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_open.c
33 * Lustre Metadata Target (mdt) open/close file handling
35 * Author: Huang Hua <huanghua@clusterfs.com>
38 #define DEBUG_SUBSYSTEM S_MDS
40 #include <lustre_acl.h>
41 #include <lustre_mds.h>
42 #include <lustre_swab.h>
43 #include "mdt_internal.h"
44 #include <lustre_nodemap.h>
46 static const char mfd_open_handle_owner[] = "mdt";
48 static int mdt_open_by_fid(struct mdt_thread_info *info, struct ldlm_reply *rep,
49 struct mdt_lock_handle *lhc);
51 /* Create a new mdt_file_data struct, initialize it,
52 * and insert it to global hash table */
53 struct mdt_file_data *mdt_mfd_new(const struct mdt_export_data *med)
55 struct mdt_file_data *mfd;
60 refcount_set(&mfd->mfd_open_handle.h_ref, 1);
61 INIT_HLIST_NODE(&mfd->mfd_open_handle.h_link);
63 INIT_LIST_HEAD(&mfd->mfd_list);
64 class_handle_hash(&mfd->mfd_open_handle, mfd_open_handle_owner);
71 * Find the mfd pointed to by handle in global hash table.
72 * In case of replay the handle is obsoleted
73 * but mfd can be found in mfd list by that handle.
74 * Callers need to be holding med_open_lock.
76 struct mdt_file_data *mdt_open_handle2mfd(struct mdt_export_data *med,
77 const struct lustre_handle *open_handle,
78 bool is_replay_or_resent)
80 struct mdt_file_data *mfd;
83 LASSERT(open_handle != NULL);
84 mfd = class_handle2object(open_handle->cookie, mfd_open_handle_owner);
86 refcount_dec(&mfd->mfd_open_handle.h_ref);
88 /* during dw/setattr replay the mfd can be found by old handle */
89 if ((!mfd || mfd->mfd_owner != med) && is_replay_or_resent) {
90 list_for_each_entry(mfd, &med->med_open_head, mfd_list) {
91 if (mfd->mfd_open_handle_old.cookie ==
102 void mdt_mfd_free(struct mdt_file_data *mfd)
104 LASSERT(refcount_read(&mfd->mfd_open_handle.h_ref) == 1);
105 LASSERT(list_empty(&mfd->mfd_list));
106 OBD_FREE_PRE(mfd, sizeof(*mfd), "rcu");
107 kfree_rcu(mfd, mfd_open_handle.h_rcu);
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)
144 spin_lock(&o->mot_write_lock);
145 rc = o->mot_write_count;
146 spin_unlock(&o->mot_write_lock);
150 int mdt_write_get(struct mdt_object *o)
155 spin_lock(&o->mot_write_lock);
156 if (o->mot_write_count < 0)
159 o->mot_write_count++;
160 spin_unlock(&o->mot_write_lock);
165 void mdt_write_put(struct mdt_object *o)
168 spin_lock(&o->mot_write_lock);
169 o->mot_write_count--;
170 spin_unlock(&o->mot_write_lock);
174 static int mdt_write_deny(struct mdt_object *o)
179 spin_lock(&o->mot_write_lock);
180 if (o->mot_write_count > 0)
183 o->mot_write_count--;
184 spin_unlock(&o->mot_write_lock);
188 static void mdt_write_allow(struct mdt_object *o)
191 spin_lock(&o->mot_write_lock);
192 o->mot_write_count++;
193 spin_unlock(&o->mot_write_lock);
197 /* there can be no real transaction so prepare the fake one */
198 static void mdt_empty_transno(struct mdt_thread_info *info, int rc)
200 struct mdt_device *mdt = info->mti_mdt;
201 struct ptlrpc_request *req = mdt_info_req(info);
202 struct tg_export_data *ted;
203 struct lsd_client_data *lcd;
206 if (mdt_rdonly(req->rq_export))
209 /* transaction has occurred already */
210 if (lustre_msg_get_transno(req->rq_repmsg) != 0)
213 if (tgt_is_multimodrpcs_client(req->rq_export)) {
216 /* generate an empty transaction to get a transno
218 th = dt_trans_create(info->mti_env, mdt->mdt_bottom);
220 rc = dt_trans_start(info->mti_env, mdt->mdt_bottom, th);
221 dt_trans_stop(info->mti_env, mdt->mdt_bottom, th);
226 spin_lock(&mdt->mdt_lut.lut_translock);
228 if (info->mti_transno != 0) {
229 struct obd_export *exp = req->rq_export;
231 CERROR("%s: replay trans %llu NID %s: rc = %d\n",
232 mdt_obd_name(mdt), info->mti_transno,
233 obd_export_nid2str(exp), rc);
234 spin_unlock(&mdt->mdt_lut.lut_translock);
237 } else if (info->mti_transno == 0) {
238 info->mti_transno = ++mdt->mdt_lut.lut_last_transno;
240 /* should be replay */
241 if (info->mti_transno > mdt->mdt_lut.lut_last_transno)
242 mdt->mdt_lut.lut_last_transno = info->mti_transno;
244 spin_unlock(&mdt->mdt_lut.lut_translock);
246 CDEBUG(D_INODE, "transno = %llu, last_committed = %llu\n",
248 req->rq_export->exp_obd->obd_last_committed);
250 req->rq_transno = info->mti_transno;
251 lustre_msg_set_transno(req->rq_repmsg, info->mti_transno);
253 /* update lcd in memory only for resent cases */
254 ted = &req->rq_export->exp_target_data;
256 mutex_lock(&ted->ted_lcd_lock);
258 if (info->mti_transno < lcd->lcd_last_transno &&
259 info->mti_transno != 0) {
260 /* This should happen during replay. Do not update
261 * last rcvd info if replay req transno < last transno,
262 * otherwise the following resend(after replay) can not
263 * be checked correctly by xid */
264 mutex_unlock(&ted->ted_lcd_lock);
265 CDEBUG(D_HA, "%s: transno = %llu < last_transno = %llu\n",
266 mdt_obd_name(mdt), info->mti_transno,
267 lcd->lcd_last_transno);
271 if (lustre_msg_get_opc(req->rq_reqmsg) == MDS_CLOSE) {
272 if (info->mti_transno != 0)
273 lcd->lcd_last_close_transno = info->mti_transno;
274 lcd->lcd_last_close_xid = req->rq_xid;
275 lcd->lcd_last_close_result = rc;
277 /* VBR: save versions in last_rcvd for reconstruct. */
278 __u64 *pre_versions = lustre_msg_get_versions(req->rq_repmsg);
281 lcd->lcd_pre_versions[0] = pre_versions[0];
282 lcd->lcd_pre_versions[1] = pre_versions[1];
283 lcd->lcd_pre_versions[2] = pre_versions[2];
284 lcd->lcd_pre_versions[3] = pre_versions[3];
286 if (info->mti_transno != 0)
287 lcd->lcd_last_transno = info->mti_transno;
289 lcd->lcd_last_xid = req->rq_xid;
290 lcd->lcd_last_result = rc;
291 lcd->lcd_last_data = info->mti_opdata;
293 mutex_unlock(&ted->ted_lcd_lock);
298 void mdt_mfd_set_mode(struct mdt_file_data *mfd, u64 open_flags)
300 LASSERT(mfd != NULL);
302 CDEBUG(D_DENTRY, DFID " Change mfd open_flags %#llo -> %#llo.\n",
303 PFID(mdt_object_fid(mfd->mfd_object)), mfd->mfd_open_flags,
306 mfd->mfd_open_flags = open_flags;
310 * prep ma_lmm/ma_lmv for md_attr from reply
312 static void mdt_prep_ma_buf_from_rep(struct mdt_thread_info *info,
313 struct mdt_object *obj,
316 if (ma->ma_lmv || ma->ma_lmm) {
317 CDEBUG(D_INFO, DFID " %s already set.\n",
318 PFID(mdt_object_fid(obj)),
319 ma->ma_lmv ? (ma->ma_lmm ? "ma_lmv and ma_lmm"
325 if (S_ISDIR(obj->mot_header.loh_attr)) {
326 ma->ma_lmv = req_capsule_server_get(info->mti_pill,
328 ma->ma_lmv_size = req_capsule_get_size(info->mti_pill,
331 if (ma->ma_lmv_size > 0)
332 ma->ma_need |= MA_LMV;
334 ma->ma_lmm = req_capsule_server_get(info->mti_pill,
336 ma->ma_lmm_size = req_capsule_get_size(info->mti_pill,
339 if (ma->ma_lmm_size > 0)
340 ma->ma_need |= MA_LOV;
344 static int mdt_mfd_open(struct mdt_thread_info *info, struct mdt_object *p,
345 struct mdt_object *o, u64 open_flags, int created,
346 struct ldlm_reply *rep)
348 struct ptlrpc_request *req = mdt_info_req(info);
349 struct mdt_export_data *med = &req->rq_export->exp_mdt_data;
350 struct mdt_file_data *mfd;
351 struct md_attr *ma = &info->mti_attr;
352 struct lu_attr *la = &ma->ma_attr;
353 struct mdt_body *repbody;
358 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
360 isreg = S_ISREG(la->la_mode);
361 isdir = S_ISDIR(la->la_mode);
362 if (isreg && !(ma->ma_valid & MA_LOV) &&
363 !(open_flags & MDS_OPEN_RELEASE)) {
365 * No EA, check whether it is will set regEA and dirEA since in
366 * above attr get, these size might be zero, so reset it, to
367 * retrieve the MD after create obj.
369 ma->ma_lmm_size = req_capsule_get_size(info->mti_pill,
372 /* in replay case, p == NULL */
373 rc = mdt_create_data(info, p, o);
377 if (exp_connect_flags(req->rq_export) & OBD_CONNECT_DISP_STRIPE)
378 mdt_set_disposition(info, rep, DISP_OPEN_STRIPE);
381 CDEBUG(D_INODE, "after open, ma_valid bit = %#llx lmm_size = %d\n",
382 ma->ma_valid, ma->ma_lmm_size);
384 if (ma->ma_valid & MA_LOV) {
385 LASSERT(ma->ma_lmm_size != 0);
386 repbody->mbo_eadatasize = ma->ma_lmm_size;
388 repbody->mbo_valid |= OBD_MD_FLDIREA;
390 repbody->mbo_valid |= OBD_MD_FLEASIZE;
393 if (ma->ma_valid & MA_LMV) {
394 LASSERT(ma->ma_lmv_size != 0);
395 repbody->mbo_eadatasize = ma->ma_lmv_size;
397 repbody->mbo_valid |= OBD_MD_FLDIREA | OBD_MD_MEA;
400 if (open_flags & MDS_FMODE_WRITE)
401 rc = mdt_write_get(o);
402 else if (open_flags & MDS_FMODE_EXEC)
403 rc = mdt_write_deny(o);
408 rc = mo_open(info->mti_env, mdt_object_child(o),
409 created ? open_flags | MDS_OPEN_CREATED : open_flags,
412 /* If we allow the client to chgrp (CFS_SETGRP_PERM), but the
413 * client does not know which suppgid should be sent to the MDS,
414 * or some other(s) changed the target file's GID after this RPC
415 * sent to the MDS with the suppgid as the original GID, then we
416 * give the client another chance to send the right suppgid. */
418 allow_client_chgrp(info, lu_ucred(info->mti_env)))
419 mdt_set_disposition(info, rep, DISP_OPEN_DENY);
424 mfd = mdt_mfd_new(med);
426 GOTO(err_out, rc = -ENOMEM);
429 * Keep a reference on this object for this open, and is
430 * released by mdt_mfd_close().
