1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
6 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
8 * This program is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License version 2 only,
10 * as published by the Free Software Foundation.
12 * This program is distributed in the hope that it will be useful, but
13 * WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * General Public License version 2 for more details (a copy is included
16 * in the LICENSE file that accompanied this code).
18 * You should have received a copy of the GNU General Public License
19 * version 2 along with this program; If not, see
20 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
22 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
23 * CA 95054 USA or visit www.sun.com if you need additional information or
29 * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
30 * Use is subject to license terms.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
38 #include <linux/sched.h>
40 #include <linux/smp_lock.h>
41 #include <linux/quotaops.h>
42 #include <linux/highmem.h>
43 #include <linux/pagemap.h>
45 #define DEBUG_SUBSYSTEM S_LLITE
47 #include <obd_support.h>
48 #include <lustre_fid.h>
49 #include <lustre_lite.h>
50 #include <lustre_dlm.h>
51 #include <lustre_ver.h>
52 #include <lustre_mdc.h>
53 #include "llite_internal.h"
56 * Check if we have something mounted at the named dchild.
57 * In such a case there would always be dentry present.
59 static int ll_d_mountpoint(struct dentry *dparent, struct dentry *dchild,
64 if (unlikely(dchild)) {
65 mounted = d_mountpoint(dchild);
67 dchild = d_lookup(dparent, name);
69 mounted = d_mountpoint(dchild);
76 int ll_unlock(__u32 mode, struct lustre_handle *lockh)
80 ldlm_lock_decref(lockh, mode);
86 /* called from iget5_locked->find_inode() under inode_lock spinlock */
87 static int ll_test_inode(struct inode *inode, void *opaque)
89 struct ll_inode_info *lli = ll_i2info(inode);
90 struct lustre_md *md = opaque;
92 if (unlikely(!(md->body->valid & OBD_MD_FLID))) {
93 CERROR("MDS body missing FID\n");
97 if (!lu_fid_eq(&lli->lli_fid, &md->body->fid1))
103 static int ll_set_inode(struct inode *inode, void *opaque)
105 struct ll_inode_info *lli = ll_i2info(inode);
106 struct mdt_body *body = ((struct lustre_md *)opaque)->body;
108 if (unlikely(!(body->valid & OBD_MD_FLID))) {
109 CERROR("MDS body missing FID\n");
113 lli->lli_fid = body->fid1;
119 * Get an inode by inode number (already instantiated by the intent lookup).
120 * Returns inode or NULL
122 struct inode *ll_iget(struct super_block *sb, ino_t hash,
123 struct lustre_md *md)
129 inode = iget5_locked(sb, hash, ll_test_inode, ll_set_inode, md);
132 if (inode->i_state & I_NEW) {
135 ll_read_inode2(inode, md);
136 rc = cl_inode_init(inode, md);
139 make_bad_inode(inode);
140 unlock_new_inode(inode);
144 unlock_new_inode(inode);
145 } else if (!(inode->i_state & (I_FREEING | I_CLEAR)))
146 ll_update_inode(inode, md);
147 CDEBUG(D_VFSTRACE, "got inode: %p for "DFID"\n",
148 inode, PFID(&md->body->fid1));
153 static void ll_drop_negative_dentry(struct inode *dir)
155 struct dentry *dentry, *tmp_alias, *tmp_subdir;
157 cfs_spin_lock(&ll_lookup_lock);
158 spin_lock(&dcache_lock);
160 list_for_each_entry_safe(dentry, tmp_alias,
161 &dir->i_dentry,d_alias) {
162 if (!list_empty(&dentry->d_subdirs)) {
163 struct dentry *child;
164 list_for_each_entry_safe(child, tmp_subdir,
167 /* XXX Print some debug here? */
169 /* Negative dentry. If we were
170 dropping dcache lock, go
171 throught the list again */
172 if (ll_drop_dentry(child))
177 spin_unlock(&dcache_lock);
178 cfs_spin_unlock(&ll_lookup_lock);
182 int ll_md_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
183 void *data, int flag)
