Right now we optionally allocate 8 counters per table entry, so in
addition to memory consumed by counters, we require 8 pointers worth of
space in each entry even when counters are not allocated (the default).
Instead, define a UMA zone that returns contiguous per-CPU counter
arrays for use in table entries. On amd64 this reduces sizeof(struct
pfr_kentry) from 216 to 160. The smaller size also results in better
slab efficiency, so memory usage for large tables is reduced by about
28%.
Reviewed by: kp
MFC after: 2 weeks
Differential Revision: https://reviews.freebsd.org/D24843
__builtin_unreachable doesn't raise any compile-time warnings/errors on its
own, so problems with its usage can't be easily detected. While it would be
nice for this situation to change and compilers to at least add a warning
for trivial cases where local state means the instruction can't be reached,
this isn't the case at the moment and likely will not happen.
This commit adds an __assert_unreachable, whose intent is incredibly clear:
it asserts that this instruction is unreachable. On INVARIANTS builds, it's
a panic(), and on non-INVARIANTS it expands to __unreachable().
Existing users of __unreachable() are converted to __assert_unreachable,
to improve debuggability if this assumption is violated.
Reviewed by: mjg
Differential Revision: https://reviews.freebsd.org/D23793
rnh_close callbackes was used by the in[6]_clsroute() handlers,
doing cleanup in the route cloning code. Route cloning was eliminated
somewhere around r186119. Last callback user was eliminated in r186215,
11 years ago.
Differential Revision: https://reviews.freebsd.org/D24793
Last user of rtalloc1() KPI has been eliminated in rS360631.
As kernel is now fully switched to use new routing KPI defined in
rS359823, remove old lookup functions.
Differential Revision: https://reviews.freebsd.org/D24776
Currently each rtentry has dst&gateway allocated separately from another zone,
bloating cache accesses.
Current 'struct rtentry' has 12 "mandatory" radix pointers in the beginning,
leaving 4 usable pointers/32 bytes in the first 2 cache lines (amd64).
Fields needed for the datapath are destination sockaddr and rt_nhop.
So far it doesn't look like there is other routable addressing protocol other
than IPv4/IPv6/MPLS, which uses keys longer than 20 bytes.
With that in mind, embed dst into struct rtentry, making the first 24 bytes
of rtentry within 128 bytes. That is enough to make IPv6 address within first
128 bytes.
It is still pretty easy to add code for supporting separately-allocated dst,
however it doesn't make a lot of sense in having such code without a use case.
As rS359823 moved the gateway to the nexthop structure, the dst embedding change
removes the need for any additional allocations done by rt_setgate().
Lastly, as a part of cleanup, remove counter(9) allocation code, as this field
is not used in packet processing anymore.
Reviewed by: ae
Differential Revision: https://reviews.freebsd.org/D24669
Create rib_lookup() wrapper around per-af dataplane lookup functions.
This will help in the cases of having control plane af-agnostic code.
Switch ifa_ifwithroute() to use this function instead of rtalloc1().
Reviewed by: ae
Differential Revision: https://reviews.freebsd.org/D24731
Eliminate the last rtalloc1() call to finish transition to the new routing
KPI defined in r359823.
Differential Revision: https://reviews.freebsd.org/D24663
After converting routing subsystem customers to use nexthop objects
defined in r359823, some fields in struct rtentry became unused.
This commit removes rt_ifp, rt_ifa, rt_gateway and rt_mtu from struct rtentry
along with the code initializing and updating these fields.
Cleanup of the remaining fields will be addressed by D24669.
This commit also changes the implementation of the RTM_CHANGE handling.
Old implementation tried to perform the whole operation under radix WLOCK,
resulting in slow performance and hacks like using RTF_RNH_LOCKED flag.
New implementation looks up the route nexthop under radix RLOCK, creates new
nexthop and tries to update rte nhop pointer. Only last part is done under
WLOCK.
In the hypothetical scenarious where multiple rtsock clients
repeatedly issue RTM_CHANGE requests for the same route, route may get
updated between read and update operation. This is addressed by retrying
the operation multiple (3) times before returning failure back to the
caller.
Differential Revision: https://reviews.freebsd.org/D24666
With a length of 16, the name ("<if name>:TX(<qid>):callout") typically
gets truncated.
PR: 245712
Reported by: ghuckriede@blackberry.com
MFC after: 1 week
Continue routing subsystem conversion to nhop objects defined in r359823.
Use fields from nhop structure instead of "struct rtentry" fields.
This is one of the last changes prior to removing rt_ifp, rt_ifa,
rt_gateway and rt_mtu from struct rtentry.
Differential Revision: https://reviews.freebsd.org/D24609
With the upcoming multipath changes described in D24141,
rt->rt_nhop can potentially point to a nexthop group instead of
an individual nhop.
To simplify caller handling of such cases, change ifa_rtrequest() callback
to pass changed nhop directly.
Differential Revision: https://reviews.freebsd.org/D24604
Currently functionality resides in rtsock.c, which is a controlling
interface, partially external to the routing subsystem.
Additionally, DDB-supporting functionality is > 100SLOC, which deserves
a separate file.
Given that, move this functionality to a newly-created net/route/ subdir.
Reviewed by: cem
Differential Revision: https://reviews.freebsd.org/D24561
Nexthop objects implementation, defined in r359823,
introduced sys/net/route directory intended to hold all
routing-related code. Move recently-introduced route_temporal.c and
private route_var.h header there.
Differential Revision: https://reviews.freebsd.org/D24597
One of the goals of the new routing KPI defined in r359823
is to entirely hide`struct rtentry` from the consumers.
It will allow to improve routing subsystem internals and deliver
features much faster.
This is one of the last changes, effectively moving struct rtentry
definition to a net/route_var.h header, internal to the routing subsystem.
Differential Revision: https://reviews.freebsd.org/D24580
New fib[46]_lookup() functions support multipath transparently.
Given that, switch the last rtalloc_mpath_fib() calls to
dib4_lookup() and eliminate the function itself.
Note: proper flowid generation (especially for the outbound traffic) is a
bigger topic and will be handled in a separate review.
This change leaves flowid generation intact.
Differential Revision: https://reviews.freebsd.org/D24595
r360292 switched most of the remaining routing customers to a new KPI,
leaving a bunch of wrappers for old routing lookup functions unused.
Remove them from the tree as a part of routing cleanup.
Differential Revision: https://reviews.freebsd.org/D24569
This patch is intended to solve a specific problem that iavf(4)
encounters, but what it does can be extended to solve other issues.
To summarize the iavf(4) issue, if the PF driver configures VLAN
anti-spoof, then the VF driver needs to make sure no untagged traffic is
sent if a VLAN is configured, and vice-versa. This can be an issue when
a VLAN is being registered or unregistered, e.g. when a packet may be on
the ring with a VLAN in it, but the VLANs are being unregistered. This
can cause that tagged packet to go out and cause an MDD event.
To fix this, include a new interface-dependent function that drivers can
implement named IFDI_NEEDS_RESTART(). Right now, this function is called
in iflib_vlan_unregister/register() to determine whether the interface
needs to be stopped and started when a VLAN is registered or
unregistered. The default return value of IFDI_NEEDS_RESTART() is true,
so this fixes the MDD problem that iavf(4) encounters, since the
interface rings are flushed during a stop/init.
A future change to iavf(4) will implement that function just in case the
default value changes, and to make it explicit that this interface reset
is required when a VLAN is added or removed.
Reviewed by: gallatin@
MFC after: 1 week
Sponsored by: Intel Corporation
Differential Revision: https://reviews.freebsd.org/D22086
Introduce new fib[46]_lookup_debugnet() functions serving as a
special interface for the crash-time operations. Underlying
implementation will try to return lookup result if
datastructures are not corrupted, avoding locking.
Convert debugnet to use fib4_lookup_debugnet() and switch it
to use nexthops instead of rtentries.
Reviewed by: cem
Differential Revision: https://reviews.freebsd.org/D24555
Run the bridge datapath under epoch, rather than under the
BRIDGE_LOCK().
We still take the BRIDGE_LOCK() whenever we insert or delete items in
the relevant lists, but we use epoch callbacks to free items so that
it's safe to iterate the lists without the BRIDGE_LOCK.
Tests on mercat5/6 shows this increases bridge throughput significantly,
from 3.7Mpps to 18.6Mpps.
Reviewed by: emaste, philip, melifaro
MFC after: 2 months
Sponsored by: The FreeBSD Foundation
Differential Revision: https://reviews.freebsd.org/D24250
This change is build on top of nexthop objects introduced in r359823.
Nexthops are separate datastructures, containing all necessary information
to perform packet forwarding such as gateway interface and mtu. Nexthops
are shared among the routes, providing more pre-computed cache-efficient
data while requiring less memory. Splitting the LPM code and the attached
data solves multiple long-standing problems in the routing layer,
drastically reduces the coupling with outher parts of the stack and allows
to transparently introduce faster lookup algorithms.
Route caching was (re)introduced to minimise (slow) routing lookups, allowing
for notably better performance for large TCP senders. Caching works by
acquiring rtentry reference, which is protected by per-rtentry mutex.
If the routing table is changed (checked by comparing the rtable generation id)
or link goes down, cache record gets withdrawn.
Nexthops have the same reference counting interface, backed by refcount(9).
This change merely replaces rtentry with the actual forwarding nextop as a
cached object, which is mostly mechanical. Other moving parts like cache
cleanup on rtable change remains the same.
Differential Revision: https://reviews.freebsd.org/D24340
One of the goals of the new routing KPI defined in r359823 is to entirely
hide`struct rtentry` from the consumers. It will allow to improve routing
subsystem internals and deliver more features much faster.
This commit is mostly mechanical change to eliminate direct struct rtentry
field accesses.
The only notable difference is AF_LINK gateway encoding.
AF_LINK gw is used in routing stack for operations with interface routes
and host loopback routes.
In the former case it indicates _some_ non-NULL gateway, as the interface
is the same as in rt_ifp in kernel and rtm_ifindex in rtsock reporting.
In the latter case the interface index inside gateway was used by the IPv6
datapath to verify address scope for link-local interfaces.
Kernel uses struct sockaddr_dl for this type of gateway. This structure
allows for specifying rich interface data, such as mac address and interface
name. However, this results in relatively large structure size - 52 bytes.
Routing stack fils in only 2 fields - sdl_index and sdl_type, which reside
in the first 8 bytes of the structure.
In the new KPI, struct nhop_object tries to be cache-efficient, hence
embodies gateway address inside the structure. In the AF_LINK case it
stores stortened version of the structure - struct sockaddr_dl_short,
which occupies 16 bytes. After D24340 changes, the data inside AF_LINK
gateway will not be used in the kernel at all, leaving rtsock as the only
potential concern.
The difference in rtsock reporting:
(old)
got message of size 240 on Thu Apr 16 03:12:13 2020
RTM_ADD: Add Route: len 240, pid: 0, seq 0, errno 0, flags:<UP,DONE,PINNED>
locks: inits:
sockaddrs: <DST,GATEWAY,NETMASK>
10.0.0.0 link#5 255.255.255.0
(new)
got message of size 200 on Sun Apr 19 09:46:32 2020
RTM_ADD: Add Route: len 200, pid: 0, seq 0, errno 0, flags:<UP,DONE,PINNED>
locks: inits:
sockaddrs: <DST,GATEWAY,NETMASK>
10.0.0.0 link#5 255.255.255.0
Note 40 bytes different (52-16 + alignment).
However, gateway is still a valid AF_LINK gateway with proper data filled in.
It is worth noting that these particular messages (interface routes) are mostly
ignored by routing daemons:
* bird/quagga/frr uses RTM_NEWADDR and ignores prefix route addition messages.
* quagga/frr ignores routes without gateway
More detailed overview on how rtsock messages are used by the
routing daemons to reconstruct the kernel view, can be found in D22974.
Differential Revision: https://reviews.freebsd.org/D24519
Unconditionally use ether_gen_addr() to generate bridge mac addresses. This
function is now less likely to generate duplicate mac addresses across jails.
The old hand rolled hostid based code adds no value.
Reviewed by: bz
Differential Revision: https://reviews.freebsd.org/D24432
If we create two (vnet) jails and create a bridge interface in each we end up
with the same mac address on both bridge interfaces.
These very often conflicts, resulting in same mac address in both jails.
Mitigate this problem by including the jail name in the mac address.
Reviewed by: kevans, melifaro
MFC after: 1 week
Differential Revision: https://reviews.freebsd.org/D24383
Generic if_output() callback signature was modified to use struct route
instead of struct rtentry in r191148, back in 2009.
Quoting commit message:
Change if_output to take a struct route as its fourth argument in order
to allow passing a cached struct llentry * down to L2
Fix bridge_output() to match this signature and update the remaining
comment in if_var.h.
Reviewed by: kp
MFC after: 2 weeks
Differential Revision: https://reviews.freebsd.org/D24394
One of the goals of the new routing KPI defined in r359823 is to
entirely hide`struct rtentry` from the consumers. It will allow to
improve routing subsystem internals and deliver more features much faster.
