Whamcloud - gitweb
LU-15850 mdt: pack default LMV in open reply
[fs/lustre-release.git] / lustre / mdt / mdt_open.c
1 /*
2  * GPL HEADER START
3  *
4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
5  *
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.
9  *
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).
15  *
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
19  *
20  * GPL HEADER END
21  */
22 /*
23  * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
24  * Use is subject to license terms.
25  *
26  * Copyright (c) 2011, 2017, Intel Corporation.
27  */
28 /*
29  * This file is part of Lustre, http://www.lustre.org/
30  *
31  * lustre/mdt/mdt_open.c
32  *
33  * Lustre Metadata Target (mdt) open/close file handling
34  *
35  * Author: Huang Hua <huanghua@clusterfs.com>
36  */
37
38 #define DEBUG_SUBSYSTEM S_MDS
39
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>
45
46 static const char mfd_open_handle_owner[] = "mdt";
47
48 static int mdt_open_by_fid(struct mdt_thread_info *info, struct ldlm_reply *rep,
49                            struct mdt_lock_handle *lhc);
50
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)
54 {
55         struct mdt_file_data *mfd;
56
57         ENTRY;
58         OBD_ALLOC_PTR(mfd);
59         if (mfd != NULL) {
60                 refcount_set(&mfd->mfd_open_handle.h_ref, 1);
61                 INIT_HLIST_NODE(&mfd->mfd_open_handle.h_link);
62                 mfd->mfd_owner = med;
63                 INIT_LIST_HEAD(&mfd->mfd_list);
64                 class_handle_hash(&mfd->mfd_open_handle, mfd_open_handle_owner);
65         }
66
67         RETURN(mfd);
68 }
69
70 /*
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.
75  */
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)
79 {
80         struct mdt_file_data   *mfd;
81
82         ENTRY;
83         LASSERT(open_handle != NULL);
84         mfd = class_handle2object(open_handle->cookie, mfd_open_handle_owner);
85         if (mfd)
86                 refcount_dec(&mfd->mfd_open_handle.h_ref);
87
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 ==
92                             open_handle->cookie)
93                                 RETURN(mfd);
94                 }
95                 mfd = NULL;
96         }
97
98         RETURN(mfd);
99 }
100
101 /* free mfd */
102 void mdt_mfd_free(struct mdt_file_data *mfd)
103 {
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);
108 }
109
110 static int mdt_create_data(struct mdt_thread_info *info,
111                            struct mdt_object *p, struct mdt_object *o)
112 {
113         struct md_op_spec *spec = &info->mti_spec;
114         struct md_attr *ma = &info->mti_attr;
115         int rc = 0;
116
117         ENTRY;
118         if (!md_should_create(spec->sp_cr_flags))
119                 RETURN(0);
120
121         ma->ma_need = MA_INODE | MA_LOV;
122         ma->ma_valid = 0;
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);
128                 if (rc == 0)
129                         rc = mdt_attr_get_complex(info, o, ma);
130
131                 if (rc == 0 && ma->ma_valid & MA_LOV)
132                         o->mot_lov_created = 1;
133         }
134
135         mutex_unlock(&o->mot_lov_mutex);
136         RETURN(rc);
137 }
138
139 int mdt_write_read(struct mdt_object *o)
140 {
141         int rc = 0;
142
143         ENTRY;
144         spin_lock(&o->mot_write_lock);
145         rc = o->mot_write_count;
146         spin_unlock(&o->mot_write_lock);
147         RETURN(rc);
148 }
149
150 int mdt_write_get(struct mdt_object *o)
151 {
152         int rc = 0;
153
154         ENTRY;
155         spin_lock(&o->mot_write_lock);
156         if (o->mot_write_count < 0)
157                 rc = -ETXTBSY;
158         else
159                 o->mot_write_count++;
160         spin_unlock(&o->mot_write_lock);
161
162         RETURN(rc);
163 }
164
165 void mdt_write_put(struct mdt_object *o)
166 {
167         ENTRY;
168         spin_lock(&o->mot_write_lock);
169         o->mot_write_count--;
170         spin_unlock(&o->mot_write_lock);
171         EXIT;
172 }
173
174 static int mdt_write_deny(struct mdt_object *o)
175 {
176         int rc = 0;
177
178         ENTRY;
179         spin_lock(&o->mot_write_lock);
180         if (o->mot_write_count > 0)
181                 rc = -ETXTBSY;
182         else
183                 o->mot_write_count--;
184         spin_unlock(&o->mot_write_lock);
185         RETURN(rc);
186 }
187
188 static void mdt_write_allow(struct mdt_object *o)
189 {
190         ENTRY;
191         spin_lock(&o->mot_write_lock);
192         o->mot_write_count++;
193         spin_unlock(&o->mot_write_lock);
194         EXIT;
195 }
196
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)
199 {
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;
204
205         ENTRY;
206         if (mdt_rdonly(req->rq_export))
207                 RETURN_EXIT;
208
209         /* transaction has occurred already */
210         if (lustre_msg_get_transno(req->rq_repmsg) != 0)
211                 RETURN_EXIT;
212
213         if (tgt_is_multimodrpcs_client(req->rq_export)) {
214                 struct thandle *th;
215
216                 /* generate an empty transaction to get a transno
217                  * and reply data */
218                 th = dt_trans_create(info->mti_env, mdt->mdt_bottom);
219                 if (!IS_ERR(th)) {
220                         rc = dt_trans_start(info->mti_env, mdt->mdt_bottom, th);
221                         dt_trans_stop(info->mti_env, mdt->mdt_bottom, th);
222                 }
223                 RETURN_EXIT;
224         }
225
226         spin_lock(&mdt->mdt_lut.lut_translock);
227         if (rc != 0) {
228                 if (info->mti_transno != 0) {
229                         struct obd_export *exp = req->rq_export;
230
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);
235                         RETURN_EXIT;
236                 }
237         } else if (info->mti_transno == 0) {
238                 info->mti_transno = ++mdt->mdt_lut.lut_last_transno;
239         } else {
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;
243         }
244         spin_unlock(&mdt->mdt_lut.lut_translock);
245
246         CDEBUG(D_INODE, "transno = %llu, last_committed = %llu\n",
247                info->mti_transno,
248                req->rq_export->exp_obd->obd_last_committed);
249
250         req->rq_transno = info->mti_transno;
251         lustre_msg_set_transno(req->rq_repmsg, info->mti_transno);
252
253         /* update lcd in memory only for resent cases */
254         ted = &req->rq_export->exp_target_data;
255         LASSERT(ted);
256         mutex_lock(&ted->ted_lcd_lock);
257         lcd = ted->ted_lcd;
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);
268                 RETURN_EXIT;
269         }
270
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;
276         } else {
277                 /* VBR: save versions in last_rcvd for reconstruct. */
278                 __u64 *pre_versions = lustre_msg_get_versions(req->rq_repmsg);
279
280                 if (pre_versions) {
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];
285                 }
286                 if (info->mti_transno != 0)
287                         lcd->lcd_last_transno = info->mti_transno;
288
289                 lcd->lcd_last_xid = req->rq_xid;
290                 lcd->lcd_last_result = rc;
291                 lcd->lcd_last_data = info->mti_opdata;
292         }
293         mutex_unlock(&ted->ted_lcd_lock);
294
295         EXIT;
296 }
297
298 void mdt_mfd_set_mode(struct mdt_file_data *mfd, u64 open_flags)
299 {
300         LASSERT(mfd != NULL);
301
302         CDEBUG(D_DENTRY, DFID " Change mfd open_flags %#llo -> %#llo.\n",
303                PFID(mdt_object_fid(mfd->mfd_object)), mfd->mfd_open_flags,
304                open_flags);
305
306         mfd->mfd_open_flags = open_flags;
307 }
308
309 /**
310  * prep ma_lmm/ma_lmv for md_attr from reply
311  */
312 static void mdt_prep_ma_buf_from_rep(struct mdt_thread_info *info,
313                                      struct mdt_object *obj,
314                                      struct md_attr *ma, __u64 open_flags)
315 {
316         struct req_capsule *pill;
317
318         if (ma->ma_lmv || ma->ma_lmm) {
319                 CDEBUG(D_INFO, DFID " %s already set.\n",
320                        PFID(mdt_object_fid(obj)),
321                        ma->ma_lmv ? (ma->ma_lmm ? "ma_lmv and ma_lmm"
322                                                 : "ma_lmv")
323                                   : "ma_lmm");
324                 return;
325         }
326
327         pill = info->mti_pill;
328         if (S_ISDIR(obj->mot_header.loh_attr)) {
329                 ma->ma_lmv = req_capsule_server_get(pill, &RMF_MDT_MD);
330                 ma->ma_lmv_size = req_capsule_get_size(pill, &RMF_MDT_MD,
331                                                        RCL_SERVER);
332                 if (ma->ma_lmv_size > 0)
333                         ma->ma_need |= MA_LMV;
334
335                 if (open_flags & MDS_OPEN_DEFAULT_LMV) {
336                         ma->ma_default_lmv = req_capsule_server_get(pill,
337                                                         &RMF_DEFAULT_MDT_MD);
338                         ma->ma_default_lmv_size = req_capsule_get_size(pill,
339                                                         &RMF_DEFAULT_MDT_MD,
340                                                         RCL_SERVER);
341                         if (ma->ma_default_lmv_size > 0)
342                                 ma->ma_need |= MA_LMV_DEF;
343                 }
344         } else {
345                 ma->ma_lmm = req_capsule_server_get(pill, &RMF_MDT_MD);
346                 ma->ma_lmm_size = req_capsule_get_size(pill, &RMF_MDT_MD,
347                                                        RCL_SERVER);
348                 if (ma->ma_lmm_size > 0)
349                         ma->ma_need |= MA_LOV;
350         }
351 }
352
353 static int mdt_mfd_open(struct mdt_thread_info *info, struct mdt_object *p,
354                         struct mdt_object *o, u64 open_flags, int created,
355                         struct ldlm_reply *rep)
356 {
357         struct ptlrpc_request *req = mdt_info_req(info);
358         struct mdt_export_data *med = &req->rq_export->exp_mdt_data;
359         struct mdt_file_data *mfd;
360         struct md_attr *ma  = &info->mti_attr;
361         struct lu_attr *la  = &ma->ma_attr;
362         struct mdt_body *repbody;
363         bool isdir, isreg;
364         int rc = 0;
365
366         ENTRY;
367         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
368
369         isreg = S_ISREG(la->la_mode);
370         isdir = S_ISDIR(la->la_mode);
371         if (isreg && !(ma->ma_valid & MA_LOV) &&
372             !(open_flags & MDS_OPEN_RELEASE)) {
373                 /*
374                  * No EA, check whether it is will set regEA and dirEA since in
375                  * above attr get, these size might be zero, so reset it, to
376                  * retrieve the MD after create obj.
377                  */
378                 ma->ma_lmm_size = req_capsule_get_size(info->mti_pill,
379                                                        &RMF_MDT_MD,
380                                                        RCL_SERVER);
381                 /* in replay case, p == NULL */
382                 rc = mdt_create_data(info, p, o);
383                 if (rc)
384                         RETURN(rc);
385
386                 if (exp_connect_flags(req->rq_export) & OBD_CONNECT_DISP_STRIPE)
387                         mdt_set_disposition(info, rep, DISP_OPEN_STRIPE);
388         }
389
390         CDEBUG(D_INODE, "after open, ma_valid bit = %#llx lmm_size = %d\n",
391                ma->ma_valid, ma->ma_lmm_size);
392
393         if (ma->ma_valid & MA_LOV) {
394                 LASSERT(ma->ma_lmm_size != 0);
395                 repbody->mbo_eadatasize = ma->ma_lmm_size;
396                 if (isdir)
397                         repbody->mbo_valid |= OBD_MD_FLDIREA;
398                 else
399                         repbody->mbo_valid |= OBD_MD_FLEASIZE;
400         }
401
402         if (ma->ma_valid & MA_LMV) {
403                 LASSERT(ma->ma_lmv_size != 0);
404                 repbody->mbo_eadatasize = ma->ma_lmv_size;
405                 LASSERT(isdir);
406                 repbody->mbo_valid |= OBD_MD_FLDIREA | OBD_MD_MEA;
407         }
408
409         if (ma->ma_valid & MA_LMV_DEF) {
410                 LASSERT(ma->ma_default_lmv_size != 0);
411                 LASSERT(isdir);
412                 repbody->mbo_valid |= OBD_MD_FLDIREA | OBD_MD_DEFAULT_MEA;
413         }
414
415         if (open_flags & MDS_FMODE_WRITE)
416                 rc = mdt_write_get(o);
417         else if (open_flags & MDS_FMODE_EXEC)
418                 rc = mdt_write_deny(o);
419
420         if (rc)
421                 RETURN(rc);
422
423         rc = mo_open(info->mti_env, mdt_object_child(o),
424                      created ? open_flags | MDS_OPEN_CREATED : open_flags,
425                      &info->mti_spec);
426         if (rc != 0) {
427                 /* If we allow the client to chgrp (CFS_SETGRP_PERM), but the
428                  * client does not know which suppgid should be sent to the MDS,
429                  * or some other(s) changed the target file's GID after this RPC
430                  * sent to the MDS with the suppgid as the original GID, then we
431                  * give the client another chance to send the right suppgid. */
432                 if (rc == -EACCES &&
433                     allow_client_chgrp(info, lu_ucred(info->mti_env)))
434                         mdt_set_disposition(info, rep, DISP_OPEN_DENY);
435
436                 GOTO(err_out, rc);
437         }
438
439         mfd = mdt_mfd_new(med);
440         if (mfd == NULL)
441                 GOTO(err_out, rc = -ENOMEM);
442
443         /*
444          * Keep a reference on this object for this open, and is
445          * released by mdt_mfd_close().
446          */
447         mdt_object_get(info->mti_env, o);
448         mfd->mfd_object = o;
449         mfd->mfd_xid = req->rq_xid;
450
451         /*
452          * @open_flags is always not zero. At least it should be FMODE_READ,
453          * FMODE_WRITE or MDS_FMODE_EXEC.
