Table of Contents
Various scripts in partman make use of function libraries: separate files, mostly containing shell functions, that are sourced by scripts so they can make use of common functions and code duplication can be avoided.
The main function library is base.sh
, provided by
partman-base
. This file is sourced by
most scripts in partman as it defines a lot of useful variables and common
functions. Some of these are documented in the remainder of this chapter.
[5]
Other packages provide more targeted function libraries. In most cases their scope and use can easily be determined from their name.
The variables TAB
and NL
have values
ASCII 9 and ASCII 10 correspondingly. They can be used as temporary values
for the variable IFS
. The function
restore_ifs restores the variable IFS
its
original value.
The function library base.sh also contains simple reimplementations of basename and dirname so that busybox doesn't have to provide them.
The function debconf_select is a high level function to ask user with a menu using a Debconf question with type `select'. Synopsis:
debconf_selectpriority
template
choices
default
The first argument is the debconf-priority of the question and the second is the name of the template to be used. The third argument is a newline-separated list of items for the menu. Each item has the form
menu_item_id
<TAB>The text for the user
Here <TAB>
is ASCII 9. The text `The text for the
user' is the text of the menu item. If
of some menu-item is
identical with the fourth argument given to debconf_select
then this menu-item will be default.
menu_item_id
If the user cancels the question debconf_select returns with
exit-code 255. Otherwise the value of the variable RET
will
be the
of the chosen
menu item. If the chosen menu item was chosen by the user then the exit-code
is 0. If the item was chosen automatically (due to the debconf-priority or to
some other reason) the exit-code is 1.
menu_item_id
The function debconf_select doesn't care to
db_fset $template
seen
false
. The template must have exactly the following type
and choices fields:
Type: select Choices-C: ${CHOICES} Choices: ${DESCRIPTIONS}
The udebs that generate menus using menu-directories use the function ask_user instead of debconf_select. Synopsis:
ask_usera_menu_directory
additional_optional_arguments
...
This function displays the menu for
. The first
argument is a menu-directory (see Section 3.2, “Menu-directories”). If the user
cancels the dialog then ask_user returns with exit code 255.
Otherwise it returns with the exit code of the script
do_option.
a_menu_directory
If ask_user is called re-entrantly from within a do_option script, then the calling do_option script should typically be careful to handle or discard exit code 255 itself (and sometimes other codes, depending on the protocol in force) to avoid a backup operation inadvertently backing up out of several nested menus at once.
The script choices is invoked with
as
arguments. The first argument given to do_option is the
additional_optional_arguments
of the chosen menu
item and the other arguments are again
menu_item_id
.
additional_optional_arguments
To set the default selected item in a menu-directory, use the function menudir_default_choice. Synopsis:
menudir_default_choicea_menu_directory
subdirectory
menu_item_id
Where the
is the
name of a subdirectory in the menu-directory with the leading sequence number
stripped off and subdirectory
is
the id of a menu-item printed by
menu_item_id
.
The specified item is set as default not forever but only for the next
invocation of ask_user. It is not an error to set as
default non-existing item; in this case the first item in the menu will be
default.
a_menu_directory
/??subdirectory
/choices
The function partition_tree_choices prints a sequence of lines in the form
menu_item_id
<TAB>The text for the user
– one for every storage device and one for every partition. The
of the storage
devices is their storage directory. The
menu_item_id
of the partitions
has the form
menu_item_id
storage_directory
//partition_id
.
The output of partition_tree_choices can be given as third
argument to debconf_select.
Notice that the sizes of most of the present storage devices are so large that we cannot measure them using 32-bit integers. Consequently we cannot use the usual shell arithmetic. The functions longint_le, longint2human, human2longint and valid_human exist in order to deal with such big numbers.
The function longint_le is used to compare two big numbers.
longint_lenumber1
number2
returns with exit code 0 if the first number is less or equal to the second and returns 1 otherwise.
The function longint2human accepts in its first argument some number of bytes, converts it to something that is more meaningful for humans and outputs the result. For example
longint2human 1234567890
gives 1.2 GB
. Notice that this function rounds its
argument.
The function human2longint is used for the opposite convertion:
human2longint 1.234Gb
gives 1234000000
.
