Add support for replaying blktrace trim/discard
Signed-off-by: Jens Axboe <jaxboe@fusionio.com>
diff --git a/blktrace.c b/blktrace.c
index 6cf8d46..f22ab1b 100644
--- a/blktrace.c
+++ b/blktrace.c
@@ -228,12 +228,41 @@
case BLK_TN_TIMESTAMP:
printf("got timestamp notify: %x, %d\n", t->action, t->pid);
break;
+ case BLK_TN_MESSAGE:
+ break;
default:
dprint(FD_BLKTRACE, "unknown trace act %x\n", t->action);
break;
}
}
+static void handle_trace_discard(struct thread_data *td, struct blk_io_trace *t,
+ unsigned long long ttime, unsigned long *ios)
+{
+ struct io_piece *ipo = malloc(sizeof(*ipo));
+
+ trace_add_file(td, t->device);
+
+ ios[DDIR_WRITE]++;
+ td->o.size += t->bytes;
+
+ memset(ipo, 0, sizeof(*ipo));
+ INIT_FLIST_HEAD(&ipo->list);
+
+ /*
+ * the 512 is wrong here, it should be the hardware sector size...
+ */
+ ipo->offset = t->sector * 512;
+ ipo->len = t->bytes;
+ ipo->delay = ttime / 1000;
+ ipo->ddir = DDIR_TRIM;
+
+ dprint(FD_BLKTRACE, "store discard, off=%llu, len=%lu, delay=%lu\n",
+ ipo->offset, ipo->len,
+ ipo->delay);
+ queue_io_piece(td, ipo);
+}
+
static void handle_trace_fs(struct thread_data *td, struct blk_io_trace *t,
unsigned long long ttime, unsigned long *ios,
unsigned int *bs)
@@ -267,6 +296,8 @@
if (t->action & BLK_TC_ACT(BLK_TC_NOTIFY))
handle_trace_notify(t);
+ else if (t->action & BLK_TC_ACT(BLK_TC_DISCARD))
+ handle_trace_discard(td, t, ttime, ios);
else
handle_trace_fs(td, t, ttime, ios, bs);
}
diff --git a/blktrace_api.h b/blktrace_api.h
index 61b405a..3df3347 100644
--- a/blktrace_api.h
+++ b/blktrace_api.h
@@ -20,6 +20,8 @@
BLK_TC_NOTIFY = 1 << 10, /* special message */
BLK_TC_AHEAD = 1 << 11, /* readahead */
BLK_TC_META = 1 << 12, /* metadata */
+ BLK_TC_DISCARD = 1 << 13, /* discard requests */
+ BLK_TC_DRV_DATA = 1 << 14, /* binary per-driver data */
BLK_TC_END = 1 << 15, /* only 16-bits, reminder */
};
@@ -46,6 +48,8 @@
__BLK_TA_SPLIT, /* bio was split */
__BLK_TA_BOUNCE, /* bio was bounced */
__BLK_TA_REMAP, /* bio was remapped */
+ __BLK_TA_ABORT, /* request aborted */
+ __BLK_TA_DRV_DATA, /* driver-specific binary data */
};
/*
@@ -54,6 +58,7 @@
enum blktrace_notify {
__BLK_TN_PROCESS = 0, /* establish pid/name mapping */
__BLK_TN_TIMESTAMP, /* include system clock */
+ __BLK_TN_MESSAGE, /* Character string message */
};
/*
@@ -74,9 +79,11 @@
#define BLK_TA_SPLIT (__BLK_TA_SPLIT)
#define BLK_TA_BOUNCE (__BLK_TA_BOUNCE)
#define BLK_TA_REMAP (__BLK_TA_REMAP | BLK_TC_ACT(BLK_TC_QUEUE))
+#define BLK_TA_DRV_DATA (__BLK_TA_DRV_DATA | BLK_TC_ACT(BLK_TC_DRV_DATA))
#define BLK_TN_PROCESS (__BLK_TN_PROCESS | BLK_TC_ACT(BLK_TC_NOTIFY))
#define BLK_TN_TIMESTAMP (__BLK_TN_TIMESTAMP | BLK_TC_ACT(BLK_TC_NOTIFY))
+#define BLK_TN_MESSAGE (__BLK_TN_MESSAGE | BLK_TC_ACT(BLK_TC_NOTIFY))
