| Age | Commit message (Collapse) | Author |
|---|
|
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
The vzalloc_node(), already rounds the total size to whole pages, and
sizeof(u64) is smaller than sizeof(struct recv_comp_data). So
round_up of recv_completion_cnt is not necessary, and may cause extra
memory allocation.
To save memory, remove this unnecessary round_up for recv_completion_cnt.
Signed-off-by: Haiyang Zhang <haiyangz@microsoft.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Pablo Neira Ayuso says:
====================
Netfilter/IPVS updates for net-next
The following patchset contains Netfilter/IPVS updates for net-next:
1) Add support to specify a stateful expression in set definitions,
this allows users to specify e.g. counters per set elements.
2) Flowtable software counter support.
3) Flowtable hardware offload counter support, from wenxu.
3) Parallelize flowtable hardware offload requests, from Paul Blakey.
This includes a patch to add one work entry per offload command.
4) Several patches to rework nf_queue refcount handling, from Florian
Westphal.
4) A few fixes for the flowtable tunnel offload: Fix crash if tunneling
information is missing and set up indirect flow block as TC_SETUP_FT,
patch from wenxu.
5) Stricter netlink attribute sanity check on filters, from Romain Bellan
and Florent Fourcot.
5) Annotations to make sparse happy, from Jules Irenge.
6) Improve icmp errors in debugging information, from Haishuang Yan.
7) Fix warning in IPVS icmp error debugging, from Haishuang Yan.
8) Fix endianess issue in tcp extension header, from Sergey Marinkevich.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Ido Schimmel says:
====================
Add packet trap policers support
Background
==========
Devices capable of offloading the kernel's datapath and perform
functions such as bridging and routing must also be able to send (trap)
specific packets to the kernel (i.e., the CPU) for processing.
For example, a device acting as a multicast-aware bridge must be able to
trap IGMP membership reports to the kernel for processing by the bridge
module.
Motivation
==========
In most cases, the underlying device is capable of handling packet rates
that are several orders of magnitude higher compared to those that can
be handled by the CPU.
Therefore, in order to prevent the underlying device from overwhelming
the CPU, devices usually include packet trap policers that are able to
police the trapped packets to rates that can be handled by the CPU.
Proposed solution
=================
This patch set allows capable device drivers to register their supported
packet trap policers with devlink. User space can then tune the
parameters of these policers (currently, rate and burst size) and read
from the device the number of packets that were dropped by the policer,
if supported.
These packet trap policers can then be bound to existing packet trap
groups, which are used to aggregate logically related packet traps. As a
result, trapped packets are policed to rates that can be handled the
host CPU.
Example usage
=============
Instantiate netdevsim:
Dump available packet trap policers:
netdevsim/netdevsim10:
policer 1 rate 1000 burst 128
policer 2 rate 2000 burst 256
policer 3 rate 3000 burst 512
Change the parameters of a packet trap policer:
Bind a packet trap policer to a packet trap group:
Dump parameters and statistics of a packet trap policer:
netdevsim/netdevsim10:
policer 3 rate 100 burst 16
stats:
rx:
dropped 92
Unbind a packet trap policer from a packet trap group:
Patch set overview
==================
Patch #1 adds the core infrastructure in devlink which allows capable
device drivers to register their supported packet trap policers with
devlink.
Patch #2 extends the existing devlink-trap documentation.
Patch #3 extends netdevsim to register a few dummy packet trap policers
with devlink. Used later on to selftests the core infrastructure.
Patches #4-#5 adds infrastructure in devlink to allow binding of packet
trap policers to packet trap groups.
Patch #6 extends netdevsim to allow such binding.
Patch #7 adds a selftest over netdevsim that verifies the core
devlink-trap policers functionality.
Patches #8-#14 gradually add devlink-trap policers support in mlxsw.
Patch #15 adds a selftest over mlxsw. All registered packet trap
policers are verified to handle the configured rate and burst size.
Future plans
============
* Allow changing default association between packet traps and packet
trap groups
* Add more packet traps. For example, for control packets (e.g., IGMP)
v3:
* Rebase
v2 (address comments from Jiri and Jakub):
* Patch #1: Add 'strict_start_type' in devlink policy
* Patch #1: Have device drivers provide max/min rate/burst size for each
policer. Use them to check validity of user provided parameters
* Patch #3: Remove check about burst size being a power of 2 and instead
add a debugfs knob to fail the operation
* Patch #3: Provide max/min rate/burst size when registering policers
and remove the validity checks from nsim_dev_devlink_trap_policer_set()
* Patch #5: Check for presence of 'DEVLINK_ATTR_TRAP_POLICER_ID' in
devlink_trap_group_set() and bail if not present
* Patch #5: Add extack error message in case trap group was partially
modified
* Patch #7: Add test case with new 'fail_trap_policer_set' knob
* Patch #7: Add test case for partially modified trap group
* Patch #10: Provide max/min rate/burst size when registering policers
* Patch #11: Remove the max/min validity checks from
__mlxsw_sp_trap_policer_set()
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Add test cases that verify that each registered packet trap policer:
* Honors that imposed limitations of rate and burst size
* Able to police trapped packets to the specified rate
* Able to police trapped packets to the specified burst size
* Able to be unbound from its trap group
Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Implement support for setting of packet trap group parameters by
invoking the trap_group_init() callback with the new parameters.
Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Reviewed-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Some packet traps are currently exposed to user space as being member of
"l3_drops" trap group, but internally they are member of a different
group.
Switch these traps to use the correct group so that they are all subject
to the same policer, as exposed to user space.
Set the trap priority of packets trapped due to loopback error during
routing to the lowest priority. Such packets are not routed again by the
kernel and therefore should not mask other traps (e.g., host miss) that
should be routed.
Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Reviewed-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
The policer is now initialized as part of the registration with devlink,
so there is no need to initialize it before the registration.
Remove the initialization.
Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Reviewed-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Register supported packet trap policers with devlink and implement
callbacks to change their parameters and read their counters.
Prevent user space from passing invalid policer parameters down to the
device by checking their validity and communicating the failure via an
appropriate extack message.
v2:
* Remove the max/min validity checks from __mlxsw_sp_trap_policer_set()
Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Reviewed-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Prepare an array of policer IDs to register with devlink and their
associated parameters.
The array is composed from both policers that are currently bound to
exposed trap groups and policers that are not bound to any trap group.
v2:
* Provide max/min rate/burst size when registering policers
Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Reviewed-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
During initialization the driver configures various packet trap groups
and binds policers to them.
Currently, most of these groups are not exposed to user space and
therefore their policers should not be exposed as well. Otherwise, user
space will be able to alter policer parameters without knowing which
packet traps are policed by the policer.
Use a bitmap to track the used policer IDs so that these policers will
not be registered with devlink in a subsequent patch.
Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Reviewed-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
The QoS Policer Configuration Register (QPCR) is used to configure
hardware policers. Extend this register with following fields and
defines which will be used by subsequent patches:
1. Violate counter: reads number of packets dropped by the policer
2. Clear counter: to ensure we start counting from 0
3. Rate and burst size limits
Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Reviewed-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Add test cases for packet trap policer set / show commands as well as
for the binding of these policers to packet trap groups.
Both good and bad flows are tested for maximum coverage.
v2:
* Add test case with new 'fail_trap_policer_set' knob
* Add test case for partially modified trap group
Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Add a dummy callback to set trap group parameters. Return an error when
the 'fail_trap_group_set' debugfs file is set in order to exercise error
paths and verify that error is propagated to user space when should.
Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Reviewed-by: Jiri Pirko <jiri@mellanox.com>
Reviewed-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
The previous patch allowed device drivers to publish their default
binding between packet trap policers and packet trap groups. However,
some users might not be content with this binding and would like to
change it.
In case user space passed a packet trap policer identifier when setting
a packet trap group, invoke the appropriate device driver callback and
pass the new policer identifier.
v2:
* Check for presence of 'DEVLINK_ATTR_TRAP_POLICER_ID' in
devlink_trap_group_set() and bail if not present
* Add extack error message in case trap group was partially modified
Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Reviewed-by: Jiri Pirko <jiri@mellanox.com>
Acked-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Packet trap groups are used to aggregate logically related packet traps.
Currently, these groups allow user space to batch operations such as
setting the trap action of all member traps.
In order to prevent the CPU from being overwhelmed by too many trapped
packets, it is desirable to bind a packet trap policer to these groups.
For example, to limit all the packets that encountered an exception
during routing to 10Kpps.
Allow device drivers to bind default packet trap policers to packet trap
groups when the latter are registered with devlink.
The next patch will enable user space to change this default binding.
Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Reviewed-by: Jiri Pirko <jiri@mellanox.com>
Reviewed-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Register three dummy packet trap policers with devlink and implement
callbacks to change their parameters and read their counters.
This will be used later on in the series to test the devlink-trap
policer infrastructure.
v2:
* Remove check about burst size being a power of 2 and instead add a
debugfs knob to fail the operation
* Provide max/min rate/burst size when registering policers and remove
the validity checks from nsim_dev_devlink_trap_policer_set()
Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Reviewed-by: Jiri Pirko <jiri@mellanox.com>
Reviewed-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Extend devlink-trap documentation with information about packet trap
policers.
Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Reviewed-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Devices capable of offloading the kernel's datapath and perform
functions such as bridging and routing must also be able to send (trap)
specific packets to the kernel (i.e., the CPU) for processing.
For example, a device acting as a multicast-aware bridge must be able to
trap IGMP membership reports to the kernel for processing by the bridge
module.
In most cases, the underlying device is capable of handling packet rates
that are several orders of magnitude higher compared to those that can
be handled by the CPU.
Therefore, in order to prevent the underlying device from overwhelming
the CPU, devices usually include packet trap policers that are able to
police the trapped packets to rates that can be handled by the CPU.
This patch allows capable device drivers to register their supported
packet trap policers with devlink. User space can then tune the
parameters of these policer (currently, rate and burst size) and read
from the device the number of packets that were dropped by the policer,
if supported.
Subsequent patches in the series will allow device drivers to create
default binding between these policers and packet trap groups and allow
user space to change the binding.
v2:
* Add 'strict_start_type' in devlink policy
* Have device drivers provide max/min rate/burst size for each policer.
Use them to check validity of user provided parameters
Signed-off-by: Ido Schimmel <idosch@mellanox.com>
Reviewed-by: Jiri Pirko <jiri@mellanox.com>
Reviewed-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Andrii Nakryiko says:
====================
bpf_link abstraction itself was formalized in [0] with justifications for why
its semantics is a good fit for attaching BPF programs of various types. This
patch set adds bpf_link-based BPF program attachment mechanism for cgroup BPF
programs.
Cgroup BPF link is semantically compatible with current BPF_F_ALLOW_MULTI
semantics of attaching cgroup BPF programs directly. Thus cgroup bpf_link can
co-exist with legacy BPF program multi-attachment.
bpf_link is destroyed and automatically detached when the last open FD holding
the reference to bpf_link is closed. This means that by default, when the
process that created bpf_link exits, attached BPF program will be
automatically detached due to bpf_link's clean up code. Cgroup bpf_link, like
any other bpf_link, can be pinned in BPF FS and by those means survive the
exit of process that created the link. This is useful in many scenarios to
provide long-living BPF program attachments. Pinning also means that there
could be many owners of bpf_link through independent FDs.
Additionally, auto-detachmet of cgroup bpf_link is implemented. When cgroup is
dying it will automatically detach all active bpf_links. This ensures that
cgroup clean up is not delayed due to active bpf_link even despite no chance
for any BPF program to be run for a given cgroup. In that sense it's similar
to existing behavior of dropping refcnt of attached bpf_prog. But in the case
of bpf_link, bpf_link is not destroyed and is still available to user as long
as at least one active FD is still open (or if it's pinned in BPF FS).
There are two main cgroup-specific differences between bpf_link-based and
direct bpf_prog-based attachment.
First, as opposed to direct bpf_prog attachment, cgroup itself doesn't "own"
bpf_link, which makes it possible to auto-clean up attached bpf_link when user
process abruptly exits without explicitly detaching BPF program. This makes
for a safe default behavior proven in BPF tracing program types. But bpf_link
doesn't bump cgroup->bpf.refcnt as well and because of that doesn't prevent
cgroup from cleaning up its BPF state.
Second, only owners of bpf_link (those who created bpf_link in the first place
or obtained a new FD by opening bpf_link from BPF FS) can detach and/or update
it. This makes sure that no other process can accidentally remove/replace BPF
program.
This patch set also implements LINK_UPDATE sub-command, which allows to
replace bpf_link's underlying bpf_prog, similarly to BPF_F_REPLACE flag
behavior for direct bpf_prog cgroup attachment. Similarly to LINK_CREATE, it
is supposed to be generic command for different types of bpf_links.
[0] https://lore.kernel.org/bpf/20200228223948.360936-1-andriin@fb.com/
v2->v3:
- revert back to just MULTI mode (Alexei);
- fix tinyconfig compilation warning (kbuild test robot);
v1->v2:
- implement exclusive and overridable exclusive modes (Andrey Ignatov);
- fix build for !CONFIG_CGROUP_BPF build;
- add more selftests for non-multi mode and inter-operability;
====================
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
|
|
Add selftests to exercise FD-based cgroup BPF program attachments and their
intermixing with legacy cgroup BPF attachments. Auto-detachment and program
replacement (both unconditional and cmpxchng-like) are tested as well.
Signed-off-by: Andrii Nakryiko <andriin@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20200330030001.2312810-5-andriin@fb.com
|
|
Add bpf_program__attach_cgroup(), which uses BPF_LINK_CREATE subcommand to
create an FD-based kernel bpf_link. Also add low-level bpf_link_create() API.
If expected_attach_type is not specified explicitly with
bpf_program__set_expected_attach_type(), libbpf will try to determine proper
attach type from BPF program's section definition.
Also add support for bpf_link's underlying BPF program replacement:
- unconditional through high-level bpf_link__update_program() API;
- cmpxchg-like with specifying expected current BPF program through
low-level bpf_link_update() API.
Signed-off-by: Andrii Nakryiko <andriin@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20200330030001.2312810-4-andriin@fb.com
|
|
Add new operation (LINK_UPDATE), which allows to replace active bpf_prog from
under given bpf_link. Currently this is only supported for bpf_cgroup_link,
but will be extended to other kinds of bpf_links in follow-up patches.
