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 /* That means the inode is evicted most likely and may cause
240 * the skipping of lock cleanups below, so print the message
243 if (lock->l_resource->lr_lvb_inode)
245 "can't take inode for the lock (%sevicted)\n",
246 lock->l_resource->lr_lvb_inode->i_state &
247 I_FREEING ? "" : "not ");
251 if (!fid_res_name_eq(ll_inode2fid(inode),
252 &lock->l_resource->lr_name)) {
253 LDLM_ERROR(lock, "data mismatch with object "DFID"(%p)",
254 PFID(ll_inode2fid(inode)), inode);
258 if (bits & MDS_INODELOCK_XATTR) {
259 if (S_ISDIR(inode->i_mode))
260 ll_i2info(inode)->lli_def_stripe_offset = -1;
261 ll_xattr_cache_destroy(inode);
262 bits &= ~MDS_INODELOCK_XATTR;
265 /* For OPEN locks we differentiate between lock modes
266 * LCK_CR, LCK_CW, LCK_PR - bug 22891 */
267 if (bits & MDS_INODELOCK_OPEN)
268 ll_have_md_lock(inode, &bits, lock->l_req_mode);
270 if (bits & MDS_INODELOCK_OPEN) {
273 switch (lock->l_req_mode) {
284 LDLM_ERROR(lock, "bad lock mode for OPEN lock");
288 ll_md_real_close(inode, fmode);
290 bits &= ~MDS_INODELOCK_OPEN;
293 if (bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
294 MDS_INODELOCK_LAYOUT | MDS_INODELOCK_PERM |
296 ll_have_md_lock(inode, &bits, LCK_MINMODE);
298 if (bits & MDS_INODELOCK_DOM) {
299 rc = ll_dom_lock_cancel(inode, lock);
301 CDEBUG(D_INODE, "cannot flush DoM data "
303 PFID(ll_inode2fid(inode)), rc);
304 lock_res_and_lock(lock);
305 ldlm_set_kms_ignore(lock);
306 unlock_res_and_lock(lock);
309 if (bits & MDS_INODELOCK_LAYOUT) {
310 struct cl_object_conf conf = {
311 .coc_opc = OBJECT_CONF_INVALIDATE,
315 rc = ll_layout_conf(inode, &conf);
317 CDEBUG(D_INODE, "cannot invalidate layout of "
319 PFID(ll_inode2fid(inode)), rc);
322 if (bits & MDS_INODELOCK_UPDATE) {
323 struct ll_inode_info *lli = ll_i2info(inode);
325 lli->lli_update_atime = 1;
328 if ((bits & MDS_INODELOCK_UPDATE) && S_ISDIR(inode->i_mode)) {
329 struct ll_inode_info *lli = ll_i2info(inode);
331 CDEBUG(D_INODE, "invalidating inode "DFID" lli = %p, "
332 "pfid = "DFID"\n", PFID(ll_inode2fid(inode)),
333 lli, PFID(&lli->lli_pfid));
334 truncate_inode_pages(inode->i_mapping, 0);
336 if (unlikely(!fid_is_zero(&lli->lli_pfid))) {
337 struct inode *master_inode = NULL;
340 /* This is slave inode, since all of the child dentry
341 * is connected on the master inode, so we have to
342 * invalidate the negative children on master inode */
343 CDEBUG(D_INODE, "Invalidate s"DFID" m"DFID"\n",
344 PFID(ll_inode2fid(inode)), PFID(&lli->lli_pfid));
346 hash = cl_fid_build_ino(&lli->lli_pfid,
347 ll_need_32bit_api(ll_i2sbi(inode)));
349 /* Do not lookup the inode with ilookup5, otherwise
350 * it will cause dead lock,
351 * 1. Client1 send chmod req to the MDT0, then on MDT0,
352 * it enqueues master and all of its slaves lock,
353 * (mdt_attr_set() -> mdt_lock_slaves()), after gets
354 * master and stripe0 lock, it will send the enqueue
355 * req (for stripe1) to MDT1, then MDT1 finds the lock
356 * has been granted to client2. Then MDT1 sends blocking
358 * 2. At the same time, client2 tries to unlink
359 * the striped dir (rm -rf striped_dir), and during
360 * lookup, it will hold the master inode of the striped
361 * directory, whose inode state is NEW, then tries to
362 * revalidate all of its slaves, (ll_prep_inode()->
363 * ll_iget()->ll_read_inode2()-> ll_update_inode().).
364 * And it will be blocked on the server side because
366 * 3. Then the client get the blocking_ast req, cancel
367 * the lock, but being blocked if using ->ilookup5()),
368 * because master inode state is NEW. */
369 master_inode = ilookup5_nowait(inode->i_sb, hash,
370 ll_test_inode_by_fid,
371 (void *)&lli->lli_pfid);
373 ll_invalidate_negative_children(master_inode);
377 ll_invalidate_negative_children(inode);
381 if ((bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM)) &&
382 inode->i_sb->s_root != NULL &&
383 inode != inode->i_sb->s_root->d_inode)
384 ll_invalidate_aliases(inode);
386 if (bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM))
387 forget_all_cached_acls(inode);
393 /* Check if the given lock may be downgraded instead of canceling and
394 * that convert is really needed. */
395 int ll_md_need_convert(struct ldlm_lock *lock)
397 struct ldlm_namespace *ns = ldlm_lock_to_ns(lock);
399 __u64 wanted = lock->l_policy_data.l_inodebits.cancel_bits;
400 __u64 bits = lock->l_policy_data.l_inodebits.bits & ~wanted;
401 enum ldlm_mode mode = LCK_MINMODE;
403 if (!lock->l_conn_export ||
404 !exp_connect_lock_convert(lock->l_conn_export))
407 if (!wanted || !bits || ldlm_is_cancel(lock))
410 /* do not convert locks other than DOM for now */
411 if (!((bits | wanted) & MDS_INODELOCK_DOM))
414 /* We may have already remaining bits in some other lock so
415 * lock convert will leave us just extra lock for the same bit.
