4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2011, 2014, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * lustre/mdt/mdt_reint.c
38 * Lustre Metadata Target (mdt) reintegration routines
40 * Author: Peter Braam <braam@clusterfs.com>
41 * Author: Andreas Dilger <adilger@clusterfs.com>
42 * Author: Phil Schwan <phil@clusterfs.com>
43 * Author: Huang Hua <huanghua@clusterfs.com>
44 * Author: Yury Umanets <umka@clusterfs.com>
47 #define DEBUG_SUBSYSTEM S_MDS
49 #include <lprocfs_status.h>
50 #include "mdt_internal.h"
51 #include <lustre_lmv.h>
53 static inline void mdt_reint_init_ma(struct mdt_thread_info *info,
56 ma->ma_need = MA_INODE;
60 static int mdt_create_pack_capa(struct mdt_thread_info *info, int rc,
61 struct mdt_object *object,
62 struct mdt_body *repbody)
66 /* for cross-ref mkdir, mds capa has been fetched from remote obj, then
67 * we won't go to below*/
68 if (repbody->mbo_valid & OBD_MD_FLMDSCAPA)
71 if (rc == 0 && info->mti_mdt->mdt_lut.lut_mds_capa &&
72 exp_connect_flags(info->mti_exp) & OBD_CONNECT_MDS_CAPA) {
73 struct lustre_capa *capa;
75 capa = req_capsule_server_get(info->mti_pill, &RMF_CAPA1);
77 capa->lc_opc = CAPA_OPC_MDS_DEFAULT;
78 rc = mo_capa_get(info->mti_env, mdt_object_child(object), capa,
81 repbody->mbo_valid |= OBD_MD_FLMDSCAPA;
88 * Get version of object by fid.
90 * Return real version or ENOENT_VERSION if object doesn't exist
92 static void mdt_obj_version_get(struct mdt_thread_info *info,
93 struct mdt_object *o, __u64 *version)
96 if (mdt_object_exists(o) && !mdt_object_remote(o) &&
97 !fid_is_obf(mdt_object_fid(o)))
98 *version = dt_version_get(info->mti_env, mdt_obj2dt(o));
100 *version = ENOENT_VERSION;
101 CDEBUG(D_INODE, "FID "DFID" version is "LPX64"\n",
102 PFID(mdt_object_fid(o)), *version);
106 * Check version is correct.
108 * Should be called only during replay.
110 static int mdt_version_check(struct ptlrpc_request *req,
111 __u64 version, int idx)
113 __u64 *pre_ver = lustre_msg_get_versions(req->rq_reqmsg);
116 if (!exp_connect_vbr(req->rq_export))
119 LASSERT(req_is_replay(req));
120 /** VBR: version is checked always because costs nothing */
121 LASSERT(idx < PTLRPC_NUM_VERSIONS);
122 /** Sanity check for malformed buffers */
123 if (pre_ver == NULL) {
124 CERROR("No versions in request buffer\n");
125 spin_lock(&req->rq_export->exp_lock);
126 req->rq_export->exp_vbr_failed = 1;
127 spin_unlock(&req->rq_export->exp_lock);
129 } else if (pre_ver[idx] != version) {
130 CDEBUG(D_INODE, "Version mismatch "LPX64" != "LPX64"\n",
131 pre_ver[idx], version);
132 spin_lock(&req->rq_export->exp_lock);
133 req->rq_export->exp_vbr_failed = 1;
134 spin_unlock(&req->rq_export->exp_lock);
141 * Save pre-versions in reply.
143 static void mdt_version_save(struct ptlrpc_request *req, __u64 version,
148 if (!exp_connect_vbr(req->rq_export))
151 LASSERT(!req_is_replay(req));
152 LASSERT(req->rq_repmsg != NULL);
153 reply_ver = lustre_msg_get_versions(req->rq_repmsg);
155 reply_ver[idx] = version;
159 * Save enoent version, it is needed when it is obvious that object doesn't
160 * exist, e.g. child during create.
162 static void mdt_enoent_version_save(struct mdt_thread_info *info, int idx)
164 /* save version of file name for replay, it must be ENOENT here */
165 if (!req_is_replay(mdt_info_req(info))) {
166 info->mti_ver[idx] = ENOENT_VERSION;
167 mdt_version_save(mdt_info_req(info), info->mti_ver[idx], idx);
172 * Get version from disk and save in reply buffer.
174 * Versions are saved in reply only during normal operations not replays.
176 void mdt_version_get_save(struct mdt_thread_info *info,
177 struct mdt_object *mto, int idx)
179 /* don't save versions during replay */
180 if (!req_is_replay(mdt_info_req(info))) {
181 mdt_obj_version_get(info, mto, &info->mti_ver[idx]);
182 mdt_version_save(mdt_info_req(info), info->mti_ver[idx], idx);
187 * Get version from disk and check it, no save in reply.
189 int mdt_version_get_check(struct mdt_thread_info *info,
190 struct mdt_object *mto, int idx)
192 /* only check versions during replay */
193 if (!req_is_replay(mdt_info_req(info)))
196 mdt_obj_version_get(info, mto, &info->mti_ver[idx]);
197 return mdt_version_check(mdt_info_req(info), info->mti_ver[idx], idx);
201 * Get version from disk and check if recovery or just save.
203 int mdt_version_get_check_save(struct mdt_thread_info *info,
204 struct mdt_object *mto, int idx)
208 mdt_obj_version_get(info, mto, &info->mti_ver[idx]);
209 if (req_is_replay(mdt_info_req(info)))
210 rc = mdt_version_check(mdt_info_req(info), info->mti_ver[idx],
213 mdt_version_save(mdt_info_req(info), info->mti_ver[idx], idx);
218 * Lookup with version checking.
220 * This checks version of 'name'. Many reint functions uses 'name' for child not
221 * FID, therefore we need to get object by name and check its version.
223 int mdt_lookup_version_check(struct mdt_thread_info *info,
224 struct mdt_object *p, const struct lu_name *lname,
225 struct lu_fid *fid, int idx)
229 rc = mdo_lookup(info->mti_env, mdt_object_child(p), lname, fid,
231 /* Check version only during replay */
232 if (!req_is_replay(mdt_info_req(info)))
235 info->mti_ver[idx] = ENOENT_VERSION;
237 struct mdt_object *child;
238 child = mdt_object_find(info->mti_env, info->mti_mdt, fid);
239 if (likely(!IS_ERR(child))) {
240 mdt_obj_version_get(info, child, &info->mti_ver[idx]);
241 mdt_object_put(info->mti_env, child);
244 vbrc = mdt_version_check(mdt_info_req(info), info->mti_ver[idx], idx);
245 return vbrc ? vbrc : rc;
250 * mdt_remote_permission: Check whether the remote operation is permitted,
252 * Before we implement async cross-MDT updates (DNE phase 2). There are a few
255 * 1.Only sysadmin can create remote directory and striped directory and
256 * migrate directory now, unless
257 * lctl set_param mdt.*.enable_remote_dir_gid=allow_gid.
258 * 2.Remote directory can only be created on MDT0, unless
259 * lctl set_param mdt.*.enable_remote_dir = 1
260 * 3.Only new clients can access remote dir( >= 2.4) and striped dir(>= 2.6),
261 * old client will return -ENOTSUPP.
263 * XXX these check are only needed for remote synchronization, once async
264 * update is supported, these check will be removed.
266 * param[in]info: execution environment.
267 * param[in]parent: the directory of this operation.
268 * param[in]child: the child of this operation.
270 * retval = 0 remote operation is allowed.
271 * < 0 remote operation is denied.
273 static int mdt_remote_permission(struct mdt_thread_info *info,
274 struct mdt_object *parent,
275 struct mdt_object *child)
277 struct mdt_device *mdt = info->mti_mdt;
278 struct lu_ucred *uc = mdt_ucred(info);
279 struct md_op_spec *spec = &info->mti_spec;
280 struct lu_attr *attr = &info->mti_attr.ma_attr;
281 struct obd_export *exp = mdt_info_req(info)->rq_export;
283 /* Only check create remote directory, striped directory and
285 if (mdt_object_remote(parent) == 0 && mdt_object_remote(child) == 0 &&
286 !(S_ISDIR(attr->la_mode) && spec->u.sp_ea.eadata != NULL &&
287 spec->u.sp_ea.eadatalen != 0) &&
288 info->mti_rr.rr_opcode != REINT_MIGRATE)
291 if (!md_capable(uc, CFS_CAP_SYS_ADMIN)) {
292 if (uc->uc_gid != mdt->mdt_enable_remote_dir_gid &&
293 mdt->mdt_enable_remote_dir_gid != -1)
297 if (mdt->mdt_enable_remote_dir == 0) {
298 struct seq_server_site *ss = mdt_seq_site(mdt);
299 struct lu_seq_range range = { 0 };
302 fld_range_set_type(&range, LU_SEQ_RANGE_MDT);
303 rc = fld_server_lookup(info->mti_env, ss->ss_server_fld,
304 fid_seq(mdt_object_fid(parent)), &range);
308 if (range.lsr_index != 0)
312 if (!mdt_is_dne_client(exp))
315 if (S_ISDIR(attr->la_mode) && spec->u.sp_ea.eadata != NULL &&
316 spec->u.sp_ea.eadatalen != 0) {
317 const struct lmv_user_md *lum = spec->u.sp_ea.eadata;
319 if (le32_to_cpu(lum->lum_stripe_count) > 1 &&
320 !mdt_is_striped_client(exp))
328 * VBR: we save three versions in reply:
329 * 0 - parent. Check that parent version is the same during replay.
