* Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
* Use is subject to license terms.
*
- * Copyright (c) 2011, 2012, Whamcloud, Inc.
+ * Copyright (c) 2011, 2013, Intel Corporation.
*/
/*
* This file is part of Lustre, http://www.lustre.org/
#include <lustre_log.h>
#include <lustre_debug.h>
#include <lustre_param.h>
+#include <lustre_fid.h>
#include "osc_internal.h"
#include "osc_cl_internal.h"
/* Pack OSC object metadata for disk storage (LE byte order). */
static int osc_packmd(struct obd_export *exp, struct lov_mds_md **lmmp,
- struct lov_stripe_md *lsm)
+ struct lov_stripe_md *lsm)
{
- int lmm_size;
- ENTRY;
+ int lmm_size;
+ ENTRY;
- lmm_size = sizeof(**lmmp);
- if (!lmmp)
- RETURN(lmm_size);
+ lmm_size = sizeof(**lmmp);
+ if (lmmp == NULL)
+ RETURN(lmm_size);
- if (*lmmp && !lsm) {
- OBD_FREE(*lmmp, lmm_size);
- *lmmp = NULL;
- RETURN(0);
- }
+ if (*lmmp != NULL && lsm == NULL) {
+ OBD_FREE(*lmmp, lmm_size);
+ *lmmp = NULL;
+ RETURN(0);
+ } else if (unlikely(lsm != NULL && ostid_id(&lsm->lsm_oi) == 0)) {
+ RETURN(-EBADF);
+ }
- if (!*lmmp) {
- OBD_ALLOC(*lmmp, lmm_size);
- if (!*lmmp)
- RETURN(-ENOMEM);
- }
+ if (*lmmp == NULL) {
+ OBD_ALLOC(*lmmp, lmm_size);
+ if (*lmmp == NULL)
+ RETURN(-ENOMEM);
+ }
- if (lsm) {
- LASSERT(lsm->lsm_object_id);
- LASSERT_SEQ_IS_MDT(lsm->lsm_object_seq);
- (*lmmp)->lmm_object_id = cpu_to_le64(lsm->lsm_object_id);
- (*lmmp)->lmm_object_seq = cpu_to_le64(lsm->lsm_object_seq);
- }
+ if (lsm)
+ ostid_cpu_to_le(&lsm->lsm_oi, &(*lmmp)->lmm_oi);
- RETURN(lmm_size);
+ RETURN(lmm_size);
}
/* Unpack OSC object metadata from disk storage (LE byte order). */
static int osc_unpackmd(struct obd_export *exp, struct lov_stripe_md **lsmp,
- struct lov_mds_md *lmm, int lmm_bytes)
+ struct lov_mds_md *lmm, int lmm_bytes)
{
- int lsm_size;
- struct obd_import *imp = class_exp2cliimp(exp);
- ENTRY;
+ int lsm_size;
+ struct obd_import *imp = class_exp2cliimp(exp);
+ ENTRY;
- if (lmm != NULL) {
- if (lmm_bytes < sizeof (*lmm)) {
- CERROR("lov_mds_md too small: %d, need %d\n",
- lmm_bytes, (int)sizeof(*lmm));
- RETURN(-EINVAL);
- }
- /* XXX LOV_MAGIC etc check? */
+ if (lmm != NULL) {
+ if (lmm_bytes < sizeof(*lmm)) {
+ CERROR("%s: lov_mds_md too small: %d, need %d\n",
+ exp->exp_obd->obd_name, lmm_bytes,
+ (int)sizeof(*lmm));
+ RETURN(-EINVAL);
+ }
+ /* XXX LOV_MAGIC etc check? */
- if (lmm->lmm_object_id == 0) {
- CERROR("lov_mds_md: zero lmm_object_id\n");
- RETURN(-EINVAL);
- }
- }
+ if (unlikely(ostid_id(&lmm->lmm_oi) == 0)) {
+ CERROR("%s: zero lmm_object_id: rc = %d\n",
+ exp->exp_obd->obd_name, -EINVAL);
+ RETURN(-EINVAL);
+ }
+ }
- lsm_size = lov_stripe_md_size(1);
- if (lsmp == NULL)
- RETURN(lsm_size);
+ lsm_size = lov_stripe_md_size(1);
+ if (lsmp == NULL)
+ RETURN(lsm_size);
- if (*lsmp != NULL && lmm == NULL) {
- OBD_FREE((*lsmp)->lsm_oinfo[0], sizeof(struct lov_oinfo));
- OBD_FREE(*lsmp, lsm_size);
- *lsmp = NULL;
- RETURN(0);
- }
+ if (*lsmp != NULL && lmm == NULL) {
+ OBD_FREE((*lsmp)->lsm_oinfo[0], sizeof(struct lov_oinfo));
+ OBD_FREE(*lsmp, lsm_size);
+ *lsmp = NULL;
+ RETURN(0);
+ }
- if (*lsmp == NULL) {
- OBD_ALLOC(*lsmp, lsm_size);
- if (*lsmp == NULL)
- RETURN(-ENOMEM);
- OBD_ALLOC((*lsmp)->lsm_oinfo[0], sizeof(struct lov_oinfo));
- if ((*lsmp)->lsm_oinfo[0] == NULL) {
- OBD_FREE(*lsmp, lsm_size);
- RETURN(-ENOMEM);
- }
- loi_init((*lsmp)->lsm_oinfo[0]);
- }
+ if (*lsmp == NULL) {
+ OBD_ALLOC(*lsmp, lsm_size);
+ if (unlikely(*lsmp == NULL))
+ RETURN(-ENOMEM);
+ OBD_ALLOC((*lsmp)->lsm_oinfo[0], sizeof(struct lov_oinfo));
+ if (unlikely((*lsmp)->lsm_oinfo[0] == NULL)) {
+ OBD_FREE(*lsmp, lsm_size);
+ RETURN(-ENOMEM);
+ }
+ loi_init((*lsmp)->lsm_oinfo[0]);
+ } else if (unlikely(ostid_id(&(*lsmp)->lsm_oi) == 0)) {
+ RETURN(-EBADF);
+ }
- if (lmm != NULL) {
- /* XXX zero *lsmp? */
- (*lsmp)->lsm_object_id = le64_to_cpu (lmm->lmm_object_id);
- (*lsmp)->lsm_object_seq = le64_to_cpu (lmm->lmm_object_seq);
- LASSERT((*lsmp)->lsm_object_id);
- LASSERT_SEQ_IS_MDT((*lsmp)->lsm_object_seq);
- }
+ if (lmm != NULL)
+ /* XXX zero *lsmp? */
+ ostid_le_to_cpu(&lmm->lmm_oi, &(*lsmp)->lsm_oi);
- if (imp != NULL &&
- (imp->imp_connect_data.ocd_connect_flags & OBD_CONNECT_MAXBYTES))
- (*lsmp)->lsm_maxbytes = imp->imp_connect_data.ocd_maxbytes;
- else
- (*lsmp)->lsm_maxbytes = LUSTRE_STRIPE_MAXBYTES;
+ if (imp != NULL &&
+ (imp->imp_connect_data.ocd_connect_flags & OBD_CONNECT_MAXBYTES))
+ (*lsmp)->lsm_maxbytes = imp->imp_connect_data.ocd_maxbytes;
+ else
+ (*lsmp)->lsm_maxbytes = LUSTRE_STRIPE_MAXBYTES;
- RETURN(lsm_size);
+ RETURN(lsm_size);
}
static inline void osc_pack_capa(struct ptlrpc_request *req,
static inline void osc_pack_req_body(struct ptlrpc_request *req,
struct obd_info *oinfo)
{
- struct ost_body *body;
+ struct ost_body *body;
- body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
- LASSERT(body);
+ body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
+ LASSERT(body);
- lustre_set_wire_obdo(&body->oa, oinfo->oi_oa);
- osc_pack_capa(req, body, oinfo->oi_capa);
+ lustre_set_wire_obdo(&req->rq_import->imp_connect_data, &body->oa,
+ oinfo->oi_oa);
+ osc_pack_capa(req, body, oinfo->oi_capa);
}
static inline void osc_set_capa_size(struct ptlrpc_request *req,
body = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
if (body) {
- CDEBUG(D_INODE, "mode: %o\n", body->oa.o_mode);
- lustre_get_wire_obdo(aa->aa_oi->oi_oa, &body->oa);
+ CDEBUG(D_INODE, "mode: %o\n", body->oa.o_mode);
+ lustre_get_wire_obdo(&req->rq_import->imp_connect_data,
+ aa->aa_oi->oi_oa, &body->oa);
- /* This should really be sent by the OST */
