1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
6 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
8 * This program is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License version 2 only,
10 * as published by the Free Software Foundation.
12 * This program is distributed in the hope that it will be useful, but
13 * WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * General Public License version 2 for more details (a copy is included
16 * in the LICENSE file that accompanied this code).
18 * You should have received a copy of the GNU General Public License
19 * version 2 along with this program; If not, see
20 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
22 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
23 * CA 95054 USA or visit www.sun.com if you need additional information or
29 * Copyright 2008 Sun Microsystems, Inc. All rights reserved
30 * Use is subject to license terms.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * lustre/mdt/mdt_reint.c
38 * Lustre Metadata Target (mdt) reintegration routines
40 * Author: Peter Braam <braam@clusterfs.com>
41 * Author: Andreas Dilger <adilger@clusterfs.com>
42 * Author: Phil Schwan <phil@clusterfs.com>
43 * Author: Huang Hua <huanghua@clusterfs.com>
44 * Author: Yury Umanets <umka@clusterfs.com>
48 # define EXPORT_SYMTAB
50 #define DEBUG_SUBSYSTEM S_MDS
52 #include "mdt_internal.h"
55 static inline void mdt_reint_init_ma(struct mdt_thread_info *info,
58 ma->ma_lmm = req_capsule_server_get(info->mti_pill, &RMF_MDT_MD);
59 ma->ma_lmm_size = req_capsule_get_size(info->mti_pill,
60 &RMF_MDT_MD, RCL_SERVER);
62 ma->ma_cookie = req_capsule_server_get(info->mti_pill,
64 ma->ma_cookie_size = req_capsule_get_size(info->mti_pill,
68 ma->ma_need = MA_INODE | MA_LOV | MA_COOKIE;
72 static int mdt_create_pack_capa(struct mdt_thread_info *info, int rc,
73 struct mdt_object *object,
74 struct mdt_body *repbody)
78 /* for cross-ref mkdir, mds capa has been fetched from remote obj, then
79 * we won't go to below*/
80 if (repbody->valid & OBD_MD_FLMDSCAPA)
83 if (rc == 0 && info->mti_mdt->mdt_opts.mo_mds_capa) {
84 struct lustre_capa *capa;
86 capa = req_capsule_server_get(info->mti_pill, &RMF_CAPA1);
88 capa->lc_opc = CAPA_OPC_MDS_DEFAULT;
89 rc = mo_capa_get(info->mti_env, mdt_object_child(object), capa,
92 repbody->valid |= OBD_MD_FLMDSCAPA;
98 static int mdt_md_create(struct mdt_thread_info *info)
100 struct mdt_device *mdt = info->mti_mdt;
101 struct mdt_object *parent;
102 struct mdt_object *child;
103 struct mdt_lock_handle *lh;
104 struct mdt_body *repbody;
105 struct md_attr *ma = &info->mti_attr;
106 struct mdt_reint_record *rr = &info->mti_rr;
107 struct lu_name *lname;
111 DEBUG_REQ(D_INODE, mdt_info_req(info), "Create (%s->"DFID") in "DFID,
112 rr->rr_name, PFID(rr->rr_fid2), PFID(rr->rr_fid1));
114 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
116 lh = &info->mti_lh[MDT_LH_PARENT];
117 mdt_lock_pdo_init(lh, LCK_PW, rr->rr_name, rr->rr_namelen);
119 parent = mdt_object_find_lock(info, rr->rr_fid1, lh,
120 MDS_INODELOCK_UPDATE);
122 RETURN(PTR_ERR(parent));
124 child = mdt_object_find(info->mti_env, mdt, rr->rr_fid2);
125 if (likely(!IS_ERR(child))) {
126 struct md_object *next = mdt_object_child(parent);
128 ma->ma_need = MA_INODE;
130 /* capa for cross-ref will be stored here */
131 ma->ma_capa = req_capsule_server_get(info->mti_pill,
133 LASSERT(ma->ma_capa);
