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 };
163 CDEBUG(D_INFO, "name: %*s, intent op: %d\n", dentry->d_name.len,
164 dentry->d_name.name, it->it_op);
166 if (dentry->d_name.len > EXT2_NAME_LEN)
167 RETURN(ERR_PTR(-ENAMETOOLONG));
169 err = ll_lock(dir, dentry, it, &lockh);
171 RETURN(ERR_PTR(err));
172 memcpy(it->it_lock_handle, &lockh, sizeof(lockh));
174 request = (struct ptlrpc_request *)it->it_data;
175 if (it->it_disposition) {
176 int mode, symlen = 0;
180 if (it->it_op & (IT_CREAT | IT_MKDIR | IT_SYMLINK | IT_MKNOD)) {
181 /* For create ops, we want the lookup to be negative */
183 GOTO(negative, NULL);
184 } else if (it->it_op & (IT_GETATTR | IT_SETATTR | IT_LOOKUP)) {
185 /* For check ops, we want the lookup to succeed */
189 } else if (it->it_op & (IT_RENAME | IT_LINK)) {
190 /* For rename, we want the lookup to succeed */
195 it->it_data = dentry;
196 } else if (it->it_op & (IT_UNLINK | IT_RMDIR)) {
197 /* For remove ops, we want the lookup to succeed */
202 } else if (it->it_op == IT_OPEN) {
204 if (it->it_status && it->it_status != -EEXIST)
206 } else if (it->it_op & (IT_RENAME2|IT_LINK2)) {
207 struct mds_body *body =
208 lustre_msg_buf(request->rq_repmsg, offset);
210 /* For rename2, this means the lookup is negative */
212 if (body->valid == 0)
214 goto iget; /* XXX not sure about this */
217 /* Do a getattr now that we have the lock */
218 lic.lic_body = lustre_msg_buf(request->rq_repmsg, offset);
219 ino = lic.lic_body->fid1.id;
220 mode = lic.lic_body->mode;
221 valid = OBD_MD_FLNOTOBD | OBD_MD_FLEASIZE;
222 if (it->it_op == IT_READLINK) {
223 valid |= OBD_MD_LINKNAME;
224 symlen = lic.lic_body->size;
226 ptlrpc_free_req(request);
228 err = mdc_getattr(&sbi->ll_mdc_conn, ino, mode,
229 valid, symlen, &request);
231 CERROR("failure %d inode %Ld\n", err, (long long)ino);
232 ptlrpc_free_req(request);
233 #warning FIXME: must release lock here
234 RETURN(ERR_PTR(-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, &type);
245 #warning FIXME: handle negative inode case (see old ll_lookup)
247 err = mdc_getattr(&sbi->ll_mdc_conn, ino, type,
248 OBD_MD_FLNOTOBD|OBD_MD_FLEASIZE, 0, &request);
250 CERROR("failure %d inode %Ld\n", err, (long long)ino);
251 ptlrpc_free_req(request);
252 #warning FIXME: must release lock here
253 RETURN(ERR_PTR(-abs(err)));
258 lic.lic_body = lustre_msg_buf(request->rq_repmsg, offset);
259 if (S_ISREG(lic.lic_body->mode) &&
260 lic.lic_body->valid & OBD_MD_FLEASIZE) {
261 LASSERT(request->rq_repmsg->bufcount > offset);
262 lic.lic_lmm = lustre_msg_buf(request->rq_repmsg, offset + 1);
266 /* No rpc's happen during iget4, -ENOMEM's are possible */
267 inode = iget4(dir->i_sb, ino, ll_find_inode, &lic);
269 LASSERT(!IS_ERR(inode));
271 ptlrpc_free_req(request);
