1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
4 * Copyright (c) 2002, 2003 Cluster File Systems, Inc.
6 * This file is part of Lustre, http://www.lustre.org.
8 * Lustre is free software; you can redistribute it and/or
9 * modify it under the terms of version 2 of the GNU General Public
10 * License as published by the Free Software Foundation.
12 * Lustre is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 * GNU General Public License for more details.
17 * You should have received a copy of the GNU General Public License
18 * along with Lustre; if not, write to the Free Software
19 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
23 #include <linux/sched.h>
25 #include <linux/smp_lock.h>
26 #include <linux/quotaops.h>
27 #include <linux/highmem.h>
28 #include <linux/pagemap.h>
30 #define DEBUG_SUBSYSTEM S_LLITE
32 #include <obd_support.h>
33 #include <lustre_lite.h>
34 #include <lustre_dlm.h>
35 #include <linux/lustre_version.h>
36 #include "llite_internal.h"
40 /* called from iget{4,5_locked}->find_inode() under inode_lock spinlock */
41 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
42 static int ll_test_inode(struct inode *inode, unsigned long ino, void *opaque)
44 static int ll_test_inode(struct inode *inode, void *opaque)
47 static int last_ino, last_gen, last_count;
48 struct lustre_md *md = opaque;
50 if (!(md->body->valid & (OBD_MD_FLGENER | OBD_MD_FLID))) {
51 CERROR("MDS body missing inum or generation\n");
55 if (last_ino == md->body->ino && last_gen == md->body->generation &&
60 CDEBUG(D_VFSTRACE, "compared %u/%u %u times\n",
61 last_ino, last_gen, last_count);
63 last_ino = md->body->ino;
64 last_gen = md->body->generation;
66 "comparing inode %p ino %lu/%u to body "LPU64"/%u\n",
67 inode, inode->i_ino, inode->i_generation,
68 md->body->ino, md->body->generation);
71 #if (LINUX_VERSION_CODE > KERNEL_VERSION(2,5,0))
72 if (inode->i_ino != md->body->ino)
75 if (inode->i_generation != md->body->generation) {
76 struct ll_sb_info *sbi = ll_i2sbi(inode);
77 struct ll_inode_info *lli = ll_i2info(inode);
79 if (inode->i_state & (I_FREEING | I_CLEAR))
81 if (inode->i_nlink == 0)
84 /* add "duplicate" inode into deathrow for destroy */
85 spin_lock(&sbi->ll_deathrow_lock);
86 if (list_empty(&lli->lli_dead_list)) {
88 list_add(&lli->lli_dead_list, &sbi->ll_deathrow);
90 spin_unlock(&sbi->ll_deathrow_lock);
95 /* Apply the attributes in 'opaque' to this inode */
96 if (!(inode->i_state & (I_FREEING | I_CLEAR)))
97 ll_update_inode(inode, md);
101 extern struct dentry_operations ll_d_ops;
103 int ll_unlock(__u32 mode, struct lustre_handle *lockh)
107 ldlm_lock_decref(lockh, mode);
112 /* Get an inode by inode number (already instantiated by the intent lookup).
113 * Returns inode or NULL
115 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,5,0))
116 int ll_set_inode(struct inode *inode, void *opaque)
118 ll_read_inode2(inode, opaque);
122 struct inode *ll_iget(struct super_block *sb, ino_t hash,
123 struct lustre_md *md)
128 inode = iget5_locked(sb, hash, ll_test_inode, ll_set_inode, md);
131 if (inode->i_state & I_NEW)
132 unlock_new_inode(inode);
133 CDEBUG(D_VFSTRACE, "inode: %lu/%u(%p)\n", inode->i_ino,
134 inode->i_generation, inode);
140 struct inode *ll_iget(struct super_block *sb, ino_t hash,
141 struct lustre_md *md)
145 inode = iget4(sb, hash, ll_test_inode, md);
147 CDEBUG(D_VFSTRACE, "inode: %lu/%u(%p)\n", inode->i_ino,
148 inode->i_generation, inode);
153 int ll_mdc_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
154 void *data, int flag)
157 struct lustre_handle lockh;
161 case LDLM_CB_BLOCKING:
162 ldlm_lock2handle(lock, &lockh);
163 rc = ldlm_cli_cancel(&lockh);
165 CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
169 case LDLM_CB_CANCELING: {
170 struct inode *inode = ll_inode_from_lock(lock);
171 __u64 bits = lock->l_policy_data.l_inodebits.bits;
173 /* Invalidate all dentries associated with this inode */
177 if (lock->l_resource->lr_name.name[0] != inode->i_ino ||