- aa->aa_oi->oi_oa->o_blksize = PTLRPC_MAX_BRW_SIZE;
- aa->aa_oi->oi_oa->o_valid |= OBD_MD_FLBLKSZ;
+ /* This should really be sent by the OST */
+ aa->aa_oi->oi_oa->o_blksize = DT_MAX_BRW_SIZE;
+ aa->aa_oi->oi_oa->o_valid |= OBD_MD_FLBLKSZ;
} else {
CDEBUG(D_INFO, "can't unpack ost_body\n");
rc = -EPROTO;
if (body == NULL)
GOTO(out, rc = -EPROTO);
- CDEBUG(D_INODE, "mode: %o\n", body->oa.o_mode);
- lustre_get_wire_obdo(oinfo->oi_oa, &body->oa);
+ CDEBUG(D_INODE, "mode: %o\n", body->oa.o_mode);
+ lustre_get_wire_obdo(&req->rq_import->imp_connect_data, oinfo->oi_oa,
+ &body->oa);
- /* This should really be sent by the OST */
- oinfo->oi_oa->o_blksize = PTLRPC_MAX_BRW_SIZE;
- oinfo->oi_oa->o_valid |= OBD_MD_FLBLKSZ;
+ oinfo->oi_oa->o_blksize = cli_brw_size(exp->exp_obd);
+ oinfo->oi_oa->o_valid |= OBD_MD_FLBLKSZ;
EXIT;
out:
if (body == NULL)
GOTO(out, rc = -EPROTO);
- lustre_get_wire_obdo(oinfo->oi_oa, &body->oa);
+ lustre_get_wire_obdo(&req->rq_import->imp_connect_data, oinfo->oi_oa,
+ &body->oa);
EXIT;
out:
if (body == NULL)
GOTO(out, rc = -EPROTO);
- lustre_get_wire_obdo(sa->sa_oa, &body->oa);
+ lustre_get_wire_obdo(&req->rq_import->imp_connect_data, sa->sa_oa,
+ &body->oa);
out:
rc = sa->sa_upcall(sa->sa_cookie, rc);
RETURN(rc);
body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
LASSERT(body);
- lustre_set_wire_obdo(&body->oa, oa);
+
+ lustre_set_wire_obdo(&req->rq_import->imp_connect_data, &body->oa, oa);
ptlrpc_request_set_replen(req);
if (body == NULL)
GOTO(out_req, rc = -EPROTO);
- lustre_get_wire_obdo(oa, &body->oa);
+ CDEBUG(D_INFO, "oa flags %x\n", oa->o_flags);
+ lustre_get_wire_obdo(&req->rq_import->imp_connect_data, oa, &body->oa);
- /* This should really be sent by the OST */
- oa->o_blksize = PTLRPC_MAX_BRW_SIZE;
- oa->o_valid |= OBD_MD_FLBLKSZ;
+ oa->o_blksize = cli_brw_size(exp->exp_obd);
+ oa->o_valid |= OBD_MD_FLBLKSZ;
- /* XXX LOV STACKING: the lsm that is passed to us from LOV does not
- * have valid lsm_oinfo data structs, so don't go touching that.
- * This needs to be fixed in a big way.
- */
- lsm->lsm_object_id = oa->o_id;
- lsm->lsm_object_seq = oa->o_seq;
- *ea = lsm;
+ /* XXX LOV STACKING: the lsm that is passed to us from LOV does not
+ * have valid lsm_oinfo data structs, so don't go touching that.
+ * This needs to be fixed in a big way.
+ */
+ lsm->lsm_oi = oa->o_oi;
+ *ea = lsm;
if (oti != NULL) {
oti->oti_transno = lustre_msg_get_transno(req->rq_repmsg);
req->rq_request_portal = OST_IO_PORTAL; /* bug 7198 */
ptlrpc_at_set_req_timeout(req);
- body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
- LASSERT(body);
- lustre_set_wire_obdo(&body->oa, oinfo->oi_oa);
- osc_pack_capa(req, body, oinfo->oi_capa);
+ body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
+ LASSERT(body);
+ lustre_set_wire_obdo(&req->rq_import->imp_connect_data, &body->oa,
+ oinfo->oi_oa);
+ osc_pack_capa(req, body, oinfo->oi_capa);
ptlrpc_request_set_replen(req);
RETURN(rc);
}
- /* overload the size and blocks fields in the oa with start/end */
- body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
- LASSERT(body);
- lustre_set_wire_obdo(&body->oa, oinfo->oi_oa);
- osc_pack_capa(req, body, oinfo->oi_capa);
+ /* overload the size and blocks fields in the oa with start/end */
+ body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
+ LASSERT(body);
+ lustre_set_wire_obdo(&req->rq_import->imp_connect_data, &body->oa,
+ oinfo->oi_oa);
+ osc_pack_capa(req, body, oinfo->oi_capa);
ptlrpc_request_set_replen(req);
req->rq_interpret_reply = osc_sync_interpret;
* @objid. Found locks are added into @cancel list. Returns the amount of
* locks added to @cancels list. */
static int osc_resource_get_unused(struct obd_export *exp, struct obdo *oa,
- cfs_list_t *cancels,
- ldlm_mode_t mode, int lock_flags)
+ cfs_list_t *cancels,
+ ldlm_mode_t mode, __u64 lock_flags)
{
struct ldlm_namespace *ns = exp->exp_obd->obd_namespace;
struct ldlm_res_id res_id;
if (exp_connect_cancelset(exp) && !ns_connect_cancelset(ns))
RETURN(0);
- osc_build_res_name(oa->o_id, oa->o_seq, &res_id);
- res = ldlm_resource_get(ns, NULL, &res_id, 0, 0);
- if (res == NULL)
- RETURN(0);
+ ostid_build_res_name(&oa->o_oi, &res_id);
+ res = ldlm_resource_get(ns, NULL, &res_id, 0, 0);
+ if (res == NULL)
+ RETURN(0);
LDLM_RESOURCE_ADDREF(res);
count = ldlm_cancel_resource_local(res, cancels, NULL, mode,
}
static int osc_destroy_interpret(const struct lu_env *env,
- struct ptlrpc_request *req, void *data,
- int rc)
+ struct ptlrpc_request *req, void *data,
+ int rc)
{
- struct client_obd *cli = &req->rq_import->imp_obd->u.cli;
+ struct client_obd *cli = &req->rq_import->imp_obd->u.cli;
- cfs_atomic_dec(&cli->cl_destroy_in_flight);
- cfs_waitq_signal(&cli->cl_destroy_waitq);
- return 0;
+ cfs_atomic_dec(&cli->cl_destroy_in_flight);
+ wake_up(&cli->cl_destroy_waitq);
+ return 0;
}
static int osc_can_send_destroy(struct client_obd *cli)
{
- if (cfs_atomic_inc_return(&cli->cl_destroy_in_flight) <=
- cli->cl_max_rpcs_in_flight) {
- /* The destroy request can be sent */
- return 1;
- }
- if (cfs_atomic_dec_return(&cli->cl_destroy_in_flight) <
- cli->cl_max_rpcs_in_flight) {
- /*
- * The counter has been modified between the two atomic
- * operations.
- */
- cfs_waitq_signal(&cli->cl_destroy_waitq);
- }
- return 0;
+ if (cfs_atomic_inc_return(&cli->cl_destroy_in_flight) <=
+ cli->cl_max_rpcs_in_flight) {
+ /* The destroy request can be sent */
+ return 1;
+ }
+ if (cfs_atomic_dec_return(&cli->cl_destroy_in_flight) <
+ cli->cl_max_rpcs_in_flight) {
+ /*
+ * The counter has been modified between the two atomic
+ * operations.
