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(&ll_lookup_lock);
145 spin_lock(&dcache_lock);
147 list_for_each_entry_safe(dentry, tmp_alias,
148 &dir->i_dentry,d_alias) {
149 if (!list_empty(&dentry->d_subdirs)) {
150 struct dentry *child;
151 list_for_each_entry_safe(child, tmp_subdir,
154 /* XXX Print some debug here? */
156 /* Negative dentry. If we were
157 dropping dcache lock, go
158 throught the list again */
159 if (ll_drop_dentry(child))
164 spin_unlock(&dcache_lock);
165 spin_unlock(&ll_lookup_lock);
169 int ll_md_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
170 void *data, int flag)
173 struct lustre_handle lockh;
177 case LDLM_CB_BLOCKING:
178 ldlm_lock2handle(lock, &lockh);
179 rc = ldlm_cli_cancel(&lockh);
181 CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
185 case LDLM_CB_CANCELING: {
186 struct inode *inode = ll_inode_from_lock(lock);
187 __u64 bits = lock->l_policy_data.l_inodebits.bits;
190 /* Invalidate all dentries associated with this inode */
194 LASSERT(lock->l_flags & LDLM_FL_CANCELING);
195 if ((bits & MDS_INODELOCK_LOOKUP) &&
196 ll_have_md_lock(inode, MDS_INODELOCK_LOOKUP))
197 bits &= ~MDS_INODELOCK_LOOKUP;
198 if ((bits & MDS_INODELOCK_UPDATE) &&
199 ll_have_md_lock(inode, MDS_INODELOCK_UPDATE))
200 bits &= ~MDS_INODELOCK_UPDATE;
201 if ((bits & MDS_INODELOCK_OPEN) &&
202 ll_have_md_lock(inode, MDS_INODELOCK_OPEN))
203 bits &= ~MDS_INODELOCK_OPEN;
205 fid = ll_inode2fid(inode);
206 if (lock->l_resource->lr_name.name[0] != fid_seq(fid) ||
207 lock->l_resource->lr_name.name[1] != fid_oid(fid) ||
208 lock->l_resource->lr_name.name[2] != fid_ver(fid)) {
209 LDLM_ERROR(lock, "data mismatch with object "
210 DFID" (%p)", PFID(fid), inode);
213 if (bits & MDS_INODELOCK_OPEN) {
215 switch (lock->l_req_mode) {
226 CERROR("Unexpected lock mode for OPEN lock "
227 "%d, inode %ld\n", lock->l_req_mode,
230 ll_md_real_close(inode, flags);
233 if (bits & MDS_INODELOCK_UPDATE)
234 ll_i2info(inode)->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
236 if (S_ISDIR(inode->i_mode) &&
237 (bits & MDS_INODELOCK_UPDATE)) {
238 CDEBUG(D_INODE, "invalidating inode %lu\n",
240 truncate_inode_pages(inode->i_mapping, 0);
241 ll_drop_negative_dentry(inode);
244 if (inode->i_sb->s_root &&
245 inode != inode->i_sb->s_root->d_inode &&
246 (bits & MDS_INODELOCK_LOOKUP))
247 ll_unhash_aliases(inode);
258 __u32 ll_i2suppgid(struct inode *i)
260 if (in_group_p(i->i_gid))
261 return (__u32)i->i_gid;
266 /* Pack the required supplementary groups into the supplied groups array.
