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);
80 static int mdt_clients_data_init(const struct lu_env *env,
81 struct mdt_device *mdt,
82 unsigned long last_size)
84 struct lr_server_data *lsd = &mdt->mdt_lut.lut_lsd;
85 struct lsd_client_data *lcd;
86 struct obd_device *obd = mdt2obd_dev(mdt);
96 /* When we do a clean MDS shutdown, we save the last_transno into
97 * the header. If we find clients with higher last_transno values
98 * then those clients may need recovery done. */
99 LASSERT(cfs_atomic_read(&obd->obd_req_replay_clients) == 0);
100 for (cl_idx = 0, off = lsd->lsd_client_start;
101 off < last_size; cl_idx++) {
103 struct obd_export *exp;
104 struct mdt_thread_info *mti;
106 off = lsd->lsd_client_start +
107 cl_idx * lsd->lsd_client_size;
109 rc = tgt_client_data_read(env, &mdt->mdt_lut, lcd, &off, cl_idx);
111 CERROR("error reading MDS %s idx %d, off %llu: rc %d\n",
112 LAST_RCVD, cl_idx, off, rc);
114 break; /* read error shouldn't cause startup to fail */
117 if (lcd->lcd_uuid[0] == '\0') {
118 CDEBUG(D_INFO, "skipping zeroed client at offset %d\n",
123 last_transno = lcd_last_transno(lcd);
125 /* These exports are cleaned up by mdt_obd_disconnect(), so
126 * they need to be set up like real exports as
127 * mdt_obd_connect() does.
129 CDEBUG(D_HA, "RCVRNG CLIENT uuid: %s idx: %d lr: "LPU64
130 " srv lr: "LPU64" lx: "LPU64"\n", lcd->lcd_uuid, cl_idx,
131 last_transno, lsd->lsd_last_transno,
134 exp = class_new_export(obd, (struct obd_uuid *)lcd->lcd_uuid);
136 if (PTR_ERR(exp) == -EALREADY) {
137 /* export already exists, zero out this one */
138 CERROR("Duplicate export %s!\n", lcd->lcd_uuid);
141 GOTO(err_client, rc = PTR_ERR(exp));
144 mti = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
145 LASSERT(mti != NULL);
147 /* copy on-disk lcd to the export */
148 *exp->exp_target_data.ted_lcd = *lcd;
149 rc = tgt_client_add(env, exp, cl_idx);
150 /* can't fail existing */
151 LASSERTF(rc == 0, "rc = %d\n", rc);
152 /* VBR: set export last committed version */
153 exp->exp_last_committed = last_transno;
154 spin_lock(&exp->exp_lock);
155 exp->exp_connecting = 0;
156 exp->exp_in_recovery = 0;
157 spin_unlock(&exp->exp_lock);
158 obd->obd_max_recoverable_clients++;
159 class_export_put(exp);
161 CDEBUG(D_OTHER, "client at idx %d has last_transno ="LPU64"\n",
162 cl_idx, last_transno);
163 /* protect __u64 value update */
164 spin_lock(&mdt->mdt_lut.lut_translock);
165 mdt->mdt_lut.lut_last_transno = max(last_transno,
166 mdt->mdt_lut.lut_last_transno);
167 spin_unlock(&mdt->mdt_lut.lut_translock);
175 static int mdt_server_data_init(const struct lu_env *env,
176 struct mdt_device *mdt,
177 struct lustre_sb_info *lsi)
179 struct lr_server_data *lsd = &mdt->mdt_lut.lut_lsd;
180 struct lsd_client_data *lcd = NULL;
181 struct obd_device *obd = mdt2obd_dev(mdt);
182 struct mdt_thread_info *mti;
183 struct dt_object *obj;
185 unsigned long last_rcvd_size;
191 /* ensure padding in the struct is the correct size */
192 CLASSERT(offsetof(struct lr_server_data, lsd_padding) +
193 sizeof(lsd->lsd_padding) == LR_SERVER_SIZE);
194 CLASSERT(offsetof(struct lsd_client_data, lcd_padding) +
