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/
33 #include <linux/sched.h>
35 #include <linux/file.h>
36 #include <linux/quotaops.h>
37 #include <linux/highmem.h>
38 #include <linux/pagemap.h>
39 #include <linux/user_namespace.h>
40 #include <linux/uidgid.h>
42 #define DEBUG_SUBSYSTEM S_LLITE
44 #include <obd_support.h>
45 #include <lustre_fid.h>
46 #include <lustre_dlm.h>
47 #include "llite_internal.h"
49 #ifndef HAVE_USER_NAMESPACE_ARG
50 #define ll_create_nd(ns, dir, de, mode, ex) ll_create_nd(dir, de, mode, ex)
51 #define ll_mkdir(ns, dir, dch, mode) ll_mkdir(dir, dch, mode)
52 #define ll_mknod(ns, dir, dch, mode, rd) ll_mknod(dir, dch, mode, rd)
53 #ifdef HAVE_IOPS_RENAME_WITH_FLAGS
54 #define ll_rename(ns, src, sdc, tgt, tdc, fl) ll_rename(src, sdc, tgt, tdc, fl)
56 #define ll_rename(ns, src, sdc, tgt, tdc) ll_rename(src, sdc, tgt, tdc)
57 #endif /* HAVE_IOPS_RENAME_WITH_FLAGS */
58 #define ll_symlink(nd, dir, dch, old) ll_symlink(dir, dch, old)
61 static int ll_create_it(struct inode *dir, struct dentry *dentry,
62 struct lookup_intent *it,
63 void *secctx, __u32 secctxlen, bool encrypt,
64 void *encctx, __u32 encctxlen, unsigned int open_flags);
66 /* called from iget5_locked->find_inode() under inode_lock spinlock */
67 static int ll_test_inode(struct inode *inode, void *opaque)
69 struct ll_inode_info *lli = ll_i2info(inode);
70 struct lustre_md *md = opaque;
72 if (unlikely(!(md->body->mbo_valid & OBD_MD_FLID))) {
73 CERROR("MDS body missing FID\n");
77 if (!lu_fid_eq(&lli->lli_fid, &md->body->mbo_fid1))
83 static int ll_set_inode(struct inode *inode, void *opaque)
85 struct ll_inode_info *lli = ll_i2info(inode);
86 struct mdt_body *body = ((struct lustre_md *)opaque)->body;
88 if (unlikely(!(body->mbo_valid & OBD_MD_FLID))) {
89 CERROR("MDS body missing FID\n");
93 lli->lli_fid = body->mbo_fid1;
94 if (unlikely(!(body->mbo_valid & OBD_MD_FLTYPE))) {
95 CERROR("Can not initialize inode "DFID" without object type: "
97 PFID(&lli->lli_fid), body->mbo_valid);
101 inode->i_mode = (inode->i_mode & ~S_IFMT) | (body->mbo_mode & S_IFMT);
102 if (unlikely(inode->i_mode == 0)) {
103 CERROR("Invalid inode "DFID" type\n", PFID(&lli->lli_fid));
114 * Get an inode by inode number(@hash), which is already instantiated by
115 * the intent lookup).
117 struct inode *ll_iget(struct super_block *sb, ino_t hash,
118 struct lustre_md *md)
126 inode = iget5_locked(sb, hash, ll_test_inode, ll_set_inode, md);
128 RETURN(ERR_PTR(-ENOMEM));
130 if (inode->i_state & I_NEW) {
131 rc = ll_read_inode2(inode, md);
132 if (rc == 0 && S_ISREG(inode->i_mode) &&
133 ll_i2info(inode)->lli_clob == NULL)
134 rc = cl_file_inode_init(inode, md);
137 /* Let's clear directory lsm here, otherwise
138 * make_bad_inode() will reset the inode mode
139 * to regular, then ll_clear_inode will not
140 * be able to clear lsm_md */
141 if (S_ISDIR(inode->i_mode))
142 ll_dir_clear_lsm_md(inode);
143 make_bad_inode(inode);
144 unlock_new_inode(inode);
148 inode_has_no_xattr(inode);
149 unlock_new_inode(inode);
151 } else if (is_bad_inode(inode)) {
153 inode = ERR_PTR(-ESTALE);
154 } else if (!(inode->i_state & (I_FREEING | I_CLEAR))) {
155 rc = ll_update_inode(inode, md);
156 CDEBUG(D_VFSTRACE, "got inode: "DFID"(%p): rc = %d\n",
157 PFID(&md->body->mbo_fid1), inode, rc);
159 if (S_ISDIR(inode->i_mode))
160 ll_dir_clear_lsm_md(inode);
169 /* mark negative sub file dentries invalid and prune unused dentries */
170 static void ll_prune_negative_children(struct inode *dir)
172 struct dentry *dentry;
173 struct dentry *child;
178 spin_lock(&dir->i_lock);
179 hlist_for_each_entry(dentry, &dir->i_dentry, d_alias) {
180 spin_lock(&dentry->d_lock);
181 list_for_each_entry(child, &dentry->d_subdirs, d_child) {
185 spin_lock_nested(&child->d_lock, DENTRY_D_LOCK_NESTED);
186 if (lld_is_init(child))
187 ll_d2d(child)->lld_invalid = 1;
188 if (!ll_d_count(child)) {
191 spin_unlock(&child->d_lock);
192 spin_unlock(&dentry->d_lock);
193 spin_unlock(&dir->i_lock);
195 CDEBUG(D_DENTRY, "prune negative dentry %pd\n",
201 spin_unlock(&child->d_lock);
203 spin_unlock(&dentry->d_lock);
205 spin_unlock(&dir->i_lock);
210 int ll_test_inode_by_fid(struct inode *inode, void *opaque)
212 return lu_fid_eq(&ll_i2info(inode)->lli_fid, opaque);
215 static int ll_dom_lock_cancel(struct inode *inode, struct ldlm_lock *lock)
218 struct ll_inode_info *lli = ll_i2info(inode);
223 env = cl_env_get(&refcheck);
225 RETURN(PTR_ERR(env));
227 CFS_FAIL_TIMEOUT(OBD_FAIL_LDLM_REPLAY_PAUSE, cfs_fail_val);
229 /* reach MDC layer to flush data under the DoM ldlm lock */
230 rc = cl_object_flush(env, lli->lli_clob, lock);
231 if (rc == -ENODATA) {
232 CDEBUG(D_INODE, "inode "DFID" layout has no DoM stripe\n",
233 PFID(ll_inode2fid(inode)));
234 /* most likely result of layout change, do nothing */
238 cl_env_put(env, &refcheck);
242 static void ll_lock_cancel_bits(struct ldlm_lock *lock, __u64 to_cancel)
244 struct inode *inode = ll_inode_from_resource_lock(lock);
245 struct ll_inode_info *lli;
246 __u64 bits = to_cancel;
252 /* That means the inode is evicted most likely and may cause
253 * the skipping of lock cleanups below, so print the message
256 if (lock->l_resource->lr_lvb_inode)
258 "can't take inode for the lock (%sevicted)",
259 lock->l_resource->lr_lvb_inode->i_state &
260 I_FREEING ? "" : "not ");
264 if (!fid_res_name_eq(ll_inode2fid(inode),
265 &lock->l_resource->lr_name)) {
266 LDLM_ERROR(lock, "data mismatch with object "DFID"(%p)",
267 PFID(ll_inode2fid(inode)), inode);
271 if (bits & MDS_INODELOCK_XATTR) {
272 ll_xattr_cache_empty(inode);
273 bits &= ~MDS_INODELOCK_XATTR;
276 /* For OPEN locks we differentiate between lock modes
277 * LCK_CR, LCK_CW, LCK_PR - bug 22891 */
278 if (bits & MDS_INODELOCK_OPEN)
279 ll_have_md_lock(lock->l_conn_export, inode, &bits,
282 if (bits & MDS_INODELOCK_OPEN) {
285 switch (lock->l_req_mode) {
296 LDLM_ERROR(lock, "bad lock mode for OPEN lock");
300 ll_md_real_close(inode, fmode);
302 bits &= ~MDS_INODELOCK_OPEN;
305 if (bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
306 MDS_INODELOCK_LAYOUT | MDS_INODELOCK_PERM |
308 ll_have_md_lock(lock->l_conn_export, inode, &bits, LCK_MINMODE);
310 if (bits & MDS_INODELOCK_DOM) {
311 rc = ll_dom_lock_cancel(inode, lock);
313 CDEBUG(D_INODE, "cannot flush DoM data "
315 PFID(ll_inode2fid(inode)), rc);
318 if (bits & MDS_INODELOCK_LAYOUT) {
319 struct cl_object_conf conf = {
320 .coc_opc = OBJECT_CONF_INVALIDATE,
324 rc = ll_layout_conf(inode, &conf);
326 CDEBUG(D_INODE, "cannot invalidate layout of "
328 PFID(ll_inode2fid(inode)), rc);
331 lli = ll_i2info(inode);
333 if (bits & MDS_INODELOCK_UPDATE)
334 set_bit(LLIF_UPDATE_ATIME, &lli->lli_flags);
336 if ((bits & MDS_INODELOCK_UPDATE) && S_ISDIR(inode->i_mode)) {
337 CDEBUG(D_INODE, "invalidating inode "DFID" lli = %p, "
338 "pfid = "DFID"\n", PFID(ll_inode2fid(inode)),
339 lli, PFID(&lli->lli_pfid));
340 truncate_inode_pages(inode->i_mapping, 0);
342 if (unlikely(!fid_is_zero(&lli->lli_pfid))) {
343 struct inode *master_inode = NULL;
346 /* This is slave inode, since all of the child dentry
347 * is connected on the master inode, so we have to
348 * invalidate the negative children on master inode */
349 CDEBUG(D_INODE, "Invalidate s"DFID" m"DFID"\n",
350 PFID(ll_inode2fid(inode)), PFID(&lli->lli_pfid));
352 hash = cl_fid_build_ino(&lli->lli_pfid,
353 ll_need_32bit_api(ll_i2sbi(inode)));
355 /* Do not lookup the inode with ilookup5, otherwise
356 * it will cause dead lock,
357 * 1. Client1 send chmod req to the MDT0, then on MDT0,
358 * it enqueues master and all of its slaves lock,
359 * (mdt_attr_set() -> mdt_lock_slaves()), after gets
360 * master and stripe0 lock, it will send the enqueue
361 * req (for stripe1) to MDT1, then MDT1 finds the lock
362 * has been granted to client2. Then MDT1 sends blocking
364 * 2. At the same time, client2 tries to unlink
365 * the striped dir (rm -rf striped_dir), and during
366 * lookup, it will hold the master inode of the striped
367 * directory, whose inode state is NEW, then tries to
368 * revalidate all of its slaves, (ll_prep_inode()->
369 * ll_iget()->ll_read_inode2()-> ll_update_inode().).
