1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
6 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
8 * This program is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License version 2 only,
10 * as published by the Free Software Foundation.
12 * This program is distributed in the hope that it will be useful, but
13 * WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * General Public License version 2 for more details (a copy is included
16 * in the LICENSE file that accompanied this code).
18 * You should have received a copy of the GNU General Public License
19 * version 2 along with this program; If not, see
20 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
22 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
23 * CA 95054 USA or visit www.sun.com if you need additional information or
29 * Copyright 2008 Sun Microsystems, Inc. All rights reserved
30 * Use is subject to license terms.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
38 #include <linux/sched.h>
40 #include <linux/smp_lock.h>
41 #include <linux/quotaops.h>
42 #include <linux/highmem.h>
43 #include <linux/pagemap.h>
45 #define DEBUG_SUBSYSTEM S_LLITE
47 #include <obd_support.h>
48 #include <lustre_fid.h>
49 #include <lustre_lite.h>
50 #include <lustre_dlm.h>
51 #include <lustre_ver.h>
52 #include <lustre_mdc.h>
53 #include "llite_internal.h"
56 * Check if we have something mounted at the named dchild.
57 * In such a case there would always be dentry present.
59 static int ll_d_mountpoint(struct dentry *dparent, struct dentry *dchild,
64 if (unlikely(dchild)) {
65 mounted = d_mountpoint(dchild);
67 dchild = d_lookup(dparent, name);
69 mounted = d_mountpoint(dchild);
76 int ll_unlock(__u32 mode, struct lustre_handle *lockh)
80 ldlm_lock_decref(lockh, mode);
86 /* called from iget5_locked->find_inode() under inode_lock spinlock */
87 static int ll_test_inode(struct inode *inode, void *opaque)
89 struct ll_inode_info *lli = ll_i2info(inode);
90 struct lustre_md *md = opaque;
92 if (unlikely(!(md->body->valid & OBD_MD_FLID))) {
93 CERROR("MDS body missing FID\n");
97 if (!lu_fid_eq(&lli->lli_fid, &md->body->fid1))
103 static int ll_set_inode(struct inode *inode, void *opaque)
105 struct ll_inode_info *lli = ll_i2info(inode);
106 struct mdt_body *body = ((struct lustre_md *)opaque)->body;
108 if (unlikely(!(body->valid & OBD_MD_FLID))) {
109 CERROR("MDS body missing FID\n");
113 lli->lli_fid = body->fid1;
119 * Get an inode by inode number (already instantiated by the intent lookup).
120 * Returns inode or NULL
122 struct inode *ll_iget(struct super_block *sb, ino_t hash,
123 struct lustre_md *md)
129 inode = iget5_locked(sb, hash, ll_test_inode, ll_set_inode, md);
132 if (inode->i_state & I_NEW) {
135 ll_read_inode2(inode, md);
136 rc = cl_inode_init(inode, md);
139 make_bad_inode(inode);
140 unlock_new_inode(inode);
144 unlock_new_inode(inode);
145 } else if (!(inode->i_state & (I_FREEING | I_CLEAR)))
146 ll_update_inode(inode, md);
147 CDEBUG(D_VFSTRACE, "got inode: %p for "DFID"\n",
148 inode, PFID(&md->body->fid1));
153 static void ll_drop_negative_dentry(struct inode *dir)
155 struct dentry *dentry, *tmp_alias, *tmp_subdir;
157 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 __u64 bits = lock->l_policy_data.l_inodebits.bits;
203 /* Invalidate all dentries associated with this inode */
207 LASSERT(lock->l_flags & LDLM_FL_CANCELING);
208 if ((bits & MDS_INODELOCK_LOOKUP) &&
209 ll_have_md_lock(inode, MDS_INODELOCK_LOOKUP))
210 bits &= ~MDS_INODELOCK_LOOKUP;
211 if ((bits & MDS_INODELOCK_UPDATE) &&
212 ll_have_md_lock(inode, MDS_INODELOCK_UPDATE))
213 bits &= ~MDS_INODELOCK_UPDATE;
214 if ((bits & MDS_INODELOCK_OPEN) &&
215 ll_have_md_lock(inode, MDS_INODELOCK_OPEN))
216 bits &= ~MDS_INODELOCK_OPEN;
218 fid = ll_inode2fid(inode);
219 if (lock->l_resource->lr_name.name[0] != fid_seq(fid) ||
220 lock->l_resource->lr_name.name[1] != fid_oid(fid) ||
221 lock->l_resource->lr_name.name[2] != fid_ver(fid)) {
222 LDLM_ERROR(lock, "data mismatch with object "
223 DFID" (%p)", PFID(fid), inode);
226 if (bits & MDS_INODELOCK_OPEN) {
228 switch (lock->l_req_mode) {
239 CERROR("Unexpected lock mode for OPEN lock "
240 "%d, inode %ld\n", lock->l_req_mode,
243 ll_md_real_close(inode, flags);
246 if (bits & MDS_INODELOCK_UPDATE)
247 ll_i2info(inode)->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
249 if (S_ISDIR(inode->i_mode) &&
250 (bits & MDS_INODELOCK_UPDATE)) {
251 CDEBUG(D_INODE, "invalidating inode %lu\n",
253 truncate_inode_pages(inode->i_mapping, 0);
254 ll_drop_negative_dentry(inode);
257 if (inode->i_sb->s_root &&
258 inode != inode->i_sb->s_root->d_inode &&
259 (bits & MDS_INODELOCK_LOOKUP))
260 ll_unhash_aliases(inode);
271 __u32 ll_i2suppgid(struct inode *i)
273 if (cfs_curproc_is_in_groups(i->i_gid))
274 return (__u32)i->i_gid;
279 /* Pack the required supplementary groups into the supplied groups array.
