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 2008 Sun Microsystems, Inc. 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>
44 #include <linux/dcache.h>
45 #include <linux/buffer_head.h>
47 #define DEBUG_SUBSYSTEM S_LLITE
49 #include <obd_support.h>
50 #include <lustre_fid.h>
51 #include <lustre_lite.h>
52 #include <lustre_dlm.h>
53 #include <lustre_ver.h>
54 #include <lustre_mdc.h>
55 #include "llite_internal.h"
58 * Check if we have something mounted at the named dchild.
59 * In such a case there would always be dentry present.
61 static int ll_d_mountpoint(struct dentry *dparent, struct dentry *dchild,
66 if (unlikely(dchild)) {
67 mounted = d_mountpoint(dchild);
69 dchild = d_lookup(dparent, name);
71 mounted = d_mountpoint(dchild);
78 int ll_unlock(__u32 mode, struct lustre_handle *lockh)
82 ldlm_lock_decref(lockh, mode);
88 /* called from iget5_locked->find_inode() under inode_lock spinlock */
89 static int ll_test_inode(struct inode *inode, void *opaque)
91 struct ll_inode_info *lli = ll_i2info(inode);
92 struct lustre_md *md = opaque;
94 if (unlikely(!(md->body->valid & OBD_MD_FLID))) {
95 CERROR("MDS body missing FID\n");
99 if (!lu_fid_eq(&lli->lli_fid, &md->body->fid1))
105 static int ll_set_inode(struct inode *inode, void *opaque)
107 struct ll_inode_info *lli = ll_i2info(inode);
108 struct mdt_body *body = ((struct lustre_md *)opaque)->body;
110 if (unlikely(!(body->valid & OBD_MD_FLID))) {
111 CERROR("MDS body missing FID\n");
115 lli->lli_fid = body->fid1;
121 * Get an inode by inode number (already instantiated by the intent lookup).
122 * Returns inode or NULL
124 struct inode *ll_iget(struct super_block *sb, ino_t hash,
125 struct lustre_md *md)
131 inode = iget5_locked(sb, hash, ll_test_inode, ll_set_inode, md);
134 if (inode->i_state & I_NEW) {
137 ll_read_inode2(inode, md);
138 rc = cl_inode_init(inode, md);
141 make_bad_inode(inode);
142 unlock_new_inode(inode);
146 unlock_new_inode(inode);
147 } else if (!(inode->i_state & (I_FREEING | I_CLEAR)))
148 ll_update_inode(inode, md);
149 CDEBUG(D_VFSTRACE, "got inode: %p for "DFID"\n",
150 inode, PFID(&md->body->fid1));
155 static void ll_drop_negative_dentry(struct inode *dir)
157 struct dentry *dentry, *tmp_alias, *tmp_subdir;
159 spin_lock(&ll_lookup_lock);
160 spin_lock(&dcache_lock);
162 list_for_each_entry_safe(dentry, tmp_alias,
163 &dir->i_dentry,d_alias) {
164 if (!list_empty(&dentry->d_subdirs)) {
165 struct dentry *child;
166 list_for_each_entry_safe(child, tmp_subdir,
169 /* XXX Print some debug here? */
171 /* Negative dentry. If we were
172 dropping dcache lock, go
173 throught the list again */
174 if (ll_drop_dentry(child))
179 spin_unlock(&dcache_lock);
180 spin_unlock(&ll_lookup_lock);
184 int ll_md_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
185 void *data, int flag)
188 struct lustre_handle lockh;
192 case LDLM_CB_BLOCKING:
193 ldlm_lock2handle(lock, &lockh);
194 rc = ldlm_cli_cancel(&lockh);
196 CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
200 case LDLM_CB_CANCELING: {
201 struct inode *inode = ll_inode_from_lock(lock);
202 __u64 bits = lock->l_policy_data.l_inodebits.bits;
205 /* Invalidate all dentries associated with this inode */
209 LASSERT(lock->l_flags & LDLM_FL_CANCELING);
210 if ((bits & MDS_INODELOCK_LOOKUP) &&
211 ll_have_md_lock(inode, MDS_INODELOCK_LOOKUP))
212 bits &= ~MDS_INODELOCK_LOOKUP;
213 if ((bits & MDS_INODELOCK_UPDATE) &&
214 ll_have_md_lock(inode, MDS_INODELOCK_UPDATE))
215 bits &= ~MDS_INODELOCK_UPDATE;
216 if ((bits & MDS_INODELOCK_OPEN) &&
217 ll_have_md_lock(inode, MDS_INODELOCK_OPEN))
218 bits &= ~MDS_INODELOCK_OPEN;
220 fid = ll_inode2fid(inode);
221 if (lock->l_resource->lr_name.name[0] != fid_seq(fid) ||
222 lock->l_resource->lr_name.name[1] != fid_oid(fid) ||
223 lock->l_resource->lr_name.name[2] != fid_ver(fid)) {
224 LDLM_ERROR(lock, "data mismatch with object "
225 DFID" (%p)", PFID(fid), inode);
228 if (bits & MDS_INODELOCK_OPEN) {
230 switch (lock->l_req_mode) {
241 CERROR("Unexpected lock mode for OPEN lock "
242 "%d, inode %ld\n", lock->l_req_mode,
245 ll_md_real_close(inode, flags);
248 if (bits & MDS_INODELOCK_UPDATE)
249 ll_i2info(inode)->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 (inode->i_sb->s_root &&
260 inode != inode->i_sb->s_root->d_inode &&
261 (bits & MDS_INODELOCK_LOOKUP))
262 ll_unhash_aliases(inode);
273 __u32 ll_i2suppgid(struct inode *i)
275 if (in_group_p(i->i_gid))
276 return (__u32)i->i_gid;
281 /* Pack the required supplementary groups into the supplied groups array.
