net: stmmac: clarify difference between "interface" and "phy_interface"

Clarify the difference between "interface" and "phy_interface" in
struct plat_stmmacenet_data, both by adding a comment, and also
renaming "interface" to be "mac_interface". The difference between
these are:

 MAC ----- optional PCS ----- SerDes ----- optional PHY ----- Media
       ^                               ^
 mac_interface                   phy_interface

Note that phylink currently only deals with phy_interface.

Signed-off-by: Russell King (Oracle) <rmk+kernel@armlinux.org.uk>
Link: https://lore.kernel.org/r/E1qZq83-005tts-6K@rmk-PC.armlinux.org.uk
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

r8152: add vendor/device ID pair for D-Link DUB-E250

The D-Link DUB-E250 is an RTL8156 based 2.5G Ethernet controller.

Add the vendor and product ID values to the driver. This makes Ethernet
work with the adapter.

Signed-off-by: Antonio Napolitano <anton@polit.no>
Link: https://lore.kernel.org/r/CV200KJEEUPC.WPKAHXCQJ05I@mercurius
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

Merge branch 'devlink-finish-file-split-and-get-retire-leftover-c'

Jiri Pirko says:

====================
devlink: finish file split and get retire leftover.c

This patchset finishes a move Jakub started and Moshe continued in the
past. I was planning to do this for a long time, so here it is, finally.

This patchset does not change any behaviour. It just splits leftover.c
into per-object files and do necessary changes, like declaring functions
used from other code, on the way.

The last 3 patches are pushing the rest of the code into appropriate
existing files.
====================

Link: https://lore.kernel.org/r/20230828061657.300667-1-jiri@resnulli.us
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

devlink: move devlink_notify_register/unregister() to dev.c

At last, move the last bits out of leftover.c,
the devlink_notify_register/unregister() functions to dev.c

Signed-off-by: Jiri Pirko <jiri@nvidia.com>
Link: https://lore.kernel.org/r/20230828061657.300667-16-jiri@resnulli.us
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

devlink: move small_ops definition into netlink.c

Move the generic netlink small_ops definition where they are consumed,
into netlink.c

Signed-off-by: Jiri Pirko <jiri@nvidia.com>
Link: https://lore.kernel.org/r/20230828061657.300667-15-jiri@resnulli.us
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

devlink: move tracepoint definitions into core.c

Move remaining tracepoint definitions to most suitable file core.c.

Signed-off-by: Jiri Pirko <jiri@nvidia.com>
Link: https://lore.kernel.org/r/20230828061657.300667-14-jiri@resnulli.us
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

devlink: push linecard related code into separate file

Cut out another chunk from leftover.c and put linecard related code
into a separate file.

Signed-off-by: Jiri Pirko <jiri@nvidia.com>
Link: https://lore.kernel.org/r/20230828061657.300667-13-jiri@resnulli.us
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

devlink: push rate related code into separate file

Cut out another chunk from leftover.c and put rate related code
into a separate file.

Signed-off-by: Jiri Pirko <jiri@nvidia.com>
Link: https://lore.kernel.org/r/20230828061657.300667-12-jiri@resnulli.us
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

devlink: push trap related code into separate file

Cut out another chunk from leftover.c and put trap related code
into a separate file.

Signed-off-by: Jiri Pirko <jiri@nvidia.com>
Link: https://lore.kernel.org/r/20230828061657.300667-11-jiri@resnulli.us
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

devlink: use tracepoint_enabled() helper

In preparation for the trap code move, use tracepoint_enabled() helper
instead of trace_devlink_trap_report_enabled() which would not be
defined in that scope.

Signed-off-by: Jiri Pirko <jiri@nvidia.com>
Link: https://lore.kernel.org/r/20230828061657.300667-10-jiri@resnulli.us
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

devlink: push region related code into separate file

Cut out another chunk from leftover.c and put region related code
into a separate file.

Signed-off-by: Jiri Pirko <jiri@nvidia.com>
Link: https://lore.kernel.org/r/20230828061657.300667-9-jiri@resnulli.us
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

devlink: push param related code into separate file

Cut out another chunk from leftover.c and put param related code
into a separate file.

Signed-off-by: Jiri Pirko <jiri@nvidia.com>
Link: https://lore.kernel.org/r/20230828061657.300667-8-jiri@resnulli.us
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

devlink: push resource related code into separate file

Cut out another chunk from leftover.c and put resource related code
into a separate file.

Signed-off-by: Jiri Pirko <jiri@nvidia.com>
Link: https://lore.kernel.org/r/20230828061657.300667-7-jiri@resnulli.us
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

devlink: push dpipe related code into separate file

Cut out another chunk from leftover.c and put dpipe related code
into a separate file.

Signed-off-by: Jiri Pirko <jiri@nvidia.com>
Link: https://lore.kernel.org/r/20230828061657.300667-6-jiri@resnulli.us
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

devlink: move and rename devlink_dpipe_send_and_alloc_skb() helper

Since both dpipe and resource code is using this helper, in preparation
for code split to separate files, move
devlink_dpipe_send_and_alloc_skb() helper into netlink.c. Rename it on
the way.

Signed-off-by: Jiri Pirko <jiri@nvidia.com>
Link: https://lore.kernel.org/r/20230828061657.300667-5-jiri@resnulli.us
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

devlink: push shared buffer related code into separate file

Cut out another chunk from leftover.c and put sb related code
into a separate file.

Signed-off-by: Jiri Pirko <jiri@nvidia.com>
Link: https://lore.kernel.org/r/20230828061657.300667-4-jiri@resnulli.us
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

devlink: push port related code into separate file

Cut out another chunk from leftover.c and put port related code
into a separate file.

Signed-off-by: Jiri Pirko <jiri@nvidia.com>
Link: https://lore.kernel.org/r/20230828061657.300667-3-jiri@resnulli.us
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

devlink: push object register/unregister notifications into separate helpers

In preparations of leftover.c split to individual files, avoid need to
have object structures exposed in devl_internal.h and allow to have them
maintained in object files.

The register/unregister notifications need to know the structures
to iterate lists. To avoid the need, introduce per-object
register/unregister notification helpers and use them.

Signed-off-by: Jiri Pirko <jiri@nvidia.com>
Link: https://lore.kernel.org/r/20230828061657.300667-2-jiri@resnulli.us
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

inet: fix IP_TRANSPARENT error handling

My recent patch forgot to change error handling for IP_TRANSPARENT
socket option.

WARNING: bad unlock balance detected!
6.5.0-rc7-syzkaller-01717-g59da9885767a #0 Not tainted
-------------------------------------
syz-executor151/5028 is trying to release lock (sk_lock-AF_INET) at:
[<ffffffff88213983>] sockopt_release_sock+0x53/0x70 net/core/sock.c:1073
but there are no more locks to release!

other info that might help us debug this:
1 lock held by syz-executor151/5028:

stack backtrace:
CPU: 0 PID: 5028 Comm: syz-executor151 Not tainted 6.5.0-rc7-syzkaller-01717-g59da9885767a #0
Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 07/26/2023
Call Trace:
<TASK>
__dump_stack lib/dump_stack.c:88 [inline]
dump_stack_lvl+0xd9/0x1b0 lib/dump_stack.c:106
__lock_release kernel/locking/lockdep.c:5438 [inline]
lock_release+0x4b5/0x680 kernel/locking/lockdep.c:5781
sock_release_ownership include/net/sock.h:1824 [inline]
release_sock+0x175/0x1b0 net/core/sock.c:3527
sockopt_release_sock+0x53/0x70 net/core/sock.c:1073
do_ip_setsockopt+0x12c1/0x3640 net/ipv4/ip_sockglue.c:1364
ip_setsockopt+0x59/0xe0 net/ipv4/ip_sockglue.c:1419
raw_setsockopt+0x218/0x290 net/ipv4/raw.c:833
__sys_setsockopt+0x2cd/0x5b0 net/socket.c:2305
__do_sys_setsockopt net/socket.c:2316 [inline]
__se_sys_setsockopt net/socket.c:2313 [inline]

Fixes: 4bd0623f04ee ("inet: move inet->transparent to inet->inet_flags")
Reported-by: syzbot <syzkaller@googlegroups.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Soheil Hassas Yeganeh <soheil@google.com>
Cc: Simon Horman <horms@kernel.org>
Cc: Matthieu Baerts <matthieu.baerts@tessares.net>
Signed-off-by: David S. Miller <davem@davemloft.net>

selftests: bonding: create directly devices in the target namespaces

If failed to set link1_1 to netns client, we should delete link1_1 in the
cleanup path. But if set link1_1 to netns client successfully, delete
link1_1 will report warning. So it will be safer creating directly the
devices in the target namespaces.

Reported-by: Hangbin Liu <liuhangbin@gmail.com>
Closes: https://lore.kernel.org/all/ZNyJx1HtXaUzOkNA@Laptop-X1/
Signed-off-by: Zhengchao Shao <shaozhengchao@huawei.com>
Acked-by: Hangbin Liu <liuhangbin@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

net: Make consumed action consistent in sch_handle_egress

While looking at TC_ACT_* handling, the TC_ACT_CONSUMED is only handled in
sch_handle_ingress but not sch_handle_egress. This was added via cd11b164073b
("net/tc: introduce TC_ACT_REINSERT.") and e5cf1baf92cb ("act_mirred: use
TC_ACT_REINSERT when possible") and later got renamed into TC_ACT_CONSUMED
via 720f22fed81b ("net: sched: refactor reinsert action").

