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 /* called from iget5_locked->find_inode() under inode_lock spinlock */
99 static int fid_test_inode(struct inode *inode, void *opaque)
101 struct lustre_md *md = opaque;
103 if (unlikely(!(md->body->valid & OBD_MD_FLID))) {
104 CERROR("MDS body missing FID\n");
108 return lu_fid_eq(ll_inode_lu_fid(inode),
109 (struct lu_fid*)&md->body->fid1);
112 static int fid_set_inode(struct inode *inode, void *opaque)
114 struct lustre_md *md = opaque;
116 *ll_inode_lu_fid(inode) = *((struct lu_fid*)&md->body->fid1);
120 struct inode *ll_iget(struct super_block *sb, ino_t hash,
121 struct lustre_md *md)
123 struct ll_inode_info *lli;
128 inode = iget5_locked(sb, hash, fid_test_inode, fid_set_inode, md);
131 lli = ll_i2info(inode);
132 if (inode->i_state & I_NEW) {
133 ll_read_inode2(inode, md);
134 unlock_new_inode(inode);
136 if (!(inode->i_state & (I_FREEING | I_CLEAR)))
137 ll_update_inode(inode, md);
139 CDEBUG(D_VFSTRACE, "got inode: %lu/%u(%p) for "DFID"\n",
140 inode->i_ino, inode->i_generation, inode,
141 PFID(ll_inode_lu_fid(inode)));
147 static void ll_drop_negative_dentry(struct inode *dir)
149 struct dentry *dentry, *tmp_alias, *tmp_subdir;
151 spin_lock(&ll_lookup_lock);
152 spin_lock(&dcache_lock);
154 list_for_each_entry_safe(dentry, tmp_alias,
155 &dir->i_dentry,d_alias) {
156 if (!list_empty(&dentry->d_subdirs)) {
157 struct dentry *child;
158 list_for_each_entry_safe(child, tmp_subdir,
161 /* XXX Print some debug here? */
163 /* Negative dentry. If we were
164 dropping dcache lock, go
165 throught the list again */
166 if (ll_drop_dentry(child))
171 spin_unlock(&dcache_lock);
172 spin_unlock(&ll_lookup_lock);
175 int ll_mdc_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
176 void *data, int flag)
179 struct lustre_handle lockh;
183 case LDLM_CB_BLOCKING:
184 ldlm_lock2handle(lock, &lockh);
185 rc = ldlm_cli_cancel(&lockh);
187 CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
191 case LDLM_CB_CANCELING: {
192 struct inode *inode = ll_inode_from_lock(lock);
193 __u64 bits = lock->l_policy_data.l_inodebits.bits;
196 /* Invalidate all dentries associated with this inode */
200 fid = ll_inode_lu_fid(inode);;
202 LASSERT(lock->l_flags & LDLM_FL_CANCELING);
203 if ((bits & MDS_INODELOCK_LOOKUP) &&
204 ll_have_md_lock(inode, MDS_INODELOCK_LOOKUP))
205 bits &= ~MDS_INODELOCK_LOOKUP;
206 if ((bits & MDS_INODELOCK_UPDATE) &&
207 ll_have_md_lock(inode, MDS_INODELOCK_UPDATE))
208 bits &= ~MDS_INODELOCK_UPDATE;
209 if ((bits & MDS_INODELOCK_OPEN) &&
210 ll_have_md_lock(inode, MDS_INODELOCK_OPEN))
211 bits &= ~MDS_INODELOCK_OPEN;
213 if (!fid_res_name_eq(fid, &lock->l_resource->lr_name)) {
214 LDLM_ERROR(lock, "data mismatch with ino %lu/%u (%p)",
215 inode->i_ino, inode->i_generation, inode);
218 if (bits & MDS_INODELOCK_OPEN) {
220 switch (lock->l_req_mode) {
227 CERROR("open PR lock without FMODE_EXEC\n");
233 CERROR("Unexpected lock mode for OPEN lock "
234 "%d, inode %ld\n", lock->l_req_mode,
237 ll_mdc_real_close(inode, flags);
240 if (bits & MDS_INODELOCK_UPDATE)
241 clear_bit(LLI_F_HAVE_MDS_SIZE_LOCK,
242 &(ll_i2info(inode)->lli_flags));
244 if (S_ISDIR(inode->i_mode) &&
245 (bits & MDS_INODELOCK_UPDATE)) {
246 CDEBUG(D_INODE, "invalidating inode %lu\n",
248 truncate_inode_pages(inode->i_mapping, 0);
249 ll_drop_negative_dentry(inode);
250 inode->i_version++; /* XXX: remove with inode version*/
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 struct obd_device *obddev = class_conn2obd(conn);
274 fid_build_reg_res_name(ll_inode_lu_fid(inode), &res_id);
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 *op_data, struct inode *i1,
320 struct inode *i2, const char *name, int namelen,
321 int mode, void *data)
325 if (namelen > ll_i2sbi(i1)->ll_namelen)
326 return -ENAMETOOLONG;
327 ll_i2gids(op_data->suppgids, i1, i2);
