iolog: remove random layout verification optimisation

Running the following fio jobs unexpectedly reports a verification
failure:
rm /tmp/tmp.fio; ./fio --iodepth=1 \
 --verify=pattern --verify_fatal=1 --size=100M --bsrange=512-128k \
 --rw=randwrite --verify_backlog=128 --filename=/tmp/tmp.fio \
 --verify_pattern="%o" --name=spuriousmismatch1

rm /tmp/tmp.fio; ./fio --iodepth=1 \
 --verify=crc32c --verify_fatal=1 --size=100M --bs=4k \
 --rw=randwrite --verify_backlog=20 --filename=/tmp/tmp.fio \
 --percentage_random=50 --randseed=86 --name=spuriousmismatch2

In the case of the first job, using a bsrange where the start and end
are different can cause random write I/O to overlap an already written
region making the original data unverifiable.

For the second job, when percentage_random is between 1 and 99 the same
offset can be generated multiple times but only the last write to that
offset should be verified.

Rather than special casing the growing number of random jobs that might
generate overlaps while still having a randommap, and given preallocation
during layout is the default where possible, just remove the overwrite=0
optimisation thus forcing all random jobs to be checked for overlaps. It
is still possible to force the old behaviour by setting verifysort=0.

Fixes https://github.com/axboe/fio/issues/335 and
https://github.com/axboe/fio/issues/344 .

Tested-by: Jeff Furlong <jeff.furlong@wdc.com>
Signed-off-by: Sitsofe Wheeler <sitsofe@yahoo.com>

diff --git a/iolog.c b/iolog.c
index 18ae436..b041eff 100644 (file)
--- a/iolog.c
+++ b/iolog.c
@@ -227,20 +227,16 @@ void log_io_piece(struct thread_data *td, struct io_u *io_u)
        }
 
        /*
-        * We don't need to sort the entries, if:
-        *
-        *      Sequential writes, or
-        *      Random writes that lay out the file as it goes along
-        *
-        * For both these cases, just reading back data in the order we
-        * wrote it out is the fastest.
+        * We don't need to sort the entries if we only performed sequential
+        * writes. In this case, just reading back data in the order we wrote
+        * it out is the faster but still safe.
         *
         * One exception is if we don't have a random map AND we are doing
         * verifies, in that case we need to check for duplicate blocks and
         * drop the old one, which we rely on the rb insert/lookup for
         * handling.
         */
-       if (((!td->o.verifysort) || !td_random(td) || !td->o.overwrite) &&
+       if (((!td->o.verifysort) || !td_random(td)) &&
              (file_randommap(td, ipo->file) || td->o.verify == VERIFY_NONE)) {
                INIT_FLIST_HEAD(&ipo->list);
                flist_add_tail(&ipo->list, &td->io_hist_list);