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 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/version.h>
35 #include <linux/module.h>
36 #include <linux/init.h>
37 #include <linux/version.h>
38 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,5,0))
39 # include <linux/buffer_head.h>
40 # include <linux/workqueue.h>
42 # include <linux/locks.h>
45 #include <linux/obd_class.h>
46 #include <linux/obd_lov.h>
47 #include <linux/lustre_fsfilt.h>
48 #include <linux/lprocfs_status.h>
50 #include "mds_internal.h"
52 /* Exported function from this file are:
54 * mds_open - called by the intent handler
55 * mds_close - an rpc handling function
56 * mds_pin - an rpc handling function - which will go away
57 * mds_mfd_close - for force closing files when a client dies
61 * MDS file data handling: file data holds a handle for a file opened
65 static void mds_mfd_addref(void *mfdp)
67 struct mds_file_data *mfd = mfdp;
69 atomic_inc(&mfd->mfd_refcount);
70 CDEBUG(D_INFO, "GETting mfd %p : new refcount %d\n", mfd,
71 atomic_read(&mfd->mfd_refcount));
74 /* Create a new mds_file_data struct.
75 * One reference for handle+med_open_head list and dropped by mds_mfd_unlink(),
76 * one reference for the caller of this function. */
77 struct mds_file_data *mds_mfd_new(void)
79 struct mds_file_data *mfd;
81 OBD_ALLOC(mfd, sizeof *mfd);
83 CERROR("mds: out of memory\n");
87 atomic_set(&mfd->mfd_refcount, 2);
89 INIT_LIST_HEAD(&mfd->mfd_handle.h_link);
90 INIT_LIST_HEAD(&mfd->mfd_list);
91 class_handle_hash(&mfd->mfd_handle, mds_mfd_addref);
96 /* Get a new reference on the mfd pointed to by handle, if handle is still
97 * valid. Caller must drop reference with mds_mfd_put(). */
98 static struct mds_file_data *mds_handle2mfd(struct lustre_handle *handle)
101 LASSERT(handle != NULL);
102 RETURN(class_handle2object(handle->cookie));
105 /* Drop mfd reference, freeing struct if this is the last one. */
106 static void mds_mfd_put(struct mds_file_data *mfd)
108 CDEBUG(D_INFO, "PUTting mfd %p : new refcount %d\n", mfd,
109 atomic_read(&mfd->mfd_refcount) - 1);
110 LASSERT(atomic_read(&mfd->mfd_refcount) > 0 &&
111 atomic_read(&mfd->mfd_refcount) < 0x5a5a);
112 if (atomic_dec_and_test(&mfd->mfd_refcount)) {
113 LASSERT(list_empty(&mfd->mfd_handle.h_link));
114 OBD_FREE(mfd, sizeof *mfd);
118 /* Remove the mfd handle so that it cannot be found by open/close again.
119 * Caller must hold med_open_lock for mfd_list manipulation. */
120 void mds_mfd_unlink(struct mds_file_data *mfd, int decref)
122 class_handle_unhash(&mfd->mfd_handle);
123 list_del_init(&mfd->mfd_list);
128 /* Caller must hold mds->mds_epoch_sem */
129 static int mds_alloc_filterdata(struct inode *inode)
131 LASSERT(inode->i_filterdata == NULL);
132 OBD_ALLOC(inode->i_filterdata, sizeof(struct mds_filter_data));
133 if (inode->i_filterdata == NULL)
135 LASSERT(igrab(inode) == inode);
139 /* Caller must hold mds->mds_epoch_sem */
140 static void mds_free_filterdata(struct inode *inode)
142 LASSERT(inode->i_filterdata != NULL);
143 OBD_FREE(inode->i_filterdata, sizeof(struct mds_filter_data));
144 inode->i_filterdata = NULL;
148 /* Write access to a file: executors cause a negative count,
149 * writers a positive count. The semaphore is needed to perform
150 * a check for the sign and then increment or decrement atomically.
152 * This code is closely tied to the allocation of the d_fsdata and the
153 * MDS epoch, so we use the same semaphore for the whole lot.
155 * We could use a different semaphore for each file, if it ever shows
156 * up in a profile, which it won't.
158 * epoch argument is nonzero during recovery */
159 static int mds_get_write_access(struct mds_obd *mds, struct inode *inode,
164 down(&mds->mds_epoch_sem);
166 if (atomic_read(&inode->i_writecount) < 0) {
167 up(&mds->mds_epoch_sem);
172 if (MDS_FILTERDATA(inode) && MDS_FILTERDATA(inode)->io_epoch != 0) {
173 CDEBUG(D_INODE, "continuing MDS epoch "LPU64" for ino %lu/%u\n",
174 MDS_FILTERDATA(inode)->io_epoch, inode->i_ino,
175 inode->i_generation);
179 if (inode->i_filterdata == NULL)
180 mds_alloc_filterdata(inode);
181 if (inode->i_filterdata == NULL) {
185 if (epoch > mds->mds_io_epoch)
186 mds->mds_io_epoch = epoch;
189 MDS_FILTERDATA(inode)->io_epoch = mds->mds_io_epoch;
190 CDEBUG(D_INODE, "starting MDS epoch "LPU64" for ino %lu/%u\n",
191 mds->mds_io_epoch, inode->i_ino, inode->i_generation);
194 atomic_inc(&inode->i_writecount);
195 up(&mds->mds_epoch_sem);
199 /* Returns EAGAIN if the client needs to get size and/or cookies and close
200 * again -- which is never true if the file is about to be unlinked. Otherwise
201 * returns the number of remaining writers. */