186 struct lustre_handle lockh;
190 case LDLM_CB_BLOCKING:
191 ldlm_lock2handle(lock, &lockh);
192 rc = ldlm_cli_cancel(&lockh);
194 CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
198 case LDLM_CB_CANCELING: {
199 struct inode *inode = ll_inode_from_lock(lock);
200 struct ll_inode_info *lli;
201 __u64 bits = lock->l_policy_data.l_inodebits.bits;
203 ldlm_mode_t mode = lock->l_req_mode;
205 /* Invalidate all dentries associated with this inode */
209 LASSERT(lock->l_flags & LDLM_FL_CANCELING);
210 /* For OPEN locks we differentiate between lock modes - CR, CW. PR - bug 22891 */
211 if ((bits & MDS_INODELOCK_LOOKUP) &&
212 ll_have_md_lock(inode, MDS_INODELOCK_LOOKUP, LCK_MINMODE))
213 bits &= ~MDS_INODELOCK_LOOKUP;
214 if ((bits & MDS_INODELOCK_UPDATE) &&
215 ll_have_md_lock(inode, MDS_INODELOCK_UPDATE, LCK_MINMODE))
216 bits &= ~MDS_INODELOCK_UPDATE;
217 if ((bits & MDS_INODELOCK_OPEN) &&
218 ll_have_md_lock(inode, MDS_INODELOCK_OPEN, mode))
219 bits &= ~MDS_INODELOCK_OPEN;
221 fid = ll_inode2fid(inode);
222 if (lock->l_resource->lr_name.name[0] != fid_seq(fid) ||
223 lock->l_resource->lr_name.name[1] != fid_oid(fid) ||
224 lock->l_resource->lr_name.name[2] != fid_ver(fid)) {
225 LDLM_ERROR(lock, "data mismatch with object "
226 DFID" (%p)", PFID(fid), inode);
229 if (bits & MDS_INODELOCK_OPEN) {
231 switch (lock->l_req_mode) {
242 CERROR("Unexpected lock mode for OPEN lock "
243 "%d, inode %ld\n", lock->l_req_mode,
246 ll_md_real_close(inode, flags);
249 lli = ll_i2info(inode);
250 if (bits & MDS_INODELOCK_UPDATE)
251 lli->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
253 if (S_ISDIR(inode->i_mode) &&
254 (bits & MDS_INODELOCK_UPDATE)) {
255 CDEBUG(D_INODE, "invalidating inode %lu\n",
257 truncate_inode_pages(inode->i_mapping, 0);
258 ll_drop_negative_dentry(inode);
261 if (inode->i_sb->s_root &&
262 inode != inode->i_sb->s_root->d_inode &&
263 (bits & MDS_INODELOCK_LOOKUP))
264 ll_unhash_aliases(inode);
275 __u32 ll_i2suppgid(struct inode *i)
277 if (cfs_curproc_is_in_groups(i->i_gid))
278 return (__u32)i->i_gid;
283 /* Pack the required supplementary groups into the supplied groups array.
284 * If we don't need to use the groups from the target inode(s) then we
285 * instead pack one or more groups from the user's supplementary group
286 * array in case it might be useful. Not needed if doing an MDS-side upcall. */
287 void ll_i2gids(__u32 *suppgids, struct inode *i1, struct inode *i2)
294 LASSERT(suppgids != NULL);
296 suppgids[0] = ll_i2suppgid(i1);
299 suppgids[1] = ll_i2suppgid(i2);
304 for (i = 0; i < current_ngroups; i++) {
305 if (suppgids[0] == -1) {
306 if (current_groups[i] != suppgids[1])
307 suppgids[0] = current_groups[i];
310 if (suppgids[1] == -1) {
311 if (current_groups[i] != suppgids[0])
312 suppgids[1] = current_groups[i];
320 static void ll_d_add(struct dentry *de, struct inode *inode)
322 CDEBUG(D_DENTRY, "adding inode %p to dentry %p\n", inode, de);
324 if (!list_empty(&de->d_alias)) {
325 spin_unlock(&dcache_lock);
326 CERROR("dentry %.*s %p alias next %p, prev %p\n",
327 de->d_name.len, de->d_name.name, de,
328 de->d_alias.next, de->d_alias.prev);
332 list_add(&de->d_alias, &inode->i_dentry);
336 if (!d_unhashed(de)) {
337 spin_unlock(&dcache_lock);
338 CERROR("dentry %.*s %p hash next %p\n",
339 de->d_name.len, de->d_name.name, de, de->d_hash.next);
342 d_rehash_cond(de, 0);
345 /* Search "inode"'s alias list for a dentry that has the same name and parent
346 * as de. If found, return it. If not found, return de.
347 * Lustre can't use d_add_unique because don't unhash aliases for directory
348 * in ll_revalidate_it. After revaliadate inode will be have hashed aliases
349 * and it triggers BUG_ON in d_instantiate_unique (bug #10954).
