Le lundi 07 novembre 2011 à 15:33 +0100, Eric Dumazet a écrit :

> At least, in recent kernels we dont change dst->refcnt in forwarding
> patch (usinf NOREF skb->dst)
>
> One particular point is the atomic_inc(dst->refcnt) we have to perform
> when queuing an UDP packet if socket asked PKTINFO stuff (for example a
> typical DNS server has to setup this option)
>
> I have one patch somewhere that stores the information in skb->cb[] and
> avoid the atomic_{inc|dec}(dst->refcnt).
>

OK I found it, I did some extra tests and believe its ready.

[PATCH net-next] ipv4: IP_PKTINFO doesnt need dst reference

When a socket uses IP_PKTINFO notifications, we currently force a dst
reference for each received skb. Reader has to access dst to get needed
information (rt_iif & rt_spec_dst) and must release dst reference.

We also forced a dst reference if skb was put in socket backlog, even
without IP_PKTINFO handling. This happens under stress/load.

We can instead store the needed information in skb->cb[], so that only
softirq handler really access dst, improving cache hit ratios.

This removes two atomic operations per packet, and false sharing as
well.

On a benchmark using a mono threaded receiver (doing only recvmsg()
calls), I can reach 720.000 pps instead of 570.000 pps.

IP_PKTINFO is typically used by DNS servers, and any multihomed aware
UDP application.
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Reading /proc/net/snmp on a machine with a lot of cpus is very expensive
(can be ~88000 us).

This is because ICMPMSG MIB uses 4096 bytes per cpu, and folding values
for all possible cpus can read 16 Mbytes of memory.

ICMP messages are not considered as fast path on a typical server, and
eventually few cpus handle them anyway. We can afford an atomic
operation instead of using percpu data.

This saves 4096 bytes per cpu and per network namespace.
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Make clear that sk_clone() and inet_csk_clone() return a locked socket.

Add _lock() prefix and kerneldoc.
Suggested-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Remove the assumption that skb_get_rxhash() makes IP header and ports
linear, and use skb_header_pointer() instead in choke_match_flow()

This permits __skb_get_rxhash() to use skb_header_pointer() eventually.
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

This patch enables the ioctl support for the driver. So userspace
programs like mii-tool can work.

Resend in merge window
Signed-off-by: Ricardo Ribalda Delgado <ricardo.ribalda@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

This just makes it possible to spoof source IPv6 address on a socket
without having to create and bind a new socket for every source IP
we wish to spoof.
Signed-off-by: Maciej Żenczykowski <maze@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

This makes native ipv6 bind follow the precedent set by:
  - native ipv4 bind behaviour
  - dual stack ipv4-mapped ipv6 bind behaviour.

This does allow an unpriviledged process to spoof its source IPv6
address, just like it currently can spoof its source IPv4 address
(for example when using UDP).
Signed-off-by: Maciej Żenczykowski <maze@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Per the mention made by Ben Hutchings that strlcpy is now the preferred
string copy routine for a .get_drvinfo routine, do a bit of floor
sweeping and convert some of the as-yet unconverted ethernet drivers to
it.
Signed-off-by: Rick Jones <rick.jones2@hp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Retransmit chunks to newly confirmed destination when ASCONF and
HEARTBEAT negotiation has success with a single-homed peer.
Signed-off-by: Michio Honda <micchie@sfc.wide.ad.jp>
Signed-off-by: David S. Miller <davem@davemloft.net>

Fast retransmission after changing the last address
with ASCONF negotiation
Signed-off-by: Michio Honda <micchie@sfc.wide.ad.jp>
Signed-off-by: David S. Miller <davem@davemloft.net>

Tunnels can force an alignment of their percpu data to reduce number of
cache lines used in fast path, or read in .ndo_get_stats()

percpu_alloc() is a very fine grained allocator, so any small hole will
be used anyway.
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

As we update the learned pmtu informations on demand, we might
report a nagative expiration time value to userspace if the
pmtu informations are already expired and we have not send a
packet to that inetpeer after expiration. With this patch we
send a expire time of null to userspace after expiration
until the next packet is send to that inetpeer.
Signed-off-by: Steffen Klassert <steffen.klassert@secunet.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Small fix in Documentation, since min_pmtu is 512 + 20 + 20 = 552
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

TCP_NODELAY is weaker than TCP_CORK, when TCP_CORK was set, small
segments will always pass Nagle test regardless of TCP_NODELAY option.
Signed-off-by: Feng King <kinwin2008@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

To prevent the SBus driver for Sun Happy Meal cards from being loaded for
PCI cards utilizing the same chipset, a filter was added to the probe
function in commit 0b492fce.

