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.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2011, 2017, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
31 * lustre/mdt/mdt_reint.c
33 * Lustre Metadata Target (mdt) reintegration routines
35 * Author: Peter Braam <braam@clusterfs.com>
36 * Author: Andreas Dilger <adilger@clusterfs.com>
37 * Author: Phil Schwan <phil@clusterfs.com>
38 * Author: Huang Hua <huanghua@clusterfs.com>
39 * Author: Yury Umanets <umka@clusterfs.com>
42 #define DEBUG_SUBSYSTEM S_MDS
44 #include <lprocfs_status.h>
45 #include "mdt_internal.h"
46 #include <lustre_lmv.h>
48 static inline void mdt_reint_init_ma(struct mdt_thread_info *info,
51 ma->ma_need = MA_INODE;
56 * Get version of object by fid.
58 * Return real version or ENOENT_VERSION if object doesn't exist
60 static void mdt_obj_version_get(struct mdt_thread_info *info,
61 struct mdt_object *o, __u64 *version)
65 if (mdt_object_exists(o) && !mdt_object_remote(o) &&
66 !fid_is_obf(mdt_object_fid(o)))
67 *version = dt_version_get(info->mti_env, mdt_obj2dt(o));
69 *version = ENOENT_VERSION;
70 CDEBUG(D_INODE, "FID "DFID" version is %#llx\n",
71 PFID(mdt_object_fid(o)), *version);
75 * Check version is correct.
77 * Should be called only during replay.
79 static int mdt_version_check(struct ptlrpc_request *req,
80 __u64 version, int idx)
82 __u64 *pre_ver = lustre_msg_get_versions(req->rq_reqmsg);
85 if (!exp_connect_vbr(req->rq_export))
88 LASSERT(req_is_replay(req));
89 /** VBR: version is checked always because costs nothing */
90 LASSERT(idx < PTLRPC_NUM_VERSIONS);
91 /** Sanity check for malformed buffers */
92 if (pre_ver == NULL) {
93 CERROR("No versions in request buffer\n");
94 spin_lock(&req->rq_export->exp_lock);
95 req->rq_export->exp_vbr_failed = 1;
96 spin_unlock(&req->rq_export->exp_lock);
98 } else if (pre_ver[idx] != version) {
99 CDEBUG(D_INODE, "Version mismatch %#llx != %#llx\n",
100 pre_ver[idx], version);
101 spin_lock(&req->rq_export->exp_lock);
102 req->rq_export->exp_vbr_failed = 1;
103 spin_unlock(&req->rq_export->exp_lock);
110 * Save pre-versions in reply.
112 static void mdt_version_save(struct ptlrpc_request *req, __u64 version,
117 if (!exp_connect_vbr(req->rq_export))
120 LASSERT(!req_is_replay(req));
121 LASSERT(req->rq_repmsg != NULL);
122 reply_ver = lustre_msg_get_versions(req->rq_repmsg);
124 reply_ver[idx] = version;
128 * Save enoent version, it is needed when it is obvious that object doesn't
129 * exist, e.g. child during create.
131 static void mdt_enoent_version_save(struct mdt_thread_info *info, int idx)
133 /* save version of file name for replay, it must be ENOENT here */
134 if (!req_is_replay(mdt_info_req(info))) {
135 info->mti_ver[idx] = ENOENT_VERSION;
136 mdt_version_save(mdt_info_req(info), info->mti_ver[idx], idx);
141 * Get version from disk and save in reply buffer.
143 * Versions are saved in reply only during normal operations not replays.
145 void mdt_version_get_save(struct mdt_thread_info *info,
146 struct mdt_object *mto, int idx)
148 /* don't save versions during replay */
149 if (!req_is_replay(mdt_info_req(info))) {
150 mdt_obj_version_get(info, mto, &info->mti_ver[idx]);
151 mdt_version_save(mdt_info_req(info), info->mti_ver[idx], idx);
156 * Get version from disk and check it, no save in reply.
158 int mdt_version_get_check(struct mdt_thread_info *info,
159 struct mdt_object *mto, int idx)
161 /* only check versions during replay */
162 if (!req_is_replay(mdt_info_req(info)))
165 mdt_obj_version_get(info, mto, &info->mti_ver[idx]);
166 return mdt_version_check(mdt_info_req(info), info->mti_ver[idx], idx);
170 * Get version from disk and check if recovery or just save.
172 int mdt_version_get_check_save(struct mdt_thread_info *info,
173 struct mdt_object *mto, int idx)
177 mdt_obj_version_get(info, mto, &info->mti_ver[idx]);
178 if (req_is_replay(mdt_info_req(info)))
179 rc = mdt_version_check(mdt_info_req(info), info->mti_ver[idx],
182 mdt_version_save(mdt_info_req(info), info->mti_ver[idx], idx);
187 * Lookup with version checking.
189 * This checks version of 'name'. Many reint functions uses 'name' for child not
190 * FID, therefore we need to get object by name and check its version.
192 int mdt_lookup_version_check(struct mdt_thread_info *info,
193 struct mdt_object *p,
194 const struct lu_name *lname,
195 struct lu_fid *fid, int idx)
199 rc = mdo_lookup(info->mti_env, mdt_object_child(p), lname, fid,
201 /* Check version only during replay */
202 if (!req_is_replay(mdt_info_req(info)))
205 info->mti_ver[idx] = ENOENT_VERSION;
207 struct mdt_object *child;
209 child = mdt_object_find(info->mti_env, info->mti_mdt, fid);
210 if (likely(!IS_ERR(child))) {
211 mdt_obj_version_get(info, child, &info->mti_ver[idx]);
212 mdt_object_put(info->mti_env, child);
215 vbrc = mdt_version_check(mdt_info_req(info), info->mti_ver[idx], idx);
216 return vbrc ? vbrc : rc;
220 static int mdt_unlock_slaves(struct mdt_thread_info *mti,
221 struct mdt_object *obj,
222 struct ldlm_enqueue_info *einfo,
225 union ldlm_policy_data *policy = &mti->mti_policy;
226 struct mdt_lock_handle *lh = &mti->mti_lh[MDT_LH_LOCAL];
227 struct lustre_handle_array *slave_locks = einfo->ei_cbdata;
230 LASSERT(S_ISDIR(obj->mot_header.loh_attr));
231 LASSERT(slave_locks);
233 memset(policy, 0, sizeof(*policy));
234 policy->l_inodebits.bits = einfo->ei_inodebits;
235 mdt_lock_handle_init(lh);
236 mdt_lock_reg_init(lh, einfo->ei_mode);
237 for (i = 0; i < slave_locks->ha_count; i++) {
238 if (test_bit(i, (void *)slave_locks->ha_map))
239 lh->mlh_rreg_lh = slave_locks->ha_handles[i];
241 lh->mlh_reg_lh = slave_locks->ha_handles[i];
242 mdt_object_unlock(mti, NULL, lh, decref);
243 slave_locks->ha_handles[i].cookie = 0ull;
246 return mo_object_unlock(mti->mti_env, mdt_object_child(obj), einfo,
250 static inline int mdt_object_striped(struct mdt_thread_info *mti,
251 struct mdt_object *obj)
253 struct lu_device *bottom_dev;
254 struct lu_object *bottom_obj;
257 if (!S_ISDIR(obj->mot_header.loh_attr))
260 /* getxattr from bottom obj to avoid reading in shard FIDs */
261 bottom_dev = dt2lu_dev(mti->mti_mdt->mdt_bottom);
262 bottom_obj = lu_object_find_slice(mti->mti_env, bottom_dev,
263 mdt_object_fid(obj), NULL);
264 if (IS_ERR(bottom_obj))
265 return PTR_ERR(bottom_obj);
267 rc = dt_xattr_get(mti->mti_env, lu2dt(bottom_obj), &LU_BUF_NULL,
269 lu_object_put(mti->mti_env, bottom_obj);
271 return (rc > 0) ? 1 : (rc == -ENODATA) ? 0 : rc;
275 * Lock slave stripes if necessary, the lock handles of slave stripes
276 * will be stored in einfo->ei_cbdata.
278 static int mdt_lock_slaves(struct mdt_thread_info *mti, struct mdt_object *obj,
279 enum ldlm_mode mode, __u64 ibits,
280 struct ldlm_enqueue_info *einfo)
282 union ldlm_policy_data *policy = &mti->mti_policy;
284 LASSERT(S_ISDIR(obj->mot_header.loh_attr));
286 einfo->ei_type = LDLM_IBITS;
287 einfo->ei_mode = mode;
288 einfo->ei_cb_bl = mdt_remote_blocking_ast;
289 einfo->ei_cb_local_bl = mdt_blocking_ast;
290 einfo->ei_cb_cp = ldlm_completion_ast;
291 einfo->ei_enq_slave = 1;
292 einfo->ei_namespace = mti->mti_mdt->mdt_namespace;
293 einfo->ei_inodebits = ibits;
294 memset(policy, 0, sizeof(*policy));
295 policy->l_inodebits.bits = ibits;
297 return mo_object_lock(mti->mti_env, mdt_object_child(obj), NULL, einfo,
301 int mdt_reint_striped_lock(struct mdt_thread_info *info,
302 struct mdt_object *o,
303 struct mdt_lock_handle *lh,
305 struct ldlm_enqueue_info *einfo,
310 LASSERT(!mdt_object_remote(o));
312 memset(einfo, 0, sizeof(*einfo));
314 rc = mdt_reint_object_lock(info, o, lh, ibits, cos_incompat);
318 rc = mdt_object_striped(info, o);
321 mdt_object_unlock(info, o, lh, rc);
325 rc = mdt_lock_slaves(info, o, lh->mlh_reg_mode, ibits, einfo);
327 mdt_object_unlock(info, o, lh, rc);
328 if (rc == -EIO && OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_SLAVE_NAME))
335 void mdt_reint_striped_unlock(struct mdt_thread_info *info,
336 struct mdt_object *o,
337 struct mdt_lock_handle *lh,
338 struct ldlm_enqueue_info *einfo, int decref)
340 if (einfo->ei_cbdata)
341 mdt_unlock_slaves(info, o, einfo, decref);
342 mdt_object_unlock(info, o, lh, decref);
345 static int mdt_restripe(struct mdt_thread_info *info,
346 struct mdt_object *parent,
347 const struct lu_name *lname,
348 const struct lu_fid *tfid,
349 struct md_op_spec *spec,
352 struct mdt_device *mdt = info->mti_mdt;
353 struct lu_fid *fid = &info->mti_tmp_fid2;
354 struct ldlm_enqueue_info *einfo = &info->mti_einfo[0];
355 struct lmv_user_md *lum = spec->u.sp_ea.eadata;
356 struct lmv_mds_md_v1 *lmv;
357 struct mdt_object *child;
358 struct mdt_lock_handle *lhp;
359 struct mdt_lock_handle *lhc;
360 struct mdt_body *repbody;
364 if (!mdt->mdt_enable_dir_restripe)
368 lum->lum_hash_type |= cpu_to_le32(LMV_HASH_FLAG_FIXED);
370 rc = mdt_version_get_check_save(info, parent, 0);
374 lhp = &info->mti_lh[MDT_LH_PARENT];
375 mdt_lock_pdo_init(lhp, LCK_PW, lname);
376 rc = mdt_reint_object_lock(info, parent, lhp, MDS_INODELOCK_UPDATE,
381 rc = mdt_stripe_get(info, parent, ma, XATTR_NAME_LMV);
383 GOTO(unlock_parent, rc);
385 if (ma->ma_valid & MA_LMV) {
386 /* don't allow restripe if parent dir layout is changing */
387 lmv = &ma->ma_lmv->lmv_md_v1;
388 if (!lmv_is_sane2(lmv))
389 GOTO(unlock_parent, rc = -EBADF);
391 if (lmv_is_layout_changing(lmv))
392 GOTO(unlock_parent, rc = -EBUSY);
396 rc = mdt_lookup_version_check(info, parent, lname, fid, 1);
398 GOTO(unlock_parent, rc);
400 child = mdt_object_find(info->mti_env, mdt, fid);
402 GOTO(unlock_parent, rc = PTR_ERR(child));
404 if (!mdt_object_exists(child))
405 GOTO(out_child, rc = -ENOENT);
407 if (mdt_object_remote(child)) {
408 struct mdt_body *repbody;
410 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
412 GOTO(out_child, rc = -EPROTO);
414 repbody->mbo_fid1 = *fid;
415 repbody->mbo_valid |= (OBD_MD_FLID | OBD_MD_MDS);
416 GOTO(out_child, rc = -EREMOTE);
419 if (!S_ISDIR(lu_object_attr(&child->mot_obj)))
420 GOTO(out_child, rc = -ENOTDIR);
422 rc = mdt_stripe_get(info, child, ma, XATTR_NAME_LMV);
426 /* race with migrate? */
427 if ((ma->ma_valid & MA_LMV) &&
428 lmv_is_migrating(&ma->ma_lmv->lmv_md_v1))
429 GOTO(out_child, rc = -EBUSY);
432 lhc = &info->mti_lh[MDT_LH_CHILD];
433 mdt_lock_reg_init(lhc, LCK_EX);
435 /* enqueue object remote LOOKUP lock */
436 if (mdt_object_remote(parent)) {
437 rc = mdt_remote_object_lock(info, parent, fid,
440 MDS_INODELOCK_LOOKUP, false);
445 rc = mdt_reint_striped_lock(info, child, lhc, MDS_INODELOCK_FULL, einfo,
448 GOTO(unlock_child, rc);
450 tgt_vbr_obj_set(info->mti_env, mdt_obj2dt(child));
451 rc = mdt_version_get_check_save(info, child, 1);
453 GOTO(unlock_child, rc);
455 spin_lock(&mdt->mdt_restriper.mdr_lock);
456 if (child->mot_restriping) {
458 spin_unlock(&mdt->mdt_restriper.mdr_lock);
459 GOTO(unlock_child, rc = -EBUSY);
461 child->mot_restriping = 1;
462 spin_unlock(&mdt->mdt_restriper.mdr_lock);
465 rc = mdt_restripe_internal(info, parent, child, lname, fid, spec, ma);
467 GOTO(restriping_clear, rc);
469 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
471 GOTO(restriping_clear, rc = -EPROTO);
473 mdt_pack_attr2body(info, repbody, &ma->ma_attr, fid);
477 child->mot_restriping = 0;
479 mdt_reint_striped_unlock(info, child, lhc, einfo, rc);
481 mdt_object_put(info->mti_env, child);
483 mdt_object_unlock(info, parent, lhp, rc);
489 * VBR: we save three versions in reply:
490 * 0 - parent. Check that parent version is the same during replay.