328 ll_inode2fid(&op_data->fid1, i1);
331 ll_inode2fid(&op_data->fid2, i2);
333 memset(&op_data->fid2, 0, sizeof(op_data->fid2));
335 op_data->name = name;
336 op_data->namelen = namelen;
337 op_data->create_mode = mode;
338 op_data->mod_time = CURRENT_SECONDS;
339 op_data->data = data;
344 static void ll_d_add(struct dentry *de, struct inode *inode)
346 CDEBUG(D_DENTRY, "adding inode %p to dentry %p\n", inode, de);
348 if (!list_empty(&de->d_alias)) {
349 spin_unlock(&dcache_lock);
350 CERROR("dentry %.*s %p alias next %p, prev %p\n",
351 de->d_name.len, de->d_name.name, de,
352 de->d_alias.next, de->d_alias.prev);
356 list_add(&de->d_alias, &inode->i_dentry);
360 if (!d_unhashed(de)) {
361 spin_unlock(&dcache_lock);
362 CERROR("dentry %.*s %p hash next %p\n",
363 de->d_name.len, de->d_name.name, de, de->d_hash.next);
366 d_rehash_cond(de, 0);
369 /* Search "inode"'s alias list for a dentry that has the same name and parent
370 * as de. If found, return it. If not found, return de.
371 * Lustre can't use d_add_unique because don't unhash aliases for directory
372 * in ll_revalidate_it. After revaliadate inode will be have hashed aliases
373 * and it triggers BUG_ON in d_instantiate_unique (bug #10954).
375 static struct dentry *ll_find_alias(struct inode *inode, struct dentry *de)
377 struct list_head *tmp;
378 struct dentry *dentry;
379 struct dentry *last_discon = NULL;
381 spin_lock(&ll_lookup_lock);
382 spin_lock(&dcache_lock);
383 list_for_each(tmp, &inode->i_dentry) {
384 dentry = list_entry(tmp, struct dentry, d_alias);
386 /* We are called here with 'de' already on the aliases list. */
387 if (unlikely(dentry == de)) {
392 if (dentry->d_flags & DCACHE_DISCONNECTED) {
393 LASSERT(last_discon == NULL);
394 last_discon = dentry;
398 if (dentry->d_parent != de->d_parent)
401 if (dentry->d_name.hash != de->d_name.hash)
404 if (dentry->d_name.len != de->d_name.len)
407 if (memcmp(dentry->d_name.name, de->d_name.name,
408 de->d_name.len) != 0)
414 #ifdef DCACHE_LUSTRE_INVALID
415 dentry->d_flags &= ~DCACHE_LUSTRE_INVALID;
417 unlock_dentry(dentry);
418 ll_dops_init(dentry, 0);
419 d_rehash_cond(dentry, 0); /* avoid taking dcache_lock inside */
420 spin_unlock(&dcache_lock);
421 spin_unlock(&ll_lookup_lock);
423 CDEBUG(D_DENTRY, "alias dentry %.*s (%p) parent %p inode %p "
424 "refc %d\n", de->d_name.len, de->d_name.name, de,
425 de->d_parent, de->d_inode, atomic_read(&de->d_count));
430 CDEBUG(D_DENTRY, "Reuse disconnected dentry %p inode %p "
431 "refc %d\n", last_discon, last_discon->d_inode,
432 atomic_read(&last_discon->d_count));
433 dget_locked(last_discon);
434 spin_unlock(&dcache_lock);
435 spin_unlock(&ll_lookup_lock);
436 ll_dops_init(last_discon, 1);
438 d_move(last_discon, de);
445 spin_unlock(&dcache_lock);
446 spin_unlock(&ll_lookup_lock);
451 int lookup_it_finish(struct ptlrpc_request *request, int offset,
452 struct lookup_intent *it, void *data)
454 struct it_cb_data *icbd = data;
455 struct dentry **de = icbd->icbd_childp;
456 struct inode *parent = icbd->icbd_parent;
457 struct ll_sb_info *sbi = ll_i2sbi(parent);
458 struct inode *inode = NULL;
462 /* NB 1 request reference will be taken away by ll_intent_lock()
464 if (!it_disposition(it, DISP_LOOKUP_NEG)) {
465 struct dentry *save = *de;
467 rc = ll_prep_inode(sbi->ll_osc_exp, &inode, request, offset,
472 CDEBUG(D_DLMTRACE, "setting l_data to inode %p (%lu/%u)\n",
473 inode, inode->i_ino, inode->i_generation);
474 mdc_set_lock_data(&it->d.lustre.it_lock_handle, inode);
476 /* We used to query real size from OSTs here, but actually
477 this is not needed. For stat() calls size would be updated
478 from subsequent do_revalidate()->ll_inode_revalidate_it() in
480 vfs_getattr_it->ll_getattr()->ll_inode_revalidate_it() in 2.6
481 Everybody else who needs correct file size would call
482 ll_glimpse_size or some equivalent themselves anyway.
