1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
4 * Copyright (c) 2002, 2003 Cluster File Systems, Inc.
6 * This file is part of Lustre, http://www.lustre.org.
8 * Lustre is free software; you can redistribute it and/or
9 * modify it under the terms of version 2 of the GNU General Public
10 * License as published by the Free Software Foundation.
12 * Lustre is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 * GNU General Public License for more details.
17 * You should have received a copy of the GNU General Public License
18 * along with Lustre; if not, write to the Free Software
19 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
23 #include <linux/sched.h>
25 #include <linux/smp_lock.h>
26 #include <linux/quotaops.h>
27 #include <linux/highmem.h>
28 #include <linux/pagemap.h>
30 #define DEBUG_SUBSYSTEM S_LLITE
32 #include <obd_support.h>
33 #include <lustre_fid.h>
34 #include <lustre_lite.h>
35 #include <lustre_dlm.h>
36 #include <lustre_ver.h>
37 #include <lustre_mdc.h>
38 #include "llite_internal.h"
41 * Check if we have something mounted at the named dchild.
42 * In such a case there would always be dentry present.
44 static int ll_d_mountpoint(struct dentry *dparent, struct dentry *dchild,
49 if (unlikely(dchild)) {
50 mounted = d_mountpoint(dchild);
52 dchild = d_lookup(dparent, name);
54 mounted = d_mountpoint(dchild);
61 int ll_unlock(__u32 mode, struct lustre_handle *lockh)
65 ldlm_lock_decref(lockh, mode);
71 /* called from iget5_locked->find_inode() under inode_lock spinlock */
72 static int ll_test_inode(struct inode *inode, void *opaque)
74 struct ll_inode_info *lli = ll_i2info(inode);
75 struct lustre_md *md = opaque;
77 if (unlikely(!(md->body->valid & OBD_MD_FLID))) {
78 CERROR("MDS body missing FID\n");
82 if (!lu_fid_eq(&lli->lli_fid, &md->body->fid1))
88 static int ll_set_inode(struct inode *inode, void *opaque)
95 * Get an inode by inode number (already instantiated by the intent lookup).
96 * Returns inode or NULL
98 struct inode *ll_iget(struct super_block *sb, ino_t hash,
101 struct ll_inode_info *lli;
106 inode = iget5_locked(sb, hash, ll_test_inode, ll_set_inode, md);
109 lli = ll_i2info(inode);
110 if (inode->i_state & I_NEW) {
111 ll_read_inode2(inode, md);
112 unlock_new_inode(inode);
114 if (!(inode->i_state & (I_FREEING | I_CLEAR)))
115 ll_update_inode(inode, md);
117 CDEBUG(D_VFSTRACE, "got inode: %lu/%u(%p) for "DFID"\n",
118 inode->i_ino, inode->i_generation, inode,
119 PFID(&lli->lli_fid));
125 static void ll_drop_negative_dentry(struct inode *dir)
127 struct dentry *dentry, *tmp_alias, *tmp_subdir;
129 spin_lock(&dcache_lock);
131 list_for_each_entry_safe(dentry, tmp_alias,
132 &dir->i_dentry,d_alias) {
133 if (!list_empty(&dentry->d_subdirs)) {
134 struct dentry *child;
135 list_for_each_entry_safe(child, tmp_subdir,
138 /* XXX Print some debug here? */
140 /* Negative dentry. If we were
141 dropping dcache lock, go
142 throught the list again */
143 if (ll_drop_dentry(child))
148 spin_unlock(&dcache_lock);
152 int ll_md_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
153 void *data, int flag)
156 struct lustre_handle lockh;
160 case LDLM_CB_BLOCKING:
161 ldlm_lock2handle(lock, &lockh);
162 rc = ldlm_cli_cancel(&lockh);
164 CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
168 case LDLM_CB_CANCELING: {
169 struct inode *inode = ll_inode_from_lock(lock);
170 __u64 bits = lock->l_policy_data.l_inodebits.bits;
173 /* Invalidate all dentries associated with this inode */
177 LASSERT(lock->l_flags & LDLM_FL_CANCELING);
178 if ((bits & MDS_INODELOCK_LOOKUP) &&
179 ll_have_md_lock(inode, MDS_INODELOCK_LOOKUP))
180 bits &= ~MDS_INODELOCK_LOOKUP;
181 if ((bits & MDS_INODELOCK_UPDATE) &&
182 ll_have_md_lock(inode, MDS_INODELOCK_UPDATE))
183 bits &= ~MDS_INODELOCK_UPDATE;
184 if ((bits & MDS_INODELOCK_OPEN) &&
185 ll_have_md_lock(inode, MDS_INODELOCK_OPEN))
186 bits &= ~MDS_INODELOCK_OPEN;
188 fid = ll_inode2fid(inode);
189 if (lock->l_resource->lr_name.name[0] != fid_seq(fid) ||
190 lock->l_resource->lr_name.name[1] != fid_oid(fid) ||
191 lock->l_resource->lr_name.name[2] != fid_ver(fid)) {
192 LDLM_ERROR(lock, "data mismatch with object "
193 DFID" (%p)", PFID(fid), inode);
196 if (bits & MDS_INODELOCK_OPEN) {
198 switch (lock->l_req_mode) {
209 CERROR("Unexpected lock mode for OPEN lock "
210 "%d, inode %ld\n", lock->l_req_mode,
213 ll_md_real_close(inode, flags);
216 if (bits & MDS_INODELOCK_UPDATE)
217 ll_i2info(inode)->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
219 if (S_ISDIR(inode->i_mode) &&
220 (bits & MDS_INODELOCK_UPDATE)) {
221 CDEBUG(D_INODE, "invalidating inode %lu\n",
223 truncate_inode_pages(inode->i_mapping, 0);
224 ll_drop_negative_dentry(inode);
227 if (inode->i_sb->s_root &&
228 inode != inode->i_sb->s_root->d_inode &&
229 (bits & MDS_INODELOCK_LOOKUP))
230 ll_unhash_aliases(inode);
241 __u32 ll_i2suppgid(struct inode *i)
243 if (in_group_p(i->i_gid))
244 return (__u32)i->i_gid;
249 /* Pack the required supplementary groups into the supplied groups array.