202 static int mds_put_write_access(struct mds_obd *mds, struct inode *inode,
203 struct mds_body *body, int unlinking)
208 down(&mds->mds_epoch_sem);
209 atomic_dec(&inode->i_writecount);
210 rc = atomic_read(&inode->i_writecount);
214 if (!unlinking && !(body->valid & OBD_MD_FLSIZE))
215 GOTO(out, rc = EAGAIN);
217 mds_free_filterdata(inode);
219 up(&mds->mds_epoch_sem);
223 static int mds_deny_write_access(struct mds_obd *mds, struct inode *inode)
226 down(&mds->mds_epoch_sem);
227 if (atomic_read(&inode->i_writecount) > 0) {
228 up(&mds->mds_epoch_sem);
231 atomic_dec(&inode->i_writecount);
232 up(&mds->mds_epoch_sem);
236 static void mds_allow_write_access(struct inode *inode)
239 atomic_inc(&inode->i_writecount);
242 int mds_query_write_access(struct inode *inode)
245 RETURN(atomic_read(&inode->i_writecount));
248 /* This replaces the VFS dentry_open, it manages mfd and writecount */
249 static struct mds_file_data *mds_dentry_open(struct dentry *dentry,
250 struct vfsmount *mnt, int flags,
251 struct ptlrpc_request *req)
253 struct mds_export_data *med = &req->rq_export->exp_mds_data;
254 struct mds_obd *mds = mds_req2mds(req);
255 struct mds_file_data *mfd;
256 struct mds_body *body;
262 CERROR("mds: out of memory\n");
263 GOTO(cleanup_dentry, error = -ENOMEM);
266 body = lustre_msg_buf(req->rq_repmsg, 1, sizeof (*body));
268 if (flags & FMODE_WRITE) {
269 /* FIXME: in recovery, need to pass old epoch here */
270 error = mds_get_write_access(mds, dentry->d_inode, 0);
272 GOTO(cleanup_mfd, error);
273 body->io_epoch = MDS_FILTERDATA(dentry->d_inode)->io_epoch;
274 } else if (flags & FMODE_EXEC) {
275 error = mds_deny_write_access(mds, dentry->d_inode);
277 GOTO(cleanup_mfd, error);
282 /* Mark the file as open to handle open-unlink. */
283 MDS_DOWN_WRITE_ORPHAN_SEM(dentry->d_inode);
284 mds_orphan_open_inc(dentry->d_inode);
285 MDS_UP_WRITE_ORPHAN_SEM(dentry->d_inode);
287 mfd->mfd_mode = flags;
288 mfd->mfd_dentry = dentry;
289 mfd->mfd_xid = req->rq_xid;
291 spin_lock(&med->med_open_lock);
292 list_add(&mfd->mfd_list, &med->med_open_head);
293 spin_unlock(&med->med_open_lock);
295 body->handle.cookie = mfd->mfd_handle.h_cookie;
301 mds_mfd_unlink(mfd, 1);
303 return ERR_PTR(error);
306 /* Must be called with i_sem held */
307 static int mds_create_objects(struct ptlrpc_request *req, int offset,
308 struct mds_update_record *rec,
309 struct mds_obd *mds, struct obd_device *obd,
310 struct dentry *dchild, void **handle,
313 struct inode *inode = dchild->d_inode;
314 struct obd_trans_info oti = { 0 };
315 struct lov_stripe_md *lsm = NULL;
316 struct lov_mds_md *lmm = NULL;
318 struct mds_body *body;
323 if (!S_ISREG(inode->i_mode))
325 if (rec->ur_flags & MDS_OPEN_DELAY_CREATE ||
326 !(rec->ur_flags & FMODE_WRITE))
329 body = lustre_msg_buf(req->rq_repmsg, 1, sizeof(*body));
331 if (body->valid & OBD_MD_FLEASIZE)
334 OBD_ALLOC(*ids, mds->mds_lov_desc.ld_tgt_count * sizeof(**ids));
338 oti.oti_objid = *ids;
341 if (lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY) {
342 if (rec->ur_fid2->id == 0) {
343 DEBUG_REQ(D_ERROR, req, "fid2 not set on open replay");
347 body->valid |= OBD_MD_FLBLKSZ | OBD_MD_FLEASIZE;
348 lmm_size = rec->ur_eadatalen;
349 lmm = rec->ur_eadata;
353 *handle = fsfilt_start(obd,inode,FSFILT_OP_CREATE,NULL);
354 if (IS_ERR(*handle)) {
355 rc = PTR_ERR(*handle);
360 mds_objids_from_lmm(*ids, lmm, &mds->mds_lov_desc);
362 rc = fsfilt_set_md(obd, inode, *handle, lmm, lmm_size, "lov");
363 lmm_buf = lustre_msg_buf(req->rq_repmsg, offset, lmm_size);
365 if (!rc) rc = -ENOMEM;
367 memcpy(lmm_buf, lmm, lmm_size);
370 CERROR("open replay failed to set md:%d\n", rc);
374 if (OBD_FAIL_CHECK_ONCE(OBD_FAIL_MDS_ALLOC_OBDO))
375 GOTO(out_ids, rc = -ENOMEM);
379 GOTO(out_ids, rc = -ENOMEM);
380 oa->o_uid = 0; /* must have 0 uid / gid on OST */
382 oa->o_mode = S_IFREG | 0600;
383 oa->o_id = inode->i_ino;
384 oa->o_valid = OBD_MD_FLID | OBD_MD_FLTYPE | OBD_MD_FLFLAGS |
385 OBD_MD_FLMODE | OBD_MD_FLUID | OBD_MD_FLGID;
388 obdo_from_inode(oa, inode, OBD_MD_FLTYPE | OBD_MD_FLATIME |
389 OBD_MD_FLMTIME | OBD_MD_FLCTIME);
391 if (!(rec->ur_flags & MDS_OPEN_HAS_OBJS)) {
392 /* check if things like lfs setstripe are sending us the ea */
393 if (rec->ur_flags & MDS_OPEN_HAS_EA) {
394 rc = obd_iocontrol(OBD_IOC_LOV_SETSTRIPE,
396 0, &lsm, rec->ur_eadata);
400 OBD_ALLOC(lmm, mds->mds_max_mdsize);
402 GOTO(out_oa, rc = -ENOMEM);
404 lmm_size = mds->mds_max_mdsize;
405 rc = mds_get_md(obd, dchild->d_parent->d_inode,
408 rc = obd_iocontrol(OBD_IOC_LOV_SETSTRIPE,
411 OBD_FREE(lmm, mds->mds_max_mdsize);
415 rc = obd_create(mds->mds_osc_exp, oa, &lsm, &oti);
420 CDEBUG(level, "error creating objects for "
