1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
4 * linux/mds/mds_reint.c
5 * Lustre Metadata Server (mds) reintegration routines
7 * Copyright (C) 2002 Cluster File Systems, Inc.
9 * This file is part of Lustre, http://www.lustre.org.
11 * Lustre is free software; you can redistribute it and/or
12 * modify it under the terms of version 2 of the GNU General Public
13 * License as published by the Free Software Foundation.
15 * Lustre is distributed in the hope that it will be useful,
16 * but WITHOUT ANY WARRANTY; without even the implied warranty of
17 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
18 * GNU General Public License for more details.
20 * You should have received a copy of the GNU General Public License
21 * along with Lustre; if not, write to the Free Software
22 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
24 * Author: Peter Braam <braam@clusterfs.com>
25 * Author: Andreas Dilger <adilger@clusterfs.com>
29 #define DEBUG_SUBSYSTEM S_MDS
31 #include <linux/obd_support.h>
32 #include <linux/obd_class.h>
33 #include <linux/obd.h>
34 #include <linux/lustre_lib.h>
35 #include <linux/lustre_idl.h>
36 #include <linux/lustre_mds.h>
37 #include <linux/lustre_dlm.h>
38 #include <linux/obd_class.h>
40 extern inline struct mds_obd *mds_req2mds(struct ptlrpc_request *req);
42 /* Assumes caller has already pushed us into the kernel context. */
43 int mds_update_last_rcvd(struct mds_obd *mds, void *handle,
44 struct ptlrpc_request *req)
46 struct mds_export_data *med = &req->rq_export->exp_mds_data;
47 struct mds_client_data *mcd = med->med_mcd;
52 /* we don't allocate new transnos for replayed requests */
53 if (req->rq_level == LUSTRE_CONN_RECOVD)
56 off = MDS_LR_CLIENT + med->med_off * MDS_LR_SIZE;
58 spin_lock(&mds->mds_last_lock);
59 last_rcvd = ++mds->mds_last_rcvd;
60 spin_unlock(&mds->mds_last_lock);
61 req->rq_repmsg->transno = HTON__u64(last_rcvd);
62 mcd->mcd_last_rcvd = cpu_to_le64(last_rcvd);
63 mcd->mcd_mount_count = cpu_to_le64(mds->mds_mount_count);
64 mcd->mcd_last_xid = cpu_to_le64(req->rq_xid);
66 mds_fs_set_last_rcvd(mds, handle);
67 rc = lustre_fwrite(mds->mds_rcvd_filp, (char *)mcd, sizeof(*mcd), &off);
68 CDEBUG(D_INODE, "wrote trans #"LPD64" for client '%s' at #%d: rc = "
69 "%d\n", last_rcvd, mcd->mcd_uuid, med->med_off, rc);
71 if (rc == sizeof(*mcd))
74 CERROR("error writing to last_rcvd file: rc = %d\n", rc);
82 /* In the write-back case, the client holds a lock on a subtree.
83 * In the intent case, the client holds a lock on the child inode.