416 * Check if client has other lock with the same bits and the same
417 * or lower mode and don't convert if any.
419 switch (lock->l_req_mode) {
431 /* do not convert other modes */
435 /* is lock is too old to be converted? */
436 lock_res_and_lock(lock);
437 if (ktime_after(ktime_get(),
438 ktime_add(lock->l_last_used,
439 ktime_set(ns->ns_dirty_age_limit, 0)))) {
440 unlock_res_and_lock(lock);
443 unlock_res_and_lock(lock);
445 inode = ll_inode_from_resource_lock(lock);
446 ll_have_md_lock(inode, &bits, mode);
451 int ll_md_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *ld,
452 void *data, int flag)
454 struct lustre_handle lockh;
460 case LDLM_CB_BLOCKING:
462 __u64 cancel_flags = LCF_ASYNC;
464 /* if lock convert is not needed then still have to
465 * pass lock via ldlm_cli_convert() to keep all states
466 * correct, set cancel_bits to full lock bits to cause
467 * full cancel to happen.
469 if (!ll_md_need_convert(lock)) {
470 lock_res_and_lock(lock);
471 lock->l_policy_data.l_inodebits.cancel_bits =
472 lock->l_policy_data.l_inodebits.bits;
473 unlock_res_and_lock(lock);
475 rc = ldlm_cli_convert(lock, cancel_flags);
478 /* continue with cancel otherwise */
479 ldlm_lock2handle(lock, &lockh);
480 rc = ldlm_cli_cancel(&lockh, cancel_flags);
482 CDEBUG(D_INODE, "ldlm_cli_cancel: rc = %d\n", rc);
487 case LDLM_CB_CANCELING:
489 __u64 to_cancel = lock->l_policy_data.l_inodebits.bits;
491 /* Nothing to do for non-granted locks */
492 if (!ldlm_is_granted(lock))
495 /* If 'ld' is supplied then bits to be cancelled are passed
496 * implicitly by lock converting and cancel_bits from 'ld'
497 * should be used. Otherwise full cancel is being performed
498 * and lock inodebits are used.
500 * Note: we cannot rely on cancel_bits in lock itself at this
501 * moment because they can be changed by concurrent thread,
502 * so ldlm_cli_inodebits_convert() pass cancel bits implicitly
506 /* partial bits cancel allowed only during convert */
507 LASSERT(ldlm_is_converting(lock));
508 /* mask cancel bits by lock bits so only no any unused
509 * bits are passed to ll_lock_cancel_bits()
511 to_cancel &= ld->l_policy_data.l_inodebits.cancel_bits;
513 ll_lock_cancel_bits(lock, to_cancel);
523 __u32 ll_i2suppgid(struct inode *i)
525 if (in_group_p(i->i_gid))
526 return (__u32)from_kgid(&init_user_ns, i->i_gid);
528 return (__u32) __kgid_val(INVALID_GID);
531 /* Pack the required supplementary groups into the supplied groups array.
532 * If we don't need to use the groups from the target inode(s) then we
533 * instead pack one or more groups from the user's supplementary group
534 * array in case it might be useful. Not needed if doing an MDS-side upcall. */
535 void ll_i2gids(__u32 *suppgids, struct inode *i1, struct inode *i2)
538 LASSERT(suppgids != NULL);
540 suppgids[0] = ll_i2suppgid(i1);
543 suppgids[1] = ll_i2suppgid(i2);
549 * try to reuse three types of dentry:
550 * 1. unhashed alias, this one is unhashed by d_invalidate (but it may be valid
551 * by concurrent .revalidate).
552 * 2. INVALID alias (common case for no valid ldlm lock held, but this flag may
553 * be cleared by others calling d_lustre_revalidate).
554 * 3. DISCONNECTED alias.
556 static struct dentry *ll_find_alias(struct inode *inode, struct dentry *dentry)
558 struct dentry *alias, *discon_alias, *invalid_alias;
559 DECLARE_LL_D_HLIST_NODE_PTR(p);
561 if (ll_d_hlist_empty(&inode->i_dentry))
564 discon_alias = invalid_alias = NULL;
566 ll_lock_dcache(inode);
567 ll_d_hlist_for_each_entry(alias, p, &inode->i_dentry) {
568 LASSERT(alias != dentry);
570 spin_lock(&alias->d_lock);
571 if ((alias->d_flags & DCACHE_DISCONNECTED) &&
572 S_ISDIR(inode->i_mode))
573 /* LASSERT(last_discon == NULL); LU-405, bz 20055 */
574 discon_alias = alias;
575 else if (alias->d_parent == dentry->d_parent &&
576 alias->d_name.hash == dentry->d_name.hash &&
577 alias->d_name.len == dentry->d_name.len &&
578 memcmp(alias->d_name.name, dentry->d_name.name,
579 dentry->d_name.len) == 0)
580 invalid_alias = alias;
581 spin_unlock(&alias->d_lock);
586 alias = invalid_alias ?: discon_alias ?: NULL;
588 spin_lock(&alias->d_lock);
590 spin_unlock(&alias->d_lock);
592 ll_unlock_dcache(inode);
598 * Similar to d_splice_alias(), but lustre treats invalid alias
599 * similar to DCACHE_DISCONNECTED, and tries to use it anyway.
