1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
6 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
8 * This program is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License version 2 only,
10 * as published by the Free Software Foundation.
12 * This program is distributed in the hope that it will be useful, but
13 * WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * General Public License version 2 for more details (a copy is included
16 * in the LICENSE file that accompanied this code).
18 * You should have received a copy of the GNU General Public License
19 * version 2 along with this program; If not, see
20 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
22 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
23 * CA 95054 USA or visit www.sun.com if you need additional information or
29 * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
30 * Use is subject to license terms.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
38 #include <linux/sched.h>
40 #include <linux/smp_lock.h>
41 #include <linux/quotaops.h>
42 #include <linux/highmem.h>
43 #include <linux/pagemap.h>
45 #define DEBUG_SUBSYSTEM S_LLITE
47 #include <obd_support.h>
48 #include <lustre_fid.h>
49 #include <lustre_lite.h>
50 #include <lustre_dlm.h>
51 #include <lustre_ver.h>
52 #include <lustre_mdc.h>
53 #include "llite_internal.h"
56 * Check if we have something mounted at the named dchild.
57 * In such a case there would always be dentry present.
59 static int ll_d_mountpoint(struct dentry *dparent, struct dentry *dchild,
64 if (unlikely(dchild)) {
65 mounted = d_mountpoint(dchild);
67 dchild = d_lookup(dparent, name);
69 mounted = d_mountpoint(dchild);
76 int ll_unlock(__u32 mode, struct lustre_handle *lockh)
80 ldlm_lock_decref(lockh, mode);
86 /* called from iget5_locked->find_inode() under inode_lock spinlock */
87 static int ll_test_inode(struct inode *inode, void *opaque)
89 struct ll_inode_info *lli = ll_i2info(inode);
90 struct lustre_md *md = opaque;
92 if (unlikely(!(md->body->valid & OBD_MD_FLID))) {
93 CERROR("MDS body missing FID\n");
97 if (!lu_fid_eq(&lli->lli_fid, &md->body->fid1))
103 static int ll_set_inode(struct inode *inode, void *opaque)
105 struct ll_inode_info *lli = ll_i2info(inode);
106 struct mdt_body *body = ((struct lustre_md *)opaque)->body;
108 if (unlikely(!(body->valid & OBD_MD_FLID))) {
109 CERROR("MDS body missing FID\n");
113 lli->lli_fid = body->fid1;
119 * Get an inode by inode number (already instantiated by the intent lookup).
120 * Returns inode or NULL
122 struct inode *ll_iget(struct super_block *sb, ino_t hash,
123 struct lustre_md *md)
129 inode = iget5_locked(sb, hash, ll_test_inode, ll_set_inode, md);
132 if (inode->i_state & I_NEW) {
135 ll_read_inode2(inode, md);
136 rc = cl_inode_init(inode, md);
139 make_bad_inode(inode);
140 unlock_new_inode(inode);
144 unlock_new_inode(inode);
145 } else if (!(inode->i_state & (I_FREEING | I_CLEAR)))
146 ll_update_inode(inode, md);
147 CDEBUG(D_VFSTRACE, "got inode: %p for "DFID"\n",
148 inode, PFID(&md->body->fid1));
153 static void ll_drop_negative_dentry(struct inode *dir)
155 struct dentry *dentry, *tmp_alias, *tmp_subdir;
157 cfs_spin_lock(&ll_lookup_lock);
158 spin_lock(&dcache_lock);
160 list_for_each_entry_safe(dentry, tmp_alias,
161 &dir->i_dentry,d_alias) {
162 if (!list_empty(&dentry->d_subdirs)) {
163 struct dentry *child;
164 list_for_each_entry_safe(child, tmp_subdir,
167 /* XXX Print some debug here? */
169 /* Negative dentry. If we were
170 dropping dcache lock, go
171 throught the list again */
172 if (ll_drop_dentry(child))
177 spin_unlock(&dcache_lock);
178 cfs_spin_unlock(&ll_lookup_lock);
182 int ll_md_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
183 void *data, int flag)
186 struct lustre_handle lockh;
190 case LDLM_CB_BLOCKING:
