base-linux uses seccomp to reduce the available system calls
to the minimum set needed to run base-linux. There are still
some syscalls that allow accessing global state which should
be further reduced.
The combination of seccomp and socket descriptor caps should
provide an intermediate level of security for base-linux
thereby enabling base-linux as a migration path from using
the Linux kernel to the use of microkernel-based Genode.
Fixes#3581
Add the option to configure the RAM quota for core on base-linux.
The environment variable GENODE_RAM_QUOTA if it exists is interpreted
as number of bytes to provide to the first init.
Issue #3762
- Since Genode::strncpy is not 100% compatible with the POSIX
strncpy function, better use a distinct name.
- Remove bogus return value from the function, easing the potential
enforcement of mandatory return-value checks later.
Fixes#3752
This patch is a workaround for the missing implementation of
'Pd_session::transfer_quota' interface by the GDB monitor's PD service.
The missing implementation becomes problematic with the changes of #3750
that enabled the cap-quota accounting for core's CPU service.
In regular scenarios without the GDB monitor, the client of
'Cpu_session::create_thread' deals with Out_of_caps or Out_of_ram by
upgrading the CPU session's cap and RAM quotas. This, in turn, results
in a sequence of 'transfer_quota' operations at the parent.
Since GDB monitor implements a custom PD service, these 'transfer_quota'
calls try to transfer quota between sessions provided by core and those
provided by the GDB monitor. This does of course not work. To fix this
issue, the GDB monitor needs a major overhaul. This patch side-steps
the problem by handing Out_of_caps and Out_of_ram from the debuging
target.
As discovered by Johannes Kliemann, peeking at buffered socket data
using 'recv' and 'MSG_PEEK' is not supported. Read a "peek" control file
from the socket directory to attempt to peek into buffers at the
socket_fs. Support for every feature of POSIX sockets cannot be
expected, but this one is trivial to implement.
Fix#2875
The driver always opens the audio device in duplex mode, i.e.,
playback and recording. Setting the 'playback' or 'recording' attribute
only influnces the service announcement. Due to changes made in a more
recent OpenBSD release recording must be set enabled explicitly anyway.
Since we already provide the interface that mirrors the one used by
OpenBSD 1:1 in the configuration use that and the remove the additional
config attributes.
Fixes#3757.
The former ldso-startup static library (now called ldso_so_support) is
used to spice each shared object/library with local support code for the
dynamic linker (execution of static constructors and ARM-EABI).
Therefore, the library must be statically linked to each dynamic
library.
As a result recipes for dynamic libraries must always depend on the "so"
API, which makes ldso_so_support.mk and so_support.c available
independent of "base". Additionally, ldso_so_support is also provided in
the libc API to cut the dependency early for libc/posix libraries.
Issue #3720
If a "cat" tool as simple as the removed one is needed, it should be
implemented with Goa, only depend on libc/posix (*not* base), and be
named "simple_cat". Up to today, the misleading naming of the removed
tool, tricked me into believing "test-pipe" tests POSIX pipes, which it
did not!
This patch fixes the handling of the corner case where the allocation of
a trace buffer throws 'Out_of_caps' or 'Out_of_ram'. Under this
circumstance, the '_buffer' would still be flagged with the 'size',
which prevented any subsequent allocation attempt. This patch fixes the
problem by initializing the 'size' after the potentially throwing
allocation.
The problem triggered with the test-trace_logger after the accounting of
core's TRACE service (replacing the 'Allocator_guard' by
'Constrained_ram_allocator') became more accurate.
Related to issue #3750
The 'WHITESPACE' case of the _calc_len method wrongly accessed the
character before checking upper bound of the token. The problem is fixed
by switching the order of both conditions.
Fixes#3756
This patch removes old 'Allocator_guard' utility and replaces its use
with the modern 'Constrained_ram_allocator'.
The adjustment of core in this respect has the side effect of a more
accurate capability accounting in core's CPU, TRACE, and RM services.
In particular, the dataspace capabilities needed for core-internal
allocations via the 'Sliced_heap' are accounted to the client now.
The same goes for nitpicker and nic_dump as other former users of the
allocator guard. Hence, the patch also touches code at the client and
server sides related to these services.
The only remaining user of the 'Allocator_guard' is the Intel GPU
driver. As the adaptation of this component would be too invasive
without testing, this patch leaves this component unchanged by keeping a
copy of the 'allocator_guard.h' locally at the component.
Fixes#3750
This patch largely reverts the commit "base: lay groundwork for
base-linux caps change" because the use of 'epoll' instead of 'select'
alleviated the need to allocate large FD sets, which motivated the
introduction of the 'Native_context' hook.
