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); */
159 if (ll_i2info(dir)->lli_mount_epoch != ll_i2sbi(dir)->ll_mount_epoch)
160 RETURN(ERR_PTR(-EIO));
167 CDEBUG(D_INFO, "name: %*s, intent: %s\n", dentry->d_name.len,
168 dentry->d_name.name, ldlm_it2str(it->it_op));
170 if (dentry->d_name.len > EXT2_NAME_LEN)
171 RETURN(ERR_PTR(-ENAMETOOLONG));
173 err = ll_lock(dir, dentry, it, &lockh);
175 RETURN(ERR_PTR(err));
176 memcpy(it->it_lock_handle, &lockh, sizeof(lockh));
178 request = (struct ptlrpc_request *)it->it_data;
179 if (it->it_disposition) {
180 int mode, symlen = 0;
184 lic.lic_body = lustre_msg_buf(request->rq_repmsg, offset);
185 ino = lic.lic_body->fid1.id;
186 mode = lic.lic_body->mode;
187 if (it->it_op & (IT_CREAT | IT_MKDIR | IT_SYMLINK | IT_MKNOD)) {
188 /* For create ops, we want the lookup to be negative,
189 * unless the create failed in a way that indicates
190 * that the file is already there */
191 if (it->it_status != -EEXIST)
192 GOTO(negative, NULL);
193 } else if (it->it_op & (IT_GETATTR | IT_SETATTR | IT_LOOKUP)) {
194 /* For check ops, we want the lookup to succeed */
198 } else if (it->it_op & (IT_RENAME | IT_LINK)) {
199 /* For rename, we want the lookup to succeed */
204 it->it_data = dentry;
205 } else if (it->it_op & (IT_UNLINK | IT_RMDIR)) {
206 /* For remove ops, we want the lookup to succeed unless
207 * the file truly doesn't exist */
209 if (it->it_status == -ENOENT)
212 } else if (it->it_op == IT_OPEN) {
214 if (it->it_status && it->it_status != -EEXIST)
216 } else if (it->it_op & (IT_RENAME2|IT_LINK2)) {
217 struct mds_body *body =
218 lustre_msg_buf(request->rq_repmsg, offset);
220 /* For rename2, this means the lookup is negative */
222 if (body->valid == 0)
227 /* Do a getattr now that we have the lock */
228 valid = OBD_MD_FLNOTOBD | OBD_MD_FLEASIZE;
229 if (it->it_op == IT_READLINK) {
230 valid |= OBD_MD_LINKNAME;
231 symlen = lic.lic_body->size;
233 ptlrpc_req_finished(request);
235 err = mdc_getattr(&sbi->ll_mdc_conn, ino, mode,
236 valid, symlen, &request);
238 CERROR("failure %d inode %Ld\n", err, (long long)ino);
239 GOTO(drop_req, err = -abs(err));
243 struct ll_inode_info *lli = ll_i2info(dir);
246 memcpy(&lli->lli_intent_lock_handle, &lockh, sizeof(lockh));
249 ino = ll_inode_by_name(dir, dentry, &mode);
251 CERROR("inode %*s not found by name\n",
252 dentry->d_name.len, dentry->d_name.name);
253 GOTO(drop_lock, err = -ENOENT);
256 err = mdc_getattr(&sbi->ll_mdc_conn, ino, mode,
257 OBD_MD_FLNOTOBD|OBD_MD_FLEASIZE, 0, &request);
259 CERROR("failure %d inode %Ld\n", err, (long long)ino);
260 GOTO(drop_req, err = -abs(err));
265 lic.lic_body = lustre_msg_buf(request->rq_repmsg, offset);
266 if (S_ISREG(lic.lic_body->mode) &&
267 lic.lic_body->valid & OBD_MD_FLEASIZE) {
268 LASSERT(request->rq_repmsg->bufcount > offset);
269 lic.lic_lmm = lustre_msg_buf(request->rq_repmsg, offset + 1);
273 /* No rpc's happen during iget4, -ENOMEM's are possible */
275 inode = iget4(dir->i_sb, ino, ll_find_inode, &lic);
278 ptlrpc_free_req(request);
279 ll_intent_release(dentry);
280 RETURN(ERR_PTR(-ENOMEM));
285 ptlrpc_req_finished(request);
287 dentry->d_op = &ll_d_ops;
288 d_add(dentry, inode);
290 if (ll_d2d(dentry) == NULL)
292 // down(&ll_d2d(dentry)->lld_it_sem);
293 // dentry->d_it = it;
295 if (it->it_op == IT_LOOKUP)
