Re: [PATCH bpf-next v8 05/11] seccomp,landlock: Enforce Landlock programs per process hierarchy

From: Alexei Starovoitov
Date: Tue Apr 10 2018 - 00:48:38 EST


On Mon, Apr 09, 2018 at 12:01:59AM +0200, Mickaël Salaün wrote:
>
> On 04/08/2018 11:06 PM, Andy Lutomirski wrote:
> > On Sun, Apr 8, 2018 at 6:13 AM, Mickaël Salaün <mic@xxxxxxxxxxx> wrote:
> >>
> >> On 02/27/2018 10:48 PM, Mickaël Salaün wrote:
> >>>
> >>> On 27/02/2018 17:39, Andy Lutomirski wrote:
> >>>> On Tue, Feb 27, 2018 at 5:32 AM, Alexei Starovoitov
> >>>> <alexei.starovoitov@xxxxxxxxx> wrote:
> >>>>> On Tue, Feb 27, 2018 at 05:20:55AM +0000, Andy Lutomirski wrote:
> >>>>>> On Tue, Feb 27, 2018 at 4:54 AM, Alexei Starovoitov
> >>>>>> <alexei.starovoitov@xxxxxxxxx> wrote:
> >>>>>>> On Tue, Feb 27, 2018 at 04:40:34AM +0000, Andy Lutomirski wrote:
> >>>>>>>> On Tue, Feb 27, 2018 at 2:08 AM, Alexei Starovoitov
> >>>>>>>> <alexei.starovoitov@xxxxxxxxx> wrote:
> >>>>>>>>> On Tue, Feb 27, 2018 at 01:41:15AM +0100, Mickaël Salaün wrote:
> >>>>>>>>>> The seccomp(2) syscall can be used by a task to apply a Landlock program
> >>>>>>>>>> to itself. As a seccomp filter, a Landlock program is enforced for the
> >>>>>>>>>> current task and all its future children. A program is immutable and a
> >>>>>>>>>> task can only add new restricting programs to itself, forming a list of
> >>>>>>>>>> programss.
> >>>>>>>>>>
> >>>>>>>>>> A Landlock program is tied to a Landlock hook. If the action on a kernel
> >>>>>>>>>> object is allowed by the other Linux security mechanisms (e.g. DAC,
> >>>>>>>>>> capabilities, other LSM), then a Landlock hook related to this kind of
> >>>>>>>>>> object is triggered. The list of programs for this hook is then
> >>>>>>>>>> evaluated. Each program return a 32-bit value which can deny the action
> >>>>>>>>>> on a kernel object with a non-zero value. If every programs of the list
> >>>>>>>>>> return zero, then the action on the object is allowed.
> >>>>>>>>>>
> >>>>>>>>>> Multiple Landlock programs can be chained to share a 64-bits value for a
> >>>>>>>>>> call chain (e.g. evaluating multiple elements of a file path). This
> >>>>>>>>>> chaining is restricted when a process construct this chain by loading a
> >>>>>>>>>> program, but additional checks are performed when it requests to apply
> >>>>>>>>>> this chain of programs to itself. The restrictions ensure that it is
> >>>>>>>>>> not possible to call multiple programs in a way that would imply to
> >>>>>>>>>> handle multiple shared values (i.e. cookies) for one chain. For now,
> >>>>>>>>>> only a fs_pick program can be chained to the same type of program,
> >>>>>>>>>> because it may make sense if they have different triggers (cf. next
> >>>>>>>>>> commits). This restrictions still allows to reuse Landlock programs in
> >>>>>>>>>> a safe way (e.g. use the same loaded fs_walk program with multiple
> >>>>>>>>>> chains of fs_pick programs).
> >>>>>>>>>>
> >>>>>>>>>> Signed-off-by: Mickaël Salaün <mic@xxxxxxxxxxx>
> >>>>>>>>>
> >>>>>>>>> ...
