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)
112 * Get an inode by inode number (already instantiated by the intent lookup).
113 * Returns inode or NULL
115 struct inode *ll_iget(struct super_block *sb, ino_t hash,
116 struct lustre_md *md)
118 struct ll_inode_info *lli;
123 inode = iget5_locked(sb, hash, ll_test_inode, ll_set_inode, md);
126 lli = ll_i2info(inode);
127 if (inode->i_state & I_NEW) {
130 ll_read_inode2(inode, md);
131 rc = cl_inode_init(inode, md);
134 make_bad_inode(inode);
135 unlock_new_inode(inode);
139 unlock_new_inode(inode);
140 } else if (!(inode->i_state & (I_FREEING | I_CLEAR)))
141 ll_update_inode(inode, md);
142 CDEBUG(D_VFSTRACE, "got inode: %p for "DFID"\n",
143 inode, PFID(&md->body->fid1));
148 static void ll_drop_negative_dentry(struct inode *dir)
150 struct dentry *dentry, *tmp_alias, *tmp_subdir;
152 spin_lock(&ll_lookup_lock);
153 spin_lock(&dcache_lock);
155 list_for_each_entry_safe(dentry, tmp_alias,
156 &dir->i_dentry,d_alias) {
157 if (!list_empty(&dentry->d_subdirs)) {
158 struct dentry *child;
159 list_for_each_entry_safe(child, tmp_subdir,
162 /* XXX Print some debug here? */
164 /* Negative dentry. If we were
165 dropping dcache lock, go
166 throught the list again */
167 if (ll_drop_dentry(child))
172 spin_unlock(&dcache_lock);
173 spin_unlock(&ll_lookup_lock);
177 int ll_md_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
178 void *data, int flag)
181 struct lustre_handle lockh;
185 case LDLM_CB_BLOCKING:
186 ldlm_lock2handle(lock, &lockh);
187 rc = ldlm_cli_cancel(&lockh);
189 CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
193 case LDLM_CB_CANCELING: {
194 struct inode *inode = ll_inode_from_lock(lock);
195 __u64 bits = lock->l_policy_data.l_inodebits.bits;
198 /* Invalidate all dentries associated with this inode */
202 LASSERT(lock->l_flags & LDLM_FL_CANCELING);
203 if ((bits & MDS_INODELOCK_LOOKUP) &&
204 ll_have_md_lock(inode, MDS_INODELOCK_LOOKUP))
205 bits &= ~MDS_INODELOCK_LOOKUP;
206 if ((bits & MDS_INODELOCK_UPDATE) &&
207 ll_have_md_lock(inode, MDS_INODELOCK_UPDATE))
208 bits &= ~MDS_INODELOCK_UPDATE;
209 if ((bits & MDS_INODELOCK_OPEN) &&
210 ll_have_md_lock(inode, MDS_INODELOCK_OPEN))
211 bits &= ~MDS_INODELOCK_OPEN;
213 fid = ll_inode2fid(inode);
214 if (lock->l_resource->lr_name.name[0] != fid_seq(fid) ||
215 lock->l_resource->lr_name.name[1] != fid_oid(fid) ||
216 lock->l_resource->lr_name.name[2] != fid_ver(fid)) {
217 LDLM_ERROR(lock, "data mismatch with object "
218 DFID" (%p)", PFID(fid), inode);
221 if (bits & MDS_INODELOCK_OPEN) {
223 switch (lock->l_req_mode) {
234 CERROR("Unexpected lock mode for OPEN lock "
235 "%d, inode %ld\n", lock->l_req_mode,
238 ll_md_real_close(inode, flags);
241 if (bits & MDS_INODELOCK_UPDATE)
242 ll_i2info(inode)->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
244 if (S_ISDIR(inode->i_mode) &&
245 (bits & MDS_INODELOCK_UPDATE)) {
246 CDEBUG(D_INODE, "invalidating inode %lu\n",
248 truncate_inode_pages(inode->i_mapping, 0);
249 ll_drop_negative_dentry(inode);
252 if (inode->i_sb->s_root &&
253 inode != inode->i_sb->s_root->d_inode &&
254 (bits & MDS_INODELOCK_LOOKUP))
255 ll_unhash_aliases(inode);
266 __u32 ll_i2suppgid(struct inode *i)
268 if (in_group_p(i->i_gid))
269 return (__u32)i->i_gid;
274 /* Pack the required supplementary groups into the supplied groups array.
