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/obd_lov.h>
44 #include <linux/lustre_fsfilt.h>
45 #include <linux/lprocfs_status.h>
47 #include "mds_internal.h"
49 /* Exported function from this file are:
51 * mds_open - called by the intent handler
52 * mds_close - an rpc handling function
53 * mds_pin - an rpc handling function - which will go away
54 * mds_mfd_close - for force closing files when a client dies
58 * MDS file data handling: file data holds a handle for a file opened
62 static void mds_mfd_addref(void *mfdp)
64 struct mds_file_data *mfd = mfdp;
66 atomic_inc(&mfd->mfd_refcount);
67 CDEBUG(D_INFO, "GETting mfd %p : new refcount %d\n", mfd,
68 atomic_read(&mfd->mfd_refcount));
71 struct mds_file_data *mds_mfd_new(void)
73 struct mds_file_data *mfd;
75 OBD_ALLOC(mfd, sizeof *mfd);
77 CERROR("mds: out of memory\n");
81 atomic_set(&mfd->mfd_refcount, 2);
83 INIT_LIST_HEAD(&mfd->mfd_handle.h_link);
84 class_handle_hash(&mfd->mfd_handle, mds_mfd_addref);
89 static struct mds_file_data *mds_handle2mfd(struct lustre_handle *handle)
92 LASSERT(handle != NULL);
93 RETURN(class_handle2object(handle->cookie));
96 static void mds_mfd_put(struct mds_file_data *mfd)
98 CDEBUG(D_INFO, "PUTting mfd %p : new refcount %d\n", mfd,
99 atomic_read(&mfd->mfd_refcount) - 1);
100 LASSERT(atomic_read(&mfd->mfd_refcount) > 0 &&
101 atomic_read(&mfd->mfd_refcount) < 0x5a5a);
102 if (atomic_dec_and_test(&mfd->mfd_refcount)) {
103 LASSERT(list_empty(&mfd->mfd_handle.h_link));
104 OBD_FREE(mfd, sizeof *mfd);
108 static void mds_mfd_destroy(struct mds_file_data *mfd)
110 class_handle_unhash(&mfd->mfd_handle);
115 /* Caller must hold mds->mds_epoch_sem */
116 static int mds_alloc_filterdata(struct inode *inode)
118 LASSERT(inode->i_filterdata == NULL);
119 OBD_ALLOC(inode->i_filterdata, sizeof(struct mds_filter_data));
120 if (inode->i_filterdata == NULL)
122 LASSERT(igrab(inode) == inode);
126 /* Caller must hold mds->mds_epoch_sem */
127 static void mds_free_filterdata(struct inode *inode)
129 LASSERT(inode->i_filterdata != NULL);
130 OBD_FREE(inode->i_filterdata, sizeof(struct mds_filter_data));
131 inode->i_filterdata = NULL;
135 /* Write access to a file: executors cause a negative count,
136 * writers a positive count. The semaphore is needed to perform
137 * a check for the sign and then increment or decrement atomically.
139 * This code is closely tied to the allocation of the d_fsdata and the
140 * MDS epoch, so we use the same semaphore for the whole lot.
142 * We could use a different semaphore for each file, if it ever shows
143 * up in a profile, which it won't.
145 * epoch argument is nonzero during recovery */
146 static int mds_get_write_access(struct mds_obd *mds, struct inode *inode,
151 down(&mds->mds_epoch_sem);
153 if (atomic_read(&inode->i_writecount) < 0) {
154 up(&mds->mds_epoch_sem);
159 if (MDS_FILTERDATA(inode) && MDS_FILTERDATA(inode)->io_epoch != 0) {
160 CDEBUG(D_INODE, "continuing MDS epoch "LPU64" for ino %lu/%u\n",
161 MDS_FILTERDATA(inode)->io_epoch, inode->i_ino,
162 inode->i_generation);
166 if (inode->i_filterdata == NULL)
167 mds_alloc_filterdata(inode);
168 if (inode->i_filterdata == NULL) {
172 if (epoch > mds->mds_io_epoch)
173 mds->mds_io_epoch = epoch;
176 MDS_FILTERDATA(inode)->io_epoch = mds->mds_io_epoch;
177 CDEBUG(D_INODE, "starting MDS epoch "LPU64" for ino %lu/%u\n",
178 mds->mds_io_epoch, inode->i_ino, inode->i_generation);
181 atomic_inc(&inode->i_writecount);
182 up(&mds->mds_epoch_sem);
186 /* Returns EAGAIN if the client needs to get size and/or cookies and close
187 * again -- which is never true if the file is about to be unlinked. Otherwise
188 * returns the number of remaining writers. */
189 static int mds_put_write_access(struct mds_obd *mds, struct inode *inode,
190 struct mds_body *body, int unlinking)
195 down(&mds->mds_epoch_sem);
196 atomic_dec(&inode->i_writecount);
197 rc = atomic_read(&inode->i_writecount);
201 if (!unlinking && !(body->valid & OBD_MD_FLSIZE))
202 GOTO(out, rc = EAGAIN);
204 mds_free_filterdata(inode);
206 up(&mds->mds_epoch_sem);
210 static int mds_deny_write_access(struct mds_obd *mds, struct inode *inode)
213 down(&mds->mds_epoch_sem);
214 if (atomic_read(&inode->i_writecount) > 0) {
215 up(&mds->mds_epoch_sem);
218 atomic_dec(&inode->i_writecount);
219 up(&mds->mds_epoch_sem);
223 static void mds_allow_write_access(struct inode *inode)
226 atomic_inc(&inode->i_writecount);
229 int mds_query_write_access(struct inode *inode)
232 RETURN(atomic_read(&inode->i_writecount));
235 /* This replaces the VFS dentry_open, it manages mfd and writecount */
236 static struct mds_file_data *mds_dentry_open(struct dentry *dentry,
