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 info->mti_exp->exp_connect_flags & OBD_CONNECT_MDS_CAPA) {
85 struct lustre_capa *capa;
87 capa = req_capsule_server_get(info->mti_pill, &RMF_CAPA1);
89 capa->lc_opc = CAPA_OPC_MDS_DEFAULT;
90 rc = mo_capa_get(info->mti_env, mdt_object_child(object), capa,
93 repbody->valid |= OBD_MD_FLMDSCAPA;
99 int mdt_version_get_check(struct mdt_thread_info *info, int index)
101 /** version recovery */
102 struct md_object *mo;
103 struct ptlrpc_request *req = mdt_info_req(info);
104 __u64 curr_version, *pre_versions;
107 if (!exp_connect_vbr(req->rq_export))
110 LASSERT(info->mti_mos[index]);
111 LASSERT(mdt_object_exists(info->mti_mos[index]));
112 mo = mdt_object_child(info->mti_mos[index]);
114 curr_version = mo_version_get(info->mti_env, mo);
115 CDEBUG(D_INODE, "Version is "LPX64"\n", curr_version);
116 /** VBR: version is checked always because costs nothing */
117 if (lustre_msg_get_transno(req->rq_reqmsg) != 0) {
118 pre_versions = lustre_msg_get_versions(req->rq_reqmsg);
119 LASSERT(index < PTLRPC_NUM_VERSIONS);
120 /** Sanity check for malformed buffers */
121 if (pre_versions == 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_versions[index] != curr_version) {
128 CDEBUG(D_INODE, "Version mismatch "LPX64" != "LPX64"\n",
129 pre_versions[index], curr_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);
136 /** save pre-versions in reply */
137 LASSERT(req->rq_repmsg != NULL);
138 pre_versions = lustre_msg_get_versions(req->rq_repmsg);
140 pre_versions[index] = curr_version;
144 static int mdt_md_create(struct mdt_thread_info *info)
146 struct mdt_device *mdt = info->mti_mdt;
147 struct mdt_object *parent;
148 struct mdt_object *child;
149 struct mdt_lock_handle *lh;
150 struct mdt_body *repbody;
151 struct md_attr *ma = &info->mti_attr;
152 struct mdt_reint_record *rr = &info->mti_rr;
153 struct lu_name *lname;
157 DEBUG_REQ(D_INODE, mdt_info_req(info), "Create (%s->"DFID") in "DFID,
158 rr->rr_name, PFID(rr->rr_fid2), PFID(rr->rr_fid1));
160 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
162 lh = &info->mti_lh[MDT_LH_PARENT];
163 mdt_lock_pdo_init(lh, LCK_PW, rr->rr_name, rr->rr_namelen);
165 parent = mdt_object_find_lock(info, rr->rr_fid1, lh,
166 MDS_INODELOCK_UPDATE);
168 RETURN(PTR_ERR(parent));
170 child = mdt_object_find(info->mti_env, mdt, rr->rr_fid2);
171 if (likely(!IS_ERR(child))) {
172 struct md_object *next = mdt_object_child(parent);
174 ma->ma_need = MA_INODE;
176 /* capa for cross-ref will be stored here */
177 ma->ma_capa = req_capsule_server_get(info->mti_pill,
179 LASSERT(ma->ma_capa);
181 mdt_fail_write(info->mti_env, info->mti_mdt->mdt_bottom,
182 OBD_FAIL_MDS_REINT_CREATE_WRITE);
184 info->mti_mos[0] = parent;
185 info->mti_mos[1] = child;
186 rc = mdt_version_get_check(info, 0);
188 GOTO(out_put_child, rc);
190 /* Let lower layer know current lock mode. */
191 info->mti_spec.sp_cr_mode =
192 mdt_dlm_mode2mdl_mode(lh->mlh_pdo_mode);
195 * Do perform lookup sanity check. We do not know if name exists
198 info->mti_spec.sp_cr_lookup = 1;
199 info->mti_spec.sp_feat = &dt_directory_features;
201 lname = mdt_name(info->mti_env, (char *)rr->rr_name,
203 rc = mdo_create(info->mti_env, next, lname,
204 mdt_object_child(child),
205 &info->mti_spec, ma);
207 /* Return fid & attr to client. */
208 if (ma->ma_valid & MA_INODE)
209 mdt_pack_attr2body(info, repbody, &ma->ma_attr,
210 mdt_object_fid(child));
213 mdt_object_put(info->mti_env, child);
217 mdt_create_pack_capa(info, rc, child, repbody);
218 mdt_object_unlock_put(info, parent, lh, rc);
222 /* Partial request to create object only */
223 static int mdt_md_mkobj(struct mdt_thread_info *info)
225 struct mdt_device *mdt = info->mti_mdt;
226 struct mdt_object *o;
227 struct mdt_body *repbody;
228 struct md_attr *ma = &info->mti_attr;
232 DEBUG_REQ(D_INODE, mdt_info_req(info), "Partial create "DFID"",
233 PFID(info->mti_rr.rr_fid2));
235 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
237 o = mdt_object_find(info->mti_env, mdt, info->mti_rr.rr_fid2);
239 struct md_object *next = mdt_object_child(o);
241 ma->ma_need = MA_INODE;
245 * Cross-ref create can encounter already created obj in case of
246 * recovery, just get attr in that case.