+ */
+ wake_up(&cli->cl_destroy_waitq);
+ }
+ return 0;
}
int osc_create(const struct lu_env *env, struct obd_export *exp,
RETURN(osc_real_create(exp, oa, ea, oti));
}
- if (!fid_seq_is_mdt(oa->o_seq))
+ if (!fid_seq_is_mdt(ostid_seq(&oa->o_oi)))
RETURN(osc_real_create(exp, oa, ea, oti));
/* we should not get here anymore */
req->rq_request_portal = OST_IO_PORTAL; /* bug 7198 */
ptlrpc_at_set_req_timeout(req);
- if (oti != NULL && oa->o_valid & OBD_MD_FLCOOKIE)
- oa->o_lcookie = *oti->oti_logcookies;
- body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
- LASSERT(body);
- lustre_set_wire_obdo(&body->oa, oa);
+ if (oti != NULL && oa->o_valid & OBD_MD_FLCOOKIE)
+ oa->o_lcookie = *oti->oti_logcookies;
+ body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
+ LASSERT(body);
+ lustre_set_wire_obdo(&req->rq_import->imp_connect_data, &body->oa, oa);
osc_pack_capa(req, body, (struct obd_capa *)capa);
ptlrpc_request_set_replen(req);
oa->o_valid |= bits;
client_obd_list_lock(&cli->cl_loi_list_lock);
oa->o_dirty = cli->cl_dirty;
- if (cli->cl_dirty - cli->cl_dirty_transit > cli->cl_dirty_max) {
- CERROR("dirty %lu - %lu > dirty_max %lu\n",
- cli->cl_dirty, cli->cl_dirty_transit, cli->cl_dirty_max);
- oa->o_undirty = 0;
- } else if (cfs_atomic_read(&obd_dirty_pages) -
- cfs_atomic_read(&obd_dirty_transit_pages) >
- obd_max_dirty_pages + 1){
- /* The cfs_atomic_read() allowing the cfs_atomic_inc() are
- * not covered by a lock thus they may safely race and trip
- * this CERROR() unless we add in a small fudge factor (+1). */
- CERROR("dirty %d - %d > system dirty_max %d\n",
- cfs_atomic_read(&obd_dirty_pages),
- cfs_atomic_read(&obd_dirty_transit_pages),
- obd_max_dirty_pages);
- oa->o_undirty = 0;
- } else if (cli->cl_dirty_max - cli->cl_dirty > 0x7fffffff) {
- CERROR("dirty %lu - dirty_max %lu too big???\n",
- cli->cl_dirty, cli->cl_dirty_max);
- oa->o_undirty = 0;
- } else {
- long max_in_flight = (cli->cl_max_pages_per_rpc << CFS_PAGE_SHIFT)*
- (cli->cl_max_rpcs_in_flight + 1);
+ if (unlikely(cli->cl_dirty - cli->cl_dirty_transit >
+ cli->cl_dirty_max)) {
+ CERROR("dirty %lu - %lu > dirty_max %lu\n",
+ cli->cl_dirty, cli->cl_dirty_transit, cli->cl_dirty_max);
+ oa->o_undirty = 0;
+ } else if (unlikely(cfs_atomic_read(&obd_unstable_pages) +
+ cfs_atomic_read(&obd_dirty_pages) -
+ cfs_atomic_read(&obd_dirty_transit_pages) >
+ (long)(obd_max_dirty_pages + 1))) {
+ /* The cfs_atomic_read() allowing the cfs_atomic_inc() are
+ * not covered by a lock thus they may safely race and trip
+ * this CERROR() unless we add in a small fudge factor (+1). */
+ CERROR("%s: dirty %d + %d - %d > system dirty_max %d\n",
+ cli->cl_import->imp_obd->obd_name,
+ cfs_atomic_read(&obd_unstable_pages),
+ cfs_atomic_read(&obd_dirty_pages),
+ cfs_atomic_read(&obd_dirty_transit_pages),
+ obd_max_dirty_pages);
+ oa->o_undirty = 0;
+ } else if (unlikely(cli->cl_dirty_max - cli->cl_dirty > 0x7fffffff)) {
+ CERROR("dirty %lu - dirty_max %lu too big???\n",
+ cli->cl_dirty, cli->cl_dirty_max);
+ oa->o_undirty = 0;
+ } else {
+ long max_in_flight = (cli->cl_max_pages_per_rpc <<
+ PAGE_CACHE_SHIFT) *
+ (cli->cl_max_rpcs_in_flight + 1);
oa->o_undirty = max(cli->cl_dirty_max, max_in_flight);
}
oa->o_grant = cli->cl_avail_grant + cli->cl_reserved_grant;
* needed, and avoids shrinking the grant piecemeal. */
static int osc_shrink_grant(struct client_obd *cli)
{
- long target = (cli->cl_max_rpcs_in_flight + 1) *
- cli->cl_max_pages_per_rpc;
+ __u64 target_bytes = (cli->cl_max_rpcs_in_flight + 1) *
+ (cli->cl_max_pages_per_rpc << PAGE_CACHE_SHIFT);
- client_obd_list_lock(&cli->cl_loi_list_lock);
- if (cli->cl_avail_grant <= target)
- target = cli->cl_max_pages_per_rpc;
- client_obd_list_unlock(&cli->cl_loi_list_lock);
+ client_obd_list_lock(&cli->cl_loi_list_lock);
+ if (cli->cl_avail_grant <= target_bytes)
+ target_bytes = cli->cl_max_pages_per_rpc << PAGE_CACHE_SHIFT;
+ client_obd_list_unlock(&cli->cl_loi_list_lock);
- return osc_shrink_grant_to_target(cli, target);
+ return osc_shrink_grant_to_target(cli, target_bytes);
}
-int osc_shrink_grant_to_target(struct client_obd *cli, long target)
+int osc_shrink_grant_to_target(struct client_obd *cli, __u64 target_bytes)
{
- int rc = 0;
- struct ost_body *body;
- ENTRY;
-
- client_obd_list_lock(&cli->cl_loi_list_lock);
- /* Don't shrink if we are already above or below the desired limit
- * We don't want to shrink below a single RPC, as that will negatively
- * impact block allocation and long-term performance. */
- if (target < cli->cl_max_pages_per_rpc)
- target = cli->cl_max_pages_per_rpc;
+ int rc = 0;
+ struct ost_body *body;
+ ENTRY;
- if (target >= cli->cl_avail_grant) {
- client_obd_list_unlock(&cli->cl_loi_list_lock);
- RETURN(0);
- }
- client_obd_list_unlock(&cli->cl_loi_list_lock);
+ client_obd_list_lock(&cli->cl_loi_list_lock);
+ /* Don't shrink if we are already above or below the desired limit
+ * We don't want to shrink below a single RPC, as that will negatively
+ * impact block allocation and long-term performance. */
+ if (target_bytes < cli->cl_max_pages_per_rpc << PAGE_CACHE_SHIFT)
+ target_bytes = cli->cl_max_pages_per_rpc << PAGE_CACHE_SHIFT;
+
+ if (target_bytes >= cli->cl_avail_grant) {
+ client_obd_list_unlock(&cli->cl_loi_list_lock);
+ RETURN(0);
+ }
+ client_obd_list_unlock(&cli->cl_loi_list_lock);
- OBD_ALLOC_PTR(body);
- if (!body)
- RETURN(-ENOMEM);
+ OBD_ALLOC_PTR(body);
+ if (!body)
+ RETURN(-ENOMEM);
- osc_announce_cached(cli, &body->oa, 0);
+ osc_announce_cached(cli, &body->oa, 0);
- client_obd_list_lock(&cli->cl_loi_list_lock);
- body->oa.o_grant = cli->cl_avail_grant - target;
- cli->cl_avail_grant = target;
+ client_obd_list_lock(&cli->cl_loi_list_lock);
+ body->oa.o_grant = cli->cl_avail_grant - target_bytes;
+ cli->cl_avail_grant = target_bytes;
client_obd_list_unlock(&cli->cl_loi_list_lock);
if (!(body->oa.o_valid & OBD_MD_FLFLAGS)) {
body->oa.o_valid |= OBD_MD_FLFLAGS;
RETURN(rc);
}
-#define GRANT_SHRINK_LIMIT PTLRPC_MAX_BRW_SIZE
static int osc_should_shrink_grant(struct client_obd *client)
{
cfs_time_t time = cfs_time_current();
OBD_CONNECT_GRANT_SHRINK) == 0)
return 0;
- if (cfs_time_aftereq(time, next_shrink - 5 * CFS_TICK)) {
- if (client->cl_import->imp_state == LUSTRE_IMP_FULL &&
- client->cl_avail_grant > GRANT_SHRINK_LIMIT)
- return 1;
- else
- osc_update_next_shrink(client);
- }
+ if (cfs_time_aftereq(time, next_shrink - 5 * CFS_TICK)) {
+ /* Get the current RPC size directly, instead of going via:
+ * cli_brw_size(obd->u.cli.cl_import->imp_obd->obd_self_export)
+ * Keep comment here so that it can be found by searching. */
+ int brw_size = client->cl_max_pages_per_rpc << PAGE_CACHE_SHIFT;
+
+ if (client->cl_import->imp_state == LUSTRE_IMP_FULL &&
+ client->cl_avail_grant > brw_size)
+ return 1;
+ else
+ osc_update_next_shrink(client);
+ }
return 0;
}
cli->cl_avail_grant = ocd->ocd_grant - cli->cl_dirty;
if (cli->cl_avail_grant < 0) {
- CWARN("%s: available grant < 0, the OSS is probably not running"
- " with patch from bug20278 (%ld) \n",
- cli->cl_import->imp_obd->obd_name, cli->cl_avail_grant);
- /* workaround for 1.6 servers which do not have
- * the patch from bug20278 */
- cli->cl_avail_grant = ocd->ocd_grant;
+ CWARN("%s: available grant < 0: avail/ocd/dirty %ld/%u/%ld\n",
+ cli->cl_import->imp_obd->obd_name, cli->cl_avail_grant,
+ ocd->ocd_grant, cli->cl_dirty);
+ /* workaround for servers which do not have the patch from
+ * LU-2679 */
+ cli->cl_avail_grant = ocd->ocd_grant;
}
/* determine the appropriate chunk size used by osc_extent. */
- cli->cl_chunkbits = max_t(int, CFS_PAGE_SHIFT, ocd->ocd_blocksize);
+ cli->cl_chunkbits = max_t(int, PAGE_CACHE_SHIFT, ocd->ocd_blocksize);
client_obd_list_unlock(&cli->cl_loi_list_lock);
CDEBUG(D_CACHE, "%s, setting cl_avail_grant: %ld cl_lost_grant: %ld."