491 * 1 - name. Version of 'name' if file exists with the same name or
492 * ENOENT_VERSION, it is needed because file may appear due to missed replays.
493 * 2 - child. Version of child by FID. Must be ENOENT. It is mostly sanity
496 static int mdt_create(struct mdt_thread_info *info)
498 struct mdt_device *mdt = info->mti_mdt;
499 struct mdt_object *parent;
500 struct mdt_object *child;
501 struct mdt_lock_handle *lh;
502 struct mdt_body *repbody;
503 struct md_attr *ma = &info->mti_attr;
504 struct mdt_reint_record *rr = &info->mti_rr;
505 struct md_op_spec *spec = &info->mti_spec;
506 bool restripe = false;
510 DEBUG_REQ(D_INODE, mdt_info_req(info),
511 "Create ("DNAME"->"DFID") in "DFID,
512 PNAME(&rr->rr_name), PFID(rr->rr_fid2), PFID(rr->rr_fid1));
514 if (!fid_is_md_operative(rr->rr_fid1))
517 if (S_ISDIR(ma->ma_attr.la_mode) &&
518 spec->u.sp_ea.eadata != NULL && spec->u.sp_ea.eadatalen != 0) {
519 const struct lmv_user_md *lum = spec->u.sp_ea.eadata;
520 struct lu_ucred *uc = mdt_ucred(info);
521 struct obd_export *exp = mdt_info_req(info)->rq_export;
523 /* Only new clients can create remote dir( >= 2.4) and
524 * striped dir(>= 2.6), old client will return -ENOTSUPP
526 if (!mdt_is_dne_client(exp))
529 if (le32_to_cpu(lum->lum_stripe_count) > 1) {
530 if (!mdt_is_striped_client(exp))
533 if (!mdt->mdt_enable_striped_dir)
535 } else if (!mdt->mdt_enable_remote_dir) {
539 if ((!(exp_connect_flags2(exp) & OBD_CONNECT2_CRUSH)) &&
540 (le32_to_cpu(lum->lum_hash_type) & LMV_HASH_TYPE_MASK) ==
544 if (!md_capable(uc, CAP_SYS_ADMIN) &&
545 uc->uc_gid != mdt->mdt_enable_remote_dir_gid &&
546 mdt->mdt_enable_remote_dir_gid != -1)
549 /* restripe if later found dir exists, MDS_OPEN_CREAT means
550 * this is create only, don't try restripe.
552 if (mdt->mdt_enable_dir_restripe &&
553 le32_to_cpu(lum->lum_stripe_offset) == LMV_OFFSET_DEFAULT &&
554 !(spec->sp_cr_flags & MDS_OPEN_CREAT))
558 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
560 parent = mdt_object_find(info->mti_env, info->mti_mdt, rr->rr_fid1);
562 RETURN(PTR_ERR(parent));
564 if (!mdt_object_exists(parent))
565 GOTO(put_parent, rc = -ENOENT);
568 * LU-10235: check if name exists locklessly first to avoid massive
569 * lock recalls on existing directories.
571 rc = mdt_lookup_version_check(info, parent, &rr->rr_name,
572 &info->mti_tmp_fid1, 1);
575 GOTO(put_parent, rc = -EEXIST);
577 rc = mdt_restripe(info, parent, &rr->rr_name, rr->rr_fid2, spec,
581 /* -ENOENT is expected here */
583 GOTO(put_parent, rc);
585 /* save version of file name for replay, it must be ENOENT here */
586 mdt_enoent_version_save(info, 1);
588 OBD_RACE(OBD_FAIL_MDS_CREATE_RACE);
590 lh = &info->mti_lh[MDT_LH_PARENT];
591 mdt_lock_pdo_init(lh, LCK_PW, &rr->rr_name);
592 rc = mdt_object_lock(info, parent, lh, MDS_INODELOCK_UPDATE);
594 GOTO(put_parent, rc);
596 if (!mdt_object_remote(parent)) {
597 rc = mdt_version_get_check_save(info, parent, 0);
599 GOTO(unlock_parent, rc);
602 child = mdt_object_new(info->mti_env, mdt, rr->rr_fid2);
603 if (unlikely(IS_ERR(child)))
604 GOTO(unlock_parent, rc = PTR_ERR(child));
606 ma->ma_need = MA_INODE;
609 mdt_fail_write(info->mti_env, info->mti_mdt->mdt_bottom,
610 OBD_FAIL_MDS_REINT_CREATE_WRITE);
612 /* Version of child will be updated on disk. */
613 tgt_vbr_obj_set(info->mti_env, mdt_obj2dt(child));
614 rc = mdt_version_get_check_save(info, child, 2);
619 * Do not perform lookup sanity check. We know that name does
622 info->mti_spec.sp_cr_lookup = 0;
623 info->mti_spec.sp_feat = &dt_directory_features;
625 rc = mdo_create(info->mti_env, mdt_object_child(parent), &rr->rr_name,
626 mdt_object_child(child), &info->mti_spec, ma);
628 rc = mdt_attr_get_complex(info, child, ma);
634 * On DNE, we need to eliminate dependey between 'mkdir a' and
635 * 'mkdir a/b' if b is a striped directory, to achieve this, two
636 * things are done below:
637 * 1. save child and slaves lock.
638 * 2. if the child is a striped directory, relock parent so to
639 * compare against with COS locks to ensure parent was
642 if (mdt_slc_is_enabled(mdt) && S_ISDIR(ma->ma_attr.la_mode)) {
643 struct mdt_lock_handle *lhc;
644 struct ldlm_enqueue_info *einfo = &info->mti_einfo[0];
647 rc = mdt_object_striped(info, child);
653 if (!mdt_object_remote(parent)) {
654 mdt_object_unlock(info, parent, lh, 1);
655 mdt_lock_pdo_init(lh, LCK_PW, &rr->rr_name);
656 rc = mdt_reint_object_lock(info, parent, lh,
657 MDS_INODELOCK_UPDATE,
664 lhc = &info->mti_lh[MDT_LH_CHILD];
665 mdt_lock_handle_init(lhc);
666 mdt_lock_reg_init(lhc, LCK_PW);
667 rc = mdt_reint_striped_lock(info, child, lhc,
668 MDS_INODELOCK_UPDATE, einfo,
673 mdt_reint_striped_unlock(info, child, lhc, einfo, rc);
676 /* Return fid & attr to client. */
677 if (ma->ma_valid & MA_INODE)
678 mdt_pack_attr2body(info, repbody, &ma->ma_attr,
679 mdt_object_fid(child));
682 mdt_object_put(info->mti_env, child);
684 mdt_object_unlock(info, parent, lh, rc);
686 mdt_object_put(info->mti_env, parent);
690 static int mdt_attr_set(struct mdt_thread_info *info, struct mdt_object *mo,
693 struct mdt_lock_handle *lh;
694 int do_vbr = ma->ma_attr.la_valid &
695 (LA_MODE | LA_UID | LA_GID | LA_PROJID | LA_FLAGS);
696 __u64 lockpart = MDS_INODELOCK_UPDATE;
697 struct ldlm_enqueue_info *einfo = &info->mti_einfo[0];
702 rc = mdt_object_striped(info, mo);
708 lh = &info->mti_lh[MDT_LH_PARENT];
709 mdt_lock_reg_init(lh, LCK_PW);
711 /* Even though the new MDT will grant PERM lock to the old
712 * client, but the old client will almost ignore that during
713 * So it needs to revoke both LOOKUP and PERM lock here, so
714 * both new and old client can cancel the dcache
716 if (ma->ma_attr.la_valid & (LA_MODE|LA_UID|LA_GID))
717 lockpart |= MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM;
719 rc = mdt_reint_striped_lock(info, mo, lh, lockpart, einfo,
724 /* all attrs are packed into mti_attr in unpack_setattr */
725 mdt_fail_write(info->mti_env, info->mti_mdt->mdt_bottom,
726 OBD_FAIL_MDS_REINT_SETATTR_WRITE);
728 /* VBR: update version if attr changed are important for recovery */
730 /* update on-disk version of changed object */
731 tgt_vbr_obj_set(info->mti_env, mdt_obj2dt(mo));
732 rc = mdt_version_get_check_save(info, mo, 0);
734 GOTO(out_unlock, rc);
737 /* Ensure constant striping during chown(). See LU-2789. */
738 if (ma->ma_attr.la_valid & (LA_UID|LA_GID|LA_PROJID))
739 mutex_lock(&mo->mot_lov_mutex);
741 /* all attrs are packed into mti_attr in unpack_setattr */
742 rc = mo_attr_set(info->mti_env, mdt_object_child(mo), ma);
744 if (ma->ma_attr.la_valid & (LA_UID|LA_GID|LA_PROJID))
745 mutex_unlock(&mo->mot_lov_mutex);
748 GOTO(out_unlock, rc);
749 mdt_dom_obj_lvb_update(info->mti_env, mo, false);
752 mdt_reint_striped_unlock(info, mo, lh, einfo, rc);
757 * Check HSM flags and add HS_DIRTY flag if relevant.
759 * A file could be set dirty only if it has a copy in the backend (HS_EXISTS)
760 * and is not RELEASED.
762 int mdt_add_dirty_flag(struct mdt_thread_info *info, struct mdt_object *mo,
765 struct lu_ucred *uc = mdt_ucred(info);
770 /* If the file was modified, add the dirty flag */
771 ma->ma_need = MA_HSM;
772 rc = mdt_attr_get_complex(info, mo, ma);
774 CERROR("file attribute read error for "DFID": %d.\n",
775 PFID(mdt_object_fid(mo)), rc);
779 /* If an up2date copy exists in the backend, add dirty flag */
780 if ((ma->ma_valid & MA_HSM) && (ma->ma_hsm.mh_flags & HS_EXISTS)
781 && !(ma->ma_hsm.mh_flags & (HS_DIRTY|HS_RELEASED))) {
782 ma->ma_hsm.mh_flags |= HS_DIRTY;
784 /* Bump cap so that closes from non-owner writers can
785 * set the HSM state to dirty.