> >>>>>>>>>
> >>>>>>>>>> +struct landlock_prog_set *landlock_prepend_prog(
> >>>>>>>>>> + struct landlock_prog_set *current_prog_set,
> >>>>>>>>>> + struct bpf_prog *prog)
> >>>>>>>>>> +{
> >>>>>>>>>> + struct landlock_prog_set *new_prog_set = current_prog_set;
> >>>>>>>>>> + unsigned long pages;
> >>>>>>>>>> + int err;
> >>>>>>>>>> + size_t i;
> >>>>>>>>>> + struct landlock_prog_set tmp_prog_set = {};
> >>>>>>>>>> +
> >>>>>>>>>> + if (prog->type != BPF_PROG_TYPE_LANDLOCK_HOOK)
> >>>>>>>>>> + return ERR_PTR(-EINVAL);
> >>>>>>>>>> +
> >>>>>>>>>> + /* validate memory size allocation */
> >>>>>>>>>> + pages = prog->pages;
> >>>>>>>>>> + if (current_prog_set) {
> >>>>>>>>>> + size_t i;
> >>>>>>>>>> +
> >>>>>>>>>> + for (i = 0; i < ARRAY_SIZE(current_prog_set->programs); i++) {
> >>>>>>>>>> + struct landlock_prog_list *walker_p;
> >>>>>>>>>> +
> >>>>>>>>>> + for (walker_p = current_prog_set->programs[i];
> >>>>>>>>>> + walker_p; walker_p = walker_p->prev)
> >>>>>>>>>> + pages += walker_p->prog->pages;
> >>>>>>>>>> + }
> >>>>>>>>>> + /* count a struct landlock_prog_set if we need to allocate one */
> >>>>>>>>>> + if (refcount_read(&current_prog_set->usage) != 1)
> >>>>>>>>>> + pages += round_up(sizeof(*current_prog_set), PAGE_SIZE)
> >>>>>>>>>> + / PAGE_SIZE;
> >>>>>>>>>> + }
> >>>>>>>>>> + if (pages > LANDLOCK_PROGRAMS_MAX_PAGES)
> >>>>>>>>>> + return ERR_PTR(-E2BIG);
> >>>>>>>>>> +
> >>>>>>>>>> + /* ensure early that we can allocate enough memory for the new
> >>>>>>>>>> + * prog_lists */
> >>>>>>>>>> + err = store_landlock_prog(&tmp_prog_set, current_prog_set, prog);
> >>>>>>>>>> + if (err)
> >>>>>>>>>> + return ERR_PTR(err);
> >>>>>>>>>> +
> >>>>>>>>>> + /*
> >>>>>>>>>> + * Each task_struct points to an array of prog list pointers. These
> >>>>>>>>>> + * tables are duplicated when additions are made (which means each
> >>>>>>>>>> + * table needs to be refcounted for the processes using it). When a new
> >>>>>>>>>> + * table is created, all the refcounters on the prog_list are bumped (to
> >>>>>>>>>> + * track each table that references the prog). When a new prog is
> >>>>>>>>>> + * added, it's just prepended to the list for the new table to point
> >>>>>>>>>> + * at.
> >>>>>>>>>> + *
> >>>>>>>>>> + * Manage all the possible errors before this step to not uselessly
> >>>>>>>>>> + * duplicate current_prog_set and avoid a rollback.
> >>>>>>>>>> + */
> >>>>>>>>>> + if (!new_prog_set) {
> >>>>>>>>>> + /*
> >>>>>>>>>> + * If there is no Landlock program set used by the current task,
> >>>>>>>>>> + * then create a new one.
> >>>>>>>>>> + */
> >>>>>>>>>> + new_prog_set = new_landlock_prog_set();
> >>>>>>>>>> + if (IS_ERR(new_prog_set))
> >>>>>>>>>> + goto put_tmp_lists;
> >>>>>>>>>> + } else if (refcount_read(&current_prog_set->usage) > 1) {
> >>>>>>>>>> + /*
> >>>>>>>>>> + * If the current task is not the sole user of its Landlock
> >>>>>>>>>> + * program set, then duplicate them.