330 * 1 - name. Version of 'name' if file exists with the same name or
331 * ENOENT_VERSION, it is needed because file may appear due to missed replays.
332 * 2 - child. Version of child by FID. Must be ENOENT. It is mostly sanity
335 static int mdt_md_create(struct mdt_thread_info *info)
337 struct mdt_device *mdt = info->mti_mdt;
338 struct mdt_object *parent;
339 struct mdt_object *child;
340 struct mdt_lock_handle *lh;
341 struct mdt_body *repbody;
342 struct md_attr *ma = &info->mti_attr;
343 struct mdt_reint_record *rr = &info->mti_rr;
347 DEBUG_REQ(D_INODE, mdt_info_req(info), "Create ("DNAME"->"DFID") "
349 PNAME(&rr->rr_name), PFID(rr->rr_fid2), PFID(rr->rr_fid1));
351 if (!fid_is_md_operative(rr->rr_fid1))
354 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
356 parent = mdt_object_find(info->mti_env, info->mti_mdt, rr->rr_fid1);
358 RETURN(PTR_ERR(parent));
360 if (!mdt_object_exists(parent))
361 GOTO(put_parent, rc = -ENOENT);
363 lh = &info->mti_lh[MDT_LH_PARENT];
364 mdt_lock_pdo_init(lh, LCK_PW, &rr->rr_name);
365 rc = mdt_object_lock(info, parent, lh, MDS_INODELOCK_UPDATE,
368 GOTO(put_parent, rc);
370 if (!mdt_object_remote(parent)) {
371 rc = mdt_version_get_check_save(info, parent, 0);
373 GOTO(unlock_parent, rc);
377 * Check child name version during replay.
378 * During create replay a file may exist with same name.
380 rc = mdt_lookup_version_check(info, parent, &rr->rr_name,
381 &info->mti_tmp_fid1, 1);
383 GOTO(unlock_parent, rc = -EEXIST);
385 /* -ENOENT is expected here */
387 GOTO(unlock_parent, rc);
389 /* save version of file name for replay, it must be ENOENT here */
390 mdt_enoent_version_save(info, 1);
392 child = mdt_object_new(info->mti_env, mdt, rr->rr_fid2);
393 if (likely(!IS_ERR(child))) {
394 struct md_object *next = mdt_object_child(parent);
396 rc = mdt_remote_permission(info, parent, child);
398 GOTO(out_put_child, rc);
400 ma->ma_need = MA_INODE;
402 /* capa for cross-ref will be stored here */
403 ma->ma_capa = req_capsule_server_get(info->mti_pill,
405 LASSERT(ma->ma_capa);
407 mdt_fail_write(info->mti_env, info->mti_mdt->mdt_bottom,
408 OBD_FAIL_MDS_REINT_CREATE_WRITE);
410 /* Version of child will be updated on disk. */
411 tgt_vbr_obj_set(info->mti_env, mdt_obj2dt(child));
412 rc = mdt_version_get_check_save(info, child, 2);
414 GOTO(out_put_child, rc);
416 /* Let lower layer know current lock mode. */
417 info->mti_spec.sp_cr_mode =
418 mdt_dlm_mode2mdl_mode(lh->mlh_pdo_mode);
421 * Do not perform lookup sanity check. We know that name does
424 info->mti_spec.sp_cr_lookup = 0;
425 info->mti_spec.sp_feat = &dt_directory_features;
427 rc = mdo_create(info->mti_env, next, &rr->rr_name,
428 mdt_object_child(child), &info->mti_spec, ma);
430 rc = mdt_attr_get_complex(info, child, ma);
433 /* Return fid & attr to client. */
434 if (ma->ma_valid & MA_INODE)
435 mdt_pack_attr2body(info, repbody, &ma->ma_attr,
436 mdt_object_fid(child));
439 mdt_create_pack_capa(info, rc, child, repbody);
440 mdt_object_put(info->mti_env, child);
443 mdt_create_pack_capa(info, rc, NULL, repbody);
446 mdt_object_unlock(info, parent, lh, rc);
448 mdt_object_put(info->mti_env, parent);
452 static int mdt_unlock_slaves(struct mdt_thread_info *mti,
453 struct mdt_object *obj, __u64 ibits,
454 struct mdt_lock_handle *s0_lh,
455 struct mdt_object *s0_obj,
456 struct ldlm_enqueue_info *einfo)
458 ldlm_policy_data_t *policy = &mti->mti_policy;
462 if (!S_ISDIR(obj->mot_header.loh_attr))
465 /* Unlock stripe 0 */
466 if (s0_lh != NULL && lustre_handle_is_used(&s0_lh->mlh_reg_lh)) {
467 LASSERT(s0_obj != NULL);
468 mdt_object_unlock_put(mti, s0_obj, s0_lh, 1);
471 memset(policy, 0, sizeof(*policy));
472 policy->l_inodebits.bits = ibits;
474 rc = mo_object_unlock(mti->mti_env, mdt_object_child(obj), einfo,
480 * Lock slave stripes if necessary, the lock handles of slave stripes
481 * will be stored in einfo->ei_cbdata.
483 static int mdt_lock_slaves(struct mdt_thread_info *mti, struct mdt_object *obj,
484 ldlm_mode_t mode, __u64 ibits,
485 struct mdt_lock_handle *s0_lh,
486 struct mdt_object **s0_objp,
487 struct ldlm_enqueue_info *einfo)
489 ldlm_policy_data_t *policy = &mti->mti_policy;
490 struct lu_buf *buf = &mti->mti_buf;
491 struct lmv_mds_md_v1 *lmv;
492 struct lu_fid *fid = &mti->mti_tmp_fid1;
496 if (!S_ISDIR(obj->mot_header.loh_attr))
499 buf->lb_buf = mti->mti_xattr_buf;
500 buf->lb_len = sizeof(mti->mti_xattr_buf);
501 rc = mo_xattr_get(mti->mti_env, mdt_object_child(obj), buf,
504 rc = mdt_big_xattr_get(mti, obj, XATTR_NAME_LMV);
506 buf->lb_buf = mti->mti_big_lmm;
507 buf->lb_len = mti->mti_big_lmmsize;
511 if (rc == -ENODATA || rc == -ENOENT)
518 if (le32_to_cpu(lmv->lmv_magic) != LMV_MAGIC_V1)
521 /* Sigh, 0_stripe and master object are different
522 * object, though they are in the same MDT, to avoid
523 * adding osd_object_lock here, so we will enqueue the
524 * stripe0 lock in MDT0 for now */
525 fid_le_to_cpu(fid, &lmv->lmv_stripe_fids[0]);
526 *s0_objp = mdt_object_find_lock(mti, fid, s0_lh, ibits);
527 if (IS_ERR(*s0_objp))
528 RETURN(PTR_ERR(*s0_objp));
530 memset(einfo, 0, sizeof(*einfo));
531 einfo->ei_type = LDLM_IBITS;
532 einfo->ei_mode = mode;
533 einfo->ei_cb_bl = mdt_remote_blocking_ast;
534 einfo->ei_cb_cp = ldlm_completion_ast;
535 einfo->ei_enq_slave = 1;
536 memset(policy, 0, sizeof(*policy));
537 policy->l_inodebits.bits = ibits;
539 rc = mo_object_lock(mti->mti_env, mdt_object_child(obj), NULL, einfo,
544 static int mdt_attr_set(struct mdt_thread_info *info, struct mdt_object *mo,
547 struct mdt_lock_handle *lh;
548 int do_vbr = ma->ma_attr.la_valid & (LA_MODE|LA_UID|LA_GID|LA_FLAGS);
549 __u64 lockpart = MDS_INODELOCK_UPDATE;
550 struct ldlm_enqueue_info *einfo = &info->mti_einfo;
551 struct mdt_lock_handle *s0_lh;
552 struct mdt_object *s0_obj = NULL;
556 lh = &info->mti_lh[MDT_LH_PARENT];
557 mdt_lock_reg_init(lh, LCK_PW);
559 /* Even though the new MDT will grant PERM lock to the old
560 * client, but the old client will almost ignore that during
561 * So it needs to revoke both LOOKUP and PERM lock here, so
562 * both new and old client can cancel the dcache */
563 if (ma->ma_attr.la_valid & (LA_MODE|LA_UID|LA_GID))
564 lockpart |= MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM;
566 rc = mdt_object_lock(info, mo, lh, lockpart, MDT_LOCAL_LOCK);
570 s0_lh = &info->mti_lh[MDT_LH_LOCAL];
571 mdt_lock_reg_init(s0_lh, LCK_PW);
572 rc = mdt_lock_slaves(info, mo, LCK_PW, lockpart, s0_lh, &s0_obj, einfo);
574 GOTO(out_unlock, rc);
576 /* all attrs are packed into mti_attr in unpack_setattr */
577 mdt_fail_write(info->mti_env, info->mti_mdt->mdt_bottom,
578 OBD_FAIL_MDS_REINT_SETATTR_WRITE);
580 /* This is only for set ctime when rename's source is on remote MDS. */
581 if (unlikely(ma->ma_attr.la_valid == LA_CTIME))
582 ma->ma_attr_flags |= MDS_VTX_BYPASS;
584 /* VBR: update version if attr changed are important for recovery */
586 /* update on-disk version of changed object */
587 tgt_vbr_obj_set(info->mti_env, mdt_obj2dt(mo));
588 rc = mdt_version_get_check_save(info, mo, 0);
590 GOTO(out_unlock, rc);
593 /* Ensure constant striping during chown(). See LU-2789. */
594 if (ma->ma_attr.la_valid & (LA_UID|LA_GID))
595 mutex_lock(&mo->mot_lov_mutex);
597 /* all attrs are packed into mti_attr in unpack_setattr */
598 rc = mo_attr_set(info->mti_env, mdt_object_child(mo), ma);
600 if (ma->ma_attr.la_valid & (LA_UID|LA_GID))
601 mutex_unlock(&mo->mot_lov_mutex);
604 GOTO(out_unlock, rc);
608 mdt_unlock_slaves(info, mo, lockpart, s0_lh, s0_obj, einfo);
609 mdt_object_unlock(info, mo, lh, rc);
614 * Check HSM flags and add HS_DIRTY flag if relevant.
