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)
110 * Get an inode by inode number (already instantiated by the intent lookup).
111 * Returns inode or NULL
113 struct inode *ll_iget(struct super_block *sb, ino_t hash,
114 struct lustre_md *md)
116 struct ll_inode_info *lli;
121 inode = iget5_locked(sb, hash, ll_test_inode, ll_set_inode, md);
124 lli = ll_i2info(inode);
125 if (inode->i_state & I_NEW) {
126 ll_read_inode2(inode, md);
127 unlock_new_inode(inode);
129 if (!(inode->i_state & (I_FREEING | I_CLEAR)))
130 ll_update_inode(inode, md);
132 CDEBUG(D_VFSTRACE, "got inode: %lu/%u(%p) for "DFID"\n",
133 inode->i_ino, inode->i_generation, inode,
134 PFID(&lli->lli_fid));
140 static void ll_drop_negative_dentry(struct inode *dir)
142 struct dentry *dentry, *tmp_alias, *tmp_subdir;
144 spin_lock(&dcache_lock);
146 list_for_each_entry_safe(dentry, tmp_alias,
147 &dir->i_dentry,d_alias) {
148 if (!list_empty(&dentry->d_subdirs)) {
149 struct dentry *child;
150 list_for_each_entry_safe(child, tmp_subdir,
153 /* XXX Print some debug here? */
155 /* Negative dentry. If we were
156 dropping dcache lock, go
157 throught the list again */
158 if (ll_drop_dentry(child))
163 spin_unlock(&dcache_lock);
167 int ll_md_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
168 void *data, int flag)
171 struct lustre_handle lockh;
175 case LDLM_CB_BLOCKING:
176 ldlm_lock2handle(lock, &lockh);
177 rc = ldlm_cli_cancel(&lockh);
179 CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
183 case LDLM_CB_CANCELING: {
184 struct inode *inode = ll_inode_from_lock(lock);
185 __u64 bits = lock->l_policy_data.l_inodebits.bits;
188 /* Invalidate all dentries associated with this inode */
192 LASSERT(lock->l_flags & LDLM_FL_CANCELING);
193 if ((bits & MDS_INODELOCK_LOOKUP) &&
194 ll_have_md_lock(inode, MDS_INODELOCK_LOOKUP))
195 bits &= ~MDS_INODELOCK_LOOKUP;
196 if ((bits & MDS_INODELOCK_UPDATE) &&
197 ll_have_md_lock(inode, MDS_INODELOCK_UPDATE))
198 bits &= ~MDS_INODELOCK_UPDATE;
199 if ((bits & MDS_INODELOCK_OPEN) &&
200 ll_have_md_lock(inode, MDS_INODELOCK_OPEN))
201 bits &= ~MDS_INODELOCK_OPEN;
203 fid = ll_inode2fid(inode);
204 if (lock->l_resource->lr_name.name[0] != fid_seq(fid) ||
205 lock->l_resource->lr_name.name[1] != fid_oid(fid) ||
206 lock->l_resource->lr_name.name[2] != fid_ver(fid)) {
207 LDLM_ERROR(lock, "data mismatch with object "
208 DFID" (%p)", PFID(fid), inode);
211 if (bits & MDS_INODELOCK_OPEN) {
213 switch (lock->l_req_mode) {
224 CERROR("Unexpected lock mode for OPEN lock "
225 "%d, inode %ld\n", lock->l_req_mode,
228 ll_md_real_close(inode, flags);
231 if (bits & MDS_INODELOCK_UPDATE)
232 ll_i2info(inode)->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
234 if (S_ISDIR(inode->i_mode) &&
235 (bits & MDS_INODELOCK_UPDATE)) {
236 CDEBUG(D_INODE, "invalidating inode %lu\n",
238 truncate_inode_pages(inode->i_mapping, 0);
239 ll_drop_negative_dentry(inode);
242 if (inode->i_sb->s_root &&
243 inode != inode->i_sb->s_root->d_inode &&
244 (bits & MDS_INODELOCK_LOOKUP))
245 ll_unhash_aliases(inode);
256 __u32 ll_i2suppgid(struct inode *i)
258 if (in_group_p(i->i_gid))
259 return (__u32)i->i_gid;
264 /* Pack the required supplementary groups into the supplied groups array.
