4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2011, 2014, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
32 * lustre/mdt/mdt_recovery.c
34 * Lustre Metadata Target (mdt) recovery-related methods
36 * Author: Huang Hua <huanghua@clusterfs.com>
37 * Author: Pershin Mike <tappro@clusterfs.com>
40 #define DEBUG_SUBSYSTEM S_MDS
42 #include "mdt_internal.h"
44 struct lu_buf *mdt_buf(const struct lu_env *env, void *area, ssize_t len)
47 struct mdt_thread_info *mti;
49 mti = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
56 const struct lu_buf *mdt_buf_const(const struct lu_env *env,
57 const void *area, ssize_t len)
60 struct mdt_thread_info *mti;
62 mti = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
65 buf->lb_buf = (void *)area;
71 * last_rcvd & last_committed update callbacks
73 extern struct lu_context_key mdt_thread_key;
75 /* This callback notifies MDT that transaction was done. This is needed by
76 * mdt_save_lock() only. It is similar to new target code and will be removed
77 * as mdt_save_lock() will be converted to use target structures */
78 static int mdt_txn_stop_cb(const struct lu_env *env,
79 struct thandle *txn, void *cookie)
81 struct mdt_thread_info *mti;
83 mti = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
86 if (mti->mti_has_trans)
87 CDEBUG(D_INFO, "More than one transaction\n");
89 mti->mti_has_trans = 1;
93 int mdt_fs_setup(const struct lu_env *env, struct mdt_device *mdt,
94 struct obd_device *obd, struct lustre_sb_info *lsi)
100 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_FS_SETUP))
103 /* prepare transactions callbacks */
104 mdt->mdt_txn_cb.dtc_txn_start = NULL;
105 mdt->mdt_txn_cb.dtc_txn_stop = mdt_txn_stop_cb;
106 mdt->mdt_txn_cb.dtc_txn_commit = NULL;
107 mdt->mdt_txn_cb.dtc_cookie = NULL;
108 mdt->mdt_txn_cb.dtc_tag = LCT_MD_THREAD;
109 INIT_LIST_HEAD(&mdt->mdt_txn_cb.dtc_linkage);
111 dt_txn_callback_add(mdt->mdt_bottom, &mdt->mdt_txn_cb);
116 void mdt_fs_cleanup(const struct lu_env *env, struct mdt_device *mdt)
120 /* Remove transaction callback */
121 dt_txn_callback_del(mdt->mdt_bottom, &mdt->mdt_txn_cb);
125 /* reconstruction code */
126 static void mdt_steal_ack_locks(struct ptlrpc_request *req)
128 struct ptlrpc_service_part *svcpt;
129 struct obd_export *exp = req->rq_export;
130 struct list_head *tmp;
131 struct ptlrpc_reply_state *oldrep;
134 /* CAVEAT EMPTOR: spinlock order */
135 spin_lock(&exp->exp_lock);
136 list_for_each(tmp, &exp->exp_outstanding_replies) {
137 oldrep = list_entry(tmp, struct ptlrpc_reply_state,
140 if (oldrep->rs_xid != req->rq_xid)
143 if (oldrep->rs_opc != lustre_msg_get_opc(req->rq_reqmsg))
144 CERROR("%s: Resent req xid %llu has mismatched opc: "
145 "new %d old %d\n", exp->exp_obd->obd_name,
146 req->rq_xid, lustre_msg_get_opc(req->rq_reqmsg),
149 svcpt = oldrep->rs_svcpt;
150 spin_lock(&svcpt->scp_rep_lock);
152 list_del_init(&oldrep->rs_exp_list);
154 CDEBUG(D_HA, "Stealing %d locks from rs %p x%lld.t%lld"
156 oldrep->rs_nlocks, oldrep,
157 oldrep->rs_xid, oldrep->rs_transno, oldrep->rs_opc,
158 libcfs_nid2str(exp->exp_connection->c_peer.nid));
160 for (i = 0; i < oldrep->rs_nlocks; i++)
161 ptlrpc_save_lock(req, &oldrep->rs_locks[i],
162 oldrep->rs_modes[i], oldrep->rs_no_ack,
163 oldrep->rs_convert_lock);
164 oldrep->rs_nlocks = 0;
166 DEBUG_REQ(D_HA, req, "stole locks for");
167 spin_lock(&oldrep->rs_lock);
168 ptlrpc_schedule_difficult_reply(oldrep);
169 spin_unlock(&oldrep->rs_lock);
171 spin_unlock(&svcpt->scp_rep_lock);
174 spin_unlock(&exp->exp_lock);
177 __u64 mdt_req_from_lrd(struct ptlrpc_request *req,
178 struct tg_reply_data *trd)
180 struct lsd_reply_data *lrd;
182 LASSERT(trd != NULL);
183 lrd = &trd->trd_reply;
185 DEBUG_REQ(D_HA, req, "restoring transno %lld/status %d",
186 lrd->lrd_transno, lrd->lrd_result);
188 req->rq_transno = lrd->lrd_transno;
189 req->rq_status = lrd->lrd_result;
191 lustre_msg_set_versions(req->rq_repmsg, trd->trd_pre_versions);
193 if (req->rq_status != 0)
195 lustre_msg_set_transno(req->rq_repmsg, req->rq_transno);
196 lustre_msg_set_status(req->rq_repmsg, req->rq_status);
