2019-05-29 01:10:09 +08:00
|
|
|
/* SPDX-License-Identifier: GPL-2.0-only */
|
2016-09-21 18:43:57 +08:00
|
|
|
/* Copyright (c) 2011-2014 PLUMgrid, http://plumgrid.com
|
|
|
|
|
*/
|
|
|
|
|
#ifndef _LINUX_BPF_VERIFIER_H
|
|
|
|
|
#define _LINUX_BPF_VERIFIER_H 1
|
|
|
|
|
|
|
|
|
|
#include <linux/bpf.h> /* for enum bpf_reg_type */
|
|
|
|
|
#include <linux/filter.h> /* for MAX_BPF_STACK */
|
2017-08-07 22:26:19 +08:00
|
|
|
#include <linux/tnum.h>
|
2016-09-21 18:43:57 +08:00
|
|
|
|
2017-08-07 22:26:36 +08:00
|
|
|
/* Maximum variable offset umax_value permitted when resolving memory accesses.
|
|
|
|
|
* In practice this is far bigger than any realistic pointer offset; this limit
|
|
|
|
|
* ensures that umax_value + (int)off + (int)size cannot overflow a u64.
|
|
|
|
|
*/
|
2017-12-19 12:12:00 +08:00
|
|
|
#define BPF_MAX_VAR_OFF (1 << 29)
|
2017-08-07 22:26:36 +08:00
|
|
|
/* Maximum variable size permitted for ARG_CONST_SIZE[_OR_ZERO]. This ensures
|
|
|
|
|
* that converting umax_value to int cannot overflow.
|
|
|
|
|
*/
|
2017-12-19 12:12:00 +08:00
|
|
|
#define BPF_MAX_VAR_SIZ (1 << 29)
|
2016-09-28 22:54:32 +08:00
|
|
|
|
2017-08-23 22:11:21 +08:00
|
|
|
/* Liveness marks, used for registers and spilled-regs (in stack slots).
|
|
|
|
|
* Read marks propagate upwards until they find a write mark; they record that
|
|
|
|
|
* "one of this state's descendants read this reg" (and therefore the reg is
|
|
|
|
|
* relevant for states_equal() checks).
|
|
|
|
|
* Write marks collect downwards and do not propagate; they record that "the
|
|
|
|
|
* straight-line code that reached this state (from its parent) wrote this reg"
|
|
|
|
|
* (and therefore that reads propagated from this state or its descendants
|
|
|
|
|
* should not propagate to its parent).
|
|
|
|
|
* A state with a write mark can receive read marks; it just won't propagate
|
|
|
|
|
* them to its parent, since the write mark is a property, not of the state,
|
|
|
|
|
* but of the link between it and its parent. See mark_reg_read() and
|
|
|
|
|
* mark_stack_slot_read() in kernel/bpf/verifier.c.
|
|
|
|
|
*/
|
2017-08-16 03:34:35 +08:00
|
|
|
enum bpf_reg_liveness {
|
|
|
|
|
REG_LIVE_NONE = 0, /* reg hasn't been read or written this branch */
|
bpf: verifier: mark verified-insn with sub-register zext flag
eBPF ISA specification requires high 32-bit cleared when low 32-bit
sub-register is written. This applies to destination register of ALU32 etc.
JIT back-ends must guarantee this semantic when doing code-gen. x86_64 and
AArch64 ISA has the same semantics, so the corresponding JIT back-end
doesn't need to do extra work.
However, 32-bit arches (arm, x86, nfp etc.) and some other 64-bit arches
(PowerPC, SPARC etc) need to do explicit zero extension to meet this
requirement, otherwise code like the following will fail.
u64_value = (u64) u32_value
... other uses of u64_value
This is because compiler could exploit the semantic described above and
save those zero extensions for extending u32_value to u64_value, these JIT
back-ends are expected to guarantee this through inserting extra zero
extensions which however could be a significant increase on the code size.
Some benchmarks show there could be ~40% sub-register writes out of total
insns, meaning at least ~40% extra code-gen.
One observation is these extra zero extensions are not always necessary.
Take above code snippet for example, it is possible u32_value will never be
casted into a u64, the value of high 32-bit of u32_value then could be
ignored and extra zero extension could be eliminated.
This patch implements this idea, insns defining sub-registers will be
marked when the high 32-bit of the defined sub-register matters. For
those unmarked insns, it is safe to eliminate high 32-bit clearnace for
them.
Algo:
- Split read flags into READ32 and READ64.
- Record index of insn that does sub-register write. Keep the index inside
reg state and update it during verifier insn walking.
- A full register read on a sub-register marks its definition insn as
needing zero extension on dst register.
A new sub-register write overrides the old one.
- When propagating read64 during path pruning, also mark any insn defining
a sub-register that is read in the pruned path as full-register.
Reviewed-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Signed-off-by: Jiong Wang <jiong.wang@netronome.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2019-05-25 06:25:12 +08:00
|
|
|
REG_LIVE_READ32 = 0x1, /* reg was read, so we're sensitive to initial value */
|
|
|
|
|
REG_LIVE_READ64 = 0x2, /* likewise, but full 64-bit content matters */
|
|
|
|
|
REG_LIVE_READ = REG_LIVE_READ32 | REG_LIVE_READ64,
|
|
|
|
|
REG_LIVE_WRITTEN = 0x4, /* reg was written first, screening off later reads */
|
|
|
|
|
REG_LIVE_DONE = 0x8, /* liveness won't be updating this register anymore */
|
2017-08-16 03:34:35 +08:00
|
|
|
};
|
|
|
|
|
|
2016-09-21 18:43:57 +08:00
|
|
|
struct bpf_reg_state {
|
2018-08-23 03:02:19 +08:00
|
|
|
/* Ordering of fields matters. See states_equal() */
|
2016-09-21 18:43:57 +08:00
|
|
|
enum bpf_reg_type type;
|
|
|
|
|
union {
|
2017-08-07 22:26:19 +08:00
|
|
|
/* valid when type == PTR_TO_PACKET */
|
|
|
|
|
u16 range;
|
2016-09-21 18:43:57 +08:00
|
|
|
|
|
|
|
|
/* valid when type == CONST_PTR_TO_MAP | PTR_TO_MAP_VALUE |
|
|
|
|
|
* PTR_TO_MAP_VALUE_OR_NULL
|
|
|
|
|
*/
|
|
|
|
|
struct bpf_map *map_ptr;
|
2018-11-01 07:05:52 +08:00
|
|
|
|
2019-10-16 11:25:00 +08:00
|
|
|
u32 btf_id; /* for PTR_TO_BTF_ID */
|
|
|
|
|
|
bpf: Implement BPF ring buffer and verifier support for it
This commit adds a new MPSC ring buffer implementation into BPF ecosystem,
which allows multiple CPUs to submit data to a single shared ring buffer. On
the consumption side, only single consumer is assumed.
Motivation
----------
There are two distinctive motivators for this work, which are not satisfied by
existing perf buffer, which prompted creation of a new ring buffer
implementation.
- more efficient memory utilization by sharing ring buffer across CPUs;
- preserving ordering of events that happen sequentially in time, even
across multiple CPUs (e.g., fork/exec/exit events for a task).
These two problems are independent, but perf buffer fails to satisfy both.
Both are a result of a choice to have per-CPU perf ring buffer. Both can be
also solved by having an MPSC implementation of ring buffer. The ordering
problem could technically be solved for perf buffer with some in-kernel
counting, but given the first one requires an MPSC buffer, the same solution
would solve the second problem automatically.
Semantics and APIs
------------------
Single ring buffer is presented to BPF programs as an instance of BPF map of
type BPF_MAP_TYPE_RINGBUF. Two other alternatives considered, but ultimately
rejected.
One way would be to, similar to BPF_MAP_TYPE_PERF_EVENT_ARRAY, make
BPF_MAP_TYPE_RINGBUF could represent an array of ring buffers, but not enforce
"same CPU only" rule. This would be more familiar interface compatible with
existing perf buffer use in BPF, but would fail if application needed more
advanced logic to lookup ring buffer by arbitrary key. HASH_OF_MAPS addresses
this with current approach. Additionally, given the performance of BPF
ringbuf, many use cases would just opt into a simple single ring buffer shared
among all CPUs, for which current approach would be an overkill.
Another approach could introduce a new concept, alongside BPF map, to
represent generic "container" object, which doesn't necessarily have key/value
interface with lookup/update/delete operations. This approach would add a lot
of extra infrastructure that has to be built for observability and verifier
support. It would also add another concept that BPF developers would have to
familiarize themselves with, new syntax in libbpf, etc. But then would really
provide no additional benefits over the approach of using a map.
BPF_MAP_TYPE_RINGBUF doesn't support lookup/update/delete operations, but so
doesn't few other map types (e.g., queue and stack; array doesn't support
delete, etc).
The approach chosen has an advantage of re-using existing BPF map
infrastructure (introspection APIs in kernel, libbpf support, etc), being
familiar concept (no need to teach users a new type of object in BPF program),
and utilizing existing tooling (bpftool). For common scenario of using
a single ring buffer for all CPUs, it's as simple and straightforward, as
would be with a dedicated "container" object. On the other hand, by being
a map, it can be combined with ARRAY_OF_MAPS and HASH_OF_MAPS map-in-maps to
implement a wide variety of topologies, from one ring buffer for each CPU
(e.g., as a replacement for perf buffer use cases), to a complicated
application hashing/sharding of ring buffers (e.g., having a small pool of
ring buffers with hashed task's tgid being a look up key to preserve order,
but reduce contention).
Key and value sizes are enforced to be zero. max_entries is used to specify
the size of ring buffer and has to be a power of 2 value.
There are a bunch of similarities between perf buffer
(BPF_MAP_TYPE_PERF_EVENT_ARRAY) and new BPF ring buffer semantics:
- variable-length records;
- if there is no more space left in ring buffer, reservation fails, no
blocking;
- memory-mappable data area for user-space applications for ease of
consumption and high performance;
- epoll notifications for new incoming data;
- but still the ability to do busy polling for new data to achieve the
lowest latency, if necessary.
BPF ringbuf provides two sets of APIs to BPF programs:
- bpf_ringbuf_output() allows to *copy* data from one place to a ring
buffer, similarly to bpf_perf_event_output();
- bpf_ringbuf_reserve()/bpf_ringbuf_commit()/bpf_ringbuf_discard() APIs
split the whole process into two steps. First, a fixed amount of space is
reserved. If successful, a pointer to a data inside ring buffer data area
is returned, which BPF programs can use similarly to a data inside
array/hash maps. Once ready, this piece of memory is either committed or
discarded. Discard is similar to commit, but makes consumer ignore the
record.
bpf_ringbuf_output() has disadvantage of incurring extra memory copy, because
record has to be prepared in some other place first. But it allows to submit
records of the length that's not known to verifier beforehand. It also closely
matches bpf_perf_event_output(), so will simplify migration significantly.
bpf_ringbuf_reserve() avoids the extra copy of memory by providing a memory
pointer directly to ring buffer memory. In a lot of cases records are larger
than BPF stack space allows, so many programs have use extra per-CPU array as
a temporary heap for preparing sample. bpf_ringbuf_reserve() avoid this needs
completely. But in exchange, it only allows a known constant size of memory to
be reserved, such that verifier can verify that BPF program can't access
memory outside its reserved record space. bpf_ringbuf_output(), while slightly
slower due to extra memory copy, covers some use cases that are not suitable
for bpf_ringbuf_reserve().
The difference between commit and discard is very small. Discard just marks
a record as discarded, and such records are supposed to be ignored by consumer
code. Discard is useful for some advanced use-cases, such as ensuring
all-or-nothing multi-record submission, or emulating temporary malloc()/free()
within single BPF program invocation.
Each reserved record is tracked by verifier through existing
reference-tracking logic, similar to socket ref-tracking. It is thus
impossible to reserve a record, but forget to submit (or discard) it.
bpf_ringbuf_query() helper allows to query various properties of ring buffer.
Currently 4 are supported:
- BPF_RB_AVAIL_DATA returns amount of unconsumed data in ring buffer;
- BPF_RB_RING_SIZE returns the size of ring buffer;
- BPF_RB_CONS_POS/BPF_RB_PROD_POS returns current logical possition of
consumer/producer, respectively.
Returned values are momentarily snapshots of ring buffer state and could be
off by the time helper returns, so this should be used only for
debugging/reporting reasons or for implementing various heuristics, that take
into account highly-changeable nature of some of those characteristics.
One such heuristic might involve more fine-grained control over poll/epoll
notifications about new data availability in ring buffer. Together with
BPF_RB_NO_WAKEUP/BPF_RB_FORCE_WAKEUP flags for output/commit/discard helpers,
it allows BPF program a high degree of control and, e.g., more efficient
batched notifications. Default self-balancing strategy, though, should be
adequate for most applications and will work reliable and efficiently already.
Design and implementation
-------------------------
This reserve/commit schema allows a natural way for multiple producers, either
on different CPUs or even on the same CPU/in the same BPF program, to reserve
independent records and work with them without blocking other producers. This
means that if BPF program was interruped by another BPF program sharing the
same ring buffer, they will both get a record reserved (provided there is
enough space left) and can work with it and submit it independently. This
applies to NMI context as well, except that due to using a spinlock during
reservation, in NMI context, bpf_ringbuf_reserve() might fail to get a lock,
in which case reservation will fail even if ring buffer is not full.
The ring buffer itself internally is implemented as a power-of-2 sized
circular buffer, with two logical and ever-increasing counters (which might
wrap around on 32-bit architectures, that's not a problem):
- consumer counter shows up to which logical position consumer consumed the
data;
- producer counter denotes amount of data reserved by all producers.
Each time a record is reserved, producer that "owns" the record will
successfully advance producer counter. At that point, data is still not yet
ready to be consumed, though. Each record has 8 byte header, which contains
the length of reserved record, as well as two extra bits: busy bit to denote
that record is still being worked on, and discard bit, which might be set at
commit time if record is discarded. In the latter case, consumer is supposed
to skip the record and move on to the next one. Record header also encodes
record's relative offset from the beginning of ring buffer data area (in
pages). This allows bpf_ringbuf_commit()/bpf_ringbuf_discard() to accept only
the pointer to the record itself, without requiring also the pointer to ring
buffer itself. Ring buffer memory location will be restored from record
metadata header. This significantly simplifies verifier, as well as improving
API usability.
Producer counter increments are serialized under spinlock, so there is
a strict ordering between reservations. Commits, on the other hand, are
completely lockless and independent. All records become available to consumer
in the order of reservations, but only after all previous records where
already committed. It is thus possible for slow producers to temporarily hold
off submitted records, that were reserved later.
Reservation/commit/consumer protocol is verified by litmus tests in
Documentation/litmus-test/bpf-rb.
One interesting implementation bit, that significantly simplifies (and thus
speeds up as well) implementation of both producers and consumers is how data
area is mapped twice contiguously back-to-back in the virtual memory. This
allows to not take any special measures for samples that have to wrap around
at the end of the circular buffer data area, because the next page after the
last data page would be first data page again, and thus the sample will still
appear completely contiguous in virtual memory. See comment and a simple ASCII
diagram showing this visually in bpf_ringbuf_area_alloc().
Another feature that distinguishes BPF ringbuf from perf ring buffer is
a self-pacing notifications of new data being availability.
bpf_ringbuf_commit() implementation will send a notification of new record
being available after commit only if consumer has already caught up right up
to the record being committed. If not, consumer still has to catch up and thus
will see new data anyways without needing an extra poll notification.
Benchmarks (see tools/testing/selftests/bpf/benchs/bench_ringbuf.c) show that
this allows to achieve a very high throughput without having to resort to
tricks like "notify only every Nth sample", which are necessary with perf
buffer. For extreme cases, when BPF program wants more manual control of
notifications, commit/discard/output helpers accept BPF_RB_NO_WAKEUP and
BPF_RB_FORCE_WAKEUP flags, which give full control over notifications of data
availability, but require extra caution and diligence in using this API.
Comparison to alternatives
--------------------------
Before considering implementing BPF ring buffer from scratch existing
alternatives in kernel were evaluated, but didn't seem to meet the needs. They
largely fell into few categores:
- per-CPU buffers (perf, ftrace, etc), which don't satisfy two motivations
outlined above (ordering and memory consumption);
- linked list-based implementations; while some were multi-producer designs,
consuming these from user-space would be very complicated and most
probably not performant; memory-mapping contiguous piece of memory is
simpler and more performant for user-space consumers;
- io_uring is SPSC, but also requires fixed-sized elements. Naively turning
SPSC queue into MPSC w/ lock would have subpar performance compared to
locked reserve + lockless commit, as with BPF ring buffer. Fixed sized
elements would be too limiting for BPF programs, given existing BPF
programs heavily rely on variable-sized perf buffer already;
- specialized implementations (like a new printk ring buffer, [0]) with lots
of printk-specific limitations and implications, that didn't seem to fit
well for intended use with BPF programs.
[0] https://lwn.net/Articles/779550/
Signed-off-by: Andrii Nakryiko <andriin@fb.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Link: https://lore.kernel.org/bpf/20200529075424.3139988-2-andriin@fb.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2020-05-29 15:54:20 +08:00
|
|
|
u32 mem_size; /* for PTR_TO_MEM | PTR_TO_MEM_OR_NULL */
|
|
|
|
|
|
2018-11-01 07:05:52 +08:00
|
|
|
/* Max size from any of the above. */
|
|
|
|
|
unsigned long raw;
|
2016-09-21 18:43:57 +08:00
|
|
|
};
|
2017-08-07 22:26:19 +08:00
|
|
|
/* Fixed part of pointer offset, pointer types only */
|
|
|
|
|
s32 off;
|
|
|
|
|
/* For PTR_TO_PACKET, used to find other pointers with the same variable
|
|
|
|
|
* offset, so they can share range knowledge.
|
|
|
|
|
* For PTR_TO_MAP_VALUE_OR_NULL this is used to share which map value we
|
|
|
|
|
* came from, when one is tested for != NULL.
|
bpf: Implement BPF ring buffer and verifier support for it
This commit adds a new MPSC ring buffer implementation into BPF ecosystem,
which allows multiple CPUs to submit data to a single shared ring buffer. On
the consumption side, only single consumer is assumed.
