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_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>
47 #define DEBUG_SUBSYSTEM S_MDS
49 #include "mdt_internal.h"
51 static inline void mdt_reint_init_ma(struct mdt_thread_info *info,
54 ma->ma_need = MA_INODE;
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_lut.lut_mds_capa &&
70 exp_connect_flags(info->mti_exp) & OBD_CONNECT_MDS_CAPA) {
71 struct lustre_capa *capa;
73 capa = req_capsule_server_get(info->mti_pill, &RMF_CAPA1);
75 capa->lc_opc = CAPA_OPC_MDS_DEFAULT;
76 rc = mo_capa_get(info->mti_env, mdt_object_child(object), capa,
79 repbody->valid |= OBD_MD_FLMDSCAPA;
86 * Get version of object by fid.
88 * Return real version or ENOENT_VERSION if object doesn't exist
90 static void mdt_obj_version_get(struct mdt_thread_info *info,
91 struct mdt_object *o, __u64 *version)
94 if (mdt_object_exists(o) && !mdt_object_remote(o) &&
95 !fid_is_obf(mdt_object_fid(o)))
96 *version = dt_version_get(info->mti_env, mdt_obj2dt(o));
98 *version = ENOENT_VERSION;
99 CDEBUG(D_INODE, "FID "DFID" version is "LPX64"\n",
100 PFID(mdt_object_fid(o)), *version);
104 * Check version is correct.
106 * Should be called only during replay.
108 static int mdt_version_check(struct ptlrpc_request *req,
109 __u64 version, int idx)
111 __u64 *pre_ver = lustre_msg_get_versions(req->rq_reqmsg);
114 if (!exp_connect_vbr(req->rq_export))
117 LASSERT(req_is_replay(req));
118 /** VBR: version is checked always because costs nothing */
119 LASSERT(idx < PTLRPC_NUM_VERSIONS);
120 /** Sanity check for malformed buffers */
121 if (pre_ver == NULL) {
122 CERROR("No versions in request buffer\n");
123 spin_lock(&req->rq_export->exp_lock);
124 req->rq_export->exp_vbr_failed = 1;
125 spin_unlock(&req->rq_export->exp_lock);
127 } else if (pre_ver[idx] != version) {
128 CDEBUG(D_INODE, "Version mismatch "LPX64" != "LPX64"\n",
129 pre_ver[idx], version);
130 spin_lock(&req->rq_export->exp_lock);
131 req->rq_export->exp_vbr_failed = 1;
132 spin_unlock(&req->rq_export->exp_lock);
139 * Save pre-versions in reply.
141 static void mdt_version_save(struct ptlrpc_request *req, __u64 version,
146 if (!exp_connect_vbr(req->rq_export))
149 LASSERT(!req_is_replay(req));
150 LASSERT(req->rq_repmsg != NULL);
151 reply_ver = lustre_msg_get_versions(req->rq_repmsg);
153 reply_ver[idx] = version;
157 * Save enoent version, it is needed when it is obvious that object doesn't
158 * exist, e.g. child during create.
160 static void mdt_enoent_version_save(struct mdt_thread_info *info, int idx)
162 /* save version of file name for replay, it must be ENOENT here */
163 if (!req_is_replay(mdt_info_req(info))) {
164 info->mti_ver[idx] = ENOENT_VERSION;
165 mdt_version_save(mdt_info_req(info), info->mti_ver[idx], idx);
170 * Get version from disk and save in reply buffer.
172 * Versions are saved in reply only during normal operations not replays.
174 void mdt_version_get_save(struct mdt_thread_info *info,
175 struct mdt_object *mto, int idx)
177 /* don't save versions during replay */
178 if (!req_is_replay(mdt_info_req(info))) {
179 mdt_obj_version_get(info, mto, &info->mti_ver[idx]);
180 mdt_version_save(mdt_info_req(info), info->mti_ver[idx], idx);
185 * Get version from disk and check it, no save in reply.
187 int mdt_version_get_check(struct mdt_thread_info *info,
188 struct mdt_object *mto, int idx)
190 /* only check versions during replay */
191 if (!req_is_replay(mdt_info_req(info)))
194 mdt_obj_version_get(info, mto, &info->mti_ver[idx]);
195 return mdt_version_check(mdt_info_req(info), info->mti_ver[idx], idx);
199 * Get version from disk and check if recovery or just save.
201 int mdt_version_get_check_save(struct mdt_thread_info *info,
202 struct mdt_object *mto, int idx)
206 mdt_obj_version_get(info, mto, &info->mti_ver[idx]);
207 if (req_is_replay(mdt_info_req(info)))
208 rc = mdt_version_check(mdt_info_req(info), info->mti_ver[idx],
211 mdt_version_save(mdt_info_req(info), info->mti_ver[idx], idx);
216 * Lookup with version checking.
218 * This checks version of 'name'. Many reint functions uses 'name' for child not
219 * FID, therefore we need to get object by name and check its version.
221 int mdt_lookup_version_check(struct mdt_thread_info *info,
222 struct mdt_object *p, struct lu_name *lname,
223 struct lu_fid *fid, int idx)
227 rc = mdo_lookup(info->mti_env, mdt_object_child(p), lname, fid,
229 /* Check version only during replay */
230 if (!req_is_replay(mdt_info_req(info)))
233 info->mti_ver[idx] = ENOENT_VERSION;
235 struct mdt_object *child;
236 child = mdt_object_find(info->mti_env, info->mti_mdt, fid);
237 if (likely(!IS_ERR(child))) {
238 mdt_obj_version_get(info, child, &info->mti_ver[idx]);
239 mdt_object_put(info->mti_env, child);
242 vbrc = mdt_version_check(mdt_info_req(info), info->mti_ver[idx], idx);
243 return vbrc ? vbrc : rc;
248 * VBR: we save three versions in reply:
249 * 0 - parent. Check that parent version is the same during replay.
250 * 1 - name. Version of 'name' if file exists with the same name or
251 * ENOENT_VERSION, it is needed because file may appear due to missed replays.
