4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2011, 2012, Whamcloud, Inc.
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>
44 #include <linux/security.h>
46 #define DEBUG_SUBSYSTEM S_LLITE
48 #include <obd_support.h>
49 #include <lustre_fid.h>
50 #include <lustre_lite.h>
51 #include <lustre_dlm.h>
52 #include <lustre_ver.h>
53 #include "llite_internal.h"
56 * Check if we have something mounted at the named dchild.
57 * In such a case there would always be dentry present.
59 static int ll_d_mountpoint(struct dentry *dparent, struct dentry *dchild,
64 if (unlikely(dchild)) {
65 mounted = d_mountpoint(dchild);
67 dchild = d_lookup(dparent, name);
69 mounted = d_mountpoint(dchild);
76 int ll_unlock(__u32 mode, struct lustre_handle *lockh)
80 ldlm_lock_decref(lockh, mode);
86 /* called from iget5_locked->find_inode() under inode_lock spinlock */
87 static int ll_test_inode(struct inode *inode, void *opaque)
89 struct ll_inode_info *lli = ll_i2info(inode);
90 struct lustre_md *md = opaque;
92 if (unlikely(!(md->body->valid & OBD_MD_FLID))) {
93 CERROR("MDS body missing FID\n");
97 if (!lu_fid_eq(&lli->lli_fid, &md->body->fid1))
103 static int ll_set_inode(struct inode *inode, void *opaque)
105 struct ll_inode_info *lli = ll_i2info(inode);
106 struct mdt_body *body = ((struct lustre_md *)opaque)->body;
108 if (unlikely(!(body->valid & OBD_MD_FLID))) {
109 CERROR("MDS body missing FID\n");
113 lli->lli_fid = body->fid1;
114 if (unlikely(!(body->valid & OBD_MD_FLTYPE))) {
115 CERROR("Can not initialize inode "DFID" without object type: "
116 "valid = "LPX64"\n", PFID(&lli->lli_fid), body->valid);
120 inode->i_mode = (inode->i_mode & ~S_IFMT) | (body->mode & S_IFMT);
121 if (unlikely(inode->i_mode == 0)) {
122 CERROR("Invalid inode "DFID" type\n", PFID(&lli->lli_fid));
133 * Get an inode by inode number (already instantiated by the intent lookup).
134 * Returns inode or NULL
136 struct inode *ll_iget(struct super_block *sb, ino_t hash,
137 struct lustre_md *md)
143 inode = iget5_locked(sb, hash, ll_test_inode, ll_set_inode, md);
146 if (inode->i_state & I_NEW) {
149 ll_read_inode2(inode, md);
150 if (S_ISREG(inode->i_mode) &&
151 ll_i2info(inode)->lli_clob == NULL)
152 rc = cl_file_inode_init(inode, md);
155 make_bad_inode(inode);
156 unlock_new_inode(inode);
160 unlock_new_inode(inode);
161 } else if (!(inode->i_state & (I_FREEING | I_CLEAR)))
162 ll_update_inode(inode, md);
163 CDEBUG(D_VFSTRACE, "got inode: %p for "DFID"\n",
164 inode, PFID(&md->body->fid1));
169 static void ll_invalidate_negative_children(struct inode *dir)
171 struct dentry *dentry, *tmp_subdir;
174 list_for_each_entry(dentry, &dir->i_dentry, d_alias) {
175 spin_lock(&dentry->d_lock);
176 if (!list_empty(&dentry->d_subdirs)) {
177 struct dentry *child;
179 list_for_each_entry_safe(child, tmp_subdir,
182 if (child->d_inode == NULL)
183 d_lustre_invalidate(child);
186 spin_unlock(&dentry->d_lock);
188 ll_unlock_dcache(dir);
191 int ll_md_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
192 void *data, int flag)
195 struct lustre_handle lockh;
199 case LDLM_CB_BLOCKING:
200 ldlm_lock2handle(lock, &lockh);
201 rc = ldlm_cli_cancel(&lockh);
203 CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
207 case LDLM_CB_CANCELING: {
208 struct inode *inode = ll_inode_from_lock(lock);
209 struct ll_inode_info *lli;
210 __u64 bits = lock->l_policy_data.l_inodebits.bits;
212 ldlm_mode_t mode = lock->l_req_mode;
214 /* Invalidate all dentries associated with this inode */
218 LASSERT(lock->l_flags & LDLM_FL_CANCELING);
219 /* For OPEN locks we differentiate between lock modes
220 * LCK_CR, LCK_CW, LCK_PR - bug 22891 */
