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,
53 void *encctx, __u32 encctxlen);
55 /* called from iget5_locked->find_inode() under inode_lock spinlock */
56 static int ll_test_inode(struct inode *inode, void *opaque)
58 struct ll_inode_info *lli = ll_i2info(inode);
59 struct lustre_md *md = opaque;
61 if (unlikely(!(md->body->mbo_valid & OBD_MD_FLID))) {
62 CERROR("MDS body missing FID\n");
66 if (!lu_fid_eq(&lli->lli_fid, &md->body->mbo_fid1))
72 static int ll_set_inode(struct inode *inode, void *opaque)
74 struct ll_inode_info *lli = ll_i2info(inode);
75 struct mdt_body *body = ((struct lustre_md *)opaque)->body;
77 if (unlikely(!(body->mbo_valid & OBD_MD_FLID))) {
78 CERROR("MDS body missing FID\n");
82 lli->lli_fid = body->mbo_fid1;
83 if (unlikely(!(body->mbo_valid & OBD_MD_FLTYPE))) {
84 CERROR("Can not initialize inode "DFID" without object type: "
86 PFID(&lli->lli_fid), body->mbo_valid);
90 inode->i_mode = (inode->i_mode & ~S_IFMT) | (body->mbo_mode & S_IFMT);
91 if (unlikely(inode->i_mode == 0)) {
92 CERROR("Invalid inode "DFID" type\n", PFID(&lli->lli_fid));
103 * Get an inode by inode number(@hash), which is already instantiated by
104 * the intent lookup).
106 struct inode *ll_iget(struct super_block *sb, ino_t hash,
107 struct lustre_md *md)
115 inode = iget5_locked(sb, hash, ll_test_inode, ll_set_inode, md);
117 RETURN(ERR_PTR(-ENOMEM));
119 if (inode->i_state & I_NEW) {
120 rc = ll_read_inode2(inode, md);
121 if (rc == 0 && S_ISREG(inode->i_mode) &&
122 ll_i2info(inode)->lli_clob == NULL)
123 rc = cl_file_inode_init(inode, md);
126 /* Let's clear directory lsm here, otherwise
127 * make_bad_inode() will reset the inode mode
128 * to regular, then ll_clear_inode will not
129 * be able to clear lsm_md */
130 if (S_ISDIR(inode->i_mode))
131 ll_dir_clear_lsm_md(inode);
132 make_bad_inode(inode);
133 unlock_new_inode(inode);
137 inode_has_no_xattr(inode);
138 unlock_new_inode(inode);
140 } else if (is_bad_inode(inode)) {
142 inode = ERR_PTR(-ESTALE);
143 } else if (!(inode->i_state & (I_FREEING | I_CLEAR))) {
144 rc = ll_update_inode(inode, md);
145 CDEBUG(D_VFSTRACE, "got inode: "DFID"(%p): rc = %d\n",
146 PFID(&md->body->mbo_fid1), inode, rc);
148 if (S_ISDIR(inode->i_mode))
149 ll_dir_clear_lsm_md(inode);
158 /* mark negative sub file dentries invalid and prune unused dentries */
159 static void ll_prune_negative_children(struct inode *dir)
161 struct dentry *dentry;
162 struct dentry *child;
167 spin_lock(&dir->i_lock);
168 hlist_for_each_entry(dentry, &dir->i_dentry, d_alias) {
169 spin_lock(&dentry->d_lock);
170 list_for_each_entry(child, &dentry->d_subdirs, d_child) {
174 spin_lock_nested(&child->d_lock, DENTRY_D_LOCK_NESTED);
175 __d_lustre_invalidate(child);
176 if (!ll_d_count(child)) {
179 spin_unlock(&child->d_lock);
180 spin_unlock(&dentry->d_lock);
181 spin_unlock(&dir->i_lock);
183 CDEBUG(D_DENTRY, "prune negative dentry %pd\n",
189 spin_unlock(&child->d_lock);
191 spin_unlock(&dentry->d_lock);
193 spin_unlock(&dir->i_lock);
198 int ll_test_inode_by_fid(struct inode *inode, void *opaque)
200 return lu_fid_eq(&ll_i2info(inode)->lli_fid, opaque);
203 static int ll_dom_lock_cancel(struct inode *inode, struct ldlm_lock *lock)
206 struct ll_inode_info *lli = ll_i2info(inode);
211 env = cl_env_get(&refcheck);
213 RETURN(PTR_ERR(env));
215 /* reach MDC layer to flush data under the DoM ldlm lock */
216 rc = cl_object_flush(env, lli->lli_clob, lock);
217 if (rc == -ENODATA) {
218 CDEBUG(D_INODE, "inode "DFID" layout has no DoM stripe\n",
219 PFID(ll_inode2fid(inode)));
220 /* most likely result of layout change, do nothing */
224 cl_env_put(env, &refcheck);
228 static void ll_lock_cancel_bits(struct ldlm_lock *lock, __u64 to_cancel)
230 struct inode *inode = ll_inode_from_resource_lock(lock);
231 struct ll_inode_info *lli;
232 __u64 bits = to_cancel;
238 /* That means the inode is evicted most likely and may cause
239 * the skipping of lock cleanups below, so print the message
242 if (lock->l_resource->lr_lvb_inode)
244 "can't take inode for the lock (%sevicted)\n",
245 lock->l_resource->lr_lvb_inode->i_state &
246 I_FREEING ? "" : "not ");
250 if (!fid_res_name_eq(ll_inode2fid(inode),
251 &lock->l_resource->lr_name)) {
252 LDLM_ERROR(lock, "data mismatch with object "DFID"(%p)",
253 PFID(ll_inode2fid(inode)), inode);
257 if (bits & MDS_INODELOCK_XATTR) {
258 ll_xattr_cache_destroy(inode);
259 bits &= ~MDS_INODELOCK_XATTR;
262 /* For OPEN locks we differentiate between lock modes
263 * LCK_CR, LCK_CW, LCK_PR - bug 22891 */
264 if (bits & MDS_INODELOCK_OPEN)
265 ll_have_md_lock(inode, &bits, lock->l_req_mode);
267 if (bits & MDS_INODELOCK_OPEN) {
270 switch (lock->l_req_mode) {
281 LDLM_ERROR(lock, "bad lock mode for OPEN lock");
285 ll_md_real_close(inode, fmode);
287 bits &= ~MDS_INODELOCK_OPEN;
290 if (bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
291 MDS_INODELOCK_LAYOUT | MDS_INODELOCK_PERM |
293 ll_have_md_lock(inode, &bits, LCK_MINMODE);
295 if (bits & MDS_INODELOCK_DOM) {
296 rc = ll_dom_lock_cancel(inode, lock);
298 CDEBUG(D_INODE, "cannot flush DoM data "
300 PFID(ll_inode2fid(inode)), rc);
303 if (bits & MDS_INODELOCK_LAYOUT) {
304 struct cl_object_conf conf = {
305 .coc_opc = OBJECT_CONF_INVALIDATE,
309 rc = ll_layout_conf(inode, &conf);
311 CDEBUG(D_INODE, "cannot invalidate layout of "
313 PFID(ll_inode2fid(inode)), rc);
316 lli = ll_i2info(inode);
318 if (bits & MDS_INODELOCK_UPDATE)
319 ll_file_set_flag(lli, LLIF_UPDATE_ATIME);
321 if ((bits & MDS_INODELOCK_UPDATE) && S_ISDIR(inode->i_mode)) {
322 CDEBUG(D_INODE, "invalidating inode "DFID" lli = %p, "
323 "pfid = "DFID"\n", PFID(ll_inode2fid(inode)),
324 lli, PFID(&lli->lli_pfid));
325 truncate_inode_pages(inode->i_mapping, 0);
327 if (unlikely(!fid_is_zero(&lli->lli_pfid))) {
328 struct inode *master_inode = NULL;
331 /* This is slave inode, since all of the child dentry
332 * is connected on the master inode, so we have to
333 * invalidate the negative children on master inode */
334 CDEBUG(D_INODE, "Invalidate s"DFID" m"DFID"\n",
335 PFID(ll_inode2fid(inode)), PFID(&lli->lli_pfid));
337 hash = cl_fid_build_ino(&lli->lli_pfid,
338 ll_need_32bit_api(ll_i2sbi(inode)));
340 /* Do not lookup the inode with ilookup5, otherwise
341 * it will cause dead lock,
342 * 1. Client1 send chmod req to the MDT0, then on MDT0,
343 * it enqueues master and all of its slaves lock,
344 * (mdt_attr_set() -> mdt_lock_slaves()), after gets
345 * master and stripe0 lock, it will send the enqueue
346 * req (for stripe1) to MDT1, then MDT1 finds the lock
347 * has been granted to client2. Then MDT1 sends blocking
349 * 2. At the same time, client2 tries to unlink
350 * the striped dir (rm -rf striped_dir), and during
351 * lookup, it will hold the master inode of the striped
352 * directory, whose inode state is NEW, then tries to
353 * revalidate all of its slaves, (ll_prep_inode()->
354 * ll_iget()->ll_read_inode2()-> ll_update_inode().).