601 struct dentry *ll_splice_alias(struct inode *inode, struct dentry *de)
607 new = ll_find_alias(inode, de);
617 "Reuse dentry %p inode %p refc %d flags %#x\n",
618 new, new->d_inode, ll_d_count(new), new->d_flags);
626 CDEBUG(D_DENTRY, "Add dentry %p inode %p refc %d flags %#x\n",
627 de, de->d_inode, ll_d_count(de), de->d_flags);
631 static int ll_lookup_it_finish(struct ptlrpc_request *request,
632 struct lookup_intent *it,
633 struct inode *parent, struct dentry **de,
636 struct inode *inode = NULL;
639 struct dentry *alias;
642 /* NB 1 request reference will be taken away by ll_intent_lock()
644 CDEBUG(D_DENTRY, "it %p it_disposition %x\n", it,
646 if (!it_disposition(it, DISP_LOOKUP_NEG)) {
647 rc = ll_prep_inode(&inode, request, (*de)->d_sb, it);
651 if (it->it_op & IT_OPEN)
652 ll_dom_finish_open(inode, request, it);
654 ll_set_lock_data(ll_i2sbi(parent)->ll_md_exp, inode, it, &bits);
656 /* We used to query real size from OSTs here, but actually
657 * this is not needed. For stat() calls size would be updated
658 * from subsequent do_revalidate()->ll_inode_revalidate_it() in
660 * vfs_getattr_it->ll_getattr()->ll_inode_revalidate_it() in 2.6
661 * Everybody else who needs correct file size would call
662 * ll_glimpse_size or some equivalent themselves anyway.
667 /* Only hash *de if it is unhashed (new dentry).
668 * Atoimc_open may passin hashed dentries for open.
670 alias = ll_splice_alias(inode, *de);
672 GOTO(out, rc = PTR_ERR(alias));
676 if (!it_disposition(it, DISP_LOOKUP_NEG)) {
677 /* we have lookup look - unhide dentry */
678 if (bits & MDS_INODELOCK_LOOKUP)
679 d_lustre_revalidate(*de);
680 } else if (!it_disposition(it, DISP_OPEN_CREATE)) {
682 * If file was created on the server, the dentry is revalidated
683 * in ll_create_it if the lock allows for it.
685 /* Check that parent has UPDATE lock. */
686 struct lookup_intent parent_it = {
688 .it_lock_handle = 0 };
689 struct lu_fid fid = ll_i2info(parent)->lli_fid;
691 /* If it is striped directory, get the real stripe parent */
692 if (unlikely(ll_i2info(parent)->lli_lsm_md != NULL)) {
693 rc = md_get_fid_from_lsm(ll_i2mdexp(parent),
694 ll_i2info(parent)->lli_lsm_md,
696 (*de)->d_name.len, &fid);
701 if (md_revalidate_lock(ll_i2mdexp(parent), &parent_it, &fid,
703 d_lustre_revalidate(*de);
704 ll_intent_release(&parent_it);
708 if (it_disposition(it, DISP_OPEN_CREATE)) {
709 ll_stats_ops_tally(ll_i2sbi(parent), LPROC_LL_MKNOD,
710 ktime_us_delta(ktime_get(), kstart));
716 if (rc != 0 && it->it_op & IT_OPEN) {
717 ll_intent_drop_lock(it);
718 ll_open_cleanup((*de)->d_sb, request);
724 static struct dentry *ll_lookup_it(struct inode *parent, struct dentry *dentry,
725 struct lookup_intent *it,
726 void **secctx, __u32 *secctxlen)
728 ktime_t kstart = ktime_get();
729 struct lookup_intent lookup_it = { .it_op = IT_LOOKUP };
730 struct dentry *save = dentry, *retval;
731 struct ptlrpc_request *req = NULL;
732 struct md_op_data *op_data = NULL;
737 if (dentry->d_name.len > ll_i2sbi(parent)->ll_namelen)
738 RETURN(ERR_PTR(-ENAMETOOLONG));
740 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), intent=%s\n",
741 dentry->d_name.len, dentry->d_name.name,
742 PFID(ll_inode2fid(parent)), parent, LL_IT2STR(it));
744 if (d_mountpoint(dentry))
745 CERROR("Tell Peter, lookup on mtpt, it %s\n", LL_IT2STR(it));
747 if (it == NULL || it->it_op == IT_GETXATTR)
750 if (it->it_op == IT_GETATTR && dentry_may_statahead(parent, dentry)) {
751 rc = ll_statahead(parent, &dentry, 0);
753 RETURN(dentry == save ? NULL : dentry);
756 if (it->it_op & IT_OPEN && it->it_flags & FMODE_WRITE &&
757 dentry->d_sb->s_flags & MS_RDONLY)
758 RETURN(ERR_PTR(-EROFS));
760 if (it->it_op & IT_CREAT)
761 opc = LUSTRE_OPC_CREATE;
763 opc = LUSTRE_OPC_ANY;
765 op_data = ll_prep_md_op_data(NULL, parent, NULL, dentry->d_name.name,
766 dentry->d_name.len, 0, opc, NULL);
768 GOTO(out, retval = ERR_CAST(op_data));
770 /* enforce umask if acl disabled or MDS doesn't support umask */
771 if (!IS_POSIXACL(parent) || !exp_connect_umask(ll_i2mdexp(parent)))
772 it->it_create_mode &= ~current_umask();
774 if (it->it_op & IT_CREAT &&
775 ll_i2sbi(parent)->ll_flags & LL_SBI_FILE_SECCTX) {
776 rc = ll_dentry_init_security(dentry, it->it_create_mode,
778 &op_data->op_file_secctx_name,
779 &op_data->op_file_secctx,
780 &op_data->op_file_secctx_size);
782 GOTO(out, retval = ERR_PTR(rc));
784 *secctx = op_data->op_file_secctx;
785 if (secctxlen != NULL)
786 *secctxlen = op_data->op_file_secctx_size;
789 rc = md_intent_lock(ll_i2mdexp(parent), op_data, it, &req,
790 &ll_md_blocking_ast, 0);
791 /* If the MDS allows the client to chgrp (CFS_SETGRP_PERM), but the
792 * client does not know which suppgid should be sent to the MDS, or
793 * some other(s) changed the target file's GID after this RPC sent
794 * to the MDS with the suppgid as the original GID, then we should
795 * try again with right suppgid. */