while (nob_read > 0) {
LASSERT (page_count > 0);
- if (pga[i]->count > nob_read) {
- /* EOF inside this page */
- ptr = cfs_kmap(pga[i]->pg) +
- (pga[i]->off & ~CFS_PAGE_MASK);
- memset(ptr + nob_read, 0, pga[i]->count - nob_read);
- cfs_kunmap(pga[i]->pg);
- page_count--;
- i++;
- break;
- }
+ if (pga[i]->count > nob_read) {
+ /* EOF inside this page */
+ ptr = kmap(pga[i]->pg) +
+ (pga[i]->off & ~CFS_PAGE_MASK);
+ memset(ptr + nob_read, 0, pga[i]->count - nob_read);
+ kunmap(pga[i]->pg);
+ page_count--;
+ i++;
+ break;
+ }
nob_read -= pga[i]->count;
page_count--;
i++;
}
- /* zero remaining pages */
- while (page_count-- > 0) {
- ptr = cfs_kmap(pga[i]->pg) + (pga[i]->off & ~CFS_PAGE_MASK);
- memset(ptr, 0, pga[i]->count);
- cfs_kunmap(pga[i]->pg);
- i++;
- }
+ /* zero remaining pages */
+ while (page_count-- > 0) {
+ ptr = kmap(pga[i]->pg) + (pga[i]->off & ~CFS_PAGE_MASK);
+ memset(ptr, 0, pga[i]->count);
+ kunmap(pga[i]->pg);
+ i++;
+ }
}
static int check_write_rcs(struct ptlrpc_request *req,
* simulate an OST->client data error */
if (i == 0 && opc == OST_READ &&
OBD_FAIL_CHECK(OBD_FAIL_OSC_CHECKSUM_RECEIVE)) {
- unsigned char *ptr = cfs_kmap(pga[i]->pg);
+ unsigned char *ptr = kmap(pga[i]->pg);
int off = pga[i]->off & ~CFS_PAGE_MASK;
memcpy(ptr + off, "bad1", min(4, nob));
- cfs_kunmap(pga[i]->pg);
+ kunmap(pga[i]->pg);
}
cfs_crypto_hash_update_page(hdesc, pga[i]->pg,
pga[i]->off & ~CFS_PAGE_MASK,
count);
- LL_CDEBUG_PAGE(D_PAGE, pga[i]->pg, "off %d checksum %x\n",
- (int)(pga[i]->off & ~CFS_PAGE_MASK), cksum);
+ LL_CDEBUG_PAGE(D_PAGE, pga[i]->pg, "off %d\n",
+ (int)(pga[i]->off & ~CFS_PAGE_MASK));
nob -= pga[i]->count;
pg_count--;
* retry logic */
req->rq_no_retry_einprogress = 1;
- if (opc == OST_WRITE)
- desc = ptlrpc_prep_bulk_imp(req, page_count,
- BULK_GET_SOURCE, OST_BULK_PORTAL);
- else
- desc = ptlrpc_prep_bulk_imp(req, page_count,
- BULK_PUT_SINK, OST_BULK_PORTAL);
+ desc = ptlrpc_prep_bulk_imp(req, page_count,
+ cli->cl_import->imp_connect_data.ocd_brw_size >> LNET_MTU_BITS,
+ opc == OST_WRITE ? BULK_GET_SOURCE : BULK_PUT_SINK,
+ OST_BULK_PORTAL);
if (desc == NULL)
GOTO(out, rc = -ENOMEM);
niobuf = req_capsule_client_get(pill, &RMF_NIOBUF_REMOTE);
LASSERT(body != NULL && ioobj != NULL && niobuf != NULL);
- lustre_set_wire_obdo(&body->oa, oa);
-
- obdo_to_ioobj(oa, ioobj);
- ioobj->ioo_bufcnt = niocount;
- osc_pack_capa(req, body, ocapa);
- LASSERT (page_count > 0);
- pg_prev = pga[0];
+ lustre_set_wire_obdo(&req->rq_import->imp_connect_data, &body->oa, oa);
+
+ obdo_to_ioobj(oa, ioobj);
+ ioobj->ioo_bufcnt = niocount;
+ /* The high bits of ioo_max_brw tells server _maximum_ number of bulks
+ * that might be send for this request. The actual number is decided
+ * when the RPC is finally sent in ptlrpc_register_bulk(). It sends
+ * "max - 1" for old client compatibility sending "0", and also so the
+ * the actual maximum is a power-of-two number, not one less. LU-1431 */
+ ioobj_max_brw_set(ioobj, desc->bd_md_max_brw);
+ osc_pack_capa(req, body, ocapa);
+ LASSERT(page_count > 0);
+ pg_prev = pga[0];
for (requested_nob = i = 0; i < page_count; i++, niobuf++) {
struct brw_page *pg = pga[i];
int poff = pg->off & ~CFS_PAGE_MASK;
LASSERT(pg->count > 0);
/* make sure there is no gap in the middle of page array */
- LASSERTF(page_count == 1 ||
- (ergo(i == 0, poff + pg->count == CFS_PAGE_SIZE) &&
- ergo(i > 0 && i < page_count - 1,
- poff == 0 && pg->count == CFS_PAGE_SIZE) &&
- ergo(i == page_count - 1, poff == 0)),
- "i: %d/%d pg: %p off: "LPU64", count: %u\n",
- i, page_count, pg, pg->off, pg->count);
+ LASSERTF(page_count == 1 ||
+ (ergo(i == 0, poff + pg->count == PAGE_CACHE_SIZE) &&
+ ergo(i > 0 && i < page_count - 1,
+ poff == 0 && pg->count == PAGE_CACHE_SIZE) &&
+ ergo(i == page_count - 1, poff == 0)),
+ "i: %d/%d pg: %p off: "LPU64", count: %u\n",
+ i, page_count, pg, pg->off, pg->count);
#ifdef __linux__
LASSERTF(i == 0 || pg->off > pg_prev->off,
"i %d p_c %u pg %p [pri %lu ind %lu] off "LPU64
msg = "changed in transit AND doesn't match the original - "
"likely false positive due to mmap IO (bug 11742)";
- LCONSOLE_ERROR_MSG(0x132, "BAD WRITE CHECKSUM: %s: from %s inode "DFID
- " object "LPU64"/"LPU64" extent ["LPU64"-"LPU64"]\n",
- msg, libcfs_nid2str(peer->nid),
- oa->o_valid & OBD_MD_FLFID ? oa->o_parent_seq : (__u64)0,
- oa->o_valid & OBD_MD_FLFID ? oa->o_parent_oid : 0,
- oa->o_valid & OBD_MD_FLFID ? oa->o_parent_ver : 0,
- oa->o_id,
- oa->o_valid & OBD_MD_FLGROUP ? oa->o_seq : (__u64)0,
- pga[0]->off,
- pga[page_count-1]->off + pga[page_count-1]->count - 1);
- CERROR("original client csum %x (type %x), server csum %x (type %x), "
- "client csum now %x\n", client_cksum, client_cksum_type,
- server_cksum, cksum_type, new_cksum);
- return 1;
+ LCONSOLE_ERROR_MSG(0x132, "BAD WRITE CHECKSUM: %s: from %s inode "DFID
+ " object "DOSTID" extent ["LPU64"-"LPU64"]\n",
+ msg, libcfs_nid2str(peer->nid),
+ oa->o_valid & OBD_MD_FLFID ? oa->o_parent_seq : (__u64)0,
+ oa->o_valid & OBD_MD_FLFID ? oa->o_parent_oid : 0,
+ oa->o_valid & OBD_MD_FLFID ? oa->o_parent_ver : 0,
+ POSTID(&oa->o_oi), pga[0]->off,
+ pga[page_count-1]->off + pga[page_count-1]->count - 1);
+ CERROR("original client csum %x (type %x), server csum %x (type %x), "
+ "client csum now %x\n", client_cksum, client_cksum_type,
+ server_cksum, cksum_type, new_cksum);
+ return 1;
}
/* Note rc enters this function as number of bytes transferred */
router = libcfs_nid2str(req->rq_bulk->bd_sender);
}
- if (server_cksum == ~0 && rc > 0) {
- CERROR("Protocol error: server %s set the 'checksum' "
- "bit, but didn't send a checksum. Not fatal, "
- "but please notify on http://bugs.whamcloud.com/\n",
- libcfs_nid2str(peer->nid));
- } else if (server_cksum != client_cksum) {
- LCONSOLE_ERROR_MSG(0x133, "%s: BAD READ CHECKSUM: from "
- "%s%s%s inode "DFID" object "
- LPU64"/"LPU64" extent "
- "["LPU64"-"LPU64"]\n",
- req->rq_import->imp_obd->obd_name,
- libcfs_nid2str(peer->nid),
- via, router,
- body->oa.o_valid & OBD_MD_FLFID ?
