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 if (mdt_object_exists(info->mti_mos[index]) == 0)
113 mo = mdt_object_child(info->mti_mos[index]);
115 curr_version = mo_version_get(info->mti_env, mo);
116 CDEBUG(D_INODE, "Version is "LPX64"\n", curr_version);
117 /** VBR: version is checked always because costs nothing */
118 if (lustre_msg_get_transno(req->rq_reqmsg) != 0) {
119 pre_versions = lustre_msg_get_versions(req->rq_reqmsg);
120 LASSERT(index < PTLRPC_NUM_VERSIONS);
121 /** Sanity check for malformed buffers */
122 if (pre_versions == NULL) {
123 CERROR("No versions in request buffer\n");
124 cfs_spin_lock(&req->rq_export->exp_lock);
125 req->rq_export->exp_vbr_failed = 1;
126 cfs_spin_unlock(&req->rq_export->exp_lock);
128 } else if (pre_versions[index] != curr_version) {
129 CDEBUG(D_INODE, "Version mismatch "LPX64" != "LPX64"\n",
130 pre_versions[index], curr_version);
131 cfs_spin_lock(&req->rq_export->exp_lock);
132 req->rq_export->exp_vbr_failed = 1;
133 cfs_spin_unlock(&req->rq_export->exp_lock);
137 /** save pre-versions in reply */
138 LASSERT(req->rq_repmsg != NULL);
139 pre_versions = lustre_msg_get_versions(req->rq_repmsg);
141 pre_versions[index] = curr_version;
145 static int mdt_md_create(struct mdt_thread_info *info)
147 struct mdt_device *mdt = info->mti_mdt;
148 struct mdt_object *parent;
149 struct mdt_object *child;
150 struct mdt_lock_handle *lh;
151 struct mdt_body *repbody;
152 struct md_attr *ma = &info->mti_attr;
153 struct mdt_reint_record *rr = &info->mti_rr;
154 struct lu_name *lname;
158 DEBUG_REQ(D_INODE, mdt_info_req(info), "Create (%s->"DFID") in "DFID,
159 rr->rr_name, PFID(rr->rr_fid2), PFID(rr->rr_fid1));
161 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
163 lh = &info->mti_lh[MDT_LH_PARENT];
164 mdt_lock_pdo_init(lh, LCK_PW, rr->rr_name, rr->rr_namelen);
166 parent = mdt_object_find_lock(info, rr->rr_fid1, lh,
167 MDS_INODELOCK_UPDATE);
169 RETURN(PTR_ERR(parent));
171 child = mdt_object_find(info->mti_env, mdt, rr->rr_fid2);
172 if (likely(!IS_ERR(child))) {
173 struct md_object *next = mdt_object_child(parent);
175 ma->ma_need = MA_INODE;
177 /* capa for cross-ref will be stored here */
178 ma->ma_capa = req_capsule_server_get(info->mti_pill,
180 LASSERT(ma->ma_capa);
182 mdt_fail_write(info->mti_env, info->mti_mdt->mdt_bottom,
183 OBD_FAIL_MDS_REINT_CREATE_WRITE);
185 info->mti_mos[0] = parent;
186 info->mti_mos[1] = child;
187 rc = mdt_version_get_check(info, 0);
189 GOTO(out_put_child, rc);
191 /* Let lower layer know current lock mode. */
192 info->mti_spec.sp_cr_mode =
193 mdt_dlm_mode2mdl_mode(lh->mlh_pdo_mode);
196 * Do perform lookup sanity check. We do not know if name exists
199 info->mti_spec.sp_cr_lookup = 1;
200 info->mti_spec.sp_feat = &dt_directory_features;
202 lname = mdt_name(info->mti_env, (char *)rr->rr_name,
204 rc = mdo_create(info->mti_env, next, lname,
205 mdt_object_child(child),
206 &info->mti_spec, ma);
208 /* Return fid & attr to client. */
209 if (ma->ma_valid & MA_INODE)
210 mdt_pack_attr2body(info, repbody, &ma->ma_attr,
211 mdt_object_fid(child));
214 mdt_object_put(info->mti_env, child);
218 mdt_create_pack_capa(info, rc, child, repbody);
219 mdt_object_unlock_put(info, parent, lh, rc);
223 /* Partial request to create object only */
224 static int mdt_md_mkobj(struct mdt_thread_info *info)
226 struct mdt_device *mdt = info->mti_mdt;
227 struct mdt_object *o;
228 struct mdt_body *repbody;
229 struct md_attr *ma = &info->mti_attr;
233 DEBUG_REQ(D_INODE, mdt_info_req(info), "Partial create "DFID"",
234 PFID(info->mti_rr.rr_fid2));
236 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
238 o = mdt_object_find(info->mti_env, mdt, info->mti_rr.rr_fid2);
240 struct md_object *next = mdt_object_child(o);
242 ma->ma_need = MA_INODE;
246 * Cross-ref create can encounter already created obj in case of
247 * recovery, just get attr in that case.