The initial work was targeted for ovs back then and only needed on ingress,
and the mirred action module also restricts it to only that. However, given
it's an API contract it would still make sense to make this consistent to
sch_handle_ingress and handle it on egress side in the same way, that is,
setting return code to "success" and returning NULL back to the caller as
otherwise an action module sitting on egress returning TC_ACT_CONSUMED could
lead to an UAF when untreated.

Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Reviewed-by: Simon Horman <horms@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

net: Fix skb consume leak in sch_handle_egress

Fix a memory leak for the tc egress path with TC_ACT_{STOLEN,QUEUED,TRAP}:

  [...]
  unreferenced object 0xffff88818bcb4f00 (size 232):
  comm "softirq", pid 0, jiffies 4299085078 (age 134.028s)
  hex dump (first 32 bytes):
    00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00  ................
    00 80 70 61 81 88 ff ff 00 41 31 14 81 88 ff ff  ..pa.....A1.....
  backtrace:
    [<ffffffff9991b938>] kmem_cache_alloc_node+0x268/0x400
    [<ffffffff9b3d9231>] __alloc_skb+0x211/0x2c0
    [<ffffffff9b3f0c7e>] alloc_skb_with_frags+0xbe/0x6b0
    [<ffffffff9b3bf9a9>] sock_alloc_send_pskb+0x6a9/0x870
    [<ffffffff9b6b3f00>] __ip_append_data+0x14d0/0x3bf0
    [<ffffffff9b6ba24e>] ip_append_data+0xee/0x190
    [<ffffffff9b7e1496>] icmp_push_reply+0xa6/0x470
    [<ffffffff9b7e4030>] icmp_reply+0x900/0xa00
    [<ffffffff9b7e42e3>] icmp_echo.part.0+0x1a3/0x230
    [<ffffffff9b7e444d>] icmp_echo+0xcd/0x190
    [<ffffffff9b7e9566>] icmp_rcv+0x806/0xe10
    [<ffffffff9b699bd1>] ip_protocol_deliver_rcu+0x351/0x3d0
    [<ffffffff9b699f14>] ip_local_deliver_finish+0x2b4/0x450
    [<ffffffff9b69a234>] ip_local_deliver+0x174/0x1f0
    [<ffffffff9b69a4b2>] ip_sublist_rcv_finish+0x1f2/0x420
    [<ffffffff9b69ab56>] ip_sublist_rcv+0x466/0x920
  [...]

I was able to reproduce this via:

  ip link add dev dummy0 type dummy
  ip link set dev dummy0 up
  tc qdisc add dev eth0 clsact
  tc filter add dev eth0 egress protocol ip prio 1 u32 match ip protocol 1 0xff action mirred egress redirect dev dummy0
  ping 1.1.1.1
  <stolen>

After the fix, there are no kmemleak reports with the reproducer. This is
in line with what is also done on the ingress side, and from debugging the
skb_unref(skb) on dummy xmit and sch_handle_egress() side, it is visible
that these are two different skbs with both skb_unref(skb) as true. The two
seen skbs are due to mirred doing a skb_clone() internally as use_reinsert
is false in tcf_mirred_act() for egress. This was initially reported by Gal.

Fixes: e420bed02507 ("bpf: Add fd-based tcx multi-prog infra with link support")
Reported-by: Gal Pressman <gal@nvidia.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Link: https://lore.kernel.org/bpf/bdfc2640-8f65-5b56-4472-db8e2b161aab@nvidia.com
Reviewed-by: Simon Horman <horms@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

Merge branch 'octeontx2-af-misc-mac-block-changes'

Hariprasad Kelam says:

====================
octeontx2-af: misc MAC block changes

This series of patches adds recent changes added in MAC (CGX/RPM) block.

Patch1: Adds new LMAC mode supported by CN10KB silicon

Patch2: In a scenario where system boots with no cgx devices, currently
        AF driver treats this as error as a result no interfaces will work.
        This patch relaxes this check, such that non cgx mapped netdev
        devices will work.

Patch3: This patch adds required lmac validation in MAC block APIs.

Patch4: Prints error message incase, no netdev is mapped with given
        cgx,lmac pair.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>

octeontx2-af: print error message incase of invalid pf mapping

During AF driver initialization, it creates a mapping between pf to
cgx,lmac pair. Whenever there is a physical link change, using this
mapping driver forwards the message to the associated netdev.

This patch prints error message incase of cgx,lmac pair is not
associated with any pf netdev.

Signed-off-by: Hariprasad Kelam <hkelam@marvell.com>
Signed-off-by: Sunil Kovvuri Goutham <sgoutham@marvell.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

octeontx2-af: Add validation of lmac

With the addition of new MAC blocks like CN10K RPM and CN10KB
RPM_USX, LMACs are noncontiguous. Though in most of the functions,
lmac validation checks exist but in few functions they are missing.
This patch adds the same.

Signed-off-by: Hariprasad Kelam <hkelam@marvell.com>
Signed-off-by: Sunil Kovvuri Goutham <sgoutham@marvell.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

octeontx2-af: Don't treat lack of CGX interfaces as error

Don't treat lack of CGX LMACs on the system as a error.
Instead ignore it so that LBK VFs are created and can be used.

Signed-off-by: Sunil Goutham <sgoutham@marvell.com>
Signed-off-by: Hariprasad Kelam <hkelam@marvell.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

octeontx2-af: CN10KB: Add USGMII LMAC mode

Upon physical link change, firmware reports to the kernel about the
change along with the details like speed, lmac_type_id, etc.
Kernel derives lmac_type based on lmac_type_id received from firmware.

This patch extends current lmac list with new USGMII mode supported
by CN10KB RPM block.

Signed-off-by: Hariprasad Kelam <hkelam@marvell.com>
Signed-off-by: Sunil Kovvuri Goutham <sgoutham@marvell.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

dt-bindings: net: xilinx_gmii2rgmii: Convert to json schema

Convert the Xilinx GMII to RGMII Converter device tree binding
documentation to json schema.
This converter is usually used as gem <---> gmii2rgmii <---> external phy
and, it's phy-handle should point to the phandle of the external phy.

Signed-off-by: Pranavi Somisetty <pranavi.somisetty@amd.com>
Signed-off-by: Harini Katakam <harini.katakam@amd.com>
Reviewed-by: Conor Dooley <conor.dooley@microchip.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Merge branch 'tls-expand-tls_cipher_size_desc-to-simplify-getsockopt-setsockopt'

Sabrina Dubroca says:

====================
tls: expand tls_cipher_size_desc to simplify getsockopt/setsockopt

Commit 2d2c5ea24243 ("net/tls: Describe ciphers sizes by const
structs") introduced tls_cipher_size_desc to describe the size of the
fields of the per-cipher crypto_info structs, and commit ea7a9d88ba21
("net/tls: Use cipher sizes structs") used it, but only in
tls_device.c and tls_device_fallback.c, and skipped converting similar
code in tls_main.c and tls_sw.c.

This series expands tls_cipher_size_desc (renamed to tls_cipher_desc
to better fit this expansion) to fully describe a cipher:
 - offset of the fields within the per-cipher crypto_info
 - size of the full struct (for copies to/from userspace)
 - offload flag
 - algorithm name used by SW crypto

With these additions, we can remove ~350L of
     switch (crypto_info->cipher_type) { ... }
from tls_set_device_offload, tls_sw_fallback_init,
do_tls_getsockopt_conf, do_tls_setsockopt_conf, tls_set_sw_offload
(mainly do_tls_getsockopt_conf and tls_set_sw_offload).

This series also adds the ARIA ciphers to the tls selftests, and some
more getsockopt/setsockopt tests to cover more of the code changed by
this series.
====================

Link: https://lore.kernel.org/r/cover.1692977948.git.sd@queasysnail.net
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

tls: get cipher_name from cipher_desc in tls_set_sw_offload

tls_cipher_desc also contains the algorithm name needed by
crypto_alloc_aead, use it.

Finally, use get_cipher_desc to check if the cipher_type coming from
userspace is valid, and remove the cipher_type switch.

Signed-off-by: Sabrina Dubroca <sd@queasysnail.net>
Link: https://lore.kernel.org/r/53d021d80138aa125a9cef4468aa5ce531975a7b.1692977948.git.sd@queasysnail.net
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

tls: use tls_cipher_desc to access per-cipher crypto_info in tls_set_sw_offload

The crypto_info_* helpers allow us to fetch pointers into the
per-cipher crypto_info's data.

Signed-off-by: Sabrina Dubroca <sd@queasysnail.net>
Link: https://lore.kernel.org/r/c23af110caf0af6b68de2f86c58064913e2e902a.1692977948.git.sd@queasysnail.net
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

tls: use tls_cipher_desc to get per-cipher sizes in tls_set_sw_offload

We can get rid of some local variables, but we have to keep nonce_size
because tls1.3 uses nonce_size = 0 for all ciphers.