Motivation
----------
There are two distinctive motivators for this work, which are not satisfied by
existing perf buffer, which prompted creation of a new ring buffer
implementation.
- more efficient memory utilization by sharing ring buffer across CPUs;
- preserving ordering of events that happen sequentially in time, even
across multiple CPUs (e.g., fork/exec/exit events for a task).
These two problems are independent, but perf buffer fails to satisfy both.
Both are a result of a choice to have per-CPU perf ring buffer. Both can be
also solved by having an MPSC implementation of ring buffer. The ordering
problem could technically be solved for perf buffer with some in-kernel
counting, but given the first one requires an MPSC buffer, the same solution
would solve the second problem automatically.
Semantics and APIs
------------------
Single ring buffer is presented to BPF programs as an instance of BPF map of
type BPF_MAP_TYPE_RINGBUF. Two other alternatives considered, but ultimately
rejected.
One way would be to, similar to BPF_MAP_TYPE_PERF_EVENT_ARRAY, make
BPF_MAP_TYPE_RINGBUF could represent an array of ring buffers, but not enforce
"same CPU only" rule. This would be more familiar interface compatible with
existing perf buffer use in BPF, but would fail if application needed more
advanced logic to lookup ring buffer by arbitrary key. HASH_OF_MAPS addresses
this with current approach. Additionally, given the performance of BPF
ringbuf, many use cases would just opt into a simple single ring buffer shared
among all CPUs, for which current approach would be an overkill.
Another approach could introduce a new concept, alongside BPF map, to
represent generic "container" object, which doesn't necessarily have key/value
interface with lookup/update/delete operations. This approach would add a lot
of extra infrastructure that has to be built for observability and verifier
support. It would also add another concept that BPF developers would have to
familiarize themselves with, new syntax in libbpf, etc. But then would really
provide no additional benefits over the approach of using a map.
BPF_MAP_TYPE_RINGBUF doesn't support lookup/update/delete operations, but so
doesn't few other map types (e.g., queue and stack; array doesn't support
delete, etc).
The approach chosen has an advantage of re-using existing BPF map
infrastructure (introspection APIs in kernel, libbpf support, etc), being
familiar concept (no need to teach users a new type of object in BPF program),
and utilizing existing tooling (bpftool). For common scenario of using
a single ring buffer for all CPUs, it's as simple and straightforward, as
would be with a dedicated "container" object. On the other hand, by being
a map, it can be combined with ARRAY_OF_MAPS and HASH_OF_MAPS map-in-maps to
implement a wide variety of topologies, from one ring buffer for each CPU
(e.g., as a replacement for perf buffer use cases), to a complicated
application hashing/sharding of ring buffers (e.g., having a small pool of
ring buffers with hashed task's tgid being a look up key to preserve order,
but reduce contention).
Key and value sizes are enforced to be zero. max_entries is used to specify
the size of ring buffer and has to be a power of 2 value.
There are a bunch of similarities between perf buffer
(BPF_MAP_TYPE_PERF_EVENT_ARRAY) and new BPF ring buffer semantics:
- variable-length records;
- if there is no more space left in ring buffer, reservation fails, no
blocking;
- memory-mappable data area for user-space applications for ease of
consumption and high performance;
- epoll notifications for new incoming data;
- but still the ability to do busy polling for new data to achieve the
lowest latency, if necessary.
BPF ringbuf provides two sets of APIs to BPF programs:
- bpf_ringbuf_output() allows to *copy* data from one place to a ring
buffer, similarly to bpf_perf_event_output();
- bpf_ringbuf_reserve()/bpf_ringbuf_commit()/bpf_ringbuf_discard() APIs
split the whole process into two steps. First, a fixed amount of space is
reserved. If successful, a pointer to a data inside ring buffer data area
is returned, which BPF programs can use similarly to a data inside
array/hash maps. Once ready, this piece of memory is either committed or
discarded. Discard is similar to commit, but makes consumer ignore the
record.
bpf_ringbuf_output() has disadvantage of incurring extra memory copy, because
record has to be prepared in some other place first. But it allows to submit
records of the length that's not known to verifier beforehand. It also closely
matches bpf_perf_event_output(), so will simplify migration significantly.
bpf_ringbuf_reserve() avoids the extra copy of memory by providing a memory
pointer directly to ring buffer memory. In a lot of cases records are larger
than BPF stack space allows, so many programs have use extra per-CPU array as
a temporary heap for preparing sample. bpf_ringbuf_reserve() avoid this needs
completely. But in exchange, it only allows a known constant size of memory to
be reserved, such that verifier can verify that BPF program can't access
memory outside its reserved record space. bpf_ringbuf_output(), while slightly
slower due to extra memory copy, covers some use cases that are not suitable
for bpf_ringbuf_reserve().
The difference between commit and discard is very small. Discard just marks
a record as discarded, and such records are supposed to be ignored by consumer
code. Discard is useful for some advanced use-cases, such as ensuring
all-or-nothing multi-record submission, or emulating temporary malloc()/free()
within single BPF program invocation.
Each reserved record is tracked by verifier through existing
reference-tracking logic, similar to socket ref-tracking. It is thus
impossible to reserve a record, but forget to submit (or discard) it.
bpf_ringbuf_query() helper allows to query various properties of ring buffer.
Currently 4 are supported:
- BPF_RB_AVAIL_DATA returns amount of unconsumed data in ring buffer;
- BPF_RB_RING_SIZE returns the size of ring buffer;
- BPF_RB_CONS_POS/BPF_RB_PROD_POS returns current logical possition of
consumer/producer, respectively.
Returned values are momentarily snapshots of ring buffer state and could be
off by the time helper returns, so this should be used only for
debugging/reporting reasons or for implementing various heuristics, that take
into account highly-changeable nature of some of those characteristics.
One such heuristic might involve more fine-grained control over poll/epoll
notifications about new data availability in ring buffer. Together with
BPF_RB_NO_WAKEUP/BPF_RB_FORCE_WAKEUP flags for output/commit/discard helpers,
it allows BPF program a high degree of control and, e.g., more efficient
batched notifications. Default self-balancing strategy, though, should be
adequate for most applications and will work reliable and efficiently already.
Design and implementation
-------------------------
This reserve/commit schema allows a natural way for multiple producers, either
on different CPUs or even on the same CPU/in the same BPF program, to reserve
independent records and work with them without blocking other producers. This
means that if BPF program was interruped by another BPF program sharing the
same ring buffer, they will both get a record reserved (provided there is
enough space left) and can work with it and submit it independently. This
applies to NMI context as well, except that due to using a spinlock during
reservation, in NMI context, bpf_ringbuf_reserve() might fail to get a lock,
in which case reservation will fail even if ring buffer is not full.
The ring buffer itself internally is implemented as a power-of-2 sized
circular buffer, with two logical and ever-increasing counters (which might
wrap around on 32-bit architectures, that's not a problem):
- consumer counter shows up to which logical position consumer consumed the
data;
- producer counter denotes amount of data reserved by all producers.
Each time a record is reserved, producer that "owns" the record will
successfully advance producer counter. At that point, data is still not yet
ready to be consumed, though. Each record has 8 byte header, which contains
the length of reserved record, as well as two extra bits: busy bit to denote
that record is still being worked on, and discard bit, which might be set at
commit time if record is discarded. In the latter case, consumer is supposed
to skip the record and move on to the next one. Record header also encodes
record's relative offset from the beginning of ring buffer data area (in
pages). This allows bpf_ringbuf_commit()/bpf_ringbuf_discard() to accept only
the pointer to the record itself, without requiring also the pointer to ring
buffer itself. Ring buffer memory location will be restored from record
metadata header. This significantly simplifies verifier, as well as improving
API usability.
Producer counter increments are serialized under spinlock, so there is
a strict ordering between reservations. Commits, on the other hand, are
completely lockless and independent. All records become available to consumer
in the order of reservations, but only after all previous records where
already committed. It is thus possible for slow producers to temporarily hold
off submitted records, that were reserved later.
Reservation/commit/consumer protocol is verified by litmus tests in
Documentation/litmus-test/bpf-rb.
One interesting implementation bit, that significantly simplifies (and thus
speeds up as well) implementation of both producers and consumers is how data
area is mapped twice contiguously back-to-back in the virtual memory. This
allows to not take any special measures for samples that have to wrap around
at the end of the circular buffer data area, because the next page after the
last data page would be first data page again, and thus the sample will still
appear completely contiguous in virtual memory. See comment and a simple ASCII
diagram showing this visually in bpf_ringbuf_area_alloc().
Another feature that distinguishes BPF ringbuf from perf ring buffer is
a self-pacing notifications of new data being availability.
bpf_ringbuf_commit() implementation will send a notification of new record
being available after commit only if consumer has already caught up right up
to the record being committed. If not, consumer still has to catch up and thus
will see new data anyways without needing an extra poll notification.
Benchmarks (see tools/testing/selftests/bpf/benchs/bench_ringbuf.c) show that
this allows to achieve a very high throughput without having to resort to
tricks like "notify only every Nth sample", which are necessary with perf
buffer. For extreme cases, when BPF program wants more manual control of
notifications, commit/discard/output helpers accept BPF_RB_NO_WAKEUP and
BPF_RB_FORCE_WAKEUP flags, which give full control over notifications of data
availability, but require extra caution and diligence in using this API.
Comparison to alternatives
--------------------------
Before considering implementing BPF ring buffer from scratch existing
alternatives in kernel were evaluated, but didn't seem to meet the needs. They
largely fell into few categores:
- per-CPU buffers (perf, ftrace, etc), which don't satisfy two motivations
outlined above (ordering and memory consumption);
- linked list-based implementations; while some were multi-producer designs,
consuming these from user-space would be very complicated and most
probably not performant; memory-mapping contiguous piece of memory is
simpler and more performant for user-space consumers;
- io_uring is SPSC, but also requires fixed-sized elements. Naively turning
SPSC queue into MPSC w/ lock would have subpar performance compared to
locked reserve + lockless commit, as with BPF ring buffer. Fixed sized
elements would be too limiting for BPF programs, given existing BPF
programs heavily rely on variable-sized perf buffer already;
- specialized implementations (like a new printk ring buffer, [0]) with lots
of printk-specific limitations and implications, that didn't seem to fit
well for intended use with BPF programs.
[0] https://lwn.net/Articles/779550/
Signed-off-by: Andrii Nakryiko <andriin@fb.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Link: https://lore.kernel.org/bpf/20200529075424.3139988-2-andriin@fb.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2020-05-29 15:54:20 +08:00
|
|
|
* For PTR_TO_MEM_OR_NULL this is used to identify memory allocation
|
|
|
|
|
* for the purpose of tracking that it's freed.
|
2018-10-03 04:35:33 +08:00
|
|
|
* For PTR_TO_SOCKET this is used to share which pointers retain the
|
|
|
|
|
* same reference to the socket, to determine proper reference freeing.
|
2017-08-07 22:26:19 +08:00
|
|
|
*/
|
2016-12-08 02:57:59 +08:00
|
|
|
u32 id;
|
bpf: Fix bpf_tcp_sock and bpf_sk_fullsock issue related to bpf_sk_release
Lorenz Bauer [thanks!] reported that a ptr returned by bpf_tcp_sock(sk)
can still be accessed after bpf_sk_release(sk).
Both bpf_tcp_sock() and bpf_sk_fullsock() have the same issue.
This patch addresses them together.
A simple reproducer looks like this:
sk = bpf_sk_lookup_tcp();
/* if (!sk) ... */
tp = bpf_tcp_sock(sk);
/* if (!tp) ... */
bpf_sk_release(sk);
snd_cwnd = tp->snd_cwnd; /* oops! The verifier does not complain. */
The problem is the verifier did not scrub the register's states of
the tcp_sock ptr (tp) after bpf_sk_release(sk).
[ Note that when calling bpf_tcp_sock(sk), the sk is not always
refcount-acquired. e.g. bpf_tcp_sock(skb->sk). The verifier works
fine for this case. ]
Currently, the verifier does not track if a helper's return ptr (in REG_0)
is "carry"-ing one of its argument's refcount status. To carry this info,
the reg1->id needs to be stored in reg0.
One approach was tried, like "reg0->id = reg1->id", when calling
"bpf_tcp_sock()". The main idea was to avoid adding another "ref_obj_id"
for the same reg. However, overlapping the NULL marking and ref
tracking purpose in one "id" does not work well:
ref_sk = bpf_sk_lookup_tcp();
fullsock = bpf_sk_fullsock(ref_sk);
tp = bpf_tcp_sock(ref_sk);
if (!fullsock) {
bpf_sk_release(ref_sk);
return 0;
}
/* fullsock_reg->id is marked for NOT-NULL.
* Same for tp_reg->id because they have the same id.
*/
/* oops. verifier did not complain about the missing !tp check */
snd_cwnd = tp->snd_cwnd;
Hence, a new "ref_obj_id" is needed in "struct bpf_reg_state".
With a new ref_obj_id, when bpf_sk_release(sk) is called, the verifier can
scrub all reg states which has a ref_obj_id match. It is done with the
changes in release_reg_references() in this patch.
While fixing it, sk_to_full_sk() is removed from bpf_tcp_sock() and
bpf_sk_fullsock() to avoid these helpers from returning
another ptr. It will make bpf_sk_release(tp) possible:
sk = bpf_sk_lookup_tcp();
/* if (!sk) ... */
tp = bpf_tcp_sock(sk);
/* if (!tp) ... */
bpf_sk_release(tp);
A separate helper "bpf_get_listener_sock()" will be added in a later
patch to do sk_to_full_sk().
Misc change notes:
- To allow bpf_sk_release(tp), the arg of bpf_sk_release() is changed
from ARG_PTR_TO_SOCKET to ARG_PTR_TO_SOCK_COMMON. ARG_PTR_TO_SOCKET
is removed from bpf.h since no helper is using it.
- arg_type_is_refcounted() is renamed to arg_type_may_be_refcounted()
because ARG_PTR_TO_SOCK_COMMON is the only one and skb->sk is not
refcounted. All bpf_sk_release(), bpf_sk_fullsock() and bpf_tcp_sock()
take ARG_PTR_TO_SOCK_COMMON.
- check_refcount_ok() ensures is_acquire_function() cannot take
arg_type_may_be_refcounted() as its argument.
- The check_func_arg() can only allow one refcount-ed arg. It is
guaranteed by check_refcount_ok() which ensures at most one arg can be
refcounted. Hence, it is a verifier internal error if >1 refcount arg
found in check_func_arg().
- In release_reference(), release_reference_state() is called
first to ensure a match on "reg->ref_obj_id" can be found before
scrubbing the reg states with release_reg_references().
- reg_is_refcounted() is no longer needed.
1. In mark_ptr_or_null_regs(), its usage is replaced by
"ref_obj_id && ref_obj_id == id" because,
when is_null == true, release_reference_state() should only be
called on the ref_obj_id obtained by a acquire helper (i.e.
is_acquire_function() == true). Otherwise, the following
would happen:
sk = bpf_sk_lookup_tcp();
/* if (!sk) { ... } */
fullsock = bpf_sk_fullsock(sk);
if (!fullsock) {
/*
* release_reference_state(fullsock_reg->ref_obj_id)
* where fullsock_reg->ref_obj_id == sk_reg->ref_obj_id.
*
* Hence, the following bpf_sk_release(sk) will fail
* because the ref state has already been released in the
* earlier release_reference_state(fullsock_reg->ref_obj_id).
*/
bpf_sk_release(sk);
}
2. In release_reg_references(), the current reg_is_refcounted() call
is unnecessary because the id check is enough.
- The type_is_refcounted() and type_is_refcounted_or_null()
are no longer needed also because reg_is_refcounted() is removed.
