* Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
* Use is subject to license terms.
*
- * Copyright (c) 2011, 2013, Intel Corporation.
+ * Copyright (c) 2011, 2014, Intel Corporation.
*/
/*
* This file is part of Lustre, http://www.lustre.org/
#include <libcfs/libcfs.h>
-#ifndef __KERNEL__
-# include <liblustre.h>
-#endif
-
#include <lustre_dlm.h>
#include <lustre_net.h>
#include <lustre/lustre_user.h>
#include <obd_cksum.h>
-#include <obd_ost.h>
-#include <obd_lov.h>
-
-#ifdef __CYGWIN__
-# include <ctype.h>
-#endif
-
#include <lustre_ha.h>
#include <lprocfs_status.h>
-#include <lustre_log.h>
+#include <lustre_ioctl.h>
#include <lustre_debug.h>
#include <lustre_param.h>
#include <lustre_fid.h>
+#include <obd_class.h>
#include "osc_internal.h"
#include "osc_cl_internal.h"
-static void osc_release_ppga(struct brw_page **ppga, obd_count count);
-static int brw_interpret(const struct lu_env *env,
- struct ptlrpc_request *req, void *data, int rc);
-int osc_cleanup(struct obd_device *obd);
-
-/* 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)
-{
- int lmm_size;
- ENTRY;
-
- lmm_size = sizeof(**lmmp);
- if (lmmp == NULL)
- RETURN(lmm_size);
-
- 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 == NULL) {
- OBD_ALLOC(*lmmp, lmm_size);
- if (*lmmp == NULL)
- RETURN(-ENOMEM);
- }
-
- if (lsm)
- ostid_cpu_to_le(&lsm->lsm_oi, &(*lmmp)->lmm_oi);
-
- 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)
-{
- int lsm_size;
- struct obd_import *imp = class_exp2cliimp(exp);
- ENTRY;
-
- 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 (unlikely(ostid_id(&lmm->lmm_oi) == 0)) {
- CERROR("%s: zero lmm_object_id: rc = %d\n",
- exp->exp_obd->obd_name, -EINVAL);
- RETURN(-EINVAL);
- }
- }
+struct osc_brw_async_args {
+ struct obdo *aa_oa;
+ int aa_requested_nob;
+ int aa_nio_count;
+ obd_count aa_page_count;
+ int aa_resends;
+ struct brw_page **aa_ppga;
+ struct client_obd *aa_cli;
+ struct list_head aa_oaps;
+ struct list_head aa_exts;
+ struct obd_capa *aa_ocapa;
+ struct cl_req *aa_clerq;
+};
- lsm_size = lov_stripe_md_size(1);
- if (lsmp == NULL)
- RETURN(lsm_size);
+#define osc_grant_args osc_brw_async_args
- if (*lsmp != NULL && lmm == NULL) {
- OBD_FREE((*lsmp)->lsm_oinfo[0], sizeof(struct lov_oinfo));
- OBD_FREE(*lsmp, lsm_size);
- *lsmp = NULL;
- RETURN(0);
- }
+struct osc_async_args {
+ struct obd_info *aa_oi;
+};
- 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);
- }
+struct osc_setattr_args {
+ struct obdo *sa_oa;
+ obd_enqueue_update_f sa_upcall;
+ void *sa_cookie;
+};
- if (lmm != NULL)
- /* XXX zero *lsmp? */
- ostid_le_to_cpu(&lmm->lmm_oi, &(*lsmp)->lsm_oi);
+struct osc_fsync_args {
+ struct obd_info *fa_oi;
+ obd_enqueue_update_f fa_upcall;
+ void *fa_cookie;
+};
- 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;
+struct osc_enqueue_args {
+ struct obd_export *oa_exp;
+ ldlm_type_t oa_type;
+ ldlm_mode_t oa_mode;
+ __u64 *oa_flags;
+ osc_enqueue_upcall_f oa_upcall;
+ void *oa_cookie;
+ struct ost_lvb *oa_lvb;
+ struct lustre_handle oa_lockh;
+ unsigned int oa_agl:1;
+};
- RETURN(lsm_size);
-}
+static void osc_release_ppga(struct brw_page **ppga, obd_count count);
+static int brw_interpret(const struct lu_env *env, struct ptlrpc_request *req,
+ void *data, int rc);
static inline void osc_pack_capa(struct ptlrpc_request *req,
struct ost_body *body, void *capa)
oinfo->oi_cb_up, oinfo, rqset);
}
-int osc_real_create(struct obd_export *exp, struct obdo *oa,
- struct lov_stripe_md **ea, struct obd_trans_info *oti)
+static int osc_create(const struct lu_env *env, struct obd_export *exp,
+ struct obdo *oa, struct obd_trans_info *oti)
{
struct ptlrpc_request *req;
struct ost_body *body;
- struct lov_stripe_md *lsm;
int rc;
ENTRY;
- LASSERT(oa);
- LASSERT(ea);
-
- lsm = *ea;
- if (!lsm) {
- rc = obd_alloc_memmd(exp, &lsm);
- if (rc < 0)
- RETURN(rc);
- }
+ LASSERT(oa != NULL);
+ LASSERT(oa->o_valid & OBD_MD_FLGROUP);
+ LASSERT(fid_seq_is_echo(ostid_seq(&oa->o_oi)));
req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_OST_CREATE);
if (req == NULL)
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_oi = oa->o_oi;
- *ea = lsm;
-
if (oti != NULL) {
- oti->oti_transno = lustre_msg_get_transno(req->rq_repmsg);
-
if (oa->o_valid & OBD_MD_FLCOOKIE) {
- if (!oti->oti_logcookies)
- oti_alloc_cookies(oti, 1);
+ if (oti->oti_logcookies == NULL)
+ oti->oti_logcookies = &oti->oti_onecookie;
+
*oti->oti_logcookies = oa->o_lcookie;
}
}
out_req:
ptlrpc_req_finished(req);
out:
- if (rc && !*ea)
- obd_free_memmd(exp, &lsm);
- RETURN(rc);
+ RETURN(rc);
}
int osc_punch_base(struct obd_export *exp, struct obd_info *oinfo,
RETURN(0);
}
-static int osc_punch(const struct lu_env *env, struct obd_export *exp,
- struct obd_info *oinfo, struct obd_trans_info *oti,
- struct ptlrpc_request_set *rqset)
-{
- oinfo->oi_oa->o_size = oinfo->oi_policy.l_extent.start;
- oinfo->oi_oa->o_blocks = oinfo->oi_policy.l_extent.end;
- oinfo->oi_oa->o_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
- return osc_punch_base(exp, oinfo,
- oinfo->oi_cb_up, oinfo, rqset);
-}
-
static int osc_sync_interpret(const struct lu_env *env,
struct ptlrpc_request *req,
void *arg, int rc)
RETURN (0);
}
-static int osc_sync(const struct lu_env *env, struct obd_export *exp,
- struct obd_info *oinfo, obd_size start, obd_size end,
- struct ptlrpc_request_set *set)
-{
- ENTRY;
-
- if (!oinfo->oi_oa) {
- CDEBUG(D_INFO, "oa NULL\n");
- RETURN(-EINVAL);
- }
-
- oinfo->oi_oa->o_size = start;
- oinfo->oi_oa->o_blocks = end;
- oinfo->oi_oa->o_valid |= (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS);
-
- RETURN(osc_sync_base(exp, oinfo, oinfo->oi_cb_up, oinfo, set));
-}
-
/* Find and cancel locally locks matched by @mode in the resource found by
* @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,
+ struct list_head *cancels,
ldlm_mode_t mode, __u64 lock_flags)
{
struct ldlm_namespace *ns = exp->exp_obd->obd_namespace;
ostid_build_res_name(&oa->o_oi, &res_id);
res = ldlm_resource_get(ns, NULL, &res_id, 0, 0);
- if (res == NULL)
+ if (IS_ERR(res))
RETURN(0);
LDLM_RESOURCE_ADDREF(res);
{
struct client_obd *cli = &req->rq_import->imp_obd->u.cli;
- cfs_atomic_dec(&cli->cl_destroy_in_flight);
+ 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) <=
+ if (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) <
+ if (atomic_dec_return(&cli->cl_destroy_in_flight) <
cli->cl_max_rpcs_in_flight) {
/*
* The counter has been modified between the two atomic
return 0;
}
-int osc_create(const struct lu_env *env, struct obd_export *exp,
- struct obdo *oa, struct lov_stripe_md **ea,
- struct obd_trans_info *oti)
-{
- int rc = 0;
- ENTRY;
-
- LASSERT(oa);
- LASSERT(ea);
- LASSERT(oa->o_valid & OBD_MD_FLGROUP);
-
- if ((oa->o_valid & OBD_MD_FLFLAGS) &&
- oa->o_flags == OBD_FL_RECREATE_OBJS) {
- RETURN(osc_real_create(exp, oa, ea, oti));
- }
-
- if (!fid_seq_is_mdt(ostid_seq(&oa->o_oi)))
- RETURN(osc_real_create(exp, oa, ea, oti));
-
- /* we should not get here anymore */
- LBUG();
-
- RETURN(rc);
-}
-
/* Destroy requests can be async always on the client, and we don't even really
* care about the return code since the client cannot do anything at all about
* a destroy failure.