250 * If we don't need to use the groups from the target inode(s) then we
251 * instead pack one or more groups from the user's supplementary group
252 * array in case it might be useful. Not needed if doing an MDS-side upcall. */
253 void ll_i2gids(__u32 *suppgids, struct inode *i1, struct inode *i2)
260 LASSERT(suppgids != NULL);
262 suppgids[0] = ll_i2suppgid(i1);
265 suppgids[1] = ll_i2suppgid(i2);
270 for (i = 0; i < current_ngroups; i++) {
271 if (suppgids[0] == -1) {
272 if (current_groups[i] != suppgids[1])
273 suppgids[0] = current_groups[i];
276 if (suppgids[1] == -1) {
277 if (current_groups[i] != suppgids[0])
278 suppgids[1] = current_groups[i];
286 static void ll_d_add(struct dentry *de, struct inode *inode)
288 CDEBUG(D_DENTRY, "adding inode %p to dentry %p\n", inode, de);
290 if (!list_empty(&de->d_alias)) {
291 spin_unlock(&dcache_lock);
292 CERROR("dentry %.*s %p alias next %p, prev %p\n",
293 de->d_name.len, de->d_name.name, de,
294 de->d_alias.next, de->d_alias.prev);
298 list_add(&de->d_alias, &inode->i_dentry);
302 if (!d_unhashed(de)) {
303 spin_unlock(&dcache_lock);
304 CERROR("dentry %.*s %p hash next %p\n",
305 de->d_name.len, de->d_name.name, de, de->d_hash.next);
308 d_rehash_cond(de, 0);
311 /* Search "inode"'s alias list for a dentry that has the same name and parent
312 * as de. If found, return it. If not found, return de.
313 * Lustre can't use d_add_unique because don't unhash aliases for directory
314 * in ll_revalidate_it. After revaliadate inode will be have hashed aliases
315 * and it triggers BUG_ON in d_instantiate_unique (bug #10954).
317 static struct dentry *ll_find_alias(struct inode *inode, struct dentry *de)
319 struct list_head *tmp;
320 struct dentry *dentry;
321 struct dentry *last_discon = NULL;
323 spin_lock(&dcache_lock);
324 list_for_each(tmp, &inode->i_dentry) {
325 dentry = list_entry(tmp, struct dentry, d_alias);
327 /* We are called here with 'de' already on the aliases list. */
328 if (unlikely(dentry == de)) {
333 if (dentry->d_flags & DCACHE_DISCONNECTED) {
334 LASSERT(last_discon == NULL);
335 last_discon = dentry;
339 if (dentry->d_parent != de->d_parent)
342 if (dentry->d_name.hash != de->d_name.hash)
345 if (dentry->d_name.len != de->d_name.len)
348 if (memcmp(dentry->d_name.name, de->d_name.name,
349 de->d_name.len) != 0)
355 #ifdef DCACHE_LUSTRE_INVALID
356 dentry->d_flags &= ~DCACHE_LUSTRE_INVALID;
358 unlock_dentry(dentry);
359 d_rehash_cond(dentry, 0); /* avoid taking dcache_lock inside */
360 spin_unlock(&dcache_lock);
362 CDEBUG(D_DENTRY, "alias dentry %.*s (%p) parent %p inode %p "
363 "refc %d\n", de->d_name.len, de->d_name.name, de,
364 de->d_parent, de->d_inode, atomic_read(&de->d_count));
369 CDEBUG(D_DENTRY, "Reuse disconnected dentry %p inode %p "
370 "refc %d\n", last_discon, last_discon->d_inode,
371 atomic_read(&last_discon->d_count));
372 dget_locked(last_discon);
373 spin_unlock(&dcache_lock);
375 d_move(last_discon, de);
382 spin_unlock(&dcache_lock);
387 static inline void ll_dop_init(struct dentry *de, int *set)
390 if (likely(de->d_op != &ll_d_ops)) {
391 de->d_op = &ll_init_d_ops;
397 static inline void ll_dop_fini(struct dentry *de, int succ)
400 if (likely(de->d_op == &ll_init_d_ops)) {
402 de->d_op = &ll_d_ops;
404 de->d_op = &ll_fini_d_ops;
410 de->d_op = &ll_d_ops;
