1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
4 * This code is issued under the GNU General Public License.
5 * See the file COPYING in this distribution
7 * Copyright (C) 1992, 1993, 1994, 1995
8 * Remy Card (card@masi.ibp.fr)
9 * Laboratoire MASI - Institut Blaise Pascal
10 * Universite Pierre et Marie Curie (Paris VI)
14 * linux/fs/ext2/namei.c
16 * Copyright (C) 1991, 1992 Linus Torvalds
18 * Big-endian to little-endian byte-swapping/bitmaps by
19 * David S. Miller (davem@caip.rutgers.edu), 1995
20 * Directory entry file type support and forward compatibility hooks
21 * for B-tree directories by Theodore Ts'o (tytso@mit.edu), 1998
23 * Changes for use in OBDFS
24 * Copyright (c) 1999, Seagate Technology Inc.
25 * Copyright (C) 2001, Cluster File Systems, Inc.
26 * Rewritten based on recent ext2 page cache use.
31 #include <linux/locks.h>
32 #include <linux/quotaops.h>
34 #define DEBUG_SUBSYSTEM S_LLITE
36 #include <linux/obd_support.h>
37 #include <linux/lustre_lite.h>
38 #include <linux/lustre_dlm.h>
39 #include <linux/obd_lov.h>
41 extern struct address_space_operations ll_aops;
44 extern void ll_change_inode(struct inode *inode);
45 extern int ll_setattr(struct dentry *de, struct iattr *attr);
48 extern int ll_add_link (struct dentry *dentry, struct inode *inode);
49 obd_id ll_inode_by_name(struct inode * dir, struct dentry *dentry, int *typ);
50 int ext2_make_empty(struct inode *inode, struct inode *parent);
51 struct ext2_dir_entry_2 * ext2_find_entry (struct inode * dir,
52 struct dentry *dentry, struct page ** res_page);
53 int ext2_delete_entry (struct ext2_dir_entry_2 * dir, struct page * page );
54 int ext2_empty_dir (struct inode * inode);
55 struct ext2_dir_entry_2 * ext2_dotdot (struct inode *dir, struct page **p);
56 void ext2_set_link(struct inode *dir, struct ext2_dir_entry_2 *de,
57 struct page *page, struct inode *inode);
60 * Couple of helper functions - make the code slightly cleaner.
62 static inline void ext2_inc_count(struct inode *inode)
67 /* postpone the disk update until the inode really goes away */
68 static inline void ext2_dec_count(struct inode *inode)
73 static inline int ext2_add_nondir(struct dentry *dentry, struct inode *inode)
76 err = ll_add_link(dentry, inode);
78 d_instantiate(dentry, inode);
81 ext2_dec_count(inode);
87 static int ll_find_inode(struct inode *inode, unsigned long ino, void *opaque)
89 struct ll_read_inode2_cookie *lic = opaque;
90 struct mds_body *body = lic->lic_body;
92 if (inode->i_generation != lic->lic_body->generation)
95 /* Apply the attributes in 'opaque' to this inode */
96 ll_update_inode(inode, body);
101 extern struct dentry_operations ll_d_ops;
103 int ll_lock(struct inode *dir, struct dentry *dentry,
104 struct lookup_intent *it, struct lustre_handle *lockh)
106 struct ll_sb_info *sbi = ll_i2sbi(dir);
111 if ((it->it_op & (IT_CREAT | IT_MKDIR | IT_SETATTR | IT_MKNOD)))
113 else if (it->it_op & (IT_READDIR | IT_GETATTR | IT_OPEN | IT_UNLINK |
114 IT_RMDIR | IT_RENAME | IT_RENAME2 | IT_READLINK|
117 else if (it->it_op & IT_SYMLINK) {