249 if (mdt_object_exists(o) == 1) {
250 rc = mo_attr_get(info->mti_env, next, ma);
253 * Here, NO permission check for object_create,
254 * such check has been done on the original MDS.
256 rc = mo_object_create(info->mti_env, next,
257 &info->mti_spec, ma);
260 /* Return fid & attr to client. */
261 if (ma->ma_valid & MA_INODE)
262 mdt_pack_attr2body(info, repbody, &ma->ma_attr,
265 mdt_object_put(info->mti_env, o);
269 mdt_create_pack_capa(info, rc, o, repbody);
273 /* In the raw-setattr case, we lock the child inode.
274 * In the write-back case or if being called from open,
275 * the client holds a lock already.
276 * We use the ATTR_FROM_OPEN (translated into MRF_SETATTR_LOCKED by
277 * mdt_setattr_unpack()) flag to tell these cases apart. */
278 int mdt_attr_set(struct mdt_thread_info *info, struct mdt_object *mo,
279 struct md_attr *ma, int flags)
281 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 lh = &info->mti_lh[MDT_LH_PARENT];
290 mdt_lock_reg_init(lh, LCK_PW);
292 if (!(flags & MRF_SETATTR_LOCKED)) {
293 __u64 lockpart = MDS_INODELOCK_UPDATE;
294 if (ma->ma_attr.la_valid & (LA_MODE|LA_UID|LA_GID))
295 lockpart |= MDS_INODELOCK_LOOKUP;
297 rc = mdt_object_lock(info, mo, lh, lockpart, MDT_LOCAL_LOCK);
302 if (mdt_object_exists(mo) == 0)
303 GOTO(out_unlock, rc = -ENOENT);
305 /* all attrs are packed into mti_attr in unpack_setattr */
306 mdt_fail_write(info->mti_env, info->mti_mdt->mdt_bottom,
307 OBD_FAIL_MDS_REINT_SETATTR_WRITE);
309 /* This is only for set ctime when rename's source is on remote MDS. */
310 if (unlikely(ma->ma_attr.la_valid == LA_CTIME))
311 ma->ma_attr_flags |= MDS_VTX_BYPASS;
313 /* VBR: update version if attr changed are important for recovery */
315 info->mti_mos[0] = mo;
316 rc = mdt_version_get_check(info, 0);
318 GOTO(out_unlock, rc);
321 /* all attrs are packed into mti_attr in unpack_setattr */
322 rc = mo_attr_set(info->mti_env, mdt_object_child(mo), ma);
324 GOTO(out_unlock, rc);
328 mdt_object_unlock(info, mo, lh, rc);
332 static int mdt_reint_setattr(struct mdt_thread_info *info,
333 struct mdt_lock_handle *lhc)
335 struct md_attr *ma = &info->mti_attr;
336 struct mdt_reint_record *rr = &info->mti_rr;
337 struct ptlrpc_request *req = mdt_info_req(info);
338 struct mdt_export_data *med = &req->rq_export->exp_mdt_data;
339 struct mdt_file_data *mfd;
340 struct mdt_object *mo;
341 struct md_object *next;
342 struct mdt_body *repbody;
346 DEBUG_REQ(D_INODE, req, "setattr "DFID" %x", PFID(rr->rr_fid1),
347 (unsigned int)ma->ma_attr.la_valid);
349 if (info->mti_dlm_req)
350 ldlm_request_cancel(req, info->mti_dlm_req, 0);
352 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
353 mo = mdt_object_find(info->mti_env, info->mti_mdt, rr->rr_fid1);
355 GOTO(out, rc = PTR_ERR(mo));
357 /* start a log jounal handle if needed */
358 if (!(mdt_conn_flags(info) & OBD_CONNECT_SOM)) {
359 if ((ma->ma_attr.la_valid & LA_SIZE) ||
360 (rr->rr_flags & MRF_SETATTR_LOCKED)) {
361 /* Check write access for the O_TRUNC case */
362 if (mdt_write_read(mo) < 0)
363 GOTO(out_put, rc = -ETXTBSY);
365 } else if (info->mti_ioepoch &&
366 (info->mti_ioepoch->flags & MF_EPOCH_OPEN)) {
367 /* Truncate case. IOEpoch is opened. */
368 rc = mdt_write_get(mo);
375 GOTO(out_put, rc = -ENOMEM);
378 mdt_ioepoch_open(info, mo, 0);
379 repbody->ioepoch = mo->mot_ioepoch;
381 mdt_object_get(info->mti_env, mo);
382 mdt_mfd_set_mode(mfd, FMODE_TRUNC);
383 mfd->mfd_object = mo;
384 mfd->mfd_xid = req->rq_xid;