483 Also see bug 7198. */
485 ll_dops_init(*de, 1);
486 *de = ll_find_alias(inode, *de);
488 struct ll_dentry_data *lld = ll_d2d(*de);
490 /* just make sure the ll_dentry_data is ready */
491 if (unlikely(lld == NULL)) {
494 if (likely(lld != NULL))
495 lld->lld_sa_generation = 0;
499 ll_dops_init(*de, 1);
500 /* Check that parent has UPDATE lock. If there is none, we
501 cannot afford to hash this dentry (done by ll_d_add) as it
502 might get picked up later when UPDATE lock will appear */
503 if (ll_have_md_lock(parent, MDS_INODELOCK_UPDATE)) {
504 spin_lock(&dcache_lock);
505 ll_d_add(*de, inode);
506 spin_unlock(&dcache_lock);
508 (*de)->d_inode = NULL;
509 /* We do not want to hash the dentry if don`t have a
510 * lock, but if this dentry is later used in d_move,
511 * we'd hit uninitialised list head d_hash, so we just
512 * do this to init d_hash field but leave dentry
513 * unhashed. (bug 10796). */
522 static struct dentry *ll_lookup_it(struct inode *parent, struct dentry *dentry,
523 struct lookup_intent *it, int lookup_flags)
525 struct dentry *save = dentry, *retval;
526 struct mdc_op_data op_data = { { 0 } };
527 struct it_cb_data icbd;
528 struct ptlrpc_request *req = NULL;
529 struct lookup_intent lookup_it = { .it_op = IT_LOOKUP };
533 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),intent=%s\n",
534 dentry->d_name.len, dentry->d_name.name, parent->i_ino,
535 parent->i_generation, parent, LL_IT2STR(it));
537 if (d_mountpoint(dentry))
538 CERROR("Tell Peter, lookup on mtpt, it %s\n", LL_IT2STR(it));
540 ll_frob_intent(&it, &lookup_it);
542 /* As do_lookup is called before follow_mount, root dentry may be left
543 * not valid, revalidate it here. */
544 if (parent->i_sb->s_root && (parent->i_sb->s_root->d_inode == parent) &&
545 (it->it_op & (IT_OPEN | IT_CREAT))) {
546 rc = ll_inode_revalidate_it(parent->i_sb->s_root, it);
551 if (it->it_op == IT_GETATTR) {
552 first = ll_statahead_enter(parent, &dentry, 1);
554 ll_statahead_exit(dentry, first);
556 RETURN(retval = dentry);
560 icbd.icbd_parent = parent;
561 icbd.icbd_childp = &dentry;
563 rc = ll_prepare_mdc_op_data(&op_data, parent, NULL, dentry->d_name.name,
564 dentry->d_name.len, lookup_flags, NULL);
568 it->it_create_mode &= ~current->fs->umask;
570 rc = mdc_intent_lock(ll_i2mdcexp(parent), &op_data, NULL, 0, it,
571 lookup_flags, &req, ll_mdc_blocking_ast, 0);
574 GOTO(out, retval = ERR_PTR(rc));
576 rc = lookup_it_finish(req, DLM_REPLY_REC_OFF, it, &icbd);
578 ll_intent_release(it);
579 GOTO(out, retval = ERR_PTR(rc));
582 if (first == -EEXIST)
583 ll_statahead_mark(dentry);
585 if ((it->it_op & IT_OPEN) && dentry->d_inode &&
586 !S_ISREG(dentry->d_inode->i_mode) &&
587 !S_ISDIR(dentry->d_inode->i_mode)) {
588 ll_release_openhandle(dentry, it);
590 ll_lookup_finish_locks(it, dentry);
593 GOTO(out, retval = NULL);
595 GOTO(out, retval = dentry);
598 ptlrpc_req_finished(req);
602 #ifdef HAVE_VFS_INTENT_PATCHES
603 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
604 struct nameidata *nd)
609 if (nd && nd->flags & LOOKUP_LAST && !(nd->flags & LOOKUP_LINK_NOTLAST))
610 de = ll_lookup_it(parent, dentry, &nd->intent, nd->flags);
612 de = ll_lookup_it(parent, dentry, NULL, 0);
617 struct lookup_intent *ll_convert_intent(struct open_intent *oit,
620 struct lookup_intent *it;
622 OBD_ALLOC(it, sizeof(*it));
624 return ERR_PTR(-ENOMEM);
626 if (lookup_flags & LOOKUP_OPEN) {
628 if (lookup_flags & LOOKUP_CREATE)
629 it->it_op |= IT_CREAT;
630 it->it_create_mode = oit->create_mode;
631 it->it_flags = oit->flags;
633 it->it_op = IT_GETATTR;
