Add support for SFP hotpluggable modules via sfp-bus and phylink.
This supports both copper and optical SFP modules, which require
different Serdes modes in order to properly negotiate the link.
Optical SFP modules typically require the Serdes link to be talking
1000BaseX mode - this is the gigabit ethernet mode defined by the
802.3 standard.
Copper SFP modules typically integrate a PHY in the module to convert
from Serdes to copper, and the PHY will be configured by the vendor
to either present a 1000BaseX Serdes link (for fixed 1000BaseT) or a
SGMII Serdes link. However, this is vendor defined, so we instead
detect the PHY, switch the link to SGMII mode, and use traditional
PHY based negotiation.
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
Signed-off-by: David S. Miller <davem@davemloft.net>
Add in-band autonegotation support for 10GBase-KR mode.
Signed-off-by: Russell King <rmk+kernel@armlinux.org.uk>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
Add support for reading and writing the clause 45 MII registers.
Signed-off-by: Russell King <rmk+kernel@armlinux.org.uk>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
Add support for reading module EEPROMs through phylink.
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Russell King <rmk+kernel@armlinux.org.uk>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
The link between the ethernet MAC and its PHY has become more complex
as the interface evolves. This is especially true with serdes links,
where the part of the PHY is effectively integrated into the MAC.
Serdes links can be connected to a variety of devices, including SFF
modules soldered down onto the board with the MAC, a SFP cage with
a hotpluggable SFP module which may contain a PHY or directly modulate
the serdes signals onto optical media with or without a PHY, or even
a classical PHY connection.
Moreover, the negotiation information on serdes links comes in two
varieties - SGMII mode, where the PHY provides its speed/duplex/flow
control information to the MAC, and 1000base-X mode where both ends
exchange their abilities and each resolve the link capabilities.
This means we need a more flexible means to support these arrangements,
particularly with the hotpluggable nature of SFP, where the PHY can
be attached or detached after the network device has been brought up.
Ethtool information can come from multiple sources:
- we may have a PHY operating in either SGMII or 1000base-X mode, in
which case we take ethtool/mii data directly from the PHY.
- we may have a optical SFP module without a PHY, with the MAC
operating in 1000base-X mode - the ethtool/mii data needs to come
from the MAC.
- we may have a copper SFP module with a PHY whic can't be accessed,
which means we need to take ethtool/mii data from the MAC.
Phylink aims to solve this by providing an intermediary between the
MAC and PHY, providing a safe way for PHYs to be hotplugged, and
allowing a SFP driver to reconfigure the serdes connection.
Phylink also takes over support of fixed link connections, where the
speed/duplex/flow control are fixed, but link status may be controlled
by a GPIO signal. By avoiding the fixed-phy implementation, phylink
can provide a faster response to link events: fixed-phy has to wait for
phylib to operate its state machine, which can take several seconds.
In comparison, phylink takes milliseconds.
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
- remove sync status
- rework supported and advertisment handling
- add 1000base-x speed for fixed links
- use functionality exported from phy-core, reworking
__phylink_ethtool_ksettings_set for it
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
Add an I2C MDIO bus bridge library, to allow phylib to access PHYs which
are connected to an I2C bus instead of the more conventional MDIO bus.
Such PHYs can be found in SFP adapters and SFF modules.
Since PHYs appear at I2C bus address 0x40..0x5f, and 0x50/0x51 are
reserved for SFP EEPROMs/diagnostics, we must not allow the MDIO bus
to access these I2C addresses.
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
phylink will need phy_start_machine exported, so lets export it as a
GPL symbol. Documentation/networking/phy.txt indicates that this
should be a PHY API function.
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
Signed-off-by: David S. Miller <davem@davemloft.net>
Sometimes, we need to do additional work between the PHY coming up and
marking the carrier present - for example, we may need to wait for the
PHY to MAC link to finish negotiation. This changes phylib to provide
a notification function pointer which avoids the built-in
netif_carrier_on() and netif_carrier_off() functions.
Standard ->adjust_link functionality is provided by hooking a helper
into the new ->phy_link_change method.
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
Signed-off-by: David S. Miller <davem@davemloft.net>
Add the missing 1000Base-X entry to the phy settings table. This was
not included because the original code could not cope with more than
32 bits of link mode mask.
