1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
4 * linux/mds/mds_reint.c
6 * Lustre Metadata Server (mds) reintegration routines
8 * Copyright (C) 2002 Cluster File Systems, Inc.
9 * author: Peter Braam <braam@clusterfs.com>
11 * This code is issued under the GNU General Public License.
12 * See the file COPYING in this distribution
16 // XXX - add transaction sequence numbers
20 #include <linux/version.h>
21 #include <linux/module.h>
23 #include <linux/stat.h>
24 #include <linux/locks.h>
25 #include <linux/quotaops.h>
26 #include <asm/unistd.h>
27 #include <asm/uaccess.h>
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/obd_class.h>
39 struct mds_client_info *mds_uuid_to_mci(struct mds_obd *mds, __u8 *uuid)
46 list_for_each(p, &mds->mds_client_info) {
47 struct mds_client_info *mci;
49 mci = list_entry(p, struct mds_client_info, mci_list);
50 CDEBUG(D_INFO, "checking client UUID '%s'\n",
51 mci->mci_mcd->mcd_uuid);
52 if (!strncmp(mci->mci_mcd->mcd_uuid, uuid,
53 sizeof(mci->mci_mcd->mcd_uuid)))
56 CDEBUG(D_INFO, "no mds client info found for UUID '%s'\n", uuid);
60 int mds_update_last_rcvd(struct mds_obd *mds, void *handle,
61 struct ptlrpc_request *req)
63 /* get from req->rq_connection-> or req->rq_client */
64 struct mds_client_info *mci;
68 mci = mds_uuid_to_mci(mds, req->rq_connection->c_remote_uuid);
70 CERROR("unable to locate MDS client data for UUID '%s'\n",
71 ptlrpc_req_to_uuid(req));
72 /* This will be a real error once everything is working */
77 off = MDS_LR_CLIENT + mci->mci_off * MDS_LR_SIZE;
79 ++mds->mds_last_rcvd; /* lock this, or make it an LDLM function? */
80 req->rq_repmsg->transno = HTON__u64(mds->mds_last_rcvd);
81 mci->mci_mcd->mcd_last_rcvd = cpu_to_le64(mds->mds_last_rcvd);
82 mci->mci_mcd->mcd_mount_count = cpu_to_le64(mds->mds_mount_count);
83 mci->mci_mcd->mcd_last_xid = cpu_to_le32(req->rq_reqmsg->xid);
85 mds_fs_set_last_rcvd(mds, handle);
86 rc = lustre_fwrite(mds->mds_rcvd_filp, (char *)mci->mci_mcd,
87 sizeof(*mci->mci_mcd), &off);
88 CDEBUG(D_INODE, "wrote trans #%Ld for client '%s' at #%d: rc = %d\n",
89 mds->mds_last_rcvd, mci->mci_mcd->mcd_uuid, mci->mci_off, rc);
90 // store new value and last committed value in req struct
92 if (rc == sizeof(mci->mci_mcd))
100 static int mds_reint_setattr(struct mds_update_record *rec,
101 struct ptlrpc_request *req)
103 struct mds_obd *mds = &req->rq_obd->u.mds;
108 de = mds_fid2dentry(mds, rec->ur_fid1, NULL);
109 if (IS_ERR(de) || OBD_FAIL_CHECK(OBD_FAIL_MDS_REINT_SETATTR)) {
110 GOTO(out_setattr, rc = -ESTALE);
113 CDEBUG(D_INODE, "ino %ld\n", de->d_inode->i_ino);
115 OBD_FAIL_WRITE(OBD_FAIL_MDS_REINT_SETATTR_WRITE,
116 de->d_inode->i_sb->s_dev);
118 handle = mds_fs_start(mds, de->d_inode, MDS_FSOP_SETATTR);
120 GOTO(out_setattr_de, rc = PTR_ERR(handle));
121 rc = mds_fs_setattr(mds, de, handle, &rec->ur_iattr);
124 rc = mds_update_last_rcvd(mds, handle, req);
125 /* FIXME: need to return last_rcvd, last_committed */
129 /* FIXME: keep rc intact */
