- 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
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.
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.
- 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
- Remove dated 'Block::Driver' front end and implement
'Block::Request_stream' front end
- Remove all dynamic memory allocations but DMA memory allocations
- Remove 'Platform_hba' and implement platform specific functions in
'spec/<platform>/*'
- Ata and Atapi don't inherit from 'Port' any more, but are a member of
'Port' as a protocol implementation
- Use platform driver for MMIO mappings (x86)
- Exchange stateful initialization of Ata/Atapi with a sequential
initialization using 'wait_for_any' and 'retry' patterns
- Fix Atapi initialization by setting the byte count limit
- Set FIS receive base only when Cmd::FRE is disabled and Cmd::FR is 0
- Put everything in namespaces ('Ahci', 'Ata', or 'Atapi')
- Ata decides during read/write operations to use native-command queuing
or normal DMA requests
- Remove port claiming logic (is now done via 'Constructibles')
fixes#3636
This patch extracts the child-management functionality from the init
component into a new library called "sandbox". The library API is
located at 'os/include/os/sandbox.h'.
The sandbox API allows for the interaction of the component with the
sandboxed children by providing locally implemented services. This
mechanism is illustrated by the new test at os/src/test/sandbox.
Issue #3601
When re-configuring a domain from using a static IP config to using a dynamic
IP config, attached interfaces didn't react with doing DHCP discover. This bug
is fixed now and the nic_router_flood test was adapted in a way that it tests
the described behavior.
Fix#3518
The rtc_drv on x86 can now by used to also set the RTC. If the config
attribute 'allow_setting_rtc' is set to 'yes' the driver will update
the RTC from the content of the 'set-rtc' ROM module. A valid ROM must
contain a top node with the following attributes: 'year', 'month',
'day', 'hour', 'minute' and 'second'.
* Only rudimentary checking of the provided values is done.
* '12H' mode is not supported.
Fixes#3438.
On some platforms (foc+pbxa9, hw+imx53_qsb_tz, hw+rpi) the default buffer size
is to much for the RAM available on the board. Thus, decrease the buffer size
and therefore the number of polls for these platforms only.
Fixes#3354
Moreover, express current requirement to hw API, which was missing until now.
Originally, in the board-specific build directory of imx53_qsb the
"KERNEL" variable was preset with "hw". Therefore, it was not perceived
that this dependency is required.
Ref #3316
* Make package buildable for ARM too
* Move usb library to src targets for explicitly named targets
* adapt remaining run-scripts to use the correctly named usb drivers
Ref #2190
* Make target binaries independent of board SPECS
* Name binaries of one architecture unambigously
* Extend include path to match board specifics
* Adapt run-scripts to use the right binary
Ref #2190
Ref #3180
- Added 'io_buffer' attribute, default is 4M
- Added 'batch' attribute, specifying the number of jobs used
in parallel, default is 1 (sequential)
- Removed 'synchronous' attribute (use batch of 1 instead)
- Added 'copy' attribute (default "yes")
- Print number of signals ("triggered")
Issue #3283
Track the dataspaces used by attach and add handling of flushing VM space
when dataspace gets destroyed (not triggered via the vm_session interface).
Issue #3111
Issue #3111
- enable vt-x in kernel configuration
Kernel patches:
- add unrestricted guest support
- avoid kernel boot failure when vt-x is not available
- avoid nullpointer in kernel when vcpu is not fully setup
- avoid vcpu scheduling bug which causes starvation on same/below prio level
- save efer register correctly from guest
The zynq nic_drv also depends on hw, we therefore adapted the folder
structure for clarity. Also renamed the binary to 'zynq_nic_drv' to
prevent conflicts and to allow removing the cadence_gem spec.
Issue #3179
Since the timer and timeout handling is part of the base library (the
dynamic linker), it belongs to the base repository.
Besides moving the timer and its related infrastructure (alarm, timeout
libs, tests) to the base repository, this patch also moves the timer
from the 'drivers' subdirectory directly to 'src' and disamibuates the
timer's build locations for the various kernels. Otherwise the different
timer implementations could interfere with each other when using one
build directory with multiple kernels.