252 * 2 - child. Version of child by FID. Must be ENOENT. It is mostly sanity
255 static int mdt_md_create(struct mdt_thread_info *info)
257 struct mdt_device *mdt = info->mti_mdt;
258 struct mdt_object *parent;
259 struct mdt_object *child;
260 struct mdt_lock_handle *lh;
261 struct mdt_body *repbody;
262 struct md_attr *ma = &info->mti_attr;
263 struct mdt_reint_record *rr = &info->mti_rr;
264 struct lu_name *lname;
268 DEBUG_REQ(D_INODE, mdt_info_req(info), "Create (%s->"DFID") in "DFID,
269 rr->rr_name, PFID(rr->rr_fid2), PFID(rr->rr_fid1));
271 if (fid_is_obf(rr->rr_fid1) || fid_is_dot_lustre(rr->rr_fid1))
274 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
276 lh = &info->mti_lh[MDT_LH_PARENT];
277 mdt_lock_pdo_init(lh, LCK_PW, rr->rr_name, rr->rr_namelen);
279 parent = mdt_object_find_lock(info, rr->rr_fid1, lh,
280 MDS_INODELOCK_UPDATE);
282 RETURN(PTR_ERR(parent));
284 rc = mdt_version_get_check_save(info, parent, 0);
286 GOTO(out_put_parent, rc);
289 * Check child name version during replay.
290 * During create replay a file may exist with same name.
292 lname = mdt_name(info->mti_env, (char *)rr->rr_name, rr->rr_namelen);
293 rc = mdt_lookup_version_check(info, parent, lname,
294 &info->mti_tmp_fid1, 1);
296 GOTO(out_put_parent, rc = -EEXIST);
298 /* -ENOENT is expected here */
300 GOTO(out_put_parent, rc);
302 /* save version of file name for replay, it must be ENOENT here */
303 mdt_enoent_version_save(info, 1);
305 child = mdt_object_new(info->mti_env, mdt, rr->rr_fid2);
306 if (likely(!IS_ERR(child))) {
307 struct md_object *next = mdt_object_child(parent);
309 if (mdt_object_remote(child)) {
310 struct seq_server_site *ss;
311 struct lu_ucred *uc = mdt_ucred(info);
313 if (!md_capable(uc, CFS_CAP_SYS_ADMIN)) {
315 mdt->mdt_enable_remote_dir_gid &&
316 mdt->mdt_enable_remote_dir_gid != -1) {
317 CERROR("%s: Creating remote dir is only"
318 " permitted for administrator or"
319 " set mdt_enable_remote_dir_gid:"
321 mdt_obd_name(mdt), -EPERM);
322 GOTO(out_put_child, rc = -EPERM);
326 ss = mdt_seq_site(mdt);
327 if (ss->ss_node_id != 0 &&
328 mdt->mdt_enable_remote_dir == 0) {
329 CERROR("%s: remote dir is only permitted on"
331 " mdt.*.enable_remote_dir=1\n",
333 GOTO(out_put_child, rc = -EPERM);
335 if (!mdt_is_dne_client(mdt_info_req(info)->rq_export)) {
336 /* Return -EIO for old client */
337 GOTO(out_put_child, rc = -EIO);
341 ma->ma_need = MA_INODE;
343 /* capa for cross-ref will be stored here */
344 ma->ma_capa = req_capsule_server_get(info->mti_pill,
346 LASSERT(ma->ma_capa);
348 mdt_fail_write(info->mti_env, info->mti_mdt->mdt_bottom,
349 OBD_FAIL_MDS_REINT_CREATE_WRITE);
351 /* Version of child will be updated on disk. */
352 tgt_vbr_obj_set(info->mti_env, mdt_obj2dt(child));
353 rc = mdt_version_get_check_save(info, child, 2);
355 GOTO(out_put_child, rc);
357 /* Let lower layer know current lock mode. */
358 info->mti_spec.sp_cr_mode =
359 mdt_dlm_mode2mdl_mode(lh->mlh_pdo_mode);
362 * Do not perform lookup sanity check. We know that name does
365 info->mti_spec.sp_cr_lookup = 0;
366 info->mti_spec.sp_feat = &dt_directory_features;
368 rc = mdo_create(info->mti_env, next, lname,
369 mdt_object_child(child),
370 &info->mti_spec, ma);
372 rc = mdt_attr_get_complex(info, child, ma);
375 /* Return fid & attr to client. */
376 if (ma->ma_valid & MA_INODE)
377 mdt_pack_attr2body(info, repbody, &ma->ma_attr,
378 mdt_object_fid(child));
381 mdt_object_put(info->mti_env, child);
385 mdt_create_pack_capa(info, rc, child, repbody);
387 mdt_object_unlock_put(info, parent, lh, rc);
391 int mdt_attr_set(struct mdt_thread_info *info, struct mdt_object *mo,
392 struct md_attr *ma, int flags)
394 struct mdt_lock_handle *lh;
395 int do_vbr = ma->ma_attr.la_valid & (LA_MODE|LA_UID|LA_GID|LA_FLAGS);
396 __u64 lockpart = MDS_INODELOCK_UPDATE;
400 /* attr shouldn't be set on remote object */
401 LASSERT(!mdt_object_remote(mo));
403 lh = &info->mti_lh[MDT_LH_PARENT];
404 mdt_lock_reg_init(lh, LCK_PW);
406 /* Even though the new MDT will grant PERM lock to the old
407 * client, but the old client will almost ignore that during
408 * So it needs to revoke both LOOKUP and PERM lock here, so
409 * both new and old client can cancel the dcache */
410 if (ma->ma_attr.la_valid & (LA_MODE|LA_UID|LA_GID))
411 lockpart |= MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM;
413 rc = mdt_object_lock(info, mo, lh, lockpart, MDT_LOCAL_LOCK);
417 if (mdt_object_exists(mo) == 0)
418 GOTO(out_unlock, rc = -ENOENT);
420 /* all attrs are packed into mti_attr in unpack_setattr */
421 mdt_fail_write(info->mti_env, info->mti_mdt->mdt_bottom,
422 OBD_FAIL_MDS_REINT_SETATTR_WRITE);
424 /* This is only for set ctime when rename's source is on remote MDS. */
425 if (unlikely(ma->ma_attr.la_valid == LA_CTIME))
426 ma->ma_attr_flags |= MDS_VTX_BYPASS;
428 /* VBR: update version if attr changed are important for recovery */
430 /* update on-disk version of changed object */
431 tgt_vbr_obj_set(info->mti_env, mdt_obj2dt(mo));
432 rc = mdt_version_get_check_save(info, mo, 0);
434 GOTO(out_unlock, rc);
437 /* Ensure constant striping during chown(). See LU-2789. */
438 if (ma->ma_attr.la_valid & (LA_UID|LA_GID))
439 mutex_lock(&mo->mot_lov_mutex);
441 /* all attrs are packed into mti_attr in unpack_setattr */
442 rc = mo_attr_set(info->mti_env, mdt_object_child(mo), ma);
444 if (ma->ma_attr.la_valid & (LA_UID|LA_GID))
445 mutex_unlock(&mo->mot_lov_mutex);
448 GOTO(out_unlock, rc);
452 mdt_object_unlock(info, mo, lh, rc);
457 * Check HSM flags and add HS_DIRTY flag if relevant.
