Initial suppor for sync_file_range()
This revs the ioengine to 11, as we now have another data direction.
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
diff --git a/engines/sync.c b/engines/sync.c
index 12b85f6..a673ae9 100644
--- a/engines/sync.c
+++ b/engines/sync.c
@@ -64,6 +64,29 @@
return FIO_Q_COMPLETED;
}
+static int fio_syncio_sync(struct thread_data *td, struct io_u *io_u)
+{
+ int ret;
+
+ if (io_u->ddir == DDIR_SYNC) {
+ ret = fsync(io_u->file->fd);
+ } else if (io_u->ddir == DDIR_DATASYNC) {
+#ifdef FIO_HAVE_FDATASYNC
+ ret = fdatasync(io_u->file->fd);
+#else
+ ret = io_u->xfer_buflen;
+ io_u->error = EINVAL;
+#endif
+ } else if (io_u->ddir == DDIR_SYNC_FILE_RANGE)
+ ret = do_sync_file_range(td, io_u->file);
+ else {
+ ret = io_u->xfer_buflen;
+ io_u->error = EINVAL;
+ }
+
+ return fio_io_end(td, io_u, ret);
+}
+
static int fio_psyncio_queue(struct thread_data *td, struct io_u *io_u)
{
struct fio_file *f = io_u->file;
@@ -76,7 +99,7 @@
else if (io_u->ddir == DDIR_WRITE)
ret = pwrite(f->fd, io_u->xfer_buf, io_u->xfer_buflen, io_u->offset);
else
- ret = fsync(f->fd);
+ return fio_syncio_sync(td, io_u);
return fio_io_end(td, io_u, ret);
}
@@ -93,7 +116,7 @@
else if (io_u->ddir == DDIR_WRITE)
ret = write(f->fd, io_u->xfer_buf, io_u->xfer_buflen);
else
- ret = fsync(f->fd);
+ return fio_syncio_sync(td, io_u);
return fio_io_end(td, io_u, ret);
}
@@ -163,22 +186,9 @@
*/
if (sd->queued)
return FIO_Q_BUSY;
- if (io_u->ddir == DDIR_SYNC) {
- int ret = fsync(io_u->file->fd);
+ if (ddir_sync(io_u->ddir))
+ return fio_syncio_sync(td, io_u);
- return fio_io_end(td, io_u, ret);
- } else if (io_u->ddir == DDIR_DATASYNC) {
- int ret;
-#ifdef FIO_HAVE_FDATASYNC
- ret = fdatasync(io_u->file->fd);
-#else
- ret = io_u->xfer_buflen;
- io_u->error = EINVAL;
-#endif
- return fio_io_end(td, io_u, ret);
-
- }
-
sd->queued = 0;
sd->queued_bytes = 0;
fio_vsyncio_set_iov(sd, io_u, 0);
diff --git a/file.h b/file.h
index 2abe3ba..30293fc 100644
--- a/file.h
+++ b/file.h
@@ -74,6 +74,9 @@
unsigned long long last_pos;
+ unsigned long long first_write;
+ unsigned long long last_write;
+
/*
* For use by the io engine
*/
diff --git a/fio.h b/fio.h
index f00f64a..05911c0 100644
--- a/fio.h
+++ b/fio.h
@@ -282,6 +282,8 @@
unsigned int uid;
unsigned int gid;
+
+ unsigned int sync_file_range;
};
#define FIO_VERROR_SIZE 128
@@ -417,6 +419,8 @@
unsigned int file_service_left;
struct fio_file *file_service_file;
+ unsigned int sync_file_range_nr;
+
/*
* For generating file sizes
*/
diff --git a/io_ddir.h b/io_ddir.h
index 03eefdb..87cded4 100644
--- a/io_ddir.h
+++ b/io_ddir.h
@@ -6,6 +6,7 @@
DDIR_WRITE,
DDIR_SYNC,
DDIR_DATASYNC,
+ DDIR_SYNC_FILE_RANGE,
DDIR_WAIT,
DDIR_INVAL = -1,
};
@@ -28,7 +29,8 @@
static inline int ddir_sync(enum fio_ddir ddir)
{
- return ddir == DDIR_SYNC || ddir == DDIR_DATASYNC;
+ return ddir == DDIR_SYNC || ddir == DDIR_DATASYNC ||
+ ddir == DDIR_SYNC_FILE_RANGE;
}
#endif
diff --git a/io_u.c b/io_u.c
index 9b9570e..a4bf0c0 100644
--- a/io_u.c
+++ b/io_u.c
@@ -394,6 +394,14 @@
td->io_issues[DDIR_WRITE] && should_fsync(td))
return DDIR_DATASYNC;
+ /*
+ * see if it's time to sync_file_range
+ */
+ if (td->sync_file_range_nr &&
+ !(td->io_issues[DDIR_WRITE] % td->sync_file_range_nr) &&
+ td->io_issues[DDIR_WRITE] && should_fsync(td))
+ return DDIR_SYNC_FILE_RANGE;
+
if (td_rw(td)) {
/*
* Check if it's time to seed a new data direction.
@@ -996,6 +1004,7 @@
* initialized, silence that warning.