282 * If we don't need to use the groups from the target inode(s) then we
283 * instead pack one or more groups from the user's supplementary group
284 * array in case it might be useful. Not needed if doing an MDS-side upcall. */
285 void ll_i2gids(__u32 *suppgids, struct inode *i1, struct inode *i2)
292 LASSERT(suppgids != NULL);
294 suppgids[0] = ll_i2suppgid(i1);
297 suppgids[1] = ll_i2suppgid(i2);
302 for (i = 0; i < current_ngroups; i++) {
303 if (suppgids[0] == -1) {
304 if (current_groups[i] != suppgids[1])
305 suppgids[0] = current_groups[i];
308 if (suppgids[1] == -1) {
309 if (current_groups[i] != suppgids[0])
310 suppgids[1] = current_groups[i];
318 static void ll_d_add(struct dentry *de, struct inode *inode)
320 CDEBUG(D_DENTRY, "adding inode %p to dentry %p\n", inode, de);
322 if (!list_empty(&de->d_alias)) {
323 spin_unlock(&dcache_lock);
324 CERROR("dentry %.*s %p alias next %p, prev %p\n",
325 de->d_name.len, de->d_name.name, de,
326 de->d_alias.next, de->d_alias.prev);
330 list_add(&de->d_alias, &inode->i_dentry);
334 if (!d_unhashed(de)) {
335 spin_unlock(&dcache_lock);
336 CERROR("dentry %.*s %p hash next %p\n",
337 de->d_name.len, de->d_name.name, de, de->d_hash.next);
340 d_rehash_cond(de, 0);
343 /* Search "inode"'s alias list for a dentry that has the same name and parent
344 * as de. If found, return it. If not found, return de.
345 * Lustre can't use d_add_unique because don't unhash aliases for directory
346 * in ll_revalidate_it. After revaliadate inode will be have hashed aliases
347 * and it triggers BUG_ON in d_instantiate_unique (bug #10954).
349 static struct dentry *ll_find_alias(struct inode *inode, struct dentry *de)
351 struct list_head *tmp;
352 struct dentry *dentry;
353 struct dentry *last_discon = NULL;
355 spin_lock(&ll_lookup_lock);
356 spin_lock(&dcache_lock);
357 list_for_each(tmp, &inode->i_dentry) {
358 dentry = list_entry(tmp, struct dentry, d_alias);
360 /* We are called here with 'de' already on the aliases list. */
361 if (unlikely(dentry == de)) {
366 if (dentry->d_flags & DCACHE_DISCONNECTED) {
367 LASSERT(last_discon == NULL);
368 last_discon = dentry;
372 if (dentry->d_parent != de->d_parent)
375 if (dentry->d_name.hash != de->d_name.hash)
378 if (dentry->d_name.len != de->d_name.len)
381 if (memcmp(dentry->d_name.name, de->d_name.name,
382 de->d_name.len) != 0)
388 unlock_dentry(dentry);
389 ll_dops_init(dentry, 0);
390 d_rehash_cond(dentry, 0); /* avoid taking dcache_lock inside */
391 spin_unlock(&dcache_lock);
392 spin_unlock(&ll_lookup_lock);
394 CDEBUG(D_DENTRY, "alias dentry %.*s (%p) parent %p inode %p "
395 "refc %d\n", de->d_name.len, de->d_name.name, de,
396 de->d_parent, de->d_inode, atomic_read(&de->d_count));
401 CDEBUG(D_DENTRY, "Reuse disconnected dentry %p inode %p "
402 "refc %d\n", last_discon, last_discon->d_inode,
403 atomic_read(&last_discon->d_count));
404 dget_locked(last_discon);
405 lock_dentry(last_discon);
406 last_discon->d_flags |= DCACHE_LUSTRE_INVALID;
407 unlock_dentry(last_discon);
408 spin_unlock(&dcache_lock);
409 spin_unlock(&ll_lookup_lock);
410 ll_dops_init(last_discon, 1);
412 d_move(last_discon, de);
416 de->d_flags |= DCACHE_LUSTRE_INVALID;
419 spin_unlock(&dcache_lock);
420 spin_unlock(&ll_lookup_lock);
425 int ll_lookup_it_finish(struct ptlrpc_request *request,
426 struct lookup_intent *it, void *data)
428 struct it_cb_data *icbd = data;
429 struct dentry **de = icbd->icbd_childp;
430 struct inode *parent = icbd->icbd_parent;
431 struct ll_sb_info *sbi = ll_i2sbi(parent);
432 struct inode *inode = NULL;
436 /* NB 1 request reference will be taken away by ll_intent_lock()
438 if (!it_disposition(it, DISP_LOOKUP_NEG)) {
439 struct dentry *save = *de;
442 rc = ll_prep_inode(&inode, request, (*de)->d_sb);
446 CDEBUG(D_DLMTRACE, "setting l_data to inode %p (%lu/%u)\n",
447 inode, inode->i_ino, inode->i_generation);
448 md_set_lock_data(sbi->ll_md_exp,
449 &it->d.lustre.it_lock_handle, inode, &bits);
451 /* We used to query real size from OSTs here, but actually
452 this is not needed. For stat() calls size would be updated
453 from subsequent do_revalidate()->ll_inode_revalidate_it() in
455 vfs_getattr_it->ll_getattr()->ll_inode_revalidate_it() in 2.6
456 Everybody else who needs correct file size would call
457 cl_glimpse_size or some equivalent themselves anyway.