459 * A file could be set dirty only if it has a copy in the backend (HS_EXISTS)
460 * and is not RELEASED.
462 int mdt_add_dirty_flag(struct mdt_thread_info *info, struct mdt_object *mo,
468 /* If the file was modified, add the dirty flag */
469 ma->ma_need = MA_HSM;
470 rc = mdt_attr_get_complex(info, mo, ma);
472 CERROR("file attribute read error for "DFID": %d.\n",
473 PFID(mdt_object_fid(mo)), rc);
477 /* If an up2date copy exists in the backend, add dirty flag */
478 if ((ma->ma_valid & MA_HSM) && (ma->ma_hsm.mh_flags & HS_EXISTS)
479 && !(ma->ma_hsm.mh_flags & (HS_DIRTY|HS_RELEASED))) {
480 struct mdt_lock_handle *lh = &info->mti_lh[MDT_LH_CHILD];
482 ma->ma_hsm.mh_flags |= HS_DIRTY;
484 mdt_lock_reg_init(lh, LCK_PW);
485 rc = mdt_object_lock(info, mo, lh, MDS_INODELOCK_XATTR,
490 rc = mdt_hsm_attr_set(info, mo, &ma->ma_hsm);
492 CERROR("file attribute change error for "DFID": %d\n",
493 PFID(mdt_object_fid(mo)), rc);
494 mdt_object_unlock(info, mo, lh, rc);
500 static int mdt_reint_setattr(struct mdt_thread_info *info,
501 struct mdt_lock_handle *lhc)
503 struct md_attr *ma = &info->mti_attr;
504 struct mdt_reint_record *rr = &info->mti_rr;
505 struct ptlrpc_request *req = mdt_info_req(info);
506 struct mdt_export_data *med = &req->rq_export->exp_mdt_data;
507 struct mdt_file_data *mfd;
508 struct mdt_object *mo;
509 struct mdt_body *repbody;
513 DEBUG_REQ(D_INODE, req, "setattr "DFID" %x", PFID(rr->rr_fid1),
514 (unsigned int)ma->ma_attr.la_valid);
516 if (info->mti_dlm_req)
517 ldlm_request_cancel(req, info->mti_dlm_req, 0);
519 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
520 mo = mdt_object_find(info->mti_env, info->mti_mdt, rr->rr_fid1);
522 GOTO(out, rc = PTR_ERR(mo));
524 /* start a log jounal handle if needed */
525 if (!(mdt_conn_flags(info) & OBD_CONNECT_SOM)) {
526 if ((ma->ma_attr.la_valid & LA_SIZE) ||
527 (rr->rr_flags & MRF_OPEN_TRUNC)) {
528 /* Check write access for the O_TRUNC case */
529 if (mdt_write_read(mo) < 0)
530 GOTO(out_put, rc = -ETXTBSY);
532 } else if (info->mti_ioepoch &&
533 (info->mti_ioepoch->flags & MF_EPOCH_OPEN)) {
534 /* Truncate case. IOEpoch is opened. */
535 rc = mdt_write_get(mo);
539 mfd = mdt_mfd_new(med);
542 GOTO(out_put, rc = -ENOMEM);
545 mdt_ioepoch_open(info, mo, 0);
546 repbody->ioepoch = mo->mot_ioepoch;
548 mdt_object_get(info->mti_env, mo);
549 mdt_mfd_set_mode(mfd, MDS_FMODE_TRUNC);
550 mfd->mfd_object = mo;
551 mfd->mfd_xid = req->rq_xid;
553 spin_lock(&med->med_open_lock);
554 cfs_list_add(&mfd->mfd_list, &med->med_open_head);
555 spin_unlock(&med->med_open_lock);
556 repbody->handle.cookie = mfd->mfd_handle.h_cookie;
559 som_au = info->mti_ioepoch && info->mti_ioepoch->flags & MF_SOM_CHANGE;
561 /* SOM Attribute update case. Find the proper mfd and update
562 * SOM attributes on the proper object. */
563 LASSERT(mdt_conn_flags(info) & OBD_CONNECT_SOM);
564 LASSERT(info->mti_ioepoch);
566 spin_lock(&med->med_open_lock);
567 mfd = mdt_handle2mfd(med, &info->mti_ioepoch->handle,
570 spin_unlock(&med->med_open_lock);
571 CDEBUG(D_INODE, "no handle for file close: "
572 "fid = "DFID": cookie = "LPX64"\n",
573 PFID(info->mti_rr.rr_fid1),
574 info->mti_ioepoch->handle.cookie);
575 GOTO(out_put, rc = -ESTALE);
577 LASSERT(mfd->mfd_mode == MDS_FMODE_SOM);
578 LASSERT(!(info->mti_ioepoch->flags & MF_EPOCH_CLOSE));
580 class_handle_unhash(&mfd->mfd_handle);
581 cfs_list_del_init(&mfd->mfd_list);
582 spin_unlock(&med->med_open_lock);
584 mdt_mfd_close(info, mfd);
585 } else if ((ma->ma_valid & MA_INODE) && ma->ma_attr.la_valid) {
586 LASSERT((ma->ma_valid & MA_LOV) == 0);
587 rc = mdt_attr_set(info, mo, ma, rr->rr_flags);
590 } else if ((ma->ma_valid & MA_LOV) && (ma->ma_valid & MA_INODE)) {
591 struct lu_buf *buf = &info->mti_buf;
592 LASSERT(ma->ma_attr.la_valid == 0);
593 buf->lb_buf = ma->ma_lmm;
594 buf->lb_len = ma->ma_lmm_size;
595 rc = mo_xattr_set(info->mti_env, mdt_object_child(mo),
