1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
4 * Copyright (c) 2002, 2003 Cluster File Systems, Inc.
6 * This file is part of Lustre, http://www.lustre.org.
8 * Lustre is free software; you can redistribute it and/or
9 * modify it under the terms of version 2 of the GNU General Public
10 * License as published by the Free Software Foundation.
12 * Lustre is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 * GNU General Public License for more details.
17 * You should have received a copy of the GNU General Public License
18 * along with Lustre; if not, write to the Free Software
19 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
23 #include <linux/sched.h>
25 #include <linux/smp_lock.h>
26 #include <linux/quotaops.h>
27 #include <linux/highmem.h>
28 #include <linux/pagemap.h>
30 #define DEBUG_SUBSYSTEM S_LLITE
32 #include <obd_support.h>
33 #include <lustre_fid.h>
34 #include <lustre_lite.h>
35 #include <lustre_dlm.h>
36 #include <lustre_ver.h>
37 #include <lustre_mdc.h>
38 #include "llite_internal.h"
41 extern struct dentry_operations ll_d_ops;
44 * Check if we have something mounted at the named dchild.
45 * In such a case there would always be dentry present.
47 static int ll_d_mountpoint(struct dentry *dparent, struct dentry *dchild,
52 if (unlikely(dchild)) {
53 mounted = d_mountpoint(dchild);
55 dchild = d_lookup(dparent, name);
57 mounted = d_mountpoint(dchild);
64 int ll_unlock(__u32 mode, struct lustre_handle *lockh)
68 ldlm_lock_decref(lockh, mode);
74 /* called from iget5_locked->find_inode() under inode_lock spinlock */
75 static int ll_test_inode(struct inode *inode, void *opaque)
77 struct ll_inode_info *lli = ll_i2info(inode);
78 struct lustre_md *md = opaque;
80 if (unlikely(!(md->body->valid & OBD_MD_FLID))) {
81 CERROR("MDS body missing FID\n");
85 if (!lu_fid_eq(&lli->lli_fid, &md->body->fid1))
91 static int ll_set_inode(struct inode *inode, void *opaque)
98 * Get an inode by inode number (already instantiated by the intent lookup).
99 * Returns inode or NULL
101 struct inode *ll_iget(struct super_block *sb, ino_t hash,
102 struct lustre_md *md)
104 struct ll_inode_info *lli;
109 inode = iget5_locked(sb, hash, ll_test_inode, ll_set_inode, md);
112 lli = ll_i2info(inode);
113 if (inode->i_state & I_NEW) {
114 ll_read_inode2(inode, md);
115 unlock_new_inode(inode);
117 if (!(inode->i_state & (I_FREEING | I_CLEAR)))
118 ll_update_inode(inode, md);
120 CDEBUG(D_VFSTRACE, "got inode: %lu/%u(%p) for "DFID"\n",
121 inode->i_ino, inode->i_generation, inode,
122 PFID(&lli->lli_fid));
128 static void ll_drop_negative_dentry(struct inode *dir)
130 struct dentry *dentry, *tmp_alias, *tmp_subdir;
132 spin_lock(&dcache_lock);
134 list_for_each_entry_safe(dentry, tmp_alias,
135 &dir->i_dentry,d_alias) {
136 if (!list_empty(&dentry->d_subdirs)) {
137 struct dentry *child;
138 list_for_each_entry_safe(child, tmp_subdir,
141 /* XXX Print some debug here? */
143 /* Negative dentry. If we were
144 dropping dcache lock, go
145 throught the list again */
146 if (ll_drop_dentry(child))
151 spin_unlock(&dcache_lock);
155 int ll_md_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
156 void *data, int flag)
159 struct lustre_handle lockh;
163 case LDLM_CB_BLOCKING:
164 ldlm_lock2handle(lock, &lockh);
165 rc = ldlm_cli_cancel(&lockh);
167 CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