275 * If we don't need to use the groups from the target inode(s) then we
276 * instead pack one or more groups from the user's supplementary group
277 * array in case it might be useful. Not needed if doing an MDS-side upcall. */
278 void ll_i2gids(__u32 *suppgids, struct inode *i1, struct inode *i2)
285 LASSERT(suppgids != NULL);
287 suppgids[0] = ll_i2suppgid(i1);
290 suppgids[1] = ll_i2suppgid(i2);
295 for (i = 0; i < current_ngroups; i++) {
296 if (suppgids[0] == -1) {
297 if (current_groups[i] != suppgids[1])
298 suppgids[0] = current_groups[i];
301 if (suppgids[1] == -1) {
302 if (current_groups[i] != suppgids[0])
303 suppgids[1] = current_groups[i];
311 static void ll_d_add(struct dentry *de, struct inode *inode)
313 CDEBUG(D_DENTRY, "adding inode %p to dentry %p\n", inode, de);
315 if (!list_empty(&de->d_alias)) {
316 spin_unlock(&dcache_lock);
317 CERROR("dentry %.*s %p alias next %p, prev %p\n",
318 de->d_name.len, de->d_name.name, de,
319 de->d_alias.next, de->d_alias.prev);
323 list_add(&de->d_alias, &inode->i_dentry);
327 if (!d_unhashed(de)) {
328 spin_unlock(&dcache_lock);
329 CERROR("dentry %.*s %p hash next %p\n",
330 de->d_name.len, de->d_name.name, de, de->d_hash.next);
333 d_rehash_cond(de, 0);
336 /* Search "inode"'s alias list for a dentry that has the same name and parent
337 * as de. If found, return it. If not found, return de.
338 * Lustre can't use d_add_unique because don't unhash aliases for directory
339 * in ll_revalidate_it. After revaliadate inode will be have hashed aliases
340 * and it triggers BUG_ON in d_instantiate_unique (bug #10954).
342 static struct dentry *ll_find_alias(struct inode *inode, struct dentry *de)
344 struct list_head *tmp;
345 struct dentry *dentry;
346 struct dentry *last_discon = NULL;
348 spin_lock(&ll_lookup_lock);
349 spin_lock(&dcache_lock);
350 list_for_each(tmp, &inode->i_dentry) {
351 dentry = list_entry(tmp, struct dentry, d_alias);
353 /* We are called here with 'de' already on the aliases list. */
354 if (unlikely(dentry == de)) {
359 if (dentry->d_flags & DCACHE_DISCONNECTED) {
360 LASSERT(last_discon == NULL);
361 last_discon = dentry;
365 if (dentry->d_parent != de->d_parent)
368 if (dentry->d_name.hash != de->d_name.hash)
371 if (dentry->d_name.len != de->d_name.len)
374 if (memcmp(dentry->d_name.name, de->d_name.name,
375 de->d_name.len) != 0)
381 #ifdef DCACHE_LUSTRE_INVALID
382 dentry->d_flags &= ~DCACHE_LUSTRE_INVALID;
384 unlock_dentry(dentry);
385 ll_dops_init(dentry, 0);
386 d_rehash_cond(dentry, 0); /* avoid taking dcache_lock inside */
387 spin_unlock(&dcache_lock);
388 spin_unlock(&ll_lookup_lock);
390 CDEBUG(D_DENTRY, "alias dentry %.*s (%p) parent %p inode %p "
391 "refc %d\n", de->d_name.len, de->d_name.name, de,
392 de->d_parent, de->d_inode, atomic_read(&de->d_count));
397 CDEBUG(D_DENTRY, "Reuse disconnected dentry %p inode %p "
398 "refc %d\n", last_discon, last_discon->d_inode,
399 atomic_read(&last_discon->d_count));
400 dget_locked(last_discon);
401 spin_unlock(&dcache_lock);
402 spin_unlock(&ll_lookup_lock);
403 ll_dops_init(last_discon, 1);
405 d_move(last_discon, de);
412 spin_unlock(&dcache_lock);
413 spin_unlock(&ll_lookup_lock);
418 int ll_lookup_it_finish(struct ptlrpc_request *request,
419 struct lookup_intent *it, void *data)
421 struct it_cb_data *icbd = data;
422 struct dentry **de = icbd->icbd_childp;
423 struct inode *parent = icbd->icbd_parent;
424 struct ll_sb_info *sbi = ll_i2sbi(parent);
425 struct inode *inode = NULL;
429 /* NB 1 request reference will be taken away by ll_intent_lock()
431 if (!it_disposition(it, DISP_LOOKUP_NEG)) {
432 struct dentry *save = *de;
434 rc = ll_prep_inode(&inode, request, (*de)->d_sb);
438 CDEBUG(D_DLMTRACE, "setting l_data to inode %p (%lu/%u)\n",
439 inode, inode->i_ino, inode->i_generation);
440 md_set_lock_data(sbi->ll_md_exp,
441 &it->d.lustre.it_lock_handle, inode);
443 /* We used to query real size from OSTs here, but actually
444 this is not needed. For stat() calls size would be updated
445 from subsequent do_revalidate()->ll_inode_revalidate_it() in
447 vfs_getattr_it->ll_getattr()->ll_inode_revalidate_it() in 2.6
448 Everybody else who needs correct file size would call
449 cl_glimpse_size or some equivalent themselves anyway.
