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 rc = lustre_fwrite(mds->mds_rcvd_filp, (char *)mci->mci_mcd,
56 sizeof(*mci->mci_mcd), &off);
57 CDEBUG(D_INODE, "wrote trans #%Ld for client '%s' at %Ld: rc = %d\n",
58 mds->mds_last_rcvd, mci->mci_mcd->mcd_uuid, mci->mci_off, rc);
59 // store new value and last committed value in req struct
61 if (rc == sizeof(mci->mci_mcd))
69 static int mds_reint_setattr(struct mds_update_record *rec,
70 struct ptlrpc_request *req)
72 struct mds_obd *mds = &req->rq_obd->u.mds;
77 de = mds_fid2dentry(mds, rec->ur_fid1, NULL);
78 if (IS_ERR(de) || OBD_FAIL_CHECK(OBD_FAIL_MDS_REINT_SETATTR)) {
79 GOTO(out_setattr, rc = -ESTALE);
82 CDEBUG(D_INODE, "ino %ld\n", de->d_inode->i_ino);
84 OBD_FAIL_WRITE(OBD_FAIL_MDS_REINT_SETATTR_WRITE,
85 de->d_inode->i_sb->s_dev);
87 handle = mds_fs_start(mds, de->d_inode, MDS_FSOP_SETATTR);
89 GOTO(out_setattr_de, rc = PTR_ERR(handle));
90 rc = mds_fs_setattr(mds, de, handle, &rec->ur_iattr);
93 rc = mds_update_last_rcvd(mds, req);
97 /* FIXME: keep rc intact */
98 rc = mds_fs_commit(mds, de->d_inode, handle);
106 static int mds_reint_create(struct mds_update_record *rec,
107 struct ptlrpc_request *req)
109 struct dentry *de = NULL;
110 struct mds_obd *mds = &req->rq_obd->u.mds;
111 struct dentry *dchild = NULL;
114 int rc = 0, type = rec->ur_mode & S_IFMT;
117 de = mds_fid2dentry(mds, rec->ur_fid1, NULL);
118 if (IS_ERR(de) || OBD_FAIL_CHECK(OBD_FAIL_MDS_REINT_CREATE)) {
120 GOTO(out_create_de, rc = -ESTALE);
123 CDEBUG(D_INODE, "ino %ld\n", dir->i_ino);
126 dchild = lookup_one_len(rec->ur_name, de, rec->ur_namelen - 1);
127 if (IS_ERR(dchild)) {
128 CERROR("child lookup error %ld\n", PTR_ERR(dchild));
131 GOTO(out_create_dchild, rc = -ESTALE);
134 if (dchild->d_inode) {
135 CERROR("child exists (dir %ld, name %s)\n",
136 dir->i_ino, rec->ur_name);
138 GOTO(out_create_dchild, rc = -EEXIST);
141 OBD_FAIL_WRITE(OBD_FAIL_MDS_REINT_CREATE_WRITE, dir->i_sb->s_dev);
145 handle = mds_fs_start(mds, dir, MDS_FSOP_CREATE);
147 GOTO(out_create_dchild, PTR_ERR(handle));
148 rc = vfs_create(dir, dchild, rec->ur_mode);
153 handle = mds_fs_start(mds, dir, MDS_FSOP_MKDIR);
155 GOTO(out_create_dchild, PTR_ERR(handle));
156 rc = vfs_mkdir(dir, dchild, rec->ur_mode);
161 handle = mds_fs_start(mds, dir, MDS_FSOP_SYMLINK);
163 GOTO(out_create_dchild, PTR_ERR(handle));
164 rc = vfs_symlink(dir, dchild, rec->ur_tgt);
172 int rdev = rec->ur_id;
173 handle = mds_fs_start(mds, dir, MDS_FSOP_MKNOD);
175 GOTO(out_create_dchild, PTR_ERR(handle));
176 rc = vfs_mknod(dir, dchild, rec->ur_mode, rdev);
181 CERROR("bad file type %d for create of %s\n",type,rec->ur_name);
182 GOTO(out_create_dchild, rc = -EINVAL);
186 CERROR("error during create: %d\n", rc);
188 GOTO(out_create_commit, rc);
191 struct inode *inode = dchild->d_inode;
192 struct mds_body *body;
194 if (type == S_IFREG) {
195 rc = mds_fs_set_objid(mds, inode, handle, rec->ur_id);
