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/smp_lock.h>
32 #include <linux/highmem.h>
33 #include <linux/pagemap.h>
35 #define DEBUG_SUBSYSTEM S_LLITE
37 #include <linux/obd_support.h>
38 #include <linux/lustre_lite.h>
39 #include <linux/lustre_dlm.h>
40 #include <linux/obd_lov.h>
42 extern struct address_space_operations ll_aops;
45 extern void ll_change_inode(struct inode *inode);
46 extern int ll_setattr(struct dentry *de, struct iattr *attr);
49 extern int ll_add_link (struct dentry *dentry, struct inode *inode);
50 obd_id ll_inode_by_name(struct inode * dir, struct dentry *dentry, int *typ);
51 int ext2_make_empty(struct inode *inode, struct inode *parent);
52 struct ext2_dir_entry_2 * ext2_find_entry (struct inode * dir,
53 struct dentry *dentry, struct page ** res_page);
54 int ext2_delete_entry (struct ext2_dir_entry_2 * dir, struct page * page );
55 int ext2_empty_dir (struct inode * inode);
56 struct ext2_dir_entry_2 * ext2_dotdot (struct inode *dir, struct page **p);
57 void ext2_set_link(struct inode *dir, struct ext2_dir_entry_2 *de,
58 struct page *page, struct inode *inode);
61 * Couple of helper functions - make the code slightly cleaner.
63 static inline void ext2_inc_count(struct inode *inode)
68 /* postpone the disk update until the inode really goes away */
69 static inline void ext2_dec_count(struct inode *inode)
74 static inline int ext2_add_nondir(struct dentry *dentry, struct inode *inode)
77 err = ll_add_link(dentry, inode);
79 d_instantiate(dentry, inode);
82 ext2_dec_count(inode);
88 static int ll_find_inode(struct inode *inode, unsigned long ino, void *opaque)
90 struct ll_read_inode2_cookie *lic = opaque;
91 struct mds_body *body = lic->lic_body;
93 if (inode->i_generation != lic->lic_body->generation)
96 /* Apply the attributes in 'opaque' to this inode */
97 ll_update_inode(inode, body);
102 extern struct dentry_operations ll_d_ops;
104 int ll_lock(struct inode *dir, struct dentry *dentry,
105 struct lookup_intent *it, struct lustre_handle *lockh)
107 struct ll_sb_info *sbi = ll_i2sbi(dir);
112 /* CREAT needs to be tested before open (both could be set) */
113 if ((it->it_op & (IT_CREAT | IT_MKDIR | IT_SETATTR | IT_MKNOD))) {
115 } else if (it->it_op & (IT_READDIR | IT_GETATTR | IT_OPEN | IT_UNLINK |
116 IT_RMDIR | IT_RENAME | IT_RENAME2 | IT_READLINK|
117 IT_LINK | IT_LINK2 | IT_LOOKUP)) {
118 /* XXXphil PW for LINK2/RENAME2? */
120 } else if (it->it_op & IT_SYMLINK) {
123 tgtlen = strlen(tgt);
130 err = mdc_enqueue(&sbi->ll_mdc_conn, LDLM_MDSINTENT, it, lock_mode,
131 dir, dentry, lockh, tgt, tgtlen, dir, sizeof(*dir));
136 int ll_unlock(__u32 mode, struct lustre_handle *lockh)
140 ldlm_lock_decref(lockh, mode);
145 static struct dentry *ll_lookup2(struct inode *dir, struct dentry *dentry,
146 struct lookup_intent *it)
148 struct ptlrpc_request *request = NULL;
149 struct inode * inode = NULL;
150 struct ll_sb_info *sbi = ll_i2sbi(dir);
151 struct ll_read_inode2_cookie lic;
152 struct lustre_handle lockh;