> >>>>>>>>>> + */
> >>>>>>>>>> + new_prog_set = new_landlock_prog_set();
> >>>>>>>>>> + if (IS_ERR(new_prog_set))
> >>>>>>>>>> + goto put_tmp_lists;
> >>>>>>>>>> + for (i = 0; i < ARRAY_SIZE(new_prog_set->programs); i++) {
> >>>>>>>>>> + new_prog_set->programs[i] =
> >>>>>>>>>> + READ_ONCE(current_prog_set->programs[i]);
> >>>>>>>>>> + if (new_prog_set->programs[i])
> >>>>>>>>>> + refcount_inc(&new_prog_set->programs[i]->usage);
> >>>>>>>>>> + }
> >>>>>>>>>> +
> >>>>>>>>>> + /*
> >>>>>>>>>> + * Landlock program set from the current task will not be freed
> >>>>>>>>>> + * here because the usage is strictly greater than 1. It is
> >>>>>>>>>> + * only prevented to be freed by another task thanks to the
> >>>>>>>>>> + * caller of landlock_prepend_prog() which should be locked if
> >>>>>>>>>> + * needed.
> >>>>>>>>>> + */
> >>>>>>>>>> + landlock_put_prog_set(current_prog_set);
> >>>>>>>>>> + }
> >>>>>>>>>> +
> >>>>>>>>>> + /* prepend tmp_prog_set to new_prog_set */
> >>>>>>>>>> + for (i = 0; i < ARRAY_SIZE(tmp_prog_set.programs); i++) {
> >>>>>>>>>> + /* get the last new list */
> >>>>>>>>>> + struct landlock_prog_list *last_list =
> >>>>>>>>>> + tmp_prog_set.programs[i];
> >>>>>>>>>> +
> >>>>>>>>>> + if (last_list) {
> >>>>>>>>>> + while (last_list->prev)
> >>>>>>>>>> + last_list = last_list->prev;
> >>>>>>>>>> + /* no need to increment usage (pointer replacement) */
> >>>>>>>>>> + last_list->prev = new_prog_set->programs[i];
> >>>>>>>>>> + new_prog_set->programs[i] = tmp_prog_set.programs[i];
> >>>>>>>>>> + }
> >>>>>>>>>> + }
> >>>>>>>>>> + new_prog_set->chain_last = tmp_prog_set.chain_last;
> >>>>>>>>>> + return new_prog_set;
> >>>>>>>>>> +
> >>>>>>>>>> +put_tmp_lists:
> >>>>>>>>>> + for (i = 0; i < ARRAY_SIZE(tmp_prog_set.programs); i++)
> >>>>>>>>>> + put_landlock_prog_list(tmp_prog_set.programs[i]);
> >>>>>>>>>> + return new_prog_set;
> >>>>>>>>>> +}
> >>>>>>>>>
> >>>>>>>>> Nack on the chaining concept.
> >>>>>>>>> Please do not reinvent the wheel.
> >>>>>>>>> There is an existing mechanism for attaching/detaching/quering multiple
> >>>>>>>>> programs attached to cgroup and tracing hooks that are also
> >>>>>>>>> efficiently executed via BPF_PROG_RUN_ARRAY.
> >>>>>>>>> Please use that instead.
> >>>>>>>>>
> >>>>>>>>
> >>>>>>>> I don't see how that would help. Suppose you add a filter, then
> >>>>>>>> fork(), and then the child adds another filter. Do you want to
> >>>>>>>> duplicate the entire array? You certainly can't *modify* the array
> >>>>>>>> because you'll affect processes that shouldn't be affected.
> >>>>>>>>
> >>>>>>>> In contrast, doing this through seccomp like the earlier patches
> >>>>>>>> seemed just fine to me, and seccomp already had the right logic.
> >>>>>>>
> >>>>>>> it doesn't look to me that existing seccomp side of managing fork
> >>>>>>> situation can be reused. Here there is an attempt to add 'chaining'
> >>>>>>> concept which sort of an extension of existing seccomp style,
> >>>>>>> but somehow heavily done on bpf side and contradicts cgroup/tracing.
> >>>>>>>
> >>>>>>
> >>>>>> I don't see why the seccomp way can't be used. I agree with you that
> >>>>>> the seccomp *style* shouldn't be used in bpf code like this, but I
> >>>>>> think that Landlock programs can and should just live in the existing
> >>>>>> seccomp chain. If the existing seccomp code needs some modification
> >>>>>> to make this work, then so be it.
> >>>>>
> >>>>> +1
> >>>>> if that was the case...