454          */
455         LASSERT(open_flags != 0);
456
457         /* Open handling. */
458         mdt_mfd_set_mode(mfd, open_flags);
459
460         atomic_inc(&o->mot_open_count);
461         if (open_flags & MDS_OPEN_LEASE)
462                 atomic_inc(&o->mot_lease_count);
463
464         /* replay handle */
465         if (req_is_replay(req)) {
466                 struct mdt_file_data *old_mfd;
467                 /* Check wheather old cookie already exist in
468                  * the list, becasue when do recovery, client
469                  * might be disconnected from server, and
470                  * restart replay, so there maybe some orphan
471                  * mfd here, we should remove them */
472                 LASSERT(info->mti_rr.rr_open_handle != NULL);
473                 spin_lock(&med->med_open_lock);
474                 old_mfd = mdt_open_handle2mfd(med, info->mti_rr.rr_open_handle,
475                                               true);
476                 if (old_mfd != NULL) {
477                         CDEBUG(D_HA, "delete orphan mfd = %p, fid = "DFID", "
478                                "cookie = %#llx\n", mfd,
479                                PFID(mdt_object_fid(mfd->mfd_object)),
480                                info->mti_rr.rr_open_handle->cookie);
481                         class_handle_unhash(&old_mfd->mfd_open_handle);
482                         list_del_init(&old_mfd->mfd_list);
483                         spin_unlock(&med->med_open_lock);
484                         /* no attr update for that close */
485                         la->la_valid = 0;
486                         ma->ma_valid |= MA_FLAGS;
487                         ma->ma_attr_flags |= MDS_RECOV_OPEN;
488                         mdt_mfd_close(info, old_mfd);
489                         ma->ma_attr_flags &= ~MDS_RECOV_OPEN;
490                         ma->ma_valid &= ~MA_FLAGS;
491                 } else {
492                         spin_unlock(&med->med_open_lock);
493                         CDEBUG(D_HA, "orphan mfd not found, fid = "DFID", "
494                                "cookie = %#llx\n",
495                                PFID(mdt_object_fid(mfd->mfd_object)),
496                                info->mti_rr.rr_open_handle->cookie);
497                 }
498
499                 CDEBUG(D_HA, "Store old cookie %#llx in new mfd\n",
500                        info->mti_rr.rr_open_handle->cookie);
501
502                 mfd->mfd_open_handle_old = *info->mti_rr.rr_open_handle;
503         }
504
505         repbody->mbo_open_handle.cookie = mfd->mfd_open_handle.h_cookie;
506
507         if (req->rq_export->exp_disconnected) {
508                 spin_lock(&med->med_open_lock);
509                 class_handle_unhash(&mfd->mfd_open_handle);
510                 list_del_init(&mfd->mfd_list);
511                 spin_unlock(&med->med_open_lock);
512                 mdt_mfd_close(info, mfd);
513         } else {
514                 spin_lock(&med->med_open_lock);
515                 list_add_tail(&mfd->mfd_list, &med->med_open_head);
516                 spin_unlock(&med->med_open_lock);
517         }
518
519         mdt_empty_transno(info, rc);
520
521         RETURN(rc);
522
523 err_out:
524         if (open_flags & MDS_FMODE_WRITE)
525                 mdt_write_put(o);
526         else if (open_flags & MDS_FMODE_EXEC)
527                 mdt_write_allow(o);
528
529         return rc;
530 }
531
532 static int mdt_finish_open(struct mdt_thread_info *info,
533                            struct mdt_object *p, struct mdt_object *o,
534                            u64 open_flags,
535                            struct ldlm_reply *rep)
536 {
537         struct ptlrpc_request *req = mdt_info_req(info);
538         struct obd_export *exp = req->rq_export;
539         struct mdt_export_data *med = &req->rq_export->exp_mdt_data;
540         struct md_attr *ma  = &info->mti_attr;
541         struct lu_attr *la  = &ma->ma_attr;
542         struct mdt_file_data *mfd;
543         struct mdt_body *repbody;
544         int created;
545         int rc = 0;
546         int isreg, isdir, islnk;
547         struct list_head *t;
548
549         ENTRY;
550         LASSERT(ma->ma_valid & MA_INODE);
551         created = mdt_get_disposition(rep, DISP_OPEN_CREATE);
552
553         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
554
555         isreg = S_ISREG(la->la_mode);
556         isdir = S_ISDIR(la->la_mode);
557         islnk = S_ISLNK(la->la_mode);
558         mdt_pack_attr2body(info, repbody, la, mdt_object_fid(o));
559
560         /* compatibility check for 2.10 clients when it tries to open mirrored
561          * files. 2.10 clients don't verify overlapping components so they
562          * would read and write mirrored files just as if they were normal
563          * PFL files, which will cause the problem that sycned mirrors actually
564          * contain different data.
565          * Older clients are not a concern here because they don't even
566          * understand PFL layout. */
567         if (isreg && !exp_connect_flr(exp) && ma->ma_valid & MA_LOV &&
568             mdt_lmm_is_flr(ma->ma_lmm)) {
569                 /* LU-10286: for simplicity clients who don't understand
570                  * mirrored layout(with connect flag OBD_CONNECT2_FLR) won't
571                  * be able to open mirrored files */
572                 RETURN(-EOPNOTSUPP);
573         }
574
575         /* Overstriped files can crash older clients */
576         if (isreg && !exp_connect_overstriping(exp) &&
577             mdt_lmm_is_overstriping(ma->ma_lmm))
578                 RETURN(-EOPNOTSUPP);
579
580         /* LU-2275, simulate broken behaviour (esp. prevalent in
581          * pre-2.4 servers where a very strange reply is sent on error
582          * that looks like it was actually almost successful and a
583          * failure at the same time.) */
584         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_NEGATIVE_POSITIVE)) {
585                 mdt_set_disposition(info, rep, DISP_OPEN_OPEN |
586                                                DISP_LOOKUP_NEG |
587                                                DISP_LOOKUP_POS);
588
589                 if (open_flags & MDS_OPEN_LOCK)
590                         mdt_set_disposition(info, rep, DISP_OPEN_LOCK);
591
592                 RETURN(-ENOENT);
593         }
594
595 #ifdef CONFIG_LUSTRE_FS_POSIX_ACL
596         if (exp_connect_flags(exp) & OBD_CONNECT_ACL) {
597                 struct lu_nodemap *nodemap = nodemap_get_from_exp(exp);
598                 if (IS_ERR(nodemap))
599                         RETURN(PTR_ERR(nodemap));
600
601                 rc = mdt_pack_acl2body(info, repbody, o, nodemap);
602                 nodemap_putref(nodemap);
603                 if (rc)
604                         RETURN(rc);
605         }
606 #endif
607
608         /*
609          * If we are following a symlink, don't open; and do not return open
610          * handle for special nodes as client required.
611          */
612         if (islnk || (!isreg && !isdir &&
613             (exp_connect_flags(req->rq_export) & OBD_CONNECT_NODEVOH))) {
614                 lustre_msg_set_transno(req->rq_repmsg, 0);
615                 RETURN(0);
616         }
617
618         /*
619          * We need to return the existing object's fid back, so it is done here,
620          * after preparing the reply.
621          */
622         if (!created && (open_flags & MDS_OPEN_EXCL) &&
623             (open_flags & MDS_OPEN_CREAT))
624                 RETURN(-EEXIST);
625
626         /* This can't be done earlier, we need to return reply body */
627         if (isdir) {
628                 if (open_flags & (MDS_OPEN_CREAT | MDS_FMODE_WRITE)) {
629                         /* We are trying to create or write an existing dir. */
630                         RETURN(-EISDIR);
631                 }
632         } else if (open_flags & MDS_OPEN_DIRECTORY)
633                 RETURN(-ENOTDIR);
634
635         if (OBD_FAIL_CHECK_RESET(OBD_FAIL_MDS_OPEN_CREATE,
636                                  OBD_FAIL_MDS_LDLM_REPLY_NET | OBD_FAIL_ONCE))
637                 RETURN(-EAGAIN);
638
639         mfd = NULL;
640         if (info->mti_rr.rr_flags & MRF_OPEN_RESEND) {
641                 spin_lock(&med->med_open_lock);
642                 list_for_each(t, &med->med_open_head) {
643                         mfd = list_entry(t, struct mdt_file_data, mfd_list);
644                         if (mfd->mfd_xid == req->rq_xid) {
645                                 repbody->mbo_open_handle.cookie =
646                                                 mfd->mfd_open_handle.h_cookie;
647                                 break;
648                         }
649                         mfd = NULL;
650                 }
651                 spin_unlock(&med->med_open_lock);
652
653                 if (mfd != NULL) {
654                         /* set repbody->ea_size for resent case */
655                         if (ma->ma_valid & MA_LOV) {
656                                 LASSERT(ma->ma_lmm_size != 0);
657                                 repbody->mbo_eadatasize = ma->ma_lmm_size;
658                                 if (isdir)
659                                         repbody->mbo_valid |= OBD_MD_FLDIREA;
660                                 else
661                                         repbody->mbo_valid |= OBD_MD_FLEASIZE;
662                         }
663                         mdt_set_disposition(info, rep, DISP_OPEN_OPEN);
664                         RETURN(0);
665                 }
666                 /* if we have a real resend (not a resend afrer failover), it
667                  * means close is already happend, so lets return error
668                  */
669                 RETURN(-ESTALE);
670         }
671
672         rc = mdt_mfd_open(info, p, o, open_flags, created, rep);
673         if (!rc)
674                 mdt_set_disposition(info, rep, DISP_OPEN_OPEN);
675
676         RETURN(rc);
677 }
678
679 void mdt_reconstruct_open(struct mdt_thread_info *info,
680                           struct mdt_lock_handle *lhc)
681 {
682         struct req_capsule *pill = info->mti_pill;
683         struct ptlrpc_request *req = mdt_info_req(info);
684         struct mdt_reint_record *rr = &info->mti_rr;
685         struct md_attr *ma = &info->mti_attr;
686         struct ldlm_reply *ldlm_rep;
687         u64 opdata;
688         int rc;
689
690         ENTRY;
691         LASSERT(pill->rc_fmt == &RQF_LDLM_INTENT_OPEN);
692         ldlm_rep = req_capsule_server_get(pill, &RMF_DLM_REP);
693
694         ma->ma_need = MA_INODE | MA_HSM;
695         ma->ma_valid = 0;
696         opdata = mdt_req_from_lrd(req, info->mti_reply_data);
697         mdt_set_disposition(info, ldlm_rep, opdata);
698
699         CDEBUG(D_INODE, "This is reconstruct open: disp=%#llx, result=%d\n",
700                ldlm_rep->lock_policy_res1, req->rq_status);
701         if (req->rq_status)
702                 /* We did not create successfully, return error to client. */
703                 GOTO(out, rc = req->rq_status);
704
705         /* tg_reply_data is just memory only  structure, so any non zero fid
706          * means a real resend not a resend after recovery which need to be
707          * handled as regular open
708          */
709         if (likely(!fid_is_zero(&info->mti_reply_data->trd_object))) {
710                 rr->rr_fid2 = &info->mti_reply_data->trd_object;
711                 rr->rr_flags |= MRF_OPEN_RESEND;
712                 rc = mdt_open_by_fid(info, ldlm_rep, lhc);
713                 if (rc)
714                         lustre_msg_set_transno(req->rq_repmsg, 0);
715         } else {
716                 /* We did not try to create, so we are a pure open */
717                 rc = mdt_reint_open(info, lhc);
718         }
719         EXIT;
720 out:
721         req->rq_status = rc;
722         lustre_msg_set_status(req->rq_repmsg, req->rq_status);
723         LASSERT(ergo(rc < 0, lustre_msg_get_transno(req->rq_repmsg) == 0));
724 }
725
726 static int mdt_open_by_fid(struct mdt_thread_info *info, struct ldlm_reply *rep,
727                            struct mdt_lock_handle *lhc)
728 {
729         u64 open_flags = info->mti_spec.sp_cr_flags;
730         struct mdt_reint_record *rr = &info->mti_rr;
731         struct md_attr *ma = &info->mti_attr;
732         struct mdt_object *o;
733         int rc;
734
735         ENTRY;
736         o = mdt_object_find(info->mti_env, info->mti_mdt, rr->rr_fid2);
737         if (IS_ERR(o))
738                 RETURN(rc = PTR_ERR(o));
739
740         rc = mdt_check_enc(info, o);
741         if (rc)
742                 GOTO(out, rc);
743
744         if (unlikely(mdt_object_remote(o))) {
745                 /* the child object was created on remote server */
746                 struct mdt_body *repbody;
747
748                 mdt_set_disposition(info, rep, (DISP_IT_EXECD |
749                                                 DISP_LOOKUP_EXECD |
750                                                 DISP_LOOKUP_POS));
751                 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
752                 repbody->mbo_fid1 = *rr->rr_fid2;
753                 repbody->mbo_valid |= (OBD_MD_FLID | OBD_MD_MDS);
754                 rc = 0;
755         } else {
756                 if (mdt_object_exists(o)) {
757                         tgt_open_obj_set(info->mti_env, mdt_obj2dt(o));
758                         mdt_set_disposition(info, rep, (DISP_IT_EXECD |
759                                                         DISP_LOOKUP_EXECD |
760                                                         DISP_LOOKUP_POS));
761                         if ((open_flags & MDS_OPEN_EXCL) &&
762                             (open_flags & MDS_OPEN_CREAT))
763                                 mdt_set_disposition(info, rep,
764                                                     DISP_OPEN_CREATE);
765
766                         mdt_prep_ma_buf_from_rep(info, o, ma, open_flags);
767                         rc = mdt_attr_get_complex(info, o, ma);
768                         if (rc)
769                                 GOTO(out, rc);
770                         rc = mdt_finish_open(info, NULL, o, open_flags, rep);
771                         if (rc)
772                                 GOTO(out, rc);
773                         mdt_pack_size2body(info, rr->rr_fid2, &lhc->mlh_reg_lh);
774                 } else {
775                         rc = -ENOENT;
776                 }
777         }
778
779 out:
780         mdt_object_put(info->mti_env, o);
781         RETURN(rc);
782 }
783
784 /* lock object for open */
785 static int mdt_object_open_lock(struct mdt_thread_info *info,
786                                 struct mdt_object *obj,
787                                 struct mdt_lock_handle *lhc,
788                                 __u64 *ibits)
789 {
790         struct md_attr *ma = &info->mti_attr;
791         __u64 open_flags = info->mti_spec.sp_cr_flags;
792         __u64 trybits = 0;
793         enum ldlm_mode lm = LCK_CR;
794         bool acq_lease = !!(open_flags & MDS_OPEN_LEASE);
795         bool try_layout = false;
796         bool create_layout = false;
797         int rc = 0;
798         __u32 dom_stripe = 0;
799         unsigned int dom_only = 0;
800         unsigned int dom_lock = 0;
801
802         ENTRY;
803         *ibits = 0;
804         mdt_lock_handle_init(lhc);
805
806         if (req_is_replay(mdt_info_req(info)))
807                 RETURN(0);
808
809         if (S_ISREG(lu_object_attr(&obj->mot_obj))) {
810                 if (ma->ma_need & MA_LOV && !(ma->ma_valid & MA_LOV) &&
811                     md_should_create(open_flags))
812                         create_layout = true;
813                 if (exp_connect_layout(info->mti_exp) && !create_layout &&
814                     ma->ma_need & MA_LOV)
815                         try_layout = true;
816
817                 /* DoM files can take IO lock at OPEN when it makes sense,
818                  * check if file has DoM stripe and ask for lock if client
819                  * no lock on that resource yet.
