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>
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 spin_lock(&dcache_lock);
159 list_for_each_entry_safe(dentry, tmp_alias,
160 &dir->i_dentry,d_alias) {
161 if (!list_empty(&dentry->d_subdirs)) {
162 struct dentry *child;
163 list_for_each_entry_safe(child, tmp_subdir,
166 /* XXX Print some debug here? */
168 /* Negative dentry. If we were
169 dropping dcache lock, go
170 throught the list again */
171 if (ll_drop_dentry(child))
176 spin_unlock(&dcache_lock);
180 int ll_md_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
181 void *data, int flag)
184 struct lustre_handle lockh;
188 case LDLM_CB_BLOCKING:
189 ldlm_lock2handle(lock, &lockh);
190 rc = ldlm_cli_cancel(&lockh);
192 CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
196 case LDLM_CB_CANCELING: {
197 struct inode *inode = ll_inode_from_lock(lock);
198 __u64 bits = lock->l_policy_data.l_inodebits.bits;
201 /* Invalidate all dentries associated with this inode */
205 LASSERT(lock->l_flags & LDLM_FL_CANCELING);
206 if ((bits & MDS_INODELOCK_LOOKUP) &&
207 ll_have_md_lock(inode, MDS_INODELOCK_LOOKUP))
208 bits &= ~MDS_INODELOCK_LOOKUP;
209 if ((bits & MDS_INODELOCK_UPDATE) &&
210 ll_have_md_lock(inode, MDS_INODELOCK_UPDATE))
211 bits &= ~MDS_INODELOCK_UPDATE;
212 if ((bits & MDS_INODELOCK_OPEN) &&
213 ll_have_md_lock(inode, MDS_INODELOCK_OPEN))
214 bits &= ~MDS_INODELOCK_OPEN;
216 fid = ll_inode2fid(inode);
217 if (lock->l_resource->lr_name.name[0] != fid_seq(fid) ||
218 lock->l_resource->lr_name.name[1] != fid_oid(fid) ||
219 lock->l_resource->lr_name.name[2] != fid_ver(fid)) {
220 LDLM_ERROR(lock, "data mismatch with object "
221 DFID" (%p)", PFID(fid), inode);
224 if (bits & MDS_INODELOCK_OPEN) {
226 switch (lock->l_req_mode) {
237 CERROR("Unexpected lock mode for OPEN lock "
238 "%d, inode %ld\n", lock->l_req_mode,
241 ll_md_real_close(inode, flags);
244 if (bits & MDS_INODELOCK_UPDATE)
245 ll_i2info(inode)->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
247 if (S_ISDIR(inode->i_mode) &&
248 (bits & MDS_INODELOCK_UPDATE)) {
249 CDEBUG(D_INODE, "invalidating inode %lu\n",
251 truncate_inode_pages(inode->i_mapping, 0);
252 ll_drop_negative_dentry(inode);
255 if (inode->i_sb->s_root &&
256 inode != inode->i_sb->s_root->d_inode &&
257 (bits & MDS_INODELOCK_LOOKUP))
258 ll_unhash_aliases(inode);
269 __u32 ll_i2suppgid(struct inode *i)
271 if (in_group_p(i->i_gid))
272 return (__u32)i->i_gid;
277 /* Pack the required supplementary groups into the supplied groups array.
278 * If we don't need to use the groups from the target inode(s) then we
279 * instead pack one or more groups from the user's supplementary group
280 * array in case it might be useful. Not needed if doing an MDS-side upcall. */
281 void ll_i2gids(__u32 *suppgids, struct inode *i1, struct inode *i2)
288 LASSERT(suppgids != NULL);
290 suppgids[0] = ll_i2suppgid(i1);
293 suppgids[1] = ll_i2suppgid(i2);
298 for (i = 0; i < current_ngroups; i++) {
299 if (suppgids[0] == -1) {
300 if (current_groups[i] != suppgids[1])
301 suppgids[0] = current_groups[i];
304 if (suppgids[1] == -1) {
305 if (current_groups[i] != suppgids[0])
306 suppgids[1] = current_groups[i];
314 static void ll_d_add(struct dentry *de, struct inode *inode)
316 CDEBUG(D_DENTRY, "adding inode %p to dentry %p\n", inode, de);
318 if (!list_empty(&de->d_alias)) {
319 spin_unlock(&dcache_lock);
320 CERROR("dentry %.*s %p alias next %p, prev %p\n",
321 de->d_name.len, de->d_name.name, de,
322 de->d_alias.next, de->d_alias.prev);
326 list_add(&de->d_alias, &inode->i_dentry);
330 if (!d_unhashed(de)) {
331 spin_unlock(&dcache_lock);
332 CERROR("dentry %.*s %p hash next %p\n",
333 de->d_name.len, de->d_name.name, de, de->d_hash.next);
336 d_rehash_cond(de, 0);
339 /* Search "inode"'s alias list for a dentry that has the same name and parent
340 * as de. If found, return it. If not found, return de.
