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/
30 * Lustre is a trademark of Sun Microsystems, Inc.
32 * lustre/mdt/mdt_open.c
34 * Lustre Metadata Target (mdt) open/close file handling
36 * Author: Huang Hua <huanghua@clusterfs.com>
39 #define DEBUG_SUBSYSTEM S_MDS
41 #include <lustre_acl.h>
42 #include <lustre_mds.h>
43 #include <lustre_swab.h>
44 #include "mdt_internal.h"
45 #include <lustre_nodemap.h>
47 /* we do nothing because we do not have refcount now */
48 static void mdt_mfd_get(void *mfdp)
52 static struct portals_handle_ops mfd_open_handle_ops = {
53 .hop_addref = mdt_mfd_get,
57 /* Create a new mdt_file_data struct, initialize it,
58 * and insert it to global hash table */
59 struct mdt_file_data *mdt_mfd_new(const struct mdt_export_data *med)
61 struct mdt_file_data *mfd;
66 INIT_LIST_HEAD_RCU(&mfd->mfd_open_handle.h_link);
67 mfd->mfd_open_handle.h_owner = med;
68 INIT_LIST_HEAD(&mfd->mfd_list);
69 class_handle_hash(&mfd->mfd_open_handle, &mfd_open_handle_ops);
76 * Find the mfd pointed to by handle in global hash table.
77 * In case of replay the handle is obsoleted
78 * but mfd can be found in mfd list by that handle.
79 * Callers need to be holding med_open_lock.
81 struct mdt_file_data *mdt_open_handle2mfd(struct mdt_export_data *med,
82 const struct lustre_handle *open_handle,
83 bool is_replay_or_resent)
85 struct mdt_file_data *mfd;
88 LASSERT(open_handle != NULL);
89 mfd = class_handle2object(open_handle->cookie, med);
90 /* during dw/setattr replay the mfd can be found by old handle */
91 if (mfd == NULL && is_replay_or_resent) {
92 list_for_each_entry(mfd, &med->med_open_head, mfd_list) {
93 if (mfd->mfd_open_handle_old.cookie ==
104 void mdt_mfd_free(struct mdt_file_data *mfd)
106 LASSERT(list_empty(&mfd->mfd_list));
107 OBD_FREE_RCU(mfd, sizeof *mfd, &mfd->mfd_open_handle);
110 static int mdt_create_data(struct mdt_thread_info *info,
111 struct mdt_object *p, struct mdt_object *o)
113 struct md_op_spec *spec = &info->mti_spec;
114 struct md_attr *ma = &info->mti_attr;
118 if (!md_should_create(spec->sp_cr_flags))
121 ma->ma_need = MA_INODE | MA_LOV;
123 mutex_lock(&o->mot_lov_mutex);
124 if (!o->mot_lov_created) {
125 rc = mdo_create_data(info->mti_env,
126 p ? mdt_object_child(p) : NULL,
127 mdt_object_child(o), spec, ma);
129 rc = mdt_attr_get_complex(info, o, ma);
131 if (rc == 0 && ma->ma_valid & MA_LOV)
132 o->mot_lov_created = 1;
135 mutex_unlock(&o->mot_lov_mutex);
139 int mdt_write_read(struct mdt_object *o)
143 spin_lock(&o->mot_write_lock);
144 rc = o->mot_write_count;
145 spin_unlock(&o->mot_write_lock);
149 int mdt_write_get(struct mdt_object *o)
153 spin_lock(&o->mot_write_lock);
154 if (o->mot_write_count < 0)
157 o->mot_write_count++;
158 spin_unlock(&o->mot_write_lock);
163 void mdt_write_put(struct mdt_object *o)
166 spin_lock(&o->mot_write_lock);
167 o->mot_write_count--;
168 spin_unlock(&o->mot_write_lock);
172 static int mdt_write_deny(struct mdt_object *o)
176 spin_lock(&o->mot_write_lock);
177 if (o->mot_write_count > 0)
180 o->mot_write_count--;
181 spin_unlock(&o->mot_write_lock);
185 static void mdt_write_allow(struct mdt_object *o)
188 spin_lock(&o->mot_write_lock);
189 o->mot_write_count++;
190 spin_unlock(&o->mot_write_lock);
194 /* there can be no real transaction so prepare the fake one */
195 static void mdt_empty_transno(struct mdt_thread_info *info, int rc)
197 struct mdt_device *mdt = info->mti_mdt;
198 struct ptlrpc_request *req = mdt_info_req(info);
199 struct tg_export_data *ted;
200 struct lsd_client_data *lcd;
203 if (mdt_rdonly(req->rq_export))
206 /* transaction has occurred already */
207 if (lustre_msg_get_transno(req->rq_repmsg) != 0)
210 if (tgt_is_multimodrpcs_client(req->rq_export)) {
213 /* generate an empty transaction to get a transno
215 th = dt_trans_create(info->mti_env, mdt->mdt_bottom);
217 rc = dt_trans_start(info->mti_env, mdt->mdt_bottom, th);
218 dt_trans_stop(info->mti_env, mdt->mdt_bottom, th);
223 spin_lock(&mdt->mdt_lut.lut_translock);
225 if (info->mti_transno != 0) {
226 struct obd_export *exp = req->rq_export;
228 CERROR("%s: replay trans %llu NID %s: rc = %d\n",
229 mdt_obd_name(mdt), info->mti_transno,
230 obd_export_nid2str(exp), rc);
231 spin_unlock(&mdt->mdt_lut.lut_translock);
234 } else if (info->mti_transno == 0) {
235 info->mti_transno = ++mdt->mdt_lut.lut_last_transno;
237 /* should be replay */
238 if (info->mti_transno > mdt->mdt_lut.lut_last_transno)
239 mdt->mdt_lut.lut_last_transno = info->mti_transno;
241 spin_unlock(&mdt->mdt_lut.lut_translock);
243 CDEBUG(D_INODE, "transno = %llu, last_committed = %llu\n",
245 req->rq_export->exp_obd->obd_last_committed);
247 req->rq_transno = info->mti_transno;
248 lustre_msg_set_transno(req->rq_repmsg, info->mti_transno);
250 /* update lcd in memory only for resent cases */
251 ted = &req->rq_export->exp_target_data;
253 mutex_lock(&ted->ted_lcd_lock);
255 if (info->mti_transno < lcd->lcd_last_transno &&
256 info->mti_transno != 0) {
257 /* This should happen during replay. Do not update
258 * last rcvd info if replay req transno < last transno,
259 * otherwise the following resend(after replay) can not
260 * be checked correctly by xid */
261 mutex_unlock(&ted->ted_lcd_lock);
262 CDEBUG(D_HA, "%s: transno = %llu < last_transno = %llu\n",
263 mdt_obd_name(mdt), info->mti_transno,
264 lcd->lcd_last_transno);
268 if (lustre_msg_get_opc(req->rq_reqmsg) == MDS_CLOSE) {
269 if (info->mti_transno != 0)
270 lcd->lcd_last_close_transno = info->mti_transno;
271 lcd->lcd_last_close_xid = req->rq_xid;
272 lcd->lcd_last_close_result = rc;
274 /* VBR: save versions in last_rcvd for reconstruct. */
275 __u64 *pre_versions = lustre_msg_get_versions(req->rq_repmsg);
277 lcd->lcd_pre_versions[0] = pre_versions[0];
278 lcd->lcd_pre_versions[1] = pre_versions[1];
279 lcd->lcd_pre_versions[2] = pre_versions[2];
280 lcd->lcd_pre_versions[3] = pre_versions[3];
282 if (info->mti_transno != 0)
283 lcd->lcd_last_transno = info->mti_transno;
285 lcd->lcd_last_xid = req->rq_xid;
286 lcd->lcd_last_result = rc;
287 lcd->lcd_last_data = info->mti_opdata;
289 mutex_unlock(&ted->ted_lcd_lock);
294 void mdt_mfd_set_mode(struct mdt_file_data *mfd, u64 open_flags)
296 LASSERT(mfd != NULL);
298 CDEBUG(D_DENTRY, DFID " Change mfd open_flags %#llo -> %#llo.\n",
299 PFID(mdt_object_fid(mfd->mfd_object)), mfd->mfd_open_flags,
302 mfd->mfd_open_flags = open_flags;
306 * prep ma_lmm/ma_lmv for md_attr from reply
308 static void mdt_prep_ma_buf_from_rep(struct mdt_thread_info *info,
309 struct mdt_object *obj,
312 LASSERT(ma->ma_lmv == NULL && ma->ma_lmm == NULL);
313 if (S_ISDIR(obj->mot_header.loh_attr)) {
314 ma->ma_lmv = req_capsule_server_get(info->mti_pill,
316 ma->ma_lmv_size = req_capsule_get_size(info->mti_pill,
319 if (ma->ma_lmv_size > 0)
320 ma->ma_need |= MA_LMV;
322 ma->ma_lmm = req_capsule_server_get(info->mti_pill,
324 ma->ma_lmm_size = req_capsule_get_size(info->mti_pill,
327 if (ma->ma_lmm_size > 0)
328 ma->ma_need |= MA_LOV;
332 static int mdt_mfd_open(struct mdt_thread_info *info, struct mdt_object *p,
333 struct mdt_object *o, u64 open_flags, int created,
334 struct ldlm_reply *rep)
336 struct ptlrpc_request *req = mdt_info_req(info);
337 struct mdt_export_data *med = &req->rq_export->exp_mdt_data;
338 struct mdt_file_data *mfd;
339 struct md_attr *ma = &info->mti_attr;
340 struct lu_attr *la = &ma->ma_attr;
341 struct mdt_body *repbody;
346 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
348 isreg = S_ISREG(la->la_mode);
349 isdir = S_ISDIR(la->la_mode);
350 if (isreg && !(ma->ma_valid & MA_LOV) &&
351 !(open_flags & MDS_OPEN_RELEASE)) {
353 * No EA, check whether it is will set regEA and dirEA since in
354 * above attr get, these size might be zero, so reset it, to
355 * retrieve the MD after create obj.
