Unify the flow control interfaces for MDC RPC and FLD RPC.
We allow to adjust the maximum inflight RPCs count via /sys
interface.
Signed-off-by: Fan Yong <fan.yong@intel.com>
Intel-bug-id: https://jira.hpdd.intel.com/browse/LU-4687
Reviewed-on: http://review.whamcloud.com/9562
Reviewed-by: Niu Yawei <yawei.niu@intel.com>
Reviewed-by: Alex Zhuravlev <alexey.zhuravlev@intel.com>
Reviewed-by: Oleg Drokin <oleg.drokin@intel.com>
Signed-off-by: James Simmons <jsimmons@infradead.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
#include "../include/lustre_mdc.h"
#include "fld_internal.h"
-/* TODO: these 3 functions are copies of flow-control code from mdc_lib.c
- * It should be common thing. The same about mdc RPC lock
- */
-static int fld_req_avail(struct client_obd *cli, struct mdc_cache_waiter *mcw)
-{
- int rc;
-
- spin_lock(&cli->cl_loi_list_lock);
- rc = list_empty(&mcw->mcw_entry);
- spin_unlock(&cli->cl_loi_list_lock);
- return rc;
-};
-
-static void fld_enter_request(struct client_obd *cli)
-{
- struct mdc_cache_waiter mcw;
- struct l_wait_info lwi = { 0 };
-
- spin_lock(&cli->cl_loi_list_lock);
- if (cli->cl_r_in_flight >= cli->cl_max_rpcs_in_flight) {
- list_add_tail(&mcw.mcw_entry, &cli->cl_cache_waiters);
- init_waitqueue_head(&mcw.mcw_waitq);
- spin_unlock(&cli->cl_loi_list_lock);
- l_wait_event(mcw.mcw_waitq, fld_req_avail(cli, &mcw), &lwi);
- } else {
- cli->cl_r_in_flight++;
- spin_unlock(&cli->cl_loi_list_lock);
- }
-}
-
-static void fld_exit_request(struct client_obd *cli)
-{
- struct list_head *l, *tmp;
- struct mdc_cache_waiter *mcw;
-
- spin_lock(&cli->cl_loi_list_lock);
- cli->cl_r_in_flight--;
- list_for_each_safe(l, tmp, &cli->cl_cache_waiters) {
- if (cli->cl_r_in_flight >= cli->cl_max_rpcs_in_flight) {
- /* No free request slots anymore */
- break;
- }
-
- mcw = list_entry(l, struct mdc_cache_waiter, mcw_entry);
- list_del_init(&mcw->mcw_entry);
- cli->cl_r_in_flight++;
- wake_up(&mcw->mcw_waitq);
- }
- spin_unlock(&cli->cl_loi_list_lock);
-}
-
static int fld_rrb_hash(struct lu_client_fld *fld, u64 seq)
{
LASSERT(fld->lcf_count > 0);
req->rq_reply_portal = MDC_REPLY_PORTAL;
ptlrpc_at_set_req_timeout(req);
- fld_enter_request(&exp->exp_obd->u.cli);
+ obd_get_request_slot(&exp->exp_obd->u.cli);
rc = ptlrpc_queue_wait(req);
- fld_exit_request(&exp->exp_obd->u.cli);
+ obd_put_request_slot(&exp->exp_obd->u.cli);
if (rc)
goto out_req;
}
}
-struct mdc_cache_waiter {
- struct list_head mcw_entry;
- wait_queue_head_t mcw_waitq;
-};
-
/* mdc/mdc_locks.c */
int it_open_error(int phase, struct lookup_intent *it);
struct list_head ti_chain;
};
-#define OSC_MAX_RIF_DEFAULT 8
-#define OSC_MAX_RIF_MAX 256
-#define OSC_MAX_DIRTY_DEFAULT (OSC_MAX_RIF_DEFAULT * 4)
-#define OSC_MAX_DIRTY_MB_MAX 2048 /* arbitrary, but < MAX_LONG bytes */
-#define OSC_DEFAULT_RESENDS 10
+#define OBD_MAX_RIF_DEFAULT 8
+#define OBD_MAX_RIF_MAX 512
+#define OSC_MAX_RIF_MAX 256
+#define OSC_MAX_DIRTY_DEFAULT (OBD_MAX_RIF_DEFAULT * 4)