237 struct vfsmount *mnt, int flags,
238 struct ptlrpc_request *req)
240 struct mds_export_data *med = &req->rq_export->exp_mds_data;
241 struct mds_obd *mds = mds_req2mds(req);
242 struct mds_file_data *mfd;
243 struct mds_body *body;
249 CERROR("mds: out of memory\n");
250 GOTO(cleanup_dentry, error = -ENOMEM);
253 body = lustre_msg_buf(req->rq_repmsg, 1, sizeof (*body));
255 if (flags & FMODE_WRITE) {
256 /* FIXME: in recovery, need to pass old epoch here */
257 error = mds_get_write_access(mds, dentry->d_inode, 0);
259 GOTO(cleanup_mfd, error);
260 body->io_epoch = MDS_FILTERDATA(dentry->d_inode)->io_epoch;
261 } else if (flags & FMODE_EXEC) {
262 error = mds_deny_write_access(mds, dentry->d_inode);
264 GOTO(cleanup_mfd, error);
269 /* Mark the file as open to handle open-unlink. */
271 DOWN_WRITE_I_ALLOC_SEM(dentry->d_inode);
272 mds_orphan_open_inc(dentry->d_inode);
273 UP_WRITE_I_ALLOC_SEM(dentry->d_inode);
275 mfd->mfd_mode = flags;
276 mfd->mfd_dentry = dentry;
277 mfd->mfd_xid = req->rq_xid;
279 spin_lock(&med->med_open_lock);
280 list_add(&mfd->mfd_list, &med->med_open_head);
281 spin_unlock(&med->med_open_lock);
284 body->handle.cookie = mfd->mfd_handle.h_cookie;
290 mds_mfd_destroy(mfd);
292 return ERR_PTR(error);
295 static void mds_objids_from_lmm(obd_id *ids, struct lov_mds_md *lmm,
296 struct lov_desc *desc)
299 for (i = 0; i < le32_to_cpu(lmm->lmm_stripe_count); i++) {
300 ids[le32_to_cpu(lmm->lmm_objects[i].l_ost_idx)] =
301 le64_to_cpu(lmm->lmm_objects[i].l_object_id);
305 /* Must be called with i_sem held */
306 static int mds_create_objects(struct ptlrpc_request *req, int offset,
307 struct mds_update_record *rec,
308 struct mds_obd *mds, struct obd_device *obd,
309 struct dentry *dchild, void **handle,
312 struct obdo *oa = NULL;
313 struct obd_trans_info oti = { 0 };
314 struct mds_body *body;
315 struct lov_stripe_md *lsm = NULL;
316 struct lov_mds_md *lmm = NULL;
317 struct inode *inode = dchild->d_inode;
319 int rc, lmm_bufsize, lmm_size;
322 if (rec->ur_flags & MDS_OPEN_DELAY_CREATE ||
323 !(rec->ur_flags & FMODE_WRITE))
326 body = lustre_msg_buf(req->rq_repmsg, 1, sizeof(*body));
328 if (!S_ISREG(inode->i_mode))
330 if (body->valid & OBD_MD_FLEASIZE)
333 OBD_ALLOC(*ids, mds->mds_lov_desc.ld_tgt_count * sizeof(**ids));
336 oti.oti_objid = *ids;
339 if (lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY) {
340 LASSERT (rec->ur_fid2->id);
341 body->valid |= OBD_MD_FLBLKSZ | OBD_MD_FLEASIZE;
342 lmm_size = rec->ur_eadatalen;
343 lmm = rec->ur_eadata;
347 *handle = fsfilt_start(obd,inode,FSFILT_OP_CREATE,NULL);
348 if (IS_ERR(*handle)) {
349 rc = PTR_ERR(*handle);
354 mds_objids_from_lmm(*ids, lmm, &mds->mds_lov_desc);
356 lmm_buf = lustre_msg_buf(req->rq_repmsg, offset, 0);
357 lmm_bufsize = req->rq_repmsg->buflens[offset];
359 LASSERT(lmm_bufsize >= lmm_size);
360 memcpy(lmm_buf, lmm, lmm_size);
361 rc = fsfilt_set_md(obd, inode, *handle, lmm, lmm_size);
363 CERROR("open replay failed to set md:%d\n", rc);
367 if (OBD_FAIL_CHECK_ONCE(OBD_FAIL_MDS_ALLOC_OBDO))
368 GOTO(out_ids, rc = -ENOMEM);
372 GOTO(out_ids, rc = -ENOMEM);
373 oa->o_mode = S_IFREG | 0600;
374 oa->o_id = inode->i_ino;
375 oa->o_gr = FILTER_GROUP_FIRST_MDS + mds->mds_num;
376 oa->o_generation = inode->i_generation;
377 oa->o_uid = 0; /* must have 0 uid / gid on OST */
379 oa->o_valid = OBD_MD_FLID | OBD_MD_FLGENER | OBD_MD_FLTYPE |
380 OBD_MD_FLMODE | OBD_MD_FLUID | OBD_MD_FLGID | OBD_MD_FLGROUP;
383 obdo_from_inode(oa, inode, OBD_MD_FLTYPE|OBD_MD_FLATIME|OBD_MD_FLMTIME|
386 if (!(rec->ur_flags & MDS_OPEN_HAS_OBJS)) {
387 /* check if things like lfs setstripe are sending us the ea */
388 if (rec->ur_flags & MDS_OPEN_HAS_EA) {
389 rc = obd_iocontrol(OBD_IOC_LOV_SETSTRIPE,
391 0, &lsm, rec->ur_eadata);
395 /* Per-directory striping default code removed, because
396 * it uses the same unnamed EA storage as the directory
397 * striping for CMD. -p */
399 LASSERT(oa->o_gr >= FILTER_GROUP_FIRST_MDS);
400 rc = obd_create(mds->mds_osc_exp, oa, &lsm, &oti);
405 CDEBUG(level, "error creating objects for "
406 "inode %lu: rc = %d\n",
409 CERROR("obd_create returned invalid "
416 rc = obd_iocontrol(OBD_IOC_LOV_SETEA, mds->mds_osc_exp,
417 0, &lsm, rec->ur_eadata);
421 lsm->lsm_object_id = oa->o_id;
422 lsm->lsm_object_gr = oa->o_gr;
425 oa->o_size = inode->i_size;
426 obdo_from_inode(oa, inode, OBD_MD_FLTYPE|OBD_MD_FLATIME|
427 OBD_MD_FLMTIME| OBD_MD_FLCTIME| OBD_MD_FLSIZE);
428 rc = obd_setattr(mds->mds_osc_exp, oa, lsm, &oti);
430 CERROR("error setting attrs for inode %lu: rc %d\n",
433 CERROR("obd_setattr returned bad rc %d\n", rc);
440 body->valid |= OBD_MD_FLBLKSZ | OBD_MD_FLEASIZE;
441 obdo_refresh_inode(inode, oa, OBD_MD_FLBLKSZ);