Fixes: 655a51e536c0 ("bpf: Add struct bpf_tcp_sock and BPF_FUNC_tcp_sock")
Reported-by: Lorenz Bauer <lmb@cloudflare.com>
Signed-off-by: Martin KaFai Lau <kafai@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2019-03-13 01:23:02 +08:00
|
|
|
/* PTR_TO_SOCKET and PTR_TO_TCP_SOCK could be a ptr returned
|
|
|
|
|
* from a pointer-cast helper, bpf_sk_fullsock() and
|
|
|
|
|
* bpf_tcp_sock().
|
|
|
|
|
*
|
|
|
|
|
* Consider the following where "sk" is a reference counted
|
|
|
|
|
* pointer returned from "sk = bpf_sk_lookup_tcp();":
|
|
|
|
|
*
|
|
|
|
|
* 1: sk = bpf_sk_lookup_tcp();
|
|
|
|
|
* 2: if (!sk) { return 0; }
|
|
|
|
|
* 3: fullsock = bpf_sk_fullsock(sk);
|
|
|
|
|
* 4: if (!fullsock) { bpf_sk_release(sk); return 0; }
|
|
|
|
|
* 5: tp = bpf_tcp_sock(fullsock);
|
|
|
|
|
* 6: if (!tp) { bpf_sk_release(sk); return 0; }
|
|
|
|
|
* 7: bpf_sk_release(sk);
|
|
|
|
|
* 8: snd_cwnd = tp->snd_cwnd; // verifier will complain
|
|
|
|
|
*
|
|
|
|
|
* After bpf_sk_release(sk) at line 7, both "fullsock" ptr and
|
|
|
|
|
* "tp" ptr should be invalidated also. In order to do that,
|
|
|
|
|
* the reg holding "fullsock" and "sk" need to remember
|
|
|
|
|
* the original refcounted ptr id (i.e. sk_reg->id) in ref_obj_id
|
|
|
|
|
* such that the verifier can reset all regs which have
|
|
|
|
|
* ref_obj_id matching the sk_reg->id.
|
|
|
|
|
*
|
|
|
|
|
* sk_reg->ref_obj_id is set to sk_reg->id at line 1.
|
|
|
|
|
* sk_reg->id will stay as NULL-marking purpose only.
|
|
|
|
|
* After NULL-marking is done, sk_reg->id can be reset to 0.
|
|
|
|
|
*
|
|
|
|
|
* After "fullsock = bpf_sk_fullsock(sk);" at line 3,
|
|
|
|
|
* fullsock_reg->ref_obj_id is set to sk_reg->ref_obj_id.
|
|
|
|
|
*
|
|
|
|
|
* After "tp = bpf_tcp_sock(fullsock);" at line 5,
|
|
|
|
|
* tp_reg->ref_obj_id is set to fullsock_reg->ref_obj_id
|
|
|
|
|
* which is the same as sk_reg->ref_obj_id.
|
|
|
|
|
*
|
|
|
|
|
* From the verifier perspective, if sk, fullsock and tp
|
|
|
|
|
* are not NULL, they are the same ptr with different
|
|
|
|
|
* reg->type. In particular, bpf_sk_release(tp) is also
|
|
|
|
|
* allowed and has the same effect as bpf_sk_release(sk).
|
|
|
|
|
*/
|
|
|
|
|
u32 ref_obj_id;
|
2017-08-07 22:26:19 +08:00
|
|
|
/* For scalar types (SCALAR_VALUE), this represents our knowledge of
|
|
|
|
|
* the actual value.
|
|
|
|
|
* For pointer types, this represents the variable part of the offset
|
|
|
|
|
* from the pointed-to object, and is shared with all bpf_reg_states
|
|
|
|
|
* with the same id as us.
|
|
|
|
|
*/
|
|
|
|
|
struct tnum var_off;
|
2016-12-08 02:57:59 +08:00
|
|
|
/* Used to determine if any memory access using this register will
|
2017-08-07 22:26:19 +08:00
|
|
|
* result in a bad access.
|
|
|
|
|
* These refer to the same value as var_off, not necessarily the actual
|
|
|
|
|
* contents of the register.
|
2016-12-08 02:57:59 +08:00
|
|
|
*/
|
2017-08-07 22:26:36 +08:00
|
|
|
s64 smin_value; /* minimum possible (s64)value */
|
|
|
|
|
s64 smax_value; /* maximum possible (s64)value */
|
|
|
|
|
u64 umin_value; /* minimum possible (u64)value */
|
|
|
|
|
u64 umax_value; /* maximum possible (u64)value */
|
bpf: Verifier, do explicit ALU32 bounds tracking
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
2020-03-31 05:36:39 +08:00
|
|
|
s32 s32_min_value; /* minimum possible (s32)value */
|
|
|
|
|
s32 s32_max_value; /* maximum possible (s32)value */
|
|
|
|
|
u32 u32_min_value; /* minimum possible (u32)value */
|
|
|
|
|
u32 u32_max_value; /* maximum possible (u32)value */
|
2018-08-23 03:02:19 +08:00
|
|
|
/* parentage chain for liveness checking */
|
|
|
|
|
struct bpf_reg_state *parent;
|
2017-12-15 09:55:06 +08:00
|
|
|
/* Inside the callee two registers can be both PTR_TO_STACK like
|
|
|
|
|
* R1=fp-8 and R2=fp-8, but one of them points to this function stack
|
|
|
|
|
* while another to the caller's stack. To differentiate them 'frameno'
|
|
|
|
|
* is used which is an index in bpf_verifier_state->frame[] array
|
|
|
|
|
* pointing to bpf_func_state.
|
|
|
|
|
*/
|
|
|
|
|
u32 frameno;
|
bpf: verifier: mark verified-insn with sub-register zext flag
eBPF ISA specification requires high 32-bit cleared when low 32-bit
sub-register is written. This applies to destination register of ALU32 etc.
JIT back-ends must guarantee this semantic when doing code-gen. x86_64 and
AArch64 ISA has the same semantics, so the corresponding JIT back-end
doesn't need to do extra work.
However, 32-bit arches (arm, x86, nfp etc.) and some other 64-bit arches
(PowerPC, SPARC etc) need to do explicit zero extension to meet this
requirement, otherwise code like the following will fail.
u64_value = (u64) u32_value
... other uses of u64_value
This is because compiler could exploit the semantic described above and
save those zero extensions for extending u32_value to u64_value, these JIT
back-ends are expected to guarantee this through inserting extra zero
extensions which however could be a significant increase on the code size.
Some benchmarks show there could be ~40% sub-register writes out of total
insns, meaning at least ~40% extra code-gen.
One observation is these extra zero extensions are not always necessary.
Take above code snippet for example, it is possible u32_value will never be
casted into a u64, the value of high 32-bit of u32_value then could be
ignored and extra zero extension could be eliminated.
This patch implements this idea, insns defining sub-registers will be
marked when the high 32-bit of the defined sub-register matters. For
those unmarked insns, it is safe to eliminate high 32-bit clearnace for
them.
Algo:
- Split read flags into READ32 and READ64.
- Record index of insn that does sub-register write. Keep the index inside
reg state and update it during verifier insn walking.
- A full register read on a sub-register marks its definition insn as
needing zero extension on dst register.
A new sub-register write overrides the old one.
- When propagating read64 during path pruning, also mark any insn defining
a sub-register that is read in the pruned path as full-register.
Reviewed-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Signed-off-by: Jiong Wang <jiong.wang@netronome.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2019-05-25 06:25:12 +08:00
|
|
|
/* Tracks subreg definition. The stored value is the insn_idx of the
|
|
|
|
|
* writing insn. This is safe because subreg_def is used before any insn
|
|
|
|
|
* patching which only happens after main verification finished.
|
|
|
|
|
*/
|
|
|
|
|
s32 subreg_def;
|
2017-08-16 03:34:35 +08:00
|
|
|
enum bpf_reg_liveness live;
|
bpf: precise scalar_value tracking
Introduce precision tracking logic that
helps cilium programs the most:
old clang old clang new clang new clang
with all patches with all patches
bpf_lb-DLB_L3.o 1838 2283 1923 1863
bpf_lb-DLB_L4.o 3218 2657 3077 2468
bpf_lb-DUNKNOWN.o 1064 545 1062 544
bpf_lxc-DDROP_ALL.o 26935 23045 166729 22629
bpf_lxc-DUNKNOWN.o 34439 35240 174607 28805
bpf_netdev.o 9721 8753 8407 6801
bpf_overlay.o 6184 7901 5420 4754
bpf_lxc_jit.o 39389 50925 39389 50925
Consider code:
654: (85) call bpf_get_hash_recalc#34
655: (bf) r7 = r0
656: (15) if r8 == 0x0 goto pc+29
657: (bf) r2 = r10
658: (07) r2 += -48
659: (18) r1 = 0xffff8881e41e1b00
661: (85) call bpf_map_lookup_elem#1
662: (15) if r0 == 0x0 goto pc+23
663: (69) r1 = *(u16 *)(r0 +0)
664: (15) if r1 == 0x0 goto pc+21
665: (bf) r8 = r7
666: (57) r8 &= 65535
667: (bf) r2 = r8
668: (3f) r2 /= r1
669: (2f) r2 *= r1
670: (bf) r1 = r8
671: (1f) r1 -= r2
672: (57) r1 &= 255
673: (25) if r1 > 0x1e goto pc+12
R0=map_value(id=0,off=0,ks=20,vs=64,imm=0) R1_w=inv(id=0,umax_value=30,var_off=(0x0; 0x1f))
674: (67) r1 <<= 1
675: (0f) r0 += r1
At this point the verifier will notice that scalar R1 is used in map pointer adjustment.
R1 has to be precise for later operations on R0 to be validated properly.
The verifier will backtrack the above code in the following way:
last_idx 675 first_idx 664
regs=2 stack=0 before 675: (0f) r0 += r1 // started backtracking R1 regs=2 is a bitmask
regs=2 stack=0 before 674: (67) r1 <<= 1
regs=2 stack=0 before 673: (25) if r1 > 0x1e goto pc+12
regs=2 stack=0 before 672: (57) r1 &= 255
regs=2 stack=0 before 671: (1f) r1 -= r2 // now both R1 and R2 has to be precise -> regs=6 mask
regs=6 stack=0 before 670: (bf) r1 = r8 // after this insn R8 and R2 has to be precise
regs=104 stack=0 before 669: (2f) r2 *= r1 // after this one R8, R2, and R1
regs=106 stack=0 before 668: (3f) r2 /= r1
regs=106 stack=0 before 667: (bf) r2 = r8
regs=102 stack=0 before 666: (57) r8 &= 65535
regs=102 stack=0 before 665: (bf) r8 = r7
regs=82 stack=0 before 664: (15) if r1 == 0x0 goto pc+21
// this is the end of verifier state. The following regs will be marked precised:
R1_rw=invP(id=0,umax_value=65535,var_off=(0x0; 0xffff)) R7_rw=invP(id=0)
parent didn't have regs=82 stack=0 marks // so backtracking continues into parent state
last_idx 663 first_idx 655
regs=82 stack=0 before 663: (69) r1 = *(u16 *)(r0 +0) // R1 was assigned no need to track it further
regs=80 stack=0 before 662: (15) if r0 == 0x0 goto pc+23 // keep tracking R7
regs=80 stack=0 before 661: (85) call bpf_map_lookup_elem#1 // keep tracking R7
regs=80 stack=0 before 659: (18) r1 = 0xffff8881e41e1b00
regs=80 stack=0 before 658: (07) r2 += -48
regs=80 stack=0 before 657: (bf) r2 = r10
regs=80 stack=0 before 656: (15) if r8 == 0x0 goto pc+29
regs=80 stack=0 before 655: (bf) r7 = r0 // here the assignment into R7
// mark R0 to be precise:
R0_rw=invP(id=0)
parent didn't have regs=1 stack=0 marks // regs=1 -> tracking R0
last_idx 654 first_idx 644
regs=1 stack=0 before 654: (85) call bpf_get_hash_recalc#34 // and in the parent frame it was a return value
// nothing further to backtrack
Two scalar registers not marked precise are equivalent from state pruning point of view.
More details in the patch comments.
It doesn't support bpf2bpf calls yet and enabled for root only.
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Andrii Nakryiko <andriin@fb.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-06-16 03:12:25 +08:00
|
|
|
/* if (!precise && SCALAR_VALUE) min/max/tnum don't affect safety */
|
|
|
|
|
bool precise;
|
2016-09-21 18:43:57 +08:00
|
|
|
};
|
|
|
|
|
|
|
|
|
|
enum bpf_stack_slot_type {
|
|
|
|
|
STACK_INVALID, /* nothing was stored in this stack slot */
|
|
|
|
|
STACK_SPILL, /* register spilled into stack */
|
2017-12-15 09:55:08 +08:00
|
|
|
STACK_MISC, /* BPF program wrote some data into this slot */
|
|
|
|
|
STACK_ZERO, /* BPF program wrote constant zero */
|
2016-09-21 18:43:57 +08:00
|
|
|
};
|
|
|
|
|
|
|
|
|
|
#define BPF_REG_SIZE 8 /* size of eBPF register in bytes */
|
|
|
|
|
|
2017-11-01 09:16:05 +08:00
|
|
|
struct bpf_stack_state {
|
|
|
|
|
struct bpf_reg_state spilled_ptr;
|
|
|
|
|
u8 slot_type[BPF_REG_SIZE];
|
|
|
|
|
};
|
|
|
|
|
|
2018-10-03 04:35:35 +08:00
|
|
|
struct bpf_reference_state {
|
|
|
|
|
/* Track each reference created with a unique id, even if the same
|
|
|
|
|
* instruction creates the reference multiple times (eg, via CALL).
|
|
|
|
|
*/
|
|
|
|
|
int id;
|
|
|
|
|
/* Instruction where the allocation of this reference occurred. This
|
|
|
|
|
* is used purely to inform the user of a reference leak.
|
|
|
|
|
*/
|
|
|
|
|
int insn_idx;
|
|
|
|
|
};
|
|
|
|
|
|
2016-09-21 18:43:57 +08:00
|
|
|
/* state of the program:
|
|
|
|
|
* type of all registers and stack info
|
|
|
|
|
*/
|
2017-12-15 09:55:06 +08:00
|
|
|
struct bpf_func_state {
|
2016-09-21 18:43:57 +08:00
|
|
|
struct bpf_reg_state regs[MAX_BPF_REG];
|
2017-12-15 09:55:06 +08:00
|
|
|
/* index of call instruction that called into this func */
|
|
|
|
|
int callsite;
|
|
|
|
|
/* stack frame number of this function state from pov of
|
|
|
|
|
* enclosing bpf_verifier_state.
|
|
|
|
|
* 0 = main function, 1 = first callee.
|
|
|
|
|
*/
|
|
|
|
|
u32 frameno;
|
|
|
|
|
/* subprog number == index within subprog_stack_depth
|
|
|
|
|
* zero == main subprog
|
|
|
|
|
*/
|
|
|
|
|
u32 subprogno;
|
|
|
|
|
|
2018-10-03 04:35:35 +08:00
|
|
|
/* The following fields should be last. See copy_func_state() */
|
|
|
|
|
int acquired_refs;
|
|
|
|
|
struct bpf_reference_state *refs;
|
2017-11-01 09:16:05 +08:00
|
|
|
int allocated_stack;
|
|
|
|
|
struct bpf_stack_state *stack;
|
2016-09-21 18:43:57 +08:00
|
|
|
};
|
|
|
|
|
|
bpf: precise scalar_value tracking
Introduce precision tracking logic that
helps cilium programs the most:
old clang old clang new clang new clang
with all patches with all patches
bpf_lb-DLB_L3.o 1838 2283 1923 1863
bpf_lb-DLB_L4.o 3218 2657 3077 2468
bpf_lb-DUNKNOWN.o 1064 545 1062 544
bpf_lxc-DDROP_ALL.o 26935 23045 166729 22629
bpf_lxc-DUNKNOWN.o 34439 35240 174607 28805
bpf_netdev.o 9721 8753 8407 6801
bpf_overlay.o 6184 7901 5420 4754
bpf_lxc_jit.o 39389 50925 39389 50925
Consider code:
654: (85) call bpf_get_hash_recalc#34
655: (bf) r7 = r0
656: (15) if r8 == 0x0 goto pc+29
657: (bf) r2 = r10
658: (07) r2 += -48
659: (18) r1 = 0xffff8881e41e1b00
661: (85) call bpf_map_lookup_elem#1
662: (15) if r0 == 0x0 goto pc+23
663: (69) r1 = *(u16 *)(r0 +0)
664: (15) if r1 == 0x0 goto pc+21
665: (bf) r8 = r7
666: (57) r8 &= 65535
667: (bf) r2 = r8
668: (3f) r2 /= r1
669: (2f) r2 *= r1
670: (bf) r1 = r8
671: (1f) r1 -= r2
672: (57) r1 &= 255
673: (25) if r1 > 0x1e goto pc+12
R0=map_value(id=0,off=0,ks=20,vs=64,imm=0) R1_w=inv(id=0,umax_value=30,var_off=(0x0; 0x1f))
674: (67) r1 <<= 1
675: (0f) r0 += r1
At this point the verifier will notice that scalar R1 is used in map pointer adjustment.
R1 has to be precise for later operations on R0 to be validated properly.
The verifier will backtrack the above code in the following way:
last_idx 675 first_idx 664
regs=2 stack=0 before 675: (0f) r0 += r1 // started backtracking R1 regs=2 is a bitmask
regs=2 stack=0 before 674: (67) r1 <<= 1
regs=2 stack=0 before 673: (25) if r1 > 0x1e goto pc+12
regs=2 stack=0 before 672: (57) r1 &= 255
regs=2 stack=0 before 671: (1f) r1 -= r2 // now both R1 and R2 has to be precise -> regs=6 mask
regs=6 stack=0 before 670: (bf) r1 = r8 // after this insn R8 and R2 has to be precise
regs=104 stack=0 before 669: (2f) r2 *= r1 // after this one R8, R2, and R1
regs=106 stack=0 before 668: (3f) r2 /= r1
regs=106 stack=0 before 667: (bf) r2 = r8
regs=102 stack=0 before 666: (57) r8 &= 65535
regs=102 stack=0 before 665: (bf) r8 = r7
regs=82 stack=0 before 664: (15) if r1 == 0x0 goto pc+21
// this is the end of verifier state. The following regs will be marked precised:
R1_rw=invP(id=0,umax_value=65535,var_off=(0x0; 0xffff)) R7_rw=invP(id=0)
parent didn't have regs=82 stack=0 marks // so backtracking continues into parent state
last_idx 663 first_idx 655
regs=82 stack=0 before 663: (69) r1 = *(u16 *)(r0 +0) // R1 was assigned no need to track it further
regs=80 stack=0 before 662: (15) if r0 == 0x0 goto pc+23 // keep tracking R7
regs=80 stack=0 before 661: (85) call bpf_map_lookup_elem#1 // keep tracking R7
regs=80 stack=0 before 659: (18) r1 = 0xffff8881e41e1b00
regs=80 stack=0 before 658: (07) r2 += -48
regs=80 stack=0 before 657: (bf) r2 = r10
regs=80 stack=0 before 656: (15) if r8 == 0x0 goto pc+29
regs=80 stack=0 before 655: (bf) r7 = r0 // here the assignment into R7
// mark R0 to be precise:
R0_rw=invP(id=0)
parent didn't have regs=1 stack=0 marks // regs=1 -> tracking R0
last_idx 654 first_idx 644
regs=1 stack=0 before 654: (85) call bpf_get_hash_recalc#34 // and in the parent frame it was a return value
// nothing further to backtrack
Two scalar registers not marked precise are equivalent from state pruning point of view.
More details in the patch comments.
It doesn't support bpf2bpf calls yet and enabled for root only.
