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 int mds_update_last_rcvd(struct mds_obd *mds, struct ptlrpc_request *req)
41 /* get from req->rq_connection-> or req->rq_client */
42 struct mds_client_info mci_data, *mci = &mci_data;
43 struct mds_client_data mcd_data, *mcd = &mcd_data;
47 /* Just a placeholder until more gets committed */
48 memset(mcd, 0, sizeof(*mcd));
50 mci->mci_off = off = MDS_LR_CLIENT;
52 ++mds->mds_last_rcvd; /* lock this, or make it an LDLM function? */
53 mci->mci_mcd->mcd_last_rcvd = cpu_to_le64(mds->mds_last_rcvd);
54 mci->mci_mcd->mcd_mount_count = cpu_to_le64(mds->mds_mount_count);
55 mci->mci_mcd->mcd_xid = cpu_to_le32(req->rq_connection->c_xid_in);
56 rc = lustre_fwrite(mds->mds_rcvd_filp, (char *)mci->mci_mcd,
57 sizeof(*mci->mci_mcd), &off);
58 CDEBUG(D_INODE, "wrote trans #%Ld for client '%s' at %Ld: rc = %d\n",
59 mds->mds_last_rcvd, mci->mci_mcd->mcd_uuid, mci->mci_off, rc);
60 // store new value and last committed value in req struct
62 if (rc == sizeof(mci->mci_mcd))
70 static int mds_reint_setattr(struct mds_update_record *rec,
71 struct ptlrpc_request *req)
73 struct mds_obd *mds = &req->rq_obd->u.mds;
78 de = mds_fid2dentry(mds, rec->ur_fid1, NULL);
79 if (IS_ERR(de) || OBD_FAIL_CHECK(OBD_FAIL_MDS_REINT_SETATTR)) {
80 GOTO(out_setattr, rc = -ESTALE);
83 CDEBUG(D_INODE, "ino %ld\n", de->d_inode->i_ino);
85 OBD_FAIL_WRITE(OBD_FAIL_MDS_REINT_SETATTR_WRITE,
86 de->d_inode->i_sb->s_dev);
88 handle = mds_fs_start(mds, de->d_inode, MDS_FSOP_SETATTR);
90 GOTO(out_setattr_de, rc = PTR_ERR(handle));
91 rc = mds_fs_setattr(mds, de, handle, &rec->ur_iattr);
94 rc = mds_update_last_rcvd(mds, req);
98 /* FIXME: keep rc intact */
99 rc = mds_fs_commit(mds, de->d_inode, handle);
107 static int mds_reint_create(struct mds_update_record *rec,
108 struct ptlrpc_request *req)
110 struct dentry *de = NULL;
111 struct mds_obd *mds = &req->rq_obd->u.mds;
112 struct dentry *dchild = NULL;
115 int rc = 0, type = rec->ur_mode & S_IFMT;
118 de = mds_fid2dentry(mds, rec->ur_fid1, NULL);
119 if (IS_ERR(de) || OBD_FAIL_CHECK(OBD_FAIL_MDS_REINT_CREATE)) {
121 GOTO(out_create_de, rc = -ESTALE);
124 CDEBUG(D_INODE, "ino %ld\n", dir->i_ino);
127 dchild = lookup_one_len(rec->ur_name, de, rec->ur_namelen - 1);
128 if (IS_ERR(dchild)) {
129 CERROR("child lookup error %ld\n", PTR_ERR(dchild));
132 GOTO(out_create_dchild, rc = -ESTALE);
135 if (dchild->d_inode) {
136 CERROR("child exists (dir %ld, name %s)\n",
137 dir->i_ino, rec->ur_name);
139 GOTO(out_create_dchild, rc = -EEXIST);
142 OBD_FAIL_WRITE(OBD_FAIL_MDS_REINT_CREATE_WRITE, dir->i_sb->s_dev);
146 handle = mds_fs_start(mds, dir, MDS_FSOP_CREATE);
148 GOTO(out_create_dchild, PTR_ERR(handle));
149 rc = vfs_create(dir, dchild, rec->ur_mode);
154 handle = mds_fs_start(mds, dir, MDS_FSOP_MKDIR);
156 GOTO(out_create_dchild, PTR_ERR(handle));
157 rc = vfs_mkdir(dir, dchild, rec->ur_mode);
162 handle = mds_fs_start(mds, dir, MDS_FSOP_SYMLINK);
164 GOTO(out_create_dchild, PTR_ERR(handle));
165 rc = vfs_symlink(dir, dchild, rec->ur_tgt);
173 int rdev = rec->ur_id;
174 handle = mds_fs_start(mds, dir, MDS_FSOP_MKNOD);