341 * Lustre can't use d_add_unique because don't unhash aliases for directory
342 * in ll_revalidate_it. After revaliadate inode will be have hashed aliases
343 * and it triggers BUG_ON in d_instantiate_unique (bug #10954).
345 static struct dentry *ll_find_alias(struct inode *inode, struct dentry *de)
347 struct list_head *tmp;
348 struct dentry *dentry;
349 struct dentry *last_discon = NULL;
351 spin_lock(&dcache_lock);
352 list_for_each(tmp, &inode->i_dentry) {
353 dentry = list_entry(tmp, struct dentry, d_alias);
355 /* We are called here with 'de' already on the aliases list. */
356 if (unlikely(dentry == de)) {
361 if (dentry->d_flags & DCACHE_DISCONNECTED) {
362 LASSERT(last_discon == NULL);
363 last_discon = dentry;
367 if (dentry->d_parent != de->d_parent)
370 if (dentry->d_name.hash != de->d_name.hash)
373 if (dentry->d_name.len != de->d_name.len)
376 if (memcmp(dentry->d_name.name, de->d_name.name,
377 de->d_name.len) != 0)
383 unlock_dentry(dentry);
384 ll_dops_init(dentry, 0);
385 d_rehash_cond(dentry, 0); /* avoid taking dcache_lock inside */
386 spin_unlock(&dcache_lock);
388 CDEBUG(D_DENTRY, "alias dentry %.*s (%p) parent %p inode %p "
389 "refc %d\n", de->d_name.len, de->d_name.name, de,
390 de->d_parent, de->d_inode, atomic_read(&de->d_count));
395 CDEBUG(D_DENTRY, "Reuse disconnected dentry %p inode %p "
396 "refc %d\n", last_discon, last_discon->d_inode,
397 atomic_read(&last_discon->d_count));
398 dget_locked(last_discon);
399 lock_dentry(last_discon);
400 last_discon->d_flags |= DCACHE_LUSTRE_INVALID;
401 unlock_dentry(last_discon);
402 spin_unlock(&dcache_lock);
403 ll_dops_init(last_discon, 1);
405 d_move(last_discon, de);
409 de->d_flags |= DCACHE_LUSTRE_INVALID;
412 spin_unlock(&dcache_lock);
417 int ll_lookup_it_finish(struct ptlrpc_request *request,
418 struct lookup_intent *it, void *data)
420 struct it_cb_data *icbd = data;
421 struct dentry **de = icbd->icbd_childp;
422 struct inode *parent = icbd->icbd_parent;
423 struct ll_sb_info *sbi = ll_i2sbi(parent);
424 struct inode *inode = NULL;
428 /* NB 1 request reference will be taken away by ll_intent_lock()
430 if (!it_disposition(it, DISP_LOOKUP_NEG)) {
431 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, &bits);
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;
465 /* we have lookup look - unhide dentry */
466 if (bits & MDS_INODELOCK_LOOKUP) {
468 (*de)->d_flags &= ~DCACHE_LUSTRE_INVALID;
472 ll_dops_init(*de, 1);
473 /* Check that parent has UPDATE lock. If there is none, we
474 cannot afford to hash this dentry (done by ll_d_add) as it
475 might get picked up later when UPDATE lock will appear */
476 if (ll_have_md_lock(parent, MDS_INODELOCK_UPDATE)) {
477 spin_lock(&dcache_lock);
479 spin_unlock(&dcache_lock);
481 /* negative lookup - and don't have update lock to
484 (*de)->d_flags |= DCACHE_LUSTRE_INVALID;
487 (*de)->d_inode = NULL;
488 /* We do not want to hash the dentry if don`t have a
489 * lock, but if this dentry is later used in d_move,
490 * we'd hit uninitialised list head d_hash, so we just
491 * do this to init d_hash field but leave dentry
492 * unhashed. (bug 10796). */
501 static struct dentry *ll_lookup_it(struct inode *parent, struct dentry *dentry,
502 struct lookup_intent *it, int lookup_flags)
504 struct lookup_intent lookup_it = { .it_op = IT_LOOKUP };
505 struct dentry *save = dentry, *retval;
506 struct ptlrpc_request *req = NULL;
507 struct md_op_data *op_data;
508 struct it_cb_data icbd;
