn <jannh@google.com>
Signed-off-by: Alexander Duyck <alexander.h.duyck@linux.intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

This patch replaces the size + 1 value introduced with the recent fix for 1
byte allocs with a constant value.

The idea here is to reduce code overhead as the previous logic would have
to read size into a register, then increment it, and write it back to
whatever field was being used. By using a constant we can avoid those
memory reads and arithmetic operations in favor of just encoding the
maximum value into the operation itself.

Fixes: 2c2ade81 ("mm: page_alloc: fix ref bias in page_frag_alloc() for 1-byte allocs")
Signed-off-by: Alexander Duyck <alexander.h.duyck@linux.intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Eric Dumazet says:

====================
tcp: fix possible crash in tcp_v4_err()

soukjin bae reported a crash in tcp_v4_err() that we
root caused to a missing initialization.

Second patch adds a sanity check in tcp_v4_err() to avoid
future potential problems. Ignoring an ICMP message
is probably better than crashing a machine.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>

ICMP handlers are not very often stressed, we should
make them more resilient to bugs that might surface in
the future.

If there is no packet in retransmit queue, we should
avoid a NULL deref.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reported-by: soukjin bae <soukjin.bae@samsung.com>
Acked-by: Neal Cardwell <ncardwell@google.com>
Acked-by: Soheil Hassas Yeganeh <soheil@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

soukjin bae reported a crash in tcp_v4_err() handling
ICMP_DEST_UNREACH after tcp_write_queue_head(sk)
returned a NULL pointer.

Current logic should have prevented this :

  if (seq != tp->snd_una  || !icsk->icsk_retransmits ||
      !icsk->icsk_backoff || fastopen)
      break;

Problem is the write queue might have been purged
and icsk_backoff has not been cleared.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reported-by: soukjin bae <soukjin.bae@samsung.com>
Acked-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

If mv643xx_eth_shared_of_probe() fails, mv643xx_eth_shared_probe()
leaves clk enabled.

Found by Linux Driver Verification project (linuxtesting.org).
Signed-off-by: Alexey Khoroshilov <khoroshilov@ispras.ru>
Signed-off-by: David S. Miller <davem@davemloft.net>

The 1199:68C0 USB ID is reused by Sierra WP7607 which requires the DTR
quirk to be detected. Apply QMI_QUIRK_SET_DTR unconditionally as
already done for other IDs shared between different devices.
Signed-off-by: Beniamino Galvani <bgalvani@redhat.com>
Acked-by: Bjørn Mork <bjorn@mork.no>
Signed-off-by: David S. Miller <davem@davemloft.net>

GMAC IP is little-endian and used on several kind of CPU (big or little
endian). Main callbacks functions of the stmmac drivers take care about
it. It was not the case for dwmac4_get_timestamp function.

Fixes: ba1ffd74 ("stmmac: fix PTP support for GMAC4")
Signed-off-by: Alexandre Torgue <alexandre.torgue@st.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

MSG_ZEROCOPY implementation for UDP was merged in v5.0,
6e360f73 ("Merge branch 'udp-msg_zerocopy'").
Signed-off-by: Petr Vorel <pvorel@suse.cz>
Signed-off-by: David S. Miller <davem@davemloft.net>

The function-local variable "delay" enters the loop interpreted as delay
in bits. However, inside the loop it gets overwritten by the result of
mlxsw_sp_pg_buf_delay_get(), and thus leaves the loop as quantity in
cells. Thus on second and further loop iterations, the headroom for a
given priority is configured with a wrong size.

Fix by introducing a loop-local variable, delay_cells. Rename thres to
thres_cells for consistency.

Fixes: f417f04d ("mlxsw: spectrum: Refactor port buffer configuration")
Signed-off-by: Petr Machata <petrm@mellanox.com>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Alexei Starovoitov says:

====================
pull-request: bpf 2019-02-16

The following pull-request contains BPF updates for your *net* tree.