We can also drop the runtime sanity checks on iv/rec_seq/tag size,
since we have compile time checks on those values.

Signed-off-by: Sabrina Dubroca <sd@queasysnail.net>
Link: https://lore.kernel.org/r/deed9c4430a62c31751a72b8c03ad66ffe710717.1692977948.git.sd@queasysnail.net
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

tls: use tls_cipher_desc to simplify do_tls_getsockopt_conf

Every cipher uses the same code to update its crypto_info struct based
on the values contained in the cctx, with only the struct type and
size/offset changing. We can get those  from tls_cipher_desc, and use
a single pair of memcpy and final copy_to_user.

Signed-off-by: Sabrina Dubroca <sd@queasysnail.net>
Link: https://lore.kernel.org/r/c21a904b91e972bdbbf9d1c6d2731ccfa1eedf72.1692977948.git.sd@queasysnail.net
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

tls: get crypto_info size from tls_cipher_desc in do_tls_setsockopt_conf

We can simplify do_tls_setsockopt_conf using tls_cipher_desc. Also use
get_cipher_desc's result to check if the cipher_type coming from
userspace is valid.

Signed-off-by: Sabrina Dubroca <sd@queasysnail.net>
Link: https://lore.kernel.org/r/e97658eb4c6a5832f8ba20a06c4f36a77763c59e.1692977948.git.sd@queasysnail.net
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

tls: expand use of tls_cipher_desc in tls_sw_fallback_init

tls_sw_fallback_init already gets the key and tag size from
tls_cipher_desc. We can now also check that the cipher type is valid,
and stop hard-coding the algorithm name passed to crypto_alloc_aead.

Signed-off-by: Sabrina Dubroca <sd@queasysnail.net>
Link: https://lore.kernel.org/r/c8c94b8fcafbfb558e09589c1f1ad48dbdf92f76.1692977948.git.sd@queasysnail.net
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

tls: allocate the fallback aead after checking that the cipher is valid

No need to allocate the aead if we're going to fail afterwards.

Signed-off-by: Sabrina Dubroca <sd@queasysnail.net>
Link: https://lore.kernel.org/r/335e32511ed55a0b30f3f81a78fa8f323b3bdf8f.1692977948.git.sd@queasysnail.net
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

tls: expand use of tls_cipher_desc in tls_set_device_offload

tls_set_device_offload is already getting iv and rec_seq sizes from
tls_cipher_desc. We can now also check if the cipher_type coming from
userspace is valid and can be offloaded.

We can also remove the runtime check on rec_seq, since we validate it
at compile time.

Signed-off-by: Sabrina Dubroca <sd@queasysnail.net>
Link: https://lore.kernel.org/r/8ab71b8eca856c7aaf981a45fe91ac649eb0e2e9.1692977948.git.sd@queasysnail.net
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

tls: validate cipher descriptions at compile time

Signed-off-by: Sabrina Dubroca <sd@queasysnail.net>
Link: https://lore.kernel.org/r/b38fb8cf60e099e82ae9979c3c9c92421042417c.1692977948.git.sd@queasysnail.net
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

tls: extend tls_cipher_desc to fully describe the ciphers

- add nonce, usually equal to iv_size but not for chacha
 - add offsets into the crypto_info for each field
 - add algorithm name
 - add offloadable flag

Also add helpers to access each field of a crypto_info struct
described by a tls_cipher_desc.

Signed-off-by: Sabrina Dubroca <sd@queasysnail.net>
Link: https://lore.kernel.org/r/39d5f476d63c171097764e8d38f6f158b7c109ae.1692977948.git.sd@queasysnail.net
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

tls: rename tls_cipher_size_desc to tls_cipher_desc

We're going to add other fields to it to fully describe a cipher, so
the "_size" name won't match the contents.

Signed-off-by: Sabrina Dubroca <sd@queasysnail.net>
Link: https://lore.kernel.org/r/76ca6c7686bd6d1534dfa188fb0f1f6fabebc791.1692977948.git.sd@queasysnail.net
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

tls: reduce size of tls_cipher_size_desc

tls_cipher_size_desc indexes ciphers by their type, but we're not
using indices 0..50 of the array. Each struct tls_cipher_size_desc is
20B, so that's a lot of unused memory. We can reindex the array
starting at the lowest used cipher_type.

Introduce the get_cipher_size_desc helper to find the right item and
avoid out-of-bounds accesses, and make tls_cipher_size_desc's size
explicit so that gcc reminds us to update TLS_CIPHER_MIN/MAX when we
add a new cipher.

Signed-off-by: Sabrina Dubroca <sd@queasysnail.net>
Link: https://lore.kernel.org/r/5e054e370e240247a5d37881a1cd93a67c15f4ca.1692977948.git.sd@queasysnail.net
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

tls: add TLS_CIPHER_ARIA_GCM_* to tls_cipher_size_desc

Signed-off-by: Sabrina Dubroca <sd@queasysnail.net>
Link: https://lore.kernel.org/r/b2e0fb79e6d0a4478be9bf33781dc9c9281c9d56.1692977948.git.sd@queasysnail.net
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

tls: move tls_cipher_size_desc to net/tls/tls.h

It's only used in net/tls/*, no need to bloat include/net/tls.h.

Signed-off-by: Sabrina Dubroca <sd@queasysnail.net>
Link: https://lore.kernel.org/r/dd9fad80415e5b3575b41f56b331871038362eab.1692977948.git.sd@queasysnail.net
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

selftests: tls: test some invalid inputs for setsockopt

This test will need to be updated if new ciphers are added.

Signed-off-by: Sabrina Dubroca <sd@queasysnail.net>
Link: https://lore.kernel.org/r/bfcfa9cffda56d2064296ab7c99a05775dd4c28e.1692977948.git.sd@queasysnail.net
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

selftests: tls: add getsockopt test

The kernel accepts fetching either just the version and cipher type,
or exactly the per-cipher struct. Also check that getsockopt returns
what we just passed to the kernel.

Signed-off-by: Sabrina Dubroca <sd@queasysnail.net>
Link: https://lore.kernel.org/r/81a007ca13de9a74f4af45635d06682cdb385a54.1692977948.git.sd@queasysnail.net
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

selftests: tls: add test variants for aria-gcm

Only supported for TLS1.2.

Signed-off-by: Sabrina Dubroca <sd@queasysnail.net>
Link: https://lore.kernel.org/r/ccf4a4d3f3820f8ff30431b7629f5210cb33fa89.1692977948.git.sd@queasysnail.net
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

Merge branch 'tools-net-ynl-add-support-for-netlink-raw-families'

Donald Hunter says:

====================
tools/net/ynl: Add support for netlink-raw families

This patchset adds support for netlink-raw families such as rtnetlink.

Patch 1 fixes a typo in existing schemas
Patch 2 contains the schema definition
Patches 3 & 4 update the schema documentation
Patches 5 - 9 extends ynl
Patches 10 - 12 add several netlink-raw specs

The netlink-raw schema is very similar to genetlink-legacy and I thought
about making the changes there and symlinking to it. On balance I
thought that might be problematic for accurate schema validation.

rtnetlink doesn't seem to fit into unified or directional message
enumeration models. It seems like an 'explicit' model would be useful,
to force the schema author to specify the message ids directly.

There is not yet support for notifications because ynl currently doesn't
support defining 'event' properties on a 'do' operation. The message ids
are shared so ops need to be both sync and async. I plan to look at this
in a future patch.

The link and route messages contain different nested attributes
dependent on the type of link or route. Decoding these will need some
kind of attr-space selection that uses the value of another attribute as
the selector key. These nested attributes have been left with type
'binary' for now.
====================

Link: https://lore.kernel.org/r/20230825122756.7603-1-donald.hunter@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

doc/netlink: Add spec for rt route messages

Add schema for rt route with support for getroute, newroute and
delroute.

Routes can be dumped with filter attributes like this:

./tools/net/ynl/cli.py \
    --spec Documentation/netlink/specs/rt_route.yaml \
    --dump getroute --json '{"rtm-family": 2, "rtm-table": 254}'

Signed-off-by: Donald Hunter <donald.hunter@gmail.com>
Reviewed-by: Jacob Keller <jacob.e.keller@intel.com>
Link: https://lore.kernel.org/r/20230825122756.7603-13-donald.hunter@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

doc/netlink: Add spec for rt link messages

Add schema for rt link with support for newlink, dellink, getlink,
setlink and getstats.

A dummy link can be created like this:

sudo ./tools/net/ynl/cli.py \
    --spec Documentation/netlink/specs/rt_link.yaml \
    --do newlink --create \
    --json '{"ifname": "dummy0", "linkinfo": {"kind": "dummy"}}'

For example, offload stats can be fetched like this:

./tools/net/ynl/cli.py \
    --spec Documentation/netlink/specs/rt_link.yaml \
    --dump getstats --json '{ "filter-mask": 8 }'

Signed-off-by: Donald Hunter <donald.hunter@gmail.com>
Reviewed-by: Jacob Keller <jacob.e.keller@intel.com>
Link: https://lore.kernel.org/r/20230825122756.7603-12-donald.hunter@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

doc/netlink: Add spec for rt addr messages

Add schema for rt addr with support for:
     - newaddr, deladdr, getaddr (dump)

Signed-off-by: Donald Hunter <donald.hunter@gmail.com>
Reviewed-by: Jacob Keller <jacob.e.keller@intel.com>
Link: https://lore.kernel.org/r/20230825122756.7603-11-donald.hunter@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

tools/net/ynl: Add support for create flags

Add support for using NLM_F_REPLACE, _EXCL, _CREATE and _APPEND flags
in requests.