Note that this patch changes the include paths for the former os/timer,
os/alarm.h, os/duration.h, and os/timed_semaphore.h to base/.
Issue #3101
If the NIC router has insufficient CAP or RAM quota for the creation of
a state object for an interface, it tries to destroy a certain amount of
existing state objects of this interface to free resources. Afterwards,
it retries handling the current packet once. If it does fail again, the
router drops the packet.
Issue #2953
Ensure that a '\0' always appears at the end of the ouput of the XML
generator. Previously, exceptions during the Node(...) constructor
might have prevented this. This commit also extends the xml_generator
test to drive a harder test on exceptions in the Xml_generator.
Issue #2953
When the functor provided to the Node constructor throws an exception,
do revert all changes in reverse order. Previously, the changes made
to the parent node were not considered by the exception handler which
caused unnecessary characters to remain in the out buffer for each
reverted node.
Issue #2953
The PS/2 driver retries to get mouse-reset results for 700 ms, sleeping
after each attempt for 10 ms. So, the driver needs a Timer session now.
Fixes#2713
The ram plugin to the VFS must send notifications to watch handles on
files when they are removed. This brings the VFS server to parity with
the ram_fs server for the sake of notifications.
Move run/fs_report to gems and use the vfs server and the vfs init
plugin in the test.
Ref #2902
This patch removes the detection of statically linked executables from
the base framework. It thereby fixes the corner cases encountered with
Sculpt when obtaining the binaries of the runtime from the depot_rom
service that is hosted within the runtime.
Statically linked binaries and hybrid Linux/Genode (lx_hybrid) binaries
can still be started by relabeling the ROM-session route of "ld.lib.so"
to the binary name, pretending that the binary is the dynamic linker.
This can be achieved via init's label rewriting mechanism:
<route>
<service name="ROM" unscoped_label="ld.lib.so">
<parent label="test-platform"/> </service>
</route>
However, as this is quite cryptic and would need to be applied for all
lx_hybrid components, the patch adds a shortcut to init's configuration.
One can simply add the 'ld="no"' attribute to the <start> node of the
corresponding component:
<start name="test-platform" ld="no"/>
Fixes#2866
Process I/O packets in batches. If a batch is processed and there are
still packets pending, send a signal locally to the packet handler and
return to the entrypoint signal dispatcher. This prevents clients from
starving each other, which happens when a client continuously submits
packets at a faster rate than the server can process.
Fix#2900
Test of File_system I/O scheduling. It appears that a client not
submitting I/O packets slower than the VFS server can process will
starve other clients.
Ref #2900
This component is contrasted with the fs_rom server that serves
independent dataspaces to each client. Using a cache was not possible
until the region map session supported the creation of read-only
attachments.
Test at run/read_only_rom.
Ref #1633Fix#2760
The test triggers the following assertion in the kernel, which will
remain unfixed as the kernel is no longer developed:
Assertion queue_state.is_set(queue_state_t::send) failed in file
.../pistachio/kernel/src/api/v4/tcb.h, line 727
This assertion is presumably similar / related to issue #1495.
The legacy PIT timer driver is too jittery for the character-repeat
test. This patch disables the test for platforms using this device as
user-level time source, which are at the moment OKL4, Pistachio, and
seL4.
Also remove 'requires_installation_of', while also checking sbin
directories in 'have_installed'. The run scripts have been adjusted
accordingly.
Fixes#2853
The 200-second timeout of the init test triggers on some test platform
(in particular on qemu) while it is overly pessimistic on others.
This patch aims at stabilizing the timing behavior of the test across
the tested kernels/machines.
This is a follow-up patch of "init: avoid too eager child restart". On
each config update of init, init re-applies child-specific configuration
changes. In the case of an already exited child, this re-evaluation
wrongly marked such a child as abandoned because the child's environment
sessions do no longer exist. Abandoning the child, in turn, triggers the
destruction and subseqent restart (because the <start> node of the
configuration still exists). The latter is bad for two reasons.