Related to issue #3581
The sandbox library supports the forwarding of session requests from the
outside to one of the hosted children according to a policy. This patch
introduces the distinction between two cases, which previously triggered
the denial of the session request.
- There exists no matching policy for the requested session
- There exists a matching policy but the referred server child
does not exist (yet)
Whereas the proper response to the first case is the denial of the
request, the second case can occur in situation where a dynamic init is
used to implement a staged startup, for example via the deploy
mechanism. In such cases, a policy may exist as a static rule while the
server has not been started yet. This patch changes the behavior such
that such requests are stalled.
The patch is accompanied with test cases for exercising both situations.
Fixes#3733
If trace is enabled for component than an attempt to put message into
trace buffer is performed using log_output policy. If it succeeds than
message is not put to logs using log service.
Fixes#3714
On Linux, Genode used to represent each RPC object by a socket
descriptor of the receiving thread (entrypoint) and a globally-unique
value that identifies the object. Because the latter was transferred as
plain message payload, clients had to be trusted to not forge the
values. For this reason, Linux could not be considered as a productive
Genode base platform but remained merely a development vehicle.
This patch changes the RPC mechanism such that each RPC object is
represented by a dedicated socket pair. Entrypoints wait on a set of
the local ends of the socket pairs of all RPC objects managed by the
respective entrypoint. The epoll kernel interface is used as the
underlying mechanism to wait for a set of socket descriptors at the
server side.
When delegating a capability, the remote end of the socket pair is
transferred to the recipient along with a plaintext copy of the
socket-descriptor value of the local end. The latter value serves as a
hint for re-identifiying a capability whenever it is delegated back to
its origin. Note that the client is not trusted to preserve this
information. The integrity of the hint value is protected by comparing
the inode values of incoming and already present capablities at the
originating site (whenever the capability is invoked or presented to the
owner of the RPC object).
The new mechanism effectively equips base-linux with Genode's capablity
model as described in the Chapter 3 of the Genode Foundations book.
That said, the sandboxing of components cannot be assumed at this point
because each component has still direct access to the Linux system-call
interface.
This patch is based on the extensive exploration work conducted by
Stefan Thoeni who strongly motivated the inclusion of this feature into
Genode.
Issue #3581
Older ARM processors like ARMv6, or Cortex A8 need to write back changes
of the page-tables to physical ram because the MMU does not use the cache.
This naturally needs to be done not only when adding a mapping,
but on removal too.
Fix#3715
The topics are either covered by the Genode Founations book for by our
tools, in particular the integration of the prepare_port mechanism with
the run tool.
- don't use 'qemu -serial mon:stdio' anymore as it no longer works as
expected
- use "bash -l" with [terminal] to read user's profile configuration,
e.g., PATH settings
- added missing boot modules and cap quotas
This commit puts all C++ runtime/support symbols of ld.lib.so in a
dedicated section of base/lib/symbols/ld and mirrors the section to
libports/lib/symbols/libc. So, the libc ABI resolves potential C++
runtime dependencies of base-ABI-agnostic components at link time. The
runtime resolution is done by the linker by symbol lookup in ld.lib.so.
Issue #3720
config_0.xml.tmp:26: element config: Schemas validity error :
Element 'config': Character content other than whitespace is
not allowed because the content type is 'element-only'
Issue #3612
This patch increases the default limit of the maximum number of open
file descriptors to the hard limit of the system. This is needed for
complex scenarios, which require more FDs than the default of 1024
at core.
Related to issue #3581Fixes#3721
This commit fixes a regression introduced by "window_layouter: add
dynamic screen handling", rendering the sub-division of screens in
columns and row unusable. The said commit removed a condition needed
for the correct window placement. This patch restores the condition.
Issues #3646
This patch also introduces the use of the RTC for creating directory
names and increases the rate of snapshot creation from 10 seconds to
one minute to make sure that directories are named differently when
reading the RTC clock at granularity of minutes.
Issue #3696
This patch avoids calls of '__sys_getstatfs' (indirectly via
'getmntinfo'), which does not return any meaningful values in Genode's
libc. Otherwise, the libc's dummy implementation prompts resize2fs to
back out.
Issue #3696
This patch makes the e2fsprogs 'mke2fs' and 'resize2fs' available as
standalone components that can be used by Sculpt OS directly without
a Unix emulation environment.
As a result of the API change the memory handling could be simplified.
Since the Block session dataspace is now directly used for DMA, we
actually only have to provide the memory for setting up PRP lists for
large requests (for the moment more than 8 KiB of data).
