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)
44 static int mds_reint_setattr(struct mds_update_record *rec,
45 struct ptlrpc_request *req)
47 struct mds_obd *mds = &req->rq_obd->u.mds;
52 de = mds_fid2dentry(mds, rec->ur_fid1, NULL);
53 if (IS_ERR(de) || OBD_FAIL_CHECK(OBD_FAIL_MDS_REINT_SETATTR)) {
54 GOTO(out_setattr, rc = -ESTALE);
57 CDEBUG(D_INODE, "ino %ld\n", de->d_inode->i_ino);
59 OBD_FAIL_WRITE(OBD_FAIL_MDS_REINT_SETATTR_WRITE,
60 de->d_inode->i_sb->s_dev);
62 handle = mds_fs_start(mds, de->d_inode, MDS_FSOP_SETATTR);
64 GOTO(out_setattr_de, rc = PTR_ERR(handle));
65 rc = mds_fs_setattr(mds, de, handle, &rec->ur_iattr);
68 rc = mds_update_last_rcvd(mds, req);
72 /* FIXME: keep rc intact */
73 rc = mds_fs_commit(mds, de->d_inode, handle);
81 static int mds_reint_create(struct mds_update_record *rec,
82 struct ptlrpc_request *req)
84 struct dentry *de = NULL;
85 struct mds_obd *mds = &req->rq_obd->u.mds;
86 struct dentry *dchild = NULL;
89 int rc = 0, type = rec->ur_mode & S_IFMT;
92 de = mds_fid2dentry(mds, rec->ur_fid1, NULL);
93 if (IS_ERR(de) || OBD_FAIL_CHECK(OBD_FAIL_MDS_REINT_CREATE)) {
95 GOTO(out_create_de, rc = -ESTALE);
98 CDEBUG(D_INODE, "ino %ld\n", dir->i_ino);
101 dchild = lookup_one_len(rec->ur_name, de, rec->ur_namelen - 1);
102 if (IS_ERR(dchild)) {
103 CERROR("child lookup error %ld\n", PTR_ERR(dchild));
106 GOTO(out_create_dchild, rc = -ESTALE);
109 if (dchild->d_inode) {
110 CERROR("child exists (dir %ld, name %s)\n",
111 dir->i_ino, rec->ur_name);
113 GOTO(out_create_dchild, rc = -EEXIST);
116 OBD_FAIL_WRITE(OBD_FAIL_MDS_REINT_CREATE_WRITE, dir->i_sb->s_dev);
120 handle = mds_fs_start(mds, dir, MDS_FSOP_CREATE);
122 GOTO(out_create_dchild, PTR_ERR(handle));
123 rc = vfs_create(dir, dchild, rec->ur_mode);
128 handle = mds_fs_start(mds, dir, MDS_FSOP_MKDIR);
130 GOTO(out_create_dchild, PTR_ERR(handle));
131 rc = vfs_mkdir(dir, dchild, rec->ur_mode);
136 handle = mds_fs_start(mds, dir, MDS_FSOP_SYMLINK);
138 GOTO(out_create_dchild, PTR_ERR(handle));
139 rc = vfs_symlink(dir, dchild, rec->ur_tgt);
147 int rdev = rec->ur_id;
148 handle = mds_fs_start(mds, dir, MDS_FSOP_MKNOD);
150 GOTO(out_create_dchild, PTR_ERR(handle));
151 rc = vfs_mknod(dir, dchild, rec->ur_mode, rdev);
156 CERROR("bad file type %d for create of %s\n",type,rec->ur_name);
157 GOTO(out_create_dchild, rc = -EINVAL);
161 CERROR("error during create: %d\n", rc);
163 GOTO(out_create_commit, rc);
166 struct inode *inode = dchild->d_inode;
167 struct mds_body *body;
169 if (type == S_IFREG) {
170 rc = mds_fs_set_objid(mds, inode, handle, rec->ur_id);
172 CERROR("error %d setting objid for %ld\n",
176 iattr.ia_atime = rec->ur_time;
177 iattr.ia_ctime = rec->ur_time;
178 iattr.ia_mtime = rec->ur_time;
179 iattr.ia_uid = rec->ur_uid;
180 iattr.ia_gid = rec->ur_gid;