351 static struct dentry *ll_find_alias(struct inode *inode, struct dentry *de)
353 struct list_head *tmp;
354 struct dentry *dentry;
355 struct dentry *last_discon = NULL;
357 cfs_spin_lock(&ll_lookup_lock);
358 spin_lock(&dcache_lock);
359 list_for_each(tmp, &inode->i_dentry) {
360 dentry = list_entry(tmp, struct dentry, d_alias);
362 /* We are called here with 'de' already on the aliases list. */
363 if (unlikely(dentry == de)) {
368 if (dentry->d_flags & DCACHE_DISCONNECTED) {
369 LASSERT(last_discon == NULL);
370 last_discon = dentry;
374 if (dentry->d_parent != de->d_parent)
377 if (dentry->d_name.hash != de->d_name.hash)
380 if (dentry->d_name.len != de->d_name.len)
383 if (memcmp(dentry->d_name.name, de->d_name.name,
384 de->d_name.len) != 0)
390 unlock_dentry(dentry);
391 ll_dops_init(dentry, 0);
392 d_rehash_cond(dentry, 0); /* avoid taking dcache_lock inside */
393 spin_unlock(&dcache_lock);
394 cfs_spin_unlock(&ll_lookup_lock);
396 CDEBUG(D_DENTRY, "alias dentry %.*s (%p) parent %p inode %p "
397 "refc %d\n", de->d_name.len, de->d_name.name, de,
398 de->d_parent, de->d_inode, atomic_read(&de->d_count));
403 CDEBUG(D_DENTRY, "Reuse disconnected dentry %p inode %p "
404 "refc %d\n", last_discon, last_discon->d_inode,
405 atomic_read(&last_discon->d_count));
406 dget_locked(last_discon);
407 lock_dentry(last_discon);
408 last_discon->d_flags |= DCACHE_LUSTRE_INVALID;
409 unlock_dentry(last_discon);
410 spin_unlock(&dcache_lock);
411 cfs_spin_unlock(&ll_lookup_lock);
412 ll_dops_init(last_discon, 1);
414 d_move(last_discon, de);
419 de->d_flags |= DCACHE_LUSTRE_INVALID;
423 spin_unlock(&dcache_lock);
424 cfs_spin_unlock(&ll_lookup_lock);
429 int ll_lookup_it_finish(struct ptlrpc_request *request,
430 struct lookup_intent *it, void *data)
432 struct it_cb_data *icbd = data;
433 struct dentry **de = icbd->icbd_childp;
434 struct inode *parent = icbd->icbd_parent;
435 struct ll_sb_info *sbi = ll_i2sbi(parent);
436 struct inode *inode = NULL;
440 /* NB 1 request reference will be taken away by ll_intent_lock()
442 if (!it_disposition(it, DISP_LOOKUP_NEG)) {
443 struct dentry *save = *de;
446 rc = ll_prep_inode(&inode, request, (*de)->d_sb);
450 CDEBUG(D_DLMTRACE, "setting l_data to inode %p (%lu/%u)\n",
451 inode, inode->i_ino, inode->i_generation);
452 md_set_lock_data(sbi->ll_md_exp,
453 &it->d.lustre.it_lock_handle, inode, &bits);
455 /* We used to query real size from OSTs here, but actually
456 this is not needed. For stat() calls size would be updated
457 from subsequent do_revalidate()->ll_inode_revalidate_it() in
459 vfs_getattr_it->ll_getattr()->ll_inode_revalidate_it() in 2.6
460 Everybody else who needs correct file size would call
461 cl_glimpse_size or some equivalent themselves anyway.
462 Also see bug 7198. */
463 ll_dops_init(*de, 1);
464 *de = ll_find_alias(inode, *de);
466 struct ll_dentry_data *lld = ll_d2d(*de);
468 /* just make sure the ll_dentry_data is ready */
469 if (unlikely(lld == NULL))
470 ll_dops_init(*de, 1);
472 /* we have lookup look - unhide dentry */
473 if (bits & MDS_INODELOCK_LOOKUP) {
475 (*de)->d_flags &= ~DCACHE_LUSTRE_INVALID;
479 ll_dops_init(*de, 1);
480 /* Check that parent has UPDATE lock. If there is none, we
481 cannot afford to hash this dentry (done by ll_d_add) as it
482 might get picked up later when UPDATE lock will appear */
483 if (ll_have_md_lock(parent, MDS_INODELOCK_UPDATE, LCK_MINMODE)) {
484 spin_lock(&dcache_lock);
486 spin_unlock(&dcache_lock);
488 /* negative lookup - and don't have update lock to
491 (*de)->d_flags |= DCACHE_LUSTRE_INVALID;
494 (*de)->d_inode = NULL;
495 /* We do not want to hash the dentry if don`t have a
496 * lock, but if this dentry is later used in d_move,
497 * we'd hit uninitialised list head d_hash, so we just
498 * do this to init d_hash field but leave dentry
499 * unhashed. (bug 10796). */
508 static struct dentry *ll_lookup_it(struct inode *parent, struct dentry *dentry,
509 struct lookup_intent *it, int lookup_flags)
511 struct lookup_intent lookup_it = { .it_op = IT_LOOKUP };
512 struct dentry *save = dentry, *retval;
513 struct ptlrpc_request *req = NULL;
514 struct md_op_data *op_data;
515 struct it_cb_data icbd;
520 if (dentry->d_name.len > ll_i2sbi(parent)->ll_namelen)
521 RETURN(ERR_PTR(-ENAMETOOLONG));
523 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),intent=%s\n",
524 dentry->d_name.len, dentry->d_name.name, parent->i_ino,
525 parent->i_generation, parent, LL_IT2STR(it));
527 if (d_mountpoint(dentry))
528 CERROR("Tell Peter, lookup on mtpt, it %s\n", LL_IT2STR(it));