787 cap_saved = uc->uc_cap;
788 uc->uc_cap |= MD_CAP_TO_MASK(CAP_FOWNER);
789 rc = mdt_hsm_attr_set(info, mo, &ma->ma_hsm);
790 uc->uc_cap = cap_saved;
792 CERROR("file attribute change error for "DFID": %d\n",
793 PFID(mdt_object_fid(mo)), rc);
799 static int mdt_reint_setattr(struct mdt_thread_info *info,
800 struct mdt_lock_handle *lhc)
802 struct mdt_device *mdt = info->mti_mdt;
803 struct md_attr *ma = &info->mti_attr;
804 struct mdt_reint_record *rr = &info->mti_rr;
805 struct ptlrpc_request *req = mdt_info_req(info);
806 struct mdt_object *mo;
807 struct mdt_body *repbody;
808 ktime_t kstart = ktime_get();
812 DEBUG_REQ(D_INODE, req, "setattr "DFID" %x", PFID(rr->rr_fid1),
813 (unsigned int)ma->ma_attr.la_valid);
815 if (info->mti_dlm_req)
816 ldlm_request_cancel(req, info->mti_dlm_req, 0, LATF_SKIP);
818 OBD_RACE(OBD_FAIL_PTLRPC_RESEND_RACE);
820 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
821 mo = mdt_object_find(info->mti_env, mdt, rr->rr_fid1);
823 GOTO(out, rc = PTR_ERR(mo));
825 if (!mdt_object_exists(mo))
826 GOTO(out_put, rc = -ENOENT);
828 if (mdt_object_remote(mo))
829 GOTO(out_put, rc = -EREMOTE);
831 ma->ma_enable_chprojid_gid = mdt->mdt_enable_chprojid_gid;
832 /* revoke lease lock if size is going to be changed */
833 if (unlikely(ma->ma_attr.la_valid & LA_SIZE &&
834 !(ma->ma_attr_flags & MDS_TRUNC_KEEP_LEASE) &&
835 atomic_read(&mo->mot_lease_count) > 0)) {
836 down_read(&mo->mot_open_sem);
838 if (atomic_read(&mo->mot_lease_count) > 0) { /* lease exists */
839 lhc = &info->mti_lh[MDT_LH_LOCAL];
840 mdt_lock_reg_init(lhc, LCK_CW);
842 rc = mdt_object_lock(info, mo, lhc, MDS_INODELOCK_OPEN);
844 up_read(&mo->mot_open_sem);
848 /* revoke lease lock */
849 mdt_object_unlock(info, mo, lhc, 1);
851 up_read(&mo->mot_open_sem);
854 if (ma->ma_attr.la_valid & LA_SIZE || rr->rr_flags & MRF_OPEN_TRUNC) {
855 /* Check write access for the O_TRUNC case */
856 if (mdt_write_read(mo) < 0)
857 GOTO(out_put, rc = -ETXTBSY);
859 /* LU-10286: compatibility check for FLR.
860 * Please check the comment in mdt_finish_open() for details
862 if (!exp_connect_flr(info->mti_exp) ||
863 !exp_connect_overstriping(info->mti_exp)) {
864 rc = mdt_big_xattr_get(info, mo, XATTR_NAME_LOV);
865 if (rc < 0 && rc != -ENODATA)
868 if (!exp_connect_flr(info->mti_exp)) {
870 mdt_lmm_is_flr(info->mti_big_lmm))
871 GOTO(out_put, rc = -EOPNOTSUPP);
874 if (!exp_connect_overstriping(info->mti_exp)) {
876 mdt_lmm_is_overstriping(info->mti_big_lmm))
877 GOTO(out_put, rc = -EOPNOTSUPP);
881 /* For truncate, the file size sent from client
882 * is believable, but the blocks are incorrect,
883 * which makes the block size in LSOM attribute
884 * inconsisent with the real block size.
886 rc = mdt_lsom_update(info, mo, true);
891 if ((ma->ma_valid & MA_INODE) && ma->ma_attr.la_valid) {
892 if (ma->ma_valid & MA_LOV)
893 GOTO(out_put, rc = -EPROTO);
895 /* MDT supports FMD for regular files due to Data-on-MDT */
896 if (S_ISREG(lu_object_attr(&mo->mot_obj)) &&
897 ma->ma_attr.la_valid & (LA_ATIME | LA_MTIME | LA_CTIME)) {
898 tgt_fmd_update(info->mti_exp, mdt_object_fid(mo),
901 if (ma->ma_attr.la_valid & LA_MTIME) {
902 rc = mdt_attr_get_pfid(info, mo, &ma->ma_pfid);
904 ma->ma_valid |= MA_PFID;
908 rc = mdt_attr_set(info, mo, ma);
911 } else if ((ma->ma_valid & (MA_LOV | MA_LMV)) &&
912 (ma->ma_valid & MA_INODE)) {
913 struct lu_buf *buf = &info->mti_buf;
914 struct lu_ucred *uc = mdt_ucred(info);
915 struct mdt_lock_handle *lh;
917 __u64 lockpart = MDS_INODELOCK_XATTR;
919 /* reject if either remote or striped dir is disabled */
920 if (ma->ma_valid & MA_LMV) {
921 if (!mdt->mdt_enable_remote_dir ||
922 !mdt->mdt_enable_striped_dir)
923 GOTO(out_put, rc = -EPERM);
925 if (!md_capable(uc, CAP_SYS_ADMIN) &&
926 uc->uc_gid != mdt->mdt_enable_remote_dir_gid &&
927 mdt->mdt_enable_remote_dir_gid != -1)
928 GOTO(out_put, rc = -EPERM);
931 if (!S_ISDIR(lu_object_attr(&mo->mot_obj)))
932 GOTO(out_put, rc = -ENOTDIR);
934 if (ma->ma_attr.la_valid != 0)
935 GOTO(out_put, rc = -EPROTO);
937 if (ma->ma_valid & MA_LOV) {
938 buf->lb_buf = ma->ma_lmm;
939 buf->lb_len = ma->ma_lmm_size;
940 name = XATTR_NAME_LOV;
942 struct lmv_user_md *lmu = &ma->ma_lmv->lmv_user_md;
945 buf->lb_len = ma->ma_lmv_size;
946 name = XATTR_NAME_DEFAULT_LMV;
947 /* force client to update dir default layout */
948 lockpart |= MDS_INODELOCK_LOOKUP;
951 lh = &info->mti_lh[MDT_LH_PARENT];
952 mdt_lock_reg_init(lh, LCK_PW);
954 rc = mdt_object_lock(info, mo, lh, lockpart);
958 rc = mo_xattr_set(info->mti_env, mdt_object_child(mo), buf,
961 mdt_object_unlock(info, mo, lh, rc);
965 GOTO(out_put, rc = -EPROTO);
968 /* If file data is modified, add the dirty flag */
969 if (ma->ma_attr_flags & MDS_DATA_MODIFIED)
970 rc = mdt_add_dirty_flag(info, mo, ma);
972 ma->ma_need = MA_INODE;
974 rc = mdt_attr_get_complex(info, mo, ma);
978 mdt_pack_attr2body(info, repbody, &ma->ma_attr, mdt_object_fid(mo));
982 mdt_object_put(info->mti_env, mo);
985 mdt_counter_incr(req, LPROC_MDT_SETATTR,
986 ktime_us_delta(ktime_get(), kstart));
988 mdt_client_compatibility(info);
989 rc2 = mdt_fix_reply(info);
995 static int mdt_reint_create(struct mdt_thread_info *info,
996 struct mdt_lock_handle *lhc)
998 struct ptlrpc_request *req = mdt_info_req(info);
999 ktime_t kstart = ktime_get();
1003 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_REINT_CREATE))
1004 RETURN(err_serious(-ESTALE));
1006 if (info->mti_dlm_req)
1007 ldlm_request_cancel(mdt_info_req(info),
1008 info->mti_dlm_req, 0, LATF_SKIP);
1010 if (!lu_name_is_valid(&info->mti_rr.rr_name))
1013 switch (info->mti_attr.ma_attr.la_mode & S_IFMT) {
1023 CERROR("%s: Unsupported mode %o\n",
1024 mdt_obd_name(info->mti_mdt),
1025 info->mti_attr.ma_attr.la_mode);
1026 RETURN(err_serious(-EOPNOTSUPP));
1029 rc = mdt_create(info);
1031 if ((info->mti_attr.ma_attr.la_mode & S_IFMT) == S_IFDIR)
1032 mdt_counter_incr(req, LPROC_MDT_MKDIR,
1033 ktime_us_delta(ktime_get(), kstart));
1035 /* Special file should stay on the same node as parent*/
1036 mdt_counter_incr(req, LPROC_MDT_MKNOD,
1037 ktime_us_delta(ktime_get(), kstart));
1044 * VBR: save parent version in reply and child version getting by its name.
1045 * Version of child is getting and checking during its lookup. If
1047 static int mdt_reint_unlink(struct mdt_thread_info *info,
1048 struct mdt_lock_handle *lhc)
1050 struct mdt_reint_record *rr = &info->mti_rr;
1051 struct ptlrpc_request *req = mdt_info_req(info);
1052 struct md_attr *ma = &info->mti_attr;
1053 struct lu_fid *child_fid = &info->mti_tmp_fid1;
1054 struct mdt_object *mp;
1055 struct mdt_object *mc;
1056 struct mdt_lock_handle *parent_lh;
1057 struct mdt_lock_handle *child_lh;
1058 struct ldlm_enqueue_info *einfo = &info->mti_einfo[0];
1060 bool cos_incompat = false;
1062 ktime_t kstart = ktime_get();
1066 DEBUG_REQ(D_INODE, req, "unlink "DFID"/"DNAME"", PFID(rr->rr_fid1),
1067 PNAME(&rr->rr_name));
1069 if (info->mti_dlm_req)
1070 ldlm_request_cancel(req, info->mti_dlm_req, 0, LATF_SKIP);
1072 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_REINT_UNLINK))
1073 RETURN(err_serious(-ENOENT));
1075 if (!fid_is_md_operative(rr->rr_fid1))
1078 mp = mdt_object_find(info->mti_env, info->mti_mdt, rr->rr_fid1);
1080 RETURN(PTR_ERR(mp));
1082 if (mdt_object_remote(mp)) {
1083 cos_incompat = true;
1085 rc = mdt_version_get_check_save(info, mp, 0);
1087 GOTO(put_parent, rc);
1090 OBD_RACE(OBD_FAIL_MDS_REINT_OPEN);
1091 OBD_RACE(OBD_FAIL_MDS_REINT_OPEN2);
1093 parent_lh = &info->mti_lh[MDT_LH_PARENT];
1094 mdt_lock_pdo_init(parent_lh, LCK_PW, &rr->rr_name);
1095 rc = mdt_reint_object_lock(info, mp, parent_lh, MDS_INODELOCK_UPDATE,
1098 GOTO(put_parent, rc);
1100 /* lookup child object along with version checking */
1101 fid_zero(child_fid);
1102 rc = mdt_lookup_version_check(info, mp, &rr->rr_name, child_fid, 1);
1104 /* Name might not be able to find during resend of
1105 * remote unlink, considering following case.
1106 * dir_A is a remote directory, the name entry of
1107 * dir_A is on MDT0, the directory is on MDT1,
1109 * 1. client sends unlink req to MDT1.
1110 * 2. MDT1 sends name delete update to MDT0.
1111 * 3. name entry is being deleted in MDT0 synchronously.
1112 * 4. MDT1 is restarted.
1113 * 5. client resends unlink req to MDT1. So it can not
1114 * find the name entry on MDT0 anymore.
1115 * In this case, MDT1 only needs to destory the local
1118 if (mdt_object_remote(mp) && rc == -ENOENT &&
1119 !fid_is_zero(rr->rr_fid2) &&
1120 lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT) {
1122 *child_fid = *rr->rr_fid2;
1124 GOTO(unlock_parent, rc);
1128 if (!fid_is_md_operative(child_fid))
1129 GOTO(unlock_parent, rc = -EPERM);
1131 /* We will lock the child regardless it is local or remote. No harm. */
1132 mc = mdt_object_find(info->mti_env, info->mti_mdt, child_fid);
1134 GOTO(unlock_parent, rc = PTR_ERR(mc));
1136 if (!cos_incompat) {
1137 rc = mdt_object_striped(info, mc);
1139 GOTO(put_child, rc);
1143 mdt_object_put(info->mti_env, mc);
1144 mdt_object_unlock(info, mp, parent_lh, -EAGAIN);
1149 child_lh = &info->mti_lh[MDT_LH_CHILD];
1150 mdt_lock_reg_init(child_lh, LCK_EX);
1151 if (info->mti_spec.sp_rm_entry) {
1152 struct lu_ucred *uc = mdt_ucred(info);
1154 if (!mdt_is_dne_client(req->rq_export))
1155 /* Return -ENOTSUPP for old client */
1156 GOTO(put_child, rc = -ENOTSUPP);
1158 if (!md_capable(uc, CAP_SYS_ADMIN))
1159 GOTO(put_child, rc = -EPERM);
1161 ma->ma_need = MA_INODE;
1163 rc = mdo_unlink(info->mti_env, mdt_object_child(mp),
1164 NULL, &rr->rr_name, ma, no_name);
1165 GOTO(put_child, rc);
1168 if (mdt_object_remote(mc)) {
1169 struct mdt_body *repbody;
1171 if (!fid_is_zero(rr->rr_fid2)) {
1172 CDEBUG(D_INFO, "%s: name "DNAME" cannot find "DFID"\n",
1173 mdt_obd_name(info->mti_mdt),
1174 PNAME(&rr->rr_name), PFID(mdt_object_fid(mc)));
1175 GOTO(put_child, rc = -ENOENT);
1177 CDEBUG(D_INFO, "%s: name "DNAME": "DFID" is on another MDT\n",
1178 mdt_obd_name(info->mti_mdt),
1179 PNAME(&rr->rr_name), PFID(mdt_object_fid(mc)));
1181 if (!mdt_is_dne_client(req->rq_export))
1182 /* Return -ENOTSUPP for old client */
1183 GOTO(put_child, rc = -ENOTSUPP);
1185 /* Revoke the LOOKUP lock of the remote object granted by
1186 * this MDT. Since the unlink will happen on another MDT,
1187 * it will release the LOOKUP lock right away. Then What
1188 * would happen if another client try to grab the LOOKUP
1189 * lock at the same time with unlink XXX
1191 mdt_object_lock(info, mc, child_lh, MDS_INODELOCK_LOOKUP);
1192 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1193 LASSERT(repbody != NULL);
1194 repbody->mbo_fid1 = *mdt_object_fid(mc);
1195 repbody->mbo_valid |= (OBD_MD_FLID | OBD_MD_MDS);
1196 GOTO(unlock_child, rc = -EREMOTE);
1198 /* We used to acquire MDS_INODELOCK_FULL here but we can't do
1199 * this now because a running HSM restore on the child (unlink
1200 * victim) will hold the layout lock. See LU-4002.