267 * If we don't need to use the groups from the target inode(s) then we
268 * instead pack one or more groups from the user's supplementary group
269 * array in case it might be useful. Not needed if doing an MDS-side upcall. */
270 void ll_i2gids(__u32 *suppgids, struct inode *i1, struct inode *i2)
277 LASSERT(suppgids != NULL);
279 suppgids[0] = ll_i2suppgid(i1);
282 suppgids[1] = ll_i2suppgid(i2);
287 for (i = 0; i < current_ngroups; i++) {
288 if (suppgids[0] == -1) {
289 if (current_groups[i] != suppgids[1])
290 suppgids[0] = current_groups[i];
293 if (suppgids[1] == -1) {
294 if (current_groups[i] != suppgids[0])
295 suppgids[1] = current_groups[i];
303 static void ll_d_add(struct dentry *de, struct inode *inode)
305 CDEBUG(D_DENTRY, "adding inode %p to dentry %p\n", inode, de);
307 if (!list_empty(&de->d_alias)) {
308 spin_unlock(&dcache_lock);
309 CERROR("dentry %.*s %p alias next %p, prev %p\n",
310 de->d_name.len, de->d_name.name, de,
311 de->d_alias.next, de->d_alias.prev);
315 list_add(&de->d_alias, &inode->i_dentry);
319 if (!d_unhashed(de)) {
320 spin_unlock(&dcache_lock);
321 CERROR("dentry %.*s %p hash next %p\n",
322 de->d_name.len, de->d_name.name, de, de->d_hash.next);
325 d_rehash_cond(de, 0);
328 /* Search "inode"'s alias list for a dentry that has the same name and parent
329 * as de. If found, return it. If not found, return de.
330 * Lustre can't use d_add_unique because don't unhash aliases for directory
331 * in ll_revalidate_it. After revaliadate inode will be have hashed aliases
332 * and it triggers BUG_ON in d_instantiate_unique (bug #10954).
334 static struct dentry *ll_find_alias(struct inode *inode, struct dentry *de)
336 struct list_head *tmp;
337 struct dentry *dentry;
338 struct dentry *last_discon = NULL;
340 spin_lock(&ll_lookup_lock);
341 spin_lock(&dcache_lock);
342 list_for_each(tmp, &inode->i_dentry) {
343 dentry = list_entry(tmp, struct dentry, d_alias);
345 /* We are called here with 'de' already on the aliases list. */
346 if (unlikely(dentry == de)) {
351 if (dentry->d_flags & DCACHE_DISCONNECTED) {
352 LASSERT(last_discon == NULL);
353 last_discon = dentry;
357 if (dentry->d_parent != de->d_parent)
360 if (dentry->d_name.hash != de->d_name.hash)
363 if (dentry->d_name.len != de->d_name.len)
366 if (memcmp(dentry->d_name.name, de->d_name.name,
367 de->d_name.len) != 0)
373 #ifdef DCACHE_LUSTRE_INVALID
374 dentry->d_flags &= ~DCACHE_LUSTRE_INVALID;
376 unlock_dentry(dentry);
377 d_rehash_cond(dentry, 0); /* avoid taking dcache_lock inside */
378 spin_unlock(&dcache_lock);
379 spin_unlock(&ll_lookup_lock);
381 CDEBUG(D_DENTRY, "alias dentry %.*s (%p) parent %p inode %p "
382 "refc %d\n", de->d_name.len, de->d_name.name, de,
383 de->d_parent, de->d_inode, atomic_read(&de->d_count));
388 CDEBUG(D_DENTRY, "Reuse disconnected dentry %p inode %p "
389 "refc %d\n", last_discon, last_discon->d_inode,
390 atomic_read(&last_discon->d_count));
391 dget_locked(last_discon);
392 spin_unlock(&dcache_lock);
393 spin_unlock(&ll_lookup_lock);
395 d_move(last_discon, de);
402 spin_unlock(&dcache_lock);
403 spin_unlock(&ll_lookup_lock);
408 static inline void ll_dop_init(struct dentry *de, int *set)
411 if (likely(de->d_op != &ll_d_ops)) {
412 de->d_op = &ll_init_d_ops;
418 static inline void ll_dop_fini(struct dentry *de, int succ)
421 if (likely(de->d_op == &ll_init_d_ops)) {
423 de->d_op = &ll_d_ops;
425 de->d_op = &ll_fini_d_ops;
431 de->d_op = &ll_d_ops;
436 int ll_lookup_it_finish(struct ptlrpc_request *request,
437 struct lookup_intent *it, void *data)
439 struct it_cb_data *icbd = data;
440 struct dentry **de = icbd->icbd_childp;
441 struct inode *parent = icbd->icbd_parent;
442 struct ll_sb_info *sbi = ll_i2sbi(parent);
443 struct inode *inode = NULL;
447 ll_dop_init(*de, &set);
449 /* NB 1 request reference will be taken away by ll_intent_lock()
451 if (!it_disposition(it, DISP_LOOKUP_NEG)) {
452 struct dentry *save = *de;
454 rc = ll_prep_inode(&inode, request, (*de)->d_sb);
461 CDEBUG(D_DLMTRACE, "setting l_data to inode %p (%lu/%u)\n",
462 inode, inode->i_ino, inode->i_generation);
463 md_set_lock_data(sbi->ll_md_exp,
464 &it->d.lustre.it_lock_handle, inode);
466 /* We used to query real size from OSTs here, but actually
467 this is not needed. For stat() calls size would be updated
468 from subsequent do_revalidate()->ll_inode_revalidate_it() in
470 vfs_getattr_it->ll_getattr()->ll_inode_revalidate_it() in 2.6
471 Everybody else who needs correct file size would call
472 ll_glimpse_size or some equivalent themselves anyway.