616 * A file could be set dirty only if it has a copy in the backend (HS_EXISTS)
617 * and is not RELEASED.
619 int mdt_add_dirty_flag(struct mdt_thread_info *info, struct mdt_object *mo,
625 /* If the file was modified, add the dirty flag */
626 ma->ma_need = MA_HSM;
627 rc = mdt_attr_get_complex(info, mo, ma);
629 CERROR("file attribute read error for "DFID": %d.\n",
630 PFID(mdt_object_fid(mo)), rc);
634 /* If an up2date copy exists in the backend, add dirty flag */
635 if ((ma->ma_valid & MA_HSM) && (ma->ma_hsm.mh_flags & HS_EXISTS)
636 && !(ma->ma_hsm.mh_flags & (HS_DIRTY|HS_RELEASED))) {
637 struct mdt_lock_handle *lh = &info->mti_lh[MDT_LH_CHILD];
639 ma->ma_hsm.mh_flags |= HS_DIRTY;
641 mdt_lock_reg_init(lh, LCK_PW);
642 rc = mdt_object_lock(info, mo, lh, MDS_INODELOCK_XATTR,
647 rc = mdt_hsm_attr_set(info, mo, &ma->ma_hsm);
649 CERROR("file attribute change error for "DFID": %d\n",
650 PFID(mdt_object_fid(mo)), rc);
651 mdt_object_unlock(info, mo, lh, rc);
657 static int mdt_reint_setattr(struct mdt_thread_info *info,
658 struct mdt_lock_handle *lhc)
660 struct md_attr *ma = &info->mti_attr;
661 struct mdt_reint_record *rr = &info->mti_rr;
662 struct ptlrpc_request *req = mdt_info_req(info);
663 struct mdt_export_data *med = &req->rq_export->exp_mdt_data;
664 struct mdt_file_data *mfd;
665 struct mdt_object *mo;
666 struct mdt_body *repbody;
670 DEBUG_REQ(D_INODE, req, "setattr "DFID" %x", PFID(rr->rr_fid1),
671 (unsigned int)ma->ma_attr.la_valid);
673 if (info->mti_dlm_req)
674 ldlm_request_cancel(req, info->mti_dlm_req, 0, LATF_SKIP);
676 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
677 mo = mdt_object_find(info->mti_env, info->mti_mdt, rr->rr_fid1);
679 GOTO(out, rc = PTR_ERR(mo));
681 if (!mdt_object_exists(mo))
682 GOTO(out_put, rc = -ENOENT);
684 if (mdt_object_remote(mo))
685 GOTO(out_put, rc = -EREMOTE);
687 /* start a log jounal handle if needed */
688 if (!(mdt_conn_flags(info) & OBD_CONNECT_SOM)) {
689 if ((ma->ma_attr.la_valid & LA_SIZE) ||
690 (rr->rr_flags & MRF_OPEN_TRUNC)) {
691 /* Check write access for the O_TRUNC case */
692 if (mdt_write_read(mo) < 0)
693 GOTO(out_put, rc = -ETXTBSY);
695 } else if (info->mti_ioepoch &&
696 (info->mti_ioepoch->flags & MF_EPOCH_OPEN)) {
697 /* Truncate case. IOEpoch is opened. */
698 rc = mdt_write_get(mo);
702 mfd = mdt_mfd_new(med);
705 GOTO(out_put, rc = -ENOMEM);
708 mdt_ioepoch_open(info, mo, 0);
709 repbody->mbo_ioepoch = mo->mot_ioepoch;
711 mdt_object_get(info->mti_env, mo);
712 mdt_mfd_set_mode(mfd, MDS_FMODE_TRUNC);
713 mfd->mfd_object = mo;
714 mfd->mfd_xid = req->rq_xid;
716 spin_lock(&med->med_open_lock);
717 list_add(&mfd->mfd_list, &med->med_open_head);
718 spin_unlock(&med->med_open_lock);
719 repbody->mbo_handle.cookie = mfd->mfd_handle.h_cookie;
722 som_au = info->mti_ioepoch && info->mti_ioepoch->flags & MF_SOM_CHANGE;
724 /* SOM Attribute update case. Find the proper mfd and update
725 * SOM attributes on the proper object. */
726 if (!(mdt_conn_flags(info) & OBD_CONNECT_SOM))
727 GOTO(out_put, rc = -EPROTO);
729 spin_lock(&med->med_open_lock);
730 mfd = mdt_handle2mfd(med, &info->mti_ioepoch->handle,
733 spin_unlock(&med->med_open_lock);
734 CDEBUG(D_INODE, "no handle for file close: "
735 "fid = "DFID": cookie = "LPX64"\n",
736 PFID(info->mti_rr.rr_fid1),
737 info->mti_ioepoch->handle.cookie);
738 GOTO(out_put, rc = -ESTALE);
741 if (mfd->mfd_mode != MDS_FMODE_SOM ||
742 (info->mti_ioepoch->flags & MF_EPOCH_CLOSE))
743 GOTO(out_put, rc = -EPROTO);
745 class_handle_unhash(&mfd->mfd_handle);
746 list_del_init(&mfd->mfd_list);
747 spin_unlock(&med->med_open_lock);
749 mdt_mfd_close(info, mfd);
750 } else if ((ma->ma_valid & MA_INODE) && ma->ma_attr.la_valid) {
751 if (ma->ma_valid & MA_LOV)
752 GOTO(out_put, rc = -EPROTO);
754 rc = mdt_attr_set(info, mo, ma);
757 } else if ((ma->ma_valid & MA_LOV) && (ma->ma_valid & MA_INODE)) {
758 struct lu_buf *buf = &info->mti_buf;
760 if (ma->ma_attr.la_valid != 0)
761 GOTO(out_put, rc = -EPROTO);
763 buf->lb_buf = ma->ma_lmm;
764 buf->lb_len = ma->ma_lmm_size;
765 rc = mo_xattr_set(info->mti_env, mdt_object_child(mo),
766 buf, XATTR_NAME_LOV, 0);
769 } else if ((ma->ma_valid & MA_LMV) && (ma->ma_valid & MA_INODE)) {
770 struct lu_buf *buf = &info->mti_buf;
772 if (ma->ma_attr.la_valid != 0)
773 GOTO(out_put, rc = -EPROTO);
775 buf->lb_buf = ma->ma_lmv;
776 buf->lb_len = ma->ma_lmv_size;
777 rc = mo_xattr_set(info->mti_env, mdt_object_child(mo),
778 buf, XATTR_NAME_DEFAULT_LMV, 0);
782 GOTO(out_put, rc = -EPROTO);
785 /* If file data is modified, add the dirty flag */
786 if (ma->ma_attr_flags & MDS_DATA_MODIFIED)
787 rc = mdt_add_dirty_flag(info, mo, ma);
789 ma->ma_need = MA_INODE;
791 rc = mdt_attr_get_complex(info, mo, ma);
795 mdt_pack_attr2body(info, repbody, &ma->ma_attr, mdt_object_fid(mo));
797 if (info->mti_mdt->mdt_lut.lut_oss_capa &&
798 exp_connect_flags(info->mti_exp) & OBD_CONNECT_OSS_CAPA &&
799 S_ISREG(lu_object_attr(&mo->mot_obj)) &&
800 (ma->ma_attr.la_valid & LA_SIZE) && !som_au) {
801 struct lustre_capa *capa;
803 capa = req_capsule_server_get(info->mti_pill, &RMF_CAPA2);
805 capa->lc_opc = CAPA_OPC_OSS_DEFAULT | CAPA_OPC_OSS_TRUNC;
806 rc = mo_capa_get(info->mti_env, mdt_object_child(mo), capa, 0);
809 repbody->mbo_valid |= OBD_MD_FLOSSCAPA;
814 mdt_object_put(info->mti_env, mo);
817 mdt_counter_incr(req, LPROC_MDT_SETATTR);
819 mdt_client_compatibility(info);
820 rc2 = mdt_fix_reply(info);
826 static int mdt_reint_create(struct mdt_thread_info *info,
827 struct mdt_lock_handle *lhc)
829 struct ptlrpc_request *req = mdt_info_req(info);
833 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_REINT_CREATE))
834 RETURN(err_serious(-ESTALE));
836 if (info->mti_dlm_req)
837 ldlm_request_cancel(mdt_info_req(info),
838 info->mti_dlm_req, 0, LATF_SKIP);
840 if (!lu_name_is_valid(&info->mti_rr.rr_name))
843 switch (info->mti_attr.ma_attr.la_mode & S_IFMT) {
845 mdt_counter_incr(req, LPROC_MDT_MKDIR);
853 /* Special file should stay on the same node as parent. */
854 mdt_counter_incr(req, LPROC_MDT_MKNOD);
857 CERROR("%s: Unsupported mode %o\n",
858 mdt_obd_name(info->mti_mdt),
859 info->mti_attr.ma_attr.la_mode);
860 RETURN(err_serious(-EOPNOTSUPP));
863 rc = mdt_md_create(info);
868 * VBR: save parent version in reply and child version getting by its name.