458 Also see bug 7198. */
460 ll_dops_init(*de, 1);
461 *de = ll_find_alias(inode, *de);
463 struct ll_dentry_data *lld = ll_d2d(*de);
465 /* just make sure the ll_dentry_data is ready */
466 if (unlikely(lld == NULL)) {
469 if (likely(lld != NULL))
470 lld->lld_sa_generation = 0;
473 /* we have lookup look - unhide dentry */
474 if (bits & MDS_INODELOCK_LOOKUP) {
476 (*de)->d_flags &= ~(DCACHE_LUSTRE_INVALID);
480 ll_dops_init(*de, 1);
481 /* Check that parent has UPDATE lock. If there is none, we
482 cannot afford to hash this dentry (done by ll_d_add) as it
483 might get picked up later when UPDATE lock will appear */
484 if (ll_have_md_lock(parent, MDS_INODELOCK_UPDATE)) {
485 spin_lock(&dcache_lock);
487 spin_unlock(&dcache_lock);
489 /* negative lookup - and don't have update lock to
492 (*de)->d_flags |= DCACHE_LUSTRE_INVALID;
495 (*de)->d_inode = NULL;
496 /* We do not want to hash the dentry if don`t have a
497 * lock, but if this dentry is later used in d_move,
498 * we'd hit uninitialised list head d_hash, so we just
499 * do this to init d_hash field but leave dentry
500 * unhashed. (bug 10796). */
509 static struct dentry *ll_lookup_it(struct inode *parent, struct dentry *dentry,
510 struct lookup_intent *it, int lookup_flags)
512 struct lookup_intent lookup_it = { .it_op = IT_LOOKUP };
513 struct dentry *save = dentry, *retval;
514 struct ptlrpc_request *req = NULL;
515 struct md_op_data *op_data;
516 struct it_cb_data icbd;
521 if (dentry->d_name.len > ll_i2sbi(parent)->ll_namelen)
522 RETURN(ERR_PTR(-ENAMETOOLONG));
524 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),intent=%s\n",
525 dentry->d_name.len, dentry->d_name.name, parent->i_ino,
526 parent->i_generation, parent, LL_IT2STR(it));
528 if (d_mountpoint(dentry))
529 CERROR("Tell Peter, lookup on mtpt, it %s\n", LL_IT2STR(it));
531 ll_frob_intent(&it, &lookup_it);
533 /* As do_lookup is called before follow_mount, root dentry may be left
534 * not valid, revalidate it here. */
535 if (parent->i_sb->s_root && (parent->i_sb->s_root->d_inode == parent) &&
536 (it->it_op & (IT_OPEN | IT_CREAT))) {
537 rc = ll_inode_revalidate_it(parent->i_sb->s_root, it);
542 if (it->it_op == IT_GETATTR) {
543 first = ll_statahead_enter(parent, &dentry, 1);
545 ll_statahead_exit(parent, dentry, first);
547 RETURN(retval = dentry);
551 icbd.icbd_childp = &dentry;
552 icbd.icbd_parent = parent;
554 if (it->it_op & IT_CREAT ||
555 (it->it_op & IT_OPEN && it->it_create_mode & O_CREAT))
556 opc = LUSTRE_OPC_CREATE;
558 opc = LUSTRE_OPC_ANY;
560 op_data = ll_prep_md_op_data(NULL, parent, NULL, dentry->d_name.name,
561 dentry->d_name.len, lookup_flags, opc,
564 RETURN((void *)op_data);
566 it->it_create_mode &= ~current->fs->umask;
568 rc = md_intent_lock(ll_i2mdexp(parent), op_data, NULL, 0, it,
569 lookup_flags, &req, ll_md_blocking_ast, 0);
570 ll_finish_md_op_data(op_data);
572 GOTO(out, retval = ERR_PTR(rc));
574 rc = ll_lookup_it_finish(req, it, &icbd);
576 ll_intent_release(it);
577 GOTO(out, retval = ERR_PTR(rc));
580 if (first == -EEXIST)
581 ll_statahead_mark(parent, dentry);