- body->oa.o_parent_seq : (__u64)0,
- body->oa.o_valid & OBD_MD_FLFID ?
- body->oa.o_parent_oid : 0,
- body->oa.o_valid & OBD_MD_FLFID ?
- body->oa.o_parent_ver : 0,
- body->oa.o_id,
- body->oa.o_valid & OBD_MD_FLGROUP ?
- body->oa.o_seq : (__u64)0,
- aa->aa_ppga[0]->off,
- aa->aa_ppga[aa->aa_page_count-1]->off +
- aa->aa_ppga[aa->aa_page_count-1]->count -
- 1);
- CERROR("client %x, server %x, cksum_type %x\n",
- client_cksum, server_cksum, cksum_type);
- cksum_counter = 0;
- aa->aa_oa->o_cksum = client_cksum;
- rc = -EAGAIN;
- } else {
- cksum_counter++;
- CDEBUG(D_PAGE, "checksum %x confirmed\n", client_cksum);
- rc = 0;
- }
+ if (server_cksum == ~0 && rc > 0) {
+ CERROR("Protocol error: server %s set the 'checksum' "
+ "bit, but didn't send a checksum. Not fatal, "
+ "but please notify on http://bugs.whamcloud.com/\n",
+ libcfs_nid2str(peer->nid));
+ } else if (server_cksum != client_cksum) {
+ LCONSOLE_ERROR_MSG(0x133, "%s: BAD READ CHECKSUM: from "
+ "%s%s%s inode "DFID" object "DOSTID
+ " extent ["LPU64"-"LPU64"]\n",
+ req->rq_import->imp_obd->obd_name,
+ libcfs_nid2str(peer->nid),
+ via, router,
+ body->oa.o_valid & OBD_MD_FLFID ?
+ body->oa.o_parent_seq : (__u64)0,
+ body->oa.o_valid & OBD_MD_FLFID ?
+ body->oa.o_parent_oid : 0,
+ body->oa.o_valid & OBD_MD_FLFID ?
+ body->oa.o_parent_ver : 0,
+ POSTID(&body->oa.o_oi),
+ aa->aa_ppga[0]->off,
+ aa->aa_ppga[aa->aa_page_count-1]->off +
+ aa->aa_ppga[aa->aa_page_count-1]->count -
+ 1);
+ CERROR("client %x, server %x, cksum_type %x\n",
+ client_cksum, server_cksum, cksum_type);
+ cksum_counter = 0;
+ aa->aa_oa->o_cksum = client_cksum;
+ rc = -EAGAIN;
+ } else {
+ cksum_counter++;
+ CDEBUG(D_PAGE, "checksum %x confirmed\n", client_cksum);
+ rc = 0;
+ }
} else if (unlikely(client_cksum)) {
static int cksum_missed;
rc = 0;
}
out:
- if (rc >= 0)
- lustre_get_wire_obdo(aa->aa_oa, &body->oa);
+ if (rc >= 0)
+ lustre_get_wire_obdo(&req->rq_import->imp_connect_data,
+ aa->aa_oa, &body->oa);
RETURN(rc);
}
obd_count page_count, struct brw_page **pga,
struct obd_capa *ocapa)
{
- struct ptlrpc_request *req;
- int rc;
- cfs_waitq_t waitq;
- int generation, resends = 0;
- struct l_wait_info lwi;
+ struct ptlrpc_request *req;
+ int rc;
+ wait_queue_head_t waitq;
+ int generation, resends = 0;
+ struct l_wait_info lwi;
- ENTRY;
+ ENTRY;
- cfs_waitq_init(&waitq);
- generation = exp->exp_obd->u.cli.cl_import->imp_generation;
+ init_waitqueue_head(&waitq);
+ generation = exp->exp_obd->u.cli.cl_import->imp_generation;
restart_bulk:
rc = osc_brw_prep_request(cmd, &exp->exp_obd->u.cli, oa, lsm,
if (rc != -EINPROGRESS &&
!client_should_resend(resends, &exp->exp_obd->u.cli)) {
CERROR("%s: too many resend retries for object: "
- ""LPU64":"LPU64", rc = %d.\n",
- exp->exp_obd->obd_name, oa->o_id, oa->o_seq, rc);
- goto out;
- }
- if (generation !=
- exp->exp_obd->u.cli.cl_import->imp_generation) {
- CDEBUG(D_HA, "%s: resend cross eviction for object: "
- ""LPU64":"LPU64", rc = %d.\n",
- exp->exp_obd->obd_name, oa->o_id, oa->o_seq, rc);
+ ""DOSTID", rc = %d.\n", exp->exp_obd->obd_name,
+ POSTID(&oa->o_oi), rc);
goto out;
}
+ if (generation !=
+ exp->exp_obd->u.cli.cl_import->imp_generation) {
+ CDEBUG(D_HA, "%s: resend cross eviction for object: "
+ ""DOSTID", rc = %d.\n", exp->exp_obd->obd_name,
+ POSTID(&oa->o_oi), rc);
+ goto out;
+ }
lwi = LWI_TIMEOUT_INTR(cfs_time_seconds(resends), NULL, NULL,
NULL);
aa->aa_resends++;
new_req->rq_interpret_reply = request->rq_interpret_reply;
new_req->rq_async_args = request->rq_async_args;
+ new_req->rq_commit_cb = request->rq_commit_cb;
/* cap resend delay to the current request timeout, this is similar to
* what ptlrpc does (see after_reply()) */
if (aa->aa_resends > new_req->rq_timeout)
cfs_list_splice_init(&aa->aa_oaps, &new_aa->aa_oaps);
CFS_INIT_LIST_HEAD(&new_aa->aa_exts);
cfs_list_splice_init(&aa->aa_exts, &new_aa->aa_exts);
+ new_aa->aa_resends = aa->aa_resends;
cfs_list_for_each_entry(oap, &new_aa->aa_oaps, oap_rpc_item) {
if (oap->oap_request) {
if (pages == 0) /* that's all */
return count;
- if (offset + pg[i]->count < CFS_PAGE_SIZE)
+ if (offset + pg[i]->count < PAGE_CACHE_SIZE)
return count; /* doesn't end on page boundary */
i++;
CDEBUG(D_INODE, "request %p aa %p rc %d\n", req, aa, rc);
/* When server return -EINPROGRESS, client should always retry
* regardless of the number of times the bulk was resent already. */
- if (osc_recoverable_error(rc)) {
- if (req->rq_import_generation !=
- req->rq_import->imp_generation) {
- CDEBUG(D_HA, "%s: resend cross eviction for object: "
- ""LPU64":"LPU64", rc = %d.\n",
- req->rq_import->imp_obd->obd_name,
- aa->aa_oa->o_id, aa->aa_oa->o_seq, rc);
- } else if (rc == -EINPROGRESS ||
- client_should_resend(aa->aa_resends, aa->aa_cli)) {
- rc = osc_brw_redo_request(req, aa, rc);
- } else {
- CERROR("%s: too many resent retries for object: "
- ""LPU64":"LPU64", rc = %d.\n",
- req->rq_import->imp_obd->obd_name,
- aa->aa_oa->o_id, aa->aa_oa->o_seq, rc);
- }
+ if (osc_recoverable_error(rc)) {
+ if (req->rq_import_generation !=
+ req->rq_import->imp_generation) {
+ CDEBUG(D_HA, "%s: resend cross eviction for object: "
+ ""DOSTID", rc = %d.\n",
+ req->rq_import->imp_obd->obd_name,
+ POSTID(&aa->aa_oa->o_oi), rc);
+ } else if (rc == -EINPROGRESS ||
+ client_should_resend(aa->aa_resends, aa->aa_cli)) {
+ rc = osc_brw_redo_request(req, aa, rc);
+ } else {
+ CERROR("%s: too many resent retries for object: "
+ ""LPU64":"LPU64", rc = %d.\n",
+ req->rq_import->imp_obd->obd_name,
+ POSTID(&aa->aa_oa->o_oi), rc);
+ }
- if (rc == 0)
- RETURN(0);
- else if (rc == -EAGAIN || rc == -EINPROGRESS)
- rc = -EIO;
- }
+ if (rc == 0)
+ RETURN(0);
+ else if (rc == -EAGAIN || rc == -EINPROGRESS)
+ rc = -EIO;
+ }
if (aa->aa_ocapa) {
capa_put(aa->aa_ocapa);
RETURN(rc);
}
+static void brw_commit(struct ptlrpc_request *req)
+{
+ spin_lock(&req->rq_lock);
+ /* If osc_inc_unstable_pages (via osc_extent_finish) races with
+ * this called via the rq_commit_cb, I need to ensure
+ * osc_dec_unstable_pages is still called. Otherwise unstable
+ * pages may be leaked. */
+ if (req->rq_unstable) {
+ spin_unlock(&req->rq_lock);
+ osc_dec_unstable_pages(req);
+ spin_lock(&req->rq_lock);
+ } else {
+ req->rq_committed = 1;
+ }
+ spin_unlock(&req->rq_lock);
+}
+
/**
* Build an RPC by the list of extent @ext_list. The caller must ensure
* that the total pages in this list are NOT over max pages per RPC.