Signed-off-by: Russell King <rmk+kernel@armlinux.org.uk>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
phy_lookup_setting() provides useful functionality in ethtool code
outside phylib. Move it to phy-core and allow it to be re-used (eg,
in phylink) rather than duplicated elsewhere. Note that this supports
the larger linkmode space.
As we move the phy settings table, we also need to move the guts of
phy_supported_speeds() as well.
Signed-off-by: Russell King <rmk+kernel@armlinux.org.uk>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
Other code would like to make use of this, so make the speed and duplex
string generation visible, and place it in a separate file.
Signed-off-by: Russell King <rmk+kernel@armlinux.org.uk>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Allow the phy settings table to support more than 32 link modes by
switching to the ethtool link mode bit number representation, rather
than storing the mask. This will allow phylink and other ethtool
code to share the settings table to look up settings.
Signed-off-by: Russell King <rmk+kernel@armlinux.org.uk>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Paolo Abeni says:
====================
IP: cleanup LSRR option processing
The __ip_options_echo() function expect a valid dst entry in skb->dst;
as result we sometimes need to preserve the dst entry for the whole IP
RX path.
The current usage of skb->dst looks more a relic from ancient past that
a real functional constraint. This patchset tries to remove such usage,
and than drops some hacks currently in place in the IP code to keep
skb->dst around.
__ip_options_echo() uses of skb->dst for two different purposes: retrieving
the netns assicated with the skb, and modify the ingress packet LSRR address
list.
The first patch removes the code modifying the ingress packet, and the second
one provides an explicit netns argument to __ip_options_echo(). The following
patches cleanup the current code keeping arund skb->dst for __ip_options_echo's
sake.
Updating the __ip_options_echo() function has been previously discussed here:
http://marc.info/?l=linux-netdev&m=150064533516348&w=2
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
__ip_options_echo() does not need anymore skb->dst, so we can
avoid explicitly preserving it for its own sake.
This is almost a revert of commit 0ddf3fb2c4 ("udp: preserve
skb->dst if required for IP options processing") plus some
lifting to fit later changes.
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
ip_options_echo() does not use anymore the skb->dst and don't
need to keep the dst around for options's sake only.
This reverts commit 34b2cef20f.
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
__ip_options_echo() uses the current network namespace, and
currently retrives it via skb->dst->dev.
This commit adds an explicit 'net' argument to __ip_options_echo()
and update all the call sites to provide it, usually via a simpler
sock_net().
After this change, __ip_options_echo() no more needs to access
skb->dst and we can drop a couple of hack to preserve such
info in the rx path.
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
While computing the response option set for LSRR, ip_options_echo()
also changes the ingress packet LSRR addresses list, setting
the last one to the dst specific address for the ingress packet
- via memset(start[ ...
The only visible effect of such change - beyond possibly damaging
shared/cloned skbs - is modifying the data carried by ICMP replies
changing the header information for reported the ingress packet,
which violates RFC1122 3.2.2.6.
All the others call sites just ignore the ingress packet IP options
after calling ip_options_echo()
Note that the last element in the LSRR option address list for the
reply packet will be properly set later in the ip output path
via ip_options_build().
This buggy memset() predates git history and apparently was present
into the initial ip_options_echo() implementation in linux 1.3.30 but
still looks wrong.
The removal of the fib_compute_spec_dst() call will help
completely dropping the skb->dst usage by __ip_options_echo() with a
later patch.
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Add support for GRO (generic receive offload) for aQuantia Atlantic driver.
This results in a perfomance improvement when GRO is enabled.
Signed-off-by: Pavel Belous <pavel.belous@aquantia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Jiri Pirko says:
====================
net: sched: summer cleanup part 1, mainly in exts area
This patchset is one of the couple cleanup patchsets I have in queue.
The motivation aside the obvious need to "make things nicer" is also
to prepare for shared filter blocks introduction. That requires tp->q
removal, and therefore removal of all tp->q users.
Patch 1 is just some small thing I spotted on the way
Patch 2 removes one user of tp->q, namely tcf_em_tree_change
Patches 3-8 do preparations for exts->nr_actions removal
Patches 9-10 do simple renames of functions in cls*
Patches 11-19 remove unnecessary calls of tcf_exts_change helper
The last patch changes tcf_exts_change to don't take lock
Tested by tools/testing/selftests/tc-testing
v1->v2:
- removed conversion of action array to list as noted by Cong
- added the past patch instead
- small rebases of patches 11-19
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
tcf_exts_change is always called on newly created exts, which are not used
on fastpath. Therefore, simple struct copy is enough.
Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
As the n struct was allocated right before u32_set_parms call,
no need to use tcf_exts_change to do atomic change, and we can just
fill-up the unused exts struct directly by tcf_exts_validate.
Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
As the f struct was allocated right before route4_set_parms call,
no need to use tcf_exts_change to do atomic change, and we can just
fill-up the unused exts struct directly by tcf_exts_validate.
Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
As the fnew struct just was allocated, so no need to use tcf_exts_change
to do atomic change, and we can just fill-up the unused exts struct
directly by tcf_exts_validate.
Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
As the new struct just was allocated, so no need to use tcf_exts_change
to do atomic change, and we can just fill-up the unused exts struct
directly by tcf_exts_validate.
Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
As the prog struct was allocated right before cls_bpf_set_parms call,
no need to use tcf_exts_change to do atomic change, and we can just
fill-up the unused exts struct directly by tcf_exts_validate.
Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
As the f struct was allocated right before basic_set_parms call, no need
to use tcf_exts_change to do atomic change, and we can just fill-up
the unused exts struct directly by tcf_exts_validate.
Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
As the head struct was allocated right before mall_set_parms call,
no need to use tcf_exts_change to do atomic change, and we can just
fill-up the unused exts struct directly by tcf_exts_validate.
Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
As the f struct was allocated right before fw_set_parms call, no need
to use tcf_exts_change to do atomic change, and we can just fill-up
the unused exts struct directly by tcf_exts_validate.
Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
As the f struct was allocated right before fl_set_parms call, no need
to use tcf_exts_change to do atomic change, and we can just fill-up
the unused exts struct directly by tcf_exts_validate.
Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Since the function name is misleading since it is not changing
anything, name it similarly to other cls.
Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The name cls_bpf_modify_existing is highly misleading, as it indeed does
not modify anything existing. It does not modify at all.
Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Acked-by: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
For check in tcf_exts_dump use tcf_exts_has_actions helper instead
of exts->nr_actions for checking if there are any actions present.
Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Leave it to tcf_action_exec to return TC_ACT_OK in case there is no
action present.
Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
These two helpers are doing the same as tcf_exts_has_actions, so remove
them and use tcf_exts_has_actions instead.
Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Use the tcf_exts_has_actions helper instead or directly testing
exts->nr_actions in tcf_exts_exec.
Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The rest of the helpers are named tcf_exts_*, so change the name of
the action number helpers to be aligned. While at it, change to inline
functions.
Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Acked-by: Jamal Hadi Salim <jhs@mojatatu.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Since tcf_em_tree_validate could be always called on a newly created
filter, there is no need for this change function.
Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Even if it is only for classid now, use this common struct a be aligned
with the rest of the classful qdiscs.
Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Acked-by: Jamal Hadi Salim <jhs@mojatatu.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch fixes the __udivdi3 undefined error reported by
test robot.
Fixes: b8c17f7088 ("net: hns: Add self-adaptive interrupt coalesce support in hns driver")
Signed-off-by: Yunsheng Lin <linyunsheng@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This was supposed to be a bitwise OR but there is a || vs | typo.
Fixes: 864dc729d5 ("net: phy: marvell: Refactor m88e1121 RGMII delay configuration")
Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
Willem de Bruijn says:
====================
socket sendmsg MSG_ZEROCOPY
Introduce zerocopy socket send flag MSG_ZEROCOPY. This extends the
shared page support (SKBTX_SHARED_FRAG) from sendpage to sendmsg.
Implement the feature for TCP initially, as large writes benefit
most.
On a send call with MSG_ZEROCOPY, the kernel pins user pages and
links these directly into the skbuff frags[] array.
Each send call with MSG_ZEROCOPY that transmits data will eventually
queue a completion notification on the error queue: a per-socket u32
incremented on each such call. A request may have to revert to copy
to succeed, for instance when a device cannot support scatter-gather
IO. In that case a flag is passed along to notify that the operation
succeeded without zerocopy optimization.