For bpf_cgroup_link, implemented functionality matches existing semantics for
direct bpf_prog attachment (including BPF_F_REPLACE flag). User can either
unconditionally set new bpf_prog regardless of which bpf_prog is currently
active under given bpf_link, or, optionally, can specify expected active
bpf_prog. If active bpf_prog doesn't match expected one, no changes are
performed, old bpf_link stays intact and attached, operation returns
a failure.
cgroup_bpf_replace() operation is resolving race between auto-detachment and
bpf_prog update in the same fashion as it's done for bpf_link detachment,
except in this case update has no way of succeeding because of target cgroup
marked as dying. So in this case error is returned.
Signed-off-by: Andrii Nakryiko <andriin@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20200330030001.2312810-3-andriin@fb.com
|
|
Implement new sub-command to attach cgroup BPF programs and return FD-based
bpf_link back on success. bpf_link, once attached to cgroup, cannot be
replaced, except by owner having its FD. Cgroup bpf_link supports only
BPF_F_ALLOW_MULTI semantics. Both link-based and prog-based BPF_F_ALLOW_MULTI
attachments can be freely intermixed.
To prevent bpf_cgroup_link from keeping cgroup alive past the point when no
BPF program can be executed, implement auto-detachment of link. When
cgroup_bpf_release() is called, all attached bpf_links are forced to release
cgroup refcounts, but they leave bpf_link otherwise active and allocated, as
well as still owning underlying bpf_prog. This is because user-space might
still have FDs open and active, so bpf_link as a user-referenced object can't
be freed yet. Once last active FD is closed, bpf_link will be freed and
underlying bpf_prog refcount will be dropped. But cgroup refcount won't be
touched, because cgroup is released already.
The inherent race between bpf_cgroup_link release (from closing last FD) and
cgroup_bpf_release() is resolved by both operations taking cgroup_mutex. So
the only additional check required is when bpf_cgroup_link attempts to detach
itself from cgroup. At that time we need to check whether there is still
cgroup associated with that link. And if not, exit with success, because
bpf_cgroup_link was already successfully detached.
Signed-off-by: Andrii Nakryiko <andriin@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Roman Gushchin <guro@fb.com>
Link: https://lore.kernel.org/bpf/20200330030001.2312810-2-andriin@fb.com
|
|
John Fastabend says:
====================
This series adds ALU32 signed and unsigned min/max bounds.
The origins of this work is to fix do_refine_retval_range() which before
this series clamps the return value bounds to [0, max]. However, this
is not correct because its possible these functions may return negative
errors so the correct bound is [*MIN, max]. Where *MIN is the signed
and unsigned min values U64_MIN and S64_MIN. And 'max' here is the max
positive value returned by this routine.
Patch 1 changes the do_refine_retval_range() to return the correct bounds
but this breaks existing programs that were depending on the old incorrect
bound. To repair these old programs we add ALU32 bounds to properly track
the return values from these helpers. The ALU32 bounds are needed because
clang realizes these helepers return 'int' type and will use jmp32 ops
with the return value. With current state of things this does little to
help 64bit bounds and with patch 1 applied will cause many programs to
fail verifier pass. See patch 5 for trace details on how this happens.
Patch 2 does the ALU32 addition it adds the new bounds and populates them
through the verifier. Design note, initially a var32 was added but as
pointed out by Alexei and Edward it is not strictly needed so it was
removed here. This worked out nicely.
Patch 3 notes that the refine return value can now also bound the 32-bit
subregister allowing better bouinds tracking in these cases.
Patches 4 adds a C test case to test_progs which will cause the verifier
to fail if new 32bit and do_refine_retval_range() is incorrect.
Patches 5 and 6 fix test cases that broke after refining the return
values from helpers. I attempted to be explicit about each failure and
why we need the change. See patches for details.
Patch 7 adds some bounds check tests to ensure bounds checking when
mixing alu32, alu64 and jmp32 ops together.
Thanks to Alexei, Edward, and Daniel for initial feedback it helped clean
this up a lot.
v2:
- rebased to bpf-next
- fixed tnum equals optimization for combining 32->64bits
- updated patch to fix verifier test correctly
- updated refine_retval_range to set both s32_*_value and s*_value we
need both to get better bounds tracking
====================
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
|
|
Its possible to have divergent ALU32 and ALU64 bounds when using JMP32
instructins and ALU64 arithmatic operations. Sometimes the clang will
even generate this code. Because the case is a bit tricky lets add
a specific test for it.
Here is pseudocode asm version to illustrate the idea,
1 r0 = 0xffffffff00000001;
2 if w0 > 1 goto %l[fail];
3 r0 += 1
5 if w0 > 2 goto %l[fail]
6 exit
The intent here is the verifier will fail the load if the 32bit bounds
are not tracked correctly through ALU64 op. Similarly we can check the
64bit bounds are correctly zero extended after ALU32 ops.
1 r0 = 0xffffffff00000001;
2 w0 += 1
2 if r0 > 3 goto %l[fail];
6 exit
The above will fail if we do not correctly zero extend 64bit bounds
after 32bit op.
Signed-off-by: John Fastabend <john.fastabend@gmail.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/158560430155.10843.514209255758200922.stgit@john-Precision-5820-Tower
|
|
After changes to add update_reg_bounds after ALU ops and 32-bit bounds
tracking truncation of boundary crossing range will fail earlier and with
a different error message. Now the test error trace is the following
11: (17) r1 -= 2147483584
12: R0_w=map_value(id=0,off=0,ks=8,vs=8,imm=0)
R1_w=invP(id=0,smin_value=-2147483584,smax_value=63)
R10=fp0 fp-8_w=mmmmmmmm
12: (17) r1 -= 2147483584
13: R0_w=map_value(id=0,off=0,ks=8,vs=8,imm=0)
R1_w=invP(id=0,
umin_value=18446744069414584448,umax_value=18446744071562068095,
var_off=(0xffffffff00000000; 0xffffffff))
R10=fp0 fp-8_w=mmmmmmmm
13: (77) r1 >>= 8
14: R0_w=map_value(id=0,off=0,ks=8,vs=8,imm=0)
R1_w=invP(id=0,
umin_value=72057594021150720,umax_value=72057594029539328,
var_off=(0xffffffff000000; 0xffffff),
s32_min_value=-16777216,s32_max_value=-1,
u32_min_value=-16777216)
R10=fp0 fp-8_w=mmmmmmmm
14: (0f) r0 += r1
value 72057594021150720 makes map_value pointer be out of bounds
Because we have 'umin_value == umax_value' instead of previously
where 'umin_value != umax_value' we can now fail earlier noting
that pointer addition is out of bounds.
Signed-off-by: John Fastabend <john.fastabend@gmail.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/158560428103.10843.6316594510312781186.stgit@john-Precision-5820-Tower
|
|
With current ALU32 subreg handling and retval refine fix from last
patches we see an expected failure in test_verifier. With verbose
verifier state being printed at each step for clarity we have the
following relavent lines [I omit register states that are not
necessarily useful to see failure cause],
#101/p bpf_get_stack return R0 within range FAIL
Failed to load prog 'Success'!
[..]
