This is done with 2 major steps:
1. Remove JailManagement singleton and use a structure that resembles
what we have with the Process object. This is required later for the
second step in this commit, but on its own, is a major change that
removes this clunky singleton that had no real usage by itself.
2. Use IntrusiveLists to keep references to Process objects in the same
Jail so it will be much more straightforward to iterate on this kind
of objects when needed. Previously we locked the entire Process list
and we did a simple pointer comparison to check if the checked
Process we iterate on is in the same Jail or not, which required
taking multiple Spinlocks in a very clumsy and heavyweight way.
This patch switches away from {Nonnull,}LockRefPtr to the non-locking
smart pointers throughout the kernel.
I've looked at the handful of places where these were being persisted
and I don't see any race situations.
Note that the process file descriptor table (Process::m_fds) was already
guarded via MutexProtected.
Reduce inclusion of limits.h as much as possible at the same time.
This does mean that kmalloc.h is now including Kernel/API/POSIX/limits.h
instead of LibC/limits.h, but the scope could be limited a lot more.
Basically every file in the kernel includes kmalloc.h, and needs the
limits.h include for PAGE_SIZE.
We really don't want callers of this function to accidentally change
the jail, or even worse - remove the Process from an attached jail.
To ensure this never happens, we can just declare this method as const
so nobody can mutate it this way.
There are places in the kernel that would like to have access
to `pgid` credentials in certain circumstances.
I haven't found any use cases for `sid` yet, but `sid` and `pgid` are
both changed with `sys$setpgid`, so it seemed sensical to add it.
In Linux, `man 7 credentials` also mentions both the session id and
process group id, so this isn't unprecedented.
Check if the process we are currently running is in a jail, and if that
is the case, fail early with the EPERM error code.
Also, as Brian noted, we should also disallow attaching to a jail in
case of already running within a setid executable, as this leaves the
user with false thinking of being secure (because you can't exec new
setid binaries), but the current program is still marked setid, which
means that at the very least we gained permissions while we didn't
expect it, so let's block it.
This patch validates that the size of the auxiliary vector does not
exceed `Process::max_auxiliary_size`. The auxiliary vector is a range
of memory in userspace stack where the kernel can pass information to
the process that will be created via `Process:do_exec`.
The reason the kernel needs to validate its size is that the about to
be created process needs to have remaining space on the stack.
Previously only `argv` and `envp` were taken into account for the
size validation, with this patch, the size of `auxv` is also
checked. All three elements contain values that a user (or an
attacker) can specify.
This patch adds the constant `Process::max_auxiliary_size` which is
defined to be one eight of the user-space stack size. This is the
approach taken by `Process:max_arguments_size` and
`Process::max_environment_size` which are used to check the sizes
of `argv` and `envp`.
Some programs explicitly ask for a different initial stack size than
what the OS provides. This is implemented in ELF by having a
PT_GNU_STACK header which has its p_memsz set to the amount that the
program requires. This commit implements this policy by reading the
p_memsz of the header and setting the main thread stack size to that.
ELF::Image::validate_program_headers ensures that the size attribute is
a reasonable value.
To accomplish this, we add another VeilState which is called
LockedInherited. The idea is to apply exec unveil data, similar to
execpromises of the pledge syscall, on the current exec'ed program
during the execve sequence. When applying the forced unveil data, the
veil state is set to be locked but the special state of LockedInherited
ensures that if the new program tries to unveil paths, the request will
silently be ignored, so the program will continue running without
receiving an error, but is still can only use the paths that were
unveiled before the exec syscall. This in turn, allows us to use the
unveil syscall with a special utility to sandbox other userland programs
in terms of what is visible to them on the filesystem, and is usable on
both programs that use or don't use the unveil syscall in their code.
Our implementation for Jails resembles much of how FreeBSD jails are
working - it's essentially only a matter of using a RefPtr in the
Process class to a Jail object. Then, when we iterate over all processes
in various cases, we could ensure if either the current process is in
jail and therefore should be restricted what is visible in terms of
PID isolation, and also to be able to expose metadata about Jails in
/sys/kernel/jails node (which does not reveal anything to a process
which is in jail).
A lifetime model for the Jail object is currently plain simple - there's
simpy no way to manually delete a Jail object once it was created. Such
feature should be carefully designed to allow safe destruction of a Jail
without the possibility of releasing a process which is in Jail from the
actual jail. Each process which is attached into a Jail cannot leave it
until the end of a Process (i.e. when finalizing a Process). All jails
are kept being referenced in the JailManagement. When a last attached
process is finalized, the Jail is automatically destroyed.
This forces anyone who wants to look into and/or manipulate an address
space to lock it. And this replaces the previous, more flimsy, manual
spinlock use.
Note that pointers *into* the address space are not safe to use after
you unlock the space. We've got many issues like this, and we'll have
to track those down as wlel.
Instead of getting credentials from Process::current(), we now require
that they be provided as input to the various VFS functions.
This ensures that an atomic set of credentials is used throughout an
entire VFS operation.
This ensures that both mutable and immutable access to the protected
data of a process is serialized.
Note that there may still be multiple TOCTOU issues around this, as we
have a bunch of convenience accessors that make it easy to introduce
them. We'll need to audit those as well.
By protecting all the RefPtr<Custody> objects that may be accessed from
multiple threads at the same time (with spinlocks), we remove the need
for using LockRefPtr<Custody> (which is basically a RefPtr with a
built-in spinlock.)
This patch adds a new object to hold a Process's user credentials:
- UID, EUID, SUID
- GID, EGID, SGID, extra GIDs
Credentials are immutable and child processes initially inherit the
Credentials object from their parent.
