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.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2011, 2017, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
34 #include <linux/sched.h>
36 #include <linux/quotaops.h>
37 #include <linux/highmem.h>
38 #include <linux/pagemap.h>
39 #include <linux/security.h>
40 #include <linux/user_namespace.h>
41 #ifdef HAVE_UIDGID_HEADER
42 # include <linux/uidgid.h>
45 #define DEBUG_SUBSYSTEM S_LLITE
47 #include <obd_support.h>
48 #include <lustre_fid.h>
49 #include <lustre_dlm.h>
50 #include "llite_internal.h"
52 static int ll_create_it(struct inode *dir, struct dentry *dentry,
53 struct lookup_intent *it,
54 void *secctx, __u32 secctxlen);
56 /* called from iget5_locked->find_inode() under inode_lock spinlock */
57 static int ll_test_inode(struct inode *inode, void *opaque)
59 struct ll_inode_info *lli = ll_i2info(inode);
60 struct lustre_md *md = opaque;
62 if (unlikely(!(md->body->mbo_valid & OBD_MD_FLID))) {
63 CERROR("MDS body missing FID\n");
67 if (!lu_fid_eq(&lli->lli_fid, &md->body->mbo_fid1))
73 static int ll_set_inode(struct inode *inode, void *opaque)
75 struct ll_inode_info *lli = ll_i2info(inode);
76 struct mdt_body *body = ((struct lustre_md *)opaque)->body;
78 if (unlikely(!(body->mbo_valid & OBD_MD_FLID))) {
79 CERROR("MDS body missing FID\n");
83 lli->lli_fid = body->mbo_fid1;
84 if (unlikely(!(body->mbo_valid & OBD_MD_FLTYPE))) {
85 CERROR("Can not initialize inode "DFID" without object type: "
87 PFID(&lli->lli_fid), body->mbo_valid);
91 inode->i_mode = (inode->i_mode & ~S_IFMT) | (body->mbo_mode & S_IFMT);
92 if (unlikely(inode->i_mode == 0)) {
93 CERROR("Invalid inode "DFID" type\n", PFID(&lli->lli_fid));
104 * Get an inode by inode number(@hash), which is already instantiated by
105 * the intent lookup).
107 struct inode *ll_iget(struct super_block *sb, ino_t hash,
108 struct lustre_md *md)
116 inode = iget5_locked(sb, hash, ll_test_inode, ll_set_inode, md);
118 RETURN(ERR_PTR(-ENOMEM));
120 if (inode->i_state & I_NEW) {
121 rc = ll_read_inode2(inode, md);
122 if (rc == 0 && S_ISREG(inode->i_mode) &&
123 ll_i2info(inode)->lli_clob == NULL)
124 rc = cl_file_inode_init(inode, md);
127 /* Let's clear directory lsm here, otherwise
128 * make_bad_inode() will reset the inode mode
129 * to regular, then ll_clear_inode will not
130 * be able to clear lsm_md */
131 if (S_ISDIR(inode->i_mode))
132 ll_dir_clear_lsm_md(inode);
133 make_bad_inode(inode);
134 unlock_new_inode(inode);
138 inode_has_no_xattr(inode);
139 unlock_new_inode(inode);
141 } else if (is_bad_inode(inode)) {
143 inode = ERR_PTR(-ESTALE);
144 } else if (!(inode->i_state & (I_FREEING | I_CLEAR))) {
145 rc = ll_update_inode(inode, md);
146 CDEBUG(D_VFSTRACE, "got inode: "DFID"(%p): rc = %d\n",
147 PFID(&md->body->mbo_fid1), inode, rc);
149 if (S_ISDIR(inode->i_mode))
150 ll_dir_clear_lsm_md(inode);
159 static void ll_invalidate_negative_children(struct inode *dir)
161 struct dentry *dentry, *tmp_subdir;
162 DECLARE_LL_D_HLIST_NODE_PTR(p);
165 ll_d_hlist_for_each_entry(dentry, p, &dir->i_dentry) {
166 spin_lock(&dentry->d_lock);
167 if (!list_empty(&dentry->d_subdirs)) {
168 struct dentry *child;
170 list_for_each_entry_safe(child, tmp_subdir,
173 if (child->d_inode == NULL)
174 d_lustre_invalidate(child, 1);
177 spin_unlock(&dentry->d_lock);
179 ll_unlock_dcache(dir);
182 int ll_test_inode_by_fid(struct inode *inode, void *opaque)
184 return lu_fid_eq(&ll_i2info(inode)->lli_fid, opaque);
187 int ll_dom_lock_cancel(struct inode *inode, struct ldlm_lock *lock)
190 struct ll_inode_info *lli = ll_i2info(inode);
191 struct cl_layout clt = { .cl_layout_gen = 0, };
198 if (!lli->lli_clob) {
199 /* due to DoM read on open, there may exist pages for Lustre
200 * regular file even though cl_object is not set up yet. */
201 truncate_inode_pages(inode->i_mapping, 0);
205 env = cl_env_get(&refcheck);
207 RETURN(PTR_ERR(env));
209 rc = cl_object_layout_get(env, lli->lli_clob, &clt);
211 CDEBUG(D_INODE, "Cannot get layout for "DFID"\n",
212 PFID(ll_inode2fid(inode)));
214 } else if (clt.cl_size == 0 || clt.cl_dom_comp_size == 0) {
215 CDEBUG(D_INODE, "DOM lock without DOM layout for "DFID"\n",
216 PFID(ll_inode2fid(inode)));
218 enum cl_fsync_mode mode;
219 loff_t end = clt.cl_dom_comp_size - 1;
221 mode = ldlm_is_discard_data(lock) ?