*/
unsigned long uninitialized_var(usec);
+ struct fio_file *f;
dprint_io_u(io_u, "io complete");
@@ -1006,6 +1015,11 @@
if (ddir_sync(io_u->ddir)) {
td->last_was_sync = 1;
+ f = io_u->file;
+ if (f) {
+ f->first_write = -1ULL;
+ f->last_write = -1ULL;
+ }
return;
}
@@ -1021,6 +1035,18 @@
td->io_bytes[idx] += bytes;
td->this_io_bytes[idx] += bytes;
+ if (idx == DDIR_WRITE) {
+ f = io_u->file;
+ if (f) {
+ if (f->first_write == -1ULL ||
+ io_u->offset < f->first_write)
+ f->first_write = io_u->offset;
+ if (f->last_write == -1ULL ||
+ ((io_u->offset + bytes) > f->last_write))
+ f->last_write = io_u->offset + bytes;
+ }
+ }
+
if (ramp_time_over(td)) {
unsigned long uninitialized_var(lusec);
diff --git a/ioengine.h b/ioengine.h
index eb6655d..5acfbd2 100644
--- a/ioengine.h
+++ b/ioengine.h
@@ -1,7 +1,7 @@
#ifndef FIO_IOENGINE_H
#define FIO_IOENGINE_H
-#define FIO_IOOPS_VERSION 10
+#define FIO_IOOPS_VERSION 11
enum {
IO_U_F_FREE = 1 << 0,
@@ -153,6 +153,8 @@
void io_u_mark_complete(struct thread_data *, unsigned int);
void io_u_mark_submit(struct thread_data *, unsigned int);
+int do_sync_file_range(struct thread_data *, struct fio_file *);
+
#ifdef FIO_INC_DEBUG
static inline void dprint_io_u(struct io_u *io_u, const char *p)
{
diff --git a/ioengines.c b/ioengines.c
index 7f0a5c4..0e89534 100644
--- a/ioengines.c
+++ b/ioengines.c
@@ -431,3 +431,16 @@
return td->io_ops->get_file_size(td, f);
}
+
+int do_sync_file_range(struct thread_data *td, struct fio_file *f)
+{
+ off64_t offset, nbytes;
+
+ offset = f->first_write;
+ nbytes = f->last_write - f->first_write;
+
+ if (nbytes)
+ return sync_file_range(f->fd, offset, nbytes, 0);
+
+ return 0;
+}
diff --git a/log.c b/log.c
index ba52f07..99f20b5 100644
--- a/log.c
+++ b/log.c
@@ -20,9 +20,10 @@
void log_io_u(struct thread_data *td, struct io_u *io_u)
{
- const char *act[] = { "read", "write", "sync", "datasync" };
+ const char *act[] = { "read", "write", "sync", "datasync",
+ "sync_file_range" };
- assert(io_u->ddir < 3);
+ assert(io_u->ddir <= 4);
if (!td->o.write_iolog_file)
return;
diff --git a/options.c b/options.c
index e2daf37..bee15cf 100644
--- a/options.c
+++ b/options.c
@@ -429,6 +429,20 @@
return 0;
}
+static int str_sfr_cb(void *data, const char *str)
+{
+ struct thread_data *td = data;
+ char *nr = get_opt_postfix(str);
+
+ td->sync_file_range_nr = 1;
+ if (nr) {
+ td->sync_file_range_nr = atoi(nr);
+ free(nr);
+ }
+
+ return 0;
+}
+
static int check_dir(struct thread_data *td, char *fname)
{
char file[PATH_MAX], *dir;
@@ -1110,6 +1124,30 @@
.help = "Issue fdatasync for writes every given number of blocks",
.def = "0",
},
+#ifdef FIO_HAVE_SYNC_FILE_RANGE
+ {
+ .name = "sync_file_range",
+ .posval = {
+ { .ival = "wait_before",
+ .oval = SYNC_FILE_RANGE_WAIT_BEFORE,
+ .help = "SYNC_FILE_RANGE_WAIT_BEFORE",
+ },
+ { .ival = "write",
+ .oval = SYNC_FILE_RANGE_WRITE,
+ .help = "SYNC_FILE_RANGE_WRITE",
+ },
+ {
+ .ival = "wait_after",
+ .oval = SYNC_FILE_RANGE_WAIT_AFTER,
+ .help = "SYNC_FILE_RANGE_WAIT_AFTER",
+ },
+ },
+ .type = FIO_OPT_STR,
+ .cb = str_sfr_cb,
+ .off1 = td_var_offset(sync_file_range),
+ .help = "Use sync_file_range()",
+ },
+#endif
{
.name = "direct",
.type = FIO_OPT_BOOL,
diff --git a/os/os-linux.h b/os/os-linux.h
index 01140e0..8c61cc0 100644
--- a/os/os-linux.h
+++ b/os/os-linux.h
@@ -34,6 +34,7 @@
#define FIO_HAVE_CL_SIZE
#define FIO_HAVE_CGROUPS
#define FIO_HAVE_FDATASYNC
+#define FIO_HAVE_SYNC_FILE_RANGE
#define OS_MAP_ANON MAP_ANONYMOUS