Skip to content

Commit

Permalink
Handle 32-bit overflows in disk utilization stats
Browse files Browse the repository at this point in the history
Linux prints [1] some of the values reported in block device's `stat`
sysfs file as 32-bit unsigned integers. fio interprets them as 64-bit
integers when reading that sysfs file and performs further arithmetics
on them in 64-bits. If the reported value overflows during fio run,
a huge bogus value is reported in the "disk utilization" block instead.

[1] https://source.corp.google.com/h/kernel/pub/scm/linux/kernel/git/torvalds/linux/+/refs/tags/v5.10:block/genhd.c;l=1301-1328;drc=7e890c37c25c7cbca37ff0ab292873d8146e713b

Signed-off-by: Alexey Neyman <aneyman@google.com>
  • Loading branch information
stilor committed Oct 4, 2023
1 parent 6f9cdcf commit f15a6ea
Showing 1 changed file with 23 additions and 4 deletions.
27 changes: 23 additions & 4 deletions diskutil.c
Original file line number Diff line number Diff line change
Expand Up @@ -77,6 +77,24 @@ static int get_io_ticks(struct disk_util *du, struct disk_util_stat *dus)
return ret != 10;
}

static uint64_t safe_32bit_diff(uint64_t nval, uint64_t oval)

This comment has been minimized.

Copy link
@vincentkfu

vincentkfu Oct 4, 2023

Collaborator

Please change the variable names to something like new_val and old_val or new and old. oval is used elsewhere in fio and it's confusing to see it here.

{
/* Linux kernel prints some of the stat fields as 32-bit integers. It is
* possible that the value overflows, but since fio uses unsigned 64-bit
* arithmetics in update_io_tick_disk(), it instead results in a huge
* bogus value being added to the respective accumulating field. Just
* in case Linux starts reporting these metrics as 64-bit values in the
* future, check that overflow actually happens around the 32-bit
* unsigned boundary; assume overflow only happens once between
* successive polls.
*/
if (oval <= nval || oval >= (1ull << 32)) {
return nval - oval;
} else {
return (1ull << 32) + nval - oval;
}
}

static void update_io_tick_disk(struct disk_util *du)
{
struct disk_util_stat __dus, *dus, *ldus;
Expand All @@ -96,10 +114,11 @@ static void update_io_tick_disk(struct disk_util *du)
dus->s.ios[1] += (__dus.s.ios[1] - ldus->s.ios[1]);
dus->s.merges[0] += (__dus.s.merges[0] - ldus->s.merges[0]);
dus->s.merges[1] += (__dus.s.merges[1] - ldus->s.merges[1]);
dus->s.ticks[0] += (__dus.s.ticks[0] - ldus->s.ticks[0]);
dus->s.ticks[1] += (__dus.s.ticks[1] - ldus->s.ticks[1]);
dus->s.io_ticks += (__dus.s.io_ticks - ldus->s.io_ticks);
dus->s.time_in_queue += (__dus.s.time_in_queue - ldus->s.time_in_queue);
dus->s.ticks[0] += safe_32bit_diff(__dus.s.ticks[0], ldus->s.ticks[0]);
dus->s.ticks[1] += safe_32bit_diff(__dus.s.ticks[1], ldus->s.ticks[1]);
dus->s.io_ticks += safe_32bit_diff(__dus.s.io_ticks, ldus->s.io_ticks);
dus->s.time_in_queue +=
safe_32bit_diff(__dus.s.time_in_queue, ldus->s.time_in_queue);

fio_gettime(&t, NULL);
dus->s.msec += mtime_since(&du->time, &t);
Expand Down

0 comments on commit f15a6ea

Please sign in to comment.