Age | Commit message (Collapse) | Author |
|
This commit moves the SELinux label code after the tree is sorted and
the inode table is generated. Sorting helps to make sure that the tree
will always be traversed in a defined, deterministic order and likewise
the creation of xattrs happens in a defined, deterministic order.
Second, we can now use the inode table instead of having to implement a
recursive tree traversal yet again.
Signed-off-by: David Oberhollenzer <david.oberhollenzer@sigma-star.at>
|
|
Signed-off-by: David Oberhollenzer <david.oberhollenzer@sigma-star.at>
|
|
Always try to gather all compressor libraries available, but only
complain about missing a one if it has been *explicitly* selected.
If a compressor has been explicityl disabled, we can still turn it
off after checking.
Also, rework gensquashfs to set the default compressor based on
what's available.
Signed-off-by: David Oberhollenzer <david.oberhollenzer@sigma-star.at>
|
|
Signed-off-by: David Oberhollenzer <david.oberhollenzer@sigma-star.at>
|
|
Signed-off-by: David Oberhollenzer <david.oberhollenzer@sigma-star.at>
|
|
Signed-off-by: David Oberhollenzer <david.oberhollenzer@sigma-star.at>
|
|
Signed-off-by: David Oberhollenzer <david.oberhollenzer@sigma-star.at>
|
|
Signed-off-by: David Oberhollenzer <david.oberhollenzer@sigma-star.at>
|
|
Signed-off-by: David Oberhollenzer <david.oberhollenzer@sigma-star.at>
|
|
Signed-off-by: David Oberhollenzer <david.oberhollenzer@sigma-star.at>
|
|
Add a generic one that reads and processes N bytes from an input
file descriptor.
Signed-off-by: David Oberhollenzer <david.oberhollenzer@sigma-star.at>
|
|
This commit tries to make the block and fragment processing code a
lot more readable by simplifying it as follows.
The following pattern is used repeatedly:
- Try to compress a block
- Write the compressed if smaller (the original otherwise) to a file
- Print errors on the way
- Set the magic (1 << 24) flag if we wrote the original
- Increment the super block byte counter
=> Move that to a seperate function called "write_compressed"
=> Move the code to grow the fragment table is moved to a seperate function.
=> Clearing the fragment buffer is pointless, so remove that
The add_fragment/add_block functions are now trivial
=> Merge them into process_file
=> Global block counter no longer needed, so remove that too
Signed-off-by: David Oberhollenzer <david.oberhollenzer@sigma-star.at>
|
|
Don't need that, we already know the index when we write add
the fragment.
Signed-off-by: David Oberhollenzer <david.oberhollenzer@sigma-star.at>
|
|
Signed-off-by: David Oberhollenzer <david.oberhollenzer@sigma-star.at>
|
|
Signed-off-by: David Oberhollenzer <david.oberhollenzer@sigma-star.at>
|
|
The idea is to move various higher level helper functions there.
Signed-off-by: David Oberhollenzer <david.oberhollenzer@sigma-star.at>
|
|
Signed-off-by: David Oberhollenzer <david.oberhollenzer@sigma-star.at>
|
|
Signed-off-by: David Oberhollenzer <david.oberhollenzer@sigma-star.at>
|
|
Signed-off-by: David Oberhollenzer <david.oberhollenzer@sigma-star.at>
|
|
Instead of allocating inode numbers as we go, generate and populat an
inode table from the fstree ahead of time. This makes processing nodes
a little bit simpler and we will need that table anyway for NFS export
support later on.
Signed-off-by: David Oberhollenzer <david.oberhollenzer@sigma-star.at>
|
|
Signed-off-by: David Oberhollenzer <david.oberhollenzer@sigma-star.at>
|
|
This commit replaces the chdir to the input directory with pushd/popd
when building the fstree and again when packing files.
This simplifies handling of other file paths given on the command line
that have to be accessed and are relative to the original working
directories.
Signed-off-by: David Oberhollenzer <david.oberhollenzer@sigma-star.at>
|
|
The problem: SquashFS uses extended inode types for inodes with xattrs,
so for directories we need to generate an ldir inode. The ldir inode
type includes a directory index which cannot be empty (the counter is
off by 1, so 0 means 1 entry) and the entry is expected to contain a
file name with the same issue (must be at least 1 byte).
