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 (c) 2002, 2010, Oracle and/or its affiliates. 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_lite.h>
49 #include <lustre_dlm.h>
50 #include <linux/lustre_version.h>
51 #include "llite_internal.h"
55 int ll_unlock(__u32 mode, struct lustre_handle *lockh)
59 ldlm_lock_decref(lockh, mode);
65 * Flatten 128-bit FID values into a 64-bit value for
66 * use as an inode number. For non-IGIF FIDs this
67 * starts just over 2^32, and continues without conflict
68 * until 2^64, at which point we wrap the high 32 bits
69 * of the SEQ into the range where there may not be many
70 * OID values in use, to minimize the risk of conflict.
72 * The time between re-used inode numbers is very long -
73 * 2^32 SEQ numbers, or about 2^32 client mounts. */
74 static inline __u64 fid_flatten(const struct lu_fid *fid)
79 if (fid_is_igif(fid)) {
80 ino = lu_igif_ino(fid);
86 ino = (seq << 24) + ((seq >> (64-8)) & 0xffffff0000ULL) + fid_oid(fid);
88 RETURN(ino ? ino : fid_oid(fid));
92 * map fid to 32 bit value for ino on 32bit systems. */
93 static inline __u32 fid_flatten32(const struct lu_fid *fid)
98 if (fid_is_igif(fid)) {
99 ino = lu_igif_ino(fid);
103 seq = fid_seq(fid) - FID_SEQ_START;
106 map the high bits of the OID into higher bits of the inode number so that
107 inodes generated at about the same time have a reduced chance of collisions.
108 This will give a period of 1024 clients and 128 k = 128M inodes without collisions.
111 ino = ((seq & 0x000fffffULL) << 12) + ((seq >> 8) & 0xfffff000) +
112 (seq >> (64 - (40-8)) & 0xffffff00) +
113 (fid_oid(fid) & 0xff000fff) + ((fid_oid(fid) & 0x00fff000) << 16);
115 RETURN(ino ? ino : fid_oid(fid));
119 * for 32 bit inode numbers directly map seq+oid to 32bit number.
121 __u32 ll_fid_build_ino32(const struct ll_fid *fid)
123 RETURN(fid_flatten32((struct lu_fid *)fid));
127 * build inode number from passed @fid */
128 __u64 ll_fid_build_ino(const struct ll_fid *fid)
130 #if BITS_PER_LONG == 32
131 RETURN(fid_flatten32((struct lu_fid *)fid));
133 RETURN(fid_flatten((struct lu_fid *)fid));
137 __u32 ll_fid_build_gen(struct ll_sb_info *sbi, struct ll_fid *fid)
142 if (fid_is_igif((struct lu_fid*)fid)) {
143 gen = lu_igif_gen((struct lu_fid*)fid);
148 /* called from iget5_locked->find_inode() under inode_lock spinlock */
149 static int fid_test_inode(struct inode *inode, void *opaque)
151 struct lustre_md *md = opaque;
152 struct lu_fid *fid = (struct lu_fid*)&md->body->fid1;
154 if (unlikely(!(md->body->valid & OBD_MD_FLID))) {
155 CERROR("MDS body missing FID\n");
159 return fid_seq(ll_inode_lu_fid(inode)) == fid_seq(fid) &&
160 fid_oid(ll_inode_lu_fid(inode)) == fid_oid(fid);
163 static int fid_set_inode(struct inode *inode, void *opaque)
165 struct lustre_md *md = opaque;
167 *ll_inode_lu_fid(inode) = *((struct lu_fid*)&md->body->fid1);
171 struct inode *ll_iget(struct super_block *sb, ino_t hash,
172 struct lustre_md *md)
178 inode = iget5_locked(sb, hash, fid_test_inode, fid_set_inode, md);
181 if (inode->i_state & I_NEW) {
182 ll_read_inode2(inode, md);
183 unlock_new_inode(inode);
185 if (!(inode->i_state & (I_FREEING | I_CLEAR)))
186 ll_update_inode(inode, md);
188 CDEBUG(D_VFSTRACE, "got inode: %lu/%u(%p) for "DFID"\n",
189 inode->i_ino, inode->i_generation, inode,
190 PFID(ll_inode_lu_fid(inode)));
196 static void ll_drop_negative_dentry(struct inode *dir)
198 struct dentry *dentry, *tmp_alias, *tmp_subdir;
200 spin_lock(&ll_lookup_lock);
201 spin_lock(&dcache_lock);
203 list_for_each_entry_safe(dentry, tmp_alias,
204 &dir->i_dentry,d_alias) {
205 if (!list_empty(&dentry->d_subdirs)) {
206 struct dentry *child;
207 list_for_each_entry_safe(child, tmp_subdir,
210 /* XXX Print some debug here? */
212 /* Negative dentry. If we were
213 dropping dcache lock, go
214 throught the list again */
215 if (ll_drop_dentry(child))
220 spin_unlock(&dcache_lock);
221 spin_unlock(&ll_lookup_lock);
224 int ll_mdc_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
225 void *data, int flag)
228 struct lustre_handle lockh;
232 case LDLM_CB_BLOCKING:
233 ldlm_lock2handle(lock, &lockh);
234 rc = ldlm_cli_cancel(&lockh);