+#define OSC_MAX_DIRTY_MB_MAX 2048 /* arbitrary, but < MAX_LONG bytes */
+#define OSC_DEFAULT_RESENDS 10
/* possible values for fo_sync_lock_cancel */
enum {
NUM_SYNC_ON_CANCEL_STATES
};
-#define MDC_MAX_RIF_DEFAULT 8
-#define MDC_MAX_RIF_MAX 512
-
enum obd_cl_sem_lock_class {
OBD_CLI_SEM_NORMAL,
OBD_CLI_SEM_MGC,
void obd_zombie_impexp_stop(void);
void obd_zombie_barrier(void);
+int obd_get_request_slot(struct client_obd *cli);
+void obd_put_request_slot(struct client_obd *cli);
+__u32 obd_get_max_rpcs_in_flight(struct client_obd *cli);
+int obd_set_max_rpcs_in_flight(struct client_obd *cli, __u32 max);
+
struct llog_handle;
struct llog_rec_hdr;
typedef int (*llog_cb_t)(const struct lu_env *, struct llog_handle *,
cli->cl_chunkbits = PAGE_SHIFT;
if (!strcmp(name, LUSTRE_MDC_NAME)) {
- cli->cl_max_rpcs_in_flight = MDC_MAX_RIF_DEFAULT;
+ cli->cl_max_rpcs_in_flight = OBD_MAX_RIF_DEFAULT;
} else if (totalram_pages >> (20 - PAGE_SHIFT) <= 128 /* MB */) {
cli->cl_max_rpcs_in_flight = 2;
} else if (totalram_pages >> (20 - PAGE_SHIFT) <= 256 /* MB */) {
} else if (totalram_pages >> (20 - PAGE_SHIFT) <= 512 /* MB */) {
cli->cl_max_rpcs_in_flight = 4;
} else {
- cli->cl_max_rpcs_in_flight = OSC_MAX_RIF_DEFAULT;
+ cli->cl_max_rpcs_in_flight = OBD_MAX_RIF_DEFAULT;
}
rc = ldlm_get_ref();
if (rc) {
int len;
struct obd_device *dev = container_of(kobj, struct obd_device,
obd_kobj);
- struct client_obd *cli = &dev->u.cli;
+ __u32 max;
- spin_lock(&cli->cl_loi_list_lock);
- len = sprintf(buf, "%u\n", cli->cl_max_rpcs_in_flight);
- spin_unlock(&cli->cl_loi_list_lock);
+ max = obd_get_max_rpcs_in_flight(&dev->u.cli);
+ len = sprintf(buf, "%u\n", max);
return len;
}
{
struct obd_device *dev = container_of(kobj, struct obd_device,
obd_kobj);
- struct client_obd *cli = &dev->u.cli;
int rc;
unsigned long val;
if (rc)
return rc;
- if (val < 1 || val > MDC_MAX_RIF_MAX)
- return -ERANGE;
-
- spin_lock(&cli->cl_loi_list_lock);
- cli->cl_max_rpcs_in_flight = val;
- spin_unlock(&cli->cl_loi_list_lock);
+ rc = obd_set_max_rpcs_in_flight(&dev->u.cli, val);
+ if (rc)
+ count = rc;
return count;
}
void mdc_rename_pack(struct ptlrpc_request *req, struct md_op_data *op_data,
const char *old, int oldlen, const char *new, int newlen);
void mdc_close_pack(struct ptlrpc_request *req, struct md_op_data *op_data);
-int mdc_enter_request(struct client_obd *cli);
-void mdc_exit_request(struct client_obd *cli);
/* mdc/mdc_locks.c */
int mdc_set_lock_data(struct obd_export *exp,
mdc_ioepoch_pack(epoch, op_data);
mdc_hsm_release_pack(req, op_data);
}
-
-static int mdc_req_avail(struct client_obd *cli, struct mdc_cache_waiter *mcw)
-{
- int rc;
-
- spin_lock(&cli->cl_loi_list_lock);
- rc = list_empty(&mcw->mcw_entry);
- spin_unlock(&cli->cl_loi_list_lock);
- return rc;
-};
-
-/* We record requests in flight in cli->cl_r_in_flight here.
- * There is only one write rpc possible in mdc anyway. If this to change
- * in the future - the code may need to be revisited.
