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>
44 #include <linux/dcache.h>
45 #include <linux/buffer_head.h>
47 #define DEBUG_SUBSYSTEM S_LLITE
49 #include <obd_support.h>
50 #include <lustre_fid.h>
51 #include <lustre_lite.h>
52 #include <lustre_dlm.h>
53 #include <lustre_ver.h>
54 #include <lustre_mdc.h>
55 #include "llite_internal.h"
58 * Check if we have something mounted at the named dchild.
59 * In such a case there would always be dentry present.
61 static int ll_d_mountpoint(struct dentry *dparent, struct dentry *dchild,
66 if (unlikely(dchild)) {
67 mounted = d_mountpoint(dchild);
69 dchild = d_lookup(dparent, name);
71 mounted = d_mountpoint(dchild);
78 int ll_unlock(__u32 mode, struct lustre_handle *lockh)
82 ldlm_lock_decref(lockh, mode);
88 /* called from iget5_locked->find_inode() under inode_lock spinlock */
89 static int ll_test_inode(struct inode *inode, void *opaque)
91 struct ll_inode_info *lli = ll_i2info(inode);
92 struct lustre_md *md = opaque;
94 if (unlikely(!(md->body->valid & OBD_MD_FLID))) {
95 CERROR("MDS body missing FID\n");
99 if (!lu_fid_eq(&lli->lli_fid, &md->body->fid1))
105 static int ll_set_inode(struct inode *inode, void *opaque)
112 * Get an inode by inode number (already instantiated by the intent lookup).
113 * Returns inode or NULL
115 struct inode *ll_iget(struct super_block *sb, ino_t hash,
116 struct lustre_md *md)
118 struct ll_inode_info *lli;
123 inode = iget5_locked(sb, hash, ll_test_inode, ll_set_inode, md);
126 lli = ll_i2info(inode);
127 if (inode->i_state & I_NEW) {
130 ll_read_inode2(inode, md);
131 rc = cl_inode_init(inode, md);
134 make_bad_inode(inode);
135 unlock_new_inode(inode);
139 unlock_new_inode(inode);
140 } else if (!(inode->i_state & (I_FREEING | I_CLEAR)))
141 ll_update_inode(inode, md);
142 CDEBUG(D_VFSTRACE, "got inode: %p for "DFID"\n",
143 inode, PFID(&md->body->fid1));
148 static void ll_drop_negative_dentry(struct inode *dir)
150 struct dentry *dentry, *tmp_alias, *tmp_subdir;
152 spin_lock(&ll_lookup_lock);
153 spin_lock(&dcache_lock);
155 list_for_each_entry_safe(dentry, tmp_alias,
156 &dir->i_dentry,d_alias) {
157 if (!list_empty(&dentry->d_subdirs)) {
158 struct dentry *child;
159 list_for_each_entry_safe(child, tmp_subdir,
162 /* XXX Print some debug here? */
164 /* Negative dentry. If we were
165 dropping dcache lock, go
166 throught the list again */
167 if (ll_drop_dentry(child))
172 spin_unlock(&dcache_lock);
173 spin_unlock(&ll_lookup_lock);
177 int ll_md_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
178 void *data, int flag)
181 struct lustre_handle lockh;
185 case LDLM_CB_BLOCKING:
186 ldlm_lock2handle(lock, &lockh);
187 rc = ldlm_cli_cancel(&lockh);
189 CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
193 case LDLM_CB_CANCELING: {
194 struct inode *inode = ll_inode_from_lock(lock);
195 __u64 bits = lock->l_policy_data.l_inodebits.bits;
198 /* Invalidate all dentries associated with this inode */
202 LASSERT(lock->l_flags & LDLM_FL_CANCELING);
203 if ((bits & MDS_INODELOCK_LOOKUP) &&
204 ll_have_md_lock(inode, MDS_INODELOCK_LOOKUP))
205 bits &= ~MDS_INODELOCK_LOOKUP;
206 if ((bits & MDS_INODELOCK_UPDATE) &&
207 ll_have_md_lock(inode, MDS_INODELOCK_UPDATE))
208 bits &= ~MDS_INODELOCK_UPDATE;
209 if ((bits & MDS_INODELOCK_OPEN) &&
210 ll_have_md_lock(inode, MDS_INODELOCK_OPEN))
211 bits &= ~MDS_INODELOCK_OPEN;
213 fid = ll_inode2fid(inode);
214 if (lock->l_resource->lr_name.name[0] != fid_seq(fid) ||
215 lock->l_resource->lr_name.name[1] != fid_oid(fid) ||
216 lock->l_resource->lr_name.name[2] != fid_ver(fid)) {
217 LDLM_ERROR(lock, "data mismatch with object "
218 DFID" (%p)", PFID(fid), inode);
221 if (bits & MDS_INODELOCK_OPEN) {
223 switch (lock->l_req_mode) {
234 CERROR("Unexpected lock mode for OPEN lock "
235 "%d, inode %ld\n", lock->l_req_mode,
238 ll_md_real_close(inode, flags);
241 if (bits & MDS_INODELOCK_UPDATE)
242 ll_i2info(inode)->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
244 if (S_ISDIR(inode->i_mode) &&
245 (bits & MDS_INODELOCK_UPDATE)) {
246 CDEBUG(D_INODE, "invalidating inode %lu\n",
248 truncate_inode_pages(inode->i_mapping, 0);
249 ll_drop_negative_dentry(inode);
252 if (inode->i_sb->s_root &&
253 inode != inode->i_sb->s_root->d_inode &&
254 (bits & MDS_INODELOCK_LOOKUP))
255 ll_unhash_aliases(inode);
266 __u32 ll_i2suppgid(struct inode *i)
268 if (in_group_p(i->i_gid))
269 return (__u32)i->i_gid;
274 /* Pack the required supplementary groups into the supplied groups array.