370 * And it will be blocked on the server side because
372 * 3. Then the client get the blocking_ast req, cancel
373 * the lock, but being blocked if using ->ilookup5()),
374 * because master inode state is NEW. */
375 master_inode = ilookup5_nowait(inode->i_sb, hash,
376 ll_test_inode_by_fid,
377 (void *)&lli->lli_pfid);
379 ll_prune_negative_children(master_inode);
383 ll_prune_negative_children(inode);
387 /* at umount s_root becomes NULL */
388 if ((bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM)) &&
389 inode->i_sb->s_root && !is_root_inode(inode))
390 ll_prune_aliases(inode);
392 if (bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM))
393 forget_all_cached_acls(inode);
399 /* Check if the given lock may be downgraded instead of canceling and
400 * that convert is really needed. */
401 static int ll_md_need_convert(struct ldlm_lock *lock)
403 struct ldlm_namespace *ns = ldlm_lock_to_ns(lock);
405 __u64 wanted = lock->l_policy_data.l_inodebits.cancel_bits;
406 __u64 bits = lock->l_policy_data.l_inodebits.bits & ~wanted;
407 enum ldlm_mode mode = LCK_MINMODE;
409 if (!lock->l_conn_export ||
410 !exp_connect_lock_convert(lock->l_conn_export))
413 if (!wanted || !bits || ldlm_is_cancel(lock))
416 /* do not convert locks other than DOM for now */
417 if (!((bits | wanted) & MDS_INODELOCK_DOM))
420 /* We may have already remaining bits in some other lock so
421 * lock convert will leave us just extra lock for the same bit.
422 * Check if client has other lock with the same bits and the same
423 * or lower mode and don't convert if any.
425 switch (lock->l_req_mode) {
439 /* do not convert other modes */
443 /* is lock is too old to be converted? */
444 lock_res_and_lock(lock);
445 if (ktime_after(ktime_get(),
446 ktime_add(lock->l_last_used, ns->ns_dirty_age_limit))) {
447 unlock_res_and_lock(lock);
450 unlock_res_and_lock(lock);
452 inode = ll_inode_from_resource_lock(lock);
453 ll_have_md_lock(lock->l_conn_export, inode, &bits, mode);
458 int ll_md_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *ld,
459 void *data, int flag)
461 struct lustre_handle lockh;
467 case LDLM_CB_BLOCKING:
469 __u64 cancel_flags = LCF_ASYNC;
471 /* if lock convert is not needed then still have to
472 * pass lock via ldlm_cli_convert() to keep all states
473 * correct, set cancel_bits to full lock bits to cause
474 * full cancel to happen.
476 if (!ll_md_need_convert(lock)) {
477 lock_res_and_lock(lock);
478 lock->l_policy_data.l_inodebits.cancel_bits =
479 lock->l_policy_data.l_inodebits.bits;
480 unlock_res_and_lock(lock);
482 rc = ldlm_cli_convert(lock, cancel_flags);
485 /* continue with cancel otherwise */
486 ldlm_lock2handle(lock, &lockh);
487 rc = ldlm_cli_cancel(&lockh, cancel_flags);
489 CDEBUG(D_INODE, "ldlm_cli_cancel: rc = %d\n", rc);
494 case LDLM_CB_CANCELING:
496 __u64 to_cancel = lock->l_policy_data.l_inodebits.bits;
498 /* Nothing to do for non-granted locks */
499 if (!ldlm_is_granted(lock))
502 /* If 'ld' is supplied then bits to be cancelled are passed
503 * implicitly by lock converting and cancel_bits from 'ld'
504 * should be used. Otherwise full cancel is being performed
505 * and lock inodebits are used.
507 * Note: we cannot rely on cancel_bits in lock itself at this
508 * moment because they can be changed by concurrent thread,
509 * so ldlm_cli_inodebits_convert() pass cancel bits implicitly
513 /* partial bits cancel allowed only during convert */
514 LASSERT(ldlm_is_converting(lock));
515 /* mask cancel bits by lock bits so only no any unused
516 * bits are passed to ll_lock_cancel_bits()
518 to_cancel &= ld->l_policy_data.l_inodebits.cancel_bits;
520 ll_lock_cancel_bits(lock, to_cancel);
530 __u32 ll_i2suppgid(struct inode *i)
532 if (in_group_p(i->i_gid))
533 return (__u32)from_kgid(&init_user_ns, i->i_gid);
535 return (__u32) __kgid_val(INVALID_GID);
538 /* Pack the required supplementary groups into the supplied groups array. */
539 void ll_i2gids(__u32 *suppgids, struct inode *i1, struct inode *i2)
542 LASSERT(suppgids != NULL);
544 suppgids[0] = ll_i2suppgid(i1);
547 suppgids[1] = ll_i2suppgid(i2);
553 * try to reuse three types of dentry:
554 * 1. unhashed alias, this one is unhashed by d_invalidate (but it may be valid
555 * by concurrent .revalidate).
556 * 2. INVALID alias (common case for no valid ldlm lock held, but this flag may
557 * be cleared by others calling d_lustre_revalidate).
558 * 3. DISCONNECTED alias.
560 static struct dentry *ll_find_alias(struct inode *inode, struct dentry *dentry)
562 struct dentry *alias, *discon_alias, *invalid_alias;
564 if (hlist_empty(&inode->i_dentry))
567 discon_alias = invalid_alias = NULL;
569 spin_lock(&inode->i_lock);
570 hlist_for_each_entry(alias, &inode->i_dentry, d_alias) {
571 LASSERT(alias != dentry);
573 spin_lock(&alias->d_lock);
574 if ((alias->d_flags & DCACHE_DISCONNECTED) &&
575 S_ISDIR(inode->i_mode))
576 /* LASSERT(last_discon == NULL); LU-405, bz 20055 */
577 discon_alias = alias;
578 else if (alias->d_parent == dentry->d_parent &&
579 alias->d_name.hash == dentry->d_name.hash &&
580 alias->d_name.len == dentry->d_name.len &&
581 memcmp(alias->d_name.name, dentry->d_name.name,
582 dentry->d_name.len) == 0)
583 invalid_alias = alias;
584 spin_unlock(&alias->d_lock);
589 alias = invalid_alias ?: discon_alias ?: NULL;
591 spin_lock(&alias->d_lock);
593 spin_unlock(&alias->d_lock);
595 spin_unlock(&inode->i_lock);
601 * Similar to d_splice_alias(), but lustre treats invalid alias
602 * similar to DCACHE_DISCONNECTED, and tries to use it anyway.
604 struct dentry *ll_splice_alias(struct inode *inode, struct dentry *de)
609 new = ll_find_alias(inode, de);
611 if (!ll_d_setup(new, true))
612 return ERR_PTR(-ENOMEM);
616 "Reuse dentry %p inode %p refc %d flags %#x\n",
617 new, new->d_inode, ll_d_count(new), new->d_flags);
621 if (!ll_d_setup(de, false))
622 return ERR_PTR(-ENOMEM);
625 /* this needs only to be done for foreign symlink dirs as
626 * DCACHE_SYMLINK_TYPE is already set by d_flags_for_inode()
627 * kernel routine for files with symlink ops (ie, real symlink)
629 if (inode && S_ISDIR(inode->i_mode) &&
630 ll_sbi_has_foreign_symlink(ll_i2sbi(inode)) &&
631 #ifdef HAVE_IOP_GET_LINK
632 inode->i_op->get_link) {
634 inode->i_op->follow_link) {
636 CDEBUG(D_INFO, "%s: inode "DFID": faking foreign dir as a symlink\n",
637 ll_i2sbi(inode)->ll_fsname, PFID(ll_inode2fid(inode)));
638 spin_lock(&de->d_lock);
639 /* like d_flags_for_inode() already does for files */
640 de->d_flags = (de->d_flags & ~DCACHE_ENTRY_TYPE) |
642 spin_unlock(&de->d_lock);
645 CDEBUG(D_DENTRY, "Add dentry %p inode %p refc %d flags %#x\n",
646 de, de->d_inode, ll_d_count(de), de->d_flags);
650 static int ll_lookup_it_finish(struct ptlrpc_request *request,
651 struct lookup_intent *it,
652 struct inode *parent, struct dentry **de,
653 void *secctx, __u32 secctxlen,
654 void *encctx, __u32 encctxlen,
655 ktime_t kstart, bool encrypt)
657 struct inode *inode = NULL;
660 struct dentry *alias;
663 /* NB 1 request reference will be taken away by ll_intent_lock()
665 CDEBUG(D_DENTRY, "it %p it_disposition %x\n", it,
667 if (!it_disposition(it, DISP_LOOKUP_NEG)) {
668 struct req_capsule *pill = &request->rq_pill;
669 struct mdt_body *body = req_capsule_server_get(pill,
672 rc = ll_prep_inode(&inode, &request->rq_pill, (*de)->d_sb, it);
676 /* If encryption context was returned by MDT, put it in
677 * inode now to save an extra getxattr and avoid deadlock.
