/*
* Just discard the pdu by seeking past it.
*/
static int discard_pdu(struct thread_data *td, struct fifo *fifo, int fd,
struct blk_io_trace *t)
{
if (t->pdu_len == 0)
return 0;
dprint(FD_BLKTRACE, "discard pdu len %u\n", t->pdu_len);
return trace_fifo_get(td, fifo, fd, NULL, t->pdu_len);
}
/*
* Load a blktrace file by reading all the blk_io_trace entries, and storing
* them as io_pieces like the fio text version would do.
*/
int load_blktrace(struct thread_data *td, const char *filename, int need_swap)
{
unsigned long long ttime, delay;
struct blk_io_trace t;
unsigned long ios[2], skipped_writes;
unsigned int cpu;
unsigned int rw_bs[2];
struct fifo *fifo;
int fd, i, old_state;
struct fio_file *f;
fd = open(filename, O_RDONLY);
if (fd < 0) {
td_verror(td, errno, "open blktrace file");
return 1;
}
fifo = fifo_alloc(TRACE_FIFO_SIZE);
old_state = td_bump_runstate(td, TD_SETTING_UP);
td->o.size = 0;
cpu = 0;
ttime = 0;
ios[0] = ios[1] = 0;
rw_bs[0] = rw_bs[1] = 0;
skipped_writes = 0;
do {
int ret = trace_fifo_get(td, fifo, fd, &t, sizeof(t));
if (ret < 0)
goto err;
else if (!ret)
break;
else if (ret < (int) sizeof(t)) {
log_err("fio: short fifo get\n");
break;
}
if (need_swap)
byteswap_trace(&t);
if ((t.magic & 0xffffff00) != BLK_IO_TRACE_MAGIC) {
log_err("fio: bad magic in blktrace data: %x\n",
t.magic);
goto err;
}
if ((t.magic & 0xff) != BLK_IO_TRACE_VERSION) {
log_err("fio: bad blktrace version %d\n",
t.magic & 0xff);
goto err;
}
ret = discard_pdu(td, fifo, fd, &t);
if (ret < 0) {
td_verror(td, ret, "blktrace lseek");
goto err;
} else if (t.pdu_len != ret) {
log_err("fio: discarded %d of %d\n", ret, t.pdu_len);
goto err;
}
if ((t.action & BLK_TC_ACT(BLK_TC_NOTIFY)) == 0) {
if (!ttime) {
ttime = t.time;
cpu = t.cpu;
}
delay = 0;
if (cpu == t.cpu)
delay = t.time - ttime;
if ((t.action & BLK_TC_ACT(BLK_TC_WRITE)) && read_only)
skipped_writes++;
else {
/*
* set delay to zero if no_stall enabled for
* fast replay
*/
if (td->o.no_stall)
delay = 0;
handle_trace(td, &t, delay, ios, rw_bs);
}
ttime = t.time;
cpu = t.cpu;
} else {
delay = 0;
handle_trace(td, &t, delay, ios, rw_bs);
}
} while (1);
for (i = 0; i < td->files_index; i++) {
f = td->files[i];
trace_add_open_close_event(td, f->fileno, FIO_LOG_CLOSE_FILE);
}
fifo_free(fifo);
close(fd);
td_restore_runstate(td, old_state);
if (!td->files_index) {
log_err("fio: did not find replay device(s)\n");
return 1;
}
if (skipped_writes)
log_err("fio: %s skips replay of %lu writes due to read-only\n",
td->o.name, skipped_writes);
if (!ios[DDIR_READ] && !ios[DDIR_WRITE]) {
log_err("fio: found no ios in blktrace data\n");
return 1;
} else if (ios[DDIR_READ] && !ios[DDIR_READ]) {
td->o.td_ddir = TD_DDIR_READ;
td->o.max_bs[DDIR_READ] = rw_bs[DDIR_READ];
} else if (!ios[DDIR_READ] && ios[DDIR_WRITE]) {
td->o.td_ddir = TD_DDIR_WRITE;
td->o.max_bs[DDIR_WRITE] = rw_bs[DDIR_WRITE];
} else {
td->o.td_ddir = TD_DDIR_RW;
td->o.max_bs[DDIR_READ] = rw_bs[DDIR_READ];
td->o.max_bs[DDIR_WRITE] = rw_bs[DDIR_WRITE];
}
/*
* We need to do direct/raw ios to the device, to avoid getting
* read-ahead in our way.
*/
td->o.odirect = 1;
return 0;
err:
close(fd);
fifo_free(fifo);
return 1;
}
/*
* Load a blktrace file by reading all the blk_io_trace entries, and storing
* them as io_pieces like the fio text version would do.