473 Also see bug 7198. */
475 *de = ll_find_alias(inode, *de);
476 if (set && *de != save)
477 ll_dop_fini(save, 0);
479 /* Check that parent has UPDATE lock. If there is none, we
480 cannot afford to hash this dentry (done by ll_d_add) as it
481 might get picked up later when UPDATE lock will appear */
482 if (ll_have_md_lock(parent, MDS_INODELOCK_UPDATE)) {
483 spin_lock(&dcache_lock);
484 ll_d_add(*de, inode);
485 spin_unlock(&dcache_lock);
487 (*de)->d_inode = NULL;
488 /* We do not want to hash the dentry if don`t have a
489 * lock, but if this dentry is later used in d_move,
490 * we'd hit uninitialised list head d_hash, so we just
491 * do this to init d_hash field but leave dentry
492 * unhashed. (bug 10796). */
505 static struct dentry *ll_lookup_it(struct inode *parent, struct dentry *dentry,
506 struct lookup_intent *it, int lookup_flags)
508 struct lookup_intent lookup_it = { .it_op = IT_LOOKUP };
509 struct dentry *save = dentry, *retval;
510 struct ptlrpc_request *req = NULL;
511 struct md_op_data *op_data;
512 struct it_cb_data icbd;
517 if (dentry->d_name.len > ll_i2sbi(parent)->ll_namelen)
518 RETURN(ERR_PTR(-ENAMETOOLONG));
520 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),intent=%s\n",
521 dentry->d_name.len, dentry->d_name.name, parent->i_ino,
522 parent->i_generation, parent, LL_IT2STR(it));
524 if (d_mountpoint(dentry))
525 CERROR("Tell Peter, lookup on mtpt, it %s\n", LL_IT2STR(it));
527 ll_frob_intent(&it, &lookup_it);
529 /* As do_lookup is called before follow_mount, root dentry may be left
530 * not valid, revalidate it here. */
531 if (parent->i_sb->s_root && (parent->i_sb->s_root->d_inode == parent) &&
532 (it->it_op & (IT_OPEN | IT_CREAT))) {
533 rc = ll_inode_revalidate_it(parent->i_sb->s_root, it);
538 if (it->it_op == IT_GETATTR) {
539 rc = ll_statahead_enter(parent, &dentry, 1);
541 ll_statahead_exit(dentry, rc);
543 RETURN(retval = dentry);
547 icbd.icbd_childp = &dentry;
548 icbd.icbd_parent = parent;
550 if (it->it_op & IT_CREAT ||
551 (it->it_op & IT_OPEN && it->it_create_mode & O_CREAT))
552 opc = LUSTRE_OPC_CREATE;
554 opc = LUSTRE_OPC_ANY;
556 op_data = ll_prep_md_op_data(NULL, parent, NULL, dentry->d_name.name,
557 dentry->d_name.len, lookup_flags, opc,
560 RETURN((void *)op_data);
562 it->it_create_mode &= ~current->fs->umask;
564 rc = md_intent_lock(ll_i2mdexp(parent), op_data, NULL, 0, it,
565 lookup_flags, &req, ll_md_blocking_ast, 0);
566 ll_finish_md_op_data(op_data);
568 GOTO(out, retval = ERR_PTR(rc));
570 rc = ll_lookup_it_finish(req, it, &icbd);
572 ll_intent_release(it);
573 GOTO(out, retval = ERR_PTR(rc));
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);
679 nd->intent.open.file->private_data = it;
680 filp =lookup_instantiate_filp(nd,dentry,
682 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,17))
683 /* 2.6.1[456] have a bug in open_namei() that forgets to check
684 * nd->intent.open.file for error, so we need to return it as lookup's result
689 de = (struct dentry *) filp;
694 #else /* HAVE_FILE_IN_STRUCT_INTENT */
695 /* Release open handle as we have no way to
696 * pass it to ll_file_open */
697 ll_release_openhandle(dentry, it);
698 #endif /* HAVE_FILE_IN_STRUCT_INTENT */
699 } else if (it_disposition(it, DISP_OPEN_CREATE)) {
700 // XXX This can only reliably work on assumption
701 // that there are NO hashed negative dentries.