583 if ((it->it_op & IT_OPEN) && dentry->d_inode &&
584 !S_ISREG(dentry->d_inode->i_mode) &&
585 !S_ISDIR(dentry->d_inode->i_mode)) {
586 ll_release_openhandle(dentry, it);
588 ll_lookup_finish_locks(it, dentry);
591 GOTO(out, retval = NULL);
593 GOTO(out, retval = dentry);
596 ptlrpc_req_finished(req);
600 #ifdef HAVE_VFS_INTENT_PATCHES
601 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
602 struct nameidata *nd)
607 if (nd && nd->flags & LOOKUP_LAST && !(nd->flags & LOOKUP_LINK_NOTLAST))
608 de = ll_lookup_it(parent, dentry, &nd->intent, nd->flags);
610 de = ll_lookup_it(parent, dentry, NULL, 0);
615 struct lookup_intent *ll_convert_intent(struct open_intent *oit,
618 struct lookup_intent *it;
620 OBD_ALLOC(it, sizeof(*it));
622 return ERR_PTR(-ENOMEM);
624 if (lookup_flags & LOOKUP_OPEN) {
626 if (lookup_flags & LOOKUP_CREATE)
627 it->it_op |= IT_CREAT;
628 it->it_create_mode = oit->create_mode;
629 it->it_flags = oit->flags;
631 it->it_op = IT_GETATTR;
634 #ifndef HAVE_FILE_IN_STRUCT_INTENT
635 /* Since there is no way to pass our intent to ll_file_open,
636 * just check the file is there. Actual open will be done
638 if (it->it_op & IT_OPEN)
639 it->it_op = IT_LOOKUP;
645 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
646 struct nameidata *nd)
651 if (nd && !(nd->flags & (LOOKUP_CONTINUE|LOOKUP_PARENT))) {
652 struct lookup_intent *it;
654 #if defined(HAVE_FILE_IN_STRUCT_INTENT) && (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,17))
655 /* Did we came here from failed revalidate just to propagate
657 if (nd->flags & LOOKUP_OPEN)
658 if (IS_ERR(nd->intent.open.file))
659 RETURN((struct dentry *)nd->intent.open.file);
662 if (ll_d2d(dentry) && ll_d2d(dentry)->lld_it) {
663 it = ll_d2d(dentry)->lld_it;
664 ll_d2d(dentry)->lld_it = NULL;
666 it = ll_convert_intent(&nd->intent.open, nd->flags);
668 RETURN((struct dentry *)it);
671 de = ll_lookup_it(parent, dentry, it, nd->flags);
674 if ((nd->flags & LOOKUP_OPEN) && !IS_ERR(dentry)) { /* Open */
675 if (dentry->d_inode &&
676 it_disposition(it, DISP_OPEN_OPEN)) { /* nocreate */
677 #ifdef HAVE_FILE_IN_STRUCT_INTENT
678 if (S_ISFIFO(dentry->d_inode->i_mode)) {
679 // We cannot call open here as it would
682 (struct ptlrpc_request *)
683 it->d.lustre.it_data);
685 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,17))
686 /* 2.6.1[456] have a bug in open_namei() that forgets to check
687 * nd->intent.open.file for error, so we need to return it as lookup's result
690 nd->intent.open.file->private_data = it;
691 filp =lookup_instantiate_filp(nd,dentry,
696 de = (struct dentry *) filp;
699 nd->intent.open.file->private_data = it;
700 (void)lookup_instantiate_filp(nd,dentry,
705 #else /* HAVE_FILE_IN_STRUCT_INTENT */
706 /* Release open handle as we have no way to
707 * pass it to ll_file_open */
708 ll_release_openhandle(dentry, it);
709 #endif /* HAVE_FILE_IN_STRUCT_INTENT */
710 } else if (it_disposition(it, DISP_OPEN_CREATE)) {
711 // XXX This can only reliably work on assumption
712 // that there are NO hashed negative dentries.