386 cfs_spin_lock(&med->med_open_lock);
387 cfs_list_add(&mfd->mfd_list, &med->med_open_head);
388 cfs_spin_unlock(&med->med_open_lock);
389 repbody->handle.cookie = mfd->mfd_handle.h_cookie;
392 som_au = info->mti_ioepoch && info->mti_ioepoch->flags & MF_SOM_CHANGE;
394 /* SOM Attribute update case. Find the proper mfd and update
395 * SOM attributes on the proper object. */
396 LASSERT(mdt_conn_flags(info) & OBD_CONNECT_SOM);
397 LASSERT(info->mti_ioepoch);
399 cfs_spin_lock(&med->med_open_lock);
400 mfd = mdt_handle2mfd(info, &info->mti_ioepoch->handle);
402 cfs_spin_unlock(&med->med_open_lock);
403 CDEBUG(D_INODE, "no handle for file close: "
404 "fid = "DFID": cookie = "LPX64"\n",
405 PFID(info->mti_rr.rr_fid1),
406 info->mti_ioepoch->handle.cookie);
407 GOTO(out_put, rc = -ESTALE);
409 LASSERT(mfd->mfd_mode == FMODE_SOM);
410 LASSERT(!(info->mti_ioepoch->flags & MF_EPOCH_CLOSE));
412 class_handle_unhash(&mfd->mfd_handle);
413 cfs_list_del_init(&mfd->mfd_list);
414 cfs_spin_unlock(&med->med_open_lock);
416 /* Close the found mfd, update attributes. */
417 ma->ma_lmm_size = info->mti_mdt->mdt_max_mdsize;
418 OBD_ALLOC(ma->ma_lmm, ma->ma_lmm_size);
419 if (ma->ma_lmm == NULL)
420 GOTO(out_put, rc = -ENOMEM);
422 mdt_mfd_close(info, mfd);
424 OBD_FREE(ma->ma_lmm, ma->ma_lmm_size);
426 rc = mdt_attr_set(info, mo, ma, rr->rr_flags);
431 ma->ma_need = MA_INODE;
433 next = mdt_object_child(mo);
434 rc = mo_attr_get(info->mti_env, next, ma);
438 mdt_pack_attr2body(info, repbody, &ma->ma_attr, mdt_object_fid(mo));
440 if (info->mti_mdt->mdt_opts.mo_oss_capa &&
441 info->mti_exp->exp_connect_flags & OBD_CONNECT_OSS_CAPA &&
442 S_ISREG(lu_object_attr(&mo->mot_obj.mo_lu)) &&
443 (ma->ma_attr.la_valid & LA_SIZE) && !som_au) {
444 struct lustre_capa *capa;
446 capa = req_capsule_server_get(info->mti_pill, &RMF_CAPA2);
448 capa->lc_opc = CAPA_OPC_OSS_DEFAULT | CAPA_OPC_OSS_TRUNC;
449 rc = mo_capa_get(info->mti_env, mdt_object_child(mo), capa, 0);
452 repbody->valid |= OBD_MD_FLOSSCAPA;
457 mdt_object_put(info->mti_env, mo);
459 mdt_shrink_reply(info);
463 static int mdt_reint_create(struct mdt_thread_info *info,
464 struct mdt_lock_handle *lhc)
469 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_REINT_CREATE))
470 RETURN(err_serious(-ESTALE));
472 if (info->mti_dlm_req)
473 ldlm_request_cancel(mdt_info_req(info), info->mti_dlm_req, 0);
475 switch (info->mti_attr.ma_attr.la_mode & S_IFMT) {
477 /* Cross-ref case. */
478 if (info->mti_cross_ref) {
479 rc = mdt_md_mkobj(info);
489 /* Special file should stay on the same node as parent. */
490 LASSERT(info->mti_rr.rr_namelen > 0);
491 rc = mdt_md_create(info);
495 rc = err_serious(-EOPNOTSUPP);
500 static int mdt_reint_unlink(struct mdt_thread_info *info,
501 struct mdt_lock_handle *lhc)
503 struct mdt_reint_record *rr = &info->mti_rr;
504 struct ptlrpc_request *req = mdt_info_req(info);
505 struct md_attr *ma = &info->mti_attr;
506 struct lu_fid *child_fid = &info->mti_tmp_fid1;
507 struct mdt_object *mp;
508 struct mdt_object *mc;
509 struct mdt_lock_handle *parent_lh;
510 struct mdt_lock_handle *child_lh;
511 struct lu_name *lname;
515 DEBUG_REQ(D_INODE, req, "unlink "DFID"/%s", PFID(rr->rr_fid1),
518 if (info->mti_dlm_req)
519 ldlm_request_cancel(req, info->mti_dlm_req, 0);
521 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_REINT_UNLINK))
522 RETURN(err_serious(-ENOENT));
525 * step 1: lock the parent. Note, this may be child in case of
526 * remote operation denoted by ->mti_cross_ref flag.