280 * If we don't need to use the groups from the target inode(s) then we
281 * instead pack one or more groups from the user's supplementary group
282 * array in case it might be useful. Not needed if doing an MDS-side upcall. */
283 void ll_i2gids(__u32 *suppgids, struct inode *i1, struct inode *i2)
290 LASSERT(suppgids != NULL);
292 suppgids[0] = ll_i2suppgid(i1);
295 suppgids[1] = ll_i2suppgid(i2);
300 for (i = 0; i < current_ngroups; i++) {
301 if (suppgids[0] == -1) {
302 if (current_groups[i] != suppgids[1])
303 suppgids[0] = current_groups[i];
306 if (suppgids[1] == -1) {
307 if (current_groups[i] != suppgids[0])
308 suppgids[1] = current_groups[i];
316 static void ll_d_add(struct dentry *de, struct inode *inode)
318 CDEBUG(D_DENTRY, "adding inode %p to dentry %p\n", inode, de);
320 if (!list_empty(&de->d_alias)) {
321 spin_unlock(&dcache_lock);
322 CERROR("dentry %.*s %p alias next %p, prev %p\n",
323 de->d_name.len, de->d_name.name, de,
324 de->d_alias.next, de->d_alias.prev);
328 list_add(&de->d_alias, &inode->i_dentry);
332 if (!d_unhashed(de)) {
333 spin_unlock(&dcache_lock);
334 CERROR("dentry %.*s %p hash next %p\n",
335 de->d_name.len, de->d_name.name, de, de->d_hash.next);
338 d_rehash_cond(de, 0);
341 /* Search "inode"'s alias list for a dentry that has the same name and parent
342 * as de. If found, return it. If not found, return de.
343 * Lustre can't use d_add_unique because don't unhash aliases for directory
344 * in ll_revalidate_it. After revaliadate inode will be have hashed aliases
345 * and it triggers BUG_ON in d_instantiate_unique (bug #10954).
347 static struct dentry *ll_find_alias(struct inode *inode, struct dentry *de)
349 struct list_head *tmp;
350 struct dentry *dentry;
351 struct dentry *last_discon = NULL;
353 cfs_spin_lock(&ll_lookup_lock);
354 spin_lock(&dcache_lock);
355 list_for_each(tmp, &inode->i_dentry) {
356 dentry = list_entry(tmp, struct dentry, d_alias);
358 /* We are called here with 'de' already on the aliases list. */
359 if (unlikely(dentry == de)) {
364 if (dentry->d_flags & DCACHE_DISCONNECTED) {
365 LASSERT(last_discon == NULL);
366 last_discon = dentry;
370 if (dentry->d_parent != de->d_parent)
373 if (dentry->d_name.hash != de->d_name.hash)
376 if (dentry->d_name.len != de->d_name.len)
379 if (memcmp(dentry->d_name.name, de->d_name.name,
380 de->d_name.len) != 0)
386 unlock_dentry(dentry);
387 ll_dops_init(dentry, 0);
388 d_rehash_cond(dentry, 0); /* avoid taking dcache_lock inside */
389 spin_unlock(&dcache_lock);
390 cfs_spin_unlock(&ll_lookup_lock);
392 CDEBUG(D_DENTRY, "alias dentry %.*s (%p) parent %p inode %p "
393 "refc %d\n", de->d_name.len, de->d_name.name, de,
394 de->d_parent, de->d_inode, atomic_read(&de->d_count));
399 CDEBUG(D_DENTRY, "Reuse disconnected dentry %p inode %p "
400 "refc %d\n", last_discon, last_discon->d_inode,
401 atomic_read(&last_discon->d_count));
402 dget_locked(last_discon);
403 lock_dentry(last_discon);
404 last_discon->d_flags |= DCACHE_LUSTRE_INVALID;
405 unlock_dentry(last_discon);
406 spin_unlock(&dcache_lock);
407 cfs_spin_unlock(&ll_lookup_lock);
408 ll_dops_init(last_discon, 1);
410 d_move(last_discon, de);
415 de->d_flags |= DCACHE_LUSTRE_INVALID;
419 spin_unlock(&dcache_lock);
420 cfs_spin_unlock(&ll_lookup_lock);
425 void ll_lookup_it_alias(struct dentry **de, struct inode *inode, __u32 bits)
427 struct dentry *save = *de;
430 ll_dops_init(*de, 1);
431 *de = ll_find_alias(inode, *de);
433 struct ll_dentry_data *lld = ll_d2d(*de);
435 /* just make sure the ll_dentry_data is ready */
436 if (unlikely(lld == NULL)) {
439 if (likely(lld != NULL))
