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_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);
64 /* Get an inode by inode number (already instantiated by the intent lookup).
65 * Returns inode or NULL
68 static inline __u64 fid_flatten(const struct lu_fid *fid)
70 return (fid_seq(fid) - 1) * LUSTRE_SEQ_MAX_WIDTH + fid_oid(fid);
72 /* Build inode number on passed @fid */
73 ino_t ll_fid_build_ino(struct ll_sb_info *sbi,
79 if (fid_is_igif((struct lu_fid*)fid)) {
80 ino = lu_igif_ino((struct lu_fid*)fid);
85 * Very stupid and having many downsides inode allocation algorithm
88 ino = fid_flatten((struct lu_fid*)fid) & 0xFFFFFFFF;
90 if (unlikely(ino == 0))
91 /* the first result ino is 0xFFC001, so this is rarely used */
93 ino = ino | 0x80000000;
98 __u32 ll_fid_build_gen(struct ll_sb_info *sbi, struct ll_fid *fid)
103 if (fid_is_igif((struct lu_fid*)fid)) {
104 gen = lu_igif_gen((struct lu_fid*)fid);
109 /* called from iget5_locked->find_inode() under inode_lock spinlock */
110 static int fid_test_inode(struct inode *inode, void *opaque)
112 struct lustre_md *md = opaque;
113 struct lu_fid *fid = (struct lu_fid*)&md->body->fid1;
115 if (unlikely(!(md->body->valid & OBD_MD_FLID))) {
116 CERROR("MDS body missing FID\n");
120 return fid_seq(ll_inode_lu_fid(inode)) == fid_seq(fid) &&
121 fid_oid(ll_inode_lu_fid(inode)) == fid_oid(fid);
124 static int fid_set_inode(struct inode *inode, void *opaque)
126 struct lustre_md *md = opaque;
128 *ll_inode_lu_fid(inode) = *((struct lu_fid*)&md->body->fid1);
132 struct inode *ll_iget(struct super_block *sb, ino_t hash,
133 struct lustre_md *md)
139 inode = iget5_locked(sb, hash, fid_test_inode, fid_set_inode, md);
142 if (inode->i_state & I_NEW) {
143 ll_read_inode2(inode, md);
144 unlock_new_inode(inode);
146 if (!(inode->i_state & (I_FREEING | I_CLEAR)))
147 ll_update_inode(inode, md);
149 CDEBUG(D_VFSTRACE, "got inode: %lu/%u(%p) for "DFID"\n",
150 inode->i_ino, inode->i_generation, inode,
151 PFID(ll_inode_lu_fid(inode)));
157 static void ll_drop_negative_dentry(struct inode *dir)
159 struct dentry *dentry, *tmp_alias, *tmp_subdir;
161 spin_lock(&ll_lookup_lock);
162 spin_lock(&dcache_lock);
164 list_for_each_entry_safe(dentry, tmp_alias,
165 &dir->i_dentry,d_alias) {
166 if (!list_empty(&dentry->d_subdirs)) {
167 struct dentry *child;
168 list_for_each_entry_safe(child, tmp_subdir,
171 /* XXX Print some debug here? */
173 /* Negative dentry. If we were
174 dropping dcache lock, go
175 throught the list again */
176 if (ll_drop_dentry(child))
181 spin_unlock(&dcache_lock);
182 spin_unlock(&ll_lookup_lock);
185 int ll_mdc_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
186 void *data, int flag)
189 struct lustre_handle lockh;
193 case LDLM_CB_BLOCKING:
194 ldlm_lock2handle(lock, &lockh);
195 rc = ldlm_cli_cancel(&lockh);
197 CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
201 case LDLM_CB_CANCELING: {
202 struct inode *inode = ll_inode_from_lock(lock);
203 __u64 bits = lock->l_policy_data.l_inodebits.bits;
206 /* Invalidate all dentries associated with this inode */
210 fid = ll_inode_lu_fid(inode);;
212 LASSERT(lock->l_flags & LDLM_FL_CANCELING);
213 if ((bits & MDS_INODELOCK_LOOKUP) &&
214 ll_have_md_lock(inode, MDS_INODELOCK_LOOKUP))
215 bits &= ~MDS_INODELOCK_LOOKUP;
216 if ((bits & MDS_INODELOCK_UPDATE) &&
217 ll_have_md_lock(inode, MDS_INODELOCK_UPDATE))
218 bits &= ~MDS_INODELOCK_UPDATE;
219 if ((bits & MDS_INODELOCK_OPEN) &&
220 ll_have_md_lock(inode, MDS_INODELOCK_OPEN))
221 bits &= ~MDS_INODELOCK_OPEN;
223 if (!fid_res_name_eq(fid, &lock->l_resource->lr_name)) {
224 LDLM_ERROR(lock, "data mismatch with ino %lu/%u (%p)",
225 inode->i_ino, inode->i_generation, inode);
228 if (bits & MDS_INODELOCK_OPEN) {
230 switch (lock->l_req_mode) {
237 CERROR("open PR lock without FMODE_EXEC\n");
243 CERROR("Unexpected lock mode for OPEN lock "
244 "%d, inode %ld\n", lock->l_req_mode,
247 ll_mdc_real_close(inode, flags);
250 if (bits & MDS_INODELOCK_UPDATE)
251 clear_bit(LLI_F_HAVE_MDS_SIZE_LOCK,
252 &(ll_i2info(inode)->lli_flags));
254 if (S_ISDIR(inode->i_mode) &&
255 (bits & MDS_INODELOCK_UPDATE)) {
256 CDEBUG(D_INODE, "invalidating inode %lu\n",
258 truncate_inode_pages(inode->i_mapping, 0);
259 ll_drop_negative_dentry(inode);
260 inode->i_version++; /* XXX: remove with inode version*/
263 if (inode->i_sb->s_root &&
264 inode != inode->i_sb->s_root->d_inode &&
265 (bits & MDS_INODELOCK_LOOKUP))
266 ll_unhash_aliases(inode);
277 int ll_mdc_cancel_unused(struct lustre_handle *conn, struct inode *inode,
278 int flags, void *opaque)
280 struct ldlm_res_id res_id;
281 struct obd_device *obddev = class_conn2obd(conn);
284 fid_build_reg_res_name(ll_inode_lu_fid(inode), &res_id);
285 RETURN(ldlm_cli_cancel_unused(obddev->obd_namespace, &res_id, flags,
289 /* Pack the required supplementary groups into the supplied groups array.
