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 rc = ll_lock(dir, dentry, it, &lockh);
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 source lookup to succeed */
204 GOTO(drop_req, rc = it->it_status);
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 /* This means the target lookup is negative */
223 if (body->valid == 0)
228 /* Do a getattr now that we have the lock */
229 valid = OBD_MD_FLNOTOBD | OBD_MD_FLEASIZE;
230 if (it->it_op == IT_READLINK) {
231 valid |= OBD_MD_LINKNAME;
232 symlen = lic.lic_body->size;
234 ptlrpc_req_finished(request);
236 rc = mdc_getattr(&sbi->ll_mdc_conn, ino, mode,
237 valid, symlen, &request);
239 CERROR("failure %d inode "LPX64"\n", rc, ino);
240 GOTO(drop_req, rc = -abs(rc));
244 struct ll_inode_info *lli = ll_i2info(dir);
247 memcpy(&lli->lli_intent_lock_handle, &lockh, sizeof(lockh));
250 ino = ll_inode_by_name(dir, dentry, &mode);
252 CERROR("inode %*s not found by name\n",
253 dentry->d_name.len, dentry->d_name.name);
254 GOTO(drop_lock, rc = -ENOENT);
257 rc = mdc_getattr(&sbi->ll_mdc_conn, ino, mode,
258 OBD_MD_FLNOTOBD|OBD_MD_FLEASIZE, 0, &request);
260 CERROR("failure %d inode "LPX64"\n", rc, ino);
261 GOTO(drop_req, rc = -abs(rc));
266 lic.lic_body = lustre_msg_buf(request->rq_repmsg, offset);
267 if (S_ISREG(lic.lic_body->mode) &&
268 lic.lic_body->valid & OBD_MD_FLEASIZE) {
269 LASSERT(request->rq_repmsg->bufcount > offset);
270 lic.lic_lmm = lustre_msg_buf(request->rq_repmsg, offset + 1);
274 /* No rpc's happen during iget4, -ENOMEM's are possible */
276 inode = iget4(dir->i_sb, ino, ll_find_inode, &lic);
279 ptlrpc_free_req(request);
280 ll_intent_release(dentry);
281 RETURN(ERR_PTR(-ENOMEM));
286 ptlrpc_req_finished(request);
288 dentry->d_op = &ll_d_ops;
289 d_add(dentry, inode);
291 if (ll_d2d(dentry) == NULL)
293 // down(&ll_d2d(dentry)->lld_it_sem);
294 // dentry->d_it = it;
296 if (it->it_op == IT_LOOKUP)
297 ll_intent_release(dentry);
302 ptlrpc_free_req(request);
304 #warning FIXME: must release lock here
308 static struct inode *ll_create_node(struct inode *dir, const char *name,
309 int namelen, const char *tgt, int tgtlen,
310 int mode, __u64 extra,
311 struct lookup_intent *it,
312 struct lov_stripe_md *lsm)
315 struct ptlrpc_request *request = NULL;
316 struct mds_body *body;
317 time_t time = CURRENT_TIME;
318 struct ll_sb_info *sbi = ll_i2sbi(dir);
319 struct ll_read_inode2_cookie lic;
320 struct lov_mds_md *lmm = NULL;
323 if (it && it->it_disposition) {
324 int rc = it->it_status;
326 CERROR("error creating MDS inode for %*s: rc = %d\n",
330 invalidate_inode_pages(dir);
331 request = it->it_data;
332 body = lustre_msg_buf(request->rq_repmsg, 1);
335 int gid = current->fsgid;
339 OBD_ALLOC(lmm, lsm->lsm_mds_easize);
341 RETURN(ERR_PTR(-ENOMEM));
342 lov_packmd(lmm, lsm);
347 if (dir->i_mode & S_ISGID) {
353 rc = mdc_create(&sbi->ll_mdc_conn, dir, name, namelen, tgt,
354 tgtlen, mode, current->fsuid, gid,
355 time, extra, lsm, &request);
360 body = lustre_msg_buf(request->rq_repmsg, 0);
365 LASSERT(body->ino != 0);
366 inode = iget4(dir->i_sb, body->ino, ll_find_inode, &lic);
368 int rc = PTR_ERR(inode);
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, lsm->lsm_mds_easize);
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 the directory cache
443 * entry for the new file, but it is so far negative - it has no inode.
444 * We defer creating the OBD object(s) until open, to keep the intent and
445 * non-intent code paths similar, and also because we do not have the MDS
446 * inode number before calling ll_create_node() (which is needed for LOV),
447 * so we would need to do yet another RPC to the MDS to store the LOV EA
450 * If the create succeeds, we fill in the inode information
451 * with d_instantiate().
453 static int ll_create(struct inode *dir, struct dentry *dentry, int mode)
455 struct lookup_intent *it = dentry->d_it;
460 CHECK_MOUNT_EPOCH(dir);
462 inode = ll_create_node(dir, dentry->d_name.name, dentry->d_name.len,
463 NULL, 0, mode, 0, it, NULL);
466 RETURN(PTR_ERR(inode));
468 if (it->it_disposition) {
469 struct ll_inode_info *lli = ll_i2info(inode);
470 memcpy(&lli->lli_intent_lock_handle, it->it_lock_handle,
471 sizeof(struct lustre_handle));
472 d_instantiate(dentry, inode);
474 /* no directory data updates when intents rule */
475 rc = ext2_add_nondir(dentry, inode);
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 CHECK_MOUNT_EPOCH(dir);
513 inode = ll_create_node(dir, dentry->d_name.name, dentry->d_name.len,
514 symname, l, S_IFLNK | S_IRWXUGO, 0,
517 RETURN(PTR_ERR(inode));
519 lli = ll_i2info(inode);
521 OBD_ALLOC(lli->lli_symlink_name, l + 1);
522 /* this _could_ be a non-fatal error, since the symlink is already
523 * stored on the MDS by this point, and we can re-get it in readlink.