528 parent_lh = &info->mti_lh[MDT_LH_PARENT];
529 if (info->mti_cross_ref) {
531 * Init reg lock for cross ref case when we need to do only
534 mdt_lock_reg_init(parent_lh, LCK_PW);
536 mdt_lock_pdo_init(parent_lh, LCK_PW, rr->rr_name,
539 mp = mdt_object_find_lock(info, rr->rr_fid1, parent_lh,
540 MDS_INODELOCK_UPDATE);
543 /* errors are possible here in cross-ref cases, see below */
544 if (info->mti_cross_ref)
549 info->mti_mos[0] = mp;
550 rc = mdt_version_get_check(info, 0);
552 GOTO(out_unlock_parent, rc);
554 mdt_reint_init_ma(info, ma);
555 if (!ma->ma_lmm || !ma->ma_cookie)
556 GOTO(out_unlock_parent, rc = -EINVAL);
558 if (info->mti_cross_ref) {
560 * Remote partial operation. It is possible that replay may
561 * happen on parent MDT and this operation will be repeated.
562 * Therefore the object absense is allowed case and nothing
563 * should be done here.
565 if (mdt_object_exists(mp) > 0) {
566 mdt_set_capainfo(info, 0, rr->rr_fid1, BYPASS_CAPA);
567 rc = mo_ref_del(info->mti_env,
568 mdt_object_child(mp), ma);
570 mdt_handle_last_unlink(info, mp, ma);
573 GOTO(out_unlock_parent, rc);
576 /* step 2: find & lock the child */
577 lname = mdt_name(info->mti_env, (char *)rr->rr_name, rr->rr_namelen);
578 rc = mdo_lookup(info->mti_env, mdt_object_child(mp),
579 lname, child_fid, &info->mti_spec);
581 GOTO(out_unlock_parent, rc);
583 /* We will lock the child regardless it is local or remote. No harm. */
584 mc = mdt_object_find(info->mti_env, info->mti_mdt, child_fid);
586 GOTO(out_unlock_parent, rc = PTR_ERR(mc));
587 child_lh = &info->mti_lh[MDT_LH_CHILD];
588 mdt_lock_reg_init(child_lh, LCK_EX);
589 rc = mdt_object_lock(info, mc, child_lh, MDS_INODELOCK_FULL,
592 mdt_object_put(info->mti_env, mc);
593 GOTO(out_unlock_parent, rc);
596 mdt_fail_write(info->mti_env, info->mti_mdt->mdt_bottom,
597 OBD_FAIL_MDS_REINT_UNLINK_WRITE);
599 info->mti_mos[1] = mc;
600 rc = mdt_version_get_check(info, 1);
602 GOTO(out_unlock_child, rc);
605 * Now we can only make sure we need MA_INODE, in mdd layer, will check
606 * whether need MA_LOV and MA_COOKIE.
