discover/udev: Additional checks in udev_handle_block_add
- Several filesystem types can appear that we won't be able to mount.
Instead of waiting to fail mounting them in device_handler_discover(),
skip processing them at all.
- Do not create dm-snapshots on top of raid arrays until we have a
reliable way of determining the sector count for a md raid device.
- Turn down the verbosity on skipping dm-devices.
Signed-off-by: Samuel Mendoza-Jonas <sam.mj@au1.ibm.com>
discover/devmapper: Disable libdm udev sync support
Certain userspace environments that Petitboot is packaged with include
a libdm built with udev sync support which hung with the previous approach.
Relying on udev to properly process dm device creation makes some
assumptions about the flavour of udev available - until that is certain
disable udev sync support and have device-mapper control device creation
itself.
Signed-off-by: Samuel Mendoza-Jonas <sam.mj@au1.ibm.com>
Create a new Petitboot option 'petitboot,write?' that specifies whether
the system is allowed to mount devices read-write. The option can be
toggled by the user in the nc-config screen.
Signed-off-by: Samuel Mendoza-Jonas <sam.mj@au1.ibm.com>
discover: Mount snapshots for all eligible disk devices
Device-mapper snapshots are created for all disk devices prior to
being mounted. If explicit writes are made to the snapshot they are
merged back to the disk once write access is released.
Signed-off-by: Samuel Mendoza-Jonas <sam.mj@au1.ibm.com>
discover: Support creation of device-mapper devices
Add discover/dm-snapshot that allows the creation of device-mapper
snapshots that support merging changes back to disk.
Device-mapper snapshots are a CoW device backed by a ramdisk, mirroring
the contents of a source device. No changes are made to the original
disk unless an explicit merge action is performed. This guarantees
read-only mounting of host disks even when writes could implicitly
occur, eg. when performing recovering a journaled filesystem.
In the event that writing back to the disk is desired, such as when
updating grubenv, the changes made to the snapshot can be merged back to
the source disk.
This patch adds support but does not change functionality.
Signed-off-by: Samuel Mendoza-Jonas <sam.mj@au1.ibm.com>
Jeremy Kerr [Fri, 26 Jun 2015 03:41:46 +0000 (11:41 +0800)]
discover: Fix raw string argument to talloc_asprintf_append
Current builds give a warning:
../discover/platform-powerpc.c: In function ‘update_bootdev_config’:
../discover/platform-powerpc.c:667:4: warning: format not a string
literal and no format arguments [-Wformat-security]
tmp = val = talloc_asprintf_append(val, boot_str);
^
ui/ncurses: Properly reposition cursor after menu update
The currently selected item in the main menu can be set to an item off
the visible portion of the menu after the additional or removal of a
boot option. Update the currently selected item and/or the current view
such that the item remains in the visible area.
Signed-off-by: Samuel Mendoza-Jonas <sam.mj@au1.ibm.com>
Send a short message to the UI to inform the user a device is being
parsed for boot options. This helps slightly in environments when the UI
appears well before devices are available for parsing, giving the user
an indication that work is still being done.
Signed-off-by: Samuel Mendoza-Jonas <sam.mj@au1.ibm.com>
discover: Clear default-boot-device when invalidated
Update the clear_ipmi_bootdev functions to optionally invalidate the
next- or default-boot-device for their respective machines.
If a client invalidates the ipmi_bootdev in the config, invalidate the
respective bootdev.
Signed-off-by: Samuel Mendoza-Jonas <sam.mj@au1.ibm.com>
Move petitboot to a more familiar 'boot-order' based autoboot system.
The discover server now reads multiple values from the petitboot,bootdev
parameter and adds them in order to config->autoboot_opts. Boot priority
is determined by the options' position in the list.
On the client, nc-config now recognises the new boot order, and allows
the user to add, remove, and reorder the devices in the list.
Signed-off-by: Samuel Mendoza-Jonas <sam.mj@au1.ibm.com>
Changing the visual order of widgets will cause their associated fields to be
out of order relative to the widgetset field array. Rather than manually
resorting the array, use sorted navigation to move according to a field's
visual positon
Signed-off-by: Samuel Mendoza-Jonas <sam.mj@au1.ibm.com>
The nc-subset screen is intended to be used as a sub-screen from the
current screen (eg. nc-config) which passes a pointer to a
nc_widget_subset struct. The nc-subset screen allows the user to select
an option from a list of 'inactive' options, before returning control
back to the current screen.
Signed-off-by: Samuel Mendoza-Jonas <sam.mj@au1.ibm.com>
The new nc_widget_subset is similar to nc_widget_select, but hides added
options until they are made active. This allows the widget to have
multiple options selected and ordered without cluttering the screen.
Signed-off-by: Samuel Mendoza-Jonas <sam.mj@au1.ibm.com>
discover: Add support for multiple bootdev arguments
To support multiple autoboot options while retaining backwards
compatability, interpret the petitboot,bootdev parameter as
optionally having several space-separated values.