This change is one of the ongoing changes to eliminate direct
struct rtentry field accesses.
Additionally, with the followup multipath changes, single rtentry can point
to multiple nexthops.
With that in mind, convert rti_filter callback used when traversing the
routing table to accept pair (rt, nhop) instead of nexthop.
Reviewed by: ae
Differential Revision: https://reviews.freebsd.org/D24440
One of the goals of the new routing KPI defined in r359823 is to entirely hide
`struct rtentry` from the consumers. Doing so will allow to improve routing
subsystem internals and deliver features more easily. This change is one of
the ongoing changes to eliminate direct struct rtentry field accesses.
It introduces rtfree_func() wrapper around RTFREE() and reorganises nd6 notification
code to avoid accessing most of the rtentry fields.
Reviewed by: ae
Differential Revision: https://reviews.freebsd.org/D24404
Before this change, lagg double-counted errors from lagg members, and counted
every drop by a lagg member as an error. Eg, if lagg sent a packet, and the
underlying hardware driver dropped it, a counter would be incremented by both
lagg and the underlying driver.
This change attempts to fix that by incrementing lagg's counters only for
errors that do not come from underlying drivers.
Reviewed by: hselasky, jhb
Sponsored by: Netflix
Differential Revision: https://reviews.freebsd.org/D24331
This is the foundational change for the routing subsytem rearchitecture.
More details and goals are available in https://reviews.freebsd.org/D24141 .
This patch introduces concept of nexthop objects and new nexthop-based
routing KPI.
Nexthops are objects, containing all necessary information for performing
the packet output decision. Output interface, mtu, flags, gw address goes
there. For most of the cases, these objects will serve the same role as
the struct rtentry is currently serving.
Typically there will be low tens of such objects for the router even with
multiple BGP full-views, as these objects will be shared between routing
entries. This allows to store more information in the nexthop.
New KPI:
struct nhop_object *fib4_lookup(uint32_t fibnum, struct in_addr dst,
uint32_t scopeid, uint32_t flags, uint32_t flowid);
struct nhop_object *fib6_lookup(uint32_t fibnum, const struct in6_addr *dst6,
uint32_t scopeid, uint32_t flags, uint32_t flowid);
These 2 function are intended to replace all all flavours of
<in_|in6_>rtalloc[1]<_ign><_fib>, mpath functions and the previous
fib[46]-generation functions.
Upon successful lookup, they return nexthop object which is guaranteed to
exist within current NET_EPOCH. If longer lifetime is desired, one can
specify NHR_REF as a flag and get a referenced version of the nexthop.
Reference semantic closely resembles rtentry one, allowing sed-style conversion.
Additionally, another 2 functions are introduced to support uRPF functionality
inside variety of our firewalls. Their primary goal is to hide the multipath
implementation details inside the routing subsystem, greatly simplifying
firewalls implementation:
int fib4_lookup_urpf(uint32_t fibnum, struct in_addr dst, uint32_t scopeid,
uint32_t flags, const struct ifnet *src_if);
int fib6_lookup_urpf(uint32_t fibnum, const struct in6_addr *dst6, uint32_t scopeid,
uint32_t flags, const struct ifnet *src_if);
All functions have a separate scopeid argument, paving way to eliminating IPv6 scope
embedding and allowing to support IPv4 link-locals in the future.
Structure changes:
* rtentry gets new 'rt_nhop' pointer, slightly growing the overall size.
* rib_head gets new 'rnh_preadd' callback pointer, slightly growing overall sz.
Old KPI:
During the transition state old and new KPI will coexists. As there are another 4-5
decent-sized conversion patches, it will probably take a couple of weeks.
To support both KPIs, fields not required by the new KPI (most of rtentry) has to be
kept, resulting in the temporary size increase.
Once conversion is finished, rtentry will notably shrink.
More details:
* architectural overview: https://reviews.freebsd.org/D24141
* list of the next changes: https://reviews.freebsd.org/D24232
Reviewed by: ae,glebius(initial version)
Differential Revision: https://reviews.freebsd.org/D24232
Split their functionality by moving random seed allocation
to SYSINIT and calling (new) generic multipath function from
standard IPv4/IPv5 RIB init handlers.
Differential Revision: https://reviews.freebsd.org/D24356
No functional changes.
* Move route addition / route deletion code from rtrequest1_fib()
to add_route() and del_route() respectively.
* Rename rtrequest1_fib_change() to change_route() for consistency.
* Shrink the scope of ugly info #defines.
MFC after: 2 weeks
Differential Revision: https://reviews.freebsd.org/D24349
Prepare the ground for a rework of the bridge locking approach. We will
use an epoch-based approach in the datapath and making it safe to
iterate over the interface, span and rtnode lists without holding the
BRIDGE_LOCK. Replace the relevant lists by their ConcurrencyKit
equivalents.
No functional change in this commit.
Reviewed by: emaste, ae, philip (previous version)
Sponsored by: The FreeBSD Foundation
Differential Revision: https://reviews.freebsd.org/D24249
taskqgroup initialization was broken into two steps:
1. allocate the taskqgroup structure, at SI_SUB_TASKQ;
2. initialize taskqueues, start taskqueue threads, enqueue "binder"
tasks to bind threads to specific CPUs, at SI_SUB_SMP.
Step 2 tries to handle the case where tasks have already been attached
to a queue, by migrating them to their intended queue. In particular,
tasks can't be enqueued before step 2 has completed. This breaks NFS
mountroot on systems using an iflib-based driver when EARLY_AP_STARTUP
is not defined, since mountroot happens before SI_SUB_SMP in this case.
Simplify initialization: do all initialization except for CPU binding at
SI_SUB_TASKQ. This means that until CPU binding is completed, group
tasks may be executed on a CPU other than that to which they were bound,
but this should not be a problem for existing users of the taskqgroup
KPIs.
Reported by: sbruno
Tested by: bdragon, sbruno
MFC after: 1 month
Sponsored by: The FreeBSD Foundation
Differential Revision: https://reviews.freebsd.org/D24188
The previous descriptor string specified that all fields were significant for
match. However, the only significant fields for in-tree drivers are
vendor:devid, and the fictitious zero values constructed by PVID() did not
match real subvendor, subdevice, revision, and/or class values, resulting in no
automatic probe.
If a future iflib driver needs to match on other criteria, the descriptor
string can be updated accordingly. (E.g., "V32" and ~0 for unspecified values
in PVID().)
Reported by: mav
Sponsored by: Dell EMC Isilon
ThunderX cluster systems are panicking on boot with a failed assertion
MPASS(gtask != NULL && gtask->gt_taskqueue != NULL). Split the
assertion so that it's clear which part is failing.
ifsd_cidx is never used, and the line removed from rxd_frag_to_sd() is
just dead code.
Reviewed by: erj, gallatin
MFC after: 1 week
Differential Revision: https://reviews.freebsd.org/D23951
The vmx driver is an example of an iflib driver that might report
packets using non-contiguous descriptors (with unused descriptors
either between received packets or between the fragments of a received
packet), so this assertion needs to be removed.
For such drivers, the freelist producer and consumer indexes don't
relate directly to driver ring slots (the driver deals directly with
freelist buffer indexes supplied by iflib during refill, and reports
them with each fragment during packet reception), but do continue to
be used by iflib for accounting, such as determining the number of
ring slots that are refillable.
PR: 243126, 243392, 240628
Reported by: avg, alexandr.oleynikov@gmail.com, Harald Schmalzbauer
Reviewed by: gallatin
MFC after: 1 week
Differential Revision: https://reviews.freebsd.org/D23946
The dmamap for zero-length fragments should not be unloaded, as doing
so breaks the the cluster-reuse logic in _iflib_fl_refill().
All zero-length fragments are now handled by the assemble_segments()
path so that the cluster-reuse logic there does not have to be
replicated in the small-single-fragment-packet path of
iflib_rxd_pkt_get().
Packets consisting entirely of zero-length fragments (which result in
a NULL mbuf pointer) are now properly tolerated. This allows drivers
(such as the vmx driver) to pass such packets to iflib when a
descriptor error occurs during packet reception, the advantage being
that the refill of descriptors associated with the error packet are
handled via the existing iflib machinery without having to duplicate
parts of that machinery in the driver to handle that error case.
Reviewed by: avg, erj, gallatin
MFC after: 1 week
Differential Revision: https://reviews.freebsd.org/D23945
This fixes a bug in iflib freelist management that breaks the required
correspondence between freelist indexes and driver ring slots.
PR: 243126, 243392, 240628
Reported by: avg, alexandr.oleynikov@gmail.com, Harald Schmalzbauer
Reviewed by: avg, gallatin
MFC after: 1 week
Differential Revision: https://reviews.freebsd.org/D23943
When I did the use_numa support, I missed the fact that there is
a separate hash function for send tag nic selection. So when
use_numa is enabled, ktls offload does not work properly, as it
does not reliably allocate a send tag on the proper egress nic
since different egress nics are selected for send-tag allocation
and packet transmit. To fix this, this change:
- refectors lacp_select_tx_port_by_hash() and
lacp_select_tx_port() to make lacp_select_tx_port_by_hash()
always called by lacp_select_tx_port()
- pre-shifts flowids to convert them to hashes when calling lacp_select_tx_port_by_hash()
- adds a numa_domain field to if_snd_tag_alloc_params
- plumbs the numa domain into places where we allocate send tags
In testing with NIC TLS setup on a NUMA machine, I see thousands
of output errors before the change when enabling
kern.ipc.tls.ifnet.permitted=1. After the change, I see no
errors, and I see the NIC sysctl counters showing active TLS
offload sessions.
Reviewed by: rrs, hselasky, jhb
Sponsored by: Netflix
Implement counting of table entries linked on a per-table base
with an optional (if set > 0) limit of the maximum number of table
entries.
For that the public lltable_link_entry() and lltable_unlink_entry()
functions as well as the internal function pointers change from void
to having an int return type.
Given no consumer currently sets the new llt_maxentries this can be
committed on its own. The moment we make use of the table limits,
the callers of the link function must check the return value as
it can change and entries might not be added.
Adjustments for IPv6 (and possibly IPv4) will follow.
Sponsored by: Netflix (originally)
Reviewed by: melifaro
MFC after: 2 weeks
Differential Revision: https://reviews.freebsd.org/D22713
When generating an cloned interface instance in edsc_clone_create(),
generate a MAC address from the FF OUI with ether_gen_addr(). This allows us
to have unique local-link addresses. Previously, the MAC address was zero.
Submitted by: Neel Chauhan <neel AT neelc DOT org>
Differential Revision: https://reviews.freebsd.org/D23842
r357614 added CTLFLAG_NEEDGIANT to make it easier to find nodes that are
still not MPSAFE (or already are but aren’t properly marked).
Use it in preparation for a general review of all nodes.
This is non-functional change that adds annotations to SYSCTL_NODE and
SYSCTL_PROC nodes using one of the soon-to-be-required flags.
Mark all obvious cases as MPSAFE. All entries that haven't been marked
as MPSAFE before are by default marked as NEEDGIANT
Approved by: kib (mentor, blanket)
Commented by: kib, gallatin, melifaro
Differential Revision: https://reviews.freebsd.org/D23718
like the mlx-c5 and c6 that require a "setup" routine before
the tcp_ratelimit code can declare and use a rate. I add the
setup routine to if_var as well as fix tcp_ratelimit to call it.
I also revisit the rates so that in the case of a mlx card
of type c5/6 we will use about 100 rates concentrated in the range
where the most gain can be had (1-200Mbps). Note that I have
tested these on a c5 and they work and perform well. In fact
in an unloaded system they pace right to the correct rate (great
job mlx!). There will be a further commit here from Hans that
will add the respective changes to the mlx driver to support this
work (which I was testing with).
Sponsored by: Netflix Inc.
Differential Revision: ttps://reviews.freebsd.org/D23647
The locking defines for if_bridge used to live in if_bridgevar.h, but
they're only ever used by the bridge implementation itself (in
if_bridge.c). Moving them into the .c file.
Reported by: philip, emaste
Sponsored by: The FreeBSD Foundation
Differential Revision: https://reviews.freebsd.org/D23808
switch over to opt-in instead of opt-out for epoch.
Instead of IFF_NEEDSEPOCH, provide IFF_KNOWSEPOCH. If driver marks
itself with IFF_KNOWSEPOCH, then ether_input() would not enter epoch
when processing its packets.
Now this will create recursive entrance in epoch in >90% network
drivers, but will guarantee safeness of the transition.
Mark several tested drivers as IFF_KNOWSEPOCH.
Reviewed by: hselasky, jeff, bz, gallatin
Differential Revision: https://reviews.freebsd.org/D23674
Revert parts of r353274 replacing vnet_state with a shutdown flag.
Not having the state flag for the current SI_SUB_* makes it harder to debug
kernel or module panics related to VNET bringup or teardown.