1202 lock_ibits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE;
1203 if (mdt_object_remote(mp)) {
1204 /* Enqueue lookup lock from parent MDT */
1205 rc = mdt_remote_object_lock(info, mp, mdt_object_fid(mc),
1206 &child_lh->mlh_rreg_lh,
1207 child_lh->mlh_rreg_mode,
1208 MDS_INODELOCK_LOOKUP, false);
1210 GOTO(put_child, rc);
1212 lock_ibits &= ~MDS_INODELOCK_LOOKUP;
1215 rc = mdt_reint_striped_lock(info, mc, child_lh, lock_ibits, einfo,
1218 GOTO(put_child, rc);
1221 * Now we can only make sure we need MA_INODE, in mdd layer, will check
1222 * whether need MA_LOV and MA_COOKIE.
1224 ma->ma_need = MA_INODE;
1227 mdt_fail_write(info->mti_env, info->mti_mdt->mdt_bottom,
1228 OBD_FAIL_MDS_REINT_UNLINK_WRITE);
1229 /* save version when object is locked */
1230 mdt_version_get_save(info, mc, 1);
1232 mutex_lock(&mc->mot_lov_mutex);
1234 rc = mdo_unlink(info->mti_env, mdt_object_child(mp),
1235 mdt_object_child(mc), &rr->rr_name, ma, no_name);
1237 mutex_unlock(&mc->mot_lov_mutex);
1239 GOTO(unlock_child, rc);
1241 if (!lu_object_is_dying(&mc->mot_header)) {
1242 rc = mdt_attr_get_complex(info, mc, ma);
1245 } else if (mdt_dom_check_for_discard(info, mc)) {
1246 mdt_dom_discard_data(info, mc);
1248 mdt_handle_last_unlink(info, mc, ma);
1251 if (ma->ma_valid & MA_INODE) {
1252 switch (ma->ma_attr.la_mode & S_IFMT) {
1254 mdt_counter_incr(req, LPROC_MDT_RMDIR,
1255 ktime_us_delta(ktime_get(), kstart));
1263 mdt_counter_incr(req, LPROC_MDT_UNLINK,
1264 ktime_us_delta(ktime_get(), kstart));
1267 LASSERTF(0, "bad file type %o unlinking\n",
1268 ma->ma_attr.la_mode);
1275 mdt_reint_striped_unlock(info, mc, child_lh, einfo, rc);
1277 mdt_object_put(info->mti_env, mc);
1279 mdt_object_unlock(info, mp, parent_lh, rc);
1281 mdt_object_put(info->mti_env, mp);
1282 CFS_RACE_WAKEUP(OBD_FAIL_OBD_ZERO_NLINK_RACE);
1287 * VBR: save versions in reply: 0 - parent; 1 - child by fid; 2 - target by
1290 static int mdt_reint_link(struct mdt_thread_info *info,
1291 struct mdt_lock_handle *lhc)
1293 struct mdt_reint_record *rr = &info->mti_rr;
1294 struct ptlrpc_request *req = mdt_info_req(info);
1295 struct md_attr *ma = &info->mti_attr;
1296 struct mdt_object *ms;
1297 struct mdt_object *mp;
1298 struct mdt_lock_handle *lhs;
1299 struct mdt_lock_handle *lhp;
1300 ktime_t kstart = ktime_get();
1305 DEBUG_REQ(D_INODE, req, "link "DFID" to "DFID"/"DNAME,
1306 PFID(rr->rr_fid1), PFID(rr->rr_fid2), PNAME(&rr->rr_name));
1308 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_REINT_LINK))
1309 RETURN(err_serious(-ENOENT));
1311 if (OBD_FAIL_PRECHECK(OBD_FAIL_PTLRPC_RESEND_RACE) ||
1312 OBD_FAIL_PRECHECK(OBD_FAIL_PTLRPC_ENQ_RESEND)) {
1313 req->rq_no_reply = 1;
1314 RETURN(err_serious(-ENOENT));
1317 if (info->mti_dlm_req)
1318 ldlm_request_cancel(req, info->mti_dlm_req, 0, LATF_SKIP);
1320 /* Invalid case so return error immediately instead of
1323 if (lu_fid_eq(rr->rr_fid1, rr->rr_fid2))
1326 if (!fid_is_md_operative(rr->rr_fid1) ||
1327 !fid_is_md_operative(rr->rr_fid2))
1330 /* step 1: find target parent dir */
1331 mp = mdt_object_find(info->mti_env, info->mti_mdt, rr->rr_fid2);
1333 RETURN(PTR_ERR(mp));
1335 rc = mdt_version_get_check_save(info, mp, 0);
1337 GOTO(put_parent, rc);
1339 /* step 2: find source */
1340 ms = mdt_object_find(info->mti_env, info->mti_mdt, rr->rr_fid1);
1342 GOTO(put_parent, rc = PTR_ERR(ms));
1344 if (!mdt_object_exists(ms)) {
1345 CDEBUG(D_INFO, "%s: "DFID" does not exist.\n",
1346 mdt_obd_name(info->mti_mdt), PFID(rr->rr_fid1));
1347 GOTO(put_source, rc = -ENOENT);
1350 cos_incompat = (mdt_object_remote(mp) || mdt_object_remote(ms));
1352 lhp = &info->mti_lh[MDT_LH_PARENT];
1353 mdt_lock_pdo_init(lhp, LCK_PW, &rr->rr_name);
1354 rc = mdt_reint_object_lock(info, mp, lhp, MDS_INODELOCK_UPDATE,
1357 GOTO(put_source, rc);
1359 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_RENAME3, 5);
1361 lhs = &info->mti_lh[MDT_LH_CHILD];
1362 mdt_lock_reg_init(lhs, LCK_EX);
1363 rc = mdt_reint_object_lock(info, ms, lhs,
1364 MDS_INODELOCK_UPDATE | MDS_INODELOCK_XATTR,
1367 GOTO(unlock_parent, rc);
1369 /* step 3: link it */
1370 mdt_fail_write(info->mti_env, info->mti_mdt->mdt_bottom,
1371 OBD_FAIL_MDS_REINT_LINK_WRITE);
1373 tgt_vbr_obj_set(info->mti_env, mdt_obj2dt(ms));
1374 rc = mdt_version_get_check_save(info, ms, 1);
1376 GOTO(unlock_source, rc);
1378 /** check target version by name during replay */
1379 rc = mdt_lookup_version_check(info, mp, &rr->rr_name,
1380 &info->mti_tmp_fid1, 2);
1381 if (rc != 0 && rc != -ENOENT)
1382 GOTO(unlock_source, rc);
1383 /* save version of file name for replay, it must be ENOENT here */
1384 if (!req_is_replay(mdt_info_req(info))) {
1385 if (rc != -ENOENT) {
1386 CDEBUG(D_INFO, "link target "DNAME" existed!\n",
1387 PNAME(&rr->rr_name));
1388 GOTO(unlock_source, rc = -EEXIST);
1390 info->mti_ver[2] = ENOENT_VERSION;
1391 mdt_version_save(mdt_info_req(info), info->mti_ver[2], 2);
1394 rc = mdo_link(info->mti_env, mdt_object_child(mp),
1395 mdt_object_child(ms), &rr->rr_name, ma);
1398 mdt_counter_incr(req, LPROC_MDT_LINK,
1399 ktime_us_delta(ktime_get(), kstart));
1403 mdt_object_unlock(info, ms, lhs, rc);
1405 mdt_object_unlock(info, mp, lhp, rc);
1407 mdt_object_put(info->mti_env, ms);
1409 mdt_object_put(info->mti_env, mp);
1413 * lock the part of the directory according to the hash of the name
1414 * (lh->mlh_pdo_hash) in parallel directory lock.
1416 static int mdt_pdir_hash_lock(struct mdt_thread_info *info,
1417 struct mdt_lock_handle *lh,
1418 struct mdt_object *obj, __u64 ibits,
1421 struct ldlm_res_id *res = &info->mti_res_id;
1422 struct ldlm_namespace *ns = info->mti_mdt->mdt_namespace;
1423 union ldlm_policy_data *policy = &info->mti_policy;
1424 __u64 dlmflags = LDLM_FL_LOCAL_ONLY | LDLM_FL_ATOMIC_CB;
1428 * Finish res_id initializing by name hash marking part of
1429 * directory which is taking modification.
1431 LASSERT(lh->mlh_pdo_hash != 0);
1432 fid_build_pdo_res_name(mdt_object_fid(obj), lh->mlh_pdo_hash, res);
1433 memset(policy, 0, sizeof(*policy));
1434 policy->l_inodebits.bits = ibits;
1436 (lh->mlh_reg_mode == LCK_PW || lh->mlh_reg_mode == LCK_EX))
1437 dlmflags |= LDLM_FL_COS_INCOMPAT;
1439 * Use LDLM_FL_LOCAL_ONLY for this lock. We do not know yet if it is
1440 * going to be sent to client. If it is - mdt_intent_policy() path will
1441 * fix it up and turn FL_LOCAL flag off.
1443 rc = mdt_fid_lock(info->mti_env, ns, &lh->mlh_reg_lh, lh->mlh_reg_mode,
1444 policy, res, dlmflags,
1445 &info->mti_exp->exp_handle.h_cookie);
1450 * Get BFL lock for rename or migrate process.
1452 static int mdt_rename_lock(struct mdt_thread_info *info,
1453 struct lustre_handle *lh)
1458 if (mdt_seq_site(info->mti_mdt)->ss_node_id != 0) {
1459 struct lu_fid *fid = &info->mti_tmp_fid1;
1460 struct mdt_object *obj;
1462 /* XXX, right now, it has to use object API to
1463 * enqueue lock cross MDT, so it will enqueue
1464 * rename lock(with LUSTRE_BFL_FID) by root object
1467 obj = mdt_object_find(info->mti_env, info->mti_mdt, fid);
1469 RETURN(PTR_ERR(obj));
1471 rc = mdt_remote_object_lock(info, obj,
1472 &LUSTRE_BFL_FID, lh,
1474 MDS_INODELOCK_UPDATE, false);
1475 mdt_object_put(info->mti_env, obj);
1477 struct ldlm_namespace *ns = info->mti_mdt->mdt_namespace;
1478 union ldlm_policy_data *policy = &info->mti_policy;
1479 struct ldlm_res_id *res_id = &info->mti_res_id;
1482 fid_build_reg_res_name(&LUSTRE_BFL_FID, res_id);
1483 memset(policy, 0, sizeof(*policy));
1484 policy->l_inodebits.bits = MDS_INODELOCK_UPDATE;
1485 flags = LDLM_FL_LOCAL_ONLY | LDLM_FL_ATOMIC_CB;
1486 rc = ldlm_cli_enqueue_local(info->mti_env, ns, res_id,
1487 LDLM_IBITS, policy, LCK_EX, &flags,
1489 ldlm_completion_ast, NULL, NULL, 0,
1491 &info->mti_exp->exp_handle.h_cookie,
1498 static void mdt_rename_unlock(struct lustre_handle *lh)
1501 LASSERT(lustre_handle_is_used(lh));
1502 /* Cancel the single rename lock right away */
1503 ldlm_lock_decref_and_cancel(lh, LCK_EX);
1507 static struct mdt_object *mdt_parent_find_check(struct mdt_thread_info *info,
1508 const struct lu_fid *fid,
1511 struct mdt_object *dir;
1515 dir = mdt_object_find(info->mti_env, info->mti_mdt, fid);
1519 /* check early, the real version will be saved after locking */
1520 rc = mdt_version_get_check(info, dir, idx);
1524 if (!mdt_object_exists(dir))
1525 GOTO(out_put, rc = -ENOENT);
1527 if (!S_ISDIR(lu_object_attr(&dir->mot_obj)))
1528 GOTO(out_put, rc = -ENOTDIR);
1532 mdt_object_put(info->mti_env, dir);
1537 * in case obj is remote obj on its parent, revoke LOOKUP lock,
1538 * herein we don't really check it, just do revoke.