415 int ll_lookup_it_finish(struct ptlrpc_request *request,
416 struct lookup_intent *it, void *data)
418 struct it_cb_data *icbd = data;
419 struct dentry **de = icbd->icbd_childp;
420 struct inode *parent = icbd->icbd_parent;
421 struct ll_sb_info *sbi = ll_i2sbi(parent);
422 struct inode *inode = NULL;
426 ll_dop_init(*de, &set);
428 /* NB 1 request reference will be taken away by ll_intent_lock()
430 if (!it_disposition(it, DISP_LOOKUP_NEG)) {
431 struct dentry *save = *de;
433 rc = ll_prep_inode(&inode, request, (*de)->d_sb);
440 CDEBUG(D_DLMTRACE, "setting l_data to inode %p (%lu/%u)\n",
441 inode, inode->i_ino, inode->i_generation);
442 md_set_lock_data(sbi->ll_md_exp,
443 &it->d.lustre.it_lock_handle, inode);
445 /* We used to query real size from OSTs here, but actually
446 this is not needed. For stat() calls size would be updated
447 from subsequent do_revalidate()->ll_inode_revalidate_it() in
449 vfs_getattr_it->ll_getattr()->ll_inode_revalidate_it() in 2.6
450 Everybody else who needs correct file size would call
451 ll_glimpse_size or some equivalent themselves anyway.
452 Also see bug 7198. */
454 *de = ll_find_alias(inode, *de);
455 if (set && *de != save)
456 ll_dop_fini(save, 0);
458 /* Check that parent has UPDATE lock. If there is none, we
459 cannot afford to hash this dentry (done by ll_d_add) as it
460 might get picked up later when UPDATE lock will appear */
461 if (ll_have_md_lock(parent, MDS_INODELOCK_UPDATE)) {
462 spin_lock(&dcache_lock);
463 ll_d_add(*de, inode);
464 spin_unlock(&dcache_lock);
466 (*de)->d_inode = NULL;
467 /* We do not want to hash the dentry if don`t have a
468 * lock, but if this dentry is later used in d_move,
469 * we'd hit uninitialised list head d_hash, so we just
470 * do this to init d_hash field but leave dentry
471 * unhashed. (bug 10796). */
484 static struct dentry *ll_lookup_it(struct inode *parent, struct dentry *dentry,
485 struct lookup_intent *it, int lookup_flags)
487 struct lookup_intent lookup_it = { .it_op = IT_LOOKUP };
488 struct dentry *save = dentry, *retval;
489 struct ptlrpc_request *req = NULL;
490 struct md_op_data *op_data;
491 struct it_cb_data icbd;
496 if (dentry->d_name.len > ll_i2sbi(parent)->ll_namelen)
497 RETURN(ERR_PTR(-ENAMETOOLONG));
499 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),intent=%s\n",
500 dentry->d_name.len, dentry->d_name.name, parent->i_ino,
501 parent->i_generation, parent, LL_IT2STR(it));
503 if (d_mountpoint(dentry))
504 CERROR("Tell Peter, lookup on mtpt, it %s\n", LL_IT2STR(it));
506 ll_frob_intent(&it, &lookup_it);
508 /* As do_lookup is called before follow_mount, root dentry may be left
509 * not valid, revalidate it here. */
510 if (parent->i_sb->s_root && (parent->i_sb->s_root->d_inode == parent) &&
511 (it->it_op & (IT_OPEN | IT_CREAT))) {
512 rc = ll_inode_revalidate_it(parent->i_sb->s_root, it);
517 if (it->it_op == IT_GETATTR) {
518 rc = ll_statahead_enter(parent, &dentry, 1);
520 ll_statahead_exit(dentry, rc);
522 RETURN(retval = dentry);
526 icbd.icbd_childp = &dentry;
527 icbd.icbd_parent = parent;
529 if (it->it_op & IT_CREAT ||
530 (it->it_op & IT_OPEN && it->it_create_mode & O_CREAT))
531 opc = LUSTRE_OPC_CREATE;
533 opc = LUSTRE_OPC_ANY;
535 op_data = ll_prep_md_op_data(NULL, parent, NULL, dentry->d_name.name,
536 dentry->d_name.len, lookup_flags, opc,