176 GOTO(out_create_dchild, PTR_ERR(handle));
177 rc = vfs_mknod(dir, dchild, rec->ur_mode, rdev);
182 CERROR("bad file type %d for create of %s\n",type,rec->ur_name);
183 GOTO(out_create_dchild, rc = -EINVAL);
187 CERROR("error during create: %d\n", rc);
189 GOTO(out_create_commit, rc);
192 struct inode *inode = dchild->d_inode;
193 struct mds_body *body;
195 if (type == S_IFREG) {
196 rc = mds_fs_set_objid(mds, inode, handle, rec->ur_id);
198 CERROR("error %d setting objid for %ld\n",
202 iattr.ia_atime = rec->ur_time;
203 iattr.ia_ctime = rec->ur_time;
204 iattr.ia_mtime = rec->ur_time;
205 iattr.ia_uid = rec->ur_uid;
206 iattr.ia_gid = rec->ur_gid;
207 iattr.ia_valid = ATTR_UID | ATTR_GID | ATTR_ATIME |
208 ATTR_MTIME | ATTR_CTIME;
210 rc = mds_fs_setattr(mds, dchild, handle, &iattr);
211 /* XXX should we abort here in case of error? */
213 body = lustre_msg_buf(req->rq_repmsg, 0);
214 body->ino = inode->i_ino;
215 body->generation = inode->i_generation;
219 rc = mds_update_last_rcvd(mds, req);
222 /* FIXME: keep rc intact */
223 rc = mds_fs_commit(mds, dir, handle);
233 static int mds_reint_unlink(struct mds_update_record *rec,
234 struct ptlrpc_request *req)
236 struct dentry *de = NULL;
237 struct dentry *dchild = NULL;
238 struct mds_obd *mds = &req->rq_obd->u.mds;
239 struct inode *dir, *inode;
244 de = mds_fid2dentry(mds, rec->ur_fid1, NULL);
245 if (IS_ERR(de) || OBD_FAIL_CHECK(OBD_FAIL_MDS_REINT_UNLINK)) {
247 GOTO(out_unlink, rc = -ESTALE);
250 CDEBUG(D_INODE, "ino %ld\n", dir->i_ino);
253 dchild = lookup_one_len(rec->ur_name, de, rec->ur_namelen - 1);
254 if (IS_ERR(dchild)) {
255 CERROR("child lookup error %ld\n", PTR_ERR(dchild));
257 GOTO(out_unlink_de, rc = -ESTALE);
260 inode = dchild->d_inode;
262 CERROR("child doesn't exist (dir %ld, name %s\n",
263 dir->i_ino, rec->ur_name);
265 GOTO(out_unlink_dchild, rc = -ESTALE);
268 if (inode->i_ino != rec->ur_fid2->id) {
269 CERROR("inode and FID ID do not match (%ld != %Ld)\n",
270 inode->i_ino, rec->ur_fid2->id);
272 GOTO(out_unlink_dchild, rc = -ESTALE);
274 if (inode->i_generation != rec->ur_fid2->generation) {
275 CERROR("inode and FID GENERATION do not match (%d != %d)\n",
276 inode->i_generation, rec->ur_fid2->generation);
278 GOTO(out_unlink_dchild, rc = -ESTALE);
281 OBD_FAIL_WRITE(OBD_FAIL_MDS_REINT_UNLINK_WRITE, dir->i_sb->s_dev);
283 switch (dchild->d_inode->i_mode & S_IFMT) {
285 handle = mds_fs_start(mds, dir, MDS_FSOP_RMDIR);
287 GOTO(out_unlink_dchild, rc = PTR_ERR(handle));
288 rc = vfs_rmdir(dir, dchild);
291 handle = mds_fs_start(mds, dir, MDS_FSOP_UNLINK);
293 GOTO(out_unlink_dchild, rc = PTR_ERR(handle));
294 rc = vfs_unlink(dir, dchild);
299 rc = mds_update_last_rcvd(mds, req);
300 /* FIXME: keep rc intact */
301 rc = mds_fs_commit(mds, dir, handle);
314 static int mds_reint_link(struct mds_update_record *rec,
315 struct ptlrpc_request *req)
317 struct dentry *de_src = NULL;
318 struct dentry *de_tgt_dir = NULL;
319 struct dentry *dchild = NULL;
320 struct mds_obd *mds = &req->rq_obd->u.mds;
325 de_src = mds_fid2dentry(mds, rec->ur_fid1, NULL);
326 if (IS_ERR(de_src) || OBD_FAIL_CHECK(OBD_FAIL_MDS_REINT_LINK)) {
