Utilize the new Thread::wait_on timeout parameter to implement
timeout support for FUTEX_WAIT.
As we compute the relative time from the user specified absolute
time, we try to delay that computation as long as possible before
we call into Thread::wait_on(..). To enable this a small bit of
refactoring was done pull futex_queue fetching out and timeout fetch
and calculation separation.
This is a special case that was previously not implemented.
The idea is that you can dispatch a signal to all other processes
the calling process has access to.
There was some minor refactoring to make the self signal logic
into a function so it could easily be easily re-used from do_killall.
Previously, when returning from a pthread's start_routine, we would
segfault. Now we instead implicitly call pthread_exit as specified in
the standard.
pthread_create now creates a thread running the new
pthread_create_helper, which properly manages the calling and exiting
of the start_routine supplied to pthread_create. To accomplish this,
the thread's stack initialization has been moved out of
sys$create_thread and into the userspace function create_thread.
POSIX says, "Conforming applications should not assume that the returned
contents of the symbolic link are null-terminated."
If we do include the null terminator into the returning string, Python
believes it to actually be a part of the returned name, and gets unhappy
about that later. This suggests other systems Python runs in don't include
it, so let's do that too.
Also, make our userspace support non-null-terminated realpath().
PT_SETTREGS sets the regsiters of the traced thread. It can only be
used when the tracee is stopped.
Also, refactor ptrace.
The implementation was getting long and cluttered the alraedy large
Process.cpp file.
This commit moves the bulk of the implementation to Kernel/Ptrace.cpp,
and factors out peek & poke to separate methods of the Process class.
This was a missing feature in the PT_TRACEME command.
This feature allows the tracer to interact with the tracee before the
tracee has started executing its program.
It will be useful for automatically inserting a breakpoint at a
debugged program's entry point.
PT_POKE writes a single word to the tracee's address space.
Some caveats:
- If the user requests to write to an address in a read-only region, we
temporarily change the page's protections to allow it.
- If the user requests to write to a region that's backed by a
SharedInodeVMObject, we replace the vmobject with a PrivateIndoeVMObject.
This patch adds the minherit() syscall originally invented by OpenBSD.
Only the MAP_INHERIT_ZERO mode is supported for now. If set on an mmap
region, that region will be zeroed out on fork().
These validate_elf_* methods really had no business being static
methods of ELF::Image. Now that the ELF namespace exists, it makes
sense to just move them to be free functions in the namespace.
If we don't support ACPI, just don't instantiate an ACPI parser.
This is way less confusing than having a special parser class whose
only purpose is to do nothing.
We now search for the RSDP in ACPI::initialize() instead of letting
the parser constructor do it. This allows us to defer the decision
to create a parser until we're sure we can make a useful one.
This commit adds a basic implementation of
the ptrace syscall, which allows one process
(the tracer) to control another process (the tracee).
While a process is being traced, it is stopped whenever a signal is
received (other than SIGCONT).
The tracer can start tracing another thread with PT_ATTACH,
which causes the tracee to stop.
From there, the tracer can use PT_CONTINUE
to continue the execution of the tracee,
or use other request codes (which haven't been implemented yet)
to modify the state of the tracee.
Additional request codes are PT_SYSCALL, which causes the tracee to
continue exection but stop at the next entry or exit from a syscall,
and PT_GETREGS which fethces the last saved register set of the tracee
(can be used to inspect syscall arguments and return value).
A special request code is PT_TRACE_ME, which is issued by the tracee
and causes it to stop when it calls execve and wait for the
tracer to attach.
This new subsystem includes better abstractions of how time will be
handled in the OS. We take advantage of the existing RTC timer to aid
in keeping time synchronized. This is standing in contrast to how we
handled time-keeping in the kernel, where the PIT was responsible for
that function in addition to update the scheduler about ticks.
With that new advantage, we can easily change the ticking dynamically
and still keep the time synchronized.
In the process context, we no longer use a fixed declaration of
TICKS_PER_SECOND, but we call the TimeManagement singleton class to
provide us the right value. This allows us to use dynamic ticking in
the future, a feature known as tickless kernel.
The scheduler no longer does by himself the calculation of real time
(Unix time), and just calls the TimeManagment singleton class to provide
the value.
Also, we can use 2 new boot arguments:
- the "time" boot argument accpets either the value "modern", or
"legacy". If "modern" is specified, the time management subsystem will
try to setup HPET. Otherwise, for "legacy" value, the time subsystem
will revert to use the PIT & RTC, leaving HPET disabled.
If this boot argument is not specified, the default pattern is to try
to setup HPET.