The main changes are:

1) fix lockdep false positive in bpf_get_stackid(), from Alexei.

2) several AF_XDP fixes, from Bjorn, Magnus, Davidlohr.

3) fix narrow load from struct bpf_sock, from Martin.

4) mips JIT fixes, from Paul.

5) gso handling fix in bpf helpers, from Willem.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>

Fixes: 3b89ea9c ("net: Fix for_each_netdev_feature on Big endian")
Suggested-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Florian Fainelli says:

====================
net: dsa: b53: VLAN and L2 fixes

This patch series contains a collection of fixes to the b53 driver in
order to:

- consistently program the same default VLAN ID when a port is bridged
  or not
- properly account for VLAN filtering being turned on/off and turning
  on ingress VID checking accordingly
- have SYSTEMPORT properly forward BPDU frames to the network stack
  (which it did not)
- do not assume that WoL is supported by the DSA master network device
  we are connected to
====================
Signed-off-by: David S. Miller <davem@davemloft.net>

The CPU port is special and does not need to obey VLAN restrictions as
far as untagged traffic goes, also, having the CPU port be part of a
particular PVID is against the idea of keeping it tagged in all VLANs.

Fixes: ca893194 ("net: dsa: b53: Keep CPU port as tagged in all VLANs")
Signed-off-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

We assume in the bcm_sf2 driver that the DSA master network device
supports ethtool_ops::{get,set}_wol operations, which is not a given.
Avoid de-referencing potentially non-existent function pointers and
check them as we should.

Fixes: 96e65d7f ("net: dsa: bcm_sf2: add support for Wake-on-LAN")
Signed-off-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

SYSTEMPORT has its RXCHK parser block that attempts to validate the
packet structures, unfortunately setting the L2 header check bit will
cause Bridge PDUs (BPDUs) to be incorrectly rejected because they look
like LLC/SNAP packets with a non-IPv4 or non-IPv6 Ethernet Type.

Fixes: 4e8aedfe78c7 ("net: systemport: Turn on offloads by default")
Signed-off-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

VLAN filtering can be built into the kernel, and also dynamically turned
on/off through the bridge master device. Allow re-configuring the switch
appropriately to account for that by deciding whether VLAN table
(v_table) misses should lead to a drop or forward.

Fixes: a2482d2c ("net: dsa: b53: Plug in VLAN support")
Signed-off-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

We were not consistent in how the default VID of a given port was
defined, b53_br_leave() would make sure the VLAN ID would be either 0/1
depending on the switch generation, but b53_configure_vlan(), which is
the default configuration would unconditionally set it to 1. The correct
value is 1 for 5325/5365 series and 0 otherwise. To avoid repeating that
mistake ever again, introduce a helper function: b53_default_pvid() to
factor that out.

Fixes: 967dd82f ("net: dsa: b53: Add support for Broadcom RoboSwitch")
Signed-off-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Syzkaller again found a path to a kernel crash through bad gso input.
By building an excessively large packet to cause an skb field to wrap.

If VIRTIO_NET_HDR_F_NEEDS_CSUM was set this would have been dropped in
skb_partial_csum_set.

GSO packets that do not set checksum offload are suspicious and rare.
Most callers of virtio_net_hdr_to_skb already pass them to
skb_probe_transport_header.

Move that test forward, change it to detect parse failure and drop
packets on failure as those cleary are not one of the legitimate
VIRTIO_NET_HDR_GSO types.

Fixes: bfd5f4a3 ("packet: Add GSO/csum offload support.")
Fixes: f43798c2 ("tun: Allow GSO using virtio_net_hdr")
Reported-by: syzbot <syzkaller@googlegroups.com>
Signed-off-by: Willem de Bruijn <willemb@google.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

The features attribute is of type u64 and stored in the native endianes on
the system. The for_each_set_bit() macro takes a pointer to a 32 bit array
and goes over the bits in this area. On little Endian systems this also
works with an u64 as the most significant bit is on the highest address,
but on big endian the words are swapped. When we expect bit 15 here we get
bit 47 (15 + 32).

