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 static int mdt_lookup_version_check(struct mdt_thread_info *info,
224 struct mdt_object *p,
225 const struct lu_name *lname,
226 struct lu_fid *fid, int idx)
230 rc = mdo_lookup(info->mti_env, mdt_object_child(p), lname, fid,
232 /* Check version only during replay */
233 if (!req_is_replay(mdt_info_req(info)))
236 info->mti_ver[idx] = ENOENT_VERSION;
238 struct mdt_object *child;
239 child = mdt_object_find(info->mti_env, info->mti_mdt, fid);
240 if (likely(!IS_ERR(child))) {
241 mdt_obj_version_get(info, child, &info->mti_ver[idx]);
242 mdt_object_put(info->mti_env, child);
245 vbrc = mdt_version_check(mdt_info_req(info), info->mti_ver[idx], idx);
246 return vbrc ? vbrc : rc;
251 * mdt_remote_permission: Check whether the remote operation is permitted,
253 * Before we implement async cross-MDT updates (DNE phase 2). There are a few
256 * 1.Only sysadmin can create remote directory and striped directory and
257 * migrate directory now, unless
258 * lctl set_param mdt.*.enable_remote_dir_gid=allow_gid.
259 * 2.Remote directory can only be created on MDT0, unless
260 * lctl set_param mdt.*.enable_remote_dir = 1
261 * 3.Only new clients can access remote dir( >= 2.4) and striped dir(>= 2.6),
262 * old client will return -ENOTSUPP.
264 * XXX these check are only needed for remote synchronization, once async
265 * update is supported, these check will be removed.
267 * param[in]info: execution environment.
268 * param[in]parent: the directory of this operation.
269 * param[in]child: the child of this operation.
271 * retval = 0 remote operation is allowed.
272 * < 0 remote operation is denied.
274 static int mdt_remote_permission(struct mdt_thread_info *info,
275 struct mdt_object *parent,
276 struct mdt_object *child)
278 struct mdt_device *mdt = info->mti_mdt;
279 struct lu_ucred *uc = mdt_ucred(info);
280 struct md_op_spec *spec = &info->mti_spec;
281 struct lu_attr *attr = &info->mti_attr.ma_attr;
282 struct obd_export *exp = mdt_info_req(info)->rq_export;
284 /* Only check create remote directory, striped directory and
286 if (mdt_object_remote(parent) == 0 && mdt_object_remote(child) == 0 &&
287 !(S_ISDIR(attr->la_mode) && spec->u.sp_ea.eadata != NULL &&
288 spec->u.sp_ea.eadatalen != 0) &&
289 info->mti_rr.rr_opcode != REINT_MIGRATE)
292 if (!md_capable(uc, CFS_CAP_SYS_ADMIN)) {
293 if (uc->uc_gid != mdt->mdt_enable_remote_dir_gid &&
294 mdt->mdt_enable_remote_dir_gid != -1)
298 if (mdt->mdt_enable_remote_dir == 0) {
299 struct seq_server_site *ss = mdt_seq_site(mdt);
300 struct lu_seq_range range = { 0 };
303 fld_range_set_type(&range, LU_SEQ_RANGE_MDT);
304 rc = fld_server_lookup(info->mti_env, ss->ss_server_fld,
305 fid_seq(mdt_object_fid(parent)), &range);
309 if (range.lsr_index != 0)
313 if (!mdt_is_dne_client(exp))
316 if (S_ISDIR(attr->la_mode) && spec->u.sp_ea.eadata != NULL &&
317 spec->u.sp_ea.eadatalen != 0) {
318 const struct lmv_user_md *lum = spec->u.sp_ea.eadata;
320 if (le32_to_cpu(lum->lum_stripe_count) > 1 &&
321 !mdt_is_striped_client(exp))
329 * VBR: we save three versions in reply:
330 * 0 - parent. Check that parent version is the same during replay.
331 * 1 - name. Version of 'name' if file exists with the same name or
332 * ENOENT_VERSION, it is needed because file may appear due to missed replays.
333 * 2 - child. Version of child by FID. Must be ENOENT. It is mostly sanity
336 static int mdt_md_create(struct mdt_thread_info *info)
338 struct mdt_device *mdt = info->mti_mdt;
339 struct mdt_object *parent;
340 struct mdt_object *child;
341 struct mdt_lock_handle *lh;
342 struct mdt_body *repbody;
343 struct md_attr *ma = &info->mti_attr;
344 struct mdt_reint_record *rr = &info->mti_rr;
348 DEBUG_REQ(D_INODE, mdt_info_req(info), "Create ("DNAME"->"DFID") "
350 PNAME(&rr->rr_name), PFID(rr->rr_fid2), PFID(rr->rr_fid1));
352 if (!fid_is_md_operative(rr->rr_fid1))
355 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
357 parent = mdt_object_find(info->mti_env, info->mti_mdt, rr->rr_fid1);
359 RETURN(PTR_ERR(parent));
361 if (!mdt_object_exists(parent))
362 GOTO(put_parent, rc = -ENOENT);
364 lh = &info->mti_lh[MDT_LH_PARENT];
365 mdt_lock_pdo_init(lh, LCK_PW, &rr->rr_name);
366 rc = mdt_object_lock(info, parent, lh, MDS_INODELOCK_UPDATE,
369 GOTO(put_parent, rc);
371 if (!mdt_object_remote(parent)) {
372 rc = mdt_version_get_check_save(info, parent, 0);
374 GOTO(unlock_parent, rc);
378 * Check child name version during replay.
379 * During create replay a file may exist with same name.
381 rc = mdt_lookup_version_check(info, parent, &rr->rr_name,
382 &info->mti_tmp_fid1, 1);
384 GOTO(unlock_parent, rc = -EEXIST);
386 /* -ENOENT is expected here */
388 GOTO(unlock_parent, rc);
390 /* save version of file name for replay, it must be ENOENT here */
391 mdt_enoent_version_save(info, 1);
393 child = mdt_object_new(info->mti_env, mdt, rr->rr_fid2);
394 if (likely(!IS_ERR(child))) {
395 struct md_object *next = mdt_object_child(parent);
397 rc = mdt_remote_permission(info, parent, child);
399 GOTO(out_put_child, rc);
401 ma->ma_need = MA_INODE;
403 /* capa for cross-ref will be stored here */
404 ma->ma_capa = req_capsule_server_get(info->mti_pill,
406 LASSERT(ma->ma_capa);
408 mdt_fail_write(info->mti_env, info->mti_mdt->mdt_bottom,
409 OBD_FAIL_MDS_REINT_CREATE_WRITE);
411 /* Version of child will be updated on disk. */
412 tgt_vbr_obj_set(info->mti_env, mdt_obj2dt(child));
413 rc = mdt_version_get_check_save(info, child, 2);
415 GOTO(out_put_child, rc);
417 /* Let lower layer know current lock mode. */
418 info->mti_spec.sp_cr_mode =
419 mdt_dlm_mode2mdl_mode(lh->mlh_pdo_mode);
422 * Do not perform lookup sanity check. We know that name does
425 info->mti_spec.sp_cr_lookup = 0;
426 info->mti_spec.sp_feat = &dt_directory_features;
428 rc = mdo_create(info->mti_env, next, &rr->rr_name,
429 mdt_object_child(child), &info->mti_spec, ma);
431 rc = mdt_attr_get_complex(info, child, ma);
434 /* Return fid & attr to client. */
435 if (ma->ma_valid & MA_INODE)
436 mdt_pack_attr2body(info, repbody, &ma->ma_attr,
437 mdt_object_fid(child));
440 mdt_create_pack_capa(info, rc, child, repbody);
441 mdt_object_put(info->mti_env, child);
444 mdt_create_pack_capa(info, rc, NULL, repbody);
447 mdt_object_unlock(info, parent, lh, rc);
449 mdt_object_put(info->mti_env, parent);
453 static int mdt_unlock_slaves(struct mdt_thread_info *mti,
454 struct mdt_object *obj, __u64 ibits,
455 struct mdt_lock_handle *s0_lh,
456 struct mdt_object *s0_obj,
457 struct ldlm_enqueue_info *einfo)
459 ldlm_policy_data_t *policy = &mti->mti_policy;
463 if (!S_ISDIR(obj->mot_header.loh_attr))
466 /* Unlock stripe 0 */
467 if (s0_lh != NULL && lustre_handle_is_used(&s0_lh->mlh_reg_lh)) {
468 LASSERT(s0_obj != NULL);
469 mdt_object_unlock_put(mti, s0_obj, s0_lh, 1);
472 memset(policy, 0, sizeof(*policy));
473 policy->l_inodebits.bits = ibits;
475 rc = mo_object_unlock(mti->mti_env, mdt_object_child(obj), einfo,
481 * Lock slave stripes if necessary, the lock handles of slave stripes
482 * will be stored in einfo->ei_cbdata.
