Split BPFMap API into managed and unmanaged

[geyslan]

Close: #354

commit 9cb8ba9 (HEAD -> map-effort, origin/map-effort)
Author: Geyslan Gregório [email protected]
Date: Mon Aug 7 17:38:03 2023 -0300

feat(map): introduce ReuseFD() to reuse a map fd

The `ReuseFD` function links the current BPFMap instance to a provided
map file descriptor. This enables the reuse of a map that was initiated
by another process.

Initial effort: #346

Co-authored-by: Francisco Javier Honduvilla Coto <[email protected]>

commit fe71839
Author: Geyslan Gregório [email protected]
Date: Wed Aug 9 16:46:41 2023 -0300

chore(map): logic refactoring and standardization

Introduced struct helpers to manage the C struct lifecycle on the C side,
which avoids problems with structs which may contain bitfields. See #244.

  These are replacements for 'BPFMapBatchOpts' and 'bpfMapBatchOptsToC()'.
  - cgo_bpf_map_batch_opts_new
  - cgo_bpf_map_batch_opts_free

  These are replacements for 'bpfMapCreateOptsToC()'
  - cgo_bpf_map_create_opts_new
  - cgo_bpf_map_create_opts_free

Restructured libbpf BPF map logic into separate files:
- 'map-common.go' contains generic BPF map types, constants, and helpers
  like 'GetMapInfoByFD()'.
- 'map-low.go' includes 'BPFMapLow' and respective logic and helpers
  like 'CreateMap()'.
- 'map-iterator.go' contains related types and logic.

In the 'BPFMap' struct:
- Removed fields like 'fd' and 'name' since they must be retrieved
  directly from 'libbpf'.
- Added 'bpfMapLow' instance for access to low-level methods.
- Deprecated 'BPFMap.GetMaxEntries' in favor of using 'MaxEntries'.
- Exposed 'InitialValue' and 'SetInitialValue' as public wrappers.
- Introduced MapFlags(), IfIndex() and MapExtra() methods.

Introduced 'misc.go' as a place for miscellaneous generic helpers.

Bug Fixes:
- Fixed cases where 'BPFMap.ValueSize()', possibly 0, was being passed
  directly to 'make()' - detected on the map of maps effort #343.

Further Refactoring and Standardization on the map related logic and on
lines touched by other changes:
- Removed check for 'BPFMap.Name()' against nil, as 'C.GoString(C.NULL)'
  returns "".
- Changed error returns to 'retC' in functions that return error values
  to avoid confusion with possible use of errno.
- Changed returns to 'valueC' in functions that return values or
  pointers to values.
- Suffixed variable names that are C types with 'C'.
- Applied the use of 'defer' for better resource management.

commit 7b73fc0
Author: Geyslan Gregório [email protected]
Date: Wed Aug 9 15:42:49 2023 -0300

fix(c): check against NULL before memcpy

commit 8b1190a
Author: Geyslan Gregório [email protected]
Date: Wed Aug 9 14:17:13 2023 -0300

chore(c): prefix helpers with 'cgo_'

Prefix all C helpers used in Go with 'cgo_', renaming two cases
where the functions does not reflect the libbpf API:

set_print_fn -> cgo_libbpf_set_print_fn
get_internal_map_init_value -> cgo_bpf_map__initial_value

commit f58df61
Author: Geyslan Gregório [email protected]
Date: Wed Aug 9 14:09:17 2023 -0300

chore(c): move C definitions to 'libbpfgo.c'

Leaving 'libbpfgo.h' only with declarations to avoid multiple
declaration errors when importing 'libbpfgo.h' from different Go files.

commit c9b7e75
Author: Geyslan Gregório [email protected]
Date: Wed Aug 9 13:53:53 2023 -0300

chore(map): relocate map logic to map.go

commit d45e8f6
Author: Geyslan Gregório [email protected]
Date: Tue Aug 8 16:12:26 2023 -0300

chore: del unneeded assignment to blank identifier

[javierhonduco]

Thanks for doing this! Tested func (m *Module) GetMap() and things seem to work fine!

[geyslan]

Since this involved creating new files and relocating logic, it required making major changes in a single commit.

[geyslan]

When this get merged, the next to come to light is the map of maps.

Despite the discussion in #343 (comment), I'm not so sure about creating another type like BPFMapOfMaps, as that would require another module method like GetMapOfMaps(). Perhaps best is to leave BPFMap and BPFMapLow with the same libbpf APIs and make usage assumptions based on respective selftest. Let's see.

[geyslan]

Since this involved creating new files and relocating logic, it required making major changes in a single commit.

I'm going redo the bigger commit splitting it by subject for better understanding and review.

[geyslan]

I'm going redo the bigger commit splitting it by subject for better understanding and review.

@yanivagman I've broken it down a bit more and left GetMapByID() and GetMapFDByID() implementations for an upcoming PR (map of maps). ReuseFD(), co-authored by @javierhonduco, is coming with this since it's a good example of map high and low level separation.

[yanivagman]

Overall LGTM,
The last commit (ReuseFD) should be feat and not chore,
Also two small questions in the following comments

[geyslan]

Overall LGTM, The last commit (ReuseFD) should be feat and not chore, Also two small questions in the following comments

Thanks for reviewing this. I'm going to address all requests and bugs.