1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
6 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
8 * This program is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License version 2 only,
10 * as published by the Free Software Foundation.
12 * This program is distributed in the hope that it will be useful, but
13 * WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * General Public License version 2 for more details (a copy is included
16 * in the LICENSE file that accompanied this code).
18 * You should have received a copy of the GNU General Public License
19 * version 2 along with this program; If not, see
20 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
22 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
23 * CA 95054 USA or visit www.sun.com if you need additional information or
29 * Copyright 2008 Sun Microsystems, Inc. All rights reserved
30 * Use is subject to license terms.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * lustre/mdt/mdt_open.c
38 * Lustre Metadata Target (mdt) open/close file handling
40 * Author: Huang Hua <huanghua@clusterfs.com>
44 # define EXPORT_SYMTAB
46 #define DEBUG_SUBSYSTEM S_MDS
48 #include <lustre_acl.h>
49 #include <lustre_mds.h>
50 #include "mdt_internal.h"
52 /* we do nothing because we do not have refcount now */
53 static void mdt_mfd_get(void *mfdp)
57 /* Create a new mdt_file_data struct, initialize it,
58 * and insert it to global hash table */
59 struct mdt_file_data *mdt_mfd_new(void)
61 struct mdt_file_data *mfd;
66 CFS_INIT_LIST_HEAD(&mfd->mfd_handle.h_link);
67 CFS_INIT_LIST_HEAD(&mfd->mfd_list);
68 class_handle_hash(&mfd->mfd_handle, mdt_mfd_get);
74 * Find the mfd pointed to by handle in global hash table.
75 * In case of replay the handle is obsoleted
76 * but mfd can be found in mfd list by that handle
78 struct mdt_file_data *mdt_handle2mfd(struct mdt_thread_info *info,
79 const struct lustre_handle *handle)
81 struct ptlrpc_request *req = mdt_info_req(info);
82 struct mdt_file_data *mfd;
85 LASSERT(handle != NULL);
86 mfd = class_handle2object(handle->cookie);
87 /* during dw/setattr replay the mfd can be found by old handle */
89 lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY) {
90 struct mdt_export_data *med = &req->rq_export->exp_mdt_data;
91 list_for_each_entry(mfd, &med->med_open_head, mfd_list) {
92 if (mfd->mfd_old_handle.cookie == handle->cookie)
101 void mdt_mfd_free(struct mdt_file_data *mfd)
103 LASSERT(list_empty(&mfd->mfd_list));
104 OBD_FREE_RCU(mfd, sizeof *mfd, &mfd->mfd_handle);
107 static int mdt_create_data(struct mdt_thread_info *info,
108 struct mdt_object *p, struct mdt_object *o)
110 struct md_op_spec *spec = &info->mti_spec;
111 struct md_attr *ma = &info->mti_attr;
115 if (!md_should_create(spec->sp_cr_flags))
118 ma->ma_need = MA_INODE | MA_LOV;
120 rc = mdo_create_data(info->mti_env,
121 p ? mdt_object_child(p) : NULL,
122 mdt_object_child(o), spec, ma);
126 static int mdt_ioepoch_opened(struct mdt_object *mo)
128 return mo->mot_ioepoch_count;
131 int mdt_object_is_som_enabled(struct mdt_object *mo)
133 return !mo->mot_ioepoch;
137 * Re-enable Size-on-MDS.
138 * Call under ->mot_ioepoch_sem.
140 static void mdt_object_som_enable(struct mdt_object *mo, __u64 ioepoch)
142 if (ioepoch == mo->mot_ioepoch) {
143 LASSERT(!mdt_ioepoch_opened(mo));
150 * Open the IOEpoch. It is allowed if @writecount is not negative.
151 * The epoch and writecount handling is performed under the mot_ioepoch_sem.
153 int mdt_ioepoch_open(struct mdt_thread_info *info, struct mdt_object *o,
156 struct mdt_device *mdt = info->mti_mdt;
161 if (!(mdt_conn_flags(info) & OBD_CONNECT_SOM) ||
162 !S_ISREG(lu_object_attr(&o->mot_obj.mo_lu)))
165 down(&o->mot_ioepoch_sem);
166 if (mdt_ioepoch_opened(o)) {
167 /* Epoch continues even if there is no writers yet. */
168 CDEBUG(D_INODE, "continue epoch "LPU64" for "DFID"\n",
169 o->mot_ioepoch, PFID(mdt_object_fid(o)));
171 /* XXX: ->mdt_ioepoch is not initialized at the mount */
172 spin_lock(&mdt->mdt_ioepoch_lock);
173 if (mdt->mdt_ioepoch < info->mti_replayepoch)
174 mdt->mdt_ioepoch = info->mti_replayepoch;
176 if (info->mti_replayepoch)
177 o->mot_ioepoch = info->mti_replayepoch;
178 else if (++mdt->mdt_ioepoch == IOEPOCH_INVAL)
179 o->mot_ioepoch = ++mdt->mdt_ioepoch;
181 o->mot_ioepoch = mdt->mdt_ioepoch;
183 spin_unlock(&mdt->mdt_ioepoch_lock);
185 CDEBUG(D_INODE, "starting epoch "LPU64" for "DFID"\n",
186 o->mot_ioepoch, PFID(mdt_object_fid(o)));
188 o->mot_flags |= MOF_SOM_CREATED;
191 o->mot_ioepoch_count++;
192 up(&o->mot_ioepoch_sem);
194 /* Cancel Size-on-MDS attributes cached on clients for the open case.
195 * In the truncate case, see mdt_reint_setattr(). */
196 if (cancel && (info->mti_rr.rr_fid1 != NULL)) {
197 struct mdt_lock_handle *lh = &info->mti_lh[MDT_LH_CHILD];
198 mdt_lock_reg_init(lh, LCK_EX);
199 rc = mdt_object_lock(info, o, lh, MDS_INODELOCK_UPDATE,
202 mdt_object_unlock(info, o, lh, 1);
208 * Update SOM on-disk attributes.
209 * If enabling, write update inodes and lustre-ea with the proper IOEpoch,
210 * mountid and attributes. If disabling, zero IOEpoch id in lustre-ea.
211 * Call under ->mot_ioepoch_sem.
213 static int mdt_som_attr_set(struct mdt_thread_info *info,
214 struct mdt_object *obj, __u64 ioepoch, int enable)
216 struct md_attr *ma = &info->mti_attr;
220 CDEBUG(D_INODE, "Size-on-MDS attribute %s for epoch "LPU64
221 " on "DFID".\n", enable ? "update" : "disabling",
222 ioepoch, PFID(mdt_object_fid(obj)));
224 ma->ma_valid |= MA_SOM;
225 ma->ma_som = &info->mti_u.som.data;
227 struct mdt_device *mdt = info->mti_mdt;
228 struct lu_attr *la = &ma->ma_attr;
230 ma->ma_som->msd_ioepoch = ioepoch;
231 ma->ma_som->msd_size = la->la_valid & LA_SIZE ? la->la_size : 0;
232 ma->ma_som->msd_blocks = la->la_valid & LA_BLOCKS ?