171 case LDLM_CB_CANCELING: {
172 struct inode *inode = ll_inode_from_lock(lock);
173 __u64 bits = lock->l_policy_data.l_inodebits.bits;
176 /* Invalidate all dentries associated with this inode */
180 LASSERT(lock->l_flags & LDLM_FL_CANCELING);
181 if ((bits & MDS_INODELOCK_LOOKUP) &&
182 ll_have_md_lock(inode, MDS_INODELOCK_LOOKUP))
183 bits &= ~MDS_INODELOCK_LOOKUP;
184 if ((bits & MDS_INODELOCK_UPDATE) &&
185 ll_have_md_lock(inode, MDS_INODELOCK_UPDATE))
186 bits &= ~MDS_INODELOCK_UPDATE;
187 if ((bits & MDS_INODELOCK_OPEN) &&
188 ll_have_md_lock(inode, MDS_INODELOCK_OPEN))
189 bits &= ~MDS_INODELOCK_OPEN;
191 fid = ll_inode2fid(inode);
192 if (lock->l_resource->lr_name.name[0] != fid_seq(fid) ||
193 lock->l_resource->lr_name.name[1] != fid_oid(fid) ||
194 lock->l_resource->lr_name.name[2] != fid_ver(fid)) {
195 LDLM_ERROR(lock, "data mismatch with object "
196 DFID" (%p)", PFID(fid), inode);
199 if (bits & MDS_INODELOCK_OPEN) {
201 switch (lock->l_req_mode) {
212 CERROR("Unexpected lock mode for OPEN lock "
213 "%d, inode %ld\n", lock->l_req_mode,
216 ll_md_real_close(inode, flags);
219 if (bits & MDS_INODELOCK_UPDATE)
220 ll_i2info(inode)->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
222 if (S_ISDIR(inode->i_mode) &&
223 (bits & MDS_INODELOCK_UPDATE)) {
224 CDEBUG(D_INODE, "invalidating inode %lu\n",
226 truncate_inode_pages(inode->i_mapping, 0);
227 ll_drop_negative_dentry(inode);
230 if (inode->i_sb->s_root &&
231 inode != inode->i_sb->s_root->d_inode &&
232 (bits & MDS_INODELOCK_LOOKUP))
233 ll_unhash_aliases(inode);
244 __u32 ll_i2suppgid(struct inode *i)
246 if (in_group_p(i->i_gid))
247 return (__u32)i->i_gid;
252 /* Pack the required supplementary groups into the supplied groups array.
253 * If we don't need to use the groups from the target inode(s) then we
254 * instead pack one or more groups from the user's supplementary group
255 * array in case it might be useful. Not needed if doing an MDS-side upcall. */
256 void ll_i2gids(__u32 *suppgids, struct inode *i1, struct inode *i2)
263 LASSERT(suppgids != NULL);
265 suppgids[0] = ll_i2suppgid(i1);
268 suppgids[1] = ll_i2suppgid(i2);
273 for (i = 0; i < current_ngroups; i++) {
274 if (suppgids[0] == -1) {
275 if (current_groups[i] != suppgids[1])
276 suppgids[0] = current_groups[i];
279 if (suppgids[1] == -1) {
280 if (current_groups[i] != suppgids[0])
281 suppgids[1] = current_groups[i];
289 static void ll_d_add(struct dentry *de, struct inode *inode)
291 CDEBUG(D_DENTRY, "adding inode %p to dentry %p\n", inode, de);
293 if (!list_empty(&de->d_alias)) {
294 spin_unlock(&dcache_lock);
295 CERROR("dentry %.*s %p alias next %p, prev %p\n",
296 de->d_name.len, de->d_name.name, de,
297 de->d_alias.next, de->d_alias.prev);
301 list_add(&de->d_alias, &inode->i_dentry);
305 if (!d_unhashed(de)) {
306 spin_unlock(&dcache_lock);
307 CERROR("dentry %.*s %p hash next %p\n",
308 de->d_name.len, de->d_name.name, de, de->d_hash.next);
311 d_rehash_cond(de, 0);
314 /* Search "inode"'s alias list for a dentry that has the same name and parent
315 * as de. If found, return it. If not found, return de.
316 * Lustre can't use d_add_unique because don't unhash aliases for directory
317 * in ll_revalidate_it. After revaliadate inode will be have hashed aliases
318 * and it triggers BUG_ON in d_instantiate_unique (bug #10954).
