This makes the pipe code use separate wait-queues and exclusive waiting
for readers and writers, avoiding a nasty thundering herd problem when
there are lots of readers waiting for data on a pipe (or, less commonly,
lots of writers waiting for a pipe to have space).

While this isn't a common occurrence in the traditional "use a pipe as a
data transport" case, where you typically only have a single reader and
a single writer process, there is one common special case: using a pipe
as a source of "locking tokens" rather than for data communication.

In particular, the GNU make jobserver code ends up using a pipe as a way
to limit parallelism, where each job consumes a token by reading a byte
from the jobserver pipe, and releases the token by writing a byte back
to the pipe.

This pattern is fairly traditional on Unix, and works very well, but
will waste a lot of time waking up a lot of processes when only a single
reader needs to be woken up when a writer releases a new token.

A simplified test-case of just this pipe interaction is to create 64
processes, and then pass a single token around between them (this
test-case also intentionally passes another token that gets ignored to
test the "wake up next" logic too, in case anybody wonders about it):

    #include <unistd.h>

    int main(int argc, char **argv)
    {
        int fd[2], counters[2];

        pipe(fd);
        counters[0] = 0;
        counters[1] = -1;
        write(fd[1], counters, sizeof(counters));

        /* 64 processes */
        fork(); fork(); fork(); fork(); fork(); fork();

        do {
                int i;
                read(fd[0], &i, sizeof(i));
                if (i < 0)
                        continue;
                counters[0] = i+1;
                write(fd[1], counters, (1+(i & 1)) *sizeof(int));
        } while (counters[0] < 1000000);
        return 0;
    }

and in a perfect world, passing that token around should only cause one
context switch per transfer, when the writer of a token causes a
directed wakeup of just a single reader.

But with the "writer wakes all readers" model we traditionally had, on
my test box the above case causes more than an order of magnitude more
scheduling: instead of the expected ~1M context switches, "perf stat"
shows

        231,852.37 msec task-clock                #   15.857 CPUs utilized
        11,250,961      context-switches          #    0.049 M/sec
           616,304      cpu-migrations            #    0.003 M/sec
             1,648      page-faults               #    0.007 K/sec
 1,097,903,998,514      cycles                    #    4.735 GHz
   120,781,778,352      instructions              #    0.11  insn per cycle
    27,997,056,043      branches                  #  120.754 M/sec
       283,581,233      branch-misses             #    1.01% of all branches

      14.621273891 seconds time elapsed

       0.018243000 seconds user
       3.611468000 seconds sys

before this commit.

After this commit, I get

          5,229.55 msec task-clock                #    3.072 CPUs utilized
         1,212,233      context-switches          #    0.232 M/sec
           103,951      cpu-migrations            #    0.020 M/sec
             1,328      page-faults               #    0.254 K/sec
    21,307,456,166      cycles                    #    4.074 GHz
    12,947,819,999      instructions              #    0.61  insn per cycle
     2,881,985,678      branches                  #  551.096 M/sec
        64,267,015      branch-misses             #    2.23% of all branches

       1.702148350 seconds time elapsed

       0.004868000 seconds user
       0.110786000 seconds sys

instead. Much better.

[ Note! This kernel improvement seems to be very good at triggering a
  race condition in the make jobserver (in GNU make 4.2.1) for me. It's
  a long known bug that was fixed back in June 2017 by GNU make commit
  b552b0525198 ("[SV 51159] Use a non-blocking read with pselect to
  avoid hangs.").

  But there wasn't a new release of GNU make until 4.3 on Jan 19 2020,
  so a number of distributions may still have the buggy version. Some
  have backported the fix to their 4.2.1 release, though, and even
  without the fix it's quite timing-dependent whether the bug actually
  is hit. ]

Josh Triplett says:
 "I've been hammering on your pipe fix patch (switching to exclusive
  wait queues) for a month or so, on several different systems, and I've
  run into no issues with it. The patch *substantially* improves
  parallel build times on large (~100 CPU) systems, both with parallel
  make and with other things that use make's pipe-based jobserver.