440 lld->lld_sa_generation = 0;
443 /* we have lookup look - unhide dentry */
444 if (bits & MDS_INODELOCK_LOOKUP) {
446 (*de)->d_flags &= ~DCACHE_LUSTRE_INVALID;
452 int ll_lookup_it_finish(struct ptlrpc_request *request,
453 struct lookup_intent *it, void *data,
454 struct inode **alias)
456 struct it_cb_data *icbd = data;
457 struct dentry **de = icbd->icbd_childp;
458 struct inode *parent = icbd->icbd_parent;
459 struct ll_sb_info *sbi = ll_i2sbi(parent);
460 struct inode *inode = NULL;
464 /* NB 1 request reference will be taken away by ll_intent_lock()
466 if (!it_disposition(it, DISP_LOOKUP_NEG)) {
469 rc = ll_prep_inode(&inode, request, (*de)->d_sb);
473 CDEBUG(D_DLMTRACE, "setting l_data to inode %p (%lu/%u)\n",
474 inode, inode->i_ino, inode->i_generation);
475 md_set_lock_data(sbi->ll_md_exp,
476 &it->d.lustre.it_lock_handle, inode, &bits);
483 /* We used to query real size from OSTs here, but actually
484 this is not needed. For stat() calls size would be updated
485 from subsequent do_revalidate()->ll_inode_revalidate_it() in
487 vfs_getattr_it->ll_getattr()->ll_inode_revalidate_it() in 2.6
488 Everybody else who needs correct file size would call
489 cl_glimpse_size or some equivalent themselves anyway.
490 Also see bug 7198. */
491 ll_lookup_it_alias(de, inode, bits);
493 ll_dops_init(*de, 1);
494 /* Check that parent has UPDATE lock. If there is none, we
495 cannot afford to hash this dentry (done by ll_d_add) as it
496 might get picked up later when UPDATE lock will appear */
497 if (ll_have_md_lock(parent, MDS_INODELOCK_UPDATE)) {
498 spin_lock(&dcache_lock);
500 spin_unlock(&dcache_lock);
502 /* negative lookup - and don't have update lock to
505 (*de)->d_flags |= DCACHE_LUSTRE_INVALID;
508 (*de)->d_inode = NULL;
509 /* We do not want to hash the dentry if don`t have a
510 * lock, but if this dentry is later used in d_move,
511 * we'd hit uninitialised list head d_hash, so we just
512 * do this to init d_hash field but leave dentry
513 * unhashed. (bug 10796). */
522 static struct dentry *ll_lookup_it(struct inode *parent, struct dentry *dentry,
523 struct lookup_intent *it, int lookup_flags)
525 struct lookup_intent lookup_it = { .it_op = IT_LOOKUP };
526 struct dentry *save = dentry, *retval;
527 struct ptlrpc_request *req = NULL;
528 struct md_op_data *op_data;
529 struct it_cb_data icbd;
534 if (dentry->d_name.len > ll_i2sbi(parent)->ll_namelen)
535 RETURN(ERR_PTR(-ENAMETOOLONG));
537 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),intent=%s\n",
538 dentry->d_name.len, dentry->d_name.name, parent->i_ino,
539 parent->i_generation, parent, LL_IT2STR(it));
541 if (d_mountpoint(dentry))
542 CERROR("Tell Peter, lookup on mtpt, it %s\n", LL_IT2STR(it));
544 ll_frob_intent(&it, &lookup_it);
546 /* As do_lookup is called before follow_mount, root dentry may be left
547 * not valid, revalidate it here. */
548 if (parent->i_sb->s_root && (parent->i_sb->s_root->d_inode == parent) &&
549 (it->it_op & (IT_OPEN | IT_CREAT))) {
550 rc = ll_inode_revalidate_it(parent->i_sb->s_root, it);
555 if (it->it_op == IT_GETATTR) {
556 first = ll_statahead_enter(parent, &dentry, 1);
558 ll_statahead_exit(parent, dentry, first);
560 RETURN(retval = dentry);
564 icbd.icbd_childp = &dentry;
565 icbd.icbd_parent = parent;
567 if (it->it_op & IT_CREAT ||
568 (it->it_op & IT_OPEN && it->it_create_mode & O_CREAT))
569 opc = LUSTRE_OPC_CREATE;
571 opc = LUSTRE_OPC_ANY;
573 op_data = ll_prep_md_op_data(NULL, parent, NULL, dentry->d_name.name,
574 dentry->d_name.len, lookup_flags, opc,
577 RETURN((void *)op_data);
579 it->it_create_mode &= ~current->fs->umask;