195 sizeof(lcd->lcd_padding) == LR_CLIENT_SIZE);
197 rc = server_name2index(obd->obd_name, &index, NULL);
199 CERROR("%s: Can not get index from obd_name: rc = %d\n",
204 mti = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
205 LASSERT(mti != NULL);
206 la = &mti->mti_attr.ma_attr;
208 obj = mdt->mdt_lut.lut_last_rcvd;
209 rc = dt_attr_get(env, obj, la, BYPASS_CAPA);
213 last_rcvd_size = (unsigned long)la->la_size;
215 if (last_rcvd_size == 0) {
216 LCONSOLE_WARN("%s: new disk, initializing\n", obd->obd_name);
218 memcpy(lsd->lsd_uuid, obd->obd_uuid.uuid,
219 sizeof(lsd->lsd_uuid));
220 lsd->lsd_last_transno = 0;
221 lsd->lsd_mount_count = 0;
222 lsd->lsd_server_size = LR_SERVER_SIZE;
223 lsd->lsd_client_start = LR_CLIENT_START;
224 lsd->lsd_client_size = LR_CLIENT_SIZE;
225 lsd->lsd_feature_compat = OBD_COMPAT_MDT;
226 lsd->lsd_feature_rocompat = OBD_ROCOMPAT_LOVOBJID;
227 lsd->lsd_feature_incompat = OBD_INCOMPAT_MDT |
228 OBD_INCOMPAT_COMMON_LR |
229 OBD_INCOMPAT_MULTI_OI;
230 lsd->lsd_osd_index = index;
232 LCONSOLE_WARN("%s: used disk, loading\n", obd->obd_name);
233 rc = tgt_server_data_read(env, &mdt->mdt_lut);
235 CERROR("error reading MDS %s: rc %d\n", LAST_RCVD, rc);
238 if (strcmp(lsd->lsd_uuid, obd->obd_uuid.uuid) != 0) {
239 LCONSOLE_ERROR_MSG(0x157, "Trying to start OBD %s using"
240 "the wrong disk %s. Were the /dev/ "
241 "assignments rearranged?\n",
242 obd->obd_uuid.uuid, lsd->lsd_uuid);
243 GOTO(out, rc = -EINVAL);
245 lsd->lsd_feature_compat |= OBD_COMPAT_MDT;
246 lsd->lsd_feature_incompat |= OBD_INCOMPAT_MDT |
247 OBD_INCOMPAT_COMMON_LR;
248 if (lsd->lsd_osd_index != index) {
249 LCONSOLE_ERROR_MSG(0x157, "%s: index %d in last rcvd is"
250 "different with the index %d in"
251 "config log, It might be disk"
252 "corruption!\n", obd->obd_name,
253 lsd->lsd_osd_index, index);
254 GOTO(out, rc = -EINVAL);
257 mount_count = lsd->lsd_mount_count;
259 if (lsd->lsd_feature_incompat & ~MDT_INCOMPAT_SUPP) {
260 CERROR("%s: unsupported incompat filesystem feature(s) %x\n",
262 lsd->lsd_feature_incompat & ~MDT_INCOMPAT_SUPP);
263 GOTO(out, rc = -EINVAL);
265 if (lsd->lsd_feature_rocompat & ~MDT_ROCOMPAT_SUPP) {
266 CERROR("%s: unsupported read-only filesystem feature(s) %x\n",
268 lsd->lsd_feature_rocompat & ~MDT_ROCOMPAT_SUPP);
269 /* XXX: Do something like remount filesystem read-only */
270 GOTO(out, rc = -EINVAL);
272 /** Interop: evict all clients at first boot with 1.8 last_rcvd */
273 if (!(lsd->lsd_feature_compat & OBD_COMPAT_20)) {
274 if (last_rcvd_size > lsd->lsd_client_start) {
275 LCONSOLE_WARN("Mounting %s at first time on 1.8 FS, "
276 "remove all clients for interop needs\n",
278 rc = tgt_truncate_last_rcvd(env, &mdt->mdt_lut,
279 lsd->lsd_client_start);
282 last_rcvd_size = lsd->lsd_client_start;
284 /** set 2.0 flag to upgrade/downgrade between 1.8 and 2.0 */
285 lsd->lsd_feature_compat |= OBD_COMPAT_20;
288 lsd->lsd_feature_incompat |= OBD_INCOMPAT_FID;
290 spin_lock(&mdt->mdt_lut.lut_translock);
291 mdt->mdt_lut.lut_last_transno = lsd->lsd_last_transno;
292 spin_unlock(&mdt->mdt_lut.lut_translock);
294 CDEBUG(D_INODE, "========BEGIN DUMPING LAST_RCVD========\n");