- the "hpet" boot argumet accepts either the value "periodic" or
"nonperiodic". If "periodic" is specified, the HPET will scan for
periodic timers, and will assert if none are found. If only one is
found, that timer will be assigned for the time-keeping task. If more
than one is found, both time-keeping task & scheduler-ticking task
will be assigned to periodic timers.
If this boot argument is not specified, the default pattern is to try
to scan for HPET periodic timers. This boot argument has no effect if
HPET is disabled.
In hardware context, PIT & RealTimeClock classes are merely inheriting
from the HardwareTimer class, and they allow to use the old i8254 (PIT)
and RTC devices, managing them via IO ports. By default, the RTC will be
programmed to a frequency of 1024Hz. The PIT will be programmed to a
frequency close to 1000Hz.
About HPET, depending if we need to scan for periodic timers or not,
we try to set a frequency close to 1000Hz for the time-keeping timer
and scheduler-ticking timer. Also, if possible, we try to enable the
Legacy replacement feature of the HPET. This feature if exists,
instructs the chipset to disconnect both i8254 (PIT) and RTC.
This behavior is observable on QEMU, and was verified against the source
code:
ce967e2f33
The HPETComparator class is inheriting from HardwareTimer class, and is
responsible for an individual HPET comparator, which is essentially a
timer. Therefore, it needs to call the singleton HPET class to perform
HPET-related operations.
The new abstraction of Hardware timers brings an opportunity of more new
features in the foreseeable future. For example, we can change the
callback function of each hardware timer, thus it makes it possible to
swap missions between hardware timers, or to allow to use a hardware
timer for other temporary missions (e.g. calibrating the LAPIC timer,
measuring the CPU frequency, etc).
This is similar to 28e1da344d
and 4dd4dd2f3c.
The crux is that wait verifies that the outvalue (siginfo* infop)
is writable *before* waiting, and writes to it *after* waiting.
In the meantime, a concurrent thread can make the output region
unwritable, e.g. by deallocating it.
This is similar to 28e1da344d
and 4dd4dd2f3c.
The crux is that select verifies that the filedescriptor sets
are writable *before* blocking, and writes to them *after* blocking.
In the meantime, a concurrent thread can make the output buffer
unwritable, e.g. by deallocating it.
This is a complete fix of clock_nanosleep, because the thread holds the
process lock again when returning from sleep()/sleep_until().
Therefore, no further concurrent invalidation can occur.
Also, duplicate data in dbg() and klog() calls were removed.
In addition, leakage of virtual address to kernel log is prevented.
This is done by replacing kprintf() calls to dbg() calls with the
leaked data instead.
Also, other kprintf() calls were replaced with klog().
This was only used by the mechanism for mapping executables into each
process's own address space. Now that we remap executables on demand
when needed for symbolication, this can go away.
Previously we would map the entire executable of a program in its own
address space (but make it unavailable to userspace code.)
This patch removes that and changes the symbolication code to remap
the executable on demand (and into the kernel's own address space
instead of the process address space.)
This opens up a couple of further simplifications that will follow.
I had the wrong idea about this. Thanks to Sergey for pointing it out!
Here's what he says (reproduced for posterity):
> Private mappings protect the underlying file from the changes made by
> you, not the other way around. To quote POSIX, "If MAP_PRIVATE is
> specified, modifications to the mapped data by the calling process
> shall be visible only to the calling process and shall not change the
> underlying object. It is unspecified whether modifications to the
> underlying object done after the MAP_PRIVATE mapping is established
> are visible through the MAP_PRIVATE mapping." In practice that means
> that the pages that were already paged in don't get updated when the
> underlying file changes, and the pages that weren't paged in yet will
> load the latest data at that moment.
> The only thing MAP_FILE | MAP_PRIVATE is really useful for is mapping
> a library and performing relocations; it's definitely useless (and
> actively harmful for the system memory usage) if you only read from
> the file.
This effectively reverts e2697c2ddd.
This will be a memory usage pessimization until we actually implement
CoW sharing of the memory pages with SharedInodeVMObject.
However, it's a huge architectural improvement, so let's take it and
improve on this incrementally.
fork() should still be neutral, since all private mappings are CoW'ed.
It's now up to the caller to provide a VMObject when constructing a new
Region object. This will make it easier to handle things going wrong,
like allocation failures, etc.
If we wrote anything we should just inform userspace that we did,
and not worry about the error code. Userspace can call us again if
it wants, and we'll give them the error then.
We don't have to log the process name/PID/TID, dbg() automatically adds
that as a prefix to every line.
Also we don't have to do .characters() on Strings passed to dbg() :^)
You can now mmap a file as private and writable, and the changes you
make will only be visible to you.