153 struct lookup_intent lookup_it = { IT_LOOKUP };
159 /* CHECK_MOUNT_EPOCH(dir); */
160 if (ll_i2info(dir)->lli_mount_epoch != ll_i2sbi(dir)->ll_mount_epoch) {
162 RETURN(ERR_PTR(-EIO));
170 CDEBUG(D_INFO, "name: %*s, intent: %s\n", dentry->d_name.len,
171 dentry->d_name.name, ldlm_it2str(it->it_op));
173 if (dentry->d_name.len > EXT2_NAME_LEN)
174 RETURN(ERR_PTR(-ENAMETOOLONG));
176 rc = ll_lock(dir, dentry, it, &lockh);
179 memcpy(it->it_lock_handle, &lockh, sizeof(lockh));
181 request = (struct ptlrpc_request *)it->it_data;
182 if (it->it_disposition) {
183 int mode, symlen = 0;
187 lic.lic_body = lustre_msg_buf(request->rq_repmsg, offset);
188 ino = lic.lic_body->fid1.id;
189 mode = lic.lic_body->mode;
190 if (it->it_op & (IT_CREAT | IT_MKDIR | IT_SYMLINK | IT_MKNOD)) {
191 /* For create ops, we want the lookup to be negative,
192 * unless the create failed in a way that indicates
193 * that the file is already there */
194 if (it->it_status != -EEXIST)
195 GOTO(negative, NULL);
196 } else if (it->it_op & (IT_GETATTR | IT_SETATTR | IT_LOOKUP |
198 /* For check ops, we want the lookup to succeed */
202 } else if (it->it_op & (IT_RENAME | IT_LINK)) {
203 /* For rename, we want the source lookup to succeed */
206 GOTO(drop_req, rc = it->it_status);
208 it->it_data = dentry;
209 } else if (it->it_op & (IT_UNLINK | IT_RMDIR)) {
210 /* For remove ops, we want the lookup to succeed unless
211 * the file truly doesn't exist */
213 if (it->it_status == -ENOENT)
216 } else if (it->it_op == IT_OPEN) {
218 if (it->it_status && it->it_status != -EEXIST)
220 } else if (it->it_op & (IT_RENAME2 | IT_LINK2)) {
221 struct mds_body *body =
222 lustre_msg_buf(request->rq_repmsg, offset);
224 /* This means the target lookup is negative */
225 if (body->valid == 0)
230 /* Do a getattr now that we have the lock */
231 valid = OBD_MD_FLNOTOBD | OBD_MD_FLEASIZE;
232 if (it->it_op == IT_READLINK) {
233 valid |= OBD_MD_LINKNAME;
234 symlen = lic.lic_body->size;
236 ptlrpc_req_finished(request);
238 rc = mdc_getattr(&sbi->ll_mdc_conn, ino, mode,
239 valid, symlen, &request);
241 CERROR("failure %d inode "LPX64"\n", rc, ino);
242 GOTO(drop_req, rc = -abs(rc));
246 struct ll_inode_info *lli = ll_i2info(dir);
249 memcpy(&lli->lli_intent_lock_handle, &lockh, sizeof(lockh));
252 ino = ll_inode_by_name(dir, dentry, &mode);
254 CERROR("inode %*s not found by name\n",
255 dentry->d_name.len, dentry->d_name.name);
256 GOTO(drop_lock, rc = -ENOENT);
259 rc = mdc_getattr(&sbi->ll_mdc_conn, ino, mode,
260 OBD_MD_FLNOTOBD|OBD_MD_FLEASIZE, 0, &request);
262 CERROR("failure %d inode "LPX64"\n", rc, ino);
263 GOTO(drop_req, rc = -abs(rc));
268 lic.lic_body = lustre_msg_buf(request->rq_repmsg, offset);
269 if (S_ISREG(lic.lic_body->mode) &&
270 lic.lic_body->valid & OBD_MD_FLEASIZE) {
271 LASSERT(request->rq_repmsg->bufcount > offset);
272 lic.lic_lmm = lustre_msg_buf(request->rq_repmsg, offset + 1);
276 /* No rpc's happen during iget4, -ENOMEM's are possible */