357 ma->ma_lmm_size = req_capsule_get_size(info->mti_pill,
360 /* in replay case, p == NULL */
361 rc = mdt_create_data(info, p, o);
365 if (exp_connect_flags(req->rq_export) & OBD_CONNECT_DISP_STRIPE)
366 mdt_set_disposition(info, rep, DISP_OPEN_STRIPE);
369 CDEBUG(D_INODE, "after open, ma_valid bit = %#llx lmm_size = %d\n",
370 ma->ma_valid, ma->ma_lmm_size);
372 if (ma->ma_valid & MA_LOV) {
373 LASSERT(ma->ma_lmm_size != 0);
374 repbody->mbo_eadatasize = ma->ma_lmm_size;
376 repbody->mbo_valid |= OBD_MD_FLDIREA;
378 repbody->mbo_valid |= OBD_MD_FLEASIZE;
381 if (ma->ma_valid & MA_LMV) {
382 LASSERT(ma->ma_lmv_size != 0);
383 repbody->mbo_eadatasize = ma->ma_lmv_size;
385 repbody->mbo_valid |= OBD_MD_FLDIREA | OBD_MD_MEA;
388 if (open_flags & MDS_FMODE_WRITE)
389 rc = mdt_write_get(o);
390 else if (open_flags & MDS_FMODE_EXEC)
391 rc = mdt_write_deny(o);
396 rc = mo_open(info->mti_env, mdt_object_child(o),
397 created ? open_flags | MDS_OPEN_CREATED : open_flags);
399 /* If we allow the client to chgrp (CFS_SETGRP_PERM), but the
400 * client does not know which suppgid should be sent to the MDS,
401 * or some other(s) changed the target file's GID after this RPC
402 * sent to the MDS with the suppgid as the original GID, then we
403 * give the client another chance to send the right suppgid. */
405 allow_client_chgrp(info, lu_ucred(info->mti_env)))
406 mdt_set_disposition(info, rep, DISP_OPEN_DENY);
411 mfd = mdt_mfd_new(med);
413 GOTO(err_out, rc = -ENOMEM);
416 * Keep a reference on this object for this open, and is
417 * released by mdt_mfd_close().
419 mdt_object_get(info->mti_env, o);
421 mfd->mfd_xid = req->rq_xid;
424 * @open_flags is always not zero. At least it should be FMODE_READ,
425 * FMODE_WRITE or MDS_FMODE_EXEC.
427 LASSERT(open_flags != 0);
430 mdt_mfd_set_mode(mfd, open_flags);
432 atomic_inc(&o->mot_open_count);
433 if (open_flags & MDS_OPEN_LEASE)
434 atomic_inc(&o->mot_lease_count);
437 if (req_is_replay(req)) {
438 struct mdt_file_data *old_mfd;
439 /* Check wheather old cookie already exist in
440 * the list, becasue when do recovery, client
441 * might be disconnected from server, and
442 * restart replay, so there maybe some orphan
443 * mfd here, we should remove them */
444 LASSERT(info->mti_rr.rr_open_handle != NULL);
445 spin_lock(&med->med_open_lock);
446 old_mfd = mdt_open_handle2mfd(med, info->mti_rr.rr_open_handle,
448 if (old_mfd != NULL) {
449 CDEBUG(D_HA, "delete orphan mfd = %p, fid = "DFID", "
450 "cookie = %#llx\n", mfd,
451 PFID(mdt_object_fid(mfd->mfd_object)),
452 info->mti_rr.rr_open_handle->cookie);
453 class_handle_unhash(&old_mfd->mfd_open_handle);
454 list_del_init(&old_mfd->mfd_list);
455 spin_unlock(&med->med_open_lock);
456 /* no attr update for that close */
458 ma->ma_valid |= MA_FLAGS;
459 ma->ma_attr_flags |= MDS_RECOV_OPEN;
460 mdt_mfd_close(info, old_mfd);
461 ma->ma_attr_flags &= ~MDS_RECOV_OPEN;
462 ma->ma_valid &= ~MA_FLAGS;
464 spin_unlock(&med->med_open_lock);
465 CDEBUG(D_HA, "orphan mfd not found, fid = "DFID", "
467 PFID(mdt_object_fid(mfd->mfd_object)),
468 info->mti_rr.rr_open_handle->cookie);
471 CDEBUG(D_HA, "Store old cookie %#llx in new mfd\n",
472 info->mti_rr.rr_open_handle->cookie);
474 mfd->mfd_open_handle_old = *info->mti_rr.rr_open_handle;
477 repbody->mbo_open_handle.cookie = mfd->mfd_open_handle.h_cookie;
479 if (req->rq_export->exp_disconnected) {
480 spin_lock(&med->med_open_lock);
481 class_handle_unhash(&mfd->mfd_open_handle);
482 list_del_init(&mfd->mfd_list);
483 spin_unlock(&med->med_open_lock);
484 mdt_mfd_close(info, mfd);
486 spin_lock(&med->med_open_lock);
487 list_add_tail(&mfd->mfd_list, &med->med_open_head);
488 spin_unlock(&med->med_open_lock);
491 mdt_empty_transno(info, rc);
496 if (open_flags & MDS_FMODE_WRITE)
498 else if (open_flags & MDS_FMODE_EXEC)
504 static int mdt_finish_open(struct mdt_thread_info *info,
505 struct mdt_object *p, struct mdt_object *o,
506 u64 open_flags, int created,
507 struct ldlm_reply *rep)
509 struct ptlrpc_request *req = mdt_info_req(info);
510 struct obd_export *exp = req->rq_export;
511 struct mdt_export_data *med = &req->rq_export->exp_mdt_data;
512 struct md_attr *ma = &info->mti_attr;
513 struct lu_attr *la = &ma->ma_attr;
514 struct mdt_file_data *mfd;
515 struct mdt_body *repbody;
517 int isreg, isdir, islnk;
521 LASSERT(ma->ma_valid & MA_INODE);
523 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
525 isreg = S_ISREG(la->la_mode);
526 isdir = S_ISDIR(la->la_mode);
527 islnk = S_ISLNK(la->la_mode);
528 mdt_pack_attr2body(info, repbody, la, mdt_object_fid(o));
530 /* compatibility check for 2.10 clients when it tries to open mirrored
531 * files. 2.10 clients don't verify overlapping components so they
532 * would read and write mirrored files just as if they were normal
533 * PFL files, which will cause the problem that sycned mirrors actually
534 * contain different data.
535 * Older clients are not a concern here because they don't even
536 * understand PFL layout. */
537 if (isreg && !exp_connect_flr(exp) && ma->ma_valid & MA_LOV &&
538 mdt_lmm_is_flr(ma->ma_lmm)) {
539 /* LU-10286: for simplicity clients who don't understand
540 * mirrored layout(with connect flag OBD_CONNECT2_FLR) won't
541 * be able to open mirrored files */
545 /* LU-2275, simulate broken behaviour (esp. prevalent in
546 * pre-2.4 servers where a very strange reply is sent on error
547 * that looks like it was actually almost successful and a
548 * failure at the same time.) */
549 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_NEGATIVE_POSITIVE)) {
550 mdt_set_disposition(info, rep, DISP_OPEN_OPEN |
554 if (open_flags & MDS_OPEN_LOCK)
555 mdt_set_disposition(info, rep, DISP_OPEN_LOCK);
560 #ifdef CONFIG_FS_POSIX_ACL
561 if (exp_connect_flags(exp) & OBD_CONNECT_ACL) {
562 struct lu_nodemap *nodemap = nodemap_get_from_exp(exp);
564 RETURN(PTR_ERR(nodemap));
566 rc = mdt_pack_acl2body(info, repbody, o, nodemap);
567 nodemap_putref(nodemap);
574 * If we are following a symlink, don't open; and do not return open
575 * handle for special nodes as client required.
577 if (islnk || (!isreg && !isdir &&
578 (exp_connect_flags(req->rq_export) & OBD_CONNECT_NODEVOH))) {
579 lustre_msg_set_transno(req->rq_repmsg, 0);
584 * We need to return the existing object's fid back, so it is done here,
585 * after preparing the reply.
587 if (!created && (open_flags & MDS_OPEN_EXCL) &&
588 (open_flags & MDS_OPEN_CREAT))
591 /* This can't be done earlier, we need to return reply body */
593 if (open_flags & (MDS_OPEN_CREAT | MDS_FMODE_WRITE)) {
594 /* We are trying to create or write an existing dir. */
597 } else if (open_flags & MDS_OPEN_DIRECTORY)
600 if (OBD_FAIL_CHECK_RESET(OBD_FAIL_MDS_OPEN_CREATE,
601 OBD_FAIL_MDS_LDLM_REPLY_NET | OBD_FAIL_ONCE))
605 if (lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT) {
606 spin_lock(&med->med_open_lock);
607 list_for_each(t, &med->med_open_head) {
608 mfd = list_entry(t, struct mdt_file_data, mfd_list);
609 if (mfd->mfd_xid == req->rq_xid)
613 spin_unlock(&med->med_open_lock);
616 repbody->mbo_open_handle.cookie =
617 mfd->mfd_open_handle.h_cookie;
618 /* set repbody->ea_size for resent case */
619 if (ma->ma_valid & MA_LOV) {
620 LASSERT(ma->ma_lmm_size != 0);
621 repbody->mbo_eadatasize = ma->ma_lmm_size;
623 repbody->mbo_valid |= OBD_MD_FLDIREA;
625 repbody->mbo_valid |= OBD_MD_FLEASIZE;
627 mdt_set_disposition(info, rep, DISP_OPEN_OPEN);
632 rc = mdt_mfd_open(info, p, o, open_flags, created, rep);
634 mdt_set_disposition(info, rep, DISP_OPEN_OPEN);
639 void mdt_reconstruct_open(struct mdt_thread_info *info,
640 struct mdt_lock_handle *lhc)
642 const struct lu_env *env = info->mti_env;
643 struct mdt_device *mdt = info->mti_mdt;
644 struct req_capsule *pill = info->mti_pill;
645 struct ptlrpc_request *req = mdt_info_req(info);
646 struct md_attr *ma = &info->mti_attr;
647 struct mdt_reint_record *rr = &info->mti_rr;
648 u64 open_flags = info->mti_spec.sp_cr_flags;
649 struct ldlm_reply *ldlm_rep;
650 struct mdt_object *parent;
651 struct mdt_object *child;
652 struct mdt_body *repbody;
657 LASSERT(pill->rc_fmt == &RQF_LDLM_INTENT_OPEN);
658 ldlm_rep = req_capsule_server_get(pill, &RMF_DLM_REP);
659 repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
661 ma->ma_need = MA_INODE | MA_HSM;
664 opdata = mdt_req_from_lrd(req, info->mti_reply_data);
665 mdt_set_disposition(info, ldlm_rep, opdata);
667 CDEBUG(D_INODE, "This is reconstruct open: disp=%#llx, result=%d\n",
668 ldlm_rep->lock_policy_res1, req->rq_status);
670 if (mdt_get_disposition(ldlm_rep, DISP_OPEN_CREATE) &&
672 /* We did not create successfully, return error to client. */
673 GOTO(out, rc = req->rq_status);
675 if (mdt_get_disposition(ldlm_rep, DISP_OPEN_CREATE)) {
676 struct obd_export *exp = req->rq_export;
678 * We failed after creation, but we do not know in which step
679 * we failed. So try to check the child object.
