This replaces manually grabbing the thread's main lock.
This lets us remove the `get_thread_name` and `set_thread_name` syscalls
from the big lock. :^)
This filesystem is based on the code of the long-lived TmpFS. It differs
from that filesystem in one keypoint - its root inode doesn't have a
sticky bit on it.
Therefore, we mount it on /dev, to ensure only root can modify files on
that directory. In addition to that, /tmp is mounted directly in the
SystemServer main (start) code, so it's no longer specified in the fstab
file. We ensure that /tmp has a sticky bit and has the value 0777 for
root directory permissions, which is certainly a special case when using
RAM-backed (and in general other) filesystems.
Because of these 2 changes, it's no longer needed to maintain the TmpFS
filesystem, hence it's removed (renamed to RAMFS), because the RAMFS
represents the purpose of this filesystem in a much better way - it
relies on being backed by RAM "storage", and therefore it's easy to
conclude it's temporary and volatile, so its content is gone on either
system shutdown or unmounting of the filesystem.
This is done by merging all scattered pieces of derived classes from the
ProcFSInode class into that one class, so we don't use inheritance but
rather simplistic checks to determine the proper code for each ProcFS
inode with its specific characteristics.
For each exposed PCI device in sysfs, there's a new node called "rom"
and by reading it, it exposes the raw data of a PCI option ROM blob to
a user for examining the blob.
There are now 2 separate classes for almost the same object type:
- EnumerableDeviceIdentifier, which is used in the enumeration code for
all PCI host controller classes. This is allowed to be moved and
copied, as it doesn't support ref-counting.
- DeviceIdentifier, which inherits from EnumerableDeviceIdentifier. This
class uses ref-counting, and is not allowed to be copied. It has a
spinlock member in its structure to allow safely executing complicated
IO sequences on a PCI device and its space configuration.
There's a static method that allows a quick conversion from
EnumerableDeviceIdentifier to DeviceIdentifier while creating a
NonnullRefPtr out of it.
The reason for doing this is for the sake of integrity and reliablity of
the system in 2 places:
- Ensure that "complicated" tasks that rely on manipulating PCI device
registers are done in a safe manner. For example, determining a PCI
BAR space size requires multiple read and writes to the same register,
and if another CPU tries to do something else with our selected
register, then the result will be a catastrophe.
- Allow the PCI API to have a united form around a shared object which
actually holds much more data than the PCI::Address structure. This is
fundamental if we want to do certain types of optimizations, and be
able to support more features of the PCI bus in the foreseeable
future.
This patch already has several implications:
- All PCI::Device(s) hold a reference to a DeviceIdentifier structure
being given originally from the PCI::Access singleton. This means that
all instances of DeviceIdentifier structures are located in one place,
and all references are pointing to that location. This ensures that
locking the operation spinlock will take effect in all the appropriate
places.
- We no longer support adding PCI host controllers and then immediately
allow for enumerating it with a lambda function. It was found that
this method is extremely broken and too much complicated to work
reliably with the new paradigm being introduced in this patch. This
means that for Volume Management Devices (Intel VMD devices), we
simply first enumerate the PCI bus for such devices in the storage
code, and if we find a device, we attach it in the PCI::Access method
which will scan for devices behind that bridge and will add new
DeviceIdentifier(s) objects to its internal Vector. Afterwards, we
just continue as usual with scanning for actual storage controllers,
so we will find a corresponding NVMe controllers if there were any
behind that VMD bridge.
A lot of places were relying on AK/Traits.h to give it strnlen, memcmp,
memcpy and other related declarations.
In the quest to remove inclusion of LibC headers from Kernel files, deal
with all the fallout of this included-everywhere header including less
things.
This header has always been fundamentally a Kernel API file. Move it
where it belongs. Include it directly in Kernel files, and make
Userland applications include it via sys/ioctl.h rather than directly.
Resolves issue where a panic would occur if the file system failed to
initialize or mount, due to how the FileSystem was already added to
VFS's list. The newly-created FileSystem destructor would fail as a
result of the object still remaining in the IntrusiveList.
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.
Use this helper function in various places to replace the old code of
acquiring the SpinlockProtected<RefPtr<Jail>> of a Process to do that
validation.
