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 #include <linux/uidgid.h>
43 #define DEBUG_SUBSYSTEM S_LLITE
45 #include <obd_support.h>
46 #include <lustre_fid.h>
47 #include <lustre_dlm.h>
48 #include "llite_internal.h"
50 static int ll_create_it(struct inode *dir, struct dentry *dentry,
51 struct lookup_intent *it,
52 void *secctx, __u32 secctxlen, bool encrypt);
54 /* called from iget5_locked->find_inode() under inode_lock spinlock */
55 static int ll_test_inode(struct inode *inode, void *opaque)
57 struct ll_inode_info *lli = ll_i2info(inode);
58 struct lustre_md *md = opaque;
60 if (unlikely(!(md->body->mbo_valid & OBD_MD_FLID))) {
61 CERROR("MDS body missing FID\n");
65 if (!lu_fid_eq(&lli->lli_fid, &md->body->mbo_fid1))
71 static int ll_set_inode(struct inode *inode, void *opaque)
73 struct ll_inode_info *lli = ll_i2info(inode);
74 struct mdt_body *body = ((struct lustre_md *)opaque)->body;
76 if (unlikely(!(body->mbo_valid & OBD_MD_FLID))) {
77 CERROR("MDS body missing FID\n");
81 lli->lli_fid = body->mbo_fid1;
82 if (unlikely(!(body->mbo_valid & OBD_MD_FLTYPE))) {
83 CERROR("Can not initialize inode "DFID" without object type: "
85 PFID(&lli->lli_fid), body->mbo_valid);
89 inode->i_mode = (inode->i_mode & ~S_IFMT) | (body->mbo_mode & S_IFMT);
90 if (unlikely(inode->i_mode == 0)) {
91 CERROR("Invalid inode "DFID" type\n", PFID(&lli->lli_fid));
102 * Get an inode by inode number(@hash), which is already instantiated by
103 * the intent lookup).
105 struct inode *ll_iget(struct super_block *sb, ino_t hash,
106 struct lustre_md *md)
114 inode = iget5_locked(sb, hash, ll_test_inode, ll_set_inode, md);
116 RETURN(ERR_PTR(-ENOMEM));
118 if (inode->i_state & I_NEW) {
119 rc = ll_read_inode2(inode, md);
120 if (rc == 0 && S_ISREG(inode->i_mode) &&
121 ll_i2info(inode)->lli_clob == NULL)
122 rc = cl_file_inode_init(inode, md);
125 /* Let's clear directory lsm here, otherwise
126 * make_bad_inode() will reset the inode mode
127 * to regular, then ll_clear_inode will not
128 * be able to clear lsm_md */
129 if (S_ISDIR(inode->i_mode))
130 ll_dir_clear_lsm_md(inode);
131 make_bad_inode(inode);
132 unlock_new_inode(inode);
136 inode_has_no_xattr(inode);
137 unlock_new_inode(inode);
139 } else if (is_bad_inode(inode)) {
141 inode = ERR_PTR(-ESTALE);
142 } else if (!(inode->i_state & (I_FREEING | I_CLEAR))) {
143 rc = ll_update_inode(inode, md);
144 CDEBUG(D_VFSTRACE, "got inode: "DFID"(%p): rc = %d\n",
145 PFID(&md->body->mbo_fid1), inode, rc);
147 if (S_ISDIR(inode->i_mode))
148 ll_dir_clear_lsm_md(inode);
157 static void ll_invalidate_negative_children(struct inode *dir)
159 struct dentry *dentry, *tmp_subdir;
161 spin_lock(&dir->i_lock);
162 hlist_for_each_entry(dentry, &dir->i_dentry, d_alias) {
163 spin_lock(&dentry->d_lock);
164 if (!list_empty(&dentry->d_subdirs)) {
165 struct dentry *child;
167 list_for_each_entry_safe(child, tmp_subdir,
170 if (child->d_inode == NULL)
171 d_lustre_invalidate(child, 1);
174 spin_unlock(&dentry->d_lock);
176 spin_unlock(&dir->i_lock);
179 int ll_test_inode_by_fid(struct inode *inode, void *opaque)
181 return lu_fid_eq(&ll_i2info(inode)->lli_fid, opaque);
184 static int ll_dom_lock_cancel(struct inode *inode, struct ldlm_lock *lock)
187 struct ll_inode_info *lli = ll_i2info(inode);
192 if (!lli->lli_clob) {
193 /* due to DoM read on open, there may exist pages for Lustre
194 * regular file even though cl_object is not set up yet. */
195 truncate_inode_pages(inode->i_mapping, 0);
199 env = cl_env_get(&refcheck);
201 RETURN(PTR_ERR(env));
203 /* reach MDC layer to flush data under the DoM ldlm lock */
204 rc = cl_object_flush(env, lli->lli_clob, lock);
205 if (rc == -ENODATA) {
206 CDEBUG(D_INODE, "inode "DFID" layout has no DoM stripe\n",
207 PFID(ll_inode2fid(inode)));
208 /* most likely result of layout change, do nothing */
212 cl_env_put(env, &refcheck);
216 static void ll_lock_cancel_bits(struct ldlm_lock *lock, __u64 to_cancel)
218 struct inode *inode = ll_inode_from_resource_lock(lock);
219 struct ll_inode_info *lli;
220 __u64 bits = to_cancel;
226 /* That means the inode is evicted most likely and may cause
227 * the skipping of lock cleanups below, so print the message
230 if (lock->l_resource->lr_lvb_inode)
232 "can't take inode for the lock (%sevicted)\n",
233 lock->l_resource->lr_lvb_inode->i_state &
234 I_FREEING ? "" : "not ");
238 if (!fid_res_name_eq(ll_inode2fid(inode),
239 &lock->l_resource->lr_name)) {