608 ma->ma_need = MA_INODE;
610 mdt_set_capainfo(info, 1, child_fid, BYPASS_CAPA);
611 rc = mdo_unlink(info->mti_env, mdt_object_child(mp),
612 mdt_object_child(mc), lname, ma);
614 mdt_handle_last_unlink(info, mc, ma);
618 mdt_object_unlock_put(info, mc, child_lh, rc);
620 mdt_object_unlock_put(info, mp, parent_lh, rc);
625 static int mdt_reint_link(struct mdt_thread_info *info,
626 struct mdt_lock_handle *lhc)
628 struct mdt_reint_record *rr = &info->mti_rr;
629 struct ptlrpc_request *req = mdt_info_req(info);
630 struct md_attr *ma = &info->mti_attr;
631 struct mdt_object *ms;
632 struct mdt_object *mp;
633 struct mdt_lock_handle *lhs;
634 struct mdt_lock_handle *lhp;
635 struct lu_name *lname;
639 DEBUG_REQ(D_INODE, req, "link "DFID" to "DFID"/%s",
640 PFID(rr->rr_fid1), PFID(rr->rr_fid2), rr->rr_name);
642 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_REINT_LINK))
643 RETURN(err_serious(-ENOENT));
645 if (info->mti_dlm_req)
646 ldlm_request_cancel(req, info->mti_dlm_req, 0);
648 if (info->mti_cross_ref) {
649 /* MDT holding name ask us to add ref. */
650 lhs = &info->mti_lh[MDT_LH_CHILD];
651 mdt_lock_reg_init(lhs, LCK_EX);
652 ms = mdt_object_find_lock(info, rr->rr_fid1, lhs,
653 MDS_INODELOCK_UPDATE);
657 mdt_set_capainfo(info, 0, rr->rr_fid1, BYPASS_CAPA);
658 rc = mo_ref_add(info->mti_env, mdt_object_child(ms), ma);
659 mdt_object_unlock_put(info, ms, lhs, rc);
663 /* Invalid case so return error immediately instead of
665 if (lu_fid_eq(rr->rr_fid1, rr->rr_fid2))
668 /* step 1: find & lock the target parent dir */
669 lhp = &info->mti_lh[MDT_LH_PARENT];
670 mdt_lock_pdo_init(lhp, LCK_EX, rr->rr_name,
672 mp = mdt_object_find_lock(info, rr->rr_fid2, lhp,
673 MDS_INODELOCK_UPDATE);
677 info->mti_mos[0] = mp;
678 rc = mdt_version_get_check(info, 0);
680 GOTO(out_unlock_parent, rc);
682 /* step 2: find & lock the source */
683 lhs = &info->mti_lh[MDT_LH_CHILD];
684 mdt_lock_reg_init(lhs, LCK_EX);
686 ms = mdt_object_find(info->mti_env, info->mti_mdt, rr->rr_fid1);
688 GOTO(out_unlock_parent, rc = PTR_ERR(ms));
690 rc = mdt_object_lock(info, ms, lhs, MDS_INODELOCK_UPDATE,
693 mdt_object_put(info->mti_env, ms);
694 GOTO(out_unlock_parent, rc);
697 /* step 3: link it */
698 mdt_fail_write(info->mti_env, info->mti_mdt->mdt_bottom,
699 OBD_FAIL_MDS_REINT_LINK_WRITE);
701 info->mti_mos[1] = ms;
702 rc = mdt_version_get_check(info, 1);
704 GOTO(out_unlock_child, rc);
706 lname = mdt_name(info->mti_env, (char *)rr->rr_name, rr->rr_namelen);
707 rc = mdo_link(info->mti_env, mdt_object_child(mp),
708 mdt_object_child(ms), lname, ma);
712 mdt_object_unlock_put(info, ms, lhs, rc);
714 mdt_object_unlock_put(info, mp, lhp, rc);
718 /* partial operation for rename */
719 static int mdt_reint_rename_tgt(struct mdt_thread_info *info)
721 struct mdt_reint_record *rr = &info->mti_rr;
722 struct ptlrpc_request *req = mdt_info_req(info);
723 struct md_attr *ma = &info->mti_attr;
724 struct mdt_object *mtgtdir;
725 struct mdt_object *mtgt = NULL;
726 struct mdt_lock_handle *lh_tgtdir;
727 struct mdt_lock_handle *lh_tgt = NULL;
728 struct lu_fid *tgt_fid = &info->mti_tmp_fid1;
729 struct lu_name *lname;
733 DEBUG_REQ(D_INODE, req, "rename_tgt: insert (%s->"DFID") in "DFID,
734 rr->rr_tgt, PFID(rr->rr_fid2), PFID(rr->rr_fid1));
736 /* step 1: lookup & lock the tgt dir. */
737 lh_tgtdir = &info->mti_lh[MDT_LH_PARENT];
738 mdt_lock_pdo_init(lh_tgtdir, LCK_PW, rr->rr_tgt,
740 mtgtdir = mdt_object_find_lock(info, rr->rr_fid1, lh_tgtdir,
741 MDS_INODELOCK_UPDATE);
743 RETURN(PTR_ERR(mtgtdir));
745 /* step 2: find & lock the target object if exists. */
746 mdt_set_capainfo(info, 0, rr->rr_fid1, BYPASS_CAPA);
747 lname = mdt_name(info->mti_env, (char *)rr->rr_tgt, rr->rr_tgtlen);
748 rc = mdo_lookup(info->mti_env, mdt_object_child(mtgtdir),
749 lname, tgt_fid, &info->mti_spec);
750 if (rc != 0 && rc != -ENOENT) {
751 GOTO(out_unlock_tgtdir, rc);
752 } else if (rc == 0) {
754 * In case of replay that name can be already inserted, check
755 * that and do nothing if so.