581 rc = md_intent_lock(ll_i2mdexp(parent), op_data, NULL, 0, it,
582 lookup_flags, &req, ll_md_blocking_ast, 0);
583 ll_finish_md_op_data(op_data);
585 GOTO(out, retval = ERR_PTR(rc));
587 rc = ll_lookup_it_finish(req, it, &icbd, NULL);
589 ll_intent_release(it);
590 GOTO(out, retval = ERR_PTR(rc));
593 if (first == -EEXIST)
594 ll_statahead_mark(parent, dentry);
596 if ((it->it_op & IT_OPEN) && dentry->d_inode &&
597 !S_ISREG(dentry->d_inode->i_mode) &&
598 !S_ISDIR(dentry->d_inode->i_mode)) {
599 ll_release_openhandle(dentry, it);
601 ll_lookup_finish_locks(it, dentry);
604 GOTO(out, retval = NULL);
606 GOTO(out, retval = dentry);
609 ptlrpc_req_finished(req);
613 #ifdef HAVE_VFS_INTENT_PATCHES
614 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
615 struct nameidata *nd)
620 if (nd && nd->flags & LOOKUP_LAST && !(nd->flags & LOOKUP_LINK_NOTLAST))
621 de = ll_lookup_it(parent, dentry, &nd->intent, nd->flags);
623 de = ll_lookup_it(parent, dentry, NULL, 0);
628 struct lookup_intent *ll_convert_intent(struct open_intent *oit,
631 struct lookup_intent *it;
633 OBD_ALLOC(it, sizeof(*it));
635 return ERR_PTR(-ENOMEM);
637 if (lookup_flags & LOOKUP_OPEN) {
639 if (lookup_flags & LOOKUP_CREATE)
640 it->it_op |= IT_CREAT;
641 it->it_create_mode = oit->create_mode;
642 it->it_flags = oit->flags;
644 it->it_op = IT_GETATTR;
647 #ifndef HAVE_FILE_IN_STRUCT_INTENT
648 /* Since there is no way to pass our intent to ll_file_open,
649 * just check the file is there. Actual open will be done
651 if (it->it_op & IT_OPEN)
652 it->it_op = IT_LOOKUP;
658 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
659 struct nameidata *nd)
664 if (nd && !(nd->flags & (LOOKUP_CONTINUE|LOOKUP_PARENT))) {
665 struct lookup_intent *it;
667 #if defined(HAVE_FILE_IN_STRUCT_INTENT) && (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,17))
668 /* Did we came here from failed revalidate just to propagate
670 if (nd->flags & LOOKUP_OPEN)
671 if (IS_ERR(nd->intent.open.file))
672 RETURN((struct dentry *)nd->intent.open.file);
675 if (ll_d2d(dentry) && ll_d2d(dentry)->lld_it) {
676 it = ll_d2d(dentry)->lld_it;
677 ll_d2d(dentry)->lld_it = NULL;
679 if ((nd->flags & LOOKUP_CREATE ) && !(nd->flags & LOOKUP_OPEN)) {
680 /* We are sure this is new dentry, so we need to create
681 our private data and set the dentry ops */
682 ll_dops_init(dentry, 1);
685 it = ll_convert_intent(&nd->intent.open, nd->flags);
687 RETURN((struct dentry *)it);
690 de = ll_lookup_it(parent, dentry, it, nd->flags);
693 if ((nd->flags & LOOKUP_OPEN) && !IS_ERR(dentry)) { /* Open */
694 if (dentry->d_inode &&
695 it_disposition(it, DISP_OPEN_OPEN)) { /* nocreate */
696 #ifdef HAVE_FILE_IN_STRUCT_INTENT
697 if (S_ISFIFO(dentry->d_inode->i_mode)) {
698 // We cannot call open here as it would
701 (struct ptlrpc_request *)
702 it->d.lustre.it_data);
704 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,17))
705 /* 2.6.1[456] have a bug in open_namei() that forgets to check
706 * nd->intent.open.file for error, so we need to return it as lookup's result
709 nd->intent.open.file->private_data = it;
710 filp =lookup_instantiate_filp(nd,dentry,
715 de = (struct dentry *) filp;
718 nd->intent.open.file->private_data = it;
719 (void)lookup_instantiate_filp(nd,dentry,
724 #else /* HAVE_FILE_IN_STRUCT_INTENT */
725 /* Release open handle as we have no way to
726 * pass it to ll_file_open */
727 ll_release_openhandle(dentry, it);
728 #endif /* HAVE_FILE_IN_STRUCT_INTENT */
729 } else if (it_disposition(it, DISP_OPEN_CREATE)) {
730 // XXX This can only reliably work on assumption
731 // that there are NO hashed negative dentries.
