1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
4 * Copyright (c) 2003 Cluster File Systems, Inc.
5 * Author: Peter Braam <braam@clusterfs.com>
6 * Author: Andreas Dilger <adilger@clusterfs.com>
7 * Author: Phil Schwan <phil@clusterfs.com>
8 * Author: Mike Shaver <shaver@clusterfs.com>
10 * This file is part of Lustre, http://www.lustre.org.
12 * Lustre is free software; you can redistribute it and/or
13 * modify it under the terms of version 2 of the GNU General Public
14 * License as published by the Free Software Foundation.
16 * Lustre is distributed in the hope that it will be useful,
17 * but WITHOUT ANY WARRANTY; without even the implied warranty of
18 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
19 * GNU General Public License for more details.
21 * You should have received a copy of the GNU General Public License
22 * along with Lustre; if not, write to the Free Software
23 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
27 # define EXPORT_SYMTAB
29 #define DEBUG_SUBSYSTEM S_MDS
31 #include <linux/version.h>
32 #include <linux/module.h>
33 #include <linux/init.h>
34 #include <linux/version.h>
35 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,5,0))
36 # include <linux/buffer_head.h>
37 # include <linux/workqueue.h>
39 # include <linux/locks.h>
42 #include <linux/obd_class.h>
43 #include <linux/lustre_mds.h>
44 #include <linux/obd_lov.h>
45 #include <linux/lustre_fsfilt.h>
46 #include <linux/lprocfs_status.h>
48 #include "mds_internal.h"
50 /* Exported function from this file are:
52 * mds_open - called by the intent handler
53 * mds_close - an rpc handling function
54 * mds_pin - an rpc handling function - which will go away
55 * mds_mfd_close - for force closing files when a client dies
59 * MDS file data handling: file data holds a handle for a file opened
63 static void mds_mfd_addref(void *mfdp)
65 struct mds_file_data *mfd = mfdp;
67 atomic_inc(&mfd->mfd_refcount);
68 CDEBUG(D_INFO, "GETting mfd %p : new refcount %d\n", mfd,
69 atomic_read(&mfd->mfd_refcount));
72 struct mds_file_data *mds_mfd_new(void)
74 struct mds_file_data *mfd;
76 OBD_ALLOC(mfd, sizeof *mfd);
78 CERROR("mds: out of memory\n");
82 atomic_set(&mfd->mfd_refcount, 2);
84 INIT_LIST_HEAD(&mfd->mfd_handle.h_link);
85 class_handle_hash(&mfd->mfd_handle, mds_mfd_addref);
90 static struct mds_file_data *mds_handle2mfd(struct lustre_handle *handle)
93 LASSERT(handle != NULL);
94 RETURN(class_handle2object(handle->cookie));
97 static void mds_mfd_put(struct mds_file_data *mfd)
99 CDEBUG(D_INFO, "PUTting mfd %p : new refcount %d\n", mfd,
100 atomic_read(&mfd->mfd_refcount) - 1);
101 LASSERT(atomic_read(&mfd->mfd_refcount) > 0 &&
102 atomic_read(&mfd->mfd_refcount) < 0x5a5a);
103 if (atomic_dec_and_test(&mfd->mfd_refcount)) {
104 LASSERT(list_empty(&mfd->mfd_handle.h_link));
105 OBD_FREE(mfd, sizeof *mfd);
109 static void mds_mfd_destroy(struct mds_file_data *mfd)
111 class_handle_unhash(&mfd->mfd_handle);
116 /* Caller must hold mds->mds_epoch_sem */
117 static int mds_alloc_filterdata(struct inode *inode)
119 LASSERT(inode->i_filterdata == NULL);
120 OBD_ALLOC(inode->i_filterdata, sizeof(struct mds_filter_data));
121 if (inode->i_filterdata == NULL)
123 LASSERT(igrab(inode) == inode);
127 /* Caller must hold mds->mds_epoch_sem */
128 static void mds_free_filterdata(struct inode *inode)
130 LASSERT(inode->i_filterdata != NULL);
131 OBD_FREE(inode->i_filterdata, sizeof(struct mds_filter_data));
132 inode->i_filterdata = NULL;
136 /* Write access to a file: executors cause a negative count,
137 * writers a positive count. The semaphore is needed to perform
138 * a check for the sign and then increment or decrement atomically.
140 * This code is closely tied to the allocation of the d_fsdata and the
141 * MDS epoch, so we use the same semaphore for the whole lot.
143 * We could use a different semaphore for each file, if it ever shows
144 * up in a profile, which it won't.
146 * epoch argument is nonzero during recovery */
147 static int mds_get_write_access(struct mds_obd *mds, struct inode *inode,
152 down(&mds->mds_epoch_sem);
154 if (atomic_read(&inode->i_writecount) < 0) {
155 up(&mds->mds_epoch_sem);
160 if (MDS_FILTERDATA(inode) && MDS_FILTERDATA(inode)->io_epoch != 0) {
161 CDEBUG(D_INODE, "continuing MDS epoch "LPU64" for ino %lu/%u\n",
162 MDS_FILTERDATA(inode)->io_epoch, inode->i_ino,
163 inode->i_generation);
167 if (inode->i_filterdata == NULL)
168 mds_alloc_filterdata(inode);
169 if (inode->i_filterdata == NULL) {
173 if (epoch > mds->mds_io_epoch)
174 mds->mds_io_epoch = epoch;
177 MDS_FILTERDATA(inode)->io_epoch = mds->mds_io_epoch;
178 CDEBUG(D_INODE, "starting MDS epoch "LPU64" for ino %lu/%u\n",
179 mds->mds_io_epoch, inode->i_ino, inode->i_generation);
182 atomic_inc(&inode->i_writecount);
