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 cfs_spin_lock(&req->rq_export->exp_lock);
124 req->rq_export->exp_vbr_failed = 1;
125 cfs_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 cfs_spin_lock(&req->rq_export->exp_lock);
131 req->rq_export->exp_vbr_failed = 1;
132 cfs_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,
278 struct md_attr *ma, int flags)
280 struct mdt_lock_handle *lh;
281 int do_vbr = ma->ma_attr.la_valid & (LA_MODE|LA_UID|LA_GID|LA_FLAGS);
285 /* attr shouldn't be set on remote object */
286 LASSERT(mdt_object_exists(mo) >= 0);
288 lh = &info->mti_lh[MDT_LH_PARENT];
289 mdt_lock_reg_init(lh, LCK_PW);
291 if (!(flags & MRF_SETATTR_LOCKED)) {
292 __u64 lockpart = MDS_INODELOCK_UPDATE;
293 if (ma->ma_attr.la_valid & (LA_MODE|LA_UID|LA_GID))
294 lockpart |= MDS_INODELOCK_LOOKUP;
296 rc = mdt_object_lock(info, mo, lh, lockpart, MDT_LOCAL_LOCK);
301 if (mdt_object_exists(mo) == 0)
302 GOTO(out_unlock, rc = -ENOENT);
304 /* all attrs are packed into mti_attr in unpack_setattr */
305 mdt_fail_write(info->mti_env, info->mti_mdt->mdt_bottom,
306 OBD_FAIL_MDS_REINT_SETATTR_WRITE);
308 /* This is only for set ctime when rename's source is on remote MDS. */
309 if (unlikely(ma->ma_attr.la_valid == LA_CTIME))
310 ma->ma_attr_flags |= MDS_VTX_BYPASS;
312 /* VBR: update version if attr changed are important for recovery */
314 info->mti_mos[0] = mo;
315 rc = mdt_version_get_check(info, 0);
317 GOTO(out_unlock, rc);
320 /* all attrs are packed into mti_attr in unpack_setattr */
321 rc = mo_attr_set(info->mti_env, mdt_object_child(mo), ma);
323 GOTO(out_unlock, rc);
327 mdt_object_unlock(info, mo, lh, rc);
331 static int mdt_reint_setattr(struct mdt_thread_info *info,
332 struct mdt_lock_handle *lhc)
334 struct md_attr *ma = &info->mti_attr;
335 struct mdt_reint_record *rr = &info->mti_rr;
336 struct ptlrpc_request *req = mdt_info_req(info);
337 struct mdt_export_data *med = &req->rq_export->exp_mdt_data;
338 struct mdt_file_data *mfd;
339 struct mdt_object *mo;
340 struct md_object *next;
341 struct mdt_body *repbody;
345 DEBUG_REQ(D_INODE, req, "setattr "DFID" %x", PFID(rr->rr_fid1),
346 (unsigned int)ma->ma_attr.la_valid);
348 if (info->mti_dlm_req)
349 ldlm_request_cancel(req, info->mti_dlm_req, 0);
351 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
352 mo = mdt_object_find(info->mti_env, info->mti_mdt, rr->rr_fid1);
354 GOTO(out, rc = PTR_ERR(mo));
356 /* start a log jounal handle if needed */
357 if (!(mdt_conn_flags(info) & OBD_CONNECT_SOM)) {
358 if ((ma->ma_attr.la_valid & LA_SIZE) ||
359 (rr->rr_flags & MRF_SETATTR_LOCKED)) {
360 /* Check write access for the O_TRUNC case */
361 if (mdt_write_read(mo) < 0)
362 GOTO(out_put, rc = -ETXTBSY);
364 } else if (info->mti_ioepoch &&
365 (info->mti_ioepoch->flags & MF_EPOCH_OPEN)) {
366 /* Truncate case. IOEpoch is opened. */
367 rc = mdt_write_get(mo);
374 GOTO(out_put, rc = -ENOMEM);
377 mdt_ioepoch_open(info, mo, 0);
378 repbody->ioepoch = mo->mot_ioepoch;
380 mdt_object_get(info->mti_env, mo);
381 mdt_mfd_set_mode(mfd, FMODE_TRUNC);
382 mfd->mfd_object = mo;
383 mfd->mfd_xid = req->rq_xid;
385 cfs_spin_lock(&med->med_open_lock);
386 cfs_list_add(&mfd->mfd_list, &med->med_open_head);
387 cfs_spin_unlock(&med->med_open_lock);
388 repbody->handle.cookie = mfd->mfd_handle.h_cookie;
