*
* You should have received a copy of the GNU General Public License
* version 2 along with this program; If not, see
- * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
- *
- * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
- * CA 95054 USA or visit www.sun.com if you need additional information or
- * have any questions.
+ * http://www.gnu.org/licenses/gpl-2.0.html
*
* GPL HEADER END
*/
* Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
* Use is subject to license terms.
*
- * Copyright (c) 2011, 2013, Intel Corporation.
+ * Copyright (c) 2011, 2017, Intel Corporation.
*/
/*
* This file is part of Lustre, http://www.lustre.org/
return buf;
}
-void mdt_trans_stop(const struct lu_env *env,
- struct mdt_device *mdt, struct thandle *th)
-{
- dt_trans_stop(env, mdt->mdt_bottom, th);
-}
-
/*
* last_rcvd & last_committed update callbacks
*/
/* prepare transactions callbacks */
mdt->mdt_txn_cb.dtc_txn_start = NULL;
mdt->mdt_txn_cb.dtc_txn_stop = mdt_txn_stop_cb;
- mdt->mdt_txn_cb.dtc_txn_commit = NULL;
mdt->mdt_txn_cb.dtc_cookie = NULL;
mdt->mdt_txn_cb.dtc_tag = LCT_MD_THREAD;
- CFS_INIT_LIST_HEAD(&mdt->mdt_txn_cb.dtc_linkage);
+ INIT_LIST_HEAD(&mdt->mdt_txn_cb.dtc_linkage);
dt_txn_callback_add(mdt->mdt_bottom, &mdt->mdt_txn_cb);
/* Remove transaction callback */
dt_txn_callback_del(mdt->mdt_bottom, &mdt->mdt_txn_cb);
- if (mdt->mdt_ck_obj)
- lu_object_put(env, &mdt->mdt_ck_obj->do_lu);
- mdt->mdt_ck_obj = NULL;
EXIT;
}
static void mdt_steal_ack_locks(struct ptlrpc_request *req)
{
struct ptlrpc_service_part *svcpt;
- struct obd_export *exp = req->rq_export;
- cfs_list_t *tmp;
- struct ptlrpc_reply_state *oldrep;
- int i;
+ struct obd_export *exp = req->rq_export;
+ struct list_head *tmp;
+ struct ptlrpc_reply_state *rs;
+ int i;
- /* CAVEAT EMPTOR: spinlock order */
+ /* CAVEAT EMPTOR: spinlock order */
spin_lock(&exp->exp_lock);
- cfs_list_for_each (tmp, &exp->exp_outstanding_replies) {
- oldrep = cfs_list_entry(tmp, struct ptlrpc_reply_state,
- rs_exp_list);
-
- if (oldrep->rs_xid != req->rq_xid)
- continue;
+ list_for_each(tmp, &exp->exp_outstanding_replies) {
+ rs = list_entry(tmp, struct ptlrpc_reply_state,
+ rs_exp_list);
- if (oldrep->rs_opc != lustre_msg_get_opc(req->rq_reqmsg))
- CERROR ("Resent req xid "LPU64" has mismatched opc: "
- "new %d old %d\n", req->rq_xid,
- lustre_msg_get_opc(req->rq_reqmsg),
- oldrep->rs_opc);
+ if (rs->rs_xid != req->rq_xid)
+ continue;
- svcpt = oldrep->rs_svcpt;
- spin_lock(&svcpt->scp_rep_lock);
+ if (rs->rs_opc != lustre_msg_get_opc(req->rq_reqmsg))
+ CERROR("%s: Resent req xid %llu has mismatched opc: "
+ "new %d old %d\n", exp->exp_obd->obd_name,
+ req->rq_xid, lustre_msg_get_opc(req->rq_reqmsg),
+ rs->rs_opc);
- cfs_list_del_init (&oldrep->rs_exp_list);
+ svcpt = rs->rs_svcpt;
- CDEBUG(D_HA, "Stealing %d locks from rs %p x"LPD64".t"LPD64
+ CDEBUG(D_HA, "Stealing %d locks from rs %p x%lld.t%lld"
" o%d NID %s\n",
- oldrep->rs_nlocks, oldrep,
- oldrep->rs_xid, oldrep->rs_transno, oldrep->rs_opc,
- libcfs_nid2str(exp->exp_connection->c_peer.nid));
+ rs->rs_nlocks, rs,
+ rs->rs_xid, rs->rs_transno, rs->rs_opc,
+ obd_export_nid2str(exp));
+
+ spin_lock(&svcpt->scp_rep_lock);
+ list_del_init(&rs->rs_exp_list);
- for (i = 0; i < oldrep->rs_nlocks; i++)
- ptlrpc_save_lock(req, &oldrep->rs_locks[i],
- oldrep->rs_modes[i], 0);
- oldrep->rs_nlocks = 0;
+ spin_lock(&rs->rs_lock);
+ for (i = 0; i < rs->rs_nlocks; i++)
+ ptlrpc_save_lock(req, &rs->rs_locks[i],