355 * And it will be blocked on the server side because
357 * 3. Then the client get the blocking_ast req, cancel
358 * the lock, but being blocked if using ->ilookup5()),
359 * because master inode state is NEW. */
360 master_inode = ilookup5_nowait(inode->i_sb, hash,
361 ll_test_inode_by_fid,
362 (void *)&lli->lli_pfid);
364 ll_prune_negative_children(master_inode);
368 ll_prune_negative_children(inode);
372 if ((bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM)) &&
373 inode->i_sb->s_root != NULL &&
374 inode != inode->i_sb->s_root->d_inode)
375 ll_prune_aliases(inode);
377 if (bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM))
378 forget_all_cached_acls(inode);
384 /* Check if the given lock may be downgraded instead of canceling and
385 * that convert is really needed. */
386 int ll_md_need_convert(struct ldlm_lock *lock)
388 struct ldlm_namespace *ns = ldlm_lock_to_ns(lock);
390 __u64 wanted = lock->l_policy_data.l_inodebits.cancel_bits;
391 __u64 bits = lock->l_policy_data.l_inodebits.bits & ~wanted;
392 enum ldlm_mode mode = LCK_MINMODE;
394 if (!lock->l_conn_export ||
395 !exp_connect_lock_convert(lock->l_conn_export))
398 if (!wanted || !bits || ldlm_is_cancel(lock))
401 /* do not convert locks other than DOM for now */
402 if (!((bits | wanted) & MDS_INODELOCK_DOM))
405 /* We may have already remaining bits in some other lock so
406 * lock convert will leave us just extra lock for the same bit.
407 * Check if client has other lock with the same bits and the same
408 * or lower mode and don't convert if any.
410 switch (lock->l_req_mode) {
424 /* do not convert other modes */
428 /* is lock is too old to be converted? */
429 lock_res_and_lock(lock);
430 if (ktime_after(ktime_get(),
431 ktime_add(lock->l_last_used, ns->ns_dirty_age_limit))) {
432 unlock_res_and_lock(lock);
435 unlock_res_and_lock(lock);
437 inode = ll_inode_from_resource_lock(lock);
438 ll_have_md_lock(inode, &bits, mode);
443 int ll_md_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *ld,
444 void *data, int flag)
446 struct lustre_handle lockh;
452 case LDLM_CB_BLOCKING:
454 __u64 cancel_flags = LCF_ASYNC;
456 /* if lock convert is not needed then still have to
457 * pass lock via ldlm_cli_convert() to keep all states
458 * correct, set cancel_bits to full lock bits to cause
459 * full cancel to happen.
461 if (!ll_md_need_convert(lock)) {
462 lock_res_and_lock(lock);
463 lock->l_policy_data.l_inodebits.cancel_bits =
464 lock->l_policy_data.l_inodebits.bits;
465 unlock_res_and_lock(lock);
467 rc = ldlm_cli_convert(lock, cancel_flags);
470 /* continue with cancel otherwise */
471 ldlm_lock2handle(lock, &lockh);
472 rc = ldlm_cli_cancel(&lockh, cancel_flags);
474 CDEBUG(D_INODE, "ldlm_cli_cancel: rc = %d\n", rc);
479 case LDLM_CB_CANCELING:
481 __u64 to_cancel = lock->l_policy_data.l_inodebits.bits;
483 /* Nothing to do for non-granted locks */
484 if (!ldlm_is_granted(lock))
487 /* If 'ld' is supplied then bits to be cancelled are passed
488 * implicitly by lock converting and cancel_bits from 'ld'
489 * should be used. Otherwise full cancel is being performed
490 * and lock inodebits are used.
492 * Note: we cannot rely on cancel_bits in lock itself at this
493 * moment because they can be changed by concurrent thread,
494 * so ldlm_cli_inodebits_convert() pass cancel bits implicitly
498 /* partial bits cancel allowed only during convert */
499 LASSERT(ldlm_is_converting(lock));
500 /* mask cancel bits by lock bits so only no any unused
501 * bits are passed to ll_lock_cancel_bits()
503 to_cancel &= ld->l_policy_data.l_inodebits.cancel_bits;
505 ll_lock_cancel_bits(lock, to_cancel);
515 __u32 ll_i2suppgid(struct inode *i)
517 if (in_group_p(i->i_gid))
518 return (__u32)from_kgid(&init_user_ns, i->i_gid);
520 return (__u32) __kgid_val(INVALID_GID);
523 /* Pack the required supplementary groups into the supplied groups array.
524 * If we don't need to use the groups from the target inode(s) then we
525 * instead pack one or more groups from the user's supplementary group
526 * array in case it might be useful. Not needed if doing an MDS-side upcall. */
527 void ll_i2gids(__u32 *suppgids, struct inode *i1, struct inode *i2)
530 LASSERT(suppgids != NULL);
532 suppgids[0] = ll_i2suppgid(i1);
535 suppgids[1] = ll_i2suppgid(i2);
541 * try to reuse three types of dentry:
542 * 1. unhashed alias, this one is unhashed by d_invalidate (but it may be valid
543 * by concurrent .revalidate).
544 * 2. INVALID alias (common case for no valid ldlm lock held, but this flag may
545 * be cleared by others calling d_lustre_revalidate).
546 * 3. DISCONNECTED alias.
