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 static void ll_invalidate_negative_children(struct inode *dir)
160 struct dentry *dentry, *tmp_subdir;
162 spin_lock(&dir->i_lock);
163 hlist_for_each_entry(dentry, &dir->i_dentry, d_alias) {
164 spin_lock(&dentry->d_lock);
165 if (!list_empty(&dentry->d_subdirs)) {
166 struct dentry *child;
168 list_for_each_entry_safe(child, tmp_subdir,
171 if (child->d_inode == NULL)
172 d_lustre_invalidate(child, 1);
175 spin_unlock(&dentry->d_lock);
177 spin_unlock(&dir->i_lock);
180 int ll_test_inode_by_fid(struct inode *inode, void *opaque)
182 return lu_fid_eq(&ll_i2info(inode)->lli_fid, opaque);
185 static int ll_dom_lock_cancel(struct inode *inode, struct ldlm_lock *lock)
188 struct ll_inode_info *lli = ll_i2info(inode);
193 env = cl_env_get(&refcheck);
195 RETURN(PTR_ERR(env));
197 /* reach MDC layer to flush data under the DoM ldlm lock */
198 rc = cl_object_flush(env, lli->lli_clob, lock);
199 if (rc == -ENODATA) {
200 CDEBUG(D_INODE, "inode "DFID" layout has no DoM stripe\n",
201 PFID(ll_inode2fid(inode)));
202 /* most likely result of layout change, do nothing */
206 cl_env_put(env, &refcheck);
210 static void ll_lock_cancel_bits(struct ldlm_lock *lock, __u64 to_cancel)
212 struct inode *inode = ll_inode_from_resource_lock(lock);
213 struct ll_inode_info *lli;
214 __u64 bits = to_cancel;
220 /* That means the inode is evicted most likely and may cause
221 * the skipping of lock cleanups below, so print the message
224 if (lock->l_resource->lr_lvb_inode)
226 "can't take inode for the lock (%sevicted)\n",
227 lock->l_resource->lr_lvb_inode->i_state &
228 I_FREEING ? "" : "not ");
232 if (!fid_res_name_eq(ll_inode2fid(inode),
233 &lock->l_resource->lr_name)) {
234 LDLM_ERROR(lock, "data mismatch with object "DFID"(%p)",
235 PFID(ll_inode2fid(inode)), inode);
239 if (bits & MDS_INODELOCK_XATTR) {
240 ll_xattr_cache_destroy(inode);
241 bits &= ~MDS_INODELOCK_XATTR;
244 /* For OPEN locks we differentiate between lock modes
245 * LCK_CR, LCK_CW, LCK_PR - bug 22891 */
246 if (bits & MDS_INODELOCK_OPEN)
247 ll_have_md_lock(inode, &bits, lock->l_req_mode);
249 if (bits & MDS_INODELOCK_OPEN) {
252 switch (lock->l_req_mode) {
263 LDLM_ERROR(lock, "bad lock mode for OPEN lock");
267 ll_md_real_close(inode, fmode);
269 bits &= ~MDS_INODELOCK_OPEN;
272 if (bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
273 MDS_INODELOCK_LAYOUT | MDS_INODELOCK_PERM |
275 ll_have_md_lock(inode, &bits, LCK_MINMODE);
277 if (bits & MDS_INODELOCK_DOM) {
278 rc = ll_dom_lock_cancel(inode, lock);
280 CDEBUG(D_INODE, "cannot flush DoM data "
282 PFID(ll_inode2fid(inode)), rc);
285 if (bits & MDS_INODELOCK_LAYOUT) {
286 struct cl_object_conf conf = {
287 .coc_opc = OBJECT_CONF_INVALIDATE,
291 rc = ll_layout_conf(inode, &conf);
293 CDEBUG(D_INODE, "cannot invalidate layout of "
295 PFID(ll_inode2fid(inode)), rc);
298 lli = ll_i2info(inode);
300 if (bits & MDS_INODELOCK_UPDATE)
301 ll_file_set_flag(lli, LLIF_UPDATE_ATIME);
303 if ((bits & MDS_INODELOCK_UPDATE) && S_ISDIR(inode->i_mode)) {
304 CDEBUG(D_INODE, "invalidating inode "DFID" lli = %p, "
305 "pfid = "DFID"\n", PFID(ll_inode2fid(inode)),
306 lli, PFID(&lli->lli_pfid));
307 truncate_inode_pages(inode->i_mapping, 0);
309 if (unlikely(!fid_is_zero(&lli->lli_pfid))) {
310 struct inode *master_inode = NULL;
313 /* This is slave inode, since all of the child dentry
314 * is connected on the master inode, so we have to
315 * invalidate the negative children on master inode */
316 CDEBUG(D_INODE, "Invalidate s"DFID" m"DFID"\n",
317 PFID(ll_inode2fid(inode)), PFID(&lli->lli_pfid));
319 hash = cl_fid_build_ino(&lli->lli_pfid,
320 ll_need_32bit_api(ll_i2sbi(inode)));
322 /* Do not lookup the inode with ilookup5, otherwise
323 * it will cause dead lock,
324 * 1. Client1 send chmod req to the MDT0, then on MDT0,
325 * it enqueues master and all of its slaves lock,
326 * (mdt_attr_set() -> mdt_lock_slaves()), after gets
327 * master and stripe0 lock, it will send the enqueue
328 * req (for stripe1) to MDT1, then MDT1 finds the lock
329 * has been granted to client2. Then MDT1 sends blocking
331 * 2. At the same time, client2 tries to unlink
332 * the striped dir (rm -rf striped_dir), and during
333 * lookup, it will hold the master inode of the striped
334 * directory, whose inode state is NEW, then tries to
335 * revalidate all of its slaves, (ll_prep_inode()->
336 * ll_iget()->ll_read_inode2()-> ll_update_inode().).