484 static int mdt_lock_slaves(struct mdt_thread_info *mti, struct mdt_object *obj,
485 ldlm_mode_t mode, __u64 ibits,
486 struct mdt_lock_handle *s0_lh,
487 struct mdt_object **s0_objp,
488 struct ldlm_enqueue_info *einfo)
490 ldlm_policy_data_t *policy = &mti->mti_policy;
491 struct lu_buf *buf = &mti->mti_buf;
492 struct lmv_mds_md_v1 *lmv;
493 struct lu_fid *fid = &mti->mti_tmp_fid1;
497 if (!S_ISDIR(obj->mot_header.loh_attr))
500 buf->lb_buf = mti->mti_xattr_buf;
501 buf->lb_len = sizeof(mti->mti_xattr_buf);
502 rc = mo_xattr_get(mti->mti_env, mdt_object_child(obj), buf,
505 rc = mdt_big_xattr_get(mti, obj, XATTR_NAME_LMV);
507 buf->lb_buf = mti->mti_big_lmm;
508 buf->lb_len = mti->mti_big_lmmsize;
512 if (rc == -ENODATA || rc == -ENOENT)
519 if (le32_to_cpu(lmv->lmv_magic) != LMV_MAGIC_V1)
522 /* Sigh, 0_stripe and master object are different
523 * object, though they are in the same MDT, to avoid
524 * adding osd_object_lock here, so we will enqueue the
525 * stripe0 lock in MDT0 for now */
526 fid_le_to_cpu(fid, &lmv->lmv_stripe_fids[0]);
527 *s0_objp = mdt_object_find_lock(mti, fid, s0_lh, ibits);
528 if (IS_ERR(*s0_objp))
529 RETURN(PTR_ERR(*s0_objp));
531 memset(einfo, 0, sizeof(*einfo));
532 einfo->ei_type = LDLM_IBITS;
533 einfo->ei_mode = mode;
534 einfo->ei_cb_bl = mdt_remote_blocking_ast;
535 einfo->ei_cb_cp = ldlm_completion_ast;
536 einfo->ei_enq_slave = 1;
537 memset(policy, 0, sizeof(*policy));
538 policy->l_inodebits.bits = ibits;
540 rc = mo_object_lock(mti->mti_env, mdt_object_child(obj), NULL, einfo,
545 static int mdt_attr_set(struct mdt_thread_info *info, struct mdt_object *mo,
548 struct mdt_lock_handle *lh;
549 int do_vbr = ma->ma_attr.la_valid & (LA_MODE|LA_UID|LA_GID|LA_FLAGS);
550 __u64 lockpart = MDS_INODELOCK_UPDATE;
551 struct ldlm_enqueue_info *einfo = &info->mti_einfo;
552 struct mdt_lock_handle *s0_lh;
553 struct mdt_object *s0_obj = NULL;
557 lh = &info->mti_lh[MDT_LH_PARENT];
558 mdt_lock_reg_init(lh, LCK_PW);
560 /* Even though the new MDT will grant PERM lock to the old
561 * client, but the old client will almost ignore that during
562 * So it needs to revoke both LOOKUP and PERM lock here, so
563 * both new and old client can cancel the dcache */
564 if (ma->ma_attr.la_valid & (LA_MODE|LA_UID|LA_GID))
565 lockpart |= MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM;
567 rc = mdt_object_lock(info, mo, lh, lockpart, MDT_LOCAL_LOCK);
571 s0_lh = &info->mti_lh[MDT_LH_LOCAL];
572 mdt_lock_reg_init(s0_lh, LCK_PW);
573 rc = mdt_lock_slaves(info, mo, LCK_PW, lockpart, s0_lh, &s0_obj, einfo);
575 GOTO(out_unlock, rc);
577 /* all attrs are packed into mti_attr in unpack_setattr */
578 mdt_fail_write(info->mti_env, info->mti_mdt->mdt_bottom,
579 OBD_FAIL_MDS_REINT_SETATTR_WRITE);
581 /* This is only for set ctime when rename's source is on remote MDS. */
582 if (unlikely(ma->ma_attr.la_valid == LA_CTIME))
583 ma->ma_attr_flags |= MDS_VTX_BYPASS;
585 /* VBR: update version if attr changed are important for recovery */
587 /* update on-disk version of changed object */
588 tgt_vbr_obj_set(info->mti_env, mdt_obj2dt(mo));
589 rc = mdt_version_get_check_save(info, mo, 0);
591 GOTO(out_unlock, rc);
594 /* Ensure constant striping during chown(). See LU-2789. */
595 if (ma->ma_attr.la_valid & (LA_UID|LA_GID))
596 mutex_lock(&mo->mot_lov_mutex);
598 /* all attrs are packed into mti_attr in unpack_setattr */
599 rc = mo_attr_set(info->mti_env, mdt_object_child(mo), ma);
601 if (ma->ma_attr.la_valid & (LA_UID|LA_GID))
602 mutex_unlock(&mo->mot_lov_mutex);
605 GOTO(out_unlock, rc);
609 mdt_unlock_slaves(info, mo, lockpart, s0_lh, s0_obj, einfo);
610 mdt_object_unlock(info, mo, lh, rc);
615 * Check HSM flags and add HS_DIRTY flag if relevant.
617 * A file could be set dirty only if it has a copy in the backend (HS_EXISTS)
618 * and is not RELEASED.
620 int mdt_add_dirty_flag(struct mdt_thread_info *info, struct mdt_object *mo,
626 /* If the file was modified, add the dirty flag */
627 ma->ma_need = MA_HSM;
628 rc = mdt_attr_get_complex(info, mo, ma);
630 CERROR("file attribute read error for "DFID": %d.\n",
631 PFID(mdt_object_fid(mo)), rc);
635 /* If an up2date copy exists in the backend, add dirty flag */
636 if ((ma->ma_valid & MA_HSM) && (ma->ma_hsm.mh_flags & HS_EXISTS)
637 && !(ma->ma_hsm.mh_flags & (HS_DIRTY|HS_RELEASED))) {
638 struct mdt_lock_handle *lh = &info->mti_lh[MDT_LH_CHILD];
640 ma->ma_hsm.mh_flags |= HS_DIRTY;
642 mdt_lock_reg_init(lh, LCK_PW);
643 rc = mdt_object_lock(info, mo, lh, MDS_INODELOCK_XATTR,
648 rc = mdt_hsm_attr_set(info, mo, &ma->ma_hsm);
650 CERROR("file attribute change error for "DFID": %d\n",
651 PFID(mdt_object_fid(mo)), rc);
652 mdt_object_unlock(info, mo, lh, rc);
658 static int mdt_reint_setattr(struct mdt_thread_info *info,
659 struct mdt_lock_handle *lhc)
661 struct md_attr *ma = &info->mti_attr;
662 struct mdt_reint_record *rr = &info->mti_rr;
663 struct ptlrpc_request *req = mdt_info_req(info);
664 struct mdt_export_data *med = &req->rq_export->exp_mdt_data;
665 struct mdt_file_data *mfd;
666 struct mdt_object *mo;
667 struct mdt_body *repbody;
671 DEBUG_REQ(D_INODE, req, "setattr "DFID" %x", PFID(rr->rr_fid1),
672 (unsigned int)ma->ma_attr.la_valid);
674 if (info->mti_dlm_req)
675 ldlm_request_cancel(req, info->mti_dlm_req, 0, LATF_SKIP);
677 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
678 mo = mdt_object_find(info->mti_env, info->mti_mdt, rr->rr_fid1);
680 GOTO(out, rc = PTR_ERR(mo));
682 if (!mdt_object_exists(mo))
683 GOTO(out_put, rc = -ENOENT);
685 if (mdt_object_remote(mo))
686 GOTO(out_put, rc = -EREMOTE);
688 /* start a log jounal handle if needed */
689 if (!(mdt_conn_flags(info) & OBD_CONNECT_SOM)) {
690 if ((ma->ma_attr.la_valid & LA_SIZE) ||
691 (rr->rr_flags & MRF_OPEN_TRUNC)) {
692 /* Check write access for the O_TRUNC case */
693 if (mdt_write_read(mo) < 0)
694 GOTO(out_put, rc = -ETXTBSY);
696 } else if (info->mti_ioepoch &&
697 (info->mti_ioepoch->flags & MF_EPOCH_OPEN)) {
698 /* Truncate case. IOEpoch is opened. */
699 rc = mdt_write_get(mo);