278 inode = iget4(dir->i_sb, ino, ll_find_inode, &lic);
281 ptlrpc_free_req(request);
282 ll_intent_release(dentry);
283 RETURN(ERR_PTR(-ENOMEM));
288 ptlrpc_req_finished(request);
290 dentry->d_op = &ll_d_ops;
291 d_add(dentry, inode);
293 if (ll_d2d(dentry) == NULL)
295 // down(&ll_d2d(dentry)->lld_it_sem);
296 // dentry->d_it = it;
298 if (it->it_op == IT_LOOKUP)
299 ll_intent_release(dentry);
304 ptlrpc_free_req(request);
306 #warning FIXME: must release lock here
310 static struct inode *ll_create_node(struct inode *dir, const char *name,
311 int namelen, const char *tgt, int tgtlen,
312 int mode, __u64 extra,
313 struct lookup_intent *it,
314 struct lov_stripe_md *lsm)
317 struct ptlrpc_request *request = NULL;
318 struct mds_body *body;
319 time_t time = CURRENT_TIME;
320 struct ll_sb_info *sbi = ll_i2sbi(dir);
321 struct ll_read_inode2_cookie lic;
322 struct lov_mds_md *lmm = NULL;
325 if (it && it->it_disposition) {
326 int rc = it->it_status;
328 CERROR("error creating MDS inode for %*s: rc = %d\n",
332 invalidate_inode_pages(dir);
333 request = it->it_data;
334 body = lustre_msg_buf(request->rq_repmsg, 1);
337 int gid = current->fsgid;
341 OBD_ALLOC(lmm, lsm->lsm_mds_easize);
343 RETURN(ERR_PTR(-ENOMEM));
344 lov_packmd(lmm, lsm);
349 if (dir->i_mode & S_ISGID) {
355 rc = mdc_create(&sbi->ll_mdc_conn, dir, name, namelen, tgt,
356 tgtlen, mode, current->fsuid, gid,
357 time, extra, lsm, &request);
362 body = lustre_msg_buf(request->rq_repmsg, 0);
367 LASSERT(body->ino != 0);
368 inode = iget4(dir->i_sb, body->ino, ll_find_inode, &lic);
370 int rc = PTR_ERR(inode);
371 CERROR("new_inode -fatal: rc %d\n", rc);
376 if (!list_empty(&inode->i_dentry)) {
377 CERROR("new_inode -fatal: inode %d, ct %d lnk %d\n",
378 body->ino, atomic_read(&inode->i_count),
382 inode = ERR_PTR(-EIO);
389 OBD_FREE(lmm, lsm->lsm_mds_easize);
390 ptlrpc_req_finished(request);
394 static int ll_mdc_unlink(struct inode *dir, struct inode *child, __u32 mode,
395 const char *name, int len)
397 struct ptlrpc_request *request = NULL;
398 struct ll_sb_info *sbi = ll_i2sbi(dir);
403 err = mdc_unlink(&sbi->ll_mdc_conn, dir, child, mode, name, len,
405 ptlrpc_req_finished(request);
410 int ll_mdc_link(struct dentry *src, struct inode *dir,
411 const char *name, int len)
413 struct ptlrpc_request *request = NULL;
415 struct ll_sb_info *sbi = ll_i2sbi(dir);
419 err = mdc_link(&sbi->ll_mdc_conn, src, dir, name, len, &request);
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 the directory cache
444 * entry for the new file, but it is so far negative - it has no inode.
445 * We defer creating the OBD object(s) until open, to keep the intent and
446 * non-intent code paths similar, and also because we do not have the MDS
447 * inode number before calling ll_create_node() (which is needed for LOV),
448 * so we would need to do yet another RPC to the MDS to store the LOV EA
451 * If the create succeeds, we fill in the inode information
452 * with d_instantiate().