> >>>>> but that's not my reading of the patch set.
> >>>>
> >>>> An earlier version of the patch set used the seccomp filter chain.
> >>>> Mickaël, what exactly was wrong with that approach other than that the
> >>>> seccomp() syscall was awkward for you to use? You could add a
> >>>> seccomp_add_landlock_rule() syscall if you needed to.
> >>>
> >>> Nothing was wrong about about that, this part did not changed (see my
> >>> next comment).
> >>>
> >>>>
> >>>> As a side comment, why is this an LSM at all, let alone a non-stacking
> >>>> LSM? It would make a lot more sense to me to make Landlock depend on
> >>>> having LSMs configured in but to call the landlock hooks directly from
> >>>> the security_xyz() hooks.
> >>>
> >>> See Casey's answer and his patch series: https://lwn.net/Articles/741963/
> >>>
> >>>>
> >>>>>
> >>>>>> In other words, the kernel already has two kinds of chaining:
> >>>>>> seccomp's and bpf's. bpf's doesn't work right for this type of usage
> >>>>>> across fork(), whereas seccomp's already handles that case correctly.
> >>>>>> (In contrast, seccomp's is totally wrong for cgroup-attached filters.)
> >>>>>> So IMO Landlock should use the seccomp core code and call into bpf
> >>>>>> for the actual filtering.
> >>>>>
> >>>>> +1
> >>>>> in cgroup we had to invent this new BPF_PROG_RUN_ARRAY mechanism,
> >>>>> since cgroup hierarchy can be complicated with bpf progs attached
> >>>>> at different levels with different override/multiprog properties,
> >>>>> so walking link list and checking all flags at run-time would have
> >>>>> been too slow. That's why we added compute_effective_progs().
> >>>>
> >>>> If we start adding override flags to Landlock, I think we're doing it
> >>>> wrong. With cgroup bpf programs, the whole mess is set up by the
> >>>> administrator. With seccomp, and with Landlock if done correctly, it
> >>>> *won't* be set up by the administrator, so the chance that everyone
> >>>> gets all the flags right is about zero. All attached filters should
> >>>> run unconditionally.
> >>>
> >>>
> >>> There is a misunderstanding about this chaining mechanism. This should
> >>> not be confused with the list of seccomp filters nor the cgroup
> >>> hierarchies. Landlock programs can be stacked the same way seccomp's
> >>> filters can (cf. struct landlock_prog_set, the "chain_last" field is an
> >>> optimization which is not used for this struct handling). This stackable
> >>> property did not changed from the previous patch series. The chaining
> >>> mechanism is for another use case, which does not make sense for seccomp
> >>> filters nor other eBPF program types, at least for now, from what I can
> >>> tell.
> >>>
> >>> You may want to get a look at my talk at FOSDEM
> >>> (https://landlock.io/talks/2018-02-04_landlock-fosdem.pdf), especially
> >>> slides 11 and 12.
> >>>
> >>> Let me explain my reasoning about this program chaining thing.
> >>>
> >>> To check if an action on a file is allowed, we first need to identify
> >>> this file and match it to the security policy. In a previous
> >>> (non-public) patch series, I tried to use one type of eBPF program to
> >>> check every kind of access to a file. To be able to identify a file, I
> >>> relied on an eBPF map, similar to the current inode map. This map store
> >>> a set of references to file descriptors. I then created a function
> >>> bpf_is_file_beneath() to check if the requested file was beneath a file
> >>> in the map. This way, no chaining, only one eBPF program type to check
> >>> an access to a file... but some issues then emerged. First, this design
> >>> create a side-channel which help an attacker using such a program to
> >>> infer some information not normally available, for example to get a hint
> >>> on where a file descriptor (received from a UNIX socket) come from.
> >>> Another issue is that this type of program would be called for each
> >>> component of a path. Indeed, when the kernel check if an access to a
> >>> file is allowed, it walk through all of the directories in its path
> >>> (checking if the current process is allowed to execute them). That first
> >>> attempt led me to rethink the way we could filter an access to a file
> >>> *path*.