This patch converts it more or less to its own for_each_set_bit()
implementation which works on 64 bit integers directly. This is then
completely in host endianness and should work like expected.

Fixes: fd867d51 ("net/core: generic support for disabling netdev features down stack")
Signed-off-by: Hauke Mehrtens <hauke.mehrtens@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Some PHY drivers like the generic one do not provide a read_status
callback on their own but rely on genphy_read_status being called
directly.

With the current code, this results in a NULL function pointer call.
Call genphy_read_status instead when there is no specific callback.
Signed-off-by: Paul Kocialkowski <paul.kocialkowski@bootlin.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

After commit c706863b ("net: ip6_gre: always reports o_key to
userspace"), ip6gre and ip6gretap tunnels started reporting TUNNEL_KEY
output flag even if it is not configured.
ip6gre_fill_info checks erspan_ver value to add TUNNEL_KEY for
erspan tunnels, however in commit 84581bda ("erspan: set
erspan_ver to 1 by default when adding an erspan dev")
erspan_ver is initialized to 1 even for ip6gre or ip6gretap
Fix the issue moving erspan_ver initialization in a dedicated routine

Fixes: c706863b ("net: ip6_gre: always reports o_key to userspace")
Signed-off-by: Lorenzo Bianconi <lorenzo.bianconi@redhat.com>
Reviewed-by: Greg Rose <gvrose8192@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Merge tag 'mac80211-for-davem-2019-02-15' of git://git.kernel.org/pub/scm/linux/kernel/git/jberg/mac80211

Johannes Berg says:

====================
Just a few fixes this time:
 * mesh rhashtable fixes from Herbert
 * a small error path fix when starting AP interfaces
====================
Signed-off-by: David S. Miller <davem@davemloft.net>

REG_32BIT_ZERO_EX and REG_64BIT are always handled in exactly the same
way, and reg_val_propagate_range() never actually sets any register to
type REG_32BIT_ZERO_EX.

Remove the redundant & unused REG_32BIT_ZERO_EX.
Signed-off-by: Paul Burton <paul.burton@mips.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>

The function prototype used to call JITed eBPF code (ie. the type of the
struct bpf_prog bpf_func field) returns an unsigned int. The MIPS n64
ABI that MIPS64 kernels target defines that 32 bit integers should
always be sign extended when passed in registers as either arguments or
return values.

This means that when returning any value which may not already be sign
extended (ie. of type REG_64BIT or REG_32BIT_ZERO_EX) we need to perform
that sign extension in order to comply with the n64 ABI. Without this we
see strange looking test failures from test_bpf.ko, such as:

  test_bpf: #65 ALU64_MOV_X:
    dst = 4294967295 jited:1 ret -1 != -1 FAIL (1 times)

Although the return value printed matches the expected value, this is
only because printf is only examining the least significant 32 bits of
the 64 bit register value we returned. The register holding the expected
value is sign extended whilst the v0 register was set to a zero extended
value by our JITed code, so when compared by a conditional branch
instruction the values are not equal.

We already handle this when the return value register is of type
REG_32BIT_ZERO_EX, so simply extend this to also cover REG_64BIT.
Signed-off-by: Paul Burton <paul.burton@mips.com>
Fixes: b6bd53f9 ("MIPS: Add missing file for eBPF JIT.")
Cc: stable@vger.kernel.org # v4.13+
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>

Pull block fixes from Jens Axboe:

 - Ensure we insert into the hctx dispatch list, if a request is marked
   as DONTPREP (Jianchao)

 - NVMe pull request, single missing unlock on error fix (Keith)

 - MD pull request, single fix for a potentially data corrupting issue
   (Nate)

 - Floppy check_events regression fix (Yufen)

* tag 'for-linus-20190215' of git://git.kernel.dk/linux-block:
  md/raid1: don't clear bitmap bits on interrupted recovery.
  floppy: check_events callback should not return a negative number
  nvme-pci: add missing unlock for reset error
  blk-mq: insert rq with DONTPREP to hctx dispatch list when requeue

Merge tag 'for-5.0/dm-fixes-3' of git://git.kernel.org/pub/scm/linux/kernel/git/device-mapper/linux-dm

Pull device mapper fixes from Mike Snitzer:

 - Fix bug in DM crypt's sizing of its block integrity tag space,
   resulting in less memory use when DM crypt layers on DM integrity.