290 * If we don't need to use the groups from the target inode(s) then we
291 * instead pack one or more groups from the user's supplementary group
292 * array in case it might be useful. Not needed if doing an MDS-side upcall. */
293 void ll_i2gids(__u32 *suppgids, struct inode *i1, struct inode *i2)
298 LASSERT(suppgids != NULL);
300 if (in_group_p(i1->i_gid))
301 suppgids[0] = i1->i_gid;
306 if (in_group_p(i2->i_gid))
307 suppgids[1] = i2->i_gid;
314 for (i = 0; i < current_ngroups; i++) {
315 if (suppgids[0] == -1) {
316 if (current_groups[i] != suppgids[1])
317 suppgids[0] = current_groups[i];
320 if (suppgids[1] == -1) {
321 if (current_groups[i] != suppgids[0])
322 suppgids[1] = current_groups[i];
329 int ll_prepare_mdc_op_data(struct mdc_op_data *op_data, struct inode *i1,
330 struct inode *i2, const char *name, int namelen,
331 int mode, void *data)
335 if (namelen > ll_i2sbi(i1)->ll_namelen)
336 return -ENAMETOOLONG;
337 ll_i2gids(op_data->suppgids, i1, i2);
338 ll_inode2fid(&op_data->fid1, i1);
341 ll_inode2fid(&op_data->fid2, i2);
343 memset(&op_data->fid2, 0, sizeof(op_data->fid2));
345 op_data->name = name;
346 op_data->namelen = namelen;
347 op_data->create_mode = mode;
348 op_data->mod_time = CURRENT_SECONDS;
349 op_data->data = data;
354 static void ll_d_add(struct dentry *de, struct inode *inode)
356 CDEBUG(D_DENTRY, "adding inode %p to dentry %p\n", inode, de);
358 if (!list_empty(&de->d_alias)) {
359 spin_unlock(&dcache_lock);
360 CERROR("dentry %.*s %p alias next %p, prev %p\n",
361 de->d_name.len, de->d_name.name, de,
362 de->d_alias.next, de->d_alias.prev);
366 list_add(&de->d_alias, &inode->i_dentry);
370 if (!d_unhashed(de)) {
371 spin_unlock(&dcache_lock);
372 CERROR("dentry %.*s %p hash next %p\n",
373 de->d_name.len, de->d_name.name, de, de->d_hash.next);
376 d_rehash_cond(de, 0);
379 /* Search "inode"'s alias list for a dentry that has the same name and parent
380 * as de. If found, return it. If not found, return de.
381 * Lustre can't use d_add_unique because don't unhash aliases for directory
382 * in ll_revalidate_it. After revaliadate inode will be have hashed aliases
383 * and it triggers BUG_ON in d_instantiate_unique (bug #10954).