191 ldlm_lock2handle(lock, &lockh);
192 rc = ldlm_cli_cancel(&lockh);
194 CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
198 case LDLM_CB_CANCELING: {
199 struct inode *inode = ll_inode_from_lock(lock);
200 struct ll_inode_info *lli;
201 __u64 bits = lock->l_policy_data.l_inodebits.bits;
204 /* Invalidate all dentries associated with this inode */
208 LASSERT(lock->l_flags & LDLM_FL_CANCELING);
209 if ((bits & MDS_INODELOCK_LOOKUP) &&
210 ll_have_md_lock(inode, MDS_INODELOCK_LOOKUP))
211 bits &= ~MDS_INODELOCK_LOOKUP;
212 if ((bits & MDS_INODELOCK_UPDATE) &&
213 ll_have_md_lock(inode, MDS_INODELOCK_UPDATE))
214 bits &= ~MDS_INODELOCK_UPDATE;
215 if ((bits & MDS_INODELOCK_OPEN) &&
216 ll_have_md_lock(inode, MDS_INODELOCK_OPEN))
217 bits &= ~MDS_INODELOCK_OPEN;
219 fid = ll_inode2fid(inode);
220 if (lock->l_resource->lr_name.name[0] != fid_seq(fid) ||
221 lock->l_resource->lr_name.name[1] != fid_oid(fid) ||
222 lock->l_resource->lr_name.name[2] != fid_ver(fid)) {
223 LDLM_ERROR(lock, "data mismatch with object "
224 DFID" (%p)", PFID(fid), inode);
227 if (bits & MDS_INODELOCK_OPEN) {
229 switch (lock->l_req_mode) {
240 CERROR("Unexpected lock mode for OPEN lock "
241 "%d, inode %ld\n", lock->l_req_mode,
244 ll_md_real_close(inode, flags);
247 lli = ll_i2info(inode);
248 if (bits & MDS_INODELOCK_UPDATE)
249 lli->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
251 if (S_ISDIR(inode->i_mode) &&
252 (bits & MDS_INODELOCK_UPDATE)) {
253 CDEBUG(D_INODE, "invalidating inode %lu\n",
255 truncate_inode_pages(inode->i_mapping, 0);
256 ll_drop_negative_dentry(inode);
259 if ((bits & MDS_INODELOCK_LOOKUP) &&
260 !cfs_list_empty(&lli->lli_sa_dentry)) {
261 struct ll_dentry_data *lld, *next;
263 cfs_spin_lock(&lli->lli_sa_lock);
264 cfs_list_for_each_entry_safe(lld, next,
267 cfs_list_del_init(&lld->lld_sa_alias);
268 cfs_spin_unlock(&lli->lli_sa_lock);
271 if (inode->i_sb->s_root &&
272 inode != inode->i_sb->s_root->d_inode &&
273 (bits & MDS_INODELOCK_LOOKUP))
274 ll_unhash_aliases(inode);
285 __u32 ll_i2suppgid(struct inode *i)
287 if (cfs_curproc_is_in_groups(i->i_gid))
288 return (__u32)i->i_gid;
293 /* Pack the required supplementary groups into the supplied groups array.
294 * If we don't need to use the groups from the target inode(s) then we
295 * instead pack one or more groups from the user's supplementary group
296 * array in case it might be useful. Not needed if doing an MDS-side upcall. */
297 void ll_i2gids(__u32 *suppgids, struct inode *i1, struct inode *i2)
304 LASSERT(suppgids != NULL);
306 suppgids[0] = ll_i2suppgid(i1);
309 suppgids[1] = ll_i2suppgid(i2);
314 for (i = 0; i < current_ngroups; i++) {
315 if (suppgids[0] == -1) {
316 if (current_groups[i] != suppgids[1])
317 suppgids[0] = current_groups[i];
320 if (suppgids[1] == -1) {
321 if (current_groups[i] != suppgids[0])
322 suppgids[1] = current_groups[i];
330 static void ll_d_add(struct dentry *de, struct inode *inode)
332 CDEBUG(D_DENTRY, "adding inode %p to dentry %p\n", inode, de);
334 if (!list_empty(&de->d_alias)) {
335 spin_unlock(&dcache_lock);
336 CERROR("dentry %.*s %p alias next %p, prev %p\n",
337 de->d_name.len, de->d_name.name, de,
338 de->d_alias.next, de->d_alias.prev);
342 list_add(&de->d_alias, &inode->i_dentry);
346 if (!d_unhashed(de)) {
347 spin_unlock(&dcache_lock);
348 CERROR("dentry %.*s %p hash next %p\n",
349 de->d_name.len, de->d_name.name, de, de->d_hash.next);
352 d_rehash_cond(de, 0);
355 /* Search "inode"'s alias list for a dentry that has the same name and parent
356 * as de. If found, return it. If not found, return de.
357 * Lustre can't use d_add_unique because don't unhash aliases for directory
358 * in ll_revalidate_it. After revaliadate inode will be have hashed aliases
359 * and it triggers BUG_ON in d_instantiate_unique (bug #10954).