548 static struct dentry *ll_find_alias(struct inode *inode, struct dentry *dentry)
550 struct dentry *alias, *discon_alias, *invalid_alias;
552 if (hlist_empty(&inode->i_dentry))
555 discon_alias = invalid_alias = NULL;
557 spin_lock(&inode->i_lock);
558 hlist_for_each_entry(alias, &inode->i_dentry, d_alias) {
559 LASSERT(alias != dentry);
561 spin_lock(&alias->d_lock);
562 if ((alias->d_flags & DCACHE_DISCONNECTED) &&
563 S_ISDIR(inode->i_mode))
564 /* LASSERT(last_discon == NULL); LU-405, bz 20055 */
565 discon_alias = alias;
566 else if (alias->d_parent == dentry->d_parent &&
567 alias->d_name.hash == dentry->d_name.hash &&
568 alias->d_name.len == dentry->d_name.len &&
569 memcmp(alias->d_name.name, dentry->d_name.name,
570 dentry->d_name.len) == 0)
571 invalid_alias = alias;
572 spin_unlock(&alias->d_lock);
577 alias = invalid_alias ?: discon_alias ?: NULL;
579 spin_lock(&alias->d_lock);
581 spin_unlock(&alias->d_lock);
583 spin_unlock(&inode->i_lock);
589 * Similar to d_splice_alias(), but lustre treats invalid alias
590 * similar to DCACHE_DISCONNECTED, and tries to use it anyway.
592 struct dentry *ll_splice_alias(struct inode *inode, struct dentry *de)
598 new = ll_find_alias(inode, de);
608 "Reuse dentry %p inode %p refc %d flags %#x\n",
609 new, new->d_inode, ll_d_count(new), new->d_flags);
617 CDEBUG(D_DENTRY, "Add dentry %p inode %p refc %d flags %#x\n",
618 de, de->d_inode, ll_d_count(de), de->d_flags);
622 static int ll_lookup_it_finish(struct ptlrpc_request *request,
623 struct lookup_intent *it,
624 struct inode *parent, struct dentry **de,
625 void *secctx, __u32 secctxlen,
626 void *encctx, __u32 encctxlen,
627 ktime_t kstart, bool encrypt)
629 struct inode *inode = NULL;
632 struct dentry *alias;
635 /* NB 1 request reference will be taken away by ll_intent_lock()
637 CDEBUG(D_DENTRY, "it %p it_disposition %x\n", it,
639 if (!it_disposition(it, DISP_LOOKUP_NEG)) {
640 struct req_capsule *pill = &request->rq_pill;
641 struct mdt_body *body = req_capsule_server_get(pill,
644 rc = ll_prep_inode(&inode, request, (*de)->d_sb, it);
648 /* If encryption context was returned by MDT, put it in
649 * inode now to save an extra getxattr and avoid deadlock.
651 if (body->mbo_valid & OBD_MD_ENCCTX) {
652 encctx = req_capsule_server_get(pill, &RMF_FILE_ENCCTX);
653 encctxlen = req_capsule_get_size(pill,
659 "server returned encryption ctx for "DFID"\n",
660 PFID(ll_inode2fid(inode)));
661 rc = ll_xattr_cache_insert(inode,
662 LL_XATTR_NAME_ENCRYPTION_CONTEXT,
665 CWARN("%s: cannot set enc ctx for "DFID": rc = %d\n",
666 ll_i2sbi(inode)->ll_fsname,
667 PFID(ll_inode2fid(inode)), rc);
669 rc = llcrypt_get_encryption_info(inode);
672 "cannot get enc info for "DFID": rc = %d\n",
673 PFID(ll_inode2fid(inode)), rc);
678 ll_set_lock_data(ll_i2sbi(parent)->ll_md_exp, inode, it, &bits);
679 /* OPEN can return data if lock has DoM+LAYOUT bits set */
680 if (it->it_op & IT_OPEN &&
681 bits & MDS_INODELOCK_DOM && bits & MDS_INODELOCK_LAYOUT)
682 ll_dom_finish_open(inode, request);
684 /* We used to query real size from OSTs here, but actually
685 * this is not needed. For stat() calls size would be updated
686 * from subsequent do_revalidate()->ll_inode_revalidate_it() in
688 * vfs_getattr_it->ll_getattr()->ll_inode_revalidate_it() in 2.6
689 * Everybody else who needs correct file size would call
690 * ll_glimpse_size or some equivalent themselves anyway.
694 /* If security context was returned by MDT, put it in
695 * inode now to save an extra getxattr from security hooks,
696 * and avoid deadlock.
698 if (body->mbo_valid & OBD_MD_SECCTX) {
699 secctx = req_capsule_server_get(pill, &RMF_FILE_SECCTX);
700 secctxlen = req_capsule_get_size(pill,
705 CDEBUG(D_SEC, "server returned security context"
707 PFID(ll_inode2fid(inode)));
710 if (secctx != NULL && secctxlen != 0) {
711 /* no need to protect selinux_inode_setsecurity() by
712 * inode_lock. Taking it would lead to a client deadlock
715 rc = security_inode_notifysecctx(inode, secctx,
718 CWARN("%s: cannot set security context for "DFID": rc = %d\n",
719 ll_i2sbi(inode)->ll_fsname,
720 PFID(ll_inode2fid(inode)),
725 /* Only hash *de if it is unhashed (new dentry).
726 * Atoimc_open may passin hashed dentries for open.
728 alias = ll_splice_alias(inode, *de);
730 GOTO(out, rc = PTR_ERR(alias));
734 if (!it_disposition(it, DISP_LOOKUP_NEG)) {
735 /* we have lookup look - unhide dentry */
736 if (bits & MDS_INODELOCK_LOOKUP)
737 d_lustre_revalidate(*de);
740 rc = llcrypt_get_encryption_info(inode);
743 if (!llcrypt_has_encryption_key(inode))
744 GOTO(out, rc = -ENOKEY);
746 } else if (!it_disposition(it, DISP_OPEN_CREATE)) {
748 * If file was created on the server, the dentry is revalidated
749 * in ll_create_it if the lock allows for it.
