1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
4 * linux/mdt/mdt_reint.c
5 * Lustre Metadata Target (mdt) reintegration routines
7 * Copyright (C) 2002-2006 Cluster File Systems, Inc.
8 * Author: Peter Braam <braam@clusterfs.com>
9 * Author: Andreas Dilger <adilger@clusterfs.com>
10 * Author: Phil Schwan <phil@clusterfs.com>
11 * Author: Huang Hua <huanghua@clusterfs.com>
12 * Author: Yury Umanets <umka@clusterfs.com>
14 * This file is part of the Lustre file system, http://www.lustre.org
15 * Lustre is a trademark of Cluster File Systems, Inc.
17 * You may have signed or agreed to another license before downloading
18 * this software. If so, you are bound by the terms and conditions
19 * of that agreement, and the following does not apply to you. See the
20 * LICENSE file included with this distribution for more information.
22 * If you did not agree to a different license, then this copy of Lustre
23 * is open source software; you can redistribute it and/or modify it
24 * under the terms of version 2 of the GNU General Public License as
25 * published by the Free Software Foundation.
27 * In either case, Lustre is distributed in the hope that it will be
28 * useful, but WITHOUT ANY WARRANTY; without even the implied warranty
29 * of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
30 * license text for more details.
34 # define EXPORT_SYMTAB
36 #define DEBUG_SUBSYSTEM S_MDS
38 #include "mdt_internal.h"
41 static inline void mdt_reint_init_ma(struct mdt_thread_info *info,
44 ma->ma_lmm = req_capsule_server_get(&info->mti_pill, &RMF_MDT_MD);
45 ma->ma_lmm_size = req_capsule_get_size(&info->mti_pill,
46 &RMF_MDT_MD, RCL_SERVER);
48 ma->ma_cookie = req_capsule_server_get(&info->mti_pill,
50 ma->ma_cookie_size = req_capsule_get_size(&info->mti_pill,
54 ma->ma_need = MA_INODE | MA_LOV | MA_COOKIE;
58 static int mdt_create_pack_capa(struct mdt_thread_info *info, int rc,
59 struct mdt_object *object,
60 struct mdt_body *repbody)
64 /* for cross-ref mkdir, mds capa has been fetched from remote obj, then
65 * we won't go to below*/
66 if (repbody->valid & OBD_MD_FLMDSCAPA)
69 if (rc == 0 && info->mti_mdt->mdt_opts.mo_mds_capa) {
70 struct lustre_capa *capa;
72 capa = req_capsule_server_get(&info->mti_pill, &RMF_CAPA1);
74 capa->lc_opc = CAPA_OPC_MDS_DEFAULT;
75 rc = mo_capa_get(info->mti_env, mdt_object_child(object), capa,
78 repbody->valid |= OBD_MD_FLMDSCAPA;
80 if (!(repbody->valid & OBD_MD_FLMDSCAPA))
81 lustre_shrink_reply(mdt_info_req(info), REPLY_REC_OFF+1, 0, 1);
86 static int mdt_md_create(struct mdt_thread_info *info)
88 struct mdt_device *mdt = info->mti_mdt;
89 struct mdt_object *parent;
90 struct mdt_object *child;
91 struct mdt_lock_handle *lh;
92 struct mdt_body *repbody;
93 struct md_attr *ma = &info->mti_attr;
94 struct mdt_reint_record *rr = &info->mti_rr;
95 struct lu_name *lname;
99 DEBUG_REQ(D_INODE, mdt_info_req(info), "Create (%s->"DFID") in "DFID,
100 rr->rr_name, PFID(rr->rr_fid2), PFID(rr->rr_fid1));
102 repbody = req_capsule_server_get(&info->mti_pill, &RMF_MDT_BODY);
104 lh = &info->mti_lh[MDT_LH_PARENT];
105 mdt_lock_pdo_init(lh, LCK_PW, rr->rr_name, rr->rr_namelen);
107 parent = mdt_object_find_lock(info, rr->rr_fid1, lh,
108 MDS_INODELOCK_UPDATE);
110 RETURN(PTR_ERR(parent));
112 child = mdt_object_find(info->mti_env, mdt, rr->rr_fid2);
113 if (likely(!IS_ERR(child))) {
114 struct md_object *next = mdt_object_child(parent);
116 ma->ma_need = MA_INODE;
118 /* capa for cross-ref will be stored here */