272 ll_intent_release(dentry);
273 RETURN(ERR_PTR(-ENOMEM));
278 ptlrpc_req_finished(request);
280 dentry->d_op = &ll_d_ops;
281 d_add(dentry, inode);
282 if (it->it_op == IT_LOOKUP)
283 ll_intent_release(dentry);
288 static struct inode *ll_create_node(struct inode *dir, const char *name,
289 int namelen, const char *tgt, int tgtlen,
290 int mode, __u64 extra,
291 struct lookup_intent *it,
292 struct lov_stripe_md *lsm)
295 struct ptlrpc_request *request = NULL;
296 struct mds_body *body;
298 time_t time = CURRENT_TIME;
299 struct ll_sb_info *sbi = ll_i2sbi(dir);
300 int gid = current->fsgid;
301 struct ll_read_inode2_cookie lic;
302 struct lov_mds_md *lmm = NULL;
307 if (dir->i_mode & S_ISGID) {
313 if (!it || !it->it_disposition) {
314 rc = mdc_create(&sbi->ll_mdc_conn, dir, name, namelen, tgt,
315 tgtlen, mode, current->fsuid,
316 gid, time, extra, lsm, &request);
321 body = lustre_msg_buf(request->rq_repmsg, 0);
323 mds_md_size = ll_mds_easize(dir->i_sb);
324 OBD_ALLOC(lmm, mds_md_size);
325 lov_packmd(lmm, lsm);
331 invalidate_inode_pages(dir);
332 request = it->it_data;
333 body = lustre_msg_buf(request->rq_repmsg, 1);
337 body->valid = OBD_MD_FLNOTOBD;
340 body->atime = body->ctime = body->mtime = time;
341 body->uid = current->fsuid;
347 inode = iget4(dir->i_sb, body->ino, ll_find_inode, &lic);
350 CERROR("new_inode -fatal: rc %d\n", rc);
355 if (!list_empty(&inode->i_dentry)) {
356 CERROR("new_inode -fatal: inode %d, ct %d lnk %d\n",
357 body->ino, atomic_read(&inode->i_count),
361 inode = ERR_PTR(-EIO);
368 OBD_FREE(lmm, mds_md_size);
369 ptlrpc_free_req(request);
373 static int ll_mdc_unlink(struct inode *dir, struct inode *child, __u32 mode,
374 const char *name, int len)
376 struct ptlrpc_request *request = NULL;
377 struct ll_sb_info *sbi = ll_i2sbi(dir);
382 err = mdc_unlink(&sbi->ll_mdc_conn, dir, child, mode, name, len,
384 ptlrpc_free_req(request);
389 int ll_mdc_link(struct dentry *src, struct inode *dir,
390 const char *name, int len)
392 struct ptlrpc_request *request = NULL;
394 struct ll_sb_info *sbi = ll_i2sbi(dir);
398 err = mdc_link(&sbi->ll_mdc_conn, src, dir, name,
400 ptlrpc_free_req(request);
405 int ll_mdc_rename(struct inode *src, struct inode *tgt,
406 struct dentry *old, struct dentry *new)
408 struct ptlrpc_request *request = NULL;
409 struct ll_sb_info *sbi = ll_i2sbi(src);
414 err = mdc_rename(&sbi->ll_mdc_conn, src, tgt,
415 old->d_name.name, old->d_name.len,
416 new->d_name.name, new->d_name.len, &request);
417 ptlrpc_free_req(request);
423 * By the time this is called, we already have created
424 * the directory cache entry for the new file, but it
425 * is so far negative - it has no inode.
427 * If the create succeeds, we fill in the inode information
428 * with d_instantiate().
