cache-tree: fix verify_cache() to catch non-adjacent D/F conflicts

verify_cache() checks that the index does not contain both "path" and
"path/file" before writing a tree.  It does this by comparing only
adjacent entries, relying on the assumption that "path/file" would
immediately follow "path" in sorted order.  Unfortunately, this
assumption does not always hold.  For example:

    docs                     <-- submodule entry
    docs-internal/README.md  <-- intervening entry
    docs/requirements.txt    <-- D/F conflict, NOT adjacent to "docs"

When this happens, verify_cache() silently misses the D/F conflict and
write-tree produces a corrupt tree object containing duplicate entries
(one for the submodule "docs" and one for the tree "docs").

I could not find any caller in current git that both allows the index to
get into this state and then tries to write it out without doing other
checks beyond the verify_cache() call in cache_tree_update(), but
verify_cache() is documented as a safety net for preventing corrupt
trees and should actually provide that guarantee.  A downstream consumer
that relied solely on cache_tree_update()'s internal checking via
verify_cache() to prevent duplicate tree entries was bitten by the gap.

Add a test that constructs a corrupt index directly (bypassing the D/F
checks in add_index_entry) and verifies that write-tree now rejects it.

Signed-off-by: Elijah Newren <newren@gmail.com>

Author: newren

cache-tree.c

@@ -161,6 +161,54 @@ void cache_tree_invalidate_path(struct index_state *istate, const char *path)
 		istate->cache_changed |= CACHE_TREE_CHANGED;
 }
 
+/*
+ * Check whether this_ce and the next entry in the index form a D/F
+ * conflict ("path" vs "path/file").  Returns the conflicting "path/..."
+ * name when one is found, or NULL otherwise.
+ *
+ * The cache is sorted, so "path/file" sorts after "path" and the
+ * conflict is usually visible as adjacent entries.  But other entries
+ * can sort between them -- e.g. "path-internal" sits between "path"
+ * and "path/file" because '-' (0x2D) precedes '/' (0x2F) -- so when
+ * the immediately following entry shares our prefix but starts with a
+ * character that sorts before '/', binary search for "path/" instead.
+ */
+static const char *find_df_conflict(struct index_state *istate,
+				    const struct cache_entry *this_ce,
+				    const struct cache_entry *next_ce)
+{
+	const char *this_name = this_ce->name;
+	const char *next_name = next_ce->name;
+	int this_len = ce_namelen(this_ce);
+	const struct cache_entry *other;
+	struct strbuf probe = STRBUF_INIT;
+	int pos;
+
+	if (this_len >= ce_namelen(next_ce) ||
+	    next_name[this_len] > '/' ||
+	    strncmp(this_name, next_name, this_len))
+		return NULL;
+
+	if (next_name[this_len] == '/')
+		return next_name;
+
+	strbuf_add(&probe, this_name, this_len);
+	strbuf_addch(&probe, '/');
+	pos = index_name_pos_sparse(istate, probe.buf, probe.len);
+	strbuf_release(&probe);
+
+	if (pos < 0)
+		pos = -pos - 1;
+	if (pos >= (int)istate->cache_nr)
+		return NULL;
+	other = istate->cache[pos];
+	if (ce_namelen(other) > this_len &&
+	    other->name[this_len] == '/' &&
+	    !strncmp(this_name, other->name, this_len))
+		return other->name;
+	return NULL;
+}
+
 static int verify_cache(struct index_state *istate, int flags)
 {
 	unsigned i, funny;
@@ -190,24 +238,18 @@ static int verify_cache(struct index_state *istate, int flags)
 	 */
 	funny = 0;
 	for (i = 0; i + 1 < istate->cache_nr; i++) {
-		/* path/file always comes after path because of the way
-		 * the cache is sorted.  Also path can appear only once,
-		 * which means conflicting one would immediately follow.
-		 */
 		const struct cache_entry *this_ce = istate->cache[i];
 		const struct cache_entry *next_ce = istate->cache[i + 1];
-		const char *this_name = this_ce->name;
-		const char *next_name = next_ce->name;
-		int this_len = ce_namelen(this_ce);
-		if (this_len < ce_namelen(next_ce) &&
-		    next_name[this_len] == '/' &&
-		    strncmp(this_name, next_name, this_len) == 0) {
+		const char *conflict_name;
+
+		conflict_name = find_df_conflict(istate, this_ce, next_ce);
+		if (conflict_name) {
 			if (10 < ++funny) {
 				fprintf(stderr, "...\n");
 				break;
 			}
 			fprintf(stderr, "You have both %s and %s\n",
-				this_name, next_name);
+				this_ce->name, conflict_name);
 		}
 	}
 	if (funny)