454 static int ll_create(struct inode *dir, struct dentry *dentry, int mode)
456 struct lookup_intent *it;
461 CHECK_MOUNT_EPOCH(dir);
465 inode = ll_create_node(dir, dentry->d_name.name, dentry->d_name.len,
466 NULL, 0, mode, 0, it, NULL);
469 RETURN(PTR_ERR(inode));
471 if (it->it_disposition) {
472 struct ll_inode_info *lli = ll_i2info(inode);
473 memcpy(&lli->lli_intent_lock_handle, it->it_lock_handle,
474 sizeof(lli->lli_intent_lock_handle));
475 d_instantiate(dentry, inode);
477 /* no directory data updates when intents rule */
478 rc = ext2_add_nondir(dentry, inode);
484 static int ll_mknod(struct inode *dir, struct dentry *dentry, int mode,
487 struct lookup_intent *it;
493 inode = ll_create_node(dir, dentry->d_name.name, dentry->d_name.len,
494 NULL, 0, mode, rdev, it, NULL);
497 RETURN(PTR_ERR(inode));
499 /* no directory data updates when intents rule */
500 if (it && it->it_disposition)
501 d_instantiate(dentry, inode);
503 rc = ext2_add_nondir(dentry, inode);
508 static int ll_symlink(struct inode *dir, struct dentry *dentry,
511 struct lookup_intent *it;
512 unsigned l = strlen(symname);
514 struct ll_inode_info *lli;
518 CHECK_MOUNT_EPOCH(dir);
522 inode = ll_create_node(dir, dentry->d_name.name, dentry->d_name.len,
523 symname, l, S_IFLNK | S_IRWXUGO, 0, it, NULL);
525 RETURN(PTR_ERR(inode));
527 lli = ll_i2info(inode);
529 OBD_ALLOC(lli->lli_symlink_name, l + 1);
530 /* this _could_ be a non-fatal error, since the symlink is already
531 * stored on the MDS by this point, and we can re-get it in readlink.
533 if (!lli->lli_symlink_name)
536 memcpy(lli->lli_symlink_name, symname, l + 1);
539 /* no directory data updates when intents rule */
540 if (it && it->it_disposition)
541 d_instantiate(dentry, inode);
543 err = ext2_add_nondir(dentry, inode);
548 static int ll_link(struct dentry *old_dentry, struct inode * dir,
549 struct dentry *dentry)
551 struct lookup_intent *it;
552 struct inode *inode = old_dentry->d_inode;
557 if (it && it->it_disposition) {
559 RETURN(it->it_status);
560 inode->i_ctime = CURRENT_TIME;
561 ext2_inc_count(inode);
562 atomic_inc(&inode->i_count);
563 d_instantiate(dentry, inode);
564 invalidate_inode_pages(dir);
568 if (S_ISDIR(inode->i_mode))
571 if (inode->i_nlink >= EXT2_LINK_MAX)
574 rc = ll_mdc_link(old_dentry, dir,
575 dentry->d_name.name, dentry->d_name.len);
579 inode->i_ctime = CURRENT_TIME;
580 ext2_inc_count(inode);
581 atomic_inc(&inode->i_count);
583 return ext2_add_nondir(dentry, inode);
586 static int ll_mkdir(struct inode *dir, struct dentry *dentry, int mode)
588 struct lookup_intent *it;
589 struct inode * inode;
593 if (dir->i_nlink >= EXT2_LINK_MAX)
600 inode = ll_create_node(dir, dentry->d_name.name, dentry->d_name.len,
601 NULL, 0, S_IFDIR | mode, 0, it, NULL);
602 err = PTR_ERR(inode);
606 ext2_inc_count(inode);
608 err = ext2_make_empty(inode, dir);
612 /* no directory data updates when intents rule */
613 if (!it || !it->it_disposition) {
614 err = ll_add_link(dentry, inode);
619 d_instantiate(dentry, inode);
625 ext2_dec_count(inode);
626 ext2_dec_count(inode);
635 static int ll_common_unlink(struct inode *dir, struct dentry *dentry,
636 struct lookup_intent *it, __u32 mode)
638 struct inode *inode = dentry->d_inode;
639 struct ext2_dir_entry_2 * de;
643 if (it && it->it_disposition) {
645 invalidate_inode_pages(dir);
651 de = ext2_find_entry(dir, dentry, &page);
653 GOTO(out, rc = -ENOENT);
654 rc = ll_mdc_unlink(dir, dentry->d_inode, mode,
655 dentry->d_name.name, dentry->d_name.len);
659 rc = ext2_delete_entry(de, page);
663 /* AED: not sure if needed - directory lock revocation should do it
664 * in the case where the client has cached it for non-intent ops.