135 mdt_fail_write(info->mti_env, info->mti_mdt->mdt_bottom,
136 OBD_FAIL_MDS_REINT_CREATE_WRITE);
138 /* Let lower layer know current lock mode. */
139 info->mti_spec.sp_cr_mode =
140 mdt_dlm_mode2mdl_mode(lh->mlh_pdo_mode);
143 * Do perform lookup sanity check. We do not know if name exists
146 info->mti_spec.sp_cr_lookup = 1;
148 lname = mdt_name(info->mti_env, (char *)rr->rr_name,
150 rc = mdo_create(info->mti_env, next, lname,
151 mdt_object_child(child),
152 &info->mti_spec, ma);
154 /* Return fid & attr to client. */
155 if (ma->ma_valid & MA_INODE)
156 mdt_pack_attr2body(info, repbody, &ma->ma_attr,
157 mdt_object_fid(child));
159 mdt_object_put(info->mti_env, child);
163 mdt_create_pack_capa(info, rc, child, repbody);
164 mdt_object_unlock_put(info, parent, lh, rc);
168 /* Partial request to create object only */
169 static int mdt_md_mkobj(struct mdt_thread_info *info)
171 struct mdt_device *mdt = info->mti_mdt;
172 struct mdt_object *o;
173 struct mdt_body *repbody;
174 struct md_attr *ma = &info->mti_attr;
178 DEBUG_REQ(D_INODE, mdt_info_req(info), "Partial create "DFID"",
179 PFID(info->mti_rr.rr_fid2));
181 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
183 o = mdt_object_find(info->mti_env, mdt, info->mti_rr.rr_fid2);
185 struct md_object *next = mdt_object_child(o);
187 ma->ma_need = MA_INODE;
191 * Cross-ref create can encounter already created obj in case of
192 * recovery, just get attr in that case.
194 if (mdt_object_exists(o) == 1) {
195 rc = mo_attr_get(info->mti_env, next, ma);
198 * Here, NO permission check for object_create,
199 * such check has been done on the original MDS.
201 rc = mo_object_create(info->mti_env, next,
202 &info->mti_spec, ma);
205 /* Return fid & attr to client. */
206 if (ma->ma_valid & MA_INODE)
207 mdt_pack_attr2body(info, repbody, &ma->ma_attr,
210 mdt_object_put(info->mti_env, o);
214 mdt_create_pack_capa(info, rc, o, repbody);
218 /* In the raw-setattr case, we lock the child inode.
219 * In the write-back case or if being called from open,
220 * the client holds a lock already.
221 * We use the ATTR_FROM_OPEN (translated into MRF_SETATTR_LOCKED by
222 * mdt_setattr_unpack()) flag to tell these cases apart. */
223 int mdt_attr_set(struct mdt_thread_info *info, struct mdt_object *mo, int flags)
225 struct md_attr *ma = &info->mti_attr;
226 struct mdt_lock_handle *lh;
231 /* attr shouldn't be set on remote object */
232 LASSERT(mdt_object_exists(mo) >= 0);
235 som_update = (info->mti_epoch->flags & MF_SOM_CHANGE);
237 /* Try to avoid object_lock if another epoch has been started
239 if (som_update && (info->mti_epoch->ioepoch != mo->mot_ioepoch))
242 lh = &info->mti_lh[MDT_LH_PARENT];
243 mdt_lock_reg_init(lh, LCK_PW);
245 if (!(flags & MRF_SETATTR_LOCKED)) {
246 __u64 lockpart = MDS_INODELOCK_UPDATE;
247 if (ma->ma_attr.la_valid & (LA_MODE|LA_UID|LA_GID))
248 lockpart |= MDS_INODELOCK_LOOKUP;
250 rc = mdt_object_lock(info, mo, lh, lockpart, MDT_LOCAL_LOCK);
255 /* Setattrs are syncronized through dlm lock taken above. If another
256 * epoch started, its attributes may be already flushed on disk,
258 if (som_update && (info->mti_epoch->ioepoch != mo->mot_ioepoch))
259 GOTO(out_unlock, rc = 0);
261 if (mdt_object_exists(mo) == 0)
262 GOTO(out_unlock, rc = -ENOENT);
264 /* all attrs are packed into mti_attr in unpack_setattr */
265 mdt_fail_write(info->mti_env, info->mti_mdt->mdt_bottom,
266 OBD_FAIL_MDS_REINT_SETATTR_WRITE);
