of needed interface when many gif interfaces are present.
Remove rmlock from gif_softc, use epoch(9) and CK_LIST instead.
Move more AF-related code into AF-related locations.
Use hash table to speedup lookup of needed softc. Interfaces
with GIF_IGNORE_SOURCE flag are stored in plain CK_LIST.
Sysctl net.link.gif.parallel_tunnels is removed. The removal was planed
16 years ago, and actually it could work only for outbound direction.
Each protocol, that can be handled by if_gif(4) interface is registered
by separate encap handler, this helps avoid invoking the handler
for unrelated protocols (GRE, PIM, etc.).
This change allows dramatically improve performance when many gif(4)
interfaces are used.
Sponsored by: Yandex LLC
Currently it has several disadvantages:
- it uses single mutex to protect internal structures. It is used by
data- and control- path, thus there are no parallelism at all.
- it uses single list to keep encap handlers for both INET and INET6
families.
- struct encaptab keeps unneeded information (src, dst, masks, protosw),
that isn't used by code in the source tree.
- matches are prioritized and when many tunneling interfaces are
registered, encapcheck handler of each interface is invoked for each
packet. The search takes O(n) for n interfaces. All this work is done
with exclusive lock held.
What this patch includes:
- the datapath is converted to be lockless using epoch(9) KPI.
- struct encaptab now linked using CK_LIST.
- all unused fields removed from struct encaptab. Several new fields
addedr: min_length is the minimum packet length, that encapsulation
handler expects to see; exact_match is maximum number of bits, that
can return an encapsulation handler, when it wants to consume a packet.
- IPv6 and IPv4 handlers are stored in separate lists;
- added new "encap_lookup_t" method, that will be used later. It is
targeted to speedup lookup of needed interface, when gif(4)/gre(4) have
many interfaces.
- the need to use protosw structure is eliminated. The only pr_input
method was used from this structure, so I don't see the need to keep
using it.
- encap_input_t method changed to avoid using mbuf tags to store softc
pointer. Now it is passed directly trough encap_input_t method.
encap_getarg() funtions is removed.
- all sockaddr structures and code that uses them removed. We don't have
any code in the tree that uses them. All consumers use encap_attach_func()
method, that relies on invoking of encapcheck() to determine the needed
handler.
- introduced struct encap_config, it contains parameters of encap handler
that is going to be registered by encap_attach() function.
- encap handlers are stored in lists ordered by exact_match value, thus
handlers that need more bits to match will be checked first, and if
encapcheck method returns exact_match value, the search will be stopped.
- all current consumers changed to use new KPI.
Reviewed by: mmacy
Sponsored by: Yandex LLC
Differential Revision: https://reviews.freebsd.org/D15617
- move harvest mask check inline
- move harvest mask to frequently_read out of actively
modified cache line
- disable ether_input collection and describe its limitations
in NOTES
Typically entropy collection in ether_input was stirring zero
in to the entropy pool while at the same time greatly reducing
max pps. This indicates that perhaps we should more closely
scrutinize how much entropy we're getting from a given source
as well as what our actual entropy collection needs are for
seeding Yarrow.
Reviewed by: cem, gallatin, delphij
Approved by: secteam
Differential Revision: https://reviews.freebsd.org/D15526
- Restore local change to include <net/bpf.h> inside pcap.h.
This fixes ports build problems.
- Update local copy of dlt.h with new DLT types.
- Revert no longer needed <net/bpf.h> includes which were added
as part of r334277.
Suggested by: antoine@, delphij@, np@
MFC after: 3 weeks
Sponsored by: Mellanox Technologies
Given that PF_RULES_LOCK is a mostly read lock, replace the rwlock with rmlock.
This change improves packet processing rate in high pps environments.
Benchmarking by olivier@ shows a 65% improvement in pps.
While here, also eliminate all appearances of "sys/rwlock.h" includes since it
is not used anymore.
Submitted by: farrokhi@
Differential Revision: https://reviews.freebsd.org/D15502
The *name parameter passed to iflib_irq_alloc_generic and
iflib_softirq_alloc_generic is never modified. Many places in code pass
string literals and thus should not be modified.
Mark the *name parameter as a const char * instead, so that we enforce
that the name is not modified before passing to bus_describe_intr()
Submitted by: Jacob Keller <jacob.e.keller@intel.com>
Reviewed by: kmacy
Sponsored by: Intel Corporation
Differential Revision: https://reviews.freebsd.org/D15343
ixl(4)'s nvmupdate utility expects the nvmupdate process to run
while the interface is down; these nvm update commands use the
admin queue, so the admin queue needs to be able to generate
interrupts and be processed while the interface is down.
So add a flag that ixl(4) sets that lets the entire admin task
run even when the interface is marked down/IFF_DRV_RUNNING isn't set.