Only do so after a brief check if we are in a Jail or not. This fixes
SMP, because apparently it is crashing when calling try_generate()
from the SysFSGlobalInformation::refresh_data method, so the fix for
this is to simply not do that inside the Process' Jail spinlock scope,
because otherwise we will simply have a possible flow of taking
multiple conflicting Spinlocks (in the wrong order multiple times), for
the SysFSOverallProcesses generation code:
Process::current().jail(), and then Process::for_each_in_same_jail being
called, we take Process::all_instances(), and Process::current().jail()
again.
Therefore, we should at the very least eliminate the first taking of the
Process::current().jail() spinlock, in the refresh_data method of the
SysFSGlobalInformation class.
* Fix bug where last character of a filename or extension would be
truncated (HELLO.TXT -> HELL.TX).
* Fix bug where additional NULL characters would be added to long
filenames that did not completely fill one of the Long Filename Entry
character fields.
These instances were detected by searching for files that include
AK/Memory.h, but don't match the regex:
\\b(fast_u32_copy|fast_u32_fill|secure_zero|timing_safe_compare)\\b
This regex is pessimistic, so there might be more files that don't
actually use any memory function.
In theory, one might use LibCPP to detect things like this
automatically, but let's do this one step after another.
These instances were detected by searching for files that include
Kernel/Debug.h, but don't match the regex:
\\bdbgln_if\(|_DEBUG\\b
This regex is pessimistic, so there might be more files that don't check
for any real *_DEBUG macro. There seem to be no corner cases anyway.
In theory, one might use LibCPP to detect things like this
automatically, but let's do this one step after another.
This step would ideally not have been necessary (increases amount of
refactoring and templates necessary, which in turn increases build
times), but it gives us a couple of nice properties:
- SpinlockProtected inside Singleton (a very common combination) can now
obtain any lock rank just via the template parameter. It was not
previously possible to do this with SingletonInstanceCreator magic.
- SpinlockProtected's lock rank is now mandatory; this is the majority
of cases and allows us to see where we're still missing proper ranks.
- The type already informs us what lock rank a lock has, which aids code
readability and (possibly, if gdb cooperates) lock mismatch debugging.
- The rank of a lock can no longer be dynamic, which is not something we
wanted in the first place (or made use of). Locks randomly changing
their rank sounds like a disaster waiting to happen.
- In some places, we might be able to statically check that locks are
taken in the right order (with the right lock rank checking
implementation) as rank information is fully statically known.
This refactoring even more exposes the fact that Mutex has no lock rank
capabilites, which is not fixed here.
We were already handling the rmdir("..") case by refusing to remove
directories that were not empty.
This patch removes a FIXME from January 2019 and adds a test. :^)
Dr. POSIX says that we should reject attempts to rmdir() the file named
"." so this patch does exactly that. We also add a test.
This solves a FIXME from January 2019. :^)
This commit makes it possible for a process to downgrade a file lock it
holds from a write (exclusive) lock to a read (shared) lock. For this,
the process must point to the exact range of the flock, and must be the
owner of the lock.
The fact that we used a Vector meant that even if creating a Mount
object succeeded, we were still at a risk that appending to the actual
mounts Vector could fail due to OOM condition. To guard against this,
the mount table is now an IntrusiveList, which always means that when
allocation of a Mount object succeeded, then inserting that object to
the list will succeed, which allows us to fail early in case of OOM
condition.
This solves one of the security issues being mentioned in issue #15996.
We simply don't allow creating hardlinks on paths that were not unveiled
as writable to prevent possible bypass on a certain path that was
unveiled as non-writable.
Instead, allow userspace to decide on the coredump directory path. By
default, SystemServer sets it to the /tmp/coredump directory, but users
can now change this by writing a new path to the sysfs node at
/sys/kernel/variables/coredump_directory, and also to read this node to
check where coredumps are currently generated at.
By default, disallow reading of values in that directory. Later on, we
will enable sparingly read access to specific files.
The idea that led to this mechanism was suggested by Jean-Baptiste
Boric (also known as boricj in GitHub), to prevent access to sensitive
information in the SysFS if someone adds a new file in the /sys/kernel
directory.