432 mdt_object_get(info->mti_env, o);
434 mfd->mfd_xid = req->rq_xid;
437 * @open_flags is always not zero. At least it should be FMODE_READ,
438 * FMODE_WRITE or MDS_FMODE_EXEC.
440 LASSERT(open_flags != 0);
443 mdt_mfd_set_mode(mfd, open_flags);
445 atomic_inc(&o->mot_open_count);
446 if (open_flags & MDS_OPEN_LEASE)
447 atomic_inc(&o->mot_lease_count);
450 if (req_is_replay(req)) {
451 struct mdt_file_data *old_mfd;
452 /* Check wheather old cookie already exist in
453 * the list, becasue when do recovery, client
454 * might be disconnected from server, and
455 * restart replay, so there maybe some orphan
456 * mfd here, we should remove them */
457 LASSERT(info->mti_rr.rr_open_handle != NULL);
458 spin_lock(&med->med_open_lock);
459 old_mfd = mdt_open_handle2mfd(med, info->mti_rr.rr_open_handle,
461 if (old_mfd != NULL) {
462 CDEBUG(D_HA, "delete orphan mfd = %p, fid = "DFID", "
463 "cookie = %#llx\n", mfd,
464 PFID(mdt_object_fid(mfd->mfd_object)),
465 info->mti_rr.rr_open_handle->cookie);
466 class_handle_unhash(&old_mfd->mfd_open_handle);
467 list_del_init(&old_mfd->mfd_list);
468 spin_unlock(&med->med_open_lock);
469 /* no attr update for that close */
471 ma->ma_valid |= MA_FLAGS;
472 ma->ma_attr_flags |= MDS_RECOV_OPEN;
473 mdt_mfd_close(info, old_mfd);
474 ma->ma_attr_flags &= ~MDS_RECOV_OPEN;
475 ma->ma_valid &= ~MA_FLAGS;
477 spin_unlock(&med->med_open_lock);
478 CDEBUG(D_HA, "orphan mfd not found, fid = "DFID", "
480 PFID(mdt_object_fid(mfd->mfd_object)),
481 info->mti_rr.rr_open_handle->cookie);
484 CDEBUG(D_HA, "Store old cookie %#llx in new mfd\n",
485 info->mti_rr.rr_open_handle->cookie);
487 mfd->mfd_open_handle_old = *info->mti_rr.rr_open_handle;
490 repbody->mbo_open_handle.cookie = mfd->mfd_open_handle.h_cookie;
492 if (req->rq_export->exp_disconnected) {
493 spin_lock(&med->med_open_lock);
494 class_handle_unhash(&mfd->mfd_open_handle);
495 list_del_init(&mfd->mfd_list);
496 spin_unlock(&med->med_open_lock);
497 mdt_mfd_close(info, mfd);
499 spin_lock(&med->med_open_lock);
500 list_add_tail(&mfd->mfd_list, &med->med_open_head);
501 spin_unlock(&med->med_open_lock);
504 mdt_empty_transno(info, rc);
509 if (open_flags & MDS_FMODE_WRITE)
511 else if (open_flags & MDS_FMODE_EXEC)
517 static int mdt_finish_open(struct mdt_thread_info *info,
518 struct mdt_object *p, struct mdt_object *o,
520 struct ldlm_reply *rep)
522 struct ptlrpc_request *req = mdt_info_req(info);
523 struct obd_export *exp = req->rq_export;
524 struct mdt_export_data *med = &req->rq_export->exp_mdt_data;
525 struct md_attr *ma = &info->mti_attr;
526 struct lu_attr *la = &ma->ma_attr;
527 struct mdt_file_data *mfd;
528 struct mdt_body *repbody;
531 int isreg, isdir, islnk;
535 LASSERT(ma->ma_valid & MA_INODE);
536 created = mdt_get_disposition(rep, DISP_OPEN_CREATE);
538 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
540 isreg = S_ISREG(la->la_mode);
541 isdir = S_ISDIR(la->la_mode);
542 islnk = S_ISLNK(la->la_mode);
543 mdt_pack_attr2body(info, repbody, la, mdt_object_fid(o));
545 /* compatibility check for 2.10 clients when it tries to open mirrored
546 * files. 2.10 clients don't verify overlapping components so they
547 * would read and write mirrored files just as if they were normal
548 * PFL files, which will cause the problem that sycned mirrors actually
549 * contain different data.
550 * Older clients are not a concern here because they don't even
551 * understand PFL layout. */
552 if (isreg && !exp_connect_flr(exp) && ma->ma_valid & MA_LOV &&
553 mdt_lmm_is_flr(ma->ma_lmm)) {
554 /* LU-10286: for simplicity clients who don't understand
555 * mirrored layout(with connect flag OBD_CONNECT2_FLR) won't
556 * be able to open mirrored files */
560 /* Overstriped files can crash older clients */
561 if (isreg && !exp_connect_overstriping(exp) &&
562 mdt_lmm_is_overstriping(ma->ma_lmm))
565 /* LU-2275, simulate broken behaviour (esp. prevalent in
566 * pre-2.4 servers where a very strange reply is sent on error
567 * that looks like it was actually almost successful and a
568 * failure at the same time.) */
569 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_NEGATIVE_POSITIVE)) {
570 mdt_set_disposition(info, rep, DISP_OPEN_OPEN |
574 if (open_flags & MDS_OPEN_LOCK)
575 mdt_set_disposition(info, rep, DISP_OPEN_LOCK);
580 #ifdef CONFIG_LUSTRE_FS_POSIX_ACL
581 if (exp_connect_flags(exp) & OBD_CONNECT_ACL) {
582 struct lu_nodemap *nodemap = nodemap_get_from_exp(exp);
584 RETURN(PTR_ERR(nodemap));
586 rc = mdt_pack_acl2body(info, repbody, o, nodemap);
587 nodemap_putref(nodemap);
594 * If we are following a symlink, don't open; and do not return open
595 * handle for special nodes as client required.
597 if (islnk || (!isreg && !isdir &&
598 (exp_connect_flags(req->rq_export) & OBD_CONNECT_NODEVOH))) {
599 lustre_msg_set_transno(req->rq_repmsg, 0);
604 * We need to return the existing object's fid back, so it is done here,
605 * after preparing the reply.
607 if (!created && (open_flags & MDS_OPEN_EXCL) &&
608 (open_flags & MDS_OPEN_CREAT))
611 /* This can't be done earlier, we need to return reply body */
613 if (open_flags & (MDS_OPEN_CREAT | MDS_FMODE_WRITE)) {
614 /* We are trying to create or write an existing dir. */
617 } else if (open_flags & MDS_OPEN_DIRECTORY)
620 if (OBD_FAIL_CHECK_RESET(OBD_FAIL_MDS_OPEN_CREATE,
621 OBD_FAIL_MDS_LDLM_REPLY_NET | OBD_FAIL_ONCE))
625 if (info->mti_rr.rr_flags & MRF_OPEN_RESEND) {
626 spin_lock(&med->med_open_lock);
627 list_for_each(t, &med->med_open_head) {
628 mfd = list_entry(t, struct mdt_file_data, mfd_list);
629 if (mfd->mfd_xid == req->rq_xid) {
630 repbody->mbo_open_handle.cookie =
631 mfd->mfd_open_handle.h_cookie;
636 spin_unlock(&med->med_open_lock);
639 /* set repbody->ea_size for resent case */
640 if (ma->ma_valid & MA_LOV) {
641 LASSERT(ma->ma_lmm_size != 0);
642 repbody->mbo_eadatasize = ma->ma_lmm_size;
644 repbody->mbo_valid |= OBD_MD_FLDIREA;
646 repbody->mbo_valid |= OBD_MD_FLEASIZE;
648 mdt_set_disposition(info, rep, DISP_OPEN_OPEN);
651 /* if we have a real resend (not a resend afrer failover), it
652 * means close is already happend, so lets return error
657 rc = mdt_mfd_open(info, p, o, open_flags, created, rep);
659 mdt_set_disposition(info, rep, DISP_OPEN_OPEN);
664 void mdt_reconstruct_open(struct mdt_thread_info *info,
665 struct mdt_lock_handle *lhc)
667 struct req_capsule *pill = info->mti_pill;
668 struct ptlrpc_request *req = mdt_info_req(info);
669 struct mdt_reint_record *rr = &info->mti_rr;
670 struct md_attr *ma = &info->mti_attr;
671 struct ldlm_reply *ldlm_rep;
676 LASSERT(pill->rc_fmt == &RQF_LDLM_INTENT_OPEN);
677 ldlm_rep = req_capsule_server_get(pill, &RMF_DLM_REP);
679 ma->ma_need = MA_INODE | MA_HSM;
681 opdata = mdt_req_from_lrd(req, info->mti_reply_data);
682 mdt_set_disposition(info, ldlm_rep, opdata);
684 CDEBUG(D_INODE, "This is reconstruct open: disp=%#llx, result=%d\n",
685 ldlm_rep->lock_policy_res1, req->rq_status);
687 /* We did not create successfully, return error to client. */
688 GOTO(out, rc = req->rq_status);
690 /* tg_reply_data is just memory only structure, so any non zero fid
691 * means a real resend not a resend after recovery which need to be
692 * handled as regular open
694 if (likely(!fid_is_zero(&info->mti_reply_data->trd_object))) {
695 rr->rr_fid2 = &info->mti_reply_data->trd_object;
696 rr->rr_flags |= MRF_OPEN_RESEND;
697 rc = mdt_open_by_fid(info, ldlm_rep, lhc);
699 lustre_msg_set_transno(req->rq_repmsg, 0);
701 /* We did not try to create, so we are a pure open */
702 rc = mdt_reint_open(info, lhc);
707 lustre_msg_set_status(req->rq_repmsg, req->rq_status);
708 LASSERT(ergo(rc < 0, lustre_msg_get_transno(req->rq_repmsg) == 0));
711 static int mdt_open_by_fid(struct mdt_thread_info *info, struct ldlm_reply *rep,
712 struct mdt_lock_handle *lhc)
714 u64 open_flags = info->mti_spec.sp_cr_flags;
715 struct mdt_reint_record *rr = &info->mti_rr;
716 struct md_attr *ma = &info->mti_attr;
717 struct mdt_object *o;
721 o = mdt_object_find(info->mti_env, info->mti_mdt, rr->rr_fid2);
723 RETURN(rc = PTR_ERR(o));
725 rc = mdt_check_enc(info, o);
729 if (unlikely(mdt_object_remote(o))) {
730 /* the child object was created on remote server */
731 struct mdt_body *repbody;
733 mdt_set_disposition(info, rep, (DISP_IT_EXECD |
736 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
737 repbody->mbo_fid1 = *rr->rr_fid2;
738 repbody->mbo_valid |= (OBD_MD_FLID | OBD_MD_MDS);
741 if (mdt_object_exists(o)) {
742 tgt_open_obj_set(info->mti_env, mdt_obj2dt(o));
743 mdt_set_disposition(info, rep, (DISP_IT_EXECD |
746 if ((open_flags & MDS_OPEN_EXCL) &&
747 (open_flags & MDS_OPEN_CREAT))
748 mdt_set_disposition(info, rep,
751 mdt_prep_ma_buf_from_rep(info, o, ma);
752 rc = mdt_attr_get_complex(info, o, ma);
755 rc = mdt_finish_open(info, NULL, o, open_flags, rep);
758 mdt_pack_size2body(info, rr->rr_fid2, &lhc->mlh_reg_lh);
765 mdt_object_put(info->mti_env, o);
769 /* lock object for open */
770 static int mdt_object_open_lock(struct mdt_thread_info *info,
771 struct mdt_object *obj,
772 struct mdt_lock_handle *lhc,
775 struct md_attr *ma = &info->mti_attr;
776 __u64 open_flags = info->mti_spec.sp_cr_flags;
778 enum ldlm_mode lm = LCK_CR;
779 bool acq_lease = !!(open_flags & MDS_OPEN_LEASE);
780 bool try_layout = false;
781 bool create_layout = false;
783 __u32 dom_stripe = 0;
784 unsigned int dom_only = 0;
785 unsigned int dom_lock = 0;
789 mdt_lock_handle_init(lhc);
791 if (req_is_replay(mdt_info_req(info)))
794 if (S_ISREG(lu_object_attr(&obj->mot_obj))) {
795 if (ma->ma_need & MA_LOV && !(ma->ma_valid & MA_LOV) &&
796 md_should_create(open_flags))
797 create_layout = true;
798 if (exp_connect_layout(info->mti_exp) && !create_layout &&
799 ma->ma_need & MA_LOV)
802 /* DoM files can take IO lock at OPEN when it makes sense,
803 * check if file has DoM stripe and ask for lock if client
804 * no lock on that resource yet.