275 * If we don't need to use the groups from the target inode(s) then we
276 * instead pack one or more groups from the user's supplementary group
277 * array in case it might be useful. Not needed if doing an MDS-side upcall. */
278 void ll_i2gids(__u32 *suppgids, struct inode *i1, struct inode *i2)
285 LASSERT(suppgids != NULL);
287 suppgids[0] = ll_i2suppgid(i1);
290 suppgids[1] = ll_i2suppgid(i2);
295 for (i = 0; i < current_ngroups; i++) {
296 if (suppgids[0] == -1) {
297 if (current_groups[i] != suppgids[1])
298 suppgids[0] = current_groups[i];
301 if (suppgids[1] == -1) {
302 if (current_groups[i] != suppgids[0])
303 suppgids[1] = current_groups[i];
311 static void ll_d_add(struct dentry *de, struct inode *inode)
313 CDEBUG(D_DENTRY, "adding inode %p to dentry %p\n", inode, de);
315 if (!list_empty(&de->d_alias)) {
316 spin_unlock(&dcache_lock);
317 CERROR("dentry %.*s %p alias next %p, prev %p\n",
318 de->d_name.len, de->d_name.name, de,
319 de->d_alias.next, de->d_alias.prev);
323 list_add(&de->d_alias, &inode->i_dentry);
327 if (!d_unhashed(de)) {
328 spin_unlock(&dcache_lock);
329 CERROR("dentry %.*s %p hash next %p\n",
330 de->d_name.len, de->d_name.name, de, de->d_hash.next);
333 d_rehash_cond(de, 0);
336 /* Search "inode"'s alias list for a dentry that has the same name and parent
337 * as de. If found, return it. If not found, return de.
338 * Lustre can't use d_add_unique because don't unhash aliases for directory
339 * in ll_revalidate_it. After revaliadate inode will be have hashed aliases
340 * and it triggers BUG_ON in d_instantiate_unique (bug #10954).
342 static struct dentry *ll_find_alias(struct inode *inode, struct dentry *de)
344 struct list_head *tmp;
345 struct dentry *dentry;
346 struct dentry *last_discon = NULL;
348 spin_lock(&ll_lookup_lock);
349 spin_lock(&dcache_lock);
350 list_for_each(tmp, &inode->i_dentry) {
351 dentry = list_entry(tmp, struct dentry, d_alias);
353 /* We are called here with 'de' already on the aliases list. */
354 if (unlikely(dentry == de)) {
359 if (dentry->d_flags & DCACHE_DISCONNECTED) {
360 LASSERT(last_discon == NULL);
361 last_discon = dentry;
365 if (dentry->d_parent != de->d_parent)
368 if (dentry->d_name.hash != de->d_name.hash)
371 if (dentry->d_name.len != de->d_name.len)
374 if (memcmp(dentry->d_name.name, de->d_name.name,
375 de->d_name.len) != 0)
381 unlock_dentry(dentry);
382 ll_dops_init(dentry, 0);
383 d_rehash_cond(dentry, 0); /* avoid taking dcache_lock inside */
384 spin_unlock(&dcache_lock);
385 spin_unlock(&ll_lookup_lock);
387 CDEBUG(D_DENTRY, "alias dentry %.*s (%p) parent %p inode %p "
388 "refc %d\n", de->d_name.len, de->d_name.name, de,
389 de->d_parent, de->d_inode, atomic_read(&de->d_count));
394 CDEBUG(D_DENTRY, "Reuse disconnected dentry %p inode %p "
395 "refc %d\n", last_discon, last_discon->d_inode,
396 atomic_read(&last_discon->d_count));
397 dget_locked(last_discon);
398 lock_dentry(last_discon);
399 last_discon->d_flags |= DCACHE_LUSTRE_INVALID;
400 unlock_dentry(last_discon);
401 spin_unlock(&dcache_lock);
402 spin_unlock(&ll_lookup_lock);
403 ll_dops_init(last_discon, 1);
405 d_move(last_discon, de);
409 de->d_flags |= DCACHE_LUSTRE_INVALID;
412 spin_unlock(&dcache_lock);
413 spin_unlock(&ll_lookup_lock);
418 int ll_lookup_it_finish(struct ptlrpc_request *request,
419 struct lookup_intent *it, void *data)
421 struct it_cb_data *icbd = data;
422 struct dentry **de = icbd->icbd_childp;
423 struct inode *parent = icbd->icbd_parent;
424 struct ll_sb_info *sbi = ll_i2sbi(parent);
425 struct inode *inode = NULL;
429 /* NB 1 request reference will be taken away by ll_intent_lock()
431 if (!it_disposition(it, DISP_LOOKUP_NEG)) {