296 ll_intent_release(dentry);
301 ptlrpc_free_req(request);
303 #warning FIXME: must release lock here
307 static struct inode *ll_create_node(struct inode *dir, const char *name,
308 int namelen, const char *tgt, int tgtlen,
309 int mode, __u64 extra,
310 struct lookup_intent *it,
311 struct lov_stripe_md *lsm)
314 struct ptlrpc_request *request = NULL;
315 struct mds_body *body;
317 time_t time = CURRENT_TIME;
318 struct ll_sb_info *sbi = ll_i2sbi(dir);
319 int gid = current->fsgid;
320 struct ll_read_inode2_cookie lic;
321 struct lov_mds_md *lmm = NULL;
326 if (dir->i_mode & S_ISGID) {
332 if (!it || !it->it_disposition) {
333 rc = mdc_create(&sbi->ll_mdc_conn, dir, name, namelen, tgt,
334 tgtlen, mode, current->fsuid,
335 gid, time, extra, lsm, &request);
340 body = lustre_msg_buf(request->rq_repmsg, 0);
342 mds_md_size = ll_mds_easize(dir->i_sb);
343 OBD_ALLOC(lmm, mds_md_size);
344 lov_packmd(lmm, lsm);
350 invalidate_inode_pages(dir);
351 request = it->it_data;
352 body = lustre_msg_buf(request->rq_repmsg, 1);
356 body->valid = OBD_MD_FLNOTOBD;
359 body->atime = body->ctime = body->mtime = time;
360 body->uid = current->fsuid;
366 inode = iget4(dir->i_sb, body->ino, ll_find_inode, &lic);
369 CERROR("new_inode -fatal: rc %d\n", rc);
374 if (!list_empty(&inode->i_dentry)) {
375 CERROR("new_inode -fatal: inode %d, ct %d lnk %d\n",
376 body->ino, atomic_read(&inode->i_count),
380 inode = ERR_PTR(-EIO);
387 OBD_FREE(lmm, mds_md_size);
388 ptlrpc_req_finished(request);
392 static int ll_mdc_unlink(struct inode *dir, struct inode *child, __u32 mode,
393 const char *name, int len)
395 struct ptlrpc_request *request = NULL;
396 struct ll_sb_info *sbi = ll_i2sbi(dir);
401 err = mdc_unlink(&sbi->ll_mdc_conn, dir, child, mode, name, len,
403 ptlrpc_req_finished(request);
408 int ll_mdc_link(struct dentry *src, struct inode *dir,
409 const char *name, int len)
411 struct ptlrpc_request *request = NULL;
413 struct ll_sb_info *sbi = ll_i2sbi(dir);
417 err = mdc_link(&sbi->ll_mdc_conn, src, dir, name,
419 ptlrpc_req_finished(request);
424 int ll_mdc_rename(struct inode *src, struct inode *tgt,
425 struct dentry *old, struct dentry *new)
427 struct ptlrpc_request *request = NULL;
428 struct ll_sb_info *sbi = ll_i2sbi(src);
433 err = mdc_rename(&sbi->ll_mdc_conn, src, tgt,
434 old->d_name.name, old->d_name.len,
435 new->d_name.name, new->d_name.len, &request);
436 ptlrpc_req_finished(request);
442 * By the time this is called, we already have created
443 * the directory cache entry for the new file, but it
444 * is so far negative - it has no inode.
446 * If the create succeeds, we fill in the inode information
447 * with d_instantiate().
450 static int ll_create(struct inode * dir, struct dentry * dentry, int mode)
453 struct obdo *oa = NULL;
455 struct lov_stripe_md *lsm = NULL;
456 struct ll_inode_info *lli = NULL;
459 CHECK_MOUNT_EPOCH(dir);
461 if (dentry->d_it->it_disposition == 0) {
462 int gid = current->fsgid;
464 if (dir->i_mode & S_ISGID)
471 oa->o_mode = S_IFREG | 0600;
472 /* FIXME: we set the UID/GID fields to 0 for now, because it
473 * fixes a bug on the BA OSTs. We should really set
474 * them properly, and this needs to be revisited when
475 * we do proper credentials checking on the OST, and
476 * set the attributes on the OST in ll_inode_setattr().