The function valid_human returns with exit code 0 when its first argument is a string that is suitable to be given to human2longint. Otherwise it returns with exit code 1.
Different components of the installer may need to get information about the
partitions. They can communicate with parted_server in
order to know the characteristics of the partition. However not everything can
be known from parted_server. Imagine an udeb that provides
the user with the option to upgrade some existing GNU/Linux installation. This
udeb analyses the fstab and knows that some partition is
used as /home
and should not be formatted. This sort of
information has nothing to do with parted_server. The udeb
stores it in a subdirectory of the device directory named after the id of the
partition.
But now a problem arises. Suppose that the user chooses to format some
partition as ext2 and mount it on /home
. The udebs
responsible for formatting and mounting create the directories
filesystem and mountpoint in the
partition. What will happen if the users change their mind and decide to use
the same partition as swap space? Swap spaces have no mount points and the
file mountpoint should be removed. Who is responsible for
removing it? The udeb that allows the user to choose a file system for the
partition doesn't have to know that swap-spaces have no mount points, only the
udeb that provides support for swap-spaces can know that the file
mountpoint should be removed.
In order to solve this difficulty every script that makes changes to some partition should invoke the function update_partition from base.sh. Synopsis:
update_partitiondevice_directory
partition_id
In order to update the contents of the directory
the function update_partition executes the scripts from the
directory device_directory
/partition_id
/lib/partman/update.d/
. Every udeb is allowed
to install scripts in this directory. Their names are prefixed by two-digit
numbers that control the order of the execution. The scripts from
update.d are given several arguments. $1 is the
device_directory
. $2 is the number of the partition
(/dev/hda6
will have number 6). $3 is the id of the
partition. $4 is the length of the partition (in bytes). $5 is the type of
the partition, it can be either `primary' or `logical'. $6 is the type of the
file system as known to parted_server, in most cases you
should ignore this argument. $7 is the device name (for example
/dev/ide/host0/bus0/target0/lun0/part6
).
"$8 $9 $10 $11 ..." is the name of the partition in partition tables that
support partition names. Otherwise $8, $9, $10,... are not defined.
The package partman-base
creates two
FIFOs – /var/lib/partman/infifo
and
/var/lib/partman/outfifo
.
Parted_server reads instructions from
infifo and responds by writting to
outfifo. Consequently the clients write to
infifo and read from outfifo. The
function library base.sh contain several functions to make
the communication with parted_server easier. Here we will
omit the details, if you want to know the exact communication protocol please
read how these functions are implemented.
The functions open_infifo, close_infifo, open_outfifo and close_outfifo are called without arguments. They open and close infifo and outfifo assigning them file descriptors 6 and 7 correspondingly. You do not need to use these low-level functions.
The function write_line prints its arguments to outfifo.
The function read_line reads from infifo
a line, splits it in fields according to $IFS
and assigns
these fields to variables whose names are given to read_line
as arguments. For example
read_line x y z
reads a line from infifo, splits it and assigns the first
field to the variable x
, the second field to the variable
y
and the rest to the variable z
. You
see that read_line is used the same way as the shell
operator read.
The function read_paragraph reads consequently lines from infifo until it reaches an empty line. It prints the read lines with the exception of the last empty line.
The function read_list reads lines the same way as the function read_paragraph. However the function read_list always prints only one line that is a comma-separated sequence of the lines read from infifo. If read_paragraph prints
This is the first line This is the second line This is the third line
read_list prints
This is the first line, This is the second line, This is the third line
In order to initiate a communication dialog with parted_server you will use the function open_dialog. You will invoke it in the device directory of the device you want to issue command about. The first argument of open_dialog is a command for parted_server. The rest arguments are arguments for the command.
You use the function close_dialog in order to terminate the communication dialog.
When you send parted_server an order to do some long operation (e.g. resize a file system) the user will be shown a progress bar. You may give a name to it by the function name_progress_bar. It may be used right before the command open_dialog and accepts only one argument – a template with type text that describes what is being done.
The function log appends its arguments to the file
/var/log/partman
. This file is used as log-file also by
parted_server.
[5] The script base.sh
has become somewhat
too long and this probably makes its sourcing slow. Although some
reorganization has already taken place, it could be a good idea to split it up
further.