* it will retrieve the llog unlink logs and then sends the log cancellation
* cookies to the MDS after committing destroy transactions. */
static int osc_destroy(const struct lu_env *env, struct obd_export *exp,
- struct obdo *oa, struct lov_stripe_md *ea,
- struct obd_trans_info *oti, struct obd_export *md_export,
- void *capa)
+ struct obdo *oa, struct obd_trans_info *oti)
{
struct client_obd *cli = &exp->exp_obd->u.cli;
struct ptlrpc_request *req;
struct ost_body *body;
- CFS_LIST_HEAD(cancels);
+ struct list_head cancels = LIST_HEAD_INIT(cancels);
int rc, count;
ENTRY;
RETURN(-ENOMEM);
}
- osc_set_capa_size(req, &RMF_CAPA1, (struct obd_capa *)capa);
+ osc_set_capa_size(req, &RMF_CAPA1, NULL);
rc = ldlm_prep_elc_req(exp, req, LUSTRE_OST_VERSION, OST_DESTROY,
0, &cancels, count);
if (rc) {
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);
/* If osc_destory is for destroying the unlink orphan,
LASSERT(!(oa->o_valid & bits));
- oa->o_valid |= bits;
- client_obd_list_lock(&cli->cl_loi_list_lock);
- oa->o_dirty = cli->cl_dirty;
- if (unlikely(cli->cl_dirty - cli->cl_dirty_transit >
- cli->cl_dirty_max)) {
+ oa->o_valid |= bits;
+ spin_lock(&cli->cl_loi_list_lock);
+ oa->o_dirty = cli->cl_dirty_pages << PAGE_CACHE_SHIFT;
+ if (unlikely(cli->cl_dirty_pages - cli->cl_dirty_transit >
+ cli->cl_dirty_max_pages)) {
CERROR("dirty %lu - %lu > dirty_max %lu\n",
- cli->cl_dirty, cli->cl_dirty_transit, cli->cl_dirty_max);
+ cli->cl_dirty_pages, cli->cl_dirty_transit,
+ cli->cl_dirty_max_pages);
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
+ } else if (unlikely(atomic_long_read(&obd_dirty_pages) -
+ atomic_long_read(&obd_dirty_transit_pages) >
+ (obd_max_dirty_pages + 1))) {
+ /* The atomic_read() allowing the 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",
+ CERROR("%s: dirty %ld - %ld > system dirty_max %lu\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),
+ atomic_long_read(&obd_dirty_pages),
+ atomic_long_read(&obd_dirty_transit_pages),
obd_max_dirty_pages);
oa->o_undirty = 0;
- } else if (unlikely(cli->cl_dirty_max - cli->cl_dirty > 0x7fffffff)) {
+ } else if (unlikely(cli->cl_dirty_max_pages - cli->cl_dirty_pages >
+ 0x7fffffff)) {
CERROR("dirty %lu - dirty_max %lu too big???\n",
- cli->cl_dirty, cli->cl_dirty_max);
+ cli->cl_dirty_pages, cli->cl_dirty_max_pages);
oa->o_undirty = 0;
} else {
- long max_in_flight = (cli->cl_max_pages_per_rpc <<
+ unsigned 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_undirty = max(cli->cl_dirty_max_pages << PAGE_CACHE_SHIFT,
+ max_in_flight);
}
oa->o_grant = cli->cl_avail_grant + cli->cl_reserved_grant;
oa->o_dropped = cli->cl_lost_grant;
cli->cl_lost_grant = 0;
- client_obd_list_unlock(&cli->cl_loi_list_lock);
+ spin_unlock(&cli->cl_loi_list_lock);
CDEBUG(D_CACHE,"dirty: "LPU64" undirty: %u dropped %u grant: "LPU64"\n",
oa->o_dirty, oa->o_undirty, oa->o_dropped, oa->o_grant);
static void __osc_update_grant(struct client_obd *cli, obd_size grant)
{
- client_obd_list_lock(&cli->cl_loi_list_lock);
- cli->cl_avail_grant += grant;
- client_obd_list_unlock(&cli->cl_loi_list_lock);
+ spin_lock(&cli->cl_loi_list_lock);
+ cli->cl_avail_grant += grant;
+ spin_unlock(&cli->cl_loi_list_lock);
}
static void osc_update_grant(struct client_obd *cli, struct ost_body *body)
static void osc_shrink_grant_local(struct client_obd *cli, struct obdo *oa)
{
- client_obd_list_lock(&cli->cl_loi_list_lock);
- oa->o_grant = cli->cl_avail_grant / 4;
- cli->cl_avail_grant -= oa->o_grant;
- client_obd_list_unlock(&cli->cl_loi_list_lock);
+ spin_lock(&cli->cl_loi_list_lock);
+ oa->o_grant = cli->cl_avail_grant / 4;
+ cli->cl_avail_grant -= oa->o_grant;
+ spin_unlock(&cli->cl_loi_list_lock);
if (!(oa->o_valid & OBD_MD_FLFLAGS)) {
oa->o_valid |= OBD_MD_FLFLAGS;
oa->o_flags = 0;
__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);
+ spin_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);
+ spin_unlock(&cli->cl_loi_list_lock);
return osc_shrink_grant_to_target(cli, target_bytes);
}
struct ost_body *body;
ENTRY;
- client_obd_list_lock(&cli->cl_loi_list_lock);
+ spin_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. */
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);
+ spin_unlock(&cli->cl_loi_list_lock);
RETURN(0);
}
- client_obd_list_unlock(&cli->cl_loi_list_lock);
+ spin_unlock(&cli->cl_loi_list_lock);
OBD_ALLOC_PTR(body);
if (!body)
osc_announce_cached(cli, &body->oa, 0);
- client_obd_list_lock(&cli->cl_loi_list_lock);
+ spin_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);
+ spin_unlock(&cli->cl_loi_list_lock);
if (!(body->oa.o_valid & OBD_MD_FLFLAGS)) {
body->oa.o_valid |= OBD_MD_FLFLAGS;
body->oa.o_flags = 0;
static int osc_grant_shrink_grant_cb(struct timeout_item *item, void *data)
{
- struct client_obd *client;
+ struct client_obd *client;
- cfs_list_for_each_entry(client, &item->ti_obd_list,
- cl_grant_shrink_list) {
- if (osc_should_shrink_grant(client))
- osc_shrink_grant(client);
- }
- return 0;
+ list_for_each_entry(client, &item->ti_obd_list, cl_grant_shrink_list) {
+ if (osc_should_shrink_grant(client))
+ osc_shrink_grant(client);
+ }
+ return 0;
}
static int osc_add_shrink_grant(struct client_obd *client)
static void osc_init_grant(struct client_obd *cli, struct obd_connect_data *ocd)
{
- /*
- * ocd_grant is the total grant amount we're expect to hold: if we've
- * been evicted, it's the new avail_grant amount, cl_dirty will drop
- * to 0 as inflight RPCs fail out; otherwise, it's avail_grant + dirty.
- *
- * race is tolerable here: if we're evicted, but imp_state already
- * left EVICTED state, then cl_dirty must be 0 already.
- */
- client_obd_list_lock(&cli->cl_loi_list_lock);
- if (cli->cl_import->imp_state == LUSTRE_IMP_EVICTED)
- cli->cl_avail_grant = ocd->ocd_grant;
- else
- cli->cl_avail_grant = ocd->ocd_grant - cli->cl_dirty;
+ /*
+ * ocd_grant is the total grant amount we're expect to hold: if we've
+ * been evicted, it's the new avail_grant amount, cl_dirty_pages will
+ * drop to 0 as inflight RPCs fail out; otherwise, it's avail_grant +
+ * dirty.
+ *
+ * race is tolerable here: if we're evicted, but imp_state already
+ * left EVICTED state, then cl_dirty_pages must be 0 already.
+ */
+ spin_lock(&cli->cl_loi_list_lock);
+ if (cli->cl_import->imp_state == LUSTRE_IMP_EVICTED)
+ cli->cl_avail_grant = ocd->ocd_grant;
+ else
+ cli->cl_avail_grant = ocd->ocd_grant -
+ (cli->cl_dirty_pages << PAGE_CACHE_SHIFT);
if (cli->cl_avail_grant < 0) {
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);
+ ocd->ocd_grant, cli->cl_dirty_pages << PAGE_CACHE_SHIFT);
/* 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, PAGE_CACHE_SHIFT, ocd->ocd_blocksize);
- client_obd_list_unlock(&cli->cl_loi_list_lock);
+ spin_unlock(&cli->cl_loi_list_lock);
CDEBUG(D_CACHE, "%s, setting cl_avail_grant: %ld cl_lost_grant: %ld."