443 LASSERT(lsm && lsm->lsm_object_id);
445 rc = obd_packmd(mds->mds_osc_exp, &lmm, lsm);
446 if (!rec->ur_fid2->id)
447 obd_free_memmd(mds->mds_osc_exp, &lsm);
450 body->eadatasize = rc;
453 *handle = fsfilt_start(obd, inode, FSFILT_OP_CREATE, NULL);
454 if (IS_ERR(*handle)) {
455 rc = PTR_ERR(*handle);
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, *dchild;
489 struct ldlm_reply *rep;
490 struct mds_body *body;
496 LASSERT(offset == 3); /* 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 LASSERTF(!IS_ERR(parent), "fid "LPU64"/%u rc %ld\n", rec->ur_fid1->id,
512 rec->ur_fid1->generation, PTR_ERR(parent));
514 dchild = ll_lookup_one_len(rec->ur_name, parent, rec->ur_namelen - 1);
515 LASSERTF(!IS_ERR(dchild), "parent "LPU64"/%u child %s rc %ld\n",
516 rec->ur_fid1->id, rec->ur_fid1->generation, rec->ur_name,
519 if (!dchild->d_inode)
520 GOTO(out_dput, 0); /* child not present to open */
522 /* At this point, we know we have a child. We'll send
523 * it back _unless_ it not created and open failed. */
524 if (intent_disposition(rep, DISP_OPEN_OPEN) &&
525 !intent_disposition(rep, DISP_OPEN_CREATE) &&
529 mds_pack_inode2fid(obd, &body->fid1, dchild->d_inode);
530 mds_pack_inode2body(obd, body, dchild->d_inode);
531 if (S_ISREG(dchild->d_inode->i_mode)) {
532 rc = mds_pack_md(obd, req->rq_repmsg, 2, body,
536 LASSERT(rc == req->rq_status);
538 /* If we have LOV EA data, the OST holds size, mtime */
539 if (!(body->valid & OBD_MD_FLEASIZE))
540 body->valid |= (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS |
541 OBD_MD_FLATIME | OBD_MD_FLMTIME);
544 /* If we have -EEXIST as the status, and we were asked to create
545 * exclusively, we can tell we failed because the file already existed.
547 if (req->rq_status == -EEXIST &&
548 ((rec->ur_flags & (MDS_OPEN_CREAT | MDS_OPEN_EXCL)) ==
549 (MDS_OPEN_CREAT | MDS_OPEN_EXCL))) {
553 /* If we didn't get as far as trying to open, then some locking thing
554 * probably went wrong, and we'll just bail here.
556 if (!intent_disposition(rep, DISP_OPEN_OPEN))
559 /* If we failed, then we must have failed opening, so don't look for
560 * file descriptor or anything, just give the client the bad news.
566 list_for_each(t, &med->med_open_head) {
567 mfd = list_entry(t, struct mds_file_data, mfd_list);
568 if (mfd->mfd_xid == req->rq_xid)
573 /* #warning "XXX fixme" bug 2991 */
574 /* Here it used to LASSERT(mfd) if exp_outstanding_reply != NULL.
575 * Now that exp_outstanding_reply is a list, it's just using mfd != NULL
576 * to detect a re-open */
578 mntget(mds->mds_vfsmnt);
579 CERROR("Re-opened file \n");
580 mfd = mds_dentry_open(dchild, mds->mds_vfsmnt,
581 rec->ur_flags & ~MDS_OPEN_TRUNC, req);
583 CERROR("mds: out of memory\n");
584 GOTO(out_dput, req->rq_status = -ENOMEM);
588 body->handle.cookie = mfd->mfd_handle.h_cookie;
589 CDEBUG(D_INODE, "resend mfd %p, cookie "LPX64"\n", mfd,
590 mfd->mfd_handle.h_cookie);
600 /* do NOT or the MAY_*'s, you'll get the weakest */
601 static int accmode(int flags)
605 if (flags & FMODE_READ)
607 if (flags & (FMODE_WRITE|MDS_OPEN_TRUNC))
609 if (flags & FMODE_EXEC)
614 /* Handles object creation, actual opening, and I/O epoch */
615 static int mds_finish_open(struct ptlrpc_request *req, struct dentry *dchild,
616 struct mds_body *body, int flags, void **handle,
617 struct mds_update_record *rec,struct ldlm_reply *rep)
619 struct mds_obd *mds = mds_req2mds(req);
620 struct obd_device *obd = req->rq_export->exp_obd;
621 struct mds_file_data *mfd = NULL;
622 obd_id *ids = NULL; /* object IDs created */
627 /* atomically create objects if necessary */
628 down(&dchild->d_inode->i_sem);
629 mode = dchild->d_inode->i_mode;
630 if ((S_ISREG(mode) && !(body->valid & OBD_MD_FLEASIZE)) ||
631 (S_ISDIR(mode) && !(body->valid & OBD_MD_FLDIREA))) {
632 rc = mds_pack_md(obd, req->rq_repmsg, 2, body,
635 up(&dchild->d_inode->i_sem);
640 /* no EA: create objects */
641 if ((body->valid & OBD_MD_FLEASIZE) &&
642 (rec->ur_flags & MDS_OPEN_HAS_EA)) {
643 up(&dchild->d_inode->i_sem);
646 if (!(body->valid & OBD_MD_FLEASIZE)) {
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 (S_ISREG(dchild->d_inode->i_mode) &&
657 (body->valid & OBD_MD_FLEASIZE)) {
658 rc = mds_revalidate_lov_ea(obd, dchild->d_inode,
661 rc = mds_pack_md(obd, req->rq_repmsg, 2, body,
664 up(&dchild->d_inode->i_sem);
669 /* If the inode has no EA data, then MDSs hold size, mtime */
670 if (S_ISREG(dchild->d_inode->i_mode) &&
671 !(body->valid & OBD_MD_FLEASIZE)) {
672 body->valid |= (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS |
673 OBD_MD_FLATIME | OBD_MD_FLMTIME);
675 up(&dchild->d_inode->i_sem);
677 intent_set_disposition(rep, DISP_OPEN_OPEN);