234 ma->ma_som->msd_mountid = mdt->mdt_mount_count;
235 ma->ma_attr.la_valid &= LA_ATIME | LA_MTIME | LA_CTIME;
237 ma->ma_som->msd_ioepoch = IOEPOCH_INVAL;
238 ma->ma_attr.la_valid &= LA_ATIME;
241 /* Since we have opened the file, it is unnecessary
242 * to check permission when close it. Between the "open"
243 * and "close", maybe someone has changed the file mode
244 * or flags, or the file created mode do not permit wirte,
245 * and so on. Just set MDS_PERM_BYPASS for all the cases. */
246 ma->ma_attr_flags |= MDS_PERM_BYPASS | MDS_SOM;
248 rc = mdt_attr_set(info, obj, ma, 0);
252 /** Perform the eviction specific actions on ioepoch close. */
253 static inline int mdt_ioepoch_close_on_eviction(struct mdt_thread_info *info,
254 struct mdt_object *o)
258 down(&o->mot_ioepoch_sem);
259 CDEBUG(D_INODE, "Eviction. Closing IOepoch "LPU64" on "DFID". "
260 "Count %d\n", o->mot_ioepoch, PFID(mdt_object_fid(o)),
261 o->mot_ioepoch_count);
262 o->mot_ioepoch_count--;
264 /* If eviction occured set MOF_SOM_RECOV,
265 * if no other epoch holders, disable SOM on disk. */
266 o->mot_flags |= MOF_SOM_CHANGE | MOF_SOM_RECOV;
267 if (!mdt_ioepoch_opened(o)) {
268 rc = mdt_som_attr_set(info, o, o->mot_ioepoch, MDT_SOM_DISABLE);
269 mdt_object_som_enable(o, o->mot_ioepoch);
271 up(&o->mot_ioepoch_sem);
276 * Perform the replay specific actions on ioepoch close.
277 * Skip SOM attribute update if obtained and just forget about the inode state
278 * for the last ioepoch holder. The SOM cache is invalidated on MDS failure.
280 static inline int mdt_ioepoch_close_on_replay(struct mdt_thread_info *info,
281 struct mdt_object *o)
285 down(&o->mot_ioepoch_sem);
286 CDEBUG(D_INODE, "Replay. Closing epoch "LPU64" on "DFID". Count %d\n",
287 o->mot_ioepoch, PFID(mdt_object_fid(o)), o->mot_ioepoch_count);
288 o->mot_ioepoch_count--;
290 if (!mdt_ioepoch_opened(o))
291 mdt_object_som_enable(o, info->mti_ioepoch->ioepoch);
292 up(&o->mot_ioepoch_sem);
298 * Regular file IOepoch close.
299 * Closes the ioepoch, checks the object state, apply obtained attributes and
300 * re-enable SOM on the object, if possible.
303 * MDT_IOEPOCH_CLOSED if ioepoch is closed.
304 * MDT_IOEPOCH_GETATTR if ioepoch is closed but another SOM update is needed.
306 static inline int mdt_ioepoch_close_reg(struct mdt_thread_info *info,
307 struct mdt_object *o)
309 struct md_attr *tmp_ma;
313 int rc = 0, ret = MDT_IOEPOCH_CLOSED;
316 la = &info->mti_attr.ma_attr;
317 achange = (info->mti_ioepoch->flags & MF_SOM_CHANGE);
319 down(&o->mot_ioepoch_sem);
320 o->mot_ioepoch_count--;
322 tmp_ma = &info->mti_u.som.attr;
323 tmp_ma->ma_lmm = info->mti_attr.ma_lmm;
324 tmp_ma->ma_lmm_size = info->mti_attr.ma_lmm_size;
325 tmp_ma->ma_som = &info->mti_u.som.data;
326 tmp_ma->ma_need = MA_INODE | MA_LOV | MA_SOM;
327 tmp_ma->ma_valid = 0;
328 rc = mo_attr_get(info->mti_env, mdt_object_child(o), tmp_ma);
332 /* Check the on-disk SOM state. */
333 if (o->mot_flags & MOF_SOM_RECOV)
335 else if (!(o->mot_flags & MOF_SOM_CREATED) &&
336 !(tmp_ma->ma_valid & MA_SOM))
339 CDEBUG(D_INODE, "Closing epoch "LPU64" on "DFID". Count %d\n",
340 o->mot_ioepoch, PFID(mdt_object_fid(o)), o->mot_ioepoch_count);
342 opened = mdt_ioepoch_opened(o);
344 * If IOEpoch is not opened, check if a Size-on-MDS update is needed.
345 * Skip the check for file with no LOV or for unlink files.
347 if (!opened && tmp_ma->ma_valid & MA_LOV &&
348 !(tmp_ma->ma_valid & MA_INODE && tmp_ma->ma_attr.la_nlink == 0)) {
350 /* If some previous writer was evicted, re-ask the
351 * client for attributes. Even if attributes are
352 * provided, we cannot believe in them.
353 * Another use case is that there is no SOM cache on
354 * disk -- first access with SOM or there was an MDS
356 ret = MDT_IOEPOCH_GETATTR;
357 else if (o->mot_flags & MOF_SOM_CHANGE)
358 /* Some previous writer changed the attribute.
359 * Do not believe to the current Size-on-MDS
360 * update, re-ask client. */
361 ret = MDT_IOEPOCH_GETATTR;
362 else if (!(la->la_valid & LA_SIZE) && achange)
363 /* Attributes were changed by the last writer
364 * only but no Size-on-MDS update is received.*/
365 ret = MDT_IOEPOCH_GETATTR;
368 if (achange || ret == MDT_IOEPOCH_GETATTR)
369 o->mot_flags |= MOF_SOM_CHANGE;
371 /* If epoch ends and relable SOM attributes are obtained, update them.
372 * Create SOM ea for new files even if there is no attributes obtained
373 * (0-length file). */
374 if (ret == MDT_IOEPOCH_CLOSED && !opened) {
375 if (achange || o->mot_flags & MOF_SOM_CREATED) {
376 LASSERT(achange || !(la->la_valid & LA_SIZE));
377 rc = mdt_som_attr_set(info, o, o->mot_ioepoch,
379 /* Avoid the following setattrs of these attributes,
380 * e.g. for atime update. */
381 info->mti_attr.ma_valid = 0;
383 mdt_object_som_enable(o, o->mot_ioepoch);
388 up(&o->mot_ioepoch_sem);
393 * Close IOEpoch (opened file or FMODE_EPOCH state). It happens if:
394 * - a client closes the IOEpoch;
395 * - a client eviction occured.
397 * MDT_IOEPOCH_OPENED if the client does not close IOEpoch.
398 * MDT_IOEPOCH_CLOSED if the client closes IOEpoch.