268 /* This is only for set ctime when rename's source is on remote MDS. */
269 if (unlikely(ma->ma_attr.la_valid == LA_CTIME))
270 ma->ma_attr_flags |= MDS_VTX_BYPASS;
272 /* all attrs are packed into mti_attr in unpack_setattr */
273 rc = mo_attr_set(info->mti_env, mdt_object_child(mo), ma);
275 GOTO(out_unlock, rc);
277 /* Re-enable SIZEONMDS. */
279 CDEBUG(D_INODE, "Closing epoch "LPU64" on "DFID". Count %d\n",
280 mo->mot_ioepoch, PFID(mdt_object_fid(mo)),
282 mdt_sizeonmds_enable(info, mo);
287 mdt_object_unlock(info, mo, lh, rc);
291 static int mdt_reint_setattr(struct mdt_thread_info *info,
292 struct mdt_lock_handle *lhc)
294 struct mdt_device *mdt = info->mti_mdt;
295 struct md_attr *ma = &info->mti_attr;
296 struct mdt_reint_record *rr = &info->mti_rr;
297 struct ptlrpc_request *req = mdt_info_req(info);
298 struct mdt_export_data *med = &req->rq_export->exp_mdt_data;
299 struct mdt_file_data *mfd;
300 struct mdt_object *mo;
301 struct md_object *next;
302 struct mdt_body *repbody;
306 DEBUG_REQ(D_INODE, req, "setattr "DFID" %x", PFID(rr->rr_fid1),
307 (unsigned int)ma->ma_attr.la_valid);
309 if (info->mti_dlm_req)
310 ldlm_request_cancel(req, info->mti_dlm_req, 0);
312 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
313 mo = mdt_object_find(info->mti_env, info->mti_mdt, rr->rr_fid1);
315 GOTO(out, rc = PTR_ERR(mo));
317 if (!(mdt_conn_flags(info) & OBD_CONNECT_SOM)) {
318 if ((ma->ma_attr.la_valid & LA_SIZE) ||
319 (rr->rr_flags & MRF_SETATTR_LOCKED)) {
320 /* Check write access for the O_TRUNC case */
321 if (mdt_write_read(info->mti_mdt, mo) < 0)
322 GOTO(out_put, rc = -ETXTBSY);
324 } else if (info->mti_epoch &&
325 (info->mti_epoch->flags & MF_EPOCH_OPEN)) {
327 rc = mdt_write_get(info->mti_mdt, mo);
333 GOTO(out_put, rc = -ENOMEM);
335 mdt_epoch_open(info, mo);
336 repbody->ioepoch = mo->mot_ioepoch;
338 mdt_object_get(info->mti_env, mo);
339 mdt_mfd_set_mode(mfd, FMODE_EPOCHLCK);
340 mfd->mfd_object = mo;
341 mfd->mfd_xid = req->rq_xid;
343 spin_lock(&med->med_open_lock);
344 list_add(&mfd->mfd_list, &med->med_open_head);
345 spin_unlock(&med->med_open_lock);
346 repbody->handle.cookie = mfd->mfd_handle.h_cookie;
349 if (info->mti_epoch && (info->mti_epoch->flags & MF_SOM_CHANGE))
350 ma->ma_attr_flags |= MDS_PERM_BYPASS | MDS_SOM;
352 rc = mdt_attr_set(info, mo, rr->rr_flags);
356 if (info->mti_epoch && (info->mti_epoch->flags & MF_SOM_CHANGE)) {
357 LASSERT(mdt_conn_flags(info) & OBD_CONNECT_SOM);
358 LASSERT(info->mti_epoch);
360 spin_lock(&med->med_open_lock);
361 /* Size-on-MDS Update. Find and free mfd. */
362 mfd = mdt_handle2mfd(info, &info->mti_epoch->handle);
364 spin_unlock(&med->med_open_lock);
365 CDEBUG(D_INODE | D_ERROR, "no handle for file close: "
366 "fid = "DFID": cookie = "LPX64"\n",
367 PFID(info->mti_rr.rr_fid1),
368 info->mti_epoch->handle.cookie);
369 GOTO(out_put, rc = -ESTALE);
371 LASSERT(mfd->mfd_mode == FMODE_SOM);
372 LASSERT(!(info->mti_epoch->flags & MF_EPOCH_CLOSE));
374 class_handle_unhash(&mfd->mfd_handle);
375 list_del_init(&mfd->mfd_list);
376 spin_unlock(&med->med_open_lock);
378 mdt_mfd_close(info, mfd);
381 ma->ma_need = MA_INODE;
383 next = mdt_object_child(mo);
384 rc = mo_attr_get(info->mti_env, next, ma);
388 mdt_pack_attr2body(info, repbody, &ma->ma_attr, mdt_object_fid(mo));
390 if (mdt->mdt_opts.mo_oss_capa &&
391 S_ISREG(lu_object_attr(&mo->mot_obj.mo_lu)) &&
392 (ma->ma_attr.la_valid & LA_SIZE)) {
