It's possible to remove the FB tunnel with the command 'ip link del ip6gre0' but
this is unsafe, the module always supposes that this device exists. For example,
ip6gre_tunnel_lookup() may use it unconditionally.
Let's add a rtnl handler for dellink, which will never remove the FB tunnel (we
let ip6gre_destroy_tunnels() do the job).
Introduced by commit c12b395a4664 ("gre: Support GRE over IPv6").
CC: Dmitry Kozlov <xeb@mail.ru> Signed-off-by: Nicolas Dichtel <nicolas.dichtel@6wind.com> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Sometimes, when the packet arrives at skb_mac_gso_segment()
its skb->mac_len already accounts for some of the mac lenght
headers in the packet. This seems to happen when forwarding
through and OpenSSL tunnel.
When we start looking for any vlan headers in skb_network_protocol()
we seem to ignore any of the already known mac headers and start
with an ETH_HLEN. This results in an incorrect offset, dropped
TSO frames and general slowness of the connection.
We can start counting from the known skb->mac_len
and return at least that much if all mac level headers
are known and accounted for.
Fixes: 53d6471cef17262d3ad1c7ce8982a234244f68ec (net: Account for all vlan headers in skb_mac_gso_segment) CC: Eric Dumazet <eric.dumazet@gmail.com> CC: Daniel Borkman <dborkman@redhat.com> Tested-by: Martin Filip <nexus+kernel@smoula.net> Signed-off-by: Vlad Yasevich <vyasevic@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
This reverts commit ef2820a735f7 ("net: sctp: Fix a_rwnd/rwnd management
to reflect real state of the receiver's buffer") as it introduced a
serious performance regression on SCTP over IPv4 and IPv6, though a not
as dramatic on the latter. Measurements are on 10Gbit/s with ixgbe NICs.
With the reverted patch applied, the SCTP/IPv4 performance is back
to normal on latest upstream for IPv4 and IPv6 and has same throughput
as 3.4.2 test kernel, steady and interval reports are smooth again.
Fixes: ef2820a735f7 ("net: sctp: Fix a_rwnd/rwnd management to reflect real state of the receiver's buffer") Reported-by: Peter Butler <pbutler@sonusnet.com> Reported-by: Dongsheng Song <dongsheng.song@gmail.com> Reported-by: Fengguang Wu <fengguang.wu@intel.com> Tested-by: Peter Butler <pbutler@sonusnet.com> Signed-off-by: Daniel Borkmann <dborkman@redhat.com> Cc: Matija Glavinic Pecotic <matija.glavinic-pecotic.ext@nsn.com> Cc: Alexander Sverdlin <alexander.sverdlin@nsn.com> Cc: Vlad Yasevich <vyasevich@gmail.com> Acked-by: Vlad Yasevich <vyasevich@gmail.com> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The BPF_S_ANC_NLATTR and BPF_S_ANC_NLATTR_NEST extensions fail to check
for a minimal message length before testing the supplied offset to be
within the bounds of the message. This allows the subtraction of the nla
header to underflow and therefore -- as the data type is unsigned --
allowing far to big offset and length values for the search of the
netlink attribute.
The remainder calculation for the BPF_S_ANC_NLATTR_NEST extension is
also wrong. It has the minuend and subtrahend mixed up, therefore
calculates a huge length value, allowing to overrun the end of the
message while looking for the netlink attribute.
The following three BPF snippets will trigger the bugs when attached to
a UNIX datagram socket and parsing a message with length 1, 2 or 3.
,-[ PoC for missing size check in BPF_S_ANC_NLATTR ]--
| ld #0x87654321
| ldx #42
| ld #nla
| ret a
`---
,-[ PoC for the same bug in BPF_S_ANC_NLATTR_NEST ]--
| ld #0x87654321
| ldx #42
| ld #nlan
| ret a
`---
,-[ PoC for wrong remainder calculation in BPF_S_ANC_NLATTR_NEST ]--
| ; (needs a fake netlink header at offset 0)
| ld #0
| ldx #42
| ld #nlan
| ret a
`---
Fix the first issue by ensuring the message length fulfills the minimal
size constrains of a nla header. Fix the second bug by getting the math
for the remainder calculation right.
Fixes: 4738c1db15 ("[SKFILTER]: Add SKF_ADF_NLATTR instruction") Fixes: d214c7537b ("filter: add SKF_AD_NLATTR_NEST to look for nested..") Cc: Patrick McHardy <kaber@trash.net> Cc: Pablo Neira Ayuso <pablo@netfilter.org> Signed-off-by: Mathias Krause <minipli@googlemail.com> Acked-by: Daniel Borkmann <dborkman@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Extend commit 13378cad02afc2adc6c0e07fca03903c7ada0b37
("ipv4: Change rt->rt_iif encoding.") from 3.6 to return valid
RTA_IIF on 'ip route get ... iif DEVICE' instead of rt_iif 0
which is displayed as 'iif *'.
inet_iif is not appropriate to use because skb_iif is not set.
Use the skb->dev->ifindex instead.
Signed-off-by: Julian Anastasov <ja@ssi.bg> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Plug a group_info refcount leak in ping_init.
group_info is only needed during initialization and
the code failed to release the reference on exit.
While here move grabbing the reference to a place
where it is actually needed.
Signed-off-by: Chuansheng Liu <chuansheng.liu@intel.com> Signed-off-by: Zhang Dongxing <dongxing.zhang@intel.com> Signed-off-by: xiaoming wang <xiaoming.wang@intel.com> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Before the patch, it was possible to add two times the same tunnel:
ip l a vti1 type vti remote 10.16.0.121 local 10.16.0.249 key 41
ip l a vti2 type vti remote 10.16.0.121 local 10.16.0.249 key 41
It was possible, because ip_tunnel_newlink() calls ip_tunnel_find() with the
argument dev->type, which was set only later (when calling ndo_init handler
in register_netdevice()). Let's set this type in the setup handler, which is
called before newlink handler.
Introduced by commit b9959fd3b0fa ("vti: switch to new ip tunnel code").