636 #ifndef HAVE_FILE_IN_STRUCT_INTENT
637 /* Since there is no way to pass our intent to ll_file_open,
638 * just check the file is there. Actual open will be done
640 if (it->it_op & IT_OPEN)
641 it->it_op = IT_LOOKUP;
647 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
648 struct nameidata *nd)
653 if (nd && !(nd->flags & (LOOKUP_CONTINUE|LOOKUP_PARENT))) {
654 struct lookup_intent *it;
656 #if defined(HAVE_FILE_IN_STRUCT_INTENT) && (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,17))
657 /* Did we came here from failed revalidate just to propagate
659 if (nd->flags & LOOKUP_OPEN)
660 if (IS_ERR(nd->intent.open.file))
661 RETURN((struct dentry *)nd->intent.open.file);
664 if (ll_d2d(dentry) && ll_d2d(dentry)->lld_it) {
665 it = ll_d2d(dentry)->lld_it;
666 ll_d2d(dentry)->lld_it = NULL;
668 it = ll_convert_intent(&nd->intent.open, nd->flags);
670 RETURN((struct dentry *)it);
673 de = ll_lookup_it(parent, dentry, it, nd->flags);
676 if ((nd->flags & LOOKUP_OPEN) && !IS_ERR(dentry)) { /* Open */
677 if (dentry->d_inode &&
678 it_disposition(it, DISP_OPEN_OPEN)) { /* nocreate */
679 #ifdef HAVE_FILE_IN_STRUCT_INTENT
680 if (S_ISFIFO(dentry->d_inode->i_mode)) {
681 // We cannot call open here as it would
684 (struct ptlrpc_request *)
685 it->d.lustre.it_data);
688 nd->intent.open.file->private_data = it;
689 filp =lookup_instantiate_filp(nd,dentry,
691 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,17))
692 /* 2.6.1[456] have a bug in open_namei() that forgets to check
693 * nd->intent.open.file for error, so we need to return it as lookup's result
698 de = (struct dentry *) filp;
703 #else /* HAVE_FILE_IN_STRUCT_INTENT */
704 /* Release open handle as we have no way to
705 * pass it to ll_file_open */
706 ll_release_openhandle(dentry, it);
707 #endif /* HAVE_FILE_IN_STRUCT_INTENT */
708 } else if (it_disposition(it, DISP_OPEN_CREATE)) {
709 // XXX This can only reliably work on assumption
710 // that there are NO hashed negative dentries.
711 ll_d2d(dentry)->lld_it = it;
712 it = NULL; /* Will be freed in ll_create_nd */
713 /* We absolutely depend on ll_create_nd to be
714 * called to not leak this intent and possible
715 * data attached to it */
720 ll_intent_release(it);
721 OBD_FREE(it, sizeof(*it));
724 de = ll_lookup_it(parent, dentry, NULL, 0);
731 /* We depend on "mode" being set with the proper file type/umask by now */
732 static struct inode *ll_create_node(struct inode *dir, const char *name,
733 int namelen, const void *data, int datalen,
734 int mode, __u64 extra,
735 struct lookup_intent *it)
737 struct inode *inode = NULL;
738 struct ptlrpc_request *request = NULL;
739 struct ll_sb_info *sbi = ll_i2sbi(dir);
743 LASSERT(it && it->d.lustre.it_disposition);
745 LASSERT(it_disposition(it, DISP_ENQ_CREATE_REF));
746 request = it->d.lustre.it_data;
747 it_clear_disposition(it, DISP_ENQ_CREATE_REF);
748 rc = ll_prep_inode(sbi->ll_osc_exp, &inode, request, DLM_REPLY_REC_OFF,
751 GOTO(out, inode = ERR_PTR(rc));
753 LASSERT(list_empty(&inode->i_dentry));
755 /* We asked for a lock on the directory, but were granted a
756 * lock on the inode. Since we finally have an inode pointer,
757 * stuff it in the lock. */
758 CDEBUG(D_DLMTRACE, "setting l_ast_data to inode %p (%lu/%u)\n",
759 inode, inode->i_ino, inode->i_generation);
760 mdc_set_lock_data(&it->d.lustre.it_lock_handle, inode);
763 ptlrpc_req_finished(request);
768 * By the time this is called, we already have created the directory cache
769 * entry for the new file, but it is so far negative - it has no inode.