"chunk bits: %d.\n", cli->cl_import->imp_obd->obd_name,
cli->cl_avail_grant, cli->cl_lost_grant, cli->cl_chunkbits);
if (ocd->ocd_connect_flags & OBD_CONNECT_GRANT_SHRINK &&
- cfs_list_empty(&cli->cl_grant_shrink_list))
+ list_empty(&cli->cl_grant_shrink_list))
osc_add_shrink_grant(cli);
}
* safe to combine */
if (unlikely((p1->flag & mask) != (p2->flag & mask))) {
CWARN("Saw flags 0x%x and 0x%x in the same brw, please "
- "report this at http://bugs.whamcloud.com/\n",
+ "report this at https://jira.hpdd.intel.com/\n",
p1->flag, p2->flag);
}
return 0;
OBD_FAIL_CHECK(OBD_FAIL_OSC_CHECKSUM_RECEIVE)) {
unsigned char *ptr = kmap(pga[i]->pg);
int off = pga[i]->off & ~CFS_PAGE_MASK;
+
memcpy(ptr + off, "bad1", min(4, nob));
kunmap(pga[i]->pg);
}
cfs_crypto_hash_update_page(hdesc, pga[i]->pg,
- pga[i]->off & ~CFS_PAGE_MASK,
- count);
+ pga[i]->off & ~CFS_PAGE_MASK,
+ count);
LL_CDEBUG_PAGE(D_PAGE, pga[i]->pg, "off %d\n",
(int)(pga[i]->off & ~CFS_PAGE_MASK));
i++;
}
- bufsize = 4;
+ bufsize = sizeof(cksum);
err = cfs_crypto_hash_final(hdesc, (unsigned char *)&cksum, &bufsize);
- if (err)
- cfs_crypto_hash_final(hdesc, NULL, NULL);
-
/* For sending we only compute the wrong checksum instead
* of corrupting the data so it is still correct on a redo */
if (opc == OST_WRITE && OBD_FAIL_CHECK(OBD_FAIL_OSC_CHECKSUM_SEND))
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
" prev_pg %p [pri %lu ind %lu] off "LPU64"\n",
pg->pg, page_private(pg->pg), pg->pg->index, pg->off,
pg_prev->pg, page_private(pg_prev->pg),
pg_prev->pg->index, pg_prev->off);
-#else
- LASSERTF(i == 0 || pg->off > pg_prev->off,
- "i %d p_c %u\n", i, page_count);
-#endif
LASSERT((pga[0]->flag & OBD_BRW_SRVLOCK) ==
(pg->flag & OBD_BRW_SRVLOCK));
if (i > 0 && can_merge_pages(pg_prev, pg)) {
niobuf--;
- niobuf->len += pg->count;
- } else {
- niobuf->offset = pg->off;
- niobuf->len = pg->count;
- niobuf->flags = pg->flag;
+ niobuf->rnb_len += pg->count;
+ } else {
+ niobuf->rnb_offset = pg->off;
+ niobuf->rnb_len = pg->count;
+ niobuf->rnb_flags = pg->flag;
}
pg_prev = pg;
}
aa->aa_resends = 0;
aa->aa_ppga = pga;
aa->aa_cli = cli;
- CFS_INIT_LIST_HEAD(&aa->aa_oaps);
+ INIT_LIST_HEAD(&aa->aa_oaps);
if (ocapa && reserve)
aa->aa_ocapa = capa_get(ocapa);
- *reqp = req;
+ *reqp = req;
+ niobuf = req_capsule_client_get(pill, &RMF_NIOBUF_REMOTE);
+ CDEBUG(D_RPCTRACE, "brw rpc %p - object "DOSTID" offset %lld<>%lld\n",
+ req, POSTID(&oa->o_oi), niobuf[0].rnb_offset,
+ niobuf[niocount - 1].rnb_offset + niobuf[niocount - 1].rnb_len);
RETURN(0);
out:
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) {
+ 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",
RETURN(rc);
}
-static int osc_brw_internal(int cmd, struct obd_export *exp, struct obdo *oa,
- struct lov_stripe_md *lsm,
- obd_count page_count, struct brw_page **pga,
- struct obd_capa *ocapa)
-{
- struct ptlrpc_request *req;
- int rc;
- wait_queue_head_t waitq;
- int generation, resends = 0;
- struct l_wait_info lwi;
-
- ENTRY;
-
- 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,
- page_count, pga, &req, ocapa, 0, resends);
- if (rc != 0)
- return (rc);
-
- if (resends) {
- req->rq_generation_set = 1;
- req->rq_import_generation = generation;
- req->rq_sent = cfs_time_current_sec() + resends;
- }
-
- rc = ptlrpc_queue_wait(req);
-
- if (rc == -ETIMEDOUT && req->rq_resend) {
- DEBUG_REQ(D_HA, req, "BULK TIMEOUT");
- ptlrpc_req_finished(req);
- goto restart_bulk;
- }
-
- rc = osc_brw_fini_request(req, rc);
-
- ptlrpc_req_finished(req);
- /* When server return -EINPROGRESS, client should always retry
- * regardless of the number of times the bulk was resent already.*/
- if (osc_recoverable_error(rc)) {
- resends++;
- if (rc != -EINPROGRESS &&
- !client_should_resend(resends, &exp->exp_obd->u.cli)) {
- CERROR("%s: too many resend retries for object: "
- ""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);
- l_wait_event(waitq, 0, &lwi);
-
- goto restart_bulk;
- }
-out:
- if (rc == -EAGAIN || rc == -EINPROGRESS)
- rc = -EIO;
- RETURN (rc);
-}
-
static int osc_brw_redo_request(struct ptlrpc_request *request,
struct osc_brw_async_args *aa, int rc)
{
if (rc)
RETURN(rc);
- cfs_list_for_each_entry(oap, &aa->aa_oaps, oap_rpc_item) {
+ list_for_each_entry(oap, &aa->aa_oaps, oap_rpc_item) {
if (oap->oap_request != NULL) {
LASSERTF(request == oap->oap_request,
"request %p != oap_request %p\n",
new_aa = ptlrpc_req_async_args(new_req);
- CFS_INIT_LIST_HEAD(&new_aa->aa_oaps);
- 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);
+ INIT_LIST_HEAD(&new_aa->aa_oaps);
+ list_splice_init(&aa->aa_oaps, &new_aa->aa_oaps);
+ INIT_LIST_HEAD(&new_aa->aa_exts);
+ 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) {
+ list_for_each_entry(oap, &new_aa->aa_oaps, oap_rpc_item) {
if (oap->oap_request) {
ptlrpc_req_finished(oap->oap_request);
oap->oap_request = ptlrpc_request_addref(new_req);
} while (stride > 1);
}
-static obd_count max_unfragmented_pages(struct brw_page **pg, obd_count pages)
-{
- int count = 1;
- int offset;
- int i = 0;
-
- LASSERT (pages > 0);
- offset = pg[i]->off & ~CFS_PAGE_MASK;
-
- for (;;) {
- pages--;
- if (pages == 0) /* that's all */
- return count;
-
- if (offset + pg[i]->count < PAGE_CACHE_SIZE)
- return count; /* doesn't end on page boundary */
-
- i++;
- offset = pg[i]->off & ~CFS_PAGE_MASK;
- if (offset != 0) /* doesn't start on page boundary */
- return count;
-
- count++;
- }
-}
-
-static struct brw_page **osc_build_ppga(struct brw_page *pga, obd_count count)
-{
- struct brw_page **ppga;
- int i;
-
- OBD_ALLOC(ppga, sizeof(*ppga) * count);
- if (ppga == NULL)
- return NULL;
-
- for (i = 0; i < count; i++)
- ppga[i] = pga + i;
- return ppga;
-}
-
static void osc_release_ppga(struct brw_page **ppga, obd_count count)
{
LASSERT(ppga != NULL);
OBD_FREE(ppga, sizeof(*ppga) * count);
}
-static int osc_brw(int cmd, struct obd_export *exp, struct obd_info *oinfo,
- obd_count page_count, struct brw_page *pga,
- struct obd_trans_info *oti)
-{
- struct obdo *saved_oa = NULL;
- struct brw_page **ppga, **orig;
- struct obd_import *imp = class_exp2cliimp(exp);
- struct client_obd *cli;
- int rc, page_count_orig;
- ENTRY;
-
- LASSERT((imp != NULL) && (imp->imp_obd != NULL));
- cli = &imp->imp_obd->u.cli;
-
- if (cmd & OBD_BRW_CHECK) {
- /* The caller just wants to know if there's a chance that this
- * I/O can succeed */
-
- if (imp->imp_invalid)
- RETURN(-EIO);
- RETURN(0);
- }
-
- /* test_brw with a failed create can trip this, maybe others. */
- LASSERT(cli->cl_max_pages_per_rpc);
-
- rc = 0;
-
- orig = ppga = osc_build_ppga(pga, page_count);
- if (ppga == NULL)
- RETURN(-ENOMEM);
- page_count_orig = page_count;
-
- sort_brw_pages(ppga, page_count);
- while (page_count) {
- obd_count pages_per_brw;
-
- if (page_count > cli->cl_max_pages_per_rpc)
- pages_per_brw = cli->cl_max_pages_per_rpc;
- else
- pages_per_brw = page_count;
-
- pages_per_brw = max_unfragmented_pages(ppga, pages_per_brw);
-
- if (saved_oa != NULL) {
- /* restore previously saved oa */
- *oinfo->oi_oa = *saved_oa;
- } else if (page_count > pages_per_brw) {
- /* save a copy of oa (brw will clobber it) */
- OBDO_ALLOC(saved_oa);
- if (saved_oa == NULL)
- GOTO(out, rc = -ENOMEM);
- *saved_oa = *oinfo->oi_oa;
- }
-
- rc = osc_brw_internal(cmd, exp, oinfo->oi_oa, oinfo->oi_md,
- pages_per_brw, ppga, oinfo->oi_capa);
