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 *handle = fsfilt_start(obd, inode, FSFILT_OP_CREATE, NULL);
338 if (IS_ERR(*handle)) {
339 rc = PTR_ERR(*handle);
345 if(lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY) {
346 LASSERT (rec->ur_fid2->id);
347 body->valid |= OBD_MD_FLBLKSZ | OBD_MD_FLEASIZE;
348 lmm_size = rec->ur_eadatalen;
349 lmm = rec->ur_eadata;
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 OBD_ALLOC(lmm, mds->mds_max_mdsize);
395 GOTO(out_oa, rc = -ENOMEM);
397 lmm_size = mds->mds_max_mdsize;
398 rc = mds_get_md(obd, dchild->d_parent->d_inode,
401 rc = obd_iocontrol(OBD_IOC_LOV_SETSTRIPE,
404 OBD_FREE(lmm, mds->mds_max_mdsize);
408 LASSERT(oa->o_gr >= FILTER_GROUP_FIRST_MDS);
409 rc = obd_create(mds->mds_osc_exp, oa, &lsm, &oti);
414 CDEBUG(level, "error creating objects for "
415 "inode %lu: rc = %d\n",
418 CERROR("obd_create returned invalid "
425 rc = obd_iocontrol(OBD_IOC_LOV_SETEA, mds->mds_osc_exp,
426 0, &lsm, rec->ur_eadata);
430 lsm->lsm_object_id = oa->o_id;
431 lsm->lsm_object_gr = oa->o_gr;
434 oa->o_size = inode->i_size;
435 obdo_from_inode(oa, inode, OBD_MD_FLTYPE|OBD_MD_FLATIME|
436 OBD_MD_FLMTIME| OBD_MD_FLCTIME| OBD_MD_FLSIZE);
437 rc = obd_setattr(mds->mds_osc_exp, oa, lsm, &oti);
439 CERROR("error setting attrs for inode %lu: rc %d\n",
442 CERROR("obd_setattr returned bad rc %d\n", rc);
449 body->valid |= OBD_MD_FLBLKSZ | OBD_MD_FLEASIZE;
450 obdo_refresh_inode(inode, oa, OBD_MD_FLBLKSZ);
452 LASSERT(lsm && lsm->lsm_object_id);
454 rc = obd_packmd(mds->mds_osc_exp, &lmm, lsm);
455 if (!rec->ur_fid2->id)
456 obd_free_memmd(mds->mds_osc_exp, &lsm);
459 body->eadatasize = rc;
460 rc = fsfilt_set_md(obd, inode, *handle, lmm, lmm_size);
461 lmm_buf = lustre_msg_buf(req->rq_repmsg, offset, 0);
462 lmm_bufsize = req->rq_repmsg->buflens[offset];
464 LASSERT(lmm_bufsize >= lmm_size);
466 memcpy(lmm_buf, lmm, lmm_size);
467 obd_free_diskmd(mds->mds_osc_exp, &lmm);
469 oti_free_cookies(&oti);
473 OBD_FREE(*ids, mds->mds_lov_desc.ld_tgt_count * sizeof(**ids));
479 static void reconstruct_open(struct mds_update_record *rec, int offset,
480 struct ptlrpc_request *req,
481 struct lustre_handle *child_lockh)
483 struct mds_export_data *med = &req->rq_export->exp_mds_data;
484 struct mds_client_data *mcd = med->med_mcd;
485 struct mds_obd *mds = mds_req2mds(req);
486 struct mds_file_data *mfd;
487 struct obd_device *obd = req->rq_export->exp_obd;
488 struct dentry *parent, *child;
489 struct ldlm_reply *rep;
490 struct mds_body *body;
496 LASSERT(offset == 2); /* only called via intent */
497 rep = lustre_msg_buf(req->rq_repmsg, 0, sizeof (*rep));
498 body = lustre_msg_buf(req->rq_repmsg, 1, sizeof (*body));
500 /* copy rc, transno and disp; steal locks */
501 mds_req_from_mcd(req, mcd);
502 intent_set_disposition(rep, mcd->mcd_last_data);
504 /* Only replay if create or open actually happened. */
505 if (!intent_disposition(rep, DISP_OPEN_CREATE | DISP_OPEN_OPEN) ) {
507 return; /* error looking up parent or child */
510 parent = mds_fid2dentry(mds, rec->ur_fid1, NULL);
511 LASSERT(!IS_ERR(parent));
513 child = ll_lookup_one_len(rec->ur_name, parent, rec->ur_namelen - 1);
514 LASSERT(!IS_ERR(child));
516 if (!child->d_inode) {
517 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.