678 mfd = mds_dentry_open(dchild, mds->mds_vfsmnt, flags, req);
680 RETURN(PTR_ERR(mfd));
682 CDEBUG(D_INODE, "mfd %p, cookie "LPX64"\n", mfd,
683 mfd->mfd_handle.h_cookie);
685 mds_lov_update_objids(obd, ids);
686 OBD_FREE(ids, sizeof(*ids) * mds->mds_lov_desc.ld_tgt_count);
689 // mds_mfd_destroy(mfd);
693 static int mds_open_by_fid(struct ptlrpc_request *req, struct ll_fid *fid,
694 struct mds_body *body, int flags,
695 struct mds_update_record *rec,struct ldlm_reply *rep)
697 struct mds_obd *mds = mds_req2mds(req);
698 struct inode *pending_dir = mds->mds_pending_dir->d_inode;
699 struct dentry *dchild;
700 char fidname[LL_FID_NAMELEN];
705 down(&pending_dir->i_sem);
706 fidlen = ll_fid2str(fidname, fid->id, fid->generation);
707 dchild = lookup_one_len(fidname, mds->mds_pending_dir, fidlen);
708 if (IS_ERR(dchild)) {
709 up(&pending_dir->i_sem);
710 rc = PTR_ERR(dchild);
711 CERROR("error looking up %s in PENDING: rc = %d\n",fidname, rc);
715 if (dchild->d_inode != NULL) {
716 up(&pending_dir->i_sem);
717 mds_inode_set_orphan(dchild->d_inode);
718 mds_pack_inode2fid(req2obd(req), &body->fid1, dchild->d_inode);
719 mds_pack_inode2body(req2obd(req), body, dchild->d_inode);
720 intent_set_disposition(rep, DISP_LOOKUP_EXECD);
721 intent_set_disposition(rep, DISP_LOOKUP_POS);
722 CWARN("Orphan %s found and opened in PENDING directory\n",
727 up(&pending_dir->i_sem);
729 /* We didn't find it in PENDING so it isn't an orphan. See
730 * if it was a regular inode that was previously created. */
731 dchild = mds_fid2dentry(mds, fid, NULL);
733 RETURN(PTR_ERR(dchild));
735 mds_pack_inode2fid(req2obd(req), &body->fid1, dchild->d_inode);
736 mds_pack_inode2body(req2obd(req), body, dchild->d_inode);
737 intent_set_disposition(rep, DISP_LOOKUP_EXECD);
738 intent_set_disposition(rep, DISP_LOOKUP_POS);
741 rc = mds_finish_open(req, dchild, body, flags, &handle, rec, rep);
742 rc = mds_finish_transno(mds, dchild ? dchild->d_inode : NULL, handle,
743 req, rc, rep ? rep->lock_policy_res1 : 0);
744 /* XXX what do we do here if mds_finish_transno itself failed? */
750 int mds_pin(struct ptlrpc_request *req, int offset)
752 struct obd_device *obd = req->rq_export->exp_obd;
753 struct mds_body *request_body, *reply_body;
754 struct lvfs_run_ctxt saved;
755 int rc, size = sizeof(*reply_body);
758 request_body = lustre_msg_buf(req->rq_reqmsg, offset,
759 sizeof(*request_body));
761 rc = lustre_pack_reply(req, 1, &size, NULL);
764 reply_body = lustre_msg_buf(req->rq_repmsg, 0, sizeof(*reply_body));
766 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
767 rc = mds_open_by_fid(req, &request_body->fid1, reply_body,
768 request_body->flags, NULL, NULL);
769 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
774 /* Get a lock on the ino to sync with creation WRT inode reuse (bug 2029).
775 * If child_lockh is NULL we just get the lock as a barrier to wait for
776 * other holders of this lock, and drop it right away again. */
777 int mds_lock_new_child(struct obd_device *obd, struct inode *inode,
778 struct lustre_handle *child_lockh)
780 struct ldlm_res_id child_res_id = { .name = { inode->i_ino, 0, 1, 0 } };
781 struct lustre_handle lockh;
785 if (child_lockh == NULL)
786 child_lockh = &lockh;
788 rc = ldlm_cli_enqueue(NULL, NULL, obd->obd_namespace, child_res_id,
789 LDLM_PLAIN, NULL, LCK_EX, &lock_flags,
790 mds_blocking_ast, ldlm_completion_ast, NULL, NULL,
791 NULL, 0, NULL, child_lockh);
793 CERROR("ldlm_cli_enqueue: %d\n", rc);
794 else if (child_lockh == &lockh)
795 ldlm_lock_decref(child_lockh, LCK_EX);
800 static int is_mount_object(struct dentry *dparent)
802 struct dentry *dchild;
804 if (!(dparent->d_inode->i_mode & S_ISUID))
807 dchild = lookup_one_len(".mntinfo", dparent, strlen(".mntinfo"));
808 if (IS_ERR(dchild) || dchild == NULL)
815 int mds_open(struct mds_update_record *rec, int offset,
816 struct ptlrpc_request *req, struct lustre_handle *child_lockh)
818 /* XXX ALLOCATE _something_ - 464 bytes on stack here */
819 struct obd_device *obd = req->rq_export->exp_obd;
820 struct mds_obd *mds = mds_req2mds(req);
821 struct ldlm_reply *rep = NULL;
822 struct mds_body *body = NULL;
823 struct dentry *dchild = NULL, *dparent = NULL;
824 struct mds_export_data *med;
825 struct lustre_handle parent_lockh[2];
826 int rc = 0, cleanup_phase = 0, acc_mode, created = 0;
827 int parent_mode = LCK_PR;
829 struct dentry_params dp;
830 struct mea *mea = NULL;
831 int mea_size, update_mode;
834 DEBUG_REQ(D_INODE, req, "parent "LPU64"/%u name %*s mode %o",
835 rec->ur_fid1->id, rec->ur_fid1->generation,
836 rec->ur_namelen - 1, rec->ur_name, rec->ur_mode);
838 parent_lockh[0].cookie = 0;
839 parent_lockh[1].cookie = 0;