751 /* Check that parent has UPDATE lock. */
752 struct lookup_intent parent_it = {
754 .it_lock_handle = 0 };
755 struct lu_fid fid = ll_i2info(parent)->lli_fid;
757 /* If it is striped directory, get the real stripe parent */
758 if (unlikely(ll_dir_striped(parent))) {
759 rc = md_get_fid_from_lsm(ll_i2mdexp(parent),
760 ll_i2info(parent)->lli_lsm_md,
762 (*de)->d_name.len, &fid);
767 if (md_revalidate_lock(ll_i2mdexp(parent), &parent_it, &fid,
769 d_lustre_revalidate(*de);
770 ll_intent_release(&parent_it);
774 if (it_disposition(it, DISP_OPEN_CREATE)) {
775 ll_stats_ops_tally(ll_i2sbi(parent), LPROC_LL_MKNOD,
776 ktime_us_delta(ktime_get(), kstart));
782 if (rc != 0 && it->it_op & IT_OPEN) {
783 ll_intent_drop_lock(it);
784 ll_open_cleanup((*de)->d_sb, request);
790 static struct dentry *ll_lookup_it(struct inode *parent, struct dentry *dentry,
791 struct lookup_intent *it,
792 void **secctx, __u32 *secctxlen,
793 struct pcc_create_attach *pca,
795 void **encctx, __u32 *encctxlen)
797 ktime_t kstart = ktime_get();
798 struct lookup_intent lookup_it = { .it_op = IT_LOOKUP };
799 struct dentry *save = dentry, *retval;
800 struct ptlrpc_request *req = NULL;
801 struct md_op_data *op_data = NULL;
802 struct lov_user_md *lum = NULL;
805 char secctx_name[XATTR_NAME_MAX + 1];
809 if (dentry->d_name.len > ll_i2sbi(parent)->ll_namelen)
810 RETURN(ERR_PTR(-ENAMETOOLONG));
812 CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p), intent=%s\n",
813 dentry, PFID(ll_inode2fid(parent)), parent, LL_IT2STR(it));
815 if (d_mountpoint(dentry))
816 CERROR("Tell Peter, lookup on mtpt, it %s\n", LL_IT2STR(it));
818 if (it == NULL || it->it_op == IT_GETXATTR)
821 if (it->it_op == IT_GETATTR && dentry_may_statahead(parent, dentry)) {
822 rc = ll_revalidate_statahead(parent, &dentry, 0);
824 RETURN(dentry == save ? NULL : dentry);
827 if (it->it_op & IT_OPEN && it->it_flags & FMODE_WRITE &&
828 dentry->d_sb->s_flags & SB_RDONLY)
829 RETURN(ERR_PTR(-EROFS));
831 if (it->it_op & IT_CREAT)
832 opc = LUSTRE_OPC_CREATE;
834 opc = LUSTRE_OPC_ANY;
836 op_data = ll_prep_md_op_data(NULL, parent, NULL, dentry->d_name.name,
837 dentry->d_name.len, 0, opc, NULL);
839 GOTO(out, retval = ERR_CAST(op_data));
841 /* enforce umask if acl disabled or MDS doesn't support umask */
842 if (!IS_POSIXACL(parent) || !exp_connect_umask(ll_i2mdexp(parent)))
843 it->it_create_mode &= ~current_umask();
845 if (it->it_op & IT_CREAT &&
846 ll_i2sbi(parent)->ll_flags & LL_SBI_FILE_SECCTX) {
847 rc = ll_dentry_init_security(dentry, it->it_create_mode,
849 &op_data->op_file_secctx_name,
850 &op_data->op_file_secctx,
851 &op_data->op_file_secctx_size);
853 GOTO(out, retval = ERR_PTR(rc));
855 *secctx = op_data->op_file_secctx;
856 if (secctxlen != NULL)
857 *secctxlen = op_data->op_file_secctx_size;
861 if (secctxlen != NULL)
864 if (it->it_op & IT_CREAT && encrypt) {
865 rc = llcrypt_inherit_context(parent, NULL, op_data, false);
867 GOTO(out, retval = ERR_PTR(rc));
869 *encctx = op_data->op_file_encctx;
870 if (encctxlen != NULL)
871 *encctxlen = op_data->op_file_encctx_size;
875 if (encctxlen != NULL)
879 /* ask for security context upon intent */
880 if (it->it_op & (IT_LOOKUP | IT_GETATTR | IT_OPEN)) {
881 /* get name of security xattr to request to server */
882 rc = ll_listsecurity(parent, secctx_name,
883 sizeof(secctx_name));
885 CDEBUG(D_SEC, "cannot get security xattr name for "
887 PFID(ll_inode2fid(parent)), rc);
889 op_data->op_file_secctx_name = secctx_name;
890 op_data->op_file_secctx_name_size = rc;
891 CDEBUG(D_SEC, "'%.*s' is security xattr for "DFID"\n",
892 rc, secctx_name, PFID(ll_inode2fid(parent)));
896 if (pca && pca->pca_dataset) {
897 struct pcc_dataset *dataset = pca->pca_dataset;
901 GOTO(out, retval = ERR_PTR(-ENOMEM));
903 lum->lmm_magic = LOV_USER_MAGIC_V1;
904 lum->lmm_pattern = LOV_PATTERN_F_RELEASED | LOV_PATTERN_RAID0;
905 op_data->op_data = lum;
906 op_data->op_data_size = sizeof(*lum);
907 op_data->op_archive_id = dataset->pccd_rwid;
909 rc = obd_fid_alloc(NULL, ll_i2mdexp(parent), &op_data->op_fid2,
912 GOTO(out, retval = ERR_PTR(rc));
914 rc = pcc_inode_create(parent->i_sb, dataset, &op_data->op_fid2,
917 GOTO(out, retval = ERR_PTR(rc));
919 it->it_flags |= MDS_OPEN_PCC;
922 rc = md_intent_lock(ll_i2mdexp(parent), op_data, it, &req,
923 &ll_md_blocking_ast, 0);
924 /* If the MDS allows the client to chgrp (CFS_SETGRP_PERM), but the
925 * client does not know which suppgid should be sent to the MDS, or
926 * some other(s) changed the target file's GID after this RPC sent
927 * to the MDS with the suppgid as the original GID, then we should
928 * try again with right suppgid. */
929 if (rc == -EACCES && it->it_op & IT_OPEN &&
930 it_disposition(it, DISP_OPEN_DENY)) {
931 struct mdt_body *body;
933 LASSERT(req != NULL);
935 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
936 if (op_data->op_suppgids[0] == body->mbo_gid ||
937 op_data->op_suppgids[1] == body->mbo_gid ||
938 !in_group_p(make_kgid(&init_user_ns, body->mbo_gid)))
939 GOTO(out, retval = ERR_PTR(-EACCES));
941 fid_zero(&op_data->op_fid2);
942 op_data->op_suppgids[1] = body->mbo_gid;
943 ptlrpc_req_finished(req);
945 ll_intent_release(it);
946 rc = md_intent_lock(ll_i2mdexp(parent), op_data, it, &req,
947 &ll_md_blocking_ast, 0);
951 GOTO(out, retval = ERR_PTR(rc));
953 /* dir layout may change */
954 ll_unlock_md_op_lsm(op_data);
955 rc = ll_lookup_it_finish(req, it, parent, &dentry,
956 secctx != NULL ? *secctx : NULL,
957 secctxlen != NULL ? *secctxlen : 0,
958 encctx != NULL ? *encctx : NULL,
959 encctxlen != NULL ? *encctxlen : 0,
962 ll_intent_release(it);
963 GOTO(out, retval = ERR_PTR(rc));
966 if ((it->it_op & IT_OPEN) && dentry->d_inode &&
967 !S_ISREG(dentry->d_inode->i_mode) &&
968 !S_ISDIR(dentry->d_inode->i_mode)) {
969 ll_release_openhandle(dentry, it);
971 ll_lookup_finish_locks(it, dentry);
973 GOTO(out, retval = (dentry == save) ? NULL : dentry);
976 if (op_data != NULL && !IS_ERR(op_data)) {
977 if (secctx != NULL && secctxlen != NULL) {
978 /* caller needs sec ctx info, so reset it in op_data to
979 * prevent it from being freed */
980 op_data->op_file_secctx = NULL;
981 op_data->op_file_secctx_size = 0;
983 if (encctx != NULL && encctxlen != NULL &&
984 it->it_op & IT_CREAT && encrypt) {
985 /* caller needs enc ctx info, so reset it in op_data to
986 * prevent it from being freed
988 op_data->op_file_encctx = NULL;
989 op_data->op_file_encctx_size = 0;
991 ll_finish_md_op_data(op_data);
997 ptlrpc_req_finished(req);
1001 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
1004 struct lookup_intent *itp, it = { .it_op = IT_GETATTR };
1007 CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p), flags=%u\n",
1008 dentry, PFID(ll_inode2fid(parent)), parent, flags);
1011 * Optimize away (CREATE && !OPEN). Let .create handle the race.