523 if (intent_disposition(rep, DISP_OPEN_OPEN) &&
524 !intent_disposition(rep, DISP_OPEN_CREATE) &&
529 /* get lock (write for O_CREAT, read otherwise) */
531 mds_pack_inode2fid(obd, &body->fid1, child->d_inode);
532 mds_pack_inode2body(obd, body, child->d_inode);
533 if (S_ISREG(child->d_inode->i_mode)) {
534 rc = mds_pack_md(obd, req->rq_repmsg, 2, body,
538 LASSERT(rc == req->rq_status);
540 /* If we have LOV EA data, the OST holds size, mtime */
541 if (!(body->valid & OBD_MD_FLEASIZE))
542 body->valid |= (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS |
543 OBD_MD_FLATIME | OBD_MD_FLMTIME);
545 /* XXX need to check this case */
548 /* If we're opening a file without an EA, change to a write
549 lock (unless we already have one). */
551 /* If we have -EEXIST as the status, and we were asked to create
552 * exclusively, we can tell we failed because the file already existed.
554 if (req->rq_status == -EEXIST &&
555 ((rec->ur_flags & (MDS_OPEN_CREAT | MDS_OPEN_EXCL)) ==
556 (MDS_OPEN_CREAT | MDS_OPEN_EXCL))) {
560 /* If we didn't get as far as trying to open, then some locking thing
561 * probably went wrong, and we'll just bail here.
563 if (!intent_disposition(rep, DISP_OPEN_OPEN))
566 /* If we failed, then we must have failed opening, so don't look for
567 * file descriptor or anything, just give the client the bad news.
573 list_for_each(t, &med->med_open_head) {
574 mfd = list_entry(t, struct mds_file_data, mfd_list);
575 if (mfd->mfd_xid == req->rq_xid)
580 /* #warning "XXX fixme" bug 2991 */
581 /* Here it used to LASSERT(mfd) if exp_outstanding_reply != NULL.
582 * Now that exp_outstanding_reply is a list, it's just using mfd != NULL
583 * to detect a re-open */
585 mntget(mds->mds_vfsmnt);
586 CERROR("Re-opened file \n");
587 mfd = mds_dentry_open(child, mds->mds_vfsmnt,
588 rec->ur_flags & ~MDS_OPEN_TRUNC, req);
590 CERROR("mds: out of memory\n");
591 GOTO(out_dput, req->rq_status = -ENOMEM);
595 body->handle.cookie = mfd->mfd_handle.h_cookie;
596 CDEBUG(D_INODE, "resend mfd %p, cookie "LPX64"\n", mfd,
597 mfd->mfd_handle.h_cookie);
607 /* do NOT or the MAY_*'s, you'll get the weakest */
608 static int accmode(int flags)
612 if (flags & FMODE_READ)
614 if (flags & (FMODE_WRITE|MDS_OPEN_TRUNC))
616 if (flags & FMODE_EXEC)
621 /* Handles object creation, actual opening, and I/O epoch */
622 static int mds_finish_open(struct ptlrpc_request *req, struct dentry *dchild,
623 struct mds_body *body, int flags, void **handle,
624 struct mds_update_record *rec,struct ldlm_reply *rep)
626 struct mds_obd *mds = mds_req2mds(req);
627 struct obd_device *obd = req->rq_export->exp_obd;
628 struct mds_file_data *mfd = NULL;
629 obd_id *ids = NULL; /* object IDs created */
634 /* atomically create objects if necessary */
635 down(&dchild->d_inode->i_sem);
636 mode = dchild->d_inode->i_mode;
637 if ((S_ISREG(mode) && !(body->valid & OBD_MD_FLEASIZE)) ||
638 (S_ISDIR(mode) && !(body->valid & OBD_MD_FLDIREA))) {
639 rc = mds_pack_md(obd, req->rq_repmsg, 2, body,
642 up(&dchild->d_inode->i_sem);
647 /* no EA: create objects */
648 rc = mds_create_objects(req, 2, rec, mds, obd,
649 dchild, handle, &ids);
651 CERROR("mds_create_objects: rc = %d\n", rc);
652 up(&dchild->d_inode->i_sem);
656 /* If the inode has EA data, then OSTs hold size, mtime */
657 if (S_ISREG(dchild->d_inode->i_mode) &&
658 !(body->valid & OBD_MD_FLEASIZE)) {
659 body->valid |= (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS |
660 OBD_MD_FLATIME | OBD_MD_FLMTIME);