385 static struct dentry *ll_find_alias(struct inode *inode, struct dentry *de)
387 struct list_head *tmp;
388 struct dentry *dentry;
389 struct dentry *last_discon = NULL;
391 spin_lock(&ll_lookup_lock);
392 spin_lock(&dcache_lock);
393 list_for_each(tmp, &inode->i_dentry) {
394 dentry = list_entry(tmp, struct dentry, d_alias);
396 /* We are called here with 'de' already on the aliases list. */
397 if (unlikely(dentry == de)) {
402 if (dentry->d_flags & DCACHE_DISCONNECTED) {
403 /* LASSERT(last_discon == NULL); see bug 20055 */
404 last_discon = dentry;
408 if (dentry->d_parent != de->d_parent)
411 if (dentry->d_name.hash != de->d_name.hash)
414 if (dentry->d_name.len != de->d_name.len)
417 if (memcmp(dentry->d_name.name, de->d_name.name,
418 de->d_name.len) != 0)
424 unlock_dentry(dentry);
425 ll_dops_init(dentry, 0);
426 d_rehash_cond(dentry, 0); /* avoid taking dcache_lock inside */
427 spin_unlock(&dcache_lock);
428 spin_unlock(&ll_lookup_lock);
430 CDEBUG(D_DENTRY, "alias dentry %.*s (%p) parent %p inode %p "
431 "refc %d\n", de->d_name.len, de->d_name.name, de,
432 de->d_parent, de->d_inode, atomic_read(&de->d_count));
437 CDEBUG(D_DENTRY, "Reuse disconnected dentry %p inode %p "
438 "refc %d\n", last_discon, last_discon->d_inode,
439 atomic_read(&last_discon->d_count));
440 dget_locked(last_discon);
441 last_discon->d_flags |= DCACHE_LUSTRE_INVALID;
442 spin_unlock(&dcache_lock);
443 spin_unlock(&ll_lookup_lock);
444 ll_dops_init(last_discon, 1);
446 d_move(last_discon, de);
451 de->d_flags |= DCACHE_LUSTRE_INVALID;
454 spin_unlock(&dcache_lock);
455 spin_unlock(&ll_lookup_lock);
460 int lookup_it_finish(struct ptlrpc_request *request, int offset,
461 struct lookup_intent *it, void *data)
463 struct it_cb_data *icbd = data;
464 struct dentry **de = icbd->icbd_childp;
465 struct inode *parent = icbd->icbd_parent;
466 struct ll_sb_info *sbi = ll_i2sbi(parent);
467 struct inode *inode = NULL;
471 /* NB 1 request reference will be taken away by ll_intent_lock()
473 if (!it_disposition(it, DISP_LOOKUP_NEG)) {
474 struct dentry *save = *de;
477 rc = ll_prep_inode(sbi->ll_osc_exp, &inode, request, offset,
482 CDEBUG(D_DLMTRACE, "setting l_data to inode %p (%lu/%u)\n",
483 inode, inode->i_ino, inode->i_generation);
484 mdc_set_lock_data(&it->d.lustre.it_lock_handle, inode, &bits);
486 /* We used to query real size from OSTs here, but actually
487 this is not needed. For stat() calls size would be updated
488 from subsequent do_revalidate()->ll_inode_revalidate_it() in
490 vfs_getattr_it->ll_getattr()->ll_inode_revalidate_it() in 2.6
491 Everybody else who needs correct file size would call
492 ll_glimpse_size or some equivalent themselves anyway.
493 Also see bug 7198. */
495 ll_dops_init(*de, 1);
496 *de = ll_find_alias(inode, *de);
498 struct ll_dentry_data *lld = ll_d2d(*de);
500 /* just make sure the ll_dentry_data is ready */
501 if (unlikely(lld == NULL)) {
504 if (likely(lld != NULL))
505 lld->lld_sa_generation = 0;
508 /* we have lookup look - unhide dentry */
509 if (bits & MDS_INODELOCK_LOOKUP) {
511 (*de)->d_flags &= ~(DCACHE_LUSTRE_INVALID);
515 ll_dops_init(*de, 1);
516 /* Check that parent has UPDATE lock. If there is none, we
517 cannot afford to hash this dentry (done by ll_d_add) as it
518 might get picked up later when UPDATE lock will appear */
519 if (ll_have_md_lock(parent, MDS_INODELOCK_UPDATE)) {
520 spin_lock(&dcache_lock);
521 ll_d_add(*de, inode);
522 spin_unlock(&dcache_lock);
524 /* negative lookup - and don't have update lock to
527 (*de)->d_flags |= DCACHE_LUSTRE_INVALID;
530 (*de)->d_inode = NULL;
531 /* We do not want to hash the dentry if don`t have a
532 * lock, but if this dentry is later used in d_move,
533 * we'd hit uninitialised list head d_hash, so we just
534 * do this to init d_hash field but leave dentry
535 * unhashed. (bug 10796). */
544 static struct dentry *ll_lookup_it(struct inode *parent, struct dentry *dentry,
545 struct lookup_intent *it, int lookup_flags)
547 struct dentry *save = dentry, *retval;
548 struct mdc_op_data op_data = { { 0 } };
549 struct it_cb_data icbd;
550 struct ptlrpc_request *req = NULL;
551 struct lookup_intent lookup_it = { .it_op = IT_LOOKUP };
555 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),intent=%s\n",
556 dentry->d_name.len, dentry->d_name.name, parent->i_ino,
557 parent->i_generation, parent, LL_IT2STR(it));
559 if (d_mountpoint(dentry))