221 if (bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
222 MDS_INODELOCK_LAYOUT))
223 ll_have_md_lock(inode, &bits, LCK_MINMODE);
225 if (bits & MDS_INODELOCK_OPEN)
226 ll_have_md_lock(inode, &bits, mode);
228 fid = ll_inode2fid(inode);
229 if (lock->l_resource->lr_name.name[0] != fid_seq(fid) ||
230 lock->l_resource->lr_name.name[1] != fid_oid(fid) ||
231 lock->l_resource->lr_name.name[2] != fid_ver(fid)) {
232 LDLM_ERROR(lock, "data mismatch with object "
233 DFID" (%p)", PFID(fid), inode);
236 if (bits & MDS_INODELOCK_OPEN) {
238 switch (lock->l_req_mode) {
249 CERROR("Unexpected lock mode for OPEN lock "
250 "%d, inode %ld\n", lock->l_req_mode,
253 ll_md_real_close(inode, flags);
256 lli = ll_i2info(inode);
257 if (bits & MDS_INODELOCK_LAYOUT) {
258 struct cl_object_conf conf = { .coc_inode = inode,
259 .coc_invalidate = true };
260 rc = ll_layout_conf(inode, &conf);
262 CDEBUG(D_INODE, "invaliding layout %d.\n", rc);
265 if (bits & MDS_INODELOCK_UPDATE)
266 lli->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
268 if (S_ISDIR(inode->i_mode) &&
269 (bits & MDS_INODELOCK_UPDATE)) {
270 CDEBUG(D_INODE, "invalidating inode %lu\n",
272 truncate_inode_pages(inode->i_mapping, 0);
273 ll_invalidate_negative_children(inode);
276 if (inode->i_sb->s_root &&
277 inode != inode->i_sb->s_root->d_inode &&
278 (bits & MDS_INODELOCK_LOOKUP))
279 ll_invalidate_aliases(inode);
290 __u32 ll_i2suppgid(struct inode *i)
292 if (cfs_curproc_is_in_groups(i->i_gid))
293 return (__u32)i->i_gid;
298 /* Pack the required supplementary groups into the supplied groups array.
299 * If we don't need to use the groups from the target inode(s) then we
300 * instead pack one or more groups from the user's supplementary group
301 * array in case it might be useful. Not needed if doing an MDS-side upcall. */
302 void ll_i2gids(__u32 *suppgids, struct inode *i1, struct inode *i2)
309 LASSERT(suppgids != NULL);
311 suppgids[0] = ll_i2suppgid(i1);
314 suppgids[1] = ll_i2suppgid(i2);
319 for (i = 0; i < current_ngroups; i++) {
320 if (suppgids[0] == -1) {
321 if (current_groups[i] != suppgids[1])
322 suppgids[0] = current_groups[i];
325 if (suppgids[1] == -1) {
326 if (current_groups[i] != suppgids[0])
327 suppgids[1] = current_groups[i];
336 * try to reuse three types of dentry:
337 * 1. unhashed alias, this one is unhashed by d_invalidate (but it may be valid
338 * by concurrent .revalidate).
339 * 2. INVALID alias (common case for no valid ldlm lock held, but this flag may
340 * be cleared by others calling d_lustre_revalidate).
341 * 3. DISCONNECTED alias.
343 static struct dentry *ll_find_alias(struct inode *inode, struct dentry *dentry)
345 struct dentry *alias, *discon_alias, *invalid_alias;
347 if (list_empty(&inode->i_dentry))
350 discon_alias = invalid_alias = NULL;
352 ll_lock_dcache(inode);
353 list_for_each_entry(alias, &inode->i_dentry, d_alias) {
354 LASSERT(alias != dentry);
356 spin_lock(&alias->d_lock);
357 if (alias->d_flags & DCACHE_DISCONNECTED)
358 /* LASSERT(last_discon == NULL); LU-405, bz 20055 */
359 discon_alias = alias;
360 else if (alias->d_parent == dentry->d_parent &&
361 alias->d_name.hash == dentry->d_name.hash &&
362 alias->d_name.len == dentry->d_name.len &&
363 memcmp(alias->d_name.name, dentry->d_name.name,
364 dentry->d_name.len) == 0)
365 invalid_alias = alias;
366 spin_unlock(&alias->d_lock);
371 alias = invalid_alias ?: discon_alias ?: NULL;
373 spin_lock(&alias->d_lock);
375 spin_unlock(&alias->d_lock);
377 ll_unlock_dcache(inode);
383 * Similar to d_splice_alias(), but lustre treats DCACHE_LUSTRE_INVALID alias
384 * similar to DCACHE_DISCONNECTED, and tries to use it anyway.