393 struct lustre_capa *capa;
395 capa = req_capsule_server_get(info->mti_pill, &RMF_CAPA2);
397 capa->lc_opc = CAPA_OPC_OSS_DEFAULT | CAPA_OPC_OSS_TRUNC;
398 rc = mo_capa_get(info->mti_env, mdt_object_child(mo), capa, 0);
401 repbody->valid |= OBD_MD_FLOSSCAPA;
406 mdt_object_put(info->mti_env, mo);
408 mdt_shrink_reply(info);
412 static int mdt_reint_create(struct mdt_thread_info *info,
413 struct mdt_lock_handle *lhc)
418 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_REINT_CREATE))
419 RETURN(err_serious(-ESTALE));
421 if (info->mti_dlm_req)
422 ldlm_request_cancel(mdt_info_req(info), info->mti_dlm_req, 0);
424 switch (info->mti_attr.ma_attr.la_mode & S_IFMT) {
426 /* Cross-ref case. */
427 if (info->mti_cross_ref) {
428 rc = mdt_md_mkobj(info);
438 /* Special file should stay on the same node as parent. */
439 LASSERT(info->mti_rr.rr_namelen > 0);
440 rc = mdt_md_create(info);
444 rc = err_serious(-EOPNOTSUPP);
449 static int mdt_reint_unlink(struct mdt_thread_info *info,
450 struct mdt_lock_handle *lhc)
452 struct mdt_reint_record *rr = &info->mti_rr;
453 struct ptlrpc_request *req = mdt_info_req(info);
454 struct md_attr *ma = &info->mti_attr;
455 struct lu_fid *child_fid = &info->mti_tmp_fid1;
456 struct mdt_object *mp;
457 struct mdt_object *mc;
458 struct mdt_lock_handle *parent_lh;
459 struct mdt_lock_handle *child_lh;
460 struct lu_name *lname;
464 DEBUG_REQ(D_INODE, req, "unlink "DFID"/%s", PFID(rr->rr_fid1),
467 if (info->mti_dlm_req)
468 ldlm_request_cancel(req, info->mti_dlm_req, 0);
470 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_REINT_UNLINK))
471 RETURN(err_serious(-ENOENT));
473 /* step 1: lock the parent */
474 parent_lh = &info->mti_lh[MDT_LH_PARENT];
475 mdt_lock_pdo_init(parent_lh, LCK_PW, rr->rr_name,
478 mp = mdt_object_find_lock(info, rr->rr_fid1, parent_lh,
479 MDS_INODELOCK_UPDATE);
482 /* errors are possible here in cross-ref cases, see below */
483 if (info->mti_cross_ref)
488 mdt_reint_init_ma(info, ma);
489 if (!ma->ma_lmm || !ma->ma_cookie)
490 GOTO(out_unlock_parent, rc = -EINVAL);
492 if (info->mti_cross_ref) {
494 * Remote partial operation. It is possible that replay may
495 * happen on parent MDT and this operation will be repeated.
496 * Therefore the object absense is allowed case and nothing
497 * should be done here.
499 if (mdt_object_exists(mp) > 0) {
500 mdt_set_capainfo(info, 0, rr->rr_fid1, BYPASS_CAPA);
501 rc = mo_ref_del(info->mti_env,
502 mdt_object_child(mp), ma);
503 mdt_handle_last_unlink(info, mp, ma);
506 GOTO(out_unlock_parent, rc);
509 /* step 2: find & lock the child */
510 lname = mdt_name(info->mti_env, (char *)rr->rr_name, rr->rr_namelen);
511 rc = mdo_lookup(info->mti_env, mdt_object_child(mp),
512 lname, child_fid, &info->mti_spec);
514 GOTO(out_unlock_parent, rc);
516 /* We will lock the child regardless it is local or remote. No harm. */
517 mc = mdt_object_find(info->mti_env, info->mti_mdt, child_fid);
519 GOTO(out_unlock_parent, rc = PTR_ERR(mc));
520 child_lh = &info->mti_lh[MDT_LH_CHILD];
521 mdt_lock_reg_init(child_lh, LCK_EX);
522 rc = mdt_object_lock(info, mc, child_lh, MDS_INODELOCK_FULL,
525 mdt_object_put(info->mti_env, mc);
526 GOTO(out_unlock_parent, rc);
529 mdt_fail_write(info->mti_env, info->mti_mdt->mdt_bottom,
530 OBD_FAIL_MDS_REINT_UNLINK_WRITE);
533 * Now we can only make sure we need MA_INODE, in mdd layer, will check
534 * whether need MA_LOV and MA_COOKIE.