361 static struct dentry *ll_find_alias(struct inode *inode, struct dentry *de)
363 struct list_head *tmp;
364 struct dentry *dentry;
365 struct dentry *last_discon = NULL;
367 cfs_spin_lock(&ll_lookup_lock);
368 spin_lock(&dcache_lock);
369 list_for_each(tmp, &inode->i_dentry) {
370 dentry = list_entry(tmp, struct dentry, d_alias);
372 /* We are called here with 'de' already on the aliases list. */
373 if (unlikely(dentry == de)) {
378 if (dentry->d_flags & DCACHE_DISCONNECTED) {
379 LASSERT(last_discon == NULL);
380 last_discon = dentry;
384 if (dentry->d_parent != de->d_parent)
387 if (dentry->d_name.hash != de->d_name.hash)
390 if (dentry->d_name.len != de->d_name.len)
393 if (memcmp(dentry->d_name.name, de->d_name.name,
394 de->d_name.len) != 0)
400 unlock_dentry(dentry);
401 ll_dops_init(dentry, 0);
402 d_rehash_cond(dentry, 0); /* avoid taking dcache_lock inside */
403 spin_unlock(&dcache_lock);
404 cfs_spin_unlock(&ll_lookup_lock);
406 CDEBUG(D_DENTRY, "alias dentry %.*s (%p) parent %p inode %p "
407 "refc %d\n", de->d_name.len, de->d_name.name, de,
408 de->d_parent, de->d_inode, atomic_read(&de->d_count));
413 CDEBUG(D_DENTRY, "Reuse disconnected dentry %p inode %p "
414 "refc %d\n", last_discon, last_discon->d_inode,
415 atomic_read(&last_discon->d_count));
416 dget_locked(last_discon);
417 lock_dentry(last_discon);
418 last_discon->d_flags |= DCACHE_LUSTRE_INVALID;
419 unlock_dentry(last_discon);
420 spin_unlock(&dcache_lock);
421 cfs_spin_unlock(&ll_lookup_lock);
422 ll_dops_init(last_discon, 1);
424 d_move(last_discon, de);
429 de->d_flags |= DCACHE_LUSTRE_INVALID;
433 spin_unlock(&dcache_lock);
434 cfs_spin_unlock(&ll_lookup_lock);
439 void ll_lookup_it_alias(struct dentry **de, struct inode *inode, __u32 bits)
441 struct dentry *save = *de;
444 ll_dops_init(*de, 1);
445 *de = ll_find_alias(inode, *de);
447 struct ll_dentry_data *lld = ll_d2d(*de);
449 /* just make sure the ll_dentry_data is ready */
450 if (unlikely(lld == NULL))
451 ll_dops_init(*de, 1);
453 /* we have lookup look - unhide dentry */
454 if (bits & MDS_INODELOCK_LOOKUP) {
456 (*de)->d_flags &= ~DCACHE_LUSTRE_INVALID;
462 int ll_lookup_it_finish(struct ptlrpc_request *request,
463 struct lookup_intent *it, void *data)
465 struct it_cb_data *icbd = data;
466 struct dentry **de = icbd->icbd_childp;
467 struct inode *parent = icbd->icbd_parent;
468 struct ll_sb_info *sbi = ll_i2sbi(parent);
469 struct inode *inode = NULL;
473 /* NB 1 request reference will be taken away by ll_intent_lock()
475 if (!it_disposition(it, DISP_LOOKUP_NEG)) {
478 rc = ll_prep_inode(&inode, request, (*de)->d_sb);
482 CDEBUG(D_DLMTRACE, "setting l_data to inode %p (%lu/%u)\n",
483 inode, inode->i_ino, inode->i_generation);
484 md_set_lock_data(sbi->ll_md_exp,
485 &it->d.lustre.it_lock_handle, inode, &bits);
487 if (icbd->bits != NULL)
489 if (icbd->icbd_alias != NULL) {
490 *icbd->icbd_alias = inode;
494 /* We used to query real size from OSTs here, but actually
495 this is not needed. For stat() calls size would be updated
496 from subsequent do_revalidate()->ll_inode_revalidate_it() in
498 vfs_getattr_it->ll_getattr()->ll_inode_revalidate_it() in 2.6
499 Everybody else who needs correct file size would call
500 cl_glimpse_size or some equivalent themselves anyway.
501 Also see bug 7198. */
502 ll_lookup_it_alias(de, inode, bits);
504 ll_dops_init(*de, 1);
505 /* Check that parent has UPDATE lock. If there is none, we
506 cannot afford to hash this dentry (done by ll_d_add) as it
507 might get picked up later when UPDATE lock will appear */
508 if (ll_have_md_lock(parent, MDS_INODELOCK_UPDATE)) {
509 spin_lock(&dcache_lock);
511 spin_unlock(&dcache_lock);
513 /* negative lookup - and don't have update lock to
516 (*de)->d_flags |= DCACHE_LUSTRE_INVALID;
519 (*de)->d_inode = NULL;
520 /* We do not want to hash the dentry if don`t have a
521 * lock, but if this dentry is later used in d_move,
522 * we'd hit uninitialised list head d_hash, so we just
523 * do this to init d_hash field but leave dentry
524 * unhashed. (bug 10796). */
533 static struct dentry *ll_lookup_it(struct inode *parent, struct dentry *dentry,
534 struct lookup_intent *it, int lookup_flags)
536 struct lookup_intent lookup_it = { .it_op = IT_LOOKUP };
537 struct dentry *save = dentry, *retval;
538 struct ptlrpc_request *req = NULL;
539 struct md_op_data *op_data;
540 struct it_cb_data icbd;
545 if (dentry->d_name.len > ll_i2sbi(parent)->ll_namelen)
546 RETURN(ERR_PTR(-ENAMETOOLONG));
548 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),intent=%s\n",
549 dentry->d_name.len, dentry->d_name.name, parent->i_ino,
550 parent->i_generation, parent, LL_IT2STR(it));