757 if (lu_fid_eq(tgt_fid, rr->rr_fid2))
758 GOTO(out_unlock_tgtdir, rc);
760 lh_tgt = &info->mti_lh[MDT_LH_CHILD];
761 mdt_lock_reg_init(lh_tgt, LCK_EX);
763 mtgt = mdt_object_find_lock(info, tgt_fid, lh_tgt,
764 MDS_INODELOCK_LOOKUP);
766 GOTO(out_unlock_tgtdir, rc = PTR_ERR(mtgt));
768 mdt_reint_init_ma(info, ma);
769 if (!ma->ma_lmm || !ma->ma_cookie)
770 GOTO(out_unlock_tgt, rc = -EINVAL);
772 rc = mdo_rename_tgt(info->mti_env, mdt_object_child(mtgtdir),
773 mdt_object_child(mtgt), rr->rr_fid2,
775 } else /* -ENOENT */ {
776 rc = mdo_name_insert(info->mti_env, mdt_object_child(mtgtdir),
777 lname, rr->rr_fid2, ma);
780 /* handle last link of tgt object */
782 mdt_handle_last_unlink(info, mtgt, ma);
787 mdt_object_unlock_put(info, mtgt, lh_tgt, rc);
789 mdt_object_unlock_put(info, mtgtdir, lh_tgtdir, rc);
793 static int mdt_rename_lock(struct mdt_thread_info *info,
794 struct lustre_handle *lh)
796 struct ldlm_namespace *ns = info->mti_mdt->mdt_namespace;
797 ldlm_policy_data_t *policy = &info->mti_policy;
798 struct ldlm_res_id *res_id = &info->mti_res_id;
803 ms = mdt_md_site(info->mti_mdt);
804 fid_build_reg_res_name(&LUSTRE_BFL_FID, res_id);
806 memset(policy, 0, sizeof *policy);
807 policy->l_inodebits.bits = MDS_INODELOCK_UPDATE;
809 if (ms->ms_control_exp == NULL) {
810 int flags = LDLM_FL_LOCAL_ONLY | LDLM_FL_ATOMIC_CB;
813 * Current node is controller, that is mdt0, where we should
816 rc = ldlm_cli_enqueue_local(ns, res_id, LDLM_IBITS, policy,
817 LCK_EX, &flags, ldlm_blocking_ast,
818 ldlm_completion_ast, NULL, NULL, 0,
819 &info->mti_exp->exp_handle.h_cookie,
822 struct ldlm_enqueue_info einfo = { LDLM_IBITS, LCK_EX,
823 ldlm_blocking_ast, ldlm_completion_ast, NULL, NULL, NULL };
827 * This is the case mdt0 is remote node, issue DLM lock like
830 rc = ldlm_cli_enqueue(ms->ms_control_exp, NULL, &einfo, res_id,
831 policy, &flags, NULL, 0, lh, 0);
837 static void mdt_rename_unlock(struct lustre_handle *lh)
840 LASSERT(lustre_handle_is_used(lh));
841 ldlm_lock_decref(lh, LCK_EX);
846 * This is is_subdir() variant, it is CMD if cmm forwards it to correct
847 * target. Source should not be ancestor of target dir. May be other rename
848 * checks can be moved here later.
