1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
5 * Lustre Metadata Target (mdt) open/close file handling
7 * Copyright (C) 2002-2006 Cluster File Systems, Inc.
8 * Author: Huang Hua <huanghua@clusterfs.com>
10 * This file is part of the Lustre file system, http://www.lustre.org
11 * Lustre is a trademark of Cluster File Systems, Inc.
13 * You may have signed or agreed to another license before downloading
14 * this software. If so, you are bound by the terms and conditions
15 * of that agreement, and the following does not apply to you. See the
16 * LICENSE file included with this distribution for more information.
18 * If you did not agree to a different license, then this copy of Lustre
19 * is open source software; you can redistribute it and/or modify it
20 * under the terms of version 2 of the GNU General Public License as
21 * published by the Free Software Foundation.
23 * In either case, Lustre is distributed in the hope that it will be
24 * useful, but WITHOUT ANY WARRANTY; without even the implied warranty
25 * of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
26 * license text for more details.
30 # define EXPORT_SYMTAB
32 #define DEBUG_SUBSYSTEM S_MDS
34 #include <linux/lustre_acl.h>
35 #include <lustre_mds.h>
36 #include "mdt_internal.h"
38 /* we do nothing because we do not have refcount now */
39 static void mdt_mfd_get(void *mfdp)
43 /* Create a new mdt_file_data struct, initialize it,
44 * and insert it to global hash table */
45 struct mdt_file_data *mdt_mfd_new(void)
47 struct mdt_file_data *mfd;
52 INIT_LIST_HEAD(&mfd->mfd_handle.h_link);
53 INIT_LIST_HEAD(&mfd->mfd_list);
54 class_handle_hash(&mfd->mfd_handle, mdt_mfd_get);
59 /* Find the mfd pointed to by handle in global hash table. */
60 struct mdt_file_data *mdt_handle2mfd(const struct lustre_handle *handle)
63 LASSERT(handle != NULL);
64 RETURN(class_handle2object(handle->cookie));
68 void mdt_mfd_free(struct mdt_file_data *mfd)
70 LASSERT(list_empty(&mfd->mfd_handle.h_link));
71 LASSERT(list_empty(&mfd->mfd_list));
75 static int mdt_create_data(struct mdt_thread_info *info,
76 struct mdt_object *p, struct mdt_object *o)
78 struct md_create_spec *spec = &info->mti_spec;
79 struct md_attr *ma = &info->mti_attr;
83 if ((spec->sp_cr_flags & MDS_OPEN_DELAY_CREATE) ||
84 !(spec->sp_cr_flags & FMODE_WRITE))
87 ma->ma_need = MA_INODE | MA_LOV;
89 rc = mdo_create_data(info->mti_env,
90 p ? mdt_object_child(p) : NULL,
91 mdt_object_child(o), spec, ma);
95 static int mdt_epoch_opened(struct mdt_object *mo)
97 return mo->mot_epochcount;
100 int mdt_sizeonmds_enabled(struct mdt_object *mo)
102 return !mo->mot_ioepoch;
105 /* Re-enable Size-on-MDS. */
106 void mdt_sizeonmds_enable(struct mdt_thread_info *info,
107 struct mdt_object *mo)
109 spin_lock(&info->mti_mdt->mdt_ioepoch_lock);
110 if (info->mti_epoch->ioepoch == mo->mot_ioepoch) {
114 spin_unlock(&info->mti_mdt->mdt_ioepoch_lock);
117 /* Open the epoch. Epoch open is allowed if @writecount is not negative.
118 * The epoch and writecount handling is performed under the mdt_ioepoch_lock. */
119 int mdt_epoch_open(struct mdt_thread_info *info, struct mdt_object *o)
121 struct mdt_device *mdt = info->mti_mdt;
126 if (!S_ISREG(lu_object_attr(&o->mot_obj.mo_lu)))
129 spin_lock(&mdt->mdt_ioepoch_lock);
130 if (mdt_epoch_opened(o)) {
131 /* Epoch continues even if there is no writers yet. */
132 CDEBUG(D_INODE, "continue epoch "LPU64" for "DFID"\n",
133 o->mot_ioepoch, PFID(mdt_object_fid(o)));
135 if (info->mti_replayepoch > mdt->mdt_ioepoch)
136 mdt->mdt_ioepoch = info->mti_replayepoch;
139 o->mot_ioepoch = info->mti_replayepoch ?
140 info->mti_replayepoch : mdt->mdt_ioepoch;
141 CDEBUG(D_INODE, "starting epoch "LPU64" for "DFID"\n",
142 mdt->mdt_ioepoch, PFID(mdt_object_fid(o)));
146 spin_unlock(&mdt->mdt_ioepoch_lock);
148 /* Cancel Size-on-MDS attributes on clients if not truncate.