391 som_au = info->mti_ioepoch && info->mti_ioepoch->flags & MF_SOM_CHANGE;
393 /* SOM Attribute update case. Find the proper mfd and update
394 * SOM attributes on the proper object. */
395 LASSERT(mdt_conn_flags(info) & OBD_CONNECT_SOM);
396 LASSERT(info->mti_ioepoch);
398 cfs_spin_lock(&med->med_open_lock);
399 mfd = mdt_handle2mfd(info, &info->mti_ioepoch->handle);
401 cfs_spin_unlock(&med->med_open_lock);
402 CDEBUG(D_INODE, "no handle for file close: "
403 "fid = "DFID": cookie = "LPX64"\n",
404 PFID(info->mti_rr.rr_fid1),
405 info->mti_ioepoch->handle.cookie);
406 GOTO(out_put, rc = -ESTALE);
408 LASSERT(mfd->mfd_mode == FMODE_SOM);
409 LASSERT(!(info->mti_ioepoch->flags & MF_EPOCH_CLOSE));
411 class_handle_unhash(&mfd->mfd_handle);
412 cfs_list_del_init(&mfd->mfd_list);
413 cfs_spin_unlock(&med->med_open_lock);
415 /* Close the found mfd, update attributes. */
416 ma->ma_lmm_size = info->mti_mdt->mdt_max_mdsize;
417 OBD_ALLOC(ma->ma_lmm, ma->ma_lmm_size);
418 if (ma->ma_lmm == NULL)
419 GOTO(out_put, rc = -ENOMEM);
421 mdt_mfd_close(info, mfd);
423 OBD_FREE(ma->ma_lmm, ma->ma_lmm_size);
425 rc = mdt_attr_set(info, mo, ma, rr->rr_flags);
430 ma->ma_need = MA_INODE;
432 next = mdt_object_child(mo);
433 rc = mo_attr_get(info->mti_env, next, ma);
437 mdt_pack_attr2body(info, repbody, &ma->ma_attr, mdt_object_fid(mo));
439 if (info->mti_mdt->mdt_opts.mo_oss_capa &&
440 info->mti_exp->exp_connect_flags & OBD_CONNECT_OSS_CAPA &&
441 S_ISREG(lu_object_attr(&mo->mot_obj.mo_lu)) &&
442 (ma->ma_attr.la_valid & LA_SIZE) && !som_au) {
443 struct lustre_capa *capa;
445 capa = req_capsule_server_get(info->mti_pill, &RMF_CAPA2);
447 capa->lc_opc = CAPA_OPC_OSS_DEFAULT | CAPA_OPC_OSS_TRUNC;
448 rc = mo_capa_get(info->mti_env, mdt_object_child(mo), capa, 0);
451 repbody->valid |= OBD_MD_FLOSSCAPA;
456 mdt_object_put(info->mti_env, mo);
458 mdt_shrink_reply(info);
462 static int mdt_reint_create(struct mdt_thread_info *info,
463 struct mdt_lock_handle *lhc)
468 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_REINT_CREATE))
469 RETURN(err_serious(-ESTALE));
471 if (info->mti_dlm_req)
472 ldlm_request_cancel(mdt_info_req(info), info->mti_dlm_req, 0);
474 switch (info->mti_attr.ma_attr.la_mode & S_IFMT) {
476 /* Cross-ref case. */
477 if (info->mti_cross_ref) {
478 rc = mdt_md_mkobj(info);
488 /* Special file should stay on the same node as parent. */
489 LASSERT(info->mti_rr.rr_namelen > 0);
490 rc = mdt_md_create(info);
494 rc = err_serious(-EOPNOTSUPP);
499 static int mdt_reint_unlink(struct mdt_thread_info *info,
500 struct mdt_lock_handle *lhc)
502 struct mdt_reint_record *rr = &info->mti_rr;
503 struct ptlrpc_request *req = mdt_info_req(info);
504 struct md_attr *ma = &info->mti_attr;
505 struct lu_fid *child_fid = &info->mti_tmp_fid1;
506 struct mdt_object *mp;
507 struct mdt_object *mc;
508 struct mdt_lock_handle *parent_lh;
509 struct mdt_lock_handle *child_lh;
510 struct lu_name *lname;
514 DEBUG_REQ(D_INODE, req, "unlink "DFID"/%s", PFID(rr->rr_fid1),
517 if (info->mti_dlm_req)
518 ldlm_request_cancel(req, info->mti_dlm_req, 0);
520 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_REINT_UNLINK))
521 RETURN(err_serious(-ENOENT));
524 * step 1: lock the parent. Note, this may be child in case of
525 * remote operation denoted by ->mti_cross_ref flag.