337 * And it will be blocked on the server side because
339 * 3. Then the client get the blocking_ast req, cancel
340 * the lock, but being blocked if using ->ilookup5()),
341 * because master inode state is NEW. */
342 master_inode = ilookup5_nowait(inode->i_sb, hash,
343 ll_test_inode_by_fid,
344 (void *)&lli->lli_pfid);
346 ll_invalidate_negative_children(master_inode);
350 ll_invalidate_negative_children(inode);
354 if ((bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM)) &&
355 inode->i_sb->s_root != NULL &&
356 inode != inode->i_sb->s_root->d_inode)
357 ll_invalidate_aliases(inode);
359 if (bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM))
360 forget_all_cached_acls(inode);
366 /* Check if the given lock may be downgraded instead of canceling and
367 * that convert is really needed. */
368 int ll_md_need_convert(struct ldlm_lock *lock)
370 struct ldlm_namespace *ns = ldlm_lock_to_ns(lock);
372 __u64 wanted = lock->l_policy_data.l_inodebits.cancel_bits;
373 __u64 bits = lock->l_policy_data.l_inodebits.bits & ~wanted;
374 enum ldlm_mode mode = LCK_MINMODE;
376 if (!lock->l_conn_export ||
377 !exp_connect_lock_convert(lock->l_conn_export))
380 if (!wanted || !bits || ldlm_is_cancel(lock))
383 /* do not convert locks other than DOM for now */
384 if (!((bits | wanted) & MDS_INODELOCK_DOM))
387 /* We may have already remaining bits in some other lock so
388 * lock convert will leave us just extra lock for the same bit.
389 * Check if client has other lock with the same bits and the same
390 * or lower mode and don't convert if any.
392 switch (lock->l_req_mode) {
406 /* do not convert other modes */
410 /* is lock is too old to be converted? */
411 lock_res_and_lock(lock);
412 if (ktime_after(ktime_get(),
413 ktime_add(lock->l_last_used, ns->ns_dirty_age_limit))) {
414 unlock_res_and_lock(lock);
417 unlock_res_and_lock(lock);
419 inode = ll_inode_from_resource_lock(lock);
420 ll_have_md_lock(inode, &bits, mode);
425 int ll_md_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *ld,
426 void *data, int flag)
428 struct lustre_handle lockh;
434 case LDLM_CB_BLOCKING:
436 __u64 cancel_flags = LCF_ASYNC;
438 /* if lock convert is not needed then still have to
439 * pass lock via ldlm_cli_convert() to keep all states
440 * correct, set cancel_bits to full lock bits to cause
441 * full cancel to happen.
443 if (!ll_md_need_convert(lock)) {
444 lock_res_and_lock(lock);
445 lock->l_policy_data.l_inodebits.cancel_bits =
446 lock->l_policy_data.l_inodebits.bits;
447 unlock_res_and_lock(lock);
449 rc = ldlm_cli_convert(lock, cancel_flags);
452 /* continue with cancel otherwise */
453 ldlm_lock2handle(lock, &lockh);
454 rc = ldlm_cli_cancel(&lockh, cancel_flags);
456 CDEBUG(D_INODE, "ldlm_cli_cancel: rc = %d\n", rc);
461 case LDLM_CB_CANCELING:
463 __u64 to_cancel = lock->l_policy_data.l_inodebits.bits;
465 /* Nothing to do for non-granted locks */
466 if (!ldlm_is_granted(lock))
469 /* If 'ld' is supplied then bits to be cancelled are passed
470 * implicitly by lock converting and cancel_bits from 'ld'
471 * should be used. Otherwise full cancel is being performed
472 * and lock inodebits are used.
