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 /* CREAT needs to be tested before open (both could be set) */
112 if ((it->it_op & (IT_CREAT | IT_MKDIR | IT_SETATTR | IT_MKNOD))) {
114 } else if (it->it_op & (IT_READDIR | IT_GETATTR | IT_OPEN | IT_UNLINK |
115 IT_RMDIR | IT_RENAME | IT_RENAME2 | IT_READLINK|
116 IT_LINK | IT_LINK2 | IT_LOOKUP)) {
117 /* XXXphil PW for LINK2/RENAME2? */
119 } else if (it->it_op & IT_SYMLINK) {
122 tgtlen = strlen(tgt);
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) {
161 RETURN(ERR_PTR(-EIO));
169 CDEBUG(D_INFO, "name: %*s, intent: %s\n", dentry->d_name.len,
170 dentry->d_name.name, ldlm_it2str(it->it_op));
172 if (dentry->d_name.len > EXT2_NAME_LEN)
173 RETURN(ERR_PTR(-ENAMETOOLONG));
175 err = ll_lock(dir, dentry, it, &lockh);
177 RETURN(ERR_PTR(err));
178 memcpy(it->it_lock_handle, &lockh, sizeof(lockh));
180 request = (struct ptlrpc_request *)it->it_data;
181 if (it->it_disposition) {
182 int mode, symlen = 0;
186 lic.lic_body = lustre_msg_buf(request->rq_repmsg, offset);
187 ino = lic.lic_body->fid1.id;
188 mode = lic.lic_body->mode;
189 if (it->it_op & (IT_CREAT | IT_MKDIR | IT_SYMLINK | IT_MKNOD)) {
190 /* For create ops, we want the lookup to be negative,
191 * unless the create failed in a way that indicates
192 * that the file is already there */
193 if (it->it_status != -EEXIST)
194 GOTO(negative, NULL);
195 } else if (it->it_op & (IT_GETATTR | IT_SETATTR | IT_LOOKUP)) {
196 /* For check ops, we want the lookup to succeed */
200 } else if (it->it_op & (IT_RENAME | IT_LINK)) {
201 /* For rename, we want the lookup to succeed */
206 it->it_data = dentry;
207 } else if (it->it_op & (IT_UNLINK | IT_RMDIR)) {
208 /* For remove ops, we want the lookup to succeed unless
209 * the file truly doesn't exist */
211 if (it->it_status == -ENOENT)
214 } else if (it->it_op == IT_OPEN) {
216 if (it->it_status && it->it_status != -EEXIST)
218 } else if (it->it_op & (IT_RENAME2|IT_LINK2)) {
219 struct mds_body *body =
220 lustre_msg_buf(request->rq_repmsg, offset);
222 /* For rename2, this means the lookup is negative */
224 if (body->valid == 0)
229 /* Do a getattr now that we have the lock */
230 valid = OBD_MD_FLNOTOBD | OBD_MD_FLEASIZE;
231 if (it->it_op == IT_READLINK) {
232 valid |= OBD_MD_LINKNAME;
233 symlen = lic.lic_body->size;
235 ptlrpc_req_finished(request);
237 err = mdc_getattr(&sbi->ll_mdc_conn, ino, mode,
238 valid, symlen, &request);
240 CERROR("failure %d inode %Ld\n", err, (long long)ino);
241 GOTO(drop_req, err = -abs(err));
245 struct ll_inode_info *lli = ll_i2info(dir);
248 memcpy(&lli->lli_intent_lock_handle, &lockh, sizeof(lockh));
251 ino = ll_inode_by_name(dir, dentry, &mode);
253 CERROR("inode %*s not found by name\n",
254 dentry->d_name.len, dentry->d_name.name);
255 GOTO(drop_lock, err = -ENOENT);
258 err = mdc_getattr(&sbi->ll_mdc_conn, ino, mode,
259 OBD_MD_FLNOTOBD|OBD_MD_FLEASIZE, 0, &request);
261 CERROR("failure %d inode %Ld\n", err, (long long)ino);
262 GOTO(drop_req, err = -abs(err));
267 lic.lic_body = lustre_msg_buf(request->rq_repmsg, offset);
268 if (S_ISREG(lic.lic_body->mode) &&
269 lic.lic_body->valid & OBD_MD_FLEASIZE) {