525 if (!lli->lli_symlink_name)
528 memcpy(lli->lli_symlink_name, symname, l + 1);
531 /* no directory data updates when intents rule */
532 if (dentry->d_it && dentry->d_it->it_disposition)
533 d_instantiate(dentry, inode);
535 err = ext2_add_nondir(dentry, inode);
540 static int ll_link(struct dentry *old_dentry, struct inode * dir,
541 struct dentry *dentry)
543 struct inode *inode = old_dentry->d_inode;
544 struct lookup_intent *it = dentry->d_it;
547 if (it && it->it_disposition) {
549 RETURN(it->it_status);
550 inode->i_ctime = CURRENT_TIME;
551 ext2_inc_count(inode);
552 atomic_inc(&inode->i_count);
553 d_instantiate(dentry, inode);
554 invalidate_inode_pages(dir);
558 if (S_ISDIR(inode->i_mode))
561 if (inode->i_nlink >= EXT2_LINK_MAX)
564 rc = ll_mdc_link(old_dentry, dir,
565 dentry->d_name.name, dentry->d_name.len);
569 inode->i_ctime = CURRENT_TIME;
570 ext2_inc_count(inode);
571 atomic_inc(&inode->i_count);
573 return ext2_add_nondir(dentry, inode);
576 static int ll_mkdir(struct inode *dir, struct dentry *dentry, int mode)
578 struct inode * inode;
582 if (dir->i_nlink >= EXT2_LINK_MAX)
587 inode = ll_create_node(dir, dentry->d_name.name, dentry->d_name.len,
588 NULL, 0, S_IFDIR | mode, 0, dentry->d_it, NULL);
589 err = PTR_ERR(inode);
593 ext2_inc_count(inode);
595 err = ext2_make_empty(inode, dir);
599 /* no directory data updates when intents rule */
600 if (dentry->d_it->it_disposition == 0) {
601 err = ll_add_link(dentry, inode);
606 d_instantiate(dentry, inode);
612 ext2_dec_count(inode);
613 ext2_dec_count(inode);
622 static int ll_common_unlink(struct inode *dir, struct dentry *dentry,__u32 mode)
624 struct lookup_intent *it = dentry->d_it;
625 struct inode *inode = dentry->d_inode;
627 struct ext2_dir_entry_2 * de;
632 if (it && it->it_disposition) {
634 invalidate_inode_pages(dir);
641 de = ext2_find_entry(dir, dentry, &page);
645 rc = ll_mdc_unlink(dir, dentry->d_inode, mode,
646 dentry->d_name.name, dentry->d_name.len);
651 err = ext2_delete_entry(de, page);
656 /* AED: not sure if needed - directory lock revocation should do it
657 * in the case where the client has cached it for non-intent ops.
659 invalidate_inode_pages(dir);
661 inode->i_ctime = dir->i_ctime;
663 ext2_dec_count(inode);
668 static int ll_unlink(struct inode *dir, struct dentry *dentry)
670 return ll_common_unlink(dir, dentry, S_IFREG);
673 static int ll_rmdir(struct inode *dir, struct dentry *dentry)
675 struct inode * inode = dentry->d_inode;
676 struct lookup_intent *it = dentry->d_it;
680 if ((!it || !it->it_disposition) && !ext2_empty_dir(inode))
683 rc = ll_common_unlink(dir, dentry, S_IFDIR);
686 ext2_dec_count(inode);
693 static int ll_rename(struct inode * old_dir, struct dentry * old_dentry,
694 struct inode * new_dir, struct dentry * new_dentry)
696 struct lookup_intent *it = new_dentry->d_it;
697 struct inode * old_inode = old_dentry->d_inode;
698 struct inode * tgt_inode = new_dentry->d_inode;
699 struct page * dir_page = NULL;
700 struct ext2_dir_entry_2 * dir_de = NULL;
701 struct ext2_dir_entry_2 * old_de;
702 struct page * old_page;
705 if (it && it->it_disposition) {
707 tgt_inode->i_ctime = CURRENT_TIME;
708 tgt_inode->i_nlink--;
710 invalidate_inode_pages(old_dir);
711 invalidate_inode_pages(new_dir);
712 GOTO(out, err = it->it_status);
715 err = ll_mdc_rename(old_dir, new_dir, old_dentry, new_dentry);
719 old_de = ext2_find_entry (old_dir, old_dentry, &old_page);
723 if (S_ISDIR(old_inode->i_mode)) {
725 dir_de = ext2_dotdot(old_inode, &dir_page);
731 struct page *new_page;
732 struct ext2_dir_entry_2 *new_de;
735 if (dir_de && !ext2_empty_dir (tgt_inode))
739 new_de = ext2_find_entry (new_dir, new_dentry, &new_page);
742 ext2_inc_count(old_inode);
743 ext2_set_link(new_dir, new_de, new_page, old_inode);
744 tgt_inode->i_ctime = CURRENT_TIME;
746 tgt_inode->i_nlink--;
747 ext2_dec_count(tgt_inode);
751 if (new_dir->i_nlink >= EXT2_LINK_MAX)
754 ext2_inc_count(old_inode);
755 err = ll_add_link(new_dentry, old_inode);
757 ext2_dec_count(old_inode);
761 ext2_inc_count(new_dir);
764 ext2_delete_entry (old_de, old_page);
765 ext2_dec_count(old_inode);
768 ext2_set_link(old_inode, dir_de, dir_page, new_dir);
769 ext2_dec_count(old_dir);
776 page_cache_release(dir_page);
780 page_cache_release(old_page);
785 struct inode_operations ll_dir_inode_operations = {