If logical volumes are active and recognised by udev, no longer ignore
them. We also do some extra handling to use user-friendly device names
and mount the /dev/mapper/foo device rather than the /dev/dm-xx device.
Additionally if we see "LMV2_member" devices start a rescan in case
LVM-formatted disks came up after the LVM initscript.
Signed-off-by: Samuel Mendoza-Jonas <sam@mendozajonas.com>
discover/devmapper: Add prefix to devmapper device names
Add a 'pb-' prefix to all device mapper devices created by Petitboot.
Beyond helping to identify Petitboot-related devices, this avoids naming
collisions if we create snapshots of LVM logical volumes which also
exist in /dev/mapper.
Signed-off-by: Samuel Mendoza-Jonas <sam@mendozajonas.com>
The nc-subset screen can exceed its maximum height if some options are
long enough to wrap around to two lines. Increaes the maximum size of
the pad to account for every line potentially wrapping once.
Signed-off-by: Samuel Mendoza-Jonas <sam@mendozajonas.com>
Load tasks that start before the network is available will fail. Rather
than just fail these tasks, add them to a queue that is processed once
the network is ready. This helps users who try to request files early in
setup, as well as very early running load tasks.
Signed-off-by: Samuel Mendoza-Jonas <sam@mendozajonas.com>
Currently over reinit events the system info is not affected. However
network and block device information can change over reinit, so clear
this information.
Signed-off-by: Samuel Mendoza-Jonas <sam@mendozajonas.com>
discover/platform-powerpc: Handle optional Get Device ID section
The 'auxiliary' section of the 'Get Device ID' response is optional,
and some platforms exclude it from the response entirely. However
Petitboot only recognises the response as valid if it includes the full
16 bytes.
Update get_ipmi_bmc_versions() to also handle responses of only 12 bytes.
Signed-off-by: Samuel Mendoza-Jonas <sam@mendozajonas.com>
discover/platform-powerpc: Correct aux revision format
The Auxiliary Firmware Revision Information should be displayed as four
hexadecimal bytes if a manufacturer-specific format is not known. Update
the "Firmware version" format to reflect this.
Signed-off-by: Samuel Mendoza-Jonas <sam@mendozajonas.com>
discover/device-handler: Include makedev() from sysmacros.h
Include sys/sysmacros.h explicitly in response to the following error
message:
../discover/device-handler.c:1001:13: warning: In the GNU C Library, "makedev" is defined
by <sys/sysmacros.h>. For historical compatibility, it is
currently defined by <sys/types.h> as well, but we plan to
remove this soon. To use "makedev", include <sys/sysmacros.h>
directly. If you did not intend to use a system-defined macro
"makedev", you should undefine it after including <sys/types.h>.
id = makedev(1, handler->n_ramdisks);
^~~~~~~~~~~~~~~~~~~~~~~~~~
Signed-off-by: Samuel Mendoza-Jonas <sam@mendozajonas.com>
discover: Track both configured and current autoboot settings
If autoboot is enabled but later disabled or cancelled by, for example,
an IPMI override then the nc-config screen will set the autoboot widget
as disabled. If the user then makes and saves a change in nc-config,
autoboot will also be saved as disabled. This accidental change is
particularly awkward if the user is attempting to remove an IPMI
override.
Instead only ever change the autoboot setting if the user explicitly
changes it. Use a new helper function 'config_autoboot_active()' to
determine the current autoboot status where needed.
Signed-off-by: Samuel Mendoza-Jonas <sam@mendozajonas.com>
If an asynchronous job is running over a reinit, the process can return
and run its callback function after the reinit. This becomes a problem
if the callback function accesses pointers that were only valid before
the reinit (eg. device structs).
If a reinit is requested explicitly stop all active asynchronous jobs
and clear their callback functions before the reinit.
Signed-off-by: Samuel Mendoza-Jonas <sam@mendozajonas.com>
discover/device-handler: Cancel pending boot on reinit
When a reinit is requested device_handler_cancel_default() is
called, however as the name suggests this only cancels the boot task if
it is the result of a default boot option. We also want to cancel a boot
task if it was executed manually because it may have outstanding
asynchronous transfers running, so explicitly cancel it during reinit.