236 CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
240 case LDLM_CB_CANCELING: {
241 struct inode *inode = ll_inode_from_lock(lock);
242 __u64 bits = lock->l_policy_data.l_inodebits.bits;
244 ldlm_mode_t mode = lock->l_req_mode;
246 /* Invalidate all dentries associated with this inode */
250 fid = ll_inode_lu_fid(inode);;
252 LASSERT(lock->l_flags & LDLM_FL_CANCELING);
253 /* For OPEN locks we differentiate between lock modes - CR, CW. PR - bug 22891 */
254 if ((bits & MDS_INODELOCK_LOOKUP) &&
255 ll_have_md_lock(inode, MDS_INODELOCK_LOOKUP, LCK_MINMODE))
256 bits &= ~MDS_INODELOCK_LOOKUP;
257 if ((bits & MDS_INODELOCK_UPDATE) &&
258 ll_have_md_lock(inode, MDS_INODELOCK_UPDATE, LCK_MINMODE))
259 bits &= ~MDS_INODELOCK_UPDATE;
260 if ((bits & MDS_INODELOCK_OPEN) &&
261 ll_have_md_lock(inode, MDS_INODELOCK_OPEN, mode))
262 bits &= ~MDS_INODELOCK_OPEN;
264 if (!fid_res_name_eq(fid, &lock->l_resource->lr_name)) {
265 LDLM_ERROR(lock, "data mismatch with ino %lu/%u (%p)",
266 inode->i_ino, inode->i_generation, inode);
269 if (bits & MDS_INODELOCK_OPEN) {
271 switch (lock->l_req_mode) {
278 CERROR("open PR lock without FMODE_EXEC\n");
284 CERROR("Unexpected lock mode for OPEN lock "
285 "%d, inode %ld\n", lock->l_req_mode,
288 ll_mdc_real_close(inode, flags);
291 if (bits & MDS_INODELOCK_UPDATE)
292 clear_bit(LLI_F_HAVE_MDS_SIZE_LOCK,
293 &(ll_i2info(inode)->lli_flags));
295 if (S_ISDIR(inode->i_mode) &&
296 (bits & MDS_INODELOCK_UPDATE)) {
297 CDEBUG(D_INODE, "invalidating inode %lu\n",
299 truncate_inode_pages(inode->i_mapping, 0);
300 ll_drop_negative_dentry(inode);
301 inode->i_version++; /* XXX: remove with inode version*/
304 if (inode->i_sb->s_root &&
305 inode != inode->i_sb->s_root->d_inode &&
306 (bits & MDS_INODELOCK_LOOKUP))
307 ll_unhash_aliases(inode);
318 int ll_mdc_cancel_unused(struct lustre_handle *conn, struct inode *inode,
319 int flags, void *opaque)
321 struct ldlm_res_id res_id;
322 struct obd_device *obddev = class_conn2obd(conn);
325 fid_build_reg_res_name(ll_inode_lu_fid(inode), &res_id);
326 RETURN(ldlm_cli_cancel_unused(obddev->obd_namespace, &res_id, flags,
330 /* Pack the required supplementary groups into the supplied groups array.
331 * If we don't need to use the groups from the target inode(s) then we
332 * instead pack one or more groups from the user's supplementary group
333 * array in case it might be useful. Not needed if doing an MDS-side upcall. */
334 void ll_i2gids(__u32 *suppgids, struct inode *i1, struct inode *i2)
339 LASSERT(suppgids != NULL);
341 if (in_group_p(i1->i_gid))
342 suppgids[0] = i1->i_gid;
347 if (in_group_p(i2->i_gid))
348 suppgids[1] = i2->i_gid;
355 for (i = 0; i < current_ngroups; i++) {
356 if (suppgids[0] == -1) {
357 if (current_groups[i] != suppgids[1])
358 suppgids[0] = current_groups[i];
361 if (suppgids[1] == -1) {
362 if (current_groups[i] != suppgids[0])
363 suppgids[1] = current_groups[i];
370 int ll_prepare_mdc_op_data(struct mdc_op_data *op_data, struct inode *i1,
371 struct inode *i2, const char *name, int namelen,
372 int mode, void *data)
376 if (namelen > ll_i2sbi(i1)->ll_namelen)
377 return -ENAMETOOLONG;
378 ll_i2gids(op_data->suppgids, i1, i2);
379 ll_inode2fid(&op_data->fid1, i1);
382 ll_inode2fid(&op_data->fid2, i2);
384 memset(&op_data->fid2, 0, sizeof(op_data->fid2));
386 op_data->name = name;
387 op_data->namelen = namelen;
388 op_data->create_mode = mode;
389 op_data->mod_time = CURRENT_SECONDS;
390 op_data->data = data;
395 static void ll_d_add(struct dentry *de, struct inode *inode)
397 CDEBUG(D_DENTRY, "adding inode %p to dentry %p\n", inode, de);
399 if (!list_empty(&de->d_alias)) {
400 spin_unlock(&dcache_lock);
401 CERROR("dentry %.*s %p alias next %p, prev %p\n",
402 de->d_name.len, de->d_name.name, de,
403 de->d_alias.next, de->d_alias.prev);
407 list_add(&de->d_alias, &inode->i_dentry);
411 if (!d_unhashed(de)) {
412 spin_unlock(&dcache_lock);
413 CERROR("dentry %.*s %p hash next %p\n",
414 de->d_name.len, de->d_name.name, de, de->d_hash.next);
417 d_rehash_cond(de, 0);
420 /* Search "inode"'s alias list for a dentry that has the same name and parent
421 * as de. If found, return it. If not found, return de.