222 CL_FSYNC_DISCARD : CL_FSYNC_LOCAL;
223 rc = cl_sync_file_range(inode, 0, end, mode, 1);
224 truncate_inode_pages_range(inode->i_mapping, 0, end);
226 cl_env_put(env, &refcheck);
230 void ll_lock_cancel_bits(struct ldlm_lock *lock, __u64 to_cancel)
232 struct inode *inode = ll_inode_from_resource_lock(lock);
233 __u64 bits = to_cancel;
239 if (!fid_res_name_eq(ll_inode2fid(inode),
240 &lock->l_resource->lr_name)) {
241 LDLM_ERROR(lock, "data mismatch with object "DFID"(%p)",
242 PFID(ll_inode2fid(inode)), inode);
246 if (bits & MDS_INODELOCK_XATTR) {
247 if (S_ISDIR(inode->i_mode))
248 ll_i2info(inode)->lli_def_stripe_offset = -1;
249 ll_xattr_cache_destroy(inode);
250 bits &= ~MDS_INODELOCK_XATTR;
253 /* For OPEN locks we differentiate between lock modes
254 * LCK_CR, LCK_CW, LCK_PR - bug 22891 */
255 if (bits & MDS_INODELOCK_OPEN)
256 ll_have_md_lock(inode, &bits, lock->l_req_mode);
258 if (bits & MDS_INODELOCK_OPEN) {
261 switch (lock->l_req_mode) {
272 LDLM_ERROR(lock, "bad lock mode for OPEN lock");
276 ll_md_real_close(inode, fmode);
278 bits &= ~MDS_INODELOCK_OPEN;
281 if (bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
282 MDS_INODELOCK_LAYOUT | MDS_INODELOCK_PERM |
284 ll_have_md_lock(inode, &bits, LCK_MINMODE);
286 if (bits & MDS_INODELOCK_DOM) {
287 rc = ll_dom_lock_cancel(inode, lock);
289 CDEBUG(D_INODE, "cannot flush DoM data "
291 PFID(ll_inode2fid(inode)), rc);
292 lock_res_and_lock(lock);
293 ldlm_set_kms_ignore(lock);
294 unlock_res_and_lock(lock);
297 if (bits & MDS_INODELOCK_LAYOUT) {
298 struct cl_object_conf conf = {
299 .coc_opc = OBJECT_CONF_INVALIDATE,
303 rc = ll_layout_conf(inode, &conf);
305 CDEBUG(D_INODE, "cannot invalidate layout of "
307 PFID(ll_inode2fid(inode)), rc);
310 if (bits & MDS_INODELOCK_UPDATE) {
311 struct ll_inode_info *lli = ll_i2info(inode);
313 lli->lli_update_atime = 1;
316 if ((bits & MDS_INODELOCK_UPDATE) && S_ISDIR(inode->i_mode)) {
317 struct ll_inode_info *lli = ll_i2info(inode);
319 CDEBUG(D_INODE, "invalidating inode "DFID" lli = %p, "
320 "pfid = "DFID"\n", PFID(ll_inode2fid(inode)),
321 lli, PFID(&lli->lli_pfid));
322 truncate_inode_pages(inode->i_mapping, 0);
324 if (unlikely(!fid_is_zero(&lli->lli_pfid))) {
325 struct inode *master_inode = NULL;
328 /* This is slave inode, since all of the child dentry
329 * is connected on the master inode, so we have to
330 * invalidate the negative children on master inode */
331 CDEBUG(D_INODE, "Invalidate s"DFID" m"DFID"\n",
332 PFID(ll_inode2fid(inode)), PFID(&lli->lli_pfid));
334 hash = cl_fid_build_ino(&lli->lli_pfid,
335 ll_need_32bit_api(ll_i2sbi(inode)));
337 /* Do not lookup the inode with ilookup5, otherwise
338 * it will cause dead lock,
339 * 1. Client1 send chmod req to the MDT0, then on MDT0,
340 * it enqueues master and all of its slaves lock,
341 * (mdt_attr_set() -> mdt_lock_slaves()), after gets
342 * master and stripe0 lock, it will send the enqueue
343 * req (for stripe1) to MDT1, then MDT1 finds the lock
344 * has been granted to client2. Then MDT1 sends blocking
346 * 2. At the same time, client2 tries to unlink
347 * the striped dir (rm -rf striped_dir), and during
348 * lookup, it will hold the master inode of the striped
349 * directory, whose inode state is NEW, then tries to
350 * revalidate all of its slaves, (ll_prep_inode()->
351 * ll_iget()->ll_read_inode2()-> ll_update_inode().).
352 * And it will be blocked on the server side because
354 * 3. Then the client get the blocking_ast req, cancel
355 * the lock, but being blocked if using ->ilookup5()),
356 * because master inode state is NEW. */
357 master_inode = ilookup5_nowait(inode->i_sb, hash,
358 ll_test_inode_by_fid,
359 (void *)&lli->lli_pfid);
361 ll_invalidate_negative_children(master_inode);
365 ll_invalidate_negative_children(inode);
369 if ((bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM)) &&
370 inode->i_sb->s_root != NULL &&
371 inode != inode->i_sb->s_root->d_inode)
372 ll_invalidate_aliases(inode);
377 /* Check if the given lock may be downgraded instead of canceling and
378 * that convert is really needed. */
379 int ll_md_need_convert(struct ldlm_lock *lock)
381 struct ldlm_namespace *ns = ldlm_lock_to_ns(lock);
383 __u64 wanted = lock->l_policy_data.l_inodebits.cancel_bits;
384 __u64 bits = lock->l_policy_data.l_inodebits.bits & ~wanted;
385 enum ldlm_mode mode = LCK_MINMODE;
387 if (!lock->l_conn_export ||
388 !exp_connect_lock_convert(lock->l_conn_export))
391 if (!wanted || !bits || ldlm_is_cancel(lock))
394 /* do not convert locks other than DOM for now */
395 if (!((bits | wanted) & MDS_INODELOCK_DOM))
398 /* We may have already remaining bits in some other lock so
399 * lock convert will leave us just extra lock for the same bit.
400 * Check if client has other lock with the same bits and the same
401 * or lower mode and don't convert if any.
403 switch (lock->l_req_mode) {
415 /* do not convert other modes */
419 /* is lock is too old to be converted? */
420 lock_res_and_lock(lock);
421 if (ktime_after(ktime_get(),
422 ktime_add(lock->l_last_used,
423 ktime_set(ns->ns_dirty_age_limit, 0)))) {
424 unlock_res_and_lock(lock);
427 unlock_res_and_lock(lock);
429 inode = ll_inode_from_resource_lock(lock);
430 ll_have_md_lock(inode, &bits, mode);
435 int ll_md_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
436 void *data, int flag)
438 struct lustre_handle lockh;
439 __u64 bits = lock->l_policy_data.l_inodebits.bits;
445 case LDLM_CB_BLOCKING:
447 __u64 cancel_flags = LCF_ASYNC;
449 if (ll_md_need_convert(lock)) {
450 cancel_flags |= LCF_CONVERT;
451 /* For lock convert some cancel actions may require
452 * this lock with non-dropped canceled bits, e.g. page
453 * flush for DOM lock. So call ll_lock_cancel_bits()
454 * here while canceled bits are still set.