820                  */
821                 if (ma->ma_valid & MA_LOV && ma->ma_lmm != NULL)
822                         dom_stripe = mdt_lmm_dom_entry_check(ma->ma_lmm,
823                                                              &dom_only);
824                 /* If only DOM stripe is being used then we can expect IO
825                  * to it after OPEN and will return corresponding DOM ibit
826                  * using default strategy from mdt_opts.mo_dom_lock.
827                  * Otherwise trylock mode is used always and DOM ibit will
828                  * be returned optionally.
829                  */
830                 if (dom_stripe &&
831                     !mdt_dom_client_has_lock(info, mdt_object_fid(obj)))
832                         dom_lock = !dom_only ? TRYLOCK_DOM_ON_OPEN :
833                                    info->mti_mdt->mdt_opts.mo_dom_lock;
834         }
835
836         if (acq_lease) {
837                 /* lease open, acquire write mode of open sem */
838                 down_write(&obj->mot_open_sem);
839
840                 /* Lease exists and ask for new lease */
841                 if (atomic_read(&obj->mot_lease_count) > 0) {
842                         /* only exclusive open is supported, so lease
843                          * are conflicted to each other */
844                         GOTO(out, rc = -EBUSY);
845                 }
846
847                 /* Lease must be with open lock */
848                 if (!(open_flags & MDS_OPEN_LOCK)) {
849                         CERROR("%s: Request lease for file:"DFID ", but open lock "
850                                "is missed, open_flags = %#llo : rc = %d\n",
851                                mdt_obd_name(info->mti_mdt),
852                                PFID(mdt_object_fid(obj)), open_flags, -EPROTO);
853                         GOTO(out, rc = -EPROTO);
854                 }
855
856                 /* should conflict with new opens for write/execute */
857                 lm = LCK_PW;
858                 *ibits = MDS_INODELOCK_OPEN;
859
860                 /* never grant LCK_EX layout lock to client */
861                 try_layout = false;
862         } else { /* normal open */
863                 /* normal open holds read mode of open sem */
864                 down_read(&obj->mot_open_sem);
865
866                 if (open_flags & MDS_OPEN_LOCK) {
867                         if (open_flags & MDS_FMODE_WRITE)
868                                 lm = LCK_CW;
869                         else if (open_flags & MDS_FMODE_EXEC)
870                                 lm = LCK_PR;
871                         else
872                                 lm = LCK_CR;
873
874                         *ibits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_OPEN;
875                 } else if (atomic_read(&obj->mot_lease_count) > 0) {
876                         if (open_flags & MDS_FMODE_WRITE)
877                                 lm = LCK_CW;
878                         else
879                                 lm = LCK_CR;
880
881                         /* revoke lease */
882                         *ibits = MDS_INODELOCK_OPEN;
883                         try_layout = false;
884
885                         lhc = &info->mti_lh[MDT_LH_LOCAL];
886                 } else if (dom_lock) {
887                         lm = (open_flags & MDS_FMODE_WRITE) ? LCK_PW : LCK_PR;
888                         trybits |= MDS_INODELOCK_DOM | MDS_INODELOCK_LAYOUT;
889                 }
890
891                 CDEBUG(D_INODE, "normal open:"DFID" lease count: %d, lm: %d\n",
892                         PFID(mdt_object_fid(obj)),
893                         atomic_read(&obj->mot_lease_count), lm);
894         }
895
896         mdt_lock_reg_init(lhc, lm);
897
898         /* Return lookup lock to validate inode at the client side.
899          * This is pretty important otherwise MDT will return layout
900          * lock for each open.
901          * However this is a double-edged sword because changing
902          * permission will revoke a huge number of LOOKUP locks.
903          */
904         if (!OBD_FAIL_CHECK(OBD_FAIL_MDS_NO_LL_OPEN) && try_layout) {
905                 if (!(*ibits & MDS_INODELOCK_LOOKUP))
906                         trybits |= MDS_INODELOCK_LOOKUP;
907                 trybits |= MDS_INODELOCK_LAYOUT;
908         }
909
910         if (*ibits | trybits)
911                 rc = mdt_object_lock_try(info, obj, lhc, ibits, trybits, false);
912
913         CDEBUG(D_INODE, "%s: Requested bits lock:"DFID ", ibits = %#llx/%#llx"
914                ", open_flags = %#llo, try_layout = %d : rc = %d\n",
915                mdt_obd_name(info->mti_mdt), PFID(mdt_object_fid(obj)),
916                *ibits, trybits, open_flags, try_layout, rc);
917
918         /* will change layout, revoke layout locks by enqueuing EX lock. */
919         if (rc == 0 && create_layout) {
920                 struct mdt_lock_handle *ll = &info->mti_lh[MDT_LH_LAYOUT];
921
922                 CDEBUG(D_INODE, "Will create layout, get EX layout lock:"DFID
923                         ", open_flags = %#llo\n",
924                         PFID(mdt_object_fid(obj)), open_flags);
925
926                 /* We cannot enqueue another lock for the same resource we
927                  * already have a lock for, due to mechanics of waiting list
928                  * iterating in ldlm, see LU-3601.
929                  * As such we'll drop the open lock we just got above here,
930                  * it's ok not to have this open lock as it's main purpose is to
931                  * flush unused cached client open handles. */
932                 if (lustre_handle_is_used(&lhc->mlh_reg_lh))
933                         mdt_object_unlock(info, obj, lhc, 1);
934
935                 LASSERT(!try_layout);
936                 mdt_lock_handle_init(ll);
937                 mdt_lock_reg_init(ll, LCK_EX);
938                 rc = mdt_object_lock(info, obj, ll, MDS_INODELOCK_LAYOUT);
939
940                 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_LL_BLOCK, 2);
941         }
942
943         /* Check if there is any other open handles after acquiring
944          * open lock. At this point, caching open handles have been revoked
945          * by open lock.
946          * XXX: Now only exclusive open is supported. Need to check the
947          * type of open for generic lease support. */
948         if (rc == 0 && acq_lease) {
949                 struct ptlrpc_request *req = mdt_info_req(info);
950                 struct mdt_export_data *med = &req->rq_export->exp_mdt_data;
951                 struct mdt_file_data *mfd;
952                 bool is_replay_or_resent;
953                 int open_count = 0;
954
955                 /* For lease: application can open a file and then apply lease,
956                  * @handle contains original open handle in that case.
957                  * In recovery, open REQ will be replayed and the lease REQ may
958                  * be resent that means the open handle is already stale, so we
959                  * need to fix it up here by finding new handle. */
960                 is_replay_or_resent = req_is_replay(req) ||
961                         lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT;
962
963                 /* if the request is _not_ a replay request, rr_open_handle
964                  * may be used to hold an open file handle which is issuing the
965                  * lease request, so that this openhandle doesn't count. */
966                 mfd = mdt_open_handle2mfd(med, info->mti_rr.rr_open_handle,
967                                           is_replay_or_resent);
968                 if (mfd != NULL)
969                         ++open_count;
970
971                 CDEBUG(D_INODE, "acq_lease "DFID": openers: %d, want: %d\n",
972                         PFID(mdt_object_fid(obj)),
973                         atomic_read(&obj->mot_open_count), open_count);
974
975                 if (atomic_read(&obj->mot_open_count) > open_count) {
976                         /* fail if anyone *else* has opened file for write */
977                         if (mdt_write_read(obj) > 1)
978                                 GOTO(out, rc = -EBUSY);
979                 }
980         }
981         GOTO(out, rc);
982
983 out:
984         RETURN(rc);
985 }
986
987 static void mdt_object_open_unlock(struct mdt_thread_info *info,
988                                    struct mdt_object *obj,
989                                    struct mdt_lock_handle *lhc,
990                                    __u64 ibits, int rc)
991 {
992         __u64 open_flags = info->mti_spec.sp_cr_flags;
993         struct mdt_lock_handle *ll = &info->mti_lh[MDT_LH_LOCAL];
994
995         ENTRY;
996         if (req_is_replay(mdt_info_req(info)))
997                 RETURN_EXIT;
998
999         /* Release local lock - the lock put in MDT_LH_LOCAL will never
1000          * return to client side. */
1001         if (lustre_handle_is_used(&ll->mlh_reg_lh))
1002                 mdt_object_unlock(info, obj, ll, 1);
1003
1004         ll = &info->mti_lh[MDT_LH_LAYOUT];
1005         /* Release local layout lock, layout was created */
1006         if (lustre_handle_is_used(&ll->mlh_reg_lh)) {
1007                 LASSERT(!(ibits & MDS_INODELOCK_LAYOUT));
1008                 mdt_object_unlock(info, obj, ll, 1);
1009         }
1010
1011         if (open_flags & MDS_OPEN_LEASE)
1012                 up_write(&obj->mot_open_sem);
1013         else
1014                 up_read(&obj->mot_open_sem);
1015
1016         /* Cross-ref case, the lock should be returned to the client */
1017         if (ibits == 0 || rc == -MDT_EREMOTE_OPEN)
1018                 RETURN_EXIT;
1019
1020         if (!(open_flags & MDS_OPEN_LOCK) && !(ibits & MDS_INODELOCK_LAYOUT) &&
1021             !(ibits & MDS_INODELOCK_DOM)) {
1022                 /* for the open request, the lock will only return to client
1023                  * if open or layout lock is granted. */
1024                 rc = 1;
1025         }
1026
1027         if (rc != 0 || !lustre_handle_is_used(&lhc->mlh_reg_lh)) {
1028                 struct ldlm_reply       *ldlm_rep;
1029
1030                 ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
1031                 mdt_clear_disposition(info, ldlm_rep, DISP_OPEN_LOCK);
1032                 if (lustre_handle_is_used(&lhc->mlh_reg_lh))
1033                         mdt_object_unlock(info, obj, lhc, 1);
1034         }
1035         RETURN_EXIT;
1036 }
1037
1038 /**
1039  * Check release is permitted for the current HSM flags.