Not having the state also does not allow us to check for other dependency
levels between components, e.g. for moving interfaces.
Expand the VNET structure with the new boolean flag indicating that we are
doing a shutdown of a given vnet and update the vnet magic cookie for the
change.
Update libkvm to compile with a bool in the kernel struct.
Bump __FreeBSD_version for (external) module builds to more easily detect
the change.
Reviewed by: hselasky
MFC after: 1 week
Differential Revision: https://reviews.freebsd.org/D23097
When VIMAGE is enabled make sure the "m_pkthdr.rcvif" pointer is set
for all mbufs being input by the IGMP/MLD6 code. Else there will be a
NULL-pointer dereference in the netisr code when trying to set the
VNET based on the incoming mbuf. Add an assert to catch this when
queueing mbufs on a netisr to make debugging of similar cases easier.
Found by: Vladislav V. Prodan
PR: 244002
Reviewed by: bz@
MFC after: 1 week
Sponsored by: Mellanox Technologies
When the receive ring cannot be filled with mbufs, due to lack of memory,
no more interrupts may be generated to fill the receive ring later on.
Make sure to have a watchdog, to try refilling the receive ring from time
to time, hopefully when more mbufs are available.
Differential Revision: https://reviews.freebsd.org/D23315
MFC after: 1 week
Reviewed by: gallatin@
Sponsored by: Mellanox Technologies
The reason for this change is to make it clear the scope of the in-kernel usage
of IFM_TYPE_DESCRIPTIONS and IFM_SUBTYPE_ETHERNET_DESCRIPTIONS macros. Also it
is somewhat better C.
Reviewed by: hselasky
Sponsored by: Mellanox Technologies
Differential revision: https://reviews.freebsd.org/D23620
Recent network epoch changes have left some drivers unexpectedly broken
and there is not yet a consensus on the correct fix. This is patch is
a minor performance impact until we can agree on the correct path
forward.
Reviewed by: core, network, imp, glebius, hselasky
Differential Revision: https://reviews.freebsd.org/D23515
During vnet cleanup vnet_if_uninit() checks that no more interfaces remain in
the vnet. Any interface borrowed from another vnet is returned by
vnet_if_return(). Other interfaces (i.e. cloned interfaces) should have been
destroyed by their cloner at this point.
The if_vlan VNET_SYSUNINIT had priority SI_ORDER_FIRST, which means it had
equal priority as vnet_if_uninit(). In other words: it was possible for it to
be called *after* vnet_if_uninit(), which would lead to assertion failures.
Set the priority to SI_ORDER_ANY, like other cloners to ensure that vlan
interfaces are destroyed before we enter vnet_if_uninit().
The sys/net/if_vlan test provoked this.
Software interrupt handlers are allowed to sleep. In swi_net() there
is a read lock behind NETISR_RLOCK() which in turn ends up calling
msleep() which means the whole of swi_net() cannot be protected by an
EPOCH(9) section. By default the NETISR_LOCKING feature is disabled.
This issue was introduced by r357004. This is a preparation step for
replacing the functionality provided by r357004.
Found by: kib@
Sponsored by: Mellanox Technologies
if_epair used the 'params' argument to pass a pointer to the b interface
through if_clone_create().
This pointer can be controlled by userspace, which means it could be abused to
trigger a panic. While this requires PRIV_NET_IFCREATE
privileges those are assigned to vnet jails, which means that vnet jails
could panic the system.
Reported by: Ilja Van Sprundel <ivansprundel@ioactive.com>
MFC after: 3 days
supposedly may call into ether_input() without network epoch.
They all need to be reviewed before 13.0-RELEASE. Some may need
be fixed. The flag is not planned to be used in the kernel for
a long time.
In upcoming changes ether_input() is going to be changed not
to enter the network epoch. It is going to be responsibility
of network interrupt. In case of iflib - its taskqueue.
Those interfaces may implicitly change their MTU on addition of parent
interface in addition to normal SIOCSIFMTU ioctl path, where the route
MTUs are updated normally.
MFC after: 2 weeks
Sponsored by: iXsystems, Inc.
Redirect (and temporal) route expiration was broken a while ago.
This change brings route expiration back, with unified IPv4/IPv6 handling code.
It introduces net.inet.icmp.redirtimeout sysctl, allowing to set
an expiration time for redirected routes. It defaults to 10 minutes,
analogues with net.inet6.icmp6.redirtimeout.
Implementation uses separate file, route_temporal.c, as route.c is already
bloated with tons of different functions.
Internally, expiration is implemented as an per-rnh callout scheduled when
route with non-zero rt_expire time is added or rt_expire is changed.
It does not add any overhead when no temporal routes are present.
Callout traverses entire routing tree under wlock, scheduling expired routes
for deletion and calculating the next time it needs to be run. The rationale
for such implemention is the following: typically workloads requiring large
amount of routes have redirects turned off already, while the systems with
small amount of routes will not inhibit large overhead during tree traversal.
This changes also fixes netstat -rn display of route expiration time, which
has been broken since the conversion from kread() to sysctl.
Reviewed by: bz
MFC after: 3 weeks
Differential Revision: https://reviews.freebsd.org/D23075
Correction after r333476:
- write this as LOG_DEBUG again instead of LOG_INFO;
- get back function name into the message;
- error may be ESRCH if an address is removed in process (by carp f.e.),
not only ENOENT;
- expression complexity grows, so try making it more readable.
MFC after: 1 week
subsystems tend to need to know about it, and including if_var.h is
huge header pollution for them. Polluting possible non-network
users with single symbol seems much lesser evil.
- Remove non-preemptible network epoch. Not used yet, and unlikely
to get used in close future.
The netmap passthrough subsystem requires proper support in the
hypervisor. In particular, two PCI device ids (from the Red Hat
PCI vendor id 0x1b36) need to be assigned to the two netmap
virtual devices. We then disable these devices until the ids have
not been assigned, in order to avoid conflicts with other
virtual devices emulated by upstream QEMU.
PR: 241774
MFC after: 3 days
Having metadata such as fibnum or vnet in the struct rib_head
is handy as it eases building functionality in the routing space.
This change is required to properly bring back route redirect support.
Reviewed by: bz
MFC after: 3 weeks
Differential Revision: https://reviews.freebsd.org/D23047
Add an option flag so that arbitrary updates to a lagg's configuration
do not clear sc_stride. Preseve compatibility for old ifconfig
binaries. Update ifconfig to use the new flag and improve the casting
used when parsing the option parameter.
Modify the RR transmit function to avoid locklessly reading sc_stride
twice. Ensure that sc_stride is always 1 or greater.
Reviewed by: hselasky
MFC after: 1 week
Sponsored by: The FreeBSD Foundation
Differential Revision: https://reviews.freebsd.org/D23092
if_vmove can fail if it lost a race and the vnet's already been moved. The
callers (and their callers) can generally cope with this, but right now
success is assumed. Plumb out the ENOENT from if_detach_internal if it
happens so that the error's properly reported to userland.
Reviewed by: bz, kp
MFC after: 1 week
Differential Revision: https://reviews.freebsd.org/D22780
While changing link state in iflib_link_state_change(), queues are
marked as IFLIB_QUEUE_IDLE to disable watchdog. Currently, iflib_timer()
watchdog does not check for previous queue status before marking it as
IFLIB_QUEUE_HUNG.
This patch adds check of queue status before marking it as hung.
Signed-off-by: Piotr Pietruszewski <piotr.pietruszewski@intel.com>
PR: 239240
Submitted by: Piotr Pietruszewski <piotr.pietruszewski@intel.com>
Reported by: ultima@
Reviewed by: gallatin@, erj@
MFC after: 3 days
Sponsored by: Intel Corporation
Differential Revision: https://reviews.freebsd.org/D21712
Amount of changes to the original code has been intentionally minimised
to ease diffing.
The changes are mostly mechanical, with the following exceptions:
* lltable handler is now called directly based of RTF_LLINFO flag presense.
* "report" logic for updating rtm in RTM_GET/RTM_DELETE has been simplified,
fixing several potential use-after-free cases in rt_addrinfo.
* llable asserts has been replaced with error-returning, preventing kernel
crashes when lltable gw af family is invalid (root required).
MFC after: 2 weeks
Differential Revision: https://reviews.freebsd.org/D22864
- Only take an ifaddr ref in in rt_exportinfo() if the caller explicitly
requests it. Take care to release it in this case.
- Don't unconditionally take a ref in rtrequest1_fib(). rt_getifa_fib()
will acquire a reference, in which case we would previously acquire
two references.
- Stop taking a reference in rtinit1() before calling rtrequest1_fib().
rtrequest1_fib() will acquire a reference for the RTM_ADD case.
PR: 242746
Reviewed by: melifaro (previous version)
Tested by: ghuckriede@blackberry.com
MFC after: 1 week
Sponsored by: The FreeBSD Foundation
Differential Revision: https://reviews.freebsd.org/D22912
- Don't allow an unprivileged user to set the stride. [1]
- Only set the stride under the softc lock.
- Rename the internal fields to accurately reflect their use. Keep
ro_bkt to avoid changing the user API.
- Simplify the implementation. The port index is just sc_seq / stride.
- Document rr_limit in ifconfig.8.
Reported by: Ilja Van Sprundel <ivansprundel@ioactive.com> [1]
MFC after: 1 week
Sponsored by: The FreeBSD Foundation
Differential Revision: https://reviews.freebsd.org/D22857
Fix some style in if_addgroup(). No functional change intended.
Reviewed by: hselasky
MFC after: 2 weeks
Sponsored by: The FreeBSD Foundation
Differential Revision: https://reviews.freebsd.org/D22892
r297816 added some bpf magic for VIMAGE unconditionally which no longer
allows kernels to compile without bpf (but with other networking).
Add the missing ifdef checks and allow a kernel to compile without bpf
again.
PR: 242136
Reported by: dave mischler.com
MFC after: 2 weeks
Add explicit SI_SUB_EPOCH, after SI_SUB_TASKQ and before SI_SUB_SMP
(EARLY_AP_STARTUP). Rename existing "SI_SUB_TASKQ + 1" to SI_SUB_EPOCH.
epoch(9) consumers cannot epoch_alloc() before SI_SUB_EPOCH:SI_ORDER_SECOND,
but likely should allocate before SI_SUB_SMP. Prior to this change,
consumers (well, epoch itself, and net/if.c) just open-coded the
SI_SUB_TASKQ + 1 order to match epoch.c, but this was fragile.
Reviewed by: mmacy
Differential Revision: https://reviews.freebsd.org/D22503
Adjust the logic in htable_unlink_entry() to the one in
htable_link_entry() saving a block indent and making it more clear
in which case we do not do any work.
No functional change.
MFC after: 3 weeks
Sponsored by: Netflix
Remove function prototypes which are not needed (no use before function
definition for these file static functions).
MFC after: 3 weeks
Sponsored by: Netflix
Remove the long (8? years ago) #if 0 marked function lltable_drain() and
while here also remove the unused function llentry_alloc() which has call
paths tools keep finding and are never used.
Sponsored by: Netflix
interface has changed.
During vlan reconfiguration without destroying interface, it is possible,
that parent interface will be changed. This usually means, that link
layer address of vlan will be different. Therefore we need to update all
associated with vlan's addresses permanent llentries - NDP for IPv6
addresses, and ARP for IPv4 addresses. This is done via lladdr_task
execution. To avoid extra work, before execution do the check, that L2
address is different.
No objection from: #network
Obtained from: Yandex LLC
MFC after: 1 week
Sponsored by: Yandex LLC
Differential Revision: https://reviews.freebsd.org/D22243
Some places in network code may need to verify that an ethernet address
is not the 'zero' address. Provide a standard macro ETHER_IS_ZERO for
this purpose, similar to the ETHER_IS_BROADCAST macro already available.
This patch also removes previous ETHER_IS_ZERO definitions in several
USB ethernet drivers, in favor of this centrally-located macro.
Signed-off-by: Jacob Keller <jacob.e.keller@intel.com>
Submitted by: Jacob Keller <jacob.e.keller@intel.com>
Reviewed by: erj@
Sponsored by: Intel Corporation
Differential Revision: https://reviews.freebsd.org/D21240
DMA memory allocations using the bus_dma.h interface are not properly
released in all cases for both Tx and Rx. This causes ~448 bytes of
M_DEVBUF allocations to be leaked.
First, the DMA maps for Rx are not properly destroyed. A slight attempt
is made in iflib_fl_bufs_free to destroy the maps if we're detaching.
However, this function may not be reliably called during detach. Indeed,
there is a comment "asking" if this should be moved out.
Fix this by moving the bus_dmamap_destroy call into iflib_rx_sds_free,
where we already sync and unload the DMA.
Second, the DMA tag associated with the ifr_ifdi descriptor DMA is not
released properly anywhere. Add a call to iflib_dma_free in
iflib_rx_structures_free.
Third, use of NULL as a canary value on the map pointer returned by
bus_dmamap_create is not valid. On some platforms, notably x86, this
value may be NULL. In this case, we fail to properly release the related
resources.
