Buffer is 64bytes, but label printing can take 66bytes printing
in hex, and will overflow when setting the string delimiter ('\0').
Fix that by increasing the print buffer size.
Example of overflowing ct_label:
ct_label 11111111111111111111111111111111/11111111111111111111111111111111
Fixes: 2fffb1c030 ("tc: flower: Add matching on conntrack info")
Signed-off-by: Paul Blakey <paulb@nvidia.com>
Signed-off-by: Stephen Hemminger <stephen@networkplumber.org>
Fix the wild bracket in the if clause leading to the error in the condition.
Fixes: d61167dd88 ("m_vlan: add pop_eth and push_eth actions")
Signed-off-by: Maxim Petrov <mmrmaximuzz@gmail.com>
Signed-off-by: Stephen Hemminger <stephen@networkplumber.org>
Provided port range in tc rule are parsed incorrectly.
Even though range is passed as min-max. It throws an error.
$ tc filter add dev eth0 ingress handle 100 priority 10000 protocol ipv4 flower ip_proto tcp dst_port 10368-61000 action pass
max value should be greater than min value
Illegal "dst_port"
Fixes: 8930840e67 ("tc: flower: Classify packets based port ranges")
Signed-off-by: Puneet Sharma <pusharma@akamai.com>
Signed-off-by: Stephen Hemminger <stephen@networkplumber.org>
Recently we added SKBMOD_F_ECN option support to the kernel; support it in
the tc-skbmod(8) front end, and update its man page accordingly.
The 2 least significant bits of the Traffic Class field in IPv4 and IPv6
headers are used to represent different ECN states [1]:
0b00: "Non ECN-Capable Transport", Non-ECT
0b10: "ECN Capable Transport", ECT(0)
0b01: "ECN Capable Transport", ECT(1)
0b11: "Congestion Encountered", CE
This new option, "ecn", marks ECT(0) and ECT(1) IPv{4,6} packets as CE,
which is useful for ECN-based rate limiting. For example:
$ tc filter add dev eth0 parent 1: protocol ip prio 10 \
u32 match ip protocol 1 0xff flowid 1:2 \
action skbmod \
ecn
The updated tc-skbmod SYNOPSIS looks like the following:
tc ... action skbmod { set SETTABLE | swap SWAPPABLE | ecn } ...
Only one of "set", "swap" or "ecn" shall be used in a single tc-skbmod
command. Trying to use more than one of them at a time is considered
undefined behavior; pipe multiple tc-skbmod commands together instead.
"set" and "swap" only affect Ethernet packets, while "ecn" only affects
IP packets.
Depends on kernel patch "net/sched: act_skbmod: Add SKBMOD_F_ECN option
support", as well as iproute2 patch "tc/skbmod: Remove misinformation
about the swap action".
[1] https://en.wikipedia.org/wiki/Explicit_Congestion_Notification
Reviewed-by: Cong Wang <cong.wang@bytedance.com>
Signed-off-by: Peilin Ye <peilin.ye@bytedance.com>
Signed-off-by: David Ahern <dsahern@kernel.org>
In between Linux kernel 2.4 and 2.6, key folding for hash tables changed
in kernel space. When iproute2 dropped support for the older algorithm,
the wrong code was removed and kernel 2.4 folding method remained in
place. To get things functional for recent kernels again, restoring the
old code alone was not sufficient - additional byteorder fixes were
needed.
While being at it, make use of ffs() and thereby align the code with how
kernel determines the shift width.
Fixes: 267480f553 ("Backout the 2.4 utsname hash patch.")
Signed-off-by: Phil Sutter <phil@nwl.cc>
Signed-off-by: Stephen Hemminger <stephen@networkplumber.org>
Currently man 8 tc-skbmod says that "...the swap action will occur after
any smac/dmac substitutions are executed, if they are present."