There's simply no benefit in allowing sandboxed programs to change the
power state of the machine, so disallow writes to the mentioned node to
prevent malicious programs to request that.
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.
Because the ".." entry in a directory is a separate inode, if a
directory is renamed to a new location, then we should update this entry
the point to the new parent directory as well.
Co-authored-by: Liav A <liavalb@gmail.com>
Each GenericInterruptHandler now tracks the number of calls that each
CPU has serviced.
This takes care of a FIXME in the /sys/kernel/interrupts generator.
Also, the lsirq command line tool now displays per-CPU call counts.
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.
Let's put the power_state global node into the /sys/kernel directory,
because that directory represents all global nodes and variables being
related to the Kernel. It's also a mutable node, that is more acceptable
being in the mentioned directory due to the fact that all other files in
the /sys/firmware directory are just firmware blobs and are not mutable
at all.
The ProcFS is an utter mess currently, so let's start move things that
are not related to processes-info. To ensure it's done in a sane manner,
we start by duplicating all /proc/ global nodes to the /sys/kernel/
directory, then we will move Userland to use the new directory so the
old directory nodes can be removed from the /proc directory.
If a program needs to execute a dynamic executable program, then it
should unveil /usr/lib/Loader.so by itself and not rely on the Kernel to
allow using this binary without any sense of respect to unveil promises
being made by the running parent program.
Previously we didn't send the SIGPIPE signal to processes when
sendto()/sendmsg()/etc. returned EPIPE. And now we do.
This also adds support for MSG_NOSIGNAL to suppress the signal.
This commit reached that goal of "safely discarding" a filesystem by
doing the following:
1. Stop using the s_file_system_map HashMap as it was an unsafe measure
to access pointers of FileSystems. Instead, make sure to register all
FileSystems at the VFS layer, with an IntrusiveList, to avoid problems
related to OOM conditions.
2. Make sure to cleanly remove the DiskCache object from a BlockBased
filesystem, so the destructor of such object will not need to do that in
the destruction point.
3. For ext2 filesystems, don't cache the root inode at m_inode_cache
HashMap. The reason for this is that when unmounting an ext2 filesystem,
we lookup at the cache to see if there's a reference to a cached inode
and if that's the case, we fail with EBUSY. If we keep the m_root_inode
also being referenced at the m_inode_cache map, we have 2 references to
that object, which will lead to fail with EBUSY. Also, it's much simpler
to always ask for a root inode and get it immediately from m_root_inode,
instead of looking up the cache for that inode.
The idea is to enable mounting FileSystem objects across multiple mounts
in contrast to what happened until now - each mount has its own unique
FileSystem object being attached to it.
Considering a situation of mounting a block device at 2 different mount
points at in system, there were a couple of critical flaws due to how
the previous "design" worked:
1. BlockBasedFileSystem(s) that pointed to the same actual device had a
separate DiskCache object being attached to them. Because both instances
were not synchronized by any means, corruption of the filesystem is most
likely achieveable by a simple cache flush of either of the instances.
2. For superblock-oriented filesystems (such as the ext2 filesystem),
lack of synchronization between both instances can lead to severe
corruption in the superblock, which could render the entire filesystem
unusable.
3. Flags of a specific filesystem implementation (for example, with xfs
on Linux, one can instruct to mount it with the discard option) must be
honored across multiple mounts, to ensure expected behavior against a
particular filesystem.
This patch put the foundations to start fix the issues mentioned above.
However, there are still major issues to solve, so this is only a start.
This flag doesn't conform to any POSIX standard nor is found in any OS
out there. The idea behind this mount flag is to ensure that only
non-regular files will be placed in a filesystem, which includes device
nodes, symbolic links, directories, FIFOs and sockets. Currently, the
only valid case for using this mount flag is for TmpFS instances, where
we want to mount a TmpFS but disallow any kind of regular file and only
allow other types of files on the filesystem.
Although this code worked quite well, it is considered to be a code
duplication with the TmpFS code which is more tested and works quite
well for a variety of cases. The only valid reason to keep this
filesystem was that it enforces that no regular files will be created at
all in the filesystem. Later on, we will re-introduce this feature in a
sane manner. Therefore, this can be safely removed after SystemServer no
longer uses this filesystem type anymore.