421 "inode %lu: rc = %d\n",
424 CERROR("obd_create returned invalid "
431 rc = obd_iocontrol(OBD_IOC_LOV_SETEA, mds->mds_osc_exp,
432 0, &lsm, rec->ur_eadata);
436 lsm->lsm_object_id = oa->o_id;
439 oa->o_size = inode->i_size;
440 obdo_from_inode(oa, inode, OBD_MD_FLTYPE | OBD_MD_FLATIME |
441 OBD_MD_FLMTIME | OBD_MD_FLCTIME | OBD_MD_FLSIZE);
443 /* pack lustre id to OST */
444 oa->o_fid = body->fid1.id;
445 oa->o_generation = body->fid1.generation;
446 oa->o_valid |= OBD_MD_FLFID | OBD_MD_FLGENER;
448 rc = obd_setattr(mds->mds_osc_exp, oa, lsm, &oti);
450 CERROR("error setting attrs for inode %lu: rc %d\n",
453 CERROR("obd_setattr returned bad rc %d\n", rc);
460 body->valid |= OBD_MD_FLBLKSZ | OBD_MD_FLEASIZE;
461 obdo_refresh_inode(inode, oa, OBD_MD_FLBLKSZ);
463 LASSERT(lsm && lsm->lsm_object_id);
465 rc = obd_packmd(mds->mds_osc_exp, &lmm, lsm);
467 CERROR("cannot pack lsm, err = %d\n", rc);
471 body->eadatasize = rc;
474 *handle = fsfilt_start(obd, inode, FSFILT_OP_CREATE, NULL);
475 if (IS_ERR(*handle)) {
476 rc = PTR_ERR(*handle);
481 rc = fsfilt_set_md(obd, inode, *handle, lmm, lmm_size, "lov");
482 lmm_buf = lustre_msg_buf(req->rq_repmsg, offset, lmm_size);
484 if (!rc) rc = -ENOMEM;
486 memcpy(lmm_buf, lmm, lmm_size);
488 obd_free_diskmd(mds->mds_osc_exp, &lmm);
490 oti_free_cookies(&oti);
494 OBD_FREE(*ids, mds->mds_lov_desc.ld_tgt_count * sizeof(**ids));
498 obd_free_memmd(mds->mds_osc_exp, &lsm);
502 static void reconstruct_open(struct mds_update_record *rec, int offset,
503 struct ptlrpc_request *req,
504 struct lustre_handle *child_lockh)
506 struct mds_export_data *med = &req->rq_export->exp_mds_data;
507 struct mds_client_data *mcd = med->med_mcd;
508 struct mds_obd *mds = mds_req2mds(req);
509 struct mds_file_data *mfd;
510 struct obd_device *obd = req->rq_export->exp_obd;
511 struct dentry *parent, *dchild;
512 struct ldlm_reply *rep;
513 struct mds_body *body;
519 LASSERT(offset == 2); /* only called via intent */
520 rep = lustre_msg_buf(req->rq_repmsg, 0, sizeof (*rep));
521 body = lustre_msg_buf(req->rq_repmsg, 1, sizeof (*body));
523 /* copy rc, transno and disp; steal locks */
524 mds_req_from_mcd(req, mcd);
525 intent_set_disposition(rep, mcd->mcd_last_data);
527 /* Only replay if create or open actually happened. */
528 if (!intent_disposition(rep, DISP_OPEN_CREATE | DISP_OPEN_OPEN) ) {
530 return; /* error looking up parent or child */
533 parent = mds_fid2dentry(mds, rec->ur_fid1, NULL);
534 LASSERT(!IS_ERR(parent));
536 dchild = ll_lookup_one_len(rec->ur_name, parent, rec->ur_namelen - 1);
537 LASSERT(!IS_ERR(dchild));
539 if (!dchild->d_inode)
540 GOTO(out_dput, 0); /* child not present to open */
542 /* At this point, we know we have a child. We'll send
543 * it back _unless_ it not created and open failed.
545 if (intent_disposition(rep, DISP_OPEN_OPEN) &&
546 !intent_disposition(rep, DISP_OPEN_CREATE) &&
551 mds_pack_inode2fid(&body->fid1, dchild->d_inode);
552 mds_pack_inode2body(body, dchild->d_inode);
553 if (S_ISREG(dchild->d_inode->i_mode)) {
554 rc = mds_pack_md(obd, req->rq_repmsg, 2, body,
558 LASSERT(rc == req->rq_status);
560 /* If we have LOV EA data, the OST holds size, mtime */
561 if (!(body->valid & OBD_MD_FLEASIZE))
562 body->valid |= (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS |
563 OBD_MD_FLATIME | OBD_MD_FLMTIME);
566 if (!(rec->ur_flags & MDS_OPEN_JOIN_FILE))
567 lustre_shrink_reply(req, 2, body->eadatasize, 0);
569 if (req->rq_export->exp_connect_flags & OBD_CONNECT_ACL &&
570 !(rec->ur_flags & MDS_OPEN_JOIN_FILE)) {
571 int acl_off = body->eadatasize ? 3 : 2;
573 rc = mds_pack_acl(med, dchild->d_inode, req->rq_repmsg,
575 lustre_shrink_reply(req, acl_off, body->aclsize, 0);
576 if (!req->rq_status && rc)
580 /* If we have -EEXIST as the status, and we were asked to create
581 * exclusively, we can tell we failed because the file already existed.
583 if (req->rq_status == -EEXIST &&
584 ((rec->ur_flags & (MDS_OPEN_CREAT | MDS_OPEN_EXCL)) ==
585 (MDS_OPEN_CREAT | MDS_OPEN_EXCL))) {
589 /* If we didn't get as far as trying to open, then some locking thing
590 * probably went wrong, and we'll just bail here.
592 if (!intent_disposition(rep, DISP_OPEN_OPEN))
595 /* If we failed, then we must have failed opening, so don't look for
596 * file descriptor or anything, just give the client the bad news.
602 spin_lock(&med->med_open_lock);
603 list_for_each(t, &med->med_open_head) {
604 mfd = list_entry(t, struct mds_file_data, mfd_list);
605 if (mfd->mfd_xid == req->rq_xid) {
611 spin_unlock(&med->med_open_lock);
613 /* #warning "XXX fixme" bug 2991 */
614 /* Here it used to LASSERT(mfd) if exp_outstanding_reply != NULL.