399 * MDT_IOEPOCH_GETATTR if the client closes IOEpoch but another SOM attribute
402 static int mdt_ioepoch_close(struct mdt_thread_info *info, struct mdt_object *o)
404 struct ptlrpc_request *req = mdt_info_req(info);
407 if (!(mdt_conn_flags(info) & OBD_CONNECT_SOM) ||
408 !S_ISREG(lu_object_attr(&o->mot_obj.mo_lu)))
411 LASSERT(o->mot_ioepoch_count);
412 LASSERT(info->mti_ioepoch == NULL ||
413 info->mti_ioepoch->ioepoch == o->mot_ioepoch);
415 /* IOEpoch is closed only if client tells about it or eviction occures.
416 * In the replay case, always close the epoch. */
418 RETURN(mdt_ioepoch_close_on_eviction(info, o));
419 if (lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY)
420 RETURN(mdt_ioepoch_close_on_replay(info, o));
421 if (info->mti_ioepoch->flags & MF_EPOCH_CLOSE)
422 RETURN(mdt_ioepoch_close_reg(info, o));
423 /* IO epoch is not closed. */
424 RETURN(MDT_IOEPOCH_OPENED);
428 * Close FMODE_SOM state, when IOEpoch is already closed and we are waiting for
429 * attribute update. It happens if:
430 * - SOM Attribute Update is obtained;
431 * - the client failed to obtain it and informs MDS about it;
432 * - a client eviction occured.
433 * Apply obtained attributes for the 1st case, wipe out the on-disk SOM
436 int mdt_som_au_close(struct mdt_thread_info *info, struct mdt_object *o)
438 struct ptlrpc_request *req = mdt_info_req(info);
440 int act = MDT_SOM_ENABLE;
444 LASSERT(!req || info->mti_ioepoch);
445 if (!(mdt_conn_flags(info) & OBD_CONNECT_SOM) ||
446 !S_ISREG(lu_object_attr(&o->mot_obj.mo_lu)))
449 /* No size whereas MF_SOM_CHANGE is set means client failed to
450 * obtain ost attributes, drop the SOM cache on disk if so. */
452 (info->mti_ioepoch &&
453 info->mti_ioepoch->flags & MF_SOM_CHANGE &&
454 !(info->mti_attr.ma_attr.la_valid & LA_SIZE)))
455 act = MDT_SOM_DISABLE;
457 down(&o->mot_ioepoch_sem);
458 /* Mark the object it is the recovery state if we failed to obtain
460 if (act == MDT_SOM_DISABLE)
461 o->mot_flags |= MOF_SOM_RECOV;
463 if (!mdt_ioepoch_opened(o)) {
464 ioepoch = info->mti_ioepoch ?
465 info->mti_ioepoch->ioepoch : o->mot_ioepoch;
467 if (!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY))
468 rc = mdt_som_attr_set(info, o, ioepoch, act);
469 mdt_object_som_enable(o, ioepoch);
471 up(&o->mot_ioepoch_sem);
475 int mdt_write_read(struct mdt_object *o)
479 down(&o->mot_ioepoch_sem);
480 rc = o->mot_writecount;
481 up(&o->mot_ioepoch_sem);
485 int mdt_write_get(struct mdt_object *o)
489 down(&o->mot_ioepoch_sem);
490 if (o->mot_writecount < 0)
494 up(&o->mot_ioepoch_sem);
498 void mdt_write_put(struct mdt_object *o)
501 down(&o->mot_ioepoch_sem);
503 up(&o->mot_ioepoch_sem);
507 static int mdt_write_deny(struct mdt_object *o)
511 down(&o->mot_ioepoch_sem);
512 if (o->mot_writecount > 0)
516 up(&o->mot_ioepoch_sem);
520 static void mdt_write_allow(struct mdt_object *o)
523 down(&o->mot_ioepoch_sem);
525 up(&o->mot_ioepoch_sem);
529 /* there can be no real transaction so prepare the fake one */
530 static void mdt_empty_transno(struct mdt_thread_info* info)
532 struct mdt_device *mdt = info->mti_mdt;
533 struct ptlrpc_request *req = mdt_info_req(info);
536 /* transaction has occurred already */
537 if (lustre_msg_get_transno(req->rq_repmsg) != 0) {
542 spin_lock(&mdt->mdt_transno_lock);
543 if (info->mti_transno == 0) {
544 info->mti_transno = ++ mdt->mdt_last_transno;
546 /* should be replay */
547 if (info->mti_transno > mdt->mdt_last_transno)
548 mdt->mdt_last_transno = info->mti_transno;
550 spin_unlock(&mdt->mdt_transno_lock);
552 CDEBUG(D_INODE, "transno = %llu, last_committed = %llu\n",
554 req->rq_export->exp_obd->obd_last_committed);
556 req->rq_transno = info->mti_transno;
557 lustre_msg_set_transno(req->rq_repmsg, info->mti_transno);
561 void mdt_mfd_set_mode(struct mdt_file_data *mfd, int mode)
563 LASSERT(mfd != NULL);
565 CDEBUG(D_HA, "Change mfd %p mode 0x%x->0x%x\n",
566 mfd, (unsigned int)mfd->mfd_mode, (unsigned int)mode);
568 mfd->mfd_mode = mode;
571 static int mdt_mfd_open(struct mdt_thread_info *info, struct mdt_object *p,
572 struct mdt_object *o, int flags, int created)
574 struct ptlrpc_request *req = mdt_info_req(info);
575 struct mdt_export_data *med = &req->rq_export->exp_mdt_data;
576 struct mdt_file_data *mfd;
577 struct md_attr *ma = &info->mti_attr;
578 struct lu_attr *la = &ma->ma_attr;
579 struct mdt_body *repbody;
580 int rc = 0, isdir, isreg;
583 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
585 isreg = S_ISREG(la->la_mode);
586 isdir = S_ISDIR(la->la_mode);
587 if (isreg && !(ma->ma_valid & MA_LOV)) {
589 * No EA, check whether it is will set regEA and dirEA since in
590 * above attr get, these size might be zero, so reset it, to
591 * retrieve the MD after create obj.
593 ma->ma_lmm_size = req_capsule_get_size(info->mti_pill,
596 /* in replay case, p == NULL */
597 rc = mdt_create_data(info, p, o);
602 CDEBUG(D_INODE, "after open, ma_valid bit = "LPX64" lmm_size = %d\n",
603 ma->ma_valid, ma->ma_lmm_size);
605 if (ma->ma_valid & MA_LOV) {
606 LASSERT(ma->ma_lmm_size != 0);
607 repbody->eadatasize = ma->ma_lmm_size;
609 repbody->valid |= OBD_MD_FLDIREA;
611 repbody->valid |= OBD_MD_FLEASIZE;
614 if (flags & FMODE_WRITE) {
615 rc = mdt_write_get(o);
617 mdt_ioepoch_open(info, o, created);
618 repbody->ioepoch = o->mot_ioepoch;
620 } else if (flags & MDS_FMODE_EXEC) {
621 rc = mdt_write_deny(o);
626 rc = mo_open(info->mti_env, mdt_object_child(o),
627 created ? flags | MDS_OPEN_CREATED : flags);
634 * Keep a reference on this object for this open, and is
635 * released by mdt_mfd_close().