Whenever a process changes one or more of its user/group IDs, a new
Credentials object is constructed.
Any code that wants to inspect and act on a set of credentials can now
do so without worrying about data races.
Until now, our kernel has reimplemented a number of AK classes to
provide automatic internal locking:
- RefPtr
- NonnullRefPtr
- WeakPtr
- Weakable
This patch renames the Kernel classes so that they can coexist with
the original AK classes:
- RefPtr => LockRefPtr
- NonnullRefPtr => NonnullLockRefPtr
- WeakPtr => LockWeakPtr
- Weakable => LockWeakable
The goal here is to eventually get rid of the Lock* classes in favor of
using external locking.
Each of these strings would previously rely on StringView's char const*
constructor overload, which would call __builtin_strlen on the string.
Since we now have operator ""sv, we can replace these with much simpler
versions. This opens the door to being able to remove
StringView(char const*).
No functional changes.
This patch move AddressSpace (the per-process memory manager) to using
the new atomic "place" APIs in RegionTree as well, just like we did for
MemoryManager in the previous commit.
This required updating quite a few places where VM allocation and
actually committing a Region object to the AddressSpace were separated
by other code.
All you have to do now is call into AddressSpace once and it'll take
care of everything for you.
These are not technically required, since the Thread constructor
already sets these, but they are set on i686, so let's try and keep
consistent behaviour between the different archs.
While investigating why gdb is failing when it calls `PT_CONTINUE`
against Serenity I noticed that the names of the programs in the
System Monitor didn't make sense. They were seemingly stale.
After inspecting the kernel code, it became apparent that the sequence
occurs as follows:
1. Debugger calls `fork()`
2. The forked child calls `PT_TRACE_ME`
3. The `PT_TRACE_ME` instructs the forked process to block in the
kernel waiting for a signal from the tracer on the next call
to `execve(..)`.
4. Debugger waits for forked child to spawn and stop, and then it
calls `PT_ATTACH` followed by `PT_CONTINUE` on the child.
5. Currently the `PT_CONTINUE` fails because of some other yet to
be found bug.
6. The process name is set immediately AFTER we are woken up by
the `PT_CONTINUE` which never happens in the case I'm debugging.
This chain of events leaves the process suspended, with the name of
the original (forked) process instead of the name we inherit from
the `execve(..)` call.
To avoid such confusion in the future, we set the new name before we
block waiting for the tracer.
This matches the likes of the adopt_{own, ref}_if_nonnull family and
also frees up the name to allow us to eventually add OOM-fallible
versions of these functions.
Move the definitions for maximum argument and environment size to
Process.h from execve.cpp. This allows sysconf(_SC_ARG_MAX) to return
the actual argument maximum of 128 KiB to userspace.
Function-local `static constexpr` variables can be `constexpr`. This
can reduce memory consumption, binary size, and offer additional
compiler optimizations.
These changes result in a stripped x86_64 kernel binary size reduction
of 592 bytes.
This commit removes the usage of HashMap in Mutex, thereby making Mutex
be allocation-free.
In order to achieve this several simplifications were made to Mutex,
removing unused code-paths and extra VERIFYs:
* We no longer support 'upgrading' a shared lock holder to an
exclusive holder when it is the only shared holder and it did not
unlock the lock before relocking it as exclusive. NOTE: Unlike the
rest of these changes, this scenario is not VERIFY-able in an
allocation-free way, as a result the new LOCK_SHARED_UPGRADE_DEBUG
debug flag was added, this flag lets Mutex allocate in order to
detect such cases when debugging a deadlock.
* We no longer support checking if a Mutex is locked by the current
thread when the Mutex was not locked exclusively, the shared version
of this check was not used anywhere.
* We no longer support force unlocking/relocking a Mutex if the Mutex
was not locked exclusively, the shared version of these functions
was not used anywhere.
We were marking the execing thread as Runnable near the end of
Process::do_exec().
This was necessary for exec in processes that had never been scheduled
yet, which is a specific edge case that only applies to the very first
userspace process (normally SystemServer). At this point, such threads
are in the Invalid state.
In the common case (normal userspace-initiated exec), making the current
thread Runnable meant that we switched away from its current state:
Running. As the thread is indeed running, that's a bogus change!
This created a short time window in which the thread state was bogus,
and any attempt to block the thread would panic the kernel (due to a
bogus thread state in Thread::block() leading to VERIFY_NOT_REACHED().)
Fix this by not touching the thread state in Process::do_exec()
and instead make the first userspace thread Runnable directly after
calling Process::exec() on it in try_create_userspace_process().
It's unfortunate that exec() can be called both on the current thread,
and on a new thread that has never been scheduled. It would be good to
not have the latter edge case, but fixing that will require larger
architectural changes outside the scope of this fix.
There are many assumptions in the stack that argc is not zero, and
argv[0] points to a valid string. The recent pwnkit exploit on Linux
was able to exploit this assumption in the `pkexec` utility
(a SUID-root binary) to escalate from any user to root.
By convention `execve(..)` should always be called with at least one
valid argument, so lets enforce that semantic to harden the system
against vulnerabilities like pwnkit.
Reference: https://www.qualys.com/2022/01/25/cve-2021-4034/pwnkit.txt
These checks in `sys$execve` could trip up the system whenever you try
to execute an `.so` file. For example, double-clicking `libwasm.so` in
Terminal crashes the kernel.
This changes the program header alignment checks to reflect the same
checks in LibELF, and passes the requested alignment on to
`::try_allocate_range()`.