513 if (dentry->d_name.len > ll_i2sbi(parent)->ll_namelen)
514 RETURN(ERR_PTR(-ENAMETOOLONG));
516 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),intent=%s\n",
517 dentry->d_name.len, dentry->d_name.name, parent->i_ino,
518 parent->i_generation, parent, LL_IT2STR(it));
520 if (d_mountpoint(dentry))
521 CERROR("Tell Peter, lookup on mtpt, it %s\n", LL_IT2STR(it));
523 ll_frob_intent(&it, &lookup_it);
525 /* As do_lookup is called before follow_mount, root dentry may be left
526 * not valid, revalidate it here. */
527 if (parent->i_sb->s_root && (parent->i_sb->s_root->d_inode == parent) &&
528 (it->it_op & (IT_OPEN | IT_CREAT))) {
529 rc = ll_inode_revalidate_it(parent->i_sb->s_root, it);
534 if (it->it_op == IT_GETATTR) {
535 first = ll_statahead_enter(parent, &dentry, 1);
537 ll_statahead_exit(parent, dentry, first);
539 RETURN(retval = dentry);
543 icbd.icbd_childp = &dentry;
544 icbd.icbd_parent = parent;
546 if (it->it_op & IT_CREAT ||
547 (it->it_op & IT_OPEN && it->it_create_mode & O_CREAT))
548 opc = LUSTRE_OPC_CREATE;
550 opc = LUSTRE_OPC_ANY;
552 op_data = ll_prep_md_op_data(NULL, parent, NULL, dentry->d_name.name,
553 dentry->d_name.len, lookup_flags, opc,
556 RETURN((void *)op_data);
558 it->it_create_mode &= ~current->fs->umask;
560 rc = md_intent_lock(ll_i2mdexp(parent), op_data, NULL, 0, it,
561 lookup_flags, &req, ll_md_blocking_ast, 0);
562 ll_finish_md_op_data(op_data);
564 GOTO(out, retval = ERR_PTR(rc));
566 rc = ll_lookup_it_finish(req, it, &icbd);
568 ll_intent_release(it);
569 GOTO(out, retval = ERR_PTR(rc));
572 if (first == -EEXIST)
573 ll_statahead_mark(parent, dentry);
575 if ((it->it_op & IT_OPEN) && dentry->d_inode &&
576 !S_ISREG(dentry->d_inode->i_mode) &&
577 !S_ISDIR(dentry->d_inode->i_mode)) {
578 ll_release_openhandle(dentry, it);
580 ll_finish_locks(it, dentry);
583 GOTO(out, retval = NULL);
585 GOTO(out, retval = dentry);
588 ptlrpc_req_finished(req);
592 #ifdef HAVE_VFS_INTENT_PATCHES
593 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
594 struct nameidata *nd)
599 if (nd && nd->flags & LOOKUP_LAST && !(nd->flags & LOOKUP_LINK_NOTLAST))
600 de = ll_lookup_it(parent, dentry, &nd->intent, nd->flags);
602 de = ll_lookup_it(parent, dentry, NULL, 0);
607 struct lookup_intent *ll_convert_intent(struct open_intent *oit,
610 struct lookup_intent *it;
612 OBD_ALLOC(it, sizeof(*it));
614 return ERR_PTR(-ENOMEM);
616 if (lookup_flags & LOOKUP_OPEN) {
618 if (lookup_flags & LOOKUP_CREATE)
619 it->it_op |= IT_CREAT;
620 it->it_create_mode = oit->create_mode;
621 it->it_flags = oit->flags;
623 it->it_op = IT_GETATTR;
626 #ifndef HAVE_FILE_IN_STRUCT_INTENT
627 /* Since there is no way to pass our intent to ll_file_open,
628 * just check the file is there. Actual open will be done
630 if (it->it_op & IT_OPEN)
631 it->it_op = IT_LOOKUP;
637 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
638 struct nameidata *nd)
643 if (nd && !(nd->flags & (LOOKUP_CONTINUE|LOOKUP_PARENT))) {
644 struct lookup_intent *it;
646 #if defined(HAVE_FILE_IN_STRUCT_INTENT) && (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,17))
647 /* Did we came here from failed revalidate just to propagate
649 if (nd->flags & LOOKUP_OPEN)
650 if (IS_ERR(nd->intent.open.file))
651 RETURN((struct dentry *)nd->intent.open.file);
654 if (ll_d2d(dentry) && ll_d2d(dentry)->lld_it) {
655 it = ll_d2d(dentry)->lld_it;
656 ll_d2d(dentry)->lld_it = NULL;
658 if ((nd->flags & LOOKUP_CREATE ) && !(nd->flags & LOOKUP_OPEN)) {