CC: Cong Wang <amwang@redhat.com> CC: Steffen Klassert <steffen.klassert@secunet.com> Signed-off-by: Nicolas Dichtel <nicolas.dichtel@6wind.com> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Before the patch, it was possible to add two times the same tunnel:
ip l a gre1 type gre remote 10.16.0.121 local 10.16.0.249
ip l a gre2 type gre remote 10.16.0.121 local 10.16.0.249
It was possible, because ip_tunnel_newlink() calls ip_tunnel_find() with the
argument dev->type, which was set only later (when calling ndo_init handler
in register_netdevice()). Let's set this type in the setup handler, which is
called before newlink handler.
Introduced by commit c54419321455 ("GRE: Refactor GRE tunneling code.").
CC: Pravin B Shelar <pshelar@nicira.com> Signed-off-by: Nicolas Dichtel <nicolas.dichtel@6wind.com> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
1. If br_allowed_ingress() is called by br_handle_frame_finish() and
vlan_untag() in br_allowed_ingress() fails, skb will be freed by both
vlan_untag() and br_handle_frame_finish().
2. If br_allowed_ingress() is called by br_dev_xmit() and
br_allowed_ingress() fails, the skb will not be freed.
Fix these two problems by freeing the skb in br_allowed_ingress()
if it fails.
Signed-off-by: Toshiaki Makita <makita.toshiaki@lab.ntt.co.jp> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Remove the bonding debug_fs entries when the
module initialization fails. The debug_fs
entries should be removed together with all other
already allocated resources.
Signed-off-by: Thomas Richter <tmricht@linux.vnet.ibm.com> Signed-off-by: Jay Vosburgh <j.vosburgh@gmail.com> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
For UFO (udp fragmentation offloading) skbs, gso_size is the fragment
payload size, i.e. we must not account for udp header size.
Otherwise, when using virtio drivers, a to-be-forwarded UFO GSO packet
will be needlessly fragmented in the forward path, because we think its
individual segments are too large for the outgoing link.
Fixes: fe6cc55f3a9a053 ("net: ip, ipv6: handle gso skbs in forwarding path") Cc: Eric Dumazet <eric.dumazet@gmail.com> Reported-by: Tobias Brunner <tobias@strongswan.org> Signed-off-by: Florian Westphal <fw@strlen.de> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
When l2tp driver tries to get PMTU for the tunnel destination, it uses
the pointer to struct sock that represents PPPoX socket, while it
should use the pointer that represents UDP socket of the tunnel.
Signed-off-by: Dmitry Petukhov <dmgenp@gmail.com> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
In function sctp_wake_up_waiters(), we need to involve a test
if the association is declared dead. If so, we don't have any
reference to a possible sibling association anymore and need
to invoke sctp_write_space() instead, and normally walk the
socket's associations and notify them of new wmem space. The
reason for special casing is that otherwise, we could run
into the following issue when a sctp_primitive_SEND() call
from sctp_sendmsg() fails, and tries to flush an association's
outq, i.e. in the following way:
Therefore, only walk the list in an 'optimized' way if we find
that the current association is still active. We could also use
list_del_init() in addition when we call sctp_association_free(),
but as Vlad suggests, we want to trap such bugs and thus leave
it poisoned as is.
Why is it safe to resolve the issue by testing for asoc->base.dead?
Parallel calls to sctp_sendmsg() are protected under socket lock,
that is lock_sock()/release_sock(). Only within that path under
lock held, we're setting skb/chunk owner via sctp_set_owner_w().
Eventually, chunks are freed directly by an association still
under that lock. So when traversing association list on destruction
time from sctp_wake_up_waiters() via sctp_wfree(), a different
CPU can't be running sctp_wfree() while another one calls
sctp_association_free() as both happens under the same lock.
Therefore, this can also not race with setting/testing against
asoc->base.dead as we are guaranteed for this to happen in order,
under lock. Further, Vlad says: the times we check asoc->base.dead
is when we've cached an association pointer for later processing.
In between cache and processing, the association may have been
freed and is simply still around due to reference counts. We check
asoc->base.dead under a lock, so it should always be safe to check
and not race against sctp_association_free(). Stress-testing seems
fine now, too.
Fixes: cd253f9f357d ("net: sctp: wake up all assocs if sndbuf policy is per socket") Signed-off-by: Daniel Borkmann <dborkman@redhat.com> Cc: Vlad Yasevich <vyasevic@redhat.com> Acked-by: Neil Horman <nhorman@tuxdriver.com> Acked-by: Vlad Yasevich <vyasevic@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
SCTP charges chunks for wmem accounting via skb->truesize in
sctp_set_owner_w(), and sctp_wfree() respectively as the
reverse operation. If a sender runs out of wmem, it needs to
wait via sctp_wait_for_sndbuf(), and gets woken up by a call
to __sctp_write_space() mostly via sctp_wfree().
__sctp_write_space() is being called per association. Although
we assign sk->sk_write_space() to sctp_write_space(), which
is then being done per socket, it is only used if send space
is increased per socket option (SO_SNDBUF), as SOCK_USE_WRITE_QUEUE
is set and therefore not invoked in sock_wfree().
Commit 4c3a5bdae293 ("sctp: Don't charge for data in sndbuf
again when transmitting packet") fixed an issue where in case
sctp_packet_transmit() manages to queue up more than sndbuf
bytes, sctp_wait_for_sndbuf() will never be woken up again
unless it is interrupted by a signal. However, a still
remaining issue is that if net.sctp.sndbuf_policy=0, that is
accounting per socket, and one-to-many sockets are in use,
the reclaimed write space from sctp_wfree() is 'unfairly'
handed back on the server to the association that is the lucky
one to be woken up again via __sctp_write_space(), while
the remaining associations are never be woken up again
(unless by a signal).
The effect disappears with net.sctp.sndbuf_policy=1, that
is wmem accounting per association, as it guarantees a fair
share of wmem among associations.
Therefore, if we have reclaimed memory in case of per socket
accounting, wake all related associations to a socket in a
fair manner, that is, traverse the socket association list
starting from the current neighbour of the association and
issue a __sctp_write_space() to everyone until we end up
waking ourselves. This guarantees that no association is
preferred over another and even if more associations are
taken into the one-to-many session, all receivers will get
messages from the server and are not stalled forever on
high load. This setting still leaves the advantage of per
socket accounting in touch as an association can still use
up global limits if unused by others.