-
- if (rc != 0)
- break;
-
- page_count -= pages_per_brw;
- ppga += pages_per_brw;
- }
-
-out:
- osc_release_ppga(orig, page_count_orig);
-
- if (saved_oa != NULL)
- OBDO_FREE(saved_oa);
-
- RETURN(rc);
-}
-
static int brw_interpret(const struct lu_env *env,
struct ptlrpc_request *req, void *data, int rc)
{
if (lustre_msg_get_opc(req->rq_reqmsg) == OST_WRITE) {
struct lov_oinfo *loi = cl2osc(obj)->oo_oinfo;
- loff_t last_off = last->oap_count + last->oap_obj_off;
+ loff_t last_off = last->oap_count + last->oap_obj_off +
+ last->oap_page_off;
/* Change file size if this is an out of quota or
* direct IO write and it extends the file size */
}
OBDO_FREE(aa->aa_oa);
- cfs_list_for_each_entry_safe(ext, tmp, &aa->aa_exts, oe_link) {
- cfs_list_del_init(&ext->oe_link);
+ if (lustre_msg_get_opc(req->rq_reqmsg) == OST_WRITE && rc == 0)
+ osc_inc_unstable_pages(req);
+
+ list_for_each_entry_safe(ext, tmp, &aa->aa_exts, oe_link) {
+ list_del_init(&ext->oe_link);
osc_extent_finish(env, ext, 1, rc);
}
- LASSERT(cfs_list_empty(&aa->aa_exts));
- LASSERT(cfs_list_empty(&aa->aa_oaps));
+ LASSERT(list_empty(&aa->aa_exts));
+ LASSERT(list_empty(&aa->aa_oaps));
cl_req_completion(env, aa->aa_clerq, rc < 0 ? rc :
req->rq_bulk->bd_nob_transferred);
osc_release_ppga(aa->aa_ppga, aa->aa_page_count);
ptlrpc_lprocfs_brw(req, req->rq_bulk->bd_nob_transferred);
- client_obd_list_lock(&cli->cl_loi_list_lock);
+ spin_lock(&cli->cl_loi_list_lock);
/* We need to decrement before osc_ap_completion->osc_wake_cache_waiters
* is called so we know whether to go to sync BRWs or wait for more
* RPCs to complete */
else
cli->cl_r_in_flight--;
osc_wake_cache_waiters(cli);
- client_obd_list_unlock(&cli->cl_loi_list_lock);
+ spin_unlock(&cli->cl_loi_list_lock);
osc_io_unplug(env, cli, NULL, PDL_POLICY_SAME);
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_lock(&req->rq_lock);
+ if (likely(req->rq_unstable)) {
+ req->rq_unstable = 0;
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);
}
- spin_unlock(&req->rq_lock);
}
/**
* Extents in the list must be in OES_RPC state.
*/
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 list_head *ext_list, int cmd, pdl_policy_t pol)
{
struct ptlrpc_request *req = NULL;
struct osc_extent *ext;
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;
+ bool soft_sync = false;
int i;
int rc;
- CFS_LIST_HEAD(rpc_list);
-
+ struct list_head rpc_list = LIST_HEAD_INIT(rpc_list);
+ struct ost_body *body;
ENTRY;
- LASSERT(!cfs_list_empty(ext_list));
+ LASSERT(!list_empty(ext_list));
/* add pages into rpc_list to build BRW rpc */
- cfs_list_for_each_entry(ext, ext_list, oe_link) {
+ list_for_each_entry(ext, ext_list, oe_link) {
LASSERT(ext->oe_state == OES_RPC);
mem_tight |= ext->oe_memalloc;
- cfs_list_for_each_entry(oap, &ext->oe_pages, oap_pending_item) {
+ list_for_each_entry(oap, &ext->oe_pages, oap_pending_item) {
++page_count;
- cfs_list_add_tail(&oap->oap_rpc_item, &rpc_list);
+ list_add_tail(&oap->oap_rpc_item, &rpc_list);
if (starting_offset > oap->oap_obj_off)
starting_offset = oap->oap_obj_off;
else
}
}
+ soft_sync = osc_over_unstable_soft_limit(cli);
if (mem_tight)
mpflag = cfs_memory_pressure_get_and_set();
GOTO(out, rc = -ENOMEM);
i = 0;
- cfs_list_for_each_entry(oap, &rpc_list, oap_rpc_item) {
+ list_for_each_entry(oap, &rpc_list, oap_rpc_item) {
struct cl_page *page = oap2cl_page(oap);
if (clerq == NULL) {
clerq = cl_req_alloc(env, page, crt,
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;
+ if (soft_sync)
+ oap->oap_brw_flags |= OBD_BRW_SOFT_SYNC;
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",
LASSERT(clerq != NULL);
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) {
* 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 */
+ body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
+ crattr->cra_oa = &body->oa;
cl_req_attr_set(env, clerq, crattr,
OBD_MD_FLMTIME|OBD_MD_FLCTIME|OBD_MD_FLATIME);
CLASSERT(sizeof(*aa) <= sizeof(req->rq_async_args));
aa = ptlrpc_req_async_args(req);
- CFS_INIT_LIST_HEAD(&aa->aa_oaps);
- cfs_list_splice_init(&rpc_list, &aa->aa_oaps);
- CFS_INIT_LIST_HEAD(&aa->aa_exts);
- cfs_list_splice_init(ext_list, &aa->aa_exts);
+ INIT_LIST_HEAD(&aa->aa_oaps);
+ list_splice_init(&rpc_list, &aa->aa_oaps);
+ INIT_LIST_HEAD(&aa->aa_exts);
+ list_splice_init(ext_list, &aa->aa_exts);
aa->aa_clerq = clerq;
/* queued sync pages can be torn down while the pages
* were between the pending list and the rpc */
tmp = NULL;
- cfs_list_for_each_entry(oap, &aa->aa_oaps, oap_rpc_item) {
+ list_for_each_entry(oap, &aa->aa_oaps, oap_rpc_item) {
/* only one oap gets a request reference */
if (tmp == NULL)
tmp = oap;
if (tmp != NULL)
tmp->oap_request = ptlrpc_request_addref(req);
- client_obd_list_lock(&cli->cl_loi_list_lock);
+ spin_lock(&cli->cl_loi_list_lock);
starting_offset >>= PAGE_CACHE_SHIFT;
if (cmd == OBD_BRW_READ) {
cli->cl_r_in_flight++;
lprocfs_oh_tally_log2(&cli->cl_write_offset_hist,
starting_offset + 1);
}
- client_obd_list_unlock(&cli->cl_loi_list_lock);
+ spin_unlock(&cli->cl_loi_list_lock);
- DEBUG_REQ(D_INODE, req, "%d pages, aa %p. now %dr/%dw in flight",
+ DEBUG_REQ(D_INODE, req, "%d pages, aa %p. now %ur/%uw in flight",
page_count, aa, cli->cl_r_in_flight,
cli->cl_w_in_flight);
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);
- cfs_list_del_init(&ext->oe_link);
+ while (!list_empty(ext_list)) {
+ ext = list_entry(ext_list->next, struct osc_extent,
+ oe_link);
+ list_del_init(&ext->oe_link);
osc_extent_finish(env, ext, 0, rc);
}
if (clerq && !IS_ERR(clerq))
LASSERT(lock->l_glimpse_ast == einfo->ei_cb_gl);
lock_res_and_lock(lock);
- spin_lock(&osc_ast_guard);
if (lock->l_ast_data == NULL)
lock->l_ast_data = data;
if (lock->l_ast_data == data)
set = 1;
- spin_unlock(&osc_ast_guard);
unlock_res_and_lock(lock);
return set;
return(rc);
}
-static int osc_enqueue_fini(struct ptlrpc_request *req, struct ost_lvb *lvb,
- obd_enqueue_update_f upcall, void *cookie,
- __u64 *flags, int agl, int rc)
+static int osc_enqueue_fini(struct ptlrpc_request *req,
+ osc_enqueue_upcall_f upcall, void *cookie,
+ struct lustre_handle *lockh, ldlm_mode_t mode,
+ __u64 *flags, int agl, int errcode)
{
- int intent = *flags & LDLM_FL_HAS_INTENT;
- ENTRY;
-
- if (intent) {
- /* The request was created before ldlm_cli_enqueue call. */
- if (rc == ELDLM_LOCK_ABORTED) {
- struct ldlm_reply *rep;
- rep = req_capsule_server_get(&req->rq_pill,
- &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;
- }
- }
+ bool intent = *flags & LDLM_FL_HAS_INTENT;
+ int rc;
+ ENTRY;
- if ((intent != 0 && rc == ELDLM_LOCK_ABORTED && agl == 0) ||
- (rc == 0)) {
- *flags |= LDLM_FL_LVB_READY;
- CDEBUG(D_INODE,"got kms "LPU64" blocks "LPU64" mtime "LPU64"\n",
- lvb->lvb_size, lvb->lvb_blocks, lvb->lvb_mtime);
- }
+ /* The request was created before ldlm_cli_enqueue call. */
+ if (intent && errcode == ELDLM_LOCK_ABORTED) {
+ struct ldlm_reply *rep;
+
+ rep = req_capsule_server_get(&req->rq_pill, &RMF_DLM_REP);
+ LASSERT(rep != NULL);
+
+ rep->lock_policy_res1 =
+ ptlrpc_status_ntoh(rep->lock_policy_res1);
+ if (rep->lock_policy_res1)
+ errcode = rep->lock_policy_res1;
+ if (!agl)
+ *flags |= LDLM_FL_LVB_READY;
+ } else if (errcode == ELDLM_OK) {
+ *flags |= LDLM_FL_LVB_READY;
+ }
/* Call the update callback. */
- rc = (*upcall)(cookie, rc);
