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;
772 struct mdt_lock_handle *lh;
774 if (ma->ma_attr.la_valid != 0)
775 GOTO(out_put, rc = -EPROTO);
777 lh = &info->mti_lh[MDT_LH_PARENT];
778 mdt_lock_reg_init(lh, LCK_PW);
780 rc = mdt_object_lock(info, mo, lh,
786 buf->lb_buf = ma->ma_lmv;
787 buf->lb_len = ma->ma_lmv_size;
788 rc = mo_xattr_set(info->mti_env, mdt_object_child(mo),
789 buf, XATTR_NAME_DEFAULT_LMV, 0);
791 mdt_object_unlock(info, mo, lh, rc);
795 GOTO(out_put, rc = -EPROTO);
798 /* If file data is modified, add the dirty flag */
799 if (ma->ma_attr_flags & MDS_DATA_MODIFIED)
800 rc = mdt_add_dirty_flag(info, mo, ma);
802 ma->ma_need = MA_INODE;
804 rc = mdt_attr_get_complex(info, mo, ma);
808 mdt_pack_attr2body(info, repbody, &ma->ma_attr, mdt_object_fid(mo));
810 if (info->mti_mdt->mdt_lut.lut_oss_capa &&
811 exp_connect_flags(info->mti_exp) & OBD_CONNECT_OSS_CAPA &&
812 S_ISREG(lu_object_attr(&mo->mot_obj)) &&
813 (ma->ma_attr.la_valid & LA_SIZE) && !som_au) {
814 struct lustre_capa *capa;
816 capa = req_capsule_server_get(info->mti_pill, &RMF_CAPA2);
818 capa->lc_opc = CAPA_OPC_OSS_DEFAULT | CAPA_OPC_OSS_TRUNC;
819 rc = mo_capa_get(info->mti_env, mdt_object_child(mo), capa, 0);
822 repbody->mbo_valid |= OBD_MD_FLOSSCAPA;
827 mdt_object_put(info->mti_env, mo);
830 mdt_counter_incr(req, LPROC_MDT_SETATTR);
832 mdt_client_compatibility(info);
833 rc2 = mdt_fix_reply(info);
839 static int mdt_reint_create(struct mdt_thread_info *info,
840 struct mdt_lock_handle *lhc)
842 struct ptlrpc_request *req = mdt_info_req(info);
846 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_REINT_CREATE))
847 RETURN(err_serious(-ESTALE));
849 if (info->mti_dlm_req)
850 ldlm_request_cancel(mdt_info_req(info),
851 info->mti_dlm_req, 0, LATF_SKIP);
853 if (!lu_name_is_valid(&info->mti_rr.rr_name))
856 switch (info->mti_attr.ma_attr.la_mode & S_IFMT) {
858 mdt_counter_incr(req, LPROC_MDT_MKDIR);
866 /* Special file should stay on the same node as parent. */
867 mdt_counter_incr(req, LPROC_MDT_MKNOD);
870 CERROR("%s: Unsupported mode %o\n",
871 mdt_obd_name(info->mti_mdt),
872 info->mti_attr.ma_attr.la_mode);
873 RETURN(err_serious(-EOPNOTSUPP));
876 rc = mdt_md_create(info);
881 * VBR: save parent version in reply and child version getting by its name.
882 * Version of child is getting and checking during its lookup. If
884 static int mdt_reint_unlink(struct mdt_thread_info *info,
885 struct mdt_lock_handle *lhc)
887 struct mdt_reint_record *rr = &info->mti_rr;
888 struct ptlrpc_request *req = mdt_info_req(info);
889 struct md_attr *ma = &info->mti_attr;
890 struct lu_fid *child_fid = &info->mti_tmp_fid1;
891 struct mdt_object *mp;
892 struct mdt_object *mc;
893 struct mdt_lock_handle *parent_lh;
894 struct mdt_lock_handle *child_lh;
895 struct ldlm_enqueue_info *einfo = &info->mti_einfo;
896 struct mdt_lock_handle *s0_lh = NULL;
897 struct mdt_object *s0_obj = NULL;
902 DEBUG_REQ(D_INODE, req, "unlink "DFID"/"DNAME"", PFID(rr->rr_fid1),
903 PNAME(&rr->rr_name));
905 if (info->mti_dlm_req)
906 ldlm_request_cancel(req, info->mti_dlm_req, 0, LATF_SKIP);
908 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_REINT_UNLINK))
909 RETURN(err_serious(-ENOENT));
911 if (!fid_is_md_operative(rr->rr_fid1))
915 * step 1: Found the parent.
917 mp = mdt_object_find(info->mti_env, info->mti_mdt, rr->rr_fid1);
923 parent_lh = &info->mti_lh[MDT_LH_PARENT];
924 mdt_lock_pdo_init(parent_lh, LCK_PW, &rr->rr_name);
925 rc = mdt_object_lock(info, mp, parent_lh, MDS_INODELOCK_UPDATE,
928 GOTO(put_parent, rc);
930 if (!mdt_object_remote(mp)) {
931 rc = mdt_version_get_check_save(info, mp, 0);
933 GOTO(unlock_parent, rc);
936 /* step 2: find & lock the child */
937 /* lookup child object along with version checking */
939 rc = mdt_lookup_version_check(info, mp, &rr->rr_name, child_fid, 1);
941 /* Name might not be able to find during resend of
942 * remote unlink, considering following case.
943 * dir_A is a remote directory, the name entry of
944 * dir_A is on MDT0, the directory is on MDT1,
946 * 1. client sends unlink req to MDT1.
947 * 2. MDT1 sends name delete update to MDT0.
948 * 3. name entry is being deleted in MDT0 synchronously.
949 * 4. MDT1 is restarted.
950 * 5. client resends unlink req to MDT1. So it can not
951 * find the name entry on MDT0 anymore.