796 if (rc == -EACCES && it->it_op & IT_OPEN &&
797 it_disposition(it, DISP_OPEN_DENY)) {
798 struct mdt_body *body;
800 LASSERT(req != NULL);
802 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
803 if (op_data->op_suppgids[0] == body->mbo_gid ||
804 op_data->op_suppgids[1] == body->mbo_gid ||
805 !in_group_p(make_kgid(&init_user_ns, body->mbo_gid)))
806 GOTO(out, retval = ERR_PTR(-EACCES));
808 fid_zero(&op_data->op_fid2);
809 op_data->op_suppgids[1] = body->mbo_gid;
810 ptlrpc_req_finished(req);
812 ll_intent_release(it);
813 rc = md_intent_lock(ll_i2mdexp(parent), op_data, it, &req,
814 &ll_md_blocking_ast, 0);
818 GOTO(out, retval = ERR_PTR(rc));
820 /* dir layout may change */
821 ll_unlock_md_op_lsm(op_data);
822 rc = ll_lookup_it_finish(req, it, parent, &dentry, kstart);
824 ll_intent_release(it);
825 GOTO(out, retval = ERR_PTR(rc));
828 if ((it->it_op & IT_OPEN) && dentry->d_inode &&
829 !S_ISREG(dentry->d_inode->i_mode) &&
830 !S_ISDIR(dentry->d_inode->i_mode)) {
831 ll_release_openhandle(dentry, it);
833 ll_lookup_finish_locks(it, dentry);
835 GOTO(out, retval = (dentry == save) ? NULL : dentry);
838 if (op_data != NULL && !IS_ERR(op_data)) {
839 if (secctx != NULL && secctxlen != NULL) {
840 /* caller needs sec ctx info, so reset it in op_data to
841 * prevent it from being freed */
842 op_data->op_file_secctx = NULL;
843 op_data->op_file_secctx_size = 0;
845 ll_finish_md_op_data(op_data);
848 ptlrpc_req_finished(req);
852 #ifdef HAVE_IOP_ATOMIC_OPEN
853 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
856 struct lookup_intent *itp, it = { .it_op = IT_GETATTR };
859 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), flags=%u\n",
860 dentry->d_name.len, dentry->d_name.name,
861 PFID(ll_inode2fid(parent)), parent, flags);
864 * Optimize away (CREATE && !OPEN). Let .create handle the race.
865 * but only if we have write permissions there, otherwise we need
866 * to proceed with lookup. LU-4185
868 if ((flags & LOOKUP_CREATE) && !(flags & LOOKUP_OPEN) &&
869 (inode_permission(parent, MAY_WRITE | MAY_EXEC) == 0))
872 if (flags & (LOOKUP_PARENT|LOOKUP_OPEN|LOOKUP_CREATE))
876 de = ll_lookup_it(parent, dentry, itp, NULL, NULL);
879 ll_intent_release(itp);
884 #ifdef FMODE_CREATED /* added in Linux v4.18-rc1-20-g73a09dd */
885 # define ll_is_opened(o, f) ((f)->f_mode & FMODE_OPENED)
886 # define ll_finish_open(f, d, o) finish_open((f), (d), NULL)
888 # define ll_set_created(o, f) \
890 (f)->f_mode |= FMODE_CREATED; \
894 # define ll_is_opened(o, f) (*(o))
895 # define ll_finish_open(f, d, o) finish_open((f), (d), NULL, (o))
896 # define ll_last_arg , int *opened
897 # define ll_set_created(o, f) \
899 *(o) |= FILE_CREATED; \
905 * For cached negative dentry and new dentry, handle lookup/create/open
908 static int ll_atomic_open(struct inode *dir, struct dentry *dentry,
909 struct file *file, unsigned open_flags,
910 umode_t mode ll_last_arg)
912 struct lookup_intent *it;
914 long long lookup_flags = LOOKUP_OPEN;
920 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), file %p,"
921 "open_flags %x, mode %x opened %d\n",
922 dentry->d_name.len, dentry->d_name.name,
923 PFID(ll_inode2fid(dir)), dir, file, open_flags, mode,
924 ll_is_opened(opened, file));
926 /* Only negative dentries enter here */
927 LASSERT(dentry->d_inode == NULL);
929 if (!d_unhashed(dentry)) {
930 /* A valid negative dentry that just passed revalidation,
931 * there's little point to try and open it server-side,
932 * even though there's a minuscule chance it might succeed.
933 * Either way it's a valid race to just return -ENOENT here.
935 if (!(open_flags & O_CREAT))
938 /* Otherwise we just unhash it to be rehashed afresh via
939 * lookup if necessary
944 OBD_ALLOC(it, sizeof(*it));
949 if (open_flags & O_CREAT) {
950 it->it_op |= IT_CREAT;
951 lookup_flags |= LOOKUP_CREATE;
953 it->it_create_mode = (mode & S_IALLUGO) | S_IFREG;
954 it->it_flags = (open_flags & ~O_ACCMODE) | OPEN_FMODE(open_flags);
955 it->it_flags &= ~MDS_OPEN_FL_INTERNAL;
957 /* Dentry added to dcache tree in ll_lookup_it */
958 de = ll_lookup_it(dir, dentry, it, &secctx, &secctxlen);
964 CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_FILE_PAUSE, cfs_fail_val);
967 if (it_disposition(it, DISP_OPEN_CREATE)) {
968 /* Dentry instantiated in ll_create_it. */
969 rc = ll_create_it(dir, dentry, it, secctx, secctxlen);
970 security_release_secctx(secctx, secctxlen);
972 /* We dget in ll_splice_alias. */
977 ll_set_created(opened, file);
979 if (dentry->d_inode && it_disposition(it, DISP_OPEN_OPEN)) {
981 if (S_ISFIFO(dentry->d_inode->i_mode)) {
982 /* We cannot call open here as it might
983 * deadlock. This case is unreachable in
984 * practice because of OBD_CONNECT_NODEVOH. */
985 rc = finish_no_open(file, de);
987 file->private_data = it;
988 rc = ll_finish_open(file, dentry, opened);
989 /* We dget in ll_splice_alias. finish_open takes
990 * care of dget for fd open.