295 CDEBUG(D_INODE, "%s: server last_transno: "LPU64"\n",
296 obd->obd_name, mdt->mdt_lut.lut_last_transno);
297 CDEBUG(D_INODE, "%s: server mount_count: "LPU64"\n",
298 obd->obd_name, mount_count + 1);
299 CDEBUG(D_INODE, "%s: server data size: %u\n",
300 obd->obd_name, lsd->lsd_server_size);
301 CDEBUG(D_INODE, "%s: per-client data start: %u\n",
302 obd->obd_name, lsd->lsd_client_start);
303 CDEBUG(D_INODE, "%s: per-client data size: %u\n",
304 obd->obd_name, lsd->lsd_client_size);
305 CDEBUG(D_INODE, "%s: last_rcvd size: %lu\n",
306 obd->obd_name, last_rcvd_size);
307 CDEBUG(D_INODE, "%s: last_rcvd clients: %lu\n", obd->obd_name,
308 last_rcvd_size <= lsd->lsd_client_start ? 0 :
309 (last_rcvd_size - lsd->lsd_client_start) /
310 lsd->lsd_client_size);
311 CDEBUG(D_INODE, "========END DUMPING LAST_RCVD========\n");
313 if (!lsd->lsd_server_size || !lsd->lsd_client_start ||
314 !lsd->lsd_client_size) {
315 CERROR("Bad last_rcvd contents!\n");
316 GOTO(out, rc = -EINVAL);
319 rc = mdt_clients_data_init(env, mdt, last_rcvd_size);
321 GOTO(err_client, rc);
323 spin_lock(&mdt->mdt_lut.lut_translock);
324 /* obd_last_committed is used for compatibility
325 * with other lustre recovery code */
326 obd->obd_last_committed = mdt->mdt_lut.lut_last_transno;
327 spin_unlock(&mdt->mdt_lut.lut_translock);
329 obd->u.obt.obt_mount_count = mount_count + 1;
330 obd->u.obt.obt_instance = (__u32)obd->u.obt.obt_mount_count;
331 lsd->lsd_mount_count = obd->u.obt.obt_mount_count;
333 /* save it, so mount count and last_transno is current */
334 rc = tgt_server_data_update(env, &mdt->mdt_lut, 0);
336 GOTO(err_client, rc);
341 class_disconnect_exports(obd);
347 * last_rcvd & last_committed update callbacks
349 static int mdt_last_rcvd_update(struct mdt_thread_info *mti,
352 struct mdt_device *mdt = mti->mti_mdt;
353 struct ptlrpc_request *req = mdt_info_req(mti);
354 struct tg_export_data *ted;
355 struct lsd_client_data *lcd;
358 __s32 rc = th->th_result;
362 LASSERT(req->rq_export);
364 ted = &req->rq_export->exp_target_data;
367 mutex_lock(&ted->ted_lcd_lock);
369 /* if the export has already been disconnected, we have no last_rcvd
370 * slot, update server data with latest transno then */
372 mutex_unlock(&ted->ted_lcd_lock);
373 CWARN("commit transaction for disconnected client %s: rc %d\n",
374 req->rq_export->exp_client_uuid.uuid, rc);
375 err = tgt_server_data_write(mti->mti_env, &mdt->mdt_lut, th);
379 off = ted->ted_lr_off;
380 LASSERT(ergo(mti->mti_transno == 0, rc != 0));
381 if (lustre_msg_get_opc(req->rq_reqmsg) == MDS_CLOSE ||
382 lustre_msg_get_opc(req->rq_reqmsg) == MDS_DONE_WRITING) {
383 if (mti->mti_transno != 0) {
384 if (lcd->lcd_last_close_transno > mti->mti_transno) {
385 CERROR("Trying to overwrite bigger transno:"
386 "on-disk: "LPU64", new: "LPU64" "
387 "replay: %d. see LU-617.\n",
388 lcd->lcd_last_close_transno,
389 mti->mti_transno, req_is_replay(req));
390 if (req_is_replay(req)) {
391 spin_lock(&req->rq_export->exp_lock);
392 req->rq_export->exp_vbr_failed = 1;
393 spin_unlock(&req->rq_export->exp_lock);
395 mutex_unlock(&ted->ted_lcd_lock);
396 RETURN(req_is_replay(req) ? -EOVERFLOW : 0);