270 LASSERT(request->rq_repmsg->bufcount > offset);
271 lic.lic_lmm = lustre_msg_buf(request->rq_repmsg, offset + 1);
275 /* No rpc's happen during iget4, -ENOMEM's are possible */
277 inode = iget4(dir->i_sb, ino, ll_find_inode, &lic);
280 ptlrpc_free_req(request);
281 ll_intent_release(dentry);
282 RETURN(ERR_PTR(-ENOMEM));
287 ptlrpc_req_finished(request);
289 dentry->d_op = &ll_d_ops;
290 d_add(dentry, inode);
292 if (ll_d2d(dentry) == NULL)
294 // down(&ll_d2d(dentry)->lld_it_sem);
295 // dentry->d_it = it;
297 if (it->it_op == IT_LOOKUP)
298 ll_intent_release(dentry);
303 ptlrpc_free_req(request);
305 #warning FIXME: must release lock here
309 static struct inode *ll_create_node(struct inode *dir, const char *name,
310 int namelen, const char *tgt, int tgtlen,
311 int mode, __u64 extra,
312 struct lookup_intent *it,
313 struct lov_stripe_md *lsm)
316 struct ptlrpc_request *request = NULL;
317 struct mds_body *body;
319 time_t time = CURRENT_TIME;
320 struct ll_sb_info *sbi = ll_i2sbi(dir);
321 int gid = current->fsgid;
322 struct ll_read_inode2_cookie lic;
323 struct lov_mds_md *lmm = NULL;
328 if (dir->i_mode & S_ISGID) {
334 if (!it || !it->it_disposition) {
335 rc = mdc_create(&sbi->ll_mdc_conn, dir, name, namelen, tgt,
336 tgtlen, mode, current->fsuid,
337 gid, time, extra, lsm, &request);
342 body = lustre_msg_buf(request->rq_repmsg, 0);
344 OBD_ALLOC(lmm, lsm->lsm_mds_easize);
345 lov_packmd(lmm, lsm);
351 invalidate_inode_pages(dir);
352 request = it->it_data;
353 body = lustre_msg_buf(request->rq_repmsg, 1);
357 body->valid = OBD_MD_FLNOTOBD;
360 body->atime = body->ctime = body->mtime = time;
361 body->uid = current->fsuid;
367 inode = iget4(dir->i_sb, body->ino, ll_find_inode, &lic);
370 CERROR("new_inode -fatal: rc %d\n", rc);
375 if (!list_empty(&inode->i_dentry)) {
376 CERROR("new_inode -fatal: inode %d, ct %d lnk %d\n",
377 body->ino, atomic_read(&inode->i_count),
381 inode = ERR_PTR(-EIO);
388 OBD_FREE(lmm, mds_md_size);
389 ptlrpc_req_finished(request);
393 static int ll_mdc_unlink(struct inode *dir, struct inode *child, __u32 mode,
394 const char *name, int len)
396 struct ptlrpc_request *request = NULL;
397 struct ll_sb_info *sbi = ll_i2sbi(dir);
402 err = mdc_unlink(&sbi->ll_mdc_conn, dir, child, mode, name, len,
404 ptlrpc_req_finished(request);
409 int ll_mdc_link(struct dentry *src, struct inode *dir,
410 const char *name, int len)
412 struct ptlrpc_request *request = NULL;
414 struct ll_sb_info *sbi = ll_i2sbi(dir);
418 err = mdc_link(&sbi->ll_mdc_conn, src, dir, name,
420 ptlrpc_req_finished(request);
425 int ll_mdc_rename(struct inode *src, struct inode *tgt,
426 struct dentry *old, struct dentry *new)
428 struct ptlrpc_request *request = NULL;
429 struct ll_sb_info *sbi = ll_i2sbi(src);
434 err = mdc_rename(&sbi->ll_mdc_conn, src, tgt,
435 old->d_name.name, old->d_name.len,
436 new->d_name.name, new->d_name.len, &request);
437 ptlrpc_req_finished(request);
443 * By the time this is called, we already have created
444 * the directory cache entry for the new file, but it
445 * is so far negative - it has no inode.
447 * If the create succeeds, we fill in the inode information
448 * with d_instantiate().