869 * Version of child is getting and checking during its lookup. If
871 static int mdt_reint_unlink(struct mdt_thread_info *info,
872 struct mdt_lock_handle *lhc)
874 struct mdt_reint_record *rr = &info->mti_rr;
875 struct ptlrpc_request *req = mdt_info_req(info);
876 struct md_attr *ma = &info->mti_attr;
877 struct lu_fid *child_fid = &info->mti_tmp_fid1;
878 struct mdt_object *mp;
879 struct mdt_object *mc;
880 struct mdt_lock_handle *parent_lh;
881 struct mdt_lock_handle *child_lh;
882 struct ldlm_enqueue_info *einfo = &info->mti_einfo;
883 struct mdt_lock_handle *s0_lh = NULL;
884 struct mdt_object *s0_obj = NULL;
889 DEBUG_REQ(D_INODE, req, "unlink "DFID"/"DNAME"", PFID(rr->rr_fid1),
890 PNAME(&rr->rr_name));
892 if (info->mti_dlm_req)
893 ldlm_request_cancel(req, info->mti_dlm_req, 0, LATF_SKIP);
895 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_REINT_UNLINK))
896 RETURN(err_serious(-ENOENT));
898 if (!fid_is_md_operative(rr->rr_fid1))
902 * step 1: Found the parent.
904 mp = mdt_object_find(info->mti_env, info->mti_mdt, rr->rr_fid1);
910 parent_lh = &info->mti_lh[MDT_LH_PARENT];
911 mdt_lock_pdo_init(parent_lh, LCK_PW, &rr->rr_name);
912 rc = mdt_object_lock(info, mp, parent_lh, MDS_INODELOCK_UPDATE,
915 GOTO(put_parent, rc);
917 if (!mdt_object_remote(mp)) {
918 rc = mdt_version_get_check_save(info, mp, 0);
920 GOTO(unlock_parent, rc);
923 /* step 2: find & lock the child */
924 /* lookup child object along with version checking */
926 rc = mdt_lookup_version_check(info, mp, &rr->rr_name, child_fid, 1);
928 /* Name might not be able to find during resend of
929 * remote unlink, considering following case.
930 * dir_A is a remote directory, the name entry of
931 * dir_A is on MDT0, the directory is on MDT1,
933 * 1. client sends unlink req to MDT1.
934 * 2. MDT1 sends name delete update to MDT0.
935 * 3. name entry is being deleted in MDT0 synchronously.
936 * 4. MDT1 is restarted.
937 * 5. client resends unlink req to MDT1. So it can not
938 * find the name entry on MDT0 anymore.
939 * In this case, MDT1 only needs to destory the local
942 if (mdt_object_remote(mp) && rc == -ENOENT &&
943 !fid_is_zero(rr->rr_fid2) &&
944 lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT) {
946 *child_fid = *rr->rr_fid2;
948 GOTO(unlock_parent, rc);
952 if (!fid_is_md_operative(child_fid))
953 GOTO(unlock_parent, rc = -EPERM);
955 /* We will lock the child regardless it is local or remote. No harm. */
956 mc = mdt_object_find(info->mti_env, info->mti_mdt, child_fid);
958 GOTO(unlock_parent, rc = PTR_ERR(mc));
960 child_lh = &info->mti_lh[MDT_LH_CHILD];
961 mdt_lock_reg_init(child_lh, LCK_EX);
962 if (mdt_object_remote(mc)) {
963 struct mdt_body *repbody;
965 if (!fid_is_zero(rr->rr_fid2)) {
966 CDEBUG(D_INFO, "%s: name "DNAME" cannot find "DFID"\n",
967 mdt_obd_name(info->mti_mdt),
968 PNAME(&rr->rr_name), PFID(mdt_object_fid(mc)));
969 GOTO(put_child, rc = -ENOENT);
971 CDEBUG(D_INFO, "%s: name "DNAME": "DFID" is on another MDT\n",
972 mdt_obd_name(info->mti_mdt),
973 PNAME(&rr->rr_name), PFID(mdt_object_fid(mc)));
975 if (!mdt_is_dne_client(req->rq_export))
976 /* Return -ENOTSUPP for old client */
977 GOTO(put_child, rc = -ENOTSUPP);
979 if (info->mti_spec.sp_rm_entry) {
980 struct lu_ucred *uc = mdt_ucred(info);
982 if (!md_capable(uc, CFS_CAP_SYS_ADMIN)) {
983 CERROR("%s: unlink remote entry is only "
984 "permitted for administrator: rc = %d\n",
985 mdt_obd_name(info->mti_mdt),
987 GOTO(put_child, rc = -EPERM);
990 ma->ma_need = MA_INODE;
992 mdt_set_capainfo(info, 1, child_fid, BYPASS_CAPA);
993 rc = mdo_unlink(info->mti_env, mdt_object_child(mp),
994 NULL, &rr->rr_name, ma, no_name);
997 /* Revoke the LOOKUP lock of the remote object granted by
998 * this MDT. Since the unlink will happen on another MDT,
999 * it will release the LOOKUP lock right away. Then What
1000 * would happen if another client try to grab the LOOKUP
1001 * lock at the same time with unlink XXX */
1002 mdt_object_lock(info, mc, child_lh, MDS_INODELOCK_LOOKUP,
1004 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1005 LASSERT(repbody != NULL);
1006 repbody->mbo_fid1 = *mdt_object_fid(mc);
1007 repbody->mbo_valid |= (OBD_MD_FLID | OBD_MD_MDS);
1008 GOTO(unlock_child, rc = -EREMOTE);
1009 } else if (info->mti_spec.sp_rm_entry) {
1011 CDEBUG(D_INFO, "%s: no rm_entry on local dir '"DNAME"': "
1013 mdt_obd_name(info->mti_mdt), PNAME(&rr->rr_name), rc);
1014 GOTO(put_child, rc);
1017 /* We used to acquire MDS_INODELOCK_FULL here but we can't do
1018 * this now because a running HSM restore on the child (unlink
1019 * victim) will hold the layout lock. See LU-4002. */
1020 rc = mdt_object_lock(info, mc, child_lh,
1021 MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE,
1024 GOTO(put_child, rc);
1026 * Now we can only make sure we need MA_INODE, in mdd layer, will check
1027 * whether need MA_LOV and MA_COOKIE.