806 if (ma->ma_valid & MA_LOV && ma->ma_lmm != NULL)
807 dom_stripe = mdt_lmm_dom_entry_check(ma->ma_lmm,
809 /* If only DOM stripe is being used then we can expect IO
810 * to it after OPEN and will return corresponding DOM ibit
811 * using default strategy from mdt_opts.mo_dom_lock.
812 * Otherwise trylock mode is used always and DOM ibit will
813 * be returned optionally.
816 !mdt_dom_client_has_lock(info, mdt_object_fid(obj)))
817 dom_lock = !dom_only ? TRYLOCK_DOM_ON_OPEN :
818 info->mti_mdt->mdt_opts.mo_dom_lock;
822 /* lease open, acquire write mode of open sem */
823 down_write(&obj->mot_open_sem);
825 /* Lease exists and ask for new lease */
826 if (atomic_read(&obj->mot_lease_count) > 0) {
827 /* only exclusive open is supported, so lease
828 * are conflicted to each other */
829 GOTO(out, rc = -EBUSY);
832 /* Lease must be with open lock */
833 if (!(open_flags & MDS_OPEN_LOCK)) {
834 CERROR("%s: Request lease for file:"DFID ", but open lock "
835 "is missed, open_flags = %#llo : rc = %d\n",
836 mdt_obd_name(info->mti_mdt),
837 PFID(mdt_object_fid(obj)), open_flags, -EPROTO);
838 GOTO(out, rc = -EPROTO);
841 /* should conflict with new opens for write/execute */
843 *ibits = MDS_INODELOCK_OPEN;
845 /* never grant LCK_EX layout lock to client */
847 } else { /* normal open */
848 /* normal open holds read mode of open sem */
849 down_read(&obj->mot_open_sem);
851 if (open_flags & MDS_OPEN_LOCK) {
852 if (open_flags & MDS_FMODE_WRITE)
854 else if (open_flags & MDS_FMODE_EXEC)
859 *ibits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_OPEN;
860 } else if (atomic_read(&obj->mot_lease_count) > 0) {
861 if (open_flags & MDS_FMODE_WRITE)
867 *ibits = MDS_INODELOCK_OPEN;
870 lhc = &info->mti_lh[MDT_LH_LOCAL];
871 } else if (dom_lock) {
872 lm = (open_flags & MDS_FMODE_WRITE) ? LCK_PW : LCK_PR;
873 trybits |= MDS_INODELOCK_DOM | MDS_INODELOCK_LAYOUT;
876 CDEBUG(D_INODE, "normal open:"DFID" lease count: %d, lm: %d\n",
877 PFID(mdt_object_fid(obj)),
878 atomic_read(&obj->mot_lease_count), lm);
881 mdt_lock_reg_init(lhc, lm);
883 /* Return lookup lock to validate inode at the client side.
884 * This is pretty important otherwise MDT will return layout
885 * lock for each open.
886 * However this is a double-edged sword because changing
887 * permission will revoke a huge number of LOOKUP locks.
889 if (!OBD_FAIL_CHECK(OBD_FAIL_MDS_NO_LL_OPEN) && try_layout) {
890 if (!(*ibits & MDS_INODELOCK_LOOKUP))
891 trybits |= MDS_INODELOCK_LOOKUP;
892 trybits |= MDS_INODELOCK_LAYOUT;
895 if (*ibits | trybits)
896 rc = mdt_object_lock_try(info, obj, lhc, ibits, trybits, false);
898 CDEBUG(D_INODE, "%s: Requested bits lock:"DFID ", ibits = %#llx/%#llx"
899 ", open_flags = %#llo, try_layout = %d : rc = %d\n",
900 mdt_obd_name(info->mti_mdt), PFID(mdt_object_fid(obj)),
901 *ibits, trybits, open_flags, try_layout, rc);
903 /* will change layout, revoke layout locks by enqueuing EX lock. */
904 if (rc == 0 && create_layout) {
905 struct mdt_lock_handle *ll = &info->mti_lh[MDT_LH_LAYOUT];
907 CDEBUG(D_INODE, "Will create layout, get EX layout lock:"DFID
908 ", open_flags = %#llo\n",
909 PFID(mdt_object_fid(obj)), open_flags);
911 /* We cannot enqueue another lock for the same resource we
912 * already have a lock for, due to mechanics of waiting list
913 * iterating in ldlm, see LU-3601.
914 * As such we'll drop the open lock we just got above here,
915 * it's ok not to have this open lock as it's main purpose is to
916 * flush unused cached client open handles. */
917 if (lustre_handle_is_used(&lhc->mlh_reg_lh))
918 mdt_object_unlock(info, obj, lhc, 1);
920 LASSERT(!try_layout);
921 mdt_lock_handle_init(ll);
922 mdt_lock_reg_init(ll, LCK_EX);
923 rc = mdt_object_lock(info, obj, ll, MDS_INODELOCK_LAYOUT);
925 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_LL_BLOCK, 2);
928 /* Check if there is any other open handles after acquiring
929 * open lock. At this point, caching open handles have been revoked
931 * XXX: Now only exclusive open is supported. Need to check the
932 * type of open for generic lease support. */
933 if (rc == 0 && acq_lease) {
934 struct ptlrpc_request *req = mdt_info_req(info);
935 struct mdt_export_data *med = &req->rq_export->exp_mdt_data;
936 struct mdt_file_data *mfd;
937 bool is_replay_or_resent;
940 /* For lease: application can open a file and then apply lease,
941 * @handle contains original open handle in that case.
942 * In recovery, open REQ will be replayed and the lease REQ may
943 * be resent that means the open handle is already stale, so we
944 * need to fix it up here by finding new handle. */
945 is_replay_or_resent = req_is_replay(req) ||
946 lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT;
948 /* if the request is _not_ a replay request, rr_open_handle
949 * may be used to hold an open file handle which is issuing the
950 * lease request, so that this openhandle doesn't count. */
951 mfd = mdt_open_handle2mfd(med, info->mti_rr.rr_open_handle,
952 is_replay_or_resent);
956 CDEBUG(D_INODE, "acq_lease "DFID": openers: %d, want: %d\n",
957 PFID(mdt_object_fid(obj)),
958 atomic_read(&obj->mot_open_count), open_count);
960 if (atomic_read(&obj->mot_open_count) > open_count) {
961 /* fail if anyone *else* has opened file for write */
962 if (mdt_write_read(obj) > 1)
963 GOTO(out, rc = -EBUSY);
972 static void mdt_object_open_unlock(struct mdt_thread_info *info,
973 struct mdt_object *obj,
974 struct mdt_lock_handle *lhc,
977 __u64 open_flags = info->mti_spec.sp_cr_flags;
978 struct mdt_lock_handle *ll = &info->mti_lh[MDT_LH_LOCAL];
981 if (req_is_replay(mdt_info_req(info)))
984 /* Release local lock - the lock put in MDT_LH_LOCAL will never
985 * return to client side. */
986 if (lustre_handle_is_used(&ll->mlh_reg_lh))
987 mdt_object_unlock(info, obj, ll, 1);
989 ll = &info->mti_lh[MDT_LH_LAYOUT];
990 /* Release local layout lock, layout was created */
991 if (lustre_handle_is_used(&ll->mlh_reg_lh)) {
992 LASSERT(!(ibits & MDS_INODELOCK_LAYOUT));
993 mdt_object_unlock(info, obj, ll, 1);
996 if (open_flags & MDS_OPEN_LEASE)
997 up_write(&obj->mot_open_sem);
999 up_read(&obj->mot_open_sem);
1001 /* Cross-ref case, the lock should be returned to the client */
1002 if (ibits == 0 || rc == -MDT_EREMOTE_OPEN)
1005 if (!(open_flags & MDS_OPEN_LOCK) && !(ibits & MDS_INODELOCK_LAYOUT) &&
1006 !(ibits & MDS_INODELOCK_DOM)) {
1007 /* for the open request, the lock will only return to client
1008 * if open or layout lock is granted. */
1012 if (rc != 0 || !lustre_handle_is_used(&lhc->mlh_reg_lh)) {
1013 struct ldlm_reply *ldlm_rep;
1015 ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
1016 mdt_clear_disposition(info, ldlm_rep, DISP_OPEN_LOCK);
1017 if (lustre_handle_is_used(&lhc->mlh_reg_lh))
1018 mdt_object_unlock(info, obj, lhc, 1);
1024 * Check release is permitted for the current HSM flags.