14: (85) call bpf_get_stack#67
R0_w=map_value(id=0,off=0,ks=8,vs=48,imm=0)
R3_w=inv48
15:
R0=inv(id=0,smax_value=48,var32_off=(0x0; 0xffffffff))
15: (b7) r1 = 0
16:
R0=inv(id=0,smax_value=48,var32_off=(0x0; 0xffffffff))
R1_w=inv0
16: (bf) r8 = r0
17:
R0=inv(id=0,smax_value=48,var32_off=(0x0; 0xffffffff))
R1_w=inv0
R8_w=inv(id=0,smax_value=48,var32_off=(0x0; 0xffffffff))
17: (67) r8 <<= 32
18:
R0=inv(id=0,smax_value=48,var32_off=(0x0; 0xffffffff))
R1_w=inv0
R8_w=inv(id=0,smax_value=9223372032559808512,
umax_value=18446744069414584320,
var_off=(0x0; 0xffffffff00000000),
s32_min_value=0,
s32_max_value=0,
u32_max_value=0,
var32_off=(0x0; 0x0))
18: (c7) r8 s>>= 32
19
R0=inv(id=0,smax_value=48,var32_off=(0x0; 0xffffffff))
R1_w=inv0
R8_w=inv(id=0,smin_value=-2147483648,
smax_value=2147483647,
var32_off=(0x0; 0xffffffff))
19: (cd) if r1 s< r8 goto pc+16
R0=inv(id=0,smax_value=48,var32_off=(0x0; 0xffffffff))
R1_w=inv0
R8_w=inv(id=0,smin_value=-2147483648,
smax_value=0,
var32_off=(0x0; 0xffffffff))
20:
R0=inv(id=0,smax_value=48,var32_off=(0x0; 0xffffffff))
R1_w=inv0
R8_w=inv(id=0,smin_value=-2147483648,
smax_value=0,
R9=inv48
20: (1f) r9 -= r8
21: (bf) r2 = r7
22:
R2_w=map_value(id=0,off=0,ks=8,vs=48,imm=0)
22: (0f) r2 += r8
value -2147483648 makes map_value pointer be out of bounds
After call bpf_get_stack() on line 14 and some moves we have at line 16
an r8 bound with max_value 48 but an unknown min value. This is to be
expected bpf_get_stack call can only return a max of the input size but
is free to return any negative error in the 32-bit register space. The
C helper is returning an int so will use lower 32-bits.
Lines 17 and 18 clear the top 32 bits with a left/right shift but use
ARSH so we still have worst case min bound before line 19 of -2147483648.
At this point the signed check 'r1 s< r8' meant to protect the addition
on line 22 where dst reg is a map_value pointer may very well return
true with a large negative number. Then the final line 22 will detect
this as an invalid operation and fail the program. What we want to do
is proceed only if r8 is positive non-error. So change 'r1 s< r8' to
'r1 s> r8' so that we jump if r8 is negative.
Next we will throw an error because we access past the end of the map
value. The map value size is 48 and sizeof(struct test_val) is 48 so
we walk off the end of the map value on the second call to
get bpf_get_stack(). Fix this by changing sizeof(struct test_val) to
24 by using 'sizeof(struct test_val) / 2'. After this everything passes
as expected.
Signed-off-by: John Fastabend <john.fastabend@gmail.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/158560426019.10843.3285429543232025187.stgit@john-Precision-5820-Tower
|
|
Before this series the verifier would clamp return bounds of
bpf_get_stack() to [0, X] and this led the verifier to believe
that a JMP_JSLT 0 would be false and so would prune that path.
The result is anything hidden behind that JSLT would be unverified.
Add a test to catch this case by hiding an goto pc-1 behind the
check which will cause an infinite loop if not rejected.
Signed-off-by: John Fastabend <john.fastabend@gmail.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/158560423908.10843.11783152347709008373.stgit@john-Precision-5820-Tower
|
|
Further refine return values range in do_refine_retval_range by noting
these are int return types (We will assume here that int is a 32-bit type).
Two reasons to pull this out of original patch. First it makes the original
fix impossible to backport. And second I've not seen this as being problematic
in practice unlike the other case.
Fixes: 849fa50662fbc ("bpf/verifier: refine retval R0 state for bpf_get_stack helper")
Signed-off-by: John Fastabend <john.fastabend@gmail.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/158560421952.10843.12496354931526965046.stgit@john-Precision-5820-Tower
|
|
It is not possible for the current verifier to track ALU32 and JMP ops
correctly. This can result in the verifier aborting with errors even though
the program should be verifiable. BPF codes that hit this can work around
it by changin int variables to 64-bit types, marking variables volatile,
etc. But this is all very ugly so it would be better to avoid these tricks.
But, the main reason to address this now is do_refine_retval_range() was
assuming return values could not be negative. Once we fixed this code that
was previously working will no longer work. See do_refine_retval_range()
patch for details. And we don't want to suddenly cause programs that used
to work to fail.
The simplest example code snippet that illustrates the problem is likely
this,
53: w8 = w0 // r8 <- [0, S32_MAX],
// w8 <- [-S32_MIN, X]
54: w8 <s 0 // r8 <- [0, U32_MAX]
// w8 <- [0, X]
The expected 64-bit and 32-bit bounds after each line are shown on the
right. The current issue is without the w* bounds we are forced to use
the worst case bound of [0, U32_MAX]. To resolve this type of case,
jmp32 creating divergent 32-bit bounds from 64-bit bounds, we add explicit
32-bit register bounds s32_{min|max}_value and u32_{min|max}_value. Then
from branch_taken logic creating new bounds we can track 32-bit bounds
explicitly.
The next case we observed is ALU ops after the jmp32,
53: w8 = w0 // r8 <- [0, S32_MAX],
// w8 <- [-S32_MIN, X]
54: w8 <s 0 // r8 <- [0, U32_MAX]
// w8 <- [0, X]
55: w8 += 1 // r8 <- [0, U32_MAX+1]
// w8 <- [0, X+1]
In order to keep the bounds accurate at this point we also need to track
ALU32 ops. To do this we add explicit ALU32 logic for each of the ALU
ops, mov, add, sub, etc.
Finally there is a question of how and when to merge bounds. The cases
enumerate here,
1. MOV ALU32 - zext 32-bit -> 64-bit
2. MOV ALU64 - copy 64-bit -> 32-bit
3. op ALU32 - zext 32-bit -> 64-bit
4. op ALU64 - n/a
5. jmp ALU32 - 64-bit: var32_off | upper_32_bits(var64_off)
6. jmp ALU64 - 32-bit: (>> (<< var64_off))
Details for each case,
For "MOV ALU32" BPF arch zero extends so we simply copy the bounds
from 32-bit into 64-bit ensuring we truncate var_off and 64-bit
bounds correctly. See zext_32_to_64.
For "MOV ALU64" copy all bounds including 32-bit into new register. If
the src register had 32-bit bounds the dst register will as well.
For "op ALU32" zero extend 32-bit into 64-bit the same as move,
see zext_32_to_64.
For "op ALU64" calculate both 32-bit and 64-bit bounds no merging
is done here. Except we have a special case. When RSH or ARSH is
done we can't simply ignore shifting bits from 64-bit reg into the
32-bit subreg. So currently just push bounds from 64-bit into 32-bit.
This will be correct in the sense that they will represent a valid
state of the register. However we could lose some accuracy if an
ARSH is following a jmp32 operation. We can handle this special
case in a follow up series.
For "jmp ALU32" mark 64-bit reg unknown and recalculate 64-bit bounds
from tnum by setting var_off to ((<<(>>var_off)) | var32_off). We
special case if 64-bit bounds has zero'd upper 32bits at which point
we can simply copy 32-bit bounds into 64-bit register. This catches
a common compiler trick where upper 32-bits are zeroed and then
32-bit ops are used followed by a 64-bit compare or 64-bit op on
a pointer. See __reg_combine_64_into_32().