713 ll_d2d(dentry)->lld_it = it;
714 it = NULL; /* Will be freed in ll_create_nd */
715 /* We absolutely depend on ll_create_nd to be
716 * called to not leak this intent and possible
717 * data attached to it */
722 ll_intent_release(it);
723 OBD_FREE(it, sizeof(*it));
726 de = ll_lookup_it(parent, dentry, NULL, 0);
734 * check new allocated inode, new inode shld not have any valid alias
736 static void ll_validate_new_inode(struct inode *new)
738 struct list_head *lp;
739 struct dentry * dentry;
743 if (list_empty(&new->i_dentry))
746 spin_lock(&dcache_lock);
747 list_for_each(lp, &new->i_dentry) {
748 dentry = list_entry(lp, struct dentry, d_alias);
749 if (!d_unhashed(dentry) && !(dentry->d_flags & DCACHE_LUSTRE_INVALID)){
756 spin_unlock(&dcache_lock);
758 if (need_inval && !in_recheck) {
759 /* kill all old inode's data pages */
760 truncate_inode_pages(new->i_mapping, 0);
762 /* invalidate all dirent and recheck inode */
763 ll_unhash_aliases(new);
769 /* We depend on "mode" being set with the proper file type/umask by now */
770 static struct inode *ll_create_node(struct inode *dir, const char *name,
771 int namelen, const void *data, int datalen,
772 int mode, __u64 extra,
773 struct lookup_intent *it)
775 struct inode *inode = NULL;
776 struct ptlrpc_request *request = NULL;
777 struct ll_sb_info *sbi = ll_i2sbi(dir);
781 LASSERT(it && it->d.lustre.it_disposition);
783 LASSERT(it_disposition(it, DISP_ENQ_CREATE_REF));
784 request = it->d.lustre.it_data;
785 it_clear_disposition(it, DISP_ENQ_CREATE_REF);
786 rc = ll_prep_inode(&inode, request, dir->i_sb);
788 GOTO(out, inode = ERR_PTR(rc));
790 ll_validate_new_inode(inode);
792 /* We asked for a lock on the directory, but were granted a
793 * lock on the inode. Since we finally have an inode pointer,
794 * stuff it in the lock. */
795 CDEBUG(D_DLMTRACE, "setting l_ast_data to inode %p (%lu/%u)\n",
796 inode, inode->i_ino, inode->i_generation);
797 md_set_lock_data(sbi->ll_md_exp,
798 &it->d.lustre.it_lock_handle, inode, NULL);
801 ptlrpc_req_finished(request);
806 * By the time this is called, we already have created the directory cache
807 * entry for the new file, but it is so far negative - it has no inode.
809 * We defer creating the OBD object(s) until open, to keep the intent and
810 * non-intent code paths similar, and also because we do not have the MDS
811 * inode number before calling ll_create_node() (which is needed for LOV),
812 * so we would need to do yet another RPC to the MDS to store the LOV EA
813 * data on the MDS. If needed, we would pass the PACKED lmm as data and
814 * lmm_size in datalen (the MDS still has code which will handle that).
816 * If the create succeeds, we fill in the inode information
817 * with d_instantiate().
819 static int ll_create_it(struct inode *dir, struct dentry *dentry, int mode,
820 struct lookup_intent *it)
826 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),intent=%s\n",
827 dentry->d_name.len, dentry->d_name.name, dir->i_ino,
828 dir->i_generation, dir, LL_IT2STR(it));
830 rc = it_open_error(DISP_OPEN_CREATE, it);
834 inode = ll_create_node(dir, dentry->d_name.name, dentry->d_name.len,
835 NULL, 0, mode, 0, it);
837 RETURN(PTR_ERR(inode));
840 /* it might been set during parent dir revalidation */
841 dentry->d_flags &= ~DCACHE_LUSTRE_INVALID;
842 d_instantiate(dentry, inode);
843 /* Negative dentry may be unhashed if parent does not have UPDATE lock,
844 * but some callers, e.g. do_coredump, expect dentry to be hashed after
845 * successful create. Hash it here. */
846 spin_lock(&dcache_lock);
847 if (d_unhashed(dentry))
848 d_rehash_cond(dentry, 0);
849 spin_unlock(&dcache_lock);
853 static void ll_update_times(struct ptlrpc_request *request,
856 struct mdt_body *body = req_capsule_server_get(&request->rq_pill,
860 /* mtime is always updated with ctime, but can be set in past.
861 As write and utime(2) may happen within 1 second, and utime's
862 mtime has a priority over write's one, so take mtime from mds
863 for the same ctimes. */
864 if (body->valid & OBD_MD_FLCTIME &&
865 body->ctime >= LTIME_S(inode->i_ctime)) {
866 LTIME_S(inode->i_ctime) = body->ctime;
868 if (body->valid & OBD_MD_FLMTIME) {
869 CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
870 "to "LPU64"\n", inode->i_ino,
871 LTIME_S(inode->i_mtime), body->mtime);
872 LTIME_S(inode->i_mtime) = body->mtime;
877 static int ll_new_node(struct inode *dir, struct qstr *name,
878 const char *tgt, int mode, int rdev,
879 struct dentry *dchild, __u32 opc)
881 struct ptlrpc_request *request = NULL;
882 struct md_op_data *op_data;
883 struct inode *inode = NULL;
884 struct ll_sb_info *sbi = ll_i2sbi(dir);
889 if (unlikely(tgt != NULL))
890 tgt_len = strlen(tgt) + 1;
892 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
893 name->len, 0, opc, NULL);
895 GOTO(err_exit, err = PTR_ERR(op_data));
897 err = md_create(sbi->ll_md_exp, op_data, tgt, tgt_len, mode,
898 current->fsuid, current->fsgid,