952 * In this case, MDT1 only needs to destory the local
955 if (mdt_object_remote(mp) && rc == -ENOENT &&
956 !fid_is_zero(rr->rr_fid2) &&
957 lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT) {
959 *child_fid = *rr->rr_fid2;
961 GOTO(unlock_parent, rc);
965 if (!fid_is_md_operative(child_fid))
966 GOTO(unlock_parent, rc = -EPERM);
968 /* We will lock the child regardless it is local or remote. No harm. */
969 mc = mdt_object_find(info->mti_env, info->mti_mdt, child_fid);
971 GOTO(unlock_parent, rc = PTR_ERR(mc));
973 child_lh = &info->mti_lh[MDT_LH_CHILD];
974 mdt_lock_reg_init(child_lh, LCK_EX);
975 if (mdt_object_remote(mc)) {
976 struct mdt_body *repbody;
978 if (!fid_is_zero(rr->rr_fid2)) {
979 CDEBUG(D_INFO, "%s: name "DNAME" cannot find "DFID"\n",
980 mdt_obd_name(info->mti_mdt),
981 PNAME(&rr->rr_name), PFID(mdt_object_fid(mc)));
982 GOTO(put_child, rc = -ENOENT);
984 CDEBUG(D_INFO, "%s: name "DNAME": "DFID" is on another MDT\n",
985 mdt_obd_name(info->mti_mdt),
986 PNAME(&rr->rr_name), PFID(mdt_object_fid(mc)));
988 if (!mdt_is_dne_client(req->rq_export))
989 /* Return -ENOTSUPP for old client */
990 GOTO(put_child, rc = -ENOTSUPP);
992 if (info->mti_spec.sp_rm_entry) {
993 struct lu_ucred *uc = mdt_ucred(info);
995 if (!md_capable(uc, CFS_CAP_SYS_ADMIN)) {
996 CERROR("%s: unlink remote entry is only "
997 "permitted for administrator: rc = %d\n",
998 mdt_obd_name(info->mti_mdt),
1000 GOTO(put_child, rc = -EPERM);
1003 ma->ma_need = MA_INODE;
1005 mdt_set_capainfo(info, 1, child_fid, BYPASS_CAPA);
1006 rc = mdo_unlink(info->mti_env, mdt_object_child(mp),
1007 NULL, &rr->rr_name, ma, no_name);
1008 GOTO(put_child, rc);
1010 /* Revoke the LOOKUP lock of the remote object granted by
1011 * this MDT. Since the unlink will happen on another MDT,
1012 * it will release the LOOKUP lock right away. Then What
1013 * would happen if another client try to grab the LOOKUP
1014 * lock at the same time with unlink XXX */
1015 mdt_object_lock(info, mc, child_lh, MDS_INODELOCK_LOOKUP,
1017 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1018 LASSERT(repbody != NULL);
1019 repbody->mbo_fid1 = *mdt_object_fid(mc);
1020 repbody->mbo_valid |= (OBD_MD_FLID | OBD_MD_MDS);
1021 GOTO(unlock_child, rc = -EREMOTE);
1022 } else if (info->mti_spec.sp_rm_entry) {
1024 CDEBUG(D_INFO, "%s: no rm_entry on local dir '"DNAME"': "
1026 mdt_obd_name(info->mti_mdt), PNAME(&rr->rr_name), rc);
1027 GOTO(put_child, rc);
1030 /* We used to acquire MDS_INODELOCK_FULL here but we can't do
1031 * this now because a running HSM restore on the child (unlink
1032 * victim) will hold the layout lock. See LU-4002. */
1033 rc = mdt_object_lock(info, mc, child_lh,
1034 MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE,
1037 GOTO(put_child, rc);
1039 * Now we can only make sure we need MA_INODE, in mdd layer, will check
1040 * whether need MA_LOV and MA_COOKIE.
1042 ma->ma_need = MA_INODE;
1045 s0_lh = &info->mti_lh[MDT_LH_LOCAL];
1046 mdt_lock_reg_init(s0_lh, LCK_EX);
1047 rc = mdt_lock_slaves(info, mc, LCK_EX, MDS_INODELOCK_UPDATE, s0_lh,
1050 GOTO(unlock_child, rc);
1052 mdt_fail_write(info->mti_env, info->mti_mdt->mdt_bottom,
1053 OBD_FAIL_MDS_REINT_UNLINK_WRITE);
1054 /* save version when object is locked */
1055 mdt_version_get_save(info, mc, 1);
1057 mdt_set_capainfo(info, 1, child_fid, BYPASS_CAPA);
1059 mutex_lock(&mc->mot_lov_mutex);
1061 rc = mdo_unlink(info->mti_env, mdt_object_child(mp),
1062 mdt_object_child(mc), &rr->rr_name, ma, no_name);
1064 mutex_unlock(&mc->mot_lov_mutex);
1066 if (rc == 0 && !lu_object_is_dying(&mc->mot_header))
1067 rc = mdt_attr_get_complex(info, mc, ma);
1069 mdt_handle_last_unlink(info, mc, ma);
1071 if (ma->ma_valid & MA_INODE) {
1072 switch (ma->ma_attr.la_mode & S_IFMT) {
1074 mdt_counter_incr(req, LPROC_MDT_RMDIR);
1082 mdt_counter_incr(req, LPROC_MDT_UNLINK);
1085 LASSERTF(0, "bad file type %o unlinking\n",
1086 ma->ma_attr.la_mode);
1093 mdt_unlock_slaves(info, mc, MDS_INODELOCK_UPDATE, s0_lh, s0_obj, einfo);
1094 mdt_object_unlock(info, mc, child_lh, rc);
1096 mdt_object_put(info->mti_env, mc);
1098 mdt_object_unlock(info, mp, parent_lh, rc);
1100 mdt_object_put(info->mti_env, mp);
1106 * VBR: save versions in reply: 0 - parent; 1 - child by fid; 2 - target by
1109 static int mdt_reint_link(struct mdt_thread_info *info,
1110 struct mdt_lock_handle *lhc)
1112 struct mdt_reint_record *rr = &info->mti_rr;