This is false. In fact, trying to "set" and "swap" in a single skbmod
command causes the "set" part to be completely ignored. As an example:
$ tc filter add dev eth0 parent 1: protocol ip prio 10 \
matchall action skbmod \
set dmac AA:AA:AA:AA:AA:AA smac BB:BB:BB:BB:BB:BB \
swap mac
The above command simply does a "swap", without setting DMAC or SMAC to
AA's or BB's. The root cause of this is in the kernel, see
net/sched/act_skbmod.c:tcf_skbmod_init():
parm = nla_data(tb[TCA_SKBMOD_PARMS]);
index = parm->index;
if (parm->flags & SKBMOD_F_SWAPMAC)
lflags = SKBMOD_F_SWAPMAC;
^^^^^^^^^^^^^^^^^^^^^^^^^^
Doing a "=" instead of "|=" clears all other "set" flags when doing a
"swap". Discourage using "set" and "swap" in the same command by
documenting it as undefined behavior, and update the "SYNOPSIS" section
as well as tc -help text accordingly.
If one really needs to e.g. "set" DMAC to all AA's then "swap" DMAC and
SMAC, one should do two separate commands and "pipe" them together.
Reviewed-by: Cong Wang <cong.wang@bytedance.com>
Signed-off-by: Peilin Ye <peilin.ye@bytedance.com>
Signed-off-by: Stephen Hemminger <stephen@networkplumber.org>
With the json support fix the normal output was
changed. set it back to what it was.
Print overhead with print_size().
Print newline before ref.
Fixes: 0d5cf51e0d ("police: Add support for json output")
Signed-off-by: Roi Dayan <roid@nvidia.com>
Signed-off-by: Stephen Hemminger <stephen@networkplumber.org>
Implement a decrement operation for ttl and hoplimit.
Since this is just syntactic sugar, it goes that:
tc filter add ... action pedit ex munge ip ttl dec ...
tc filter add ... action pedit ex munge ip6 hoplimit dec ...
is just a more readable version of this:
tc filter add ... action pedit ex munge ip ttl add 0xff ...
tc filter add ... action pedit ex munge ip6 hoplimit add 0xff ...
This feature was suggested by some pseudo tc examples in Mellanox's
documentation[1], but wasn't present in neither their mlnx-iproute2
nor iproute2.
Tested with skip_sw on Mellanox ConnectX-6 Dx.
[1] https://docs.mellanox.com/pages/viewpage.action?pageId=47033989
v3:
- Use dedicated flags argument in parse_cmd() (David Ahern)
- Minor rewording of the man page
v2:
- Fix whitespace issue (Stephen Hemminger)
- Add to usage info in explain()
Signed-off-by: Asbjørn Sloth Tønnesen <asbjorn@asbjorn.st>
Acked-by: Jamal Hadi Salim <jhs@mojatatu.com>
Signed-off-by: David Ahern <dsahern@kernel.org>
This patch just prepares the flags argument, so it's
available to the next patch.
Signed-off-by: Asbjørn Sloth Tønnesen <asbjorn@asbjorn.st>
Signed-off-by: David Ahern <dsahern@kernel.org>
Add support for matching on ct_state flag related.
The related state indicates a packet is associated with an existing
connection.
Example:
$ tc filter add dev ens1f0_0 ingress prio 1 chain 1 proto ip flower \
ct_state -est-rel+trk \
action mirred egress redirect dev ens1f0_1
$ tc filter add dev ens1f0_0 ingress prio 1 chain 1 proto ip flower \
ct_state +rel+trk \
action mirred egress redirect dev ens1f0_1
Signed-off-by: Ariel Levkovich <lariel@nvidia.com>
Reviewed-by: Jiri Pirko <jiri@nvidia.com>
Signed-off-by: David Ahern <dsahern@kernel.org>
When a wrong value is provided for "burst" or "cburst" parameters, the
resulting error message is unclear and can be misleading:
$ tc class add dev dummy0 parent 1: classid 1:1 htb rate 100KBps burst errtrigger
Illegal "buffer"
The message claims an illegal "buffer" is provided, but neither the
inline help nor the man page list "buffer" among the htb parameters, and
the only way to know that "burst", "maxburst" and "buffer" are synonyms
is to look into tc/q_htb.c.
