static void
zpios_remove_objset(run_args_t *run_args)
{
zpios_time_t *t = &(run_args->stats.rm_time);
zpios_region_t *region;
char name[32];
int rc = 0, i;
(void)zpios_upcall(run_args->pre, PHASE_PRE_REMOVE, run_args, 0);
t->start = zpios_timespec_now();
(void)snprintf(name, 32, "%s/id_%d", run_args->pool, run_args->id);
if (run_args->flags & DMU_REMOVE) {
if (run_args->flags & DMU_FPP) {
for (i = 0; i < run_args->region_count; i++) {
region = &run_args->regions[i];
rc = zpios_dmu_object_free(run_args,
region->obj.os,
region->obj.obj);
if (rc)
zpios_print(run_args->file, "Error "
"removing object %d, %d\n",
(int)region->obj.obj, rc);
}
} else {
region = &run_args->regions[0];
rc = zpios_dmu_object_free(run_args,
region->obj.os,
region->obj.obj);
if (rc)
zpios_print(run_args->file, "Error "
"removing object %d, %d\n",
(int)region->obj.obj, rc);
}
}
dmu_objset_disown(run_args->os, zpios_tag);
if (run_args->flags & DMU_REMOVE) {
rc = dsl_destroy_head(name);
if (rc)
zpios_print(run_args->file, "Error dsl_destroy_head"
"(%s, ...) failed: %d\n", name, rc);
}
t->stop = zpios_timespec_now();
t->delta = zpios_timespec_sub(t->stop, t->start);
(void)zpios_upcall(run_args->post, PHASE_POST_REMOVE, run_args, rc);
}
static int
zpios_setup_run(run_args_t **run_args, zpios_cmd_t *kcmd, struct file *file)
{
run_args_t *ra;
int rc, size;
size = sizeof (*ra) + kcmd->cmd_region_count * sizeof (zpios_region_t);
ra = vmem_zalloc(size, KM_SLEEP);
if (ra == NULL) {
zpios_print(file, "Unable to vmem_zalloc() %d bytes "
"for regions\n", size);
return (-ENOMEM);
}
*run_args = ra;
strncpy(ra->pool, kcmd->cmd_pool, ZPIOS_NAME_SIZE - 1);
strncpy(ra->pre, kcmd->cmd_pre, ZPIOS_PATH_SIZE - 1);
strncpy(ra->post, kcmd->cmd_post, ZPIOS_PATH_SIZE - 1);
strncpy(ra->log, kcmd->cmd_log, ZPIOS_PATH_SIZE - 1);
ra->id = kcmd->cmd_id;
ra->chunk_size = kcmd->cmd_chunk_size;
ra->thread_count = kcmd->cmd_thread_count;
ra->region_count = kcmd->cmd_region_count;
ra->region_size = kcmd->cmd_region_size;
ra->offset = kcmd->cmd_offset;
ra->region_noise = kcmd->cmd_region_noise;
ra->chunk_noise = kcmd->cmd_chunk_noise;
ra->thread_delay = kcmd->cmd_thread_delay;
ra->flags = kcmd->cmd_flags;
ra->block_size = kcmd->cmd_block_size;
ra->stats.wr_data = 0;
ra->stats.wr_chunks = 0;
ra->stats.rd_data = 0;
ra->stats.rd_chunks = 0;
ra->region_next = 0;
ra->file = file;
mutex_init(&ra->lock_work, NULL, MUTEX_DEFAULT, NULL);
mutex_init(&ra->lock_ctl, NULL, MUTEX_DEFAULT, NULL);
(void) zpios_upcall(ra->pre, PHASE_PRE_RUN, ra, 0);
rc = zpios_dmu_setup(ra);
if (rc) {
mutex_destroy(&ra->lock_ctl);
mutex_destroy(&ra->lock_work);
vmem_free(ra, size);
*run_args = NULL;
}
return (rc);
}
static int
zpios_do_one_run(struct file *file, zpios_cmd_t *kcmd,
int data_size, void *data)
{
run_args_t *run_args = { 0 };
zpios_stats_t *stats = (zpios_stats_t *)data;
int i, n, m, size, rc;
if ((!kcmd->cmd_chunk_size) || (!kcmd->cmd_region_size) ||
(!kcmd->cmd_thread_count) || (!kcmd->cmd_region_count)) {
zpios_print(file, "Invalid chunk_size, region_size, "
"thread_count, or region_count, %d\n", -EINVAL);
return (-EINVAL);
}
if (!(kcmd->cmd_flags & DMU_WRITE) ||
!(kcmd->cmd_flags & DMU_READ)) {
zpios_print(file, "Invalid flags, minimally DMU_WRITE "
"and DMU_READ must be set, %d\n", -EINVAL);
return (-EINVAL);
}
if ((kcmd->cmd_flags & (DMU_WRITE_ZC | DMU_READ_ZC)) &&
(kcmd->cmd_flags & DMU_VERIFY)) {
zpios_print(file, "Invalid flags, DMU_*_ZC incompatible "
"with DMU_VERIFY, used for performance analysis "
"only, %d\n", -EINVAL);
return (-EINVAL);
}
/*
* Opaque data on return contains structs of the following form:
*
* zpios_stat_t stats[];
* stats[0] = run_args->stats;
* stats[1-N] = threads[N]->stats;
* stats[N+1-M] = regions[M]->stats;
*
* Where N is the number of threads, and M is the number of regions.