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Andrii Nakryiko <andriin@fb.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-06-16 03:12:25 +08:00
|
|
|
struct bpf_idx_pair {
|
|
|
|
|
u32 prev_idx;
|
|
|
|
|
u32 idx;
|
|
|
|
|
};
|
|
|
|
|
|
2017-12-15 09:55:06 +08:00
|
|
|
#define MAX_CALL_FRAMES 8
|
|
|
|
|
struct bpf_verifier_state {
|
|
|
|
|
/* call stack tracking */
|
|
|
|
|
struct bpf_func_state *frame[MAX_CALL_FRAMES];
|
2019-06-16 03:12:20 +08:00
|
|
|
struct bpf_verifier_state *parent;
|
|
|
|
|
/*
|
|
|
|
|
* 'branches' field is the number of branches left to explore:
|
|
|
|
|
* 0 - all possible paths from this state reached bpf_exit or
|
|
|
|
|
* were safely pruned
|
|
|
|
|
* 1 - at least one path is being explored.
|
|
|
|
|
* This state hasn't reached bpf_exit
|
|
|
|
|
* 2 - at least two paths are being explored.
|
|
|
|
|
* This state is an immediate parent of two children.
|
|
|
|
|
* One is fallthrough branch with branches==1 and another
|
|
|
|
|
* state is pushed into stack (to be explored later) also with
|
|
|
|
|
* branches==1. The parent of this state has branches==1.
|
|
|
|
|
* The verifier state tree connected via 'parent' pointer looks like:
|
|
|
|
|
* 1
|
|
|
|
|
* 1
|
|
|
|
|
* 2 -> 1 (first 'if' pushed into stack)
|
|
|
|
|
* 1
|
|
|
|
|
* 2 -> 1 (second 'if' pushed into stack)
|
|
|
|
|
* 1
|
|
|
|
|
* 1
|
|
|
|
|
* 1 bpf_exit.
|
|
|
|
|
*
|
|
|
|
|
* Once do_check() reaches bpf_exit, it calls update_branch_counts()
|
|
|
|
|
* and the verifier state tree will look:
|
|
|
|
|
* 1
|
|
|
|
|
* 1
|
|
|
|
|
* 2 -> 1 (first 'if' pushed into stack)
|
|
|
|
|
* 1
|
|
|
|
|
* 1 -> 1 (second 'if' pushed into stack)
|
|
|
|
|
* 0
|
|
|
|
|
* 0
|
|
|
|
|
* 0 bpf_exit.
|
|
|
|
|
* After pop_stack() the do_check() will resume at second 'if'.
|
|
|
|
|
*
|
|
|
|
|
* If is_state_visited() sees a state with branches > 0 it means
|
|
|
|
|
* there is a loop. If such state is exactly equal to the current state
|
|
|
|
|
* it's an infinite loop. Note states_equal() checks for states
|
|
|
|
|
* equvalency, so two states being 'states_equal' does not mean
|
|
|
|
|
* infinite loop. The exact comparison is provided by
|
|
|
|
|
* states_maybe_looping() function. It's a stronger pre-check and
|
|
|
|
|
* much faster than states_equal().
|
|
|
|
|
*
|
|
|
|
|
* This algorithm may not find all possible infinite loops or
|
|
|
|
|
* loop iteration count may be too high.
|
|
|
|
|
* In such cases BPF_COMPLEXITY_LIMIT_INSNS limit kicks in.
|
|
|
|
|
*/
|
|
|
|
|
u32 branches;
|
2019-05-22 11:17:07 +08:00
|
|
|
u32 insn_idx;
|
2017-12-15 09:55:06 +08:00
|
|
|
u32 curframe;
|
2019-02-01 07:40:04 +08:00
|
|
|
u32 active_spin_lock;
|
bpf: prevent out of bounds speculation on pointer arithmetic
Jann reported that the original commit back in b2157399cc98
("bpf: prevent out-of-bounds speculation") was not sufficient
to stop CPU from speculating out of bounds memory access:
While b2157399cc98 only focussed on masking array map access
for unprivileged users for tail calls and data access such
that the user provided index gets sanitized from BPF program
and syscall side, there is still a more generic form affected
from BPF programs that applies to most maps that hold user
data in relation to dynamic map access when dealing with
unknown scalars or "slow" known scalars as access offset, for
example:
- Load a map value pointer into R6
- Load an index into R7
- Do a slow computation (e.g. with a memory dependency) that
loads a limit into R8 (e.g. load the limit from a map for
high latency, then mask it to make the verifier happy)
- Exit if R7 >= R8 (mispredicted branch)
- Load R0 = R6[R7]
- Load R0 = R6[R0]
For unknown scalars there are two options in the BPF verifier
where we could derive knowledge from in order to guarantee
safe access to the memory: i) While </>/<=/>= variants won't
allow to derive any lower or upper bounds from the unknown
scalar where it would be safe to add it to the map value
pointer, it is possible through ==/!= test however. ii) another
option is to transform the unknown scalar into a known scalar,
for example, through ALU ops combination such as R &= <imm>
followed by R |= <imm> or any similar combination where the
original information from the unknown scalar would be destroyed
entirely leaving R with a constant. The initial slow load still
precedes the latter ALU ops on that register, so the CPU
executes speculatively from that point. Once we have the known
scalar, any compare operation would work then. A third option
only involving registers with known scalars could be crafted
as described in [0] where a CPU port (e.g. Slow Int unit)
would be filled with many dependent computations such that
the subsequent condition depending on its outcome has to wait
for evaluation on its execution port and thereby executing
speculatively if the speculated code can be scheduled on a
different execution port, or any other form of mistraining
as described in [1], for example. Given this is not limited
to only unknown scalars, not only map but also stack access
is affected since both is accessible for unprivileged users
and could potentially be used for out of bounds access under
speculation.
In order to prevent any of these cases, the verifier is now
sanitizing pointer arithmetic on the offset such that any
out of bounds speculation would be masked in a way where the
pointer arithmetic result in the destination register will
stay unchanged, meaning offset masked into zero similar as
in array_index_nospec() case. With regards to implementation,
there are three options that were considered: i) new insn
for sanitation, ii) push/pop insn and sanitation as inlined
BPF, iii) reuse of ax register and sanitation as inlined BPF.
Option i) has the downside that we end up using from reserved
bits in the opcode space, but also that we would require
each JIT to emit masking as native arch opcodes meaning
mitigation would have slow adoption till everyone implements
it eventually which is counter-productive. Option ii) and iii)
have both in common that a temporary register is needed in
order to implement the sanitation as inlined BPF since we
are not allowed to modify the source register. While a push /
pop insn in ii) would be useful to have in any case, it
requires once again that every JIT needs to implement it
first. While possible, amount of changes needed would also
be unsuitable for a -stable patch. Therefore, the path which
has fewer changes, less BPF instructions for the mitigation
and does not require anything to be changed in the JITs is
option iii) which this work is pursuing. The ax register is
already mapped to a register in all JITs (modulo arm32 where
it's mapped to stack as various other BPF registers there)
and used in constant blinding for JITs-only so far. It can
be reused for verifier rewrites under certain constraints.
The interpreter's tmp "register" has therefore been remapped
into extending the register set with hidden ax register and
reusing that for a number of instructions that needed the
prior temporary variable internally (e.g. div, mod). This
allows for zero increase in stack space usage in the interpreter,
and enables (restricted) generic use in rewrites otherwise as
long as such a patchlet does not make use of these instructions.
The sanitation mask is dynamic and relative to the offset the
map value or stack pointer currently holds.
There are various cases that need to be taken under consideration
for the masking, e.g. such operation could look as follows:
ptr += val or val += ptr or ptr -= val. Thus, the value to be
sanitized could reside either in source or in destination
register, and the limit is different depending on whether
the ALU op is addition or subtraction and depending on the
current known and bounded offset. The limit is derived as
follows: limit := max_value_size - (smin_value + off). For
subtraction: limit := umax_value + off. This holds because
we do not allow any pointer arithmetic that would
temporarily go out of bounds or would have an unknown
value with mixed signed bounds where it is unclear at
verification time whether the actual runtime value would
be either negative or positive. For example, we have a
derived map pointer value with constant offset and bounded
one, so limit based on smin_value works because the verifier
requires that statically analyzed arithmetic on the pointer
must be in bounds, and thus it checks if resulting
smin_value + off and umax_value + off is still within map
value bounds at time of arithmetic in addition to time of
access. Similarly, for the case of stack access we derive
the limit as follows: MAX_BPF_STACK + off for subtraction
and -off for the case of addition where off := ptr_reg->off +
ptr_reg->var_off.value. Subtraction is a special case for
the masking which can be in form of ptr += -val, ptr -= -val,
or ptr -= val. In the first two cases where we know that
the value is negative, we need to temporarily negate the
value in order to do the sanitation on a positive value
where we later swap the ALU op, and restore original source
register if the value was in source.
The sanitation of pointer arithmetic alone is still not fully
sufficient as is, since a scenario like the following could
happen ...
PTR += 0x1000 (e.g. K-based imm)
PTR -= BIG_NUMBER_WITH_SLOW_COMPARISON
PTR += 0x1000
PTR -= BIG_NUMBER_WITH_SLOW_COMPARISON
[...]
... which under speculation could end up as ...
PTR += 0x1000
PTR -= 0 [ truncated by mitigation ]
PTR += 0x1000
PTR -= 0 [ truncated by mitigation ]
[...]
... and therefore still access out of bounds. To prevent such
case, the verifier is also analyzing safety for potential out
of bounds access under speculative execution. Meaning, it is
also simulating pointer access under truncation. We therefore
"branch off" and push the current verification state after the
ALU operation with known 0 to the verification stack for later
analysis. Given the current path analysis succeeded it is
likely that the one under speculation can be pruned. In any
case, it is also subject to existing complexity limits and
therefore anything beyond this point will be rejected. In
terms of pruning, it needs to be ensured that the verification
state from speculative execution simulation must never prune
a non-speculative execution path, therefore, we mark verifier
state accordingly at the time of push_stack(). If verifier
detects out of bounds access under speculative execution from
one of the possible paths that includes a truncation, it will
reject such program.
Given we mask every reg-based pointer arithmetic for
unprivileged programs, we've been looking into how it could
affect real-world programs in terms of size increase. As the
majority of programs are targeted for privileged-only use
case, we've unconditionally enabled masking (with its alu
restrictions on top of it) for privileged programs for the
sake of testing in order to check i) whether they get rejected
in its current form, and ii) by how much the number of
instructions and size will increase. We've tested this by
using Katran, Cilium and test_l4lb from the kernel selftests.
For Katran we've evaluated balancer_kern.o, Cilium bpf_lxc.o
and an older test object bpf_lxc_opt_-DUNKNOWN.o and l4lb
we've used test_l4lb.o as well as test_l4lb_noinline.o. We
found that none of the programs got rejected by the verifier
with this change, and that impact is rather minimal to none.
balancer_kern.o had 13,904 bytes (1,738 insns) xlated and
7,797 bytes JITed before and after the change. Most complex
program in bpf_lxc.o had 30,544 bytes (3,817 insns) xlated
and 18,538 bytes JITed before and after and none of the other
tail call programs in bpf_lxc.o had any changes either. For
the older bpf_lxc_opt_-DUNKNOWN.o object we found a small
increase from 20,616 bytes (2,576 insns) and 12,536 bytes JITed
before to 20,664 bytes (2,582 insns) and 12,558 bytes JITed
after the change. Other programs from that object file had
similar small increase. Both test_l4lb.o had no change and
remained at 6,544 bytes (817 insns) xlated and 3,401 bytes
JITed and for test_l4lb_noinline.o constant at 5,080 bytes
(634 insns) xlated and 3,313 bytes JITed. This can be explained
in that LLVM typically optimizes stack based pointer arithmetic
by using K-based operations and that use of dynamic map access
is not overly frequent. However, in future we may decide to
optimize the algorithm further under known guarantees from
branch and value speculation. Latter seems also unclear in
terms of prediction heuristics that today's CPUs apply as well
as whether there could be collisions in e.g. the predictor's
Value History/Pattern Table for triggering out of bounds access,
thus masking is performed unconditionally at this point but could
be subject to relaxation later on. We were generally also
brainstorming various other approaches for mitigation, but the
blocker was always lack of available registers at runtime and/or
overhead for runtime tracking of limits belonging to a specific
pointer. Thus, we found this to be minimally intrusive under
given constraints.
With that in place, a simple example with sanitized access on
unprivileged load at post-verification time looks as follows:
# bpftool prog dump xlated id 282
[...]
28: (79) r1 = *(u64 *)(r7 +0)
29: (79) r2 = *(u64 *)(r7 +8)
30: (57) r1 &= 15
31: (79) r3 = *(u64 *)(r0 +4608)
32: (57) r3 &= 1
33: (47) r3 |= 1
34: (2d) if r2 > r3 goto pc+19
35: (b4) (u32) r11 = (u32) 20479 |
36: (1f) r11 -= r2 | Dynamic sanitation for pointer
37: (4f) r11 |= r2 | arithmetic with registers
38: (87) r11 = -r11 | containing bounded or known
39: (c7) r11 s>>= 63 | scalars in order to prevent
40: (5f) r11 &= r2 | out of bounds speculation.
41: (0f) r4 += r11 |
42: (71) r4 = *(u8 *)(r4 +0)
43: (6f) r4 <<= r1
[...]
For the case where the scalar sits in the destination register
as opposed to the source register, the following code is emitted
for the above example:
[...]
16: (b4) (u32) r11 = (u32) 20479
17: (1f) r11 -= r2
18: (4f) r11 |= r2
19: (87) r11 = -r11
20: (c7) r11 s>>= 63
21: (5f) r2 &= r11
22: (0f) r2 += r0
23: (61) r0 = *(u32 *)(r2 +0)
[...]
JIT blinding example with non-conflicting use of r10:
[...]
d5: je 0x0000000000000106 _
d7: mov 0x0(%rax),%edi |
da: mov $0xf153246,%r10d | Index load from map value and
e0: xor $0xf153259,%r10 | (const blinded) mask with 0x1f.
e7: and %r10,%rdi |_
ea: mov $0x2f,%r10d |
f0: sub %rdi,%r10 | Sanitized addition. Both use r10
f3: or %rdi,%r10 | but do not interfere with each
f6: neg %r10 | other. (Neither do these instructions
f9: sar $0x3f,%r10 | interfere with the use of ax as temp
fd: and %r10,%rdi | in interpreter.)
100: add %rax,%rdi |_
103: mov 0x0(%rdi),%eax
[...]
Tested that it fixes Jann's reproducer, and also checked that test_verifier
and test_progs suite with interpreter, JIT and JIT with hardening enabled
on x86-64 and arm64 runs successfully.
[0] Speculose: Analyzing the Security Implications of Speculative
Execution in CPUs, Giorgi Maisuradze and Christian Rossow,
https://arxiv.org/pdf/1801.04084.pdf
[1] A Systematic Evaluation of Transient Execution Attacks and
Defenses, Claudio Canella, Jo Van Bulck, Michael Schwarz,
Moritz Lipp, Benjamin von Berg, Philipp Ortner, Frank Piessens,
Dmitry Evtyushkin, Daniel Gruss,
https://arxiv.org/pdf/1811.05441.pdf
Fixes: b2157399cc98 ("bpf: prevent out-of-bounds speculation")
Reported-by: Jann Horn <jannh@google.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2019-01-03 07:58:34 +08:00
|
|
|
bool speculative;
|
bpf: precise scalar_value tracking
Introduce precision tracking logic that
helps cilium programs the most:
old clang old clang new clang new clang
with all patches with all patches
bpf_lb-DLB_L3.o 1838 2283 1923 1863
bpf_lb-DLB_L4.o 3218 2657 3077 2468
bpf_lb-DUNKNOWN.o 1064 545 1062 544
bpf_lxc-DDROP_ALL.o 26935 23045 166729 22629
bpf_lxc-DUNKNOWN.o 34439 35240 174607 28805
bpf_netdev.o 9721 8753 8407 6801
bpf_overlay.o 6184 7901 5420 4754
bpf_lxc_jit.o 39389 50925 39389 50925
Consider code:
654: (85) call bpf_get_hash_recalc#34
655: (bf) r7 = r0
656: (15) if r8 == 0x0 goto pc+29
657: (bf) r2 = r10
658: (07) r2 += -48
659: (18) r1 = 0xffff8881e41e1b00
661: (85) call bpf_map_lookup_elem#1
662: (15) if r0 == 0x0 goto pc+23
663: (69) r1 = *(u16 *)(r0 +0)
664: (15) if r1 == 0x0 goto pc+21
665: (bf) r8 = r7
666: (57) r8 &= 65535
667: (bf) r2 = r8
668: (3f) r2 /= r1
669: (2f) r2 *= r1
670: (bf) r1 = r8
671: (1f) r1 -= r2
672: (57) r1 &= 255
673: (25) if r1 > 0x1e goto pc+12
R0=map_value(id=0,off=0,ks=20,vs=64,imm=0) R1_w=inv(id=0,umax_value=30,var_off=(0x0; 0x1f))
674: (67) r1 <<= 1
675: (0f) r0 += r1
At this point the verifier will notice that scalar R1 is used in map pointer adjustment.
R1 has to be precise for later operations on R0 to be validated properly.