539 RETURN((void *)op_data);
541 it->it_create_mode &= ~current->fs->umask;
543 rc = md_intent_lock(ll_i2mdexp(parent), op_data, NULL, 0, it,
544 lookup_flags, &req, ll_md_blocking_ast, 0);
545 ll_finish_md_op_data(op_data);
547 GOTO(out, retval = ERR_PTR(rc));
549 rc = ll_lookup_it_finish(req, it, &icbd);
551 ll_intent_release(it);
552 GOTO(out, retval = ERR_PTR(rc));
555 if ((it->it_op & IT_OPEN) && dentry->d_inode &&
556 !S_ISREG(dentry->d_inode->i_mode) &&
557 !S_ISDIR(dentry->d_inode->i_mode)) {
558 ll_release_openhandle(dentry, it);
560 ll_lookup_finish_locks(it, dentry);
563 GOTO(out, retval = NULL);
565 GOTO(out, retval = dentry);
568 ptlrpc_req_finished(req);
572 #ifdef HAVE_VFS_INTENT_PATCHES
573 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
574 struct nameidata *nd)
579 if (nd && nd->flags & LOOKUP_LAST && !(nd->flags & LOOKUP_LINK_NOTLAST))
580 de = ll_lookup_it(parent, dentry, &nd->intent, nd->flags);
582 de = ll_lookup_it(parent, dentry, NULL, 0);
587 struct lookup_intent *ll_convert_intent(struct open_intent *oit,
590 struct lookup_intent *it;
592 OBD_ALLOC(it, sizeof(*it));
594 return ERR_PTR(-ENOMEM);
596 if (lookup_flags & LOOKUP_OPEN) {
598 if (lookup_flags & LOOKUP_CREATE)
599 it->it_op |= IT_CREAT;
600 it->it_create_mode = oit->create_mode;
601 it->it_flags = oit->flags;
603 it->it_op = IT_GETATTR;
606 #ifndef HAVE_FILE_IN_STRUCT_INTENT
607 /* Since there is no way to pass our intent to ll_file_open,
608 * just check the file is there. Actual open will be done
610 if (it->it_op & IT_OPEN)
611 it->it_op = IT_LOOKUP;
617 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
618 struct nameidata *nd)
623 if (nd && !(nd->flags & (LOOKUP_CONTINUE|LOOKUP_PARENT))) {
624 struct lookup_intent *it;
626 #if defined(HAVE_FILE_IN_STRUCT_INTENT) && (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,17))
627 /* Did we came here from failed revalidate just to propagate
629 if (nd->flags & LOOKUP_OPEN)
630 if (IS_ERR(nd->intent.open.file))
631 RETURN((struct dentry *)nd->intent.open.file);
634 if (ll_d2d(dentry) && ll_d2d(dentry)->lld_it) {
635 it = ll_d2d(dentry)->lld_it;
636 ll_d2d(dentry)->lld_it = NULL;
638 it = ll_convert_intent(&nd->intent.open, nd->flags);
640 RETURN((struct dentry *)it);
643 de = ll_lookup_it(parent, dentry, it, nd->flags);
646 if ((nd->flags & LOOKUP_OPEN) && !IS_ERR(dentry)) { /* Open */
647 if (dentry->d_inode &&
648 it_disposition(it, DISP_OPEN_OPEN)) { /* nocreate */
649 #ifdef HAVE_FILE_IN_STRUCT_INTENT
650 if (S_ISFIFO(dentry->d_inode->i_mode)) {
651 // We cannot call open here as it would
654 (struct ptlrpc_request *)
655 it->d.lustre.it_data);
658 nd->intent.open.file->private_data = it;
659 filp =lookup_instantiate_filp(nd,dentry,
661 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,17))
662 /* 2.6.1[456] have a bug in open_namei() that forgets to check
663 * nd->intent.open.file for error, so we need to return it as lookup's result
668 de = (struct dentry *) filp;
673 #else /* HAVE_FILE_IN_STRUCT_INTENT */
674 /* Release open handle as we have no way to
675 * pass it to ll_file_open */
676 ll_release_openhandle(dentry, it);
677 #endif /* HAVE_FILE_IN_STRUCT_INTENT */
678 } else if (it_disposition(it, DISP_OPEN_CREATE)) {
679 // XXX This can only reliably work on assumption
680 // that there are NO hashed negative dentries.