This has probably not really been thought through. What the mksquashfs
actually does is generating an ldir inode without an index. The kernel
does not bother to read the index, since the size of the directory is 0.
Its a hack that just so happens to work.
Signed-off-by: David Oberhollenzer <david.oberhollenzer@sigma-star.at>
|
|
Signed-off-by: David Oberhollenzer <david.oberhollenzer@sigma-star.at>
|
|
If packdir and packfile are both specified, use packdir as
alternate root.
Signed-off-by: David Oberhollenzer <david.oberhollenzer@sigma-star.at>
|
|
Signed-off-by: David Oberhollenzer <david.oberhollenzer@sigma-star.at>
|
|
Signed-off-by: David Oberhollenzer <david.oberhollenzer@sigma-star.at>
|
|
Signed-off-by: David Oberhollenzer <david.oberhollenzer@sigma-star.at>
|
|
For now, it is enough to reduce the comparison value, since the inode
numbers are increasing.
Signed-off-by: David Oberhollenzer <david.oberhollenzer@sigma-star.at>
|
|
Signed-off-by: David Oberhollenzer <david.oberhollenzer@sigma-star.at>
|
|
Signed-off-by: David Oberhollenzer <david.oberhollenzer@sigma-star.at>
|
|
Signed-off-by: David Oberhollenzer <david.oberhollenzer@sigma-star.at>
|
|
Signed-off-by: David Oberhollenzer <david.oberhollenzer@sigma-star.at>
|
|
Signed-off-by: David Oberhollenzer <david.oberhollenzer@sigma-star.at>
|
|
Signed-off-by: David Oberhollenzer <david.oberhollenzer@sigma-star.at>
|
|
Signed-off-by: David Oberhollenzer <david.oberhollenzer@sigma-star.at>
|
|
Use libselinux to lookup the context attributes from a file.
Signed-off-by: David Oberhollenzer <david.oberhollenzer@sigma-star.at>
|
|
Turns out (after countless hours of debugging since the squashfs kernel
code is so convoluted) that the offset is relative to the first directory
header for no apparant reason (the field holding the offset is way larger
than neccessary to store absolute values).
Signed-off-by: David Oberhollenzer <david.oberhollenzer@sigma-star.at>
|
|
Signed-off-by: David Oberhollenzer <david.oberhollenzer@sigma-star.at>
|
|
Signed-off-by: David Oberhollenzer <david.oberhollenzer@sigma-star.at>
|
|
Signed-off-by: David Oberhollenzer <david.oberhollenzer@sigma-star.at>
|
|
- gensquashfs simply asks the backend compressor to write its options
to the file and does accounting
- rdsquasfs simply asks the backend compressor to transparentyl snort
the options from the file
- not implemented in any compressor backend yet
Signed-off-by: David Oberhollenzer <david.oberhollenzer@sigma-star.at>
|
|
Signed-off-by: David Oberhollenzer <david.oberhollenzer@sigma-star.at>
|
|
Signed-off-by: David Oberhollenzer <david.oberhollenzer@sigma-star.at>
|
|
This commit changes the names of the tools to gensquashfs and rdsquashfs
so they don't collide with the names used by the squashfs-tools package
and the two can be installed side by side.
Signed-off-by: David Oberhollenzer <david.oberhollenzer@sigma-star.at>
|
|
Stick to the terminology that SquashFS uses. SquashFS uses "lzma" to
refere to lzma1, "xz" to refere to lzma2 and confusingly "gzip" to
refere to raw zlib deflate. We can avoid a lot of confusion by
_consistently_ using the same terminology.
Signed-off-by: David Oberhollenzer <david.oberhollenzer@sigma-star.at>
|
|
Pass in an external destination buffer + size and allow for propper
bounds checking (especially when unpacking).
Signed-off-by: David Oberhollenzer <david.oberhollenzer@sigma-star.at>
|
|
Signed-off-by: David Oberhollenzer <david.oberhollenzer@sigma-star.at>
|
|
- remove debug prints
- move id table write out to id table code
- dummy remove wrapper function for fragment table
Signed-off-by: David Oberhollenzer <david.oberhollenzer@sigma-star.at>
|
|
Signed-off-by: David Oberhollenzer <david.oberhollenzer@sigma-star.at>
|