  All current distributions (including stable and long-term stable
  distributions) have versions of GNU make that no longer have the
  jobserver bug"
Tested-by: Josh Triplett <josh@joshtriplett.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

Pull io_uring fixes from Jens Axboe:
 "Fix for two regressions in this cycle, both reported by the postgresql
  use case.

  One removes the added restriction on who can submit IO, making it
  possible for rings shared across forks to do so. The other fixes an
  issue for the same kind of use case, where one exiting process would
  cancel all IO"

* tag 'io_uring-5.5-2020-01-26' of git://git.kernel.dk/linux-block:
  io_uring: don't cancel all work on process exit
  Revert "io_uring: only allow submit from owning task"

Pull block fix from Jens Axboe:
 "Unfortunately this weekend we had a few last minute reports, one was
  for block.

  The partition disable for zoned devices was overly restrictive, it can
  work (and be supported) just fine for host-aware variants.

  Here's a fix ensuring that's the case so we don't break existing users
  of that"

* tag 'block-5.5-2020-01-26' of git://git.kernel.dk/linux-block:
  block: allow partitions on host aware zone devices

Pull SCSI fixes from James Bottomley:
 "Two last minute fixes, both in drivers.

  The fnic one is a highly unlikely condition, but the RDMA one is a
  recently introduced regression that causes a kernel warning to trigger
  in every RDMA logon, which would be unsightly if it got into the final
  release"

* tag 'scsi-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/jejb/scsi:
  scsi: RDMA/isert: Fix a recently introduced regression related to logout
  scsi: fnic: do not queue commands during fwreset

Pull vfs fix from Al Viro:
 "Fix a use-after-free in do_last() handling of sysctl_protected_...
  checks.

  The use-after-free normally doesn't happen there, but race with
  rename() and it becomes possible"

* 'fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs:
  do_last(): fetch directory ->i_mode and ->i_uid before it's too late

If we're sharing the ring across forks, then one process exiting means
that we cancel ALL work and prevent future work. This is overly
restrictive. As long as we cancel the work associated with the files
from the current task, it's safe to let others persist. Normal fd close
on exit will still wait (and cancel) pending work.

Fixes: fcb323cc ("io_uring: io_uring: add support for async work inheriting files")
Reported-by: Andres Freund <andres@anarazel.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>

Host-aware SMR drives can be used with the commands to explicitly manage
zone state, but they can also be used as normal disks.  In the former
case it makes perfect sense to allow partitions on them, in the latter
it does not, just like for host managed devices.  Add a check to
add_partition to allow partitions on host aware devices, but give
up any zone management capabilities in that case, which also catches
the previously missed case of adding a partition vs just scanning it.

Because sd can rescan the attribute at runtime it needs to check if
a disk has partitions, for which a new helper is added to genhd.h.

Fixes: 5eac3eb3 ("block: Remove partition support for zoned block devices")
Reported-by: Borislav Petkov <bp@suse.de>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Tested-by: Damien Le Moal <damien.lemoal@wdc.com>
Reviewed-by: Damien Le Moal <damien.lemoal@wdc.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>

This ends up being too restrictive for tasks that willingly fork and
share the ring between forks. Andres reports that this breaks his
postgresql work. Since we're close to 5.5 release, revert this change
for now.

Cc: stable@vger.kernel.org
Fixes: 44d28279 ("io_uring: only allow submit from owning task")
Reported-by: Andres Freund <andres@anarazel.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>

The afs filesystem needs to prohibit certain characters from cell names,
such as '/', as these are used to form filenames in procfs, leading to
the following warning being generated:

	WARNING: CPU: 0 PID: 3489 at fs/proc/generic.c:178

Fix afs_alloc_cell() to disallow nonprintable characters, '/', '@' and
names that begin with a dot.