996 rc = finish_no_open(file, de);
1001 ll_intent_release(it);
1002 OBD_FREE(it, sizeof(*it));
1007 #else /* !HAVE_IOP_ATOMIC_OPEN */
1008 static struct lookup_intent *
1009 ll_convert_intent(struct open_intent *oit, int lookup_flags, bool is_readonly)
1011 struct lookup_intent *it;
1015 return ERR_PTR(-ENOMEM);
1017 if (lookup_flags & LOOKUP_OPEN) {
1018 it->it_op = IT_OPEN;
1019 /* Avoid file creation for ro bind mount point(is_readonly) */
1020 if ((lookup_flags & LOOKUP_CREATE) && !is_readonly)
1021 it->it_op |= IT_CREAT;
1022 it->it_create_mode = (oit->create_mode & S_IALLUGO) | S_IFREG;
1023 it->it_flags = ll_namei_to_lookup_intent_flag(oit->flags &
1024 ~(is_readonly ? O_CREAT : 0));
1025 it->it_flags &= ~MDS_OPEN_FL_INTERNAL;
1027 it->it_op = IT_GETATTR;
1033 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
1034 struct nameidata *nd)
1039 if (nd && !(nd->flags & (LOOKUP_CONTINUE|LOOKUP_PARENT))) {
1040 struct lookup_intent *it;
1042 if (ll_d2d(dentry) && ll_d2d(dentry)->lld_it) {
1043 it = ll_d2d(dentry)->lld_it;
1044 ll_d2d(dentry)->lld_it = NULL;
1047 * Optimize away (CREATE && !OPEN). Let .create handle
1048 * the race. But only if we have write permissions
1049 * there, otherwise we need to proceed with lookup.
1052 if ((nd->flags & LOOKUP_CREATE) &&
1053 !(nd->flags & LOOKUP_OPEN) &&
1054 (inode_permission(parent,
1055 MAY_WRITE | MAY_EXEC) == 0))
1058 it = ll_convert_intent(&nd->intent.open, nd->flags,
1059 (nd->path.mnt->mnt_flags & MNT_READONLY) ||
1060 (nd->path.mnt->mnt_sb->s_flags & MS_RDONLY));
1062 RETURN((struct dentry *)it);
1065 de = ll_lookup_it(parent, dentry, it, NULL, NULL);
1068 if ((nd->flags & LOOKUP_OPEN) && !IS_ERR(dentry)) { /* Open */
1069 if (dentry->d_inode &&
1070 it_disposition(it, DISP_OPEN_OPEN)) { /* nocreate */
1071 if (S_ISFIFO(dentry->d_inode->i_mode)) {
1072 /* We cannot call open here as it might
1073 * deadlock. This case is unreachable in
1074 * practice because of
1075 * OBD_CONNECT_NODEVOH. */
1079 nd->intent.open.file->private_data = it;
1080 filp = lookup_instantiate_filp(nd,
1086 de = (struct dentry *)filp;
1089 } else if (it_disposition(it, DISP_OPEN_CREATE)) {
1090 /* XXX This can only reliably work on assumption
1091 * that there are NO hashed negative dentries.*/
1092 ll_d2d(dentry)->lld_it = it;
1093 it = NULL; /* Will be freed in ll_create_nd */
1094 /* We absolutely depend on ll_create_nd to be
1095 * called to not leak this intent and possible
1096 * data attached to it */
1101 ll_intent_release(it);
1102 OBD_FREE(it, sizeof(*it));
1105 de = ll_lookup_it(parent, dentry, NULL, NULL, NULL);
1110 #endif /* HAVE_IOP_ATOMIC_OPEN */
1112 /* We depend on "mode" being set with the proper file type/umask by now */
1113 static struct inode *ll_create_node(struct inode *dir, struct lookup_intent *it)
1115 struct inode *inode = NULL;
1116 struct ptlrpc_request *request = NULL;
1117 struct ll_sb_info *sbi = ll_i2sbi(dir);
1121 LASSERT(it && it->it_disposition);
1123 LASSERT(it_disposition(it, DISP_ENQ_CREATE_REF));
1124 request = it->it_request;
1125 it_clear_disposition(it, DISP_ENQ_CREATE_REF);
1126 rc = ll_prep_inode(&inode, request, dir->i_sb, it);
1128 GOTO(out, inode = ERR_PTR(rc));
1130 /* Pause to allow for a race with concurrent access by fid */
1131 OBD_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_NODE_PAUSE, cfs_fail_val);
1133 /* We asked for a lock on the directory, but were granted a
1134 * lock on the inode. Since we finally have an inode pointer,
1135 * stuff it in the lock. */
1136 CDEBUG(D_DLMTRACE, "setting l_ast_data to inode "DFID"(%p)\n",
1137 PFID(ll_inode2fid(inode)), inode);
1138 ll_set_lock_data(sbi->ll_md_exp, inode, it, NULL);
1141 ptlrpc_req_finished(request);
1146 * By the time this is called, we already have created the directory cache
1147 * entry for the new file, but it is so far negative - it has no inode.
1149 * We defer creating the OBD object(s) until open, to keep the intent and
1150 * non-intent code paths similar, and also because we do not have the MDS
1151 * inode number before calling ll_create_node() (which is needed for LOV),
1152 * so we would need to do yet another RPC to the MDS to store the LOV EA
1153 * data on the MDS. If needed, we would pass the PACKED lmm as data and
1154 * lmm_size in datalen (the MDS still has code which will handle that).