1113 struct ptlrpc_request *req = mdt_info_req(info);
1114 struct md_attr *ma = &info->mti_attr;
1115 struct mdt_object *ms;
1116 struct mdt_object *mp;
1117 struct mdt_lock_handle *lhs;
1118 struct mdt_lock_handle *lhp;
1122 DEBUG_REQ(D_INODE, req, "link "DFID" to "DFID"/"DNAME,
1123 PFID(rr->rr_fid1), PFID(rr->rr_fid2), PNAME(&rr->rr_name));
1125 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_REINT_LINK))
1126 RETURN(err_serious(-ENOENT));
1128 if (info->mti_dlm_req)
1129 ldlm_request_cancel(req, info->mti_dlm_req, 0, LATF_SKIP);
1131 /* Invalid case so return error immediately instead of
1133 if (lu_fid_eq(rr->rr_fid1, rr->rr_fid2))
1136 if (!fid_is_md_operative(rr->rr_fid1) ||
1137 !fid_is_md_operative(rr->rr_fid2))
1140 /* step 1: find & lock the target parent dir */
1141 lhp = &info->mti_lh[MDT_LH_PARENT];
1142 mdt_lock_pdo_init(lhp, LCK_PW, &rr->rr_name);
1143 mp = mdt_object_find_lock(info, rr->rr_fid2, lhp,
1144 MDS_INODELOCK_UPDATE);
1146 RETURN(PTR_ERR(mp));
1148 rc = mdt_version_get_check_save(info, mp, 0);
1150 GOTO(out_unlock_parent, rc);
1152 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_RENAME3, 5);
1154 /* step 2: find & lock the source */
1155 lhs = &info->mti_lh[MDT_LH_CHILD];
1156 mdt_lock_reg_init(lhs, LCK_EX);
1158 ms = mdt_object_find(info->mti_env, info->mti_mdt, rr->rr_fid1);
1160 GOTO(out_unlock_parent, rc = PTR_ERR(ms));
1162 if (!mdt_object_exists(ms)) {
1163 mdt_object_put(info->mti_env, ms);
1164 CDEBUG(D_INFO, "%s: "DFID" does not exist.\n",
1165 mdt_obd_name(info->mti_mdt), PFID(rr->rr_fid1));
1166 GOTO(out_unlock_parent, rc = -ENOENT);
1169 if (mdt_object_remote(ms)) {
1170 mdt_object_put(info->mti_env, ms);
1171 CERROR("%s: source inode "DFID" on remote MDT from "DFID"\n",
1172 mdt_obd_name(info->mti_mdt), PFID(rr->rr_fid1),
1174 GOTO(out_unlock_parent, rc = -EXDEV);
1177 rc = mdt_object_lock(info, ms, lhs, MDS_INODELOCK_UPDATE |
1178 MDS_INODELOCK_XATTR, MDT_LOCAL_LOCK);
1180 mdt_object_put(info->mti_env, ms);
1181 GOTO(out_unlock_parent, rc);
1184 /* step 3: link it */
1185 mdt_fail_write(info->mti_env, info->mti_mdt->mdt_bottom,
1186 OBD_FAIL_MDS_REINT_LINK_WRITE);
1188 tgt_vbr_obj_set(info->mti_env, mdt_obj2dt(ms));
1189 rc = mdt_version_get_check_save(info, ms, 1);
1191 GOTO(out_unlock_child, rc);
1193 /** check target version by name during replay */
1194 rc = mdt_lookup_version_check(info, mp, &rr->rr_name,
1195 &info->mti_tmp_fid1, 2);
1196 if (rc != 0 && rc != -ENOENT)
1197 GOTO(out_unlock_child, rc);
1198 /* save version of file name for replay, it must be ENOENT here */
1199 if (!req_is_replay(mdt_info_req(info))) {
1200 if (rc != -ENOENT) {
1201 CDEBUG(D_INFO, "link target "DNAME" existed!\n",
1202 PNAME(&rr->rr_name));
1203 GOTO(out_unlock_child, rc = -EEXIST);
1205 info->mti_ver[2] = ENOENT_VERSION;
1206 mdt_version_save(mdt_info_req(info), info->mti_ver[2], 2);
1209 rc = mdo_link(info->mti_env, mdt_object_child(mp),
1210 mdt_object_child(ms), &rr->rr_name, ma);
1213 mdt_counter_incr(req, LPROC_MDT_LINK);
1217 mdt_object_unlock_put(info, ms, lhs, rc);
1219 mdt_object_unlock_put(info, mp, lhp, rc);
1223 * lock the part of the directory according to the hash of the name
1224 * (lh->mlh_pdo_hash) in parallel directory lock.
1226 static int mdt_pdir_hash_lock(struct mdt_thread_info *info,
1227 struct mdt_lock_handle *lh,
1228 struct mdt_object *obj, __u64 ibits)
1230 struct ldlm_res_id *res = &info->mti_res_id;
1231 struct ldlm_namespace *ns = info->mti_mdt->mdt_namespace;
1232 ldlm_policy_data_t *policy = &info->mti_policy;
1236 * Finish res_id initializing by name hash marking part of
1237 * directory which is taking modification.
1239 LASSERT(lh->mlh_pdo_hash != 0);
1240 fid_build_pdo_res_name(mdt_object_fid(obj), lh->mlh_pdo_hash, res);
1241 memset(policy, 0, sizeof(*policy));
1242 policy->l_inodebits.bits = ibits;
1244 * Use LDLM_FL_LOCAL_ONLY for this lock. We do not know yet if it is
1245 * going to be sent to client. If it is - mdt_intent_policy() path will
1246 * fix it up and turn FL_LOCAL flag off.
1248 rc = mdt_fid_lock(ns, &lh->mlh_reg_lh, lh->mlh_reg_mode, policy,
1249 res, LDLM_FL_LOCAL_ONLY | LDLM_FL_ATOMIC_CB,
1250 &info->mti_exp->exp_handle.h_cookie);
1254 enum mdt_rename_lock {
1260 * Get BFL lock for rename or migrate process, right now, it does not support
1261 * cross-MDT rename, so we only need global rename lock during migration.