560 CERROR("Tell Peter, lookup on mtpt, it %s\n", LL_IT2STR(it));
562 ll_frob_intent(&it, &lookup_it);
564 /* As do_lookup is called before follow_mount, root dentry may be left
565 * not valid, revalidate it here. */
566 if (parent->i_sb->s_root && (parent->i_sb->s_root->d_inode == parent) &&
567 (it->it_op & (IT_OPEN | IT_CREAT))) {
568 rc = ll_inode_revalidate_it(parent->i_sb->s_root, it);
573 if (it->it_op == IT_GETATTR) {
574 first = ll_statahead_enter(parent, &dentry, 1);
576 ll_statahead_exit(parent, dentry, first);
578 RETURN(retval = dentry);
582 icbd.icbd_parent = parent;
583 icbd.icbd_childp = &dentry;
585 rc = ll_prepare_mdc_op_data(&op_data, parent, NULL, dentry->d_name.name,
586 dentry->d_name.len, lookup_flags, NULL);
590 it->it_create_mode &= ~current->fs->umask;
592 rc = mdc_intent_lock(ll_i2mdcexp(parent), &op_data, NULL, 0, it,
593 lookup_flags, &req, ll_mdc_blocking_ast, 0);
596 GOTO(out, retval = ERR_PTR(rc));
598 rc = lookup_it_finish(req, DLM_REPLY_REC_OFF, it, &icbd);
600 ll_intent_release(it);
601 GOTO(out, retval = ERR_PTR(rc));
604 if (first == -EEXIST)
605 ll_statahead_mark(parent, dentry);
607 if ((it->it_op & IT_OPEN) && dentry->d_inode &&
608 !S_ISREG(dentry->d_inode->i_mode) &&
609 !S_ISDIR(dentry->d_inode->i_mode)) {
610 ll_release_openhandle(dentry, it);
612 ll_lookup_finish_locks(it, dentry);
615 GOTO(out, retval = NULL);
617 GOTO(out, retval = dentry);
620 ptlrpc_req_finished(req);
624 #ifdef HAVE_VFS_INTENT_PATCHES
625 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
626 struct nameidata *nd)
631 if (nd && nd->flags & LOOKUP_LAST && !(nd->flags & LOOKUP_LINK_NOTLAST))
632 de = ll_lookup_it(parent, dentry, &nd->intent, nd->flags);
634 de = ll_lookup_it(parent, dentry, NULL, 0);
639 struct lookup_intent *ll_convert_intent(struct open_intent *oit,
642 struct lookup_intent *it;
644 OBD_ALLOC(it, sizeof(*it));
646 return ERR_PTR(-ENOMEM);
648 if (lookup_flags & LOOKUP_OPEN) {
650 if (lookup_flags & LOOKUP_CREATE)
651 it->it_op |= IT_CREAT;
652 it->it_create_mode = oit->create_mode;
653 it->it_flags = oit->flags;
655 it->it_op = IT_GETATTR;
658 #ifndef HAVE_FILE_IN_STRUCT_INTENT
659 /* Since there is no way to pass our intent to ll_file_open,
660 * just check the file is there. Actual open will be done
662 if (it->it_op & IT_OPEN)
663 it->it_op = IT_LOOKUP;
669 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
670 struct nameidata *nd)
675 if (nd && !(nd->flags & (LOOKUP_CONTINUE|LOOKUP_PARENT))) {
676 struct lookup_intent *it;
678 #if defined(HAVE_FILE_IN_STRUCT_INTENT) && (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,17))
679 /* Did we came here from failed revalidate just to propagate
681 if (nd->flags & LOOKUP_OPEN)
682 if (IS_ERR(nd->intent.open.file))
683 RETURN((struct dentry *)nd->intent.open.file);
686 if (ll_d2d(dentry) && ll_d2d(dentry)->lld_it) {
687 it = ll_d2d(dentry)->lld_it;
688 ll_d2d(dentry)->lld_it = NULL;
690 if ((nd->flags & LOOKUP_CREATE ) && !(nd->flags & LOOKUP_OPEN)) {
691 /* We are sure this is new dentry, so we need to create
692 our private data and set the dentry ops */
693 ll_dops_init(dentry, 1);
696 it = ll_convert_intent(&nd->intent.open, nd->flags);
698 RETURN((struct dentry *)it);
701 de = ll_lookup_it(parent, dentry, it, nd->flags);
704 if ((nd->flags & LOOKUP_OPEN) && !IS_ERR(dentry)) { /* Open */
705 if (dentry->d_inode &&
706 it_disposition(it, DISP_OPEN_OPEN)) { /* nocreate */
707 #ifdef HAVE_FILE_IN_STRUCT_INTENT
708 if (S_ISFIFO(dentry->d_inode->i_mode)) {
709 // We cannot call open here as it would
712 (struct ptlrpc_request *)
713 it->d.lustre.it_data);
715 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,17))
716 /* 2.6.1[456] have a bug in open_namei() that forgets to check
717 * nd->intent.open.file for error, so we need to return it as lookup's result
720 nd->intent.open.file->private_data = it;
721 filp =lookup_instantiate_filp(nd,dentry,
726 de = (struct dentry *) filp;
729 nd->intent.open.file->private_data = it;
730 (void)lookup_instantiate_filp(nd,dentry,
735 #else /* HAVE_FILE_IN_STRUCT_INTENT */
736 /* Release open handle as we have no way to
737 * pass it to ll_file_open */
738 ll_release_openhandle(dentry, it);
739 #endif /* HAVE_FILE_IN_STRUCT_INTENT */
740 } else if (it_disposition(it, DISP_OPEN_CREATE)) {
741 // XXX This can only reliably work on assumption
742 // that there are NO hashed negative dentries.
