[DTrace-devel] [PATCH 11/12] Abort if provider population fails for any provider

[Eugene Loh]

It seems to me, we have to go through the providers and clean up error 
returns if we want to do this.

Most providers seems to have populate() return 0 on error.  E.g., 
dt_prov_dtrace.c.  It calls dt_provider_create(), which if it runs out 
of memory, calls dt_set_errno(EDT_NOMEM) and returns NULL.  So 
dt_prov_dtrace.c's populate() sees the NULL and then returns 0. This 
patch decides that populate ran just fine.

One solution may be to fix this patch so that n<=0 is an error.

But we might still want it okay for a provider not to return any 
probes.  E.g., lockstat with a back-rev kernel, etc.  So then it's a 
matter of going through and having providers' populate() functions 
return -1 when they intend error.  E.g., dt_sdt_populate() should.  That 
would fix ip, lockstat, and sched.  Then go through and fix cpc, dtrace, 
fbt, proc, profile, rawtp, sdt, and syscall.

Anyhow, looks like basically all the other providers need to switch to 
-1 for error for this patch to work;  they currently return 0 for error.

On 1/5/24 00:31, Kris Van Hees via DTrace-devel wrote:
> diff --git a/libdtrace/dt_open.c b/libdtrace/dt_open.c
> @@ -1208,6 +1208,9 @@ dtrace_init(dtrace_hdl_t *dtp)
>   		int	n;
>   
>   		n = dt_providers[i]->populate(dtp);
> +		if (n < 0)
> +			return -1;		/* errno is already set */
> +
>   		dt_dprintf("loaded %d probes for %s\n", n,
>   			   dt_providers[i]->name);
>   	}