527 parent_lh = &info->mti_lh[MDT_LH_PARENT];
528 if (info->mti_cross_ref) {
530 * Init reg lock for cross ref case when we need to do only
533 mdt_lock_reg_init(parent_lh, LCK_PW);
535 mdt_lock_pdo_init(parent_lh, LCK_PW, rr->rr_name,
538 mp = mdt_object_find_lock(info, rr->rr_fid1, parent_lh,
539 MDS_INODELOCK_UPDATE);
542 /* errors are possible here in cross-ref cases, see below */
543 if (info->mti_cross_ref)
548 info->mti_mos[0] = mp;
549 rc = mdt_version_get_check(info, 0);
551 GOTO(out_unlock_parent, rc);
553 mdt_reint_init_ma(info, ma);
554 if (!ma->ma_lmm || !ma->ma_cookie)
555 GOTO(out_unlock_parent, rc = -EINVAL);
557 if (info->mti_cross_ref) {
559 * Remote partial operation. It is possible that replay may
560 * happen on parent MDT and this operation will be repeated.
561 * Therefore the object absense is allowed case and nothing
562 * should be done here.
564 if (mdt_object_exists(mp) > 0) {
565 mdt_set_capainfo(info, 0, rr->rr_fid1, BYPASS_CAPA);
566 rc = mo_ref_del(info->mti_env,
567 mdt_object_child(mp), ma);
569 mdt_handle_last_unlink(info, mp, ma);
572 GOTO(out_unlock_parent, rc);
575 /* step 2: find & lock the child */
576 lname = mdt_name(info->mti_env, (char *)rr->rr_name, rr->rr_namelen);
577 rc = mdo_lookup(info->mti_env, mdt_object_child(mp),
578 lname, child_fid, &info->mti_spec);
580 GOTO(out_unlock_parent, rc);
582 /* We will lock the child regardless it is local or remote. No harm. */
583 mc = mdt_object_find(info->mti_env, info->mti_mdt, child_fid);
585 GOTO(out_unlock_parent, rc = PTR_ERR(mc));
586 child_lh = &info->mti_lh[MDT_LH_CHILD];
587 mdt_lock_reg_init(child_lh, LCK_EX);
588 rc = mdt_object_lock(info, mc, child_lh, MDS_INODELOCK_FULL,
591 mdt_object_put(info->mti_env, mc);
592 GOTO(out_unlock_parent, rc);
595 mdt_fail_write(info->mti_env, info->mti_mdt->mdt_bottom,
596 OBD_FAIL_MDS_REINT_UNLINK_WRITE);
598 info->mti_mos[1] = mc;
599 rc = mdt_version_get_check(info, 1);
601 GOTO(out_unlock_child, rc);
604 * Now we can only make sure we need MA_INODE, in mdd layer, will check
605 * whether need MA_LOV and MA_COOKIE.