Fixes: 4eb701dfc618 ("[SCTP] Fix SCTP sendbuffer accouting.") Signed-off-by: Daniel Borkmann <dborkman@redhat.com> Cc: Thomas Graf <tgraf@suug.ch> Cc: Neil Horman <nhorman@tuxdriver.com> Cc: Vlad Yasevich <vyasevic@redhat.com> Acked-by: Vlad Yasevich <vyasevic@redhat.com> Acked-by: Neil Horman <nhorman@tuxdriver.com> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The TP_printk() should never dereference any pointers, because the ring
buffer can be read at some unknown time in the future. If a device no
longer exists, it can cause a kernel oops. This also makes this
event useless when saving the ring buffer in userspaces tools such as
perf and trace-cmd.
The i915_gem_evict_vm dereferences the vm pointer which may also not
exist when the ring buffer is read sometime in the future.
Link: http://lkml.kernel.org/r/1395095198-20034-3-git-send-email-artagnon@gmail.com Reported-by: Ramkumar Ramachandra <artagnon@gmail.com> Fixes: bcccff847d1f "drm/i915: trace vm eviction instead of everything" Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
[danvet: Try to make it actually compile] Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com> Reviewed-by: Christian König <christian.koenig@amd.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com> Reviewed-by: Christian König <christian.koenig@amd.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
This needs to be done to update some of the fields in
the connector structure used by the audio code.
Noticed by several users on irc.
Signed-off-by: Alex Deucher <alexander.deucher@amd.com> Signed-off-by: Christian König <christian.koenig@amd.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Signed-off-by: Christian König <christian.koenig@amd.com> Reviewed-by: Alex Deucher <alexander.deucher@amd.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
We already check that the buffer object we're accessing is registered with
the file. Now also make sure that we can't DMA across buffer object boundaries.
v2: Code commenting update.
Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com> Reviewed-by: Jakob Bornecrantz <jakob@vmware.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The query buffers were reserved while holding the binding mutex, which
caused a circular locking dependency.
Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com> Reviewed-by: Brian Paul <brianp@vmware.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Previously, the vmwgfx_fb driver would allow users to call FBIOSET_VINFO, but it would not adjust
the FINFO properly, resulting in distorted screen rendering. The patch corrects that behaviour.
See https://bugs.gentoo.org/show_bug.cgi?id=494794 for examples.
Signed-off-by: Christopher Friedt <chrisfriedt@gmail.com> Reviewed-by: Thomas Hellstrom <thellstrom@vmware.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Some fields are missing from the event mailbox
struct definitions, which cause issues when
trying to handle some events.
Add the missing fields in order to align the
struct size (without adding actual support
for the new fields).
Reported-and-tested-by: Imre Kaloz <kaloz@openwrt.org> Fixes: 028e724 ("wl18xx: move to new firmware (wl18xx-fw-3.bin)") Signed-off-by: Eliad Peller <eliad@wizery.com> Signed-off-by: John W. Linville <linville@tuxdriver.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Commit 9e30cc9595303b27b48 removed an internal mount. This
has the side-effect that rootfs now has FSID 0. Many
userspace utilities assume that st_dev in struct stat
is never 0, so this change breaks a number of tools in
early userspace.
Since we don't know how many userspace programs are affected,
make sure that FSID is at least 1.
References: http://article.gmane.org/gmane.linux.kernel/1666905
References: http://permalink.gmane.org/gmane.linux.utilities.util-linux-ng/8557 Signed-off-by: Thomas Bächler <thomas@archlinux.org> Acked-by: Tejun Heo <tj@kernel.org> Acked-by: H. Peter Anvin <hpa@zytor.com> Tested-by: Alexandre Demers <alexandre.f.demers@gmail.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
This was part of a patch by Alexander Guller from Mellanox in 2011,
but it still isn't upstream.
Signed-off-by: Chris Mason <clm@fb.com> Acked-By: Amir Vadai <amirv@mellanox.com> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The lack of pm_runtime_resume handling for the device state leads into
device wake-up interrupts not working after a while for runtime PM.
Also, serial-omap is confused about the use of device_may_wakeup.
The checks for device_may_wakeup should only be done for suspend and
resume, not for pm_runtime_suspend and pm_runtime_resume. The wake-up
events for PM runtime should always be enabled.
The lack of pm_runtime_resume handling leads into device wake-up
interrupts not working after a while for runtime PM.
Rather than try to patch over the issue of adding complex tests to
the pm_runtime_resume, let's fix the issues properly:
1. Make serial_omap_enable_wakeup deal with all internal PM state
handling so we don't need to test for up->wakeups_enabled elsewhere.
Later on once omap3 boots in device tree only mode we can also
remove the up->wakeups_enabled flag and rely on the wake-up
interrupt enable/disable state alone.
2. Do the device_may_wakeup checks in suspend and resume only,
for runtime PM the wake-up events need to be always enabled.
3. Finally just call serial_omap_enable_wakeup and make sure we
call it also in pm_runtime_resume.
4. Note that we also have to use disable_irq_nosync as serial_omap_irq
calls pm_runtime_get_sync.
Fixes: 2a0b965cfb6e (serial: omap: Add support for optional wake-up) Signed-off-by: Tony Lindgren <tony@atomide.com> Acked-by: Felipe Balbi <balbi@ti.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
During firmware download the device expects memory addresses in
big-endian byte order. As the wIndex parameter which hold the address is
sent in little-endian byte order regardless of host byte order, we need
to use swab16 rather than cpu_to_be16.
Also make sure to handle the struct ti_i2c_desc size parameter which is
returned in little-endian byte order.
Fix driver new_id sysfs-attribute removal deadlock by making sure to
not hold any locks that the attribute operations grab when removing the
attribute.
Specifically, usb_serial_deregister holds the table mutex when
deregistering the driver, which includes removing the new_id attribute.
This can lead to a deadlock as writing to new_id increments the
attribute's active count before trying to grab the same mutex in
usb_serial_probe.
The deadlock can easily be triggered by inserting a sleep in
usb_serial_deregister and writing the id of an unbound device to new_id
during module unload.
As the table mutex (in this case) is used to prevent subdriver unload
during probe, it should be sufficient to only hold the lock while
manipulating the usb-serial driver list during deregister. A racing
probe will then either fail to find a matching subdriver or fail to get
the corresponding module reference.