841 if (offset == 3) { /* intent */
842 rep = lustre_msg_buf(req->rq_repmsg, 0, sizeof (*rep));
843 body = lustre_msg_buf(req->rq_repmsg, 1, sizeof (*body));
844 } else if (offset == 1) { /* non-intent reint */
845 body = lustre_msg_buf(req->rq_repmsg, 0, sizeof (*body));
851 MDS_CHECK_RESENT(req, reconstruct_open(rec, offset, req, child_lockh));
853 MDS_UPDATE_COUNTER(mds, MDS_OPEN_COUNT);
855 /* Step 0: If we are passed a fid, then we assume the client already
856 * opened this file and is only replaying the RPC, so we open the
857 * inode by fid (at some large expense in security). */
858 if (lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY) {
859 CDEBUG(D_HA, "open fid "LPU64"/%u name %*s mode %o\n",
860 rec->ur_fid2->id, rec->ur_fid2->generation,
861 rec->ur_namelen - 1, rec->ur_name, rec->ur_mode);
863 LASSERT(rec->ur_fid2->id);
865 rc = mds_open_by_fid(req, rec->ur_fid2, body, rec->ur_flags,
869 /* We didn't find the correct inode on disk either, so we
870 * need to re-create it via a regular replay. */
871 LASSERT(rec->ur_flags & MDS_OPEN_CREAT);
873 LASSERT(!rec->ur_fid2->id);
876 LASSERT(offset == 3); /* If we got here, we must be called via intent */
878 med = &req->rq_export->exp_mds_data;
879 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_OPEN_PACK)) {
880 CERROR("test case OBD_FAIL_MDS_OPEN_PACK\n");
884 acc_mode = accmode(rec->ur_flags);
886 /* Step 1: Find and lock the parent */
887 if (rec->ur_flags & O_CREAT) {
888 /* XXX Well, in fact we only need this lock mode change if
889 in addition to O_CREAT, the file does not exist.
890 But we do not know if it exists or not yet */
891 parent_mode = LCK_PW;
894 if (rec->ur_namelen == 1) {
895 /* client (LMV) wants to open the file by fid */
896 CDEBUG(D_OTHER, "OPEN by fid %u/%u/%u\n",
897 (unsigned) rec->ur_fid1->mds,
898 (unsigned) rec->ur_fid1->id,
899 (unsigned) rec->ur_fid1->generation);
900 dchild = mds_fid2dentry(mds, rec->ur_fid1, NULL);
901 if (IS_ERR(dchild)) {
902 rc = PTR_ERR(dparent);
903 CERROR("child lookup by a fid error %d\n", rc);
909 dparent = mds_fid2locked_dentry(obd, rec->ur_fid1, NULL, parent_mode,
910 parent_lockh, &update_mode, rec->ur_name,
912 MDS_INODELOCK_UPDATE);
913 if (IS_ERR(dparent)) {
914 rc = PTR_ERR(dparent);
916 CERROR("parent lookup error %d\n", rc);
919 LASSERT(dparent->d_inode != NULL);
921 cleanup_phase = 1; /* parent dentry and lock */
923 /* try to retrieve MEA data for this dir */
924 rc = mds_get_lmv_attr(obd, dparent->d_inode, &mea, &mea_size);
929 /* dir is already splitted, check is requested filename
930 * should live at this MDS or at another one */
932 i = mea_name2idx(mea, rec->ur_name, rec->ur_namelen - 1);
933 if (mea->mea_master != mea->mea_fids[i].mds) {
935 "%s: inapropriate MDS(%d) for %lu/%u:%s."
936 " should be %d(%d)\n", obd->obd_name,
937 mea->mea_master, dparent->d_inode->i_ino,
938 dparent->d_inode->i_generation, rec->ur_name,
939 mea->mea_fids[i].mds, i);
940 GOTO(cleanup, rc = -ERESTART);
944 /* Step 2: Lookup the child */
945 dchild = ll_lookup_one_len(rec->ur_name, dparent, rec->ur_namelen - 1);
946 if (IS_ERR(dchild)) {
947 rc = PTR_ERR(dchild);
948 dchild = NULL; /* don't confuse mds_finish_transno() below */
953 cleanup_phase = 2; /* child dentry */
955 intent_set_disposition(rep, DISP_LOOKUP_EXECD);
957 if (dchild->d_flags & DCACHE_CROSS_REF) {
958 struct ldlm_res_id res_id = { . name = {0} };
959 ldlm_policy_data_t policy;
961 CDEBUG(D_OTHER, "cross reference: %lu/%lu/%lu\n",
962 (unsigned long) dchild->d_mdsnum,
963 (unsigned long) dchild->d_inum,
964 (unsigned long) dchild->d_generation);
965 body->valid |= OBD_MD_FLID | OBD_MD_MDS;
966 body->fid1.id = dchild->d_inum;
967 body->fid1.mds = dchild->d_mdsnum;
968 body->fid1.generation = dchild->d_generation;
969 intent_set_disposition(rep, DISP_LOOKUP_POS);
970 res_id.name[0] = dchild->d_inum;
971 res_id.name[1] = dchild->d_generation;
972 policy.l_inodebits.bits = MDS_INODELOCK_LOOKUP;
973 rc = ldlm_cli_enqueue(NULL, NULL, obd->obd_namespace,
974 res_id, LDLM_IBITS, &policy,
977 ldlm_completion_ast, NULL, NULL,
978 NULL, 0, NULL, child_lockh);
980 if (parent_lockh[1].cookie != 0)
981 ldlm_lock_decref(parent_lockh + 1, update_mode);
983 ldlm_lock_decref(parent_lockh, parent_mode);
985 OBD_FREE(mea, mea_size);
992 intent_set_disposition(rep, DISP_LOOKUP_POS);
994 intent_set_disposition(rep, DISP_LOOKUP_NEG);
996 /*Step 3: If the child was negative, and we're supposed to, create it.*/
997 if (dchild->d_inode == NULL) {
998 unsigned long ino = rec->ur_fid2->id;
1000 struct inode *inode;
1001 rc = mds_try_to_split_dir(obd, dparent, &mea, 0, update_mode);
1002 CDEBUG(D_OTHER, "%s: splitted %lu/%u - %d\n",