386 struct dentry *ll_splice_alias(struct inode *inode, struct dentry *de)
391 new = ll_find_alias(inode, de);
393 ll_dops_init(new, 1, 1);
397 "Reuse dentry %p inode %p refc %d flags %#x\n",
398 new, new->d_inode, d_refcount(new), new->d_flags);
402 __d_lustre_invalidate(de);
403 ll_dops_init(de, 1, 1);
405 CDEBUG(D_DENTRY, "Add dentry %p inode %p refc %d flags %#x\n",
406 de, de->d_inode, d_refcount(de), de->d_flags);
410 int ll_lookup_it_finish(struct ptlrpc_request *request,
411 struct lookup_intent *it, void *data)
413 struct it_cb_data *icbd = data;
414 struct dentry **de = icbd->icbd_childp;
415 struct inode *parent = icbd->icbd_parent;
416 struct inode *inode = NULL;
421 /* NB 1 request reference will be taken away by ll_intent_lock()
423 if (!it_disposition(it, DISP_LOOKUP_NEG)) {
424 rc = ll_prep_inode(&inode, request, (*de)->d_sb);
428 ll_set_lock_data(ll_i2sbi(parent)->ll_md_exp, inode, it, &bits);
430 /* We used to query real size from OSTs here, but actually
431 this is not needed. For stat() calls size would be updated
432 from subsequent do_revalidate()->ll_inode_revalidate_it() in
434 vfs_getattr_it->ll_getattr()->ll_inode_revalidate_it() in 2.6
435 Everybody else who needs correct file size would call
436 ll_glimpse_size or some equivalent themselves anyway.
437 Also see bug 7198. */
440 *de = ll_splice_alias(inode, *de);
442 if (!it_disposition(it, DISP_LOOKUP_NEG)) {
443 /* we have lookup look - unhide dentry */
444 if (bits & MDS_INODELOCK_LOOKUP)
445 d_lustre_revalidate(*de);
447 /* Check that parent has UPDATE lock. */
448 struct lookup_intent parent_it = {
450 .d.lustre.it_lock_handle = 0 };
452 if (md_revalidate_lock(ll_i2mdexp(parent), &parent_it,
453 &ll_i2info(parent)->lli_fid, NULL)) {
454 d_lustre_revalidate(*de);
455 ll_intent_release(&parent_it);
462 static struct dentry *ll_lookup_it(struct inode *parent, struct dentry *dentry,
463 struct lookup_intent *it, int lookup_flags)
465 struct lookup_intent lookup_it = { .it_op = IT_LOOKUP };
466 struct dentry *save = dentry, *retval;
467 struct ptlrpc_request *req = NULL;
468 struct md_op_data *op_data;
469 struct it_cb_data icbd;
474 if (dentry->d_name.len > ll_i2sbi(parent)->ll_namelen)
475 RETURN(ERR_PTR(-ENAMETOOLONG));
477 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),intent=%s\n",
478 dentry->d_name.len, dentry->d_name.name, parent->i_ino,
479 parent->i_generation, parent, LL_IT2STR(it));
481 if (d_mountpoint(dentry))
482 CERROR("Tell Peter, lookup on mtpt, it %s\n", LL_IT2STR(it));
484 ll_frob_intent(&it, &lookup_it);
486 /* As do_lookup is called before follow_mount, root dentry may be left
487 * not valid, revalidate it here. */
488 if (parent->i_sb->s_root && (parent->i_sb->s_root->d_inode == parent) &&
489 (it->it_op & (IT_OPEN | IT_CREAT))) {
490 rc = ll_inode_revalidate_it(parent->i_sb->s_root, it,
491 MDS_INODELOCK_LOOKUP);
496 if (it->it_op == IT_GETATTR) {
497 rc = ll_statahead_enter(parent, &dentry, 0);
500 GOTO(out, retval = NULL);
501 GOTO(out, retval = dentry);
