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;
91 if (inode->i_generation != lic->lic_body->generation)
97 extern struct dentry_operations ll_d_ops;
99 int ll_lock(struct inode *dir, struct dentry *dentry,
100 struct lookup_intent *it, struct lustre_handle *lockh)
102 struct ll_sb_info *sbi = ll_i2sbi(dir);
107 if ((it->it_op & (IT_CREAT | IT_MKDIR | IT_SETATTR | IT_MKNOD)))
109 else if (it->it_op & (IT_READDIR | IT_GETATTR | IT_OPEN | IT_UNLINK |
110 IT_RMDIR | IT_RENAME | IT_RENAME2 | IT_READLINK))
112 else if (it->it_op & IT_SYMLINK) {
115 tgtlen = strlen(tgt);
117 } else if (it->it_op & IT_LOOKUP)
124 err = mdc_enqueue(&sbi->ll_mdc_conn, LDLM_MDSINTENT, it, lock_mode,
125 dir, dentry, lockh, tgt, tgtlen, dir, sizeof(*dir));
130 int ll_unlock(__u32 mode, struct lustre_handle *lockh)
134 ldlm_lock_decref(lockh, mode);
139 static struct dentry *ll_lookup2(struct inode *dir, struct dentry *dentry,
140 struct lookup_intent *it)
142 struct ptlrpc_request *request = NULL;
143 struct inode * inode = NULL;
144 struct ll_sb_info *sbi = ll_i2sbi(dir);
145 struct ll_read_inode2_cookie lic;
146 struct lustre_handle lockh;
147 struct lookup_intent lookup_it = { IT_LOOKUP };
158 CDEBUG(D_INFO, "name: %*s, intent op: %d\n", dentry->d_name.len,
159 dentry->d_name.name, it->it_op);
161 if (dentry->d_name.len > EXT2_NAME_LEN)
162 RETURN(ERR_PTR(-ENAMETOOLONG));
164 err = ll_lock(dir, dentry, it, &lockh);
166 RETURN(ERR_PTR(err));
167 memcpy(it->it_lock_handle, &lockh, sizeof(lockh));
169 request = (struct ptlrpc_request *)it->it_data;
170 if (it->it_disposition) {
171 int mode, symlen = 0;
175 if (it->it_op & (IT_CREAT | IT_MKDIR | IT_SYMLINK | IT_MKNOD)) {
176 /* For create ops, we want the lookup to be negative */
178 GOTO(negative, NULL);
179 } else if (it->it_op & (IT_GETATTR | IT_SETATTR | IT_LOOKUP)) {
180 /* For check ops, we want the lookup to succeed */
184 } else if (it->it_op & IT_RENAME) {
185 /* For rename, we want the lookup to succeed */
190 it->it_data = dentry;
191 } else if (it->it_op & (IT_UNLINK | IT_RMDIR)) {
192 /* For remove ops, we want the lookup to succeed */
197 } else if (it->it_op == IT_OPEN) {
199 if (it->it_status && it->it_status != -EEXIST)
201 } else if (it->it_op == IT_RENAME2) {
202 struct mds_body *body =
203 lustre_msg_buf(request->rq_repmsg, offset);
205 /* For rename2, this means the lookup is negative */
206 if (body->valid == 0)
208 goto iget; /* XXX not sure about this */
211 /* Do a getattr now that we have the lock */
212 lic.lic_body = lustre_msg_buf(request->rq_repmsg, offset);
213 ino = lic.lic_body->fid1.id;
214 mode = lic.lic_body->mode;
215 valid = OBD_MD_FLNOTOBD | OBD_MD_FLEASIZE;
216 if (it->it_op == IT_READLINK) {
217 valid |= OBD_MD_LINKNAME;
218 symlen = lic.lic_body->size;
220 ptlrpc_free_req(request);
222 err = mdc_getattr(&sbi->ll_mdc_conn, ino, mode,
223 valid, symlen, &request);
225 CERROR("failure %d inode %Ld\n", err, (long long)ino);
226 ptlrpc_free_req(request);
227 #warning FIXME: must release lock here
228 RETURN(ERR_PTR(-abs(err)));
232 struct ll_inode_info *lli = ll_i2info(dir);
235 memcpy(&lli->lli_intent_lock_handle, &lockh, sizeof(lockh));
238 ino = ll_inode_by_name(dir, dentry, &type);
239 #warning FIXME: handle negative inode case (see old ll_lookup)
241 err = mdc_getattr(&sbi->ll_mdc_conn, ino, type,