1026 static bool mdt_hsm_release_allow(const struct md_attr *ma)
1028 if (!(ma->ma_valid & MA_HSM))
1031 if (ma->ma_hsm.mh_flags & (HS_DIRTY|HS_NORELEASE|HS_LOST))
1034 if (!(ma->ma_hsm.mh_flags & HS_ARCHIVED))
1040 static int mdt_open_by_fid_lock(struct mdt_thread_info *info,
1041 struct ldlm_reply *rep,
1042 struct mdt_lock_handle *lhc)
1044 const struct lu_env *env = info->mti_env;
1045 struct mdt_device *mdt = info->mti_mdt;
1046 u64 open_flags = info->mti_spec.sp_cr_flags;
1047 struct mdt_reint_record *rr = &info->mti_rr;
1048 struct md_attr *ma = &info->mti_attr;
1049 struct mdt_object *parent = NULL;
1050 struct mdt_object *o;
1051 bool object_locked = false;
1056 if (md_should_create(open_flags)) {
1057 if (!lu_fid_eq(rr->rr_fid1, rr->rr_fid2)) {
1058 parent = mdt_object_find(env, mdt, rr->rr_fid1);
1059 if (IS_ERR(parent)) {
1060 CDEBUG(D_INODE, "Fail to find parent "DFID
1061 " for anonymous created %ld, try to"
1062 " use server-side parent.\n",
1063 PFID(rr->rr_fid1), PTR_ERR(parent));
1068 ma->ma_need |= MA_PFID;
1071 o = mdt_object_find(env, mdt, rr->rr_fid2);
1073 GOTO(out_parent_put, rc = PTR_ERR(o));
1075 if (mdt_object_remote(o)) {
1076 CDEBUG(D_INFO, "%s: "DFID" is on remote MDT.\n",
1077 mdt_obd_name(info->mti_mdt),
1079 GOTO(out, rc = -EREMOTE);
1080 } else if (!mdt_object_exists(o)) {
1081 mdt_set_disposition(info, rep,
1085 GOTO(out, rc = -ENOENT);
1088 /* do not check enc for directory: always allow open */
1089 if (!S_ISDIR(lu_object_attr(&o->mot_obj))) {
1090 rc = mdt_check_enc(info, o);
1095 mdt_set_disposition(info, rep, (DISP_IT_EXECD | DISP_LOOKUP_EXECD));
1097 mdt_prep_ma_buf_from_rep(info, o, ma);
1098 if (open_flags & MDS_OPEN_RELEASE)
1099 ma->ma_need |= MA_HSM;
1100 rc = mdt_attr_get_complex(info, o, ma);
1104 /* We should not change file's existing LOV EA */
1105 if (S_ISREG(lu_object_attr(&o->mot_obj)) &&
1106 open_flags & MDS_OPEN_HAS_EA && ma->ma_valid & MA_LOV)
1107 GOTO(out, rc = -EEXIST);
1109 /* If a release request, check file open flags are fine and ask for an
1110 * exclusive open access. */
1111 if (open_flags & MDS_OPEN_RELEASE && !mdt_hsm_release_allow(ma))
1112 GOTO(out, rc = -EPERM);
1114 rc = mdt_check_resent_lock(info, o, lhc);
1117 } else if (rc > 0) {
1118 rc = mdt_object_open_lock(info, o, lhc, &ibits);
1119 object_locked = true;
1121 GOTO(out_unlock, rc);
1124 if (ma->ma_valid & MA_PFID) {
1125 parent = mdt_object_find(env, mdt, &ma->ma_pfid);
1126 if (IS_ERR(parent)) {
1127 CDEBUG(D_INODE, "Fail to find parent "DFID
1128 " for anonymous created %ld, try to"
1129 " use system default.\n",
1130 PFID(&ma->ma_pfid), PTR_ERR(parent));
1135 tgt_open_obj_set(info->mti_env, mdt_obj2dt(o));
1136 rc = mdt_finish_open(info, parent, o, open_flags, rep);
1138 mdt_set_disposition(info, rep, DISP_LOOKUP_POS);
1139 if (open_flags & MDS_OPEN_LOCK)
1140 mdt_set_disposition(info, rep, DISP_OPEN_LOCK);
1141 if (open_flags & MDS_OPEN_LEASE)
1142 mdt_set_disposition(info, rep, DISP_OPEN_LEASE);
1144 GOTO(out_unlock, rc);
1148 mdt_object_open_unlock(info, o, lhc, ibits, rc);
1150 mdt_object_put(env, o);
1152 mdt_pack_size2body(info, rr->rr_fid2, &lhc->mlh_reg_lh);
1155 mdt_object_put(env, parent);
1159 /* Cross-ref request. Currently it can only be a pure open (w/o create) */
1160 static int mdt_cross_open(struct mdt_thread_info *info,
1161 const struct lu_fid *parent_fid,
1162 const struct lu_fid *fid,
1163 struct ldlm_reply *rep, u64 open_flags)
1165 struct md_attr *ma = &info->mti_attr;
1166 struct mdt_object *o;
1170 o = mdt_object_find(info->mti_env, info->mti_mdt, fid);
1172 RETURN(rc = PTR_ERR(o));
1174 rc = mdt_check_enc(info, o);
1178 if (mdt_object_remote(o)) {
1179 /* Something is wrong here, the object is on another MDS! */
1180 CERROR("%s: "DFID" isn't on this server!: rc = %d\n",
1181 mdt_obd_name(info->mti_mdt), PFID(fid), -EFAULT);
1182 LU_OBJECT_DEBUG(D_WARNING, info->mti_env,
1184 "Object isn't on this server! FLD error?");
1187 if (mdt_object_exists(o)) {
1190 /* Do permission check for cross-open after converting
1191 * MDS_OPEN_* flags to MAY_* permission mask.
1193 mask = mds_accmode(open_flags);
1195 rc = mo_permission(info->mti_env, NULL,
1196 mdt_object_child(o), NULL, mask);
1200 mdt_prep_ma_buf_from_rep(info, o, ma);
1201 rc = mdt_attr_get_complex(info, o, ma);
1205 rc = mdt_pack_secctx_in_reply(info, o);
1209 rc = mdt_pack_encctx_in_reply(info, o);
1213 rc = mdt_finish_open(info, NULL, o, open_flags, rep);
1216 * Something is wrong here. lookup was positive but
1217 * there is no object!
1219 CERROR("%s: "DFID" doesn't exist!: rc = %d\n",
1220 mdt_obd_name(info->mti_mdt), PFID(fid), -EFAULT);
1225 mdt_object_put(info->mti_env, o);
1231 * find root object and take its xattr lock if it's on remote MDT, later create
1232 * may use fs default striping (which is stored in root xattr).
1234 static int mdt_lock_root_xattr(struct mdt_thread_info *info,
1235 struct mdt_device *mdt)
1237 struct mdt_object *md_root = mdt->mdt_md_root;
1238 struct lustre_handle lhroot;
1241 if (md_root == NULL) {
1242 lu_root_fid(&info->mti_tmp_fid1);
1243 md_root = mdt_object_find(info->mti_env, mdt,
1244 &info->mti_tmp_fid1);
1245 if (IS_ERR(md_root))
1246 return PTR_ERR(md_root);
1248 spin_lock(&mdt->mdt_lock);
1249 if (mdt->mdt_md_root != NULL) {
1250 spin_unlock(&mdt->mdt_lock);
1252 LASSERTF(mdt->mdt_md_root == md_root,
1253 "Different root object ("
1254 DFID") instances, %p, %p\n",
1255 PFID(&info->mti_tmp_fid1),
1256 mdt->mdt_md_root, md_root);
1257 LASSERT(atomic_read(
1258 &md_root->mot_obj.lo_header->loh_ref) > 1);
1260 mdt_object_put(info->mti_env, md_root);
1262 mdt->mdt_md_root = md_root;
1263 spin_unlock(&mdt->mdt_lock);
1267 if (md_root->mot_cache_attr || !mdt_object_remote(md_root))
1270 rc = mdt_remote_object_lock(info, md_root, mdt_object_fid(md_root),
1271 &lhroot, LCK_PR, MDS_INODELOCK_XATTR,
1276 md_root->mot_cache_attr = 1;
1278 /* don't cancel this lock, so that we know the cached xattr is valid. */
1279 ldlm_lock_decref(&lhroot, LCK_PR);
1284 static inline enum ldlm_mode mdt_open_lock_mode(struct mdt_thread_info *info,
1285 struct mdt_object *p,
1286 struct lu_name *name,
1292 /* We don't need to take the DLM lock for a volatile */
1293 if (open_flags & MDS_OPEN_VOLATILE)
1296 if (!(open_flags & MDS_OPEN_CREAT))
1299 result = mdo_lookup(info->mti_env, mdt_object_child(p), name, &fid,
1302 /* If the file exists we only need a read lock on the parent */
1303 return (result == 0) ? LCK_PR : LCK_PW;
1306 int mdt_reint_open(struct mdt_thread_info *info, struct mdt_lock_handle *lhc)
1308 struct mdt_device *mdt = info->mti_mdt;
1309 struct ptlrpc_request *req = mdt_info_req(info);
1310 struct mdt_object *parent;
1311 struct mdt_object *child;
1312 struct mdt_lock_handle *lh = NULL;
1313 struct ldlm_reply *ldlm_rep;
1314 struct mdt_body *repbody;
1315 struct lu_fid *child_fid = &info->mti_tmp_fid1;
1316 struct md_attr *ma = &info->mti_attr;
1317 u64 open_flags = info->mti_spec.sp_cr_flags;
1319 struct mdt_reint_record *rr = &info->mti_rr;
1322 int object_locked = 0;
1323 enum ldlm_mode lock_mode;
1325 ktime_t kstart = ktime_get();
1328 OBD_FAIL_TIMEOUT_ORSET(OBD_FAIL_MDS_PAUSE_OPEN, OBD_FAIL_ONCE,
1329 (obd_timeout + 1) / 4);
1331 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1333 ma->ma_need = MA_INODE;
1336 LASSERT(info->mti_pill->rc_fmt == &RQF_LDLM_INTENT_OPEN);
1337 ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
1339 if (unlikely(open_flags & MDS_OPEN_JOIN_FILE)) {
1340 CERROR("file join is not supported anymore.\n");
1341 GOTO(out, result = err_serious(-EOPNOTSUPP));
1343 msg_flags = lustre_msg_get_flags(req->rq_reqmsg);
1345 if ((open_flags & (MDS_OPEN_HAS_EA | MDS_OPEN_HAS_OBJS)) &&
1346 info->mti_spec.u.sp_ea.eadata == NULL)
1347 GOTO(out, result = err_serious(-EINVAL));
1349 if (open_flags & MDS_FMODE_WRITE &&
1350 exp_connect_flags(req->rq_export) & OBD_CONNECT_RDONLY)
1351 GOTO(out, result = -EROFS);
1353 CDEBUG(D_INODE, "I am going to open "DFID"/("DNAME"->"DFID") "
1354 "cr_flag=%#llo mode=0%06o msg_flag=0x%x\n",
1355 PFID(rr->rr_fid1), PNAME(&rr->rr_name), PFID(rr->rr_fid2),
1356 open_flags, ma->ma_attr.la_mode, msg_flags);
1358 if (info->mti_cross_ref) {
1359 /* This is cross-ref open */
1360 mdt_set_disposition(info, ldlm_rep,
1361 (DISP_IT_EXECD | DISP_LOOKUP_EXECD |
1363 result = mdt_cross_open(info, rr->rr_fid2, rr->rr_fid1,
1364 ldlm_rep, open_flags);
1366 } else if (req_is_replay(req)) {
1367 result = mdt_open_by_fid(info, ldlm_rep, lhc);
1369 if (result != -ENOENT)
1372 /* We didn't find the correct object, so we need to re-create it
1373 * via a regular replay. */
1374 if (!(open_flags & MDS_OPEN_CREAT)) {
1375 DEBUG_REQ(D_ERROR, req,
1376 "OPEN & CREAT not in open replay/by_fid");
1377 GOTO(out, result = -EFAULT);
1379 CDEBUG(D_INFO, "No object(1), continue as regular open.\n");
1380 } else if (open_flags & MDS_OPEN_BY_FID) {
1381 result = mdt_open_by_fid_lock(info, ldlm_rep, lhc);
1383 CDEBUG(D_INFO, "no object for "DFID": %d\n",
1384 PFID(rr->rr_fid2), result);
1388 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_OPEN_PACK))
1389 GOTO(out, result = err_serious(-ENOMEM));
1391 mdt_set_disposition(info, ldlm_rep,
1392 (DISP_IT_EXECD | DISP_LOOKUP_EXECD));
1394 if (!lu_name_is_valid(&rr->rr_name))
1395 GOTO(out, result = -EPROTO);
1397 result = mdt_lock_root_xattr(info, mdt);
1401 parent = mdt_object_find(info->mti_env, mdt, rr->rr_fid1);
1403 GOTO(out, result = PTR_ERR(parent));
1405 /* get and check version of parent */
1406 result = mdt_version_get_check(info, parent, 0);
1408 mdt_object_put(info->mti_env, parent);
1412 result = mdt_check_enc(info, parent);
1414 GOTO(out_parent, result);
1416 fid_zero(child_fid);
1418 lock_mode = mdt_open_lock_mode(info, parent, &rr->rr_name, open_flags);
1420 OBD_RACE(OBD_FAIL_MDS_REINT_OPEN);
1422 if (lock_mode != LCK_NL) {
1423 lh = &info->mti_lh[MDT_LH_PARENT];
1424 mdt_lock_pdo_init(lh, lock_mode, &rr->rr_name);
1425 result = mdt_object_lock(info, parent, lh, MDS_INODELOCK_UPDATE);
1427 GOTO(out_parent, result);
1429 result = mdo_lookup(info->mti_env, mdt_object_child(parent),
1430 &rr->rr_name, child_fid, &info->mti_spec);
1433 LASSERTF(ergo(result == 0, fid_is_sane(child_fid)),
1434 "looking for "DFID"/"DNAME", found FID = "DFID"\n",
1435 PFID(mdt_object_fid(parent)), PNAME(&rr->rr_name),
1438 if (result != 0 && result != -ENOENT)
1439 GOTO(out_parent_unlock, result);
1441 OBD_RACE(OBD_FAIL_MDS_REINT_OPEN2);
1443 if (result == -ENOENT) {
1444 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
1445 if (!(open_flags & MDS_OPEN_CREAT))
1446 GOTO(out_parent_unlock, result);
1447 if (mdt_rdonly(req->rq_export))
1448 GOTO(out_parent_unlock, result = -EROFS);
1450 LASSERT(equi(lh == NULL, lock_mode == LCK_NL));
1452 if (lock_mode == LCK_PR) {
1453 /* unlink vs create race: get write lock and restart */
1454 mdt_object_unlock(info, parent, lh, 1);
1455 mdt_clear_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
1456 mdt_lock_handle_init(lh);
1461 *child_fid = *info->mti_rr.rr_fid2;
1462 LASSERTF(fid_is_sane(child_fid), "fid="DFID"\n",
1464 /* In the function below, .hs_keycmp resolves to
1465 * lu_obj_hop_keycmp() */
1466 /* coverity[overrun-buffer-val] */
1467 child = mdt_object_new(info->mti_env, mdt, child_fid);
1470 * Check for O_EXCL is moved to the mdt_finish_open(),
1471 * we need to return FID back in that case.