178 lock->l_resource->lr_name.name[1] != inode->i_generation) {
179 LDLM_ERROR(lock, "data mismatch with ino %lu/%u (%p)",
180 inode->i_ino, inode->i_generation, inode);
183 if (bits & MDS_INODELOCK_OPEN) {
185 switch (lock->l_req_mode) {
192 CERROR("open PR lock without FMODE_EXEC\n");
198 CERROR("Unexpected lock mode for OPEN lock "
199 "%d, inode %ld\n", lock->l_req_mode,
202 ll_mdc_real_close(inode, flags);
205 if (bits & MDS_INODELOCK_UPDATE)
206 clear_bit(LLI_F_HAVE_MDS_SIZE_LOCK,
207 &(ll_i2info(inode)->lli_flags));
209 if (S_ISDIR(inode->i_mode) &&
210 (bits & MDS_INODELOCK_UPDATE)) {
211 struct dentry *dentry, *tmp, *dir;
212 int alias_counter = 0;
214 CDEBUG(D_INODE, "invalidating inode %lu\n",
216 truncate_inode_pages(inode->i_mapping, 0);
218 /* Drop possible cached negative dentries */
220 spin_lock(&dcache_lock);
222 /* It is possible to have several dentries (with
224 list_for_each_entry_safe(dentry, tmp,
225 &inode->i_dentry,d_alias) {
226 if (!list_empty(&dentry->d_subdirs))
231 if (alias_counter > 1)
232 CWARN("More than 1 alias dir %lu alias %d\n",
233 inode->i_ino, alias_counter);
237 list_for_each_entry_safe(dentry, tmp,
241 /* XXX Print some debug here? */
242 if (!dentry->d_inode)
243 /* Negative dentry. If we were
244 dropping dcache lock, go
245 throught the list again */
246 if (ll_drop_dentry(dentry))
250 spin_unlock(&dcache_lock);
253 if (inode->i_sb->s_root &&
254 inode != inode->i_sb->s_root->d_inode &&
255 (bits & MDS_INODELOCK_LOOKUP))
256 ll_unhash_aliases(inode);
267 int ll_mdc_cancel_unused(struct lustre_handle *conn, struct inode *inode,
268 int flags, void *opaque)
270 struct ldlm_res_id res_id =
271 { .name = {inode->i_ino, inode->i_generation} };
272 struct obd_device *obddev = class_conn2obd(conn);
275 RETURN(ldlm_cli_cancel_unused(obddev->obd_namespace, &res_id, flags,
279 /* Pack the required supplementary groups into the supplied groups array.
280 * If we don't need to use the groups from the target inode(s) then we
281 * instead pack one or more groups from the user's supplementary group
282 * array in case it might be useful. Not needed if doing an MDS-side upcall. */
283 void ll_i2gids(__u32 *suppgids, struct inode *i1, struct inode *i2)
288 LASSERT(suppgids != NULL);
290 if (in_group_p(i1->i_gid))
291 suppgids[0] = i1->i_gid;
296 if (in_group_p(i2->i_gid))
297 suppgids[1] = i2->i_gid;
304 for (i = 0; i < current_ngroups; i++) {
305 if (suppgids[0] == -1) {
306 if (current_groups[i] != suppgids[1])
307 suppgids[0] = current_groups[i];
310 if (suppgids[1] == -1) {
311 if (current_groups[i] != suppgids[0])
312 suppgids[1] = current_groups[i];
319 int ll_prepare_mdc_op_data(struct mdc_op_data *data, struct inode *i1,
320 struct inode *i2, const char *name, int namelen,
325 if (namelen > ll_i2sbi(i1)->ll_namelen)
326 return -ENAMETOOLONG;
327 ll_i2gids(data->suppgids, i1, i2);
328 ll_inode2fid(&data->fid1, i1);
331 ll_inode2fid(&data->fid2, i2);
333 memset(&data->fid2, 0, sizeof(data->fid2));
336 data->namelen = namelen;
337 data->create_mode = mode;
338 data->mod_time = CURRENT_SECONDS;
343 static void ll_d_add(struct dentry *de, struct inode *inode)
345 CDEBUG(D_DENTRY, "adding inode %p to dentry %p\n", inode, de);
347 if (!list_empty(&de->d_alias)) {
348 spin_unlock(&dcache_lock);
349 CERROR("dentry %.*s %p alias next %p, prev %p\n",
350 de->d_name.len, de->d_name.name, de,
351 de->d_alias.next, de->d_alias.prev);
355 list_add(&de->d_alias, &inode->i_dentry);
359 if (!d_unhashed(de)) {
360 spin_unlock(&dcache_lock);
361 CERROR("dentry %.*s %p hash next %p\n",
362 de->d_name.len, de->d_name.name, de, de->d_hash.next);
368 /* Search "inode"'s alias list for a dentry that has the same name and parent
369 * as de. If found, return it. If not found, return de.
370 * Lustre can't use d_add_unique because don't unhash aliases for directory
371 * in ll_revalidate_it. After revaliadate inode will be have hashed aliases
372 * and it triggers BUG_ON in d_instantiate_unique (bug #10954).