552 if (d_mountpoint(dentry))
553 CERROR("Tell Peter, lookup on mtpt, it %s\n", LL_IT2STR(it));
555 ll_frob_intent(&it, &lookup_it);
557 /* As do_lookup is called before follow_mount, root dentry may be left
558 * not valid, revalidate it here. */
559 if (parent->i_sb->s_root && (parent->i_sb->s_root->d_inode == parent) &&
560 (it->it_op & (IT_OPEN | IT_CREAT))) {
561 rc = ll_inode_revalidate_it(parent->i_sb->s_root, it);
566 if (it->it_op == IT_GETATTR) {
567 first = ll_statahead_enter(parent, &dentry, 1);
569 ll_statahead_exit(parent, dentry, first);
571 RETURN(retval = dentry);
575 icbd.icbd_childp = &dentry;
576 icbd.icbd_parent = parent;
577 icbd.icbd_alias = NULL;
580 if (it->it_op & IT_CREAT ||
581 (it->it_op & IT_OPEN && it->it_create_mode & O_CREAT))
582 opc = LUSTRE_OPC_CREATE;
584 opc = LUSTRE_OPC_ANY;
586 op_data = ll_prep_md_op_data(NULL, parent, NULL, dentry->d_name.name,
587 dentry->d_name.len, lookup_flags, opc,
590 RETURN((void *)op_data);
592 it->it_create_mode &= ~current->fs->umask;
594 rc = md_intent_lock(ll_i2mdexp(parent), op_data, NULL, 0, it,
595 lookup_flags, &req, ll_md_blocking_ast, 0);
596 ll_finish_md_op_data(op_data);
598 GOTO(out, retval = ERR_PTR(rc));
600 rc = ll_lookup_it_finish(req, it, &icbd);
602 ll_intent_release(it);
603 GOTO(out, retval = ERR_PTR(rc));
606 if (first == -EEXIST)
607 ll_statahead_mark(parent, dentry);
609 if ((it->it_op & IT_OPEN) && dentry->d_inode &&
610 !S_ISREG(dentry->d_inode->i_mode) &&
611 !S_ISDIR(dentry->d_inode->i_mode)) {
612 ll_release_openhandle(dentry, it);
614 ll_lookup_finish_locks(it, dentry);
617 GOTO(out, retval = NULL);
619 GOTO(out, retval = dentry);
622 ptlrpc_req_finished(req);
626 #ifdef HAVE_VFS_INTENT_PATCHES
627 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
628 struct nameidata *nd)
633 if (nd && nd->flags & LOOKUP_LAST && !(nd->flags & LOOKUP_LINK_NOTLAST))
634 de = ll_lookup_it(parent, dentry, &nd->intent, nd->flags);
636 de = ll_lookup_it(parent, dentry, NULL, 0);
641 struct lookup_intent *ll_convert_intent(struct open_intent *oit,
644 struct lookup_intent *it;
646 OBD_ALLOC(it, sizeof(*it));
648 return ERR_PTR(-ENOMEM);
650 if (lookup_flags & LOOKUP_OPEN) {
652 if (lookup_flags & LOOKUP_CREATE)
653 it->it_op |= IT_CREAT;
654 it->it_create_mode = oit->create_mode;
655 it->it_flags = oit->flags;
657 it->it_op = IT_GETATTR;
660 #ifndef HAVE_FILE_IN_STRUCT_INTENT
661 /* Since there is no way to pass our intent to ll_file_open,
662 * just check the file is there. Actual open will be done
664 if (it->it_op & IT_OPEN)
665 it->it_op = IT_LOOKUP;
671 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
672 struct nameidata *nd)
677 if (nd && !(nd->flags & (LOOKUP_CONTINUE|LOOKUP_PARENT))) {
678 struct lookup_intent *it;
680 #if defined(HAVE_FILE_IN_STRUCT_INTENT) && (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,17))
681 /* Did we came here from failed revalidate just to propagate
683 if (nd->flags & LOOKUP_OPEN)
684 if (IS_ERR(nd->intent.open.file))
685 RETURN((struct dentry *)nd->intent.open.file);
688 if (ll_d2d(dentry) && ll_d2d(dentry)->lld_it) {
689 it = ll_d2d(dentry)->lld_it;
690 ll_d2d(dentry)->lld_it = NULL;
692 if ((nd->flags & LOOKUP_CREATE ) && !(nd->flags & LOOKUP_OPEN)) {
693 /* We are sure this is new dentry, so we need to create
694 our private data and set the dentry ops */
695 ll_dops_init(dentry, 1);
698 it = ll_convert_intent(&nd->intent.open, nd->flags);
700 RETURN((struct dentry *)it);
703 de = ll_lookup_it(parent, dentry, it, nd->flags);
706 if ((nd->flags & LOOKUP_OPEN) && !IS_ERR(dentry)) { /* Open */
707 if (dentry->d_inode &&
708 it_disposition(it, DISP_OPEN_OPEN)) { /* nocreate */
709 #ifdef HAVE_FILE_IN_STRUCT_INTENT
710 if (S_ISFIFO(dentry->d_inode->i_mode)) {
711 // We cannot call open here as it would
714 (struct ptlrpc_request *)
715 it->d.lustre.it_data);
717 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,17))
718 /* 2.6.1[456] have a bug in open_namei() that forgets to check
719 * nd->intent.open.file for error, so we need to return it as lookup's result
722 nd->intent.open.file->private_data = it;
723 filp =lookup_instantiate_filp(nd,dentry,
728 de = (struct dentry *) filp;
731 nd->intent.open.file->private_data = it;
732 (void)lookup_instantiate_filp(nd,dentry,
737 #else /* HAVE_FILE_IN_STRUCT_INTENT */
738 /* Release open handle as we have no way to
739 * pass it to ll_file_open */
740 ll_release_openhandle(dentry, it);
741 #endif /* HAVE_FILE_IN_STRUCT_INTENT */
742 } else if (it_disposition(it, DISP_OPEN_CREATE)) {
743 // XXX This can only reliably work on assumption
744 // that there are NO hashed negative dentries.