1473 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1474 child = mdt_object_find(info->mti_env, mdt, child_fid);
1477 GOTO(out_parent_unlock, result = PTR_ERR(child));
1479 /** check version of child */
1480 rc = mdt_version_get_check(info, child, 1);
1482 GOTO(out_child, result = rc);
1484 tgt_open_obj_set(info->mti_env, mdt_obj2dt(child));
1486 if (result == -ENOENT) {
1487 /* Create under OBF and .lustre is not permitted */
1488 if (!fid_is_md_operative(rr->rr_fid1) &&
1489 (open_flags & MDS_OPEN_VOLATILE) == 0)
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 /* Don't do lookup sanity check. We know name doesn't exist. */
1503 info->mti_spec.sp_cr_lookup = 0;
1504 info->mti_spec.sp_feat = &dt_directory_features;
1506 result = mdo_create(info->mti_env, mdt_object_child(parent),
1507 &rr->rr_name, mdt_object_child(child),
1508 &info->mti_spec, &info->mti_attr);
1509 if (result == -ERESTART) {
1510 mdt_clear_disposition(info, ldlm_rep, DISP_OPEN_CREATE);
1511 GOTO(out_child, result);
1513 mdt_prep_ma_buf_from_rep(info, child, ma);
1514 /* XXX: we should call this once, see few lines below */
1516 result = mdt_attr_get_complex(info, child, ma);
1519 GOTO(out_child, result);
1522 mdt_counter_incr(req, LPROC_MDT_MKNOD,
1523 ktime_us_delta(ktime_get(), kstart));
1526 * The object is on remote node, return its FID for remote open.
1528 if (mdt_object_remote(child)) {
1530 * Check if this lock already was sent to client and
1531 * this is resent case. For resent case do not take lock
1532 * again, use what is already granted.
1534 LASSERT(lhc != NULL);
1536 rc = mdt_check_resent_lock(info, child, lhc);
1538 GOTO(out_child, result = rc);
1539 } else if (rc > 0) {
1540 mdt_lock_handle_init(lhc);
1541 mdt_lock_reg_init(lhc, LCK_PR);
1543 rc = mdt_object_lock(info, child, lhc,
1544 MDS_INODELOCK_LOOKUP);
1546 repbody->mbo_fid1 = *mdt_object_fid(child);
1547 repbody->mbo_valid |= (OBD_MD_FLID | OBD_MD_MDS);
1551 result = -MDT_EREMOTE_OPEN;
1552 GOTO(out_child, result);
1553 } else if (mdt_object_exists(child)) {
1554 /* Check early for MDS_OPEN_DIRECTORY/O_DIRECTORY to
1555 * avoid opening regular files from lfs getstripe
1556 * since doing so breaks the leases used by lfs
1557 * mirror. See LU-13693. */
1558 if (open_flags & MDS_OPEN_DIRECTORY &&
1559 S_ISREG(lu_object_attr(&child->mot_obj)))
1560 GOTO(out_child, result = -ENOTDIR);
1562 /* We have to get attr & LOV EA & HSM for this
1564 mdt_prep_ma_buf_from_rep(info, child, ma);
1565 ma->ma_need |= MA_HSM;
1566 result = mdt_attr_get_complex(info, child, ma);
1568 GOTO(out_child, result);
1570 /* Object does not exist. Likely FS corruption. */
1571 CERROR("%s: name '"DNAME"' present, but FID "
1572 DFID" is invalid\n", mdt_obd_name(info->mti_mdt),
1573 PNAME(&rr->rr_name), PFID(child_fid));
1574 GOTO(out_child, result = -EIO);
1578 repbody->mbo_max_mdsize = info->mti_mdt->mdt_max_mdsize;
1579 repbody->mbo_valid |= OBD_MD_FLMODEASIZE;
1581 rc = mdt_pack_secctx_in_reply(info, child);
1583 GOTO(out_child, result = rc);
1585 rc = mdt_pack_encctx_in_reply(info, child);
1587 GOTO(out_child, result = rc);
1589 rc = mdt_check_resent_lock(info, child, lhc);
1591 GOTO(out_child, result = rc);
1592 } else if (rc == 0) {
1593 /* the open lock might already be gotten in
1594 * ldlm_handle_enqueue() */
1595 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT);
1596 if (open_flags & MDS_OPEN_LOCK)
1597 mdt_set_disposition(info, ldlm_rep, DISP_OPEN_LOCK);
1599 /* get openlock if this isn't replay and client requested it */
1600 if (!req_is_replay(req)) {
1601 rc = mdt_object_open_lock(info, child, lhc, &ibits);
1604 GOTO(out_child_unlock, result = rc);
1605 else if (open_flags & MDS_OPEN_LOCK)
1606 mdt_set_disposition(info, ldlm_rep,
1610 /* Try to open it now. */
1611 rc = mdt_finish_open(info, parent, child, open_flags, ldlm_rep);
1614 /* openlock will be released if mdt_finish_open() failed */
1615 mdt_clear_disposition(info, ldlm_rep, DISP_OPEN_LOCK);
1617 if (created && (open_flags & MDS_OPEN_VOLATILE)) {
1618 CERROR("%s: cannot open volatile file "DFID", orphan "
1619 "file will be left in PENDING directory until "
1620 "next reboot, rc = %d\n", mdt_obd_name(mdt),
1621 PFID(mdt_object_fid(child)), rc);
1622 GOTO(out_child_unlock, result);
1628 rc = mdo_unlink(info->mti_env,
1629 mdt_object_child(parent),
1630 mdt_object_child(child),
1632 &info->mti_attr, 0);
1634 CERROR("%s: "DFID" cleanup of open: rc = %d\n",
1635 mdt_obd_name(info->mti_mdt),
1636 PFID(mdt_object_fid(child)), rc);
1637 mdt_clear_disposition(info, ldlm_rep, DISP_OPEN_CREATE);
1641 mdt_counter_incr(req, LPROC_MDT_OPEN,
1642 ktime_us_delta(ktime_get(), kstart));
1647 mdt_object_open_unlock(info, child, lhc, ibits, result);
1649 mdt_object_put(info->mti_env, child);
1651 mdt_pack_size2body(info, child_fid, &lhc->mlh_reg_lh);
1654 mdt_object_unlock(info, parent, lh, result || !created);
1657 mdt_object_put(info->mti_env, parent);
1660 lustre_msg_set_transno(req->rq_repmsg, 0);
1665 * Create an orphan object use local root.
