GitLab

In case the memory allocation fails in the command parser, there
will might remain some allocated memory upon return. Fix this.
Signed-off-by: Marek Vasut <marex@denx.de>

No functional change in the code, just reorder the functions so
they are not in such a state of chaos.
Signed-off-by: Marek Vasut <marex@denx.de>

The variable was initialized and reinitialized twice as a remnant
of the rework, initialize it only once.
Signed-off-by: Marek Vasut <marex@denx.de>

There is no need to pass the IV explicitly as another argument,
the function can pull this information from the image context.
Signed-off-by: Marek Vasut <marex@denx.de>

Update the checksum of the command right after we have all necessary
information filled in it instead of doing it late in the function that
encrypts the command.
Signed-off-by: Marek Vasut <marex@denx.de>

Split the code that encrypts portions of the image and the code
that assembles the resulting image in memory.
Signed-off-by: Marek Vasut <marex@denx.de>

We need to store image digest in image context to let it
traverse between the image creation function and image write
function.
Signed-off-by: Marek Vasut <marex@denx.de>

Store image dictionary in image context, as this is where it should
be according to the documentation. We also need it here to let it
traverse between the image creation function and image write function.
Signed-off-by: Marek Vasut <marex@denx.de>

This patch pulls out the function calls that dump the final image
at the end of function that creates the image. This will now allow
easy separation of them into self-standing function.
Signed-off-by: Marek Vasut <marex@denx.de>

The encryption of command data corrupted the contents of the
structure. To prevent that, store the encrypted command data
in a separate entry in the command context structure.
Signed-off-by: Marek Vasut <marex@denx.de>

This instruction does nothing and is easy to implement.
Signed-off-by: Marek Vasut <marex@denx.de>

When loading the IVT, make a local copy of it instead of pointing
directly onto the static structure. This makes it easier to operate
with the file buffers associated with LOAD command in mostly uniform
manner.
Signed-off-by: Marek Vasut <marex@denx.de>

The TAG command has to be encrypted using IV from the header,
so reinit the AES with the IV for each section.
Signed-off-by: Marek Vasut <marex@denx.de>

This patch fixes the generation of section header block for multiple
sections. Before this patch, only one section was always generated,
which was incorrect.
Signed-off-by: Marek Vasut <marex@denx.de>

Move the function lower in mxssb.c so there is no longer need for
forward declaration of sb_load_file().
Signed-off-by: Marek Vasut <marex@denx.de>

Move the loading of a file associated with a LOAD instruction into
sb_tag_to_command, since this is where the LOAD instruction is filled.
This will allow us to keep the handling of LOAD instruction in one
place.
Signed-off-by: Marek Vasut <marex@denx.de>

The command stream now only contains the filename as it should.
The IVTs are currently handled via little hack, but this will
change once there is a proper file parser.
Signed-off-by: Marek Vasut <marex@denx.de>

The image tree will be free'd in main(), so there is no need to
worry about freeing it here.
Signed-off-by: Marek Vasut <marex@denx.de>

The cmd allocation in sb_tag_to_command() is no longer needed.
It was needed temporarily for compatibility reasons during the
transition, but this is resolved now.
Signed-off-by: Marek Vasut <marex@denx.de>

The TAG instruction is wholy filled later and the LOAD instruction
fields need to be updated after a file was loaded. Thus, clean up
bogus assignments into fields of these instructions.
Signed-off-by: Marek Vasut <marex@denx.de>

The sb_load_binary() call is now redundant, since the files are
now loaded while preparing the SB image header. Reduce this to
a stub function that fills the sb instruction stream.
Signed-off-by: Marek Vasut <marex@denx.de>

Pull the SB command blocks from the image tree.
Signed-off-by: Marek Vasut <marex@denx.de>

Pull out the section header information from the SB image tree.
This allows us to remove some ad-hoc calls.
Signed-off-by: Marek Vasut <marex@denx.de>

The section offset computation has to be done after the
SB image header is filled in. Moreover, the TAG command
can be filled in only afterwards as well, since it's almost
exact copy of the section header. Implement a function that
fixes up the section headers and TAG commands.
Signed-off-by: Marek Vasut <marex@denx.de>

Pass the sb_image_ctx context into the cryptographic functions
instead of passing the standalone cryptographic structures. This
makes the code a little cleaner.
Signed-off-by: Marek Vasut <marex@denx.de>

Swap the malloc upwards as it can now no longer depends on SB header
crypto init which filled the SB header with additional data. Now
the SB header crypto operation and malloc of the image are truly
separate entities. This also groups the SB header crypto operation
together nicely.
Signed-off-by: Marek Vasut <marex@denx.de>

Calculate the size of the section block while reading it. This
value can be later reused when calculating the placement of the
final image blocks.
Signed-off-by: Marek Vasut <marex@denx.de>

This patch loads a file associated with a LOAD command and puts
it correctly into the tree-like structure representing the SB image.
Note that the loading operation is now done twice, but this will be
fixed shortly.
Signed-off-by: Marek Vasut <marex@denx.de>

Rework the mechanism behind generating the SB image header generation.
This also puts down the groundwork for further rework of the mechanism
for contructing the image.

The basic idea here is to parse the whole image, construct a tree-like
structure in RAM that represents the image and operate with it throughout
the whole life-cycle of the program. Right now, this only serves the
purpose of filling the SB image header, but this will change in subsequent
patches.

NOTE that this patch introduces a lot of FIXMEs, since this is a huge
rework of the whole tool. This patch revealed a lot of weak spots too.
Signed-off-by: Marek Vasut <marex@denx.de>

There is only one SB image header, so make it part of the image
context instead of having separate variable for this purpose.
Moreover, it's now possible to constify the SB header template.
Signed-off-by: Marek Vasut <marex@denx.de>

The program uses many variables throughout it's operation, implement
a program context structure to hold these variables.
Signed-off-by: Marek Vasut <marex@denx.de>

Do not abuse the start of image directly as the CBC-MAC IV, but
create separate overlaid field for this purpose.
Signed-off-by: Marek Vasut <marex@denx.de>

Implement a function that allocates space for an instructions
and fills in the fields on a per-command basis. Note that some
commands are still not implemented.
Signed-off-by: Marek Vasut <marex@denx.de>

Certain fields of the SB header are only available since v1.1
of the SB file format, document that.
Signed-off-by: Marek Vasut <marex@denx.de>

This patch adds all possible command flags that can be used with the
SB file format.
Signed-off-by: Marek Vasut <marex@denx.de>

Add details about the fields for all the commands, moreover align
the names of the fields with the documentation where it was not
confusing.
Signed-off-by: Marek Vasut <marex@denx.de>

Fill in the full table of SB command tags as they are listed in the
SB image specification.
Signed-off-by: Marek Vasut <marex@denx.de>

The SB image format specification calls the command identifier
a tag, so align the source this way. Also rename the commands
so they match the SB format specification.
Signed-off-by: Marek Vasut <marex@denx.de>

This will allow GCC to lint the sources even more.
Signed-off-by: Marek Vasut <marex@denx.de>

The type must be unsigned int, otherwise gcc -Wextra produces a warning.
Signed-off-by: Marek Vasut <marex@denx.de>