248 if (mdt_object_exists(o) == 1) {
249 rc = mo_attr_get(info->mti_env, next, ma);
252 * Here, NO permission check for object_create,
253 * such check has been done on the original MDS.
255 rc = mo_object_create(info->mti_env, next,
256 &info->mti_spec, ma);
259 /* Return fid & attr to client. */
260 if (ma->ma_valid & MA_INODE)
261 mdt_pack_attr2body(info, repbody, &ma->ma_attr,
264 mdt_object_put(info->mti_env, o);
268 mdt_create_pack_capa(info, rc, o, repbody);
272 /* In the raw-setattr case, we lock the child inode.
273 * In the write-back case or if being called from open,
274 * the client holds a lock already.
275 * We use the ATTR_FROM_OPEN (translated into MRF_SETATTR_LOCKED by
276 * mdt_setattr_unpack()) flag to tell these cases apart. */
277 int mdt_attr_set(struct mdt_thread_info *info, struct mdt_object *mo, int flags)
279 struct md_attr *ma = &info->mti_attr;
280 struct mdt_lock_handle *lh;
282 int do_vbr = ma->ma_attr.la_valid & (LA_MODE|LA_UID|LA_GID|LA_FLAGS);
286 /* attr shouldn't be set on remote object */
287 LASSERT(mdt_object_exists(mo) >= 0);
289 if (exp_connect_som(info->mti_exp) && info->mti_epoch)
290 som_update = (info->mti_epoch->flags & MF_SOM_CHANGE);
292 /* Try to avoid object_lock if another epoch has been started
294 if (som_update && (info->mti_epoch->ioepoch != mo->mot_ioepoch))
297 lh = &info->mti_lh[MDT_LH_PARENT];
298 mdt_lock_reg_init(lh, LCK_PW);
300 if (!(flags & MRF_SETATTR_LOCKED)) {
301 __u64 lockpart = MDS_INODELOCK_UPDATE;
302 if (ma->ma_attr.la_valid & (LA_MODE|LA_UID|LA_GID))
303 lockpart |= MDS_INODELOCK_LOOKUP;
305 rc = mdt_object_lock(info, mo, lh, lockpart, MDT_LOCAL_LOCK);
310 /* Setattrs are syncronized through dlm lock taken above. If another
311 * epoch started, its attributes may be already flushed on disk,
313 if (som_update && (info->mti_epoch->ioepoch != mo->mot_ioepoch))
314 GOTO(out_unlock, rc = 0);
316 if (mdt_object_exists(mo) == 0)
317 GOTO(out_unlock, rc = -ENOENT);
319 /* all attrs are packed into mti_attr in unpack_setattr */
320 mdt_fail_write(info->mti_env, info->mti_mdt->mdt_bottom,
321 OBD_FAIL_MDS_REINT_SETATTR_WRITE);
323 /* This is only for set ctime when rename's source is on remote MDS. */
324 if (unlikely(ma->ma_attr.la_valid == LA_CTIME))
325 ma->ma_attr_flags |= MDS_VTX_BYPASS;
327 /* VBR: update version if attr changed are important for recovery */
329 info->mti_mos[0] = mo;
330 rc = mdt_version_get_check(info, 0);
332 GOTO(out_unlock, rc);
335 /* all attrs are packed into mti_attr in unpack_setattr */
336 rc = mo_attr_set(info->mti_env, mdt_object_child(mo), ma);
338 GOTO(out_unlock, rc);
340 /* Re-enable SIZEONMDS. */
342 CDEBUG(D_INODE, "Closing epoch "LPU64" on "DFID". Count %d\n",
343 mo->mot_ioepoch, PFID(mdt_object_fid(mo)),
345 mdt_sizeonmds_enable(info, mo);
350 mdt_object_unlock(info, mo, lh, rc);
354 static int mdt_reint_setattr(struct mdt_thread_info *info,
355 struct mdt_lock_handle *lhc)
357 struct md_attr *ma = &info->mti_attr;
358 struct mdt_reint_record *rr = &info->mti_rr;
359 struct ptlrpc_request *req = mdt_info_req(info);
360 struct mdt_export_data *med = &req->rq_export->exp_mdt_data;
361 struct mdt_file_data *mfd;
362 struct mdt_object *mo;
363 struct md_object *next;