398 lcd->lcd_last_close_transno = mti->mti_transno;
400 lcd->lcd_last_close_xid = req->rq_xid;
401 lcd->lcd_last_close_result = rc;
403 /* VBR: save versions in last_rcvd for reconstruct. */
404 __u64 *pre_versions = lustre_msg_get_versions(req->rq_repmsg);
406 lcd->lcd_pre_versions[0] = pre_versions[0];
407 lcd->lcd_pre_versions[1] = pre_versions[1];
408 lcd->lcd_pre_versions[2] = pre_versions[2];
409 lcd->lcd_pre_versions[3] = pre_versions[3];
411 if (mti->mti_transno != 0) {
412 if (lcd->lcd_last_transno > mti->mti_transno) {
413 CERROR("Trying to overwrite bigger transno:"
414 "on-disk: "LPU64", new: "LPU64" "
415 "replay: %d. see LU-617.\n",
416 lcd->lcd_last_transno,
417 mti->mti_transno, req_is_replay(req));
418 if (req_is_replay(req)) {
419 spin_lock(&req->rq_export->exp_lock);
420 req->rq_export->exp_vbr_failed = 1;
421 spin_unlock(&req->rq_export->exp_lock);
423 mutex_unlock(&ted->ted_lcd_lock);
424 RETURN(req_is_replay(req) ? -EOVERFLOW : 0);
426 lcd->lcd_last_transno = mti->mti_transno;
428 lcd->lcd_last_xid = req->rq_xid;
429 lcd->lcd_last_result = rc;
430 /*XXX: save intent_disposition in mdt_thread_info?
431 * also there is bug - intent_dispostion is __u64,
432 * see struct ldlm_reply->lock_policy_res1; */
433 lcd->lcd_last_data = mti->mti_opdata;
436 if (exp_connect_flags(mti->mti_exp) & OBD_CONNECT_LIGHTWEIGHT) {
437 /* Although lightweight (LW) connections have no slot in
438 * last_rcvd, we still want to maintain the in-memory
439 * lsd_client_data structure in order to properly handle reply
441 struct lu_target *tg = &mdt->mdt_lut;
444 mutex_unlock(&ted->ted_lcd_lock);
447 /* All operations performed by LW clients are synchronous and
448 * we store the committed transno in the last_rcvd header */
449 spin_lock(&tg->lut_translock);
450 if (mti->mti_transno > tg->lut_lsd.lsd_last_transno) {
451 tg->lut_lsd.lsd_last_transno = mti->mti_transno;
454 spin_unlock(&tg->lut_translock);
457 err = tgt_server_data_write(mti->mti_env, tg, th);
458 } else if (off <= 0) {
459 CERROR("%s: client idx %d has offset %lld\n",
460 mdt2obd_dev(mdt)->obd_name, ted->ted_lr_idx, off);
461 mutex_unlock(&ted->ted_lcd_lock);
464 err = tgt_client_data_write(mti->mti_env, &mdt->mdt_lut, lcd,
466 mutex_unlock(&ted->ted_lcd_lock);
471 extern struct lu_context_key mdt_thread_key;
473 /* add credits for last_rcvd update */
474 static int mdt_txn_start_cb(const struct lu_env *env,
475 struct thandle *th, void *cookie)
477 struct mdt_device *mdt = cookie;
478 struct mdt_thread_info *mti;
482 mti = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
484 LASSERT(mdt->mdt_lut.lut_last_rcvd);
485 if (mti->mti_exp == NULL)
488 rc = dt_declare_record_write(env, mdt->mdt_lut.lut_last_rcvd,
489 sizeof(struct lsd_client_data),
490 mti->mti_exp->exp_target_data.ted_lr_off,
495 rc = dt_declare_record_write(env, mdt->mdt_lut.lut_last_rcvd,
496 sizeof(struct lr_server_data), 0, th);
500 /* we probably should not set local transno to the remote object
501 * on another storage, What about VBR on remote object? XXX */
502 if (mti->mti_mos != NULL && !mdt_object_remote(mti->mti_mos))