703 mfd = mdt_mfd_new(med);
706 GOTO(out_put, rc = -ENOMEM);
709 mdt_ioepoch_open(info, mo, 0);
710 repbody->mbo_ioepoch = mo->mot_ioepoch;
712 mdt_object_get(info->mti_env, mo);
713 mdt_mfd_set_mode(mfd, MDS_FMODE_TRUNC);
714 mfd->mfd_object = mo;
715 mfd->mfd_xid = req->rq_xid;
717 spin_lock(&med->med_open_lock);
718 list_add(&mfd->mfd_list, &med->med_open_head);
719 spin_unlock(&med->med_open_lock);
720 repbody->mbo_handle.cookie = mfd->mfd_handle.h_cookie;
723 som_au = info->mti_ioepoch && info->mti_ioepoch->flags & MF_SOM_CHANGE;
725 /* SOM Attribute update case. Find the proper mfd and update
726 * SOM attributes on the proper object. */
727 if (!(mdt_conn_flags(info) & OBD_CONNECT_SOM))
728 GOTO(out_put, rc = -EPROTO);
730 spin_lock(&med->med_open_lock);
731 mfd = mdt_handle2mfd(med, &info->mti_ioepoch->handle,
734 spin_unlock(&med->med_open_lock);
735 CDEBUG(D_INODE, "no handle for file close: "
736 "fid = "DFID": cookie = "LPX64"\n",
737 PFID(info->mti_rr.rr_fid1),
738 info->mti_ioepoch->handle.cookie);
739 GOTO(out_put, rc = -ESTALE);
742 if (mfd->mfd_mode != MDS_FMODE_SOM ||
743 (info->mti_ioepoch->flags & MF_EPOCH_CLOSE))
744 GOTO(out_put, rc = -EPROTO);
746 class_handle_unhash(&mfd->mfd_handle);
747 list_del_init(&mfd->mfd_list);
748 spin_unlock(&med->med_open_lock);
750 mdt_mfd_close(info, mfd);
751 } else if ((ma->ma_valid & MA_INODE) && ma->ma_attr.la_valid) {
752 if (ma->ma_valid & MA_LOV)
753 GOTO(out_put, rc = -EPROTO);
755 rc = mdt_attr_set(info, mo, ma);
758 } else if ((ma->ma_valid & MA_LOV) && (ma->ma_valid & MA_INODE)) {
759 struct lu_buf *buf = &info->mti_buf;
761 if (ma->ma_attr.la_valid != 0)
762 GOTO(out_put, rc = -EPROTO);
764 buf->lb_buf = ma->ma_lmm;
765 buf->lb_len = ma->ma_lmm_size;
766 rc = mo_xattr_set(info->mti_env, mdt_object_child(mo),
767 buf, XATTR_NAME_LOV, 0);
770 } else if ((ma->ma_valid & MA_LMV) && (ma->ma_valid & MA_INODE)) {
771 struct lu_buf *buf = &info->mti_buf;
773 if (ma->ma_attr.la_valid != 0)
774 GOTO(out_put, rc = -EPROTO);
776 buf->lb_buf = ma->ma_lmv;
777 buf->lb_len = ma->ma_lmv_size;
778 rc = mo_xattr_set(info->mti_env, mdt_object_child(mo),
779 buf, XATTR_NAME_DEFAULT_LMV, 0);
783 GOTO(out_put, rc = -EPROTO);
786 /* If file data is modified, add the dirty flag */
787 if (ma->ma_attr_flags & MDS_DATA_MODIFIED)
788 rc = mdt_add_dirty_flag(info, mo, ma);
790 ma->ma_need = MA_INODE;
792 rc = mdt_attr_get_complex(info, mo, ma);
796 mdt_pack_attr2body(info, repbody, &ma->ma_attr, mdt_object_fid(mo));
798 if (info->mti_mdt->mdt_lut.lut_oss_capa &&
799 exp_connect_flags(info->mti_exp) & OBD_CONNECT_OSS_CAPA &&
800 S_ISREG(lu_object_attr(&mo->mot_obj)) &&
801 (ma->ma_attr.la_valid & LA_SIZE) && !som_au) {
802 struct lustre_capa *capa;
804 capa = req_capsule_server_get(info->mti_pill, &RMF_CAPA2);
806 capa->lc_opc = CAPA_OPC_OSS_DEFAULT | CAPA_OPC_OSS_TRUNC;
807 rc = mo_capa_get(info->mti_env, mdt_object_child(mo), capa, 0);
810 repbody->mbo_valid |= OBD_MD_FLOSSCAPA;
815 mdt_object_put(info->mti_env, mo);
818 mdt_counter_incr(req, LPROC_MDT_SETATTR);
820 mdt_client_compatibility(info);
821 rc2 = mdt_fix_reply(info);
827 static int mdt_reint_create(struct mdt_thread_info *info,
828 struct mdt_lock_handle *lhc)
830 struct ptlrpc_request *req = mdt_info_req(info);
834 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_REINT_CREATE))
835 RETURN(err_serious(-ESTALE));
837 if (info->mti_dlm_req)
838 ldlm_request_cancel(mdt_info_req(info),
839 info->mti_dlm_req, 0, LATF_SKIP);
841 if (!lu_name_is_valid(&info->mti_rr.rr_name))
844 switch (info->mti_attr.ma_attr.la_mode & S_IFMT) {
846 mdt_counter_incr(req, LPROC_MDT_MKDIR);
854 /* Special file should stay on the same node as parent. */
855 mdt_counter_incr(req, LPROC_MDT_MKNOD);
858 CERROR("%s: Unsupported mode %o\n",
859 mdt_obd_name(info->mti_mdt),
860 info->mti_attr.ma_attr.la_mode);
861 RETURN(err_serious(-EOPNOTSUPP));
864 rc = mdt_md_create(info);
869 * VBR: save parent version in reply and child version getting by its name.
870 * Version of child is getting and checking during its lookup. If
872 static int mdt_reint_unlink(struct mdt_thread_info *info,
873 struct mdt_lock_handle *lhc)
875 struct mdt_reint_record *rr = &info->mti_rr;
876 struct ptlrpc_request *req = mdt_info_req(info);
877 struct md_attr *ma = &info->mti_attr;
878 struct lu_fid *child_fid = &info->mti_tmp_fid1;
879 struct mdt_object *mp;
880 struct mdt_object *mc;
881 struct mdt_lock_handle *parent_lh;
882 struct mdt_lock_handle *child_lh;
883 struct ldlm_enqueue_info *einfo = &info->mti_einfo;
884 struct mdt_lock_handle *s0_lh = NULL;
885 struct mdt_object *s0_obj = NULL;
890 DEBUG_REQ(D_INODE, req, "unlink "DFID"/"DNAME"", PFID(rr->rr_fid1),
891 PNAME(&rr->rr_name));
893 if (info->mti_dlm_req)
894 ldlm_request_cancel(req, info->mti_dlm_req, 0, LATF_SKIP);
896 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_REINT_UNLINK))
897 RETURN(err_serious(-ENOENT));
899 if (!fid_is_md_operative(rr->rr_fid1))
903 * step 1: Found the parent.
905 mp = mdt_object_find(info->mti_env, info->mti_mdt, rr->rr_fid1);
911 parent_lh = &info->mti_lh[MDT_LH_PARENT];
912 mdt_lock_pdo_init(parent_lh, LCK_PW, &rr->rr_name);
913 rc = mdt_object_lock(info, mp, parent_lh, MDS_INODELOCK_UPDATE,
916 GOTO(put_parent, rc);
918 if (!mdt_object_remote(mp)) {
919 rc = mdt_version_get_check_save(info, mp, 0);
921 GOTO(unlock_parent, rc);
924 /* step 2: find & lock the child */
925 /* lookup child object along with version checking */
927 rc = mdt_lookup_version_check(info, mp, &rr->rr_name, child_fid, 1);
929 /* Name might not be able to find during resend of
930 * remote unlink, considering following case.
931 * dir_A is a remote directory, the name entry of
932 * dir_A is on MDT0, the directory is on MDT1,
934 * 1. client sends unlink req to MDT1.
935 * 2. MDT1 sends name delete update to MDT0.
936 * 3. name entry is being deleted in MDT0 synchronously.
937 * 4. MDT1 is restarted.
938 * 5. client resends unlink req to MDT1. So it can not
939 * find the name entry on MDT0 anymore.