Remove the check for "@cell" as that is then redundant.

This can be tested by running:

	echo add foo/.bar 1.2.3.4 >/proc/fs/afs/cells

Note that we will also need to deal with:

 - Names ending in ".invalid" shouldn't be passed to the DNS.

 - Names that contain non-valid domainname chars shouldn't be passed to
   the DNS.

 - DNS replies that say "your-dns-needs-immediate-attention.<gTLD>" and
   replies containing A records that say 127.0.53.53 should be
   considered invalid.
   [https://www.icann.org/en/system/files/files/name-collision-mitigation-01aug14-en.pdf]

but these need to be dealt with by the kafs-client DNS program rather
than the kernel.

Reported-by: syzbot+b904ba7c947a37b4b291@syzkaller.appspotmail.com
Cc: stable@kernel.org
Signed-off-by: David Howells <dhowells@redhat.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

may_create_in_sticky() call is done when we already have dropped the
reference to dir.

Fixes: 30aba665 (namei: allow restricted O_CREAT of FIFOs and regular files)
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

Pull ARM fixes from Russell King:

 - fix ftrace relocation type filtering

 - relax arch timer version check

* tag 'for-linus' of git://git.armlinux.org.uk/~rmk/linux-arm:
  ARM: 8955/1: virt: Relax arch timer version check during early boot
  ARM: 8950/1: ftrace/recordmcount: filter relocation types

Pull networking fixes from David Miller:

 1) Off by one in mt76 airtime calculation, from Dan Carpenter.

 2) Fix TLV fragment allocation loop condition in iwlwifi, from Luca
    Coelho.

 3) Don't confirm neigh entries when doing ipsec pmtu updates, from Xu
    Wang.

 4) More checks to make sure we only send TSO packets to lan78xx chips
    that they can actually handle. From James Hughes.

 5) Fix ip_tunnel namespace move, from William Dauchy.

 6) Fix unintended packet reordering due to cooperation between
    listification done by GRO and non-GRO paths. From Maxim
    Mikityanskiy.

 7) Add Jakub Kicincki formally as networking co-maintainer.

 8) Info leak in airo ioctls, from Michael Ellerman.

 9) IFLA_MTU attribute needs validation during rtnl_create_link(), from
    Eric Dumazet.

10) Use after free during reload in mlxsw, from Ido Schimmel.

11) Dangling pointers are possible in tp->highest_sack, fix from Eric
    Dumazet.

12) Missing *pos++ in various networking seq_next handlers, from Vasily
    Averin.

13) CHELSIO_GET_MEM operation neds CAP_NET_ADMIN check, from Michael
    Ellerman.

* git://git.kernel.org/pub/scm/linux/kernel/git/netdev/net: (109 commits)
  firestream: fix memory leaks
  net: cxgb3_main: Add CAP_NET_ADMIN check to CHELSIO_GET_MEM
  net: bcmgenet: Use netif_tx_napi_add() for TX NAPI
  tipc: change maintainer email address
  net: stmmac: platform: fix probe for ACPI devices
  net/mlx5e: kTLS, Do not send decrypted-marked SKBs via non-accel path
  net/mlx5e: kTLS, Remove redundant posts in TX resync flow
  net/mlx5e: kTLS, Fix corner-case checks in TX resync flow
  net/mlx5e: Clear VF config when switching modes
  net/mlx5: DR, use non preemptible call to get the current cpu number
  net/mlx5: E-Switch, Prevent ingress rate configuration of uplink rep
  net/mlx5: DR, Enable counter on non-fwd-dest objects
  net/mlx5: Update the list of the PCI supported devices
  net/mlx5: Fix lowest FDB pool size
  net: Fix skb->csum update in inet_proto_csum_replace16().
  netfilter: nf_tables: autoload modules from the abort path
  netfilter: nf_tables: add __nft_chain_type_get()
  netfilter: nf_tables_offload: fix check the chain offload flag
  netfilter: conntrack: sctp: use distinct states for new SCTP connections
  ipv6_route_seq_next should increase position index
  ...