503 rc = dt_declare_version_set(env, mdt_obj2dt(mti->mti_mos), th);
508 /* Update last_rcvd records with latests transaction data */
509 static int mdt_txn_stop_cb(const struct lu_env *env,
510 struct thandle *txn, void *cookie)
512 struct mdt_device *mdt = cookie;
513 struct mdt_thread_info *mti;
514 struct ptlrpc_request *req;
516 mti = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
517 req = mdt_info_req(mti);
519 if (mti->mti_mdt == NULL || req == NULL)
522 if (mti->mti_has_trans) {
523 /* XXX: currently there are allowed cases, but the wrong cases
524 * are also possible, so better check is needed here */
525 CDEBUG(D_INFO, "More than one transaction "LPU64"\n",
530 mti->mti_has_trans = 1;
531 spin_lock(&mdt->mdt_lut.lut_translock);
532 if (txn->th_result != 0) {
533 if (mti->mti_transno != 0) {
534 CERROR("Replay transno "LPU64" failed: rc %d\n",
535 mti->mti_transno, txn->th_result);
538 } else if (mti->mti_transno == 0) {
539 mti->mti_transno = ++ mdt->mdt_lut.lut_last_transno;
541 /* should be replay */
542 if (mti->mti_transno > mdt->mdt_lut.lut_last_transno)
543 mdt->mdt_lut.lut_last_transno = mti->mti_transno;
545 spin_unlock(&mdt->mdt_lut.lut_translock);
546 /* sometimes the reply message has not been successfully packed */
547 LASSERT(req != NULL && req->rq_repmsg != NULL);
549 /** VBR: set new versions */
550 /* we probably should not set local transno to the remote object
551 * on another storage, What about VBR on remote object? XXX */
552 if (txn->th_result == 0 && mti->mti_mos != NULL &&
553 !mdt_object_remote(mti->mti_mos)) {
555 dt_version_set(env, mdt_obj2dt(mti->mti_mos),
556 mti->mti_transno, txn);
560 /* filling reply data */
561 CDEBUG(D_INODE, "transno = "LPU64", last_committed = "LPU64"\n",
562 mti->mti_transno, req->rq_export->exp_obd->obd_last_committed);
564 req->rq_transno = mti->mti_transno;
565 lustre_msg_set_transno(req->rq_repmsg, mti->mti_transno);
566 /* if can't add callback, do sync write */
567 txn->th_sync |= !!tgt_last_commit_cb_add(txn, &mdt->mdt_lut,
570 return mdt_last_rcvd_update(mti, txn);
573 int mdt_fs_setup(const struct lu_env *env, struct mdt_device *mdt,
574 struct obd_device *obd,
575 struct lustre_sb_info *lsi)
580 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_FS_SETUP))
583 /* prepare transactions callbacks */
584 mdt->mdt_txn_cb.dtc_txn_start = mdt_txn_start_cb;
585 mdt->mdt_txn_cb.dtc_txn_stop = mdt_txn_stop_cb;
586 mdt->mdt_txn_cb.dtc_txn_commit = NULL;
587 mdt->mdt_txn_cb.dtc_cookie = mdt;
588 mdt->mdt_txn_cb.dtc_tag = LCT_MD_THREAD;
589 CFS_INIT_LIST_HEAD(&mdt->mdt_txn_cb.dtc_linkage);
591 dt_txn_callback_add(mdt->mdt_bottom, &mdt->mdt_txn_cb);
593 rc = mdt_server_data_init(env, mdt, lsi);
598 void mdt_fs_cleanup(const struct lu_env *env, struct mdt_device *mdt)
602 /* Remove transaction callback */
603 dt_txn_callback_del(mdt->mdt_bottom, &mdt->mdt_txn_cb);
605 lu_object_put(env, &mdt->mdt_ck_obj->do_lu);
606 mdt->mdt_ck_obj = NULL;
610 /* reconstruction code */
611 static void mdt_steal_ack_locks(struct ptlrpc_request *req)
613 struct ptlrpc_service_part *svcpt;
614 struct obd_export *exp = req->rq_export;
616 struct ptlrpc_reply_state *oldrep;