First, the exit state of the original instance becomes lost. Second, the
restart may have unexpected side effects due to sessions created by the
new instance. I.e., when resizing a partition in sculpt, init would
wrongly restart the gpt-write tool after the tool successfully exited.
This collides with a newly started instance of part_blk/resize2fs, which
now competes with the second gpt-write instance for the exclusive access
of the targeted block device.
The patch prevents init from re-applying configurations to exited
children. The accompanied test case covers the corner case.
Do not simply wait for the good ping test to finish, but for the other
flood tests to trigger the RAM exhaustion. This makes the test more
robust with slower platforms or a different timing.
Issue #2857
Introduce the uplink tag:
! <config>
! <uplink label="wifi" domain="uplink">
! <uplink label="wired" domain="wired_bridge">
! <uplink domain="wired_bridge">
! <config/>
For each uplink tag, the NIC router requests a NIC session with the
corresponding label or an empty label if there is no label attribute.
These NIC sessions get attached to the domain that is set in their
uplink tag as soon as the domain appears. This means their lifetime is
not bound to the domain. Uplink NIC sessions can be safely moved from
one domain to another without being closed by reconfiguring the
corresponding domain attribute.
Attention: This may render previously valid NIC router configurations
useless. A domain named "uplink" doesn't automatically request a NIC
session anymore. To fix these configurations, just add
! <uplink domain="uplink"/>
or
! <uplink label="[LABEL]" domain="uplink"/>
as direct subtag of the <config> tag.
Issue #2840
Currently has three clients that continuously create new UDP/TCP/ICMP
connections through the NIC router with NAT to the outer world and they
get never closed. A fourth client does normal ping through the same
domain to the outer world that must succeed even after the RAM quota of
the other session at the router is exhausted. The test is restricted to
Qemu to not being at risk to flood real networks.
Issue #2857
This test monitors the RAM quota of a dynamic init and a server hosted
within the dynamic init in the presence of a repeatedly created and
destructed client.
On Linux, we have a tap device as NIC back end but there is no one to
ping to in the subnet of the tap device. On FOC, the tests seem to trigger
a bug in the destruction of components with parent.exit(X); .
Fixes#2848
Clients may wish to act on missing files. In any case the fs_rom
needs to reopen a watch handle when a file is deleted, and this
sort of change to the internal state machine is propelled by
client RPC requests.
Fix#2839
This tests the case that behind a NIC router domain there are not only
directly connected clients of the subnet but also Layer 2 indirections like
a switch or a bridge which might be interesting for example for testing the
NIC routers DHCP server implementation.
Fixes#2837
When an environment session is provided by a async service such as a
sibling component, the session metadata must be preserved until end of
the lifetime of the session at the server has been acknowledged by the
server. Since the session meta data of env sessions are always part of
the 'Child' object, the destruction of this object must be deferred
until this point.
On real hardware, the tests expect an IPv4 subnet such that UDP requests
to 10.0.0.2 port 12345 get answered with an ICMP destination port unreachable.
Issue #2775
support USB NIC
Issue #2788
Set DHCP discover timeout to 1 second because, for some reason, the first
DHCP discover attempt of the NIC router on the PandaBoard times out with the
nightly test infrastructure.
Issue #2788
Adaption to mac-address allocation changes
Instead of handing over the maximum available size to the packet data
accessors, hand over a size guard that keeps track of the packets
boundaries.
This commit also moves the size-guard utilitiy header of Ping and NIC
Router to the include/net directory making it a part of the net library.
It applies the new approach to all net-lib users in the basic repositories.
Ping looses its configurability regarding the ICMP data size as this would
require an additional method in the size guard which would be used only by
Ping.
The size guard was also re-worked to fit the fact that a packet can
bring a tail as well as a header (Ethernet).
Issue #2788
This commit changes the 'Input::Event' type to be more safe and to
deliver symbolic character information along with press events.
Issue #2761Fixes#2786
The new 'attribute' and 'value' attributes of input nodes
can be used to select input sub nodes that match the presence and value
of the specified attribute.