242 OBD_MD_FLNOTOBD|OBD_MD_FLEASIZE, 0, &request);
244 CERROR("failure %d inode %Ld\n", err, (long long)ino);
245 ptlrpc_free_req(request);
246 #warning FIXME: must release lock here
247 RETURN(ERR_PTR(-abs(err)));
252 lic.lic_body = lustre_msg_buf(request->rq_repmsg, offset);
253 if (S_ISREG(lic.lic_body->mode) &&
254 lic.lic_body->valid & OBD_MD_FLEASIZE) {
255 LASSERT(request->rq_repmsg->bufcount > offset);
256 lic.lic_lmm = lustre_msg_buf(request->rq_repmsg, offset + 1);
260 /* No rpc's happen during iget4, -ENOMEM's are possible */
261 inode = iget4(dir->i_sb, ino, ll_find_inode, &lic);
263 LASSERT(!IS_ERR(inode));
265 ptlrpc_free_req(request);
266 ll_intent_release(dentry);
267 RETURN(ERR_PTR(-ENOMEM));
272 ptlrpc_req_finished(request);
274 dentry->d_op = &ll_d_ops;
275 d_add(dentry, inode);
276 if (it->it_op == IT_LOOKUP)
277 ll_intent_release(dentry);
282 static struct inode *ll_create_node(struct inode *dir, const char *name,
283 int namelen, const char *tgt, int tgtlen,
284 int mode, __u64 extra,
285 struct lookup_intent *it,
286 struct lov_stripe_md *lsm)
289 struct ptlrpc_request *request = NULL;
290 struct mds_body *body;
292 time_t time = CURRENT_TIME;
293 struct ll_sb_info *sbi = ll_i2sbi(dir);
294 int gid = current->fsgid;
295 struct ll_read_inode2_cookie lic;
296 struct lov_mds_md *lmm = NULL;
301 if (dir->i_mode & S_ISGID) {
307 if (!it || !it->it_disposition) {
308 rc = mdc_create(&sbi->ll_mdc_conn, dir, name, namelen, tgt,
309 tgtlen, mode, current->fsuid,
310 gid, time, extra, lsm, &request);
315 body = lustre_msg_buf(request->rq_repmsg, 0);
317 mds_md_size = ll_mds_easize(dir->i_sb);
318 OBD_ALLOC(lmm, mds_md_size);
319 lov_packmd(lmm, lsm);
325 invalidate_inode_pages(dir);
326 request = it->it_data;
327 body = lustre_msg_buf(request->rq_repmsg, 1);
331 body->valid = OBD_MD_FLNOTOBD;
334 body->atime = body->ctime = body->mtime = time;
335 body->uid = current->fsuid;
341 inode = iget4(dir->i_sb, body->ino, ll_find_inode, &lic);
344 CERROR("new_inode -fatal: rc %d\n", rc);
349 if (!list_empty(&inode->i_dentry)) {
350 CERROR("new_inode -fatal: inode %d, ct %d lnk %d\n",
351 body->ino, atomic_read(&inode->i_count),
355 inode = ERR_PTR(-EIO);
362 OBD_FREE(lmm, mds_md_size);
363 ptlrpc_free_req(request);
367 static int ll_mdc_unlink(struct inode *dir, struct inode *child, __u32 mode,
368 const char *name, int len)
370 struct ptlrpc_request *request = NULL;
371 struct ll_sb_info *sbi = ll_i2sbi(dir);
376 err = mdc_unlink(&sbi->ll_mdc_conn, dir, child, mode, name, len,
378 ptlrpc_free_req(request);
383 int ll_mdc_link(struct dentry *src, struct inode *dir,
384 const char *name, int len)
386 struct ptlrpc_request *request = NULL;
388 struct ll_sb_info *sbi = ll_i2sbi(dir);
392 err = mdc_link(&sbi->ll_mdc_conn, src, dir, name,
394 ptlrpc_free_req(request);
399 int ll_mdc_rename(struct inode *src, struct inode *tgt,
400 struct dentry *old, struct dentry *new)
402 struct ptlrpc_request *request = NULL;
403 struct ll_sb_info *sbi = ll_i2sbi(src);
408 err = mdc_rename(&sbi->ll_mdc_conn, src, tgt,
409 old->d_name.name, old->d_name.len,
410 new->d_name.name, new->d_name.len, &request);
411 ptlrpc_free_req(request);
417 * By the time this is called, we already have created
418 * the directory cache entry for the new file, but it
419 * is so far negative - it has no inode.