Pull ARM SoC fixes from Olof Johansson:
 "A couple of fixes have come in that would be good to include in this
  release:

   - A fix for amount of memory on Beaglebone Black. Surfaced now since
     GRUB2 doesn't update memory size in the booted kernel.

   - A fix to make SPI interfaces work on am43x-epos-evm.

   - Small Kconfig fix for OPTEE (adds a depend on MMU) to avoid build
     failures"

* tag 'armsoc-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/soc/soc:
  ARM: dts: am43x-epos-evm: set data pin directions for spi0 and spi1
  tee: optee: Fix compilation issue with nommu
  ARM: dts: am335x-boneblack-common: fix memory size

In fs_open(), 'vcc' is allocated through kmalloc() and assigned to
'atm_vcc->dev_data.' In the following execution, if an error occurs, e.g.,
there is no more free channel, an error code EBUSY or ENOMEM will be
returned. However, 'vcc' is not deallocated, leading to memory leaks. Note
that, in normal cases where fs_open() returns 0, 'vcc' will be deallocated
in fs_close(). But, if fs_open() fails, there is no guarantee that
fs_close() will be invoked.

To fix this issue, deallocate 'vcc' before the error code is returned.
Signed-off-by: Wenwen Wang <wenwen@cs.uga.edu>
Signed-off-by: David S. Miller <davem@davemloft.net>

Pablo Neira Ayuso says:

====================
Netfilter fixes for net

The following patchset contains Netfilter fixes for net:

1) Missing netlink attribute sanity check for NFTA_OSF_DREG,
   from Florian Westphal.

2) Use bitmap infrastructure in ipset to fix KASAN slab-out-of-bounds
   reads, from Jozsef Kadlecsik.

3) Missing initial CLOSED state in new sctp connection through
   ctnetlink events, from Jiri Wiesner.

4) Missing check for NFT_CHAIN_HW_OFFLOAD in nf_tables offload
   indirect block infrastructure, from wenxu.

5) Add __nft_chain_type_get() to sanity check family and chain type.

6) Autoload modules from the nf_tables abort path to fix races
   reported by syzbot.

7) Remove unnecessary skb->csum update on inet_proto_csum_replace16(),
   from Praveen Chaudhary.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>

Pull btrfs fix from David Sterba:
 "Here's a last minute fix for a regression introduced in this
  development cycle.

  There's a small chance of a silent corruption when device replace and
  NOCOW data writes happen at the same time in one block group. Metadata
  or COW data writes are unaffected.

  The extra fixup patch is there to silence an unnecessary warning"

* tag 'for-5.5-rc8-tag' of git://git.kernel.org/pub/scm/linux/kernel/git/kdave/linux:
  btrfs: dev-replace: remove warning for unknown return codes when finished
  btrfs: scrub: Require mandatory block group RO for dev-replace

Pull pin control fix from Linus Walleij:
 "A single fix for the Intel Sunrisepoint pin controller that makes the
  interrupts work properly on it"

* tag 'pinctrl-v5.5-5' of git://git.kernel.org/pub/scm/linux/kernel/git/linusw/linux-pinctrl:
  pinctrl: sunrisepoint: Add missing Interrupt Status register offset

Saeed Mahameed says:

====================
Mellanox, mlx5 fixes 2020-01-24

This series introduces some fixes to mlx5 driver.

Please pull and let me know if there is any problem.

Merge conflict: once merge with net-next, a contextual conflict will
appear in drivers/net/ethernet/mellanox/mlx5/core/eswitch_offloads.c
since the code moved in net-next.
To resolve, just delete ALL of the conflicting hunk from net.
So sorry for the small mess ..