119 ma->ma_capa = req_capsule_server_get(&info->mti_pill,
121 LASSERT(ma->ma_capa);
123 mdt_fail_write(info->mti_env, info->mti_mdt->mdt_bottom,
124 OBD_FAIL_MDS_REINT_CREATE_WRITE);
126 /* Let lower layer know current lock mode. */
127 info->mti_spec.sp_cr_mode =
128 mdt_dlm_mode2mdl_mode(lh->mlh_pdo_mode);
131 * Do perform lookup sanity check. We do not know if name exists
134 info->mti_spec.sp_cr_lookup = 1;
136 lname = mdt_name(info->mti_env, (char *)rr->rr_name,
138 rc = mdo_create(info->mti_env, next, lname,
139 mdt_object_child(child),
140 &info->mti_spec, ma);
142 /* Return fid & attr to client. */
143 if (ma->ma_valid & MA_INODE)
144 mdt_pack_attr2body(info, repbody, &ma->ma_attr,
145 mdt_object_fid(child));
147 mdt_object_put(info->mti_env, child);
151 mdt_create_pack_capa(info, rc, child, repbody);
152 mdt_object_unlock_put(info, parent, lh, rc);
156 /* Partial request to create object only */
157 static int mdt_md_mkobj(struct mdt_thread_info *info)
159 struct mdt_device *mdt = info->mti_mdt;
160 struct mdt_object *o;
161 struct mdt_body *repbody;
162 struct md_attr *ma = &info->mti_attr;
166 DEBUG_REQ(D_INODE, mdt_info_req(info), "Partial create "DFID"",
167 PFID(info->mti_rr.rr_fid2));
169 repbody = req_capsule_server_get(&info->mti_pill, &RMF_MDT_BODY);
171 o = mdt_object_find(info->mti_env, mdt, info->mti_rr.rr_fid2);
173 struct md_object *next = mdt_object_child(o);
175 ma->ma_need = MA_INODE;
179 * Cross-ref create can encounter already created obj in case of
180 * recovery, just get attr in that case.
182 if (mdt_object_exists(o) == 1) {
183 rc = mo_attr_get(info->mti_env, next, ma);
186 * Here, NO permission check for object_create,
187 * such check has been done on the original MDS.
189 rc = mo_object_create(info->mti_env, next,
190 &info->mti_spec, ma);
193 /* Return fid & attr to client. */
194 if (ma->ma_valid & MA_INODE)
195 mdt_pack_attr2body(info, repbody, &ma->ma_attr,
198 mdt_object_put(info->mti_env, o);
202 mdt_create_pack_capa(info, rc, o, repbody);
206 /* In the raw-setattr case, we lock the child inode.
207 * In the write-back case or if being called from open,
208 * the client holds a lock already.
209 * We use the ATTR_FROM_OPEN (translated into MRF_SETATTR_LOCKED by
210 * mdt_setattr_unpack()) flag to tell these cases apart. */
211 int mdt_attr_set(struct mdt_thread_info *info, struct mdt_object *mo, int flags)
213 struct md_attr *ma = &info->mti_attr;
214 struct mdt_lock_handle *lh;
219 /* attr shouldn't be set on remote object */
220 LASSERT(mdt_object_exists(mo) >= 0);
223 som_update = (info->mti_epoch->flags & MF_SOM_CHANGE);
225 /* Try to avoid object_lock if another epoch has been started
227 if (som_update && (info->mti_epoch->ioepoch != mo->mot_ioepoch))
230 lh = &info->mti_lh[MDT_LH_PARENT];
231 mdt_lock_reg_init(lh, LCK_PW);
233 if (!(flags & MRF_SETATTR_LOCKED)) {
234 __u64 lockpart = MDS_INODELOCK_UPDATE;
235 if (ma->ma_attr.la_valid & (LA_MODE|LA_UID|LA_GID))
236 lockpart |= MDS_INODELOCK_LOOKUP;
238 rc = mdt_object_lock(info, mo, lh, lockpart, MDT_LOCAL_LOCK);
243 /* Setattrs are syncronized through dlm lock taken above. If another
244 * epoch started, its attributes may be already flushed on disk,
246 if (som_update && (info->mti_epoch->ioepoch != mo->mot_ioepoch))
247 GOTO(out_unlock, rc = 0);
249 if (mdt_object_exists(mo) == 0)
250 GOTO(out_unlock, rc = -ENOENT);
252 /* all attrs are packed into mti_attr in unpack_setattr */
253 mdt_fail_write(info->mti_env, info->mti_mdt->mdt_bottom,