265 * If we don't need to use the groups from the target inode(s) then we
266 * instead pack one or more groups from the user's supplementary group
267 * array in case it might be useful. Not needed if doing an MDS-side upcall. */
268 void ll_i2gids(__u32 *suppgids, struct inode *i1, struct inode *i2)
275 LASSERT(suppgids != NULL);
277 suppgids[0] = ll_i2suppgid(i1);
280 suppgids[1] = ll_i2suppgid(i2);
285 for (i = 0; i < current_ngroups; i++) {
286 if (suppgids[0] == -1) {
287 if (current_groups[i] != suppgids[1])
288 suppgids[0] = current_groups[i];
291 if (suppgids[1] == -1) {
292 if (current_groups[i] != suppgids[0])
293 suppgids[1] = current_groups[i];
301 static void ll_d_add(struct dentry *de, struct inode *inode)
303 CDEBUG(D_DENTRY, "adding inode %p to dentry %p\n", inode, de);
305 if (!list_empty(&de->d_alias)) {
306 spin_unlock(&dcache_lock);
307 CERROR("dentry %.*s %p alias next %p, prev %p\n",
308 de->d_name.len, de->d_name.name, de,
309 de->d_alias.next, de->d_alias.prev);
313 list_add(&de->d_alias, &inode->i_dentry);
317 if (!d_unhashed(de)) {
318 spin_unlock(&dcache_lock);
319 CERROR("dentry %.*s %p hash next %p\n",
320 de->d_name.len, de->d_name.name, de, de->d_hash.next);
323 d_rehash_cond(de, 0);
326 /* Search "inode"'s alias list for a dentry that has the same name and parent
327 * as de. If found, return it. If not found, return de.
328 * Lustre can't use d_add_unique because don't unhash aliases for directory
329 * in ll_revalidate_it. After revaliadate inode will be have hashed aliases
330 * and it triggers BUG_ON in d_instantiate_unique (bug #10954).
332 static struct dentry *ll_find_alias(struct inode *inode, struct dentry *de)
334 struct list_head *tmp;
335 struct dentry *dentry;
336 struct dentry *last_discon = NULL;
338 spin_lock(&dcache_lock);
339 list_for_each(tmp, &inode->i_dentry) {
340 dentry = list_entry(tmp, struct dentry, d_alias);
342 /* We are called here with 'de' already on the aliases list. */
343 if (unlikely(dentry == de)) {
348 if (dentry->d_flags & DCACHE_DISCONNECTED) {
349 LASSERT(last_discon == NULL);
350 last_discon = dentry;
354 if (dentry->d_parent != de->d_parent)
357 if (dentry->d_name.hash != de->d_name.hash)
360 if (dentry->d_name.len != de->d_name.len)
363 if (memcmp(dentry->d_name.name, de->d_name.name,
364 de->d_name.len) != 0)
370 #ifdef DCACHE_LUSTRE_INVALID
371 dentry->d_flags &= ~DCACHE_LUSTRE_INVALID;
373 unlock_dentry(dentry);
374 d_rehash_cond(dentry, 0); /* avoid taking dcache_lock inside */
375 spin_unlock(&dcache_lock);
377 CDEBUG(D_DENTRY, "alias dentry %.*s (%p) parent %p inode %p "
378 "refc %d\n", de->d_name.len, de->d_name.name, de,
379 de->d_parent, de->d_inode, atomic_read(&de->d_count));
384 CDEBUG(D_DENTRY, "Reuse disconnected dentry %p inode %p "
385 "refc %d\n", last_discon, last_discon->d_inode,
386 atomic_read(&last_discon->d_count));
387 dget_locked(last_discon);
388 spin_unlock(&dcache_lock);
390 d_move(last_discon, de);
397 spin_unlock(&dcache_lock);
402 static inline void ll_dop_init(struct dentry *de, int *set)
405 if (likely(de->d_op != &ll_d_ops)) {
406 de->d_op = &ll_init_d_ops;
412 static inline void ll_dop_fini(struct dentry *de, int succ)
415 if (likely(de->d_op == &ll_init_d_ops)) {
417 de->d_op = &ll_d_ops;
419 de->d_op = &ll_fini_d_ops;
425 de->d_op = &ll_d_ops;
430 int ll_lookup_it_finish(struct ptlrpc_request *request,
431 struct lookup_intent *it, void *data)
433 struct it_cb_data *icbd = data;
434 struct dentry **de = icbd->icbd_childp;
435 struct inode *parent = icbd->icbd_parent;
436 struct ll_sb_info *sbi = ll_i2sbi(parent);
437 struct inode *inode = NULL;
441 ll_dop_init(*de, &set);
443 /* NB 1 request reference will be taken away by ll_intent_lock()
445 if (!it_disposition(it, DISP_LOOKUP_NEG)) {
446 struct dentry *save = *de;
448 rc = ll_prep_inode(&inode, request, (*de)->d_sb);
455 CDEBUG(D_DLMTRACE, "setting l_data to inode %p (%lu/%u)\n",
456 inode, inode->i_ino, inode->i_generation);
457 md_set_lock_data(sbi->ll_md_exp,
458 &it->d.lustre.it_lock_handle, inode);
460 /* We used to query real size from OSTs here, but actually
461 this is not needed. For stat() calls size would be updated
462 from subsequent do_revalidate()->ll_inode_revalidate_it() in
464 vfs_getattr_it->ll_getattr()->ll_inode_revalidate_it() in 2.6
465 Everybody else who needs correct file size would call
466 ll_glimpse_size or some equivalent themselves anyway.