For -stable v5.4:
 ('net/mlx5: Update the list of the PCI supported devices')
 ('net/mlx5: Fix lowest FDB pool size')
 ('net/mlx5e: kTLS, Fix corner-case checks in TX resync flow')
 ('net/mlx5e: kTLS, Do not send decrypted-marked SKBs via non-accel path')
 ('net/mlx5: Eswitch, Prevent ingress rate configuration of uplink rep')
 ('net/mlx5e: kTLS, Remove redundant posts in TX resync flow')
 ('net/mlx5: DR, Enable counter on non-fwd-dest objects')
 ('net/mlx5: DR, use non preemptible call to get the current cpu number')
====================
Signed-off-by: David S. Miller <davem@davemloft.net>

The fstests btrfs/011 triggered a warning at the end of device replace,

  [ 1891.998975] BTRFS warning (device vdd): failed setting block group ro: -28
  [ 1892.038338] BTRFS error (device vdd): btrfs_scrub_dev(/dev/vdd, 1, /dev/vdb) failed -28
  [ 1892.059993] ------------[ cut here ]------------
  [ 1892.063032] WARNING: CPU: 2 PID: 2244 at fs/btrfs/dev-replace.c:506 btrfs_dev_replace_start.cold+0xf9/0x140 [btrfs]
  [ 1892.074346] CPU: 2 PID: 2244 Comm: btrfs Not tainted 5.5.0-rc7-default+ #942
  [ 1892.079956] RIP: 0010:btrfs_dev_replace_start.cold+0xf9/0x140 [btrfs]

  [ 1892.096576] RSP: 0018:ffffbb58c7b3fd10 EFLAGS: 00010286
  [ 1892.098311] RAX: 00000000ffffffe4 RBX: 0000000000000001 RCX: 8888888888888889
  [ 1892.100342] RDX: 0000000000000001 RSI: ffff9e889645f5d8 RDI: ffffffff92821080
  [ 1892.102291] RBP: ffff9e889645c000 R08: 000001b8878fe1f6 R09: 0000000000000000
  [ 1892.104239] R10: ffffbb58c7b3fd08 R11: 0000000000000000 R12: ffff9e88a0017000
  [ 1892.106434] R13: ffff9e889645f608 R14: ffff9e88794e1000 R15: ffff9e88a07b5200
  [ 1892.108642] FS:  00007fcaed3f18c0(0000) GS:ffff9e88bda00000(0000) knlGS:0000000000000000
  [ 1892.111558] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
  [ 1892.113492] CR2: 00007f52509ff420 CR3: 00000000603dd002 CR4: 0000000000160ee0

  [ 1892.115814] Call Trace:
  [ 1892.116896]  btrfs_dev_replace_by_ioctl+0x35/0x60 [btrfs]
  [ 1892.118962]  btrfs_ioctl+0x1d62/0x2550 [btrfs]

caused by the previous patch ("btrfs: scrub: Require mandatory block
group RO for dev-replace"). Hitting ENOSPC is possible and could happen
when the block group is set read-only, preventing NOCOW writes to the
area that's being accessed by dev-replace.

This has happend with scratch devices of size 12G but not with 5G and
20G, so this is depends on timing and other activity on the filesystem.
The whole replace operation is restartable, the space state should be
examined by the user in any case.

The error code is propagated back to the ioctl caller so the kernel
warning is causing false alerts.
Signed-off-by: David Sterba <dsterba@suse.com>

The cxgb3 driver for "Chelsio T3-based gigabit and 10Gb Ethernet
adapters" implements a custom ioctl as SIOCCHIOCTL/SIOCDEVPRIVATE in
cxgb_extension_ioctl().

One of the subcommands of the ioctl is CHELSIO_GET_MEM, which appears
to read memory directly out of the adapter and return it to userspace.
It's not entirely clear what the contents of the adapter memory
contains, but the assumption is that it shouldn't be accessible to all
users.