615 * Now that exp_outstanding_reply is a list, it's just using mfd != NULL
616 * to detect a re-open */
618 if (rec->ur_flags & MDS_OPEN_JOIN_FILE) {
619 rc = mds_join_file(rec, req, dchild, NULL);
623 mntget(mds->mds_vfsmnt);
624 CERROR("Re-opened file \n");
625 mfd = mds_dentry_open(dchild, mds->mds_vfsmnt,
626 rec->ur_flags & ~MDS_OPEN_TRUNC, req);
628 CERROR("mds: out of memory\n");
629 GOTO(out_dput, req->rq_status = -ENOMEM);
633 body->handle.cookie = mfd->mfd_handle.h_cookie;
634 CDEBUG(D_INODE, "resend mfd %p, cookie "LPX64"\n", mfd,
635 mfd->mfd_handle.h_cookie);
647 /* do NOT or the MAY_*'s, you'll get the weakest */
648 static int accmode(struct inode *inode, int flags)
652 /* Sadly, NFSD reopens a file repeatedly during operation, so the
653 * "acc_mode = 0" allowance for newly-created files isn't honoured.
654 * NFSD uses the MDS_OPEN_OWNEROVERRIDE flag to say that a file
655 * owner can write to a file even if it is marked readonly to hide
656 * its brokenness. (bug 5781) */
657 if (flags & MDS_OPEN_OWNEROVERRIDE && inode->i_uid == current->fsuid)
660 if (flags & FMODE_READ)
662 if (flags & (FMODE_WRITE|MDS_OPEN_TRUNC))
664 if (flags & FMODE_EXEC)
669 /* Handles object creation, actual opening, and I/O epoch */
670 static int mds_finish_open(struct ptlrpc_request *req, struct dentry *dchild,
671 struct mds_body *body, int flags, void **handle,
672 struct mds_update_record *rec,struct ldlm_reply *rep,
673 struct lustre_handle *lockh)
675 struct mds_obd *mds = mds_req2mds(req);
676 struct obd_device *obd = req->rq_export->exp_obd;
677 struct mds_file_data *mfd = NULL;
678 obd_id *ids = NULL; /* object IDs created */
682 /* atomically create objects if necessary */
683 down(&dchild->d_inode->i_sem);
685 if (S_ISREG(dchild->d_inode->i_mode) &&
686 !(body->valid & OBD_MD_FLEASIZE)) {
687 rc = mds_pack_md(obd, req->rq_repmsg, 2, body,
690 up(&dchild->d_inode->i_sem);
695 if ((body->valid & OBD_MD_FLEASIZE) &&
696 (rec->ur_flags & MDS_OPEN_HAS_EA)) {
697 up(&dchild->d_inode->i_sem);
700 if (rec->ur_flags & MDS_OPEN_JOIN_FILE) {
701 up(&dchild->d_inode->i_sem);
702 rc = mds_join_file(rec, req, dchild, lockh);
705 down(&dchild->d_inode->i_sem);
707 if (!(body->valid & OBD_MD_FLEASIZE) &&
708 !(body->valid & OBD_MD_FLMODEASIZE)) {
709 /* no EA: create objects */
710 rc = mds_create_objects(req, 2, rec, mds, obd,
711 dchild, handle, &ids);
713 CERROR("mds_create_objects: rc = %d\n", rc);
714 up(&dchild->d_inode->i_sem);
719 /* If the inode has no EA data, then MDS holds size, mtime */
720 if (S_ISREG(dchild->d_inode->i_mode) &&
721 !(body->valid & OBD_MD_FLEASIZE)) {
722 body->valid |= (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS |
723 OBD_MD_FLATIME | OBD_MD_FLMTIME);
725 up(&dchild->d_inode->i_sem);
727 if (!(rec->ur_flags & MDS_OPEN_JOIN_FILE))
728 lustre_shrink_reply(req, 2, body->eadatasize, 0);
730 if (req->rq_export->exp_connect_flags & OBD_CONNECT_ACL &&
731 !(rec->ur_flags & MDS_OPEN_JOIN_FILE)) {
732 int acl_off = body->eadatasize ? 3 : 2;
734 rc = mds_pack_acl(&req->rq_export->exp_mds_data,
735 dchild->d_inode, req->rq_repmsg,
737 lustre_shrink_reply(req, acl_off, body->aclsize, 0);
742 intent_set_disposition(rep, DISP_OPEN_OPEN);
743 mfd = mds_dentry_open(dchild, mds->mds_vfsmnt, flags, req);
745 RETURN(PTR_ERR(mfd));
747 CDEBUG(D_INODE, "mfd %p, cookie "LPX64"\n", mfd,
748 mfd->mfd_handle.h_cookie);
751 mds_lov_update_objids(obd, ids);
752 OBD_FREE(ids, sizeof(*ids) * mds->mds_lov_desc.ld_tgt_count);
755 mds_mfd_unlink(mfd, 1);
760 static int mds_open_by_fid(struct ptlrpc_request *req, struct lu_fid *fid,
761 struct mds_body *body, int flags,
762 struct mds_update_record *rec,struct ldlm_reply *rep)
764 struct mds_obd *mds = mds_req2mds(req);
765 struct dentry *dchild;
766 char fidname[LU_FID_NAMELEN];
771 fidlen = lu_fid2str(fidname, fid->id, fid->generation);
772 dchild = ll_lookup_one_len(fidname, mds->mds_pending_dir, fidlen);
773 if (IS_ERR(dchild)) {
774 rc = PTR_ERR(dchild);
775 CERROR("error looking up %s in PENDING: rc = %d\n",fidname, rc);
779 if (dchild->d_inode != NULL) {
780 mds_inode_set_orphan(dchild->d_inode);
781 CWARN("Orphan %s found and opened in PENDING directory\n",
786 /* We didn't find it in PENDING so it isn't an orphan. See
787 * if it was a regular inode that was previously created. */
788 dchild = mds_fid2dentry(mds, fid, NULL);
790 RETURN(PTR_ERR(dchild));
793 mds_pack_inode2fid(&body->fid1, dchild->d_inode);
794 mds_pack_inode2body(body, dchild->d_inode);
795 intent_set_disposition(rep, DISP_LOOKUP_EXECD);
796 intent_set_disposition(rep, DISP_LOOKUP_POS);
798 rc = mds_finish_open(req, dchild, body, flags, &handle, rec, rep, NULL);
799 rc = mds_finish_transno(mds, dchild->d_inode, handle,
800 req, rc, rep ? rep->lock_policy_res1 : 0);
801 /* XXX what do we do here if mds_finish_transno itself failed? */
807 int mds_pin(struct ptlrpc_request *req, int offset)
809 struct obd_device *obd = req->rq_export->exp_obd;