1040  */
1041 static bool mdt_hsm_release_allow(const struct md_attr *ma)
1042 {
1043         if (!(ma->ma_valid & MA_HSM))
1044                 return false;
1045
1046         if (ma->ma_hsm.mh_flags & (HS_DIRTY|HS_NORELEASE|HS_LOST))
1047                 return false;
1048
1049         if (!(ma->ma_hsm.mh_flags & HS_ARCHIVED))
1050                 return false;
1051
1052         return true;
1053 }
1054
1055 static int mdt_open_by_fid_lock(struct mdt_thread_info *info,
1056                                 struct ldlm_reply *rep,
1057                                 struct mdt_lock_handle *lhc)
1058 {
1059         const struct lu_env *env = info->mti_env;
1060         struct mdt_device *mdt = info->mti_mdt;
1061         u64 open_flags = info->mti_spec.sp_cr_flags;
1062         struct mdt_reint_record *rr = &info->mti_rr;
1063         struct md_attr *ma = &info->mti_attr;
1064         struct mdt_object *parent = NULL;
1065         struct mdt_object *o;
1066         bool object_locked = false;
1067         u64 ibits = 0;
1068         int rc;
1069
1070         ENTRY;
1071         if (md_should_create(open_flags)) {
1072                 if (!lu_fid_eq(rr->rr_fid1, rr->rr_fid2)) {
1073                         parent = mdt_object_find(env, mdt, rr->rr_fid1);
1074                         if (IS_ERR(parent)) {
1075                                 CDEBUG(D_INODE, "Fail to find parent "DFID
1076                                        " for anonymous created %ld, try to"
1077                                        " use server-side parent.\n",
1078                                        PFID(rr->rr_fid1), PTR_ERR(parent));
1079                                 parent = NULL;
1080                         }
1081                 }
1082                 if (parent == NULL)
1083                         ma->ma_need |= MA_PFID;
1084         }
1085
1086         o = mdt_object_find(env, mdt, rr->rr_fid2);
1087         if (IS_ERR(o))
1088                 GOTO(out_parent_put, rc = PTR_ERR(o));
1089
1090         if (mdt_object_remote(o)) {
1091                 CDEBUG(D_INFO, "%s: "DFID" is on remote MDT.\n",
1092                        mdt_obd_name(info->mti_mdt),
1093                        PFID(rr->rr_fid2));
1094                 GOTO(out, rc = -EREMOTE);
1095         } else if (!mdt_object_exists(o)) {
1096                 mdt_set_disposition(info, rep,
1097                                     DISP_IT_EXECD |
1098                                     DISP_LOOKUP_EXECD |
1099                                     DISP_LOOKUP_NEG);
1100                 GOTO(out, rc = -ENOENT);
1101         }
1102
1103         /* do not check enc for directory: always allow open */
1104         if (!S_ISDIR(lu_object_attr(&o->mot_obj))) {
1105                 rc = mdt_check_enc(info, o);
1106                 if (rc)
1107                         GOTO(out, rc);
1108         }
1109
1110         mdt_set_disposition(info, rep, (DISP_IT_EXECD | DISP_LOOKUP_EXECD));
1111
1112         mdt_prep_ma_buf_from_rep(info, o, ma, open_flags);
1113         if (open_flags & MDS_OPEN_RELEASE)
1114                 ma->ma_need |= MA_HSM;
1115         rc = mdt_attr_get_complex(info, o, ma);
1116         if (rc)
1117                 GOTO(out, rc);
1118
1119         /* We should not change file's existing LOV EA */
1120         if (S_ISREG(lu_object_attr(&o->mot_obj)) &&
1121             open_flags & MDS_OPEN_HAS_EA && ma->ma_valid & MA_LOV)
1122                 GOTO(out, rc = -EEXIST);
1123
1124         /* If a release request, check file open flags are fine and ask for an
1125          * exclusive open access. */
1126         if (open_flags & MDS_OPEN_RELEASE && !mdt_hsm_release_allow(ma))
1127                 GOTO(out, rc = -EPERM);
1128
1129         rc = mdt_check_resent_lock(info, o, lhc);
1130         if (rc < 0) {
1131                 GOTO(out, rc);
1132         } else if (rc > 0) {
1133                 rc = mdt_object_open_lock(info, o, lhc, &ibits);
1134                 object_locked = true;
1135                 if (rc)
1136                         GOTO(out_unlock, rc);
1137         }
1138
1139         if (ma->ma_valid & MA_PFID) {
1140                 parent = mdt_object_find(env, mdt, &ma->ma_pfid);
1141                 if (IS_ERR(parent)) {
1142                         CDEBUG(D_INODE, "Fail to find parent "DFID
1143                                " for anonymous created %ld, try to"
1144                                " use system default.\n",
1145                                PFID(&ma->ma_pfid), PTR_ERR(parent));
1146                         parent = NULL;
1147                 }
1148         }
1149
1150         tgt_open_obj_set(info->mti_env, mdt_obj2dt(o));
1151         rc = mdt_finish_open(info, parent, o, open_flags, rep);
1152         if (!rc) {
1153                 mdt_set_disposition(info, rep, DISP_LOOKUP_POS);
1154                 if (open_flags & MDS_OPEN_LOCK)
1155                         mdt_set_disposition(info, rep, DISP_OPEN_LOCK);
1156                 if (open_flags & MDS_OPEN_LEASE)
1157                         mdt_set_disposition(info, rep, DISP_OPEN_LEASE);
1158         }
1159         GOTO(out_unlock, rc);
1160
1161 out_unlock:
1162         if (object_locked)
1163                 mdt_object_open_unlock(info, o, lhc, ibits, rc);
1164 out:
1165         mdt_object_put(env, o);
1166         if (rc == 0)
1167                 mdt_pack_size2body(info, rr->rr_fid2, &lhc->mlh_reg_lh);
1168 out_parent_put:
1169         if (parent != NULL)
1170                 mdt_object_put(env, parent);
1171         return rc;
1172 }
1173
1174 /* Cross-ref request. Currently it can only be a pure open (w/o create) */
1175 static int mdt_cross_open(struct mdt_thread_info *info,
1176                           const struct lu_fid *parent_fid,
1177                           const struct lu_fid *fid,
1178                           struct ldlm_reply *rep, u64 open_flags)
1179 {
1180         struct md_attr *ma = &info->mti_attr;
1181         struct mdt_object *o;
1182         int rc;
1183
1184         ENTRY;
1185         o = mdt_object_find(info->mti_env, info->mti_mdt, fid);
1186         if (IS_ERR(o))
1187                 RETURN(rc = PTR_ERR(o));
1188
1189         rc = mdt_check_enc(info, o);
1190         if (rc)
1191                 GOTO(out, rc);
1192
1193         if (mdt_object_remote(o)) {
1194                 /* Something is wrong here, the object is on another MDS! */
1195                 CERROR("%s: "DFID" isn't on this server!: rc = %d\n",
1196                        mdt_obd_name(info->mti_mdt), PFID(fid), -EFAULT);
1197                 LU_OBJECT_DEBUG(D_WARNING, info->mti_env,
1198                                 &o->mot_obj,
1199                                 "Object isn't on this server! FLD error?");
1200                 rc = -EFAULT;
1201         } else {
1202                 if (mdt_object_exists(o)) {
1203                         int mask;
1204
1205                         /* Do permission check for cross-open after converting
1206                          * MDS_OPEN_* flags to MAY_* permission mask.
1207                          */
1208                         mask = mds_accmode(open_flags);
1209
1210                         rc = mo_permission(info->mti_env, NULL,
1211                                            mdt_object_child(o), NULL, mask);
1212                         if (rc)
1213                                 goto out;
1214
1215                         mdt_prep_ma_buf_from_rep(info, o, ma, open_flags);
1216                         rc = mdt_attr_get_complex(info, o, ma);
1217                         if (rc != 0)
1218                                 GOTO(out, rc);
1219
1220                         rc = mdt_pack_secctx_in_reply(info, o);
1221                         if (unlikely(rc))
1222                                 GOTO(out, rc);
1223
1224                         rc = mdt_pack_encctx_in_reply(info, o);
1225                         if (unlikely(rc))
1226                                 GOTO(out, rc);
1227
1228                         rc = mdt_finish_open(info, NULL, o, open_flags, rep);
1229                 } else {
1230                         /*
1231                          * Something is wrong here. lookup was positive but
1232                          * there is no object!
1233                          */
1234                         CERROR("%s: "DFID" doesn't exist!: rc = %d\n",
1235                               mdt_obd_name(info->mti_mdt), PFID(fid), -EFAULT);
1236                         rc = -EFAULT;
1237                 }
1238         }
1239 out:
1240         mdt_object_put(info->mti_env, o);
1241
1242         RETURN(rc);
1243 }
1244
1245 /*
1246  * find root object and take its xattr lock if it's on remote MDT, later create
1247  * may use fs default striping (which is stored in root xattr).
1248  */
1249 static int mdt_lock_root_xattr(struct mdt_thread_info *info,
1250                                struct mdt_device *mdt)
1251 {
1252         struct mdt_object *md_root = mdt->mdt_md_root;
1253         struct lustre_handle lhroot;
1254         int rc;
1255
1256         if (md_root == NULL) {
1257                 lu_root_fid(&info->mti_tmp_fid1);
1258                 md_root = mdt_object_find(info->mti_env, mdt,
1259                                           &info->mti_tmp_fid1);
1260                 if (IS_ERR(md_root))
1261                         return PTR_ERR(md_root);
1262
1263                 spin_lock(&mdt->mdt_lock);
1264                 if (mdt->mdt_md_root != NULL) {
1265                         spin_unlock(&mdt->mdt_lock);
1266
1267                         LASSERTF(mdt->mdt_md_root == md_root,
1268                                  "Different root object ("
1269                                  DFID") instances, %p, %p\n",
1270                                  PFID(&info->mti_tmp_fid1),
1271                                  mdt->mdt_md_root, md_root);
1272                         LASSERT(atomic_read(
1273                                 &md_root->mot_obj.lo_header->loh_ref) > 1);
1274
1275                         mdt_object_put(info->mti_env, md_root);
1276                 } else {
1277                         mdt->mdt_md_root = md_root;
1278                         spin_unlock(&mdt->mdt_lock);
1279                 }
1280         }
1281
1282         if (md_root->mot_cache_attr || !mdt_object_remote(md_root))
1283                 return 0;
1284
1285         rc = mdt_remote_object_lock(info, md_root, mdt_object_fid(md_root),
1286                                     &lhroot, LCK_PR, MDS_INODELOCK_XATTR,
1287                                     true);
1288         if (rc < 0)
1289                 return rc;
1290
1291         md_root->mot_cache_attr = 1;
1292
1293         /* don't cancel this lock, so that we know the cached xattr is valid. */
1294         ldlm_lock_decref(&lhroot, LCK_PR);
1295
1296         return 0;
1297 }
1298
1299 static inline enum ldlm_mode mdt_open_lock_mode(struct mdt_thread_info *info,
1300                                                 struct mdt_object *p,
1301                                                 struct lu_name *name,
1302                                                 u64 open_flags)
1303 {
1304         int result;
1305         struct lu_fid fid;
1306
1307         /* We don't need to take the DLM lock for a volatile */
1308         if (open_flags & MDS_OPEN_VOLATILE)
1309                 return LCK_NL;
1310
1311         if (!(open_flags & MDS_OPEN_CREAT))
1312                 return LCK_PR;
1313
1314         result = mdo_lookup(info->mti_env, mdt_object_child(p), name, &fid,
1315                             &info->mti_spec);
1316
1317         /* If the file exists we only need a read lock on the parent */
1318         return (result == 0) ? LCK_PR : LCK_PW;
1319 }
1320
1321 int mdt_reint_open(struct mdt_thread_info *info, struct mdt_lock_handle *lhc)
1322 {
1323         struct mdt_device *mdt = info->mti_mdt;
1324         struct ptlrpc_request *req = mdt_info_req(info);
1325         struct mdt_object *parent;
1326         struct mdt_object *child;
1327         struct mdt_lock_handle *lh = NULL;
1328         struct ldlm_reply *ldlm_rep;
1329         struct mdt_body *repbody;
1330         struct lu_fid *child_fid = &info->mti_tmp_fid1;
1331         struct md_attr *ma = &info->mti_attr;
1332         u64 open_flags = info->mti_spec.sp_cr_flags;
1333         u64 ibits = 0;
1334         struct mdt_reint_record *rr = &info->mti_rr;
1335         int result, rc;
1336         int created = 0;
1337         int object_locked = 0;
1338         enum ldlm_mode lock_mode;
1339         u32 msg_flags;
1340         ktime_t kstart = ktime_get();
1341
1342         ENTRY;
1343         OBD_FAIL_TIMEOUT_ORSET(OBD_FAIL_MDS_PAUSE_OPEN, OBD_FAIL_ONCE,
1344                                (obd_timeout + 1) / 4);
1345
1346         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1347
1348         ma->ma_need = MA_INODE;
1349         ma->ma_valid = 0;
1350
1351         LASSERT(info->mti_pill->rc_fmt == &RQF_LDLM_INTENT_OPEN);
1352         ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
1353
1354         if (unlikely(open_flags & MDS_OPEN_JOIN_FILE)) {
1355                 CERROR("file join is not supported anymore.\n");
1356                 GOTO(out, result = err_serious(-EOPNOTSUPP));
1357         }
1358         msg_flags = lustre_msg_get_flags(req->rq_reqmsg);
1359
1360         if ((open_flags & (MDS_OPEN_HAS_EA | MDS_OPEN_HAS_OBJS)) &&
1361             info->mti_spec.u.sp_ea.eadata == NULL)
1362                 GOTO(out, result = err_serious(-EINVAL));
1363
1364         if (open_flags & MDS_FMODE_WRITE &&
1365             exp_connect_flags(req->rq_export) & OBD_CONNECT_RDONLY)
1366                 GOTO(out, result = -EROFS);
1367
1368         CDEBUG(D_INODE, "I am going to open "DFID"/("DNAME"->"DFID") "
1369                "cr_flag=%#llo mode=0%06o msg_flag=0x%x\n",
1370                PFID(rr->rr_fid1), PNAME(&rr->rr_name), PFID(rr->rr_fid2),
1371                open_flags, ma->ma_attr.la_mode, msg_flags);
1372
1373         if (info->mti_cross_ref) {
1374                 /* This is cross-ref open */
1375                 mdt_set_disposition(info, ldlm_rep,
1376                             (DISP_IT_EXECD | DISP_LOOKUP_EXECD |
1377                              DISP_LOOKUP_POS));
1378                 result = mdt_cross_open(info, rr->rr_fid2, rr->rr_fid1,
1379                                         ldlm_rep, open_flags);
1380                 GOTO(out, result);
1381         } else if (req_is_replay(req)) {
1382                 result = mdt_open_by_fid(info, ldlm_rep, lhc);
1383
1384                 if (result != -ENOENT)
1385                         GOTO(out, result);
1386
1387                 /* We didn't find the correct object, so we need to re-create it
1388                  * via a regular replay. */
1389                 if (!(open_flags & MDS_OPEN_CREAT)) {
1390                         DEBUG_REQ(D_ERROR, req,
1391                                   "OPEN & CREAT not in open replay/by_fid");
1392                         GOTO(out, result = -EFAULT);
1393                 }
1394                 CDEBUG(D_INFO, "No object(1), continue as regular open.\n");
1395         } else if (open_flags & MDS_OPEN_BY_FID) {
1396                 result = mdt_open_by_fid_lock(info, ldlm_rep, lhc);
1397                 if (result < 0)
1398                         CDEBUG(D_INFO, "no object for "DFID": %d\n",
1399                                PFID(rr->rr_fid2), result);
1400                 GOTO(out, result);
1401         }
1402
1403         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_OPEN_PACK))
1404                 GOTO(out, result = err_serious(-ENOMEM));
1405
1406         mdt_set_disposition(info, ldlm_rep,
1407                             (DISP_IT_EXECD | DISP_LOOKUP_EXECD));
1408
1409         if (!lu_name_is_valid(&rr->rr_name))
1410                 GOTO(out, result = -EPROTO);
1411
1412         result = mdt_lock_root_xattr(info, mdt);
1413         if (result < 0)
1414                 GOTO(out, result);
1415
1416         parent = mdt_object_find(info->mti_env, mdt, rr->rr_fid1);
1417         if (IS_ERR(parent))
1418                 GOTO(out, result = PTR_ERR(parent));
1419
1420         /* get and check version of parent */
1421         result = mdt_version_get_check(info, parent, 0);
1422         if (result) {
1423                 mdt_object_put(info->mti_env, parent);
1424                 GOTO(out, result);
1425         }
1426
1427         result = mdt_check_enc(info, parent);
1428         if (result)
1429                 GOTO(out_parent, result);
1430
1431         fid_zero(child_fid);
1432         result = -ENOENT;
1433         lock_mode = mdt_open_lock_mode(info, parent, &rr->rr_name, open_flags);
1434
1435         OBD_RACE(OBD_FAIL_MDS_REINT_OPEN);
1436 again_pw:
1437         if (lock_mode != LCK_NL) {
1438                 lh = &info->mti_lh[MDT_LH_PARENT];
1439                 mdt_lock_pdo_init(lh, lock_mode, &rr->rr_name);
1440                 result = mdt_object_lock(info, parent, lh, MDS_INODELOCK_UPDATE);
1441                 if (result != 0)
1442                         GOTO(out_parent, result);
1443
1444                 result = mdo_lookup(info->mti_env, mdt_object_child(parent),
1445                                     &rr->rr_name, child_fid, &info->mti_spec);
1446         }
1447
1448         LASSERTF(ergo(result == 0, fid_is_sane(child_fid)),
1449                  "looking for "DFID"/"DNAME", found FID = "DFID"\n",
1450                  PFID(mdt_object_fid(parent)), PNAME(&rr->rr_name),
1451                  PFID(child_fid));
1452
1453         if (result != 0 && result != -ENOENT)
1454                 GOTO(out_parent_unlock, result);
1455
1456         OBD_RACE(OBD_FAIL_MDS_REINT_OPEN2);
1457
1458         if (result == -ENOENT) {
1459                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
1460                 if (!(open_flags & MDS_OPEN_CREAT))
1461                         GOTO(out_parent_unlock, result);
1462                 if (mdt_rdonly(req->rq_export))
1463                         GOTO(out_parent_unlock, result = -EROFS);
1464
1465                 LASSERT(equi(lh == NULL, lock_mode == LCK_NL));
1466
1467                 if (lock_mode == LCK_PR) {
1468                         /* unlink vs create race: get write lock and restart */
1469                         mdt_object_unlock(info, parent, lh, 1);
1470                         mdt_clear_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
1471                         mdt_lock_handle_init(lh);
1472                         lock_mode = LCK_PW;
1473                         goto again_pw;
1474                 }
1475
1476                 *child_fid = *info->mti_rr.rr_fid2;
1477                 LASSERTF(fid_is_sane(child_fid), "fid="DFID"\n",
1478                          PFID(child_fid));
1479                 /* In the function below, .hs_keycmp resolves to
1480                  * lu_obj_hop_keycmp() */
1481                 /* coverity[overrun-buffer-val] */
1482                 child = mdt_object_new(info->mti_env, mdt, child_fid);
1483         } else {
1484                 /*
1485                  * Check for O_EXCL is moved to the mdt_finish_open(),
1486                  * we need to return FID back in that case.