254 OBD_FAIL_MDS_REINT_SETATTR_WRITE);
256 /* This is only for set ctime when rename's source is on remote MDS. */
257 if (unlikely(ma->ma_attr.la_valid == LA_CTIME))
258 ma->ma_attr_flags |= MDS_VTX_BYPASS;
260 /* all attrs are packed into mti_attr in unpack_setattr */
261 rc = mo_attr_set(info->mti_env, mdt_object_child(mo), ma);
263 GOTO(out_unlock, rc);
265 /* Re-enable SIZEONMDS. */
267 CDEBUG(D_INODE, "Closing epoch "LPU64" on "DFID". Count %d\n",
268 mo->mot_ioepoch, PFID(mdt_object_fid(mo)),
270 mdt_sizeonmds_enable(info, mo);
275 mdt_object_unlock(info, mo, lh, rc);
279 static int mdt_reint_setattr(struct mdt_thread_info *info,
280 struct mdt_lock_handle *lhc)
282 struct mdt_device *mdt = info->mti_mdt;
283 struct md_attr *ma = &info->mti_attr;
284 struct mdt_reint_record *rr = &info->mti_rr;
285 struct ptlrpc_request *req = mdt_info_req(info);
286 struct mdt_export_data *med = &req->rq_export->exp_mdt_data;
287 struct mdt_file_data *mfd;
288 struct mdt_object *mo;
289 struct md_object *next;
290 struct mdt_body *repbody;
294 DEBUG_REQ(D_INODE, req, "setattr "DFID" %x", PFID(rr->rr_fid1),
295 (unsigned int)ma->ma_attr.la_valid);
297 if (info->mti_dlm_req)
298 ldlm_request_cancel(req, info->mti_dlm_req, 0);
300 repbody = req_capsule_server_get(&info->mti_pill, &RMF_MDT_BODY);
301 mo = mdt_object_find(info->mti_env, info->mti_mdt, rr->rr_fid1);
303 GOTO(out, rc = PTR_ERR(mo));
305 if (!(mdt_conn_flags(info) & OBD_CONNECT_SOM)) {
306 if ((ma->ma_attr.la_valid & LA_SIZE) ||
307 (rr->rr_flags & MRF_SETATTR_LOCKED)) {
308 /* Check write access for the O_TRUNC case */
309 if (mdt_write_read(info->mti_mdt, mo) < 0)
310 GOTO(out_put, rc = -ETXTBSY);
312 } else if (info->mti_epoch &&
313 (info->mti_epoch->flags & MF_EPOCH_OPEN)) {
315 rc = mdt_write_get(info->mti_mdt, mo);
321 GOTO(out_put, rc = -ENOMEM);
323 mdt_epoch_open(info, mo);
324 repbody->ioepoch = mo->mot_ioepoch;
326 mdt_object_get(info->mti_env, mo);
327 mdt_mfd_set_mode(mfd, FMODE_EPOCHLCK);
328 mfd->mfd_object = mo;
329 mfd->mfd_xid = req->rq_xid;
331 spin_lock(&med->med_open_lock);
332 list_add(&mfd->mfd_list, &med->med_open_head);
333 spin_unlock(&med->med_open_lock);
334 repbody->handle.cookie = mfd->mfd_handle.h_cookie;
337 rc = mdt_attr_set(info, mo, rr->rr_flags);
341 if (info->mti_epoch && (info->mti_epoch->flags & MF_SOM_CHANGE)) {
342 LASSERT(mdt_conn_flags(info) & OBD_CONNECT_SOM);
343 LASSERT(info->mti_epoch);
345 spin_lock(&med->med_open_lock);
346 /* Size-on-MDS Update. Find and free mfd. */
347 mfd = mdt_handle2mfd(info, &info->mti_epoch->handle);
349 spin_unlock(&med->med_open_lock);
350 CDEBUG(D_INODE | D_ERROR, "no handle for file close: "
351 "fid = "DFID": cookie = "LPX64"\n",
352 PFID(info->mti_rr.rr_fid1),
353 info->mti_epoch->handle.cookie);
354 GOTO(out_put, rc = -ESTALE);
356 if (mfd->mfd_mode != FMODE_SOM) {
357 CWARN("mfd_mode %d not right should be in replay\n",
359 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) &
361 LASSERT(mfd->mfd_mode == FMODE_EPOCH);
362 mfd->mfd_mode = FMODE_SOM;
365 LASSERT(ma->ma_attr.la_valid & LA_SIZE);
366 LASSERT(!(info->mti_epoch->flags & MF_EPOCH_CLOSE));
368 class_handle_unhash(&mfd->mfd_handle);
369 list_del_init(&mfd->mfd_list);
370 spin_unlock(&med->med_open_lock);
372 mdt_mfd_close(info, mfd);
375 ma->ma_need = MA_INODE;
377 next = mdt_object_child(mo);
378 rc = mo_attr_get(info->mti_env, next, ma);
382 mdt_pack_attr2body(info, repbody, &ma->ma_attr, mdt_object_fid(mo));