Issue #2691
This driver component provides support for using consumer NVMe storage
devices, i.e. it omits name space managment and will always use the
first name space, on Genode. For now it defaults to a reasonable low
configuration:
- 1 I/O queue (completion/submission tuple)
- 128 entries in the I/O queue
- 4096 as the only I/O transaction memory page size
Fixes#2747.
This tests ping with simple IP forwarding, ping with NAPT as well as
forwarding of ICMP "Destination Unreachable" messages through the NIC
router.
Issue #2732
The 'ping' component continuously sends ICMP Echo requests to a given IP host
and waits for the corresponding ICMP Echo replies. For each successfull ICMP
Echo handshake it prints a short statistic. By now, it can be used only with a
static IP configuration. The size of the ICMP data field can be configured. It
gets filled with the letters of the alphabet ('a' to 'z') repeatedly.
Issue #2732
Add a new 'Vfs_watch_handle' type to the VFS interface. This handle type
will pass a handle context up through the I/O handler to the application
when a notification event occurs.
Watch support implemented for RAM and File_system plugins, all other
file-systems return WATCH_ERR_STATIC by default.
Test at run/fs_rom_update_ram and run/fs_rom_update_fs.
Fix#1934
This patch improves the `Text_painter` utility that is commonly used by
native Genode components to render text:
- Support for subpixel positioning
- Generic interface for accessing font data
- Basic UTF-8 support
Since the change decouples the font format from the 'Text_painter' and
changes the API to use the sub-pixel accurate 'Text_painter::Position'
type, all users of the utility require an adaptation.
Fixes#2716
File_system clients may now watch files and directories for changes by
opening a 'Watch_handle' rather than submitting a 'CONTENT_CHANGED'
packet to the server. When a change happens at a node with an open
Watch_handle a CONTENT_CHANGED packet will be sent from the server to
the client. This serializes registration with other handle operations
and separates I/O handle state from notification handle state.
Test at run/fs_rom_update.
Ref #1934
The 'trace_logger' component can be used to easily gather, process and export
different types of tracing data. Which subjects to select is configurable via
session label policies and thread names. Which data to collect from the
selected subjects can be configured for each subject individually, for groups
of subjects, or for all subjects. The gathered data can be exported as log
output.
This is an example configuration of the 'trace_logger' component which shows
the default value for each attribute except the policy.thread and
policy.label:
! <config verbose="no"
! session_ram="10M"
! session_arg_buffer="4K"
! session_parent_levels="0"
! period_sec="5"
! activity="no"
! affinity="no"
! default_policy="null"
! default_buffer="4K">
!
! <policy label="init -> timer" />
! <policy label_suffix=" -> ram_fs" />
! <policy label_prefix="init -> encryption -> "
! thread="worker"
! buffer="4K"
! policy="null" />
! </config>
For more details see os/src/app/trace_logger/README.
Fixes#2654
On x86 64 bit with SeL4, the test needs around 80MB that must be
completely composed of 4KB-pages due to current limitations of the SeL4
port. Thus, Core must flush the page table caches pretty often during
the test which is an expensive high-prior operation and makes it
impossible to provide a highly precise time.
This commit replaces the old xray_trigger component by a new component
called global_keys_handler. For details, please refer to the issue text
and the accompanied README file.
Fixes#2554
The timer driver should always be of the highest priority to avoid
problem with timers that have low max-counter values like the PIT
with only 53 ms.
Ref #2400
Add a "writeable" policy option to the ahci_drv and part_blk Block
servers and default from writeable to ready-only. Should a policy
permit write acesss the session request argument "writeable" may still
downgrade a session to ready-only.
Fix#2469
Currently, init does not test wether a service is abandoned on a new
configuration if the service was routed via an any-child route. Trigger
this behaviour in the init test.
Ref #2483
The new version of the test exercises the combination of fs_report with
ram_fs and fs_rom as a more flexible alternative to report_rom.