681 parent = mdt_object_find(env, mdt, rr->rr_fid1);
682 if (IS_ERR(parent)) {
683 rc = PTR_ERR(parent);
684 LCONSOLE_WARN("Parent "DFID" lookup error %d."
685 " Evicting client %s with export %s.\n",
686 PFID(rr->rr_fid1), rc,
687 obd_uuid2str(&exp->exp_client_uuid),
688 obd_export_nid2str(exp));
689 mdt_export_evict(exp);
693 child = mdt_object_find(env, mdt, rr->rr_fid2);
696 LCONSOLE_WARN("cannot lookup child "DFID": rc = %d; "
697 "evicting client %s with export %s\n",
698 PFID(rr->rr_fid2), rc,
699 obd_uuid2str(&exp->exp_client_uuid),
700 obd_export_nid2str(exp));
701 mdt_object_put(env, parent);
702 mdt_export_evict(exp);
706 if (unlikely(mdt_object_remote(child))) {
707 mdt_object_put(env, parent);
708 mdt_object_put(env, child);
709 /* the child object was created on remote server */
710 if (!mdt_is_dne_client(exp))
711 /* Return -EIO for old client */
712 GOTO(out, rc = -EIO);
713 repbody->mbo_fid1 = *rr->rr_fid2;
714 repbody->mbo_valid |= (OBD_MD_FLID | OBD_MD_MDS);
717 if (mdt_object_exists(child)) {
718 mdt_prep_ma_buf_from_rep(info, child, ma);
719 rc = mdt_attr_get_complex(info, child, ma);
721 rc = mdt_finish_open(info, parent, child,
722 open_flags, 1, ldlm_rep);
723 mdt_object_put(env, parent);
724 mdt_object_put(env, child);
726 mdt_pack_size2body(info, rr->rr_fid2,
730 /* the child does not exist, we should do regular open */
731 mdt_object_put(env, parent);
732 mdt_object_put(env, child);
734 /* We did not try to create, so we are a pure open */
735 rc = mdt_reint_open(info, lhc);
739 lustre_msg_set_status(req->rq_repmsg, req->rq_status);
740 LASSERT(ergo(rc < 0, lustre_msg_get_transno(req->rq_repmsg) == 0));
743 static int mdt_open_by_fid(struct mdt_thread_info *info, struct ldlm_reply *rep)
745 u64 open_flags = info->mti_spec.sp_cr_flags;
746 struct mdt_reint_record *rr = &info->mti_rr;
747 struct md_attr *ma = &info->mti_attr;
748 struct mdt_object *o;
752 o = mdt_object_find(info->mti_env, info->mti_mdt, rr->rr_fid2);
754 RETURN(rc = PTR_ERR(o));
756 if (unlikely(mdt_object_remote(o))) {
757 /* the child object was created on remote server */
758 struct mdt_body *repbody;
760 mdt_set_disposition(info, rep, (DISP_IT_EXECD |
763 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
764 repbody->mbo_fid1 = *rr->rr_fid2;
765 repbody->mbo_valid |= (OBD_MD_FLID | OBD_MD_MDS);
768 if (mdt_object_exists(o)) {
769 mdt_set_disposition(info, rep, (DISP_IT_EXECD |
772 mdt_prep_ma_buf_from_rep(info, o, ma);
773 rc = mdt_attr_get_complex(info, o, ma);
775 rc = mdt_finish_open(info, NULL, o, open_flags,
782 mdt_object_put(info->mti_env, o);
786 /* lock object for open */
787 static int mdt_object_open_lock(struct mdt_thread_info *info,
788 struct mdt_object *obj,
789 struct mdt_lock_handle *lhc,
792 struct md_attr *ma = &info->mti_attr;
793 __u64 open_flags = info->mti_spec.sp_cr_flags;
795 enum ldlm_mode lm = LCK_CR;
796 bool acq_lease = !!(open_flags & MDS_OPEN_LEASE);
797 bool try_layout = false;
798 bool create_layout = false;
800 int dom_stripes = LMM_NO_DOM;
801 bool dom_lock = false;
806 mdt_lock_handle_init(lhc);
808 if (req_is_replay(mdt_info_req(info)))
811 if (S_ISREG(lu_object_attr(&obj->mot_obj))) {
812 if (ma->ma_need & MA_LOV && !(ma->ma_valid & MA_LOV) &&
813 md_should_create(open_flags))
814 create_layout = true;
815 if (exp_connect_layout(info->mti_exp) && !create_layout &&
816 ma->ma_need & MA_LOV)
819 /* DoM files can have just MDT stripe or combined MDT + OST
821 * - In the first case the open for read/write will do IO to
822 * the MDT stripe and it makes sense to take IO lock in
823 * advance along with OPEN even if it is blocking lock.
824 * - In the second case it is just size of MDT stripe and it
825 * is quite unlikely that client will write into it, though
826 * it may read it. So IO lock will be taken optionally if it
827 * is non-blocking one.
829 if (ma->ma_valid & MA_LOV && ma->ma_lmm != NULL)
830 dom_stripes = mdt_lmm_dom_entry(ma->ma_lmm);
832 if (dom_stripes == LMM_DOM_ONLY &&
833 info->mti_mdt->mdt_opts.mo_dom_lock > 0 &&
834 !mdt_dom_client_has_lock(info, mdt_object_fid(obj)))
839 /* lease open, acquire write mode of open sem */
840 down_write(&obj->mot_open_sem);
842 /* Lease exists and ask for new lease */
843 if (atomic_read(&obj->mot_lease_count) > 0) {
844 /* only exclusive open is supported, so lease
845 * are conflicted to each other */
846 GOTO(out, rc = -EBUSY);
849 /* Lease must be with open lock */
850 if (!(open_flags & MDS_OPEN_LOCK)) {
851 CERROR("%s: Request lease for file:"DFID ", but open lock "
852 "is missed, open_flags = %#llo : rc = %d\n",
853 mdt_obd_name(info->mti_mdt),
854 PFID(mdt_object_fid(obj)), open_flags, -EPROTO);
855 GOTO(out, rc = -EPROTO);
858 /* XXX: only exclusive open is supported. */
860 *ibits = MDS_INODELOCK_OPEN;
862 /* never grant LCK_EX layout lock to client */
864 } else { /* normal open */
865 /* normal open holds read mode of open sem */
866 down_read(&obj->mot_open_sem);
868 if (open_flags & MDS_OPEN_LOCK) {
869 if (open_flags & MDS_FMODE_WRITE)
871 else if (open_flags & MDS_FMODE_EXEC)
876 *ibits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_OPEN;
877 } else if (atomic_read(&obj->mot_lease_count) > 0) {
878 if (open_flags & MDS_FMODE_WRITE)
884 *ibits = MDS_INODELOCK_OPEN;
887 lhc = &info->mti_lh[MDT_LH_LOCAL];
888 } else if (dom_lock) {
889 lm = (open_flags & MDS_FMODE_WRITE) ? LCK_PW : LCK_PR;
890 if (info->mti_mdt->mdt_opts.mo_dom_lock ==
891 TRYLOCK_DOM_ON_OPEN) {
892 trybits |= MDS_INODELOCK_DOM |
893 MDS_INODELOCK_LAYOUT;
895 /* mo_dom_lock == ALWAYS_DOM_LOCK_ON_OPEN */
896 *ibits = MDS_INODELOCK_DOM;
897 if (info->mti_mdt->mdt_opts.mo_dom_read_open) {
898 trybits |= MDS_INODELOCK_LAYOUT;
903 CDEBUG(D_INODE, "normal open:"DFID" lease count: %d, lm: %d\n",
904 PFID(mdt_object_fid(obj)),
905 atomic_read(&obj->mot_lease_count), lm);
908 mdt_lock_reg_init(lhc, lm);
910 /* Return lookup lock to validate inode at the client side.
911 * This is pretty important otherwise MDT will return layout
912 * lock for each open.
913 * However this is a double-edged sword because changing
914 * permission will revoke a huge number of LOOKUP locks.
916 if (!OBD_FAIL_CHECK(OBD_FAIL_MDS_NO_LL_OPEN) && try_layout) {
917 if (!(*ibits & MDS_INODELOCK_LOOKUP))
918 trybits |= MDS_INODELOCK_LOOKUP;
919 trybits |= MDS_INODELOCK_LAYOUT;
922 if (*ibits | trybits)
923 rc = mdt_object_lock_try(info, obj, lhc, ibits, trybits, false);
925 CDEBUG(D_INODE, "%s: Requested bits lock:"DFID ", ibits = %#llx/%#llx"
926 ", open_flags = %#llo, try_layout = %d : rc = %d\n",
927 mdt_obd_name(info->mti_mdt), PFID(mdt_object_fid(obj)),
928 *ibits, trybits, open_flags, try_layout, rc);
930 /* will change layout, revoke layout locks by enqueuing EX lock. */
931 if (rc == 0 && create_layout) {
932 struct mdt_lock_handle *ll = &info->mti_lh[MDT_LH_LAYOUT];
934 CDEBUG(D_INODE, "Will create layout, get EX layout lock:"DFID
935 ", open_flags = %#llo\n",
936 PFID(mdt_object_fid(obj)), open_flags);
938 /* We cannot enqueue another lock for the same resource we
939 * already have a lock for, due to mechanics of waiting list
940 * iterating in ldlm, see LU-3601.
941 * As such we'll drop the open lock we just got above here,
942 * it's ok not to have this open lock as it's main purpose is to
943 * flush unused cached client open handles. */
944 if (lustre_handle_is_used(&lhc->mlh_reg_lh))
945 mdt_object_unlock(info, obj, lhc, 1);
947 LASSERT(!try_layout);
948 mdt_lock_handle_init(ll);
949 mdt_lock_reg_init(ll, LCK_EX);
950 rc = mdt_object_lock(info, obj, ll, MDS_INODELOCK_LAYOUT);
952 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_LL_BLOCK, 2);
955 /* Check if there is any other open handles after acquiring
956 * open lock. At this point, caching open handles have been revoked
958 * XXX: Now only exclusive open is supported. Need to check the
959 * type of open for generic lease support. */
960 if (rc == 0 && acq_lease) {
961 struct ptlrpc_request *req = mdt_info_req(info);
962 struct mdt_export_data *med = &req->rq_export->exp_mdt_data;
963 struct mdt_file_data *mfd;
964 bool is_replay_or_resent;
967 /* For lease: application can open a file and then apply lease,
968 * @handle contains original open handle in that case.