422 * Lustre can't use d_add_unique because don't unhash aliases for directory
423 * in ll_revalidate_it. After revaliadate inode will be have hashed aliases
424 * and it triggers BUG_ON in d_instantiate_unique (bug #10954).
426 static struct dentry *ll_find_alias(struct inode *inode, struct dentry *de)
428 struct list_head *tmp;
429 struct dentry *dentry;
430 struct dentry *last_discon = NULL;
432 spin_lock(&ll_lookup_lock);
433 spin_lock(&dcache_lock);
434 list_for_each(tmp, &inode->i_dentry) {
435 dentry = list_entry(tmp, struct dentry, d_alias);
437 /* We are called here with 'de' already on the aliases list. */
438 if (unlikely(dentry == de)) {
443 if (dentry->d_flags & DCACHE_DISCONNECTED) {
444 /* LASSERT(last_discon == NULL); see bug 20055 */
445 last_discon = dentry;
449 if (dentry->d_parent != de->d_parent)
452 if (dentry->d_name.hash != de->d_name.hash)
455 if (dentry->d_name.len != de->d_name.len)
458 if (memcmp(dentry->d_name.name, de->d_name.name,
459 de->d_name.len) != 0)
465 unlock_dentry(dentry);
466 ll_dops_init(dentry, 0);
467 d_rehash_cond(dentry, 0); /* avoid taking dcache_lock inside */
468 spin_unlock(&dcache_lock);
469 spin_unlock(&ll_lookup_lock);
471 CDEBUG(D_DENTRY, "alias dentry %.*s (%p) parent %p inode %p "
472 "refc %d\n", de->d_name.len, de->d_name.name, de,
473 de->d_parent, de->d_inode, atomic_read(&de->d_count));
478 CDEBUG(D_DENTRY, "Reuse disconnected dentry %p inode %p "
479 "refc %d\n", last_discon, last_discon->d_inode,
480 atomic_read(&last_discon->d_count));
481 dget_locked(last_discon);
482 last_discon->d_flags |= DCACHE_LUSTRE_INVALID;
483 spin_unlock(&dcache_lock);
484 spin_unlock(&ll_lookup_lock);
485 ll_dops_init(last_discon, 1);
487 d_move(last_discon, de);
492 de->d_flags |= DCACHE_LUSTRE_INVALID;
495 spin_unlock(&dcache_lock);
496 spin_unlock(&ll_lookup_lock);
501 int lookup_it_finish(struct ptlrpc_request *request, int offset,
502 struct lookup_intent *it, void *data)
504 struct it_cb_data *icbd = data;
505 struct dentry **de = icbd->icbd_childp;
506 struct inode *parent = icbd->icbd_parent;
507 struct ll_sb_info *sbi = ll_i2sbi(parent);
508 struct inode *inode = NULL;
512 /* NB 1 request reference will be taken away by ll_intent_lock()
514 if (!it_disposition(it, DISP_LOOKUP_NEG)) {
515 struct dentry *save = *de;
518 rc = ll_prep_inode(sbi->ll_osc_exp, &inode, request, offset,
523 CDEBUG(D_DLMTRACE, "setting l_data to inode %p (%lu/%u)\n",
524 inode, inode->i_ino, inode->i_generation);
525 mdc_set_lock_data(&it->d.lustre.it_lock_handle, inode, &bits);
527 /* We used to query real size from OSTs here, but actually
528 this is not needed. For stat() calls size would be updated
529 from subsequent do_revalidate()->ll_inode_revalidate_it() in
531 vfs_getattr_it->ll_getattr()->ll_inode_revalidate_it() in 2.6
532 Everybody else who needs correct file size would call
533 ll_glimpse_size or some equivalent themselves anyway.