467 Also see bug 7198. */
469 *de = ll_find_alias(inode, *de);
470 if (set && *de != save)
471 ll_dop_fini(save, 0);
473 /* Check that parent has UPDATE lock. If there is none, we
474 cannot afford to hash this dentry (done by ll_d_add) as it
475 might get picked up later when UPDATE lock will appear */
476 if (ll_have_md_lock(parent, MDS_INODELOCK_UPDATE)) {
477 spin_lock(&dcache_lock);
478 ll_d_add(*de, inode);
479 spin_unlock(&dcache_lock);
481 (*de)->d_inode = NULL;
482 /* We do not want to hash the dentry if don`t have a
483 * lock, but if this dentry is later used in d_move,
484 * we'd hit uninitialised list head d_hash, so we just
485 * do this to init d_hash field but leave dentry
486 * unhashed. (bug 10796). */
499 static struct dentry *ll_lookup_it(struct inode *parent, struct dentry *dentry,
500 struct lookup_intent *it, int lookup_flags)
502 struct lookup_intent lookup_it = { .it_op = IT_LOOKUP };
503 struct dentry *save = dentry, *retval;
504 struct ptlrpc_request *req = NULL;
505 struct md_op_data *op_data;
506 struct it_cb_data icbd;
511 if (dentry->d_name.len > ll_i2sbi(parent)->ll_namelen)
512 RETURN(ERR_PTR(-ENAMETOOLONG));
514 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),intent=%s\n",
515 dentry->d_name.len, dentry->d_name.name, parent->i_ino,
516 parent->i_generation, parent, LL_IT2STR(it));
518 if (d_mountpoint(dentry))
519 CERROR("Tell Peter, lookup on mtpt, it %s\n", LL_IT2STR(it));
521 ll_frob_intent(&it, &lookup_it);
523 /* As do_lookup is called before follow_mount, root dentry may be left
524 * not valid, revalidate it here. */
525 if (parent->i_sb->s_root && (parent->i_sb->s_root->d_inode == parent) &&
526 (it->it_op & (IT_OPEN | IT_CREAT))) {
527 rc = ll_inode_revalidate_it(parent->i_sb->s_root, it);
532 if (it->it_op == IT_GETATTR) {
533 rc = ll_statahead_enter(parent, &dentry, 1);
535 ll_statahead_exit(dentry, rc);
537 RETURN(retval = dentry);
541 icbd.icbd_childp = &dentry;
542 icbd.icbd_parent = parent;
544 if (it->it_op & IT_CREAT ||
545 (it->it_op & IT_OPEN && it->it_create_mode & O_CREAT))
546 opc = LUSTRE_OPC_CREATE;
548 opc = LUSTRE_OPC_ANY;
550 op_data = ll_prep_md_op_data(NULL, parent, NULL, dentry->d_name.name,
551 dentry->d_name.len, lookup_flags, opc,
554 RETURN((void *)op_data);
556 it->it_create_mode &= ~current->fs->umask;
558 rc = md_intent_lock(ll_i2mdexp(parent), op_data, NULL, 0, it,
559 lookup_flags, &req, ll_md_blocking_ast, 0);
560 ll_finish_md_op_data(op_data);
562 GOTO(out, retval = ERR_PTR(rc));
564 rc = ll_lookup_it_finish(req, it, &icbd);
566 ll_intent_release(it);
567 GOTO(out, retval = ERR_PTR(rc));
570 if ((it->it_op & IT_OPEN) && dentry->d_inode &&
571 !S_ISREG(dentry->d_inode->i_mode) &&
572 !S_ISDIR(dentry->d_inode->i_mode)) {
573 ll_release_openhandle(dentry, it);
575 ll_lookup_finish_locks(it, dentry);
578 GOTO(out, retval = NULL);
580 GOTO(out, retval = dentry);
583 ptlrpc_req_finished(req);
587 #ifdef HAVE_VFS_INTENT_PATCHES
588 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
589 struct nameidata *nd)
594 if (nd && nd->flags & LOOKUP_LAST && !(nd->flags & LOOKUP_LINK_NOTLAST))
595 de = ll_lookup_it(parent, dentry, &nd->intent, nd->flags);
597 de = ll_lookup_it(parent, dentry, NULL, 0);
602 struct lookup_intent *ll_convert_intent(struct open_intent *oit,
605 struct lookup_intent *it;
607 OBD_ALLOC(it, sizeof(*it));
609 return ERR_PTR(-ENOMEM);
611 if (lookup_flags & LOOKUP_OPEN) {
613 if (lookup_flags & LOOKUP_CREATE)
614 it->it_op |= IT_CREAT;
615 it->it_create_mode = oit->create_mode;
616 it->it_flags = oit->flags;