240 LDLM_ERROR(lock, "data mismatch with object "DFID"(%p)",
241 PFID(ll_inode2fid(inode)), inode);
245 if (bits & MDS_INODELOCK_XATTR) {
246 ll_xattr_cache_destroy(inode);
247 bits &= ~MDS_INODELOCK_XATTR;
250 /* For OPEN locks we differentiate between lock modes
251 * LCK_CR, LCK_CW, LCK_PR - bug 22891 */
252 if (bits & MDS_INODELOCK_OPEN)
253 ll_have_md_lock(inode, &bits, lock->l_req_mode);
255 if (bits & MDS_INODELOCK_OPEN) {
258 switch (lock->l_req_mode) {
269 LDLM_ERROR(lock, "bad lock mode for OPEN lock");
273 ll_md_real_close(inode, fmode);
275 bits &= ~MDS_INODELOCK_OPEN;
278 if (bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
279 MDS_INODELOCK_LAYOUT | MDS_INODELOCK_PERM |
281 ll_have_md_lock(inode, &bits, LCK_MINMODE);
283 if (bits & MDS_INODELOCK_DOM) {
284 rc = ll_dom_lock_cancel(inode, lock);
286 CDEBUG(D_INODE, "cannot flush DoM data "
288 PFID(ll_inode2fid(inode)), rc);
291 if (bits & MDS_INODELOCK_LAYOUT) {
292 struct cl_object_conf conf = {
293 .coc_opc = OBJECT_CONF_INVALIDATE,
297 rc = ll_layout_conf(inode, &conf);
299 CDEBUG(D_INODE, "cannot invalidate layout of "
301 PFID(ll_inode2fid(inode)), rc);
304 lli = ll_i2info(inode);
306 if (bits & MDS_INODELOCK_UPDATE)
307 ll_file_set_flag(lli, LLIF_UPDATE_ATIME);
309 if ((bits & MDS_INODELOCK_UPDATE) && S_ISDIR(inode->i_mode)) {
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) {
412 /* do not convert other modes */
416 /* is lock is too old to be converted? */
417 lock_res_and_lock(lock);
418 if (ktime_after(ktime_get(),
419 ktime_add(lock->l_last_used, ns->ns_dirty_age_limit))) {
420 unlock_res_and_lock(lock);
423 unlock_res_and_lock(lock);
425 inode = ll_inode_from_resource_lock(lock);
426 ll_have_md_lock(inode, &bits, mode);
431 int ll_md_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *ld,
432 void *data, int flag)
434 struct lustre_handle lockh;
440 case LDLM_CB_BLOCKING:
442 __u64 cancel_flags = LCF_ASYNC;
444 /* if lock convert is not needed then still have to
445 * pass lock via ldlm_cli_convert() to keep all states
446 * correct, set cancel_bits to full lock bits to cause
447 * full cancel to happen.
449 if (!ll_md_need_convert(lock)) {
450 lock_res_and_lock(lock);
451 lock->l_policy_data.l_inodebits.cancel_bits =
452 lock->l_policy_data.l_inodebits.bits;
453 unlock_res_and_lock(lock);
455 rc = ldlm_cli_convert(lock, cancel_flags);
458 /* continue with cancel otherwise */
459 ldlm_lock2handle(lock, &lockh);
460 rc = ldlm_cli_cancel(&lockh, cancel_flags);
462 CDEBUG(D_INODE, "ldlm_cli_cancel: rc = %d\n", rc);
467 case LDLM_CB_CANCELING:
469 __u64 to_cancel = lock->l_policy_data.l_inodebits.bits;
471 /* Nothing to do for non-granted locks */
472 if (!ldlm_is_granted(lock))
475 /* If 'ld' is supplied then bits to be cancelled are passed
476 * implicitly by lock converting and cancel_bits from 'ld'
477 * should be used. Otherwise full cancel is being performed
478 * and lock inodebits are used.
480 * Note: we cannot rely on cancel_bits in lock itself at this
481 * moment because they can be changed by concurrent thread,
482 * so ldlm_cli_inodebits_convert() pass cancel bits implicitly
486 /* partial bits cancel allowed only during convert */
487 LASSERT(ldlm_is_converting(lock));
488 /* mask cancel bits by lock bits so only no any unused
489 * bits are passed to ll_lock_cancel_bits()
491 to_cancel &= ld->l_policy_data.l_inodebits.cancel_bits;
493 ll_lock_cancel_bits(lock, to_cancel);
503 __u32 ll_i2suppgid(struct inode *i)
505 if (in_group_p(i->i_gid))
506 return (__u32)from_kgid(&init_user_ns, i->i_gid);
508 return (__u32) __kgid_val(INVALID_GID);
511 /* Pack the required supplementary groups into the supplied groups array.
512 * If we don't need to use the groups from the target inode(s) then we
513 * instead pack one or more groups from the user's supplementary group
514 * array in case it might be useful. Not needed if doing an MDS-side upcall. */
515 void ll_i2gids(__u32 *suppgids, struct inode *i1, struct inode *i2)
518 LASSERT(suppgids != NULL);
520 suppgids[0] = ll_i2suppgid(i1);
523 suppgids[1] = ll_i2suppgid(i2);
529 * try to reuse three types of dentry:
530 * 1. unhashed alias, this one is unhashed by d_invalidate (but it may be valid
531 * by concurrent .revalidate).
532 * 2. INVALID alias (common case for no valid ldlm lock held, but this flag may
533 * be cleared by others calling d_lustre_revalidate).
534 * 3. DISCONNECTED alias.