450 Also see bug 7198. */
452 ll_dops_init(*de, 1);
453 *de = ll_find_alias(inode, *de);
455 struct ll_dentry_data *lld = ll_d2d(*de);
457 /* just make sure the ll_dentry_data is ready */
458 if (unlikely(lld == NULL)) {
461 if (likely(lld != NULL))
462 lld->lld_sa_generation = 0;
466 ll_dops_init(*de, 1);
467 /* Check that parent has UPDATE lock. If there is none, we
468 cannot afford to hash this dentry (done by ll_d_add) as it
469 might get picked up later when UPDATE lock will appear */
470 if (ll_have_md_lock(parent, MDS_INODELOCK_UPDATE)) {
471 spin_lock(&dcache_lock);
473 spin_unlock(&dcache_lock);
475 (*de)->d_inode = NULL;
476 /* We do not want to hash the dentry if don`t have a
477 * lock, but if this dentry is later used in d_move,
478 * we'd hit uninitialised list head d_hash, so we just
479 * do this to init d_hash field but leave dentry
480 * unhashed. (bug 10796). */
489 static struct dentry *ll_lookup_it(struct inode *parent, struct dentry *dentry,
490 struct lookup_intent *it, int lookup_flags)
492 struct lookup_intent lookup_it = { .it_op = IT_LOOKUP };
493 struct dentry *save = dentry, *retval;
494 struct ptlrpc_request *req = NULL;
495 struct md_op_data *op_data;
496 struct it_cb_data icbd;
501 if (dentry->d_name.len > ll_i2sbi(parent)->ll_namelen)
502 RETURN(ERR_PTR(-ENAMETOOLONG));
504 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),intent=%s\n",
505 dentry->d_name.len, dentry->d_name.name, parent->i_ino,
506 parent->i_generation, parent, LL_IT2STR(it));
508 if (d_mountpoint(dentry))
509 CERROR("Tell Peter, lookup on mtpt, it %s\n", LL_IT2STR(it));
511 ll_frob_intent(&it, &lookup_it);
513 /* As do_lookup is called before follow_mount, root dentry may be left
514 * not valid, revalidate it here. */
515 if (parent->i_sb->s_root && (parent->i_sb->s_root->d_inode == parent) &&
516 (it->it_op & (IT_OPEN | IT_CREAT))) {
517 rc = ll_inode_revalidate_it(parent->i_sb->s_root, it);
522 if (it->it_op == IT_GETATTR) {
523 first = ll_statahead_enter(parent, &dentry, 1);
525 ll_statahead_exit(dentry, first);
527 RETURN(retval = dentry);
531 icbd.icbd_childp = &dentry;
532 icbd.icbd_parent = parent;
534 if (it->it_op & IT_CREAT ||
535 (it->it_op & IT_OPEN && it->it_create_mode & O_CREAT))
536 opc = LUSTRE_OPC_CREATE;
538 opc = LUSTRE_OPC_ANY;
540 op_data = ll_prep_md_op_data(NULL, parent, NULL, dentry->d_name.name,
541 dentry->d_name.len, lookup_flags, opc,
544 RETURN((void *)op_data);
546 it->it_create_mode &= ~current->fs->umask;
548 rc = md_intent_lock(ll_i2mdexp(parent), op_data, NULL, 0, it,
549 lookup_flags, &req, ll_md_blocking_ast, 0);
550 ll_finish_md_op_data(op_data);
552 GOTO(out, retval = ERR_PTR(rc));
554 rc = ll_lookup_it_finish(req, it, &icbd);
556 ll_intent_release(it);
557 GOTO(out, retval = ERR_PTR(rc));
560 if (first == -EEXIST)
561 ll_statahead_mark(dentry);
563 if ((it->it_op & IT_OPEN) && dentry->d_inode &&
564 !S_ISREG(dentry->d_inode->i_mode) &&
565 !S_ISDIR(dentry->d_inode->i_mode)) {
566 ll_release_openhandle(dentry, it);
568 ll_lookup_finish_locks(it, dentry);
571 GOTO(out, retval = NULL);
573 GOTO(out, retval = dentry);
576 ptlrpc_req_finished(req);
580 #ifdef HAVE_VFS_INTENT_PATCHES
581 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
582 struct nameidata *nd)
587 if (nd && nd->flags & LOOKUP_LAST && !(nd->flags & LOOKUP_LINK_NOTLAST))
588 de = ll_lookup_it(parent, dentry, &nd->intent, nd->flags);
590 de = ll_lookup_it(parent, dentry, NULL, 0);
595 struct lookup_intent *ll_convert_intent(struct open_intent *oit,
598 struct lookup_intent *it;
600 OBD_ALLOC(it, sizeof(*it));
602 return ERR_PTR(-ENOMEM);
604 if (lookup_flags & LOOKUP_OPEN) {
606 if (lookup_flags & LOOKUP_CREATE)
607 it->it_op |= IT_CREAT;
608 it->it_create_mode = oit->create_mode;
609 it->it_flags = oit->flags;
611 it->it_op = IT_GETATTR;
614 #ifndef HAVE_FILE_IN_STRUCT_INTENT
615 /* Since there is no way to pass our intent to ll_file_open,
616 * just check the file is there. Actual open will be done
618 if (it->it_op & IT_OPEN)
619 it->it_op = IT_LOOKUP;