432 struct dentry *save = *de;
435 rc = ll_prep_inode(&inode, request, (*de)->d_sb);
439 CDEBUG(D_DLMTRACE, "setting l_data to inode %p (%lu/%u)\n",
440 inode, inode->i_ino, inode->i_generation);
441 md_set_lock_data(sbi->ll_md_exp,
442 &it->d.lustre.it_lock_handle, inode, &bits);
444 /* We used to query real size from OSTs here, but actually
445 this is not needed. For stat() calls size would be updated
446 from subsequent do_revalidate()->ll_inode_revalidate_it() in
448 vfs_getattr_it->ll_getattr()->ll_inode_revalidate_it() in 2.6
449 Everybody else who needs correct file size would call
450 cl_glimpse_size or some equivalent themselves anyway.
451 Also see bug 7198. */
453 ll_dops_init(*de, 1);
454 *de = ll_find_alias(inode, *de);
456 struct ll_dentry_data *lld = ll_d2d(*de);
458 /* just make sure the ll_dentry_data is ready */
459 if (unlikely(lld == NULL)) {
462 if (likely(lld != NULL))
463 lld->lld_sa_generation = 0;
466 /* we have lookup look - unhide dentry */
467 if (bits & MDS_INODELOCK_LOOKUP) {
469 (*de)->d_flags &= ~(DCACHE_LUSTRE_INVALID);
473 ll_dops_init(*de, 1);
474 /* Check that parent has UPDATE lock. If there is none, we
475 cannot afford to hash this dentry (done by ll_d_add) as it
476 might get picked up later when UPDATE lock will appear */
477 if (ll_have_md_lock(parent, MDS_INODELOCK_UPDATE)) {
478 spin_lock(&dcache_lock);
480 spin_unlock(&dcache_lock);
482 /* negative lookup - and don't have update lock to
485 (*de)->d_flags |= DCACHE_LUSTRE_INVALID;
488 (*de)->d_inode = NULL;
489 /* We do not want to hash the dentry if don`t have a
490 * lock, but if this dentry is later used in d_move,
491 * we'd hit uninitialised list head d_hash, so we just
492 * do this to init d_hash field but leave dentry
493 * unhashed. (bug 10796). */
502 static struct dentry *ll_lookup_it(struct inode *parent, struct dentry *dentry,
503 struct lookup_intent *it, int lookup_flags)
505 struct lookup_intent lookup_it = { .it_op = IT_LOOKUP };
506 struct dentry *save = dentry, *retval;
507 struct ptlrpc_request *req = NULL;
508 struct md_op_data *op_data;
509 struct it_cb_data icbd;
514 if (dentry->d_name.len > ll_i2sbi(parent)->ll_namelen)
515 RETURN(ERR_PTR(-ENAMETOOLONG));
517 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),intent=%s\n",
518 dentry->d_name.len, dentry->d_name.name, parent->i_ino,
519 parent->i_generation, parent, LL_IT2STR(it));
521 if (d_mountpoint(dentry))
522 CERROR("Tell Peter, lookup on mtpt, it %s\n", LL_IT2STR(it));
524 ll_frob_intent(&it, &lookup_it);
526 /* As do_lookup is called before follow_mount, root dentry may be left
527 * not valid, revalidate it here. */
528 if (parent->i_sb->s_root && (parent->i_sb->s_root->d_inode == parent) &&
529 (it->it_op & (IT_OPEN | IT_CREAT))) {
530 rc = ll_inode_revalidate_it(parent->i_sb->s_root, it);
535 if (it->it_op == IT_GETATTR) {
536 first = ll_statahead_enter(parent, &dentry, 1);
538 ll_statahead_exit(dentry, first);
540 RETURN(retval = dentry);
544 icbd.icbd_childp = &dentry;
545 icbd.icbd_parent = parent;
547 if (it->it_op & IT_CREAT ||
548 (it->it_op & IT_OPEN && it->it_create_mode & O_CREAT))
549 opc = LUSTRE_OPC_CREATE;
551 opc = LUSTRE_OPC_ANY;
553 op_data = ll_prep_md_op_data(NULL, parent, NULL, dentry->d_name.name,
554 dentry->d_name.len, lookup_flags, opc,
557 RETURN((void *)op_data);
559 it->it_create_mode &= ~current->fs->umask;
561 rc = md_intent_lock(ll_i2mdexp(parent), op_data, NULL, 0, it,
562 lookup_flags, &req, ll_md_blocking_ast, 0);
563 ll_finish_md_op_data(op_data);
565 GOTO(out, retval = ERR_PTR(rc));
567 rc = ll_lookup_it_finish(req, it, &icbd);
569 ll_intent_release(it);
570 GOTO(out, retval = ERR_PTR(rc));
