# include <ctype.h>
#endif
+#include <libcfs/kp30.h>
#include <lustre_net.h>
# include <lustre_lib.h>
#include <lustre_ha.h>
#include <obd_class.h> /* for obd_zombie */
#include <obd_support.h> /* for OBD_FAIL_CHECK */
-#include <cl_object.h> /* cl_env_{get,put}() */
#include <lprocfs_status.h>
-enum pscope_thread {
- PT_NORMAL,
- PT_RECOVERY,
- PT_NR
-};
-
-struct ptlrpcd_scope_ctl {
- struct ptlrpcd_thread {
- const char *pt_name;
- struct ptlrpcd_ctl pt_ctl;
- } pscope_thread[PT_NR];
-};
-
-static struct ptlrpcd_scope_ctl ptlrpcd_scopes[PSCOPE_NR] = {
- [PSCOPE_BRW] = {
- .pscope_thread = {
- [PT_NORMAL] = {
- .pt_name = "ptlrpcd-brw"
- },
- [PT_RECOVERY] = {
- .pt_name = "ptlrpcd-brw-rcv"
- }
- }
- },
- [PSCOPE_OTHER] = {
- .pscope_thread = {
- [PT_NORMAL] = {
- .pt_name = "ptlrpcd"
- },
- [PT_RECOVERY] = {
- .pt_name = "ptlrpcd-rcv"
- }
- }
- }
-};
+static struct ptlrpcd_ctl ptlrpcd_pc;
+static struct ptlrpcd_ctl ptlrpcd_recovery_pc;
struct semaphore ptlrpcd_sem;
static int ptlrpcd_users = 0;
}
/*
+ * Move all request from an existing request set to the ptlrpcd queue.
+ * All requests from the set must be in phase RQ_PHASE_NEW.
+ */
+void ptlrpcd_add_rqset(struct ptlrpc_request_set *set)
+{
+ struct list_head *tmp, *pos;
+
+ list_for_each_safe(pos, tmp, &set->set_requests) {
+ struct ptlrpc_request *req =
+ list_entry(pos, struct ptlrpc_request, rq_set_chain);
+
+ LASSERT(req->rq_phase == RQ_PHASE_NEW);
+ list_del_init(&req->rq_set_chain);
+ req->rq_set = NULL;
+ ptlrpcd_add_req(req);
+ set->set_remaining--;
+ }
+ LASSERT(set->set_remaining == 0);
+}
+EXPORT_SYMBOL(ptlrpcd_add_rqset);
+
+/*
* Requests that are added to the ptlrpcd queue are sent via
* ptlrpcd_check->ptlrpc_check_set().
*/
-void ptlrpcd_add_req(struct ptlrpc_request *req, enum ptlrpcd_scope scope)
+int ptlrpcd_add_req(struct ptlrpc_request *req)
{
struct ptlrpcd_ctl *pc;
- enum pscope_thread pt;
int rc;
- LASSERT(scope < PSCOPE_NR);
- pt = req->rq_send_state == LUSTRE_IMP_FULL ? PT_NORMAL : PT_RECOVERY;
- pc = &ptlrpcd_scopes[scope].pscope_thread[pt].pt_ctl;
+ if (req->rq_send_state == LUSTRE_IMP_FULL)
+ pc = &ptlrpcd_pc;
+ else
+ pc = &ptlrpcd_recovery_pc;
rc = ptlrpc_set_add_new_req(pc, req);
- /*
- * XXX disable this for CLIO: environment is needed for interpreter.
- */
- if (rc && 0) {
- ptlrpc_interpterer_t interpreter;
-
- interpreter = req->rq_interpret_reply;
-
+ if (rc) {
/*
* Thread is probably in stop now so we need to
* kill this rpc as it was not added. Let's call
* interpret for it to let know we're killing it
* so that higher levels might free assosiated
* resources.
- */
- req->rq_status = -EBADR;
- interpreter(NULL, req, &req->rq_async_args,
- req->rq_status);
+ */
+
+ ptlrpc_req_interpret(req, -EBADR);
req->rq_set = NULL;
ptlrpc_req_finished(req);
}
+ return rc;
}
-static int ptlrpcd_check(const struct lu_env *env, struct ptlrpcd_ctl *pc)
+static int ptlrpcd_check(struct ptlrpcd_ctl *pc)
{
struct list_head *tmp, *pos;
struct ptlrpc_request *req;
int rc = 0;
ENTRY;
- if (test_bit(LIOD_STOP, &pc->pc_flags))
- RETURN(1);
-
spin_lock(&pc->pc_set->set_new_req_lock);
list_for_each_safe(pos, tmp, &pc->pc_set->set_new_requests) {
req = list_entry(pos, struct ptlrpc_request, rq_set_chain);
spin_unlock(&pc->pc_set->set_new_req_lock);
if (pc->pc_set->set_remaining) {
- rc = rc | ptlrpc_check_set(env, pc->pc_set);
+ rc = rc | ptlrpc_check_set(pc->pc_set);
/*
* XXX: our set never completes, so we prune the completed
static int ptlrpcd(void *arg)
{
struct ptlrpcd_ctl *pc = arg;
- struct lu_env env = { .le_ses = NULL };
- int rc;
+ int rc, exit = 0;
ENTRY;
- rc = cfs_daemonize_ctxt(pc->pc_name);
- if (rc == 0) {
- /*
- * XXX So far only "client" ptlrpcd uses an environment. In
- * the future, ptlrpcd thread (or a thread-set) has to given
- * an argument, describing its "scope".