743 ll_d2d(dentry)->lld_it = it;
744 it = NULL; /* Will be freed in ll_create_nd */
745 /* We absolutely depend on ll_create_nd to be
746 * called to not leak this intent and possible
747 * data attached to it */
752 ll_intent_release(it);
753 OBD_FREE(it, sizeof(*it));
756 de = ll_lookup_it(parent, dentry, NULL, 0);
763 /* We depend on "mode" being set with the proper file type/umask by now */
764 static struct inode *ll_create_node(struct inode *dir, const char *name,
765 int namelen, const void *data, int datalen,
766 int mode, __u64 extra,
767 struct lookup_intent *it)
769 struct inode *inode = NULL;
770 struct ptlrpc_request *request = NULL;
771 struct ll_sb_info *sbi = ll_i2sbi(dir);
775 LASSERT(it && it->d.lustre.it_disposition);
777 LASSERT(it_disposition(it, DISP_ENQ_CREATE_REF));
778 request = it->d.lustre.it_data;
779 it_clear_disposition(it, DISP_ENQ_CREATE_REF);
780 rc = ll_prep_inode(sbi->ll_osc_exp, &inode, request, DLM_REPLY_REC_OFF,
783 GOTO(out, inode = ERR_PTR(rc));
785 LASSERT(list_empty(&inode->i_dentry));
787 /* We asked for a lock on the directory, but were granted a
788 * lock on the inode. Since we finally have an inode pointer,
789 * stuff it in the lock. */
790 CDEBUG(D_DLMTRACE, "setting l_ast_data to inode %p (%lu/%u)\n",
791 inode, inode->i_ino, inode->i_generation);
792 mdc_set_lock_data(&it->d.lustre.it_lock_handle, inode, NULL);
795 ptlrpc_req_finished(request);
800 * By the time this is called, we already have created the directory cache
801 * entry for the new file, but it is so far negative - it has no inode.
803 * We defer creating the OBD object(s) until open, to keep the intent and
804 * non-intent code paths similar, and also because we do not have the MDS
805 * inode number before calling ll_create_node() (which is needed for LOV),
806 * so we would need to do yet another RPC to the MDS to store the LOV EA
807 * data on the MDS. If needed, we would pass the PACKED lmm as data and
808 * lmm_size in datalen (the MDS still has code which will handle that).
810 * If the create succeeds, we fill in the inode information
811 * with d_instantiate().
813 static int ll_create_it(struct inode *dir, struct dentry *dentry, int mode,
814 struct lookup_intent *it)
817 struct ptlrpc_request *request = it->d.lustre.it_data;
821 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),intent=%s\n",
822 dentry->d_name.len, dentry->d_name.name, dir->i_ino,
823 dir->i_generation, dir, LL_IT2STR(it));
825 rc = it_open_error(DISP_OPEN_CREATE, it);
829 mdc_store_inode_generation(request, DLM_INTENT_REC_OFF,
831 inode = ll_create_node(dir, dentry->d_name.name, dentry->d_name.len,
832 NULL, 0, mode, 0, it);
834 RETURN(PTR_ERR(inode));
837 d_instantiate(dentry, inode);
838 /* Negative dentry may be unhashed if parent does not have UPDATE lock,
839 * but some callers, e.g. do_coredump, expect dentry to be hashed after
840 * successful create. Hash it here. */
841 spin_lock(&dcache_lock);
842 if (d_unhashed(dentry))
843 d_rehash_cond(dentry, 0);
844 spin_unlock(&dcache_lock);
848 static void ll_update_times(struct ptlrpc_request *request, int offset,
851 struct mds_body *body = lustre_msg_buf(request->rq_repmsg, offset,
855 if (body->valid & OBD_MD_FLMTIME &&
856 body->mtime > LTIME_S(inode->i_mtime)) {
857 CDEBUG(D_INODE, "setting ino %lu mtime from %lu to "LPU64"\n",
858 inode->i_ino, LTIME_S(inode->i_mtime), body->mtime);
859 LTIME_S(inode->i_mtime) = body->mtime;
861 if (body->valid & OBD_MD_FLCTIME &&
862 body->ctime > LTIME_S(inode->i_ctime))
863 LTIME_S(inode->i_ctime) = body->ctime;
866 static int ll_new_node(struct inode *dir, struct qstr *name,