702 ll_d2d(dentry)->lld_it = it;
703 it = NULL; /* Will be freed in ll_create_nd */
704 /* We absolutely depend on ll_create_nd to be
705 * called to not leak this intent and possible
706 * data attached to it */
711 ll_intent_release(it);
712 OBD_FREE(it, sizeof(*it));
715 de = ll_lookup_it(parent, dentry, NULL, 0);
722 /* We depend on "mode" being set with the proper file type/umask by now */
723 static struct inode *ll_create_node(struct inode *dir, const char *name,
724 int namelen, const void *data, int datalen,
725 int mode, __u64 extra,
726 struct lookup_intent *it)
728 struct inode *inode = NULL;
729 struct ptlrpc_request *request = NULL;
730 struct ll_sb_info *sbi = ll_i2sbi(dir);
734 LASSERT(it && it->d.lustre.it_disposition);
736 LASSERT(it_disposition(it, DISP_ENQ_CREATE_REF));
737 request = it->d.lustre.it_data;
738 it_clear_disposition(it, DISP_ENQ_CREATE_REF);
739 rc = ll_prep_inode(&inode, request, dir->i_sb);
741 GOTO(out, inode = ERR_PTR(rc));
743 LASSERT(list_empty(&inode->i_dentry));
745 /* We asked for a lock on the directory, but were granted a
746 * lock on the inode. Since we finally have an inode pointer,
747 * stuff it in the lock. */
748 CDEBUG(D_DLMTRACE, "setting l_ast_data to inode %p (%lu/%u)\n",
749 inode, inode->i_ino, inode->i_generation);
750 md_set_lock_data(sbi->ll_md_exp,
751 &it->d.lustre.it_lock_handle, inode);
754 ptlrpc_req_finished(request);
759 * By the time this is called, we already have created the directory cache
760 * entry for the new file, but it is so far negative - it has no inode.
762 * We defer creating the OBD object(s) until open, to keep the intent and
763 * non-intent code paths similar, and also because we do not have the MDS
764 * inode number before calling ll_create_node() (which is needed for LOV),
765 * so we would need to do yet another RPC to the MDS to store the LOV EA
766 * data on the MDS. If needed, we would pass the PACKED lmm as data and
767 * lmm_size in datalen (the MDS still has code which will handle that).
769 * If the create succeeds, we fill in the inode information
770 * with d_instantiate().
772 static int ll_create_it(struct inode *dir, struct dentry *dentry, int mode,
773 struct lookup_intent *it)
779 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),intent=%s\n",
780 dentry->d_name.len, dentry->d_name.name, dir->i_ino,
781 dir->i_generation, dir, LL_IT2STR(it));
783 rc = it_open_error(DISP_OPEN_CREATE, it);
787 inode = ll_create_node(dir, dentry->d_name.name, dentry->d_name.len,
788 NULL, 0, mode, 0, it);
790 RETURN(PTR_ERR(inode));
793 d_instantiate(dentry, inode);
794 /* Negative dentry may be unhashed if parent does not have UPDATE lock,
795 * but some callers, e.g. do_coredump, expect dentry to be hashed after
796 * successful create. Hash it here. */
797 spin_lock(&dcache_lock);
798 if (d_unhashed(dentry))
799 d_rehash_cond(dentry, 0);
800 spin_unlock(&dcache_lock);
804 static void ll_update_times(struct ptlrpc_request *request,
807 struct mdt_body *body = req_capsule_server_get(&request->rq_pill,
811 /* mtime is always updated with ctime, but can be set in past.