384 if (mdt->mdt_opts.mo_oss_capa &&
385 S_ISREG(lu_object_attr(&mo->mot_obj.mo_lu)) &&
386 (ma->ma_attr.la_valid & LA_SIZE)) {
387 struct lustre_capa *capa;
389 capa = req_capsule_server_get(&info->mti_pill, &RMF_CAPA1);
391 capa->lc_opc = CAPA_OPC_OSS_DEFAULT | CAPA_OPC_OSS_TRUNC;
392 rc = mo_capa_get(info->mti_env, mdt_object_child(mo), capa, 0);
395 repbody->valid |= OBD_MD_FLOSSCAPA;
400 mdt_object_put(info->mti_env, mo);
405 static int mdt_reint_create(struct mdt_thread_info *info,
406 struct mdt_lock_handle *lhc)
411 if (MDT_FAIL_CHECK(OBD_FAIL_MDS_REINT_CREATE))
412 GOTO(out, rc = err_serious(-ESTALE));
414 if (info->mti_dlm_req)
415 ldlm_request_cancel(mdt_info_req(info), info->mti_dlm_req, 0);
417 switch (info->mti_attr.ma_attr.la_mode & S_IFMT) {
419 /* Cross-ref case. */
420 if (info->mti_cross_ref) {
421 rc = mdt_md_mkobj(info);
431 /* Special file should stay on the same node as parent. */
432 LASSERT(info->mti_rr.rr_namelen > 0);
433 rc = mdt_md_create(info);
437 rc = err_serious(-EOPNOTSUPP);
444 static int mdt_reint_unlink(struct mdt_thread_info *info,
445 struct mdt_lock_handle *lhc)
447 struct mdt_reint_record *rr = &info->mti_rr;
448 struct ptlrpc_request *req = mdt_info_req(info);
449 struct md_attr *ma = &info->mti_attr;
450 struct lu_fid *child_fid = &info->mti_tmp_fid1;
451 struct mdt_object *mp;
452 struct mdt_object *mc;
453 struct mdt_lock_handle *parent_lh;
454 struct mdt_lock_handle *child_lh;
455 struct lu_name *lname;
459 DEBUG_REQ(D_INODE, req, "unlink "DFID"/%s", PFID(rr->rr_fid1),
462 if (info->mti_dlm_req)
463 ldlm_request_cancel(req, info->mti_dlm_req, 0);
465 if (MDT_FAIL_CHECK(OBD_FAIL_MDS_REINT_UNLINK))
466 GOTO(out, rc = err_serious(-ENOENT));
468 /* step 1: lock the parent */
469 parent_lh = &info->mti_lh[MDT_LH_PARENT];
470 mdt_lock_pdo_init(parent_lh, LCK_PW, rr->rr_name,
473 mp = mdt_object_find_lock(info, rr->rr_fid1, parent_lh,
474 MDS_INODELOCK_UPDATE);
477 /* errors are possible here in cross-ref cases, see below */
478 if (info->mti_cross_ref)
483 mdt_reint_init_ma(info, ma);
484 if (!ma->ma_lmm || !ma->ma_cookie)
485 GOTO(out_unlock_parent, rc = -EINVAL);
487 if (info->mti_cross_ref) {
489 * Remote partial operation. It is possible that replay may
490 * happen on parent MDT and this operation will be repeated.
491 * Therefore the object absense is allowed case and nothing
492 * should be done here.
494 if (mdt_object_exists(mp) > 0) {
495 mdt_set_capainfo(info, 0, rr->rr_fid1, BYPASS_CAPA);
496 rc = mo_ref_del(info->mti_env,
497 mdt_object_child(mp), ma);
498 mdt_handle_last_unlink(info, mp, ma);
501 GOTO(out_unlock_parent, rc);
504 /* step 2: find & lock the child */
505 lname = mdt_name(info->mti_env, (char *)rr->rr_name, rr->rr_namelen);
506 rc = mdo_lookup(info->mti_env, mdt_object_child(mp),
507 lname, child_fid, &info->mti_spec);
509 GOTO(out_unlock_parent, rc);
511 /* We will lock the child regardless it is local or remote. No harm. */
512 mc = mdt_object_find(info->mti_env, info->mti_mdt, child_fid);
514 GOTO(out_unlock_parent, rc = PTR_ERR(mc));
515 child_lh = &info->mti_lh[MDT_LH_CHILD];
516 mdt_lock_reg_init(child_lh, LCK_EX);
517 rc = mdt_object_lock(info, mc, child_lh, MDS_INODELOCK_FULL,
520 mdt_object_put(info->mti_env, mc);
521 GOTO(out_unlock_parent, rc);
524 mdt_fail_write(info->mti_env, info->mti_mdt->mdt_bottom,
525 OBD_FAIL_MDS_REINT_UNLINK_WRITE);
528 * Now we can only make sure we need MA_INODE, in mdd layer, will check
529 * whether need MA_LOV and MA_COOKIE.