As we limit the maximum data transfer length to 2 MiB, we get by with
just a page per request. Those memory is allocated beforehand for the
maximum number of I/O requests, which got bumbed to 512 entries. Since
not all NVMe controllers support such large a maximum data transfer
length and this many entries, especially older ones, the values are
capped according to the properties of the controller during
initialization. (The memory demands of the component are around 3 MiB
due to setting up for the common case, even if a particular controller
is only able to make use of less.)
(Although there are controllers whose maximum memory page size is more
than 4K, the driver is hardcoded to solely use 4K pages.)
In addition to those changes, the driver now supports the 'SYNC' and
'TRIM' operations of the Block session by using the NVMe 'FLUSH' and
'WRITE_ZEROS' commands.
Fixes#3702.
The former scheme left open a race window between
_process_incoming_signals() and wait_and_dispatch_one_io_signal()
resulting in both threads calling block_for_signal() and blocking
forever with one unprocessed signal.
Fixes#3704
This patch adds support for booting base-hw kernel on qemu-arm virt
machines. The arm_virt machine has 2GB of RAM, 2 Cortex A15 cores and
uses GICv2 interrupt controller. The arm_64_virt machine also has 2GB of
RAM, but has 4 Cortex A53 cores and uses GICv3. Both machines use PSCI
to boot additional CPU cores.
Fixes#3673
So far this only exposes two functions allowing the code to turn on
additional CPUs. There is much more functionality defined by PSCI spec
but so far its not needed by Genode.
Fixes#3672
The update modification timestamp is implemented as one-shot where
the caller is expected to try again if the first attempt failed (see
current libc implementation). So the message is misleading as further
attempts might succeed.
Issue #3713.
This patch fixes a regression with run/log_core introduced by
2064ffd64b nova: support multidimensional affinity space
The run tool uses a log message 'run_boot_string' to detect successful
boot as well as to detect unexpected reboots. This message should never
be logged twice and, thus, should not be part of the core_log. The patch
mentioned above moved the former run_boot_string after the core_log
initialization.
This is a follow-up fix for commit 202333c881.
It checks for the diagnostic registers being already setup correctly.
Otherwise on platforms with secure firmware, like Pandaboard it will stuck.
Ref #3639
ASLR collides with the libc's fork mechanism on 32-bit. In particular,
the tool_chain_auto.run scenario would sporadically fail while mirroring
the parent's address space.
Fixes#3710
This is needed to execute the tool-chain scenario on base-nova.
Otherwise, the fork mechanism stumbles upon a region conflict
between ldso allocations and the application heap.
Fixes#3706
This patch reflects the Out_of_caps exception from core to the client,
in addition to the already covered Out_of_ram exception. It thereby
eliminates a potential abort in core, which I observed with the
tool_chain_auto.run script:
Kernel: RPC upgrade_cap_slab
Error: Uncaught exception of type 'Genode::Quota_guard<Genode::Cap_quota>::Limit_exceeded'
Warning: abort called - thread: entrypoint
In addition to propagating the exception, the patch add the client-side
exception-handling code to the base library.
Fixes#3703
This run script is outdated and not regularly tested. In fact, the
tested mechanism (obtaining values from a '.sysctl/' directory) is
nowhere to be found, neither in noux nor the libc. The test still
returns success as it merely checks for the completion of the sequence,
not the printed results. Hence, it is not worth preserving.
Issue #3696
This run script tested the terminal VFS plugin as a new feature of noux.
In the meantime, the plugin remains as the only way for the interaction
of command-line-based applications with the terminal session. Hence,
the mechanism is now stressed by many other scenarios.
Issue #3696
This patch prefixes a few symbols in the make binary that are offered by
both the libc and the make binary. The clash of symbol names produces
confusing runtime linkage otherwise. This becomes a problem during
fork/evecve.
This patch reduces the size of session labels for all services other
than LOG, keeping only the last element. This avoids exceeding the
maximum label length in the presence of deep fork hierarchies, e.g., for
running the tool chain.
Fixes#3700
When the last writer closes the pipe, we must trigger the I/O progress
handler to deliver the final EOF to the client (READ_OK with out count
of 0).
Fixes#3697
This patch replaces the former Noux instance by a sub init that hosts
the combination of VFS + fs_rom + bash. If bash exits, the whole sub
init exits, which implicitly tears down the terminal session used by the
sub init. The latter is expected by the test sequence.
Issue #3696
The tar vfs plugin returns an inode value of zero and a type directory
for null records, which wrongly triggers the file-system loop detection
of the find utility. This patch returns the node pointer as inode value
instead, which is a unique value.