120 tgtlen = strlen(tgt);
122 } else if (it->it_op & IT_LOOKUP)
129 err = mdc_enqueue(&sbi->ll_mdc_conn, LDLM_MDSINTENT, it, lock_mode,
130 dir, dentry, lockh, tgt, tgtlen, dir, sizeof(*dir));
135 int ll_unlock(__u32 mode, struct lustre_handle *lockh)
139 ldlm_lock_decref(lockh, mode);
144 static struct dentry *ll_lookup2(struct inode *dir, struct dentry *dentry,
145 struct lookup_intent *it)
147 struct ptlrpc_request *request = NULL;
148 struct inode * inode = NULL;
149 struct ll_sb_info *sbi = ll_i2sbi(dir);
150 struct ll_read_inode2_cookie lic;
151 struct lustre_handle lockh;
152 struct lookup_intent lookup_it = { IT_LOOKUP };
158 CHECK_MOUNT_EPOCH(dir);
165 CDEBUG(D_INFO, "name: %*s, intent op: %d\n", dentry->d_name.len,
166 dentry->d_name.name, it->it_op);
168 if (dentry->d_name.len > EXT2_NAME_LEN)
169 RETURN(ERR_PTR(-ENAMETOOLONG));
171 err = ll_lock(dir, dentry, it, &lockh);
173 RETURN(ERR_PTR(err));
174 memcpy(it->it_lock_handle, &lockh, sizeof(lockh));
176 request = (struct ptlrpc_request *)it->it_data;
177 if (it->it_disposition) {
178 int mode, symlen = 0;
182 lic.lic_body = lustre_msg_buf(request->rq_repmsg, offset);
183 ino = lic.lic_body->fid1.id;
184 mode = lic.lic_body->mode;
185 if (it->it_op & (IT_CREAT | IT_MKDIR | IT_SYMLINK | IT_MKNOD)) {
186 /* For create ops, we want the lookup to be negative */
188 GOTO(negative, NULL);
189 } else if (it->it_op & (IT_GETATTR | IT_SETATTR | IT_LOOKUP)) {
190 /* For check ops, we want the lookup to succeed */
194 } else if (it->it_op & (IT_RENAME | IT_LINK)) {
195 /* For rename, we want the lookup to succeed */
200 it->it_data = dentry;
201 } else if (it->it_op & (IT_UNLINK | IT_RMDIR)) {
202 /* For remove ops, we want the lookup to succeed */
207 } else if (it->it_op == IT_OPEN) {
209 if (it->it_status && it->it_status != -EEXIST)
211 } else if (it->it_op & (IT_RENAME2|IT_LINK2)) {
212 struct mds_body *body =
213 lustre_msg_buf(request->rq_repmsg, offset);
215 /* For rename2, this means the lookup is negative */
217 if (body->valid == 0)
222 /* Do a getattr now that we have the lock */
223 valid = OBD_MD_FLNOTOBD | OBD_MD_FLEASIZE;
224 if (it->it_op == IT_READLINK) {
225 valid |= OBD_MD_LINKNAME;
226 symlen = lic.lic_body->size;
228 ptlrpc_free_req(request);
230 err = mdc_getattr(&sbi->ll_mdc_conn, ino, mode,
231 valid, symlen, &request);
233 CERROR("failure %d inode %Ld\n", err, (long long)ino);
234 GOTO(drop_req, err = -abs(err));
238 struct ll_inode_info *lli = ll_i2info(dir);
241 memcpy(&lli->lli_intent_lock_handle, &lockh, sizeof(lockh));
244 ino = ll_inode_by_name(dir, dentry, &mode);
246 CERROR("inode %*s not found by name\n",
247 dentry->d_name.len, dentry->d_name.name);
248 GOTO(drop_lock, err = -ENOENT);
251 err = mdc_getattr(&sbi->ll_mdc_conn, ino, mode,
252 OBD_MD_FLNOTOBD|OBD_MD_FLEASIZE, 0, &request);
254 CERROR("failure %d inode %Ld\n", err, (long long)ino);
255 GOTO(drop_req, err = -abs(err));
260 lic.lic_body = lustre_msg_buf(request->rq_repmsg, offset);
261 if (S_ISREG(lic.lic_body->mode) &&
262 lic.lic_body->valid & OBD_MD_FLEASIZE) {
263 LASSERT(request->rq_repmsg->bufcount > offset);