130 rc = mds_fs_commit(mds, de->d_inode, handle);
138 static int mds_reint_recreate(struct mds_update_record *rec,
139 struct ptlrpc_request *req)
141 struct dentry *de = NULL;
142 struct mds_obd *mds = &req->rq_obd->u.mds;
143 struct dentry *dchild = NULL;
148 de = mds_fid2dentry(mds, rec->ur_fid1, NULL);
149 if (IS_ERR(de) || OBD_FAIL_CHECK(OBD_FAIL_MDS_REINT_CREATE)) {
151 GOTO(out_create_de, rc = -ESTALE);
154 CDEBUG(D_INODE, "parent ino %ld\n", dir->i_ino);
157 dchild = lookup_one_len(rec->ur_name, de, rec->ur_namelen - 1);
158 if (IS_ERR(dchild)) {
159 CERROR("child lookup error %ld\n", PTR_ERR(dchild));
162 GOTO(out_create_dchild, rc = -ESTALE);
165 if (dchild->d_inode) {
166 struct mds_body *body;
168 body = lustre_msg_buf(req->rq_repmsg, 0);
169 body->ino = dchild->d_inode->i_ino;
170 body->generation = dchild->d_inode->i_generation;
172 CERROR("child doesn't exist (dir %ld, name %s)\n",
173 dir->i_ino, rec->ur_name);
186 static int mds_reint_create(struct mds_update_record *rec,
187 struct ptlrpc_request *req)
189 struct dentry *de = NULL;
190 struct mds_obd *mds = &req->rq_obd->u.mds;
191 struct dentry *dchild = NULL;
194 int rc = 0, type = rec->ur_mode & S_IFMT;
197 de = mds_fid2dentry(mds, rec->ur_fid1, NULL);
198 if (IS_ERR(de) || OBD_FAIL_CHECK(OBD_FAIL_MDS_REINT_CREATE)) {
200 GOTO(out_create_de, rc = -ESTALE);
203 CDEBUG(D_INODE, "parent ino %ld name %s mode %o\n",
204 dir->i_ino, rec->ur_name, rec->ur_mode);
207 dchild = lookup_one_len(rec->ur_name, de, rec->ur_namelen - 1);
208 if (IS_ERR(dchild)) {
209 CERROR("child lookup error %ld\n", PTR_ERR(dchild));
212 GOTO(out_create_dchild, rc = -ESTALE);
215 if (dchild->d_inode) {
216 CERROR("child exists (dir %ld, name %s, ino %ld)\n",
217 dir->i_ino, rec->ur_name, dchild->d_inode->i_ino);
219 GOTO(out_create_dchild, rc = -EEXIST);
222 OBD_FAIL_WRITE(OBD_FAIL_MDS_REINT_CREATE_WRITE, dir->i_sb->s_dev);
226 handle = mds_fs_start(mds, dir, MDS_FSOP_CREATE);
228 GOTO(out_create_dchild, PTR_ERR(handle));
229 rc = vfs_create(dir, dchild, rec->ur_mode);
234 handle = mds_fs_start(mds, dir, MDS_FSOP_MKDIR);
236 GOTO(out_create_dchild, PTR_ERR(handle));
237 rc = vfs_mkdir(dir, dchild, rec->ur_mode);
242 handle = mds_fs_start(mds, dir, MDS_FSOP_SYMLINK);
244 GOTO(out_create_dchild, PTR_ERR(handle));
245 rc = vfs_symlink(dir, dchild, rec->ur_tgt);
253 int rdev = rec->ur_id;
254 handle = mds_fs_start(mds, dir, MDS_FSOP_MKNOD);
256 GOTO(out_create_dchild, PTR_ERR(handle));
257 rc = vfs_mknod(dir, dchild, rec->ur_mode, rdev);
262 CERROR("bad file type %d for create of %s\n",type,rec->ur_name);
263 GOTO(out_create_dchild, rc = -EINVAL);
267 CERROR("error during create: %d\n", rc);
268 if (rc != -ENOSPC) LBUG();
269 GOTO(out_create_commit, rc);
272 struct inode *inode = dchild->d_inode;
273 struct mds_body *body;
275 CDEBUG(D_INODE, "created ino %ld\n", dchild->d_inode->i_ino);
276 if (type == S_IFREG) {
277 rc = mds_fs_set_objid(mds, inode, handle, rec->ur_id);
279 CERROR("error %d setting objid for %ld\n",
283 iattr.ia_atime = rec->ur_time;