607 ma->ma_need = MA_INODE;
609 mdt_set_capainfo(info, 1, child_fid, BYPASS_CAPA);
610 rc = mdo_unlink(info->mti_env, mdt_object_child(mp),
611 mdt_object_child(mc), lname, ma);
613 mdt_handle_last_unlink(info, mc, ma);
617 mdt_object_unlock_put(info, mc, child_lh, rc);
619 mdt_object_unlock_put(info, mp, parent_lh, rc);
624 static int mdt_reint_link(struct mdt_thread_info *info,
625 struct mdt_lock_handle *lhc)
627 struct mdt_reint_record *rr = &info->mti_rr;
628 struct ptlrpc_request *req = mdt_info_req(info);
629 struct md_attr *ma = &info->mti_attr;
630 struct mdt_object *ms;
631 struct mdt_object *mp;
632 struct mdt_lock_handle *lhs;
633 struct mdt_lock_handle *lhp;
634 struct lu_name *lname;
638 DEBUG_REQ(D_INODE, req, "link "DFID" to "DFID"/%s",
639 PFID(rr->rr_fid1), PFID(rr->rr_fid2), rr->rr_name);
641 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_REINT_LINK))
642 RETURN(err_serious(-ENOENT));
644 if (info->mti_dlm_req)
645 ldlm_request_cancel(req, info->mti_dlm_req, 0);
647 if (info->mti_cross_ref) {
648 /* MDT holding name ask us to add ref. */
649 lhs = &info->mti_lh[MDT_LH_CHILD];
650 mdt_lock_reg_init(lhs, LCK_EX);
651 ms = mdt_object_find_lock(info, rr->rr_fid1, lhs,
652 MDS_INODELOCK_UPDATE);
656 mdt_set_capainfo(info, 0, rr->rr_fid1, BYPASS_CAPA);
657 rc = mo_ref_add(info->mti_env, mdt_object_child(ms), ma);
658 mdt_object_unlock_put(info, ms, lhs, rc);
662 /* Invalid case so return error immediately instead of
664 if (lu_fid_eq(rr->rr_fid1, rr->rr_fid2))
667 /* step 1: find & lock the target parent dir */
668 lhp = &info->mti_lh[MDT_LH_PARENT];
669 mdt_lock_pdo_init(lhp, LCK_EX, rr->rr_name,
671 mp = mdt_object_find_lock(info, rr->rr_fid2, lhp,
672 MDS_INODELOCK_UPDATE);
676 info->mti_mos[0] = mp;
677 rc = mdt_version_get_check(info, 0);
679 GOTO(out_unlock_parent, rc);
681 /* step 2: find & lock the source */
682 lhs = &info->mti_lh[MDT_LH_CHILD];
683 mdt_lock_reg_init(lhs, LCK_EX);
685 ms = mdt_object_find(info->mti_env, info->mti_mdt, rr->rr_fid1);
687 GOTO(out_unlock_parent, rc = PTR_ERR(ms));
689 rc = mdt_object_lock(info, ms, lhs, MDS_INODELOCK_UPDATE,
692 mdt_object_put(info->mti_env, ms);
693 GOTO(out_unlock_parent, rc);
696 /* step 3: link it */
697 mdt_fail_write(info->mti_env, info->mti_mdt->mdt_bottom,
698 OBD_FAIL_MDS_REINT_LINK_WRITE);
700 info->mti_mos[1] = ms;
701 rc = mdt_version_get_check(info, 1);
703 GOTO(out_unlock_child, rc);
705 lname = mdt_name(info->mti_env, (char *)rr->rr_name, rr->rr_namelen);
706 rc = mdo_link(info->mti_env, mdt_object_child(mp),
707 mdt_object_child(ms), lname, ma);
711 mdt_object_unlock_put(info, ms, lhs, rc);
713 mdt_object_unlock_put(info, mp, lhp, rc);
717 /* partial operation for rename */
718 static int mdt_reint_rename_tgt(struct mdt_thread_info *info)
720 struct mdt_reint_record *rr = &info->mti_rr;
721 struct ptlrpc_request *req = mdt_info_req(info);
722 struct md_attr *ma = &info->mti_attr;
723 struct mdt_object *mtgtdir;
724 struct mdt_object *mtgt = NULL;
725 struct mdt_lock_handle *lh_tgtdir;
726 struct mdt_lock_handle *lh_tgt = NULL;
727 struct lu_fid *tgt_fid = &info->mti_tmp_fid1;
728 struct lu_name *lname;
732 DEBUG_REQ(D_INODE, req, "rename_tgt: insert (%s->"DFID") in "DFID,
733 rr->rr_tgt, PFID(rr->rr_fid2), PFID(rr->rr_fid1));
735 /* step 1: lookup & lock the tgt dir. */
736 lh_tgtdir = &info->mti_lh[MDT_LH_PARENT];
737 mdt_lock_pdo_init(lh_tgtdir, LCK_PW, rr->rr_tgt,
739 mtgtdir = mdt_object_find_lock(info, rr->rr_fid1, lh_tgtdir,
740 MDS_INODELOCK_UPDATE);
742 RETURN(PTR_ERR(mtgtdir));
744 /* step 2: find & lock the target object if exists. */
745 mdt_set_capainfo(info, 0, rr->rr_fid1, BYPASS_CAPA);
746 lname = mdt_name(info->mti_env, (char *)rr->rr_tgt, rr->rr_tgtlen);
747 rc = mdo_lookup(info->mti_env, mdt_object_child(mtgtdir),
748 lname, tgt_fid, &info->mti_spec);
749 if (rc != 0 && rc != -ENOENT) {
750 GOTO(out_unlock_tgtdir, rc);
751 } else if (rc == 0) {
753 * In case of replay that name can be already inserted, check
754 * that and do nothing if so.
