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.
32 * Copyright (c) 2011 Whamcloud, Inc.
36 * This file is part of Lustre, http://www.lustre.org/
37 * Lustre is a trademark of Sun Microsystems, Inc.
41 #include <linux/sched.h>
43 #include <linux/smp_lock.h>
44 #include <linux/quotaops.h>
45 #include <linux/highmem.h>
46 #include <linux/pagemap.h>
47 #include <linux/security.h>
49 #define DEBUG_SUBSYSTEM S_LLITE
51 #include <obd_support.h>
52 #include <lustre_fid.h>
53 #include <lustre_lite.h>
54 #include <lustre_dlm.h>
55 #include <lustre_ver.h>
56 #include "llite_internal.h"
59 * Check if we have something mounted at the named dchild.
60 * In such a case there would always be dentry present.
62 static int ll_d_mountpoint(struct dentry *dparent, struct dentry *dchild,
67 if (unlikely(dchild)) {
68 mounted = d_mountpoint(dchild);
70 dchild = d_lookup(dparent, name);
72 mounted = d_mountpoint(dchild);
79 int ll_unlock(__u32 mode, struct lustre_handle *lockh)
83 ldlm_lock_decref(lockh, mode);
89 /* called from iget5_locked->find_inode() under inode_lock spinlock */
90 static int ll_test_inode(struct inode *inode, void *opaque)
92 struct ll_inode_info *lli = ll_i2info(inode);
93 struct lustre_md *md = opaque;
95 if (unlikely(!(md->body->valid & OBD_MD_FLID))) {
96 CERROR("MDS body missing FID\n");
100 if (!lu_fid_eq(&lli->lli_fid, &md->body->fid1))
106 static int ll_set_inode(struct inode *inode, void *opaque)
108 struct ll_inode_info *lli = ll_i2info(inode);
109 struct mdt_body *body = ((struct lustre_md *)opaque)->body;
111 if (unlikely(!(body->valid & OBD_MD_FLID))) {
112 CERROR("MDS body missing FID\n");
116 lli->lli_fid = body->fid1;
117 if (unlikely(!(body->valid & OBD_MD_FLTYPE))) {
118 CERROR("Can not initialize inode "DFID" without object type: "
119 "valid = "LPX64"\n", PFID(&lli->lli_fid), body->valid);
123 inode->i_mode = (inode->i_mode & ~S_IFMT) | (body->mode & S_IFMT);
124 if (unlikely(inode->i_mode == 0)) {
125 CERROR("Invalid inode "DFID" type\n", PFID(&lli->lli_fid));
136 * Get an inode by inode number (already instantiated by the intent lookup).
137 * Returns inode or NULL
139 struct inode *ll_iget(struct super_block *sb, ino_t hash,
140 struct lustre_md *md)
146 inode = iget5_locked(sb, hash, ll_test_inode, ll_set_inode, md);
149 if (inode->i_state & I_NEW) {
152 ll_read_inode2(inode, md);
153 if (S_ISREG(inode->i_mode) &&
154 ll_i2info(inode)->lli_clob == NULL)
155 rc = cl_file_inode_init(inode, md);
158 make_bad_inode(inode);
159 unlock_new_inode(inode);
163 unlock_new_inode(inode);
164 } else if (!(inode->i_state & (I_FREEING | I_CLEAR)))
165 ll_update_inode(inode, md);
166 CDEBUG(D_VFSTRACE, "got inode: %p for "DFID"\n",
167 inode, PFID(&md->body->fid1));
172 static void ll_drop_negative_dentry(struct inode *dir)
174 struct dentry *dentry, *tmp_alias, *tmp_subdir;
176 cfs_spin_lock(&ll_lookup_lock);
177 spin_lock(&dcache_lock);
179 list_for_each_entry_safe(dentry, tmp_alias,
180 &dir->i_dentry,d_alias) {
181 if (!list_empty(&dentry->d_subdirs)) {
182 struct dentry *child;
183 list_for_each_entry_safe(child, tmp_subdir,
186 /* XXX Print some debug here? */
188 /* Negative dentry. If we were
189 dropping dcache lock, go
190 throught the list again */
191 if (ll_drop_dentry(child))
196 spin_unlock(&dcache_lock);
197 cfs_spin_unlock(&ll_lookup_lock);
201 int ll_md_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
202 void *data, int flag)
205 struct lustre_handle lockh;
209 case LDLM_CB_BLOCKING:
210 ldlm_lock2handle(lock, &lockh);