536 ma->ma_need = MA_INODE;
538 mdt_set_capainfo(info, 1, child_fid, BYPASS_CAPA);
539 rc = mdo_unlink(info->mti_env, mdt_object_child(mp),
540 mdt_object_child(mc), lname, ma);
542 mdt_handle_last_unlink(info, mc, ma);
545 mdt_object_unlock_put(info, mc, child_lh, rc);
547 mdt_object_unlock_put(info, mp, parent_lh, rc);
552 static int mdt_reint_link(struct mdt_thread_info *info,
553 struct mdt_lock_handle *lhc)
555 struct mdt_reint_record *rr = &info->mti_rr;
556 struct ptlrpc_request *req = mdt_info_req(info);
557 struct md_attr *ma = &info->mti_attr;
558 struct mdt_object *ms;
559 struct mdt_object *mp;
560 struct mdt_lock_handle *lhs;
561 struct mdt_lock_handle *lhp;
562 struct lu_name *lname;
566 DEBUG_REQ(D_INODE, req, "link "DFID" to "DFID"/%s",
567 PFID(rr->rr_fid1), PFID(rr->rr_fid2), rr->rr_name);
569 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_REINT_LINK))
570 RETURN(err_serious(-ENOENT));
572 if (info->mti_dlm_req)
573 ldlm_request_cancel(req, info->mti_dlm_req, 0);
575 if (info->mti_cross_ref) {
576 /* MDT holding name ask us to add ref. */
577 lhs = &info->mti_lh[MDT_LH_CHILD];
578 mdt_lock_reg_init(lhs, LCK_EX);
579 ms = mdt_object_find_lock(info, rr->rr_fid1, lhs,
580 MDS_INODELOCK_UPDATE);
584 mdt_set_capainfo(info, 0, rr->rr_fid1, BYPASS_CAPA);
585 rc = mo_ref_add(info->mti_env, mdt_object_child(ms), ma);
586 mdt_object_unlock_put(info, ms, lhs, rc);
590 /* Invalid case so return error immediately instead of
592 if (lu_fid_eq(rr->rr_fid1, rr->rr_fid2))
595 /* step 1: find & lock the target parent dir */
596 lhp = &info->mti_lh[MDT_LH_PARENT];
597 mdt_lock_pdo_init(lhp, LCK_EX, rr->rr_name,
599 mp = mdt_object_find_lock(info, rr->rr_fid2, lhp,
600 MDS_INODELOCK_UPDATE);
604 /* step 2: find & lock the source */
605 lhs = &info->mti_lh[MDT_LH_CHILD];
606 mdt_lock_reg_init(lhs, LCK_EX);
608 ms = mdt_object_find(info->mti_env, info->mti_mdt, rr->rr_fid1);
610 GOTO(out_unlock_parent, rc = PTR_ERR(ms));
612 rc = mdt_object_lock(info, ms, lhs, MDS_INODELOCK_UPDATE,
615 mdt_object_put(info->mti_env, ms);
616 GOTO(out_unlock_parent, rc);
619 /* step 3: link it */
620 mdt_fail_write(info->mti_env, info->mti_mdt->mdt_bottom,
621 OBD_FAIL_MDS_REINT_LINK_WRITE);
623 lname = mdt_name(info->mti_env, (char *)rr->rr_name, rr->rr_namelen);
624 rc = mdo_link(info->mti_env, mdt_object_child(mp),
625 mdt_object_child(ms), lname, ma);
628 mdt_object_unlock_put(info, ms, lhs, rc);
630 mdt_object_unlock_put(info, mp, lhp, rc);
634 /* partial operation for rename */
635 static int mdt_reint_rename_tgt(struct mdt_thread_info *info)
637 struct mdt_reint_record *rr = &info->mti_rr;
638 struct ptlrpc_request *req = mdt_info_req(info);
639 struct md_attr *ma = &info->mti_attr;
640 struct mdt_object *mtgtdir;
641 struct mdt_object *mtgt = NULL;
642 struct mdt_lock_handle *lh_tgtdir;
643 struct mdt_lock_handle *lh_tgt = NULL;
644 struct lu_fid *tgt_fid = &info->mti_tmp_fid1;
645 struct lu_name *lname;
649 DEBUG_REQ(D_INODE, req, "rename_tgt: insert (%s->"DFID") in "DFID,
650 rr->rr_tgt, PFID(rr->rr_fid2), PFID(rr->rr_fid1));
652 /* step 1: lookup & lock the tgt dir. */
653 lh_tgtdir = &info->mti_lh[MDT_LH_PARENT];