198 DEBUG_REQ(D_RPCTRACE, req, "restoring transno %lld/status %d",
199 req->rq_transno, req->rq_status);
201 mdt_steal_ack_locks(req);
203 return lrd->lrd_data;
207 void mdt_reconstruct_generic(struct mdt_thread_info *mti,
208 struct mdt_lock_handle *lhc)
210 struct ptlrpc_request *req = mdt_info_req(mti);
212 mdt_req_from_lrd(req, mti->mti_reply_data);
217 * Generate fake attributes for a non-existing object
219 * While the client was waiting for the reply, the original transaction
220 * got committed and corresponding rep-ack lock got released, then another
221 * client was able to destroy the object. But we still need to send some
222 * attributes back. So we fake them and set nlink=0, so the client will
223 * be able to detect a non-existing object and drop it from the cache
226 * \param[out] ma attributes to fill
228 static void mdt_fake_ma(struct md_attr *ma)
230 ma->ma_valid = MA_INODE;
231 memset(&ma->ma_attr, 0, sizeof(ma->ma_attr));
232 ma->ma_attr.la_valid = LA_NLINK;
233 ma->ma_attr.la_mode = S_IFREG;
236 static void mdt_reconstruct_create(struct mdt_thread_info *mti,
237 struct mdt_lock_handle *lhc)
239 struct ptlrpc_request *req = mdt_info_req(mti);
240 struct obd_export *exp = req->rq_export;
241 struct mdt_device *mdt = mti->mti_mdt;
242 struct mdt_object *child;
243 struct mdt_body *body;
246 mdt_req_from_lrd(req, mti->mti_reply_data);
250 /* if no error, so child was created with requested fid */
251 child = mdt_object_find(mti->mti_env, mdt, mti->mti_rr.rr_fid2);
254 LCONSOLE_WARN("cannot lookup child "DFID": rc = %d; "
255 "evicting client %s with export %s\n",
256 PFID(mti->mti_rr.rr_fid2), rc,
257 obd_uuid2str(&exp->exp_client_uuid),
258 obd_export_nid2str(exp));
259 mdt_export_evict(exp);
263 body = req_capsule_server_get(mti->mti_pill, &RMF_MDT_BODY);
264 mti->mti_attr.ma_need = MA_INODE;
265 mti->mti_attr.ma_valid = 0;
266 rc = mdt_attr_get_complex(mti, child, &mti->mti_attr);
268 mdt_fake_ma(&mti->mti_attr);
269 } else if (rc == -EREMOTE) {
270 /* object was created on remote server */
271 if (!mdt_is_dne_client(exp))
272 /* Return -EIO for old client */
276 body->mbo_valid |= OBD_MD_MDS;
278 mdt_pack_attr2body(mti, body, &mti->mti_attr.ma_attr,
279 mdt_object_fid(child));
280 mdt_object_put(mti->mti_env, child);
283 static void mdt_reconstruct_setattr(struct mdt_thread_info *mti,
284 struct mdt_lock_handle *lhc)
286 struct ptlrpc_request *req = mdt_info_req(mti);
287 struct obd_export *exp = req->rq_export;
288 struct mdt_device *mdt = mti->mti_mdt;
289 struct mdt_object *obj;
290 struct mdt_body *body;
293 mdt_req_from_lrd(req, mti->mti_reply_data);
297 body = req_capsule_server_get(mti->mti_pill, &RMF_MDT_BODY);
298 obj = mdt_object_find(mti->mti_env, mdt, mti->mti_rr.rr_fid1);
301 LCONSOLE_WARN("cannot lookup "DFID": rc = %d; "
302 "evicting client %s with export %s\n",
303 PFID(mti->mti_rr.rr_fid1), rc,
304 obd_uuid2str(&exp->exp_client_uuid),
305 obd_export_nid2str(exp));
306 mdt_export_evict(exp);
310 mti->mti_attr.ma_need = MA_INODE;
311 mti->mti_attr.ma_valid = 0;
313 rc = mdt_attr_get_complex(mti, obj, &mti->mti_attr);
315 mdt_fake_ma(&mti->mti_attr);
316 mdt_pack_attr2body(mti, body, &mti->mti_attr.ma_attr,
317 mdt_object_fid(obj));
319 mdt_object_put(mti->mti_env, obj);
322 typedef void (*mdt_reconstructor)(struct mdt_thread_info *mti,
323 struct mdt_lock_handle *lhc);
325 static mdt_reconstructor reconstructors[REINT_MAX] = {
326 [REINT_SETATTR] = mdt_reconstruct_setattr,
327 [REINT_CREATE] = mdt_reconstruct_create,
328 [REINT_LINK] = mdt_reconstruct_generic,
329 [REINT_UNLINK] = mdt_reconstruct_generic,
330 [REINT_RENAME] = mdt_reconstruct_generic,
331 [REINT_OPEN] = mdt_reconstruct_open,
332 [REINT_SETXATTR] = mdt_reconstruct_generic,
333 [REINT_RMENTRY] = mdt_reconstruct_generic,
334 [REINT_MIGRATE] = mdt_reconstruct_generic
337 void mdt_reconstruct(struct mdt_thread_info *mti,
338 struct mdt_lock_handle *lhc)
340 mdt_reconstructor reconst;
342 LASSERT(mti->mti_rr.rr_opcode < REINT_MAX &&
343 (reconst = reconstructors[mti->mti_rr.rr_opcode]) != NULL);