899 cfs_curproc_cap_pack(), rdev, &request);
900 ll_finish_md_op_data(op_data);
904 ll_update_times(request, dir);
907 err = ll_prep_inode(&inode, request, dchild->d_sb);
912 d_instantiate(dchild, inode);
916 ptlrpc_req_finished(request);
921 static int ll_mknod_generic(struct inode *dir, struct qstr *name, int mode,
922 unsigned rdev, struct dentry *dchild)
927 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p) mode %o dev %x\n",
928 name->len, name->name, dir->i_ino, dir->i_generation, dir,
931 mode &= ~current->fs->umask;
933 switch (mode & S_IFMT) {
935 mode |= S_IFREG; /* for mode = 0 case, fallthrough */
941 err = ll_new_node(dir, name, NULL, mode, rdev, dchild,
953 #ifndef HAVE_VFS_INTENT_PATCHES
954 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
955 int mode, struct nameidata *nd)
957 struct lookup_intent *it = ll_d2d(dentry)->lld_it;
961 return ll_mknod_generic(dir, &dentry->d_name, mode, 0, dentry);
963 ll_d2d(dentry)->lld_it = NULL;
965 /* Was there an error? Propagate it! */
966 if (it->d.lustre.it_status) {
967 rc = it->d.lustre.it_status;
971 rc = ll_create_it(dir, dentry, mode, it);
972 #ifdef HAVE_FILE_IN_STRUCT_INTENT
973 if (nd && (nd->flags & LOOKUP_OPEN) && dentry->d_inode) { /* Open */
974 nd->intent.open.file->private_data = it;
975 lookup_instantiate_filp(nd, dentry, NULL);
978 ll_release_openhandle(dentry,it);
982 ll_intent_release(it);
983 OBD_FREE(it, sizeof(*it));
988 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
989 int mode, struct nameidata *nd)
991 if (!nd || !nd->intent.d.lustre.it_disposition)
992 /* No saved request? Just mknod the file */
993 return ll_mknod_generic(dir, &dentry->d_name, mode, 0, dentry);
995 return ll_create_it(dir, dentry, mode, &nd->intent);
999 static int ll_symlink_generic(struct inode *dir, struct qstr *name,
1000 const char *tgt, struct dentry *dchild)
1005 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),target=%.*s\n",
1006 name->len, name->name, dir->i_ino, dir->i_generation,
1009 err = ll_new_node(dir, name, (char *)tgt, S_IFLNK | S_IRWXUGO,
1010 0, dchild, LUSTRE_OPC_SYMLINK);
1014 static int ll_link_generic(struct inode *src, struct inode *dir,
1015 struct qstr *name, struct dentry *dchild)
1017 struct ll_sb_info *sbi = ll_i2sbi(dir);
1018 struct ptlrpc_request *request = NULL;
1019 struct md_op_data *op_data;
1024 "VFS Op: inode=%lu/%u(%p), dir=%lu/%u(%p), target=%.*s\n",
1025 src->i_ino, src->i_generation, src, dir->i_ino,
1026 dir->i_generation, dir, name->len, name->name);
1028 op_data = ll_prep_md_op_data(NULL, src, dir, name->name, name->len,
1029 0, LUSTRE_OPC_ANY, NULL);
1030 if (IS_ERR(op_data))
1031 RETURN(PTR_ERR(op_data));
1033 err = md_link(sbi->ll_md_exp, op_data, &request);
1034 ll_finish_md_op_data(op_data);
1040 ll_update_times(request, dir);
1043 ptlrpc_req_finished(request);
1047 static int ll_mkdir_generic(struct inode *dir, struct qstr *name,
1048 int mode, struct dentry *dchild)
1054 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
1055 name->len, name->name, dir->i_ino, dir->i_generation, dir);
1057 mode = (mode & (S_IRWXUGO|S_ISVTX) & ~current->fs->umask) | S_IFDIR;
1058 err = ll_new_node(dir, name, NULL, mode, 0, dchild, LUSTRE_OPC_MKDIR);
1063 /* Try to find the child dentry by its name.
1064 If found, put the result fid into @fid. */
1065 static void ll_get_child_fid(struct inode * dir, struct qstr *name,
1068 struct dentry *parent, *child;
1070 parent = list_entry(dir->i_dentry.next, struct dentry, d_alias);
1071 child = d_lookup(parent, name);
1074 *fid = *ll_inode2fid(child->d_inode);
1079 static int ll_rmdir_generic(struct inode *dir, struct dentry *dparent,
1080 struct dentry *dchild, struct qstr *name)
1082 struct ptlrpc_request *request = NULL;
1083 struct md_op_data *op_data;
1087 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
1088 name->len, name->name, dir->i_ino, dir->i_generation, dir);
1090 if (unlikely(ll_d_mountpoint(dparent, dchild, name)))
1093 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len,
1094 S_IFDIR, LUSTRE_OPC_ANY, NULL);
1095 if (IS_ERR(op_data))
1096 RETURN(PTR_ERR(op_data));
1098 ll_get_child_fid(dir, name, &op_data->op_fid3);
1099 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1100 ll_finish_md_op_data(op_data);
1102 ll_update_times(request, dir);
1103 ptlrpc_req_finished(request);
1107 int ll_objects_destroy(struct ptlrpc_request *request, struct inode *dir)
1109 struct mdt_body *body;
1110 struct lov_mds_md *eadata;
1111 struct lov_stripe_md *lsm = NULL;
1112 struct obd_trans_info oti = { 0 };
1114 struct obd_capa *oc = NULL;
1118 /* req is swabbed so this is safe */
1119 body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
1120 if (!(body->valid & OBD_MD_FLEASIZE))
1123 if (body->eadatasize == 0) {
1124 CERROR("OBD_MD_FLEASIZE set but eadatasize zero\n");
1125 GOTO(out, rc = -EPROTO);
1128 /* The MDS sent back the EA because we unlinked the last reference
1129 * to this file. Use this EA to unlink the objects on the OST.