+ rc = (*upcall)(cookie, lockh, errcode);
+
+ /* release the reference taken in ldlm_cli_enqueue() */
+ if (errcode == ELDLM_LOCK_MATCHED)
+ errcode = ELDLM_OK;
+ if (errcode == ELDLM_OK && lustre_handle_is_used(lockh))
+ ldlm_lock_decref(lockh, mode);
+
RETURN(rc);
}
struct ptlrpc_request *req,
struct osc_enqueue_args *aa, int rc)
{
- struct ldlm_lock *lock;
- struct lustre_handle handle;
- __u32 mode;
- struct ost_lvb *lvb;
- __u32 lvb_len;
- __u64 *flags = aa->oa_flags;
-
- /* Make a local copy of a lock handle and a mode, because aa->oa_*
- * might be freed anytime after lock upcall has been called. */
- lustre_handle_copy(&handle, aa->oa_lockh);
- mode = aa->oa_ei->ei_mode;
-
- /* ldlm_cli_enqueue is holding a reference on the lock, so it must
- * be valid. */
- lock = ldlm_handle2lock(&handle);
-
- /* Take an additional reference so that a blocking AST that
- * ldlm_cli_enqueue_fini() might post for a failed lock, is guaranteed
- * to arrive after an upcall has been executed by
- * osc_enqueue_fini(). */
- ldlm_lock_addref(&handle, mode);
-
- /* Let CP AST to grant the lock first. */
- OBD_FAIL_TIMEOUT(OBD_FAIL_OSC_CP_ENQ_RACE, 1);
-
- if (aa->oa_agl && rc == ELDLM_LOCK_ABORTED) {
- lvb = NULL;
- lvb_len = 0;
- } else {
- lvb = aa->oa_lvb;
- lvb_len = sizeof(*aa->oa_lvb);
- }
-
- /* Complete obtaining the lock procedure. */
- rc = ldlm_cli_enqueue_fini(aa->oa_exp, req, aa->oa_ei->ei_type, 1,
- mode, flags, lvb, lvb_len, &handle, rc);
- /* Complete osc stuff. */
- rc = osc_enqueue_fini(req, aa->oa_lvb, aa->oa_upcall, aa->oa_cookie,
- flags, aa->oa_agl, rc);
-
- OBD_FAIL_TIMEOUT(OBD_FAIL_OSC_CP_CANCEL_RACE, 10);
+ struct ldlm_lock *lock;
+ struct lustre_handle *lockh = &aa->oa_lockh;
+ ldlm_mode_t mode = aa->oa_mode;
+ struct ost_lvb *lvb = aa->oa_lvb;
+ __u32 lvb_len = sizeof(*lvb);
+ __u64 flags = 0;
- /* Release the lock for async request. */
- if (lustre_handle_is_used(&handle) && rc == ELDLM_OK)
- /*
- * Releases a reference taken by ldlm_cli_enqueue(), if it is
- * not already released by
- * ldlm_cli_enqueue_fini()->failed_lock_cleanup()
- */
- ldlm_lock_decref(&handle, mode);
-
- LASSERTF(lock != NULL, "lockh %p, req %p, aa %p - client evicted?\n",
- aa->oa_lockh, req, aa);
- ldlm_lock_decref(&handle, mode);
- LDLM_LOCK_PUT(lock);
- return rc;
-}
-
-void osc_update_enqueue(struct lustre_handle *lov_lockhp,
- struct lov_oinfo *loi, __u64 flags,
- struct ost_lvb *lvb, __u32 mode, int rc)
-{
- struct ldlm_lock *lock = ldlm_handle2lock(lov_lockhp);
+ ENTRY;
- if (rc == ELDLM_OK) {
- __u64 tmp;
+ /* ldlm_cli_enqueue is holding a reference on the lock, so it must
+ * be valid. */
+ lock = ldlm_handle2lock(lockh);
+ LASSERTF(lock != NULL,
+ "lockh "LPX64", req %p, aa %p - client evicted?\n",
+ lockh->cookie, req, aa);
+
+ /* Take an additional reference so that a blocking AST that
+ * ldlm_cli_enqueue_fini() might post for a failed lock, is guaranteed
+ * to arrive after an upcall has been executed by
+ * osc_enqueue_fini(). */
+ ldlm_lock_addref(lockh, mode);
+
+ /* Let cl_lock_state_wait fail with -ERESTARTSYS to unuse sublocks. */
+ OBD_FAIL_TIMEOUT(OBD_FAIL_LDLM_ENQUEUE_HANG, 2);
+
+ /* Let CP AST to grant the lock first. */
+ OBD_FAIL_TIMEOUT(OBD_FAIL_OSC_CP_ENQ_RACE, 1);
+
+ if (aa->oa_agl) {
+ LASSERT(aa->oa_lvb == NULL);
+ LASSERT(aa->oa_flags == NULL);
+ aa->oa_flags = &flags;
+ }
- LASSERT(lock != NULL);
- loi->loi_lvb = *lvb;
- tmp = loi->loi_lvb.lvb_size;
- /* Extend KMS up to the end of this lock and no further
- * A lock on [x,y] means a KMS of up to y + 1 bytes! */
- if (tmp > lock->l_policy_data.l_extent.end)
- tmp = lock->l_policy_data.l_extent.end + 1;
- if (tmp >= loi->loi_kms) {
- LDLM_DEBUG(lock, "lock acquired, setting rss="LPU64
- ", kms="LPU64, loi->loi_lvb.lvb_size, tmp);
- loi_kms_set(loi, tmp);
- } else {
- LDLM_DEBUG(lock, "lock acquired, setting rss="
- LPU64"; leaving kms="LPU64", end="LPU64,
- loi->loi_lvb.lvb_size, loi->loi_kms,
- lock->l_policy_data.l_extent.end);
- }
- ldlm_lock_allow_match(lock);
- } else if (rc == ELDLM_LOCK_ABORTED && (flags & LDLM_FL_HAS_INTENT)) {
- LASSERT(lock != NULL);
- loi->loi_lvb = *lvb;
- ldlm_lock_allow_match(lock);
- CDEBUG(D_INODE, "glimpsed, setting rss="LPU64"; leaving"
- " kms="LPU64"\n", loi->loi_lvb.lvb_size, loi->loi_kms);
- rc = ELDLM_OK;
- }
+ /* Complete obtaining the lock procedure. */
+ rc = ldlm_cli_enqueue_fini(aa->oa_exp, req, aa->oa_type, 1,
+ aa->oa_mode, aa->oa_flags, lvb, lvb_len,
+ lockh, rc);
+ /* Complete osc stuff. */
+ rc = osc_enqueue_fini(req, aa->oa_upcall, aa->oa_cookie, lockh, mode,
+ aa->oa_flags, aa->oa_agl, rc);
- if (lock != NULL) {
- if (rc != ELDLM_OK)
- ldlm_lock_fail_match(lock);
+ OBD_FAIL_TIMEOUT(OBD_FAIL_OSC_CP_CANCEL_RACE, 10);
- LDLM_LOCK_PUT(lock);
- }
+ ldlm_lock_decref(lockh, mode);
+ LDLM_LOCK_PUT(lock);
+ RETURN(rc);
}
-EXPORT_SYMBOL(osc_update_enqueue);
struct ptlrpc_request_set *PTLRPCD_SET = (void *)1;
* other synchronous requests, however keeping some locks and trying to obtain
* others may take a considerable amount of time in a case of ost failure; and
* when other sync requests do not get released lock from a client, the client
- * is excluded from the cluster -- such scenarious make the life difficult, so
+ * is evicted from the cluster -- such scenarious make the life difficult, so
* 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,
+ osc_enqueue_upcall_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 lustre_handle lockh = { 0 };
struct ptlrpc_request *req = NULL;
int intent = *flags & LDLM_FL_HAS_INTENT;
- __u64 match_lvb = (agl != 0 ? 0 : LDLM_FL_LVB_READY);
+ __u64 match_lvb = agl ? 0 : LDLM_FL_LVB_READY;
ldlm_mode_t mode;
int rc;
ENTRY;
if (einfo->ei_mode == LCK_PR)
mode |= LCK_PW;
mode = ldlm_lock_match(obd->obd_namespace, *flags | match_lvb, res_id,
- einfo->ei_type, policy, mode, lockh, 0);
- if (mode) {
- struct ldlm_lock *matched = ldlm_handle2lock(lockh);
-
- 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. */
- ldlm_lock_decref(lockh, mode);
- LDLM_LOCK_PUT(matched);
- RETURN(-ECANCELED);
- } else if (osc_set_lock_data_with_check(matched, einfo)) {
- *flags |= LDLM_FL_LVB_READY;
- /* addref the lock only if not async requests and PW
- * lock is matched whereas we asked for PR. */
- if (!rqset && einfo->ei_mode != mode)
- ldlm_lock_addref(lockh, LCK_PR);
- if (intent) {
- /* I would like to be able to ASSERT here that
- * rss <= kms, but I can't, for reasons which
- * are explained in lov_enqueue() */
- }
+ einfo->ei_type, policy, mode, &lockh, 0);
+ if (mode) {
+ struct ldlm_lock *matched;
+
+ if (*flags & LDLM_FL_TEST_LOCK)
+ RETURN(ELDLM_OK);
+
+ matched = ldlm_handle2lock(&lockh);
+ if (agl) {
+ /* AGL enqueues DLM locks speculatively. Therefore if
+ * it already exists a DLM lock, it wll just inform the
+ * caller to cancel the AGL process for this stripe. */
+ ldlm_lock_decref(&lockh, mode);
+ LDLM_LOCK_PUT(matched);
+ RETURN(-ECANCELED);
+ } else if (osc_set_lock_data_with_check(matched, einfo)) {
+ *flags |= LDLM_FL_LVB_READY;
+
+ /* We already have a lock, and it's referenced. */
+ (*upcall)(cookie, &lockh, ELDLM_LOCK_MATCHED);
+
+ ldlm_lock_decref(&lockh, mode);
+ LDLM_LOCK_PUT(matched);
+ RETURN(ELDLM_OK);
+ } else {
+ ldlm_lock_decref(&lockh, mode);
+ LDLM_LOCK_PUT(matched);
+ }
+ }
- /* We already have a lock, and it's referenced.
