Note: This was last sent 6 years ago. It really belongs upstream and I believe it is useful functionality for libfdt, so I am trying again. Please take a fresh look at this. It is cut back from the previous series. If accepted we can expand the feature set piece by piece from here. This series adds two new functions, fdt_first_region() and fdt_next_region() which map FDT parts such as nodes and properties to their regions in the FDT binary. The function is then used to implement a grep utility for FDTs. The core code is quite simple and small, but it grew a little due to the need to make it iterative (first/next/next). Also this series adds tests and a grep utility, so quite a bit of code is built on it. The use for this feature is twofold. Firstly it provides a convenient way of performing a structure-aware grep of the tree. For example it is possible to grep for a node and get all the properties associated with that node. Trees can be subsetted easily, by specifying the nodes that are required, and then writing out the regions returned by this function. This is useful for small resource-constrained systems, such as boot loaders, which want to use an FDT but do not need to know about all of it. The full FDT can be grepped to pull out the few things that are needed - this can be an automatic part of the build system and does not require the FDT source code. This first use makes it easy to implement an FDT grep tool. Options are provided to search for named nodes and properties. It is possible to search for non-matches also (useful for excluding a particular property from the FDT, for example). The output is like fdtdump, but only with the regions selected by the grep. The fdtgrep utility cannot currently output valid source in all cases, which can be used by dtc. That will come later. Secondly it makes it easy to hash parts of the tree and detect changes. The intent is to get a list of regions which will be invariant provided those parts are invariant. For example, if you request a list of regions for all nodes but exclude the property "data", then you will get the same region contents regardless of any change to "data" properties. An assumption is made here that the tree ordering remains the same. This second use is the mechanism behind U-Boot's verified-boot feature using Flat Image Tree (FIT) images, introduced 6 years ago. Briefly, this works by signing configurations (consisting of particular kernel and FDT combinations) so that the boot loader can verify that these combinations are valid and permitted. Since a FIT image is in fact an FDT, we need to be able to hash particular regions of the FDT for the signing and verification process. This is done by using the region functions to select the data that needs to be hashed for a particular configuration. A third use is currently under discussion in the boot-architecture mailing list, which is to sign vanilla FDT files by adding a signature node. This will avoid the need for a separate signature for every device tree file, or inventing yet another file format. The fdtgrep utility could be used to replace all of the functions of fdtdump. However fdtdump is intended as a separate, simple way of dumping the tree (for verifying against dtc output for example). So fdtdump remains a separate program and this series leaves it alone. Note: a somewhat unfortunately feature of this implementation is that a state structure needs to be kept around between calls of fdt_next_region(). This is declared in libfdt.h but really should be opaque. Ideas welcome. Changes in v4: - Chop back to the minimal useful functionality - Rebase to master - Update cover letter for developments over the past 6 years Changes in v3: - Add -f option to display offset; make -a display an absolute file address - Add -s option to include all subnodes - Add a feature to include all subnodes - Add documentation on fdtgrep - Adjust help and command line processing to follow new approach - Rename -V to -I to avoid using -V for a different purpose to other tools - Rename -s to -e since it only 'enters' the node and does not include it all Changes in v2: - Add local fdt_find_regions() function since libfdt no longer has it - Add long comment explaining theory of operation - Add more comments about the -1 return value from h_include - Add new FDT_ERR_TOODEEP error type and use it - Add note that changes in node/property order can cause false hash misses - Change returned error from BADLAYOUT to BADSTRUCTURE - Drop FDT_IS_COMPAT and pass node offset to h_include function - Drop stale comment about names / wildcards - Fix info->count <= info->max_regions in fdt_add_region() merge case - Move region code to separate fdt_region.c file - Move to a model with fdt_first_region()/fdt_next_region() - Return FDT_ERR_BADLAYOUT error if strings block is before structure block Simon Glass (4): README: Explain how to add a new API function libfdt: Add functions to find regions in an FDT Add fdtgrep to grep and hash FDTs Add tests for fdtgrep .gitignore | 1 + Documentation/manual.txt | 32 +++ Makefile | 5 + Makefile.utils | 6 + README | 9 + fdtgrep.c | 554 +++++++++++++++++++++++++++++++++++++++ libfdt/Makefile.libfdt | 2 +- libfdt/fdt_region.c | 369 ++++++++++++++++++++++++++ libfdt/libfdt.h | 223 +++++++++++++++- libfdt/meson.build | 1 + libfdt/version.lds | 2 + tests/grep.dts | 23 ++ tests/run_tests.sh | 111 +++++++- tests/testutils.sh | 1 + 14 files changed, 1336 insertions(+), 3 deletions(-) create mode 100644 fdtgrep.c create mode 100644 libfdt/fdt_region.c create mode 100644 tests/grep.dts -- 2.34.0.rc0.344.g81b53c2807-goog