431 static int ll_create(struct inode * dir, struct dentry * dentry, int mode)
436 struct lov_stripe_md *lsm = NULL;
437 struct ll_inode_info *lli = NULL;
440 if (dentry->d_it->it_disposition == 0) {
441 memset(&oa, 0, sizeof(oa));
442 oa.o_valid = OBD_MD_FLMODE;
443 oa.o_mode = S_IFREG | 0600;
444 rc = obd_create(ll_i2obdconn(dir), &oa, &lsm);
445 CDEBUG(D_DENTRY, "name %s mode %o o_id %lld: rc = %d\n",
446 dentry->d_name.name, mode, (long long)oa.o_id, rc);
451 inode = ll_create_node(dir, dentry->d_name.name, dentry->d_name.len,
452 NULL, 0, mode, 0, dentry->d_it, lsm);
456 CERROR("error creating MDS object for id %Ld: rc = %d\n",
457 (unsigned long long)oa.o_id, rc);
458 GOTO(out_destroy, rc);
461 if (dentry->d_it->it_disposition) {
462 lli = ll_i2info(inode);
463 memcpy(&lli->lli_intent_lock_handle,
464 dentry->d_it->it_lock_handle,
465 sizeof(struct lustre_handle));
466 d_instantiate(dentry, inode);
468 /* no directory data updates when intents rule */
469 rc = ext2_add_nondir(dentry, inode);
476 oa.o_easize = ll_mds_easize(inode->i_sb);
477 oa.o_valid |= OBD_MD_FLEASIZE;
478 err = obd_destroy(ll_i2obdconn(dir), &oa, lsm);
480 CERROR("error destroying objid %Ld on error: err %d\n",
481 (unsigned long long)oa.o_id, err);
487 static int ll_mknod(struct inode *dir, struct dentry *dentry, int mode,
493 inode = ll_create_node(dir, dentry->d_name.name, dentry->d_name.len,
494 NULL, 0, mode, rdev, dentry->d_it, NULL);
497 RETURN(PTR_ERR(inode));
499 /* no directory data updates when intents rule */
500 if (dentry->d_it && dentry->d_it->it_disposition)
501 d_instantiate(dentry, inode);
503 err = ext2_add_nondir(dentry, inode);
508 static int ll_symlink(struct inode *dir, struct dentry *dentry,
511 unsigned l = strlen(symname);
513 struct ll_inode_info *lli;
517 inode = ll_create_node(dir, dentry->d_name.name, dentry->d_name.len,
518 symname, l, S_IFLNK | S_IRWXUGO, 0,
521 RETURN(PTR_ERR(inode));
523 lli = ll_i2info(inode);
525 OBD_ALLOC(lli->lli_symlink_name, l + 1);
526 /* this _could_ be a non-fatal error, since the symlink is already
527 * stored on the MDS by this point, and we can re-get it in readlink.
529 if (!lli->lli_symlink_name)
532 memcpy(lli->lli_symlink_name, symname, l + 1);
535 /* no directory data updates when intents rule */
536 if (dentry->d_it && dentry->d_it->it_disposition)
537 d_instantiate(dentry, inode);
539 err = ext2_add_nondir(dentry, inode);
544 static int ll_link(struct dentry *old_dentry, struct inode * dir,
545 struct dentry *dentry)
548 struct inode *inode = old_dentry->d_inode;
550 if (dentry->d_it && dentry->d_it->it_disposition) {
551 int err = dentry->d_it->it_status;
554 inode->i_ctime = CURRENT_TIME;
555 ext2_inc_count(inode);
556 atomic_inc(&inode->i_count);
557 d_instantiate(dentry, inode);
558 invalidate_inode_pages(dir);
562 if (S_ISDIR(inode->i_mode))
565 if (inode->i_nlink >= EXT2_LINK_MAX)
568 err = ll_mdc_link(old_dentry, dir,
569 dentry->d_name.name, dentry->d_name.len);
573 inode->i_ctime = CURRENT_TIME;
574 ext2_inc_count(inode);
575 atomic_inc(&inode->i_count);
577 return ext2_add_nondir(dentry, inode);
580 static int ll_mkdir(struct inode * dir, struct dentry * dentry, int mode)
