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 static int ll_dom_lock_cancel(struct inode *inode, struct ldlm_lock *lock)
190 struct ll_inode_info *lli = ll_i2info(inode);
195 env = cl_env_get(&refcheck);
197 RETURN(PTR_ERR(env));
199 /* reach MDC layer to flush data under the DoM ldlm lock */
200 rc = cl_object_flush(env, lli->lli_clob, lock);
201 if (rc == -ENODATA) {
202 CDEBUG(D_INODE, "inode "DFID" layout has no DoM stripe\n",
203 PFID(ll_inode2fid(inode)));
204 /* most likely result of layout change, do nothing */
208 cl_env_put(env, &refcheck);
212 static void ll_lock_cancel_bits(struct ldlm_lock *lock, __u64 to_cancel)
214 struct inode *inode = ll_inode_from_resource_lock(lock);
215 __u64 bits = to_cancel;
221 /* That means the inode is evicted most likely and may cause
222 * the skipping of lock cleanups below, so print the message
225 if (lock->l_resource->lr_lvb_inode)
227 "can't take inode for the lock (%sevicted)\n",
228 lock->l_resource->lr_lvb_inode->i_state &
229 I_FREEING ? "" : "not ");
233 if (!fid_res_name_eq(ll_inode2fid(inode),
234 &lock->l_resource->lr_name)) {
235 LDLM_ERROR(lock, "data mismatch with object "DFID"(%p)",
236 PFID(ll_inode2fid(inode)), inode);
240 if (bits & MDS_INODELOCK_XATTR) {
241 if (S_ISDIR(inode->i_mode))
242 ll_i2info(inode)->lli_def_stripe_offset = -1;
243 ll_xattr_cache_destroy(inode);
244 bits &= ~MDS_INODELOCK_XATTR;
247 /* For OPEN locks we differentiate between lock modes
248 * LCK_CR, LCK_CW, LCK_PR - bug 22891 */
249 if (bits & MDS_INODELOCK_OPEN)
250 ll_have_md_lock(inode, &bits, lock->l_req_mode);
252 if (bits & MDS_INODELOCK_OPEN) {
255 switch (lock->l_req_mode) {
266 LDLM_ERROR(lock, "bad lock mode for OPEN lock");
270 ll_md_real_close(inode, fmode);
272 bits &= ~MDS_INODELOCK_OPEN;
275 if (bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
276 MDS_INODELOCK_LAYOUT | MDS_INODELOCK_PERM |
278 ll_have_md_lock(inode, &bits, LCK_MINMODE);
280 if (bits & MDS_INODELOCK_DOM) {
281 rc = ll_dom_lock_cancel(inode, lock);
283 CDEBUG(D_INODE, "cannot flush DoM data "
285 PFID(ll_inode2fid(inode)), rc);
288 if (bits & MDS_INODELOCK_LAYOUT) {
289 struct cl_object_conf conf = {
290 .coc_opc = OBJECT_CONF_INVALIDATE,
294 rc = ll_layout_conf(inode, &conf);
296 CDEBUG(D_INODE, "cannot invalidate layout of "
298 PFID(ll_inode2fid(inode)), rc);
301 if (bits & MDS_INODELOCK_UPDATE) {
302 struct ll_inode_info *lli = ll_i2info(inode);
304 lli->lli_update_atime = 1;
307 if ((bits & MDS_INODELOCK_UPDATE) && S_ISDIR(inode->i_mode)) {
308 struct ll_inode_info *lli = ll_i2info(inode);
310 CDEBUG(D_INODE, "invalidating inode "DFID" lli = %p, "
311 "pfid = "DFID"\n", PFID(ll_inode2fid(inode)),
312 lli, PFID(&lli->lli_pfid));
313 truncate_inode_pages(inode->i_mapping, 0);
315 if (unlikely(!fid_is_zero(&lli->lli_pfid))) {
316 struct inode *master_inode = NULL;
319 /* This is slave inode, since all of the child dentry
320 * is connected on the master inode, so we have to
321 * invalidate the negative children on master inode */
322 CDEBUG(D_INODE, "Invalidate s"DFID" m"DFID"\n",
323 PFID(ll_inode2fid(inode)), PFID(&lli->lli_pfid));
325 hash = cl_fid_build_ino(&lli->lli_pfid,
326 ll_need_32bit_api(ll_i2sbi(inode)));
328 /* Do not lookup the inode with ilookup5, otherwise
329 * it will cause dead lock,
330 * 1. Client1 send chmod req to the MDT0, then on MDT0,
331 * it enqueues master and all of its slaves lock,
332 * (mdt_attr_set() -> mdt_lock_slaves()), after gets
333 * master and stripe0 lock, it will send the enqueue
334 * req (for stripe1) to MDT1, then MDT1 finds the lock
335 * has been granted to client2. Then MDT1 sends blocking
337 * 2. At the same time, client2 tries to unlink
338 * the striped dir (rm -rf striped_dir), and during
339 * lookup, it will hold the master inode of the striped
340 * directory, whose inode state is NEW, then tries to
341 * revalidate all of its slaves, (ll_prep_inode()->
342 * ll_iget()->ll_read_inode2()-> ll_update_inode().).
343 * And it will be blocked on the server side because
345 * 3. Then the client get the blocking_ast req, cancel
346 * the lock, but being blocked if using ->ilookup5()),
347 * because master inode state is NEW. */
348 master_inode = ilookup5_nowait(inode->i_sb, hash,
349 ll_test_inode_by_fid,
350 (void *)&lli->lli_pfid);
352 ll_invalidate_negative_children(master_inode);
356 ll_invalidate_negative_children(inode);
360 if ((bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM)) &&
361 inode->i_sb->s_root != NULL &&
362 inode != inode->i_sb->s_root->d_inode)
363 ll_invalidate_aliases(inode);
365 if (bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM))
366 forget_all_cached_acls(inode);
372 /* Check if the given lock may be downgraded instead of canceling and
373 * that convert is really needed. */
374 int ll_md_need_convert(struct ldlm_lock *lock)
376 struct ldlm_namespace *ns = ldlm_lock_to_ns(lock);
378 __u64 wanted = lock->l_policy_data.l_inodebits.cancel_bits;
379 __u64 bits = lock->l_policy_data.l_inodebits.bits & ~wanted;
380 enum ldlm_mode mode = LCK_MINMODE;
382 if (!lock->l_conn_export ||
383 !exp_connect_lock_convert(lock->l_conn_export))
386 if (!wanted || !bits || ldlm_is_cancel(lock))
389 /* do not convert locks other than DOM for now */
390 if (!((bits | wanted) & MDS_INODELOCK_DOM))
393 /* We may have already remaining bits in some other lock so
394 * lock convert will leave us just extra lock for the same bit.