681 ll_d2d(dentry)->lld_it = it;
682 it = NULL; /* Will be freed in ll_create_nd */
683 /* We absolutely depend on ll_create_nd to be
684 * called to not leak this intent and possible
685 * data attached to it */
690 ll_intent_release(it);
691 OBD_FREE(it, sizeof(*it));
694 de = ll_lookup_it(parent, dentry, NULL, 0);
701 /* We depend on "mode" being set with the proper file type/umask by now */
702 static struct inode *ll_create_node(struct inode *dir, const char *name,
703 int namelen, const void *data, int datalen,
704 int mode, __u64 extra,
705 struct lookup_intent *it)
707 struct inode *inode = NULL;
708 struct ptlrpc_request *request = NULL;
709 struct ll_sb_info *sbi = ll_i2sbi(dir);
713 LASSERT(it && it->d.lustre.it_disposition);
715 LASSERT(it_disposition(it, DISP_ENQ_CREATE_REF));
716 request = it->d.lustre.it_data;
717 it_clear_disposition(it, DISP_ENQ_CREATE_REF);
718 rc = ll_prep_inode(&inode, request, dir->i_sb);
720 GOTO(out, inode = ERR_PTR(rc));
722 LASSERT(list_empty(&inode->i_dentry));
724 /* We asked for a lock on the directory, but were granted a
725 * lock on the inode. Since we finally have an inode pointer,
726 * stuff it in the lock. */
727 CDEBUG(D_DLMTRACE, "setting l_ast_data to inode %p (%lu/%u)\n",
728 inode, inode->i_ino, inode->i_generation);
729 md_set_lock_data(sbi->ll_md_exp,
730 &it->d.lustre.it_lock_handle, inode);
733 ptlrpc_req_finished(request);
738 * By the time this is called, we already have created the directory cache
739 * entry for the new file, but it is so far negative - it has no inode.
741 * We defer creating the OBD object(s) until open, to keep the intent and
742 * non-intent code paths similar, and also because we do not have the MDS
743 * inode number before calling ll_create_node() (which is needed for LOV),
744 * so we would need to do yet another RPC to the MDS to store the LOV EA
745 * data on the MDS. If needed, we would pass the PACKED lmm as data and
746 * lmm_size in datalen (the MDS still has code which will handle that).
748 * If the create succeeds, we fill in the inode information
749 * with d_instantiate().
751 static int ll_create_it(struct inode *dir, struct dentry *dentry, int mode,
752 struct lookup_intent *it)
758 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),intent=%s\n",
759 dentry->d_name.len, dentry->d_name.name, dir->i_ino,
760 dir->i_generation, dir, LL_IT2STR(it));
762 rc = it_open_error(DISP_OPEN_CREATE, it);
766 inode = ll_create_node(dir, dentry->d_name.name, dentry->d_name.len,
767 NULL, 0, mode, 0, it);
769 RETURN(PTR_ERR(inode));
772 d_instantiate(dentry, inode);
773 /* Negative dentry may be unhashed if parent does not have UPDATE lock,
774 * but some callers, e.g. do_coredump, expect dentry to be hashed after
775 * successful create. Hash it here. */
776 spin_lock(&dcache_lock);
777 if (d_unhashed(dentry))
778 d_rehash_cond(dentry, 0);
779 spin_unlock(&dcache_lock);
783 static void ll_update_times(struct ptlrpc_request *request,
786 struct mdt_body *body = req_capsule_server_get(&request->rq_pill,
790 /* mtime is always updated with ctime, but can be set in past.
791 As write and utime(2) may happen within 1 second, and utime's
792 mtime has a priority over write's one, so take mtime from mds
793 for the same ctimes. */
794 if (body->valid & OBD_MD_FLCTIME &&
795 body->ctime >= LTIME_S(inode->i_ctime)) {
796 LTIME_S(inode->i_ctime) = body->ctime;
798 if (body->valid & OBD_MD_FLMTIME) {
799 CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
800 "to "LPU64"\n", inode->i_ino,
801 LTIME_S(inode->i_mtime), body->mtime);
802 LTIME_S(inode->i_mtime) = body->mtime;
807 static int ll_new_node(struct inode *dir, struct qstr *name,
808 const char *tgt, int mode, int rdev,
809 struct dentry *dchild, __u32 opc)
811 struct ptlrpc_request *request = NULL;
812 struct md_op_data *op_data;
813 struct inode *inode = NULL;
814 struct ll_sb_info *sbi = ll_i2sbi(dir);
819 if (unlikely(tgt != NULL))
820 tgt_len = strlen(tgt) + 1;
822 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
823 name->len, 0, opc, NULL);
825 GOTO(err_exit, err = PTR_ERR(op_data));
827 err = md_create(sbi->ll_md_exp, op_data, tgt, tgt_len, mode,
828 current->fsuid, current->fsgid,
829 current->cap_effective, rdev, &request);
830 ll_finish_md_op_data(op_data);
834 ll_update_times(request, dir);
837 err = ll_prep_inode(&inode, request, dchild->d_sb);
842 d_instantiate(dchild, inode);
846 ptlrpc_req_finished(request);
851 static int ll_mknod_generic(struct inode *dir, struct qstr *name, int mode,
852 unsigned rdev, struct dentry *dchild)