659 /* We are sure this is new dentry, so we need to create
660 our private data and set the dentry ops */
661 ll_dops_init(dentry, 1);
664 it = ll_convert_intent(&nd->intent.open, nd->flags);
666 RETURN((struct dentry *)it);
669 de = ll_lookup_it(parent, dentry, it, nd->flags);
672 if ((nd->flags & LOOKUP_OPEN) && !IS_ERR(dentry)) { /* Open */
673 if (dentry->d_inode &&
674 it_disposition(it, DISP_OPEN_OPEN)) { /* nocreate */
675 #ifdef HAVE_FILE_IN_STRUCT_INTENT
676 if (S_ISFIFO(dentry->d_inode->i_mode)) {
677 // We cannot call open here as it would
680 (struct ptlrpc_request *)
681 it->d.lustre.it_data);
683 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,17))
684 /* 2.6.1[456] have a bug in open_namei() that forgets to check
685 * nd->intent.open.file for error, so we need to return it as lookup's result
688 nd->intent.open.file->private_data = it;
689 filp =lookup_instantiate_filp(nd,dentry,
694 de = (struct dentry *) filp;
697 nd->intent.open.file->private_data = it;
698 (void)lookup_instantiate_filp(nd,dentry,
703 #else /* HAVE_FILE_IN_STRUCT_INTENT */
704 /* Release open handle as we have no way to
705 * pass it to ll_file_open */
706 ll_release_openhandle(dentry, it);
707 #endif /* HAVE_FILE_IN_STRUCT_INTENT */
708 } else if (it_disposition(it, DISP_OPEN_CREATE)) {
709 // XXX This can only reliably work on assumption
710 // that there are NO hashed negative dentries.
711 ll_d2d(dentry)->lld_it = it;
712 it = NULL; /* Will be freed in ll_create_nd */
713 /* We absolutely depend on ll_create_nd to be
714 * called to not leak this intent and possible
715 * data attached to it */
720 ll_intent_release(it);
721 OBD_FREE(it, sizeof(*it));
724 de = ll_lookup_it(parent, dentry, NULL, 0);
731 /* We depend on "mode" being set with the proper file type/umask by now */
732 static struct inode *ll_create_node(struct inode *dir, const char *name,
733 int namelen, const void *data, int datalen,
734 int mode, __u64 extra,
735 struct lookup_intent *it)
737 struct inode *inode = NULL;
738 struct ptlrpc_request *request = NULL;
739 struct ll_sb_info *sbi = ll_i2sbi(dir);
743 LASSERT(it && it->d.lustre.it_disposition);
745 LASSERT(it_disposition(it, DISP_ENQ_CREATE_REF));
746 request = it->d.lustre.it_data;
747 it_clear_disposition(it, DISP_ENQ_CREATE_REF);
748 rc = ll_prep_inode(&inode, request, dir->i_sb);
750 GOTO(out, inode = ERR_PTR(rc));
752 LASSERT(list_empty(&inode->i_dentry));
754 /* We asked for a lock on the directory, but were granted a
755 * lock on the inode. Since we finally have an inode pointer,
756 * stuff it in the lock. */
757 CDEBUG(D_DLMTRACE, "setting l_ast_data to inode %p (%lu/%u)\n",
758 inode, inode->i_ino, inode->i_generation);
759 md_set_lock_data(sbi->ll_md_exp,
760 &it->d.lustre.it_lock_handle, inode, NULL);
763 ptlrpc_req_finished(request);
768 * By the time this is called, we already have created the directory cache
769 * entry for the new file, but it is so far negative - it has no inode.
771 * We defer creating the OBD object(s) until open, to keep the intent and
772 * non-intent code paths similar, and also because we do not have the MDS
773 * inode number before calling ll_create_node() (which is needed for LOV),
774 * so we would need to do yet another RPC to the MDS to store the LOV EA
775 * data on the MDS. If needed, we would pass the PACKED lmm as data and
776 * lmm_size in datalen (the MDS still has code which will handle that).
778 * If the create succeeds, we fill in the inode information
779 * with d_instantiate().