395 * Check if client has other lock with the same bits and the same
396 * or lower mode and don't convert if any.
398 switch (lock->l_req_mode) {
410 /* do not convert other modes */
414 /* is lock is too old to be converted? */
415 lock_res_and_lock(lock);
416 if (ktime_after(ktime_get(),
417 ktime_add(lock->l_last_used,
418 ktime_set(ns->ns_dirty_age_limit, 0)))) {
419 unlock_res_and_lock(lock);
422 unlock_res_and_lock(lock);
424 inode = ll_inode_from_resource_lock(lock);
425 ll_have_md_lock(inode, &bits, mode);
430 int ll_md_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *ld,
431 void *data, int flag)
433 struct lustre_handle lockh;
439 case LDLM_CB_BLOCKING:
441 __u64 cancel_flags = LCF_ASYNC;
443 /* if lock convert is not needed then still have to
444 * pass lock via ldlm_cli_convert() to keep all states
445 * correct, set cancel_bits to full lock bits to cause
446 * full cancel to happen.
448 if (!ll_md_need_convert(lock)) {
449 lock_res_and_lock(lock);
450 lock->l_policy_data.l_inodebits.cancel_bits =
451 lock->l_policy_data.l_inodebits.bits;
452 unlock_res_and_lock(lock);
454 rc = ldlm_cli_convert(lock, cancel_flags);
457 /* continue with cancel otherwise */
458 ldlm_lock2handle(lock, &lockh);
459 rc = ldlm_cli_cancel(&lockh, cancel_flags);
461 CDEBUG(D_INODE, "ldlm_cli_cancel: rc = %d\n", rc);
466 case LDLM_CB_CANCELING:
468 __u64 to_cancel = lock->l_policy_data.l_inodebits.bits;
470 /* Nothing to do for non-granted locks */
471 if (!ldlm_is_granted(lock))
474 /* If 'ld' is supplied then bits to be cancelled are passed
475 * implicitly by lock converting and cancel_bits from 'ld'
476 * should be used. Otherwise full cancel is being performed
477 * and lock inodebits are used.
479 * Note: we cannot rely on cancel_bits in lock itself at this
480 * moment because they can be changed by concurrent thread,
481 * so ldlm_cli_inodebits_convert() pass cancel bits implicitly
485 /* partial bits cancel allowed only during convert */
486 LASSERT(ldlm_is_converting(lock));
487 /* mask cancel bits by lock bits so only no any unused
488 * bits are passed to ll_lock_cancel_bits()
490 to_cancel &= ld->l_policy_data.l_inodebits.cancel_bits;
492 ll_lock_cancel_bits(lock, to_cancel);
502 __u32 ll_i2suppgid(struct inode *i)
504 if (in_group_p(i->i_gid))
505 return (__u32)from_kgid(&init_user_ns, i->i_gid);
507 return (__u32) __kgid_val(INVALID_GID);
510 /* Pack the required supplementary groups into the supplied groups array.
511 * If we don't need to use the groups from the target inode(s) then we
512 * instead pack one or more groups from the user's supplementary group
513 * array in case it might be useful. Not needed if doing an MDS-side upcall. */
514 void ll_i2gids(__u32 *suppgids, struct inode *i1, struct inode *i2)
517 LASSERT(suppgids != NULL);
519 suppgids[0] = ll_i2suppgid(i1);
522 suppgids[1] = ll_i2suppgid(i2);
528 * try to reuse three types of dentry:
529 * 1. unhashed alias, this one is unhashed by d_invalidate (but it may be valid
530 * by concurrent .revalidate).
531 * 2. INVALID alias (common case for no valid ldlm lock held, but this flag may
532 * be cleared by others calling d_lustre_revalidate).
533 * 3. DISCONNECTED alias.
535 static struct dentry *ll_find_alias(struct inode *inode, struct dentry *dentry)
537 struct dentry *alias, *discon_alias, *invalid_alias;
538 DECLARE_LL_D_HLIST_NODE_PTR(p);
540 if (ll_d_hlist_empty(&inode->i_dentry))
543 discon_alias = invalid_alias = NULL;
545 ll_lock_dcache(inode);
546 ll_d_hlist_for_each_entry(alias, p, &inode->i_dentry) {
547 LASSERT(alias != dentry);
549 spin_lock(&alias->d_lock);
550 if ((alias->d_flags & DCACHE_DISCONNECTED) &&
551 S_ISDIR(inode->i_mode))
552 /* LASSERT(last_discon == NULL); LU-405, bz 20055 */
553 discon_alias = alias;
554 else if (alias->d_parent == dentry->d_parent &&
555 alias->d_name.hash == dentry->d_name.hash &&
556 alias->d_name.len == dentry->d_name.len &&
557 memcmp(alias->d_name.name, dentry->d_name.name,
558 dentry->d_name.len) == 0)
559 invalid_alias = alias;
560 spin_unlock(&alias->d_lock);
565 alias = invalid_alias ?: discon_alias ?: NULL;
567 spin_lock(&alias->d_lock);
569 spin_unlock(&alias->d_lock);
571 ll_unlock_dcache(inode);
577 * Similar to d_splice_alias(), but lustre treats invalid alias
578 * similar to DCACHE_DISCONNECTED, and tries to use it anyway.
