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;
84 static int mdt_md_create(struct mdt_thread_info *info)
86 struct mdt_device *mdt = info->mti_mdt;
87 struct mdt_object *parent;
88 struct mdt_object *child;
89 struct mdt_lock_handle *lh;
90 struct mdt_body *repbody;
91 struct md_attr *ma = &info->mti_attr;
92 struct mdt_reint_record *rr = &info->mti_rr;
93 struct lu_name *lname;
97 DEBUG_REQ(D_INODE, mdt_info_req(info), "Create (%s->"DFID") in "DFID,
98 rr->rr_name, PFID(rr->rr_fid2), PFID(rr->rr_fid1));
100 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
102 lh = &info->mti_lh[MDT_LH_PARENT];
103 mdt_lock_pdo_init(lh, LCK_PW, rr->rr_name, rr->rr_namelen);
105 parent = mdt_object_find_lock(info, rr->rr_fid1, lh,
106 MDS_INODELOCK_UPDATE);
108 RETURN(PTR_ERR(parent));
110 child = mdt_object_find(info->mti_env, mdt, rr->rr_fid2);
111 if (likely(!IS_ERR(child))) {
112 struct md_object *next = mdt_object_child(parent);
114 ma->ma_need = MA_INODE;
116 /* capa for cross-ref will be stored here */
117 ma->ma_capa = req_capsule_server_get(info->mti_pill,
119 LASSERT(ma->ma_capa);
121 mdt_fail_write(info->mti_env, info->mti_mdt->mdt_bottom,
122 OBD_FAIL_MDS_REINT_CREATE_WRITE);
124 /* Let lower layer know current lock mode. */
125 info->mti_spec.sp_cr_mode =
126 mdt_dlm_mode2mdl_mode(lh->mlh_pdo_mode);
129 * Do perform lookup sanity check. We do not know if name exists
132 info->mti_spec.sp_cr_lookup = 1;
134 lname = mdt_name(info->mti_env, (char *)rr->rr_name,
136 rc = mdo_create(info->mti_env, next, lname,
137 mdt_object_child(child),
138 &info->mti_spec, ma);
140 /* Return fid & attr to client. */
141 if (ma->ma_valid & MA_INODE)
142 mdt_pack_attr2body(info, repbody, &ma->ma_attr,
143 mdt_object_fid(child));
145 mdt_object_put(info->mti_env, child);
149 mdt_create_pack_capa(info, rc, child, repbody);
150 mdt_object_unlock_put(info, parent, lh, rc);
154 /* Partial request to create object only */
155 static int mdt_md_mkobj(struct mdt_thread_info *info)
157 struct mdt_device *mdt = info->mti_mdt;
158 struct mdt_object *o;
159 struct mdt_body *repbody;
160 struct md_attr *ma = &info->mti_attr;
164 DEBUG_REQ(D_INODE, mdt_info_req(info), "Partial create "DFID"",
165 PFID(info->mti_rr.rr_fid2));
167 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
169 o = mdt_object_find(info->mti_env, mdt, info->mti_rr.rr_fid2);
171 struct md_object *next = mdt_object_child(o);
173 ma->ma_need = MA_INODE;
177 * Cross-ref create can encounter already created obj in case of
178 * recovery, just get attr in that case.
180 if (mdt_object_exists(o) == 1) {
181 rc = mo_attr_get(info->mti_env, next, ma);
184 * Here, NO permission check for object_create,
185 * such check has been done on the original MDS.
187 rc = mo_object_create(info->mti_env, next,
188 &info->mti_spec, ma);
191 /* Return fid & attr to client. */
192 if (ma->ma_valid & MA_INODE)
193 mdt_pack_attr2body(info, repbody, &ma->ma_attr,
196 mdt_object_put(info->mti_env, o);
200 mdt_create_pack_capa(info, rc, o, repbody);
204 /* In the raw-setattr case, we lock the child inode.
205 * In the write-back case or if being called from open,
206 * the client holds a lock already.