1263 static int mdt_rename_lock(struct mdt_thread_info *info,
1264 struct lustre_handle *lh,
1265 enum mdt_rename_lock rename_lock)
1267 struct ldlm_namespace *ns = info->mti_mdt->mdt_namespace;
1268 ldlm_policy_data_t *policy = &info->mti_policy;
1269 struct ldlm_res_id *res_id = &info->mti_res_id;
1274 /* XXX only do global rename lock for migration */
1275 if (mdt_seq_site(info->mti_mdt)->ss_node_id != 0 &&
1276 rename_lock == MRL_MIGRATE) {
1277 struct lu_fid *fid = &info->mti_tmp_fid1;
1278 struct mdt_object *obj;
1280 /* XXX, right now, it has to use object API to
1281 * enqueue lock cross MDT, so it will enqueue
1282 * rename lock(with LUSTRE_BFL_FID) by root object */
1284 obj = mdt_object_find(info->mti_env, info->mti_mdt, fid);
1286 RETURN(PTR_ERR(obj));
1288 LASSERT(mdt_object_remote(obj));
1289 rc = mdt_remote_object_lock(info, obj,
1290 &LUSTRE_BFL_FID, lh,
1292 MDS_INODELOCK_UPDATE);
1293 mdt_object_put(info->mti_env, obj);
1295 fid_build_reg_res_name(&LUSTRE_BFL_FID, res_id);
1296 memset(policy, 0, sizeof *policy);
1297 policy->l_inodebits.bits = MDS_INODELOCK_UPDATE;
1298 flags = LDLM_FL_LOCAL_ONLY | LDLM_FL_ATOMIC_CB;
1299 rc = ldlm_cli_enqueue_local(ns, res_id, LDLM_IBITS, policy,
1300 LCK_EX, &flags, ldlm_blocking_ast,
1301 ldlm_completion_ast, NULL, NULL, 0,
1303 &info->mti_exp->exp_handle.h_cookie,
1310 static void mdt_rename_unlock(struct lustre_handle *lh)
1313 LASSERT(lustre_handle_is_used(lh));
1314 /* Cancel the single rename lock right away */
1315 ldlm_lock_decref_and_cancel(lh, LCK_EX);
1320 * This is is_subdir() variant, it is CMD if cmm forwards it to correct
1321 * target. Source should not be ancestor of target dir. May be other rename
1322 * checks can be moved here later.
1324 static int mdt_is_subdir(struct mdt_thread_info *info,
1325 struct mdt_object *dir,
1326 const struct lu_fid *fid)
1328 struct lu_fid dir_fid = dir->mot_header.loh_fid;
1332 /* If the source and target are in the same directory, they can not
1333 * be parent/child relationship, so subdir check is not needed */
1334 if (lu_fid_eq(&dir_fid, fid))
1337 if (!mdt_object_exists(dir))
1340 rc = mdo_is_subdir(info->mti_env, mdt_object_child(dir),
1343 CERROR("%s: failed subdir check in "DFID" for "DFID
1344 ": rc = %d\n", mdt_obd_name(info->mti_mdt),
1345 PFID(&dir_fid), PFID(fid), rc);
1346 /* Return EINVAL only if a parent is the @fid */
1350 /* check the found fid */
1351 if (lu_fid_eq(&dir_fid, fid))
1358 /* Update object linkEA */
1359 struct mdt_lock_list {
1360 struct mdt_object *mll_obj;
1361 struct mdt_lock_handle mll_lh;
1362 struct list_head mll_list;
1365 static void mdt_unlock_list(struct mdt_thread_info *info,
1366 struct list_head *list, int rc)
1368 struct mdt_lock_list *mll;
1369 struct mdt_lock_list *mll2;
1371 list_for_each_entry_safe(mll, mll2, list, mll_list) {
1372 mdt_object_unlock_put(info, mll->mll_obj, &mll->mll_lh, rc);
1373 list_del(&mll->mll_list);
1378 static int mdt_lock_objects_in_linkea(struct mdt_thread_info *info,
1379 struct mdt_object *obj,
1380 struct mdt_object *pobj,
1381 struct list_head *lock_list)
1383 struct lu_buf *buf = &info->mti_big_buf;
1384 struct linkea_data ldata = { NULL };
1389 if (S_ISDIR(lu_object_attr(&obj->mot_obj)))
1392 buf = lu_buf_check_and_alloc(buf, PATH_MAX);
1393 if (buf->lb_buf == NULL)
1397 rc = mdt_links_read(info, obj, &ldata);
1399 if (rc == -ENOENT || rc == -ENODATA)
1404 LASSERT(ldata.ld_leh != NULL);
1405 ldata.ld_lee = (struct link_ea_entry *)(ldata.ld_leh + 1);
1406 for (count = 0; count < ldata.ld_leh->leh_reccount; count++) {
1407 struct mdt_device *mdt = info->mti_mdt;
1408 struct mdt_object *mdt_pobj;
1409 struct mdt_lock_list *mll;
1410 struct lu_name name;
1413 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen,
1415 ldata.ld_lee = (struct link_ea_entry *)((char *)ldata.ld_lee +
1417 mdt_pobj = mdt_object_find(info->mti_env, mdt, &fid);
1418 if (IS_ERR(mdt_pobj)) {
1419 CWARN("%s: cannot find obj "DFID": rc = %ld\n",
1420 mdt_obd_name(mdt), PFID(&fid), PTR_ERR(mdt_pobj));
1424 if (!mdt_object_exists(mdt_pobj)) {
1425 CDEBUG(D_INFO, "%s: obj "DFID" does not exist\n",
1426 mdt_obd_name(mdt), PFID(&fid));
1427 mdt_object_put(info->mti_env, mdt_pobj);
1431 if (mdt_pobj == pobj) {
1432 CDEBUG(D_INFO, "%s: skipping parent obj "DFID"\n",
1433 mdt_obd_name(mdt), PFID(&fid));
1434 mdt_object_put(info->mti_env, mdt_pobj);
1440 mdt_object_put(info->mti_env, mdt_pobj);
1441 GOTO(out, rc = -ENOMEM);
1444 mdt_lock_pdo_init(&mll->mll_lh, LCK_PW, &name);