857 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p) mode %o dev %x\n",
858 name->len, name->name, dir->i_ino, dir->i_generation, dir,
861 mode &= ~current->fs->umask;
863 switch (mode & S_IFMT) {
865 mode |= S_IFREG; /* for mode = 0 case, fallthrough */
871 err = ll_new_node(dir, name, NULL, mode, rdev, dchild,
883 #ifndef HAVE_VFS_INTENT_PATCHES
884 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
885 int mode, struct nameidata *nd)
887 struct lookup_intent *it = ll_d2d(dentry)->lld_it;
891 return ll_mknod_generic(dir, &dentry->d_name, mode, 0, dentry);
893 ll_d2d(dentry)->lld_it = NULL;
895 /* Was there an error? Propagate it! */
896 if (it->d.lustre.it_status) {
897 rc = it->d.lustre.it_status;
901 rc = ll_create_it(dir, dentry, mode, it);
902 #ifdef HAVE_FILE_IN_STRUCT_INTENT
903 if (nd && (nd->flags & LOOKUP_OPEN) && dentry->d_inode) { /* Open */
904 nd->intent.open.file->private_data = it;
905 lookup_instantiate_filp(nd, dentry, NULL);
908 ll_release_openhandle(dentry,it);
912 ll_intent_release(it);
913 OBD_FREE(it, sizeof(*it));
918 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
919 int mode, struct nameidata *nd)
921 if (!nd || !nd->intent.d.lustre.it_disposition)
922 /* No saved request? Just mknod the file */
923 return ll_mknod_generic(dir, &dentry->d_name, mode, 0, dentry);
925 return ll_create_it(dir, dentry, mode, &nd->intent);
929 static int ll_symlink_generic(struct inode *dir, struct qstr *name,
930 const char *tgt, struct dentry *dchild)
935 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),target=%.*s\n",
936 name->len, name->name, dir->i_ino, dir->i_generation,
939 err = ll_new_node(dir, name, (char *)tgt, S_IFLNK | S_IRWXUGO,
940 0, dchild, LUSTRE_OPC_SYMLINK);
944 static int ll_link_generic(struct inode *src, struct inode *dir,
945 struct qstr *name, struct dentry *dchild)
947 struct ll_sb_info *sbi = ll_i2sbi(dir);
948 struct ptlrpc_request *request = NULL;
949 struct md_op_data *op_data;
954 "VFS Op: inode=%lu/%u(%p), dir=%lu/%u(%p), target=%.*s\n",
955 src->i_ino, src->i_generation, src, dir->i_ino,
956 dir->i_generation, dir, name->len, name->name);
958 op_data = ll_prep_md_op_data(NULL, src, dir, name->name, name->len,
959 0, LUSTRE_OPC_ANY, NULL);
961 RETURN(PTR_ERR(op_data));
963 err = md_link(sbi->ll_md_exp, op_data, &request);
964 ll_finish_md_op_data(op_data);
970 ll_update_times(request, dir);
973 ptlrpc_req_finished(request);
977 static int ll_mkdir_generic(struct inode *dir, struct qstr *name,
978 int mode, struct dentry *dchild)
984 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
985 name->len, name->name, dir->i_ino, dir->i_generation, dir);
987 mode = (mode & (S_IRWXUGO|S_ISVTX) & ~current->fs->umask) | S_IFDIR;
988 err = ll_new_node(dir, name, NULL, mode, 0, dchild, LUSTRE_OPC_MKDIR);
993 /* Try to find the child dentry by its name.
994 If found, put the result fid into @fid. */
995 static void ll_get_child_fid(struct inode * dir, struct qstr *name,
998 struct dentry *parent, *child;
1000 parent = list_entry(dir->i_dentry.next, struct dentry, d_alias);
1001 child = d_lookup(parent, name);
1004 *fid = *ll_inode2fid(child->d_inode);
1009 static int ll_rmdir_generic(struct inode *dir, struct dentry *dparent,
1010 struct dentry *dchild, struct qstr *name)
1012 struct ptlrpc_request *request = NULL;
1013 struct md_op_data *op_data;
1017 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
1018 name->len, name->name, dir->i_ino, dir->i_generation, dir);
1020 if (unlikely(ll_d_mountpoint(dparent, dchild, name)))
1023 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len,
1024 S_IFDIR, LUSTRE_OPC_ANY, NULL);
1025 if (IS_ERR(op_data))
1026 RETURN(PTR_ERR(op_data));
1028 ll_get_child_fid(dir, name, &op_data->op_fid3);
1029 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1030 ll_finish_md_op_data(op_data);
1032 ll_update_times(request, dir);
1033 ptlrpc_req_finished(request);
1037 int ll_objects_destroy(struct ptlrpc_request *request, struct inode *dir)
1039 struct mdt_body *body;
1040 struct lov_mds_md *eadata;
1041 struct lov_stripe_md *lsm = NULL;
1042 struct obd_trans_info oti = { 0 };
1047 /* req is swabbed so this is safe */
1048 body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
1049 if (!(body->valid & OBD_MD_FLEASIZE))
1052 if (body->eadatasize == 0) {
1053 CERROR("OBD_MD_FLEASIZE set but eadatasize zero\n");
1054 GOTO(out, rc = -EPROTO);
1057 /* The MDS sent back the EA because we unlinked the last reference
1058 * to this file. Use this EA to unlink the objects on the OST.