534 Also see bug 7198. */
536 ll_dops_init(*de, 1);
537 *de = ll_find_alias(inode, *de);
539 struct ll_dentry_data *lld = ll_d2d(*de);
541 /* just make sure the ll_dentry_data is ready */
542 if (unlikely(lld == NULL)) {
545 if (likely(lld != NULL))
546 lld->lld_sa_generation = 0;
549 /* we have lookup look - unhide dentry */
550 if (bits & MDS_INODELOCK_LOOKUP) {
552 (*de)->d_flags &= ~(DCACHE_LUSTRE_INVALID);
556 ll_dops_init(*de, 1);
557 /* Check that parent has UPDATE lock. If there is none, we
558 cannot afford to hash this dentry (done by ll_d_add) as it
559 might get picked up later when UPDATE lock will appear */
560 if (ll_have_md_lock(parent, MDS_INODELOCK_UPDATE, LCK_MINMODE)) {
561 spin_lock(&dcache_lock);
562 ll_d_add(*de, inode);
563 spin_unlock(&dcache_lock);
565 /* negative lookup - and don't have update lock to
568 (*de)->d_flags |= DCACHE_LUSTRE_INVALID;
571 (*de)->d_inode = NULL;
572 /* We do not want to hash the dentry if don`t have a
573 * lock, but if this dentry is later used in d_move,
574 * we'd hit uninitialised list head d_hash, so we just
575 * do this to init d_hash field but leave dentry
576 * unhashed. (bug 10796). */
585 static struct dentry *ll_lookup_it(struct inode *parent, struct dentry *dentry,
586 struct lookup_intent *it, int lookup_flags)
588 struct dentry *save = dentry, *retval;
589 struct mdc_op_data op_data = { { 0 } };
590 struct it_cb_data icbd;
591 struct ptlrpc_request *req = NULL;
592 struct lookup_intent lookup_it = { .it_op = IT_LOOKUP };
596 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),intent=%s\n",
597 dentry->d_name.len, dentry->d_name.name, parent->i_ino,
598 parent->i_generation, parent, LL_IT2STR(it));
600 if (d_mountpoint(dentry))
601 CERROR("Tell Peter, lookup on mtpt, it %s\n", LL_IT2STR(it));
603 ll_frob_intent(&it, &lookup_it);
605 /* As do_lookup is called before follow_mount, root dentry may be left
606 * not valid, revalidate it here. */
607 if (parent->i_sb->s_root && (parent->i_sb->s_root->d_inode == parent) &&
608 (it->it_op & (IT_OPEN | IT_CREAT))) {
609 rc = ll_inode_revalidate_it(parent->i_sb->s_root, it);
614 if (it->it_op == IT_GETATTR) {
615 first = ll_statahead_enter(parent, &dentry, 1);
617 ll_statahead_exit(parent, dentry, first);
619 RETURN(retval = dentry);
623 icbd.icbd_parent = parent;
624 icbd.icbd_childp = &dentry;
626 rc = ll_prepare_mdc_op_data(&op_data, parent, NULL, dentry->d_name.name,
627 dentry->d_name.len, lookup_flags, NULL);
631 it->it_create_mode &= ~current->fs->umask;
633 rc = mdc_intent_lock(ll_i2mdcexp(parent), &op_data, NULL, 0, it,
634 lookup_flags, &req, ll_mdc_blocking_ast, 0);
637 GOTO(out, retval = ERR_PTR(rc));
639 rc = lookup_it_finish(req, DLM_REPLY_REC_OFF, it, &icbd);
641 ll_intent_release(it);
642 GOTO(out, retval = ERR_PTR(rc));
645 if (first == -EEXIST)
646 ll_statahead_mark(parent, dentry);
648 if ((it->it_op & IT_OPEN) && dentry->d_inode &&
649 !S_ISREG(dentry->d_inode->i_mode) &&
650 !S_ISDIR(dentry->d_inode->i_mode)) {
651 ll_release_openhandle(dentry, it);
653 ll_lookup_finish_locks(it, dentry);
656 GOTO(out, retval = NULL);
658 GOTO(out, retval = dentry);
661 ptlrpc_req_finished(req);
665 #ifdef HAVE_VFS_INTENT_PATCHES
666 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
667 struct nameidata *nd)
672 if (nd && nd->flags & LOOKUP_LAST && !(nd->flags & LOOKUP_LINK_NOTLAST))
673 de = ll_lookup_it(parent, dentry, &nd->intent, nd->flags);
675 de = ll_lookup_it(parent, dentry, NULL, 0);
680 struct lookup_intent *ll_convert_intent(struct open_intent *oit,
683 struct lookup_intent *it;
685 OBD_ALLOC(it, sizeof(*it));
687 return ERR_PTR(-ENOMEM);
689 if (lookup_flags & LOOKUP_OPEN) {
691 if (lookup_flags & LOOKUP_CREATE)
692 it->it_op |= IT_CREAT;
693 it->it_create_mode = oit->create_mode;
694 it->it_flags = oit->flags;
696 it->it_op = IT_GETATTR;
699 #ifndef HAVE_FILE_IN_STRUCT_INTENT
700 /* Since there is no way to pass our intent to ll_file_open,
701 * just check the file is there. Actual open will be done
703 if (it->it_op & IT_OPEN)
704 it->it_op = IT_LOOKUP;
710 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
711 struct nameidata *nd)
716 if (nd && !(nd->flags & (LOOKUP_CONTINUE|LOOKUP_PARENT))) {
717 struct lookup_intent *it;
719 #if defined(HAVE_FILE_IN_STRUCT_INTENT) && (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,17))
720 /* Did we came here from failed revalidate just to propagate
722 if (nd->flags & LOOKUP_OPEN)
723 if (IS_ERR(nd->intent.open.file))
724 RETURN((struct dentry *)nd->intent.open.file);
727 if (ll_d2d(dentry) && ll_d2d(dentry)->lld_it) {
728 it = ll_d2d(dentry)->lld_it;
729 ll_d2d(dentry)->lld_it = NULL;
731 if ((nd->flags & LOOKUP_CREATE ) && !(nd->flags & LOOKUP_OPEN)) {
732 /* We are sure this is new dentry, so we need to create
733 our private data and set the dentry ops */
734 ll_dops_init(dentry, 1);
737 it = ll_convert_intent(&nd->intent.open, nd->flags);
739 RETURN((struct dentry *)it);
742 de = ll_lookup_it(parent, dentry, it, nd->flags);
745 if ((nd->flags & LOOKUP_OPEN) && !IS_ERR(dentry)) { /* Open */
746 if (dentry->d_inode &&
747 it_disposition(it, DISP_OPEN_OPEN)) { /* nocreate */
748 #ifdef HAVE_FILE_IN_STRUCT_INTENT
749 if (S_ISFIFO(dentry->d_inode->i_mode)) {
750 // We cannot call open here as it would
753 (struct ptlrpc_request *)
754 it->d.lustre.it_data);
756 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,17))
757 /* 2.6.1[456] have a bug in open_namei() that forgets to check
758 * nd->intent.open.file for error, so we need to return it as lookup's result
761 nd->intent.open.file->private_data = it;
762 filp =lookup_instantiate_filp(nd,dentry,
767 de = (struct dentry *) filp;
770 nd->intent.open.file->private_data = it;
771 (void)lookup_instantiate_filp(nd,dentry,
776 #else /* HAVE_FILE_IN_STRUCT_INTENT */
777 /* Release open handle as we have no way to
778 * pass it to ll_file_open */
779 ll_release_openhandle(dentry, it);
780 #endif /* HAVE_FILE_IN_STRUCT_INTENT */
781 } else if (it_disposition(it, DISP_OPEN_CREATE)) {
782 // XXX This can only reliably work on assumption
783 // that there are NO hashed negative dentries.