1156 * If the create succeeds, we fill in the inode information
1157 * with d_instantiate().
1159 static int ll_create_it(struct inode *dir, struct dentry *dentry,
1160 struct lookup_intent *it,
1161 void *secctx, __u32 secctxlen)
1163 struct inode *inode;
1168 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), intent=%s\n",
1169 dentry->d_name.len, dentry->d_name.name,
1170 PFID(ll_inode2fid(dir)), dir, LL_IT2STR(it));
1172 rc = it_open_error(DISP_OPEN_CREATE, it);
1176 inode = ll_create_node(dir, it);
1178 RETURN(PTR_ERR(inode));
1180 if ((ll_i2sbi(inode)->ll_flags & LL_SBI_FILE_SECCTX) &&
1183 /* must be done before d_instantiate, because it calls
1184 * security_d_instantiate, which means a getxattr if security
1185 * context is not set yet */
1186 rc = security_inode_notifysecctx(inode, secctx, secctxlen);
1187 inode_unlock(inode);
1192 d_instantiate(dentry, inode);
1194 if (!(ll_i2sbi(inode)->ll_flags & LL_SBI_FILE_SECCTX)) {
1195 rc = ll_inode_init_security(dentry, inode, dir);
1200 ll_set_lock_data(ll_i2sbi(dir)->ll_md_exp, inode, it, &bits);
1201 if (bits & MDS_INODELOCK_LOOKUP)
1202 d_lustre_revalidate(dentry);
1207 void ll_update_times(struct ptlrpc_request *request, struct inode *inode)
1209 struct mdt_body *body = req_capsule_server_get(&request->rq_pill,
1213 if (body->mbo_valid & OBD_MD_FLMTIME &&
1214 body->mbo_mtime > inode->i_mtime.tv_sec) {
1216 "setting fid " DFID " mtime from %lld to %llu\n",
1217 PFID(ll_inode2fid(inode)),
1218 (s64)inode->i_mtime.tv_sec, body->mbo_mtime);
1219 inode->i_mtime.tv_sec = body->mbo_mtime;
1222 if (body->mbo_valid & OBD_MD_FLCTIME &&
1223 body->mbo_ctime > inode->i_ctime.tv_sec)
1224 inode->i_ctime.tv_sec = body->mbo_ctime;
1227 static int ll_new_node(struct inode *dir, struct dentry *dchild,
1228 const char *tgt, umode_t mode, int rdev, __u32 opc)
1230 struct qstr *name = &dchild->d_name;
1231 struct ptlrpc_request *request = NULL;
1232 struct md_op_data *op_data;
1233 struct inode *inode = NULL;
1234 struct ll_sb_info *sbi = ll_i2sbi(dir);
1239 if (unlikely(tgt != NULL))
1240 tgt_len = strlen(tgt) + 1;
1243 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
1244 name->len, 0, opc, NULL);
1245 if (IS_ERR(op_data))
1246 GOTO(err_exit, err = PTR_ERR(op_data));
1248 if (sbi->ll_flags & LL_SBI_FILE_SECCTX) {
1249 err = ll_dentry_init_security(dchild, mode, &dchild->d_name,
1250 &op_data->op_file_secctx_name,
1251 &op_data->op_file_secctx,
1252 &op_data->op_file_secctx_size);
1254 GOTO(err_exit, err);
1257 err = md_create(sbi->ll_md_exp, op_data, tgt, tgt_len, mode,
1258 from_kuid(&init_user_ns, current_fsuid()),
1259 from_kgid(&init_user_ns, current_fsgid()),
1260 cfs_curproc_cap_pack(), rdev, &request);
1261 if (err < 0 && err != -EREMOTE)
1262 GOTO(err_exit, err);
1264 /* If the client doesn't know where to create a subdirectory (or
1265 * in case of a race that sends the RPC to the wrong MDS), the
1266 * MDS will return -EREMOTE and the client will fetch the layout
1267 * of the directory, then create the directory on the right MDT. */
1268 if (unlikely(err == -EREMOTE)) {
1269 struct ll_inode_info *lli = ll_i2info(dir);
1270 struct lmv_user_md *lum;
1274 ptlrpc_req_finished(request);
1277 err2 = ll_dir_getstripe(dir, (void **)&lum, &lumsize, &request,
1278 OBD_MD_DEFAULT_MEA);
1280 /* Update stripe_offset and retry */
1281 lli->lli_def_stripe_offset = lum->lum_stripe_offset;
1282 } else if (err2 == -ENODATA &&
1283 lli->lli_def_stripe_offset != -1) {
1284 /* If there are no default stripe EA on the MDT, but the
1285 * client has default stripe, then it probably means
1286 * default stripe EA has just been deleted. */
1287 lli->lli_def_stripe_offset = -1;
1289 GOTO(err_exit, err);
1292 ptlrpc_req_finished(request);
1294 ll_finish_md_op_data(op_data);
1298 ll_update_times(request, dir);
1300 CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_NEWNODE_PAUSE, cfs_fail_val);
1302 err = ll_prep_inode(&inode, request, dchild->d_sb, NULL);
1304 GOTO(err_exit, err);
1306 if (sbi->ll_flags & LL_SBI_FILE_SECCTX) {
1308 /* must be done before d_instantiate, because it calls
1309 * security_d_instantiate, which means a getxattr if security
1310 * context is not set yet */
1311 err = security_inode_notifysecctx(inode,
1312 op_data->op_file_secctx,
1313 op_data->op_file_secctx_size);
1314 inode_unlock(inode);
1316 GOTO(err_exit, err);
1319 d_instantiate(dchild, inode);
1321 if (!(sbi->ll_flags & LL_SBI_FILE_SECCTX)) {
1322 err = ll_inode_init_security(dchild, inode, dir);
1324 GOTO(err_exit, err);
1329 if (request != NULL)
1330 ptlrpc_req_finished(request);
1332 if (!IS_ERR_OR_NULL(op_data))
1333 ll_finish_md_op_data(op_data);
1338 static int ll_mknod(struct inode *dir, struct dentry *dchild, ll_umode_t mode,
1341 struct qstr *name = &dchild->d_name;
1345 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p) mode %o dev %x\n",
1346 name->len, name->name, PFID(ll_inode2fid(dir)), dir,
1349 if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
1350 mode &= ~current_umask();
1352 switch (mode & S_IFMT) {
1354 mode |= S_IFREG; /* for mode = 0 case, fallthrough */
1360 err = ll_new_node(dir, dchild, NULL, mode, old_encode_dev(rdev),
1371 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKNOD, 1);
1376 #ifdef HAVE_IOP_ATOMIC_OPEN
1378 * Plain create. Intent create is handled in atomic_open.