969 * In recovery, open REQ will be replayed and the lease REQ may
970 * be resent that means the open handle is already stale, so we
971 * need to fix it up here by finding new handle. */
972 is_replay_or_resent = req_is_replay(req) ||
973 lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT;
975 /* if the request is _not_ a replay request, rr_open_handle
976 * may be used to hold an open file handle which is issuing the
977 * lease request, so that this openhandle doesn't count. */
978 mfd = mdt_open_handle2mfd(med, info->mti_rr.rr_open_handle,
979 is_replay_or_resent);
983 CDEBUG(D_INODE, "acq_lease "DFID": openers: %d, want: %d\n",
984 PFID(mdt_object_fid(obj)),
985 atomic_read(&obj->mot_open_count), open_count);
987 if (atomic_read(&obj->mot_open_count) > open_count)
988 GOTO(out, rc = -EBUSY);
996 static void mdt_object_open_unlock(struct mdt_thread_info *info,
997 struct mdt_object *obj,
998 struct mdt_lock_handle *lhc,
1001 __u64 open_flags = info->mti_spec.sp_cr_flags;
1002 struct mdt_lock_handle *ll = &info->mti_lh[MDT_LH_LOCAL];
1005 if (req_is_replay(mdt_info_req(info)))
1008 /* Release local lock - the lock put in MDT_LH_LOCAL will never
1009 * return to client side. */
1010 if (lustre_handle_is_used(&ll->mlh_reg_lh))
1011 mdt_object_unlock(info, obj, ll, 1);
1013 ll = &info->mti_lh[MDT_LH_LAYOUT];
1014 /* Release local layout lock, layout was created */
1015 if (lustre_handle_is_used(&ll->mlh_reg_lh)) {
1016 LASSERT(!(ibits & MDS_INODELOCK_LAYOUT));
1017 mdt_object_unlock(info, obj, ll, 1);
1020 if (open_flags & MDS_OPEN_LEASE)
1021 up_write(&obj->mot_open_sem);
1023 up_read(&obj->mot_open_sem);
1025 /* Cross-ref case, the lock should be returned to the client */
1026 if (ibits == 0 || rc == -MDT_EREMOTE_OPEN)
1029 if (!(open_flags & MDS_OPEN_LOCK) && !(ibits & MDS_INODELOCK_LAYOUT) &&
1030 !(ibits & MDS_INODELOCK_DOM)) {
1031 /* for the open request, the lock will only return to client
1032 * if open or layout lock is granted. */
1036 if (rc != 0 || !lustre_handle_is_used(&lhc->mlh_reg_lh)) {
1037 struct ldlm_reply *ldlm_rep;
1039 ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
1040 mdt_clear_disposition(info, ldlm_rep, DISP_OPEN_LOCK);
1041 if (lustre_handle_is_used(&lhc->mlh_reg_lh))
1042 mdt_object_unlock(info, obj, lhc, 1);
1048 * Check release is permitted for the current HSM flags.
1050 static bool mdt_hsm_release_allow(const struct md_attr *ma)
1052 if (!(ma->ma_valid & MA_HSM))
1055 if (ma->ma_hsm.mh_flags & (HS_DIRTY|HS_NORELEASE|HS_LOST))
1058 if (!(ma->ma_hsm.mh_flags & HS_ARCHIVED))
1064 static int mdt_open_by_fid_lock(struct mdt_thread_info *info,
1065 struct ldlm_reply *rep,
1066 struct mdt_lock_handle *lhc)
1068 const struct lu_env *env = info->mti_env;
1069 struct mdt_device *mdt = info->mti_mdt;
1070 u64 open_flags = info->mti_spec.sp_cr_flags;
1071 struct mdt_reint_record *rr = &info->mti_rr;
1072 struct md_attr *ma = &info->mti_attr;
1073 struct mdt_object *parent = NULL;
1074 struct mdt_object *o;
1075 bool object_locked = false;
1080 if (md_should_create(open_flags)) {
1081 if (!lu_fid_eq(rr->rr_fid1, rr->rr_fid2)) {
1082 parent = mdt_object_find(env, mdt, rr->rr_fid1);
1083 if (IS_ERR(parent)) {
1084 CDEBUG(D_INODE, "Fail to find parent "DFID
1085 " for anonymous created %ld, try to"
1086 " use server-side parent.\n",
1087 PFID(rr->rr_fid1), PTR_ERR(parent));
1092 ma->ma_need |= MA_PFID;
1095 o = mdt_object_find(env, mdt, rr->rr_fid2);
1097 GOTO(out_parent_put, rc = PTR_ERR(o));
1099 if (mdt_object_remote(o)) {
1100 CDEBUG(D_INFO, "%s: "DFID" is on remote MDT.\n",
1101 mdt_obd_name(info->mti_mdt),
1103 GOTO(out, rc = -EREMOTE);
1104 } else if (!mdt_object_exists(o)) {
1105 mdt_set_disposition(info, rep,
1109 GOTO(out, rc = -ENOENT);
1112 mdt_set_disposition(info, rep, (DISP_IT_EXECD | DISP_LOOKUP_EXECD));
1114 mdt_prep_ma_buf_from_rep(info, o, ma);
1115 if (open_flags & MDS_OPEN_RELEASE)
1116 ma->ma_need |= MA_HSM;
1117 rc = mdt_attr_get_complex(info, o, ma);
1121 /* We should not change file's existing LOV EA */
1122 if (S_ISREG(lu_object_attr(&o->mot_obj)) &&
1123 open_flags & MDS_OPEN_HAS_EA && ma->ma_valid & MA_LOV)
1124 GOTO(out, rc = -EEXIST);
1126 /* If a release request, check file open flags are fine and ask for an
1127 * exclusive open access. */
1128 if (open_flags & MDS_OPEN_RELEASE && !mdt_hsm_release_allow(ma))
1129 GOTO(out, rc = -EPERM);
1131 rc = mdt_check_resent_lock(info, o, lhc);
1134 } else if (rc > 0) {
1135 rc = mdt_object_open_lock(info, o, lhc, &ibits);
1136 object_locked = true;
1138 GOTO(out_unlock, rc);
1141 if (ma->ma_valid & MA_PFID) {
1142 parent = mdt_object_find(env, mdt, &ma->ma_pfid);
1143 if (IS_ERR(parent)) {
1144 CDEBUG(D_INODE, "Fail to find parent "DFID
1145 " for anonymous created %ld, try to"
1146 " use system default.\n",
1147 PFID(&ma->ma_pfid), PTR_ERR(parent));
1152 rc = mdt_finish_open(info, parent, o, open_flags, 0, rep);
1154 mdt_set_disposition(info, rep, DISP_LOOKUP_POS);
1155 if (open_flags & MDS_OPEN_LOCK)
1156 mdt_set_disposition(info, rep, DISP_OPEN_LOCK);
1157 if (open_flags & MDS_OPEN_LEASE)
1158 mdt_set_disposition(info, rep, DISP_OPEN_LEASE);
1160 GOTO(out_unlock, rc);
1164 mdt_object_open_unlock(info, o, lhc, ibits, rc);
1166 mdt_object_put(env, o);
1168 mdt_pack_size2body(info, rr->rr_fid2, &lhc->mlh_reg_lh);
1171 mdt_object_put(env, parent);
1175 /* Cross-ref request. Currently it can only be a pure open (w/o create) */
1176 static int mdt_cross_open(struct mdt_thread_info *info,
1177 const struct lu_fid *parent_fid,
1178 const struct lu_fid *fid,
1179 struct ldlm_reply *rep, u64 open_flags)
1181 struct md_attr *ma = &info->mti_attr;
1182 struct mdt_object *o;
1186 o = mdt_object_find(info->mti_env, info->mti_mdt, fid);
1188 RETURN(rc = PTR_ERR(o));
1190 if (mdt_object_remote(o)) {
1191 /* Something is wrong here, the object is on another MDS! */
1192 CERROR("%s: "DFID" isn't on this server!: rc = %d\n",
1193 mdt_obd_name(info->mti_mdt), PFID(fid), -EFAULT);
1194 LU_OBJECT_DEBUG(D_WARNING, info->mti_env,
1196 "Object isn't on this server! FLD error?");
1199 if (mdt_object_exists(o)) {
1202 /* Do permission check for cross-open after converting
1203 * MDS_OPEN_* flags to MAY_* permission mask.
1205 mask = mds_accmode(open_flags);
1207 rc = mo_permission(info->mti_env, NULL,
1208 mdt_object_child(o), NULL, mask);
1212 mdt_prep_ma_buf_from_rep(info, o, ma);
1213 rc = mdt_attr_get_complex(info, o, ma);
1217 mdt_pack_secctx_in_reply(info, o);
1219 rc = mdt_finish_open(info, NULL, o, open_flags, 0, rep);
1222 * Something is wrong here. lookup was positive but
1223 * there is no object!
1225 CERROR("%s: "DFID" doesn't exist!: rc = %d\n",
1226 mdt_obd_name(info->mti_mdt), PFID(fid), -EFAULT);
1231 mdt_object_put(info->mti_env, o);
1237 * find root object and take its xattr lock if it's on remote MDT, later create
1238 * may use fs default striping (which is stored in root xattr).