183 up(&mds->mds_epoch_sem);
187 /* Returns EAGAIN if the client needs to get size and/or cookies and close
188 * again -- which is never true if the file is about to be unlinked. Otherwise
189 * returns the number of remaining writers. */
190 static int mds_put_write_access(struct mds_obd *mds, struct inode *inode,
191 struct mds_body *body, int unlinking)
196 down(&mds->mds_epoch_sem);
197 atomic_dec(&inode->i_writecount);
198 rc = atomic_read(&inode->i_writecount);
202 if (!unlinking && !(body->valid & OBD_MD_FLSIZE))
203 GOTO(out, rc = EAGAIN);
205 mds_free_filterdata(inode);
207 up(&mds->mds_epoch_sem);
211 static int mds_deny_write_access(struct mds_obd *mds, struct inode *inode)
214 down(&mds->mds_epoch_sem);
215 if (atomic_read(&inode->i_writecount) > 0) {
216 up(&mds->mds_epoch_sem);
219 atomic_dec(&inode->i_writecount);
220 up(&mds->mds_epoch_sem);
224 static void mds_allow_write_access(struct inode *inode)
227 atomic_inc(&inode->i_writecount);
230 int mds_query_write_access(struct inode *inode)
233 RETURN(atomic_read(&inode->i_writecount));
236 /* This replaces the VFS mds_dentry_open, it manages mfd and writecount */
237 static struct mds_file_data *mds_dentry_open(struct dentry *dentry,
238 struct vfsmount *mnt, int flags,
239 struct ptlrpc_request *req)
241 struct mds_export_data *med = &req->rq_export->exp_mds_data;
242 struct mds_obd *mds = mds_req2mds(req);
243 struct mds_file_data *mfd;
244 struct mds_body *body;
250 CERROR("mds: out of memory\n");
251 GOTO(cleanup_dentry, error = -ENOMEM);
254 body = lustre_msg_buf(req->rq_repmsg, 1, sizeof (*body));
256 if (flags & FMODE_WRITE) {
257 /* FIXME: in recovery, need to pass old epoch here */
258 error = mds_get_write_access(mds, dentry->d_inode, 0);
260 GOTO(cleanup_mfd, error);
261 body->io_epoch = MDS_FILTERDATA(dentry->d_inode)->io_epoch;
262 } else if (flags & FMODE_EXEC) {
263 error = mds_deny_write_access(mds, dentry->d_inode);
265 GOTO(cleanup_mfd, error);
270 /* Mark the file as open to handle open-unlink. */
271 mds_open_orphan_inc(dentry->d_inode);
273 mfd->mfd_mode = flags;
274 mfd->mfd_dentry = dentry;
275 mfd->mfd_xid = req->rq_xid;
277 spin_lock(&med->med_open_lock);
278 list_add(&mfd->mfd_list, &med->med_open_head);
279 spin_unlock(&med->med_open_lock);
282 body->handle.cookie = mfd->mfd_handle.h_cookie;
288 mds_mfd_destroy(mfd);
290 return ERR_PTR(error);
293 static void mds_objids_from_lmm(obd_id *ids, struct lov_mds_md *lmm,
294 struct lov_desc *desc)
297 for (i = 0; i < le32_to_cpu(lmm->lmm_stripe_count); i++) {
298 ids[le32_to_cpu(lmm->lmm_objects[i].l_ost_idx)] =
299 le64_to_cpu(lmm->lmm_objects[i].l_object_id);
303 /* Must be called with i_sem held */
304 static int mds_create_objects(struct ptlrpc_request *req, int offset,
305 struct mds_update_record *rec,
306 struct mds_obd *mds, struct obd_device *obd,
307 struct dentry *dchild, void **handle,
311 struct obd_trans_info oti = { 0 };
312 struct mds_body *body;
313 struct lov_stripe_md *lsm = NULL;
314 struct lov_mds_md *lmm = NULL;
315 struct inode *inode = dchild->d_inode;
317 int rc, lmm_bufsize, lmm_size;
320 if (rec->ur_flags & MDS_OPEN_DELAY_CREATE ||
321 !(rec->ur_flags & FMODE_WRITE))
324 body = lustre_msg_buf(req->rq_repmsg, 1, sizeof(*body));
326 if (!S_ISREG(inode->i_mode))
328 if (body->valid & OBD_MD_FLEASIZE)
331 OBD_ALLOC(*ids, mds->mds_lov_desc.ld_tgt_count * sizeof(**ids));
334 oti.oti_objid = *ids;
337 if(lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY) {
338 LASSERT (rec->ur_fid2->id);
339 body->valid |= OBD_MD_FLBLKSZ | OBD_MD_FLEASIZE;
340 lmm_size = rec->ur_eadatalen;
341 lmm = rec->ur_eadata;
345 *handle = fsfilt_start(obd,inode,FSFILT_OP_CREATE,NULL);
346 if (IS_ERR(*handle)) {
347 rc = PTR_ERR(*handle);
352 mds_objids_from_lmm(*ids, lmm, &mds->mds_lov_desc);
354 lmm_buf = lustre_msg_buf(req->rq_repmsg, offset, 0);
355 lmm_bufsize = req->rq_repmsg->buflens[offset];
357 LASSERT(lmm_bufsize >= lmm_size);
358 memcpy(lmm_buf, lmm, lmm_size);
359 rc = fsfilt_set_md(obd, inode, *handle, lmm, lmm_size);
361 CERROR("open replay failed to set md:%d\n", rc);
365 if (OBD_FAIL_CHECK_ONCE(OBD_FAIL_MDS_ALLOC_OBDO))
366 GOTO(out_ids, rc = -ENOMEM);
370 GOTO(out_ids, rc = -ENOMEM);
371 oa->o_mode = S_IFREG | 0600;
372 oa->o_id = inode->i_ino;
373 oa->o_gr = FILTER_GROUP_FIRST_MDS + mds->mds_num;
374 oa->o_generation = inode->i_generation;
375 oa->o_uid = 0; /* must have 0 uid / gid on OST */
377 oa->o_valid = OBD_MD_FLID | OBD_MD_FLGENER | OBD_MD_FLTYPE |
378 OBD_MD_FLMODE | OBD_MD_FLUID | OBD_MD_FLGID | OBD_MD_FLGROUP;
381 obdo_from_inode(oa, inode, OBD_MD_FLTYPE|OBD_MD_FLATIME|OBD_MD_FLMTIME|
384 if (!(rec->ur_flags & MDS_OPEN_HAS_OBJS)) {
385 /* check if things like lfs setstripe are sending us the ea */
386 if (rec->ur_flags & MDS_OPEN_HAS_EA) {