771 * We defer creating the OBD object(s) until open, to keep the intent and
772 * non-intent code paths similar, and also because we do not have the MDS
773 * inode number before calling ll_create_node() (which is needed for LOV),
774 * so we would need to do yet another RPC to the MDS to store the LOV EA
775 * data on the MDS. If needed, we would pass the PACKED lmm as data and
776 * lmm_size in datalen (the MDS still has code which will handle that).
778 * If the create succeeds, we fill in the inode information
779 * with d_instantiate().
781 static int ll_create_it(struct inode *dir, struct dentry *dentry, int mode,
782 struct lookup_intent *it)
785 struct ptlrpc_request *request = it->d.lustre.it_data;
789 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),intent=%s\n",
790 dentry->d_name.len, dentry->d_name.name, dir->i_ino,
791 dir->i_generation, dir, LL_IT2STR(it));
793 rc = it_open_error(DISP_OPEN_CREATE, it);
797 mdc_store_inode_generation(request, DLM_INTENT_REC_OFF,
799 inode = ll_create_node(dir, dentry->d_name.name, dentry->d_name.len,
800 NULL, 0, mode, 0, it);
802 RETURN(PTR_ERR(inode));
805 d_instantiate(dentry, inode);
806 /* Negative dentry may be unhashed if parent does not have UPDATE lock,
807 * but some callers, e.g. do_coredump, expect dentry to be hashed after
808 * successful create. Hash it here. */
809 spin_lock(&dcache_lock);
810 if (d_unhashed(dentry))
811 d_rehash_cond(dentry, 0);
812 spin_unlock(&dcache_lock);
816 static void ll_update_times(struct ptlrpc_request *request, int offset,
819 struct mds_body *body = lustre_msg_buf(request->rq_repmsg, offset,
823 if (body->valid & OBD_MD_FLMTIME &&
824 body->mtime > LTIME_S(inode->i_mtime)) {
825 CDEBUG(D_INODE, "setting ino %lu mtime from %lu to "LPU64"\n",
826 inode->i_ino, LTIME_S(inode->i_mtime), body->mtime);
827 LTIME_S(inode->i_mtime) = body->mtime;
829 if (body->valid & OBD_MD_FLCTIME &&
830 body->ctime > LTIME_S(inode->i_ctime))
831 LTIME_S(inode->i_ctime) = body->ctime;
834 static int ll_new_node(struct inode *dir, struct qstr *name,
835 const char *tgt, int mode,
836 int rdev, struct dentry *dchild)
838 struct ptlrpc_request *request = NULL;
839 struct inode *inode = NULL;
840 struct ll_sb_info *sbi = ll_i2sbi(dir);
841 struct mdc_op_data op_data = { { 0 } };
846 if (unlikely(tgt != NULL))
847 tgt_len = strlen(tgt)+1;
849 err = ll_prepare_mdc_op_data(&op_data, dir, NULL, name->name,
854 err = mdc_create(sbi->ll_mdc_exp, &op_data, tgt, tgt_len,
855 mode, current->fsuid, current->fsgid,
856 cfs_curproc_cap_pack(), rdev, &request);
860 ll_update_times(request, REPLY_REC_OFF, dir);
863 err = ll_prep_inode(sbi->ll_osc_exp, &inode, request,
864 REPLY_REC_OFF, dchild->d_sb);
869 d_instantiate(dchild, inode);
873 ptlrpc_req_finished(request);
879 static int ll_mknod_generic(struct inode *dir, struct qstr *name, int mode,
880 unsigned rdev, struct dentry *dchild)
885 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p) mode %o dev %x\n",
886 name->len, name->name, dir->i_ino, dir->i_generation, dir,
889 mode &= ~current->fs->umask;
891 switch (mode & S_IFMT) {
893 mode |= S_IFREG; /* for mode = 0 case, fallthrough */
899 err = ll_new_node(dir, name, NULL, mode, rdev, dchild);
910 #ifndef HAVE_VFS_INTENT_PATCHES
911 static int ll_create_nd(struct inode *dir, struct dentry *dentry, int mode, struct nameidata *nd)
913 struct lookup_intent *it = ll_d2d(dentry)->lld_it;
917 return ll_mknod_generic(dir, &dentry->d_name, mode, 0, dentry);
919 ll_d2d(dentry)->lld_it = NULL;
921 /* Was there an error? Propagate it! */
922 if (it->d.lustre.it_status) {
923 rc = it->d.lustre.it_status;
927 rc = ll_create_it(dir, dentry, mode, it);
928 #ifdef HAVE_FILE_IN_STRUCT_INTENT