531 ma->ma_need = MA_INODE;
533 mdt_set_capainfo(info, 1, child_fid, BYPASS_CAPA);
534 rc = mdo_unlink(info->mti_env, mdt_object_child(mp),
535 mdt_object_child(mc), lname, ma);
537 mdt_handle_last_unlink(info, mc, ma);
540 mdt_object_unlock_put(info, mc, child_lh, rc);
542 mdt_object_unlock_put(info, mp, parent_lh, rc);
544 mdt_shrink_reply(info);
548 static int mdt_reint_link(struct mdt_thread_info *info,
549 struct mdt_lock_handle *lhc)
551 struct mdt_reint_record *rr = &info->mti_rr;
552 struct ptlrpc_request *req = mdt_info_req(info);
553 struct md_attr *ma = &info->mti_attr;
554 struct mdt_object *ms;
555 struct mdt_object *mp;
556 struct mdt_lock_handle *lhs;
557 struct mdt_lock_handle *lhp;
558 struct lu_name *lname;
562 DEBUG_REQ(D_INODE, req, "link "DFID" to "DFID"/%s",
563 PFID(rr->rr_fid1), PFID(rr->rr_fid2), rr->rr_name);
565 if (MDT_FAIL_CHECK(OBD_FAIL_MDS_REINT_LINK))
566 GOTO(out, rc = err_serious(-ENOENT));
568 if (info->mti_dlm_req)
569 ldlm_request_cancel(req, info->mti_dlm_req, 0);
571 if (info->mti_cross_ref) {
572 /* MDT holding name ask us to add ref. */
573 lhs = &info->mti_lh[MDT_LH_CHILD];
574 mdt_lock_reg_init(lhs, LCK_EX);
575 ms = mdt_object_find_lock(info, rr->rr_fid1, lhs,
576 MDS_INODELOCK_UPDATE);
578 GOTO(out, rc = PTR_ERR(ms));
580 mdt_set_capainfo(info, 0, rr->rr_fid1, BYPASS_CAPA);
581 rc = mo_ref_add(info->mti_env, mdt_object_child(ms), ma);
582 mdt_object_unlock_put(info, ms, lhs, rc);
586 /* Invalid case so return error immediately instead of
588 if (lu_fid_eq(rr->rr_fid1, rr->rr_fid2))
589 GOTO(out, rc = -EPERM);
591 /* step 1: find & lock the target parent dir */
592 lhp = &info->mti_lh[MDT_LH_PARENT];
593 mdt_lock_pdo_init(lhp, LCK_EX, rr->rr_name,
595 mp = mdt_object_find_lock(info, rr->rr_fid2, lhp,
596 MDS_INODELOCK_UPDATE);
598 GOTO(out, rc = PTR_ERR(mp));
600 /* step 2: find & lock the source */
601 lhs = &info->mti_lh[MDT_LH_CHILD];
602 mdt_lock_reg_init(lhs, LCK_EX);
604 ms = mdt_object_find(info->mti_env, info->mti_mdt, rr->rr_fid1);
606 GOTO(out_unlock_parent, rc = PTR_ERR(ms));
608 rc = mdt_object_lock(info, ms, lhs, MDS_INODELOCK_UPDATE,
611 mdt_object_put(info->mti_env, ms);
612 GOTO(out_unlock_parent, rc);
615 /* step 3: link it */
616 mdt_fail_write(info->mti_env, info->mti_mdt->mdt_bottom,
617 OBD_FAIL_MDS_REINT_LINK_WRITE);
619 lname = mdt_name(info->mti_env, (char *)rr->rr_name, rr->rr_namelen);
620 rc = mdo_link(info->mti_env, mdt_object_child(mp),
621 mdt_object_child(ms), lname, ma);
624 mdt_object_unlock_put(info, ms, lhs, rc);
626 mdt_object_unlock_put(info, mp, lhp, rc);
631 /* partial operation for rename */
632 static int mdt_reint_rename_tgt(struct mdt_thread_info *info)
634 struct mdt_reint_record *rr = &info->mti_rr;
635 struct ptlrpc_request *req = mdt_info_req(info);
636 struct md_attr *ma = &info->mti_attr;
637 struct mdt_object *mtgtdir;
638 struct mdt_object *mtgt = NULL;
639 struct mdt_lock_handle *lh_tgtdir;
640 struct mdt_lock_handle *lh_tgt = NULL;
641 struct lu_fid *tgt_fid = &info->mti_tmp_fid1;
642 struct lu_name *lname;
646 DEBUG_REQ(D_INODE, req, "rename_tgt: insert (%s->"DFID") in "DFID,
647 rr->rr_tgt, PFID(rr->rr_fid2), PFID(rr->rr_fid1));
649 /* step 1: lookup & lock the tgt dir. */
650 lh_tgtdir = &info->mti_lh[MDT_LH_PARENT];