387 rc = obd_iocontrol(OBD_IOC_LOV_SETSTRIPE,
389 0, &lsm, rec->ur_eadata);
393 /* Per-directory striping default code removed, because
394 * it uses the same unnamed EA storage as the directory
395 * striping for CMD. -p */
397 LASSERT(oa->o_gr >= FILTER_GROUP_FIRST_MDS);
398 rc = obd_create(mds->mds_osc_exp, oa, &lsm, &oti);
403 CDEBUG(level, "error creating objects for "
404 "inode %lu: rc = %d\n",
407 CERROR("obd_create returned invalid "
414 rc = obd_iocontrol(OBD_IOC_LOV_SETEA, mds->mds_osc_exp,
415 0, &lsm, rec->ur_eadata);
419 lsm->lsm_object_id = oa->o_id;
420 lsm->lsm_object_gr = oa->o_gr;
423 oa->o_size = inode->i_size;
424 obdo_from_inode(oa, inode, OBD_MD_FLTYPE|OBD_MD_FLATIME|
425 OBD_MD_FLMTIME| OBD_MD_FLCTIME| OBD_MD_FLSIZE);
426 rc = obd_setattr(mds->mds_osc_exp, oa, lsm, &oti);
428 CERROR("error setting attrs for inode %lu: rc %d\n",
431 CERROR("obd_setattr returned bad rc %d\n", rc);
438 body->valid |= OBD_MD_FLBLKSZ | OBD_MD_FLEASIZE;
439 obdo_refresh_inode(inode, oa, OBD_MD_FLBLKSZ);
441 LASSERT(lsm && lsm->lsm_object_id);
443 rc = obd_packmd(mds->mds_osc_exp, &lmm, lsm);
444 if (!rec->ur_fid2->id)
445 obd_free_memmd(mds->mds_osc_exp, &lsm);
448 body->eadatasize = rc;
451 *handle = fsfilt_start(obd, inode, FSFILT_OP_CREATE, NULL);
452 if (IS_ERR(*handle)) {
453 rc = PTR_ERR(*handle);
458 rc = fsfilt_set_md(obd, inode, *handle, lmm, lmm_size);
459 lmm_buf = lustre_msg_buf(req->rq_repmsg, offset, 0);
460 lmm_bufsize = req->rq_repmsg->buflens[offset];
462 LASSERT(lmm_bufsize >= lmm_size);
464 memcpy(lmm_buf, lmm, lmm_size);
465 obd_free_diskmd(mds->mds_osc_exp, &lmm);
467 oti_free_cookies(&oti);
471 OBD_FREE(*ids, mds->mds_lov_desc.ld_tgt_count * sizeof(**ids));
477 static void reconstruct_open(struct mds_update_record *rec, int offset,
478 struct ptlrpc_request *req,
479 struct lustre_handle *child_lockh)
481 struct mds_export_data *med = &req->rq_export->exp_mds_data;
482 struct mds_client_data *mcd = med->med_mcd;
483 struct mds_obd *mds = mds_req2mds(req);
484 struct mds_file_data *mfd;
485 struct obd_device *obd = req->rq_export->exp_obd;
486 struct dentry *parent, *child;
487 struct ldlm_reply *rep;
488 struct mds_body *body;
494 LASSERT(offset == 3); /* only called via intent */
495 rep = lustre_msg_buf(req->rq_repmsg, 0, sizeof (*rep));
496 body = lustre_msg_buf(req->rq_repmsg, 1, sizeof (*body));
498 /* copy rc, transno and disp; steal locks */
499 mds_req_from_mcd(req, mcd);
500 intent_set_disposition(rep, mcd->mcd_last_data);
502 /* Only replay if create or open actually happened. */
503 if (!intent_disposition(rep, DISP_OPEN_CREATE | DISP_OPEN_OPEN) ) {
505 return; /* error looking up parent or child */
508 parent = mds_fid2dentry(mds, rec->ur_fid1, NULL);
509 LASSERTF(!IS_ERR(parent), "fid "LPU64"/%u rc %ld\n", rec->ur_fid1->id,
510 rec->ur_fid1->generation, PTR_ERR(parent));
512 child = ll_lookup_one_len(rec->ur_name, parent, rec->ur_namelen - 1);
513 LASSERTF(!IS_ERR(child), "parent "LPU64"/%u child %s rc %ld\n",
514 rec->ur_fid1->id, rec->ur_fid1->generation, rec->ur_name,
518 GOTO(out_dput, 0); /* child not present to open */
520 /* At this point, we know we have a child. We'll send
521 * it back _unless_ it not created and open failed. */
522 if (intent_disposition(rep, DISP_OPEN_OPEN) &&
523 !intent_disposition(rep, DISP_OPEN_CREATE) &&
527 mds_pack_inode2fid(obd, &body->fid1, child->d_inode);
528 mds_pack_inode2body(obd, body, child->d_inode);
529 if (S_ISREG(child->d_inode->i_mode)) {
530 rc = mds_pack_md(obd, req->rq_repmsg, 2, body,
534 LASSERT(rc == req->rq_status);
536 /* If we have LOV EA data, the OST holds size, mtime */
537 if (!(body->valid & OBD_MD_FLEASIZE))
538 body->valid |= (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS |
539 OBD_MD_FLATIME | OBD_MD_FLMTIME);
541 /* XXX need to check this case */
544 /* If we're opening a file without an EA, change to a write
545 lock (unless we already have one). */
547 /* If we have -EEXIST as the status, and we were asked to create
548 * exclusively, we can tell we failed because the file already existed.
550 if (req->rq_status == -EEXIST &&
551 ((rec->ur_flags & (MDS_OPEN_CREAT | MDS_OPEN_EXCL)) ==
552 (MDS_OPEN_CREAT | MDS_OPEN_EXCL))) {
556 /* If we didn't get as far as trying to open, then some locking thing
557 * probably went wrong, and we'll just bail here.
559 if (!intent_disposition(rep, DISP_OPEN_OPEN))
562 /* If we failed, then we must have failed opening, so don't look for
563 * file descriptor or anything, just give the client the bad news.
569 list_for_each(t, &med->med_open_head) {
570 mfd = list_entry(t, struct mds_file_data, mfd_list);
571 if (mfd->mfd_xid == req->rq_xid)
576 /* #warning "XXX fixme" bug 2991 */
577 /* Here it used to LASSERT(mfd) if exp_outstanding_reply != NULL.