625 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
626 struct nameidata *nd)
631 if (nd && !(nd->flags & (LOOKUP_CONTINUE|LOOKUP_PARENT))) {
632 struct lookup_intent *it;
634 #if defined(HAVE_FILE_IN_STRUCT_INTENT) && (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,17))
635 /* Did we came here from failed revalidate just to propagate
637 if (nd->flags & LOOKUP_OPEN)
638 if (IS_ERR(nd->intent.open.file))
639 RETURN((struct dentry *)nd->intent.open.file);
642 if (ll_d2d(dentry) && ll_d2d(dentry)->lld_it) {
643 it = ll_d2d(dentry)->lld_it;
644 ll_d2d(dentry)->lld_it = NULL;
646 it = ll_convert_intent(&nd->intent.open, nd->flags);
648 RETURN((struct dentry *)it);
651 de = ll_lookup_it(parent, dentry, it, nd->flags);
654 if ((nd->flags & LOOKUP_OPEN) && !IS_ERR(dentry)) { /* Open */
655 if (dentry->d_inode &&
656 it_disposition(it, DISP_OPEN_OPEN)) { /* nocreate */
657 #ifdef HAVE_FILE_IN_STRUCT_INTENT
658 if (S_ISFIFO(dentry->d_inode->i_mode)) {
659 // We cannot call open here as it would
662 (struct ptlrpc_request *)
663 it->d.lustre.it_data);
665 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,17))
666 /* 2.6.1[456] have a bug in open_namei() that forgets to check
667 * nd->intent.open.file for error, so we need to return it as lookup's result
670 nd->intent.open.file->private_data = it;
671 filp =lookup_instantiate_filp(nd,dentry,
676 de = (struct dentry *) filp;
679 nd->intent.open.file->private_data = it;
680 (void)lookup_instantiate_filp(nd,dentry,
685 #else /* HAVE_FILE_IN_STRUCT_INTENT */
686 /* Release open handle as we have no way to
687 * pass it to ll_file_open */
688 ll_release_openhandle(dentry, it);
689 #endif /* HAVE_FILE_IN_STRUCT_INTENT */
690 } else if (it_disposition(it, DISP_OPEN_CREATE)) {
691 // XXX This can only reliably work on assumption
692 // that there are NO hashed negative dentries.
693 ll_d2d(dentry)->lld_it = it;
694 it = NULL; /* Will be freed in ll_create_nd */
695 /* We absolutely depend on ll_create_nd to be
696 * called to not leak this intent and possible
697 * data attached to it */
702 ll_intent_release(it);
703 OBD_FREE(it, sizeof(*it));
706 de = ll_lookup_it(parent, dentry, NULL, 0);
714 * check new allocated inode, new inode shld not have any valid alias
716 static void ll_validate_new_inode(struct inode *new)
718 struct list_head *lp;
719 struct dentry * dentry;
723 if (list_empty(&new->i_dentry))
726 spin_lock(&dcache_lock);
727 list_for_each(lp, &new->i_dentry) {
728 dentry = list_entry(lp, struct dentry, d_alias);
729 if (!d_unhashed(dentry) && !(dentry->d_flags & DCACHE_LUSTRE_INVALID)){
736 spin_unlock(&dcache_lock);
738 if (need_inval && !in_recheck) {
739 /* kill all old inode's data pages */
740 truncate_inode_pages(new->i_mapping, 0);
742 /* invalidate all dirent and recheck inode */
743 ll_unhash_aliases(new);
749 /* We depend on "mode" being set with the proper file type/umask by now */
750 static struct inode *ll_create_node(struct inode *dir, const char *name,
751 int namelen, const void *data, int datalen,
752 int mode, __u64 extra,
753 struct lookup_intent *it)
755 struct inode *inode = NULL;
756 struct ptlrpc_request *request = NULL;
757 struct ll_sb_info *sbi = ll_i2sbi(dir);
761 LASSERT(it && it->d.lustre.it_disposition);
763 LASSERT(it_disposition(it, DISP_ENQ_CREATE_REF));
764 request = it->d.lustre.it_data;
765 it_clear_disposition(it, DISP_ENQ_CREATE_REF);
766 rc = ll_prep_inode(&inode, request, dir->i_sb);
768 GOTO(out, inode = ERR_PTR(rc));
770 ll_validate_new_inode(inode);
772 /* We asked for a lock on the directory, but were granted a
773 * lock on the inode. Since we finally have an inode pointer,
774 * stuff it in the lock. */
775 CDEBUG(D_DLMTRACE, "setting l_ast_data to inode %p (%lu/%u)\n",
776 inode, inode->i_ino, inode->i_generation);
777 md_set_lock_data(sbi->ll_md_exp,
778 &it->d.lustre.it_lock_handle, inode);
781 ptlrpc_req_finished(request);
786 * By the time this is called, we already have created the directory cache
787 * entry for the new file, but it is so far negative - it has no inode.