637 mdt_object_get(info->mti_env, o);
640 * @flags is always not zero. At least it should be FMODE_READ,
641 * FMODE_WRITE or FMODE_EXEC.
646 mdt_mfd_set_mode(mfd, flags);
649 mfd->mfd_xid = req->rq_xid;
652 if (lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY) {
653 struct mdt_file_data *old_mfd;
654 /* Check wheather old cookie already exist in
655 * the list, becasue when do recovery, client
656 * might be disconnected from server, and
657 * restart replay, so there maybe some orphan
658 * mfd here, we should remove them */
659 LASSERT(info->mti_rr.rr_handle != NULL);
660 old_mfd = mdt_handle2mfd(info, info->mti_rr.rr_handle);
662 CDEBUG(D_HA, "del orph mfd %p fid=("DFID") "
665 PFID(mdt_object_fid(mfd->mfd_object)),
666 info->mti_rr.rr_handle->cookie);
667 spin_lock(&med->med_open_lock);
668 class_handle_unhash(&old_mfd->mfd_handle);
669 list_del_init(&old_mfd->mfd_list);
670 spin_unlock(&med->med_open_lock);
671 mdt_mfd_close(info, old_mfd);
673 CDEBUG(D_HA, "Store old cookie "LPX64" in new mfd\n",
674 info->mti_rr.rr_handle->cookie);
675 mfd->mfd_old_handle.cookie =
676 info->mti_rr.rr_handle->cookie;
678 repbody->handle.cookie = mfd->mfd_handle.h_cookie;
680 if (req->rq_export->exp_disconnected) {
681 spin_lock(&med->med_open_lock);
682 class_handle_unhash(&mfd->mfd_handle);
683 list_del_init(&mfd->mfd_list);
684 spin_unlock(&med->med_open_lock);
685 mdt_mfd_close(info, mfd);
687 spin_lock(&med->med_open_lock);
688 list_add(&mfd->mfd_list, &med->med_open_head);
689 spin_unlock(&med->med_open_lock);
692 mdt_empty_transno(info);
700 static int mdt_finish_open(struct mdt_thread_info *info,
701 struct mdt_object *p, struct mdt_object *o,
702 int flags, int created, struct ldlm_reply *rep)
704 struct ptlrpc_request *req = mdt_info_req(info);
705 struct obd_export *exp = req->rq_export;
706 struct mdt_export_data *med = &req->rq_export->exp_mdt_data;
707 struct md_attr *ma = &info->mti_attr;
708 struct lu_attr *la = &ma->ma_attr;
709 struct mdt_file_data *mfd;
710 struct mdt_body *repbody;
712 int isreg, isdir, islnk;
716 LASSERT(ma->ma_valid & MA_INODE);
718 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
720 isreg = S_ISREG(la->la_mode);
721 isdir = S_ISDIR(la->la_mode);
722 islnk = S_ISLNK(la->la_mode);
723 mdt_pack_attr2body(info, repbody, la, mdt_object_fid(o));
725 if (exp_connect_rmtclient(exp)) {
726 void *buf = req_capsule_server_get(info->mti_pill, &RMF_ACL);
728 rc = mdt_pack_remote_perm(info, o, buf);
730 repbody->valid &= ~OBD_MD_FLRMTPERM;
731 repbody->aclsize = 0;
733 repbody->valid |= OBD_MD_FLRMTPERM;
734 repbody->aclsize = sizeof(struct mdt_remote_perm);
737 #ifdef CONFIG_FS_POSIX_ACL
738 else if (exp->exp_connect_flags & OBD_CONNECT_ACL) {
739 const struct lu_env *env = info->mti_env;
740 struct md_object *next = mdt_object_child(o);
741 struct lu_buf *buf = &info->mti_buf;
743 buf->lb_buf = req_capsule_server_get(info->mti_pill, &RMF_ACL);
744 buf->lb_len = req_capsule_get_size(info->mti_pill, &RMF_ACL,
746 if (buf->lb_len > 0) {
747 rc = mo_xattr_get(env, next, buf,
748 XATTR_NAME_ACL_ACCESS);
750 if (rc == -ENODATA) {
751 repbody->aclsize = 0;
752 repbody->valid |= OBD_MD_FLACL;
754 } else if (rc == -EOPNOTSUPP) {
757 CERROR("got acl size: %d\n", rc);
760 repbody->aclsize = rc;
761 repbody->valid |= OBD_MD_FLACL;
768 if (info->mti_mdt->mdt_opts.mo_mds_capa &&
769 exp->exp_connect_flags & OBD_CONNECT_MDS_CAPA) {
770 struct lustre_capa *capa;
772 capa = req_capsule_server_get(info->mti_pill, &RMF_CAPA1);
774 capa->lc_opc = CAPA_OPC_MDS_DEFAULT;
775 rc = mo_capa_get(info->mti_env, mdt_object_child(o), capa, 0);
778 repbody->valid |= OBD_MD_FLMDSCAPA;
780 if (info->mti_mdt->mdt_opts.mo_oss_capa &&
781 exp->exp_connect_flags & OBD_CONNECT_OSS_CAPA &&
782 S_ISREG(lu_object_attr(&o->mot_obj.mo_lu))) {
783 struct lustre_capa *capa;
785 capa = req_capsule_server_get(info->mti_pill, &RMF_CAPA2);
787 capa->lc_opc = CAPA_OPC_OSS_DEFAULT | capa_open_opc(flags);
788 rc = mo_capa_get(info->mti_env, mdt_object_child(o), capa, 0);
791 repbody->valid |= OBD_MD_FLOSSCAPA;
795 * If we are following a symlink, don't open; and do not return open
796 * handle for special nodes as client required.
798 if (islnk || (!isreg && !isdir &&
799 (req->rq_export->exp_connect_flags & OBD_CONNECT_NODEVOH))) {
800 lustre_msg_set_transno(req->rq_repmsg, 0);
804 mdt_set_disposition(info, rep, DISP_OPEN_OPEN);
807 * We need to return the existing object's fid back, so it is done here,
808 * after preparing the reply.
