Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Proposal on better exception handling #1376

Merged
merged 7 commits into from
Jan 8, 2021
Merged
Changes from 6 commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
241 changes: 241 additions & 0 deletions proposals/20200828-structured-exception-handling.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,241 @@
# Proposal for First Class Structured Exceptions in Falco Rules

## Summary

## Motivation

Almost all Falco Rules have cases where the behavior detected by the
rule should be allowed. For example, The rule Write Below Binary Dir
has exceptions for specific programs that are known to write below
these directories as a part of software installation/management:

```yaml
- rule: Write below binary dir
desc: an attempt to write to any file below a set of binary directories
condition: >
bin_dir and evt.dir = < and open_write
and not package_mgmt_procs
and not exe_running_docker_save
and not python_running_get_pip
and not python_running_ms_oms
and not user_known_write_below_binary_dir_activities
...
```
In most cases, these exceptions are expressed as concatenations to the original rule's condition. For example, looking at the macro package_mgmt_procs:

```yaml
- macro: package_mgmt_procs
condition: proc.name in (package_mgmt_binaries)
```

The result is appending `and not proc.name in (package_mgmt_binaries)` to the condition of the rule.

A more extreme case of this is the write_below_etc macro used by Write below etc rule. It has tens of exceptions:

```
...
and not sed_temporary_file
and not exe_running_docker_save
and not ansible_running_python
and not python_running_denyhosts
and not fluentd_writing_conf_files
and not user_known_write_etc_conditions
and not run_by_centrify
and not run_by_adclient
and not qualys_writing_conf_files
and not git_writing_nssdb
...
```

The exceptions all generally follow the same structure--naming a program and a directory prefix below /etc where that program is allowed to write files.

### Using Appends/Overwrites to Customize Rules

An important way to customize rules and macros is to use `append: true` to add to them, or `append: false` to define a new rule/macro, overwriting the original rule/macro. Here's an example from Update Package Repository:

```yaml
- list: package_mgmt_binaries
items: [rpm_binaries, deb_binaries, update-alternat, gem, pip, pip3, sane-utils.post, alternatives, chef-client, apk, snapd]

- macro: package_mgmt_procs
condition: proc.name in (package_mgmt_binaries)

- macro: user_known_update_package_registry
condition: (never_true)

- rule: Update Package Repository
desc: Detect package repositories get updated
condition: >
((open_write and access_repositories) or (modify and modify_repositories))
` and not package_mgmt_procs
leogr marked this conversation as resolved.
Show resolved Hide resolved
and not exe_running_docker_save
and not user_known_update_package_registry
```

If someone wanted to add additional exceptions to this rule, they could add the following to the user_rules file:

```yaml
- list: package_mgmt_binaries
items: [puppet]
append: true

- macro: package_mgmt_procs
condition: and not proc.pname=chef
append: true

- macro: user_known_update_package_registry
condition: (proc.name in (npm))
append: false
mstemm marked this conversation as resolved.
Show resolved Hide resolved
```

This adds an 3 different exceptions:
* an additional binary to package_mgmt_binaries (because append is true),
* adds to package_mgmt_procs, adding an exception for programs spawned by chef (because append is true)
* overrides the macro user_known_update_package_registry to add an exception for npm (because append is false).
mstemm marked this conversation as resolved.
Show resolved Hide resolved

### Problems with Appends/Overrides to Define Exceptions

Although the concepts of macros and lists in condition fields, combined with appending to lists/conditions in macros/rules, is very general purpose, it can be unwieldy:

* Appending to conditions can result in incorrect behavior, unless the original condition has its logical operators set up properly with parentheses. For example:

```yaml
rule: my_rule
condition: (evt.type=open and (fd.name=/tmp/foo or fd.name=/tmp/bar))

rule: my_rule
condition: or fd.name=/tmp/baz
append: true
```

Results in unintended behavior. It will match any fd related event where the name is /tmp/baz, when the intent was probably to add /tmp/baz as an additional opened file.

* A good convention many rules use is to have a clause "and not user_known_xxxx" built into the condition field. However, it's not in all rules and its use is a bit haphazard.

* Appends and overrides can get confusing if you try to apply them multiple times. For example:

```yaml
macro: allowed_files
condition: fd.name=/tmp/foo

...

macro: allowed_files
condition: and fd.name=/tmp/bar
append: true
```

If someone wanted to override the original behavior of allowed_files, they would have to use `append: false` in a third definition of allowed_files, but this would result in losing the append: true override.