Remove the NULL checks on map values in both iflib_fl_bufs_free and
iflib_txsd_destroy.
With all of these fixes applied, the leaks to M_DEVBUF are squelched,
and iflib drivers now seem to properly cleanup when detaching.
Signed-off-by: Jacob Keller <jacob.e.keller@intel.com>
Submitted by: Jacob Keller <jacob.e.keller@intel.com>
Reviewed by: erj@, gallatin@
MFC after: 1 week
Sponsored by: Intel Corporation
Differential Revision: https://reviews.freebsd.org/D22203
The valectl(4) program is used to manage vale(4) switches.
Add it to the system commands so that it can be used right away.
This program was previously called vale-ctl, and stored in
tools/tools/netmap
Reviewed by: hrs, bcr, lwhsu, kevans
MFC after: 1 week
Differential Revision: https://reviews.freebsd.org/D22146
From Jake:
The iflib stack failed to release all of the memory allocated under
M_IFLIB during device detach.
Specifically, the ifmp_ring, the ift_ifdi Tx DMA info, and the ifr_ifdi Rx
DMA info were not being released.
Release this memory so that iflib won't leak memory when a device
detaches.
Since we're freeing the ift_ifdi pointer during iflib_txq_destroy we
need to call this only after iflib_dma_free in iflib_tx_structures_free.
Additionally, also ensure that we destroy the callout mutex associated
with each Tx queue when we free it.
Signed-off-by: Jacob Keller <jacob.e.keller@intel.com>
Submitted by: Jacob Keller <jacob.e.keller@intel.com>
Reviewed by: erj@, gallatin@
MFC after: 1 week
Sponsored by: Intel Corporation
Differential Revision: https://reviews.freebsd.org/D22157
and IPv6. With IPv6 we may call if_addmulti() in context of processing
of an incoming packet. Usually this is interrupt context. While most
of the NIC drivers are able to reprogram multicast filters without
sleeping, some of them can't. An example is e1000 family of drivers.
With iflib conversion the problem was somewhat hidden. Iflib processes
packets in private taskqueue, so going to sleep doesn't trigger an
assertion. However, the sleep would block operation of the driver and
following incoming packets would fill the ring and eventually would
start being dropped. Enabling epoch for the full time of a packet
processing again started to trigger assertions for e1000.
Fix this problem once and for all using a general taskqueue to call
if_ioctl() method in all cases when if_addmulti() is called in a
non sleeping context. Note that nobody cares about returned value.
Reviewed by: hselasky, kib
Differential Revision: https://reviews.freebsd.org/D22154
From Jake:
Calling ether_ifdetach after iflib_stop leads to a potential race where
a stale ifp pointer can remain in the route entry list for IPv6 traffic.
This will potentially cause a page fault or other system instability if
the ifp pointer is accessed.
Move both iflib_netmap_detach and ether_ifdetach to be called prior to
iflib_stop. This avoids the race above, and helps ensure that other ifp
references are removed before stopping the interface.
Submitted by: Jacob Keller <jacob.e.keller@intel.com>
Reviewed by: erj@, gallatin@, jhb@
MFC after: 3 days
Sponsored by: Intel Corporation
Differential Revision: https://reviews.freebsd.org/D22071
Shuffle headers around to more appropriate #ifdef OPTION blocks (INET vs.
INET6) -- double checked LINT-{NOINET,NOINET6,NOIP}, all seem good.
Reported by: cem
Simply adding MODULE_VERSION does not do the trick, because the modules
haven't been declared. This should actually fix modfind/kldstat, which
r351229 aimed and failed to do.
This should make vm-bhyve do the right thing again when using the ports
version, rather than the latest version not in ports.
MFC after: 3 days
net.link.tap.user_open has historically allowed non-root users to do devfs
cloning and open /dev/tap* nodes based on permissions. Loosen this up to
make it only allow users to do devfs cloning -- we no longer check it in
tunopen.
This allows tap devices to be created that can actually be opened by a user,
rather than swiftly restricting them to root because the magic sysctl has
not been set.
The sysctl has not yet been completely deprecated, because more thought is
needed for how to handle the devfs cloning case. There is not an easy
suitable replacement for the sysctl there, and more care needs to be placed
in determining whether that's OK or not.
PR: 200185
cdevpriv dtors will be called when the reference count on the associated
struct file drops to 0, while d_close can be unreliable for cleaning up
state at "last close" for a number of reasons. As far as tunclose/tundtor is
concerned the difference is minimal, so make the switch.
This allows us to avoid some dance in tunopen for dealing with the
possibility of dev->si_drv1 being NULL as it's set prior to the devfs node
being created in all cases.
There's still the possibility that the tun device hasn't been fully
initialized, since that's done after the devfs node was created. Alleviate
this by returning ENXIO if we're not to that point of tuncreate yet.
This work is what sparked r353128, full initialization of cloned devices
w/ specified make_dev_args.
Thanks to cem@ for discussing the issue which resulted in this patch.
Reviewed by: cem@
Sponsored by: Netflix, Inc.
Differential Revision: https://reviews.freebsd.org/D22089
This patch is part of an effort to make bhyve networking (in particular TCP)
faster. The key strategy to enhance TCP throughput is to let the whole packet
datapath work with TSO/LRO packets (up to 64KB each), so that the per-packet
overhead is amortized over a large number of bytes.
This capability is supported in the guest by means of the vtnet(4) driver,
which is able to handle TSO/LRO packets leveraging the virtio-net header
(see struct virtio_net_hdr and struct virtio_net_hdr_mrg_rxbuf).
A bhyve VM exchanges packets with the host through a network backend,
which can be vale(4) or if_tap(4).
While vale(4) supports TSO/LRO packets, if_tap(4) does not.
This patch extends if_tap(4) with the ability to understand the virtio-net
header, so that a tapX interface can process TSO/LRO packets.
A couple of ioctl commands have been added to configure and probe the
virtio-net header. Once the virtio-net header is set, the tapX interface
acquires all the IFCAP capabilities necessary for TSO/LRO.
Reviewed by: kevans
Differential Revision: https://reviews.freebsd.org/D21263
NetGDB(4) is a component of a system using a panic-time network stack to
remotely debug crashed FreeBSD kernels over the network, instead of
traditional serial interfaces.
There are three pieces in the complete NetGDB system.
First, a dedicated proxy server must be running to accept connections from
both NetGDB and gdb(1), and pass bidirectional traffic between the two
protocols.
Second, the NetGDB client is activated much like ordinary 'gdb' and
similarly to 'netdump' in ddb(4) after a panic. Like other debugnet(4)
clients (netdump(4)), the network interface on the route to the proxy server
must be online and support debugnet(4).
Finally, the remote (k)gdb(1) uses 'target remote <proxy>:<port>' (like any
other TCP remote) to connect to the proxy server.
The NetGDB v1 protocol speaks the literal GDB remote serial protocol, and
uses a 1:1 relationship between GDB packets and sequences of debugnet
packets (fragmented by MTU). There is no encryption utilized to keep
debugging sessions private, so this is only appropriate for local
segments or trusted networks.
Submitted by: John Reimer <john.reimer AT emc.com> (earlier version)
Discussed some with: emaste, markj
Relnotes: sure
Differential Revision: https://reviews.freebsd.org/D21568
It remains unattached to any client protocol. Netdump is unaffected
(remaining half-duplex). The intended consumer is NetGDB.
Submitted by: John Reimer <john.reimer AT emc.com> (earlier version)
Discussed with: markj
Differential Revision: https://reviews.freebsd.org/D21541
Loosen requirements for connecting to debugnet-type servers. Only require a
destination address; the rest can theoretically be inferred from the routing
table.
Relax corresponding constraints in netdump(4) and move ifp validation to
debugnet connection time.
Submitted by: John Reimer <john.reimer AT emc.com> (earlier version)
Reviewed by: markj
Differential Revision: https://reviews.freebsd.org/D21482
Add a 'X -s <server> -c <client> [-g <gateway>] -i <interface>' subroutine
to the generic debugnet code. The imagined use is both netdump, shown here,
and NetGDB (vaporware). It uses the ddb(4) lexer, with some new extensions,
to parse out IPv4 addresses.
'Netdump' uses the generic debugnet routine to load a configuration and
start a dump, without any netdump configuration prior to panic.
Loosely derived from work by: John Reimer <john.reimer AT emc.com>
Reviewed by: markj
Differential Revision: https://reviews.freebsd.org/D21460
The in-tree netdump code has always ignored non-directed ARP requests, and
that seems to work most of the time for netdump.
In my work and testing on NetGDB, it seems like sometimes the remote FreeBSD
conversant (the non-panic system) will send broadcast-destination ARP
requests to the debugnet kernel; without this change, those are dropped and
the remote will see EHOSTDOWN "Host is down" errors from the userspace
interface of the network stack.
Discussed with: markj
Similar to INET checksums, lazily validate UDP checksums when the driver has
already performed the check for us. Like debugnet(4) INET checksums,
validation in software is left as future work.
Reviewed by: markj
Differential Revision: https://reviews.freebsd.org/D21745
Debugnet is a simplistic and specialized panic- or debug-time reliable
datagram transport. It can drive a single connection at a time and is
currently unidirectional (debug/panic machine transmit to remote server
only).
It is mostly a verbatim code lift from netdump(4). Netdump(4) remains
the only consumer (until the rest of this patch series lands).
The INET-specific logic has been extracted somewhat more thoroughly than
previously in netdump(4), into debugnet_inet.c. UDP-layer logic and up, as
much as possible as is protocol-independent, remains in debugnet.c. The
separation is not perfect and future improvement is welcome. Supporting
INET6 is a long-term goal.
Much of the diff is "gratuitous" renaming from 'netdump_' or 'nd_' to
'debugnet_' or 'dn_' -- sorry. I thought keeping the netdump name on the
generic module would be more confusing than the refactoring.
The only functional change here is the mbuf allocation / tracking. Instead
of initiating solely on netdump-configured interface(s) at dumpon(8)
configuration time, we watch for any debugnet-enabled NIC for link
activation and query it for mbuf parameters at that time. If they exceed
the existing high-water mark allocation, we re-allocate and track the new
high-water mark. Otherwise, we leave the pre-panic mbuf allocation alone.
In a future patch in this series, this will allow initiating netdump from
panic ddb(4) without pre-panic configuration.
No other functional change intended.
Reviewed by: markj (earlier version)
Some discussion with: emaste, jhb
Objection from: marius
Differential Revision: https://reviews.freebsd.org/D21421
Older network equipment used the ethertypes 0x9100, 0x9200, and 0x9300 for
outer VLANs, before standardisation introduced 0x88a8.
Submitted by: Lutz Donnerhacke <lutz_donnerhacke.de>
Differential Revision: https://reviews.freebsd.org/D21846
general is allowed to sleep. Don't enter the epoch for the
whole duration. If some event handlers need the epoch, they
should handle that theirselves.
Discussed with: hselasky
same after the network stack was epochified. Merge the two into one function
and cleanup all uses of ifnet_byindex_locked().
While at it:
- Add branch prediction macros.
- Make sure the ifnet pointer is only deferred once,
also when code optimisation is disabled.
Sponsored by: Mellanox Technologies
This fixes the following assert when "options RATELIMIT" is used:
panic()
malloc()
sysctl_add_oid()
tcp_rl_ifnet_link()
do_link_state_change()
taskqueue_run_locked()
Sponsored by: Mellanox Technologies
instead of calling an SCTP specific function from the IP code.
This is a requirement of supporting SCTP as a kernel loadable module.
This patch was developed by markj@, I tweaked a bit the SCTP related
code.
Submitted by: markj@
MFC after: 3 days
on interface. Such function could been implemented on top of
the if_foreach_llm?addr(), but several drivers need counting,
so avoid copy-n-paste inside the drivers.
addresses. The KPI doesn't reveal neither how addresses are stored,
how the access to them is synchronized, neither reveal struct ifaddr
and struct ifmaddr.
Reviewed by: gallatin, erj, hselasky, philip, stevek
Differential Revision: https://reviews.freebsd.org/D21943
protected by IF_ADDR_LOCK(), which was a mutex, so that two simultaneous
if_setlladdr() can't execute. Later it was switched to IF_ADDR_RLOCK(),
likely by a mistake. Later it was switched to NET_EPOCH_ENTER(). Then I
incorrectly added NET_EPOCH_ASSERT() here.
In reality ifp->if_addr never goes away and never changes its length. So,
doing bcopy() in it is always "safe", meaning it won't dereference a wrong
pointer or write into someone's else memory. Of course doing two bcopy() in
parallel would result in a mess of two addresses, but net epoch doesn't
protect against that, neither IF_ADDR_RLOCK() did.
So for now, just remove the assertion and leave for later a proper fix.
Reported by: markj
Make sure the vnet_shutdown field is not set until after all
VNET_SYSUNINIT()'s in the SI_SUB_VNET_DONE subsystem have been
executed. Especially the vnet_if_return() functions requires that
if_move() is still operational.