582 struct inode * inode;
586 if (dir->i_nlink >= EXT2_LINK_MAX)
591 inode = ll_create_node(dir, dentry->d_name.name, dentry->d_name.len,
592 NULL, 0, S_IFDIR | mode, 0, dentry->d_it, NULL);
593 err = PTR_ERR(inode);
597 ext2_inc_count(inode);
599 err = ext2_make_empty(inode, dir);
603 /* no directory data updates when intents rule */
604 if (dentry->d_it->it_disposition == 0) {
605 err = ll_add_link(dentry, inode);
610 d_instantiate(dentry, inode);
616 ext2_dec_count(inode);
617 ext2_dec_count(inode);
626 static int ll_common_unlink(struct inode *dir, struct dentry *dentry,
629 struct inode * inode = dentry->d_inode;
631 struct ext2_dir_entry_2 * de;
636 if (dentry->d_it && dentry->d_it->it_disposition) {
637 err = dentry->d_it->it_status;
638 invalidate_inode_pages(dir);
643 de = ext2_find_entry(dir, dentry, &page);
647 err = ll_mdc_unlink(dir, dentry->d_inode, mode,
648 dentry->d_name.name, dentry->d_name.len);
653 err = ext2_delete_entry(de, page);
657 invalidate_inode_pages(dir);
659 inode->i_ctime = dir->i_ctime;
661 ext2_dec_count(inode);
665 static int ll_unlink(struct inode *dir, struct dentry *dentry)
667 return ll_common_unlink(dir, dentry, S_IFREG);
670 static int ll_rmdir(struct inode *dir, struct dentry *dentry)
672 struct inode * inode = dentry->d_inode;
676 if (!dentry->d_it || dentry->d_it->it_disposition == 0) {
677 if (!ext2_empty_dir(inode))
679 err = ll_common_unlink(dir, dentry, S_IFDIR);
681 err = dentry->d_it->it_status;
685 ext2_dec_count(inode);
690 static int ll_rename(struct inode * old_dir, struct dentry * old_dentry,
691 struct inode * new_dir, struct dentry * new_dentry)
693 struct inode * old_inode = old_dentry->d_inode;
694 struct inode * tgt_inode = new_dentry->d_inode;
695 struct page * dir_page = NULL;
696 struct ext2_dir_entry_2 * dir_de = NULL;
697 struct ext2_dir_entry_2 * old_de;
698 struct page * old_page;
701 if (new_dentry->d_it && new_dentry->d_it->it_disposition) {
703 tgt_inode->i_ctime = CURRENT_TIME;
704 tgt_inode->i_nlink--;
706 invalidate_inode_pages(old_dir);
707 invalidate_inode_pages(new_dir);
708 GOTO(out, err = new_dentry->d_it->it_status);
711 err = ll_mdc_rename(old_dir, new_dir, old_dentry, new_dentry);
715 old_de = ext2_find_entry (old_dir, old_dentry, &old_page);
719 if (S_ISDIR(old_inode->i_mode)) {
721 dir_de = ext2_dotdot(old_inode, &dir_page);
727 struct page *new_page;
728 struct ext2_dir_entry_2 *new_de;
731 if (dir_de && !ext2_empty_dir (tgt_inode))
735 new_de = ext2_find_entry (new_dir, new_dentry, &new_page);
738 ext2_inc_count(old_inode);
739 ext2_set_link(new_dir, new_de, new_page, old_inode);
740 tgt_inode->i_ctime = CURRENT_TIME;
742 tgt_inode->i_nlink--;
743 ext2_dec_count(tgt_inode);
747 if (new_dir->i_nlink >= EXT2_LINK_MAX)
750 ext2_inc_count(old_inode);
751 err = ll_add_link(new_dentry, old_inode);
753 ext2_dec_count(old_inode);
757 ext2_inc_count(new_dir);
760 ext2_delete_entry (old_de, old_page);
761 ext2_dec_count(old_inode);
764 ext2_set_link(old_inode, dir_de, dir_page, new_dir);
765 ext2_dec_count(old_dir);
772 page_cache_release(dir_page);
776 page_cache_release(old_page);
781 struct inode_operations ll_dir_inode_operations = {