654 mdt_lock_pdo_init(lh_tgtdir, LCK_PW, rr->rr_tgt,
656 mtgtdir = mdt_object_find_lock(info, rr->rr_fid1, lh_tgtdir,
657 MDS_INODELOCK_UPDATE);
659 RETURN(PTR_ERR(mtgtdir));
661 /* step 2: find & lock the target object if exists. */
662 mdt_set_capainfo(info, 0, rr->rr_fid1, BYPASS_CAPA);
663 lname = mdt_name(info->mti_env, (char *)rr->rr_tgt, rr->rr_tgtlen);
664 rc = mdo_lookup(info->mti_env, mdt_object_child(mtgtdir),
665 lname, tgt_fid, &info->mti_spec);
666 if (rc != 0 && rc != -ENOENT) {
667 GOTO(out_unlock_tgtdir, rc);
668 } else if (rc == 0) {
670 * In case of replay that name can be already inserted, check
671 * that and do nothing if so.
673 if (lu_fid_eq(tgt_fid, rr->rr_fid2))
674 GOTO(out_unlock_tgtdir, rc);
676 lh_tgt = &info->mti_lh[MDT_LH_CHILD];
677 mdt_lock_reg_init(lh_tgt, LCK_EX);
679 mtgt = mdt_object_find_lock(info, tgt_fid, lh_tgt,
680 MDS_INODELOCK_LOOKUP);
682 GOTO(out_unlock_tgtdir, rc = PTR_ERR(mtgt));
684 mdt_reint_init_ma(info, ma);
685 if (!ma->ma_lmm || !ma->ma_cookie)
686 GOTO(out_unlock_tgt, rc = -EINVAL);
688 rc = mdo_rename_tgt(info->mti_env, mdt_object_child(mtgtdir),
689 mdt_object_child(mtgt), rr->rr_fid2,
691 } else /* -ENOENT */ {
692 rc = mdo_name_insert(info->mti_env, mdt_object_child(mtgtdir),
693 lname, rr->rr_fid2, ma);
696 /* handle last link of tgt object */
698 mdt_handle_last_unlink(info, mtgt, ma);
703 mdt_object_unlock_put(info, mtgt, lh_tgt, rc);
705 mdt_object_unlock_put(info, mtgtdir, lh_tgtdir, rc);
709 static int mdt_rename_lock(struct mdt_thread_info *info,
710 struct lustre_handle *lh)
712 struct ldlm_namespace *ns = info->mti_mdt->mdt_namespace;
713 ldlm_policy_data_t *policy = &info->mti_policy;
714 struct ldlm_res_id *res_id = &info->mti_res_id;
720 * Disable global rename BFL lock temporarily because
721 * when a mds do rename recoverying, which might enqueue
722 * BFL lock to the controller mds. and this req might be
723 * replay req for controller mds. but we did not have
724 * such handling in controller mds. XXX
727 ls = info->mti_mdt->mdt_md_dev.md_lu_dev.ld_site;
728 fid_build_reg_res_name(&LUSTRE_BFL_FID, res_id);
730 memset(policy, 0, sizeof *policy);
731 policy->l_inodebits.bits = MDS_INODELOCK_UPDATE;
733 if (ls->ls_control_exp == NULL) {
734 int flags = LDLM_FL_LOCAL_ONLY | LDLM_FL_ATOMIC_CB;
737 * Current node is controller, that is mdt0, where we should
740 rc = ldlm_cli_enqueue_local(ns, res_id, LDLM_IBITS, policy,
741 LCK_EX, &flags, ldlm_blocking_ast,
742 ldlm_completion_ast, NULL, NULL, 0,
745 struct ldlm_enqueue_info einfo = { LDLM_IBITS, LCK_EX,
746 ldlm_blocking_ast, ldlm_completion_ast, NULL, NULL };
750 * This is the case mdt0 is remote node, issue DLM lock like
753 rc = ldlm_cli_enqueue(ls->ls_control_exp, NULL, &einfo, res_id,
754 policy, &flags, NULL, 0, NULL, lh, 0);
760 static void mdt_rename_unlock(struct lustre_handle *lh)
763 /* Disable global rename BFL lock temporarily. see above XXX*/
766 LASSERT(lustre_handle_is_used(lh));
767 ldlm_lock_decref(lh, LCK_EX);
772 * This is is_subdir() variant, it is CMD if cmm forwards it to correct
773 * target. Source should not be ancestor of target dir. May be other rename
774 * checks can be moved here later.
