libcare-ctl.rst

The Doctor: libcare-ctl

Detailed description is in the internals.

All the job is done by the libcare-ctl. It is called doctor hereafter and the targets of operations are thus called patients.

The doctor accepts a few arguments that are common for all types of operations:

-v	enable verbose output
-h	show commands list

Applying patches via patch

The patch mode patches a process with ID given as an argument to -p option or all of them except self and init when the argument is all. The patch (or directory with patches) to be applied should be specified as the only positional argument:

$ libcare-ctl patch -p <PID_or_all> some_patch_file.kpatch

The patches are basically ELF files of relocatable type REL with binary meta-information such as BuildID and name of the patch target prepended. Loading patches is thus a lot like loading a shared object (library) into a process. Except we are puppeting it by strings going through a keyhole in other process' memory.

First, the memory near the original object is allocated, then all the relocations and symbols are resolved in a local copy of patch content. This pre-baked patch is copied to the patient's memory and, finally, original functions are overwritten with the unconditional jumps to the patched version.

For more details look at the Patching.

Cancelling patches via unpatch

The unpatch mode makes doctor remove patches listed by target BuildID from the patients' memory. It simply restores the original code of the patched functions from a stash allocated along with the patch and puppets patients to munmap the memory areas used by patches.

Showing info via info

The last entry to the libcare-ctl is the info command that lists all the objects and their BuildIDs for the set of the processes requested. Its primary use is as the utility for the book-keeping software.

Patchlevel support

Since patches to the objects such as libraries can be updated, there is a way to distinguish them, called patchlevel. This information is parsed from the layout of the directory where the patches are stored. If on patching stage a patch with a bigger patchlevel is found, the old one is removed and the new one is applied.