596 buf, XATTR_NAME_LOV, 0);
602 /* If file data is modified, add the dirty flag */
603 if (ma->ma_attr_flags & MDS_DATA_MODIFIED)
604 rc = mdt_add_dirty_flag(info, mo, ma);
606 ma->ma_need = MA_INODE;
608 rc = mdt_attr_get_complex(info, mo, ma);
612 mdt_pack_attr2body(info, repbody, &ma->ma_attr, mdt_object_fid(mo));
614 if (info->mti_mdt->mdt_lut.lut_oss_capa &&
615 exp_connect_flags(info->mti_exp) & OBD_CONNECT_OSS_CAPA &&
616 S_ISREG(lu_object_attr(&mo->mot_obj)) &&
617 (ma->ma_attr.la_valid & LA_SIZE) && !som_au) {
618 struct lustre_capa *capa;
620 capa = req_capsule_server_get(info->mti_pill, &RMF_CAPA2);
622 capa->lc_opc = CAPA_OPC_OSS_DEFAULT | CAPA_OPC_OSS_TRUNC;
623 rc = mo_capa_get(info->mti_env, mdt_object_child(mo), capa, 0);
626 repbody->valid |= OBD_MD_FLOSSCAPA;
631 mdt_object_put(info->mti_env, mo);
634 mdt_counter_incr(req, LPROC_MDT_SETATTR);
636 mdt_client_compatibility(info);
637 rc2 = mdt_fix_reply(info);
643 static int mdt_reint_create(struct mdt_thread_info *info,
644 struct mdt_lock_handle *lhc)
646 struct ptlrpc_request *req = mdt_info_req(info);
650 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_REINT_CREATE))
651 RETURN(err_serious(-ESTALE));
653 if (info->mti_dlm_req)
654 ldlm_request_cancel(mdt_info_req(info), info->mti_dlm_req, 0);
656 LASSERT(info->mti_rr.rr_namelen > 0);
657 switch (info->mti_attr.ma_attr.la_mode & S_IFMT) {
659 mdt_counter_incr(req, LPROC_MDT_MKDIR);
667 /* Special file should stay on the same node as parent. */
668 mdt_counter_incr(req, LPROC_MDT_MKNOD);
671 CERROR("%s: Unsupported mode %o\n",
672 mdt_obd_name(info->mti_mdt),
673 info->mti_attr.ma_attr.la_mode);
674 RETURN(err_serious(-EOPNOTSUPP));
677 rc = mdt_md_create(info);
682 * VBR: save parent version in reply and child version getting by its name.
683 * Version of child is getting and checking during its lookup. If
685 static int mdt_reint_unlink(struct mdt_thread_info *info,
686 struct mdt_lock_handle *lhc)
688 struct mdt_reint_record *rr = &info->mti_rr;
689 struct ptlrpc_request *req = mdt_info_req(info);
690 struct md_attr *ma = &info->mti_attr;
691 struct lu_fid *child_fid = &info->mti_tmp_fid1;
692 struct mdt_object *mp;
693 struct mdt_object *mc;
694 struct mdt_lock_handle *parent_lh;
695 struct mdt_lock_handle *child_lh;
696 struct lu_name *lname;
701 DEBUG_REQ(D_INODE, req, "unlink "DFID"/%s", PFID(rr->rr_fid1),
704 if (info->mti_dlm_req)
705 ldlm_request_cancel(req, info->mti_dlm_req, 0);
707 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_REINT_UNLINK))
708 RETURN(err_serious(-ENOENT));
710 if (fid_is_obf(rr->rr_fid1) || fid_is_dot_lustre(rr->rr_fid1))
713 * step 1: Found the parent.
715 mp = mdt_object_find(info->mti_env, info->mti_mdt, rr->rr_fid1);
721 parent_lh = &info->mti_lh[MDT_LH_PARENT];
722 lname = mdt_name(info->mti_env, (char *)rr->rr_name, rr->rr_namelen);
723 if (mdt_object_remote(mp)) {
724 mdt_lock_reg_init(parent_lh, LCK_EX);
725 rc = mdt_remote_object_lock(info, mp, &parent_lh->mlh_rreg_lh,
726 parent_lh->mlh_rreg_mode,
727 MDS_INODELOCK_UPDATE);
729 GOTO(put_parent, rc);
732 mdt_lock_pdo_init(parent_lh, LCK_PW, rr->rr_name,
734 rc = mdt_object_lock(info, mp, parent_lh, MDS_INODELOCK_UPDATE,
737 GOTO(put_parent, rc);
739 rc = mdt_version_get_check_save(info, mp, 0);
741 GOTO(unlock_parent, rc);
744 /* step 2: find & lock the child */
745 /* lookup child object along with version checking */
747 rc = mdt_lookup_version_check(info, mp, lname, child_fid, 1);
749 /* Name might not be able to find during resend of
750 * remote unlink, considering following case.
751 * dir_A is a remote directory, the name entry of
752 * dir_A is on MDT0, the directory is on MDT1,
754 * 1. client sends unlink req to MDT1.
755 * 2. MDT1 sends name delete update to MDT0.
756 * 3. name entry is being deleted in MDT0 synchronously.
757 * 4. MDT1 is restarted.
758 * 5. client resends unlink req to MDT1. So it can not
759 * find the name entry on MDT0 anymore.