776 static int mdt_rename_sanity(struct mdt_thread_info *info, struct lu_fid *fid)
778 struct mdt_reint_record *rr = &info->mti_rr;
779 struct lu_fid dst_fid = *rr->rr_fid2;
780 struct mdt_object *dst;
785 LASSERT(fid_is_sane(&dst_fid));
786 dst = mdt_object_find(info->mti_env, info->mti_mdt, &dst_fid);
788 rc = mdo_is_subdir(info->mti_env,
789 mdt_object_child(dst), fid,
791 mdt_object_put(info->mti_env, dst);
792 if (rc != -EREMOTE && rc < 0) {
793 CERROR("Failed mdo_is_subdir(), rc %d\n", rc);
795 /* check the found fid */
796 if (lu_fid_eq(&dst_fid, fid))
802 } while (rc == -EREMOTE);
807 static int mdt_reint_rename(struct mdt_thread_info *info,
808 struct mdt_lock_handle *lhc)
810 struct mdt_reint_record *rr = &info->mti_rr;
811 struct md_attr *ma = &info->mti_attr;
812 struct ptlrpc_request *req = mdt_info_req(info);
813 struct mdt_object *msrcdir;
814 struct mdt_object *mtgtdir;
815 struct mdt_object *mold;
816 struct mdt_object *mnew = NULL;
817 struct mdt_lock_handle *lh_srcdirp;
818 struct mdt_lock_handle *lh_tgtdirp;
819 struct mdt_lock_handle *lh_oldp;
820 struct mdt_lock_handle *lh_newp;
821 struct lu_fid *old_fid = &info->mti_tmp_fid1;
822 struct lu_fid *new_fid = &info->mti_tmp_fid2;
823 struct lustre_handle rename_lh = { 0 };
824 struct lu_name slname = { 0 };
825 struct lu_name *lname;
829 if (info->mti_dlm_req)
830 ldlm_request_cancel(mdt_info_req(info), info->mti_dlm_req, 0);
832 if (info->mti_cross_ref) {
833 rc = mdt_reint_rename_tgt(info);
837 DEBUG_REQ(D_INODE, req, "rename "DFID"/%s to "DFID"/%s",
838 PFID(rr->rr_fid1), rr->rr_name,
839 PFID(rr->rr_fid2), rr->rr_tgt);
841 rc = mdt_rename_lock(info, &rename_lh);
843 CERROR("Can't lock FS for rename, rc %d\n", rc);
847 lh_newp = &info->mti_lh[MDT_LH_NEW];
849 /* step 1: lock the source dir. */
850 lh_srcdirp = &info->mti_lh[MDT_LH_PARENT];
851 mdt_lock_pdo_init(lh_srcdirp, LCK_PW, rr->rr_name,
853 msrcdir = mdt_object_find_lock(info, rr->rr_fid1, lh_srcdirp,
854 MDS_INODELOCK_UPDATE);
856 GOTO(out_rename_lock, rc = PTR_ERR(msrcdir));
858 /* step 2: find & lock the target dir. */
859 lh_tgtdirp = &info->mti_lh[MDT_LH_CHILD];
860 mdt_lock_pdo_init(lh_tgtdirp, LCK_PW, rr->rr_tgt,
862 if (lu_fid_eq(rr->rr_fid1, rr->rr_fid2)) {
863 mdt_object_get(info->mti_env, msrcdir);
866 mtgtdir = mdt_object_find(info->mti_env, info->mti_mdt,
869 GOTO(out_unlock_source, rc = PTR_ERR(mtgtdir));
871 rc = mdt_object_exists(mtgtdir);
873 GOTO(out_unlock_target, rc = -ESTALE);
875 /* we lock the target dir if it is local */
876 rc = mdt_object_lock(info, mtgtdir, lh_tgtdirp,
877 MDS_INODELOCK_UPDATE,
880 GOTO(out_unlock_target, rc);
884 /* step 3: find & lock the old object. */
885 lname = mdt_name(info->mti_env, (char *)rr->rr_name, rr->rr_namelen);
886 mdt_name_copy(&slname, lname);
887 rc = mdo_lookup(info->mti_env, mdt_object_child(msrcdir),