789 * We defer creating the OBD object(s) until open, to keep the intent and
790 * non-intent code paths similar, and also because we do not have the MDS
791 * inode number before calling ll_create_node() (which is needed for LOV),
792 * so we would need to do yet another RPC to the MDS to store the LOV EA
793 * data on the MDS. If needed, we would pass the PACKED lmm as data and
794 * lmm_size in datalen (the MDS still has code which will handle that).
796 * If the create succeeds, we fill in the inode information
797 * with d_instantiate().
799 static int ll_create_it(struct inode *dir, struct dentry *dentry, int mode,
800 struct lookup_intent *it)
806 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),intent=%s\n",
807 dentry->d_name.len, dentry->d_name.name, dir->i_ino,
808 dir->i_generation, dir, LL_IT2STR(it));
810 rc = it_open_error(DISP_OPEN_CREATE, it);
814 inode = ll_create_node(dir, dentry->d_name.name, dentry->d_name.len,
815 NULL, 0, mode, 0, it);
817 RETURN(PTR_ERR(inode));
820 /* it might been set during parent dir revalidation */
821 dentry->d_flags &= ~DCACHE_LUSTRE_INVALID;
822 d_instantiate(dentry, inode);
823 /* Negative dentry may be unhashed if parent does not have UPDATE lock,
824 * but some callers, e.g. do_coredump, expect dentry to be hashed after
825 * successful create. Hash it here. */
826 spin_lock(&dcache_lock);
827 if (d_unhashed(dentry))
828 d_rehash_cond(dentry, 0);
829 spin_unlock(&dcache_lock);
833 static void ll_update_times(struct ptlrpc_request *request,
836 struct mdt_body *body = req_capsule_server_get(&request->rq_pill,
840 /* mtime is always updated with ctime, but can be set in past.
841 As write and utime(2) may happen within 1 second, and utime's
842 mtime has a priority over write's one, so take mtime from mds
843 for the same ctimes. */
844 if (body->valid & OBD_MD_FLCTIME &&
845 body->ctime >= LTIME_S(inode->i_ctime)) {
846 LTIME_S(inode->i_ctime) = body->ctime;
848 if (body->valid & OBD_MD_FLMTIME) {
849 CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
850 "to "LPU64"\n", inode->i_ino,
851 LTIME_S(inode->i_mtime), body->mtime);
852 LTIME_S(inode->i_mtime) = body->mtime;
857 static int ll_new_node(struct inode *dir, struct qstr *name,
858 const char *tgt, int mode, int rdev,
859 struct dentry *dchild, __u32 opc)
861 struct ptlrpc_request *request = NULL;
862 struct md_op_data *op_data;
863 struct inode *inode = NULL;
864 struct ll_sb_info *sbi = ll_i2sbi(dir);
869 if (unlikely(tgt != NULL))
870 tgt_len = strlen(tgt) + 1;
872 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
873 name->len, 0, opc, NULL);
875 GOTO(err_exit, err = PTR_ERR(op_data));
877 err = md_create(sbi->ll_md_exp, op_data, tgt, tgt_len, mode,
878 current->fsuid, current->fsgid,
879 cfs_curproc_cap_pack(), rdev, &request);
880 ll_finish_md_op_data(op_data);
884 ll_update_times(request, dir);
887 err = ll_prep_inode(&inode, request, dchild->d_sb);
892 d_instantiate(dchild, inode);
896 ptlrpc_req_finished(request);
901 static int ll_mknod_generic(struct inode *dir, struct qstr *name, int mode,
902 unsigned rdev, struct dentry *dchild)
907 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p) mode %o dev %x\n",
908 name->len, name->name, dir->i_ino, dir->i_generation, dir,
911 mode &= ~current->fs->umask;
913 switch (mode & S_IFMT) {
915 mode |= S_IFREG; /* for mode = 0 case, fallthrough */
921 err = ll_new_node(dir, name, NULL, mode, rdev, dchild,
933 #ifndef HAVE_VFS_INTENT_PATCHES
934 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
935 int mode, struct nameidata *nd)
937 struct lookup_intent *it = ll_d2d(dentry)->lld_it;
941 return ll_mknod_generic(dir, &dentry->d_name, mode, 0, dentry);
943 ll_d2d(dentry)->lld_it = NULL;
945 /* Was there an error? Propagate it! */
946 if (it->d.lustre.it_status) {
947 rc = it->d.lustre.it_status;
951 rc = ll_create_it(dir, dentry, mode, it);