 - Fix a long-standing DM thinp crash consistency bug that was due to
   improper handling of FUA. This issue is specific to writes that fill
   an entire thinp block which needs to be allocated.

* tag 'for-5.0/dm-fixes-3' of git://git.kernel.org/pub/scm/linux/kernel/git/device-mapper/linux-dm:
  dm thin: fix bug where bio that overwrites thin block ignores FUA
  dm crypt: don't overallocate the integrity tag space

Pull MMC fixes from Ulf Hansson:
 "A couple of MMC fixes intended for v5.0-rc7.

  MMC core:
   - Fix deadlock bug for block I/O requests

  MMC host:
   - sunxi: Disable broken HS-DDR mode for H5 by default
   - sunxi: Avoid unsupported speed modes declared via DT
   - meson-gx: Restore interrupt name"

* tag 'mmc-v5.0-rc5' of git://git.kernel.org/pub/scm/linux/kernel/git/ulfh/mmc:
  mmc: meson-gx: fix interrupt name
  mmc: block: handle complete_work on separate workqueue
  mmc: sunxi: Filter out unsupported modes declared in the device tree
  mmc: sunxi: Disable HS-DDR mode for H5 eMMC controller by default

Pull drm fixes from Dave Airlie:
 "Usual pull request, little larger than I'd like but nothing too
  strange in it. Willy found an bug in the lease ioctl calculations, but
  it's a drm master only ioctl which makes it harder to mess with.

 i915:
   - combo phy programming fix
   - opregion version check fix for VBT RVDA lookup
   - gem mmap ioctl race fix
   - fbdev hpd during suspend fix
   - array size bounds check fix in pmu

  amdgpu:
   - Vega20 psp fix
   - Add vrr range to debugfs for freesync debugging

  sched:
   - Scheduler race fix

  vkms:
   - license header fixups

  imx:
   - Fix CSI register offsets for i.MX51 and i.MX53.
   - Fix delayed page flip completion events on i.MX6QP due to
     unexpected behaviour of the PRE when issuing NOP buffer updates to
     the same buffer address.
   - Stop throwing errors for plane updates on disabled CRTCs when a
     userspace process is killed while a plane update is pending.
   - Add missing of_node_put cleanup in imx_ldb_bind"

* tag 'drm-fixes-2019-02-15-1' of git://anongit.freedesktop.org/drm/drm:
  drm: Use array_size() when creating lease
  drm/amdgpu/psp11: TA firmware is optional (v3)
  drm/i915/opregion: rvda is relative from opregion base in opregion 2.1+
  drm/i915/opregion: fix version check
  drm/i915: Prevent a race during I915_GEM_MMAP ioctl with WC set
  drm/i915: Block fbdev HPD processing during suspend
  drm/i915/pmu: Fix enable count array size and bounds checking
  drm/i915/cnl: Fix CNL macros for Voltage Swing programming
  drm/i915/icl: combo port vswing programming changes per BSPEC
  drm/vkms: Fix license inconsistent
  drm/amd/display: Expose connector VRR range via debugfs
  drm/sched: Always trace the dependencies we wait on, to fix a race.
  gpu: ipu-v3: pre: don't trigger update if buffer address doesn't change
  gpu: ipu-v3: Fix CSI offsets for imx53
  drm/imx: imx-ldb: add missing of_node_puts
  gpu: ipu-v3: Fix i.MX51 CSI control registers offset
  drm/imx: ignore plane updates on disabled crtcs

Pull crypto fix from Herbert Xu:
 "This fixes a crash on resume in the ccree driver"

* 'linus' of git://git.kernel.org/pub/scm/linux/kernel/git/herbert/crypto-2.6:
  crypto: ccree - fix resume race condition on init

Pull networking fixes from David Miller:

 1) Fix MAC address setting in mac80211 pmsr code, from Johannes Berg.