374 struct dentry *ll_find_alias(struct inode *inode, struct dentry *de)
376 struct list_head *tmp;
377 struct dentry *dentry;
378 struct dentry *last_discon = NULL;
380 spin_lock(&dcache_lock);
381 list_for_each(tmp, &inode->i_dentry) {
382 dentry = list_entry(tmp, struct dentry, d_alias);
384 /* We are called here with 'de' already on the aliases list. */
385 if (unlikely(dentry == de)) {
390 if (dentry->d_flags & DCACHE_DISCONNECTED) {
391 LASSERT(last_discon == NULL);
392 last_discon = dentry;
396 if (dentry->d_parent != de->d_parent)
399 if (dentry->d_name.hash != de->d_name.hash)
402 if (dentry->d_name.len != de->d_name.len)
405 if (memcmp(dentry->d_name.name, de->d_name.name,
406 de->d_name.len) != 0)
412 #ifdef LUSTRE_KERNEL_VERSION
413 dentry->d_flags &= ~DCACHE_LUSTRE_INVALID;
415 unlock_dentry(dentry);
416 __d_rehash(dentry, 0); /* avoid taking dcache_lock inside */
417 spin_unlock(&dcache_lock);
419 CDEBUG(D_DENTRY, "alias dentry %.*s (%p) parent %p inode %p "
420 "refc %d\n", de->d_name.len, de->d_name.name, de,
421 de->d_parent, de->d_inode, atomic_read(&de->d_count));
425 CDEBUG(D_DENTRY, "Reuse disconnected dentry %p inode %p "
426 "refc %d\n", last_discon, last_discon->d_inode,
427 atomic_read(&last_discon->d_count));
428 dget_locked(last_discon);
429 spin_unlock(&dcache_lock);
431 d_move(last_discon, de);
438 spin_unlock(&dcache_lock);
443 static int lookup_it_finish(struct ptlrpc_request *request, int offset,
444 struct lookup_intent *it, void *data)
446 struct it_cb_data *icbd = data;
447 struct dentry **de = icbd->icbd_childp;
448 struct inode *parent = icbd->icbd_parent;
449 struct ll_sb_info *sbi = ll_i2sbi(parent);
450 struct inode *inode = NULL;
453 /* NB 1 request reference will be taken away by ll_intent_lock()
455 if (!it_disposition(it, DISP_LOOKUP_NEG)) {
458 rc = ll_prep_inode(sbi->ll_osc_exp, &inode, request, offset,
463 CDEBUG(D_DLMTRACE, "setting l_data to inode %p (%lu/%u)\n",
464 inode, inode->i_ino, inode->i_generation);
465 mdc_set_lock_data(&it->d.lustre.it_lock_handle, inode);
467 /* We used to query real size from OSTs here, but actually
468 this is not needed. For stat() calls size would be updated
469 from subsequent do_revalidate()->ll_inode_revalidate_it() in
471 vfs_getattr_it->ll_getattr()->ll_inode_revalidate_it() in 2.6
472 Everybody else who needs correct file size would call
473 ll_glimpse_size or some equivalent themselves anyway.
474 Also see bug 7198. */
475 *de = ll_find_alias(inode, *de);
478 /* Check that parent has UPDATE lock. If there is none, we
479 cannot afford to hash this dentry (done by ll_d_add) as it
480 might get picked up later when UPDATE lock will appear */
481 if (ll_have_md_lock(parent, MDS_INODELOCK_UPDATE)) {
482 spin_lock(&dcache_lock);
483 ll_d_add(*de, inode);
484 spin_unlock(&dcache_lock);
486 /* We do not want to hash the dentry if don`t have a
487 * lock, but if this dentry is later used in d_move,
488 * we'd hit uninitialised list head d_hash, so we just
489 * do this to init d_hash field but leave dentry
490 * unhashed. (bug 10796). */
497 (*de)->d_op = &ll_d_ops;
502 static struct dentry *ll_lookup_it(struct inode *parent, struct dentry *dentry,
503 struct lookup_intent *it, int lookup_flags)
505 struct dentry *save = dentry, *retval;
506 struct mdc_op_data op_data;
507 struct it_cb_data icbd;
508 struct ptlrpc_request *req = NULL;
509 struct lookup_intent lookup_it = { .it_op = IT_LOOKUP };
513 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),intent=%s\n",
514 dentry->d_name.len, dentry->d_name.name, parent->i_ino,
515 parent->i_generation, parent, LL_IT2STR(it));
517 if (d_mountpoint(dentry))
518 CERROR("Tell Peter, lookup on mtpt, it %s\n", LL_IT2STR(it));
520 ll_frob_intent(&it, &lookup_it);