264 lic.lic_lmm = lustre_msg_buf(request->rq_repmsg, offset + 1);
268 /* No rpc's happen during iget4, -ENOMEM's are possible */
270 inode = iget4(dir->i_sb, ino, ll_find_inode, &lic);
273 ptlrpc_free_req(request);
274 ll_intent_release(dentry);
275 RETURN(ERR_PTR(-ENOMEM));
280 ptlrpc_req_finished(request);
282 dentry->d_op = &ll_d_ops;
283 d_add(dentry, inode);
284 if (it->it_op == IT_LOOKUP)
285 ll_intent_release(dentry);
290 ptlrpc_free_req(request);
292 #warning FIXME: must release lock here
296 static struct inode *ll_create_node(struct inode *dir, const char *name,
297 int namelen, const char *tgt, int tgtlen,
298 int mode, __u64 extra,
299 struct lookup_intent *it,
300 struct lov_stripe_md *lsm)
303 struct ptlrpc_request *request = NULL;
304 struct mds_body *body;
306 time_t time = CURRENT_TIME;
307 struct ll_sb_info *sbi = ll_i2sbi(dir);
308 int gid = current->fsgid;
309 struct ll_read_inode2_cookie lic;
310 struct lov_mds_md *lmm = NULL;
315 if (dir->i_mode & S_ISGID) {
321 if (!it || !it->it_disposition) {
322 rc = mdc_create(&sbi->ll_mdc_conn, dir, name, namelen, tgt,
323 tgtlen, mode, current->fsuid,
324 gid, time, extra, lsm, &request);
329 body = lustre_msg_buf(request->rq_repmsg, 0);
331 mds_md_size = ll_mds_easize(dir->i_sb);
332 OBD_ALLOC(lmm, mds_md_size);
333 lov_packmd(lmm, lsm);
339 invalidate_inode_pages(dir);
340 request = it->it_data;
341 body = lustre_msg_buf(request->rq_repmsg, 1);
345 body->valid = OBD_MD_FLNOTOBD;
348 body->atime = body->ctime = body->mtime = time;
349 body->uid = current->fsuid;
355 inode = iget4(dir->i_sb, body->ino, ll_find_inode, &lic);
358 CERROR("new_inode -fatal: rc %d\n", rc);
363 if (!list_empty(&inode->i_dentry)) {
364 CERROR("new_inode -fatal: inode %d, ct %d lnk %d\n",
365 body->ino, atomic_read(&inode->i_count),
369 inode = ERR_PTR(-EIO);
376 OBD_FREE(lmm, mds_md_size);
377 ptlrpc_free_req(request);
381 static int ll_mdc_unlink(struct inode *dir, struct inode *child, __u32 mode,
382 const char *name, int len)
384 struct ptlrpc_request *request = NULL;
385 struct ll_sb_info *sbi = ll_i2sbi(dir);
390 err = mdc_unlink(&sbi->ll_mdc_conn, dir, child, mode, name, len,
392 ptlrpc_free_req(request);
397 int ll_mdc_link(struct dentry *src, struct inode *dir,
398 const char *name, int len)
400 struct ptlrpc_request *request = NULL;
402 struct ll_sb_info *sbi = ll_i2sbi(dir);
406 err = mdc_link(&sbi->ll_mdc_conn, src, dir, name,
408 ptlrpc_free_req(request);
413 int ll_mdc_rename(struct inode *src, struct inode *tgt,
414 struct dentry *old, struct dentry *new)
416 struct ptlrpc_request *request = NULL;
417 struct ll_sb_info *sbi = ll_i2sbi(src);
422 err = mdc_rename(&sbi->ll_mdc_conn, src, tgt,
423 old->d_name.name, old->d_name.len,
424 new->d_name.name, new->d_name.len, &request);
425 ptlrpc_free_req(request);
431 * By the time this is called, we already have created
432 * the directory cache entry for the new file, but it
433 * is so far negative - it has no inode.
435 * If the create succeeds, we fill in the inode information
436 * with d_instantiate().
