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>
48 #define DEBUG_SUBSYSTEM S_LLITE
50 #include <obd_support.h>
51 #include <lustre_fid.h>
52 #include <lustre_lite.h>
53 #include <lustre_dlm.h>
54 #include <lustre_ver.h>
55 #include <lustre_mdc.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;
122 * Get an inode by inode number (already instantiated by the intent lookup).
123 * Returns inode or NULL
125 struct inode *ll_iget(struct super_block *sb, ino_t hash,
126 struct lustre_md *md)
132 inode = iget5_locked(sb, hash, ll_test_inode, ll_set_inode, md);
135 if (inode->i_state & I_NEW) {
138 ll_read_inode2(inode, md);
139 rc = cl_inode_init(inode, md);
142 make_bad_inode(inode);
143 unlock_new_inode(inode);
147 unlock_new_inode(inode);
148 } else if (!(inode->i_state & (I_FREEING | I_CLEAR)))
149 ll_update_inode(inode, md);
150 CDEBUG(D_VFSTRACE, "got inode: %p for "DFID"\n",
151 inode, PFID(&md->body->fid1));
156 static void ll_drop_negative_dentry(struct inode *dir)
158 struct dentry *dentry, *tmp_alias, *tmp_subdir;
160 cfs_spin_lock(&ll_lookup_lock);
161 spin_lock(&dcache_lock);
163 list_for_each_entry_safe(dentry, tmp_alias,
164 &dir->i_dentry,d_alias) {
165 if (!list_empty(&dentry->d_subdirs)) {
166 struct dentry *child;
167 list_for_each_entry_safe(child, tmp_subdir,
170 /* XXX Print some debug here? */
172 /* Negative dentry. If we were
173 dropping dcache lock, go
174 throught the list again */
175 if (ll_drop_dentry(child))
180 spin_unlock(&dcache_lock);
181 cfs_spin_unlock(&ll_lookup_lock);
185 int ll_md_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
186 void *data, int flag)
189 struct lustre_handle lockh;
193 case LDLM_CB_BLOCKING:
194 ldlm_lock2handle(lock, &lockh);
195 rc = ldlm_cli_cancel(&lockh);
197 CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
201 case LDLM_CB_CANCELING: {
202 struct inode *inode = ll_inode_from_lock(lock);
203 struct ll_inode_info *lli;
204 __u64 bits = lock->l_policy_data.l_inodebits.bits;
206 ldlm_mode_t mode = lock->l_req_mode;
208 /* Invalidate all dentries associated with this inode */
212 LASSERT(lock->l_flags & LDLM_FL_CANCELING);
213 /* For OPEN locks we differentiate between lock modes - CR, CW. PR - bug 22891 */
214 if ((bits & MDS_INODELOCK_LOOKUP) &&
215 ll_have_md_lock(inode, MDS_INODELOCK_LOOKUP, LCK_MINMODE))
216 bits &= ~MDS_INODELOCK_LOOKUP;
217 if ((bits & MDS_INODELOCK_UPDATE) &&
218 ll_have_md_lock(inode, MDS_INODELOCK_UPDATE, LCK_MINMODE))
219 bits &= ~MDS_INODELOCK_UPDATE;
220 if ((bits & MDS_INODELOCK_OPEN) &&
221 ll_have_md_lock(inode, MDS_INODELOCK_OPEN, mode))
222 bits &= ~MDS_INODELOCK_OPEN;
224 fid = ll_inode2fid(inode);
225 if (lock->l_resource->lr_name.name[0] != fid_seq(fid) ||
226 lock->l_resource->lr_name.name[1] != fid_oid(fid) ||
227 lock->l_resource->lr_name.name[2] != fid_ver(fid)) {
228 LDLM_ERROR(lock, "data mismatch with object "
229 DFID" (%p)", PFID(fid), inode);
232 if (bits & MDS_INODELOCK_OPEN) {
234 switch (lock->l_req_mode) {
245 CERROR("Unexpected lock mode for OPEN lock "
246 "%d, inode %ld\n", lock->l_req_mode,
249 ll_md_real_close(inode, flags);
252 lli = ll_i2info(inode);
253 if (bits & MDS_INODELOCK_UPDATE)
254 lli->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
256 if (S_ISDIR(inode->i_mode) &&
257 (bits & MDS_INODELOCK_UPDATE)) {
258 CDEBUG(D_INODE, "invalidating inode %lu\n",
260 truncate_inode_pages(inode->i_mapping, 0);
261 ll_drop_negative_dentry(inode);
264 if (inode->i_sb->s_root &&
265 inode != inode->i_sb->s_root->d_inode &&
266 (bits & MDS_INODELOCK_LOOKUP))
267 ll_unhash_aliases(inode);
278 __u32 ll_i2suppgid(struct inode *i)
280 if (cfs_curproc_is_in_groups(i->i_gid))
281 return (__u32)i->i_gid;
286 /* Pack the required supplementary groups into the supplied groups array.