662 up(&dchild->d_inode->i_sem);
664 intent_set_disposition(rep, DISP_OPEN_OPEN);
665 mfd = mds_dentry_open(dchild, mds->mds_vfsmnt, flags, req);
667 RETURN(PTR_ERR(mfd));
669 CDEBUG(D_INODE, "mfd %p, cookie "LPX64"\n", mfd,
670 mfd->mfd_handle.h_cookie);
672 mds_lov_update_objids(obd, ids);
673 OBD_FREE(ids, sizeof(*ids) * mds->mds_lov_desc.ld_tgt_count);
676 // mds_mfd_destroy(mfd);
680 static int mds_open_by_fid(struct ptlrpc_request *req, struct ll_fid *fid,
681 struct mds_body *body, int flags,
682 struct mds_update_record *rec,struct ldlm_reply *rep)
684 struct mds_obd *mds = mds_req2mds(req);
685 struct inode *pending_dir = mds->mds_pending_dir->d_inode;
686 struct dentry *dchild;
687 char fidname[LL_FID_NAMELEN];
692 down(&pending_dir->i_sem);
693 fidlen = ll_fid2str(fidname, fid->id, fid->generation);
694 dchild = lookup_one_len(fidname, mds->mds_pending_dir, fidlen);
695 if (IS_ERR(dchild)) {
696 up(&pending_dir->i_sem);
697 rc = PTR_ERR(dchild);
698 CERROR("error looking up %s in PENDING: rc = %d\n",fidname, rc);
702 if (dchild->d_inode != NULL) {
703 up(&pending_dir->i_sem);
704 mds_inode_set_orphan(dchild->d_inode);
705 mds_pack_inode2fid(req2obd(req), &body->fid1, dchild->d_inode);
706 mds_pack_inode2body(req2obd(req), body, dchild->d_inode);
707 intent_set_disposition(rep, DISP_LOOKUP_EXECD);
708 intent_set_disposition(rep, DISP_LOOKUP_POS);
709 CWARN("Orphan %s found and opened in PENDING directory\n",
714 up(&pending_dir->i_sem);
716 /* We didn't find it in PENDING so it isn't an orphan. See
717 * if it was a regular inode that was previously created. */
718 dchild = mds_fid2dentry(mds, fid, NULL);
720 RETURN(PTR_ERR(dchild));
722 mds_pack_inode2fid(req2obd(req), &body->fid1, dchild->d_inode);
723 mds_pack_inode2body(req2obd(req), body, dchild->d_inode);
724 intent_set_disposition(rep, DISP_LOOKUP_EXECD);
725 intent_set_disposition(rep, DISP_LOOKUP_POS);
728 rc = mds_finish_open(req, dchild, body, flags, &handle, rec, rep);
729 rc = mds_finish_transno(mds, dchild ? dchild->d_inode : NULL, handle,
730 req, rc, rep ? rep->lock_policy_res1 : 0);
731 /* XXX what do we do here if mds_finish_transno itself failed? */
737 int mds_pin(struct ptlrpc_request *req)
739 struct obd_device *obd = req->rq_export->exp_obd;
740 struct mds_body *request_body, *reply_body;
741 struct lvfs_run_ctxt saved;
742 int rc, size = sizeof(*reply_body);
745 request_body = lustre_msg_buf(req->rq_reqmsg, 0, sizeof(*request_body));
747 rc = lustre_pack_reply(req, 1, &size, NULL);
750 reply_body = lustre_msg_buf(req->rq_repmsg, 0, sizeof(*reply_body));
752 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
753 rc = mds_open_by_fid(req, &request_body->fid1, reply_body,
754 request_body->flags, NULL, NULL);
755 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
760 /* Get a lock on the ino to sync with creation WRT inode reuse (bug 2029).
761 * If child_lockh is NULL we just get the lock as a barrier to wait for
762 * other holders of this lock, and drop it right away again. */
763 int mds_lock_new_child(struct obd_device *obd, struct inode *inode,
764 struct lustre_handle *child_lockh)
766 struct ldlm_res_id child_res_id = { .name = { inode->i_ino, 0, 1, 0 } };
767 struct lustre_handle lockh;
771 if (child_lockh == NULL)
772 child_lockh = &lockh;
774 rc = ldlm_cli_enqueue(NULL, NULL, obd->obd_namespace, child_res_id,
775 LDLM_PLAIN, NULL, LCK_EX, &lock_flags,
776 mds_blocking_ast, ldlm_completion_ast, NULL, NULL,
777 NULL, 0, NULL, child_lockh);