84 * In the pathname case, the client (may) hold a lock on the child inode. */
85 static int mds_reint_setattr(struct mds_update_record *rec, int offset,
86 struct ptlrpc_request *req)
88 struct mds_obd *mds = mds_req2mds(req);
89 struct obd_device *obd = req->rq_export->exp_obd;
90 struct mds_body *body;
94 struct lustre_handle child_lockh;
97 if (req->rq_reqmsg->bufcount > offset + 1) {
99 struct lustre_handle dir_lockh;
103 /* a name was supplied by the client; fid1 is the directory */
104 dir = mds_fid2locked_dentry(obd, rec->ur_fid1, NULL, LCK_PR,
108 GOTO(out_setattr, rc = PTR_ERR(dir));
111 name = lustre_msg_buf(req->rq_reqmsg, offset + 1);
112 namelen = req->rq_reqmsg->buflens[offset + 1] - 1;
113 de = mds_name2locked_dentry(obd, dir, NULL, name, namelen,
114 0, &child_lockh, LCK_PR);
118 GOTO(out_setattr_de, rc = PTR_ERR(de));
121 de = mds_fid2dentry(mds, rec->ur_fid1, NULL);
122 if (!de || IS_ERR(de)) {
124 GOTO(out_setattr_de, rc = -ESTALE);
128 CDEBUG(D_INODE, "ino %ld\n", inode->i_ino);
130 OBD_FAIL_WRITE(OBD_FAIL_MDS_REINT_SETATTR_WRITE,
131 to_kdev_t(inode->i_sb->s_dev));
133 handle = mds_fs_start(mds, inode, MDS_FSOP_SETATTR);
135 GOTO(out_setattr_de, rc = PTR_ERR(handle));
137 rc = mds_fs_setattr(mds, de, handle, &rec->ur_iattr);
140 body = lustre_msg_buf(req->rq_repmsg, 1);
141 mds_pack_inode2fid(&body->fid1, inode);
142 mds_pack_inode2body(body, inode);
146 rc = mds_update_last_rcvd(mds, handle, req);
148 err = mds_fs_commit(mds, de->d_inode, handle);
150 CERROR("error on commit: err = %d\n", err);
163 static int mds_reint_create(struct mds_update_record *rec, int offset,
164 struct ptlrpc_request *req)
166 struct dentry *de = NULL;
167 struct mds_obd *mds = mds_req2mds(req);
168 struct obd_device *obd = req->rq_export->exp_obd;
169 struct dentry *dchild = NULL;
172 struct lustre_handle lockh;
173 int rc = 0, err, lock_mode, type = rec->ur_mode & S_IFMT;
176 /* requests were at offset 2, replies go back at 1 */
180 LASSERT(!strcmp(req->rq_export->exp_obd->obd_type->typ_name, "mds"));
182 lock_mode = (req->rq_reqmsg->opc == MDS_REINT) ? LCK_CW : LCK_PW;
184 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_REINT_CREATE))
185 GOTO(out_create, rc = -ESTALE);
187 de = mds_fid2locked_dentry(obd, rec->ur_fid1, NULL, lock_mode, &lockh);
190 CERROR("parent lookup error %d\n", rc);
192 GOTO(out_create, rc);
195 CDEBUG(D_INODE, "parent ino %ld name %s mode %o\n",
196 dir->i_ino, rec->ur_name, rec->ur_mode);
198 ldlm_lock_dump((void *)(unsigned long)lockh.addr);
201 dchild = lookup_one_len(rec->ur_name, de, rec->ur_namelen - 1);
202 if (IS_ERR(dchild)) {
203 CERROR("child lookup error %ld\n", PTR_ERR(dchild));
205 GOTO(out_create_de, rc = -ESTALE);
208 if (dchild->d_inode) {
209 struct mds_body *body;
210 struct inode *inode = dchild->d_inode;
212 CDEBUG(D_INODE, "child exists (dir %ld, name %s, ino %ld)\n",
213 dir->i_ino, rec->ur_name, dchild->d_inode->i_ino);
215 /* XXX check that mode is correct? */
217 body = lustre_msg_buf(req->rq_repmsg, offset);
218 mds_pack_inode2fid(&body->fid1, inode);
219 mds_pack_inode2body(body, inode);