810 struct mds_body *request_body, *reply_body;
811 struct lvfs_run_ctxt saved;
812 int rc, size = sizeof(*reply_body);
815 request_body = lustre_msg_buf(req->rq_reqmsg, offset,
816 sizeof(*request_body));
818 rc = lustre_pack_reply(req, 1, &size, NULL);
821 reply_body = lustre_msg_buf(req->rq_repmsg, 0, sizeof(*reply_body));
823 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
824 rc = mds_open_by_fid(req, &request_body->fid1, reply_body,
825 request_body->flags, NULL, NULL);
826 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
831 /* Get an internal lock on the inode number (but not generation) to sync
832 * new inode creation with inode unlink (bug 2029). If child_lockh is NULL
833 * we just get the lock as a barrier to wait for other holders of this lock,
834 * and drop it right away again. */
835 int mds_lock_new_child(struct obd_device *obd, struct inode *inode,
836 struct lustre_handle *child_lockh)
838 struct ldlm_res_id child_res_id = { .name = { inode->i_ino, 0, 1, 0 } };
839 struct lustre_handle lockh;
843 if (child_lockh == NULL)
844 child_lockh = &lockh;
846 rc = ldlm_cli_enqueue(NULL, NULL, obd->obd_namespace, child_res_id,
847 LDLM_PLAIN, NULL, LCK_EX, &lock_flags,
848 ldlm_blocking_ast, ldlm_completion_ast,
849 NULL, NULL, NULL, 0, NULL, child_lockh);
851 CERROR("ldlm_cli_enqueue: %d\n", rc);
852 else if (child_lockh == &lockh)
853 ldlm_lock_decref(child_lockh, LCK_EX);
858 int mds_open(struct mds_update_record *rec, int offset,
859 struct ptlrpc_request *req, struct lustre_handle *child_lockh)
861 /* XXX ALLOCATE _something_ - 464 bytes on stack here */
862 struct obd_device *obd = req->rq_export->exp_obd;
863 struct mds_obd *mds = mds_req2mds(req);
864 struct ldlm_reply *rep = NULL;
865 struct mds_body *body = NULL;
866 struct dentry *dchild = NULL, *dparent = NULL;
867 struct mds_export_data *med;
868 struct lustre_handle parent_lockh;
869 int rc = 0, cleanup_phase = 0, acc_mode, created = 0;
870 int parent_mode = LCK_CR;
872 struct dentry_params dp;
873 unsigned int qcids[MAXQUOTAS] = {current->fsuid, current->fsgid};
874 unsigned int qpids[MAXQUOTAS] = {0, 0};
877 CLASSERT(MAXQUOTAS < 4);
878 if (offset == 2) { /* intent */
879 rep = lustre_msg_buf(req->rq_repmsg, 0, sizeof (*rep));
880 body = lustre_msg_buf(req->rq_repmsg, 1, sizeof (*body));
881 } else if (offset == MDS_REQ_REC_OFF) { /* non-intent reint */
882 body = lustre_msg_buf(req->rq_repmsg, 0, sizeof (*body));
888 MDS_CHECK_RESENT(req, reconstruct_open(rec, offset, req, child_lockh));
890 /* Step 0: If we are passed a fid, then we assume the client already
891 * opened this file and is only replaying the RPC, so we open the
892 * inode by fid (at some large expense in security). */
893 if (lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY &&
894 !(rec->ur_flags & MDS_OPEN_JOIN_FILE)) {
895 if (rec->ur_fid2->id == 0) {
896 struct ldlm_lock *lock = ldlm_handle2lock(child_lockh);
898 LDLM_ERROR(lock, "fid2 not set on open replay");
901 DEBUG_REQ(D_ERROR, req, "fid2 not set on open replay");
905 rc = mds_open_by_fid(req, rec->ur_fid2, body, rec->ur_flags,
910 /* We didn't find the correct inode on disk either, so we
911 * need to re-create it via a regular replay. */
912 if (!(rec->ur_flags & MDS_OPEN_CREAT)) {
913 DEBUG_REQ(D_ERROR, req,"OPEN_CREAT not in open replay");
916 } else if (rec->ur_fid2->id) {
917 DEBUG_REQ(D_ERROR, req, "fid2 "LPU64"/%u on open non-replay",
918 rec->ur_fid2->id, rec->ur_fid2->generation);
922 LASSERT(offset == 2); /* If we got here, we must be called via intent */
924 med = &req->rq_export->exp_mds_data;
925 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_OPEN_PACK)) {
926 CERROR("test case OBD_FAIL_MDS_OPEN_PACK\n");
930 /* Step 1: Find and lock the parent */
931 if (rec->ur_flags & (MDS_OPEN_CREAT | MDS_OPEN_JOIN_FILE))
932 parent_mode = LCK_EX;
933 dparent = mds_fid2locked_dentry(obd, rec->ur_fid1, NULL, parent_mode,
934 &parent_lockh, MDS_INODELOCK_UPDATE);
935 if (IS_ERR(dparent)) {
936 rc = PTR_ERR(dparent);
938 CERROR("parent "LPU64"/%u lookup error %d\n",
939 rec->ur_fid1->id, rec->ur_fid1->generation, rc);
941 /* Just cannot find parent - make it look like
942 * usual negative lookup to avoid extra MDS RPC */
943 intent_set_disposition(rep, DISP_LOOKUP_EXECD);
944 intent_set_disposition(rep, DISP_LOOKUP_NEG);
948 LASSERT(dparent->d_inode != NULL);
950 cleanup_phase = 1; /* parent dentry and lock */
952 if (rec->ur_flags & MDS_OPEN_JOIN_FILE) {
953 dchild = dget(dparent);
954 cleanup_phase = 2; /* child dentry */
955 acc_mode = accmode(dchild->d_inode, rec->ur_flags);
956 GOTO(found_child, rc);
959 /* Step 2: Lookup the child */
960 dchild = ll_lookup_one_len(rec->ur_name, dparent, rec->ur_namelen - 1);
961 if (IS_ERR(dchild)) {
962 rc = PTR_ERR(dchild);
963 dchild = NULL; /* don't confuse mds_finish_transno() below */
967 cleanup_phase = 2; /* child dentry */
969 intent_set_disposition(rep, DISP_LOOKUP_EXECD);