1487                  */
1488                 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1489                 child = mdt_object_find(info->mti_env, mdt, child_fid);
1490         }
1491         if (IS_ERR(child))
1492                 GOTO(out_parent_unlock, result = PTR_ERR(child));
1493
1494         /** check version of child  */
1495         rc = mdt_version_get_check(info, child, 1);
1496         if (rc)
1497                 GOTO(out_child, result = rc);
1498
1499         tgt_open_obj_set(info->mti_env, mdt_obj2dt(child));
1500
1501         if (result == -ENOENT) {
1502                 /* Create under OBF and .lustre is not permitted */
1503                 if (!fid_is_md_operative(rr->rr_fid1) &&
1504                     (open_flags & MDS_OPEN_VOLATILE) == 0)
1505                         GOTO(out_child, result = -EPERM);
1506
1507                 /* save versions in reply */
1508                 mdt_version_get_save(info, parent, 0);
1509                 mdt_version_get_save(info, child, 1);
1510
1511                 /* version of child will be changed */
1512                 tgt_vbr_obj_set(info->mti_env, mdt_obj2dt(child));
1513
1514                 /* Not found and with MDS_OPEN_CREAT: let's create it. */
1515                 mdt_set_disposition(info, ldlm_rep, DISP_OPEN_CREATE);
1516
1517                 /* Don't do lookup sanity check. We know name doesn't exist. */
1518                 info->mti_spec.sp_cr_lookup = 0;
1519                 info->mti_spec.sp_feat = &dt_directory_features;
1520
1521                 result = mdo_create(info->mti_env, mdt_object_child(parent),
1522                                     &rr->rr_name, mdt_object_child(child),
1523                                     &info->mti_spec, &info->mti_attr);
1524                 if (result == -ERESTART) {
1525                         mdt_clear_disposition(info, ldlm_rep, DISP_OPEN_CREATE);
1526                         GOTO(out_child, result);
1527                 } else {
1528                         mdt_prep_ma_buf_from_rep(info, child, ma, open_flags);
1529                         /* XXX: we should call this once, see few lines below */
1530                         if (result == 0)
1531                                 result = mdt_attr_get_complex(info, child, ma);
1532
1533                         if (result != 0)
1534                                 GOTO(out_child, result);
1535                 }
1536                 created = 1;
1537                 mdt_counter_incr(req, LPROC_MDT_MKNOD,
1538                                  ktime_us_delta(ktime_get(), kstart));
1539         } else {
1540                 /*
1541                  * The object is on remote node, return its FID for remote open.
1542                  */
1543                 if (mdt_object_remote(child)) {
1544                         /*
1545                          * Check if this lock already was sent to client and
1546                          * this is resent case. For resent case do not take lock
1547                          * again, use what is already granted.
1548                          */
1549                         LASSERT(lhc != NULL);
1550
1551                         rc = mdt_check_resent_lock(info, child, lhc);
1552                         if (rc < 0) {
1553                                 GOTO(out_child, result = rc);
1554                         } else if (rc > 0) {
1555                                 mdt_lock_handle_init(lhc);
1556                                 mdt_lock_reg_init(lhc, LCK_PR);
1557
1558                                 rc = mdt_object_lock(info, child, lhc,
1559                                                      MDS_INODELOCK_LOOKUP);
1560                         }
1561                         repbody->mbo_fid1 = *mdt_object_fid(child);
1562                         repbody->mbo_valid |= (OBD_MD_FLID | OBD_MD_MDS);
1563                         if (rc != 0)
1564                                 result = rc;
1565                         else
1566                                 result = -MDT_EREMOTE_OPEN;
1567                         GOTO(out_child, result);
1568                 } else if (mdt_object_exists(child)) {
1569                         /* Check early for MDS_OPEN_DIRECTORY/O_DIRECTORY to
1570                          * avoid opening regular files from lfs getstripe
1571                          * since doing so breaks the leases used by lfs
1572                          * mirror. See LU-13693. */
1573                         if (open_flags & MDS_OPEN_DIRECTORY &&
1574                             S_ISREG(lu_object_attr(&child->mot_obj)))
1575                                 GOTO(out_child, result = -ENOTDIR);
1576
1577                         /* We have to get attr & LOV EA & HSM for this
1578                          * object. */
1579                         mdt_prep_ma_buf_from_rep(info, child, ma, open_flags);
1580                         ma->ma_need |= MA_HSM;
1581                         result = mdt_attr_get_complex(info, child, ma);
1582                         if (result != 0)
1583                                 GOTO(out_child, result);
1584                 } else {
1585                         /* Object does not exist. Likely FS corruption. */
1586                         CERROR("%s: name '"DNAME"' present, but FID "
1587                                DFID" is invalid\n", mdt_obd_name(info->mti_mdt),
1588                                PNAME(&rr->rr_name), PFID(child_fid));
1589                         GOTO(out_child, result = -EIO);
1590                 }
1591         }
1592
1593         repbody->mbo_max_mdsize = info->mti_mdt->mdt_max_mdsize;
1594         repbody->mbo_valid |= OBD_MD_FLMODEASIZE;
1595
1596         rc = mdt_pack_secctx_in_reply(info, child);
1597         if (unlikely(rc))
1598                 GOTO(out_child, result = rc);
1599
1600         rc = mdt_pack_encctx_in_reply(info, child);
1601         if (unlikely(rc))
1602                 GOTO(out_child, result = rc);
1603
1604         rc = mdt_check_resent_lock(info, child, lhc);
1605         if (rc < 0) {
1606                 GOTO(out_child, result = rc);
1607         } else if (rc == 0) {
1608                 /* the open lock might already be gotten in
1609                  * ldlm_handle_enqueue() */
1610                 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT);
1611                 if (open_flags & MDS_OPEN_LOCK)
1612                         mdt_set_disposition(info, ldlm_rep, DISP_OPEN_LOCK);
1613         } else {
1614                 /* get openlock if this isn't replay and client requested it */
1615                 if (!req_is_replay(req)) {
1616                         rc = mdt_object_open_lock(info, child, lhc, &ibits);
1617                         object_locked = 1;
1618                         if (rc != 0)
1619                                 GOTO(out_child_unlock, result = rc);
1620                         else if (open_flags & MDS_OPEN_LOCK)
1621                                 mdt_set_disposition(info, ldlm_rep,
1622                                                     DISP_OPEN_LOCK);
1623                 }
1624         }
1625         /* Try to open it now. */
1626         rc = mdt_finish_open(info, parent, child, open_flags, ldlm_rep);
1627         if (rc) {
1628                 result = rc;
1629                 /* openlock will be released if mdt_finish_open() failed */
1630                 mdt_clear_disposition(info, ldlm_rep, DISP_OPEN_LOCK);
1631
1632                 if (created && (open_flags & MDS_OPEN_VOLATILE)) {
1633                         CERROR("%s: cannot open volatile file "DFID", orphan "
1634                                "file will be left in PENDING directory until "
1635                                "next reboot, rc = %d\n", mdt_obd_name(mdt),
1636                                PFID(mdt_object_fid(child)), rc);
1637                         GOTO(out_child_unlock, result);
1638                 }
1639
1640                 if (created) {
1641                         ma->ma_need = 0;
1642                         ma->ma_valid = 0;
1643                         rc = mdo_unlink(info->mti_env,
1644                                         mdt_object_child(parent),
1645                                         mdt_object_child(child),
1646                                         &rr->rr_name,
1647                                         &info->mti_attr, 0);
1648                         if (rc != 0)
1649                                 CERROR("%s: "DFID" cleanup of open: rc = %d\n",
1650                                        mdt_obd_name(info->mti_mdt),
1651                                        PFID(mdt_object_fid(child)), rc);
1652                         mdt_clear_disposition(info, ldlm_rep, DISP_OPEN_CREATE);
1653                 }
1654         }
1655
1656         mdt_counter_incr(req, LPROC_MDT_OPEN,
1657                          ktime_us_delta(ktime_get(), kstart));
1658
1659         EXIT;
1660 out_child_unlock:
1661         if (object_locked)
1662                 mdt_object_open_unlock(info, child, lhc, ibits, result);
1663 out_child:
1664         mdt_object_put(info->mti_env, child);
1665         if (result == 0)
1666                 mdt_pack_size2body(info, child_fid, &lhc->mlh_reg_lh);
1667 out_parent_unlock:
1668         if (lh != NULL)
1669                 mdt_object_unlock(info, parent, lh, result || !created);
1670
1671 out_parent:
1672         mdt_object_put(info->mti_env, parent);
1673 out:
1674         if (result)
1675                 lustre_msg_set_transno(req->rq_repmsg, 0);
1676         return result;
1677 }
1678
1679 /**
1680  * Create an orphan object use local root.