- *
- * At this point, the cl_lock::cll_state is CLS_QUEUING,
- * AGL upcall may change it to CLS_HELD directly. */
- (*upcall)(cookie, ELDLM_OK);
-
- if (einfo->ei_mode != mode)
- ldlm_lock_decref(lockh, LCK_PW);
- else if (rqset)
- /* For async requests, decref the lock. */
- ldlm_lock_decref(lockh, einfo->ei_mode);
- LDLM_LOCK_PUT(matched);
- RETURN(ELDLM_OK);
- } else {
- ldlm_lock_decref(lockh, mode);
- LDLM_LOCK_PUT(matched);
- }
- }
+no_match:
+ if (*flags & LDLM_FL_TEST_LOCK)
+ RETURN(-ENOLCK);
- no_match:
- if (intent) {
- CFS_LIST_HEAD(cancels);
- req = ptlrpc_request_alloc(class_exp2cliimp(exp),
- &RQF_LDLM_ENQUEUE_LVB);
- if (req == NULL)
- RETURN(-ENOMEM);
+ if (intent) {
+ req = ptlrpc_request_alloc(class_exp2cliimp(exp),
+ &RQF_LDLM_ENQUEUE_LVB);
+ if (req == NULL)
+ RETURN(-ENOMEM);
- rc = ldlm_prep_enqueue_req(exp, req, &cancels, 0);
- if (rc) {
+ rc = ptlrpc_request_pack(req, LUSTRE_DLM_VERSION, LDLM_ENQUEUE);
+ if (rc < 0) {
ptlrpc_request_free(req);
RETURN(rc);
}
*flags &= ~LDLM_FL_BLOCK_GRANTED;
rc = ldlm_cli_enqueue(exp, &req, einfo, res_id, policy, flags, lvb,
- sizeof(*lvb), LVB_T_OST, lockh, async);
- if (rqset) {
- if (!rc) {
- struct osc_enqueue_args *aa;
- CLASSERT (sizeof(*aa) <= sizeof(req->rq_async_args));
- aa = ptlrpc_req_async_args(req);
- aa->oa_ei = einfo;
- aa->oa_exp = exp;
- aa->oa_flags = flags;
- aa->oa_upcall = upcall;
- aa->oa_cookie = cookie;
- aa->oa_lvb = lvb;
- aa->oa_lockh = lockh;
- aa->oa_agl = !!agl;
+ sizeof(*lvb), LVB_T_OST, &lockh, async);
+ if (async) {
+ if (!rc) {
+ struct osc_enqueue_args *aa;
+ CLASSERT(sizeof(*aa) <= sizeof(req->rq_async_args));
+ aa = ptlrpc_req_async_args(req);
+ aa->oa_exp = exp;
+ aa->oa_mode = einfo->ei_mode;
+ aa->oa_type = einfo->ei_type;
+ lustre_handle_copy(&aa->oa_lockh, &lockh);
+ aa->oa_upcall = upcall;
+ aa->oa_cookie = cookie;
+ aa->oa_agl = !!agl;
+ if (!agl) {
+ aa->oa_flags = flags;
+ aa->oa_lvb = lvb;
+ } else {
+ /* AGL is essentially to enqueue an DLM lock
+ * in advance, so we don't care about the
+ * result of AGL enqueue. */
+ aa->oa_lvb = NULL;
+ aa->oa_flags = NULL;
+ }
req->rq_interpret_reply =
(ptlrpc_interpterer_t)osc_enqueue_interpret;
RETURN(rc);
}
- rc = osc_enqueue_fini(req, lvb, upcall, cookie, flags, agl, rc);
- if (intent)
- ptlrpc_req_finished(req);
-
- RETURN(rc);
-}
-
-static int osc_enqueue(struct obd_export *exp, struct obd_info *oinfo,
- struct ldlm_enqueue_info *einfo,
- struct ptlrpc_request_set *rqset)
-{
- struct ldlm_res_id res_id;
- int rc;
- ENTRY;
+ rc = osc_enqueue_fini(req, upcall, cookie, &lockh, einfo->ei_mode,
+ flags, agl, rc);
+ if (intent)
+ ptlrpc_req_finished(req);
- 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,
- oinfo->oi_cb_up, oinfo, einfo, oinfo->oi_lockh,
- rqset, rqset != NULL, 0);
- RETURN(rc);
+ RETURN(rc);
}
int osc_match_base(struct obd_export *exp, struct ldlm_res_id *res_id,
RETURN(0);
}
-static int osc_cancel(struct obd_export *exp, struct lov_stripe_md *md,
- __u32 mode, struct lustre_handle *lockh)
-{
- ENTRY;
- RETURN(osc_cancel_base(lockh, mode));
-}
-
-static int osc_cancel_unused(struct obd_export *exp,
- struct lov_stripe_md *lsm,
- ldlm_cancel_flags_t flags,
- void *opaque)
-{
- struct obd_device *obd = class_exp2obd(exp);
- struct ldlm_res_id res_id, *resp = NULL;
-
- 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);
-}
-
static int osc_statfs_interpret(const struct lu_env *env,
struct ptlrpc_request *req,
struct osc_async_args *aa, int rc)
return rc;
}
-/* Retrieve object striping information.
- *
- * @lmmu is a pointer to an in-core struct with lmm_ost_count indicating
- * the maximum number of OST indices which will fit in the user buffer.
- * lmm_magic must be LOV_MAGIC (we only use 1 slot here).