578 * Now that exp_outstanding_reply is a list, it's just using mfd != NULL
579 * to detect a re-open */
581 mntget(mds->mds_vfsmnt);
582 CERROR("Re-opened file \n");
583 mfd = mds_dentry_open(child, mds->mds_vfsmnt,
584 rec->ur_flags & ~MDS_OPEN_TRUNC, req);
586 CERROR("mds: out of memory\n");
587 GOTO(out_dput, req->rq_status = -ENOMEM);
591 body->handle.cookie = mfd->mfd_handle.h_cookie;
592 CDEBUG(D_INODE, "resend mfd %p, cookie "LPX64"\n", mfd,
593 mfd->mfd_handle.h_cookie);
603 /* do NOT or the MAY_*'s, you'll get the weakest */
604 static int accmode(int flags)
608 if (flags & FMODE_READ)
610 if (flags & (FMODE_WRITE|MDS_OPEN_TRUNC))
612 if (flags & FMODE_EXEC)
617 /* Handles object creation, actual opening, and I/O epoch */
618 static int mds_finish_open(struct ptlrpc_request *req, struct dentry *dchild,
619 struct mds_body *body, int flags, void **handle,
620 struct mds_update_record *rec,struct ldlm_reply *rep)
622 struct mds_obd *mds = mds_req2mds(req);
623 struct obd_device *obd = req->rq_export->exp_obd;
624 struct mds_file_data *mfd = NULL;
625 obd_id *ids = NULL; /* object IDs created */
630 /* atomically create objects if necessary */
631 down(&dchild->d_inode->i_sem);
632 mode = dchild->d_inode->i_mode;
633 if ((S_ISREG(mode) && !(body->valid & OBD_MD_FLEASIZE)) ||
634 (S_ISDIR(mode) && !(body->valid & OBD_MD_FLDIREA))) {
635 rc = mds_pack_md(obd, req->rq_repmsg, 2, body,
638 up(&dchild->d_inode->i_sem);
643 /* no EA: create objects */
644 rc = mds_create_objects(req, 2, rec, mds, obd,
645 dchild, handle, &ids);
647 CERROR("mds_create_objects: rc = %d\n", rc);
648 up(&dchild->d_inode->i_sem);
651 if (S_ISREG(dchild->d_inode->i_mode) &&
652 (body->valid & OBD_MD_FLEASIZE)) {
653 rc = mds_revalidate_lov_ea(obd, dchild->d_inode,
656 rc = mds_pack_md(obd, req->rq_repmsg, 2, body,
659 up(&dchild->d_inode->i_sem);
664 /* If the inode has no EA data, then MDSs hold size, mtime */
665 if (S_ISREG(dchild->d_inode->i_mode) &&
666 !(body->valid & OBD_MD_FLEASIZE)) {
667 body->valid |= (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS |
668 OBD_MD_FLATIME | OBD_MD_FLMTIME);
670 up(&dchild->d_inode->i_sem);
672 intent_set_disposition(rep, DISP_OPEN_OPEN);
673 mfd = mds_dentry_open(dchild, mds->mds_vfsmnt, flags, req);
675 RETURN(PTR_ERR(mfd));
677 CDEBUG(D_INODE, "mfd %p, cookie "LPX64"\n", mfd,
678 mfd->mfd_handle.h_cookie);
680 mds_lov_update_objids(obd, ids);
681 OBD_FREE(ids, sizeof(*ids) * mds->mds_lov_desc.ld_tgt_count);
684 // mds_mfd_destroy(mfd);
688 static int mds_open_by_fid(struct ptlrpc_request *req, struct ll_fid *fid,
689 struct mds_body *body, int flags,
690 struct mds_update_record *rec,struct ldlm_reply *rep)
692 struct mds_obd *mds = mds_req2mds(req);
693 struct inode *pending_dir = mds->mds_pending_dir->d_inode;
694 struct dentry *dchild;
695 char fidname[LL_FID_NAMELEN];
700 down(&pending_dir->i_sem);
701 fidlen = ll_fid2str(fidname, fid->id, fid->generation);
702 dchild = lookup_one_len(fidname, mds->mds_pending_dir, fidlen);
703 if (IS_ERR(dchild)) {
704 up(&pending_dir->i_sem);
705 rc = PTR_ERR(dchild);
706 CERROR("error looking up %s in PENDING: rc = %d\n",fidname, rc);
710 if (dchild->d_inode != NULL) {
711 up(&pending_dir->i_sem);
712 mds_inode_set_orphan(dchild->d_inode);
713 mds_pack_inode2fid(req2obd(req), &body->fid1, dchild->d_inode);
714 mds_pack_inode2body(req2obd(req), body, dchild->d_inode);
715 intent_set_disposition(rep, DISP_LOOKUP_EXECD);
716 intent_set_disposition(rep, DISP_LOOKUP_POS);
717 CWARN("Orphan %s found and opened in PENDING directory\n",
722 up(&pending_dir->i_sem);
724 /* We didn't find it in PENDING so it isn't an orphan. See
725 * if it was a regular inode that was previously created. */
726 dchild = mds_fid2dentry(mds, fid, NULL);
728 RETURN(PTR_ERR(dchild));
730 mds_pack_inode2fid(req2obd(req), &body->fid1, dchild->d_inode);
731 mds_pack_inode2body(req2obd(req), body, dchild->d_inode);
732 intent_set_disposition(rep, DISP_LOOKUP_EXECD);
733 intent_set_disposition(rep, DISP_LOOKUP_POS);
736 rc = mds_finish_open(req, dchild, body, flags, &handle, rec, rep);
737 rc = mds_finish_transno(mds, dchild ? dchild->d_inode : NULL, handle,
738 req, rc, rep ? rep->lock_policy_res1 : 0);
739 /* XXX what do we do here if mds_finish_transno itself failed? */
745 int mds_pin(struct ptlrpc_request *req, int offset)
747 struct obd_device *obd = req->rq_export->exp_obd;
748 struct mds_body *request_body, *reply_body;
749 struct lvfs_run_ctxt saved;
750 int rc, size = sizeof(*reply_body);
753 request_body = lustre_msg_buf(req->rq_reqmsg, offset,
754 sizeof(*request_body));
756 rc = lustre_pack_reply(req, 1, &size, NULL);
759 reply_body = lustre_msg_buf(req->rq_repmsg, 0, sizeof(*reply_body));
761 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
762 rc = mds_open_by_fid(req, &request_body->fid1, reply_body,
763 request_body->flags, NULL, NULL);
764 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
769 /* Get a lock on the ino to sync with creation WRT inode reuse (bug 2029).