1130 * It's opaque so we don't swab here; we leave it to obd_unpackmd() to
1131 * check it is complete and sensible. */
1132 eadata = req_capsule_server_sized_get(&request->rq_pill, &RMF_MDT_MD,
1134 LASSERT(eadata != NULL);
1136 rc = obd_unpackmd(ll_i2dtexp(dir), &lsm, eadata, body->eadatasize);
1138 CERROR("obd_unpackmd: %d\n", rc);
1141 LASSERT(rc >= sizeof(*lsm));
1143 rc = obd_checkmd(ll_i2dtexp(dir), ll_i2mdexp(dir), lsm);
1145 GOTO(out_free_memmd, rc);
1149 GOTO(out_free_memmd, rc = -ENOMEM);
1151 oa->o_id = lsm->lsm_object_id;
1152 oa->o_gr = lsm->lsm_object_gr;
1153 oa->o_mode = body->mode & S_IFMT;
1154 oa->o_valid = OBD_MD_FLID | OBD_MD_FLTYPE | OBD_MD_FLGROUP;
1156 if (body->valid & OBD_MD_FLCOOKIE) {
1157 oa->o_valid |= OBD_MD_FLCOOKIE;
1158 oti.oti_logcookies =
1159 req_capsule_server_sized_get(&request->rq_pill,
1161 sizeof(struct llog_cookie) *
1162 lsm->lsm_stripe_count);
1163 if (oti.oti_logcookies == NULL) {
1164 oa->o_valid &= ~OBD_MD_FLCOOKIE;
1165 body->valid &= ~OBD_MD_FLCOOKIE;
1169 if (body->valid & OBD_MD_FLOSSCAPA) {
1170 rc = md_unpack_capa(ll_i2mdexp(dir), request, &RMF_CAPA2, &oc);
1172 GOTO(out_free_memmd, rc);
1175 rc = obd_destroy(ll_i2dtexp(dir), oa, lsm, &oti, ll_i2mdexp(dir), oc);
1179 CERROR("obd destroy objid "LPX64" error %d\n",
1180 lsm->lsm_object_id, rc);
1182 obd_free_memmd(ll_i2dtexp(dir), &lsm);
1187 static int ll_unlink_generic(struct inode *dir, struct dentry *dparent,
1188 struct dentry *dchild, struct qstr *name)
1190 struct ptlrpc_request *request = NULL;
1191 struct md_op_data *op_data;
1194 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
1195 name->len, name->name, dir->i_ino, dir->i_generation, dir);
1198 * XXX: unlink bind mountpoint maybe call to here,
1199 * just check it as vfs_unlink does.
1201 if (unlikely(ll_d_mountpoint(dparent, dchild, name)))
1204 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
1205 name->len, 0, LUSTRE_OPC_ANY, NULL);
1206 if (IS_ERR(op_data))
1207 RETURN(PTR_ERR(op_data));
1209 ll_get_child_fid(dir, name, &op_data->op_fid3);
1210 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1211 ll_finish_md_op_data(op_data);
1215 ll_update_times(request, dir);
1217 rc = ll_objects_destroy(request, dir);
1219 ptlrpc_req_finished(request);
1223 static int ll_rename_generic(struct inode *src, struct dentry *src_dparent,
1224 struct dentry *src_dchild, struct qstr *src_name,
1225 struct inode *tgt, struct dentry *tgt_dparent,
1226 struct dentry *tgt_dchild, struct qstr *tgt_name)
1228 struct ptlrpc_request *request = NULL;
1229 struct ll_sb_info *sbi = ll_i2sbi(src);
1230 struct md_op_data *op_data;
1233 CDEBUG(D_VFSTRACE,"VFS Op:oldname=%.*s,src_dir=%lu/%u(%p),newname=%.*s,"
1234 "tgt_dir=%lu/%u(%p)\n", src_name->len, src_name->name,
1235 src->i_ino, src->i_generation, src, tgt_name->len,
1236 tgt_name->name, tgt->i_ino, tgt->i_generation, tgt);
1238 if (unlikely(ll_d_mountpoint(src_dparent, src_dchild, src_name) ||
1239 ll_d_mountpoint(tgt_dparent, tgt_dchild, tgt_name)))
1242 op_data = ll_prep_md_op_data(NULL, src, tgt, NULL, 0, 0,
1243 LUSTRE_OPC_ANY, NULL);
1244 if (IS_ERR(op_data))
1245 RETURN(PTR_ERR(op_data));
1247 ll_get_child_fid(src, src_name, &op_data->op_fid3);
1248 ll_get_child_fid(tgt, tgt_name, &op_data->op_fid4);
1249 err = md_rename(sbi->ll_md_exp, op_data,
1250 src_name->name, src_name->len,
1251 tgt_name->name, tgt_name->len, &request);
1252 ll_finish_md_op_data(op_data);
1254 ll_update_times(request, src);
1255 ll_update_times(request, tgt);
1256 err = ll_objects_destroy(request, src);
1259 ptlrpc_req_finished(request);