1003 obd->obd_name, dparent->d_inode->i_ino,
1004 dparent->d_inode->i_generation, rc);
1006 /* dir got splitted */
1007 GOTO(cleanup, rc = -ERESTART);
1008 } else if (rc < 0) {
1009 /* error happened during spitting */
1013 if (!(rec->ur_flags & MDS_OPEN_CREAT)) {
1014 /* It's negative and we weren't supposed to create it */
1015 GOTO(cleanup, rc = -ENOENT);
1018 intent_set_disposition(rep, DISP_OPEN_CREATE);
1019 handle = fsfilt_start(obd, dparent->d_inode, FSFILT_OP_CREATE,
1021 if (IS_ERR(handle)) {
1022 rc = PTR_ERR(handle);
1026 dchild->d_fsdata = (void *) &dp;
1030 rc = ll_vfs_create(dparent->d_inode, dchild, rec->ur_mode,NULL);
1031 if (dchild->d_fsdata == (void *)(unsigned long)ino)
1032 dchild->d_fsdata = NULL;
1035 CDEBUG(D_INODE, "error during create: %d\n", rc);
1038 inode = dchild->d_inode;
1040 LASSERT(ino == inode->i_ino);
1041 /* Written as part of setattr */
1042 inode->i_generation = rec->ur_fid2->generation;
1043 CDEBUG(D_HA, "recreated ino %lu with gen %u\n",
1044 inode->i_ino, inode->i_generation);
1048 LTIME_S(iattr.ia_atime) = rec->ur_time;
1049 LTIME_S(iattr.ia_ctime) = rec->ur_time;
1050 LTIME_S(iattr.ia_mtime) = rec->ur_time;
1052 iattr.ia_uid = rec->_ur_fsuid;
1053 if (dparent->d_inode->i_mode & S_ISGID)
1054 iattr.ia_gid = dparent->d_inode->i_gid;
1056 iattr.ia_gid = rec->_ur_fsgid;
1058 iattr.ia_valid = ATTR_UID | ATTR_GID | ATTR_ATIME |
1059 ATTR_MTIME | ATTR_CTIME;
1061 rc = fsfilt_setattr(obd, dchild, handle, &iattr, 0);
1063 CERROR("error on child setattr: rc = %d\n", rc);
1065 iattr.ia_valid = ATTR_MTIME | ATTR_CTIME;
1067 rc = fsfilt_setattr(obd, dparent, handle, &iattr, 0);
1069 CERROR("error on parent setattr: rc = %d\n", rc);
1071 MDS_UPDATE_COUNTER(mds, MDS_CREATE_COUNT);
1073 if (!(rec->ur_flags & O_EXCL)) { /* bug 3313 */
1074 rc = fsfilt_commit(obd, dchild->d_inode->i_sb,
1075 dchild->d_inode, handle, 0);
1079 acc_mode = 0; /* Don't check for permissions */
1081 LASSERTF(!mds_inode_is_orphan(dchild->d_inode),
1082 "dchild %*s (%p) inode %p\n", dchild->d_name.len,
1083 dchild->d_name.name, dchild, dchild->d_inode);
1085 mds_pack_inode2fid(obd, &body->fid1, dchild->d_inode);
1086 mds_pack_inode2body(obd, body, dchild->d_inode);
1088 if (S_ISREG(dchild->d_inode->i_mode)) {
1089 /* Check permissions etc */
1090 rc = ll_permission(dchild->d_inode, acc_mode, NULL);
1094 /* Can't write to a read-only file */
1095 if (IS_RDONLY(dchild->d_inode) && (acc_mode & MAY_WRITE) != 0)
1096 GOTO(cleanup, rc = -EPERM);
1098 /* An append-only file must be opened in append mode for
1100 if (IS_APPEND(dchild->d_inode) && (acc_mode & MAY_WRITE) != 0 &&
1101 ((rec->ur_flags & MDS_OPEN_APPEND) == 0 ||
1102 (rec->ur_flags & MDS_OPEN_TRUNC) != 0))
1103 GOTO(cleanup, rc = -EPERM);
1106 if (!created && (rec->ur_flags & MDS_OPEN_CREAT) &&
1107 (rec->ur_flags & MDS_OPEN_EXCL)) {
1108 /* File already exists, we didn't just create it, and we
1109 * were passed O_EXCL; err-or. */
1110 GOTO(cleanup, rc = -EEXIST); // returns a lock to the client
1113 /* if we are following a symlink, don't open */
1114 if (S_ISLNK(dchild->d_inode->i_mode))
1115 GOTO(cleanup, rc = 0);
1117 if ((rec->ur_flags & MDS_OPEN_DIRECTORY) &&
1118 !S_ISDIR(dchild->d_inode->i_mode))
1119 GOTO(cleanup, rc = -ENOTDIR);
1121 if (S_ISDIR(dchild->d_inode->i_mode)) {
1122 if (rec->ur_flags & MDS_OPEN_CREAT ||
1123 rec->ur_flags & FMODE_WRITE) {
1124 /*we are trying to create or write a exist dir*/
1125 GOTO(cleanup, rc = -EISDIR);
1127 if (ll_permission(dchild->d_inode, acc_mode, NULL)) {
1128 GOTO(cleanup, rc = -EACCES);
1130 if (is_mount_object(dchild)) {
1131 CERROR("Found possible GNS mount object %*s; not "
1132 "opening.\n", dchild->d_name.len,
1133 dchild->d_name.name);
1134 GOTO(cleanup, rc = 0); // success, but don't really open
1138 /* Step 5: mds_open it */
1139 rc = mds_finish_open(req, dchild, body, rec->ur_flags, &handle, rec,
1144 rc = mds_finish_transno(mds, dchild ? dchild->d_inode : NULL, handle,
1145 req, rc, rep ? rep->lock_policy_res1 : 0);
1147 switch (cleanup_phase) {
1149 if (rc && created) {
1150 int err = vfs_unlink(dparent->d_inode, dchild);
1152 CERROR("unlink(%*s) in error path: %d\n",
1153 dchild->d_name.len, dchild->d_name.name,
1156 } else if (created) {
1158 mds_lock_new_child(obd, dchild->d_inode, NULL);
1163 if (dparent == NULL)
1168 if (parent_lockh[1].cookie != 0)
1169 ldlm_lock_decref(parent_lockh + 1, update_mode);
1172 ldlm_lock_decref(parent_lockh, parent_mode);
1174 ptlrpc_save_lock (req, parent_lockh, parent_mode);
1177 OBD_FREE(mea, mea_size);
1181 /* Close a "file descriptor" and possibly unlink an orphan from the
1182 * PENDING directory. Caller must hold child->i_sem, this drops it.