1240 static int mdt_lock_root_xattr(struct mdt_thread_info *info,
1241 struct mdt_device *mdt)
1243 struct mdt_object *md_root = mdt->mdt_md_root;
1244 struct lustre_handle lhroot;
1247 if (md_root == NULL) {
1248 lu_root_fid(&info->mti_tmp_fid1);
1249 md_root = mdt_object_find(info->mti_env, mdt,
1250 &info->mti_tmp_fid1);
1251 if (IS_ERR(md_root))
1252 return PTR_ERR(md_root);
1254 spin_lock(&mdt->mdt_lock);
1255 if (mdt->mdt_md_root != NULL) {
1256 spin_unlock(&mdt->mdt_lock);
1258 LASSERTF(mdt->mdt_md_root == md_root,
1259 "Different root object ("
1260 DFID") instances, %p, %p\n",
1261 PFID(&info->mti_tmp_fid1),
1262 mdt->mdt_md_root, md_root);
1263 LASSERT(atomic_read(
1264 &md_root->mot_obj.lo_header->loh_ref) > 1);
1266 mdt_object_put(info->mti_env, md_root);
1268 mdt->mdt_md_root = md_root;
1269 spin_unlock(&mdt->mdt_lock);
1273 if (md_root->mot_cache_attr || !mdt_object_remote(md_root))
1276 rc = mdt_remote_object_lock(info, md_root, mdt_object_fid(md_root),
1277 &lhroot, LCK_PR, MDS_INODELOCK_XATTR,
1282 md_root->mot_cache_attr = 1;
1284 /* don't cancel this lock, so that we know the cached xattr is valid. */
1285 ldlm_lock_decref(&lhroot, LCK_PR);
1290 int mdt_reint_open(struct mdt_thread_info *info, struct mdt_lock_handle *lhc)
1292 struct mdt_device *mdt = info->mti_mdt;
1293 struct ptlrpc_request *req = mdt_info_req(info);
1294 struct mdt_object *parent;
1295 struct mdt_object *child;
1296 struct mdt_lock_handle *lh;
1297 struct ldlm_reply *ldlm_rep;
1298 struct mdt_body *repbody;
1299 struct lu_fid *child_fid = &info->mti_tmp_fid1;
1300 struct md_attr *ma = &info->mti_attr;
1301 u64 open_flags = info->mti_spec.sp_cr_flags;
1303 struct mdt_reint_record *rr = &info->mti_rr;
1306 int object_locked = 0;
1310 OBD_FAIL_TIMEOUT_ORSET(OBD_FAIL_MDS_PAUSE_OPEN, OBD_FAIL_ONCE,
1311 (obd_timeout + 1) / 4);
1313 mdt_counter_incr(req, LPROC_MDT_OPEN);
1314 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1316 ma->ma_need = MA_INODE;
1319 LASSERT(info->mti_pill->rc_fmt == &RQF_LDLM_INTENT_OPEN);
1320 ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
1322 if (unlikely(open_flags & MDS_OPEN_JOIN_FILE)) {
1323 CERROR("file join is not supported anymore.\n");
1324 GOTO(out, result = err_serious(-EOPNOTSUPP));
1326 msg_flags = lustre_msg_get_flags(req->rq_reqmsg);
1328 if ((open_flags & (MDS_OPEN_HAS_EA | MDS_OPEN_HAS_OBJS)) &&
1329 info->mti_spec.u.sp_ea.eadata == NULL)
1330 GOTO(out, result = err_serious(-EINVAL));
1332 if (open_flags & MDS_FMODE_WRITE &&
1333 exp_connect_flags(req->rq_export) & OBD_CONNECT_RDONLY)
1334 GOTO(out, result = -EROFS);
1336 CDEBUG(D_INODE, "I am going to open "DFID"/("DNAME"->"DFID") "
1337 "cr_flag=%#llo mode=0%06o msg_flag=0x%x\n",
1338 PFID(rr->rr_fid1), PNAME(&rr->rr_name), PFID(rr->rr_fid2),
1339 open_flags, ma->ma_attr.la_mode, msg_flags);
1341 if (info->mti_cross_ref) {
1342 /* This is cross-ref open */
1343 mdt_set_disposition(info, ldlm_rep,
1344 (DISP_IT_EXECD | DISP_LOOKUP_EXECD |
1346 result = mdt_cross_open(info, rr->rr_fid2, rr->rr_fid1,
1347 ldlm_rep, open_flags);
1349 } else if (req_is_replay(req)) {
1350 result = mdt_open_by_fid(info, ldlm_rep);
1352 if (result != -ENOENT)
1355 /* We didn't find the correct object, so we need to re-create it
1356 * via a regular replay. */
1357 if (!(open_flags & MDS_OPEN_CREAT)) {
1358 DEBUG_REQ(D_ERROR, req,
1359 "OPEN & CREAT not in open replay/by_fid.");
1360 GOTO(out, result = -EFAULT);
1362 CDEBUG(D_INFO, "No object(1), continue as regular open.\n");
1363 } else if (open_flags & (MDS_OPEN_BY_FID | MDS_OPEN_LOCK)) {
1365 * MDS_OPEN_LOCK is checked for backward compatibility with 2.1
1368 result = mdt_open_by_fid_lock(info, ldlm_rep, lhc);
1370 CDEBUG(D_INFO, "no object for "DFID": %d\n",
1371 PFID(rr->rr_fid2), result);
1375 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_OPEN_PACK))
1376 GOTO(out, result = err_serious(-ENOMEM));
1378 mdt_set_disposition(info, ldlm_rep,
1379 (DISP_IT_EXECD | DISP_LOOKUP_EXECD));
1381 if (!lu_name_is_valid(&rr->rr_name))
1382 GOTO(out, result = -EPROTO);
1384 result = mdt_lock_root_xattr(info, mdt);
1389 lh = &info->mti_lh[MDT_LH_PARENT];
1390 mdt_lock_pdo_init(lh, (open_flags & MDS_OPEN_CREAT) ? LCK_PW : LCK_PR,
1393 parent = mdt_object_find(info->mti_env, mdt, rr->rr_fid1);
1395 GOTO(out, result = PTR_ERR(parent));
1397 result = mdt_object_lock(info, parent, lh, MDS_INODELOCK_UPDATE);
1399 mdt_object_put(info->mti_env, parent);
1403 /* get and check version of parent */
1404 result = mdt_version_get_check(info, parent, 0);
1406 GOTO(out_parent, result);
1408 fid_zero(child_fid);
1410 result = mdo_lookup(info->mti_env, mdt_object_child(parent),
1411 &rr->rr_name, child_fid, &info->mti_spec);
1413 LASSERTF(ergo(result == 0, fid_is_sane(child_fid)),
1414 "looking for "DFID"/"DNAME", found FID = "DFID"\n",
1415 PFID(mdt_object_fid(parent)), PNAME(&rr->rr_name),
1418 if (result != 0 && result != -ENOENT && result != -ESTALE)
1419 GOTO(out_parent, result);
1421 if (result == -ENOENT || result == -ESTALE) {
1422 /* If the object is dead, let's check if the object
1423 * is being migrated to a new object */
1424 if (result == -ESTALE) {
1425 struct lu_buf lmv_buf;
1427 lmv_buf.lb_buf = info->mti_xattr_buf;
1428 lmv_buf.lb_len = sizeof(info->mti_xattr_buf);
1429 rc = mo_xattr_get(info->mti_env,
1430 mdt_object_child(parent),
1431 &lmv_buf, XATTR_NAME_LMV);
1433 struct lmv_mds_md_v1 *lmv;
1435 lmv = lmv_buf.lb_buf;
1436 if (le32_to_cpu(lmv->lmv_hash_type) &
1437 LMV_HASH_FLAG_MIGRATION) {
1438 /* Get the new parent FID and retry */
1439 mdt_object_unlock_put(info, parent,
1441 mdt_lock_handle_init(lh);
1443 (struct lu_fid *)rr->rr_fid1,
1444 &lmv->lmv_stripe_fids[1]);
1450 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
1451 if (result == -ESTALE) {
1453 * -ESTALE means the parent is a dead(unlinked) dir, so
1454 * it should return -ENOENT to in accordance with the
1455 * original mds implementaion.
1457 GOTO(out_parent, result = -ENOENT);
1460 if (!(open_flags & MDS_OPEN_CREAT))
1461 GOTO(out_parent, result);
1462 if (mdt_rdonly(req->rq_export))
1463 GOTO(out_parent, result = -EROFS);
1464 *child_fid = *info->mti_rr.rr_fid2;
1465 LASSERTF(fid_is_sane(child_fid), "fid="DFID"\n",
1467 /* In the function below, .hs_keycmp resolves to
1468 * lu_obj_hop_keycmp() */
1469 /* coverity[overrun-buffer-val] */
1470 child = mdt_object_new(info->mti_env, mdt, child_fid);
1473 * Check for O_EXCL is moved to the mdt_finish_open(),
1474 * we need to return FID back in that case.
1476 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1477 child = mdt_object_find(info->mti_env, mdt, child_fid);
1480 GOTO(out_parent, result = PTR_ERR(child));
1482 /** check version of child */
1483 rc = mdt_version_get_check(info, child, 1);
1485 GOTO(out_child, result = rc);
1487 if (result == -ENOENT) {
1488 /* Create under OBF and .lustre is not permitted */
1489 if (!fid_is_md_operative(rr->rr_fid1))
1490 GOTO(out_child, result = -EPERM);
1492 /* save versions in reply */
1493 mdt_version_get_save(info, parent, 0);
1494 mdt_version_get_save(info, child, 1);
1496 /* version of child will be changed */
1497 tgt_vbr_obj_set(info->mti_env, mdt_obj2dt(child));
1499 /* Not found and with MDS_OPEN_CREAT: let's create it. */
1500 mdt_set_disposition(info, ldlm_rep, DISP_OPEN_CREATE);
1502 /* Let lower layers know what is lock mode on directory. */
1503 info->mti_spec.sp_cr_mode =
1504 mdt_dlm_mode2mdl_mode(lh->mlh_pdo_mode);
1506 /* Don't do lookup sanity check. We know name doesn't exist. */
1507 info->mti_spec.sp_cr_lookup = 0;
1508 info->mti_spec.sp_feat = &dt_directory_features;
1510 result = mdo_create(info->mti_env, mdt_object_child(parent),
1511 &rr->rr_name, mdt_object_child(child),
1512 &info->mti_spec, &info->mti_attr);
1513 if (result == -ERESTART) {
1514 mdt_clear_disposition(info, ldlm_rep, DISP_OPEN_CREATE);
1515 GOTO(out_child, result);
1517 mdt_prep_ma_buf_from_rep(info, child, ma);
1518 /* XXX: we should call this once, see few lines below */
1520 result = mdt_attr_get_complex(info, child, ma);
1523 GOTO(out_child, result);
1526 mdt_counter_incr(req, LPROC_MDT_MKNOD);
1529 * The object is on remote node, return its FID for remote open.
1531 if (mdt_object_remote(child)) {
1533 * Check if this lock already was sent to client and
1534 * this is resent case. For resent case do not take lock
1535 * again, use what is already granted.