220 if (S_ISREG(inode->i_mode)) {
221 struct lov_mds_md *lmm;
223 lmm = lustre_msg_buf(req->rq_repmsg, offset + 1);
224 lmm->lmm_easize = mds->mds_max_mdsize;
226 if (mds_fs_get_md(mds, inode, lmm) < 0) {
227 CDEBUG(D_INFO,"No md for %ld: rc %d\n",
229 memset(lmm, 0, lmm->lmm_easize);
231 body->valid |= OBD_MD_FLEASIZE;
234 /* This isn't an error for RECREATE. */
235 if (rec->ur_opcode & REINT_REPLAYING) {
236 CDEBUG(D_INODE, "EEXIST suppressed for REPLAYING\n");
241 GOTO(out_create_dchild, rc);
244 OBD_FAIL_WRITE(OBD_FAIL_MDS_REINT_CREATE_WRITE,
245 to_kdev_t(dir->i_sb->s_dev));
247 if (dir->i_mode & S_ISGID) {
248 rec->ur_gid = dir->i_gid;
249 if (S_ISDIR(rec->ur_mode))
250 rec->ur_mode |= S_ISGID;
255 handle = mds_fs_start(mds, dir, MDS_FSOP_CREATE);
257 GOTO(out_create_dchild, PTR_ERR(handle));
258 rc = vfs_create(dir, dchild, rec->ur_mode);
263 handle = mds_fs_start(mds, dir, MDS_FSOP_MKDIR);
265 GOTO(out_create_dchild, PTR_ERR(handle));
266 rc = vfs_mkdir(dir, dchild, rec->ur_mode);
271 handle = mds_fs_start(mds, dir, MDS_FSOP_SYMLINK);
273 GOTO(out_create_dchild, PTR_ERR(handle));
274 rc = vfs_symlink(dir, dchild, rec->ur_name);
282 int rdev = rec->ur_rdev;
283 handle = mds_fs_start(mds, dir, MDS_FSOP_MKNOD);
285 GOTO(out_create_dchild, PTR_ERR(handle));
286 rc = vfs_mknod(dir, dchild, rec->ur_mode, rdev);
291 CERROR("bad file type %o creating %s\n", type, rec->ur_name);
292 GOTO(out_create_dchild, rc = -EINVAL);
296 CDEBUG(D_INODE, "error during create: %d\n", rc);
297 GOTO(out_create_commit, rc);
300 struct inode *inode = dchild->d_inode;
301 struct mds_body *body;
303 iattr.ia_atime = rec->ur_time;
304 iattr.ia_ctime = rec->ur_time;
305 iattr.ia_mtime = rec->ur_time;
306 iattr.ia_uid = rec->ur_uid;
307 iattr.ia_gid = rec->ur_gid;
308 iattr.ia_valid = ATTR_UID | ATTR_GID | ATTR_ATIME |
309 ATTR_MTIME | ATTR_CTIME;
311 if (rec->ur_fid2->id) {
312 LASSERT(rec->ur_opcode & REINT_REPLAYING);
313 inode->i_generation = rec->ur_fid2->generation;
314 /* Dirtied and committed by this setattr: */
315 CDEBUG(D_INODE, "recreated ino %ld with gen %ld\n",
316 inode->i_ino, inode->i_generation);
318 CDEBUG(D_INODE, "created ino %ld\n", inode->i_ino);
321 rc = mds_fs_setattr(mds, dchild, handle, &iattr);
323 CERROR("error on setattr: rc = %d\n", rc);
324 /* XXX should we abort here in case of error? */
327 rc = mds_update_last_rcvd(mds, handle, req);
329 CERROR("error on mds_update_last_rcvd: rc = %d\n", rc);
330 GOTO(out_create_unlink, rc);
333 body = lustre_msg_buf(req->rq_repmsg, offset);
334 mds_pack_inode2fid(&body->fid1, inode);
335 mds_pack_inode2body(body, inode);
339 err = mds_fs_commit(mds, dir, handle);
341 CERROR("error on commit: err = %d\n", err);
347 ldlm_lock_decref(&lockh, lock_mode);
356 /* Destroy the file we just created. This should not need extra
357 * journal credits, as we have already modified all of the blocks
358 * needed in order to create the file in the first place.
362 err = vfs_rmdir(dir, dchild);
364 CERROR("failed rmdir in error path: rc = %d\n", err);
367 err = vfs_unlink(dir, dchild);
369 CERROR("failed unlink in error path: rc = %d\n", err);