530 ll_frob_intent(&it, &lookup_it);
532 /* As do_lookup is called before follow_mount, root dentry may be left
533 * not valid, revalidate it here. */
534 if (parent->i_sb->s_root && (parent->i_sb->s_root->d_inode == parent) &&
535 (it->it_op & (IT_OPEN | IT_CREAT))) {
536 rc = ll_inode_revalidate_it(parent->i_sb->s_root, it);
541 if (it->it_op == IT_GETATTR) {
542 first = ll_statahead_enter(parent, &dentry, 1);
544 ll_statahead_exit(parent, dentry, first);
546 RETURN(retval = dentry);
550 icbd.icbd_childp = &dentry;
551 icbd.icbd_parent = parent;
553 if (it->it_op & IT_CREAT ||
554 (it->it_op & IT_OPEN && it->it_create_mode & O_CREAT))
555 opc = LUSTRE_OPC_CREATE;
557 opc = LUSTRE_OPC_ANY;
559 op_data = ll_prep_md_op_data(NULL, parent, NULL, dentry->d_name.name,
560 dentry->d_name.len, lookup_flags, opc,
563 RETURN((void *)op_data);
565 it->it_create_mode &= ~cfs_curproc_umask();
567 rc = md_intent_lock(ll_i2mdexp(parent), op_data, NULL, 0, it,
568 lookup_flags, &req, ll_md_blocking_ast, 0);
569 ll_finish_md_op_data(op_data);
571 GOTO(out, retval = ERR_PTR(rc));
573 rc = ll_lookup_it_finish(req, it, &icbd);
575 ll_intent_release(it);
576 GOTO(out, retval = ERR_PTR(rc));
579 if (first == -EEXIST)
580 ll_statahead_mark(parent, dentry);
582 if ((it->it_op & IT_OPEN) && dentry->d_inode &&
583 !S_ISREG(dentry->d_inode->i_mode) &&
584 !S_ISDIR(dentry->d_inode->i_mode)) {
585 ll_release_openhandle(dentry, it);
587 ll_lookup_finish_locks(it, dentry);
590 GOTO(out, retval = NULL);
592 GOTO(out, retval = dentry);
595 ptlrpc_req_finished(req);
599 #ifdef HAVE_VFS_INTENT_PATCHES
600 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
601 struct nameidata *nd)
606 if (nd && nd->flags & LOOKUP_LAST && !(nd->flags & LOOKUP_LINK_NOTLAST))
607 de = ll_lookup_it(parent, dentry, &nd->intent, nd->flags);
609 de = ll_lookup_it(parent, dentry, NULL, 0);
614 struct lookup_intent *ll_convert_intent(struct open_intent *oit,
617 struct lookup_intent *it;
619 OBD_ALLOC(it, sizeof(*it));
621 return ERR_PTR(-ENOMEM);
623 if (lookup_flags & LOOKUP_OPEN) {
625 if (lookup_flags & LOOKUP_CREATE)
626 it->it_op |= IT_CREAT;
627 it->it_create_mode = oit->create_mode;
628 it->it_flags = oit->flags;
630 it->it_op = IT_GETATTR;
633 #ifndef HAVE_FILE_IN_STRUCT_INTENT
634 /* Since there is no way to pass our intent to ll_file_open,
635 * just check the file is there. Actual open will be done
637 if (it->it_op & IT_OPEN)
638 it->it_op = IT_LOOKUP;
644 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
645 struct nameidata *nd)
650 if (nd && !(nd->flags & (LOOKUP_CONTINUE|LOOKUP_PARENT))) {
651 struct lookup_intent *it;
653 #if defined(HAVE_FILE_IN_STRUCT_INTENT) && (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,17))
654 /* Did we came here from failed revalidate just to propagate
656 if (nd->flags & LOOKUP_OPEN)
657 if (IS_ERR(nd->intent.open.file))
658 RETURN((struct dentry *)nd->intent.open.file);
660 if (ll_d2d(dentry) && ll_d2d(dentry)->lld_it) {
661 it = ll_d2d(dentry)->lld_it;
662 ll_d2d(dentry)->lld_it = NULL;
664 if ((nd->flags & LOOKUP_CREATE ) && !(nd->flags & LOOKUP_OPEN)) {
665 /* We are sure this is new dentry, so we need to create
666 our private data and set the dentry ops */
667 ll_dops_init(dentry, 1);
670 it = ll_convert_intent(&nd->intent.open, nd->flags);
672 RETURN((struct dentry *)it);
675 de = ll_lookup_it(parent, dentry, it, nd->flags);
678 if ((nd->flags & LOOKUP_OPEN) && !IS_ERR(dentry)) { /* Open */
679 if (dentry->d_inode &&
680 it_disposition(it, DISP_OPEN_OPEN)) { /* nocreate */
681 #ifdef HAVE_FILE_IN_STRUCT_INTENT
682 if (S_ISFIFO(dentry->d_inode->i_mode)) {
683 // We cannot call open here as it would
686 (struct ptlrpc_request *)
687 it->d.lustre.it_data);
689 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,17))
690 /* 2.6.1[456] have a bug in open_namei() that forgets to check
691 * nd->intent.open.file for error, so we need to return it as lookup's result
694 nd->intent.open.file->private_data = it;
695 filp =lookup_instantiate_filp(nd,dentry,
700 de = (struct dentry *) filp;
703 nd->intent.open.file->private_data = it;
704 (void)lookup_instantiate_filp(nd,dentry,
709 #else /* HAVE_FILE_IN_STRUCT_INTENT */
710 /* Release open handle as we have no way to
711 * pass it to ll_file_open */
712 ll_release_openhandle(dentry, it);
713 #endif /* HAVE_FILE_IN_STRUCT_INTENT */
714 } else if (it_disposition(it, DISP_OPEN_CREATE)) {
715 // XXX This can only reliably work on assumption
716 // that there are NO hashed negative dentries.