Signed-off-by: Samuel Mendoza-Jonas <sam@mendozajonas.com>
discover/device-handler: Process queue after device added
In device_handler_discover() we process the unresolved boot options
queue first. However the discover_device in question has not yet been
added to handler->devices so when a parser tries to search for a
matching device it will fail.
The discover_device will be added to the handler if it has not already
in device_handler_discover_context_commit() so move the call to
process_boot_option_queue() after it.
Signed-off-by: Samuel Mendoza-Jonas <sam@mendozajonas.com>
discover/network: Ignore interfaces with pre-existing MAC address
Petitboot uses the MAC address of network interfaces as a unique
identifier. This can cause a crash in pb-discover on a machine that has
multiple interfaces with the same MAC address.
While duplicate MAC addresses are rare and imply an issue with the
larger system configuration Petitboot should handle this gracefully, so
log a warning and ignore any interfaces other than the first to appear
that share a MAC address.
Signed-off-by: Samuel Mendoza-Jonas <sam@mendozajonas.com>
discover/network: Ensure interfaces have device before configuring
Reorganise network_handle_nlmsg() slightly to create interface->dev just
before calling configure_interface() rather than only for brand new
interfaces. This ensures existing interfaces which have had ->dev
removed but receive a new configure event do not access a NULL pointer
during the configuration process.
Signed-off-by: Samuel Mendoza-Jonas <sam@mendozajonas.com>
discover/network: Search by UUID only if available
When registering a new discover device it is possible the device does
not have an associated UUID, for example when created via
device_handler_process_url(). Fall back to find_interface_by_name() in
this case.
Signed-off-by: Samuel Mendoza-Jonas <sam@mendozajonas.com>
The "Connected to pb-discover!" message is more useful for development
than actual use; for users the more important messages are related to
device and configuration parsing. Drop the message to slightly reduce
the level of noise on start up.
Signed-off-by: Samuel Mendoza-Jonas <sam@mendozajonas.com>
Add status updates to a persistent list in the discover_server struct,
and send each client the backlog on connect. This avoids clients missing
useful messages from early init. Clients will only show this in the
backlog screen to avoid flooding the client's status line.
Signed-off-by: Samuel Mendoza-Jonas <sam@mendozajonas.com>
discover/paths: Parse Busybox progress information
Several busybox utilities (tftp and wget in particular) use a common
format for progress bar output. Add a stdout callback that recognises
this format and passes progress information to
device_handler_status_download().
If Petitboot has been explicitly built with busybox support set
busybox_progress_cb() as the default stdout callback for
load_url_async().
Signed-off-by: Samuel Mendoza-Jonas <sam@mendozajonas.com>
Several processes run by Petitboot output progress information while
running. Add device_handler_status_download() which process callers can
call to register and update progress information (percentage and current
size).
A list of 'progress_info' structs holds this progress information, and
on each call to device_handler_status_download() the information is
combined and displayed as a single status update for readability.
On completion device_handler_status_download_remove() is called to
remove old progress information from the list.
Signed-off-by: Samuel Mendoza-Jonas <sam@mendozajonas.com>
lib/process: Allow process output to be retrieved on each event
Allow a custom callback function to be set when registering the IO
waiter for asynchronous processes.
To allow output from processes to be parsed as it is received, add
process_stdout_custom() which passes a new "line" parameter to
process_read_stdout_once() in order to consume output as it appears.
Users of a custom IO callback will only have access to the process_info
struct which is internal to lib/process; the function
procinfo_get_process() is added to allow these callers to access process
information.
Signed-off-by: Samuel Mendoza-Jonas <sam@mendozajonas.com>
Update kexec_load() to preserve output from the call to `kexec -l`. On
error retrieve the resulting error message and update the status line
with it to provide a more informative error message.
Signed-off-by: Samuel Mendoza-Jonas <sam@mendozajonas.com>
Jeremy Kerr [Mon, 12 Dec 2016 09:34:42 +0000 (17:34 +0800)]
discover/status: remove completion messages
The completion messages are unconditional, so don't really indicate
anything. In fact, the dhcp completion status is misleading, as we may
still be processing the context through pxe callbacks.