+ rs->rs_modes[i], rs->rs_no_ack,
+ rs->rs_convert_lock);
+ rs->rs_nlocks = 0;
- DEBUG_REQ(D_HA, req, "stole locks for");
- spin_lock(&oldrep->rs_lock);
- ptlrpc_schedule_difficult_reply(oldrep);
- spin_unlock(&oldrep->rs_lock);
+ DEBUG_REQ(D_HA, req, "stole locks for");
+ ptlrpc_schedule_difficult_reply(rs);
+ spin_unlock(&rs->rs_lock);
spin_unlock(&svcpt->scp_rep_lock);
break;
}
spin_unlock(&exp->exp_lock);
-}
-/**
- * VBR: restore versions
- */
-void mdt_vbr_reconstruct(struct ptlrpc_request *req,
- struct lsd_client_data *lcd)
-{
- __u64 pre_versions[4] = {0};
- pre_versions[0] = lcd->lcd_pre_versions[0];
- pre_versions[1] = lcd->lcd_pre_versions[1];
- pre_versions[2] = lcd->lcd_pre_versions[2];
- pre_versions[3] = lcd->lcd_pre_versions[3];
- lustre_msg_set_versions(req->rq_repmsg, pre_versions);
+ /* if exp_disconnected, decref stolen locks */
+ if (exp->exp_disconnected) {
+ rs = req->rq_reply_state;
+
+ for (i = 0; i < rs->rs_nlocks; i++)
+ ldlm_lock_decref(&rs->rs_locks[i], rs->rs_modes[i]);
+
+ rs->rs_nlocks = 0;
+ }
}
-void mdt_req_from_lcd(struct ptlrpc_request *req,
- struct lsd_client_data *lcd)
+__u64 mdt_req_from_lrd(struct ptlrpc_request *req,
+ struct tg_reply_data *trd)
{
- DEBUG_REQ(D_HA, req, "restoring transno "LPD64"/status %d",
- lcd->lcd_last_transno, lcd->lcd_last_result);
-
- if (lustre_msg_get_opc(req->rq_reqmsg) == MDS_CLOSE ||
- lustre_msg_get_opc(req->rq_repmsg) == MDS_DONE_WRITING) {
- req->rq_transno = lcd->lcd_last_close_transno;
- req->rq_status = lcd->lcd_last_close_result;
- } else {
- req->rq_transno = lcd->lcd_last_transno;
- req->rq_status = lcd->lcd_last_result;
- mdt_vbr_reconstruct(req, lcd);
- }
- if (req->rq_status != 0)
- req->rq_transno = 0;
- lustre_msg_set_transno(req->rq_repmsg, req->rq_transno);
- lustre_msg_set_status(req->rq_repmsg, req->rq_status);
- DEBUG_REQ(D_RPCTRACE, req, "restoring transno "LPD64"/status %d",
- req->rq_transno, req->rq_status);
-
- mdt_steal_ack_locks(req);
+ struct lsd_reply_data *lrd;
+
+ LASSERT(trd != NULL);
+ lrd = &trd->trd_reply;
+
+ DEBUG_REQ(D_HA, req, "restoring transno %lld/status %d",
+ lrd->lrd_transno, lrd->lrd_result);
+
+ req->rq_transno = lrd->lrd_transno;
+ req->rq_status = lrd->lrd_result;
+
+ lustre_msg_set_versions(req->rq_repmsg, trd->trd_pre_versions);
+
+ if (req->rq_status != 0)
+ req->rq_transno = 0;
+ lustre_msg_set_transno(req->rq_repmsg, req->rq_transno);
+ lustre_msg_set_status(req->rq_repmsg, req->rq_status);
+
+ DEBUG_REQ(D_RPCTRACE, req, "restoring transno %lld/status %d",
+ req->rq_transno, req->rq_status);
+
+ mdt_steal_ack_locks(req);
+
+ return lrd->lrd_data;
}
+
void mdt_reconstruct_generic(struct mdt_thread_info *mti,
- struct mdt_lock_handle *lhc)
+ struct mdt_lock_handle *lhc)
{
- struct ptlrpc_request *req = mdt_info_req(mti);
- struct tg_export_data *ted = &req->rq_export->exp_target_data;
+ struct ptlrpc_request *req = mdt_info_req(mti);
- return mdt_req_from_lcd(req, ted->ted_lcd);
+ mdt_req_from_lrd(req, mti->mti_reply_data);
+ return;
+}
+
+/**
+ * Generate fake attributes for a non-existing object
+ *
+ * While the client was waiting for the reply, the original transaction
+ * got committed and corresponding rep-ack lock got released, then another
+ * client was able to destroy the object. But we still need to send some
+ * attributes back. So we fake them and set nlink=0, so the client will
+ * be able to detect a non-existing object and drop it from the cache
+ * immediately.