The verifier will backtrack the above code in the following way:
last_idx 675 first_idx 664
regs=2 stack=0 before 675: (0f) r0 += r1 // started backtracking R1 regs=2 is a bitmask
regs=2 stack=0 before 674: (67) r1 <<= 1
regs=2 stack=0 before 673: (25) if r1 > 0x1e goto pc+12
regs=2 stack=0 before 672: (57) r1 &= 255
regs=2 stack=0 before 671: (1f) r1 -= r2 // now both R1 and R2 has to be precise -> regs=6 mask
regs=6 stack=0 before 670: (bf) r1 = r8 // after this insn R8 and R2 has to be precise
regs=104 stack=0 before 669: (2f) r2 *= r1 // after this one R8, R2, and R1
regs=106 stack=0 before 668: (3f) r2 /= r1
regs=106 stack=0 before 667: (bf) r2 = r8
regs=102 stack=0 before 666: (57) r8 &= 65535
regs=102 stack=0 before 665: (bf) r8 = r7
regs=82 stack=0 before 664: (15) if r1 == 0x0 goto pc+21
// this is the end of verifier state. The following regs will be marked precised:
R1_rw=invP(id=0,umax_value=65535,var_off=(0x0; 0xffff)) R7_rw=invP(id=0)
parent didn't have regs=82 stack=0 marks // so backtracking continues into parent state
last_idx 663 first_idx 655
regs=82 stack=0 before 663: (69) r1 = *(u16 *)(r0 +0) // R1 was assigned no need to track it further
regs=80 stack=0 before 662: (15) if r0 == 0x0 goto pc+23 // keep tracking R7
regs=80 stack=0 before 661: (85) call bpf_map_lookup_elem#1 // keep tracking R7
regs=80 stack=0 before 659: (18) r1 = 0xffff8881e41e1b00
regs=80 stack=0 before 658: (07) r2 += -48
regs=80 stack=0 before 657: (bf) r2 = r10
regs=80 stack=0 before 656: (15) if r8 == 0x0 goto pc+29
regs=80 stack=0 before 655: (bf) r7 = r0 // here the assignment into R7
// mark R0 to be precise:
R0_rw=invP(id=0)
parent didn't have regs=1 stack=0 marks // regs=1 -> tracking R0
last_idx 654 first_idx 644
regs=1 stack=0 before 654: (85) call bpf_get_hash_recalc#34 // and in the parent frame it was a return value
// nothing further to backtrack
Two scalar registers not marked precise are equivalent from state pruning point of view.
More details in the patch comments.
It doesn't support bpf2bpf calls yet and enabled for root only.
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Andrii Nakryiko <andriin@fb.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-06-16 03:12:25 +08:00
|
|
|
|
|
|
|
|
/* first and last insn idx of this verifier state */
|
|
|
|
|
u32 first_insn_idx;
|
|
|
|
|
u32 last_insn_idx;
|
|
|
|
|
/* jmp history recorded from first to last.
|
|
|
|
|
* backtracking is using it to go from last to first.
|
|
|
|
|
* For most states jmp_history_cnt is [0-3].
|
|
|
|
|
* For loops can go up to ~40.
|
|
|
|
|
*/
|
|
|
|
|
struct bpf_idx_pair *jmp_history;
|
|
|
|
|
u32 jmp_history_cnt;
|
2017-12-15 09:55:06 +08:00
|
|
|
};
|
|
|
|
|
|
2018-10-03 04:35:29 +08:00
|
|
|
#define bpf_get_spilled_reg(slot, frame) \
|
|
|
|
|
(((slot < frame->allocated_stack / BPF_REG_SIZE) && \
|
|
|
|
|
(frame->stack[slot].slot_type[0] == STACK_SPILL)) \
|
|
|
|
|
? &frame->stack[slot].spilled_ptr : NULL)
|
|
|
|
|
|
|
|
|
|
/* Iterate over 'frame', setting 'reg' to either NULL or a spilled register. */
|
|
|
|
|
#define bpf_for_each_spilled_reg(iter, frame, reg) \
|
|
|
|
|
for (iter = 0, reg = bpf_get_spilled_reg(iter, frame); \
|
|
|
|
|
iter < frame->allocated_stack / BPF_REG_SIZE; \
|
|
|
|
|
iter++, reg = bpf_get_spilled_reg(iter, frame))
|
|
|
|
|
|
2016-09-21 18:43:57 +08:00
|
|
|
/* linked list of verifier states used to prune search */
|
|
|
|
|
struct bpf_verifier_state_list {
|
|
|
|
|
struct bpf_verifier_state state;
|
|
|
|
|
struct bpf_verifier_state_list *next;
|
2019-04-02 12:27:41 +08:00
|
|
|
int miss_cnt, hit_cnt;
|
2016-09-21 18:43:57 +08:00
|
|
|
};
|
|
|
|
|
|
bpf: prevent out of bounds speculation on pointer arithmetic
Jann reported that the original commit back in b2157399cc98
("bpf: prevent out-of-bounds speculation") was not sufficient
to stop CPU from speculating out of bounds memory access:
While b2157399cc98 only focussed on masking array map access
for unprivileged users for tail calls and data access such
that the user provided index gets sanitized from BPF program
and syscall side, there is still a more generic form affected
from BPF programs that applies to most maps that hold user
data in relation to dynamic map access when dealing with
unknown scalars or "slow" known scalars as access offset, for
example:
- Load a map value pointer into R6
- Load an index into R7
- Do a slow computation (e.g. with a memory dependency) that
loads a limit into R8 (e.g. load the limit from a map for
high latency, then mask it to make the verifier happy)
- Exit if R7 >= R8 (mispredicted branch)
- Load R0 = R6[R7]
- Load R0 = R6[R0]
For unknown scalars there are two options in the BPF verifier
where we could derive knowledge from in order to guarantee
safe access to the memory: i) While </>/<=/>= variants won't
allow to derive any lower or upper bounds from the unknown
scalar where it would be safe to add it to the map value
pointer, it is possible through ==/!= test however. ii) another
option is to transform the unknown scalar into a known scalar,
for example, through ALU ops combination such as R &= <imm>
followed by R |= <imm> or any similar combination where the
original information from the unknown scalar would be destroyed
entirely leaving R with a constant. The initial slow load still
precedes the latter ALU ops on that register, so the CPU
executes speculatively from that point. Once we have the known
scalar, any compare operation would work then. A third option
only involving registers with known scalars could be crafted
as described in [0] where a CPU port (e.g. Slow Int unit)
would be filled with many dependent computations such that
the subsequent condition depending on its outcome has to wait
for evaluation on its execution port and thereby executing
speculatively if the speculated code can be scheduled on a
different execution port, or any other form of mistraining
as described in [1], for example. Given this is not limited
to only unknown scalars, not only map but also stack access
is affected since both is accessible for unprivileged users
and could potentially be used for out of bounds access under
speculation.
In order to prevent any of these cases, the verifier is now
sanitizing pointer arithmetic on the offset such that any
out of bounds speculation would be masked in a way where the
pointer arithmetic result in the destination register will
stay unchanged, meaning offset masked into zero similar as
in array_index_nospec() case. With regards to implementation,
there are three options that were considered: i) new insn
for sanitation, ii) push/pop insn and sanitation as inlined
BPF, iii) reuse of ax register and sanitation as inlined BPF.
Option i) has the downside that we end up using from reserved
bits in the opcode space, but also that we would require
each JIT to emit masking as native arch opcodes meaning
mitigation would have slow adoption till everyone implements
it eventually which is counter-productive. Option ii) and iii)
have both in common that a temporary register is needed in
order to implement the sanitation as inlined BPF since we
are not allowed to modify the source register. While a push /
pop insn in ii) would be useful to have in any case, it
requires once again that every JIT needs to implement it
first. While possible, amount of changes needed would also
be unsuitable for a -stable patch. Therefore, the path which
has fewer changes, less BPF instructions for the mitigation
and does not require anything to be changed in the JITs is
option iii) which this work is pursuing. The ax register is
already mapped to a register in all JITs (modulo arm32 where
it's mapped to stack as various other BPF registers there)
and used in constant blinding for JITs-only so far. It can
be reused for verifier rewrites under certain constraints.
The interpreter's tmp "register" has therefore been remapped
into extending the register set with hidden ax register and
reusing that for a number of instructions that needed the
prior temporary variable internally (e.g. div, mod). This
allows for zero increase in stack space usage in the interpreter,
and enables (restricted) generic use in rewrites otherwise as
long as such a patchlet does not make use of these instructions.
The sanitation mask is dynamic and relative to the offset the
map value or stack pointer currently holds.
There are various cases that need to be taken under consideration
for the masking, e.g. such operation could look as follows:
ptr += val or val += ptr or ptr -= val. Thus, the value to be
sanitized could reside either in source or in destination
register, and the limit is different depending on whether
the ALU op is addition or subtraction and depending on the
current known and bounded offset. The limit is derived as
follows: limit := max_value_size - (smin_value + off). For
subtraction: limit := umax_value + off. This holds because
we do not allow any pointer arithmetic that would
temporarily go out of bounds or would have an unknown
value with mixed signed bounds where it is unclear at
verification time whether the actual runtime value would
be either negative or positive. For example, we have a
derived map pointer value with constant offset and bounded
one, so limit based on smin_value works because the verifier
requires that statically analyzed arithmetic on the pointer
must be in bounds, and thus it checks if resulting
smin_value + off and umax_value + off is still within map
value bounds at time of arithmetic in addition to time of
access. Similarly, for the case of stack access we derive
the limit as follows: MAX_BPF_STACK + off for subtraction
and -off for the case of addition where off := ptr_reg->off +
ptr_reg->var_off.value. Subtraction is a special case for
the masking which can be in form of ptr += -val, ptr -= -val,
or ptr -= val. In the first two cases where we know that
the value is negative, we need to temporarily negate the
value in order to do the sanitation on a positive value
where we later swap the ALU op, and restore original source
register if the value was in source.
The sanitation of pointer arithmetic alone is still not fully
sufficient as is, since a scenario like the following could
happen ...
PTR += 0x1000 (e.g. K-based imm)
PTR -= BIG_NUMBER_WITH_SLOW_COMPARISON
PTR += 0x1000
PTR -= BIG_NUMBER_WITH_SLOW_COMPARISON
[...]
... which under speculation could end up as ...
PTR += 0x1000
PTR -= 0 [ truncated by mitigation ]
PTR += 0x1000
PTR -= 0 [ truncated by mitigation ]
[...]
... and therefore still access out of bounds. To prevent such
case, the verifier is also analyzing safety for potential out
of bounds access under speculative execution. Meaning, it is
also simulating pointer access under truncation. We therefore
"branch off" and push the current verification state after the
ALU operation with known 0 to the verification stack for later
analysis. Given the current path analysis succeeded it is
likely that the one under speculation can be pruned. In any
case, it is also subject to existing complexity limits and
therefore anything beyond this point will be rejected. In
terms of pruning, it needs to be ensured that the verification
state from speculative execution simulation must never prune
a non-speculative execution path, therefore, we mark verifier
state accordingly at the time of push_stack(). If verifier
detects out of bounds access under speculative execution from
one of the possible paths that includes a truncation, it will
reject such program.
Given we mask every reg-based pointer arithmetic for
unprivileged programs, we've been looking into how it could
affect real-world programs in terms of size increase. As the
majority of programs are targeted for privileged-only use
case, we've unconditionally enabled masking (with its alu
restrictions on top of it) for privileged programs for the
sake of testing in order to check i) whether they get rejected
in its current form, and ii) by how much the number of
instructions and size will increase. We've tested this by
using Katran, Cilium and test_l4lb from the kernel selftests.
For Katran we've evaluated balancer_kern.o, Cilium bpf_lxc.o
and an older test object bpf_lxc_opt_-DUNKNOWN.o and l4lb
we've used test_l4lb.o as well as test_l4lb_noinline.o. We
found that none of the programs got rejected by the verifier
with this change, and that impact is rather minimal to none.
balancer_kern.o had 13,904 bytes (1,738 insns) xlated and
7,797 bytes JITed before and after the change. Most complex
program in bpf_lxc.o had 30,544 bytes (3,817 insns) xlated
and 18,538 bytes JITed before and after and none of the other
tail call programs in bpf_lxc.o had any changes either. For
the older bpf_lxc_opt_-DUNKNOWN.o object we found a small
increase from 20,616 bytes (2,576 insns) and 12,536 bytes JITed
before to 20,664 bytes (2,582 insns) and 12,558 bytes JITed
after the change. Other programs from that object file had
similar small increase. Both test_l4lb.o had no change and
remained at 6,544 bytes (817 insns) xlated and 3,401 bytes
JITed and for test_l4lb_noinline.o constant at 5,080 bytes
(634 insns) xlated and 3,313 bytes JITed. This can be explained
in that LLVM typically optimizes stack based pointer arithmetic
by using K-based operations and that use of dynamic map access
is not overly frequent. However, in future we may decide to
optimize the algorithm further under known guarantees from
branch and value speculation. Latter seems also unclear in
terms of prediction heuristics that today's CPUs apply as well
as whether there could be collisions in e.g. the predictor's
Value History/Pattern Table for triggering out of bounds access,
thus masking is performed unconditionally at this point but could
be subject to relaxation later on. We were generally also
brainstorming various other approaches for mitigation, but the
blocker was always lack of available registers at runtime and/or
overhead for runtime tracking of limits belonging to a specific
pointer. Thus, we found this to be minimally intrusive under
given constraints.
With that in place, a simple example with sanitized access on
unprivileged load at post-verification time looks as follows:
# bpftool prog dump xlated id 282
[...]
28: (79) r1 = *(u64 *)(r7 +0)
29: (79) r2 = *(u64 *)(r7 +8)
30: (57) r1 &= 15
31: (79) r3 = *(u64 *)(r0 +4608)
32: (57) r3 &= 1
33: (47) r3 |= 1
34: (2d) if r2 > r3 goto pc+19
35: (b4) (u32) r11 = (u32) 20479 |
36: (1f) r11 -= r2 | Dynamic sanitation for pointer
37: (4f) r11 |= r2 | arithmetic with registers
38: (87) r11 = -r11 | containing bounded or known
39: (c7) r11 s>>= 63 | scalars in order to prevent
40: (5f) r11 &= r2 | out of bounds speculation.
41: (0f) r4 += r11 |
42: (71) r4 = *(u8 *)(r4 +0)
43: (6f) r4 <<= r1
[...]
For the case where the scalar sits in the destination register
as opposed to the source register, the following code is emitted
for the above example:
[...]
16: (b4) (u32) r11 = (u32) 20479
17: (1f) r11 -= r2
18: (4f) r11 |= r2
19: (87) r11 = -r11
20: (c7) r11 s>>= 63
21: (5f) r2 &= r11
22: (0f) r2 += r0
23: (61) r0 = *(u32 *)(r2 +0)
[...]
JIT blinding example with non-conflicting use of r10:
[...]
d5: je 0x0000000000000106 _
d7: mov 0x0(%rax),%edi |
da: mov $0xf153246,%r10d | Index load from map value and
e0: xor $0xf153259,%r10 | (const blinded) mask with 0x1f.
e7: and %r10,%rdi |_
ea: mov $0x2f,%r10d |
f0: sub %rdi,%r10 | Sanitized addition. Both use r10
f3: or %rdi,%r10 | but do not interfere with each
f6: neg %r10 | other. (Neither do these instructions
f9: sar $0x3f,%r10 | interfere with the use of ax as temp
fd: and %r10,%rdi | in interpreter.)
100: add %rax,%rdi |_
103: mov 0x0(%rdi),%eax
[...]
Tested that it fixes Jann's reproducer, and also checked that test_verifier
and test_progs suite with interpreter, JIT and JIT with hardening enabled
on x86-64 and arm64 runs successfully.
[0] Speculose: Analyzing the Security Implications of Speculative
Execution in CPUs, Giorgi Maisuradze and Christian Rossow,
https://arxiv.org/pdf/1801.04084.pdf
[1] A Systematic Evaluation of Transient Execution Attacks and
Defenses, Claudio Canella, Jo Van Bulck, Michael Schwarz,
Moritz Lipp, Benjamin von Berg, Philipp Ortner, Frank Piessens,
Dmitry Evtyushkin, Daniel Gruss,
https://arxiv.org/pdf/1811.05441.pdf
Fixes: b2157399cc98 ("bpf: prevent out-of-bounds speculation")
Reported-by: Jann Horn <jannh@google.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2019-01-03 07:58:34 +08:00
|
|
|
/* Possible states for alu_state member. */
|
|
|
|
|
#define BPF_ALU_SANITIZE_SRC 1U
|
|
|
|
|
#define BPF_ALU_SANITIZE_DST 2U
|
|
|
|
|
#define BPF_ALU_NEG_VALUE (1U << 2)
|
2019-01-06 07:54:37 +08:00
|
|
|
#define BPF_ALU_NON_POINTER (1U << 3)
|
bpf: prevent out of bounds speculation on pointer arithmetic
Jann reported that the original commit back in b2157399cc98
("bpf: prevent out-of-bounds speculation") was not sufficient
to stop CPU from speculating out of bounds memory access:
While b2157399cc98 only focussed on masking array map access
for unprivileged users for tail calls and data access such
that the user provided index gets sanitized from BPF program
and syscall side, there is still a more generic form affected
from BPF programs that applies to most maps that hold user
data in relation to dynamic map access when dealing with
unknown scalars or "slow" known scalars as access offset, for
example:
- Load a map value pointer into R6
- Load an index into R7
- Do a slow computation (e.g. with a memory dependency) that
loads a limit into R8 (e.g. load the limit from a map for
high latency, then mask it to make the verifier happy)
- Exit if R7 >= R8 (mispredicted branch)
- Load R0 = R6[R7]
- Load R0 = R6[R0]
For unknown scalars there are two options in the BPF verifier
where we could derive knowledge from in order to guarantee
safe access to the memory: i) While </>/<=/>= variants won't
allow to derive any lower or upper bounds from the unknown
scalar where it would be safe to add it to the map value
pointer, it is possible through ==/!= test however. ii) another
option is to transform the unknown scalar into a known scalar,
for example, through ALU ops combination such as R &= <imm>
followed by R |= <imm> or any similar combination where the
original information from the unknown scalar would be destroyed
entirely leaving R with a constant. The initial slow load still
precedes the latter ALU ops on that register, so the CPU
executes speculatively from that point. Once we have the known
scalar, any compare operation would work then. A third option
only involving registers with known scalars could be crafted
as described in [0] where a CPU port (e.g. Slow Int unit)
would be filled with many dependent computations such that
the subsequent condition depending on its outcome has to wait
for evaluation on its execution port and thereby executing
speculatively if the speculated code can be scheduled on a
different execution port, or any other form of mistraining
as described in [1], for example. Given this is not limited
to only unknown scalars, not only map but also stack access
is affected since both is accessible for unprivileged users
and could potentially be used for out of bounds access under
speculation.