> >>>
> >>> To minimize the number of called to an eBPF program dedicated to
> >>> validate an access to a file path, I decided to create three subtype of
> >>> eBPF programs. The FS_WALK type is called when walking through every
> >>> directory of a file path (except the last one if it is the target). We
> >>> can then restrict this type of program to the minimum set of functions
> >>> it is allowed to call and the minimum set of data available from its
> >>> context. The first implicit chaining is for this type of program. To be
> >>> able to evaluate a path while being called for all its components, this
> >>> program need to store a state (to remember what was the parent directory
> >>> of this path). There is no "previous" field in the subtype for this
> >>> program because it is chained with itself, for each directories. This
> >>> enable to create a FS_WALK program to evaluate a file hierarchy, thank
> >>> to the inode map which can be used to check if a directory of this
> >>> hierarchy is part of an allowed (or denied) list of directories. This
> >>> design enables to express a file hierarchy in a programmatic way,
> >>> without requiring an eBPF helper to do the job (unlike my first experiment).
> >>>
> >>> The explicit chaining is used to tied a path evaluation (with a FS_WALK
> >>> program) to an access to the actual file being requested (the last
> >>> component of a file path), with a FS_PICK program. It is only at this
> >>> time that the kernel check for the requested action (e.g. read, write,
> >>> chdir, append...). To be able to filter such access request we can have
> >>> one call to the same program for every action and let this program check
> >>> for which action it was called. However, this design does not allow the
> >>> kernel to know if the current action is indeed handled by this program.
> >>> Hence, it is not possible to implement a cache mechanism to only call
> >>> this program if it knows how to handle this action.
> >>>
> >>> The approach I took for this FS_PICK type of program is to add to its
> >>> subtype which action it can handle (with the "triggers" bitfield, seen
> >>> as ORed actions). This way, the kernel knows if a call to a FS_PICK
> >>> program is necessary. If the user wants to enforce a different security
> >>> policy according to the action requested on a file, then it needs
> >>> multiple FS_PICK programs. However, to reduce the number of such
> >>> programs, this patch series allow a FS_PICK program to be chained with
> >>> another, the same way a FS_WALK is chained with itself. This way, if the
> >>> user want to check if the action is a for example an "open" and a "read"
> >>> and not a "map" and a "read", then it can chain multiple FS_PICK
> >>> programs with different triggers actions. The OR check performed by the
> >>> kernel is not a limitation then, only a way to know if a call to an eBPF
> >>> program is needed.
> >>>
> >>> The last type of program is FS_GET. This one is called when a process
> >>> get a struct file or change its working directory. This is the only
> >>> program type able (and allowed) to tag a file. This restriction is
> >>> important to not being subject to resource exhaustion attacks (i.e.
> >>> tagging every inode accessible to an attacker, which would allocate too
> >>> much kernel memory).
> >>>
> >>> This design gives room for improvements to create a cache of eBPF
> >>> context (input data, including maps if any), with the result of an eBPF
> >>> program. This would help limit the number of call to an eBPF program the
> >>> same way SELinux or other kernel components do to limit costly checks.
> >>>
> >>> The eBPF maps of progs are useful to call the same type of eBPF
> >>> program. It does not fit with this use case because we may want multiple
> >>> eBPF program according to the action requested on a kernel object (e.g.
> >>> FS_GET). The other reason is because the eBPF program does not know what
> >>> will be the next (type of) access check performed by the kernel.
> >>>
> >>> To say it another way, this chaining mechanism is a way to split a
> >>> kernel object evaluation with multiple specialized programs, each of
> >>> them being able to deal with data tied to their type. Using a monolithic
> >>> eBPF program to check everything does not scale and does not fit with
> >>> unprivileged use either.
> >>>
> >>> As a side note, the cookie value is only an ephemeral value to keep a
> >>> state between multiple programs call. It can be used to create a state
> >>> machine for an object evaluation.
> >>>
> >>> I don't see a way to do an efficient and programmatic path evaluation,
> >>> with different access checks, with the current eBPF features. Please let
> >>> me know if you know how to do it another way.
> >>>
> >>
> >> Andy, Alexei, Daniel, what do you think about this Landlock program
> >> chaining and cookie?
> >>
> >
> > Can you give a small pseudocode real world example that acutally needs
> > chaining? The mechanism is quite complicated and I'd like to
> > understand how it'll be used.
