engines/libblkio: Add option libblkio_wait_mode

[fio.git] / io_u.c

diff --git a/io_u.c b/io_u.c
index 059637e592d2b8f6d9b5bff8617be961ca919c0f..8035f4b725fc13b64be0c11aa5b45853e19a7982 100644 (file)
--- a/io_u.c
+++ b/io_u.c
@@ -355,7 +355,7 @@ static int get_next_seq_offset(struct thread_data *td, struct fio_file *f,
         * and invalidate the cache, if we need to.
         */
        if (f->last_pos[ddir] >= f->io_size + get_start_offset(td, f) &&
-           o->time_based) {
+           o->time_based && o->nr_files == 1) {
                f->last_pos[ddir] = f->file_offset;
                loop_cache_invalidate(td, f);
        }
@@ -417,7 +417,13 @@ static int get_next_block(struct thread_data *td, struct io_u *io_u,
 
        b = offset = -1ULL;
 
-       if (rw_seq) {
+       if (td_randtrimwrite(td) && ddir == DDIR_WRITE) {
+               /* don't mark randommap for these writes */
+               io_u_set(td, io_u, IO_U_F_BUSY_OK);
+               offset = f->last_start[DDIR_TRIM];
+               *is_random = true;
+               ret = 0;
+       } else if (rw_seq) {
                if (td_random(td)) {
                        if (should_do_random(td, ddir)) {
                                ret = get_next_rand_block(td, f, ddir, &b);
@@ -507,6 +513,24 @@ static int get_next_offset(struct thread_data *td, struct io_u *io_u,
                return 1;
        }
 
+       /*
+        * For randtrimwrite, we decide whether to issue a trim or a write
+        * based on whether the offsets for the most recent trim and write
+        * operations match. If they don't match that means we just issued a
+        * new trim and the next operation should be a write. If they *do*
+        * match that means we just completed a trim+write pair and the next
+        * command should be a trim.
+        *
+        * This works fine for sequential workloads but for random workloads
+        * it's possible to complete a trim+write pair and then have the next
+        * randomly generated offset match the previous offset. If that happens
+        * we need to alter the offset for the last write operation in order
+        * to ensure that we issue a write operation the next time through.
+        */
+       if (td_randtrimwrite(td) && ddir == DDIR_TRIM &&
+           f->last_start[DDIR_TRIM] == io_u->offset)
+               f->last_start[DDIR_WRITE]--;
+
        io_u->verify_offset = io_u->offset;
        return 0;
 }
@@ -530,6 +554,12 @@ static unsigned long long get_next_buflen(struct thread_data *td, struct io_u *i
 
        assert(ddir_rw(ddir));
 
+       if (td_randtrimwrite(td) && ddir == DDIR_WRITE) {
+               struct fio_file *f = io_u->file;
+
+               return f->last_pos[DDIR_TRIM] - f->last_start[DDIR_TRIM];
+       }
+
        if (td->o.bs_is_seq_rand)
                ddir = is_random ? DDIR_WRITE : DDIR_READ;
 
@@ -768,7 +798,7 @@ static void set_rw_ddir(struct thread_data *td, struct io_u *io_u)
 
        if (td_trimwrite(td)) {
                struct fio_file *f = io_u->file;
-               if (f->last_pos[DDIR_WRITE] == f->last_pos[DDIR_TRIM])
+               if (f->last_start[DDIR_WRITE] == f->last_start[DDIR_TRIM])
                        ddir = DDIR_TRIM;
                else
                        ddir = DDIR_WRITE;
@@ -993,7 +1023,7 @@ static void __io_u_mark_map(uint64_t *map, unsigned int nr)
                break;
        case 1 ... 4:
                idx = 1;
-               fallthrough;
+               fio_fallthrough;
        case 0:
                break;
        }
@@ -1035,7 +1065,7 @@ void io_u_mark_depth(struct thread_data *td, unsigned int nr)
                break;
        case 2 ... 3:
                idx = 1;
-               fallthrough;
+               fio_fallthrough;
        case 1:
                break;
        }
@@ -1076,7 +1106,7 @@ static void io_u_mark_lat_nsec(struct thread_data *td, unsigned long long nsec)
                break;
        case 2 ... 3:
                idx = 1;
-               fallthrough;
+               fio_fallthrough;
        case 0 ... 1:
                break;
        }
@@ -1118,7 +1148,7 @@ static void io_u_mark_lat_usec(struct thread_data *td, unsigned long long usec)
                break;
        case 2 ... 3:
                idx = 1;
-               fallthrough;
+               fio_fallthrough;
        case 0 ... 1:
                break;
        }
@@ -1166,7 +1196,7 @@ static void io_u_mark_lat_msec(struct thread_data *td, unsigned long long msec)
                break;
        case 2 ... 3:
                idx = 1;
-               fallthrough;
+               fio_fallthrough;
        case 0 ... 1:
                break;
        }
@@ -2091,13 +2121,26 @@ static void ios_completed(struct thread_data *td,
        }
 }
 
+static void io_u_update_bytes_done(struct thread_data *td,
+                                  struct io_completion_data *icd)
+{
+       int ddir;
+
+       if (td->runstate == TD_VERIFYING) {
+               td->bytes_verified += icd->bytes_done[DDIR_READ];
+               return;
+       }
+
+       for (ddir = 0; ddir < DDIR_RWDIR_CNT; ddir++)
+               td->bytes_done[ddir] += icd->bytes_done[ddir];
+}
+
 /*
  * Complete a single io_u for the sync engines.
  */
 int io_u_sync_complete(struct thread_data *td, struct io_u *io_u)
 {
        struct io_completion_data icd;
-       int ddir;
 
        init_icd(td, &icd, 1);
        io_completed(td, &io_u, &icd);
@@ -2110,8 +2153,7 @@ int io_u_sync_complete(struct thread_data *td, struct io_u *io_u)
                return -1;
        }
 
-       for (ddir = 0; ddir < DDIR_RWDIR_CNT; ddir++)
-               td->bytes_done[ddir] += icd.bytes_done[ddir];
+       io_u_update_bytes_done(td, &icd);
 
        return 0;
 }
@@ -2123,7 +2165,7 @@ int io_u_queued_complete(struct thread_data *td, int min_evts)
 {
        struct io_completion_data icd;
        struct timespec *tvp = NULL;
-       int ret, ddir;
+       int ret;
        struct timespec ts = { .tv_sec = 0, .tv_nsec = 0, };
 
        dprint(FD_IO, "io_u_queued_complete: min=%d\n", min_evts);
@@ -2149,8 +2191,7 @@ int io_u_queued_complete(struct thread_data *td, int min_evts)
                return -1;
        }
 
-       for (ddir = 0; ddir < DDIR_RWDIR_CNT; ddir++)
-               td->bytes_done[ddir] += icd.bytes_done[ddir];
+       io_u_update_bytes_done(td, &icd);
 
        return ret;
 }
@@ -2297,7 +2338,11 @@ int do_io_u_sync(const struct thread_data *td, struct io_u *io_u)
        int ret;
 
        if (io_u->ddir == DDIR_SYNC) {
+#ifdef CONFIG_FCNTL_SYNC
+               ret = fcntl(io_u->file->fd, F_FULLFSYNC);
+#else
                ret = fsync(io_u->file->fd);
+#endif
        } else if (io_u->ddir == DDIR_DATASYNC) {
 #ifdef CONFIG_FDATASYNC
                ret = fdatasync(io_u->file->fd);