779 CERROR("ldlm_cli_enqueue: %d\n", rc);
780 else if (child_lockh == &lockh)
781 ldlm_lock_decref(child_lockh, LCK_EX);
786 static int is_mount_object(struct dentry *dparent)
788 struct dentry *dchild;
790 if (!(dparent->d_inode->i_mode & S_ISUID))
793 dchild = lookup_one_len(".mntinfo", dparent, strlen(".mntinfo"));
794 if (IS_ERR(dchild) || dchild == NULL)
801 int mds_open(struct mds_update_record *rec, int offset,
802 struct ptlrpc_request *req, struct lustre_handle *child_lockh)
804 /* XXX ALLOCATE _something_ - 464 bytes on stack here */
805 struct obd_device *obd = req->rq_export->exp_obd;
806 struct mds_obd *mds = mds_req2mds(req);
807 struct ldlm_reply *rep = NULL;
808 struct mds_body *body = NULL;
809 struct dentry *dchild = NULL, *dparent = NULL;
810 struct mds_export_data *med;
811 struct lustre_handle parent_lockh[2];
812 int rc = 0, cleanup_phase = 0, acc_mode, created = 0;
813 int parent_mode = LCK_PR;
815 struct dentry_params dp;
816 struct mea *mea = NULL;
817 int mea_size, update_mode;
820 parent_lockh[0].cookie = 0;
821 parent_lockh[1].cookie = 0;
823 if (offset == 2) { /* intent */
824 rep = lustre_msg_buf(req->rq_repmsg, 0, sizeof (*rep));
825 body = lustre_msg_buf(req->rq_repmsg, 1, sizeof (*body));
826 } else if (offset == 0) { /* non-intent reint */
827 body = lustre_msg_buf(req->rq_repmsg, 0, sizeof (*body));
833 MDS_CHECK_RESENT(req, reconstruct_open(rec, offset, req, child_lockh));
835 /* Step 0: If we are passed a fid, then we assume the client already
836 * opened this file and is only replaying the RPC, so we open the
837 * inode by fid (at some large expense in security). */
838 if (lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY) {
839 DEBUG_REQ(D_HA, req, "open replay, disp: "LPX64"\n",
840 rep->lock_policy_res1);
842 LASSERT(rec->ur_fid2->id);
844 rc = mds_open_by_fid(req, rec->ur_fid2, body, rec->ur_flags,
848 /* We didn't find the correct inode on disk either, so we
849 * need to re-create it via a regular replay. */
850 LASSERT(rec->ur_flags & MDS_OPEN_CREAT);
852 LASSERT(!rec->ur_fid2->id);
855 LASSERT(offset == 2); /* If we got here, we must be called via intent */
857 med = &req->rq_export->exp_mds_data;
858 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_OPEN_PACK)) {
859 CERROR("test case OBD_FAIL_MDS_OPEN_PACK\n");
863 acc_mode = accmode(rec->ur_flags);
865 /* Step 1: Find and lock the parent */
866 if (rec->ur_flags & O_CREAT) {
867 /* XXX Well, in fact we only need this lock mode change if
868 in addition to O_CREAT, the file does not exist.
869 But we do not know if it exists or not yet */
870 parent_mode = LCK_PW;
873 if (rec->ur_namelen == 1) {
874 /* client (LMV) wants to open the file by fid */
875 CDEBUG(D_OTHER, "OPEN by fid %u/%u/%u\n",
876 (unsigned) rec->ur_fid1->mds,
877 (unsigned) rec->ur_fid1->id,
878 (unsigned) rec->ur_fid1->generation);
879 dchild = mds_fid2dentry(mds, rec->ur_fid1, NULL);
880 if (IS_ERR(dchild)) {
881 rc = PTR_ERR(dparent);
882 CERROR("child lookup by a fid error %d\n", rc);
888 dparent = mds_fid2locked_dentry(obd, rec->ur_fid1, NULL, parent_mode,
889 parent_lockh, &update_mode, rec->ur_name,
891 MDS_INODELOCK_UPDATE);
892 if (IS_ERR(dparent)) {
893 rc = PTR_ERR(dparent);
894 CERROR("parent lookup error %d\n", rc);
897 LASSERT(dparent->d_inode != NULL);
899 cleanup_phase = 1; /* parent dentry and lock */
901 /* try to retrieve MEA data for this dir */
902 rc = mds_get_lmv_attr(obd, dparent->d_inode, &mea, &mea_size);
907 /* dir is already splitted, check is requested filename