451 static int ll_create(struct inode *dir, struct dentry *dentry, int mode)
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 /* FIXME: we set the UID/GID fields to 0 for now, because it
463 * fixes a bug on the BA OSTs. We should really set
464 * them properly, and this needs to be revisited when
465 * we do proper credentials checking on the OST, and
466 * set the attributes on the OST in ll_inode_setattr().
467 * See also ll_file_open() and ll_lov_setstripe().
468 gid_t gid = current->fsgid;
470 if (dir->i_mode & S_ISGID)
472 rc = ll_create_objects(ll_i2obdconn(dir), 0, current->fsuid,
475 rc = ll_create_objects(dir->i_sb, 0, 0, 0, &lsm);
476 CDEBUG(D_DENTRY, "name %*s mode %o: rc = %d\n",
477 dentry->d_name.len, dentry->d_name.name, mode, rc);
482 inode = ll_create_node(dir, dentry->d_name.name, dentry->d_name.len,
483 NULL, 0, mode, 0, dentry->d_it, lsm);
487 CERROR("error creating MDS object for %*s: rc = %d\n",
488 dentry->d_name.len, dentry->d_name.name, rc);
489 GOTO(out_destroy, rc);
492 if (dentry->d_it->it_disposition) {
493 lli = ll_i2info(inode);
494 memcpy(&lli->lli_intent_lock_handle,
495 dentry->d_it->it_lock_handle,
496 sizeof(struct lustre_handle));
497 d_instantiate(dentry, inode);
499 /* no directory data updates when intents rule */
500 rc = ext2_add_nondir(dentry, inode);
511 oa->o_easize = lsm->lsm_mds_easize;
512 oa->o_id = lsm->lsm_object_id;
513 oa->o_valid |= OBD_MD_FLID | OBD_MD_FLEASIZE;
514 err = obd_destroy(ll_i2obdconn(dir), oa, lsm);
519 CERROR("error uncreating objid "LPX64": err %d\n",
520 lsm->lsm_object_id, err);
521 OBD_FREE(lsm, ll_ost_easize(dir->i_sb));
527 static int ll_mknod(struct inode *dir, struct dentry *dentry, int mode,
533 inode = ll_create_node(dir, dentry->d_name.name, dentry->d_name.len,
534 NULL, 0, mode, rdev, dentry->d_it, NULL);
537 RETURN(PTR_ERR(inode));
539 /* no directory data updates when intents rule */
540 if (dentry->d_it && dentry->d_it->it_disposition)
541 d_instantiate(dentry, inode);
543 err = ext2_add_nondir(dentry, inode);
548 static int ll_symlink(struct inode *dir, struct dentry *dentry,
551 unsigned l = strlen(symname);
553 struct ll_inode_info *lli;
557 CHECK_MOUNT_EPOCH(dir);
559 inode = ll_create_node(dir, dentry->d_name.name, dentry->d_name.len,
560 symname, l, S_IFLNK | S_IRWXUGO, 0,
563 RETURN(PTR_ERR(inode));
565 lli = ll_i2info(inode);
567 OBD_ALLOC(lli->lli_symlink_name, l + 1);
568 /* this _could_ be a non-fatal error, since the symlink is already
569 * stored on the MDS by this point, and we can re-get it in readlink.