 2) Probe SFP modules after being attached, from Russell King.

 3) Byte ordering bug in SMC rx_curs_confirmed code, from Ursula Braun.

 4) Revert some r8169 changes that are causing regressions, from Heiner
    Kallweit.

 5) Fix spurious connection timeouts in netfilter nat code, from Florian
    Westphal.

 6) SKB leak in tipc, from Hoang Le.

 7) Short packet checkum issue in mlx4, similar to a previous mlx5
    change, from Saeed Mahameed. The issue is that whilst padding bytes
    are usually zero, it is not guarateed and the hardware doesn't take
    the padding bytes into consideration when generating the checksum.

 8) Fix various races in cls_tcindex, from Cong Wang.

 9) Need to set stream ext to NULL before freeing in SCTP code, from Xin
    Long.

10) Fix locking in phy_is_started, from Heiner Kallweit.

* git://git.kernel.org/pub/scm/linux/kernel/git/davem/net: (54 commits)
  net: ethernet: freescale: set FEC ethtool regs version
  net: hns: Fix object reference leaks in hns_dsaf_roce_reset()
  mm: page_alloc: fix ref bias in page_frag_alloc() for 1-byte allocs
  net: phy: fix potential race in the phylib state machine
  net: phy: don't use locking in phy_is_started
  selftests: fix timestamping Makefile
  net: dsa: bcm_sf2: potential array overflow in bcm_sf2_sw_suspend()
  net: fix possible overflow in __sk_mem_raise_allocated()
  dsa: mv88e6xxx: Ensure all pending interrupts are handled prior to exit
  net: phy: fix interrupt handling in non-started states
  sctp: set stream ext to NULL after freeing it in sctp_stream_outq_migrate
  sctp: call gso_reset_checksum when computing checksum in sctp_gso_segment
  net/mlx5e: XDP, fix redirect resources availability check
  net/mlx5: Fix a compilation warning in events.c
  net/mlx5: No command allowed when command interface is not ready
  net/mlx5e: Fix NULL pointer derefernce in set channels error flow
  netfilter: nft_compat: use-after-free when deleting targets
  team: avoid complex list operations in team_nl_cmd_options_set()
  net_sched: fix two more memory leaks in cls_tcindex
  net_sched: fix a memory leak in cls_tcindex
  ...

Pull signal fix from Eric Biederman:
 "Just a single patch that restores PTRACE_EVENT_EXIT functionality that
  was accidentally broken by last weeks fixes"

* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/ebiederm/user-namespace:
  signal: Restore the stop PTRACE_EVENT_EXIT

The starting of AP interface can fail due to invalid
beacon interval, which does not match the minimum gcd
requirement set by the wifi driver. In such case, the
beacon interval of that interface gets updated with
that invalid beacon interval.

The next time that interface is brought up in AP mode,
an interface combination check is performed and the
beacon interval is taken from the previously set value.

In a case where an invalid beacon interval, i.e. a beacon
interval value which does not satisfy the minimum gcd criteria
set by the driver, is set, all the subsequent trials to
bring that interface in AP mode will fail, even if the
subsequent trials have a valid beacon interval.

To avoid this, in case of a failure in bringing up an
interface in AP mode due to interface combination error,
the interface beacon interval which is stored in bss
conf, needs to be restored with the last working value
of beacon interval.

Tested on ath10k using WCN3990.

Cc: stable@vger.kernel.org
Fixes: 0c317a02 ("cfg80211: support virtual interfaces with different beacon intervals")
Signed-off-by: Rakesh Pillai <pillair@codeaurora.org>
Signed-off-by: Johannes Berg <johannes.berg@intel.com>

When rhashtable insertion fails the mesh table code doesn't free
the now-orphan mesh path object.  This patch fixes that.