908 * should live at this MDS or at another one */
910 i = mea_name2idx(mea, rec->ur_name, rec->ur_namelen - 1);
911 if (mea->mea_master != i) {
912 CERROR("inapropriate MDS(%d) for %s. should be %d\n",
913 mea->mea_master, rec->ur_name, i);
914 GOTO(cleanup, rc = -ERESTART);
918 /* Step 2: Lookup the child */
919 dchild = ll_lookup_one_len(rec->ur_name, dparent, rec->ur_namelen - 1);
920 if (IS_ERR(dchild)) {
921 rc = PTR_ERR(dchild);
922 dchild = NULL; /* don't confuse mds_finish_transno() below */
927 cleanup_phase = 2; /* child dentry */
929 intent_set_disposition(rep, DISP_LOOKUP_EXECD);
931 if (dchild->d_flags & DCACHE_CROSS_REF) {
932 struct ldlm_res_id res_id = { . name = {0} };
933 ldlm_policy_data_t policy;
935 CDEBUG(D_OTHER, "cross reference: %lu/%lu/%lu\n",
936 (unsigned long) dchild->d_mdsnum,
937 (unsigned long) dchild->d_inum,
938 (unsigned long) dchild->d_generation);
939 body->valid |= OBD_MD_FLID | OBD_MD_MDS;
940 body->fid1.id = dchild->d_inum;
941 body->fid1.mds = dchild->d_mdsnum;
942 body->fid1.generation = dchild->d_generation;
943 intent_set_disposition(rep, DISP_LOOKUP_POS);
944 res_id.name[0] = dchild->d_inum;
945 res_id.name[1] = dchild->d_generation;
946 policy.l_inodebits.bits = MDS_INODELOCK_LOOKUP;
947 rc = ldlm_cli_enqueue(NULL, NULL, obd->obd_namespace,
948 res_id, LDLM_IBITS, &policy,
951 ldlm_completion_ast, NULL, NULL,
952 NULL, 0, NULL, child_lockh);
954 if (parent_lockh[1].cookie != 0)
955 ldlm_lock_decref(parent_lockh + 1, update_mode);
957 ldlm_lock_decref(parent_lockh, parent_mode);
959 OBD_FREE(mea, mea_size);
966 intent_set_disposition(rep, DISP_LOOKUP_POS);
968 intent_set_disposition(rep, DISP_LOOKUP_NEG);
970 /*Step 3: If the child was negative, and we're supposed to, create it.*/
971 if (dchild->d_inode == NULL) {
972 unsigned long ino = rec->ur_fid2->id;
976 if ((rc = mds_try_to_split_dir(obd, dparent, &mea, 0))) {
978 /* dir got splitted */
979 GOTO(cleanup, rc = -ERESTART);
981 /* error happened during spitting */
986 if (!(rec->ur_flags & MDS_OPEN_CREAT)) {
987 /* It's negative and we weren't supposed to create it */
988 GOTO(cleanup, rc = -ENOENT);
991 intent_set_disposition(rep, DISP_OPEN_CREATE);
992 handle = fsfilt_start(obd, dparent->d_inode, FSFILT_OP_CREATE,
994 if (IS_ERR(handle)) {
995 rc = PTR_ERR(handle);
999 dchild->d_fsdata = (void *) &dp;
1003 rc = ll_vfs_create(dparent->d_inode, dchild, rec->ur_mode,NULL);
1004 if (dchild->d_fsdata == (void *)(unsigned long)ino)
1005 dchild->d_fsdata = NULL;
1008 CDEBUG(D_INODE, "error during create: %d\n", rc);
1011 inode = dchild->d_inode;
1013 LASSERT(ino == inode->i_ino);
1014 /* Written as part of setattr */
1015 inode->i_generation = rec->ur_fid2->generation;
1016 CDEBUG(D_HA, "recreated ino %lu with gen %u\n",
1017 inode->i_ino, inode->i_generation);
1021 LTIME_S(iattr.ia_atime) = LTIME_S(rec->ur_time);
1022 LTIME_S(iattr.ia_ctime) = LTIME_S(rec->ur_time);
1023 LTIME_S(iattr.ia_mtime) = LTIME_S(rec->ur_time);
1025 iattr.ia_uid = rec->_ur_fsuid;
1026 if (dparent->d_inode->i_mode & S_ISGID)
1027 iattr.ia_gid = dparent->d_inode->i_gid;
1029 iattr.ia_gid = rec->_ur_fsgid;
1031 iattr.ia_valid = ATTR_UID | ATTR_GID | ATTR_ATIME |
1032 ATTR_MTIME | ATTR_CTIME;
1034 rc = fsfilt_setattr(obd, dchild, handle, &iattr, 0);
1036 CERROR("error on child setattr: rc = %d\n", rc);
1038 iattr.ia_valid = ATTR_MTIME | ATTR_CTIME;
1040 rc = fsfilt_setattr(obd, dparent, handle, &iattr, 0);
1042 CERROR("error on parent setattr: rc = %d\n", rc);