320 struct dentry *ll_find_alias(struct inode *inode, struct dentry *de)
322 struct list_head *tmp;
323 struct dentry *dentry;
324 struct dentry *last_discon = NULL;
326 spin_lock(&dcache_lock);
327 list_for_each(tmp, &inode->i_dentry) {
328 dentry = list_entry(tmp, struct dentry, d_alias);
330 /* We are called here with 'de' already on the aliases list. */
331 if (unlikely(dentry == de)) {
336 if (dentry->d_flags & DCACHE_DISCONNECTED) {
337 LASSERT(last_discon == NULL);
338 last_discon = dentry;
342 if (dentry->d_parent != de->d_parent)
345 if (dentry->d_name.hash != de->d_name.hash)
348 if (dentry->d_name.len != de->d_name.len)
351 if (memcmp(dentry->d_name.name, de->d_name.name,
352 de->d_name.len) != 0)
358 #ifdef DCACHE_LUSTRE_INVALID
359 dentry->d_flags &= ~DCACHE_LUSTRE_INVALID;
361 unlock_dentry(dentry);
362 d_rehash_cond(dentry, 0); /* avoid taking dcache_lock inside */
363 spin_unlock(&dcache_lock);
365 CDEBUG(D_DENTRY, "alias dentry %.*s (%p) parent %p inode %p "
366 "refc %d\n", de->d_name.len, de->d_name.name, de,
367 de->d_parent, de->d_inode, atomic_read(&de->d_count));
372 CDEBUG(D_DENTRY, "Reuse disconnected dentry %p inode %p "
373 "refc %d\n", last_discon, last_discon->d_inode,
374 atomic_read(&last_discon->d_count));
375 dget_locked(last_discon);
376 spin_unlock(&dcache_lock);
378 d_move(last_discon, de);
385 spin_unlock(&dcache_lock);
390 static int lookup_it_finish(struct ptlrpc_request *request,
391 struct lookup_intent *it,
394 struct it_cb_data *icbd = data;
395 struct dentry **de = icbd->icbd_childp;
396 struct inode *parent = icbd->icbd_parent;
397 struct ll_sb_info *sbi = ll_i2sbi(parent);
398 struct inode *inode = NULL;
401 /* NB 1 request reference will be taken away by ll_intent_lock()
403 if (!it_disposition(it, DISP_LOOKUP_NEG)) {
406 rc = ll_prep_inode(&inode, request, (*de)->d_sb);
410 CDEBUG(D_DLMTRACE, "setting l_data to inode %p (%lu/%u)\n",
411 inode, inode->i_ino, inode->i_generation);
412 md_set_lock_data(sbi->ll_md_exp,
413 &it->d.lustre.it_lock_handle, inode);
415 /* We used to query real size from OSTs here, but actually
416 this is not needed. For stat() calls size would be updated
417 from subsequent do_revalidate()->ll_inode_revalidate_it() in
419 vfs_getattr_it->ll_getattr()->ll_inode_revalidate_it() in 2.6
420 Everybody else who needs correct file size would call
421 ll_glimpse_size or some equivalent themselves anyway.
422 Also see bug 7198. */
424 *de = ll_find_alias(inode, *de);
427 /* Check that parent has UPDATE lock. If there is none, we
428 cannot afford to hash this dentry (done by ll_d_add) as it
429 might get picked up later when UPDATE lock will appear */
430 if (ll_have_md_lock(parent, MDS_INODELOCK_UPDATE)) {
431 spin_lock(&dcache_lock);
432 ll_d_add(*de, inode);
433 spin_unlock(&dcache_lock);
435 (*de)->d_inode = NULL;
436 /* We do not want to hash the dentry if don`t have a
437 * lock, but if this dentry is later used in d_move,
438 * we'd hit uninitialised list head d_hash, so we just
439 * do this to init d_hash field but leave dentry
440 * unhashed. (bug 10796). */
447 (*de)->d_op = &ll_d_ops;
452 static struct dentry *ll_lookup_it(struct inode *parent, struct dentry *dentry,
453 struct lookup_intent *it, int lookup_flags)
455 struct lookup_intent lookup_it = { .it_op = IT_LOOKUP };
456 struct dentry *save = dentry, *retval;
457 struct ptlrpc_request *req = NULL;
458 struct md_op_data *op_data;
459 struct it_cb_data icbd;
464 if (dentry->d_name.len > ll_i2sbi(parent)->ll_namelen)
465 RETURN(ERR_PTR(-ENAMETOOLONG));
467 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),intent=%s\n",
468 dentry->d_name.len, dentry->d_name.name, parent->i_ino,
469 parent->i_generation, parent, LL_IT2STR(it));
471 if (d_mountpoint(dentry))
472 CERROR("Tell Peter, lookup on mtpt, it %s\n", LL_IT2STR(it));
474 ll_frob_intent(&it, &lookup_it);
476 /* As do_lookup is called before follow_mount, root dentry may be left