211 rc = ldlm_cli_cancel(&lockh);
213 CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
217 case LDLM_CB_CANCELING: {
218 struct inode *inode = ll_inode_from_lock(lock);
219 struct ll_inode_info *lli;
220 __u64 bits = lock->l_policy_data.l_inodebits.bits;
222 ldlm_mode_t mode = lock->l_req_mode;
224 /* Invalidate all dentries associated with this inode */
228 LASSERT(lock->l_flags & LDLM_FL_CANCELING);
229 /* For OPEN locks we differentiate between lock modes - CR, CW. PR - bug 22891 */
230 if (bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE))
231 ll_have_md_lock(inode, &bits, LCK_MINMODE);
233 if (bits & MDS_INODELOCK_OPEN)
234 ll_have_md_lock(inode, &bits, mode);
236 fid = ll_inode2fid(inode);
237 if (lock->l_resource->lr_name.name[0] != fid_seq(fid) ||
238 lock->l_resource->lr_name.name[1] != fid_oid(fid) ||
239 lock->l_resource->lr_name.name[2] != fid_ver(fid)) {
240 LDLM_ERROR(lock, "data mismatch with object "
241 DFID" (%p)", PFID(fid), inode);
244 if (bits & MDS_INODELOCK_OPEN) {
246 switch (lock->l_req_mode) {
257 CERROR("Unexpected lock mode for OPEN lock "
258 "%d, inode %ld\n", lock->l_req_mode,
261 ll_md_real_close(inode, flags);
264 lli = ll_i2info(inode);
265 if (bits & MDS_INODELOCK_UPDATE)
266 lli->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
268 if (S_ISDIR(inode->i_mode) &&
269 (bits & MDS_INODELOCK_UPDATE)) {
270 CDEBUG(D_INODE, "invalidating inode %lu\n",
272 truncate_inode_pages(inode->i_mapping, 0);
273 ll_drop_negative_dentry(inode);
276 if (inode->i_sb->s_root &&
277 inode != inode->i_sb->s_root->d_inode &&
278 (bits & MDS_INODELOCK_LOOKUP))
279 ll_unhash_aliases(inode);
290 __u32 ll_i2suppgid(struct inode *i)
292 if (cfs_curproc_is_in_groups(i->i_gid))
293 return (__u32)i->i_gid;
298 /* Pack the required supplementary groups into the supplied groups array.
299 * If we don't need to use the groups from the target inode(s) then we
300 * instead pack one or more groups from the user's supplementary group
301 * array in case it might be useful. Not needed if doing an MDS-side upcall. */
302 void ll_i2gids(__u32 *suppgids, struct inode *i1, struct inode *i2)
309 LASSERT(suppgids != NULL);
311 suppgids[0] = ll_i2suppgid(i1);
314 suppgids[1] = ll_i2suppgid(i2);
319 for (i = 0; i < current_ngroups; i++) {
320 if (suppgids[0] == -1) {
321 if (current_groups[i] != suppgids[1])
322 suppgids[0] = current_groups[i];
325 if (suppgids[1] == -1) {
326 if (current_groups[i] != suppgids[0])
327 suppgids[1] = current_groups[i];
335 static void ll_d_add(struct dentry *de, struct inode *inode)
337 CDEBUG(D_DENTRY, "adding inode %p to dentry %p\n", inode, de);
339 if (!list_empty(&de->d_alias)) {
340 spin_unlock(&dcache_lock);
341 CERROR("dentry %.*s %p alias next %p, prev %p\n",
342 de->d_name.len, de->d_name.name, de,
343 de->d_alias.next, de->d_alias.prev);
347 list_add(&de->d_alias, &inode->i_dentry);
349 /* d_instantiate() replacement code should initialize security
351 security_d_instantiate(de, inode);
354 if (!d_unhashed(de)) {
355 spin_unlock(&dcache_lock);
356 CERROR("dentry %.*s %p hash next %p\n",
357 de->d_name.len, de->d_name.name, de, de->d_hash.next);
360 d_rehash_cond(de, 0);
363 /* Search "inode"'s alias list for a dentry that has the same name and parent
364 * as de. If found, return it. If not found, return de.
365 * Lustre can't use d_add_unique because don't unhash aliases for directory
366 * in ll_revalidate_it. After revaliadate inode will be have hashed aliases
367 * and it triggers BUG_ON in d_instantiate_unique (bug #10954).