This works because internally a MAP_PRIVATE region is backed by a
unique PrivateInodeVMObject instead of using the globally shared
SharedInodeVMObject like we always did before. :^)
Fixes#1045.
We now have PrivateInodeVMObject and SharedInodeVMObject, corresponding
to MAP_PRIVATE and MAP_SHARED respectively.
Note that PrivateInodeVMObject is not used yet.
Add an extra out-parameter to shbuf_get() that receives the size of the
shared buffer. That way we don't need to make a separate syscall to
get the size, which we always did immediately after.
This feels a lot more consistent and Unixy:
create_shared_buffer() => shbuf_create()
share_buffer_with() => shbuf_allow_pid()
share_buffer_globally() => shbuf_allow_all()
get_shared_buffer() => shbuf_get()
release_shared_buffer() => shbuf_release()
seal_shared_buffer() => shbuf_seal()
get_shared_buffer_size() => shbuf_get_size()
Also, "shared_buffer_id" is shortened to "shbuf_id" all around.
set_interrupted_by_death was never called whenever a thread that had
a joiner died, so the joiner remained with the joinee pointer there,
resulting in an assertion fail in JoinBlocker: m_joinee pointed to
a freed task, filled with garbage.
Thread::current->m_joinee may not be valid after the unblock
Properly return the joinee exit value to the joiner thread.
On 32-bit platforms, INT32_MIN == -INT32_MIN, so we can't expect this
to always work:
if (pid < 0)
positive_pid = -pid; // may still be negative!
This happens because the -INT32_MIN expression becomes a long and is
then truncated back to an int.
Fixes#1312.
This allows a process wich has more than 1 thread to call exec, even
from a thread. This kills all the other threads, but it won't wait for
them to finish, just makes sure that they are not in a running/runable
state.
In the case where a thread does exec, the new program PID will be the
thread TID, to keep the PID == TID in the new process.
This introduces a new function inside the Process class,
kill_threads_except_self which is called on exit() too (exit with
multiple threads wasn't properly working either).
Inside the Lock class, there is the need for a new function,
clear_waiters, which removes all the waiters from the
Process::big_lock. This is needed since after a exit/exec, there should
be no other threads waiting for this lock, the threads should be simply
killed. Only queued threads should wait for this lock at this point,
since blocked threads are handled in set_should_die.
Each process has a 1-level lookup cache for fast repeated lookups of
the same VM region (which tends to be the majority of lookups.)
The cache is used by the following syscalls: munmap, madvise, mprotect
and set_mmap_name.
After a succesful exec(), there could be a stale Region* in the lookup
cache, and the new executable was able to manipulate it using a number
of use-after-free code paths.
When committing to a new executable, disown any shared buffers that the
process was previously co-owning.
Otherwise accessing the same shared buffer ID from the new program
would cause the kernel to find a cached (and stale!) reference to the
previous program's VM region corresponding to that shared buffer,
leading to a Region* use-after-free.
Fixes#1270.
Since we're gonna throw away these stacks at the end of exec anyway,
we might as well disable profiling before starting to mess with the
process page tables. One less weird situation to worry about in the
sampling code.
Process teardown is divided into two main stages: finalize and reap.
Finalization happens in the "Finalizer" kernel and runs with interrupts
enabled, allowing destructors to take locks, etc.
Reaping happens either in sys$waitid() or in the scheduler for orphans.
The more work we can do in finalization, the better, since it's fully
pre-emptible and reduces the amount of time the system runs without
interrupts enabled.
Calling shutdown prevents further reads and/or writes on a socket.
We should do a few more things based on the type of socket, but this
initial implementation just puts the basic mechanism in place.
Work towards #428.
If there's not enough space in the output buffer for the whole sockaddr
we now simply truncate the address instead of returning EINVAL.
This patch also makes getpeername() actually return the peer address
rather than the local address.. :^)
According to POSIX, waitid() should fill si_signo and si_pid members
with zeroes if there are no children that have already changed their
state by the time of the call. Let's just fill the whole structure
with zeroes to avoid leaking kernel memory.
sys$waitid() takes an explicit description of whether it's waiting for a single
process with the given PID, all of the children, a group, etc., and returns its
info as a siginfo_t.
It also doesn't automatically imply WEXITED, which clears up the confusion in
the kernel.
This patch introduces sys$perf_event() with two event types:
- PERF_EVENT_MALLOC
- PERF_EVENT_FREE
After the first call to sys$perf_event(), a process will begin keeping
these events in a buffer. When the process dies, that buffer will be
written out to "perfcore" in the current directory unless that filename
is already taken.