Fixes#3695
This patch changes the 'Single_file_system' to return NO_PERM only if
the to-be-unlinked file corresponds to the single file. This way, a
<rom> co-mounted with a <ram> file-system does not stand in the way of
unlinking files from the <ram>. The concrete symptom occurred the
following scenario:
<vfs>
<dir name="home">
<ram/>
<rom name="..."/>
</dir>
</vfs>
The following sequence of commands wrongly resulted in "Operation not
permitted":
$ mkdir -p /home/a/b/c
$ rm -f /home/a/b/c/d
In this case, rm should not fail (unlink should return ENOENT)
Fixes#3690
When updating the domain object of interfaces that stay with the same domain
during a reconfiguration, until now, the normal "detach raw" function was used.
This caused the old domain object to discard a dynamic IP config as all
interfaces detached. This caused interfaces also to discard network links
established with the old configuration although it wasn't necessary. Thus, now
we use a dedicated "detach" in case that an interface actually stays with its
domain. This new "detach" doesn't decrease the interface counter of the domain,
so, it'll not discard its dynamic IP config. If, however, during a
reconfiguration, there's no interface calling this function (all interfaces
move to another or no domain), the dynamic IP config is still discarded as
expected.
Fixes#3686
Until now, the DHCP client was called also for DHCP requests when an interface
had a domain but yet no IP config. Now, an interface distinguishes between DHCP
requests and replies first and then accordingly calls the DHCP server or the
DHCP client if they're available. This also prevents that the DHCP client has
to handle packet headers other than that of DHCP.
Fixes#3681
Let the DHCP client be a constructible member of Interface that is constructed
only as long as the interface is attached to a domain with a dynamic IP config.
This prevents DHCP client timeouts from a period with dynamic IP config to
trigger after a reconfiguration to a static IP config. Furthermore, handle
DHCP-reply packets at an interface only when the DHCP client its constructed.
Otherwise drop such packets.
Ref #3681
This commit fixes the following issues regarding cache maintainance
under ARM:
* read out I-, and D-cache line size at runtime and use the correct one
* remove 'update_data_region' call from unprivileged syscalls
* rename 'update_instr_region' syscall to 'cache_coherent_region' to
reflect what it doing, namely make I-, and D-cache coherent
* restrict 'cache_coherent_region' syscall to one page at a time
* lookup the region given in a 'cache_coherent_region' syscall in the
page-table of the PD to prevent machine exceptions in the kernel
* only clean D-cache lines, do not invalidate them when pages where
added on Cortex-A8 and ARMv6 (MMU sees phys. memory here)
* remove unused code relicts of cache maintainance
In addition it introduces per architecture memory clearance functions
used by core, when preparing new dataspaces. Thereby, it optimizes:
* on ARMv7 using per-word assignments
* on ARMv8 using cacheline zeroing
* on x86_64 using 'rept stosq' assembler instruction
Fix#3685
This depot package is the runtime for rtc_drv and system_rtc server with
coordinated update of system and hardware RTC via reports. It replaces
drivers_rtc which was never freestanding drivers package and, therefore,
has to be accompanied with a running platform_drv etc.
Fixes#3680
- use two disks with two partitions each
- one disk uses MBR and one GPT
- connect one 'part_block' to each disk
- run 'block_tester' and 'test-block-client' connecting to one partiion
on each disk
- x86 uses 'ahci_drv' as disk back-end, Linux 'lx_block'
- on autopilot enable for Qemu only
issue #3671
Formerly, _next was always updated to the last free'd allocation, which
left large gaps on alloc/free bursts. Now, we try keep allocation
density high from the start of the array and ensure that allocations
happen at the lowest available index in the bit array.
Fixes#3679
The function is exposed in libc headers provided by Genode, the code for
the function is being compiled and actually works, but the symbol is
missing from the symbols file resulting in linking failures. Add it to
the libc symbols file.
Fixes#3676Fixes#3677
This patch increases the Sculpt version to avoid the risk of binary
compatibility clashes with index files published for the just-released
Sculpt 20.02 version.
Without this patch, the text area would mistakenly delete the characters
types after placing the cursor with the mouse as soon as shift is used
as modifier.
- Don't reset sculpt partition on USB changes whenever ram fs is
selected for use.
- Reconfigure runtime immediately when plugging USB devices.
This prevents the usb-block driver and part_block from
dangling in the graph. They disappear immediately now.
This patch restores the anchoring of the popup menu to the original
behavior, placing the popup to the right of the '+' button. The
anchoring got lost due to the recent structural changes of the GUI,
which affect the hover XML structure.
The new implementation relieves the main entrypoint from monitor jobs
for contended lock primitives and is based on custom applicant data
structures, per-lock resp. per-semaphore applicant lists, and a
libc-internal blockade with timeouts based on libc kernel primitives.
This patch makes the 'with_raw_node' method more useful in situations
where content of an Xml_node is fed into an Xml_generator, i.e., the
rules report/rom mechanism of the window layouter.