474 * Note: we cannot rely on cancel_bits in lock itself at this
475 * moment because they can be changed by concurrent thread,
476 * so ldlm_cli_inodebits_convert() pass cancel bits implicitly
480 /* partial bits cancel allowed only during convert */
481 LASSERT(ldlm_is_converting(lock));
482 /* mask cancel bits by lock bits so only no any unused
483 * bits are passed to ll_lock_cancel_bits()
485 to_cancel &= ld->l_policy_data.l_inodebits.cancel_bits;
487 ll_lock_cancel_bits(lock, to_cancel);
497 __u32 ll_i2suppgid(struct inode *i)
499 if (in_group_p(i->i_gid))
500 return (__u32)from_kgid(&init_user_ns, i->i_gid);
502 return (__u32) __kgid_val(INVALID_GID);
505 /* Pack the required supplementary groups into the supplied groups array.
506 * If we don't need to use the groups from the target inode(s) then we
507 * instead pack one or more groups from the user's supplementary group
508 * array in case it might be useful. Not needed if doing an MDS-side upcall. */
509 void ll_i2gids(__u32 *suppgids, struct inode *i1, struct inode *i2)
512 LASSERT(suppgids != NULL);
514 suppgids[0] = ll_i2suppgid(i1);
517 suppgids[1] = ll_i2suppgid(i2);
523 * try to reuse three types of dentry:
524 * 1. unhashed alias, this one is unhashed by d_invalidate (but it may be valid
525 * by concurrent .revalidate).
526 * 2. INVALID alias (common case for no valid ldlm lock held, but this flag may
527 * be cleared by others calling d_lustre_revalidate).
528 * 3. DISCONNECTED alias.
530 static struct dentry *ll_find_alias(struct inode *inode, struct dentry *dentry)
532 struct dentry *alias, *discon_alias, *invalid_alias;
534 if (hlist_empty(&inode->i_dentry))
537 discon_alias = invalid_alias = NULL;
539 spin_lock(&inode->i_lock);
540 hlist_for_each_entry(alias, &inode->i_dentry, d_alias) {
541 LASSERT(alias != dentry);
543 spin_lock(&alias->d_lock);
544 if ((alias->d_flags & DCACHE_DISCONNECTED) &&
545 S_ISDIR(inode->i_mode))
546 /* LASSERT(last_discon == NULL); LU-405, bz 20055 */
547 discon_alias = alias;
548 else if (alias->d_parent == dentry->d_parent &&
549 alias->d_name.hash == dentry->d_name.hash &&
550 alias->d_name.len == dentry->d_name.len &&
551 memcmp(alias->d_name.name, dentry->d_name.name,
552 dentry->d_name.len) == 0)
553 invalid_alias = alias;
554 spin_unlock(&alias->d_lock);
559 alias = invalid_alias ?: discon_alias ?: NULL;
561 spin_lock(&alias->d_lock);
563 spin_unlock(&alias->d_lock);
565 spin_unlock(&inode->i_lock);
571 * Similar to d_splice_alias(), but lustre treats invalid alias
572 * similar to DCACHE_DISCONNECTED, and tries to use it anyway.
574 struct dentry *ll_splice_alias(struct inode *inode, struct dentry *de)
580 new = ll_find_alias(inode, de);
590 "Reuse dentry %p inode %p refc %d flags %#x\n",
591 new, new->d_inode, ll_d_count(new), new->d_flags);
599 CDEBUG(D_DENTRY, "Add dentry %p inode %p refc %d flags %#x\n",
600 de, de->d_inode, ll_d_count(de), de->d_flags);
604 static int ll_lookup_it_finish(struct ptlrpc_request *request,
605 struct lookup_intent *it,
606 struct inode *parent, struct dentry **de,
607 void *secctx, __u32 secctxlen,
608 void *encctx, __u32 encctxlen,
609 ktime_t kstart, bool encrypt)
611 struct inode *inode = NULL;
614 struct dentry *alias;
617 /* NB 1 request reference will be taken away by ll_intent_lock()
619 CDEBUG(D_DENTRY, "it %p it_disposition %x\n", it,
621 if (!it_disposition(it, DISP_LOOKUP_NEG)) {
622 struct req_capsule *pill = &request->rq_pill;
623 struct mdt_body *body = req_capsule_server_get(pill,
626 rc = ll_prep_inode(&inode, request, (*de)->d_sb, it);
630 /* If encryption context was returned by MDT, put it in
631 * inode now to save an extra getxattr and avoid deadlock.