This is probably not the best way to do this, but it's a start and will
make it possible to start doing memory allocation profiling. :^)
Before putting itself back on the wait queue, the finalizer task will
now check if there's more work to do, and if so, do it first. :^)
This patch also puts a bunch of process/thread debug logging behind
PROCESS_DEBUG and THREAD_DEBUG since it was unbearable to debug this
stuff with all the spam.
If the waitee process is dead, we don't need to inspect the thread.
This fixes an issue with sys$waitpid() failing before reap() since
dead processes will have no remaining threads alive.
There was a race window in a bunch of syscalls between calling
Thread::from_tid() and checking if the found thread was in the same
process as the calling thread.
If the found thread object was destroyed at that point, there was a
use-after-free that could be exploited by filling the kernel heap with
something that looked like a thread object.
Memory validation is used to verify that user syscalls are allowed to
access a given memory range. Ring 0 threads never make syscalls, and
so will never end up in validation anyway.
The reason we were allowing kmalloc memory accesses is because kernel
thread stacks used to be allocated in kmalloc memory. Since that's no
longer the case, we can stop making exceptions for kmalloc in the
validation code.
Move timeout management to the ReadBlocker and WriteBlocker classes.
Also get rid of the specialized ReceiveBlocker since it no longer does
anything that ReadBlocker can't do.
Vector::ensure_capacity() makes sure the underlying vector buffer can
contain all the data, but it doesn't update the Vector::size().
As a result, writev() would simply collect all the buffers to write,
and then do nothing.
Move all the fork-specific inheritance logic to sys$fork(), and all the
stuff for setting up stdio for non-fork ring 3 processes moves to
Process::create_user_process().
Also: we were setting up the PGID, SID and umask twice. Also the code
for copying the open file descriptors was overly complicated. Now it's
just a simple Vector copy assignment. :^)
When using dbg() in the kernel, the output is automatically prefixed
with [Process(PID:TID)]. This makes it a lot easier to understand which
thread is generating the output.
This patch also cleans up some common logging messages and removes the
now-unnecessary "dbg() << *current << ..." pattern.
Sergey suggested that having a non-zero O_RDONLY would make some things
less confusing, and it seems like he's right about that.
We can now easily check read/write permissions separately instead of
dancing around with the bits.
This patch also fixes unveil() validation for O_RDWR which previously
forgot to check for "r" permission.
This syscall is a complement to pledge() and adds the same sort of
incremental relinquishing of capabilities for filesystem access.
The first call to unveil() will "drop a veil" on the process, and from
now on, only unveiled parts of the filesystem are visible to it.
Each call to unveil() specifies a path to either a directory or a file
along with permissions for that path. The permissions are a combination
of the following:
- r: Read access (like the "rpath" promise)
- w: Write access (like the "wpath" promise)
- x: Execute access
- c: Create/remove access (like the "cpath" promise)
Attempts to open a path that has not been unveiled with fail with
ENOENT. If the unveiled path lacks sufficient permissions, it will fail
with EACCES.
Like pledge(), subsequent calls to unveil() with the same path can only
remove permissions, not add them.
Once you call unveil(nullptr, nullptr), the veil is locked, and it's no
longer possible to unveil any more paths for the process, ever.
This concept comes from OpenBSD, and their implementation does various
things differently, I'm sure. This is just a first implementation for
SerenityOS, and we'll keep improving on it as we go. :^)
uintptr_t is 32-bit or 64-bit depending on the target platform.
This will help us write pointer size agnostic code so that when the day
comes that we want to do a 64-bit port, we'll be in better shape.
Instead of restoring CR3 to the current process's paging scope when a
ProcessPagingScope goes out of scope, we now restore exactly whatever
the CR3 value was when we created the ProcessPagingScope.
This fixes breakage in situations where a process ends up with nested
ProcessPagingScopes. This was making profiling very fragile, and with
this change it's now possible to profile g++! :^)
This will panic the kernel immediately if these functions are misused
so we can catch it and fix the misuse.
This patch fixes a couple of misuses:
- create_signal_trampolines() writes to a user-accessible page
above the 3GB address mark. We should really get rid of this
page but that's a whole other thing.
- CoW faults need to use copy_from_user rather than copy_to_user
since it's the *source* pointer that points to user memory.
- Inode faults need to use memcpy rather than copy_to_user since
we're copying a kernel stack buffer into a quickmapped page.
This should make the copy_to/from_user() functions slightly less useful
for exploitation. Before this, they were essentially just glorified
memcpy() with SMAP disabled. :^)
Previously, VFS::open() would only use the passed flags for permission checking
purposes, and Process::sys$open() would set them on the created FileDescription
explicitly. Now, they should be set by VFS::open() on any files being opened,
including files that the kernel opens internally.