364 struct mdt_body *repbody;
368 DEBUG_REQ(D_INODE, req, "setattr "DFID" %x", PFID(rr->rr_fid1),
369 (unsigned int)ma->ma_attr.la_valid);
371 if (info->mti_dlm_req)
372 ldlm_request_cancel(req, info->mti_dlm_req, 0);
374 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
375 mo = mdt_object_find(info->mti_env, info->mti_mdt, rr->rr_fid1);
377 GOTO(out, rc = PTR_ERR(mo));
379 /* start a log jounal handle if needed */
380 if (!(mdt_conn_flags(info) & OBD_CONNECT_SOM)) {
381 if ((ma->ma_attr.la_valid & LA_SIZE) ||
382 (rr->rr_flags & MRF_SETATTR_LOCKED)) {
383 /* Check write access for the O_TRUNC case */
384 if (mdt_write_read(info->mti_mdt, mo) < 0)
385 GOTO(out_put, rc = -ETXTBSY);
387 } else if (info->mti_epoch &&
388 (info->mti_epoch->flags & MF_EPOCH_OPEN)) {
390 rc = mdt_write_get(info->mti_mdt, mo);
396 GOTO(out_put, rc = -ENOMEM);
398 mdt_epoch_open(info, mo);
399 repbody->ioepoch = mo->mot_ioepoch;
401 mdt_object_get(info->mti_env, mo);
402 mdt_mfd_set_mode(mfd, FMODE_EPOCHLCK);
403 mfd->mfd_object = mo;
404 mfd->mfd_xid = req->rq_xid;
406 spin_lock(&med->med_open_lock);
407 list_add(&mfd->mfd_list, &med->med_open_head);
408 spin_unlock(&med->med_open_lock);
409 repbody->handle.cookie = mfd->mfd_handle.h_cookie;
412 if (info->mti_epoch && (info->mti_epoch->flags & MF_SOM_CHANGE))
413 ma->ma_attr_flags |= MDS_PERM_BYPASS | MDS_SOM;
415 rc = mdt_attr_set(info, mo, rr->rr_flags);
419 if (info->mti_epoch && (info->mti_epoch->flags & MF_SOM_CHANGE)) {
420 LASSERT(mdt_conn_flags(info) & OBD_CONNECT_SOM);
421 LASSERT(info->mti_epoch);
423 spin_lock(&med->med_open_lock);
424 /* Size-on-MDS Update. Find and free mfd. */
425 mfd = mdt_handle2mfd(info, &info->mti_epoch->handle);
427 spin_unlock(&med->med_open_lock);
428 CDEBUG(D_INODE | D_ERROR, "no handle for file close: "
429 "fid = "DFID": cookie = "LPX64"\n",
430 PFID(info->mti_rr.rr_fid1),
431 info->mti_epoch->handle.cookie);
432 GOTO(out_put, rc = -ESTALE);
434 LASSERT(mfd->mfd_mode == FMODE_SOM);
435 LASSERT(!(info->mti_epoch->flags & MF_EPOCH_CLOSE));
437 class_handle_unhash(&mfd->mfd_handle);
438 list_del_init(&mfd->mfd_list);
439 spin_unlock(&med->med_open_lock);
441 mdt_mfd_close(info, mfd);
444 ma->ma_need = MA_INODE;
446 next = mdt_object_child(mo);
447 rc = mo_attr_get(info->mti_env, next, ma);
451 mdt_pack_attr2body(info, repbody, &ma->ma_attr, mdt_object_fid(mo));
453 if (info->mti_mdt->mdt_opts.mo_oss_capa &&
454 info->mti_exp->exp_connect_flags & OBD_CONNECT_OSS_CAPA &&
455 S_ISREG(lu_object_attr(&mo->mot_obj.mo_lu)) &&
456 (ma->ma_attr.la_valid & LA_SIZE)) {
457 struct lustre_capa *capa;
459 capa = req_capsule_server_get(info->mti_pill, &RMF_CAPA2);
461 capa->lc_opc = CAPA_OPC_OSS_DEFAULT | CAPA_OPC_OSS_TRUNC;
462 rc = mo_capa_get(info->mti_env, mdt_object_child(mo), capa, 0);
465 repbody->valid |= OBD_MD_FLOSSCAPA;
470 mdt_object_put(info->mti_env, mo);
472 mdt_shrink_reply(info);
476 static int mdt_reint_create(struct mdt_thread_info *info,
477 struct mdt_lock_handle *lhc)
482 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_REINT_CREATE))
483 RETURN(err_serious(-ESTALE));
485 if (info->mti_dlm_req)
486 ldlm_request_cancel(mdt_info_req(info), info->mti_dlm_req, 0);