952 #ifdef HAVE_FILE_IN_STRUCT_INTENT
953 if (nd && (nd->flags & LOOKUP_OPEN) && dentry->d_inode) { /* Open */
954 nd->intent.open.file->private_data = it;
955 lookup_instantiate_filp(nd, dentry, NULL);
958 ll_release_openhandle(dentry,it);
962 ll_intent_release(it);
963 OBD_FREE(it, sizeof(*it));
968 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
969 int mode, struct nameidata *nd)
971 if (!nd || !nd->intent.d.lustre.it_disposition)
972 /* No saved request? Just mknod the file */
973 return ll_mknod_generic(dir, &dentry->d_name, mode, 0, dentry);
975 return ll_create_it(dir, dentry, mode, &nd->intent);
979 static int ll_symlink_generic(struct inode *dir, struct qstr *name,
980 const char *tgt, struct dentry *dchild)
985 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),target=%.*s\n",
986 name->len, name->name, dir->i_ino, dir->i_generation,
989 err = ll_new_node(dir, name, (char *)tgt, S_IFLNK | S_IRWXUGO,
990 0, dchild, LUSTRE_OPC_SYMLINK);
994 static int ll_link_generic(struct inode *src, struct inode *dir,
995 struct qstr *name, struct dentry *dchild)
997 struct ll_sb_info *sbi = ll_i2sbi(dir);
998 struct ptlrpc_request *request = NULL;
999 struct md_op_data *op_data;
1004 "VFS Op: inode=%lu/%u(%p), dir=%lu/%u(%p), target=%.*s\n",
1005 src->i_ino, src->i_generation, src, dir->i_ino,
1006 dir->i_generation, dir, name->len, name->name);
1008 op_data = ll_prep_md_op_data(NULL, src, dir, name->name, name->len,
1009 0, LUSTRE_OPC_ANY, NULL);
1010 if (IS_ERR(op_data))
1011 RETURN(PTR_ERR(op_data));
1013 err = md_link(sbi->ll_md_exp, op_data, &request);
1014 ll_finish_md_op_data(op_data);
1020 ll_update_times(request, dir);
1023 ptlrpc_req_finished(request);
1027 static int ll_mkdir_generic(struct inode *dir, struct qstr *name,
1028 int mode, struct dentry *dchild)
1034 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
1035 name->len, name->name, dir->i_ino, dir->i_generation, dir);
1037 mode = (mode & (S_IRWXUGO|S_ISVTX) & ~current->fs->umask) | S_IFDIR;
1038 err = ll_new_node(dir, name, NULL, mode, 0, dchild, LUSTRE_OPC_MKDIR);
1043 /* Try to find the child dentry by its name.
1044 If found, put the result fid into @fid. */
1045 static void ll_get_child_fid(struct inode * dir, struct qstr *name,
1048 struct dentry *parent, *child;
1050 parent = list_entry(dir->i_dentry.next, struct dentry, d_alias);
1051 child = d_lookup(parent, name);
1054 *fid = *ll_inode2fid(child->d_inode);
1059 static int ll_rmdir_generic(struct inode *dir, struct dentry *dparent,
1060 struct dentry *dchild, struct qstr *name)
1062 struct ptlrpc_request *request = NULL;
1063 struct md_op_data *op_data;
1067 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
1068 name->len, name->name, dir->i_ino, dir->i_generation, dir);
1070 if (unlikely(ll_d_mountpoint(dparent, dchild, name)))
1073 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len,
1074 S_IFDIR, LUSTRE_OPC_ANY, NULL);
1075 if (IS_ERR(op_data))
1076 RETURN(PTR_ERR(op_data));
1078 ll_get_child_fid(dir, name, &op_data->op_fid3);
1079 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1080 ll_finish_md_op_data(op_data);
1082 ll_update_times(request, dir);
1083 ptlrpc_req_finished(request);
1087 int ll_objects_destroy(struct ptlrpc_request *request, struct inode *dir)
1089 struct mdt_body *body;
1090 struct lov_mds_md *eadata;
1091 struct lov_stripe_md *lsm = NULL;
1092 struct obd_trans_info oti = { 0 };
1094 struct obd_capa *oc = NULL;
1098 /* req is swabbed so this is safe */
1099 body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
1100 if (!(body->valid & OBD_MD_FLEASIZE))
1103 if (body->eadatasize == 0) {
1104 CERROR("OBD_MD_FLEASIZE set but eadatasize zero\n");
1105 GOTO(out, rc = -EPROTO);
1108 /* The MDS sent back the EA because we unlinked the last reference
1109 * to this file. Use this EA to unlink the objects on the OST.