Signed-off-by: Jeremy Kerr <jk@ozlabs.org> Signed-off-by: Samuel Mendoza-Jonas <sam@mendozajonas.com>
Jeremy Kerr [Mon, 12 Dec 2016 11:58:32 +0000 (19:58 +0800)]
discover: add handler reference to struct discover_context
Since the device handler provides the status message functions, we need
a pointer to it for device discovery (which we use a struct
discover_context for).
This change adds a 'handler' member to struct discover_context, to allow
status reporting. Since we now have a handler, there's no need for the
network pointer, so provide an accessor function instead.
Signed-off-by: Jeremy Kerr <jk@ozlabs.org> Signed-off-by: Samuel Mendoza-Jonas <sam@mendozajonas.com>
Jeremy Kerr [Mon, 12 Dec 2016 07:43:21 +0000 (15:43 +0800)]
discover: Add device-specific status reporting functions
Most of our status reporting is against a specific device, so add
status reporting functions that take a struct discover_device and use a
stnadard prefix.
Signed-off-by: Jeremy Kerr <jk@ozlabs.org> Signed-off-by: Samuel Mendoza-Jonas <sam@mendozajonas.com>
Jeremy Kerr [Thu, 8 Dec 2016 02:08:15 +0000 (13:08 +1100)]
discover: Add helpers for status reporting
This change adds a couple of helpers for the status reporting API,
allowing callers to provide just a set of printf-style arguments, rather
than having to build up a struct status.
Signed-off-by: Jeremy Kerr <jk@ozlabs.org> Signed-off-by: Samuel Mendoza-Jonas <sam@mendozajonas.com>
Jeremy Kerr [Wed, 7 Dec 2016 00:43:08 +0000 (11:43 +1100)]
discover: separate status-reporting function from boot() callback
Currently, the device_discover_boot_status function is both used for
internal status updates, as well as the callback passed to boot().
This change splits this into two functions; one for the latter and one
for the former. The latter just has a void * for its first argument, to
match the boot_status_fn type.
Signed-off-by: Jeremy Kerr <jk@ozlabs.org> Signed-off-by: Samuel Mendoza-Jonas <sam@mendozajonas.com>
Jeremy Kerr [Tue, 6 Dec 2016 06:23:28 +0000 (17:23 +1100)]
ui/ncurses: Add status log UI
Currently, status messages from the server are displayed in a single
line at the bottom of the main menu UI, and are lost once a new status
is reported.
This change adds a facility for the UI to collect and display the status
messages from the server, in a dedicated UI screen. This allows a user
to look back through the discovery & boot process.
Signed-off-by: Jeremy Kerr <jk@ozlabs.org> Signed-off-by: Samuel Mendoza-Jonas <sam@mendozajonas.com>
ui/ncurses: Make server connect message more clear
The current message mentions a "server" which can give the misleading
impression that the UI is waiting for a remote network server. The delay
is actually in waiting for the pb-discover process to be ready, so
update the message to reflect that.
Signed-off-by: Samuel Mendoza-Jonas <sam@mendozajonas.com>
discover/platform-powerpc: Reject bootdevs with empty UUIDs
If a "uuid:" label is set in the petitboot,bootdevs parameter without a
matching UUID, the UUID is unintentionally accepted and set to NULL.
This can cause a segfault in nc-config when device UUIDs are compared
against the autoboot option. Instead treat options like this as
malformed.
Signed-off-by: Samuel Mendoza-Jonas <sam@mendozajonas.com>
The "petitboot,bootdevs" parameter has been around long enough now that
there shouldn't be anyone still transitioning over from the old
"petitboot,bootdev" parameter. Drop this parameter to simplify the
populate_bootdev_config() logic.
Signed-off-by: Samuel Mendoza-Jonas <sam@mendozajonas.com>
Support HTTP(S) proxies when downloading resources
Allow the user to specify a HTTP and HTTPS proxy server. The discover
server will set the http_proxy and https_proxy environment variables,
enabling the proxy servers for any further HTTP(S) requests.
Signed-off-by: Samuel Mendoza-Jonas <sam@mendozajonas.com>
ui/ncurses: Spin child to ensure autoboot cancelled on exit
If the client is not connected to the server instance when exiting, fork
and have the child process spin until the server is available and can be
told to cancel autoboot. This prevents the scenario of a user exiting
the UI and having the server continue to autoboot while they are using
the command line.