745 ll_d2d(dentry)->lld_it = it;
746 it = NULL; /* Will be freed in ll_create_nd */
747 /* We absolutely depend on ll_create_nd to be
748 * called to not leak this intent and possible
749 * data attached to it */
754 ll_intent_release(it);
755 OBD_FREE(it, sizeof(*it));
758 de = ll_lookup_it(parent, dentry, NULL, 0);
765 /* We depend on "mode" being set with the proper file type/umask by now */
766 static struct inode *ll_create_node(struct inode *dir, const char *name,
767 int namelen, const void *data, int datalen,
768 int mode, __u64 extra,
769 struct lookup_intent *it)
771 struct inode *inode = NULL;
772 struct ptlrpc_request *request = NULL;
773 struct ll_sb_info *sbi = ll_i2sbi(dir);
777 LASSERT(it && it->d.lustre.it_disposition);
779 LASSERT(it_disposition(it, DISP_ENQ_CREATE_REF));
780 request = it->d.lustre.it_data;
781 it_clear_disposition(it, DISP_ENQ_CREATE_REF);
782 rc = ll_prep_inode(&inode, request, dir->i_sb);
784 GOTO(out, inode = ERR_PTR(rc));
786 LASSERT(list_empty(&inode->i_dentry));
788 /* We asked for a lock on the directory, but were granted a
789 * lock on the inode. Since we finally have an inode pointer,
790 * stuff it in the lock. */
791 CDEBUG(D_DLMTRACE, "setting l_ast_data to inode %p (%lu/%u)\n",
792 inode, inode->i_ino, inode->i_generation);
793 md_set_lock_data(sbi->ll_md_exp,
794 &it->d.lustre.it_lock_handle, inode, NULL);
797 ptlrpc_req_finished(request);
802 * By the time this is called, we already have created the directory cache
803 * entry for the new file, but it is so far negative - it has no inode.
805 * We defer creating the OBD object(s) until open, to keep the intent and
806 * non-intent code paths similar, and also because we do not have the MDS
807 * inode number before calling ll_create_node() (which is needed for LOV),
808 * so we would need to do yet another RPC to the MDS to store the LOV EA
809 * data on the MDS. If needed, we would pass the PACKED lmm as data and
810 * lmm_size in datalen (the MDS still has code which will handle that).
812 * If the create succeeds, we fill in the inode information
813 * with d_instantiate().
815 static int ll_create_it(struct inode *dir, struct dentry *dentry, int mode,
816 struct lookup_intent *it)
822 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),intent=%s\n",
823 dentry->d_name.len, dentry->d_name.name, dir->i_ino,
824 dir->i_generation, dir, LL_IT2STR(it));
826 rc = it_open_error(DISP_OPEN_CREATE, it);
830 inode = ll_create_node(dir, dentry->d_name.name, dentry->d_name.len,
831 NULL, 0, mode, 0, it);
833 RETURN(PTR_ERR(inode));
836 d_instantiate(dentry, inode);
837 /* Negative dentry may be unhashed if parent does not have UPDATE lock,
838 * but some callers, e.g. do_coredump, expect dentry to be hashed after
839 * successful create. Hash it here. */
840 spin_lock(&dcache_lock);
841 if (d_unhashed(dentry))
842 d_rehash_cond(dentry, 0);
843 spin_unlock(&dcache_lock);
847 static void ll_update_times(struct ptlrpc_request *request,
850 struct mdt_body *body = req_capsule_server_get(&request->rq_pill,
854 if (body->valid & OBD_MD_FLMTIME &&
855 body->mtime > LTIME_S(inode->i_mtime)) {
856 CDEBUG(D_INODE, "setting ino %lu mtime from %lu to "LPU64"\n",
857 inode->i_ino, LTIME_S(inode->i_mtime), body->mtime);
858 LTIME_S(inode->i_mtime) = body->mtime;
860 if (body->valid & OBD_MD_FLCTIME &&
861 body->ctime > LTIME_S(inode->i_ctime))
862 LTIME_S(inode->i_ctime) = body->ctime;
865 static int ll_new_node(struct inode *dir, struct qstr *name,
866 const char *tgt, int mode, int rdev,
867 struct dentry *dchild, __u32 opc)
869 struct ptlrpc_request *request = NULL;
870 struct md_op_data *op_data;
871 struct inode *inode = NULL;
872 struct ll_sb_info *sbi = ll_i2sbi(dir);
877 if (unlikely(tgt != NULL))
878 tgt_len = strlen(tgt) + 1;
880 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
881 name->len, 0, opc, NULL);
883 GOTO(err_exit, err = PTR_ERR(op_data));
885 err = md_create(sbi->ll_md_exp, op_data, tgt, tgt_len, mode,
886 cfs_curproc_fsuid(), cfs_curproc_fsgid(),
887 cfs_curproc_cap_pack(), rdev, &request);
888 ll_finish_md_op_data(op_data);
892 ll_update_times(request, dir);
895 err = ll_prep_inode(&inode, request, dchild->d_sb);
900 d_instantiate(dchild, inode);
904 ptlrpc_req_finished(request);
909 static int ll_mknod_generic(struct inode *dir, struct qstr *name, int mode,
910 unsigned rdev, struct dentry *dchild)
915 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p) mode %o dev %x\n",