Since v3.15-rc1 this also triggers the following lockdep warning:
======================================================
[ INFO: possible circular locking dependency detected ]
3.15.0-rc2 #123 Tainted: G W
-------------------------------------------------------
modprobe/190 is trying to acquire lock:
(s_active#4){++++.+}, at: [<c0167aa0>] kernfs_remove_by_name_ns+0x4c/0x94
but task is already holding lock:
(table_lock){+.+.+.}, at: [<bf004d84>] usb_serial_deregister+0x3c/0x78 [usbserial]
which lock already depends on the new lock.
the existing dependency chain (in reverse order) is:
Fix regression introduced by commit 8e493ca1767d ("USB: usb_wwan: fix
bulk-urb allocation") by making sure to require both bulk-in and out
endpoints during port probe.
The original option driver (which usb_wwan is based on) was written
under the assumption that either endpoint could be missing, but
evidently this cannot have been tested properly. Specifically, it would
handle opening a device without bulk-in (but would blow up during resume
which was implemented later), but not a missing bulk-out in write()
(although it is handled in some places such as write_room()).
Fortunately (?), the driver also got the test for missing endpoints
wrong so the urbs were in fact always allocated, although they would be
initialised using the wrong endpoint address (0) and any submission of
such an urb would fail.
The commit mentioned above fixed the test for missing endpoints but
thereby exposed the other bugs which would now generate null-pointer
exceptions rather than failed urb submissions.
The regression was introduced in v3.7, but the offending commit was also
marked for stable.
By specifying NO_UNION_NORMAL the ACM driver does only use the first two
USB interfaces (modem data & control). The AT Port, Diagnostic and NMEA
interfaces are left to the USB serial driver.
Signed-off-by: Michael Ulbricht <michael.ulbricht@systec-electronic.com> Signed-off-by: Alexander Stein <alexander.stein@systec-electronic.com> Signed-off-by: Oliver Neukum <oliver@neukum.org> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
David Vrabel identified a regression when using automatic NUMA balancing
under Xen whereby page table entries were getting corrupted due to the
use of native PTE operations. Quoting him
Xen PV guest page tables require that their entries use machine
addresses if the preset bit (_PAGE_PRESENT) is set, and (for
successful migration) non-present PTEs must use pseudo-physical
addresses. This is because on migration MFNs in present PTEs are
translated to PFNs (canonicalised) so they may be translated back
to the new MFN in the destination domain (uncanonicalised).
pte_mknonnuma(), pmd_mknonnuma(), pte_mknuma() and pmd_mknuma()
set and clear the _PAGE_PRESENT bit using pte_set_flags(),
pte_clear_flags(), etc.
In a Xen PV guest, these functions must translate MFNs to PFNs
when clearing _PAGE_PRESENT and translate PFNs to MFNs when setting
_PAGE_PRESENT.
His suggested fix converted p[te|md]_[set|clear]_flags to using
paravirt-friendly ops but this is overkill. He suggested an alternative
of using p[te|md]_modify in the NUMA page table operations but this is
does more work than necessary and would require looking up a VMA for
protections.
This patch modifies the NUMA page table operations to use paravirt
friendly operations to set/clear the flags of interest. Unfortunately
this will take a performance hit when updating the PTEs on
CONFIG_PARAVIRT but I do not see a way around it that does not break
Xen.
Signed-off-by: Mel Gorman <mgorman@suse.de> Acked-by: David Vrabel <david.vrabel@citrix.com> Tested-by: David Vrabel <david.vrabel@citrix.com> Cc: Ingo Molnar <mingo@kernel.org> Cc: Peter Anvin <hpa@zytor.com> Cc: Fengguang Wu <fengguang.wu@intel.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Steven Noonan <steven@uplinklabs.net> Cc: Rik van Riel <riel@redhat.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Andrea Arcangeli <aarcange@redhat.com> Cc: Dave Hansen <dave.hansen@intel.com> Cc: Srikar Dronamraju <srikar@linux.vnet.ibm.com> Cc: Cyrill Gorcunov <gorcunov@gmail.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
soft lockup in freeing gigantic hugepage fixed in commit 55f67141a892 "mm:
hugetlb: fix softlockup when a large number of hugepages are freed." can
happen in return_unused_surplus_pages(), so let's fix it.
Sasha Levin has reported two THP BUGs[1][2]. I believe both of them
have the same root cause. Let's look to them one by one.
The first bug[1] is "kernel BUG at mm/huge_memory.c:1829!". It's
BUG_ON(mapcount != page_mapcount(page)) in __split_huge_page(). From my
testing I see that page_mapcount() is higher than mapcount here.
I think it happens due to race between zap_huge_pmd() and
page_check_address_pmd(). page_check_address_pmd() misses PMD which is
under zap:
CPU0 CPU1
zap_huge_pmd()
pmdp_get_and_clear()
__split_huge_page()
anon_vma_interval_tree_foreach()
__split_huge_page_splitting()
page_check_address_pmd()
mm_find_pmd()
/*
* We check if PMD present without taking ptl: no
* serialization against zap_huge_pmd(). We miss this PMD,
* it's not accounted to 'mapcount' in __split_huge_page().
*/
pmd_present(pmd) == 0
The second bug[2] is "kernel BUG at mm/huge_memory.c:1371!".
It's VM_BUG_ON_PAGE(!PageHead(page), page) in zap_huge_pmd().
This happens in similar way:
CPU0 CPU1
zap_huge_pmd()
pmdp_get_and_clear()
page_remove_rmap(page)
atomic_add_negative(-1, &page->_mapcount)
__split_huge_page()
anon_vma_interval_tree_foreach()
__split_huge_page_splitting()
page_check_address_pmd()
mm_find_pmd()
pmd_present(pmd) == 0 /* The same comment as above */
/*
* No crash this time since we already decremented page->_mapcount in
* zap_huge_pmd().
*/
BUG_ON(mapcount != page_mapcount(page))
/*
* We split the compound page here into small pages without
* serialization against zap_huge_pmd()
*/
__split_huge_page_refcount()
VM_BUG_ON_PAGE(!PageHead(page), page); // CRASH!!!
So my understanding the problem is pmd_present() check in mm_find_pmd()
without taking page table lock.
The bug was introduced by me commit with commit 117b0791ac42. Sorry for
that. :(
Let's open code mm_find_pmd() in page_check_address_pmd() and do the
check under page table lock.