Signed-off-by: Samuel Mendoza-Jonas <sam@mendozajonas.com>
Commit 2163af5 "discover/pxe-parser: Retrieve configs asynchronously"
added asynchronous loading of remote pxe filenames, but made an
unintended change in behaviour to the PXE parser. Previously the parser
would try a list of possible filenames, and parse the first one it
found. However the above commit spawns an asynchronous job for every
filename, and parses any that can be retrieved. It is a common
configuration to have a machine-specific config and a 'fallback' default
config, and the change means we could erroneously retrieve and parse
both configs.
Update the PXE parser so that asynchronous jobs are spawned
sequentially. That is, spawn a job for the first filename and if not
successful spawn another job for the next filename, and so on. Once a
remote config is successfully retrieved, parse it and stop.
Signed-off-by: Samuel Mendoza-Jonas <sam@mendozajonas.com>
Commit ce54f86 "Add petitboot,tty and track available consoles" added
the petitboot,tty parameter, but the petitboot,console parameter is also
recognised by Petitboot. These are ultimately handled by the 30-add-offb
and 80-set-stdout hooks respectively, but exist for mostly the same
purpose.
We consolidate these down to just the original petitboot,console
parameter. If the contents of petitboot,console have been configured by
Petitboot (ie. it is of the form /dev/dev# [ Description ]) we behave as
normal, otherwise we assume that petitboot,console contains a full
OF path to the intended console device and do not allow it to be
modified. This follows petitboot,console's original intent to be a debug
aid, and takes precedence over any other use.
The 80-set-stdout hook is removed as 30-add-offb now accounts for both
use cases.
Signed-off-by: Samuel Mendoza-Jonas <sam@mendozajonas.com>
ui/ncurses: Display warning when net override active
If the current interface config has been set by an IPMI network
override, display a warning in nc-config that saving the current config
will overwrite any saved interface config.
Signed-off-by: Samuel Mendoza-Jonas <sam@mendozajonas.com>
discover: Avoid writing network overrides to NVRAM
Explicitly keep track of whether the current interface config was set by
an IPMI network override, and avoid overwriting any saved config unless
the override was marked persistent.
Signed-off-by: Samuel Mendoza-Jonas <sam@mendozajonas.com>
Currently in network_register_device() and network_unregister_device()
the appropriate interface is searched for by name. However it is
possible in some scenarios for multiple interfaces to have the same
name, so instead search by UUID to be sure that the correct interface is
being selected.
Signed-off-by: Samuel Mendoza-Jonas <sam@mendozajonas.com>
Currently discover_device_create() will search for existing discover
devices by id to determine if a new device is required. However it is
possible under some circumstances for distinct devices to have the same
name. This is especially troublesome if the following network events are
seen in network_handle_nlmsg():
- New interface, 'foo' with uuid x:x:x:x:x:x
-> new discover device created with
dev->device->id = 'foo'
dev->uuid = x:x:x:x:x:x
- New interface, 'foo' with uuid y:y:y:y:y:y
-> existing device 'foo' found
dev->uuid = y:y:y:y:y:y
This can occur if an interface rename event arrives *after* an old name
is reused, where temporarily Petitboot will see two distinct network
interfaces with the same name. Now the two interfaces point to the same
discover device, which can quickly result in a segfault if a 'remove'
event occurs for one of the interfaces and the discover device is freed.
To generally avoid this a 'uuid' parameter is added to
discover_device_create(), which if present allows existing devices to be
looked up by UUID rather than just their name.
Signed-off-by: Samuel Mendoza-Jonas <sam@mendozajonas.com>
pb-udhcpc currently only passes the interface's MAC address for an 'add'
event, however it is useful to know it for any user event concerning a
network interface. Pass it to pb-event in all cases.
Signed-off-by: Samuel Mendoza-Jonas <sam@mendozajonas.com>
The 'conf' user event is functionally very similar to the 'url' event,
in that both events result in downloading a specified configuration file
and passing it to iterate_parsers().