This also lets us get rid of the explicit check for whether or not the returned
FileDescription was a preopen fd, and in fact, fixes a bug where a read-only
preopen fd without any other flags would be considered freshly opened (due to
O_RDONLY being indistinguishable from 0) and granted a new set of flags.
Kernel processes just do not need them.
This also avoids touching the file (sub)system early in the boot process when
initializing the colonel process.
Right now, permission flags passed to VFS::open() are effectively ignored, but
that is going to change.
* O_RDONLY is 0, but it's still nicer to pass it explicitly
* POSIX says that binding a Unix socket to a symlink shall fail with EADDRINUSE
It's now an error to sys$mmap() a file as writable if it's currently
mapped executable by anyone else.
It's also an error to sys$execve() a file that's currently mapped
writable by anyone else.
This fixes a race condition vulnerability where one program could make
modifications to an executable while another process was in the kernel,
in the middle of exec'ing the same executable.
Test: Kernel/elf-execve-mmap-race.cpp
As suggested by Joshua, this commit adds the 2-clause BSD license as a
comment block to the top of every source file.
For the first pass, I've just added myself for simplicity. I encourage
everyone to add themselves as copyright holders of any file they've
added or modified in some significant way. If I've added myself in
error somewhere, feel free to replace it with the appropriate copyright
holder instead.
Going forward, all new source files should include a license header.
The syscall is now called sys$open(), but it behaves like the old sys$openat().
In userspace, open_with_path_length() is made a wrapper over openat_with_path_length().
This patch adds a new "accept" promise that allows you to call accept()
on an already listening socket. This lets programs set up a socket for
for listening and then dropping "inet" and/or "unix" so that only
incoming (and existing) connections are allowed from that point on.
No new outgoing connections or listening server sockets can be created.
In addition to accept() it also allows getsockopt() with SOL_SOCKET
and SO_PEERCRED, which is used to find the PID/UID/GID of the socket
peer. This is used by our IPC library when creating shared buffers that
should only be accessible to a specific peer process.
This allows us to drop "unix" in WindowServer and LookupServer. :^)
It also makes the debugging/introspection RPC sockets in CEventLoop
based programs work again.
It was possible to craft a custom ELF executable that when symbolicated
would cause the kernel to read from user-controlled addresses anywhere
in memory. You could then fetch this memory via /proc/PID/stack
We fix this by making ELFImage hand out StringView rather than raw
const char* for symbol names. In case a symbol offset is outside the
ELF image, you get a null StringView. :^)
Test: Kernel/elf-symbolication-kernel-read-exploit.cpp
The O_NOFOLLOW_NOERROR is an internal kernel mechanism used for the
implementation of sys$readlink() and sys$lstat().
There is no reason to allow userspace to open symlinks directly.
We now can create a cacheable Region, so when map() is called, if a
Region is cacheable then all the virtual memory space being allocated
to it will be marked as not cache disabled.
In addition to that, OS components can create a Region that will be
mapped to a specific physical address by using the appropriate helper
method.
This is needed to eliminate a race in Thread::wait_on() where we'd
otherwise have to wait until after unlocking the process lock before
we can disable interrupts.
This patch changes how exec() figures out which program image to
actually load. Previously, we opened the path to our main executable in
find_shebang_interpreter_for_executable, read the first page (or less,
if the file was smaller) and then decided whether to recurse with the
interpreter instead. We then then re-opened the main executable in
do_exec.
However, since we now want to parse the ELF header and Program Headers
of an elf image before even doing any memory region work, we can change
the way this whole process works. We open the file and read (up to) the
first page in exec() itself, then pass just the page and the amount read
to find_shebang_interpreter_for_executable. Since we now have that page
and the FileDescription for the main executable handy, we can do a few
things. First, validate the ELF header and ELF program headers for any
shenanigans. ELF32 Little Endian i386 only, please. Second, we can grab
the PT_INTERP interpreter from any ET_DYN files, and open that guy right
away if it exists. Finally, we can pass the main executable's and
optionally the PT_INTERP interpreter's file descriptions down to do_exec
and not have to feel guilty about opening the file twice.
In do_exec, we now have a choice. Are we going to load the main
executable, or the interpreter? We could load both, but it'll be way
easier for the inital pass on the RTLD if we only load the interpreter.
Then it can load the main executable itself like any old shared object,
just, the one with main in it :). Later on we can load both of them
into memory and the RTLD can relocate itself before trying to do
anything. The way it's written now the RTLD will get dibs on its
requested virtual addresses being the actual virtual addresses.