618 it->it_op = IT_GETATTR;
621 #ifndef HAVE_FILE_IN_STRUCT_INTENT
622 /* Since there is no way to pass our intent to ll_file_open,
623 * just check the file is there. Actual open will be done
625 if (it->it_op & IT_OPEN)
626 it->it_op = IT_LOOKUP;
632 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
633 struct nameidata *nd)
638 if (nd && !(nd->flags & (LOOKUP_CONTINUE|LOOKUP_PARENT))) {
639 struct lookup_intent *it;
641 #if defined(HAVE_FILE_IN_STRUCT_INTENT) && (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,17))
642 /* Did we came here from failed revalidate just to propagate
644 if (nd->flags & LOOKUP_OPEN)
645 if (IS_ERR(nd->intent.open.file))
646 RETURN((struct dentry *)nd->intent.open.file);
649 if (ll_d2d(dentry) && ll_d2d(dentry)->lld_it) {
650 it = ll_d2d(dentry)->lld_it;
651 ll_d2d(dentry)->lld_it = NULL;
653 it = ll_convert_intent(&nd->intent.open, nd->flags);
655 RETURN((struct dentry *)it);
658 de = ll_lookup_it(parent, dentry, it, nd->flags);
661 if ((nd->flags & LOOKUP_OPEN) && !IS_ERR(dentry)) { /* Open */
662 if (dentry->d_inode &&
663 it_disposition(it, DISP_OPEN_OPEN)) { /* nocreate */
664 #ifdef HAVE_FILE_IN_STRUCT_INTENT
665 if (S_ISFIFO(dentry->d_inode->i_mode)) {
666 // We cannot call open here as it would
669 (struct ptlrpc_request *)
670 it->d.lustre.it_data);
673 nd->intent.open.file->private_data = it;
674 filp =lookup_instantiate_filp(nd,dentry,
676 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,17))
677 /* 2.6.1[456] have a bug in open_namei() that forgets to check
678 * nd->intent.open.file for error, so we need to return it as lookup's result
683 de = (struct dentry *) filp;
688 #else /* HAVE_FILE_IN_STRUCT_INTENT */
689 /* Release open handle as we have no way to
690 * pass it to ll_file_open */
691 ll_release_openhandle(dentry, it);
692 #endif /* HAVE_FILE_IN_STRUCT_INTENT */
693 } else if (it_disposition(it, DISP_OPEN_CREATE)) {
694 // XXX This can only reliably work on assumption
695 // that there are NO hashed negative dentries.
696 ll_d2d(dentry)->lld_it = it;
697 it = NULL; /* Will be freed in ll_create_nd */
698 /* We absolutely depend on ll_create_nd to be
699 * called to not leak this intent and possible
700 * data attached to it */
705 ll_intent_release(it);
706 OBD_FREE(it, sizeof(*it));
709 de = ll_lookup_it(parent, dentry, NULL, 0);
716 /* We depend on "mode" being set with the proper file type/umask by now */
717 static struct inode *ll_create_node(struct inode *dir, const char *name,
718 int namelen, const void *data, int datalen,
719 int mode, __u64 extra,
720 struct lookup_intent *it)
722 struct inode *inode = NULL;
723 struct ptlrpc_request *request = NULL;
724 struct ll_sb_info *sbi = ll_i2sbi(dir);
728 LASSERT(it && it->d.lustre.it_disposition);
730 LASSERT(it_disposition(it, DISP_ENQ_CREATE_REF));
731 request = it->d.lustre.it_data;
732 it_clear_disposition(it, DISP_ENQ_CREATE_REF);
733 rc = ll_prep_inode(&inode, request, dir->i_sb);
735 GOTO(out, inode = ERR_PTR(rc));
737 LASSERT(list_empty(&inode->i_dentry));
739 /* We asked for a lock on the directory, but were granted a
740 * lock on the inode. Since we finally have an inode pointer,
741 * stuff it in the lock. */
742 CDEBUG(D_DLMTRACE, "setting l_ast_data to inode %p (%lu/%u)\n",
743 inode, inode->i_ino, inode->i_generation);
744 md_set_lock_data(sbi->ll_md_exp,
745 &it->d.lustre.it_lock_handle, inode);
748 ptlrpc_req_finished(request);
753 * By the time this is called, we already have created the directory cache
754 * entry for the new file, but it is so far negative - it has no inode.