Note that __page_check_address() does the same for PTE entires
if sync != 0.
I've stress tested split and zap code paths for 36+ hours by now and
don't see crashes with the patch applied. Before it took <20 min to
trigger the first bug and few hours for second one (if we ignore
first).
The SYSTEM_TRUSTED_KEYRING config option is not in any menu, causing it
to show up in the toplevel of the kernel configuration. Fix this by
moving it under the General Setup menu.
Signed-off-by: Peter Foley <pefoley2@pefoley.com> Cc: David Howells <dhowells@redhat.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
As trace event triggers are now part of the mainline kernel, I added
my trace event trigger tests to my test suite I run on all my kernels.
Now these tests get run under different config options, and one of
those options is CONFIG_PROVE_RCU, which checks under lockdep that
the rcu locking primitives are being used correctly. This triggered
the following splat:
The cause is that the triggers are protected by rcu_read_lock_sched() but
the data is dereferenced with rcu_dereference() which expects it to
be protected with rcu_read_lock(). The proper reference should be
rcu_dereference_sched().
Cc: Tom Zanussi <tom.zanussi@linux.intel.com> Signed-off-by: Steven Rostedt <rostedt@goodmis.org> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
When performing a user-request check/repair (MD_RECOVERY_REQUEST is set)
on a raid1, we allocate multiple bios each with their own set of pages.
If the page allocations for one bio fails, we currently do *not* free
the pages allocated for the previous bios, nor do we free the bio itself.
This patch frees all the already-allocate pages, and makes sure that
all the bios are freed as well.
This bug can cause a memory leak which can ultimately OOM a machine.
It was introduced in 3.10-rc1.
Fixes: a07876064a0b73ab5ef1ebcf14b1cf0231c07858 Cc: Kent Overstreet <koverstreet@google.com> Reported-by: Russell King - ARM Linux <linux@arm.linux.org.uk> Signed-off-by: NeilBrown <neilb@suse.de> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The Microsoft Surface Type/Touch Cover 2 is a fancy device which advertised
itself as a multitouch device but with constant input reports.
This way, hid_scan_report() gives the group MULTITOUCH to it, but
hid-multitouch can not handle it due to the constant collection ignored
by hid-input.
To prevent such crap in the future, and while we do not fix this particular
device, make the scan_report coherent with hid-input.c, and ignore constant
input reports.
The MS Surface Pro 2 has an USB composite device with 3 interfaces
- interface 0 - sensor hub
- interface 1 - wacom digitizer
- interface 2 - the keyboard cover, if one is attached
This USB composite device changes it product id dependent on if and which
keyboard cover is attached. Adding the covers to hid_have_special_driver
prevents loading the right hid drivers for the other two interfaces, all 3
get loaded with hid-microsoft. We don't even need hid-microsoft for the
keyboards. We have to revert this to load the right hid modules for each
interface.
The git commit a945928ea2709bc0e8e8165d33aed855a0110279
('xen: Do not enable spinlocks before jump_label_init() has executed')
was added to deal with the jump machinery. Earlier the code
that turned on the jump label was only called by Xen specific
functions. But now that it had been moved to the initcall machinery
it gets called on Xen, KVM, and baremetal - ouch!. And the detection
machinery to only call it on Xen wasn't remembered in the heat
of merge window excitement.
This means that the slowpath is enabled on baremetal while it should
not be.
Reported-by: Waiman Long <waiman.long@hp.com> Acked-by: Steven Rostedt <rostedt@goodmis.org> CC: Boris Ostrovsky <boris.ostrovsky@oracle.com> Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com> Signed-off-by: David Vrabel <david.vrabel@citrix.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Problem reported in Red Hat bz 1040329 for strict writes where we cache
only when we hold oplock and write direct to the server when we don't.
When we receive an oplock break, we first change the oplock value for
the inode in cifsInodeInfo->oplock to indicate that we no longer hold
the oplock before we enqueue a task to flush changes to the backing
device. Once we have completed flushing the changes, we return the
oplock to the server.
There are 2 ways here where we can have data corruption
1) While we flush changes to the backing device as part of the oplock
break, we can have processes write to the file. These writes check for
the oplock, find none and attempt to write directly to the server.
These direct writes made while we are flushing from cache could be
overwritten by data being flushed from the cache causing data
corruption.
2) While a thread runs in cifs_strict_writev, the machine could receive
and process an oplock break after the thread has checked the oplock and
found that it allows us to cache and before we have made changes to the
cache. In that case, we end up with a dirty page in cache when we
shouldn't have any. This will be flushed later and will overwrite all
subsequent writes to the part of the file represented by this page.
Before making any writes to the server, we need to confirm that we are
not in the process of flushing data to the server and if we are, we
should wait until the process is complete before we attempt the write.
We should also wait for existing writes to complete before we process
an oplock break request which changes oplock values.
We add a version specific downgrade_oplock() operation to allow for
differences in the oplock values set for the different smb versions.
Signed-off-by: Sachin Prabhu <sprabhu@redhat.com> Reviewed-by: Jeff Layton <jlayton@redhat.com> Reviewed-by: Pavel Shilovsky <piastry@etersoft.ru> Signed-off-by: Steve French <smfrench@gmail.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
it's pointless and actually leads to wrong behaviour in at least one
moderately convoluted case (pipe(), close one end, try to get to
another via /proc/*/fd and run into ETXTBUSY).
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
There are only a couple of architectures that override _STK_LIM_MAX to
a non-infinity value. This changes the stack allocation semantics in
subtle ways. For example, GNU make changes its stack allocation to the
hard maximum defined by _STK_LIM_MAX. As a results, threads executed
by processes running under make are allocated a stack size of
_STK_LIM_MAX rather than a sensible default value. This causes various
thread stress tests to fail when they can't muster more than about 50
threads.
The attached change implements the default behavior used by the
majority of architectures.
Signed-off-by: John David Anglin <dave.anglin@bell.net> Reviewed-by: Carlos O'Donell <carlos@systemhalted.org> Signed-off-by: Helge Deller <deller@gmx.de> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
On parisc, SHMLBA was defined to 0x00400000 (4MB) to reflect that we need to
take care of our caches for shared mappings. But actually, we can map a file at
any multiple address of PAGE_SIZE, so let us correct that now with a value of
PAGE_SIZE for SHMLBA. Instead we now take care of this cache colouring via the
constant SHM_COLOUR while we map shared pages.