197 CERROR("error %d setting objid for %ld\n",
201 iattr.ia_atime = rec->ur_time;
202 iattr.ia_ctime = rec->ur_time;
203 iattr.ia_mtime = rec->ur_time;
204 iattr.ia_uid = rec->ur_uid;
205 iattr.ia_gid = rec->ur_gid;
206 iattr.ia_valid = ATTR_UID | ATTR_GID | ATTR_ATIME |
207 ATTR_MTIME | ATTR_CTIME;
209 rc = mds_fs_setattr(mds, dchild, handle, &iattr);
210 /* XXX should we abort here in case of error? */
212 body = lustre_msg_buf(req->rq_repmsg, 0);
213 body->ino = inode->i_ino;
214 body->generation = inode->i_generation;
218 rc = mds_update_last_rcvd(mds, req);
221 /* FIXME: keep rc intact */
222 rc = mds_fs_commit(mds, dir, handle);
232 static int mds_reint_unlink(struct mds_update_record *rec,
233 struct ptlrpc_request *req)
235 struct dentry *de = NULL;
236 struct dentry *dchild = NULL;
237 struct mds_obd *mds = &req->rq_obd->u.mds;
238 struct inode *dir, *inode;
243 de = mds_fid2dentry(mds, rec->ur_fid1, NULL);
244 if (IS_ERR(de) || OBD_FAIL_CHECK(OBD_FAIL_MDS_REINT_UNLINK)) {
246 GOTO(out_unlink, rc = -ESTALE);
249 CDEBUG(D_INODE, "ino %ld\n", dir->i_ino);
252 dchild = lookup_one_len(rec->ur_name, de, rec->ur_namelen - 1);
253 if (IS_ERR(dchild)) {
254 CERROR("child lookup error %ld\n", PTR_ERR(dchild));
256 GOTO(out_unlink_de, rc = -ESTALE);
259 inode = dchild->d_inode;
261 CERROR("child doesn't exist (dir %ld, name %s\n",
262 dir->i_ino, rec->ur_name);
264 GOTO(out_unlink_dchild, rc = -ESTALE);
267 if (inode->i_ino != rec->ur_fid2->id) {
268 CERROR("inode and FID ID do not match (%ld != %Ld)\n",
269 inode->i_ino, rec->ur_fid2->id);
271 GOTO(out_unlink_dchild, rc = -ESTALE);
273 if (inode->i_generation != rec->ur_fid2->generation) {
274 CERROR("inode and FID GENERATION do not match (%d != %d)\n",
275 inode->i_generation, rec->ur_fid2->generation);
277 GOTO(out_unlink_dchild, rc = -ESTALE);
280 OBD_FAIL_WRITE(OBD_FAIL_MDS_REINT_UNLINK_WRITE, dir->i_sb->s_dev);
282 switch (dchild->d_inode->i_mode & S_IFMT) {
284 handle = mds_fs_start(mds, dir, MDS_FSOP_RMDIR);
286 GOTO(out_unlink_dchild, rc = PTR_ERR(handle));
287 rc = vfs_rmdir(dir, dchild);
290 handle = mds_fs_start(mds, dir, MDS_FSOP_UNLINK);
292 GOTO(out_unlink_dchild, rc = PTR_ERR(handle));
293 rc = vfs_unlink(dir, dchild);
298 rc = mds_update_last_rcvd(mds, req);
299 /* FIXME: keep rc intact */
300 rc = mds_fs_commit(mds, dir, handle);
313 static int mds_reint_link(struct mds_update_record *rec,
314 struct ptlrpc_request *req)
316 struct dentry *de_src = NULL;
317 struct dentry *de_tgt_dir = NULL;
318 struct dentry *dchild = NULL;
319 struct mds_obd *mds = &req->rq_obd->u.mds;
324 de_src = mds_fid2dentry(mds, rec->ur_fid1, NULL);
325 if (IS_ERR(de_src) || OBD_FAIL_CHECK(OBD_FAIL_MDS_REINT_LINK)) {
326 GOTO(out_link, rc = -ESTALE);
329 de_tgt_dir = mds_fid2dentry(mds, rec->ur_fid2, NULL);
330 if (IS_ERR(de_tgt_dir)) {
331 GOTO(out_link_de_src, rc = -ESTALE);
334 down(&de_tgt_dir->d_inode->i_sem);