Signed-off-by: Samuel Mendoza-Jonas <sam.mj@au1.ibm.com>
Add the new autoboot_option struct, and helper functions for working
with device_type enums. device_type_name() returns exact strings as used
by platform code to read/write nvram params, so
device_type_display_name() is added for use in user-visible strings.
Signed-off-by: Samuel Mendoza-Jonas <sam.mj@au1.ibm.com>
Joel Stanley [Thu, 30 Apr 2015 06:02:03 +0000 (16:02 +1000)]
discover: Set OS Boot sensor with 0xa9
If we don't set all the things, we only get an assertion event, without
the state being asserted.
Despite the IPMI spec suggesting we only needed to set 0x10, aka the
assertion bits, testing reveals we want to set all of the things to
cause the BMC to assert the state.
Tested-by: Nick Bofferding <bofferdn@us.ibm.com> Signed-off-by: Joel Stanley <joel@jms.id.au> Signed-off-by: Jeremy Kerr <jk@ozlabs.org>
Joel Stanley [Thu, 30 Apr 2015 06:02:03 +0000 (16:02 +1000)]
discover: Set OS Boot sensor with 0xa9
If we don't set all the things, we only get an assertion event, without
the state being asserted.
Despite the IPMI spec suggesting we only needed to set 0x10, aka the
assertion bits, testing reveals we want to set all of the things to
cause the BMC to assert the state.
Tested-by: Nick Bofferding <bofferdn@us.ibm.com> Signed-off-by: Joel Stanley <joel@jms.id.au> Signed-off-by: Samuel Mendoza-Jonas <sam.mj@au1.ibm.com>
discover: Mount with norecovery, avoid rw mount of XFS filesystems
Journaled filesytems may still write to their disk even if the disk is
mounted read only. Petitboot should avoid modifying any disks
automatically, and in mixed-endian systems this can also cause journal
operations to fail. Use the 'norecovery' option on filesystems that
support it to skip the journal replay.
Additionally, mounting an XFS filesystem as read-write in such a case
will cause the call to mount to hang indefinitely. Avoid this generally
by explicitly unmounting and (re)mounting when mounting read-write.
Signed-off-by: Samuel Mendoza-Jonas <sam.mj@au1.ibm.com>
Udev may change the name of network interfaces if persistent interface
naming is used. It is important to recognise this since udev may re-use
the old name for a different interface.
If we receive an RTM_NEWLINK message for an existing interface, check if
the name has been updated.
Signed-off-by: Samuel Mendoza-Jonas <sam.mj@au1.ibm.com>
All current *_post() methods in ui/ncurses call redrawwin() and
wrefresh() together. wrefresh() updates any lines on the screen that
have been marked as changed or invalid. However redrawwin() marks the
entire screen as invalid unconditionally. We can reduce the amount of
data written to the screen by avoiding calls to redrawwin().
Screen transitions are the primary use case of redrawwin(), where the
whole screen must be invalidated to avoid stale data remaining on
screen. All other 'in-screen' updates such as changes to widgets or
changing focus do not require a call to redrawwin(). The most noticeable
performance improvement is in nc-menu, which makes an unnecssary call to
redrawwin() after every addition to the boot option menu.
eg. The number of bytes written to STDOUT in the main menu:
# Boot options | Before | After
--------------------------------
8 | 5488 | 1149
133 | 422454 | 4652
Signed-off-by: Samuel Mendoza-Jonas <sam.mj@au1.ibm.com>
A boot option requires an image path at minimum. If an option is
created without an image, trying to edit the resulting user item causes
a segfault. Prevent the user from creating a boot option without at
least an image path.
Signed-off-by: Samuel Mendoza-Jonas <sam.mj@au1.ibm.com>
Jeremy Kerr [Thu, 11 Dec 2014 08:38:21 +0000 (16:38 +0800)]
discover: Integrate ipmi bootdev settings into the priority system
Currently, we expose the boot device priorities through an array in
struct config, which will either be the default (network -> disk), or a
single device type specified by the IPMI code.
Rather than hide the implementation details in this array, we'd like to
expose the details of the machine configuration instead. This allows
user visibility of the real boot configuration (for example, if an IPMI
boot preference is set).
This change removes the priority array, and replaces it with the
ipmi_bootdev data (and a persistent flag). We update the
default-conflict-resolution code to reflect the priorities between IPMI
and UUID preferences.
Jeremy Kerr [Tue, 2 Dec 2014 07:04:38 +0000 (15:04 +0800)]
discover/powerpc: Separate ipmi bootdev handling into separate functions
We'd like to add a new backend to the bootdev storage, so move the
common bootdev-handling code into separate functions, moving
ipmi_bootdev_is_valid to ipmi.c.
Jeremy Kerr [Thu, 11 Dec 2014 01:42:58 +0000 (09:42 +0800)]
discover/grub2: Allow unset and invalid defaults
If the default environment variable is unset or invalid (i.e.,
references a non-existent boot option), then GRUB2 will fallback to the
first boot option present. This is preventing petitboot from autobooting
where no default is explicitly set, or is stale.