717 ll_d2d(dentry)->lld_it = it;
718 it = NULL; /* Will be freed in ll_create_nd */
719 /* We absolutely depend on ll_create_nd to be
720 * called to not leak this intent and possible
721 * data attached to it */
726 ll_intent_release(it);
727 OBD_FREE(it, sizeof(*it));
730 de = ll_lookup_it(parent, dentry, NULL, 0);
737 /* We depend on "mode" being set with the proper file type/umask by now */
738 static struct inode *ll_create_node(struct inode *dir, const char *name,
739 int namelen, const void *data, int datalen,
740 int mode, __u64 extra,
741 struct lookup_intent *it)
743 struct inode *inode = NULL;
744 struct ptlrpc_request *request = NULL;
745 struct ll_sb_info *sbi = ll_i2sbi(dir);
749 LASSERT(it && it->d.lustre.it_disposition);
751 LASSERT(it_disposition(it, DISP_ENQ_CREATE_REF));
752 request = it->d.lustre.it_data;
753 it_clear_disposition(it, DISP_ENQ_CREATE_REF);
754 rc = ll_prep_inode(&inode, request, dir->i_sb);
756 GOTO(out, inode = ERR_PTR(rc));
758 LASSERT(list_empty(&inode->i_dentry));
760 /* We asked for a lock on the directory, but were granted a
761 * lock on the inode. Since we finally have an inode pointer,
762 * stuff it in the lock. */
763 CDEBUG(D_DLMTRACE, "setting l_ast_data to inode %p (%lu/%u)\n",
764 inode, inode->i_ino, inode->i_generation);
765 md_set_lock_data(sbi->ll_md_exp,
766 &it->d.lustre.it_lock_handle, inode, NULL);
769 ptlrpc_req_finished(request);
774 * By the time this is called, we already have created the directory cache
775 * entry for the new file, but it is so far negative - it has no inode.
777 * We defer creating the OBD object(s) until open, to keep the intent and
778 * non-intent code paths similar, and also because we do not have the MDS
779 * inode number before calling ll_create_node() (which is needed for LOV),
780 * so we would need to do yet another RPC to the MDS to store the LOV EA
781 * data on the MDS. If needed, we would pass the PACKED lmm as data and
782 * lmm_size in datalen (the MDS still has code which will handle that).
784 * If the create succeeds, we fill in the inode information
785 * with d_instantiate().
787 static int ll_create_it(struct inode *dir, struct dentry *dentry, int mode,
788 struct lookup_intent *it)
794 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),intent=%s\n",
795 dentry->d_name.len, dentry->d_name.name, dir->i_ino,
796 dir->i_generation, dir, LL_IT2STR(it));
798 rc = it_open_error(DISP_OPEN_CREATE, it);
802 inode = ll_create_node(dir, dentry->d_name.name, dentry->d_name.len,
803 NULL, 0, mode, 0, it);
805 RETURN(PTR_ERR(inode));
807 d_instantiate(dentry, inode);
808 /* Negative dentry may be unhashed if parent does not have UPDATE lock,
809 * but some callers, e.g. do_coredump, expect dentry to be hashed after
810 * successful create. Hash it here. */
811 spin_lock(&dcache_lock);
812 if (d_unhashed(dentry))
813 d_rehash_cond(dentry, 0);
814 spin_unlock(&dcache_lock);
818 static void ll_update_times(struct ptlrpc_request *request,
821 struct mdt_body *body = req_capsule_server_get(&request->rq_pill,
825 if (body->valid & OBD_MD_FLMTIME &&
826 body->mtime > LTIME_S(inode->i_mtime)) {
827 CDEBUG(D_INODE, "setting ino %lu mtime from %lu to "LPU64"\n",
828 inode->i_ino, LTIME_S(inode->i_mtime), body->mtime);
829 LTIME_S(inode->i_mtime) = body->mtime;
831 if (body->valid & OBD_MD_FLCTIME &&
832 body->ctime > LTIME_S(inode->i_ctime))
833 LTIME_S(inode->i_ctime) = body->ctime;
836 static int ll_new_node(struct inode *dir, struct qstr *name,
837 const char *tgt, int mode, int rdev,
838 struct dentry *dchild, __u32 opc)
840 struct ptlrpc_request *request = NULL;
841 struct md_op_data *op_data;
842 struct inode *inode = NULL;
843 struct ll_sb_info *sbi = ll_i2sbi(dir);
848 if (unlikely(tgt != NULL))
849 tgt_len = strlen(tgt) + 1;
851 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
852 name->len, 0, opc, NULL);
854 GOTO(err_exit, err = PTR_ERR(op_data));
856 err = md_create(sbi->ll_md_exp, op_data, tgt, tgt_len, mode,
857 cfs_curproc_fsuid(), cfs_curproc_fsgid(),
858 cfs_curproc_cap_pack(), rdev, &request);
859 ll_finish_md_op_data(op_data);
863 ll_update_times(request, dir);
866 err = ll_prep_inode(&inode, request, dchild->d_sb);
871 d_instantiate(dchild, inode);
875 ptlrpc_req_finished(request);
880 static int ll_mknod_generic(struct inode *dir, struct qstr *name, int mode,
881 unsigned rdev, struct dentry *dchild)
886 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p) mode %o dev %x\n",