1667 static struct mdt_object *mdt_orphan_open(struct mdt_thread_info *info,
1668 struct mdt_device *mdt,
1669 const struct lu_fid *fid,
1670 struct md_attr *attr, fmode_t fmode)
1672 const struct lu_env *env = info->mti_env;
1673 struct md_op_spec *spec = &info->mti_spec;
1674 struct lu_fid *local_root_fid = &info->mti_tmp_fid1;
1675 struct mdt_object *obj = NULL;
1676 struct mdt_object *local_root;
1677 static const struct lu_name lname = {
1678 .ln_name = "i_am_nobody",
1679 .ln_namelen = sizeof("i_am_nobody") - 1,
1681 struct lu_ucred *uc;
1682 kernel_cap_t uc_cap_save;
1686 rc = dt_root_get(env, mdt->mdt_bottom, local_root_fid);
1688 RETURN(ERR_PTR(rc));
1690 local_root = mdt_object_find(env, mdt, local_root_fid);
1691 if (IS_ERR(local_root))
1694 obj = mdt_object_new(env, mdt, fid);
1696 GOTO(out, rc = PTR_ERR(obj));
1698 spec->sp_cr_lookup = 0;
1699 spec->sp_feat = &dt_directory_features;
1700 spec->sp_cr_flags = MDS_OPEN_VOLATILE | fmode;
1701 if (attr->ma_valid & MA_LOV) {
1702 spec->u.sp_ea.eadata = attr->ma_lmm;
1703 spec->u.sp_ea.eadatalen = attr->ma_lmm_size;
1704 spec->sp_cr_flags |= MDS_OPEN_HAS_EA;
1706 spec->sp_cr_flags |= MDS_OPEN_DELAY_CREATE;
1710 uc_cap_save = uc->uc_cap;
1711 cap_raise(uc->uc_cap, CAP_DAC_OVERRIDE);
1712 rc = mdo_create(env, mdt_object_child(local_root), &lname,
1713 mdt_object_child(obj), spec, attr);
1714 uc->uc_cap = uc_cap_save;
1716 CERROR("%s: cannot create volatile file "DFID": rc = %d\n",
1717 mdt_obd_name(mdt), PFID(fid), rc);
1721 rc = mo_open(env, mdt_object_child(obj), MDS_OPEN_CREATED, spec);
1723 CERROR("%s: cannot open volatile file "DFID", orphan "
1724 "file will be left in PENDING directory until "
1725 "next reboot, rc = %d\n", mdt_obd_name(mdt),
1732 mdt_object_put(env, obj);
1735 mdt_object_put(env, local_root);
1739 /* XXX Look into layout in MDT layer. */
1740 static inline int mdt_hsm_set_released(struct lov_mds_md *lmm)
1742 struct lov_comp_md_v1 *comp_v1;
1743 struct lov_mds_md *v1;
1747 if (lmm->lmm_magic == cpu_to_le32(LOV_MAGIC_COMP_V1_DEFINED)) {
1748 comp_v1 = (struct lov_comp_md_v1 *)lmm;
1750 if (comp_v1->lcm_entry_count == 0)
1753 for (i = 0; i < le16_to_cpu(comp_v1->lcm_entry_count); i++) {
1754 off = le32_to_cpu(comp_v1->lcm_entries[i].lcme_offset);
1755 v1 = (struct lov_mds_md *)((char *)comp_v1 + off);
1756 v1->lmm_pattern |= cpu_to_le32(LOV_PATTERN_F_RELEASED);
1759 lmm->lmm_pattern |= cpu_to_le32(LOV_PATTERN_F_RELEASED);
1764 static inline int mdt_get_lmm_gen(struct lov_mds_md *lmm, __u32 *gen)
1766 struct lov_comp_md_v1 *comp_v1;
1768 if (le32_to_cpu(lmm->lmm_magic == LOV_MAGIC_COMP_V1)) {
1769 comp_v1 = (struct lov_comp_md_v1 *)lmm;
1770 *gen = le32_to_cpu(comp_v1->lcm_layout_gen);
1771 } else if (le32_to_cpu(lmm->lmm_magic) == LOV_MAGIC_V1 ||
1772 le32_to_cpu(lmm->lmm_magic) == LOV_MAGIC_V3) {
1773 *gen = le16_to_cpu(lmm->lmm_layout_gen);
1780 static int mdt_hsm_release(struct mdt_thread_info *info, struct mdt_object *o,
1783 struct mdt_lock_handle *lh = &info->mti_lh[MDT_LH_LAYOUT];
1784 struct lu_ucred *uc = mdt_ucred(info);
1785 struct close_data *data;
1786 struct ldlm_lock *lease;
1787 struct mdt_object *orphan;
1788 struct md_attr *orp_ma;
1796 if (mdt_rdonly(info->mti_exp))
1799 data = req_capsule_client_get(info->mti_pill, &RMF_CLOSE_DATA);
1803 lease = ldlm_handle2lock(&data->cd_handle);
1807 /* try to hold open_sem so that nobody else can open the file */
1808 if (!down_write_trylock(&o->mot_open_sem)) {
1809 ldlm_lock_cancel(lease);
1810 GOTO(out_reprocess, rc = -EBUSY);
1813 /* Check if the lease open lease has already canceled */
1814 lock_res_and_lock(lease);
1815 lease_broken = ldlm_is_cancel(lease);
1816 unlock_res_and_lock(lease);
1818 LDLM_DEBUG(lease, DFID " lease broken? %d",
1819 PFID(mdt_object_fid(o)), lease_broken);
1821 /* Cancel server side lease. Client side counterpart should
1822 * have been cancelled. It's okay to cancel it now as we've
1823 * held mot_open_sem. */
1824 ldlm_lock_cancel(lease);
1826 if (lease_broken) /* don't perform release task */
1827 GOTO(out_unlock, rc = -ESTALE);
1829 if (fid_is_zero(&data->cd_fid) || !fid_is_sane(&data->cd_fid))
1830 GOTO(out_unlock, rc = -EINVAL);
1832 /* ma_need was set before but it seems fine to change it in order to
1833 * avoid modifying the one from RPC */
1834 ma->ma_need = MA_HSM;
1835 rc = mdt_attr_get_complex(info, o, ma);
1837 GOTO(out_unlock, rc);
1839 if (ma->ma_attr_flags & MDS_PCC_ATTACH) {
1840 if (ma->ma_valid & MA_HSM) {
1841 if (ma->ma_hsm.mh_flags & HS_RELEASED)
1842 GOTO(out_unlock, rc = -EALREADY);
1844 if (ma->ma_hsm.mh_arch_id != data->cd_archive_id)
1846 DFID" archive id diff: %llu:%u\n",
1847 PFID(mdt_object_fid(o)),
1848 ma->ma_hsm.mh_arch_id,
1849 data->cd_archive_id);
1851 if (!(ma->ma_hsm.mh_flags & HS_DIRTY) &&
1852 ma->ma_hsm.mh_arch_ver == data->cd_data_version) {
1854 DFID" data version matches: packed=%llu "
1855 "and on-disk=%llu\n",
1856 PFID(mdt_object_fid(o)),
1857 data->cd_data_version,
1858 ma->ma_hsm.mh_arch_ver);
1859 ma->ma_hsm.mh_flags = HS_ARCHIVED | HS_EXISTS;
1862 if (ma->ma_hsm.mh_flags & HS_DIRTY)
1863 ma->ma_hsm.mh_flags = HS_ARCHIVED | HS_EXISTS;
1865 /* Set up HSM attribte for PCC archived object */
1866 BUILD_BUG_ON(sizeof(struct hsm_attrs) >
1867 sizeof(info->mti_xattr_buf));
1868 buf = &info->mti_buf;
1869 buf->lb_buf = info->mti_xattr_buf;
1870 buf->lb_len = sizeof(struct hsm_attrs);
1871 memset(&ma->ma_hsm, 0, sizeof(ma->ma_hsm));
1872 ma->ma_hsm.mh_flags = HS_ARCHIVED | HS_EXISTS;
1873 ma->ma_hsm.mh_arch_id = data->cd_archive_id;
1874 ma->ma_hsm.mh_arch_ver = data->cd_data_version;
1875 lustre_hsm2buf(buf->lb_buf, &ma->ma_hsm);
1877 rc = mo_xattr_set(info->mti_env, mdt_object_child(o),
1878 buf, XATTR_NAME_HSM, 0);
1880 GOTO(out_unlock, rc);
1883 if (!mdt_hsm_release_allow(ma))
1884 GOTO(out_unlock, rc = -EPERM);
1886 /* already released? */
1887 if (ma->ma_hsm.mh_flags & HS_RELEASED)
1888 GOTO(out_unlock, rc = 0);
1890 /* Compare on-disk and packed data_version */
1891 if (data->cd_data_version != ma->ma_hsm.mh_arch_ver) {
1892 CDEBUG(D_HSM, DFID" data_version mismatches: "
1893 "packed=%llu and on-disk=%llu\n",
1894 PFID(mdt_object_fid(o)),
1895 data->cd_data_version,
1896 ma->ma_hsm.mh_arch_ver);
1897 GOTO(out_unlock, rc = -EPERM);
1901 ma->ma_valid = MA_INODE;
1902 ma->ma_attr.la_valid &= LA_ATIME | LA_MTIME | LA_CTIME | LA_SIZE;
1903 rc = mo_attr_set(info->mti_env, mdt_object_child(o), ma);
1905 GOTO(out_unlock, rc);
1907 mutex_lock(&o->mot_som_mutex);
1908 rc2 = mdt_set_som(info, o, SOM_FL_STRICT, ma->ma_attr.la_size,
1909 ma->ma_attr.la_blocks);
1910 mutex_unlock(&o->mot_som_mutex);
1913 "%s: File "DFID" SOM update failed: rc = %d\n",
1914 mdt_obd_name(info->mti_mdt),
1915 PFID(mdt_object_fid(o)), rc2);
1918 ma->ma_need = MA_INODE | MA_LOV;
1919 rc = mdt_attr_get_complex(info, o, ma);
1921 GOTO(out_unlock, rc);
1923 if (!(ma->ma_valid & MA_LOV)) {
1924 /* Even empty file are released */
1925 memset(ma->ma_lmm, 0, sizeof(*ma->ma_lmm));
1926 ma->ma_lmm->lmm_magic = cpu_to_le32(LOV_MAGIC_V1_DEFINED);
1927 ma->ma_lmm->lmm_pattern = cpu_to_le32(LOV_PATTERN_RAID0);
1928 ma->ma_lmm->lmm_stripe_size = cpu_to_le32(LOV_MIN_STRIPE_SIZE);
1929 ma->ma_lmm_size = sizeof(*ma->ma_lmm);
1931 /* Magic must be LOV_MAGIC_*_DEFINED or LOD will interpret
1932 * ma_lmm as lov_user_md, then it will be confused by union of
1933 * layout_gen and stripe_offset. */
1934 if ((le32_to_cpu(ma->ma_lmm->lmm_magic) & LOV_MAGIC_MASK) ==
1936 ma->ma_lmm->lmm_magic |= cpu_to_le32(LOV_MAGIC_DEFINED);
1938 GOTO(out_unlock, rc = -EINVAL);
1941 /* Set file as released. */
1942 rc = mdt_hsm_set_released(ma->ma_lmm);
1944 GOTO(out_unlock, rc);
1946 orp_ma = &info->mti_u.hsm.attr;
1947 orp_ma->ma_attr.la_mode = S_IFREG | S_IWUSR;
1948 /* We use root ownership to bypass potential quota
1949 * restrictions on the user and group of the file. */
1950 orp_ma->ma_attr.la_uid = 0;
1951 orp_ma->ma_attr.la_gid = 0;
1952 orp_ma->ma_attr.la_valid = LA_MODE | LA_UID | LA_GID;
1953 orp_ma->ma_lmm = ma->ma_lmm;
1954 orp_ma->ma_lmm_size = ma->ma_lmm_size;
1955 orp_ma->ma_valid = MA_INODE | MA_LOV;
1956 orphan = mdt_orphan_open(info, info->mti_mdt, &data->cd_fid, orp_ma,