1380 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
1381 umode_t mode, bool want_excl)
1385 CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_FILE_PAUSE, cfs_fail_val);
1387 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), "
1388 "flags=%u, excl=%d\n", dentry->d_name.len,
1389 dentry->d_name.name, PFID(ll_inode2fid(dir)),
1390 dir, mode, want_excl);
1392 /* Using mknod(2) to create a regular file is designed to not recognize
1393 * volatile file name, so we use ll_mknod() here. */
1394 rc = ll_mknod(dir, dentry, mode, 0);
1396 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_CREATE, 1);
1398 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, unhashed %d\n",
1399 dentry->d_name.len, dentry->d_name.name, d_unhashed(dentry));
1403 #else /* !HAVE_IOP_ATOMIC_OPEN */
1404 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
1405 ll_umode_t mode, struct nameidata *nd)
1407 struct ll_dentry_data *lld = ll_d2d(dentry);
1408 struct lookup_intent *it = NULL;
1411 CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_FILE_PAUSE, cfs_fail_val);
1417 /* LU-8559: use LUSTRE_OPC_CREATE for non atomic open case
1418 * so that volatile file name is recoginized.
1419 * Mknod(2), however, is designed to not recognize volatile
1420 * file name to avoid inode leak under orphan directory until
1422 return ll_new_node(dir, dentry, NULL, mode, 0,
1428 /* Was there an error? Propagate it! */
1429 if (it->it_status) {
1434 rc = ll_create_it(dir, dentry, it, NULL, 0);
1435 if (nd && (nd->flags & LOOKUP_OPEN) && dentry->d_inode) { /* Open */
1438 nd->intent.open.file->private_data = it;
1439 filp = lookup_instantiate_filp(nd, dentry, NULL);
1445 ll_intent_release(it);
1446 OBD_FREE(it, sizeof(*it));
1449 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_CREATE, 1);
1453 #endif /* HAVE_IOP_ATOMIC_OPEN */
1455 static int ll_symlink(struct inode *dir, struct dentry *dchild,
1456 const char *oldpath)
1458 struct qstr *name = &dchild->d_name;
1462 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), target=%.*s\n",
1463 name->len, name->name, PFID(ll_inode2fid(dir)),
1464 dir, 3000, oldpath);
1466 err = ll_new_node(dir, dchild, oldpath, S_IFLNK | S_IRWXUGO, 0,
1467 LUSTRE_OPC_SYMLINK);
1470 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_SYMLINK, 1);
1475 static int ll_link(struct dentry *old_dentry, struct inode *dir,
1476 struct dentry *new_dentry)
1478 struct inode *src = old_dentry->d_inode;
1479 struct qstr *name = &new_dentry->d_name;
1480 struct ll_sb_info *sbi = ll_i2sbi(dir);
1481 struct ptlrpc_request *request = NULL;
1482 struct md_op_data *op_data;
1486 CDEBUG(D_VFSTRACE, "VFS Op: inode="DFID"(%p), dir="DFID"(%p), "
1487 "target=%.*s\n", PFID(ll_inode2fid(src)), src,
1488 PFID(ll_inode2fid(dir)), dir, name->len, name->name);
1490 op_data = ll_prep_md_op_data(NULL, src, dir, name->name, name->len,
1491 0, LUSTRE_OPC_ANY, NULL);
1492 if (IS_ERR(op_data))
1493 RETURN(PTR_ERR(op_data));
1495 err = md_link(sbi->ll_md_exp, op_data, &request);
1496 ll_finish_md_op_data(op_data);
1500 ll_update_times(request, dir);
1501 ll_stats_ops_tally(sbi, LPROC_LL_LINK, 1);
1504 ptlrpc_req_finished(request);
1508 static int ll_mkdir(struct inode *dir, struct dentry *dchild, ll_umode_t mode)
1510 struct qstr *name = &dchild->d_name;
1514 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1515 name->len, name->name, PFID(ll_inode2fid(dir)), dir);
1517 if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
1518 mode &= ~current_umask();
1520 mode = (mode & (S_IRWXUGO|S_ISVTX)) | S_IFDIR;
1522 err = ll_new_node(dir, dchild, NULL, mode, 0, LUSTRE_OPC_MKDIR);
1524 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKDIR, 1);
1529 static int ll_rmdir(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;
1537 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1538 name->len, name->name, PFID(ll_inode2fid(dir)), dir);
1540 if (unlikely(d_mountpoint(dchild)))
1543 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len,
1544 S_IFDIR, LUSTRE_OPC_ANY, NULL);
1545 if (IS_ERR(op_data))
1546 RETURN(PTR_ERR(op_data));
1548 if (dchild->d_inode != NULL)
1549 op_data->op_fid3 = *ll_inode2fid(dchild->d_inode);
1551 op_data->op_fid2 = op_data->op_fid3;
1552 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1553 ll_finish_md_op_data(op_data);
1555 ll_update_times(request, dir);
1556 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR, 1);
1559 ptlrpc_req_finished(request);
1566 int ll_rmdir_entry(struct inode *dir, char *name, int namelen)
1568 struct ptlrpc_request *request = NULL;
1569 struct md_op_data *op_data;
1573 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1574 namelen, name, PFID(ll_inode2fid(dir)), dir);
1576 op_data = ll_prep_md_op_data(NULL, dir, NULL, name, strlen(name),
1577 S_IFDIR, LUSTRE_OPC_ANY, NULL);
1578 if (IS_ERR(op_data))
1579 RETURN(PTR_ERR(op_data));
1580 op_data->op_cli_flags |= CLI_RM_ENTRY;
1581 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1582 ll_finish_md_op_data(op_data);
1584 ll_update_times(request, dir);
1585 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR, 1);
1588 ptlrpc_req_finished(request);
1592 static int ll_unlink(struct inode *dir, struct dentry *dchild)
1594 struct qstr *name = &dchild->d_name;
1595 struct ptlrpc_request *request = NULL;
1596 struct md_op_data *op_data;
1597 struct mdt_body *body;
1600 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1601 name->len, name->name, PFID(ll_inode2fid(dir)), dir);
1604 * XXX: unlink bind mountpoint maybe call to here,
1605 * just check it as vfs_unlink does.