1029 ma->ma_need = MA_INODE;
1032 s0_lh = &info->mti_lh[MDT_LH_LOCAL];
1033 mdt_lock_reg_init(s0_lh, LCK_EX);
1034 rc = mdt_lock_slaves(info, mc, LCK_EX, MDS_INODELOCK_UPDATE, s0_lh,
1037 GOTO(unlock_child, rc);
1039 mdt_fail_write(info->mti_env, info->mti_mdt->mdt_bottom,
1040 OBD_FAIL_MDS_REINT_UNLINK_WRITE);
1041 /* save version when object is locked */
1042 mdt_version_get_save(info, mc, 1);
1044 mdt_set_capainfo(info, 1, child_fid, BYPASS_CAPA);
1046 mutex_lock(&mc->mot_lov_mutex);
1048 rc = mdo_unlink(info->mti_env, mdt_object_child(mp),
1049 mdt_object_child(mc), &rr->rr_name, ma, no_name);
1051 mutex_unlock(&mc->mot_lov_mutex);
1053 if (rc == 0 && !lu_object_is_dying(&mc->mot_header))
1054 rc = mdt_attr_get_complex(info, mc, ma);
1056 mdt_handle_last_unlink(info, mc, ma);
1058 if (ma->ma_valid & MA_INODE) {
1059 switch (ma->ma_attr.la_mode & S_IFMT) {
1061 mdt_counter_incr(req, LPROC_MDT_RMDIR);
1069 mdt_counter_incr(req, LPROC_MDT_UNLINK);
1072 LASSERTF(0, "bad file type %o unlinking\n",
1073 ma->ma_attr.la_mode);
1080 mdt_unlock_slaves(info, mc, MDS_INODELOCK_UPDATE, s0_lh, s0_obj, einfo);
1081 mdt_object_unlock(info, mc, child_lh, rc);
1083 mdt_object_put(info->mti_env, mc);
1085 mdt_object_unlock(info, mp, parent_lh, rc);
1087 mdt_object_put(info->mti_env, mp);
1093 * VBR: save versions in reply: 0 - parent; 1 - child by fid; 2 - target by
1096 static int mdt_reint_link(struct mdt_thread_info *info,
1097 struct mdt_lock_handle *lhc)
1099 struct mdt_reint_record *rr = &info->mti_rr;
1100 struct ptlrpc_request *req = mdt_info_req(info);
1101 struct md_attr *ma = &info->mti_attr;
1102 struct mdt_object *ms;
1103 struct mdt_object *mp;
1104 struct mdt_lock_handle *lhs;
1105 struct mdt_lock_handle *lhp;
1109 DEBUG_REQ(D_INODE, req, "link "DFID" to "DFID"/"DNAME,
1110 PFID(rr->rr_fid1), PFID(rr->rr_fid2), PNAME(&rr->rr_name));
1112 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_REINT_LINK))
1113 RETURN(err_serious(-ENOENT));
1115 if (info->mti_dlm_req)
1116 ldlm_request_cancel(req, info->mti_dlm_req, 0, LATF_SKIP);
1118 /* Invalid case so return error immediately instead of
1120 if (lu_fid_eq(rr->rr_fid1, rr->rr_fid2))
1123 if (!fid_is_md_operative(rr->rr_fid1) ||
1124 !fid_is_md_operative(rr->rr_fid2))
1127 /* step 1: find & lock the target parent dir */
1128 lhp = &info->mti_lh[MDT_LH_PARENT];
1129 mdt_lock_pdo_init(lhp, LCK_PW, &rr->rr_name);
1130 mp = mdt_object_find_lock(info, rr->rr_fid2, lhp,
1131 MDS_INODELOCK_UPDATE);
1133 RETURN(PTR_ERR(mp));
1135 rc = mdt_version_get_check_save(info, mp, 0);
1137 GOTO(out_unlock_parent, rc);
1139 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_RENAME3, 5);
1141 /* step 2: find & lock the source */
1142 lhs = &info->mti_lh[MDT_LH_CHILD];
1143 mdt_lock_reg_init(lhs, LCK_EX);
1145 ms = mdt_object_find(info->mti_env, info->mti_mdt, rr->rr_fid1);
1147 GOTO(out_unlock_parent, rc = PTR_ERR(ms));
1149 if (!mdt_object_exists(ms)) {
1150 mdt_object_put(info->mti_env, ms);
1151 CDEBUG(D_INFO, "%s: "DFID" does not exist.\n",
1152 mdt_obd_name(info->mti_mdt), PFID(rr->rr_fid1));
1153 GOTO(out_unlock_parent, rc = -ENOENT);
1156 if (mdt_object_remote(ms)) {
1157 mdt_object_put(info->mti_env, ms);
1158 CERROR("%s: source inode "DFID" on remote MDT from "DFID"\n",
1159 mdt_obd_name(info->mti_mdt), PFID(rr->rr_fid1),
1161 GOTO(out_unlock_parent, rc = -EXDEV);
1164 rc = mdt_object_lock(info, ms, lhs, MDS_INODELOCK_UPDATE |
1165 MDS_INODELOCK_XATTR, MDT_LOCAL_LOCK);
1167 mdt_object_put(info->mti_env, ms);
1168 GOTO(out_unlock_parent, rc);
1171 /* step 3: link it */
1172 mdt_fail_write(info->mti_env, info->mti_mdt->mdt_bottom,
1173 OBD_FAIL_MDS_REINT_LINK_WRITE);
1175 tgt_vbr_obj_set(info->mti_env, mdt_obj2dt(ms));
1176 rc = mdt_version_get_check_save(info, ms, 1);
1178 GOTO(out_unlock_child, rc);
1180 /** check target version by name during replay */
1181 rc = mdt_lookup_version_check(info, mp, &rr->rr_name,
1182 &info->mti_tmp_fid1, 2);
1183 if (rc != 0 && rc != -ENOENT)
1184 GOTO(out_unlock_child, rc);
1185 /* save version of file name for replay, it must be ENOENT here */
1186 if (!req_is_replay(mdt_info_req(info))) {
1187 if (rc != -ENOENT) {
1188 CDEBUG(D_INFO, "link target "DNAME" existed!\n",
1189 PNAME(&rr->rr_name));
1190 GOTO(out_unlock_child, rc = -EEXIST);
1192 info->mti_ver[2] = ENOENT_VERSION;
1193 mdt_version_save(mdt_info_req(info), info->mti_ver[2], 2);
1196 rc = mdo_link(info->mti_env, mdt_object_child(mp),
1197 mdt_object_child(ms), &rr->rr_name, ma);
1200 mdt_counter_incr(req, LPROC_MDT_LINK);
1204 mdt_object_unlock_put(info, ms, lhs, rc);
1206 mdt_object_unlock_put(info, mp, lhp, rc);
1210 * lock the part of the directory according to the hash of the name
1211 * (lh->mlh_pdo_hash) in parallel directory lock.
1213 static int mdt_pdir_hash_lock(struct mdt_thread_info *info,
1214 struct mdt_lock_handle *lh,
1215 struct mdt_object *obj, __u64 ibits)
1217 struct ldlm_res_id *res = &info->mti_res_id;
1218 struct ldlm_namespace *ns = info->mti_mdt->mdt_namespace;
1219 ldlm_policy_data_t *policy = &info->mti_policy;
1223 * Finish res_id initializing by name hash marking part of
1224 * directory which is taking modification.
1226 LASSERT(lh->mlh_pdo_hash != 0);
1227 fid_build_pdo_res_name(mdt_object_fid(obj), lh->mlh_pdo_hash, res);
1228 memset(policy, 0, sizeof(*policy));
1229 policy->l_inodebits.bits = ibits;
1231 * Use LDLM_FL_LOCAL_ONLY for this lock. We do not know yet if it is
1232 * going to be sent to client. If it is - mdt_intent_policy() path will
1233 * fix it up and turn FL_LOCAL flag off.
1235 rc = mdt_fid_lock(ns, &lh->mlh_reg_lh, lh->mlh_reg_mode, policy,
1236 res, LDLM_FL_LOCAL_ONLY | LDLM_FL_ATOMIC_CB,
1237 &info->mti_exp->exp_handle.h_cookie);
1241 enum mdt_rename_lock {
1247 * Get BFL lock for rename or migrate process, right now, it does not support
1248 * cross-MDT rename, so we only need global rename lock during migration.
1250 static int mdt_rename_lock(struct mdt_thread_info *info,
1251 struct lustre_handle *lh,
1252 enum mdt_rename_lock rename_lock)
1254 struct ldlm_namespace *ns = info->mti_mdt->mdt_namespace;
1255 ldlm_policy_data_t *policy = &info->mti_policy;
1256 struct ldlm_res_id *res_id = &info->mti_res_id;
1261 /* XXX only do global rename lock for migration */
1262 if (mdt_seq_site(info->mti_mdt)->ss_node_id != 0 &&
1263 rename_lock == MRL_MIGRATE) {
1264 struct lu_fid *fid = &info->mti_tmp_fid1;
1265 struct mdt_object *obj;
1267 /* XXX, right now, it has to use object API to
1268 * enqueue lock cross MDT, so it will enqueue
1269 * rename lock(with LUSTRE_BFL_FID) by root object */
1271 obj = mdt_object_find(info->mti_env, info->mti_mdt, fid);
1273 RETURN(PTR_ERR(obj));
1275 LASSERT(mdt_object_remote(obj));
1276 rc = mdt_remote_object_lock(info, obj,
1277 &LUSTRE_BFL_FID, lh,
1279 MDS_INODELOCK_UPDATE);
1280 mdt_object_put(info->mti_env, obj);
1282 fid_build_reg_res_name(&LUSTRE_BFL_FID, res_id);
1283 memset(policy, 0, sizeof *policy);
1284 policy->l_inodebits.bits = MDS_INODELOCK_UPDATE;
1285 flags = LDLM_FL_LOCAL_ONLY | LDLM_FL_ATOMIC_CB;
1286 rc = ldlm_cli_enqueue_local(ns, res_id, LDLM_IBITS, policy,
1287 LCK_EX, &flags, ldlm_blocking_ast,
1288 ldlm_completion_ast, NULL, NULL, 0,
1290 &info->mti_exp->exp_handle.h_cookie,
1297 static void mdt_rename_unlock(struct lustre_handle *lh)
1300 LASSERT(lustre_handle_is_used(lh));
1301 /* Cancel the single rename lock right away */
1302 ldlm_lock_decref_and_cancel(lh, LCK_EX);
1307 * This is is_subdir() variant, it is CMD if cmm forwards it to correct
1308 * target. Source should not be ancestor of target dir. May be other rename
1309 * checks can be moved here later.