287 * If we don't need to use the groups from the target inode(s) then we
288 * instead pack one or more groups from the user's supplementary group
289 * array in case it might be useful. Not needed if doing an MDS-side upcall. */
290 void ll_i2gids(__u32 *suppgids, struct inode *i1, struct inode *i2)
297 LASSERT(suppgids != NULL);
299 suppgids[0] = ll_i2suppgid(i1);
302 suppgids[1] = ll_i2suppgid(i2);
307 for (i = 0; i < current_ngroups; i++) {
308 if (suppgids[0] == -1) {
309 if (current_groups[i] != suppgids[1])
310 suppgids[0] = current_groups[i];
313 if (suppgids[1] == -1) {
314 if (current_groups[i] != suppgids[0])
315 suppgids[1] = current_groups[i];
323 static void ll_d_add(struct dentry *de, struct inode *inode)
325 CDEBUG(D_DENTRY, "adding inode %p to dentry %p\n", inode, de);
327 if (!list_empty(&de->d_alias)) {
328 spin_unlock(&dcache_lock);
329 CERROR("dentry %.*s %p alias next %p, prev %p\n",
330 de->d_name.len, de->d_name.name, de,
331 de->d_alias.next, de->d_alias.prev);
335 list_add(&de->d_alias, &inode->i_dentry);
339 if (!d_unhashed(de)) {
340 spin_unlock(&dcache_lock);
341 CERROR("dentry %.*s %p hash next %p\n",
342 de->d_name.len, de->d_name.name, de, de->d_hash.next);
345 d_rehash_cond(de, 0);
348 /* Search "inode"'s alias list for a dentry that has the same name and parent
349 * as de. If found, return it. If not found, return de.
350 * Lustre can't use d_add_unique because don't unhash aliases for directory
351 * in ll_revalidate_it. After revaliadate inode will be have hashed aliases
352 * and it triggers BUG_ON in d_instantiate_unique (bug #10954).
354 static struct dentry *ll_find_alias(struct inode *inode, struct dentry *de)
356 struct list_head *tmp;
357 struct dentry *dentry;
358 struct dentry *last_discon = NULL;
360 cfs_spin_lock(&ll_lookup_lock);
361 spin_lock(&dcache_lock);
362 list_for_each(tmp, &inode->i_dentry) {
363 dentry = list_entry(tmp, struct dentry, d_alias);
365 /* We are called here with 'de' already on the aliases list. */
366 if (unlikely(dentry == de)) {
371 if (dentry->d_flags & DCACHE_DISCONNECTED) {
372 /* LASSERT(last_discon == NULL); LU-405, bz 20055 */
373 last_discon = dentry;
377 if (dentry->d_parent != de->d_parent)
380 if (dentry->d_name.hash != de->d_name.hash)
383 if (dentry->d_name.len != de->d_name.len)
386 if (memcmp(dentry->d_name.name, de->d_name.name,
387 de->d_name.len) != 0)
393 unlock_dentry(dentry);
394 ll_dops_init(dentry, 0, 1);
395 d_rehash_cond(dentry, 0); /* avoid taking dcache_lock inside */
396 spin_unlock(&dcache_lock);
397 cfs_spin_unlock(&ll_lookup_lock);
399 CDEBUG(D_DENTRY, "alias dentry %.*s (%p) parent %p inode %p "
400 "refc %d\n", de->d_name.len, de->d_name.name, de,
401 de->d_parent, de->d_inode, atomic_read(&de->d_count));
406 CDEBUG(D_DENTRY, "Reuse disconnected dentry %p inode %p "
407 "refc %d\n", last_discon, last_discon->d_inode,
408 atomic_read(&last_discon->d_count));
409 dget_locked(last_discon);
410 lock_dentry(last_discon);
411 last_discon->d_flags |= DCACHE_LUSTRE_INVALID;
412 unlock_dentry(last_discon);
413 spin_unlock(&dcache_lock);
414 cfs_spin_unlock(&ll_lookup_lock);
415 ll_dops_init(last_discon, 1, 1);
417 d_move(last_discon, de);
422 de->d_flags |= DCACHE_LUSTRE_INVALID;
426 spin_unlock(&dcache_lock);
427 cfs_spin_unlock(&ll_lookup_lock);
432 int ll_lookup_it_finish(struct ptlrpc_request *request,
433 struct lookup_intent *it, void *data)
435 struct it_cb_data *icbd = data;
436 struct dentry **de = icbd->icbd_childp;