## Solution: Exceptions as first class objects

To address some of these problems, we will add the notion of Exceptions as top level objects alongside Rules, Macros, and Lists. A rule that supports exceptions must define a new key `exceptions` in the rule. The exceptions key is a list of identifier plus list of tuples of filtercheck fields. Here's an example:

```yaml
- rule: Write below binary dir
desc: an attempt to write to any file below a set of binary directories
condition: >
bin_dir and evt.dir = < and open_write
and not package_mgmt_procs
and not exe_running_docker_save
and not python_running_get_pip
and not python_running_ms_oms
and not user_known_write_below_binary_dir_activities
exceptions:
- name: proc_writer
fields: [proc.name, fd.directory]
- name: container_writer
fields: [container.image.repository, fd.directory]
comps: [=, startswith]
- name: proc_filenames
fields: [proc.name, fd.name]
comps: [=, in]
leogr marked this conversation as resolved.
Show resolved Hide resolved
- name: filenames
fields: fd.filename
comps: in
```

This rule defines four kinds of exceptions:
* proc_writer: uses a combination of proc.name and fd.directory
* container_writer: uses a combination of container.image.repository and fd.directory
* proc_filenames: uses a combination of process and list of filenames.
* filenames: uses a list of filenames

The specific strings "proc_writer"/"container_writer"/"proc_filenames"/"filenames" are arbitrary strings and don't have a special meaning to the rules file parser. They're only used to link together the list of field names with the list of field values that exist in the exception object.

proc_writer does not have any comps property, so the fields are directly compared to values using the = operator. container_writer does have a comps property, so each field will be compared to the corresponding exception items using the corresponding comparison operator.

proc_filenames uses the in comparison operator, so the corresponding values entry should be a list of filenames.

filenames differs from the others in that it names a single field and single comp operator. This changes how the exception condition snippet is constructed (see below).

Notice that exceptions are defined as a part of the rule. This is important because the author of the rule defines what construes a valid exception to the rule. In this case, an exception can consist of a process and file directory (actor and target), but not a process name only (too broad).

Exception values will most commonly be defined in rules with append: true. Here's an example:

```yaml
- list: apt_files
items: [/bin/ls, /bin/rm]

- rule: Write below binary dir
exceptions:
- name: proc_writer
values:
- [apk, /usr/lib/alpine]
- [npm, /usr/node/bin]
- name: container_writer
values:
- [docker.io/alpine, /usr/libexec/alpine]
- name: proc_filenames
values:
- [apt, apt_files]
- [rpm, [/bin/cp, /bin/pwd]]
leogr marked this conversation as resolved.
Show resolved Hide resolved
- name: filenames
values: [python, go]
```

A rule exception applies if for a given event, the fields in a rule.exception match all of the values in some exception.item. For example, if a program `apk` writes to a file below `/usr/lib/alpine`, the rule will not trigger, even if the condition is met.

Notice that an item in a values list can be a list. This allows building exceptions with operators like "in", "pmatch", etc. that work on a list of items. The item can also be a name of an existing list. If not present surrounding parantheses will be added.

Finally, note that the structure of the values property differs between the items where fields is a list of fields (proc_writer/container_writer/proc_filenames) and when it is a single field (procs_only). This changes how the condition snippet is constructed.

### Implementation

For exception items where the fields property is a list of field names, each exception can be thought of as an implicit "and not (field1 cmp1 val1 and field2 cmp2 val2 and...)" appended to the rule's condition. For exception items where the fields property is a single field name, the exception can be thought of as an implict "and not field cmp (val1, val2, ...)". In practice, that's how exceptions will be implemented.

When a rule is parsed, the original condition will be wrapped in an extra layer of parentheses and all exception values will be appended to the condition. For example, using the example above, the resulting condition will be:

```
(<Write below binary dir condition>) and not (
(proc.name = apk and fd.directory = /usr/lib/alpine) or (proc.name = npm and fd.directory = /usr/node/bin) or
(container.image.repository = docker.io/alpine and fd.directory startswith /usr/libexec/alpine) or
(proc.name=apt and fd.name in (apt_files))) or
(fd.filename in (python, go))))
```

The exceptions are effectively syntatic sugar that allows expressing sets of exceptions in a concise way.

### Advantages
mstemm marked this conversation as resolved.
Show resolved Hide resolved

Adding Exception objects as described here has several advantages:

* All rules will implicitly support exceptions. A rule writer doesn't need to define a user_known_xxx macro and add it to the condition.
* The rule writer has some controls on what defines a valid exception. The rule author knows best what is a good exception, and can define the fields that make up the exception.
* With this approach, it's much easier to add and manage multiple sets of exceptions from multiple sources. You're just combining lists of tuples of filtercheck field values.

## Backwards compatibility
mstemm marked this conversation as resolved.
Show resolved Hide resolved

To take advantage of these new features, users will need to upgrade Falco to a version that supports exception objects and exception keys in rules. For the most part, however, the rules file structure is unchanged.

This approach does not remove the ability to append to exceptions nor the existing use of user_xxx macros to define exceptions to rules. It only provides an additional way to express exceptions. Hopefully, we can migrate existing exceptions to use this approach, but there isn't any plan to make wholesale rules changes as a part of this.

This approach is for the most part backwards compatible with older Falco releases. To implement exceptions, we'll add a preprocessing element to rule parsing. The main Falco engine is unchanged.

However, there are a few changes we'll have to make to Falco rules file parsing:

* Currently, Falco will reject files containing anything other than rule/macro/list top-level objects. As a result, `exception` objects would be rejected. We'll probably want to make a one-time change to Falco to allow arbitrary top level objects.
* Similarly, Falco will reject rule objects with exception keys. We'll also probably want to change Falco to allow unknown keys inside rule/macro/list/exception objects.