Signed-off-by: Helge Deller <deller@gmx.de> CC: Jeroen Roovers <jer@gentoo.org> CC: John David Anglin <dave.anglin@bell.net> CC: Carlos O'Donell <carlos@systemhalted.org> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Since commit d689fe222 (NOHZ: Check for nohz active instead of nohz
enabled) the tick_nohz_switch_to_nohz() function returns because it
checks for the tick_nohz_active flag. This can't be set, because the
function itself sets it.
In tick_do_update_jiffies64() we are processing ticks only if delta is
greater than tick_period. This is what we are supposed to do here and
it broke a bit with this patch:
commit 47a1b796 (tick/timekeeping: Call update_wall_time outside the
jiffies lock)
With above patch, we might end up calling update_wall_time() even if
delta is found to be smaller that tick_period. Fix this by returning
when the delta is less than tick period.
[ tglx: Made it a 3 liner and massaged changelog ]
drm/i915: do not setup backlight if not available according to VBT
While the commits are within the stable rules otherwise, and fix more
machines than just the regressed Dell XPS models, we feel backporting
them to stable may be too risky. The revert is limited to the broken
machines, and the impact should be effectively the same as what the
upstream commits do more generally.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=76276 Reported-by: Romain Francoise <romain@orebokech.com> Acked-by: Daniel Vetter <daniel@ffwll.ch> Tested-by: Kamal Mostafa <kamal@canonical.com> Signed-off-by: Jani Nikula <jani.nikula@intel.com> Tested-by: Romain Francoise <romain@orebokech.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Doing rbd_obj_request_put() in rbd_img_request_fill() error paths is
not only insufficient, but also triggers an rbd_assert() in
rbd_obj_request_destroy():
Assertion failure in rbd_obj_request_destroy() at line 1867:
rbd_assert(obj_request->img_request == NULL);
rbd_img_obj_request_add() adds obj_requests to the img_request, the
opposite is rbd_img_obj_request_del(). Use it.
Peter Zijlstra [Wed, 9 Apr 2014 14:24:47 +0000 (16:24 +0200)]
x86,preempt: Fix preemption for i386
Many people reported preemption/reschedule problems with i386 kernels
for .13 and .14. After Michele bisected this to a combination of
3e8e42c69bb ("sched: Revert need_resched() to look at TIF_NEED_RESCHED") ded79754754 ("irq: Force hardirq exit's softirq processing on its own stack")
it finally dawned on me that i386's current_thread_info() was to
blame.
When we are on interrupt/exception stacks, we fail to observe the
right TIF_NEED_RESCHED bit and therefore the PREEMPT_NEED_RESCHED
folding malfunctions.
Current upstream fixes this by making i386 behave the same as x86_64
already did:
2432e1364bbe ("x86: Nuke the supervisor_stack field in i386 thread_info") b807902a88c4 ("x86: Nuke GET_THREAD_INFO_WITH_ESP() macro for i386") 0788aa6a23cb ("x86: Prepare removal of previous_esp from i386 thread_info structure") 198d208df437 ("x86: Keep thread_info on thread stack in x86_32")
However, that is far too much to stuff into -stable. Therefore I
propose we merge the below patch which uses task_thread_info(current)
for tif_need_resched() instead of the ESP based current_thread_info().
This makes sure we always observe the one true TIF_NEED_RESCHED bit
and things will work as expected again.
Currently, nf_tables trims off the set name if it exceeeds 15
bytes, so explicitly reject set names that are too large.
Reported-by: Giuseppe Longo <giuseppelng@gmail.com> Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
All xtables variants suffer from the defect that the copy_to_user()
to copy the counters to user memory may fail after the table has
already been exchanged and thus exposed. Return an error at this
point will result in freeing the already exposed table. Any
subsequent packet processing will result in a kernel panic.
We can't copy the counters before exposing the new tables as we
want provide the counter state after the old table has been
unhooked. Therefore convert this into a silent error.
Cc: Florian Westphal <fw@strlen.de> Signed-off-by: Thomas Graf <tgraf@suug.ch> Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
"len" contains sizeof(nf_ct_ext) and size of extensions. In a worst
case it can contain all extensions. Bellow you can find sizes for all
types of extensions. Their sum is definitely bigger than 256.
I have seen "len" up to 280 and my host has crashes w/o this patch.
The right way to fix this problem is reducing the size of the ecache
extension (4) and Florian is going to do this, but these changes will
be quite large to be appropriate for a stable tree.
Fixes: 5b423f6a40a0 (netfilter: nf_conntrack: fix racy timer handling with reliable) Cc: Pablo Neira Ayuso <pablo@netfilter.org> Cc: Patrick McHardy <kaber@trash.net> Cc: Jozsef Kadlecsik <kadlec@blackhole.kfki.hu> Cc: "David S. Miller" <davem@davemloft.net> Signed-off-by: Andrey Vagin <avagin@openvz.org> Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
nft_cmp_fast is used for equality comparisions of size <= 4. For
comparisions of size < 4 byte a mask is calculated that is applied to
both the data from userspace (during initialization) and the register
value (during runtime). Both values are stored using (in effect) memcpy
to a memory area that is then interpreted as u32 by nft_cmp_fast.
This works fine on little endian since smaller types have the same base
address, however on big endian this is not true and the smaller types
are interpreted as a big number with trailing zero bytes.
The mask therefore must not include the lower bytes, but the higher bytes
on big endian. Add a helper function that does a cpu_to_le32 to switch
the bytes on big endian. Since we're dealing with a mask of just consequitive
bits, this works out fine.
Signed-off-by: Patrick McHardy <kaber@trash.net> Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
sys_getppid() returns the parent pid of the current process in its own pid
namespace. Since audit filters are based in the init pid namespace, a process
could avoid a filter or trigger an unintended one by being in an alternate pid
namespace or log meaningless information.
Switch to task_ppid_nr() for PPIDs to anchor all audit filters in the
init_pid_ns.
(informed by ebiederman's 6c621b7e) Cc: Eric W. Biederman <ebiederm@xmission.com> Signed-off-by: Richard Guy Briggs <rgb@redhat.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Added the functions task_ppid_nr_ns() and task_ppid_nr() to abstract the lookup
of the PPID (real_parent's pid_t) of a process, including rcu locking, in the
arbitrary and init_pid_ns.