The implementation extends the existing zerocopy infra for tuntap,
vhost and xen with features needed for TCP, notably reference
counting to handle cloning on retransmit and GSO.
For more details, see also the netdev 2.1 paper and presentation at
https://netdevconf.org/2.1/session.html?debruijn
Changelog:
v3 -> v4:
- dropped UDP, RAW and PF_PACKET for now
Without loopback support, datagrams are usually smaller than
the ~8KB size threshold needed to benefit from zerocopy.
- style: a few reverse chrismas tree
- minor: SO_ZEROCOPY returns ENOTSUPP on unsupported protocols
- minor: squashed SO_EE_CODE_ZEROCOPY_COPIED patch
- minor: rebased on top of net-next with kmap_atomic fix
v2 -> v3:
- fix rebase conflict: SO_ZEROCOPY 59 -> 60
v1 -> v2:
- fix (kbuild-bot): do not remove uarg until patch 5
- fix (kbuild-bot): move zerocopy_sg_from_iter doc with function
- fix: remove unused extern in header file
RFCv2 -> v1:
- patch 2
- review comment: in skb_copy_ubufs, always allocate order-0
page, also when replacing compound source pages.
- patch 3
- fix: always queue completion notification on MSG_ZEROCOPY,
also if revert to copy.
- fix: on syscall abort, correctly revert notification state
- minor: skip queue notification on SOCK_DEAD
- minor: replace BUG_ON with WARN_ON in recoverable error
- patch 4
- new: add socket option SOCK_ZEROCOPY.
only honor MSG_ZEROCOPY if set, ignore for legacy apps.
- patch 5
- fix: clear zerocopy state on skb_linearize
- patch 6
- fix: only coalesce if prev errqueue elem is zerocopy
- minor: try coalescing with list tail instead of head
- minor: merge bytelen limit patch
- patch 7
- new: signal when data had to be copied
- patch 8 (tcp)
- optimize: avoid setting PSH bit when exceeding max frags.
that limits GRO on the client. do not goto new_segment.
- fix: fail on MSG_ZEROCOPY | MSG_FASTOPEN
- minor: do not wait for memory: does not work for optmem
- minor: simplify alloc
- patch 9 (udp)
- new: add PF_INET6
- fix: attach zerocopy notification even if revert to copy
- minor: simplify alloc size arithmetic
- patch 10 (raw hdrinc)
- new: add PF_INET6
- patch 11 (pf_packet)
- minor: simplify slightly
- patch 12
- new msg_zerocopy regression test: use veth pair to test
all protocols: ipv4/ipv6/packet, tcp/udp/raw, cork
all relevant ethtool settings: rx off, sg off
all relevant packet lengths: 0, <MAX_HEADER, max size
RFC -> RFCv2:
- review comment: do not loop skb with zerocopy frags onto rx:
add skb_orphan_frags_rx to orphan even refcounted frags
call this in __netif_receive_skb_core, deliver_skb and tun:
same as commit 1080e512d4 ("net: orphan frags on receive")
- fix: hold an explicit sk reference on each notification skb.
previously relied on the reference (or wmem) held by the
data skb that would trigger notification, but this breaks
on skb_orphan.
- fix: when aborting a send, do not inc the zerocopy counter
this caused gaps in the notification chain
- fix: in packet with SOCK_DGRAM, pull ll headers before calling
zerocopy_sg_from_iter
- fix: if sock_zerocopy_realloc does not allow coalescing,
do not fail, just allocate a new ubuf
- fix: in tcp, check return value of second allocation attempt
- chg: allocate notification skbs from optmem
to avoid affecting tcp write queue accounting (TSQ)
- chg: limit #locked pages (ulimit) per user instead of per process
- chg: grow notification ids from 16 to 32 bit
- pass range [lo, hi] through 32 bit fields ee_info and ee_data
- chg: rebased to davem-net-next on top of v4.10-rc7
- add: limit notification coalescing
sharing ubufs limits overhead, but delays notification until
the last packet is released, possibly unbounded. Add a cap.
- tests: add snd_zerocopy_lo pf_packet test
- tests: two bugfixes (add do_flush_tcp, ++sent not only in debug)
Limitations / Known Issues:
- TCP may build slightly smaller than max TSO packets due to
exceeding MAX_SKB_FRAGS frags when zerocopy pages are unaligned.