456 bits = lock->l_policy_data.l_inodebits.cancel_bits;
457 if (bits & MDS_INODELOCK_DOM)
458 ll_lock_cancel_bits(lock, MDS_INODELOCK_DOM);
460 ldlm_lock2handle(lock, &lockh);
461 rc = ldlm_cli_cancel(&lockh, cancel_flags);
463 CDEBUG(D_INODE, "ldlm_cli_cancel: rc = %d\n", rc);
468 case LDLM_CB_CANCELING:
469 /* Nothing to do for non-granted locks */
470 if (!ldlm_is_granted(lock))
473 if (ldlm_is_converting(lock)) {
474 /* this is called on already converted lock, so
475 * ibits has remained bits only and cancel_bits
476 * are bits that were dropped.
477 * Note that DOM lock is handled prior lock convert
478 * and is excluded here.
480 bits = lock->l_policy_data.l_inodebits.cancel_bits &
483 LASSERT(ldlm_is_canceling(lock));
485 ll_lock_cancel_bits(lock, bits);
494 __u32 ll_i2suppgid(struct inode *i)
496 if (in_group_p(i->i_gid))
497 return (__u32)from_kgid(&init_user_ns, i->i_gid);
499 return (__u32) __kgid_val(INVALID_GID);
502 /* Pack the required supplementary groups into the supplied groups array.
503 * If we don't need to use the groups from the target inode(s) then we
504 * instead pack one or more groups from the user's supplementary group
505 * array in case it might be useful. Not needed if doing an MDS-side upcall. */
506 void ll_i2gids(__u32 *suppgids, struct inode *i1, struct inode *i2)
509 LASSERT(suppgids != NULL);
511 suppgids[0] = ll_i2suppgid(i1);
514 suppgids[1] = ll_i2suppgid(i2);
520 * try to reuse three types of dentry:
521 * 1. unhashed alias, this one is unhashed by d_invalidate (but it may be valid
522 * by concurrent .revalidate).
523 * 2. INVALID alias (common case for no valid ldlm lock held, but this flag may
524 * be cleared by others calling d_lustre_revalidate).
525 * 3. DISCONNECTED alias.
527 static struct dentry *ll_find_alias(struct inode *inode, struct dentry *dentry)
529 struct dentry *alias, *discon_alias, *invalid_alias;
530 DECLARE_LL_D_HLIST_NODE_PTR(p);
532 if (ll_d_hlist_empty(&inode->i_dentry))
535 discon_alias = invalid_alias = NULL;
537 ll_lock_dcache(inode);
538 ll_d_hlist_for_each_entry(alias, p, &inode->i_dentry) {
539 LASSERT(alias != dentry);
541 spin_lock(&alias->d_lock);
542 if ((alias->d_flags & DCACHE_DISCONNECTED) &&
543 S_ISDIR(inode->i_mode))
544 /* LASSERT(last_discon == NULL); LU-405, bz 20055 */
545 discon_alias = alias;
546 else if (alias->d_parent == dentry->d_parent &&
547 alias->d_name.hash == dentry->d_name.hash &&
548 alias->d_name.len == dentry->d_name.len &&
549 memcmp(alias->d_name.name, dentry->d_name.name,
550 dentry->d_name.len) == 0)
551 invalid_alias = alias;
552 spin_unlock(&alias->d_lock);
557 alias = invalid_alias ?: discon_alias ?: NULL;
559 spin_lock(&alias->d_lock);
561 spin_unlock(&alias->d_lock);
563 ll_unlock_dcache(inode);
569 * Similar to d_splice_alias(), but lustre treats invalid alias
570 * similar to DCACHE_DISCONNECTED, and tries to use it anyway.
572 struct dentry *ll_splice_alias(struct inode *inode, struct dentry *de)
578 new = ll_find_alias(inode, de);
588 "Reuse dentry %p inode %p refc %d flags %#x\n",
589 new, new->d_inode, ll_d_count(new), new->d_flags);
597 CDEBUG(D_DENTRY, "Add dentry %p inode %p refc %d flags %#x\n",
598 de, de->d_inode, ll_d_count(de), de->d_flags);
602 static int ll_lookup_it_finish(struct ptlrpc_request *request,
603 struct lookup_intent *it,
604 struct inode *parent, struct dentry **de)
606 struct inode *inode = NULL;
609 struct dentry *alias;
612 /* NB 1 request reference will be taken away by ll_intent_lock()
614 CDEBUG(D_DENTRY, "it %p it_disposition %x\n", it,
616 if (!it_disposition(it, DISP_LOOKUP_NEG)) {
617 rc = ll_prep_inode(&inode, request, (*de)->d_sb, it);
621 if (it->it_op & IT_OPEN)
622 ll_dom_finish_open(inode, request, it);
624 ll_set_lock_data(ll_i2sbi(parent)->ll_md_exp, inode, it, &bits);
626 /* We used to query real size from OSTs here, but actually
627 * this is not needed. For stat() calls size would be updated
628 * from subsequent do_revalidate()->ll_inode_revalidate_it() in
630 * vfs_getattr_it->ll_getattr()->ll_inode_revalidate_it() in 2.6
631 * Everybody else who needs correct file size would call
632 * ll_glimpse_size or some equivalent themselves anyway.
637 /* Only hash *de if it is unhashed (new dentry).
638 * Atoimc_open may passin hashed dentries for open.
640 alias = ll_splice_alias(inode, *de);
642 GOTO(out, rc = PTR_ERR(alias));
646 if (!it_disposition(it, DISP_LOOKUP_NEG)) {
647 /* we have lookup look - unhide dentry */
648 if (bits & MDS_INODELOCK_LOOKUP)
649 d_lustre_revalidate(*de);
650 } else if (!it_disposition(it, DISP_OPEN_CREATE)) {
651 /* If file created on server, don't depend on parent UPDATE
652 * lock to unhide it. It is left hidden and next lookup can
653 * find it in ll_splice_alias.