573 if (first == -EEXIST)
574 ll_statahead_mark(dentry);
576 if ((it->it_op & IT_OPEN) && dentry->d_inode &&
577 !S_ISREG(dentry->d_inode->i_mode) &&
578 !S_ISDIR(dentry->d_inode->i_mode)) {
579 ll_release_openhandle(dentry, it);
581 ll_lookup_finish_locks(it, dentry);
584 GOTO(out, retval = NULL);
586 GOTO(out, retval = dentry);
589 ptlrpc_req_finished(req);
593 #ifdef HAVE_VFS_INTENT_PATCHES
594 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
595 struct nameidata *nd)
600 if (nd && nd->flags & LOOKUP_LAST && !(nd->flags & LOOKUP_LINK_NOTLAST))
601 de = ll_lookup_it(parent, dentry, &nd->intent, nd->flags);
603 de = ll_lookup_it(parent, dentry, NULL, 0);
608 struct lookup_intent *ll_convert_intent(struct open_intent *oit,
611 struct lookup_intent *it;
613 OBD_ALLOC(it, sizeof(*it));
615 return ERR_PTR(-ENOMEM);
617 if (lookup_flags & LOOKUP_OPEN) {
619 if (lookup_flags & LOOKUP_CREATE)
620 it->it_op |= IT_CREAT;
621 it->it_create_mode = oit->create_mode;
622 it->it_flags = oit->flags;
624 it->it_op = IT_GETATTR;
627 #ifndef HAVE_FILE_IN_STRUCT_INTENT
628 /* Since there is no way to pass our intent to ll_file_open,
629 * just check the file is there. Actual open will be done
631 if (it->it_op & IT_OPEN)
632 it->it_op = IT_LOOKUP;
638 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
639 struct nameidata *nd)
644 if (nd && !(nd->flags & (LOOKUP_CONTINUE|LOOKUP_PARENT))) {
645 struct lookup_intent *it;
647 #if defined(HAVE_FILE_IN_STRUCT_INTENT) && (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,17))
648 /* Did we came here from failed revalidate just to propagate
650 if (nd->flags & LOOKUP_OPEN)
651 if (IS_ERR(nd->intent.open.file))
652 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 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);
727 * check new allocated inode, new inode shld not have any valid alias
729 static void ll_validate_new_inode(struct inode *new)
731 struct list_head *lp;
732 struct dentry * dentry;
736 if (list_empty(&new->i_dentry))
739 spin_lock(&dcache_lock);
740 list_for_each(lp, &new->i_dentry) {
741 dentry = list_entry(lp, struct dentry, d_alias);
742 if (!d_unhashed(dentry) && !(dentry->d_flags & DCACHE_LUSTRE_INVALID)){
749 spin_unlock(&dcache_lock);
751 if (need_inval && !in_recheck) {
752 /* kill all old inode's data pages */
753 truncate_inode_pages(new->i_mapping, 0);
755 /* invalidate all dirent and recheck inode */
756 ll_unhash_aliases(new);
762 /* We depend on "mode" being set with the proper file type/umask by now */
763 static struct inode *ll_create_node(struct inode *dir, const char *name,
764 int namelen, const void *data, int datalen,
765 int mode, __u64 extra,
766 struct lookup_intent *it)
768 struct inode *inode = NULL;
769 struct ptlrpc_request *request = NULL;
770 struct ll_sb_info *sbi = ll_i2sbi(dir);
774 LASSERT(it && it->d.lustre.it_disposition);
776 LASSERT(it_disposition(it, DISP_ENQ_CREATE_REF));
777 request = it->d.lustre.it_data;
778 it_clear_disposition(it, DISP_ENQ_CREATE_REF);
779 rc = ll_prep_inode(&inode, request, dir->i_sb);
781 GOTO(out, inode = ERR_PTR(rc));
783 ll_validate_new_inode(inode);
785 /* We asked for a lock on the directory, but were granted a
786 * lock on the inode. Since we finally have an inode pointer,
787 * stuff it in the lock. */
788 CDEBUG(D_DLMTRACE, "setting l_ast_data to inode %p (%lu/%u)\n",
789 inode, inode->i_ino, inode->i_generation);
790 md_set_lock_data(sbi->ll_md_exp,
791 &it->d.lustre.it_lock_handle, inode, NULL);
794 ptlrpc_req_finished(request);
799 * By the time this is called, we already have created the directory cache
800 * entry for the new file, but it is so far negative - it has no inode.
