[FFmpeg-user] How to correctly free a double-linked list

[Paul B Mahol]

On 7/7/19, Ulf Zibis <Ulf.Zibis at gmx.de> wrote:
> Hi,
>
> when running my code, at the end I irregularly get messages like this:
> [.....]
> [Parsed_lineshiftrecover_0 @ 0x55e5afb83e80] start freeing memory ...
> [Parsed_lineshiftrecover_0 @ 0x55e5afb83e80] filler and shifts freed!
> [Parsed_lineshiftrecover_0 @ 0x55e5afb83e80] now start freeing
> shifts_sums ...
> corrupted double-linked list
> debug/lineshiftrecover.sh: Zeile 99: 13359 Abgebrochen
> (Speicherabzug geschrieben) ./ffmpeg -y -v warning -i debug/$sample -vf
> lineshiftrecover${options[o]} -q 5 $output
>
> debug/lineshiftrecover.sh: Zeile 99: 10524 Abgebrochen
> (Speicherabzug geschrieben) ./ffmpeg -y -v warning -i debug/$sample -vf
> lineshiftrecover${options[o]} -q 5 $output
>
>
> As you can see I'm running my code from the script debug/lineshiftrecover.sh
> lineshiftrecover is a new filter I'm working on. It's purpose is to
> repair shifted line glitches in videos recoded from VHS tape.
>
> The code runs fine until I'm coming to free the allocated memory in
> function "uninit". The code is:
> static av_cold void uninit(AVFilterContext *ctx)
> {
>     LineShiftContext *s = ctx->priv;
>     av_log(ctx, s->report >= R_SHIFTS ? AV_LOG_WARNING : AV_LOG_INFO,
> "start freeing memory ...\n");
>     for (int p = 0; p < s->nb_planes; p++)
>         av_freep(&s->filler[p]);
>     av_freep(&s->shifts);
> /*
>     printf("printf(): filler and shifts freed!\n");
> */
>     av_log(ctx, s->report >= R_SHIFTS ? AV_LOG_WARNING : AV_LOG_INFO,
> "filler and shifts freed!\n");
>     if (s->report >= R_SHIFTS) {
> /*
>         printf("do printf() before statistic of found shifts!\n");
>         av_log(ctx, AV_LOG_WARNING, "statistic of found shifts ...\n");
>         av_log(NULL, AV_LOG_WARNING, "shift:    "); // * increment);
>         for (int shift = 0; shift <= s->span_r -  s->span_l; shift++)
>             av_log(NULL, AV_LOG_WARNING, "%6d", s->span_l + shift);
>         for (int l = 0; l <= s->lines; l++) {
>             av_log(NULL, AV_LOG_WARNING, "\nline %4d:", s->start + l);
>             for (int shift = 0; shift <= s->span_r -  s->span_l; shift++)
>                 av_log(NULL, AV_LOG_WARNING, "%6d",
> s->shifts_sums[l][shift]);
>             av_log(NULL, AV_LOG_WARNING, "\n");
>         }
> */
> /*
>         printf("printf(): now start freeing shifts_sums ...\n");
> */
>         av_log(ctx, AV_LOG_WARNING, "now start freeing shifts_sums ...\n");
>         for (int l = 0; l <= s->lines; l++) {
>             av_freep(&(s->shifts_sums[l]));
>         }
>         av_log(ctx, AV_LOG_WARNING, "lines of shifts_sums freed!\n");
>         av_freep(&(s->shifts_sums));
> /*
>         printf("printf(): shifts_sums freed!\n");
> */
>         av_log(ctx, AV_LOG_WARNING, "shifts_sums freed!\n");
>     }
> }
>
> The out-commented printf() lines are to check if there is a problem if
> the av_log print buffer is not flushed already, but this seems not the case.
>
> The definition is:
>     int16_t **shifts_sums;
> The initialization is:
> static int config_props(AVFilterLink *inlink)
> {
>     [.....]
>     if (s->report >= R_SHIFTS) {
>         s->shifts_sums = av_malloc((s->lines + 1) *
> sizeof(*s->shifts_sums));
>         for (int l = s->lines; l >= 0; l--) {
>             s->shifts_sums[l] = av_malloc((s->span_r - s->span_l + 1) *
> sizeof(**s->shifts_sums));
>             memset(s->shifts_sums[l], 0, (s->span_r - s->span_l + 1) *
> sizeof(**s->shifts_sums));
>         }

There is av_calloc, it calls memset for you after allocation.

It looks like you access unallocated array in uninit because your
allocation is under condition.
Before freeing lines you need to check if array is not NULL.


>     }
> }
>
>
> --
> Von meinem Seibert gesendet
>
> _______________________________________________
> ffmpeg-user mailing list
> ffmpeg-user at ffmpeg.org
> https://ffmpeg.org/mailman/listinfo/ffmpeg-user
>
> To unsubscribe, visit link above, or email
> ffmpeg-user-request at ffmpeg.org with subject "unsubscribe".