- */
-int mdc_enter_request(struct client_obd *cli)
-{
- int rc = 0;
- struct mdc_cache_waiter mcw;
- struct l_wait_info lwi = LWI_INTR(LWI_ON_SIGNAL_NOOP, NULL);
-
- spin_lock(&cli->cl_loi_list_lock);
- if (cli->cl_r_in_flight >= cli->cl_max_rpcs_in_flight) {
- list_add_tail(&mcw.mcw_entry, &cli->cl_cache_waiters);
- init_waitqueue_head(&mcw.mcw_waitq);
- spin_unlock(&cli->cl_loi_list_lock);
- rc = l_wait_event(mcw.mcw_waitq, mdc_req_avail(cli, &mcw),
- &lwi);
- if (rc) {
- spin_lock(&cli->cl_loi_list_lock);
- if (list_empty(&mcw.mcw_entry))
- cli->cl_r_in_flight--;
- list_del_init(&mcw.mcw_entry);
- spin_unlock(&cli->cl_loi_list_lock);
- }
- } else {
- cli->cl_r_in_flight++;
- spin_unlock(&cli->cl_loi_list_lock);
- }
- return rc;
-}
-
-void mdc_exit_request(struct client_obd *cli)
-{
- struct list_head *l, *tmp;
- struct mdc_cache_waiter *mcw;
-
- spin_lock(&cli->cl_loi_list_lock);
- cli->cl_r_in_flight--;
- list_for_each_safe(l, tmp, &cli->cl_cache_waiters) {
- if (cli->cl_r_in_flight >= cli->cl_max_rpcs_in_flight) {
- /* No free request slots anymore */
- break;
- }
-
- mcw = list_entry(l, struct mdc_cache_waiter, mcw_entry);
- list_del_init(&mcw->mcw_entry);
- cli->cl_r_in_flight++;
- wake_up(&mcw->mcw_waitq);
- }
- /* Empty waiting list? Decrease reqs in-flight number */
-
- spin_unlock(&cli->cl_loi_list_lock);
-}
*/
if (it) {
mdc_get_rpc_lock(obddev->u.cli.cl_rpc_lock, it);
- rc = mdc_enter_request(&obddev->u.cli);
+ rc = obd_get_request_slot(&obddev->u.cli);
if (rc != 0) {
mdc_put_rpc_lock(obddev->u.cli.cl_rpc_lock, it);
mdc_clear_replay_flag(req, 0);
return rc;
}
- mdc_exit_request(&obddev->u.cli);
+ obd_put_request_slot(&obddev->u.cli);
mdc_put_rpc_lock(obddev->u.cli.cl_rpc_lock, it);
if (rc < 0) {
obddev = class_exp2obd(exp);
- mdc_exit_request(&obddev->u.cli);
+ obd_put_request_slot(&obddev->u.cli);
if (OBD_FAIL_CHECK(OBD_FAIL_MDC_GETATTR_ENQUEUE))
rc = -ETIMEDOUT;
if (IS_ERR(req))
return PTR_ERR(req);
- rc = mdc_enter_request(&obddev->u.cli);
+ rc = obd_get_request_slot(&obddev->u.cli);
if (rc != 0) {
ptlrpc_req_finished(req);
return rc;
rc = ldlm_cli_enqueue(exp, &req, einfo, &res_id, &policy, &flags, NULL,
0, LVB_T_NONE, &minfo->mi_lockh, 1);
if (rc < 0) {
- mdc_exit_request(&obddev->u.cli);
+ obd_put_request_slot(&obddev->u.cli);
ptlrpc_req_finished(req);
return rc;
}
struct client_obd *cli = &req->rq_import->imp_obd->u.cli;
int rc;
- /* mdc_enter_request() ensures that this client has no more
+ /* obd_get_request_slot() ensures that this client has no more
* than cl_max_rpcs_in_flight RPCs simultaneously inf light
* against an MDT.
*/
- rc = mdc_enter_request(cli);
+ rc = obd_get_request_slot(cli);
if (rc != 0)
return rc;
rc = ptlrpc_queue_wait(req);
- mdc_exit_request(cli);
+ obd_put_request_slot(cli);
return rc;
}
obd_zombie_impexp_notify();
wait_for_completion(&obd_zombie_stop);
}
+
+struct obd_request_slot_waiter {
+ struct list_head orsw_entry;
+ wait_queue_head_t orsw_waitq;
+ bool orsw_signaled;
+};
+
+static bool obd_request_slot_avail(struct client_obd *cli,
+ struct obd_request_slot_waiter *orsw)
+{
+ bool avail;
+
+ spin_lock(&cli->cl_loi_list_lock);
+ avail = !!list_empty(&orsw->orsw_entry);
+ spin_unlock(&cli->cl_loi_list_lock);
+
+ return avail;
+};
+
+/*
+ * For network flow control, the RPC sponsor needs to acquire a credit
+ * before sending the RPC. The credits count for a connection is defined
+ * by the "cl_max_rpcs_in_flight". If all the credits are occpuied, then
+ * the subsequent RPC sponsors need to wait until others released their
+ * credits, or the administrator increased the "cl_max_rpcs_in_flight".