810 if (!created && (flags & MDS_OPEN_EXCL) && (flags & MDS_OPEN_CREAT))
813 /* This can't be done earlier, we need to return reply body */
815 if (flags & (MDS_OPEN_CREAT | FMODE_WRITE)) {
816 /* We are trying to create or write an existing dir. */
819 } else if (flags & MDS_OPEN_DIRECTORY)
822 if (OBD_FAIL_CHECK_RESET(OBD_FAIL_MDS_OPEN_CREATE,
823 OBD_FAIL_LDLM_REPLY | OBD_FAIL_ONCE)) {
828 if (lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT) {
829 spin_lock(&med->med_open_lock);
830 list_for_each(t, &med->med_open_head) {
831 mfd = list_entry(t, struct mdt_file_data, mfd_list);
832 if (mfd->mfd_xid == req->rq_xid) {
837 spin_unlock(&med->med_open_lock);
840 repbody->handle.cookie = mfd->mfd_handle.h_cookie;
841 /*set repbody->ea_size for resent case*/
842 if (ma->ma_valid & MA_LOV) {
843 LASSERT(ma->ma_lmm_size != 0);
844 repbody->eadatasize = ma->ma_lmm_size;
846 repbody->valid |= OBD_MD_FLDIREA;
848 repbody->valid |= OBD_MD_FLEASIZE;
854 rc = mdt_mfd_open(info, p, o, flags, created);
858 extern void mdt_req_from_lcd(struct ptlrpc_request *req,
859 struct lsd_client_data *lcd);
861 void mdt_reconstruct_open(struct mdt_thread_info *info,
862 struct mdt_lock_handle *lhc)
864 const struct lu_env *env = info->mti_env;
865 struct mdt_device *mdt = info->mti_mdt;
866 struct req_capsule *pill = info->mti_pill;
867 struct ptlrpc_request *req = mdt_info_req(info);
868 struct mdt_export_data *med = &req->rq_export->exp_mdt_data;
869 struct lsd_client_data *lcd = med->med_lcd;
870 struct md_attr *ma = &info->mti_attr;
871 struct mdt_reint_record *rr = &info->mti_rr;
872 __u32 flags = info->mti_spec.sp_cr_flags;
873 struct ldlm_reply *ldlm_rep;
874 struct mdt_object *parent;
875 struct mdt_object *child;
876 struct mdt_body *repbody;
880 LASSERT(pill->rc_fmt == &RQF_LDLM_INTENT_OPEN);
881 ldlm_rep = req_capsule_server_get(pill, &RMF_DLM_REP);
882 repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
884 ma->ma_lmm = req_capsule_server_get(pill, &RMF_MDT_MD);
885 ma->ma_lmm_size = req_capsule_get_size(pill, &RMF_MDT_MD,
887 ma->ma_need = MA_INODE;
888 if (ma->ma_lmm_size > 0)
889 ma->ma_need |= MA_LOV;
893 mdt_req_from_lcd(req, med->med_lcd);
894 mdt_set_disposition(info, ldlm_rep, lcd->lcd_last_data);
896 CERROR("This is reconstruct open: disp="LPX64", result=%d\n",
897 ldlm_rep->lock_policy_res1, req->rq_status);
899 if (mdt_get_disposition(ldlm_rep, DISP_OPEN_CREATE) &&
901 /* We did not create successfully, return error to client. */
902 GOTO(out, rc = req->rq_status);
904 if (mdt_get_disposition(ldlm_rep, DISP_OPEN_CREATE)) {
905 struct obd_export *exp = req->rq_export;
907 * We failed after creation, but we do not know in which step
908 * we failed. So try to check the child object.
910 parent = mdt_object_find(env, mdt, rr->rr_fid1);
911 if (IS_ERR(parent)) {
912 rc = PTR_ERR(parent);
913 LCONSOLE_WARN("Parent "DFID" lookup error %d."
914 " Evicting client %s with export %s.\n",
915 PFID(mdt_object_fid(parent)), rc,
916 obd_uuid2str(&exp->exp_client_uuid),
917 obd_export_nid2str(exp));
918 mdt_export_evict(exp);
922 child = mdt_object_find(env, mdt, rr->rr_fid2);
925 LCONSOLE_WARN("Child "DFID" lookup error %d."
926 " Evicting client %s with export %s.\n",
927 PFID(mdt_object_fid(child)), rc,
928 obd_uuid2str(&exp->exp_client_uuid),
929 obd_export_nid2str(exp));
930 mdt_object_put(env, parent);
931 mdt_export_evict(exp);
935 rc = mdt_object_exists(child);
937 struct md_object *next;
939 mdt_set_capainfo(info, 1, rr->rr_fid2, BYPASS_CAPA);
940 next = mdt_object_child(child);
941 rc = mo_attr_get(env, next, ma);
943 rc = mdt_finish_open(info, parent, child,
946 /* the child object was created on remote server */
947 repbody->fid1 = *rr->rr_fid2;
948 repbody->valid |= (OBD_MD_FLID | OBD_MD_MDS);
950 } else if (rc == 0) {
951 /* the child does not exist, we should do regular open */
952 mdt_object_put(env, parent);
953 mdt_object_put(env, child);
954 GOTO(regular_open, 0);
956 mdt_object_put(env, parent);
957 mdt_object_put(env, child);
961 /* We did not try to create, so we are a pure open */
962 rc = mdt_reint_open(info, lhc);
968 lustre_msg_set_status(req->rq_repmsg, req->rq_status);
969 LASSERT(ergo(rc < 0, lustre_msg_get_transno(req->rq_repmsg) == 0));
972 static int mdt_open_by_fid(struct mdt_thread_info* info,
973 struct ldlm_reply *rep)
975 const struct lu_env *env = info->mti_env;
976 __u32 flags = info->mti_spec.sp_cr_flags;
977 struct mdt_reint_record *rr = &info->mti_rr;
978 struct md_attr *ma = &info->mti_attr;
979 struct mdt_object *o;
983 o = mdt_object_find(info->mti_env, info->mti_mdt, rr->rr_fid2);
985 RETURN(rc = PTR_ERR(o));
987 rc = mdt_object_exists(o);
989 mdt_set_disposition(info, rep, (DISP_IT_EXECD |
993 rc = mo_attr_get(env, mdt_object_child(o), ma);
995 rc = mdt_finish_open(info, NULL, o, flags, 0, rep);
996 } else if (rc == 0) {
999 /* the child object was created on remote server */
1000 struct mdt_body *repbody;
1001 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1002 repbody->fid1 = *rr->rr_fid2;
1003 repbody->valid |= (OBD_MD_FLID | OBD_MD_MDS);
1007 mdt_object_put(info->mti_env, o);
1011 static int mdt_open_anon_by_fid(struct mdt_thread_info* info,
1012 struct ldlm_reply *rep,
1013 struct mdt_lock_handle *lhc)
1015 __u32 flags = info->mti_spec.sp_cr_flags;
1016 struct mdt_reint_record *rr = &info->mti_rr;
1017 struct md_attr *ma = &info->mti_attr;
1018 struct mdt_object *o;
1023 o = mdt_object_find(info->mti_env, info->mti_mdt, rr->rr_fid2);
1025 RETURN(rc = PTR_ERR(o));
1027 rc = mdt_object_exists(o);
1029 mdt_set_disposition(info, rep, (DISP_LOOKUP_EXECD |
1031 GOTO(out, rc = -ENOENT);
1032 } else if (rc < 0) {
1033 CERROR("NFS remote open shouldn't happen.\n");