781 static int ll_create_it(struct inode *dir, struct dentry *dentry, int mode,
782 struct lookup_intent *it)
788 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),intent=%s\n",
789 dentry->d_name.len, dentry->d_name.name, dir->i_ino,
790 dir->i_generation, dir, LL_IT2STR(it));
792 rc = it_open_error(DISP_OPEN_CREATE, it);
796 inode = ll_create_node(dir, dentry->d_name.name, dentry->d_name.len,
797 NULL, 0, mode, 0, it);
799 RETURN(PTR_ERR(inode));
802 d_instantiate(dentry, inode);
803 /* Negative dentry may be unhashed if parent does not have UPDATE lock,
804 * but some callers, e.g. do_coredump, expect dentry to be hashed after
805 * successful create. Hash it here. */
806 spin_lock(&dcache_lock);
807 if (d_unhashed(dentry))
808 d_rehash_cond(dentry, 0);
809 spin_unlock(&dcache_lock);
813 static void ll_update_times(struct ptlrpc_request *request,
816 struct mdt_body *body = req_capsule_server_get(&request->rq_pill,
820 /* mtime is always updated with ctime, but can be set in past.
821 As write and utime(2) may happen within 1 second, and utime's
822 mtime has a priority over write's one, so take mtime from mds
823 for the same ctimes. */
824 if (body->valid & OBD_MD_FLCTIME &&
825 body->ctime >= LTIME_S(inode->i_ctime)) {
826 LTIME_S(inode->i_ctime) = body->ctime;
828 if (body->valid & OBD_MD_FLMTIME) {
829 CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
830 "to "LPU64"\n", inode->i_ino,
831 LTIME_S(inode->i_mtime), body->mtime);
832 LTIME_S(inode->i_mtime) = body->mtime;
837 static int ll_new_node(struct inode *dir, struct qstr *name,
838 const char *tgt, int mode, int rdev,
839 struct dentry *dchild, __u32 opc)
841 struct ptlrpc_request *request = NULL;
842 struct md_op_data *op_data;
843 struct inode *inode = NULL;
844 struct ll_sb_info *sbi = ll_i2sbi(dir);
849 if (unlikely(tgt != NULL))
850 tgt_len = strlen(tgt) + 1;
852 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
853 name->len, 0, opc, NULL);
855 GOTO(err_exit, err = PTR_ERR(op_data));
857 err = md_create(sbi->ll_md_exp, op_data, tgt, tgt_len, mode,
858 current->fsuid, current->fsgid,
859 cfs_curproc_cap_pack(), rdev, &request);
860 ll_finish_md_op_data(op_data);
864 ll_update_times(request, dir);
867 err = ll_prep_inode(&inode, request, dchild->d_sb);
872 d_instantiate(dchild, inode);
876 ptlrpc_req_finished(request);
881 static int ll_mknod_generic(struct inode *dir, struct qstr *name, int mode,
882 unsigned rdev, struct dentry *dchild)
887 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p) mode %o dev %x\n",
888 name->len, name->name, dir->i_ino, dir->i_generation, dir,
891 mode &= ~current->fs->umask;
893 switch (mode & S_IFMT) {
895 mode |= S_IFREG; /* for mode = 0 case, fallthrough */
901 err = ll_new_node(dir, name, NULL, mode, rdev, dchild,
913 #ifndef HAVE_VFS_INTENT_PATCHES
914 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
915 int mode, struct nameidata *nd)
917 struct lookup_intent *it = ll_d2d(dentry)->lld_it;
921 return ll_mknod_generic(dir, &dentry->d_name, mode, 0, dentry);
923 ll_d2d(dentry)->lld_it = NULL;
925 /* Was there an error? Propagate it! */
926 if (it->d.lustre.it_status) {
927 rc = it->d.lustre.it_status;
931 rc = ll_create_it(dir, dentry, mode, it);
932 #ifdef HAVE_FILE_IN_STRUCT_INTENT
933 if (nd && (nd->flags & LOOKUP_OPEN) && dentry->d_inode) { /* Open */
934 nd->intent.open.file->private_data = it;
935 lookup_instantiate_filp(nd, dentry, NULL);
938 ll_release_openhandle(dentry,it);
942 ll_intent_release(it);
943 OBD_FREE(it, sizeof(*it));