802 * We defer creating the OBD object(s) until open, to keep the intent and
803 * non-intent code paths similar, and also because we do not have the MDS
804 * inode number before calling ll_create_node() (which is needed for LOV),
805 * so we would need to do yet another RPC to the MDS to store the LOV EA
806 * data on the MDS. If needed, we would pass the PACKED lmm as data and
807 * lmm_size in datalen (the MDS still has code which will handle that).
809 * If the create succeeds, we fill in the inode information
810 * with d_instantiate().
812 static int ll_create_it(struct inode *dir, struct dentry *dentry, int mode,
813 struct lookup_intent *it)
819 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),intent=%s\n",
820 dentry->d_name.len, dentry->d_name.name, dir->i_ino,
821 dir->i_generation, dir, LL_IT2STR(it));
823 rc = it_open_error(DISP_OPEN_CREATE, it);
827 inode = ll_create_node(dir, dentry->d_name.name, dentry->d_name.len,
828 NULL, 0, mode, 0, it);
830 RETURN(PTR_ERR(inode));
833 /* it might been set during parent dir revalidation */
834 dentry->d_flags &= ~DCACHE_LUSTRE_INVALID;
835 d_instantiate(dentry, inode);
836 /* Negative dentry may be unhashed if parent does not have UPDATE lock,
837 * but some callers, e.g. do_coredump, expect dentry to be hashed after
838 * successful create. Hash it here. */
839 spin_lock(&dcache_lock);
840 if (d_unhashed(dentry))
841 d_rehash_cond(dentry, 0);
842 spin_unlock(&dcache_lock);
846 static void ll_update_times(struct ptlrpc_request *request,
849 struct mdt_body *body = req_capsule_server_get(&request->rq_pill,
853 /* mtime is always updated with ctime, but can be set in past.
854 As write and utime(2) may happen within 1 second, and utime's
855 mtime has a priority over write's one, so take mtime from mds
856 for the same ctimes. */
857 if (body->valid & OBD_MD_FLCTIME &&
858 body->ctime >= LTIME_S(inode->i_ctime)) {
859 LTIME_S(inode->i_ctime) = body->ctime;
861 if (body->valid & OBD_MD_FLMTIME) {
862 CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
863 "to "LPU64"\n", inode->i_ino,
864 LTIME_S(inode->i_mtime), body->mtime);
865 LTIME_S(inode->i_mtime) = body->mtime;
870 static int ll_new_node(struct inode *dir, struct qstr *name,
871 const char *tgt, int mode, int rdev,
872 struct dentry *dchild, __u32 opc)
874 struct ptlrpc_request *request = NULL;
875 struct md_op_data *op_data;
876 struct inode *inode = NULL;
877 struct ll_sb_info *sbi = ll_i2sbi(dir);
882 if (unlikely(tgt != NULL))
883 tgt_len = strlen(tgt) + 1;
885 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
886 name->len, 0, opc, NULL);
888 GOTO(err_exit, err = PTR_ERR(op_data));
890 err = md_create(sbi->ll_md_exp, op_data, tgt, tgt_len, mode,
891 current->fsuid, current->fsgid,
892 cfs_curproc_cap_pack(), rdev, &request);
893 ll_finish_md_op_data(op_data);
897 ll_update_times(request, dir);
900 err = ll_prep_inode(&inode, request, dchild->d_sb);
905 d_instantiate(dchild, inode);
909 ptlrpc_req_finished(request);
914 static int ll_mknod_generic(struct inode *dir, struct qstr *name, int mode,
915 unsigned rdev, struct dentry *dchild)
920 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p) mode %o dev %x\n",
921 name->len, name->name, dir->i_ino, dir->i_generation, dir,
924 mode &= ~current->fs->umask;
926 switch (mode & S_IFMT) {
928 mode |= S_IFREG; /* for mode = 0 case, fallthrough */
934 err = ll_new_node(dir, name, NULL, mode, rdev, dchild,
946 #ifndef HAVE_VFS_INTENT_PATCHES
947 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
948 int mode, struct nameidata *nd)
950 struct lookup_intent *it = ll_d2d(dentry)->lld_it;
954 return ll_mknod_generic(dir, &dentry->d_name, mode, 0, dentry);
956 ll_d2d(dentry)->lld_it = NULL;
958 /* Was there an error? Propagate it! */
959 if (it->d.lustre.it_status) {
960 rc = it->d.lustre.it_status;
964 rc = ll_create_it(dir, dentry, mode, it);