536 static struct dentry *ll_find_alias(struct inode *inode, struct dentry *dentry)
538 struct dentry *alias, *discon_alias, *invalid_alias;
540 if (hlist_empty(&inode->i_dentry))
543 discon_alias = invalid_alias = NULL;
545 spin_lock(&inode->i_lock);
546 hlist_for_each_entry(alias, &inode->i_dentry, d_alias) {
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 spin_unlock(&inode->i_lock);
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,
614 ktime_t kstart, bool encrypt)
616 struct inode *inode = NULL;
619 struct dentry *alias;
622 /* NB 1 request reference will be taken away by ll_intent_lock()
624 CDEBUG(D_DENTRY, "it %p it_disposition %x\n", it,
626 if (!it_disposition(it, DISP_LOOKUP_NEG)) {
627 struct req_capsule *pill = &request->rq_pill;
628 struct mdt_body *body = req_capsule_server_get(pill,
631 rc = ll_prep_inode(&inode, request, (*de)->d_sb, it);
635 if (it->it_op & IT_OPEN)
636 ll_dom_finish_open(inode, request, it);
638 ll_set_lock_data(ll_i2sbi(parent)->ll_md_exp, inode, it, &bits);
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 != NULL && secctxlen != 0) {
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);
693 rc = llcrypt_get_encryption_info(inode);
696 if (!llcrypt_has_encryption_key(inode))
697 GOTO(out, rc = -ENOKEY);
699 } else if (!it_disposition(it, DISP_OPEN_CREATE)) {
701 * If file was created on the server, the dentry is revalidated
702 * in ll_create_it if the lock allows for it.
704 /* Check that parent has UPDATE lock. */
705 struct lookup_intent parent_it = {
707 .it_lock_handle = 0 };
708 struct lu_fid fid = ll_i2info(parent)->lli_fid;
710 /* If it is striped directory, get the real stripe parent */
711 if (unlikely(ll_dir_striped(parent))) {
712 rc = md_get_fid_from_lsm(ll_i2mdexp(parent),
713 ll_i2info(parent)->lli_lsm_md,
715 (*de)->d_name.len, &fid);
720 if (md_revalidate_lock(ll_i2mdexp(parent), &parent_it, &fid,
722 d_lustre_revalidate(*de);
723 ll_intent_release(&parent_it);
727 if (it_disposition(it, DISP_OPEN_CREATE)) {
728 ll_stats_ops_tally(ll_i2sbi(parent), LPROC_LL_MKNOD,
729 ktime_us_delta(ktime_get(), kstart));
735 if (rc != 0 && it->it_op & IT_OPEN) {
736 ll_intent_drop_lock(it);
737 ll_open_cleanup((*de)->d_sb, request);
743 static struct dentry *ll_lookup_it(struct inode *parent, struct dentry *dentry,
744 struct lookup_intent *it,
745 void **secctx, __u32 *secctxlen,
746 struct pcc_create_attach *pca,
749 ktime_t kstart = ktime_get();
750 struct lookup_intent lookup_it = { .it_op = IT_LOOKUP };
751 struct dentry *save = dentry, *retval;
752 struct ptlrpc_request *req = NULL;
753 struct md_op_data *op_data = NULL;
754 struct lov_user_md *lum = NULL;
757 char secctx_name[XATTR_NAME_MAX + 1];
761 if (dentry->d_name.len > ll_i2sbi(parent)->ll_namelen)
762 RETURN(ERR_PTR(-ENAMETOOLONG));
764 CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p), intent=%s\n",
765 dentry, PFID(ll_inode2fid(parent)), parent, LL_IT2STR(it));
767 if (d_mountpoint(dentry))
768 CERROR("Tell Peter, lookup on mtpt, it %s\n", LL_IT2STR(it));
770 if (it == NULL || it->it_op == IT_GETXATTR)
773 if (it->it_op == IT_GETATTR && dentry_may_statahead(parent, dentry)) {
774 rc = ll_revalidate_statahead(parent, &dentry, 0);
776 RETURN(dentry == save ? NULL : dentry);
779 if (it->it_op & IT_OPEN && it->it_flags & FMODE_WRITE &&
780 dentry->d_sb->s_flags & SB_RDONLY)
781 RETURN(ERR_PTR(-EROFS));
783 if (it->it_op & IT_CREAT)
784 opc = LUSTRE_OPC_CREATE;
786 opc = LUSTRE_OPC_ANY;
788 op_data = ll_prep_md_op_data(NULL, parent, NULL, dentry->d_name.name,
789 dentry->d_name.len, 0, opc, NULL);
791 GOTO(out, retval = ERR_CAST(op_data));
793 /* enforce umask if acl disabled or MDS doesn't support umask */
794 if (!IS_POSIXACL(parent) || !exp_connect_umask(ll_i2mdexp(parent)))
795 it->it_create_mode &= ~current_umask();
797 if (it->it_op & IT_CREAT &&
798 ll_i2sbi(parent)->ll_flags & LL_SBI_FILE_SECCTX) {
799 rc = ll_dentry_init_security(dentry, it->it_create_mode,
801 &op_data->op_file_secctx_name,
802 &op_data->op_file_secctx,
803 &op_data->op_file_secctx_size);
805 GOTO(out, retval = ERR_PTR(rc));
807 *secctx = op_data->op_file_secctx;
808 if (secctxlen != NULL)
809 *secctxlen = op_data->op_file_secctx_size;
813 if (secctxlen != NULL)
817 /* ask for security context upon intent */
818 if (it->it_op & (IT_LOOKUP | IT_GETATTR | IT_OPEN)) {
819 /* get name of security xattr to request to server */
820 rc = ll_listsecurity(parent, secctx_name,
821 sizeof(secctx_name));
823 CDEBUG(D_SEC, "cannot get security xattr name for "
825 PFID(ll_inode2fid(parent)), rc);
827 op_data->op_file_secctx_name = secctx_name;
828 op_data->op_file_secctx_name_size = rc;
829 CDEBUG(D_SEC, "'%.*s' is security xattr for "DFID"\n",