948 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
949 int mode, struct nameidata *nd)
951 if (!nd || !nd->intent.d.lustre.it_disposition)
952 /* No saved request? Just mknod the file */
953 return ll_mknod_generic(dir, &dentry->d_name, mode, 0, dentry);
955 return ll_create_it(dir, dentry, mode, &nd->intent);
959 static int ll_symlink_generic(struct inode *dir, struct qstr *name,
960 const char *tgt, struct dentry *dchild)
965 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),target=%.*s\n",
966 name->len, name->name, dir->i_ino, dir->i_generation,
969 err = ll_new_node(dir, name, (char *)tgt, S_IFLNK | S_IRWXUGO,
970 0, dchild, LUSTRE_OPC_SYMLINK);
974 static int ll_link_generic(struct inode *src, struct inode *dir,
975 struct qstr *name, struct dentry *dchild)
977 struct ll_sb_info *sbi = ll_i2sbi(dir);
978 struct ptlrpc_request *request = NULL;
979 struct md_op_data *op_data;
984 "VFS Op: inode=%lu/%u(%p), dir=%lu/%u(%p), target=%.*s\n",
985 src->i_ino, src->i_generation, src, dir->i_ino,
986 dir->i_generation, dir, name->len, name->name);
988 op_data = ll_prep_md_op_data(NULL, src, dir, name->name, name->len,
989 0, LUSTRE_OPC_ANY, NULL);
991 RETURN(PTR_ERR(op_data));
993 err = md_link(sbi->ll_md_exp, op_data, &request);
994 ll_finish_md_op_data(op_data);
1000 ll_update_times(request, dir);
1003 ptlrpc_req_finished(request);
1007 static int ll_mkdir_generic(struct inode *dir, struct qstr *name,
1008 int mode, struct dentry *dchild)
1014 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
1015 name->len, name->name, dir->i_ino, dir->i_generation, dir);
1017 mode = (mode & (S_IRWXUGO|S_ISVTX) & ~current->fs->umask) | S_IFDIR;
1018 err = ll_new_node(dir, name, NULL, mode, 0, dchild, LUSTRE_OPC_MKDIR);
1023 /* Try to find the child dentry by its name.
1024 If found, put the result fid into @fid. */
1025 static void ll_get_child_fid(struct inode * dir, struct qstr *name,
1028 struct dentry *parent, *child;
1030 parent = list_entry(dir->i_dentry.next, struct dentry, d_alias);
1031 child = d_lookup(parent, name);
1034 *fid = *ll_inode2fid(child->d_inode);
1039 static int ll_rmdir_generic(struct inode *dir, struct dentry *dparent,
1040 struct dentry *dchild, struct qstr *name)
1042 struct ptlrpc_request *request = NULL;
1043 struct md_op_data *op_data;
1047 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
1048 name->len, name->name, dir->i_ino, dir->i_generation, dir);
1050 if (unlikely(ll_d_mountpoint(dparent, dchild, name)))
1053 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len,
1054 S_IFDIR, LUSTRE_OPC_ANY, NULL);
1055 if (IS_ERR(op_data))
1056 RETURN(PTR_ERR(op_data));
1058 ll_get_child_fid(dir, name, &op_data->op_fid3);
1059 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1060 ll_finish_md_op_data(op_data);
1062 ll_update_times(request, dir);
1063 ptlrpc_req_finished(request);
1067 int ll_objects_destroy(struct ptlrpc_request *request, struct inode *dir)
1069 struct mdt_body *body;
1070 struct lov_mds_md *eadata;
1071 struct lov_stripe_md *lsm = NULL;
1072 struct obd_trans_info oti = { 0 };
1074 struct obd_capa *oc = NULL;
1078 /* req is swabbed so this is safe */
1079 body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
1080 if (!(body->valid & OBD_MD_FLEASIZE))
1083 if (body->eadatasize == 0) {
1084 CERROR("OBD_MD_FLEASIZE set but eadatasize zero\n");
1085 GOTO(out, rc = -EPROTO);
1088 /* The MDS sent back the EA because we unlinked the last reference
1089 * to this file. Use this EA to unlink the objects on the OST.