1012 * but only if we have write permissions there, otherwise we need
1013 * to proceed with lookup. LU-4185
1015 if ((flags & LOOKUP_CREATE) && !(flags & LOOKUP_OPEN) &&
1016 (inode_permission(parent, MAY_WRITE | MAY_EXEC) == 0))
1019 if (flags & (LOOKUP_PARENT|LOOKUP_OPEN|LOOKUP_CREATE))
1023 de = ll_lookup_it(parent, dentry, itp, NULL, NULL, NULL, false,
1027 ll_intent_release(itp);
1032 #ifdef FMODE_CREATED /* added in Linux v4.18-rc1-20-g73a09dd */
1033 # define ll_is_opened(o, f) ((f)->f_mode & FMODE_OPENED)
1034 # define ll_finish_open(f, d, o) finish_open((f), (d), NULL)
1035 # define ll_last_arg
1036 # define ll_set_created(o, f) \
1038 (f)->f_mode |= FMODE_CREATED; \
1042 # define ll_is_opened(o, f) (*(o))
1043 # define ll_finish_open(f, d, o) finish_open((f), (d), NULL, (o))
1044 # define ll_last_arg , int *opened
1045 # define ll_set_created(o, f) \
1047 *(o) |= FILE_CREATED; \
1053 * For cached negative dentry and new dentry, handle lookup/create/open
1056 static int ll_atomic_open(struct inode *dir, struct dentry *dentry,
1057 struct file *file, unsigned open_flags,
1058 umode_t mode ll_last_arg)
1060 struct lookup_intent *it;
1062 long long lookup_flags = LOOKUP_OPEN;
1063 void *secctx = NULL;
1064 __u32 secctxlen = 0;
1065 void *encctx = NULL;
1066 __u32 encctxlen = 0;
1067 struct ll_sb_info *sbi = NULL;
1068 struct pcc_create_attach pca = { NULL, NULL };
1069 bool encrypt = false;
1074 "VFS Op:name=%pd, dir="DFID"(%p), file %p, open_flags %x, mode %x opened %d\n",
1075 dentry, PFID(ll_inode2fid(dir)), dir, file, open_flags, mode,
1076 ll_is_opened(opened, file));
1078 /* Only negative dentries enter here */
1079 LASSERT(dentry->d_inode == NULL);
1081 if (!d_unhashed(dentry)) {
1082 /* A valid negative dentry that just passed revalidation,
1083 * there's little point to try and open it server-side,
1084 * even though there's a minuscule chance it might succeed.
1085 * Either way it's a valid race to just return -ENOENT here.
1087 if (!(open_flags & O_CREAT))
1090 /* Otherwise we just unhash it to be rehashed afresh via
1091 * lookup if necessary
1096 OBD_ALLOC(it, sizeof(*it));
1100 it->it_op = IT_OPEN;
1101 if (open_flags & O_CREAT) {
1102 it->it_op |= IT_CREAT;
1103 lookup_flags |= LOOKUP_CREATE;
1104 sbi = ll_i2sbi(dir);
1105 /* Volatile file is used for HSM restore, so do not use PCC */
1106 if (!filename_is_volatile(dentry->d_name.name,
1107 dentry->d_name.len, NULL)) {
1108 struct pcc_matcher item;
1109 struct pcc_dataset *dataset;
1111 item.pm_uid = from_kuid(&init_user_ns, current_uid());
1112 item.pm_gid = from_kgid(&init_user_ns, current_gid());
1113 item.pm_projid = ll_i2info(dir)->lli_projid;
1114 item.pm_name = &dentry->d_name;
1115 dataset = pcc_dataset_match_get(&sbi->ll_pcc_super,
1117 pca.pca_dataset = dataset;
1120 it->it_create_mode = (mode & S_IALLUGO) | S_IFREG;
1121 it->it_flags = (open_flags & ~O_ACCMODE) | OPEN_FMODE(open_flags);
1122 it->it_flags &= ~MDS_OPEN_FL_INTERNAL;
1124 if (ll_sbi_has_encrypt(ll_i2sbi(dir)) && IS_ENCRYPTED(dir)) {
1125 /* in case of create, this is going to be a regular file because
1126 * we set S_IFREG bit on it->it_create_mode above
1128 rc = llcrypt_get_encryption_info(dir);
1130 GOTO(out_release, rc);
1131 if (open_flags & O_CREAT) {
1132 if (!llcrypt_has_encryption_key(dir))
1133 GOTO(out_release, rc = -ENOKEY);
1138 OBD_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_FILE_PAUSE2, cfs_fail_val);
1140 /* Dentry added to dcache tree in ll_lookup_it */
1141 de = ll_lookup_it(dir, dentry, it, &secctx, &secctxlen, &pca, encrypt,
1142 &encctx, &encctxlen);
1145 else if (de != NULL)
1148 CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_FILE_PAUSE, cfs_fail_val);
1151 if (it_disposition(it, DISP_OPEN_CREATE)) {
1152 /* Dentry instantiated in ll_create_it. */
1153 rc = ll_create_it(dir, dentry, it, secctx, secctxlen,
1154 encrypt, encctx, encctxlen);
1155 security_release_secctx(secctx, secctxlen);
1156 llcrypt_free_ctx(encctx, encctxlen);
1158 /* We dget in ll_splice_alias. */
1164 rc = pcc_inode_create_fini(dentry->d_inode, &pca);
1168 GOTO(out_release, rc);
1171 ll_set_created(opened, file);
1173 /* Open the file with O_CREAT, but the file already
1174 * existed on MDT. This may happend in the case that
1175 * the LOOKUP ibits lock is revoked and the
1176 * corresponding dentry cache is deleted.
1177 * i.e. In the current Lustre, the truncate operation
1178 * will revoke the LOOKUP ibits lock, and the file
1179 * dentry cache will be invalidated. The following open
1180 * with O_CREAT flag will call into ->atomic_open, the
1181 * file was wrongly though as newly created file and
1182 * try to auto cache the file. So after client knows it
1183 * is not a DISP_OPEN_CREATE, it should cleanup the
1184 * already created PCC copy.
1186 pcc_create_attach_cleanup(dir->i_sb, &pca);
1188 if (open_flags & O_CREAT && encrypt &&
1190 rc = ll_set_encflags(dentry->d_inode, encctx,
1192 llcrypt_free_ctx(encctx, encctxlen);
1194 GOTO(out_release, rc);
1198 if (dentry->d_inode && it_disposition(it, DISP_OPEN_OPEN)) {
1200 if (S_ISFIFO(dentry->d_inode->i_mode)) {
1201 /* We cannot call open here as it might
1202 * deadlock. This case is unreachable in
1203 * practice because of OBD_CONNECT_NODEVOH. */
1204 rc = finish_no_open(file, de);
1206 file->private_data = it;
1207 rc = ll_finish_open(file, dentry, opened);
1208 /* We dget in ll_splice_alias. finish_open takes
1209 * care of dget for fd open.