So add a CAP_NET_ADMIN check to the CHELSIO_GET_MEM case. Put it after
the is_offload() check, which matches two of the other subcommands in
the same function which also check for is_offload() and CAP_NET_ADMIN.

Found by Ilja by code inspection, not tested as I don't have the
required hardware.
Reported-by: Ilja Van Sprundel <ivansprundel@ioactive.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Signed-off-by: David S. Miller <davem@davemloft.net>

Before commit 7587935c ("net: bcmgenet: move NAPI initialization to
ring initialization") moved the code, this used to be
netif_tx_napi_add(), but we lost that small semantic change in the
process, restore that.

Fixes: 7587935c ("net: bcmgenet: move NAPI initialization to ring initialization")
Signed-off-by: Florian Fainelli <f.fainelli@gmail.com>
Acked-by: Doug Berger <opendmb@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Reflecting new realities.
Signed-off-by: Jon Maloy <jmaloy@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Use generic device API to get phy mode to fix probe failure
with ACPI based devices.
Signed-off-by: Ajay Gupta <ajayg@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Pull input fixes from Dmitry Torokhov:

 - add sanity checks to USB endpoints in various dirvers

 - max77650-onkey was missing an OF table which was preventing module
   autoloading

 - a revert and a different fix for F54 handling in Synaptics dirver

 - a fixup for handling register in pm8xxx vibrator driver

* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/dtor/input:
  Input: pm8xxx-vib - fix handling of separate enable register
  Input: keyspan-remote - fix control-message timeouts
  Input: max77650-onkey - add of_match table
  Input: rmi_f54 - read from FIFO in 32 byte blocks
  Revert "Input: synaptics-rmi4 - don't increment rmiaddr for SMBus transfers"
  Input: sur40 - fix interface sanity checks
  Input: gtco - drop redundant variable reinit
  Input: gtco - fix extra-descriptor debug message
  Input: gtco - fix endpoint sanity check
  Input: aiptek - use descriptors of current altsetting
  Input: aiptek - fix endpoint sanity check
  Input: pegasus_notetaker - fix endpoint sanity check
  Input: sun4i-ts - add a check for devm_thermal_zone_of_sensor_register
  Input: evdev - convert kzalloc()/vzalloc() to kvzalloc()

Merge tag 'omap-for-fixes-whenever-signed' of git://git.kernel.org/pub/scm/linux/kernel/git/tmlind/linux-omap into arm/fixes

Few minor fixes for omaps

Looks like we have wrong default memory size for beaglebone black,
it has at least 512 MB of RAM and not 256 MB. This causes an issue
when booted with GRUB2 that does not seem to pass memory info to
the kernel.

And for am43x-epos-evm the SPI pin directions need to be configured
for SPI to work.

* tag 'omap-for-fixes-whenever-signed' of git://git.kernel.org/pub/scm/linux/kernel/git/tmlind/linux-omap:
  ARM: dts: am43x-epos-evm: set data pin directions for spi0 and spi1
  ARM: dts: am335x-boneblack-common: fix memory size

Link: https://lore.kernel.org/r/pull-1579895109-287828@atomide.comSigned-off-by: Olof Johansson <olof@lixom.net>

Merge tag 'tee-optee-fix2-for-5.5' of https://git.linaro.org:/people/jens.wiklander/linux-tee into arm/fixes

Fix OP-TEE compile error with nommu

* tag 'tee-optee-fix2-for-5.5' of https://git.linaro.org:/people/jens.wiklander/linux-tee:
  tee: optee: Fix compilation issue with nommu

Link: https://lore.kernel.org/r/20200123101310.GA10320@jaxSigned-off-by: Olof Johansson <olof@lixom.net>

When TCP out-of-order is identified (unexpected tcp seq mismatch), driver
analyzes the packet and decides what handling should it get:
1. go to accelerated path (to be encrypted in HW),
2. go to regular xmit path (send w/o encryption),
3. drop.