Signed-off-by: Donald Hunter <donald.hunter@gmail.com>
Reviewed-by: Jacob Keller <jacob.e.keller@intel.com>
Link: https://lore.kernel.org/r/20230825122756.7603-10-donald.hunter@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

tools/net/ynl: Implement nlattr array-nest decoding in ynl

Add support for the 'array-nest' attribute type that is used by several
netlink-raw families.

Signed-off-by: Donald Hunter <donald.hunter@gmail.com>
Reviewed-by: Jakub Kicinski <kuba@kernel.org>
Reviewed-by: Jacob Keller <jacob.e.keller@intel.com>
Link: https://lore.kernel.org/r/20230825122756.7603-9-donald.hunter@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

tools/net/ynl: Add support for netlink-raw families

Refactor the ynl code to encapsulate protocol specifics into
NetlinkProtocol and GenlProtocol.

Signed-off-by: Donald Hunter <donald.hunter@gmail.com>
Link: https://lore.kernel.org/r/20230825122756.7603-8-donald.hunter@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

tools/net/ynl: Fix extack parsing with fixed header genlmsg

Move decode_fixed_header into YnlFamily and add a _fixed_header_size
method to allow extack decoding to skip the fixed header.

Signed-off-by: Donald Hunter <donald.hunter@gmail.com>
Reviewed-by: Jacob Keller <jacob.e.keller@intel.com>
Link: https://lore.kernel.org/r/20230825122756.7603-7-donald.hunter@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

tools/ynl: Add mcast-group schema parsing to ynl

Add a SpecMcastGroup class to the nlspec lib.

Signed-off-by: Donald Hunter <donald.hunter@gmail.com>
Reviewed-by: Jacob Keller <jacob.e.keller@intel.com>
Link: https://lore.kernel.org/r/20230825122756.7603-6-donald.hunter@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

doc/netlink: Document the netlink-raw schema extensions

Add a doc page for netlink-raw that describes the schema attributes
needed for netlink-raw.

Signed-off-by: Donald Hunter <donald.hunter@gmail.com>
Reviewed-by: Jacob Keller <jacob.e.keller@intel.com>
Link: https://lore.kernel.org/r/20230825122756.7603-5-donald.hunter@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

doc/netlink: Update genetlink-legacy documentation

Add documentation for recently added genetlink-legacy schema attributes.
Remove statements about 'work in progress' and 'todo'.

Signed-off-by: Donald Hunter <donald.hunter@gmail.com>
Reviewed-by: Jacob Keller <jacob.e.keller@intel.com>
Link: https://lore.kernel.org/r/20230825122756.7603-4-donald.hunter@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

doc/netlink: Add a schema for netlink-raw families

This schema is largely a copy of the genetlink-legacy schema with the
following modifications:

 - change the schema id to netlink-raw
 - add a top-level protonum property, e.g. 0 (for NETLINK_ROUTE)
 - change the protocol enumeration to netlink-raw, removing the
   genetlink options.
 - replace doc references to generic netlink with raw netlink
 - add a value property to mcast-group definitions

Signed-off-by: Donald Hunter <donald.hunter@gmail.com>
Reviewed-by: Jacob Keller <jacob.e.keller@intel.com>
Link: https://lore.kernel.org/r/20230825122756.7603-3-donald.hunter@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

doc/netlink: Fix typo in genetlink-* schemas

Fix typo verion -> version in genetlink-c and genetlink-legacy.

Signed-off-by: Donald Hunter <donald.hunter@gmail.com>
Reviewed-by: Jacob Keller <jacob.e.keller@intel.com>
Link: https://lore.kernel.org/r/20230825122756.7603-2-donald.hunter@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

Merge branch 'devlink-mlx5-add-port-function-attributes-for-ipsec'

Saeed Mahameed says:

====================
{devlink,mlx5}: Add port function attributes for ipsec

From Dima:

Introduce hypervisor-level control knobs to set the functionality of PCI
VF devices passed through to guests. The administrator of a hypervisor
host may choose to change the settings of a port function from the
defaults configured by the device firmware.

The software stack has two types of IPsec offload - crypto and packet.
Specifically, the ip xfrm command has sub-commands for "state" and
"policy" that have an "offload" parameter. With ip xfrm state, both
crypto and packet offload types are supported, while ip xfrm policy can
only be offloaded in packet mode.

The series introduces two new boolean attributes of a port function:
ipsec_crypto and ipsec_packet. The goal is to provide a similar level of
granularity for controlling VF IPsec offload capabilities, which would
be aligned with the software model. This will allow users to decide if
they want both types of offload enabled for a VF, just one of them, or
none at all (which is the default).

At a high level, the difference between the two knobs is that with
ipsec_crypto, only XFRM state can be offloaded. Specifically, only the
crypto operation (Encrypt/Decrypt) is offloaded. With ipsec_packet, both
XFRM state and policy can be offloaded. Furthermore, in addition to
crypto operation offload, IPsec encapsulation is also offloaded. For
XFRM state, choosing between crypto and packet offload types is
possible. From the HW perspective, different resources may be required
for each offload type.

Examples of when a user prefers to enable IPsec packet offload for a VF
when using switchdev mode:

  $ devlink port show pci/0000:06:00.0/1
      pci/0000:06:00.0/1: type eth netdev enp6s0pf0vf0 flavour pcivf pfnum 0 vfnum 0
          function:
          hw_addr 00:00:00:00:00:00 roce enable migratable disable ipsec_crypto disable ipsec_packet disable

  $ devlink port function set pci/0000:06:00.0/1 ipsec_packet enable

  $ devlink port show pci/0000:06:00.0/1
      pci/0000:06:00.0/1: type eth netdev enp6s0pf0vf0 flavour pcivf pfnum 0 vfnum 0
          function:
          hw_addr 00:00:00:00:00:00 roce enable migratable disable ipsec_crypto disable ipsec_packet enable

This enables the corresponding IPsec capability of the function before
it's enumerated, so when the driver reads the capability from the device
firmware, it is enabled. The driver is then able to configure
corresponding features and ops of the VF net device to support IPsec
state and policy offloading.

v2: https://lore.kernel.org/netdev/20230421104901.897946-1-dchumak@nvidia.com/
====================

Link: https://lore.kernel.org/r/20230825062836.103744-1-saeed@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net/mlx5: Implement devlink port function cmds to control ipsec_packet

Implement devlink port function commands to enable / disable IPsec
packet offloads. This is used to control the IPsec capability of the
device.

When ipsec_offload is enabled for a VF, it prevents adding IPsec packet
offloads on the PF, because the two cannot be active simultaneously due
to HW constraints. Conversely, if there are any active IPsec packet
offloads on the PF, it's not allowed to enable ipsec_packet on a VF,
until PF IPsec offloads are cleared.

Signed-off-by: Dima Chumak <dchumak@nvidia.com>
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
Link: https://lore.kernel.org/r/20230825062836.103744-9-saeed@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net/mlx5: Implement devlink port function cmds to control ipsec_crypto

Implement devlink port function commands to enable / disable IPsec
crypto offloads.  This is used to control the IPsec capability of the
device.

When ipsec_crypto is enabled for a VF, it prevents adding IPsec crypto
offloads on the PF, because the two cannot be active simultaneously due
to HW constraints. Conversely, if there are any active IPsec crypto
offloads on the PF, it's not allowed to enable ipsec_crypto on a VF,
until PF IPsec offloads are cleared.

Signed-off-by: Dima Chumak <dchumak@nvidia.com>
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
Link: https://lore.kernel.org/r/20230825062836.103744-8-saeed@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net/mlx5: Provide an interface to block change of IPsec capabilities

mlx5 HW can't perform IPsec offload operation simultaneously both on PF
and VFs at the same time. While the previous patches added devlink knobs
to change IPsec capabilities dynamically, there is a need to add a logic
to block such IPsec capabilities for the cases when IPsec is already
configured.

Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
Link: https://lore.kernel.org/r/20230825062836.103744-7-saeed@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net/mlx5: Add IFC bits to support IPsec enable/disable

Add hardware definitions to allow to control IPSec capabilities.

Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
Link: https://lore.kernel.org/r/20230825062836.103744-6-saeed@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net/mlx5e: Rewrite IPsec vs. TC block interface

In the commit 366e46242b8e ("net/mlx5e: Make IPsec offload work together
with eswitch and TC"), new API to block IPsec vs. TC creation was introduced.

Internally, that API used devlink lock to avoid races with userspace, but it is
not really needed as dev->priv.eswitch is stable and can't be changed. So remove
dependency on devlink lock and move block encap code back to its original place.

Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
Link: https://lore.kernel.org/r/20230825062836.103744-5-saeed@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net/mlx5: Drop extra layer of locks in IPsec

There is no need in holding devlink lock as it gives nothing
compared to already used write mode_lock.

Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
Link: https://lore.kernel.org/r/20230825062836.103744-4-saeed@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

devlink: Expose port function commands to control IPsec packet offloads

Expose port function commands to enable / disable IPsec packet offloads,
this is used to control the port IPsec capabilities.

When IPsec packet is disabled for a function of the port (default),
function cannot offload IPsec packet operations (encapsulation and XFRM
policy offload). When enabled, IPsec packet operations can be offloaded
by the function of the port, which includes crypto operation
(Encrypt/Decrypt), IPsec encapsulation and XFRM state and policy
offload.

Example of a PCI VF port which supports IPsec packet offloads:

$ devlink port show pci/0000:06:00.0/1
    pci/0000:06:00.0/1: type eth netdev enp6s0pf0vf0 flavour pcivf pfnum 0 vfnum 0
        function:
        hw_addr 00:00:00:00:00:00 roce enable ipsec_packet disable

$ devlink port function set pci/0000:06:00.0/1 ipsec_packet enable

$ devlink port show pci/0000:06:00.0/1
    pci/0000:06:00.0/1: type eth netdev enp6s0pf0vf0 flavour pcivf pfnum 0 vfnum 0
        function:
        hw_addr 00:00:00:00:00:00 roce enable ipsec_packet enable

Signed-off-by: Dima Chumak <dchumak@nvidia.com>
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
Reviewed-by: Jiri Pirko <jiri@nvidia.com>
Link: https://lore.kernel.org/r/20230825062836.103744-3-saeed@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

devlink: Expose port function commands to control IPsec crypto offloads

Expose port function commands to enable / disable IPsec crypto offloads,
this is used to control the port IPsec capabilities.

When IPsec crypto is disabled for a function of the port (default),
function cannot offload any IPsec crypto operations (Encrypt/Decrypt and
XFRM state offloading). When enabled, IPsec crypto operations can be
offloaded by the function of the port.

Example of a PCI VF port which supports IPsec crypto offloads:

$ devlink port show pci/0000:06:00.0/1
    pci/0000:06:00.0/1: type eth netdev enp6s0pf0vf0 flavour pcivf pfnum 0 vfnum 0
        function:
        hw_addr 00:00:00:00:00:00 roce enable ipsec_crypto disable

$ devlink port function set pci/0000:06:00.0/1 ipsec_crypto enable

$ devlink port show pci/0000:06:00.0/1
    pci/0000:06:00.0/1: type eth netdev enp6s0pf0vf0 flavour pcivf pfnum 0 vfnum 0
        function:
        hw_addr 00:00:00:00:00:00 roce enable ipsec_crypto enable

Signed-off-by: Dima Chumak <dchumak@nvidia.com>
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
Reviewed-by: Jiri Pirko <jiri@nvidia.com>
Link: https://lore.kernel.org/r/20230825062836.103744-2-saeed@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

Merge branch 'iep-drver-timestamping-support'

MD Danish Anwar says:

====================
Introduce IEP driver and packet timestamping support

This series introduces Industrial Ethernet Peripheral (IEP) driver to
support timestamping of ethernet packets and thus support PTP and PPS
for PRU ICSSG ethernet ports.

This series also adds 10M full duplex support for ICSSG ethernet driver.

There are two IEP instances. IEP0 is used for packet timestamping while IEP1
is used for 10M full duplex support.

This is v7 of the series [v1]. It addresses comments made on [v6].
This series is based on linux-next(#next-20230823).

Changes from v6 to v7:
*) Dropped blank line in example section of patch 1.
*) Patch 1 previously had three examples, removed two examples and kept only
   one example as asked by Krzysztof.
*) Added Jacob Keller's RB tag in patch 5.
*) Dropped Roger's RB tags from the patches that he has authored (Patch 3 and 4)

Changes from v5 to v6:
*) Added description of IEP in commit messages of patch 2 as asked by Rob.
*) Described the items constraints properly for iep property in patch 2 as
   asked by Rob.
*) Added Roger and Simon's RB tags.

Changes from v4 to v5:
*) Added comments on why we are using readl / writel instead of regmap_read()
   / write() in icss_iep_gettime() / settime() APIs as asked by Roger.
*) Added Conor's RB tag in patch 1 and 2.

Change from v3 to v4:
*) Changed compatible in iep dt bindings. Now each SoC has their own compatible
   in the binding with "ti,am654-icss-iep" as a fallback as asked by Conor.
*) Addressed Andew's comments and removed helper APIs icss_iep_readl() /
   writel(). Now the settime/gettime APIs directly use readl() / writel().
*) Moved selecting TI_ICSS_IEP in Kconfig from patch 3 to patch 4.
*) Removed forward declaration of icss_iep_of_match in patch 3.
*) Replaced use of of_device_get_match_data() to device_get_match_data() in
   patch 3.
*) Removed of_match_ptr() from patch 3 as it is not needed.

Changes from v2 to v3:
*) Addressed Roger's comment and moved IEP1 related changes in patch 5.
*) Addressed Roger's comment and moved icss_iep.c / .h changes from patch 4
   to patch 3.
*) Added support for multiple timestamping in patch 4 as asked by Roger.
*) Addressed Andrew's comment and added comment in case SPEED_10 in
   icssg_config_ipg() API.
*) Kept compatible as "ti,am654-icss-iep" for all TI K3 SoCs

Changes from v1 to v2:
*) Addressed Simon's comment to fix reverse xmas tree declaration. Some APIs
   in patch 3 and 4 were not following reverse xmas tree variable declaration.
   Fixed it in this version.
*) Addressed Conor's comments and removed unsupported SoCs from compatible
   comment in patch 1.
*) Addded patch 2 which was not part of v1. Patch 2, adds IEP node to dt
   bindings for ICSSG.

[v1] https://lore.kernel.org/all/20230803110153.3309577-1-danishanwar@ti.com/
[v2] https://lore.kernel.org/all/20230807110048.2611456-1-danishanwar@ti.com/
[v3] https://lore.kernel.org/all/20230809114906.21866-1-danishanwar@ti.com/
[v4] https://lore.kernel.org/all/20230814100847.3531480-1-danishanwar@ti.com/
[v5] https://lore.kernel.org/all/20230817114527.1585631-1-danishanwar@ti.com/
[v6] https://lore.kernel.org/all/20230823113254.292603-1-danishanwar@ti.com/
====================

Signed-off-by: David S. Miller <davem@davemloft.net>

net: ti: icssg-prueth: am65x SR2.0 add 10M full duplex support

For AM65x SR2.0 it's required to enable IEP1 in raw 64bit mode which is
used by PRU FW to monitor the link and apply w/a for 10M link issue.
Note. No public errata available yet.

Without this w/a the PRU FW will stuck if link state changes under TX
traffic pressure.

Hence, add support for 10M full duplex for AM65x SR2.0:
 - add new IEP API to enable IEP, but without PTP support
 - add pdata quirk_10m_link_issue to enable 10M link issue w/a.

Signed-off-by: Grygorii Strashko <grygorii.strashko@ti.com>
Signed-off-by: Vignesh Raghavendra <vigneshr@ti.com>
Reviewed-by: Roger Quadros <rogerq@kernel.org>
Reviewed-by: Simon Horman <horms@kernel.org>
Reviewed-by: Jacob Keller <jacob.e.keller@intel.com>
Signed-off-by: MD Danish Anwar <danishanwar@ti.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

net: ti: icssg-prueth: add packet timestamping and ptp support

Add packet timestamping TS and PTP PHC clock support.

For AM65x and AM64x:
 - IEP1 is not used
 - IEP0 is configured in shadow mode with 1ms cycle and shared between
Linux and FW. It provides time and TS in number cycles, so special
conversation in ns is required.
 - IEP0 shared between PRUeth ports.
 - IEP0 supports PPS, periodic output.
 - IEP0 settime() and enabling PPS required FW interraction.
 - RX TS provided with each packet in CPPI5 descriptor.
 - TX TS returned through separate ICSSG hw queues for each port. TX TS
readiness is signaled by INTC IRQ. Only one packet at time can be requested
for TX TS.

Signed-off-by: Roger Quadros <rogerq@ti.com>
Co-developed-by: Grygorii Strashko <grygorii.strashko@ti.com>
Signed-off-by: Grygorii Strashko <grygorii.strashko@ti.com>
Signed-off-by: Vignesh Raghavendra <vigneshr@ti.com>
Reviewed-by: Simon Horman <horms@kernel.org>
Signed-off-by: MD Danish Anwar <danishanwar@ti.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

net: ti: icss-iep: Add IEP driver

Add a driver for Industrial Ethernet Peripheral (IEP) block of PRUSS to
support timestamping of ethernet packets and thus support PTP and PPS
for PRU ethernet ports.

Signed-off-by: Roger Quadros <rogerq@ti.com>
Signed-off-by: Lokesh Vutla <lokeshvutla@ti.com>
Signed-off-by: Murali Karicheri <m-karicheri2@ti.com>
Signed-off-by: Vignesh Raghavendra <vigneshr@ti.com>
Reviewed-by: Simon Horman <horms@kernel.org>
Signed-off-by: MD Danish Anwar <danishanwar@ti.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

dt-bindings: net: Add IEP property in ICSSG

Add IEP property in ICSSG hardware DT binding document.
ICSSG uses IEP (Industrial Ethernet Peripheral) to support timestamping
of ethernet packets, PTP and PPS.