812 As write and utime(2) may happen within 1 second, and utime's
813 mtime has a priority over write's one, so take mtime from mds
814 for the same ctimes. */
815 if (body->valid & OBD_MD_FLCTIME &&
816 body->ctime >= LTIME_S(inode->i_ctime)) {
817 LTIME_S(inode->i_ctime) = body->ctime;
819 if (body->valid & OBD_MD_FLMTIME) {
820 CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
821 "to "LPU64"\n", inode->i_ino,
822 LTIME_S(inode->i_mtime), body->mtime);
823 LTIME_S(inode->i_mtime) = body->mtime;
828 static int ll_new_node(struct inode *dir, struct qstr *name,
829 const char *tgt, int mode, int rdev,
830 struct dentry *dchild, __u32 opc)
832 struct ptlrpc_request *request = NULL;
833 struct md_op_data *op_data;
834 struct inode *inode = NULL;
835 struct ll_sb_info *sbi = ll_i2sbi(dir);
840 if (unlikely(tgt != NULL))
841 tgt_len = strlen(tgt) + 1;
843 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
844 name->len, 0, opc, NULL);
846 GOTO(err_exit, err = PTR_ERR(op_data));
848 err = md_create(sbi->ll_md_exp, op_data, tgt, tgt_len, mode,
849 current->fsuid, current->fsgid,
850 cfs_curproc_cap_pack(), rdev, &request);
851 ll_finish_md_op_data(op_data);
855 ll_update_times(request, dir);
858 err = ll_prep_inode(&inode, request, dchild->d_sb);
863 d_instantiate(dchild, inode);
867 ptlrpc_req_finished(request);
872 static int ll_mknod_generic(struct inode *dir, struct qstr *name, int mode,
873 unsigned rdev, struct dentry *dchild)
878 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p) mode %o dev %x\n",
879 name->len, name->name, dir->i_ino, dir->i_generation, dir,
882 mode &= ~current->fs->umask;
884 switch (mode & S_IFMT) {
886 mode |= S_IFREG; /* for mode = 0 case, fallthrough */
892 err = ll_new_node(dir, name, NULL, mode, rdev, dchild,
904 #ifndef HAVE_VFS_INTENT_PATCHES
905 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
906 int mode, struct nameidata *nd)
908 struct lookup_intent *it = ll_d2d(dentry)->lld_it;
912 return ll_mknod_generic(dir, &dentry->d_name, mode, 0, dentry);
914 ll_d2d(dentry)->lld_it = NULL;
916 /* Was there an error? Propagate it! */
917 if (it->d.lustre.it_status) {
918 rc = it->d.lustre.it_status;
922 rc = ll_create_it(dir, dentry, mode, it);
923 #ifdef HAVE_FILE_IN_STRUCT_INTENT
924 if (nd && (nd->flags & LOOKUP_OPEN) && dentry->d_inode) { /* Open */
925 nd->intent.open.file->private_data = it;
926 lookup_instantiate_filp(nd, dentry, NULL);
929 ll_release_openhandle(dentry,it);
933 ll_intent_release(it);
934 OBD_FREE(it, sizeof(*it));
939 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
940 int mode, struct nameidata *nd)
942 if (!nd || !nd->intent.d.lustre.it_disposition)
943 /* No saved request? Just mknod the file */
944 return ll_mknod_generic(dir, &dentry->d_name, mode, 0, dentry);
946 return ll_create_it(dir, dentry, mode, &nd->intent);
950 static int ll_symlink_generic(struct inode *dir, struct qstr *name,
951 const char *tgt, struct dentry *dchild)
956 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),target=%.*s\n",
957 name->len, name->name, dir->i_ino, dir->i_generation,
960 err = ll_new_node(dir, name, (char *)tgt, S_IFLNK | S_IRWXUGO,
961 0, dchild, LUSTRE_OPC_SYMLINK);
965 static int ll_link_generic(struct inode *src, struct inode *dir,
966 struct qstr *name, struct dentry *dchild)
968 struct ll_sb_info *sbi = ll_i2sbi(dir);
969 struct ptlrpc_request *request = NULL;
970 struct md_op_data *op_data;
975 "VFS Op: inode=%lu/%u(%p), dir=%lu/%u(%p), target=%.*s\n",
976 src->i_ino, src->i_generation, src, dir->i_ino,
977 dir->i_generation, dir, name->len, name->name);
979 op_data = ll_prep_md_op_data(NULL, src, dir, name->name, name->len,
980 0, LUSTRE_OPC_ANY, NULL);
982 RETURN(PTR_ERR(op_data));
984 err = md_link(sbi->ll_md_exp, op_data, &request);
985 ll_finish_md_op_data(op_data);
991 ll_update_times(request, dir);
994 ptlrpc_req_finished(request);
998 static int ll_mkdir_generic(struct inode *dir, struct qstr *name,
999 int mode, struct dentry *dchild)
1005 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
1006 name->len, name->name, dir->i_ino, dir->i_generation, dir);
1008 mode = (mode & (S_IRWXUGO|S_ISVTX) & ~current->fs->umask) | S_IFDIR;
1009 err = ll_new_node(dir, name, NULL, mode, 0, dchild, LUSTRE_OPC_MKDIR);
1014 /* Try to find the child dentry by its name.
