rsync: fix reallocarray() usage in blkhash_set()

The well-named ERR() macro doesn't error out. Therefore an incorrect use
of reallocarray() is actually a leak that is easily overlooked. Do it the
right way by assigning to a temporary variable and preserve behavior by
freeing and NULL-ing.

ok claudio

diff --git a/usr.bin/rsync/blocks.c b/usr.bin/rsync/blocks.c
index b300e41..7b774e6 100644 (file)
--- a/usr.bin/rsync/blocks.c
+++ b/usr.bin/rsync/blocks.c
@@ -1,4 +1,4 @@
-/*     $OpenBSD: blocks.c,v 1.24 2024/09/18 10:22:36 job Exp $ */
+/*     $OpenBSD: blocks.c,v 1.25 2024/09/27 13:06:21 tb Exp $ */
 /*
  * Copyright (c) 2019 Kristaps Dzonsons <kristaps@bsd.lv>
  *
@@ -88,7 +88,8 @@ blkhash_alloc(void)
 int
 blkhash_set(struct blktab *p, const struct blkset *bset)
 {
-       size_t   i, idx;
+       struct blkhash  *blks;
+       size_t           i, idx;
 
        if (bset == NULL)
                return 1;
@@ -100,11 +101,14 @@ blkhash_set(struct blktab *p, const struct blkset *bset)
 
        /* Fill in the hashtable. */
 
-       p->blks = reallocarray(p->blks, bset->blksz, sizeof(struct blkhash));
-       if (p->blks == NULL) {
+       blks = reallocarray(p->blks, bset->blksz, sizeof(struct blkhash));
+       if (blks == NULL) {
                ERR("reallocarray");
+               free(p->blks);
+               p->blks = NULL;
                return 0;
        }
+       p->blks = blks;
        for (i = 0; i < bset->blksz; i++) {
                p->blks[i].blk = &bset->blks[i];
                idx = bset->blks[i].chksum_short % p->qsz;