940 * In this case, MDT1 only needs to destory the local
943 if (mdt_object_remote(mp) && rc == -ENOENT &&
944 !fid_is_zero(rr->rr_fid2) &&
945 lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT) {
947 *child_fid = *rr->rr_fid2;
949 GOTO(unlock_parent, rc);
953 if (!fid_is_md_operative(child_fid))
954 GOTO(unlock_parent, rc = -EPERM);
956 /* We will lock the child regardless it is local or remote. No harm. */
957 mc = mdt_object_find(info->mti_env, info->mti_mdt, child_fid);
959 GOTO(unlock_parent, rc = PTR_ERR(mc));
961 child_lh = &info->mti_lh[MDT_LH_CHILD];
962 mdt_lock_reg_init(child_lh, LCK_EX);
963 if (mdt_object_remote(mc)) {
964 struct mdt_body *repbody;
966 if (!fid_is_zero(rr->rr_fid2)) {
967 CDEBUG(D_INFO, "%s: name "DNAME" cannot find "DFID"\n",
968 mdt_obd_name(info->mti_mdt),
969 PNAME(&rr->rr_name), PFID(mdt_object_fid(mc)));
970 GOTO(put_child, rc = -ENOENT);
972 CDEBUG(D_INFO, "%s: name "DNAME": "DFID" is on another MDT\n",
973 mdt_obd_name(info->mti_mdt),
974 PNAME(&rr->rr_name), PFID(mdt_object_fid(mc)));
976 if (!mdt_is_dne_client(req->rq_export))
977 /* Return -ENOTSUPP for old client */
978 GOTO(put_child, rc = -ENOTSUPP);
980 if (info->mti_spec.sp_rm_entry) {
981 struct lu_ucred *uc = mdt_ucred(info);
983 if (!md_capable(uc, CFS_CAP_SYS_ADMIN)) {
984 CERROR("%s: unlink remote entry is only "
985 "permitted for administrator: rc = %d\n",
986 mdt_obd_name(info->mti_mdt),
988 GOTO(put_child, rc = -EPERM);
991 ma->ma_need = MA_INODE;
993 mdt_set_capainfo(info, 1, child_fid, BYPASS_CAPA);
994 rc = mdo_unlink(info->mti_env, mdt_object_child(mp),
995 NULL, &rr->rr_name, ma, no_name);
998 /* Revoke the LOOKUP lock of the remote object granted by
999 * this MDT. Since the unlink will happen on another MDT,
1000 * it will release the LOOKUP lock right away. Then What
1001 * would happen if another client try to grab the LOOKUP
1002 * lock at the same time with unlink XXX */
1003 mdt_object_lock(info, mc, child_lh, MDS_INODELOCK_LOOKUP,
1005 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1006 LASSERT(repbody != NULL);
1007 repbody->mbo_fid1 = *mdt_object_fid(mc);
1008 repbody->mbo_valid |= (OBD_MD_FLID | OBD_MD_MDS);
1009 GOTO(unlock_child, rc = -EREMOTE);
1010 } else if (info->mti_spec.sp_rm_entry) {
1012 CDEBUG(D_INFO, "%s: no rm_entry on local dir '"DNAME"': "
1014 mdt_obd_name(info->mti_mdt), PNAME(&rr->rr_name), rc);
1015 GOTO(put_child, rc);
1018 /* We used to acquire MDS_INODELOCK_FULL here but we can't do
1019 * this now because a running HSM restore on the child (unlink
1020 * victim) will hold the layout lock. See LU-4002. */
1021 rc = mdt_object_lock(info, mc, child_lh,
1022 MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE,
1025 GOTO(put_child, rc);
1027 * Now we can only make sure we need MA_INODE, in mdd layer, will check
1028 * whether need MA_LOV and MA_COOKIE.
1030 ma->ma_need = MA_INODE;
1033 s0_lh = &info->mti_lh[MDT_LH_LOCAL];
1034 mdt_lock_reg_init(s0_lh, LCK_EX);
1035 rc = mdt_lock_slaves(info, mc, LCK_EX, MDS_INODELOCK_UPDATE, s0_lh,
1038 GOTO(unlock_child, rc);
1040 mdt_fail_write(info->mti_env, info->mti_mdt->mdt_bottom,
1041 OBD_FAIL_MDS_REINT_UNLINK_WRITE);
1042 /* save version when object is locked */
1043 mdt_version_get_save(info, mc, 1);
1045 mdt_set_capainfo(info, 1, child_fid, BYPASS_CAPA);
1047 mutex_lock(&mc->mot_lov_mutex);
1049 rc = mdo_unlink(info->mti_env, mdt_object_child(mp),
1050 mdt_object_child(mc), &rr->rr_name, ma, no_name);
1052 mutex_unlock(&mc->mot_lov_mutex);
1054 if (rc == 0 && !lu_object_is_dying(&mc->mot_header))
1055 rc = mdt_attr_get_complex(info, mc, ma);
1057 mdt_handle_last_unlink(info, mc, ma);
1059 if (ma->ma_valid & MA_INODE) {
1060 switch (ma->ma_attr.la_mode & S_IFMT) {
1062 mdt_counter_incr(req, LPROC_MDT_RMDIR);
1070 mdt_counter_incr(req, LPROC_MDT_UNLINK);
1073 LASSERTF(0, "bad file type %o unlinking\n",
1074 ma->ma_attr.la_mode);
1081 mdt_unlock_slaves(info, mc, MDS_INODELOCK_UPDATE, s0_lh, s0_obj, einfo);
1082 mdt_object_unlock(info, mc, child_lh, rc);
1084 mdt_object_put(info->mti_env, mc);
1086 mdt_object_unlock(info, mp, parent_lh, rc);
1088 mdt_object_put(info->mti_env, mp);
1094 * VBR: save versions in reply: 0 - parent; 1 - child by fid; 2 - target by
1097 static int mdt_reint_link(struct mdt_thread_info *info,
1098 struct mdt_lock_handle *lhc)
1100 struct mdt_reint_record *rr = &info->mti_rr;
1101 struct ptlrpc_request *req = mdt_info_req(info);
1102 struct md_attr *ma = &info->mti_attr;
1103 struct mdt_object *ms;
1104 struct mdt_object *mp;
1105 struct mdt_lock_handle *lhs;
1106 struct mdt_lock_handle *lhp;
1110 DEBUG_REQ(D_INODE, req, "link "DFID" to "DFID"/"DNAME,
1111 PFID(rr->rr_fid1), PFID(rr->rr_fid2), PNAME(&rr->rr_name));
1113 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_REINT_LINK))
1114 RETURN(err_serious(-ENOENT));
1116 if (info->mti_dlm_req)
1117 ldlm_request_cancel(req, info->mti_dlm_req, 0, LATF_SKIP);
1119 /* Invalid case so return error immediately instead of
1121 if (lu_fid_eq(rr->rr_fid1, rr->rr_fid2))
1124 if (!fid_is_md_operative(rr->rr_fid1) ||
1125 !fid_is_md_operative(rr->rr_fid2))
1128 /* step 1: find & lock the target parent dir */
1129 lhp = &info->mti_lh[MDT_LH_PARENT];
1130 mdt_lock_pdo_init(lhp, LCK_PW, &rr->rr_name);
1131 mp = mdt_object_find_lock(info, rr->rr_fid2, lhp,
1132 MDS_INODELOCK_UPDATE);
1134 RETURN(PTR_ERR(mp));
1136 rc = mdt_version_get_check_save(info, mp, 0);
1138 GOTO(out_unlock_parent, rc);
1140 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_RENAME3, 5);
1142 /* step 2: find & lock the source */
1143 lhs = &info->mti_lh[MDT_LH_CHILD];
1144 mdt_lock_reg_init(lhs, LCK_EX);
1146 ms = mdt_object_find(info->mti_env, info->mti_mdt, rr->rr_fid1);
1148 GOTO(out_unlock_parent, rc = PTR_ERR(ms));
1150 if (!mdt_object_exists(ms)) {
1151 mdt_object_put(info->mti_env, ms);
1152 CDEBUG(D_INFO, "%s: "DFID" does not exist.\n",
1153 mdt_obd_name(info->mti_mdt), PFID(rr->rr_fid1));
1154 GOTO(out_unlock_parent, rc = -ENOENT);
1157 if (mdt_object_remote(ms)) {
1158 mdt_object_put(info->mti_env, ms);
1159 CERROR("%s: source inode "DFID" on remote MDT from "DFID"\n",
1160 mdt_obd_name(info->mti_mdt), PFID(rr->rr_fid1),
1162 GOTO(out_unlock_parent, rc = -EXDEV);
1165 rc = mdt_object_lock(info, ms, lhs, MDS_INODELOCK_UPDATE |
1166 MDS_INODELOCK_XATTR, MDT_LOCAL_LOCK);
1168 mdt_object_put(info->mti_env, ms);
1169 GOTO(out_unlock_parent, rc);
1172 /* step 3: link it */
1173 mdt_fail_write(info->mti_env, info->mti_mdt->mdt_bottom,
1174 OBD_FAIL_MDS_REINT_LINK_WRITE);
1176 tgt_vbr_obj_set(info->mti_env, mdt_obj2dt(ms));
1177 rc = mdt_version_get_check_save(info, ms, 1);
1179 GOTO(out_unlock_child, rc);
1181 /** check target version by name during replay */
1182 rc = mdt_lookup_version_check(info, mp, &rr->rr_name,
1183 &info->mti_tmp_fid1, 2);
1184 if (rc != 0 && rc != -ENOENT)
1185 GOTO(out_unlock_child, rc);
1186 /* save version of file name for replay, it must be ENOENT here */
1187 if (!req_is_replay(mdt_info_req(info))) {
1188 if (rc != -ENOENT) {
1189 CDEBUG(D_INFO, "link target "DNAME" existed!\n",
1190 PNAME(&rr->rr_name));
1191 GOTO(out_unlock_child, rc = -EEXIST);
1193 info->mti_ver[2] = ENOENT_VERSION;
1194 mdt_version_save(mdt_info_req(info), info->mti_ver[2], 2);
1197 rc = mdo_link(info->mti_env, mdt_object_child(mp),
1198 mdt_object_child(ms), &rr->rr_name, ma);
1201 mdt_counter_incr(req, LPROC_MDT_LINK);
1205 mdt_object_unlock_put(info, ms, lhs, rc);
1207 mdt_object_unlock_put(info, mp, lhp, rc);
1211 * lock the part of the directory according to the hash of the name
1212 * (lh->mlh_pdo_hash) in parallel directory lock.