488 switch (info->mti_attr.ma_attr.la_mode & S_IFMT) {
490 /* Cross-ref case. */
491 if (info->mti_cross_ref) {
492 rc = mdt_md_mkobj(info);
502 /* Special file should stay on the same node as parent. */
503 LASSERT(info->mti_rr.rr_namelen > 0);
504 rc = mdt_md_create(info);
508 rc = err_serious(-EOPNOTSUPP);
513 static int mdt_reint_unlink(struct mdt_thread_info *info,
514 struct mdt_lock_handle *lhc)
516 struct mdt_reint_record *rr = &info->mti_rr;
517 struct ptlrpc_request *req = mdt_info_req(info);
518 struct md_attr *ma = &info->mti_attr;
519 struct lu_fid *child_fid = &info->mti_tmp_fid1;
520 struct mdt_object *mp;
521 struct mdt_object *mc;
522 struct mdt_lock_handle *parent_lh;
523 struct mdt_lock_handle *child_lh;
524 struct lu_name *lname;
528 DEBUG_REQ(D_INODE, req, "unlink "DFID"/%s", PFID(rr->rr_fid1),
531 if (info->mti_dlm_req)
532 ldlm_request_cancel(req, info->mti_dlm_req, 0);
534 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_REINT_UNLINK))
535 RETURN(err_serious(-ENOENT));
538 * step 1: lock the parent. Note, this may be child in case of
539 * remote operation denoted by ->mti_cross_ref flag.
541 parent_lh = &info->mti_lh[MDT_LH_PARENT];
542 if (info->mti_cross_ref) {
544 * Init reg lock for cross ref case when we need to do only
547 mdt_lock_reg_init(parent_lh, LCK_PW);
549 mdt_lock_pdo_init(parent_lh, LCK_PW, rr->rr_name,
552 mp = mdt_object_find_lock(info, rr->rr_fid1, parent_lh,
553 MDS_INODELOCK_UPDATE);
556 /* errors are possible here in cross-ref cases, see below */
557 if (info->mti_cross_ref)
562 info->mti_mos[0] = mp;
563 rc = mdt_version_get_check(info, 0);
565 GOTO(out_unlock_parent, rc);
567 mdt_reint_init_ma(info, ma);
568 if (!ma->ma_lmm || !ma->ma_cookie)
569 GOTO(out_unlock_parent, rc = -EINVAL);
571 if (info->mti_cross_ref) {
573 * Remote partial operation. It is possible that replay may
574 * happen on parent MDT and this operation will be repeated.
575 * Therefore the object absense is allowed case and nothing
576 * should be done here.
578 if (mdt_object_exists(mp) > 0) {
579 mdt_set_capainfo(info, 0, rr->rr_fid1, BYPASS_CAPA);
580 rc = mo_ref_del(info->mti_env,
581 mdt_object_child(mp), ma);
583 mdt_handle_last_unlink(info, mp, ma);
586 GOTO(out_unlock_parent, rc);
589 /* step 2: find & lock the child */
590 lname = mdt_name(info->mti_env, (char *)rr->rr_name, rr->rr_namelen);
591 rc = mdo_lookup(info->mti_env, mdt_object_child(mp),
592 lname, child_fid, &info->mti_spec);
594 GOTO(out_unlock_parent, rc);
596 /* We will lock the child regardless it is local or remote. No harm. */
597 mc = mdt_object_find(info->mti_env, info->mti_mdt, child_fid);
599 GOTO(out_unlock_parent, rc = PTR_ERR(mc));
600 child_lh = &info->mti_lh[MDT_LH_CHILD];
601 mdt_lock_reg_init(child_lh, LCK_EX);
602 rc = mdt_object_lock(info, mc, child_lh, MDS_INODELOCK_FULL,
605 mdt_object_put(info->mti_env, mc);
606 GOTO(out_unlock_parent, rc);
609 mdt_fail_write(info->mti_env, info->mti_mdt->mdt_bottom,
610 OBD_FAIL_MDS_REINT_UNLINK_WRITE);
612 info->mti_mos[1] = mc;
613 rc = mdt_version_get_check(info, 1);
615 GOTO(out_unlock_child, rc);
618 * Now we can only make sure we need MA_INODE, in mdd layer, will check
619 * whether need MA_LOV and MA_COOKIE.