651 mdt_lock_pdo_init(lh_tgtdir, LCK_PW, rr->rr_tgt,
653 mtgtdir = mdt_object_find_lock(info, rr->rr_fid1, lh_tgtdir,
654 MDS_INODELOCK_UPDATE);
656 GOTO(out, rc = PTR_ERR(mtgtdir));
658 /* step 2: find & lock the target object if exists. */
659 mdt_set_capainfo(info, 0, rr->rr_fid1, BYPASS_CAPA);
660 lname = mdt_name(info->mti_env, (char *)rr->rr_tgt, rr->rr_tgtlen);
661 rc = mdo_lookup(info->mti_env, mdt_object_child(mtgtdir),
662 lname, tgt_fid, &info->mti_spec);
663 if (rc != 0 && rc != -ENOENT) {
664 GOTO(out_unlock_tgtdir, rc);
665 } else if (rc == 0) {
667 * In case of replay that name can be already inserted, check
668 * that and do nothing if so.
670 if (lu_fid_eq(tgt_fid, rr->rr_fid2))
671 GOTO(out_unlock_tgtdir, rc);
673 lh_tgt = &info->mti_lh[MDT_LH_CHILD];
674 mdt_lock_reg_init(lh_tgt, LCK_EX);
676 mtgt = mdt_object_find_lock(info, tgt_fid, lh_tgt,
677 MDS_INODELOCK_LOOKUP);
679 GOTO(out_unlock_tgtdir, rc = PTR_ERR(mtgt));
681 mdt_reint_init_ma(info, ma);
682 if (!ma->ma_lmm || !ma->ma_cookie)
683 GOTO(out_unlock_tgt, rc = -EINVAL);
685 rc = mdo_rename_tgt(info->mti_env, mdt_object_child(mtgtdir),
686 mdt_object_child(mtgt), rr->rr_fid2,
688 } else /* -ENOENT */ {
689 rc = mdo_name_insert(info->mti_env, mdt_object_child(mtgtdir),
690 lname, rr->rr_fid2, ma);
693 /* handle last link of tgt object */
695 mdt_handle_last_unlink(info, mtgt, ma);
700 mdt_object_unlock_put(info, mtgt, lh_tgt, rc);
702 mdt_object_unlock_put(info, mtgtdir, lh_tgtdir, rc);
704 mdt_shrink_reply(info);
708 static int mdt_rename_lock(struct mdt_thread_info *info,
709 struct lustre_handle *lh)
711 struct ldlm_namespace *ns = info->mti_mdt->mdt_namespace;
712 ldlm_policy_data_t *policy = &info->mti_policy;
713 struct ldlm_res_id *res_id = &info->mti_res_id;
719 * Disable global rename BFL lock temporarily because
720 * when a mds do rename recoverying, which might enqueue
721 * BFL lock to the controller mds. and this req might be
722 * replay req for controller mds. but we did not have
723 * such handling in controller mds. XXX
726 ls = info->mti_mdt->mdt_md_dev.md_lu_dev.ld_site;
727 fid_build_reg_res_name(&LUSTRE_BFL_FID, res_id);
729 memset(policy, 0, sizeof *policy);
730 policy->l_inodebits.bits = MDS_INODELOCK_UPDATE;
732 if (ls->ls_control_exp == NULL) {
733 int flags = LDLM_FL_LOCAL_ONLY | LDLM_FL_ATOMIC_CB;
736 * Current node is controller, that is mdt0, where we should
739 rc = ldlm_cli_enqueue_local(ns, res_id, LDLM_IBITS, policy,
740 LCK_EX, &flags, ldlm_blocking_ast,
741 ldlm_completion_ast, NULL, NULL, 0,
744 struct ldlm_enqueue_info einfo = { LDLM_IBITS, LCK_EX,
745 ldlm_blocking_ast, ldlm_completion_ast, NULL, NULL };
749 * This is the case mdt0 is remote node, issue DLM lock like
752 rc = ldlm_cli_enqueue(ls->ls_control_exp, NULL, &einfo, res_id,
753 policy, &flags, NULL, 0, NULL, lh, 0);
759 static void mdt_rename_unlock(struct lustre_handle *lh)
762 /* Disable global rename BFL lock temporarily. see above XXX*/
765 LASSERT(lustre_handle_is_used(lh));
766 ldlm_lock_decref(lh, LCK_EX);
771 * This is is_subdir() variant, it is CMD if cmm forwards it to correct
772 * target. Source should not be ancestor of target dir. May be other rename
773 * checks can be moved here later.