1215 rc = finish_no_open(file, de);
1218 pcc_create_attach_cleanup(dir->i_sb, &pca);
1222 ll_intent_release(it);
1223 OBD_FREE(it, sizeof(*it));
1228 /* We depend on "mode" being set with the proper file type/umask by now */
1229 static struct inode *ll_create_node(struct inode *dir, struct lookup_intent *it)
1231 struct inode *inode = NULL;
1232 struct ptlrpc_request *request = NULL;
1233 struct ll_sb_info *sbi = ll_i2sbi(dir);
1237 LASSERT(it && it->it_disposition);
1239 LASSERT(it_disposition(it, DISP_ENQ_CREATE_REF));
1240 request = it->it_request;
1241 it_clear_disposition(it, DISP_ENQ_CREATE_REF);
1242 rc = ll_prep_inode(&inode, request, dir->i_sb, it);
1244 GOTO(out, inode = ERR_PTR(rc));
1246 /* Pause to allow for a race with concurrent access by fid */
1247 OBD_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_NODE_PAUSE, cfs_fail_val);
1249 /* We asked for a lock on the directory, but were granted a
1250 * lock on the inode. Since we finally have an inode pointer,
1251 * stuff it in the lock. */
1252 CDEBUG(D_DLMTRACE, "setting l_ast_data to inode "DFID"(%p)\n",
1253 PFID(ll_inode2fid(inode)), inode);
1254 ll_set_lock_data(sbi->ll_md_exp, inode, it, NULL);
1257 ptlrpc_req_finished(request);
1262 * By the time this is called, we already have created the directory cache
1263 * entry for the new file, but it is so far negative - it has no inode.
1265 * We defer creating the OBD object(s) until open, to keep the intent and
1266 * non-intent code paths similar, and also because we do not have the MDS
1267 * inode number before calling ll_create_node() (which is needed for LOV),
1268 * so we would need to do yet another RPC to the MDS to store the LOV EA
1269 * data on the MDS. If needed, we would pass the PACKED lmm as data and
1270 * lmm_size in datalen (the MDS still has code which will handle that).
1272 * If the create succeeds, we fill in the inode information
1273 * with d_instantiate().
1275 static int ll_create_it(struct inode *dir, struct dentry *dentry,
1276 struct lookup_intent *it,
1277 void *secctx, __u32 secctxlen, bool encrypt,
1278 void *encctx, __u32 encctxlen)
1280 struct inode *inode;
1285 CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p), intent=%s\n",
1286 dentry, PFID(ll_inode2fid(dir)), dir, LL_IT2STR(it));
1288 rc = it_open_error(DISP_OPEN_CREATE, it);
1292 inode = ll_create_node(dir, it);
1294 RETURN(PTR_ERR(inode));
1296 if ((ll_i2sbi(inode)->ll_flags & LL_SBI_FILE_SECCTX) &&
1298 /* must be done before d_instantiate, because it calls
1299 * security_d_instantiate, which means a getxattr if security
1300 * context is not set yet */
1301 /* no need to protect selinux_inode_setsecurity() by
1302 * inode_lock. Taking it would lead to a client deadlock
1305 rc = security_inode_notifysecctx(inode, secctx, secctxlen);
1310 d_instantiate(dentry, inode);
1313 rc = ll_set_encflags(inode, encctx, encctxlen, true);
1318 if (!(ll_i2sbi(inode)->ll_flags & LL_SBI_FILE_SECCTX)) {
1319 rc = ll_inode_init_security(dentry, inode, dir);
1324 ll_set_lock_data(ll_i2sbi(dir)->ll_md_exp, inode, it, &bits);
1325 if (bits & MDS_INODELOCK_LOOKUP)
1326 d_lustre_revalidate(dentry);
1331 void ll_update_times(struct ptlrpc_request *request, struct inode *inode)
1333 struct mdt_body *body = req_capsule_server_get(&request->rq_pill,
1337 if (body->mbo_valid & OBD_MD_FLMTIME &&
1338 body->mbo_mtime > inode->i_mtime.tv_sec) {
1340 "setting fid " DFID " mtime from %lld to %llu\n",
1341 PFID(ll_inode2fid(inode)),
1342 (s64)inode->i_mtime.tv_sec, body->mbo_mtime);
1343 inode->i_mtime.tv_sec = body->mbo_mtime;
1346 if (body->mbo_valid & OBD_MD_FLCTIME &&
1347 body->mbo_ctime > inode->i_ctime.tv_sec)
1348 inode->i_ctime.tv_sec = body->mbo_ctime;
1351 static int ll_new_node(struct inode *dir, struct dentry *dchild,
1352 const char *tgt, umode_t mode, int rdev, __u32 opc)
1354 struct qstr *name = &dchild->d_name;
1355 struct ptlrpc_request *request = NULL;
1356 struct md_op_data *op_data = NULL;
1357 struct inode *inode = NULL;
1358 struct ll_sb_info *sbi = ll_i2sbi(dir);
1360 bool encrypt = false;
1364 if (unlikely(tgt != NULL))
1365 tgt_len = strlen(tgt) + 1;
1368 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
1369 name->len, 0, opc, NULL);
1370 if (IS_ERR(op_data))
1371 GOTO(err_exit, err = PTR_ERR(op_data));
1373 if (sbi->ll_flags & LL_SBI_FILE_SECCTX) {
1374 err = ll_dentry_init_security(dchild, mode, &dchild->d_name,
1375 &op_data->op_file_secctx_name,
1376 &op_data->op_file_secctx,
1377 &op_data->op_file_secctx_size);
1379 GOTO(err_exit, err);
1382 if (ll_sbi_has_encrypt(sbi) &&
1383 ((IS_ENCRYPTED(dir) &&
1384 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) ||
1385 (unlikely(llcrypt_dummy_context_enabled(dir)) && S_ISDIR(mode)))) {
1386 err = llcrypt_get_encryption_info(dir);
1388 GOTO(err_exit, err);
1389 if (!llcrypt_has_encryption_key(dir))
1390 GOTO(err_exit, err = -ENOKEY);
1395 err = llcrypt_inherit_context(dir, NULL, op_data, false);
1397 GOTO(err_exit, err);
1400 err = md_create(sbi->ll_md_exp, op_data, tgt, tgt_len, mode,
1401 from_kuid(&init_user_ns, current_fsuid()),
1402 from_kgid(&init_user_ns, current_fsgid()),
1403 cfs_curproc_cap_pack(), rdev, &request);
1404 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 14, 58, 0)
1406 * server < 2.12.58 doesn't pack default LMV in intent_getattr reply,
1407 * fetch default LMV here.
