readelf(1): fix -wR option, memory leaks, and -wf minor bug

[christos]

Summary

Fixed broken -wR option, and various memory leaks. There's also
a memory leak in search_loclist_at() (-wo option) which I haven't
been able to fix.

[emaste]

Is there a PR or other reference?

[christos]

In D38419#874131, @emaste wrote:

Is there a PR or other reference?

Nothing that I can find after a quick search.

[markj]

In D38419#874131, @emaste wrote:

Is there a PR or other reference?

For context, Christos is working on a libdwarf program which enumerates inline function call locations. The eventual goal is to embed that functionality into libdtrace, so that we can use the kinst provider to probe inline function calls and returns.

[markj]

contrib/elftoolchain/readelf/readelf.c

5193

Are we still leaking the DIE in these error cases?

[christos]

contrib/elftoolchain/readelf/readelf.c

5193

Are we still leaking the DIE in these error cases?

No, as far as I know, deallocating the srcfiles with DW_DLA_STRING is a NOP in FreeBSD's implementation. Libdwarf takes care of this.

[christos]

contrib/elftoolchain/readelf/readelf.c

5193

Scratch that, I thought you meant something else.
Well, in this case, perhaps we can try deallocating the DIE before continuing.

[christos]

Fixed a few more DIE leaks.

[markj]

contrib/elftoolchain/readelf/readelf.c

4942

Where does this DIE get freed in the non-error case?

5201

Note that these continues apply to the inner for loop, so there's no leak. It's not clear to me that breaking out of the inner loop is the right thing to do?

[christos]

contrib/elftoolchain/readelf/readelf.c

5201

You're right. The dwarf_dealloc() should apply only to the outer loop's continues.

[emaste]

Looks reasonable to me