505 icbd.icbd_childp = &dentry;
506 icbd.icbd_parent = parent;
508 if (it->it_op & IT_CREAT ||
509 (it->it_op & IT_OPEN && it->it_create_mode & O_CREAT))
510 opc = LUSTRE_OPC_CREATE;
512 opc = LUSTRE_OPC_ANY;
514 op_data = ll_prep_md_op_data(NULL, parent, NULL, dentry->d_name.name,
515 dentry->d_name.len, lookup_flags, opc,
518 RETURN((void *)op_data);
520 it->it_create_mode &= ~cfs_curproc_umask();
522 rc = md_intent_lock(ll_i2mdexp(parent), op_data, NULL, 0, it,
523 lookup_flags, &req, ll_md_blocking_ast, 0);
524 ll_finish_md_op_data(op_data);
526 GOTO(out, retval = ERR_PTR(rc));
528 rc = ll_lookup_it_finish(req, it, &icbd);
530 ll_intent_release(it);
531 GOTO(out, retval = ERR_PTR(rc));
534 if ((it->it_op & IT_OPEN) && dentry->d_inode &&
535 !S_ISREG(dentry->d_inode->i_mode) &&
536 !S_ISDIR(dentry->d_inode->i_mode)) {
537 ll_release_openhandle(dentry, it);
539 ll_lookup_finish_locks(it, dentry);
542 GOTO(out, retval = NULL);
544 GOTO(out, retval = dentry);
547 ptlrpc_req_finished(req);
548 if (it->it_op == IT_GETATTR && (retval == NULL || retval == dentry))
549 ll_statahead_mark(parent, dentry);
553 struct lookup_intent *ll_convert_intent(struct open_intent *oit,
556 struct lookup_intent *it;
558 OBD_ALLOC(it, sizeof(*it));
560 return ERR_PTR(-ENOMEM);
562 if (lookup_flags & LOOKUP_OPEN) {
564 if (lookup_flags & LOOKUP_CREATE)
565 it->it_op |= IT_CREAT;
566 it->it_create_mode = (oit->create_mode & S_IALLUGO) | S_IFREG;
567 it->it_flags = oit->flags;
569 it->it_op = IT_GETATTR;
575 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
576 struct nameidata *nd)
581 if (nd && !(nd->flags & (LOOKUP_CONTINUE|LOOKUP_PARENT))) {
582 struct lookup_intent *it;
584 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,17)
585 /* Did we came here from failed revalidate just to propagate
587 if (nd->flags & LOOKUP_OPEN)
588 if (IS_ERR(nd->intent.open.file))
589 RETURN((struct dentry *)nd->intent.open.file);
591 if (ll_d2d(dentry) && ll_d2d(dentry)->lld_it) {
592 it = ll_d2d(dentry)->lld_it;
593 ll_d2d(dentry)->lld_it = NULL;
595 if ((nd->flags & LOOKUP_CREATE ) && !(nd->flags & LOOKUP_OPEN)) {
596 /* We are sure this is new dentry, so we need to create
597 our private data and set the dentry ops */
598 ll_dops_init(dentry, 1, 1);
599 __d_lustre_invalidate(dentry);
603 it = ll_convert_intent(&nd->intent.open, nd->flags);
605 RETURN((struct dentry *)it);
608 de = ll_lookup_it(parent, dentry, it, nd->flags);
611 if ((nd->flags & LOOKUP_OPEN) && !IS_ERR(dentry)) { /* Open */
612 if (dentry->d_inode &&
613 it_disposition(it, DISP_OPEN_OPEN)) { /* nocreate */
614 if (S_ISFIFO(dentry->d_inode->i_mode)) {
615 // We cannot call open here as it would
618 (struct ptlrpc_request *)
619 it->d.lustre.it_data);
623 nd->intent.open.file->private_data = it;
624 filp = lookup_instantiate_filp(nd,
630 de = (struct dentry *)filp;
633 } else if (it_disposition(it, DISP_OPEN_CREATE)) {
634 // XXX This can only reliably work on assumption
635 // that there are NO hashed negative dentries.