830 rc, secctx_name, PFID(ll_inode2fid(parent)));
834 if (pca && pca->pca_dataset) {
835 struct pcc_dataset *dataset = pca->pca_dataset;
839 GOTO(out, retval = ERR_PTR(-ENOMEM));
841 lum->lmm_magic = LOV_USER_MAGIC_V1;
842 lum->lmm_pattern = LOV_PATTERN_F_RELEASED | LOV_PATTERN_RAID0;
843 op_data->op_data = lum;
844 op_data->op_data_size = sizeof(*lum);
845 op_data->op_archive_id = dataset->pccd_rwid;
847 rc = obd_fid_alloc(NULL, ll_i2mdexp(parent), &op_data->op_fid2,
850 GOTO(out, retval = ERR_PTR(rc));
852 rc = pcc_inode_create(parent->i_sb, dataset, &op_data->op_fid2,
855 GOTO(out, retval = ERR_PTR(rc));
857 it->it_flags |= MDS_OPEN_PCC;
860 rc = md_intent_lock(ll_i2mdexp(parent), op_data, it, &req,
861 &ll_md_blocking_ast, 0);
862 /* If the MDS allows the client to chgrp (CFS_SETGRP_PERM), but the
863 * client does not know which suppgid should be sent to the MDS, or
864 * some other(s) changed the target file's GID after this RPC sent
865 * to the MDS with the suppgid as the original GID, then we should
866 * try again with right suppgid. */
867 if (rc == -EACCES && it->it_op & IT_OPEN &&
868 it_disposition(it, DISP_OPEN_DENY)) {
869 struct mdt_body *body;
871 LASSERT(req != NULL);
873 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
874 if (op_data->op_suppgids[0] == body->mbo_gid ||
875 op_data->op_suppgids[1] == body->mbo_gid ||
876 !in_group_p(make_kgid(&init_user_ns, body->mbo_gid)))
877 GOTO(out, retval = ERR_PTR(-EACCES));
879 fid_zero(&op_data->op_fid2);
880 op_data->op_suppgids[1] = body->mbo_gid;
881 ptlrpc_req_finished(req);
883 ll_intent_release(it);
884 rc = md_intent_lock(ll_i2mdexp(parent), op_data, it, &req,
885 &ll_md_blocking_ast, 0);
889 GOTO(out, retval = ERR_PTR(rc));
891 /* dir layout may change */
892 ll_unlock_md_op_lsm(op_data);
893 rc = ll_lookup_it_finish(req, it, parent, &dentry,
894 secctx != NULL ? *secctx : NULL,
895 secctxlen != NULL ? *secctxlen : 0,
898 ll_intent_release(it);
899 GOTO(out, retval = ERR_PTR(rc));
902 if ((it->it_op & IT_OPEN) && dentry->d_inode &&
903 !S_ISREG(dentry->d_inode->i_mode) &&
904 !S_ISDIR(dentry->d_inode->i_mode)) {
905 ll_release_openhandle(dentry, it);
907 ll_lookup_finish_locks(it, dentry);
909 GOTO(out, retval = (dentry == save) ? NULL : dentry);
912 if (op_data != NULL && !IS_ERR(op_data)) {
913 if (secctx != NULL && secctxlen != NULL) {
914 /* caller needs sec ctx info, so reset it in op_data to
915 * prevent it from being freed */
916 op_data->op_file_secctx = NULL;
917 op_data->op_file_secctx_size = 0;
919 ll_finish_md_op_data(op_data);
925 ptlrpc_req_finished(req);
929 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
932 struct lookup_intent *itp, it = { .it_op = IT_GETATTR };
935 CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p), flags=%u\n",
936 dentry, PFID(ll_inode2fid(parent)), parent, flags);
939 * Optimize away (CREATE && !OPEN). Let .create handle the race.
940 * but only if we have write permissions there, otherwise we need
941 * to proceed with lookup. LU-4185
943 if ((flags & LOOKUP_CREATE) && !(flags & LOOKUP_OPEN) &&
944 (inode_permission(parent, MAY_WRITE | MAY_EXEC) == 0))
947 if (flags & (LOOKUP_PARENT|LOOKUP_OPEN|LOOKUP_CREATE))
951 de = ll_lookup_it(parent, dentry, itp, NULL, NULL, NULL, false);
954 ll_intent_release(itp);
959 #ifdef FMODE_CREATED /* added in Linux v4.18-rc1-20-g73a09dd */
960 # define ll_is_opened(o, f) ((f)->f_mode & FMODE_OPENED)
961 # define ll_finish_open(f, d, o) finish_open((f), (d), NULL)
963 # define ll_set_created(o, f) \
965 (f)->f_mode |= FMODE_CREATED; \
969 # define ll_is_opened(o, f) (*(o))
970 # define ll_finish_open(f, d, o) finish_open((f), (d), NULL, (o))
971 # define ll_last_arg , int *opened
972 # define ll_set_created(o, f) \
974 *(o) |= FILE_CREATED; \
980 * For cached negative dentry and new dentry, handle lookup/create/open
983 static int ll_atomic_open(struct inode *dir, struct dentry *dentry,
984 struct file *file, unsigned open_flags,
985 umode_t mode ll_last_arg)
987 struct lookup_intent *it;
989 long long lookup_flags = LOOKUP_OPEN;
992 struct ll_sb_info *sbi = NULL;
993 struct pcc_create_attach pca = { NULL, NULL };
994 bool encrypt = false;
999 "VFS Op:name=%pd, dir="DFID"(%p), file %p, open_flags %x, mode %x opened %d\n",
1000 dentry, PFID(ll_inode2fid(dir)), dir, file, open_flags, mode,
1001 ll_is_opened(opened, file));
1003 /* Only negative dentries enter here */
1004 LASSERT(dentry->d_inode == NULL);
1006 if (!d_unhashed(dentry)) {
1007 /* A valid negative dentry that just passed revalidation,
1008 * there's little point to try and open it server-side,
1009 * even though there's a minuscule chance it might succeed.
1010 * Either way it's a valid race to just return -ENOENT here.