1214 static int mdt_pdir_hash_lock(struct mdt_thread_info *info,
1215 struct mdt_lock_handle *lh,
1216 struct mdt_object *obj, __u64 ibits)
1218 struct ldlm_res_id *res = &info->mti_res_id;
1219 struct ldlm_namespace *ns = info->mti_mdt->mdt_namespace;
1220 ldlm_policy_data_t *policy = &info->mti_policy;
1224 * Finish res_id initializing by name hash marking part of
1225 * directory which is taking modification.
1227 LASSERT(lh->mlh_pdo_hash != 0);
1228 fid_build_pdo_res_name(mdt_object_fid(obj), lh->mlh_pdo_hash, res);
1229 memset(policy, 0, sizeof(*policy));
1230 policy->l_inodebits.bits = ibits;
1232 * Use LDLM_FL_LOCAL_ONLY for this lock. We do not know yet if it is
1233 * going to be sent to client. If it is - mdt_intent_policy() path will
1234 * fix it up and turn FL_LOCAL flag off.
1236 rc = mdt_fid_lock(ns, &lh->mlh_reg_lh, lh->mlh_reg_mode, policy,
1237 res, LDLM_FL_LOCAL_ONLY | LDLM_FL_ATOMIC_CB,
1238 &info->mti_exp->exp_handle.h_cookie);
1242 enum mdt_rename_lock {
1248 * Get BFL lock for rename or migrate process, right now, it does not support
1249 * cross-MDT rename, so we only need global rename lock during migration.
1251 static int mdt_rename_lock(struct mdt_thread_info *info,
1252 struct lustre_handle *lh,
1253 enum mdt_rename_lock rename_lock)
1255 struct ldlm_namespace *ns = info->mti_mdt->mdt_namespace;
1256 ldlm_policy_data_t *policy = &info->mti_policy;
1257 struct ldlm_res_id *res_id = &info->mti_res_id;
1262 /* XXX only do global rename lock for migration */
1263 if (mdt_seq_site(info->mti_mdt)->ss_node_id != 0 &&
1264 rename_lock == MRL_MIGRATE) {
1265 struct lu_fid *fid = &info->mti_tmp_fid1;
1266 struct mdt_object *obj;
1268 /* XXX, right now, it has to use object API to
1269 * enqueue lock cross MDT, so it will enqueue
1270 * rename lock(with LUSTRE_BFL_FID) by root object */
1272 obj = mdt_object_find(info->mti_env, info->mti_mdt, fid);
1274 RETURN(PTR_ERR(obj));
1276 LASSERT(mdt_object_remote(obj));
1277 rc = mdt_remote_object_lock(info, obj,
1278 &LUSTRE_BFL_FID, lh,
1280 MDS_INODELOCK_UPDATE);
1281 mdt_object_put(info->mti_env, obj);
1283 fid_build_reg_res_name(&LUSTRE_BFL_FID, res_id);
1284 memset(policy, 0, sizeof *policy);
1285 policy->l_inodebits.bits = MDS_INODELOCK_UPDATE;
1286 flags = LDLM_FL_LOCAL_ONLY | LDLM_FL_ATOMIC_CB;
1287 rc = ldlm_cli_enqueue_local(ns, res_id, LDLM_IBITS, policy,
1288 LCK_EX, &flags, ldlm_blocking_ast,
1289 ldlm_completion_ast, NULL, NULL, 0,
1291 &info->mti_exp->exp_handle.h_cookie,
1298 static void mdt_rename_unlock(struct lustre_handle *lh)
1301 LASSERT(lustre_handle_is_used(lh));
1302 /* Cancel the single rename lock right away */
1303 ldlm_lock_decref_and_cancel(lh, LCK_EX);
1308 * This is is_subdir() variant, it is CMD if cmm forwards it to correct
1309 * target. Source should not be ancestor of target dir. May be other rename
1310 * checks can be moved here later.
1312 static int mdt_is_subdir(struct mdt_thread_info *info,
1313 struct mdt_object *dir,
1314 const struct lu_fid *fid)
1316 struct lu_fid dir_fid = dir->mot_header.loh_fid;
1320 /* If the source and target are in the same directory, they can not
1321 * be parent/child relationship, so subdir check is not needed */
1322 if (lu_fid_eq(&dir_fid, fid))
1325 if (!mdt_object_exists(dir))
1328 rc = mdo_is_subdir(info->mti_env, mdt_object_child(dir),
1331 CERROR("%s: failed subdir check in "DFID" for "DFID
1332 ": rc = %d\n", mdt_obd_name(info->mti_mdt),
1333 PFID(&dir_fid), PFID(fid), rc);
1334 /* Return EINVAL only if a parent is the @fid */
1338 /* check the found fid */
1339 if (lu_fid_eq(&dir_fid, fid))
1346 /* Update object linkEA */
1347 struct mdt_lock_list {
1348 struct mdt_object *mll_obj;
1349 struct mdt_lock_handle mll_lh;
1350 struct list_head mll_list;
1353 static void mdt_unlock_list(struct mdt_thread_info *info,
1354 struct list_head *list, int rc)
1356 struct mdt_lock_list *mll;
1357 struct mdt_lock_list *mll2;
1359 list_for_each_entry_safe(mll, mll2, list, mll_list) {
1360 mdt_object_unlock_put(info, mll->mll_obj, &mll->mll_lh, rc);
1361 list_del(&mll->mll_list);
1366 static int mdt_lock_objects_in_linkea(struct mdt_thread_info *info,
1367 struct mdt_object *obj,
1368 struct mdt_object *pobj,
1369 struct list_head *lock_list)
1371 struct lu_buf *buf = &info->mti_big_buf;
1372 struct linkea_data ldata = { NULL };
1377 if (S_ISDIR(lu_object_attr(&obj->mot_obj)))
1380 buf = lu_buf_check_and_alloc(buf, PATH_MAX);
1381 if (buf->lb_buf == NULL)
1385 rc = mdt_links_read(info, obj, &ldata);
1387 if (rc == -ENOENT || rc == -ENODATA)
1392 LASSERT(ldata.ld_leh != NULL);
1393 ldata.ld_lee = (struct link_ea_entry *)(ldata.ld_leh + 1);
1394 for (count = 0; count < ldata.ld_leh->leh_reccount; count++) {
1395 struct mdt_device *mdt = info->mti_mdt;
1396 struct mdt_object *mdt_pobj;
1397 struct mdt_lock_list *mll;
1398 struct lu_name name;
1401 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen,
1403 ldata.ld_lee = (struct link_ea_entry *)((char *)ldata.ld_lee +
1405 mdt_pobj = mdt_object_find(info->mti_env, mdt, &fid);
1406 if (IS_ERR(mdt_pobj)) {
1407 CWARN("%s: cannot find obj "DFID": rc = %ld\n",
1408 mdt_obd_name(mdt), PFID(&fid), PTR_ERR(mdt_pobj));
1412 if (!mdt_object_exists(mdt_pobj)) {
1413 CDEBUG(D_INFO, "%s: obj "DFID" does not exist\n",
1414 mdt_obd_name(mdt), PFID(&fid));
1415 mdt_object_put(info->mti_env, mdt_pobj);
1419 if (mdt_pobj == pobj) {
1420 CDEBUG(D_INFO, "%s: skipping parent obj "DFID"\n",
1421 mdt_obd_name(mdt), PFID(&fid));
1422 mdt_object_put(info->mti_env, mdt_pobj);
1428 mdt_object_put(info->mti_env, mdt_pobj);
1429 GOTO(out, rc = -ENOMEM);
1432 mdt_lock_pdo_init(&mll->mll_lh, LCK_PW, &name);
1433 rc = mdt_object_lock(info, mdt_pobj, &mll->mll_lh,
1434 MDS_INODELOCK_UPDATE,
1437 CERROR("%s: cannot lock "DFID": rc =%d\n",
1438 mdt_obd_name(mdt), PFID(&fid), rc);