439 static int ll_create(struct inode * dir, struct dentry * dentry, int mode)
444 struct lov_stripe_md *lsm = NULL;
445 struct ll_inode_info *lli = NULL;
448 CHECK_MOUNT_EPOCH(dir);
450 if (dentry->d_it->it_disposition == 0) {
451 memset(&oa, 0, sizeof(oa));
452 oa.o_mode = S_IFREG | 0600;
453 oa.o_valid = OBD_MD_FLTYPE | OBD_MD_FLMODE;
454 rc = obd_create(ll_i2obdconn(dir), &oa, &lsm);
455 CDEBUG(D_DENTRY, "name %s mode %o o_id %lld: rc = %d\n",
456 dentry->d_name.name, mode, (long long)oa.o_id, rc);
461 inode = ll_create_node(dir, dentry->d_name.name, dentry->d_name.len,
462 NULL, 0, mode, 0, dentry->d_it, lsm);
466 CERROR("error creating MDS object for id %Ld: rc = %d\n",
467 (unsigned long long)oa.o_id, rc);
468 GOTO(out_destroy, rc);
471 if (dentry->d_it->it_disposition) {
472 lli = ll_i2info(inode);
473 memcpy(&lli->lli_intent_lock_handle,
474 dentry->d_it->it_lock_handle,
475 sizeof(struct lustre_handle));
476 d_instantiate(dentry, inode);
478 /* no directory data updates when intents rule */
479 rc = ext2_add_nondir(dentry, inode);
486 oa.o_easize = ll_mds_easize(inode->i_sb);
487 oa.o_valid |= OBD_MD_FLEASIZE;
488 err = obd_destroy(ll_i2obdconn(dir), &oa, lsm);
490 CERROR("error destroying objid %Ld on error: err %d\n",
491 (unsigned long long)oa.o_id, err);
497 static int ll_mknod(struct inode *dir, struct dentry *dentry, int mode,
503 inode = ll_create_node(dir, dentry->d_name.name, dentry->d_name.len,
504 NULL, 0, mode, rdev, dentry->d_it, NULL);
507 RETURN(PTR_ERR(inode));
509 /* no directory data updates when intents rule */
510 if (dentry->d_it && dentry->d_it->it_disposition)
511 d_instantiate(dentry, inode);
513 err = ext2_add_nondir(dentry, inode);
518 static int ll_symlink(struct inode *dir, struct dentry *dentry,
521 unsigned l = strlen(symname);
523 struct ll_inode_info *lli;
527 CHECK_MOUNT_EPOCH(dir);
529 inode = ll_create_node(dir, dentry->d_name.name, dentry->d_name.len,
530 symname, l, S_IFLNK | S_IRWXUGO, 0,
533 RETURN(PTR_ERR(inode));
535 lli = ll_i2info(inode);
537 OBD_ALLOC(lli->lli_symlink_name, l + 1);
538 /* this _could_ be a non-fatal error, since the symlink is already
539 * stored on the MDS by this point, and we can re-get it in readlink.
541 if (!lli->lli_symlink_name)
544 memcpy(lli->lli_symlink_name, symname, l + 1);
547 /* no directory data updates when intents rule */
548 if (dentry->d_it && dentry->d_it->it_disposition)
549 d_instantiate(dentry, inode);
551 err = ext2_add_nondir(dentry, inode);
556 static int ll_link(struct dentry *old_dentry, struct inode * dir,
557 struct dentry *dentry)
560 struct inode *inode = old_dentry->d_inode;
562 if (dentry->d_it && dentry->d_it->it_disposition) {
563 int err = dentry->d_it->it_status;
566 inode->i_ctime = CURRENT_TIME;
567 ext2_inc_count(inode);
568 atomic_inc(&inode->i_count);
569 d_instantiate(dentry, inode);
570 invalidate_inode_pages(dir);
574 if (S_ISDIR(inode->i_mode))
577 if (inode->i_nlink >= EXT2_LINK_MAX)
580 err = ll_mdc_link(old_dentry, dir,
581 dentry->d_name.name, dentry->d_name.len);
585 inode->i_ctime = CURRENT_TIME;
586 ext2_inc_count(inode);
587 atomic_inc(&inode->i_count);
589 return ext2_add_nondir(dentry, inode);