580 struct dentry *ll_splice_alias(struct inode *inode, struct dentry *de)
586 new = ll_find_alias(inode, de);
596 "Reuse dentry %p inode %p refc %d flags %#x\n",
597 new, new->d_inode, ll_d_count(new), new->d_flags);
605 CDEBUG(D_DENTRY, "Add dentry %p inode %p refc %d flags %#x\n",
606 de, de->d_inode, ll_d_count(de), de->d_flags);
610 static int ll_lookup_it_finish(struct ptlrpc_request *request,
611 struct lookup_intent *it,
612 struct inode *parent, struct dentry **de,
613 void *secctx, __u32 secctxlen, ktime_t kstart)
615 struct inode *inode = NULL;
618 struct dentry *alias;
621 /* NB 1 request reference will be taken away by ll_intent_lock()
623 CDEBUG(D_DENTRY, "it %p it_disposition %x\n", it,
625 if (!it_disposition(it, DISP_LOOKUP_NEG)) {
626 struct req_capsule *pill = &request->rq_pill;
627 struct mdt_body *body = req_capsule_server_get(pill,
630 rc = ll_prep_inode(&inode, request, (*de)->d_sb, it);
634 ll_set_lock_data(ll_i2sbi(parent)->ll_md_exp, inode, it, &bits);
635 /* OPEN can return data if lock has DoM+LAYOUT bits set */
636 if (it->it_op & IT_OPEN &&
637 bits & MDS_INODELOCK_DOM && bits & MDS_INODELOCK_LAYOUT)
638 ll_dom_finish_open(inode, request);
640 /* We used to query real size from OSTs here, but actually
641 * this is not needed. For stat() calls size would be updated
642 * from subsequent do_revalidate()->ll_inode_revalidate_it() in
644 * vfs_getattr_it->ll_getattr()->ll_inode_revalidate_it() in 2.6
645 * Everybody else who needs correct file size would call
646 * ll_glimpse_size or some equivalent themselves anyway.
650 /* If security context was returned by MDT, put it in
651 * inode now to save an extra getxattr from security hooks,
652 * and avoid deadlock.
654 if (body->mbo_valid & OBD_MD_SECCTX) {
655 secctx = req_capsule_server_get(pill, &RMF_FILE_SECCTX);
656 secctxlen = req_capsule_get_size(pill,
661 CDEBUG(D_SEC, "server returned security context"
663 PFID(ll_inode2fid(inode)));
666 if (secctx && secctxlen) {
668 rc = security_inode_notifysecctx(inode, secctx,
672 CWARN("cannot set security context for "
674 PFID(ll_inode2fid(inode)), rc);
678 /* Only hash *de if it is unhashed (new dentry).
679 * Atoimc_open may passin hashed dentries for open.
681 alias = ll_splice_alias(inode, *de);
683 GOTO(out, rc = PTR_ERR(alias));
687 if (!it_disposition(it, DISP_LOOKUP_NEG)) {
688 /* we have lookup look - unhide dentry */
689 if (bits & MDS_INODELOCK_LOOKUP)
690 d_lustre_revalidate(*de);
691 } else if (!it_disposition(it, DISP_OPEN_CREATE)) {
693 * If file was created on the server, the dentry is revalidated
694 * in ll_create_it if the lock allows for it.
696 /* Check that parent has UPDATE lock. */
697 struct lookup_intent parent_it = {
699 .it_lock_handle = 0 };
700 struct lu_fid fid = ll_i2info(parent)->lli_fid;
702 /* If it is striped directory, get the real stripe parent */
703 if (unlikely(ll_i2info(parent)->lli_lsm_md != NULL)) {
704 rc = md_get_fid_from_lsm(ll_i2mdexp(parent),
705 ll_i2info(parent)->lli_lsm_md,
707 (*de)->d_name.len, &fid);
712 if (md_revalidate_lock(ll_i2mdexp(parent), &parent_it, &fid,
714 d_lustre_revalidate(*de);
715 ll_intent_release(&parent_it);
719 if (it_disposition(it, DISP_OPEN_CREATE)) {
720 ll_stats_ops_tally(ll_i2sbi(parent), LPROC_LL_MKNOD,
721 ktime_us_delta(ktime_get(), kstart));
727 if (rc != 0 && it->it_op & IT_OPEN) {
728 ll_intent_drop_lock(it);
729 ll_open_cleanup((*de)->d_sb, request);
735 static struct dentry *ll_lookup_it(struct inode *parent, struct dentry *dentry,
736 struct lookup_intent *it,
737 void **secctx, __u32 *secctxlen)
739 ktime_t kstart = ktime_get();
740 struct lookup_intent lookup_it = { .it_op = IT_LOOKUP };
741 struct dentry *save = dentry, *retval;
742 struct ptlrpc_request *req = NULL;
743 struct md_op_data *op_data = NULL;
746 char secctx_name[XATTR_NAME_MAX + 1];
750 if (dentry->d_name.len > ll_i2sbi(parent)->ll_namelen)
751 RETURN(ERR_PTR(-ENAMETOOLONG));
753 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), intent=%s\n",
754 dentry->d_name.len, dentry->d_name.name,
755 PFID(ll_inode2fid(parent)), parent, LL_IT2STR(it));
757 if (d_mountpoint(dentry))
758 CERROR("Tell Peter, lookup on mtpt, it %s\n", LL_IT2STR(it));
760 if (it == NULL || it->it_op == IT_GETXATTR)
763 if (it->it_op == IT_GETATTR && dentry_may_statahead(parent, dentry)) {
764 rc = ll_statahead(parent, &dentry, 0);
766 RETURN(dentry == save ? NULL : dentry);
769 if (it->it_op & IT_OPEN && it->it_flags & FMODE_WRITE &&
770 dentry->d_sb->s_flags & SB_RDONLY)
771 RETURN(ERR_PTR(-EROFS));
773 if (it->it_op & IT_CREAT)
774 opc = LUSTRE_OPC_CREATE;
776 opc = LUSTRE_OPC_ANY;
778 op_data = ll_prep_md_op_data(NULL, parent, NULL, dentry->d_name.name,
779 dentry->d_name.len, 0, opc, NULL);
781 GOTO(out, retval = ERR_CAST(op_data));
783 /* enforce umask if acl disabled or MDS doesn't support umask */
784 if (!IS_POSIXACL(parent) || !exp_connect_umask(ll_i2mdexp(parent)))
785 it->it_create_mode &= ~current_umask();