971 intent_set_disposition(rep, DISP_LOOKUP_POS);
973 intent_set_disposition(rep, DISP_LOOKUP_NEG);
975 /*Step 3: If the child was negative, and we're supposed to, create it.*/
976 if (dchild->d_inode == NULL) {
977 unsigned long ino = rec->ur_fid2->id;
981 if (!(rec->ur_flags & MDS_OPEN_CREAT)) {
982 /* It's negative and we weren't supposed to create it */
983 GOTO(cleanup, rc = -ENOENT);
986 if (req->rq_export->exp_connect_flags & OBD_CONNECT_RDONLY)
987 GOTO(cleanup, rc = -EROFS);
989 intent_set_disposition(rep, DISP_OPEN_CREATE);
990 handle = fsfilt_start(obd, dparent->d_inode, FSFILT_OP_CREATE,
992 if (IS_ERR(handle)) {
993 rc = PTR_ERR(handle);
997 dchild->d_fsdata = (void *) &dp;
1001 rc = ll_vfs_create(dparent->d_inode, dchild, rec->ur_mode,NULL);
1002 if (dchild->d_fsdata == (void *)(unsigned long)ino)
1003 dchild->d_fsdata = NULL;
1006 CDEBUG(D_INODE, "error during create: %d\n", rc);
1009 inode = dchild->d_inode;
1011 LASSERT(ino == inode->i_ino);
1012 /* Written as part of setattr */
1013 inode->i_generation = rec->ur_fid2->generation;
1014 CDEBUG(D_HA, "recreated ino %lu with gen %u\n",
1015 inode->i_ino, inode->i_generation);
1019 LTIME_S(iattr.ia_atime) = rec->ur_time;
1020 LTIME_S(iattr.ia_ctime) = rec->ur_time;
1021 LTIME_S(iattr.ia_mtime) = rec->ur_time;
1023 iattr.ia_uid = current->fsuid; /* set by push_ctxt already */
1024 if (dparent->d_inode->i_mode & S_ISGID)
1025 iattr.ia_gid = dparent->d_inode->i_gid;
1027 iattr.ia_gid = current->fsgid;
1029 iattr.ia_valid = ATTR_UID | ATTR_GID | ATTR_ATIME |
1030 ATTR_MTIME | ATTR_CTIME;
1032 rc = fsfilt_setattr(obd, dchild, handle, &iattr, 0);
1034 CERROR("error on child setattr: rc = %d\n", rc);
1036 iattr.ia_valid = ATTR_MTIME | ATTR_CTIME;
1038 rc = fsfilt_setattr(obd, dparent, handle, &iattr, 0);
1040 CERROR("error on parent setattr: rc = %d\n", rc);
1042 rc = fsfilt_commit(obd, dchild->d_inode, handle, 0);
1044 acc_mode = 0; /* Don't check for permissions */
1046 acc_mode = accmode(dchild->d_inode, rec->ur_flags);
1049 LASSERTF(!mds_inode_is_orphan(dchild->d_inode),
1050 "dchild %.*s (%p) inode %p/%lu/%u\n", dchild->d_name.len,
1051 dchild->d_name.name, dchild, dchild->d_inode,
1052 dchild->d_inode->i_ino, dchild->d_inode->i_generation);
1055 mds_pack_inode2fid(&body->fid1, dchild->d_inode);
1056 mds_pack_inode2body(body, dchild->d_inode);
1058 if (S_ISREG(dchild->d_inode->i_mode)) {
1059 /* Check permissions etc */
1060 rc = ll_permission(dchild->d_inode, acc_mode, NULL);
1064 if ((req->rq_export->exp_connect_flags & OBD_CONNECT_RDONLY) &&
1065 (acc_mode & MAY_WRITE))
1066 GOTO(cleanup, rc = -EROFS);
1068 /* An append-only file must be opened in append mode for
1070 if (IS_APPEND(dchild->d_inode) && (acc_mode & MAY_WRITE) != 0 &&
1071 ((rec->ur_flags & MDS_OPEN_APPEND) == 0 ||
1072 (rec->ur_flags & MDS_OPEN_TRUNC) != 0))
1073 GOTO(cleanup, rc = -EPERM);
1076 if (!created && (rec->ur_flags & MDS_OPEN_CREAT) &&
1077 (rec->ur_flags & MDS_OPEN_EXCL)) {
1078 /* File already exists, we didn't just create it, and we
1079 * were passed O_EXCL; err-or. */
1080 GOTO(cleanup, rc = -EEXIST); // returns a lock to the client
1083 /* if we are following a symlink, don't open */
1084 if (S_ISLNK(dchild->d_inode->i_mode))
1085 GOTO(cleanup_no_trans, rc = 0);
1087 if (S_ISDIR(dchild->d_inode->i_mode)) {
1088 if (rec->ur_flags & MDS_OPEN_CREAT ||
1089 rec->ur_flags & FMODE_WRITE) {
1090 /* we are trying to create or write a exist dir */
1091 GOTO(cleanup, rc = -EISDIR);
1093 if (ll_permission(dchild->d_inode, acc_mode, NULL)) {
1094 intent_set_disposition(rep, DISP_OPEN_OPEN);
1095 GOTO(cleanup, rc = -EACCES);
1097 } else if (rec->ur_flags & MDS_OPEN_DIRECTORY) {
1098 GOTO(cleanup, rc = -ENOTDIR);
1101 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_OPEN_CREATE)) {
1102 obd_fail_loc = OBD_FAIL_LDLM_REPLY | OBD_FAIL_ONCE;
1103 GOTO(cleanup, rc = -EAGAIN);
1106 /* Step 5: mds_open it */
1107 rc = mds_finish_open(req, dchild, body, rec->ur_flags, &handle, rec,
1108 rep, &parent_lockh);
1112 rc = mds_finish_transno(mds, dchild ? dchild->d_inode : NULL, handle,
1113 req, rc, rep ? rep->lock_policy_res1 : 0);
1116 switch (cleanup_phase) {
1118 if (rc && created) {
1119 int err = vfs_unlink(dparent->d_inode, dchild);
1121 CERROR("unlink(%.*s) in error path: %d\n",
1122 dchild->d_name.len, dchild->d_name.name,
1125 } else if (created) {
1126 mds_lock_new_child(obd, dchild->d_inode, NULL);
1127 /* save uid/gid for quota acquire/release */
1128 qpids[USRQUOTA] = dparent->d_inode->i_uid;
1129 qpids[GRPQUOTA] = dparent->d_inode->i_gid;
1133 if (dparent == NULL)
1138 ldlm_lock_decref(&parent_lockh, parent_mode);
1140 ptlrpc_save_lock (req, &parent_lockh, parent_mode);
1143 /* trigger dqacq on the owner of child and parent */
1144 lquota_adjust(quota_interface, obd, qcids, qpids, rc, FSFILT_OP_CREATE);
1148 /* Close a "file descriptor" and possibly unlink an orphan from the
1149 * PENDING directory. Caller must hold child->i_sem, this drops it.