679 if (body->mbo_valid & OBD_MD_ENCCTX) {
680 encctx = req_capsule_server_get(pill, &RMF_FILE_ENCCTX);
681 encctxlen = req_capsule_get_size(pill,
687 "server returned encryption ctx for "DFID"\n",
688 PFID(ll_inode2fid(inode)));
689 rc = ll_xattr_cache_insert(inode,
690 xattr_for_enc(inode),
693 CWARN("%s: cannot set enc ctx for "DFID": rc = %d\n",
694 ll_i2sbi(inode)->ll_fsname,
695 PFID(ll_inode2fid(inode)), rc);
697 rc = llcrypt_prepare_readdir(inode);
700 "cannot get enc info for "DFID": rc = %d\n",
701 PFID(ll_inode2fid(inode)), rc);
706 ll_set_lock_data(ll_i2sbi(parent)->ll_md_exp, inode, it, &bits);
707 /* OPEN can return data if lock has DoM+LAYOUT bits set */
708 if (it->it_op & IT_OPEN &&
709 bits & MDS_INODELOCK_DOM && bits & MDS_INODELOCK_LAYOUT)
710 ll_dom_finish_open(inode, request);
712 /* We used to query real size from OSTs here, but actually
713 * this is not needed. For stat() calls size would be updated
714 * from subsequent do_revalidate()->ll_inode_revalidate_it() in
716 * vfs_getattr_it->ll_getattr()->ll_inode_revalidate_it() in 2.6
717 * Everybody else who needs correct file size would call
718 * ll_glimpse_size or some equivalent themselves anyway.
722 /* If security context was returned by MDT, put it in
723 * inode now to save an extra getxattr from security hooks,
724 * and avoid deadlock.
726 if (body->mbo_valid & OBD_MD_SECCTX) {
727 secctx = req_capsule_server_get(pill, &RMF_FILE_SECCTX);
728 secctxlen = req_capsule_get_size(pill,
733 CDEBUG(D_SEC, "server returned security context"
735 PFID(ll_inode2fid(inode)));
738 /* resume normally on error */
739 ll_inode_notifysecctx(inode, secctx, secctxlen);
742 /* Only hash *de if it is unhashed (new dentry).
743 * Atoimc_open may passin hashed dentries for open.
745 alias = ll_splice_alias(inode, *de);
747 GOTO(out, rc = PTR_ERR(alias));
751 if (!it_disposition(it, DISP_LOOKUP_NEG)) {
752 /* We have the "lookup" lock, so unhide dentry */
753 if (bits & MDS_INODELOCK_LOOKUP)
754 d_lustre_revalidate(*de);
755 /* open may not fetch LOOKUP lock, update dir depth/dmv anyway
756 * in case it's used uninitialized.
758 if (S_ISDIR(inode->i_mode))
759 ll_update_dir_depth_dmv(parent, *de);
762 rc = llcrypt_prepare_readdir(inode);
766 } else if (!it_disposition(it, DISP_OPEN_CREATE)) {
768 * If file was created on the server, the dentry is revalidated
769 * in ll_create_it if the lock allows for it.
771 /* Check that parent has UPDATE lock. */
772 struct lookup_intent parent_it = {
774 .it_lock_handle = 0 };
775 struct lu_fid fid = ll_i2info(parent)->lli_fid;
777 /* If it is striped directory, get the real stripe parent */
778 if (unlikely(ll_dir_striped(parent))) {
779 down_read(&ll_i2info(parent)->lli_lsm_sem);
780 rc = md_get_fid_from_lsm(ll_i2mdexp(parent),
781 ll_i2info(parent)->lli_lsm_obj,
783 (*de)->d_name.len, &fid);
784 up_read(&ll_i2info(parent)->lli_lsm_sem);
789 if (md_revalidate_lock(ll_i2mdexp(parent), &parent_it, &fid,
791 d_lustre_revalidate(*de);
792 ll_intent_release(&parent_it);
796 if (it_disposition(it, DISP_OPEN_CREATE)) {
797 ll_stats_ops_tally(ll_i2sbi(parent), LPROC_LL_MKNOD,
798 ktime_us_delta(ktime_get(), kstart));
804 if (rc != 0 && it->it_op & IT_OPEN) {
805 ll_intent_drop_lock(it);
806 ll_open_cleanup((*de)->d_sb, &request->rq_pill);
812 static struct dentry *ll_lookup_it(struct inode *parent, struct dentry *dentry,
813 struct lookup_intent *it,
814 void **secctx, __u32 *secctxlen,
816 struct pcc_create_attach *pca,
818 void **encctx, __u32 *encctxlen)
820 ktime_t kstart = ktime_get();
821 struct lookup_intent lookup_it = { .it_op = IT_LOOKUP };
822 struct dentry *save = dentry, *retval;
823 struct ptlrpc_request *req = NULL;
824 struct md_op_data *op_data = NULL;
825 struct lov_user_md *lum = NULL;
826 struct ll_sb_info *sbi = ll_i2sbi(parent);
829 struct llcrypt_name fname;
833 if (dentry->d_name.len > sbi->ll_namelen)
834 RETURN(ERR_PTR(-ENAMETOOLONG));
836 CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p), intent=%s\n",
837 dentry, PFID(ll_inode2fid(parent)), parent, LL_IT2STR(it));
839 if (d_mountpoint(dentry))
840 CERROR("Tell Peter, lookup on mtpt, it %s\n", LL_IT2STR(it));
842 if (it == NULL || it->it_op == IT_GETXATTR)
845 if (it->it_op == IT_GETATTR && dentry_may_statahead(parent, dentry)) {
846 rc = ll_revalidate_statahead(parent, &dentry, 0);
848 RETURN(dentry == save ? NULL : dentry);
851 if (it->it_op & IT_OPEN && it->it_flags & FMODE_WRITE &&
852 dentry->d_sb->s_flags & SB_RDONLY)
853 RETURN(ERR_PTR(-EROFS));
855 if (it->it_op & IT_CREAT)
856 opc = LUSTRE_OPC_CREATE;
858 opc = LUSTRE_OPC_LOOKUP;
860 rc = ll_prepare_lookup(parent, dentry, &fname, &fid);
862 RETURN(rc != -ENOENT ? ERR_PTR(rc) : NULL);
864 op_data = ll_prep_md_op_data(NULL, parent, NULL, fname.disk_name.name,
865 fname.disk_name.len, 0, opc, NULL);
866 if (IS_ERR(op_data)) {
867 llcrypt_free_filename(&fname);
868 RETURN(ERR_CAST(op_data));
870 if (!fid_is_zero(&fid)) {
871 op_data->op_fid2 = fid;
872 op_data->op_bias = MDS_FID_OP;
873 if (it->it_op & IT_OPEN)
874 it->it_flags |= MDS_OPEN_BY_FID;
877 /* enforce umask if acl disabled or MDS doesn't support umask */
878 if (!IS_POSIXACL(parent) || !exp_connect_umask(ll_i2mdexp(parent)))
879 it->it_create_mode &= ~current_umask();
881 if (it->it_op & IT_CREAT &&
882 test_bit(LL_SBI_FILE_SECCTX, sbi->ll_flags)) {
883 rc = ll_dentry_init_security(dentry, it->it_create_mode,
885 &op_data->op_file_secctx_name,
886 &op_data->op_file_secctx_name_size,
887 &op_data->op_file_secctx,
888 &op_data->op_file_secctx_size,
889 &op_data->op_file_secctx_slot);
891 GOTO(out, retval = ERR_PTR(rc));
893 *secctx = op_data->op_file_secctx;
894 if (secctxlen != NULL)
895 *secctxlen = op_data->op_file_secctx_size;
896 if (secctxslot != NULL)
897 *secctxslot = op_data->op_file_secctx_slot;
901 if (secctxlen != NULL)
903 if (secctxslot != NULL)
906 if (it->it_op & IT_CREAT && encrypt) {
907 if (unlikely(filename_is_volatile(dentry->d_name.name,
908 dentry->d_name.len, NULL))) {
909 /* get encryption context from reference file */
910 int ctx_size = LLCRYPT_ENC_CTX_SIZE;
911 struct lustre_sb_info *lsi;
912 struct file *ref_file;
913 struct inode *ref_inode;
916 rc = volatile_ref_file(dentry->d_name.name,
920 GOTO(out, retval = ERR_PTR(rc));
922 ref_inode = file_inode(ref_file);
925 GOTO(inherit, rc = -EINVAL);
928 lsi = s2lsi(ref_inode->i_sb);
931 OBD_ALLOC(ctx, ctx_size);
933 GOTO(out, retval = ERR_PTR(-ENOMEM));
935 #ifdef CONFIG_LL_ENCRYPTION