Reviewed-by: Conor Dooley <conor.dooley@microchip.com>
Reviewed-by: Roger Quadros <rogerq@kernel.org>
Reviewed-by: Simon Horman <horms@kernel.org>
Signed-off-by: MD Danish Anwar <danishanwar@ti.com>
Reviewed-by: Rob Herring <robh@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

dt-bindings: net: Add ICSS IEP

Add a DT binding document for the ICSS Industrial Ethernet Peripheral(IEP)
hardware. IEP supports packet timestamping, PTP and PPS.

Reviewed-by: Conor Dooley <conor.dooley@microchip.com>
Reviewed-by: Roger Quadros <rogerq@kernel.org>
Reviewed-by: Simon Horman <horms@kernel.org>
Signed-off-by: MD Danish Anwar <danishanwar@ti.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Merge branch 'sfc-pedit-offloads'

Pieter Jansen van Vuuren says:

====================
sfc: introduce eth, ipv4 and ipv6 pedit offloads

This set introduces mac source and destination pedit set action offloads.
It also adds offload for ipv4 ttl and ipv6 hop limit pedit set action as
well pedit add actions that would result in the same semantics as
decrementing the ttl and hop limit.

v2:
- fix 'efx_tc_mangle' kdoc which was orphaned when adding 'efx_tc_pedit_add'.
- add description of 'match' in 'efx_tc_mangle' kdoc.
- correct some inconsistent kdoc indentation.

v1: https://lore.kernel.org/netdev/20230823111725.28090-1-pieter.jansen-van-vuuren@amd.com/
====================

Reviewed-by: Simon Horman <horms@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

sfc: extend pedit add action to handle decrement ipv6 hop limit

Extend the pedit add actions to handle this case for ipv6. Similar to ipv4
dec ttl, decrementing ipv6 hop limit can be achieved by adding 0xff to the
hop limit field.

Co-developed-by: Edward Cree <ecree.xilinx@gmail.com>
Signed-off-by: Edward Cree <ecree.xilinx@gmail.com>
Signed-off-by: Pieter Jansen van Vuuren <pieter.jansen-van-vuuren@amd.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

sfc: introduce pedit add actions on the ipv4 ttl field

Introduce pedit add actions and use it to achieve decrement ttl offload.
Decrement ttl can be achieved by adding 0xff to the ttl field.

Co-developed-by: Edward Cree <ecree.xilinx@gmail.com>
Signed-off-by: Edward Cree <ecree.xilinx@gmail.com>
Signed-off-by: Pieter Jansen van Vuuren <pieter.jansen-van-vuuren@amd.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

sfc: add decrement ipv6 hop limit by offloading set hop limit actions

Offload pedit set ipv6 hop limit, where the hop limit has already been
matched and the new value is one less, by translating it to a decrement.

Co-developed-by: Edward Cree <ecree.xilinx@gmail.com>
Signed-off-by: Edward Cree <ecree.xilinx@gmail.com>
Signed-off-by: Pieter Jansen van Vuuren <pieter.jansen-van-vuuren@amd.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

sfc: add decrement ttl by offloading set ipv4 ttl actions

Offload pedit set ipv4 ttl field, where the ttl field has already been
matched and the new value is one less, by translating it to a decrement.

Co-developed-by: Edward Cree <ecree.xilinx@gmail.com>
Signed-off-by: Edward Cree <ecree.xilinx@gmail.com>
Signed-off-by: Pieter Jansen van Vuuren <pieter.jansen-van-vuuren@amd.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

sfc: add mac source and destination pedit action offload

Introduce the first pedit set offload functionality for the sfc driver.
In addition to this, add offload functionality for both mac source and
destination pedit set actions.

Co-developed-by: Edward Cree <ecree.xilinx@gmail.com>
Signed-off-by: Edward Cree <ecree.xilinx@gmail.com>
Signed-off-by: Pieter Jansen van Vuuren <pieter.jansen-van-vuuren@amd.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

sfc: introduce ethernet pedit set action infrastructure

Introduce the initial ethernet pedit set action infrastructure in
preparation for adding mac src and dst pedit action offloads.

Co-developed-by: Edward Cree <ecree.xilinx@gmail.com>
Signed-off-by: Edward Cree <ecree.xilinx@gmail.com>
Signed-off-by: Pieter Jansen van Vuuren <pieter.jansen-van-vuuren@amd.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Merge branch '1GbE' of git://git./linux/kernel/git/tnguy/next-queue

Tony Nguyen says:

====================
Intel Wired LAN Driver Updates 2023-08-24 (igc, e1000e)

This series contains updates to igc and e1000e drivers.

Vinicius adds support for utilizing multiple PTP registers on igc.

Sasha reduces interval time for PTM on igc and adds new device support
on e1000e.

* '1GbE' of git://git.kernel.org/pub/scm/linux/kernel/git/tnguy/next-queue:
  e1000e: Add support for the next LOM generation
  igc: Decrease PTM short interval from 10 us to 1 us
  igc: Add support for multiple in-flight TX timestamps
====================

Link: https://lore.kernel.org/r/20230824204418.1551093-1-anthony.l.nguyen@intel.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

doc/netlink: Add delete operation to ovs_vport spec

Add del operation to the spec to help with testing.

Signed-off-by: Donald Hunter <donald.hunter@gmail.com>
Link: https://lore.kernel.org/r/20230824142221.71339-1-donald.hunter@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

tools: ynl-gen: fix uAPI generation after tempfile changes

We use a tempfile for code generation, to avoid wiping the target
file out if the code generator crashes. File contents are copied
from tempfile to actual destination at the end of main().

uAPI generation is relatively simple so when generating the uAPI
header we return from main() early, and never reach the "copy code
over" stage. Since commit under Fixes uAPI headers are not updated
by ynl-gen.

Move the copy/commit of the code into CodeWriter, to make it
easier to call at any point in time. Hook it into the destructor
to make sure we don't miss calling it.

Fixes: f65f305ae008 ("tools: ynl-gen: use temporary file for rendering")
Link: https://lore.kernel.org/r/20230824212431.1683612-1-kuba@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

Merge branch 'stmmac-cleanups'

Russell King says:

====================
stmmac cleanups

One of the comments I had on Feiyang Chen's series was concerning the
initialisation of phylink... and so I've decided to do something about
it, cleaning it up a bit.

This series:

1) adds a new phylink function to limit the MAC capabilities according
   to a maximum speed. This allows us to greatly simplify stmmac's
   initialisation of phylink's mac capabilities.

2) everywhere that uses priv->plat->phylink_node first converts this
   to a fwnode before doing anything with it. This is silly. Let's
   instead store it as a fwnode to eliminate these conversions in
   multiple places.

3) clean up passing the fwnode to phylink - it might as well happen
   at the phylink_create() callsite, rather than being scattered
   throughout the entire function.

4) same for mdio_bus_data

5) use phylink_limit_mac_speed() to handle the priv->plat->max_speed
   restriction.

6) add a method to get the MAC-specific capabilities from the code
   dealing with the MACs, and arrange to call it at an appropriate
   time.

7) convert the gmac4 users to use the MAC specific method.

8) same for xgmac.

9) group all the simple phylink_config initialisations together.

10) convert half-duplex logic to being positive logic.

While looking into all of this, this raised eyebrows:

        if (priv->plat->tx_queues_to_use > 1)
                priv->phylink_config.mac_capabilities &=
                        ~(MAC_10HD | MAC_100HD | MAC_1000HD);

priv->plat->tx_queues_to_use is initialised by platforms to either 1,
4 or 8, and can be controlled from userspace via the --set-channels
ethtool op. The implementation of this op in this driver limits the
number of channels to priv->dma_cap.number_tx_queues, which is derived
from the DMA hwcap.

So, the obvious questions are:

1) what guarantees that the static initialisation of tx_queues_to_use
will always be less than or equal to number_tx_queues from the DMA hw
cap?

2) tx_queues_to_use starts off as 1, but number_tx_queues is larger,
we will leave the half-duplex capabilities in place, but userspace can
increase tx_queues_to_use above 1. Does that mean half-duplex is then
not supported?

3) Should we be basing the decision whether half-duplex is supported
off the DMA capabilities?

4) What about priv->dma_cap.half_duplex? Doesn't that get a say in
whether half-duplex is supported or not? Why isn't this used? Why is
it only reported via debugfs? If it's not being used by the driver,
what's the point of reporting it via debugfs?
====================

Link: https://lore.kernel.org/r/ZOddFH22PWmOmbT5@shell.armlinux.org.uk
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net: stmmac: convert half-duplex support to positive logic

Rather than detecting when half-duplex is not supported, and clearing
the MAC capabilities, reverse the if() condition and use it to set the
capabilities instead.

Signed-off-by: Russell King (Oracle) <rmk+kernel@armlinux.org.uk>
Link: https://lore.kernel.org/r/E1qZAXn-005pUb-SP@rmk-PC.armlinux.org.uk
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net: stmmac: move priv->phylink_config.mac_managed_pm

Move priv->phylink_config.mac_managed_pm to be along side the other
phylink initialisations.