369 struct dentry *ll_find_alias(struct inode *inode, struct dentry *de)
371 struct list_head *tmp;
372 struct dentry *dentry;
373 struct dentry *last_discon = NULL;
375 cfs_spin_lock(&ll_lookup_lock);
376 spin_lock(&dcache_lock);
377 list_for_each(tmp, &inode->i_dentry) {
378 dentry = list_entry(tmp, struct dentry, d_alias);
380 /* We are called here with 'de' already on the aliases list. */
381 if (unlikely(dentry == de)) {
386 if (dentry->d_flags & DCACHE_DISCONNECTED) {
387 /* LASSERT(last_discon == NULL); LU-405, bz 20055 */
388 last_discon = dentry;
392 if (dentry->d_parent != de->d_parent)
395 if (dentry->d_name.hash != de->d_name.hash)
398 if (dentry->d_name.len != de->d_name.len)
401 if (memcmp(dentry->d_name.name, de->d_name.name,
402 de->d_name.len) != 0)
406 ll_dops_init(dentry, 0, 1);
407 ll_dentry_rehash(dentry, 1);
408 spin_unlock(&dcache_lock);
409 cfs_spin_unlock(&ll_lookup_lock);
411 CDEBUG(D_DENTRY, "alias dentry %.*s (%p) parent %p inode %p "
412 "refc %d\n", de->d_name.len, de->d_name.name, de,
413 de->d_parent, de->d_inode, atomic_read(&de->d_count));
418 CDEBUG(D_DENTRY, "Reuse disconnected dentry %p inode %p "
419 "refc %d\n", last_discon, last_discon->d_inode,
420 atomic_read(&last_discon->d_count));
421 dget_locked(last_discon);
422 lock_dentry(last_discon);
423 last_discon->d_flags |= DCACHE_LUSTRE_INVALID;
424 unlock_dentry(last_discon);
425 spin_unlock(&dcache_lock);
426 cfs_spin_unlock(&ll_lookup_lock);
427 ll_dops_init(last_discon, 1, 1);
429 d_move(last_discon, de);
434 de->d_flags |= DCACHE_LUSTRE_INVALID;
438 spin_unlock(&dcache_lock);
439 cfs_spin_unlock(&ll_lookup_lock);
444 int ll_lookup_it_finish(struct ptlrpc_request *request,
445 struct lookup_intent *it, void *data)
447 struct it_cb_data *icbd = data;
448 struct dentry **de = icbd->icbd_childp;
449 struct inode *parent = icbd->icbd_parent;
450 struct inode *inode = NULL;
454 /* NB 1 request reference will be taken away by ll_intent_lock()
456 if (!it_disposition(it, DISP_LOOKUP_NEG)) {
457 struct dentry *save = *de;
460 rc = ll_prep_inode(&inode, request, (*de)->d_sb);
464 ll_set_lock_data(ll_i2sbi(parent)->ll_md_exp, inode, it, &bits);
466 /* We used to query real size from OSTs here, but actually
467 this is not needed. For stat() calls size would be updated
468 from subsequent do_revalidate()->ll_inode_revalidate_it() in
470 vfs_getattr_it->ll_getattr()->ll_inode_revalidate_it() in 2.6
471 Everybody else who needs correct file size would call
472 ll_glimpse_size or some equivalent themselves anyway.
473 Also see bug 7198. */
474 ll_dops_init(*de, 1, 1);
475 *de = ll_find_alias(inode, *de);
477 struct ll_dentry_data *lld = ll_d2d(*de);
479 /* just make sure the ll_dentry_data is ready */
480 if (unlikely(lld == NULL))
481 ll_dops_init(*de, 1, 1);
483 /* we have lookup look - unhide dentry */
484 ll_dentry_reset_flags(*de, bits);
488 ll_dops_init(*de, 1, 1);
489 /* Check that parent has UPDATE lock. If there is none, we
490 cannot afford to hash this dentry (done by ll_d_add) as it
491 might get picked up later when UPDATE lock will appear */
492 ibits = MDS_INODELOCK_UPDATE;
493 if (ll_have_md_lock(parent, &ibits, LCK_MINMODE)) {
494 spin_lock(&dcache_lock);
496 spin_unlock(&dcache_lock);
498 /* negative lookup - and don't have update lock to
501 (*de)->d_flags |= DCACHE_LUSTRE_INVALID;
504 (*de)->d_inode = NULL;
505 /* We do not want to hash the dentry if don`t have a
506 * lock, but if this dentry is later used in d_move,
507 * we'd hit uninitialised list head d_hash, so we just
508 * do this to init d_hash field but leave dentry
509 * unhashed. (bug 10796). */