- */
- rc = lu_context_init(&env.le_ctx,
- LCT_CL_THREAD|LCT_REMEMBER|LCT_NOREF);
+ if ((rc = cfs_daemonize_ctxt(pc->pc_name))) {
+ complete(&pc->pc_starting);
+ goto out;
}
complete(&pc->pc_starting);
- if (rc != 0)
- RETURN(rc);
- env.le_ctx.lc_cookie = 0x7;
/*
* This mainloop strongly resembles ptlrpc_set_wait() except that our
* set never completes. ptlrpcd_check() calls ptlrpc_check_set() when
* there are requests in the set. New requests come in on the set's
* new_req_list and ptlrpcd_check() moves them into the set.
*/
- while (1) {
+ do {
struct l_wait_info lwi;
int timeout;
- rc = lu_env_refill(&env);
- if (rc != 0) {
- /*
- * XXX This is very awkward situation, because
- * execution can neither continue (request
- * interpreters assume that env is set up), nor repeat
- * the loop (as this potentially results in a tight
- * loop of -ENOMEM's).
- *
- * Fortunately, refill only ever does something when
- * new modules are loaded, i.e., early during boot up.
- */
- CERROR("Failure to refill session: %d\n", rc);
- continue;
- }
-
timeout = ptlrpc_set_next_timeout(pc->pc_set);
lwi = LWI_TIMEOUT(cfs_time_seconds(timeout ? timeout : 1),
ptlrpc_expired_set, pc->pc_set);
- lu_context_enter(&env.le_ctx);
- l_wait_event(pc->pc_set->set_waitq,
- ptlrpcd_check(&env, pc), &lwi);
- lu_context_exit(&env.le_ctx);
+ l_wait_event(pc->pc_set->set_waitq, ptlrpcd_check(pc), &lwi);
/*
* Abort inflight rpcs for forced stop case.
*/
- if (test_bit(LIOD_STOP_FORCE, &pc->pc_flags))
- ptlrpc_abort_set(pc->pc_set);
+ if (test_bit(LIOD_STOP, &pc->pc_flags)) {
+ if (test_bit(LIOD_FORCE, &pc->pc_flags))
+ ptlrpc_abort_set(pc->pc_set);
+ exit++;
+ }
- if (test_bit(LIOD_STOP, &pc->pc_flags))
- break;
- }
+ /*
+ * Let's make one more loop to make sure that ptlrpcd_check()
+ * copied all raced new rpcs into the set so we can kill them.
+ */
+ } while (exit < 2);
/*
* Wait for inflight requests to drain.
*/
if (!list_empty(&pc->pc_set->set_requests))
ptlrpc_set_wait(pc->pc_set);
- lu_context_fini(&env.le_ctx);
- complete(&pc->pc_finishing);
+ complete(&pc->pc_finishing);
+out:
clear_bit(LIOD_START, &pc->pc_flags);
clear_bit(LIOD_STOP, &pc->pc_flags);
+ clear_bit(LIOD_FORCE, &pc->pc_flags);
return 0;
}
-#else /* !__KERNEL__ */
+#else
int ptlrpcd_check_async_rpcs(void *arg)
{
struct ptlrpcd_ctl *pc = arg;
- int rc = 0;
+ int rc = 0;
/*
* Single threaded!!
pc->pc_recurred++;
if (pc->pc_recurred == 1) {
- lu_context_enter(&pc->pc_env.le_ctx);
- rc = ptlrpcd_check(&pc->pc_env, pc);
- lu_context_exit(&pc->pc_env.le_ctx);
+ rc = ptlrpcd_check(pc);
if (!rc)
ptlrpc_expired_set(pc->pc_set);
/*
* XXX: send replay requests.
*/
- if (test_bit(LIOD_RECOVERY, &pc->pc_flags))
- rc = ptlrpcd_check(&pc->pc_env, pc);
+ if (pc == &ptlrpcd_recovery_pc)
+ rc = ptlrpcd_check(pc);
}
pc->pc_recurred--;
#endif
-int ptlrpcd_start(const char *name, struct ptlrpcd_ctl *pc)
+int ptlrpcd_start(char *name, struct ptlrpcd_ctl *pc)
{
- int rc;
+ int rc = 0;
ENTRY;
/*
* Do not allow start second thread for one pc.