Signed-off-by: Russell King (Oracle) <rmk+kernel@armlinux.org.uk>
Link: https://lore.kernel.org/r/E1qZAXi-005pUV-Nq@rmk-PC.armlinux.org.uk
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net: stmmac: move xgmac specific phylink caps to dwxgmac2 core

Move the xgmac specific phylink capabilities to the dwxgmac2 support
core.

Signed-off-by: Russell King (Oracle) <rmk+kernel@armlinux.org.uk>
Link: https://lore.kernel.org/r/E1qZAXd-005pUP-JL@rmk-PC.armlinux.org.uk
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net: stmmac: move gmac4 specific phylink capabilities to gmac4

Move the setup of gmac4 speicifc phylink capabilities into gmac4 code.

Signed-off-by: Russell King (Oracle) <rmk+kernel@armlinux.org.uk>
Link: https://lore.kernel.org/r/E1qZAXY-005pUJ-Ez@rmk-PC.armlinux.org.uk
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net: stmmac: provide stmmac_mac_phylink_get_caps()

Allow MACs to provide their own capabilities via the MAC operations
struct.

Signed-off-by: Russell King (Oracle) <rmk+kernel@armlinux.org.uk>
Link: https://lore.kernel.org/r/E1qZAXT-005pUD-Aj@rmk-PC.armlinux.org.uk
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net: stmmac: use phylink_limit_mac_speed()

Use phylink_limit_mac_speed() to limit the MAC capabilities rather
than coding this for each speed.

Signed-off-by: Russell King (Oracle) <rmk+kernel@armlinux.org.uk>
Link: https://lore.kernel.org/r/E1qZAXO-005pU7-61@rmk-PC.armlinux.org.uk
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net: stmmac: use "mdio_bus_data" local variable

We have a local variable for priv->plat->mdio_bus_data, which we use
later in the conditional if() block, but we evaluate the above within
the conditional expression. Use mdio_bus_data instead. Since these
will be the only two users of this local variable, move its assignment
just before the if().

Signed-off-by: Russell King (Oracle) <rmk+kernel@armlinux.org.uk>
Link: https://lore.kernel.org/r/E1qZAXJ-005pU1-1z@rmk-PC.armlinux.org.uk
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net: stmmac: clean up passing fwnode to phylink

Move the initialisation of the fwnode variable closer to its use
site, rather than scattered throughout stmmac_phy_setup().

Signed-off-by: Russell King (Oracle) <rmk+kernel@armlinux.org.uk>
Link: https://lore.kernel.org/r/E1qZAXD-005pTv-TN@rmk-PC.armlinux.org.uk
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net: stmmac: convert plat->phylink_node to fwnode

All users of plat->phylink_node first convert it to a fwnode. Rather
than repeatedly convert to a fwnode, store it as a fwnode. To reflect
this change, call it plat->port_node instead - it is used for more
than just phylink.

Signed-off-by: Russell King (Oracle) <rmk+kernel@armlinux.org.uk>
Link: https://lore.kernel.org/r/E1qZAX8-005pTo-OT@rmk-PC.armlinux.org.uk
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net: phylink: add phylink_limit_mac_speed()

Add a function which can be used to limit the phylink MAC capabilities
to an upper speed limit.

Signed-off-by: Russell King (Oracle) <rmk+kernel@armlinux.org.uk>
Link: https://lore.kernel.org/r/E1qZAX3-005pTi-K1@rmk-PC.armlinux.org.uk
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

veth: Avoid NAPI scheduling on failed SKB forwarding

When an skb fails to be forwarded to the peer(e.g., skb data buffer
length exceeds MTU), it will not be added to the peer's receive queue.
Therefore, we should schedule the peer's NAPI poll function only when
skb forwarding is successful to avoid unnecessary overhead.

Signed-off-by: Liang Chen <liangchen.linux@gmail.com>
Link: https://lore.kernel.org/r/20230824123131.7673-1-liangchen.linux@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

Merge tag 'for-netdev' of https://git./linux/kernel/git/bpf/bpf-next

Daniel Borkmann says:

====================
pull-request: bpf-next 2023-08-25

We've added 87 non-merge commits during the last 8 day(s) which contain
a total of 104 files changed, 3719 insertions(+), 4212 deletions(-).

The main changes are:

1) Add multi uprobe BPF links for attaching multiple uprobes
   and usdt probes, which is significantly faster and saves extra fds,
   from Jiri Olsa.

2) Add support BPF cpu v4 instructions for arm64 JIT compiler,
   from Xu Kuohai.

3) Add support BPF cpu v4 instructions for riscv64 JIT compiler,
   from Pu Lehui.

4) Fix LWT BPF xmit hooks wrt their return values where propagating
   the result from skb_do_redirect() would trigger a use-after-free,
   from Yan Zhai.

5) Fix a BPF verifier issue related to bpf_kptr_xchg() with local kptr
   where the map's value kptr type and locally allocated obj type
   mismatch, from Yonghong Song.

6) Fix BPF verifier's check_func_arg_reg_off() function wrt graph
   root/node which bypassed reg->off == 0 enforcement,
   from Kumar Kartikeya Dwivedi.

7) Lift BPF verifier restriction in networking BPF programs to treat
   comparison of packet pointers not as a pointer leak,
   from Yafang Shao.

8) Remove unmaintained XDP BPF samples as they are maintained
   in xdp-tools repository out of tree, from Toke Høiland-Jørgensen.

9) Batch of fixes for the tracing programs from BPF samples in order
   to make them more libbpf-aware, from Daniel T. Lee.

10) Fix a libbpf signedness determination bug in the CO-RE relocation
    handling logic, from Andrii Nakryiko.

11) Extend libbpf to support CO-RE kfunc relocations. Also follow-up
    fixes for bpf_refcount shared ownership implementation,
    both from Dave Marchevsky.

12) Add a new bpf_object__unpin() API function to libbpf,
    from Daniel Xu.

13) Fix a memory leak in libbpf to also free btf_vmlinux
    when the bpf_object gets closed, from Hao Luo.

14) Small error output improvements to test_bpf module, from Helge Deller.

* tag 'for-netdev' of https://git.kernel.org/pub/scm/linux/kernel/git/bpf/bpf-next: (87 commits)
  selftests/bpf: Add tests for rbtree API interaction in sleepable progs
  bpf: Allow bpf_spin_{lock,unlock} in sleepable progs
  bpf: Consider non-owning refs to refcounted nodes RCU protected
  bpf: Reenable bpf_refcount_acquire
  bpf: Use bpf_mem_free_rcu when bpf_obj_dropping refcounted nodes
  bpf: Consider non-owning refs trusted
  bpf: Ensure kptr_struct_meta is non-NULL for collection insert and refcount_acquire
  selftests/bpf: Enable cpu v4 tests for RV64
  riscv, bpf: Support unconditional bswap insn
  riscv, bpf: Support signed div/mod insns
  riscv, bpf: Support 32-bit offset jmp insn
  riscv, bpf: Support sign-extension mov insns
  riscv, bpf: Support sign-extension load insns
  riscv, bpf: Fix missing exception handling and redundant zext for LDX_B/H/W
  samples/bpf: Add note to README about the XDP utilities moved to xdp-tools
  samples/bpf: Cleanup .gitignore
  samples/bpf: Remove the xdp_sample_pkts utility
  samples/bpf: Remove the xdp1 and xdp2 utilities
  samples/bpf: Remove the xdp_rxq_info utility
  samples/bpf: Remove the xdp_redirect* utilities
  ...
====================

Link: https://lore.kernel.org/r/20230825194319.12727-1-daniel@iogearbox.net
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

Merge tag 'wireless-next-2023-08-25' of git://git./linux/kernel/git/wireless/wireless-next

Kalle Valo says:

====================
wireless-next patches for v6.6

The second pull request for v6.6, this time with both stack and driver
changes. Unusually we have only one major new feature but lots of
small cleanup all over, I guess this is due to people have been on
vacation the last month.