1540 int mdt_revoke_remote_lookup_lock(struct mdt_thread_info *info,
1541 struct mdt_object *pobj,
1542 struct mdt_object *obj)
1544 struct mdt_lock_handle *lh = &info->mti_lh[MDT_LH_LOCAL];
1547 mdt_lock_handle_init(lh);
1548 mdt_lock_reg_init(lh, LCK_EX);
1550 if (mdt_object_remote(pobj)) {
1551 /* don't bother to check if pobj and obj are on the same MDT. */
1552 rc = mdt_remote_object_lock(info, pobj, mdt_object_fid(obj),
1553 &lh->mlh_rreg_lh, LCK_EX,
1554 MDS_INODELOCK_LOOKUP, false);
1555 } else if (mdt_object_remote(obj)) {
1556 struct ldlm_res_id *res = &info->mti_res_id;
1557 union ldlm_policy_data *policy = &info->mti_policy;
1558 __u64 dlmflags = LDLM_FL_LOCAL_ONLY | LDLM_FL_ATOMIC_CB |
1559 LDLM_FL_COS_INCOMPAT;
1561 fid_build_reg_res_name(mdt_object_fid(obj), res);
1562 memset(policy, 0, sizeof(*policy));
1563 policy->l_inodebits.bits = MDS_INODELOCK_LOOKUP;
1564 rc = mdt_fid_lock(info->mti_env, info->mti_mdt->mdt_namespace,
1565 &lh->mlh_reg_lh, LCK_EX, policy, res,
1568 /* do nothing if both are local */
1576 * TODO, currently we don't save this lock because there is no place to
1577 * hold this lock handle, but to avoid race we need to save this lock.
1579 mdt_object_unlock(info, NULL, lh, 1);
1585 * operation may takes locks of linkea, or directory stripes, group them in
1588 struct mdt_sub_lock {
1589 struct mdt_object *msl_obj;
1590 struct mdt_lock_handle msl_lh;
1591 struct list_head msl_linkage;
1594 static void mdt_unlock_list(struct mdt_thread_info *info,
1595 struct list_head *list, int decref)
1597 struct mdt_sub_lock *msl;
1598 struct mdt_sub_lock *tmp;
1600 list_for_each_entry_safe(msl, tmp, list, msl_linkage) {
1601 mdt_object_unlock_put(info, msl->msl_obj, &msl->msl_lh, decref);
1602 list_del(&msl->msl_linkage);
1607 static inline void mdt_migrate_object_unlock(struct mdt_thread_info *info,
1608 struct mdt_object *obj,
1609 struct mdt_lock_handle *lh,
1610 struct ldlm_enqueue_info *einfo,
1611 struct list_head *slave_locks,
1614 if (mdt_object_remote(obj)) {
1615 mdt_unlock_list(info, slave_locks, decref);
1616 mdt_object_unlock(info, obj, lh, decref);
1618 mdt_reint_striped_unlock(info, obj, lh, einfo, decref);
1623 * lock parents of links, and also check whether total locks don't exceed
1626 * \retval 0 on success, and locks can be saved in ptlrpc_reply_stat
1627 * \retval 1 on success, but total lock count may exceed RS_MAX_LOCKS
1628 * \retval -ev negative errno upon error
1630 static int mdt_link_parents_lock(struct mdt_thread_info *info,
1631 struct mdt_object *pobj,
1632 const struct md_attr *ma,
1633 struct mdt_object *obj,
1634 struct mdt_lock_handle *lhp,
1635 struct ldlm_enqueue_info *peinfo,
1636 struct list_head *parent_slave_locks,
1637 struct list_head *link_locks)
1639 struct mdt_device *mdt = info->mti_mdt;
1640 struct lu_buf *buf = &info->mti_big_buf;
1641 struct lu_name *lname = &info->mti_name;
1642 struct linkea_data ldata = { NULL };
1643 bool blocked = false;
1644 int local_lnkp_cnt = 0;
1648 if (S_ISDIR(lu_object_attr(&obj->mot_obj)))
1651 buf = lu_buf_check_and_alloc(buf, MAX_LINKEA_SIZE);
1652 if (buf->lb_buf == NULL)
1656 rc = mdt_links_read(info, obj, &ldata);
1658 if (rc == -ENOENT || rc == -ENODATA)
1663 for (linkea_first_entry(&ldata); ldata.ld_lee && !rc;
1664 linkea_next_entry(&ldata)) {
1665 struct mdt_object *lnkp;
1666 struct mdt_sub_lock *msl;
1670 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen, lname,
1673 /* check if it's also linked to parent */
1674 if (lu_fid_eq(mdt_object_fid(pobj), &fid)) {
1675 CDEBUG(D_INFO, "skip parent "DFID", reovke "DNAME"\n",
1676 PFID(&fid), PNAME(lname));
1677 /* in case link is remote object, revoke LOOKUP lock */
1678 rc = mdt_revoke_remote_lookup_lock(info, pobj, obj);
1684 /* check if it's linked to a stripe of parent */
1685 if (ma->ma_valid & MA_LMV) {
1686 struct lmv_mds_md_v1 *lmv = &ma->ma_lmv->lmv_md_v1;
1687 struct lu_fid *stripe_fid = &info->mti_tmp_fid1;
1690 for (; j < le32_to_cpu(lmv->lmv_stripe_count); j++) {
1691 fid_le_to_cpu(stripe_fid,
1692 &lmv->lmv_stripe_fids[j]);
1693 if (lu_fid_eq(stripe_fid, &fid)) {
1694 CDEBUG(D_INFO, "skip stripe "DFID
1695 ", reovke "DNAME"\n",
1696 PFID(&fid), PNAME(lname));
1697 lnkp = mdt_object_find(info->mti_env,
1700 GOTO(out, rc = PTR_ERR(lnkp));
1706 rc = mdt_revoke_remote_lookup_lock(info, lnkp,
1708 mdt_object_put(info->mti_env, lnkp);
1713 /* Check if it's already locked */
1714 list_for_each_entry(msl, link_locks, msl_linkage) {
1715 if (lu_fid_eq(mdt_object_fid(msl->msl_obj), &fid)) {
1717 DFID" was locked, revoke "DNAME"\n",
1718 PFID(&fid), PNAME(lname));
1719 lnkp = msl->msl_obj;
1725 rc = mdt_revoke_remote_lookup_lock(info, lnkp, obj);
1729 CDEBUG(D_INFO, "lock "DFID":"DNAME"\n",
1730 PFID(&fid), PNAME(lname));
1732 lnkp = mdt_object_find(info->mti_env, mdt, &fid);
1734 CWARN("%s: cannot find obj "DFID": %ld\n",
1735 mdt_obd_name(mdt), PFID(&fid), PTR_ERR(lnkp));
1739 if (!mdt_object_exists(lnkp)) {
1740 CDEBUG(D_INFO, DFID" doesn't exist, skip "DNAME"\n",
1741 PFID(&fid), PNAME(lname));
1742 mdt_object_put(info->mti_env, lnkp);
1746 if (!mdt_object_remote(lnkp))
1751 GOTO(out, rc = -ENOMEM);
1754 * we can't follow parent-child lock order like other MD
1755 * operations, use lock_try here to avoid deadlock, if the lock
1756 * cannot be taken, drop all locks taken, revoke the blocked
1757 * one, and continue processing the remaining entries, and in
1758 * the end of the loop restart from beginning.
1760 mdt_lock_pdo_init(&msl->msl_lh, LCK_PW, lname);
1762 rc = mdt_object_lock_try(info, lnkp, &msl->msl_lh, &ibits,
1763 MDS_INODELOCK_UPDATE, true);
1764 if (!(ibits & MDS_INODELOCK_UPDATE)) {
1766 CDEBUG(D_INFO, "busy lock on "DFID" "DNAME"\n",
1767 PFID(&fid), PNAME(lname));
1769 mdt_unlock_list(info, link_locks, 1);
1770 /* also unlock parent locks to avoid deadlock */
1772 mdt_migrate_object_unlock(info, pobj, lhp,
1779 mdt_lock_pdo_init(&msl->msl_lh, LCK_PW, lname);
1780 rc = mdt_object_lock(info, lnkp, &msl->msl_lh,
1781 MDS_INODELOCK_UPDATE);
1783 mdt_object_put(info->mti_env, lnkp);
1788 if (mdt_object_remote(lnkp)) {
1789 struct ldlm_lock *lock;
1792 * for remote object, set lock cb_atomic,
1793 * so lock can be released in blocking_ast()
1794 * immediately, then the next lock_try will
1795 * have better chance of success.
1797 lock = ldlm_handle2lock(
1798 &msl->msl_lh.mlh_rreg_lh);
1799 LASSERT(lock != NULL);
1800 lock_res_and_lock(lock);
1801 ldlm_set_atomic_cb(lock);
1802 unlock_res_and_lock(lock);
1803 LDLM_LOCK_PUT(lock);
1806 mdt_object_unlock_put(info, lnkp, &msl->msl_lh, 1);
1811 INIT_LIST_HEAD(&msl->msl_linkage);
1812 msl->msl_obj = lnkp;
1813 list_add_tail(&msl->msl_linkage, link_locks);
1815 rc = mdt_revoke_remote_lookup_lock(info, lnkp, obj);
1819 GOTO(out, rc = -EBUSY);
1824 mdt_unlock_list(info, link_locks, rc);
1825 } else if (local_lnkp_cnt > RS_MAX_LOCKS - 5) {
1826 CDEBUG(D_INFO, "Too many links (%d), sync operations\n",
1829 * parent may have 3 local objects: master object and 2 stripes
1830 * (if it's being migrated too); source may have 1 local objects
1831 * as regular file; target has 1 local object.
1832 * Note, source may have 2 local locks if it is directory but it
1833 * can't have hardlinks, so it is not considered here.
1840 static int mdt_lock_remote_slaves(struct mdt_thread_info *info,
1841 struct mdt_object *obj,
1842 const struct md_attr *ma,
1843 struct list_head *slave_locks)
1845 struct mdt_device *mdt = info->mti_mdt;
1846 const struct lmv_mds_md_v1 *lmv = &ma->ma_lmv->lmv_md_v1;
1847 struct lu_fid *fid = &info->mti_tmp_fid1;
1848 struct mdt_object *slave;
1849 struct mdt_sub_lock *msl;
1854 LASSERT(mdt_object_remote(obj));
1855 LASSERT(ma->ma_valid & MA_LMV);
1858 if (!lmv_is_sane(lmv))
1861 for (i = 0; i < le32_to_cpu(lmv->lmv_stripe_count); i++) {
1862 fid_le_to_cpu(fid, &lmv->lmv_stripe_fids[i]);
1864 if (!fid_is_sane(fid))
1867 slave = mdt_object_find(info->mti_env, mdt, fid);
1869 GOTO(out, rc = PTR_ERR(slave));
1873 mdt_object_put(info->mti_env, slave);
1874 GOTO(out, rc = -ENOMEM);
1877 mdt_lock_reg_init(&msl->msl_lh, LCK_EX);
1878 rc = mdt_reint_object_lock(info, slave, &msl->msl_lh,
1879 MDS_INODELOCK_UPDATE, true);
1882 mdt_object_put(info->mti_env, slave);
1886 INIT_LIST_HEAD(&msl->msl_linkage);
1887 msl->msl_obj = slave;
1888 list_add_tail(&msl->msl_linkage, slave_locks);
1894 mdt_unlock_list(info, slave_locks, rc);
1898 /* lock parent and its stripes */
1899 static int mdt_migrate_parent_lock(struct mdt_thread_info *info,
1900 struct mdt_object *obj,
1901 const struct md_attr *ma,
1902 struct mdt_lock_handle *lh,
1903 struct ldlm_enqueue_info *einfo,
1904 struct list_head *slave_locks)
1908 if (mdt_object_remote(obj)) {
1909 rc = mdt_remote_object_lock(info, obj, mdt_object_fid(obj),
1910 &lh->mlh_rreg_lh, LCK_PW,
1911 MDS_INODELOCK_UPDATE, false);
1916 * if obj is remote and striped, lock its stripes explicitly
1917 * because it's not striped in LOD layer on this MDT.
1919 if (ma->ma_valid & MA_LMV) {
1920 rc = mdt_lock_remote_slaves(info, obj, ma, slave_locks);
1922 mdt_object_unlock(info, obj, lh, rc);
1925 rc = mdt_reint_striped_lock(info, obj, lh, MDS_INODELOCK_UPDATE,
1933 * in migration, object may be remote, and we need take full lock of it and its
1934 * stripes if it's directory, besides, object may be a remote object on its
1935 * parent, revoke its LOOKUP lock on where its parent is located.
1937 static int mdt_migrate_object_lock(struct mdt_thread_info *info,
1938 struct mdt_object *pobj,
1939 struct mdt_object *obj,
1940 struct mdt_lock_handle *lh,
1941 struct ldlm_enqueue_info *einfo,
1942 struct list_head *slave_locks)
1946 if (mdt_object_remote(obj)) {
1947 rc = mdt_revoke_remote_lookup_lock(info, pobj, obj);
1951 rc = mdt_remote_object_lock(info, obj, mdt_object_fid(obj),
1952 &lh->mlh_rreg_lh, LCK_EX,
1953 MDS_INODELOCK_FULL, false);
1958 * if obj is remote and striped, lock its stripes explicitly
1959 * because it's not striped in LOD layer on this MDT.