1439 mdt_object_put(info->mti_env, mdt_pobj);
1444 INIT_LIST_HEAD(&mll->mll_list);
1445 mll->mll_obj = mdt_pobj;
1446 list_add_tail(&mll->mll_list, lock_list);
1450 mdt_unlock_list(info, lock_list, rc);
1454 /* migrate files from one MDT to another MDT */
1455 static int mdt_reint_migrate_internal(struct mdt_thread_info *info,
1456 struct mdt_lock_handle *lhc)
1458 struct mdt_reint_record *rr = &info->mti_rr;
1459 struct md_attr *ma = &info->mti_attr;
1460 struct mdt_object *msrcdir;
1461 struct mdt_object *mold;
1462 struct mdt_object *mnew = NULL;
1463 struct mdt_lock_handle *lh_dirp;
1464 struct mdt_lock_handle *lh_childp;
1465 struct mdt_lock_handle *lh_tgtp = NULL;
1466 struct lu_fid *old_fid = &info->mti_tmp_fid1;
1467 struct list_head lock_list;
1471 CDEBUG(D_INODE, "migrate "DFID"/"DNAME" to "DFID"\n", PFID(rr->rr_fid1),
1472 PNAME(&rr->rr_name), PFID(rr->rr_fid2));
1474 /* 1: lock the source dir. */
1475 msrcdir = mdt_object_find(info->mti_env, info->mti_mdt, rr->rr_fid1);
1476 if (IS_ERR(msrcdir)) {
1477 CERROR("%s: cannot find source dir "DFID" : rc = %d\n",
1478 mdt_obd_name(info->mti_mdt), PFID(rr->rr_fid1),
1479 (int)PTR_ERR(msrcdir));
1480 RETURN(PTR_ERR(msrcdir));
1483 lh_dirp = &info->mti_lh[MDT_LH_PARENT];
1484 mdt_lock_pdo_init(lh_dirp, LCK_PW, &rr->rr_name);
1485 rc = mdt_object_lock(info, msrcdir, lh_dirp,
1486 MDS_INODELOCK_UPDATE,
1489 GOTO(out_put_parent, rc);
1491 if (!mdt_object_remote(msrcdir)) {
1492 rc = mdt_version_get_check_save(info, msrcdir, 0);
1494 GOTO(out_unlock_parent, rc);
1497 /* 2: sanity check and find the object to be migrated. */
1499 rc = mdt_lookup_version_check(info, msrcdir, &rr->rr_name, old_fid, 2);
1501 GOTO(out_unlock_parent, rc);
1503 if (lu_fid_eq(old_fid, rr->rr_fid1) || lu_fid_eq(old_fid, rr->rr_fid2))
1504 GOTO(out_unlock_parent, rc = -EINVAL);
1506 if (!fid_is_md_operative(old_fid))
1507 GOTO(out_unlock_parent, rc = -EPERM);
1509 mold = mdt_object_find(info->mti_env, info->mti_mdt, old_fid);
1511 GOTO(out_unlock_parent, rc = PTR_ERR(mold));
1513 if (mdt_object_remote(mold)) {
1514 CERROR("%s: source "DFID" is on the remote MDT\n",
1515 mdt_obd_name(info->mti_mdt), PFID(old_fid));
1516 GOTO(out_put_child, rc = -EREMOTE);
1519 if (S_ISREG(lu_object_attr(&mold->mot_obj)) &&
1520 !mdt_object_remote(msrcdir)) {
1521 CERROR("%s: parent "DFID" is still on the same"
1522 " MDT, which should be migrated first:"
1523 " rc = %d\n", mdt_obd_name(info->mti_mdt),
1524 PFID(mdt_object_fid(msrcdir)), -EPERM);
1525 GOTO(out_put_child, rc = -EPERM);
1528 rc = mdt_remote_permission(info, msrcdir, mold);
1530 GOTO(out_put_child, rc);
1532 /* 3: iterate the linkea of the object and lock all of the objects */
1533 INIT_LIST_HEAD(&lock_list);
1534 rc = mdt_lock_objects_in_linkea(info, mold, msrcdir, &lock_list);
1536 GOTO(out_put_child, rc);
1538 /* 4: lock of the object migrated object */
1539 lh_childp = &info->mti_lh[MDT_LH_OLD];
1540 mdt_lock_reg_init(lh_childp, LCK_EX);
1541 rc = mdt_object_lock(info, mold, lh_childp,
1542 MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
1543 MDS_INODELOCK_LAYOUT, MDT_CROSS_LOCK);
1545 GOTO(out_unlock_list, rc);
1547 ma->ma_need = MA_LMV;
1549 ma->ma_lmv = (union lmv_mds_md *)info->mti_xattr_buf;
1550 ma->ma_lmv_size = sizeof(info->mti_xattr_buf);
1551 rc = mdt_stripe_get(info, mold, ma, XATTR_NAME_LMV);
1553 GOTO(out_unlock_list, rc);
1555 if ((ma->ma_valid & MA_LMV)) {
1556 struct lmv_mds_md_v1 *lmm1;
1558 lmv_le_to_cpu(ma->ma_lmv, ma->ma_lmv);
1559 lmm1 = &ma->ma_lmv->lmv_md_v1;
1560 if (!(lmm1->lmv_hash_type & LMV_HASH_FLAG_MIGRATION)) {
1561 CERROR("%s: can not migrate striped dir "DFID
1562 ": rc = %d\n", mdt_obd_name(info->mti_mdt),
1563 PFID(mdt_object_fid(mold)), -EPERM);
1564 GOTO(out_unlock_child, rc = -EPERM);
1567 if (!fid_is_sane(&lmm1->lmv_stripe_fids[1]))
1568 GOTO(out_unlock_child, rc = -EINVAL);
1570 mnew = mdt_object_find(info->mti_env, info->mti_mdt,
1571 &lmm1->lmv_stripe_fids[1]);
1573 GOTO(out_unlock_child, rc = PTR_ERR(mnew));
1575 if (!mdt_object_remote(mnew)) {
1576 CERROR("%s: "DFID" being migrated is on this MDT:"
1577 " rc = %d\n", mdt_obd_name(info->mti_mdt),
1578 PFID(rr->rr_fid2), -EPERM);
1579 GOTO(out_put_new, rc = -EPERM);
1582 lh_tgtp = &info->mti_lh[MDT_LH_CHILD];
1583 mdt_lock_reg_init(lh_tgtp, LCK_EX);
1584 rc = mdt_remote_object_lock(info, mnew,
1585 mdt_object_fid(mnew),
1586 &lh_tgtp->mlh_rreg_lh,
1587 lh_tgtp->mlh_rreg_mode,
1588 MDS_INODELOCK_UPDATE);
1591 GOTO(out_put_new, rc);
1594 mnew = mdt_object_find(info->mti_env, info->mti_mdt,
1597 GOTO(out_unlock_child, rc = PTR_ERR(mnew));
1598 if (!mdt_object_remote(mnew)) {
1599 CERROR("%s: Migration "DFID" is on this MDT:"
1600 " rc = %d\n", mdt_obd_name(info->mti_mdt),
1601 PFID(rr->rr_fid2), -EXDEV);
1602 GOTO(out_put_new, rc = -EXDEV);
1607 mdt_reint_init_ma(info, ma);
1609 mdt_fail_write(info->mti_env, info->mti_mdt->mdt_bottom,
1610 OBD_FAIL_MDS_REINT_RENAME_WRITE);
1612 rc = mdo_migrate(info->mti_env, mdt_object_child(msrcdir),
1613 mdt_object_child(mold), &rr->rr_name,
1614 mdt_object_child(mnew), ma);
1616 GOTO(out_unlock_new, rc);
1618 if (lh_tgtp != NULL)
1619 mdt_object_unlock(info, mnew, lh_tgtp, rc);
1622 mdt_object_put(info->mti_env, mnew);
1624 mdt_object_unlock(info, mold, lh_childp, rc);
1626 mdt_unlock_list(info, &lock_list, rc);
1628 mdt_object_put(info->mti_env, mold);
1630 mdt_object_unlock(info, msrcdir, lh_dirp, rc);
1632 mdt_object_put(info->mti_env, msrcdir);
1637 static struct mdt_object *mdt_object_find_check(struct mdt_thread_info *info,
1638 const struct lu_fid *fid,
1641 struct mdt_object *dir;
1645 dir = mdt_object_find(info->mti_env, info->mti_mdt, fid);
1649 /* check early, the real version will be saved after locking */
1650 rc = mdt_version_get_check(info, dir, idx);
1656 mdt_object_put(info->mti_env, dir);