732 ll_d2d(dentry)->lld_it = it;
733 it = NULL; /* Will be freed in ll_create_nd */
734 /* We absolutely depend on ll_create_nd to be
735 * called to not leak this intent and possible
736 * data attached to it */
741 ll_intent_release(it);
742 OBD_FREE(it, sizeof(*it));
745 de = ll_lookup_it(parent, dentry, NULL, 0);
752 /* We depend on "mode" being set with the proper file type/umask by now */
753 static struct inode *ll_create_node(struct inode *dir, const char *name,
754 int namelen, const void *data, int datalen,
755 int mode, __u64 extra,
756 struct lookup_intent *it)
758 struct inode *inode = NULL;
759 struct ptlrpc_request *request = NULL;
760 struct ll_sb_info *sbi = ll_i2sbi(dir);
764 LASSERT(it && it->d.lustre.it_disposition);
766 LASSERT(it_disposition(it, DISP_ENQ_CREATE_REF));
767 request = it->d.lustre.it_data;
768 it_clear_disposition(it, DISP_ENQ_CREATE_REF);
769 rc = ll_prep_inode(&inode, request, dir->i_sb);
771 GOTO(out, inode = ERR_PTR(rc));
773 LASSERT(list_empty(&inode->i_dentry));
775 /* We asked for a lock on the directory, but were granted a
776 * lock on the inode. Since we finally have an inode pointer,
777 * stuff it in the lock. */
778 CDEBUG(D_DLMTRACE, "setting l_ast_data to inode %p (%lu/%u)\n",
779 inode, inode->i_ino, inode->i_generation);
780 md_set_lock_data(sbi->ll_md_exp,
781 &it->d.lustre.it_lock_handle, inode, NULL);
784 ptlrpc_req_finished(request);
789 * By the time this is called, we already have created the directory cache
790 * entry for the new file, but it is so far negative - it has no inode.
792 * We defer creating the OBD object(s) until open, to keep the intent and
793 * non-intent code paths similar, and also because we do not have the MDS
794 * inode number before calling ll_create_node() (which is needed for LOV),
795 * so we would need to do yet another RPC to the MDS to store the LOV EA
796 * data on the MDS. If needed, we would pass the PACKED lmm as data and
797 * lmm_size in datalen (the MDS still has code which will handle that).
799 * If the create succeeds, we fill in the inode information
800 * with d_instantiate().
802 static int ll_create_it(struct inode *dir, struct dentry *dentry, int mode,
803 struct lookup_intent *it)
809 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),intent=%s\n",
810 dentry->d_name.len, dentry->d_name.name, dir->i_ino,
811 dir->i_generation, dir, LL_IT2STR(it));
813 rc = it_open_error(DISP_OPEN_CREATE, it);
817 inode = ll_create_node(dir, dentry->d_name.name, dentry->d_name.len,
818 NULL, 0, mode, 0, it);
820 RETURN(PTR_ERR(inode));
823 d_instantiate(dentry, inode);
824 /* Negative dentry may be unhashed if parent does not have UPDATE lock,
825 * but some callers, e.g. do_coredump, expect dentry to be hashed after
826 * successful create. Hash it here. */
827 spin_lock(&dcache_lock);
828 if (d_unhashed(dentry))
829 d_rehash_cond(dentry, 0);
830 spin_unlock(&dcache_lock);
834 static void ll_update_times(struct ptlrpc_request *request,
837 struct mdt_body *body = req_capsule_server_get(&request->rq_pill,
841 /* mtime is always updated with ctime, but can be set in past.
842 As write and utime(2) may happen within 1 second, and utime's
843 mtime has a priority over write's one, so take mtime from mds
844 for the same ctimes. */
845 if (body->valid & OBD_MD_FLCTIME &&
846 body->ctime >= LTIME_S(inode->i_ctime)) {
847 LTIME_S(inode->i_ctime) = body->ctime;
849 if (body->valid & OBD_MD_FLMTIME) {
850 CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
851 "to "LPU64"\n", inode->i_ino,
852 LTIME_S(inode->i_mtime), body->mtime);
853 LTIME_S(inode->i_mtime) = body->mtime;
858 static int ll_new_node(struct inode *dir, struct qstr *name,
859 const char *tgt, int mode, int rdev,
860 struct dentry *dchild, __u32 opc)
862 struct ptlrpc_request *request = NULL;
863 struct md_op_data *op_data;
864 struct inode *inode = NULL;
865 struct ll_sb_info *sbi = ll_i2sbi(dir);
870 if (unlikely(tgt != NULL))
871 tgt_len = strlen(tgt) + 1;
873 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
874 name->len, 0, opc, NULL);
876 GOTO(err_exit, err = PTR_ERR(op_data));
878 err = md_create(sbi->ll_md_exp, op_data, tgt, tgt_len, mode,
879 cfs_curproc_fsuid(), cfs_curproc_fsgid(),
880 cfs_curproc_cap_pack(), rdev, &request);
881 ll_finish_md_op_data(op_data);
885 ll_update_times(request, dir);
888 err = ll_prep_inode(&inode, request, dchild->d_sb);
893 d_instantiate(dchild, inode);
897 ptlrpc_req_finished(request);
902 static int ll_mknod_generic(struct inode *dir, struct qstr *name, int mode,
903 unsigned rdev, struct dentry *dchild)