477 oa->o_uid = current->fsuid;
482 oa->o_valid = OBD_MD_FLTYPE | OBD_MD_FLMODE | OBD_MD_FLUID |
484 rc = obd_create(ll_i2obdconn(dir), oa, &lsm);
485 CDEBUG(D_DENTRY, "name %s mode %o o_id "LPX64": rc = %d\n",
486 dentry->d_name.name, mode, oa->o_id, rc);
491 inode = ll_create_node(dir, dentry->d_name.name, dentry->d_name.len,
492 NULL, 0, mode, 0, dentry->d_it, lsm);
496 CERROR("error creating MDS object for id "LPX64": rc = %d\n",
498 GOTO(out_destroy, rc);
501 if (dentry->d_it->it_disposition) {
502 lli = ll_i2info(inode);
503 memcpy(&lli->lli_intent_lock_handle,
504 dentry->d_it->it_lock_handle,
505 sizeof(struct lustre_handle));
506 d_instantiate(dentry, inode);
508 /* no directory data updates when intents rule */
509 rc = ext2_add_nondir(dentry, inode);
523 oa->o_easize = ll_mds_easize(inode->i_sb);
524 oa->o_valid |= OBD_MD_FLEASIZE;
525 err = obd_destroy(ll_i2obdconn(dir), oa, lsm);
527 CERROR("error uncreating objid "LPX64": err %d\n",
534 static int ll_mknod(struct inode *dir, struct dentry *dentry, int mode,
540 inode = ll_create_node(dir, dentry->d_name.name, dentry->d_name.len,
541 NULL, 0, mode, rdev, dentry->d_it, NULL);
544 RETURN(PTR_ERR(inode));
546 /* no directory data updates when intents rule */
547 if (dentry->d_it && dentry->d_it->it_disposition)
548 d_instantiate(dentry, inode);
550 err = ext2_add_nondir(dentry, inode);
555 static int ll_symlink(struct inode *dir, struct dentry *dentry,
558 unsigned l = strlen(symname);
560 struct ll_inode_info *lli;
564 CHECK_MOUNT_EPOCH(dir);
566 inode = ll_create_node(dir, dentry->d_name.name, dentry->d_name.len,
567 symname, l, S_IFLNK | S_IRWXUGO, 0,
570 RETURN(PTR_ERR(inode));
572 lli = ll_i2info(inode);
574 OBD_ALLOC(lli->lli_symlink_name, l + 1);
575 /* this _could_ be a non-fatal error, since the symlink is already
576 * stored on the MDS by this point, and we can re-get it in readlink.
578 if (!lli->lli_symlink_name)
581 memcpy(lli->lli_symlink_name, symname, l + 1);
584 /* no directory data updates when intents rule */
585 if (dentry->d_it && dentry->d_it->it_disposition)
586 d_instantiate(dentry, inode);
588 err = ext2_add_nondir(dentry, inode);
593 static int ll_link(struct dentry *old_dentry, struct inode * dir,
594 struct dentry *dentry)
597 struct inode *inode = old_dentry->d_inode;
599 if (dentry->d_it && dentry->d_it->it_disposition) {
600 int err = dentry->d_it->it_status;
603 inode->i_ctime = CURRENT_TIME;
604 ext2_inc_count(inode);
605 atomic_inc(&inode->i_count);
606 d_instantiate(dentry, inode);
607 invalidate_inode_pages(dir);
611 if (S_ISDIR(inode->i_mode))
614 if (inode->i_nlink >= EXT2_LINK_MAX)
617 err = ll_mdc_link(old_dentry, dir,
618 dentry->d_name.name, dentry->d_name.len);
622 inode->i_ctime = CURRENT_TIME;
623 ext2_inc_count(inode);
624 atomic_inc(&inode->i_count);
626 return ext2_add_nondir(dentry, inode);
629 static int ll_mkdir(struct inode * dir, struct dentry * dentry, int mode)
631 struct inode * inode;
635 if (dir->i_nlink >= EXT2_LINK_MAX)
640 inode = ll_create_node(dir, dentry->d_name.name, dentry->d_name.len,