965 #ifdef HAVE_FILE_IN_STRUCT_INTENT
966 if (nd && (nd->flags & LOOKUP_OPEN) && dentry->d_inode) { /* Open */
967 nd->intent.open.file->private_data = it;
968 lookup_instantiate_filp(nd, dentry, NULL);
971 ll_release_openhandle(dentry,it);
975 ll_intent_release(it);
976 OBD_FREE(it, sizeof(*it));
981 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
982 int mode, struct nameidata *nd)
984 if (!nd || !nd->intent.d.lustre.it_disposition)
985 /* No saved request? Just mknod the file */
986 return ll_mknod_generic(dir, &dentry->d_name, mode, 0, dentry);
988 return ll_create_it(dir, dentry, mode, &nd->intent);
992 static int ll_symlink_generic(struct inode *dir, struct qstr *name,
993 const char *tgt, struct dentry *dchild)
998 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),target=%.*s\n",
999 name->len, name->name, dir->i_ino, dir->i_generation,
1002 err = ll_new_node(dir, name, (char *)tgt, S_IFLNK | S_IRWXUGO,
1003 0, dchild, LUSTRE_OPC_SYMLINK);
1007 static int ll_link_generic(struct inode *src, struct inode *dir,
1008 struct qstr *name, struct dentry *dchild)
1010 struct ll_sb_info *sbi = ll_i2sbi(dir);
1011 struct ptlrpc_request *request = NULL;
1012 struct md_op_data *op_data;
1017 "VFS Op: inode=%lu/%u(%p), dir=%lu/%u(%p), target=%.*s\n",
1018 src->i_ino, src->i_generation, src, dir->i_ino,
1019 dir->i_generation, dir, name->len, name->name);
1021 op_data = ll_prep_md_op_data(NULL, src, dir, name->name, name->len,
1022 0, LUSTRE_OPC_ANY, NULL);
1023 if (IS_ERR(op_data))
1024 RETURN(PTR_ERR(op_data));
1026 err = md_link(sbi->ll_md_exp, op_data, &request);
1027 ll_finish_md_op_data(op_data);
1033 ll_update_times(request, dir);
1036 ptlrpc_req_finished(request);
1040 static int ll_mkdir_generic(struct inode *dir, struct qstr *name,
1041 int mode, struct dentry *dchild)
1047 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
1048 name->len, name->name, dir->i_ino, dir->i_generation, dir);
1050 mode = (mode & (S_IRWXUGO|S_ISVTX) & ~current->fs->umask) | S_IFDIR;
1051 err = ll_new_node(dir, name, NULL, mode, 0, dchild, LUSTRE_OPC_MKDIR);
1056 /* Try to find the child dentry by its name.
1057 If found, put the result fid into @fid. */
1058 static void ll_get_child_fid(struct inode * dir, struct qstr *name,
1061 struct dentry *parent, *child;
1063 parent = list_entry(dir->i_dentry.next, struct dentry, d_alias);
1064 child = d_lookup(parent, name);
1067 *fid = *ll_inode2fid(child->d_inode);
1072 static int ll_rmdir_generic(struct inode *dir, struct dentry *dparent,
1073 struct dentry *dchild, struct qstr *name)
1075 struct ptlrpc_request *request = NULL;
1076 struct md_op_data *op_data;
1080 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
1081 name->len, name->name, dir->i_ino, dir->i_generation, dir);
1083 if (unlikely(ll_d_mountpoint(dparent, dchild, name)))
1086 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len,
1087 S_IFDIR, LUSTRE_OPC_ANY, NULL);
1088 if (IS_ERR(op_data))
1089 RETURN(PTR_ERR(op_data));
1091 ll_get_child_fid(dir, name, &op_data->op_fid3);
1092 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1093 ll_finish_md_op_data(op_data);
1095 ll_update_times(request, dir);
1096 ptlrpc_req_finished(request);
1100 int ll_objects_destroy(struct ptlrpc_request *request, struct inode *dir)
1102 struct mdt_body *body;
1103 struct lov_mds_md *eadata;
1104 struct lov_stripe_md *lsm = NULL;
1105 struct obd_trans_info oti = { 0 };
1107 struct obd_capa *oc = NULL;
1111 /* req is swabbed so this is safe */
1112 body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
1113 if (!(body->valid & OBD_MD_FLEASIZE))
1116 if (body->eadatasize == 0) {
1117 CERROR("OBD_MD_FLEASIZE set but eadatasize zero\n");
1118 GOTO(out, rc = -EPROTO);
1121 /* The MDS sent back the EA because we unlinked the last reference
1122 * to this file. Use this EA to unlink the objects on the OST.