787 if (it->it_op & IT_CREAT &&
788 ll_i2sbi(parent)->ll_flags & LL_SBI_FILE_SECCTX) {
789 rc = ll_dentry_init_security(dentry, it->it_create_mode,
791 &op_data->op_file_secctx_name,
792 &op_data->op_file_secctx,
793 &op_data->op_file_secctx_size);
795 GOTO(out, retval = ERR_PTR(rc));
797 *secctx = op_data->op_file_secctx;
799 *secctxlen = op_data->op_file_secctx_size;
807 /* ask for security context upon intent */
808 if (it->it_op & (IT_LOOKUP | IT_GETATTR | IT_OPEN)) {
809 /* get name of security xattr to request to server */
810 rc = ll_listsecurity(parent, secctx_name,
811 sizeof(secctx_name));
813 CDEBUG(D_SEC, "cannot get security xattr name for "
815 PFID(ll_inode2fid(parent)), rc);
817 op_data->op_file_secctx_name = secctx_name;
818 op_data->op_file_secctx_name_size = rc;
819 CDEBUG(D_SEC, "'%.*s' is security xattr for "DFID"\n",
820 rc, secctx_name, PFID(ll_inode2fid(parent)));
824 rc = md_intent_lock(ll_i2mdexp(parent), op_data, it, &req,
825 &ll_md_blocking_ast, 0);
826 /* If the MDS allows the client to chgrp (CFS_SETGRP_PERM), but the
827 * client does not know which suppgid should be sent to the MDS, or
828 * some other(s) changed the target file's GID after this RPC sent
829 * to the MDS with the suppgid as the original GID, then we should
830 * try again with right suppgid. */
831 if (rc == -EACCES && it->it_op & IT_OPEN &&
832 it_disposition(it, DISP_OPEN_DENY)) {
833 struct mdt_body *body;
835 LASSERT(req != NULL);
837 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
838 if (op_data->op_suppgids[0] == body->mbo_gid ||
839 op_data->op_suppgids[1] == body->mbo_gid ||
840 !in_group_p(make_kgid(&init_user_ns, body->mbo_gid)))
841 GOTO(out, retval = ERR_PTR(-EACCES));
843 fid_zero(&op_data->op_fid2);
844 op_data->op_suppgids[1] = body->mbo_gid;
845 ptlrpc_req_finished(req);
847 ll_intent_release(it);
848 rc = md_intent_lock(ll_i2mdexp(parent), op_data, it, &req,
849 &ll_md_blocking_ast, 0);
853 GOTO(out, retval = ERR_PTR(rc));
855 /* dir layout may change */
856 ll_unlock_md_op_lsm(op_data);
857 rc = ll_lookup_it_finish(req, it, parent, &dentry,
858 secctx ? *secctx : NULL,
859 secctxlen ? *secctxlen : 0, kstart);
861 ll_intent_release(it);
862 GOTO(out, retval = ERR_PTR(rc));
865 if ((it->it_op & IT_OPEN) && dentry->d_inode &&
866 !S_ISREG(dentry->d_inode->i_mode) &&
867 !S_ISDIR(dentry->d_inode->i_mode)) {
868 ll_release_openhandle(dentry, it);
870 ll_lookup_finish_locks(it, dentry);
872 GOTO(out, retval = (dentry == save) ? NULL : dentry);
875 if (op_data != NULL && !IS_ERR(op_data)) {
876 if (secctx && secctxlen) {
877 /* caller needs sec ctx info, so reset it in op_data to
878 * prevent it from being freed */
879 op_data->op_file_secctx = NULL;
880 op_data->op_file_secctx_size = 0;
882 ll_finish_md_op_data(op_data);
885 ptlrpc_req_finished(req);
889 #ifdef HAVE_IOP_ATOMIC_OPEN
890 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
893 struct lookup_intent *itp, it = { .it_op = IT_GETATTR };
896 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), flags=%u\n",
897 dentry->d_name.len, dentry->d_name.name,
898 PFID(ll_inode2fid(parent)), parent, flags);
901 * Optimize away (CREATE && !OPEN). Let .create handle the race.
902 * but only if we have write permissions there, otherwise we need
903 * to proceed with lookup. LU-4185
905 if ((flags & LOOKUP_CREATE) && !(flags & LOOKUP_OPEN) &&
906 (inode_permission(parent, MAY_WRITE | MAY_EXEC) == 0))
909 if (flags & (LOOKUP_PARENT|LOOKUP_OPEN|LOOKUP_CREATE))
913 de = ll_lookup_it(parent, dentry, itp, NULL, NULL);
916 ll_intent_release(itp);
921 #ifdef FMODE_CREATED /* added in Linux v4.18-rc1-20-g73a09dd */
922 # define ll_is_opened(o, f) ((f)->f_mode & FMODE_OPENED)
923 # define ll_finish_open(f, d, o) finish_open((f), (d), NULL)
925 # define ll_set_created(o, f) \
927 (f)->f_mode |= FMODE_CREATED; \
931 # define ll_is_opened(o, f) (*(o))
932 # define ll_finish_open(f, d, o) finish_open((f), (d), NULL, (o))
933 # define ll_last_arg , int *opened
934 # define ll_set_created(o, f) \
936 *(o) |= FILE_CREATED; \
942 * For cached negative dentry and new dentry, handle lookup/create/open
945 static int ll_atomic_open(struct inode *dir, struct dentry *dentry,
946 struct file *file, unsigned open_flags,
947 umode_t mode ll_last_arg)
949 struct lookup_intent *it;
951 long long lookup_flags = LOOKUP_OPEN;
957 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), file %p,"
958 "open_flags %x, mode %x opened %d\n",
959 dentry->d_name.len, dentry->d_name.name,
960 PFID(ll_inode2fid(dir)), dir, file, open_flags, mode,
961 ll_is_opened(opened, file));
963 /* Only negative dentries enter here */
964 LASSERT(dentry->d_inode == NULL);
966 if (!d_unhashed(dentry)) {
967 /* A valid negative dentry that just passed revalidation,
968 * there's little point to try and open it server-side,
969 * even though there's a minuscule chance it might succeed.
970 * Either way it's a valid race to just return -ENOENT here.