207 * We use the ATTR_FROM_OPEN (translated into MRF_SETATTR_LOCKED by
208 * mdt_setattr_unpack()) flag to tell these cases apart. */
209 int mdt_attr_set(struct mdt_thread_info *info, struct mdt_object *mo, int flags)
211 struct md_attr *ma = &info->mti_attr;
212 struct mdt_lock_handle *lh;
217 /* attr shouldn't be set on remote object */
218 LASSERT(mdt_object_exists(mo) >= 0);
221 som_update = (info->mti_epoch->flags & MF_SOM_CHANGE);
223 /* Try to avoid object_lock if another epoch has been started
225 if (som_update && (info->mti_epoch->ioepoch != mo->mot_ioepoch))
228 lh = &info->mti_lh[MDT_LH_PARENT];
229 mdt_lock_reg_init(lh, LCK_PW);
231 if (!(flags & MRF_SETATTR_LOCKED)) {
232 __u64 lockpart = MDS_INODELOCK_UPDATE;
233 if (ma->ma_attr.la_valid & (LA_MODE|LA_UID|LA_GID))
234 lockpart |= MDS_INODELOCK_LOOKUP;
236 rc = mdt_object_lock(info, mo, lh, lockpart, MDT_LOCAL_LOCK);
241 /* Setattrs are syncronized through dlm lock taken above. If another
242 * epoch started, its attributes may be already flushed on disk,
244 if (som_update && (info->mti_epoch->ioepoch != mo->mot_ioepoch))
245 GOTO(out_unlock, rc = 0);
247 if (mdt_object_exists(mo) == 0)
248 GOTO(out_unlock, rc = -ENOENT);
250 /* all attrs are packed into mti_attr in unpack_setattr */
251 mdt_fail_write(info->mti_env, info->mti_mdt->mdt_bottom,
252 OBD_FAIL_MDS_REINT_SETATTR_WRITE);
254 /* This is only for set ctime when rename's source is on remote MDS. */
255 if (unlikely(ma->ma_attr.la_valid == LA_CTIME))
256 ma->ma_attr_flags |= MDS_VTX_BYPASS;
258 /* all attrs are packed into mti_attr in unpack_setattr */
259 rc = mo_attr_set(info->mti_env, mdt_object_child(mo), ma);
261 GOTO(out_unlock, rc);
263 /* Re-enable SIZEONMDS. */
265 CDEBUG(D_INODE, "Closing epoch "LPU64" on "DFID". Count %d\n",
266 mo->mot_ioepoch, PFID(mdt_object_fid(mo)),
268 mdt_sizeonmds_enable(info, mo);
273 mdt_object_unlock(info, mo, lh, rc);
277 static int mdt_reint_setattr(struct mdt_thread_info *info,
278 struct mdt_lock_handle *lhc)
280 struct mdt_device *mdt = info->mti_mdt;
281 struct md_attr *ma = &info->mti_attr;
282 struct mdt_reint_record *rr = &info->mti_rr;
283 struct ptlrpc_request *req = mdt_info_req(info);
284 struct mdt_export_data *med = &req->rq_export->exp_mdt_data;
285 struct mdt_file_data *mfd;
286 struct mdt_object *mo;
287 struct md_object *next;
288 struct mdt_body *repbody;
292 DEBUG_REQ(D_INODE, req, "setattr "DFID" %x", PFID(rr->rr_fid1),
293 (unsigned int)ma->ma_attr.la_valid);
295 if (info->mti_dlm_req)
296 ldlm_request_cancel(req, info->mti_dlm_req, 0);
298 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
299 mo = mdt_object_find(info->mti_env, info->mti_mdt, rr->rr_fid1);
301 GOTO(out, rc = PTR_ERR(mo));