For "jmp ALU64" cast the bounds of the 64bit to their 32-bit
counterpart. For example s32_min_value = (s32)reg->smin_value. For
tnum use only the lower 32bits via, (>>(<<var_off)). See
__reg_combine_64_into_32().
Signed-off-by: John Fastabend <john.fastabend@gmail.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/158560419880.10843.11448220440809118343.stgit@john-Precision-5820-Tower
|
|
do_refine_retval_range() is called to refine return values from specified
helpers, probe_read_str and get_stack at the moment, the reasoning is
because both have a max value as part of their input arguments and
because the helper ensure the return value will not be larger than this
we can set smax values of the return register, r0.
However, the return value is a signed integer so setting umax is incorrect
It leads to further confusion when the do_refine_retval_range() then calls,
__reg_deduce_bounds() which will see a umax value as meaning the value is
unsigned and then assuming it is unsigned set the smin = umin which in this
case results in 'smin = 0' and an 'smax = X' where X is the input argument
from the helper call.
Here are the comments from _reg_deduce_bounds() on why this would be safe
to do.
/* Learn sign from unsigned bounds. Signed bounds cross the sign
* boundary, so we must be careful.
*/
if ((s64)reg->umax_value >= 0) {
/* Positive. We can't learn anything from the smin, but smax
* is positive, hence safe.
*/
reg->smin_value = reg->umin_value;
reg->smax_value = reg->umax_value = min_t(u64, reg->smax_value,
reg->umax_value);
But now we incorrectly have a return value with type int with the
signed bounds (0,X). Suppose the return value is negative, which is
possible the we have the verifier and reality out of sync. Among other
things this may result in any error handling code being falsely detected
as dead-code and removed. For instance the example below shows using
bpf_probe_read_str() causes the error path to be identified as dead
code and removed.
>From the 'llvm-object -S' dump,
r2 = 100
call 45
if r0 s< 0 goto +4
r4 = *(u32 *)(r7 + 0)
But from dump xlate
(b7) r2 = 100
(85) call bpf_probe_read_compat_str#-96768
(61) r4 = *(u32 *)(r7 +0) <-- dropped if goto
Due to verifier state after call being
R0=inv(id=0,umax_value=100,var_off=(0x0; 0x7f))
To fix omit setting the umax value because its not safe. The only
actual bounds we know is the smax. This results in the correct bounds
(SMIN, X) where X is the max length from the helper. After this the
new verifier state looks like the following after call 45.
R0=inv(id=0,smax_value=100)
Then xlated version no longer removed dead code giving the expected
result,
(b7) r2 = 100
(85) call bpf_probe_read_compat_str#-96768
(c5) if r0 s< 0x0 goto pc+4
(61) r4 = *(u32 *)(r7 +0)
Note, bpf_probe_read_* calls are root only so we wont hit this case
with non-root bpf users.
v3: comment had some documentation about meta set to null case which
is not relevant here and confusing to include in the comment.
v2 note: In original version we set msize_smax_value from check_func_arg()
and propagated this into smax of retval. The logic was smax is the bound
on the retval we set and because the type in the helper is ARG_CONST_SIZE
we know that the reg is a positive tnum_const() so umax=smax. Alexei
pointed out though this is a bit odd to read because the register in
check_func_arg() has a C type of u32 and the umax bound would be the
normally relavent bound here. Pulling in extra knowledge about future
checks makes reading the code a bit tricky. Further having a signed
meta data that can only ever be positive is also a bit odd. So dropped
the msize_smax_value metadata and made it a u64 msize_max_value to
indicate its unsigned. And additionally save bound from umax value in
check_arg_funcs which is the same as smax due to as noted above tnumx_cont
and negative check but reads better. By my analysis nothing functionally
changes in v2 but it does get easier to read so that is win.
Fixes: 849fa50662fbc ("bpf/verifier: refine retval R0 state for bpf_get_stack helper")
Signed-off-by: John Fastabend <john.fastabend@gmail.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/158560417900.10843.14351995140624628941.stgit@john-Precision-5820-Tower
|
|
LSM and tracing programs share their helpers with bpf_tracing_func_proto
which is only defined (in bpf_trace.c) when BPF_EVENTS is enabled.
Instead of adding __weak symbol, make BPF_LSM depend on BPF_EVENTS so
that both tracing and LSM programs can actually share helpers.
Fixes: fc611f47f218 ("bpf: Introduce BPF_PROG_TYPE_LSM")
Reported-by: Randy Dunlap <rdunlap@infradead.org>
Signed-off-by: KP Singh <kpsingh@google.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Link: https://lore.kernel.org/bpf/20200330204059.13024-1-kpsingh@chromium.org
|
|
Joe Stringer says:
====================
Introduce a new helper that allows assigning a previously-found socket
to the skb as the packet is received towards the stack, to cause the
stack to guide the packet towards that socket subject to local routing
configuration. The intention is to support TProxy use cases more
directly from eBPF programs attached at TC ingress, to simplify and
streamline Linux stack configuration in scale environments with Cilium.
Normally in ip{,6}_rcv_core(), the skb will be orphaned, dropping any
existing socket reference associated with the skb. Existing tproxy
implementations in netfilter get around this restriction by running the
tproxy logic after ip_rcv_core() in the PREROUTING table. However, this
is not an option for TC-based logic (including eBPF programs attached at
TC ingress).
This series introduces the BPF helper bpf_sk_assign() to associate the
socket with the skb on the ingress path as the packet is passed up the
stack. The initial patch in the series simply takes a reference on the
socket to ensure safety, but later patches relax this for listen
sockets.
To ensure delivery to the relevant socket, we still consult the routing
table, for full examples of how to configure see the tests in patch #5;
the simplest form of the route would look like this:
$ ip route add local default dev lo
This series is laid out as follows:
* Patch 1 extends the eBPF API to add sk_assign() and defines a new
socket free function to allow the later paths to understand when the
socket associated with the skb should be kept through receive.
* Patches 2-3 optimize the receive path to avoid taking a reference on
listener sockets during receive.
* Patches 4-5 extends the selftests with examples of the new
functionality and validation of correct behaviour.
Changes since v4:
* Fix build with CONFIG_INET disabled
* Rebase
Changes since v3:
* Use sock_gen_put() directly instead of sock_edemux() from sock_pfree()
* Commit message wording fixups
* Add acks from Martin, Lorenz
* Rebase
Changes since v2:
* Add selftests for UDP socket redirection
* Drop the early demux optimization patch (defer for more testing)
* Fix check for orphaning after TC act return
* Tidy up the tests to clean up properly and be less noisy.
Changes since v1:
* Replace the metadata_dst approach with using the skb->destructor to
determine whether the socket has been prefetched. This is much
simpler.
* Avoid taking a reference on listener sockets during receive
* Restrict assigning sockets across namespaces
* Restrict assigning SO_REUSEPORT sockets
* Fix cookie usage for socket dst check
* Rebase the tests against test_progs infrastructure
* Tidy up commit messages
====================
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
|
|
Add support for testing UDP sk_assign to the existing tests.
Signed-off-by: Joe Stringer <joe@wand.net.nz>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Lorenz Bauer <lmb@cloudflare.com>
Acked-by: Martin KaFai Lau <kafai@fb.com>
Link: https://lore.kernel.org/bpf/20200329225342.16317-6-joe@wand.net.nz
|
|
Attach a tc direct-action classifier to lo in a fresh network
namespace, and rewrite all connection attempts to localhost:4321
to localhost:1234 (for port tests) and connections to unreachable
IPv4/IPv6 IPs to the local socket (for address tests). Includes
implementations for both TCP and UDP.