284 iattr.ia_ctime = rec->ur_time;
285 iattr.ia_mtime = rec->ur_time;
286 iattr.ia_uid = rec->ur_uid;
287 iattr.ia_gid = rec->ur_gid;
288 iattr.ia_valid = ATTR_UID | ATTR_GID | ATTR_ATIME |
289 ATTR_MTIME | ATTR_CTIME;
291 rc = mds_fs_setattr(mds, dchild, handle, &iattr);
292 /* XXX should we abort here in case of error? */
295 rc = mds_update_last_rcvd(mds, handle, req);
297 body = lustre_msg_buf(req->rq_repmsg, 0);
298 body->ino = inode->i_ino;
299 body->generation = inode->i_generation;
303 /* FIXME: keep rc intact */
304 rc = mds_fs_commit(mds, dir, handle);
314 static int mds_reint_unlink(struct mds_update_record *rec,
315 struct ptlrpc_request *req)
317 struct dentry *de = NULL;
318 struct dentry *dchild = NULL;
319 struct mds_obd *mds = &req->rq_obd->u.mds;
320 struct inode *dir, *inode;
325 de = mds_fid2dentry(mds, rec->ur_fid1, NULL);
326 if (IS_ERR(de) || OBD_FAIL_CHECK(OBD_FAIL_MDS_REINT_UNLINK)) {
328 GOTO(out_unlink, rc = -ESTALE);
331 CDEBUG(D_INODE, "parent ino %ld\n", dir->i_ino);
334 dchild = lookup_one_len(rec->ur_name, de, rec->ur_namelen - 1);
335 if (IS_ERR(dchild)) {
336 CERROR("child lookup error %ld\n", PTR_ERR(dchild));
338 GOTO(out_unlink_de, rc = -ESTALE);
341 inode = dchild->d_inode;
343 CERROR("child doesn't exist (dir %ld, name %s\n",
344 dir->i_ino, rec->ur_name);
346 GOTO(out_unlink_dchild, rc = -ESTALE);
349 if (inode->i_ino != rec->ur_fid2->id) {
350 CERROR("inode and FID ID do not match (%ld != %Ld)\n",
351 inode->i_ino, rec->ur_fid2->id);
353 GOTO(out_unlink_dchild, rc = -ESTALE);
355 if (inode->i_generation != rec->ur_fid2->generation) {
356 CERROR("inode and FID GENERATION do not match (%d != %d)\n",
357 inode->i_generation, rec->ur_fid2->generation);
359 GOTO(out_unlink_dchild, rc = -ESTALE);
362 OBD_FAIL_WRITE(OBD_FAIL_MDS_REINT_UNLINK_WRITE, dir->i_sb->s_dev);
364 switch (dchild->d_inode->i_mode & S_IFMT) {
366 handle = mds_fs_start(mds, dir, MDS_FSOP_RMDIR);
368 GOTO(out_unlink_dchild, rc = PTR_ERR(handle));
369 rc = vfs_rmdir(dir, dchild);
372 handle = mds_fs_start(mds, dir, MDS_FSOP_UNLINK);
374 GOTO(out_unlink_dchild, rc = PTR_ERR(handle));
375 rc = vfs_unlink(dir, dchild);
380 rc = mds_update_last_rcvd(mds, handle, req);
381 /* FIXME: need to return last_rcvd, last_committed */
382 /* FIXME: keep rc intact */
383 rc = mds_fs_commit(mds, dir, handle);
396 static int mds_reint_link(struct mds_update_record *rec,
397 struct ptlrpc_request *req)
399 struct dentry *de_src = NULL;
400 struct dentry *de_tgt_dir = NULL;
401 struct dentry *dchild = NULL;
402 struct mds_obd *mds = &req->rq_obd->u.mds;
407 de_src = mds_fid2dentry(mds, rec->ur_fid1, NULL);
408 if (IS_ERR(de_src) || OBD_FAIL_CHECK(OBD_FAIL_MDS_REINT_LINK)) {
409 GOTO(out_link, rc = -ESTALE);
412 de_tgt_dir = mds_fid2dentry(mds, rec->ur_fid2, NULL);
413 if (IS_ERR(de_tgt_dir)) {
414 GOTO(out_link_de_src, rc = -ESTALE);
417 down(&de_tgt_dir->d_inode->i_sem);
418 dchild = lookup_one_len(rec->ur_name, de_tgt_dir, rec->ur_namelen - 1);
419 if (IS_ERR(dchild)) {