1445 rc = mdt_object_lock(info, mdt_pobj, &mll->mll_lh,
1446 MDS_INODELOCK_UPDATE,
1449 CERROR("%s: cannot lock "DFID": rc =%d\n",
1450 mdt_obd_name(mdt), PFID(&fid), rc);
1451 mdt_object_put(info->mti_env, mdt_pobj);
1456 INIT_LIST_HEAD(&mll->mll_list);
1457 mll->mll_obj = mdt_pobj;
1458 list_add_tail(&mll->mll_list, lock_list);
1462 mdt_unlock_list(info, lock_list, rc);
1466 /* migrate files from one MDT to another MDT */
1467 static int mdt_reint_migrate_internal(struct mdt_thread_info *info,
1468 struct mdt_lock_handle *lhc)
1470 struct mdt_reint_record *rr = &info->mti_rr;
1471 struct md_attr *ma = &info->mti_attr;
1472 struct mdt_object *msrcdir;
1473 struct mdt_object *mold;
1474 struct mdt_object *mnew = NULL;
1475 struct mdt_lock_handle *lh_dirp;
1476 struct mdt_lock_handle *lh_childp;
1477 struct mdt_lock_handle *lh_tgtp = NULL;
1478 struct lu_fid *old_fid = &info->mti_tmp_fid1;
1479 struct list_head lock_list;
1483 CDEBUG(D_INODE, "migrate "DFID"/"DNAME" to "DFID"\n", PFID(rr->rr_fid1),
1484 PNAME(&rr->rr_name), PFID(rr->rr_fid2));
1486 /* 1: lock the source dir. */
1487 msrcdir = mdt_object_find(info->mti_env, info->mti_mdt, rr->rr_fid1);
1488 if (IS_ERR(msrcdir)) {
1489 CERROR("%s: cannot find source dir "DFID" : rc = %d\n",
1490 mdt_obd_name(info->mti_mdt), PFID(rr->rr_fid1),
1491 (int)PTR_ERR(msrcdir));
1492 RETURN(PTR_ERR(msrcdir));
1495 lh_dirp = &info->mti_lh[MDT_LH_PARENT];
1496 mdt_lock_pdo_init(lh_dirp, LCK_PW, &rr->rr_name);
1497 rc = mdt_object_lock(info, msrcdir, lh_dirp,
1498 MDS_INODELOCK_UPDATE,
1501 GOTO(out_put_parent, rc);
1503 if (!mdt_object_remote(msrcdir)) {
1504 rc = mdt_version_get_check_save(info, msrcdir, 0);
1506 GOTO(out_unlock_parent, rc);
1509 /* 2: sanity check and find the object to be migrated. */
1511 rc = mdt_lookup_version_check(info, msrcdir, &rr->rr_name, old_fid, 2);
1513 GOTO(out_unlock_parent, rc);
1515 if (lu_fid_eq(old_fid, rr->rr_fid1) || lu_fid_eq(old_fid, rr->rr_fid2))
1516 GOTO(out_unlock_parent, rc = -EINVAL);
1518 if (!fid_is_md_operative(old_fid))
1519 GOTO(out_unlock_parent, rc = -EPERM);
1521 mold = mdt_object_find(info->mti_env, info->mti_mdt, old_fid);
1523 GOTO(out_unlock_parent, rc = PTR_ERR(mold));
1525 if (mdt_object_remote(mold)) {
1526 CERROR("%s: source "DFID" is on the remote MDT\n",
1527 mdt_obd_name(info->mti_mdt), PFID(old_fid));
1528 GOTO(out_put_child, rc = -EREMOTE);
1531 if (S_ISREG(lu_object_attr(&mold->mot_obj)) &&
1532 !mdt_object_remote(msrcdir)) {
1533 CERROR("%s: parent "DFID" is still on the same"
1534 " MDT, which should be migrated first:"
1535 " rc = %d\n", mdt_obd_name(info->mti_mdt),
1536 PFID(mdt_object_fid(msrcdir)), -EPERM);
1537 GOTO(out_put_child, rc = -EPERM);
1540 rc = mdt_remote_permission(info, msrcdir, mold);
1542 GOTO(out_put_child, rc);
1544 /* 3: iterate the linkea of the object and lock all of the objects */
1545 INIT_LIST_HEAD(&lock_list);
1546 rc = mdt_lock_objects_in_linkea(info, mold, msrcdir, &lock_list);
1548 GOTO(out_put_child, rc);
1550 /* 4: lock of the object migrated object */
1551 lh_childp = &info->mti_lh[MDT_LH_OLD];
1552 mdt_lock_reg_init(lh_childp, LCK_EX);
1553 rc = mdt_object_lock(info, mold, lh_childp,
1554 MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
1555 MDS_INODELOCK_LAYOUT, MDT_CROSS_LOCK);
1557 GOTO(out_unlock_list, rc);
1559 ma->ma_need = MA_LMV;
1561 ma->ma_lmv = (union lmv_mds_md *)info->mti_xattr_buf;
1562 ma->ma_lmv_size = sizeof(info->mti_xattr_buf);
1563 rc = mdt_stripe_get(info, mold, ma, XATTR_NAME_LMV);
1565 GOTO(out_unlock_list, rc);
1567 if ((ma->ma_valid & MA_LMV)) {
1568 struct lmv_mds_md_v1 *lmm1;
1570 lmv_le_to_cpu(ma->ma_lmv, ma->ma_lmv);
1571 lmm1 = &ma->ma_lmv->lmv_md_v1;
1572 if (!(lmm1->lmv_hash_type & LMV_HASH_FLAG_MIGRATION)) {
1573 CERROR("%s: can not migrate striped dir "DFID
1574 ": rc = %d\n", mdt_obd_name(info->mti_mdt),
1575 PFID(mdt_object_fid(mold)), -EPERM);
1576 GOTO(out_unlock_child, rc = -EPERM);
1579 if (!fid_is_sane(&lmm1->lmv_stripe_fids[1]))
1580 GOTO(out_unlock_child, rc = -EINVAL);
1582 mnew = mdt_object_find(info->mti_env, info->mti_mdt,
1583 &lmm1->lmv_stripe_fids[1]);
1585 GOTO(out_unlock_child, rc = PTR_ERR(mnew));
1587 if (!mdt_object_remote(mnew)) {
1588 CERROR("%s: "DFID" being migrated is on this MDT:"
1589 " rc = %d\n", mdt_obd_name(info->mti_mdt),
1590 PFID(rr->rr_fid2), -EPERM);
1591 GOTO(out_put_new, rc = -EPERM);
1594 lh_tgtp = &info->mti_lh[MDT_LH_CHILD];
1595 mdt_lock_reg_init(lh_tgtp, LCK_EX);