This commit tries to improve this simply changing the error string to
the parameter name provided in the user-given command, clearly pointing
out where the wrong value is.
$ tc class add dev dummy0 parent 1: classid 1:1 htb rate 100KBps burst errtrigger
Illegal "burst"
$ tc class add dev dummy0 parent 1: classid 1:1 htb rate 100Kbps maxburst errtrigger
Illegal "maxburst"
Reported-by: Sebastian Mitterle <smitterl@redhat.com>
Signed-off-by: Andrea Claudi <aclaudi@redhat.com>
Signed-off-by: David Ahern <dsahern@kernel.org>
Checking for nbands to be at least 1 at this point is useless. Indeed:
- ets requires "bands", "quanta" or "strict" to be specified
- if "bands" is specified, nbands cannot be negative, see parse_nbands()
- if "strict" is specified, nstrict cannot be negative, see
parse_nbands()
- if "quantum" is specified, nquanta cannot be negative, see
parse_quantum()
- if "bands" is not specified, nbands is set to nstrict+nquanta
- the previous if statement takes care of the case when none of them are
specified and nbands is 0, terminating execution.
Thus nbands cannot be < 1 at this point and this code cannot be executed.
Signed-off-by: Andrea Claudi <aclaudi@redhat.com>
Signed-off-by: David Ahern <dsahern@kernel.org>
envp_run is dinamically allocated with a malloc, and not freed in the
out: return path. This commit fix it.
Signed-off-by: Andrea Claudi <aclaudi@redhat.com>
Signed-off-by: Stephen Hemminger <stephen@networkplumber.org>
In cake_parse_opt(), *argv is checked not to be null when parsing for
overhead and mpu parameters. However this is useless, since *argv
matches right before for "overhead" or "mpu".
Signed-off-by: Andrea Claudi <aclaudi@redhat.com>
Signed-off-by: Stephen Hemminger <stephen@networkplumber.org>
Allow a policer action to enforce a rate-limit based on packets-per-second,
configurable using a packet-per-second rate and burst parameters.
e.g.
# $TC actions add action police pkts_rate 1000 pkts_burst 200 index 1
# $TC actions ls action police
total acts 1
action order 0: police 0x1 rate 0bit burst 0b mtu 4096Mb pkts_rate 1000 pkts_burst 200
ref 1 bind 0
Signed-off-by: Baowen Zheng <baowen.zheng@corigine.com>
Signed-off-by: Simon Horman <simon.horman@netronome.com>
Signed-off-by: Louis Peens <louis.peens@netronome.com>
Signed-off-by: David Ahern <dsahern@kernel.org>
The deficit returned from the kernel is signed, but was printed with a %u
specifier in the format string, leading to negative values to be printed as
high unsigned values instead. In addition, we passed a negative value to
sprint_time() even though that expects an unsigned value. Fix this by
changing the format specifier and reversing the sign of negative time
values.
Fixes: 714444c0cb ("Add support for CAKE qdisc")
Signed-off-by: Toke Høiland-Jørgensen <toke@redhat.com>
Signed-off-by: Stephen Hemminger <stephen@networkplumber.org>
sprint_time64() uses SPRINT_BSIZE-1 as a constant buffer lenght in its
implementation, however m_gate uses shorter buffers when calling it.
Fix this using SPRINT_BUF macro to get the buffer, thus getting a
SPRINT_BSIZE-long buffer.
Fixes: 07d5ee70b5 ("iproute2-next:tc:action: add a gate control action")
Signed-off-by: Andrea Claudi <aclaudi@redhat.com>
Signed-off-by: Stephen Hemminger <stephen@networkplumber.org>
The json output of the TCA_FLOWER_KEY_MPLS_OPTS attribute was invalid.
Example:
$ tc filter add dev eth0 ingress protocol mpls_uc flower mpls \
lse depth 1 label 100 \
lse depth 2 label 200
$ tc -json filter show dev eth0 ingress
...{"eth_type":"8847",
" mpls":[" lse":["depth":1,"label":100],
" lse":["depth":2,"label":200]]}...