With this change, nvmupdate should function like it did pre-iflib.
Reviewed by: gallatin@, sbruno@
MFC after: 1 week
Sponsored by: Intel Corporation
Differential Revision: https://reviews.freebsd.org/D15575
an rtentry. r334118 introduced a case when this was not done.
While we're here make the intent more obvious by moving the refcount
bump down to when we know we'll actually need it.
Reported by: markj
As reported in PR184149, it can happen that epair devices can have the same
MAC address.
This solution is based on a 32-bit hash, obtained combining the if_index of
the a interface and the hostid.
If the hostid is zero, a random number is used.
PR: 184149
Reviewed by: wollman, eugen
Approved by: cognet
Differential Revision: https://reviews.freebsd.org/D15329
Even though 64-bit atomics are supported on i386 there are panics
indicating that the code does not work correctly there. Switch
to mutex based variant (and fix that while we're here).
Reported by: pho, kib
There are risks associated with waiting on a preemptible epoch section.
Change the name to make them not be the default and document the issue
under CAVEATS.
Reported by: markj
adds:
- epoch_enter_critical() - can be called inside a different epoch,
starts a section that will acquire any MTX_DEF mutexes or do
anything that might sleep.
- epoch_exit_critical() - corresponding exit call
- epoch_wait_critical() - wait variant that is guaranteed that any
threads in a section are running.
- epoch_global_critical - an epoch_wait_critical safe epoch instance
Requested by: markj
Approved by: sbruno
When poll() is called via netmap, txsync is initially called,
and if there are no available buffers to reclaim, it waits for the driver
to notify of new buffers. Since the TX IRQ is generally not used in iflib
drivers, this ends up causing a timeout.
Work around this by having the reclaim DELAY(1) if it's initially unable
to reclaim anything, then schedule the tx task, which will spin by
continuously rescheduling itself until some buffers are reclaimed. In
general, the delay is enough to allow some buffers to be reclaimed, so
spinning is minimized.
Reported by: Johannes Lundberg <johalun0@gmail.com>
Reviewed by: sbruno
Sponsored by: Limelight Networks
Differential Revision: https://reviews.freebsd.org/D15455
Use the new epoch based reclamation API. Now the hot paths will not
block at all, and the sx lock is used for the softc data. This fixes LORs
reported where the rwlock was obtained when the sxlock was held.
Submitted by: mmacy
Reported by: Harry Schmalzbauer <freebsd@omnilan.de>
Reviewed by: sbruno
Sponsored by: Limelight Networks
Differential Revision: https://reviews.freebsd.org/D15355
Part 3 of many ...
The VPC framework relies heavily on cloning pseudo interfaces
(vmnics, vpc switch, vcpswitch port, hostif, vxlan if, etc).
This pulls in that piece. Some ancillary changes get pulled
in as a side effect.
Reviewed by: shurd@
Approved by: sbruno@
Sponsored by: Joyent, Inc.
Differential Revision: https://reviews.freebsd.org/D15347
It is guaranteed that if_ipsec(4) interface is used only for tunnel
mode IPsec, i.e. decrypted and decapsultaed packet has its own IP header.
Thus we can consider it as new packet and clear the protocols flags.
This allows ICMP/ICMPv6 properly handle errors that may cause this packet.
PR: 228108
MFC after: 1 week
if_bridge has a lot of limitations that make it scale poorly to higher data
rates. In my projects/VPC branch I leverage the bridge interface between
layers for my high speed soft switch as well as for purposes of stacking
in general.
Reviewed by: sbruno@
Approved by: sbruno@
Differential Revision: https://reviews.freebsd.org/D15344
Make if_printf() use vlog() instead of vprintf(). This means it can no
longer return the number of characters printed, as it used to, but every
single call to if_printf() in the entire kernel ignores the return value
anyway; just return 0 so we don't have to change the prototype.
Consistently use if_printf() throughout sys/net/if.c, instead of a
mixture of if_printf() and log().
In ifa_maintain_loopback_route(), don't needlessly log an error if we
either failed to add a route because it already existed or failed to
remove one because it did not. We still return an error code, though.
MFC after: 1 week
Print a message when iflib_tx_structures_setup fails, like we do for
iflib_rx_structures_setup.
Now that we always print a message from within
iflib_qset_structures_setup when it fails, stop printing one in
iflib_device_register() at the call site.
Submitted by: Jacob Keller <jacob.e.keller@intel.com>
Reviewed by: gallatin
MFC after: 3 days
Sponsored by: Intel Corporation
Differential Revision: https://reviews.freebsd.org/D15300
The canonical check for whether or not a ring is drainable is
TXQ_AVAIL() > MAX_TX_DESC() + 2. Use this same construct here,
in order to avoid a potential off-by-one error where we might otherwise
fail to request an interrupt.