655 /* Check that parent has UPDATE lock. */
656 struct lookup_intent parent_it = {
658 .it_lock_handle = 0 };
659 struct lu_fid fid = ll_i2info(parent)->lli_fid;
661 /* If it is striped directory, get the real stripe parent */
662 if (unlikely(ll_i2info(parent)->lli_lsm_md != NULL)) {
663 rc = md_get_fid_from_lsm(ll_i2mdexp(parent),
664 ll_i2info(parent)->lli_lsm_md,
666 (*de)->d_name.len, &fid);
671 if (md_revalidate_lock(ll_i2mdexp(parent), &parent_it, &fid,
673 d_lustre_revalidate(*de);
674 ll_intent_release(&parent_it);
681 if (rc != 0 && it->it_op & IT_OPEN)
682 ll_open_cleanup((*de)->d_sb, request);
687 static struct dentry *ll_lookup_it(struct inode *parent, struct dentry *dentry,
688 struct lookup_intent *it,
689 void **secctx, __u32 *secctxlen)
691 struct lookup_intent lookup_it = { .it_op = IT_LOOKUP };
692 struct dentry *save = dentry, *retval;
693 struct ptlrpc_request *req = NULL;
694 struct md_op_data *op_data = NULL;
699 if (dentry->d_name.len > ll_i2sbi(parent)->ll_namelen)
700 RETURN(ERR_PTR(-ENAMETOOLONG));
702 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), intent=%s\n",
703 dentry->d_name.len, dentry->d_name.name,
704 PFID(ll_inode2fid(parent)), parent, LL_IT2STR(it));
706 if (d_mountpoint(dentry))
707 CERROR("Tell Peter, lookup on mtpt, it %s\n", LL_IT2STR(it));
709 if (it == NULL || it->it_op == IT_GETXATTR)
712 if (it->it_op == IT_GETATTR && dentry_may_statahead(parent, dentry)) {
713 rc = ll_statahead(parent, &dentry, 0);
715 RETURN(dentry == save ? NULL : dentry);
718 if (it->it_op & IT_OPEN && it->it_flags & FMODE_WRITE &&
719 dentry->d_sb->s_flags & MS_RDONLY)
720 RETURN(ERR_PTR(-EROFS));
722 if (it->it_op & IT_CREAT)
723 opc = LUSTRE_OPC_CREATE;
725 opc = LUSTRE_OPC_ANY;
727 op_data = ll_prep_md_op_data(NULL, parent, NULL, dentry->d_name.name,
728 dentry->d_name.len, 0, opc, NULL);
730 GOTO(out, retval = ERR_CAST(op_data));
732 /* enforce umask if acl disabled or MDS doesn't support umask */
733 if (!IS_POSIXACL(parent) || !exp_connect_umask(ll_i2mdexp(parent)))
734 it->it_create_mode &= ~current_umask();
736 if (it->it_op & IT_CREAT &&
737 ll_i2sbi(parent)->ll_flags & LL_SBI_FILE_SECCTX) {
738 rc = ll_dentry_init_security(dentry, it->it_create_mode,
740 &op_data->op_file_secctx_name,
741 &op_data->op_file_secctx,
742 &op_data->op_file_secctx_size);
744 GOTO(out, retval = ERR_PTR(rc));
746 *secctx = op_data->op_file_secctx;
747 if (secctxlen != NULL)
748 *secctxlen = op_data->op_file_secctx_size;
751 rc = md_intent_lock(ll_i2mdexp(parent), op_data, it, &req,
752 &ll_md_blocking_ast, 0);
753 /* If the MDS allows the client to chgrp (CFS_SETGRP_PERM), but the
754 * client does not know which suppgid should be sent to the MDS, or
755 * some other(s) changed the target file's GID after this RPC sent
756 * to the MDS with the suppgid as the original GID, then we should
757 * try again with right suppgid. */
758 if (rc == -EACCES && it->it_op & IT_OPEN &&
759 it_disposition(it, DISP_OPEN_DENY)) {
760 struct mdt_body *body;
762 LASSERT(req != NULL);
764 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
765 if (op_data->op_suppgids[0] == body->mbo_gid ||
766 op_data->op_suppgids[1] == body->mbo_gid ||
767 !in_group_p(make_kgid(&init_user_ns, body->mbo_gid)))
768 GOTO(out, retval = ERR_PTR(-EACCES));
770 fid_zero(&op_data->op_fid2);
771 op_data->op_suppgids[1] = body->mbo_gid;
772 ptlrpc_req_finished(req);
774 ll_intent_release(it);
775 rc = md_intent_lock(ll_i2mdexp(parent), op_data, it, &req,
776 &ll_md_blocking_ast, 0);
780 GOTO(out, retval = ERR_PTR(rc));
782 /* dir layout may change */
783 ll_unlock_md_op_lsm(op_data);
784 rc = ll_lookup_it_finish(req, it, parent, &dentry);
786 ll_intent_release(it);
787 GOTO(out, retval = ERR_PTR(rc));
790 if ((it->it_op & IT_OPEN) && dentry->d_inode &&
791 !S_ISREG(dentry->d_inode->i_mode) &&
792 !S_ISDIR(dentry->d_inode->i_mode)) {
793 ll_release_openhandle(dentry, it);
795 ll_lookup_finish_locks(it, dentry);
797 GOTO(out, retval = (dentry == save) ? NULL : dentry);
800 if (op_data != NULL && !IS_ERR(op_data)) {
801 if (secctx != NULL && secctxlen != NULL) {
802 /* caller needs sec ctx info, so reset it in op_data to
803 * prevent it from being freed */
804 op_data->op_file_secctx = NULL;
805 op_data->op_file_secctx_size = 0;
807 ll_finish_md_op_data(op_data);
810 ptlrpc_req_finished(req);
814 #ifdef HAVE_IOP_ATOMIC_OPEN
815 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
818 struct lookup_intent *itp, it = { .it_op = IT_GETATTR };
821 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), flags=%u\n",
822 dentry->d_name.len, dentry->d_name.name,
823 PFID(ll_inode2fid(parent)), parent, flags);
826 * Optimize away (CREATE && !OPEN). Let .create handle the race.