887 name->len, name->name, dir->i_ino, dir->i_generation, dir,
890 mode &= ~cfs_curproc_umask();
892 switch (mode & S_IFMT) {
894 mode |= S_IFREG; /* for mode = 0 case, fallthrough */
900 err = ll_new_node(dir, name, NULL, mode, rdev, dchild,
912 #ifndef HAVE_VFS_INTENT_PATCHES
913 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
914 int mode, struct nameidata *nd)
916 struct lookup_intent *it = ll_d2d(dentry)->lld_it;
920 return ll_mknod_generic(dir, &dentry->d_name, mode, 0, dentry);
922 ll_d2d(dentry)->lld_it = NULL;
924 /* Was there an error? Propagate it! */
925 if (it->d.lustre.it_status) {
926 rc = it->d.lustre.it_status;
930 rc = ll_create_it(dir, dentry, mode, it);
931 #ifdef HAVE_FILE_IN_STRUCT_INTENT
932 if (nd && (nd->flags & LOOKUP_OPEN) && dentry->d_inode) { /* Open */
933 nd->intent.open.file->private_data = it;
934 lookup_instantiate_filp(nd, dentry, NULL);
937 ll_release_openhandle(dentry,it);
941 ll_intent_release(it);
942 OBD_FREE(it, sizeof(*it));
947 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
948 int mode, struct nameidata *nd)
950 if (!nd || !nd->intent.d.lustre.it_disposition)
951 /* No saved request? Just mknod the file */
952 return ll_mknod_generic(dir, &dentry->d_name, mode, 0, dentry);
954 return ll_create_it(dir, dentry, mode, &nd->intent);
958 static int ll_symlink_generic(struct inode *dir, struct qstr *name,
959 const char *tgt, struct dentry *dchild)
964 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),target=%.*s\n",
965 name->len, name->name, dir->i_ino, dir->i_generation,
968 err = ll_new_node(dir, name, (char *)tgt, S_IFLNK | S_IRWXUGO,
969 0, dchild, LUSTRE_OPC_SYMLINK);
973 static int ll_link_generic(struct inode *src, struct inode *dir,
974 struct qstr *name, struct dentry *dchild)
976 struct ll_sb_info *sbi = ll_i2sbi(dir);
977 struct ptlrpc_request *request = NULL;
978 struct md_op_data *op_data;
983 "VFS Op: inode=%lu/%u(%p), dir=%lu/%u(%p), target=%.*s\n",
984 src->i_ino, src->i_generation, src, dir->i_ino,
985 dir->i_generation, dir, name->len, name->name);
987 op_data = ll_prep_md_op_data(NULL, src, dir, name->name, name->len,
988 0, LUSTRE_OPC_ANY, NULL);
990 RETURN(PTR_ERR(op_data));
992 err = md_link(sbi->ll_md_exp, op_data, &request);
993 ll_finish_md_op_data(op_data);
999 ll_update_times(request, dir);
1002 ptlrpc_req_finished(request);
1006 static int ll_mkdir_generic(struct inode *dir, struct qstr *name,
1007 int mode, struct dentry *dchild)
1013 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
1014 name->len, name->name, dir->i_ino, dir->i_generation, dir);
1016 mode = (mode & (S_IRWXUGO|S_ISVTX) & ~cfs_curproc_umask()) | S_IFDIR;
1017 err = ll_new_node(dir, name, NULL, mode, 0, dchild, LUSTRE_OPC_MKDIR);
1022 /* Try to find the child dentry by its name.
1023 If found, put the result fid into @fid. */
1024 static void ll_get_child_fid(struct inode * dir, struct qstr *name,
1027 struct dentry *parent, *child;
1029 parent = list_entry(dir->i_dentry.next, struct dentry, d_alias);
1030 child = d_lookup(parent, name);
1033 *fid = *ll_inode2fid(child->d_inode);
1038 static int ll_rmdir_generic(struct inode *dir, struct dentry *dparent,
1039 struct dentry *dchild, struct qstr *name)
1041 struct ptlrpc_request *request = NULL;
1042 struct md_op_data *op_data;
1046 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
1047 name->len, name->name, dir->i_ino, dir->i_generation, dir);
1049 if (unlikely(ll_d_mountpoint(dparent, dchild, name)))
1052 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len,
1053 S_IFDIR, LUSTRE_OPC_ANY, NULL);
1054 if (IS_ERR(op_data))
1055 RETURN(PTR_ERR(op_data));
1057 ll_get_child_fid(dir, name, &op_data->op_fid3);
1058 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1059 ll_finish_md_op_data(op_data);
1061 ll_update_times(request, dir);
1062 ptlrpc_req_finished(request);
1066 int ll_objects_destroy(struct ptlrpc_request *request, struct inode *dir)
1068 struct mdt_body *body;
1069 struct lov_mds_md *eadata;
1070 struct lov_stripe_md *lsm = NULL;
1071 struct obd_trans_info oti = { 0 };
1073 struct obd_capa *oc = NULL;
1077 /* req is swabbed so this is safe */
1078 body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
1079 if (!(body->valid & OBD_MD_FLEASIZE))
1082 if (body->eadatasize == 0) {
1083 CERROR("OBD_MD_FLEASIZE set but eadatasize zero\n");
1084 GOTO(out, rc = -EPROTO);
1087 /* The MDS sent back the EA because we unlinked the last reference
1088 * to this file. Use this EA to unlink the objects on the OST.