Reported by: lwhsu@
MFC after: 1 week
Sponsored by: Mellanox Technologies
When epoch(9) was introduced to network stack, it was basically
dropped in place of existing locking, which was mutexes and
rwlocks. For the sake of performance mutex covered areas were
as small as possible, so became epoch covered areas.
However, epoch doesn't introduce any contention, it just delays
memory reclaim. So, there is no point to minimise epoch covered
areas in sense of performance. Meanwhile entering/exiting epoch
also has non-zero CPU usage, so doing this less often is a win.
Not the least is also code maintainability. In the new paradigm
we can assume that at any stage of processing a packet, we are
inside network epoch. This makes coding both input and output
path way easier.
On output path we already enter epoch quite early - in the
ip_output(), in the ip6_output().
This patch does the same for the input path. All ISR processing,
network related callouts, other ways of packet injection to the
network stack shall be performed in net_epoch. Any leaf function
that walks network configuration now asserts epoch.
Tricky part is configuration code paths - ioctls, sysctls. They
also call into leaf functions, so some need to be changed.
This patch would introduce more epoch recursions (see EPOCH_TRACE)
than we had before. They will be cleaned up separately, as several
of them aren't trivial. Note, that unlike a lock recursion the
epoch recursion is safe and just wastes a bit of resources.
Reviewed by: gallatin, hselasky, cy, adrian, kristof
Differential Revision: https://reviews.freebsd.org/D19111
Remove the now obsolete vnet_state field. This greatly simplifies the
detection of VNET shutdown and avoids code duplication.
Discussed with: bz@
MFC after: 1 week
Sponsored by: Mellanox Technologies
Realistically, this cannot work. We don't allow the tun to be opened twice,
so it must be done via fd passing, fork, dup, some mechanism like these.
Applications demonstrably do not enforce strict ordering when they're
handing off tun devices, so the parent closing before the child will easily
leave the tun/tap device in a bad state where it can't be destroyed and a
confused user because they did nothing wrong.
Concede that we can't leave the tun/tap device in this kind of state because
of software not playing the TUNSIFPID game, but it is still good to find and
fix this kind of thing to keep ifconfig(8) up-to-date and help ensure good
discipline in tun handling.
MFC after: 3 days
Currently, if you do:
$ ifconfig tun0 create
$ ifconfig tun0 name wg0
$ ls -l /dev | egrep 'wg|tun'
You will see tun0, but no wg0. In fact, it's slightly more annoying to make
the association between the new name and the old name in order to open the
device (if it hadn't been opened during the rename).
Register an eventhandler for ifnet_arrival_events and catch interface
renames. We can determine if the ifnet is a tun easily enough from the
if_dname, which matches the cevsw.d_name from the associated tuntap_driver.
Some locking dance is required because renames don't require the device to
be opened, so it could go away in the middle of handling the ioctl, but as
soon as we've verified this isn't the case we can attempt to busy the tun
and either bail out if the tun device is dying, or we can proceed with the
rename.
We only create these aliases on a best-effort basis. Renaming a tun device
to "usbctl", which doesn't exist as an ifnet but does as a /dev, is clearly
not that disastrous, but we can't and won't create a /dev for that.
A future commit will create device aliases when a tuntap device is renamed
so that it's still easily found in /dev after the rename. Said mechanism
will want to keep the tun alive long enough to either realize that it's
about to go away or complete the alias creation, even if the alias is about
to get destroyed.
While we're introducing it, using it to prevent open devices from going away
makes plenty of sense and keeps the logic on waking up tun_destroy clean, so
we don't have multiple places trying to cv_broadcast unless it's still in
use elsewhere.
As of r347221 the iflib legacy interrupt mode setup assumes that drivers
perform both receive and transmit processing from the interrupt handler.
This assumption is invalid in the vmxnet3 driver, so introduce the
IFLIB_SINGLE_IRQ_RX_ONLY flag to make iflib avoid tx processing in the
interrupt handler.
PR: 239118
Reported and tested by: Juraj Lutter <otis@sk.freebsd.org>
Obtained from: marius
Reviewed by: gallatin
MFC after: 3 days
Differential Revision: https://reviews.freebsd.org/D21831
TLS 1.3 requires a few changes because 1.3 pretends to be 1.2
with a record type of application data. The "real" record type is
then included at the end of the user-supplied plaintext
data. This required adding a field to the mbuf_ext_pgs struct to
save the record type, and passing the real record type to the
sw_encrypt() ktls backend functions.
Reviewed by: jhb, hselasky
Sponsored by: Netflix
Differential Revision: D21801
properly nested and warns about recursive entrances. Unlike with locks,
there is nothing fundamentally wrong with such use, the intent of tracer
is to help to review complex epoch-protected code paths, and we mean the
network stack here.
Reviewed by: hselasky
Sponsored by: Netflix
Pull Request: https://reviews.freebsd.org/D21610
From Piotr:
r351152 introduced iflib_deregister() function calling
EVENTHANDLER_DEREGISTER() to unregister VLAN events. This patch removes
duplicate of EVENTHANDLER_DEREGISTER() calls placed in
iflib_device_deregister() as this function is now calling
iflib_deregister(). This is to avoid deregistering same event twice.
This patch also adds check in iflib_vlan_register() to prevent
registering VLAN while being in detach.
Patch co-authored by Krzysztof Galazka <krzysztof.galazka@intel.com>,
erj <erj@FreeBSD.org> and Jacob Keller <jacob.e.keller@intel.com>.
Signed-off-by: Piotr Pietruszewski <piotr.pietruszewski@intel.com>
Submitted by: Piotr Pietruszewski <piotr.pietruszewski@intel.com>
Reviewed by: gallatin@, erj@
MFC after: 3 days
Sponsored by: Intel Corporation
Differential Revision: https://reviews.freebsd.org/D21711
The ioctl(2) is intended to provide more details about the cause of
the down for the link.
Eventually we might define a comprehensive list of codes for the
situations. But interface also allows the driver to provide free-form
null-terminated ASCII string to provide arbitrary non-formalized
information. Sample implementation exists for mlx5(4), where the
string is fetched from firmware controlling the port.
Reviewed by: hselasky, rrs
Sponsored by: Mellanox Technologies
MFC after: 1 week
Differential revision: https://reviews.freebsd.org/D21527
Instead of throwing EEXIST, just succeed if the name isn't actually
changing. We don't need to trigger departure or any of that because there's
no change from consumers' perspective.
PR: 240539
Reviewed by: brooks
MFC after: 5 days
Differential Revision: https://reviews.freebsd.org/D21618
These commands show the route resolved for a specified destination, or
print out the entire routing table for a given address family (or all
families, if none is explicitly provided).
Discussed with: emaste
Differential Revision: https://reviews.freebsd.org/D21510
There are several mechanisms by which a vm_page reference is held,
preventing the page from being freed back to the page allocator. In
particular, holding the page's object lock is sufficient to prevent the
page from being freed; holding the busy lock or a wiring is sufficent as
well. These references are protected by the page lock, which must
therefore be acquired for many per-page operations. This results in
false sharing since the page locks are external to the vm_page
structures themselves and each lock protects multiple structures.
Transition to using an atomically updated per-page reference counter.
The object's reference is counted using a flag bit in the counter. A
second flag bit is used to atomically block new references via
pmap_extract_and_hold() while removing managed mappings of a page.
Thus, the reference count of a page is guaranteed not to increase if the
page is unbusied, unmapped, and the object's write lock is held. As
a consequence of this, the page lock no longer protects a page's
identity; operations which move pages between objects are now
synchronized solely by the objects' locks.
The vm_page_wire() and vm_page_unwire() KPIs are changed. The former
requires that either the object lock or the busy lock is held. The
latter no longer has a return value and may free the page if it releases
the last reference to that page. vm_page_unwire_noq() behaves the same
as before; the caller is responsible for checking its return value and
freeing or enqueuing the page as appropriate. vm_page_wire_mapped() is
introduced for use in pmap_extract_and_hold(). It fails if the page is
concurrently being unmapped, typically triggering a fallback to the
fault handler. vm_page_wire() no longer requires the page lock and
vm_page_unwire() now internally acquires the page lock when releasing
the last wiring of a page (since the page lock still protects a page's
queue state). In particular, synchronization details are no longer
leaked into the caller.
The change excises the page lock from several frequently executed code
paths. In particular, vm_object_terminate() no longer bounces between
page locks as it releases an object's pages, and direct I/O and
sendfile(SF_NOCACHE) completions no longer require the page lock. In
these latter cases we now get linear scalability in the common scenario
where different threads are operating on different files.
__FreeBSD_version is bumped. The DRM ports have been updated to
accomodate the KPI changes.
Reviewed by: jeff (earlier version)
Tested by: gallatin (earlier version), pho
Sponsored by: Netflix
Differential Revision: https://reviews.freebsd.org/D20486
When a vlan interface is created, its if_output is set directly to the
parent interface's if_output. This is fine in the normal case but has an
unfortunate consequence if you end up with a certain combination of vlan
and lagg interfaces.
Consider you have a lagg interface with a single laggport member. When
an interface is added to a lagg its if_output is set to
lagg_port_output, which blackholes traffic from the normal networking
stack but not certain frames from BPF (pseudo_AF_HDRCMPLT). If you now
create a vlan with the laggport member (not the lagg interface) as its
parent, its if_output is set to lagg_port_output as well. While this is
confusing conceptually and likely represents a misconfigured system, it
is not itself a problem. The problem arises when you then remove the
lagg interface. Doing this resets the if_output of the laggport member
back to its original state, but the vlan's if_output is left pointing to
lagg_port_output. This gives rise to the possibility that the system
will panic when e.g. bpf is used to send any frames on the vlan
interface.
Fix this by creating a new function, vlan_output, which simply wraps the
parent's current if_output. That way when the parent's if_output is
restored there is no stale usage of lagg_port_output.
Reviewed by: rstone
Differential Revision: D21209
KTLS adds support for in-kernel framing and encryption of Transport
Layer Security (1.0-1.2) data on TCP sockets. KTLS only supports
offload of TLS for transmitted data. Key negotation must still be
performed in userland. Once completed, transmit session keys for a
connection are provided to the kernel via a new TCP_TXTLS_ENABLE
socket option. All subsequent data transmitted on the socket is
placed into TLS frames and encrypted using the supplied keys.
Any data written to a KTLS-enabled socket via write(2), aio_write(2),
or sendfile(2) is assumed to be application data and is encoded in TLS
frames with an application data type. Individual records can be sent
with a custom type (e.g. handshake messages) via sendmsg(2) with a new
control message (TLS_SET_RECORD_TYPE) specifying the record type.
At present, rekeying is not supported though the in-kernel framework
should support rekeying.
KTLS makes use of the recently added unmapped mbufs to store TLS
frames in the socket buffer. Each TLS frame is described by a single
ext_pgs mbuf. The ext_pgs structure contains the header of the TLS
record (and trailer for encrypted records) as well as references to
the associated TLS session.
KTLS supports two primary methods of encrypting TLS frames: software
TLS and ifnet TLS.
Software TLS marks mbufs holding socket data as not ready via
M_NOTREADY similar to sendfile(2) when TLS framing information is
added to an unmapped mbuf in ktls_frame(). ktls_enqueue() is then
called to schedule TLS frames for encryption. In the case of
sendfile_iodone() calls ktls_enqueue() instead of pru_ready() leaving
the mbufs marked M_NOTREADY until encryption is completed. For other
writes (vn_sendfile when pages are available, write(2), etc.), the
PRUS_NOTREADY is set when invoking pru_send() along with invoking
ktls_enqueue().
A pool of worker threads (the "KTLS" kernel process) encrypts TLS
frames queued via ktls_enqueue(). Each TLS frame is temporarily
mapped using the direct map and passed to a software encryption
backend to perform the actual encryption.
(Note: The use of PHYS_TO_DMAP could be replaced with sf_bufs if
someone wished to make this work on architectures without a direct
map.)
KTLS supports pluggable software encryption backends. Internally,
Netflix uses proprietary pure-software backends. This commit includes
a simple backend in a new ktls_ocf.ko module that uses the kernel's
OpenCrypto framework to provide AES-GCM encryption of TLS frames. As
a result, software TLS is now a bit of a misnomer as it can make use
of hardware crypto accelerators.