908 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p) mode %o dev %x\n",
909 name->len, name->name, dir->i_ino, dir->i_generation, dir,
912 mode &= ~current->fs->umask;
914 switch (mode & S_IFMT) {
916 mode |= S_IFREG; /* for mode = 0 case, fallthrough */
922 err = ll_new_node(dir, name, NULL, mode, rdev, dchild,
934 #ifndef HAVE_VFS_INTENT_PATCHES
935 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
936 int mode, struct nameidata *nd)
938 struct lookup_intent *it = ll_d2d(dentry)->lld_it;
942 return ll_mknod_generic(dir, &dentry->d_name, mode, 0, dentry);
944 ll_d2d(dentry)->lld_it = NULL;
946 /* Was there an error? Propagate it! */
947 if (it->d.lustre.it_status) {
948 rc = it->d.lustre.it_status;
952 rc = ll_create_it(dir, dentry, mode, it);
953 #ifdef HAVE_FILE_IN_STRUCT_INTENT
954 if (nd && (nd->flags & LOOKUP_OPEN) && dentry->d_inode) { /* Open */
955 nd->intent.open.file->private_data = it;
956 lookup_instantiate_filp(nd, dentry, NULL);
959 ll_release_openhandle(dentry,it);
963 ll_intent_release(it);
964 OBD_FREE(it, sizeof(*it));
969 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
970 int mode, struct nameidata *nd)
972 if (!nd || !nd->intent.d.lustre.it_disposition)
973 /* No saved request? Just mknod the file */
974 return ll_mknod_generic(dir, &dentry->d_name, mode, 0, dentry);
976 return ll_create_it(dir, dentry, mode, &nd->intent);
980 static int ll_symlink_generic(struct inode *dir, struct qstr *name,
981 const char *tgt, struct dentry *dchild)
986 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),target=%.*s\n",
987 name->len, name->name, dir->i_ino, dir->i_generation,
990 err = ll_new_node(dir, name, (char *)tgt, S_IFLNK | S_IRWXUGO,
991 0, dchild, LUSTRE_OPC_SYMLINK);
995 static int ll_link_generic(struct inode *src, struct inode *dir,
996 struct qstr *name, struct dentry *dchild)
998 struct ll_sb_info *sbi = ll_i2sbi(dir);
999 struct ptlrpc_request *request = NULL;
1000 struct md_op_data *op_data;
1005 "VFS Op: inode=%lu/%u(%p), dir=%lu/%u(%p), target=%.*s\n",
1006 src->i_ino, src->i_generation, src, dir->i_ino,
1007 dir->i_generation, dir, name->len, name->name);
1009 op_data = ll_prep_md_op_data(NULL, src, dir, name->name, name->len,
1010 0, LUSTRE_OPC_ANY, NULL);
1011 if (IS_ERR(op_data))
1012 RETURN(PTR_ERR(op_data));
1014 err = md_link(sbi->ll_md_exp, op_data, &request);
1015 ll_finish_md_op_data(op_data);
1021 ll_update_times(request, dir);
1024 ptlrpc_req_finished(request);
1028 static int ll_mkdir_generic(struct inode *dir, struct qstr *name,
1029 int mode, struct dentry *dchild)
1035 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
1036 name->len, name->name, dir->i_ino, dir->i_generation, dir);
1038 mode = (mode & (S_IRWXUGO|S_ISVTX) & ~current->fs->umask) | S_IFDIR;
1039 err = ll_new_node(dir, name, NULL, mode, 0, dchild, LUSTRE_OPC_MKDIR);
1044 /* Try to find the child dentry by its name.
1045 If found, put the result fid into @fid. */
1046 static void ll_get_child_fid(struct inode * dir, struct qstr *name,
1049 struct dentry *parent, *child;
1051 parent = list_entry(dir->i_dentry.next, struct dentry, d_alias);
1052 child = d_lookup(parent, name);
1055 *fid = *ll_inode2fid(child->d_inode);
1060 static int ll_rmdir_generic(struct inode *dir, struct dentry *dparent,
1061 struct dentry *dchild, struct qstr *name)
1063 struct ptlrpc_request *request = NULL;
1064 struct md_op_data *op_data;
1068 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
1069 name->len, name->name, dir->i_ino, dir->i_generation, dir);
1071 if (unlikely(ll_d_mountpoint(dparent, dchild, name)))
1074 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len,
1075 S_IFDIR, LUSTRE_OPC_ANY, NULL);
1076 if (IS_ERR(op_data))
1077 RETURN(PTR_ERR(op_data));
1079 ll_get_child_fid(dir, name, &op_data->op_fid3);
1080 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1081 ll_finish_md_op_data(op_data);
1083 ll_update_times(request, dir);
1084 ptlrpc_req_finished(request);
1088 int ll_objects_destroy(struct ptlrpc_request *request, struct inode *dir)
1090 struct mdt_body *body;
1091 struct lov_mds_md *eadata;
1092 struct lov_stripe_md *lsm = NULL;
1093 struct obd_trans_info oti = { 0 };
1095 struct obd_capa *oc = NULL;
1099 /* req is swabbed so this is safe */
1100 body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
1101 if (!(body->valid & OBD_MD_FLEASIZE))
1104 if (body->eadatasize == 0) {
1105 CERROR("OBD_MD_FLEASIZE set but eadatasize zero\n");
1106 GOTO(out, rc = -EPROTO);
1109 /* The MDS sent back the EA because we unlinked the last reference
1110 * to this file. Use this EA to unlink the objects on the OST.