1537 LASSERT(lhc != NULL);
1539 rc = mdt_check_resent_lock(info, child, lhc);
1541 GOTO(out_child, result = rc);
1542 } else if (rc > 0) {
1543 mdt_lock_handle_init(lhc);
1544 mdt_lock_reg_init(lhc, LCK_PR);
1546 rc = mdt_object_lock(info, child, lhc,
1547 MDS_INODELOCK_LOOKUP);
1549 repbody->mbo_fid1 = *mdt_object_fid(child);
1550 repbody->mbo_valid |= (OBD_MD_FLID | OBD_MD_MDS);
1554 result = -MDT_EREMOTE_OPEN;
1555 GOTO(out_child, result);
1556 } else if (mdt_object_exists(child)) {
1557 /* We have to get attr & LOV EA & HSM for this
1559 mdt_prep_ma_buf_from_rep(info, child, ma);
1560 ma->ma_need |= MA_HSM;
1561 result = mdt_attr_get_complex(info, child, ma);
1563 GOTO(out_child, result);
1565 /* Object does not exist. Likely FS corruption. */
1566 CERROR("%s: name '"DNAME"' present, but FID "
1567 DFID" is invalid\n", mdt_obd_name(info->mti_mdt),
1568 PNAME(&rr->rr_name), PFID(child_fid));
1569 GOTO(out_child, result = -EIO);
1573 mdt_pack_secctx_in_reply(info, child);
1575 rc = mdt_check_resent_lock(info, child, lhc);
1577 GOTO(out_child, result = rc);
1578 } else if (rc == 0) {
1579 /* the open lock might already be gotten in
1580 * ldlm_handle_enqueue() */
1581 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT);
1582 if (open_flags & MDS_OPEN_LOCK)
1583 mdt_set_disposition(info, ldlm_rep, DISP_OPEN_LOCK);
1585 /* get openlock if this isn't replay and client requested it */
1586 if (!req_is_replay(req)) {
1587 rc = mdt_object_open_lock(info, child, lhc, &ibits);
1590 GOTO(out_child_unlock, result = rc);
1591 else if (open_flags & MDS_OPEN_LOCK)
1592 mdt_set_disposition(info, ldlm_rep,
1596 /* Try to open it now. */
1597 rc = mdt_finish_open(info, parent, child, open_flags,
1601 /* openlock will be released if mdt_finish_open() failed */
1602 mdt_clear_disposition(info, ldlm_rep, DISP_OPEN_LOCK);
1604 if (created && (open_flags & MDS_OPEN_VOLATILE)) {
1605 CERROR("%s: cannot open volatile file "DFID", orphan "
1606 "file will be left in PENDING directory until "
1607 "next reboot, rc = %d\n", mdt_obd_name(mdt),
1608 PFID(mdt_object_fid(child)), rc);
1609 GOTO(out_child_unlock, result);
1615 rc = mdo_unlink(info->mti_env,
1616 mdt_object_child(parent),
1617 mdt_object_child(child),
1619 &info->mti_attr, 0);
1621 CERROR("%s: "DFID" cleanup of open: rc = %d\n",
1622 mdt_obd_name(info->mti_mdt),
1623 PFID(mdt_object_fid(child)), rc);
1624 mdt_clear_disposition(info, ldlm_rep, DISP_OPEN_CREATE);
1630 mdt_object_open_unlock(info, child, lhc, ibits, result);
1632 mdt_object_put(info->mti_env, child);
1634 mdt_pack_size2body(info, child_fid, &lhc->mlh_reg_lh);
1636 mdt_object_unlock_put(info, parent, lh, result || !created);
1639 lustre_msg_set_transno(req->rq_repmsg, 0);
1644 * Create an orphan object use local root.
1646 static struct mdt_object *mdt_orphan_open(struct mdt_thread_info *info,
1647 struct mdt_device *mdt,
1648 const struct lu_fid *fid,
1649 struct md_attr *attr, fmode_t fmode)
1651 const struct lu_env *env = info->mti_env;
1652 struct md_op_spec *spec = &info->mti_spec;
1653 struct lu_fid *local_root_fid = &info->mti_tmp_fid1;
1654 struct mdt_object *obj = NULL;
1655 struct mdt_object *local_root;
1656 static const struct lu_name lname = {
1657 .ln_name = "i_am_nobody",
1658 .ln_namelen = sizeof("i_am_nobody") - 1,
1660 struct lu_ucred *uc;
1661 cfs_cap_t uc_cap_save;
1665 rc = dt_root_get(env, mdt->mdt_bottom, local_root_fid);
1667 RETURN(ERR_PTR(rc));
1669 local_root = mdt_object_find(env, mdt, local_root_fid);
1670 if (IS_ERR(local_root))
1673 obj = mdt_object_new(env, mdt, fid);
1675 GOTO(out, rc = PTR_ERR(obj));
1677 spec->sp_cr_lookup = 0;
1678 spec->sp_feat = &dt_directory_features;
1679 spec->sp_cr_mode = MDL_MINMODE; /* no lock */
1680 spec->sp_cr_flags = MDS_OPEN_VOLATILE | fmode;
1681 if (attr->ma_valid & MA_LOV) {
1682 spec->u.sp_ea.eadata = attr->ma_lmm;
1683 spec->u.sp_ea.eadatalen = attr->ma_lmm_size;
1684 spec->sp_cr_flags |= MDS_OPEN_HAS_EA;
1686 spec->sp_cr_flags |= MDS_OPEN_DELAY_CREATE;
1690 uc_cap_save = uc->uc_cap;
1691 uc->uc_cap |= 1 << CFS_CAP_DAC_OVERRIDE;
1692 rc = mdo_create(env, mdt_object_child(local_root), &lname,
1693 mdt_object_child(obj), spec, attr);
1694 uc->uc_cap = uc_cap_save;
1696 CERROR("%s: cannot create volatile file "DFID": rc = %d\n",
1697 mdt_obd_name(mdt), PFID(fid), rc);
1701 rc = mo_open(env, mdt_object_child(obj), MDS_OPEN_CREATED);
1703 CERROR("%s: cannot open volatile file "DFID", orphan "
1704 "file will be left in PENDING directory until "
1705 "next reboot, rc = %d\n", mdt_obd_name(mdt),
1712 mdt_object_put(env, obj);
1715 mdt_object_put(env, local_root);
1719 /* XXX Look into layout in MDT layer. */
1720 static inline int mdt_hsm_set_released(struct lov_mds_md *lmm)
1722 struct lov_comp_md_v1 *comp_v1;
1723 struct lov_mds_md *v1;
1727 if (lmm->lmm_magic == cpu_to_le32(LOV_MAGIC_COMP_V1_DEFINED)) {
1728 comp_v1 = (struct lov_comp_md_v1 *)lmm;
1730 if (comp_v1->lcm_entry_count == 0)
1733 for (i = 0; i < le16_to_cpu(comp_v1->lcm_entry_count); i++) {
1734 off = le32_to_cpu(comp_v1->lcm_entries[i].lcme_offset);
1735 v1 = (struct lov_mds_md *)((char *)comp_v1 + off);
1736 v1->lmm_pattern |= cpu_to_le32(LOV_PATTERN_F_RELEASED);
1739 lmm->lmm_pattern |= cpu_to_le32(LOV_PATTERN_F_RELEASED);
1744 static int mdt_hsm_release(struct mdt_thread_info *info, struct mdt_object *o,
1747 struct mdt_lock_handle *lh = &info->mti_lh[MDT_LH_LAYOUT];
1748 struct lu_ucred *uc = mdt_ucred(info);
1749 struct close_data *data;
1750 struct ldlm_lock *lease;
1751 struct mdt_object *orphan;
1752 struct md_attr *orp_ma;
1760 if (mdt_rdonly(info->mti_exp))
1763 data = req_capsule_client_get(info->mti_pill, &RMF_CLOSE_DATA);
1767 lease = ldlm_handle2lock(&data->cd_handle);
1771 /* try to hold open_sem so that nobody else can open the file */
1772 if (!down_write_trylock(&o->mot_open_sem)) {
1773 ldlm_lock_cancel(lease);
1774 GOTO(out_reprocess, rc = -EBUSY);
1777 /* Check if the lease open lease has already canceled */
1778 lock_res_and_lock(lease);
1779 lease_broken = ldlm_is_cancel(lease);
1780 unlock_res_and_lock(lease);
1782 LDLM_DEBUG(lease, DFID " lease broken? %d",
1783 PFID(mdt_object_fid(o)), lease_broken);
1785 /* Cancel server side lease. Client side counterpart should
1786 * have been cancelled. It's okay to cancel it now as we've
1787 * held mot_open_sem. */
1788 ldlm_lock_cancel(lease);
1790 if (lease_broken) /* don't perform release task */
1791 GOTO(out_unlock, rc = -ESTALE);
1793 if (fid_is_zero(&data->cd_fid) || !fid_is_sane(&data->cd_fid))
1794 GOTO(out_unlock, rc = -EINVAL);
1796 /* ma_need was set before but it seems fine to change it in order to
1797 * avoid modifying the one from RPC */
1798 ma->ma_need = MA_HSM;
1799 rc = mdt_attr_get_complex(info, o, ma);
1801 GOTO(out_unlock, rc);
1803 if (!mdt_hsm_release_allow(ma))
1804 GOTO(out_unlock, rc = -EPERM);
1806 /* already released? */
1807 if (ma->ma_hsm.mh_flags & HS_RELEASED)
1808 GOTO(out_unlock, rc = 0);