929 if (nd && (nd->flags & LOOKUP_OPEN) && dentry->d_inode) { /* Open */
930 nd->intent.open.file->private_data = it;
931 lookup_instantiate_filp(nd, dentry, NULL);
934 ll_release_openhandle(dentry,it);
938 ll_intent_release(it);
939 OBD_FREE(it, sizeof(*it));
944 static int ll_create_nd(struct inode *dir, struct dentry *dentry, int mode, struct nameidata *nd)
947 if (!nd || !nd->intent.d.lustre.it_disposition)
948 /* No saved request? Just mknod the file */
949 return ll_mknod_generic(dir, &dentry->d_name, mode, 0, dentry);
951 return ll_create_it(dir, dentry, mode, &nd->intent);
955 static int ll_symlink_generic(struct inode *dir, struct qstr *name,
956 const char *tgt, struct dentry *dchild)
961 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),target=%.*s\n",
962 name->len, name->name, dir->i_ino, dir->i_generation,
965 err = ll_new_node(dir, name, (char *)tgt, S_IFLNK | S_IRWXUGO,
970 static int ll_link_generic(struct inode *src, struct inode *dir,
971 struct qstr *name, struct dentry *dchild)
973 struct ptlrpc_request *request = NULL;
974 struct mdc_op_data op_data = { { 0 } };
976 struct ll_sb_info *sbi = ll_i2sbi(dir);
980 "VFS Op: inode=%lu/%u(%p), dir=%lu/%u(%p), target=%.*s\n",
981 src->i_ino, src->i_generation, src, dir->i_ino,
982 dir->i_generation, dir, name->len, name->name);
984 err = ll_prepare_mdc_op_data(&op_data, src, dir, name->name,
988 err = mdc_link(sbi->ll_mdc_exp, &op_data, &request);
995 ll_update_times(request, REPLY_REC_OFF, dir);
999 ptlrpc_req_finished(request);
1003 static int ll_mkdir_generic(struct inode *dir, struct qstr *name, int mode,
1004 struct dentry *dchild)
1010 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
1011 name->len, name->name, dir->i_ino, dir->i_generation, dir);
1013 mode = (mode & (S_IRWXUGO|S_ISVTX) & ~current->fs->umask) | S_IFDIR;
1014 err = ll_new_node(dir, name, NULL, mode, 0, dchild);
1019 /* Try to find the child dentry by its name.
1020 If found, put the result fid into @fid. */
1021 static void ll_get_child_fid(struct inode * dir, struct qstr *name,
1024 struct dentry *parent, *child;
1026 parent = list_entry(dir->i_dentry.next, struct dentry, d_alias);
1027 child = d_lookup(parent, name);
1030 ll_inode2fid(fid, child->d_inode);
1035 static int ll_rmdir_generic(struct inode *dir, struct dentry *dparent,
1038 struct ptlrpc_request *request = NULL;
1039 struct mdc_op_data op_data = { { 0 } };
1040 struct dentry *dentry;
1043 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
1044 name->len, name->name, dir->i_ino, dir->i_generation, dir);
1046 /* Check if we have something mounted at the dir we are going to delete
1047 * In such a case there would always be dentry present. */
1049 dentry = d_lookup(dparent, name);
1051 int mounted = d_mountpoint(dentry);
1054 GOTO(out, rc = -EBUSY);
1058 rc = ll_prepare_mdc_op_data(&op_data, dir, NULL, name->name,
1059 name->len, S_IFDIR, NULL);
1063 ll_get_child_fid(dir, name, &op_data.fid3);
1064 rc = mdc_unlink(ll_i2sbi(dir)->ll_mdc_exp, &op_data, &request);
1067 ll_update_times(request, REPLY_REC_OFF, dir);
1071 ptlrpc_req_finished(request);
1075 int ll_objects_destroy(struct ptlrpc_request *request, struct inode *dir)
1077 struct mds_body *body;
1078 struct lov_mds_md *eadata;
1079 struct lov_stripe_md *lsm = NULL;
1080 struct obd_trans_info oti = { 0 };
1085 /* req is swabbed so this is safe */
1086 body = lustre_msg_buf(request->rq_repmsg, REPLY_REC_OFF, sizeof(*body));
1088 if (!(body->valid & OBD_MD_FLEASIZE))
1091 if (body->eadatasize == 0) {
1092 CERROR("OBD_MD_FLEASIZE set but eadatasize zero\n");
1093 GOTO(out, rc = -EPROTO);
1096 /* The MDS sent back the EA because we unlinked the last reference
1097 * to this file. Use this EA to unlink the objects on the OST.