633 if (body->mbo_valid & OBD_MD_ENCCTX) {
634 encctx = req_capsule_server_get(pill, &RMF_FILE_ENCCTX);
635 encctxlen = req_capsule_get_size(pill,
641 "server returned encryption ctx for "DFID"\n",
642 PFID(ll_inode2fid(inode)));
643 rc = ll_xattr_cache_insert(inode,
644 LL_XATTR_NAME_ENCRYPTION_CONTEXT,
647 CWARN("%s: cannot set enc ctx for "DFID": rc = %d\n",
648 ll_i2sbi(inode)->ll_fsname,
649 PFID(ll_inode2fid(inode)), rc);
651 rc = llcrypt_get_encryption_info(inode);
654 "cannot get enc info for "DFID": rc = %d\n",
655 PFID(ll_inode2fid(inode)), rc);
660 ll_set_lock_data(ll_i2sbi(parent)->ll_md_exp, inode, it, &bits);
661 /* OPEN can return data if lock has DoM+LAYOUT bits set */
662 if (it->it_op & IT_OPEN &&
663 bits & MDS_INODELOCK_DOM && bits & MDS_INODELOCK_LAYOUT)
664 ll_dom_finish_open(inode, request);
666 /* We used to query real size from OSTs here, but actually
667 * this is not needed. For stat() calls size would be updated
668 * from subsequent do_revalidate()->ll_inode_revalidate_it() in
670 * vfs_getattr_it->ll_getattr()->ll_inode_revalidate_it() in 2.6
671 * Everybody else who needs correct file size would call
672 * ll_glimpse_size or some equivalent themselves anyway.
676 /* If security context was returned by MDT, put it in
677 * inode now to save an extra getxattr from security hooks,
678 * and avoid deadlock.
680 if (body->mbo_valid & OBD_MD_SECCTX) {
681 secctx = req_capsule_server_get(pill, &RMF_FILE_SECCTX);
682 secctxlen = req_capsule_get_size(pill,
687 CDEBUG(D_SEC, "server returned security context"
689 PFID(ll_inode2fid(inode)));
692 if (secctx != NULL && secctxlen != 0) {
693 /* no need to protect selinux_inode_setsecurity() by
694 * inode_lock. Taking it would lead to a client deadlock
697 rc = security_inode_notifysecctx(inode, secctx,
700 CWARN("%s: cannot set security context for "DFID": rc = %d\n",
701 ll_i2sbi(inode)->ll_fsname,
702 PFID(ll_inode2fid(inode)),
707 /* Only hash *de if it is unhashed (new dentry).
708 * Atoimc_open may passin hashed dentries for open.
710 alias = ll_splice_alias(inode, *de);
712 GOTO(out, rc = PTR_ERR(alias));
716 if (!it_disposition(it, DISP_LOOKUP_NEG)) {
717 /* we have lookup look - unhide dentry */
718 if (bits & MDS_INODELOCK_LOOKUP)
719 d_lustre_revalidate(*de);
722 rc = llcrypt_get_encryption_info(inode);
725 if (!llcrypt_has_encryption_key(inode))
726 GOTO(out, rc = -ENOKEY);
728 } else if (!it_disposition(it, DISP_OPEN_CREATE)) {
730 * If file was created on the server, the dentry is revalidated
731 * in ll_create_it if the lock allows for it.