518 static struct dentry *ll_lookup_it(struct inode *parent, struct dentry *dentry,
519 struct lookup_intent *it, int lookup_flags)
521 struct lookup_intent lookup_it = { .it_op = IT_LOOKUP };
522 struct dentry *save = dentry, *retval;
523 struct ptlrpc_request *req = NULL;
524 struct md_op_data *op_data;
525 struct it_cb_data icbd;
530 if (dentry->d_name.len > ll_i2sbi(parent)->ll_namelen)
531 RETURN(ERR_PTR(-ENAMETOOLONG));
533 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),intent=%s\n",
534 dentry->d_name.len, dentry->d_name.name, parent->i_ino,
535 parent->i_generation, parent, LL_IT2STR(it));
537 if (d_mountpoint(dentry))
538 CERROR("Tell Peter, lookup on mtpt, it %s\n", LL_IT2STR(it));
540 ll_frob_intent(&it, &lookup_it);
542 /* As do_lookup is called before follow_mount, root dentry may be left
543 * not valid, revalidate it here. */
544 if (parent->i_sb->s_root && (parent->i_sb->s_root->d_inode == parent) &&
545 (it->it_op & (IT_OPEN | IT_CREAT))) {
546 rc = ll_inode_revalidate_it(parent->i_sb->s_root, it,
547 MDS_INODELOCK_LOOKUP);
552 if (it->it_op == IT_GETATTR) {
553 rc = ll_statahead_enter(parent, &dentry, 0);
556 GOTO(out, retval = NULL);
557 GOTO(out, retval = dentry);
561 icbd.icbd_childp = &dentry;
562 icbd.icbd_parent = parent;
564 if (it->it_op & IT_CREAT ||
565 (it->it_op & IT_OPEN && it->it_create_mode & O_CREAT))
566 opc = LUSTRE_OPC_CREATE;
568 opc = LUSTRE_OPC_ANY;
570 op_data = ll_prep_md_op_data(NULL, parent, NULL, dentry->d_name.name,
571 dentry->d_name.len, lookup_flags, opc,
574 RETURN((void *)op_data);
576 it->it_create_mode &= ~cfs_curproc_umask();
578 rc = md_intent_lock(ll_i2mdexp(parent), op_data, NULL, 0, it,
579 lookup_flags, &req, ll_md_blocking_ast, 0);
580 ll_finish_md_op_data(op_data);
582 GOTO(out, retval = ERR_PTR(rc));
584 rc = ll_lookup_it_finish(req, it, &icbd);
586 ll_intent_release(it);
587 GOTO(out, retval = ERR_PTR(rc));
590 if ((it->it_op & IT_OPEN) && dentry->d_inode &&
591 !S_ISREG(dentry->d_inode->i_mode) &&
592 !S_ISDIR(dentry->d_inode->i_mode)) {
593 ll_release_openhandle(dentry, it);
595 ll_lookup_finish_locks(it, dentry);
598 GOTO(out, retval = NULL);
600 GOTO(out, retval = dentry);
603 ptlrpc_req_finished(req);
604 if (it->it_op == IT_GETATTR && (retval == NULL || retval == dentry))
605 ll_statahead_mark(parent, dentry);
609 struct lookup_intent *ll_convert_intent(struct open_intent *oit,
612 struct lookup_intent *it;
614 OBD_ALLOC(it, sizeof(*it));
616 return ERR_PTR(-ENOMEM);
618 if (lookup_flags & LOOKUP_OPEN) {
620 if (lookup_flags & LOOKUP_CREATE)
621 it->it_op |= IT_CREAT;
622 it->it_create_mode = (oit->create_mode & S_IALLUGO) | S_IFREG;
623 it->it_flags = oit->flags;
625 it->it_op = IT_GETATTR;
628 #ifndef HAVE_FILE_IN_STRUCT_INTENT
629 /* Since there is no way to pass our intent to ll_file_open,
630 * just check the file is there. Actual open will be done
632 if (it->it_op & IT_OPEN)
633 it->it_op = IT_LOOKUP;
639 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
640 struct nameidata *nd)
645 if (nd && !(nd->flags & (LOOKUP_CONTINUE|LOOKUP_PARENT))) {
646 struct lookup_intent *it;
648 #if defined(HAVE_FILE_IN_STRUCT_INTENT) && (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,17))
649 /* Did we came here from failed revalidate just to propagate
651 if (nd->flags & LOOKUP_OPEN)
652 if (IS_ERR(nd->intent.open.file))
653 RETURN((struct dentry *)nd->intent.open.file);
655 if (ll_d2d(dentry) && ll_d2d(dentry)->lld_it) {
656 it = ll_d2d(dentry)->lld_it;
657 ll_d2d(dentry)->lld_it = NULL;
659 if ((nd->flags & LOOKUP_CREATE ) && !(nd->flags & LOOKUP_OPEN)) {
660 /* We are sure this is new dentry, so we need to create