1958 if (IS_ERR(orphan)) {
1959 CERROR("%s: cannot open orphan file "DFID": rc = %ld\n",
1960 mdt_obd_name(info->mti_mdt), PFID(&data->cd_fid),
1962 GOTO(out_unlock, rc = PTR_ERR(orphan));
1965 /* Set up HSM attribute for orphan object */
1966 BUILD_BUG_ON(sizeof(struct hsm_attrs) > sizeof(info->mti_xattr_buf));
1967 buf = &info->mti_buf;
1968 buf->lb_buf = info->mti_xattr_buf;
1969 buf->lb_len = sizeof(struct hsm_attrs);
1970 ma->ma_hsm.mh_flags |= HS_RELEASED;
1971 lustre_hsm2buf(buf->lb_buf, &ma->ma_hsm);
1972 ma->ma_hsm.mh_flags &= ~HS_RELEASED;
1974 mdt_lock_reg_init(lh, LCK_EX);
1975 rc = mdt_object_lock(info, o, lh, MDS_INODELOCK_LAYOUT |
1976 MDS_INODELOCK_XATTR);
1978 GOTO(out_close, rc);
1980 /* The orphan has root ownership so we need to raise
1981 * CAP_FOWNER to set the HSM attributes. */
1983 cap_raise(uc->uc_cap, CAP_FOWNER);
1984 rc = mo_xattr_set(info->mti_env, mdt_object_child(orphan), buf,
1988 GOTO(out_layout_lock, rc);
1990 /* Swap layout with orphan objects. */
1991 rc = mo_swap_layouts(info->mti_env, mdt_object_child(o),
1992 mdt_object_child(orphan),
1993 SWAP_LAYOUTS_MDS_HSM);
1995 if (!rc && ma->ma_attr_flags & MDS_PCC_ATTACH) {
1996 ma->ma_need = MA_LOV;
1997 rc = mdt_attr_get_complex(info, o, ma);
2003 /* Release exclusive LL */
2004 mdt_object_unlock(info, o, lh, 1);
2006 /* Close orphan object anyway */
2007 rc2 = mo_close(info->mti_env, mdt_object_child(orphan), orp_ma,
2010 CERROR("%s: error closing volatile file "DFID": rc = %d\n",
2011 mdt_obd_name(info->mti_mdt), PFID(&data->cd_fid), rc2);
2012 LU_OBJECT_DEBUG(D_HSM, info->mti_env, &orphan->mot_obj,
2014 mdt_object_put(info->mti_env, orphan);
2017 up_write(&o->mot_open_sem);
2019 /* already released */
2021 struct mdt_body *repbody;
2023 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
2024 LASSERT(repbody != NULL);
2025 repbody->mbo_valid |= OBD_MD_CLOSE_INTENT_EXECED;
2026 if (ma->ma_attr_flags & MDS_PCC_ATTACH) {
2027 LASSERT(ma->ma_valid & MA_LOV);
2028 rc = mdt_get_lmm_gen(ma->ma_lmm,
2029 &repbody->mbo_layout_gen);
2031 repbody->mbo_valid |= OBD_MD_LAYOUT_VERSION;
2036 ldlm_reprocess_all(lease->l_resource,
2037 lease->l_policy_data.l_inodebits.bits);
2038 LDLM_LOCK_PUT(lease);
2046 int mdt_close_handle_layouts(struct mdt_thread_info *info,
2047 struct mdt_object *o, struct md_attr *ma)
2049 struct mdt_lock_handle *lh1 = &info->mti_lh[MDT_LH_NEW];
2050 struct mdt_lock_handle *lh2 = &info->mti_lh[MDT_LH_OLD];
2051 struct close_data *data;
2052 struct ldlm_lock *lease;
2053 struct mdt_object *o1 = o, *o2 = NULL;
2055 bool swap_objects = false;
2059 if (exp_connect_flags(info->mti_exp) & OBD_CONNECT_RDONLY)
2062 if (!S_ISREG(lu_object_attr(&o1->mot_obj)))
2065 data = req_capsule_client_get(info->mti_pill, &RMF_CLOSE_DATA);
2069 if (fid_is_zero(&data->cd_fid) || !fid_is_sane(&data->cd_fid))
2072 rc = lu_fid_cmp(&data->cd_fid, mdt_object_fid(o));
2075 * only MDS_CLOSE_LAYOUT_SPLIT use the same fid to indicate
2076 * mirror deletion, so we'd zero cd_fid, and keeps o2 be NULL.
2078 if (!(ma->ma_attr_flags & MDS_CLOSE_LAYOUT_SPLIT))
2081 /* zero cd_fid to keeps o2 be NULL */
2082 fid_zero(&data->cd_fid);
2083 } else if (rc < 0) {
2084 /* Exchange o1 and o2, to enforce locking order */
2085 swap_objects = true;
2088 lease = ldlm_handle2lock(&data->cd_handle);
2092 if (!fid_is_zero(&data->cd_fid)) {
2093 o2 = mdt_object_find(info->mti_env, info->mti_mdt,
2096 GOTO(out_lease, rc = PTR_ERR(o2));
2098 if (!mdt_object_exists(o2))
2099 GOTO(out_obj, rc = -ENOENT);
2101 if (!S_ISREG(lu_object_attr(&o2->mot_obj)))
2102 GOTO(out_obj, rc = -EINVAL);
2108 rc = mo_permission(info->mti_env, NULL, mdt_object_child(o1), NULL,
2114 rc = mo_permission(info->mti_env, NULL, mdt_object_child(o2),
2120 /* try to hold open_sem so that nobody else can open the file */
2121 if (!down_write_trylock(&o->mot_open_sem)) {
2122 ldlm_lock_cancel(lease);
2123 GOTO(out_obj, rc = -EBUSY);
2126 /* Check if the lease open lease has already canceled */
2127 lock_res_and_lock(lease);
2128 lease_broken = ldlm_is_cancel(lease);
2129 unlock_res_and_lock(lease);
2131 LDLM_DEBUG(lease, DFID " lease broken? %d",
2132 PFID(mdt_object_fid(o)), lease_broken);
2134 /* Cancel server side lease. Client side counterpart should
2135 * have been cancelled. It's okay to cancel it now as we've
2136 * held mot_open_sem. */
2137 ldlm_lock_cancel(lease);
2140 GOTO(out_unlock_sem, rc = -ESTALE);
2142 mdt_lock_reg_init(lh1, LCK_EX);
2143 rc = mdt_object_lock(info, o1, lh1, MDS_INODELOCK_LAYOUT |
2144 MDS_INODELOCK_XATTR);
2146 GOTO(out_unlock_sem, rc);
2149 mdt_lock_reg_init(lh2, LCK_EX);
2150 rc = mdt_object_lock(info, o2, lh2, MDS_INODELOCK_LAYOUT |
2151 MDS_INODELOCK_XATTR);
2153 GOTO(out_unlock1, rc);
2156 /* Swap layout with orphan object */
2157 if (ma->ma_attr_flags & MDS_CLOSE_LAYOUT_SWAP) {
2158 rc = mo_swap_layouts(info->mti_env, mdt_object_child(o1),
2159 mdt_object_child(o2), 0);
2160 } else if (ma->ma_attr_flags & MDS_CLOSE_LAYOUT_MERGE ||
2161 ma->ma_attr_flags & MDS_CLOSE_LAYOUT_SPLIT) {
2162 struct lu_buf *buf = &info->mti_buf;
2163 struct md_rejig_data mrd;
2166 mrd.mrd_obj = mdt_object_child(o == o1 ? o2 : o1);
2168 if (!(ma->ma_attr_flags & MDS_CLOSE_LAYOUT_SPLIT)) {
2169 /* paranoid check again */
2171 ":only mirror split support NULL o2 object\n",
2172 PFID(mdt_object_fid(o)));
2173 GOTO(out_unlock1, rc = -EINVAL);
2176 /* set NULL mrd_obj for deleting mirror objects */
2180 if (ma->ma_attr_flags & MDS_CLOSE_LAYOUT_SPLIT) {
2181 mrd.mrd_mirror_id = data->cd_mirror_id;
2182 /* set a small enough blocks in the SoM */
2183 ma->ma_attr.la_blocks >>= 1;
2186 buf->lb_len = sizeof(mrd);
2188 rc = mo_xattr_set(info->mti_env, mdt_object_child(o), buf,
2190 ma->ma_attr_flags & MDS_CLOSE_LAYOUT_SPLIT ?
2191 LU_XATTR_SPLIT : LU_XATTR_MERGE);
2192 if (rc == 0 && ma->ma_attr.la_valid & (LA_SIZE | LA_BLOCKS |
2193 LA_LSIZE | LA_LBLOCKS)) {
2195 enum lustre_som_flags lsf;
2197 if (ma->ma_attr.la_valid & (LA_SIZE | LA_BLOCKS))
2198 lsf = SOM_FL_STRICT;
2202 mutex_lock(&o->mot_som_mutex);
2203 rc2 = mdt_set_som(info, o, lsf,
2204 ma->ma_attr.la_size,
2205 ma->ma_attr.la_blocks);
2206 mutex_unlock(&o->mot_som_mutex);
2208 CERROR(DFID": Setting i_blocks error: %d, "
2209 "i_blocks will be reported wrongly and "
2210 "can only be fixed in next resync\n",
2211 PFID(mdt_object_fid(o)), rc2);
2215 GOTO(out_unlock2, rc);
2220 /* Release exclusive LL */
2222 mdt_object_unlock(info, o2, lh2, 1);
2225 mdt_object_unlock(info, o1, lh1, 1);
2228 up_write(&o->mot_open_sem);
2230 /* already swapped */
2232 struct mdt_body *repbody;
2234 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
2235 LASSERT(repbody != NULL);
2236 repbody->mbo_valid |= OBD_MD_CLOSE_INTENT_EXECED;
2241 /* the 2nd object has been used, and swapped to o1 */
2242 mdt_object_put(info->mti_env, o1);
2244 /* the 2nd object has been used, and not swapped */
2245 mdt_object_put(info->mti_env, o2);
2247 ldlm_reprocess_all(lease->l_resource,
2248 lease->l_policy_data.l_inodebits.bits);
2251 LDLM_LOCK_PUT(lease);
2261 static int mdt_close_resync_done(struct mdt_thread_info *info,
2262 struct mdt_object *o, struct md_attr *ma)
2264 struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_LOCAL];
2265 struct close_data *data;
2266 struct ldlm_lock *lease;
2267 struct md_layout_change layout = { 0 };
2268 __u32 *resync_ids = NULL;
2269 size_t resync_count = 0;
2274 if (exp_connect_flags(info->mti_exp) & OBD_CONNECT_RDONLY)
2277 if (!S_ISREG(lu_object_attr(&o->mot_obj)))
2280 data = req_capsule_client_get(info->mti_pill, &RMF_CLOSE_DATA);
2284 if (req_capsule_req_need_swab(info->mti_pill))
2285 lustre_swab_close_data_resync_done(&data->cd_resync);
2287 if (!fid_is_zero(&data->cd_fid))
2290 lease = ldlm_handle2lock(&data->cd_handle);
2294 /* try to hold open_sem so that nobody else can open the file */
2295 if (!down_write_trylock(&o->mot_open_sem)) {
2296 ldlm_lock_cancel(lease);
2297 GOTO(out_reprocess, rc = -EBUSY);
2300 /* Check if the lease open lease has already canceled */
2301 lock_res_and_lock(lease);
2302 lease_broken = ldlm_is_cancel(lease);
2303 unlock_res_and_lock(lease);