1409 if (unlikely(err == -EREMOTE)) {
1410 struct ll_inode_info *lli = ll_i2info(dir);
1411 struct lmv_user_md *lum;
1415 ptlrpc_req_finished(request);
1418 err2 = ll_dir_getstripe(dir, (void **)&lum, &lumsize, &request,
1419 OBD_MD_DEFAULT_MEA);
1420 ll_finish_md_op_data(op_data);
1423 struct lustre_md md = { NULL };
1425 md.body = req_capsule_server_get(&request->rq_pill,
1428 GOTO(err_exit, err = -EPROTO);
1430 OBD_ALLOC_PTR(md.default_lmv);
1431 if (!md.default_lmv)
1432 GOTO(err_exit, err = -ENOMEM);
1434 md.default_lmv->lsm_md_magic = lum->lum_magic;
1435 md.default_lmv->lsm_md_stripe_count =
1436 lum->lum_stripe_count;
1437 md.default_lmv->lsm_md_master_mdt_index =
1438 lum->lum_stripe_offset;
1439 md.default_lmv->lsm_md_hash_type = lum->lum_hash_type;
1441 err = ll_update_inode(dir, &md);
1442 md_free_lustre_md(sbi->ll_md_exp, &md);
1444 GOTO(err_exit, err);
1445 } else if (err2 == -ENODATA && lli->lli_default_lsm_md) {
1447 * If there are no default stripe EA on the MDT, but the
1448 * client has default stripe, then it probably means
1449 * default stripe EA has just been deleted.
1451 down_write(&lli->lli_lsm_sem);
1452 if (lli->lli_default_lsm_md)
1453 OBD_FREE_PTR(lli->lli_default_lsm_md);
1454 lli->lli_default_lsm_md = NULL;
1455 up_write(&lli->lli_lsm_sem);
1457 GOTO(err_exit, err);
1460 ptlrpc_req_finished(request);
1467 GOTO(err_exit, err);
1469 ll_update_times(request, dir);
1471 CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_NEWNODE_PAUSE, cfs_fail_val);
1473 err = ll_prep_inode(&inode, request, dchild->d_sb, NULL);
1475 GOTO(err_exit, err);
1477 if (sbi->ll_flags & LL_SBI_FILE_SECCTX) {
1478 /* must be done before d_instantiate, because it calls
1479 * security_d_instantiate, which means a getxattr if security
1480 * context is not set yet */
1481 /* no need to protect selinux_inode_setsecurity() by
1482 * inode_lock. Taking it would lead to a client deadlock
1485 err = security_inode_notifysecctx(inode,
1486 op_data->op_file_secctx,
1487 op_data->op_file_secctx_size);
1489 GOTO(err_exit, err);
1492 d_instantiate(dchild, inode);
1495 err = ll_set_encflags(inode, op_data->op_file_encctx,
1496 op_data->op_file_encctx_size, true);
1498 GOTO(err_exit, err);
1501 if (!(sbi->ll_flags & LL_SBI_FILE_SECCTX)) {
1502 err = ll_inode_init_security(dchild, inode, dir);
1504 GOTO(err_exit, err);
1509 if (request != NULL)
1510 ptlrpc_req_finished(request);
1512 if (!IS_ERR_OR_NULL(op_data))
1513 ll_finish_md_op_data(op_data);
1518 static int ll_mknod(struct inode *dir, struct dentry *dchild, umode_t mode,
1521 ktime_t kstart = ktime_get();
1525 CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p) mode %o dev %x\n",
1526 dchild, PFID(ll_inode2fid(dir)), dir, mode, rdev);
1528 if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
1529 mode &= ~current_umask();
1531 switch (mode & S_IFMT) {
1540 err = ll_new_node(dir, dchild, NULL, mode, old_encode_dev(rdev),
1551 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKNOD,
1552 ktime_us_delta(ktime_get(), kstart));
1558 * Plain create. Intent create is handled in atomic_open.
1560 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
1561 umode_t mode, bool want_excl)
1563 ktime_t kstart = ktime_get();
1566 CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_FILE_PAUSE, cfs_fail_val);
1569 "VFS Op:name=%pd, dir="DFID"(%p), flags=%u, excl=%d\n",
1570 dentry, PFID(ll_inode2fid(dir)), dir, mode, want_excl);
1572 /* Using mknod(2) to create a regular file is designed to not recognize
1573 * volatile file name, so we use ll_mknod() here. */
1574 rc = ll_mknod(dir, dentry, mode, 0);
1576 CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, unhashed %d\n",
1577 dentry, d_unhashed(dentry));
1580 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_CREATE,
1581 ktime_us_delta(ktime_get(), kstart));
1586 static int ll_symlink(struct inode *dir, struct dentry *dchild,
1587 const char *oldpath)
1589 ktime_t kstart = ktime_get();
1593 CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p), target=%.*s\n",
1594 dchild, PFID(ll_inode2fid(dir)), dir, 3000, oldpath);
1596 err = ll_new_node(dir, dchild, oldpath, S_IFLNK | S_IRWXUGO, 0,
1597 LUSTRE_OPC_SYMLINK);
1600 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_SYMLINK,
1601 ktime_us_delta(ktime_get(), kstart));
1606 static int ll_link(struct dentry *old_dentry, struct inode *dir,
1607 struct dentry *new_dentry)
1609 struct inode *src = old_dentry->d_inode;
1610 struct qstr *name = &new_dentry->d_name;
1611 struct ll_sb_info *sbi = ll_i2sbi(dir);
1612 struct ptlrpc_request *request = NULL;
1613 struct md_op_data *op_data;
1614 ktime_t kstart = ktime_get();
1619 "VFS Op: inode="DFID"(%p), dir="DFID"(%p), target=%pd\n",
1620 PFID(ll_inode2fid(src)), src,
1621 PFID(ll_inode2fid(dir)), dir, new_dentry);
1623 err = llcrypt_prepare_link(old_dentry, dir, new_dentry);
1627 op_data = ll_prep_md_op_data(NULL, src, dir, name->name, name->len,
1628 0, LUSTRE_OPC_ANY, NULL);
1629 if (IS_ERR(op_data))
1630 RETURN(PTR_ERR(op_data));
1632 err = md_link(sbi->ll_md_exp, op_data, &request);
1633 ll_finish_md_op_data(op_data);
1637 ll_update_times(request, dir);
1638 ll_stats_ops_tally(sbi, LPROC_LL_LINK,
1639 ktime_us_delta(ktime_get(), kstart));
1642 ptlrpc_req_finished(request);
1646 static int ll_mkdir(struct inode *dir, struct dentry *dchild, umode_t mode)
1648 ktime_t kstart = ktime_get();
1652 CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p)\n",
1653 dchild, PFID(ll_inode2fid(dir)), dir);
1655 if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
1656 mode &= ~current_umask();
1658 mode = (mode & (S_IRWXUGO|S_ISVTX)) | S_IFDIR;
1660 err = ll_new_node(dir, dchild, NULL, mode, 0, LUSTRE_OPC_MKDIR);
1662 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKDIR,
1663 ktime_us_delta(ktime_get(), kstart));
1668 static int ll_rmdir(struct inode *dir, struct dentry *dchild)
1670 struct qstr *name = &dchild->d_name;
1671 struct ptlrpc_request *request = NULL;
1672 struct md_op_data *op_data;
1673 ktime_t kstart = ktime_get();
1678 CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p)\n",
1679 dchild, PFID(ll_inode2fid(dir)), dir);
1681 if (unlikely(d_mountpoint(dchild)))
1684 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len,
1685 S_IFDIR, LUSTRE_OPC_ANY, NULL);
1686 if (IS_ERR(op_data))
1687 RETURN(PTR_ERR(op_data));
1689 if (dchild->d_inode != NULL)
1690 op_data->op_fid3 = *ll_inode2fid(dchild->d_inode);
1692 op_data->op_fid2 = op_data->op_fid3;
1693 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1694 ll_finish_md_op_data(op_data);
1696 struct mdt_body *body;
1698 ll_update_times(request, dir);
1699 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR,
1700 ktime_us_delta(ktime_get(), kstart));
1703 * The server puts attributes in on the last unlink, use them
1704 * to update the link count so the inode can be freed
1707 body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
1708 if (body->mbo_valid & OBD_MD_FLNLINK)
1709 set_nlink(dchild->d_inode, body->mbo_nlink);
1712 ptlrpc_req_finished(request);
1720 int ll_rmdir_entry(struct inode *dir, char *name, int namelen)
1722 struct ptlrpc_request *request = NULL;
1723 struct md_op_data *op_data;
1724 ktime_t kstart = ktime_get();
1728 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1729 namelen, name, PFID(ll_inode2fid(dir)), dir);
1731 op_data = ll_prep_md_op_data(NULL, dir, NULL, name, strlen(name),
1732 S_IFDIR, LUSTRE_OPC_ANY, NULL);
1733 if (IS_ERR(op_data))
1734 RETURN(PTR_ERR(op_data));
1735 op_data->op_cli_flags |= CLI_RM_ENTRY;
1736 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1737 ll_finish_md_op_data(op_data);
1739 ll_update_times(request, dir);
1741 ptlrpc_req_finished(request);
1743 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR,
1744 ktime_us_delta(ktime_get(), kstart));
1748 static int ll_unlink(struct inode *dir, struct dentry *dchild)
1750 struct qstr *name = &dchild->d_name;
1751 struct ptlrpc_request *request = NULL;
1752 struct md_op_data *op_data;
1753 struct mdt_body *body;
1754 ktime_t kstart = ktime_get();
1759 CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p)\n",
1760 dchild, PFID(ll_inode2fid(dir)), dir);
1763 * XXX: unlink bind mountpoint maybe call to here,
1764 * just check it as vfs_unlink does.