437 struct inode *parent = icbd->icbd_parent;
438 struct ll_sb_info *sbi = ll_i2sbi(parent);
439 struct inode *inode = NULL;
443 /* NB 1 request reference will be taken away by ll_intent_lock()
445 if (!it_disposition(it, DISP_LOOKUP_NEG)) {
446 struct dentry *save = *de;
449 rc = ll_prep_inode(&inode, request, (*de)->d_sb);
453 CDEBUG(D_DLMTRACE, "setting l_data to inode %p (%lu/%u)\n",
454 inode, inode->i_ino, inode->i_generation);
455 md_set_lock_data(sbi->ll_md_exp,
456 &it->d.lustre.it_lock_handle, inode, &bits);
458 /* We used to query real size from OSTs here, but actually
459 this is not needed. For stat() calls size would be updated
460 from subsequent do_revalidate()->ll_inode_revalidate_it() in
462 vfs_getattr_it->ll_getattr()->ll_inode_revalidate_it() in 2.6
463 Everybody else who needs correct file size would call
464 cl_glimpse_size or some equivalent themselves anyway.
465 Also see bug 7198. */
466 ll_dops_init(*de, 1, 1);
467 *de = ll_find_alias(inode, *de);
469 struct ll_dentry_data *lld = ll_d2d(*de);
471 /* just make sure the ll_dentry_data is ready */
472 if (unlikely(lld == NULL))
473 ll_dops_init(*de, 1, 1);
475 /* we have lookup look - unhide dentry */
476 if (bits & MDS_INODELOCK_LOOKUP) {
478 (*de)->d_flags &= ~DCACHE_LUSTRE_INVALID;
482 ll_dops_init(*de, 1, 1);
483 /* Check that parent has UPDATE lock. If there is none, we
484 cannot afford to hash this dentry (done by ll_d_add) as it
485 might get picked up later when UPDATE lock will appear */
486 if (ll_have_md_lock(parent, MDS_INODELOCK_UPDATE, LCK_MINMODE)) {
487 spin_lock(&dcache_lock);
489 spin_unlock(&dcache_lock);
491 /* negative lookup - and don't have update lock to
494 (*de)->d_flags |= DCACHE_LUSTRE_INVALID;
497 (*de)->d_inode = NULL;
498 /* We do not want to hash the dentry if don`t have a
499 * lock, but if this dentry is later used in d_move,
500 * we'd hit uninitialised list head d_hash, so we just
501 * do this to init d_hash field but leave dentry
502 * unhashed. (bug 10796). */
511 static struct dentry *ll_lookup_it(struct inode *parent, struct dentry *dentry,
512 struct lookup_intent *it, int lookup_flags)
514 struct lookup_intent lookup_it = { .it_op = IT_LOOKUP };
515 struct dentry *save = dentry, *retval;
516 struct ptlrpc_request *req = NULL;
517 struct md_op_data *op_data;
518 struct it_cb_data icbd;
523 if (dentry->d_name.len > ll_i2sbi(parent)->ll_namelen)
524 RETURN(ERR_PTR(-ENAMETOOLONG));
526 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),intent=%s\n",
527 dentry->d_name.len, dentry->d_name.name, parent->i_ino,
528 parent->i_generation, parent, LL_IT2STR(it));
530 if (d_mountpoint(dentry))
531 CERROR("Tell Peter, lookup on mtpt, it %s\n", LL_IT2STR(it));
533 ll_frob_intent(&it, &lookup_it);
535 /* As do_lookup is called before follow_mount, root dentry may be left
536 * not valid, revalidate it here. */
537 if (parent->i_sb->s_root && (parent->i_sb->s_root->d_inode == parent) &&
538 (it->it_op & (IT_OPEN | IT_CREAT))) {
539 rc = ll_inode_revalidate_it(parent->i_sb->s_root, it,
540 MDS_INODELOCK_LOOKUP);
545 if (it->it_op == IT_GETATTR) {
546 first = ll_statahead_enter(parent, &dentry, 1);
548 ll_statahead_exit(parent, dentry, first);
550 RETURN(retval = dentry);
554 icbd.icbd_childp = &dentry;
555 icbd.icbd_parent = parent;
557 if (it->it_op & IT_CREAT ||
558 (it->it_op & IT_OPEN && it->it_create_mode & O_CREAT))
559 opc = LUSTRE_OPC_CREATE;
561 opc = LUSTRE_OPC_ANY;
563 op_data = ll_prep_md_op_data(NULL, parent, NULL, dentry->d_name.name,