It covers two corner cases that remained unaddressed by fs_rom and
ram_fs so far: First, the late installation of a ROM-update signal
handler at fs_rom right before the content of the file is modified.
Second, the case where the requested file is not present on the file
system at the creation time of the ROM session. Here, the ram_fs missed
to inform listeners for the compound directory about the later created
file.
On platforms that use the PIT timer driver, 'elapsed_ms' is pretty
inprecise/unsteady (up to 3 ms deviation) for a reason that is not
clearly determined yet. On Fiasco and Fiasco.OC, that use kernel timing,
it is the same. So, on these platforms, our locally interpolated time
seems to be fine but the reference time is bad. Until this is fixed, we
raise the error tolerance for these platforms in the run script.
Ref #2400
Appending a suffix to report filenames was behavior inherited from
fs_log, it prevents creating files where directories need to be created
later. But unlike logs, only a subset of the hierarchy will report and
those that do append a component-local label, so the risk of collision
is low.
By removing the suffix fs_rom can serve reports back as ROM just as
report_rom does.
Ref #2422
This patch increases init's preserved RAM and capability quota to
account for a current limitation of init with respect to the creation of
sessions to parent services:
In contrast to regular routed services, sessions to parent services are
created via 'Env::session'. The implementation of 'Env::session'
automatically upgrades session quotas on demand, which is the desired
behavior for regular 'Connection' objects. However, for sessions
established on the behalf of init's children, we would need to reflect
the error condition to the child instead of resolving it locally within
init (by subsidizing the session with init's quota). This patch leaves
this issue unresolved but fixes the symptom for the bomb test. It is
meant as an interim solution until the handling of parent sessions is
revised.
On QEMU, NOVA uses the pretty unstable TSC emulation as primary time
source. Thus, timeouts do not trigger with the common precision (< 50
ms). Use an error tolerance of 200 ms for this platform constellation.
Ref #2400
On platforms were we do not have local time interpolation we can simply
skip the first test stage in the timeout test. This way, we can at least
test the rest.
Fixes#2435
Previously, the Genode::Timer::curr_time always used the
Timer_session::elapsed_ms RPC as back end. Now, Genode::Timer reads
this remote time only in a periodic fashion independently from the calls
to Genode::Timer::curr_time. If now one calls Genode::Timer::curr_time,
the function takes the last read remote time value and adapts it using
the timestamp difference since the remote-time read. The conversion
factor from timestamps to time is estimated on every remote-time read
using the last read remote-time value and the timestamp difference since
the last remote time read.
This commit also re-works the timeout test. The test now has two stages.
In the first stage, it tests fast polling of the
Genode::Timer::curr_time. This stage checks the error between locally
interpolated and timer-driver time as well as wether the locally
interpolated time is monotone and sufficiently homogeneous. In the
second stage several periodic and one-shot timeouts are scheduled at
once. This stage checks if the timeouts trigger sufficiently precise.
This commit adds the new Kernel::time syscall to base-hw. The syscall is
solely used by the Genode::Timer on base-hw as substitute for the
timestamp. This is because on ARM, the timestamp function uses the ARM
performance counter that stops counting when the WFI (wait for
interrupt) instruction is active. This instruction, however is used by
the base-hw idle contexts that get active when no user thread needs to
be scheduled. Thus, the ARM performance counter is not a good choice for
time interpolation and we use the kernel internal time instead.
With this commit, the timeout library becomes a basic library. That means
that it is linked against the LDSO which then provides it to the program it
serves. Furthermore, you can't use the timeout library anymore without the
LDSO because through the kernel-dependent LDSO make-files we can achieve a
kernel-dependent timeout implementation.
This commit introduces a structured Duration type that shall successively
replace the use of Microseconds, Milliseconds, and integer types for duration
values.
Open issues:
* The timeout test fails on Raspberry PI because of precision errors in the
first stage. However, this does not render the framework unusable in general
on the RPI but merely is an issue when speaking of microseconds precision.