775 static int mdt_rename_sanity(struct mdt_thread_info *info, struct lu_fid *fid)
777 struct mdt_reint_record *rr = &info->mti_rr;
778 struct lu_fid dst_fid = *rr->rr_fid2;
779 struct mdt_object *dst;
784 LASSERT(fid_is_sane(&dst_fid));
785 dst = mdt_object_find(info->mti_env, info->mti_mdt, &dst_fid);
787 rc = mdo_is_subdir(info->mti_env,
788 mdt_object_child(dst), fid,
790 mdt_object_put(info->mti_env, dst);
791 if (rc != -EREMOTE && rc < 0) {
792 CERROR("Failed mdo_is_subdir(), rc %d\n", rc);
794 /* check the found fid */
795 if (lu_fid_eq(&dst_fid, fid))
801 } while (rc == -EREMOTE);
806 static int mdt_reint_rename(struct mdt_thread_info *info,
807 struct mdt_lock_handle *lhc)
809 struct mdt_reint_record *rr = &info->mti_rr;
810 struct md_attr *ma = &info->mti_attr;
811 struct ptlrpc_request *req = mdt_info_req(info);
812 struct mdt_object *msrcdir;
813 struct mdt_object *mtgtdir;
814 struct mdt_object *mold;
815 struct mdt_object *mnew = NULL;
816 struct mdt_lock_handle *lh_srcdirp;
817 struct mdt_lock_handle *lh_tgtdirp;
818 struct mdt_lock_handle *lh_oldp;
819 struct mdt_lock_handle *lh_newp;
820 struct lu_fid *old_fid = &info->mti_tmp_fid1;
821 struct lu_fid *new_fid = &info->mti_tmp_fid2;
822 struct lustre_handle rename_lh = { 0 };
823 struct lu_name slname = { 0 };
824 struct lu_name *lname;
828 if (info->mti_dlm_req)
829 ldlm_request_cancel(mdt_info_req(info), info->mti_dlm_req, 0);
831 if (info->mti_cross_ref) {
832 rc = mdt_reint_rename_tgt(info);
836 DEBUG_REQ(D_INODE, req, "rename "DFID"/%s to "DFID"/%s",
837 PFID(rr->rr_fid1), rr->rr_name,
838 PFID(rr->rr_fid2), rr->rr_tgt);
840 rc = mdt_rename_lock(info, &rename_lh);
842 CERROR("Can't lock FS for rename, rc %d\n", rc);
846 lh_newp = &info->mti_lh[MDT_LH_NEW];
848 /* step 1: lock the source dir. */
849 lh_srcdirp = &info->mti_lh[MDT_LH_PARENT];
850 mdt_lock_pdo_init(lh_srcdirp, LCK_PW, rr->rr_name,
852 msrcdir = mdt_object_find_lock(info, rr->rr_fid1, lh_srcdirp,
853 MDS_INODELOCK_UPDATE);
855 GOTO(out_rename_lock, rc = PTR_ERR(msrcdir));
857 /* step 2: find & lock the target dir. */
858 lh_tgtdirp = &info->mti_lh[MDT_LH_CHILD];
859 mdt_lock_pdo_init(lh_tgtdirp, LCK_PW, rr->rr_tgt,
861 if (lu_fid_eq(rr->rr_fid1, rr->rr_fid2)) {
862 mdt_object_get(info->mti_env, msrcdir);
865 mtgtdir = mdt_object_find(info->mti_env, info->mti_mdt,
868 GOTO(out_unlock_source, rc = PTR_ERR(mtgtdir));
870 rc = mdt_object_exists(mtgtdir);
872 GOTO(out_unlock_target, rc = -ESTALE);
874 /* we lock the target dir if it is local */
875 rc = mdt_object_lock(info, mtgtdir, lh_tgtdirp,
876 MDS_INODELOCK_UPDATE,
879 GOTO(out_unlock_target, rc);
883 /* step 3: find & lock the old object. */
884 lname = mdt_name(info->mti_env, (char *)rr->rr_name, rr->rr_namelen);
885 mdt_name_copy(&slname, lname);
886 rc = mdo_lookup(info->mti_env, mdt_object_child(msrcdir),
887 &slname, old_fid, &info->mti_spec);