1111 * It's opaque so we don't swab here; we leave it to obd_unpackmd() to
1112 * check it is complete and sensible. */
1113 eadata = req_capsule_server_sized_get(&request->rq_pill, &RMF_MDT_MD,
1115 LASSERT(eadata != NULL);
1117 rc = obd_unpackmd(ll_i2dtexp(dir), &lsm, eadata, body->eadatasize);
1119 CERROR("obd_unpackmd: %d\n", rc);
1122 LASSERT(rc >= sizeof(*lsm));
1126 GOTO(out_free_memmd, rc = -ENOMEM);
1128 oa->o_id = lsm->lsm_object_id;
1129 oa->o_gr = lsm->lsm_object_gr;
1130 oa->o_mode = body->mode & S_IFMT;
1131 oa->o_valid = OBD_MD_FLID | OBD_MD_FLTYPE | OBD_MD_FLGROUP;
1133 if (body->valid & OBD_MD_FLCOOKIE) {
1134 oa->o_valid |= OBD_MD_FLCOOKIE;
1135 oti.oti_logcookies =
1136 req_capsule_server_sized_get(&request->rq_pill,
1138 sizeof(struct llog_cookie) *
1139 lsm->lsm_stripe_count);
1140 if (oti.oti_logcookies == NULL) {
1141 oa->o_valid &= ~OBD_MD_FLCOOKIE;
1142 body->valid &= ~OBD_MD_FLCOOKIE;
1146 if (body->valid & OBD_MD_FLOSSCAPA) {
1147 rc = md_unpack_capa(ll_i2mdexp(dir), request, &RMF_CAPA2, &oc);
1149 GOTO(out_free_memmd, rc);
1152 rc = obd_destroy(ll_i2dtexp(dir), oa, lsm, &oti, ll_i2mdexp(dir), oc);
1156 CERROR("obd destroy objid "LPX64" error %d\n",
1157 lsm->lsm_object_id, rc);
1159 obd_free_memmd(ll_i2dtexp(dir), &lsm);
1164 static int ll_unlink_generic(struct inode *dir, struct dentry *dparent,
1165 struct dentry *dchild, struct qstr *name)
1167 struct ptlrpc_request *request = NULL;
1168 struct md_op_data *op_data;
1171 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
1172 name->len, name->name, dir->i_ino, dir->i_generation, dir);
1175 * XXX: unlink bind mountpoint maybe call to here,
1176 * just check it as vfs_unlink does.
1178 if (unlikely(ll_d_mountpoint(dparent, dchild, name)))
1181 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
1182 name->len, 0, LUSTRE_OPC_ANY, NULL);
1183 if (IS_ERR(op_data))
1184 RETURN(PTR_ERR(op_data));
1186 ll_get_child_fid(dir, name, &op_data->op_fid3);
1187 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1188 ll_finish_md_op_data(op_data);
1192 ll_update_times(request, dir);
1194 rc = ll_objects_destroy(request, dir);
1196 ptlrpc_req_finished(request);
1200 static int ll_rename_generic(struct inode *src, struct dentry *src_dparent,
1201 struct dentry *src_dchild, struct qstr *src_name,
1202 struct inode *tgt, struct dentry *tgt_dparent,
1203 struct dentry *tgt_dchild, struct qstr *tgt_name)
1205 struct ptlrpc_request *request = NULL;
1206 struct ll_sb_info *sbi = ll_i2sbi(src);
1207 struct md_op_data *op_data;
1210 CDEBUG(D_VFSTRACE,"VFS Op:oldname=%.*s,src_dir=%lu/%u(%p),newname=%.*s,"
1211 "tgt_dir=%lu/%u(%p)\n", src_name->len, src_name->name,
1212 src->i_ino, src->i_generation, src, tgt_name->len,
1213 tgt_name->name, tgt->i_ino, tgt->i_generation, tgt);
1215 if (unlikely(ll_d_mountpoint(src_dparent, src_dchild, src_name) ||
1216 ll_d_mountpoint(tgt_dparent, tgt_dchild, tgt_name)))
1219 op_data = ll_prep_md_op_data(NULL, src, tgt, NULL, 0, 0,
1220 LUSTRE_OPC_ANY, NULL);
1221 if (IS_ERR(op_data))
1222 RETURN(PTR_ERR(op_data));
1224 ll_get_child_fid(src, src_name, &op_data->op_fid3);
1225 ll_get_child_fid(tgt, tgt_name, &op_data->op_fid4);
1226 err = md_rename(sbi->ll_md_exp, op_data,
1227 src_name->name, src_name->len,
1228 tgt_name->name, tgt_name->len, &request);
1229 ll_finish_md_op_data(op_data);
1231 ll_update_times(request, src);
1232 ll_update_times(request, tgt);
1233 err = ll_objects_destroy(request, src);
1236 ptlrpc_req_finished(request);
1241 #ifdef HAVE_VFS_INTENT_PATCHES