522 /* As do_lookup is called before follow_mount, root dentry may be left
523 * not valid, revalidate it here. */
524 if (parent->i_sb->s_root && (parent->i_sb->s_root->d_inode == parent) &&
525 (it->it_op & (IT_OPEN | IT_CREAT))) {
526 rc = ll_inode_revalidate_it(parent->i_sb->s_root, it);
531 icbd.icbd_childp = &dentry;
532 icbd.icbd_parent = parent;
534 rc = ll_prepare_mdc_op_data(&op_data, parent, NULL, dentry->d_name.name,
535 dentry->d_name.len, lookup_flags);
539 it->it_create_mode &= ~current->fs->umask;
541 rc = mdc_intent_lock(ll_i2mdcexp(parent), &op_data, NULL, 0, it,
542 lookup_flags, &req, ll_mdc_blocking_ast, 0);
545 GOTO(out, retval = ERR_PTR(rc));
547 rc = lookup_it_finish(req, DLM_REPLY_REC_OFF, it, &icbd);
549 ll_intent_release(it);
550 GOTO(out, retval = ERR_PTR(rc));
553 if ((it->it_op & IT_OPEN) && dentry->d_inode &&
554 !S_ISREG(dentry->d_inode->i_mode) &&
555 !S_ISDIR(dentry->d_inode->i_mode)) {
556 ll_release_openhandle(dentry, it);
558 ll_lookup_finish_locks(it, dentry);
561 GOTO(out, retval = NULL);
563 GOTO(out, retval = dentry);
566 ptlrpc_req_finished(req);
570 #if (LINUX_VERSION_CODE > KERNEL_VERSION(2,5,0))
571 #ifdef LUSTRE_KERNEL_VERSION
572 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
573 struct nameidata *nd)
578 if (nd && nd->flags & LOOKUP_LAST && !(nd->flags & LOOKUP_LINK_NOTLAST))
579 de = ll_lookup_it(parent, dentry, &nd->intent, nd->flags);
581 de = ll_lookup_it(parent, dentry, NULL, 0);
586 struct lookup_intent *ll_convert_intent(struct open_intent *oit,
589 struct lookup_intent *it;
591 OBD_ALLOC(it, sizeof(*it));
593 return ERR_PTR(-ENOMEM);
595 if (lookup_flags & LOOKUP_OPEN) {
597 if (lookup_flags & LOOKUP_CREATE)
598 it->it_op |= IT_CREAT;
599 it->it_create_mode = oit->create_mode;
600 it->it_flags = oit->flags;
602 it->it_op = IT_GETATTR;
605 #ifndef HAVE_FILE_IN_STRUCT_INTENT
606 /* Since there is no way to pass our intent to ll_file_open,
607 * just check the file is there. Actual open will be done
609 if (it->it_op & IT_OPEN)
610 it->it_op = IT_LOOKUP;
616 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
617 struct nameidata *nd)
622 if (nd && !(nd->flags & (LOOKUP_CONTINUE|LOOKUP_PARENT))) {
623 struct lookup_intent *it;
625 #if defined(HAVE_FILE_IN_STRUCT_INTENT) && (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,17))
626 /* Did we came here from failed revalidate just to propagate
628 if (nd->flags & LOOKUP_OPEN)
629 if (IS_ERR(nd->intent.open.file))
630 RETURN((struct dentry *)nd->intent.open.file);
633 if (ll_d2d(dentry) && ll_d2d(dentry)->lld_it) {
634 it = ll_d2d(dentry)->lld_it;
635 ll_d2d(dentry)->lld_it = NULL;
637 it = ll_convert_intent(&nd->intent.open, nd->flags);
639 RETURN((struct dentry *)it);
642 de = ll_lookup_it(parent, dentry, it, nd->flags);
645 if ((nd->flags & LOOKUP_OPEN) && !IS_ERR(dentry)) { /* Open */
646 if (dentry->d_inode &&
647 it_disposition(it, DISP_OPEN_OPEN)) { /* nocreate */
648 #ifdef HAVE_FILE_IN_STRUCT_INTENT
649 if (S_ISFIFO(dentry->d_inode->i_mode)) {
650 // We cannot call open here as it would
653 (struct ptlrpc_request *)
654 it->d.lustre.it_data);
657 nd->intent.open.file->private_data = it;
658 filp =lookup_instantiate_filp(nd,dentry,
660 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,17))
661 /* 2.6.1[456] have a bug in open_namei() that forgets to check
662 * nd->intent.open.file for error, so we need to return it as lookup's result
667 de = (struct dentry *) filp;
672 #else /* HAVE_FILE_IN_STRUCT_INTENT */
673 /* Release open handle as we have no way to
674 * pass it to ll_file_open */
675 ll_release_openhandle(dentry, it);
676 #endif /* HAVE_FILE_IN_STRUCT_INTENT */
677 } else if (it_disposition(it, DISP_OPEN_CREATE)) {
678 // XXX This can only reliably work on assumption
679 // that there are NO hashed negative dentries.