1044 acc_mode = 0; /* Don't check for permissions */
1047 LASSERT(!mds_inode_is_orphan(dchild->d_inode));
1049 mds_pack_inode2fid(obd, &body->fid1, dchild->d_inode);
1050 mds_pack_inode2body(obd, body, dchild->d_inode);
1052 if (S_ISREG(dchild->d_inode->i_mode)) {
1053 /* Check permissions etc */
1054 rc = ll_permission(dchild->d_inode, acc_mode, NULL);
1058 /* Can't write to a read-only file */
1059 if (IS_RDONLY(dchild->d_inode) && (acc_mode & MAY_WRITE) != 0)
1060 GOTO(cleanup, rc = -EPERM);
1062 /* An append-only file must be opened in append mode for
1064 if (IS_APPEND(dchild->d_inode) && (acc_mode & MAY_WRITE) != 0 &&
1065 ((rec->ur_flags & MDS_OPEN_APPEND) == 0 ||
1066 (rec->ur_flags & MDS_OPEN_TRUNC) != 0))
1067 GOTO(cleanup, rc = -EPERM);
1070 if (!created && (rec->ur_flags & MDS_OPEN_CREAT) &&
1071 (rec->ur_flags & MDS_OPEN_EXCL)) {
1072 /* File already exists, we didn't just create it, and we
1073 * were passed O_EXCL; err-or. */
1074 GOTO(cleanup, rc = -EEXIST); // returns a lock to the client
1077 /* if we are following a symlink, don't open */
1078 if (S_ISLNK(dchild->d_inode->i_mode))
1079 GOTO(cleanup, rc = 0);
1081 if ((rec->ur_flags & MDS_OPEN_DIRECTORY) &&
1082 !S_ISDIR(dchild->d_inode->i_mode))
1083 GOTO(cleanup, rc = -ENOTDIR);
1085 if (S_ISDIR(dchild->d_inode->i_mode)) {
1086 if (rec->ur_flags & MDS_OPEN_CREAT ||
1087 rec->ur_flags & FMODE_WRITE) {
1088 /*we are trying to create or write a exist dir*/
1089 GOTO(cleanup, rc = -EISDIR);
1091 if (ll_permission(dchild->d_inode, acc_mode, NULL)) {
1092 GOTO(cleanup, rc = -EACCES);
1094 if (is_mount_object(dchild)) {
1095 CERROR("Found possible GNS mount object %*s; not "
1096 "opening.\n", dchild->d_name.len,
1097 dchild->d_name.name);
1098 GOTO(cleanup, rc = 0); // success, but don't really open
1103 struct ldlm_res_id res_id = { . name = {0} };
1104 ldlm_policy_data_t policy;
1106 res_id.name[0] = dchild->d_inode->i_ino;
1107 res_id.name[1] = dchild->d_inode->i_generation;
1108 policy.l_inodebits.bits = MDS_INODELOCK_LOOKUP |
1109 MDS_INODELOCK_UPDATE;
1110 rc = ldlm_cli_enqueue(NULL, NULL, obd->obd_namespace,
1111 res_id, LDLM_IBITS, &policy,
1114 ldlm_completion_ast, NULL, NULL,
1115 NULL, 0, NULL, child_lockh);
1120 /* Step 5: mds_open it */
1121 rc = mds_finish_open(req, dchild, body, rec->ur_flags, &handle, rec,
1126 rc = mds_finish_transno(mds, dchild ? dchild->d_inode : NULL, handle,
1127 req, rc, rep ? rep->lock_policy_res1 : 0);
1129 switch (cleanup_phase) {
1132 ldlm_lock_decref(child_lockh, LCK_PR);
1134 if (rc && created) {
1135 int err = vfs_unlink(dparent->d_inode, dchild);
1137 CERROR("unlink(%*s) in error path: %d\n",
1138 dchild->d_name.len, dchild->d_name.name,
1141 } else if (created) {
1142 mds_lock_new_child(obd, dchild->d_inode, NULL);
1146 if (dparent == NULL)
1151 if (parent_lockh[1].cookie != 0)
1152 ldlm_lock_decref(parent_lockh + 1, update_mode);
1155 ldlm_lock_decref(parent_lockh, parent_mode);
1157 ptlrpc_save_lock (req, parent_lockh, parent_mode);
1160 atomic_inc(&mds->mds_open_count);
1162 OBD_FREE(mea, mea_size);
1163 if ((cleanup_phase != 3) && !rc)
1168 /* Close a "file descriptor" and possibly unlink an orphan from the
1169 * PENDING directory.
1171 * If we are being called from mds_disconnect() because the client has
1172 * disappeared, then req == NULL and we do not update last_rcvd because
1173 * there is nothing that could be recovered by the client at this stage
1174 * (it will not even _have_ an entry in last_rcvd anymore).