149 * In the later case, mdt_reint_setattr will do it. */
150 if (cancel && (info->mti_rr.rr_fid1 != NULL)) {
151 struct mdt_lock_handle *lh = &info->mti_lh[MDT_LH_CHILD];
152 mdt_lock_reg_init(lh, LCK_EX);
153 rc = mdt_object_lock(info, o, lh, MDS_INODELOCK_UPDATE,
156 mdt_object_unlock(info, o, lh, 1);
161 /* Update the on-disk attributes if needed and re-enable Size-on-MDS caching. */
162 static int mdt_sizeonmds_update(struct mdt_thread_info *info,
163 struct mdt_object *o)
167 CDEBUG(D_INODE, "Closing epoch "LPU64" on "DFID". Count %d\n",
168 o->mot_ioepoch, PFID(mdt_object_fid(o)), o->mot_epochcount);
170 if (info->mti_attr.ma_attr.la_valid & LA_SIZE)
171 /* Do Size-on-MDS attribute update.
172 * Size-on-MDS is re-enabled inside. */
173 RETURN(mdt_attr_set(info, o, 0));
175 mdt_sizeonmds_enable(info, o);
180 * Returns 1 if epoch does not close.
181 * Returns 0 if epoch closes.
182 * Returns -EAGAIN if epoch closes but an Size-on-MDS Update is still needed
183 * from the client. */
184 static int mdt_epoch_close(struct mdt_thread_info *info, struct mdt_object *o)
186 int eviction = (mdt_info_req(info) == NULL ? 1 : 0);
187 struct lu_attr *la = &info->mti_attr.ma_attr;
193 if (!S_ISREG(lu_object_attr(&o->mot_obj.mo_lu)))
196 spin_lock(&info->mti_mdt->mdt_ioepoch_lock);
198 /* Epoch closes only if client tells about it or eviction occures. */
199 if (eviction || (info->mti_epoch->flags & MF_EPOCH_CLOSE)) {
200 LASSERT(o->mot_epochcount);
203 CDEBUG(D_INODE, "Closing epoch "LPU64" on "DFID". Count %d\n",
204 o->mot_ioepoch, PFID(mdt_object_fid(o)),
208 achange = (info->mti_epoch->flags & MF_SOM_CHANGE);
211 if (!eviction && !mdt_epoch_opened(o)) {
212 /* Epoch ends. Is an Size-on-MDS update needed? */
213 if (o->mot_flags & MF_SOM_CHANGE) {
214 /* Some previous writer changed the attribute.
215 * Do not believe to the current Size-on-MDS
216 * update, re-ask client. */
218 } else if (!(la->la_valid & LA_SIZE) && achange) {
219 /* Attributes were changed by the last writer
220 * only but no Size-on-MDS update is received.*/
225 if (achange || eviction)
226 o->mot_flags |= MF_SOM_CHANGE;
229 opened = mdt_epoch_opened(o);
230 spin_unlock(&info->mti_mdt->mdt_ioepoch_lock);
232 /* XXX: if eviction occured, do nothing yet. */
233 if ((rc == 0) && !opened && !eviction) {
234 /* Epoch ends and wanted Size-on-MDS update is obtained. */
235 rc = mdt_sizeonmds_update(info, o);
240 int mdt_write_get(struct mdt_device *mdt, struct mdt_object *o)
244 spin_lock(&mdt->mdt_ioepoch_lock);
245 if (o->mot_writecount < 0)
249 spin_unlock(&mdt->mdt_ioepoch_lock);
253 static void mdt_write_put(struct mdt_device *mdt, struct mdt_object *o)
256 spin_lock(&mdt->mdt_ioepoch_lock);
258 spin_unlock(&mdt->mdt_ioepoch_lock);
262 static int mdt_write_deny(struct mdt_device *mdt, struct mdt_object *o)
266 spin_lock(&mdt->mdt_ioepoch_lock);
267 if (o->mot_writecount > 0)
271 spin_unlock(&mdt->mdt_ioepoch_lock);
275 static void mdt_write_allow(struct mdt_device *mdt, struct mdt_object *o)
278 spin_lock(&mdt->mdt_ioepoch_lock);
280 spin_unlock(&mdt->mdt_ioepoch_lock);
284 /* there can be no real transaction so prepare the fake one */
285 static void mdt_open_transno(struct mdt_thread_info* info)
287 struct mdt_device *mdt = info->mti_mdt;
288 struct ptlrpc_request *req = mdt_info_req(info);
290 spin_lock(&mdt->mdt_transno_lock);
291 if (info->mti_transno == 0) {
292 info->mti_transno = ++ mdt->mdt_last_transno;
294 /* should be replay */
295 if (info->mti_transno > mdt->mdt_last_transno)
296 mdt->mdt_last_transno = info->mti_transno;
298 spin_unlock(&mdt->mdt_transno_lock);
300 CDEBUG(D_INODE, "open only: transno = %llu, last_committed = %llu\n",