756 * We defer creating the OBD object(s) until open, to keep the intent and
757 * non-intent code paths similar, and also because we do not have the MDS
758 * inode number before calling ll_create_node() (which is needed for LOV),
759 * so we would need to do yet another RPC to the MDS to store the LOV EA
760 * data on the MDS. If needed, we would pass the PACKED lmm as data and
761 * lmm_size in datalen (the MDS still has code which will handle that).
763 * If the create succeeds, we fill in the inode information
764 * with d_instantiate().
766 static int ll_create_it(struct inode *dir, struct dentry *dentry, int mode,
767 struct lookup_intent *it)
773 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),intent=%s\n",
774 dentry->d_name.len, dentry->d_name.name, dir->i_ino,
775 dir->i_generation, dir, LL_IT2STR(it));
777 rc = it_open_error(DISP_OPEN_CREATE, it);
781 inode = ll_create_node(dir, dentry->d_name.name, dentry->d_name.len,
782 NULL, 0, mode, 0, it);
784 RETURN(PTR_ERR(inode));
787 d_instantiate(dentry, inode);
788 /* Negative dentry may be unhashed if parent does not have UPDATE lock,
789 * but some callers, e.g. do_coredump, expect dentry to be hashed after
790 * successful create. Hash it here. */
791 spin_lock(&dcache_lock);
792 if (d_unhashed(dentry))
793 d_rehash_cond(dentry, 0);
794 spin_unlock(&dcache_lock);
798 static void ll_update_times(struct ptlrpc_request *request,
801 struct mdt_body *body = req_capsule_server_get(&request->rq_pill,
805 /* mtime is always updated with ctime, but can be set in past.
806 As write and utime(2) may happen within 1 second, and utime's
807 mtime has a priority over write's one, so take mtime from mds
808 for the same ctimes. */
809 if (body->valid & OBD_MD_FLCTIME &&
810 body->ctime >= LTIME_S(inode->i_ctime)) {
811 LTIME_S(inode->i_ctime) = body->ctime;
813 if (body->valid & OBD_MD_FLMTIME) {
814 CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
815 "to "LPU64"\n", inode->i_ino,
816 LTIME_S(inode->i_mtime), body->mtime);
817 LTIME_S(inode->i_mtime) = body->mtime;
822 static int ll_new_node(struct inode *dir, struct qstr *name,
823 const char *tgt, int mode, int rdev,
824 struct dentry *dchild, __u32 opc)
826 struct ptlrpc_request *request = NULL;
827 struct md_op_data *op_data;
828 struct inode *inode = NULL;
829 struct ll_sb_info *sbi = ll_i2sbi(dir);
834 if (unlikely(tgt != NULL))
835 tgt_len = strlen(tgt) + 1;
837 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
838 name->len, 0, opc, NULL);
840 GOTO(err_exit, err = PTR_ERR(op_data));
842 err = md_create(sbi->ll_md_exp, op_data, tgt, tgt_len, mode,
843 current->fsuid, current->fsgid,
844 current->cap_effective, rdev, &request);
845 ll_finish_md_op_data(op_data);
849 ll_update_times(request, dir);
852 err = ll_prep_inode(&inode, request, dchild->d_sb);
857 d_instantiate(dchild, inode);
861 ptlrpc_req_finished(request);
866 static int ll_mknod_generic(struct inode *dir, struct qstr *name, int mode,
867 unsigned rdev, struct dentry *dchild)
872 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p) mode %o dev %x\n",
873 name->len, name->name, dir->i_ino, dir->i_generation, dir,
876 mode &= ~current->fs->umask;
878 switch (mode & S_IFMT) {
880 mode |= S_IFREG; /* for mode = 0 case, fallthrough */
886 err = ll_new_node(dir, name, NULL, mode, rdev, dchild,
898 #ifndef HAVE_VFS_INTENT_PATCHES
899 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
900 int mode, struct nameidata *nd)
902 struct lookup_intent *it = ll_d2d(dentry)->lld_it;
906 return ll_mknod_generic(dir, &dentry->d_name, mode, 0, dentry);
908 ll_d2d(dentry)->lld_it = NULL;
910 /* Was there an error? Propagate it! */
911 if (it->d.lustre.it_status) {
912 rc = it->d.lustre.it_status;
916 rc = ll_create_it(dir, dentry, mode, it);
