Re: [BUG] git diff --no-index segfaults on large files (NULL object database)

From: Tian Yuchen <cat@malon.dev>
Date: Apr 4, 2026, 16:45 UTC
Message-ID: <d6c63949-1998-4cde-8cb0-902fd7db988c@malon.dev>
In-Reply-To: <CAO0HQ0X_pQmew5tJReOL=u+CMxCjAQynx8JfjykoYAUE59YNzw@mail.gmail.com>

On 4/4/26 18:39, Luca Stefani wrote:

Show 108 quoted lines

> Thank you for filling out a Git bug report!
> Please answer the following questions to help us understand your issue.
> 
> What did you do before the bug happened? (Steps to reproduce your issue)
> 
> `git diff --color -- file1 file2` SIGSEGV in case file1 and file2
> are "big enough", that is when the file size is bigger than
> repo_settings_get_big_file_threshold()
> 
> What did you expect to happen? (Expected behavior)
> 
> Not a crash, it to say "binary files are different"
> 
> What happened instead? (Actual behavior)
> 
> Program received signal SIGSEGV, Segmentation fault.
> index_fd (istate=istate@entry=0x5555559ccb40,
> oid=oid@entry=0x5555559d3c80, fd=4, st=st@entry=0x7fffffffc5c0,
> type=type@entry=OBJ_BLOB, path=path@entry=0x5555559d3ce0
> "4/root_part", flags=0) at /usr/src/debug/git/git/object-file.c:1634
> 1634                    transaction =
> odb_transaction_begin(the_repository->objects);
> (gdb) bt
> #0  index_fd (istate=istate@entry=0x5555559ccb40,
> oid=oid@entry=0x5555559d3c80, fd=4, st=st@entry=0x7fffffffc5c0,
> type=type@entry=OBJ_BLOB, path=path@entry=0x5555559d3ce0
> "4/root_part", flags=0)
>     at /usr/src/debug/git/git/object-file.c:1634
> #1  0x000055555579ab75 in index_path (istate=0x5555559ccb40,
> oid=0x5555559d3c80, path=0x5555559d3ce0 "4/root_part",
> st=0x7fffffffc5c0, flags=0) at
> /usr/src/debug/git/git/object-file.c:1658
> #2  0x0000555555720f83 in diff_fill_oid_info (one=0x5555559d3c80,
> istate=0x5555559ccb40) at /usr/src/debug/git/git/diff.c:4690
> #3  diff_fill_oid_info (one=one@entry=0x5555559d3c80,
> istate=0x5555559ccb40) at /usr/src/debug/git/git/diff.c:4679
> #4  0x0000555555724aba in run_diff (p=0x5555559ce900,
> o=0x7fffffffd188) at /usr/src/debug/git/git/diff.c:4738
> #5  diff_flush_patch (p=0x5555559ce900, o=0x7fffffffd188) at
> /usr/src/debug/git/git/diff.c:6247
> #6  0x0000555555729850 in diff_flush_patch_all_file_pairs
> (o=0x7fffffffd188) at /usr/src/debug/git/git/diff.c:6802
> #7  diff_flush (options=<optimized out>) at /usr/src/debug/git/git/diff.c:6942
> #8  0x00005555555b0d34 in diff_no_index (revs=0x7fffffffcbd0,
> algop=0x55555598dbd0 <hash_algos+112>, implicit_no_index=<optimized
> out>, argc=2, argv=0x5555559ce2a0) at
> /usr/src/debug/git/git/diff-no-index.c:427
> #9  cmd_diff (argc=<optimized out>, argv=0x5555559ce2a0,
> prefix=<optimized out>, repo=<optimized out>) at builtin/diff.c:516
> #10 0x000055555555fe05 in run_builtin (p=0x555555994ec8
> <commands.lto_priv+840>, argc=<optimized out>, argv=<optimized out>,
> repo=0x5555559a3c00 <the_repo.lto_priv>) at
> /usr/src/debug/git/git/git.c:506
> #11 handle_builtin (args=args@entry=0x7fffffffd980) at
> /usr/src/debug/git/git/git.c:779
> #12 0x00005555555612cc in run_argv (args=0x7fffffffd980) at
> /usr/src/debug/git/git/git.c:862
> #13 cmd_main (argc=<optimized out>, argv=<optimized out>) at
> /usr/src/debug/git/git/git.c:984
> #14 0x000055555555d794 in main (argc=6, argv=0x7fffffffdc58) at
> /usr/src/debug/git/git/common-main.c:9
> (gdb) p the_repository->objects
> $1 = (struct object_database *) 0x0
> (gdb) p istate->repo->objects
> $2 = (struct object_database *) 0x0
> (gdb) p istate->repo
> $3 = (struct repository *) 0x5555559a3c00 <the_repo.lto_priv>
> (gdb) p the_repository
> $4 = (struct repository *) 0x5555559a3c00 <the_repo.lto_priv>
> 
> The same issue happens in master, the same path is taken and when we
> finally get into
> `odb_transaction_begin` where odb is NULL, and reading ->transaction
> causes the illegal
> segment access once again.
> 
> Program received signal SIGSEGV, Segmentation fault.
> 0x00005555557c6283 in odb_transaction_begin (odb=0x0) at odb.c:1075
> 1075            if (odb->transaction)
> (gdb) bt
> #0  0x00005555557c6283 in odb_transaction_begin (odb=0x0) at odb.c:1075
> #1  0x00005555557b94f9 in index_fd (istate=0x555555a88b60,
> oid=0x555555a8fcb0, fd=4, st=0x7fffffffc730, type=OBJ_BLOB,
> path=0x555555a8fd10 "4/root_part", flags=0) at object-file.c:1665
> #2  0x00005555557b9651 in index_path (istate=0x555555a88b60,
> oid=0x555555a8fcb0, path=0x555555a8fd10 "4/root_part",
> st=0x7fffffffc730, flags=0) at object-file.c:1691
> #3  0x0000555555730937 in diff_fill_oid_info (one=0x555555a8fcb0,
> istate=0x555555a88b60) at diff.c:4699
> #4  0x0000555555730b81 in run_diff (p=0x555555a8a9a0,
> o=0x7fffffffd1a8) at diff.c:4747
> #5  0x00005555557361d5 in diff_flush_patch (p=0x555555a8a9a0,
> o=0x7fffffffd1a8) at diff.c:6258
> #6  0x0000555555737cae in diff_flush_patch_all_file_pairs
> (o=0x7fffffffd1a8) at diff.c:6813
> #7  0x00005555557382b2 in diff_flush (options=0x7fffffffd1a8) at diff.c:6953
> #8  0x00005555557228cc in diff_no_index (revs=0x7fffffffcbf0,
> algop=0x555555a49b50 <hash_algos+112>, implicit_no_index=1, argc=2,
> argv=0x555555a8a2f0) at diff-no-index.c:427
> #9  0x00005555555c04b1 in cmd_diff (argc=5, argv=0x555555a8a2f0,
> prefix=0x0, repo=0x0) at builtin/diff.c:516
> #10 0x0000555555574f47 in run_builtin (p=0x555555a4aaa8
> <commands+840>, argc=5, argv=0x555555a8a2f0, repo=0x555555a7e620
> <the_repo>) at git.c:506
> #11 0x000055555557544e in handle_builtin (args=0x7fffffffdab0) at git.c:780
> #12 0x0000555555575778 in run_argv (args=0x7fffffffdab0) at git.c:863
> #13 0x0000555555575c0f in cmd_main (argc=5, argv=0x7fffffffdc50) at git.c:984
> #14 0x00005555556a7414 in main (argc=6, argv=0x7fffffffdc48) at common-main.c:9