*/
size = (sizeof (zpios_stats_t) +
(kcmd->cmd_thread_count * sizeof (zpios_stats_t)) +
(kcmd->cmd_region_count * sizeof (zpios_stats_t)));
if (data_size < size) {
zpios_print(file, "Invalid size, command data buffer "
"size too small, (%d < %d)\n", data_size, size);
return (-ENOSPC);
}
rc = zpios_setup_run(&run_args, kcmd, file);
if (rc)
return (rc);
rc = zpios_threads_run(run_args);
zpios_remove_objset(run_args);
if (rc)
goto cleanup;
if (stats) {
n = 1;
m = 1 + kcmd->cmd_thread_count;
stats[0] = run_args->stats;
for (i = 0; i < kcmd->cmd_thread_count; i++)
stats[n+i] = run_args->threads[i]->stats;
for (i = 0; i < kcmd->cmd_region_count; i++)
stats[m+i] = run_args->regions[i].stats;
}
cleanup:
zpios_cleanup_run(run_args);
(void) zpios_upcall(kcmd->cmd_post, PHASE_POST_RUN, run_args, 0);
return (rc);
}
static int
zpios_threads_run(run_args_t *run_args)
{
struct task_struct *tsk, **tsks;
thread_data_t *thr = NULL;
zpios_time_t *tt = &(run_args->stats.total_time);
zpios_time_t *tw = &(run_args->stats.wr_time);
zpios_time_t *tr = &(run_args->stats.rd_time);
int i, rc = 0, tc = run_args->thread_count;
tsks = kmem_zalloc(sizeof (struct task_struct *) * tc, KM_SLEEP);
if (tsks == NULL) {
rc = -ENOMEM;
goto cleanup2;
}
run_args->threads = kmem_zalloc(sizeof (thread_data_t *)*tc, KM_SLEEP);
if (run_args->threads == NULL) {
rc = -ENOMEM;
goto cleanup;
}
init_waitqueue_head(&run_args->waitq);
run_args->threads_done = 0;
/* Create all the needed threads which will sleep until awoken */
for (i = 0; i < tc; i++) {
thr = kmem_zalloc(sizeof (thread_data_t), KM_SLEEP);
if (thr == NULL) {
rc = -ENOMEM;
goto taskerr;
}
thr->thread_no = i;
thr->run_args = run_args;
thr->rc = 0;
mutex_init(&thr->lock, NULL, MUTEX_DEFAULT, NULL);
run_args->threads[i] = thr;
tsk = kthread_create(zpios_thread_main, (void *)thr,
"%s/%d", "zpios_io", i);
if (IS_ERR(tsk)) {
rc = -EINVAL;
goto taskerr;
}
tsks[i] = tsk;
}
tt->start = zpios_timespec_now();
/* Wake up all threads for write phase */
(void) zpios_upcall(run_args->pre, PHASE_PRE_WRITE, run_args, 0);
for (i = 0; i < tc; i++)
wake_up_process(tsks[i]);
/* Wait for write phase to complete */
tw->start = zpios_timespec_now();
wait_event(run_args->waitq, zpios_thread_done(run_args));
tw->stop = zpios_timespec_now();
(void) zpios_upcall(run_args->post, PHASE_POST_WRITE, run_args, rc);
for (i = 0; i < tc; i++) {
thr = run_args->threads[i];
mutex_enter(&thr->lock);
if (!rc && thr->rc)
rc = thr->rc;
run_args->stats.wr_data += thr->stats.wr_data;
run_args->stats.wr_chunks += thr->stats.wr_chunks;
mutex_exit(&thr->lock);
}
if (rc) {
/* Wake up all threads and tell them to exit */
for (i = 0; i < tc; i++) {
mutex_enter(&thr->lock);
thr->rc = rc;
mutex_exit(&thr->lock);
wake_up_process(tsks[i]);
}
goto out;
}
mutex_enter(&run_args->lock_ctl);
ASSERT(run_args->threads_done == run_args->thread_count);
run_args->threads_done = 0;
mutex_exit(&run_args->lock_ctl);
/* Wake up all threads for read phase */
(void) zpios_upcall(run_args->pre, PHASE_PRE_READ, run_args, 0);
for (i = 0; i < tc; i++)