564 dentry->d_name.len, lookup_flags, opc,
567 RETURN((void *)op_data);
569 it->it_create_mode &= ~cfs_curproc_umask();
571 rc = md_intent_lock(ll_i2mdexp(parent), op_data, NULL, 0, it,
572 lookup_flags, &req, ll_md_blocking_ast, 0);
573 ll_finish_md_op_data(op_data);
575 GOTO(out, retval = ERR_PTR(rc));
577 rc = ll_lookup_it_finish(req, it, &icbd);
579 ll_intent_release(it);
580 GOTO(out, retval = ERR_PTR(rc));
583 if (first == -EEXIST)
584 ll_statahead_mark(parent, dentry);
586 if ((it->it_op & IT_OPEN) && dentry->d_inode &&
587 !S_ISREG(dentry->d_inode->i_mode) &&
588 !S_ISDIR(dentry->d_inode->i_mode)) {
589 ll_release_openhandle(dentry, it);
591 ll_lookup_finish_locks(it, dentry);
594 GOTO(out, retval = NULL);
596 GOTO(out, retval = dentry);
599 ptlrpc_req_finished(req);
603 struct lookup_intent *ll_convert_intent(struct open_intent *oit,
606 struct lookup_intent *it;
608 OBD_ALLOC(it, sizeof(*it));
610 return ERR_PTR(-ENOMEM);
612 if (lookup_flags & LOOKUP_OPEN) {
614 if (lookup_flags & LOOKUP_CREATE)
615 it->it_op |= IT_CREAT;
616 it->it_create_mode = (oit->create_mode & S_IALLUGO) | S_IFREG;
617 it->it_flags = oit->flags;
619 it->it_op = IT_GETATTR;
622 #ifndef HAVE_FILE_IN_STRUCT_INTENT
623 /* Since there is no way to pass our intent to ll_file_open,
624 * just check the file is there. Actual open will be done
626 if (it->it_op & IT_OPEN)
627 it->it_op = IT_LOOKUP;
633 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
634 struct nameidata *nd)
639 if (nd && !(nd->flags & (LOOKUP_CONTINUE|LOOKUP_PARENT))) {
640 struct lookup_intent *it;
642 #if defined(HAVE_FILE_IN_STRUCT_INTENT) && (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,17))
643 /* Did we came here from failed revalidate just to propagate
645 if (nd->flags & LOOKUP_OPEN)
646 if (IS_ERR(nd->intent.open.file))
647 RETURN((struct dentry *)nd->intent.open.file);
649 if (ll_d2d(dentry) && ll_d2d(dentry)->lld_it) {
650 it = ll_d2d(dentry)->lld_it;
651 ll_d2d(dentry)->lld_it = NULL;
653 if ((nd->flags & LOOKUP_CREATE ) && !(nd->flags & LOOKUP_OPEN)) {
654 /* We are sure this is new dentry, so we need to create
655 our private data and set the dentry ops */
656 ll_dops_init(dentry, 1, 1);
659 it = ll_convert_intent(&nd->intent.open, nd->flags);
661 RETURN((struct dentry *)it);
664 de = ll_lookup_it(parent, dentry, it, nd->flags);
667 if ((nd->flags & LOOKUP_OPEN) && !IS_ERR(dentry)) { /* Open */
668 if (dentry->d_inode &&
669 it_disposition(it, DISP_OPEN_OPEN)) { /* nocreate */
670 #ifdef HAVE_FILE_IN_STRUCT_INTENT
671 if (S_ISFIFO(dentry->d_inode->i_mode)) {
672 // We cannot call open here as it would
675 (struct ptlrpc_request *)
676 it->d.lustre.it_data);
678 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,17))
679 /* 2.6.1[456] have a bug in open_namei() that forgets to check
680 * nd->intent.open.file for error, so we need to return it as lookup's result
683 nd->intent.open.file->private_data = it;
684 filp =lookup_instantiate_filp(nd,dentry,
689 de = (struct dentry *) filp;
692 nd->intent.open.file->private_data = it;
693 (void)lookup_instantiate_filp(nd,dentry,
698 #else /* HAVE_FILE_IN_STRUCT_INTENT */
699 /* Release open handle as we have no way to
700 * pass it to ll_file_open */
701 ll_release_openhandle(dentry, it);
702 #endif /* HAVE_FILE_IN_STRUCT_INTENT */
703 } else if (it_disposition(it, DISP_OPEN_CREATE)) {
704 // XXX This can only reliably work on assumption
705 // that there are NO hashed negative dentries.