* If we run on ARM with another Kernel than HW the timestamp speed may
continuously vary from almost 0 up to CPU speed. The Timer, however,
only uses interpolation if the timestamp speed remained stable (12.5%
tolerance) for at least 3 observation periods. Currently, one period is
100ms, so its 300ms. As long as this is not the case,
Timer_session::elapsed_ms is called instead.
Anyway, it might happen that the CPU load was stable for some time so
interpolation becomes active and now the timestamp speed drops. In the
worst case, we would now have 100ms of slowed down time. The bad thing
about it would be, that this also affects the timeout of the period.
Thus, it might "freeze" the local time for more than 100ms.
On the other hand, if the timestamp speed suddenly raises after some
stable time, interpolated time can get too fast. This would shorten the
period but nonetheless may result in drifting away into the far future.
Now we would have the problem that we can't deliver the real time
anymore until it has caught up because the output of Timer::curr_time
shall be monotone. So, effectively local time might "freeze" again for
more than 100ms.
It would be a solution to not use the Trace::timestamp on ARM w/o HW but
a function whose return value causes the Timer to never use
interpolation because of its stability policy.
Fixes#2400
Removes the following Fiasco.OC specific features:
* GDB extensions for Fiasco.OC
* i.MX53 support for Fiasco.OC
* Kernel debugger terminal driver
* Obsolete interface Native_pd
* Obsolete function of interface Native_cpu
This patch reduces the number of exception types by facilitating
globally defined exceptions for common usage patterns shared by most
services. In particular, RPC functions that demand a session-resource
upgrade not longer reflect this condition via a session-specific
exception but via the 'Out_of_ram' or 'Out_of_caps' types.
Furthermore, the 'Parent::Service_denied', 'Parent::Unavailable',
'Root::Invalid_args', 'Root::Unavailable', 'Service::Invalid_args',
'Service::Unavailable', and 'Local_service::Factory::Denied' types have
been replaced by the single 'Service_denied' exception type defined in
'session/session.h'.
This consolidation eases the error handling (there are fewer exceptions
to handle), alleviates the need to convert exceptions along the
session-creation call chain, and avoids possible aliasing problems
(catching the wrong type with the same name but living in a different
scope).
This patch mirrors the accounting and trading scheme that Genode employs
for physical memory to the accounting of capability allocations.
Capability quotas must now be explicitly assigned to subsystems by
specifying a 'caps=<amount>' attribute to init's start nodes.
Analogously to RAM quotas, cap quotas can be traded between clients and
servers as part of the session protocol. The capability budget of each
component is maintained by the component's corresponding PD session at
core.
At the current stage, the accounting is applied to RPC capabilities,
signal-context capabilities, and dataspace capabilities. Capabilities
that are dynamically allocated via core's CPU and TRACE service are not
yet covered. Also, the capabilities allocated by resource multiplexers
outside of core (like nitpicker) must be accounted by the respective
servers, which is not covered yet.
If a component runs out of capabilities, core's PD service prints a
warning to the log. To observe the consumption of capabilities per
component in detail, the PD service is equipped with a diagnostic
mode, which can be enabled via the 'diag' attribute in the target
node of init's routing rules. E.g., the following route enables the
diagnostic mode for the PD session of the "timer" component:
<default-route>
<service name="PD" unscoped_label="timer">
<parent diag="yes"/>
</service>
...
</default-route>
For subsystems based on a sub-init instance, init can be configured
to report the capability-quota information of its subsystems by
adding the attribute 'child_caps="yes"' to init's '<report>'
config node. Init's own capability quota can be reported by adding
the attribute 'init_caps="yes"'.
Fixes#2398
For asynchronously provided sessions, the parent has to maintain the
session state as long as the server hasn't explicitly responded to a
close request. For this reason, the lifetime of such session states is
bound to the server, not the client.
When the server responds to a close request, the session state gets
freed. The 'session_response' implementation does not immediately
destroy the session state but delegates the destruction to a client-side
callback, which thereby also notifies the client. However, the code did
not consider the case where the client has completely vanished at
session-response time. In this case, we need to drop the session state
immediately.
Fixes#2391