936 rc = lsi->lsi_cop->get_context(ref_inode,
938 #elif defined(HAVE_LUSTRE_CRYPTO)
939 rc = ref_inode->i_sb->s_cop->get_context(ref_inode,
945 OBD_FREE(ctx, ctx_size);
951 OBD_FREE(ctx, ctx_size);
955 op_data->op_file_encctx_size = rc;
956 if (rc == ctx_size) {
957 op_data->op_file_encctx = ctx;
959 OBD_ALLOC(op_data->op_file_encctx,
960 op_data->op_file_encctx_size);
961 if (!op_data->op_file_encctx) {
962 OBD_FREE(ctx, ctx_size);
963 GOTO(out, retval = ERR_PTR(-ENOMEM));
965 memcpy(op_data->op_file_encctx, ctx,
966 op_data->op_file_encctx_size);
967 OBD_FREE(ctx, ctx_size);
971 rc = llcrypt_inherit_context(parent, NULL, op_data,
974 GOTO(out, retval = ERR_PTR(rc));
977 *encctx = op_data->op_file_encctx;
978 if (encctxlen != NULL)
979 *encctxlen = op_data->op_file_encctx_size;
983 if (encctxlen != NULL)
987 /* ask for security context upon intent:
988 * get name of security xattr to request to server
990 if (it->it_op & (IT_LOOKUP | IT_GETATTR | IT_OPEN))
991 op_data->op_file_secctx_name_size =
992 ll_secctx_name_get(sbi, &op_data->op_file_secctx_name);
994 if (pca && pca->pca_dataset) {
997 GOTO(out, retval = ERR_PTR(-ENOMEM));
999 lum->lmm_magic = LOV_USER_MAGIC_V1;
1000 lum->lmm_pattern = LOV_PATTERN_F_RELEASED | LOV_PATTERN_RAID0;
1001 op_data->op_data = lum;
1002 op_data->op_data_size = sizeof(*lum);
1003 op_data->op_archive_id = pca->pca_dataset->pccd_rwid;
1004 it->it_flags |= MDS_OPEN_PCC;
1007 rc = md_intent_lock(ll_i2mdexp(parent), op_data, it, &req,
1008 &ll_md_blocking_ast, 0);
1009 /* If the MDS allows the client to chgrp (CFS_SETGRP_PERM), but the
1010 * client does not know which suppgid should be sent to the MDS, or
1011 * some other(s) changed the target file's GID after this RPC sent
1012 * to the MDS with the suppgid as the original GID, then we should
1013 * try again with right suppgid. */
1014 if (rc == -EACCES && it->it_op & IT_OPEN &&
1015 it_disposition(it, DISP_OPEN_DENY)) {
1016 struct mdt_body *body;
1018 LASSERT(req != NULL);
1020 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1021 if (op_data->op_suppgids[0] == body->mbo_gid ||
1022 op_data->op_suppgids[1] == body->mbo_gid ||
1023 !in_group_p(make_kgid(&init_user_ns, body->mbo_gid)))
1024 GOTO(out, retval = ERR_PTR(-EACCES));
1026 fid_zero(&op_data->op_fid2);
1027 op_data->op_suppgids[1] = body->mbo_gid;
1028 ptlrpc_req_finished(req);
1030 ll_intent_release(it);
1031 rc = md_intent_lock(ll_i2mdexp(parent), op_data, it, &req,
1032 &ll_md_blocking_ast, 0);
1036 GOTO(out, retval = ERR_PTR(rc));
1038 if (pca && pca->pca_dataset) {
1039 rc = pcc_inode_create(parent->i_sb, pca->pca_dataset,
1043 GOTO(out, retval = ERR_PTR(rc));
1046 /* dir layout may change */
1047 ll_unlock_md_op_lsm(op_data);
1048 rc = ll_lookup_it_finish(req, it, parent, &dentry,
1049 secctx != NULL ? *secctx : NULL,
1050 secctxlen != NULL ? *secctxlen : 0,
1051 encctx != NULL ? *encctx : NULL,
1052 encctxlen != NULL ? *encctxlen : 0,
1055 ll_intent_release(it);
1056 GOTO(out, retval = ERR_PTR(rc));
1059 if ((it->it_op & IT_OPEN) && dentry->d_inode &&
1060 !S_ISREG(dentry->d_inode->i_mode) &&
1061 !S_ISDIR(dentry->d_inode->i_mode)) {
1062 ll_release_openhandle(dentry, it);
1064 ll_lookup_finish_locks(it, dentry);
1066 GOTO(out, retval = (dentry == save) ? NULL : dentry);
1069 if (op_data != NULL && !IS_ERR(op_data)) {
1070 if (secctx != NULL && secctxlen != NULL) {
1071 /* caller needs sec ctx info, so reset it in op_data to
1072 * prevent it from being freed */
1073 op_data->op_file_secctx = NULL;
1074 op_data->op_file_secctx_size = 0;
1076 if (encctx != NULL && encctxlen != NULL &&
1077 it->it_op & IT_CREAT && encrypt) {
1078 /* caller needs enc ctx info, so reset it in op_data to
1079 * prevent it from being freed
1081 op_data->op_file_encctx = NULL;
1082 op_data->op_file_encctx_size = 0;
1084 llcrypt_free_filename(&fname);
1085 ll_finish_md_op_data(op_data);
1091 ptlrpc_req_finished(req);
1095 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
1098 struct lookup_intent *itp, it = { .it_op = IT_GETATTR };
1099 struct dentry *de = NULL;
1101 /* VFS has locked the inode before calling this */
1102 ll_set_inode_lock_owner(parent);
1104 CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p), flags=%u\n",
1105 dentry, PFID(ll_inode2fid(parent)), parent, flags);
1108 * Optimize away (CREATE && !OPEN). Let .create handle the race.
1109 * but only if we have write permissions there, otherwise we need
1110 * to proceed with lookup. LU-4185
1112 if ((flags & LOOKUP_CREATE) && !(flags & LOOKUP_OPEN) &&
1113 (inode_permission(&nop_mnt_idmap,
1114 parent, MAY_WRITE | MAY_EXEC) == 0))
1117 if (flags & (LOOKUP_PARENT|LOOKUP_OPEN|LOOKUP_CREATE))
1121 de = ll_lookup_it(parent, dentry, itp, NULL, NULL, NULL, NULL, false,
1125 ll_intent_release(itp);
1128 ll_clear_inode_lock_owner(parent);
1133 #ifdef FMODE_CREATED /* added in Linux v4.18-rc1-20-g73a09dd */
1134 # define ll_is_opened(o, f) ((f)->f_mode & FMODE_OPENED)
1135 # define ll_finish_open(f, d, o) finish_open((f), (d), NULL)
1136 # define ll_last_arg
1137 # define ll_set_created(o, f) \
1139 (f)->f_mode |= FMODE_CREATED; \
1143 # define ll_is_opened(o, f) (*(o))
1144 # define ll_finish_open(f, d, o) finish_open((f), (d), NULL, (o))
1145 # define ll_last_arg , int *opened
1146 # define ll_set_created(o, f) \
1148 *(o) |= FILE_CREATED; \
1154 * For cached negative dentry and new dentry, handle lookup/create/open
1157 static int ll_atomic_open(struct inode *dir, struct dentry *dentry,
1158 struct file *file, unsigned open_flags,
1159 umode_t mode ll_last_arg)
1161 struct lookup_intent *it;
1163 long long lookup_flags = LOOKUP_OPEN;
1164 void *secctx = NULL;
1165 __u32 secctxlen = 0;
1167 void *encctx = NULL;
1168 __u32 encctxlen = 0;
1169 struct ll_sb_info *sbi = NULL;
1170 struct pcc_create_attach pca = { NULL, NULL };
1171 bool encrypt = false;
1176 /* VFS has locked the inode before calling this */
1177 ll_set_inode_lock_owner(dir);
1180 "VFS Op:name=%pd, dir="DFID"(%p), file %p, open_flags %x, mode %x opened %d\n",
1181 dentry, PFID(ll_inode2fid(dir)), dir, file, open_flags, mode,
1182 ll_is_opened(opened, file));
1184 /* Only negative dentries enter here */
1185 LASSERT(dentry->d_inode == NULL);
1187 if (!d_unhashed(dentry)) {
1188 /* A valid negative dentry that just passed revalidation,
1189 * there's little point to try and open it server-side,
1190 * even though there's a minuscule chance it might succeed.
1191 * Either way it's a valid race to just return -ENOENT here.