421 * If the create succeeds, we fill in the inode information
422 * with d_instantiate().
425 static int ll_create(struct inode * dir, struct dentry * dentry, int mode)
430 struct lov_stripe_md *lsm = NULL;
431 struct ll_inode_info *lli = NULL;
434 if (dentry->d_it->it_disposition == 0) {
435 memset(&oa, 0, sizeof(oa));
436 oa.o_valid = OBD_MD_FLMODE;
437 oa.o_mode = S_IFREG | 0600;
438 rc = obd_create(ll_i2obdconn(dir), &oa, &lsm);
439 CDEBUG(D_DENTRY, "name %s mode %o o_id %lld: rc = %d\n",
440 dentry->d_name.name, mode, (long long)oa.o_id, rc);
445 inode = ll_create_node(dir, dentry->d_name.name, dentry->d_name.len,
446 NULL, 0, mode, 0, dentry->d_it, lsm);
450 CERROR("error creating MDS object for id %Ld: rc = %d\n",
451 (unsigned long long)oa.o_id, rc);
452 GOTO(out_destroy, rc);
455 if (dentry->d_it->it_disposition) {
456 lli = ll_i2info(inode);
457 memcpy(&lli->lli_intent_lock_handle,
458 dentry->d_it->it_lock_handle,
459 sizeof(struct lustre_handle));
460 d_instantiate(dentry, inode);
462 /* no directory data updates when intents rule */
463 rc = ext2_add_nondir(dentry, inode);
470 oa.o_easize = ll_mds_easize(inode->i_sb);
471 oa.o_valid |= OBD_MD_FLEASIZE;
472 err = obd_destroy(ll_i2obdconn(dir), &oa, lsm);
474 CERROR("error destroying objid %Ld on error: err %d\n",
475 (unsigned long long)oa.o_id, err);
481 static int ll_mknod(struct inode *dir, struct dentry *dentry, int mode,
487 inode = ll_create_node(dir, dentry->d_name.name, dentry->d_name.len,
488 NULL, 0, mode, rdev, dentry->d_it, NULL);
491 RETURN(PTR_ERR(inode));
493 /* no directory data updates when intents rule */
494 if (dentry->d_it && dentry->d_it->it_disposition)
495 d_instantiate(dentry, inode);
497 err = ext2_add_nondir(dentry, inode);
502 static int ll_symlink(struct inode *dir, struct dentry *dentry,
505 unsigned l = strlen(symname);
507 struct ll_inode_info *lli;
511 inode = ll_create_node(dir, dentry->d_name.name, dentry->d_name.len,
512 symname, l, S_IFLNK | S_IRWXUGO, 0,
515 RETURN(PTR_ERR(inode));
517 lli = ll_i2info(inode);
519 OBD_ALLOC(lli->lli_symlink_name, l + 1);
520 /* this _could_ be a non-fatal error, since the symlink is already
521 * stored on the MDS by this point, and we can re-get it in readlink.