972 if (!(open_flags & O_CREAT))
975 /* Otherwise we just unhash it to be rehashed afresh via
976 * lookup if necessary
981 OBD_ALLOC(it, sizeof(*it));
986 if (open_flags & O_CREAT) {
987 it->it_op |= IT_CREAT;
988 lookup_flags |= LOOKUP_CREATE;
990 it->it_create_mode = (mode & S_IALLUGO) | S_IFREG;
991 it->it_flags = (open_flags & ~O_ACCMODE) | OPEN_FMODE(open_flags);
992 it->it_flags &= ~MDS_OPEN_FL_INTERNAL;
994 /* Dentry added to dcache tree in ll_lookup_it */
995 de = ll_lookup_it(dir, dentry, it, &secctx, &secctxlen);
1001 CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_FILE_PAUSE, cfs_fail_val);
1004 if (it_disposition(it, DISP_OPEN_CREATE)) {
1005 /* Dentry instantiated in ll_create_it. */
1006 rc = ll_create_it(dir, dentry, it, secctx, secctxlen);
1007 security_release_secctx(secctx, secctxlen);
1009 /* We dget in ll_splice_alias. */
1014 ll_set_created(opened, file);
1016 if (dentry->d_inode && it_disposition(it, DISP_OPEN_OPEN)) {
1018 if (S_ISFIFO(dentry->d_inode->i_mode)) {
1019 /* We cannot call open here as it might
1020 * deadlock. This case is unreachable in
1021 * practice because of OBD_CONNECT_NODEVOH. */
1022 rc = finish_no_open(file, de);
1024 file->private_data = it;
1025 rc = ll_finish_open(file, dentry, opened);
1026 /* We dget in ll_splice_alias. finish_open takes
1027 * care of dget for fd open.
1033 rc = finish_no_open(file, de);
1038 ll_intent_release(it);
1039 OBD_FREE(it, sizeof(*it));
1044 #else /* !HAVE_IOP_ATOMIC_OPEN */
1045 static struct lookup_intent *
1046 ll_convert_intent(struct open_intent *oit, int lookup_flags, bool is_readonly)
1048 struct lookup_intent *it;
1052 return ERR_PTR(-ENOMEM);
1054 if (lookup_flags & LOOKUP_OPEN) {
1055 it->it_op = IT_OPEN;
1056 /* Avoid file creation for ro bind mount point(is_readonly) */
1057 if ((lookup_flags & LOOKUP_CREATE) && !is_readonly)
1058 it->it_op |= IT_CREAT;
1059 it->it_create_mode = (oit->create_mode & S_IALLUGO) | S_IFREG;
1060 it->it_flags = ll_namei_to_lookup_intent_flag(oit->flags &
1061 ~(is_readonly ? O_CREAT : 0));
1062 it->it_flags &= ~MDS_OPEN_FL_INTERNAL;
1064 it->it_op = IT_GETATTR;
1070 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
1071 struct nameidata *nd)
1076 if (nd && !(nd->flags & (LOOKUP_CONTINUE|LOOKUP_PARENT))) {
1077 struct lookup_intent *it;
1079 if (ll_d2d(dentry) && ll_d2d(dentry)->lld_it) {
1080 it = ll_d2d(dentry)->lld_it;
1081 ll_d2d(dentry)->lld_it = NULL;
1084 * Optimize away (CREATE && !OPEN). Let .create handle
1085 * the race. But only if we have write permissions
1086 * there, otherwise we need to proceed with lookup.
1089 if ((nd->flags & LOOKUP_CREATE) &&
1090 !(nd->flags & LOOKUP_OPEN) &&
1091 (inode_permission(parent,
1092 MAY_WRITE | MAY_EXEC) == 0))
1095 it = ll_convert_intent(&nd->intent.open, nd->flags,
1096 (nd->path.mnt->mnt_flags & MNT_READONLY) ||
1097 (nd->path.mnt->mnt_sb->s_flags & SB_RDONLY));
1099 RETURN((struct dentry *)it);
1102 de = ll_lookup_it(parent, dentry, it, NULL, NULL);
1105 if ((nd->flags & LOOKUP_OPEN) && !IS_ERR(dentry)) { /* Open */
1106 if (dentry->d_inode &&
1107 it_disposition(it, DISP_OPEN_OPEN)) { /* nocreate */
1108 if (S_ISFIFO(dentry->d_inode->i_mode)) {
1109 /* We cannot call open here as it might
1110 * deadlock. This case is unreachable in
1111 * practice because of
1112 * OBD_CONNECT_NODEVOH. */
1116 nd->intent.open.file->private_data = it;
1117 filp = lookup_instantiate_filp(nd,
1123 de = (struct dentry *)filp;
1126 } else if (it_disposition(it, DISP_OPEN_CREATE)) {
1127 /* XXX This can only reliably work on assumption
1128 * that there are NO hashed negative dentries.*/
1129 ll_d2d(dentry)->lld_it = it;
1130 it = NULL; /* Will be freed in ll_create_nd */
1131 /* We absolutely depend on ll_create_nd to be
1132 * called to not leak this intent and possible
1133 * data attached to it */
1138 ll_intent_release(it);
1139 OBD_FREE(it, sizeof(*it));
1142 de = ll_lookup_it(parent, dentry, NULL, NULL, NULL);
1147 #endif /* HAVE_IOP_ATOMIC_OPEN */
1149 /* We depend on "mode" being set with the proper file type/umask by now */
1150 static struct inode *ll_create_node(struct inode *dir, struct lookup_intent *it)
1152 struct inode *inode = NULL;
1153 struct ptlrpc_request *request = NULL;
1154 struct ll_sb_info *sbi = ll_i2sbi(dir);
1158 LASSERT(it && it->it_disposition);
1160 LASSERT(it_disposition(it, DISP_ENQ_CREATE_REF));
1161 request = it->it_request;
1162 it_clear_disposition(it, DISP_ENQ_CREATE_REF);
1163 rc = ll_prep_inode(&inode, request, dir->i_sb, it);
1165 GOTO(out, inode = ERR_PTR(rc));
1167 /* Pause to allow for a race with concurrent access by fid */
1168 OBD_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_NODE_PAUSE, cfs_fail_val);
1170 /* We asked for a lock on the directory, but were granted a
1171 * lock on the inode. Since we finally have an inode pointer,
1172 * stuff it in the lock. */
1173 CDEBUG(D_DLMTRACE, "setting l_ast_data to inode "DFID"(%p)\n",
1174 PFID(ll_inode2fid(inode)), inode);
1175 ll_set_lock_data(sbi->ll_md_exp, inode, it, NULL);
1178 ptlrpc_req_finished(request);
1183 * By the time this is called, we already have created the directory cache
1184 * entry for the new file, but it is so far negative - it has no inode.