373 goto out_create_commit;
376 static int mds_reint_unlink(struct mds_update_record *rec, int offset,
377 struct ptlrpc_request *req)
379 struct dentry *de = NULL;
380 struct dentry *dchild = NULL;
381 struct mds_obd *mds = mds_req2mds(req);
382 struct obd_device *obd = req->rq_export->exp_obd;
383 struct mds_body *body = NULL;
385 struct inode *dir, *inode;
386 struct lustre_handle lockh, child_lockh;
388 int namelen, lock_mode, err, rc = 0;
391 /* a name was supplied by the client; fid1 is the directory */
392 lock_mode = (req->rq_reqmsg->opc == MDS_REINT) ? LCK_PW : LCK_PW;
393 de = mds_fid2locked_dentry(obd, rec->ur_fid1, NULL, lock_mode, &lockh);
399 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_REINT_UNLINK))
400 GOTO(out_unlink, rc = -ENOENT);
402 name = lustre_msg_buf(req->rq_reqmsg, offset + 1);
403 namelen = req->rq_reqmsg->buflens[offset + 1] - 1;
404 #warning "FIXME: if mds_name2locked_dentry decrefs this lock, we must not"
405 memcpy(&child_lockh, &lockh, sizeof(child_lockh));
406 dchild = mds_name2locked_dentry(obd, de, NULL, name, namelen,
407 LCK_EX, &child_lockh, lock_mode);
409 if (IS_ERR(dchild)) {
411 GOTO(out_unlink, rc = PTR_ERR(dchild));
415 inode = dchild->d_inode;
416 CDEBUG(D_INODE, "parent ino %ld\n", dir->i_ino);
419 if (rec->ur_opcode & REINT_REPLAYING) {
421 "child missing (%ld/%s); OK for REPLAYING\n",
422 dir->i_ino, rec->ur_name);
426 "child doesn't exist (dir %ld, name %s)\n",
427 dir->i_ino, rec->ur_name);
430 /* going to out_unlink_cancel causes an LBUG, don't know why */
431 GOTO(out_unlink_dchild, rc);
438 body = lustre_msg_buf(req->rq_repmsg, offset);
439 mds_pack_inode2fid(&body->fid1, inode);
440 mds_pack_inode2body(body, inode);
443 OBD_FAIL_WRITE(OBD_FAIL_MDS_REINT_UNLINK_WRITE,
444 to_kdev_t(dir->i_sb->s_dev));
446 switch (rec->ur_mode /* & S_IFMT ? */) {
448 handle = mds_fs_start(mds, dir, MDS_FSOP_RMDIR);
450 GOTO(out_unlink_cancel, rc = PTR_ERR(handle));
451 rc = vfs_rmdir(dir, dchild);
454 /* get OBD EA data first so client can also destroy object */
455 if ((inode->i_mode & S_IFMT) == S_IFREG && offset) {
456 struct lov_mds_md *lmm;
458 lmm = lustre_msg_buf(req->rq_repmsg, offset + 1);
459 lmm->lmm_easize = mds->mds_max_mdsize;
460 if ((rc = mds_fs_get_md(mds, inode, lmm)) < 0) {
461 CDEBUG(D_INFO, "No md for ino %ld: rc = %d\n",
463 memset(lmm, 0, lmm->lmm_easize);
465 body->valid |= OBD_MD_FLEASIZE;
473 handle = mds_fs_start(mds, dir, MDS_FSOP_UNLINK);
475 GOTO(out_unlink_cancel, rc = PTR_ERR(handle));
476 rc = vfs_unlink(dir, dchild);
479 CERROR("bad file type %o unlinking %s\n", rec->ur_mode, name);
482 GOTO(out_unlink_cancel, rc = -EINVAL);
486 rc = mds_update_last_rcvd(mds, handle, req);
487 err = mds_fs_commit(mds, dir, handle);
489 CERROR("error on commit: err = %d\n", err);
497 ldlm_lock_decref(&child_lockh, LCK_EX);
498 err = ldlm_cli_cancel(&child_lockh);
500 CERROR("failed to cancel child inode lock: err = %d\n", err);
502 rc = -ENOLCK; /*XXX translate LDLM lock error */
508 ldlm_lock_decref(&lockh, lock_mode);
514 static int mds_reint_link(struct mds_update_record *rec, int offset,