770 * If child_lockh is NULL we just get the lock as a barrier to wait for
771 * other holders of this lock, and drop it right away again. */
772 int mds_lock_new_child(struct obd_device *obd, struct inode *inode,
773 struct lustre_handle *child_lockh)
775 struct ldlm_res_id child_res_id = { .name = { inode->i_ino, 0, 1, 0 } };
776 struct lustre_handle lockh;
780 if (child_lockh == NULL)
781 child_lockh = &lockh;
783 rc = ldlm_cli_enqueue(NULL, NULL, obd->obd_namespace, child_res_id,
784 LDLM_PLAIN, NULL, LCK_EX, &lock_flags,
785 mds_blocking_ast, ldlm_completion_ast, NULL, NULL,
786 NULL, 0, NULL, child_lockh);
788 CERROR("ldlm_cli_enqueue: %d\n", rc);
789 else if (child_lockh == &lockh)
790 ldlm_lock_decref(child_lockh, LCK_EX);
795 static int is_mount_object(struct dentry *dparent)
797 struct dentry *dchild;
799 if (!(dparent->d_inode->i_mode & S_ISUID))
802 dchild = lookup_one_len(".mntinfo", dparent, strlen(".mntinfo"));
803 if (IS_ERR(dchild) || dchild == NULL)
810 int mds_open(struct mds_update_record *rec, int offset,
811 struct ptlrpc_request *req, struct lustre_handle *child_lockh)
813 /* XXX ALLOCATE _something_ - 464 bytes on stack here */
814 struct obd_device *obd = req->rq_export->exp_obd;
815 struct mds_obd *mds = mds_req2mds(req);
816 struct ldlm_reply *rep = NULL;
817 struct mds_body *body = NULL;
818 struct dentry *dchild = NULL, *dparent = NULL;
819 struct mds_export_data *med;
820 struct lustre_handle parent_lockh[2];
821 int rc = 0, cleanup_phase = 0, acc_mode, created = 0;
822 int parent_mode = LCK_PR;
824 struct dentry_params dp;
825 struct mea *mea = NULL;
826 int mea_size, update_mode;
829 DEBUG_REQ(D_INODE, req, "parent "LPU64"/%u name %*s mode %o",
830 rec->ur_fid1->id, rec->ur_fid1->generation,
831 rec->ur_namelen - 1, rec->ur_name, rec->ur_mode);
833 parent_lockh[0].cookie = 0;
834 parent_lockh[1].cookie = 0;
836 if (offset == 3) { /* intent */
837 rep = lustre_msg_buf(req->rq_repmsg, 0, sizeof (*rep));
838 body = lustre_msg_buf(req->rq_repmsg, 1, sizeof (*body));
839 } else if (offset == 1) { /* non-intent reint */
840 body = lustre_msg_buf(req->rq_repmsg, 0, sizeof (*body));
846 MDS_CHECK_RESENT(req, reconstruct_open(rec, offset, req, child_lockh));
848 MDS_UPDATE_COUNTER(mds, MDS_OPEN_COUNT);
850 /* Step 0: If we are passed a fid, then we assume the client already
851 * opened this file and is only replaying the RPC, so we open the
852 * inode by fid (at some large expense in security). */
853 if (lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY) {
854 DEBUG_REQ(D_HA, req, "open replay");
855 CDEBUG(D_HA, "open fid "LPU64"/%u name %*s mode %o\n",
856 rec->ur_fid2->id, rec->ur_fid2->generation,
857 rec->ur_namelen - 1, rec->ur_name, rec->ur_mode);
859 LASSERT(rec->ur_fid2->id);
861 rc = mds_open_by_fid(req, rec->ur_fid2, body, rec->ur_flags,
865 /* We didn't find the correct inode on disk either, so we
866 * need to re-create it via a regular replay. */
867 LASSERT(rec->ur_flags & MDS_OPEN_CREAT);
869 LASSERT(!rec->ur_fid2->id);
872 LASSERT(offset == 3); /* If we got here, we must be called via intent */
874 med = &req->rq_export->exp_mds_data;
875 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_OPEN_PACK)) {
876 CERROR("test case OBD_FAIL_MDS_OPEN_PACK\n");
880 acc_mode = accmode(rec->ur_flags);
882 /* Step 1: Find and lock the parent */
883 if (rec->ur_flags & O_CREAT) {
884 /* XXX Well, in fact we only need this lock mode change if
885 in addition to O_CREAT, the file does not exist.
886 But we do not know if it exists or not yet */
887 parent_mode = LCK_PW;
890 if (rec->ur_namelen == 1) {
891 /* client (LMV) wants to open the file by fid */
892 CDEBUG(D_OTHER, "OPEN by fid %u/%u/%u\n",
893 (unsigned) rec->ur_fid1->mds,
894 (unsigned) rec->ur_fid1->id,
895 (unsigned) rec->ur_fid1->generation);
896 dchild = mds_fid2dentry(mds, rec->ur_fid1, NULL);
897 if (IS_ERR(dchild)) {
898 rc = PTR_ERR(dparent);
899 CERROR("child lookup by a fid error %d\n", rc);
905 dparent = mds_fid2locked_dentry(obd, rec->ur_fid1, NULL, parent_mode,
906 parent_lockh, &update_mode, rec->ur_name,
908 MDS_INODELOCK_UPDATE);
909 if (IS_ERR(dparent)) {
910 rc = PTR_ERR(dparent);
911 CERROR("parent lookup error %d\n", rc);
914 LASSERT(dparent->d_inode != NULL);
916 cleanup_phase = 1; /* parent dentry and lock */
918 /* try to retrieve MEA data for this dir */
919 rc = mds_get_lmv_attr(obd, dparent->d_inode, &mea, &mea_size);
924 /* dir is already splitted, check is requested filename
925 * should live at this MDS or at another one */
927 i = mea_name2idx(mea, rec->ur_name, rec->ur_namelen - 1);
928 if (mea->mea_master != mea->mea_fids[i].mds) {
930 "%s: inapropriate MDS(%d) for %lu/%u:%s."