Instead of just having a giant KBuffer that is not resizeable easily, we
use multiple AnonymousVMObjects in one Vector to store them.
The idea is to not have to do giant memcpy or memset each time we need
to allocate or de-allocate memory for TmpFS inodes, but instead, we can
allocate only the desired block range when trying to write to it.
Therefore, it is also possible to have data holes in the inode content
in case of skipping an entire set of one data block or more when writing
to the inode content, thus, making memory usage much more efficient.
To ensure we don't run out of virtual memory range, don't allocate a
Region in advance to each TmpFSInode, but instead try to allocate a
Region on IO operation, and then use that Region to map the VMObjects
in IO loop.
We no longer require to lock the m_inode_lock in the SharedInodeVMObject
code as the methods write_bytes and read_bytes of the Inode class do
this for us now.
We move QEMU and VirtualBox shutdown sequences to a separate file, as
well as moving the i8042 reboot code sequence too to another file.
This allows us to abstract specific methods from the power state node
code of the SysFS filesystem, to allow other architectures to put their
methods there too in the future.
We make these methods non-virtual because we want to ensure we properly
enforce locking of the m_inode_lock mutex. Also, for write operations,
we want to call prepare_to_write_data before the actual write. The
previous design required us to ensure the callers do that at various
places which lead to hard-to-find bugs. By moving everything to a place
where we call prepare_to_write_data only once, we eliminate a possibilty
of forgeting to call it on some code path in the kernel.
The block list required a bit of work, and now the only method being
declared const to bypass its const-iness is the read_bytes method that
calls a new method called compute_block_list_with_exclusive_locking that
takes care of proper locking before trying to update the block list data
of the ext2 inode.
Before this change, we had File::mmap() which did all the work of
setting up a VMObject, and then creating a Region in the current
process's address space.
This patch simplifies the interface by removing the region part.
Files now only have to return a suitable VMObject from
vmobject_for_mmap(), and then sys$mmap() itself will take care of
actually mapping it into the address space.
This fixes an issue where we'd try to block on I/O (for inode metadata
lookup) while holding the address space spinlock. It also reduces time
spent holding the address space lock.
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 having three separate APIs (one for each timestamp),
there's now only Inode::update_timestamps() and it takes 3x optional
timestamps. The non-empty timestamps are updated while holding the inode
mutex, and the outside world no longer has to look at intermediate
timestamp states.
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.
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.
Instead of having two separate implementations of AK::RefCounted, one
for userspace and one for kernelspace, there is now RefCounted and
AtomicRefCounted.
All users which relied on the default constructor use a None lock rank
for now. This will make it easier to in the future remove LockRank and
actually annotate the ranks by searching for None.
We ensure that when we call SharedInodeVMObject::sync we lock the inode
lock before calling Inode virtual write_bytes method directly to avoid
assertion on the unlocked inode lock, as it was regressed recently. This
is not a complete fix as the need to lock from each path before calling
the write_bytes method should be avoided because it can lead to
hard-to-find bugs, and this commit only fixes the problem temporarily.
This enum was created to help put distinction between the commandset and
the interface type, as ATAPI devices are simply ATA devices utilizing
the SCSI commandset. Because we don't support ATAPI, putting such type
of distinction is pointless, so let's remove this for now.
Instead of requiring each FileSystem implementation to call this method
when trying to write data, do the calls at 2 points to avoid further
calls (or lack of them due to not remembering to use it) at other files
and locations in the codebase.
We should actually start counting from the parent directory and not from
the symbolic link as it will represent a wrong count of hops from the
actual mountpoint.
The symlinks in /sys/dev/block and /sys/dev/char worked only by luck,
because I have set it to the wrong parent directory which is the
/sys/dev directory, so with the symlink it was 3 hops to /sys, together
with the root directory, therefore, everything seemed to work.
Now that the device symlinks in /sys/dev/block and /sys/dev/char are set
to the right parent directory and we start measure hops from root
directory with the parent directory of a symlink, everything seem to
work correctly now.