867 const char *tgt, int mode,
868 int rdev, struct dentry *dchild)
870 struct ptlrpc_request *request = NULL;
871 struct inode *inode = NULL;
872 struct ll_sb_info *sbi = ll_i2sbi(dir);
873 struct mdc_op_data op_data = { { 0 } };
878 if (unlikely(tgt != NULL))
879 tgt_len = strlen(tgt)+1;
881 err = ll_prepare_mdc_op_data(&op_data, dir, NULL, name->name,
886 err = mdc_create(sbi->ll_mdc_exp, &op_data, tgt, tgt_len,
887 mode, cfs_curproc_fsuid(), cfs_curproc_fsgid(),
888 cfs_curproc_cap_pack(), rdev, &request);
892 ll_update_times(request, REPLY_REC_OFF, dir);
895 err = ll_prep_inode(sbi->ll_osc_exp, &inode, request,
896 REPLY_REC_OFF, dchild->d_sb);
901 d_instantiate(dchild, inode);
905 ptlrpc_req_finished(request);
911 static int ll_mknod_generic(struct inode *dir, struct qstr *name, int mode,
912 unsigned rdev, struct dentry *dchild)
917 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p) mode %o dev %x\n",
918 name->len, name->name, dir->i_ino, dir->i_generation, dir,
921 mode &= ~current->fs->umask;
923 switch (mode & S_IFMT) {
925 mode |= S_IFREG; /* for mode = 0 case, fallthrough */
931 err = ll_new_node(dir, name, NULL, mode, rdev, dchild);
942 #ifndef HAVE_VFS_INTENT_PATCHES
943 static int ll_create_nd(struct inode *dir, struct dentry *dentry, int mode, struct nameidata *nd)
945 struct lookup_intent *it = ll_d2d(dentry)->lld_it;
949 return ll_mknod_generic(dir, &dentry->d_name, mode, 0, dentry);
951 ll_d2d(dentry)->lld_it = NULL;
953 /* Was there an error? Propagate it! */
954 if (it->d.lustre.it_status) {
955 rc = it->d.lustre.it_status;
959 rc = ll_create_it(dir, dentry, mode, it);
960 #ifdef HAVE_FILE_IN_STRUCT_INTENT
961 if (nd && (nd->flags & LOOKUP_OPEN) && dentry->d_inode) { /* Open */
962 nd->intent.open.file->private_data = it;
963 lookup_instantiate_filp(nd, dentry, NULL);
966 ll_release_openhandle(dentry,it);
970 ll_intent_release(it);
971 OBD_FREE(it, sizeof(*it));
976 static int ll_create_nd(struct inode *dir, struct dentry *dentry, int mode, struct nameidata *nd)
979 if (!nd || !nd->intent.d.lustre.it_disposition)
980 /* No saved request? Just mknod the file */
981 return ll_mknod_generic(dir, &dentry->d_name, mode, 0, dentry);
983 return ll_create_it(dir, dentry, mode, &nd->intent);
987 static int ll_symlink_generic(struct inode *dir, struct qstr *name,
988 const char *tgt, struct dentry *dchild)
993 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),target=%.*s\n",
994 name->len, name->name, dir->i_ino, dir->i_generation,
997 err = ll_new_node(dir, name, (char *)tgt, S_IFLNK | S_IRWXUGO,
1002 static int ll_link_generic(struct inode *src, struct inode *dir,
1003 struct qstr *name, struct dentry *dchild)
1005 struct ptlrpc_request *request = NULL;
1006 struct mdc_op_data op_data = { { 0 } };
1008 struct ll_sb_info *sbi = ll_i2sbi(dir);
1012 "VFS Op: inode=%lu/%u(%p), dir=%lu/%u(%p), target=%.*s\n",
1013 src->i_ino, src->i_generation, src, dir->i_ino,
1014 dir->i_generation, dir, name->len, name->name);
1016 err = ll_prepare_mdc_op_data(&op_data, src, dir, name->name,
1017 name->len, 0, NULL);
1020 err = mdc_link(sbi->ll_mdc_exp, &op_data, &request);
1027 ll_update_times(request, REPLY_REC_OFF, dir);
1031 ptlrpc_req_finished(request);
1035 static int ll_mkdir_generic(struct inode *dir, struct qstr *name, int mode,
1036 struct dentry *dchild)
1042 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
1043 name->len, name->name, dir->i_ino, dir->i_generation, dir);
1045 mode = (mode & (S_IRWXUGO|S_ISVTX) & ~current->fs->umask) | S_IFDIR;
1046 err = ll_new_node(dir, name, NULL, mode, 0, dchild);
1051 /* Try to find the child dentry by its name.