931 " should be %d(%d)\n", obd->obd_name,
932 mea->mea_master, dparent->d_inode->i_ino,
933 dparent->d_inode->i_generation, rec->ur_name,
934 mea->mea_fids[i].mds, i);
935 GOTO(cleanup, rc = -ERESTART);
939 /* Step 2: Lookup the child */
940 dchild = ll_lookup_one_len(rec->ur_name, dparent, rec->ur_namelen - 1);
941 if (IS_ERR(dchild)) {
942 rc = PTR_ERR(dchild);
943 dchild = NULL; /* don't confuse mds_finish_transno() below */
948 cleanup_phase = 2; /* child dentry */
950 intent_set_disposition(rep, DISP_LOOKUP_EXECD);
952 if (dchild->d_flags & DCACHE_CROSS_REF) {
953 struct ldlm_res_id res_id = { . name = {0} };
954 ldlm_policy_data_t policy;
956 CDEBUG(D_OTHER, "cross reference: %lu/%lu/%lu\n",
957 (unsigned long) dchild->d_mdsnum,
958 (unsigned long) dchild->d_inum,
959 (unsigned long) dchild->d_generation);
960 body->valid |= OBD_MD_FLID | OBD_MD_MDS;
961 body->fid1.id = dchild->d_inum;
962 body->fid1.mds = dchild->d_mdsnum;
963 body->fid1.generation = dchild->d_generation;
964 intent_set_disposition(rep, DISP_LOOKUP_POS);
965 res_id.name[0] = dchild->d_inum;
966 res_id.name[1] = dchild->d_generation;
967 policy.l_inodebits.bits = MDS_INODELOCK_LOOKUP;
968 rc = ldlm_cli_enqueue(NULL, NULL, obd->obd_namespace,
969 res_id, LDLM_IBITS, &policy,
972 ldlm_completion_ast, NULL, NULL,
973 NULL, 0, NULL, child_lockh);
975 if (parent_lockh[1].cookie != 0)
976 ldlm_lock_decref(parent_lockh + 1, update_mode);
978 ldlm_lock_decref(parent_lockh, parent_mode);
980 OBD_FREE(mea, mea_size);
987 intent_set_disposition(rep, DISP_LOOKUP_POS);
989 intent_set_disposition(rep, DISP_LOOKUP_NEG);
991 /*Step 3: If the child was negative, and we're supposed to, create it.*/
992 if (dchild->d_inode == NULL) {
993 unsigned long ino = rec->ur_fid2->id;
996 rc = mds_try_to_split_dir(obd, dparent, &mea, 0, update_mode);
997 CDEBUG(D_OTHER, "%s: splitted %lu/%u - %d\n",
998 obd->obd_name, dparent->d_inode->i_ino,
999 dparent->d_inode->i_generation, rc);
1001 /* dir got splitted */
1002 GOTO(cleanup, rc = -ERESTART);
1003 } else if (rc < 0) {
1004 /* error happened during spitting */
1008 if (!(rec->ur_flags & MDS_OPEN_CREAT)) {
1009 /* It's negative and we weren't supposed to create it */
1010 GOTO(cleanup, rc = -ENOENT);
1013 intent_set_disposition(rep, DISP_OPEN_CREATE);
1014 handle = fsfilt_start(obd, dparent->d_inode, FSFILT_OP_CREATE,
1016 if (IS_ERR(handle)) {
1017 rc = PTR_ERR(handle);
1021 dchild->d_fsdata = (void *) &dp;
1025 rc = ll_vfs_create(dparent->d_inode, dchild, rec->ur_mode,NULL);
1026 if (dchild->d_fsdata == (void *)(unsigned long)ino)
1027 dchild->d_fsdata = NULL;
1030 CDEBUG(D_INODE, "error during create: %d\n", rc);
1033 inode = dchild->d_inode;
1035 LASSERT(ino == inode->i_ino);
1036 /* Written as part of setattr */
1037 inode->i_generation = rec->ur_fid2->generation;
1038 CDEBUG(D_HA, "recreated ino %lu with gen %u\n",
1039 inode->i_ino, inode->i_generation);
1043 LTIME_S(iattr.ia_atime) = rec->ur_time;
1044 LTIME_S(iattr.ia_ctime) = rec->ur_time;
1045 LTIME_S(iattr.ia_mtime) = rec->ur_time;
1047 iattr.ia_uid = rec->_ur_fsuid;
1048 if (dparent->d_inode->i_mode & S_ISGID)
1049 iattr.ia_gid = dparent->d_inode->i_gid;
1051 iattr.ia_gid = rec->_ur_fsgid;
1053 iattr.ia_valid = ATTR_UID | ATTR_GID | ATTR_ATIME |
1054 ATTR_MTIME | ATTR_CTIME;
1056 rc = fsfilt_setattr(obd, dchild, handle, &iattr, 0);
1058 CERROR("error on child setattr: rc = %d\n", rc);
1060 iattr.ia_valid = ATTR_MTIME | ATTR_CTIME;
1062 rc = fsfilt_setattr(obd, dparent, handle, &iattr, 0);
1064 CERROR("error on parent setattr: rc = %d\n", rc);
1066 MDS_UPDATE_COUNTER(mds, MDS_CREATE_COUNT);
1068 if (!(rec->ur_flags & O_EXCL)) { /* bug 3313 */
1069 rc = fsfilt_commit(obd, dchild->d_inode->i_sb,
1070 dchild->d_inode, handle, 0);
1074 acc_mode = 0; /* Don't check for permissions */
1077 LASSERT(!mds_inode_is_orphan(dchild->d_inode));
1079 mds_pack_inode2fid(obd, &body->fid1, dchild->d_inode);
1080 mds_pack_inode2body(obd, body, dchild->d_inode);
1082 if (S_ISREG(dchild->d_inode->i_mode)) {
1083 /* Check permissions etc */
1084 rc = ll_permission(dchild->d_inode, acc_mode, NULL);
1088 /* Can't write to a read-only file */
1089 if (IS_RDONLY(dchild->d_inode) && (acc_mode & MAY_WRITE) != 0)
1090 GOTO(cleanup, rc = -EPERM);
1092 /* An append-only file must be opened in append mode for
1094 if (IS_APPEND(dchild->d_inode) && (acc_mode & MAY_WRITE) != 0 &&
1095 ((rec->ur_flags & MDS_OPEN_APPEND) == 0 ||
1096 (rec->ur_flags & MDS_OPEN_TRUNC) != 0))
1097 GOTO(cleanup, rc = -EPERM);
1100 if (!created && (rec->ur_flags & MDS_OPEN_CREAT) &&
1101 (rec->ur_flags & MDS_OPEN_EXCL)) {
1102 /* File already exists, we didn't just create it, and we
1103 * were passed O_EXCL; err-or. */
1104 GOTO(cleanup, rc = -EEXIST); // returns a lock to the client
1107 /* if we are following a symlink, don't open */
1108 if (S_ISLNK(dchild->d_inode->i_mode))
1109 GOTO(cleanup, rc = 0);
1111 if ((rec->ur_flags & MDS_OPEN_DIRECTORY) &&
1112 !S_ISDIR(dchild->d_inode->i_mode))
1113 GOTO(cleanup, rc = -ENOTDIR);
1115 if (S_ISDIR(dchild->d_inode->i_mode)) {
1116 if (rec->ur_flags & MDS_OPEN_CREAT ||
1117 rec->ur_flags & FMODE_WRITE) {
1118 /*we are trying to create or write a exist dir*/
1119 GOTO(cleanup, rc = -EISDIR);
1121 if (ll_permission(dchild->d_inode, acc_mode, NULL)) {
1122 GOTO(cleanup, rc = -EACCES);
1124 if (is_mount_object(dchild)) {
1125 CERROR("Found possible GNS mount object %*s; not "
1126 "opening.\n", dchild->d_name.len,
1127 dchild->d_name.name);
1128 GOTO(cleanup, rc = 0); // success, but don't really open
1132 /* Step 5: mds_open it */
1133 rc = mds_finish_open(req, dchild, body, rec->ur_flags, &handle, rec,
1138 rc = mds_finish_transno(mds, dchild ? dchild->d_inode : NULL, handle,
1139 req, rc, rep ? rep->lock_policy_res1 : 0);
1141 switch (cleanup_phase) {
1143 if (rc && created) {
1144 int err = vfs_unlink(dparent->d_inode, dchild);
1146 CERROR("unlink(%*s) in error path: %d\n",
1147 dchild->d_name.len, dchild->d_name.name,
1150 } else if (created) {
1152 mds_lock_new_child(obd, dchild->d_inode, NULL);
1157 if (dparent == NULL)
1162 if (parent_lockh[1].cookie != 0)
1163 ldlm_lock_decref(parent_lockh + 1, update_mode);
1166 ldlm_lock_decref(parent_lockh, parent_mode);
1168 ptlrpc_save_lock (req, parent_lockh, parent_mode);
1171 OBD_FREE(mea, mea_size);
1175 /* Close a "file descriptor" and possibly unlink an orphan from the
1176 * PENDING directory.