#define BLK_IO_TRACE_MAGIC 0x65617400
#define BLK_IO_TRACE_VERSION 0x07
diff --git a/engines/mmap.c b/engines/mmap.c
index 53fd358..002918c 100644
--- a/engines/mmap.c
+++ b/engines/mmap.c
@@ -154,12 +154,18 @@
io_u->error = errno;
td_verror(td, io_u->error, "msync");
}
+ } else if (io_u->ddir == DDIR_TRIM) {
+ int ret = do_io_u_trim(td, io_u);
+
+ if (!ret)
+ td_verror(td, io_u->error, "trim");
}
+
/*
* not really direct, but should drop the pages from the cache
*/
- if (td->o.odirect && !ddir_sync(io_u->ddir)) {
+ if (td->o.odirect && ddir_rw(io_u->ddir)) {
if (msync(io_u->mmap_data, io_u->xfer_buflen, MS_SYNC) < 0) {
io_u->error = errno;
td_verror(td, io_u->error, "msync");
diff --git a/engines/sync.c b/engines/sync.c
index 3dbce47..4eea2f9 100644
--- a/engines/sync.c
+++ b/engines/sync.c
@@ -30,7 +30,7 @@
{
struct fio_file *f = io_u->file;
- if (ddir_sync(io_u->ddir))
+ if (!ddir_rw(io_u->ddir))
return 0;
if (f->file_pos != -1ULL && f->file_pos == io_u->offset)
@@ -46,7 +46,7 @@
static int fio_io_end(struct thread_data *td, struct io_u *io_u, int ret)
{
- if (io_u->file && ret >= 0)
+ if (io_u->file && ret >= 0 && ddir_rw(io_u->ddir))
io_u->file->file_pos = io_u->offset + ret;
if (ret != (int) io_u->xfer_buflen) {
diff --git a/fio.c b/fio.c
index 6ab0f4a..1d20cf7 100644
--- a/fio.c
+++ b/fio.c
@@ -176,6 +176,8 @@
unsigned int rate_iops = 0;
unsigned int rate_iops_min = 0;
+ assert(ddir_rw(ddir));
+
if (!td->o.ratemin[ddir] && !td->o.rate_iops_min[ddir])
return 0;
@@ -491,7 +493,8 @@
io_u->xfer_buf += bytes;
io_u->offset += bytes;
- td->ts.short_io_u[io_u->ddir]++;
+ if (ddir_rw(io_u->ddir))
+ td->ts.short_io_u[io_u->ddir]++;
if (io_u->offset == f->real_file_size)
goto sync_done;
@@ -636,7 +639,8 @@
io_u->xfer_buf += bytes;
io_u->offset += bytes;
- td->ts.short_io_u[io_u->ddir]++;
+ if (ddir_rw(io_u->ddir))
+ td->ts.short_io_u[io_u->ddir]++;
if (io_u->offset == f->real_file_size)
goto sync_done;
diff --git a/init.c b/init.c
index ff7da40..90e2063 100644
--- a/init.c
+++ b/init.c
@@ -194,6 +194,8 @@
unsigned int bs = td->o.min_bs[ddir];
unsigned long long bytes_per_sec;
+ assert(ddir_rw(ddir));
+
if (td->o.rate[ddir])
bytes_per_sec = td->o.rate[ddir];
else
diff --git a/io_ddir.h b/io_ddir.h
index f83fc9b..b234256 100644
--- a/io_ddir.h
+++ b/io_ddir.h
@@ -34,4 +34,9 @@
ddir == DDIR_SYNC_FILE_RANGE;
}
+static inline int ddir_rw(enum fio_ddir ddir)
+{
+ return ddir == DDIR_READ || ddir == DDIR_WRITE;
+}
+
#endif
diff --git a/io_u.c b/io_u.c
index 53bf492..21a801f 100644
--- a/io_u.c
+++ b/io_u.c
@@ -103,6 +103,8 @@
unsigned long long max_blocks;
unsigned long long max_size;
+ assert(ddir_rw(ddir));
+
/*
* Hmm, should we make sure that ->io_size <= ->real_file_size?