1151 * If we are being called from mds_disconnect() because the client has
1152 * disappeared, then req == NULL and we do not update last_rcvd because
1153 * there is nothing that could be recovered by the client at this stage
1154 * (it will not even _have_ an entry in last_rcvd anymore).
1156 * Returns EAGAIN if the client needs to get more data and re-close. */
1157 int mds_mfd_close(struct ptlrpc_request *req, int offset,struct obd_device *obd,
1158 struct mds_file_data *mfd, int unlink_orphan)
1160 struct inode *inode = mfd->mfd_dentry->d_inode;
1161 char fidname[LU_FID_NAMELEN];
1162 int last_orphan, fidlen, rc = 0, cleanup_phase = 0;
1163 struct dentry *pending_child = NULL;
1164 struct mds_obd *mds = &obd->u.mds;
1165 struct inode *pending_dir = mds->mds_pending_dir->d_inode;
1166 void *handle = NULL;
1167 struct mds_body *request_body = NULL, *reply_body = NULL;
1168 struct dentry_params dp;
1169 struct iattr iattr = { 0 };
1172 if (req && req->rq_reqmsg != NULL)
1173 request_body = lustre_msg_buf(req->rq_reqmsg, offset,
1174 sizeof(*request_body));
1175 if (req && req->rq_repmsg != NULL)
1176 reply_body = lustre_msg_buf(req->rq_repmsg, 0,
1177 sizeof(*reply_body));
1179 fidlen = lu_fid2str(fidname, inode->i_ino, inode->i_generation);
1181 CDEBUG(D_INODE, "inode %p ino %s nlink %d orphan %d\n", inode, fidname,
1182 inode->i_nlink, mds_orphan_open_count(inode));
1184 last_orphan = mds_orphan_open_dec_test(inode) &&
1185 mds_inode_is_orphan(inode);
1186 MDS_UP_WRITE_ORPHAN_SEM(inode);
1188 /* this is half of the actual "close" */
1189 if (mfd->mfd_mode & FMODE_WRITE) {
1190 rc = mds_put_write_access(mds, inode, request_body,
1191 last_orphan && unlink_orphan);
1192 } else if (mfd->mfd_mode & FMODE_EXEC) {
1193 mds_allow_write_access(inode);
1196 if (last_orphan && unlink_orphan) {
1197 struct lov_mds_md *lmm = NULL;
1198 int stripe_count = 0;
1199 LASSERT(rc == 0); /* mds_put_write_access must have succeeded */
1201 CDEBUG(D_HA, "destroying orphan object %s\n", fidname);
1203 if ((S_ISREG(inode->i_mode) && inode->i_nlink != 1) ||
1204 (S_ISDIR(inode->i_mode) && inode->i_nlink != 2))
1205 CERROR("found \"orphan\" %s %s with link count %d\n",
1206 S_ISREG(inode->i_mode) ? "file" : "dir",
1207 fidname, inode->i_nlink);
1209 /* Sadly, there is no easy way to save pending_child from
1210 * mds_reint_unlink() into mfd, so we need to re-lookup,
1211 * but normally it will still be in the dcache. */
1212 down(&pending_dir->i_sem);
1213 cleanup_phase = 1; /* up(&pending_dir->i_sem) when finished */
1214 pending_child = lookup_one_len(fidname, mds->mds_pending_dir,
1216 if (IS_ERR(pending_child))
1217 GOTO(cleanup, rc = PTR_ERR(pending_child));
1218 LASSERT(pending_child->d_inode != NULL);
1220 cleanup_phase = 2; /* dput(pending_child) when finished */
1221 if (S_ISDIR(pending_child->d_inode->i_mode)) {
1222 rc = vfs_rmdir(pending_dir, pending_child);
1224 CERROR("error unlinking orphan dir %s: rc %d\n",
1229 if (req != NULL && req->rq_repmsg != NULL) {
1230 lmm = lustre_msg_buf(req->rq_repmsg, 1, 0);
1231 stripe_count = le32_to_cpu(lmm->lmm_stripe_count);
1234 handle = fsfilt_start_log(obd, pending_dir, FSFILT_OP_UNLINK,
1235 NULL, stripe_count);
1236 if (IS_ERR(handle)) {
1237 rc = PTR_ERR(handle);
1242 if (req != NULL && req->rq_repmsg != NULL &&
1243 (reply_body->valid & OBD_MD_FLEASIZE) &&
1244 mds_log_op_unlink(obd, pending_child->d_inode, lmm,
1245 req->rq_repmsg->buflens[1],
1246 lustre_msg_buf(req->rq_repmsg, 2, 0),
1247 req->rq_repmsg->buflens[2]) > 0) {
1248 reply_body->valid |= OBD_MD_FLCOOKIE;
1251 pending_child->d_fsdata = (void *) &dp;
1254 rc = vfs_unlink(pending_dir, pending_child);
1256 CERROR("error unlinking orphan %s: rc %d\n",fidname,rc);
1258 goto out; /* Don't bother updating attrs on unlinked inode */
1262 if (request_body != NULL && mfd->mfd_mode & FMODE_WRITE && rc == 0) {
1263 /* Update the on-disk attributes if this was the last write
1264 * close, and all information was provided (i.e., rc == 0)
1266 * XXX this should probably be abstracted with mds_reint_setattr
1269 if (request_body->valid & OBD_MD_FLMTIME &&
1270 LTIME_S(iattr.ia_mtime) > LTIME_S(inode->i_mtime)) {
1271 LTIME_S(iattr.ia_mtime) = request_body->mtime;
1272 iattr.ia_valid |= ATTR_MTIME;
1274 if (request_body->valid & OBD_MD_FLCTIME &&
1275 LTIME_S(iattr.ia_ctime) > LTIME_S(inode->i_ctime)) {
1276 LTIME_S(iattr.ia_ctime) = request_body->ctime;
1277 iattr.ia_valid |= ATTR_CTIME;
1280 /* XXX can't set block count with fsfilt_setattr (!) */
1281 if (request_body->valid & OBD_MD_FLSIZE) {
1282 iattr.ia_valid |= ATTR_SIZE;
1283 iattr.ia_size = request_body->size;
1285 /* iattr.ia_blocks = request_body->blocks */
1289 if (request_body != NULL && request_body->valid & OBD_MD_FLATIME) {
1290 /* Only start a transaction to write out only the atime if
1291 * it is more out-of-date than the specified limit. If we
1292 * are already going to write out the atime then do it anyway.