- */
-static int osc_getstripe(struct lov_stripe_md *lsm, struct lov_user_md *lump)
-{
- /* we use lov_user_md_v3 because it is larger than lov_user_md_v1 */
- struct lov_user_md_v3 lum, *lumk;
- struct lov_user_ost_data_v1 *lmm_objects;
- int rc = 0, lum_size;
- ENTRY;
-
- if (!lsm)
- RETURN(-ENODATA);
-
- /* 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 (copy_from_user(&lum, lump, lum_size))
- RETURN(-EFAULT);
-
- if ((lum.lmm_magic != LOV_USER_MAGIC_V1) &&
- (lum.lmm_magic != LOV_USER_MAGIC_V3))
- RETURN(-EINVAL);
-
- /* lov_user_md_vX and lov_mds_md_vX must have the same size */
- LASSERT(sizeof(struct lov_user_md_v1) == sizeof(struct lov_mds_md_v1));
- LASSERT(sizeof(struct lov_user_md_v3) == sizeof(struct lov_mds_md_v3));
- LASSERT(sizeof(lum.lmm_objects[0]) == sizeof(lumk->lmm_objects[0]));
-
- /* we can use lov_mds_md_size() to compute lum_size
- * because lov_user_md_vX and lov_mds_md_vX have the same size */
- if (lum.lmm_stripe_count > 0) {
- lum_size = lov_mds_md_size(lum.lmm_stripe_count, lum.lmm_magic);
- OBD_ALLOC(lumk, lum_size);
- 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_ost_oi = lsm->lsm_oi;
- } else {
- lum_size = lov_mds_md_size(0, lum.lmm_magic);
- lumk = &lum;
- }
-
- lumk->lmm_oi = lsm->lsm_oi;
- lumk->lmm_stripe_count = 1;
-
- if (copy_to_user(lump, lumk, lum_size))
- rc = -EFAULT;
-
- if (lumk != &lum)
- OBD_FREE(lumk, lum_size);
-
- RETURN(rc);
-}
-
-
static int osc_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
void *karg, void *uarg)
{
ENTRY;
if (!try_module_get(THIS_MODULE)) {
- CERROR("Can't get module. Is it alive?");
+ CERROR("%s: cannot get module '%s'\n", obd->obd_name,
+ module_name(THIS_MODULE));
return -EINVAL;
}
switch (cmd) {
- case OBD_IOC_LOV_GET_CONFIG: {
- char *buf;
- struct lov_desc *desc;
- struct obd_uuid uuid;
-
- buf = NULL;
- len = 0;
- if (obd_ioctl_getdata(&buf, &len, (void *)uarg))
- GOTO(out, err = -EINVAL);
-
- data = (struct obd_ioctl_data *)buf;
-
- if (sizeof(*desc) > data->ioc_inllen1) {
- obd_ioctl_freedata(buf, len);
- GOTO(out, err = -EINVAL);
- }
-
- if (data->ioc_inllen2 < sizeof(uuid)) {
- obd_ioctl_freedata(buf, len);
- GOTO(out, err = -EINVAL);
- }
-
- desc = (struct lov_desc *)data->ioc_inlbuf1;
- desc->ld_tgt_count = 1;
- desc->ld_active_tgt_count = 1;
- desc->ld_default_stripe_count = 1;
- desc->ld_default_stripe_size = 0;
- desc->ld_default_stripe_offset = 0;
- desc->ld_pattern = 0;
- memcpy(&desc->ld_uuid, &obd->obd_uuid, sizeof(uuid));
-
- memcpy(data->ioc_inlbuf2, &obd->obd_uuid, sizeof(uuid));
-
- err = copy_to_user((void *)uarg, buf, len);
- if (err)
- err = -EFAULT;
- obd_ioctl_freedata(buf, len);
- GOTO(out, err);
- }
- case LL_IOC_LOV_SETSTRIPE:
- err = obd_alloc_memmd(exp, karg);
- if (err > 0)
- err = 0;
- GOTO(out, err);
- case LL_IOC_LOV_GETSTRIPE:
- err = osc_getstripe(karg, uarg);
- GOTO(out, err);
case OBD_IOC_CLIENT_RECOVER:
err = ptlrpc_recover_import(obd->u.cli.cl_import,
data->ioc_inlbuf1, 0);
if (!vallen || !val)
RETURN(-EFAULT);
- if (KEY_IS(KEY_LOCK_TO_STRIPE)) {
- __u32 *stripe = val;
- *vallen = sizeof(*stripe);
- *stripe = 0;
- RETURN(0);
- } else if (KEY_IS(KEY_LAST_ID)) {
- struct ptlrpc_request *req;
- obd_id *reply;
- char *tmp;
- int rc;
-
- req = ptlrpc_request_alloc(class_exp2cliimp(exp),
- &RQF_OST_GET_INFO_LAST_ID);
- if (req == NULL)
- RETURN(-ENOMEM);
-
- req_capsule_set_size(&req->rq_pill, &RMF_SETINFO_KEY,
- RCL_CLIENT, keylen);
- rc = ptlrpc_request_pack(req, LUSTRE_OST_VERSION, OST_GET_INFO);
- if (rc) {
- ptlrpc_request_free(req);
- RETURN(rc);
- }
-
- tmp = req_capsule_client_get(&req->rq_pill, &RMF_SETINFO_KEY);
- memcpy(tmp, key, keylen);
-
- req->rq_no_delay = req->rq_no_resend = 1;
- ptlrpc_request_set_replen(req);
- rc = ptlrpc_queue_wait(req);
- if (rc)
- GOTO(out, rc);
-
- reply = req_capsule_server_get(&req->rq_pill, &RMF_OBD_ID);
- if (reply == NULL)
- GOTO(out, rc = -EPROTO);
-
- *((obd_id *)val) = *reply;
- out:
- ptlrpc_req_finished(req);
- RETURN(rc);
- } else if (KEY_IS(KEY_FIEMAP)) {
+ if (KEY_IS(KEY_FIEMAP)) {
struct ll_fiemap_info_key *fm_key =
(struct ll_fiemap_info_key *)key;
struct ldlm_res_id res_id;
LASSERT(cli->cl_cache == NULL); /* only once */
cli->cl_cache = (struct cl_client_cache *)val;
- cfs_atomic_inc(&cli->cl_cache->ccc_users);
+ atomic_inc(&cli->cl_cache->ccc_users);
cli->cl_lru_left = &cli->cl_cache->ccc_lru_left;
/* add this osc into entity list */
- LASSERT(cfs_list_empty(&cli->cl_lru_osc));
+ LASSERT(list_empty(&cli->cl_lru_osc));
spin_lock(&cli->cl_cache->ccc_lru_lock);
- cfs_list_add(&cli->cl_lru_osc, &cli->cl_cache->ccc_lru);
+ list_add(&cli->cl_lru_osc, &cli->cl_cache->ccc_lru);
spin_unlock(&cli->cl_cache->ccc_lru_lock);
RETURN(0);
if (KEY_IS(KEY_CACHE_LRU_SHRINK)) {
struct client_obd *cli = &obd->u.cli;
- int nr = cfs_atomic_read(&cli->cl_lru_in_list) >> 1;
- int target = *(int *)val;
+ long nr = atomic_long_read(&cli->cl_lru_in_list) >> 1;
+ long target = *(long *)val;
nr = osc_lru_shrink(env, cli, min(nr, target), true);
- *(int *)val -= nr;
+ *(long *)val -= nr;
RETURN(0);
}
RETURN(0);
}
-
-static int osc_llog_init(struct obd_device *obd, struct obd_llog_group *olg,
- struct obd_device *disk_obd, int *index)
-{
- /* this code is not supposed to be used with LOD/OSP
- * to be removed soon */
- LBUG();
- return 0;
-}
-
-static int osc_llog_finish(struct obd_device *obd, int count)
-{
- struct llog_ctxt *ctxt;
-
- ENTRY;
-
- ctxt = llog_get_context(obd, LLOG_MDS_OST_ORIG_CTXT);
- if (ctxt) {
- llog_cat_close(NULL, ctxt->loc_handle);
- llog_cleanup(NULL, ctxt);
- }
-
- ctxt = llog_get_context(obd, LLOG_SIZE_REPL_CTXT);
- if (ctxt)
- llog_cleanup(NULL, ctxt);
- RETURN(0);
-}
-
static int osc_reconnect(const struct lu_env *env,
struct obd_export *exp, struct obd_device *obd,
struct obd_uuid *cluuid,
if (data != NULL && (data->ocd_connect_flags & OBD_CONNECT_GRANT)) {
long lost_grant;
- client_obd_list_lock(&cli->cl_loi_list_lock);
- data->ocd_grant = (cli->cl_avail_grant + cli->cl_dirty) ?:
- 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);
+ spin_lock(&cli->cl_loi_list_lock);
+ data->ocd_grant = (cli->cl_avail_grant +
+ (cli->cl_dirty_pages << PAGE_CACHE_SHIFT)) ?:
+ 2 * cli_brw_size(obd);
+ lost_grant = cli->cl_lost_grant;
+ cli->cl_lost_grant = 0;
+ spin_unlock(&cli->cl_loi_list_lock);
CDEBUG(D_RPCTRACE, "ocd_connect_flags: "LPX64" ocd_version: %d"
" ocd_grant: %d, lost: %ld.\n", data->ocd_connect_flags,
static int osc_disconnect(struct obd_export *exp)
{
- struct obd_device *obd = class_exp2obd(exp);
- struct llog_ctxt *ctxt;
- int rc;
-
- ctxt = llog_get_context(obd, LLOG_SIZE_REPL_CTXT);
- if (ctxt) {
- if (obd->u.cli.cl_conn_count == 1) {
- /* Flush any remaining cancel messages out to the
- * target */
- llog_sync(ctxt, exp, 0);
- }
- llog_ctxt_put(ctxt);
- } else {
- CDEBUG(D_HA, "No LLOG_SIZE_REPL_CTXT found in obd %p\n",
- obd);
- }
+ struct obd_device *obd = class_exp2obd(exp);
+ int rc;
rc = client_disconnect_export(exp);
/**
switch (event) {
case IMP_EVENT_DISCON: {
cli = &obd->u.cli;
- client_obd_list_lock(&cli->cl_loi_list_lock);
- cli->cl_avail_grant = 0;
- cli->cl_lost_grant = 0;
- client_obd_list_unlock(&cli->cl_loi_list_lock);
+ spin_lock(&cli->cl_loi_list_lock);
+ cli->cl_avail_grant = 0;
+ cli->cl_lost_grant = 0;
+ spin_unlock(&cli->cl_loi_list_lock);
break;
}
case IMP_EVENT_INACTIVE: {
* \retval zero the lock can't be canceled
* \retval other ok to cancel
*/
-static int osc_cancel_for_recovery(struct ldlm_lock *lock)
+static int osc_cancel_weight(struct ldlm_lock *lock)
{
- check_res_locked(lock->l_resource);
-
- /*
- * Cancel all unused extent lock in granted mode LCK_PR or LCK_CR.