1607 if (unlikely(d_mountpoint(dchild)))
1610 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len, 0,
1611 LUSTRE_OPC_ANY, NULL);
1612 if (IS_ERR(op_data))
1613 RETURN(PTR_ERR(op_data));
1615 op_data->op_fid3 = *ll_inode2fid(dchild->d_inode);
1617 op_data->op_fid2 = op_data->op_fid3;
1618 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1619 ll_finish_md_op_data(op_data);
1624 * The server puts attributes in on the last unlink, use them to update
1625 * the link count so the inode can be freed immediately.
1627 body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
1628 if (body->mbo_valid & OBD_MD_FLNLINK)
1629 set_nlink(dchild->d_inode, body->mbo_nlink);
1631 ll_update_times(request, dir);
1632 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_UNLINK, 1);
1635 ptlrpc_req_finished(request);
1639 static int ll_rename(struct inode *src, struct dentry *src_dchild,
1640 struct inode *tgt, struct dentry *tgt_dchild
1641 #ifdef HAVE_IOPS_RENAME_WITH_FLAGS
1642 , unsigned int flags
1646 struct qstr *src_name = &src_dchild->d_name;
1647 struct qstr *tgt_name = &tgt_dchild->d_name;
1648 struct ptlrpc_request *request = NULL;
1649 struct ll_sb_info *sbi = ll_i2sbi(src);
1650 struct md_op_data *op_data;
1654 #ifdef HAVE_IOPS_RENAME_WITH_FLAGS
1659 CDEBUG(D_VFSTRACE, "VFS Op:oldname=%.*s, src_dir="DFID
1660 "(%p), newname=%.*s, tgt_dir="DFID"(%p)\n",
1661 src_name->len, src_name->name,
1662 PFID(ll_inode2fid(src)), src, tgt_name->len,
1663 tgt_name->name, PFID(ll_inode2fid(tgt)), tgt);
1665 if (unlikely(d_mountpoint(src_dchild) || d_mountpoint(tgt_dchild)))
1668 op_data = ll_prep_md_op_data(NULL, src, tgt, NULL, 0, 0,
1669 LUSTRE_OPC_ANY, NULL);
1670 if (IS_ERR(op_data))
1671 RETURN(PTR_ERR(op_data));
1673 if (src_dchild->d_inode != NULL)
1674 op_data->op_fid3 = *ll_inode2fid(src_dchild->d_inode);
1676 if (tgt_dchild->d_inode != NULL)
1677 op_data->op_fid4 = *ll_inode2fid(tgt_dchild->d_inode);
1679 err = md_rename(sbi->ll_md_exp, op_data,
1680 src_name->name, src_name->len,
1681 tgt_name->name, tgt_name->len, &request);
1682 ll_finish_md_op_data(op_data);
1684 ll_update_times(request, src);
1685 ll_update_times(request, tgt);
1686 ll_stats_ops_tally(sbi, LPROC_LL_RENAME, 1);
1689 ptlrpc_req_finished(request);
1692 d_move(src_dchild, tgt_dchild);
1697 const struct inode_operations ll_dir_inode_operations = {
1699 #ifdef HAVE_IOP_ATOMIC_OPEN
1700 .atomic_open = ll_atomic_open,
1702 .lookup = ll_lookup_nd,
1703 .create = ll_create_nd,
1704 /* We need all these non-raw things for NFSD, to not patch it. */
1705 .unlink = ll_unlink,
1708 .symlink = ll_symlink,
1710 .rename = ll_rename,
1711 .setattr = ll_setattr,
1712 .getattr = ll_getattr,
1713 .permission = ll_inode_permission,
1714 #ifdef HAVE_IOP_XATTR
1715 .setxattr = ll_setxattr,
1716 .getxattr = ll_getxattr,
1717 .removexattr = ll_removexattr,
1719 .listxattr = ll_listxattr,
1720 #ifdef HAVE_IOP_GET_ACL
1721 .get_acl = ll_get_acl,
1723 #ifdef HAVE_IOP_SET_ACL
1724 .set_acl = ll_set_acl,
1728 const struct inode_operations ll_special_inode_operations = {
1729 .setattr = ll_setattr,
1730 .getattr = ll_getattr,
1731 .permission = ll_inode_permission,
1732 #ifdef HAVE_IOP_XATTR
1733 .setxattr = ll_setxattr,
1734 .getxattr = ll_getxattr,
1735 .removexattr = ll_removexattr,
1737 .listxattr = ll_listxattr,
1738 #ifdef HAVE_IOP_GET_ACL
1739 .get_acl = ll_get_acl,
1741 #ifdef HAVE_IOP_SET_ACL
1742 .set_acl = ll_set_acl,