1012 if (!(open_flags & O_CREAT))
1015 /* Otherwise we just unhash it to be rehashed afresh via
1016 * lookup if necessary
1021 OBD_ALLOC(it, sizeof(*it));
1025 it->it_op = IT_OPEN;
1026 if (open_flags & O_CREAT) {
1027 it->it_op |= IT_CREAT;
1028 lookup_flags |= LOOKUP_CREATE;
1029 sbi = ll_i2sbi(dir);
1030 /* Volatile file is used for HSM restore, so do not use PCC */
1031 if (!filename_is_volatile(dentry->d_name.name,
1032 dentry->d_name.len, NULL)) {
1033 struct pcc_matcher item;
1034 struct pcc_dataset *dataset;
1036 item.pm_uid = from_kuid(&init_user_ns, current_uid());
1037 item.pm_gid = from_kgid(&init_user_ns, current_gid());
1038 item.pm_projid = ll_i2info(dir)->lli_projid;
1039 item.pm_name = &dentry->d_name;
1040 dataset = pcc_dataset_match_get(&sbi->ll_pcc_super,
1042 pca.pca_dataset = dataset;
1045 it->it_create_mode = (mode & S_IALLUGO) | S_IFREG;
1046 it->it_flags = (open_flags & ~O_ACCMODE) | OPEN_FMODE(open_flags);
1047 it->it_flags &= ~MDS_OPEN_FL_INTERNAL;
1049 if (IS_ENCRYPTED(dir)) {
1050 /* we know that we are going to create a regular file because
1051 * we set S_IFREG bit on it->it_create_mode above
1053 rc = llcrypt_get_encryption_info(dir);
1055 GOTO(out_release, rc);
1056 if (!llcrypt_has_encryption_key(dir))
1057 GOTO(out_release, rc = -ENOKEY);
1062 /* Dentry added to dcache tree in ll_lookup_it */
1063 de = ll_lookup_it(dir, dentry, it, &secctx, &secctxlen, &pca, encrypt);
1066 else if (de != NULL)
1069 CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_FILE_PAUSE, cfs_fail_val);
1072 if (it_disposition(it, DISP_OPEN_CREATE)) {
1073 /* Dentry instantiated in ll_create_it. */
1074 rc = ll_create_it(dir, dentry, it, secctx, secctxlen,
1076 security_release_secctx(secctx, secctxlen);
1078 /* We dget in ll_splice_alias. */
1084 rc = pcc_inode_create_fini(dentry->d_inode, &pca);
1088 GOTO(out_release, rc);
1091 ll_set_created(opened, file);
1093 /* Open the file with O_CREAT, but the file already
1094 * existed on MDT. This may happend in the case that
1095 * the LOOKUP ibits lock is revoked and the
1096 * corresponding dentry cache is deleted.
1097 * i.e. In the current Lustre, the truncate operation
1098 * will revoke the LOOKUP ibits lock, and the file
1099 * dentry cache will be invalidated. The following open
1100 * with O_CREAT flag will call into ->atomic_open, the
1101 * file was wrongly though as newly created file and
1102 * try to auto cache the file. So after client knows it
1103 * is not a DISP_OPEN_CREATE, it should cleanup the
1104 * already created PCC copy.
1106 pcc_create_attach_cleanup(dir->i_sb, &pca);
1109 if (dentry->d_inode && it_disposition(it, DISP_OPEN_OPEN)) {
1111 if (S_ISFIFO(dentry->d_inode->i_mode)) {
1112 /* We cannot call open here as it might
1113 * deadlock. This case is unreachable in
1114 * practice because of OBD_CONNECT_NODEVOH. */
1115 rc = finish_no_open(file, de);
1117 file->private_data = it;
1118 rc = ll_finish_open(file, dentry, opened);
1119 /* We dget in ll_splice_alias. finish_open takes
1120 * care of dget for fd open.
1126 rc = finish_no_open(file, de);
1129 pcc_create_attach_cleanup(dir->i_sb, &pca);
1133 ll_intent_release(it);
1134 OBD_FREE(it, sizeof(*it));
1139 /* We depend on "mode" being set with the proper file type/umask by now */
1140 static struct inode *ll_create_node(struct inode *dir, struct lookup_intent *it)
1142 struct inode *inode = NULL;
1143 struct ptlrpc_request *request = NULL;
1144 struct ll_sb_info *sbi = ll_i2sbi(dir);
1148 LASSERT(it && it->it_disposition);
1150 LASSERT(it_disposition(it, DISP_ENQ_CREATE_REF));
1151 request = it->it_request;
1152 it_clear_disposition(it, DISP_ENQ_CREATE_REF);
1153 rc = ll_prep_inode(&inode, request, dir->i_sb, it);
1155 GOTO(out, inode = ERR_PTR(rc));
1157 /* Pause to allow for a race with concurrent access by fid */
1158 OBD_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_NODE_PAUSE, cfs_fail_val);
1160 /* We asked for a lock on the directory, but were granted a
1161 * lock on the inode. Since we finally have an inode pointer,
1162 * stuff it in the lock. */
1163 CDEBUG(D_DLMTRACE, "setting l_ast_data to inode "DFID"(%p)\n",
1164 PFID(ll_inode2fid(inode)), inode);
1165 ll_set_lock_data(sbi->ll_md_exp, inode, it, NULL);
1168 ptlrpc_req_finished(request);
1173 * By the time this is called, we already have created the directory cache
1174 * entry for the new file, but it is so far negative - it has no inode.
1176 * We defer creating the OBD object(s) until open, to keep the intent and
1177 * non-intent code paths similar, and also because we do not have the MDS
1178 * inode number before calling ll_create_node() (which is needed for LOV),
1179 * so we would need to do yet another RPC to the MDS to store the LOV EA
1180 * data on the MDS. If needed, we would pass the PACKED lmm as data and
1181 * lmm_size in datalen (the MDS still has code which will handle that).
1183 * If the create succeeds, we fill in the inode information
1184 * with d_instantiate().