int osc_build_rpc(const struct lu_env *env, struct client_obd *cli,
cfs_list_t *ext_list, int cmd, pdl_policy_t pol)
{
- struct ptlrpc_request *req = NULL;
- struct osc_extent *ext;
+ struct ptlrpc_request *req = NULL;
+ struct osc_extent *ext;
+ struct brw_page **pga = NULL;
+ struct osc_brw_async_args *aa = NULL;
+ struct obdo *oa = NULL;
+ struct osc_async_page *oap;
+ struct osc_async_page *tmp;
+ struct cl_req *clerq = NULL;
+ enum cl_req_type crt = (cmd & OBD_BRW_WRITE) ? CRT_WRITE :
+ CRT_READ;
+ struct ldlm_lock *lock = NULL;
+ struct cl_req_attr *crattr = NULL;
+ obd_off starting_offset = OBD_OBJECT_EOF;
+ obd_off ending_offset = 0;
+ int mpflag = 0;
+ int mem_tight = 0;
+ int page_count = 0;
+ int i;
+ int rc;
CFS_LIST_HEAD(rpc_list);
- struct brw_page **pga = NULL;
- struct osc_brw_async_args *aa = NULL;
- struct obdo *oa = NULL;
- struct osc_async_page *oap;
- struct osc_async_page *tmp;
- struct cl_req *clerq = NULL;
- enum cl_req_type crt = (cmd & OBD_BRW_WRITE) ? CRT_WRITE : CRT_READ;
- struct ldlm_lock *lock = NULL;
- struct cl_req_attr crattr;
- obd_off starting_offset = OBD_OBJECT_EOF;
- obd_off ending_offset = 0;
- int i, rc, mpflag = 0, mem_tight = 0, page_count = 0;
ENTRY;
LASSERT(!cfs_list_empty(ext_list));
oap->oap_count;
else
LASSERT(oap->oap_page_off + oap->oap_count ==
- CFS_PAGE_SIZE);
+ PAGE_CACHE_SIZE);
}
}
if (mem_tight)
mpflag = cfs_memory_pressure_get_and_set();
- memset(&crattr, 0, sizeof crattr);
+ OBD_ALLOC(crattr, sizeof(*crattr));
+ if (crattr == NULL)
+ GOTO(out, rc = -ENOMEM);
+
OBD_ALLOC(pga, sizeof(*pga) * page_count);
if (pga == NULL)
GOTO(out, rc = -ENOMEM);
struct cl_page *page = oap2cl_page(oap);
if (clerq == NULL) {
clerq = cl_req_alloc(env, page, crt,
- 1 /* only 1-object rpcs for
- * now */);
+ 1 /* only 1-object rpcs for now */);
if (IS_ERR(clerq))
GOTO(out, rc = PTR_ERR(clerq));
lock = oap->oap_ldlm_lock;
}
if (mem_tight)
oap->oap_brw_flags |= OBD_BRW_MEMALLOC;
- pga[i] = &oap->oap_brw_page;
- pga[i]->off = oap->oap_obj_off + oap->oap_page_off;
- CDEBUG(0, "put page %p index %lu oap %p flg %x to pga\n",
- pga[i]->pg, cfs_page_index(oap->oap_page), oap, pga[i]->flag);
- i++;
- cl_req_page_add(env, clerq, page);
- }
+ pga[i] = &oap->oap_brw_page;
+ pga[i]->off = oap->oap_obj_off + oap->oap_page_off;
+ CDEBUG(0, "put page %p index %lu oap %p flg %x to pga\n",
+ pga[i]->pg, page_index(oap->oap_page), oap,
+ pga[i]->flag);
+ i++;
+ cl_req_page_add(env, clerq, page);
+ }
- /* always get the data for the obdo for the rpc */
+ /* always get the data for the obdo for the rpc */
LASSERT(clerq != NULL);
- crattr.cra_oa = oa;
- crattr.cra_capa = NULL;
- memset(crattr.cra_jobid, 0, JOBSTATS_JOBID_SIZE);
- cl_req_attr_set(env, clerq, &crattr, ~0ULL);
- if (lock) {
- oa->o_handle = lock->l_remote_handle;
- oa->o_valid |= OBD_MD_FLHANDLE;
- }
+ crattr->cra_oa = oa;
+ cl_req_attr_set(env, clerq, crattr, ~0ULL);
+ if (lock) {
+ oa->o_handle = lock->l_remote_handle;
+ oa->o_valid |= OBD_MD_FLHANDLE;
+ }
- rc = cl_req_prep(env, clerq);
- if (rc != 0) {
- CERROR("cl_req_prep failed: %d\n", rc);
+ rc = cl_req_prep(env, clerq);
+ if (rc != 0) {
+ CERROR("cl_req_prep failed: %d\n", rc);
GOTO(out, rc);
}
sort_brw_pages(pga, page_count);
rc = osc_brw_prep_request(cmd, cli, oa, NULL, page_count,
- pga, &req, crattr.cra_capa, 1, 0);
+ pga, &req, crattr->cra_capa, 1, 0);
if (rc != 0) {
CERROR("prep_req failed: %d\n", rc);
GOTO(out, rc);
}
+ req->rq_commit_cb = brw_commit;
req->rq_interpret_reply = brw_interpret;
+
if (mem_tight != 0)
- req->rq_memalloc = 1;
+ req->rq_memalloc = 1;
- /* Need to update the timestamps after the request is built in case
- * we race with setattr (locally or in queue at OST). If OST gets
- * later setattr before earlier BRW (as determined by the request xid),
- * the OST will not use BRW timestamps. Sadly, there is no obvious
- * way to do this in a single call. bug 10150 */
- cl_req_attr_set(env, clerq, &crattr,
- OBD_MD_FLMTIME|OBD_MD_FLCTIME|OBD_MD_FLATIME);
+ /* Need to update the timestamps after the request is built in case
+ * we race with setattr (locally or in queue at OST). If OST gets
+ * later setattr before earlier BRW (as determined by the request xid),
+ * the OST will not use BRW timestamps. Sadly, there is no obvious
+ * way to do this in a single call. bug 10150 */
+ cl_req_attr_set(env, clerq, crattr,
+ OBD_MD_FLMTIME|OBD_MD_FLCTIME|OBD_MD_FLATIME);
- lustre_msg_set_jobid(req->rq_reqmsg, crattr.cra_jobid);
+ lustre_msg_set_jobid(req->rq_reqmsg, crattr->cra_jobid);
CLASSERT(sizeof(*aa) <= sizeof(req->rq_async_args));
aa = ptlrpc_req_async_args(req);
tmp->oap_request = ptlrpc_request_addref(req);
client_obd_list_lock(&cli->cl_loi_list_lock);
- starting_offset >>= CFS_PAGE_SHIFT;
+ starting_offset >>= PAGE_CACHE_SHIFT;
if (cmd == OBD_BRW_READ) {
cli->cl_r_in_flight++;
lprocfs_oh_tally_log2(&cli->cl_read_page_hist, page_count);
if (mem_tight != 0)
cfs_memory_pressure_restore(mpflag);
- capa_put(crattr.cra_capa);
+ if (crattr != NULL) {
+ capa_put(crattr->cra_capa);
+ OBD_FREE(crattr, sizeof(*crattr));
+ }
+
if (rc != 0) {
LASSERT(req == NULL);
- if (oa)
- OBDO_FREE(oa);
- if (pga)
- OBD_FREE(pga, sizeof(*pga) * page_count);
- /* this should happen rarely and is pretty bad, it makes the
