Re: [PATCH 1/1] pathspec: fix memleak
- From
Junio C Hamano <gitster@pobox.com>
- Date
- Aug 12, 2022, 17:41 UTC
- Message-ID
- <xmqqsfm1bbrf.fsf@gitster.g>
- In-Reply-To
- <20220812081744.456280-2-anthony.2lannoy@gmail.com>
Anthony Delannoy <anthony.2lannoy@gmail.com> writes:
Show 13 quoted lines
> diff --git a/preload-index.c b/preload-index.c
> index e5529a5863..a05f4d1390 100644
> --- a/preload-index.c
> +++ b/preload-index.c
> @@ -148,6 +148,9 @@ void preload_index(struct index_state *index,
> if (pthread_join(p->pthread, NULL))
> die("unable to join threaded lstat");
> t2_sum_lstat += p->t2_nr_lstat;
> +
> + if (pathspec)
> + free(p->pathspec.items);
> }
> stop_progress(&pd.progress);Given the way how copy_pathspec() makes a deep copy of a pathspec, I suspect that this is still leaking all the resources held by the array that is freed here. Let's take a look:
void copy_pathspec(struct pathspec *dst, const struct pathspec *src)
{
int i, j; *dst = *src;
ALLOC_ARRAY(dst->items, dst->nr);
COPY_ARRAY(dst->items, src->items, dst->nr);Here, we copy the array of "struct pathspec_item". But that is not enough because ...
for (i = 0; i < dst->nr; i++) {
struct pathspec_item *d = &dst->items[i];
struct pathspec_item *s = &src->items[i]; d->match = xstrdup(s->match);
d->original = xstrdup(s->original);... each "struct pathspec_item" instance has pointer members like these, and the copying of the array made these strings shared between the src and dst arrays. Here we make a copy of the string owned by the element in the src array and give the copy to the element in the dst array.
ALLOC_ARRAY(d->attr_match, d->attr_match_nr);
COPY_ARRAY(d->attr_match, s->attr_match, d->attr_match_nr);Likewise for a separate array pointed by a member in "struct pathspec_item" ...
for (j = 0; j < d->attr_match_nr; j++) {
const char *value = s->attr_match[j].value;
d->attr_match[j].value = xstrdup_or_null(value);... which has a pointer member here ...
}
d->attr_check = attr_check_dup(s->attr_check);
... and here. Both are deep-copied.
}
}There is pathspec.c::clear_pathspec() API function, which looks as if it was made for this exact use case.
I wonder if this is a good place to use it, perhaps like the attached patch.
preload-index.c | 6 ++++++ 1 file changed, 6 insertions(+)
diff --git c/preload-index.c w/preload-index.c index e5529a5863..100f7a374d 100644 --- c/preload-index.c +++ w/preload-index.c @@ -151,6 +151,12 @@ void preload_index(struct index_state *index, } stop_progress(&pd.progress); + if (pathspec) { + /* earlier we made deep copies for each thread to work with */ + for (i = 0; i < threads; i++) + clear_pathspec(&data[i].pathspec); + } + trace_performance_leave("preload index"); trace2_data_intmax("index", NULL, "preload/sum_lstat", t2_sum_lstat);