302 req->rq_export->exp_obd->obd_last_committed);
304 req->rq_transno = info->mti_transno;
305 lustre_msg_set_transno(req->rq_repmsg, info->mti_transno);
306 lustre_msg_set_last_xid(req->rq_repmsg, req->rq_xid);
309 static int mdt_mfd_open(struct mdt_thread_info *info,
310 struct mdt_object *p,
311 struct mdt_object *o,
314 struct ldlm_reply *rep)
316 struct ptlrpc_request *req = mdt_info_req(info);
317 struct mdt_export_data *med = &req->rq_export->exp_mdt_data;
318 struct mdt_device *mdt = info->mti_mdt;
319 struct md_attr *ma = &info->mti_attr;
320 struct lu_attr *la = &ma->ma_attr;
321 struct mdt_file_data *mfd;
322 struct mdt_body *repbody;
324 int isreg, isdir, islnk;
328 LASSERT(ma->ma_valid & MA_INODE);
330 repbody = req_capsule_server_get(&info->mti_pill, &RMF_MDT_BODY);
332 isreg = S_ISREG(la->la_mode);
333 isdir = S_ISDIR(la->la_mode);
334 islnk = S_ISLNK(la->la_mode);
335 mdt_pack_attr2body(info, repbody, la, mdt_object_fid(o));
337 if (med->med_rmtclient) {
338 void *buf = req_capsule_server_get(&info->mti_pill, &RMF_ACL);
340 rc = mdt_pack_remote_perm(info, o, buf);
342 repbody->valid &= ~OBD_MD_FLRMTPERM;
343 repbody->aclsize = 0;
345 repbody->valid |= OBD_MD_FLRMTPERM;
346 repbody->aclsize = sizeof(struct mdt_remote_perm);
349 #ifdef CONFIG_FS_POSIX_ACL
350 else if (req->rq_export->exp_connect_flags & OBD_CONNECT_ACL) {
351 const struct lu_env *env = info->mti_env;
352 struct md_object *next = mdt_object_child(o);
353 struct lu_buf *buf = &info->mti_buf;
355 buf->lb_buf = req_capsule_server_get(&info->mti_pill, &RMF_ACL);
356 buf->lb_len = req_capsule_get_size(&info->mti_pill, &RMF_ACL,
358 if (buf->lb_len > 0) {
359 rc = mo_xattr_get(env, next, buf,
360 XATTR_NAME_ACL_ACCESS);
362 if (rc == -ENODATA) {
363 repbody->aclsize = 0;
364 repbody->valid |= OBD_MD_FLACL;
366 } else if (rc == -EOPNOTSUPP) {
369 CERROR("got acl size: %d\n", rc);
372 repbody->aclsize = rc;
373 repbody->valid |= OBD_MD_FLACL;
380 if (mdt->mdt_opts.mo_mds_capa) {
381 struct lustre_capa *capa;
383 capa = req_capsule_server_get(&info->mti_pill, &RMF_CAPA1);
385 capa->lc_opc = CAPA_OPC_MDS_DEFAULT;
386 rc = mo_capa_get(info->mti_env, mdt_object_child(o), capa, 0);
389 repbody->valid |= OBD_MD_FLMDSCAPA;
391 if (mdt->mdt_opts.mo_oss_capa &&
392 S_ISREG(lu_object_attr(&o->mot_obj.mo_lu))) {
393 struct lustre_capa *capa;
395 capa = req_capsule_server_get(&info->mti_pill, &RMF_CAPA2);
397 capa->lc_opc = CAPA_OPC_OSS_DEFAULT | capa_open_opc(flags);
398 rc = mo_capa_get(info->mti_env, mdt_object_child(o), capa, 0);
401 repbody->valid |= OBD_MD_FLOSSCAPA;
404 /* if we are following a symlink, don't open; and
405 * do not return open handle for special nodes as client required
407 if (islnk || (!isreg && !isdir &&
408 (req->rq_export->exp_connect_flags & OBD_CONNECT_NODEVOH))) {
409 lustre_msg_set_transno(req->rq_repmsg, 0);
413 mdt_set_disposition(info, rep, DISP_OPEN_OPEN);
414 /* we need to return the existing object's fid back, so it is done
415 * here, after preparing the reply */
416 if (!created && (flags & MDS_OPEN_EXCL) && (flags & MDS_OPEN_CREAT))
419 /* This can't be done earlier, we need to return reply body */
421 if (flags & (MDS_OPEN_CREAT | FMODE_WRITE)) {
422 /* we are trying to create or
423 * write an existing dir. */
426 } else if (flags & MDS_OPEN_DIRECTORY)
429 if (MDT_FAIL_CHECK(OBD_FAIL_MDS_OPEN_CREATE)) {
430 obd_fail_loc = OBD_FAIL_LDLM_REPLY | OBD_FAIL_ONCE;
434 if (isreg && !(ma->ma_valid & MA_LOV)) {
436 * No EA, check whether it is will set regEA and dirEA since in
437 * above attr get, these size might be zero, so reset it, to
438 * retrieve the MD after create obj.