760 * In this case, MDT1 only needs to destory the local
763 if (mdt_object_remote(mp) && rc == -ENOENT &&
764 !fid_is_zero(rr->rr_fid2) &&
765 lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT) {
767 *child_fid = *rr->rr_fid2;
769 GOTO(unlock_parent, rc);
773 if (fid_is_obf(child_fid) || fid_is_dot_lustre(child_fid))
774 GOTO(unlock_parent, rc = -EPERM);
776 mdt_reint_init_ma(info, ma);
778 /* We will lock the child regardless it is local or remote. No harm. */
779 mc = mdt_object_find(info->mti_env, info->mti_mdt, child_fid);
781 GOTO(unlock_parent, rc = PTR_ERR(mc));
783 child_lh = &info->mti_lh[MDT_LH_CHILD];
784 mdt_lock_reg_init(child_lh, LCK_EX);
785 if (mdt_object_remote(mc)) {
786 struct mdt_body *repbody;
788 if (!fid_is_zero(rr->rr_fid2)) {
789 CDEBUG(D_INFO, "%s: name %s can not find "DFID"\n",
790 mdt_obd_name(info->mti_mdt),
791 (char *)rr->rr_name, PFID(mdt_object_fid(mc)));
792 GOTO(put_child, rc = -ENOENT);
794 CDEBUG(D_INFO, "%s: name %s: "DFID" is another MDT\n",
795 mdt_obd_name(info->mti_mdt),
796 (char *)rr->rr_name, PFID(mdt_object_fid(mc)));
798 if (!mdt_is_dne_client(req->rq_export))
799 /* Return -EIO for old client */
800 GOTO(put_child, rc = -EIO);
802 if (info->mti_spec.sp_rm_entry) {
803 struct lu_ucred *uc = mdt_ucred(info);
805 if (!md_capable(uc, CFS_CAP_SYS_ADMIN)) {
806 CERROR("%s: unlink remote entry is only "
807 "permitted for administrator: rc = %d\n",
808 mdt_obd_name(info->mti_mdt),
810 GOTO(put_child, rc = -EPERM);
813 ma->ma_need = MA_INODE;
815 mdt_set_capainfo(info, 1, child_fid, BYPASS_CAPA);
816 rc = mdo_unlink(info->mti_env, mdt_object_child(mp),
817 NULL, lname, ma, no_name);
820 /* Revoke the LOOKUP lock of the remote object granted by
821 * this MDT. Since the unlink will happen on another MDT,
822 * it will release the LOOKUP lock right away. Then What
823 * would happen if another client try to grab the LOOKUP
824 * lock at the same time with unlink XXX */
825 mdt_object_lock(info, mc, child_lh, MDS_INODELOCK_LOOKUP,
827 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
828 LASSERT(repbody != NULL);
829 repbody->fid1 = *mdt_object_fid(mc);
830 repbody->valid |= (OBD_MD_FLID | OBD_MD_MDS);
831 GOTO(unlock_child, rc = -EREMOTE);
832 } else if (info->mti_spec.sp_rm_entry) {
834 CDEBUG(D_INFO, "%s: no rm_entry on local dir '%s': rc = %d\n",
835 mdt_obd_name(info->mti_mdt), (char *)rr->rr_name, rc);
839 /* We used to acquire MDS_INODELOCK_FULL here but we can't do
840 * this now because a running HSM restore on the child (unlink
841 * victim) will hold the layout lock. See LU-4002. */
842 rc = mdt_object_lock(info, mc, child_lh,
843 MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE,
848 mdt_fail_write(info->mti_env, info->mti_mdt->mdt_bottom,
849 OBD_FAIL_MDS_REINT_UNLINK_WRITE);
850 /* save version when object is locked */
851 mdt_version_get_save(info, mc, 1);
853 * Now we can only make sure we need MA_INODE, in mdd layer, will check
854 * whether need MA_LOV and MA_COOKIE.
856 ma->ma_need = MA_INODE;
858 mdt_set_capainfo(info, 1, child_fid, BYPASS_CAPA);
860 mutex_lock(&mc->mot_lov_mutex);
862 rc = mdo_unlink(info->mti_env, mdt_object_child(mp),
863 mdt_object_child(mc), lname, ma, no_name);
865 mutex_unlock(&mc->mot_lov_mutex);
867 if (rc == 0 && !lu_object_is_dying(&mc->mot_header))
868 rc = mdt_attr_get_complex(info, mc, ma);
870 mdt_handle_last_unlink(info, mc, ma);
872 if (ma->ma_valid & MA_INODE) {
873 switch (ma->ma_attr.la_mode & S_IFMT) {
875 mdt_counter_incr(req, LPROC_MDT_RMDIR);
883 mdt_counter_incr(req, LPROC_MDT_UNLINK);
886 LASSERTF(0, "bad file type %o unlinking\n",
887 ma->ma_attr.la_mode);
893 mdt_object_unlock(info, mc, child_lh, rc);
895 mdt_object_put(info->mti_env, mc);
897 mdt_object_unlock(info, mp, parent_lh, rc);
899 mdt_object_put(info->mti_env, mp);
905 * VBR: save versions in reply: 0 - parent; 1 - child by fid; 2 - target by
908 static int mdt_reint_link(struct mdt_thread_info *info,
909 struct mdt_lock_handle *lhc)
911 struct mdt_reint_record *rr = &info->mti_rr;
912 struct ptlrpc_request *req = mdt_info_req(info);
913 struct md_attr *ma = &info->mti_attr;
914 struct mdt_object *ms;
915 struct mdt_object *mp;
916 struct mdt_lock_handle *lhs;
917 struct mdt_lock_handle *lhp;
918 struct lu_name *lname;
922 DEBUG_REQ(D_INODE, req, "link "DFID" to "DFID"/%s",
923 PFID(rr->rr_fid1), PFID(rr->rr_fid2), rr->rr_name);
925 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_REINT_LINK))
926 RETURN(err_serious(-ENOENT));
928 if (info->mti_dlm_req)
929 ldlm_request_cancel(req, info->mti_dlm_req, 0);
931 /* Invalid case so return error immediately instead of
933 if (lu_fid_eq(rr->rr_fid1, rr->rr_fid2))
936 if (fid_is_obf(rr->rr_fid1) || fid_is_dot_lustre(rr->rr_fid1) ||