Right now there is a significant amount of boiler plate code required
to validate user mode parameters in syscalls. In an attempt to reduce
this a bit, introduce validate_read_and_copy_typed which combines the
usermode address check and does the copy internally if the validation
passes. This cleans up a little bit of code from a significant amount
of syscalls.
It looks like setkeymap was missed when
the SMAP functionality was introduced.
Disable SMAP only in the scope where we
actually read the usermode addresses.
Now that the templated version of copy_from_user exists
their is normally no reason to use the version which
takes the number of bytes to copy. Move to the templated
version where possible.
Since a chroot is in many ways similar to a separate root mount, we can also
apply mount flags to it as if it was an actual mount. These flags will apply
whenever the chrooted process accesses its root directory, but not when other
processes access this same directory for the outside. Since it's common to
chdir("/") immediately after chrooting (so that files accessed through the
current directory inherit the same mount flags), this effectively allows one to
apply additional limitations to a process confined inside a chroot.
To this effect, sys$chroot() gains a mount_flags argument (exposed as
chroot_with_mount_flags() in userspace) which can be set to all the same values
as the flags argument for sys$mount(), and additionally to -1 to keep the flags
set for that file system. Note that passing 0 as mount_flags will unset any
flags that may have been set for the file system, not keep them.
Instead of looking up device metadata and then looking up a device by that
metadata explicitly, just use VFS::open(). This also means that attempting to
mount a device residing on a MS_NODEV file system will properly fail.
You can now bind-mount files and directories. This essentially exposes an
existing part of the file system in another place, and can be used as an
alternative to symlinks or hardlinks.
Here's an example of doing this:
# mkdir /tmp/foo
# mount /home/anon/myfile.txt /tmp/foo -o bind
# cat /tmp/foo
This is anon's file.
We now support these mount flags:
* MS_NODEV: disallow opening any devices from this file system
* MS_NOEXEC: disallow executing any executables from this file system
* MS_NOSUID: ignore set-user-id bits on executables from this file system
The fourth flag, MS_BIND, is defined, but currently ignored.
O_EXEC is mentioned by POSIX, so let's have it. Currently, it is only used
inside the kernel to ensure the process has the right permissions when opening
an executable.
At the moment, the actual flags are ignored, but we correctly propagate them all
the way from the original mount() syscall to each custody that resides on the
mounted FS.
While I was updating syscalls to stop passing null-terminated strings,
I added some helpful struct types:
- StringArgument { const char*; size_t; }
- ImmutableBuffer<Data, Size> { const Data*; Size; }
- MutableBuffer<Data, Size> { Data*; Size; }
The Process class has some convenience functions for validating and
optionally extracting the contents from these structs:
- get_syscall_path_argument(StringArgument)
- validate_and_copy_string_from_user(StringArgument)
- validate(ImmutableBuffer)
- validate(MutableBuffer)
There's still so much code around this and I'm wondering if we should
generate most of it instead. Possible nice little project.
In order to preserve the absolute path of the process root, we save the
custody used by chroot() before stripping it to become the new "/".
There's probably a better way to do this.
The chroot() syscall now allows the superuser to isolate a process into
a specific subtree of the filesystem. This is not strictly permanent,
as it is also possible for a superuser to break *out* of a chroot, but
it is a useful mechanism for isolating unprivileged processes.
The VFS now uses the current process's root_directory() as the root for
path resolution purposes. The root directory is stored as an uncached
Custody in the Process object.
Note that I'm developing some helper types in the Syscall namespace as
I go here. Once I settle on some nice types, I will convert all the
other syscalls to use them as well.
The join_thread() syscall is not supposed to be interruptible by
signals, but it was. And since the process death mechanism piggybacked
on signal interrupts, it was possible to interrupt a pthread_join() by
killing the process that was doing it, leading to confusing due to some
assumptions being made by Thread::finalize() for threads that have a
pending joiner.
This patch fixes the issue by making "interrupted by death" a distinct
block result separate from "interrupted by signal". Then we handle that
state in join_thread() and tidy things up so that thread finalization
doesn't get confused by the pending joiner being gone.
Test: Tests/Kernel/null-deref-crash-during-pthread_join.cpp
The userspace execve() wrapper now measures all the strings and puts
them in a neat and tidy structure on the stack.
This way we know exactly how much to copy in the kernel, and we don't
have to use the SMAP-violating validate_read_str(). :^)
When loading a new executable, we now map the ELF image in kernel-only
memory and parse it there. Then we use copy_to_user() when initializing
writable regions with data from the executable.
Note that the exec() syscall still disables SMAP protection and will
require additional work. This patch only affects kernel-originated
process spawns.