827 * but only if we have write permissions there, otherwise we need
828 * to proceed with lookup. LU-4185
830 if ((flags & LOOKUP_CREATE) && !(flags & LOOKUP_OPEN) &&
831 (inode_permission(parent, MAY_WRITE | MAY_EXEC) == 0))
834 if (flags & (LOOKUP_PARENT|LOOKUP_OPEN|LOOKUP_CREATE))
838 de = ll_lookup_it(parent, dentry, itp, NULL, NULL);
841 ll_intent_release(itp);
847 * For cached negative dentry and new dentry, handle lookup/create/open
850 static int ll_atomic_open(struct inode *dir, struct dentry *dentry,
851 struct file *file, unsigned open_flags,
852 umode_t mode, int *opened)
854 struct lookup_intent *it;
856 long long lookup_flags = LOOKUP_OPEN;
862 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), file %p,"
863 "open_flags %x, mode %x opened %d\n",
864 dentry->d_name.len, dentry->d_name.name,
865 PFID(ll_inode2fid(dir)), dir, file, open_flags, mode, *opened);
867 /* Only negative dentries enter here */
868 LASSERT(dentry->d_inode == NULL);
870 if (!d_unhashed(dentry)) {
871 /* A valid negative dentry that just passed revalidation,
872 * there's little point to try and open it server-side,
873 * even though there's a minuscule chance it might succeed.
874 * Either way it's a valid race to just return -ENOENT here.
876 if (!(open_flags & O_CREAT))
879 /* Otherwise we just unhash it to be rehashed afresh via
880 * lookup if necessary
885 OBD_ALLOC(it, sizeof(*it));
890 if (open_flags & O_CREAT) {
891 it->it_op |= IT_CREAT;
892 lookup_flags |= LOOKUP_CREATE;
894 it->it_create_mode = (mode & S_IALLUGO) | S_IFREG;
895 it->it_flags = (open_flags & ~O_ACCMODE) | OPEN_FMODE(open_flags);
896 it->it_flags &= ~MDS_OPEN_FL_INTERNAL;
898 /* Dentry added to dcache tree in ll_lookup_it */
899 de = ll_lookup_it(dir, dentry, it, &secctx, &secctxlen);
905 CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_FILE_PAUSE, cfs_fail_val);
908 if (it_disposition(it, DISP_OPEN_CREATE)) {
909 /* Dentry instantiated in ll_create_it. */
910 rc = ll_create_it(dir, dentry, it, secctx, secctxlen);
911 security_release_secctx(secctx, secctxlen);
913 /* We dget in ll_splice_alias. */
919 *opened |= FILE_CREATED;
921 if (dentry->d_inode && it_disposition(it, DISP_OPEN_OPEN)) {
923 if (S_ISFIFO(dentry->d_inode->i_mode)) {
924 /* We cannot call open here as it might
925 * deadlock. This case is unreachable in
926 * practice because of OBD_CONNECT_NODEVOH. */
927 rc = finish_no_open(file, de);
929 file->private_data = it;
930 rc = finish_open(file, dentry, NULL, opened);
931 /* We dget in ll_splice_alias. finish_open takes
932 * care of dget for fd open.
938 rc = finish_no_open(file, de);
943 ll_intent_release(it);
944 OBD_FREE(it, sizeof(*it));
949 #else /* !HAVE_IOP_ATOMIC_OPEN */
950 static struct lookup_intent *
951 ll_convert_intent(struct open_intent *oit, int lookup_flags, bool is_readonly)
953 struct lookup_intent *it;
957 return ERR_PTR(-ENOMEM);
959 if (lookup_flags & LOOKUP_OPEN) {
961 /* Avoid file creation for ro bind mount point(is_readonly) */
962 if ((lookup_flags & LOOKUP_CREATE) && !is_readonly)
963 it->it_op |= IT_CREAT;
964 it->it_create_mode = (oit->create_mode & S_IALLUGO) | S_IFREG;
965 it->it_flags = ll_namei_to_lookup_intent_flag(oit->flags &
966 ~(is_readonly ? O_CREAT : 0));
967 it->it_flags &= ~MDS_OPEN_FL_INTERNAL;
969 it->it_op = IT_GETATTR;
975 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
976 struct nameidata *nd)
981 if (nd && !(nd->flags & (LOOKUP_CONTINUE|LOOKUP_PARENT))) {
982 struct lookup_intent *it;
984 if (ll_d2d(dentry) && ll_d2d(dentry)->lld_it) {
985 it = ll_d2d(dentry)->lld_it;
986 ll_d2d(dentry)->lld_it = NULL;
989 * Optimize away (CREATE && !OPEN). Let .create handle
990 * the race. But only if we have write permissions
991 * there, otherwise we need to proceed with lookup.
994 if ((nd->flags & LOOKUP_CREATE) &&
995 !(nd->flags & LOOKUP_OPEN) &&
996 (inode_permission(parent,
997 MAY_WRITE | MAY_EXEC) == 0))
1000 it = ll_convert_intent(&nd->intent.open, nd->flags,
1001 (nd->path.mnt->mnt_flags & MNT_READONLY) ||
1002 (nd->path.mnt->mnt_sb->s_flags & MS_RDONLY));
1004 RETURN((struct dentry *)it);
1007 de = ll_lookup_it(parent, dentry, it, NULL, NULL);
1010 if ((nd->flags & LOOKUP_OPEN) && !IS_ERR(dentry)) { /* Open */
1011 if (dentry->d_inode &&
1012 it_disposition(it, DISP_OPEN_OPEN)) { /* nocreate */
1013 if (S_ISFIFO(dentry->d_inode->i_mode)) {
1014 /* We cannot call open here as it might
1015 * deadlock. This case is unreachable in
1016 * practice because of
1017 * OBD_CONNECT_NODEVOH. */
1021 nd->intent.open.file->private_data = it;
1022 filp = lookup_instantiate_filp(nd,
1028 de = (struct dentry *)filp;
1031 } else if (it_disposition(it, DISP_OPEN_CREATE)) {
1032 /* XXX This can only reliably work on assumption
1033 * that there are NO hashed negative dentries.*/
1034 ll_d2d(dentry)->lld_it = it;
1035 it = NULL; /* Will be freed in ll_create_nd */
1036 /* We absolutely depend on ll_create_nd to be
1037 * called to not leak this intent and possible
1038 * data attached to it */
1043 ll_intent_release(it);
1044 OBD_FREE(it, sizeof(*it));
1047 de = ll_lookup_it(parent, dentry, NULL, NULL, NULL);
1052 #endif /* HAVE_IOP_ATOMIC_OPEN */
1054 /* We depend on "mode" being set with the proper file type/umask by now */
1055 static struct inode *ll_create_node(struct inode *dir, struct lookup_intent *it)
1057 struct inode *inode = NULL;
1058 struct ptlrpc_request *request = NULL;
1059 struct ll_sb_info *sbi = ll_i2sbi(dir);
1063 LASSERT(it && it->it_disposition);
1065 LASSERT(it_disposition(it, DISP_ENQ_CREATE_REF));
1066 request = it->it_request;
1067 it_clear_disposition(it, DISP_ENQ_CREATE_REF);
1068 rc = ll_prep_inode(&inode, request, dir->i_sb, it);
1070 GOTO(out, inode = ERR_PTR(rc));
1072 /* Pause to allow for a race with concurrent access by fid */
1073 OBD_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_NODE_PAUSE, cfs_fail_val);
1075 /* We asked for a lock on the directory, but were granted a
1076 * lock on the inode. Since we finally have an inode pointer,
1077 * stuff it in the lock. */
1078 CDEBUG(D_DLMTRACE, "setting l_ast_data to inode "DFID"(%p)\n",
1079 PFID(ll_inode2fid(inode)), inode);
1080 ll_set_lock_data(sbi->ll_md_exp, inode, it, NULL);
1083 ptlrpc_req_finished(request);
1088 * By the time this is called, we already have created the directory cache
1089 * entry for the new file, but it is so far negative - it has no inode.