Once software encryption has finished, the TLS frame mbufs are marked
ready via pru_ready(). At this point, the encrypted data appears as
regular payload to the TCP stack stored in unmapped mbufs.
ifnet TLS permits a NIC to offload the TLS encryption and TCP
segmentation. In this mode, a new send tag type (IF_SND_TAG_TYPE_TLS)
is allocated on the interface a socket is routed over and associated
with a TLS session. TLS records for a TLS session using ifnet TLS are
not marked M_NOTREADY but are passed down the stack unencrypted. The
ip_output_send() and ip6_output_send() helper functions that apply
send tags to outbound IP packets verify that the send tag of the TLS
record matches the outbound interface. If so, the packet is tagged
with the TLS send tag and sent to the interface. The NIC device
driver must recognize packets with the TLS send tag and schedule them
for TLS encryption and TCP segmentation. If the the outbound
interface does not match the interface in the TLS send tag, the packet
is dropped. In addition, a task is scheduled to refresh the TLS send
tag for the TLS session. If a new TLS send tag cannot be allocated,
the connection is dropped. If a new TLS send tag is allocated,
however, subsequent packets will be tagged with the correct TLS send
tag. (This latter case has been tested by configuring both ports of a
Chelsio T6 in a lagg and failing over from one port to another. As
the connections migrated to the new port, new TLS send tags were
allocated for the new port and connections resumed without being
dropped.)
ifnet TLS can be enabled and disabled on supported network interfaces
via new '[-]txtls[46]' options to ifconfig(8). ifnet TLS is supported
across both vlan devices and lagg interfaces using failover, lacp with
flowid enabled, or lacp with flowid enabled.
Applications may request the current KTLS mode of a connection via a
new TCP_TXTLS_MODE socket option. They can also use this socket
option to toggle between software and ifnet TLS modes.
In addition, a testing tool is available in tools/tools/switch_tls.
This is modeled on tcpdrop and uses similar syntax. However, instead
of dropping connections, -s is used to force KTLS connections to
switch to software TLS and -i is used to switch to ifnet TLS.
Various sysctls and counters are available under the kern.ipc.tls
sysctl node. The kern.ipc.tls.enable node must be set to true to
enable KTLS (it is off by default). The use of unmapped mbufs must
also be enabled via kern.ipc.mb_use_ext_pgs to enable KTLS.
KTLS is enabled via the KERN_TLS kernel option.
This patch is the culmination of years of work by several folks
including Scott Long and Randall Stewart for the original design and
implementation; Drew Gallatin for several optimizations including the
use of ext_pgs mbufs, the M_NOTREADY mechanism for TLS records
awaiting software encryption, and pluggable software crypto backends;
and John Baldwin for modifications to support hardware TLS offload.
Reviewed by: gallatin, hselasky, rrs
Obtained from: Netflix
Sponsored by: Netflix, Chelsio Communications
Differential Revision: https://reviews.freebsd.org/D21277
When tun/tap were merged, appropriate MODULE_VERSION should have been added
for things like modfind(2) to continue to do the right thing with the old
names.
Reported by: jhb
Commit message by Jake:
The iflib_register function exists to allocate and setup some common
structures used by both iflib_device_register and iflib_pseudo_register.
There is no associated cleanup function used to undo the steps taken in
this function.
Both iflib_device_deregister and iflib_pseudo_deregister have some of
the necessary steps scattered in their flow. However, most of the
necessary cleanup is not done during the error path of
iflib_device_register and iflib_pseudo_register.
Some examples of missed cleanup include:
the ifp pointer is not free'd during error cleanup
the STATE and CTX locks are not destroyed during error cleanup
the vlan event handlers are not removed during error cleanup
media added to the ifmedia structure is not removed
the kobject reference is never deleted
Additionally, when initializing the kobject class reference counter is
increased even though kobj_init already increases it. This results in
the class never being free'd again because the reference count would
never hit zero even after all driver instances are unloaded.
To aid in proper cleanup, implement an iflib_deregister function that
goes through the reverse steps taken by iflib_register.
Call this function during the error cleanup for iflib_device_register
and iflib_pseudo_register. Additionally call the function in the
iflib_device_deregister and iflib_pseudo_deregister functions near the
end of their flow. This helps reduce code duplication and ensures that
proper steps are taken to cleanup allocations and references in both the
regular and error cleanup flows.
Signed-off-by: Jacob Keller <jacob.e.keller@intel.com>
Submitted by: Jacob Keller <jacob.e.keller@intel.com>
Reviewed by: shurd@, erj@
MFC after: 3 days
Sponsored by: Intel Corporation
Differential Revision: https://reviews.freebsd.org/D21005
If a device has only 1 MSI-X interrupt available and does not support either
MSI or legacy interrupts, iflib_device_register() will fail, leak memory and
MSI resources, and the driver will not load. Worse, if another iflib-using
driver tries to unload afterwards, a kernel panic will occur because the
previous failed iflib driver loead did not properly call "taskqgroup_detach()"
during it's cleanup.
This patch is band-aid for this situation -- don't try allocating MSI or legacy
interrupts if a single MSI-X interrupt was allocated, but fail to load instead.
As well, during the cleanup, properly call taskqgroup_detach() on the admin
task to prevent panics when other iflib drivers unload.
This whole interrupt allocation process actually needs re-doing to properly
support devices with only a single MSI-X interrupt, devices that only support
MSI-X, non-PCI devices, and multiple non-MSIX interrupts, as well.
Signed-off-by: Eric Joyner <erj@freebsd.org>
Reviewed by: marius@
MFC after: 1 week
Sponsored by: Intel Corporation
Differential Revision: https://reviews.freebsd.org/D20747
Commit message from Jake:
In iflib_register, the context is initialized as a kobject using the
device driver's "driver" kobject class. As part of this, the function
mistakenly increments the ref counter.
The ref counter is incremented twice, once in the code directly, and
once again by kobj_class_compile. However, there is no associated
decrement in the detach path. Because of this, the ref counter will
never go back down to zero, and thus the kobject method table will never
be released.
Remove this unnecessary reference count increment.
Signed-off-by: Jacob Keller <jacob.e.keller@intel.com>
Submitted by: Jacob Keller <jacob.e.keller@intel.com>
Reviewed by: jhb@, erj@
MFC after: 3 days
Sponsored by: Intel Corporation
Differential Revision: https://reviews.freebsd.org/D21125
an updated rack depend on having access to the new
ratelimit api in this commit.
Sponsored by: Netflix Inc.
Differential Revision: https://reviews.freebsd.org/D20953
The magic number is a 32-bit quantity; use uint32_t to match hton's
return type and avoid sending zeros (upper 32 bits) on big-endian
architectures.
PR: 184141
MFC after: 1 week
Sponsored by: The FreeBSD Foundation
Commit text by Jake:
If a driver's IFDI_ATTACH_PRE function fails, the iflib_device_register
function will free the ctx pointer. However, it does not reset the
device softc pointer to NULL.
This will result in memory corruption as a future access to the now
invalid pointer will corrupt memory that is later allocated on top of
the same memory location.
The iflib_device_deregister function correctly resets the softc pointer
by using device_set_softc().
This clears up the invalid dangling pointer and prevents memory
corruption that could lead to a panic or undefined behavior if the
device's driver failed to attach.
Signed-off-by: Jacob Keller <jacob.e.keller@intel.com>
Submitted by: Jacob Keller <jacob.e.keller@intel.com>
Reviewed by: erj@, gallatin@
MFC after: 1 week
Sponsored by: Intel Corporation
Differential Revision: https://reviews.freebsd.org/D21003
Accept an IEEE Extended Unique Identifier (EUI-64) from the command
line for each NVMe namespace. If one isn't provided, it will create one
based on the CRC16 of:
- the FreeBSD IEEE OUI
- PCI bus, device/slot, function values
- Namespace ID
Reviewed by: imp, araujo, jhb, rgrimes
Approved by: imp (mentor), jhb (maintainer)
MFC after: 2 weeks
Differential Revision: https://reviews.freebsd.org/D19905
The hold_count and wire_count fields of struct vm_page are separate
reference counters with similar semantics. The remaining essential
differences are that holds are not counted as a reference with respect
to LRU, and holds have an implicit free-on-last unhold semantic whereas
vm_page_unwire() callers must explicitly determine whether to free the
page once the last reference to the page is released.
This change removes the KPIs which directly manipulate hold_count.
Functions such as vm_fault_quick_hold_pages() now return wired pages
instead. Since r328977 the overhead of maintaining LRU for wired pages
is lower, and in many cases vm_fault_quick_hold_pages() callers would
swap holds for wirings on the returned pages anyway, so with this change
we remove a number of page lock acquisitions.
No functional change is intended. __FreeBSD_version is bumped.
Reviewed by: alc, kib
Discussed with: jeff
Discussed with: jhb, np (cxgbe)
Tested by: pho (previous version)
Sponsored by: Netflix
Differential Revision: https://reviews.freebsd.org/D19247
Enable IFCAP_NOMAP for a vlan interface if it is supported by the
underlying trunk device.
Reviewed by: gallatin, hselasky, rrs
Sponsored by: Netflix
Differential Revision: https://reviews.freebsd.org/D20616
Unmapped mbufs allow sendfile to carry multiple pages of data in a
single mbuf, without mapping those pages. It is a requirement for
Netflix's in-kernel TLS, and provides a 5-10% CPU savings on heavy web
serving workloads when used by sendfile, due to effectively
compressing socket buffers by an order of magnitude, and hence
reducing cache misses.
For this new external mbuf buffer type (EXT_PGS), the ext_buf pointer
now points to a struct mbuf_ext_pgs structure instead of a data
buffer. This structure contains an array of physical addresses (this
reduces cache misses compared to an earlier version that stored an
array of vm_page_t pointers). It also stores additional fields needed
for in-kernel TLS such as the TLS header and trailer data that are
currently unused. To more easily detect these mbufs, the M_NOMAP flag
is set in m_flags in addition to M_EXT.
Various functions like m_copydata() have been updated to safely access
packet contents (using uiomove_fromphys()), to make things like BPF
safe.
NIC drivers advertise support for unmapped mbufs on transmit via a new
IFCAP_NOMAP capability. This capability can be toggled via the new
'nomap' and '-nomap' ifconfig(8) commands. For NIC drivers that only
transmit packet contents via DMA and use bus_dma, adding the
capability to if_capabilities and if_capenable should be all that is
required.
If a NIC does not support unmapped mbufs, they are converted to a
chain of mapped mbufs (using sf_bufs to provide the mapping) in
ip_output or ip6_output. If an unmapped mbuf requires software
checksums, it is also converted to a chain of mapped mbufs before
computing the checksum.
Submitted by: gallatin (earlier version)
Reviewed by: gallatin, hselasky, rrs
Discussed with: ae, kp (firewalls)
Relnotes: yes
Sponsored by: Netflix
Differential Revision: https://reviews.freebsd.org/D20616
network interface.
This particularly manifests itself when an INP has multicast options
attached during a network interface detach. Then the IPv4 and IPv6
leave group call which results from freeing the multicast address, may
access a freed ifnet structure. These are the steps to reproduce:
service mdnsd onestart # installed from ports
ifconfig epair create
ifconfig epair0a 0/24 up
ifconfig epair0a destroy
Tested by: pho @
MFC after: 1 week
Sponsored by: Mellanox Technologies
- Remove desc_used, which is only ever written to.
- Remove a dead store to reclaimed.
- Don't recycle avail.
- Sort variables according to style(9).
These changes will make a subsequent commit easier to read.
o In iflib_tx_credits_update(), don't bother checking whether the
ift_txd_credits_update method pointer is NULL; _iflib_pre_assert()
asserts upfront that this method has been assigned and functions
like iflib_{fast_intr_rxtx,netmap_timer_adjust,txq_can_drain}()
and _task_fn_tx() were already unconditionally relying on the
method being callable.
Fixes panic when loading ipfw.ko and if_epair.ko built with modern compiler.
Similar to arm64 and riscv, when using a modern compiler (!gcc4.2), code
generated tries to access data in the wrong location, causing kernel panic
(data storage interrupt trap) when loading if_epair and ipfw.
Issue was reproduced with kernel/module compiled using gcc8 and clang8. It
affects both ELFv1 and ELFv2 ABI environments.
PR: 232387
Submitted by: alfredo.junior_eldorado.org.br
Reported by: Mark Millard
Reviewed by: jhibbits
Differential Revision: https://reviews.freebsd.org/D20461
instead of a linear array.
The multicast memberships for the inpcb structure are protected by a
non-sleepable lock, INP_WLOCK(), which needs to be dropped when
calling the underlying possibly sleeping if_ioctl() method. When using
a linear array to keep track of multicast memberships, the computed
memory location of the multicast filter may suddenly change, due to
concurrent insertion or removal of elements in the linear array. This
in turn leads to various invalid memory access issues and kernel
panics.
To avoid this problem, put all multicast memberships on a STAILQ based
list. Then the memory location of the IPv4 and IPv6 multicast filters
become fixed during their lifetime and use after free and memory leak
issues are easier to track, for example by: vmstat -m | grep multi
All list manipulation has been factored into inline functions
including some macros, to easily allow for a future hash-list
implementation, if needed.
This patch has been tested by pho@ .
Differential Revision: https://reviews.freebsd.org/D20080
Reviewed by: markj @
MFC after: 1 week
Sponsored by: Mellanox Technologies
at runtime. This change removes a dependency on a barrel shifter pass
before branch resolution, while reducing the instruction stream size
by 9 bytes on amd64.