477 * not valid, revalidate it here. */
478 if (parent->i_sb->s_root && (parent->i_sb->s_root->d_inode == parent) &&
479 (it->it_op & (IT_OPEN | IT_CREAT))) {
480 rc = ll_inode_revalidate_it(parent->i_sb->s_root, it);
485 icbd.icbd_childp = &dentry;
486 icbd.icbd_parent = parent;
488 if (it->it_op & IT_CREAT ||
489 (it->it_op & IT_OPEN && it->it_create_mode & O_CREAT))
490 opc = LUSTRE_OPC_CREATE;
492 opc = LUSTRE_OPC_ANY;
494 op_data = ll_prep_md_op_data(NULL, parent, NULL, dentry->d_name.name,
495 dentry->d_name.len, lookup_flags, opc,
498 RETURN((void *)op_data);
500 it->it_create_mode &= ~current->fs->umask;
502 rc = md_intent_lock(ll_i2mdexp(parent), op_data, NULL, 0, it,
503 lookup_flags, &req, ll_md_blocking_ast, 0);
504 ll_finish_md_op_data(op_data);
506 GOTO(out, retval = ERR_PTR(rc));
508 rc = lookup_it_finish(req, it, &icbd);
510 ll_intent_release(it);
511 GOTO(out, retval = ERR_PTR(rc));
514 if ((it->it_op & IT_OPEN) && dentry->d_inode &&
515 !S_ISREG(dentry->d_inode->i_mode) &&
516 !S_ISDIR(dentry->d_inode->i_mode)) {
517 ll_release_openhandle(dentry, it);
519 ll_lookup_finish_locks(it, dentry);
522 GOTO(out, retval = NULL);
524 GOTO(out, retval = dentry);
527 ptlrpc_req_finished(req);
531 #ifdef HAVE_VFS_INTENT_PATCHES
532 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
533 struct nameidata *nd)
538 if (nd && nd->flags & LOOKUP_LAST && !(nd->flags & LOOKUP_LINK_NOTLAST))
539 de = ll_lookup_it(parent, dentry, &nd->intent, nd->flags);
541 de = ll_lookup_it(parent, dentry, NULL, 0);
546 struct lookup_intent *ll_convert_intent(struct open_intent *oit,
549 struct lookup_intent *it;
551 OBD_ALLOC(it, sizeof(*it));
553 return ERR_PTR(-ENOMEM);
555 if (lookup_flags & LOOKUP_OPEN) {
557 if (lookup_flags & LOOKUP_CREATE)
558 it->it_op |= IT_CREAT;
559 it->it_create_mode = oit->create_mode;
560 it->it_flags = oit->flags;
562 it->it_op = IT_GETATTR;
565 #ifndef HAVE_FILE_IN_STRUCT_INTENT
566 /* Since there is no way to pass our intent to ll_file_open,
567 * just check the file is there. Actual open will be done
569 if (it->it_op & IT_OPEN)
570 it->it_op = IT_LOOKUP;
576 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
577 struct nameidata *nd)
582 if (nd && !(nd->flags & (LOOKUP_CONTINUE|LOOKUP_PARENT))) {
583 struct lookup_intent *it;
585 #if defined(HAVE_FILE_IN_STRUCT_INTENT) && (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,17))
586 /* Did we came here from failed revalidate just to propagate
588 if (nd->flags & LOOKUP_OPEN)
589 if (IS_ERR(nd->intent.open.file))
590 RETURN((struct dentry *)nd->intent.open.file);
593 if (ll_d2d(dentry) && ll_d2d(dentry)->lld_it) {
594 it = ll_d2d(dentry)->lld_it;
595 ll_d2d(dentry)->lld_it = NULL;
597 it = ll_convert_intent(&nd->intent.open, nd->flags);
599 RETURN((struct dentry *)it);
602 de = ll_lookup_it(parent, dentry, it, nd->flags);
605 if ((nd->flags & LOOKUP_OPEN) && !IS_ERR(dentry)) { /* Open */
606 if (dentry->d_inode &&
607 it_disposition(it, DISP_OPEN_OPEN)) { /* nocreate */
608 #ifdef HAVE_FILE_IN_STRUCT_INTENT
609 if (S_ISFIFO(dentry->d_inode->i_mode)) {
610 // We cannot call open here as it would
613 (struct ptlrpc_request *)
614 it->d.lustre.it_data);
617 nd->intent.open.file->private_data = it;
618 filp =lookup_instantiate_filp(nd,dentry,
620 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,17))
621 /* 2.6.1[456] have a bug in open_namei() that forgets to check
622 * nd->intent.open.file for error, so we need to return it as lookup's result
627 de = (struct dentry *) filp;
632 #else /* HAVE_FILE_IN_STRUCT_INTENT */
633 /* Release open handle as we have no way to
634 * pass it to ll_file_open */
635 ll_release_openhandle(dentry, it);
636 #endif /* HAVE_FILE_IN_STRUCT_INTENT */
637 } else if (it_disposition(it, DISP_OPEN_CREATE)) {
638 // XXX This can only reliably work on assumption
639 // that there are NO hashed negative dentries.