*/
@@ -211,6 +213,8 @@
static int get_next_seq_block(struct thread_data *td, struct fio_file *f,
enum fio_ddir ddir, unsigned long long *b)
{
+ assert(ddir_rw(ddir));
+
if (f->last_pos < f->real_file_size) {
*b = (f->last_pos - f->file_offset) / td->o.min_bs[ddir];
return 0;
@@ -225,6 +229,8 @@
struct fio_file *f = io_u->file;
int ret;
+ assert(ddir_rw(ddir));
+
if (rw_seq) {
if (td_random(td))
ret = get_next_rand_block(td, f, ddir, b);
@@ -264,6 +270,8 @@
enum fio_ddir ddir = io_u->ddir;
int rw_seq_hit = 0;
+ assert(ddir_rw(ddir));
+
if (td->o.ddir_seq_nr && !--td->ddir_seq_nr) {
rw_seq_hit = 1;
td->ddir_seq_nr = td->o.ddir_seq_nr;
@@ -308,6 +316,8 @@
unsigned int minbs, maxbs;
long r;
+ assert(ddir_rw(ddir));
+
minbs = td->o.min_bs[ddir];
maxbs = td->o.max_bs[ddir];
@@ -388,6 +398,8 @@
struct timeval t;
long usec;
+ assert(ddir_rw(ddir));
+
if (td->rate_pending_usleep[ddir] <= 0)
return ddir;
@@ -531,7 +543,7 @@
td_io_u_lock(td);
__io_u->flags |= IO_U_F_FREE;
- if ((__io_u->flags & IO_U_F_FLIGHT) && !ddir_sync(__io_u->ddir))
+ if ((__io_u->flags & IO_U_F_FLIGHT) && ddir_rw(__io_u->ddir))
td->io_issues[__io_u->ddir]--;
__io_u->flags &= ~IO_U_F_FLIGHT;
@@ -551,9 +563,9 @@
io_u->ddir = get_rw_ddir(td);
/*
- * fsync() or fdatasync(), we are done
+ * fsync() or fdatasync() or trim etc, we are done
*/
- if (ddir_sync(io_u->ddir))
+ if (!ddir_rw(io_u->ddir))
goto out;
/*
@@ -1023,7 +1035,7 @@
f = io_u->file;
assert(fio_file_open(f));
- if (!ddir_sync(io_u->ddir)) {
+ if (ddir_rw(io_u->ddir)) {
if (!io_u->buflen && !(td->io_ops->flags & FIO_NOIO)) {
dprint(FD_IO, "get_io_u: zero buflen on %p\n", io_u);
goto err_put;
@@ -1114,7 +1126,7 @@
td->last_was_sync = 0;
td->last_ddir = io_u->ddir;
- if (!io_u->error) {
+ if (!io_u->error && ddir_rw(io_u->ddir)) {
unsigned int bytes = io_u->buflen - io_u->resid;
const enum fio_ddir idx = io_u->ddir;
const enum fio_ddir odx = io_u->ddir ^ 1;
@@ -1180,7 +1192,7 @@
if (ret && !icd->error)
icd->error = ret;
}
- } else {
+ } else if (io_u->error) {
icd->error = io_u->error;
io_u_log_error(td, io_u);
}
diff --git a/ioengines.c b/ioengines.c
index 4e059a8..f976efb 100644
--- a/ioengines.c
+++ b/ioengines.c
@@ -241,7 +241,7 @@
sizeof(struct timeval));
}
- if (!ddir_sync(io_u->ddir))
+ if (ddir_rw(io_u->ddir))
td->io_issues[io_u->ddir]++;
ret = td->io_ops->queue(td, io_u);
@@ -254,7 +254,7 @@
* IO, then it's likely an alignment problem or because the host fs
* does not support O_DIRECT
*/
- if (io_u->error == EINVAL && td->io_issues[io_u->ddir] == 1 &&
+ if (io_u->error == EINVAL && td->io_issues[io_u->ddir & 1] == 1 &&
td->o.odirect) {
log_info("fio: first direct IO errored. File system may not "