1810 /* Compare on-disk and packed data_version */
1811 if (data->cd_data_version != ma->ma_hsm.mh_arch_ver) {
1812 CDEBUG(D_HSM, DFID" data_version mismatches: packed=%llu"
1813 " and on-disk=%llu\n", PFID(mdt_object_fid(o)),
1814 data->cd_data_version, ma->ma_hsm.mh_arch_ver);
1815 GOTO(out_unlock, rc = -EPERM);
1818 ma->ma_valid = MA_INODE;
1819 ma->ma_attr.la_valid &= LA_ATIME | LA_MTIME | LA_CTIME | LA_SIZE;
1820 rc = mo_attr_set(info->mti_env, mdt_object_child(o), ma);
1822 GOTO(out_unlock, rc);
1824 mutex_lock(&o->mot_som_mutex);
1825 rc2 = mdt_set_som(info, o, SOM_FL_STRICT, ma->ma_attr.la_size,
1826 ma->ma_attr.la_blocks);
1827 mutex_unlock(&o->mot_som_mutex);
1830 "%s: File "DFID" SOM update failed: rc = %d\n",
1831 mdt_obd_name(info->mti_mdt),
1832 PFID(mdt_object_fid(o)), rc2);
1835 ma->ma_need = MA_INODE | MA_LOV;
1836 rc = mdt_attr_get_complex(info, o, ma);
1838 GOTO(out_unlock, rc);
1840 if (!(ma->ma_valid & MA_LOV)) {
1841 /* Even empty file are released */
1842 memset(ma->ma_lmm, 0, sizeof(*ma->ma_lmm));
1843 ma->ma_lmm->lmm_magic = cpu_to_le32(LOV_MAGIC_V1_DEFINED);
1844 ma->ma_lmm->lmm_pattern = cpu_to_le32(LOV_PATTERN_RAID0);
1845 ma->ma_lmm->lmm_stripe_size = cpu_to_le32(LOV_MIN_STRIPE_SIZE);
1846 ma->ma_lmm_size = sizeof(*ma->ma_lmm);
1848 /* Magic must be LOV_MAGIC_*_DEFINED or LOD will interpret
1849 * ma_lmm as lov_user_md, then it will be confused by union of
1850 * layout_gen and stripe_offset. */
1851 if ((le32_to_cpu(ma->ma_lmm->lmm_magic) & LOV_MAGIC_MASK) ==
1853 ma->ma_lmm->lmm_magic |= cpu_to_le32(LOV_MAGIC_DEFINED);
1855 GOTO(out_unlock, rc = -EINVAL);
1858 /* Set file as released. */
1859 rc = mdt_hsm_set_released(ma->ma_lmm);
1861 GOTO(out_unlock, rc);
1863 orp_ma = &info->mti_u.hsm.attr;
1864 orp_ma->ma_attr.la_mode = S_IFREG | S_IWUSR;
1865 /* We use root ownership to bypass potential quota
1866 * restrictions on the user and group of the file. */
1867 orp_ma->ma_attr.la_uid = 0;
1868 orp_ma->ma_attr.la_gid = 0;
1869 orp_ma->ma_attr.la_valid = LA_MODE | LA_UID | LA_GID;
1870 orp_ma->ma_lmm = ma->ma_lmm;
1871 orp_ma->ma_lmm_size = ma->ma_lmm_size;
1872 orp_ma->ma_valid = MA_INODE | MA_LOV;
1873 orphan = mdt_orphan_open(info, info->mti_mdt, &data->cd_fid, orp_ma,
1875 if (IS_ERR(orphan)) {
1876 CERROR("%s: cannot open orphan file "DFID": rc = %ld\n",
1877 mdt_obd_name(info->mti_mdt), PFID(&data->cd_fid),
1879 GOTO(out_unlock, rc = PTR_ERR(orphan));
1882 /* Set up HSM attribute for orphan object */
1883 CLASSERT(sizeof(struct hsm_attrs) <= sizeof(info->mti_xattr_buf));
1884 buf = &info->mti_buf;
1885 buf->lb_buf = info->mti_xattr_buf;
1886 buf->lb_len = sizeof(struct hsm_attrs);
1887 ma->ma_hsm.mh_flags |= HS_RELEASED;
1888 lustre_hsm2buf(buf->lb_buf, &ma->ma_hsm);
1889 ma->ma_hsm.mh_flags &= ~HS_RELEASED;
1891 mdt_lock_reg_init(lh, LCK_EX);
1892 rc = mdt_object_lock(info, o, lh, MDS_INODELOCK_LAYOUT |
1893 MDS_INODELOCK_XATTR);
1895 GOTO(out_close, rc);
1897 /* The orphan has root ownership so we need to raise
1898 * CAP_FOWNER to set the HSM attributes. */
1900 uc->uc_cap |= MD_CAP_TO_MASK(CFS_CAP_FOWNER);
1901 rc = mo_xattr_set(info->mti_env, mdt_object_child(orphan), buf,
1905 GOTO(out_layout_lock, rc);
1907 /* Swap layout with orphan objects. */
1908 rc = mo_swap_layouts(info->mti_env, mdt_object_child(o),
1909 mdt_object_child(orphan),
1910 SWAP_LAYOUTS_MDS_HSM);
1914 /* Release exclusive LL */
1915 mdt_object_unlock(info, o, lh, 1);
1917 /* Close orphan object anyway */
1918 rc2 = mo_close(info->mti_env, mdt_object_child(orphan), orp_ma,
1921 CERROR("%s: error closing volatile file "DFID": rc = %d\n",
1922 mdt_obd_name(info->mti_mdt), PFID(&data->cd_fid), rc2);
1923 LU_OBJECT_DEBUG(D_HSM, info->mti_env, &orphan->mot_obj,
1925 mdt_object_put(info->mti_env, orphan);
1928 up_write(&o->mot_open_sem);
1930 /* already released */
1932 struct mdt_body *repbody;
1934 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1935 LASSERT(repbody != NULL);
1936 repbody->mbo_valid |= OBD_MD_CLOSE_INTENT_EXECED;
1940 ldlm_reprocess_all(lease->l_resource);
1941 LDLM_LOCK_PUT(lease);
1949 int mdt_close_handle_layouts(struct mdt_thread_info *info,
1950 struct mdt_object *o, struct md_attr *ma)
1952 struct mdt_lock_handle *lh1 = &info->mti_lh[MDT_LH_NEW];
1953 struct mdt_lock_handle *lh2 = &info->mti_lh[MDT_LH_OLD];
1954 struct close_data *data;
1955 struct ldlm_lock *lease;
1956 struct mdt_object *o1 = o, *o2;
1962 if (exp_connect_flags(info->mti_exp) & OBD_CONNECT_RDONLY)
1965 if (!S_ISREG(lu_object_attr(&o1->mot_obj)))
1968 data = req_capsule_client_get(info->mti_pill, &RMF_CLOSE_DATA);
1972 if (fid_is_zero(&data->cd_fid) || !fid_is_sane(&data->cd_fid))
1975 rc = lu_fid_cmp(&data->cd_fid, mdt_object_fid(o));
1976 if (unlikely(rc == 0))
1979 /* Exchange o1 and o2, to enforce locking order */
1980 swap_objects = (rc < 0);
1982 lease = ldlm_handle2lock(&data->cd_handle);
1986 o2 = mdt_object_find(info->mti_env, info->mti_mdt, &data->cd_fid);
1988 GOTO(out_lease, rc = PTR_ERR(o2));
1990 if (!S_ISREG(lu_object_attr(&o2->mot_obj))) {
1991 swap_objects = false; /* not swapped yet */
1992 GOTO(out_obj, rc = -EINVAL);
1998 rc = mo_permission(info->mti_env, NULL, mdt_object_child(o1), NULL,
2003 rc = mo_permission(info->mti_env, NULL, mdt_object_child(o2), NULL,
2008 /* try to hold open_sem so that nobody else can open the file */
2009 if (!down_write_trylock(&o->mot_open_sem)) {
2010 ldlm_lock_cancel(lease);
2011 GOTO(out_obj, rc = -EBUSY);
2014 /* Check if the lease open lease has already canceled */
2015 lock_res_and_lock(lease);
2016 lease_broken = ldlm_is_cancel(lease);
2017 unlock_res_and_lock(lease);
2019 LDLM_DEBUG(lease, DFID " lease broken? %d",
2020 PFID(mdt_object_fid(o)), lease_broken);
2022 /* Cancel server side lease. Client side counterpart should
2023 * have been cancelled. It's okay to cancel it now as we've
2024 * held mot_open_sem. */
2025 ldlm_lock_cancel(lease);
2028 GOTO(out_unlock_sem, rc = -ESTALE);
2030 mdt_lock_reg_init(lh1, LCK_EX);
2031 rc = mdt_object_lock(info, o1, lh1, MDS_INODELOCK_LAYOUT |
2032 MDS_INODELOCK_XATTR);
2034 GOTO(out_unlock_sem, rc);
2036 mdt_lock_reg_init(lh2, LCK_EX);
2037 rc = mdt_object_lock(info, o2, lh2, MDS_INODELOCK_LAYOUT |
2038 MDS_INODELOCK_XATTR);
2040 GOTO(out_unlock1, rc);
2042 /* Swap layout with orphan object */
2043 if (ma->ma_attr_flags & MDS_CLOSE_LAYOUT_SWAP) {
2044 rc = mo_swap_layouts(info->mti_env, mdt_object_child(o1),
2045 mdt_object_child(o2), 0);
2046 } else if (ma->ma_attr_flags & MDS_CLOSE_LAYOUT_MERGE ||
2047 ma->ma_attr_flags & MDS_CLOSE_LAYOUT_SPLIT) {
2048 struct lu_buf *buf = &info->mti_buf;
2049 struct md_rejig_data mrd;
2051 mrd.mrd_obj = mdt_object_child(o == o1 ? o2 : o1);
2052 if (ma->ma_attr_flags & MDS_CLOSE_LAYOUT_SPLIT)
2053 mrd.mrd_mirror_id = data->cd_mirror_id;
2055 buf->lb_len = sizeof(mrd);
2057 rc = mo_xattr_set(info->mti_env, mdt_object_child(o), buf,
2059 ma->ma_attr_flags & MDS_CLOSE_LAYOUT_SPLIT ?