636 ll_d2d(dentry)->lld_it = it;
637 it = NULL; /* Will be freed in ll_create_nd */
638 /* We absolutely depend on ll_create_nd to be
639 * called to not leak this intent and possible
640 * data attached to it */
645 ll_intent_release(it);
646 OBD_FREE(it, sizeof(*it));
649 de = ll_lookup_it(parent, dentry, NULL, 0);
655 /* We depend on "mode" being set with the proper file type/umask by now */
656 static struct inode *ll_create_node(struct inode *dir, const char *name,
657 int namelen, const void *data, int datalen,
658 int mode, __u64 extra,
659 struct lookup_intent *it)
661 struct inode *inode = NULL;
662 struct ptlrpc_request *request = NULL;
663 struct ll_sb_info *sbi = ll_i2sbi(dir);
667 LASSERT(it && it->d.lustre.it_disposition);
669 LASSERT(it_disposition(it, DISP_ENQ_CREATE_REF));
670 request = it->d.lustre.it_data;
671 it_clear_disposition(it, DISP_ENQ_CREATE_REF);
672 rc = ll_prep_inode(&inode, request, dir->i_sb);
674 GOTO(out, inode = ERR_PTR(rc));
676 LASSERT(list_empty(&inode->i_dentry));
678 /* We asked for a lock on the directory, but were granted a
679 * lock on the inode. Since we finally have an inode pointer,
680 * stuff it in the lock. */
681 CDEBUG(D_DLMTRACE, "setting l_ast_data to inode %p (%lu/%u)\n",
682 inode, inode->i_ino, inode->i_generation);
683 ll_set_lock_data(sbi->ll_md_exp, inode, it, NULL);
686 ptlrpc_req_finished(request);
691 * By the time this is called, we already have created the directory cache
692 * entry for the new file, but it is so far negative - it has no inode.
694 * We defer creating the OBD object(s) until open, to keep the intent and
695 * non-intent code paths similar, and also because we do not have the MDS
696 * inode number before calling ll_create_node() (which is needed for LOV),
697 * so we would need to do yet another RPC to the MDS to store the LOV EA
698 * data on the MDS. If needed, we would pass the PACKED lmm as data and
699 * lmm_size in datalen (the MDS still has code which will handle that).
701 * If the create succeeds, we fill in the inode information
702 * with d_instantiate().
704 static int ll_create_it(struct inode *dir, struct dentry *dentry, int mode,
705 struct lookup_intent *it)
711 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),intent=%s\n",
712 dentry->d_name.len, dentry->d_name.name, dir->i_ino,
713 dir->i_generation, dir, LL_IT2STR(it));
715 rc = it_open_error(DISP_OPEN_CREATE, it);
719 inode = ll_create_node(dir, dentry->d_name.name, dentry->d_name.len,
720 NULL, 0, mode, 0, it);
722 RETURN(PTR_ERR(inode));
724 d_instantiate(dentry, inode);
728 static void ll_update_times(struct ptlrpc_request *request,
731 struct mdt_body *body = req_capsule_server_get(&request->rq_pill,
735 if (body->valid & OBD_MD_FLMTIME &&
736 body->mtime > LTIME_S(inode->i_mtime)) {
737 CDEBUG(D_INODE, "setting ino %lu mtime from %lu to "LPU64"\n",
738 inode->i_ino, LTIME_S(inode->i_mtime), body->mtime);
739 LTIME_S(inode->i_mtime) = body->mtime;
741 if (body->valid & OBD_MD_FLCTIME &&
742 body->ctime > LTIME_S(inode->i_ctime))
743 LTIME_S(inode->i_ctime) = body->ctime;
746 static int ll_new_node(struct inode *dir, struct qstr *name,
747 const char *tgt, int mode, int rdev,
748 struct dentry *dchild, __u32 opc)
750 struct ptlrpc_request *request = NULL;
751 struct md_op_data *op_data;
752 struct inode *inode = NULL;
753 struct ll_sb_info *sbi = ll_i2sbi(dir);
758 if (unlikely(tgt != NULL))
759 tgt_len = strlen(tgt) + 1;
761 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
762 name->len, 0, opc, NULL);
764 GOTO(err_exit, err = PTR_ERR(op_data));
766 err = md_create(sbi->ll_md_exp, op_data, tgt, tgt_len, mode,
767 cfs_curproc_fsuid(), cfs_curproc_fsgid(),
768 cfs_curproc_cap_pack(), rdev, &request);
769 ll_finish_md_op_data(op_data);
773 ll_update_times(request, dir);
776 err = ll_prep_inode(&inode, request, dchild->d_sb);
780 d_instantiate(dchild, inode);
784 ptlrpc_req_finished(request);
789 static int ll_mknod_generic(struct inode *dir, struct qstr *name, int mode,
790 unsigned rdev, struct dentry *dchild)
795 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p) mode %o dev %x\n",
796 name->len, name->name, dir->i_ino, dir->i_generation, dir,
799 mode &= ~cfs_curproc_umask();
801 switch (mode & S_IFMT) {