1123 * It's opaque so we don't swab here; we leave it to obd_unpackmd() to
1124 * check it is complete and sensible. */
1125 eadata = req_capsule_server_sized_get(&request->rq_pill, &RMF_MDT_MD,
1127 LASSERT(eadata != NULL);
1129 rc = obd_unpackmd(ll_i2dtexp(dir), &lsm, eadata, body->eadatasize);
1131 CERROR("obd_unpackmd: %d\n", rc);
1134 LASSERT(rc >= sizeof(*lsm));
1136 rc = obd_checkmd(ll_i2dtexp(dir), ll_i2mdexp(dir), lsm);
1138 GOTO(out_free_memmd, rc);
1142 GOTO(out_free_memmd, rc = -ENOMEM);
1144 oa->o_id = lsm->lsm_object_id;
1145 oa->o_gr = lsm->lsm_object_gr;
1146 oa->o_mode = body->mode & S_IFMT;
1147 oa->o_valid = OBD_MD_FLID | OBD_MD_FLTYPE | OBD_MD_FLGROUP;
1149 if (body->valid & OBD_MD_FLCOOKIE) {
1150 oa->o_valid |= OBD_MD_FLCOOKIE;
1151 oti.oti_logcookies =
1152 req_capsule_server_sized_get(&request->rq_pill,
1154 sizeof(struct llog_cookie) *
1155 lsm->lsm_stripe_count);
1156 if (oti.oti_logcookies == NULL) {
1157 oa->o_valid &= ~OBD_MD_FLCOOKIE;
1158 body->valid &= ~OBD_MD_FLCOOKIE;
1162 if (body->valid & OBD_MD_FLOSSCAPA) {
1163 rc = md_unpack_capa(ll_i2mdexp(dir), request, &RMF_CAPA2, &oc);
1165 GOTO(out_free_memmd, rc);
1168 rc = obd_destroy(ll_i2dtexp(dir), oa, lsm, &oti, ll_i2mdexp(dir), oc);
1172 CERROR("obd destroy objid "LPX64" error %d\n",
1173 lsm->lsm_object_id, rc);
1175 obd_free_memmd(ll_i2dtexp(dir), &lsm);
1180 static int ll_unlink_generic(struct inode *dir, struct dentry *dparent,
1181 struct dentry *dchild, struct qstr *name)
1183 struct ptlrpc_request *request = NULL;
1184 struct md_op_data *op_data;
1187 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
1188 name->len, name->name, dir->i_ino, dir->i_generation, dir);
1191 * XXX: unlink bind mountpoint maybe call to here,
1192 * just check it as vfs_unlink does.
1194 if (unlikely(ll_d_mountpoint(dparent, dchild, name)))
1197 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
1198 name->len, 0, LUSTRE_OPC_ANY, NULL);
1199 if (IS_ERR(op_data))
1200 RETURN(PTR_ERR(op_data));
1202 ll_get_child_fid(dir, name, &op_data->op_fid3);
1203 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1204 ll_finish_md_op_data(op_data);
1208 ll_update_times(request, dir);
1210 rc = ll_objects_destroy(request, dir);
1212 ptlrpc_req_finished(request);
1216 static int ll_rename_generic(struct inode *src, struct dentry *src_dparent,
1217 struct dentry *src_dchild, struct qstr *src_name,
1218 struct inode *tgt, struct dentry *tgt_dparent,
1219 struct dentry *tgt_dchild, struct qstr *tgt_name)
1221 struct ptlrpc_request *request = NULL;
1222 struct ll_sb_info *sbi = ll_i2sbi(src);
1223 struct md_op_data *op_data;
1226 CDEBUG(D_VFSTRACE,"VFS Op:oldname=%.*s,src_dir=%lu/%u(%p),newname=%.*s,"
1227 "tgt_dir=%lu/%u(%p)\n", src_name->len, src_name->name,
1228 src->i_ino, src->i_generation, src, tgt_name->len,
1229 tgt_name->name, tgt->i_ino, tgt->i_generation, tgt);
1231 if (unlikely(ll_d_mountpoint(src_dparent, src_dchild, src_name) ||
1232 ll_d_mountpoint(tgt_dparent, tgt_dchild, tgt_name)))
1235 op_data = ll_prep_md_op_data(NULL, src, tgt, NULL, 0, 0,
1236 LUSTRE_OPC_ANY, NULL);
1237 if (IS_ERR(op_data))
1238 RETURN(PTR_ERR(op_data));
1240 ll_get_child_fid(src, src_name, &op_data->op_fid3);
1241 ll_get_child_fid(tgt, tgt_name, &op_data->op_fid4);
1242 err = md_rename(sbi->ll_md_exp, op_data,
1243 src_name->name, src_name->len,
1244 tgt_name->name, tgt_name->len, &request);
1245 ll_finish_md_op_data(op_data);
1247 ll_update_times(request, src);
1248 ll_update_times(request, tgt);
1249 err = ll_objects_destroy(request, src);
1252 ptlrpc_req_finished(request);
