git/list[1] front-page[2] threads[3] people[4] search[5] about
 

[PATCH] history: do not dereference NULL when parent tree is missing

From
Zzkd18cjb@mail.ustc.edu.cn <zkd18cjb@mail.ustc.edu.cn>
Date
Sep 2, 2026, 12:07 UTC
Message-ID
<5438d465.ab31e.1a062047bd5.Coremail.zkd18cjb@mail.ustc.edu.cn>

write_ondisk_index() dereferences the return value of repo_parse_tree_indirect() unconditionally. If the parent commit's tree object is missing from the object store (corrupt repository, object removed by tooling, or incomplete restore), the function returns NULL and "git history split" crashes with a SIGSEGV (release build; UBSan reports a null-pointer member access at builtin/history.c:789).

Guard the parse result and error out gracefully, following the codebase convention for objects that cannot be loaded.

Signed-off-by: Jinbao Chen <zkd18cjb@mail.ustc.edu.cn>
---
Hi,

(This was reported via the Git security contact, which suggested posting here. The security team asked me to post the fix on this list, as the crash requires a missing object in a local repository and is not considered a security issue.)

"git history split" crashes with a SIGSEGV when the commit's parent tree object is missing from the object store (corrupt repository, object removed by tooling, incomplete backup/mirror restore): write_ondisk_index() dereferences the NULL return value of repo_parse_tree_indirect(). The fix below guards the parse result, matching the codebase convention for objects that cannot be loaded ("if (!tree) return error(...)").

Reproduction (verified on master @ f78ce2f7b6, x86-64 Linux):
    git init r && cd r
    git config user.email t@t && git config user.name t
    echo a > f && git add f && git commit -qm one
    echo b > f && git commit -qam two
    tree=$(git rev-parse 'HEAD^^{tree}')
    rm .git/objects/$(echo "$tree" | cut -c1-2)/$(echo "$tree" | cut -c3-)
    GIT_EDITOR=true git history split HEAD
Before: release build SIGSEGV (exit 139, core dumped); UBSan reports
"member access within null pointer of type 'struct tree'" at
builtin/history.c:789.
After: "error: unable to parse tree <oid>", exit 255, no crash.
Control (tree object present) is unchanged.
 1 file changed, 4 insertions(+)
diff --git a/builtin/history.c b/builtin/history.c
index 000155ad9c..097631f5ba 100644
--- a/builtin/history.c
+++ b/builtin/history.c
@@ -786,6 +786,10 @@ static int write_ondisk_index(struct repository *repo,
 	opts.dst_index = &index;
 
 	tree = repo_parse_tree_indirect(repo, oid);
+	if (!tree) {
+		ret = error(_("unable to parse tree %s"), oid_to_hex(oid));
+		goto out;
+	}
 	init_tree_desc(&tree_desc, &tree->object.oid, tree->buffer, tree->size);
 
 	if (unpack_trees(1, &tree_desc, &opts)) {
-- 
2.53.0
Next: Patrick Steinhardt
Message 1 of 4 in “history: do not dereference NULL when parent tree is missing”
  1. history: do not dereference NULL when parent tree is missingzkd18cjb@mail.ustc.edu.cn, Sep 2, 2026
  2. Patrick SteinhardtSep 3, 2026
  3. Jinbao ChenSep 3, 2026
  4. Patrick SteinhardtSep 3, 2026

Read the whole thread, see it on lore, or plain text.

$ cat FOOTERMessages come from the public archive at lore.kernel.org/git, fetched every hour. The front page is chosen and written each morning by an AI editor and can be wrong; the threads themselves are the record. About and API. For agents: an MCP server at https://gitlist.dev/mcp, and any thread, story or person page as Markdown by adding .md to its URL (or sending Accept: text/markdown). Details in /llms.txt.