The filter was implemented by checking the name of the parent node in
the OF tree. This patch extends this filter, so that the driver will
load on SBus systems that are based upon SBI SBus Bridges.
Signed-off-by: Kjetil Oftedal <oftedal@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

pskb_may_pull() can change skb->data, so we have to load ptr/optr at the
right place.
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

The driver used NET_IP_ALIGN to remove some additional padding inside of
the rx_fixup function. On many architectures NET_IP_ALIGN defaults to 2
which removed the correct amount of bytes.

On MCORE2-machines commit ea812ca1
introduces a change which sets NET_IP_ALIGN to 0 by default. Which
triggered the bug on these machines.

This fix introduces a new RXW_PADDING define and uses this instead of
NET_IP_ALIGN. The name was taken from the original SMSC7500 driver which
is provided by SMSC.
Signed-off-by: Nico Erfurth <ne@erfurth.eu>
Tested-by: Phil Sutter <phil@nwl.cc>
Signed-off-by: David S. Miller <davem@davemloft.net>

RFC 4291 Section 2.7 says Multicast addresses must not be used as source
addresses in IPv6 packets - drop them on input so we don't process the
packet further.
Signed-off-by: Brian Haley <brian.haley@hp.com>
Reported-and-Tested-by: Kumar Sanghvi <divinekumar@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

The link down would occur when reseting PHY. And it would take about 2 ~ 5 seconds
from link down to link up. If the delay of pm_schedule_suspend is not long enough,
the device would enter runtime_suspend before link up. After link up, the device
would wake up and reset PHY again. Then, you would find the driver keep in a loop
of runtime_suspend and rumtime_resume.
Signed-off-by: Hayes Wang <hayeswang@realtek.com>
Acked-by: Francois Romieu <romieu@fr.zoreil.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

.. with new name.  Because nothing says "really solid kernel release"
like naming it after an extinct animal that just happened to be in the
news lately.

* 'fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/tmlind/linux-omap: (31 commits)
  ARM: OMAP: Fix export.h or module.h includes
  ARM: OMAP: omap_device: Include linux/export.h
  ARM: OMAP2: Fix H4 matrix keyboard warning
  ARM: OMAP1: Remove unused omap-alsa.h
  ARM: OMAP1: Fix warnings about enabling 32 KiHz timer
  ARM: OMAP2+: timer: Remove omap_device_pm_latency
  ARM: OMAP2+: clock data: Remove redundant timer clkdev
  ARM: OMAP: Devkit8000: Remove double omap_mux_init_gpio
  ARM: OMAP: usb: musb: OMAP: Delete unused function
  MAINTAINERS: Update linux-omap git repository
  ARM: OMAP: change get_context_loss_count ret value to int
  ARM: OMAP4: hsmmc: configure SDMMC1_DR0 properly
  ARM: OMAP4: hsmmc: Fix Pbias configuration on regulator OFF
  ARM: OMAP3: hwmod: fix variant registration and remove SmartReflex from common list
  ARM: OMAP: I2C: Fix omap_register_i2c_bus() return value on success
  ARM: OMAP: dmtimer: Include linux/module.h
  ARM: OMAP2+: l3-noc: Include linux/module.h
  ARM: OMAP2+: devices: Fixes for McPDM
  ARM: OMAP: Fix errors and warnings when building for one board
  ARM: OMAP3: PM: restrict erratum i443 handling to OMAP3430 only
  ...

Mountpoint crossing is similar to following procfs symlinks - we do
not get ->d_revalidate() called for dentry we have arrived at, with
unpleasant consequences for NFS4.

Simple way to reproduce the problem in mainline:

    cat >/tmp/a.c <<'EOF'
    #include <unistd.h>
    #include <fcntl.h>
    #include <stdio.h>
    main()
    {
            struct flock fl = {.l_type = F_RDLCK, .l_whence = SEEK_SET, .l_len = 1};
            if (fcntl(0, F_SETLK, &fl))
                    perror("setlk");
    }
    EOF
    cc /tmp/a.c -o /tmp/test

then on nfs4:

    mount --bind file1 file2
    /tmp/test < file1		# ok
    /tmp/test < file2		# spews "setlk: No locks available"...