+ */
+int obd_get_request_slot(struct client_obd *cli)
+{
+ struct obd_request_slot_waiter orsw;
+ struct l_wait_info lwi;
+ int rc;
+
+ spin_lock(&cli->cl_loi_list_lock);
+ if (cli->cl_r_in_flight < cli->cl_max_rpcs_in_flight) {
+ cli->cl_r_in_flight++;
+ spin_unlock(&cli->cl_loi_list_lock);
+ return 0;
+ }
+
+ init_waitqueue_head(&orsw.orsw_waitq);
+ list_add_tail(&orsw.orsw_entry, &cli->cl_loi_read_list);
+ orsw.orsw_signaled = false;
+ spin_unlock(&cli->cl_loi_list_lock);
+
+ lwi = LWI_INTR(LWI_ON_SIGNAL_NOOP, NULL);
+ rc = l_wait_event(orsw.orsw_waitq,
+ obd_request_slot_avail(cli, &orsw) ||
+ orsw.orsw_signaled,
+ &lwi);
+
+ /*
+ * Here, we must take the lock to avoid the on-stack 'orsw' to be
+ * freed but other (such as obd_put_request_slot) is using it.
+ */
+ spin_lock(&cli->cl_loi_list_lock);
+ if (rc) {
+ if (!orsw.orsw_signaled) {
+ if (list_empty(&orsw.orsw_entry))
+ cli->cl_r_in_flight--;
+ else
+ list_del(&orsw.orsw_entry);
+ }
+ }
+
+ if (orsw.orsw_signaled) {
+ LASSERT(list_empty(&orsw.orsw_entry));
+
+ rc = -EINTR;
+ }
+ spin_unlock(&cli->cl_loi_list_lock);
+
+ return rc;
+}
+EXPORT_SYMBOL(obd_get_request_slot);
+
+void obd_put_request_slot(struct client_obd *cli)
+{
+ struct obd_request_slot_waiter *orsw;
+
+ spin_lock(&cli->cl_loi_list_lock);
+ cli->cl_r_in_flight--;
+
+ /* If there is free slot, wakeup the first waiter. */
+ if (!list_empty(&cli->cl_loi_read_list) &&
+ likely(cli->cl_r_in_flight < cli->cl_max_rpcs_in_flight)) {
+ orsw = list_entry(cli->cl_loi_read_list.next,
+ struct obd_request_slot_waiter, orsw_entry);
+ list_del_init(&orsw->orsw_entry);
+ cli->cl_r_in_flight++;
+ wake_up(&orsw->orsw_waitq);
+ }
+ spin_unlock(&cli->cl_loi_list_lock);
+}
+EXPORT_SYMBOL(obd_put_request_slot);
+
+__u32 obd_get_max_rpcs_in_flight(struct client_obd *cli)
+{
+ return cli->cl_max_rpcs_in_flight;
+}
+EXPORT_SYMBOL(obd_get_max_rpcs_in_flight);
+
+int obd_set_max_rpcs_in_flight(struct client_obd *cli, __u32 max)
+{
+ struct obd_request_slot_waiter *orsw;
+ __u32 old;
+ int diff;
+ int i;
+
+ if (max > OBD_MAX_RIF_MAX || max < 1)
+ return -ERANGE;
+
+ spin_lock(&cli->cl_loi_list_lock);
+ old = cli->cl_max_rpcs_in_flight;
+ cli->cl_max_rpcs_in_flight = max;
+ diff = max - old;
+
+ /* We increase the max_rpcs_in_flight, then wakeup some waiters. */
+ for (i = 0; i < diff; i++) {
+ if (list_empty(&cli->cl_loi_read_list))
+ break;
+
+ orsw = list_entry(cli->cl_loi_read_list.next,
+ struct obd_request_slot_waiter, orsw_entry);
+ list_del_init(&orsw->orsw_entry);
+ cli->cl_r_in_flight++;
+ wake_up(&orsw->orsw_waitq);
+ }
+ spin_unlock(&cli->cl_loi_list_lock);
+
+ return 0;
+}
+EXPORT_SYMBOL(obd_set_max_rpcs_in_flight);
projects / linux.git / commitdiff