1311 static int mdt_is_subdir(struct mdt_thread_info *info,
1312 struct mdt_object *dir,
1313 const struct lu_fid *fid)
1315 struct lu_fid dir_fid = dir->mot_header.loh_fid;
1319 /* If the source and target are in the same directory, they can not
1320 * be parent/child relationship, so subdir check is not needed */
1321 if (lu_fid_eq(&dir_fid, fid))
1324 if (!mdt_object_exists(dir))
1327 rc = mdo_is_subdir(info->mti_env, mdt_object_child(dir),
1330 CERROR("%s: failed subdir check in "DFID" for "DFID
1331 ": rc = %d\n", mdt_obd_name(info->mti_mdt),
1332 PFID(&dir_fid), PFID(fid), rc);
1333 /* Return EINVAL only if a parent is the @fid */
1337 /* check the found fid */
1338 if (lu_fid_eq(&dir_fid, fid))
1345 /* Update object linkEA */
1346 struct mdt_lock_list {
1347 struct mdt_object *mll_obj;
1348 struct mdt_lock_handle mll_lh;
1349 struct list_head mll_list;
1352 static void mdt_unlock_list(struct mdt_thread_info *info,
1353 struct list_head *list, int rc)
1355 struct mdt_lock_list *mll;
1356 struct mdt_lock_list *mll2;
1358 list_for_each_entry_safe(mll, mll2, list, mll_list) {
1359 mdt_object_unlock_put(info, mll->mll_obj, &mll->mll_lh, rc);
1360 list_del(&mll->mll_list);
1365 static int mdt_lock_objects_in_linkea(struct mdt_thread_info *info,
1366 struct mdt_object *obj,
1367 struct mdt_object *pobj,
1368 struct list_head *lock_list)
1370 struct lu_buf *buf = &info->mti_big_buf;
1371 struct linkea_data ldata = { 0 };
1376 if (S_ISDIR(lu_object_attr(&obj->mot_obj)))
1379 buf = lu_buf_check_and_alloc(buf, PATH_MAX);
1380 if (buf->lb_buf == NULL)
1384 rc = mdt_links_read(info, obj, &ldata);
1386 if (rc == -ENOENT || rc == -ENODATA)
1391 LASSERT(ldata.ld_leh != NULL);
1392 ldata.ld_lee = (struct link_ea_entry *)(ldata.ld_leh + 1);
1393 for (count = 0; count < ldata.ld_leh->leh_reccount; count++) {
1394 struct mdt_device *mdt = info->mti_mdt;
1395 struct mdt_object *mdt_pobj;
1396 struct mdt_lock_list *mll;
1397 struct lu_name name;
1400 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen,
1402 ldata.ld_lee = (struct link_ea_entry *)((char *)ldata.ld_lee +
1404 mdt_pobj = mdt_object_find(info->mti_env, mdt, &fid);
1405 if (IS_ERR(mdt_pobj)) {
1406 CWARN("%s: cannot find obj "DFID": rc = %ld\n",
1407 mdt_obd_name(mdt), PFID(&fid), PTR_ERR(mdt_pobj));
1411 if (!mdt_object_exists(mdt_pobj)) {
1412 CDEBUG(D_INFO, "%s: obj "DFID" does not exist\n",
1413 mdt_obd_name(mdt), PFID(&fid));
1414 mdt_object_put(info->mti_env, mdt_pobj);
1418 if (mdt_pobj == pobj) {
1419 CDEBUG(D_INFO, "%s: skipping parent obj "DFID"\n",
1420 mdt_obd_name(mdt), PFID(&fid));
1421 mdt_object_put(info->mti_env, mdt_pobj);
1427 mdt_object_put(info->mti_env, mdt_pobj);
1428 GOTO(out, rc = -ENOMEM);
1431 mdt_lock_pdo_init(&mll->mll_lh, LCK_PW, &name);
1432 rc = mdt_object_lock(info, mdt_pobj, &mll->mll_lh,
1433 MDS_INODELOCK_UPDATE,
1436 CERROR("%s: cannot lock "DFID": rc =%d\n",
1437 mdt_obd_name(mdt), PFID(&fid), rc);
1438 mdt_object_put(info->mti_env, mdt_pobj);
1443 INIT_LIST_HEAD(&mll->mll_list);
1444 mll->mll_obj = mdt_pobj;
1445 list_add_tail(&mll->mll_list, lock_list);
1449 mdt_unlock_list(info, lock_list, rc);
1453 /* migrate files from one MDT to another MDT */
1454 static int mdt_reint_migrate_internal(struct mdt_thread_info *info,
1455 struct mdt_lock_handle *lhc)
1457 struct mdt_reint_record *rr = &info->mti_rr;
1458 struct md_attr *ma = &info->mti_attr;
1459 struct mdt_object *msrcdir;
1460 struct mdt_object *mold;
1461 struct mdt_object *mnew = NULL;
1462 struct mdt_lock_handle *lh_dirp;
1463 struct mdt_lock_handle *lh_childp;
1464 struct mdt_lock_handle *lh_tgtp = NULL;
1465 struct lu_fid *old_fid = &info->mti_tmp_fid1;
1466 struct list_head lock_list;
1470 CDEBUG(D_INODE, "migrate "DFID"/"DNAME" to "DFID"\n", PFID(rr->rr_fid1),
1471 PNAME(&rr->rr_name), PFID(rr->rr_fid2));
1473 /* 1: lock the source dir. */
1474 msrcdir = mdt_object_find(info->mti_env, info->mti_mdt, rr->rr_fid1);
1475 if (IS_ERR(msrcdir)) {
1476 CERROR("%s: cannot find source dir "DFID" : rc = %d\n",
1477 mdt_obd_name(info->mti_mdt), PFID(rr->rr_fid1),
1478 (int)PTR_ERR(msrcdir));
1479 RETURN(PTR_ERR(msrcdir));
1482 lh_dirp = &info->mti_lh[MDT_LH_PARENT];
1483 mdt_lock_pdo_init(lh_dirp, LCK_PW, &rr->rr_name);
1484 rc = mdt_object_lock(info, msrcdir, lh_dirp,
1485 MDS_INODELOCK_UPDATE,
1488 GOTO(out_put_parent, rc);
1490 if (!mdt_object_remote(msrcdir)) {
1491 rc = mdt_version_get_check_save(info, msrcdir, 0);
1493 GOTO(out_unlock_parent, rc);
1496 /* 2: sanity check and find the object to be migrated. */
1498 rc = mdt_lookup_version_check(info, msrcdir, &rr->rr_name, old_fid, 2);
1500 GOTO(out_unlock_parent, rc);
1502 if (lu_fid_eq(old_fid, rr->rr_fid1) || lu_fid_eq(old_fid, rr->rr_fid2))
1503 GOTO(out_unlock_parent, rc = -EINVAL);
1505 if (!fid_is_md_operative(old_fid))
1506 GOTO(out_unlock_parent, rc = -EPERM);
1508 mold = mdt_object_find(info->mti_env, info->mti_mdt, old_fid);
1510 GOTO(out_unlock_parent, rc = PTR_ERR(mold));
1512 if (mdt_object_remote(mold)) {
1513 CERROR("%s: source "DFID" is on the remote MDT\n",
1514 mdt_obd_name(info->mti_mdt), PFID(old_fid));
1515 GOTO(out_put_child, rc = -EREMOTE);
1518 if (S_ISREG(lu_object_attr(&mold->mot_obj)) &&
1519 !mdt_object_remote(msrcdir)) {
1520 CERROR("%s: parent "DFID" is still on the same"
1521 " MDT, which should be migrated first:"
1522 " rc = %d\n", mdt_obd_name(info->mti_mdt),
1523 PFID(mdt_object_fid(msrcdir)), -EPERM);
1524 GOTO(out_put_child, rc = -EPERM);
1527 rc = mdt_remote_permission(info, msrcdir, mold);
1529 GOTO(out_put_child, rc);
1531 /* 3: iterate the linkea of the object and lock all of the objects */
1532 INIT_LIST_HEAD(&lock_list);
1533 rc = mdt_lock_objects_in_linkea(info, mold, msrcdir, &lock_list);
1535 GOTO(out_put_child, rc);
1537 /* 4: lock of the object migrated object */
1538 lh_childp = &info->mti_lh[MDT_LH_OLD];
1539 mdt_lock_reg_init(lh_childp, LCK_EX);
1540 rc = mdt_object_lock(info, mold, lh_childp,