1961 if (S_ISDIR(lu_object_attr(&obj->mot_obj))) {
1962 struct md_attr *ma = &info->mti_attr;
1964 rc = mdt_stripe_get(info, obj, ma, XATTR_NAME_LMV);
1966 mdt_object_unlock(info, obj, lh, rc);
1970 if (ma->ma_valid & MA_LMV) {
1971 rc = mdt_lock_remote_slaves(info, obj, ma,
1974 mdt_object_unlock(info, obj, lh, rc);
1978 if (mdt_object_remote(pobj)) {
1979 rc = mdt_revoke_remote_lookup_lock(info, pobj, obj);
1984 rc = mdt_reint_striped_lock(info, obj, lh, MDS_INODELOCK_FULL,
1992 * lookup source by name, if parent is striped directory, we need to find the
1993 * corresponding stripe where source is located, and then lookup there.
1995 * besides, if parent is migrating too, and file is already in target stripe,
1996 * this should be a redo of 'lfs migrate' on client side.
1998 static int mdt_migrate_lookup(struct mdt_thread_info *info,
1999 struct mdt_object *pobj,
2000 const struct md_attr *ma,
2001 const struct lu_name *lname,
2002 struct mdt_object **spobj,
2003 struct mdt_object **sobj)
2005 const struct lu_env *env = info->mti_env;
2006 struct lu_fid *fid = &info->mti_tmp_fid1;
2007 struct mdt_object *stripe;
2010 if (ma->ma_valid & MA_LMV) {
2011 /* if parent is striped, lookup on corresponding stripe */
2012 struct lmv_mds_md_v1 *lmv = &ma->ma_lmv->lmv_md_v1;
2014 if (!lmv_is_sane(lmv))
2017 rc = lmv_name_to_stripe_index_old(lmv, lname->ln_name,
2022 fid_le_to_cpu(fid, &lmv->lmv_stripe_fids[rc]);
2024 stripe = mdt_object_find(env, info->mti_mdt, fid);
2026 return PTR_ERR(stripe);
2029 rc = mdo_lookup(env, mdt_object_child(stripe), lname, fid,
2031 if (rc == -ENOENT && lmv_is_layout_changing(lmv)) {
2033 * if parent layout is changeing, and lookup child
2034 * failed on source stripe, lookup again on target
2035 * stripe, if it exists, it means previous migration
2036 * was interrupted, and current file was migrated
2039 mdt_object_put(env, stripe);
2041 rc = lmv_name_to_stripe_index(lmv, lname->ln_name,
2046 fid_le_to_cpu(fid, &lmv->lmv_stripe_fids[rc]);
2048 stripe = mdt_object_find(env, info->mti_mdt, fid);
2050 return PTR_ERR(stripe);
2053 rc = mdo_lookup(env, mdt_object_child(stripe), lname,
2054 fid, &info->mti_spec);
2055 mdt_object_put(env, stripe);
2056 return rc ?: -EALREADY;
2058 mdt_object_put(env, stripe);
2063 rc = mdo_lookup(env, mdt_object_child(pobj), lname, fid,
2069 mdt_object_get(env, stripe);
2074 *sobj = mdt_object_find(env, info->mti_mdt, fid);
2075 if (IS_ERR(*sobj)) {
2076 mdt_object_put(env, stripe);
2077 rc = PTR_ERR(*sobj);
2085 /* end lease and close file for regular file */
2086 static int mdd_migrate_close(struct mdt_thread_info *info,
2087 struct mdt_object *obj)
2089 struct close_data *data;
2090 struct mdt_body *repbody;
2091 struct ldlm_lock *lease;
2096 if (!req_capsule_field_present(info->mti_pill, &RMF_MDT_EPOCH,
2098 !req_capsule_field_present(info->mti_pill, &RMF_CLOSE_DATA,
2102 data = req_capsule_client_get(info->mti_pill, &RMF_CLOSE_DATA);
2107 lease = ldlm_handle2lock(&data->cd_handle);
2111 /* check if the lease was already canceled */
2112 lock_res_and_lock(lease);
2113 rc = ldlm_is_cancel(lease);
2114 unlock_res_and_lock(lease);
2118 LDLM_DEBUG(lease, DFID" lease broken",
2119 PFID(mdt_object_fid(obj)));
2123 * cancel server side lease, client side counterpart should have been
2124 * cancelled, it's okay to cancel it now as we've held mot_open_sem.
2126 ldlm_lock_cancel(lease);
2127 ldlm_reprocess_all(lease->l_resource, lease);
2128 LDLM_LOCK_PUT(lease);
2131 rc2 = mdt_close_internal(info, mdt_info_req(info), NULL);
2132 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
2133 repbody->mbo_valid |= OBD_MD_CLOSE_INTENT_EXECED;
2139 * migrate file in below steps:
2140 * 1. lock parent and its stripes
2141 * 2. lookup source by name
2142 * 3. lock parents of source links if source is not directory
2143 * 4. reject if source is in HSM
2144 * 5. take source open_sem and close file if source is regular file
2145 * 6. lock source and its stripes if it's directory
2146 * 7. lock target so subsequent change to it can trigger COS
2148 * 9. unlock above locks
2149 * 10. sync device if source has links
2151 int mdt_reint_migrate(struct mdt_thread_info *info,
2152 struct mdt_lock_handle *unused)
2154 const struct lu_env *env = info->mti_env;
2155 struct mdt_device *mdt = info->mti_mdt;
2156 struct ptlrpc_request *req = mdt_info_req(info);
2157 struct mdt_reint_record *rr = &info->mti_rr;
2158 struct lu_ucred *uc = mdt_ucred(info);
2159 struct md_attr *ma = &info->mti_attr;
2160 struct ldlm_enqueue_info *peinfo = &info->mti_einfo[0];
2161 struct ldlm_enqueue_info *seinfo = &info->mti_einfo[1];
2162 struct mdt_object *pobj;
2163 struct mdt_object *spobj = NULL;
2164 struct mdt_object *sobj = NULL;
2165 struct mdt_object *tobj;
2166 struct lustre_handle rename_lh = { 0 };
2167 struct mdt_lock_handle *lhp;
2168 struct mdt_lock_handle *lhs;
2169 struct mdt_lock_handle *lht;
2170 LIST_HEAD(parent_slave_locks);
2171 LIST_HEAD(child_slave_locks);
2172 LIST_HEAD(link_locks);
2173 int lock_retries = 5;
2174 bool open_sem_locked = false;
2175 bool do_sync = false;
2179 CDEBUG(D_INODE, "migrate "DFID"/"DNAME" to "DFID"\n", PFID(rr->rr_fid1),
2180 PNAME(&rr->rr_name), PFID(rr->rr_fid2));
2182 if (info->mti_dlm_req)
2183 ldlm_request_cancel(req, info->mti_dlm_req, 0, LATF_SKIP);
2185 if (!fid_is_md_operative(rr->rr_fid1) ||
2186 !fid_is_md_operative(rr->rr_fid2))
2189 /* don't allow migrate . or .. */
2190 if (lu_name_is_dot_or_dotdot(&rr->rr_name))
2193 if (!mdt->mdt_enable_remote_dir || !mdt->mdt_enable_dir_migration)
2196 if (uc && !md_capable(uc, CAP_SYS_ADMIN) &&
2197 uc->uc_gid != mdt->mdt_enable_remote_dir_gid &&
2198 mdt->mdt_enable_remote_dir_gid != -1)
2202 * Note: do not enqueue rename lock for replay request, because
2203 * if other MDT holds rename lock, but being blocked to wait for
2204 * this MDT to finish its recovery, and the failover MDT can not
2205 * get rename lock, which will cause deadlock.
2207 * req is NULL if this is called by directory auto-split.
2209 if (req && !req_is_replay(req)) {
2210 rc = mdt_rename_lock(info, &rename_lh);
2212 CERROR("%s: can't lock FS for rename: rc = %d\n",
2213 mdt_obd_name(info->mti_mdt), rc);
2218 /* pobj is master object of parent */
2219 pobj = mdt_object_find(env, mdt, rr->rr_fid1);
2221 GOTO(unlock_rename, rc = PTR_ERR(pobj));
2224 rc = mdt_version_get_check(info, pobj, 0);
2226 GOTO(put_parent, rc);
2229 if (!mdt_object_exists(pobj))
2230 GOTO(put_parent, rc = -ENOENT);
2232 if (!S_ISDIR(lu_object_attr(&pobj->mot_obj)))
2233 GOTO(put_parent, rc = -ENOTDIR);
2235 rc = mdt_stripe_get(info, pobj, ma, XATTR_NAME_LMV);
2237 GOTO(put_parent, rc);
2240 /* lock parent object */
2241 lhp = &info->mti_lh[MDT_LH_PARENT];
2242 mdt_lock_reg_init(lhp, LCK_PW);
2243 rc = mdt_migrate_parent_lock(info, pobj, ma, lhp, peinfo,
2244 &parent_slave_locks);
2246 GOTO(put_parent, rc);
2249 * spobj is the corresponding stripe against name if pobj is striped
2250 * directory, which is the real parent, and no need to lock, because
2251 * we've taken full lock of pobj.
2253 rc = mdt_migrate_lookup(info, pobj, ma, &rr->rr_name, &spobj, &sobj);
2255 GOTO(unlock_parent, rc);
2257 /* lock parents of source links, and revoke LOOKUP lock of links */
2258 rc = mdt_link_parents_lock(info, pobj, ma, sobj, lhp, peinfo,
2259 &parent_slave_locks, &link_locks);
2260 if (rc == -EBUSY && lock_retries-- > 0) {
2261 mdt_object_put(env, sobj);
2262 mdt_object_put(env, spobj);
2267 GOTO(put_source, rc);
2270 * RS_MAX_LOCKS is the limit of number of locks that can be saved along
2271 * with one request, if total lock count exceeds this limit, we will
2272 * drop all locks after migration, and synchronous device in the end.