MFC after: 3 days
structs with placeholders (in the latter case, IFLIB_MAX_TX_BYTES
etc. are also only ever used for these write-only members if at all,
so both these macros and members can just go). Using these spares
may render it possible to merge certain iflib(9) fixes to stable/12.
Otherwise, changes extending struct if_irq or struct if_shared_ctx
in any way would break KBI as instances of these are allocated by
the driver front-ends (by contrast, struct if_pkt_info as well as
struct if_softc_ctx instances are provided by iflib(9) and, thus,
may grow at least at the end without breaking KBI).
- Make the pvi_name in struct pci_vendor_info const char * as device
identifiers in hardware lookup tables aren't to be expected to ever
change at runtime.
- Similarly, make the pci_vendor_info_t of struct if_shared_ctx which
is used to point to the struct pci_vendor_info arrays provided by
the driver front-ends const.
- Remove the ETH_ADDR_LEN macro from iflib.h; this was duplicating
ETHER_ADDR_LEN of <net/ethernet.h> with iflib(9) actually only
consuming the latter macro.
- Make the name argument of iflib_io_tqg_attach(9) const, matching
the taskqgroup_attach_cpu(9) this function wraps as well as e. g.
iflib_config_gtask_init(9).
- Remove the orphaned iflib_qset_lock_get() prototype.
- Remove some extraneous empty lines.
Its result is only used to determine whether to perform further
INVARIANTS-only checks. Remove a stale comment while here.
Submitted by: Sebastian Huber <sebastian.huber@embedded-brains.de>
MFC after: 1 week
From Jake:
Vendor drivers that exist out-of-tree generally should return
BUS_PROBE_VENDOR from their device probe functions. This helps ensure
that a vendor replacement driver will supersede the in-kernel driver for
a given device.
Currently, if a vendor wants to implement a driver based on iflib, it
will always report BUS_PROBE_DEFAULT.
Add a wrapper function, iflib_device_probe_vendor() which can be used in
place of iflib_device_probe(). This function will just return
BUS_PROBE_VENDOR whenever iflib_device_probe() would return
BUS_PROBE_DEFAULT.
While vendor drivers can already implement such a wrapper themselves,
providing it in the iflib.h header makes it easier for the vendor driver
to do the right thing.
Submitted by: Jacob Keller <jacob.e.keller@intel.com>
Reviewed by: erj@, gallatin@, marius@
MFC after: 1 week
Sponsored by: Intel Corporation
Differential Revision: https://reviews.freebsd.org/D20221
There were two remaining "gaps" in auditing local bridge traffic with
bpf(4):
Locally originated outbound traffic from a member interface is invisible to
the bridge's bpf(4) interface. Inbound traffic locally destined to a member
interface is invisible to the member's bpf(4) interface -- this traffic has
no chance after bridge_input to otherwise pass it over, and it wasn't
originally received on this interface.
I call these "gaps" because they don't affect conventional bridge setups.
Alas, being able to establish an audit trail of all locally destined traffic
for setups that can function like this is useful in some scenarios.
Reviewed by: kp
MFC after: 1 week
Differential Revision: https://reviews.freebsd.org/D19757
It appeared that using NET_EPOCH_WAIT() while holding global BPF lock
can lead to another panic:
spin lock 0xfffff800183c9840 (turnstile lock) held by 0xfffff80018e2c5a0 (tid 100325) too long
panic: spin lock held too long
...
#0 sched_switch (td=0xfffff80018e2c5a0, newtd=0xfffff8000389e000, flags=<optimized out>) at /usr/src/sys/kern/sched_ule.c:2133
#1 0xffffffff80bf9912 in mi_switch (flags=256, newtd=0x0) at /usr/src/sys/kern/kern_synch.c:439
#2 0xffffffff80c21db7 in sched_bind (td=<optimized out>, cpu=<optimized out>) at /usr/src/sys/kern/sched_ule.c:2704
#3 0xffffffff80c34c33 in epoch_block_handler_preempt (global=<optimized out>, cr=0xfffffe00005a1a00, arg=<optimized out>)
at /usr/src/sys/kern/subr_epoch.c:394
#4 0xffffffff803c741b in epoch_block (global=<optimized out>, cr=<optimized out>, cb=<optimized out>, ct=<optimized out>)
at /usr/src/sys/contrib/ck/src/ck_epoch.c:416
#5 ck_epoch_synchronize_wait (global=0xfffff8000380cd80, cb=<optimized out>, ct=<optimized out>) at /usr/src/sys/contrib/ck/src/ck_epoch.c:465
#6 0xffffffff80c3475e in epoch_wait_preempt (epoch=0xfffff8000380cd80) at /usr/src/sys/kern/subr_epoch.c:513
#7 0xffffffff80ce970b in bpf_detachd_locked (d=0xfffff801d309cc00, detached_ifp=<optimized out>) at /usr/src/sys/net/bpf.c:856
#8 0xffffffff80ced166 in bpf_detachd (d=<optimized out>) at /usr/src/sys/net/bpf.c:836
#9 bpf_dtor (data=0xfffff801d309cc00) at /usr/src/sys/net/bpf.c:914
To fix this add the check to the catchpacket() that BPF descriptor was
not detached just before we acquired BPFD_LOCK().
Reported by: slavash
Tested by: slavash
MFC after: 1 week
bpf_mtap() can invoke catchpacket() for already detached descriptor.
And this can lead to NULL pointer dereference, since bd_bif pointer
was reset to NULL in bpf_detachd_locked(). To avoid this, use
NET_EPOCH_WAIT() when descriptor is removed from interface's descriptors
list. After the wait it is safe to modify descriptor's content.
Submitted by: kib
Reported by: slavash
MFC after: 1 week
- Perform ifp mismatch checks (to determine if a send tag is allocated
for a different ifp than the one the packet is being output on), in
ip_output() and ip6_output(). This avoids sending packets with send
tags to ifnet drivers that don't support send tags.
Since we are now checking for ifp mismatches before invoking
if_output, we can now try to allocate a new tag before invoking
if_output sending the original packet on the new tag if allocation
succeeds.
To avoid code duplication for the fragment and unfragmented cases,
add ip_output_send() and ip6_output_send() as wrappers around
if_output and nd6_output_ifp, respectively. All of the logic for
setting send tags and dealing with send tag-related errors is done
in these wrapper functions.
For pseudo interfaces that wrap other network interfaces (vlan and
lagg), wrapper send tags are now allocated so that ip*_output see
the wrapper ifp as the ifp in the send tag. The if_transmit
routines rewrite the send tags after performing an ifp mismatch
check. If an ifp mismatch is detected, the transmit routines fail
with EAGAIN.
- To provide clearer life cycle management of send tags, especially
in the presence of vlan and lagg wrapper tags, add a reference count
to send tags managed via m_snd_tag_ref() and m_snd_tag_rele().
Provide a helper function (m_snd_tag_init()) for use by drivers
supporting send tags. m_snd_tag_init() takes care of the if_ref
on the ifp meaning that code alloating send tags via if_snd_tag_alloc
no longer has to manage that manually. Similarly, m_snd_tag_rele
drops the refcount on the ifp after invoking if_snd_tag_free when
the last reference to a send tag is dropped.
This also closes use after free races if there are pending packets in
driver tx rings after the socket is closed (e.g. from tcpdrop).
In order for m_free to work reliably, add a new CSUM_SND_TAG flag in
csum_flags to indicate 'snd_tag' is set (rather than 'rcvif').
Drivers now also check this flag instead of checking snd_tag against
NULL. This avoids false positive matches when a forwarded packet
has a non-NULL rcvif that was treated as a send tag.
- cxgbe was relying on snd_tag_free being called when the inp was
detached so that it could kick the firmware to flush any pending
work on the flow. This is because the driver doesn't require ACK
messages from the firmware for every request, but instead does a
kind of manual interrupt coalescing by only setting a flag to
request a completion on a subset of requests. If all of the
in-flight requests don't have the flag when the tag is detached from
the inp, the flow might never return the credits. The current
snd_tag_free command issues a flush command to force the credits to
return. However, the credit return is what also frees the mbufs,
and since those mbufs now hold references on the tag, this meant
that snd_tag_free would never be called.
To fix, explicitly drop the mbuf's reference on the snd tag when the
mbuf is queued in the firmware work queue. This means that once the
inp's reference on the tag goes away and all in-flight mbufs have
been queued to the firmware, tag's refcount will drop to zero and
snd_tag_free will kick in and send the flush request. Note that we
need to avoid doing this in the middle of ethofld_tx(), so the
driver grabs a temporary reference on the tag around that loop to
defer the free to the end of the function in case it sends the last
mbuf to the queue after the inp has dropped its reference on the
tag.
- mlx5 preallocates send tags and was using the ifp pointer even when
the send tag wasn't in use. Explicitly use the ifp from other data
structures instead.
- Sprinkle some assertions in various places to assert that received
packets don't have a send tag, and that other places that overwrite
rcvif (e.g. 802.11 transmit) don't clobber a send tag pointer.
Reviewed by: gallatin, hselasky, rgrimes, ae
Sponsored by: Netflix
Differential Revision: https://reviews.freebsd.org/D20117
Currently rinit1() and its IPv6 counterpart
nd6_prefix_onlink_rtrequest() uses dummy null_sdl gateway address
during route insertion and change it afterwards. This behaviour
brings complications to the routing stack and the users of its
upcoming notification system.
This change fixes both rinit1() and nd6_prefix_onlink_rtrequest()
by filling in proper gateway in the beginning. It does not change any
of the userland notifications as in both cases, they happen after
the insertion and fixup process (rt_newaddrmsg_fib() and nd6_rtmsg()).
MFC after: 2 weeks
Differential Revision: https://reviews.freebsd.org/D20328
This allows replacing "sys/eventfilter.h" includes with "sys/_eventfilter.h"
in other header files (e.g., sys/{bus,conf,cpu}.h) and reduces header
pollution substantially.
EVENTHANDLER_DECLARE and EVENTHANDLER_LIST_DECLAREs were moved out of .c
files into appropriate headers (e.g., sys/proc.h, powernv/opal.h).
As a side effect of reduced header pollution, many .c files and headers no
longer contain needed definitions. The remainder of the patch addresses
adding appropriate includes to fix those files.
LOCK_DEBUG and LOCK_FILE_LINE_ARG are moved to sys/_lock.h, as required by
sys/mutex.h since r326106 (but silently protected by header pollution prior
to this change).
No functional change (intended). Of course, any out of tree modules that
relied on header pollution for sys/eventhandler.h, sys/lock.h, or
sys/mutex.h inclusion need to be fixed. __FreeBSD_version has been bumped.
Currently such routes are added with a link-level IFA, which is
plain wrong. Only after the insertion they get fixed by the special
link_rtrequest() ifa handler. This behaviour complicates routing code
and makes ifa selection more complex.
Streamline this process by explicitly moving link_rtrequest() logic
to the pre-insertion rt_getifa_fib() ifa selector. Avoid calling all
this logic in the loopback route case by explicitly specifying
proper rt_ifa inside the ifa_maintain_loopback_route().§
MFC after: 2 weeks
Differential Revision: https://reviews.freebsd.org/D20076
r346670 added an sx to close a race between the ifioctl handler and
interface destruction. Unfortunately, it clears if_softc immediately after
the interface is closed, but before if_detach has been invoked.
Any time before detachment, an interface that's part of a bridge may still
receive traffic that's pushed through tunstart/tunstart_l2 and promptly
lead to a panic because if_softc is now NULL.
Fix it by deferring the clearing of if_softc until after the interface has
detached and thus been removed from the bridge. if_softc still gets cleared
in case another thread has already entered the ioctl handler before it's
replaced with ifdead_ioctl.
Reported by: markj
MFC after: 3 days
Release BPF_LOCK() before invoking if_output() and if_input().
Also enter epoch section before releasing lock, this should prevent
access to ifnet that may be freed on interface detach.
Reported by: markj
On high packets rate the contention on rwlock in bpf_*tap*() functions
can lead to packets dropping. To avoid this, migrate this code to use
epoch(9) KPI and ConcurrencyKit's lists.
* all lists changed to use CK_LIST;
* reference counting added to bpf_if and bpf_d;
* now bpf_if references ifnet and releases this reference on destroy;
* each bpf_d descriptor references bpf_if when it is attached;
* new struct bpf_program_buffer introduced to keep BPF filter programs;
* bpf_program_buffer, bpf_d and bpf_if structures are freed by
epoch_call();
* bpf_freelist and ifnet_departure event are no longer needed, thus
both are removed;
Reviewed by: melifaro
Sponsored by: Yandex LLC
Differential Revision: https://reviews.freebsd.org/D20224
No functional change.
tun_flags of the tuntap_driver was renamed to ident_flags to reflect the
fact that it's a subset of the tun_flags that identifies a tuntap device.
This maps more easily (visually) to the TUN_DRIVER_IDENT_MASK that masks off
the bits of tun_flags that are applicable to tuntap driver ident. This is a
purely cosmetic change.