The 'url' event additionally allows downloading files from a directory
path and is also accessed by the UI via pb-protocol, so remove the
'conf' event and associated functions in favour of 'url' and
device_handler_process_url().
Signed-off-by: Samuel Mendoza-Jonas <sam@mendozajonas.com>
discover/event: Ensure event struct exists for async callers
When handling an event, user_event_handle_message() creates an event
struct with relevant parameters. Once user_event_handle_message() is
finished it frees the struct.
However in the case of a dhcp or add_url event, asynchronous jobs may be
spawned that will later reference the event struct. In particular this
becomes a problem when pxe_process_pair() handles an IPAPPEND name/value
pair and tries to access event->device.
In the case of dhcp and add_url events, we avoid this by changing the
event struct's talloc parent to the discover_context struct which
persists until all async pxe jobs have completed.
Signed-off-by: Samuel Mendoza-Jonas <sam@mendozajonas.com>
In certain cases, such as network booting over an untrusted connection,
it may be useful to fully encrypt and sign the kernel files.
Enable fully encrypted boot using builtin keyring via the addition of
the string "ENCRYPTED" to the first line of the /etc/pb-lockdown file.
This disables detached (plaintext) signature verification.
Signed-off-by: Timothy Pearson <tpearson@raptorengineering.com> Signed-off-by: Samuel Mendoza-Jonas <sam@mendozajonas.com>
Add support for GPG signature enforcement on booted
kernels and related blobs
This can be used to implement a form of organization-controlled secure boot,
whereby kernels may be loaded from a variety of sources but they will only
boot if a valid signature file is found for each component, and only if the
signature is listed in the /etc/pb-lockdown file.
Signed-off-by: Timothy Pearson <tpearson@raptorengineering.com> Signed-off-by: Samuel Mendoza-Jonas <sam@mendozajonas.com>
(Minor build fixes and gpgme.m4, comment on secure boot in gpg.c)
In some environments a default 'tun' device is created. Petitboot
doesn't use this and it clutters up the list of network devices, so
ignore it if it is encountered.
Signed-off-by: Samuel Mendoza-Jonas <sam@mendozajonas.com>
It turns out ncurses helpfully provides REQ_LEFT_FIELD and
REQ_RIGHT_FIELD to navigate between visually horizontal fields. Update
widgetset_process_key() to use these for KEY_LEFT and KEY_RIGHT
respectively.
Signed-off-by: Samuel Mendoza-Jonas <sam@mendozajonas.com>
If we don't specify command line arguments for the next kernel, kexec
will add the contents of /chosen/bootargs if present. This is unintended
and not obvious to the user, so explicitly add append="" to the kexec
arguments if we have none to add instead.
Signed-off-by: Samuel Mendoza-Jonas <sam@mendozajonas.com>
Jeremy Kerr [Fri, 5 Aug 2016 07:05:12 +0000 (15:05 +0800)]
hooks/set-stdout: move hook to after automatic stdout settings
If a console is defined in NVRAM, we want it to take precedence over
anything that 30-add-offb calculates automatically. This change shifts
the 20-set-stdout script to later in the hook processing, so that it
will override any automatic settings.
Signed-off-by: Jeremy Kerr <jk@ozlabs.org> Signed-off-by: Samuel Mendoza-Jonas <sam@mendozajonas.com>
Jeremy Kerr [Fri, 5 Aug 2016 07:05:11 +0000 (15:05 +0800)]
hooks/add-offb: Add leading slash to HVC node paths
We need an absolute path to the serial@N device. It looks like some
kernels aren't reading the current path correctly, leading to no console
output.
Signed-off-by: Jeremy Kerr <jk@ozlabs.org> Acked-by: Stewart Smith <stewart@linux.vnet.ibm.com> Signed-off-by: Samuel Mendoza-Jonas <sam@mendozajonas.com>
Add a user event named 'sync' that causes the discover server to merge
the devicemapper snapshots of mounted devices. This is particularly
useful as a debug aid (for example, when copying logs to a USB device),
as the server will otherwise only sync changes to mounted devices in
response to parser actions.
The command can be called as
pb-event sync@device
to sync a particular device, or as
pb-event sync@all
to sync all devices with snapshots.