420 CERROR("child lookup error %ld\n", PTR_ERR(dchild));
421 GOTO(out_link_de_tgt_dir, rc = -ESTALE);
424 if (dchild->d_inode) {
425 CERROR("child exists (dir %ld, name %s\n",
426 de_tgt_dir->d_inode->i_ino, rec->ur_name);
427 GOTO(out_link_dchild, rc = -EEXIST);
430 OBD_FAIL_WRITE(OBD_FAIL_MDS_REINT_LINK_WRITE,
431 dchild->d_inode->i_sb->s_dev);
433 handle = mds_fs_start(mds, de_tgt_dir->d_inode, MDS_FSOP_LINK);
435 GOTO(out_link_dchild, rc = PTR_ERR(handle));
437 rc = vfs_link(de_src, de_tgt_dir->d_inode, dchild);
440 rc = mds_update_last_rcvd(mds, handle, req);
442 /* FIXME: need to return last_rcvd, last_committed */
443 /* FIXME: keep rc intact */
444 rc = mds_fs_commit(mds, de_tgt_dir->d_inode, handle);
450 up(&de_tgt_dir->d_inode->i_sem);
459 static int mds_reint_rename(struct mds_update_record *rec,
460 struct ptlrpc_request *req)
462 struct dentry *de_srcdir = NULL;
463 struct dentry *de_tgtdir = NULL;
464 struct dentry *de_old = NULL;
465 struct dentry *de_new = NULL;
466 struct mds_obd *mds = &req->rq_obd->u.mds;
471 de_srcdir = mds_fid2dentry(mds, rec->ur_fid1, NULL);
472 if (IS_ERR(de_srcdir)) {
473 GOTO(out_rename, rc = -ESTALE);
476 de_tgtdir = mds_fid2dentry(mds, rec->ur_fid2, NULL);
477 if (IS_ERR(de_tgtdir)) {
478 GOTO(out_rename_srcdir, rc = -ESTALE);
481 de_old = lookup_one_len(rec->ur_name, de_srcdir, rec->ur_namelen - 1);
482 if (IS_ERR(de_old)) {
483 CERROR("old child lookup error %ld\n", PTR_ERR(de_old));
484 GOTO(out_rename_tgtdir, rc = -ESTALE);
487 de_new = lookup_one_len(rec->ur_tgt, de_tgtdir, rec->ur_tgtlen - 1);
488 if (IS_ERR(de_new)) {
489 CERROR("new child lookup error %ld\n", PTR_ERR(de_new));
490 GOTO(out_rename_deold, rc = -ESTALE);
493 OBD_FAIL_WRITE(OBD_FAIL_MDS_REINT_RENAME_WRITE,
494 de_srcdir->d_inode->i_sb->s_dev);
496 handle = mds_fs_start(mds, de_tgtdir->d_inode, MDS_FSOP_RENAME);
498 GOTO(out_rename_denew, rc = PTR_ERR(handle));
499 rc = vfs_rename(de_srcdir->d_inode, de_old, de_tgtdir->d_inode, de_new);
502 rc = mds_update_last_rcvd(mds, handle, req);
504 /* FIXME: need to return last_rcvd, last_committed */
505 /* FIXME: keep rc intact */
506 rc = mds_fs_commit(mds, de_tgtdir->d_inode, handle);
522 typedef int (*mds_reinter)(struct mds_update_record *, struct ptlrpc_request*);
524 static mds_reinter reinters[REINT_MAX+1] = {
525 [REINT_SETATTR] mds_reint_setattr,
526 [REINT_CREATE] mds_reint_create,
527 [REINT_UNLINK] mds_reint_unlink,
528 [REINT_LINK] mds_reint_link,
529 [REINT_RENAME] mds_reint_rename,
530 [REINT_RECREATE] mds_reint_recreate,
533 int mds_reint_rec(struct mds_update_record *rec, struct ptlrpc_request *req)
535 int rc, size = sizeof(struct mds_body);
537 if (rec->ur_opcode < 1 || rec->ur_opcode > REINT_MAX) {
538 CERROR("opcode %d not valid\n", rec->ur_opcode);
539 rc = req->rq_status = -EINVAL;
543 rc = lustre_pack_msg(1, &size, NULL, &req->rq_replen, &req->rq_repmsg);
545 CERROR("mds: out of memory\n");
546 rc = req->rq_status = -ENOMEM;
550 rc = reinters[rec->ur_opcode](rec, req);