181 iattr.ia_valid = ATTR_UID | ATTR_GID | ATTR_ATIME |
182 ATTR_MTIME | ATTR_CTIME;
184 rc = mds_fs_setattr(mds, dchild, handle, &iattr);
185 /* XXX should we abort here in case of error? */
187 body = lustre_msg_buf(req->rq_repmsg, 0);
188 body->ino = inode->i_ino;
189 body->generation = inode->i_generation;
193 rc = mds_update_last_rcvd(mds, req);
196 /* FIXME: keep rc intact */
197 rc = mds_fs_commit(mds, dir, handle);
207 static int mds_reint_unlink(struct mds_update_record *rec,
208 struct ptlrpc_request *req)
210 struct dentry *de = NULL;
211 struct dentry *dchild = NULL;
212 struct mds_obd *mds = &req->rq_obd->u.mds;
213 struct inode *dir, *inode;
218 de = mds_fid2dentry(mds, rec->ur_fid1, NULL);
219 if (IS_ERR(de) || OBD_FAIL_CHECK(OBD_FAIL_MDS_REINT_UNLINK)) {
221 GOTO(out_unlink, rc = -ESTALE);
224 CDEBUG(D_INODE, "ino %ld\n", dir->i_ino);
227 dchild = lookup_one_len(rec->ur_name, de, rec->ur_namelen - 1);
228 if (IS_ERR(dchild)) {
229 CERROR("child lookup error %ld\n", PTR_ERR(dchild));
231 GOTO(out_unlink_de, rc = -ESTALE);
234 inode = dchild->d_inode;
236 CERROR("child doesn't exist (dir %ld, name %s\n",
237 dir->i_ino, rec->ur_name);
239 GOTO(out_unlink_dchild, rc = -ESTALE);
242 if (inode->i_ino != rec->ur_fid2->id) {
243 CERROR("inode and FID ID do not match (%ld != %Ld)\n",
244 inode->i_ino, rec->ur_fid2->id);
246 GOTO(out_unlink_dchild, rc = -ESTALE);
248 if (inode->i_generation != rec->ur_fid2->generation) {
249 CERROR("inode and FID GENERATION do not match (%d != %d)\n",
250 inode->i_generation, rec->ur_fid2->generation);
252 GOTO(out_unlink_dchild, rc = -ESTALE);
255 OBD_FAIL_WRITE(OBD_FAIL_MDS_REINT_UNLINK_WRITE, dir->i_sb->s_dev);
257 switch (dchild->d_inode->i_mode & S_IFMT) {
259 handle = mds_fs_start(mds, dir, MDS_FSOP_RMDIR);
261 GOTO(out_unlink_dchild, rc = PTR_ERR(handle));
262 rc = vfs_rmdir(dir, dchild);
265 handle = mds_fs_start(mds, dir, MDS_FSOP_UNLINK);
267 GOTO(out_unlink_dchild, rc = PTR_ERR(handle));
268 rc = vfs_unlink(dir, dchild);
273 rc = mds_update_last_rcvd(mds, req);
274 /* FIXME: keep rc intact */
275 rc = mds_fs_commit(mds, dir, handle);
288 static int mds_reint_link(struct mds_update_record *rec,
289 struct ptlrpc_request *req)
291 struct dentry *de_src = NULL;
292 struct dentry *de_tgt_dir = NULL;
293 struct dentry *dchild = NULL;
294 struct mds_obd *mds = &req->rq_obd->u.mds;
299 de_src = mds_fid2dentry(mds, rec->ur_fid1, NULL);
300 if (IS_ERR(de_src) || OBD_FAIL_CHECK(OBD_FAIL_MDS_REINT_LINK)) {
301 GOTO(out_link, rc = -ESTALE);
304 de_tgt_dir = mds_fid2dentry(mds, rec->ur_fid2, NULL);
305 if (IS_ERR(de_tgt_dir)) {
306 GOTO(out_link_de_src, rc = -ESTALE);
309 down(&de_tgt_dir->d_inode->i_sem);
310 dchild = lookup_one_len(rec->ur_name, de_tgt_dir, rec->ur_namelen - 1);
311 if (IS_ERR(dchild)) {