1037 mdt_set_disposition(info, rep, (DISP_IT_EXECD |
1041 if (flags & FMODE_WRITE)
1043 else if (flags & MDS_FMODE_EXEC)
1048 mdt_lock_handle_init(lhc);
1049 mdt_lock_reg_init(lhc, lm);
1050 rc = mdt_object_lock(info, o, lhc,
1051 MDS_INODELOCK_LOOKUP | MDS_INODELOCK_OPEN,
1056 rc = mo_attr_get(info->mti_env, mdt_object_child(o), ma);
1060 if (flags & MDS_OPEN_LOCK)
1061 mdt_set_disposition(info, rep, DISP_OPEN_LOCK);
1062 rc = mdt_finish_open(info, NULL, o, flags, 0, rep);
1064 if (!(flags & MDS_OPEN_LOCK))
1065 mdt_object_unlock(info, o, lhc, 1);
1069 mdt_object_put(info->mti_env, o);
1073 int mdt_pin(struct mdt_thread_info* info)
1076 RETURN(err_serious(-EOPNOTSUPP));
1079 /* Cross-ref request. Currently it can only be a pure open (w/o create) */
1080 static int mdt_cross_open(struct mdt_thread_info* info,
1081 const struct lu_fid *fid,
1082 struct ldlm_reply *rep, __u32 flags)
1084 struct md_attr *ma = &info->mti_attr;
1085 struct mdt_object *o;
1089 o = mdt_object_find(info->mti_env, info->mti_mdt, fid);
1091 RETURN(rc = PTR_ERR(o));
1093 rc = mdt_object_exists(o);
1095 /* Do permission check for cross-open. */
1096 rc = mo_permission(info->mti_env, NULL, mdt_object_child(o),
1097 NULL, flags | MDS_OPEN_CROSS);
1101 mdt_set_capainfo(info, 0, fid, BYPASS_CAPA);
1102 rc = mo_attr_get(info->mti_env, mdt_object_child(o), ma);
1104 rc = mdt_finish_open(info, NULL, o, flags, 0, rep);
1105 } else if (rc == 0) {
1107 * Something is wrong here. lookup was positive but there is
1110 CERROR("Cross-ref object doesn't exist!\n");
1113 /* Something is wrong here, the object is on another MDS! */
1114 CERROR("The object isn't on this server! FLD error?\n");
1115 LU_OBJECT_DEBUG(D_WARNING, info->mti_env,
1117 "Object isn't on this server! FLD error?\n");
1123 mdt_object_put(info->mti_env, o);
1127 int mdt_reint_open(struct mdt_thread_info *info, struct mdt_lock_handle *lhc)
1129 struct mdt_device *mdt = info->mti_mdt;
1130 struct ptlrpc_request *req = mdt_info_req(info);
1131 struct mdt_object *parent;
1132 struct mdt_object *child;
1133 struct mdt_lock_handle *lh;
1134 struct ldlm_reply *ldlm_rep;
1135 struct mdt_body *repbody;
1136 struct lu_fid *child_fid = &info->mti_tmp_fid1;
1137 struct md_attr *ma = &info->mti_attr;
1138 __u32 create_flags = info->mti_spec.sp_cr_flags;
1139 struct mdt_reint_record *rr = &info->mti_rr;
1140 struct lu_name *lname;
1146 OBD_FAIL_TIMEOUT_ORSET(OBD_FAIL_MDS_PAUSE_OPEN, OBD_FAIL_ONCE,
1147 (obd_timeout + 1) / 4);
1149 repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1151 ma->ma_lmm = req_capsule_server_get(info->mti_pill, &RMF_MDT_MD);
1152 ma->ma_lmm_size = req_capsule_get_size(info->mti_pill, &RMF_MDT_MD,
1154 ma->ma_need = MA_INODE;
1155 if (ma->ma_lmm_size > 0)
1156 ma->ma_need |= MA_LOV;
1160 LASSERT(info->mti_pill->rc_fmt == &RQF_LDLM_INTENT_OPEN);
1161 ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
1163 if (unlikely(create_flags & MDS_OPEN_JOIN_FILE)) {
1164 CERROR("file join is not supported anymore.\n");
1165 GOTO(out, result = err_serious(-EOPNOTSUPP));
1167 msg_flags = lustre_msg_get_flags(req->rq_reqmsg);
1169 if ((create_flags & (MDS_OPEN_HAS_EA | MDS_OPEN_HAS_OBJS)) &&
1170 info->mti_spec.u.sp_ea.eadata == NULL)
1171 GOTO(out, result = err_serious(-EINVAL));
1173 CDEBUG(D_INODE, "I am going to open "DFID"/(%s->"DFID") "
1174 "cr_flag=0%o mode=0%06o msg_flag=0x%x\n",
1175 PFID(rr->rr_fid1), rr->rr_name,
1176 PFID(rr->rr_fid2), create_flags,
1177 ma->ma_attr.la_mode, msg_flags);
1179 if (msg_flags & MSG_REPLAY ||
1180 (req->rq_export->exp_libclient && create_flags&MDS_OPEN_HAS_EA)) {
1181 /* This is a replay request or from liblustre with ea. */
1182 result = mdt_open_by_fid(info, ldlm_rep);
1184 if (result != -ENOENT) {
1185 if (req->rq_export->exp_libclient &&
1186 create_flags&MDS_OPEN_HAS_EA)
1187 GOTO(out, result = 0);
1191 * We didn't find the correct object, so we need to re-create it
1192 * via a regular replay.
1194 if (!(create_flags & MDS_OPEN_CREAT)) {
1195 DEBUG_REQ(D_ERROR, req,"OPEN & CREAT not in open replay.");
1196 GOTO(out, result = -EFAULT);
1198 CDEBUG(D_INFO, "Open replay did find object, continue as "
1200 } else if (rr->rr_namelen == 0 && !info->mti_cross_ref) {
1201 result = mdt_open_anon_by_fid(info, ldlm_rep, lhc);
1202 if (result != -ENOENT)
1206 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_OPEN_PACK))
1207 GOTO(out, result = err_serious(-ENOMEM));
1209 mdt_set_disposition(info, ldlm_rep,
1210 (DISP_IT_EXECD | DISP_LOOKUP_EXECD));
1212 if (info->mti_cross_ref) {
1213 /* This is cross-ref open */
1214 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1215 result = mdt_cross_open(info, rr->rr_fid1, ldlm_rep,
1220 lh = &info->mti_lh[MDT_LH_PARENT];
1221 mdt_lock_pdo_init(lh, (create_flags & MDS_OPEN_CREAT) ?
1222 LCK_PW : LCK_PR, rr->rr_name, rr->rr_namelen);
1224 parent = mdt_object_find_lock(info, rr->rr_fid1, lh,
1225 MDS_INODELOCK_UPDATE);
1227 GOTO(out, result = PTR_ERR(parent));
1229 fid_zero(child_fid);
1231 lname = mdt_name(info->mti_env, (char *)rr->rr_name, rr->rr_namelen);
1233 result = mdo_lookup(info->mti_env, mdt_object_child(parent),
1234 lname, child_fid, &info->mti_spec);
1235 LASSERTF(ergo(result == 0, fid_is_sane(child_fid)),
1236 "looking for "DFID"/%s, result fid="DFID"\n",
1237 PFID(mdt_object_fid(parent)), rr->rr_name, PFID(child_fid));
1239 if (result != 0 && result != -ENOENT && result != -ESTALE)
1240 GOTO(out_parent, result);
1242 if (result == -ENOENT || result == -ESTALE) {
1243 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
1244 if (result == -ESTALE) {
1246 * -ESTALE means the parent is a dead(unlinked) dir, so
1247 * it should return -ENOENT to in accordance with the
1248 * original mds implementaion.