1242 static int ll_mknod_raw(struct nameidata *nd, int mode, dev_t rdev)
1244 return ll_mknod_generic(nd->dentry->d_inode, &nd->last, mode,rdev,NULL);
1246 static int ll_rename_raw(struct nameidata *srcnd, struct nameidata *tgtnd)
1248 return ll_rename_generic(srcnd->dentry->d_inode, srcnd->dentry,
1250 tgtnd->dentry->d_inode, tgtnd->dentry,
1251 NULL, &tgtnd->last);
1253 static int ll_link_raw(struct nameidata *srcnd, struct nameidata *tgtnd)
1255 return ll_link_generic(srcnd->dentry->d_inode, tgtnd->dentry->d_inode,
1256 &tgtnd->last, NULL);
1258 static int ll_symlink_raw(struct nameidata *nd, const char *tgt)
1260 return ll_symlink_generic(nd->dentry->d_inode, &nd->last, tgt, NULL);
1262 static int ll_rmdir_raw(struct nameidata *nd)
1264 return ll_rmdir_generic(nd->dentry->d_inode, nd->dentry, NULL,
1267 static int ll_mkdir_raw(struct nameidata *nd, int mode)
1269 return ll_mkdir_generic(nd->dentry->d_inode, &nd->last, mode, NULL);
1271 static int ll_unlink_raw(struct nameidata *nd)
1273 return ll_unlink_generic(nd->dentry->d_inode, nd->dentry, NULL,
1278 static int ll_mknod(struct inode *dir, struct dentry *dchild, int mode,
1281 return ll_mknod_generic(dir, &dchild->d_name, mode,
1282 old_encode_dev(rdev), dchild);
1285 static int ll_unlink(struct inode * dir, struct dentry *dentry)
1287 return ll_unlink_generic(dir, NULL, dentry, &dentry->d_name);
1289 static int ll_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1291 return ll_mkdir_generic(dir, &dentry->d_name, mode, dentry);
1293 static int ll_rmdir(struct inode *dir, struct dentry *dentry)
1295 return ll_rmdir_generic(dir, NULL, dentry, &dentry->d_name);
1297 static int ll_symlink(struct inode *dir, struct dentry *dentry,
1298 const char *oldname)
1300 return ll_symlink_generic(dir, &dentry->d_name, oldname, dentry);
1302 static int ll_link(struct dentry *old_dentry, struct inode *dir,
1303 struct dentry *new_dentry)
1305 return ll_link_generic(old_dentry->d_inode, dir, &new_dentry->d_name,
1308 static int ll_rename(struct inode *old_dir, struct dentry *old_dentry,
1309 struct inode *new_dir, struct dentry *new_dentry)
1312 err = ll_rename_generic(old_dir, NULL,
1313 old_dentry, &old_dentry->d_name,
1314 new_dir, NULL, new_dentry,
1315 &new_dentry->d_name);
1317 #ifndef HAVE_FS_RENAME_DOES_D_MOVE
1318 if (!S_ISDIR(old_dentry->d_inode->i_mode))
1320 d_move(old_dentry, new_dentry);
1325 struct inode_operations ll_dir_inode_operations = {
1326 #ifdef HAVE_VFS_INTENT_PATCHES
1327 .link_raw = ll_link_raw,
1328 .unlink_raw = ll_unlink_raw,
1329 .symlink_raw = ll_symlink_raw,
1330 .mkdir_raw = ll_mkdir_raw,
1331 .rmdir_raw = ll_rmdir_raw,
1332 .mknod_raw = ll_mknod_raw,
1333 .rename_raw = ll_rename_raw,
1334 .setattr = ll_setattr,
1335 .setattr_raw = ll_setattr_raw,
1338 .lookup = ll_lookup_nd,
1339 .create = ll_create_nd,
1340 /* We need all these non-raw things for NFSD, to not patch it. */
1341 .unlink = ll_unlink,
1344 .symlink = ll_symlink,
1346 .rename = ll_rename,
1347 .setattr = ll_setattr,
1348 .getattr = ll_getattr,
1349 .permission = ll_inode_permission,
1350 .setxattr = ll_setxattr,
1351 .getxattr = ll_getxattr,
1352 .listxattr = ll_listxattr,
1353 .removexattr = ll_removexattr,
1356 struct inode_operations ll_special_inode_operations = {
1357 #ifdef HAVE_VFS_INTENT_PATCHES
1358 .setattr_raw = ll_setattr_raw,
1360 .setattr = ll_setattr,
1361 .getattr = ll_getattr,
1362 .permission = ll_inode_permission,
1363 .setxattr = ll_setxattr,
1364 .getxattr = ll_getxattr,
1365 .listxattr = ll_listxattr,
1366 .removexattr = ll_removexattr,