784 ll_d2d(dentry)->lld_it = it;
785 it = NULL; /* Will be freed in ll_create_nd */
786 /* We absolutely depend on ll_create_nd to be
787 * called to not leak this intent and possible
788 * data attached to it */
793 ll_intent_release(it);
794 OBD_FREE(it, sizeof(*it));
797 de = ll_lookup_it(parent, dentry, NULL, 0);
804 /* We depend on "mode" being set with the proper file type/umask by now */
805 static struct inode *ll_create_node(struct inode *dir, const char *name,
806 int namelen, const void *data, int datalen,
807 int mode, __u64 extra,
808 struct lookup_intent *it)
810 struct inode *inode = NULL;
811 struct ptlrpc_request *request = NULL;
812 struct ll_sb_info *sbi = ll_i2sbi(dir);
816 LASSERT(it && it->d.lustre.it_disposition);
818 LASSERT(it_disposition(it, DISP_ENQ_CREATE_REF));
819 request = it->d.lustre.it_data;
820 it_clear_disposition(it, DISP_ENQ_CREATE_REF);
821 rc = ll_prep_inode(sbi->ll_osc_exp, &inode, request, DLM_REPLY_REC_OFF,
824 GOTO(out, inode = ERR_PTR(rc));
826 LASSERT(list_empty(&inode->i_dentry));
828 /* We asked for a lock on the directory, but were granted a
829 * lock on the inode. Since we finally have an inode pointer,
830 * stuff it in the lock. */
831 CDEBUG(D_DLMTRACE, "setting l_ast_data to inode %p (%lu/%u)\n",
832 inode, inode->i_ino, inode->i_generation);
833 mdc_set_lock_data(&it->d.lustre.it_lock_handle, inode, NULL);
836 ptlrpc_req_finished(request);
841 * By the time this is called, we already have created the directory cache
842 * entry for the new file, but it is so far negative - it has no inode.
844 * We defer creating the OBD object(s) until open, to keep the intent and
845 * non-intent code paths similar, and also because we do not have the MDS
846 * inode number before calling ll_create_node() (which is needed for LOV),
847 * so we would need to do yet another RPC to the MDS to store the LOV EA
848 * data on the MDS. If needed, we would pass the PACKED lmm as data and
849 * lmm_size in datalen (the MDS still has code which will handle that).
851 * If the create succeeds, we fill in the inode information
852 * with d_instantiate().
854 static int ll_create_it(struct inode *dir, struct dentry *dentry, int mode,
855 struct lookup_intent *it)
858 struct ptlrpc_request *request = it->d.lustre.it_data;
862 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),intent=%s\n",
863 dentry->d_name.len, dentry->d_name.name, dir->i_ino,
864 dir->i_generation, dir, LL_IT2STR(it));
866 rc = it_open_error(DISP_OPEN_CREATE, it);
870 mdc_store_inode_generation(request, DLM_INTENT_REC_OFF,
872 inode = ll_create_node(dir, dentry->d_name.name, dentry->d_name.len,
873 NULL, 0, mode, 0, it);
875 RETURN(PTR_ERR(inode));
878 d_instantiate(dentry, inode);
879 /* Negative dentry may be unhashed if parent does not have UPDATE lock,
880 * but some callers, e.g. do_coredump, expect dentry to be hashed after
881 * successful create. Hash it here. */
882 spin_lock(&dcache_lock);
883 if (d_unhashed(dentry))
884 d_rehash_cond(dentry, 0);
885 spin_unlock(&dcache_lock);
889 static void ll_update_times(struct ptlrpc_request *request, int offset,
892 struct mds_body *body = lustre_msg_buf(request->rq_repmsg, offset,
896 if (body->valid & OBD_MD_FLMTIME &&
897 body->mtime > LTIME_S(inode->i_mtime)) {
898 CDEBUG(D_INODE, "setting ino %lu mtime from %lu to "LPU64"\n",
899 inode->i_ino, LTIME_S(inode->i_mtime), body->mtime);
900 LTIME_S(inode->i_mtime) = body->mtime;
902 if (body->valid & OBD_MD_FLCTIME &&
903 body->ctime > LTIME_S(inode->i_ctime))
904 LTIME_S(inode->i_ctime) = body->ctime;
907 static int ll_new_node(struct inode *dir, struct qstr *name,