661 our private data and set the dentry ops */
662 ll_dops_init(dentry, 1, 1);
665 it = ll_convert_intent(&nd->intent.open, nd->flags);
667 RETURN((struct dentry *)it);
670 de = ll_lookup_it(parent, dentry, it, nd->flags);
673 if ((nd->flags & LOOKUP_OPEN) && !IS_ERR(dentry)) { /* Open */
674 if (dentry->d_inode &&
675 it_disposition(it, DISP_OPEN_OPEN)) { /* nocreate */
676 #ifdef HAVE_FILE_IN_STRUCT_INTENT
677 if (S_ISFIFO(dentry->d_inode->i_mode)) {
678 // We cannot call open here as it would
681 (struct ptlrpc_request *)
682 it->d.lustre.it_data);
684 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,17))
685 /* 2.6.1[456] have a bug in open_namei() that forgets to check
686 * nd->intent.open.file for error, so we need to return it as lookup's result
689 nd->intent.open.file->private_data = it;
690 filp =lookup_instantiate_filp(nd,dentry,
695 de = (struct dentry *) filp;
698 nd->intent.open.file->private_data = it;
699 (void)lookup_instantiate_filp(nd,dentry,
704 #else /* HAVE_FILE_IN_STRUCT_INTENT */
705 /* Release open handle as we have no way to
706 * pass it to ll_file_open */
707 ll_release_openhandle(dentry, it);
708 #endif /* HAVE_FILE_IN_STRUCT_INTENT */
709 } else if (it_disposition(it, DISP_OPEN_CREATE)) {
710 // XXX This can only reliably work on assumption
711 // that there are NO hashed negative dentries.
712 ll_d2d(dentry)->lld_it = it;
713 it = NULL; /* Will be freed in ll_create_nd */
714 /* We absolutely depend on ll_create_nd to be
715 * called to not leak this intent and possible
716 * data attached to it */
721 ll_intent_release(it);
722 OBD_FREE(it, sizeof(*it));
725 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 ll_set_lock_data(sbi->ll_md_exp, inode, it, NULL);
762 ptlrpc_req_finished(request);
767 * By the time this is called, we already have created the directory cache
768 * entry for the new file, but it is so far negative - it has no inode.
770 * We defer creating the OBD object(s) until open, to keep the intent and
771 * non-intent code paths similar, and also because we do not have the MDS
772 * inode number before calling ll_create_node() (which is needed for LOV),
773 * so we would need to do yet another RPC to the MDS to store the LOV EA
774 * data on the MDS. If needed, we would pass the PACKED lmm as data and
775 * lmm_size in datalen (the MDS still has code which will handle that).
777 * If the create succeeds, we fill in the inode information
778 * with d_instantiate().
780 static int ll_create_it(struct inode *dir, struct dentry *dentry, int mode,
781 struct lookup_intent *it)
787 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),intent=%s\n",
788 dentry->d_name.len, dentry->d_name.name, dir->i_ino,
789 dir->i_generation, dir, LL_IT2STR(it));
791 rc = it_open_error(DISP_OPEN_CREATE, it);
795 inode = ll_create_node(dir, dentry->d_name.name, dentry->d_name.len,
796 NULL, 0, mode, 0, it);
798 RETURN(PTR_ERR(inode));
800 d_instantiate(dentry, inode);
801 /* Negative dentry may be unhashed if parent does not have UPDATE lock,
802 * but some callers, e.g. do_coredump, expect dentry to be hashed after
803 * successful create. Hash it here. */
804 spin_lock(&dcache_lock);
805 if (d_unhashed(dentry))
806 d_rehash_cond(dentry, 0);
807 spin_unlock(&dcache_lock);
811 static void ll_update_times(struct ptlrpc_request *request,
814 struct mdt_body *body = req_capsule_server_get(&request->rq_pill,
818 if (body->valid & OBD_MD_FLMTIME &&
819 body->mtime > LTIME_S(inode->i_mtime)) {
820 CDEBUG(D_INODE, "setting ino %lu mtime from %lu to "LPU64"\n",
821 inode->i_ino, LTIME_S(inode->i_mtime), body->mtime);