803 mode |= S_IFREG; /* for mode = 0 case, fallthrough */
809 err = ll_new_node(dir, name, NULL, mode, rdev, dchild,
820 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKNOD, 1);
825 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
826 int mode, struct nameidata *nd)
828 struct lookup_intent *it = ll_d2d(dentry)->lld_it;
832 return ll_mknod_generic(dir, &dentry->d_name, mode, 0, dentry);
834 ll_d2d(dentry)->lld_it = NULL;
836 /* Was there an error? Propagate it! */
837 if (it->d.lustre.it_status) {
838 rc = it->d.lustre.it_status;
842 rc = ll_create_it(dir, dentry, mode, it);
843 if (nd && (nd->flags & LOOKUP_OPEN) && dentry->d_inode) { /* Open */
846 nd->intent.open.file->private_data = it;
847 filp = lookup_instantiate_filp(nd, dentry, NULL);
853 ll_intent_release(it);
854 OBD_FREE(it, sizeof(*it));
857 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_CREATE, 1);
862 static int ll_symlink_generic(struct inode *dir, struct qstr *name,
863 const char *tgt, struct dentry *dchild)
868 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p),target=%.*s\n",
869 name->len, name->name, dir->i_ino, dir->i_generation,
872 err = ll_new_node(dir, name, (char *)tgt, S_IFLNK | S_IRWXUGO,
873 0, dchild, LUSTRE_OPC_SYMLINK);
876 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_SYMLINK, 1);
881 static int ll_link_generic(struct inode *src, struct inode *dir,
882 struct qstr *name, struct dentry *dchild)
884 struct ll_sb_info *sbi = ll_i2sbi(dir);
885 struct ptlrpc_request *request = NULL;
886 struct md_op_data *op_data;
891 "VFS Op: inode=%lu/%u(%p), dir=%lu/%u(%p), target=%.*s\n",
892 src->i_ino, src->i_generation, src, dir->i_ino,
893 dir->i_generation, dir, name->len, name->name);
895 op_data = ll_prep_md_op_data(NULL, src, dir, name->name, name->len,
896 0, LUSTRE_OPC_ANY, NULL);
898 RETURN(PTR_ERR(op_data));
900 err = md_link(sbi->ll_md_exp, op_data, &request);
901 ll_finish_md_op_data(op_data);
905 ll_update_times(request, dir);
906 ll_stats_ops_tally(sbi, LPROC_LL_LINK, 1);
909 ptlrpc_req_finished(request);
913 static int ll_mkdir_generic(struct inode *dir, struct qstr *name,
914 int mode, struct dentry *dchild)
920 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
921 name->len, name->name, dir->i_ino, dir->i_generation, dir);
923 mode = (mode & (S_IRWXUGO|S_ISVTX) & ~cfs_curproc_umask()) | S_IFDIR;
924 err = ll_new_node(dir, name, NULL, mode, 0, dchild, LUSTRE_OPC_MKDIR);
927 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKDIR, 1);
932 /* Try to find the child dentry by its name.
933 If found, put the result fid into @fid. */
934 static void ll_get_child_fid(struct inode * dir, struct qstr *name,
937 struct dentry *parent, *child;
939 parent = list_entry(dir->i_dentry.next, struct dentry, d_alias);
940 child = d_lookup(parent, name);
943 *fid = *ll_inode2fid(child->d_inode);
948 static int ll_rmdir_generic(struct inode *dir, struct dentry *dparent,
949 struct dentry *dchild, struct qstr *name)
951 struct ptlrpc_request *request = NULL;
952 struct md_op_data *op_data;
956 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
957 name->len, name->name, dir->i_ino, dir->i_generation, dir);
959 if (unlikely(ll_d_mountpoint(dparent, dchild, name)))
962 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len,
963 S_IFDIR, LUSTRE_OPC_ANY, NULL);
965 RETURN(PTR_ERR(op_data));
967 ll_get_child_fid(dir, name, &op_data->op_fid3);
968 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
969 ll_finish_md_op_data(op_data);
971 ll_update_times(request, dir);
972 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR, 1);
975 ptlrpc_req_finished(request);
979 int ll_objects_destroy(struct ptlrpc_request *request, struct inode *dir)
981 struct mdt_body *body;
982 struct lov_mds_md *eadata;
983 struct lov_stripe_md *lsm = NULL;
984 struct obd_trans_info oti = { 0 };
986 struct obd_capa *oc = NULL;
990 /* req is swabbed so this is safe */
991 body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
992 if (!(body->valid & OBD_MD_FLEASIZE))
995 if (body->eadatasize == 0) {
996 CERROR("OBD_MD_FLEASIZE set but eadatasize zero\n");
997 GOTO(out, rc = -EPROTO);
1000 /* The MDS sent back the EA because we unlinked the last reference
1001 * to this file. Use this EA to unlink the objects on the OST.