Major changes:

rtw89
 - Introduce Time Averaged SAR (TAS) support

* tag 'wireless-next-2023-08-25' of git://git.kernel.org/pub/scm/linux/kernel/git/wireless/wireless-next: (114 commits)
  wifi: rtlwifi: rtl8723: Remove unused function rtl8723_cmd_send_packet()
  wifi: rtw88: usb: kill and free rx urbs on probe failure
  wifi: rtw89: Fix clang -Wimplicit-fallthrough in rtw89_query_sar()
  wifi: rtw89: phy: modify register setting of ENV_MNTR, PHYSTS and DIG
  wifi: rtw89: phy: add phy_gen_def::cr_base to support WiFi 7 chips
  wifi: rtw89: mac: define register address of rx_filter to generalize code
  wifi: rtw89: mac: define internal memory address for WiFi 7 chip
  wifi: rtw89: mac: generalize code to indirectly access WiFi internal memory
  wifi: rtw89: mac: add mac_gen_def::band1_offset to map MAC band1 register address
  wifi: wlcore: sdio: Use module_sdio_driver macro to simplify the code
  wifi: rtw89: initialize multi-channel handling
  wifi: rtw89: provide functions to configure NoA for beacon update
  wifi: rtw89: call rtw89_chan_get() by vif chanctx if aware of vif
  wifi: rtw89: sar: let caller decide the center frequency to query
  wifi: rtw89: refine rtw89_correct_cck_chan() by rtw89_hw_to_nl80211_band()
  wifi: rtw89: add function prototype for coex request duration
  Fix nomenclature for USB and PCI wireless devices
  wifi: ath: Use is_multicast_ether_addr() to check multicast Ether address
  wifi: ath12k: Remove unused declarations
  wifi: ath12k: add check max message length while scanning with extraie
  ...
====================

Link: https://lore.kernel.org/r/20230825132230.A0833C433C8@smtp.kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

Merge tag 'for-net-next-2023-08-24' of git://git./linux/kernel/git/bluetooth/bluetooth-next

Luiz Augusto von Dentz says:

====================
bluetooth-next pull request for net-next:

 - Introduce HCI_QUIRK_BROKEN_LE_CODED
 - Add support for PA/BIG sync
 - Add support for NXP IW624 chipset
 - Add support for Qualcomm WCN7850

* tag 'for-net-next-2023-08-24' of git://git.kernel.org/pub/scm/linux/kernel/git/bluetooth/bluetooth-next:
  Bluetooth: btusb: Do not call kfree_skb() under spin_lock_irqsave()
  Bluetooth: btusb: Fix quirks table naming
  Bluetooth: HCI: Introduce HCI_QUIRK_BROKEN_LE_CODED
  Bluetooth: btintel: Send new command for PPAG
  Bluetooth: ISO: Add support for periodic adv reports processing
  Bluetooth: hci_conn: fail SCO/ISO via hci_conn_failed if ACL gone early
  Bluetooth: hci_core: Fix missing instances using HCI_MAX_AD_LENGTH
  Bluetooth: ISO: Use defer setup to separate PA sync and BIG sync
  Bluetooth: qca: add support for WCN7850
  Bluetooth: qca: use switch case for soc type behavior
  dt-bindings: net: bluetooth: qualcomm: document WCN7850 chipset
  Bluetooth: hci_conn: Fix sending BT_HCI_CMD_LE_CREATE_CONN_CANCEL
  Bluetooth: hci_sync: Fix UAF in hci_disconnect_all_sync
  Bluetooth: btnxpuart: Improve inband Independent Reset handling
  Bluetooth: btnxpuart: Add support for IW624 chipset
  Bluetooth: btnxpuart: Remove check for CTS low after FW download
====================

Link: https://lore.kernel.org/r/20230824201458.2577-1-luiz.dentz@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

Merge branch 'bpf-refcount-followups-3-bpf_mem_free_rcu-refcounted-nodes'

Dave Marchevsky says:

====================
BPF Refcount followups 3: bpf_mem_free_rcu refcounted nodes

This series is the third of three (or more) followups to address issues
in the bpf_refcount shared ownership implementation discovered by Kumar.
This series addresses the use-after-free scenario described in [0]. The
first followup series ([1]) also attempted to address the same
use-after-free, but only got rid of the splat without addressing the
underlying issue. After this series the underyling issue is fixed and
bpf_refcount_acquire can be re-enabled.

The main fix here is migration of bpf_obj_drop to use
bpf_mem_free_rcu. To understand why this fixes the issue, let us consider
the example interleaving provided by Kumar in [0]:

CPU 0                                   CPU 1
n = bpf_obj_new
lock(lock1)
bpf_rbtree_add(rbtree1, n)
m = bpf_rbtree_acquire(n)
unlock(lock1)

kptr_xchg(map, m) // move to map
// at this point, refcount = 2
m = kptr_xchg(map, NULL)
lock(lock2)
lock(lock1) bpf_rbtree_add(rbtree2, m)
p = bpf_rbtree_first(rbtree1) if (!RB_EMPTY_NODE) bpf_obj_drop_impl(m) // A
bpf_rbtree_remove(rbtree1, p)
unlock(lock1)
bpf_obj_drop(p) // B
bpf_refcount_acquire(m) // use-after-free
...

Before this series, bpf_obj_drop returns memory to the allocator using
bpf_mem_free. At this point (B in the example) there might be some
non-owning references to that memory which the verifier believes are valid,
but where the underlying memory was reused for some other allocation.
Commit 7793fc3babe9 ("bpf: Make bpf_refcount_acquire fallible for
non-owning refs") attempted to fix this by doing refcount_inc_non_zero
on refcount_acquire in instead of refcount_inc under the assumption that
preventing erroneous incr-on-0 would be sufficient. This isn't true,
though: refcount_inc_non_zero must *check* if the refcount is zero, and
the memory it's checking could have been reused, so the check may look
at and incr random reused bytes.

If we wait to reuse this memory until all non-owning refs that could
point to it are gone, there is no possibility of this scenario
happening. Migrating bpf_obj_drop to use bpf_mem_free_rcu for refcounted
nodes accomplishes this.

For such nodes, the validity of their underlying memory is now tied to
RCU critical section. This matches MEM_RCU trustedness
expectations, so the series takes the opportunity to more explicitly
mark this trustedness state.

The functional effects of trustedness changes here are rather small.
This is largely due to local kptrs having separate verifier handling -
with implicit trustedness assumptions - than arbitrary kptrs.
Regardless, let's take the opportunity to move towards a world where
trustedness is more explicitly handled.

Changelog:

v1 -> v2: https://lore.kernel.org/bpf/20230801203630.3581291-1-davemarchevsky@fb.com/

Patch 1 ("bpf: Ensure kptr_struct_meta is non-NULL for collection insert and refcount_acquire")
  * Spent some time experimenting with a better approach as per convo w/
    Yonghong on v1's patch. It started getting too complex, so left unchanged
    for now. Yonghong was fine with this approach being shipped.

Patch 2 ("bpf: Consider non-owning refs trusted")
  * Add Yonghong ack
Patch 3 ("bpf: Use bpf_mem_free_rcu when bpf_obj_dropping refcounted nodes")
  * Add Yonghong ack
Patch 4 ("bpf: Reenable bpf_refcount_acquire")
  * Add Yonghong ack

Patch 5 ("bpf: Consider non-owning refs to refcounted nodes RCU protected")
  * Undo a nonfunctional whitespace change that shouldn't have been included
    (Yonghong)
  * Better logging message when complaining about rcu_read_{lock,unlock} in
    rbtree cb (Alexei)
  * Don't invalidate_non_owning_refs when processing bpf_rcu_read_unlock
    (Yonghong, Alexei)

Patch 6 ("[RFC] bpf: Allow bpf_spin_{lock,unlock} in sleepable prog's RCU CS")
  * preempt_{disable,enable} in __bpf_spin_{lock,unlock} (Alexei)
    * Due to this we can consider spin_lock CS an RCU-sched read-side CS (per
      RCU/Design/Requirements/Requirements.rst). Modify in_rcu_cs accordingly.
  * no need to check for !in_rcu_cs before allowing bpf_spin_{lock,unlock}
    (Alexei)
  * RFC tag removed and renamed to "bpf: Allow bpf_spin_{lock,unlock} in
    sleepable progs"

Patch 7 ("selftests/bpf: Add tests for rbtree API interaction in sleepable progs")
  * Remove "no explicit bpf_rcu_read_lock" failure test, add similar success
    test (Alexei)

Summary of patch contents, with sub-bullets being leading questions and
comments I think are worth reviewer attention:

  * Patches 1 and 2 are moreso documententation - and
    enforcement, in patch 1's case - of existing semantics / expectations

  * Patch 3 changes bpf_obj_drop behavior for refcounted nodes such that
    their underlying memory is not reused until RCU grace period elapses
    * Perhaps it makes sense to move to mem_free_rcu for _all_
      non-owning refs in the future, not just refcounted. This might
      allow custom non-owning ref lifetime + invalidation logic to be
      entirely subsumed by MEM_RCU handling. IMO this needs a bit more
      thought and should be tackled outside of a fix series, so it's not
      attempted here.

  * Patch 4 re-enables bpf_refcount_acquire as changes in patch 3 fix
    the remaining use-after-free
    * One might expect this patch to be last in the series, or last
      before selftest changes. Patches 5 and 6 don't change
      verification or runtime behavior for existing BPF progs, though.

  * Patch 5 brings the verifier's understanding of refcounted node
    trustedness in line with Patch 4's changes

  * Patch 6 allows some bpf_spin_{lock, unlock} calls in sleepable
    progs. Marked RFC for a few reasons:
    * bpf_spin_{lock,unlock} haven't been usable in sleepable progs
      since before the introduction of bpf linked list and rbtree. As
      such this feels more like a new feature that may not belong in
      this fixes series.

  * Patch 7 adds tests

  [0]: https://lore.kernel.org/bpf/atfviesiidev4hu53hzravmtlau3wdodm2vqs7rd7tnwft34e3@xktodqeqevir/
  [1]: https://lore.kernel.org/bpf/20230602022647.1571784-1-davemarchevsky@fb.com/
====================

Link: https://lore.kernel.org/r/20230821193311.3290257-1-davemarchevsky@fb.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>