1193 if (!(open_flags & O_CREAT))
1194 GOTO(clear, rc = -ENOENT);
1196 /* Otherwise we just unhash it to be rehashed afresh via
1197 * lookup if necessary
1202 OBD_ALLOC(it, sizeof(*it));
1204 GOTO(clear, rc = -ENOMEM);
1206 it->it_op = IT_OPEN;
1207 if (open_flags & O_CREAT) {
1208 it->it_op |= IT_CREAT;
1209 lookup_flags |= LOOKUP_CREATE;
1210 sbi = ll_i2sbi(dir);
1211 /* Volatile file is used for HSM restore, so do not use PCC */
1212 if (!filename_is_volatile(dentry->d_name.name,
1213 dentry->d_name.len, NULL)) {
1214 struct pcc_matcher item;
1215 struct pcc_dataset *dataset;
1217 item.pm_uid = from_kuid(&init_user_ns, current_uid());
1218 item.pm_gid = from_kgid(&init_user_ns, current_gid());
1219 item.pm_projid = ll_i2info(dir)->lli_projid;
1220 item.pm_name = &dentry->d_name;
1221 dataset = pcc_dataset_match_get(&sbi->ll_pcc_super,
1223 pca.pca_dataset = dataset;
1226 it->it_create_mode = (mode & S_IALLUGO) | S_IFREG;
1227 it->it_flags = (open_flags & ~O_ACCMODE) | OPEN_FMODE(open_flags);
1228 it->it_flags &= ~MDS_OPEN_FL_INTERNAL;
1230 if (ll_sbi_has_encrypt(ll_i2sbi(dir)) && IS_ENCRYPTED(dir)) {
1231 /* in case of create, this is going to be a regular file because
1232 * we set S_IFREG bit on it->it_create_mode above
1234 rc = llcrypt_prepare_readdir(dir);
1236 GOTO(out_release, rc);
1238 if (open_flags & O_CREAT) {
1239 /* For migration or mirroring without enc key, we still
1240 * need to be able to create a volatile file.
1242 if (!llcrypt_has_encryption_key(dir) &&
1243 (!filename_is_volatile(dentry->d_name.name,
1244 dentry->d_name.len, NULL) ||
1245 (open_flags & O_FILE_ENC) != O_FILE_ENC ||
1246 !(open_flags & O_DIRECT)))
1247 GOTO(out_release, rc = -ENOKEY);
1251 CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_FILE_PAUSE2, cfs_fail_val);
1253 /* We can only arrive at this path when we have no inode, so
1254 * we only need to request open lock if it was requested
1257 if (ll_i2info(dir)->lli_open_thrsh_count != UINT_MAX)
1258 open_threshold = ll_i2info(dir)->lli_open_thrsh_count;
1260 open_threshold = ll_i2sbi(dir)->ll_oc_thrsh_count;
1262 if (open_threshold == 1 &&
1263 exp_connect_flags2(ll_i2mdexp(dir)) &
1264 OBD_CONNECT2_ATOMIC_OPEN_LOCK)
1265 it->it_flags |= MDS_OPEN_LOCK;
1267 /* Dentry added to dcache tree in ll_lookup_it */
1268 de = ll_lookup_it(dir, dentry, it, &secctx, &secctxlen, &secctxslot,
1269 &pca, encrypt, &encctx, &encctxlen);
1272 else if (de != NULL)
1275 CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_FILE_PAUSE, cfs_fail_val);
1278 if (it_disposition(it, DISP_OPEN_CREATE)) {
1279 /* Dentry instantiated in ll_create_it. */
1280 rc = ll_create_it(dir, dentry, it, secctx, secctxlen,
1281 encrypt, encctx, encctxlen,
1283 ll_security_release_secctx(secctx, secctxlen,
1285 llcrypt_free_ctx(encctx, encctxlen);
1287 /* We dget in ll_splice_alias. */
1293 rc = pcc_inode_create_fini(dentry->d_inode, &pca);
1297 GOTO(out_release, rc);
1300 ll_set_created(opened, file);
1302 /* Open the file with O_CREAT, but the file already
1303 * existed on MDT. This may happend in the case that
1304 * the LOOKUP ibits lock is revoked and the
1305 * corresponding dentry cache is deleted.
1306 * i.e. In the current Lustre, the truncate operation
1307 * will revoke the LOOKUP ibits lock, and the file
1308 * dentry cache will be invalidated. The following open
1309 * with O_CREAT flag will call into ->atomic_open, the
1310 * file was wrongly though as newly created file and
1311 * try to auto cache the file. So after client knows it
1312 * is not a DISP_OPEN_CREATE, it should cleanup the
1313 * already created PCC copy.
1315 pcc_create_attach_cleanup(dir->i_sb, &pca);
1317 if (open_flags & O_CREAT && encrypt &&
1319 rc = ll_set_encflags(dentry->d_inode, encctx,
1321 llcrypt_free_ctx(encctx, encctxlen);
1323 GOTO(out_release, rc);
1327 /* check also if a foreign file is openable */
1328 if (dentry->d_inode && it_disposition(it, DISP_OPEN_OPEN) &&
1329 ll_foreign_is_openable(dentry, open_flags)) {
1331 if (S_ISFIFO(dentry->d_inode->i_mode)) {
1332 /* We cannot call open here as it might
1333 * deadlock. This case is unreachable in
1334 * practice because of OBD_CONNECT_NODEVOH. */
1335 rc = finish_no_open(file, de);
1337 file->private_data = it;
1338 rc = ll_finish_open(file, dentry, opened);
1339 /* We dget in ll_splice_alias. finish_open takes
1340 * care of dget for fd open.
1346 rc = finish_no_open(file, de);
1349 pcc_create_attach_cleanup(dir->i_sb, &pca);
1353 ll_intent_release(it);
1354 OBD_FREE(it, sizeof(*it));
1356 ll_clear_inode_lock_owner(dir);
1361 /* We depend on "mode" being set with the proper file type/umask by now */
1362 static struct inode *ll_create_node(struct inode *dir, struct lookup_intent *it)
1364 struct inode *inode = NULL;
1365 struct ptlrpc_request *request = NULL;
1366 struct ll_sb_info *sbi = ll_i2sbi(dir);
1370 LASSERT(it && it->it_disposition);
1372 LASSERT(it_disposition(it, DISP_ENQ_CREATE_REF));
1373 request = it->it_request;
1374 it_clear_disposition(it, DISP_ENQ_CREATE_REF);
1375 rc = ll_prep_inode(&inode, &request->rq_pill, dir->i_sb, it);
1377 GOTO(out, inode = ERR_PTR(rc));
1379 /* Pause to allow for a race with concurrent access by fid */
1380 CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_NODE_PAUSE, cfs_fail_val);
1382 /* We asked for a lock on the directory, but were granted a
1383 * lock on the inode. Since we finally have an inode pointer,
1384 * stuff it in the lock. */
1385 CDEBUG(D_DLMTRACE, "setting l_ast_data to inode "DFID"(%p)\n",
1386 PFID(ll_inode2fid(inode)), inode);
1387 ll_set_lock_data(sbi->ll_md_exp, inode, it, NULL);
1390 ptlrpc_req_finished(request);
1395 * By the time this is called, we already have created the directory cache
1396 * entry for the new file, but it is so far negative - it has no inode.
1398 * We defer creating the OBD object(s) until open, to keep the intent and
1399 * non-intent code paths similar, and also because we do not have the MDS
1400 * inode number before calling ll_create_node() (which is needed for LOV),
1401 * so we would need to do yet another RPC to the MDS to store the LOV EA
1402 * data on the MDS. If needed, we would pass the PACKED lmm as data and
1403 * lmm_size in datalen (the MDS still has code which will handle that).
1405 * If the create succeeds, we fill in the inode information
1406 * with d_instantiate().
1408 static int ll_create_it(struct inode *dir, struct dentry *dentry,
1409 struct lookup_intent *it,
1410 void *secctx, __u32 secctxlen, bool encrypt,
1411 void *encctx, __u32 encctxlen, unsigned int open_flags)
1413 struct inode *inode;
1418 CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p), intent=%s\n",
1419 dentry, PFID(ll_inode2fid(dir)), dir, LL_IT2STR(it));
1421 rc = it_open_error(DISP_OPEN_CREATE, it);
1425 inode = ll_create_node(dir, it);
1427 RETURN(PTR_ERR(inode));
1429 /* must be done before d_instantiate, because it calls
1430 * security_d_instantiate, which means a getxattr if security
1431 * context is not set yet
1433 rc = ll_inode_notifysecctx(inode, secctx, secctxlen);
1437 d_instantiate(dentry, inode);
1440 bool preload = true;
1442 /* For migration or mirroring without enc key, we
1443 * create a volatile file without enc context.
1445 if (!llcrypt_has_encryption_key(dir) &&
1446 filename_is_volatile(dentry->d_name.name,
1447 dentry->d_name.len, NULL) &&
1448 (open_flags & O_FILE_ENC) == O_FILE_ENC &&
1449 open_flags & O_DIRECT)
1451 rc = ll_set_encflags(inode, encctx, encctxlen, preload);
1456 if (!test_bit(LL_SBI_FILE_SECCTX, ll_i2sbi(inode)->ll_flags)) {
1457 rc = ll_inode_init_security(dentry, inode, dir);
1462 ll_set_lock_data(ll_i2sbi(dir)->ll_md_exp, inode, it, &bits);
1463 if (bits & MDS_INODELOCK_LOOKUP) {
1464 d_lustre_revalidate(dentry);
1465 if (S_ISDIR(inode->i_mode))
1466 ll_update_dir_depth_dmv(dir, dentry);
1472 void ll_update_times(struct ptlrpc_request *request, struct inode *inode)
1474 struct mdt_body *body = req_capsule_server_get(&request->rq_pill,
1478 if (body->mbo_valid & OBD_MD_FLMTIME &&
1479 body->mbo_mtime > inode->i_mtime.tv_sec) {
1481 "setting fid " DFID " mtime from %lld to %llu\n",
1482 PFID(ll_inode2fid(inode)),
1483 (s64)inode->i_mtime.tv_sec, body->mbo_mtime);
1484 inode->i_mtime.tv_sec = body->mbo_mtime;
1487 if (body->mbo_valid & OBD_MD_FLCTIME &&
1488 body->mbo_ctime > inode->i_ctime.tv_sec)
1489 inode->i_ctime.tv_sec = body->mbo_ctime;
1492 /* once default LMV (space balanced) is set on ROOT, it should take effect if
1493 * default LMV is not set on parent directory.