303 if (!(mdt_conn_flags(info) & OBD_CONNECT_SOM)) {
304 if ((ma->ma_attr.la_valid & LA_SIZE) ||
305 (rr->rr_flags & MRF_SETATTR_LOCKED)) {
306 /* Check write access for the O_TRUNC case */
307 if (mdt_write_read(info->mti_mdt, mo) < 0)
308 GOTO(out_put, rc = -ETXTBSY);
310 } else if (info->mti_epoch &&
311 (info->mti_epoch->flags & MF_EPOCH_OPEN)) {
313 rc = mdt_write_get(info->mti_mdt, mo);
319 GOTO(out_put, rc = -ENOMEM);
321 mdt_epoch_open(info, mo);
322 repbody->ioepoch = mo->mot_ioepoch;
324 mdt_object_get(info->mti_env, mo);
325 mdt_mfd_set_mode(mfd, FMODE_EPOCHLCK);
326 mfd->mfd_object = mo;
327 mfd->mfd_xid = req->rq_xid;
329 spin_lock(&med->med_open_lock);
330 list_add(&mfd->mfd_list, &med->med_open_head);
331 spin_unlock(&med->med_open_lock);
332 repbody->handle.cookie = mfd->mfd_handle.h_cookie;
335 if (info->mti_epoch && (info->mti_epoch->flags & MF_SOM_CHANGE))
336 ma->ma_attr_flags |= MDS_PERM_BYPASS | MDS_SOM;
338 rc = mdt_attr_set(info, mo, rr->rr_flags);
342 if (info->mti_epoch && (info->mti_epoch->flags & MF_SOM_CHANGE)) {
343 LASSERT(mdt_conn_flags(info) & OBD_CONNECT_SOM);
344 LASSERT(info->mti_epoch);
346 spin_lock(&med->med_open_lock);
347 /* Size-on-MDS Update. Find and free mfd. */
348 mfd = mdt_handle2mfd(info, &info->mti_epoch->handle);
350 spin_unlock(&med->med_open_lock);
351 CDEBUG(D_INODE | D_ERROR, "no handle for file close: "
352 "fid = "DFID": cookie = "LPX64"\n",
353 PFID(info->mti_rr.rr_fid1),
354 info->mti_epoch->handle.cookie);
355 GOTO(out_put, rc = -ESTALE);
357 LASSERT(mfd->mfd_mode == FMODE_SOM);
358 LASSERT(!(info->mti_epoch->flags & MF_EPOCH_CLOSE));
360 class_handle_unhash(&mfd->mfd_handle);
361 list_del_init(&mfd->mfd_list);
362 spin_unlock(&med->med_open_lock);
364 mdt_mfd_close(info, mfd);
367 ma->ma_need = MA_INODE;
369 next = mdt_object_child(mo);
370 rc = mo_attr_get(info->mti_env, next, ma);
374 mdt_pack_attr2body(info, repbody, &ma->ma_attr, mdt_object_fid(mo));
376 if (mdt->mdt_opts.mo_oss_capa &&
377 S_ISREG(lu_object_attr(&mo->mot_obj.mo_lu)) &&
378 (ma->ma_attr.la_valid & LA_SIZE)) {
379 struct lustre_capa *capa;
381 capa = req_capsule_server_get(info->mti_pill, &RMF_CAPA2);
383 capa->lc_opc = CAPA_OPC_OSS_DEFAULT | CAPA_OPC_OSS_TRUNC;
384 rc = mo_capa_get(info->mti_env, mdt_object_child(mo), capa, 0);
387 repbody->valid |= OBD_MD_FLOSSCAPA;
392 mdt_object_put(info->mti_env, mo);
394 mdt_shrink_reply(info);
398 static int mdt_reint_create(struct mdt_thread_info *info,
399 struct mdt_lock_handle *lhc)
404 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_REINT_CREATE))
405 RETURN(err_serious(-ESTALE));
407 if (info->mti_dlm_req)
408 ldlm_request_cancel(mdt_info_req(info), info->mti_dlm_req, 0);
410 switch (info->mti_attr.ma_attr.la_mode & S_IFMT) {