2276 /* TODO: DoM migration is not supported, migrate dirent only */
2277 if (S_ISREG(lu_object_attr(&sobj->mot_obj))) {
2278 rc = mdt_stripe_get(info, sobj, ma, XATTR_NAME_LOV);
2280 GOTO(unlock_links, rc);
2282 if (ma->ma_valid & MA_LOV && mdt_lmm_dom_stripesize(ma->ma_lmm))
2283 info->mti_spec.sp_migrate_nsonly = 1;
2284 } else if (S_ISDIR(lu_object_attr(&sobj->mot_obj))) {
2285 rc = mdt_stripe_get(info, sobj, ma, XATTR_NAME_LMV);
2287 GOTO(unlock_links, rc);
2289 /* race with restripe/auto-split? */
2290 if ((ma->ma_valid & MA_LMV) &&
2291 lmv_is_restriping(&ma->ma_lmv->lmv_md_v1))
2292 GOTO(unlock_links, rc = -EBUSY);
2295 /* if migration HSM is allowed */
2296 if (!mdt->mdt_opts.mo_migrate_hsm_allowed) {
2297 ma->ma_need = MA_HSM;
2299 rc = mdt_attr_get_complex(info, sobj, ma);
2301 GOTO(unlock_links, rc);
2303 if ((ma->ma_valid & MA_HSM) && ma->ma_hsm.mh_flags != 0)
2304 GOTO(unlock_links, rc = -EOPNOTSUPP);
2307 /* end lease and close file for regular file */
2308 if (info->mti_spec.sp_migrate_close) {
2309 /* try to hold open_sem so that nobody else can open the file */
2310 if (!down_write_trylock(&sobj->mot_open_sem)) {
2312 mdd_migrate_close(info, sobj);
2313 GOTO(unlock_links, rc = -EBUSY);
2315 open_sem_locked = true;
2316 rc = mdd_migrate_close(info, sobj);
2318 GOTO(unlock_open_sem, rc);
2323 lhs = &info->mti_lh[MDT_LH_OLD];
2324 mdt_lock_reg_init(lhs, LCK_EX);
2325 rc = mdt_migrate_object_lock(info, spobj, sobj, lhs, seinfo,
2326 &child_slave_locks);
2328 GOTO(unlock_open_sem, rc);
2331 tobj = mdt_object_find(env, mdt, rr->rr_fid2);
2333 GOTO(unlock_source, rc = PTR_ERR(tobj));
2335 lht = &info->mti_lh[MDT_LH_NEW];
2336 mdt_lock_reg_init(lht, LCK_EX);
2337 rc = mdt_reint_object_lock(info, tobj, lht, MDS_INODELOCK_FULL, true);
2339 GOTO(put_target, rc);
2341 /* Don't do lookup sanity check. We know name doesn't exist. */
2342 info->mti_spec.sp_cr_lookup = 0;
2343 info->mti_spec.sp_feat = &dt_directory_features;
2345 rc = mdo_migrate(env, mdt_object_child(pobj),
2346 mdt_object_child(sobj), &rr->rr_name,
2347 mdt_object_child(tobj),
2348 &info->mti_spec, ma);
2350 lprocfs_counter_incr(mdt->mdt_lu_dev.ld_obd->obd_md_stats,
2351 LPROC_MDT_MIGRATE + LPROC_MD_LAST_OPC);
2354 mdt_object_unlock(info, tobj, lht, rc);
2356 mdt_object_put(env, tobj);
2358 mdt_migrate_object_unlock(info, sobj, lhs, seinfo,
2359 &child_slave_locks, rc);
2361 if (open_sem_locked)
2362 up_write(&sobj->mot_open_sem);
2364 /* if we've got too many locks to save into RPC,
2365 * then just commit before the locks are released
2368 mdt_device_sync(env, mdt);
2369 mdt_unlock_list(info, &link_locks, do_sync ? 1 : rc);
2371 mdt_object_put(env, sobj);
2372 mdt_object_put(env, spobj);
2374 mdt_migrate_object_unlock(info, pobj, lhp, peinfo,
2375 &parent_slave_locks, rc);
2377 mdt_object_put(env, pobj);
2379 if (lustre_handle_is_used(&rename_lh))
2380 mdt_rename_unlock(&rename_lh);
2385 static int mdt_object_lock_save(struct mdt_thread_info *info,
2386 struct mdt_object *dir,
2387 struct mdt_lock_handle *lh,
2388 int idx, bool cos_incompat)
2392 /* we lock the target dir if it is local */
2393 rc = mdt_reint_object_lock(info, dir, lh, MDS_INODELOCK_UPDATE,
2398 /* get and save correct version after locking */
2399 mdt_version_get_save(info, dir, idx);
2404 * determine lock order of sobj and tobj
2406 * there are two situations we need to lock tobj before sobj:
2407 * 1. sobj is child of tobj
2408 * 2. sobj and tobj are stripes of a directory, and stripe index of sobj is
2409 * larger than that of tobj
2411 * \retval 1 lock tobj before sobj
2412 * \retval 0 lock sobj before tobj
2413 * \retval -ev negative errno upon error
2415 static int mdt_rename_determine_lock_order(struct mdt_thread_info *info,
2416 struct mdt_object *sobj,
2417 struct mdt_object *tobj)
2419 struct md_attr *ma = &info->mti_attr;
2420 struct lu_fid *spfid = &info->mti_tmp_fid1;
2421 struct lu_fid *tpfid = &info->mti_tmp_fid2;
2422 struct lmv_mds_md_v1 *lmv;
2427 /* sobj and tobj are the same */
2431 if (fid_is_root(mdt_object_fid(sobj)))
2434 if (fid_is_root(mdt_object_fid(tobj)))
2437 /* check whether sobj is child of tobj */
2438 rc = mdo_is_subdir(info->mti_env, mdt_object_child(sobj),
2439 mdt_object_fid(tobj));
2446 /* check whether sobj and tobj are children of the same parent */
2447 rc = mdt_attr_get_pfid(info, sobj, spfid);
2451 rc = mdt_attr_get_pfid(info, tobj, tpfid);
2455 if (!lu_fid_eq(spfid, tpfid))
2458 /* check whether sobj and tobj are sibling stripes */
2459 rc = mdt_stripe_get(info, sobj, ma, XATTR_NAME_LMV);
2463 if (!(ma->ma_valid & MA_LMV))
2466 lmv = &ma->ma_lmv->lmv_md_v1;
2467 if (!(le32_to_cpu(lmv->lmv_magic) & LMV_MAGIC_STRIPE))
2469 sindex = le32_to_cpu(lmv->lmv_master_mdt_index);
2472 rc = mdt_stripe_get(info, tobj, ma, XATTR_NAME_LMV);
2476 if (!(ma->ma_valid & MA_LMV))
2479 lmv = &ma->ma_lmv->lmv_md_v1;
2480 if (!(le32_to_cpu(lmv->lmv_magic) & LMV_MAGIC_STRIPE))
2482 tindex = le32_to_cpu(lmv->lmv_master_mdt_index);
2484 /* check stripe index of sobj and tobj */
2485 if (sindex == tindex)
2488 return sindex < tindex ? 0 : 1;
2492 * lock rename source object.
2494 * Both source and source parent may be remote, and source may be a remote
2495 * object on source parent, to avoid overriding lock handle, store remote
2496 * LOOKUP lock separately in @lhr.
2498 * \retval 0 on success
2499 * \retval -ev negative errno upon error
2501 static int mdt_rename_source_lock(struct mdt_thread_info *info,
2502 struct mdt_object *parent,
2503 struct mdt_object *child,
2504 struct mdt_lock_handle *lhc,
2505 struct mdt_lock_handle *lhr,
2511 rc = mdt_is_remote_object(info, parent, child);
2516 /* enqueue remote LOOKUP lock from the parent MDT */
2517 __u64 rmt_ibits = MDS_INODELOCK_LOOKUP;
2519 if (mdt_object_remote(parent)) {
2520 rc = mdt_remote_object_lock(info, parent,
2521 mdt_object_fid(child),
2528 LASSERT(mdt_object_remote(child));
2529 rc = mdt_object_local_lock(info, child, lhr,
2530 &rmt_ibits, 0, true);
2535 ibits &= ~MDS_INODELOCK_LOOKUP;
2538 if (mdt_object_remote(child)) {
2539 rc = mdt_remote_object_lock(info, child, mdt_object_fid(child),
2546 rc = mdt_reint_object_lock(info, child, lhc, ibits,
2551 mdt_object_unlock(info, child, lhr, rc);
2557 * VBR: rename versions in reply: 0 - srcdir parent; 1 - tgtdir parent;
2558 * 2 - srcdir child; 3 - tgtdir child.
2559 * Update on disk version of srcdir child.
2561 static int mdt_reint_rename(struct mdt_thread_info *info,
2562 struct mdt_lock_handle *unused)
2564 struct mdt_device *mdt = info->mti_mdt;
2565 struct mdt_reint_record *rr = &info->mti_rr;
2566 struct md_attr *ma = &info->mti_attr;
2567 struct ptlrpc_request *req = mdt_info_req(info);
2568 struct mdt_object *msrcdir = NULL;
2569 struct mdt_object *mtgtdir = NULL;
2570 struct mdt_object *mold;
2571 struct mdt_object *mnew = NULL;
2572 struct lustre_handle rename_lh = { 0 };
2573 struct mdt_lock_handle *lh_srcdirp;
2574 struct mdt_lock_handle *lh_tgtdirp;
2575 struct mdt_lock_handle *lh_oldp = NULL;
2576 struct mdt_lock_handle *lh_rmt = NULL;
2577 struct mdt_lock_handle *lh_newp = NULL;
2578 struct lu_fid *old_fid = &info->mti_tmp_fid1;
2579 struct lu_fid *new_fid = &info->mti_tmp_fid2;
2581 bool reverse = false, discard = false;
2583 ktime_t kstart = ktime_get();
2587 DEBUG_REQ(D_INODE, req, "rename "DFID"/"DNAME" to "DFID"/"DNAME,
2588 PFID(rr->rr_fid1), PNAME(&rr->rr_name),
2589 PFID(rr->rr_fid2), PNAME(&rr->rr_tgt_name));
2591 if (info->mti_dlm_req)
2592 ldlm_request_cancel(req, info->mti_dlm_req, 0, LATF_SKIP);
2594 if (!fid_is_md_operative(rr->rr_fid1) ||
2595 !fid_is_md_operative(rr->rr_fid2))
2598 /* find both parents. */
2599 msrcdir = mdt_parent_find_check(info, rr->rr_fid1, 0);
2600 if (IS_ERR(msrcdir))
2601 RETURN(PTR_ERR(msrcdir));
2603 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_RENAME3, 5);
2605 if (lu_fid_eq(rr->rr_fid1, rr->rr_fid2)) {
2607 mdt_object_get(info->mti_env, mtgtdir);
2609 mtgtdir = mdt_parent_find_check(info, rr->rr_fid2, 1);
2610 if (IS_ERR(mtgtdir))
2611 GOTO(out_put_srcdir, rc = PTR_ERR(mtgtdir));
2615 * Note: do not enqueue rename lock for replay request, because
2616 * if other MDT holds rename lock, but being blocked to wait for
2617 * this MDT to finish its recovery, and the failover MDT can not
2618 * get rename lock, which will cause deadlock.
2620 if (!req_is_replay(req)) {
2622 * Normally rename RPC is handled on the MDT with the target
2623 * directory (if target exists, it's on the MDT with the
2624 * target), if the source directory is remote, it's a hint that
2625 * source is remote too (this may not be true, but it won't
2626 * cause any issue), return -EXDEV early to avoid taking
2629 if (!mdt->mdt_enable_remote_rename &&
2630 mdt_object_remote(msrcdir))
2631 GOTO(out_put_tgtdir, rc = -EXDEV);
2633 /* This might be further relaxed in the future for regular file
2634 * renames in different source and target parents. Start with
2635 * only same-directory renames for simplicity and because this
2636 * is by far the most the common use case.
2638 if (msrcdir != mtgtdir) {
2639 rc = mdt_rename_lock(info, &rename_lh);
2641 CERROR("%s: cannot lock for rename: rc = %d\n",
2642 mdt_obd_name(mdt), rc);
2643 GOTO(out_put_tgtdir, rc);
2646 CDEBUG(D_INFO, "%s: samedir rename "DFID"/"DNAME"\n",
2647 mdt_obd_name(mdt), PFID(rr->rr_fid1),
2648 PNAME(&rr->rr_name));
2652 rc = mdt_rename_determine_lock_order(info, msrcdir, mtgtdir);
2654 GOTO(out_unlock_rename, rc);
2658 /* source needs to be looked up after locking source parent, otherwise
2659 * this rename may race with unlink source, and cause rename hang, see
2660 * sanityn.sh 55b, so check parents first, if later we found source is
2661 * remote, relock parents.
2663 cos_incompat = (mdt_object_remote(msrcdir) ||
2664 mdt_object_remote(mtgtdir));
2666 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_RENAME4, 5);
2668 /* lock parents in the proper order. */
2669 lh_srcdirp = &info->mti_lh[MDT_LH_PARENT];
2670 lh_tgtdirp = &info->mti_lh[MDT_LH_CHILD];
2672 OBD_RACE(OBD_FAIL_MDS_REINT_OPEN);
2673 OBD_RACE(OBD_FAIL_MDS_REINT_OPEN2);
2675 mdt_lock_pdo_init(lh_srcdirp, LCK_PW, &rr->rr_name);
2676 mdt_lock_pdo_init(lh_tgtdirp, LCK_PW, &rr->rr_tgt_name);
2679 rc = mdt_object_lock_save(info, mtgtdir, lh_tgtdirp, 1,
2682 GOTO(out_unlock_rename, rc);
2684 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_RENAME, 5);
2686 rc = mdt_object_lock_save(info, msrcdir, lh_srcdirp, 0,
2689 mdt_object_unlock(info, mtgtdir, lh_tgtdirp, rc);
2690 GOTO(out_unlock_rename, rc);
2693 rc = mdt_object_lock_save(info, msrcdir, lh_srcdirp, 0,
2696 GOTO(out_unlock_rename, rc);
2698 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_RENAME, 5);
2700 if (mtgtdir != msrcdir) {
2701 rc = mdt_object_lock_save(info, mtgtdir, lh_tgtdirp, 1,
2703 } else if (!mdt_object_remote(mtgtdir) &&
2704 lh_srcdirp->mlh_pdo_hash !=
2705 lh_tgtdirp->mlh_pdo_hash) {
2706 rc = mdt_pdir_hash_lock(info, lh_tgtdirp, mtgtdir,
2707 MDS_INODELOCK_UPDATE,
2709 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_PDO_LOCK2, 10);
2712 mdt_object_unlock(info, msrcdir, lh_srcdirp, rc);
2713 GOTO(out_unlock_rename, rc);
2717 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_RENAME4, 5);
2718 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_RENAME2, 5);
2720 /* find mold object. */
2722 rc = mdt_lookup_version_check(info, msrcdir, &rr->rr_name, old_fid, 2);
2724 GOTO(out_unlock_parents, rc);
2726 if (lu_fid_eq(old_fid, rr->rr_fid1) || lu_fid_eq(old_fid, rr->rr_fid2))
2727 GOTO(out_unlock_parents, rc = -EINVAL);
2729 if (!fid_is_md_operative(old_fid))
2730 GOTO(out_unlock_parents, rc = -EPERM);
2732 mold = mdt_object_find(info->mti_env, info->mti_mdt, old_fid);
2734 GOTO(out_unlock_parents, rc = PTR_ERR(mold));
2736 if (!mdt_object_exists(mold)) {
2737 LU_OBJECT_DEBUG(D_INODE, info->mti_env,
2739 "object does not exist");
2740 GOTO(out_put_old, rc = -ENOENT);
2743 if (mdt_object_remote(mold) && !mdt->mdt_enable_remote_rename)
2744 GOTO(out_put_old, rc = -EXDEV);
2746 /* Check if @mtgtdir is subdir of @mold, before locking child
2747 * to avoid reverse locking.