916 name->len, name->name, dir->i_ino, dir->i_generation, dir,
919 mode &= ~current->fs->umask;
921 switch (mode & S_IFMT) {
923 mode |= S_IFREG; /* for mode = 0 case, fallthrough */
929 err = ll_new_node(dir, name, NULL, mode, rdev, dchild,
941 #ifndef HAVE_VFS_INTENT_PATCHES
942 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
943 int mode, struct nameidata *nd)
945 struct lookup_intent *it = ll_d2d(dentry)->lld_it;
949 return ll_mknod_generic(dir, &dentry->d_name, mode, 0, dentry);
951 ll_d2d(dentry)->lld_it = NULL;
953 /* Was there an error? Propagate it! */
954 if (it->d.lustre.it_status) {
955 rc = it->d.lustre.it_status;
959 rc = ll_create_it(dir, dentry, mode, it);
960 #ifdef HAVE_FILE_IN_STRUCT_INTENT
961 if (nd && (nd->flags & LOOKUP_OPEN) && dentry->d_inode) { /* Open */
962 nd->intent.open.file->private_data = it;
963 lookup_instantiate_filp(nd, dentry, NULL);
966 ll_release_openhandle(dentry,it);
970 ll_intent_release(it);
971 OBD_FREE(it, sizeof(*it));
976 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
977 int mode, struct nameidata *nd)
979 if (!nd || !nd->intent.d.lustre.it_disposition)
980 /* No saved request? Just mknod the file */
981 return ll_mknod_generic(dir, &dentry->d_name, mode, 0, dentry);
983 return ll_create_it(dir, dentry, mode, &nd->intent);
987 static int ll_symlink_generic(struct inode *dir, struct qstr *name,
988 const char *tgt, struct dentry *dchild)
993 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),target=%.*s\n",
994 name->len, name->name, dir->i_ino, dir->i_generation,
997 err = ll_new_node(dir, name, (char *)tgt, S_IFLNK | S_IRWXUGO,
998 0, dchild, LUSTRE_OPC_SYMLINK);
1002 static int ll_link_generic(struct inode *src, struct inode *dir,
1003 struct qstr *name, struct dentry *dchild)
1005 struct ll_sb_info *sbi = ll_i2sbi(dir);
1006 struct ptlrpc_request *request = NULL;
1007 struct md_op_data *op_data;
1012 "VFS Op: inode=%lu/%u(%p), dir=%lu/%u(%p), target=%.*s\n",
1013 src->i_ino, src->i_generation, src, dir->i_ino,
1014 dir->i_generation, dir, name->len, name->name);
1016 op_data = ll_prep_md_op_data(NULL, src, dir, name->name, name->len,
1017 0, LUSTRE_OPC_ANY, NULL);
1018 if (IS_ERR(op_data))
1019 RETURN(PTR_ERR(op_data));
1021 err = md_link(sbi->ll_md_exp, op_data, &request);
1022 ll_finish_md_op_data(op_data);
1028 ll_update_times(request, dir);
1031 ptlrpc_req_finished(request);
1035 static int ll_mkdir_generic(struct inode *dir, struct qstr *name,
1036 int mode, struct dentry *dchild)
1042 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
1043 name->len, name->name, dir->i_ino, dir->i_generation, dir);
1045 mode = (mode & (S_IRWXUGO|S_ISVTX) & ~current->fs->umask) | S_IFDIR;
1046 err = ll_new_node(dir, name, NULL, mode, 0, dchild, LUSTRE_OPC_MKDIR);
1051 /* Try to find the child dentry by its name.
1052 If found, put the result fid into @fid. */
1053 static void ll_get_child_fid(struct inode * dir, struct qstr *name,
1056 struct dentry *parent, *child;
1058 parent = list_entry(dir->i_dentry.next, struct dentry, d_alias);
1059 child = d_lookup(parent, name);
1062 *fid = *ll_inode2fid(child->d_inode);
1067 static int ll_rmdir_generic(struct inode *dir, struct dentry *dparent,
1068 struct dentry *dchild, struct qstr *name)
1070 struct ptlrpc_request *request = NULL;
1071 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);
1078 if (unlikely(ll_d_mountpoint(dparent, dchild, name)))
1081 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len,
1082 S_IFDIR, LUSTRE_OPC_ANY, NULL);
1083 if (IS_ERR(op_data))
1084 RETURN(PTR_ERR(op_data));
1086 ll_get_child_fid(dir, name, &op_data->op_fid3);
1087 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1088 ll_finish_md_op_data(op_data);
1090 ll_update_times(request, dir);
1091 ptlrpc_req_finished(request);
1095 int ll_objects_destroy(struct ptlrpc_request *request, struct inode *dir)
1097 struct mdt_body *body;
1098 struct lov_mds_md *eadata;
1099 struct lov_stripe_md *lsm = NULL;
1100 struct obd_trans_info oti = { 0 };
1102 struct obd_capa *oc = NULL;
1106 /* req is swabbed so this is safe */
1107 body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
1108 if (!(body->valid & OBD_MD_FLEASIZE))
1111 if (body->eadatasize == 0) {
1112 CERROR("OBD_MD_FLEASIZE set but eadatasize zero\n");
1113 GOTO(out, rc = -EPROTO);
1116 /* The MDS sent back the EA because we unlinked the last reference
1117 * to this file. Use this EA to unlink the objects on the OST.