515 struct ptlrpc_request *req)
517 struct obd_device *obd = req->rq_export->exp_obd;
518 struct dentry *de_src = NULL;
519 struct dentry *de_tgt_dir = NULL;
520 struct dentry *dchild = NULL;
521 struct mds_obd *mds = mds_req2mds(req);
522 struct lustre_handle *handle, tgtlockh, srclockh;
524 __u64 res_id[3] = { 0 };
529 de_src = mds_fid2dentry(mds, rec->ur_fid1, NULL);
530 if (IS_ERR(de_src) || OBD_FAIL_CHECK(OBD_FAIL_MDS_REINT_LINK)) {
531 GOTO(out_link, rc = -ESTALE);
534 /* plan to change the link count on this inode: write lock */
535 lock_mode = (req->rq_reqmsg->opc == MDS_REINT) ? LCK_PW : LCK_PW;
536 res_id[0] = de_src->d_inode->i_ino;
537 res_id[1] = de_src->d_inode->i_generation;
539 rc = ldlm_lock_match(obd->obd_namespace, res_id, LDLM_PLAIN,
540 NULL, 0, lock_mode, &srclockh);
542 LDLM_DEBUG_NOLOCK("enqueue res "LPU64, res_id[0]);
543 rc = ldlm_cli_enqueue(NULL, NULL, obd->obd_namespace, NULL,
544 res_id, LDLM_PLAIN, NULL, 0, lock_mode,
545 &flags, ldlm_completion_ast,
546 mds_blocking_ast, NULL, 0, &srclockh);
547 if (rc != ELDLM_OK) {
548 CERROR("lock enqueue: err: %d\n", rc);
549 GOTO(out_link_src_put, rc = -EIO);
552 ldlm_lock_dump((void *)(unsigned long)srclockh.addr);
554 de_tgt_dir = mds_fid2dentry(mds, rec->ur_fid2, NULL);
555 if (IS_ERR(de_tgt_dir)) {
556 GOTO(out_link_src, rc = -ESTALE);
559 lock_mode = (req->rq_reqmsg->opc == MDS_REINT) ? LCK_PW : LCK_PW;
560 res_id[0] = de_tgt_dir->d_inode->i_ino;
561 res_id[1] = de_tgt_dir->d_inode->i_generation;
563 rc = ldlm_lock_match(obd->obd_namespace, res_id, LDLM_PLAIN,
564 NULL, 0, lock_mode, &tgtlockh);
566 LDLM_DEBUG_NOLOCK("enqueue res "LPU64, res_id[0]);
567 rc = ldlm_cli_enqueue(NULL, NULL, obd->obd_namespace, NULL,
568 res_id, LDLM_PLAIN, NULL, 0, lock_mode,
569 &flags, ldlm_completion_ast,
570 mds_blocking_ast, NULL, 0, &tgtlockh);
571 if (rc != ELDLM_OK) {
572 CERROR("lock enqueue: err: %d\n", rc);
573 GOTO(out_link_tgt_dir_put, rc = -EIO);
576 ldlm_lock_dump((void *)(unsigned long)tgtlockh.addr);
578 down(&de_tgt_dir->d_inode->i_sem);
579 dchild = lookup_one_len(rec->ur_name, de_tgt_dir, rec->ur_namelen - 1);
580 if (IS_ERR(dchild)) {
581 CERROR("child lookup error %ld\n", PTR_ERR(dchild));
582 GOTO(out_link_tgt_dir, rc = -ESTALE);
585 if (dchild->d_inode) {
586 struct inode *inode = dchild->d_inode;
587 /* in intent case ship back attributes to client */
589 struct mds_body *body =
590 lustre_msg_buf(req->rq_repmsg, 1);
592 mds_pack_inode2fid(&body->fid1, inode);
593 mds_pack_inode2body(body, inode);
594 if (S_ISREG(inode->i_mode)) {
595 struct lov_mds_md *lmm;
597 lmm = lustre_msg_buf(req->rq_repmsg, 2);
598 lmm->lmm_easize = mds->mds_max_mdsize;
599 if ((rc = mds_fs_get_md(mds, inode, lmm)) < 0) {
600 CDEBUG(D_INFO,"No md for %ld: rc %d\n",
602 memset(lmm, 0, lmm->lmm_easize);
604 body->valid |= OBD_MD_FLEASIZE;
607 if (rec->ur_opcode & REINT_REPLAYING) {
608 /* XXX verify that the link is to the the right file? */
611 "child exists (dir %ld, name %s) (REPLAYING)\n",
612 de_tgt_dir->d_inode->i_ino, rec->ur_name);
615 CERROR("child exists (dir %ld, name %s)\n",