"support direct IO, or iomem_align= is bad.\n");
@@ -266,14 +266,14 @@
}
if (ret == FIO_Q_COMPLETED) {
- if (!ddir_sync(io_u->ddir)) {
+ if (ddir_rw(io_u->ddir)) {
io_u_mark_depth(td, 1);
td->ts.total_io_u[io_u->ddir]++;
}
} else if (ret == FIO_Q_QUEUED) {
int r;
- if (!ddir_sync(io_u->ddir)) {
+ if (ddir_rw(io_u->ddir)) {
td->io_u_queued++;
td->ts.total_io_u[io_u->ddir]++;
}
@@ -483,16 +483,16 @@
{
#ifndef FIO_HAVE_TRIM
io_u->error = EINVAL;
- return io_u->xfer_buflen;
+ return 0;
#else
struct fio_file *f = io_u->file;
int ret;
ret = os_trim(f->fd, io_u->offset + f->file_offset, io_u->xfer_buflen);
if (!ret)
- return 0;
+ return io_u->xfer_buflen;;
- io_u->error = errno;
- return io_u->xfer_buflen;
+ io_u->error = ret;
+ return 0;
#endif
}
diff --git a/log.c b/log.c
index 829de94..ce4ac9f 100644
--- a/log.c
+++ b/log.c
@@ -294,6 +294,8 @@
rw = DDIR_SYNC;
else if (!strcmp(act, "datasync"))
rw = DDIR_DATASYNC;
+ else if (!strcmp(act, "trim"))
+ rw = DDIR_TRIM;
else {
log_err("fio: bad iolog file action: %s\n",
act);
diff --git a/stat.c b/stat.c
index 8a0fab0..8e9fba0 100644
--- a/stat.c
+++ b/stat.c
@@ -157,6 +157,8 @@
char *io_p, *bw_p, *iops_p;
int i2p;
+ assert(ddir_rw(ddir));
+
if (!ts->runtime[ddir])
return;
@@ -370,6 +372,8 @@
unsigned long long bw;
double mean, dev;
+ assert(ddir_rw(ddir));
+
bw = 0;
if (ts->runtime[ddir])
bw = ts->io_bytes[ddir] / ts->runtime[ddir];
@@ -735,13 +739,20 @@
unsigned long val, enum fio_ddir ddir,
unsigned int bs)
{
+ if (!ddir_rw(ddir))
+ return;
+
__add_log_sample(iolog, val, ddir, bs, mtime_since_now(&td->epoch));
}
void add_agg_sample(unsigned long val, enum fio_ddir ddir, unsigned int bs)
{
- struct io_log *iolog = agg_io_log[ddir];
+ struct io_log *iolog;
+ if (!ddir_rw(ddir))
+ return;
+
+ iolog = agg_io_log[ddir];
__add_log_sample(iolog, val, ddir, bs, mtime_since_genesis());
}
@@ -750,6 +761,9 @@
{
struct thread_stat *ts = &td->ts;
+ if (!ddir_rw(ddir))
+ return;
+
add_stat_sample(&ts->clat_stat[ddir], usec);
if (ts->clat_log)
@@ -761,6 +775,9 @@
{
struct thread_stat *ts = &td->ts;
+ if (!ddir_rw(ddir))
+ return;
+
add_stat_sample(&ts->slat_stat[ddir], usec);
if (ts->slat_log)
@@ -772,6 +789,9 @@
{
struct thread_stat *ts = &td->ts;
+ if (!ddir_rw(ddir))
+ return;
+
add_stat_sample(&ts->lat_stat[ddir], usec);
if (ts->lat_log)
@@ -782,9 +802,12 @@
struct timeval *t)
{
struct thread_stat *ts = &td->ts;
- unsigned long spent = mtime_since(&ts->stat_sample_time[ddir], t);
- unsigned long rate;
+ unsigned long spent, rate;
+ if (!ddir_rw(ddir))
+ return;
+
+ spent = mtime_since(&ts->stat_sample_time[ddir], t);
if (spent < td->o.bw_avg_time)
return;