1186 * We defer creating the OBD object(s) until open, to keep the intent and
1187 * non-intent code paths similar, and also because we do not have the MDS
1188 * inode number before calling ll_create_node() (which is needed for LOV),
1189 * so we would need to do yet another RPC to the MDS to store the LOV EA
1190 * data on the MDS. If needed, we would pass the PACKED lmm as data and
1191 * lmm_size in datalen (the MDS still has code which will handle that).
1193 * If the create succeeds, we fill in the inode information
1194 * with d_instantiate().
1196 static int ll_create_it(struct inode *dir, struct dentry *dentry,
1197 struct lookup_intent *it,
1198 void *secctx, __u32 secctxlen)
1200 struct inode *inode;
1205 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), intent=%s\n",
1206 dentry->d_name.len, dentry->d_name.name,
1207 PFID(ll_inode2fid(dir)), dir, LL_IT2STR(it));
1209 rc = it_open_error(DISP_OPEN_CREATE, it);
1213 inode = ll_create_node(dir, it);
1215 RETURN(PTR_ERR(inode));
1217 if ((ll_i2sbi(inode)->ll_flags & LL_SBI_FILE_SECCTX) && secctx) {
1219 /* must be done before d_instantiate, because it calls
1220 * security_d_instantiate, which means a getxattr if security
1221 * context is not set yet */
1222 rc = security_inode_notifysecctx(inode, secctx, secctxlen);
1223 inode_unlock(inode);
1228 d_instantiate(dentry, inode);
1230 if (!(ll_i2sbi(inode)->ll_flags & LL_SBI_FILE_SECCTX)) {
1231 rc = ll_inode_init_security(dentry, inode, dir);
1236 ll_set_lock_data(ll_i2sbi(dir)->ll_md_exp, inode, it, &bits);
1237 if (bits & MDS_INODELOCK_LOOKUP)
1238 d_lustre_revalidate(dentry);
1243 void ll_update_times(struct ptlrpc_request *request, struct inode *inode)
1245 struct mdt_body *body = req_capsule_server_get(&request->rq_pill,
1249 if (body->mbo_valid & OBD_MD_FLMTIME &&
1250 body->mbo_mtime > inode->i_mtime.tv_sec) {
1252 "setting fid " DFID " mtime from %lld to %llu\n",
1253 PFID(ll_inode2fid(inode)),
1254 (s64)inode->i_mtime.tv_sec, body->mbo_mtime);
1255 inode->i_mtime.tv_sec = body->mbo_mtime;
1258 if (body->mbo_valid & OBD_MD_FLCTIME &&
1259 body->mbo_ctime > inode->i_ctime.tv_sec)
1260 inode->i_ctime.tv_sec = body->mbo_ctime;
1263 static int ll_new_node(struct inode *dir, struct dentry *dchild,
1264 const char *tgt, umode_t mode, int rdev, __u32 opc)
1266 struct qstr *name = &dchild->d_name;
1267 struct ptlrpc_request *request = NULL;
1268 struct md_op_data *op_data;
1269 struct inode *inode = NULL;
1270 struct ll_sb_info *sbi = ll_i2sbi(dir);
1275 if (unlikely(tgt != NULL))
1276 tgt_len = strlen(tgt) + 1;
1279 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
1280 name->len, 0, opc, NULL);
1281 if (IS_ERR(op_data))
1282 GOTO(err_exit, err = PTR_ERR(op_data));
1284 if (sbi->ll_flags & LL_SBI_FILE_SECCTX) {
1285 err = ll_dentry_init_security(dchild, mode, &dchild->d_name,
1286 &op_data->op_file_secctx_name,
1287 &op_data->op_file_secctx,
1288 &op_data->op_file_secctx_size);
1290 GOTO(err_exit, err);
1293 err = md_create(sbi->ll_md_exp, op_data, tgt, tgt_len, mode,
1294 from_kuid(&init_user_ns, current_fsuid()),
1295 from_kgid(&init_user_ns, current_fsgid()),
1296 cfs_curproc_cap_pack(), rdev, &request);
1297 if (err < 0 && err != -EREMOTE)
1298 GOTO(err_exit, err);
1300 /* If the client doesn't know where to create a subdirectory (or
1301 * in case of a race that sends the RPC to the wrong MDS), the
1302 * MDS will return -EREMOTE and the client will fetch the layout
1303 * of the directory, then create the directory on the right MDT. */
1304 if (unlikely(err == -EREMOTE)) {
1305 struct ll_inode_info *lli = ll_i2info(dir);
1306 struct lmv_user_md *lum;
1310 ptlrpc_req_finished(request);
1313 err2 = ll_dir_getstripe(dir, (void **)&lum, &lumsize, &request,
1314 OBD_MD_DEFAULT_MEA);
1316 /* Update stripe_offset and retry */
1317 lli->lli_def_stripe_offset = lum->lum_stripe_offset;
1318 } else if (err2 == -ENODATA &&
1319 lli->lli_def_stripe_offset != -1) {
1320 /* If there are no default stripe EA on the MDT, but the
1321 * client has default stripe, then it probably means
1322 * default stripe EA has just been deleted. */
1323 lli->lli_def_stripe_offset = -1;
1325 GOTO(err_exit, err);
1328 ptlrpc_req_finished(request);
1330 ll_finish_md_op_data(op_data);
1334 ll_update_times(request, dir);
1336 CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_NEWNODE_PAUSE, cfs_fail_val);
1338 err = ll_prep_inode(&inode, request, dchild->d_sb, NULL);
1340 GOTO(err_exit, err);
1342 if (sbi->ll_flags & LL_SBI_FILE_SECCTX) {
1344 /* must be done before d_instantiate, because it calls
1345 * security_d_instantiate, which means a getxattr if security
1346 * context is not set yet */
1347 err = security_inode_notifysecctx(inode,
1348 op_data->op_file_secctx,
1349 op_data->op_file_secctx_size);
1350 inode_unlock(inode);
1352 GOTO(err_exit, err);
1355 d_instantiate(dchild, inode);
1357 if (!(sbi->ll_flags & LL_SBI_FILE_SECCTX)) {
1358 err = ll_inode_init_security(dchild, inode, dir);
1360 GOTO(err_exit, err);
1365 if (request != NULL)
1366 ptlrpc_req_finished(request);
1368 if (!IS_ERR_OR_NULL(op_data))
1369 ll_finish_md_op_data(op_data);
1374 static int ll_mknod(struct inode *dir, struct dentry *dchild, ll_umode_t mode,
1377 struct qstr *name = &dchild->d_name;
1381 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p) mode %o dev %x\n",
1382 name->len, name->name, PFID(ll_inode2fid(dir)), dir,
1385 if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
1386 mode &= ~current_umask();
1388 switch (mode & S_IFMT) {
1390 mode |= S_IFREG; /* for mode = 0 case, fallthrough */
1396 err = ll_new_node(dir, dchild, NULL, mode, old_encode_dev(rdev),
1407 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKNOD, 1);
1412 #ifdef HAVE_IOP_ATOMIC_OPEN
1414 * Plain create. Intent create is handled in atomic_open.