Cc: stable@vger.kernel.org
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: Johannes Berg <johannes.berg@intel.com>

The mesh table code walks over hash tables for two purposes.  First of
all it's used as part of a netlink dump process, but it is also used
for looking up entries to delete using criteria other than the hash
key.

The second purpose is directly contrary to the design specification
of rhashtable walks.  It is only meant for use by netlink dumps.

This is because rhashtable is resizable and you cannot obtain a
stable walk over it during a resize process.

In fact mesh's use of rhashtable for dumping is bogus too.  Rather
than using rhashtable walk's iterator to keep track of the current
position, it always converts the current position to an integer
which defeats the purpose of the iterator.

Therefore this patch converts all uses of rhashtable walk into a
simple linked list.

This patch also adds a new spin lock to protect the hash table
insertion/removal as well as the walk list modifications.  In fact
the previous code was buggy as the removals can race with each
other, potentially resulting in a double-free.

Cc: stable@vger.kernel.org
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: Johannes Berg <johannes.berg@intel.com>

Passing an object_count of sufficient size will make
object_count * 4 wrap around to be very small, then a later function
will happily iterate off the end of the object_ids array.  Using
array_size() will saturate at SIZE_MAX, the kmalloc() will fail and
we'll return an -ENOMEM to the norty userspace.

Fixes: 62884cd3 ("drm: Add four ioctls for managing drm mode object leases [v7]")
Signed-off-by: Matthew Wilcox <willy@infradead.org>
Acked-by: Kees Cook <keescook@chromium.org>
Acked-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Cc: <stable@vger.kernel.org> # v4.15+
Signed-off-by: Dave Airlie <airlied@redhat.com>

amdgpu:
- Vega20 psp fix
- Add vrr range to debugfs for freesync debugging

sched:
- Scheduler race fix
Signed-off-by: Dave Airlie <airlied@redhat.com>
From: Alex Deucher <alexdeucher@gmail.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190213202958.3336-1-alexander.deucher@amd.com

Merge tag 'drm-intel-fixes-2019-02-13' of git://anongit.freedesktop.org/drm/drm-intel into drm-fixes

drm/i915 fixes for v5.0-rc7:
- combo phy programming fix
- opregion version check fix for VBT RVDA lookup
- gem mmap ioctl race fix
- fbdev hpd during suspend fix
- array size bounds check fix in pmu
Signed-off-by: Dave Airlie <airlied@redhat.com>

From: Jani Nikula <jani.nikula@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/877ee3504b.fsf@intel.com

drm-misc-fixes for v5.0:
- Fix license inconsistency in vkms.
Signed-off-by: Dave Airlie <airlied@redhat.com>

From: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/812e2f53-d72a-8fba-6c8c-fde8f44cf141@linux.intel.com

When provisioning a new data block for a virtual block, either because
the block was previously unallocated or because we are breaking sharing,
if the whole block of data is being overwritten the bio that triggered
the provisioning is issued immediately, skipping copying or zeroing of
the data block.

When this bio completes the new mapping is inserted in to the pool's
metadata by process_prepared_mapping(), where the bio completion is
signaled to the upper layers.

This completion is signaled without first committing the metadata.  If
the bio in question has the REQ_FUA flag set and the system crashes
right after its completion and before the next metadata commit, then the
write is lost despite the REQ_FUA flag requiring that I/O completion for
this request must only be signaled after the data has been committed to
non-volatile storage.

Fix this by deferring the completion of overwrite bios, with the REQ_FUA
flag set, until after the metadata has been committed.

Cc: stable@vger.kernel.org
Signed-off-by: Nikos Tsironis <ntsironis@arrikto.com>
Acked-by: Joe Thornber <ejt@redhat.com>
Acked-by: Mikulas Patocka <mpatocka@redhat.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>