412 /* Cross-ref case. */
413 if (info->mti_cross_ref) {
414 rc = mdt_md_mkobj(info);
424 /* Special file should stay on the same node as parent. */
425 LASSERT(info->mti_rr.rr_namelen > 0);
426 rc = mdt_md_create(info);
430 rc = err_serious(-EOPNOTSUPP);
435 static int mdt_reint_unlink(struct mdt_thread_info *info,
436 struct mdt_lock_handle *lhc)
438 struct mdt_reint_record *rr = &info->mti_rr;
439 struct ptlrpc_request *req = mdt_info_req(info);
440 struct md_attr *ma = &info->mti_attr;
441 struct lu_fid *child_fid = &info->mti_tmp_fid1;
442 struct mdt_object *mp;
443 struct mdt_object *mc;
444 struct mdt_lock_handle *parent_lh;
445 struct mdt_lock_handle *child_lh;
446 struct lu_name *lname;
450 DEBUG_REQ(D_INODE, req, "unlink "DFID"/%s", PFID(rr->rr_fid1),
453 if (info->mti_dlm_req)
454 ldlm_request_cancel(req, info->mti_dlm_req, 0);
456 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_REINT_UNLINK))
457 RETURN(err_serious(-ENOENT));
459 /* step 1: lock the parent */
460 parent_lh = &info->mti_lh[MDT_LH_PARENT];
461 mdt_lock_pdo_init(parent_lh, LCK_PW, rr->rr_name,
464 mp = mdt_object_find_lock(info, rr->rr_fid1, parent_lh,
465 MDS_INODELOCK_UPDATE);
468 /* errors are possible here in cross-ref cases, see below */
469 if (info->mti_cross_ref)
474 mdt_reint_init_ma(info, ma);
475 if (!ma->ma_lmm || !ma->ma_cookie)
476 GOTO(out_unlock_parent, rc = -EINVAL);
478 if (info->mti_cross_ref) {
480 * Remote partial operation. It is possible that replay may
481 * happen on parent MDT and this operation will be repeated.
482 * Therefore the object absense is allowed case and nothing
483 * should be done here.
485 if (mdt_object_exists(mp) > 0) {
486 mdt_set_capainfo(info, 0, rr->rr_fid1, BYPASS_CAPA);
487 rc = mo_ref_del(info->mti_env,
488 mdt_object_child(mp), ma);
489 mdt_handle_last_unlink(info, mp, ma);
492 GOTO(out_unlock_parent, rc);
495 /* step 2: find & lock the child */
496 lname = mdt_name(info->mti_env, (char *)rr->rr_name, rr->rr_namelen);
497 rc = mdo_lookup(info->mti_env, mdt_object_child(mp),
498 lname, child_fid, &info->mti_spec);
500 GOTO(out_unlock_parent, rc);
502 /* We will lock the child regardless it is local or remote. No harm. */
503 mc = mdt_object_find(info->mti_env, info->mti_mdt, child_fid);
505 GOTO(out_unlock_parent, rc = PTR_ERR(mc));
506 child_lh = &info->mti_lh[MDT_LH_CHILD];
507 mdt_lock_reg_init(child_lh, LCK_EX);
508 rc = mdt_object_lock(info, mc, child_lh, MDS_INODELOCK_FULL,
511 mdt_object_put(info->mti_env, mc);
512 GOTO(out_unlock_parent, rc);
515 mdt_fail_write(info->mti_env, info->mti_mdt->mdt_bottom,
516 OBD_FAIL_MDS_REINT_UNLINK_WRITE);
519 * Now we can only make sure we need MA_INODE, in mdd layer, will check
520 * whether need MA_LOV and MA_COOKIE.