888 &slname, old_fid, &info->mti_spec);
890 GOTO(out_unlock_target, rc);
892 if (lu_fid_eq(old_fid, rr->rr_fid1) || lu_fid_eq(old_fid, rr->rr_fid2))
893 GOTO(out_unlock_target, rc = -EINVAL);
895 mold = mdt_object_find(info->mti_env, info->mti_mdt, old_fid);
897 GOTO(out_unlock_target, rc = PTR_ERR(mold));
899 lh_oldp = &info->mti_lh[MDT_LH_OLD];
900 mdt_lock_reg_init(lh_oldp, LCK_EX);
901 rc = mdt_object_lock(info, mold, lh_oldp, MDS_INODELOCK_LOOKUP,
904 mdt_object_put(info->mti_env, mold);
905 GOTO(out_unlock_target, rc);
907 mdt_set_capainfo(info, 2, old_fid, BYPASS_CAPA);
909 /* step 4: find & lock the new object. */
910 /* new target object may not exist now */
911 lname = mdt_name(info->mti_env, (char *)rr->rr_tgt, rr->rr_tgtlen);
912 rc = mdo_lookup(info->mti_env, mdt_object_child(mtgtdir),
913 lname, new_fid, &info->mti_spec);
915 /* the new_fid should have been filled at this moment */
916 if (lu_fid_eq(old_fid, new_fid))
917 GOTO(out_unlock_old, rc);
919 if (lu_fid_eq(new_fid, rr->rr_fid1) ||
920 lu_fid_eq(new_fid, rr->rr_fid2))
921 GOTO(out_unlock_old, rc = -EINVAL);
923 mdt_lock_reg_init(lh_newp, LCK_EX);
924 mnew = mdt_object_find(info->mti_env, info->mti_mdt, new_fid);
926 GOTO(out_unlock_old, rc = PTR_ERR(mnew));
928 rc = mdt_object_lock(info, mnew, lh_newp,
929 MDS_INODELOCK_FULL, MDT_CROSS_LOCK);
931 mdt_object_put(info->mti_env, mnew);
932 GOTO(out_unlock_old, rc);
934 mdt_set_capainfo(info, 3, new_fid, BYPASS_CAPA);
935 } else if (rc != -EREMOTE && rc != -ENOENT)
936 GOTO(out_unlock_old, rc);
938 /* step 5: rename it */
939 mdt_reint_init_ma(info, ma);
940 if (!ma->ma_lmm || !ma->ma_cookie)
941 GOTO(out_unlock_new, rc = -EINVAL);
943 mdt_fail_write(info->mti_env, info->mti_mdt->mdt_bottom,
944 OBD_FAIL_MDS_REINT_RENAME_WRITE);
947 /* Check if @dst is subdir of @src. */
948 rc = mdt_rename_sanity(info, old_fid);
950 GOTO(out_unlock_new, rc);
952 rc = mdo_rename(info->mti_env, mdt_object_child(msrcdir),
953 mdt_object_child(mtgtdir), old_fid, &slname,
954 (mnew ? mdt_object_child(mnew) : NULL),
957 /* handle last link of tgt object */
959 mdt_handle_last_unlink(info, mnew, ma);
964 mdt_object_unlock_put(info, mnew, lh_newp, rc);
966 mdt_object_unlock_put(info, mold, lh_oldp, rc);
968 mdt_object_unlock_put(info, mtgtdir, lh_tgtdirp, rc);
970 mdt_object_unlock_put(info, msrcdir, lh_srcdirp, rc);
972 mdt_rename_unlock(&rename_lh);
976 typedef int (*mdt_reinter)(struct mdt_thread_info *info,
977 struct mdt_lock_handle *lhc);
979 static mdt_reinter reinters[REINT_MAX] = {
980 [REINT_SETATTR] = mdt_reint_setattr,
981 [REINT_CREATE] = mdt_reint_create,
982 [REINT_LINK] = mdt_reint_link,
983 [REINT_UNLINK] = mdt_reint_unlink,
984 [REINT_RENAME] = mdt_reint_rename,
985 [REINT_OPEN] = mdt_reint_open,
986 [REINT_SETXATTR] = mdt_reint_setxattr
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);