440 ma->ma_lmm_size = req_capsule_get_size(&info->mti_pill,
443 /* in replay case, p == NULL */
444 rc = mdt_create_data(info, p, o);
449 CDEBUG(D_INODE, "after open, ma_valid bit = "LPX64" lmm_size = %d\n",
450 ma->ma_valid, ma->ma_lmm_size);
452 if (ma->ma_valid & MA_LOV) {
453 LASSERT(ma->ma_lmm_size != 0);
454 repbody->eadatasize = ma->ma_lmm_size;
456 repbody->valid |= OBD_MD_FLDIREA;
458 repbody->valid |= OBD_MD_FLEASIZE;
462 if (lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT) {
463 spin_lock(&med->med_open_lock);
464 list_for_each(t, &med->med_open_head) {
465 mfd = list_entry(t, struct mdt_file_data, mfd_list);
466 if (mfd->mfd_xid == req->rq_xid) {
471 spin_unlock(&med->med_open_lock);
474 repbody->handle.cookie = mfd->mfd_handle.h_cookie;
479 if (flags & FMODE_WRITE) {
480 rc = mdt_write_get(info->mti_mdt, o);
482 mdt_epoch_open(info, o);
483 repbody->ioepoch = o->mot_ioepoch;
485 } else if (flags & MDS_FMODE_EXEC) {
486 rc = mdt_write_deny(info->mti_mdt, o);
491 rc = mo_open(info->mti_env, mdt_object_child(o),
492 created ? flags | MDS_OPEN_CREATED : flags);
499 /* keep a reference on this object for this open,
500 * and is released by mdt_mfd_close() */
501 mdt_object_get(info->mti_env, o);
504 mfd->mfd_mode = flags;
506 mfd->mfd_xid = req->rq_xid;
508 spin_lock(&med->med_open_lock);
509 list_add(&mfd->mfd_list, &med->med_open_head);
510 spin_unlock(&med->med_open_lock);
512 repbody->handle.cookie = mfd->mfd_handle.h_cookie;
513 mdt_open_transno(info);
520 extern void mdt_req_from_mcd(struct ptlrpc_request *req,
521 struct mdt_client_data *mcd);
523 void mdt_reconstruct_open(struct mdt_thread_info *info,
524 struct mdt_lock_handle *lhc)
526 const struct lu_env *env = info->mti_env;
527 struct mdt_device *mdt = info->mti_mdt;
528 struct req_capsule *pill = &info->mti_pill;
529 struct ptlrpc_request *req = mdt_info_req(info);
530 struct mdt_export_data *med = &req->rq_export->exp_mdt_data;
531 struct mdt_client_data *mcd = med->med_mcd;
532 struct md_attr *ma = &info->mti_attr;
533 struct mdt_reint_record *rr = &info->mti_rr;
534 __u32 flags = info->mti_spec.sp_cr_flags;
535 struct ldlm_reply *ldlm_rep;
536 struct mdt_object *parent;
537 struct mdt_object *child;
538 struct mdt_body *repbody;
542 LASSERT(pill->rc_fmt == &RQF_LDLM_INTENT_OPEN);
543 ldlm_rep = req_capsule_server_get(&info->mti_pill, &RMF_DLM_REP);
544 repbody = req_capsule_server_get(&info->mti_pill, &RMF_MDT_BODY);
546 ma->ma_lmm = req_capsule_server_get(&info->mti_pill, &RMF_MDT_MD);
547 ma->ma_lmm_size = req_capsule_get_size(&info->mti_pill, &RMF_MDT_MD,
549 ma->ma_need = MA_INODE | MA_LOV;
552 mdt_req_from_mcd(req, med->med_mcd);
553 mdt_set_disposition(info, ldlm_rep, mcd->mcd_last_data);
555 CERROR("This is reconstruct open: disp="LPX64", result=%d\n",
556 ldlm_rep->lock_policy_res1, req->rq_status);
558 if (mdt_get_disposition(ldlm_rep, DISP_OPEN_CREATE) &&
559 req->rq_status != 0) {
560 /* We did not create successfully, return error to client. */
561 mdt_shrink_reply(info, DLM_REPLY_REC_OFF + 1, 1, 1);
562 GOTO(out, rc = req->rq_status);
565 if (mdt_get_disposition(ldlm_rep, DISP_OPEN_CREATE)) {
567 * We failed after creation, but we do not know in which step
568 * we failed. So try to check the child object.