621 ma->ma_need = MA_INODE;
623 mdt_set_capainfo(info, 1, child_fid, BYPASS_CAPA);
624 rc = mdo_unlink(info->mti_env, mdt_object_child(mp),
625 mdt_object_child(mc), lname, ma);
627 mdt_handle_last_unlink(info, mc, ma);
631 mdt_object_unlock_put(info, mc, child_lh, rc);
633 mdt_object_unlock_put(info, mp, parent_lh, rc);
638 static int mdt_reint_link(struct mdt_thread_info *info,
639 struct mdt_lock_handle *lhc)
641 struct mdt_reint_record *rr = &info->mti_rr;
642 struct ptlrpc_request *req = mdt_info_req(info);
643 struct md_attr *ma = &info->mti_attr;
644 struct mdt_object *ms;
645 struct mdt_object *mp;
646 struct mdt_lock_handle *lhs;
647 struct mdt_lock_handle *lhp;
648 struct lu_name *lname;
652 DEBUG_REQ(D_INODE, req, "link "DFID" to "DFID"/%s",
653 PFID(rr->rr_fid1), PFID(rr->rr_fid2), rr->rr_name);
655 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_REINT_LINK))
656 RETURN(err_serious(-ENOENT));
658 if (info->mti_dlm_req)
659 ldlm_request_cancel(req, info->mti_dlm_req, 0);
661 if (info->mti_cross_ref) {
662 /* MDT holding name ask us to add ref. */
663 lhs = &info->mti_lh[MDT_LH_CHILD];
664 mdt_lock_reg_init(lhs, LCK_EX);
665 ms = mdt_object_find_lock(info, rr->rr_fid1, lhs,
666 MDS_INODELOCK_UPDATE);
670 mdt_set_capainfo(info, 0, rr->rr_fid1, BYPASS_CAPA);
671 rc = mo_ref_add(info->mti_env, mdt_object_child(ms), ma);
672 mdt_object_unlock_put(info, ms, lhs, rc);
676 /* Invalid case so return error immediately instead of
678 if (lu_fid_eq(rr->rr_fid1, rr->rr_fid2))
681 /* step 1: find & lock the target parent dir */
682 lhp = &info->mti_lh[MDT_LH_PARENT];
683 mdt_lock_pdo_init(lhp, LCK_EX, rr->rr_name,
685 mp = mdt_object_find_lock(info, rr->rr_fid2, lhp,
686 MDS_INODELOCK_UPDATE);
690 info->mti_mos[0] = mp;
691 rc = mdt_version_get_check(info, 0);
693 GOTO(out_unlock_parent, rc);
695 /* step 2: find & lock the source */
696 lhs = &info->mti_lh[MDT_LH_CHILD];
697 mdt_lock_reg_init(lhs, LCK_EX);
699 ms = mdt_object_find(info->mti_env, info->mti_mdt, rr->rr_fid1);
701 GOTO(out_unlock_parent, rc = PTR_ERR(ms));
703 rc = mdt_object_lock(info, ms, lhs, MDS_INODELOCK_UPDATE,
706 mdt_object_put(info->mti_env, ms);
707 GOTO(out_unlock_parent, rc);
710 /* step 3: link it */
711 mdt_fail_write(info->mti_env, info->mti_mdt->mdt_bottom,
712 OBD_FAIL_MDS_REINT_LINK_WRITE);
714 info->mti_mos[1] = ms;
715 rc = mdt_version_get_check(info, 1);
717 GOTO(out_unlock_child, rc);
719 lname = mdt_name(info->mti_env, (char *)rr->rr_name, rr->rr_namelen);
720 rc = mdo_link(info->mti_env, mdt_object_child(mp),
721 mdt_object_child(ms), lname, ma);
725 mdt_object_unlock_put(info, ms, lhs, rc);
727 mdt_object_unlock_put(info, mp, lhp, rc);
731 /* partial operation for rename */
732 static int mdt_reint_rename_tgt(struct mdt_thread_info *info)
734 struct mdt_reint_record *rr = &info->mti_rr;
735 struct ptlrpc_request *req = mdt_info_req(info);
736 struct md_attr *ma = &info->mti_attr;
737 struct mdt_object *mtgtdir;
738 struct mdt_object *mtgt = NULL;
739 struct mdt_lock_handle *lh_tgtdir;
740 struct mdt_lock_handle *lh_tgt = NULL;
741 struct lu_fid *tgt_fid = &info->mti_tmp_fid1;
742 struct lu_name *lname;
746 DEBUG_REQ(D_INODE, req, "rename_tgt: insert (%s->"DFID") in "DFID,
747 rr->rr_tgt, PFID(rr->rr_fid2), PFID(rr->rr_fid1));
749 /* step 1: lookup & lock the tgt dir. */
750 lh_tgtdir = &info->mti_lh[MDT_LH_PARENT];
751 mdt_lock_pdo_init(lh_tgtdir, LCK_PW, rr->rr_tgt,
753 mtgtdir = mdt_object_find_lock(info, rr->rr_fid1, lh_tgtdir,
754 MDS_INODELOCK_UPDATE);
756 RETURN(PTR_ERR(mtgtdir));
758 /* step 2: find & lock the target object if exists. */
759 mdt_set_capainfo(info, 0, rr->rr_fid1, BYPASS_CAPA);
760 lname = mdt_name(info->mti_env, (char *)rr->rr_tgt, rr->rr_tgtlen);
761 rc = mdo_lookup(info->mti_env, mdt_object_child(mtgtdir),
762 lname, tgt_fid, &info->mti_spec);
763 if (rc != 0 && rc != -ENOENT) {
764 GOTO(out_unlock_tgtdir, rc);
765 } else if (rc == 0) {
767 * In case of replay that name can be already inserted, check
768 * that and do nothing if so.