1098 * It's opaque so we don't swab here; we leave it to obd_unpackmd() to
1099 * check it is complete and sensible. */
1100 eadata = lustre_swab_repbuf(request, REPLY_REC_OFF + 1,
1101 body->eadatasize, NULL);
1102 LASSERT(eadata != NULL);
1103 if (eadata == NULL) {
1104 CERROR("Can't unpack MDS EA data\n");
1105 GOTO(out, rc = -EPROTO);
1108 rc = obd_unpackmd(ll_i2obdexp(dir), &lsm, eadata, body->eadatasize);
1110 CERROR("obd_unpackmd: %d\n", rc);
1113 LASSERT(rc >= sizeof(*lsm));
1115 rc = obd_checkmd(ll_i2obdexp(dir), ll_i2mdcexp(dir), lsm);
1117 GOTO(out_free_memmd, rc);
1121 GOTO(out_free_memmd, rc = -ENOMEM);
1123 oa->o_id = lsm->lsm_object_id;
1124 oa->o_gr = lsm->lsm_object_gr;
1125 oa->o_mode = body->mode & S_IFMT;
1126 oa->o_valid = OBD_MD_FLID | OBD_MD_FLGROUP | OBD_MD_FLTYPE;
1128 if (body->valid & OBD_MD_FLCOOKIE) {
1129 oa->o_valid |= OBD_MD_FLCOOKIE;
1130 oti.oti_logcookies =
1131 lustre_msg_buf(request->rq_repmsg, REPLY_REC_OFF + 2,
1132 sizeof(struct llog_cookie) *
1133 lsm->lsm_stripe_count);
1134 if (oti.oti_logcookies == NULL) {
1135 oa->o_valid &= ~OBD_MD_FLCOOKIE;
1136 body->valid &= ~OBD_MD_FLCOOKIE;
1140 rc = obd_destroy(ll_i2obdexp(dir), oa, lsm, &oti, ll_i2mdcexp(dir));
1143 CERROR("obd destroy objid "LPX64"@"LPX64" error %d\n",
1144 lsm->lsm_object_id, lsm->lsm_object_gr, rc);
1146 obd_free_memmd(ll_i2obdexp(dir), &lsm);
1151 static int ll_unlink_generic(struct inode * dir, struct qstr *name)
1153 struct ptlrpc_request *request = NULL;
1154 struct mdc_op_data op_data = { { 0 } };
1158 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
1159 name->len, name->name, dir->i_ino, dir->i_generation, dir);
1161 rc = ll_prepare_mdc_op_data(&op_data, dir, NULL, name->name,
1162 name->len, 0, NULL);
1166 ll_get_child_fid(dir, name, &op_data.fid3);
1167 rc = mdc_unlink(ll_i2sbi(dir)->ll_mdc_exp, &op_data, &request);
1171 ll_update_times(request, REPLY_REC_OFF, dir);
1173 rc = ll_objects_destroy(request, dir);
1178 ptlrpc_req_finished(request);
1182 static int ll_rename_generic(struct inode *src, struct qstr *src_name,
1183 struct inode *tgt, struct qstr *tgt_name)
1185 struct ptlrpc_request *request = NULL;
1186 struct ll_sb_info *sbi = ll_i2sbi(src);
1187 struct mdc_op_data op_data = { { 0 } };
1191 CDEBUG(D_VFSTRACE,"VFS Op:oldname=%.*s,src_dir=%lu/%u(%p),newname=%.*s,"
1192 "tgt_dir=%lu/%u(%p)\n", src_name->len, src_name->name,
1193 src->i_ino, src->i_generation, src, tgt_name->len,
1194 tgt_name->name, tgt->i_ino, tgt->i_generation, tgt);
1196 err = ll_prepare_mdc_op_data(&op_data, src, tgt, NULL, 0, 0, NULL);
1200 ll_get_child_fid(src, src_name, &op_data.fid3);
1201 ll_get_child_fid(tgt, tgt_name, &op_data.fid4);
1202 err = mdc_rename(sbi->ll_mdc_exp, &op_data,
1203 src_name->name, src_name->len,
1204 tgt_name->name, tgt_name->len, &request);
1207 ll_update_times(request, REPLY_REC_OFF, src);
1208 ll_update_times(request, REPLY_REC_OFF, tgt);
1209 err = ll_objects_destroy(request, src);
1215 ptlrpc_req_finished(request);
1220 #ifdef HAVE_VFS_INTENT_PATCHES
1221 static int ll_mknod_raw(struct nameidata *nd, int mode, dev_t rdev)
1223 return ll_mknod_generic(nd->dentry->d_inode, &nd->last, mode,rdev,NULL);
1225 static int ll_rename_raw(struct nameidata *srcnd, struct nameidata *tgtnd)