1089 * It's opaque so we don't swab here; we leave it to obd_unpackmd() to
1090 * check it is complete and sensible. */
1091 eadata = req_capsule_server_sized_get(&request->rq_pill, &RMF_MDT_MD,
1093 LASSERT(eadata != NULL);
1095 rc = obd_unpackmd(ll_i2dtexp(dir), &lsm, eadata, body->eadatasize);
1097 CERROR("obd_unpackmd: %d\n", rc);
1100 LASSERT(rc >= sizeof(*lsm));
1104 GOTO(out_free_memmd, rc = -ENOMEM);
1106 oa->o_id = lsm->lsm_object_id;
1107 oa->o_seq = lsm->lsm_object_seq;
1108 oa->o_mode = body->mode & S_IFMT;
1109 oa->o_valid = OBD_MD_FLID | OBD_MD_FLTYPE | OBD_MD_FLGROUP;
1111 if (body->valid & OBD_MD_FLCOOKIE) {
1112 oa->o_valid |= OBD_MD_FLCOOKIE;
1113 oti.oti_logcookies =
1114 req_capsule_server_sized_get(&request->rq_pill,
1116 sizeof(struct llog_cookie) *
1117 lsm->lsm_stripe_count);
1118 if (oti.oti_logcookies == NULL) {
1119 oa->o_valid &= ~OBD_MD_FLCOOKIE;
1120 body->valid &= ~OBD_MD_FLCOOKIE;
1124 if (body->valid & OBD_MD_FLOSSCAPA) {
1125 rc = md_unpack_capa(ll_i2mdexp(dir), request, &RMF_CAPA2, &oc);
1127 GOTO(out_free_memmd, rc);
1130 rc = obd_destroy(ll_i2dtexp(dir), oa, lsm, &oti, ll_i2mdexp(dir), oc);
1134 CERROR("obd destroy objid "LPX64" error %d\n",
1135 lsm->lsm_object_id, rc);
1137 obd_free_memmd(ll_i2dtexp(dir), &lsm);
1142 /* ll_unlink_generic() doesn't update the inode with the new link count.
1143 * Instead, ll_ddelete() and ll_d_iput() will update it based upon if there
1144 * is any lock existing. They will recycle dentries and inodes based upon locks
1146 static int ll_unlink_generic(struct inode *dir, struct dentry *dparent,
1147 struct dentry *dchild, struct qstr *name)
1149 struct ptlrpc_request *request = NULL;
1150 struct md_op_data *op_data;
1153 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
1154 name->len, name->name, dir->i_ino, dir->i_generation, dir);
1157 * XXX: unlink bind mountpoint maybe call to here,
1158 * just check it as vfs_unlink does.
1160 if (unlikely(ll_d_mountpoint(dparent, dchild, name)))
1163 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
1164 name->len, 0, LUSTRE_OPC_ANY, NULL);
1165 if (IS_ERR(op_data))
1166 RETURN(PTR_ERR(op_data));
1168 ll_get_child_fid(dir, name, &op_data->op_fid3);
1169 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1170 ll_finish_md_op_data(op_data);
1174 ll_update_times(request, dir);
1176 rc = ll_objects_destroy(request, dir);
1178 ptlrpc_req_finished(request);
1182 static int ll_rename_generic(struct inode *src, struct dentry *src_dparent,
1183 struct dentry *src_dchild, struct qstr *src_name,
1184 struct inode *tgt, struct dentry *tgt_dparent,
1185 struct dentry *tgt_dchild, struct qstr *tgt_name)
1187 struct ptlrpc_request *request = NULL;
1188 struct ll_sb_info *sbi = ll_i2sbi(src);
1189 struct md_op_data *op_data;
1192 CDEBUG(D_VFSTRACE,"VFS Op:oldname=%.*s,src_dir=%lu/%u(%p),newname=%.*s,"
1193 "tgt_dir=%lu/%u(%p)\n", src_name->len, src_name->name,
1194 src->i_ino, src->i_generation, src, tgt_name->len,
1195 tgt_name->name, tgt->i_ino, tgt->i_generation, tgt);
1197 if (unlikely(ll_d_mountpoint(src_dparent, src_dchild, src_name) ||
1198 ll_d_mountpoint(tgt_dparent, tgt_dchild, tgt_name)))
1201 op_data = ll_prep_md_op_data(NULL, src, tgt, NULL, 0, 0,
1202 LUSTRE_OPC_ANY, NULL);
1203 if (IS_ERR(op_data))
1204 RETURN(PTR_ERR(op_data));
1206 ll_get_child_fid(src, src_name, &op_data->op_fid3);
1207 ll_get_child_fid(tgt, tgt_name, &op_data->op_fid4);
1208 err = md_rename(sbi->ll_md_exp, op_data,
1209 src_name->name, src_name->len,
1210 tgt_name->name, tgt_name->len, &request);
1211 ll_finish_md_op_data(op_data);
1213 ll_update_times(request, src);
1214 ll_update_times(request, tgt);