This is invalid as the arrays, introduced by "[", can't contain raw
string:value pairs. Those must be enclosed into "{}" to form valid json
ojects. Also, there are spurious whitespaces before the mpls and lse
strings because of the indentation used for normal output.
Fix this by putting all LSE parameters (depth, label, tc, bos and ttl)
into the same json object. The "mpls" key now directly contains a list
of such objects.
Also, handle strings differently for normal and json output, so that
json strings don't get spurious indentation whitespaces.
Normal output isn't modified.
The json output now looks like:
$ tc -json filter show dev eth0 ingress
...{"eth_type":"8847",
"mpls":[{"depth":1,"label":100},
{"depth":2,"label":200}]}...
Fixes: eb09a15c12 ("tc: flower: support multiple MPLS LSE match")
Signed-off-by: Guillaume Nault <gnault@redhat.com>
Signed-off-by: Stephen Hemminger <stephen@networkplumber.org>
keys_ex is dinamically allocated with calloc on line 770, but
is not freed in case of error at line 823.
Fixes: 081d6c310d ("tc: pedit: Support JSON dumping")
Signed-off-by: Andrea Claudi <aclaudi@redhat.com>
Signed-off-by: Stephen Hemminger <stephen@networkplumber.org>
The function get_size() serves for parsing of sizes using a handly notation
that supports units and their prefixes, such as 10Kbit. This will be useful
for the DCB buffer size parsing. Move the function from TC to the general
library, so that it can be reused.
Signed-off-by: Petr Machata <me@pmachata.org>
Signed-off-by: David Ahern <dsahern@gmail.com>
The functions get_rate() and get_rate64() are useful for parsing rate-like
values. The DCB tool will find these useful in the maxrate subtool.
Move them over to lib so that they can be easily reused.
Signed-off-by: Petr Machata <me@pmachata.org>
Signed-off-by: David Ahern <dsahern@gmail.com>
When displaying sizes of various sorts, tc commonly uses the function
sprint_size() to format the size into a buffer as a human-readable string.
This string is then displayed either using print_string(), or in some code
even fprintf(). As a result, a typical sequence of code when formatting a
size is something like the following:
SPRINT_BUF(b);
print_uint(PRINT_JSON, "foo", NULL, foo);
print_string(PRINT_FP, NULL, "foo %s ", sprint_size(foo, b));
For a concept as broadly useful as size, it would be better to have a
dedicated function in json_print.
To that end, move sprint_size() from tc_util to json_print. Add helpers
print_size() and print_color_size() that wrap arount sprint_size() and
provide the JSON dispatch as appropriate.
Since print_size() should be the preferred interface, convert vast majority
of uses of sprint_size() to print_size(). Two notable exceptions are:
- q_tbf, which does not show the size as such, but uses the string
"$human_readable_size/$cell_size" even in JSON. There is simply no way to
have print_size() emit the same text, because print_size() in JSON mode
should of course just use the raw number, without human-readable frills.
- q_cake, which relies on the existence of sprint_size() in its macro-based
formatting helpers. There might be ways to convert this particular case,
but given q_tbf simply cannot be converted, leave it as is.
Signed-off-by: Petr Machata <me@pmachata.org>
Signed-off-by: David Ahern <dsahern@gmail.com>
The functions print_rate() and sprint_rate() are useful for formatting
rate-like values. The DCB tool would find these useful in the maxrate
subtool. However, the current interface to these functions uses a global
variable use_iec as a flag indicating whether 1024- or 1000-based powers
should be used when formatting the rate value. For general use, a global
variable is not a great way of passing arguments to a function. Besides, it
is unlike most other printing functions in that it deals in buffers and
ignores JSON.
Therefore make the interface to print_rate() explicit by converting use_iec
to an ordinary parameter. Since the interface changes anyway, convert it to
follow the pattern of other json_print functions (except for the
now-explicit use_iec parameter). Move to json_print.c.