1495 static void ll_qos_mkdir_prep(struct md_op_data *op_data, struct inode *dir)
1497 struct inode *root = dir->i_sb->s_root->d_inode;
1498 struct ll_inode_info *rlli = ll_i2info(root);
1499 struct ll_inode_info *lli = ll_i2info(dir);
1500 struct lmv_stripe_md *lsm;
1501 unsigned short depth;
1503 op_data->op_dir_depth = lli->lli_inherit_depth ?: lli->lli_dir_depth;
1504 depth = lli->lli_dir_depth;
1506 /* parent directory is striped */
1507 if (unlikely(ll_dir_striped(dir)))
1510 /* default LMV set on parent directory */
1511 if (unlikely(lli->lli_def_lsm_obj))
1514 /* parent is ROOT */
1515 if (unlikely(dir == root))
1518 /* default LMV not set on ROOT */
1519 if (!rlli->lli_def_lsm_obj)
1522 down_read(&rlli->lli_lsm_sem);
1523 if (!rlli->lli_def_lsm_obj)
1525 lsm = &rlli->lli_def_lsm_obj->lso_lsm;
1527 /* not space balanced */
1528 if (lsm->lsm_md_master_mdt_index != LMV_OFFSET_DEFAULT)
1532 * Check if the fs default is to be applied.
1533 * depth == 0 means 'not inited' for not root dir.
1535 if (lsm->lsm_md_max_inherit != LMV_INHERIT_NONE &&
1536 (lsm->lsm_md_max_inherit == LMV_INHERIT_UNLIMITED ||
1537 (depth && lsm->lsm_md_max_inherit > depth))) {
1538 op_data->op_flags |= MF_QOS_MKDIR;
1539 if (lsm->lsm_md_max_inherit_rr != LMV_INHERIT_RR_NONE &&
1540 (lsm->lsm_md_max_inherit_rr == LMV_INHERIT_RR_UNLIMITED ||
1541 (depth && lsm->lsm_md_max_inherit_rr > depth)))
1542 op_data->op_flags |= MF_RR_MKDIR;
1543 CDEBUG(D_INODE, DFID" requests qos mkdir %#x\n",
1544 PFID(&lli->lli_fid), op_data->op_flags);
1547 up_read(&rlli->lli_lsm_sem);
1550 static int ll_new_node(struct inode *dir, struct dentry *dchild,
1551 const char *tgt, umode_t mode, __u64 rdev, __u32 opc)
1553 struct qstr *name = &dchild->d_name;
1554 struct ptlrpc_request *request = NULL;
1555 struct md_op_data *op_data = NULL;
1556 struct inode *inode = NULL;
1557 struct ll_sb_info *sbi = ll_i2sbi(dir);
1558 struct llcrypt_str *disk_link = NULL;
1559 bool encrypt = false;
1560 struct lmv_user_md *lum = NULL;
1561 const void *data = NULL;
1566 if (unlikely(tgt != NULL)) {
1567 disk_link = (struct llcrypt_str *)rdev;
1571 data = disk_link->name;
1572 datalen = disk_link->len;
1576 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
1577 name->len, 0, opc, NULL);
1578 if (IS_ERR(op_data))
1579 GOTO(err_exit, err = PTR_ERR(op_data));
1581 if (S_ISDIR(mode)) {
1582 ll_qos_mkdir_prep(op_data, dir);
1583 if ((exp_connect_flags2(ll_i2mdexp(dir)) &
1584 OBD_CONNECT2_DMV_IMP_INHERIT) &&
1585 op_data->op_default_lso1 && !lum) {
1586 const struct lmv_stripe_md *lsm;
1588 /* once DMV_IMP_INHERIT is set, pack default LMV in
1593 GOTO(err_exit, err = -ENOMEM);
1595 lsm = &op_data->op_default_lso1->lso_lsm;
1596 lum->lum_magic = cpu_to_le32(lsm->lsm_md_magic);
1597 lum->lum_stripe_count =
1598 cpu_to_le32(lsm->lsm_md_stripe_count);
1599 lum->lum_stripe_offset =
1600 cpu_to_le32(lsm->lsm_md_master_mdt_index);
1601 lum->lum_hash_type =
1602 cpu_to_le32(lsm->lsm_md_hash_type);
1603 lum->lum_max_inherit = lsm->lsm_md_max_inherit;
1604 lum->lum_max_inherit_rr = lsm->lsm_md_max_inherit_rr;
1605 lum->lum_pool_name[0] = 0;
1606 op_data->op_bias |= MDS_CREATE_DEFAULT_LMV;
1608 datalen = sizeof(*lum);
1612 if (test_bit(LL_SBI_FILE_SECCTX, sbi->ll_flags)) {
1613 err = ll_dentry_init_security(dchild, mode, &dchild->d_name,
1614 &op_data->op_file_secctx_name,
1615 &op_data->op_file_secctx_name_size,
1616 &op_data->op_file_secctx,
1617 &op_data->op_file_secctx_size,
1618 &op_data->op_file_secctx_slot);
1620 GOTO(err_exit, err);
1623 if (ll_sbi_has_encrypt(sbi) &&
1624 ((IS_ENCRYPTED(dir) &&
1625 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) ||
1626 (unlikely(ll_sb_has_test_dummy_encryption(dir->i_sb)) &&
1628 err = llcrypt_prepare_readdir(dir);
1630 GOTO(err_exit, err);
1631 if (!llcrypt_has_encryption_key(dir))
1632 GOTO(err_exit, err = -ENOKEY);
1637 err = llcrypt_inherit_context(dir, NULL, op_data, false);
1639 GOTO(err_exit, err);
1641 if (S_ISLNK(mode)) {
1642 /* llcrypt needs inode to encrypt target name, so create
1643 * a fake inode and associate encryption context got
1644 * from llcrypt_inherit_context.
1646 struct inode *fakeinode =
1647 dchild->d_sb->s_op->alloc_inode(dchild->d_sb);
1650 GOTO(err_exit, err = -ENOMEM);
1651 fakeinode->i_sb = dchild->d_sb;
1652 fakeinode->i_mode |= S_IFLNK;
1654 fakeinode->i_opflags |= IOP_XATTR;
1656 ll_lli_init(ll_i2info(fakeinode));
1657 err = ll_set_encflags(fakeinode,
1658 op_data->op_file_encctx,
1659 op_data->op_file_encctx_size,
1662 err = __llcrypt_encrypt_symlink(fakeinode, tgt,
1666 ll_xattr_cache_destroy(fakeinode);
1667 llcrypt_put_encryption_info(fakeinode);
1668 dchild->d_sb->s_op->destroy_inode(fakeinode);
1670 GOTO(err_exit, err);
1672 data = disk_link->name;
1673 datalen = disk_link->len;
1677 err = md_create(sbi->ll_md_exp, op_data, data, datalen, mode,
1678 from_kuid(&init_user_ns, current_fsuid()),
1679 from_kgid(&init_user_ns, current_fsgid()),
1680 current_cap(), rdev, &request);
1681 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 17, 58, 0)
1683 * server < 2.12.58 doesn't pack default LMV in intent_getattr reply,
1684 * fetch default LMV here.
1686 if (unlikely(err == -EREMOTE)) {
1687 struct ll_inode_info *lli = ll_i2info(dir);
1688 struct lmv_user_md *lum;
1692 ptlrpc_req_finished(request);
1694 ll_finish_md_op_data(op_data);
1697 err2 = ll_dir_getstripe(dir, (void **)&lum, &lumsize, &request,
1698 OBD_MD_DEFAULT_MEA);
1700 struct lustre_md md = { NULL };
1703 md.body = req_capsule_server_get(&request->rq_pill,
1706 GOTO(err_exit, err = -EPROTO);
1708 OBD_ALLOC_PTR(md.def_lsm_obj);
1709 if (!md.def_lsm_obj)
1710 GOTO(err_exit, err = -ENOMEM);
1712 md.def_lsm_obj->lso_lsm.lsm_md_magic = lum->lum_magic;
1713 md.def_lsm_obj->lso_lsm.lsm_md_stripe_count =
1714 lum->lum_stripe_count;
1715 md.def_lsm_obj->lso_lsm.lsm_md_master_mdt_index =
1716 lum->lum_stripe_offset;
1717 md.def_lsm_obj->lso_lsm.lsm_md_hash_type =
1719 md.def_lsm_obj->lso_lsm.lsm_md_max_inherit =
1720 lum->lum_max_inherit;
1721 md.def_lsm_obj->lso_lsm.lsm_md_max_inherit_rr =
1722 lum->lum_max_inherit_rr;
1723 atomic_set(&md.def_lsm_obj->lso_refs, 1);
1725 err = ll_update_inode(dir, &md);
1726 md_put_lustre_md(sbi->ll_md_exp, &md);
1728 GOTO(err_exit, err);
1729 } else if (err2 == -ENODATA && lli->lli_def_lsm_obj) {
1731 * If there are no default stripe EA on the MDT, but the
1732 * client has default stripe, then it probably means
1733 * default stripe EA has just been deleted.