1118 * It's opaque so we don't swab here; we leave it to obd_unpackmd() to
1119 * check it is complete and sensible. */
1120 eadata = req_capsule_server_sized_get(&request->rq_pill, &RMF_MDT_MD,
1122 LASSERT(eadata != NULL);
1124 rc = obd_unpackmd(ll_i2dtexp(dir), &lsm, eadata, body->eadatasize);
1126 CERROR("obd_unpackmd: %d\n", rc);
1129 LASSERT(rc >= sizeof(*lsm));
1133 GOTO(out_free_memmd, rc = -ENOMEM);
1135 oa->o_id = lsm->lsm_object_id;
1136 oa->o_seq = lsm->lsm_object_seq;
1137 oa->o_mode = body->mode & S_IFMT;
1138 oa->o_valid = OBD_MD_FLID | OBD_MD_FLTYPE | OBD_MD_FLGROUP;
1140 if (body->valid & OBD_MD_FLCOOKIE) {
1141 oa->o_valid |= OBD_MD_FLCOOKIE;
1142 oti.oti_logcookies =
1143 req_capsule_server_sized_get(&request->rq_pill,
1145 sizeof(struct llog_cookie) *
1146 lsm->lsm_stripe_count);
1147 if (oti.oti_logcookies == NULL) {
1148 oa->o_valid &= ~OBD_MD_FLCOOKIE;
1149 body->valid &= ~OBD_MD_FLCOOKIE;
1153 if (body->valid & OBD_MD_FLOSSCAPA) {
1154 rc = md_unpack_capa(ll_i2mdexp(dir), request, &RMF_CAPA2, &oc);
1156 GOTO(out_free_memmd, rc);
1159 rc = obd_destroy(ll_i2dtexp(dir), oa, lsm, &oti, ll_i2mdexp(dir), oc);
1163 CERROR("obd destroy objid "LPX64" error %d\n",
1164 lsm->lsm_object_id, rc);
1166 obd_free_memmd(ll_i2dtexp(dir), &lsm);
1171 /* ll_unlink_generic() doesn't update the inode with the new link count.
1172 * Instead, ll_ddelete() and ll_d_iput() will update it based upon if there
1173 * is any lock existing. They will recycle dentries and inodes based upon locks
1175 static int ll_unlink_generic(struct inode *dir, struct dentry *dparent,
1176 struct dentry *dchild, struct qstr *name)
1178 struct ptlrpc_request *request = NULL;
1179 struct md_op_data *op_data;
1182 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
1183 name->len, name->name, dir->i_ino, dir->i_generation, dir);
1186 * XXX: unlink bind mountpoint maybe call to here,
1187 * just check it as vfs_unlink does.
1189 if (unlikely(ll_d_mountpoint(dparent, dchild, name)))
1192 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
1193 name->len, 0, LUSTRE_OPC_ANY, NULL);
1194 if (IS_ERR(op_data))
1195 RETURN(PTR_ERR(op_data));
1197 ll_get_child_fid(dir, name, &op_data->op_fid3);
1198 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1199 ll_finish_md_op_data(op_data);
1203 ll_update_times(request, dir);
1205 rc = ll_objects_destroy(request, dir);
1207 ptlrpc_req_finished(request);
1211 static int ll_rename_generic(struct inode *src, struct dentry *src_dparent,
1212 struct dentry *src_dchild, struct qstr *src_name,
1213 struct inode *tgt, struct dentry *tgt_dparent,
1214 struct dentry *tgt_dchild, struct qstr *tgt_name)
1216 struct ptlrpc_request *request = NULL;
1217 struct ll_sb_info *sbi = ll_i2sbi(src);
1218 struct md_op_data *op_data;
1221 CDEBUG(D_VFSTRACE,"VFS Op:oldname=%.*s,src_dir=%lu/%u(%p),newname=%.*s,"
1222 "tgt_dir=%lu/%u(%p)\n", src_name->len, src_name->name,
1223 src->i_ino, src->i_generation, src, tgt_name->len,
1224 tgt_name->name, tgt->i_ino, tgt->i_generation, tgt);
1226 if (unlikely(ll_d_mountpoint(src_dparent, src_dchild, src_name) ||
1227 ll_d_mountpoint(tgt_dparent, tgt_dchild, tgt_name)))
1230 op_data = ll_prep_md_op_data(NULL, src, tgt, NULL, 0, 0,
1231 LUSTRE_OPC_ANY, NULL);
1232 if (IS_ERR(op_data))
1233 RETURN(PTR_ERR(op_data));
1235 ll_get_child_fid(src, src_name, &op_data->op_fid3);
1236 ll_get_child_fid(tgt, tgt_name, &op_data->op_fid4);
1237 err = md_rename(sbi->ll_md_exp, op_data,
1238 src_name->name, src_name->len,
1239 tgt_name->name, tgt_name->len, &request);
1240 ll_finish_md_op_data(op_data);
1242 ll_update_times(request, src);
1243 ll_update_times(request, tgt);