Add a wrapper to tc, so that all the call sites do not need to repeat the
use_iec global variable argument, and convert all call sites.
In q_cake.c, the conversion is not straightforward due to usage of a macro
that is shared across numerous data types. Simply hand-roll the
corresponding code, which seems better than making an extra helper for one
call site.
Drop sprint_rate() now that everybody just uses print_rate().
Signed-off-by: Petr Machata <me@pmachata.org>
Signed-off-by: David Ahern <dsahern@gmail.com>
The tools "ip" and "tc" use a flag "use_iec", which indicates whether, when
formatting rate values, the prefixes "K", "M", etc. should refer to powers
of 1024, or powers of 1000. The flag is currently kept as a global variable
in "ip" and "tc", but is nonetheless declared in util.h.
Instead, move the declaration to tool-specific headers ip/ip_common.h and
tc/tc_common.h.
Signed-off-by: Petr Machata <me@pmachata.org>
Signed-off-by: David Ahern <dsahern@gmail.com>
Implement support for action terse dump using new TCA_ACT_FLAG_TERSE_DUMP
value of TCA_ROOT_FLAGS tlv. Set the flag when user requested it with
following example CLI (-br for 'brief'):
$ tc -s -br actions ls action tunnel_key
total acts 2
action order 0: tunnel_key index 1
Action statistics:
Sent 0 bytes 0 pkt (dropped 0, overlimits 0 requeues 0)
backlog 0b 0p requeues 0
action order 1: tunnel_key index 2
Action statistics:
Sent 0 bytes 0 pkt (dropped 0, overlimits 0 requeues 0)
backlog 0b 0p requeues 0
In terse mode dump only outputs essential data needed to identify the
action (kind, index) and stats, if requested by the user.
Signed-off-by: Vlad Buslov <vlad@buslov.dev>
Suggested-by: Jamal Hadi Salim <jhs@mojatatu.com>
Signed-off-by: David Ahern <dsahern@gmail.com>
Use TCA_ACT_FLAG_LARGE_DUMP_ON alias according to new preferred naming for
action flags.
Signed-off-by: Vlad Buslov <vlad@buslov.dev>
Signed-off-by: David Ahern <dsahern@gmail.com>
With gcc-10 it complains about array subscript error.
f_u32.c: In function ‘u32_parse_opt’:
f_u32.c:1113:24: warning: array subscript 0 is outside the bounds of an interior zero-length array ‘struct tc_u32_key[0]’ [-Wzero-length-bounds]
1113 | hash = sel2.sel.keys[0].val & sel2.sel.keys[0].mask;
| ~~~~~~~~~~~~~^~~
In file included from tc_util.h:11,
from f_u32.c:26:
../include/uapi/linux/pkt_cls.h:253:20: note: while referencing ‘keys’
253 | struct tc_u32_key keys[0];
|
This is because the keys are actually allocated in the second element
of the parent structure.
Simplest way to address the warning is to assign directly to the keys
in the containing structure.
This has always been in iproute2 (pre-git) so no Fixes.
Signed-off-by: Stephen Hemminger <stephen@networkplumber.org>
Gcc-10 complains about referencing a zero size array.
This occurs because the array of keys is actually in the following
structure which is part of the overall selector.
The original code was safe, but better to just use the key
array directly.
Fixes: 2d9a8dc439 ("tc: p_ip6: Support pedit of IPv6 dsfield")
Cc: petrm@mellanox.com
Signed-off-by: Stephen Hemminger <stephen@networkplumber.org>
This patch aim to add basic checking functions for later iproute2
libbpf support.
First we add check_libbpf() in configure to see if we have bpf library
support. By default the system libbpf will be used, but static linking
against a custom libbpf version can be achieved by passing libbpf DESTDIR
to variable LIBBPF_DIR for configure.
Another variable LIBBPF_FORCE is used to control whether to build iproute2
with libbpf. If set to on, then force to build with libbpf and exit if
not available. If set to off, then force to not build with libbpf.