917 #ifdef HAVE_FILE_IN_STRUCT_INTENT
918 if (nd && (nd->flags & LOOKUP_OPEN) && dentry->d_inode) { /* Open */
919 nd->intent.open.file->private_data = it;
920 lookup_instantiate_filp(nd, dentry, NULL);
923 ll_release_openhandle(dentry,it);
927 ll_intent_release(it);
928 OBD_FREE(it, sizeof(*it));
933 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
934 int mode, struct nameidata *nd)
936 if (!nd || !nd->intent.d.lustre.it_disposition)
937 /* No saved request? Just mknod the file */
938 return ll_mknod_generic(dir, &dentry->d_name, mode, 0, dentry);
940 return ll_create_it(dir, dentry, mode, &nd->intent);
944 static int ll_symlink_generic(struct inode *dir, struct qstr *name,
945 const char *tgt, struct dentry *dchild)
950 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),target=%.*s\n",
951 name->len, name->name, dir->i_ino, dir->i_generation,
954 err = ll_new_node(dir, name, (char *)tgt, S_IFLNK | S_IRWXUGO,
955 0, dchild, LUSTRE_OPC_SYMLINK);
959 static int ll_link_generic(struct inode *src, struct inode *dir,
960 struct qstr *name, struct dentry *dchild)
962 struct ll_sb_info *sbi = ll_i2sbi(dir);
963 struct ptlrpc_request *request = NULL;
964 struct md_op_data *op_data;
969 "VFS Op: inode=%lu/%u(%p), dir=%lu/%u(%p), target=%.*s\n",
970 src->i_ino, src->i_generation, src, dir->i_ino,
971 dir->i_generation, dir, name->len, name->name);
973 op_data = ll_prep_md_op_data(NULL, src, dir, name->name, name->len,
974 0, LUSTRE_OPC_ANY, NULL);
976 RETURN(PTR_ERR(op_data));
978 err = md_link(sbi->ll_md_exp, op_data, &request);
979 ll_finish_md_op_data(op_data);
985 ll_update_times(request, dir);
988 ptlrpc_req_finished(request);
992 static int ll_mkdir_generic(struct inode *dir, struct qstr *name,
993 int mode, struct dentry *dchild)
999 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
1000 name->len, name->name, dir->i_ino, dir->i_generation, dir);
1002 mode = (mode & (S_IRWXUGO|S_ISVTX) & ~current->fs->umask) | S_IFDIR;
1003 err = ll_new_node(dir, name, NULL, mode, 0, dchild, LUSTRE_OPC_MKDIR);
1008 /* Try to find the child dentry by its name.
1009 If found, put the result fid into @fid. */
1010 static void ll_get_child_fid(struct inode * dir, struct qstr *name,
1013 struct dentry *parent, *child;
1015 parent = list_entry(dir->i_dentry.next, struct dentry, d_alias);
1016 child = d_lookup(parent, name);
1019 *fid = *ll_inode2fid(child->d_inode);
1024 static int ll_rmdir_generic(struct inode *dir, struct dentry *dparent,
1025 struct dentry *dchild, struct qstr *name)
1027 struct ptlrpc_request *request = NULL;
1028 struct md_op_data *op_data;
1032 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
1033 name->len, name->name, dir->i_ino, dir->i_generation, dir);
1035 if (unlikely(ll_d_mountpoint(dparent, dchild, name)))
1038 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len,
1039 S_IFDIR, LUSTRE_OPC_ANY, NULL);
1040 if (IS_ERR(op_data))
1041 RETURN(PTR_ERR(op_data));
1043 ll_get_child_fid(dir, name, &op_data->op_fid3);
1044 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1045 ll_finish_md_op_data(op_data);
1047 ll_update_times(request, dir);
1048 ptlrpc_req_finished(request);
1052 int ll_objects_destroy(struct ptlrpc_request *request, struct inode *dir)
1054 struct mdt_body *body;
1055 struct lov_mds_md *eadata;
1056 struct lov_stripe_md *lsm = NULL;
1057 struct obd_trans_info oti = { 0 };
1062 /* req is swabbed so this is safe */
1063 body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
1064 if (!(body->valid & OBD_MD_FLEASIZE))
1067 if (body->eadatasize == 0) {
1068 CERROR("OBD_MD_FLEASIZE set but eadatasize zero\n");
1069 GOTO(out, rc = -EPROTO);
1072 /* The MDS sent back the EA because we unlinked the last reference
1073 * to this file. Use this EA to unlink the objects on the OST.