1090 * It's opaque so we don't swab here; we leave it to obd_unpackmd() to
1091 * check it is complete and sensible. */
1092 eadata = req_capsule_server_sized_get(&request->rq_pill, &RMF_MDT_MD,
1094 LASSERT(eadata != NULL);
1096 rc = obd_unpackmd(ll_i2dtexp(dir), &lsm, eadata, body->eadatasize);
1098 CERROR("obd_unpackmd: %d\n", rc);
1101 LASSERT(rc >= sizeof(*lsm));
1105 GOTO(out_free_memmd, rc = -ENOMEM);
1107 oa->o_id = lsm->lsm_object_id;
1108 oa->o_gr = lsm->lsm_object_gr;
1109 oa->o_mode = body->mode & S_IFMT;
1110 oa->o_valid = OBD_MD_FLID | OBD_MD_FLTYPE | OBD_MD_FLGROUP;
1112 if (body->valid & OBD_MD_FLCOOKIE) {
1113 oa->o_valid |= OBD_MD_FLCOOKIE;
1114 oti.oti_logcookies =
1115 req_capsule_server_sized_get(&request->rq_pill,
1117 sizeof(struct llog_cookie) *
1118 lsm->lsm_stripe_count);
1119 if (oti.oti_logcookies == NULL) {
1120 oa->o_valid &= ~OBD_MD_FLCOOKIE;
1121 body->valid &= ~OBD_MD_FLCOOKIE;
1125 if (body->valid & OBD_MD_FLOSSCAPA) {
1126 rc = md_unpack_capa(ll_i2mdexp(dir), request, &RMF_CAPA2, &oc);
1128 GOTO(out_free_memmd, rc);
1131 rc = obd_destroy(ll_i2dtexp(dir), oa, lsm, &oti, ll_i2mdexp(dir), oc);
1135 CERROR("obd destroy objid "LPX64" error %d\n",
1136 lsm->lsm_object_id, rc);
1138 obd_free_memmd(ll_i2dtexp(dir), &lsm);
1143 static int ll_unlink_generic(struct inode *dir, struct dentry *dparent,
1144 struct dentry *dchild, struct qstr *name)
1146 struct ptlrpc_request *request = NULL;
1147 struct md_op_data *op_data;
1150 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
1151 name->len, name->name, dir->i_ino, dir->i_generation, dir);
1154 * XXX: unlink bind mountpoint maybe call to here,
1155 * just check it as vfs_unlink does.
1157 if (unlikely(ll_d_mountpoint(dparent, dchild, name)))
1160 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
1161 name->len, 0, LUSTRE_OPC_ANY, NULL);
1162 if (IS_ERR(op_data))
1163 RETURN(PTR_ERR(op_data));
1165 ll_get_child_fid(dir, name, &op_data->op_fid3);
1166 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1167 ll_finish_md_op_data(op_data);
1171 ll_update_times(request, dir);
1173 rc = ll_objects_destroy(request, dir);
1175 ptlrpc_req_finished(request);
1179 static int ll_rename_generic(struct inode *src, struct dentry *src_dparent,
1180 struct dentry *src_dchild, struct qstr *src_name,
1181 struct inode *tgt, struct dentry *tgt_dparent,
1182 struct dentry *tgt_dchild, struct qstr *tgt_name)
1184 struct ptlrpc_request *request = NULL;
1185 struct ll_sb_info *sbi = ll_i2sbi(src);
1186 struct md_op_data *op_data;
1189 CDEBUG(D_VFSTRACE,"VFS Op:oldname=%.*s,src_dir=%lu/%u(%p),newname=%.*s,"
1190 "tgt_dir=%lu/%u(%p)\n", src_name->len, src_name->name,
1191 src->i_ino, src->i_generation, src, tgt_name->len,
1192 tgt_name->name, tgt->i_ino, tgt->i_generation, tgt);
1194 if (unlikely(ll_d_mountpoint(src_dparent, src_dchild, src_name) ||
1195 ll_d_mountpoint(tgt_dparent, tgt_dchild, tgt_name)))
1198 op_data = ll_prep_md_op_data(NULL, src, tgt, NULL, 0, 0,
1199 LUSTRE_OPC_ANY, NULL);
1200 if (IS_ERR(op_data))
1201 RETURN(PTR_ERR(op_data));
1203 ll_get_child_fid(src, src_name, &op_data->op_fid3);
1204 ll_get_child_fid(tgt, tgt_name, &op_data->op_fid4);
1205 err = md_rename(sbi->ll_md_exp, op_data,
1206 src_name->name, src_name->len,
1207 tgt_name->name, tgt_name->len, &request);
1208 ll_finish_md_op_data(op_data);
1210 ll_update_times(request, src);
1211 ll_update_times(request, tgt);
1212 err = ll_objects_destroy(request, src);
1215 ptlrpc_req_finished(request);
1220 #ifdef HAVE_VFS_INTENT_PATCHES