When dynamically linking against libbpf, we can't be sure that the
version we discovered at compile time is actually the one we are
using at runtime. This can lead to hard-to-debug errors. So we add
a new file lib/bpf_glue.c and a helper function get_libbpf_version()
to get correct libbpf version at runtime.
Signed-off-by: Hangbin Liu <haliu@redhat.com>
Signed-off-by: David Ahern <dsahern@gmail.com>
When protocol is vlan then eth_type is set to the vlan eth type.
So when parsing vlan_id and vlan_prio need to check tc_proto
is vlan and not eth_type.
Fixes: 4c551369e0 ("tc flower: use right ethertype in icmp/arp parsing")
Signed-off-by: Roi Dayan <roid@nvidia.com>
Signed-off-by: David Ahern <dsahern@gmail.com>
Currently the icmp and arp parsing functions are called with incorrect
ethtype in case of vlan or cvlan filter options. In this case either
cvlan_ethtype or vlan_ethtype has to be used. The ethtype is now updated
each time a vlan ethtype is matched during parsing.
Signed-off-by: Zahari Doychev <zahari.doychev@linux.com>
Signed-off-by: David Ahern <dsahern@gmail.com>
The code for handling batches is largely the same across iproute2 tools.
Extract a helper to handle the batch, and adjust the tools to dispatch to
this helper. Sandwitch the invocation between prologue / epilogue code
specific for each tool.
Signed-off-by: Petr Machata <me@pmachata.org>
Signed-off-by: David Ahern <dsahern@gmail.com>
Manpage:
* Remove the extra "and to ip packets" part from command description
to make it more understandable.
* Redirect packets to eth1, instead of eth0, as told in the
description.
Help string:
* "mpls pop" can be followed by a CONTROL keyword.
* "mpls modify" can also set the MPLS_BOS field.
Signed-off-by: Guillaume Nault <gnault@redhat.com>
Signed-off-by: Stephen Hemminger <stephen@networkplumber.org>
* "vlan pop" can be followed by a CONTROL keyword.
* Add missing space in error message.
Signed-off-by: Guillaume Nault <gnault@redhat.com>
Signed-off-by: Stephen Hemminger <stephen@networkplumber.org>
Implement support for classifier/action terse dump using new TCA_DUMP_FLAGS
tlv with only available flag value TCA_DUMP_FLAGS_TERSE. Set the flag when
user requested it with following example CLI (-br for 'brief'):
$ tc -s -br filter show dev ens1f0 ingress
filter protocol ip pref 49151 flower chain 0
filter protocol ip pref 49151 flower chain 0 handle 0x1
not_in_hw
action order 1: gact Action statistics:
Sent 0 bytes 0 pkt (dropped 0, overlimits 0 requeues 0)
backlog 0b 0p requeues 0
filter protocol ip pref 49152 flower chain 0
filter protocol ip pref 49152 flower chain 0 handle 0x1
not_in_hw
action order 1: gact Action statistics:
Sent 0 bytes 0 pkt (dropped 0, overlimits 0 requeues 0)
backlog 0b 0p requeues 0
In terse mode dump only outputs essential data needed to identify the
filter and action (handle, cookie, etc.) and stats, if requested by the
user. The intention is to significantly improve rule dump rate by omitting
all static data that do not change after rule is created.
Signed-off-by: Vlad Buslov <vladbu@mellanox.com>
Signed-off-by: David Ahern <dsahern@gmail.com>
Paul Blakey
Maxim Petrov
Puneet Sharma
Luca Boccassi
Peilin Ye
Phil Sutter
Roi Dayan
Asbjørn Sloth Tønnesen
Eric Dumazet
Ariel Levkovich
Andrea Claudi
Stephen Hemminger
Baowen Zheng
Toke Høiland-Jørgensen
Maxim Mikityanskiy
wenxu
Guillaume Nault
David Ahern
Petr Machata
Vlad Buslov
Hangbin Liu
Zahari Doychev
Vlad Buslov