1074 * It's opaque so we don't swab here; we leave it to obd_unpackmd() to
1075 * check it is complete and sensible. */
1076 eadata = req_capsule_server_sized_get(&request->rq_pill, &RMF_MDT_MD,
1078 LASSERT(eadata != NULL);
1080 rc = obd_unpackmd(ll_i2dtexp(dir), &lsm, eadata, body->eadatasize);
1082 CERROR("obd_unpackmd: %d\n", rc);
1085 LASSERT(rc >= sizeof(*lsm));
1087 rc = obd_checkmd(ll_i2dtexp(dir), ll_i2mdexp(dir), lsm);
1089 GOTO(out_free_memmd, rc);
1093 GOTO(out_free_memmd, rc = -ENOMEM);
1095 oa->o_id = lsm->lsm_object_id;
1096 oa->o_gr = lsm->lsm_object_gr;
1097 oa->o_mode = body->mode & S_IFMT;
1098 oa->o_valid = OBD_MD_FLID | OBD_MD_FLTYPE | OBD_MD_FLGROUP;
1100 if (body->valid & OBD_MD_FLCOOKIE) {
1101 oa->o_valid |= OBD_MD_FLCOOKIE;
1102 oti.oti_logcookies =
1103 req_capsule_server_sized_get(&request->rq_pill,
1105 sizeof(struct llog_cookie) *
1106 lsm->lsm_stripe_count);
1107 if (oti.oti_logcookies == NULL) {
1108 oa->o_valid &= ~OBD_MD_FLCOOKIE;
1109 body->valid &= ~OBD_MD_FLCOOKIE;
1113 rc = obd_destroy(ll_i2dtexp(dir), oa, lsm, &oti, ll_i2mdexp(dir));
1116 CERROR("obd destroy objid "LPX64" error %d\n",
1117 lsm->lsm_object_id, rc);
1119 obd_free_memmd(ll_i2dtexp(dir), &lsm);
1124 static int ll_unlink_generic(struct inode *dir, struct dentry *dparent,
1125 struct dentry *dchild, struct qstr *name)
1127 struct ptlrpc_request *request = NULL;
1128 struct md_op_data *op_data;
1131 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
1132 name->len, name->name, dir->i_ino, dir->i_generation, dir);
1135 * XXX: unlink bind mountpoint maybe call to here,
1136 * just check it as vfs_unlink does.
1138 if (unlikely(ll_d_mountpoint(dparent, dchild, name)))
1141 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
1142 name->len, 0, LUSTRE_OPC_ANY, NULL);
1143 if (IS_ERR(op_data))
1144 RETURN(PTR_ERR(op_data));
1146 ll_get_child_fid(dir, name, &op_data->op_fid3);
1147 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1148 ll_finish_md_op_data(op_data);
1152 ll_update_times(request, dir);
1154 rc = ll_objects_destroy(request, dir);
1156 ptlrpc_req_finished(request);
1160 static int ll_rename_generic(struct inode *src, struct dentry *src_dparent,
1161 struct dentry *src_dchild, struct qstr *src_name,
1162 struct inode *tgt, struct dentry *tgt_dparent,
1163 struct dentry *tgt_dchild, struct qstr *tgt_name)
1165 struct ptlrpc_request *request = NULL;
1166 struct ll_sb_info *sbi = ll_i2sbi(src);
1167 struct md_op_data *op_data;
1170 CDEBUG(D_VFSTRACE,"VFS Op:oldname=%.*s,src_dir=%lu/%u(%p),newname=%.*s,"
1171 "tgt_dir=%lu/%u(%p)\n", src_name->len, src_name->name,
1172 src->i_ino, src->i_generation, src, tgt_name->len,
1173 tgt_name->name, tgt->i_ino, tgt->i_generation, tgt);
1175 if (unlikely(ll_d_mountpoint(src_dparent, src_dchild, src_name) ||
1176 ll_d_mountpoint(tgt_dparent, tgt_dchild, tgt_name)))
1179 op_data = ll_prep_md_op_data(NULL, src, tgt, NULL, 0, 0,
1180 LUSTRE_OPC_ANY, NULL);
1181 if (IS_ERR(op_data))
1182 RETURN(PTR_ERR(op_data));
1184 ll_get_child_fid(src, src_name, &op_data->op_fid3);
1185 ll_get_child_fid(tgt, tgt_name, &op_data->op_fid4);
1186 err = md_rename(sbi->ll_md_exp, op_data,
1187 src_name->name, src_name->len,
1188 tgt_name->name, tgt_name->len, &request);
1189 ll_finish_md_op_data(op_data);
1191 ll_update_times(request, src);
1192 ll_update_times(request, tgt);
1193 err = ll_objects_destroy(request, src);
1196 ptlrpc_req_finished(request);
1201 #ifdef HAVE_VFS_INTENT_PATCHES
1202 static int ll_mknod_raw(struct nameidata *nd, int mode, dev_t rdev)