640 ll_d2d(dentry)->lld_it = it;
641 it = NULL; /* Will be freed in ll_create_nd */
642 /* We absolutely depend on ll_create_nd to be
643 * called to not leak this intent and possible
644 * data attached to it */
649 ll_intent_release(it);
650 OBD_FREE(it, sizeof(*it));
653 de = ll_lookup_it(parent, dentry, NULL, 0);
660 /* We depend on "mode" being set with the proper file type/umask by now */
661 static struct inode *ll_create_node(struct inode *dir, const char *name,
662 int namelen, const void *data, int datalen,
663 int mode, __u64 extra,
664 struct lookup_intent *it)
666 struct inode *inode = NULL;
667 struct ptlrpc_request *request = NULL;
668 struct ll_sb_info *sbi = ll_i2sbi(dir);
672 LASSERT(it && it->d.lustre.it_disposition);
674 LASSERT(it_disposition(it, DISP_ENQ_CREATE_REF));
675 request = it->d.lustre.it_data;
676 it_clear_disposition(it, DISP_ENQ_CREATE_REF);
677 rc = ll_prep_inode(&inode, request, dir->i_sb);
679 GOTO(out, inode = ERR_PTR(rc));
681 LASSERT(list_empty(&inode->i_dentry));
683 /* We asked for a lock on the directory, but were granted a
684 * lock on the inode. Since we finally have an inode pointer,
685 * stuff it in the lock. */
686 CDEBUG(D_DLMTRACE, "setting l_ast_data to inode %p (%lu/%u)\n",
687 inode, inode->i_ino, inode->i_generation);
688 md_set_lock_data(sbi->ll_md_exp,
689 &it->d.lustre.it_lock_handle, inode);
692 ptlrpc_req_finished(request);
697 * By the time this is called, we already have created the directory cache
698 * entry for the new file, but it is so far negative - it has no inode.
700 * We defer creating the OBD object(s) until open, to keep the intent and
701 * non-intent code paths similar, and also because we do not have the MDS
702 * inode number before calling ll_create_node() (which is needed for LOV),
703 * so we would need to do yet another RPC to the MDS to store the LOV EA
704 * data on the MDS. If needed, we would pass the PACKED lmm as data and
705 * lmm_size in datalen (the MDS still has code which will handle that).
707 * If the create succeeds, we fill in the inode information
708 * with d_instantiate().
710 static int ll_create_it(struct inode *dir, struct dentry *dentry, int mode,
711 struct lookup_intent *it)
717 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),intent=%s\n",
718 dentry->d_name.len, dentry->d_name.name, dir->i_ino,
719 dir->i_generation, dir, LL_IT2STR(it));
721 rc = it_open_error(DISP_OPEN_CREATE, it);
725 inode = ll_create_node(dir, dentry->d_name.name, dentry->d_name.len,
726 NULL, 0, mode, 0, it);
728 RETURN(PTR_ERR(inode));
731 d_instantiate(dentry, inode);
732 /* Negative dentry may be unhashed if parent does not have UPDATE lock,
733 * but some callers, e.g. do_coredump, expect dentry to be hashed after
734 * successful create. Hash it here. */
735 spin_lock(&dcache_lock);
736 if (d_unhashed(dentry))
737 d_rehash_cond(dentry, 0);
738 spin_unlock(&dcache_lock);
742 static void ll_update_times(struct ptlrpc_request *request,
745 struct mdt_body *body = req_capsule_server_get(&request->rq_pill,
749 /* mtime is always updated with ctime, but can be set in past.
750 As write and utime(2) may happen within 1 second, and utime's
751 mtime has a priority over write's one, so take mtime from mds
752 for the same ctimes. */
753 if (body->valid & OBD_MD_FLCTIME &&
754 body->ctime >= LTIME_S(inode->i_ctime)) {
755 LTIME_S(inode->i_ctime) = body->ctime;
757 if (body->valid & OBD_MD_FLMTIME) {
758 CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
759 "to "LPU64"\n", inode->i_ino,
760 LTIME_S(inode->i_mtime), body->mtime);
761 LTIME_S(inode->i_mtime) = body->mtime;
766 static int ll_new_node(struct inode *dir, struct qstr *name,
767 const char *tgt, int mode, int rdev,
768 struct dentry *dchild, __u32 opc)
770 struct ptlrpc_request *request = NULL;
771 struct md_op_data *op_data;
772 struct inode *inode = NULL;
773 struct ll_sb_info *sbi = ll_i2sbi(dir);
778 if (unlikely(tgt != NULL))
779 tgt_len = strlen(tgt) + 1;
781 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
782 name->len, 0, opc, NULL);
784 GOTO(err_exit, err = PTR_ERR(op_data));
786 err = md_create(sbi->ll_md_exp, op_data, tgt, tgt_len, mode,
787 current->fsuid, current->fsgid,
788 current->cap_effective, rdev, &request);
789 ll_finish_md_op_data(op_data);
793 ll_update_times(request, dir);
796 err = ll_prep_inode(&inode, request, dchild->d_sb);
801 d_instantiate(dchild, inode);
805 ptlrpc_req_finished(request);
810 static int ll_mknod_generic(struct inode *dir, struct qstr *name, int mode,