1541 MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
1542 MDS_INODELOCK_LAYOUT, MDT_CROSS_LOCK);
1544 GOTO(out_unlock_list, rc);
1546 ma->ma_need = MA_LMV;
1548 ma->ma_lmv = (union lmv_mds_md *)info->mti_xattr_buf;
1549 ma->ma_lmv_size = sizeof(info->mti_xattr_buf);
1550 rc = mdt_stripe_get(info, mold, ma, XATTR_NAME_LMV);
1552 GOTO(out_unlock_list, rc);
1554 if ((ma->ma_valid & MA_LMV)) {
1555 struct lmv_mds_md_v1 *lmm1;
1557 lmv_le_to_cpu(ma->ma_lmv, ma->ma_lmv);
1558 lmm1 = &ma->ma_lmv->lmv_md_v1;
1559 if (!(lmm1->lmv_hash_type & LMV_HASH_FLAG_MIGRATION)) {
1560 CERROR("%s: can not migrate striped dir "DFID
1561 ": rc = %d\n", mdt_obd_name(info->mti_mdt),
1562 PFID(mdt_object_fid(mold)), -EPERM);
1563 GOTO(out_unlock_child, rc = -EPERM);
1566 if (!fid_is_sane(&lmm1->lmv_stripe_fids[1]))
1567 GOTO(out_unlock_child, rc = -EINVAL);
1569 mnew = mdt_object_find(info->mti_env, info->mti_mdt,
1570 &lmm1->lmv_stripe_fids[1]);
1572 GOTO(out_unlock_child, rc = PTR_ERR(mnew));
1574 if (!mdt_object_remote(mnew)) {
1575 CERROR("%s: "DFID" being migrated is on this MDT:"
1576 " rc = %d\n", mdt_obd_name(info->mti_mdt),
1577 PFID(rr->rr_fid2), -EPERM);
1578 GOTO(out_put_new, rc = -EPERM);
1581 lh_tgtp = &info->mti_lh[MDT_LH_CHILD];
1582 mdt_lock_reg_init(lh_tgtp, LCK_EX);
1583 rc = mdt_remote_object_lock(info, mnew,
1584 mdt_object_fid(mnew),
1585 &lh_tgtp->mlh_rreg_lh,
1586 lh_tgtp->mlh_rreg_mode,
1587 MDS_INODELOCK_UPDATE);
1590 GOTO(out_put_new, rc);
1593 mnew = mdt_object_find(info->mti_env, info->mti_mdt,
1596 GOTO(out_unlock_child, rc = PTR_ERR(mnew));
1597 if (!mdt_object_remote(mnew)) {
1598 CERROR("%s: Migration "DFID" is on this MDT:"
1599 " rc = %d\n", mdt_obd_name(info->mti_mdt),
1600 PFID(rr->rr_fid2), -EXDEV);
1601 GOTO(out_put_new, rc = -EXDEV);
1606 mdt_reint_init_ma(info, ma);
1608 mdt_fail_write(info->mti_env, info->mti_mdt->mdt_bottom,
1609 OBD_FAIL_MDS_REINT_RENAME_WRITE);
1611 rc = mdo_migrate(info->mti_env, mdt_object_child(msrcdir),
1612 mdt_object_child(mold), &rr->rr_name,
1613 mdt_object_child(mnew), ma);
1615 GOTO(out_unlock_new, rc);
1617 if (lh_tgtp != NULL)
1618 mdt_object_unlock(info, mnew, lh_tgtp, rc);
1621 mdt_object_put(info->mti_env, mnew);
1623 mdt_object_unlock(info, mold, lh_childp, rc);
1625 mdt_unlock_list(info, &lock_list, rc);
1627 mdt_object_put(info->mti_env, mold);
1629 mdt_object_unlock(info, msrcdir, lh_dirp, rc);
1631 mdt_object_put(info->mti_env, msrcdir);
1636 static struct mdt_object *mdt_object_find_check(struct mdt_thread_info *info,
1637 const struct lu_fid *fid,
1640 struct mdt_object *dir;
1644 dir = mdt_object_find(info->mti_env, info->mti_mdt, fid);
1648 /* check early, the real version will be saved after locking */
1649 rc = mdt_version_get_check(info, dir, idx);
1655 mdt_object_put(info->mti_env, dir);
1659 static int mdt_object_lock_save(struct mdt_thread_info *info,
1660 struct mdt_object *dir,
1661 struct mdt_lock_handle *lh,
1666 /* we lock the target dir if it is local */
1667 rc = mdt_object_lock(info, dir, lh, MDS_INODELOCK_UPDATE,
1672 /* get and save correct version after locking */
1673 mdt_version_get_save(info, dir, idx);
1678 static int mdt_rename_parents_lock(struct mdt_thread_info *info,
1679 struct mdt_object **srcp,
1680 struct mdt_object **tgtp)
1682 struct mdt_reint_record *rr = &info->mti_rr;
1683 const struct lu_fid *fid_src = rr->rr_fid1;
1684 const struct lu_fid *fid_tgt = rr->rr_fid2;
1685 struct mdt_lock_handle *lh_src = &info->mti_lh[MDT_LH_PARENT];
1686 struct mdt_lock_handle *lh_tgt = &info->mti_lh[MDT_LH_CHILD];
1687 struct mdt_object *src;
1688 struct mdt_object *tgt;
1693 /* find both parents. */
1694 src = mdt_object_find_check(info, fid_src, 0);
1696 RETURN(PTR_ERR(src));
1698 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_RENAME3, 5);
1700 if (lu_fid_eq(fid_src, fid_tgt)) {
1702 mdt_object_get(info->mti_env, tgt);
1704 /* Check if the @src is not a child of the @tgt, otherwise a
1705 * reverse locking must take place. */
1706 rc = mdt_is_subdir(info, src, fid_tgt);
1710 GOTO(err_src_put, rc);
1712 tgt = mdt_object_find_check(info, fid_tgt, 1);
1714 GOTO(err_src_put, rc = PTR_ERR(tgt));
1716 if (unlikely(mdt_object_remote(tgt))) {
1717 CDEBUG(D_INFO, "Source dir "DFID" target dir "DFID
1718 "on different MDTs\n", PFID(fid_src),
1720 GOTO(err_tgt_put, rc = -EXDEV);
1724 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_RENAME4, 5);
1726 /* lock parents in the proper order. */
1728 rc = mdt_object_lock_save(info, tgt, lh_tgt, 1);
1730 GOTO(err_tgt_put, rc);
1732 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_RENAME, 5);
1734 rc = mdt_object_lock_save(info, src, lh_src, 0);
1736 rc = mdt_object_lock_save(info, src, lh_src, 0);
1738 GOTO(err_tgt_put, rc);
1740 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_RENAME, 5);
1743 rc = mdt_object_lock_save(info, tgt, lh_tgt, 1);
1744 else if (lh_src->mlh_pdo_hash != lh_tgt->mlh_pdo_hash) {
1745 rc = mdt_pdir_hash_lock(info, lh_tgt, tgt,
1746 MDS_INODELOCK_UPDATE);
1747 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_PDO_LOCK2, 10);
1751 GOTO(err_unlock, rc);
1753 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_RENAME4, 5);
1760 /* The order does not matter as the handle is checked inside,
1761 * as well as not used handle. */
1762 mdt_object_unlock(info, src, lh_src, rc);
1763 mdt_object_unlock(info, tgt, lh_tgt, rc);
1765 mdt_object_put(info->mti_env, tgt);
1767 mdt_object_put(info->mti_env, src);
1772 * VBR: rename versions in reply: 0 - src parent; 1 - tgt parent;
1773 * 2 - src child; 3 - tgt child.
1774 * Update on disk version of src child.
1777 * For DNE phase I, only these renames are allowed
1778 * mv src_p/src_c tgt_p/tgt_c
1779 * 1. src_p/src_c/tgt_p/tgt_c are in the same MDT.
1780 * 2. src_p and tgt_p are same directory, and tgt_c does not
1781 * exists. In this case, all of modification will happen
1782 * in the MDT where ithesource parent is, only one remote
1783 * update is needed, i.e. set c_time/m_time on the child.