680 ll_d2d(dentry)->lld_it = it;
681 it = NULL; /* Will be freed in ll_create_nd */
682 /* We absolutely depend on ll_create_nd to be
683 * called to not leak this intent and possible
684 * data attached to it */
689 ll_intent_release(it);
690 OBD_FREE(it, sizeof(*it));
693 de = ll_lookup_it(parent, dentry, NULL, 0);
701 /* We depend on "mode" being set with the proper file type/umask by now */
702 static struct inode *ll_create_node(struct inode *dir, const char *name,
703 int namelen, const void *data, int datalen,
704 int mode, __u64 extra,
705 struct lookup_intent *it)
707 struct inode *inode = NULL;
708 struct ptlrpc_request *request = NULL;
709 struct ll_sb_info *sbi = ll_i2sbi(dir);
713 LASSERT(it && it->d.lustre.it_disposition);
715 LASSERT(it_disposition(it, DISP_ENQ_CREATE_REF));
716 request = it->d.lustre.it_data;
717 it_clear_disposition(it, DISP_ENQ_CREATE_REF);
718 rc = ll_prep_inode(sbi->ll_osc_exp, &inode, request, DLM_REPLY_REC_OFF,
721 GOTO(out, inode = ERR_PTR(rc));
723 LASSERT(list_empty(&inode->i_dentry));
725 /* We asked for a lock on the directory, but were granted a
726 * lock on the inode. Since we finally have an inode pointer,
727 * stuff it in the lock. */
728 CDEBUG(D_DLMTRACE, "setting l_ast_data to inode %p (%lu/%u)\n",
729 inode, inode->i_ino, inode->i_generation);
730 mdc_set_lock_data(&it->d.lustre.it_lock_handle, inode);
733 ptlrpc_req_finished(request);
738 * By the time this is called, we already have created the directory cache
739 * entry for the new file, but it is so far negative - it has no inode.
741 * We defer creating the OBD object(s) until open, to keep the intent and
742 * non-intent code paths similar, and also because we do not have the MDS
743 * inode number before calling ll_create_node() (which is needed for LOV),
744 * so we would need to do yet another RPC to the MDS to store the LOV EA
745 * data on the MDS. If needed, we would pass the PACKED lmm as data and
746 * lmm_size in datalen (the MDS still has code which will handle that).
748 * If the create succeeds, we fill in the inode information
749 * with d_instantiate().
751 static int ll_create_it(struct inode *dir, struct dentry *dentry, int mode,
752 struct lookup_intent *it)
755 struct ptlrpc_request *request = it->d.lustre.it_data;
759 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),intent=%s\n",
760 dentry->d_name.len, dentry->d_name.name, dir->i_ino,
761 dir->i_generation, dir, LL_IT2STR(it));
763 rc = it_open_error(DISP_OPEN_CREATE, it);
767 mdc_store_inode_generation(request, DLM_INTENT_REC_OFF,
769 inode = ll_create_node(dir, dentry->d_name.name, dentry->d_name.len,
770 NULL, 0, mode, 0, it);
772 RETURN(PTR_ERR(inode));
775 d_instantiate(dentry, inode);
776 /* Negative dentry may be unhashed if parent does not have UPDATE lock,
777 * but some callers, e.g. do_coredump, expect dentry to be hashed after
778 * successful create. Hash it here. */
779 spin_lock(&dcache_lock);
780 if (d_unhashed(dentry))
781 __d_rehash(dentry, 0);
782 spin_unlock(&dcache_lock);
786 static void ll_update_times(struct ptlrpc_request *request, int offset,
789 struct mds_body *body = lustre_msg_buf(request->rq_repmsg, offset,
793 /* mtime is always updated with ctime, but can be set in past.