1596 rc = mdt_remote_object_lock(info, mnew,
1597 mdt_object_fid(mnew),
1598 &lh_tgtp->mlh_rreg_lh,
1599 lh_tgtp->mlh_rreg_mode,
1600 MDS_INODELOCK_UPDATE);
1603 GOTO(out_put_new, rc);
1606 mnew = mdt_object_find(info->mti_env, info->mti_mdt,
1609 GOTO(out_unlock_child, rc = PTR_ERR(mnew));
1610 if (!mdt_object_remote(mnew)) {
1611 CERROR("%s: Migration "DFID" is on this MDT:"
1612 " rc = %d\n", mdt_obd_name(info->mti_mdt),
1613 PFID(rr->rr_fid2), -EXDEV);
1614 GOTO(out_put_new, rc = -EXDEV);
1619 mdt_reint_init_ma(info, ma);
1621 mdt_fail_write(info->mti_env, info->mti_mdt->mdt_bottom,
1622 OBD_FAIL_MDS_REINT_RENAME_WRITE);
1624 rc = mdo_migrate(info->mti_env, mdt_object_child(msrcdir),
1625 mdt_object_child(mold), &rr->rr_name,
1626 mdt_object_child(mnew), ma);
1628 GOTO(out_unlock_new, rc);
1630 if (lh_tgtp != NULL)
1631 mdt_object_unlock(info, mnew, lh_tgtp, rc);
1634 mdt_object_put(info->mti_env, mnew);
1636 mdt_object_unlock(info, mold, lh_childp, rc);
1638 mdt_unlock_list(info, &lock_list, rc);
1640 mdt_object_put(info->mti_env, mold);
1642 mdt_object_unlock(info, msrcdir, lh_dirp, rc);
1644 mdt_object_put(info->mti_env, msrcdir);
1649 static struct mdt_object *mdt_object_find_check(struct mdt_thread_info *info,
1650 const struct lu_fid *fid,
1653 struct mdt_object *dir;
1657 dir = mdt_object_find(info->mti_env, info->mti_mdt, fid);
1661 /* check early, the real version will be saved after locking */
1662 rc = mdt_version_get_check(info, dir, idx);
1668 mdt_object_put(info->mti_env, dir);
1672 static int mdt_object_lock_save(struct mdt_thread_info *info,
1673 struct mdt_object *dir,
1674 struct mdt_lock_handle *lh,
1679 /* we lock the target dir if it is local */
1680 rc = mdt_object_lock(info, dir, lh, MDS_INODELOCK_UPDATE,
1685 /* get and save correct version after locking */
1686 mdt_version_get_save(info, dir, idx);
1691 static int mdt_rename_parents_lock(struct mdt_thread_info *info,
1692 struct mdt_object **srcp,
1693 struct mdt_object **tgtp)
1695 struct mdt_reint_record *rr = &info->mti_rr;
1696 const struct lu_fid *fid_src = rr->rr_fid1;
1697 const struct lu_fid *fid_tgt = rr->rr_fid2;
1698 struct mdt_lock_handle *lh_src = &info->mti_lh[MDT_LH_PARENT];
1699 struct mdt_lock_handle *lh_tgt = &info->mti_lh[MDT_LH_CHILD];
1700 struct mdt_object *src;
1701 struct mdt_object *tgt;
1706 /* find both parents. */
1707 src = mdt_object_find_check(info, fid_src, 0);
1709 RETURN(PTR_ERR(src));
1711 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_RENAME3, 5);
1713 if (lu_fid_eq(fid_src, fid_tgt)) {
1715 mdt_object_get(info->mti_env, tgt);
1717 /* Check if the @src is not a child of the @tgt, otherwise a
1718 * reverse locking must take place. */
1719 rc = mdt_is_subdir(info, src, fid_tgt);
1723 GOTO(err_src_put, rc);
1725 tgt = mdt_object_find_check(info, fid_tgt, 1);
1727 GOTO(err_src_put, rc = PTR_ERR(tgt));
1729 if (unlikely(mdt_object_remote(tgt))) {
1730 CDEBUG(D_INFO, "Source dir "DFID" target dir "DFID
1731 "on different MDTs\n", PFID(fid_src),
1733 GOTO(err_tgt_put, rc = -EXDEV);
1737 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_RENAME4, 5);
1739 /* lock parents in the proper order. */
1741 rc = mdt_object_lock_save(info, tgt, lh_tgt, 1);
1743 GOTO(err_tgt_put, rc);
1745 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_RENAME, 5);
1747 rc = mdt_object_lock_save(info, src, lh_src, 0);
1749 rc = mdt_object_lock_save(info, src, lh_src, 0);
1751 GOTO(err_tgt_put, rc);
1753 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_RENAME, 5);
1756 rc = mdt_object_lock_save(info, tgt, lh_tgt, 1);
1757 else if (lh_src->mlh_pdo_hash != lh_tgt->mlh_pdo_hash) {
1758 rc = mdt_pdir_hash_lock(info, lh_tgt, tgt,
1759 MDS_INODELOCK_UPDATE);
1760 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_PDO_LOCK2, 10);
1764 GOTO(err_unlock, rc);
1766 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_RENAME4, 5);
1773 /* The order does not matter as the handle is checked inside,
1774 * as well as not used handle. */
1775 mdt_object_unlock(info, src, lh_src, rc);
1776 mdt_object_unlock(info, tgt, lh_tgt, rc);
1778 mdt_object_put(info->mti_env, tgt);
1780 mdt_object_put(info->mti_env, src);
1785 * VBR: rename versions in reply: 0 - src parent; 1 - tgt parent;
1786 * 2 - src child; 3 - tgt child.
1787 * Update on disk version of src child.
1790 * For DNE phase I, only these renames are allowed
1791 * mv src_p/src_c tgt_p/tgt_c
1792 * 1. src_p/src_c/tgt_p/tgt_c are in the same MDT.