666 invalidate_inode_pages(dir);
668 inode->i_ctime = dir->i_ctime;
670 ext2_dec_count(inode);
675 static int ll_unlink(struct inode *dir, struct dentry *dentry)
677 struct lookup_intent *it;
681 return ll_common_unlink(dir, dentry, it, S_IFREG);
684 static int ll_rmdir(struct inode *dir, struct dentry *dentry)
686 struct inode * inode = dentry->d_inode;
687 struct lookup_intent *it;
693 if ((!it || !it->it_disposition) && !ext2_empty_dir(inode))
696 rc = ll_common_unlink(dir, dentry, it, S_IFDIR);
699 ext2_dec_count(inode);
706 static int ll_rename(struct inode * old_dir, struct dentry * old_dentry,
707 struct inode * new_dir, struct dentry * new_dentry)
709 struct lookup_intent *it;
710 struct inode * old_inode = old_dentry->d_inode;
711 struct inode * tgt_inode = new_dentry->d_inode;
712 struct page * dir_page = NULL;
713 struct ext2_dir_entry_2 * dir_de = NULL;
714 struct ext2_dir_entry_2 * old_de;
715 struct page * old_page;
718 it = new_dentry->d_it;
720 if (it && it->it_disposition) {
722 tgt_inode->i_ctime = CURRENT_TIME;
723 tgt_inode->i_nlink--;
725 invalidate_inode_pages(old_dir);
726 invalidate_inode_pages(new_dir);
727 GOTO(out, err = it->it_status);
730 err = ll_mdc_rename(old_dir, new_dir, old_dentry, new_dentry);
734 old_de = ext2_find_entry (old_dir, old_dentry, &old_page);
738 if (S_ISDIR(old_inode->i_mode)) {
740 dir_de = ext2_dotdot(old_inode, &dir_page);
746 struct page *new_page;
747 struct ext2_dir_entry_2 *new_de;
750 if (dir_de && !ext2_empty_dir (tgt_inode))
754 new_de = ext2_find_entry (new_dir, new_dentry, &new_page);
757 ext2_inc_count(old_inode);
758 ext2_set_link(new_dir, new_de, new_page, old_inode);
759 tgt_inode->i_ctime = CURRENT_TIME;
761 tgt_inode->i_nlink--;
762 ext2_dec_count(tgt_inode);
766 if (new_dir->i_nlink >= EXT2_LINK_MAX)
769 ext2_inc_count(old_inode);
770 err = ll_add_link(new_dentry, old_inode);
772 ext2_dec_count(old_inode);
776 ext2_inc_count(new_dir);
779 ext2_delete_entry (old_de, old_page);
780 ext2_dec_count(old_inode);
783 ext2_set_link(old_inode, dir_de, dir_page, new_dir);
784 ext2_dec_count(old_dir);
791 page_cache_release(dir_page);
795 page_cache_release(old_page);
800 struct inode_operations ll_dir_inode_operations = {