1091 * We defer creating the OBD object(s) until open, to keep the intent and
1092 * non-intent code paths similar, and also because we do not have the MDS
1093 * inode number before calling ll_create_node() (which is needed for LOV),
1094 * so we would need to do yet another RPC to the MDS to store the LOV EA
1095 * data on the MDS. If needed, we would pass the PACKED lmm as data and
1096 * lmm_size in datalen (the MDS still has code which will handle that).
1098 * If the create succeeds, we fill in the inode information
1099 * with d_instantiate().
1101 static int ll_create_it(struct inode *dir, struct dentry *dentry,
1102 struct lookup_intent *it,
1103 void *secctx, __u32 secctxlen)
1105 struct inode *inode;
1109 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), intent=%s\n",
1110 dentry->d_name.len, dentry->d_name.name,
1111 PFID(ll_inode2fid(dir)), dir, LL_IT2STR(it));
1113 rc = it_open_error(DISP_OPEN_CREATE, it);
1117 inode = ll_create_node(dir, it);
1119 RETURN(PTR_ERR(inode));
1121 if ((ll_i2sbi(inode)->ll_flags & LL_SBI_FILE_SECCTX) &&
1124 /* must be done before d_instantiate, because it calls
1125 * security_d_instantiate, which means a getxattr if security
1126 * context is not set yet */
1127 rc = security_inode_notifysecctx(inode, secctx, secctxlen);
1128 inode_unlock(inode);
1133 d_instantiate(dentry, inode);
1135 if (!(ll_i2sbi(inode)->ll_flags & LL_SBI_FILE_SECCTX)) {
1136 rc = ll_inode_init_security(dentry, inode, dir);
1144 void ll_update_times(struct ptlrpc_request *request, struct inode *inode)
1146 struct mdt_body *body = req_capsule_server_get(&request->rq_pill,
1150 if (body->mbo_valid & OBD_MD_FLMTIME &&
1151 body->mbo_mtime > inode->i_mtime.tv_sec) {
1153 "setting fid " DFID " mtime from %lld to %llu\n",
1154 PFID(ll_inode2fid(inode)),
1155 (s64)inode->i_mtime.tv_sec, body->mbo_mtime);
1156 inode->i_mtime.tv_sec = body->mbo_mtime;
1159 if (body->mbo_valid & OBD_MD_FLCTIME &&
1160 body->mbo_ctime > inode->i_ctime.tv_sec)
1161 inode->i_ctime.tv_sec = body->mbo_ctime;
1164 static int ll_new_node(struct inode *dir, struct dentry *dchild,
1165 const char *tgt, umode_t mode, int rdev, __u32 opc)
1167 struct qstr *name = &dchild->d_name;
1168 struct ptlrpc_request *request = NULL;
1169 struct md_op_data *op_data;
1170 struct inode *inode = NULL;
1171 struct ll_sb_info *sbi = ll_i2sbi(dir);
1176 if (unlikely(tgt != NULL))
1177 tgt_len = strlen(tgt) + 1;
1180 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
1181 name->len, 0, opc, NULL);
1182 if (IS_ERR(op_data))
1183 GOTO(err_exit, err = PTR_ERR(op_data));
1185 if (sbi->ll_flags & LL_SBI_FILE_SECCTX) {
1186 err = ll_dentry_init_security(dchild, mode, &dchild->d_name,
1187 &op_data->op_file_secctx_name,
1188 &op_data->op_file_secctx,
1189 &op_data->op_file_secctx_size);
1191 GOTO(err_exit, err);
1194 err = md_create(sbi->ll_md_exp, op_data, tgt, tgt_len, mode,
1195 from_kuid(&init_user_ns, current_fsuid()),
1196 from_kgid(&init_user_ns, current_fsgid()),
1197 cfs_curproc_cap_pack(), rdev, &request);
1198 if (err < 0 && err != -EREMOTE)
1199 GOTO(err_exit, err);
1201 /* If the client doesn't know where to create a subdirectory (or
1202 * in case of a race that sends the RPC to the wrong MDS), the
1203 * MDS will return -EREMOTE and the client will fetch the layout
1204 * of the directory, then create the directory on the right MDT. */
1205 if (unlikely(err == -EREMOTE)) {
1206 struct ll_inode_info *lli = ll_i2info(dir);
1207 struct lmv_user_md *lum;
1211 ptlrpc_req_finished(request);
1214 err2 = ll_dir_getstripe(dir, (void **)&lum, &lumsize, &request,
1215 OBD_MD_DEFAULT_MEA);
1217 /* Update stripe_offset and retry */
1218 lli->lli_def_stripe_offset = lum->lum_stripe_offset;
1219 } else if (err2 == -ENODATA &&
1220 lli->lli_def_stripe_offset != -1) {
1221 /* If there are no default stripe EA on the MDT, but the
1222 * client has default stripe, then it probably means
1223 * default stripe EA has just been deleted. */
1224 lli->lli_def_stripe_offset = -1;
1226 GOTO(err_exit, err);
1229 ptlrpc_req_finished(request);
1231 ll_finish_md_op_data(op_data);
1235 ll_update_times(request, dir);
1237 CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_NEWNODE_PAUSE, cfs_fail_val);
1239 err = ll_prep_inode(&inode, request, dchild->d_sb, NULL);
1241 GOTO(err_exit, err);
1243 if (sbi->ll_flags & LL_SBI_FILE_SECCTX) {
1245 /* must be done before d_instantiate, because it calls
1246 * security_d_instantiate, which means a getxattr if security
1247 * context is not set yet */
1248 err = security_inode_notifysecctx(inode,
1249 op_data->op_file_secctx,
1250 op_data->op_file_secctx_size);
1251 inode_unlock(inode);
1253 GOTO(err_exit, err);
1256 d_instantiate(dchild, inode);
1258 if (!(sbi->ll_flags & LL_SBI_FILE_SECCTX)) {
1259 err = ll_inode_init_security(dchild, inode, dir);
1261 GOTO(err_exit, err);
1266 if (request != NULL)
1267 ptlrpc_req_finished(request);
1269 if (!IS_ERR_OR_NULL(op_data))
1270 ll_finish_md_op_data(op_data);
1275 static int ll_mknod(struct inode *dir, struct dentry *dchild, ll_umode_t mode,
1278 struct qstr *name = &dchild->d_name;
1282 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p) mode %o dev %x\n",
1283 name->len, name->name, PFID(ll_inode2fid(dir)), dir,
1286 if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
1287 mode &= ~current_umask();
1289 switch (mode & S_IFMT) {
1291 mode |= S_IFREG; /* for mode = 0 case, fallthrough */
1297 err = ll_new_node(dir, dchild, NULL, mode, old_encode_dev(rdev),
1308 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKNOD, 1);
1313 #ifdef HAVE_IOP_ATOMIC_OPEN
1315 * Plain create. Intent create is handled in atomic_open.