wake_up_process(tsks[i]);
/* Wait for read phase to complete */
tr->start = zpios_timespec_now();
wait_event(run_args->waitq, zpios_thread_done(run_args));
tr->stop = zpios_timespec_now();
(void) zpios_upcall(run_args->post, PHASE_POST_READ, run_args, rc);
for (i = 0; i < tc; i++) {
thr = run_args->threads[i];
mutex_enter(&thr->lock);
if (!rc && thr->rc)
rc = thr->rc;
run_args->stats.rd_data += thr->stats.rd_data;
run_args->stats.rd_chunks += thr->stats.rd_chunks;
mutex_exit(&thr->lock);
}
out:
tt->stop = zpios_timespec_now();
tt->delta = zpios_timespec_sub(tt->stop, tt->start);
tw->delta = zpios_timespec_sub(tw->stop, tw->start);
tr->delta = zpios_timespec_sub(tr->stop, tr->start);
cleanup:
kmem_free(tsks, sizeof (struct task_struct *) * tc);
cleanup2:
/* Returns first encountered thread error (if any) */
return (rc);
taskerr:
/* Destroy all threads that were created successfully */
for (i = 0; i < tc; i++)
if (tsks[i] != NULL)
(void) kthread_stop(tsks[i]);
goto cleanup;
}
static int
zpios_dmu_setup(run_args_t *run_args)
{
zpios_time_t *t = &(run_args->stats.cr_time);
objset_t *os;
char name[32];
uint64_t obj = 0ULL;
int i, rc = 0, rc2;
(void) zpios_upcall(run_args->pre, PHASE_PRE_CREATE, run_args, 0);
t->start = zpios_timespec_now();
(void) snprintf(name, 32, "%s/id_%d", run_args->pool, run_args->id);
rc = dmu_objset_create(name, DMU_OST_OTHER, 0, NULL, NULL);
if (rc) {
zpios_print(run_args->file, "Error dmu_objset_create(%s, ...) "
"failed: %d\n", name, rc);
goto out;
}
rc = dmu_objset_own(name, DMU_OST_OTHER, 0, zpios_tag, &os);
if (rc) {
zpios_print(run_args->file, "Error dmu_objset_own(%s, ...) "
"failed: %d\n", name, rc);
goto out_destroy;
}
if (!(run_args->flags & DMU_FPP)) {
obj = zpios_dmu_object_create(run_args, os);
if (obj == 0) {
rc = -EBADF;
zpios_print(run_args->file, "Error zpios_dmu_"
"object_create() failed, %d\n", rc);
goto out_destroy;
}
}
for (i = 0; i < run_args->region_count; i++) {
zpios_region_t *region;
region = &run_args->regions[i];
mutex_init(®ion->lock, NULL, MUTEX_DEFAULT, NULL);
if (run_args->flags & DMU_FPP) {
/* File per process */
region->obj.os = os;
region->obj.obj = zpios_dmu_object_create(run_args, os);
ASSERT(region->obj.obj > 0); /* XXX - Handle this */
region->wr_offset = run_args->offset;
region->rd_offset = run_args->offset;
region->init_offset = run_args->offset;
region->max_offset = run_args->offset +
run_args->region_size;
} else {
/* Single shared file */
region->obj.os = os;
region->obj.obj = obj;
region->wr_offset = run_args->offset * i;
region->rd_offset = run_args->offset * i;
region->init_offset = run_args->offset * i;
region->max_offset = run_args->offset *
i + run_args->region_size;
}
}
run_args->os = os;
out_destroy:
if (rc) {
rc2 = dsl_destroy_head(name);
if (rc2)
zpios_print(run_args->file, "Error dsl_destroy_head"
"(%s, ...) failed: %d\n", name, rc2);
}
out:
t->stop = zpios_timespec_now();
t->delta = zpios_timespec_sub(t->stop, t->start);
(void) zpios_upcall(run_args->post, PHASE_POST_CREATE, run_args, rc);
return (rc);
}