1416 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
1417 umode_t mode, bool want_excl)
1421 CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_FILE_PAUSE, cfs_fail_val);
1423 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), "
1424 "flags=%u, excl=%d\n", dentry->d_name.len,
1425 dentry->d_name.name, PFID(ll_inode2fid(dir)),
1426 dir, mode, want_excl);
1428 /* Using mknod(2) to create a regular file is designed to not recognize
1429 * volatile file name, so we use ll_mknod() here. */
1430 rc = ll_mknod(dir, dentry, mode, 0);
1432 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_CREATE, 1);
1434 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, unhashed %d\n",
1435 dentry->d_name.len, dentry->d_name.name, d_unhashed(dentry));
1439 #else /* !HAVE_IOP_ATOMIC_OPEN */
1440 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
1441 ll_umode_t mode, struct nameidata *nd)
1443 struct ll_dentry_data *lld = ll_d2d(dentry);
1444 struct lookup_intent *it = NULL;
1447 CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_FILE_PAUSE, cfs_fail_val);
1453 /* LU-8559: use LUSTRE_OPC_CREATE for non atomic open case
1454 * so that volatile file name is recoginized.
1455 * Mknod(2), however, is designed to not recognize volatile
1456 * file name to avoid inode leak under orphan directory until
1458 return ll_new_node(dir, dentry, NULL, mode, 0,
1464 /* Was there an error? Propagate it! */
1465 if (it->it_status) {
1470 rc = ll_create_it(dir, dentry, it, NULL, 0);
1471 if (nd && (nd->flags & LOOKUP_OPEN) && dentry->d_inode) { /* Open */
1474 nd->intent.open.file->private_data = it;
1475 filp = lookup_instantiate_filp(nd, dentry, NULL);
1481 ll_intent_release(it);
1482 OBD_FREE(it, sizeof(*it));
1485 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_CREATE, 1);
1489 #endif /* HAVE_IOP_ATOMIC_OPEN */
1491 static int ll_symlink(struct inode *dir, struct dentry *dchild,
1492 const char *oldpath)
1494 struct qstr *name = &dchild->d_name;
1498 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), target=%.*s\n",
1499 name->len, name->name, PFID(ll_inode2fid(dir)),
1500 dir, 3000, oldpath);
1502 err = ll_new_node(dir, dchild, oldpath, S_IFLNK | S_IRWXUGO, 0,
1503 LUSTRE_OPC_SYMLINK);
1506 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_SYMLINK, 1);
1511 static int ll_link(struct dentry *old_dentry, struct inode *dir,
1512 struct dentry *new_dentry)
1514 struct inode *src = old_dentry->d_inode;
1515 struct qstr *name = &new_dentry->d_name;
1516 struct ll_sb_info *sbi = ll_i2sbi(dir);
1517 struct ptlrpc_request *request = NULL;
1518 struct md_op_data *op_data;
1522 CDEBUG(D_VFSTRACE, "VFS Op: inode="DFID"(%p), dir="DFID"(%p), "
1523 "target=%.*s\n", PFID(ll_inode2fid(src)), src,
1524 PFID(ll_inode2fid(dir)), dir, name->len, name->name);
1526 op_data = ll_prep_md_op_data(NULL, src, dir, name->name, name->len,
1527 0, LUSTRE_OPC_ANY, NULL);
1528 if (IS_ERR(op_data))
1529 RETURN(PTR_ERR(op_data));
1531 err = md_link(sbi->ll_md_exp, op_data, &request);
1532 ll_finish_md_op_data(op_data);
1536 ll_update_times(request, dir);
1537 ll_stats_ops_tally(sbi, LPROC_LL_LINK, 1);
1540 ptlrpc_req_finished(request);
1544 static int ll_mkdir(struct inode *dir, struct dentry *dchild, ll_umode_t mode)
1546 struct qstr *name = &dchild->d_name;
1550 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1551 name->len, name->name, PFID(ll_inode2fid(dir)), dir);
1553 if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
1554 mode &= ~current_umask();
1556 mode = (mode & (S_IRWXUGO|S_ISVTX)) | S_IFDIR;
1558 err = ll_new_node(dir, dchild, NULL, mode, 0, LUSTRE_OPC_MKDIR);
1560 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKDIR, 1);
1565 static int ll_rmdir(struct inode *dir, struct dentry *dchild)
1567 struct qstr *name = &dchild->d_name;
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 name->len, name->name, PFID(ll_inode2fid(dir)), dir);
1576 if (unlikely(d_mountpoint(dchild)))
1579 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len,
1580 S_IFDIR, LUSTRE_OPC_ANY, NULL);
1581 if (IS_ERR(op_data))
1582 RETURN(PTR_ERR(op_data));
1584 if (dchild->d_inode != NULL)
1585 op_data->op_fid3 = *ll_inode2fid(dchild->d_inode);
1587 op_data->op_fid2 = op_data->op_fid3;
1588 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1589 ll_finish_md_op_data(op_data);
1591 ll_update_times(request, dir);
1592 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR, 1);
1595 ptlrpc_req_finished(request);
1602 int ll_rmdir_entry(struct inode *dir, char *name, int namelen)
1604 struct ptlrpc_request *request = NULL;
1605 struct md_op_data *op_data;
1609 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1610 namelen, name, PFID(ll_inode2fid(dir)), dir);
1612 op_data = ll_prep_md_op_data(NULL, dir, NULL, name, strlen(name),
1613 S_IFDIR, LUSTRE_OPC_ANY, NULL);
1614 if (IS_ERR(op_data))
1615 RETURN(PTR_ERR(op_data));
1616 op_data->op_cli_flags |= CLI_RM_ENTRY;
1617 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1618 ll_finish_md_op_data(op_data);
1620 ll_update_times(request, dir);
1621 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR, 1);
1624 ptlrpc_req_finished(request);
1628 static int ll_unlink(struct inode *dir, struct dentry *dchild)
1630 struct qstr *name = &dchild->d_name;
1631 struct ptlrpc_request *request = NULL;
1632 struct md_op_data *op_data;
1633 struct mdt_body *body;
1636 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1637 name->len, name->name, PFID(ll_inode2fid(dir)), dir);
1640 * XXX: unlink bind mountpoint maybe call to here,
1641 * just check it as vfs_unlink does.