312 CERROR("child lookup error %ld\n", PTR_ERR(dchild));
313 GOTO(out_link_de_tgt_dir, rc = -ESTALE);
316 if (dchild->d_inode) {
317 CERROR("child exists (dir %ld, name %s\n",
318 de_tgt_dir->d_inode->i_ino, rec->ur_name);
319 GOTO(out_link_dchild, rc = -EEXIST);
322 OBD_FAIL_WRITE(OBD_FAIL_MDS_REINT_LINK_WRITE,
323 dchild->d_inode->i_sb->s_dev);
325 handle = mds_fs_start(mds, de_tgt_dir->d_inode, MDS_FSOP_LINK);
327 GOTO(out_link_dchild, rc = PTR_ERR(handle));
329 rc = vfs_link(de_src, de_tgt_dir->d_inode, dchild);
332 rc = mds_update_last_rcvd(mds, req);
334 /* FIXME: keep rc intact */
335 rc = mds_fs_commit(mds, de_tgt_dir->d_inode, handle);
341 up(&de_tgt_dir->d_inode->i_sem);
350 static int mds_reint_rename(struct mds_update_record *rec,
351 struct ptlrpc_request *req)
353 struct dentry *de_srcdir = NULL;
354 struct dentry *de_tgtdir = NULL;
355 struct dentry *de_old = NULL;
356 struct dentry *de_new = NULL;
357 struct mds_obd *mds = &req->rq_obd->u.mds;
362 de_srcdir = mds_fid2dentry(mds, rec->ur_fid1, NULL);
363 if (IS_ERR(de_srcdir)) {
364 GOTO(out_rename, rc = -ESTALE);
367 de_tgtdir = mds_fid2dentry(mds, rec->ur_fid2, NULL);
368 if (IS_ERR(de_tgtdir)) {
369 GOTO(out_rename_srcdir, rc = -ESTALE);
372 de_old = lookup_one_len(rec->ur_name, de_srcdir, rec->ur_namelen - 1);
373 if (IS_ERR(de_old)) {
374 CERROR("old child lookup error %ld\n", PTR_ERR(de_old));
375 GOTO(out_rename_tgtdir, rc = -ESTALE);
378 de_new = lookup_one_len(rec->ur_tgt, de_tgtdir, rec->ur_tgtlen - 1);
379 if (IS_ERR(de_new)) {
380 CERROR("new child lookup error %ld\n", PTR_ERR(de_new));
381 GOTO(out_rename_deold, rc = -ESTALE);
384 OBD_FAIL_WRITE(OBD_FAIL_MDS_REINT_RENAME_WRITE,
385 de_srcdir->d_inode->i_sb->s_dev);
387 handle = mds_fs_start(mds, de_tgtdir->d_inode, MDS_FSOP_RENAME);
389 GOTO(out_rename_denew, rc = PTR_ERR(handle));
390 rc = vfs_rename(de_srcdir->d_inode, de_old, de_tgtdir->d_inode, de_new);
393 rc = mds_update_last_rcvd(mds, req);
395 /* FIXME: keep rc intact */
396 rc = mds_fs_commit(mds, de_tgtdir->d_inode, handle);
412 typedef int (*mds_reinter)(struct mds_update_record *, struct ptlrpc_request*);
414 static mds_reinter reinters[REINT_MAX+1] = {
415 [REINT_SETATTR] mds_reint_setattr,
416 [REINT_CREATE] mds_reint_create,
417 [REINT_UNLINK] mds_reint_unlink,
418 [REINT_LINK] mds_reint_link,
419 [REINT_RENAME] mds_reint_rename,
422 int mds_reint_rec(struct mds_update_record *rec, struct ptlrpc_request *req)
424 int rc, size = sizeof(struct mds_body);
426 if (rec->ur_opcode < 1 || rec->ur_opcode > REINT_MAX) {
427 CERROR("opcode %d not valid\n", rec->ur_opcode);
428 rc = req->rq_status = -EINVAL;
432 rc = lustre_pack_msg(1, &size, NULL, &req->rq_replen, &req->rq_repmsg);
434 CERROR("mds: out of memory\n");
435 rc = req->rq_status = -ENOMEM;
439 rc = reinters[rec->ur_opcode](rec, req);