1015 If found, put the result fid into @fid. */
1016 static void ll_get_child_fid(struct inode * dir, struct qstr *name,
1019 struct dentry *parent, *child;
1021 parent = list_entry(dir->i_dentry.next, struct dentry, d_alias);
1022 child = d_lookup(parent, name);
1025 *fid = *ll_inode2fid(child->d_inode);
1030 static int ll_rmdir_generic(struct inode *dir, struct dentry *dparent,
1031 struct dentry *dchild, struct qstr *name)
1033 struct ptlrpc_request *request = NULL;
1034 struct md_op_data *op_data;
1038 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
1039 name->len, name->name, dir->i_ino, dir->i_generation, dir);
1041 if (unlikely(ll_d_mountpoint(dparent, dchild, name)))
1044 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len,
1045 S_IFDIR, LUSTRE_OPC_ANY, NULL);
1046 if (IS_ERR(op_data))
1047 RETURN(PTR_ERR(op_data));
1049 ll_get_child_fid(dir, name, &op_data->op_fid3);
1050 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1051 ll_finish_md_op_data(op_data);
1053 ll_update_times(request, dir);
1054 ptlrpc_req_finished(request);
1058 int ll_objects_destroy(struct ptlrpc_request *request, struct inode *dir)
1060 struct mdt_body *body;
1061 struct lov_mds_md *eadata;
1062 struct lov_stripe_md *lsm = NULL;
1063 struct obd_trans_info oti = { 0 };
1068 /* req is swabbed so this is safe */
1069 body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
1070 if (!(body->valid & OBD_MD_FLEASIZE))
1073 if (body->eadatasize == 0) {
1074 CERROR("OBD_MD_FLEASIZE set but eadatasize zero\n");
1075 GOTO(out, rc = -EPROTO);
1078 /* The MDS sent back the EA because we unlinked the last reference
1079 * to this file. Use this EA to unlink the objects on the OST.
1080 * It's opaque so we don't swab here; we leave it to obd_unpackmd() to
1081 * check it is complete and sensible. */
1082 eadata = req_capsule_server_sized_get(&request->rq_pill, &RMF_MDT_MD,
1084 LASSERT(eadata != NULL);
1086 rc = obd_unpackmd(ll_i2dtexp(dir), &lsm, eadata, body->eadatasize);
1088 CERROR("obd_unpackmd: %d\n", rc);
1091 LASSERT(rc >= sizeof(*lsm));
1093 rc = obd_checkmd(ll_i2dtexp(dir), ll_i2mdexp(dir), lsm);
1095 GOTO(out_free_memmd, rc);
1099 GOTO(out_free_memmd, rc = -ENOMEM);
1101 oa->o_id = lsm->lsm_object_id;
1102 oa->o_gr = lsm->lsm_object_gr;
1103 oa->o_mode = body->mode & S_IFMT;
1104 oa->o_valid = OBD_MD_FLID | OBD_MD_FLTYPE | OBD_MD_FLGROUP;
1106 if (body->valid & OBD_MD_FLCOOKIE) {
1107 oa->o_valid |= OBD_MD_FLCOOKIE;
1108 oti.oti_logcookies =
1109 req_capsule_server_sized_get(&request->rq_pill,
1111 sizeof(struct llog_cookie) *
1112 lsm->lsm_stripe_count);
1113 if (oti.oti_logcookies == NULL) {
1114 oa->o_valid &= ~OBD_MD_FLCOOKIE;
1115 body->valid &= ~OBD_MD_FLCOOKIE;
1119 rc = obd_destroy(ll_i2dtexp(dir), oa, lsm, &oti, ll_i2mdexp(dir));
1122 CERROR("obd destroy objid "LPX64" error %d\n",
1123 lsm->lsm_object_id, rc);
1125 obd_free_memmd(ll_i2dtexp(dir), &lsm);
1130 static int ll_unlink_generic(struct inode *dir, struct dentry *dparent,
1131 struct dentry *dchild, struct qstr *name)
1133 struct ptlrpc_request *request = NULL;
1134 struct md_op_data *op_data;
1137 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
1138 name->len, name->name, dir->i_ino, dir->i_generation, dir);
1141 * XXX: unlink bind mountpoint maybe call to here,
1142 * just check it as vfs_unlink does.