706 ll_d2d(dentry)->lld_it = it;
707 it = NULL; /* Will be freed in ll_create_nd */
708 /* We absolutely depend on ll_create_nd to be
709 * called to not leak this intent and possible
710 * data attached to it */
715 ll_intent_release(it);
716 OBD_FREE(it, sizeof(*it));
719 de = ll_lookup_it(parent, dentry, NULL, 0);
725 /* We depend on "mode" being set with the proper file type/umask by now */
726 static struct inode *ll_create_node(struct inode *dir, const char *name,
727 int namelen, const void *data, int datalen,
728 int mode, __u64 extra,
729 struct lookup_intent *it)
731 struct inode *inode = NULL;
732 struct ptlrpc_request *request = NULL;
733 struct ll_sb_info *sbi = ll_i2sbi(dir);
737 LASSERT(it && it->d.lustre.it_disposition);
739 LASSERT(it_disposition(it, DISP_ENQ_CREATE_REF));
740 request = it->d.lustre.it_data;
741 it_clear_disposition(it, DISP_ENQ_CREATE_REF);
742 rc = ll_prep_inode(&inode, request, dir->i_sb);
744 GOTO(out, inode = ERR_PTR(rc));
746 LASSERT(list_empty(&inode->i_dentry));
748 /* We asked for a lock on the directory, but were granted a
749 * lock on the inode. Since we finally have an inode pointer,
750 * stuff it in the lock. */
751 CDEBUG(D_DLMTRACE, "setting l_ast_data to inode %p (%lu/%u)\n",
752 inode, inode->i_ino, inode->i_generation);
753 md_set_lock_data(sbi->ll_md_exp,
754 &it->d.lustre.it_lock_handle, inode, NULL);
757 ptlrpc_req_finished(request);
762 * By the time this is called, we already have created the directory cache
763 * entry for the new file, but it is so far negative - it has no inode.
765 * We defer creating the OBD object(s) until open, to keep the intent and
766 * non-intent code paths similar, and also because we do not have the MDS
767 * inode number before calling ll_create_node() (which is needed for LOV),
768 * so we would need to do yet another RPC to the MDS to store the LOV EA
769 * data on the MDS. If needed, we would pass the PACKED lmm as data and
770 * lmm_size in datalen (the MDS still has code which will handle that).
772 * If the create succeeds, we fill in the inode information
773 * with d_instantiate().
775 static int ll_create_it(struct inode *dir, struct dentry *dentry, int mode,
776 struct lookup_intent *it)
782 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),intent=%s\n",
783 dentry->d_name.len, dentry->d_name.name, dir->i_ino,
784 dir->i_generation, dir, LL_IT2STR(it));
786 rc = it_open_error(DISP_OPEN_CREATE, it);
790 inode = ll_create_node(dir, dentry->d_name.name, dentry->d_name.len,
791 NULL, 0, mode, 0, it);
793 RETURN(PTR_ERR(inode));
795 d_instantiate(dentry, inode);
796 /* Negative dentry may be unhashed if parent does not have UPDATE lock,
797 * but some callers, e.g. do_coredump, expect dentry to be hashed after
798 * successful create. Hash it here. */
799 spin_lock(&dcache_lock);
800 if (d_unhashed(dentry))
801 d_rehash_cond(dentry, 0);
802 spin_unlock(&dcache_lock);
806 static void ll_update_times(struct ptlrpc_request *request,
809 struct mdt_body *body = req_capsule_server_get(&request->rq_pill,
813 if (body->valid & OBD_MD_FLMTIME &&
814 body->mtime > LTIME_S(inode->i_mtime)) {
815 CDEBUG(D_INODE, "setting ino %lu mtime from %lu to "LPU64"\n",
816 inode->i_ino, LTIME_S(inode->i_mtime), body->mtime);
817 LTIME_S(inode->i_mtime) = body->mtime;
819 if (body->valid & OBD_MD_FLCTIME &&
820 body->ctime > LTIME_S(inode->i_ctime))
821 LTIME_S(inode->i_ctime) = body->ctime;
