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.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2011, 2013, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * lustre/mdt/mdt_recovery.c
38 * Lustre Metadata Target (mdt) recovery-related methods
40 * Author: Huang Hua <huanghua@clusterfs.com>
41 * Author: Pershin Mike <tappro@clusterfs.com>
44 #define DEBUG_SUBSYSTEM S_MDS
46 #include "mdt_internal.h"
48 struct lu_buf *mdt_buf(const struct lu_env *env, void *area, ssize_t len)
51 struct mdt_thread_info *mti;
53 mti = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
60 const struct lu_buf *mdt_buf_const(const struct lu_env *env,
61 const void *area, ssize_t len)
64 struct mdt_thread_info *mti;
66 mti = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
69 buf->lb_buf = (void *)area;
74 void mdt_trans_stop(const struct lu_env *env,
75 struct mdt_device *mdt, struct thandle *th)
77 dt_trans_stop(env, mdt->mdt_bottom, th);
81 * last_rcvd & last_committed update callbacks
83 extern struct lu_context_key mdt_thread_key;
85 /* This callback notifies MDT that transaction was done. This is needed by
86 * mdt_save_lock() only. It is similar to new target code and will be removed
87 * as mdt_save_lock() will be converted to use target structures */
88 static int mdt_txn_stop_cb(const struct lu_env *env,
89 struct thandle *txn, void *cookie)
91 struct mdt_thread_info *mti;
93 mti = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
96 if (mti->mti_has_trans)
97 CDEBUG(D_INFO, "More than one transaction\n");
99 mti->mti_has_trans = 1;
103 int mdt_fs_setup(const struct lu_env *env, struct mdt_device *mdt,
104 struct obd_device *obd, struct lustre_sb_info *lsi)
110 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_FS_SETUP))
113 /* prepare transactions callbacks */
114 mdt->mdt_txn_cb.dtc_txn_start = NULL;
115 mdt->mdt_txn_cb.dtc_txn_stop = mdt_txn_stop_cb;
116 mdt->mdt_txn_cb.dtc_txn_commit = NULL;
117 mdt->mdt_txn_cb.dtc_cookie = NULL;
118 mdt->mdt_txn_cb.dtc_tag = LCT_MD_THREAD;
119 CFS_INIT_LIST_HEAD(&mdt->mdt_txn_cb.dtc_linkage);
121 dt_txn_callback_add(mdt->mdt_bottom, &mdt->mdt_txn_cb);
126 void mdt_fs_cleanup(const struct lu_env *env, struct mdt_device *mdt)
130 /* Remove transaction callback */
131 dt_txn_callback_del(mdt->mdt_bottom, &mdt->mdt_txn_cb);
133 lu_object_put(env, &mdt->mdt_ck_obj->do_lu);
134 mdt->mdt_ck_obj = NULL;
138 /* reconstruction code */
139 static void mdt_steal_ack_locks(struct ptlrpc_request *req)
141 struct ptlrpc_service_part *svcpt;
142 struct obd_export *exp = req->rq_export;
144 struct ptlrpc_reply_state *oldrep;
147 /* CAVEAT EMPTOR: spinlock order */
148 spin_lock(&exp->exp_lock);
149 cfs_list_for_each (tmp, &exp->exp_outstanding_replies) {
150 oldrep = cfs_list_entry(tmp, struct ptlrpc_reply_state,
153 if (oldrep->rs_xid != req->rq_xid)
156 if (oldrep->rs_opc != lustre_msg_get_opc(req->rq_reqmsg))
157 CERROR ("Resent req xid "LPU64" has mismatched opc: "
158 "new %d old %d\n", req->rq_xid,
159 lustre_msg_get_opc(req->rq_reqmsg),
162 svcpt = oldrep->rs_svcpt;
163 spin_lock(&svcpt->scp_rep_lock);
165 cfs_list_del_init (&oldrep->rs_exp_list);
167 CDEBUG(D_HA, "Stealing %d locks from rs %p x"LPD64".t"LPD64
169 oldrep->rs_nlocks, oldrep,
170 oldrep->rs_xid, oldrep->rs_transno, oldrep->rs_opc,
171 libcfs_nid2str(exp->exp_connection->c_peer.nid));
173 for (i = 0; i < oldrep->rs_nlocks; i++)
174 ptlrpc_save_lock(req, &oldrep->rs_locks[i],
175 oldrep->rs_modes[i], 0);
176 oldrep->rs_nlocks = 0;
178 DEBUG_REQ(D_HA, req, "stole locks for");
179 spin_lock(&oldrep->rs_lock);
180 ptlrpc_schedule_difficult_reply(oldrep);
181 spin_unlock(&oldrep->rs_lock);
183 spin_unlock(&svcpt->scp_rep_lock);
186 spin_unlock(&exp->exp_lock);
190 * VBR: restore versions
192 void mdt_vbr_reconstruct(struct ptlrpc_request *req,
193 struct lsd_client_data *lcd)
195 __u64 pre_versions[4] = {0};
196 pre_versions[0] = lcd->lcd_pre_versions[0];
197 pre_versions[1] = lcd->lcd_pre_versions[1];
198 pre_versions[2] = lcd->lcd_pre_versions[2];
199 pre_versions[3] = lcd->lcd_pre_versions[3];