1766 if (unlikely(d_mountpoint(dchild)))
1769 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len, 0,
1770 LUSTRE_OPC_ANY, NULL);
1771 if (IS_ERR(op_data))
1772 RETURN(PTR_ERR(op_data));
1774 op_data->op_fid3 = *ll_inode2fid(dchild->d_inode);
1775 /* notify lower layer if inode has dirty pages */
1776 if (S_ISREG(dchild->d_inode->i_mode) &&
1777 ll_i2info(dchild->d_inode)->lli_clob &&
1778 dirty_cnt(dchild->d_inode))
1779 op_data->op_cli_flags |= CLI_DIRTY_DATA;
1780 op_data->op_fid2 = op_data->op_fid3;
1781 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1782 ll_finish_md_op_data(op_data);
1787 * The server puts attributes in on the last unlink, use them to update
1788 * the link count so the inode can be freed immediately.
1790 body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
1791 if (body->mbo_valid & OBD_MD_FLNLINK)
1792 set_nlink(dchild->d_inode, body->mbo_nlink);
1794 ll_update_times(request, dir);
1797 ptlrpc_req_finished(request);
1799 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_UNLINK,
1800 ktime_us_delta(ktime_get(), kstart));
1804 static int ll_rename(struct inode *src, struct dentry *src_dchild,
1805 struct inode *tgt, struct dentry *tgt_dchild
1806 #ifdef HAVE_IOPS_RENAME_WITH_FLAGS
1807 , unsigned int flags
1811 struct qstr *src_name = &src_dchild->d_name;
1812 struct qstr *tgt_name = &tgt_dchild->d_name;
1813 struct ptlrpc_request *request = NULL;
1814 struct ll_sb_info *sbi = ll_i2sbi(src);
1815 struct md_op_data *op_data;
1816 ktime_t kstart = ktime_get();
1821 #ifdef HAVE_IOPS_RENAME_WITH_FLAGS
1827 "VFS Op:oldname=%pd, src_dir="DFID"(%p), newname=%pd, tgt_dir="DFID"(%p)\n",
1828 src_dchild, PFID(ll_inode2fid(src)), src,
1829 tgt_dchild, PFID(ll_inode2fid(tgt)), tgt);
1831 if (unlikely(d_mountpoint(src_dchild) || d_mountpoint(tgt_dchild)))
1834 #ifdef HAVE_IOPS_RENAME_WITH_FLAGS
1835 err = llcrypt_prepare_rename(src, src_dchild, tgt, tgt_dchild, flags);
1837 err = llcrypt_prepare_rename(src, src_dchild, tgt, tgt_dchild, 0);
1842 if (src_dchild->d_inode)
1843 mode = src_dchild->d_inode->i_mode;
1845 if (tgt_dchild->d_inode)
1846 mode = tgt_dchild->d_inode->i_mode;
1848 op_data = ll_prep_md_op_data(NULL, src, tgt, NULL, 0, mode,
1849 LUSTRE_OPC_ANY, NULL);
1850 if (IS_ERR(op_data))
1851 RETURN(PTR_ERR(op_data));
1853 if (src_dchild->d_inode)
1854 op_data->op_fid3 = *ll_inode2fid(src_dchild->d_inode);
1856 if (tgt_dchild->d_inode)
1857 op_data->op_fid4 = *ll_inode2fid(tgt_dchild->d_inode);
1859 err = md_rename(sbi->ll_md_exp, op_data,
1860 src_name->name, src_name->len,
1861 tgt_name->name, tgt_name->len, &request);
1862 ll_finish_md_op_data(op_data);
1864 ll_update_times(request, src);
1865 ll_update_times(request, tgt);
1868 ptlrpc_req_finished(request);
1871 d_move(src_dchild, tgt_dchild);
1872 ll_stats_ops_tally(sbi, LPROC_LL_RENAME,
1873 ktime_us_delta(ktime_get(), kstart));
1879 const struct inode_operations ll_dir_inode_operations = {
1881 .atomic_open = ll_atomic_open,
1882 .lookup = ll_lookup_nd,
1883 .create = ll_create_nd,
1884 /* We need all these non-raw things for NFSD, to not patch it. */
1885 .unlink = ll_unlink,
1888 .symlink = ll_symlink,
1890 .rename = ll_rename,
1891 .setattr = ll_setattr,
1892 .getattr = ll_getattr,
1893 .permission = ll_inode_permission,
1894 #ifdef HAVE_IOP_XATTR
1895 .setxattr = ll_setxattr,
1896 .getxattr = ll_getxattr,
1897 .removexattr = ll_removexattr,
1899 .listxattr = ll_listxattr,
1900 .get_acl = ll_get_acl,
1901 #ifdef HAVE_IOP_SET_ACL
1902 .set_acl = ll_set_acl,
1906 const struct inode_operations ll_special_inode_operations = {
1907 .setattr = ll_setattr,
1908 .getattr = ll_getattr,
1909 .permission = ll_inode_permission,
1910 #ifdef HAVE_IOP_XATTR
1911 .setxattr = ll_setxattr,
1912 .getxattr = ll_getxattr,
1913 .removexattr = ll_removexattr,
1915 .listxattr = ll_listxattr,
1916 .get_acl = ll_get_acl,
1917 #ifdef HAVE_IOP_SET_ACL
1918 .set_acl = ll_set_acl,