1059 * It's opaque so we don't swab here; we leave it to obd_unpackmd() to
1060 * check it is complete and sensible. */
1061 eadata = req_capsule_server_sized_get(&request->rq_pill, &RMF_MDT_MD,
1063 LASSERT(eadata != NULL);
1065 rc = obd_unpackmd(ll_i2dtexp(dir), &lsm, eadata, body->eadatasize);
1067 CERROR("obd_unpackmd: %d\n", rc);
1070 LASSERT(rc >= sizeof(*lsm));
1072 rc = obd_checkmd(ll_i2dtexp(dir), ll_i2mdexp(dir), lsm);
1074 GOTO(out_free_memmd, rc);
1078 GOTO(out_free_memmd, rc = -ENOMEM);
1080 oa->o_id = lsm->lsm_object_id;
1081 oa->o_gr = lsm->lsm_object_gr;
1082 oa->o_mode = body->mode & S_IFMT;
1083 oa->o_valid = OBD_MD_FLID | OBD_MD_FLTYPE | OBD_MD_FLGROUP;
1085 if (body->valid & OBD_MD_FLCOOKIE) {
1086 oa->o_valid |= OBD_MD_FLCOOKIE;
1087 oti.oti_logcookies =
1088 req_capsule_server_sized_get(&request->rq_pill,
1090 sizeof(struct llog_cookie) *
1091 lsm->lsm_stripe_count);
1092 if (oti.oti_logcookies == NULL) {
1093 oa->o_valid &= ~OBD_MD_FLCOOKIE;
1094 body->valid &= ~OBD_MD_FLCOOKIE;
1098 rc = obd_destroy(ll_i2dtexp(dir), oa, lsm, &oti, ll_i2mdexp(dir));
1101 CERROR("obd destroy objid "LPX64" error %d\n",
1102 lsm->lsm_object_id, rc);
1104 obd_free_memmd(ll_i2dtexp(dir), &lsm);
1109 static int ll_unlink_generic(struct inode *dir, struct dentry *dparent,
1110 struct dentry *dchild, struct qstr *name)
1112 struct ptlrpc_request *request = NULL;
1113 struct md_op_data *op_data;
1116 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
1117 name->len, name->name, dir->i_ino, dir->i_generation, dir);
1120 * XXX: unlink bind mountpoint maybe call to here,
1121 * just check it as vfs_unlink does.
1123 if (unlikely(ll_d_mountpoint(dparent, dchild, name)))
1126 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
1127 name->len, 0, LUSTRE_OPC_ANY, NULL);
1128 if (IS_ERR(op_data))
1129 RETURN(PTR_ERR(op_data));
1131 ll_get_child_fid(dir, name, &op_data->op_fid3);
1132 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1133 ll_finish_md_op_data(op_data);
1137 ll_update_times(request, dir);
1139 rc = ll_objects_destroy(request, dir);
1141 ptlrpc_req_finished(request);
1145 static int ll_rename_generic(struct inode *src, struct dentry *src_dparent,
1146 struct dentry *src_dchild, struct qstr *src_name,
1147 struct inode *tgt, struct dentry *tgt_dparent,
1148 struct dentry *tgt_dchild, struct qstr *tgt_name)
1150 struct ptlrpc_request *request = NULL;
1151 struct ll_sb_info *sbi = ll_i2sbi(src);
1152 struct md_op_data *op_data;
1155 CDEBUG(D_VFSTRACE,"VFS Op:oldname=%.*s,src_dir=%lu/%u(%p),newname=%.*s,"
1156 "tgt_dir=%lu/%u(%p)\n", src_name->len, src_name->name,
1157 src->i_ino, src->i_generation, src, tgt_name->len,
1158 tgt_name->name, tgt->i_ino, tgt->i_generation, tgt);
1160 if (unlikely(ll_d_mountpoint(src_dparent, src_dchild, src_name) ||
1161 ll_d_mountpoint(tgt_dparent, tgt_dchild, tgt_name)))
1164 op_data = ll_prep_md_op_data(NULL, src, tgt, NULL, 0, 0,
1165 LUSTRE_OPC_ANY, NULL);
1166 if (IS_ERR(op_data))
1167 RETURN(PTR_ERR(op_data));
1169 ll_get_child_fid(src, src_name, &op_data->op_fid3);
1170 ll_get_child_fid(tgt, tgt_name, &op_data->op_fid4);
1171 err = md_rename(sbi->ll_md_exp, op_data,
1172 src_name->name, src_name->len,
1173 tgt_name->name, tgt_name->len, &request);
1174 ll_finish_md_op_data(op_data);
1176 ll_update_times(request, src);
1177 ll_update_times(request, tgt);
1178 err = ll_objects_destroy(request, src);
1181 ptlrpc_req_finished(request);