616 de_tgt_dir->d_inode->i_ino, rec->ur_name);
618 GOTO(out_link_dchild, rc);
621 OBD_FAIL_WRITE(OBD_FAIL_MDS_REINT_LINK_WRITE,
622 to_kdev_t(de_src->d_inode->i_sb->s_dev));
624 handle = mds_fs_start(mds, de_tgt_dir->d_inode, MDS_FSOP_LINK);
626 GOTO(out_link_dchild, rc = PTR_ERR(handle));
628 rc = vfs_link(de_src, de_tgt_dir->d_inode, dchild);
630 CERROR("link error %d\n", rc);
632 rc = mds_update_last_rcvd(mds, handle, req);
634 err = mds_fs_commit(mds, de_tgt_dir->d_inode, handle);
636 CERROR("error on commit: err = %d\n", err);
645 ldlm_lock_decref(&tgtlockh, lock_mode);
646 out_link_tgt_dir_put:
647 up(&de_tgt_dir->d_inode->i_sem);
650 ldlm_lock_decref(&srclockh, lock_mode);
658 static int mds_reint_rename(struct mds_update_record *rec, int offset,
659 struct ptlrpc_request *req)
661 struct obd_device *obd = req->rq_export->exp_obd;
662 struct dentry *de_srcdir = NULL;
663 struct dentry *de_tgtdir = NULL;
664 struct dentry *de_old = NULL;
665 struct dentry *de_new = NULL;
666 struct mds_obd *mds = mds_req2mds(req);
667 struct lustre_handle tgtlockh, srclockh, oldhandle;
668 int flags = 0, lock_mode, rc = 0, err;
670 __u64 res_id[3] = { 0 };
673 de_srcdir = mds_fid2dentry(mds, rec->ur_fid1, NULL);
674 if (IS_ERR(de_srcdir))
675 GOTO(out_rename, rc = -ESTALE);
677 lock_mode = (req->rq_reqmsg->opc == MDS_REINT) ? LCK_PW : LCK_PW;
678 res_id[0] = de_srcdir->d_inode->i_ino;
679 res_id[1] = de_srcdir->d_inode->i_generation;
681 rc = ldlm_lock_match(obd->obd_namespace, res_id, LDLM_PLAIN,
682 NULL, 0, lock_mode, &srclockh);
684 LDLM_DEBUG_NOLOCK("enqueue res "LPU64, res_id[0]);
685 rc = ldlm_cli_enqueue(NULL, NULL, obd->obd_namespace, NULL,
686 res_id, LDLM_PLAIN, NULL, 0, lock_mode,
687 &flags, ldlm_completion_ast,
688 mds_blocking_ast, NULL, 0, &srclockh);
689 if (rc != ELDLM_OK) {
690 CERROR("lock enqueue: err: %d\n", rc);
691 GOTO(out_rename_srcput, rc = -EIO);
694 ldlm_lock_dump((void *)(unsigned long)srclockh.addr);
696 de_tgtdir = mds_fid2dentry(mds, rec->ur_fid2, NULL);
697 if (IS_ERR(de_tgtdir))
698 GOTO(out_rename_srcdir, rc = -ESTALE);
700 lock_mode = (req->rq_reqmsg->opc == MDS_REINT) ? LCK_PW : LCK_PW;
701 res_id[0] = de_tgtdir->d_inode->i_ino;
702 res_id[1] = de_tgtdir->d_inode->i_generation;
704 rc = ldlm_lock_match(obd->obd_namespace, res_id, LDLM_PLAIN,
705 NULL, 0, lock_mode, &tgtlockh);
708 LDLM_DEBUG_NOLOCK("enqueue res "LPU64, res_id[0]);
709 rc = ldlm_cli_enqueue(NULL, NULL, obd->obd_namespace, NULL,
710 res_id, LDLM_PLAIN, NULL, 0, lock_mode,
711 &flags, ldlm_completion_ast,
712 mds_blocking_ast, NULL, 0, &tgtlockh);
713 if (rc != ELDLM_OK) {
714 CERROR("lock enqueue: err: %d\n", rc);
715 GOTO(out_rename_tgtput, rc = -EIO);
718 ldlm_lock_dump((void *)(unsigned long)tgtlockh.addr);
720 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
721 double_lock(de_tgtdir, de_srcdir);
723 de_old = lookup_one_len(rec->ur_name, de_srcdir, rec->ur_namelen - 1);
724 if (IS_ERR(de_old)) {
725 CERROR("old child lookup error (%*s): %ld\n",
726 rec->ur_namelen - 1, rec->ur_name, PTR_ERR(de_old));
727 GOTO(out_rename_tgtdir, rc = -ENOENT);