Make mmap return -ENOTSUP in this case to make sure users don't get
confused and think they're using a private mapping when it's actually
shared. It's currenlty not possible to open a file and mmap it
MAP_PRIVATE, and change the perms of the private mapping to ones that
don't match the permissions of the underlying file.
This patch fixes some issues with the mmap() and mprotect() syscalls,
neither of whom were checking the permission bits of the underlying
files when mapping an inode MAP_SHARED.
This made it possible to subvert execution of any running program
by simply memory-mapping its executable and replacing some of the code.
Test: Kernel/mmap-write-into-running-programs-executable-file.cpp
This encourages callers to strongly reference file descriptions while
working with them.
This fixes a use-after-free issue where one thread would close() an
open fd while another thread was blocked on it becoming readable.
Test: Kernel/uaf-close-while-blocked-in-read.cpp
Before this, you could make the kernel copy memory from anywhere by
setting up an ELF executable with a program header specifying file
offsets outside the file.
Since ELFImage didn't even know how large it was, we had no clue that
we were copying things from outside the ELF.
Fix this by adding a size field to ELFImage and validating program
header ranges before memcpy()'ing to them.
The ELF code is definitely going to need more validation and checking.
This code had been misinterpreting the Multiboot ELF section headers
since the beginning. Furthermore QEMU wasn't even passing us any
headers at all, so this wasn't checking anything.
This patch introduces a helpful copy_string_from_user() function
that takes a bounded null-terminated string from userspace memory
and copies it into a String object.
Supervisor Mode Access Prevention (SMAP) is an x86 CPU feature that
prevents the kernel from accessing userspace memory. With SMAP enabled,
trying to read/write a userspace memory address while in the kernel
will now generate a page fault.
Since it's sometimes necessary to read/write userspace memory, there
are two new instructions that quickly switch the protection on/off:
STAC (disables protection) and CLAC (enables protection.)
These are exposed in kernel code via the stac() and clac() helpers.
There's also a SmapDisabler RAII object that can be used to ensure
that you don't forget to re-enable protection before returning to
userspace code.
THis patch also adds copy_to_user(), copy_from_user() and memset_user()
which are the "correct" way of doing things. These functions allow us
to briefly disable protection for a specific purpose, and then turn it
back on immediately after it's done. Going forward all kernel code
should be moved to using these and all uses of SmapDisabler are to be
considered FIXME's.
Note that we're not realizing the full potential of this feature since
I've used SmapDisabler quite liberally in this initial bring-up patch.
Our syscall calling convention only allows passing up to 3 arguments in
registers. For syscalls that take more arguments, we bake them into a
struct and pass a pointer to that struct instead.
When doing pointer validation, this is what we would do:
1) Validate the "params" struct
2) Validate "params->some_pointer"
3) ... other stuff ...
4) Use "params->some_pointer"
Since the parameter struct is stored in userspace, it can be modified
by userspace after validation has completed.
This was a recurring pattern in many syscalls that was further hidden
by me using structured binding declarations to give convenient local
names to things in the parameter struct:
auto& [some_pointer, ...] = *params;
memcpy(some_pointer, ...);
This devilishly makes "some_pointer" look like a local variable but
it's actually more like an alias for "params->some_pointer" and will
expand to a dereference when accessed!
This patch fixes the issues by explicitly copying out each member from
the parameter structs before validating them, and then never using
the "param" pointers beyond that.
Thanks to braindead for finding this bug! :^)
In order to ensure a specific owner and mode when the local socket
filesystem endpoint is instantiated, we need to be able to call
fchmod() and fchown() on a socket fd between socket() and bind().
This is because until we call bind(), there is no filesystem inode
for the socket yet.
We now have these API's in <Kernel/Random.h>:
- get_fast_random_bytes(u8* buffer, size_t buffer_size)
- get_good_random_bytes(u8* buffer, size_t buffer_size)
- get_fast_random<T>()
- get_good_random<T>()
Internally they both use x86 RDRAND if available, otherwise they fall
back to the same LCG we had in RandomDevice all along.
The main purpose of this patch is to give kernel code a way to better
express its needs for random data.
Randomness is something that will require a lot more work, but this is
hopefully a step in the right direction.
It was previously possible to write to read-only file descriptors,
and read from write-only file descriptors.
All FileDescription objects now start out non-readable + non-writable,
and whoever is creating them has to "manually" enable reading/writing
by calling set_readable() and/or set_writable() on them.
This code never worked, as was never used for anything. We can build
a much better SHM implementation on top of TmpFS or similar when we
get to the point when we need one.
Split a region into two/three if the desired mprotect range is a strict
subset of an existing region. We can then set the access bits on a new
region that is just our desired range and add both the new
desired subregion and the leftovers back to our page tables.