1660 static int mdt_object_lock_save(struct mdt_thread_info *info,
1661 struct mdt_object *dir,
1662 struct mdt_lock_handle *lh,
1667 /* we lock the target dir if it is local */
1668 rc = mdt_object_lock(info, dir, lh, MDS_INODELOCK_UPDATE,
1673 /* get and save correct version after locking */
1674 mdt_version_get_save(info, dir, idx);
1679 static int mdt_rename_parents_lock(struct mdt_thread_info *info,
1680 struct mdt_object **srcp,
1681 struct mdt_object **tgtp)
1683 struct mdt_reint_record *rr = &info->mti_rr;
1684 const struct lu_fid *fid_src = rr->rr_fid1;
1685 const struct lu_fid *fid_tgt = rr->rr_fid2;
1686 struct mdt_lock_handle *lh_src = &info->mti_lh[MDT_LH_PARENT];
1687 struct mdt_lock_handle *lh_tgt = &info->mti_lh[MDT_LH_CHILD];
1688 struct mdt_object *src;
1689 struct mdt_object *tgt;
1694 /* find both parents. */
1695 src = mdt_object_find_check(info, fid_src, 0);
1697 RETURN(PTR_ERR(src));
1699 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_RENAME3, 5);
1701 if (lu_fid_eq(fid_src, fid_tgt)) {
1703 mdt_object_get(info->mti_env, tgt);
1705 /* Check if the @src is not a child of the @tgt, otherwise a
1706 * reverse locking must take place. */
1707 rc = mdt_is_subdir(info, src, fid_tgt);
1711 GOTO(err_src_put, rc);
1713 tgt = mdt_object_find_check(info, fid_tgt, 1);
1715 GOTO(err_src_put, rc = PTR_ERR(tgt));
1717 if (unlikely(mdt_object_remote(tgt))) {
1718 CDEBUG(D_INFO, "Source dir "DFID" target dir "DFID
1719 "on different MDTs\n", PFID(fid_src),
1721 GOTO(err_tgt_put, rc = -EXDEV);
1725 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_RENAME4, 5);
1727 /* lock parents in the proper order. */
1729 rc = mdt_object_lock_save(info, tgt, lh_tgt, 1);
1731 GOTO(err_tgt_put, rc);
1733 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_RENAME, 5);
1735 rc = mdt_object_lock_save(info, src, lh_src, 0);
1737 rc = mdt_object_lock_save(info, src, lh_src, 0);
1739 GOTO(err_tgt_put, rc);
1741 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_RENAME, 5);
1744 rc = mdt_object_lock_save(info, tgt, lh_tgt, 1);
1745 else if (lh_src->mlh_pdo_hash != lh_tgt->mlh_pdo_hash) {
1746 rc = mdt_pdir_hash_lock(info, lh_tgt, tgt,
1747 MDS_INODELOCK_UPDATE);
1748 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_PDO_LOCK2, 10);
1752 GOTO(err_unlock, rc);
1754 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_RENAME4, 5);
1761 /* The order does not matter as the handle is checked inside,
1762 * as well as not used handle. */
1763 mdt_object_unlock(info, src, lh_src, rc);
1764 mdt_object_unlock(info, tgt, lh_tgt, rc);
1766 mdt_object_put(info->mti_env, tgt);
1768 mdt_object_put(info->mti_env, src);
1773 * VBR: rename versions in reply: 0 - src parent; 1 - tgt parent;
1774 * 2 - src child; 3 - tgt child.
1775 * Update on disk version of src child.
1778 * For DNE phase I, only these renames are allowed
1779 * mv src_p/src_c tgt_p/tgt_c
1780 * 1. src_p/src_c/tgt_p/tgt_c are in the same MDT.
1781 * 2. src_p and tgt_p are same directory, and tgt_c does not
1782 * exists. In this case, all of modification will happen
1783 * in the MDT where ithesource parent is, only one remote
1784 * update is needed, i.e. set c_time/m_time on the child.
1785 * And tgt_c will be still in the same MDT as the original
1788 static int mdt_reint_rename_internal(struct mdt_thread_info *info,
1789 struct mdt_lock_handle *lhc)
1791 struct mdt_reint_record *rr = &info->mti_rr;
1792 struct md_attr *ma = &info->mti_attr;
1793 struct ptlrpc_request *req = mdt_info_req(info);
1794 struct mdt_object *msrcdir = NULL;
1795 struct mdt_object *mtgtdir = NULL;
1796 struct mdt_object *mold;
1797 struct mdt_object *mnew = NULL;
1798 struct mdt_lock_handle *lh_srcdirp;
1799 struct mdt_lock_handle *lh_tgtdirp;
1800 struct mdt_lock_handle *lh_oldp = NULL;
1801 struct mdt_lock_handle *lh_newp = NULL;
1802 struct lu_fid *old_fid = &info->mti_tmp_fid1;
1803 struct lu_fid *new_fid = &info->mti_tmp_fid2;
1807 DEBUG_REQ(D_INODE, req, "rename "DFID"/"DNAME" to "DFID"/"DNAME,
1808 PFID(rr->rr_fid1), PNAME(&rr->rr_name),
1809 PFID(rr->rr_fid2), PNAME(&rr->rr_tgt_name));
1811 lh_srcdirp = &info->mti_lh[MDT_LH_PARENT];
1812 mdt_lock_pdo_init(lh_srcdirp, LCK_PW, &rr->rr_name);
1813 lh_tgtdirp = &info->mti_lh[MDT_LH_CHILD];
1814 mdt_lock_pdo_init(lh_tgtdirp, LCK_PW, &rr->rr_tgt_name);
1816 /* step 1&2: lock the source and target dirs. */
1817 rc = mdt_rename_parents_lock(info, &msrcdir, &mtgtdir);
1821 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_RENAME2, 5);
1823 /* step 3: find & lock the old object. */
1825 rc = mdt_lookup_version_check(info, msrcdir, &rr->rr_name, old_fid, 2);
1827 GOTO(out_unlock_parents, rc);
1829 if (lu_fid_eq(old_fid, rr->rr_fid1) || lu_fid_eq(old_fid, rr->rr_fid2))
1830 GOTO(out_unlock_parents, rc = -EINVAL);
1832 if (!fid_is_md_operative(old_fid))
1833 GOTO(out_unlock_parents, rc = -EPERM);
1835 mold = mdt_object_find(info->mti_env, info->mti_mdt, old_fid);
1837 GOTO(out_unlock_parents, rc = PTR_ERR(mold));
1839 /* Check if @mtgtdir is subdir of @mold, before locking child
1840 * to avoid reverse locking. */
1841 rc = mdt_is_subdir(info, mtgtdir, old_fid);
1843 GOTO(out_put_old, rc);
1845 tgt_vbr_obj_set(info->mti_env, mdt_obj2dt(mold));
1846 /* save version after locking */
1847 mdt_version_get_save(info, mold, 2);
1848 mdt_set_capainfo(info, 2, old_fid, BYPASS_CAPA);
1850 /* step 4: find & lock the new object. */
1851 /* new target object may not exist now */
1852 /* lookup with version checking */
1854 rc = mdt_lookup_version_check(info, mtgtdir, &rr->rr_tgt_name, new_fid,
1857 /* the new_fid should have been filled at this moment */
1858 if (lu_fid_eq(old_fid, new_fid))
1859 GOTO(out_put_old, rc);
1861 if (lu_fid_eq(new_fid, rr->rr_fid1) ||
1862 lu_fid_eq(new_fid, rr->rr_fid2))
1863 GOTO(out_put_old, rc = -EINVAL);