Reviewed by: mmacy
Sponsored by: Netflix
to sleep on commands to the NIC when updating multicast filters. More generally this permitted
driver's to use an sx as a softc lock. Unfortunately this change introduced a race whereby a
a multicast update would still be queued for deletion when ifconfig deleted the interface
thus calling down in to _purgemaddrs and synchronously deleting _all_ of the multicast addresses
on the interface.
Synchronously remove all external references to a multicast address before enqueueing for delete.
Reported by: lwhsu
Approved by: sbruno
em(4) and igb(4) were tested by me, and ixgbe(4) and bnxt(4) were
tested by sbruno.
Reviewed by: mmacy, shurd
MFC after: 1 month
Sponsored by: Dell EMC Isilon
Differential Revision: https://reviews.freebsd.org/D15262
This is a component of a system which lets the kernel dump core to
a remote host after a panic, rather than to a local storage device.
The server component is available in the ports tree. netdump is
particularly useful on diskless systems.
The netdump(4) man page contains some details describing the protocol.
Support for configuring netdump will be added to dumpon(8) in a future
commit. To use netdump, the kernel must have been compiled with the
NETDUMP option.
The initial revision of netdump was written by Darrell Anderson and
was integrated into Sandvine's OS, from which this version was derived.
Reviewed by: bdrewery, cem (earlier versions), julian, sbruno
MFC after: 1 month
X-MFC note: use a spare field in struct ifnet
Sponsored by: Dell EMC Isilon
Differential Revision: https://reviews.freebsd.org/D15253
In r301567, code was added to cleanup to prevent memory leaks for the
Tx and Rx ring structs. This code carefully tracked txq and rxq, and
made sure to free them properly during cleanup.
Because we assigned the txq and rxq pointers into the ctx->ifc_txqs and
ctx->ifc_rxqs, we carefully reset these pointers to NULL, so that
cleanup code would not accidentally free the memory twice.
This was changed by r304021 ("Update iflib to support more NIC designs"),
which removed this resetting of the pointers to NULL, because it re-used
the txq and rxq pointers as an index into the queue set array.
Unfortunately, the cleanup code was left alone. Thus, if we fail to
allocate DMA or fail to configure the queues using the drivers ifdi
methods, we will attempt to free txq and rxq. These variables would now
incorrectly point to the wrong location, resulting in a page fault.
There are a number of methods to correct this, but ultimately the root
cause was that we reuse the txq and rxq pointers for two different
purposes.
Instead, when allocating, store the returned pointer directly into
ctx->ifc_txqs and ctx->ifc_rxqs. Then, assign this to txq and rxq as
index pointers before starting the loop to allocate each queue.
Drop the cleanup code for txq and rxq, and only use ctx->ifc_txqs and
ctx->ifc_rxqs.
Thus, we no longer need to free txq or rxq under any error flow, and
intsead rely solely on the pointers stored in ctx->ifc_txqs and
ctx->ifc_rxqs. This prevents the invalid free(), and ensures that we
still properly cleanup after ourselves as before when failing to
allocate.
Submitted by: Jacob Keller
Reviewed by: gallatin, sbruno
Sponsored by: Intel Corporation
Differential Revision: https://reviews.freebsd.org/D15285
This pointer was no longer written to as of r315217. Since nothing writes
to the variable, remove it.
Submitted by: Jacob Keller <jacob.e.keller@intel.com>
Reviewed by: gallatin, kmacy, sbruno
Differential Revision: https://reviews.freebsd.org/D15284
Since the move to SMP NIC driver locking has had to go through serious
contortions using mtx around long running hardware operations. This moves
iflib past that.
Individual drivers may now sleep when appropriate.
Submitted by: Matthew Macy <mmacy@mattmacy.io>
Reviewed by: shurd
Sponsored by: Limelight Networks
Differential Revision: https://reviews.freebsd.org/D14983
Multicast incorrectly calls in to drivers with a mutex held causing drivers
to have to go through all manner of contortions to use a non sleepable lock.
Serialize multicast updates instead.
Submitted by: mmacy <mmacy@mattmacy.io>
Reviewed by: shurd, sbruno
Sponsored by: Limelight Networks
Differential Revision: https://reviews.freebsd.org/D14969
1) Don't give up if m_collapse() fails. Rather than giving up, try
m_defrag() immediately.
2) Fix a leak where, if the NIC driver rejected the defrag'ed chain
as having too many segments, we would fail to free the chain.
Reviewed by: Matthew Macy <mmacy@mattmacy.io> (this version of patch)
Submitted by: Matthew Macy <mmacy@mattmacy.io> (early version of leak fix)