570 parent = mdt_object_find(env, mdt, rr->rr_fid1);
571 LASSERT(!IS_ERR(parent));
573 child = mdt_object_find(env, mdt, rr->rr_fid2);
574 LASSERT(!IS_ERR(child));
576 rc = mdt_object_exists(child);
578 struct md_object *next;
579 next = mdt_object_child(child);
580 rc = mo_attr_get(env, next, ma);
582 rc = mdt_mfd_open(info, parent, child,
585 /* the child object was created on remote server */
586 repbody->fid1 = *rr->rr_fid2;
587 repbody->valid |= (OBD_MD_FLID | OBD_MD_MDS);
589 } else if (rc == 0) {
590 /* the child does not exist, we should do regular open */
591 mdt_object_put(env, parent);
592 mdt_object_put(env, child);
593 GOTO(regular_open, 0);
595 mdt_object_put(env, parent);
596 mdt_object_put(env, child);
597 mdt_shrink_reply(info, DLM_REPLY_REC_OFF + 1, 1, 1);
601 /* We did not try to create, so we are a pure open */
602 rc = mdt_reint_open(info, lhc);
608 lustre_msg_set_status(req->rq_repmsg, req->rq_status);
611 static int mdt_open_by_fid(struct mdt_thread_info* info,
612 struct ldlm_reply *rep)
614 const struct lu_env *env = info->mti_env;
615 struct mdt_device *mdt = info->mti_mdt;
616 struct md_device *next = mdt->mdt_child;
617 __u32 flags = info->mti_spec.sp_cr_flags;
618 struct mdt_reint_record *rr = &info->mti_rr;
619 struct md_attr *ma = &info->mti_attr;
620 struct mdt_object *o;
624 o = mdt_object_find(info->mti_env, info->mti_mdt, rr->rr_fid2);
626 RETURN(rc = PTR_ERR(o));
628 /* FIXME: capability in replay open request might have expired, disable
629 * capability check for this kind of request. security hole?
631 if (mdt->mdt_opts.mo_mds_capa)
632 mdt->mdt_child->md_ops->mdo_init_capa_ctxt(env, next, 0, 0, 0,
635 rc = mdt_object_exists(o);
637 mdt_set_disposition(info, rep, (DISP_IT_EXECD |
641 rc = mo_attr_get(env, mdt_object_child(o), ma);
643 rc = mdt_mfd_open(info, NULL, o, flags, 0, rep);
644 } else if (rc == 0) {
647 /* the child object was created on remote server */
648 struct mdt_body *repbody;
649 repbody = req_capsule_server_get(&info->mti_pill, &RMF_MDT_BODY);
650 repbody->fid1 = *rr->rr_fid2;
651 repbody->valid |= (OBD_MD_FLID | OBD_MD_MDS);
655 if (mdt->mdt_opts.mo_mds_capa)
656 mdt->mdt_child->md_ops->mdo_init_capa_ctxt(env, next,
657 mdt->mdt_opts.mo_mds_capa,
658 mdt->mdt_capa_timeout,
662 mdt_object_put(info->mti_env, o);
666 int mdt_pin(struct mdt_thread_info* info)
669 RETURN(err_serious(-EOPNOTSUPP));
672 /* Cross-ref request. Currently it can only be a pure open (w/o create) */
673 static int mdt_cross_open(struct mdt_thread_info* info,
674 const struct lu_fid *fid,
675 struct ldlm_reply *rep, __u32 flags)
677 struct md_attr *ma = &info->mti_attr;
678 struct mdt_object *o;
682 o = mdt_object_find(info->mti_env, info->mti_mdt, fid);
684 RETURN(rc = PTR_ERR(o));
686 rc = mdt_object_exists(o);
688 mdt_set_capainfo(info, 0, fid, BYPASS_CAPA);
689 rc = mo_attr_get(info->mti_env, mdt_object_child(o), ma);
691 rc = mdt_mfd_open(info, NULL, o, flags, 0, rep);
692 } else if (rc == 0) {
694 * FIXME: something wrong here, lookup was positive but there is
697 CERROR("Cross-ref object doesn't exist!\n");
700 /* FIXME: something wrong here, the object is on another MDS! */
701 CERROR("The object isn't on this server! FLD error?\n");
704 mdt_object_put(info->mti_env, o);
708 int mdt_reint_open(struct mdt_thread_info *info, struct mdt_lock_handle *lhc)
710 struct mdt_device *mdt = info->mti_mdt;
711 struct ptlrpc_request *req = mdt_info_req(info);
712 struct mdt_object *parent;
713 struct mdt_object *child;
714 struct mdt_lock_handle *lh;
715 struct ldlm_reply *ldlm_rep;
716 struct mdt_body *repbody;
717 struct lu_fid *child_fid = &info->mti_tmp_fid1;
718 struct md_attr *ma = &info->mti_attr;
719 struct lu_attr *la = &ma->ma_attr;
720 __u32 create_flags = info->mti_spec.sp_cr_flags;
721 struct mdt_reint_record *rr = &info->mti_rr;