770 if (lu_fid_eq(tgt_fid, rr->rr_fid2))
771 GOTO(out_unlock_tgtdir, rc);
773 lh_tgt = &info->mti_lh[MDT_LH_CHILD];
774 mdt_lock_reg_init(lh_tgt, LCK_EX);
776 mtgt = mdt_object_find_lock(info, tgt_fid, lh_tgt,
777 MDS_INODELOCK_LOOKUP);
779 GOTO(out_unlock_tgtdir, rc = PTR_ERR(mtgt));
781 mdt_reint_init_ma(info, ma);
782 if (!ma->ma_lmm || !ma->ma_cookie)
783 GOTO(out_unlock_tgt, rc = -EINVAL);
785 rc = mdo_rename_tgt(info->mti_env, mdt_object_child(mtgtdir),
786 mdt_object_child(mtgt), rr->rr_fid2,
788 } else /* -ENOENT */ {
789 rc = mdo_name_insert(info->mti_env, mdt_object_child(mtgtdir),
790 lname, rr->rr_fid2, ma);
793 /* handle last link of tgt object */
795 mdt_handle_last_unlink(info, mtgt, ma);
800 mdt_object_unlock_put(info, mtgt, lh_tgt, rc);
802 mdt_object_unlock_put(info, mtgtdir, lh_tgtdir, rc);
806 static int mdt_rename_lock(struct mdt_thread_info *info,
807 struct lustre_handle *lh)
809 struct ldlm_namespace *ns = info->mti_mdt->mdt_namespace;
810 ldlm_policy_data_t *policy = &info->mti_policy;
811 struct ldlm_res_id *res_id = &info->mti_res_id;
817 * Disable global rename BFL lock temporarily because
818 * when a mds do rename recoverying, which might enqueue
819 * BFL lock to the controller mds. and this req might be
820 * replay req for controller mds. but we did not have
821 * such handling in controller mds. XXX
824 ms = mdt_md_site(info->mti_mdt);
825 fid_build_reg_res_name(&LUSTRE_BFL_FID, res_id);
827 memset(policy, 0, sizeof *policy);
828 policy->l_inodebits.bits = MDS_INODELOCK_UPDATE;
830 if (ms->ms_control_exp == NULL) {
831 int flags = LDLM_FL_LOCAL_ONLY | LDLM_FL_ATOMIC_CB;
834 * Current node is controller, that is mdt0, where we should
837 rc = ldlm_cli_enqueue_local(ns, res_id, LDLM_IBITS, policy,
838 LCK_EX, &flags, ldlm_blocking_ast,
839 ldlm_completion_ast, NULL, NULL, 0,
841 &info->mti_exp->exp_handle.h_cookie,
844 struct ldlm_enqueue_info einfo = { LDLM_IBITS, LCK_EX,
845 ldlm_blocking_ast, ldlm_completion_ast, NULL, NULL, NULL };
849 * This is the case mdt0 is remote node, issue DLM lock like
852 rc = ldlm_cli_enqueue(ms->ms_control_exp,
853 NULL, &einfo, res_id,
854 policy, &flags, NULL, 0, NULL, lh, 0);
860 static void mdt_rename_unlock(struct lustre_handle *lh)
863 /* Disable global rename BFL lock temporarily. see above XXX*/
866 LASSERT(lustre_handle_is_used(lh));
867 ldlm_lock_decref(lh, LCK_EX);
872 * This is is_subdir() variant, it is CMD if cmm forwards it to correct
873 * target. Source should not be ancestor of target dir. May be other rename
874 * checks can be moved here later.