What happens is the missing call of ->d_revalidate() after mountpoint
crossing and that's where NFS4 would issue OPEN request to server.

The fix is simple - treat mountpoint crossing the same way we deal with
following procfs-style symlinks.  I.e.  set LOOKUP_JUMPED...

Cc: stable@kernel.org
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

* 'perf-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  perf top: Fix live annotation in the --stdio interface
  perf top tui: Don't recalc column widths considering just the first page
  perf report: Add progress bar when processing time ordered events
  perf hists browser: Warn about lost events
  perf tools: Fix a typo of command name as trace-cmd
  perf hists: Fix recalculation of total_period when sorting entries
  perf header: Fix build on old systems
  perf ui browser: Handle K_RESIZE in dialog windows
  perf ui browser: No need to switch char sets that often
  perf hists browser: Use K_TIMER
  perf ui: Rename ui__warning_paranoid to ui__error_paranoid
  perf ui: Reimplement the popup windows using libslang
  perf ui: Reimplement ui__popup_menu using ui__browser
  perf ui: Reimplement ui_helpline using libslang
  perf ui: Improve handling sigwinch a bit
  perf ui progress: Reimplement using slang
  perf evlist: Fix grouping of multiple events

Commit 32aaeffb (Merge branch
'modsplit-Oct31_2011'...) caused some build errors. Fix these
and make sure we always have export.h or module.h included
for MODULE_ and EXPORT_SYMBOL users:

$ grep -rl ^MODULE_ arch/arm/*omap*/*.c | xargs \
  grep -L linux/module.h
  arch/arm/mach-omap2/dsp.c
  arch/arm/mach-omap2/mailbox.c
  arch/arm/mach-omap2/omap-iommu.c
  arch/arm/mach-omap2/smartreflex.c

Also check we either have export.h or module.h included
for the files exporting symbols:

$ grep -rl EXPORT_SYMBOL arch/arm/*omap*/*.c | xargs \
  grep -L linux/export.h | xargs grep -L linux/module.h

Cc: Russell King <rmk+kernel@arm.linux.org.uk>
Signed-off-by: Tony Lindgren <tony@atomide.com>

Include linux/export.h to fix below build warning:

  CC      arch/arm/plat-omap/omap_device.o
arch/arm/plat-omap/omap_device.c:1055: warning: data definition has no type or storage class
arch/arm/plat-omap/omap_device.c:1055: warning: type defaults to 'int' in declaration of 'EXPORT_SYMBOL'
arch/arm/plat-omap/omap_device.c:1055: warning: parameter names (without types) in function declaration
Signed-off-by: Axel Lin <axel.lin@gmail.com>
Signed-off-by: Tony Lindgren <tony@atomide.com>

* git://git.kernel.org/pub/scm/linux/kernel/git/davem/net: (47 commits)
  forcedeth: fix a few sparse warnings (variable shadowing)
  forcedeth: Improve stats counters
  forcedeth: remove unneeded stats updates
  forcedeth: Acknowledge only interrupts that are being processed
  forcedeth: fix race when unloading module
  MAINTAINERS/rds: update maintainer
  wanrouter: Remove kernel_lock annotations
  usbnet: fix oops in usbnet_start_xmit
  ixgbe: Fix compile for kernel without CONFIG_PCI_IOV defined
  etherh: Add MAINTAINERS entry for etherh
  bonding: comparing a u8 with -1 is always false
  sky2: fix regression on Yukon Optima
  netlink: clarify attribute length check documentation
  netlink: validate NLA_MSECS length
  i825xx:xscale:8390:freescale: Fix Kconfig dependancies
  macvlan: receive multicast with local address
  tg3: Update version to 3.121
  tg3: Eliminate timer race with reset_task
  tg3: Schedule at most one tg3_reset_task run
  tg3: Obtain PCI function number from device
  ...