1052 If found, put the result fid into @fid. */
1053 static void ll_get_child_fid(struct inode * dir, struct qstr *name,
1056 struct dentry *parent, *child;
1058 parent = list_entry(dir->i_dentry.next, struct dentry, d_alias);
1059 child = d_lookup(parent, name);
1062 ll_inode2fid(fid, child->d_inode);
1067 static int ll_rmdir_generic(struct inode *dir, struct dentry *dparent,
1070 struct ptlrpc_request *request = NULL;
1071 struct mdc_op_data op_data = { { 0 } };
1072 struct dentry *dentry;
1075 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
1076 name->len, name->name, dir->i_ino, dir->i_generation, dir);
1078 /* Check if we have something mounted at the dir we are going to delete
1079 * In such a case there would always be dentry present. */
1081 dentry = d_lookup(dparent, name);
1083 int mounted = d_mountpoint(dentry);
1086 GOTO(out, rc = -EBUSY);
1090 rc = ll_prepare_mdc_op_data(&op_data, dir, NULL, name->name,
1091 name->len, S_IFDIR, NULL);
1095 ll_get_child_fid(dir, name, &op_data.fid3);
1096 rc = mdc_unlink(ll_i2sbi(dir)->ll_mdc_exp, &op_data, &request);
1099 ll_update_times(request, REPLY_REC_OFF, dir);
1103 ptlrpc_req_finished(request);
1107 int ll_objects_destroy(struct ptlrpc_request *request, struct inode *dir)
1109 struct mds_body *body;
1110 struct lov_mds_md *eadata;
1111 struct lov_stripe_md *lsm = NULL;
1112 struct obd_trans_info oti = { 0 };
1117 /* req is swabbed so this is safe */
1118 body = lustre_msg_buf(request->rq_repmsg, REPLY_REC_OFF, sizeof(*body));
1120 if (!(body->valid & OBD_MD_FLEASIZE))
1123 if (body->eadatasize == 0) {
1124 CERROR("OBD_MD_FLEASIZE set but eadatasize zero\n");
1125 GOTO(out, rc = -EPROTO);
1128 /* The MDS sent back the EA because we unlinked the last reference
1129 * to this file. Use this EA to unlink the objects on the OST.
1130 * It's opaque so we don't swab here; we leave it to obd_unpackmd() to
1131 * check it is complete and sensible. */
1132 eadata = lustre_swab_repbuf(request, REPLY_REC_OFF + 1,
1133 body->eadatasize, NULL);
1134 LASSERT(eadata != NULL);
1135 if (eadata == NULL) {
1136 CERROR("Can't unpack MDS EA data\n");
1137 GOTO(out, rc = -EPROTO);
1140 rc = obd_unpackmd(ll_i2obdexp(dir), &lsm, eadata, body->eadatasize);
1142 CERROR("obd_unpackmd: %d\n", rc);
1145 LASSERT(rc >= sizeof(*lsm));
1147 rc = obd_checkmd(ll_i2obdexp(dir), ll_i2mdcexp(dir), lsm);
1149 GOTO(out_free_memmd, rc);
1153 GOTO(out_free_memmd, rc = -ENOMEM);
1155 oa->o_id = lsm->lsm_object_id;
1156 oa->o_gr = lsm->lsm_object_gr;
1157 oa->o_mode = body->mode & S_IFMT;
1158 oa->o_valid = OBD_MD_FLID | OBD_MD_FLGROUP | OBD_MD_FLTYPE;
1160 if (body->valid & OBD_MD_FLCOOKIE) {
1161 oa->o_valid |= OBD_MD_FLCOOKIE;
1162 oti.oti_logcookies =
1163 lustre_msg_buf(request->rq_repmsg, REPLY_REC_OFF + 2,
1164 sizeof(struct llog_cookie) *
1165 lsm->lsm_stripe_count);
1166 if (oti.oti_logcookies == NULL) {
1167 oa->o_valid &= ~OBD_MD_FLCOOKIE;
1168 body->valid &= ~OBD_MD_FLCOOKIE;
1172 rc = obd_destroy(ll_i2obdexp(dir), oa, lsm, &oti, ll_i2mdcexp(dir));
1175 CERROR("obd destroy objid "LPX64"@"LPX64" error %d\n",
1176 lsm->lsm_object_id, lsm->lsm_object_gr, rc);
1178 obd_free_memmd(ll_i2obdexp(dir), &lsm);
1183 static int ll_unlink_generic(struct inode * dir, struct qstr *name)
1185 struct ptlrpc_request *request = NULL;
1186 struct mdc_op_data op_data = { { 0 } };
1190 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
1191 name->len, name->name, dir->i_ino, dir->i_generation, dir);
1193 rc = ll_prepare_mdc_op_data(&op_data, dir, NULL, name->name,
1194 name->len, 0, NULL);
1198 ll_get_child_fid(dir, name, &op_data.fid3);
1199 rc = mdc_unlink(ll_i2sbi(dir)->ll_mdc_exp, &op_data, &request);
1203 ll_update_times(request, REPLY_REC_OFF, dir);
1205 rc = ll_objects_destroy(request, dir);
1210 ptlrpc_req_finished(request);
1214 static int ll_rename_generic(struct inode *src, struct qstr *src_name,
1215 struct inode *tgt, struct qstr *tgt_name)
1217 struct ptlrpc_request *request = NULL;
1218 struct ll_sb_info *sbi = ll_i2sbi(src);
1219 struct mdc_op_data op_data = { { 0 } };
1223 CDEBUG(D_VFSTRACE,"VFS Op:oldname=%.*s,src_dir=%lu/%u(%p),newname=%.*s,"