1176 * Returns EAGAIN if the client needs to get more data and re-close. */
1177 int mds_mfd_close(struct ptlrpc_request *req, struct obd_device *obd,
1178 struct mds_file_data *mfd, int unlink_orphan)
1180 struct inode *inode = mfd->mfd_dentry->d_inode;
1181 char fidname[LL_FID_NAMELEN];
1182 int last_orphan, fidlen, rc = 0, cleanup_phase = 0;
1183 struct dentry *pending_child = NULL;
1184 struct mds_obd *mds = &obd->u.mds;
1185 struct inode *pending_dir = mds->mds_pending_dir->d_inode;
1186 void *handle = NULL;
1187 struct mds_body *request_body = NULL, *reply_body = NULL;
1188 struct dentry_params dp;
1189 struct iattr iattr = { 0 };
1192 if (req && req->rq_reqmsg != NULL)
1193 request_body = lustre_msg_buf(req->rq_reqmsg, 0,
1194 sizeof(*request_body));
1195 if (req && req->rq_repmsg != NULL)
1196 reply_body = lustre_msg_buf(req->rq_repmsg, 0,
1197 sizeof(*reply_body));
1199 fidlen = ll_fid2str(fidname, inode->i_ino, inode->i_generation);
1201 last_orphan = mds_open_orphan_dec_test(inode) &&
1202 mds_inode_is_orphan(inode);
1204 /* this is half of the actual "close" */
1205 if (mfd->mfd_mode & FMODE_WRITE) {
1206 rc = mds_put_write_access(mds, inode, request_body,
1207 last_orphan && unlink_orphan);
1208 } else if (mfd->mfd_mode & FMODE_EXEC) {
1209 mds_allow_write_access(inode);
1212 if (last_orphan && unlink_orphan) {
1213 struct lov_mds_md *lmm = NULL;
1214 int stripe_count = 0;
1215 LASSERT(rc == 0); /* mds_put_write_access must have succeeded */
1217 if (obd->obd_recovering) {
1218 CDEBUG(D_HA, "not remove orphan %s until recovery"
1219 " is over\n", fidname);
1223 CDEBUG(D_HA, "destroying orphan object %s\n", fidname);
1225 /* Sadly, there is no easy way to save pending_child from
1226 * mds_reint_unlink() into mfd, so we need to re-lookup,
1227 * but normally it will still be in the dcache. */
1228 pending_dir = mds->mds_pending_dir->d_inode;
1229 down(&pending_dir->i_sem);
1230 cleanup_phase = 1; /* up(i_sem) when finished */
1231 pending_child = lookup_one_len(fidname, mds->mds_pending_dir,
1233 if (IS_ERR(pending_child))
1234 GOTO(cleanup, rc = PTR_ERR(pending_child));
1235 LASSERT(pending_child->d_inode != NULL);
1237 cleanup_phase = 2; /* dput(pending_child) when finished */
1238 if (req != NULL && req->rq_repmsg != NULL) {
1239 lmm = lustre_msg_buf(req->rq_repmsg, 1, 0);
1240 stripe_count = le32_to_cpu(lmm->lmm_stripe_count);
1243 handle = fsfilt_start_log(obd, pending_dir, FSFILT_OP_UNLINK,
1244 NULL, stripe_count);
1245 if (IS_ERR(handle)) {
1246 rc = PTR_ERR(handle);
1251 if (req != NULL && req->rq_repmsg != NULL &&
1252 (reply_body->valid & OBD_MD_FLEASIZE) &&
1253 mds_log_op_unlink(obd, pending_child->d_inode, lmm,
1254 req->rq_repmsg->buflens[1],
1255 lustre_msg_buf(req->rq_repmsg, 2, 0),
1256 req->rq_repmsg->buflens[2]) > 0) {
1257 reply_body->valid |= OBD_MD_FLCOOKIE;
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 goto out; /* Don't bother updating attrs on unlinked inode */
1273 if (request_body != NULL && mfd->mfd_mode & FMODE_WRITE && rc == 0) {
1274 /* Update the on-disk attributes if this was the last write
1275 * close, and all information was provided (i.e., rc == 0)
1277 * XXX this should probably be abstracted with mds_reint_setattr
1281 if (request_body->valid & OBD_MD_FLMTIME &&
1282 LTIME_S(request_body->mtime) > LTIME_S(inode->i_mtime)) {
1283 LTIME_S(iattr.ia_mtime) = LTIME_S(request_body->mtime);
1284 iattr.ia_valid |= ATTR_MTIME;
1286 if (request_body->valid & OBD_MD_FLCTIME &&
1287 LTIME_S(request_body->ctime) > LTIME_S(inode->i_ctime)) {
1288 LTIME_S(iattr.ia_ctime) = LTIME_S(request_body->ctime);
1289 iattr.ia_valid |= ATTR_CTIME;
1292 /* XXX can't set block count with fsfilt_setattr (!) */
1293 if (request_body->valid & OBD_MD_FLSIZE) {
1294 iattr.ia_valid |= ATTR_SIZE;
1295 iattr.ia_size = request_body->size;
1297 /* if (request_body->valid & OBD_MD_FLBLOCKS) {
1298 iattr.ia_valid |= ATTR_BLOCKS;
1299 iattr.ia_blocks = request_body->blocks
1305 if (request_body != NULL && request_body->valid & OBD_MD_FLATIME) {
1306 /* Only start a transaction to write out only the atime if
1307 * it is more out-of-date than the specified limit. If we
1308 * are already going to write out the atime then do it anyway.