726 OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_PAUSE_OPEN | OBD_FAIL_ONCE,
727 (obd_timeout + 1) / 4);
729 repbody = req_capsule_server_get(&info->mti_pill, &RMF_MDT_BODY);
731 ma->ma_lmm = req_capsule_server_get(&info->mti_pill, &RMF_MDT_MD);
732 ma->ma_lmm_size = req_capsule_get_size(&info->mti_pill, &RMF_MDT_MD,
734 ma->ma_need = MA_INODE | MA_LOV;
737 LASSERT(info->mti_pill.rc_fmt == &RQF_LDLM_INTENT_OPEN);
738 ldlm_rep = req_capsule_server_get(&info->mti_pill, &RMF_DLM_REP);
740 /* TODO: JOIN file */
741 if (create_flags & MDS_OPEN_JOIN_FILE) {
742 CERROR("JOIN file will be supported soon\n");
743 GOTO(out, result = err_serious(-EOPNOTSUPP));
746 CDEBUG(D_INODE, "I am going to open "DFID"/(%s->"DFID") "
747 "cr_flag=0%o mode=0%06o msg_flag=0x%x\n",
748 PFID(rr->rr_fid1), rr->rr_name,
749 PFID(rr->rr_fid2), create_flags,
750 la->la_mode, lustre_msg_get_flags(req->rq_reqmsg));
752 if (lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY) {
753 /* This is a replay request. */
754 result = mdt_open_by_fid(info, ldlm_rep);
756 if (result != -ENOENT)
760 * We didn't find the correct object, so we need to re-create it
761 * via a regular replay.
763 if (!(create_flags & MDS_OPEN_CREAT)) {
764 DEBUG_REQ(D_ERROR, req,"OPEN & CREAT not in open replay.");
765 GOTO(out, result = -EFAULT);
767 CDEBUG(D_INFO, "Open replay did find object, continue as "
771 if (MDT_FAIL_CHECK(OBD_FAIL_MDS_OPEN_PACK))
772 GOTO(out, result = err_serious(-ENOMEM));
774 mdt_set_disposition(info, ldlm_rep,
775 (DISP_IT_EXECD | DISP_LOOKUP_EXECD));
777 if (rr->rr_name[0] == 0) {
778 /* this is cross-ref open */
779 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
780 result = mdt_cross_open(info, rr->rr_fid1, ldlm_rep,
785 lh = &info->mti_lh[MDT_LH_PARENT];
786 if (!(create_flags & MDS_OPEN_CREAT)) {
787 mdt_lock_pdo_init(lh, LCK_PR, rr->rr_name,
790 mdt_lock_pdo_init(lh, LCK_PW, rr->rr_name,
793 parent = mdt_object_find_lock(info, rr->rr_fid1, lh,
794 MDS_INODELOCK_UPDATE);
796 GOTO(out, result = PTR_ERR(parent));
798 result = mdo_lookup(info->mti_env, mdt_object_child(parent),
799 rr->rr_name, child_fid);
800 if (result != 0 && result != -ENOENT && result != -ESTALE)
801 GOTO(out_parent, result);
803 if (result == -ENOENT || result == -ESTALE) {
804 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
805 if (result == -ESTALE) {
807 * ESTALE means the parent is a dead(unlinked) dir, so
808 * it should return -ENOENT to in accordance with the
809 * original mds implementaion.
811 GOTO(out_parent, result = -ENOENT);
813 if (!(create_flags & MDS_OPEN_CREAT))
814 GOTO(out_parent, result);
815 *child_fid = *info->mti_rr.rr_fid2;
816 /* new object will be created. see the following */
819 * Check for O_EXCL is moved to the mdt_mfd_open, we need to
820 * return FID back in that case.
822 mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
825 LASSERT(fid_is_sane(child_fid));
826 child = mdt_object_find(info->mti_env, mdt, child_fid);
828 GOTO(out_parent, result = PTR_ERR(child));
830 mdt_set_capainfo(info, 1, child_fid, BYPASS_CAPA);
831 if (result == -ENOENT) {
832 /* Not found and with MDS_OPEN_CREAT: let's create it. */
833 mdt_set_disposition(info, ldlm_rep, DISP_OPEN_CREATE);
835 /* Let lower layers know what is lock mode on directory. */
836 info->mti_spec.sp_cr_mode =
837 mdt_dlm_mode2mdl_mode(lh->mlh_pdo_mode);
838 result = mdo_create(info->mti_env,
839 mdt_object_child(parent),
841 mdt_object_child(child),
844 if (result == -ERESTART) {
845 mdt_clear_disposition(info, ldlm_rep, DISP_OPEN_CREATE);
846 GOTO(out_child, result);
850 GOTO(out_child, result);
854 /* We have to get attr & lov ea for this object */
855 result = mo_attr_get(info->mti_env, mdt_object_child(child),
858 * The object is on remote node, return its FID for remote open.