1144 if (unlikely(ll_d_mountpoint(dparent, dchild, name)))
1147 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
1148 name->len, 0, LUSTRE_OPC_ANY, NULL);
1149 if (IS_ERR(op_data))
1150 RETURN(PTR_ERR(op_data));
1152 ll_get_child_fid(dir, name, &op_data->op_fid3);
1153 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1154 ll_finish_md_op_data(op_data);
1158 ll_update_times(request, dir);
1160 rc = ll_objects_destroy(request, dir);
1162 ptlrpc_req_finished(request);
1166 static int ll_rename_generic(struct inode *src, struct dentry *src_dparent,
1167 struct dentry *src_dchild, struct qstr *src_name,
1168 struct inode *tgt, struct dentry *tgt_dparent,
1169 struct dentry *tgt_dchild, struct qstr *tgt_name)
1171 struct ptlrpc_request *request = NULL;
1172 struct ll_sb_info *sbi = ll_i2sbi(src);
1173 struct md_op_data *op_data;
1176 CDEBUG(D_VFSTRACE,"VFS Op:oldname=%.*s,src_dir=%lu/%u(%p),newname=%.*s,"
1177 "tgt_dir=%lu/%u(%p)\n", src_name->len, src_name->name,
1178 src->i_ino, src->i_generation, src, tgt_name->len,
1179 tgt_name->name, tgt->i_ino, tgt->i_generation, tgt);
1181 if (unlikely(ll_d_mountpoint(src_dparent, src_dchild, src_name) ||
1182 ll_d_mountpoint(tgt_dparent, tgt_dchild, tgt_name)))
1185 op_data = ll_prep_md_op_data(NULL, src, tgt, NULL, 0, 0,
1186 LUSTRE_OPC_ANY, NULL);
1187 if (IS_ERR(op_data))
1188 RETURN(PTR_ERR(op_data));
1190 ll_get_child_fid(src, src_name, &op_data->op_fid3);
1191 ll_get_child_fid(tgt, tgt_name, &op_data->op_fid4);
1192 err = md_rename(sbi->ll_md_exp, op_data,
1193 src_name->name, src_name->len,
1194 tgt_name->name, tgt_name->len, &request);
1195 ll_finish_md_op_data(op_data);
1197 ll_update_times(request, src);
1198 ll_update_times(request, tgt);
1199 err = ll_objects_destroy(request, src);
1202 ptlrpc_req_finished(request);
1207 #ifdef HAVE_VFS_INTENT_PATCHES
1208 static int ll_mknod_raw(struct nameidata *nd, int mode, dev_t rdev)
1210 return ll_mknod_generic(nd->dentry->d_inode, &nd->last, mode,rdev,NULL);
1212 static int ll_rename_raw(struct nameidata *srcnd, struct nameidata *tgtnd)
1214 return ll_rename_generic(srcnd->dentry->d_inode, srcnd->dentry,
1216 tgtnd->dentry->d_inode, tgtnd->dentry,
1217 NULL, &tgtnd->last);
1219 static int ll_link_raw(struct nameidata *srcnd, struct nameidata *tgtnd)
1221 return ll_link_generic(srcnd->dentry->d_inode, tgtnd->dentry->d_inode,
1222 &tgtnd->last, NULL);
1224 static int ll_symlink_raw(struct nameidata *nd, const char *tgt)
1226 return ll_symlink_generic(nd->dentry->d_inode, &nd->last, tgt, NULL);
1228 static int ll_rmdir_raw(struct nameidata *nd)
1230 return ll_rmdir_generic(nd->dentry->d_inode, nd->dentry, NULL,
1233 static int ll_mkdir_raw(struct nameidata *nd, int mode)