1224 "tgt_dir=%lu/%u(%p)\n", src_name->len, src_name->name,
1225 src->i_ino, src->i_generation, src, tgt_name->len,
1226 tgt_name->name, tgt->i_ino, tgt->i_generation, tgt);
1228 err = ll_prepare_mdc_op_data(&op_data, src, tgt, NULL, 0, 0, NULL);
1232 ll_get_child_fid(src, src_name, &op_data.fid3);
1233 ll_get_child_fid(tgt, tgt_name, &op_data.fid4);
1234 err = mdc_rename(sbi->ll_mdc_exp, &op_data,
1235 src_name->name, src_name->len,
1236 tgt_name->name, tgt_name->len, &request);
1239 ll_update_times(request, REPLY_REC_OFF, src);
1240 ll_update_times(request, REPLY_REC_OFF, tgt);
1241 err = ll_objects_destroy(request, src);
1247 ptlrpc_req_finished(request);
1252 #ifdef HAVE_VFS_INTENT_PATCHES
1253 static int ll_mknod_raw(struct nameidata *nd, int mode, dev_t rdev)
1255 return ll_mknod_generic(nd->dentry->d_inode, &nd->last, mode,rdev,NULL);
1257 static int ll_rename_raw(struct nameidata *srcnd, struct nameidata *tgtnd)
1259 return ll_rename_generic(srcnd->dentry->d_inode, &srcnd->last,
1260 tgtnd->dentry->d_inode, &tgtnd->last);
1262 static int ll_link_raw(struct nameidata *srcnd, struct nameidata *tgtnd)
1264 return ll_link_generic(srcnd->dentry->d_inode, tgtnd->dentry->d_inode,
1265 &tgtnd->last, NULL);
1267 static int ll_symlink_raw(struct nameidata *nd, const char *tgt)
1269 return ll_symlink_generic(nd->dentry->d_inode, &nd->last, tgt, NULL);
1271 static int ll_rmdir_raw(struct nameidata *nd)
1273 return ll_rmdir_generic(nd->dentry->d_inode, nd->dentry, &nd->last);
1275 static int ll_mkdir_raw(struct nameidata *nd, int mode)
1277 return ll_mkdir_generic(nd->dentry->d_inode, &nd->last, mode, NULL);
1279 static int ll_unlink_raw(struct nameidata *nd)
1281 return ll_unlink_generic(nd->dentry->d_inode, &nd->last);
1285 static int ll_mknod(struct inode *dir, struct dentry *dchild, int mode,
1288 return ll_mknod_generic(dir, &dchild->d_name, mode,
1289 old_encode_dev(rdev), dchild);
1292 static int ll_unlink(struct inode * dir, struct dentry *dentry)
1294 return ll_unlink_generic(dir, &dentry->d_name);
1296 static int ll_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1298 return ll_mkdir_generic(dir, &dentry->d_name, mode, dentry);
1300 static int ll_rmdir(struct inode *dir, struct dentry *dentry)
1302 return ll_rmdir_generic(dir, NULL, &dentry->d_name);
1304 static int ll_symlink(struct inode *dir, struct dentry *dentry,
1305 const char *oldname)
1307 return ll_symlink_generic(dir, &dentry->d_name, oldname, dentry);
1309 static int ll_link(struct dentry *old_dentry, struct inode *dir,
1310 struct dentry *new_dentry)
1312 return ll_link_generic(old_dentry->d_inode, dir,
1313 &new_dentry->d_name, new_dentry);
1315 static int ll_rename(struct inode *old_dir, struct dentry *old_dentry,
1316 struct inode *new_dir, struct dentry *new_dentry)
1319 err = ll_rename_generic(old_dir, &old_dentry->d_name, new_dir,
1320 &new_dentry->d_name);
1322 #ifndef HAVE_FS_RENAME_DOES_D_MOVE
1323 if (!S_ISDIR(old_dentry->d_inode->i_mode))
1325 d_move(old_dentry, new_dentry);
1330 struct inode_operations ll_dir_inode_operations = {
1331 #ifdef HAVE_VFS_INTENT_PATCHES
1332 .link_raw = ll_link_raw,
1333 .unlink_raw = ll_unlink_raw,
1334 .symlink_raw = ll_symlink_raw,
1335 .mkdir_raw = ll_mkdir_raw,
1336 .rmdir_raw = ll_rmdir_raw,
1337 .mknod_raw = ll_mknod_raw,
1338 .rename_raw = ll_rename_raw,
1339 .setattr = ll_setattr,
1340 .setattr_raw = ll_setattr_raw,
1343 .lookup = ll_lookup_nd,
1344 .create = ll_create_nd,
1345 /* We need all these non-raw things for NFSD, to not patch it. */
1346 .unlink = ll_unlink,
1349 .symlink = ll_symlink,
1351 .rename = ll_rename,
1352 .setattr = ll_setattr,
1353 .getattr = ll_getattr,
1354 .permission = ll_inode_permission,
1355 .setxattr = ll_setxattr,
1356 .getxattr = ll_getxattr,
1357 .listxattr = ll_listxattr,
1358 .removexattr = ll_removexattr,
1361 struct inode_operations ll_special_inode_operations = {
1362 #ifdef HAVE_VFS_INTENT_PATCHES
1363 .setattr_raw = ll_setattr_raw,
1365 .setattr = ll_setattr,
1366 .getattr = ll_getattr,
1367 .permission = ll_inode_permission,
1368 .setxattr = ll_setxattr,
1369 .getxattr = ll_getxattr,
1370 .listxattr = ll_listxattr,
1371 .removexattr = ll_removexattr,