327 GOTO(out_link, rc = -ESTALE);
330 de_tgt_dir = mds_fid2dentry(mds, rec->ur_fid2, NULL);
331 if (IS_ERR(de_tgt_dir)) {
332 GOTO(out_link_de_src, rc = -ESTALE);
335 down(&de_tgt_dir->d_inode->i_sem);
336 dchild = lookup_one_len(rec->ur_name, de_tgt_dir, rec->ur_namelen - 1);
337 if (IS_ERR(dchild)) {
338 CERROR("child lookup error %ld\n", PTR_ERR(dchild));
339 GOTO(out_link_de_tgt_dir, rc = -ESTALE);
342 if (dchild->d_inode) {
343 CERROR("child exists (dir %ld, name %s\n",
344 de_tgt_dir->d_inode->i_ino, rec->ur_name);
345 GOTO(out_link_dchild, rc = -EEXIST);
348 OBD_FAIL_WRITE(OBD_FAIL_MDS_REINT_LINK_WRITE,
349 dchild->d_inode->i_sb->s_dev);
351 handle = mds_fs_start(mds, de_tgt_dir->d_inode, MDS_FSOP_LINK);
353 GOTO(out_link_dchild, rc = PTR_ERR(handle));
355 rc = vfs_link(de_src, de_tgt_dir->d_inode, dchild);
358 rc = mds_update_last_rcvd(mds, req);
360 /* FIXME: keep rc intact */
361 rc = mds_fs_commit(mds, de_tgt_dir->d_inode, handle);
367 up(&de_tgt_dir->d_inode->i_sem);
376 static int mds_reint_rename(struct mds_update_record *rec,
377 struct ptlrpc_request *req)
379 struct dentry *de_srcdir = NULL;
380 struct dentry *de_tgtdir = NULL;
381 struct dentry *de_old = NULL;
382 struct dentry *de_new = NULL;
383 struct mds_obd *mds = &req->rq_obd->u.mds;
388 de_srcdir = mds_fid2dentry(mds, rec->ur_fid1, NULL);
389 if (IS_ERR(de_srcdir)) {
390 GOTO(out_rename, rc = -ESTALE);
393 de_tgtdir = mds_fid2dentry(mds, rec->ur_fid2, NULL);
394 if (IS_ERR(de_tgtdir)) {
395 GOTO(out_rename_srcdir, rc = -ESTALE);
398 de_old = lookup_one_len(rec->ur_name, de_srcdir, rec->ur_namelen - 1);
399 if (IS_ERR(de_old)) {
400 CERROR("old child lookup error %ld\n", PTR_ERR(de_old));
401 GOTO(out_rename_tgtdir, rc = -ESTALE);
404 de_new = lookup_one_len(rec->ur_tgt, de_tgtdir, rec->ur_tgtlen - 1);
405 if (IS_ERR(de_new)) {
406 CERROR("new child lookup error %ld\n", PTR_ERR(de_new));
407 GOTO(out_rename_deold, rc = -ESTALE);
410 OBD_FAIL_WRITE(OBD_FAIL_MDS_REINT_RENAME_WRITE,
411 de_srcdir->d_inode->i_sb->s_dev);
413 handle = mds_fs_start(mds, de_tgtdir->d_inode, MDS_FSOP_RENAME);
415 GOTO(out_rename_denew, rc = PTR_ERR(handle));
416 rc = vfs_rename(de_srcdir->d_inode, de_old, de_tgtdir->d_inode, de_new);
419 rc = mds_update_last_rcvd(mds, req);
421 /* FIXME: keep rc intact */
422 rc = mds_fs_commit(mds, de_tgtdir->d_inode, handle);
438 typedef int (*mds_reinter)(struct mds_update_record *, struct ptlrpc_request*);
440 static mds_reinter reinters[REINT_MAX+1] = {
441 [REINT_SETATTR] mds_reint_setattr,
442 [REINT_CREATE] mds_reint_create,
443 [REINT_UNLINK] mds_reint_unlink,
444 [REINT_LINK] mds_reint_link,
445 [REINT_RENAME] mds_reint_rename,
448 int mds_reint_rec(struct mds_update_record *rec, struct ptlrpc_request *req)
450 int rc, size = sizeof(struct mds_body);
452 if (rec->ur_opcode < 1 || rec->ur_opcode > REINT_MAX) {
453 CERROR("opcode %d not valid\n", rec->ur_opcode);
454 rc = req->rq_status = -EINVAL;
458 rc = lustre_pack_msg(1, &size, NULL, &req->rq_replen, &req->rq_repmsg);
460 CERROR("mds: out of memory\n");
461 rc = req->rq_status = -ENOMEM;
465 rc = reinters[rec->ur_opcode](rec, req);