*/
int load_blktrace(struct thread_data *td, const char *filename, int need_swap)
{
struct blk_io_trace t;
unsigned long ios[DDIR_RWDIR_CNT], skipped_writes;
unsigned int rw_bs[DDIR_RWDIR_CNT];
struct fifo *fifo;
int fd, i, old_state;
struct fio_file *f;
int this_depth, depth;
fd = open(filename, O_RDONLY);
if (fd < 0) {
td_verror(td, errno, "open blktrace file");
return 1;
}
fifo = fifo_alloc(TRACE_FIFO_SIZE);
old_state = td_bump_runstate(td, TD_SETTING_UP);
td->o.size = 0;
ios[0] = ios[1] = 0;
rw_bs[0] = rw_bs[1] = 0;
skipped_writes = 0;
this_depth = depth = 0;
do {
int ret = trace_fifo_get(td, fifo, fd, &t, sizeof(t));
if (ret < 0)
goto err;
else if (!ret)
break;
else if (ret < (int) sizeof(t)) {
log_err("fio: short fifo get\n");
break;
}
if (need_swap)
byteswap_trace(&t);
if ((t.magic & 0xffffff00) != BLK_IO_TRACE_MAGIC) {
log_err("fio: bad magic in blktrace data: %x\n",
t.magic);
goto err;
}
if ((t.magic & 0xff) != BLK_IO_TRACE_VERSION) {
log_err("fio: bad blktrace version %d\n",
t.magic & 0xff);
goto err;
}
ret = discard_pdu(td, fifo, fd, &t);
if (ret < 0) {
td_verror(td, ret, "blktrace lseek");
goto err;
} else if (t.pdu_len != ret) {
log_err("fio: discarded %d of %d\n", ret, t.pdu_len);
goto err;
}
if ((t.action & BLK_TC_ACT(BLK_TC_NOTIFY)) == 0) {
if ((t.action & 0xffff) == __BLK_TA_QUEUE)
this_depth++;
else if ((t.action & 0xffff) == __BLK_TA_COMPLETE) {
depth = max(depth, this_depth);
this_depth = 0;
}
if (t_is_write(&t) && read_only) {
skipped_writes++;
continue;
}
}
handle_trace(td, &t, ios, rw_bs);
} while (1);
for (i = 0; i < td->files_index; i++) {
f = td->files[i];
trace_add_open_close_event(td, f->fileno, FIO_LOG_CLOSE_FILE);
}
fifo_free(fifo);
close(fd);
td_restore_runstate(td, old_state);
if (!td->files_index) {
log_err("fio: did not find replay device(s)\n");
return 1;
}
/*
* For stacked devices, we don't always get a COMPLETE event so
* the depth grows to insane values. Limit it to something sane(r).
*/
if (!depth || depth > 1024)
depth = 1024;
if (skipped_writes)
log_err("fio: %s skips replay of %lu writes due to read-only\n",
td->o.name, skipped_writes);
if (!ios[DDIR_READ] && !ios[DDIR_WRITE]) {
log_err("fio: found no ios in blktrace data\n");
return 1;
} else if (ios[DDIR_READ] && !ios[DDIR_WRITE]) {
td->o.td_ddir = TD_DDIR_READ;
td->o.max_bs[DDIR_READ] = rw_bs[DDIR_READ];
} else if (!ios[DDIR_READ] && ios[DDIR_WRITE]) {
td->o.td_ddir = TD_DDIR_WRITE;
td->o.max_bs[DDIR_WRITE] = rw_bs[DDIR_WRITE];
} else {
td->o.td_ddir = TD_DDIR_RW;
td->o.max_bs[DDIR_READ] = rw_bs[DDIR_READ];
td->o.max_bs[DDIR_WRITE] = rw_bs[DDIR_WRITE];
td->o.max_bs[DDIR_TRIM] = rw_bs[DDIR_TRIM];
}
/*
* We need to do direct/raw ios to the device, to avoid getting
* read-ahead in our way.
*/
td->o.odirect = 1;
/*
* we don't know if this option was set or not. it defaults to 1,
* so we'll just guess that we should override it if it's still 1
*/
if (td->o.iodepth != 1)
td->o.iodepth = depth;
return 0;
err:
close(fd);
fifo_free(fifo);
return 1;
}
/*
* Load a blktrace file by reading all the blk_io_trace entries, and storing
* them as io_pieces like the fio text version would do.