1002 * It's opaque so we don't swab here; we leave it to obd_unpackmd() to
1003 * check it is complete and sensible. */
1004 eadata = req_capsule_server_sized_get(&request->rq_pill, &RMF_MDT_MD,
1006 LASSERT(eadata != NULL);
1008 rc = obd_unpackmd(ll_i2dtexp(dir), &lsm, eadata, body->eadatasize);
1010 CERROR("obd_unpackmd: %d\n", rc);
1013 LASSERT(rc >= sizeof(*lsm));
1017 GOTO(out_free_memmd, rc = -ENOMEM);
1019 oa->o_id = lsm->lsm_object_id;
1020 oa->o_seq = lsm->lsm_object_seq;
1021 oa->o_mode = body->mode & S_IFMT;
1022 oa->o_valid = OBD_MD_FLID | OBD_MD_FLTYPE | OBD_MD_FLGROUP;
1024 if (body->valid & OBD_MD_FLCOOKIE) {
1025 oa->o_valid |= OBD_MD_FLCOOKIE;
1026 oti.oti_logcookies =
1027 req_capsule_server_sized_get(&request->rq_pill,
1029 sizeof(struct llog_cookie) *
1030 lsm->lsm_stripe_count);
1031 if (oti.oti_logcookies == NULL) {
1032 oa->o_valid &= ~OBD_MD_FLCOOKIE;
1033 body->valid &= ~OBD_MD_FLCOOKIE;
1037 if (body->valid & OBD_MD_FLOSSCAPA) {
1038 rc = md_unpack_capa(ll_i2mdexp(dir), request, &RMF_CAPA2, &oc);
1040 GOTO(out_free_memmd, rc);
1043 rc = obd_destroy(NULL, ll_i2dtexp(dir), oa, lsm, &oti,
1044 ll_i2mdexp(dir), oc);
1048 CERROR("obd destroy objid "LPX64" error %d\n",
1049 lsm->lsm_object_id, rc);
1051 obd_free_memmd(ll_i2dtexp(dir), &lsm);
1056 /* ll_unlink_generic() doesn't update the inode with the new link count.
1057 * Instead, ll_ddelete() and ll_d_iput() will update it based upon if there
1058 * is any lock existing. They will recycle dentries and inodes based upon locks
1060 static int ll_unlink_generic(struct inode *dir, struct dentry *dparent,
1061 struct dentry *dchild, struct qstr *name)
1063 struct ptlrpc_request *request = NULL;
1064 struct md_op_data *op_data;
1067 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s,dir=%lu/%u(%p)\n",
1068 name->len, name->name, dir->i_ino, dir->i_generation, dir);
1071 * XXX: unlink bind mountpoint maybe call to here,
1072 * just check it as vfs_unlink does.