2749 if (mtgtdir != msrcdir) {
2750 rc = mdo_is_subdir(info->mti_env, mdt_object_child(mtgtdir),
2755 GOTO(out_put_old, rc);
2759 tgt_vbr_obj_set(info->mti_env, mdt_obj2dt(mold));
2760 /* save version after locking */
2761 mdt_version_get_save(info, mold, 2);
2763 if (!cos_incompat && mdt_object_remote(mold)) {
2764 cos_incompat = true;
2765 mdt_object_put(info->mti_env, mold);
2766 mdt_object_unlock(info, mtgtdir, lh_tgtdirp, -EAGAIN);
2767 mdt_object_unlock(info, msrcdir, lh_srcdirp, -EAGAIN);
2771 /* find mnew object:
2772 * mnew target object may not exist now
2773 * lookup with version checking
2776 rc = mdt_lookup_version_check(info, mtgtdir, &rr->rr_tgt_name, new_fid,
2779 /* the new_fid should have been filled at this moment */
2780 if (lu_fid_eq(old_fid, new_fid))
2781 GOTO(out_put_old, rc);
2783 if (lu_fid_eq(new_fid, rr->rr_fid1) ||
2784 lu_fid_eq(new_fid, rr->rr_fid2))
2785 GOTO(out_put_old, rc = -EINVAL);
2787 if (!fid_is_md_operative(new_fid))
2788 GOTO(out_put_old, rc = -EPERM);
2790 mnew = mdt_object_find(info->mti_env, info->mti_mdt, new_fid);
2792 GOTO(out_put_old, rc = PTR_ERR(mnew));
2794 if (!mdt_object_exists(mnew)) {
2795 LU_OBJECT_DEBUG(D_INODE, info->mti_env,
2797 "object does not exist");
2798 GOTO(out_put_new, rc = -ENOENT);
2801 if (mdt_object_remote(mnew)) {
2802 struct mdt_body *repbody;
2804 /* Always send rename req to the target child MDT */
2805 repbody = req_capsule_server_get(info->mti_pill,
2807 LASSERT(repbody != NULL);
2808 repbody->mbo_fid1 = *new_fid;
2809 repbody->mbo_valid |= (OBD_MD_FLID | OBD_MD_MDS);
2810 GOTO(out_put_new, rc = -EXDEV);
2812 /* Before locking the target dir, check we do not replace
2813 * a dir with a non-dir, otherwise it may deadlock with
2814 * link op which tries to create a link in this dir
2815 * back to this non-dir.
2817 if (S_ISDIR(lu_object_attr(&mnew->mot_obj)) &&
2818 !S_ISDIR(lu_object_attr(&mold->mot_obj)))
2819 GOTO(out_put_new, rc = -EISDIR);
2821 lh_oldp = &info->mti_lh[MDT_LH_OLD];
2822 lh_rmt = &info->mti_lh[MDT_LH_RMT];
2823 mdt_lock_reg_init(lh_oldp, LCK_EX);
2824 mdt_lock_reg_init(lh_rmt, LCK_EX);
2825 lock_ibits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_XATTR;
2826 rc = mdt_rename_source_lock(info, msrcdir, mold, lh_oldp,
2827 lh_rmt, lock_ibits, cos_incompat);
2829 GOTO(out_put_new, rc);
2831 /* Check if @msrcdir is subdir of @mnew, before locking child
2832 * to avoid reverse locking.
2834 if (mtgtdir != msrcdir) {
2835 rc = mdo_is_subdir(info->mti_env,
2836 mdt_object_child(msrcdir), new_fid);
2840 GOTO(out_unlock_old, rc);
2844 /* We used to acquire MDS_INODELOCK_FULL here but we
2845 * can't do this now because a running HSM restore on
2846 * the rename onto victim will hold the layout
2847 * lock. See LU-4002.
2850 lh_newp = &info->mti_lh[MDT_LH_NEW];
2851 mdt_lock_reg_init(lh_newp, LCK_EX);
2852 lock_ibits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE;
2853 if (mdt_object_remote(mtgtdir)) {
2854 rc = mdt_remote_object_lock(info, mtgtdir,
2855 mdt_object_fid(mnew),
2856 &lh_newp->mlh_rreg_lh,
2857 lh_newp->mlh_rreg_mode,
2858 MDS_INODELOCK_LOOKUP,
2861 GOTO(out_unlock_old, rc);
2863 lock_ibits &= ~MDS_INODELOCK_LOOKUP;
2865 rc = mdt_reint_object_lock(info, mnew, lh_newp, lock_ibits,
2868 GOTO(out_unlock_new, rc);
2870 /* get and save version after locking */
2871 mdt_version_get_save(info, mnew, 3);
2872 } else if (rc != -ENOENT) {
2873 GOTO(out_put_old, rc);
2875 lh_oldp = &info->mti_lh[MDT_LH_OLD];
2876 lh_rmt = &info->mti_lh[MDT_LH_RMT];
2877 mdt_lock_reg_init(lh_oldp, LCK_EX);
2878 mdt_lock_reg_init(lh_rmt, LCK_EX);
2879 lock_ibits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_XATTR;
2880 rc = mdt_rename_source_lock(info, msrcdir, mold, lh_oldp,
2881 lh_rmt, lock_ibits, cos_incompat);
2883 GOTO(out_put_old, rc);
2885 mdt_enoent_version_save(info, 3);
2888 /* step 5: rename it */
2889 mdt_reint_init_ma(info, ma);
2891 mdt_fail_write(info->mti_env, info->mti_mdt->mdt_bottom,
2892 OBD_FAIL_MDS_REINT_RENAME_WRITE);
2895 mutex_lock(&mnew->mot_lov_mutex);
2897 rc = mdo_rename(info->mti_env, mdt_object_child(msrcdir),
2898 mdt_object_child(mtgtdir), old_fid, &rr->rr_name,
2899 mnew != NULL ? mdt_object_child(mnew) : NULL,
2900 &rr->rr_tgt_name, ma);
2903 mutex_unlock(&mnew->mot_lov_mutex);
2905 /* handle last link of tgt object */
2907 mdt_counter_incr(req, LPROC_MDT_RENAME,
2908 ktime_us_delta(ktime_get(), kstart));
2910 mdt_handle_last_unlink(info, mnew, ma);
2911 discard = mdt_dom_check_for_discard(info, mnew);
2913 mdt_rename_counter_tally(info, info->mti_mdt, req,
2915 ktime_us_delta(ktime_get(), kstart));
2921 mdt_object_unlock(info, mnew, lh_newp, rc);
2923 mdt_object_unlock(info, NULL, lh_rmt, rc);
2924 mdt_object_unlock(info, mold, lh_oldp, rc);
2926 if (mnew && !discard)
2927 mdt_object_put(info->mti_env, mnew);
2929 mdt_object_put(info->mti_env, mold);
2931 mdt_object_unlock(info, mtgtdir, lh_tgtdirp, rc);
2932 mdt_object_unlock(info, msrcdir, lh_srcdirp, rc);
2934 if (lustre_handle_is_used(&rename_lh))
2935 mdt_rename_unlock(&rename_lh);
2937 mdt_object_put(info->mti_env, mtgtdir);
2939 mdt_object_put(info->mti_env, msrcdir);
2941 /* The DoM discard can be done right in the place above where it is
2942 * assigned, meanwhile it is done here after rename unlock due to
2943 * compatibility with old clients, for them the discard blocks
2944 * the main thread until completion. Check LU-11359 for details.
2947 mdt_dom_discard_data(info, mnew);
2948 mdt_object_put(info->mti_env, mnew);
2953 static int mdt_reint_resync(struct mdt_thread_info *info,
2954 struct mdt_lock_handle *lhc)
2956 struct mdt_reint_record *rr = &info->mti_rr;
2957 struct ptlrpc_request *req = mdt_info_req(info);
2958 struct md_attr *ma = &info->mti_attr;
2959 struct mdt_object *mo;
2960 struct ldlm_lock *lease;
2961 struct mdt_body *repbody;
2962 struct md_layout_change layout = { .mlc_mirror_id = rr->rr_mirror_id };
2967 DEBUG_REQ(D_INODE, req, DFID", FLR file resync", PFID(rr->rr_fid1));
2969 if (info->mti_dlm_req)
2970 ldlm_request_cancel(req, info->mti_dlm_req, 0, LATF_SKIP);
2972 mo = mdt_object_find(info->mti_env, info->mti_mdt, rr->rr_fid1);
2974 GOTO(out, rc = PTR_ERR(mo));
2976 if (!mdt_object_exists(mo))
2977 GOTO(out_obj, rc = -ENOENT);
2979 if (!S_ISREG(lu_object_attr(&mo->mot_obj)))
2980 GOTO(out_obj, rc = -EINVAL);
2982 if (mdt_object_remote(mo))
2983 GOTO(out_obj, rc = -EREMOTE);
2985 lease = ldlm_handle2lock(rr->rr_lease_handle);
2987 GOTO(out_obj, rc = -ESTALE);
2989 /* It's really necessary to grab open_sem and check if the lease lock
2990 * has been lost. There would exist a concurrent writer coming in and
2991 * generating some dirty data in memory cache, the writeback would fail
2992 * after the layout version is increased by MDS_REINT_RESYNC RPC.
2994 if (!down_write_trylock(&mo->mot_open_sem))
2995 GOTO(out_put_lease, rc = -EBUSY);
2997 lock_res_and_lock(lease);
2998 lease_broken = ldlm_is_cancel(lease);
2999 unlock_res_and_lock(lease);
3001 GOTO(out_unlock, rc = -EBUSY);
3003 /* the file has yet opened by anyone else after we took the lease. */
3004 layout.mlc_opc = MD_LAYOUT_RESYNC;
3005 lhc = &info->mti_lh[MDT_LH_LOCAL];
3006 rc = mdt_layout_change(info, mo, lhc, &layout);
3008 GOTO(out_unlock, rc);
3010 mdt_object_unlock(info, mo, lhc, 0);
3012 ma->ma_need = MA_INODE;
3014 rc = mdt_attr_get_complex(info, mo, ma);
3016 GOTO(out_unlock, rc);
3018 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
3019 mdt_pack_attr2body(info, repbody, &ma->ma_attr, mdt_object_fid(mo));
3023 up_write(&mo->mot_open_sem);
3025 LDLM_LOCK_PUT(lease);
3027 mdt_object_put(info->mti_env, mo);
3029 mdt_client_compatibility(info);
3030 rc2 = mdt_fix_reply(info);
3036 struct mdt_reinter {
3037 int (*mr_handler)(struct mdt_thread_info *, struct mdt_lock_handle *);
3038 enum lprocfs_extra_opc mr_extra_opc;
3041 static const struct mdt_reinter mdt_reinters[] = {
3043 .mr_handler = &mdt_reint_setattr,
3044 .mr_extra_opc = MDS_REINT_SETATTR,
3047 .mr_handler = &mdt_reint_create,
3048 .mr_extra_opc = MDS_REINT_CREATE,
3051 .mr_handler = &mdt_reint_link,
3052 .mr_extra_opc = MDS_REINT_LINK,
3055 .mr_handler = &mdt_reint_unlink,
3056 .mr_extra_opc = MDS_REINT_UNLINK,
3059 .mr_handler = &mdt_reint_rename,
3060 .mr_extra_opc = MDS_REINT_RENAME,
3063 .mr_handler = &mdt_reint_open,
3064 .mr_extra_opc = MDS_REINT_OPEN,
3066 [REINT_SETXATTR] = {
3067 .mr_handler = &mdt_reint_setxattr,
3068 .mr_extra_opc = MDS_REINT_SETXATTR,
3071 .mr_handler = &mdt_reint_unlink,
3072 .mr_extra_opc = MDS_REINT_UNLINK,
3075 .mr_handler = &mdt_reint_migrate,
3076 .mr_extra_opc = MDS_REINT_RENAME,
3079 .mr_handler = &mdt_reint_resync,
3080 .mr_extra_opc = MDS_REINT_RESYNC,
3084 int mdt_reint_rec(struct mdt_thread_info *info,
3085 struct mdt_lock_handle *lhc)
3087 const struct mdt_reinter *mr;
3091 if (!(info->mti_rr.rr_opcode < ARRAY_SIZE(mdt_reinters)))
3094 mr = &mdt_reinters[info->mti_rr.rr_opcode];
3095 if (mr->mr_handler == NULL)
3098 rc = (*mr->mr_handler)(info, lhc);
3100 lprocfs_counter_incr(ptlrpc_req2svc(mdt_info_req(info))->srv_stats,
3101 PTLRPC_LAST_CNTR + mr->mr_extra_opc);