937 fid_is_obf(rr->rr_fid2) || fid_is_dot_lustre(rr->rr_fid2))
940 /* step 1: find & lock the target parent dir */
941 lhp = &info->mti_lh[MDT_LH_PARENT];
942 mdt_lock_pdo_init(lhp, LCK_PW, rr->rr_name,
944 mp = mdt_object_find_lock(info, rr->rr_fid2, lhp,
945 MDS_INODELOCK_UPDATE);
949 rc = mdt_version_get_check_save(info, mp, 0);
951 GOTO(out_unlock_parent, rc);
953 /* step 2: find & lock the source */
954 lhs = &info->mti_lh[MDT_LH_CHILD];
955 mdt_lock_reg_init(lhs, LCK_EX);
957 ms = mdt_object_find(info->mti_env, info->mti_mdt, rr->rr_fid1);
959 GOTO(out_unlock_parent, rc = PTR_ERR(ms));
961 if (!mdt_object_exists(ms)) {
962 mdt_object_put(info->mti_env, ms);
963 CDEBUG(D_INFO, "%s: "DFID" does not exist.\n",
964 mdt_obd_name(info->mti_mdt), PFID(rr->rr_fid1));
965 GOTO(out_unlock_parent, rc = -ENOENT);
968 if (mdt_object_remote(ms)) {
969 mdt_object_put(info->mti_env, ms);
970 CERROR("%s: source inode "DFID" on remote MDT from "DFID"\n",
971 mdt_obd_name(info->mti_mdt), PFID(rr->rr_fid1),
973 GOTO(out_unlock_parent, rc = -EXDEV);
976 rc = mdt_object_lock(info, ms, lhs, MDS_INODELOCK_UPDATE |
977 MDS_INODELOCK_XATTR, MDT_CROSS_LOCK);
979 mdt_object_put(info->mti_env, ms);
980 GOTO(out_unlock_parent, rc);
983 /* step 3: link it */
984 mdt_fail_write(info->mti_env, info->mti_mdt->mdt_bottom,
985 OBD_FAIL_MDS_REINT_LINK_WRITE);
987 tgt_vbr_obj_set(info->mti_env, mdt_obj2dt(ms));
988 rc = mdt_version_get_check_save(info, ms, 1);
990 GOTO(out_unlock_child, rc);
992 lname = mdt_name(info->mti_env, (char *)rr->rr_name, rr->rr_namelen);
993 /** check target version by name during replay */
994 rc = mdt_lookup_version_check(info, mp, lname, &info->mti_tmp_fid1, 2);
995 if (rc != 0 && rc != -ENOENT)
996 GOTO(out_unlock_child, rc);
997 /* save version of file name for replay, it must be ENOENT here */
998 if (!req_is_replay(mdt_info_req(info))) {
1000 CDEBUG(D_INFO, "link target %.*s existed!\n",
1001 rr->rr_namelen, (char *)rr->rr_name);
1002 GOTO(out_unlock_child, rc = -EEXIST);
1004 info->mti_ver[2] = ENOENT_VERSION;
1005 mdt_version_save(mdt_info_req(info), info->mti_ver[2], 2);
1008 rc = mdo_link(info->mti_env, mdt_object_child(mp),
1009 mdt_object_child(ms), lname, ma);
1012 mdt_counter_incr(req, LPROC_MDT_LINK);
1016 mdt_object_unlock_put(info, ms, lhs, rc);
1018 mdt_object_unlock_put(info, mp, lhp, rc);
1022 * lock the part of the directory according to the hash of the name
1023 * (lh->mlh_pdo_hash) in parallel directory lock.
1025 static int mdt_pdir_hash_lock(struct mdt_thread_info *info,
1026 struct mdt_lock_handle *lh,
1027 struct mdt_object *obj, __u64 ibits)
1029 struct ldlm_res_id *res = &info->mti_res_id;
1030 struct ldlm_namespace *ns = info->mti_mdt->mdt_namespace;
1031 ldlm_policy_data_t *policy = &info->mti_policy;
1035 * Finish res_id initializing by name hash marking part of
1036 * directory which is taking modification.
1038 LASSERT(lh->mlh_pdo_hash != 0);
1039 fid_build_pdo_res_name(mdt_object_fid(obj), lh->mlh_pdo_hash, res);
1040 memset(policy, 0, sizeof(*policy));
1041 policy->l_inodebits.bits = ibits;
1043 * Use LDLM_FL_LOCAL_ONLY for this lock. We do not know yet if it is
1044 * going to be sent to client. If it is - mdt_intent_policy() path will
1045 * fix it up and turn FL_LOCAL flag off.
1047 rc = mdt_fid_lock(ns, &lh->mlh_reg_lh, lh->mlh_reg_mode, policy,
1048 res, LDLM_FL_LOCAL_ONLY | LDLM_FL_ATOMIC_CB,
1049 &info->mti_exp->exp_handle.h_cookie);
1053 static int mdt_rename_lock(struct mdt_thread_info *info,
1054 struct lustre_handle *lh)
1056 struct ldlm_namespace *ns = info->mti_mdt->mdt_namespace;
1057 ldlm_policy_data_t *policy = &info->mti_policy;
1058 struct ldlm_res_id *res_id = &info->mti_res_id;
1063 fid_build_reg_res_name(&LUSTRE_BFL_FID, res_id);
1065 memset(policy, 0, sizeof *policy);
1066 policy->l_inodebits.bits = MDS_INODELOCK_UPDATE;
1067 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 5, 53, 0)
1068 /* In phase I, we will not do cross-rename, so local BFL lock would
1071 flags = LDLM_FL_LOCAL_ONLY | LDLM_FL_ATOMIC_CB;
1073 * Current node is controller, that is mdt0, where we should
1076 rc = ldlm_cli_enqueue_local(ns, res_id, LDLM_IBITS, policy,
1077 LCK_EX, &flags, ldlm_blocking_ast,
1078 ldlm_completion_ast, NULL, NULL, 0,
1080 &info->mti_exp->exp_handle.h_cookie,
1083 #warning "Local rename lock is invalid for DNE phase II."
1088 static void mdt_rename_unlock(struct lustre_handle *lh)
1091 LASSERT(lustre_handle_is_used(lh));
1092 ldlm_lock_decref(lh, LCK_EX);
1097 * This is is_subdir() variant, it is CMD if cmm forwards it to correct
1098 * target. Source should not be ancestor of target dir. May be other rename
1099 * checks can be moved here later.