This provides an alternative to sys_getppid(), which is relative to the child
process' pid namespace.
(informed by ebiederman's 6c621b7e) Cc: Eric W. Biederman <ebiederm@xmission.com> Signed-off-by: Richard Guy Briggs <rgb@redhat.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Commit 12e55569a244 "tools lib traceevent: Use helper trace-seq in print
functions like kernel does" added a extra trace_seq helper to process
string arguments like the kernel does it. But the difference between the
kernel and the userspace library is that the kernel's trace_seq structure
has a static allocated buffer. The userspace one has a dynamically
allocated one. It requires a trace_seq_destroy(), otherwise it produces
a nasty memory leak.
In the highly unusual case where two threads are running concurrently through
the scanning code scanning the same target, we run into the situation where
one may allocate the target while the other is still using it. In this case,
because the reap checks for STARGET_CREATED and kills the target without
reference counting, the second thread will do the wrong thing on reap.
Fix this by reference counting even creates and doing the STARGET_CREATED
check in the final put.
Tested-by: Sarah Sharp <sarah.a.sharp@linux.intel.com> Signed-off-by: James Bottomley <JBottomley@Parallels.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
This patch eliminates the reap_ref and replaces it with a proper kref.
On last put of this kref, the target is removed from visibility in
sysfs. The final call to scsi_target_reap() for the device is done from
__scsi_remove_device() and only if the device was made visible. This
ensures that the target disappears as soon as the last device is gone
rather than waiting until final release of the device (which is often
too long).
Reviewed-by: Alan Stern <stern@rowland.harvard.edu> Tested-by: Sarah Sharp <sarah.a.sharp@linux.intel.com> Signed-off-by: James Bottomley <JBottomley@Parallels.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
If you are using a 64-bit kernel with 32-bit userland, then
scripts/gcc-x86_64-has-stack-protector.sh invokes 32-bit gcc
with -mcmodel=kernel, which produces:
<stdin>:1:0: error: code model 'kernel' not supported in the 32 bit mode
and trips the "broken compiler" test at arch/x86/Makefile:120.
There are several places a fix is possible, but the following seems
cleanest. (But it's minimal; it would also be possible to factor
out a bunch of stuff from the two branches of the if.)
The related error with allmodconfig for unicore32:
CC drivers/cpufreq/unicore2-cpufreq.o
drivers/cpufreq/unicore2-cpufreq.c: In function ‘ucv2_target’:
drivers/cpufreq/unicore2-cpufreq.c:48: error: ‘struct cpufreq_policy’ has no member named ‘mclk’
make[2]: *** [drivers/cpufreq/unicore2-cpufreq.o] Error 1
make[1]: *** [drivers/cpufreq] Error 2
make: *** [drivers] Error 2
Fixes: 652ed95d5fa6 (cpufreq: introduce cpufreq_generic_get() routine) Signed-off-by: Chen Gang <gang.chen.5i5j@gmail.com> Acked-by: Viresh Kumar <viresh.kumar@linaro.org> Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
According commit d640113fe (ACPI: processor: fix acpi_get_cpuid for UP
processor), BIOS may not provide _MAT or MADT tables and acpi_get_apicid()
always returns -1. For these cases, original code will pass apic_id with
vaule of -1 to acpi_map_cpuid() and it will check the acpi_id. If acpi_id
is equal to zero, ignores apic_id and return zero for CPU0.
Commit b981513 (ACPI / scan: bail out early if failed to parse APIC
ID for CPU) changed the behavior. Return ENODEV when find apic_id is
less than zero after calling acpi_get_apicid(). This causes acpi-cpufreq
driver fails to be loaded on some machines. This patch is to fix it.
Fixes: b981513f806d (ACPI / scan: bail out early if failed to parse APIC ID for CPU)
References: https://bugzilla.kernel.org/show_bug.cgi?id=73781 Reported-and-tested-by: KATO Hiroshi <katoh@mikage.ne.jp> Reported-and-tested-by: Stuart Foster <smf.linux@ntlworld.com> Signed-off-by: Lan Tianyu <tianyu.lan@intel.com> Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
`sdhci_bcm_kona_remove' referenced in section `.data' of drivers/built-in.o: defined in discarded section `.exit.text' of drivers/built-in.o
Fixes: 058feb53666f ("mmc: sdhci-bcm-kona: make linker-section warning go away") Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk> Tested-by: Markus Mayer <markus.mayer@linaro.org> Acked-by: Matt Porter <mporter@linaro.org> Signed-off-by: Chris Ball <chris@printf.net> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
I got a bug report yesterday from Laszlo Ersek in which he states that
his kvm instance fails to suspend. Laszlo bisected it down to this
commit 1cf7e9c68fe8 ("virtio_blk: blk-mq support") where virtio-blk is
converted to use the blk-mq infrastructure.
After digging a bit, it became clear that the issue was with the queue
drain. blk-mq tracks queue usage in a percpu counter, which is
incremented on request alloc and decremented when the request is freed.
The initial hunt was for an inconsistency in blk-mq, but everything
seemed fine. In fact, the counter only returned crazy values when
suspend was in progress.
When a CPU is unplugged, the percpu counters merges that CPU state with
the general state. blk-mq takes care to register a hotcpu notifier with
the appropriate priority, so we know it runs after the percpu counter
notifier. However, the percpu counter notifier only merges the state
when the CPU is fully gone. This leaves a state transition where the
CPU going away is no longer in the online mask, yet it still holds
private values. This means that in this state, percpu_counter_sum()
returns invalid results, and the suspend then hangs waiting for
abs(dead-cpu-value) requests to complete which of course will never
happen.
Fix this by clearing the state earlier, so we never have a case where
the CPU isn't in online mask but still holds private state. This bug
has been there since forever, I guess we don't have a lot of users where
percpu counters needs to be reliable during the suspend cycle.
Evidently, we created some lockd sockets and then failed to create
others. make_socks then returned an error and we tried to tear down the
svc, but svc->sv_permsocks was not empty so we ended up tripping over
the BUG() in svc_destroy().
Fix this by ensuring that we tear down any live sockets we created when
socket creation is going to return an error.