876 static int mdt_rename_sanity(struct mdt_thread_info *info, struct lu_fid *fid)
878 struct mdt_reint_record *rr = &info->mti_rr;
879 struct lu_fid dst_fid = *rr->rr_fid2;
880 struct mdt_object *dst;
885 LASSERT(fid_is_sane(&dst_fid));
886 dst = mdt_object_find(info->mti_env, info->mti_mdt, &dst_fid);
888 rc = mdo_is_subdir(info->mti_env,
889 mdt_object_child(dst), fid,
891 mdt_object_put(info->mti_env, dst);
892 if (rc != -EREMOTE && rc < 0) {
893 CERROR("Failed mdo_is_subdir(), rc %d\n", rc);
895 /* check the found fid */
896 if (lu_fid_eq(&dst_fid, fid))
902 } while (rc == -EREMOTE);
907 static int mdt_reint_rename(struct mdt_thread_info *info,
908 struct mdt_lock_handle *lhc)
910 struct mdt_reint_record *rr = &info->mti_rr;
911 struct md_attr *ma = &info->mti_attr;
912 struct ptlrpc_request *req = mdt_info_req(info);
913 struct mdt_object *msrcdir;
914 struct mdt_object *mtgtdir;
915 struct mdt_object *mold;
916 struct mdt_object *mnew = NULL;
917 struct mdt_lock_handle *lh_srcdirp;
918 struct mdt_lock_handle *lh_tgtdirp;
919 struct mdt_lock_handle *lh_oldp;
920 struct mdt_lock_handle *lh_newp;
921 struct lu_fid *old_fid = &info->mti_tmp_fid1;
922 struct lu_fid *new_fid = &info->mti_tmp_fid2;
923 struct lustre_handle rename_lh = { 0 };
924 struct lu_name slname = { 0 };
925 struct lu_name *lname;
929 if (info->mti_dlm_req)
930 ldlm_request_cancel(mdt_info_req(info), info->mti_dlm_req, 0);
932 if (info->mti_cross_ref) {
933 rc = mdt_reint_rename_tgt(info);
937 DEBUG_REQ(D_INODE, req, "rename "DFID"/%s to "DFID"/%s",
938 PFID(rr->rr_fid1), rr->rr_name,
939 PFID(rr->rr_fid2), rr->rr_tgt);
941 rc = mdt_rename_lock(info, &rename_lh);
943 CERROR("Can't lock FS for rename, rc %d\n", rc);
947 lh_newp = &info->mti_lh[MDT_LH_NEW];
949 /* step 1: lock the source dir. */
950 lh_srcdirp = &info->mti_lh[MDT_LH_PARENT];
951 mdt_lock_pdo_init(lh_srcdirp, LCK_PW, rr->rr_name,
953 msrcdir = mdt_object_find_lock(info, rr->rr_fid1, lh_srcdirp,
954 MDS_INODELOCK_UPDATE);
956 GOTO(out_rename_lock, rc = PTR_ERR(msrcdir));
958 info->mti_mos[0] = msrcdir;
959 rc = mdt_version_get_check(info, 0);
961 GOTO(out_unlock_source, rc);
963 /* step 2: find & lock the target dir. */
964 lh_tgtdirp = &info->mti_lh[MDT_LH_CHILD];
965 mdt_lock_pdo_init(lh_tgtdirp, LCK_PW, rr->rr_tgt,
967 if (lu_fid_eq(rr->rr_fid1, rr->rr_fid2)) {
968 mdt_object_get(info->mti_env, msrcdir);
971 mtgtdir = mdt_object_find(info->mti_env, info->mti_mdt,
974 GOTO(out_unlock_source, rc = PTR_ERR(mtgtdir));
976 rc = mdt_object_exists(mtgtdir);
978 GOTO(out_unlock_target, rc = -ESTALE);
980 /* we lock the target dir if it is local */
981 rc = mdt_object_lock(info, mtgtdir, lh_tgtdirp,
982 MDS_INODELOCK_UPDATE,
985 mdt_object_put(info->mti_env, mtgtdir);
986 GOTO(out_unlock_source, rc);
989 info->mti_mos[1] = mtgtdir;
990 rc = mdt_version_get_check(info, 1);
992 GOTO(out_unlock_target, rc);
996 /* step 3: find & lock the old object. */
997 lname = mdt_name(info->mti_env, (char *)rr->rr_name, rr->rr_namelen);
998 mdt_name_copy(&slname, lname);
999 rc = mdo_lookup(info->mti_env, mdt_object_child(msrcdir),
1000 &slname, old_fid, &info->mti_spec);
1002 GOTO(out_unlock_target, rc);