1735 down_write(&lli->lli_lsm_sem);
1736 lmv_stripe_object_put(&lli->lli_def_lsm_obj);
1737 up_write(&lli->lli_lsm_sem);
1739 GOTO(err_exit, err);
1742 ptlrpc_req_finished(request);
1749 GOTO(err_exit, err);
1751 ll_update_times(request, dir);
1753 CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_NEWNODE_PAUSE, cfs_fail_val);
1755 err = ll_prep_inode(&inode, &request->rq_pill, dchild->d_sb, NULL);
1757 GOTO(err_exit, err);
1759 /* must be done before d_instantiate, because it calls
1760 * security_d_instantiate, which means a getxattr if security
1761 * context is not set yet
1763 err = ll_inode_notifysecctx(inode,
1764 op_data->op_file_secctx,
1765 op_data->op_file_secctx_size);
1767 GOTO(err_exit, err);
1769 d_instantiate(dchild, inode);
1772 err = ll_set_encflags(inode, op_data->op_file_encctx,
1773 op_data->op_file_encctx_size, true);
1775 GOTO(err_exit, err);
1777 if (S_ISLNK(mode)) {
1778 struct ll_inode_info *lli = ll_i2info(inode);
1780 /* Cache the plaintext symlink target
1781 * for later use by get_link()
1783 OBD_ALLOC(lli->lli_symlink_name, strlen(tgt) + 1);
1784 /* do not return an error if we cannot
1785 * cache the symlink locally
1787 if (lli->lli_symlink_name)
1788 memcpy(lli->lli_symlink_name,
1789 tgt, strlen(tgt) + 1);
1793 if (!test_bit(LL_SBI_FILE_SECCTX, sbi->ll_flags)) {
1794 err = ll_inode_init_security(dchild, inode, dir);
1796 GOTO(err_exit, err);
1801 if (request != NULL)
1802 ptlrpc_req_finished(request);
1803 if (!IS_ERR_OR_NULL(op_data))
1804 ll_finish_md_op_data(op_data);
1811 static int ll_mknod(struct mnt_idmap *map, struct inode *dir,
1812 struct dentry *dchild, umode_t mode, dev_t rdev)
1814 ktime_t kstart = ktime_get();
1818 /* VFS has locked the inode before calling this */
1819 ll_set_inode_lock_owner(dir);
1821 CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p) mode %o dev %x\n",
1822 dchild, PFID(ll_inode2fid(dir)), dir, mode, rdev);
1824 if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
1825 mode &= ~current_umask();
1827 switch (mode & S_IFMT) {
1836 err = ll_new_node(dir, dchild, NULL, mode, old_encode_dev(rdev),
1847 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKNOD,
1848 ktime_us_delta(ktime_get(), kstart));
1849 ll_clear_inode_lock_owner(dir);
1855 * Plain create. Intent create is handled in atomic_open.
1857 static int ll_create_nd(struct mnt_idmap *map, struct inode *dir,
1858 struct dentry *dentry, umode_t mode, bool want_excl)
1860 ktime_t kstart = ktime_get();
1863 /* VFS has locked the inode before calling this */
1864 ll_set_inode_lock_owner(dir);
1866 CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_FILE_PAUSE, cfs_fail_val);
1869 "VFS Op:name=%pd, dir="DFID"(%p), flags=%u, excl=%d\n",
1870 dentry, PFID(ll_inode2fid(dir)), dir, mode, want_excl);
1872 /* Using mknod(2) to create a regular file is designed to not recognize
1873 * volatile file name, so we use ll_mknod() here. */
1874 rc = ll_mknod(map, dir, dentry, mode, 0);
1876 CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, unhashed %d\n",
1877 dentry, d_unhashed(dentry));
1880 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_CREATE,
1881 ktime_us_delta(ktime_get(), kstart));
1883 ll_clear_inode_lock_owner(dir);
1888 static int ll_symlink(struct mnt_idmap *map, struct inode *dir,
1889 struct dentry *dchild, const char *oldpath)
1891 ktime_t kstart = ktime_get();
1892 int len = strlen(oldpath);
1893 struct llcrypt_str disk_link;
1897 /* VFS has locked the inode before calling this */
1898 ll_set_inode_lock_owner(dir);
1900 CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p), target=%.*s\n",
1901 dchild, PFID(ll_inode2fid(dir)), dir, 3000, oldpath);
1903 err = llcrypt_prepare_symlink(dir, oldpath, len, dir->i_sb->s_blocksize,
1908 err = ll_new_node(dir, dchild, oldpath, S_IFLNK | S_IRWXUGO,
1909 (__u64)&disk_link, LUSTRE_OPC_SYMLINK);
1911 if (disk_link.name != (unsigned char *)oldpath)
1912 kfree(disk_link.name);
1915 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_SYMLINK,
1916 ktime_us_delta(ktime_get(), kstart));
1919 ll_clear_inode_lock_owner(dir);
1924 static int ll_link(struct dentry *old_dentry, struct inode *dir,
1925 struct dentry *new_dentry)
1927 struct inode *src = old_dentry->d_inode;
1928 struct qstr *name = &new_dentry->d_name;
1929 struct ll_sb_info *sbi = ll_i2sbi(dir);
1930 struct ptlrpc_request *request = NULL;
1931 struct md_op_data *op_data;
1932 ktime_t kstart = ktime_get();
1936 /* VFS has locked the inodes before calling this */
1937 ll_set_inode_lock_owner(src);
1938 ll_set_inode_lock_owner(dir);
1941 "VFS Op: inode="DFID"(%p), dir="DFID"(%p), target=%pd\n",
1942 PFID(ll_inode2fid(src)), src,
1943 PFID(ll_inode2fid(dir)), dir, new_dentry);
1945 err = llcrypt_prepare_link(old_dentry, dir, new_dentry);
1949 op_data = ll_prep_md_op_data(NULL, src, dir, name->name, name->len,
1950 0, LUSTRE_OPC_ANY, NULL);
1951 if (IS_ERR(op_data))
1952 GOTO(clear, err = PTR_ERR(op_data));
1954 err = md_link(sbi->ll_md_exp, op_data, &request);
1955 ll_finish_md_op_data(op_data);
1959 ll_update_times(request, dir);
1960 ll_stats_ops_tally(sbi, LPROC_LL_LINK,
1961 ktime_us_delta(ktime_get(), kstart));
1964 ptlrpc_req_finished(request);
1966 ll_clear_inode_lock_owner(src);
1967 ll_clear_inode_lock_owner(dir);
1972 static int ll_mkdir(struct mnt_idmap *map, struct inode *dir,
1973 struct dentry *dchild, umode_t mode)
1975 ktime_t kstart = ktime_get();
1979 /* VFS has locked the inode before calling this */
1980 ll_set_inode_lock_owner(dir);
1982 CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p)\n",
1983 dchild, PFID(ll_inode2fid(dir)), dir);
1985 if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
1986 mode &= ~current_umask();
1988 mode = (mode & (S_IRWXUGO|S_ISVTX)) | S_IFDIR;
1990 err = ll_new_node(dir, dchild, NULL, mode, 0, LUSTRE_OPC_MKDIR);
1992 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKDIR,
1993 ktime_us_delta(ktime_get(), kstart));
1995 ll_clear_inode_lock_owner(dir);
2000 static int ll_rmdir(struct inode *dir, struct dentry *dchild)
2002 struct qstr *name = &dchild->d_name;
2003 struct ptlrpc_request *request = NULL;
2004 struct md_op_data *op_data;
2005 ktime_t kstart = ktime_get();
2010 /* VFS has locked the inodes before calling this */
2011 ll_set_inode_lock_owner(dir);
2012 ll_set_inode_lock_owner(dchild->d_inode);
2014 CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p)\n",
2015 dchild, PFID(ll_inode2fid(dir)), dir);
2017 if (unlikely(d_mountpoint(dchild)))
2018 GOTO(out, rc = -EBUSY);
2020 /* some foreign dir may not be allowed to be removed */
2021 if (!ll_foreign_is_removable(dchild, false))
2022 GOTO(out, rc = -EPERM);
2024 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len,
2025 S_IFDIR, LUSTRE_OPC_ANY, NULL);
2026 if (IS_ERR(op_data))
2027 GOTO(out, rc = PTR_ERR(op_data));
2029 if (dchild->d_inode != NULL)
2030 op_data->op_fid3 = *ll_inode2fid(dchild->d_inode);
2032 if (fid_is_zero(&op_data->op_fid2))
2033 op_data->op_fid2 = op_data->op_fid3;
2034 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
2035 ll_finish_md_op_data(op_data);
2037 struct mdt_body *body;
2039 ll_update_times(request, dir);
2040 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR,
2041 ktime_us_delta(ktime_get(), kstart));
2044 * The server puts attributes in on the last unlink, use them
2045 * to update the link count so the inode can be freed
2048 body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
2049 if (body->mbo_valid & OBD_MD_FLNLINK) {
2050 spin_lock(&dchild->d_inode->i_lock);
2051 set_nlink(dchild->d_inode, body->mbo_nlink);
2052 spin_unlock(&dchild->d_inode->i_lock);
2056 ptlrpc_req_finished(request);
2058 ll_clear_inode_lock_owner(dir);
2059 ll_clear_inode_lock_owner(dchild->d_inode);
2067 int ll_rmdir_entry(struct inode *dir, char *name, int namelen)
2069 struct ptlrpc_request *request = NULL;
2070 struct md_op_data *op_data;
2071 ktime_t kstart = ktime_get();
2075 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
2076 namelen, name, PFID(ll_inode2fid(dir)), dir);
2078 op_data = ll_prep_md_op_data(NULL, dir, NULL, name, strlen(name),
2079 S_IFDIR, LUSTRE_OPC_ANY, NULL);
2080 if (IS_ERR(op_data))
2081 RETURN(PTR_ERR(op_data));
2082 op_data->op_cli_flags |= CLI_RM_ENTRY;
2083 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
2084 ll_finish_md_op_data(op_data);
2086 ll_update_times(request, dir);
2088 ptlrpc_req_finished(request);
2090 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR,
2091 ktime_us_delta(ktime_get(), kstart));
2095 static int ll_unlink(struct inode *dir, struct dentry *dchild)
2097 struct qstr *name = &dchild->d_name;
2098 struct ptlrpc_request *request = NULL;
2099 struct md_op_data *op_data;
2100 struct mdt_body *body;
2101 ktime_t kstart = ktime_get();
2106 /* VFS has locked the inodes before calling this */
2107 ll_set_inode_lock_owner(dir);
2108 ll_set_inode_lock_owner(dchild->d_inode);
2110 CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p)\n",
2111 dchild, PFID(ll_inode2fid(dir)), dir);
2114 * XXX: unlink bind mountpoint maybe call to here,
2115 * just check it as vfs_unlink does.