Signed-off-by: Samuel Mendoza-Jonas <sam@mendozajonas.com>
ui/ncurses: Call widget process_key handlers first
Adding KEY_LEFT and KEY_RIGHT brought to light the problem that
widgetset_process_keys() may handle keystrokes that would have also been
handled by a widget's process_keys function. In particular the cursor
in a textbox widget could no longer be moved with the left/right keys.
This updates widgetset_process_keys() to call the focussed widget's
process_keys function before handling the key in any other way. All of
the widget process_keys functions correctly return false if the key is
not relevant to the widget except for textbox_process_key() which is
updated to ignore the main navigational keys.
Signed-off-by: Samuel Mendoza-Jonas <sam@mendozajonas.com>
discover/network: Mark interfaces configured once configured
In some cases additional netlink messages can be received for an
already-configured interface without any relevant changes. This can
result in multiple DHCP requests for the same interface.
Once an interface has been configured mark it as IFSTATE_CONFIGURED
to avoid configuring it again.
Signed-off-by: Samuel Mendoza-Jonas <sam@mendozajonas.com>
discover/platform-powerpc: Use IPMI on BMC-based machines only
With a kernel that has IPMI support Petitboot will try to use direct
IPMI, however on FSP machines this is not fully functional. Use direct
IPMI only on BMC-based machines, and use sysparams otherwise.
Signed-off-by: Samuel Mendoza-Jonas <sam@mendozajonas.com>
ui/ncurses: Properly set focus when updating subset
When a subset widget is made empty focus is switched to the first
visible field. It should actually be set to the first visible and active
field, otherwise we can try to focus an un-selectable label.
While we're here, also properly set the visibility for the autoboot
enable/disable widget.
Signed-off-by: Samuel Mendoza-Jonas <sam@mendozajonas.com>
utils/hooks: Set linux,stdout-path for primary console
If the boot_tty environment variable is set, determine which device path
to set in the linux,stdout-path property, which will instruct the next
kernel to use it as the primary console.
Signed-off-by: Samuel Mendoza-Jonas <sam@mendozajonas.com>
Include a list of available consoles as specified by the discover server
in nc-config from which the user can select a default. The selected
console specifies the which console to be set as the primary console
when a kernel is kexec'd.
Signed-off-by: Samuel Mendoza-Jonas <sam@mendozajonas.com>
If boot_tty is set or a boot command is sent manually from a certain
console, set the boot_tty environment variable to be used by a boot hook
before kexec.
Signed-off-by: Samuel Mendoza-Jonas <sam@mendozajonas.com>
discover: Add petitboot,tty and track available consoles
Add the NVRAM parameter petitboot,tty which sets the default console to
use when booting a kernel.
In load_config() construct a list of available consoles depending on the
current platform. A future patch depending on firmware changes will
allow this list to be constructed dynamically.
Signed-off-by: Samuel Mendoza-Jonas <sam@mendozajonas.com>
Add tty_list and boot_tty to the config struct to keep track of
available console interfaces and the default console to set as primary
respectively.
Also add a tty field to the boot_command struct so that the current
console can be sent to the discover server during a manual boot command.
Signed-off-by: Samuel Mendoza-Jonas <sam@mendozajonas.com>
Depending on the configuration of the DHCP server and the network, tftp
requests made by the pxe parser can timeout. The pxe parser makes these
requests synchronously so several timeouts can block the server
completely for several minutes, leaving the server unresponsive to UI
requests.
Rework the pxe parser such that it handles the result of each tftp
request in a callback, which can complete after iterate_parsers() has
returned. Each callback is allocated its own conf_context which takes a
talloc reference on the discover_context so that each callback can
commit new boot options after the initial iterate loop has completed.
This also means talloc_unlink must be used instead by the original
parent of the discover_context.
Signed-off-by: Samuel Mendoza-Jonas <sam@mendozajonas.com>
Substitute load_url_async() when running tests to support direct
callers of load_url_async() who will expect to read a file in a
callback.
Stub out device_handler_discover_context_commit() since it will remove
discover_options from the given discover_context, but the tests will
check the discover_context to count boot_options.
Signed-off-by: Samuel Mendoza-Jonas <sam@mendozajonas.com>
projects / petitboot / log