1310 if ((LTIME_S(request_body->atime) >
1311 LTIME_S(inode->i_atime) + MAX_ATIME_DIFF) ||
1312 (iattr.ia_valid != 0 &&
1313 LTIME_S(request_body->atime) > LTIME_S(inode->i_atime))) {
1314 LTIME_S(iattr.ia_atime) = LTIME_S(request_body->atime);
1315 iattr.ia_valid |= ATTR_ATIME;
1319 if (iattr.ia_valid != 0) {
1320 handle = fsfilt_start(obd, inode, FSFILT_OP_SETATTR, NULL);
1322 GOTO(cleanup, rc = PTR_ERR(handle));
1323 rc = fsfilt_setattr(obd, mfd->mfd_dentry, handle, &iattr, 0);
1325 CERROR("error in setattr(%s): rc %d\n", fidname, rc);
1328 /* If other clients have this file open for write, rc will be > 0 */
1331 l_dput(mfd->mfd_dentry);
1332 mds_mfd_destroy(mfd);
1335 atomic_dec(&mds->mds_open_count);
1336 if (req != NULL && reply_body != NULL) {
1337 rc = mds_finish_transno(mds, pending_dir, handle, req, rc, 0);
1338 } else if (handle) {
1339 int err = fsfilt_commit(obd, mds->mds_sb, pending_dir, handle, 0);
1341 CERROR("error committing close: %d\n", err);
1347 switch (cleanup_phase) {
1349 dput(pending_child);
1351 up(&pending_dir->i_sem);
1356 int mds_close(struct ptlrpc_request *req)
1358 struct mds_export_data *med = &req->rq_export->exp_mds_data;
1359 struct obd_device *obd = req->rq_export->exp_obd;
1360 struct mds_body *body;
1361 struct mds_file_data *mfd;
1362 struct lvfs_run_ctxt saved;
1363 struct inode *inode;
1364 int rc, repsize[3] = {sizeof(struct mds_body),
1365 obd->u.mds.mds_max_mdsize,
1366 obd->u.mds.mds_max_cookiesize};
1369 rc = lustre_pack_reply(req, 3, repsize, NULL);
1371 CERROR("lustre_pack_reply: rc = %d\n", rc);
1372 req->rq_status = rc;
1374 MDS_CHECK_RESENT(req, mds_reconstruct_generic(req));
1377 body = lustre_swab_reqbuf(req, 0, sizeof(*body), lustre_swab_mds_body);
1379 CERROR("Can't unpack body\n");
1380 req->rq_status = -EFAULT;
1384 if (body->flags & MDS_BFLAG_UNCOMMITTED_WRITES)
1385 /* do some stuff */ ;
1387 mfd = mds_handle2mfd(&body->handle);
1389 DEBUG_REQ(D_ERROR, req, "no handle for file close ino "LPD64
1390 ": cookie "LPX64, body->fid1.id, body->handle.cookie);
1391 req->rq_status = -ESTALE;
1395 inode = mfd->mfd_dentry->d_inode;
1396 if (mds_inode_is_orphan(inode) && mds_open_orphan_count(inode) == 1) {
1397 body = lustre_msg_buf(req->rq_repmsg, 0, sizeof (*body));
1398 LASSERT(body != NULL);
1400 mds_pack_inode2fid(obd, &body->fid1, inode);
1401 mds_pack_inode2body(obd, body, inode);
1402 mds_pack_md(obd, req->rq_repmsg, 1, body, inode, 1);
1404 spin_lock(&med->med_open_lock);
1405 list_del(&mfd->mfd_list);
1406 spin_unlock(&med->med_open_lock);
1408 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
1409 req->rq_status = mds_mfd_close(req, obd, mfd, 1);
1410 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
1412 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_CLOSE_PACK)) {
1413 CERROR("test case OBD_FAIL_MDS_CLOSE_PACK\n");
1414 req->rq_status = -ENOMEM;
1423 int mds_done_writing(struct ptlrpc_request *req)
1425 struct mds_body *body;
1426 int rc, size = sizeof(struct mds_body);
1429 MDS_CHECK_RESENT(req, mds_reconstruct_generic(req));
1431 body = lustre_swab_reqbuf(req, 0, sizeof(*body), lustre_swab_mds_body);
1433 CERROR("Can't unpack body\n");
1434 req->rq_status = -EFAULT;
1438 rc = lustre_pack_reply(req, 1, &size, NULL);
1440 CERROR("lustre_pack_reply: rc = %d\n", rc);
1441 req->rq_status = rc;