794 As write and utime(2) may happen within 1 second, and utime's
795 mtime has a priority over write's one, so take mtime from mds
796 for the same ctimes. */
797 if (body->valid & OBD_MD_FLCTIME &&
798 body->ctime >= LTIME_S(inode->i_ctime)) {
799 LTIME_S(inode->i_ctime) = body->ctime;
801 if (body->valid & OBD_MD_FLMTIME) {
802 CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
803 "to "LPU64"\n", inode->i_ino,
804 LTIME_S(inode->i_mtime), body->mtime);
805 LTIME_S(inode->i_mtime) = body->mtime;
810 static int ll_new_node(struct inode *dir, struct qstr *name,
811 const char *tgt, int mode,
812 int rdev, struct dentry *dchild)
814 struct ptlrpc_request *request = NULL;
815 struct inode *inode = NULL;
816 struct ll_sb_info *sbi = ll_i2sbi(dir);
817 struct mdc_op_data op_data;
822 if (unlikely(tgt != NULL))
823 tgt_len = strlen(tgt)+1;
825 err = ll_prepare_mdc_op_data(&op_data, dir, NULL, name->name,
830 err = mdc_create(sbi->ll_mdc_exp, &op_data, tgt, tgt_len,
831 mode, current->fsuid, current->fsgid,
832 current->cap_effective, rdev, &request);
836 ll_update_times(request, REPLY_REC_OFF, dir);
839 err = ll_prep_inode(sbi->ll_osc_exp, &inode, request,
840 REPLY_REC_OFF, dchild->d_sb);
845 d_instantiate(dchild, inode);
849 ptlrpc_req_finished(request);
855 static int ll_mknod_generic(struct inode *dir, struct qstr *name, int mode,
856 unsigned rdev, struct dentry *dchild)
861 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p) mode %o dev %x\n",
862 name->len, name->name, dir->i_ino, dir->i_generation, dir,
865 mode &= ~current->fs->umask;
867 switch (mode & S_IFMT) {
869 mode |= S_IFREG; /* for mode = 0 case, fallthrough */
875 err = ll_new_node(dir, name, NULL, mode, rdev, dchild);
886 #if (LINUX_VERSION_CODE > KERNEL_VERSION(2,5,0))
887 #ifndef LUSTRE_KERNEL_VERSION
888 static int ll_create_nd(struct inode *dir, struct dentry *dentry, int mode, struct nameidata *nd)
890 struct lookup_intent *it = ll_d2d(dentry)->lld_it;
894 return ll_mknod_generic(dir, &dentry->d_name, mode, 0, dentry);
896 ll_d2d(dentry)->lld_it = NULL;
898 /* Was there an error? Propagate it! */
899 if (it->d.lustre.it_status) {
900 rc = it->d.lustre.it_status;
904 rc = ll_create_it(dir, dentry, mode, it);
905 #ifdef HAVE_FILE_IN_STRUCT_INTENT
906 if (nd && (nd->flags & LOOKUP_OPEN) && dentry->d_inode) { /* Open */
907 nd->intent.open.file->private_data = it;
908 lookup_instantiate_filp(nd, dentry, NULL);
911 ll_release_openhandle(dentry,it);
915 ll_intent_release(it);
916 OBD_FREE(it, sizeof(*it));
921 static int ll_create_nd(struct inode *dir, struct dentry *dentry, int mode, struct nameidata *nd)
924 if (!nd || !nd->intent.d.lustre.it_disposition)
925 /* No saved request? Just mknod the file */
926 return ll_mknod_generic(dir, &dentry->d_name, mode, 0, dentry);
928 return ll_create_it(dir, dentry, mode, &nd->intent);
933 static int ll_symlink_generic(struct inode *dir, struct qstr *name,
934 const char *tgt, struct dentry *dchild)
939 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),target=%.*s\n",
940 name->len, name->name, dir->i_ino, dir->i_generation,
943 err = ll_new_node(dir, name, (char *)tgt, S_IFLNK | S_IRWXUGO,
948 static int ll_link_generic(struct inode *src, struct inode *dir,
949 struct qstr *name, struct dentry *dchild)
951 struct ptlrpc_request *request = NULL;
952 struct mdc_op_data op_data;
954 struct ll_sb_info *sbi = ll_i2sbi(dir);
958 "VFS Op: inode=%lu/%u(%p), dir=%lu/%u(%p), target=%.*s\n",
959 src->i_ino, src->i_generation, src, dir->i_ino,
960 dir->i_generation, dir, name->len, name->name);
962 err = ll_prepare_mdc_op_data(&op_data, src, dir, name->name,
966 err = mdc_link(sbi->ll_mdc_exp, &op_data, &request);
973 ll_update_times(request, REPLY_REC_OFF, dir);
977 ptlrpc_req_finished(request);
981 static int ll_mkdir_generic(struct inode *dir, struct qstr *name, int mode,
982 struct dentry *dchild)
988 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
989 name->len, name->name, dir->i_ino, dir->i_generation, dir);
991 mode = (mode & (S_IRWXUGO|S_ISVTX) & ~current->fs->umask) | S_IFDIR;
992 err = ll_new_node(dir, name, NULL, mode, 0, dchild);
997 static int ll_rmdir_generic(struct inode *dir, struct dentry *dparent,
1000 struct ptlrpc_request *request = NULL;
1001 struct mdc_op_data op_data;
1002 struct dentry *dentry;
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 /* Check if we have something mounted at the dir we are going to delete
1009 * In such a case there would always be dentry present. */
1011 dentry = d_lookup(dparent, name);
1013 int mounted = d_mountpoint(dentry);
1016 GOTO(out, rc = -EBUSY);
1020 rc = ll_prepare_mdc_op_data(&op_data, dir, NULL, name->name,
1021 name->len, S_IFDIR);
1024 rc = mdc_unlink(ll_i2sbi(dir)->ll_mdc_exp, &op_data, &request);
1027 ll_update_times(request, REPLY_REC_OFF, dir);
1031 ptlrpc_req_finished(request);
1035 int ll_objects_destroy(struct ptlrpc_request *request, struct inode *dir)
1037 struct mds_body *body;
1038 struct lov_mds_md *eadata;
1039 struct lov_stripe_md *lsm = NULL;
1040 struct obd_trans_info oti = { 0 };
1045 /* req is swabbed so this is safe */
1046 body = lustre_msg_buf(request->rq_repmsg, REPLY_REC_OFF, sizeof(*body));
1048 if (!(body->valid & OBD_MD_FLEASIZE))
1051 if (body->eadatasize == 0) {
1052 CERROR("OBD_MD_FLEASIZE set but eadatasize zero\n");
1053 GOTO(out, rc = -EPROTO);
1056 /* The MDS sent back the EA because we unlinked the last reference
1057 * to this file. Use this EA to unlink the objects on the OST.