811 unsigned rdev, struct dentry *dchild)
816 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p) mode %o dev %x\n",
817 name->len, name->name, dir->i_ino, dir->i_generation, dir,
820 mode &= ~current->fs->umask;
822 switch (mode & S_IFMT) {
824 mode |= S_IFREG; /* for mode = 0 case, fallthrough */
830 err = ll_new_node(dir, name, NULL, mode, rdev, dchild,
842 #ifndef HAVE_VFS_INTENT_PATCHES
843 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
844 int mode, struct nameidata *nd)
846 struct lookup_intent *it = ll_d2d(dentry)->lld_it;
850 return ll_mknod_generic(dir, &dentry->d_name, mode, 0, dentry);
852 ll_d2d(dentry)->lld_it = NULL;
854 /* Was there an error? Propagate it! */
855 if (it->d.lustre.it_status) {
856 rc = it->d.lustre.it_status;
860 rc = ll_create_it(dir, dentry, mode, it);
861 #ifdef HAVE_FILE_IN_STRUCT_INTENT
862 if (nd && (nd->flags & LOOKUP_OPEN) && dentry->d_inode) { /* Open */
863 nd->intent.open.file->private_data = it;
864 lookup_instantiate_filp(nd, dentry, NULL);
867 ll_release_openhandle(dentry,it);
871 ll_intent_release(it);
872 OBD_FREE(it, sizeof(*it));
877 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
878 int mode, struct nameidata *nd)
880 if (!nd || !nd->intent.d.lustre.it_disposition)
881 /* No saved request? Just mknod the file */
882 return ll_mknod_generic(dir, &dentry->d_name, mode, 0, dentry);
884 return ll_create_it(dir, dentry, mode, &nd->intent);
888 static int ll_symlink_generic(struct inode *dir, struct qstr *name,
889 const char *tgt, struct dentry *dchild)
894 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),target=%.*s\n",
895 name->len, name->name, dir->i_ino, dir->i_generation,
898 err = ll_new_node(dir, name, (char *)tgt, S_IFLNK | S_IRWXUGO,
899 0, dchild, LUSTRE_OPC_SYMLINK);
903 static int ll_link_generic(struct inode *src, struct inode *dir,
904 struct qstr *name, struct dentry *dchild)
906 struct ll_sb_info *sbi = ll_i2sbi(dir);
907 struct ptlrpc_request *request = NULL;
908 struct md_op_data *op_data;
913 "VFS Op: inode=%lu/%u(%p), dir=%lu/%u(%p), target=%.*s\n",
914 src->i_ino, src->i_generation, src, dir->i_ino,
915 dir->i_generation, dir, name->len, name->name);
917 op_data = ll_prep_md_op_data(NULL, src, dir, name->name, name->len,
918 0, LUSTRE_OPC_ANY, NULL);
920 RETURN(PTR_ERR(op_data));
922 err = md_link(sbi->ll_md_exp, op_data, &request);
923 ll_finish_md_op_data(op_data);
929 ll_update_times(request, dir);
932 ptlrpc_req_finished(request);
936 static int ll_mkdir_generic(struct inode *dir, struct qstr *name,
937 int mode, struct dentry *dchild)
943 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
944 name->len, name->name, dir->i_ino, dir->i_generation, dir);
946 mode = (mode & (S_IRWXUGO|S_ISVTX) & ~current->fs->umask) | S_IFDIR;
947 err = ll_new_node(dir, name, NULL, mode, 0, dchild, LUSTRE_OPC_MKDIR);
952 /* Try to find the child dentry by its name.
953 If found, put the result fid into @fid. */
954 static void ll_get_child_fid(struct inode * dir, struct qstr *name,
957 struct dentry *parent, *child;
959 parent = list_entry(dir->i_dentry.next, struct dentry, d_alias);
960 child = d_lookup(parent, name);
963 *fid = *ll_inode2fid(child->d_inode);
968 static int ll_rmdir_generic(struct inode *dir, struct dentry *dparent,
969 struct dentry *dchild, struct qstr *name)
971 struct ptlrpc_request *request = NULL;
972 struct md_op_data *op_data;
976 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
977 name->len, name->name, dir->i_ino, dir->i_generation, dir);
979 if (unlikely(ll_d_mountpoint(dparent, dchild, name)))
982 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len,
983 S_IFDIR, LUSTRE_OPC_ANY, NULL);
985 RETURN(PTR_ERR(op_data));
987 ll_get_child_fid(dir, name, &op_data->op_fid3);
988 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
989 ll_finish_md_op_data(op_data);
991 ll_update_times(request, dir);
992 ptlrpc_req_finished(request);
996 int ll_objects_destroy(struct ptlrpc_request *request, struct inode *dir)
998 struct mdt_body *body;
999 struct lov_mds_md *eadata;
1000 struct lov_stripe_md *lsm = NULL;
1001 struct obd_trans_info oti = { 0 };
1006 /* req is swabbed so this is safe */
1007 body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
1008 if (!(body->valid & OBD_MD_FLEASIZE))
1011 if (body->eadatasize == 0) {
1012 CERROR("OBD_MD_FLEASIZE set but eadatasize zero\n");
1013 GOTO(out, rc = -EPROTO);
1016 /* The MDS sent back the EA because we unlinked the last reference
1017 * to this file. Use this EA to unlink the objects on the OST.