1681  */
1682 static struct mdt_object *mdt_orphan_open(struct mdt_thread_info *info,
1683                                           struct mdt_device *mdt,
1684                                           const struct lu_fid *fid,
1685                                           struct md_attr *attr, fmode_t fmode)
1686 {
1687         const struct lu_env *env = info->mti_env;
1688         struct md_op_spec *spec = &info->mti_spec;
1689         struct lu_fid *local_root_fid = &info->mti_tmp_fid1;
1690         struct mdt_object *obj = NULL;
1691         struct mdt_object *local_root;
1692         static const struct lu_name lname = {
1693                 .ln_name = "i_am_nobody",
1694                 .ln_namelen = sizeof("i_am_nobody") - 1,
1695         };
1696         struct lu_ucred *uc;
1697         kernel_cap_t uc_cap_save;
1698         int rc;
1699
1700         ENTRY;
1701         rc = dt_root_get(env, mdt->mdt_bottom, local_root_fid);
1702         if (rc != 0)
1703                 RETURN(ERR_PTR(rc));
1704
1705         local_root = mdt_object_find(env, mdt, local_root_fid);
1706         if (IS_ERR(local_root))
1707                 RETURN(local_root);
1708
1709         obj = mdt_object_new(env, mdt, fid);
1710         if (IS_ERR(obj))
1711                 GOTO(out, rc = PTR_ERR(obj));
1712
1713         spec->sp_cr_lookup = 0;
1714         spec->sp_feat = &dt_directory_features;
1715         spec->sp_cr_flags = MDS_OPEN_VOLATILE | fmode;
1716         if (attr->ma_valid & MA_LOV) {
1717                 spec->u.sp_ea.eadata = attr->ma_lmm;
1718                 spec->u.sp_ea.eadatalen = attr->ma_lmm_size;
1719                 spec->sp_cr_flags |= MDS_OPEN_HAS_EA;
1720         } else {
1721                 spec->sp_cr_flags |= MDS_OPEN_DELAY_CREATE;
1722         }
1723
1724         uc = lu_ucred(env);
1725         uc_cap_save = uc->uc_cap;
1726         cap_raise(uc->uc_cap, CAP_DAC_OVERRIDE);
1727         rc = mdo_create(env, mdt_object_child(local_root), &lname,
1728                         mdt_object_child(obj), spec, attr);
1729         uc->uc_cap = uc_cap_save;
1730         if (rc < 0) {
1731                 CERROR("%s: cannot create volatile file "DFID": rc = %d\n",
1732                        mdt_obd_name(mdt), PFID(fid), rc);
1733                 GOTO(out, rc);
1734         }
1735
1736         rc = mo_open(env, mdt_object_child(obj), MDS_OPEN_CREATED, spec);
1737         if (rc < 0)
1738                 CERROR("%s: cannot open volatile file "DFID", orphan "
1739                        "file will be left in PENDING directory until "
1740                        "next reboot, rc = %d\n", mdt_obd_name(mdt),
1741                        PFID(fid), rc);
1742         GOTO(out, rc);
1743
1744 out:
1745         if (rc < 0) {
1746                 if (!IS_ERR(obj))
1747                         mdt_object_put(env, obj);
1748                 obj = ERR_PTR(rc);
1749         }
1750         mdt_object_put(env, local_root);
1751         return obj;
1752 }
1753
1754 /* XXX Look into layout in MDT layer. */
1755 static inline int mdt_hsm_set_released(struct lov_mds_md *lmm)
1756 {
1757         struct lov_comp_md_v1 *comp_v1;
1758         struct lov_mds_md *v1;
1759         __u32 off;
1760         int i;
1761
1762         if (lmm->lmm_magic == cpu_to_le32(LOV_MAGIC_COMP_V1_DEFINED)) {
1763                 comp_v1 = (struct lov_comp_md_v1 *)lmm;
1764
1765                 if (comp_v1->lcm_entry_count == 0)
1766                         return -EINVAL;
1767
1768                 for (i = 0; i < le16_to_cpu(comp_v1->lcm_entry_count); i++) {
1769                         off = le32_to_cpu(comp_v1->lcm_entries[i].lcme_offset);
1770                         v1 = (struct lov_mds_md *)((char *)comp_v1 + off);
1771                         v1->lmm_pattern |= cpu_to_le32(LOV_PATTERN_F_RELEASED);
1772                 }
1773         } else {
1774                 lmm->lmm_pattern |= cpu_to_le32(LOV_PATTERN_F_RELEASED);
1775         }
1776         return 0;
1777 }
1778
1779 static inline int mdt_get_lmm_gen(struct lov_mds_md *lmm, __u32 *gen)
1780 {
1781         struct lov_comp_md_v1 *comp_v1;
1782
1783         if (le32_to_cpu(lmm->lmm_magic == LOV_MAGIC_COMP_V1)) {
1784                 comp_v1 = (struct lov_comp_md_v1 *)lmm;
1785                 *gen = le32_to_cpu(comp_v1->lcm_layout_gen);
1786         } else if (le32_to_cpu(lmm->lmm_magic) == LOV_MAGIC_V1 ||
1787                    le32_to_cpu(lmm->lmm_magic) == LOV_MAGIC_V3) {
1788                 *gen = le16_to_cpu(lmm->lmm_layout_gen);
1789         } else {
1790                 return -EINVAL;
1791         }
1792         return 0;
1793 }
1794
1795 static int mdt_hsm_release(struct mdt_thread_info *info, struct mdt_object *o,
1796                            struct md_attr *ma)
1797 {
1798         struct mdt_lock_handle *lh = &info->mti_lh[MDT_LH_LAYOUT];
1799         struct lu_ucred *uc = mdt_ucred(info);
1800         struct close_data *data;
1801         struct ldlm_lock *lease;
1802         struct mdt_object *orphan;
1803         struct md_attr *orp_ma;
1804         struct lu_buf *buf;
1805         kernel_cap_t cap;
1806         bool lease_broken;
1807         int rc;
1808         int rc2;
1809
1810         ENTRY;
1811         if (mdt_rdonly(info->mti_exp))
1812                 RETURN(-EROFS);
1813
1814         data = req_capsule_client_get(info->mti_pill, &RMF_CLOSE_DATA);
1815         if (data == NULL)
1816                 RETURN(-EPROTO);
1817
1818         lease = ldlm_handle2lock(&data->cd_handle);
1819         if (lease == NULL)
1820                 RETURN(-ESTALE);
1821
1822         /* try to hold open_sem so that nobody else can open the file */
1823         if (!down_write_trylock(&o->mot_open_sem)) {
1824                 ldlm_lock_cancel(lease);
1825                 GOTO(out_reprocess, rc = -EBUSY);
1826         }
1827
1828         /* Check if the lease open lease has already canceled */
1829         lock_res_and_lock(lease);
1830         lease_broken = ldlm_is_cancel(lease);
1831         unlock_res_and_lock(lease);
1832
1833         LDLM_DEBUG(lease, DFID " lease broken? %d",
1834                    PFID(mdt_object_fid(o)), lease_broken);
1835
1836         /* Cancel server side lease. Client side counterpart should
1837          * have been cancelled. It's okay to cancel it now as we've
1838          * held mot_open_sem. */
1839         ldlm_lock_cancel(lease);
1840
1841         if (lease_broken) /* don't perform release task */
1842                 GOTO(out_unlock, rc = -ESTALE);
1843
1844         if (fid_is_zero(&data->cd_fid) || !fid_is_sane(&data->cd_fid))
1845                 GOTO(out_unlock, rc = -EINVAL);
1846
1847         /* ma_need was set before but it seems fine to change it in order to
1848          * avoid modifying the one from RPC */
1849         ma->ma_need = MA_HSM;
1850         rc = mdt_attr_get_complex(info, o, ma);
1851         if (rc != 0)
1852                 GOTO(out_unlock, rc);
1853
1854         if (ma->ma_attr_flags & MDS_PCC_ATTACH) {
1855                 if (ma->ma_valid & MA_HSM) {
1856                         if (ma->ma_hsm.mh_flags & HS_RELEASED)
1857                                 GOTO(out_unlock, rc = -EALREADY);
1858
1859                         if (ma->ma_hsm.mh_arch_id != data->cd_archive_id)
1860                                 CDEBUG(D_CACHE,
1861                                        DFID" archive id diff: %llu:%u\n",
1862                                        PFID(mdt_object_fid(o)),
1863                                        ma->ma_hsm.mh_arch_id,
1864                                        data->cd_archive_id);
1865
1866                         if (!(ma->ma_hsm.mh_flags & HS_DIRTY) &&
1867                             ma->ma_hsm.mh_arch_ver == data->cd_data_version) {
1868                                 CDEBUG(D_CACHE,
1869                                        DFID" data version matches: packed=%llu "
1870                                        "and on-disk=%llu\n",
1871                                        PFID(mdt_object_fid(o)),
1872                                        data->cd_data_version,
1873                                        ma->ma_hsm.mh_arch_ver);
1874                                 ma->ma_hsm.mh_flags = HS_ARCHIVED | HS_EXISTS;
1875                         }
1876
1877                         if (ma->ma_hsm.mh_flags & HS_DIRTY)
1878                                 ma->ma_hsm.mh_flags = HS_ARCHIVED | HS_EXISTS;
1879                 } else {
1880                         /* Set up HSM attribte for PCC archived object */
1881                         BUILD_BUG_ON(sizeof(struct hsm_attrs) >
1882                                      sizeof(info->mti_xattr_buf));
1883                         buf = &info->mti_buf;
1884                         buf->lb_buf = info->mti_xattr_buf;
1885                         buf->lb_len = sizeof(struct hsm_attrs);
1886                         memset(&ma->ma_hsm, 0, sizeof(ma->ma_hsm));
1887                         ma->ma_hsm.mh_flags = HS_ARCHIVED | HS_EXISTS;
1888                         ma->ma_hsm.mh_arch_id = data->cd_archive_id;
1889                         ma->ma_hsm.mh_arch_ver = data->cd_data_version;
1890                         lustre_hsm2buf(buf->lb_buf, &ma->ma_hsm);
1891
1892                         rc = mo_xattr_set(info->mti_env, mdt_object_child(o),
1893                                           buf, XATTR_NAME_HSM, 0);
1894                         if (rc)
1895                                 GOTO(out_unlock, rc);
1896                 }
1897         } else {
1898                 if (!mdt_hsm_release_allow(ma))
1899                         GOTO(out_unlock, rc = -EPERM);
1900
1901                 /* already released? */
1902                 if (ma->ma_hsm.mh_flags & HS_RELEASED)
1903                         GOTO(out_unlock, rc = 0);
1904
1905                 /* Compare on-disk and packed data_version */
1906                 if (data->cd_data_version != ma->ma_hsm.mh_arch_ver) {
1907                         CDEBUG(D_HSM, DFID" data_version mismatches: "
1908                                "packed=%llu and on-disk=%llu\n",
1909                                PFID(mdt_object_fid(o)),
1910                                data->cd_data_version,
1911                                ma->ma_hsm.mh_arch_ver);
1912                         GOTO(out_unlock, rc = -EPERM);
1913                 }
1914         }
1915
1916         ma->ma_valid = MA_INODE;
1917         ma->ma_attr.la_valid &= LA_ATIME | LA_MTIME | LA_CTIME | LA_SIZE;
1918         rc = mo_attr_set(info->mti_env, mdt_object_child(o), ma);
1919         if (rc < 0)
1920                 GOTO(out_unlock, rc);
1921
1922         mutex_lock(&o->mot_som_mutex);
1923         rc2 = mdt_set_som(info, o, SOM_FL_STRICT, ma->ma_attr.la_size,
1924                            ma->ma_attr.la_blocks);
1925         mutex_unlock(&o->mot_som_mutex);
1926         if (rc2 < 0)
1927                 CDEBUG(D_INODE,
1928                        "%s: File "DFID" SOM update failed: rc = %d\n",
1929                        mdt_obd_name(info->mti_mdt),
1930                        PFID(mdt_object_fid(o)), rc2);
1931
1932
1933         ma->ma_need = MA_INODE | MA_LOV;
1934         rc = mdt_attr_get_complex(info, o, ma);
1935         if (rc < 0)
1936                 GOTO(out_unlock, rc);
1937
1938         if (!(ma->ma_valid & MA_LOV)) {
1939                 /* Even empty file are released */
1940                 memset(ma->ma_lmm, 0, sizeof(*ma->ma_lmm));
1941                 ma->ma_lmm->lmm_magic = cpu_to_le32(LOV_MAGIC_V1_DEFINED);
1942                 ma->ma_lmm->lmm_pattern = cpu_to_le32(LOV_PATTERN_RAID0);
1943                 ma->ma_lmm->lmm_stripe_size = cpu_to_le32(LOV_MIN_STRIPE_SIZE);
1944                 ma->ma_lmm_size = sizeof(*ma->ma_lmm);
1945         } else {
1946                 /* Magic must be LOV_MAGIC_*_DEFINED or LOD will interpret
1947                  * ma_lmm as lov_user_md, then it will be confused by union of
1948                  * layout_gen and stripe_offset. */
1949                 if ((le32_to_cpu(ma->ma_lmm->lmm_magic) & LOV_MAGIC_MASK) ==
1950                     LOV_MAGIC_MAGIC)
1951                         ma->ma_lmm->lmm_magic |= cpu_to_le32(LOV_MAGIC_DEFINED);
1952                 else
1953                         GOTO(out_unlock, rc = -EINVAL);
1954         }
1955
1956         /* Set file as released. */
1957         rc = mdt_hsm_set_released(ma->ma_lmm);
1958         if (rc)
1959                 GOTO(out_unlock, rc);
1960
1961         orp_ma = &info->mti_u.hsm.attr;
1962         orp_ma->ma_attr.la_mode = S_IFREG | S_IWUSR;
1963         /* We use root ownership to bypass potential quota
1964          * restrictions on the user and group of the file. */
1965         orp_ma->ma_attr.la_uid = 0;
1966         orp_ma->ma_attr.la_gid = 0;
1967         orp_ma->ma_attr.la_valid = LA_MODE | LA_UID | LA_GID;
1968         orp_ma->ma_lmm = ma->ma_lmm;
1969         orp_ma->ma_lmm_size = ma->ma_lmm_size;
1970         orp_ma->ma_valid = MA_INODE | MA_LOV;
1971         orphan = mdt_orphan_open(info, info->mti_mdt, &data->cd_fid, orp_ma,
1972                                  MDS_FMODE_WRITE);
1973         if (IS_ERR(orphan)) {
1974                 CERROR("%s: cannot open orphan file "DFID": rc = %ld\n",
1975                        mdt_obd_name(info->mti_mdt), PFID(&data->cd_fid),
1976                        PTR_ERR(orphan));
1977                 GOTO(out_unlock, rc = PTR_ERR(orphan));
1978         }
1979
1980         /* Set up HSM attribute for orphan object */
1981         BUILD_BUG_ON(sizeof(struct hsm_attrs) > sizeof(info->mti_xattr_buf));
1982         buf = &info->mti_buf;
1983         buf->lb_buf = info->mti_xattr_buf;
1984         buf->lb_len = sizeof(struct hsm_attrs);
1985         ma->ma_hsm.mh_flags |= HS_RELEASED;
1986         lustre_hsm2buf(buf->lb_buf, &ma->ma_hsm);
1987         ma->ma_hsm.mh_flags &= ~HS_RELEASED;
1988
1989         mdt_lock_reg_init(lh, LCK_EX);
1990         rc = mdt_object_lock(info, o, lh, MDS_INODELOCK_LAYOUT |
1991                              MDS_INODELOCK_XATTR);
1992         if (rc != 0)
1993                 GOTO(out_close, rc);
1994
1995         /* The orphan has root ownership so we need to raise
1996          * CAP_FOWNER to set the HSM attributes. */
1997         cap = uc->uc_cap;
1998         cap_raise(uc->uc_cap, CAP_FOWNER);
1999         rc = mo_xattr_set(info->mti_env, mdt_object_child(orphan), buf,
2000                           XATTR_NAME_HSM, 0);
2001         uc->uc_cap = cap;
2002         if (rc != 0)
2003                 GOTO(out_layout_lock, rc);
2004
2005         /* Swap layout with orphan objects. */
2006         rc = mo_swap_layouts(info->mti_env, mdt_object_child(o),
2007                              mdt_object_child(orphan),
2008                              SWAP_LAYOUTS_MDS_HSM);
2009
2010         if (!rc && ma->ma_attr_flags & MDS_PCC_ATTACH) {
2011                 ma->ma_need = MA_LOV;
2012                 rc = mdt_attr_get_complex(info, o, ma);
2013         }
2014
2015         EXIT;
2016
2017 out_layout_lock:
2018         /* Release exclusive LL */
2019         mdt_object_unlock(info, o, lh, 1);
2020 out_close:
2021         /* Close orphan object anyway */
2022         rc2 = mo_close(info->mti_env, mdt_object_child(orphan), orp_ma,
2023                        MDS_FMODE_WRITE);
2024         if (rc2 < 0)
2025                 CERROR("%s: error closing volatile file "DFID": rc = %d\n",
2026                        mdt_obd_name(info->mti_mdt), PFID(&data->cd_fid), rc2);
2027         LU_OBJECT_DEBUG(D_HSM, info->mti_env, &orphan->mot_obj,
2028                         "object closed");
2029         mdt_object_put(info->mti_env, orphan);
2030
2031 out_unlock:
2032         up_write(&o->mot_open_sem);
2033
2034         /* already released */
2035         if (rc == 0) {
2036                 struct mdt_body *repbody;
2037
2038                 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
2039                 LASSERT(repbody != NULL);
2040                 repbody->mbo_valid |= OBD_MD_CLOSE_INTENT_EXECED;
2041                 if (ma->ma_attr_flags & MDS_PCC_ATTACH) {
2042                         LASSERT(ma->ma_valid & MA_LOV);
2043                         rc = mdt_get_lmm_gen(ma->ma_lmm,
2044                                              &repbody->mbo_layout_gen);
2045                         if (!rc)
2046                                 repbody->mbo_valid |= OBD_MD_LAYOUT_VERSION;
2047                 }
2048         }
2049
2050 out_reprocess:
2051         ldlm_reprocess_all(lease->l_resource,
2052                            lease->l_policy_data.l_inodebits.bits);
2053         LDLM_LOCK_PUT(lease);
2054
2055         ma->ma_valid = 0;
2056         ma->ma_need = 0;
2057
2058         return rc;
2059 }
2060
2061 int mdt_close_handle_layouts(struct mdt_thread_info *info,
2062                              struct mdt_object *o, struct md_attr *ma)
2063 {
2064         struct mdt_lock_handle *lh1 = &info->mti_lh[MDT_LH_NEW];
2065         struct mdt_lock_handle *lh2 = &info->mti_lh[MDT_LH_OLD];
2066         struct close_data *data;
2067         struct ldlm_lock *lease;
2068         struct mdt_object *o1 = o, *o2 = NULL;
2069         bool lease_broken;
2070         bool swap_objects = false;
2071         int rc;
2072
2073         ENTRY;
2074         if (exp_connect_flags(info->mti_exp) & OBD_CONNECT_RDONLY)
2075                 RETURN(-EROFS);
2076
2077         if (!S_ISREG(lu_object_attr(&o1->mot_obj)))
2078                 RETURN(-EINVAL);
2079
2080         data = req_capsule_client_get(info->mti_pill, &RMF_CLOSE_DATA);
2081         if (data == NULL)
2082                 RETURN(-EPROTO);
2083
2084         if (fid_is_zero(&data->cd_fid) || !fid_is_sane(&data->cd_fid))
2085                 RETURN(-EINVAL);
2086
2087         rc = lu_fid_cmp(&data->cd_fid, mdt_object_fid(o));
2088         if (rc == 0) {
2089                 /**
2090                  * only MDS_CLOSE_LAYOUT_SPLIT use the same fid to indicate
2091                  * mirror deletion, so we'd zero cd_fid, and keeps o2 be NULL.