1101 static int mdt_rename_sanity(struct mdt_thread_info *info, struct lu_fid *fid)
1103 struct mdt_reint_record *rr = &info->mti_rr;
1104 struct lu_fid dst_fid = *rr->rr_fid2;
1105 struct mdt_object *dst;
1110 LASSERT(fid_is_sane(&dst_fid));
1111 dst = mdt_object_find(info->mti_env, info->mti_mdt, &dst_fid);
1113 rc = mdo_is_subdir(info->mti_env,
1114 mdt_object_child(dst), fid,
1116 mdt_object_put(info->mti_env, dst);
1117 if (rc != -EREMOTE && rc < 0) {
1118 CERROR("Failed mdo_is_subdir(), rc %d\n", rc);
1120 /* check the found fid */
1121 if (lu_fid_eq(&dst_fid, fid))
1127 } while (rc == -EREMOTE);
1133 * VBR: rename versions in reply: 0 - src parent; 1 - tgt parent;
1134 * 2 - src child; 3 - tgt child.
1135 * Update on disk version of src child.
1138 * For DNE phase I, only these renames are allowed
1139 * mv src_p/src_c tgt_p/tgt_c
1140 * 1. src_p/src_c/tgt_p/tgt_c are in the same MDT.
1141 * 2. src_p and tgt_p are same directory, and tgt_c does not
1142 * exists. In this case, all of modification will happen
1143 * in the MDT where ithesource parent is, only one remote
1144 * update is needed, i.e. set c_time/m_time on the child.
1145 * And tgt_c will be still in the same MDT as the original
1148 static int mdt_reint_rename(struct mdt_thread_info *info,
1149 struct mdt_lock_handle *lhc)
1151 struct mdt_reint_record *rr = &info->mti_rr;
1152 struct md_attr *ma = &info->mti_attr;
1153 struct ptlrpc_request *req = mdt_info_req(info);
1154 struct mdt_object *msrcdir;
1155 struct mdt_object *mtgtdir;
1156 struct mdt_object *mold;
1157 struct mdt_object *mnew = NULL;
1158 struct mdt_lock_handle *lh_srcdirp;
1159 struct mdt_lock_handle *lh_tgtdirp;
1160 struct mdt_lock_handle *lh_oldp;
1161 struct mdt_lock_handle *lh_newp;
1162 struct lu_fid *old_fid = &info->mti_tmp_fid1;
1163 struct lu_fid *new_fid = &info->mti_tmp_fid2;
1164 struct lustre_handle rename_lh = { 0 };
1165 struct lu_name slname = { 0 };
1166 struct lu_name *lname;
1170 if (info->mti_dlm_req)
1171 ldlm_request_cancel(req, info->mti_dlm_req, 0);
1173 DEBUG_REQ(D_INODE, req, "rename "DFID"/%s to "DFID"/%s",
1174 PFID(rr->rr_fid1), rr->rr_name,
1175 PFID(rr->rr_fid2), rr->rr_tgt);
1177 if (fid_is_obf(rr->rr_fid1) || fid_is_dot_lustre(rr->rr_fid1) ||
1178 fid_is_obf(rr->rr_fid2) || fid_is_dot_lustre(rr->rr_fid2))
1181 rc = mdt_rename_lock(info, &rename_lh);
1183 CERROR("Can't lock FS for rename, rc %d\n", rc);
1187 lh_newp = &info->mti_lh[MDT_LH_NEW];
1189 /* step 1: lock the source dir. */
1190 lh_srcdirp = &info->mti_lh[MDT_LH_PARENT];
1191 mdt_lock_pdo_init(lh_srcdirp, LCK_PW, rr->rr_name,
1193 msrcdir = mdt_object_find_lock(info, rr->rr_fid1, lh_srcdirp,
1194 MDS_INODELOCK_UPDATE);
1195 if (IS_ERR(msrcdir))
1196 GOTO(out_rename_lock, rc = PTR_ERR(msrcdir));
1198 rc = mdt_version_get_check_save(info, msrcdir, 0);
1200 GOTO(out_unlock_source, rc);
1202 /* step 2: find & lock the target dir. */
1203 lh_tgtdirp = &info->mti_lh[MDT_LH_CHILD];
1204 mdt_lock_pdo_init(lh_tgtdirp, LCK_PW, rr->rr_tgt,
1206 if (lu_fid_eq(rr->rr_fid1, rr->rr_fid2)) {
1207 mdt_object_get(info->mti_env, msrcdir);
1209 if (lh_tgtdirp->mlh_pdo_hash != lh_srcdirp->mlh_pdo_hash) {
1210 rc = mdt_pdir_hash_lock(info, lh_tgtdirp, mtgtdir,
1211 MDS_INODELOCK_UPDATE);
1213 GOTO(out_unlock_source, rc);
1214 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_PDO_LOCK2, 10);
1217 mtgtdir = mdt_object_find(info->mti_env, info->mti_mdt,
1219 if (IS_ERR(mtgtdir))
1220 GOTO(out_unlock_source, rc = PTR_ERR(mtgtdir));
1222 /* check early, the real version will be saved after locking */
1223 rc = mdt_version_get_check(info, mtgtdir, 1);
1225 GOTO(out_put_target, rc);
1227 if (unlikely(mdt_object_remote(mtgtdir))) {
1228 CDEBUG(D_INFO, "Source dir "DFID" target dir "DFID
1229 "on different MDTs\n", PFID(rr->rr_fid1),
1231 GOTO(out_put_target, rc = -EXDEV);
1233 if (likely(mdt_object_exists(mtgtdir))) {
1234 /* we lock the target dir if it is local */
1235 rc = mdt_object_lock(info, mtgtdir, lh_tgtdirp,
1236 MDS_INODELOCK_UPDATE,
1239 GOTO(out_put_target, rc);
1240 /* get and save correct version after locking */
1241 mdt_version_get_save(info, mtgtdir, 1);
1243 GOTO(out_put_target, rc = -ESTALE);
1248 /* step 3: find & lock the old object. */
1249 lname = mdt_name(info->mti_env, (char *)rr->rr_name, rr->rr_namelen);
1250 mdt_name_copy(&slname, lname);
1252 rc = mdt_lookup_version_check(info, msrcdir, &slname, old_fid, 2);
1254 GOTO(out_unlock_target, rc);