- All SKBTX_SHARED_FRAG may require additional __skb_linearize or
skb_copy_ubufs calls in u32, skb_find_text, similar to
skb_checksum_help.
Notification skbuffs are allocated from optmem. For sockets that
cannot effectively coalesce notifications, the optmem max may need
to be increased to avoid hitting -ENOBUFS:
sysctl -w net.core.optmem_max=1048576
In application load, copy avoidance shows a roughly 5% systemwide
reduction in cycles when streaming large flows and a 4-8% reduction in
wall clock time on early tensorflow test workloads.
For the single-machine veth tests to succeed, loopback support has to
be temporarily enabled by making skb_orphan_frags_rx map to
skb_orphan_frags.
* Performance
The below table shows cycles reported by perf for a netperf process
sending a single 10 Gbps TCP_STREAM. The first three columns show
Mcycles spent in the netperf process context. The second three columns
show time spent systemwide (-a -C A,B) on the two cpus that run the
process and interrupt handler. Reported is the median of at least 3
runs. std is a standard netperf, zc uses zerocopy and % is the ratio.
Netperf is pinned to cpu 2, network interrupts to cpu3, rps and rfs
are disabled and the kernel is booted with idle=halt.
NETPERF=./netperf -t TCP_STREAM -H $host -T 2 -l 30 -- -m $size
perf stat -e cycles $NETPERF
perf stat -C 2,3 -a -e cycles $NETPERF
--process cycles-- ----cpu cycles----
std zc % std zc %
4K 27,609 11,217 41 49,217 39,175 79
16K 21,370 3,823 18 43,540 29,213 67
64K 20,557 2,312 11 42,189 26,910 64
256K 21,110 2,134 10 43,006 27,104 63
1M 20,987 1,610 8 42,759 25,931 61
Perf record indicates the main source of these differences. Process
cycles only at 1M writes (perf record; perf report -n):
std:
Samples: 42K of event 'cycles', Event count (approx.): 21258597313
79.41% 33884 netperf [kernel.kallsyms] [k] copy_user_generic_string
3.27% 1396 netperf [kernel.kallsyms] [k] tcp_sendmsg
1.66% 694 netperf [kernel.kallsyms] [k] get_page_from_freelist
0.79% 325 netperf [kernel.kallsyms] [k] tcp_ack
0.43% 188 netperf [kernel.kallsyms] [k] __alloc_skb
zc:
Samples: 1K of event 'cycles', Event count (approx.): 1439509124
30.36% 584 netperf.zerocop [kernel.kallsyms] [k] gup_pte_range
14.63% 284 netperf.zerocop [kernel.kallsyms] [k] __zerocopy_sg_from_iter
8.03% 159 netperf.zerocop [kernel.kallsyms] [k] skb_zerocopy_add_frags_iter
4.84% 96 netperf.zerocop [kernel.kallsyms] [k] __alloc_skb
3.10% 60 netperf.zerocop [kernel.kallsyms] [k] kmem_cache_alloc_node
* Safety
The number of pages that can be pinned on behalf of a user with
MSG_ZEROCOPY is bound by the locked memory ulimit.
While the kernel holds process memory pinned, a process cannot safely
reuse those pages for other purposes. Packets looped onto the receive
stack and queued to a socket can be held indefinitely. Avoid unbounded
notification latency by restricting user pages to egress paths only.
skb_orphan_frags_rx() will create a private copy of pages even for
refcounted packets when these are looped, as did skb_orphan_frags for
the original tun zerocopy implementation.
Pages are not remapped read-only. Processes can modify packet contents
while packets are in flight in the kernel path. Bytes on which kernel
control flow depends (headers) are copied to avoid TOCTTOU attacks.
Datapath integrity does not otherwise depend on payload, with three
exceptions: checksums, optional sk_filter/tc u32/.. and device +
driver logic. The effect of wrong checksums is limited to the
misbehaving process. TC filters that access contents may have to be
excluded by adding an skb_orphan_frags_rx.
Processes can also safely avoid OOM conditions by bounding the number
of bytes passed with MSG_ZEROCOPY and by removing shared pages after
transmission from their own memory map.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
Introduce regression test for msg_zerocopy feature. Send traffic from
one process to another with and without zerocopy.