523 if (!lli->lli_symlink_name)
526 memcpy(lli->lli_symlink_name, symname, l + 1);
529 /* no directory data updates when intents rule */
530 if (dentry->d_it && dentry->d_it->it_disposition)
531 d_instantiate(dentry, inode);
533 err = ext2_add_nondir(dentry, inode);
538 static int ll_link(struct dentry *old_dentry, struct inode * dir,
539 struct dentry *dentry)
542 struct inode *inode = old_dentry->d_inode;
544 #warning FIXME: still needs intent support
545 if (S_ISDIR(inode->i_mode))
548 if (inode->i_nlink >= EXT2_LINK_MAX)
551 err = ll_mdc_link(old_dentry, dir,
552 dentry->d_name.name, dentry->d_name.len);
556 inode->i_ctime = CURRENT_TIME;
557 ext2_inc_count(inode);
558 atomic_inc(&inode->i_count);
560 return ext2_add_nondir(dentry, inode);
563 static int ll_mkdir(struct inode * dir, struct dentry * dentry, int mode)
565 struct inode * inode;
569 if (dir->i_nlink >= EXT2_LINK_MAX)
574 inode = ll_create_node(dir, dentry->d_name.name, dentry->d_name.len,
575 NULL, 0, S_IFDIR | mode, 0, dentry->d_it, NULL);
576 err = PTR_ERR(inode);
580 ext2_inc_count(inode);
582 err = ext2_make_empty(inode, dir);
586 /* no directory data updates when intents rule */
587 if (dentry->d_it->it_disposition == 0) {
588 err = ll_add_link(dentry, inode);
593 d_instantiate(dentry, inode);
599 ext2_dec_count(inode);
600 ext2_dec_count(inode);
609 static int ll_common_unlink(struct inode *dir, struct dentry *dentry,
612 struct inode * inode = dentry->d_inode;
614 struct ext2_dir_entry_2 * de;
619 if (dentry->d_it && dentry->d_it->it_disposition) {
620 err = dentry->d_it->it_status;
621 invalidate_inode_pages(dir);
626 de = ext2_find_entry(dir, dentry, &page);
630 err = ll_mdc_unlink(dir, dentry->d_inode, mode,
631 dentry->d_name.name, dentry->d_name.len);
636 err = ext2_delete_entry(de, page);
640 invalidate_inode_pages(dir);
642 inode->i_ctime = dir->i_ctime;
644 ext2_dec_count(inode);
648 static int ll_unlink(struct inode *dir, struct dentry *dentry)
650 return ll_common_unlink(dir, dentry, S_IFREG);
653 static int ll_rmdir(struct inode *dir, struct dentry *dentry)
655 struct inode * inode = dentry->d_inode;
659 if (!dentry->d_it || dentry->d_it->it_disposition == 0) {
660 if (!ext2_empty_dir(inode))
662 err = ll_common_unlink(dir, dentry, S_IFDIR);
664 err = dentry->d_it->it_status;
668 ext2_dec_count(inode);
673 static int ll_rename(struct inode * old_dir, struct dentry * old_dentry,
674 struct inode * new_dir, struct dentry * new_dentry)
676 struct inode * old_inode = old_dentry->d_inode;
677 struct inode * new_inode = new_dentry->d_inode;
678 struct page * dir_page = NULL;
679 struct ext2_dir_entry_2 * dir_de = NULL;
680 struct ext2_dir_entry_2 * old_de;
681 struct page * old_page;
684 if (new_dentry->d_it && new_dentry->d_it->it_disposition) {
686 new_inode->i_ctime = CURRENT_TIME;
687 new_inode->i_nlink--;
689 invalidate_inode_pages(old_dir);
690 invalidate_inode_pages(new_dir);
691 GOTO(out, err = new_dentry->d_it->it_status);
694 err = ll_mdc_rename(old_dir, new_dir, old_dentry, new_dentry);
698 old_de = ext2_find_entry (old_dir, old_dentry, &old_page);
702 if (S_ISDIR(old_inode->i_mode)) {
704 dir_de = ext2_dotdot(old_inode, &dir_page);
710 struct page *new_page;
711 struct ext2_dir_entry_2 *new_de;
714 if (dir_de && !ext2_empty_dir (new_inode))
718 new_de = ext2_find_entry (new_dir, new_dentry, &new_page);
721 ext2_inc_count(old_inode);
722 ext2_set_link(new_dir, new_de, new_page, old_inode);
723 new_inode->i_ctime = CURRENT_TIME;
725 new_inode->i_nlink--;
726 ext2_dec_count(new_inode);
730 if (new_dir->i_nlink >= EXT2_LINK_MAX)
733 ext2_inc_count(old_inode);
734 err = ll_add_link(new_dentry, old_inode);
736 ext2_dec_count(old_inode);
740 ext2_inc_count(new_dir);
743 ext2_delete_entry (old_de, old_page);
744 ext2_dec_count(old_inode);
747 ext2_set_link(old_inode, dir_de, dir_page, new_dir);
748 ext2_dec_count(old_dir);
755 page_cache_release(dir_page);
759 page_cache_release(old_page);
764 struct inode_operations ll_dir_inode_operations = {