1018 * It's opaque so we don't swab here; we leave it to obd_unpackmd() to
1019 * check it is complete and sensible. */
1020 eadata = req_capsule_server_sized_get(&request->rq_pill, &RMF_MDT_MD,
1022 LASSERT(eadata != NULL);
1024 rc = obd_unpackmd(ll_i2dtexp(dir), &lsm, eadata, body->eadatasize);
1026 CERROR("obd_unpackmd: %d\n", rc);
1029 LASSERT(rc >= sizeof(*lsm));
1031 rc = obd_checkmd(ll_i2dtexp(dir), ll_i2mdexp(dir), lsm);
1033 GOTO(out_free_memmd, rc);
1037 GOTO(out_free_memmd, rc = -ENOMEM);
1039 oa->o_id = lsm->lsm_object_id;
1040 oa->o_gr = lsm->lsm_object_gr;
1041 oa->o_mode = body->mode & S_IFMT;
1042 oa->o_valid = OBD_MD_FLID | OBD_MD_FLTYPE | OBD_MD_FLGROUP;
1044 if (body->valid & OBD_MD_FLCOOKIE) {
1045 oa->o_valid |= OBD_MD_FLCOOKIE;
1046 oti.oti_logcookies =
1047 req_capsule_server_sized_get(&request->rq_pill,
1049 sizeof(struct llog_cookie) *
1050 lsm->lsm_stripe_count);
1051 if (oti.oti_logcookies == NULL) {
1052 oa->o_valid &= ~OBD_MD_FLCOOKIE;
1053 body->valid &= ~OBD_MD_FLCOOKIE;
1057 rc = obd_destroy(ll_i2dtexp(dir), oa, lsm, &oti, ll_i2mdexp(dir));
1060 CERROR("obd destroy objid "LPX64" error %d\n",
1061 lsm->lsm_object_id, rc);
1063 obd_free_memmd(ll_i2dtexp(dir), &lsm);
1068 static int ll_unlink_generic(struct inode *dir, struct dentry *dparent,
1069 struct dentry *dchild, struct qstr *name)
1071 struct ptlrpc_request *request = NULL;
1072 struct md_op_data *op_data;
1075 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
1076 name->len, name->name, dir->i_ino, dir->i_generation, dir);
1079 * XXX: unlink bind mountpoint maybe call to here,
1080 * just check it as vfs_unlink does.
1082 if (unlikely(ll_d_mountpoint(dparent, dchild, name)))
1085 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
1086 name->len, 0, LUSTRE_OPC_ANY, NULL);
1087 if (IS_ERR(op_data))
1088 RETURN(PTR_ERR(op_data));
1090 ll_get_child_fid(dir, name, &op_data->op_fid3);
1091 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1092 ll_finish_md_op_data(op_data);
1096 ll_update_times(request, dir);
1098 rc = ll_objects_destroy(request, dir);
1100 ptlrpc_req_finished(request);
1104 static int ll_rename_generic(struct inode *src, struct dentry *src_dparent,
1105 struct dentry *src_dchild, struct qstr *src_name,
1106 struct inode *tgt, struct dentry *tgt_dparent,
1107 struct dentry *tgt_dchild, struct qstr *tgt_name)
1109 struct ptlrpc_request *request = NULL;
1110 struct ll_sb_info *sbi = ll_i2sbi(src);
1111 struct md_op_data *op_data;
1114 CDEBUG(D_VFSTRACE,"VFS Op:oldname=%.*s,src_dir=%lu/%u(%p),newname=%.*s,"
1115 "tgt_dir=%lu/%u(%p)\n", src_name->len, src_name->name,
1116 src->i_ino, src->i_generation, src, tgt_name->len,
1117 tgt_name->name, tgt->i_ino, tgt->i_generation, tgt);
1119 if (unlikely(ll_d_mountpoint(src_dparent, src_dchild, src_name) ||
1120 ll_d_mountpoint(tgt_dparent, tgt_dchild, tgt_name)))
1123 op_data = ll_prep_md_op_data(NULL, src, tgt, NULL, 0, 0,
1124 LUSTRE_OPC_ANY, NULL);
1125 if (IS_ERR(op_data))
1126 RETURN(PTR_ERR(op_data));
1128 ll_get_child_fid(src, src_name, &op_data->op_fid3);
1129 ll_get_child_fid(tgt, tgt_name, &op_data->op_fid4);
1130 err = md_rename(sbi->ll_md_exp, op_data,
1131 src_name->name, src_name->len,
1132 tgt_name->name, tgt_name->len, &request);
1133 ll_finish_md_op_data(op_data);
1135 ll_update_times(request, src);
1136 ll_update_times(request, tgt);
1137 err = ll_objects_destroy(request, src);
1140 ptlrpc_req_finished(request);