1257 #ifdef HAVE_VFS_INTENT_PATCHES
1258 static int ll_mknod_raw(struct nameidata *nd, int mode, dev_t rdev)
1260 return ll_mknod_generic(nd->dentry->d_inode, &nd->last, mode,rdev,NULL);
1262 static int ll_rename_raw(struct nameidata *srcnd, struct nameidata *tgtnd)
1264 return ll_rename_generic(srcnd->dentry->d_inode, srcnd->dentry,
1266 tgtnd->dentry->d_inode, tgtnd->dentry,
1267 NULL, &tgtnd->last);
1269 static int ll_link_raw(struct nameidata *srcnd, struct nameidata *tgtnd)
1271 return ll_link_generic(srcnd->dentry->d_inode, tgtnd->dentry->d_inode,
1272 &tgtnd->last, NULL);
1274 static int ll_symlink_raw(struct nameidata *nd, const char *tgt)
1276 return ll_symlink_generic(nd->dentry->d_inode, &nd->last, tgt, NULL);
1278 static int ll_rmdir_raw(struct nameidata *nd)
1280 return ll_rmdir_generic(nd->dentry->d_inode, nd->dentry, NULL,
1283 static int ll_mkdir_raw(struct nameidata *nd, int mode)
1285 return ll_mkdir_generic(nd->dentry->d_inode, &nd->last, mode, NULL);
1287 static int ll_unlink_raw(struct nameidata *nd)
1289 return ll_unlink_generic(nd->dentry->d_inode, nd->dentry, NULL,
1294 static int ll_mknod(struct inode *dir, struct dentry *dchild, int mode,
1297 return ll_mknod_generic(dir, &dchild->d_name, mode,
1298 old_encode_dev(rdev), dchild);
1301 static int ll_unlink(struct inode * dir, struct dentry *dentry)
1303 return ll_unlink_generic(dir, NULL, dentry, &dentry->d_name);
1305 static int ll_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1307 return ll_mkdir_generic(dir, &dentry->d_name, mode, dentry);
1309 static int ll_rmdir(struct inode *dir, struct dentry *dentry)
1311 return ll_rmdir_generic(dir, NULL, dentry, &dentry->d_name);
1313 static int ll_symlink(struct inode *dir, struct dentry *dentry,
1314 const char *oldname)
1316 return ll_symlink_generic(dir, &dentry->d_name, oldname, dentry);
1318 static int ll_link(struct dentry *old_dentry, struct inode *dir,
1319 struct dentry *new_dentry)
1321 return ll_link_generic(old_dentry->d_inode, dir, &new_dentry->d_name,
1324 static int ll_rename(struct inode *old_dir, struct dentry *old_dentry,
1325 struct inode *new_dir, struct dentry *new_dentry)
1328 err = ll_rename_generic(old_dir, NULL,
1329 old_dentry, &old_dentry->d_name,
1330 new_dir, NULL, new_dentry,
1331 &new_dentry->d_name);
1333 #ifndef HAVE_FS_RENAME_DOES_D_MOVE
1334 if (!S_ISDIR(old_dentry->d_inode->i_mode))
1336 d_move(old_dentry, new_dentry);
1341 struct inode_operations ll_dir_inode_operations = {
1342 #ifdef HAVE_VFS_INTENT_PATCHES
1343 .link_raw = ll_link_raw,
1344 .unlink_raw = ll_unlink_raw,
1345 .symlink_raw = ll_symlink_raw,
1346 .mkdir_raw = ll_mkdir_raw,
1347 .rmdir_raw = ll_rmdir_raw,
1348 .mknod_raw = ll_mknod_raw,
1349 .rename_raw = ll_rename_raw,
1350 .setattr = ll_setattr,
1351 .setattr_raw = ll_setattr_raw,
1354 .lookup = ll_lookup_nd,
1355 .create = ll_create_nd,
1356 /* We need all these non-raw things for NFSD, to not patch it. */
1357 .unlink = ll_unlink,
1360 .symlink = ll_symlink,
1362 .rename = ll_rename,
1363 .setattr = ll_setattr,
1364 .getattr = ll_getattr,
1365 .permission = ll_inode_permission,
1366 .setxattr = ll_setxattr,
1367 .getxattr = ll_getxattr,
1368 .listxattr = ll_listxattr,
1369 .removexattr = ll_removexattr,
1372 struct inode_operations ll_special_inode_operations = {
1373 #ifdef HAVE_VFS_INTENT_PATCHES
1374 .setattr_raw = ll_setattr_raw,
1376 .setattr = ll_setattr,
1377 .getattr = ll_getattr,
1378 .permission = ll_inode_permission,
1379 .setxattr = ll_setxattr,
1380 .getxattr = ll_getxattr,
1381 .listxattr = ll_listxattr,
1382 .removexattr = ll_removexattr,