908 const char *tgt, int mode,
909 int rdev, struct dentry *dchild)
911 struct ptlrpc_request *request = NULL;
912 struct inode *inode = NULL;
913 struct ll_sb_info *sbi = ll_i2sbi(dir);
914 struct mdc_op_data op_data = { { 0 } };
919 if (unlikely(tgt != NULL))
920 tgt_len = strlen(tgt)+1;
922 err = ll_prepare_mdc_op_data(&op_data, dir, NULL, name->name,
927 err = mdc_create(sbi->ll_mdc_exp, &op_data, tgt, tgt_len,
928 mode, cfs_curproc_fsuid(), cfs_curproc_fsgid(),
929 cfs_curproc_cap_pack(), rdev, &request);
933 ll_update_times(request, REPLY_REC_OFF, dir);
936 err = ll_prep_inode(sbi->ll_osc_exp, &inode, request,
937 REPLY_REC_OFF, dchild->d_sb);
942 d_instantiate(dchild, inode);
946 ptlrpc_req_finished(request);
952 static int ll_mknod_generic(struct inode *dir, struct qstr *name, int mode,
953 unsigned rdev, struct dentry *dchild)
958 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p) mode %o dev %x\n",
959 name->len, name->name, dir->i_ino, dir->i_generation, dir,
962 mode &= ~current->fs->umask;
964 switch (mode & S_IFMT) {
966 mode |= S_IFREG; /* for mode = 0 case, fallthrough */
972 err = ll_new_node(dir, name, NULL, mode, rdev, dchild);
983 #ifndef HAVE_VFS_INTENT_PATCHES
984 static int ll_create_nd(struct inode *dir, struct dentry *dentry, int mode, struct nameidata *nd)
986 struct lookup_intent *it = ll_d2d(dentry)->lld_it;
990 return ll_mknod_generic(dir, &dentry->d_name, mode, 0, dentry);
992 ll_d2d(dentry)->lld_it = NULL;
994 /* Was there an error? Propagate it! */
995 if (it->d.lustre.it_status) {
996 rc = it->d.lustre.it_status;
1000 rc = ll_create_it(dir, dentry, mode, it);
1001 #ifdef HAVE_FILE_IN_STRUCT_INTENT
1002 if (nd && (nd->flags & LOOKUP_OPEN) && dentry->d_inode) { /* Open */
1003 nd->intent.open.file->private_data = it;
1004 lookup_instantiate_filp(nd, dentry, NULL);
1007 ll_release_openhandle(dentry,it);
1011 ll_intent_release(it);
1012 OBD_FREE(it, sizeof(*it));
1017 static int ll_create_nd(struct inode *dir, struct dentry *dentry, int mode, struct nameidata *nd)
1020 if (!nd || !nd->intent.d.lustre.it_disposition)
1021 /* No saved request? Just mknod the file */
1022 return ll_mknod_generic(dir, &dentry->d_name, mode, 0, dentry);
1024 return ll_create_it(dir, dentry, mode, &nd->intent);
1028 static int ll_symlink_generic(struct inode *dir, struct qstr *name,
1029 const char *tgt, struct dentry *dchild)
1034 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),target=%.*s\n",
1035 name->len, name->name, dir->i_ino, dir->i_generation,
1038 err = ll_new_node(dir, name, (char *)tgt, S_IFLNK | S_IRWXUGO,
1043 static int ll_link_generic(struct inode *src, struct inode *dir,
1044 struct qstr *name, struct dentry *dchild)
1046 struct ptlrpc_request *request = NULL;
1047 struct mdc_op_data op_data = { { 0 } };
1049 struct ll_sb_info *sbi = ll_i2sbi(dir);
1053 "VFS Op: inode=%lu/%u(%p), dir=%lu/%u(%p), target=%.*s\n",
1054 src->i_ino, src->i_generation, src, dir->i_ino,
1055 dir->i_generation, dir, name->len, name->name);
1057 err = ll_prepare_mdc_op_data(&op_data, src, dir, name->name,
1058 name->len, 0, NULL);
1061 err = mdc_link(sbi->ll_mdc_exp, &op_data, &request);
1068 ll_update_times(request, REPLY_REC_OFF, dir);
1072 ptlrpc_req_finished(request);
1076 static int ll_mkdir_generic(struct inode *dir, struct qstr *name, int mode,
1077 struct dentry *dchild)
1083 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
1084 name->len, name->name, dir->i_ino, dir->i_generation, dir);
1086 mode = (mode & (S_IRWXUGO|S_ISVTX) & ~current->fs->umask) | S_IFDIR;
1087 err = ll_new_node(dir, name, NULL, mode, 0, dchild);
1092 /* Try to find the child dentry by its name.