571 if (!lli->lli_symlink_name)
574 memcpy(lli->lli_symlink_name, symname, l + 1);
577 /* no directory data updates when intents rule */
578 if (dentry->d_it && dentry->d_it->it_disposition)
579 d_instantiate(dentry, inode);
581 err = ext2_add_nondir(dentry, inode);
586 static int ll_link(struct dentry *old_dentry, struct inode * dir,
587 struct dentry *dentry)
590 struct inode *inode = old_dentry->d_inode;
592 if (dentry->d_it && dentry->d_it->it_disposition) {
593 int err = dentry->d_it->it_status;
596 inode->i_ctime = CURRENT_TIME;
597 ext2_inc_count(inode);
598 atomic_inc(&inode->i_count);
599 d_instantiate(dentry, inode);
600 invalidate_inode_pages(dir);
604 if (S_ISDIR(inode->i_mode))
607 if (inode->i_nlink >= EXT2_LINK_MAX)
610 err = ll_mdc_link(old_dentry, dir,
611 dentry->d_name.name, dentry->d_name.len);
615 inode->i_ctime = CURRENT_TIME;
616 ext2_inc_count(inode);
617 atomic_inc(&inode->i_count);
619 return ext2_add_nondir(dentry, inode);
622 static int ll_mkdir(struct inode * dir, struct dentry * dentry, int mode)
624 struct inode * inode;
628 if (dir->i_nlink >= EXT2_LINK_MAX)
633 inode = ll_create_node(dir, dentry->d_name.name, dentry->d_name.len,
634 NULL, 0, S_IFDIR | mode, 0, dentry->d_it, NULL);
635 err = PTR_ERR(inode);
639 ext2_inc_count(inode);
641 err = ext2_make_empty(inode, dir);
645 /* no directory data updates when intents rule */
646 if (dentry->d_it->it_disposition == 0) {
647 err = ll_add_link(dentry, inode);
652 d_instantiate(dentry, inode);
658 ext2_dec_count(inode);
659 ext2_dec_count(inode);
668 static int ll_common_unlink(struct inode *dir, struct dentry *dentry,
671 struct inode * inode = dentry->d_inode;
673 struct ext2_dir_entry_2 * de;
678 if (dentry->d_it && dentry->d_it->it_disposition) {
679 err = dentry->d_it->it_status;
680 invalidate_inode_pages(dir);
685 de = ext2_find_entry(dir, dentry, &page);
689 err = ll_mdc_unlink(dir, dentry->d_inode, mode,
690 dentry->d_name.name, dentry->d_name.len);
695 err = ext2_delete_entry(de, page);
699 invalidate_inode_pages(dir);
701 inode->i_ctime = dir->i_ctime;
703 ext2_dec_count(inode);
707 static int ll_unlink(struct inode *dir, struct dentry *dentry)
709 return ll_common_unlink(dir, dentry, S_IFREG);
712 static int ll_rmdir(struct inode *dir, struct dentry *dentry)
714 struct inode * inode = dentry->d_inode;
718 if (!dentry->d_it || dentry->d_it->it_disposition == 0) {
719 if (!ext2_empty_dir(inode))
721 err = ll_common_unlink(dir, dentry, S_IFDIR);
723 err = dentry->d_it->it_status;
727 ext2_dec_count(inode);
732 static int ll_rename(struct inode * old_dir, struct dentry * old_dentry,
733 struct inode * new_dir, struct dentry * new_dentry)
735 struct inode * old_inode = old_dentry->d_inode;
736 struct inode * tgt_inode = new_dentry->d_inode;
737 struct page * dir_page = NULL;
738 struct ext2_dir_entry_2 * dir_de = NULL;
739 struct ext2_dir_entry_2 * old_de;
740 struct page * old_page;
743 if (new_dentry->d_it && new_dentry->d_it->it_disposition) {
745 tgt_inode->i_ctime = CURRENT_TIME;
746 tgt_inode->i_nlink--;
748 invalidate_inode_pages(old_dir);
749 invalidate_inode_pages(new_dir);
750 GOTO(out, err = new_dentry->d_it->it_status);
753 err = ll_mdc_rename(old_dir, new_dir, old_dentry, new_dentry);
757 old_de = ext2_find_entry (old_dir, old_dentry, &old_page);
761 if (S_ISDIR(old_inode->i_mode)) {
763 dir_de = ext2_dotdot(old_inode, &dir_page);
769 struct page *new_page;
770 struct ext2_dir_entry_2 *new_de;
773 if (dir_de && !ext2_empty_dir (tgt_inode))
777 new_de = ext2_find_entry (new_dir, new_dentry, &new_page);
780 ext2_inc_count(old_inode);
781 ext2_set_link(new_dir, new_de, new_page, old_inode);
782 tgt_inode->i_ctime = CURRENT_TIME;
784 tgt_inode->i_nlink--;
785 ext2_dec_count(tgt_inode);
789 if (new_dir->i_nlink >= EXT2_LINK_MAX)
792 ext2_inc_count(old_inode);
793 err = ll_add_link(new_dentry, old_inode);
795 ext2_dec_count(old_inode);
799 ext2_inc_count(new_dir);
802 ext2_delete_entry (old_de, old_page);
803 ext2_dec_count(old_inode);
806 ext2_set_link(old_inode, dir_de, dir_page, new_dir);
807 ext2_dec_count(old_dir);
814 page_cache_release(dir_page);
818 page_cache_release(old_page);
823 struct inode_operations ll_dir_inode_operations = {