1235 return ll_mkdir_generic(nd->dentry->d_inode, &nd->last, mode, NULL);
1237 static int ll_unlink_raw(struct nameidata *nd)
1239 return ll_unlink_generic(nd->dentry->d_inode, nd->dentry, NULL,
1244 static int ll_mknod(struct inode *dir, struct dentry *dchild, int mode,
1247 return ll_mknod_generic(dir, &dchild->d_name, mode,
1248 old_encode_dev(rdev), dchild);
1251 static int ll_unlink(struct inode * dir, struct dentry *dentry)
1253 return ll_unlink_generic(dir, NULL, dentry, &dentry->d_name);
1255 static int ll_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1257 return ll_mkdir_generic(dir, &dentry->d_name, mode, dentry);
1259 static int ll_rmdir(struct inode *dir, struct dentry *dentry)
1261 return ll_rmdir_generic(dir, NULL, dentry, &dentry->d_name);
1263 static int ll_symlink(struct inode *dir, struct dentry *dentry,
1264 const char *oldname)
1266 return ll_symlink_generic(dir, &dentry->d_name, oldname, dentry);
1268 static int ll_link(struct dentry *old_dentry, struct inode *dir,
1269 struct dentry *new_dentry)
1271 return ll_link_generic(old_dentry->d_inode, dir, &new_dentry->d_name,
1274 static int ll_rename(struct inode *old_dir, struct dentry *old_dentry,
1275 struct inode *new_dir, struct dentry *new_dentry)
1278 err = ll_rename_generic(old_dir, NULL,
1279 old_dentry, &old_dentry->d_name,
1280 new_dir, NULL, new_dentry,
1281 &new_dentry->d_name);
1283 #ifndef HAVE_FS_RENAME_DOES_D_MOVE
1284 if (!S_ISDIR(old_dentry->d_inode->i_mode))
1286 d_move(old_dentry, new_dentry);
1291 struct inode_operations ll_dir_inode_operations = {
1292 #ifdef HAVE_VFS_INTENT_PATCHES
1293 .link_raw = ll_link_raw,
1294 .unlink_raw = ll_unlink_raw,
1295 .symlink_raw = ll_symlink_raw,
1296 .mkdir_raw = ll_mkdir_raw,
1297 .rmdir_raw = ll_rmdir_raw,
1298 .mknod_raw = ll_mknod_raw,
1299 .rename_raw = ll_rename_raw,
1300 .setattr = ll_setattr,
1301 .setattr_raw = ll_setattr_raw,
1304 .lookup = ll_lookup_nd,
1305 .create = ll_create_nd,
1306 /* We need all these non-raw things for NFSD, to not patch it. */
1307 .unlink = ll_unlink,
1310 .symlink = ll_symlink,
1312 .rename = ll_rename,
1313 .setattr = ll_setattr,
1314 .getattr = ll_getattr,
1315 .permission = ll_inode_permission,
1316 .setxattr = ll_setxattr,
1317 .getxattr = ll_getxattr,
1318 .listxattr = ll_listxattr,
1319 .removexattr = ll_removexattr,
1322 struct inode_operations ll_special_inode_operations = {
1323 #ifdef HAVE_VFS_INTENT_PATCHES
1324 .setattr_raw = ll_setattr_raw,
1326 .setattr = ll_setattr,
1327 .getattr = ll_getattr,
1328 .permission = ll_inode_permission,
1329 .setxattr = ll_setxattr,
1330 .getxattr = ll_getxattr,
1331 .listxattr = ll_listxattr,
1332 .removexattr = ll_removexattr,