1793 * 2. src_p and tgt_p are same directory, and tgt_c does not
1794 * exists. In this case, all of modification will happen
1795 * in the MDT where ithesource parent is, only one remote
1796 * update is needed, i.e. set c_time/m_time on the child.
1797 * And tgt_c will be still in the same MDT as the original
1800 static int mdt_reint_rename_internal(struct mdt_thread_info *info,
1801 struct mdt_lock_handle *lhc)
1803 struct mdt_reint_record *rr = &info->mti_rr;
1804 struct md_attr *ma = &info->mti_attr;
1805 struct ptlrpc_request *req = mdt_info_req(info);
1806 struct mdt_object *msrcdir = NULL;
1807 struct mdt_object *mtgtdir = NULL;
1808 struct mdt_object *mold;
1809 struct mdt_object *mnew = NULL;
1810 struct mdt_lock_handle *lh_srcdirp;
1811 struct mdt_lock_handle *lh_tgtdirp;
1812 struct mdt_lock_handle *lh_oldp = NULL;
1813 struct mdt_lock_handle *lh_newp = NULL;
1814 struct lu_fid *old_fid = &info->mti_tmp_fid1;
1815 struct lu_fid *new_fid = &info->mti_tmp_fid2;
1819 DEBUG_REQ(D_INODE, req, "rename "DFID"/"DNAME" to "DFID"/"DNAME,
1820 PFID(rr->rr_fid1), PNAME(&rr->rr_name),
1821 PFID(rr->rr_fid2), PNAME(&rr->rr_tgt_name));
1823 lh_srcdirp = &info->mti_lh[MDT_LH_PARENT];
1824 mdt_lock_pdo_init(lh_srcdirp, LCK_PW, &rr->rr_name);
1825 lh_tgtdirp = &info->mti_lh[MDT_LH_CHILD];
1826 mdt_lock_pdo_init(lh_tgtdirp, LCK_PW, &rr->rr_tgt_name);
1828 /* step 1&2: lock the source and target dirs. */
1829 rc = mdt_rename_parents_lock(info, &msrcdir, &mtgtdir);
1833 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_RENAME2, 5);
1835 /* step 3: find & lock the old object. */
1837 rc = mdt_lookup_version_check(info, msrcdir, &rr->rr_name, old_fid, 2);
1839 GOTO(out_unlock_parents, rc);
1841 if (lu_fid_eq(old_fid, rr->rr_fid1) || lu_fid_eq(old_fid, rr->rr_fid2))
1842 GOTO(out_unlock_parents, rc = -EINVAL);
1844 if (!fid_is_md_operative(old_fid))
1845 GOTO(out_unlock_parents, rc = -EPERM);
1847 mold = mdt_object_find(info->mti_env, info->mti_mdt, old_fid);
1849 GOTO(out_unlock_parents, rc = PTR_ERR(mold));
1851 /* Check if @mtgtdir is subdir of @mold, before locking child
1852 * to avoid reverse locking. */
1853 rc = mdt_is_subdir(info, mtgtdir, old_fid);
1855 GOTO(out_put_old, rc);
1857 tgt_vbr_obj_set(info->mti_env, mdt_obj2dt(mold));
1858 /* save version after locking */
1859 mdt_version_get_save(info, mold, 2);
1860 mdt_set_capainfo(info, 2, old_fid, BYPASS_CAPA);
1862 /* step 4: find & lock the new object. */
1863 /* new target object may not exist now */
1864 /* lookup with version checking */
1866 rc = mdt_lookup_version_check(info, mtgtdir, &rr->rr_tgt_name, new_fid,
1869 /* the new_fid should have been filled at this moment */
1870 if (lu_fid_eq(old_fid, new_fid))
1871 GOTO(out_put_old, rc);
1873 if (lu_fid_eq(new_fid, rr->rr_fid1) ||
1874 lu_fid_eq(new_fid, rr->rr_fid2))
1875 GOTO(out_put_old, rc = -EINVAL);
1877 if (!fid_is_md_operative(new_fid))
1878 GOTO(out_put_old, rc = -EPERM);
1880 if (mdt_object_remote(mold)) {
1881 CDEBUG(D_INFO, "Src child "DFID" is on another MDT\n",
1883 GOTO(out_put_old, rc = -EXDEV);
1886 mnew = mdt_object_find(info->mti_env, info->mti_mdt, new_fid);
1888 GOTO(out_put_old, rc = PTR_ERR(mnew));
1890 if (mdt_object_remote(mnew)) {
1891 CDEBUG(D_INFO, "src child "DFID" is on another MDT\n",
1893 GOTO(out_put_new, rc = -EXDEV);
1896 /* Before locking the target dir, check we do not replace
1897 * a dir with a non-dir, otherwise it may deadlock with
1898 * link op which tries to create a link in this dir
1899 * back to this non-dir. */
1900 if (S_ISDIR(lu_object_attr(&mnew->mot_obj)) &&
1901 !S_ISDIR(lu_object_attr(&mold->mot_obj)))
1902 GOTO(out_put_new, rc = -EISDIR);
1904 lh_oldp = &info->mti_lh[MDT_LH_OLD];
1905 mdt_lock_reg_init(lh_oldp, LCK_EX);
1906 rc = mdt_object_lock(info, mold, lh_oldp, MDS_INODELOCK_LOOKUP |
1907 MDS_INODELOCK_XATTR, MDT_CROSS_LOCK);
1909 GOTO(out_put_new, rc);
1911 /* Check if @msrcdir is subdir of @mnew, before locking child
1912 * to avoid reverse locking. */
1913 rc = mdt_is_subdir(info, msrcdir, new_fid);
1915 GOTO(out_unlock_old, rc);
1917 /* We used to acquire MDS_INODELOCK_FULL here but we
1918 * can't do this now because a running HSM restore on
1919 * the rename onto victim will hold the layout
1920 * lock. See LU-4002. */
1922 lh_newp = &info->mti_lh[MDT_LH_NEW];
1923 mdt_lock_reg_init(lh_newp, LCK_EX);