850 static int mdt_rename_sanity(struct mdt_thread_info *info, struct lu_fid *fid)
852 struct mdt_reint_record *rr = &info->mti_rr;
853 struct lu_fid dst_fid = *rr->rr_fid2;
854 struct mdt_object *dst;
859 LASSERT(fid_is_sane(&dst_fid));
860 dst = mdt_object_find(info->mti_env, info->mti_mdt, &dst_fid);
862 rc = mdo_is_subdir(info->mti_env,
863 mdt_object_child(dst), fid,
865 mdt_object_put(info->mti_env, dst);
866 if (rc != -EREMOTE && rc < 0) {
867 CERROR("Failed mdo_is_subdir(), rc %d\n", rc);
869 /* check the found fid */
870 if (lu_fid_eq(&dst_fid, fid))
876 } while (rc == -EREMOTE);
881 static int mdt_reint_rename(struct mdt_thread_info *info,
882 struct mdt_lock_handle *lhc)
884 struct mdt_reint_record *rr = &info->mti_rr;
885 struct md_attr *ma = &info->mti_attr;
886 struct ptlrpc_request *req = mdt_info_req(info);
887 struct mdt_object *msrcdir;
888 struct mdt_object *mtgtdir;
889 struct mdt_object *mold;
890 struct mdt_object *mnew = NULL;
891 struct mdt_lock_handle *lh_srcdirp;
892 struct mdt_lock_handle *lh_tgtdirp;
893 struct mdt_lock_handle *lh_oldp;
894 struct mdt_lock_handle *lh_newp;
895 struct lu_fid *old_fid = &info->mti_tmp_fid1;
896 struct lu_fid *new_fid = &info->mti_tmp_fid2;
897 struct lustre_handle rename_lh = { 0 };
898 struct lu_name slname = { 0 };
899 struct lu_name *lname;
903 if (info->mti_dlm_req)
904 ldlm_request_cancel(mdt_info_req(info), info->mti_dlm_req, 0);
906 if (info->mti_cross_ref) {
907 rc = mdt_reint_rename_tgt(info);
911 DEBUG_REQ(D_INODE, req, "rename "DFID"/%s to "DFID"/%s",
912 PFID(rr->rr_fid1), rr->rr_name,
913 PFID(rr->rr_fid2), rr->rr_tgt);
915 rc = mdt_rename_lock(info, &rename_lh);
917 CERROR("Can't lock FS for rename, rc %d\n", rc);
921 lh_newp = &info->mti_lh[MDT_LH_NEW];
923 /* step 1: lock the source dir. */
924 lh_srcdirp = &info->mti_lh[MDT_LH_PARENT];
925 mdt_lock_pdo_init(lh_srcdirp, LCK_PW, rr->rr_name,
927 msrcdir = mdt_object_find_lock(info, rr->rr_fid1, lh_srcdirp,
928 MDS_INODELOCK_UPDATE);
930 GOTO(out_rename_lock, rc = PTR_ERR(msrcdir));
932 info->mti_mos[0] = msrcdir;
933 rc = mdt_version_get_check(info, 0);
935 GOTO(out_unlock_source, rc);
937 /* step 2: find & lock the target dir. */
938 lh_tgtdirp = &info->mti_lh[MDT_LH_CHILD];
939 mdt_lock_pdo_init(lh_tgtdirp, LCK_PW, rr->rr_tgt,
941 if (lu_fid_eq(rr->rr_fid1, rr->rr_fid2)) {
942 mdt_object_get(info->mti_env, msrcdir);
945 mtgtdir = mdt_object_find(info->mti_env, info->mti_mdt,
948 GOTO(out_unlock_source, rc = PTR_ERR(mtgtdir));
950 rc = mdt_object_exists(mtgtdir);
952 GOTO(out_unlock_target, rc = -ESTALE);
954 /* we lock the target dir if it is local */
955 rc = mdt_object_lock(info, mtgtdir, lh_tgtdirp,
956 MDS_INODELOCK_UPDATE,
959 mdt_object_put(info->mti_env, mtgtdir);
960 GOTO(out_unlock_source, rc);
963 info->mti_mos[1] = mtgtdir;
964 rc = mdt_version_get_check(info, 1);
966 GOTO(out_unlock_target, rc);
970 /* step 3: find & lock the old object. */
971 lname = mdt_name(info->mti_env, (char *)rr->rr_name, rr->rr_namelen);
972 mdt_name_copy(&slname, lname);
973 rc = mdo_lookup(info->mti_env, mdt_object_child(msrcdir),
974 &slname, old_fid, &info->mti_spec);
976 GOTO(out_unlock_target, rc);