Keep in mind that both client to server and server to client traffic
passes the classifier.
Signed-off-by: Lorenz Bauer <lmb@cloudflare.com>
Signed-off-by: Joe Stringer <joe@wand.net.nz>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Martin KaFai Lau <kafai@fb.com>
Link: https://lore.kernel.org/bpf/20200329225342.16317-5-joe@wand.net.nz
Co-authored-by: Joe Stringer <joe@wand.net.nz>
|
|
Avoid taking a reference on listen sockets by checking the socket type
in the sk_assign and in the corresponding skb_steal_sock() code in the
the transport layer, and by ensuring that the prefetch free (sock_pfree)
function uses the same logic to check whether the socket is refcounted.
Suggested-by: Martin KaFai Lau <kafai@fb.com>
Signed-off-by: Joe Stringer <joe@wand.net.nz>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Martin KaFai Lau <kafai@fb.com>
Link: https://lore.kernel.org/bpf/20200329225342.16317-4-joe@wand.net.nz
|
|
Refactor the UDP/TCP handlers slightly to allow skb_steal_sock() to make
the determination of whether the socket is reference counted in the case
where it is prefetched by earlier logic such as early_demux.
Signed-off-by: Joe Stringer <joe@wand.net.nz>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Martin KaFai Lau <kafai@fb.com>
Link: https://lore.kernel.org/bpf/20200329225342.16317-3-joe@wand.net.nz
|
|
Add support for TPROXY via a new bpf helper, bpf_sk_assign().
This helper requires the BPF program to discover the socket via a call
to bpf_sk*_lookup_*(), then pass this socket to the new helper. The
helper takes its own reference to the socket in addition to any existing
reference that may or may not currently be obtained for the duration of
BPF processing. For the destination socket to receive the traffic, the
traffic must be routed towards that socket via local route. The
simplest example route is below, but in practice you may want to route
traffic more narrowly (eg by CIDR):
$ ip route add local default dev lo
This patch avoids trying to introduce an extra bit into the skb->sk, as
that would require more invasive changes to all code interacting with
the socket to ensure that the bit is handled correctly, such as all
error-handling cases along the path from the helper in BPF through to
the orphan path in the input. Instead, we opt to use the destructor
variable to switch on the prefetch of the socket.
Signed-off-by: Joe Stringer <joe@wand.net.nz>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Martin KaFai Lau <kafai@fb.com>
Link: https://lore.kernel.org/bpf/20200329225342.16317-2-joe@wand.net.nz
|
|
We've added John Fastabend to our weekly BPF patch review rotation over
last months now where he provided excellent and timely feedback on BPF
patches. Therefore, add him to the BPF core reviewer team to the MAINTAINERS
file to reflect that.
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: John Fastabend <john.fastabend@gmail.com>
Link: https://lore.kernel.org/bpf/0e9a74933b3f21f4c5b5a3bc7f8e900b39805639.1585556231.git.daniel@iogearbox.net
|
|
The bounds checking for the arguments accessed in the BPF program breaks
when the expected_attach_type is not BPF_TRACE_FEXIT, BPF_LSM_MAC or
BPF_MODIFY_RETURN resulting in no check being done for the default case
(the programs which do not receive the return value of the attached
function in its arguments) when the index of the argument being accessed
is equal to the number of arguments (nr_args).
This was a result of a misplaced "else if" block introduced by the
Commit 6ba43b761c41 ("bpf: Attachment verification for
BPF_MODIFY_RETURN")
Fixes: 6ba43b761c41 ("bpf: Attachment verification for BPF_MODIFY_RETURN")
Reported-by: Jann Horn <jannh@google.com>
Signed-off-by: KP Singh <kpsingh@google.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20200330144246.338-1-kpsingh@chromium.org
|
|
If outer_proto is not set, GCC warning as following:
In file included from net/netfilter/ipvs/ip_vs_core.c:52:
net/netfilter/ipvs/ip_vs_core.c: In function 'ip_vs_in_icmp':
include/net/ip_vs.h:233:4: warning: 'outer_proto' may be used uninitialized in this function [-Wmaybe-uninitialized]
233 | printk(KERN_DEBUG pr_fmt(msg), ##__VA_ARGS__); \
| ^~~~~~
net/netfilter/ipvs/ip_vs_core.c:1666:8: note: 'outer_proto' was declared here
1666 | char *outer_proto;
| ^~~~~~~~~~~
Fixes: 73348fed35d0 ("ipvs: optimize tunnel dumps for icmp errors")
Signed-off-by: Haishuang Yan <yanhaishuang@cmss.chinamobile.com>
Acked-by: Julian Anastasov <ja@ssi.bg>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
|
|
I got a problem on MIPS with Big-Endian is turned on: every time when
NF trying to change TCP MSS it returns because of new.v16 was greater
than old.v16. But real MSS was 1460 and my rule was like this:
add rule table chain tcp option maxseg size set 1400
And 1400 is lesser that 1460, not greater.
Later I founded that main causer is cast from u32 to __be16.
Debugging:
In example MSS = 1400(HEX: 0x578). Here is representation of each byte
like it is in memory by addresses from left to right(e.g. [0x0 0x1 0x2
0x3]). LE — Little-Endian system, BE — Big-Endian, left column is type.
LE BE
u32: [78 05 00 00] [00 00 05 78]
As you can see, u32 representation will be casted to u16 from different
half of 4-byte address range. But actually nf_tables uses registers and
store data of various size. Actually TCP MSS stored in 2 bytes. But
registers are still u32 in definition:
struct nft_regs {
union {
u32 data[20];
struct nft_verdict verdict;
};
};
So, access like regs->data[priv->sreg] exactly u32. So, according to
table presents above, per-byte representation of stored TCP MSS in
register will be:
LE BE
(u32)regs->data[]: [78 05 00 00] [05 78 00 00]
^^ ^^
We see that register uses just half of u32 and other 2 bytes may be
used for some another data. But in nft_exthdr_tcp_set_eval() it casted
just like u32 -> __be16:
new.v16 = src
But u32 overfill __be16, so it get 2 low bytes. For clarity draw
one more table(<xx xx> means that bytes will be used for cast).
LE BE
u32: [<78 05> 00 00] [00 00 <05 78>]
(u32)regs->data[]: [<78 05> 00 00] [05 78 <00 00>]
As you can see, for Little-Endian nothing changes, but for Big-endian we
take the wrong half. In my case there is some other data instead of
zeros, so new MSS was wrongly greater.
For shooting this bug I used solution for ports ranges. Applying of this
patch does not affect Little-Endian systems.
Signed-off-by: Sergey Marinkevich <sergey.marinkevich@eltex-co.ru>
Acked-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
|
|
reg_set_min_max_inv() contains exactly the same logic as reg_set_min_max(),
just flipped around. While this makes sense in a cBPF verifier (where ALU
operations are not symmetric), it does not make sense for eBPF.
Replace reg_set_min_max_inv() with a helper that flips the opcode around,
then lets reg_set_min_max() do the complicated work.