> >
>
> Here is the interesting part from the example (patch 09/11):
>
> +SEC("maps")
> +struct bpf_map_def inode_map = {
> + .type = BPF_MAP_TYPE_INODE,
> + .key_size = sizeof(u32),
> + .value_size = sizeof(u64),
> + .max_entries = 20,
> +};
> +
> +SEC("subtype/landlock1")
> +static union bpf_prog_subtype _subtype1 = {
> + .landlock_hook = {
> + .type = LANDLOCK_HOOK_FS_WALK,
> + }
> +};
> +
> +static __always_inline __u64 update_cookie(__u64 cookie, __u8 lookup,
> + void *inode, void *chain, bool freeze)
> +{
> + __u64 map_allow = 0;
> +
> + if (cookie == 0) {
> + cookie = bpf_inode_get_tag(inode, chain);
> + if (cookie)
> + return cookie;
> + /* only look for the first match in the map, ignore nested
> + * paths in this example */
> + map_allow = bpf_inode_map_lookup(&inode_map, inode);
> + if (map_allow)
> + cookie = 1 | map_allow;
> + } else {
> + if (cookie & COOKIE_VALUE_FREEZED)
> + return cookie;
> + map_allow = cookie & _MAP_MARK_MASK;
> + cookie &= ~_MAP_MARK_MASK;
> + switch (lookup) {
> + case LANDLOCK_CTX_FS_WALK_INODE_LOOKUP_DOTDOT:
> + cookie--;
> + break;
> + case LANDLOCK_CTX_FS_WALK_INODE_LOOKUP_DOT:
> + break;
> + default:
> + /* ignore _MAP_MARK_MASK overflow in this example */
> + cookie++;
> + break;
> + }
> + if (cookie >= 1)
> + cookie |= map_allow;
> + }
> + /* do not modify the cookie for each fs_pick */
> + if (freeze && cookie)
> + cookie |= COOKIE_VALUE_FREEZED;
> + return cookie;
> +}
> +
> +SEC("landlock1")
> +int fs_walk(struct landlock_ctx_fs_walk *ctx)
> +{
> + ctx->cookie = update_cookie(ctx->cookie, ctx->inode_lookup,
> + (void *)ctx->inode, (void *)ctx->chain, false);
> + return LANDLOCK_RET_ALLOW;
> +}
>
> The program "landlock1" is called for every directory execution (except
> the last one if it is the leaf of a path). This enables to identify a
> file hierarchy with only a (one dimension) list of file descriptors
> (i.e. inode_map).
>
> Underneath, the Landlock LSM part looks if there is an associated path
> walk (nameidata) with each inode access request. If there is one, then
> the cookie associated with the path walk (if any) is made available
> through the eBPF program context. This enables to develop a state
> machine with an eBPF program to "evaluate" a file path (without string
> parsing).
>
> The goal with this chaining mechanism is to be able to express a complex
> kernel object like a file, with multiple run of one or more eBPF
> programs, as a multilayer evaluation. This semantic may only make sense
> for the user/developer and his security policy. We must keep in mind
> that this object identification should be available to unprivileged
> processes. This means that we must be very careful to what kind of
> information are available to an eBPF program because this can then leak
> to a process (e.g. through a map). With this mechanism, only information
> already available to user space is available to the eBPF program.
>
> In this example, the complexity of the path evaluation is in the eBPF
> program. We can then keep the kernel code more simple and generic. This
> enables more flexibility for a security policy definition.

it all sounds correct on paper, but it's pretty novel
approach and I'm not sure I see all the details in the patch.
When people say "inode" they most of the time mean inode integer number,
whereas in this patch do you mean a raw pointer to in-kernel
'struct inode' ?
To avoid confusion it should probably be called differently.

If you meant inode as a number then why inode only?
where is superblock, device, mount point?
How bpf side can compare inodes without this additional info?
How bpf side will know what inode to compare to?
What if inode number is reused?
This approach is an optimization to compare inodes
instead of strings passed into sys_open ?

If you meant inode as a pointer how bpf side will
know the pointer before the walk begins?
What guarantees that it's not a stale pointer?