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

This fixes the following sparse warnings:
drivers/net/ethernet/nvidia/forcedeth.c:2113:7: warning: symbol 'size' shadows an earlier one
drivers/net/ethernet/nvidia/forcedeth.c:2102:6: originally declared here
drivers/net/ethernet/nvidia/forcedeth.c:2155:7: warning: symbol 'size' shadows an earlier one
drivers/net/ethernet/nvidia/forcedeth.c:2102:6: originally declared here
drivers/net/ethernet/nvidia/forcedeth.c:2227:7: warning: symbol 'size' shadows an earlier one
drivers/net/ethernet/nvidia/forcedeth.c:2215:6: originally declared here
drivers/net/ethernet/nvidia/forcedeth.c:2271:7: warning: symbol 'size' shadows an earlier one
drivers/net/ethernet/nvidia/forcedeth.c:2215:6: originally declared here
drivers/net/ethernet/nvidia/forcedeth.c:2986:20: warning: symbol 'addr' shadows an earlier one
drivers/net/ethernet/nvidia/forcedeth.c:2963:6: originally declared here
Signed-off-by: David Decotigny <david.decotigny@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Rx byte count was off; instead use the hardware's count.  Tx packet
count was counting pre-TSO packets; instead count on-the-wire packets.
Report hardware dropped frame count as rx_fifo_errors.

- The count of transmitted packets reported by the forcedeth driver
  reports pre-TSO (TCP Segmentation Offload) packet counts and not the
  count of the number of packets sent on the wire. This change fixes
  the forcedeth driver to report the correct count. Fixed the code by
  copying the count stored in the NIC H/W to the value reported by the
  driver.

- Count rx_drop_frame errors as rx_fifo_errors:
  We see a lot of rx_drop_frame errors if we disable the rx bottom-halves
  for too long.  Normally, rx_fifo_errors would be counted in this case.
  The rx_drop_frame error count is private to forcedeth and is not
  reported by ifconfig or sysfs.  The rx_fifo_errors count is currently
  unused in the forcedeth driver.  It is reported by ifconfig as overruns.
  This change reports rx_drop_frame errors as rx_fifo_errors.
Signed-off-by: David Decotigny <david.decotigny@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Function ndo_get_stats() updates most of the stats from hardware
registers, making the manual updates un-needed. This change removes
these manual updates. Main exception is rx_missed_errors which needs
manual update.

Another exception is rx_packets, still updated manually in this commit
to make sure this patch doesn't change behavior of driver. This will
be addressed by a future patch.
Signed-off-by: David Decotigny <david.decotigny@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

This is to avoid a race, accidentally acknowledging an interrupt that
we didn't notice and won't immediately process.  This is based solely
on code inspection; it is not known if there was an actual bug here.
Signed-off-by: David Decotigny <david.decotigny@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

When forcedeth module is unloaded, there exists a path that can lead
to mod_timer() after del_timer_sync(), causing an oops. This patch
short-circuits this unneeded path, which originates in
nv_get_ethtool_stats().

Tested:
  x86_64 16-way + 3 ethtool -S infinite loops + 100Mbps incoming traffic
  + rmmod/modprobe/ifconfig in a loop

Initial-Author: Salman Qazi <sqazi@google.com>
Discussion: http://patchwork.ozlabs.org/patch/123548/Signed-off-by: David Decotigny <david.decotigny@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Reported-by: Witold Baryluk <baryluk@smp.if.uj.edu.pl>
Signed-off-by: Stephen Rothwell <sfr@canb.auug.org.au>
Signed-off-by: Paul Gortmaker <paul.gortmaker@windriver.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

since it uses the module facilities.
Reported-by: Witold Baryluk <baryluk@smp.if.uj.edu.pl>
Signed-off-by: Stephen Rothwell <sfr@canb.auug.org.au>
Signed-off-by: Paul Gortmaker <paul.gortmaker@windriver.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

For the files which are not themselves modular, we can change
them to include only the smaller export.h since all they are
doing is looking for EXPORT_SYMBOL.
Reported-by: Stephen Rothwell <sfr@canb.auug.org.au>
Signed-off-by: Paul Gortmaker <paul.gortmaker@windriver.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

These files didn't exist at the time of the module.h split, and
so were not fixed by the commits on that baseline.  Since they use
the EXPORT_SYMBOL and/or THIS_MODULE macros, they will need the
new export.h file included that provides them.
Signed-off-by: Paul Gortmaker <paul.gortmaker@windriver.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

update for the actual maintainer
Signed-off-by: Or Gerlitz <ogerlitz@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

The BKL is gone, these annotations are useless.
Signed-off-by: Richard Weinberger <richard@nod.at>
Signed-off-by: David S. Miller <davem@davemloft.net>

This patch fixes the bug added in commit v3.1-rc7-1055-gf9b491ec
SKB can be NULL at this point, at least for cdc-ncm.
Signed-off-by: Konstantin Khlebnikov <khlebnikov@openvz.org>
Acked-by: Richard Cochran <richardcochran@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>