860 if (result == -EREMOTE) {
864 * Check if this lock already was sent to client and
865 * this is resent case. For resent case do not take lock
866 * again, use what is already granted.
868 LASSERT(lhc != NULL);
870 if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
871 struct ldlm_lock *lock;
873 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) &
876 lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
878 CERROR("Invalid lock handle "LPX64"\n",
879 lhc->mlh_reg_lh.cookie);
882 LASSERT(fid_res_name_eq(mdt_object_fid(child),
883 &lock->l_resource->lr_name));
887 mdt_lock_handle_init(lhc);
888 mdt_lock_reg_init(lhc, LCK_PR);
890 rc = mdt_object_lock(info, child, lhc,
891 MDS_INODELOCK_LOOKUP,
894 repbody->fid1 = *mdt_object_fid(child);
895 repbody->valid |= (OBD_MD_FLID | OBD_MD_MDS);
898 GOTO(out_child, result);
902 /* Try to open it now. */
903 result = mdt_mfd_open(info, parent, child, create_flags,
905 GOTO(finish_open, result);
908 if (result != 0 && created) {
912 ma->ma_cookie_size = 0;
913 rc2 = mdo_unlink(info->mti_env,
914 mdt_object_child(parent),
915 mdt_object_child(child),
919 CERROR("error in cleanup of open");
922 mdt_object_put(info->mti_env, child);
924 mdt_object_unlock_put(info, parent, lh, result);
926 mdt_shrink_reply(info, DLM_REPLY_REC_OFF + 1, 1, 1);
928 lustre_msg_set_transno(req->rq_repmsg, 0);
932 #define MFD_CLOSED(mode) (((mode) & ~(FMODE_EPOCH | FMODE_SOM | \
933 FMODE_EPOCHLCK)) == FMODE_CLOSED)
935 static int mdt_mfd_closed(struct mdt_file_data *mfd)
937 return ((mfd == NULL) || MFD_CLOSED(mfd->mfd_mode));
940 int mdt_mfd_close(struct mdt_thread_info *info, struct mdt_file_data *mfd)
942 struct mdt_object *o = mfd->mfd_object;
943 struct md_object *next = mdt_object_child(o);
944 struct md_attr *ma = &info->mti_attr;
949 mode = mfd->mfd_mode;
950 mfd->mfd_mode = FMODE_CLOSED;
952 if ((mode & FMODE_WRITE) || (mode & FMODE_EPOCHLCK)) {
953 mdt_write_put(info->mti_mdt, o);
954 ret = mdt_epoch_close(info, o);
955 } else if (mode & MDS_FMODE_EXEC) {
956 mdt_write_allow(info->mti_mdt, o);
957 } else if (mode & FMODE_EPOCH) {
958 ret = mdt_epoch_close(info, o);
961 ma->ma_need |= MA_INODE;
963 if (!MFD_CLOSED(mode))
964 rc = mo_close(info->mti_env, next, ma);
965 else if (ret == -EAGAIN)
966 rc = mo_attr_get(info->mti_env, next, ma);
968 /* If the object is unlinked, do not try to re-enable SIZEONMDS */
969 if ((ret == -EAGAIN) && (ma->ma_valid & MA_INODE) &&
970 (ma->ma_attr.la_nlink == 0))
975 if ((ret == -EAGAIN) || (ret == 1)) {
976 struct mdt_export_data *med;
977 /* The epoch has not closed or Size-on-MDS update is needed.