756 if (lu_fid_eq(tgt_fid, rr->rr_fid2))
757 GOTO(out_unlock_tgtdir, rc);
759 lh_tgt = &info->mti_lh[MDT_LH_CHILD];
760 mdt_lock_reg_init(lh_tgt, LCK_EX);
762 mtgt = mdt_object_find_lock(info, tgt_fid, lh_tgt,
763 MDS_INODELOCK_LOOKUP);
765 GOTO(out_unlock_tgtdir, rc = PTR_ERR(mtgt));
767 mdt_reint_init_ma(info, ma);
768 if (!ma->ma_lmm || !ma->ma_cookie)
769 GOTO(out_unlock_tgt, rc = -EINVAL);
771 rc = mdo_rename_tgt(info->mti_env, mdt_object_child(mtgtdir),
772 mdt_object_child(mtgt), rr->rr_fid2,
774 } else /* -ENOENT */ {
775 rc = mdo_name_insert(info->mti_env, mdt_object_child(mtgtdir),
776 lname, rr->rr_fid2, ma);
779 /* handle last link of tgt object */
781 mdt_handle_last_unlink(info, mtgt, ma);
786 mdt_object_unlock_put(info, mtgt, lh_tgt, rc);
788 mdt_object_unlock_put(info, mtgtdir, lh_tgtdir, rc);
792 static int mdt_rename_lock(struct mdt_thread_info *info,
793 struct lustre_handle *lh)
795 struct ldlm_namespace *ns = info->mti_mdt->mdt_namespace;
796 ldlm_policy_data_t *policy = &info->mti_policy;
797 struct ldlm_res_id *res_id = &info->mti_res_id;
802 ms = mdt_md_site(info->mti_mdt);
803 fid_build_reg_res_name(&LUSTRE_BFL_FID, res_id);
805 memset(policy, 0, sizeof *policy);
806 policy->l_inodebits.bits = MDS_INODELOCK_UPDATE;
808 if (ms->ms_control_exp == NULL) {
809 int flags = LDLM_FL_LOCAL_ONLY | LDLM_FL_ATOMIC_CB;
812 * Current node is controller, that is mdt0, where we should
815 rc = ldlm_cli_enqueue_local(ns, res_id, LDLM_IBITS, policy,
816 LCK_EX, &flags, ldlm_blocking_ast,
817 ldlm_completion_ast, NULL, NULL, 0,
818 &info->mti_exp->exp_handle.h_cookie,
821 struct ldlm_enqueue_info einfo = { LDLM_IBITS, LCK_EX,
822 ldlm_blocking_ast, ldlm_completion_ast, NULL, NULL, NULL };
826 * This is the case mdt0 is remote node, issue DLM lock like
829 rc = ldlm_cli_enqueue(ms->ms_control_exp, NULL, &einfo, res_id,
830 policy, &flags, NULL, 0, lh, 0);
836 static void mdt_rename_unlock(struct lustre_handle *lh)
839 LASSERT(lustre_handle_is_used(lh));
840 ldlm_lock_decref(lh, LCK_EX);
845 * This is is_subdir() variant, it is CMD if cmm forwards it to correct
846 * target. Source should not be ancestor of target dir. May be other rename
847 * checks can be moved here later.