The problem is most likely to occur in object-file.c:

...
int index_fd(struct index_state *istate, struct object_id *oid,
	     int fd, struct stat *st,
	     enum object_type type, const char *path, unsigned flags)
{
	int ret;

	/*
	 * Call xsize_t() only when needed to avoid potentially unnecessary
	 * die() for large files.
	 */
	if (type == OBJ_BLOB && path && would_convert_to_git_filter_fd(istate, 
path)) {
		ret = index_stream_convert_blob(istate, oid, fd, path, flags);
	} else if (!S_ISREG(st->st_mode)) {
		ret = index_pipe(istate, oid, fd, type, path, flags);
	} else if ((st->st_size >= 0 &&
		    (size_t)st->st_size <= 
repo_settings_get_big_file_threshold(istate->repo)) ||
		   type != OBJ_BLOB ||
		   (path && would_convert_to_git(istate, path))) {
		ret = index_core(istate, oid, fd, xsize_t(st->st_size),
				 type, path, flags);
	} else {
		struct object_database *odb = the_repository->objects;
		struct odb_transaction_files *files_transaction;
		struct odb_transaction *transaction;

		transaction = odb_transaction_begin(odb);
		files_transaction = container_of(odb->transaction,
						 struct odb_transaction_files,
						 base);
		ret = index_blob_packfile_transaction(files_transaction, oid, fd,
						      xsize_t(st->st_size),
						      path, flags);
		odb_transaction_commit(transaction);
	}

	close(fd);
	return ret;
}
...

Note that in 'if (type == OBJ_BLOB && st->st_size > big_file_threshold)' block, there is no check to see if 'the_repository->objects' is NULL. It assumes that an object database is available, which is not true in no-repository scenarios.

One possible solution would be as follows: only when 'flags & HASH_WRITE_OBJECT' (or something like that) is true does this indicate that the file needs to be written in; if it is false only the hash value is required and the file should not be written in.

Will send a patch to fix it, soon.

Thanks, Yuchen