1250 GOTO(out_parent, result = -ENOENT);
1252 if (!(create_flags & MDS_OPEN_CREAT))
1253 GOTO(out_parent, result);
1254 *child_fid = *info->mti_rr.rr_fid2;
1255 LASSERTF(fid_is_sane(child_fid), "fid="DFID"\n",
1259 * Check for O_EXCL is moved to the mdt_finish_open(), we need to
1260 * return FID back in that case.
1262 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
1265 child = mdt_object_find(info->mti_env, mdt, child_fid);
1267 GOTO(out_parent, result = PTR_ERR(child));
1269 mdt_set_capainfo(info, 1, child_fid, BYPASS_CAPA);
1270 if (result == -ENOENT) {
1271 /* Not found and with MDS_OPEN_CREAT: let's create it. */
1272 mdt_set_disposition(info, ldlm_rep, DISP_OPEN_CREATE);
1274 info->mti_mos[0] = parent;
1275 info->mti_mos[1] = child;
1276 result = mdt_version_get_check(info, 0);
1278 GOTO(out_child, result);
1280 /* Let lower layers know what is lock mode on directory. */
1281 info->mti_spec.sp_cr_mode =
1282 mdt_dlm_mode2mdl_mode(lh->mlh_pdo_mode);
1285 * Do not perform lookup sanity check. We know that name does
1288 info->mti_spec.sp_cr_lookup = 0;
1289 info->mti_spec.sp_feat = &dt_directory_features;
1291 result = mdo_create(info->mti_env,
1292 mdt_object_child(parent),
1294 mdt_object_child(child),
1297 if (result == -ERESTART) {
1298 mdt_clear_disposition(info, ldlm_rep, DISP_OPEN_CREATE);
1299 GOTO(out_child, result);
1302 GOTO(out_child, result);
1306 /* We have to get attr & lov ea for this object */
1307 result = mo_attr_get(info->mti_env, mdt_object_child(child),
1310 * The object is on remote node, return its FID for remote open.
1312 if (result == -EREMOTE) {
1314 * Check if this lock already was sent to client and
1315 * this is resent case. For resent case do not take lock
1316 * again, use what is already granted.
1318 LASSERT(lhc != NULL);
1320 if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
1321 struct ldlm_lock *lock;
1323 LASSERT(msg_flags & MSG_RESENT);
1325 lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
1327 CERROR("Invalid lock handle "LPX64"\n",
1328 lhc->mlh_reg_lh.cookie);
1331 LASSERT(fid_res_name_eq(mdt_object_fid(child),
1332 &lock->l_resource->lr_name));
1333 LDLM_LOCK_PUT(lock);
1336 mdt_lock_handle_init(lhc);
1337 mdt_lock_reg_init(lhc, LCK_PR);
1339 rc = mdt_object_lock(info, child, lhc,
1340 MDS_INODELOCK_LOOKUP,
1343 repbody->fid1 = *mdt_object_fid(child);
1344 repbody->valid |= (OBD_MD_FLID | OBD_MD_MDS);
1347 GOTO(out_child, result);
1351 LASSERT(!lustre_handle_is_used(&lhc->mlh_reg_lh));
1353 /* get openlock if this is not replay and if a client requested it */
1354 if (!(msg_flags & MSG_REPLAY) && create_flags & MDS_OPEN_LOCK) {
1358 if (create_flags & FMODE_WRITE)
1360 else if (create_flags & MDS_FMODE_EXEC)
1364 mdt_lock_handle_init(lhc);
1365 mdt_lock_reg_init(lhc, lm);
1366 rc = mdt_object_lock(info, child, lhc,
1367 MDS_INODELOCK_LOOKUP | MDS_INODELOCK_OPEN,
1371 GOTO(out_child, result);
1374 mdt_set_disposition(info, ldlm_rep, DISP_OPEN_LOCK);
1378 /* Try to open it now. */
1379 rc = mdt_finish_open(info, parent, child, create_flags,
1383 if (lustre_handle_is_used(&lhc->mlh_reg_lh))
1384 /* openlock was acquired and mdt_finish_open failed -
1385 drop the openlock */
1386 mdt_object_unlock(info, child, lhc, 1);
1390 ma->ma_cookie_size = 0;
1391 info->mti_no_need_trans = 1;
1392 rc = mdo_unlink(info->mti_env,
1393 mdt_object_child(parent),
1394 mdt_object_child(child),
1398 CERROR("Error in cleanup of open\n");
1403 mdt_object_put(info->mti_env, child);
1405 mdt_object_unlock_put(info, parent, lh, result);
1407 if (result && result != -EREMOTE)
1408 lustre_msg_set_transno(req->rq_repmsg, 0);
1412 #define MFD_CLOSED(mode) (((mode) & ~(FMODE_EPOCH | FMODE_SOM | \
1413 FMODE_TRUNC)) == FMODE_CLOSED)
1415 static int mdt_mfd_closed(struct mdt_file_data *mfd)
1417 return ((mfd == NULL) || MFD_CLOSED(mfd->mfd_mode));
1420 int mdt_mfd_close(struct mdt_thread_info *info, struct mdt_file_data *mfd)
1422 struct mdt_object *o = mfd->mfd_object;
1423 struct md_object *next = mdt_object_child(o);
1424 struct md_attr *ma = &info->mti_attr;
1425 int ret = MDT_IOEPOCH_CLOSED;
1430 mode = mfd->mfd_mode;
1432 if ((mode & FMODE_WRITE) || (mode & FMODE_TRUNC)) {
1434 ret = mdt_ioepoch_close(info, o);
1435 } else if (mode & MDS_FMODE_EXEC) {
1437 } else if (mode & FMODE_EPOCH) {
1438 ret = mdt_ioepoch_close(info, o);
1439 } else if (mode & FMODE_SOM) {
1440 ret = mdt_som_au_close(info, o);
1443 /* Update atime on close only. */
1444 if ((mode & MDS_FMODE_EXEC || mode & FMODE_READ || mode & FMODE_WRITE)
1445 && (ma->ma_valid & MA_INODE) && (ma->ma_attr.la_valid & LA_ATIME)) {
1446 /* Set the atime only. */
1447 ma->ma_attr.la_valid = LA_ATIME;
1448 rc = mo_attr_set(info->mti_env, next, ma);
1451 ma->ma_need |= MA_INODE;
1454 if (!MFD_CLOSED(mode))
1455 rc = mo_close(info->mti_env, next, ma);
1457 if (ret == MDT_IOEPOCH_GETATTR || ret == MDT_IOEPOCH_OPENED) {
1458 struct mdt_export_data *med;
1460 /* The IOepoch is still opened or SOM update is needed.
1461 * Put mfd back into the list. */
1462 LASSERT(mdt_conn_flags(info) & OBD_CONNECT_SOM);
1463 mdt_mfd_set_mode(mfd, ret == MDT_IOEPOCH_OPENED ?