1865 if (!fid_is_md_operative(new_fid))
1866 GOTO(out_put_old, rc = -EPERM);
1868 if (mdt_object_remote(mold)) {
1869 CDEBUG(D_INFO, "Src child "DFID" is on another MDT\n",
1871 GOTO(out_put_old, rc = -EXDEV);
1874 mnew = mdt_object_find(info->mti_env, info->mti_mdt, new_fid);
1876 GOTO(out_put_old, rc = PTR_ERR(mnew));
1878 if (mdt_object_remote(mnew)) {
1879 CDEBUG(D_INFO, "src child "DFID" is on another MDT\n",
1881 GOTO(out_put_new, rc = -EXDEV);
1884 /* Before locking the target dir, check we do not replace
1885 * a dir with a non-dir, otherwise it may deadlock with
1886 * link op which tries to create a link in this dir
1887 * back to this non-dir. */
1888 if (S_ISDIR(lu_object_attr(&mnew->mot_obj)) &&
1889 !S_ISDIR(lu_object_attr(&mold->mot_obj)))
1890 GOTO(out_put_new, rc = -EISDIR);
1892 lh_oldp = &info->mti_lh[MDT_LH_OLD];
1893 mdt_lock_reg_init(lh_oldp, LCK_EX);
1894 rc = mdt_object_lock(info, mold, lh_oldp, MDS_INODELOCK_LOOKUP |
1895 MDS_INODELOCK_XATTR, MDT_CROSS_LOCK);
1897 GOTO(out_put_new, rc);
1899 /* Check if @msrcdir is subdir of @mnew, before locking child
1900 * to avoid reverse locking. */
1901 rc = mdt_is_subdir(info, msrcdir, new_fid);
1903 GOTO(out_unlock_old, rc);
1905 /* We used to acquire MDS_INODELOCK_FULL here but we
1906 * can't do this now because a running HSM restore on
1907 * the rename onto victim will hold the layout
1908 * lock. See LU-4002. */
1910 lh_newp = &info->mti_lh[MDT_LH_NEW];
1911 mdt_lock_reg_init(lh_newp, LCK_EX);
1912 rc = mdt_object_lock(info, mnew, lh_newp,
1913 MDS_INODELOCK_LOOKUP |
1914 MDS_INODELOCK_UPDATE,
1917 GOTO(out_unlock_old, rc);
1919 /* get and save version after locking */
1920 mdt_version_get_save(info, mnew, 3);
1921 mdt_set_capainfo(info, 3, new_fid, BYPASS_CAPA);
1922 } else if (rc != -EREMOTE && rc != -ENOENT) {
1923 GOTO(out_put_old, rc);
1925 /* If mnew does not exist and mold are remote directory,
1926 * it only allows rename if they are under same directory */
1927 if (mtgtdir != msrcdir && mdt_object_remote(mold)) {
1928 CDEBUG(D_INFO, "Src child "DFID" is on another MDT\n",
1930 GOTO(out_put_old, rc = -EXDEV);
1933 lh_oldp = &info->mti_lh[MDT_LH_OLD];
1934 mdt_lock_reg_init(lh_oldp, LCK_EX);
1935 rc = mdt_object_lock(info, mold, lh_oldp, MDS_INODELOCK_LOOKUP |
1936 MDS_INODELOCK_XATTR, MDT_CROSS_LOCK);
1938 GOTO(out_put_old, rc);
1940 mdt_enoent_version_save(info, 3);
1943 /* step 5: rename it */
1944 mdt_reint_init_ma(info, ma);
1946 mdt_fail_write(info->mti_env, info->mti_mdt->mdt_bottom,
1947 OBD_FAIL_MDS_REINT_RENAME_WRITE);
1950 mutex_lock(&mnew->mot_lov_mutex);
1952 rc = mdo_rename(info->mti_env, mdt_object_child(msrcdir),
1953 mdt_object_child(mtgtdir), old_fid, &rr->rr_name,
1954 mnew != NULL ? mdt_object_child(mnew) : NULL,
1955 &rr->rr_tgt_name, ma);
1958 mutex_unlock(&mnew->mot_lov_mutex);
1960 /* handle last link of tgt object */
1962 mdt_counter_incr(req, LPROC_MDT_RENAME);
1964 mdt_handle_last_unlink(info, mnew, ma);
1966 mdt_rename_counter_tally(info, info->mti_mdt, req,
1972 mdt_object_unlock(info, mnew, lh_newp, rc);
1974 mdt_object_unlock(info, mold, lh_oldp, rc);
1977 mdt_object_put(info->mti_env, mnew);
1979 mdt_object_put(info->mti_env, mold);
1981 mdt_object_unlock_put(info, mtgtdir, lh_tgtdirp, rc);
1982 mdt_object_unlock_put(info, msrcdir, lh_srcdirp, rc);
1986 static int mdt_reint_rename_or_migrate(struct mdt_thread_info *info,
1987 struct mdt_lock_handle *lhc,
1988 enum mdt_rename_lock rename_lock)
1990 struct mdt_reint_record *rr = &info->mti_rr;
1991 struct ptlrpc_request *req = mdt_info_req(info);
1992 struct lustre_handle rename_lh = { 0 };
1996 if (info->mti_dlm_req)
1997 ldlm_request_cancel(req, info->mti_dlm_req, 0, LATF_SKIP);
1999 if (!fid_is_md_operative(rr->rr_fid1) ||
2000 !fid_is_md_operative(rr->rr_fid2))
2003 rc = mdt_rename_lock(info, &rename_lh, rename_lock);
2005 CERROR("%s: can't lock FS for rename: rc = %d\n",
2006 mdt_obd_name(info->mti_mdt), rc);
2010 if (rename_lock == MRL_RENAME)
2011 rc = mdt_reint_rename_internal(info, lhc);
2013 rc = mdt_reint_migrate_internal(info, lhc);
2015 if (lustre_handle_is_used(&rename_lh))
2016 mdt_rename_unlock(&rename_lh);
2021 static int mdt_reint_rename(struct mdt_thread_info *info,
2022 struct mdt_lock_handle *lhc)
2024 return mdt_reint_rename_or_migrate(info, lhc, MRL_RENAME);
2027 static int mdt_reint_migrate(struct mdt_thread_info *info,
2028 struct mdt_lock_handle *lhc)
2030 return mdt_reint_rename_or_migrate(info, lhc, MRL_MIGRATE);
2033 struct mdt_reinter {
2034 int (*mr_handler)(struct mdt_thread_info *, struct mdt_lock_handle *);
2035 enum lprocfs_extra_opc mr_extra_opc;
2038 static const struct mdt_reinter mdt_reinters[] = {
2040 .mr_handler = &mdt_reint_setattr,
2041 .mr_extra_opc = MDS_REINT_SETATTR,
2044 .mr_handler = &mdt_reint_create,
2045 .mr_extra_opc = MDS_REINT_CREATE,
2048 .mr_handler = &mdt_reint_link,
2049 .mr_extra_opc = MDS_REINT_LINK,
2052 .mr_handler = &mdt_reint_unlink,
2053 .mr_extra_opc = MDS_REINT_UNLINK,
2056 .mr_handler = &mdt_reint_rename,
2057 .mr_extra_opc = MDS_REINT_RENAME,
2060 .mr_handler = &mdt_reint_open,
2061 .mr_extra_opc = MDS_REINT_OPEN,
2063 [REINT_SETXATTR] = {
2064 .mr_handler = &mdt_reint_setxattr,
2065 .mr_extra_opc = MDS_REINT_SETXATTR,
2068 .mr_handler = &mdt_reint_unlink,
2069 .mr_extra_opc = MDS_REINT_UNLINK,
2072 .mr_handler = &mdt_reint_migrate,
2073 .mr_extra_opc = MDS_REINT_RENAME,
2077 int mdt_reint_rec(struct mdt_thread_info *info,
2078 struct mdt_lock_handle *lhc)
2080 const struct mdt_reinter *mr;
2084 if (!(info->mti_rr.rr_opcode < ARRAY_SIZE(mdt_reinters)))
2087 mr = &mdt_reinters[info->mti_rr.rr_opcode];
2088 if (mr->mr_handler == NULL)
2091 rc = (*mr->mr_handler)(info, lhc);
2093 lprocfs_counter_incr(ptlrpc_req2svc(mdt_info_req(info))->srv_stats,
2094 PTLRPC_LAST_CNTR + mr->mr_extra_opc);