2092                  */
2093                 if (!(ma->ma_attr_flags & MDS_CLOSE_LAYOUT_SPLIT))
2094                         RETURN(-EINVAL);
2095
2096                 /* zero cd_fid to keeps o2 be NULL */
2097                 fid_zero(&data->cd_fid);
2098         } else if (rc < 0) {
2099                 /* Exchange o1 and o2, to enforce locking order */
2100                 swap_objects = true;
2101         }
2102
2103         lease = ldlm_handle2lock(&data->cd_handle);
2104         if (lease == NULL)
2105                 RETURN(-ESTALE);
2106
2107         if (!fid_is_zero(&data->cd_fid)) {
2108                 o2 = mdt_object_find(info->mti_env, info->mti_mdt,
2109                                      &data->cd_fid);
2110                 if (IS_ERR(o2))
2111                         GOTO(out_lease, rc = PTR_ERR(o2));
2112
2113                 if (!mdt_object_exists(o2))
2114                         GOTO(out_obj, rc = -ENOENT);
2115
2116                 if (!S_ISREG(lu_object_attr(&o2->mot_obj)))
2117                         GOTO(out_obj, rc = -EINVAL);
2118
2119                 if (swap_objects)
2120                         swap(o1, o2);
2121         }
2122
2123         rc = mo_permission(info->mti_env, NULL, mdt_object_child(o1), NULL,
2124                            MAY_WRITE);
2125         if (rc < 0)
2126                 GOTO(out_obj, rc);
2127
2128         if (o2) {
2129                 rc = mo_permission(info->mti_env, NULL, mdt_object_child(o2),
2130                                    NULL, MAY_WRITE);
2131                 if (rc < 0)
2132                         GOTO(out_obj, rc);
2133         }
2134
2135         /* try to hold open_sem so that nobody else can open the file */
2136         if (!down_write_trylock(&o->mot_open_sem)) {
2137                 ldlm_lock_cancel(lease);
2138                 GOTO(out_obj, rc = -EBUSY);
2139         }
2140
2141         /* Check if the lease open lease has already canceled */
2142         lock_res_and_lock(lease);
2143         lease_broken = ldlm_is_cancel(lease);
2144         unlock_res_and_lock(lease);
2145
2146         LDLM_DEBUG(lease, DFID " lease broken? %d",
2147                    PFID(mdt_object_fid(o)), lease_broken);
2148
2149         /* Cancel server side lease. Client side counterpart should
2150          * have been cancelled. It's okay to cancel it now as we've
2151          * held mot_open_sem. */
2152         ldlm_lock_cancel(lease);
2153
2154         if (lease_broken)
2155                 GOTO(out_unlock_sem, rc = -ESTALE);
2156
2157         mdt_lock_reg_init(lh1, LCK_EX);
2158         rc = mdt_object_lock(info, o1, lh1, MDS_INODELOCK_LAYOUT |
2159                              MDS_INODELOCK_XATTR);
2160         if (rc < 0)
2161                 GOTO(out_unlock_sem, rc);
2162
2163         if (o2) {
2164                 mdt_lock_reg_init(lh2, LCK_EX);
2165                 rc = mdt_object_lock(info, o2, lh2, MDS_INODELOCK_LAYOUT |
2166                                      MDS_INODELOCK_XATTR);
2167                 if (rc < 0)
2168                         GOTO(out_unlock1, rc);
2169         }
2170
2171         /* Swap layout with orphan object */
2172         if (ma->ma_attr_flags & MDS_CLOSE_LAYOUT_SWAP) {
2173                 rc = mo_swap_layouts(info->mti_env, mdt_object_child(o1),
2174                                      mdt_object_child(o2), 0);
2175         } else if (ma->ma_attr_flags & MDS_CLOSE_LAYOUT_MERGE ||
2176                    ma->ma_attr_flags & MDS_CLOSE_LAYOUT_SPLIT) {
2177                 struct lu_buf *buf = &info->mti_buf;
2178                 struct md_rejig_data mrd;
2179
2180                 if (o2) {
2181                         mrd.mrd_obj = mdt_object_child(o == o1 ? o2 : o1);
2182                 } else {
2183                         if (!(ma->ma_attr_flags & MDS_CLOSE_LAYOUT_SPLIT)) {
2184                                 /* paranoid check again */
2185                                 CERROR(DFID
2186                                   ":only mirror split support NULL o2 object\n",
2187                                         PFID(mdt_object_fid(o)));
2188                                 GOTO(out_unlock1, rc = -EINVAL);
2189                         }
2190
2191                         /* set NULL mrd_obj for deleting mirror objects */
2192                         mrd.mrd_obj = NULL;
2193                 }
2194
2195                 if (ma->ma_attr_flags & MDS_CLOSE_LAYOUT_SPLIT) {
2196                         mrd.mrd_mirror_id = data->cd_mirror_id;
2197                         /* set a small enough blocks in the SoM */
2198                         ma->ma_attr.la_blocks >>= 1;
2199                 }
2200
2201                 buf->lb_len = sizeof(mrd);
2202                 buf->lb_buf = &mrd;
2203                 rc = mo_xattr_set(info->mti_env, mdt_object_child(o), buf,
2204                                   XATTR_LUSTRE_LOV,
2205                                   ma->ma_attr_flags & MDS_CLOSE_LAYOUT_SPLIT ?
2206                                   LU_XATTR_SPLIT : LU_XATTR_MERGE);
2207                 if (rc == 0 && ma->ma_attr.la_valid & (LA_SIZE | LA_BLOCKS |
2208                                                        LA_LSIZE | LA_LBLOCKS)) {
2209                         int rc2;
2210                         enum lustre_som_flags lsf;
2211
2212                         if (ma->ma_attr.la_valid & (LA_SIZE | LA_BLOCKS))
2213                                 lsf = SOM_FL_STRICT;
2214                         else
2215                                 lsf = SOM_FL_LAZY;
2216
2217                         mutex_lock(&o->mot_som_mutex);
2218                         rc2 = mdt_set_som(info, o, lsf,
2219                                           ma->ma_attr.la_size,
2220                                           ma->ma_attr.la_blocks);
2221                         mutex_unlock(&o->mot_som_mutex);
2222                         if (rc2 < 0)
2223                                 CERROR(DFID": Setting i_blocks error: %d, "
2224                                        "i_blocks will be reported wrongly and "
2225                                        "can only be fixed in next resync\n",
2226                                        PFID(mdt_object_fid(o)), rc2);
2227                 }
2228         }
2229         if (rc < 0)
2230                 GOTO(out_unlock2, rc);
2231
2232         EXIT;
2233
2234 out_unlock2:
2235         /* Release exclusive LL */
2236         if (o2)
2237                 mdt_object_unlock(info, o2, lh2, 1);
2238
2239 out_unlock1:
2240         mdt_object_unlock(info, o1, lh1, 1);
2241
2242 out_unlock_sem:
2243         up_write(&o->mot_open_sem);
2244
2245         /* already swapped */
2246         if (rc == 0) {
2247                 struct mdt_body *repbody;
2248
2249                 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
2250                 LASSERT(repbody != NULL);
2251                 repbody->mbo_valid |= OBD_MD_CLOSE_INTENT_EXECED;
2252         }
2253
2254 out_obj:
2255         if (o1 != o)
2256                 /* the 2nd object has been used, and swapped to o1 */
2257                 mdt_object_put(info->mti_env, o1);
2258         else if (o2)
2259                 /* the 2nd object has been used, and not swapped */
2260                 mdt_object_put(info->mti_env, o2);
2261
2262         ldlm_reprocess_all(lease->l_resource,
2263                            lease->l_policy_data.l_inodebits.bits);
2264
2265 out_lease:
2266         LDLM_LOCK_PUT(lease);
2267
2268         if (ma != NULL) {
2269                 ma->ma_valid = 0;
2270                 ma->ma_need = 0;
2271         }
2272
2273         return rc;
2274 }
2275
2276 static int mdt_close_resync_done(struct mdt_thread_info *info,
2277                                  struct mdt_object *o, struct md_attr *ma)
2278 {
2279         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_LOCAL];
2280         struct close_data *data;
2281         struct ldlm_lock *lease;
2282         struct md_layout_change layout = { 0 };
2283         __u32 *resync_ids = NULL;
2284         size_t resync_count = 0;
2285         bool lease_broken;
2286         int rc;
2287
2288         ENTRY;
2289         if (exp_connect_flags(info->mti_exp) & OBD_CONNECT_RDONLY)
2290                 RETURN(-EROFS);
2291
2292         if (!S_ISREG(lu_object_attr(&o->mot_obj)))
2293                 RETURN(-EINVAL);
2294
2295         data = req_capsule_client_get(info->mti_pill, &RMF_CLOSE_DATA);
2296         if (data == NULL)
2297                 RETURN(-EPROTO);
2298
2299         if (req_capsule_req_need_swab(info->mti_pill))
2300                 lustre_swab_close_data_resync_done(&data->cd_resync);
2301
2302         if (!fid_is_zero(&data->cd_fid))
2303                 RETURN(-EPROTO);
2304
2305         lease = ldlm_handle2lock(&data->cd_handle);
2306         if (lease == NULL)
2307                 RETURN(-ESTALE);
2308
2309         /* try to hold open_sem so that nobody else can open the file */
2310         if (!down_write_trylock(&o->mot_open_sem)) {
2311                 ldlm_lock_cancel(lease);
2312                 GOTO(out_reprocess, rc = -EBUSY);
2313         }
2314
2315         /* Check if the lease open lease has already canceled */
2316         lock_res_and_lock(lease);
2317         lease_broken = ldlm_is_cancel(lease);
2318         unlock_res_and_lock(lease);
2319
2320         LDLM_DEBUG(lease, DFID " lease broken? %d\n",
2321                    PFID(mdt_object_fid(o)), lease_broken);
2322
2323         /* Cancel server side lease. Client side counterpart should
2324          * have been cancelled. It's okay to cancel it now as we've
2325          * held mot_open_sem. */
2326         ldlm_lock_cancel(lease);
2327
2328         if (lease_broken) /* don't perform release task */
2329                 GOTO(out_unlock, rc = -ESTALE);
2330
2331         resync_count = data->cd_resync.resync_count;
2332
2333         if (resync_count > INLINE_RESYNC_ARRAY_SIZE) {
2334                 void *data;
2335
2336                 if (!req_capsule_has_field(info->mti_pill, &RMF_U32,
2337                                            RCL_CLIENT))
2338                         GOTO(out_unlock, rc = -EPROTO);
2339
2340                 OBD_ALLOC_PTR_ARRAY(resync_ids, resync_count);
2341                 if (!resync_ids)
2342                         GOTO(out_unlock, rc = -ENOMEM);
2343
2344                 data = req_capsule_client_get(info->mti_pill, &RMF_U32);
2345                 memcpy(resync_ids, data, resync_count * sizeof(__u32));
2346
2347                 layout.mlc_resync_ids = resync_ids;
2348         } else {
2349                 layout.mlc_resync_ids = data->cd_resync.resync_ids_inline;
2350         }
2351
2352         layout.mlc_opc = MD_LAYOUT_RESYNC_DONE;
2353         layout.mlc_resync_count = resync_count;
2354         if (ma->ma_attr.la_valid & (LA_SIZE | LA_BLOCKS)) {
2355                 layout.mlc_som.lsa_valid = SOM_FL_STRICT;
2356                 layout.mlc_som.lsa_size = ma->ma_attr.la_size;
2357                 layout.mlc_som.lsa_blocks = ma->ma_attr.la_blocks;
2358         }
2359         rc = mdt_layout_change(info, o, lhc, &layout);
2360         if (rc)
2361                 GOTO(out_unlock, rc);
2362
2363         mdt_object_unlock(info, o, lhc, 0);
2364
2365         EXIT;
2366
2367 out_unlock:
2368         up_write(&o->mot_open_sem);
2369
2370         /* already released */
2371         if (rc == 0) {
2372                 struct mdt_body *repbody;
2373
2374                 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
2375                 LASSERT(repbody != NULL);
2376                 repbody->mbo_valid |= OBD_MD_CLOSE_INTENT_EXECED;
2377         }
2378
2379         if (resync_ids)
2380                 OBD_FREE_PTR_ARRAY(resync_ids, resync_count);