1093 If found, put the result fid into @fid. */
1094 static void ll_get_child_fid(struct inode * dir, struct qstr *name,
1097 struct dentry *parent, *child;
1099 parent = list_entry(dir->i_dentry.next, struct dentry, d_alias);
1100 child = d_lookup(parent, name);
1103 ll_inode2fid(fid, child->d_inode);
1108 static int ll_rmdir_generic(struct inode *dir, struct dentry *dparent,
1111 struct ptlrpc_request *request = NULL;
1112 struct mdc_op_data op_data = { { 0 } };
1113 struct dentry *dentry;
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);
1119 /* Check if we have something mounted at the dir we are going to delete
1120 * In such a case there would always be dentry present. */
1122 dentry = d_lookup(dparent, name);
1124 int mounted = d_mountpoint(dentry);
1127 GOTO(out, rc = -EBUSY);
1131 rc = ll_prepare_mdc_op_data(&op_data, dir, NULL, name->name,
1132 name->len, S_IFDIR, NULL);
1136 ll_get_child_fid(dir, name, &op_data.fid3);
1137 rc = mdc_unlink(ll_i2sbi(dir)->ll_mdc_exp, &op_data, &request);
1140 ll_update_times(request, REPLY_REC_OFF, dir);
1144 ptlrpc_req_finished(request);
1148 int ll_objects_destroy(struct ptlrpc_request *request, struct inode *dir)
1150 struct mds_body *body;
1151 struct lov_mds_md *eadata;
1152 struct lov_stripe_md *lsm = NULL;
1153 struct obd_trans_info oti = { 0 };
1158 /* req is swabbed so this is safe */
1159 body = lustre_msg_buf(request->rq_repmsg, REPLY_REC_OFF, sizeof(*body));
1161 if (!(body->valid & OBD_MD_FLEASIZE))
1164 if (body->eadatasize == 0) {
1165 CERROR("OBD_MD_FLEASIZE set but eadatasize zero\n");
1166 GOTO(out, rc = -EPROTO);
1169 /* The MDS sent back the EA because we unlinked the last reference
1170 * to this file. Use this EA to unlink the objects on the OST.
1171 * It's opaque so we don't swab here; we leave it to obd_unpackmd() to
1172 * check it is complete and sensible. */
1173 eadata = lustre_swab_repbuf(request, REPLY_REC_OFF + 1,
1174 body->eadatasize, NULL);
1175 LASSERT(eadata != NULL);
1176 if (eadata == NULL) {
1177 CERROR("Can't unpack MDS EA data\n");
1178 GOTO(out, rc = -EPROTO);
1181 rc = obd_unpackmd(ll_i2obdexp(dir), &lsm, eadata, body->eadatasize);
1183 CERROR("obd_unpackmd: %d\n", rc);
1186 LASSERT(rc >= sizeof(*lsm));
1188 rc = obd_checkmd(ll_i2obdexp(dir), ll_i2mdcexp(dir), lsm);
1190 GOTO(out_free_memmd, rc);
1194 GOTO(out_free_memmd, rc = -ENOMEM);
1196 oa->o_id = lsm->lsm_object_id;
1197 oa->o_gr = lsm->lsm_object_gr;
1198 oa->o_mode = body->mode & S_IFMT;
1199 oa->o_valid = OBD_MD_FLID | OBD_MD_FLGROUP | OBD_MD_FLTYPE;
1201 if (body->valid & OBD_MD_FLCOOKIE) {
1202 oa->o_valid |= OBD_MD_FLCOOKIE;
1203 oti.oti_logcookies =
1204 lustre_msg_buf(request->rq_repmsg, REPLY_REC_OFF + 2,
1205 sizeof(struct llog_cookie) *
1206 lsm->lsm_stripe_count);
1207 if (oti.oti_logcookies == NULL) {
1208 oa->o_valid &= ~OBD_MD_FLCOOKIE;
1209 body->valid &= ~OBD_MD_FLCOOKIE;
1213 rc = obd_destroy(ll_i2obdexp(dir), oa, lsm, &oti, ll_i2mdcexp(dir));
1216 CERROR("obd destroy objid "LPX64"@"LPX64" error %d\n",
1217 lsm->lsm_object_id, lsm->lsm_object_gr, rc);
1219 obd_free_memmd(ll_i2obdexp(dir), &lsm);
1224 /* ll_unlink_generic() doesn't update the inode with the new link count.
1225 * Instead, ll_ddelete() and ll_d_iput() will update it based upon if there
1226 * is any lock existing. They will recycle dentries and inodes based upon locks
1228 static int ll_unlink_generic(struct inode * dir, struct qstr *name)
1230 struct ptlrpc_request *request = NULL;
1231 struct mdc_op_data op_data = { { 0 } };
1235 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
1236 name->len, name->name, dir->i_ino, dir->i_generation, dir);
1238 rc = ll_prepare_mdc_op_data(&op_data, dir, NULL, name->name,
1239 name->len, 0, NULL);
1243 ll_get_child_fid(dir, name, &op_data.fid3);
1244 rc = mdc_unlink(ll_i2sbi(dir)->ll_mdc_exp, &op_data, &request);
1248 ll_update_times(request, REPLY_REC_OFF, dir);
1250 rc = ll_objects_destroy(request, dir);
1255 ptlrpc_req_finished(request);
1259 static int ll_rename_generic(struct inode *src, struct qstr *src_name,
1260 struct inode *tgt, struct qstr *tgt_name)
1262 struct ptlrpc_request *request = NULL;
1263 struct ll_sb_info *sbi = ll_i2sbi(src);
1264 struct mdc_op_data op_data = { { 0 } };