In order to prevent any of these cases, the verifier is now
sanitizing pointer arithmetic on the offset such that any
out of bounds speculation would be masked in a way where the
pointer arithmetic result in the destination register will
stay unchanged, meaning offset masked into zero similar as
in array_index_nospec() case. With regards to implementation,
there are three options that were considered: i) new insn
for sanitation, ii) push/pop insn and sanitation as inlined
BPF, iii) reuse of ax register and sanitation as inlined BPF.
Option i) has the downside that we end up using from reserved
bits in the opcode space, but also that we would require
each JIT to emit masking as native arch opcodes meaning
mitigation would have slow adoption till everyone implements
it eventually which is counter-productive. Option ii) and iii)
have both in common that a temporary register is needed in
order to implement the sanitation as inlined BPF since we
are not allowed to modify the source register. While a push /
pop insn in ii) would be useful to have in any case, it
requires once again that every JIT needs to implement it
first. While possible, amount of changes needed would also
be unsuitable for a -stable patch. Therefore, the path which
has fewer changes, less BPF instructions for the mitigation
and does not require anything to be changed in the JITs is
option iii) which this work is pursuing. The ax register is
already mapped to a register in all JITs (modulo arm32 where
it's mapped to stack as various other BPF registers there)
and used in constant blinding for JITs-only so far. It can
be reused for verifier rewrites under certain constraints.
The interpreter's tmp "register" has therefore been remapped
into extending the register set with hidden ax register and
reusing that for a number of instructions that needed the
prior temporary variable internally (e.g. div, mod). This
allows for zero increase in stack space usage in the interpreter,
and enables (restricted) generic use in rewrites otherwise as
long as such a patchlet does not make use of these instructions.
The sanitation mask is dynamic and relative to the offset the
map value or stack pointer currently holds.
There are various cases that need to be taken under consideration
for the masking, e.g. such operation could look as follows:
ptr += val or val += ptr or ptr -= val. Thus, the value to be
sanitized could reside either in source or in destination
register, and the limit is different depending on whether
the ALU op is addition or subtraction and depending on the
current known and bounded offset. The limit is derived as
follows: limit := max_value_size - (smin_value + off). For
subtraction: limit := umax_value + off. This holds because
we do not allow any pointer arithmetic that would
temporarily go out of bounds or would have an unknown
value with mixed signed bounds where it is unclear at
verification time whether the actual runtime value would
be either negative or positive. For example, we have a
derived map pointer value with constant offset and bounded
one, so limit based on smin_value works because the verifier
requires that statically analyzed arithmetic on the pointer
must be in bounds, and thus it checks if resulting
smin_value + off and umax_value + off is still within map
value bounds at time of arithmetic in addition to time of
access. Similarly, for the case of stack access we derive
the limit as follows: MAX_BPF_STACK + off for subtraction
and -off for the case of addition where off := ptr_reg->off +
ptr_reg->var_off.value. Subtraction is a special case for
the masking which can be in form of ptr += -val, ptr -= -val,
or ptr -= val. In the first two cases where we know that
the value is negative, we need to temporarily negate the
value in order to do the sanitation on a positive value
where we later swap the ALU op, and restore original source
register if the value was in source.
The sanitation of pointer arithmetic alone is still not fully
sufficient as is, since a scenario like the following could
happen ...
PTR += 0x1000 (e.g. K-based imm)
PTR -= BIG_NUMBER_WITH_SLOW_COMPARISON
PTR += 0x1000
PTR -= BIG_NUMBER_WITH_SLOW_COMPARISON
[...]
... which under speculation could end up as ...
PTR += 0x1000
PTR -= 0 [ truncated by mitigation ]
PTR += 0x1000
PTR -= 0 [ truncated by mitigation ]
[...]
... and therefore still access out of bounds. To prevent such
case, the verifier is also analyzing safety for potential out
of bounds access under speculative execution. Meaning, it is
also simulating pointer access under truncation. We therefore
"branch off" and push the current verification state after the
ALU operation with known 0 to the verification stack for later
analysis. Given the current path analysis succeeded it is
likely that the one under speculation can be pruned. In any
case, it is also subject to existing complexity limits and
therefore anything beyond this point will be rejected. In
terms of pruning, it needs to be ensured that the verification
state from speculative execution simulation must never prune
a non-speculative execution path, therefore, we mark verifier
state accordingly at the time of push_stack(). If verifier
detects out of bounds access under speculative execution from
one of the possible paths that includes a truncation, it will
reject such program.
Given we mask every reg-based pointer arithmetic for
unprivileged programs, we've been looking into how it could
affect real-world programs in terms of size increase. As the
majority of programs are targeted for privileged-only use
case, we've unconditionally enabled masking (with its alu
restrictions on top of it) for privileged programs for the
sake of testing in order to check i) whether they get rejected
in its current form, and ii) by how much the number of
instructions and size will increase. We've tested this by
using Katran, Cilium and test_l4lb from the kernel selftests.
For Katran we've evaluated balancer_kern.o, Cilium bpf_lxc.o
and an older test object bpf_lxc_opt_-DUNKNOWN.o and l4lb
we've used test_l4lb.o as well as test_l4lb_noinline.o. We
found that none of the programs got rejected by the verifier
with this change, and that impact is rather minimal to none.
balancer_kern.o had 13,904 bytes (1,738 insns) xlated and
7,797 bytes JITed before and after the change. Most complex
program in bpf_lxc.o had 30,544 bytes (3,817 insns) xlated
and 18,538 bytes JITed before and after and none of the other
tail call programs in bpf_lxc.o had any changes either. For
the older bpf_lxc_opt_-DUNKNOWN.o object we found a small
increase from 20,616 bytes (2,576 insns) and 12,536 bytes JITed
before to 20,664 bytes (2,582 insns) and 12,558 bytes JITed
after the change. Other programs from that object file had
similar small increase. Both test_l4lb.o had no change and
remained at 6,544 bytes (817 insns) xlated and 3,401 bytes
JITed and for test_l4lb_noinline.o constant at 5,080 bytes
(634 insns) xlated and 3,313 bytes JITed. This can be explained
in that LLVM typically optimizes stack based pointer arithmetic
by using K-based operations and that use of dynamic map access
is not overly frequent. However, in future we may decide to
optimize the algorithm further under known guarantees from
branch and value speculation. Latter seems also unclear in
terms of prediction heuristics that today's CPUs apply as well
as whether there could be collisions in e.g. the predictor's
Value History/Pattern Table for triggering out of bounds access,
thus masking is performed unconditionally at this point but could
be subject to relaxation later on. We were generally also
brainstorming various other approaches for mitigation, but the
blocker was always lack of available registers at runtime and/or
overhead for runtime tracking of limits belonging to a specific
pointer. Thus, we found this to be minimally intrusive under
given constraints.
With that in place, a simple example with sanitized access on
unprivileged load at post-verification time looks as follows:
# bpftool prog dump xlated id 282
[...]
28: (79) r1 = *(u64 *)(r7 +0)
29: (79) r2 = *(u64 *)(r7 +8)
30: (57) r1 &= 15
31: (79) r3 = *(u64 *)(r0 +4608)
32: (57) r3 &= 1
33: (47) r3 |= 1
34: (2d) if r2 > r3 goto pc+19
35: (b4) (u32) r11 = (u32) 20479 |
36: (1f) r11 -= r2 | Dynamic sanitation for pointer
37: (4f) r11 |= r2 | arithmetic with registers
38: (87) r11 = -r11 | containing bounded or known
39: (c7) r11 s>>= 63 | scalars in order to prevent
40: (5f) r11 &= r2 | out of bounds speculation.
41: (0f) r4 += r11 |
42: (71) r4 = *(u8 *)(r4 +0)
43: (6f) r4 <<= r1
[...]
For the case where the scalar sits in the destination register
as opposed to the source register, the following code is emitted
for the above example:
[...]
16: (b4) (u32) r11 = (u32) 20479
17: (1f) r11 -= r2
18: (4f) r11 |= r2
19: (87) r11 = -r11
20: (c7) r11 s>>= 63
21: (5f) r2 &= r11
22: (0f) r2 += r0
23: (61) r0 = *(u32 *)(r2 +0)
[...]
JIT blinding example with non-conflicting use of r10:
[...]
d5: je 0x0000000000000106 _
d7: mov 0x0(%rax),%edi |
da: mov $0xf153246,%r10d | Index load from map value and
e0: xor $0xf153259,%r10 | (const blinded) mask with 0x1f.
e7: and %r10,%rdi |_
ea: mov $0x2f,%r10d |
f0: sub %rdi,%r10 | Sanitized addition. Both use r10
f3: or %rdi,%r10 | but do not interfere with each
f6: neg %r10 | other. (Neither do these instructions
f9: sar $0x3f,%r10 | interfere with the use of ax as temp
fd: and %r10,%rdi | in interpreter.)
100: add %rax,%rdi |_
103: mov 0x0(%rdi),%eax
[...]
Tested that it fixes Jann's reproducer, and also checked that test_verifier
and test_progs suite with interpreter, JIT and JIT with hardening enabled
on x86-64 and arm64 runs successfully.
[0] Speculose: Analyzing the Security Implications of Speculative
Execution in CPUs, Giorgi Maisuradze and Christian Rossow,
https://arxiv.org/pdf/1801.04084.pdf
[1] A Systematic Evaluation of Transient Execution Attacks and
Defenses, Claudio Canella, Jo Van Bulck, Michael Schwarz,
Moritz Lipp, Benjamin von Berg, Philipp Ortner, Frank Piessens,
Dmitry Evtyushkin, Daniel Gruss,
https://arxiv.org/pdf/1811.05441.pdf
Fixes: b2157399cc98 ("bpf: prevent out-of-bounds speculation")
Reported-by: Jann Horn <jannh@google.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2019-01-03 07:58:34 +08:00
|
|
|
#define BPF_ALU_SANITIZE (BPF_ALU_SANITIZE_SRC | \
|
|
|
|
|
BPF_ALU_SANITIZE_DST)
|
|
|
|
|
|
2016-09-21 18:43:57 +08:00
|
|
|
struct bpf_insn_aux_data {
|
2017-03-16 09:26:42 +08:00
|
|
|
union {
|
|
|
|
|
enum bpf_reg_type ptr_type; /* pointer type for load/store insns */
|
2019-11-23 04:07:59 +08:00
|
|
|
unsigned long map_ptr_state; /* pointer/poison value for maps */
|
2017-12-15 09:55:15 +08:00
|
|
|
s32 call_imm; /* saved imm field of call insn */
|
bpf: prevent out of bounds speculation on pointer arithmetic
Jann reported that the original commit back in b2157399cc98
("bpf: prevent out-of-bounds speculation") was not sufficient
to stop CPU from speculating out of bounds memory access:
While b2157399cc98 only focussed on masking array map access
for unprivileged users for tail calls and data access such
that the user provided index gets sanitized from BPF program
and syscall side, there is still a more generic form affected
from BPF programs that applies to most maps that hold user
data in relation to dynamic map access when dealing with
unknown scalars or "slow" known scalars as access offset, for
example:
- Load a map value pointer into R6
- Load an index into R7
- Do a slow computation (e.g. with a memory dependency) that
loads a limit into R8 (e.g. load the limit from a map for
high latency, then mask it to make the verifier happy)
- Exit if R7 >= R8 (mispredicted branch)
- Load R0 = R6[R7]
- Load R0 = R6[R0]
For unknown scalars there are two options in the BPF verifier
where we could derive knowledge from in order to guarantee
safe access to the memory: i) While </>/<=/>= variants won't
allow to derive any lower or upper bounds from the unknown
scalar where it would be safe to add it to the map value
pointer, it is possible through ==/!= test however. ii) another
option is to transform the unknown scalar into a known scalar,
for example, through ALU ops combination such as R &= <imm>
followed by R |= <imm> or any similar combination where the
original information from the unknown scalar would be destroyed
entirely leaving R with a constant. The initial slow load still
precedes the latter ALU ops on that register, so the CPU
executes speculatively from that point. Once we have the known
scalar, any compare operation would work then. A third option
only involving registers with known scalars could be crafted
as described in [0] where a CPU port (e.g. Slow Int unit)
would be filled with many dependent computations such that
the subsequent condition depending on its outcome has to wait
for evaluation on its execution port and thereby executing
speculatively if the speculated code can be scheduled on a
different execution port, or any other form of mistraining
as described in [1], for example. Given this is not limited
to only unknown scalars, not only map but also stack access
is affected since both is accessible for unprivileged users
and could potentially be used for out of bounds access under
speculation.
In order to prevent any of these cases, the verifier is now
sanitizing pointer arithmetic on the offset such that any
out of bounds speculation would be masked in a way where the
pointer arithmetic result in the destination register will
stay unchanged, meaning offset masked into zero similar as
in array_index_nospec() case. With regards to implementation,
there are three options that were considered: i) new insn
for sanitation, ii) push/pop insn and sanitation as inlined
BPF, iii) reuse of ax register and sanitation as inlined BPF.
Option i) has the downside that we end up using from reserved
bits in the opcode space, but also that we would require
each JIT to emit masking as native arch opcodes meaning
mitigation would have slow adoption till everyone implements
it eventually which is counter-productive. Option ii) and iii)
have both in common that a temporary register is needed in
order to implement the sanitation as inlined BPF since we
are not allowed to modify the source register. While a push /
pop insn in ii) would be useful to have in any case, it
requires once again that every JIT needs to implement it
first. While possible, amount of changes needed would also
be unsuitable for a -stable patch. Therefore, the path which
has fewer changes, less BPF instructions for the mitigation
and does not require anything to be changed in the JITs is
option iii) which this work is pursuing. The ax register is
already mapped to a register in all JITs (modulo arm32 where
it's mapped to stack as various other BPF registers there)
and used in constant blinding for JITs-only so far. It can
be reused for verifier rewrites under certain constraints.
The interpreter's tmp "register" has therefore been remapped
into extending the register set with hidden ax register and
reusing that for a number of instructions that needed the
prior temporary variable internally (e.g. div, mod). This
allows for zero increase in stack space usage in the interpreter,
and enables (restricted) generic use in rewrites otherwise as
long as such a patchlet does not make use of these instructions.
The sanitation mask is dynamic and relative to the offset the
map value or stack pointer currently holds.
There are various cases that need to be taken under consideration
for the masking, e.g. such operation could look as follows:
ptr += val or val += ptr or ptr -= val. Thus, the value to be
sanitized could reside either in source or in destination
register, and the limit is different depending on whether
the ALU op is addition or subtraction and depending on the
current known and bounded offset. The limit is derived as
follows: limit := max_value_size - (smin_value + off). For
subtraction: limit := umax_value + off. This holds because
we do not allow any pointer arithmetic that would
temporarily go out of bounds or would have an unknown
value with mixed signed bounds where it is unclear at
verification time whether the actual runtime value would
be either negative or positive. For example, we have a
derived map pointer value with constant offset and bounded
one, so limit based on smin_value works because the verifier
requires that statically analyzed arithmetic on the pointer
must be in bounds, and thus it checks if resulting
smin_value + off and umax_value + off is still within map
value bounds at time of arithmetic in addition to time of
access. Similarly, for the case of stack access we derive
the limit as follows: MAX_BPF_STACK + off for subtraction
and -off for the case of addition where off := ptr_reg->off +
ptr_reg->var_off.value. Subtraction is a special case for
the masking which can be in form of ptr += -val, ptr -= -val,
or ptr -= val. In the first two cases where we know that
the value is negative, we need to temporarily negate the
value in order to do the sanitation on a positive value
where we later swap the ALU op, and restore original source
register if the value was in source.
The sanitation of pointer arithmetic alone is still not fully
sufficient as is, since a scenario like the following could
happen ...
PTR += 0x1000 (e.g. K-based imm)
PTR -= BIG_NUMBER_WITH_SLOW_COMPARISON
PTR += 0x1000
PTR -= BIG_NUMBER_WITH_SLOW_COMPARISON
[...]
... which under speculation could end up as ...
PTR += 0x1000
PTR -= 0 [ truncated by mitigation ]
PTR += 0x1000
PTR -= 0 [ truncated by mitigation ]
[...]
... and therefore still access out of bounds. To prevent such
case, the verifier is also analyzing safety for potential out
of bounds access under speculative execution. Meaning, it is
also simulating pointer access under truncation. We therefore
"branch off" and push the current verification state after the
ALU operation with known 0 to the verification stack for later
analysis. Given the current path analysis succeeded it is
likely that the one under speculation can be pruned. In any
case, it is also subject to existing complexity limits and
therefore anything beyond this point will be rejected. In
terms of pruning, it needs to be ensured that the verification
state from speculative execution simulation must never prune
a non-speculative execution path, therefore, we mark verifier
state accordingly at the time of push_stack(). If verifier
detects out of bounds access under speculative execution from
one of the possible paths that includes a truncation, it will
reject such program.
Given we mask every reg-based pointer arithmetic for
unprivileged programs, we've been looking into how it could
affect real-world programs in terms of size increase. As the
majority of programs are targeted for privileged-only use
case, we've unconditionally enabled masking (with its alu
restrictions on top of it) for privileged programs for the
sake of testing in order to check i) whether they get rejected
in its current form, and ii) by how much the number of
instructions and size will increase. We've tested this by
using Katran, Cilium and test_l4lb from the kernel selftests.