849 static int mdt_rename_sanity(struct mdt_thread_info *info, struct lu_fid *fid)
851 struct mdt_reint_record *rr = &info->mti_rr;
852 struct lu_fid dst_fid = *rr->rr_fid2;
853 struct mdt_object *dst;
858 LASSERT(fid_is_sane(&dst_fid));
859 dst = mdt_object_find(info->mti_env, info->mti_mdt, &dst_fid);
861 rc = mdo_is_subdir(info->mti_env,
862 mdt_object_child(dst), fid,
864 mdt_object_put(info->mti_env, dst);
865 if (rc != -EREMOTE && rc < 0) {
866 CERROR("Failed mdo_is_subdir(), rc %d\n", rc);
868 /* check the found fid */
869 if (lu_fid_eq(&dst_fid, fid))
875 } while (rc == -EREMOTE);
880 static int mdt_reint_rename(struct mdt_thread_info *info,
881 struct mdt_lock_handle *lhc)
883 struct mdt_reint_record *rr = &info->mti_rr;
884 struct md_attr *ma = &info->mti_attr;
885 struct ptlrpc_request *req = mdt_info_req(info);
886 struct mdt_object *msrcdir;
887 struct mdt_object *mtgtdir;
888 struct mdt_object *mold;
889 struct mdt_object *mnew = NULL;
890 struct mdt_lock_handle *lh_srcdirp;
891 struct mdt_lock_handle *lh_tgtdirp;
892 struct mdt_lock_handle *lh_oldp;
893 struct mdt_lock_handle *lh_newp;
894 struct lu_fid *old_fid = &info->mti_tmp_fid1;
895 struct lu_fid *new_fid = &info->mti_tmp_fid2;
896 struct lustre_handle rename_lh = { 0 };
897 struct lu_name slname = { 0 };
898 struct lu_name *lname;
902 if (info->mti_dlm_req)
903 ldlm_request_cancel(mdt_info_req(info), info->mti_dlm_req, 0);
905 if (info->mti_cross_ref) {
906 rc = mdt_reint_rename_tgt(info);
910 DEBUG_REQ(D_INODE, req, "rename "DFID"/%s to "DFID"/%s",
911 PFID(rr->rr_fid1), rr->rr_name,
912 PFID(rr->rr_fid2), rr->rr_tgt);
914 rc = mdt_rename_lock(info, &rename_lh);
916 CERROR("Can't lock FS for rename, rc %d\n", rc);
920 lh_newp = &info->mti_lh[MDT_LH_NEW];
922 /* step 1: lock the source dir. */
923 lh_srcdirp = &info->mti_lh[MDT_LH_PARENT];
924 mdt_lock_pdo_init(lh_srcdirp, LCK_PW, rr->rr_name,
926 msrcdir = mdt_object_find_lock(info, rr->rr_fid1, lh_srcdirp,
927 MDS_INODELOCK_UPDATE);
929 GOTO(out_rename_lock, rc = PTR_ERR(msrcdir));
931 info->mti_mos[0] = msrcdir;
932 rc = mdt_version_get_check(info, 0);
934 GOTO(out_unlock_source, rc);
936 /* step 2: find & lock the target dir. */
937 lh_tgtdirp = &info->mti_lh[MDT_LH_CHILD];
938 mdt_lock_pdo_init(lh_tgtdirp, LCK_PW, rr->rr_tgt,
940 if (lu_fid_eq(rr->rr_fid1, rr->rr_fid2)) {
941 mdt_object_get(info->mti_env, msrcdir);
944 mtgtdir = mdt_object_find(info->mti_env, info->mti_mdt,
947 GOTO(out_unlock_source, rc = PTR_ERR(mtgtdir));
949 rc = mdt_object_exists(mtgtdir);
951 GOTO(out_unlock_target, rc = -ESTALE);
953 /* we lock the target dir if it is local */
954 rc = mdt_object_lock(info, mtgtdir, lh_tgtdirp,
955 MDS_INODELOCK_UPDATE,
958 mdt_object_put(info->mti_env, mtgtdir);
959 GOTO(out_unlock_source, rc);
962 info->mti_mos[1] = mtgtdir;
963 rc = mdt_version_get_check(info, 1);
965 GOTO(out_unlock_target, rc);
969 /* step 3: find & lock the old object. */
970 lname = mdt_name(info->mti_env, (char *)rr->rr_name, rr->rr_namelen);
971 mdt_name_copy(&slname, lname);
972 rc = mdo_lookup(info->mti_env, mdt_object_child(msrcdir),
973 &slname, old_fid, &info->mti_spec);
975 GOTO(out_unlock_target, rc);