1215 err = ll_objects_destroy(request, src);
1218 ptlrpc_req_finished(request);
1223 #ifdef HAVE_VFS_INTENT_PATCHES
1224 static int ll_mknod_raw(struct nameidata *nd, int mode, dev_t rdev)
1226 return ll_mknod_generic(nd->dentry->d_inode, &nd->last, mode,rdev,NULL);
1228 static int ll_rename_raw(struct nameidata *srcnd, struct nameidata *tgtnd)
1230 return ll_rename_generic(srcnd->dentry->d_inode, srcnd->dentry,
1232 tgtnd->dentry->d_inode, tgtnd->dentry,
1233 NULL, &tgtnd->last);
1235 static int ll_link_raw(struct nameidata *srcnd, struct nameidata *tgtnd)
1237 return ll_link_generic(srcnd->dentry->d_inode, tgtnd->dentry->d_inode,
1238 &tgtnd->last, NULL);
1240 static int ll_symlink_raw(struct nameidata *nd, const char *tgt)
1242 return ll_symlink_generic(nd->dentry->d_inode, &nd->last, tgt, NULL);
1244 static int ll_rmdir_raw(struct nameidata *nd)
1246 return ll_rmdir_generic(nd->dentry->d_inode, nd->dentry, NULL,
1249 static int ll_mkdir_raw(struct nameidata *nd, int mode)
1251 return ll_mkdir_generic(nd->dentry->d_inode, &nd->last, mode, NULL);
1253 static int ll_unlink_raw(struct nameidata *nd)
1255 return ll_unlink_generic(nd->dentry->d_inode, nd->dentry, NULL,
1260 static int ll_mknod(struct inode *dir, struct dentry *dchild, int mode,
1263 return ll_mknod_generic(dir, &dchild->d_name, mode,
1264 old_encode_dev(rdev), dchild);
1267 static int ll_unlink(struct inode * dir, struct dentry *dentry)
1269 return ll_unlink_generic(dir, NULL, dentry, &dentry->d_name);
1271 static int ll_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1273 return ll_mkdir_generic(dir, &dentry->d_name, mode, dentry);
1275 static int ll_rmdir(struct inode *dir, struct dentry *dentry)
1277 return ll_rmdir_generic(dir, NULL, dentry, &dentry->d_name);
1279 static int ll_symlink(struct inode *dir, struct dentry *dentry,
1280 const char *oldname)
1282 return ll_symlink_generic(dir, &dentry->d_name, oldname, dentry);
1284 static int ll_link(struct dentry *old_dentry, struct inode *dir,
1285 struct dentry *new_dentry)
1287 return ll_link_generic(old_dentry->d_inode, dir, &new_dentry->d_name,
1290 static int ll_rename(struct inode *old_dir, struct dentry *old_dentry,
1291 struct inode *new_dir, struct dentry *new_dentry)
1294 err = ll_rename_generic(old_dir, NULL,
1295 old_dentry, &old_dentry->d_name,
1296 new_dir, NULL, new_dentry,
1297 &new_dentry->d_name);
1299 #ifndef HAVE_FS_RENAME_DOES_D_MOVE
1300 if (!S_ISDIR(old_dentry->d_inode->i_mode))
1302 d_move(old_dentry, new_dentry);
1307 struct inode_operations ll_dir_inode_operations = {
1308 #ifdef HAVE_VFS_INTENT_PATCHES
1309 .link_raw = ll_link_raw,
1310 .unlink_raw = ll_unlink_raw,
1311 .symlink_raw = ll_symlink_raw,
1312 .mkdir_raw = ll_mkdir_raw,
1313 .rmdir_raw = ll_rmdir_raw,
1314 .mknod_raw = ll_mknod_raw,
1315 .rename_raw = ll_rename_raw,
1316 .setattr = ll_setattr,
1317 .setattr_raw = ll_setattr_raw,
1320 .lookup = ll_lookup_nd,
1321 .create = ll_create_nd,
1322 /* We need all these non-raw things for NFSD, to not patch it. */
1323 .unlink = ll_unlink,
1326 .symlink = ll_symlink,
1328 .rename = ll_rename,
1329 .setattr = ll_setattr,
1330 .getattr = ll_getattr,
1331 .permission = ll_inode_permission,
1332 .setxattr = ll_setxattr,
1333 .getxattr = ll_getxattr,
1334 .listxattr = ll_listxattr,
1335 .removexattr = ll_removexattr,
1338 struct inode_operations ll_special_inode_operations = {
1339 #ifdef HAVE_VFS_INTENT_PATCHES
1340 .setattr_raw = ll_setattr_raw,
1342 .setattr = ll_setattr,
1343 .getattr = ll_getattr,
1344 .permission = ll_inode_permission,
1345 .setxattr = ll_setxattr,
1346 .getxattr = ll_getxattr,
1347 .listxattr = ll_listxattr,
1348 .removexattr = ll_removexattr,