1317 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
1318 umode_t mode, bool want_excl)
1322 CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_FILE_PAUSE, cfs_fail_val);
1324 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), "
1325 "flags=%u, excl=%d\n", dentry->d_name.len,
1326 dentry->d_name.name, PFID(ll_inode2fid(dir)),
1327 dir, mode, want_excl);
1329 /* Using mknod(2) to create a regular file is designed to not recognize
1330 * volatile file name, so we use ll_mknod() here. */
1331 rc = ll_mknod(dir, dentry, mode, 0);
1333 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_CREATE, 1);
1335 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, unhashed %d\n",
1336 dentry->d_name.len, dentry->d_name.name, d_unhashed(dentry));
1340 #else /* !HAVE_IOP_ATOMIC_OPEN */
1341 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
1342 ll_umode_t mode, struct nameidata *nd)
1344 struct ll_dentry_data *lld = ll_d2d(dentry);
1345 struct lookup_intent *it = NULL;
1348 CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_FILE_PAUSE, cfs_fail_val);
1354 /* LU-8559: use LUSTRE_OPC_CREATE for non atomic open case
1355 * so that volatile file name is recoginized.
1356 * Mknod(2), however, is designed to not recognize volatile
1357 * file name to avoid inode leak under orphan directory until
1359 return ll_new_node(dir, dentry, NULL, mode, 0,
1365 /* Was there an error? Propagate it! */
1366 if (it->it_status) {
1371 rc = ll_create_it(dir, dentry, it, NULL, 0);
1372 if (nd && (nd->flags & LOOKUP_OPEN) && dentry->d_inode) { /* Open */
1375 nd->intent.open.file->private_data = it;
1376 filp = lookup_instantiate_filp(nd, dentry, NULL);
1382 ll_intent_release(it);
1383 OBD_FREE(it, sizeof(*it));
1386 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_CREATE, 1);
1390 #endif /* HAVE_IOP_ATOMIC_OPEN */
1392 static int ll_symlink(struct inode *dir, struct dentry *dchild,
1393 const char *oldpath)
1395 struct qstr *name = &dchild->d_name;
1399 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), target=%.*s\n",
1400 name->len, name->name, PFID(ll_inode2fid(dir)),
1401 dir, 3000, oldpath);
1403 err = ll_new_node(dir, dchild, oldpath, S_IFLNK | S_IRWXUGO, 0,
1404 LUSTRE_OPC_SYMLINK);
1407 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_SYMLINK, 1);
1412 static int ll_link(struct dentry *old_dentry, struct inode *dir,
1413 struct dentry *new_dentry)
1415 struct inode *src = old_dentry->d_inode;
1416 struct qstr *name = &new_dentry->d_name;
1417 struct ll_sb_info *sbi = ll_i2sbi(dir);
1418 struct ptlrpc_request *request = NULL;
1419 struct md_op_data *op_data;
1423 CDEBUG(D_VFSTRACE, "VFS Op: inode="DFID"(%p), dir="DFID"(%p), "
1424 "target=%.*s\n", PFID(ll_inode2fid(src)), src,
1425 PFID(ll_inode2fid(dir)), dir, name->len, name->name);
1427 op_data = ll_prep_md_op_data(NULL, src, dir, name->name, name->len,
1428 0, LUSTRE_OPC_ANY, NULL);
1429 if (IS_ERR(op_data))
1430 RETURN(PTR_ERR(op_data));
1432 err = md_link(sbi->ll_md_exp, op_data, &request);
1433 ll_finish_md_op_data(op_data);
1437 ll_update_times(request, dir);
1438 ll_stats_ops_tally(sbi, LPROC_LL_LINK, 1);
1441 ptlrpc_req_finished(request);
1445 static int ll_mkdir(struct inode *dir, struct dentry *dchild, ll_umode_t mode)
1447 struct qstr *name = &dchild->d_name;
1451 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1452 name->len, name->name, PFID(ll_inode2fid(dir)), dir);
1454 if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
1455 mode &= ~current_umask();
1457 mode = (mode & (S_IRWXUGO|S_ISVTX)) | S_IFDIR;
1459 err = ll_new_node(dir, dchild, NULL, mode, 0, LUSTRE_OPC_MKDIR);
1461 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKDIR, 1);
1466 static int ll_rmdir(struct inode *dir, struct dentry *dchild)
1468 struct qstr *name = &dchild->d_name;
1469 struct ptlrpc_request *request = NULL;
1470 struct md_op_data *op_data;
1474 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1475 name->len, name->name, PFID(ll_inode2fid(dir)), dir);
1477 if (unlikely(d_mountpoint(dchild)))
1480 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len,
1481 S_IFDIR, LUSTRE_OPC_ANY, NULL);
1482 if (IS_ERR(op_data))
1483 RETURN(PTR_ERR(op_data));
1485 if (dchild->d_inode != NULL)
1486 op_data->op_fid3 = *ll_inode2fid(dchild->d_inode);
1488 op_data->op_fid2 = op_data->op_fid3;
1489 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1490 ll_finish_md_op_data(op_data);
1492 ll_update_times(request, dir);
1493 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR, 1);
1496 ptlrpc_req_finished(request);
1503 int ll_rmdir_entry(struct inode *dir, char *name, int namelen)
1505 struct ptlrpc_request *request = NULL;
1506 struct md_op_data *op_data;
1510 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1511 namelen, name, PFID(ll_inode2fid(dir)), dir);
1513 op_data = ll_prep_md_op_data(NULL, dir, NULL, name, strlen(name),
1514 S_IFDIR, LUSTRE_OPC_ANY, NULL);
1515 if (IS_ERR(op_data))
1516 RETURN(PTR_ERR(op_data));
1517 op_data->op_cli_flags |= CLI_RM_ENTRY;
1518 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1519 ll_finish_md_op_data(op_data);
1521 ll_update_times(request, dir);
1522 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR, 1);
1525 ptlrpc_req_finished(request);
1529 static int ll_unlink(struct inode *dir, struct dentry *dchild)
1531 struct qstr *name = &dchild->d_name;
1532 struct ptlrpc_request *request = NULL;
1533 struct md_op_data *op_data;
1534 struct mdt_body *body;
1537 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1538 name->len, name->name, PFID(ll_inode2fid(dir)), dir);
1541 * XXX: unlink bind mountpoint maybe call to here,
1542 * just check it as vfs_unlink does.