978 * Put mfd back into the list. */
979 mfd->mfd_mode = (ret == 1 ? FMODE_EPOCH : FMODE_SOM);
981 LASSERT(mdt_info_req(info));
982 med = &mdt_info_req(info)->rq_export->exp_mdt_data;
983 spin_lock(&med->med_open_lock);
984 list_add(&mfd->mfd_list, &med->med_open_head);
985 class_handle_hash_back(&mfd->mfd_handle);
986 spin_unlock(&med->med_open_lock);
990 CDEBUG(D_INODE, "Size-on-MDS attribute update is "
991 "needed on "DFID"\n", PFID(mdt_object_fid(o)));
995 mdt_object_put(info->mti_env, o);
998 RETURN(rc ? rc : ret);
1001 int mdt_close(struct mdt_thread_info *info)
1003 struct mdt_export_data *med;
1004 struct mdt_file_data *mfd;
1005 struct mdt_object *o;
1006 struct md_attr *ma = &info->mti_attr;
1007 struct mdt_body *repbody = NULL;
1011 /* Close may come with the Size-on-MDS update. Unpack it. */
1012 rc = mdt_close_unpack(info);
1014 RETURN(err_serious(rc));
1016 LASSERT(info->mti_epoch);
1018 req_capsule_set_size(&info->mti_pill, &RMF_MDT_MD, RCL_SERVER,
1019 info->mti_mdt->mdt_max_mdsize);
1020 req_capsule_set_size(&info->mti_pill, &RMF_LOGCOOKIES, RCL_SERVER,
1021 info->mti_mdt->mdt_max_cookiesize);
1022 rc = req_capsule_pack(&info->mti_pill);
1023 /* Continue to close handle even if we can not pack reply */
1025 repbody = req_capsule_server_get(&info->mti_pill,
1027 ma->ma_lmm = req_capsule_server_get(&info->mti_pill,
1029 ma->ma_lmm_size = req_capsule_get_size(&info->mti_pill,
1032 ma->ma_cookie = req_capsule_server_get(&info->mti_pill,
1034 ma->ma_cookie_size = req_capsule_get_size(&info->mti_pill,
1037 ma->ma_need = MA_INODE | MA_LOV | MA_COOKIE;
1039 repbody->eadatasize = 0;
1040 repbody->aclsize = 0;
1042 rc = err_serious(rc);
1044 med = &mdt_info_req(info)->rq_export->exp_mdt_data;
1046 spin_lock(&med->med_open_lock);
1047 mfd = mdt_handle2mfd(&(info->mti_epoch->handle));
1048 if (mdt_mfd_closed(mfd)) {
1049 spin_unlock(&med->med_open_lock);
1050 CDEBUG(D_INODE, "no handle for file close: fid = "DFID
1051 ": cookie = "LPX64"\n", PFID(info->mti_rr.rr_fid1),
1052 info->mti_epoch->handle.cookie);
1053 rc = err_serious(-ESTALE);
1055 class_handle_unhash(&mfd->mfd_handle);
1056 list_del_init(&mfd->mfd_list);
1057 spin_unlock(&med->med_open_lock);
1059 /* Do not lose object before last unlink. */
1060 o = mfd->mfd_object;
1061 mdt_object_get(info->mti_env, o);
1062 ret = mdt_mfd_close(info, mfd);
1063 if (repbody != NULL)
1064 rc = mdt_handle_last_unlink(info, o, ma);
1065 mdt_object_put(info->mti_env, o);
1067 if (repbody != NULL)
1068 mdt_shrink_reply(info, REPLY_REC_OFF + 1, 0, 0);
1070 if (MDT_FAIL_CHECK(OBD_FAIL_MDS_CLOSE_PACK))
1071 RETURN(err_serious(-ENOMEM));
1073 RETURN(rc ? rc : ret);
1076 int mdt_done_writing(struct mdt_thread_info *info)
1078 struct mdt_body *repbody = NULL;
1079 struct mdt_export_data *med;
1080 struct mdt_file_data *mfd;
1084 rc = req_capsule_pack(&info->mti_pill);
1086 RETURN(err_serious(rc));
1088 repbody = req_capsule_server_get(&info->mti_pill,
1090 repbody->eadatasize = 0;
1091 repbody->aclsize = 0;
1093 /* Done Writing may come with the Size-on-MDS update. Unpack it. */
1094 rc = mdt_close_unpack(info);
1096 RETURN(err_serious(rc));
1098 med = &mdt_info_req(info)->rq_export->exp_mdt_data;
1099 spin_lock(&med->med_open_lock);
1100 mfd = mdt_handle2mfd(&(info->mti_epoch->handle));
1102 spin_unlock(&med->med_open_lock);
1103 CDEBUG(D_INODE, "no handle for file close: fid = "DFID
1104 ": cookie = "LPX64"\n", PFID(info->mti_rr.rr_fid1),
1105 info->mti_epoch->handle.cookie);
1108 LASSERT((mfd->mfd_mode == FMODE_EPOCH) ||
1109 (mfd->mfd_mode == FMODE_EPOCHLCK));
1110 class_handle_unhash(&mfd->mfd_handle);
1111 list_del_init(&mfd->mfd_list);
1112 spin_unlock(&med->med_open_lock);
1114 /* Set EPOCH CLOSE flag if not set by client. */
1115 info->mti_epoch->flags |= MF_EPOCH_CLOSE;
1116 info->mti_attr.ma_valid = 0;
1117 rc = mdt_mfd_close(info, mfd);