1058 * It's opaque so we don't swab here; we leave it to obd_unpackmd() to
1059 * check it is complete and sensible. */
1060 eadata = lustre_swab_repbuf(request, REPLY_REC_OFF + 1,
1061 body->eadatasize, NULL);
1062 LASSERT(eadata != NULL);
1063 if (eadata == NULL) {
1064 CERROR("Can't unpack MDS EA data\n");
1065 GOTO(out, rc = -EPROTO);
1068 rc = obd_unpackmd(ll_i2obdexp(dir), &lsm, eadata, body->eadatasize);
1070 CERROR("obd_unpackmd: %d\n", rc);
1073 LASSERT(rc >= sizeof(*lsm));
1075 rc = obd_checkmd(ll_i2obdexp(dir), ll_i2mdcexp(dir), lsm);
1077 GOTO(out_free_memmd, rc);
1081 GOTO(out_free_memmd, rc = -ENOMEM);
1083 oa->o_id = lsm->lsm_object_id;
1084 oa->o_mode = body->mode & S_IFMT;
1085 oa->o_valid = OBD_MD_FLID | OBD_MD_FLTYPE;
1087 if (body->valid & OBD_MD_FLCOOKIE) {
1088 oa->o_valid |= OBD_MD_FLCOOKIE;
1089 oti.oti_logcookies =
1090 lustre_msg_buf(request->rq_repmsg, REPLY_REC_OFF + 2,
1091 sizeof(struct llog_cookie) *
1092 lsm->lsm_stripe_count);
1093 if (oti.oti_logcookies == NULL) {
1094 oa->o_valid &= ~OBD_MD_FLCOOKIE;
1095 body->valid &= ~OBD_MD_FLCOOKIE;
1099 rc = obd_destroy(ll_i2obdexp(dir), oa, lsm, &oti, ll_i2mdcexp(dir));
1102 CERROR("obd destroy objid "LPX64" error %d\n",
1103 lsm->lsm_object_id, rc);
1105 obd_free_memmd(ll_i2obdexp(dir), &lsm);
1110 static int ll_unlink_generic(struct inode * dir, struct qstr *name)
1112 struct ptlrpc_request *request = NULL;
1113 struct mdc_op_data op_data;
1117 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
1118 name->len, name->name, dir->i_ino, dir->i_generation, dir);
1120 rc = ll_prepare_mdc_op_data(&op_data, dir, NULL, name->name,
1124 rc = mdc_unlink(ll_i2sbi(dir)->ll_mdc_exp, &op_data, &request);
1128 ll_update_times(request, REPLY_REC_OFF, dir);
1130 rc = ll_objects_destroy(request, dir);
1135 ptlrpc_req_finished(request);
1139 static int ll_rename_generic(struct inode *src, struct qstr *src_name,
1140 struct inode *tgt, struct qstr *tgt_name)
1142 struct ptlrpc_request *request = NULL;
1143 struct ll_sb_info *sbi = ll_i2sbi(src);
1144 struct mdc_op_data op_data;
1148 CDEBUG(D_VFSTRACE,"VFS Op:oldname=%.*s,src_dir=%lu/%u(%p),newname=%.*s,"
1149 "tgt_dir=%lu/%u(%p)\n", src_name->len, src_name->name,
1150 src->i_ino, src->i_generation, src, tgt_name->len,
1151 tgt_name->name, tgt->i_ino, tgt->i_generation, tgt);
1153 err = ll_prepare_mdc_op_data(&op_data, src, tgt, NULL, 0, 0);
1156 err = mdc_rename(sbi->ll_mdc_exp, &op_data,
1157 src_name->name, src_name->len,
1158 tgt_name->name, tgt_name->len, &request);
1161 ll_update_times(request, REPLY_REC_OFF, src);
1162 ll_update_times(request, REPLY_REC_OFF, tgt);
1163 err = ll_objects_destroy(request, src);
1169 ptlrpc_req_finished(request);
1174 #ifdef LUSTRE_KERNEL_VERSION
1175 static int ll_mknod_raw(struct nameidata *nd, int mode, dev_t rdev)
1177 return ll_mknod_generic(nd->dentry->d_inode, &nd->last, mode,rdev,NULL);
1179 static int ll_rename_raw(struct nameidata *srcnd, struct nameidata *tgtnd)
1181 return ll_rename_generic(srcnd->dentry->d_inode, &srcnd->last,
1182 tgtnd->dentry->d_inode, &tgtnd->last);
1184 static int ll_link_raw(struct nameidata *srcnd, struct nameidata *tgtnd)
1186 return ll_link_generic(srcnd->dentry->d_inode, tgtnd->dentry->d_inode,