1244 err = ll_objects_destroy(request, src);
1247 ptlrpc_req_finished(request);
1252 #ifdef HAVE_VFS_INTENT_PATCHES
1253 static int ll_mknod_raw(struct nameidata *nd, int mode, dev_t rdev)
1255 return ll_mknod_generic(nd->dentry->d_inode, &nd->last, mode,rdev,NULL);
1257 static int ll_rename_raw(struct nameidata *srcnd, struct nameidata *tgtnd)
1259 return ll_rename_generic(srcnd->dentry->d_inode, srcnd->dentry,
1261 tgtnd->dentry->d_inode, tgtnd->dentry,
1262 NULL, &tgtnd->last);
1264 static int ll_link_raw(struct nameidata *srcnd, struct nameidata *tgtnd)
1266 return ll_link_generic(srcnd->dentry->d_inode, tgtnd->dentry->d_inode,
1267 &tgtnd->last, NULL);
1269 static int ll_symlink_raw(struct nameidata *nd, const char *tgt)
1271 return ll_symlink_generic(nd->dentry->d_inode, &nd->last, tgt, NULL);
1273 static int ll_rmdir_raw(struct nameidata *nd)
1275 return ll_rmdir_generic(nd->dentry->d_inode, nd->dentry, NULL,
1278 static int ll_mkdir_raw(struct nameidata *nd, int mode)
1280 return ll_mkdir_generic(nd->dentry->d_inode, &nd->last, mode, NULL);
1282 static int ll_unlink_raw(struct nameidata *nd)
1284 return ll_unlink_generic(nd->dentry->d_inode, nd->dentry, NULL,
1289 static int ll_mknod(struct inode *dir, struct dentry *dchild, int mode,
1292 return ll_mknod_generic(dir, &dchild->d_name, mode,
1293 old_encode_dev(rdev), dchild);
1296 static int ll_unlink(struct inode * dir, struct dentry *dentry)
1298 return ll_unlink_generic(dir, NULL, dentry, &dentry->d_name);
1300 static int ll_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1302 return ll_mkdir_generic(dir, &dentry->d_name, mode, dentry);
1304 static int ll_rmdir(struct inode *dir, struct dentry *dentry)
1306 return ll_rmdir_generic(dir, NULL, dentry, &dentry->d_name);
1308 static int ll_symlink(struct inode *dir, struct dentry *dentry,
1309 const char *oldname)
1311 return ll_symlink_generic(dir, &dentry->d_name, oldname, dentry);
1313 static int ll_link(struct dentry *old_dentry, struct inode *dir,
1314 struct dentry *new_dentry)
1316 return ll_link_generic(old_dentry->d_inode, dir, &new_dentry->d_name,
1319 static int ll_rename(struct inode *old_dir, struct dentry *old_dentry,
1320 struct inode *new_dir, struct dentry *new_dentry)
1323 err = ll_rename_generic(old_dir, NULL,
1324 old_dentry, &old_dentry->d_name,
1325 new_dir, NULL, new_dentry,
1326 &new_dentry->d_name);
1328 #ifndef HAVE_FS_RENAME_DOES_D_MOVE
1329 if (!S_ISDIR(old_dentry->d_inode->i_mode))
1331 d_move(old_dentry, new_dentry);
1336 struct inode_operations ll_dir_inode_operations = {
1337 #ifdef HAVE_VFS_INTENT_PATCHES
1338 .link_raw = ll_link_raw,
1339 .unlink_raw = ll_unlink_raw,
1340 .symlink_raw = ll_symlink_raw,
1341 .mkdir_raw = ll_mkdir_raw,
1342 .rmdir_raw = ll_rmdir_raw,
1343 .mknod_raw = ll_mknod_raw,
1344 .rename_raw = ll_rename_raw,
1345 .setattr = ll_setattr,
1346 .setattr_raw = ll_setattr_raw,
1349 .lookup = ll_lookup_nd,
1350 .create = ll_create_nd,
1351 /* We need all these non-raw things for NFSD, to not patch it. */
1352 .unlink = ll_unlink,
1355 .symlink = ll_symlink,
1357 .rename = ll_rename,
1358 .setattr = ll_setattr,
1359 .getattr = ll_getattr,
1360 .permission = ll_inode_permission,
1361 .setxattr = ll_setxattr,
1362 .getxattr = ll_getxattr,
1363 .listxattr = ll_listxattr,
1364 .removexattr = ll_removexattr,
1367 struct inode_operations ll_special_inode_operations = {
1368 #ifdef HAVE_VFS_INTENT_PATCHES
1369 .setattr_raw = ll_setattr_raw,
1371 .setattr = ll_setattr,
1372 .getattr = ll_getattr,
1373 .permission = ll_inode_permission,
1374 .setxattr = ll_setxattr,
1375 .getxattr = ll_getxattr,
1376 .listxattr = ll_listxattr,
1377 .removexattr = ll_removexattr,