- * pending list not follow the dirty order */
+ if (oa)
+ OBDO_FREE(oa);
+ if (pga)
+ OBD_FREE(pga, sizeof(*pga) * page_count);
+ /* this should happen rarely and is pretty bad, it makes the
+ * pending list not follow the dirty order */
while (!cfs_list_empty(ext_list)) {
ext = cfs_list_entry(ext_list->next, struct osc_extent,
oe_link);
struct ldlm_res_id res_id;
struct obd_device *obd = class_exp2obd(exp);
- osc_build_res_name(lsm->lsm_object_id, lsm->lsm_object_seq, &res_id);
+ ostid_build_res_name(&lsm->lsm_oi, &res_id);
ldlm_resource_iterate(obd->obd_namespace, &res_id, replace, data);
return 0;
}
struct obd_device *obd = class_exp2obd(exp);
int rc = 0;
- osc_build_res_name(lsm->lsm_object_id, lsm->lsm_object_seq, &res_id);
+ ostid_build_res_name(&lsm->lsm_oi, &res_id);
rc = ldlm_resource_iterate(obd->obd_namespace, &res_id, replace, data);
if (rc == LDLM_ITER_STOP)
return(1);
&RMF_DLM_REP);
LASSERT(rep != NULL);
+ rep->lock_policy_res1 =
+ ptlrpc_status_ntoh(rep->lock_policy_res1);
if (rep->lock_policy_res1)
rc = rep->lock_policy_res1;
}
}
void osc_update_enqueue(struct lustre_handle *lov_lockhp,
- struct lov_oinfo *loi, int flags,
- struct ost_lvb *lvb, __u32 mode, int rc)
+ struct lov_oinfo *loi, __u64 flags,
+ struct ost_lvb *lvb, __u32 mode, int rc)
{
struct ldlm_lock *lock = ldlm_handle2lock(lov_lockhp);
* release locks just after they are obtained. */
int osc_enqueue_base(struct obd_export *exp, struct ldlm_res_id *res_id,
__u64 *flags, ldlm_policy_data_t *policy,
- struct ost_lvb *lvb, int kms_valid,
- obd_enqueue_update_f upcall, void *cookie,
- struct ldlm_enqueue_info *einfo,
- struct lustre_handle *lockh,
- struct ptlrpc_request_set *rqset, int async, int agl)
+ struct ost_lvb *lvb, int kms_valid,
+ obd_enqueue_update_f upcall, void *cookie,
+ struct ldlm_enqueue_info *einfo,
+ struct lustre_handle *lockh,
+ struct ptlrpc_request_set *rqset, int async, int agl)
{
- struct obd_device *obd = exp->exp_obd;
- struct ptlrpc_request *req = NULL;
- int intent = *flags & LDLM_FL_HAS_INTENT;
- int match_lvb = (agl != 0 ? 0 : LDLM_FL_LVB_READY);
- ldlm_mode_t mode;
- int rc;
- ENTRY;
+ struct obd_device *obd = exp->exp_obd;
+ struct ptlrpc_request *req = NULL;
+ int intent = *flags & LDLM_FL_HAS_INTENT;
+ __u64 match_lvb = (agl != 0 ? 0 : LDLM_FL_LVB_READY);
+ ldlm_mode_t mode;
+ int rc;
+ ENTRY;
/* Filesystem lock extents are extended to page boundaries so that
* dealing with the page cache is a little smoother. */
if (mode) {
struct ldlm_lock *matched = ldlm_handle2lock(lockh);
- if ((agl != 0) && !(matched->l_flags & LDLM_FL_LVB_READY)) {
+ if ((agl != 0) && !ldlm_is_lvb_ready(matched)) {
/* For AGL, if enqueue RPC is sent but the lock is not
* granted, then skip to process this strpe.
* Return -ECANCELED to tell the caller. */
int rc;
ENTRY;
- osc_build_res_name(oinfo->oi_md->lsm_object_id,
- oinfo->oi_md->lsm_object_seq, &res_id);
-
+ ostid_build_res_name(&oinfo->oi_md->lsm_oi, &res_id);
rc = osc_enqueue_base(exp, &res_id, &oinfo->oi_flags, &oinfo->oi_policy,
&oinfo->oi_md->lsm_oinfo[0]->loi_lvb,
oinfo->oi_md->lsm_oinfo[0]->loi_kms_valid,
}
int osc_match_base(struct obd_export *exp, struct ldlm_res_id *res_id,
- __u32 type, ldlm_policy_data_t *policy, __u32 mode,
- int *flags, void *data, struct lustre_handle *lockh,
- int unref)
+ __u32 type, ldlm_policy_data_t *policy, __u32 mode,
+ __u64 *flags, void *data, struct lustre_handle *lockh,
+ int unref)
{
- struct obd_device *obd = exp->exp_obd;
- int lflags = *flags;
- ldlm_mode_t rc;
- ENTRY;
+ struct obd_device *obd = exp->exp_obd;
+ __u64 lflags = *flags;
+ ldlm_mode_t rc;
+ ENTRY;
if (OBD_FAIL_CHECK(OBD_FAIL_OSC_MATCH))
RETURN(-EIO);
struct obd_device *obd = class_exp2obd(exp);
struct ldlm_res_id res_id, *resp = NULL;
- if (lsm != NULL) {
- resp = osc_build_res_name(lsm->lsm_object_id,
- lsm->lsm_object_seq, &res_id);
- }
+ if (lsm != NULL) {
+ ostid_build_res_name(&lsm->lsm_oi, &res_id);
+ resp = &res_id;
+ }
return ldlm_cli_cancel_unused(obd->obd_namespace, resp, flags, opaque);
}
/* we only need the header part from user space to get lmm_magic and
* lmm_stripe_count, (the header part is common to v1 and v3) */
lum_size = sizeof(struct lov_user_md_v1);
- if (cfs_copy_from_user(&lum, lump, lum_size))
+ if (copy_from_user(&lum, lump, lum_size))
RETURN(-EFAULT);
if ((lum.lmm_magic != LOV_USER_MAGIC_V1) &&
if (!lumk)
RETURN(-ENOMEM);
- if (lum.lmm_magic == LOV_USER_MAGIC_V1)
- lmm_objects = &(((struct lov_user_md_v1 *)lumk)->lmm_objects[0]);
- else
- lmm_objects = &(lumk->lmm_objects[0]);
- lmm_objects->l_object_id = lsm->lsm_object_id;
- } else {
- lum_size = lov_mds_md_size(0, lum.lmm_magic);
- lumk = &lum;
- }
+ if (lum.lmm_magic == LOV_USER_MAGIC_V1)
+ lmm_objects =
+ &(((struct lov_user_md_v1 *)lumk)->lmm_objects[0]);
+ else
+ lmm_objects = &(lumk->lmm_objects[0]);
+ lmm_objects->l_ost_oi = lsm->lsm_oi;
+ } else {
+ lum_size = lov_mds_md_size(0, lum.lmm_magic);
+ lumk = &lum;
+ }
- lumk->lmm_object_id = lsm->lsm_object_id;
- lumk->lmm_object_seq = lsm->lsm_object_seq;
- lumk->lmm_stripe_count = 1;
+ lumk->lmm_oi = lsm->lsm_oi;
+ lumk->lmm_stripe_count = 1;
- if (cfs_copy_to_user(lump, lumk, lum_size))
- rc = -EFAULT;