335 dchild = lookup_one_len(rec->ur_name, de_tgt_dir, rec->ur_namelen - 1);
336 if (IS_ERR(dchild)) {
337 CERROR("child lookup error %ld\n", PTR_ERR(dchild));
338 GOTO(out_link_de_tgt_dir, rc = -ESTALE);
341 if (dchild->d_inode) {
342 CERROR("child exists (dir %ld, name %s\n",
343 de_tgt_dir->d_inode->i_ino, rec->ur_name);
344 GOTO(out_link_dchild, rc = -EEXIST);
347 OBD_FAIL_WRITE(OBD_FAIL_MDS_REINT_LINK_WRITE,
348 dchild->d_inode->i_sb->s_dev);
350 handle = mds_fs_start(mds, de_tgt_dir->d_inode, MDS_FSOP_LINK);
352 GOTO(out_link_dchild, rc = PTR_ERR(handle));
354 rc = vfs_link(de_src, de_tgt_dir->d_inode, dchild);
357 rc = mds_update_last_rcvd(mds, req);
359 /* FIXME: keep rc intact */
360 rc = mds_fs_commit(mds, de_tgt_dir->d_inode, handle);
366 up(&de_tgt_dir->d_inode->i_sem);
375 static int mds_reint_rename(struct mds_update_record *rec,
376 struct ptlrpc_request *req)
378 struct dentry *de_srcdir = NULL;
379 struct dentry *de_tgtdir = NULL;
380 struct dentry *de_old = NULL;
381 struct dentry *de_new = NULL;
382 struct mds_obd *mds = &req->rq_obd->u.mds;
387 de_srcdir = mds_fid2dentry(mds, rec->ur_fid1, NULL);
388 if (IS_ERR(de_srcdir)) {
389 GOTO(out_rename, rc = -ESTALE);
392 de_tgtdir = mds_fid2dentry(mds, rec->ur_fid2, NULL);
393 if (IS_ERR(de_tgtdir)) {
394 GOTO(out_rename_srcdir, rc = -ESTALE);
397 de_old = lookup_one_len(rec->ur_name, de_srcdir, rec->ur_namelen - 1);
398 if (IS_ERR(de_old)) {
399 CERROR("old child lookup error %ld\n", PTR_ERR(de_old));
400 GOTO(out_rename_tgtdir, rc = -ESTALE);
403 de_new = lookup_one_len(rec->ur_tgt, de_tgtdir, rec->ur_tgtlen - 1);
404 if (IS_ERR(de_new)) {
405 CERROR("new child lookup error %ld\n", PTR_ERR(de_new));
406 GOTO(out_rename_deold, rc = -ESTALE);
409 OBD_FAIL_WRITE(OBD_FAIL_MDS_REINT_RENAME_WRITE,
410 de_srcdir->d_inode->i_sb->s_dev);
412 handle = mds_fs_start(mds, de_tgtdir->d_inode, MDS_FSOP_RENAME);
414 GOTO(out_rename_denew, rc = PTR_ERR(handle));
415 rc = vfs_rename(de_srcdir->d_inode, de_old, de_tgtdir->d_inode, de_new);
418 rc = mds_update_last_rcvd(mds, req);
420 /* FIXME: keep rc intact */
421 rc = mds_fs_commit(mds, de_tgtdir->d_inode, handle);
437 typedef int (*mds_reinter)(struct mds_update_record *, struct ptlrpc_request*);
439 static mds_reinter reinters[REINT_MAX+1] = {
440 [REINT_SETATTR] mds_reint_setattr,
441 [REINT_CREATE] mds_reint_create,
442 [REINT_UNLINK] mds_reint_unlink,
443 [REINT_LINK] mds_reint_link,
444 [REINT_RENAME] mds_reint_rename,
447 int mds_reint_rec(struct mds_update_record *rec, struct ptlrpc_request *req)
449 int rc, size = sizeof(struct mds_body);
451 if (rec->ur_opcode < 1 || rec->ur_opcode > REINT_MAX) {
452 CERROR("opcode %d not valid\n", rec->ur_opcode);
453 rc = req->rq_status = -EINVAL;
457 rc = lustre_pack_msg(1, &size, NULL, &req->rq_replen, &req->rq_repmsg);
459 CERROR("mds: out of memory\n");
460 rc = req->rq_status = -ENOMEM;
464 rc = reinters[rec->ur_opcode](rec, req);