1184 * If we are being called from mds_disconnect() because the client has
1185 * disappeared, then req == NULL and we do not update last_rcvd because
1186 * there is nothing that could be recovered by the client at this stage
1187 * (it will not even _have_ an entry in last_rcvd anymore).
1189 * Returns EAGAIN if the client needs to get more data and re-close. */
1190 int mds_mfd_close(struct ptlrpc_request *req, int offset,
1191 struct obd_device *obd, struct mds_file_data *mfd,
1194 struct inode *inode = mfd->mfd_dentry->d_inode;
1195 char fidname[LL_FID_NAMELEN];
1196 int last_orphan, fidlen, rc = 0, cleanup_phase = 0;
1197 struct dentry *pending_child = NULL;
1198 struct mds_obd *mds = &obd->u.mds;
1199 struct inode *pending_dir = mds->mds_pending_dir->d_inode;
1200 void *handle = NULL;
1201 struct mds_body *request_body = NULL, *reply_body = NULL;
1202 struct dentry_params dp;
1203 struct iattr iattr = { 0 };
1204 struct llog_create_locks *lcl = NULL;
1207 if (req && req->rq_reqmsg != NULL)
1208 request_body = lustre_msg_buf(req->rq_reqmsg, offset,
1209 sizeof(*request_body));
1210 if (req && req->rq_repmsg != NULL)
1211 reply_body = lustre_msg_buf(req->rq_repmsg, 0,
1212 sizeof(*reply_body));
1214 fidlen = ll_fid2str(fidname, inode->i_ino, inode->i_generation);
1215 CDEBUG(D_INODE, "inode %p ino %s nlink %d orphan %d\n", inode, fidname,
1216 inode->i_nlink, mds_orphan_open_count(inode));
1217 last_orphan = mds_orphan_open_dec_test(inode) &&
1218 mds_inode_is_orphan(inode);
1219 UP_WRITE_I_ALLOC_SEM(inode);
1221 /* this is half of the actual "close" */
1222 if (mfd->mfd_mode & FMODE_WRITE) {
1223 rc = mds_put_write_access(mds, inode, request_body,
1224 last_orphan && unlink_orphan);
1225 } else if (mfd->mfd_mode & FMODE_EXEC) {
1226 mds_allow_write_access(inode);
1229 if (last_orphan && unlink_orphan) {
1230 struct lov_mds_md *lmm = NULL;
1231 int stripe_count = 0;
1232 LASSERT(rc == 0); /* mds_put_write_access must have succeeded */
1234 if (obd->obd_recovering) {
1235 CDEBUG(D_HA, "not remove orphan %s until recovery"
1236 " is over\n", fidname);
1240 CDEBUG(D_HA, "destroying orphan object %s\n", fidname);
1242 if ((S_ISREG(inode->i_mode) && inode->i_nlink != 1) ||
1243 (S_ISDIR(inode->i_mode) && inode->i_nlink != 2))
1244 CERROR("found \"orphan\" %s %s with link count %d\n",
1245 S_ISREG(inode->i_mode) ? "file" : "dir",
1246 fidname, inode->i_nlink);
1247 /* Sadly, there is no easy way to save pending_child from
1248 * mds_reint_unlink() into mfd, so we need to re-lookup,
1249 * but normally it will still be in the dcache. */
1250 down(&pending_dir->i_sem);
1251 cleanup_phase = 1; /* up(&pending_dir->i_sem) when finished */
1252 pending_child = lookup_one_len(fidname, mds->mds_pending_dir,
1254 if (IS_ERR(pending_child))
1255 GOTO(cleanup, rc = PTR_ERR(pending_child));
1256 LASSERT(pending_child->d_inode != NULL);
1258 cleanup_phase = 2; /* dput(pending_child) when finished */
1259 if (req != NULL && req->rq_repmsg != NULL) {
1260 lmm = lustre_msg_buf(req->rq_repmsg, 1, 0);
1261 stripe_count = le32_to_cpu(lmm->lmm_stripe_count);
1264 handle = fsfilt_start_log(obd, pending_dir, FSFILT_OP_UNLINK,
1265 NULL, stripe_count);
1266 if (IS_ERR(handle)) {
1267 rc = PTR_ERR(handle);
1272 pending_child->d_fsdata = (void *) &dp;
1275 if (S_ISDIR(pending_child->d_inode->i_mode))
1276 rc = vfs_rmdir(pending_dir, pending_child);
1278 rc = vfs_unlink(pending_dir, pending_child);
1280 CERROR("error unlinking orphan %s: rc %d\n",fidname,rc);
1282 if (req != NULL && req->rq_repmsg != NULL &&
1283 (reply_body->valid & OBD_MD_FLEASIZE) &&
1284 mds_log_op_unlink(obd, pending_child->d_inode,
1285 lmm, req->rq_repmsg->buflens[1],
1286 lustre_msg_buf(req->rq_repmsg, 2, 0),
1287 req->rq_repmsg->buflens[2], &lcl) > 0) {
1288 reply_body->valid |= OBD_MD_FLCOOKIE;
1291 goto out; /* Don't bother updating attrs on unlinked inode */
1295 if (request_body != NULL && mfd->mfd_mode & FMODE_WRITE && rc == 0) {
1296 /* Update the on-disk attributes if this was the last write
1297 * close, and all information was provided (i.e., rc == 0)
1299 * XXX this should probably be abstracted with mds_reint_setattr
1302 if (request_body->valid & OBD_MD_FLMTIME &&
1303 LTIME_S(request_body->mtime) > LTIME_S(inode->i_mtime)) {
1304 LTIME_S(iattr.ia_mtime) = LTIME_S(request_body->mtime);
1305 iattr.ia_valid |= ATTR_MTIME;
1307 if (request_body->valid & OBD_MD_FLCTIME &&
1308 LTIME_S(request_body->ctime) > LTIME_S(inode->i_ctime)) {
1309 LTIME_S(iattr.ia_ctime) = LTIME_S(request_body->ctime);
1310 iattr.ia_valid |= ATTR_CTIME;
1313 /* XXX can't set block count with fsfilt_setattr (!) */
1314 if (request_body->valid & OBD_MD_FLSIZE) {
1315 iattr.ia_valid |= ATTR_SIZE;
1316 iattr.ia_size = request_body->size;
1318 /* if (request_body->valid & OBD_MD_FLBLOCKS) {
1319 iattr.ia_valid |= ATTR_BLOCKS;
1320 iattr.ia_blocks = request_body->blocks
1325 if (request_body != NULL && request_body->valid & OBD_MD_FLATIME) {
1326 /* Only start a transaction to write out only the atime if
1327 * it is more out-of-date than the specified limit. If we
1328 * are already going to write out the atime then do it anyway.
