[dm-devel] [PATCH v2] libmultipath: fix memory leaks in coalesce_paths
Martin Wilck
mwilck at suse.com
Sun Nov 1 21:23:27 UTC 2020
Hi lixiaokeng,
On Thu, 2020-10-22 at 15:28 +0800, lixiaokeng wrote:
> When multipath -F are executed first and multipath -v2 or
> -d are executed later, asan will warn memory leaks. The
> reason is that the mpp allocated in coalesce_paths isn't
> freed. Here we use newmp to store mpp. If newmp need not
> be copied to mpvec, we free newmp at the end of the func.
>
> Signed-off-by: Lixiaokeng <lixiaokeng at huawei.com>
> Signed-off-by: Zhiqiang Liu <liuzhiqiang26 at huawei.com>
> Signed-off-by: Linfeilong <linfeilong at huawei.com>
Thanks. Please see below.
> ---
> libmultipath/configure.c | 66 +++++++++++++++++++++++++-------------
> --
> 1 file changed, 41 insertions(+), 25 deletions(-)
>
> diff --git a/libmultipath/configure.c b/libmultipath/configure.c
> index 6fb477fc..9d6eeba1 100644
> --- a/libmultipath/configure.c
> +++ b/libmultipath/configure.c
> @@ -1132,7 +1132,7 @@ out:
> * FORCE_RELOAD_WEAK: existing maps are compared to the current conf
> and only
> * reloaded in DM if there's a difference. This is useful during
> startup.
> */
> -int coalesce_paths (struct vectors * vecs, vector newmp, char *
> refwwid,
> +int coalesce_paths (struct vectors *vecs, vector mpvec, char
> *refwwid,
> int force_reload, enum mpath_cmds cmd)
> {
> int ret = CP_FAIL;
> @@ -1144,10 +1144,20 @@ int coalesce_paths (struct vectors * vecs,
> vector newmp, char * refwwid,
> struct path * pp2;
> vector curmp = vecs->mpvec;
> vector pathvec = vecs->pathvec;
> + vector newmp = NULL;
This initialization is not necessary.
> struct config *conf;
> int allow_queueing;
> struct bitfield *size_mismatch_seen;
>
> + if (mpvec)
> + newmp = mpvec;
> + else
> + newmp = vector_alloc();
> + if (!newmp) {
> + condlog(0, "can not allocate newmp");
> + return ret;
> + }
> +
> /* ignore refwwid if it's empty */
> if (refwwid && !strlen(refwwid))
> refwwid = NULL;
> @@ -1270,8 +1280,14 @@ int coalesce_paths (struct vectors * vecs,
> vector newmp, char * refwwid,
> goto out;
> }
> }
> - if (r == DOMAP_DRY)
> + if (r == DOMAP_DRY) {
> + if (!vector_alloc_slot(newmp)) {
> + remove_map(mpp, vecs->pathvec, vecs-
> >mpvec, KEEP_VEC);
> + goto out;
> + }
> + vector_set_slot(newmp, mpp);
> continue;
> + }
>
> if (r == DOMAP_EXIST && mpp->action == ACT_NOTHING &&
> force_reload == FORCE_RELOAD_WEAK)
> @@ -1307,44 +1323,44 @@ int coalesce_paths (struct vectors * vecs,
> vector newmp, char * refwwid,
> print_multipath_topology(mpp, verbosity);
> }
>
> - if (newmp) {
> - if (mpp->action != ACT_REJECT) {
> - if (!vector_alloc_slot(newmp))
> - goto out;
> - vector_set_slot(newmp, mpp);
> + if (mpp->action != ACT_REJECT) {
> + if (!vector_alloc_slot(newmp)) {
> + remove_map(mpp, vecs->pathvec, vecs-
> >mpvec, KEEP_VEC);
> + goto out;
> }
> - else
> - remove_map(mpp, vecs->pathvec, vecs-
> >mpvec,
> - KEEP_VEC);
> + vector_set_slot(newmp, mpp);
> }
> + else
> + remove_map(mpp, vecs->pathvec, vecs->mpvec,
> + KEEP_VEC);
> }
> /*
> * Flush maps with only dead paths (ie not in sysfs)
> * Keep maps with only failed paths
> */
> - if (newmp) {
> - vector_foreach_slot (newmp, mpp, i) {
> - char alias[WWID_SIZE];
> + vector_foreach_slot (newmp, mpp, i) {
> + char alias[WWID_SIZE];
>
> - if (!deadmap(mpp))
> - continue;
> + if (!deadmap(mpp))
> + continue;
>
> - strlcpy(alias, mpp->alias, WWID_SIZE);
> + strlcpy(alias, mpp->alias, WWID_SIZE);
>
> - vector_del_slot(newmp, i);
> - i--;
> - remove_map(mpp, vecs->pathvec, vecs->mpvec,
> KEEP_VEC);
> + vector_del_slot(newmp, i);
> + i--;
> + remove_map(mpp, vecs->pathvec, vecs->mpvec, KEEP_VEC);
>
> - if (dm_flush_map(alias))
> - condlog(2, "%s: remove failed (dead)",
> - alias);
> - else
> - condlog(2, "%s: remove (dead)", alias);
> - }
> + if (dm_flush_map(alias))
> + condlog(2, "%s: remove failed (dead)",
> + alias);
> + else
> + condlog(2, "%s: remove (dead)", alias);
Previously, this code, which flushes maps that aren't valid any more,
would only be called from multipathd() -> reconfigure(), because in all
other cases newmp was NULL.
AFAICS, you will now call dm_flush_map() even if called with
newmp==NULL (this means, from cli_add_map(), or from multipath, even if
called with CMD_DRY_RUN). I don't think that's what we want.
Regards,
Martin
More information about the dm-devel
mailing list