1221 static int ll_mknod_raw(struct nameidata *nd, int mode, dev_t rdev)
1223 return ll_mknod_generic(nd->dentry->d_inode, &nd->last, mode,rdev,NULL);
1225 static int ll_rename_raw(struct nameidata *srcnd, struct nameidata *tgtnd)
1227 return ll_rename_generic(srcnd->dentry->d_inode, srcnd->dentry,
1229 tgtnd->dentry->d_inode, tgtnd->dentry,
1230 NULL, &tgtnd->last);
1232 static int ll_link_raw(struct nameidata *srcnd, struct nameidata *tgtnd)
1234 return ll_link_generic(srcnd->dentry->d_inode, tgtnd->dentry->d_inode,
1235 &tgtnd->last, NULL);
1237 static int ll_symlink_raw(struct nameidata *nd, const char *tgt)
1239 return ll_symlink_generic(nd->dentry->d_inode, &nd->last, tgt, NULL);
1241 static int ll_rmdir_raw(struct nameidata *nd)
1243 return ll_rmdir_generic(nd->dentry->d_inode, nd->dentry, NULL,
1246 static int ll_mkdir_raw(struct nameidata *nd, int mode)
1248 return ll_mkdir_generic(nd->dentry->d_inode, &nd->last, mode, NULL);
1250 static int ll_unlink_raw(struct nameidata *nd)
1252 return ll_unlink_generic(nd->dentry->d_inode, nd->dentry, NULL,
1257 static int ll_mknod(struct inode *dir, struct dentry *dchild, int mode,
1260 return ll_mknod_generic(dir, &dchild->d_name, mode,
1261 old_encode_dev(rdev), dchild);
1264 static int ll_unlink(struct inode * dir, struct dentry *dentry)
1266 return ll_unlink_generic(dir, NULL, dentry, &dentry->d_name);
1268 static int ll_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1270 return ll_mkdir_generic(dir, &dentry->d_name, mode, dentry);
1272 static int ll_rmdir(struct inode *dir, struct dentry *dentry)
1274 return ll_rmdir_generic(dir, NULL, dentry, &dentry->d_name);
1276 static int ll_symlink(struct inode *dir, struct dentry *dentry,
1277 const char *oldname)
1279 return ll_symlink_generic(dir, &dentry->d_name, oldname, dentry);
1281 static int ll_link(struct dentry *old_dentry, struct inode *dir,
1282 struct dentry *new_dentry)
1284 return ll_link_generic(old_dentry->d_inode, dir, &new_dentry->d_name,
1287 static int ll_rename(struct inode *old_dir, struct dentry *old_dentry,
1288 struct inode *new_dir, struct dentry *new_dentry)
1291 err = ll_rename_generic(old_dir, NULL,
1292 old_dentry, &old_dentry->d_name,
1293 new_dir, NULL, new_dentry,
1294 &new_dentry->d_name);
1296 #ifndef HAVE_FS_RENAME_DOES_D_MOVE
1297 if (!S_ISDIR(old_dentry->d_inode->i_mode))
1299 d_move(old_dentry, new_dentry);
1304 struct inode_operations ll_dir_inode_operations = {
1305 #ifdef HAVE_VFS_INTENT_PATCHES
1306 .link_raw = ll_link_raw,
1307 .unlink_raw = ll_unlink_raw,
1308 .symlink_raw = ll_symlink_raw,
1309 .mkdir_raw = ll_mkdir_raw,
1310 .rmdir_raw = ll_rmdir_raw,
1311 .mknod_raw = ll_mknod_raw,
1312 .rename_raw = ll_rename_raw,
1313 .setattr = ll_setattr,
1314 .setattr_raw = ll_setattr_raw,
1317 .lookup = ll_lookup_nd,
1318 .create = ll_create_nd,
1319 /* We need all these non-raw things for NFSD, to not patch it. */
1320 .unlink = ll_unlink,
1323 .symlink = ll_symlink,
1325 .rename = ll_rename,
1326 .setattr = ll_setattr,
1327 .getattr = ll_getattr,
1328 .permission = ll_inode_permission,
1329 .setxattr = ll_setxattr,
1330 .getxattr = ll_getxattr,
1331 .listxattr = ll_listxattr,
1332 .removexattr = ll_removexattr,
1335 struct inode_operations ll_special_inode_operations = {
1336 #ifdef HAVE_VFS_INTENT_PATCHES
1337 .setattr_raw = ll_setattr_raw,
1339 .setattr = ll_setattr,
1340 .getattr = ll_getattr,
1341 .permission = ll_inode_permission,
1342 .setxattr = ll_setxattr,
1343 .getxattr = ll_getxattr,
1344 .listxattr = ll_listxattr,
1345 .removexattr = ll_removexattr,