1004 if (lu_fid_eq(old_fid, rr->rr_fid1) || lu_fid_eq(old_fid, rr->rr_fid2))
1005 GOTO(out_unlock_target, rc = -EINVAL);
1007 mold = mdt_object_find(info->mti_env, info->mti_mdt, old_fid);
1009 GOTO(out_unlock_target, rc = PTR_ERR(mold));
1011 lh_oldp = &info->mti_lh[MDT_LH_OLD];
1012 mdt_lock_reg_init(lh_oldp, LCK_EX);
1013 rc = mdt_object_lock(info, mold, lh_oldp, MDS_INODELOCK_LOOKUP,
1016 mdt_object_put(info->mti_env, mold);
1017 GOTO(out_unlock_target, rc);
1020 info->mti_mos[2] = mold;
1021 rc = mdt_version_get_check(info, 2);
1023 GOTO(out_unlock_old, rc);
1025 mdt_set_capainfo(info, 2, old_fid, BYPASS_CAPA);
1027 /* step 4: find & lock the new object. */
1028 /* new target object may not exist now */
1029 lname = mdt_name(info->mti_env, (char *)rr->rr_tgt, rr->rr_tgtlen);
1030 rc = mdo_lookup(info->mti_env, mdt_object_child(mtgtdir),
1031 lname, new_fid, &info->mti_spec);
1033 /* the new_fid should have been filled at this moment */
1034 if (lu_fid_eq(old_fid, new_fid))
1035 GOTO(out_unlock_old, rc);
1037 if (lu_fid_eq(new_fid, rr->rr_fid1) ||
1038 lu_fid_eq(new_fid, rr->rr_fid2))
1039 GOTO(out_unlock_old, rc = -EINVAL);
1041 mdt_lock_reg_init(lh_newp, LCK_EX);
1042 mnew = mdt_object_find(info->mti_env, info->mti_mdt, new_fid);
1044 GOTO(out_unlock_old, rc = PTR_ERR(mnew));
1046 rc = mdt_object_lock(info, mnew, lh_newp,
1047 MDS_INODELOCK_FULL, MDT_CROSS_LOCK);
1049 mdt_object_put(info->mti_env, mnew);
1050 GOTO(out_unlock_old, rc);
1053 info->mti_mos[3] = mnew;
1054 rc = mdt_version_get_check(info, 3);
1056 GOTO(out_unlock_new, rc);
1058 mdt_set_capainfo(info, 3, new_fid, BYPASS_CAPA);
1059 } else if (rc != -EREMOTE && rc != -ENOENT)
1060 GOTO(out_unlock_old, rc);
1062 /* step 5: rename it */
1063 mdt_reint_init_ma(info, ma);
1064 if (!ma->ma_lmm || !ma->ma_cookie)
1065 GOTO(out_unlock_new, rc = -EINVAL);
1067 mdt_fail_write(info->mti_env, info->mti_mdt->mdt_bottom,
1068 OBD_FAIL_MDS_REINT_RENAME_WRITE);
1071 /* Check if @dst is subdir of @src. */
1072 rc = mdt_rename_sanity(info, old_fid);
1074 GOTO(out_unlock_new, rc);
1076 rc = mdo_rename(info->mti_env, mdt_object_child(msrcdir),
1077 mdt_object_child(mtgtdir), old_fid, &slname,
1078 (mnew ? mdt_object_child(mnew) : NULL),
1081 /* handle last link of tgt object */
1082 if (rc == 0 && mnew)
1083 mdt_handle_last_unlink(info, mnew, ma);
1088 mdt_object_unlock_put(info, mnew, lh_newp, rc);
1090 mdt_object_unlock_put(info, mold, lh_oldp, rc);
1092 mdt_object_unlock_put(info, mtgtdir, lh_tgtdirp, rc);
1094 mdt_object_unlock_put(info, msrcdir, lh_srcdirp, rc);
1096 mdt_rename_unlock(&rename_lh);
1100 typedef int (*mdt_reinter)(struct mdt_thread_info *info,
1101 struct mdt_lock_handle *lhc);
1103 static mdt_reinter reinters[REINT_MAX] = {
1104 [REINT_SETATTR] = mdt_reint_setattr,
1105 [REINT_CREATE] = mdt_reint_create,
1106 [REINT_LINK] = mdt_reint_link,
1107 [REINT_UNLINK] = mdt_reint_unlink,
1108 [REINT_RENAME] = mdt_reint_rename,
1109 [REINT_OPEN] = mdt_reint_open,
1110 [REINT_SETXATTR] = mdt_reint_setxattr
1113 int mdt_reint_rec(struct mdt_thread_info *info,
1114 struct mdt_lock_handle *lhc)
1119 rc = reinters[info->mti_rr.rr_opcode](info, lhc);