1544 if (unlikely(d_mountpoint(dchild)))
1547 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len, 0,
1548 LUSTRE_OPC_ANY, NULL);
1549 if (IS_ERR(op_data))
1550 RETURN(PTR_ERR(op_data));
1552 op_data->op_fid3 = *ll_inode2fid(dchild->d_inode);
1554 op_data->op_fid2 = op_data->op_fid3;
1555 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1556 ll_finish_md_op_data(op_data);
1561 * The server puts attributes in on the last unlink, use them to update
1562 * the link count so the inode can be freed immediately.
1564 body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
1565 if (body->mbo_valid & OBD_MD_FLNLINK)
1566 set_nlink(dchild->d_inode, body->mbo_nlink);
1568 ll_update_times(request, dir);
1569 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_UNLINK, 1);
1572 ptlrpc_req_finished(request);
1576 static int ll_rename(struct inode *src, struct dentry *src_dchild,
1577 struct inode *tgt, struct dentry *tgt_dchild
1578 #ifdef HAVE_IOPS_RENAME_WITH_FLAGS
1579 , unsigned int flags
1583 struct qstr *src_name = &src_dchild->d_name;
1584 struct qstr *tgt_name = &tgt_dchild->d_name;
1585 struct ptlrpc_request *request = NULL;
1586 struct ll_sb_info *sbi = ll_i2sbi(src);
1587 struct md_op_data *op_data;
1591 #ifdef HAVE_IOPS_RENAME_WITH_FLAGS
1596 CDEBUG(D_VFSTRACE, "VFS Op:oldname=%.*s, src_dir="DFID
1597 "(%p), newname=%.*s, tgt_dir="DFID"(%p)\n",
1598 src_name->len, src_name->name,
1599 PFID(ll_inode2fid(src)), src, tgt_name->len,
1600 tgt_name->name, PFID(ll_inode2fid(tgt)), tgt);
1602 if (unlikely(d_mountpoint(src_dchild) || d_mountpoint(tgt_dchild)))
1605 op_data = ll_prep_md_op_data(NULL, src, tgt, NULL, 0, 0,
1606 LUSTRE_OPC_ANY, NULL);
1607 if (IS_ERR(op_data))
1608 RETURN(PTR_ERR(op_data));
1610 if (src_dchild->d_inode != NULL)
1611 op_data->op_fid3 = *ll_inode2fid(src_dchild->d_inode);
1613 if (tgt_dchild->d_inode != NULL)
1614 op_data->op_fid4 = *ll_inode2fid(tgt_dchild->d_inode);
1616 err = md_rename(sbi->ll_md_exp, op_data,
1617 src_name->name, src_name->len,
1618 tgt_name->name, tgt_name->len, &request);
1619 ll_finish_md_op_data(op_data);
1621 ll_update_times(request, src);
1622 ll_update_times(request, tgt);
1623 ll_stats_ops_tally(sbi, LPROC_LL_RENAME, 1);
1626 ptlrpc_req_finished(request);
1629 d_move(src_dchild, tgt_dchild);
1634 const struct inode_operations ll_dir_inode_operations = {
1636 #ifdef HAVE_IOP_ATOMIC_OPEN
1637 .atomic_open = ll_atomic_open,
1639 .lookup = ll_lookup_nd,
1640 .create = ll_create_nd,
1641 /* We need all these non-raw things for NFSD, to not patch it. */
1642 .unlink = ll_unlink,
1645 .symlink = ll_symlink,
1647 .rename = ll_rename,
1648 .setattr = ll_setattr,
1649 .getattr = ll_getattr,
1650 .permission = ll_inode_permission,
1651 #ifdef HAVE_IOP_XATTR
1652 .setxattr = ll_setxattr,
1653 .getxattr = ll_getxattr,
1654 .removexattr = ll_removexattr,
1656 .listxattr = ll_listxattr,
1657 #ifdef HAVE_IOP_GET_ACL
1658 .get_acl = ll_get_acl,
1660 #ifdef HAVE_IOP_SET_ACL
1661 .set_acl = ll_set_acl,
1665 const struct inode_operations ll_special_inode_operations = {
1666 .setattr = ll_setattr,
1667 .getattr = ll_getattr,
1668 .permission = ll_inode_permission,
1669 #ifdef HAVE_IOP_XATTR
1670 .setxattr = ll_setxattr,
1671 .getxattr = ll_getxattr,
1672 .removexattr = ll_removexattr,
1674 .listxattr = ll_listxattr,
1675 #ifdef HAVE_IOP_GET_ACL
1676 .get_acl = ll_get_acl,
1678 #ifdef HAVE_IOP_SET_ACL
1679 .set_acl = ll_set_acl,