This change adds this fallback behaviour to petitboot. Because we don't
know if the first option will be a default at parse time (as no other
options matched the default env var), we need to keep options in a list,
and register them with the discover server once the parse is complete.
discover/pxe: Format IPAPPEND mac addresses correctly
The SYSAPPEND/IPAPPEND option 2 in PXE configs requires
the MAC address of the booting interface to be appended
to the boot options. Previously we formatted this as
"BOOTIF=01:02:03:04:05:06",
but syslinux/pxelinux implementation use this format:
"BOOTIF=01-01-02-03-04-05-06",
where the leading '01' represents the hardware type.
Jeremy Kerr [Mon, 1 Dec 2014 03:47:16 +0000 (11:47 +0800)]
discover/grub2: improve handling of word & delimiter tokens
Currently, the delimiter token handling is a little fragile: we try to
ignore non-inter-word delimiters in the lexer with a selective set of
regexes on the possible delimiter characters.
This means we don't need to handle potential delimiters in every grammar
rule, but there are other situations (not regex-able) where we may see
delimters, and this will cause a parse error.
Instead of relying on the regex behaviour, we have an 'inter_word' flag,
which is set when we see the first word token, and cleared when we see
an end-of-line token. We only emit TOKEN_DELIM when this flag is set.
This means that we only get the delim tokens when they're required -
when we're looking for word separators (becuase WORD DELIM WORD is
distinct from WORD WORD - eg "linux /vmlinux" and "x$var").
We add a few new tests for the "menuentry" and "if" syntax, with
different delimiter configurations.
Jeremy Kerr [Fri, 28 Nov 2014 03:42:06 +0000 (11:42 +0800)]
lib/fold: fix warning with -DDEBUG
When compiling the fold code with DEBUG defined, we get:
../lib/fold/fold.c: In function 'fold_text':
../lib/fold/fold.c:37:3: error: comparison of unsigned expression >= 0 is always true [-Werror=type-limits]
assert(bytes >= 0);
^
We should be comparing with != (size_t)-1, not >= 0
Widgets running over the horizontal width of the pad will cause the
screen to blank. Expand the (non-viewable) size of the pad to prevent
blanking until shorter translations are available for nc-config.
Signed-off-by: Samuel Mendoza-Jonas <sam.mj@au1.ibm.com>
Jeremy Kerr [Thu, 18 Sep 2014 11:21:00 +0000 (19:21 +0800)]
ui/ncurses: Fix clearing a boot device UUID
Currently, we have a bug where we can't clear a boot device UUID, as we
only update config->boot_device if we're in AUTOBOOT_ONE state. This
means that the config UI can't be used to change out of "autoboot from a
single device" mode.
This sets the autoboot device unconditionally, allowing it to be
cleared.
ui/ncurses: Exit safe mode on configuration change
Selecting 'OK' in the configuration or language select screen will send
a configuration change to the server and exit safe mode.
Exit safe mode explicity and warn the user that they are about to exit
it.
Signed-off-by: Samuel Mendoza-Jonas <sam.mj@au1.ibm.com> Signed-off-by: Jeremy Kerr <jk@ozlabs.org>
Jeremy Kerr [Wed, 23 Jul 2014 06:03:42 +0000 (14:03 +0800)]
discover: Use platform code to read sysinfo type and identifier
This change uses the platform-specific code to read the system type and
identifier, rather than the pb-sysinfo utility. We introduce a new
callback for struct platform:
int (*get_sysinfo)(struct platform *, struct system_info *);
- which populates struct system_info with appropriate information.
This means that the system-specific code is kept in one place; rather
than having powerpc-specific device-tree-reading code in the pb-sysinfo
shell script.
Jeremy Kerr [Mon, 4 Aug 2014 01:46:00 +0000 (09:46 +0800)]
discover/grub2: distinguish parser.h from grub2/parser.h
'make distcheck' will do a 'make all' srcdir == objdir, then later a
'make check' sith srcdir != objdir. Since gcc's cpp always assumes that
a source file's directory is first in the include paths, we may see
discover/parser.h included when we wanted the generated
discover/grub2/parser.h.
This change renames the grub2 lexer and parser files, to work-around
this behaviour, and fix 'make distcheck'.
Jeremy Kerr [Fri, 1 Aug 2014 03:55:45 +0000 (11:55 +0800)]
autotools: Don't link udev into everything
By default, AC_CHECK_LIB will append to $LIBS, which is used for all
link stages. Intstead, we should popuate $UDEV_LIBS, and just use that
for the single pb-discover link stage.
Jeremy Kerr [Mon, 28 Jul 2014 08:20:04 +0000 (16:20 +0800)]
autotools: Use non-recursive make
With the current testing infrastructure, we don't have a strictly
hierarchical set of dependencies. This causes problems with a recursive
make, and means we have to hack around some of the dependencies.
This change generates a single, top-level makefile from all of the
Makefile.am fragments. We still need the po/ directory as a separate
SUBDIR, but all others can be converted to non-recursive.
projects / petitboot / log