1464 FMODE_EPOCH : FMODE_SOM);
1466 LASSERT(mdt_info_req(info));
1467 med = &mdt_info_req(info)->rq_export->exp_mdt_data;
1468 spin_lock(&med->med_open_lock);
1469 list_add(&mfd->mfd_list, &med->med_open_head);
1470 class_handle_hash_back(&mfd->mfd_handle);
1471 spin_unlock(&med->med_open_lock);
1473 if (ret == MDT_IOEPOCH_OPENED) {
1477 CDEBUG(D_INODE, "Size-on-MDS attribute update is "
1478 "needed on "DFID"\n", PFID(mdt_object_fid(o)));
1482 mdt_object_put(info->mti_env, o);
1485 RETURN(rc ? rc : ret);
1488 int mdt_close(struct mdt_thread_info *info)
1490 struct mdt_export_data *med;
1491 struct mdt_file_data *mfd;
1492 struct mdt_object *o;
1493 struct md_attr *ma = &info->mti_attr;
1494 struct mdt_body *repbody = NULL;
1495 struct ptlrpc_request *req = mdt_info_req(info);
1499 /* Close may come with the Size-on-MDS update. Unpack it. */
1500 rc = mdt_close_unpack(info);
1502 RETURN(err_serious(rc));
1504 LASSERT(info->mti_ioepoch);
1506 req_capsule_set_size(info->mti_pill, &RMF_MDT_MD, RCL_SERVER,
1507 info->mti_mdt->mdt_max_mdsize);
1508 req_capsule_set_size(info->mti_pill, &RMF_LOGCOOKIES, RCL_SERVER,
1509 info->mti_mdt->mdt_max_cookiesize);
1510 rc = req_capsule_server_pack(info->mti_pill);
1511 if (mdt_check_resent(info, mdt_reconstruct_generic, NULL)) {
1513 mdt_shrink_reply(info);
1514 RETURN(lustre_msg_get_status(req->rq_repmsg));
1517 /* Continue to close handle even if we can not pack reply */
1519 repbody = req_capsule_server_get(info->mti_pill,
1521 ma->ma_lmm = req_capsule_server_get(info->mti_pill,
1523 ma->ma_lmm_size = req_capsule_get_size(info->mti_pill,
1526 ma->ma_cookie = req_capsule_server_get(info->mti_pill,
1528 ma->ma_cookie_size = req_capsule_get_size(info->mti_pill,
1531 ma->ma_need = MA_INODE | MA_LOV | MA_COOKIE;
1532 repbody->eadatasize = 0;
1533 repbody->aclsize = 0;
1535 rc = err_serious(rc);
1538 med = &req->rq_export->exp_mdt_data;
1539 spin_lock(&med->med_open_lock);
1540 mfd = mdt_handle2mfd(info, &info->mti_ioepoch->handle);
1541 if (mdt_mfd_closed(mfd)) {
1542 spin_unlock(&med->med_open_lock);
1543 CDEBUG(D_INODE, "no handle for file close: fid = "DFID
1544 ": cookie = "LPX64"\n", PFID(info->mti_rr.rr_fid1),
1545 info->mti_ioepoch->handle.cookie);
1546 /** not serious error since bug 3633 */
1549 class_handle_unhash(&mfd->mfd_handle);
1550 list_del_init(&mfd->mfd_list);
1551 spin_unlock(&med->med_open_lock);
1553 /* Do not lose object before last unlink. */
1554 o = mfd->mfd_object;
1555 mdt_object_get(info->mti_env, o);
1556 ret = mdt_mfd_close(info, mfd);
1557 if (repbody != NULL)
1558 rc = mdt_handle_last_unlink(info, o, ma);
1559 mdt_empty_transno(info);
1560 mdt_object_put(info->mti_env, o);
1562 if (repbody != NULL)
1563 mdt_shrink_reply(info);
1565 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_CLOSE_PACK))
1566 RETURN(err_serious(-ENOMEM));
1568 if (OBD_FAIL_CHECK_RESET(OBD_FAIL_MDS_CLOSE_NET_REP,
1569 OBD_FAIL_MDS_CLOSE_NET_REP))
1570 info->mti_fail_id = OBD_FAIL_MDS_CLOSE_NET_REP;
1571 RETURN(rc ? rc : ret);
1575 * DONE_WRITING rpc handler.
1577 * As mfd is not kept after replayed CLOSE (see mdt_ioepoch_close_on_replay()),
1578 * only those DONE_WRITING rpc will be replayed which really wrote smth on disk,
1579 * and got a trasid. Waiting for such DONE_WRITING is not reliable, so just
1580 * skip attributes and reconstruct the reply here.
1582 int mdt_done_writing(struct mdt_thread_info *info)
1584 struct ptlrpc_request *req = mdt_info_req(info);
1585 struct mdt_body *repbody = NULL;
1586 struct mdt_export_data *med;
1587 struct mdt_file_data *mfd;
1591 rc = req_capsule_server_pack(info->mti_pill);
1593 RETURN(err_serious(rc));
1595 repbody = req_capsule_server_get(info->mti_pill,
1597 repbody->eadatasize = 0;
1598 repbody->aclsize = 0;
1600 /* Done Writing may come with the Size-on-MDS update. Unpack it. */
1601 rc = mdt_close_unpack(info);
1603 RETURN(err_serious(rc));
1605 if (mdt_check_resent(info, mdt_reconstruct_generic, NULL))
1606 RETURN(lustre_msg_get_status(req->rq_repmsg));
1608 med = &info->mti_exp->exp_mdt_data;
1609 spin_lock(&med->med_open_lock);
1610 mfd = mdt_handle2mfd(info, &info->mti_ioepoch->handle);
1612 spin_unlock(&med->med_open_lock);
1613 CDEBUG(D_INODE, "no handle for done write: fid = "DFID
1614 ": cookie = "LPX64" ioepoch = "LPU64"\n",
1615 PFID(info->mti_rr.rr_fid1),
1616 info->mti_ioepoch->handle.cookie,
1617 info->mti_ioepoch->ioepoch);
1618 /* If this is a replay, reconstruct the transno. */
1619 if (lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY) {
1620 mdt_empty_transno(info);
1626 LASSERT(mfd->mfd_mode == FMODE_EPOCH ||
1627 mfd->mfd_mode == FMODE_TRUNC);
1628 class_handle_unhash(&mfd->mfd_handle);
1629 list_del_init(&mfd->mfd_list);
1630 spin_unlock(&med->med_open_lock);
1632 /* Set EPOCH CLOSE flag if not set by client. */
1633 info->mti_ioepoch->flags |= MF_EPOCH_CLOSE;
1634 info->mti_attr.ma_valid = 0;
1636 info->mti_attr.ma_lmm_size = info->mti_mdt->mdt_max_mdsize;
1637 OBD_ALLOC(info->mti_attr.ma_lmm, info->mti_attr.ma_lmm_size);
1638 if (info->mti_attr.ma_lmm == NULL)
1641 rc = mdt_mfd_close(info, mfd);
1643 OBD_FREE(info->mti_attr.ma_lmm, info->mti_attr.ma_lmm_size);
1644 mdt_empty_transno(info);