1227 return ll_rename_generic(srcnd->dentry->d_inode, &srcnd->last,
1228 tgtnd->dentry->d_inode, &tgtnd->last);
1230 static int ll_link_raw(struct nameidata *srcnd, struct nameidata *tgtnd)
1232 return ll_link_generic(srcnd->dentry->d_inode, tgtnd->dentry->d_inode,
1233 &tgtnd->last, NULL);
1235 static int ll_symlink_raw(struct nameidata *nd, const char *tgt)
1237 return ll_symlink_generic(nd->dentry->d_inode, &nd->last, tgt, NULL);
1239 static int ll_rmdir_raw(struct nameidata *nd)
1241 return ll_rmdir_generic(nd->dentry->d_inode, nd->dentry, &nd->last);
1243 static int ll_mkdir_raw(struct nameidata *nd, int mode)
1245 return ll_mkdir_generic(nd->dentry->d_inode, &nd->last, mode, NULL);
1247 static int ll_unlink_raw(struct nameidata *nd)
1249 return ll_unlink_generic(nd->dentry->d_inode, &nd->last);
1253 static int ll_mknod(struct inode *dir, struct dentry *dchild, int mode,
1256 return ll_mknod_generic(dir, &dchild->d_name, mode,
1257 old_encode_dev(rdev), dchild);
1260 static int ll_unlink(struct inode * dir, struct dentry *dentry)
1262 return ll_unlink_generic(dir, &dentry->d_name);
1264 static int ll_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1266 return ll_mkdir_generic(dir, &dentry->d_name, mode, dentry);
1268 static int ll_rmdir(struct inode *dir, struct dentry *dentry)
1270 return ll_rmdir_generic(dir, NULL, &dentry->d_name);
1272 static int ll_symlink(struct inode *dir, struct dentry *dentry,
1273 const char *oldname)
1275 return ll_symlink_generic(dir, &dentry->d_name, oldname, dentry);
1277 static int ll_link(struct dentry *old_dentry, struct inode *dir,
1278 struct dentry *new_dentry)
1280 return ll_link_generic(old_dentry->d_inode, dir,
1281 &new_dentry->d_name, new_dentry);
1283 static int ll_rename(struct inode *old_dir, struct dentry *old_dentry,
1284 struct inode *new_dir, struct dentry *new_dentry)
1287 err = ll_rename_generic(old_dir, &old_dentry->d_name, new_dir,
1288 &new_dentry->d_name);
1290 #ifndef HAVE_FS_RENAME_DOES_D_MOVE
1291 if (!S_ISDIR(old_dentry->d_inode->i_mode))
1293 d_move(old_dentry, new_dentry);
1298 struct inode_operations ll_dir_inode_operations = {
1299 #ifdef HAVE_VFS_INTENT_PATCHES
1300 .link_raw = ll_link_raw,
1301 .unlink_raw = ll_unlink_raw,
1302 .symlink_raw = ll_symlink_raw,
1303 .mkdir_raw = ll_mkdir_raw,
1304 .rmdir_raw = ll_rmdir_raw,
1305 .mknod_raw = ll_mknod_raw,
1306 .rename_raw = ll_rename_raw,
1307 .setattr = ll_setattr,
1308 .setattr_raw = ll_setattr_raw,
1311 .lookup = ll_lookup_nd,
1312 .create = ll_create_nd,
1313 /* We need all these non-raw things for NFSD, to not patch it. */
1314 .unlink = ll_unlink,
1317 .symlink = ll_symlink,
1319 .rename = ll_rename,
1320 .setattr = ll_setattr,
1321 .getattr = ll_getattr,
1322 .permission = ll_inode_permission,
1323 .setxattr = ll_setxattr,
1324 .getxattr = ll_getxattr,
1325 .listxattr = ll_listxattr,
1326 .removexattr = ll_removexattr,
1329 struct inode_operations ll_special_inode_operations = {
1330 #ifdef HAVE_VFS_INTENT_PATCHES
1331 .setattr_raw = ll_setattr_raw,
1333 .setattr = ll_setattr,
1334 .getattr = ll_getattr,
1335 .permission = ll_inode_permission,
1336 .setxattr = ll_setxattr,
1337 .getxattr = ll_getxattr,
1338 .listxattr = ll_listxattr,
1339 .removexattr = ll_removexattr,