For Katran we've evaluated balancer_kern.o, Cilium bpf_lxc.o
and an older test object bpf_lxc_opt_-DUNKNOWN.o and l4lb
we've used test_l4lb.o as well as test_l4lb_noinline.o. We
found that none of the programs got rejected by the verifier
with this change, and that impact is rather minimal to none.
balancer_kern.o had 13,904 bytes (1,738 insns) xlated and
7,797 bytes JITed before and after the change. Most complex
program in bpf_lxc.o had 30,544 bytes (3,817 insns) xlated
and 18,538 bytes JITed before and after and none of the other
tail call programs in bpf_lxc.o had any changes either. For
the older bpf_lxc_opt_-DUNKNOWN.o object we found a small
increase from 20,616 bytes (2,576 insns) and 12,536 bytes JITed
before to 20,664 bytes (2,582 insns) and 12,558 bytes JITed
after the change. Other programs from that object file had
similar small increase. Both test_l4lb.o had no change and
remained at 6,544 bytes (817 insns) xlated and 3,401 bytes
JITed and for test_l4lb_noinline.o constant at 5,080 bytes
(634 insns) xlated and 3,313 bytes JITed. This can be explained
in that LLVM typically optimizes stack based pointer arithmetic
by using K-based operations and that use of dynamic map access
is not overly frequent. However, in future we may decide to
optimize the algorithm further under known guarantees from
branch and value speculation. Latter seems also unclear in
terms of prediction heuristics that today's CPUs apply as well
as whether there could be collisions in e.g. the predictor's
Value History/Pattern Table for triggering out of bounds access,
thus masking is performed unconditionally at this point but could
be subject to relaxation later on. We were generally also
brainstorming various other approaches for mitigation, but the
blocker was always lack of available registers at runtime and/or
overhead for runtime tracking of limits belonging to a specific
pointer. Thus, we found this to be minimally intrusive under
given constraints.
With that in place, a simple example with sanitized access on
unprivileged load at post-verification time looks as follows:
# bpftool prog dump xlated id 282
[...]
28: (79) r1 = *(u64 *)(r7 +0)
29: (79) r2 = *(u64 *)(r7 +8)
30: (57) r1 &= 15
31: (79) r3 = *(u64 *)(r0 +4608)
32: (57) r3 &= 1
33: (47) r3 |= 1
34: (2d) if r2 > r3 goto pc+19
35: (b4) (u32) r11 = (u32) 20479 |
36: (1f) r11 -= r2 | Dynamic sanitation for pointer
37: (4f) r11 |= r2 | arithmetic with registers
38: (87) r11 = -r11 | containing bounded or known
39: (c7) r11 s>>= 63 | scalars in order to prevent
40: (5f) r11 &= r2 | out of bounds speculation.
41: (0f) r4 += r11 |
42: (71) r4 = *(u8 *)(r4 +0)
43: (6f) r4 <<= r1
[...]
For the case where the scalar sits in the destination register
as opposed to the source register, the following code is emitted
for the above example:
[...]
16: (b4) (u32) r11 = (u32) 20479
17: (1f) r11 -= r2
18: (4f) r11 |= r2
19: (87) r11 = -r11
20: (c7) r11 s>>= 63
21: (5f) r2 &= r11
22: (0f) r2 += r0
23: (61) r0 = *(u32 *)(r2 +0)
[...]
JIT blinding example with non-conflicting use of r10:
[...]
d5: je 0x0000000000000106 _
d7: mov 0x0(%rax),%edi |
da: mov $0xf153246,%r10d | Index load from map value and
e0: xor $0xf153259,%r10 | (const blinded) mask with 0x1f.
e7: and %r10,%rdi |_
ea: mov $0x2f,%r10d |
f0: sub %rdi,%r10 | Sanitized addition. Both use r10
f3: or %rdi,%r10 | but do not interfere with each
f6: neg %r10 | other. (Neither do these instructions
f9: sar $0x3f,%r10 | interfere with the use of ax as temp
fd: and %r10,%rdi | in interpreter.)
100: add %rax,%rdi |_
103: mov 0x0(%rdi),%eax
[...]
Tested that it fixes Jann's reproducer, and also checked that test_verifier
and test_progs suite with interpreter, JIT and JIT with hardening enabled
on x86-64 and arm64 runs successfully.
[0] Speculose: Analyzing the Security Implications of Speculative
Execution in CPUs, Giorgi Maisuradze and Christian Rossow,
https://arxiv.org/pdf/1801.04084.pdf
[1] A Systematic Evaluation of Transient Execution Attacks and
Defenses, Claudio Canella, Jo Van Bulck, Michael Schwarz,
Moritz Lipp, Benjamin von Berg, Philipp Ortner, Frank Piessens,
Dmitry Evtyushkin, Daniel Gruss,
https://arxiv.org/pdf/1811.05441.pdf
Fixes: b2157399cc98 ("bpf: prevent out-of-bounds speculation")
Reported-by: Jann Horn <jannh@google.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2019-01-03 07:58:34 +08:00
|
|
|
u32 alu_limit; /* limit for add/sub register with pointer */
|
bpf: implement lookup-free direct value access for maps
This generic extension to BPF maps allows for directly loading
an address residing inside a BPF map value as a single BPF
ldimm64 instruction!
The idea is similar to what BPF_PSEUDO_MAP_FD does today, which
is a special src_reg flag for ldimm64 instruction that indicates
that inside the first part of the double insns's imm field is a
file descriptor which the verifier then replaces as a full 64bit
address of the map into both imm parts. For the newly added
BPF_PSEUDO_MAP_VALUE src_reg flag, the idea is the following:
the first part of the double insns's imm field is again a file
descriptor corresponding to the map, and the second part of the
imm field is an offset into the value. The verifier will then
replace both imm parts with an address that points into the BPF
map value at the given value offset for maps that support this
operation. Currently supported is array map with single entry.
It is possible to support more than just single map element by
reusing both 16bit off fields of the insns as a map index, so
full array map lookup could be expressed that way. It hasn't
been implemented here due to lack of concrete use case, but
could easily be done so in future in a compatible way, since
both off fields right now have to be 0 and would correctly
denote a map index 0.
The BPF_PSEUDO_MAP_VALUE is a distinct flag as otherwise with
BPF_PSEUDO_MAP_FD we could not differ offset 0 between load of
map pointer versus load of map's value at offset 0, and changing
BPF_PSEUDO_MAP_FD's encoding into off by one to differ between
regular map pointer and map value pointer would add unnecessary
complexity and increases barrier for debugability thus less
suitable. Using the second part of the imm field as an offset
into the value does /not/ come with limitations since maximum
possible value size is in u32 universe anyway.
This optimization allows for efficiently retrieving an address
to a map value memory area without having to issue a helper call
which needs to prepare registers according to calling convention,
etc, without needing the extra NULL test, and without having to
add the offset in an additional instruction to the value base
pointer. The verifier then treats the destination register as
PTR_TO_MAP_VALUE with constant reg->off from the user passed
offset from the second imm field, and guarantees that this is
within bounds of the map value. Any subsequent operations are
normally treated as typical map value handling without anything
extra needed from verification side.
The two map operations for direct value access have been added to
array map for now. In future other types could be supported as
well depending on the use case. The main use case for this commit
is to allow for BPF loader support for global variables that
reside in .data/.rodata/.bss sections such that we can directly
load the address of them with minimal additional infrastructure
required. Loader support has been added in subsequent commits for
libbpf library.
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2019-04-10 05:20:03 +08:00
|
|
|
struct {
|
|
|
|
|
u32 map_index; /* index into used_maps[] */
|
|
|
|
|
u32 map_off; /* offset from value base address */
|
|
|
|
|
};
|
2017-03-16 09:26:42 +08:00
|
|
|
};
|
2019-11-23 04:07:59 +08:00
|
|
|
u64 map_key_state; /* constant (32 bit) key tracking for maps */
|
2017-06-23 06:07:39 +08:00
|
|
|
int ctx_field_size; /* the ctx field size for load insn, maybe 0 */
|
2018-05-16 00:27:05 +08:00
|
|
|
int sanitize_stack_off; /* stack slot to be cleared */
|
2020-01-10 14:41:20 +08:00
|
|
|
u32 seen; /* this insn was processed by the verifier at env->pass_cnt */
|
bpf: verifier: mark verified-insn with sub-register zext flag
eBPF ISA specification requires high 32-bit cleared when low 32-bit
sub-register is written. This applies to destination register of ALU32 etc.
JIT back-ends must guarantee this semantic when doing code-gen. x86_64 and
AArch64 ISA has the same semantics, so the corresponding JIT back-end
doesn't need to do extra work.
However, 32-bit arches (arm, x86, nfp etc.) and some other 64-bit arches
(PowerPC, SPARC etc) need to do explicit zero extension to meet this
requirement, otherwise code like the following will fail.
u64_value = (u64) u32_value
... other uses of u64_value
This is because compiler could exploit the semantic described above and
save those zero extensions for extending u32_value to u64_value, these JIT
back-ends are expected to guarantee this through inserting extra zero
extensions which however could be a significant increase on the code size.
Some benchmarks show there could be ~40% sub-register writes out of total
insns, meaning at least ~40% extra code-gen.
One observation is these extra zero extensions are not always necessary.
Take above code snippet for example, it is possible u32_value will never be
casted into a u64, the value of high 32-bit of u32_value then could be
ignored and extra zero extension could be eliminated.
This patch implements this idea, insns defining sub-registers will be
marked when the high 32-bit of the defined sub-register matters. For
those unmarked insns, it is safe to eliminate high 32-bit clearnace for
them.
Algo:
- Split read flags into READ32 and READ64.
- Record index of insn that does sub-register write. Keep the index inside
reg state and update it during verifier insn walking.
- A full register read on a sub-register marks its definition insn as
needing zero extension on dst register.
A new sub-register write overrides the old one.
- When propagating read64 during path pruning, also mark any insn defining
a sub-register that is read in the pruned path as full-register.
Reviewed-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Signed-off-by: Jiong Wang <jiong.wang@netronome.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2019-05-25 06:25:12 +08:00
|
|
|
bool zext_dst; /* this insn zero extends dst reg */
|
bpf: prevent out of bounds speculation on pointer arithmetic
Jann reported that the original commit back in b2157399cc98
("bpf: prevent out-of-bounds speculation") was not sufficient
to stop CPU from speculating out of bounds memory access:
While b2157399cc98 only focussed on masking array map access
for unprivileged users for tail calls and data access such
that the user provided index gets sanitized from BPF program
and syscall side, there is still a more generic form affected
from BPF programs that applies to most maps that hold user
data in relation to dynamic map access when dealing with
unknown scalars or "slow" known scalars as access offset, for
example:
- Load a map value pointer into R6
- Load an index into R7
- Do a slow computation (e.g. with a memory dependency) that
loads a limit into R8 (e.g. load the limit from a map for
high latency, then mask it to make the verifier happy)
- Exit if R7 >= R8 (mispredicted branch)
- Load R0 = R6[R7]
- Load R0 = R6[R0]
For unknown scalars there are two options in the BPF verifier
where we could derive knowledge from in order to guarantee
safe access to the memory: i) While </>/<=/>= variants won't
allow to derive any lower or upper bounds from the unknown
scalar where it would be safe to add it to the map value
pointer, it is possible through ==/!= test however. ii) another
option is to transform the unknown scalar into a known scalar,
for example, through ALU ops combination such as R &= <imm>
followed by R |= <imm> or any similar combination where the
original information from the unknown scalar would be destroyed
entirely leaving R with a constant. The initial slow load still
precedes the latter ALU ops on that register, so the CPU
executes speculatively from that point. Once we have the known
scalar, any compare operation would work then. A third option
only involving registers with known scalars could be crafted
as described in [0] where a CPU port (e.g. Slow Int unit)
would be filled with many dependent computations such that
the subsequent condition depending on its outcome has to wait
for evaluation on its execution port and thereby executing
speculatively if the speculated code can be scheduled on a
different execution port, or any other form of mistraining
as described in [1], for example. Given this is not limited
to only unknown scalars, not only map but also stack access
is affected since both is accessible for unprivileged users
and could potentially be used for out of bounds access under
speculation.
In order to prevent any of these cases, the verifier is now
sanitizing pointer arithmetic on the offset such that any
out of bounds speculation would be masked in a way where the
pointer arithmetic result in the destination register will
stay unchanged, meaning offset masked into zero similar as
in array_index_nospec() case. With regards to implementation,
there are three options that were considered: i) new insn
for sanitation, ii) push/pop insn and sanitation as inlined
BPF, iii) reuse of ax register and sanitation as inlined BPF.
Option i) has the downside that we end up using from reserved
bits in the opcode space, but also that we would require
each JIT to emit masking as native arch opcodes meaning
mitigation would have slow adoption till everyone implements
it eventually which is counter-productive. Option ii) and iii)
have both in common that a temporary register is needed in
order to implement the sanitation as inlined BPF since we
are not allowed to modify the source register. While a push /
pop insn in ii) would be useful to have in any case, it
requires once again that every JIT needs to implement it
first. While possible, amount of changes needed would also
be unsuitable for a -stable patch. Therefore, the path which
has fewer changes, less BPF instructions for the mitigation
and does not require anything to be changed in the JITs is
option iii) which this work is pursuing. The ax register is
already mapped to a register in all JITs (modulo arm32 where
it's mapped to stack as various other BPF registers there)
and used in constant blinding for JITs-only so far. It can
be reused for verifier rewrites under certain constraints.
The interpreter's tmp "register" has therefore been remapped
into extending the register set with hidden ax register and
reusing that for a number of instructions that needed the
prior temporary variable internally (e.g. div, mod). This
allows for zero increase in stack space usage in the interpreter,
and enables (restricted) generic use in rewrites otherwise as
long as such a patchlet does not make use of these instructions.
The sanitation mask is dynamic and relative to the offset the
map value or stack pointer currently holds.
There are various cases that need to be taken under consideration
for the masking, e.g. such operation could look as follows:
ptr += val or val += ptr or ptr -= val. Thus, the value to be
sanitized could reside either in source or in destination
register, and the limit is different depending on whether
the ALU op is addition or subtraction and depending on the
current known and bounded offset. The limit is derived as
follows: limit := max_value_size - (smin_value + off). For
subtraction: limit := umax_value + off. This holds because
we do not allow any pointer arithmetic that would
temporarily go out of bounds or would have an unknown
value with mixed signed bounds where it is unclear at
verification time whether the actual runtime value would
be either negative or positive. For example, we have a
derived map pointer value with constant offset and bounded
one, so limit based on smin_value works because the verifier
requires that statically analyzed arithmetic on the pointer
must be in bounds, and thus it checks if resulting
smin_value + off and umax_value + off is still within map
value bounds at time of arithmetic in addition to time of
access. Similarly, for the case of stack access we derive
the limit as follows: MAX_BPF_STACK + off for subtraction
and -off for the case of addition where off := ptr_reg->off +
ptr_reg->var_off.value. Subtraction is a special case for
the masking which can be in form of ptr += -val, ptr -= -val,
or ptr -= val. In the first two cases where we know that
the value is negative, we need to temporarily negate the
value in order to do the sanitation on a positive value
where we later swap the ALU op, and restore original source
register if the value was in source.
The sanitation of pointer arithmetic alone is still not fully
sufficient as is, since a scenario like the following could
happen ...
PTR += 0x1000 (e.g. K-based imm)
PTR -= BIG_NUMBER_WITH_SLOW_COMPARISON
PTR += 0x1000
PTR -= BIG_NUMBER_WITH_SLOW_COMPARISON
[...]
... which under speculation could end up as ...
PTR += 0x1000
PTR -= 0 [ truncated by mitigation ]
PTR += 0x1000
PTR -= 0 [ truncated by mitigation ]
[...]
... and therefore still access out of bounds. To prevent such
case, the verifier is also analyzing safety for potential out
of bounds access under speculative execution. Meaning, it is
also simulating pointer access under truncation. We therefore
"branch off" and push the current verification state after the
ALU operation with known 0 to the verification stack for later
analysis. Given the current path analysis succeeded it is
likely that the one under speculation can be pruned. In any
case, it is also subject to existing complexity limits and
therefore anything beyond this point will be rejected. In
terms of pruning, it needs to be ensured that the verification
state from speculative execution simulation must never prune
a non-speculative execution path, therefore, we mark verifier
state accordingly at the time of push_stack(). If verifier
detects out of bounds access under speculative execution from
one of the possible paths that includes a truncation, it will
reject such program.
Given we mask every reg-based pointer arithmetic for
unprivileged programs, we've been looking into how it could
affect real-world programs in terms of size increase. As the
majority of programs are targeted for privileged-only use
case, we've unconditionally enabled masking (with its alu
restrictions on top of it) for privileged programs for the
sake of testing in order to check i) whether they get rejected
in its current form, and ii) by how much the number of
instructions and size will increase. We've tested this by
using Katran, Cilium and test_l4lb from the kernel selftests.
For Katran we've evaluated balancer_kern.o, Cilium bpf_lxc.o
and an older test object bpf_lxc_opt_-DUNKNOWN.o and l4lb
we've used test_l4lb.o as well as test_l4lb_noinline.o. We
found that none of the programs got rejected by the verifier
with this change, and that impact is rather minimal to none.
balancer_kern.o had 13,904 bytes (1,738 insns) xlated and
7,797 bytes JITed before and after the change. Most complex
program in bpf_lxc.o had 30,544 bytes (3,817 insns) xlated
and 18,538 bytes JITed before and after and none of the other
tail call programs in bpf_lxc.o had any changes either. For
the older bpf_lxc_opt_-DUNKNOWN.o object we found a small
increase from 20,616 bytes (2,576 insns) and 12,536 bytes JITed
before to 20,664 bytes (2,582 insns) and 12,558 bytes JITed
after the change. Other programs from that object file had
similar small increase. Both test_l4lb.o had no change and
remained at 6,544 bytes (817 insns) xlated and 3,401 bytes
JITed and for test_l4lb_noinline.o constant at 5,080 bytes
(634 insns) xlated and 3,313 bytes JITed. This can be explained
in that LLVM typically optimizes stack based pointer arithmetic
by using K-based operations and that use of dynamic map access
is not overly frequent. However, in future we may decide to
optimize the algorithm further under known guarantees from
branch and value speculation. Latter seems also unclear in
terms of prediction heuristics that today's CPUs apply as well
as whether there could be collisions in e.g. the predictor's
Value History/Pattern Table for triggering out of bounds access,
thus masking is performed unconditionally at this point but could
be subject to relaxation later on. We were generally also
brainstorming various other approaches for mitigation, but the
blocker was always lack of available registers at runtime and/or
overhead for runtime tracking of limits belonging to a specific
pointer. Thus, we found this to be minimally intrusive under
given constraints.