824 static int ll_new_node(struct inode *dir, struct qstr *name,
825 const char *tgt, int mode, int rdev,
826 struct dentry *dchild, __u32 opc)
828 struct ptlrpc_request *request = NULL;
829 struct md_op_data *op_data;
830 struct inode *inode = NULL;
831 struct ll_sb_info *sbi = ll_i2sbi(dir);
836 if (unlikely(tgt != NULL))
837 tgt_len = strlen(tgt) + 1;
839 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
840 name->len, 0, opc, NULL);
842 GOTO(err_exit, err = PTR_ERR(op_data));
844 err = md_create(sbi->ll_md_exp, op_data, tgt, tgt_len, mode,
845 cfs_curproc_fsuid(), cfs_curproc_fsgid(),
846 cfs_curproc_cap_pack(), rdev, &request);
847 ll_finish_md_op_data(op_data);
851 ll_update_times(request, dir);
854 err = ll_prep_inode(&inode, request, dchild->d_sb);
859 d_instantiate(dchild, inode);
863 ptlrpc_req_finished(request);
868 static int ll_mknod_generic(struct inode *dir, struct qstr *name, int mode,
869 unsigned rdev, struct dentry *dchild)
874 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p) mode %o dev %x\n",
875 name->len, name->name, dir->i_ino, dir->i_generation, dir,
878 mode &= ~cfs_curproc_umask();
880 switch (mode & S_IFMT) {
882 mode |= S_IFREG; /* for mode = 0 case, fallthrough */
888 err = ll_new_node(dir, name, NULL, mode, rdev, dchild,
900 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
901 int mode, struct nameidata *nd)
903 struct lookup_intent *it = ll_d2d(dentry)->lld_it;
907 return ll_mknod_generic(dir, &dentry->d_name, mode, 0, dentry);
909 ll_d2d(dentry)->lld_it = NULL;
911 /* Was there an error? Propagate it! */
912 if (it->d.lustre.it_status) {
913 rc = it->d.lustre.it_status;
917 rc = ll_create_it(dir, dentry, mode, it);
918 #ifdef HAVE_FILE_IN_STRUCT_INTENT
919 if (nd && (nd->flags & LOOKUP_OPEN) && dentry->d_inode) { /* Open */
920 nd->intent.open.file->private_data = it;
921 lookup_instantiate_filp(nd, dentry, NULL);
924 ll_release_openhandle(dentry,it);
928 ll_intent_release(it);
929 OBD_FREE(it, sizeof(*it));
934 static int ll_symlink_generic(struct inode *dir, struct qstr *name,
935 const char *tgt, struct dentry *dchild)
940 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),target=%.*s\n",
941 name->len, name->name, dir->i_ino, dir->i_generation,
944 err = ll_new_node(dir, name, (char *)tgt, S_IFLNK | S_IRWXUGO,
945 0, dchild, LUSTRE_OPC_SYMLINK);
949 static int ll_link_generic(struct inode *src, struct inode *dir,
950 struct qstr *name, struct dentry *dchild)
952 struct ll_sb_info *sbi = ll_i2sbi(dir);
953 struct ptlrpc_request *request = NULL;
954 struct md_op_data *op_data;
959 "VFS Op: inode=%lu/%u(%p), dir=%lu/%u(%p), target=%.*s\n",
960 src->i_ino, src->i_generation, src, dir->i_ino,
961 dir->i_generation, dir, name->len, name->name);
963 op_data = ll_prep_md_op_data(NULL, src, dir, name->name, name->len,
964 0, LUSTRE_OPC_ANY, NULL);
966 RETURN(PTR_ERR(op_data));
968 err = md_link(sbi->ll_md_exp, op_data, &request);
969 ll_finish_md_op_data(op_data);
975 ll_update_times(request, dir);
978 ptlrpc_req_finished(request);
982 static int ll_mkdir_generic(struct inode *dir, struct qstr *name,
983 int mode, struct dentry *dchild)
989 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
990 name->len, name->name, dir->i_ino, dir->i_generation, dir);
992 mode = (mode & (S_IRWXUGO|S_ISVTX) & ~cfs_curproc_umask()) | S_IFDIR;
993 err = ll_new_node(dir, name, NULL, mode, 0, dchild, LUSTRE_OPC_MKDIR);
998 /* Try to find the child dentry by its name.