522 ma->ma_need = MA_INODE;
524 mdt_set_capainfo(info, 1, child_fid, BYPASS_CAPA);
525 rc = mdo_unlink(info->mti_env, mdt_object_child(mp),
526 mdt_object_child(mc), lname, ma);
528 mdt_handle_last_unlink(info, mc, ma);
531 mdt_object_unlock_put(info, mc, child_lh, rc);
533 mdt_object_unlock_put(info, mp, parent_lh, rc);
538 static int mdt_reint_link(struct mdt_thread_info *info,
539 struct mdt_lock_handle *lhc)
541 struct mdt_reint_record *rr = &info->mti_rr;
542 struct ptlrpc_request *req = mdt_info_req(info);
543 struct md_attr *ma = &info->mti_attr;
544 struct mdt_object *ms;
545 struct mdt_object *mp;
546 struct mdt_lock_handle *lhs;
547 struct mdt_lock_handle *lhp;
548 struct lu_name *lname;
552 DEBUG_REQ(D_INODE, req, "link "DFID" to "DFID"/%s",
553 PFID(rr->rr_fid1), PFID(rr->rr_fid2), rr->rr_name);
555 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_REINT_LINK))
556 RETURN(err_serious(-ENOENT));
558 if (info->mti_dlm_req)
559 ldlm_request_cancel(req, info->mti_dlm_req, 0);
561 if (info->mti_cross_ref) {
562 /* MDT holding name ask us to add ref. */
563 lhs = &info->mti_lh[MDT_LH_CHILD];
564 mdt_lock_reg_init(lhs, LCK_EX);
565 ms = mdt_object_find_lock(info, rr->rr_fid1, lhs,
566 MDS_INODELOCK_UPDATE);
570 mdt_set_capainfo(info, 0, rr->rr_fid1, BYPASS_CAPA);
571 rc = mo_ref_add(info->mti_env, mdt_object_child(ms), ma);
572 mdt_object_unlock_put(info, ms, lhs, rc);
576 /* Invalid case so return error immediately instead of
578 if (lu_fid_eq(rr->rr_fid1, rr->rr_fid2))
581 /* step 1: find & lock the target parent dir */
582 lhp = &info->mti_lh[MDT_LH_PARENT];
583 mdt_lock_pdo_init(lhp, LCK_EX, rr->rr_name,
585 mp = mdt_object_find_lock(info, rr->rr_fid2, lhp,
586 MDS_INODELOCK_UPDATE);
590 /* step 2: find & lock the source */
591 lhs = &info->mti_lh[MDT_LH_CHILD];
592 mdt_lock_reg_init(lhs, LCK_EX);
594 ms = mdt_object_find(info->mti_env, info->mti_mdt, rr->rr_fid1);
596 GOTO(out_unlock_parent, rc = PTR_ERR(ms));
598 rc = mdt_object_lock(info, ms, lhs, MDS_INODELOCK_UPDATE,
601 mdt_object_put(info->mti_env, ms);
602 GOTO(out_unlock_parent, rc);
605 /* step 3: link it */
606 mdt_fail_write(info->mti_env, info->mti_mdt->mdt_bottom,
607 OBD_FAIL_MDS_REINT_LINK_WRITE);
609 lname = mdt_name(info->mti_env, (char *)rr->rr_name, rr->rr_namelen);
610 rc = mdo_link(info->mti_env, mdt_object_child(mp),
611 mdt_object_child(ms), lname, ma);
614 mdt_object_unlock_put(info, ms, lhs, rc);
616 mdt_object_unlock_put(info, mp, lhp, rc);
620 /* partial operation for rename */
621 static int mdt_reint_rename_tgt(struct mdt_thread_info *info)
623 struct mdt_reint_record *rr = &info->mti_rr;
624 struct ptlrpc_request *req = mdt_info_req(info);
625 struct md_attr *ma = &info->mti_attr;
626 struct mdt_object *mtgtdir;
627 struct mdt_object *mtgt = NULL;
628 struct mdt_lock_handle *lh_tgtdir;
629 struct mdt_lock_handle *lh_tgt = NULL;
630 struct lu_fid *tgt_fid = &info->mti_tmp_fid1;
631 struct lu_name *lname;
635 DEBUG_REQ(D_INODE, req, "rename_tgt: insert (%s->"DFID") in "DFID,
636 rr->rr_tgt, PFID(rr->rr_fid2), PFID(rr->rr_fid1));
638 /* step 1: lookup & lock the tgt dir. */
639 lh_tgtdir = &info->mti_lh[MDT_LH_PARENT];
640 mdt_lock_pdo_init(lh_tgtdir, LCK_PW, rr->rr_tgt,
642 mtgtdir = mdt_object_find_lock(info, rr->rr_fid1, lh_tgtdir,
643 MDS_INODELOCK_UPDATE);
645 RETURN(PTR_ERR(mtgtdir));
647 /* step 2: find & lock the target object if exists. */
648 mdt_set_capainfo(info, 0, rr->rr_fid1, BYPASS_CAPA);
649 lname = mdt_name(info->mti_env, (char *)rr->rr_tgt, rr->rr_tgtlen);
650 rc = mdo_lookup(info->mti_env, mdt_object_child(mtgtdir),
651 lname, tgt_fid, &info->mti_spec);
652 if (rc != 0 && rc != -ENOENT) {
653 GOTO(out_unlock_tgtdir, rc);
654 } else if (rc == 0) {
656 * In case of replay that name can be already inserted, check
657 * that and do nothing if so.