1145 #ifdef HAVE_VFS_INTENT_PATCHES
1146 static int ll_mknod_raw(struct nameidata *nd, int mode, dev_t rdev)
1148 return ll_mknod_generic(nd->dentry->d_inode, &nd->last, mode,rdev,NULL);
1150 static int ll_rename_raw(struct nameidata *srcnd, struct nameidata *tgtnd)
1152 return ll_rename_generic(srcnd->dentry->d_inode, srcnd->dentry,
1154 tgtnd->dentry->d_inode, tgtnd->dentry,
1155 NULL, &tgtnd->last);
1157 static int ll_link_raw(struct nameidata *srcnd, struct nameidata *tgtnd)
1159 return ll_link_generic(srcnd->dentry->d_inode, tgtnd->dentry->d_inode,
1160 &tgtnd->last, NULL);
1162 static int ll_symlink_raw(struct nameidata *nd, const char *tgt)
1164 return ll_symlink_generic(nd->dentry->d_inode, &nd->last, tgt, NULL);
1166 static int ll_rmdir_raw(struct nameidata *nd)
1168 return ll_rmdir_generic(nd->dentry->d_inode, nd->dentry, NULL,
1171 static int ll_mkdir_raw(struct nameidata *nd, int mode)
1173 return ll_mkdir_generic(nd->dentry->d_inode, &nd->last, mode, NULL);
1175 static int ll_unlink_raw(struct nameidata *nd)
1177 return ll_unlink_generic(nd->dentry->d_inode, nd->dentry, NULL,
1182 static int ll_mknod(struct inode *dir, struct dentry *dchild, int mode,
1185 return ll_mknod_generic(dir, &dchild->d_name, mode,
1186 old_encode_dev(rdev), dchild);
1189 static int ll_unlink(struct inode * dir, struct dentry *dentry)
1191 return ll_unlink_generic(dir, NULL, dentry, &dentry->d_name);
1193 static int ll_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1195 return ll_mkdir_generic(dir, &dentry->d_name, mode, dentry);
1197 static int ll_rmdir(struct inode *dir, struct dentry *dentry)
1199 return ll_rmdir_generic(dir, NULL, dentry, &dentry->d_name);
1201 static int ll_symlink(struct inode *dir, struct dentry *dentry,
1202 const char *oldname)
1204 return ll_symlink_generic(dir, &dentry->d_name, oldname, dentry);
1206 static int ll_link(struct dentry *old_dentry, struct inode *dir,
1207 struct dentry *new_dentry)
1209 return ll_link_generic(old_dentry->d_inode, dir, &new_dentry->d_name,
1212 static int ll_rename(struct inode *old_dir, struct dentry *old_dentry,
1213 struct inode *new_dir, struct dentry *new_dentry)
1215 return ll_rename_generic(old_dir, NULL,
1216 old_dentry, &old_dentry->d_name,
1217 new_dir, NULL, new_dentry,
1218 &new_dentry->d_name);
1221 struct inode_operations ll_dir_inode_operations = {
1222 #ifdef HAVE_VFS_INTENT_PATCHES
1223 .link_raw = ll_link_raw,
1224 .unlink_raw = ll_unlink_raw,
1225 .symlink_raw = ll_symlink_raw,
1226 .mkdir_raw = ll_mkdir_raw,
1227 .rmdir_raw = ll_rmdir_raw,
1228 .mknod_raw = ll_mknod_raw,
1229 .rename_raw = ll_rename_raw,
1230 .setattr = ll_setattr,
1231 .setattr_raw = ll_setattr_raw,
1234 .lookup = ll_lookup_nd,
1235 .create = ll_create_nd,
1236 /* We need all these non-raw things for NFSD, to not patch it. */
1237 .unlink = ll_unlink,
1240 .symlink = ll_symlink,
1242 .rename = ll_rename,
1243 .setattr = ll_setattr,
1244 .getattr = ll_getattr,
1245 .permission = ll_inode_permission,
1246 .setxattr = ll_setxattr,
1247 .getxattr = ll_getxattr,
1248 .listxattr = ll_listxattr,
1249 .removexattr = ll_removexattr,
1252 struct inode_operations ll_special_inode_operations = {
1253 #ifdef HAVE_VFS_INTENT_PATCHES
1254 .setattr_raw = ll_setattr_raw,
1256 .setattr = ll_setattr,
1257 .getattr = ll_getattr,
1258 .permission = ll_inode_permission,
1259 .setxattr = ll_setxattr,
1260 .getxattr = ll_getxattr,
1261 .listxattr = ll_listxattr,
1262 .removexattr = ll_removexattr,