From Jake:
A user may set a sysctl to override the default number of Tx or Rx
descriptors. However, certain calculations in the iflib core expect the
number of descriptors to be a power of 2.
Update _iflib_assert to verify that all of the shared context parameters
for the number of descriptors are powers of 2.
Modify iflib_reset_qvalues to check that the provided isc_nrxd value is
a power of 2. If it's not, print a warning message and then use the
default value.
An alternative might be to try rounding the number down instead.
However, this creates problems in case the rounded down value is below
the minimum value that the driver would support.
Submitted by: Jacob Keller <jacob.e.keller@intel.com>
Reviewed by: marius@
MFC after: 1 week
Sponsored by: Intel Corporation
Differential Revision: https://reviews.freebsd.org/D19880
lookup KPI in ip_output() like it is already used in ip_forward().
However, when there is no PCB provided we can use fast KPI, gaining
performance advantage.
Typical case when ip_output() is called without a PCB pointer is a
sendto(2) on a not connected UDP socket. In practice DNS servers do
this.
Reviewed by: melifaro
Differential Revision: https://reviews.freebsd.org/D19804
tun(4) and tap(4) share the same general management interface and have a lot
in common. Bugs exist in tap(4) that have been fixed in tun(4), and
vice-versa. Let's reduce the maintenance requirements by merging them
together and using flags to differentiate between the three interface types
(tun, tap, vmnet).
This fixes a couple of tap(4)/vmnet(4) issues right out of the gate:
- tap devices may no longer be destroyed while they're open [0]
- VIMAGE issues already addressed in tun by kp
[0] emaste had removed an easy-panic-button in r240938 due to devdrn
blocking. A naive glance over this leads me to believe that this isn't quite
complete -- destroy_devl will only block while executing d_* functions, but
doesn't block the device from being destroyed while a process has it open.
The latter is the intent of the condvar in tun, so this is "fixed" (for
certain definitions of the word -- it wasn't really broken in tap, it just
wasn't quite ideal).
ifconfig(8) also grew the ability to map an interface name to a kld, so
that `ifconfig {tun,tap}0` can continue to autoload the correct module, and
`ifconfig vmnet0 create` will now autoload the correct module. This is a
low overhead addition.
(MFC commentary)
This may get MFC'd if many bugs in tun(4)/tap(4) are discovered after this,
and how critical they are. Changes after this are likely easily MFC'd
without taking this merge, but the merge will be easier.
I have no plans to do this MFC as of now.
Reviewed by: bcr (manpages), tuexen (testing, syzkaller/packetdrill)
Input also from: melifaro
Relnotes: yes
Differential Revision: https://reviews.freebsd.org/D20044
MSI. Unlike as with iflib_fast_intr_ctx(), the former will also enqueue
_task_fn_tx() in addition to _task_fn_rx() if appropriate, bringing TCP
TX throughput of EM-class devices on par with the MSI-X case and, thus,
close to wirespeed/pre-iflib(4) times again. [1]
Note that independently of the interrupt type, the UDP performance with
these MACs still is abysmal and nowhere near to where it was before the
conversion of em(4) to iflib(4).
o In iflib_init_locked(), announce which free list failed to set up.
o In _task_fn_tx() when running netmap(4), issue ifdi_intr_enable instead
of the ifdi_tx_queue_intr_enable method in case of a "legacy" interrupt
as the latter is valid with MSI-X only.
o Instead of adding the missing - and apparently convoluted enough that a
DBG_COUNTER_INC was put into a wrong spot in _task_fn_rx() - checks for
ifdi_{r,t}x_queue_intr_enable being available in the MSI-X case also to
iflib_fast_intr_rxtx(), factor these out to iflib_device_register() and
make the checks fail gracefully rather than panic. This avoids invoking
the checks at runtime over and over again in iflib_fast_intr_rxtx() and
_task_fn_{r,t}x() - even if it's just in case of INVARIANTS - and makes
these functions more readable.
o In iflib_rx_structures_setup(), only initialize LRO resources if device
and driver have LRO capability in order to not waste memory. Also, free
the LRO resources again if setting them up fails for one of the queues.
However, don't bother invoking iflib_rx_sds_free() in that case because
iflib_rx_structures_setup() doesn't call iflib_rxsd_alloc() either (and
iflib_{device,pseudo}_register() will issue iflib_rx_sds_free() in case
of failure via iflib_rx_structures_free(), but there definitely is some
asymmetry left to be fixed, though).
o Similarly, free LRO resources again in iflib_rx_structures_free().
o In iflib_irq_set_affinity(), handle get_core_offset() errors gracefully
instead of panicing (but only in case of INVARIANTS). This is a follow-
up to r344132, as such driver bugs shouldn't be fatal.
o Likewise, handle unknown iflib_intr_type_t in iflib_irq_alloc_generic()
gracefully, too.
o Bring yet more sanity to iflib_msix_init():
- If the device doesn't provide enough MSI-X vectors or not all vectors
can be allocate so the expected number of queues in addition to admin
interrupts can't be supported, try MSI next (and then INTx) as proper
MSI-X vector distribution can't be assured in such cases. In essence,
this change brings r254008 forward to iflib(4). Also, this is the fix
alluded to in the commit message of r343934.
- If the MSI-X allocation has failed, don't prematurely announce MSI is
going to be used as the latter in fact may not be available either.
- When falling back to MSI, only release the MSI-X table resource again
if it was allocated in iflib_msix_init(), i. e. isn't supplied by the
driver, in the first place.
o In mp_ndesc_handler(), handle unknown type arguments gracefully, too.
PR: 235031 (likely) [1]
Reviewed by: shurd
Differential Revision: https://reviews.freebsd.org/D20175
- Remove the only ever written to ift_db_mtx_name member of struct iflib_txq.
- Remove the unused or only ever written to ifr_size, ifr_cq_pidx, ifr_cq_gen
and ifr_lro_enabled members of struct iflib_rxq.
- Consistently spell DMA, RX and TX uppercase in comments, messages etc.
instead of mixing with some lowercase variants.
- Consistently use if_t instead of a mix of if_t and struct ifnet pointers.
- Bring the function comments of _iflib_fl_refill(), iflib_rx_sds_free() and
iflib_fl_setup() in line with reality.
- Judging problem reports, people are wondering what on earth messages like:
"TX(0) desc avail = 1024, pidx = 0"
are trying to indicate. Thus, extend this string to be more like that of
non-iflib(4) Ethernet MAC drivers, notifying about a watchdog timeout due
to which the interface will be reset.
- Take advantage of the M_HAS_VLANTAG macro.
- Use false/true rather than FALSE/TRUE for variables of type bool.
- Use FALLTHROUGH as advocated by style(9).
This change creates an array of port maps indexed by numa domain
for lacp port selection. If we have lacp interfaces in more than
one domain, then we select the egress port by indexing into the
numa port maps and picking a port on the appropriate numa domain.
This is behavior is controlled by the new ifconfig use_numa flag
and net.link.lagg.use_numa sysctl/tunable (both modeled after the
existing use_flowid), which default to enabled.
Reviewed by: bz, hselasky, markj (and scottl, earlier version)
Sponsored by: Netflix
Differential Revision: https://reviews.freebsd.org/D20060
It's atypical, but not invalid, for a driver to pass no capabilities.
Submitted by: Gerald Aryeetey <aryeeteygerald_rogers.com>
Reviewed by: shurd
MFC after: 1 week
Sponsored by: The FreeBSD Foundation
Differential Revision: https://reviews.freebsd.org/D20142
By default, cores are now assigned to queues in a sequential
manner rather than all NICs starting at the first core. On a four-core
system with two NICs each using two queue pairs, the nic:queue -> core
mapping has changed from this:
0:0 -> 0, 0:1 -> 1
1:0 -> 0, 1:1 -> 1
To this:
0:0 -> 0, 0:1 -> 1
1:0 -> 2, 1:1 -> 3
Additionally, a device can now be configured to use separate cores for TX
and RX queues.
Two new tunables have been added, dev.X.Y.iflib.separate_txrx and
dev.X.Y.iflib.core_offset. If core_offset is set, the NIC is not part
of the auto-assigned sequence.
Reviewed by: marius
MFC after: 2 weeks
Sponsored by: Limelight Networks
Differential Revision: https://reviews.freebsd.org/D20029
Otherwise tap(4) can be loaded by loader despite being compiled into the
kernel, causing a panic as things try to double-initialize.
PR: 220867
MFC after: 3 days
Previously, a pid check was used to prevent open of the tun(4); this works,
but may not make the most sense as we don't prevent the owner process from
opening the tun device multiple times.
The potential race described near tun_pid should not be an issue: if a
tun(4) is to be handed off, its fd has to have been sent via control message
or some other mechanism that duplicates the fd to the receiving process so
that it may set the pid. Otherwise, the pid gets cleared when the original
process closes it and you have no effective handoff mechanism.
Close up another potential issue with handing a tun(4) off by not clobbering
state if the closer isn't the controller anymore. If we want some state to
be cleared, we should do that a little more surgically.
Additionally, nothing prevents a dying tun(4) from being "reopened" in the
middle of tun_destroy as soon as the mutex is unlocked, quickly leading to a
bad time. Return EBUSY if we're marked for destruction, as well, and the
consumer will need to deal with it. The associated character device will be
destroyed in short order.
MFC after: 2 weeks
Differential Revision: https://reviews.freebsd.org/D20033
It seems that there should be a better way to handle this, but this seems to
be the more common approach and it should likely get replaced in all of the
places it happens... Basically, thread 1 is in the process of destroying the
tun/tap while thread 2 is executing one of the ioctls that requires the
tun/tap mutex and the mutex is destroyed before the ioctl handler can
acquire it.
This is only one of the races described/found in PR 233955.
PR: 233955
Reviewed by: ae
MFC after: 2 weeks
Differential Revision: https://reviews.freebsd.org/D20027
As with mlx5en, the idea is to drop unwanted traffic as early
in receive as possible, before mbufs are allocated and anything
is passed up the stack. This can save considerable CPU time
when a machine is under a flooding style DOS attack.
The major change here is to remove the unneeded abstraction where
callers of rxd_frag_to_sd() get back a pointer to the mbuf ring, and
are responsible for NULL'ing that mbuf themselves. Now this happens
directly in rxd_frag_to_sd(), and it returns an mbuf. This allows us
to use the decision (and potentially mbuf) returned by the pfil
hooks. The driver can now recycle mbufs to avoid re-allocation when
packets are dropped.
Reviewed by: marius (shurd and erj also provided feedback)
Sponsored by: Netflix
Differential Revision: https://reviews.freebsd.org/D19645
This GRE-in-UDP encapsulation allows the UDP source port field to be
used as an entropy field for load-balancing of GRE traffic in transit
networks. Also most of multiqueue network cards are able distribute
incoming UDP datagrams to different NIC queues, while very little are
able do this for GRE packets.
When an administrator enables UDP encapsulation with command
`ifconfig gre0 udpencap`, the driver creates kernel socket, that binds
to tunnel source address and after udp_set_kernel_tunneling() starts
receiving of all UDP packets destined to 4754 port. Each kernel socket
maintains list of tunnels with different destination addresses. Thus
when several tunnels use the same source address, they all handled by
single socket. The IP[V6]_BINDANY socket option is used to be able bind
socket to source address even if it is not yet available in the system.
This may happen on system boot, when gre(4) interface is created before
source address become available. The encapsulation and sending of packets
is done directly from gre(4) into ip[6]_output() without using sockets.
Reviewed by: eugen
MFC after: 1 month
Relnotes: yes
Differential Revision: https://reviews.freebsd.org/D19921
tun destruction will not continue until TUN_OPEN is cleared. There are brief
moments in tunclose where the mutex is dropped and we've already cleared
TUN_OPEN, so tun_destroy would be able to proceed while we're in the middle
of cleaning up the tun still. tun_destroy should be blocked until these
parts (address/route purges, mostly) are complete.
PR: 233955
MFC after: 2 weeks
This commit adds new if_alloc_domain() and if_alloc_dev() methods to
allocate ifnets. When called with a domain on a NUMA machine,
ifalloc_domain() will record the NUMA domain in the ifnet, and it will
allocate the ifnet struct from memory which is local to that NUMA
node. Similarly, if_alloc_dev() is a wrapper for if_alloc_domain
which uses a driver supplied device_t to call ifalloc_domain() with
the appropriate domain.
Note that the new if_numa_domain field fits in an alignment pad in
struct ifnet, and so does not alter the size of the structure.
Reviewed by: glebius, kib, markj
Sponsored by: Netflix
Differential Revision: https://reviews.freebsd.org/D19930
Give devices that need a MAC a 16-bit allocation out of the FreeBSD
Foundation OUI range. Change the name ether_fakeaddr to ether_gen_addr now
that we're dealing real MAC addresses with a real OUI rather than random
locally-administered addresses.
Reviewed by: bz, rgrimes
Differential Revision: https://reviews.freebsd.org/D19587