1256 if (lu_fid_eq(old_fid, rr->rr_fid1) || lu_fid_eq(old_fid, rr->rr_fid2))
1257 GOTO(out_unlock_target, rc = -EINVAL);
1259 if (fid_is_obf(old_fid) || fid_is_dot_lustre(old_fid))
1260 GOTO(out_unlock_target, rc = -EPERM);
1262 mold = mdt_object_find(info->mti_env, info->mti_mdt, old_fid);
1264 GOTO(out_unlock_target, rc = PTR_ERR(mold));
1266 lh_oldp = &info->mti_lh[MDT_LH_OLD];
1267 mdt_lock_reg_init(lh_oldp, LCK_EX);
1268 rc = mdt_object_lock(info, mold, lh_oldp, MDS_INODELOCK_LOOKUP |
1269 MDS_INODELOCK_XATTR, MDT_CROSS_LOCK);
1271 mdt_object_put(info->mti_env, mold);
1272 GOTO(out_unlock_target, rc);
1275 tgt_vbr_obj_set(info->mti_env, mdt_obj2dt(mold));
1276 /* save version after locking */
1277 mdt_version_get_save(info, mold, 2);
1278 mdt_set_capainfo(info, 2, old_fid, BYPASS_CAPA);
1280 /* step 4: find & lock the new object. */
1281 /* new target object may not exist now */
1282 lname = mdt_name(info->mti_env, (char *)rr->rr_tgt, rr->rr_tgtlen);
1283 /* lookup with version checking */
1285 rc = mdt_lookup_version_check(info, mtgtdir, lname, new_fid, 3);
1287 /* the new_fid should have been filled at this moment */
1288 if (lu_fid_eq(old_fid, new_fid))
1289 GOTO(out_unlock_old, rc);
1291 if (lu_fid_eq(new_fid, rr->rr_fid1) ||
1292 lu_fid_eq(new_fid, rr->rr_fid2))
1293 GOTO(out_unlock_old, rc = -EINVAL);
1295 if (fid_is_obf(new_fid) || fid_is_dot_lustre(new_fid))
1296 GOTO(out_unlock_old, rc = -EPERM);
1298 if (mdt_object_remote(mold)) {
1299 CDEBUG(D_INFO, "Src child "DFID" is on another MDT\n",
1301 GOTO(out_unlock_old, rc = -EXDEV);
1304 mdt_lock_reg_init(lh_newp, LCK_EX);
1305 mnew = mdt_object_find(info->mti_env, info->mti_mdt, new_fid);
1307 GOTO(out_unlock_old, rc = PTR_ERR(mnew));
1309 if (mdt_object_remote(mnew)) {
1310 mdt_object_put(info->mti_env, mnew);
1311 CDEBUG(D_INFO, "src child "DFID" is on another MDT\n",
1313 GOTO(out_unlock_old, rc = -EXDEV);
1316 /* We used to acquire MDS_INODELOCK_FULL here but we
1317 * can't do this now because a running HSM restore on
1318 * the rename onto victim will hold the layout
1319 * lock. See LU-4002. */
1320 rc = mdt_object_lock(info, mnew, lh_newp,
1321 MDS_INODELOCK_LOOKUP |
1322 MDS_INODELOCK_UPDATE,
1325 mdt_object_put(info->mti_env, mnew);
1326 GOTO(out_unlock_old, rc);
1328 /* get and save version after locking */
1329 mdt_version_get_save(info, mnew, 3);
1330 mdt_set_capainfo(info, 3, new_fid, BYPASS_CAPA);
1331 } else if (rc != -EREMOTE && rc != -ENOENT) {
1332 GOTO(out_unlock_old, rc);
1334 /* If mnew does not exist and mold are remote directory,
1335 * it only allows rename if they are under same directory */
1336 if (mtgtdir != msrcdir && mdt_object_remote(mold)) {
1337 CDEBUG(D_INFO, "Src child "DFID" is on another MDT\n",
1339 GOTO(out_unlock_old, rc = -EXDEV);
1341 mdt_enoent_version_save(info, 3);
1344 /* step 5: rename it */
1345 mdt_reint_init_ma(info, ma);
1347 mdt_fail_write(info->mti_env, info->mti_mdt->mdt_bottom,
1348 OBD_FAIL_MDS_REINT_RENAME_WRITE);
1350 /* Check if @dst is subdir of @src. */
1351 rc = mdt_rename_sanity(info, old_fid);
1353 GOTO(out_unlock_new, rc);
1356 mutex_lock(&mnew->mot_lov_mutex);
1358 rc = mdo_rename(info->mti_env, mdt_object_child(msrcdir),
1359 mdt_object_child(mtgtdir), old_fid, &slname,
1360 (mnew ? mdt_object_child(mnew) : NULL),
1364 mutex_unlock(&mnew->mot_lov_mutex);
1366 /* handle last link of tgt object */
1368 mdt_counter_incr(req, LPROC_MDT_RENAME);
1370 mdt_handle_last_unlink(info, mnew, ma);
1372 mdt_rename_counter_tally(info, info->mti_mdt, req,
1379 mdt_object_unlock_put(info, mnew, lh_newp, rc);
1381 mdt_object_unlock_put(info, mold, lh_oldp, rc);
1383 mdt_object_unlock(info, mtgtdir, lh_tgtdirp, rc);
1385 mdt_object_put(info->mti_env, mtgtdir);
1387 mdt_object_unlock_put(info, msrcdir, lh_srcdirp, rc);
1389 if (lustre_handle_is_used(&rename_lh))
1390 mdt_rename_unlock(&rename_lh);
1394 typedef int (*mdt_reinter)(struct mdt_thread_info *info,
1395 struct mdt_lock_handle *lhc);
1397 static mdt_reinter reinters[REINT_MAX] = {
1398 [REINT_SETATTR] = mdt_reint_setattr,
1399 [REINT_CREATE] = mdt_reint_create,
1400 [REINT_LINK] = mdt_reint_link,
1401 [REINT_UNLINK] = mdt_reint_unlink,
1402 [REINT_RENAME] = mdt_reint_rename,
1403 [REINT_OPEN] = mdt_reint_open,
1404 [REINT_SETXATTR] = mdt_reint_setxattr,
1405 [REINT_RMENTRY] = mdt_reint_unlink
1408 int mdt_reint_rec(struct mdt_thread_info *info,
1409 struct mdt_lock_handle *lhc)
1414 rc = reinters[info->mti_rr.rr_opcode](info, lhc);