822 LTIME_S(inode->i_mtime) = body->mtime;
824 if (body->valid & OBD_MD_FLCTIME &&
825 body->ctime > LTIME_S(inode->i_ctime))
826 LTIME_S(inode->i_ctime) = body->ctime;
829 static int ll_new_node(struct inode *dir, struct qstr *name,
830 const char *tgt, int mode, int rdev,
831 struct dentry *dchild, __u32 opc)
833 struct ptlrpc_request *request = NULL;
834 struct md_op_data *op_data;
835 struct inode *inode = NULL;
836 struct ll_sb_info *sbi = ll_i2sbi(dir);
841 if (unlikely(tgt != NULL))
842 tgt_len = strlen(tgt) + 1;
844 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
845 name->len, 0, opc, NULL);
847 GOTO(err_exit, err = PTR_ERR(op_data));
849 err = md_create(sbi->ll_md_exp, op_data, tgt, tgt_len, mode,
850 cfs_curproc_fsuid(), cfs_curproc_fsgid(),
851 cfs_curproc_cap_pack(), rdev, &request);
852 ll_finish_md_op_data(op_data);
856 ll_update_times(request, dir);
859 err = ll_prep_inode(&inode, request, dchild->d_sb);
864 d_instantiate(dchild, inode);
868 ptlrpc_req_finished(request);
873 static int ll_mknod_generic(struct inode *dir, struct qstr *name, int mode,
874 unsigned rdev, struct dentry *dchild)
879 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p) mode %o dev %x\n",
880 name->len, name->name, dir->i_ino, dir->i_generation, dir,
883 mode &= ~cfs_curproc_umask();
885 switch (mode & S_IFMT) {
887 mode |= S_IFREG; /* for mode = 0 case, fallthrough */
893 err = ll_new_node(dir, name, NULL, mode, rdev, dchild,
905 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
906 int mode, struct nameidata *nd)
908 struct lookup_intent *it = ll_d2d(dentry)->lld_it;
912 return ll_mknod_generic(dir, &dentry->d_name, mode, 0, dentry);
914 ll_d2d(dentry)->lld_it = NULL;
916 /* Was there an error? Propagate it! */
917 if (it->d.lustre.it_status) {
918 rc = it->d.lustre.it_status;
922 rc = ll_create_it(dir, dentry, mode, it);
923 #ifdef HAVE_FILE_IN_STRUCT_INTENT
924 if (nd && (nd->flags & LOOKUP_OPEN) && dentry->d_inode) { /* Open */
925 nd->intent.open.file->private_data = it;
926 lookup_instantiate_filp(nd, dentry, NULL);
929 ll_release_openhandle(dentry,it);
933 ll_intent_release(it);
934 OBD_FREE(it, sizeof(*it));
939 static int ll_symlink_generic(struct inode *dir, struct qstr *name,
940 const char *tgt, struct dentry *dchild)
945 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),target=%.*s\n",
946 name->len, name->name, dir->i_ino, dir->i_generation,
949 err = ll_new_node(dir, name, (char *)tgt, S_IFLNK | S_IRWXUGO,
950 0, dchild, LUSTRE_OPC_SYMLINK);
954 static int ll_link_generic(struct inode *src, struct inode *dir,
955 struct qstr *name, struct dentry *dchild)
957 struct ll_sb_info *sbi = ll_i2sbi(dir);
958 struct ptlrpc_request *request = NULL;
959 struct md_op_data *op_data;
964 "VFS Op: inode=%lu/%u(%p), dir=%lu/%u(%p), target=%.*s\n",
965 src->i_ino, src->i_generation, src, dir->i_ino,
966 dir->i_generation, dir, name->len, name->name);
968 op_data = ll_prep_md_op_data(NULL, src, dir, name->name, name->len,
969 0, LUSTRE_OPC_ANY, NULL);
971 RETURN(PTR_ERR(op_data));
973 err = md_link(sbi->ll_md_exp, op_data, &request);
974 ll_finish_md_op_data(op_data);
980 ll_update_times(request, dir);
983 ptlrpc_req_finished(request);
987 static int ll_mkdir_generic(struct inode *dir, struct qstr *name,
988 int mode, struct dentry *dchild)
994 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
995 name->len, name->name, dir->i_ino, dir->i_generation, dir);
997 mode = (mode & (S_IRWXUGO|S_ISVTX) & ~cfs_curproc_umask()) | S_IFDIR;
998 err = ll_new_node(dir, name, NULL, mode, 0, dchild, LUSTRE_OPC_MKDIR);
1003 /* Try to find the child dentry by its name.