1178 * If we are being called from mds_disconnect() because the client has
1179 * disappeared, then req == NULL and we do not update last_rcvd because
1180 * there is nothing that could be recovered by the client at this stage
1181 * (it will not even _have_ an entry in last_rcvd anymore).
1183 * Returns EAGAIN if the client needs to get more data and re-close. */
1184 int mds_mfd_close(struct ptlrpc_request *req, int offset,
1185 struct obd_device *obd, struct mds_file_data *mfd,
1188 struct inode *inode = mfd->mfd_dentry->d_inode;
1189 char fidname[LL_FID_NAMELEN];
1190 int last_orphan, fidlen, rc = 0, cleanup_phase = 0;
1191 struct dentry *pending_child = NULL;
1192 struct mds_obd *mds = &obd->u.mds;
1193 struct inode *pending_dir = mds->mds_pending_dir->d_inode;
1194 void *handle = NULL;
1195 struct mds_body *request_body = NULL, *reply_body = NULL;
1196 struct dentry_params dp;
1197 struct iattr iattr = { 0 };
1198 struct llog_create_locks *lcl = NULL;
1201 if (req && req->rq_reqmsg != NULL)
1202 request_body = lustre_msg_buf(req->rq_reqmsg, offset,
1203 sizeof(*request_body));
1204 if (req && req->rq_repmsg != NULL)
1205 reply_body = lustre_msg_buf(req->rq_repmsg, 0,
1206 sizeof(*reply_body));
1208 fidlen = ll_fid2str(fidname, inode->i_ino, inode->i_generation);
1210 last_orphan = mds_open_orphan_dec_test(inode) &&
1211 mds_inode_is_orphan(inode);
1213 /* this is half of the actual "close" */
1214 if (mfd->mfd_mode & FMODE_WRITE) {
1215 rc = mds_put_write_access(mds, inode, request_body,
1216 last_orphan && unlink_orphan);
1217 } else if (mfd->mfd_mode & FMODE_EXEC) {
1218 mds_allow_write_access(inode);
1221 if (last_orphan && unlink_orphan) {
1222 struct lov_mds_md *lmm = NULL;
1223 int stripe_count = 0;
1224 LASSERT(rc == 0); /* mds_put_write_access must have succeeded */
1226 if (obd->obd_recovering) {
1227 CDEBUG(D_HA, "not remove orphan %s until recovery"
1228 " is over\n", fidname);
1232 CDEBUG(D_HA, "destroying orphan object %s\n", fidname);
1234 /* Sadly, there is no easy way to save pending_child from
1235 * mds_reint_unlink() into mfd, so we need to re-lookup,
1236 * but normally it will still be in the dcache. */
1237 pending_dir = mds->mds_pending_dir->d_inode;
1238 down(&pending_dir->i_sem);
1239 cleanup_phase = 1; /* up(i_sem) when finished */
1240 pending_child = lookup_one_len(fidname, mds->mds_pending_dir,
1242 if (IS_ERR(pending_child))
1243 GOTO(cleanup, rc = PTR_ERR(pending_child));
1244 LASSERT(pending_child->d_inode != NULL);
1246 cleanup_phase = 2; /* dput(pending_child) when finished */
1247 if (req != NULL && req->rq_repmsg != NULL) {
1248 lmm = lustre_msg_buf(req->rq_repmsg, 1, 0);
1249 stripe_count = le32_to_cpu(lmm->lmm_stripe_count);
1252 handle = fsfilt_start_log(obd, pending_dir, FSFILT_OP_UNLINK,
1253 NULL, stripe_count);
1254 if (IS_ERR(handle)) {
1255 rc = PTR_ERR(handle);
1260 pending_child->d_fsdata = (void *) &dp;
1263 if (S_ISDIR(pending_child->d_inode->i_mode))
1264 rc = vfs_rmdir(pending_dir, pending_child);
1266 rc = vfs_unlink(pending_dir, pending_child);
1268 CERROR("error unlinking orphan %s: rc %d\n",fidname,rc);
1270 if (req != NULL && req->rq_repmsg != NULL &&
1271 (reply_body->valid & OBD_MD_FLEASIZE) &&
1272 mds_log_op_unlink(obd, pending_child->d_inode,
1273 lmm, req->rq_repmsg->buflens[1],
1274 lustre_msg_buf(req->rq_repmsg, 2, 0),
1275 req->rq_repmsg->buflens[2], &lcl) > 0) {
1276 reply_body->valid |= OBD_MD_FLCOOKIE;
1279 goto out; /* Don't bother updating attrs on unlinked inode */
1282 if (request_body != NULL && mfd->mfd_mode & FMODE_WRITE && rc == 0) {
1283 /* Update the on-disk attributes if this was the last write
1284 * close, and all information was provided (i.e., rc == 0)
1286 * XXX this should probably be abstracted with mds_reint_setattr
1290 if (request_body->valid & OBD_MD_FLMTIME &&
1291 LTIME_S(request_body->mtime) > LTIME_S(inode->i_mtime)) {
1292 LTIME_S(iattr.ia_mtime) = LTIME_S(request_body->mtime);
1293 iattr.ia_valid |= ATTR_MTIME;
1295 if (request_body->valid & OBD_MD_FLCTIME &&
1296 LTIME_S(request_body->ctime) > LTIME_S(inode->i_ctime)) {
1297 LTIME_S(iattr.ia_ctime) = LTIME_S(request_body->ctime);
1298 iattr.ia_valid |= ATTR_CTIME;
1301 /* XXX can't set block count with fsfilt_setattr (!) */
1302 if (request_body->valid & OBD_MD_FLSIZE) {
1303 iattr.ia_valid |= ATTR_SIZE;
1304 iattr.ia_size = request_body->size;
1306 /* if (request_body->valid & OBD_MD_FLBLOCKS) {
1307 iattr.ia_valid |= ATTR_BLOCKS;
1308 iattr.ia_blocks = request_body->blocks
1314 if (request_body != NULL && request_body->valid & OBD_MD_FLATIME) {
1315 /* Only start a transaction to write out only the atime if
1316 * it is more out-of-date than the specified limit. If we
1317 * are already going to write out the atime then do it anyway.