619 /* CAVEAT EMPTOR: spinlock order */
620 spin_lock(&exp->exp_lock);
621 cfs_list_for_each (tmp, &exp->exp_outstanding_replies) {
622 oldrep = cfs_list_entry(tmp, struct ptlrpc_reply_state,
625 if (oldrep->rs_xid != req->rq_xid)
628 if (oldrep->rs_opc != lustre_msg_get_opc(req->rq_reqmsg))
629 CERROR ("Resent req xid "LPU64" has mismatched opc: "
630 "new %d old %d\n", req->rq_xid,
631 lustre_msg_get_opc(req->rq_reqmsg),
634 svcpt = oldrep->rs_svcpt;
635 spin_lock(&svcpt->scp_rep_lock);
637 cfs_list_del_init (&oldrep->rs_exp_list);
639 CDEBUG(D_HA, "Stealing %d locks from rs %p x"LPD64".t"LPD64
641 oldrep->rs_nlocks, oldrep,
642 oldrep->rs_xid, oldrep->rs_transno, oldrep->rs_opc,
643 libcfs_nid2str(exp->exp_connection->c_peer.nid));
645 for (i = 0; i < oldrep->rs_nlocks; i++)
646 ptlrpc_save_lock(req, &oldrep->rs_locks[i],
647 oldrep->rs_modes[i], 0);
648 oldrep->rs_nlocks = 0;
650 DEBUG_REQ(D_HA, req, "stole locks for");
651 spin_lock(&oldrep->rs_lock);
652 ptlrpc_schedule_difficult_reply(oldrep);
653 spin_unlock(&oldrep->rs_lock);
655 spin_unlock(&svcpt->scp_rep_lock);
658 spin_unlock(&exp->exp_lock);
662 * VBR: restore versions
664 void mdt_vbr_reconstruct(struct ptlrpc_request *req,
665 struct lsd_client_data *lcd)
667 __u64 pre_versions[4] = {0};
668 pre_versions[0] = lcd->lcd_pre_versions[0];
669 pre_versions[1] = lcd->lcd_pre_versions[1];
670 pre_versions[2] = lcd->lcd_pre_versions[2];
671 pre_versions[3] = lcd->lcd_pre_versions[3];
672 lustre_msg_set_versions(req->rq_repmsg, pre_versions);
675 void mdt_req_from_lcd(struct ptlrpc_request *req,
676 struct lsd_client_data *lcd)
678 DEBUG_REQ(D_HA, req, "restoring transno "LPD64"/status %d",
679 lcd->lcd_last_transno, lcd->lcd_last_result);
681 if (lustre_msg_get_opc(req->rq_reqmsg) == MDS_CLOSE ||
682 lustre_msg_get_opc(req->rq_repmsg) == MDS_DONE_WRITING) {
683 req->rq_transno = lcd->lcd_last_close_transno;
684 req->rq_status = lcd->lcd_last_close_result;
686 req->rq_transno = lcd->lcd_last_transno;
687 req->rq_status = lcd->lcd_last_result;
688 mdt_vbr_reconstruct(req, lcd);
690 if (req->rq_status != 0)
692 lustre_msg_set_transno(req->rq_repmsg, req->rq_transno);
693 lustre_msg_set_status(req->rq_repmsg, req->rq_status);
694 DEBUG_REQ(D_RPCTRACE, req, "restoring transno "LPD64"/status %d",
695 req->rq_transno, req->rq_status);
697 mdt_steal_ack_locks(req);
700 void mdt_reconstruct_generic(struct mdt_thread_info *mti,
701 struct mdt_lock_handle *lhc)
703 struct ptlrpc_request *req = mdt_info_req(mti);
704 struct tg_export_data *ted = &req->rq_export->exp_target_data;
706 return mdt_req_from_lcd(req, ted->ted_lcd);
709 static void mdt_reconstruct_create(struct mdt_thread_info *mti,
710 struct mdt_lock_handle *lhc)
712 struct ptlrpc_request *req = mdt_info_req(mti);
713 struct obd_export *exp = req->rq_export;
714 struct tg_export_data *ted = &exp->exp_target_data;
715 struct mdt_device *mdt = mti->mti_mdt;
716 struct mdt_object *child;
717 struct mdt_body *body;
720 mdt_req_from_lcd(req, ted->ted_lcd);
724 /* if no error, so child was created with requested fid */
725 child = mdt_object_find(mti->mti_env, mdt, mti->mti_rr.rr_fid2);