1186 static int ll_create_it(struct inode *dir, struct dentry *dentry,
1187 struct lookup_intent *it,
1188 void *secctx, __u32 secctxlen, bool encrypt)
1190 struct inode *inode;
1195 CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p), intent=%s\n",
1196 dentry, PFID(ll_inode2fid(dir)), dir, LL_IT2STR(it));
1198 rc = it_open_error(DISP_OPEN_CREATE, it);
1202 inode = ll_create_node(dir, it);
1204 RETURN(PTR_ERR(inode));
1206 if ((ll_i2sbi(inode)->ll_flags & LL_SBI_FILE_SECCTX) &&
1209 /* must be done before d_instantiate, because it calls
1210 * security_d_instantiate, which means a getxattr if security
1211 * context is not set yet */
1212 rc = security_inode_notifysecctx(inode, secctx, secctxlen);
1213 inode_unlock(inode);
1218 d_instantiate(dentry, inode);
1221 rc = llcrypt_inherit_context(dir, inode, dentry, true);
1226 if (!(ll_i2sbi(inode)->ll_flags & LL_SBI_FILE_SECCTX)) {
1227 rc = ll_inode_init_security(dentry, inode, dir);
1232 ll_set_lock_data(ll_i2sbi(dir)->ll_md_exp, inode, it, &bits);
1233 if (bits & MDS_INODELOCK_LOOKUP)
1234 d_lustre_revalidate(dentry);
1239 void ll_update_times(struct ptlrpc_request *request, struct inode *inode)
1241 struct mdt_body *body = req_capsule_server_get(&request->rq_pill,
1245 if (body->mbo_valid & OBD_MD_FLMTIME &&
1246 body->mbo_mtime > inode->i_mtime.tv_sec) {
1248 "setting fid " DFID " mtime from %lld to %llu\n",
1249 PFID(ll_inode2fid(inode)),
1250 (s64)inode->i_mtime.tv_sec, body->mbo_mtime);
1251 inode->i_mtime.tv_sec = body->mbo_mtime;
1254 if (body->mbo_valid & OBD_MD_FLCTIME &&
1255 body->mbo_ctime > inode->i_ctime.tv_sec)
1256 inode->i_ctime.tv_sec = body->mbo_ctime;
1259 static int ll_new_node(struct inode *dir, struct dentry *dchild,
1260 const char *tgt, umode_t mode, int rdev, __u32 opc)
1262 struct qstr *name = &dchild->d_name;
1263 struct ptlrpc_request *request = NULL;
1264 struct md_op_data *op_data = NULL;
1265 struct inode *inode = NULL;
1266 struct ll_sb_info *sbi = ll_i2sbi(dir);
1272 if (unlikely(tgt != NULL))
1273 tgt_len = strlen(tgt) + 1;
1276 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
1277 name->len, 0, opc, NULL);
1278 if (IS_ERR(op_data))
1279 GOTO(err_exit, err = PTR_ERR(op_data));
1281 if (sbi->ll_flags & LL_SBI_FILE_SECCTX) {
1282 err = ll_dentry_init_security(dchild, mode, &dchild->d_name,
1283 &op_data->op_file_secctx_name,
1284 &op_data->op_file_secctx,
1285 &op_data->op_file_secctx_size);
1287 GOTO(err_exit, err);
1290 if ((IS_ENCRYPTED(dir) &&
1291 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) ||
1292 (unlikely(llcrypt_dummy_context_enabled(dir)) && S_ISDIR(mode))) {
1293 err = llcrypt_get_encryption_info(dir);
1295 GOTO(err_exit, err);
1296 if (!llcrypt_has_encryption_key(dir))
1297 GOTO(err_exit, err = -ENOKEY);
1301 err = md_create(sbi->ll_md_exp, op_data, tgt, tgt_len, mode,
1302 from_kuid(&init_user_ns, current_fsuid()),
1303 from_kgid(&init_user_ns, current_fsgid()),
1304 cfs_curproc_cap_pack(), rdev, &request);
1305 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 14, 58, 0)
1307 * server < 2.12.58 doesn't pack default LMV in intent_getattr reply,
1308 * fetch default LMV here.
1310 if (unlikely(err == -EREMOTE)) {
1311 struct ll_inode_info *lli = ll_i2info(dir);
1312 struct lmv_user_md *lum;
1316 ptlrpc_req_finished(request);
1319 err2 = ll_dir_getstripe(dir, (void **)&lum, &lumsize, &request,
1320 OBD_MD_DEFAULT_MEA);
1321 ll_finish_md_op_data(op_data);
1324 struct lustre_md md = { NULL };
1326 md.body = req_capsule_server_get(&request->rq_pill,
1329 GOTO(err_exit, err = -EPROTO);
1331 OBD_ALLOC_PTR(md.default_lmv);
1332 if (!md.default_lmv)
1333 GOTO(err_exit, err = -ENOMEM);
1335 md.default_lmv->lsm_md_magic = lum->lum_magic;
1336 md.default_lmv->lsm_md_stripe_count =
1337 lum->lum_stripe_count;
1338 md.default_lmv->lsm_md_master_mdt_index =
1339 lum->lum_stripe_offset;
1340 md.default_lmv->lsm_md_hash_type = lum->lum_hash_type;
1342 err = ll_update_inode(dir, &md);
1343 md_free_lustre_md(sbi->ll_md_exp, &md);
1345 GOTO(err_exit, err);
1346 } else if (err2 == -ENODATA && lli->lli_default_lsm_md) {
1348 * If there are no default stripe EA on the MDT, but the
1349 * client has default stripe, then it probably means
1350 * default stripe EA has just been deleted.