With that in place, a simple example with sanitized access on
unprivileged load at post-verification time looks as follows:
# bpftool prog dump xlated id 282
[...]
28: (79) r1 = *(u64 *)(r7 +0)
29: (79) r2 = *(u64 *)(r7 +8)
30: (57) r1 &= 15
31: (79) r3 = *(u64 *)(r0 +4608)
32: (57) r3 &= 1
33: (47) r3 |= 1
34: (2d) if r2 > r3 goto pc+19
35: (b4) (u32) r11 = (u32) 20479 |
36: (1f) r11 -= r2 | Dynamic sanitation for pointer
37: (4f) r11 |= r2 | arithmetic with registers
38: (87) r11 = -r11 | containing bounded or known
39: (c7) r11 s>>= 63 | scalars in order to prevent
40: (5f) r11 &= r2 | out of bounds speculation.
41: (0f) r4 += r11 |
42: (71) r4 = *(u8 *)(r4 +0)
43: (6f) r4 <<= r1
[...]
For the case where the scalar sits in the destination register
as opposed to the source register, the following code is emitted
for the above example:
[...]
16: (b4) (u32) r11 = (u32) 20479
17: (1f) r11 -= r2
18: (4f) r11 |= r2
19: (87) r11 = -r11
20: (c7) r11 s>>= 63
21: (5f) r2 &= r11
22: (0f) r2 += r0
23: (61) r0 = *(u32 *)(r2 +0)
[...]
JIT blinding example with non-conflicting use of r10:
[...]
d5: je 0x0000000000000106 _
d7: mov 0x0(%rax),%edi |
da: mov $0xf153246,%r10d | Index load from map value and
e0: xor $0xf153259,%r10 | (const blinded) mask with 0x1f.
e7: and %r10,%rdi |_
ea: mov $0x2f,%r10d |
f0: sub %rdi,%r10 | Sanitized addition. Both use r10
f3: or %rdi,%r10 | but do not interfere with each
f6: neg %r10 | other. (Neither do these instructions
f9: sar $0x3f,%r10 | interfere with the use of ax as temp
fd: and %r10,%rdi | in interpreter.)
100: add %rax,%rdi |_
103: mov 0x0(%rdi),%eax
[...]
Tested that it fixes Jann's reproducer, and also checked that test_verifier
and test_progs suite with interpreter, JIT and JIT with hardening enabled
on x86-64 and arm64 runs successfully.
[0] Speculose: Analyzing the Security Implications of Speculative
Execution in CPUs, Giorgi Maisuradze and Christian Rossow,
https://arxiv.org/pdf/1801.04084.pdf
[1] A Systematic Evaluation of Transient Execution Attacks and
Defenses, Claudio Canella, Jo Van Bulck, Michael Schwarz,
Moritz Lipp, Benjamin von Berg, Philipp Ortner, Frank Piessens,
Dmitry Evtyushkin, Daniel Gruss,
https://arxiv.org/pdf/1811.05441.pdf
Fixes: b2157399cc98 ("bpf: prevent out-of-bounds speculation")
Reported-by: Jann Horn <jannh@google.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2019-01-03 07:58:34 +08:00
|
|
|
u8 alu_state; /* used in combination with alu_limit */
|
2020-01-10 14:41:20 +08:00
|
|
|
|
|
|
|
|
/* below fields are initialized once */
|
2019-01-23 14:45:23 +08:00
|
|
|
unsigned int orig_idx; /* original instruction index */
|
2020-01-10 14:41:20 +08:00
|
|
|
bool prune_point;
|
2016-09-21 18:43:57 +08:00
|
|
|
};
|
|
|
|
|
|
|
|
|
|
#define MAX_USED_MAPS 64 /* max number of maps accessed by one eBPF program */
|
|
|
|
|
|
2017-10-10 01:30:15 +08:00
|
|
|
#define BPF_VERIFIER_TMP_LOG_SIZE 1024
|
|
|
|
|
|
2018-03-25 02:44:22 +08:00
|
|
|
struct bpf_verifier_log {
|
2017-10-10 01:30:10 +08:00
|
|
|
u32 level;
|
2017-10-10 01:30:15 +08:00
|
|
|
char kbuf[BPF_VERIFIER_TMP_LOG_SIZE];
|
2017-10-10 01:30:10 +08:00
|
|
|
char __user *ubuf;
|
|
|
|
|
u32 len_used;
|
|
|
|
|
u32 len_total;
|
|
|
|
|
};
|
|
|
|
|
|
2018-03-25 02:44:22 +08:00
|
|
|
static inline bool bpf_verifier_log_full(const struct bpf_verifier_log *log)
|
2017-10-10 01:30:10 +08:00
|
|
|
{
|
|
|
|
|
return log->len_used >= log->len_total - 1;
|
|
|
|
|
}
|
|
|
|
|
|
2019-04-02 12:27:40 +08:00
|
|
|
#define BPF_LOG_LEVEL1 1
|
|
|
|
|
#define BPF_LOG_LEVEL2 2
|
|
|
|
|
#define BPF_LOG_STATS 4
|
|
|
|
|
#define BPF_LOG_LEVEL (BPF_LOG_LEVEL1 | BPF_LOG_LEVEL2)
|
|
|
|
|
#define BPF_LOG_MASK (BPF_LOG_LEVEL | BPF_LOG_STATS)
|
2019-10-16 11:24:57 +08:00
|
|
|
#define BPF_LOG_KERNEL (BPF_LOG_MASK + 1) /* kernel internal flag */
|
2019-04-02 12:27:40 +08:00
|
|
|
|
2018-03-25 02:44:23 +08:00
|
|
|
static inline bool bpf_verifier_log_needed(const struct bpf_verifier_log *log)
|
|
|
|
|
{
|
2019-10-16 11:24:57 +08:00
|
|
|
return (log->level && log->ubuf && !bpf_verifier_log_full(log)) ||
|
|
|
|
|
log->level == BPF_LOG_KERNEL;
|
2018-03-25 02:44:23 +08:00
|
|
|
}
|
|
|
|
|
|
2017-12-15 09:55:05 +08:00
|
|
|
#define BPF_MAX_SUBPROGS 256
|
|
|
|
|
|
2018-05-03 04:17:18 +08:00
|
|
|
struct bpf_subprog_info {
|
2019-11-15 02:57:16 +08:00
|
|
|
/* 'start' has to be the first field otherwise find_subprog() won't work */
|
2018-05-03 04:17:18 +08:00
|
|
|
u32 start; /* insn idx of function entry point */
|
2018-12-08 08:42:25 +08:00
|
|
|
u32 linfo_idx; /* The idx to the main_prog->aux->linfo */
|
2018-05-03 04:17:18 +08:00
|
|
|
u16 stack_depth; /* max. stack depth used by this function */
|
|
|
|
|
};
|
|
|
|
|
|
2016-09-21 18:43:57 +08:00
|
|
|
/* single container for all structs
|
|
|
|
|
* one verifier_env per bpf_check() call
|
|
|
|
|
*/
|
|
|
|
|
struct bpf_verifier_env {
|
2019-01-03 07:58:27 +08:00
|
|
|
u32 insn_idx;
|
|
|
|
|
u32 prev_insn_idx;
|
2016-09-21 18:43:57 +08:00
|
|
|
struct bpf_prog *prog; /* eBPF program being verified */
|
2017-10-17 07:40:54 +08:00
|
|
|
const struct bpf_verifier_ops *ops;
|
2016-09-21 18:43:57 +08:00
|
|
|
struct bpf_verifier_stack_elem *head; /* stack of verifier states to be processed */
|
|
|
|
|
int stack_size; /* number of states to be processed */
|
2017-05-11 02:38:07 +08:00
|
|
|
bool strict_alignment; /* perform strict pointer alignment checks */
|
2019-08-23 13:52:12 +08:00
|
|
|
bool test_state_freq; /* test verifier with different pruning frequency */
|
2017-11-01 09:16:05 +08:00
|
|
|
struct bpf_verifier_state *cur_state; /* current verifier state */
|
2016-09-21 18:43:57 +08:00
|
|
|
struct bpf_verifier_state_list **explored_states; /* search pruning optimization */
|
2019-04-02 12:27:41 +08:00
|
|
|
struct bpf_verifier_state_list *free_list;
|
2016-09-21 18:43:57 +08:00
|
|
|
struct bpf_map *used_maps[MAX_USED_MAPS]; /* array of map's used by eBPF program */
|
|
|
|
|
u32 used_map_cnt; /* number of used maps */
|
|
|
|
|
u32 id_gen; /* used to generate unique reg IDs */
|
|
|
|
|
bool allow_ptr_leaks;
|
bpf: Support access to bpf map fields
There are multiple use-cases when it's convenient to have access to bpf
map fields, both `struct bpf_map` and map type specific struct-s such as
`struct bpf_array`, `struct bpf_htab`, etc.
For example while working with sock arrays it can be necessary to
calculate the key based on map->max_entries (some_hash % max_entries).
Currently this is solved by communicating max_entries via "out-of-band"
channel, e.g. via additional map with known key to get info about target
map. That works, but is not very convenient and error-prone while
working with many maps.
In other cases necessary data is dynamic (i.e. unknown at loading time)
and it's impossible to get it at all. For example while working with a
hash table it can be convenient to know how much capacity is already
used (bpf_htab.count.counter for BPF_F_NO_PREALLOC case).
At the same time kernel knows this info and can provide it to bpf
program.
Fill this gap by adding support to access bpf map fields from bpf
program for both `struct bpf_map` and map type specific fields.
Support is implemented via btf_struct_access() so that a user can define
their own `struct bpf_map` or map type specific struct in their program
with only necessary fields and preserve_access_index attribute, cast a
map to this struct and use a field.
For example:
struct bpf_map {
__u32 max_entries;
} __attribute__((preserve_access_index));
struct bpf_array {
struct bpf_map map;
__u32 elem_size;
} __attribute__((preserve_access_index));
struct {
__uint(type, BPF_MAP_TYPE_ARRAY);
__uint(max_entries, 4);
__type(key, __u32);
__type(value, __u32);
} m_array SEC(".maps");
SEC("cgroup_skb/egress")
int cg_skb(void *ctx)
{
struct bpf_array *array = (struct bpf_array *)&m_array;
struct bpf_map *map = (struct bpf_map *)&m_array;
/* .. use map->max_entries or array->map.max_entries .. */
}
Similarly to other btf_struct_access() use-cases (e.g. struct tcp_sock
in net/ipv4/bpf_tcp_ca.c) the patch allows access to any fields of
corresponding struct. Only reading from map fields is supported.
For btf_struct_access() to work there should be a way to know btf id of
a struct that corresponds to a map type. To get btf id there should be a
way to get a stringified name of map-specific struct, such as
"bpf_array", "bpf_htab", etc for a map type. Two new fields are added to
`struct bpf_map_ops` to handle it:
* .map_btf_name keeps a btf name of a struct returned by map_alloc();
* .map_btf_id is used to cache btf id of that struct.
To make btf ids calculation cheaper they're calculated once while
preparing btf_vmlinux and cached same way as it's done for btf_id field
of `struct bpf_func_proto`
While calculating btf ids, struct names are NOT checked for collision.
Collisions will be checked as a part of the work to prepare btf ids used
in verifier in compile time that should land soon. The only known
collision for `struct bpf_htab` (kernel/bpf/hashtab.c vs
net/core/sock_map.c) was fixed earlier.
Both new fields .map_btf_name and .map_btf_id must be set for a map type
for the feature to work. If neither is set for a map type, verifier will
return ENOTSUPP on a try to access map_ptr of corresponding type. If
just one of them set, it's verifier misconfiguration.
Only `struct bpf_array` for BPF_MAP_TYPE_ARRAY and `struct bpf_htab` for
BPF_MAP_TYPE_HASH are supported by this patch. Other map types will be
supported separately.
The feature is available only for CONFIG_DEBUG_INFO_BTF=y and gated by
perfmon_capable() so that unpriv programs won't have access to bpf map
fields.
Signed-off-by: Andrey Ignatov <rdna@fb.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: John Fastabend <john.fastabend@gmail.com>
Acked-by: Martin KaFai Lau <kafai@fb.com>
Link: https://lore.kernel.org/bpf/6479686a0cd1e9067993df57b4c3eef0e276fec9.1592600985.git.rdna@fb.com
2020-06-20 05:11:43 +08:00
|
|
|
bool allow_ptr_to_map_access;
|
2020-05-14 07:03:54 +08:00
|
|
|
bool bpf_capable;
|
|
|
|
|
bool bypass_spec_v1;
|
|
|
|
|
bool bypass_spec_v4;
|
2016-09-21 18:43:57 +08:00
|
|
|
bool seen_direct_write;
|
|
|
|
|
struct bpf_insn_aux_data *insn_aux_data; /* array of per-insn state */
|
2018-12-14 02:41:48 +08:00
|
|
|
const struct bpf_line_info *prev_linfo;
|
2018-03-25 02:44:22 +08:00
|
|
|
struct bpf_verifier_log log;
|
2018-05-03 04:17:18 +08:00
|
|
|
struct bpf_subprog_info subprog_info[BPF_MAX_SUBPROGS + 1];
|
2019-04-19 22:44:54 +08:00
|
|
|
struct {
|
|
|
|
|
int *insn_state;
|
|
|
|
|
int *insn_stack;
|
|
|
|
|
int cur_stack;
|
|
|
|
|
} cfg;
|
2020-01-10 14:41:20 +08:00
|
|
|
u32 pass_cnt; /* number of times do_check() was called */
|
2017-12-15 09:55:05 +08:00
|
|
|
u32 subprog_cnt;
|
2019-04-02 12:27:40 +08:00
|
|
|
/* number of instructions analyzed by the verifier */
|
2019-06-16 03:12:20 +08:00
|
|
|
u32 prev_insn_processed, insn_processed;
|
|
|
|
|
/* number of jmps, calls, exits analyzed so far */
|
|
|
|
|
u32 prev_jmps_processed, jmps_processed;
|
2019-04-02 12:27:40 +08:00
|
|
|
/* total verification time */
|
|
|
|
|
u64 verification_time;
|
|
|
|
|
/* maximum number of verifier states kept in 'branching' instructions */
|
|
|
|
|
u32 max_states_per_insn;
|
|
|
|
|
/* total number of allocated verifier states */
|
|
|
|
|
u32 total_states;
|
|
|
|
|
/* some states are freed during program analysis.
|
|
|
|
|
* this is peak number of states. this number dominates kernel
|
|
|
|
|
* memory consumption during verification
|
|
|
|
|
*/
|
|
|
|
|
u32 peak_states;
|
|
|
|
|
/* longest register parentage chain walked for liveness marking */
|
|
|
|
|
u32 longest_mark_read_walk;
|
2016-09-21 18:43:57 +08:00
|
|
|
};
|
|
|
|
|
|
2018-05-17 04:27:41 +08:00
|
|
|
__printf(2, 0) void bpf_verifier_vlog(struct bpf_verifier_log *log,
|
|
|
|
|
const char *fmt, va_list args);
|
2018-01-10 20:26:06 +08:00
|
|
|
__printf(2, 3) void bpf_verifier_log_write(struct bpf_verifier_env *env,
|
|
|
|
|
const char *fmt, ...);
|
2019-10-16 11:25:00 +08:00
|
|
|
__printf(2, 3) void bpf_log(struct bpf_verifier_log *log,
|
|
|
|
|
const char *fmt, ...);
|
2018-01-10 20:26:06 +08:00
|
|
|
|
2018-10-03 04:35:35 +08:00
|
|
|
static inline struct bpf_func_state *cur_func(struct bpf_verifier_env *env)
|
2017-11-01 09:16:05 +08:00
|
|
|
{
|
2017-12-15 09:55:06 +08:00
|
|
|
struct bpf_verifier_state *cur = env->cur_state;
|
|
|
|
|
|
2018-10-03 04:35:35 +08:00
|
|
|
return cur->frame[cur->curframe];
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
static inline struct bpf_reg_state *cur_regs(struct bpf_verifier_env *env)
|
|
|
|
|
{
|
|
|
|
|
return cur_func(env)->regs;
|
2017-11-01 09:16:05 +08:00
|
|
|
}
|
|
|
|
|
|
2018-11-09 21:03:31 +08:00
|
|
|
int bpf_prog_offload_verifier_prep(struct bpf_prog *prog);
|
2017-12-28 10:39:05 +08:00
|
|
|
int bpf_prog_offload_verify_insn(struct bpf_verifier_env *env,
|
|
|
|
|
int insn_idx, int prev_insn_idx);
|
2018-10-07 19:56:47 +08:00
|
|
|
int bpf_prog_offload_finalize(struct bpf_verifier_env *env);
|
2019-01-23 14:45:24 +08:00
|
|
|
void
|
|
|
|
|
bpf_prog_offload_replace_insn(struct bpf_verifier_env *env, u32 off,
|
|
|
|
|
struct bpf_insn *insn);
|
|
|
|
|
void
|
|
|
|
|
bpf_prog_offload_remove_insns(struct bpf_verifier_env *env, u32 off, u32 cnt);
|
2017-11-04 04:56:17 +08:00
|
|
|
|
2020-01-10 14:41:20 +08:00
|
|
|
int check_ctx_reg(struct bpf_verifier_env *env,
|
|
|
|
|
const struct bpf_reg_state *reg, int regno);
|
|
|
|
|
|
2016-09-21 18:43:57 +08:00
|
|
|
#endif /* _LINUX_BPF_VERIFIER_H */
|