977 if (lu_fid_eq(old_fid, rr->rr_fid1) || lu_fid_eq(old_fid, rr->rr_fid2))
978 GOTO(out_unlock_target, rc = -EINVAL);
980 mold = mdt_object_find(info->mti_env, info->mti_mdt, old_fid);
982 GOTO(out_unlock_target, rc = PTR_ERR(mold));
984 lh_oldp = &info->mti_lh[MDT_LH_OLD];
985 mdt_lock_reg_init(lh_oldp, LCK_EX);
986 rc = mdt_object_lock(info, mold, lh_oldp, MDS_INODELOCK_LOOKUP,
989 mdt_object_put(info->mti_env, mold);
990 GOTO(out_unlock_target, rc);
993 info->mti_mos[2] = mold;
994 rc = mdt_version_get_check(info, 2);
996 GOTO(out_unlock_old, rc);
998 mdt_set_capainfo(info, 2, old_fid, BYPASS_CAPA);
1000 /* step 4: find & lock the new object. */
1001 /* new target object may not exist now */
1002 lname = mdt_name(info->mti_env, (char *)rr->rr_tgt, rr->rr_tgtlen);
1003 rc = mdo_lookup(info->mti_env, mdt_object_child(mtgtdir),
1004 lname, new_fid, &info->mti_spec);
1006 /* the new_fid should have been filled at this moment */
1007 if (lu_fid_eq(old_fid, new_fid))
1008 GOTO(out_unlock_old, rc);
1010 if (lu_fid_eq(new_fid, rr->rr_fid1) ||
1011 lu_fid_eq(new_fid, rr->rr_fid2))
1012 GOTO(out_unlock_old, rc = -EINVAL);
1014 mdt_lock_reg_init(lh_newp, LCK_EX);
1015 mnew = mdt_object_find(info->mti_env, info->mti_mdt, new_fid);
1017 GOTO(out_unlock_old, rc = PTR_ERR(mnew));
1019 rc = mdt_object_lock(info, mnew, lh_newp,
1020 MDS_INODELOCK_FULL, MDT_CROSS_LOCK);
1022 mdt_object_put(info->mti_env, mnew);
1023 GOTO(out_unlock_old, rc);
1026 info->mti_mos[3] = mnew;
1027 rc = mdt_version_get_check(info, 3);
1029 GOTO(out_unlock_new, rc);
1031 mdt_set_capainfo(info, 3, new_fid, BYPASS_CAPA);
1032 } else if (rc != -EREMOTE && rc != -ENOENT)
1033 GOTO(out_unlock_old, rc);
1035 /* step 5: rename it */
1036 mdt_reint_init_ma(info, ma);
1037 if (!ma->ma_lmm || !ma->ma_cookie)
1038 GOTO(out_unlock_new, rc = -EINVAL);
1040 mdt_fail_write(info->mti_env, info->mti_mdt->mdt_bottom,
1041 OBD_FAIL_MDS_REINT_RENAME_WRITE);
1044 /* Check if @dst is subdir of @src. */
1045 rc = mdt_rename_sanity(info, old_fid);
1047 GOTO(out_unlock_new, rc);
1049 rc = mdo_rename(info->mti_env, mdt_object_child(msrcdir),
1050 mdt_object_child(mtgtdir), old_fid, &slname,
1051 (mnew ? mdt_object_child(mnew) : NULL),
1054 /* handle last link of tgt object */
1055 if (rc == 0 && mnew)
1056 mdt_handle_last_unlink(info, mnew, ma);
1061 mdt_object_unlock_put(info, mnew, lh_newp, rc);
1063 mdt_object_unlock_put(info, mold, lh_oldp, rc);
1065 mdt_object_unlock_put(info, mtgtdir, lh_tgtdirp, rc);
1067 mdt_object_unlock_put(info, msrcdir, lh_srcdirp, rc);
1069 mdt_rename_unlock(&rename_lh);
1073 typedef int (*mdt_reinter)(struct mdt_thread_info *info,
1074 struct mdt_lock_handle *lhc);
1076 static mdt_reinter reinters[REINT_MAX] = {
1077 [REINT_SETATTR] = mdt_reint_setattr,
1078 [REINT_CREATE] = mdt_reint_create,
1079 [REINT_LINK] = mdt_reint_link,
1080 [REINT_UNLINK] = mdt_reint_unlink,
1081 [REINT_RENAME] = mdt_reint_rename,
1082 [REINT_OPEN] = mdt_reint_open,
1083 [REINT_SETXATTR] = mdt_reint_setxattr
1086 int mdt_reint_rec(struct mdt_thread_info *info,
1087 struct mdt_lock_handle *lhc)
1092 rc = reinters[info->mti_rr.rr_opcode](info, lhc);