2381
2382 out_reprocess:
2383         ldlm_reprocess_all(lease->l_resource,
2384                            lease->l_policy_data.l_inodebits.bits);
2385         LDLM_LOCK_PUT(lease);
2386
2387         ma->ma_valid = 0;
2388         ma->ma_need = 0;
2389
2390         return rc;
2391 }
2392
2393 #define MFD_CLOSED(open_flags) ((open_flags) == MDS_FMODE_CLOSED)
2394
2395 static int mdt_mfd_closed(struct mdt_file_data *mfd)
2396 {
2397         return ((mfd == NULL) || MFD_CLOSED(mfd->mfd_open_flags));
2398 }
2399
2400 int mdt_mfd_close(struct mdt_thread_info *info, struct mdt_file_data *mfd)
2401 {
2402         struct mdt_object *o = mfd->mfd_object;
2403         struct md_object *next = mdt_object_child(o);
2404         struct md_attr *ma = &info->mti_attr;
2405         struct lu_fid *ofid = &info->mti_tmp_fid1;
2406         int rc = 0;
2407         int rc2;
2408         u64 open_flags;
2409         u64 intent;
2410
2411         ENTRY;
2412         open_flags = mfd->mfd_open_flags;
2413         intent = ma->ma_attr_flags & MDS_CLOSE_INTENT;
2414         *ofid = *mdt_object_fid(o);
2415
2416         /* the below message is checked in replay-single.sh test_46 */
2417         CDEBUG(D_INODE, "%s: %sclosing file handle "DFID" with intent: %llx\n",
2418                mdt_obd_name(info->mti_mdt),
2419                ma->ma_valid & MA_FORCE_LOG ? "force " : "", PFID(ofid), intent);
2420
2421         switch (intent) {
2422         case MDS_HSM_RELEASE: {
2423                 rc = mdt_hsm_release(info, o, ma);
2424                 if (rc < 0) {
2425                         CDEBUG(D_HSM, "%s: File " DFID " release failed: %d\n",
2426                                mdt_obd_name(info->mti_mdt),
2427                                PFID(ofid), rc);
2428                         /* continue to close even error occurred. */
2429                 }
2430                 break;
2431         }
2432         case MDS_CLOSE_LAYOUT_MERGE:
2433         case MDS_CLOSE_LAYOUT_SPLIT:
2434         case MDS_CLOSE_LAYOUT_SWAP: {
2435                 rc = mdt_close_handle_layouts(info, o, ma);
2436                 if (rc < 0) {
2437                         CDEBUG(D_INODE,
2438                                "%s: cannot %s layout of "DFID": rc = %d\n",
2439                                mdt_obd_name(info->mti_mdt),
2440                                intent == MDS_CLOSE_LAYOUT_MERGE ? "merge" :
2441                                intent == MDS_CLOSE_LAYOUT_SPLIT ? "split" :
2442                                "swap",
2443                                PFID(ofid), rc);
2444                         /* continue to close even if error occurred. */
2445                 }
2446                 break;
2447         }
2448         case MDS_CLOSE_RESYNC_DONE:
2449                 rc = mdt_close_resync_done(info, o, ma);
2450                 if (rc < 0) {
2451                         CDEBUG(D_INODE,
2452                                "%s: cannot resync layout of "DFID": rc = %d\n",
2453                                mdt_obd_name(info->mti_mdt),
2454                                PFID(ofid), rc);
2455                         /* continue to close even if error occurred. */
2456                 }
2457                 break;
2458         default:
2459                 /* nothing */
2460                 break;
2461         }
2462
2463         if (S_ISREG(lu_object_attr(&o->mot_obj)) &&
2464             ma->ma_attr.la_valid & (LA_LSIZE | LA_LBLOCKS)) {
2465                 rc2 = mdt_lsom_update(info, o, false);
2466                 if (rc2 < 0) {
2467                         CDEBUG(D_INODE,
2468                                "%s: File " DFID " LSOM failed: rc = %d\n",
2469                                mdt_obd_name(info->mti_mdt),
2470                                PFID(ofid), rc2);
2471                         if (rc == 0)
2472                                 rc = rc2;
2473                         /* continue to close even if error occurred. */
2474                 }
2475         }
2476
2477         if (open_flags & MDS_FMODE_WRITE)
2478                 mdt_write_put(o);
2479         else if (open_flags & MDS_FMODE_EXEC)
2480                 mdt_write_allow(o);
2481
2482         /* Update atime|mtime|ctime on close. */
2483         if ((open_flags & MDS_FMODE_EXEC || open_flags & MDS_FMODE_READ ||
2484              open_flags & MDS_FMODE_WRITE) && (ma->ma_valid & MA_INODE) &&
2485             (ma->ma_attr.la_valid & LA_ATIME ||
2486              ma->ma_attr.la_valid & LA_MTIME ||
2487              ma->ma_attr.la_valid & LA_CTIME)) {
2488                 ma->ma_valid = MA_INODE;
2489                 ma->ma_attr_flags |= MDS_CLOSE_UPDATE_TIMES;
2490                 ma->ma_attr.la_valid &= (LA_ATIME | LA_MTIME | LA_CTIME);
2491
2492                 if (ma->ma_attr.la_valid & LA_MTIME) {
2493                         if (mdt_attr_get_pfid(info, o, &ma->ma_pfid) == 0)
2494                                 ma->ma_valid |= MA_PFID;
2495                 }
2496
2497                 rc2 = mo_attr_set(info->mti_env, next, ma);
2498                 if (rc2 != 0) {
2499                         CDEBUG(D_INODE,
2500                            "%s: File "DFID" set attr (%#llx) failed: rc = %d\n",
2501                                mdt_obd_name(info->mti_mdt), PFID(ofid),
2502                                ma->ma_attr.la_valid, rc2);
2503                         if (rc == 0)
2504                                 rc = rc2;
2505                 }
2506         }
2507
2508         /* If file data is modified, add the dirty flag. */
2509         if (ma->ma_attr_flags & MDS_DATA_MODIFIED) {
2510                 rc2 = mdt_add_dirty_flag(info, o, ma);
2511                 if (rc2 != 0) {
2512                         CDEBUG(D_INODE,
2513                              "%s: File "DFID" add dirty flag failed: rc = %d\n",
2514                                mdt_obd_name(info->mti_mdt), PFID(ofid), rc2);
2515                         if (rc == 0)
2516                                 rc = rc2;
2517                 }
2518         }
2519
2520         ma->ma_need |= MA_INODE;
2521         ma->ma_valid &= ~MA_INODE;
2522
2523         LASSERT(atomic_read(&o->mot_open_count) > 0);
2524         atomic_dec(&o->mot_open_count);
2525         mdt_handle_last_unlink(info, o, ma);
2526
2527         if (!MFD_CLOSED(open_flags)) {
2528                 rc2 = mo_close(info->mti_env, next, ma, open_flags);
2529                 if (rc2 != 0) {
2530                         CDEBUG(D_INODE,
2531                                "%s: File "DFID" close failed: rc = %d\n",
2532                                mdt_obd_name(info->mti_mdt), PFID(ofid), rc2);
2533                         if (rc == 0)
2534                                 rc = rc2;
2535                 }
2536                 if (mdt_dom_check_for_discard(info, o))
2537                         mdt_dom_discard_data(info, o);
2538         }
2539
2540         /* adjust open and lease count */
2541         if (open_flags & MDS_OPEN_LEASE) {
2542                 LASSERT(atomic_read(&o->mot_lease_count) > 0);
2543                 atomic_dec(&o->mot_lease_count);
2544         }
2545
2546         mdt_mfd_free(mfd);
2547         mdt_object_put(info->mti_env, o);
2548
2549         RETURN(rc);
2550 }
2551
2552 int mdt_close_internal(struct mdt_thread_info *info, struct ptlrpc_request *req,
2553                        struct mdt_body *repbody)
2554 {
2555         struct mdt_export_data *med;
2556         struct mdt_file_data   *mfd;
2557         int ret = 0;
2558         int rc = 0;
2559
2560         ENTRY;
2561         med = &req->rq_export->exp_mdt_data;
2562         spin_lock(&med->med_open_lock);
2563         mfd = mdt_open_handle2mfd(med, &info->mti_open_handle,
2564                                   req_is_replay(req));
2565         if (mdt_mfd_closed(mfd)) {
2566                 spin_unlock(&med->med_open_lock);
2567                 CDEBUG(D_INODE, "no handle for file close: fid = "DFID
2568                        ": cookie = %#llx\n", PFID(info->mti_rr.rr_fid1),
2569                        info->mti_open_handle.cookie);
2570                 /** not serious error since bug 3633 */
2571                 rc = -ESTALE;
2572         } else {
2573                 class_handle_unhash(&mfd->mfd_open_handle);
2574                 list_del_init(&mfd->mfd_list);
2575                 spin_unlock(&med->med_open_lock);
2576                 ret = mdt_mfd_close(info, mfd);
2577         }
2578
2579         RETURN(rc ? rc : ret);
2580 }
2581
2582 int mdt_close(struct tgt_session_info *tsi)
2583 {
2584         struct mdt_thread_info *info = tsi2mdt_info(tsi);
2585         struct ptlrpc_request *req = tgt_ses_req(tsi);
2586         struct md_attr *ma = &info->mti_attr;
2587         struct mdt_body *repbody = NULL;
2588         ktime_t kstart = ktime_get();
2589         int rc;
2590         int rc2;
2591
2592         ENTRY;
2593         /* Close may come with the Size-on-MDS update. Unpack it. */
2594         rc = mdt_close_unpack(info);
2595         if (rc)
2596                 GOTO(out, rc = err_serious(rc));
2597
2598         /* These fields are no longer used and are left for compatibility.
2599          * size is always zero */
2600         req_capsule_set_size(info->mti_pill, &RMF_MDT_MD, RCL_SERVER,
2601                              0);
2602         req_capsule_set_size(info->mti_pill, &RMF_LOGCOOKIES, RCL_SERVER,
2603                              0);
2604         rc = req_capsule_server_pack(info->mti_pill);
2605         if (mdt_check_resent(info, mdt_reconstruct_generic, NULL)) {
2606                 mdt_client_compatibility(info);
2607                 if (rc == 0)
2608                         mdt_fix_reply(info);
2609                 mdt_exit_ucred(info);
2610                 GOTO(out, rc = lustre_msg_get_status(req->rq_repmsg));
2611         }
2612
2613         /* Continue to close handle even if we can not pack reply */
2614         if (rc == 0) {
2615                 repbody = req_capsule_server_get(info->mti_pill,
2616                                                  &RMF_MDT_BODY);
2617                 ma->ma_lmm = req_capsule_server_get(info->mti_pill,
2618                                                     &RMF_MDT_MD);
2619                 ma->ma_lmm_size = req_capsule_get_size(info->mti_pill,
2620                                                        &RMF_MDT_MD,
2621                                                        RCL_SERVER);
2622                 ma->ma_need = MA_INODE | MA_LOV;
2623                 repbody->mbo_eadatasize = 0;
2624                 repbody->mbo_aclsize = 0;
2625         } else {
2626                 rc = err_serious(rc);
2627         }
2628
2629         rc2 = mdt_close_internal(info, req, repbody);
2630         if (rc2 != -ESTALE)
2631                 mdt_empty_transno(info, rc2);
2632         if (rc2 != 0 && rc == 0)
2633                 rc = rc2;
2634
2635         if (repbody != NULL) {
2636                 mdt_client_compatibility(info);
2637                 rc2 = mdt_fix_reply(info);
2638                 if (rc2 != 0 && rc == 0)
2639                         rc = rc2;
2640         }
2641
2642         mdt_exit_ucred(info);
2643         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_CLOSE_PACK))
2644                 GOTO(out, rc = err_serious(-ENOMEM));
2645
2646         if (OBD_FAIL_CHECK_RESET(OBD_FAIL_MDS_CLOSE_NET_REP,
2647                                  OBD_FAIL_MDS_CLOSE_NET_REP))
2648                 tsi->tsi_reply_fail_id = OBD_FAIL_MDS_CLOSE_NET_REP;
2649 out:
2650         mdt_thread_info_fini(info);
2651         if (rc == 0)
2652                 mdt_counter_incr(req, LPROC_MDT_CLOSE,
2653                                  ktime_us_delta(ktime_get(), kstart));
2654         RETURN(rc);
2655 }