[Feature resurrection] pkg check -r, recompute the checksum

[Emrion]

Hello,

Would it be possible to enable anew the recompute feature? I agree it's not often used, but when we need it.... We need it.

Unless there are others reasons like a security concern or too time-consuming to rewrite for the version 2?

[bapt]

for what use case would it be useful for?

[Emrion]

The general use case is: one modifies an installed software. There could be many reasons.

Recent example: I asked a new feature for the mail server mox and the author did it. Up to now, he doesn't make a new release and the FreeBSD port stays without this feature. I just recompiled the executable and replaced it. Now, pkg complains at each check.

[Emrion]

I worked to modify the code of the release-2.1 branch and it seems I succeded. I would like to have your opinion on the pkg_recompute function. I looked at your codes, past and present, and tried to do something that works. I haven't programmed in C for some time. Something may escaped me.

I changed the following files: libpkg/libpkg.ver, libpkg/pkg.c, libpkg/pkg.h and src/check.c. If you're ok, I can make a patch (adapted to the main branch of course).

int
pkg_recompute(struct pkgdb *db, struct pkg *pkg)
{
	struct pkg_file *f = NULL;
	int64_t flatsize = 0;
	struct stat st;
	pkg_checksum_type_t type;
	char *newsum;
	int rc = EPKG_OK;

	while (pkg_files(pkg, &f) == EPKG_OK) {

		if (lstat(f->path, &st) == -1) 
			continue;
		if (!S_ISLNK(st.st_mode)) 
			flatsize += st.st_size;

		type = pkg_checksum_file_get_type(f->sum, strlen(f->sum));
		if (type == PKG_HASH_TYPE_UNKNOWN) 
			type = PKG_HASH_TYPE_SHA256_HEX;
		newsum = pkg_checksum_generate_file(f->path, type);
		if (newsum == NULL) {
			rc = EPKG_FATAL;
			break;
		}
		if (strcmp(newsum, f->sum) != 0) 
			pkgdb_file_set_cksum(db, f, newsum);
		free(newsum);
	}

	if (rc == EPKG_OK && flatsize != pkg->flatsize)
		pkg->flatsize = flatsize;
	return (rc);
}

Edit: changed the pkg_recompute function. The first version worked but didn't prefix the hash with the used pkg_checksum_type. I should have used pkg_checksum_generate_file instead of pkg_checksum_symlinkat / pkg_checksum_fileat. This is what this new version does.

[bapt]

This is not complicated to resurect the feature, but I think this feature should just not exists. if one it modifying a packages, imho he should just create a package for the new thing and install that version rather than faking the currently installed package. This is the reason why I removed the feature. The other reason is in environement like PCIDSS and things like that, the availability of the feature make the auditors nervous.

[Emrion]

You're the best placed to know what pkg should and shouldn't do.

Just a remark: it was complicated for me! I never looked at the pkg code before and I'm far to understand all its logics and mecanisms.

Will pkg_recompute work well? I mean, do I overlooked an important part of the internal logic behind the checksum?

At the base, I wanted to write an utility rather than ask for a feature. But I failed to use /usr/local/include/pkg.h to work on checksum computation. Some parts of libpkg are "private"... [edited] That said, conversely at what I believed, there are many functions (all?) in libpkg.so. Maybe it's possible...

[AUnixman]

@bapt please see https://forums.freebsd.org/threads/package-checksum-errors.98857/ at your convenience.

[bapt]

this is a good reason not to readd the recompute, it does not fix the checksums! if the checksums are invalid it means there is a really issue to fix, why did it happen, is there an issue with the package, etc.

[bapt]

I ll try to have a look at what s wrong with caja in the next couple of days

[bapt]

fixed in ports

[AUnixman]

Thank you. The mime mess on mate-control-center is also FUBAR immediately after install / update: mate-control-center-1.28.1_3: checksum mismatch for /usr/local/share/applications/mimeinfo.cache

[bapt]

another packaging issue now also fixed.