1784 * And tgt_c will be still in the same MDT as the original
1787 static int mdt_reint_rename_internal(struct mdt_thread_info *info,
1788 struct mdt_lock_handle *lhc)
1790 struct mdt_reint_record *rr = &info->mti_rr;
1791 struct md_attr *ma = &info->mti_attr;
1792 struct ptlrpc_request *req = mdt_info_req(info);
1793 struct mdt_object *msrcdir = NULL;
1794 struct mdt_object *mtgtdir = NULL;
1795 struct mdt_object *mold;
1796 struct mdt_object *mnew = NULL;
1797 struct mdt_lock_handle *lh_srcdirp;
1798 struct mdt_lock_handle *lh_tgtdirp;
1799 struct mdt_lock_handle *lh_oldp = NULL;
1800 struct mdt_lock_handle *lh_newp = NULL;
1801 struct lu_fid *old_fid = &info->mti_tmp_fid1;
1802 struct lu_fid *new_fid = &info->mti_tmp_fid2;
1806 DEBUG_REQ(D_INODE, req, "rename "DFID"/"DNAME" to "DFID"/"DNAME,
1807 PFID(rr->rr_fid1), PNAME(&rr->rr_name),
1808 PFID(rr->rr_fid2), PNAME(&rr->rr_tgt_name));
1810 lh_srcdirp = &info->mti_lh[MDT_LH_PARENT];
1811 mdt_lock_pdo_init(lh_srcdirp, LCK_PW, &rr->rr_name);
1812 lh_tgtdirp = &info->mti_lh[MDT_LH_CHILD];
1813 mdt_lock_pdo_init(lh_tgtdirp, LCK_PW, &rr->rr_tgt_name);
1815 /* step 1&2: lock the source and target dirs. */
1816 rc = mdt_rename_parents_lock(info, &msrcdir, &mtgtdir);
1820 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_RENAME2, 5);
1822 /* step 3: find & lock the old object. */
1824 rc = mdt_lookup_version_check(info, msrcdir, &rr->rr_name, old_fid, 2);
1826 GOTO(out_unlock_parents, rc);
1828 if (lu_fid_eq(old_fid, rr->rr_fid1) || lu_fid_eq(old_fid, rr->rr_fid2))
1829 GOTO(out_unlock_parents, rc = -EINVAL);
1831 if (!fid_is_md_operative(old_fid))
1832 GOTO(out_unlock_parents, rc = -EPERM);
1834 mold = mdt_object_find(info->mti_env, info->mti_mdt, old_fid);
1836 GOTO(out_unlock_parents, rc = PTR_ERR(mold));
1838 /* Check if @mtgtdir is subdir of @mold, before locking child
1839 * to avoid reverse locking. */
1840 rc = mdt_is_subdir(info, mtgtdir, old_fid);
1842 GOTO(out_put_old, rc);
1844 tgt_vbr_obj_set(info->mti_env, mdt_obj2dt(mold));
1845 /* save version after locking */
1846 mdt_version_get_save(info, mold, 2);
1847 mdt_set_capainfo(info, 2, old_fid, BYPASS_CAPA);
1849 /* step 4: find & lock the new object. */
1850 /* new target object may not exist now */
1851 /* lookup with version checking */
1853 rc = mdt_lookup_version_check(info, mtgtdir, &rr->rr_tgt_name, new_fid,
1856 /* the new_fid should have been filled at this moment */
1857 if (lu_fid_eq(old_fid, new_fid))
1858 GOTO(out_put_old, rc);
1860 if (lu_fid_eq(new_fid, rr->rr_fid1) ||
1861 lu_fid_eq(new_fid, rr->rr_fid2))
1862 GOTO(out_put_old, rc = -EINVAL);
1864 if (!fid_is_md_operative(new_fid))
1865 GOTO(out_put_old, rc = -EPERM);
1867 if (mdt_object_remote(mold)) {
1868 CDEBUG(D_INFO, "Src child "DFID" is on another MDT\n",
1870 GOTO(out_put_old, rc = -EXDEV);
1873 mnew = mdt_object_find(info->mti_env, info->mti_mdt, new_fid);
1875 GOTO(out_put_old, rc = PTR_ERR(mnew));
1877 if (mdt_object_remote(mnew)) {
1878 CDEBUG(D_INFO, "src child "DFID" is on another MDT\n",
1880 GOTO(out_put_new, rc = -EXDEV);
1883 /* Before locking the target dir, check we do not replace
1884 * a dir with a non-dir, otherwise it may deadlock with
1885 * link op which tries to create a link in this dir
1886 * back to this non-dir. */
1887 if (S_ISDIR(lu_object_attr(&mnew->mot_obj)) &&
1888 !S_ISDIR(lu_object_attr(&mold->mot_obj)))
1889 GOTO(out_put_new, rc = -EISDIR);
1891 lh_oldp = &info->mti_lh[MDT_LH_OLD];
1892 mdt_lock_reg_init(lh_oldp, LCK_EX);
1893 rc = mdt_object_lock(info, mold, lh_oldp, MDS_INODELOCK_LOOKUP |
1894 MDS_INODELOCK_XATTR, MDT_CROSS_LOCK);
1896 GOTO(out_put_new, rc);
1898 /* Check if @msrcdir is subdir of @mnew, before locking child
1899 * to avoid reverse locking. */
1900 rc = mdt_is_subdir(info, msrcdir, new_fid);
1902 GOTO(out_unlock_old, rc);
1904 /* We used to acquire MDS_INODELOCK_FULL here but we
1905 * can't do this now because a running HSM restore on
1906 * the rename onto victim will hold the layout
1907 * lock. See LU-4002. */
1909 lh_newp = &info->mti_lh[MDT_LH_NEW];
1910 mdt_lock_reg_init(lh_newp, LCK_EX);
1911 rc = mdt_object_lock(info, mnew, lh_newp,
1912 MDS_INODELOCK_LOOKUP |
1913 MDS_INODELOCK_UPDATE,
1916 GOTO(out_unlock_old, rc);
1918 /* get and save version after locking */
1919 mdt_version_get_save(info, mnew, 3);
1920 mdt_set_capainfo(info, 3, new_fid, BYPASS_CAPA);
1921 } else if (rc != -EREMOTE && rc != -ENOENT) {
1922 GOTO(out_put_old, rc);
1924 /* If mnew does not exist and mold are remote directory,
1925 * it only allows rename if they are under same directory */
1926 if (mtgtdir != msrcdir && mdt_object_remote(mold)) {
1927 CDEBUG(D_INFO, "Src child "DFID" is on another MDT\n",
1929 GOTO(out_put_old, rc = -EXDEV);
1932 lh_oldp = &info->mti_lh[MDT_LH_OLD];
1933 mdt_lock_reg_init(lh_oldp, LCK_EX);
1934 rc = mdt_object_lock(info, mold, lh_oldp, MDS_INODELOCK_LOOKUP |
1935 MDS_INODELOCK_XATTR, MDT_CROSS_LOCK);
1937 GOTO(out_put_old, rc);
1939 mdt_enoent_version_save(info, 3);
1942 /* step 5: rename it */
1943 mdt_reint_init_ma(info, ma);
1945 mdt_fail_write(info->mti_env, info->mti_mdt->mdt_bottom,
1946 OBD_FAIL_MDS_REINT_RENAME_WRITE);
1949 mutex_lock(&mnew->mot_lov_mutex);
1951 rc = mdo_rename(info->mti_env, mdt_object_child(msrcdir),
1952 mdt_object_child(mtgtdir), old_fid, &rr->rr_name,
1953 mnew != NULL ? mdt_object_child(mnew) : NULL,
1954 &rr->rr_tgt_name, ma);
1957 mutex_unlock(&mnew->mot_lov_mutex);
1959 /* handle last link of tgt object */
1961 mdt_counter_incr(req, LPROC_MDT_RENAME);
1963 mdt_handle_last_unlink(info, mnew, ma);
1965 mdt_rename_counter_tally(info, info->mti_mdt, req,
1971 mdt_object_unlock(info, mnew, lh_newp, rc);
1973 mdt_object_unlock(info, mold, lh_oldp, rc);
1976 mdt_object_put(info->mti_env, mnew);
1978 mdt_object_put(info->mti_env, mold);
1980 mdt_object_unlock_put(info, mtgtdir, lh_tgtdirp, rc);
1981 mdt_object_unlock_put(info, msrcdir, lh_srcdirp, rc);
1985 static int mdt_reint_rename_or_migrate(struct mdt_thread_info *info,
1986 struct mdt_lock_handle *lhc,
1987 enum mdt_rename_lock rename_lock)
1989 struct mdt_reint_record *rr = &info->mti_rr;
1990 struct ptlrpc_request *req = mdt_info_req(info);
1991 struct lustre_handle rename_lh = { 0 };
1995 if (info->mti_dlm_req)
1996 ldlm_request_cancel(req, info->mti_dlm_req, 0, LATF_SKIP);
1998 if (!fid_is_md_operative(rr->rr_fid1) ||
1999 !fid_is_md_operative(rr->rr_fid2))
2002 rc = mdt_rename_lock(info, &rename_lh, rename_lock);
2004 CERROR("%s: can't lock FS for rename: rc = %d\n",
2005 mdt_obd_name(info->mti_mdt), rc);
2009 if (rename_lock == MRL_RENAME)
2010 rc = mdt_reint_rename_internal(info, lhc);
2012 rc = mdt_reint_migrate_internal(info, lhc);
2014 if (lustre_handle_is_used(&rename_lh))
2015 mdt_rename_unlock(&rename_lh);
2020 static int mdt_reint_rename(struct mdt_thread_info *info,
2021 struct mdt_lock_handle *lhc)
2023 return mdt_reint_rename_or_migrate(info, lhc, MRL_RENAME);
2026 static int mdt_reint_migrate(struct mdt_thread_info *info,
2027 struct mdt_lock_handle *lhc)
2029 return mdt_reint_rename_or_migrate(info, lhc, MRL_MIGRATE);
2032 struct mdt_reinter {
2033 int (*mr_handler)(struct mdt_thread_info *, struct mdt_lock_handle *);
2034 enum lprocfs_extra_opc mr_extra_opc;
2037 static const struct mdt_reinter mdt_reinters[] = {
2039 .mr_handler = &mdt_reint_setattr,
2040 .mr_extra_opc = MDS_REINT_SETATTR,
2043 .mr_handler = &mdt_reint_create,
2044 .mr_extra_opc = MDS_REINT_CREATE,
2047 .mr_handler = &mdt_reint_link,
2048 .mr_extra_opc = MDS_REINT_LINK,
2051 .mr_handler = &mdt_reint_unlink,
2052 .mr_extra_opc = MDS_REINT_UNLINK,
2055 .mr_handler = &mdt_reint_rename,
2056 .mr_extra_opc = MDS_REINT_RENAME,
2059 .mr_handler = &mdt_reint_open,
2060 .mr_extra_opc = MDS_REINT_OPEN,
2062 [REINT_SETXATTR] = {
2063 .mr_handler = &mdt_reint_setxattr,
2064 .mr_extra_opc = MDS_REINT_SETXATTR,
2067 .mr_handler = &mdt_reint_unlink,
2068 .mr_extra_opc = MDS_REINT_UNLINK,
2071 .mr_handler = &mdt_reint_migrate,
2072 .mr_extra_opc = MDS_REINT_RENAME,
2076 int mdt_reint_rec(struct mdt_thread_info *info,
2077 struct mdt_lock_handle *lhc)
2079 const struct mdt_reinter *mr;
2083 if (!(info->mti_rr.rr_opcode < ARRAY_SIZE(mdt_reinters)))
2086 mr = &mdt_reinters[info->mti_rr.rr_opcode];
2087 if (mr->mr_handler == NULL)
2090 rc = (*mr->mr_handler)(info, lhc);
2092 lprocfs_counter_incr(ptlrpc_req2svc(mdt_info_req(info))->srv_stats,
2093 PTLRPC_LAST_CNTR + mr->mr_extra_opc);