1204 return ll_mknod_generic(nd->dentry->d_inode, &nd->last, mode,rdev,NULL);
1206 static int ll_rename_raw(struct nameidata *srcnd, struct nameidata *tgtnd)
1208 return ll_rename_generic(srcnd->dentry->d_inode, srcnd->dentry,
1210 tgtnd->dentry->d_inode, tgtnd->dentry,
1211 NULL, &tgtnd->last);
1213 static int ll_link_raw(struct nameidata *srcnd, struct nameidata *tgtnd)
1215 return ll_link_generic(srcnd->dentry->d_inode, tgtnd->dentry->d_inode,
1216 &tgtnd->last, NULL);
1218 static int ll_symlink_raw(struct nameidata *nd, const char *tgt)
1220 return ll_symlink_generic(nd->dentry->d_inode, &nd->last, tgt, NULL);
1222 static int ll_rmdir_raw(struct nameidata *nd)
1224 return ll_rmdir_generic(nd->dentry->d_inode, nd->dentry, NULL,
1227 static int ll_mkdir_raw(struct nameidata *nd, int mode)
1229 return ll_mkdir_generic(nd->dentry->d_inode, &nd->last, mode, NULL);
1231 static int ll_unlink_raw(struct nameidata *nd)
1233 return ll_unlink_generic(nd->dentry->d_inode, nd->dentry, NULL,
1238 static int ll_mknod(struct inode *dir, struct dentry *dchild, int mode,
1241 return ll_mknod_generic(dir, &dchild->d_name, mode,
1242 old_encode_dev(rdev), dchild);
1245 static int ll_unlink(struct inode * dir, struct dentry *dentry)
1247 return ll_unlink_generic(dir, NULL, dentry, &dentry->d_name);
1249 static int ll_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1251 return ll_mkdir_generic(dir, &dentry->d_name, mode, dentry);
1253 static int ll_rmdir(struct inode *dir, struct dentry *dentry)
1255 return ll_rmdir_generic(dir, NULL, dentry, &dentry->d_name);
1257 static int ll_symlink(struct inode *dir, struct dentry *dentry,
1258 const char *oldname)
1260 return ll_symlink_generic(dir, &dentry->d_name, oldname, dentry);
1262 static int ll_link(struct dentry *old_dentry, struct inode *dir,
1263 struct dentry *new_dentry)
1265 return ll_link_generic(old_dentry->d_inode, dir, &new_dentry->d_name,
1268 static int ll_rename(struct inode *old_dir, struct dentry *old_dentry,
1269 struct inode *new_dir, struct dentry *new_dentry)
1272 err = ll_rename_generic(old_dir, NULL,
1273 old_dentry, &old_dentry->d_name,
1274 new_dir, NULL, new_dentry,
1275 &new_dentry->d_name);
1277 #ifndef HAVE_FS_RENAME_DOES_D_MOVE
1278 if (!S_ISDIR(old_dentry->d_inode->i_mode))
1280 d_move(old_dentry, new_dentry);
1285 struct inode_operations ll_dir_inode_operations = {
1286 #ifdef HAVE_VFS_INTENT_PATCHES
1287 .link_raw = ll_link_raw,
1288 .unlink_raw = ll_unlink_raw,
1289 .symlink_raw = ll_symlink_raw,
1290 .mkdir_raw = ll_mkdir_raw,
1291 .rmdir_raw = ll_rmdir_raw,
1292 .mknod_raw = ll_mknod_raw,
1293 .rename_raw = ll_rename_raw,
1294 .setattr = ll_setattr,
1295 .setattr_raw = ll_setattr_raw,
1298 .lookup = ll_lookup_nd,
1299 .create = ll_create_nd,
1300 /* We need all these non-raw things for NFSD, to not patch it. */
1301 .unlink = ll_unlink,
1304 .symlink = ll_symlink,
1306 .rename = ll_rename,
1307 .setattr = ll_setattr,
1308 .getattr = ll_getattr,
1309 .permission = ll_inode_permission,
1310 .setxattr = ll_setxattr,
1311 .getxattr = ll_getxattr,
1312 .listxattr = ll_listxattr,
1313 .removexattr = ll_removexattr,
1316 struct inode_operations ll_special_inode_operations = {
1317 #ifdef HAVE_VFS_INTENT_PATCHES
1318 .setattr_raw = ll_setattr_raw,
1320 .setattr = ll_setattr,
1321 .getattr = ll_getattr,
1322 .permission = ll_inode_permission,
1323 .setxattr = ll_setxattr,
1324 .getxattr = ll_getxattr,
1325 .listxattr = ll_listxattr,
1326 .removexattr = ll_removexattr,