2117 if (unlikely(d_mountpoint(dchild)))
2118 GOTO(clear, rc = -EBUSY);
2120 /* some foreign file/dir may not be allowed to be unlinked */
2121 if (!ll_foreign_is_removable(dchild, false))
2122 GOTO(clear, rc = -EPERM);
2124 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len, 0,
2125 LUSTRE_OPC_ANY, NULL);
2126 if (IS_ERR(op_data))
2127 GOTO(clear, rc = PTR_ERR(op_data));
2129 op_data->op_fid3 = *ll_inode2fid(dchild->d_inode);
2130 /* notify lower layer if inode has dirty pages */
2131 if (S_ISREG(dchild->d_inode->i_mode) &&
2132 ll_i2info(dchild->d_inode)->lli_clob &&
2133 dirty_cnt(dchild->d_inode))
2134 op_data->op_cli_flags |= CLI_DIRTY_DATA;
2135 if (fid_is_zero(&op_data->op_fid2))
2136 op_data->op_fid2 = op_data->op_fid3;
2137 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
2138 ll_finish_md_op_data(op_data);
2143 * The server puts attributes in on the last unlink, use them to update
2144 * the link count so the inode can be freed immediately.
2146 body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
2147 if (body->mbo_valid & OBD_MD_FLNLINK) {
2148 spin_lock(&dchild->d_inode->i_lock);
2149 set_nlink(dchild->d_inode, body->mbo_nlink);
2150 spin_unlock(&dchild->d_inode->i_lock);
2153 ll_update_times(request, dir);
2156 ptlrpc_req_finished(request);
2158 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_UNLINK,
2159 ktime_us_delta(ktime_get(), kstart));
2161 ll_clear_inode_lock_owner(dir);
2162 ll_clear_inode_lock_owner(dchild->d_inode);
2166 static int ll_rename(struct mnt_idmap *map,
2167 struct inode *src, struct dentry *src_dchild,
2168 struct inode *tgt, struct dentry *tgt_dchild
2169 #if defined(HAVE_USER_NAMESPACE_ARG) || defined(HAVE_IOPS_RENAME_WITH_FLAGS)
2170 , unsigned int flags
2174 struct ptlrpc_request *request = NULL;
2175 struct ll_sb_info *sbi = ll_i2sbi(src);
2176 struct md_op_data *op_data;
2177 ktime_t kstart = ktime_get();
2179 struct llcrypt_name foldname, fnewname;
2183 /* VFS has locked the inodes before calling this */
2184 ll_set_inode_lock_owner(src);
2185 ll_set_inode_lock_owner(tgt);
2186 if (tgt_dchild->d_inode)
2187 ll_set_inode_lock_owner(tgt_dchild->d_inode);
2189 #if defined(HAVE_USER_NAMESPACE_ARG) || defined(HAVE_IOPS_RENAME_WITH_FLAGS)
2191 GOTO(out, err = -EINVAL);
2195 "VFS Op:oldname=%pd, src_dir="DFID"(%p), newname=%pd, tgt_dir="DFID"(%p)\n",
2196 src_dchild, PFID(ll_inode2fid(src)), src,
2197 tgt_dchild, PFID(ll_inode2fid(tgt)), tgt);
2199 if (unlikely(d_mountpoint(src_dchild) || d_mountpoint(tgt_dchild)))
2200 GOTO(out, err = -EBUSY);
2202 #if defined(HAVE_USER_NAMESPACE_ARG) || defined(HAVE_IOPS_RENAME_WITH_FLAGS)
2203 err = llcrypt_prepare_rename(src, src_dchild, tgt, tgt_dchild, flags);
2205 err = llcrypt_prepare_rename(src, src_dchild, tgt, tgt_dchild, 0);
2209 /* we prevent an encrypted file from being renamed
2210 * into an unencrypted dir
2212 if (IS_ENCRYPTED(src) && !IS_ENCRYPTED(tgt))
2213 GOTO(out, err = -EXDEV);
2215 if (src_dchild->d_inode)
2216 mode = src_dchild->d_inode->i_mode;
2218 if (tgt_dchild->d_inode)
2219 mode = tgt_dchild->d_inode->i_mode;
2221 op_data = ll_prep_md_op_data(NULL, src, tgt, NULL, 0, mode,
2222 LUSTRE_OPC_ANY, NULL);
2223 if (IS_ERR(op_data))
2224 GOTO(out, err = PTR_ERR(op_data));
2226 /* If the client is using a subdir mount and does a rename to what it
2227 * sees as /.fscrypt, interpret it as the .fscrypt dir at fs root.
2229 if (unlikely(is_root_inode(tgt) && !fid_is_root(ll_inode2fid(tgt)) &&
2230 tgt_dchild->d_name.len == strlen(dot_fscrypt_name) &&
2231 strncmp(tgt_dchild->d_name.name, dot_fscrypt_name,
2232 tgt_dchild->d_name.len) == 0))
2233 lu_root_fid(&op_data->op_fid2);
2235 if (src_dchild->d_inode)
2236 op_data->op_fid3 = *ll_inode2fid(src_dchild->d_inode);
2238 if (tgt_dchild->d_inode)
2239 op_data->op_fid4 = *ll_inode2fid(tgt_dchild->d_inode);
2241 err = ll_setup_filename(src, &src_dchild->d_name, 1, &foldname, NULL);
2244 err = ll_setup_filename(tgt, &tgt_dchild->d_name, 1, &fnewname, NULL);
2246 llcrypt_free_filename(&foldname);
2249 err = md_rename(sbi->ll_md_exp, op_data,
2250 foldname.disk_name.name, foldname.disk_name.len,
2251 fnewname.disk_name.name, fnewname.disk_name.len,
2253 llcrypt_free_filename(&foldname);
2254 llcrypt_free_filename(&fnewname);
2255 ll_finish_md_op_data(op_data);
2257 ll_update_times(request, src);
2258 ll_update_times(request, tgt);
2261 ptlrpc_req_finished(request);
2264 d_move(src_dchild, tgt_dchild);
2265 ll_stats_ops_tally(sbi, LPROC_LL_RENAME,
2266 ktime_us_delta(ktime_get(), kstart));
2269 ll_clear_inode_lock_owner(src);
2270 ll_clear_inode_lock_owner(tgt);
2271 if (tgt_dchild->d_inode)
2272 ll_clear_inode_lock_owner(tgt_dchild->d_inode);
2276 const struct inode_operations ll_dir_inode_operations = {
2278 .atomic_open = ll_atomic_open,
2279 .lookup = ll_lookup_nd,
2280 .create = ll_create_nd,
2281 /* We need all these non-raw things for NFSD, to not patch it. */
2282 .unlink = ll_unlink,
2285 .symlink = ll_symlink,
2287 .rename = ll_rename,
2288 .setattr = ll_setattr,
2289 .getattr = ll_getattr,
2290 .permission = ll_inode_permission,
2291 #ifdef HAVE_IOP_XATTR
2292 .setxattr = ll_setxattr,
2293 .getxattr = ll_getxattr,
2294 .removexattr = ll_removexattr,
2296 .listxattr = ll_listxattr,
2297 .get_acl = ll_get_acl,
2298 #ifdef HAVE_IOP_SET_ACL
2299 .set_acl = ll_set_acl,
2301 #ifdef HAVE_FILEATTR_GET
2302 .fileattr_get = ll_fileattr_get,
2303 .fileattr_set = ll_fileattr_set,
2307 const struct inode_operations ll_special_inode_operations = {
2308 .setattr = ll_setattr,
2309 .getattr = ll_getattr,
2310 .permission = ll_inode_permission,
2311 #ifdef HAVE_IOP_XATTR
2312 .setxattr = ll_setxattr,
2313 .getxattr = ll_getxattr,
2314 .removexattr = ll_removexattr,
2316 .listxattr = ll_listxattr,
2317 .get_acl = ll_get_acl,
2318 #ifdef HAVE_IOP_SET_ACL
2319 .set_acl = ll_set_acl,