- *
- * XXX as a future improvement, we can also cancel unused write lock
- * if it doesn't have dirty data and active mmaps.
- */
- if (lock->l_resource->lr_type == LDLM_EXTENT &&
- (lock->l_granted_mode == LCK_PR ||
- lock->l_granted_mode == LCK_CR) &&
- (osc_dlm_lock_pageref(lock) == 0))
- RETURN(1);
+ /*
+ * Cancel all unused and granted extent lock.
+ */
+ if (lock->l_resource->lr_type == LDLM_EXTENT &&
+ lock->l_granted_mode == lock->l_req_mode &&
+ osc_ldlm_weigh_ast(lock) == 0)
+ RETURN(1);
- RETURN(0);
+ RETURN(0);
}
static int brw_queue_work(const struct lu_env *env, void *data)
int osc_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
{
- struct lprocfs_static_vars lvars = { 0 };
- struct client_obd *cli = &obd->u.cli;
- void *handler;
- int rc;
+ struct client_obd *cli = &obd->u.cli;
+ struct obd_type *type;
+ void *handler;
+ int rc;
ENTRY;
rc = ptlrpcd_addref();
GOTO(out_ptlrpcd_work, rc);
cli->cl_grant_shrink_interval = GRANT_SHRINK_INTERVAL;
- lprocfs_osc_init_vars(&lvars);
- if (lprocfs_obd_setup(obd, lvars.obd_vars) == 0) {
+
+#ifdef LPROCFS
+ obd->obd_vars = lprocfs_osc_obd_vars;
+#endif
+ /* If this is true then both client (osc) and server (osp) are on the
+ * same node. The osp layer if loaded first will register the osc proc
+ * directory. In that case this obd_device will be attached its proc
+ * tree to type->typ_procsym instead of obd->obd_type->typ_procroot. */
+ type = class_search_type(LUSTRE_OSP_NAME);
+ if (type && type->typ_procsym) {
+ obd->obd_proc_entry = lprocfs_seq_register(obd->obd_name,
+ type->typ_procsym,
+ obd->obd_vars, obd);
+ if (IS_ERR(obd->obd_proc_entry)) {
+ rc = PTR_ERR(obd->obd_proc_entry);
+ CERROR("error %d setting up lprocfs for %s\n", rc,
+ obd->obd_name);
+ obd->obd_proc_entry = NULL;
+ }
+ } else {
+ rc = lprocfs_obd_setup(obd);
+ }
+
+ /* If the basic OSC proc tree construction succeeded then
+ * lets do the rest. */
+ if (rc == 0) {
lproc_osc_attach_seqstat(obd);
sptlrpc_lprocfs_cliobd_attach(obd);
ptlrpc_lprocfs_register_obd(obd);
OST_MAXREQSIZE,
ptlrpc_add_rqs_to_pool);
- CFS_INIT_LIST_HEAD(&cli->cl_grant_shrink_list);
- ns_register_cancel(obd->obd_namespace, osc_cancel_for_recovery);
- RETURN(rc);
+ INIT_LIST_HEAD(&cli->cl_grant_shrink_list);
+ ns_register_cancel(obd->obd_namespace, osc_cancel_weight);
+ RETURN(0);
out_ptlrpcd_work:
if (cli->cl_writeback_work != NULL) {
obd_cleanup_client_import(obd);
ptlrpc_lprocfs_unregister_obd(obd);
lprocfs_obd_cleanup(obd);
- rc = obd_llog_finish(obd, 0);
- if (rc != 0)
- CERROR("failed to cleanup llogging subsystems\n");
break;
}
}
/* lru cleanup */
if (cli->cl_cache != NULL) {
- LASSERT(cfs_atomic_read(&cli->cl_cache->ccc_users) > 0);
+ LASSERT(atomic_read(&cli->cl_cache->ccc_users) > 0);
spin_lock(&cli->cl_cache->ccc_lru_lock);
- cfs_list_del_init(&cli->cl_lru_osc);
+ list_del_init(&cli->cl_lru_osc);
spin_unlock(&cli->cl_cache->ccc_lru_lock);
cli->cl_lru_left = NULL;
- cfs_atomic_dec(&cli->cl_cache->ccc_users);
+ atomic_dec(&cli->cl_cache->ccc_users);
cli->cl_cache = NULL;
}
int osc_process_config_base(struct obd_device *obd, struct lustre_cfg *lcfg)
{
- struct lprocfs_static_vars lvars = { 0 };
- int rc = 0;
-
- lprocfs_osc_init_vars(&lvars);
-
- switch (lcfg->lcfg_command) {
- default:
- rc = class_process_proc_param(PARAM_OSC, lvars.obd_vars,
- lcfg, obd);
- if (rc > 0)
- rc = 0;
- break;
- }
-
- return(rc);
+ int rc = class_process_proc_param(PARAM_OSC, obd->obd_vars, lcfg, obd);
+ return rc > 0 ? 0: rc;
}
-static int osc_process_config(struct obd_device *obd, obd_count len, void *buf)
+static int osc_process_config(struct obd_device *obd, size_t len, void *buf)
{
return osc_process_config_base(obd, buf);
}
-struct obd_ops osc_obd_ops = {
+static struct obd_ops osc_obd_ops = {
.o_owner = THIS_MODULE,
.o_setup = osc_setup,
.o_precleanup = osc_precleanup,
.o_disconnect = osc_disconnect,
.o_statfs = osc_statfs,
.o_statfs_async = osc_statfs_async,
- .o_packmd = osc_packmd,
- .o_unpackmd = osc_unpackmd,
.o_create = osc_create,
.o_destroy = osc_destroy,
.o_getattr = osc_getattr,
.o_getattr_async = osc_getattr_async,
.o_setattr = osc_setattr,
.o_setattr_async = osc_setattr_async,
- .o_brw = osc_brw,
- .o_punch = osc_punch,
- .o_sync = osc_sync,
- .o_enqueue = osc_enqueue,
.o_change_cbdata = osc_change_cbdata,
.o_find_cbdata = osc_find_cbdata,
- .o_cancel = osc_cancel,
- .o_cancel_unused = osc_cancel_unused,
.o_iocontrol = osc_iocontrol,
.o_get_info = osc_get_info,
.o_set_info_async = osc_set_info_async,
.o_import_event = osc_import_event,
- .o_llog_init = osc_llog_init,
- .o_llog_finish = osc_llog_finish,
.o_process_config = osc_process_config,
.o_quotactl = osc_quotactl,
.o_quotacheck = osc_quotacheck,
};
-extern struct lu_kmem_descr osc_caches[];
-extern spinlock_t osc_ast_guard;
-extern struct lock_class_key osc_ast_guard_class;
-
-int __init osc_init(void)
+static int __init osc_init(void)
{
- struct lprocfs_static_vars lvars = { 0 };
- int rc;
- ENTRY;
+ bool enable_proc = true;
+ struct obd_type *type;
+ int rc;
+ ENTRY;
/* print an address of _any_ initialized kernel symbol from this
* module, to allow debugging with gdb that doesn't support data
if (rc)
RETURN(rc);
- lprocfs_osc_init_vars(&lvars);
+ type = class_search_type(LUSTRE_OSP_NAME);
+ if (type != NULL && type->typ_procsym != NULL)
+ enable_proc = false;
- rc = class_register_type(&osc_obd_ops, NULL, NULL,
-#ifndef HAVE_ONLY_PROCFS_SEQ
- lvars.module_vars,
-#endif
- LUSTRE_OSC_NAME, &osc_device_type);
+ rc = class_register_type(&osc_obd_ops, NULL, enable_proc, NULL,
+ LUSTRE_OSC_NAME, &osc_device_type);
if (rc) {
lu_kmem_fini(osc_caches);
RETURN(rc);
}
- spin_lock_init(&osc_ast_guard);
- lockdep_set_class(&osc_ast_guard, &osc_ast_guard_class);
-
RETURN(rc);
}
-#ifdef __KERNEL__
static void /*__exit*/ osc_exit(void)
{
class_unregister_type(LUSTRE_OSC_NAME);
MODULE_LICENSE("GPL");
cfs_module(osc, LUSTRE_VERSION_STRING, osc_init, osc_exit);
-#endif