It is starting to get a little messy with how each device can try to add
or remove itself to either /sys/dev/block or /sys/dev/char directories.
To better do this, we introduce 4 virtual methods to take care of that,
so until we ensure all nodes in /sys/dev/block and /sys/dev/char are
actual symlinks, we allow the Device base class to call virtual methods
upon insertion or before being destroying, so it add itself elegantly to
either of these directories or remove itself when needed.
For special cases where we need to create symlinks, we have two virtual
methods to be called otherwise to do almost the same thing mentioned
before, but to use symlinks instead.
Under normal conditions (when mounting SysFS in /sys), there will be a
new directory in the /sys/devices directory called "graphics".
For now, under that directory there will be only a sub-directory called
"connectors" which will contain all DisplayConnectors' details, each in
its own sub-directory too, distinguished in naming with its minor
number.
Therefore, /sys/devices/graphics/connectors/MINOR_NUMBER/ will contain:
- General device attributes such as mutable_mode_setting_capable,
double_buffering_capable, flush_support, partial_flush_support and
refresh_rate_support. These values are exposed in the ioctl interface
of the DisplayConnector class too, but these can be useful later on
for command line utilities that want/need to expose these basic
settings.
- The EDID blob, simply named "edid". This will help userspace to fetch
the edid without the need of using the ioctl interface later on.
This change in fact does the following:
1. Use support for symlinks between /sys/dev/block/ storage device
identifier nodes and devices in /sys/devices/storage/{LUN}.
2. Add basic nodes in a /sys/devices/storage/{LUN} directory, to let
userspace to know about the device and its details.
This enforces us to remove duplicated code across the SysFS code. This
results in great simplification of how the SysFS works now, because we
enforce one way to treat SysFSDirectory objects.
This will be used later on to help connecting a node at /sys/dev/block/
that represents a Storage device to a directory in /sys/devices/storage/
with details on that device in that directory.
These methods will be used later on to introduce symbolic links support
in the SysFS, so the kernel will be able to resolve relative paths of
components in filesystem based on using the m_parent_directory pointer
in each SysFSComponent object.
This folder in the SysFS code represents everything related to /sys/dev,
which is a directory meant to be a convenient interface to track all IDs
of all block and character devices (ID = major:minor numbers).
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.
I haven't found any POSIX specification on this, but the Linux kernel
appears to handle it like that.
This is required by QEMU, as it just bulk-unlocks all its file locking
bytes without checking first if they are held.
In a previous commit I moved everything into the new subdirectories in
FileSystem/SysFS directory without trying to actually make changes in
the code itself too much. Now it's time to split the code to make it
more readable and understandable, hence this change occurs now.
As with the previous commit, we put a distinction between filesystems
that require a file description and those which don't, but now in a much
more readable mechanism - all initialization properties as well as the
create static method are grouped to create the FileSystemInitializer
structure. Then when we need to initialize an instance, we iterate over
a table of these structures, checking for matching structure and then
validating the given arguments from userspace against the requirements
to ensure we can create a valid instance of the requested filesystem.
This exposes the child processes for a process as a directory
of symlinks to the respective /proc entries for each child.
This makes for an easier and possibly more efficient way
to find and count a process's children. Previously the only
method was to parse the entire /proc/all JSON file.
Coverage tools like LLVM's source-based coverage or GNU's --coverage
need to be able to write out coverage files from any binary, regardless
of its security posture. Not ignoring these pledges and veils means we
can't get our coverage data out without playing some serious tricks.
However this is pretty terrible for normal exeuction, so only skip these
checks when we explicitly configured userspace for coverage.
AnonymousFile always allocates in multiples of a page size when created
with anon_create. This is especially an issue if we use AnonymousFile
shared memory to store a shared data structure that isn't exactly a
multiple of a page in size. Therefore, we can just allow mmaps of
AnonymousFile to map only an initial part of the shared memory.
This makes SharedSingleProducerCircularQueue work when it's introduced
later.
This is important for dmidecode because it does an fstat on the DMI
blobs, trying to figure out their size. Because we already know the size
of the blobs when creating the SysFS components, there's no performance
penalty whatsoever, and this allows dmidecode to not use the /dev/mem
device as a fallback.