1330 if ((request_body->atime >
1331 LTIME_S(inode->i_atime) + MAX_ATIME_DIFF) ||
1332 (iattr.ia_valid != 0 &&
1333 request_body->atime > LTIME_S(inode->i_atime))) {
1334 LTIME_S(iattr.ia_atime) = request_body->atime;
1335 iattr.ia_valid |= ATTR_ATIME;
1339 if (iattr.ia_valid != 0) {
1340 handle = fsfilt_start(obd, inode, FSFILT_OP_SETATTR, NULL);
1342 GOTO(cleanup, rc = PTR_ERR(handle));
1343 rc = fsfilt_setattr(obd, mfd->mfd_dentry, handle, &iattr, 0);
1345 CERROR("error in setattr(%s): rc %d\n", fidname, rc);
1348 /* If other clients have this file open for write, rc will be > 0 */
1351 l_dput(mfd->mfd_dentry);
1352 mds_mfd_destroy(mfd);
1355 if (req != NULL && reply_body != NULL) {
1356 rc = mds_finish_transno(mds, pending_dir, handle, req, rc, 0);
1357 } else if (handle) {
1358 int err = fsfilt_commit(obd, mds->mds_sb, pending_dir, handle, 0);
1360 CERROR("error committing close: %d\n", err);
1366 switch (cleanup_phase) {
1369 ptlrpc_save_llog_lock(req, lcl);
1370 dput(pending_child);
1372 up(&pending_dir->i_sem);
1377 int mds_close(struct ptlrpc_request *req, int offset)
1379 struct mds_export_data *med = &req->rq_export->exp_mds_data;
1380 struct obd_device *obd = req->rq_export->exp_obd;
1381 struct mds_body *body;
1382 struct mds_file_data *mfd;
1383 struct lvfs_run_ctxt saved;
1384 struct inode *inode;
1385 int rc, repsize[3] = {sizeof(struct mds_body),
1386 obd->u.mds.mds_max_mdsize,
1387 obd->u.mds.mds_max_cookiesize};
1390 MDS_UPDATE_COUNTER((&obd->u.mds), MDS_CLOSE_COUNT);
1392 rc = lustre_pack_reply(req, 3, repsize, NULL);
1394 CERROR("lustre_pack_reply: rc = %d\n", rc);
1395 req->rq_status = rc;
1397 MDS_CHECK_RESENT(req, mds_reconstruct_generic(req));
1400 if (lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY) {
1401 DEBUG_REQ(D_HA, req, "close replay\n");
1402 memcpy(lustre_msg_buf(req->rq_repmsg, 2, 0),
1403 lustre_msg_buf(req->rq_reqmsg, offset + 1, 0),
1404 req->rq_repmsg->buflens[2]);
1408 body = lustre_swab_reqbuf(req, offset, sizeof(*body),
1409 lustre_swab_mds_body);
1411 CERROR("Can't unpack body\n");
1412 req->rq_status = -EFAULT;
1416 if (body->flags & MDS_BFLAG_UNCOMMITTED_WRITES)
1417 /* do some stuff */ ;
1419 mfd = mds_handle2mfd(&body->handle);
1421 DEBUG_REQ(D_ERROR, req, "no handle for file close ino "LPD64
1422 ": cookie "LPX64, body->fid1.id, body->handle.cookie);
1423 req->rq_status = -ESTALE;
1427 inode = mfd->mfd_dentry->d_inode;
1428 /* child i_alloc_sem protects orphan_dec_test && is_orphan race */
1429 DOWN_WRITE_I_ALLOC_SEM(inode); /* mds_mfd_close drops this */
1430 if (mds_inode_is_orphan(inode) && mds_orphan_open_count(inode) == 1) {
1431 body = lustre_msg_buf(req->rq_repmsg, 0, sizeof (*body));
1432 LASSERT(body != NULL);
1434 mds_pack_inode2fid(obd, &body->fid1, inode);
1435 mds_pack_inode2body(obd, body, inode);
1436 mds_pack_md(obd, req->rq_repmsg, 1, body, inode, MDS_PACK_MD_LOCK);
1438 spin_lock(&med->med_open_lock);
1439 list_del(&mfd->mfd_list);
1440 spin_unlock(&med->med_open_lock);
1442 push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
1443 req->rq_status = mds_mfd_close(req, offset, obd, mfd, 1);
1444 pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
1446 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_CLOSE_PACK)) {
1447 CERROR("test case OBD_FAIL_MDS_CLOSE_PACK\n");
1448 req->rq_status = -ENOMEM;
1457 int mds_done_writing(struct ptlrpc_request *req, int offset)
1459 struct mds_body *body;
1460 int rc, size = sizeof(struct mds_body);
1463 MDS_CHECK_RESENT(req, mds_reconstruct_generic(req));
1465 body = lustre_swab_reqbuf(req, offset, sizeof(*body),
1466 lustre_swab_mds_body);
1468 CERROR("Can't unpack body\n");
1469 req->rq_status = -EFAULT;
1473 rc = lustre_pack_reply(req, 1, &size, NULL);
1475 CERROR("lustre_pack_reply: rc = %d\n", rc);
1476 req->rq_status = rc;