1294 LTIME_S(iattr.ia_atime) = request_body->atime;
1295 if ((LTIME_S(iattr.ia_atime) >
1296 LTIME_S(inode->i_atime) + mds->mds_atime_diff) ||
1297 (iattr.ia_valid != 0 &&
1298 LTIME_S(iattr.ia_atime) > LTIME_S(inode->i_atime)))
1299 iattr.ia_valid |= ATTR_ATIME;
1302 if (iattr.ia_valid != 0) {
1303 handle = fsfilt_start(obd, inode, FSFILT_OP_SETATTR, NULL);
1305 GOTO(cleanup, rc = PTR_ERR(handle));
1306 rc = fsfilt_setattr(obd, mfd->mfd_dentry, handle, &iattr, 0);
1308 CERROR("error in setattr(%s): rc %d\n", fidname, rc);
1311 /* If other clients have this file open for write, rc will be > 0 */
1314 l_dput(mfd->mfd_dentry);
1318 if (req != NULL && reply_body != NULL) {
1319 rc = mds_finish_transno(mds, pending_dir, handle, req, rc, 0);
1320 } else if (handle) {
1321 int err = fsfilt_commit(obd, pending_dir, handle, 0);
1323 CERROR("error committing close: %d\n", err);
1329 switch (cleanup_phase) {
1331 dput(pending_child);
1333 up(&pending_dir->i_sem);
1338 int mds_close(struct ptlrpc_request *req, int offset)
1340 struct mds_export_data *med = &req->rq_export->exp_mds_data;
1341 struct obd_device *obd = req->rq_export->exp_obd;
1342 struct mds_body *body;
1343 struct mds_file_data *mfd;
1344 struct lvfs_run_ctxt saved;
1345 struct inode *inode;
1346 int rc, repsize[3] = {sizeof(struct mds_body),
1347 obd->u.mds.mds_max_mdsize,
1348 obd->u.mds.mds_max_cookiesize};
1351 rc = lustre_pack_reply(req, 3, repsize, NULL);
1353 CERROR("lustre_pack_reply: rc = %d\n", rc);
1354 req->rq_status = rc;
1355 /* continue on to drop local open even if we can't send reply */
1357 MDS_CHECK_RESENT(req, mds_reconstruct_generic(req));
1360 CDEBUG(D_HA, "close req->rep_len %d mdsize %d cookiesize %d\n",
1362 obd->u.mds.mds_max_mdsize, obd->u.mds.mds_max_cookiesize);
1364 body = lustre_swab_reqbuf(req, offset, sizeof(*body),
1365 lustre_swab_mds_body);
1367 CERROR("Can't unpack body\n");
1368 req->rq_status = -EFAULT;
1372 if (body->flags & MDS_BFLAG_UNCOMMITTED_WRITES)
1373 /* do some stuff */ ;
1375 spin_lock(&med->med_open_lock);
1376 mfd = mds_handle2mfd(&body->handle);
1378 spin_unlock(&med->med_open_lock);
1379 DEBUG_REQ(D_ERROR, req, "no handle for file close ino "LPD64
1380 ": cookie "LPX64, body->fid1.id, body->handle.cookie);
1381 req->rq_status = -ESTALE;
1384 /* Remove mfd handle so it can't be found again. We consume mfd_list
1385 * reference here, but still have mds_handle2mfd ref until mfd_close. */
1386 mds_mfd_unlink(mfd, 1);
1387 spin_unlock(&med->med_open_lock);
1389 inode = mfd->mfd_dentry->d_inode;
1390 /* child orphan sem protects orphan_dec_test && is_orphan race */
1391 MDS_DOWN_WRITE_ORPHAN_SEM(inode); /* mds_mfd_close drops this */
1392 if (mds_inode_is_orphan(inode) && mds_orphan_open_count(inode) == 1) {
1393 body = lustre_msg_buf(req->rq_repmsg, 0, sizeof (*body));
1394 LASSERT(body != NULL);
1396 mds_pack_inode2fid(&body->fid1, inode);
1397 mds_pack_inode2body(body, inode);
1398 mds_pack_md(obd, req->rq_repmsg, 1,body,inode,MDS_PACK_MD_LOCK);
1401 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
1402 req->rq_status = mds_mfd_close(req, offset, obd, mfd, 1);
1403 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
1405 mds_shrink_reply(obd, req, body);
1406 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_CLOSE_PACK)) {
1407 CERROR("test case OBD_FAIL_MDS_CLOSE_PACK\n");
1408 req->rq_status = -ENOMEM;
1415 int mds_done_writing(struct ptlrpc_request *req, int offset)
1417 struct mds_body *body;
1418 int rc, size = sizeof(struct mds_body);
1421 MDS_CHECK_RESENT(req, mds_reconstruct_generic(req));
1423 body = lustre_swab_reqbuf(req, offset, sizeof(*body),
1424 lustre_swab_mds_body);
1426 CERROR("Can't unpack body\n");
1427 req->rq_status = -EFAULT;
1431 rc = lustre_pack_reply(req, 1, &size, NULL);
1433 CERROR("lustre_pack_reply: rc = %d\n", rc);
1434 req->rq_status = rc;