1004 If found, put the result fid into @fid. */
1005 static void ll_get_child_fid(struct inode * dir, struct qstr *name,
1008 struct dentry *parent, *child;
1010 parent = list_entry(dir->i_dentry.next, struct dentry, d_alias);
1011 child = d_lookup(parent, name);
1014 *fid = *ll_inode2fid(child->d_inode);
1019 static int ll_rmdir_generic(struct inode *dir, struct dentry *dparent,
1020 struct dentry *dchild, struct qstr *name)
1022 struct ptlrpc_request *request = NULL;
1023 struct md_op_data *op_data;
1027 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
1028 name->len, name->name, dir->i_ino, dir->i_generation, dir);
1030 if (unlikely(ll_d_mountpoint(dparent, dchild, name)))
1033 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len,
1034 S_IFDIR, LUSTRE_OPC_ANY, NULL);
1035 if (IS_ERR(op_data))
1036 RETURN(PTR_ERR(op_data));
1038 ll_get_child_fid(dir, name, &op_data->op_fid3);
1039 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1040 ll_finish_md_op_data(op_data);
1042 ll_update_times(request, dir);
1043 ptlrpc_req_finished(request);
1047 int ll_objects_destroy(struct ptlrpc_request *request, struct inode *dir)
1049 struct mdt_body *body;
1050 struct lov_mds_md *eadata;
1051 struct lov_stripe_md *lsm = NULL;
1052 struct obd_trans_info oti = { 0 };
1054 struct obd_capa *oc = NULL;
1058 /* req is swabbed so this is safe */
1059 body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
1060 if (!(body->valid & OBD_MD_FLEASIZE))
1063 if (body->eadatasize == 0) {
1064 CERROR("OBD_MD_FLEASIZE set but eadatasize zero\n");
1065 GOTO(out, rc = -EPROTO);
1068 /* The MDS sent back the EA because we unlinked the last reference
1069 * to this file. Use this EA to unlink the objects on the OST.
1070 * It's opaque so we don't swab here; we leave it to obd_unpackmd() to
1071 * check it is complete and sensible. */
1072 eadata = req_capsule_server_sized_get(&request->rq_pill, &RMF_MDT_MD,
1074 LASSERT(eadata != NULL);
1076 rc = obd_unpackmd(ll_i2dtexp(dir), &lsm, eadata, body->eadatasize);
1078 CERROR("obd_unpackmd: %d\n", rc);
1081 LASSERT(rc >= sizeof(*lsm));
1085 GOTO(out_free_memmd, rc = -ENOMEM);
1087 oa->o_id = lsm->lsm_object_id;
1088 oa->o_seq = lsm->lsm_object_seq;
1089 oa->o_mode = body->mode & S_IFMT;
1090 oa->o_valid = OBD_MD_FLID | OBD_MD_FLTYPE | OBD_MD_FLGROUP;
1092 if (body->valid & OBD_MD_FLCOOKIE) {
1093 oa->o_valid |= OBD_MD_FLCOOKIE;
1094 oti.oti_logcookies =
1095 req_capsule_server_sized_get(&request->rq_pill,
1097 sizeof(struct llog_cookie) *
1098 lsm->lsm_stripe_count);
1099 if (oti.oti_logcookies == NULL) {
1100 oa->o_valid &= ~OBD_MD_FLCOOKIE;
1101 body->valid &= ~OBD_MD_FLCOOKIE;
1105 if (body->valid & OBD_MD_FLOSSCAPA) {
1106 rc = md_unpack_capa(ll_i2mdexp(dir), request, &RMF_CAPA2, &oc);
1108 GOTO(out_free_memmd, rc);
1111 rc = obd_destroy(ll_i2dtexp(dir), oa, lsm, &oti, ll_i2mdexp(dir), oc);
1115 CERROR("obd destroy objid "LPX64" error %d\n",
1116 lsm->lsm_object_id, rc);
1118 obd_free_memmd(ll_i2dtexp(dir), &lsm);
1123 /* ll_unlink_generic() doesn't update the inode with the new link count.
1124 * Instead, ll_ddelete() and ll_d_iput() will update it based upon if there
1125 * is any lock existing. They will recycle dentries and inodes based upon locks
1127 static int ll_unlink_generic(struct inode *dir, struct dentry *dparent,
1128 struct dentry *dchild, struct qstr *name)
1130 struct ptlrpc_request *request = NULL;
1131 struct md_op_data *op_data;
1134 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
1135 name->len, name->name, dir->i_ino, dir->i_generation, dir);
1138 * XXX: unlink bind mountpoint maybe call to here,
1139 * just check it as vfs_unlink does.