1187 &tgtnd->last, NULL);
1189 static int ll_symlink_raw(struct nameidata *nd, const char *tgt)
1191 return ll_symlink_generic(nd->dentry->d_inode, &nd->last, tgt, NULL);
1193 static int ll_rmdir_raw(struct nameidata *nd)
1195 return ll_rmdir_generic(nd->dentry->d_inode, nd->dentry, &nd->last);
1197 static int ll_mkdir_raw(struct nameidata *nd, int mode)
1199 return ll_mkdir_generic(nd->dentry->d_inode, &nd->last, mode, NULL);
1201 static int ll_unlink_raw(struct nameidata *nd)
1203 return ll_unlink_generic(nd->dentry->d_inode, &nd->last);
1207 static int ll_mknod(struct inode *dir, struct dentry *dchild, int mode,
1210 return ll_mknod_generic(dir, &dchild->d_name, mode,
1211 old_encode_dev(rdev), dchild);
1214 #if (LINUX_VERSION_CODE > KERNEL_VERSION(2,5,0))
1215 static int ll_unlink(struct inode * dir, struct dentry *dentry)
1217 return ll_unlink_generic(dir, &dentry->d_name);
1219 static int ll_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1221 return ll_mkdir_generic(dir, &dentry->d_name, mode, dentry);
1223 static int ll_rmdir(struct inode *dir, struct dentry *dentry)
1225 return ll_rmdir_generic(dir, NULL, &dentry->d_name);
1227 static int ll_symlink(struct inode *dir, struct dentry *dentry,
1228 const char *oldname)
1230 return ll_symlink_generic(dir, &dentry->d_name, oldname, dentry);
1232 static int ll_link(struct dentry *old_dentry, struct inode *dir,
1233 struct dentry *new_dentry)
1235 return ll_link_generic(old_dentry->d_inode, dir,
1236 &new_dentry->d_name, new_dentry);
1238 static int ll_rename(struct inode *old_dir, struct dentry *old_dentry,
1239 struct inode *new_dir, struct dentry *new_dentry)
1241 return ll_rename_generic(old_dir, &old_dentry->d_name, new_dir,
1242 &new_dentry->d_name);
1246 struct inode_operations ll_dir_inode_operations = {
1247 #ifdef LUSTRE_KERNEL_VERSION
1248 .link_raw = ll_link_raw,
1249 .unlink_raw = ll_unlink_raw,
1250 .symlink_raw = ll_symlink_raw,
1251 .mkdir_raw = ll_mkdir_raw,
1252 .rmdir_raw = ll_rmdir_raw,
1253 .mknod_raw = ll_mknod_raw,
1254 .rename_raw = ll_rename_raw,
1255 .setattr = ll_setattr,
1256 .setattr_raw = ll_setattr_raw,
1259 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
1260 .create_it = ll_create_it,
1261 .lookup_it = ll_lookup_it,
1262 .revalidate_it = ll_inode_revalidate_it,
1264 .lookup = ll_lookup_nd,
1265 .create = ll_create_nd,
1266 /* We need all these non-raw things for NFSD, to not patch it. */
1267 .unlink = ll_unlink,
1270 .symlink = ll_symlink,
1272 .rename = ll_rename,
1273 .setattr = ll_setattr,
1274 .getattr = ll_getattr,
1276 .permission = ll_inode_permission,
1277 .setxattr = ll_setxattr,
1278 .getxattr = ll_getxattr,
1279 .listxattr = ll_listxattr,
1280 .removexattr = ll_removexattr,
1283 struct inode_operations ll_special_inode_operations = {
1284 #ifdef LUSTRE_KERNEL_VERSION
1285 .setattr_raw = ll_setattr_raw,
1287 .setattr = ll_setattr,
1288 #if (LINUX_VERSION_CODE > KERNEL_VERSION(2,5,0))
1289 .getattr = ll_getattr,
1291 .revalidate_it = ll_inode_revalidate_it,
1293 .permission = ll_inode_permission,
1294 .setxattr = ll_setxattr,
1295 .getxattr = ll_getxattr,
1296 .listxattr = ll_listxattr,
1297 .removexattr = ll_removexattr,