Evaluate tcp, udp, raw and packet sockets, including variants
- udp: corking and corking with mixed copy/zerocopy calls
- raw: with and without hdrincl
- packet: at both raw and dgram level
Test on both ipv4 and ipv6, optionally with ethtool changes to
disable scatter-gather, tx checksum or tso offload. All of these
can affect zerocopy behavior.
The regression test can be run on a single machine if over a veth
pair. Then skb_orphan_frags_rx must be modified to be identical to
skb_orphan_frags to allow forwarding zerocopy locally.
The msg_zerocopy.sh script will setup the veth pair in network
namespaces and run all tests.
Signed-off-by: Willem de Bruijn <willemb@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Enable support for MSG_ZEROCOPY to the TCP stack. TSO and GSO are
both supported. Only data sent to remote destinations is sent without
copying. Packets looped onto a local destination have their payload
copied to avoid unbounded latency.
Tested:
A 10x TCP_STREAM between two hosts showed a reduction in netserver
process cycles by up to 70%, depending on packet size. Systemwide,
savings are of course much less pronounced, at up to 20% best case.
msg_zerocopy.sh 4 tcp:
without zerocopy
tx=121792 (7600 MB) txc=0 zc=n
rx=60458 (7600 MB)
with zerocopy
tx=286257 (17863 MB) txc=286257 zc=y
rx=140022 (17863 MB)
This test opens a pair of sockets over veth, one one calls send with
64KB and optionally MSG_ZEROCOPY and on the other reads the initial
bytes. The receiver truncates, so this is strictly an upper bound on
what is achievable. It is more representative of sending data out of
a physical NIC (when payload is not touched, either).
Signed-off-by: Willem de Bruijn <willemb@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Bound the number of pages that a user may pin.
Follow the lead of perf tools to maintain a per-user bound on memory
locked pages commit 789f90fcf6 ("perf_counter: per user mlock gift")
Signed-off-by: Willem de Bruijn <willemb@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
In the simple case, each sendmsg() call generates data and eventually
a zerocopy ready notification N, where N indicates the Nth successful
invocation of sendmsg() with the MSG_ZEROCOPY flag on this socket.
TCP and corked sockets can cause send() calls to append new data to an
existing sk_buff and, thus, ubuf_info. In that case the notification
must hold a range. odify ubuf_info to store a inclusive range [N..N+m]
and add skb_zerocopy_realloc() to optionally extend an existing range.
Also coalesce notifications in this common case: if a notification
[1, 1] is about to be queued while [0, 0] is the queue tail, just modify
the head of the queue to read [0, 1].
Coalescing is limited to a few TSO frames worth of data to bound
notification latency.
Signed-off-by: Willem de Bruijn <willemb@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Prepare the datapath for refcounted ubuf_info. Clone ubuf_info with
skb_zerocopy_clone() wherever needed due to skb split, merge, resize
or clone.
Split skb_orphan_frags into two variants. The split, merge, .. paths
support reference counted zerocopy buffers, so do not do a deep copy.
Add skb_orphan_frags_rx for paths that may loop packets to receive
sockets. That is not allowed, as it may cause unbounded latency.
Deep copy all zerocopy copy buffers, ref-counted or not, in this path.
The exact locations to modify were chosen by exhaustively searching
through all code that might modify skb_frag references and/or the
the SKBTX_DEV_ZEROCOPY tx_flags bit.
The changes err on the safe side, in two ways.
(1) legacy ubuf_info paths virtio and tap are not modified. They keep
a 1:1 ubuf_info to sk_buff relationship. Calls to skb_orphan_frags
still call skb_copy_ubufs and thus copy frags in this case.
(2) not all copies deep in the stack are addressed yet. skb_shift,
skb_split and skb_try_coalesce can be refined to avoid copying.
These are not in the hot path and this patch is hairy enough as
is, so that is left for future refinement.
Signed-off-by: Willem de Bruijn <willemb@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The send call ignores unknown flags. Legacy applications may already
unwittingly pass MSG_ZEROCOPY. Continue to ignore this flag unless a
socket opts in to zerocopy.
Introduce socket option SO_ZEROCOPY to enable MSG_ZEROCOPY processing.
Processes can also query this socket option to detect kernel support
for the feature. Older kernels will return ENOPROTOOPT.
Signed-off-by: Willem de Bruijn <willemb@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