+ *
+ * \param[out] ma attributes to fill
+ */
+static void mdt_fake_ma(struct md_attr *ma)
+{
+ ma->ma_valid = MA_INODE;
+ memset(&ma->ma_attr, 0, sizeof(ma->ma_attr));
+ ma->ma_attr.la_valid = LA_NLINK;
+ ma->ma_attr.la_mode = S_IFREG;
}
static void mdt_reconstruct_create(struct mdt_thread_info *mti,
struct mdt_lock_handle *lhc)
{
- struct ptlrpc_request *req = mdt_info_req(mti);
- struct obd_export *exp = req->rq_export;
- struct tg_export_data *ted = &exp->exp_target_data;
- struct mdt_device *mdt = mti->mti_mdt;
- struct mdt_object *child;
- struct mdt_body *body;
- int rc;
+ struct ptlrpc_request *req = mdt_info_req(mti);
+ struct obd_export *exp = req->rq_export;
+ struct mdt_device *mdt = mti->mti_mdt;
+ struct mdt_object *child;
+ struct mdt_body *body;
+ int rc;
- mdt_req_from_lcd(req, ted->ted_lcd);
- if (req->rq_status)
- return;
+ mdt_req_from_lrd(req, mti->mti_reply_data);
+ if (req->rq_status)
+ return;
/* if no error, so child was created with requested fid */
child = mdt_object_find(mti->mti_env, mdt, mti->mti_rr.rr_fid2);
mti->mti_attr.ma_need = MA_INODE;
mti->mti_attr.ma_valid = 0;
rc = mdt_attr_get_complex(mti, child, &mti->mti_attr);
- if (rc == -EREMOTE) {
+ if (rc == -ENOENT) {
+ mdt_fake_ma(&mti->mti_attr);
+ } else if (rc == -EREMOTE) {
/* object was created on remote server */
if (!mdt_is_dne_client(exp))
/* Return -EIO for old client */
{
struct ptlrpc_request *req = mdt_info_req(mti);
struct obd_export *exp = req->rq_export;
- struct mdt_export_data *med = &exp->exp_mdt_data;
struct mdt_device *mdt = mti->mti_mdt;
struct mdt_object *obj;
struct mdt_body *body;
int rc;
- mdt_req_from_lcd(req, med->med_ted.ted_lcd);
- if (req->rq_status)
- return;
+ mdt_req_from_lrd(req, mti->mti_reply_data);
+ if (req->rq_status)
+ return;
body = req_capsule_server_get(mti->mti_pill, &RMF_MDT_BODY);
obj = mdt_object_find(mti->mti_env, mdt, mti->mti_rr.rr_fid1);
mti->mti_attr.ma_need = MA_INODE;
mti->mti_attr.ma_valid = 0;
- mdt_attr_get_complex(mti, obj, &mti->mti_attr);
+
+ rc = mdt_attr_get_complex(mti, obj, &mti->mti_attr);
+ if (rc == -ENOENT)
+ mdt_fake_ma(&mti->mti_attr);
mdt_pack_attr2body(mti, body, &mti->mti_attr.ma_attr,
mdt_object_fid(obj));
- if (mti->mti_ioepoch && (mti->mti_ioepoch->flags & MF_EPOCH_OPEN)) {
- struct mdt_file_data *mfd;
- struct mdt_body *repbody;
-
- repbody = req_capsule_server_get(mti->mti_pill, &RMF_MDT_BODY);
- repbody->mbo_ioepoch = obj->mot_ioepoch;
- spin_lock(&med->med_open_lock);
- cfs_list_for_each_entry(mfd, &med->med_open_head, mfd_list) {
- if (mfd->mfd_xid == req->rq_xid)
- break;
- }
- LASSERT(&mfd->mfd_list != &med->med_open_head);
- spin_unlock(&med->med_open_lock);
- repbody->mbo_handle.cookie = mfd->mfd_handle.h_cookie;
- }
mdt_object_put(mti->mti_env, obj);
}
[REINT_UNLINK] = mdt_reconstruct_generic,
[REINT_RENAME] = mdt_reconstruct_generic,
[REINT_OPEN] = mdt_reconstruct_open,
- [REINT_SETXATTR] = mdt_reconstruct_generic
+ [REINT_SETXATTR] = mdt_reconstruct_generic,
+ [REINT_RMENTRY] = mdt_reconstruct_generic,
+ [REINT_MIGRATE] = mdt_reconstruct_generic
};
void mdt_reconstruct(struct mdt_thread_info *mti,
struct mdt_lock_handle *lhc)
{
+ mdt_reconstructor reconst;
ENTRY;
- reconstructors[mti->mti_rr.rr_opcode](mti, lhc);
+ LASSERT(mti->mti_rr.rr_opcode < REINT_MAX &&
+ (reconst = reconstructors[mti->mti_rr.rr_opcode]) != NULL);
+ reconst(mti, lhc);
EXIT;
}
git://git.whamcloud.com / fs / lustre-release.git / blobdiff