1141 if (unlikely(ll_d_mountpoint(dparent, dchild, name)))
1144 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
1145 name->len, 0, LUSTRE_OPC_ANY, NULL);
1146 if (IS_ERR(op_data))
1147 RETURN(PTR_ERR(op_data));
1149 ll_get_child_fid(dir, name, &op_data->op_fid3);
1150 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1151 ll_finish_md_op_data(op_data);
1155 ll_update_times(request, dir);
1157 rc = ll_objects_destroy(request, dir);
1159 ptlrpc_req_finished(request);
1163 static int ll_rename_generic(struct inode *src, struct dentry *src_dparent,
1164 struct dentry *src_dchild, struct qstr *src_name,
1165 struct inode *tgt, struct dentry *tgt_dparent,
1166 struct dentry *tgt_dchild, struct qstr *tgt_name)
1168 struct ptlrpc_request *request = NULL;
1169 struct ll_sb_info *sbi = ll_i2sbi(src);
1170 struct md_op_data *op_data;
1173 CDEBUG(D_VFSTRACE,"VFS Op:oldname=%.*s,src_dir=%lu/%u(%p),newname=%.*s,"
1174 "tgt_dir=%lu/%u(%p)\n", src_name->len, src_name->name,
1175 src->i_ino, src->i_generation, src, tgt_name->len,
1176 tgt_name->name, tgt->i_ino, tgt->i_generation, tgt);
1178 if (unlikely(ll_d_mountpoint(src_dparent, src_dchild, src_name) ||
1179 ll_d_mountpoint(tgt_dparent, tgt_dchild, tgt_name)))
1182 op_data = ll_prep_md_op_data(NULL, src, tgt, NULL, 0, 0,
1183 LUSTRE_OPC_ANY, NULL);
1184 if (IS_ERR(op_data))
1185 RETURN(PTR_ERR(op_data));
1187 ll_get_child_fid(src, src_name, &op_data->op_fid3);
1188 ll_get_child_fid(tgt, tgt_name, &op_data->op_fid4);
1189 err = md_rename(sbi->ll_md_exp, op_data,
1190 src_name->name, src_name->len,
1191 tgt_name->name, tgt_name->len, &request);
1192 ll_finish_md_op_data(op_data);
1194 ll_update_times(request, src);
1195 ll_update_times(request, tgt);
1196 err = ll_objects_destroy(request, src);
1199 ptlrpc_req_finished(request);
1204 static int ll_mknod(struct inode *dir, struct dentry *dchild, int mode,
1207 return ll_mknod_generic(dir, &dchild->d_name, mode,
1208 old_encode_dev(rdev), dchild);
1211 static int ll_unlink(struct inode * dir, struct dentry *dentry)
1213 return ll_unlink_generic(dir, NULL, dentry, &dentry->d_name);
1215 static int ll_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1217 return ll_mkdir_generic(dir, &dentry->d_name, mode, dentry);
1219 static int ll_rmdir(struct inode *dir, struct dentry *dentry)
1221 return ll_rmdir_generic(dir, NULL, dentry, &dentry->d_name);
1223 static int ll_symlink(struct inode *dir, struct dentry *dentry,
1224 const char *oldname)
1226 return ll_symlink_generic(dir, &dentry->d_name, oldname, dentry);
1228 static int ll_link(struct dentry *old_dentry, struct inode *dir,
1229 struct dentry *new_dentry)
1231 return ll_link_generic(old_dentry->d_inode, dir, &new_dentry->d_name,
1234 static int ll_rename(struct inode *old_dir, struct dentry *old_dentry,
1235 struct inode *new_dir, struct dentry *new_dentry)
1238 err = ll_rename_generic(old_dir, NULL,
1239 old_dentry, &old_dentry->d_name,
1240 new_dir, NULL, new_dentry,
1241 &new_dentry->d_name);
1243 #ifndef HAVE_FS_RENAME_DOES_D_MOVE
1244 if (!S_ISDIR(old_dentry->d_inode->i_mode))
1246 d_move(old_dentry, new_dentry);
1251 struct inode_operations ll_dir_inode_operations = {
1253 .lookup = ll_lookup_nd,
1254 .create = ll_create_nd,
1255 /* We need all these non-raw things for NFSD, to not patch it. */
1256 .unlink = ll_unlink,
1259 .symlink = ll_symlink,
1261 .rename = ll_rename,
1262 .setattr = ll_setattr,
1263 .getattr = ll_getattr,
1264 .permission = ll_inode_permission,
1265 .setxattr = ll_setxattr,
1266 .getxattr = ll_getxattr,
1267 .listxattr = ll_listxattr,
1268 .removexattr = ll_removexattr,
1271 struct inode_operations ll_special_inode_operations = {
1272 .setattr = ll_setattr,
1273 .getattr = ll_getattr,
1274 .permission = ll_inode_permission,
1275 .setxattr = ll_setxattr,
1276 .getxattr = ll_getxattr,
1277 .listxattr = ll_listxattr,
1278 .removexattr = ll_removexattr,