1352 down_write(&lli->lli_lsm_sem);
1353 if (lli->lli_default_lsm_md)
1354 OBD_FREE_PTR(lli->lli_default_lsm_md);
1355 lli->lli_default_lsm_md = NULL;
1356 up_write(&lli->lli_lsm_sem);
1358 GOTO(err_exit, err);
1361 ptlrpc_req_finished(request);
1368 GOTO(err_exit, err);
1370 ll_update_times(request, dir);
1372 CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_NEWNODE_PAUSE, cfs_fail_val);
1374 err = ll_prep_inode(&inode, request, dchild->d_sb, NULL);
1376 GOTO(err_exit, err);
1378 if (sbi->ll_flags & LL_SBI_FILE_SECCTX) {
1380 /* must be done before d_instantiate, because it calls
1381 * security_d_instantiate, which means a getxattr if security
1382 * context is not set yet */
1383 err = security_inode_notifysecctx(inode,
1384 op_data->op_file_secctx,
1385 op_data->op_file_secctx_size);
1386 inode_unlock(inode);
1388 GOTO(err_exit, err);
1391 d_instantiate(dchild, inode);
1394 err = llcrypt_inherit_context(dir, inode, NULL, true);
1396 GOTO(err_exit, err);
1399 if (!(sbi->ll_flags & LL_SBI_FILE_SECCTX)) {
1400 err = ll_inode_init_security(dchild, inode, dir);
1402 GOTO(err_exit, err);
1407 if (request != NULL)
1408 ptlrpc_req_finished(request);
1410 if (!IS_ERR_OR_NULL(op_data))
1411 ll_finish_md_op_data(op_data);
1416 static int ll_mknod(struct inode *dir, struct dentry *dchild, umode_t mode,
1419 ktime_t kstart = ktime_get();
1423 CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p) mode %o dev %x\n",
1424 dchild, PFID(ll_inode2fid(dir)), dir, mode, rdev);
1426 if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
1427 mode &= ~current_umask();
1429 switch (mode & S_IFMT) {
1438 err = ll_new_node(dir, dchild, NULL, mode, old_encode_dev(rdev),
1449 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKNOD,
1450 ktime_us_delta(ktime_get(), kstart));
1456 * Plain create. Intent create is handled in atomic_open.
1458 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
1459 umode_t mode, bool want_excl)
1461 ktime_t kstart = ktime_get();
1464 CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_FILE_PAUSE, cfs_fail_val);
1467 "VFS Op:name=%pd, dir="DFID"(%p), flags=%u, excl=%d\n",
1468 dentry, PFID(ll_inode2fid(dir)), dir, mode, want_excl);
1470 /* Using mknod(2) to create a regular file is designed to not recognize
1471 * volatile file name, so we use ll_mknod() here. */
1472 rc = ll_mknod(dir, dentry, mode, 0);
1474 CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, unhashed %d\n",
1475 dentry, d_unhashed(dentry));
1478 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_CREATE,
1479 ktime_us_delta(ktime_get(), kstart));
1484 static int ll_symlink(struct inode *dir, struct dentry *dchild,
1485 const char *oldpath)
1487 ktime_t kstart = ktime_get();
1491 CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p), target=%.*s\n",
1492 dchild, PFID(ll_inode2fid(dir)), dir, 3000, oldpath);
1494 err = ll_new_node(dir, dchild, oldpath, S_IFLNK | S_IRWXUGO, 0,
1495 LUSTRE_OPC_SYMLINK);
1498 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_SYMLINK,
1499 ktime_us_delta(ktime_get(), kstart));
1504 static int ll_link(struct dentry *old_dentry, struct inode *dir,
1505 struct dentry *new_dentry)
1507 struct inode *src = old_dentry->d_inode;
1508 struct qstr *name = &new_dentry->d_name;
1509 struct ll_sb_info *sbi = ll_i2sbi(dir);
1510 struct ptlrpc_request *request = NULL;
1511 struct md_op_data *op_data;
1512 ktime_t kstart = ktime_get();
1517 "VFS Op: inode="DFID"(%p), dir="DFID"(%p), target=%pd\n",
1518 PFID(ll_inode2fid(src)), src,
1519 PFID(ll_inode2fid(dir)), dir, new_dentry);
1521 err = llcrypt_prepare_link(old_dentry, dir, new_dentry);
1525 op_data = ll_prep_md_op_data(NULL, src, dir, name->name, name->len,
1526 0, LUSTRE_OPC_ANY, NULL);
1527 if (IS_ERR(op_data))
1528 RETURN(PTR_ERR(op_data));
1530 err = md_link(sbi->ll_md_exp, op_data, &request);
1531 ll_finish_md_op_data(op_data);
1535 ll_update_times(request, dir);
1536 ll_stats_ops_tally(sbi, LPROC_LL_LINK,
1537 ktime_us_delta(ktime_get(), kstart));
1540 ptlrpc_req_finished(request);
1544 static int ll_mkdir(struct inode *dir, struct dentry *dchild, umode_t mode)
1546 ktime_t kstart = ktime_get();
1550 CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p)\n",
1551 dchild, 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,
1561 ktime_us_delta(ktime_get(), kstart));
1566 static int ll_rmdir(struct inode *dir, struct dentry *dchild)
1568 struct qstr *name = &dchild->d_name;
1569 struct ptlrpc_request *request = NULL;
1570 struct md_op_data *op_data;
1571 ktime_t kstart = ktime_get();
1576 CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p)\n",
1577 dchild, PFID(ll_inode2fid(dir)), dir);
1579 if (unlikely(d_mountpoint(dchild)))
1582 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len,
1583 S_IFDIR, LUSTRE_OPC_ANY, NULL);
1584 if (IS_ERR(op_data))
1585 RETURN(PTR_ERR(op_data));
1587 if (dchild->d_inode != NULL)
1588 op_data->op_fid3 = *ll_inode2fid(dchild->d_inode);
1590 op_data->op_fid2 = op_data->op_fid3;
1591 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1592 ll_finish_md_op_data(op_data);
1594 ll_update_times(request, dir);
1596 ptlrpc_req_finished(request);
1598 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR,
1599 ktime_us_delta(ktime_get(), kstart));
1606 int ll_rmdir_entry(struct inode *dir, char *name, int namelen)
1608 struct ptlrpc_request *request = NULL;
1609 struct md_op_data *op_data;
1610 ktime_t kstart = ktime_get();
1614 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1615 namelen, name, PFID(ll_inode2fid(dir)), dir);
1617 op_data = ll_prep_md_op_data(NULL, dir, NULL, name, strlen(name),
1618 S_IFDIR, LUSTRE_OPC_ANY, NULL);
1619 if (IS_ERR(op_data))
1620 RETURN(PTR_ERR(op_data));
1621 op_data->op_cli_flags |= CLI_RM_ENTRY;
1622 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1623 ll_finish_md_op_data(op_data);
1625 ll_update_times(request, dir);
1627 ptlrpc_req_finished(request);
1629 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR,
1630 ktime_us_delta(ktime_get(), kstart));
1634 static int ll_unlink(struct inode *dir, struct dentry *dchild)
1636 struct qstr *name = &dchild->d_name;
1637 struct ptlrpc_request *request = NULL;
1638 struct md_op_data *op_data;
1639 struct mdt_body *body;
1640 ktime_t kstart = ktime_get();
1645 CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p)\n",
1646 dchild, PFID(ll_inode2fid(dir)), dir);
1649 * XXX: unlink bind mountpoint maybe call to here,
1650 * just check it as vfs_unlink does.