592 static int ll_mkdir(struct inode * dir, struct dentry * dentry, int mode)
594 struct inode * inode;
598 if (dir->i_nlink >= EXT2_LINK_MAX)
603 inode = ll_create_node(dir, dentry->d_name.name, dentry->d_name.len,
604 NULL, 0, S_IFDIR | mode, 0, dentry->d_it, NULL);
605 err = PTR_ERR(inode);
609 ext2_inc_count(inode);
611 err = ext2_make_empty(inode, dir);
615 /* no directory data updates when intents rule */
616 if (dentry->d_it->it_disposition == 0) {
617 err = ll_add_link(dentry, inode);
622 d_instantiate(dentry, inode);
628 ext2_dec_count(inode);
629 ext2_dec_count(inode);
638 static int ll_common_unlink(struct inode *dir, struct dentry *dentry,
641 struct inode * inode = dentry->d_inode;
643 struct ext2_dir_entry_2 * de;
648 if (dentry->d_it && dentry->d_it->it_disposition) {
649 err = dentry->d_it->it_status;
650 invalidate_inode_pages(dir);
655 de = ext2_find_entry(dir, dentry, &page);
659 err = ll_mdc_unlink(dir, dentry->d_inode, mode,
660 dentry->d_name.name, dentry->d_name.len);
665 err = ext2_delete_entry(de, page);
669 invalidate_inode_pages(dir);
671 inode->i_ctime = dir->i_ctime;
673 ext2_dec_count(inode);
677 static int ll_unlink(struct inode *dir, struct dentry *dentry)
679 return ll_common_unlink(dir, dentry, S_IFREG);
682 static int ll_rmdir(struct inode *dir, struct dentry *dentry)
684 struct inode * inode = dentry->d_inode;
688 if (!dentry->d_it || dentry->d_it->it_disposition == 0) {
689 if (!ext2_empty_dir(inode))
691 err = ll_common_unlink(dir, dentry, S_IFDIR);
693 err = dentry->d_it->it_status;
697 ext2_dec_count(inode);
702 static int ll_rename(struct inode * old_dir, struct dentry * old_dentry,
703 struct inode * new_dir, struct dentry * new_dentry)
705 struct inode * old_inode = old_dentry->d_inode;
706 struct inode * tgt_inode = new_dentry->d_inode;
707 struct page * dir_page = NULL;
708 struct ext2_dir_entry_2 * dir_de = NULL;
709 struct ext2_dir_entry_2 * old_de;
710 struct page * old_page;
713 if (new_dentry->d_it && new_dentry->d_it->it_disposition) {
715 tgt_inode->i_ctime = CURRENT_TIME;
716 tgt_inode->i_nlink--;
718 invalidate_inode_pages(old_dir);
719 invalidate_inode_pages(new_dir);
720 GOTO(out, err = new_dentry->d_it->it_status);
723 err = ll_mdc_rename(old_dir, new_dir, old_dentry, new_dentry);
727 old_de = ext2_find_entry (old_dir, old_dentry, &old_page);
731 if (S_ISDIR(old_inode->i_mode)) {
733 dir_de = ext2_dotdot(old_inode, &dir_page);
739 struct page *new_page;
740 struct ext2_dir_entry_2 *new_de;
743 if (dir_de && !ext2_empty_dir (tgt_inode))
747 new_de = ext2_find_entry (new_dir, new_dentry, &new_page);
750 ext2_inc_count(old_inode);
751 ext2_set_link(new_dir, new_de, new_page, old_inode);
752 tgt_inode->i_ctime = CURRENT_TIME;
754 tgt_inode->i_nlink--;
755 ext2_dec_count(tgt_inode);
759 if (new_dir->i_nlink >= EXT2_LINK_MAX)
762 ext2_inc_count(old_inode);
763 err = ll_add_link(new_dentry, old_inode);
765 ext2_dec_count(old_inode);
769 ext2_inc_count(new_dir);
772 ext2_delete_entry (old_de, old_page);
773 ext2_dec_count(old_inode);
776 ext2_set_link(old_inode, dir_de, dir_page, new_dir);
777 ext2_dec_count(old_dir);
784 page_cache_release(dir_page);
788 page_cache_release(old_page);
793 struct inode_operations ll_dir_inode_operations = {