1110 * It's opaque so we don't swab here; we leave it to obd_unpackmd() to
1111 * check it is complete and sensible. */
1112 eadata = req_capsule_server_sized_get(&request->rq_pill, &RMF_MDT_MD,
1114 LASSERT(eadata != NULL);
1116 rc = obd_unpackmd(ll_i2dtexp(dir), &lsm, eadata, body->eadatasize);
1118 CERROR("obd_unpackmd: %d\n", rc);
1121 LASSERT(rc >= sizeof(*lsm));
1123 rc = obd_checkmd(ll_i2dtexp(dir), ll_i2mdexp(dir), lsm);
1125 GOTO(out_free_memmd, rc);
1129 GOTO(out_free_memmd, rc = -ENOMEM);
1131 oa->o_id = lsm->lsm_object_id;
1132 oa->o_gr = lsm->lsm_object_gr;
1133 oa->o_mode = body->mode & S_IFMT;
1134 oa->o_valid = OBD_MD_FLID | OBD_MD_FLTYPE | OBD_MD_FLGROUP;
1136 if (body->valid & OBD_MD_FLCOOKIE) {
1137 oa->o_valid |= OBD_MD_FLCOOKIE;
1138 oti.oti_logcookies =
1139 req_capsule_server_sized_get(&request->rq_pill,
1141 sizeof(struct llog_cookie) *
1142 lsm->lsm_stripe_count);
1143 if (oti.oti_logcookies == NULL) {
1144 oa->o_valid &= ~OBD_MD_FLCOOKIE;
1145 body->valid &= ~OBD_MD_FLCOOKIE;
1149 if (body->valid & OBD_MD_FLOSSCAPA) {
1150 rc = md_unpack_capa(ll_i2mdexp(dir), request, &RMF_CAPA2, &oc);
1152 GOTO(out_free_memmd, rc);
1155 rc = obd_destroy(ll_i2dtexp(dir), oa, lsm, &oti, ll_i2mdexp(dir), oc);
1159 CERROR("obd destroy objid "LPX64" error %d\n",
1160 lsm->lsm_object_id, rc);
1162 obd_free_memmd(ll_i2dtexp(dir), &lsm);
1167 static int ll_unlink_generic(struct inode *dir, struct dentry *dparent,
1168 struct dentry *dchild, struct qstr *name)
1170 struct ptlrpc_request *request = NULL;
1171 struct md_op_data *op_data;
1174 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
1175 name->len, name->name, dir->i_ino, dir->i_generation, dir);
1178 * XXX: unlink bind mountpoint maybe call to here,
1179 * just check it as vfs_unlink does.
1181 if (unlikely(ll_d_mountpoint(dparent, dchild, name)))
1184 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
1185 name->len, 0, LUSTRE_OPC_ANY, NULL);
1186 if (IS_ERR(op_data))
1187 RETURN(PTR_ERR(op_data));
1189 ll_get_child_fid(dir, name, &op_data->op_fid3);
1190 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1191 ll_finish_md_op_data(op_data);
1195 ll_update_times(request, dir);
1197 rc = ll_objects_destroy(request, dir);
1199 ptlrpc_req_finished(request);
1203 static int ll_rename_generic(struct inode *src, struct dentry *src_dparent,
1204 struct dentry *src_dchild, struct qstr *src_name,
1205 struct inode *tgt, struct dentry *tgt_dparent,
1206 struct dentry *tgt_dchild, struct qstr *tgt_name)
1208 struct ptlrpc_request *request = NULL;
1209 struct ll_sb_info *sbi = ll_i2sbi(src);
1210 struct md_op_data *op_data;
1213 CDEBUG(D_VFSTRACE,"VFS Op:oldname=%.*s,src_dir=%lu/%u(%p),newname=%.*s,"
1214 "tgt_dir=%lu/%u(%p)\n", src_name->len, src_name->name,
1215 src->i_ino, src->i_generation, src, tgt_name->len,
1216 tgt_name->name, tgt->i_ino, tgt->i_generation, tgt);
1218 if (unlikely(ll_d_mountpoint(src_dparent, src_dchild, src_name) ||
1219 ll_d_mountpoint(tgt_dparent, tgt_dchild, tgt_name)))
1222 op_data = ll_prep_md_op_data(NULL, src, tgt, NULL, 0, 0,
1223 LUSTRE_OPC_ANY, NULL);
1224 if (IS_ERR(op_data))
1225 RETURN(PTR_ERR(op_data));
1227 ll_get_child_fid(src, src_name, &op_data->op_fid3);
1228 ll_get_child_fid(tgt, tgt_name, &op_data->op_fid4);
1229 err = md_rename(sbi->ll_md_exp, op_data,
1230 src_name->name, src_name->len,
1231 tgt_name->name, tgt_name->len, &request);
1232 ll_finish_md_op_data(op_data);
1234 ll_update_times(request, src);
1235 ll_update_times(request, tgt);
1236 err = ll_objects_destroy(request, src);
1239 ptlrpc_req_finished(request);