1268 CDEBUG(D_VFSTRACE,"VFS Op:oldname=%.*s,src_dir=%lu/%u(%p),newname=%.*s,"
1269 "tgt_dir=%lu/%u(%p)\n", src_name->len, src_name->name,
1270 src->i_ino, src->i_generation, src, tgt_name->len,
1271 tgt_name->name, tgt->i_ino, tgt->i_generation, tgt);
1273 err = ll_prepare_mdc_op_data(&op_data, src, tgt, NULL, 0, 0, NULL);
1277 ll_get_child_fid(src, src_name, &op_data.fid3);
1278 ll_get_child_fid(tgt, tgt_name, &op_data.fid4);
1279 err = mdc_rename(sbi->ll_mdc_exp, &op_data,
1280 src_name->name, src_name->len,
1281 tgt_name->name, tgt_name->len, &request);
1284 ll_update_times(request, REPLY_REC_OFF, src);
1285 ll_update_times(request, REPLY_REC_OFF, tgt);
1286 err = ll_objects_destroy(request, src);
1292 ptlrpc_req_finished(request);
1297 #ifdef HAVE_VFS_INTENT_PATCHES
1298 static int ll_mknod_raw(struct nameidata *nd, int mode, dev_t rdev)
1300 return ll_mknod_generic(nd->dentry->d_inode, &nd->last, mode,rdev,NULL);
1302 static int ll_rename_raw(struct nameidata *srcnd, struct nameidata *tgtnd)
1304 return ll_rename_generic(srcnd->dentry->d_inode, &srcnd->last,
1305 tgtnd->dentry->d_inode, &tgtnd->last);
1307 static int ll_link_raw(struct nameidata *srcnd, struct nameidata *tgtnd)
1309 return ll_link_generic(srcnd->dentry->d_inode, tgtnd->dentry->d_inode,
1310 &tgtnd->last, NULL);
1312 static int ll_symlink_raw(struct nameidata *nd, const char *tgt)
1314 return ll_symlink_generic(nd->dentry->d_inode, &nd->last, tgt, NULL);
1316 static int ll_rmdir_raw(struct nameidata *nd)
1318 return ll_rmdir_generic(nd->dentry->d_inode, nd->dentry, &nd->last);
1320 static int ll_mkdir_raw(struct nameidata *nd, int mode)
1322 return ll_mkdir_generic(nd->dentry->d_inode, &nd->last, mode, NULL);
1324 static int ll_unlink_raw(struct nameidata *nd)
1326 return ll_unlink_generic(nd->dentry->d_inode, &nd->last);
1330 static int ll_mknod(struct inode *dir, struct dentry *dchild, int mode,
1333 return ll_mknod_generic(dir, &dchild->d_name, mode,
1334 old_encode_dev(rdev), dchild);
1337 static int ll_unlink(struct inode * dir, struct dentry *dentry)
1339 return ll_unlink_generic(dir, &dentry->d_name);
1341 static int ll_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1343 return ll_mkdir_generic(dir, &dentry->d_name, mode, dentry);
1345 static int ll_rmdir(struct inode *dir, struct dentry *dentry)
1347 return ll_rmdir_generic(dir, NULL, &dentry->d_name);
1349 static int ll_symlink(struct inode *dir, struct dentry *dentry,
1350 const char *oldname)
1352 return ll_symlink_generic(dir, &dentry->d_name, oldname, dentry);
1354 static int ll_link(struct dentry *old_dentry, struct inode *dir,
1355 struct dentry *new_dentry)
1357 return ll_link_generic(old_dentry->d_inode, dir,
1358 &new_dentry->d_name, new_dentry);
1360 static int ll_rename(struct inode *old_dir, struct dentry *old_dentry,
1361 struct inode *new_dir, struct dentry *new_dentry)
1364 err = ll_rename_generic(old_dir, &old_dentry->d_name, new_dir,
1365 &new_dentry->d_name);
1367 #ifndef HAVE_FS_RENAME_DOES_D_MOVE
1368 if (!S_ISDIR(old_dentry->d_inode->i_mode))
1370 d_move(old_dentry, new_dentry);
1375 struct inode_operations ll_dir_inode_operations = {
1376 #ifdef HAVE_VFS_INTENT_PATCHES
1377 .link_raw = ll_link_raw,
1378 .unlink_raw = ll_unlink_raw,
1379 .symlink_raw = ll_symlink_raw,
1380 .mkdir_raw = ll_mkdir_raw,
1381 .rmdir_raw = ll_rmdir_raw,
1382 .mknod_raw = ll_mknod_raw,
1383 .rename_raw = ll_rename_raw,
1384 .setattr = ll_setattr,
1385 .setattr_raw = ll_setattr_raw,
1388 .lookup = ll_lookup_nd,
1389 .create = ll_create_nd,
1390 /* We need all these non-raw things for NFSD, to not patch it. */
1391 .unlink = ll_unlink,
1394 .symlink = ll_symlink,
1396 .rename = ll_rename,
1397 .setattr = ll_setattr,
1398 .getattr = ll_getattr,
1399 .permission = ll_inode_permission,
1400 .setxattr = ll_setxattr,
1401 .getxattr = ll_getxattr,
1402 .listxattr = ll_listxattr,
1403 .removexattr = ll_removexattr,
1406 struct inode_operations ll_special_inode_operations = {
1407 #ifdef HAVE_VFS_INTENT_PATCHES
1408 .setattr_raw = ll_setattr_raw,
1410 .setattr = ll_setattr,
1411 .getattr = ll_getattr,
1412 .permission = ll_inode_permission,
1413 .setxattr = ll_setxattr,
1414 .getxattr = ll_getxattr,
1415 .listxattr = ll_listxattr,
1416 .removexattr = ll_removexattr,