2305 LDLM_DEBUG(lease, DFID " lease broken? %d\n",
2306 PFID(mdt_object_fid(o)), lease_broken);
2308 /* Cancel server side lease. Client side counterpart should
2309 * have been cancelled. It's okay to cancel it now as we've
2310 * held mot_open_sem. */
2311 ldlm_lock_cancel(lease);
2313 if (lease_broken) /* don't perform release task */
2314 GOTO(out_unlock, rc = -ESTALE);
2316 resync_count = data->cd_resync.resync_count;
2318 if (resync_count > INLINE_RESYNC_ARRAY_SIZE) {
2321 if (!req_capsule_has_field(info->mti_pill, &RMF_U32,
2323 GOTO(out_unlock, rc = -EPROTO);
2325 OBD_ALLOC_PTR_ARRAY(resync_ids, resync_count);
2327 GOTO(out_unlock, rc = -ENOMEM);
2329 data = req_capsule_client_get(info->mti_pill, &RMF_U32);
2330 memcpy(resync_ids, data, resync_count * sizeof(__u32));
2332 layout.mlc_resync_ids = resync_ids;
2334 layout.mlc_resync_ids = data->cd_resync.resync_ids_inline;
2337 layout.mlc_opc = MD_LAYOUT_RESYNC_DONE;
2338 layout.mlc_resync_count = resync_count;
2339 if (ma->ma_attr.la_valid & (LA_SIZE | LA_BLOCKS)) {
2340 layout.mlc_som.lsa_valid = SOM_FL_STRICT;
2341 layout.mlc_som.lsa_size = ma->ma_attr.la_size;
2342 layout.mlc_som.lsa_blocks = ma->ma_attr.la_blocks;
2344 rc = mdt_layout_change(info, o, lhc, &layout);
2346 GOTO(out_unlock, rc);
2348 mdt_object_unlock(info, o, lhc, 0);
2353 up_write(&o->mot_open_sem);
2355 /* already released */
2357 struct mdt_body *repbody;
2359 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
2360 LASSERT(repbody != NULL);
2361 repbody->mbo_valid |= OBD_MD_CLOSE_INTENT_EXECED;
2365 OBD_FREE_PTR_ARRAY(resync_ids, resync_count);
2368 ldlm_reprocess_all(lease->l_resource,
2369 lease->l_policy_data.l_inodebits.bits);
2370 LDLM_LOCK_PUT(lease);
2378 #define MFD_CLOSED(open_flags) ((open_flags) == MDS_FMODE_CLOSED)
2380 static int mdt_mfd_closed(struct mdt_file_data *mfd)
2382 return ((mfd == NULL) || MFD_CLOSED(mfd->mfd_open_flags));
2385 int mdt_mfd_close(struct mdt_thread_info *info, struct mdt_file_data *mfd)
2387 struct mdt_object *o = mfd->mfd_object;
2388 struct md_object *next = mdt_object_child(o);
2389 struct md_attr *ma = &info->mti_attr;
2390 struct lu_fid *ofid = &info->mti_tmp_fid1;
2397 open_flags = mfd->mfd_open_flags;
2398 intent = ma->ma_attr_flags & MDS_CLOSE_INTENT;
2399 *ofid = *mdt_object_fid(o);
2401 /* the below message is checked in replay-single.sh test_46 */
2402 CDEBUG(D_INODE, "%s: %sclosing file handle "DFID" with intent: %llx\n",
2403 mdt_obd_name(info->mti_mdt),
2404 ma->ma_valid & MA_FORCE_LOG ? "force " : "", PFID(ofid), intent);
2407 case MDS_HSM_RELEASE: {
2408 rc = mdt_hsm_release(info, o, ma);
2410 CDEBUG(D_HSM, "%s: File " DFID " release failed: %d\n",
2411 mdt_obd_name(info->mti_mdt),
2413 /* continue to close even error occurred. */
2417 case MDS_CLOSE_LAYOUT_MERGE:
2418 case MDS_CLOSE_LAYOUT_SPLIT:
2419 case MDS_CLOSE_LAYOUT_SWAP: {
2420 rc = mdt_close_handle_layouts(info, o, ma);
2423 "%s: cannot %s layout of "DFID": rc = %d\n",
2424 mdt_obd_name(info->mti_mdt),
2425 intent == MDS_CLOSE_LAYOUT_MERGE ? "merge" :
2426 intent == MDS_CLOSE_LAYOUT_SPLIT ? "split" :
2429 /* continue to close even if error occurred. */
2433 case MDS_CLOSE_RESYNC_DONE:
2434 rc = mdt_close_resync_done(info, o, ma);
2437 "%s: cannot resync layout of "DFID": rc = %d\n",
2438 mdt_obd_name(info->mti_mdt),
2440 /* continue to close even if error occurred. */
2448 if (S_ISREG(lu_object_attr(&o->mot_obj)) &&
2449 ma->ma_attr.la_valid & (LA_LSIZE | LA_LBLOCKS)) {
2450 rc2 = mdt_lsom_update(info, o, false);
2453 "%s: File " DFID " LSOM failed: rc = %d\n",
2454 mdt_obd_name(info->mti_mdt),
2458 /* continue to close even if error occurred. */
2462 if (open_flags & MDS_FMODE_WRITE)
2464 else if (open_flags & MDS_FMODE_EXEC)
2467 /* Update atime|mtime|ctime on close. */
2468 if ((open_flags & MDS_FMODE_EXEC || open_flags & MDS_FMODE_READ ||
2469 open_flags & MDS_FMODE_WRITE) && (ma->ma_valid & MA_INODE) &&
2470 (ma->ma_attr.la_valid & LA_ATIME ||
2471 ma->ma_attr.la_valid & LA_MTIME ||
2472 ma->ma_attr.la_valid & LA_CTIME)) {
2473 ma->ma_valid = MA_INODE;
2474 ma->ma_attr_flags |= MDS_CLOSE_UPDATE_TIMES;
2475 ma->ma_attr.la_valid &= (LA_ATIME | LA_MTIME | LA_CTIME);
2477 if (ma->ma_attr.la_valid & LA_MTIME) {
2478 if (mdt_attr_get_pfid(info, o, &ma->ma_pfid) == 0)
2479 ma->ma_valid |= MA_PFID;
2482 rc2 = mo_attr_set(info->mti_env, next, ma);
2485 "%s: File "DFID" set attr (%#llx) failed: rc = %d\n",
2486 mdt_obd_name(info->mti_mdt), PFID(ofid),
2487 ma->ma_attr.la_valid, rc2);
2493 /* If file data is modified, add the dirty flag. */
2494 if (ma->ma_attr_flags & MDS_DATA_MODIFIED) {
2495 rc2 = mdt_add_dirty_flag(info, o, ma);
2498 "%s: File "DFID" add dirty flag failed: rc = %d\n",
2499 mdt_obd_name(info->mti_mdt), PFID(ofid), rc2);
2505 ma->ma_need |= MA_INODE;
2506 ma->ma_valid &= ~MA_INODE;
2508 LASSERT(atomic_read(&o->mot_open_count) > 0);
2509 atomic_dec(&o->mot_open_count);
2510 mdt_handle_last_unlink(info, o, ma);
2512 if (!MFD_CLOSED(open_flags)) {
2513 rc2 = mo_close(info->mti_env, next, ma, open_flags);
2516 "%s: File "DFID" close failed: rc = %d\n",
2517 mdt_obd_name(info->mti_mdt), PFID(ofid), rc2);
2521 if (mdt_dom_check_for_discard(info, o))
2522 mdt_dom_discard_data(info, o);
2525 /* adjust open and lease count */
2526 if (open_flags & MDS_OPEN_LEASE) {
2527 LASSERT(atomic_read(&o->mot_lease_count) > 0);
2528 atomic_dec(&o->mot_lease_count);
2532 mdt_object_put(info->mti_env, o);
2537 int mdt_close_internal(struct mdt_thread_info *info, struct ptlrpc_request *req,
2538 struct mdt_body *repbody)
2540 struct mdt_export_data *med;
2541 struct mdt_file_data *mfd;
2546 med = &req->rq_export->exp_mdt_data;
2547 spin_lock(&med->med_open_lock);
2548 mfd = mdt_open_handle2mfd(med, &info->mti_open_handle,
2549 req_is_replay(req));
2550 if (mdt_mfd_closed(mfd)) {
2551 spin_unlock(&med->med_open_lock);
2552 CDEBUG(D_INODE, "no handle for file close: fid = "DFID
2553 ": cookie = %#llx\n", PFID(info->mti_rr.rr_fid1),
2554 info->mti_open_handle.cookie);
2555 /** not serious error since bug 3633 */
2558 class_handle_unhash(&mfd->mfd_open_handle);
2559 list_del_init(&mfd->mfd_list);
2560 spin_unlock(&med->med_open_lock);
2561 ret = mdt_mfd_close(info, mfd);
2564 RETURN(rc ? rc : ret);
2567 int mdt_close(struct tgt_session_info *tsi)
2569 struct mdt_thread_info *info = tsi2mdt_info(tsi);
2570 struct ptlrpc_request *req = tgt_ses_req(tsi);
2571 struct md_attr *ma = &info->mti_attr;
2572 struct mdt_body *repbody = NULL;
2573 ktime_t kstart = ktime_get();
2578 /* Close may come with the Size-on-MDS update. Unpack it. */
2579 rc = mdt_close_unpack(info);
2581 GOTO(out, rc = err_serious(rc));
2583 /* These fields are no longer used and are left for compatibility.
2584 * size is always zero */
2585 req_capsule_set_size(info->mti_pill, &RMF_MDT_MD, RCL_SERVER,
2587 req_capsule_set_size(info->mti_pill, &RMF_LOGCOOKIES, RCL_SERVER,
2589 rc = req_capsule_server_pack(info->mti_pill);
2590 if (mdt_check_resent(info, mdt_reconstruct_generic, NULL)) {
2591 mdt_client_compatibility(info);
2593 mdt_fix_reply(info);
2594 mdt_exit_ucred(info);
2595 GOTO(out, rc = lustre_msg_get_status(req->rq_repmsg));
2598 /* Continue to close handle even if we can not pack reply */
2600 repbody = req_capsule_server_get(info->mti_pill,
2602 ma->ma_lmm = req_capsule_server_get(info->mti_pill,
2604 ma->ma_lmm_size = req_capsule_get_size(info->mti_pill,
2607 ma->ma_need = MA_INODE | MA_LOV;
2608 repbody->mbo_eadatasize = 0;
2609 repbody->mbo_aclsize = 0;
2611 rc = err_serious(rc);
2614 rc2 = mdt_close_internal(info, req, repbody);
2616 mdt_empty_transno(info, rc2);
2617 if (rc2 != 0 && rc == 0)
2620 if (repbody != NULL) {
2621 mdt_client_compatibility(info);
2622 rc2 = mdt_fix_reply(info);
2623 if (rc2 != 0 && rc == 0)
2627 mdt_exit_ucred(info);
2628 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_CLOSE_PACK))
2629 GOTO(out, rc = err_serious(-ENOMEM));
2631 if (OBD_FAIL_CHECK_RESET(OBD_FAIL_MDS_CLOSE_NET_REP,
2632 OBD_FAIL_MDS_CLOSE_NET_REP))
2633 tsi->tsi_reply_fail_id = OBD_FAIL_MDS_CLOSE_NET_REP;
2635 mdt_thread_info_fini(info);
2637 mdt_counter_incr(req, LPROC_MDT_CLOSE,
2638 ktime_us_delta(ktime_get(), kstart));