1924 rc = mdt_object_lock(info, mnew, lh_newp,
1925 MDS_INODELOCK_LOOKUP |
1926 MDS_INODELOCK_UPDATE,
1929 GOTO(out_unlock_old, rc);
1931 /* get and save version after locking */
1932 mdt_version_get_save(info, mnew, 3);
1933 mdt_set_capainfo(info, 3, new_fid, BYPASS_CAPA);
1934 } else if (rc != -EREMOTE && rc != -ENOENT) {
1935 GOTO(out_put_old, rc);
1937 /* If mnew does not exist and mold are remote directory,
1938 * it only allows rename if they are under same directory */
1939 if (mtgtdir != msrcdir && mdt_object_remote(mold)) {
1940 CDEBUG(D_INFO, "Src child "DFID" is on another MDT\n",
1942 GOTO(out_put_old, rc = -EXDEV);
1945 lh_oldp = &info->mti_lh[MDT_LH_OLD];
1946 mdt_lock_reg_init(lh_oldp, LCK_EX);
1947 rc = mdt_object_lock(info, mold, lh_oldp, MDS_INODELOCK_LOOKUP |
1948 MDS_INODELOCK_XATTR, MDT_CROSS_LOCK);
1950 GOTO(out_put_old, rc);
1952 mdt_enoent_version_save(info, 3);
1955 /* step 5: rename it */
1956 mdt_reint_init_ma(info, ma);
1958 mdt_fail_write(info->mti_env, info->mti_mdt->mdt_bottom,
1959 OBD_FAIL_MDS_REINT_RENAME_WRITE);
1962 mutex_lock(&mnew->mot_lov_mutex);
1964 rc = mdo_rename(info->mti_env, mdt_object_child(msrcdir),
1965 mdt_object_child(mtgtdir), old_fid, &rr->rr_name,
1966 mnew != NULL ? mdt_object_child(mnew) : NULL,
1967 &rr->rr_tgt_name, ma);
1970 mutex_unlock(&mnew->mot_lov_mutex);
1972 /* handle last link of tgt object */
1974 mdt_counter_incr(req, LPROC_MDT_RENAME);
1976 mdt_handle_last_unlink(info, mnew, ma);
1978 mdt_rename_counter_tally(info, info->mti_mdt, req,
1984 mdt_object_unlock(info, mnew, lh_newp, rc);
1986 mdt_object_unlock(info, mold, lh_oldp, rc);
1989 mdt_object_put(info->mti_env, mnew);
1991 mdt_object_put(info->mti_env, mold);
1993 mdt_object_unlock_put(info, mtgtdir, lh_tgtdirp, rc);
1994 mdt_object_unlock_put(info, msrcdir, lh_srcdirp, rc);
1998 static int mdt_reint_rename_or_migrate(struct mdt_thread_info *info,
1999 struct mdt_lock_handle *lhc,
2000 enum mdt_rename_lock rename_lock)
2002 struct mdt_reint_record *rr = &info->mti_rr;
2003 struct ptlrpc_request *req = mdt_info_req(info);
2004 struct lustre_handle rename_lh = { 0 };
2008 if (info->mti_dlm_req)
2009 ldlm_request_cancel(req, info->mti_dlm_req, 0, LATF_SKIP);
2011 if (!fid_is_md_operative(rr->rr_fid1) ||
2012 !fid_is_md_operative(rr->rr_fid2))
2015 rc = mdt_rename_lock(info, &rename_lh, rename_lock);
2017 CERROR("%s: can't lock FS for rename: rc = %d\n",
2018 mdt_obd_name(info->mti_mdt), rc);
2022 if (rename_lock == MRL_RENAME)
2023 rc = mdt_reint_rename_internal(info, lhc);
2025 rc = mdt_reint_migrate_internal(info, lhc);
2027 if (lustre_handle_is_used(&rename_lh))
2028 mdt_rename_unlock(&rename_lh);
2033 static int mdt_reint_rename(struct mdt_thread_info *info,
2034 struct mdt_lock_handle *lhc)
2036 return mdt_reint_rename_or_migrate(info, lhc, MRL_RENAME);
2039 static int mdt_reint_migrate(struct mdt_thread_info *info,
2040 struct mdt_lock_handle *lhc)
2042 return mdt_reint_rename_or_migrate(info, lhc, MRL_MIGRATE);
2045 struct mdt_reinter {
2046 int (*mr_handler)(struct mdt_thread_info *, struct mdt_lock_handle *);
2047 enum lprocfs_extra_opc mr_extra_opc;
2050 static const struct mdt_reinter mdt_reinters[] = {
2052 .mr_handler = &mdt_reint_setattr,
2053 .mr_extra_opc = MDS_REINT_SETATTR,
2056 .mr_handler = &mdt_reint_create,
2057 .mr_extra_opc = MDS_REINT_CREATE,
2060 .mr_handler = &mdt_reint_link,
2061 .mr_extra_opc = MDS_REINT_LINK,
2064 .mr_handler = &mdt_reint_unlink,
2065 .mr_extra_opc = MDS_REINT_UNLINK,
2068 .mr_handler = &mdt_reint_rename,
2069 .mr_extra_opc = MDS_REINT_RENAME,
2072 .mr_handler = &mdt_reint_open,
2073 .mr_extra_opc = MDS_REINT_OPEN,
2075 [REINT_SETXATTR] = {
2076 .mr_handler = &mdt_reint_setxattr,
2077 .mr_extra_opc = MDS_REINT_SETXATTR,
2080 .mr_handler = &mdt_reint_unlink,
2081 .mr_extra_opc = MDS_REINT_UNLINK,
2084 .mr_handler = &mdt_reint_migrate,
2085 .mr_extra_opc = MDS_REINT_RENAME,
2089 int mdt_reint_rec(struct mdt_thread_info *info,
2090 struct mdt_lock_handle *lhc)
2092 const struct mdt_reinter *mr;
2096 if (!(info->mti_rr.rr_opcode < ARRAY_SIZE(mdt_reinters)))
2099 mr = &mdt_reinters[info->mti_rr.rr_opcode];
2100 if (mr->mr_handler == NULL)
2103 rc = (*mr->mr_handler)(info, lhc);
2105 lprocfs_counter_incr(ptlrpc_req2svc(mdt_info_req(info))->srv_stats,
2106 PTLRPC_LAST_CNTR + mr->mr_extra_opc);