Signed-off-by: Jann Horn <jannh@google.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20200330160324.15259-4-daniel@iogearbox.net
|
|
The BPF verifier tried to track values based on 32-bit comparisons by
(ab)using the tnum state via 581738a681b6 ("bpf: Provide better register
bounds after jmp32 instructions"). The idea is that after a check like
this:
if ((u32)r0 > 3)
exit
We can't meaningfully constrain the arithmetic-range-based tracking, but
we can update the tnum state to (value=0,mask=0xffff'ffff'0000'0003).
However, the implementation from 581738a681b6 didn't compute the tnum
constraint based on the fixed operand, but instead derives it from the
arithmetic-range-based tracking. This means that after the following
sequence of operations:
if (r0 >= 0x1'0000'0001)
exit
if ((u32)r0 > 7)
exit
The verifier assumed that the lower half of r0 is in the range (0, 0)
and apply the tnum constraint (value=0,mask=0xffff'ffff'0000'0000) thus
causing the overall tnum to be (value=0,mask=0x1'0000'0000), which was
incorrect. Provide a fixed implementation.
Fixes: 581738a681b6 ("bpf: Provide better register bounds after jmp32 instructions")
Signed-off-by: Jann Horn <jannh@google.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20200330160324.15259-3-daniel@iogearbox.net
|
|
Anatoly has been fuzzing with kBdysch harness and reported a hang in
one of the outcomes:
0: (b7) r0 = 808464432
1: (7f) r0 >>= r0
2: (14) w0 -= 808464432
3: (07) r0 += 808464432
4: (b7) r1 = 808464432
5: (de) if w1 s<= w0 goto pc+0
R0_w=invP(id=0,umin_value=808464432,umax_value=5103431727,var_off=(0x30303020;0x10000001f)) R1_w=invP808464432 R10=fp0
6: (07) r0 += -2144337872
7: (14) w0 -= -1607454672
8: (25) if r0 > 0x30303030 goto pc+0
R0_w=invP(id=0,umin_value=271581184,umax_value=271581311,var_off=(0x10300000;0x7f)) R1_w=invP808464432 R10=fp0
9: (76) if w0 s>= 0x303030 goto pc+2
12: (95) exit
from 8 to 9: safe
from 5 to 6: R0_w=invP(id=0,umin_value=808464432,umax_value=5103431727,var_off=(0x30303020;0x10000001f)) R1_w=invP808464432 R10=fp0
6: (07) r0 += -2144337872
7: (14) w0 -= -1607454672
8: (25) if r0 > 0x30303030 goto pc+0
R0_w=invP(id=0,umin_value=271581184,umax_value=271581311,var_off=(0x10300000;0x7f)) R1_w=invP808464432 R10=fp0
9: safe
from 8 to 9: safe
verification time 589 usec
stack depth 0
processed 17 insns (limit 1000000) [...]
The underlying program was xlated as follows:
# bpftool p d x i 9
0: (b7) r0 = 808464432
1: (7f) r0 >>= r0
2: (14) w0 -= 808464432
3: (07) r0 += 808464432
4: (b7) r1 = 808464432
5: (de) if w1 s<= w0 goto pc+0
6: (07) r0 += -2144337872
7: (14) w0 -= -1607454672
8: (25) if r0 > 0x30303030 goto pc+0
9: (76) if w0 s>= 0x303030 goto pc+2
10: (05) goto pc-1
11: (05) goto pc-1
12: (95) exit
The verifier rewrote original instructions it recognized as dead code with
'goto pc-1', but reality differs from verifier simulation in that we're
actually able to trigger a hang due to hitting the 'goto pc-1' instructions.
Taking different examples to make the issue more obvious: in this example
we're probing bounds on a completely unknown scalar variable in r1:
[...]
5: R0_w=inv1 R1_w=inv(id=0) R10=fp0
5: (18) r2 = 0x4000000000
7: R0_w=inv1 R1_w=inv(id=0) R2_w=inv274877906944 R10=fp0
7: (18) r3 = 0x2000000000
9: R0_w=inv1 R1_w=inv(id=0) R2_w=inv274877906944 R3_w=inv137438953472 R10=fp0
9: (18) r4 = 0x400
11: R0_w=inv1 R1_w=inv(id=0) R2_w=inv274877906944 R3_w=inv137438953472 R4_w=inv1024 R10=fp0
11: (18) r5 = 0x200
13: R0_w=inv1 R1_w=inv(id=0) R2_w=inv274877906944 R3_w=inv137438953472 R4_w=inv1024 R5_w=inv512 R10=fp0
13: (2d) if r1 > r2 goto pc+4
R0_w=inv1 R1_w=inv(id=0,umax_value=274877906944,var_off=(0x0; 0x7fffffffff)) R2_w=inv274877906944 R3_w=inv137438953472 R4_w=inv1024 R5_w=inv512 R10=fp0
14: R0_w=inv1 R1_w=inv(id=0,umax_value=274877906944,var_off=(0x0; 0x7fffffffff)) R2_w=inv274877906944 R3_w=inv137438953472 R4_w=inv1024 R5_w=inv512 R10=fp0
14: (ad) if r1 < r3 goto pc+3
R0_w=inv1 R1_w=inv(id=0,umin_value=137438953472,umax_value=274877906944,var_off=(0x0; 0x7fffffffff)) R2_w=inv274877906944 R3_w=inv137438953472 R4_w=inv1024 R5_w=inv512 R10=fp0
15: R0=inv1 R1=inv(id=0,umin_value=137438953472,umax_value=274877906944,var_off=(0x0; 0x7fffffffff)) R2=inv274877906944 R3=inv137438953472 R4=inv1024 R5=inv512 R10=fp0
15: (2e) if w1 > w4 goto pc+2
R0=inv1 R1=inv(id=0,umin_value=137438953472,umax_value=274877906944,var_off=(0x0; 0x7f00000000)) R2=inv274877906944 R3=inv137438953472 R4=inv1024 R5=inv512 R10=fp0
16: R0=inv1 R1=inv(id=0,umin_value=137438953472,umax_value=274877906944,var_off=(0x0; 0x7f00000000)) R2=inv274877906944 R3=inv137438953472 R4=inv1024 R5=inv512 R10=fp0
16: (ae) if w1 < w5 goto pc+1
R0=inv1 R1=inv(id=0,umin_value=137438953472,umax_value=274877906944,var_off=(0x0; 0x7f00000000)) R2=inv274877906944 R3=inv137438953472 R4=inv1024 R5=inv512 R10=fp0
[...]
We're first probing lower/upper bounds via jmp64, later we do a similar
check via jmp32 and examine the resulting var_off there. After fall-through
in insn 14, we get the following bounded r1 with 0x7fffffffff unknown marked
bits in the variable section.
Thus, after knowing r1 <= 0x4000000000 and r1 >= 0x2000000000:
max: 0b100000000000000000000000000000000000000 / 0x4000000000
var: 0b111111111111111111111111111111111111111 / 0x7fffffffff
min: 0b010000000000000000000000000000000000000 / 0x2000000000
Now, in insn 15 and 16, we perform a similar probe with lower/upper bounds
in jmp32.
Thus, after knowing r1 <= 0x4000000000 and r1 >= 0x2000000000 and
w1 <= 0x400 and w1 >= 0x200:
max: 0b100000000000000000000000000000000000000 / 0x4000000000
var: 0b111111100000000000000000000000000000000 / 0x7f00000000
min: 0b010000000000000000000000000000000000000 / 0x2000000000
The lower/upper bounds haven't changed since they have high bits set in
u64 space and the jmp32 tests can only refine bounds in the low bits.
