[RFC PATCH v2 2/4] Add eventfd option to --decode-fds

[Dmitry V. Levin]

On Wed, May 01, 2024 at 10:57:59PM +0530, Sahil Siddiq wrote:
[...]
> +static bool
> +parse_fdinfo_efd_counter(const char *value, void *data)
> +{
> +	uint64_t *efd_counter = data;
> +	*efd_counter = string_to_uint_ex(value, NULL, ULLONG_MAX, "\n", 16);

string_to_uint_ex() internally handles it as a "signed long long" value
and rejects negative values as invalid, so even though ULLONG_MAX was
specified, the actual limit imposed by this implementation is LLONG_MAX.

That is, you have either to open-code string_to_uint_ex here, or just
treat it as a string.

> +	return true;
> +}
> +
> +static bool
> +printeventfd(pid_t pid_of_fd, int fd, const char *path)
> +{
> +	static const char eventfd_path[] = "anon_inode:[eventfd]";
> +	static const char efd_counter_pfx[] = "eventfd-count:";
> +	static const char efd_id_pfx[] = "eventfd-id:";
> +
> +	if (strcmp(path, eventfd_path))
> +		return false;
> +
> +	uint64_t efd_counter = -1ULL;
> +	int efd_id = -1;
> +
> +	scan_fdinfo(pid_of_fd, fd, efd_counter_pfx, sizeof(efd_counter_pfx) - 1,
> +			   parse_fdinfo_efd_counter, &efd_counter);
> +
> +	scan_fdinfo(pid_of_fd, fd, efd_id_pfx, sizeof(efd_id_pfx) - 1,
> +			   parse_fdinfo_efd_id, &efd_id);
> +
> +	if (efd_counter != -1ULL && efd_id != -1) {
> +		tprint_associated_info_begin();
> +		tprints_string("eventfd-count:");
> +		PRINT_VAL_U(efd_counter);
> +		tprint_arg_next();

tprint_arg_next() is for syscall and function arguments, here we probably
need to introduce tprint_associated_info_next().

> +		tprints_string("eventfd-id:");
> +		PRINT_VAL_U(efd_id);

For the record, "eventfd-count:" was introduced in v3.8-rc1~74^2~8, while
"eventfd-id:" is available since v5.2-rc1~62^2~38.  It seems to me that it
is worth printing eventfd-count if the kernel is not fresh enough.

I would even considered printing eventfd-semaphore that was introduced
much later in v6.5-rc1~246^2~5.


-- 
ldv