+ if (copy_to_user(lump, lumk, lum_size))
+ rc = -EFAULT;
- if (lumk != &lum)
- OBD_FREE(lumk, lum_size);
+ if (lumk != &lum)
+ OBD_FREE(lumk, lum_size);
- RETURN(rc);
+ RETURN(rc);
}
int err = 0;
ENTRY;
- if (!cfs_try_module_get(THIS_MODULE)) {
- CERROR("Can't get module. Is it alive?");
- return -EINVAL;
- }
+ if (!try_module_get(THIS_MODULE)) {
+ CERROR("Can't get module. Is it alive?");
+ return -EINVAL;
+ }
switch (cmd) {
case OBD_IOC_LOV_GET_CONFIG: {
char *buf;
memcpy(data->ioc_inlbuf2, &obd->obd_uuid, sizeof(uuid));
- err = cfs_copy_to_user((void *)uarg, buf, len);
+ err = copy_to_user((void *)uarg, buf, len);
if (err)
err = -EFAULT;
obd_ioctl_freedata(buf, len);
case OBD_IOC_PING_TARGET:
err = ptlrpc_obd_ping(obd);
GOTO(out, err);
- default:
- CDEBUG(D_INODE, "unrecognised ioctl %#x by %s\n",
- cmd, cfs_curproc_comm());
- GOTO(out, err = -ENOTTY);
- }
+ default:
+ CDEBUG(D_INODE, "unrecognised ioctl %#x by %s\n",
+ cmd, current_comm());
+ GOTO(out, err = -ENOTTY);
+ }
out:
- cfs_module_put(THIS_MODULE);
- return err;
+ module_put(THIS_MODULE);
+ return err;
}
static int osc_get_info(const struct lu_env *env, struct obd_export *exp,
ptlrpc_req_finished(req);
RETURN(rc);
} else if (KEY_IS(KEY_FIEMAP)) {
- struct ptlrpc_request *req;
- struct ll_user_fiemap *reply;
- char *tmp;
- int rc;
+ struct ll_fiemap_info_key *fm_key =
+ (struct ll_fiemap_info_key *)key;
+ struct ldlm_res_id res_id;
+ ldlm_policy_data_t policy;
+ struct lustre_handle lockh;
+ ldlm_mode_t mode = 0;
+ struct ptlrpc_request *req;
+ struct ll_user_fiemap *reply;
+ char *tmp;
+ int rc;
+
+ if (!(fm_key->fiemap.fm_flags & FIEMAP_FLAG_SYNC))
+ goto skip_locking;
+
+ policy.l_extent.start = fm_key->fiemap.fm_start &
+ CFS_PAGE_MASK;
+
+ if (OBD_OBJECT_EOF - fm_key->fiemap.fm_length <=
+ fm_key->fiemap.fm_start + PAGE_CACHE_SIZE - 1)
+ policy.l_extent.end = OBD_OBJECT_EOF;
+ else
+ policy.l_extent.end = (fm_key->fiemap.fm_start +
+ fm_key->fiemap.fm_length +
+ PAGE_CACHE_SIZE - 1) & CFS_PAGE_MASK;
+
+ ostid_build_res_name(&fm_key->oa.o_oi, &res_id);
+ mode = ldlm_lock_match(exp->exp_obd->obd_namespace,
+ LDLM_FL_BLOCK_GRANTED |
+ LDLM_FL_LVB_READY,
+ &res_id, LDLM_EXTENT, &policy,
+ LCK_PR | LCK_PW, &lockh, 0);
+ if (mode) { /* lock is cached on client */
+ if (mode != LCK_PR) {
+ ldlm_lock_addref(&lockh, LCK_PR);
+ ldlm_lock_decref(&lockh, LCK_PW);
+ }
+ } else { /* no cached lock, needs acquire lock on server side */
+ fm_key->oa.o_valid |= OBD_MD_FLFLAGS;
+ fm_key->oa.o_flags |= OBD_FL_SRVLOCK;
+ }
+skip_locking:
req = ptlrpc_request_alloc(class_exp2cliimp(exp),
&RQF_OST_GET_INFO_FIEMAP);
if (req == NULL)
- RETURN(-ENOMEM);
+ GOTO(drop_lock, rc = -ENOMEM);
req_capsule_set_size(&req->rq_pill, &RMF_FIEMAP_KEY,
RCL_CLIENT, keylen);
rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, OST_GET_INFO);
if (rc) {
ptlrpc_request_free(req);
- RETURN(rc);
+ GOTO(drop_lock, rc);
}
tmp = req_capsule_client_get(&req->rq_pill, &RMF_FIEMAP_KEY);
ptlrpc_request_set_replen(req);
rc = ptlrpc_queue_wait(req);
if (rc)
- GOTO(out1, rc);
+ GOTO(fini_req, rc);
reply = req_capsule_server_get(&req->rq_pill, &RMF_FIEMAP_VAL);
if (reply == NULL)
- GOTO(out1, rc = -EPROTO);
+ GOTO(fini_req, rc = -EPROTO);
memcpy(val, reply, *vallen);
- out1:
+fini_req:
ptlrpc_req_finished(req);
-
+drop_lock:
+ if (mode)
+ ldlm_lock_decref(&lockh, LCK_PR);
RETURN(rc);
}
Even if something bad goes through, we'd get a -EINVAL from OST
anyway. */
- if (KEY_IS(KEY_GRANT_SHRINK))
- req = ptlrpc_request_alloc(imp, &RQF_OST_SET_GRANT_INFO);
- else
- req = ptlrpc_request_alloc(imp, &RQF_OBD_SET_INFO);
-
- if (req == NULL)
- RETURN(-ENOMEM);
-
- req_capsule_set_size(&req->rq_pill, &RMF_SETINFO_KEY,
- RCL_CLIENT, keylen);
- req_capsule_set_size(&req->rq_pill, &RMF_SETINFO_VAL,
- RCL_CLIENT, vallen);
- rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, OST_SET_INFO);
- if (rc) {
- ptlrpc_request_free(req);
- RETURN(rc);
- }
+ req = ptlrpc_request_alloc(imp, KEY_IS(KEY_GRANT_SHRINK) ?
+ &RQF_OST_SET_GRANT_INFO :
+ &RQF_OBD_SET_INFO);
+ if (req == NULL)
+ RETURN(-ENOMEM);
+
+ req_capsule_set_size(&req->rq_pill, &RMF_SETINFO_KEY,
+ RCL_CLIENT, keylen);
+ if (!KEY_IS(KEY_GRANT_SHRINK))
+ req_capsule_set_size(&req->rq_pill, &RMF_SETINFO_VAL,
+ RCL_CLIENT, vallen);
+ rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, OST_SET_INFO);
+ if (rc) {
+ ptlrpc_request_free(req);
+ RETURN(rc);
+ }
- tmp = req_capsule_client_get(&req->rq_pill, &RMF_SETINFO_KEY);
- memcpy(tmp, key, keylen);
- tmp = req_capsule_client_get(&req->rq_pill, &RMF_SETINFO_VAL);
+ tmp = req_capsule_client_get(&req->rq_pill, &RMF_SETINFO_KEY);
+ memcpy(tmp, key, keylen);
+ tmp = req_capsule_client_get(&req->rq_pill, KEY_IS(KEY_GRANT_SHRINK) ?
+ &RMF_OST_BODY :
+ &RMF_SETINFO_VAL);
memcpy(tmp, val, vallen);
if (KEY_IS(KEY_GRANT_SHRINK)) {
client_obd_list_lock(&cli->cl_loi_list_lock);
data->ocd_grant = (cli->cl_avail_grant + cli->cl_dirty) ?:
- 2 * cli->cl_max_pages_per_rpc << CFS_PAGE_SHIFT;
+ 2 * cli_brw_size(obd);
lost_grant = cli->cl_lost_grant;
cli->cl_lost_grant = 0;
client_obd_list_unlock(&cli->cl_loi_list_lock);
CDEBUG(D_INFO, "Lustre OSC module (%p).\n", &osc_caches);
rc = lu_kmem_init(osc_caches);
+ if (rc)
+ RETURN(rc);
lprocfs_osc_init_vars(&lvars);