Fixes: 786185b5f8abefa (SUNRPC: move per-net operations from...) Reported-by: Raphos <raphoszap@laposte.net> Signed-off-by: Jeff Layton <jlayton@redhat.com> Reviewed-by: Stanislav Kinsbursky <skinsbursky@parallels.com> Signed-off-by: J. Bruce Fields <bfields@redhat.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
As reported by Tang Chen, Gu Zheng and Yasuaki Isimatsu, the following issues
exist in the aio ring page migration support.
As a result, for example, we have the following problem:
thread 1 | thread 2
|
aio_migratepage() |
|-> take ctx->completion_lock |
|-> migrate_page_copy(new, old) |
| *NOW*, ctx->ring_pages[idx] == old |
|
| *NOW*, ctx->ring_pages[idx] == old
| aio_read_events_ring()
| |-> ring = kmap_atomic(ctx->ring_pages[0])
| |-> ring->head = head; *HERE, write to the old ring page*
| |-> kunmap_atomic(ring);
|
|-> ctx->ring_pages[idx] = new |
| *BUT NOW*, the content of |
| ring_pages[idx] is old. |
|-> release ctx->completion_lock |
As above, the new ring page will not be updated.
Fix this issue, as well as prevent races in aio_ring_setup() by holding
the ring_lock mutex during kioctx setup and page migration. This avoids
the overhead of taking another spinlock in aio_read_events_ring() as Tang's
and Gu's original fix did, pushing the overhead into the migration code.
Note that to handle the nesting of ring_lock inside of mmap_sem, the
migratepage operation uses mutex_trylock(). Page migration is not a 100%
critical operation in this case, so the ocassional failure can be
tolerated. This issue was reported by Sasha Levin.
Based on feedback from Linus, avoid the extra taking of ctx->completion_lock.
Instead, make page migration fully serialised by mapping->private_lock, and
have aio_free_ring() simply disconnect the kioctx from the mapping by calling
put_aio_ring_file() before touching ctx->ring_pages[]. This simplifies the
error handling logic in aio_migratepage(), and should improve robustness.
v4: always do mutex_unlock() in cases when kioctx setup fails.
The code to handle any length SG lists calls edma_resume()
even before edma_start() is called. This is incorrect
because edma_resume() enables edma events on the channel
after which CPU (in edma_start) cannot clear posted
events by writing to ECR (per the EDMA user's guide).
Because of this EDMA transfers fail to start if due
to some reason there is a pending EDMA event registered
even before EDMA transfers are started. This can happen if
an EDMA event is a byproduct of device initialization.
Fix this by calling edma_resume() only if it is not the
first batch of MAX_NR_SG elements.
Without this patch, MMC/SD fails to function on DA850 EVM
with DMA. The behaviour is triggered by specific IP and
this can explain why the issue was not reported before
(example with MMC/SD on AM335x).
Tested on DA850 EVM and AM335x EVM-SK using MMC/SD card.
Cc: Joel Fernandes <joelf@ti.com> Acked-by: Joel Fernandes <joelf@ti.com> Tested-by: Jon Ringle <jringle@gridpoint.com> Tested-by: Alexander Holler <holler@ahsoftware.de> Reported-by: Jon Ringle <jringle@gridpoint.com> Signed-off-by: Sekhar Nori <nsekhar@ti.com> Signed-off-by: Vinod Koul <vinod.koul@intel.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
When suspending a cache the policy is walked and the individual policy
hints written to the metadata via sync_metadata(). This led to this
lock order:
policy->lock
cache_metadata->root_lock
When loading the cache target the policy is populated while the metadata
lock is held:
cache_metadata->root_lock
policy->lock
Fix this potential lock-inversion (ABBA) deadlock in sync_metadata() by
ensuring the cache_metadata root_lock is held whilst all the hints are
written, rather than being repeatedly locked while policy->lock is held
(as was the case with each callout that policy_walk_mappings() made to
the old save_hint() method).
Found by turning on the CONFIG_PROVE_LOCKING ("Lock debugging: prove
locking correctness") build option. However, it is not clear how the
LOCKDEP reported paths can lead to a deadlock since the two paths,
suspending a target and loading a target, never occur at the same time.
But that doesn't mean the same lock-inversion couldn't have occurred
elsewhere.
Reported-by: Marian Csontos <mcsontos@redhat.com> Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
If unable to ensure_next_mapping() we must add the current bio, which
was removed from the @bios list via bio_list_pop, back to the
deferred_bios list before all the remaining @bios.
Signed-off-by: Mike Snitzer <snitzer@redhat.com> Acked-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The persistent-data library used by dm-thin, dm-cache, etc is
transactional. If anything goes wrong, such as an io error when writing
new metadata or a power failure, then we roll back to the last
transaction.
Atomicity when committing a transaction is achieved by:
a) Never overwriting data from the previous transaction.
b) Writing the superblock last, after all other metadata has hit the
disk.
This commit and the following commit ("dm: take care to copy the space
map roots before locking the superblock") fix a bug associated with (b).
When committing it was possible for the superblock to still be written
in spite of an io error occurring during the preceeding metadata flush.
With these commits we're careful not to take the write lock out on the
superblock until after the metadata flush has completed.
Change the transaction manager's semantics for dm_tm_commit() to assume
all data has been flushed _before_ the single superblock that is passed
in.
As a prerequisite, split the block manager's block unlocking and
flushing by simplifying dm_bm_flush_and_unlock() to dm_bm_flush(). Now
the unlocking must be done separately.
This issue was discovered by forcing io errors at the crucial time
using dm-flakey.
Signed-off-by: Joe Thornber <ejt@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
If the discard block size is larger than the cache block size we will
not properly quiesce IO to a region that is about to be discarded. This
results in a race between a cache migration where no copy is needed, and
a write to an adjacent cache block that's within the same large discard
block.
Workaround this by limiting the discard_block_size to cache_block_size.
Also limit the max_discard_sectors to cache_block_size.
A more comprehensive fix that introduces range locking support in the
bio_prison and proper quiescing of a discard range that spans multiple
cache blocks is already in development.
Reported-by: Morgan Mears <Morgan.Mears@netapp.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com> Acked-by: Joe Thornber <ejt@redhat.com> Acked-by: Heinz Mauelshagen <heinzm@redhat.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>