999 If found, put the result fid into @fid. */
1000 static void ll_get_child_fid(struct inode * dir, struct qstr *name,
1003 struct dentry *parent, *child;
1005 parent = list_entry(dir->i_dentry.next, struct dentry, d_alias);
1006 child = d_lookup(parent, name);
1009 *fid = *ll_inode2fid(child->d_inode);
1014 static int ll_rmdir_generic(struct inode *dir, struct dentry *dparent,
1015 struct dentry *dchild, struct qstr *name)
1017 struct ptlrpc_request *request = NULL;
1018 struct md_op_data *op_data;
1022 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
1023 name->len, name->name, dir->i_ino, dir->i_generation, dir);
1025 if (unlikely(ll_d_mountpoint(dparent, dchild, name)))
1028 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len,
1029 S_IFDIR, LUSTRE_OPC_ANY, NULL);
1030 if (IS_ERR(op_data))
1031 RETURN(PTR_ERR(op_data));
1033 ll_get_child_fid(dir, name, &op_data->op_fid3);
1034 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1035 ll_finish_md_op_data(op_data);
1037 ll_update_times(request, dir);
1038 ptlrpc_req_finished(request);
1042 int ll_objects_destroy(struct ptlrpc_request *request, struct inode *dir)
1044 struct mdt_body *body;
1045 struct lov_mds_md *eadata;
1046 struct lov_stripe_md *lsm = NULL;
1047 struct obd_trans_info oti = { 0 };
1049 struct obd_capa *oc = NULL;
1053 /* req is swabbed so this is safe */
1054 body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
1055 if (!(body->valid & OBD_MD_FLEASIZE))
1058 if (body->eadatasize == 0) {
1059 CERROR("OBD_MD_FLEASIZE set but eadatasize zero\n");
1060 GOTO(out, rc = -EPROTO);
1063 /* The MDS sent back the EA because we unlinked the last reference
1064 * to this file. Use this EA to unlink the objects on the OST.
1065 * It's opaque so we don't swab here; we leave it to obd_unpackmd() to
1066 * check it is complete and sensible. */
1067 eadata = req_capsule_server_sized_get(&request->rq_pill, &RMF_MDT_MD,
1069 LASSERT(eadata != NULL);
1071 rc = obd_unpackmd(ll_i2dtexp(dir), &lsm, eadata, body->eadatasize);
1073 CERROR("obd_unpackmd: %d\n", rc);
1076 LASSERT(rc >= sizeof(*lsm));
1080 GOTO(out_free_memmd, rc = -ENOMEM);
1082 oa->o_id = lsm->lsm_object_id;
1083 oa->o_seq = lsm->lsm_object_seq;
1084 oa->o_mode = body->mode & S_IFMT;
1085 oa->o_valid = OBD_MD_FLID | OBD_MD_FLTYPE | OBD_MD_FLGROUP;
1087 if (body->valid & OBD_MD_FLCOOKIE) {
1088 oa->o_valid |= OBD_MD_FLCOOKIE;
1089 oti.oti_logcookies =
1090 req_capsule_server_sized_get(&request->rq_pill,
1092 sizeof(struct llog_cookie) *
1093 lsm->lsm_stripe_count);
1094 if (oti.oti_logcookies == NULL) {
1095 oa->o_valid &= ~OBD_MD_FLCOOKIE;
1096 body->valid &= ~OBD_MD_FLCOOKIE;
1100 if (body->valid & OBD_MD_FLOSSCAPA) {
1101 rc = md_unpack_capa(ll_i2mdexp(dir), request, &RMF_CAPA2, &oc);
1103 GOTO(out_free_memmd, rc);
1106 rc = obd_destroy(ll_i2dtexp(dir), oa, lsm, &oti, ll_i2mdexp(dir), oc);
1110 CERROR("obd destroy objid "LPX64" error %d\n",
1111 lsm->lsm_object_id, rc);
1113 obd_free_memmd(ll_i2dtexp(dir), &lsm);
1118 /* ll_unlink_generic() doesn't update the inode with the new link count.
1119 * Instead, ll_ddelete() and ll_d_iput() will update it based upon if there
1120 * is any lock existing. They will recycle dentries and inodes based upon locks
1122 static int ll_unlink_generic(struct inode *dir, struct dentry *dparent,
1123 struct dentry *dchild, struct qstr *name)
1125 struct ptlrpc_request *request = NULL;
1126 struct md_op_data *op_data;
1129 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
1130 name->len, name->name, dir->i_ino, dir->i_generation, dir);
1133 * XXX: unlink bind mountpoint maybe call to here,
1134 * just check it as vfs_unlink does.