978 if (lu_fid_eq(old_fid, rr->rr_fid1) || lu_fid_eq(old_fid, rr->rr_fid2))
979 GOTO(out_unlock_target, rc = -EINVAL);
981 mold = mdt_object_find(info->mti_env, info->mti_mdt, old_fid);
983 GOTO(out_unlock_target, rc = PTR_ERR(mold));
985 lh_oldp = &info->mti_lh[MDT_LH_OLD];
986 mdt_lock_reg_init(lh_oldp, LCK_EX);
987 rc = mdt_object_lock(info, mold, lh_oldp, MDS_INODELOCK_LOOKUP,
990 mdt_object_put(info->mti_env, mold);
991 GOTO(out_unlock_target, rc);
994 info->mti_mos[2] = mold;
995 rc = mdt_version_get_check(info, 2);
997 GOTO(out_unlock_old, rc);
999 mdt_set_capainfo(info, 2, old_fid, BYPASS_CAPA);
1001 /* step 4: find & lock the new object. */
1002 /* new target object may not exist now */
1003 lname = mdt_name(info->mti_env, (char *)rr->rr_tgt, rr->rr_tgtlen);
1004 rc = mdo_lookup(info->mti_env, mdt_object_child(mtgtdir),
1005 lname, new_fid, &info->mti_spec);
1007 /* the new_fid should have been filled at this moment */
1008 if (lu_fid_eq(old_fid, new_fid))
1009 GOTO(out_unlock_old, rc);
1011 if (lu_fid_eq(new_fid, rr->rr_fid1) ||
1012 lu_fid_eq(new_fid, rr->rr_fid2))
1013 GOTO(out_unlock_old, rc = -EINVAL);
1015 mdt_lock_reg_init(lh_newp, LCK_EX);
1016 mnew = mdt_object_find(info->mti_env, info->mti_mdt, new_fid);
1018 GOTO(out_unlock_old, rc = PTR_ERR(mnew));
1020 rc = mdt_object_lock(info, mnew, lh_newp,
1021 MDS_INODELOCK_FULL, MDT_CROSS_LOCK);
1023 mdt_object_put(info->mti_env, mnew);
1024 GOTO(out_unlock_old, rc);
1027 info->mti_mos[3] = mnew;
1028 rc = mdt_version_get_check(info, 3);
1030 GOTO(out_unlock_new, rc);
1032 mdt_set_capainfo(info, 3, new_fid, BYPASS_CAPA);
1033 } else if (rc != -EREMOTE && rc != -ENOENT)
1034 GOTO(out_unlock_old, rc);
1036 /* step 5: rename it */
1037 mdt_reint_init_ma(info, ma);
1038 if (!ma->ma_lmm || !ma->ma_cookie)
1039 GOTO(out_unlock_new, rc = -EINVAL);
1041 mdt_fail_write(info->mti_env, info->mti_mdt->mdt_bottom,
1042 OBD_FAIL_MDS_REINT_RENAME_WRITE);
1045 /* Check if @dst is subdir of @src. */
1046 rc = mdt_rename_sanity(info, old_fid);
1048 GOTO(out_unlock_new, rc);
1050 rc = mdo_rename(info->mti_env, mdt_object_child(msrcdir),
1051 mdt_object_child(mtgtdir), old_fid, &slname,
1052 (mnew ? mdt_object_child(mnew) : NULL),
1055 /* handle last link of tgt object */
1056 if (rc == 0 && mnew)
1057 mdt_handle_last_unlink(info, mnew, ma);
1062 mdt_object_unlock_put(info, mnew, lh_newp, rc);
1064 mdt_object_unlock_put(info, mold, lh_oldp, rc);
1066 mdt_object_unlock_put(info, mtgtdir, lh_tgtdirp, rc);
1068 mdt_object_unlock_put(info, msrcdir, lh_srcdirp, rc);
1070 mdt_rename_unlock(&rename_lh);
1074 typedef int (*mdt_reinter)(struct mdt_thread_info *info,
1075 struct mdt_lock_handle *lhc);
1077 static mdt_reinter reinters[REINT_MAX] = {
1078 [REINT_SETATTR] = mdt_reint_setattr,
1079 [REINT_CREATE] = mdt_reint_create,
1080 [REINT_LINK] = mdt_reint_link,
1081 [REINT_UNLINK] = mdt_reint_unlink,
1082 [REINT_RENAME] = mdt_reint_rename,
1083 [REINT_OPEN] = mdt_reint_open,
1084 [REINT_SETXATTR] = mdt_reint_setxattr
1087 int mdt_reint_rec(struct mdt_thread_info *info,
1088 struct mdt_lock_handle *lhc)
1093 rc = reinters[info->mti_rr.rr_opcode](info, lhc);