1186 #ifdef HAVE_VFS_INTENT_PATCHES
1187 static int ll_mknod_raw(struct nameidata *nd, int mode, dev_t rdev)
1189 return ll_mknod_generic(nd->dentry->d_inode, &nd->last, mode,rdev,NULL);
1191 static int ll_rename_raw(struct nameidata *srcnd, struct nameidata *tgtnd)
1193 return ll_rename_generic(srcnd->dentry->d_inode, srcnd->dentry,
1195 tgtnd->dentry->d_inode, tgtnd->dentry,
1196 NULL, &tgtnd->last);
1198 static int ll_link_raw(struct nameidata *srcnd, struct nameidata *tgtnd)
1200 return ll_link_generic(srcnd->dentry->d_inode, tgtnd->dentry->d_inode,
1201 &tgtnd->last, NULL);
1203 static int ll_symlink_raw(struct nameidata *nd, const char *tgt)
1205 return ll_symlink_generic(nd->dentry->d_inode, &nd->last, tgt, NULL);
1207 static int ll_rmdir_raw(struct nameidata *nd)
1209 return ll_rmdir_generic(nd->dentry->d_inode, nd->dentry, NULL,
1212 static int ll_mkdir_raw(struct nameidata *nd, int mode)
1214 return ll_mkdir_generic(nd->dentry->d_inode, &nd->last, mode, NULL);
1216 static int ll_unlink_raw(struct nameidata *nd)
1218 return ll_unlink_generic(nd->dentry->d_inode, nd->dentry, NULL,
1223 static int ll_mknod(struct inode *dir, struct dentry *dchild, int mode,
1226 return ll_mknod_generic(dir, &dchild->d_name, mode,
1227 old_encode_dev(rdev), dchild);
1230 static int ll_unlink(struct inode * dir, struct dentry *dentry)
1232 return ll_unlink_generic(dir, NULL, dentry, &dentry->d_name);
1234 static int ll_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1236 return ll_mkdir_generic(dir, &dentry->d_name, mode, dentry);
1238 static int ll_rmdir(struct inode *dir, struct dentry *dentry)
1240 return ll_rmdir_generic(dir, NULL, dentry, &dentry->d_name);
1242 static int ll_symlink(struct inode *dir, struct dentry *dentry,
1243 const char *oldname)
1245 return ll_symlink_generic(dir, &dentry->d_name, oldname, dentry);
1247 static int ll_link(struct dentry *old_dentry, struct inode *dir,
1248 struct dentry *new_dentry)
1250 return ll_link_generic(old_dentry->d_inode, dir, &new_dentry->d_name,
1253 static int ll_rename(struct inode *old_dir, struct dentry *old_dentry,
1254 struct inode *new_dir, struct dentry *new_dentry)
1256 return ll_rename_generic(old_dir, NULL,
1257 old_dentry, &old_dentry->d_name,
1258 new_dir, NULL, new_dentry,
1259 &new_dentry->d_name);
1262 struct inode_operations ll_dir_inode_operations = {
1263 #ifdef HAVE_VFS_INTENT_PATCHES
1264 .link_raw = ll_link_raw,
1265 .unlink_raw = ll_unlink_raw,
1266 .symlink_raw = ll_symlink_raw,
1267 .mkdir_raw = ll_mkdir_raw,
1268 .rmdir_raw = ll_rmdir_raw,
1269 .mknod_raw = ll_mknod_raw,
1270 .rename_raw = ll_rename_raw,
1271 .setattr = ll_setattr,
1272 .setattr_raw = ll_setattr_raw,
1275 .lookup = ll_lookup_nd,
1276 .create = ll_create_nd,
1277 /* We need all these non-raw things for NFSD, to not patch it. */
1278 .unlink = ll_unlink,
1281 .symlink = ll_symlink,
1283 .rename = ll_rename,
1284 .setattr = ll_setattr,
1285 .getattr = ll_getattr,
1286 .permission = ll_inode_permission,
1287 .setxattr = ll_setxattr,
1288 .getxattr = ll_getxattr,
1289 .listxattr = ll_listxattr,
1290 .removexattr = ll_removexattr,
1293 struct inode_operations ll_special_inode_operations = {
1294 #ifdef HAVE_VFS_INTENT_PATCHES
1295 .setattr_raw = ll_setattr_raw,
1297 .setattr = ll_setattr,
1298 .getattr = ll_getattr,
1299 .permission = ll_inode_permission,
1300 .setxattr = ll_setxattr,
1301 .getxattr = ll_getxattr,
1302 .listxattr = ll_listxattr,
1303 .removexattr = ll_removexattr,