1136 if (unlikely(ll_d_mountpoint(dparent, dchild, name)))
1139 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
1140 name->len, 0, LUSTRE_OPC_ANY, NULL);
1141 if (IS_ERR(op_data))
1142 RETURN(PTR_ERR(op_data));
1144 ll_get_child_fid(dir, name, &op_data->op_fid3);
1145 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1146 ll_finish_md_op_data(op_data);
1150 ll_update_times(request, dir);
1152 rc = ll_objects_destroy(request, dir);
1154 ptlrpc_req_finished(request);
1158 static int ll_rename_generic(struct inode *src, struct dentry *src_dparent,
1159 struct dentry *src_dchild, struct qstr *src_name,
1160 struct inode *tgt, struct dentry *tgt_dparent,
1161 struct dentry *tgt_dchild, struct qstr *tgt_name)
1163 struct ptlrpc_request *request = NULL;
1164 struct ll_sb_info *sbi = ll_i2sbi(src);
1165 struct md_op_data *op_data;
1168 CDEBUG(D_VFSTRACE,"VFS Op:oldname=%.*s,src_dir=%lu/%u(%p),newname=%.*s,"
1169 "tgt_dir=%lu/%u(%p)\n", src_name->len, src_name->name,
1170 src->i_ino, src->i_generation, src, tgt_name->len,
1171 tgt_name->name, tgt->i_ino, tgt->i_generation, tgt);
1173 if (unlikely(ll_d_mountpoint(src_dparent, src_dchild, src_name) ||
1174 ll_d_mountpoint(tgt_dparent, tgt_dchild, tgt_name)))
1177 op_data = ll_prep_md_op_data(NULL, src, tgt, NULL, 0, 0,
1178 LUSTRE_OPC_ANY, NULL);
1179 if (IS_ERR(op_data))
1180 RETURN(PTR_ERR(op_data));
1182 ll_get_child_fid(src, src_name, &op_data->op_fid3);
1183 ll_get_child_fid(tgt, tgt_name, &op_data->op_fid4);
1184 err = md_rename(sbi->ll_md_exp, op_data,
1185 src_name->name, src_name->len,
1186 tgt_name->name, tgt_name->len, &request);
1187 ll_finish_md_op_data(op_data);
1189 ll_update_times(request, src);
1190 ll_update_times(request, tgt);
1191 err = ll_objects_destroy(request, src);
1194 ptlrpc_req_finished(request);
1199 static int ll_mknod(struct inode *dir, struct dentry *dchild, int mode,
1202 return ll_mknod_generic(dir, &dchild->d_name, mode,
1203 old_encode_dev(rdev), dchild);
1206 static int ll_unlink(struct inode * dir, struct dentry *dentry)
1208 return ll_unlink_generic(dir, NULL, dentry, &dentry->d_name);
1210 static int ll_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1212 return ll_mkdir_generic(dir, &dentry->d_name, mode, dentry);
1214 static int ll_rmdir(struct inode *dir, struct dentry *dentry)
1216 return ll_rmdir_generic(dir, NULL, dentry, &dentry->d_name);
1218 static int ll_symlink(struct inode *dir, struct dentry *dentry,
1219 const char *oldname)
1221 return ll_symlink_generic(dir, &dentry->d_name, oldname, dentry);
1223 static int ll_link(struct dentry *old_dentry, struct inode *dir,
1224 struct dentry *new_dentry)
1226 return ll_link_generic(old_dentry->d_inode, dir, &new_dentry->d_name,
1229 static int ll_rename(struct inode *old_dir, struct dentry *old_dentry,
1230 struct inode *new_dir, struct dentry *new_dentry)
1233 err = ll_rename_generic(old_dir, NULL,
1234 old_dentry, &old_dentry->d_name,
1235 new_dir, NULL, new_dentry,
1236 &new_dentry->d_name);
1238 #ifndef HAVE_FS_RENAME_DOES_D_MOVE
1239 if (!S_ISDIR(old_dentry->d_inode->i_mode))
1241 d_move(old_dentry, new_dentry);
1246 struct inode_operations ll_dir_inode_operations = {
1248 .lookup = ll_lookup_nd,
1249 .create = ll_create_nd,
1250 /* We need all these non-raw things for NFSD, to not patch it. */
1251 .unlink = ll_unlink,
1254 .symlink = ll_symlink,
1256 .rename = ll_rename,
1257 .setattr = ll_setattr,
1258 .getattr = ll_getattr,
1259 .permission = ll_inode_permission,
1260 .setxattr = ll_setxattr,
1261 .getxattr = ll_getxattr,
1262 .listxattr = ll_listxattr,
1263 .removexattr = ll_removexattr,
1266 struct inode_operations ll_special_inode_operations = {
1267 .setattr = ll_setattr,
1268 .getattr = ll_getattr,
1269 .permission = ll_inode_permission,
1270 .setxattr = ll_setxattr,
1271 .getxattr = ll_getxattr,
1272 .listxattr = ll_listxattr,
1273 .removexattr = ll_removexattr,