Packets marked with skb->decrypted by the TLS stack in the TX flow skips
SW encryption, and rely on the HW offload.
Verify that such packets are never sent un-encrypted on the wire.
Add a WARN to catch such bugs, and prefer dropping the packet in these cases.

Fixes: 46a3ea98 ("net/mlx5e: kTLS, Enhance TX resync flow")
Signed-off-by: Tariq Toukan <tariqt@mellanox.com>
Signed-off-by: Boris Pismenny <borisp@mellanox.com>
Reviewed-by: Boris Pismenny <borisp@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>

The call to tx_post_resync_params() is done earlier in the flow,
the post of the control WQEs is unnecessarily repeated. Remove it.

Fixes: 700ec497 ("net/mlx5e: kTLS, Fix missing SQ edge fill")
Signed-off-by: Tariq Toukan <tariqt@mellanox.com>
Signed-off-by: Boris Pismenny <borisp@mellanox.com>
Reviewed-by: Boris Pismenny <borisp@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>

There are the following cases:

1. Packet ends before start marker: bypass offload.
2. Packet starts before start marker and ends after it: drop,
   not supported, breaks contract with kernel.
3. packet ends before tls record info starts: drop,
   this packet was already acknowledged and its record info
   was released.

Add the above as comment in code.

Mind possible wraparounds of the TCP seq, replace the simple comparison
with a call to the TCP before() method.

In addition, remove logic that handles negative sync_len values,
as it became impossible.

Fixes: d2ead1f3 ("net/mlx5e: Add kTLS TX HW offload support")
Fixes: 46a3ea98 ("net/mlx5e: kTLS, Enhance TX resync flow")
Signed-off-by: Tariq Toukan <tariqt@mellanox.com>
Signed-off-by: Boris Pismenny <borisp@mellanox.com>
Reviewed-by: Boris Pismenny <borisp@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>

Currently VF in LEGACY mode are not able to go up. Also in OFFLOADS
mode, when switching to it first time, VF can go up independently to
his representor, which is not expected.
Perform clearing of VF config when switching modes and set link state
to AUTO as default value. Also, when switching to OFFLOADS mode set
link state to DOWN, which allow VF link state to be controlled by its
REP.

Fixes: 1ab2068a ("net/mlx5: Implement vports admin state backup/restore")
Fixes: 556b9d16 ("net/mlx5: Clear VF's configuration on disabling SRIOV")
Signed-off-by: Dmytro Linkin <dmitrolin@mellanox.com>
Signed-off-by: Roi Dayan <roid@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>

Use raw_smp_processor_id instead of smp_processor_id() otherwise we will
get the following trace in debug-kernel:
	BUG: using smp_processor_id() in preemptible [00000000] code: devlink
	caller is dr_create_cq.constprop.2+0x31d/0x970 [mlx5_core]
	Call Trace:
	dump_stack+0x9a/0xf0
	debug_smp_processor_id+0x1f3/0x200
	dr_create_cq.constprop.2+0x31d/0x970
	genl_family_rcv_msg+0x5fd/0x1170
	genl_rcv_msg+0xb8/0x160
	netlink_rcv_skb+0x11e/0x340

Fixes: 297ccceb ("net/mlx5: DR, Expose an internal API to issue RDMA operations")
Signed-off-by: Erez Shitrit <erezsh@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>

Since the implementation relies on limiting the VF transmit rate to
simulate ingress rate limiting, and since either uplink representor or
ecpf are not associated with a VF, we limit the rate limit configuration
for those ports.

Fixes: fcb64c0f ("net/mlx5: E-Switch, add ingress rate support")
Signed-off-by: Eli Cohen <eli@mellanox.com>
Reviewed-by: Roi Dayan <roid@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>

The current code handles only counters that attached to dest, we still
have the cases where we have counter on non-dest, like over drop etc.