1074 if (unlikely(ll_d_mountpoint(dparent, dchild, name)))
1077 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
1078 name->len, 0, LUSTRE_OPC_ANY, NULL);
1079 if (IS_ERR(op_data))
1080 RETURN(PTR_ERR(op_data));
1082 ll_get_child_fid(dir, name, &op_data->op_fid3);
1083 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1084 ll_finish_md_op_data(op_data);
1088 ll_update_times(request, dir);
1089 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_UNLINK, 1);
1091 rc = ll_objects_destroy(request, dir);
1093 ptlrpc_req_finished(request);
1097 static int ll_rename_generic(struct inode *src, struct dentry *src_dparent,
1098 struct dentry *src_dchild, struct qstr *src_name,
1099 struct inode *tgt, struct dentry *tgt_dparent,
1100 struct dentry *tgt_dchild, struct qstr *tgt_name)
1102 struct ptlrpc_request *request = NULL;
1103 struct ll_sb_info *sbi = ll_i2sbi(src);
1104 struct md_op_data *op_data;
1107 CDEBUG(D_VFSTRACE,"VFS Op:oldname=%.*s,src_dir=%lu/%u(%p),newname=%.*s,"
1108 "tgt_dir=%lu/%u(%p)\n", src_name->len, src_name->name,
1109 src->i_ino, src->i_generation, src, tgt_name->len,
1110 tgt_name->name, tgt->i_ino, tgt->i_generation, tgt);
1112 if (unlikely(ll_d_mountpoint(src_dparent, src_dchild, src_name) ||
1113 ll_d_mountpoint(tgt_dparent, tgt_dchild, tgt_name)))
1116 op_data = ll_prep_md_op_data(NULL, src, tgt, NULL, 0, 0,
1117 LUSTRE_OPC_ANY, NULL);
1118 if (IS_ERR(op_data))
1119 RETURN(PTR_ERR(op_data));
1121 ll_get_child_fid(src, src_name, &op_data->op_fid3);
1122 ll_get_child_fid(tgt, tgt_name, &op_data->op_fid4);
1123 err = md_rename(sbi->ll_md_exp, op_data,
1124 src_name->name, src_name->len,
1125 tgt_name->name, tgt_name->len, &request);
1126 ll_finish_md_op_data(op_data);
1128 ll_update_times(request, src);
1129 ll_update_times(request, tgt);
1130 ll_stats_ops_tally(sbi, LPROC_LL_RENAME, 1);
1131 err = ll_objects_destroy(request, src);
1134 ptlrpc_req_finished(request);
1139 static int ll_mknod(struct inode *dir, struct dentry *dchild, int mode,
1142 return ll_mknod_generic(dir, &dchild->d_name, mode,
1143 old_encode_dev(rdev), dchild);
1146 static int ll_unlink(struct inode * dir, struct dentry *dentry)
1148 return ll_unlink_generic(dir, NULL, dentry, &dentry->d_name);
1150 static int ll_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1152 return ll_mkdir_generic(dir, &dentry->d_name, mode, dentry);
1154 static int ll_rmdir(struct inode *dir, struct dentry *dentry)
1156 return ll_rmdir_generic(dir, NULL, dentry, &dentry->d_name);
1158 static int ll_symlink(struct inode *dir, struct dentry *dentry,
1159 const char *oldname)
1161 return ll_symlink_generic(dir, &dentry->d_name, oldname, dentry);
1163 static int ll_link(struct dentry *old_dentry, struct inode *dir,
1164 struct dentry *new_dentry)
1166 return ll_link_generic(old_dentry->d_inode, dir, &new_dentry->d_name,
1169 static int ll_rename(struct inode *old_dir, struct dentry *old_dentry,
1170 struct inode *new_dir, struct dentry *new_dentry)
1173 err = ll_rename_generic(old_dir, NULL,
1174 old_dentry, &old_dentry->d_name,
1175 new_dir, NULL, new_dentry,
1176 &new_dentry->d_name);
1178 #ifndef HAVE_FS_RENAME_DOES_D_MOVE
1179 if (!S_ISDIR(old_dentry->d_inode->i_mode))
1181 d_move(old_dentry, new_dentry);
1186 struct inode_operations ll_dir_inode_operations = {
1188 .lookup = ll_lookup_nd,
1189 .create = ll_create_nd,
1190 /* We need all these non-raw things for NFSD, to not patch it. */
1191 .unlink = ll_unlink,
1194 .symlink = ll_symlink,
1196 .rename = ll_rename,
1197 .setattr = ll_setattr,
1198 .getattr = ll_getattr,
1199 .permission = ll_inode_permission,
1200 .setxattr = ll_setxattr,
1201 .getxattr = ll_getxattr,
1202 .listxattr = ll_listxattr,
1203 .removexattr = ll_removexattr,
1206 struct inode_operations ll_special_inode_operations = {
1207 .setattr = ll_setattr,
1208 .getattr = ll_getattr,
1209 .permission = ll_inode_permission,
1210 .setxattr = ll_setxattr,
1211 .getxattr = ll_getxattr,
1212 .listxattr = ll_listxattr,
1213 .removexattr = ll_removexattr,