*/
- if (test_and_set_bit(LIOD_START, &pc->pc_flags)) {
+ if (test_bit(LIOD_START, &pc->pc_flags)) {
CERROR("Starting second thread (%s) for same pc %p\n",
name, pc);
RETURN(-EALREADY);
}
+ set_bit(LIOD_START, &pc->pc_flags);
init_completion(&pc->pc_starting);
init_completion(&pc->pc_finishing);
spin_lock_init(&pc->pc_lock);
- snprintf (pc->pc_name, sizeof (pc->pc_name), name);
+ strncpy(pc->pc_name, name, sizeof(pc->pc_name) - 1);
+
pc->pc_set = ptlrpc_prep_set();
if (pc->pc_set == NULL)
GOTO(out, rc = -ENOMEM);
- /*
- * So far only "client" ptlrpcd uses an environment. In the future,
- * ptlrpcd thread (or a thread-set) has to be given an argument,
- * describing its "scope".
- */
- rc = lu_context_init(&pc->pc_env.le_ctx, LCT_CL_THREAD|LCT_REMEMBER);
- if (rc != 0) {
- ptlrpc_set_destroy(pc->pc_set);
- GOTO(out, rc);
- }
#ifdef __KERNEL__
rc = cfs_kernel_thread(ptlrpcd, pc, 0);
if (rc < 0) {
- lu_context_fini(&pc->pc_env.le_ctx);
ptlrpc_set_destroy(pc->pc_set);
GOTO(out, rc);
}
set_bit(LIOD_STOP, &pc->pc_flags);
if (force)
- set_bit(LIOD_STOP_FORCE, &pc->pc_flags);
+ set_bit(LIOD_FORCE, &pc->pc_flags);
cfs_waitq_signal(&pc->pc_set->set_waitq);
#ifdef __KERNEL__
wait_for_completion(&pc->pc_finishing);
liblustre_deregister_wait_callback(pc->pc_wait_callback);
liblustre_deregister_idle_callback(pc->pc_idle_callback);
#endif
- lu_context_fini(&pc->pc_env.le_ctx);
ptlrpc_set_destroy(pc->pc_set);
}
-void ptlrpcd_fini(void)
-{
- int i;
- int j;
-
- ENTRY;
-
- for (i = 0; i < PSCOPE_NR; ++i) {
- for (j = 0; j < PT_NR; ++j) {
- struct ptlrpcd_ctl *pc;
-
- pc = &ptlrpcd_scopes[i].pscope_thread[j].pt_ctl;
-
- if (test_bit(LIOD_START, &pc->pc_flags))
- ptlrpcd_stop(pc, 0);
- }
- }
- EXIT;
-}
-
int ptlrpcd_addref(void)
{
int rc = 0;
- int i;
- int j;
ENTRY;
mutex_down(&ptlrpcd_sem);
- if (++ptlrpcd_users == 1) {
- for (i = 0; rc == 0 && i < PSCOPE_NR; ++i) {
- for (j = 0; rc == 0 && j < PT_NR; ++j) {
- struct ptlrpcd_thread *pt;
- struct ptlrpcd_ctl *pc;
-
- pt = &ptlrpcd_scopes[i].pscope_thread[j];
- pc = &pt->pt_ctl;
- if (j == PT_RECOVERY)
- set_bit(LIOD_RECOVERY, &pc->pc_flags);
- rc = ptlrpcd_start(pt->pt_name, pc);
- }
- }
- if (rc != 0) {
- --ptlrpcd_users;
- ptlrpcd_fini();
- }
+ if (++ptlrpcd_users != 1)
+ GOTO(out, rc);
+
+ rc = ptlrpcd_start("ptlrpcd", &ptlrpcd_pc);
+ if (rc) {
+ --ptlrpcd_users;
+ GOTO(out, rc);
+ }
+
+ rc = ptlrpcd_start("ptlrpcd-recov", &ptlrpcd_recovery_pc);
+ if (rc) {
+ ptlrpcd_stop(&ptlrpcd_pc, 0);
+ --ptlrpcd_users;
+ GOTO(out, rc);
}
+out:
mutex_up(&ptlrpcd_sem);
RETURN(rc);
}
void ptlrpcd_decref(void)
{
mutex_down(&ptlrpcd_sem);
- if (--ptlrpcd_users == 0)
- ptlrpcd_fini();
+ if (--ptlrpcd_users == 0) {
+ ptlrpcd_stop(&ptlrpcd_pc, 0);
+ ptlrpcd_stop(&ptlrpcd_recovery_pc, 0);
+ }
mutex_up(&ptlrpcd_sem);
}