1652 if (unlikely(d_mountpoint(dchild)))
1655 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len, 0,
1656 LUSTRE_OPC_ANY, NULL);
1657 if (IS_ERR(op_data))
1658 RETURN(PTR_ERR(op_data));
1660 op_data->op_fid3 = *ll_inode2fid(dchild->d_inode);
1661 /* notify lower layer if inode has dirty pages */
1662 if (S_ISREG(dchild->d_inode->i_mode) &&
1663 ll_i2info(dchild->d_inode)->lli_clob &&
1664 dirty_cnt(dchild->d_inode))
1665 op_data->op_cli_flags |= CLI_DIRTY_DATA;
1666 op_data->op_fid2 = op_data->op_fid3;
1667 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1668 ll_finish_md_op_data(op_data);
1673 * The server puts attributes in on the last unlink, use them to update
1674 * the link count so the inode can be freed immediately.
1676 body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
1677 if (body->mbo_valid & OBD_MD_FLNLINK)
1678 set_nlink(dchild->d_inode, body->mbo_nlink);
1680 ll_update_times(request, dir);
1683 ptlrpc_req_finished(request);
1685 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_UNLINK,
1686 ktime_us_delta(ktime_get(), kstart));
1690 static int ll_rename(struct inode *src, struct dentry *src_dchild,
1691 struct inode *tgt, struct dentry *tgt_dchild
1692 #ifdef HAVE_IOPS_RENAME_WITH_FLAGS
1693 , unsigned int flags
1697 struct qstr *src_name = &src_dchild->d_name;
1698 struct qstr *tgt_name = &tgt_dchild->d_name;
1699 struct ptlrpc_request *request = NULL;
1700 struct ll_sb_info *sbi = ll_i2sbi(src);
1701 struct md_op_data *op_data;
1702 ktime_t kstart = ktime_get();
1706 #ifdef HAVE_IOPS_RENAME_WITH_FLAGS
1712 "VFS Op:oldname=%pd, src_dir="DFID"(%p), newname=%pd, tgt_dir="DFID"(%p)\n",
1713 src_dchild, PFID(ll_inode2fid(src)), src,
1714 tgt_dchild, PFID(ll_inode2fid(tgt)), tgt);
1716 if (unlikely(d_mountpoint(src_dchild) || d_mountpoint(tgt_dchild)))
1719 #ifdef HAVE_IOPS_RENAME_WITH_FLAGS
1720 err = llcrypt_prepare_rename(src, src_dchild, tgt, tgt_dchild, flags);
1722 err = llcrypt_prepare_rename(src, src_dchild, tgt, tgt_dchild, 0);
1727 op_data = ll_prep_md_op_data(NULL, src, tgt, NULL, 0, 0,
1728 LUSTRE_OPC_ANY, NULL);
1729 if (IS_ERR(op_data))
1730 RETURN(PTR_ERR(op_data));
1732 if (src_dchild->d_inode != NULL)
1733 op_data->op_fid3 = *ll_inode2fid(src_dchild->d_inode);
1735 if (tgt_dchild->d_inode != NULL)
1736 op_data->op_fid4 = *ll_inode2fid(tgt_dchild->d_inode);
1738 err = md_rename(sbi->ll_md_exp, op_data,
1739 src_name->name, src_name->len,
1740 tgt_name->name, tgt_name->len, &request);
1741 ll_finish_md_op_data(op_data);
1743 ll_update_times(request, src);
1744 ll_update_times(request, tgt);
1747 ptlrpc_req_finished(request);
1750 d_move(src_dchild, tgt_dchild);
1751 ll_stats_ops_tally(sbi, LPROC_LL_RENAME,
1752 ktime_us_delta(ktime_get(), kstart));
1758 const struct inode_operations ll_dir_inode_operations = {
1760 .atomic_open = ll_atomic_open,
1761 .lookup = ll_lookup_nd,
1762 .create = ll_create_nd,
1763 /* We need all these non-raw things for NFSD, to not patch it. */
1764 .unlink = ll_unlink,
1767 .symlink = ll_symlink,
1769 .rename = ll_rename,
1770 .setattr = ll_setattr,
1771 .getattr = ll_getattr,
1772 .permission = ll_inode_permission,
1773 #ifdef HAVE_IOP_XATTR
1774 .setxattr = ll_setxattr,
1775 .getxattr = ll_getxattr,
1776 .removexattr = ll_removexattr,
1778 .listxattr = ll_listxattr,
1779 .get_acl = ll_get_acl,
1780 #ifdef HAVE_IOP_SET_ACL
1781 .set_acl = ll_set_acl,
1785 const struct inode_operations ll_special_inode_operations = {
1786 .setattr = ll_setattr,
1787 .getattr = ll_getattr,
1788 .permission = ll_inode_permission,
1789 #ifdef HAVE_IOP_XATTR
1790 .setxattr = ll_setxattr,
1791 .getxattr = ll_getxattr,
1792 .removexattr = ll_removexattr,
1794 .listxattr = ll_listxattr,
1795 .get_acl = ll_get_acl,
1796 #ifdef HAVE_IOP_SET_ACL
1797 .set_acl = ll_set_acl,