200 lustre_msg_set_versions(req->rq_repmsg, pre_versions);
203 void mdt_req_from_lcd(struct ptlrpc_request *req,
204 struct lsd_client_data *lcd)
206 DEBUG_REQ(D_HA, req, "restoring transno "LPD64"/status %d",
207 lcd->lcd_last_transno, lcd->lcd_last_result);
209 if (lustre_msg_get_opc(req->rq_reqmsg) == MDS_CLOSE ||
210 lustre_msg_get_opc(req->rq_repmsg) == MDS_DONE_WRITING) {
211 req->rq_transno = lcd->lcd_last_close_transno;
212 req->rq_status = lcd->lcd_last_close_result;
214 req->rq_transno = lcd->lcd_last_transno;
215 req->rq_status = lcd->lcd_last_result;
216 mdt_vbr_reconstruct(req, lcd);
218 if (req->rq_status != 0)
220 lustre_msg_set_transno(req->rq_repmsg, req->rq_transno);
221 lustre_msg_set_status(req->rq_repmsg, req->rq_status);
222 DEBUG_REQ(D_RPCTRACE, req, "restoring transno "LPD64"/status %d",
223 req->rq_transno, req->rq_status);
225 mdt_steal_ack_locks(req);
228 void mdt_reconstruct_generic(struct mdt_thread_info *mti,
229 struct mdt_lock_handle *lhc)
231 struct ptlrpc_request *req = mdt_info_req(mti);
232 struct tg_export_data *ted = &req->rq_export->exp_target_data;
234 return mdt_req_from_lcd(req, ted->ted_lcd);
237 static void mdt_reconstruct_create(struct mdt_thread_info *mti,
238 struct mdt_lock_handle *lhc)
240 struct ptlrpc_request *req = mdt_info_req(mti);
241 struct obd_export *exp = req->rq_export;
242 struct tg_export_data *ted = &exp->exp_target_data;
243 struct mdt_device *mdt = mti->mti_mdt;
244 struct mdt_object *child;
245 struct mdt_body *body;
248 mdt_req_from_lcd(req, ted->ted_lcd);
252 /* if no error, so child was created with requested fid */
253 child = mdt_object_find(mti->mti_env, mdt, mti->mti_rr.rr_fid2);
256 LCONSOLE_WARN("cannot lookup child "DFID": rc = %d; "
257 "evicting client %s with export %s\n",
258 PFID(mti->mti_rr.rr_fid2), rc,
259 obd_uuid2str(&exp->exp_client_uuid),
260 obd_export_nid2str(exp));
261 mdt_export_evict(exp);
265 body = req_capsule_server_get(mti->mti_pill, &RMF_MDT_BODY);
266 mti->mti_attr.ma_need = MA_INODE;
267 mti->mti_attr.ma_valid = 0;
268 rc = mdt_attr_get_complex(mti, child, &mti->mti_attr);
269 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_export_data *med = &exp->exp_mdt_data;
289 struct mdt_device *mdt = mti->mti_mdt;
290 struct mdt_object *obj;
291 struct mdt_body *body;
294 mdt_req_from_lcd(req, med->med_ted.ted_lcd);
298 body = req_capsule_server_get(mti->mti_pill, &RMF_MDT_BODY);
299 obj = mdt_object_find(mti->mti_env, mdt, mti->mti_rr.rr_fid1);
302 LCONSOLE_WARN("cannot lookup "DFID": rc = %d; "
303 "evicting client %s with export %s\n",
304 PFID(mti->mti_rr.rr_fid1), rc,
305 obd_uuid2str(&exp->exp_client_uuid),
306 obd_export_nid2str(exp));
307 mdt_export_evict(exp);
311 mti->mti_attr.ma_need = MA_INODE;
312 mti->mti_attr.ma_valid = 0;
313 mdt_attr_get_complex(mti, obj, &mti->mti_attr);
314 mdt_pack_attr2body(mti, body, &mti->mti_attr.ma_attr,
315 mdt_object_fid(obj));
316 if (mti->mti_ioepoch && (mti->mti_ioepoch->flags & MF_EPOCH_OPEN)) {
317 struct mdt_file_data *mfd;
318 struct mdt_body *repbody;
320 repbody = req_capsule_server_get(mti->mti_pill, &RMF_MDT_BODY);
321 repbody->mbo_ioepoch = obj->mot_ioepoch;
322 spin_lock(&med->med_open_lock);
323 cfs_list_for_each_entry(mfd, &med->med_open_head, mfd_list) {
324 if (mfd->mfd_xid == req->rq_xid)
327 LASSERT(&mfd->mfd_list != &med->med_open_head);
328 spin_unlock(&med->med_open_lock);
329 repbody->mbo_handle.cookie = mfd->mfd_handle.h_cookie;
332 mdt_object_put(mti->mti_env, obj);
335 typedef void (*mdt_reconstructor)(struct mdt_thread_info *mti,
336 struct mdt_lock_handle *lhc);
338 static mdt_reconstructor reconstructors[REINT_MAX] = {
339 [REINT_SETATTR] = mdt_reconstruct_setattr,
340 [REINT_CREATE] = mdt_reconstruct_create,
341 [REINT_LINK] = mdt_reconstruct_generic,
342 [REINT_UNLINK] = mdt_reconstruct_generic,
343 [REINT_RENAME] = mdt_reconstruct_generic,
344 [REINT_OPEN] = mdt_reconstruct_open,
345 [REINT_SETXATTR] = mdt_reconstruct_generic
348 void mdt_reconstruct(struct mdt_thread_info *mti,
349 struct mdt_lock_handle *lhc)
352 reconstructors[mti->mti_rr.rr_opcode](mti, lhc);