t/meson.build

@@ -124,6 +124,7 @@ integration_tests = [
   't0090-cache-tree.sh',
   't0091-bugreport.sh',
   't0092-diagnose.sh',
+  't0093-verify-cache-df-gap.sh',
   't0095-bloom.sh',
   't0100-previous.sh',
   't0101-at-syntax.sh',

t/t0093-direct-index-write.pl

@@ -0,0 +1,38 @@
+#!/usr/bin/perl
+#
+# Build a v2 index file from entries listed on stdin.
+# Each line: "octalmode hex-oid name"
+# Output: binary index written to stdout.
+#
+# This bypasses all D/F safety checks in add_index_entry(), simulating
+# what happens when code uses ADD_CACHE_JUST_APPEND to bulk-load entries.
+use strict;
+use warnings;
+use Digest::SHA qw(sha1 sha256);
+
+my $hash_algo = $ENV{'GIT_DEFAULT_HASH'} || 'sha1';
+my $hash_func = $hash_algo eq 'sha256' ? \&sha256 : \&sha1;
+
+my @entries;
+while (my $line = <STDIN>) {
+	chomp $line;
+	my ($mode, $oid_hex, $name) = split(/ /, $line, 3);
+	push @entries, [$mode, $oid_hex, $name];
+}
+
+my $body = "DIRC" . pack("NN", 2, scalar @entries);
+
+for my $ent (@entries) {
+	my ($mode, $oid_hex, $name) = @{$ent};
+	# 10 x 32-bit stat fields (zeroed), with mode in position 7
+	my $stat = pack("N10", 0, 0, 0, 0, 0, 0, oct($mode), 0, 0, 0);
+	my $oid = pack("H*", $oid_hex);
+	my $flags = pack("n", length($name) & 0xFFF);
+	my $entry = $stat . $oid . $flags . $name . "\0";
+	# Pad to 8-byte boundary
+	while (length($entry) % 8) { $entry .= "\0"; }
+	$body .= $entry;
+}
+
+binmode STDOUT;
+print $body . $hash_func->($body);

t/t0093-verify-cache-df-gap.sh

@@ -0,0 +1,59 @@
+#!/bin/sh
+
+test_description='verify_cache() must catch non-adjacent D/F conflicts
+
+Ensure that verify_cache() can complain about bad entries like:
+
+  docs               <-- submodule
+  docs-internal/...  <-- sorts here because "-" < "/"
+  docs/...           <-- D/F conflict with "docs" above, not adjacent
+
+In order to test verify_cache, we directly construct a corrupt index
+(bypassing the D/F safety checks in add_index_entry) and verify that
+write-tree rejects it.
+'
+
+. ./test-lib.sh
+
+if ! test_have_prereq PERL
+then
+	skip_all='skipping verify_cache D/F tests; Perl not available'
+	test_done
+fi
+
+# Build a v2 index from entries on stdin, bypassing D/F checks.
+# Each line: "octalmode hex-oid name" (entries must be pre-sorted).
+build_corrupt_index () {
+	perl "$TEST_DIRECTORY/t0093-direct-index-write.pl" >"$1"
+}
+
+test_expect_success 'setup objects' '
+	test_commit base &&
+	BLOB=$(git rev-parse HEAD:base.t) &&
+	SUB_COMMIT=$(git rev-parse HEAD)
+'
+
+test_expect_success 'adjacent D/F conflict is caught by verify_cache' '
+	cat >index-entries <<-EOF &&
+	0160000 $SUB_COMMIT docs
+	0100644 $BLOB docs/requirements.txt
+	EOF
+	build_corrupt_index .git/index <index-entries &&
+
+	test_must_fail git write-tree 2>err &&
+	test_grep "You have both docs and docs/requirements.txt" err
+'
+
+test_expect_success 'non-adjacent D/F conflict is caught by verify_cache' '
+	cat >index-entries <<-EOF &&
+	0160000 $SUB_COMMIT docs
+	0100644 $BLOB docs-internal/README.md
+	0100644 $BLOB docs/requirements.txt
+	EOF
+	build_corrupt_index .git/index <index-entries &&
+
+	test_must_fail git write-tree 2>err &&
+	test_grep "You have both docs and docs/requirements.txt" err
+'
+
+test_done