1643 if (unlikely(d_mountpoint(dchild)))
1646 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len, 0,
1647 LUSTRE_OPC_ANY, NULL);
1648 if (IS_ERR(op_data))
1649 RETURN(PTR_ERR(op_data));
1651 op_data->op_fid3 = *ll_inode2fid(dchild->d_inode);
1653 op_data->op_fid2 = op_data->op_fid3;
1654 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1655 ll_finish_md_op_data(op_data);
1660 * The server puts attributes in on the last unlink, use them to update
1661 * the link count so the inode can be freed immediately.
1663 body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
1664 if (body->mbo_valid & OBD_MD_FLNLINK)
1665 set_nlink(dchild->d_inode, body->mbo_nlink);
1667 ll_update_times(request, dir);
1668 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_UNLINK, 1);
1671 ptlrpc_req_finished(request);
1675 static int ll_rename(struct inode *src, struct dentry *src_dchild,
1676 struct inode *tgt, struct dentry *tgt_dchild
1677 #ifdef HAVE_IOPS_RENAME_WITH_FLAGS
1678 , unsigned int flags
1682 struct qstr *src_name = &src_dchild->d_name;
1683 struct qstr *tgt_name = &tgt_dchild->d_name;
1684 struct ptlrpc_request *request = NULL;
1685 struct ll_sb_info *sbi = ll_i2sbi(src);
1686 struct md_op_data *op_data;
1690 #ifdef HAVE_IOPS_RENAME_WITH_FLAGS
1695 CDEBUG(D_VFSTRACE, "VFS Op:oldname=%.*s, src_dir="DFID
1696 "(%p), newname=%.*s, tgt_dir="DFID"(%p)\n",
1697 src_name->len, src_name->name,
1698 PFID(ll_inode2fid(src)), src, tgt_name->len,
1699 tgt_name->name, PFID(ll_inode2fid(tgt)), tgt);
1701 if (unlikely(d_mountpoint(src_dchild) || d_mountpoint(tgt_dchild)))
1704 op_data = ll_prep_md_op_data(NULL, src, tgt, NULL, 0, 0,
1705 LUSTRE_OPC_ANY, NULL);
1706 if (IS_ERR(op_data))
1707 RETURN(PTR_ERR(op_data));
1709 if (src_dchild->d_inode != NULL)
1710 op_data->op_fid3 = *ll_inode2fid(src_dchild->d_inode);
1712 if (tgt_dchild->d_inode != NULL)
1713 op_data->op_fid4 = *ll_inode2fid(tgt_dchild->d_inode);
1715 err = md_rename(sbi->ll_md_exp, op_data,
1716 src_name->name, src_name->len,
1717 tgt_name->name, tgt_name->len, &request);
1718 ll_finish_md_op_data(op_data);
1720 ll_update_times(request, src);
1721 ll_update_times(request, tgt);
1722 ll_stats_ops_tally(sbi, LPROC_LL_RENAME, 1);
1725 ptlrpc_req_finished(request);
1728 d_move(src_dchild, tgt_dchild);
1733 const struct inode_operations ll_dir_inode_operations = {
1735 #ifdef HAVE_IOP_ATOMIC_OPEN
1736 .atomic_open = ll_atomic_open,
1738 .lookup = ll_lookup_nd,
1739 .create = ll_create_nd,
1740 /* We need all these non-raw things for NFSD, to not patch it. */
1741 .unlink = ll_unlink,
1744 .symlink = ll_symlink,
1746 .rename = ll_rename,
1747 .setattr = ll_setattr,
1748 .getattr = ll_getattr,
1749 .permission = ll_inode_permission,
1750 #ifdef HAVE_IOP_XATTR
1751 .setxattr = ll_setxattr,
1752 .getxattr = ll_getxattr,
1753 .removexattr = ll_removexattr,
1755 .listxattr = ll_listxattr,
1756 #ifdef HAVE_IOP_GET_ACL
1757 .get_acl = ll_get_acl,
1759 #ifdef HAVE_IOP_SET_ACL
1760 .set_acl = ll_set_acl,
1764 const struct inode_operations ll_special_inode_operations = {
1765 .setattr = ll_setattr,
1766 .getattr = ll_getattr,
1767 .permission = ll_inode_permission,
1768 #ifdef HAVE_IOP_XATTR
1769 .setxattr = ll_setxattr,
1770 .getxattr = ll_getxattr,
1771 .removexattr = ll_removexattr,
1773 .listxattr = ll_listxattr,
1774 #ifdef HAVE_IOP_GET_ACL
1775 .get_acl = ll_get_acl,
1777 #ifdef HAVE_IOP_SET_ACL
1778 .set_acl = ll_set_acl,