659 if (lu_fid_eq(tgt_fid, rr->rr_fid2))
660 GOTO(out_unlock_tgtdir, rc);
662 lh_tgt = &info->mti_lh[MDT_LH_CHILD];
663 mdt_lock_reg_init(lh_tgt, LCK_EX);
665 mtgt = mdt_object_find_lock(info, tgt_fid, lh_tgt,
666 MDS_INODELOCK_LOOKUP);
668 GOTO(out_unlock_tgtdir, rc = PTR_ERR(mtgt));
670 mdt_reint_init_ma(info, ma);
671 if (!ma->ma_lmm || !ma->ma_cookie)
672 GOTO(out_unlock_tgt, rc = -EINVAL);
674 rc = mdo_rename_tgt(info->mti_env, mdt_object_child(mtgtdir),
675 mdt_object_child(mtgt), rr->rr_fid2,
677 } else /* -ENOENT */ {
678 rc = mdo_name_insert(info->mti_env, mdt_object_child(mtgtdir),
679 lname, rr->rr_fid2, ma);
682 /* handle last link of tgt object */
684 mdt_handle_last_unlink(info, mtgt, ma);
689 mdt_object_unlock_put(info, mtgt, lh_tgt, rc);
691 mdt_object_unlock_put(info, mtgtdir, lh_tgtdir, rc);
695 static int mdt_rename_lock(struct mdt_thread_info *info,
696 struct lustre_handle *lh)
698 struct ldlm_namespace *ns = info->mti_mdt->mdt_namespace;
699 ldlm_policy_data_t *policy = &info->mti_policy;
700 struct ldlm_res_id *res_id = &info->mti_res_id;
706 * Disable global rename BFL lock temporarily because
707 * when a mds do rename recoverying, which might enqueue
708 * BFL lock to the controller mds. and this req might be
709 * replay req for controller mds. but we did not have
710 * such handling in controller mds. XXX
713 ls = info->mti_mdt->mdt_md_dev.md_lu_dev.ld_site;
714 fid_build_reg_res_name(&LUSTRE_BFL_FID, res_id);
716 memset(policy, 0, sizeof *policy);
717 policy->l_inodebits.bits = MDS_INODELOCK_UPDATE;
719 if (ls->ls_control_exp == NULL) {
720 int flags = LDLM_FL_LOCAL_ONLY | LDLM_FL_ATOMIC_CB;
723 * Current node is controller, that is mdt0, where we should
726 rc = ldlm_cli_enqueue_local(ns, res_id, LDLM_IBITS, policy,
727 LCK_EX, &flags, ldlm_blocking_ast,
728 ldlm_completion_ast, NULL, NULL, 0,
731 struct ldlm_enqueue_info einfo = { LDLM_IBITS, LCK_EX,
732 ldlm_blocking_ast, ldlm_completion_ast, NULL, NULL };
736 * This is the case mdt0 is remote node, issue DLM lock like
739 rc = ldlm_cli_enqueue(ls->ls_control_exp, NULL, &einfo, res_id,
740 policy, &flags, NULL, 0, NULL, lh, 0);
746 static void mdt_rename_unlock(struct lustre_handle *lh)
749 /* Disable global rename BFL lock temporarily. see above XXX*/
752 LASSERT(lustre_handle_is_used(lh));
753 ldlm_lock_decref(lh, LCK_EX);
758 * This is is_subdir() variant, it is CMD if cmm forwards it to correct
759 * target. Source should not be ancestor of target dir. May be other rename
760 * checks can be moved here later.