1319 if ((request_body->atime >
1320 LTIME_S(inode->i_atime) + MAX_ATIME_DIFF) ||
1321 (iattr.ia_valid != 0 &&
1322 request_body->atime > LTIME_S(inode->i_atime))) {
1323 LTIME_S(iattr.ia_atime) = request_body->atime;
1324 iattr.ia_valid |= ATTR_ATIME;
1328 if (iattr.ia_valid != 0) {
1329 handle = fsfilt_start(obd, inode, FSFILT_OP_SETATTR, NULL);
1331 GOTO(cleanup, rc = PTR_ERR(handle));
1332 rc = fsfilt_setattr(obd, mfd->mfd_dentry, handle, &iattr, 0);
1334 CERROR("error in setattr(%s): rc %d\n", fidname, rc);
1337 /* If other clients have this file open for write, rc will be > 0 */
1340 l_dput(mfd->mfd_dentry);
1341 mds_mfd_destroy(mfd);
1344 if (req != NULL && reply_body != NULL) {
1345 rc = mds_finish_transno(mds, pending_dir, handle, req, rc, 0);
1346 } else if (handle) {
1347 int err = fsfilt_commit(obd, mds->mds_sb, pending_dir, handle, 0);
1349 CERROR("error committing close: %d\n", err);
1355 switch (cleanup_phase) {
1358 ptlrpc_save_llog_lock(req, lcl);
1359 dput(pending_child);
1361 up(&pending_dir->i_sem);
1366 int mds_close(struct ptlrpc_request *req, int offset)
1368 struct mds_export_data *med = &req->rq_export->exp_mds_data;
1369 struct obd_device *obd = req->rq_export->exp_obd;
1370 struct mds_body *body;
1371 struct mds_file_data *mfd;
1372 struct lvfs_run_ctxt saved;
1373 struct inode *inode;
1374 int rc, repsize[3] = {sizeof(struct mds_body),
1375 obd->u.mds.mds_max_mdsize,
1376 obd->u.mds.mds_max_cookiesize};
1379 MDS_UPDATE_COUNTER((&obd->u.mds), MDS_CLOSE_COUNT);
1381 rc = lustre_pack_reply(req, 3, repsize, NULL);
1383 CERROR("lustre_pack_reply: rc = %d\n", rc);
1384 req->rq_status = rc;
1386 MDS_CHECK_RESENT(req, mds_reconstruct_generic(req));
1389 if (lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY) {
1390 DEBUG_REQ(D_HA, req, "close replay\n");
1391 memcpy(lustre_msg_buf(req->rq_repmsg, 2, 0),
1392 lustre_msg_buf(req->rq_reqmsg, offset + 1, 0),
1393 req->rq_repmsg->buflens[2]);
1397 body = lustre_swab_reqbuf(req, offset, sizeof(*body),
1398 lustre_swab_mds_body);
1400 CERROR("Can't unpack body\n");
1401 req->rq_status = -EFAULT;
1405 if (body->flags & MDS_BFLAG_UNCOMMITTED_WRITES)
1406 /* do some stuff */ ;
1408 mfd = mds_handle2mfd(&body->handle);
1410 DEBUG_REQ(D_ERROR, req, "no handle for file close ino "LPD64
1411 ": cookie "LPX64, body->fid1.id, body->handle.cookie);
1412 req->rq_status = -ESTALE;
1416 inode = mfd->mfd_dentry->d_inode;
1417 if (mds_inode_is_orphan(inode) && mds_open_orphan_count(inode) == 1) {
1418 body = lustre_msg_buf(req->rq_repmsg, 0, sizeof (*body));
1419 LASSERT(body != NULL);
1421 mds_pack_inode2fid(obd, &body->fid1, inode);
1422 mds_pack_inode2body(obd, body, inode);
1423 mds_pack_md(obd, req->rq_repmsg, 1, body, inode, 1);
1425 spin_lock(&med->med_open_lock);
1426 list_del(&mfd->mfd_list);
1427 spin_unlock(&med->med_open_lock);
1429 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
1430 req->rq_status = mds_mfd_close(req, offset, obd, mfd, 1);
1431 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
1433 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_CLOSE_PACK)) {
1434 CERROR("test case OBD_FAIL_MDS_CLOSE_PACK\n");
1435 req->rq_status = -ENOMEM;
1444 int mds_done_writing(struct ptlrpc_request *req, int offset)
1446 struct mds_body *body;
1447 int rc, size = sizeof(struct mds_body);
1450 MDS_CHECK_RESENT(req, mds_reconstruct_generic(req));
1452 body = lustre_swab_reqbuf(req, offset, sizeof(*body),
1453 lustre_swab_mds_body);
1455 CERROR("Can't unpack body\n");
1456 req->rq_status = -EFAULT;
1460 rc = lustre_pack_reply(req, 1, &size, NULL);
1462 CERROR("lustre_pack_reply: rc = %d\n", rc);
1463 req->rq_status = rc;