1264 #ifdef HAVE_VFS_INTENT_PATCHES
1265 static int ll_mknod_raw(struct nameidata *nd, int mode, dev_t rdev)
1267 return ll_mknod_generic(nd->dentry->d_inode, &nd->last, mode,rdev,NULL);
1269 static int ll_rename_raw(struct nameidata *srcnd, struct nameidata *tgtnd)
1271 return ll_rename_generic(srcnd->dentry->d_inode, srcnd->dentry,
1273 tgtnd->dentry->d_inode, tgtnd->dentry,
1274 NULL, &tgtnd->last);
1276 static int ll_link_raw(struct nameidata *srcnd, struct nameidata *tgtnd)
1278 return ll_link_generic(srcnd->dentry->d_inode, tgtnd->dentry->d_inode,
1279 &tgtnd->last, NULL);
1281 static int ll_symlink_raw(struct nameidata *nd, const char *tgt)
1283 return ll_symlink_generic(nd->dentry->d_inode, &nd->last, tgt, NULL);
1285 static int ll_rmdir_raw(struct nameidata *nd)
1287 return ll_rmdir_generic(nd->dentry->d_inode, nd->dentry, NULL,
1290 static int ll_mkdir_raw(struct nameidata *nd, int mode)
1292 return ll_mkdir_generic(nd->dentry->d_inode, &nd->last, mode, NULL);
1294 static int ll_unlink_raw(struct nameidata *nd)
1296 return ll_unlink_generic(nd->dentry->d_inode, nd->dentry, NULL,
1301 static int ll_mknod(struct inode *dir, struct dentry *dchild, int mode,
1304 return ll_mknod_generic(dir, &dchild->d_name, mode,
1305 old_encode_dev(rdev), dchild);
1308 static int ll_unlink(struct inode * dir, struct dentry *dentry)
1310 return ll_unlink_generic(dir, NULL, dentry, &dentry->d_name);
1312 static int ll_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1314 return ll_mkdir_generic(dir, &dentry->d_name, mode, dentry);
1316 static int ll_rmdir(struct inode *dir, struct dentry *dentry)
1318 return ll_rmdir_generic(dir, NULL, dentry, &dentry->d_name);
1320 static int ll_symlink(struct inode *dir, struct dentry *dentry,
1321 const char *oldname)
1323 return ll_symlink_generic(dir, &dentry->d_name, oldname, dentry);
1325 static int ll_link(struct dentry *old_dentry, struct inode *dir,
1326 struct dentry *new_dentry)
1328 return ll_link_generic(old_dentry->d_inode, dir, &new_dentry->d_name,
1331 static int ll_rename(struct inode *old_dir, struct dentry *old_dentry,
1332 struct inode *new_dir, struct dentry *new_dentry)
1335 err = ll_rename_generic(old_dir, NULL,
1336 old_dentry, &old_dentry->d_name,
1337 new_dir, NULL, new_dentry,
1338 &new_dentry->d_name);
1340 #ifndef HAVE_FS_RENAME_DOES_D_MOVE
1341 if (!S_ISDIR(old_dentry->d_inode->i_mode))
1343 d_move(old_dentry, new_dentry);
1348 struct inode_operations ll_dir_inode_operations = {
1349 #ifdef HAVE_VFS_INTENT_PATCHES
1350 .link_raw = ll_link_raw,
1351 .unlink_raw = ll_unlink_raw,
1352 .symlink_raw = ll_symlink_raw,
1353 .mkdir_raw = ll_mkdir_raw,
1354 .rmdir_raw = ll_rmdir_raw,
1355 .mknod_raw = ll_mknod_raw,
1356 .rename_raw = ll_rename_raw,
1357 .setattr = ll_setattr,
1358 .setattr_raw = ll_setattr_raw,
1361 .lookup = ll_lookup_nd,
1362 .create = ll_create_nd,
1363 /* We need all these non-raw things for NFSD, to not patch it. */
1364 .unlink = ll_unlink,
1367 .symlink = ll_symlink,
1369 .rename = ll_rename,
1370 .setattr = ll_setattr,
1371 .getattr = ll_getattr,
1372 .permission = ll_inode_permission,
1373 .setxattr = ll_setxattr,
1374 .getxattr = ll_getxattr,
1375 .listxattr = ll_listxattr,
1376 .removexattr = ll_removexattr,
1379 struct inode_operations ll_special_inode_operations = {
1380 #ifdef HAVE_VFS_INTENT_PATCHES
1381 .setattr_raw = ll_setattr_raw,
1383 .setattr = ll_setattr,
1384 .getattr = ll_getattr,
1385 .permission = ll_inode_permission,
1386 .setxattr = ll_setxattr,
1387 .getxattr = ll_getxattr,
1388 .listxattr = ll_listxattr,
1389 .removexattr = ll_removexattr,