We now validate the full range of userspace memory passed into syscalls
instead of just checking that the first and last byte of the memory are
in process-owned regions.
This fixes an issue where it was possible to avoid rejection of invalid
addresses that sat between two valid ones, simply by passing a valid
address and a size large enough to put the end of the range at another
valid address.
I added a little test utility that tries to provoke EFAULT in various
ways to help verify this. I'm sure we can think of more ways to test
this but it's at least a start. :^)
Thanks to mozjag for pointing out that this code was still lacking!
Incidentally this also makes backtraces work again.
Fixes#989.
All threads were running with iomapbase=0 in their TSS, which the CPU
interprets as "there's an I/O permission bitmap starting at offset 0
into my TSS".
Because of that, any bits that were 1 inside the TSS would allow the
thread to execute I/O instructions on the port with that bit index.
Fix this by always setting the iomapbase to sizeof(TSS32), and also
setting the TSS descriptor's limit to sizeof(TSS32), effectively making
the I/O permissions bitmap zero-length.
This should make it no longer possible to do I/O from userspace. :^)
This patch introduces a syscall:
int set_thread_boost(int tid, int amount)
You can use this to add a permanent boost value to the effective thread
priority of any thread with your UID (or any thread in the system if
you are the superuser.)
This is quite crude, but opens up some interesting opportunities. :^)
Threads now have numeric priorities with a base priority in the 1-99
range.
Whenever a runnable thread is *not* scheduled, its effective priority
is incremented by 1. This is tracked in Thread::m_extra_priority.
The effective priority of a thread is m_priority + m_extra_priority.
When a runnable thread *is* scheduled, its m_extra_priority is reset to
zero and the effective priority returns to base.
This means that lower-priority threads will always eventually get
scheduled to run, once its effective priority becomes high enough to
exceed the base priority of threads "above" it.
The previous values for ThreadPriority (Low, Normal and High) are now
replaced as follows:
Low -> 10
Normal -> 30
High -> 50
In other words, it will take 20 ticks for a "Low" priority thread to
get to "Normal" effective priority, and another 20 to reach "High".
This is not perfect, and I've used some quite naive data structures,
but I think the mechanism will allow us to build various new and
interesting optimizations, and we can figure out better data structures
later on. :^)
If an mmap fails to allocate a region, but the addr passed in was
non-zero, non-fixed mmaps should attempt to allocate at any available
virtual address.
This is memory that's loaded from an inode (file) but not modified in
memory, so still identical to what's on disk. This kind of memory can
be freed and reloaded transparently from disk if needed.
Dirty private memory is all memory in non-inode-backed mappings that's
process-private, meaning it's not shared with any other process.
This patch exposes that number via SystemMonitor, giving us an idea of
how much memory each process is responsible for all on its own.
This patch implements a simple version of the futex (fast userspace
mutex) API in the kernel and uses it to make the pthread_cond_t API's
block instead of busily sched_yield().
An arbitrary userspace address is passed to the kernel as a "token"
that identifies the futex and you can then FUTEX_WAIT and FUTEX_WAKE
that specific userspace address.
FUTEX_WAIT corresponds to pthread_cond_wait() and FUTEX_WAKE is used
for pthread_cond_signal() and pthread_cond_broadcast().
I'm pretty sure I'm missing something in this implementation, but it's
hopefully okay for a start. :^)
This is a little strange, but it's how I understand things should work.
The first thread in a new process now has TID == PID.
Additional threads subsequently spawned in that process all have unique
TID's generated by the PID allocator. TIDs are now globally unique.
The idea of all processes reliably having a main thread was nice in
some ways, but cumbersome in others. More importantly, it didn't match
up with POSIX thread semantics, so let's move away from it.
This thread gets rid of Process::main_thread() and you now we just have
a bunch of Thread objects floating around each Process.
When the finalizer nukes the last Thread in a Process, it will also
tear down the Process.
There's a bunch of more things to fix around this, but this is where we
get started :^)
While setting up the main thread stack for a new process, we'd incur
some zero-fill page faults. This was to be expected, since we allocate
a huge stack but lazily populate it with physical pages.
The problem is that page fault handlers may enable interrupts in order
to grab a VMObject lock (or to page in from an inode.)
During exec(), a process is reorganizing itself and will be in a very
unrunnable state if the scheduler should interrupt it and then later
ask it to run again. Which is exactly what happens if the process gets
pre-empted while the new stack's zero-fill page fault grabs the lock.
This patch fixes the issue by creating new main thread stacks before
disabling interrupts and going into the critical part of exec().