762 static int mdt_rename_sanity(struct mdt_thread_info *info, struct lu_fid *fid)
764 struct mdt_reint_record *rr = &info->mti_rr;
765 struct lu_fid dst_fid = *rr->rr_fid2;
766 struct mdt_object *dst;
771 LASSERT(fid_is_sane(&dst_fid));
772 dst = mdt_object_find(info->mti_env, info->mti_mdt, &dst_fid);
774 rc = mdo_is_subdir(info->mti_env,
775 mdt_object_child(dst), fid,
777 mdt_object_put(info->mti_env, dst);
778 if (rc != -EREMOTE && rc < 0) {
779 CERROR("Failed mdo_is_subdir(), rc %d\n", rc);
781 /* check the found fid */
782 if (lu_fid_eq(&dst_fid, fid))
788 } while (rc == -EREMOTE);
793 static int mdt_reint_rename(struct mdt_thread_info *info,
794 struct mdt_lock_handle *lhc)
796 struct mdt_reint_record *rr = &info->mti_rr;
797 struct md_attr *ma = &info->mti_attr;
798 struct ptlrpc_request *req = mdt_info_req(info);
799 struct mdt_object *msrcdir;
800 struct mdt_object *mtgtdir;
801 struct mdt_object *mold;
802 struct mdt_object *mnew = NULL;
803 struct mdt_lock_handle *lh_srcdirp;
804 struct mdt_lock_handle *lh_tgtdirp;
805 struct mdt_lock_handle *lh_oldp;
806 struct mdt_lock_handle *lh_newp;
807 struct lu_fid *old_fid = &info->mti_tmp_fid1;
808 struct lu_fid *new_fid = &info->mti_tmp_fid2;
809 struct lustre_handle rename_lh = { 0 };
810 struct lu_name slname = { 0 };
811 struct lu_name *lname;
815 if (info->mti_dlm_req)
816 ldlm_request_cancel(mdt_info_req(info), info->mti_dlm_req, 0);
818 if (info->mti_cross_ref) {
819 rc = mdt_reint_rename_tgt(info);
823 DEBUG_REQ(D_INODE, req, "rename "DFID"/%s to "DFID"/%s",
824 PFID(rr->rr_fid1), rr->rr_name,
825 PFID(rr->rr_fid2), rr->rr_tgt);
827 rc = mdt_rename_lock(info, &rename_lh);
829 CERROR("Can't lock FS for rename, rc %d\n", rc);
833 lh_newp = &info->mti_lh[MDT_LH_NEW];
835 /* step 1: lock the source dir. */
836 lh_srcdirp = &info->mti_lh[MDT_LH_PARENT];
837 mdt_lock_pdo_init(lh_srcdirp, LCK_PW, rr->rr_name,
839 msrcdir = mdt_object_find_lock(info, rr->rr_fid1, lh_srcdirp,
840 MDS_INODELOCK_UPDATE);
842 GOTO(out_rename_lock, rc = PTR_ERR(msrcdir));
844 /* step 2: find & lock the target dir. */
845 lh_tgtdirp = &info->mti_lh[MDT_LH_CHILD];
846 mdt_lock_pdo_init(lh_tgtdirp, LCK_PW, rr->rr_tgt,
848 if (lu_fid_eq(rr->rr_fid1, rr->rr_fid2)) {
849 mdt_object_get(info->mti_env, msrcdir);
852 mtgtdir = mdt_object_find(info->mti_env, info->mti_mdt,
855 GOTO(out_unlock_source, rc = PTR_ERR(mtgtdir));
857 rc = mdt_object_exists(mtgtdir);
859 GOTO(out_unlock_target, rc = -ESTALE);
861 /* we lock the target dir if it is local */
862 rc = mdt_object_lock(info, mtgtdir, lh_tgtdirp,
863 MDS_INODELOCK_UPDATE,
866 GOTO(out_unlock_target, rc);
870 /* step 3: find & lock the old object. */
871 lname = mdt_name(info->mti_env, (char *)rr->rr_name, rr->rr_namelen);
872 mdt_name_copy(&slname, lname);
873 rc = mdo_lookup(info->mti_env, mdt_object_child(msrcdir),
874 &slname, old_fid, &info->mti_spec);