1244 #ifdef HAVE_VFS_INTENT_PATCHES
1245 static int ll_mknod_raw(struct nameidata *nd, int mode, dev_t rdev)
1247 return ll_mknod_generic(nd->dentry->d_inode, &nd->last, mode,rdev,NULL);
1249 static int ll_rename_raw(struct nameidata *srcnd, struct nameidata *tgtnd)
1251 return ll_rename_generic(srcnd->dentry->d_inode, srcnd->dentry,
1253 tgtnd->dentry->d_inode, tgtnd->dentry,
1254 NULL, &tgtnd->last);
1256 static int ll_link_raw(struct nameidata *srcnd, struct nameidata *tgtnd)
1258 return ll_link_generic(srcnd->dentry->d_inode, tgtnd->dentry->d_inode,
1259 &tgtnd->last, NULL);
1261 static int ll_symlink_raw(struct nameidata *nd, const char *tgt)
1263 return ll_symlink_generic(nd->dentry->d_inode, &nd->last, tgt, NULL);
1265 static int ll_rmdir_raw(struct nameidata *nd)
1267 return ll_rmdir_generic(nd->dentry->d_inode, nd->dentry, NULL,
1270 static int ll_mkdir_raw(struct nameidata *nd, int mode)
1272 return ll_mkdir_generic(nd->dentry->d_inode, &nd->last, mode, NULL);
1274 static int ll_unlink_raw(struct nameidata *nd)
1276 return ll_unlink_generic(nd->dentry->d_inode, nd->dentry, NULL,
1281 static int ll_mknod(struct inode *dir, struct dentry *dchild, int mode,
1284 return ll_mknod_generic(dir, &dchild->d_name, mode,
1285 old_encode_dev(rdev), dchild);
1288 static int ll_unlink(struct inode * dir, struct dentry *dentry)
1290 return ll_unlink_generic(dir, NULL, dentry, &dentry->d_name);
1292 static int ll_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1294 return ll_mkdir_generic(dir, &dentry->d_name, mode, dentry);
1296 static int ll_rmdir(struct inode *dir, struct dentry *dentry)
1298 return ll_rmdir_generic(dir, NULL, dentry, &dentry->d_name);
1300 static int ll_symlink(struct inode *dir, struct dentry *dentry,
1301 const char *oldname)
1303 return ll_symlink_generic(dir, &dentry->d_name, oldname, dentry);
1305 static int ll_link(struct dentry *old_dentry, struct inode *dir,
1306 struct dentry *new_dentry)
1308 return ll_link_generic(old_dentry->d_inode, dir, &new_dentry->d_name,
1311 static int ll_rename(struct inode *old_dir, struct dentry *old_dentry,
1312 struct inode *new_dir, struct dentry *new_dentry)
1315 err = ll_rename_generic(old_dir, NULL,
1316 old_dentry, &old_dentry->d_name,
1317 new_dir, NULL, new_dentry,
1318 &new_dentry->d_name);
1320 #ifndef HAVE_FS_RENAME_DOES_D_MOVE
1321 if (!S_ISDIR(old_dentry->d_inode->i_mode))
1323 d_move(old_dentry, new_dentry);
1328 struct inode_operations ll_dir_inode_operations = {
1329 #ifdef HAVE_VFS_INTENT_PATCHES
1330 .link_raw = ll_link_raw,
1331 .unlink_raw = ll_unlink_raw,
1332 .symlink_raw = ll_symlink_raw,
1333 .mkdir_raw = ll_mkdir_raw,
1334 .rmdir_raw = ll_rmdir_raw,
1335 .mknod_raw = ll_mknod_raw,
1336 .rename_raw = ll_rename_raw,
1337 .setattr = ll_setattr,
1338 .setattr_raw = ll_setattr_raw,
1341 .lookup = ll_lookup_nd,
1342 .create = ll_create_nd,
1343 /* We need all these non-raw things for NFSD, to not patch it. */
1344 .unlink = ll_unlink,
1347 .symlink = ll_symlink,
1349 .rename = ll_rename,
1350 .setattr = ll_setattr,
1351 .getattr = ll_getattr,
1352 .permission = ll_inode_permission,
1353 .setxattr = ll_setxattr,
1354 .getxattr = ll_getxattr,
1355 .listxattr = ll_listxattr,
1356 .removexattr = ll_removexattr,
1359 struct inode_operations ll_special_inode_operations = {
1360 #ifdef HAVE_VFS_INTENT_PATCHES
1361 .setattr_raw = ll_setattr_raw,
1363 .setattr = ll_setattr,
1364 .getattr = ll_getattr,
1365 .permission = ll_inode_permission,
1366 .setxattr = ll_setxattr,
1367 .getxattr = ll_getxattr,
1368 .listxattr = ll_listxattr,
1369 .removexattr = ll_removexattr,