641 NULL, 0, S_IFDIR | mode, 0, dentry->d_it, NULL);
642 err = PTR_ERR(inode);
646 ext2_inc_count(inode);
648 err = ext2_make_empty(inode, dir);
652 /* no directory data updates when intents rule */
653 if (dentry->d_it->it_disposition == 0) {
654 err = ll_add_link(dentry, inode);
659 d_instantiate(dentry, inode);
665 ext2_dec_count(inode);
666 ext2_dec_count(inode);
675 static int ll_common_unlink(struct inode *dir, struct dentry *dentry,
678 struct inode * inode = dentry->d_inode;
680 struct ext2_dir_entry_2 * de;
685 if (dentry->d_it && dentry->d_it->it_disposition) {
686 err = dentry->d_it->it_status;
687 invalidate_inode_pages(dir);
692 de = ext2_find_entry(dir, dentry, &page);
696 err = ll_mdc_unlink(dir, dentry->d_inode, mode,
697 dentry->d_name.name, dentry->d_name.len);
702 err = ext2_delete_entry(de, page);
706 invalidate_inode_pages(dir);
708 inode->i_ctime = dir->i_ctime;
710 ext2_dec_count(inode);
714 static int ll_unlink(struct inode *dir, struct dentry *dentry)
716 return ll_common_unlink(dir, dentry, S_IFREG);
719 static int ll_rmdir(struct inode *dir, struct dentry *dentry)
721 struct inode * inode = dentry->d_inode;
725 if (!dentry->d_it || dentry->d_it->it_disposition == 0) {
726 if (!ext2_empty_dir(inode))
728 err = ll_common_unlink(dir, dentry, S_IFDIR);
730 err = dentry->d_it->it_status;
734 ext2_dec_count(inode);
739 static int ll_rename(struct inode * old_dir, struct dentry * old_dentry,
740 struct inode * new_dir, struct dentry * new_dentry)
742 struct inode * old_inode = old_dentry->d_inode;
743 struct inode * tgt_inode = new_dentry->d_inode;
744 struct page * dir_page = NULL;
745 struct ext2_dir_entry_2 * dir_de = NULL;
746 struct ext2_dir_entry_2 * old_de;
747 struct page * old_page;
750 if (new_dentry->d_it && new_dentry->d_it->it_disposition) {
752 tgt_inode->i_ctime = CURRENT_TIME;
753 tgt_inode->i_nlink--;
755 invalidate_inode_pages(old_dir);
756 invalidate_inode_pages(new_dir);
757 GOTO(out, err = new_dentry->d_it->it_status);
760 err = ll_mdc_rename(old_dir, new_dir, old_dentry, new_dentry);
764 old_de = ext2_find_entry (old_dir, old_dentry, &old_page);
768 if (S_ISDIR(old_inode->i_mode)) {
770 dir_de = ext2_dotdot(old_inode, &dir_page);
776 struct page *new_page;
777 struct ext2_dir_entry_2 *new_de;
780 if (dir_de && !ext2_empty_dir (tgt_inode))
784 new_de = ext2_find_entry (new_dir, new_dentry, &new_page);
787 ext2_inc_count(old_inode);
788 ext2_set_link(new_dir, new_de, new_page, old_inode);
789 tgt_inode->i_ctime = CURRENT_TIME;
791 tgt_inode->i_nlink--;
792 ext2_dec_count(tgt_inode);
796 if (new_dir->i_nlink >= EXT2_LINK_MAX)
799 ext2_inc_count(old_inode);
800 err = ll_add_link(new_dentry, old_inode);
802 ext2_dec_count(old_inode);
806 ext2_inc_count(new_dir);
809 ext2_delete_entry (old_de, old_page);
810 ext2_dec_count(old_inode);
813 ext2_set_link(old_inode, dir_de, dir_page, new_dir);
814 ext2_dec_count(old_dir);
821 page_cache_release(dir_page);
825 page_cache_release(old_page);
830 struct inode_operations ll_dir_inode_operations = {