876 GOTO(out_unlock_target, rc);
878 if (lu_fid_eq(old_fid, rr->rr_fid1) || lu_fid_eq(old_fid, rr->rr_fid2))
879 GOTO(out_unlock_target, rc = -EINVAL);
881 mold = mdt_object_find(info->mti_env, info->mti_mdt, old_fid);
883 GOTO(out_unlock_target, rc = PTR_ERR(mold));
885 lh_oldp = &info->mti_lh[MDT_LH_OLD];
886 mdt_lock_reg_init(lh_oldp, LCK_EX);
887 rc = mdt_object_lock(info, mold, lh_oldp, MDS_INODELOCK_LOOKUP,
890 mdt_object_put(info->mti_env, mold);
891 GOTO(out_unlock_target, rc);
893 mdt_set_capainfo(info, 2, old_fid, BYPASS_CAPA);
895 /* step 4: find & lock the new object. */
896 /* new target object may not exist now */
897 lname = mdt_name(info->mti_env, (char *)rr->rr_tgt, rr->rr_tgtlen);
898 rc = mdo_lookup(info->mti_env, mdt_object_child(mtgtdir),
899 lname, new_fid, &info->mti_spec);
901 /* the new_fid should have been filled at this moment */
902 if (lu_fid_eq(old_fid, new_fid))
903 GOTO(out_unlock_old, rc);
905 if (lu_fid_eq(new_fid, rr->rr_fid1) ||
906 lu_fid_eq(new_fid, rr->rr_fid2))
907 GOTO(out_unlock_old, rc = -EINVAL);
909 mdt_lock_reg_init(lh_newp, LCK_EX);
910 mnew = mdt_object_find(info->mti_env, info->mti_mdt, new_fid);
912 GOTO(out_unlock_old, rc = PTR_ERR(mnew));
914 rc = mdt_object_lock(info, mnew, lh_newp,
915 MDS_INODELOCK_FULL, MDT_CROSS_LOCK);
917 mdt_object_put(info->mti_env, mnew);
918 GOTO(out_unlock_old, rc);
920 mdt_set_capainfo(info, 3, new_fid, BYPASS_CAPA);
921 } else if (rc != -EREMOTE && rc != -ENOENT)
922 GOTO(out_unlock_old, rc);
924 /* step 5: rename it */
925 mdt_reint_init_ma(info, ma);
926 if (!ma->ma_lmm || !ma->ma_cookie)
927 GOTO(out_unlock_new, rc = -EINVAL);
929 mdt_fail_write(info->mti_env, info->mti_mdt->mdt_bottom,
930 OBD_FAIL_MDS_REINT_RENAME_WRITE);
933 /* Check if @dst is subdir of @src. */
934 rc = mdt_rename_sanity(info, old_fid);
936 GOTO(out_unlock_new, rc);
938 rc = mdo_rename(info->mti_env, mdt_object_child(msrcdir),
939 mdt_object_child(mtgtdir), old_fid, &slname,
940 (mnew ? mdt_object_child(mnew) : NULL),
943 /* handle last link of tgt object */
945 mdt_handle_last_unlink(info, mnew, ma);
950 mdt_object_unlock_put(info, mnew, lh_newp, rc);
952 mdt_object_unlock_put(info, mold, lh_oldp, rc);
954 mdt_object_unlock_put(info, mtgtdir, lh_tgtdirp, rc);
956 mdt_object_unlock_put(info, msrcdir, lh_srcdirp, rc);
958 mdt_rename_unlock(&rename_lh);
962 typedef int (*mdt_reinter)(struct mdt_thread_info *info,
963 struct mdt_lock_handle *lhc);
965 static mdt_reinter reinters[REINT_MAX] = {
966 [REINT_SETATTR] = mdt_reint_setattr,
967 [REINT_CREATE] = mdt_reint_create,
968 [REINT_LINK] = mdt_reint_link,
969 [REINT_UNLINK] = mdt_reint_unlink,
970 [REINT_RENAME] = mdt_reint_rename,
971 [REINT_OPEN] = mdt_reint_open,
972 [REINT_SETXATTR] = mdt_reint_setxattr
975 int mdt_reint_rec(struct mdt_thread_info *info,
976 struct mdt_lock_handle *lhc)
981 rc = reinters[info->mti_rr.rr_opcode](info, lhc);