Fixes: 6a48faee ("net/mlx5: Add direct rule fs_cmd implementation")
Signed-off-by: Hamdan Igbaria <hamdani@mellanox.com>
Signed-off-by: Erez Shitrit <erezsh@mellanox.com>
Reviewed-by: Alex Vesker <valex@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>

Add the upcoming ConnectX-7 device ID.

Fixes: 85327a9c ("net/mlx5: Update the list of the PCI supported devices")
Signed-off-by: Meir Lichtinger <meirl@mellanox.com>
Reviewed-by: Eran Ben Elisha <eranbe@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>

The pool sizes represent the pool sizes in the fw. when we request
a pool size from fw, it will return the next possible group.
We track how many pools the fw has left and start requesting groups
from the big to the small.
When we start request 4k group, which doesn't exists in fw, fw
wants to allocate the next possible size, 64k, but will fail since
its exhausted. The correct smallest pool size in fw is 128 and not 4k.

Fixes: e52c2802 ("net/mlx5: E-Switch, Add chains and priorities")
Signed-off-by: Paul Blakey <paulb@mellanox.com>
Reviewed-by: Roi Dayan <roid@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>

skb->csum is updated incorrectly, when manipulation for
NF_NAT_MANIP_SRC\DST is done on IPV6 packet.

Fix:
There is no need to update skb->csum in inet_proto_csum_replace16(),
because update in two fields a.) IPv6 src/dst address and b.) L4 header
checksum cancels each other for skb->csum calculation. Whereas
inet_proto_csum_replace4 function needs to update skb->csum, because
update in 3 fields a.) IPv4 src/dst address, b.) IPv4 Header checksum
and c.) L4 header checksum results in same diff as L4 Header checksum
for skb->csum calculation.

[ pablo@netfilter.org: a few comestic documentation edits ]
Signed-off-by: Praveen Chaudhary <pchaudhary@linkedin.com>
Signed-off-by: Zhenggen Xu <zxu@linkedin.com>
Signed-off-by: Andy Stracner <astracner@linkedin.com>
Reviewed-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>

This patch introduces a list of pending module requests. This new module
list is composed of nft_module_request objects that contain the module
name and one status field that tells if the module has been already
loaded (the 'done' field).

In the first pass, from the preparation phase, the netlink command finds
that a module is missing on this list. Then, a module request is
allocated and added to this list and nft_request_module() returns
-EAGAIN. This triggers the abort path with the autoload parameter set on
from nfnetlink, request_module() is called and the module request enters
the 'done' state. Since the mutex is released when loading modules from
the abort phase, the module list is zapped so this is iteration occurs
over a local list. Therefore, the request_module() calls happen when
object lists are in consistent state (after fulling aborting the
transaction) and the commit list is empty.

On the second pass, the netlink command will find that it already tried
to load the module, so it does not request it again and
nft_request_module() returns 0. Then, there is a look up to find the
object that the command was missing. If the module was successfully
loaded, the command proceeds normally since it finds the missing object
in place, otherwise -ENOENT is reported to userspace.

This patch also updates nfnetlink to include the reason to enter the
abort phase, which is required for this new autoload module rationale.

Fixes: ec7470b8 ("netfilter: nf_tables: store transaction list locally while requesting module")
Reported-by: syzbot+29125d208b3dae9a7019@syzkaller.appspotmail.com
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>

This new helper function validates that unknown family and chain type
coming from userspace do not trigger an out-of-bound array access. Bail
out in case __nft_chain_type_get() returns NULL from
nft_chain_parse_hook().

Fixes: 9370761c ("netfilter: nf_tables: convert built-in tables/chains to chain types")
Reported-by: syzbot+156a04714799b1d480bc@syzkaller.appspotmail.com
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>

In the nft_indr_block_cb the chain should check the flag with
NFT_CHAIN_HW_OFFLOAD.

Fixes: 9a32669f ("netfilter: nf_tables_offload: support indr block call")
Signed-off-by: wenxu <wenxu@ucloud.cn>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>