730 de_new = lookup_one_len(rec->ur_tgt, de_tgtdir, rec->ur_tgtlen - 1);
731 if (IS_ERR(de_new)) {
732 CERROR("new child lookup error (%*s): %ld\n",
733 rec->ur_tgtlen - 1, rec->ur_tgt, PTR_ERR(de_new));
734 GOTO(out_rename_deold, rc = -ENOENT);
737 /* in intent case ship back attributes to client */
739 struct mds_body *body = lustre_msg_buf(req->rq_repmsg, 1);
740 struct inode *inode = de_new->d_inode;
745 mds_pack_inode2fid(&body->fid1, inode);
746 mds_pack_inode2body(body, inode);
747 if (S_ISREG(inode->i_mode)) {
748 struct lov_mds_md *lmm;
750 lmm = lustre_msg_buf(req->rq_repmsg, 2);
751 lmm->lmm_easize = mds->mds_max_mdsize;
752 if ((rc = mds_fs_get_md(mds, inode, lmm)) < 0) {
753 CDEBUG(D_INFO,"No md for %ld: rc %d\n",
755 memset(lmm, 0, lmm->lmm_easize);
757 body->valid |= OBD_MD_FLEASIZE;
762 OBD_FAIL_WRITE(OBD_FAIL_MDS_REINT_RENAME_WRITE,
763 to_kdev_t(de_srcdir->d_inode->i_sb->s_dev));
765 handle = mds_fs_start(mds, de_tgtdir->d_inode, MDS_FSOP_RENAME);
767 GOTO(out_rename_denew, rc = PTR_ERR(handle));
769 rc = vfs_rename(de_srcdir->d_inode, de_old, de_tgtdir->d_inode, de_new,
774 rc = mds_update_last_rcvd(mds, handle, req);
776 err = mds_fs_commit(mds, de_tgtdir->d_inode, handle);
778 CERROR("error on commit: err = %d\n", err);
788 res_id[0] = de_old->d_inode->i_ino;
789 res_id[1] = de_old->d_inode->i_generation;
791 /* Take an exclusive lock on the resource that we're
792 * about to free, to force everyone to drop their
794 LDLM_DEBUG_NOLOCK("getting EX lock res "LPU64, res_id[0]);
795 rc = ldlm_cli_enqueue(NULL, NULL, obd->obd_namespace, NULL,
796 res_id, LDLM_PLAIN, NULL, 0, LCK_EX,
797 &flags, ldlm_completion_ast,
798 mds_blocking_ast, NULL, 0, &oldhandle);
800 CERROR("failed to get child inode lock (child ino "
801 LPD64" dir ino %ld)\n",
802 res_id[0], de_old->d_inode->i_ino);
808 ldlm_lock_decref(&oldhandle, LCK_EX);
809 rc = ldlm_cli_cancel(&oldhandle);
811 CERROR("failed to cancel child inode lock ino "
812 LPD64": %d\n", res_id[0], rc);
815 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
816 double_up(&de_srcdir->d_inode->i_sem, &de_tgtdir->d_inode->i_sem);
818 ldlm_lock_decref(&tgtlockh, lock_mode);
822 ldlm_lock_decref(&srclockh, lock_mode);
830 typedef int (*mds_reinter) (struct mds_update_record *, int offset,
831 struct ptlrpc_request *);
833 static mds_reinter reinters[REINT_MAX + 1] = {
834 [REINT_SETATTR] mds_reint_setattr,
835 [REINT_CREATE] mds_reint_create,
836 [REINT_UNLINK] mds_reint_unlink,
837 [REINT_LINK] mds_reint_link,
838 [REINT_RENAME] mds_reint_rename,
841 int mds_reint_rec(struct mds_update_record *rec, int offset,
842 struct ptlrpc_request *req)
844 struct mds_obd *mds = mds_req2mds(req);
845 struct obd_run_ctxt saved;
847 int realop = rec->ur_opcode & REINT_OPCODE_MASK;
850 if (realop < 1 || realop > REINT_MAX) {
851 CERROR("opcode %d not valid (%sREPLAYING)\n", realop,
852 rec->ur_opcode & REINT_REPLAYING ? "" : "not ");
853 rc = req->rq_status = -EINVAL;
857 uc.ouc_fsuid = rec->ur_fsuid;
858 uc.ouc_fsgid = rec->ur_fsgid;
860 push_ctxt(&saved, &mds->mds_ctxt, &uc);
861 rc = reinters[realop] (rec, offset, req);