2060 LU_XATTR_SPLIT : LU_XATTR_MERGE);
2061 if (rc == 0 && ma->ma_attr.la_valid & (LA_SIZE | LA_BLOCKS)) {
2064 mutex_lock(&o->mot_som_mutex);
2065 rc2 = mdt_set_som(info, o, SOM_FL_STRICT,
2066 ma->ma_attr.la_size,
2067 ma->ma_attr.la_blocks);
2068 mutex_unlock(&o->mot_som_mutex);
2070 CERROR(DFID": Setting i_blocks error: %d, "
2071 "i_blocks will be reported wrongly and "
2072 "can only be fixed in next resync\n",
2073 PFID(mdt_object_fid(o)), rc2);
2077 GOTO(out_unlock2, rc);
2082 /* Release exclusive LL */
2083 mdt_object_unlock(info, o2, lh2, 1);
2086 mdt_object_unlock(info, o1, lh1, 1);
2089 up_write(&o->mot_open_sem);
2091 /* already swapped */
2093 struct mdt_body *repbody;
2095 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
2096 LASSERT(repbody != NULL);
2097 repbody->mbo_valid |= OBD_MD_CLOSE_INTENT_EXECED;
2101 mdt_object_put(info->mti_env, swap_objects ? o1 : o2);
2103 ldlm_reprocess_all(lease->l_resource);
2106 LDLM_LOCK_PUT(lease);
2116 static int mdt_close_resync_done(struct mdt_thread_info *info,
2117 struct mdt_object *o, struct md_attr *ma)
2119 struct close_data *data;
2120 struct ldlm_lock *lease;
2121 struct md_layout_change layout = { 0 };
2122 __u32 *resync_ids = NULL;
2123 size_t resync_count = 0;
2128 if (exp_connect_flags(info->mti_exp) & OBD_CONNECT_RDONLY)
2131 if (!S_ISREG(lu_object_attr(&o->mot_obj)))
2134 data = req_capsule_client_get(info->mti_pill, &RMF_CLOSE_DATA);
2138 if (ptlrpc_req_need_swab(mdt_info_req(info)))
2139 lustre_swab_close_data_resync_done(&data->cd_resync);
2141 if (!fid_is_zero(&data->cd_fid))
2144 lease = ldlm_handle2lock(&data->cd_handle);
2148 /* try to hold open_sem so that nobody else can open the file */
2149 if (!down_write_trylock(&o->mot_open_sem)) {
2150 ldlm_lock_cancel(lease);
2151 GOTO(out_reprocess, rc = -EBUSY);
2154 /* Check if the lease open lease has already canceled */
2155 lock_res_and_lock(lease);
2156 lease_broken = ldlm_is_cancel(lease);
2157 unlock_res_and_lock(lease);
2159 LDLM_DEBUG(lease, DFID " lease broken? %d\n",
2160 PFID(mdt_object_fid(o)), lease_broken);
2162 /* Cancel server side lease. Client side counterpart should
2163 * have been cancelled. It's okay to cancel it now as we've
2164 * held mot_open_sem. */
2165 ldlm_lock_cancel(lease);
2167 if (lease_broken) /* don't perform release task */
2168 GOTO(out_unlock, rc = -ESTALE);
2170 resync_count = data->cd_resync.resync_count;
2172 if (resync_count > INLINE_RESYNC_ARRAY_SIZE) {
2175 if (!req_capsule_has_field(info->mti_pill, &RMF_U32,
2177 GOTO(out_unlock, rc = -EPROTO);
2179 OBD_ALLOC(resync_ids, resync_count * sizeof(__u32));
2181 GOTO(out_unlock, rc = -ENOMEM);
2183 data = req_capsule_client_get(info->mti_pill, &RMF_U32);
2184 memcpy(resync_ids, data, resync_count * sizeof(__u32));
2186 layout.mlc_resync_ids = resync_ids;
2188 layout.mlc_resync_ids = data->cd_resync.resync_ids_inline;
2191 layout.mlc_opc = MD_LAYOUT_RESYNC_DONE;
2192 layout.mlc_resync_count = resync_count;
2193 if (ma->ma_attr.la_valid & (LA_SIZE | LA_BLOCKS)) {
2194 layout.mlc_som.lsa_valid = SOM_FL_STRICT;
2195 layout.mlc_som.lsa_size = ma->ma_attr.la_size;
2196 layout.mlc_som.lsa_blocks = ma->ma_attr.la_blocks;
2198 rc = mdt_layout_change(info, o, &layout);
2200 GOTO(out_unlock, rc);
2205 up_write(&o->mot_open_sem);
2207 /* already released */
2209 struct mdt_body *repbody;
2211 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
2212 LASSERT(repbody != NULL);
2213 repbody->mbo_valid |= OBD_MD_CLOSE_INTENT_EXECED;
2217 OBD_FREE(resync_ids, resync_count * sizeof(__u32));
2220 ldlm_reprocess_all(lease->l_resource);
2221 LDLM_LOCK_PUT(lease);
2229 #define MFD_CLOSED(open_flags) ((open_flags) == MDS_FMODE_CLOSED)
2231 static int mdt_mfd_closed(struct mdt_file_data *mfd)
2233 return ((mfd == NULL) || MFD_CLOSED(mfd->mfd_open_flags));
2236 int mdt_mfd_close(struct mdt_thread_info *info, struct mdt_file_data *mfd)
2238 struct mdt_object *o = mfd->mfd_object;
2239 struct md_object *next = mdt_object_child(o);
2240 struct md_attr *ma = &info->mti_attr;
2241 struct lu_fid *ofid = &info->mti_tmp_fid1;
2245 bool discard = false;
2249 open_flags = mfd->mfd_open_flags;
2250 intent = ma->ma_attr_flags & MDS_CLOSE_INTENT;
2251 *ofid = *mdt_object_fid(o);
2253 CDEBUG(D_INODE, "%s: close file "DFID" with intent: %llx\n",
2254 mdt_obd_name(info->mti_mdt), PFID(ofid), intent);
2257 case MDS_HSM_RELEASE: {
2258 rc = mdt_hsm_release(info, o, ma);
2260 CDEBUG(D_HSM, "%s: File " DFID " release failed: %d\n",
2261 mdt_obd_name(info->mti_mdt),
2263 /* continue to close even error occurred. */
2267 case MDS_CLOSE_LAYOUT_MERGE:
2268 case MDS_CLOSE_LAYOUT_SPLIT:
2269 case MDS_CLOSE_LAYOUT_SWAP: {
2270 rc = mdt_close_handle_layouts(info, o, ma);
2273 "%s: cannot swap layout of "DFID": rc = %d\n",
2274 mdt_obd_name(info->mti_mdt),
2276 /* continue to close even if error occurred. */
2280 case MDS_CLOSE_RESYNC_DONE:
2281 rc = mdt_close_resync_done(info, o, ma);
2288 if (S_ISREG(lu_object_attr(&o->mot_obj)) &&
2289 ma->ma_attr.la_valid & (LA_LSIZE | LA_LBLOCKS)) {
2292 rc2 = mdt_lsom_update(info, o, false);
2295 "%s: File " DFID " LSOM failed: rc = %d\n",
2296 mdt_obd_name(info->mti_mdt),
2298 /* continue to close even if error occured. */
2301 if (open_flags & MDS_FMODE_WRITE)
2303 else if (open_flags & MDS_FMODE_EXEC)
2306 /* Update atime on close only. */
2307 if ((open_flags & MDS_FMODE_EXEC || open_flags & MDS_FMODE_READ ||
2308 open_flags & MDS_FMODE_WRITE) && (ma->ma_valid & MA_INODE) &&
2309 (ma->ma_attr.la_valid & LA_ATIME)) {
2310 /* Set the atime only. */
2311 ma->ma_valid = MA_INODE;
2312 ma->ma_attr.la_valid = LA_ATIME;
2313 rc = mo_attr_set(info->mti_env, next, ma);
2316 /* If file data is modified, add the dirty flag. */
2317 if (ma->ma_attr_flags & MDS_DATA_MODIFIED)
2318 rc = mdt_add_dirty_flag(info, o, ma);
2320 ma->ma_need |= MA_INODE;
2321 ma->ma_valid &= ~MA_INODE;
2323 LASSERT(atomic_read(&o->mot_open_count) > 0);
2324 atomic_dec(&o->mot_open_count);
2325 mdt_handle_last_unlink(info, o, ma);
2327 if (!MFD_CLOSED(open_flags)) {
2328 rc = mo_close(info->mti_env, next, ma, open_flags);
2329 discard = mdt_dom_check_for_discard(info, o);
2332 /* adjust open and lease count */
2333 if (open_flags & MDS_OPEN_LEASE) {
2334 LASSERT(atomic_read(&o->mot_lease_count) > 0);
2335 atomic_dec(&o->mot_lease_count);
2339 mdt_object_put(info->mti_env, o);
2342 mdt_dom_discard_data(info, ofid);
2347 int mdt_close_internal(struct mdt_thread_info *info, struct ptlrpc_request *req,
2348 struct mdt_body *repbody)
2350 struct mdt_export_data *med;
2351 struct mdt_file_data *mfd;
2356 med = &req->rq_export->exp_mdt_data;
2357 spin_lock(&med->med_open_lock);
2358 mfd = mdt_open_handle2mfd(med, &info->mti_open_handle,
2359 req_is_replay(req));
2360 if (mdt_mfd_closed(mfd)) {
2361 spin_unlock(&med->med_open_lock);
2362 CDEBUG(D_INODE, "no handle for file close: fid = "DFID
2363 ": cookie = %#llx\n", PFID(info->mti_rr.rr_fid1),
2364 info->mti_open_handle.cookie);
2365 /** not serious error since bug 3633 */
2368 class_handle_unhash(&mfd->mfd_open_handle);
2369 list_del_init(&mfd->mfd_list);
2370 spin_unlock(&med->med_open_lock);
2371 ret = mdt_mfd_close(info, mfd);
2374 RETURN(rc ? rc : ret);
2377 int mdt_close(struct tgt_session_info *tsi)
2379 struct mdt_thread_info *info = tsi2mdt_info(tsi);
2380 struct ptlrpc_request *req = tgt_ses_req(tsi);
2381 struct md_attr *ma = &info->mti_attr;
2382 struct mdt_body *repbody = NULL;
2386 mdt_counter_incr(req, LPROC_MDT_CLOSE);
2387 /* Close may come with the Size-on-MDS update. Unpack it. */
2388 rc = mdt_close_unpack(info);
2390 GOTO(out, rc = err_serious(rc));
2392 /* These fields are no longer used and are left for compatibility.
2393 * size is always zero */
2394 req_capsule_set_size(info->mti_pill, &RMF_MDT_MD, RCL_SERVER,
2396 req_capsule_set_size(info->mti_pill, &RMF_LOGCOOKIES, RCL_SERVER,
2398 rc = req_capsule_server_pack(info->mti_pill);
2399 if (mdt_check_resent(info, mdt_reconstruct_generic, NULL)) {
2400 mdt_client_compatibility(info);
2402 mdt_fix_reply(info);
2403 mdt_exit_ucred(info);
2404 GOTO(out, rc = lustre_msg_get_status(req->rq_repmsg));
2407 /* Continue to close handle even if we can not pack reply */
2409 repbody = req_capsule_server_get(info->mti_pill,
2411 ma->ma_lmm = req_capsule_server_get(info->mti_pill,
2413 ma->ma_lmm_size = req_capsule_get_size(info->mti_pill,
2416 ma->ma_need = MA_INODE | MA_LOV;
2417 repbody->mbo_eadatasize = 0;
2418 repbody->mbo_aclsize = 0;
2420 rc = err_serious(rc);
2423 rc = mdt_close_internal(info, req, repbody);
2425 mdt_empty_transno(info, rc);
2427 if (repbody != NULL) {
2428 mdt_client_compatibility(info);
2429 rc = mdt_fix_reply(info);
2432 mdt_exit_ucred(info);
2433 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_CLOSE_PACK))
2434 GOTO(out, rc = err_serious(-ENOMEM));
2436 if (OBD_FAIL_CHECK_RESET(OBD_FAIL_MDS_CLOSE_NET_REP,
2437 OBD_FAIL_MDS_CLOSE_NET_REP))
2438 tsi->tsi_reply_fail_id = OBD_FAIL_MDS_CLOSE_NET_REP;
2440 mdt_thread_info_fini(info);
2441 RETURN(rc ? rc : ret);