However, for the var part the expectation would have been 0x7f000007ff
or something less precise up to 0x7fffffffff. A outcome of 0x7f00000000
is not correct since it would contradict the earlier probed bounds
where we know that the result should have been in [0x200,0x400] in u32
space. Therefore, tests with such info will lead to wrong verifier
assumptions later on like falsely predicting conditional jumps to be
always taken, etc.
The issue here is that __reg_bound_offset32()'s implementation from
commit 581738a681b6 ("bpf: Provide better register bounds after jmp32
instructions") makes an incorrect range assumption:
static void __reg_bound_offset32(struct bpf_reg_state *reg)
{
u64 mask = 0xffffFFFF;
struct tnum range = tnum_range(reg->umin_value & mask,
reg->umax_value & mask);
struct tnum lo32 = tnum_cast(reg->var_off, 4);
struct tnum hi32 = tnum_lshift(tnum_rshift(reg->var_off, 32), 32);
reg->var_off = tnum_or(hi32, tnum_intersect(lo32, range));
}
In the above walk-through example, __reg_bound_offset32() as-is chose
a range after masking with 0xffffffff of [0x0,0x0] since umin:0x2000000000
and umax:0x4000000000 and therefore the lo32 part was clamped to 0x0 as
well. However, in the umin:0x2000000000 and umax:0x4000000000 range above
we'd end up with an actual possible interval of [0x0,0xffffffff] for u32
space instead.
In case of the original reproducer, the situation looked as follows at
insn 5 for r0:
[...]
5: R0_w=invP(id=0,umin_value=808464432,umax_value=5103431727,var_off=(0x0; 0x1ffffffff)) R1_w=invP808464432 R10=fp0
0x30303030 0x13030302f
5: (de) if w1 s<= w0 goto pc+0
R0_w=invP(id=0,umin_value=808464432,umax_value=5103431727,var_off=(0x30303020; 0x10000001f)) R1_w=invP808464432 R10=fp0
0x30303030 0x13030302f
[...]
After the fall-through, we similarly forced the var_off result into
the wrong range [0x30303030,0x3030302f] suggesting later on that fixed
bits must only be of 0x30303020 with 0x10000001f unknowns whereas such
assumption can only be made when both bounds in hi32 range match.
Originally, I was thinking to fix this by moving reg into a temp reg and
use proper coerce_reg_to_size() helper on the temp reg where we can then
based on that define the range tnum for later intersection:
static void __reg_bound_offset32(struct bpf_reg_state *reg)
{
struct bpf_reg_state tmp = *reg;
struct tnum lo32, hi32, range;
coerce_reg_to_size(&tmp, 4);
range = tnum_range(tmp.umin_value, tmp.umax_value);
lo32 = tnum_cast(reg->var_off, 4);
hi32 = tnum_lshift(tnum_rshift(reg->var_off, 32), 32);
reg->var_off = tnum_or(hi32, tnum_intersect(lo32, range));
}
In the case of the concrete example, this gives us a more conservative unknown
section. Thus, after knowing r1 <= 0x4000000000 and r1 >= 0x2000000000 and
w1 <= 0x400 and w1 >= 0x200:
max: 0b100000000000000000000000000000000000000 / 0x4000000000
var: 0b111111111111111111111111111111111111111 / 0x7fffffffff
min: 0b010000000000000000000000000000000000000 / 0x2000000000
However, above new __reg_bound_offset32() has no effect on refining the
knowledge of the register contents. Meaning, if the bounds in hi32 range
mismatch we'll get the identity function given the range reg spans
[0x0,0xffffffff] and we cast var_off into lo32 only to later on binary
or it again with the hi32.
Likewise, if the bounds in hi32 range match, then we mask both bounds
with 0xffffffff, use the resulting umin/umax for the range to later
intersect the lo32 with it. However, _prior_ called __reg_bound_offset()
did already such intersection on the full reg and we therefore would only
repeat the same operation on the lo32 part twice.
Given this has no effect and the original commit had false assumptions,
this patch reverts the code entirely which is also more straight forward
for stable trees: apparently 581738a681b6 got auto-selected by Sasha's
ML system and misclassified as a fix, so it got sucked into v5.4 where
it should never have landed. A revert is low-risk also from a user PoV
since it requires a recent kernel and llc to opt-into -mcpu=v3 BPF CPU
to generate jmp32 instructions. A proper bounds refinement would need a
significantly more complex approach which is currently being worked, but
no stable material [0]. Hence revert is best option for stable. After the
revert, the original reported program gets rejected as follows:
1: (7f) r0 >>= r0
2: (14) w0 -= 808464432
3: (07) r0 += 808464432
4: (b7) r1 = 808464432
5: (de) if w1 s<= w0 goto pc+0
R0_w=invP(id=0,umin_value=808464432,umax_value=5103431727,var_off=(0x0; 0x1ffffffff)) R1_w=invP808464432 R10=fp0
6: (07) r0 += -2144337872
7: (14) w0 -= -1607454672
8: (25) if r0 > 0x30303030 goto pc+0
R0_w=invP(id=0,umax_value=808464432,var_off=(0x0; 0x3fffffff)) R1_w=invP808464432 R10=fp0
9: (76) if w0 s>= 0x303030 goto pc+2
R0=invP(id=0,umax_value=3158063,var_off=(0x0; 0x3fffff)) R1=invP808464432 R10=fp0
10: (30) r0 = *(u8 *)skb[808464432]
BPF_LD_[ABS|IND] uses reserved fields
processed 11 insns (limit 1000000) [...]
[0] https://lore.kernel.org/bpf/158507130343.15666.8018068546764556975.stgit@john-Precision-5820-Tower/T/
Fixes: 581738a681b6 ("bpf: Provide better register bounds after jmp32 instructions")
Reported-by: Anatoly Trosinenko <anatoly.trosinenko@gmail.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20200330160324.15259-2-daniel@iogearbox.net
|
|
Russell King says:
====================
split phylink PCS operations
This series splits the phylink_mac_ops structure so that PCS can be
supported separately with their own PCS operations, separating them
from the MAC layer. This may need adaption later as more users come
along.
v2: change pcs_config() and associated called function prototypes to
only pass the information that is required, and add some documention.
v3: change phylink_create() prototype
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Add a separate set of PCS operations, which MAC drivers can use to
couple phylink with their associated MAC PCS layer. The PCS
operations include:
- pcs_get_state() - reads the link up/down, resolved speed, duplex
and pause from the PCS.
- pcs_config() - configures the PCS for the specified mode, PHY
interface type, and setting the advertisement.
- pcs_an_restart() - restarts 802.3 in-band negotiation with the
link partner
- pcs_link_up() - informs the PCS that link has come up, and the
parameters of the link. Link parameters are used to program the
PCS for fixed speed and non-inband modes.
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: Russell King <rmk+kernel@armlinux.org.uk>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Rename the bland 'ops' member of struct phylink to be a more
descriptive 'mac_ops' - this is necessary as we're about to introduce
another set of operations.
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: Russell King <rmk+kernel@armlinux.org.uk>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
|
Change phylink_mii_c22_pcs_set_advertisement() to take only the PHY
interface and advertisement mask, rather than the full phylink state.
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: Russell King <rmk+kernel@armlinux.org.uk>
Signed-off-by: David S. Miller <davem@davemloft.net>
|