*/
int load_blktrace(struct thread_data *td, const char *filename, int need_swap)
{
struct blk_io_trace t;
unsigned long ios[DDIR_RWDIR_CNT], skipped_writes;
unsigned int rw_bs[DDIR_RWDIR_CNT];
struct fifo *fifo;
int fd, i, old_state;
struct fio_file *f;
int this_depth[DDIR_RWDIR_CNT], depth[DDIR_RWDIR_CNT], max_depth;
fd = open(filename, O_RDONLY);
if (fd < 0) {
td_verror(td, errno, "open blktrace file");
return 1;
}
fifo = fifo_alloc(TRACE_FIFO_SIZE);
old_state = td_bump_runstate(td, TD_SETTING_UP);
td->o.size = 0;
for (i = 0; i < DDIR_RWDIR_CNT; i++) {
ios[i] = 0;
rw_bs[i] = 0;
this_depth[i] = 0;
depth[i] = 0;
}
skipped_writes = 0;
do {
int ret = trace_fifo_get(td, fifo, fd, &t, sizeof(t));
if (ret < 0)
goto err;
else if (!ret)
break;
else if (ret < (int) sizeof(t)) {
log_err("fio: short fifo get\n");
break;
}
if (need_swap)
byteswap_trace(&t);
if ((t.magic & 0xffffff00) != BLK_IO_TRACE_MAGIC) {
log_err("fio: bad magic in blktrace data: %x\n",
t.magic);
goto err;
}
if ((t.magic & 0xff) != BLK_IO_TRACE_VERSION) {
log_err("fio: bad blktrace version %d\n",
t.magic & 0xff);
goto err;
}
ret = discard_pdu(td, fifo, fd, &t);
if (ret < 0) {
td_verror(td, ret, "blktrace lseek");
goto err;
} else if (t.pdu_len != ret) {
log_err("fio: discarded %d of %d\n", ret, t.pdu_len);
goto err;
}
if ((t.action & BLK_TC_ACT(BLK_TC_NOTIFY)) == 0) {
if ((t.action & 0xffff) == __BLK_TA_QUEUE)
depth_inc(&t, this_depth);
else if (((t.action & 0xffff) == __BLK_TA_BACKMERGE) ||
((t.action & 0xffff) == __BLK_TA_FRONTMERGE))
depth_dec(&t, this_depth);
else if ((t.action & 0xffff) == __BLK_TA_COMPLETE)
depth_end(&t, this_depth, depth);
if (t_is_write(&t) && read_only) {
skipped_writes++;
continue;
}
}
handle_trace(td, &t, ios, rw_bs);
} while (1);
for (i = 0; i < td->files_index; i++) {
f = td->files[i];
trace_add_open_close_event(td, f->fileno, FIO_LOG_CLOSE_FILE);
}
fifo_free(fifo);
close(fd);
td_restore_runstate(td, old_state);
if (!td->files_index) {
log_err("fio: did not find replay device(s)\n");
return 1;
}
/*
* For stacked devices, we don't always get a COMPLETE event so
* the depth grows to insane values. Limit it to something sane(r).
*/
max_depth = 0;
for (i = 0; i < DDIR_RWDIR_CNT; i++) {
if (depth[i] > 1024)
depth[i] = 1024;
else if (!depth[i] && ios[i])
depth[i] = 1;
max_depth = max(depth[i], max_depth);
}
if (skipped_writes)
log_err("fio: %s skips replay of %lu writes due to read-only\n",
td->o.name, skipped_writes);
if (!ios[DDIR_READ] && !ios[DDIR_WRITE]) {
log_err("fio: found no ios in blktrace data\n");
return 1;
} else if (ios[DDIR_READ] && !ios[DDIR_WRITE]) {
td->o.td_ddir = TD_DDIR_READ;
td->o.max_bs[DDIR_READ] = rw_bs[DDIR_READ];
} else if (!ios[DDIR_READ] && ios[DDIR_WRITE]) {
td->o.td_ddir = TD_DDIR_WRITE;
td->o.max_bs[DDIR_WRITE] = rw_bs[DDIR_WRITE];
} else {
td->o.td_ddir = TD_DDIR_RW;
td->o.max_bs[DDIR_READ] = rw_bs[DDIR_READ];
td->o.max_bs[DDIR_WRITE] = rw_bs[DDIR_WRITE];
td->o.max_bs[DDIR_TRIM] = rw_bs[DDIR_TRIM];
}
/*
* We need to do direct/raw ios to the device, to avoid getting
* read-ahead in our way. But only do so if the minimum block size
* is a multiple of 4k, otherwise we don't know if it's safe to do so.
*/
if (!fio_option_is_set(&td->o, odirect) && !(td_min_bs(td) & 4095))
td->o.odirect = 1;
/*
* If depth wasn't manually set, use probed depth
*/
if (!fio_option_is_set(&td->o, iodepth))
td->o.iodepth = td->o.iodepth_low = max_depth;
return 0;
err:
close(fd);
fifo_free(fifo);
return 1;
}