728 LCONSOLE_WARN("Child "DFID" lookup error %d."
729 " Evicting client %s with export %s.\n",
730 PFID(mdt_object_fid(child)), rc,
731 obd_uuid2str(&exp->exp_client_uuid),
732 obd_export_nid2str(exp));
733 mdt_export_evict(exp);
738 body = req_capsule_server_get(mti->mti_pill, &RMF_MDT_BODY);
739 mti->mti_attr.ma_need = MA_INODE;
740 mti->mti_attr.ma_valid = 0;
741 rc = mdt_attr_get_complex(mti, child, &mti->mti_attr);
742 if (rc == -EREMOTE) {
743 /* object was created on remote server */
744 if (!mdt_is_dne_client(exp))
745 /* Return -EIO for old client */
749 body->valid |= OBD_MD_MDS;
751 mdt_pack_attr2body(mti, body, &mti->mti_attr.ma_attr,
752 mdt_object_fid(child));
753 mdt_object_put(mti->mti_env, child);
756 static void mdt_reconstruct_setattr(struct mdt_thread_info *mti,
757 struct mdt_lock_handle *lhc)
759 struct ptlrpc_request *req = mdt_info_req(mti);
760 struct obd_export *exp = req->rq_export;
761 struct mdt_export_data *med = &exp->exp_mdt_data;
762 struct mdt_device *mdt = mti->mti_mdt;
763 struct mdt_object *obj;
764 struct mdt_body *body;
766 mdt_req_from_lcd(req, med->med_ted.ted_lcd);
770 body = req_capsule_server_get(mti->mti_pill, &RMF_MDT_BODY);
771 obj = mdt_object_find(mti->mti_env, mdt, mti->mti_rr.rr_fid1);
773 int rc = PTR_ERR(obj);
774 LCONSOLE_WARN(""DFID" lookup error %d."
775 " Evicting client %s with export %s.\n",
776 PFID(mdt_object_fid(obj)), rc,
777 obd_uuid2str(&exp->exp_client_uuid),
778 obd_export_nid2str(exp));
779 mdt_export_evict(exp);
783 mti->mti_attr.ma_need = MA_INODE;
784 mti->mti_attr.ma_valid = 0;
785 mdt_attr_get_complex(mti, obj, &mti->mti_attr);
786 mdt_pack_attr2body(mti, body, &mti->mti_attr.ma_attr,
787 mdt_object_fid(obj));
788 if (mti->mti_ioepoch && (mti->mti_ioepoch->flags & MF_EPOCH_OPEN)) {
789 struct mdt_file_data *mfd;
790 struct mdt_body *repbody;
792 repbody = req_capsule_server_get(mti->mti_pill, &RMF_MDT_BODY);
793 repbody->ioepoch = obj->mot_ioepoch;
794 spin_lock(&med->med_open_lock);
795 cfs_list_for_each_entry(mfd, &med->med_open_head, mfd_list) {
796 if (mfd->mfd_xid == req->rq_xid)
799 LASSERT(&mfd->mfd_list != &med->med_open_head);
800 spin_unlock(&med->med_open_lock);
801 repbody->handle.cookie = mfd->mfd_handle.h_cookie;
804 mdt_object_put(mti->mti_env, obj);
807 typedef void (*mdt_reconstructor)(struct mdt_thread_info *mti,
808 struct mdt_lock_handle *lhc);
810 static mdt_reconstructor reconstructors[REINT_MAX] = {
811 [REINT_SETATTR] = mdt_reconstruct_setattr,
812 [REINT_CREATE] = mdt_reconstruct_create,
813 [REINT_LINK] = mdt_reconstruct_generic,
814 [REINT_UNLINK] = mdt_reconstruct_generic,
815 [REINT_RENAME] = mdt_reconstruct_generic,
816 [REINT_OPEN] = mdt_reconstruct_open,
817 [REINT_SETXATTR] = mdt_reconstruct_generic
820 void mdt_reconstruct(struct mdt_thread_info *mti,
821 struct mdt_lock_handle *lhc)
824 reconstructors[mti->mti_rr.rr_opcode](mti, lhc);