dissect: add --manifest switch to generate a UAPI.16 manifest file fr…

[poettering]

…om a directory tree

The UAPI.16 spec is being discussed here:

uapi-group/specifications#213

[poettering]

Do not merge until the spec is also good to be merged.

[github-actions]

Claude review of PR #41990 (334eca8)

Must fix

• Unused variable escaped — src/dissect/dissect.c:1348 — declared but never used in manifest_print_item(), will fail with -Werror
• S_ISDIR(sx->stx_mask) should be S_ISDIR(sx->stx_mode) — src/dissect/dissect.c:1386 — uses wrong struct field for file type check, causing directories to incorrectly enter the hardlink processing path
• fd leak on lseek failure in rm_rf_children — src/shared/rm-rf.c:313-315 — when REMOVE_REWIND lseek fails, returns -errno without closing the fd, violating the function's ownership contract

Suggestions

• NULL from inode_type_to_string() — src/dissect/dissect.c:1397 — can return NULL for unknown inode types, producing a JSON null instead of a string
• name_to_handle_at_try_fid hard failure — src/dissect/dissect.c:1418 — non-fatal errors (e.g. -EOPNOTSUPP) abort the entire manifest, could gracefully skip hardlink detection instead
• Missing log on sd_json_buildo failure — src/dissect/dissect.c:1543 — error returned without logging, unlike the analogous call earlier in the file
• stx_size accessed without STATX_SIZE mask check — src/dissect/dissect.c:1543 — in the S_IFREG case, sx->stx_size is used after only checking STATX_TYPE, inconsistent with mtree_print_item which checks both
• --mtree-hash flag reused for manifest — src/dissect/dissect.c:1453 — the flag name and description reference mtree output only, but it now also controls manifest hashing
• get_entry() helper has no failure guard — test/units/TEST-50-DISSECT.dissect.sh:1143 — if no entry matches, produces empty output leading to confusing assertion failures
• --manifest on image only checks exit code — test/units/TEST-50-DISSECT.dissect.sh:1219 — output is discarded; consider verifying non-empty output or expected content

Nits

• Double blank line — src/dissect/dissect.c:1380-1382 — extra blank line between break; and case S_IFLNK:
• Misleading error message — src/dissect/dissect.c:1427 — says "Failed to symlink" but the operation is readlinkat_malloc on the hardlink database
• Pointer star style — src/dissect/dissect.c:1383 — const char* type should be const char *type
• Operator spacing — src/dissect/dissect.c:1475 — colon +1 should be colon + 1
• Hardcoded SHA256 for empty file — test/units/TEST-50-DISSECT.dissect.sh:1162 — could compute dynamically like done for regular.txt
• Hardcoded symlink size — test/units/TEST-50-DISSECT.dissect.sh:1179 — "11" is strlen of target, would be more maintainable computed
• Missing comma in man page — man/systemd-dissect.xml:272 — "Like --mtree this" should be "Like --mtree, this"
• FID error message missing path — src/dissect/dissect.c:1395 — doesn't include the file path for debugging
• Cycle avoidance startswith too broad — src/dissect/dissect.c:1525 — startswith(path, "Uapi16Manifest.") also skips non-manifest files

Workflow run