889 GOTO(out_unlock_target, rc);
891 if (lu_fid_eq(old_fid, rr->rr_fid1) || lu_fid_eq(old_fid, rr->rr_fid2))
892 GOTO(out_unlock_target, rc = -EINVAL);
894 mold = mdt_object_find(info->mti_env, info->mti_mdt, old_fid);
896 GOTO(out_unlock_target, rc = PTR_ERR(mold));
898 lh_oldp = &info->mti_lh[MDT_LH_OLD];
899 mdt_lock_reg_init(lh_oldp, LCK_EX);
900 rc = mdt_object_lock(info, mold, lh_oldp, MDS_INODELOCK_LOOKUP,
903 mdt_object_put(info->mti_env, mold);
904 GOTO(out_unlock_target, rc);
906 mdt_set_capainfo(info, 2, old_fid, BYPASS_CAPA);
908 /* step 4: find & lock the new object. */
909 /* new target object may not exist now */
910 lname = mdt_name(info->mti_env, (char *)rr->rr_tgt, rr->rr_tgtlen);
911 rc = mdo_lookup(info->mti_env, mdt_object_child(mtgtdir),
912 lname, new_fid, &info->mti_spec);
914 /* the new_fid should have been filled at this moment */
915 if (lu_fid_eq(old_fid, new_fid))
916 GOTO(out_unlock_old, rc);
918 if (lu_fid_eq(new_fid, rr->rr_fid1) ||
919 lu_fid_eq(new_fid, rr->rr_fid2))
920 GOTO(out_unlock_old, rc = -EINVAL);
922 mdt_lock_reg_init(lh_newp, LCK_EX);
923 mnew = mdt_object_find(info->mti_env, info->mti_mdt, new_fid);
925 GOTO(out_unlock_old, rc = PTR_ERR(mnew));
927 rc = mdt_object_lock(info, mnew, lh_newp,
928 MDS_INODELOCK_FULL, MDT_CROSS_LOCK);
930 mdt_object_put(info->mti_env, mnew);
931 GOTO(out_unlock_old, rc);
933 mdt_set_capainfo(info, 3, new_fid, BYPASS_CAPA);
934 } else if (rc != -EREMOTE && rc != -ENOENT)
935 GOTO(out_unlock_old, rc);
937 /* step 5: rename it */
938 mdt_reint_init_ma(info, ma);
939 if (!ma->ma_lmm || !ma->ma_cookie)
940 GOTO(out_unlock_new, rc = -EINVAL);
942 mdt_fail_write(info->mti_env, info->mti_mdt->mdt_bottom,
943 OBD_FAIL_MDS_REINT_RENAME_WRITE);
946 /* Check if @dst is subdir of @src. */
947 rc = mdt_rename_sanity(info, old_fid);
949 GOTO(out_unlock_new, rc);
951 rc = mdo_rename(info->mti_env, mdt_object_child(msrcdir),
952 mdt_object_child(mtgtdir), old_fid, &slname,
953 (mnew ? mdt_object_child(mnew) : NULL),
956 /* handle last link of tgt object */
958 mdt_handle_last_unlink(info, mnew, ma);
963 mdt_object_unlock_put(info, mnew, lh_newp, rc);
965 mdt_object_unlock_put(info, mold, lh_oldp, rc);
967 mdt_object_unlock_put(info, mtgtdir, lh_tgtdirp, rc);
969 mdt_object_unlock_put(info, msrcdir, lh_srcdirp, rc);
971 mdt_rename_unlock(&rename_lh);
973 mdt_shrink_reply(info);
977 typedef int (*mdt_reinter)(struct mdt_thread_info *info,
978 struct mdt_lock_handle *lhc);
980 static mdt_reinter reinters[REINT_MAX] = {
981 [REINT_SETATTR] = mdt_reint_setattr,
982 [REINT_CREATE] = mdt_reint_create,
983 [REINT_LINK] = mdt_reint_link,
984 [REINT_UNLINK] = mdt_reint_unlink,
985 [REINT_RENAME] = mdt_reint_rename,
986 [REINT_OPEN] = mdt_reint_open
989 int mdt_reint_rec(struct mdt_thread_info *info,
990 struct mdt_lock_handle *lhc)
995 rc = reinters[info->mti_rr.rr_opcode](info, lhc);