733 /* Check that parent has UPDATE lock. */
734 struct lookup_intent parent_it = {
736 .it_lock_handle = 0 };
737 struct lu_fid fid = ll_i2info(parent)->lli_fid;
739 /* If it is striped directory, get the real stripe parent */
740 if (unlikely(ll_dir_striped(parent))) {
741 rc = md_get_fid_from_lsm(ll_i2mdexp(parent),
742 ll_i2info(parent)->lli_lsm_md,
744 (*de)->d_name.len, &fid);
749 if (md_revalidate_lock(ll_i2mdexp(parent), &parent_it, &fid,
751 d_lustre_revalidate(*de);
752 ll_intent_release(&parent_it);
756 if (it_disposition(it, DISP_OPEN_CREATE)) {
757 ll_stats_ops_tally(ll_i2sbi(parent), LPROC_LL_MKNOD,
758 ktime_us_delta(ktime_get(), kstart));
764 if (rc != 0 && it->it_op & IT_OPEN) {
765 ll_intent_drop_lock(it);
766 ll_open_cleanup((*de)->d_sb, request);
772 static struct dentry *ll_lookup_it(struct inode *parent, struct dentry *dentry,
773 struct lookup_intent *it,
774 void **secctx, __u32 *secctxlen,
775 struct pcc_create_attach *pca,
777 void **encctx, __u32 *encctxlen)
779 ktime_t kstart = ktime_get();
780 struct lookup_intent lookup_it = { .it_op = IT_LOOKUP };
781 struct dentry *save = dentry, *retval;
782 struct ptlrpc_request *req = NULL;
783 struct md_op_data *op_data = NULL;
784 struct lov_user_md *lum = NULL;
787 char secctx_name[XATTR_NAME_MAX + 1];
791 if (dentry->d_name.len > ll_i2sbi(parent)->ll_namelen)
792 RETURN(ERR_PTR(-ENAMETOOLONG));
794 CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p), intent=%s\n",
795 dentry, PFID(ll_inode2fid(parent)), parent, LL_IT2STR(it));
797 if (d_mountpoint(dentry))
798 CERROR("Tell Peter, lookup on mtpt, it %s\n", LL_IT2STR(it));
800 if (it == NULL || it->it_op == IT_GETXATTR)
803 if (it->it_op == IT_GETATTR && dentry_may_statahead(parent, dentry)) {
804 rc = ll_revalidate_statahead(parent, &dentry, 0);
806 RETURN(dentry == save ? NULL : dentry);
809 if (it->it_op & IT_OPEN && it->it_flags & FMODE_WRITE &&
810 dentry->d_sb->s_flags & SB_RDONLY)
811 RETURN(ERR_PTR(-EROFS));
813 if (it->it_op & IT_CREAT)
814 opc = LUSTRE_OPC_CREATE;
816 opc = LUSTRE_OPC_ANY;
818 op_data = ll_prep_md_op_data(NULL, parent, NULL, dentry->d_name.name,
819 dentry->d_name.len, 0, opc, NULL);
821 GOTO(out, retval = ERR_CAST(op_data));
823 /* enforce umask if acl disabled or MDS doesn't support umask */
824 if (!IS_POSIXACL(parent) || !exp_connect_umask(ll_i2mdexp(parent)))
825 it->it_create_mode &= ~current_umask();
827 if (it->it_op & IT_CREAT &&
828 ll_i2sbi(parent)->ll_flags & LL_SBI_FILE_SECCTX) {
829 rc = ll_dentry_init_security(dentry, it->it_create_mode,
831 &op_data->op_file_secctx_name,
832 &op_data->op_file_secctx,
833 &op_data->op_file_secctx_size);
835 GOTO(out, retval = ERR_PTR(rc));
837 *secctx = op_data->op_file_secctx;
838 if (secctxlen != NULL)
839 *secctxlen = op_data->op_file_secctx_size;
843 if (secctxlen != NULL)
846 if (it->it_op & IT_CREAT && encrypt) {
847 rc = llcrypt_inherit_context(parent, NULL, op_data, false);
849 GOTO(out, retval = ERR_PTR(rc));
851 *encctx = op_data->op_file_encctx;
852 if (encctxlen != NULL)
853 *encctxlen = op_data->op_file_encctx_size;
857 if (encctxlen != NULL)
861 /* ask for security context upon intent */
862 if (it->it_op & (IT_LOOKUP | IT_GETATTR | IT_OPEN)) {
863 /* get name of security xattr to request to server */
864 rc = ll_listsecurity(parent, secctx_name,
865 sizeof(secctx_name));
867 CDEBUG(D_SEC, "cannot get security xattr name for "
869 PFID(ll_inode2fid(parent)), rc);
871 op_data->op_file_secctx_name = secctx_name;
872 op_data->op_file_secctx_name_size = rc;
873 CDEBUG(D_SEC, "'%.*s' is security xattr for "DFID"\n",
874 rc, secctx_name, PFID(ll_inode2fid(parent)));
878 if (pca && pca->pca_dataset) {
879 struct pcc_dataset *dataset = pca->pca_dataset;
883 GOTO(out, retval = ERR_PTR(-ENOMEM));
885 lum->lmm_magic = LOV_USER_MAGIC_V1;
886 lum->lmm_pattern = LOV_PATTERN_F_RELEASED | LOV_PATTERN_RAID0;
887 op_data->op_data = lum;
888 op_data->op_data_size = sizeof(*lum);
889 op_data->op_archive_id = dataset->pccd_rwid;
891 rc = obd_fid_alloc(NULL, ll_i2mdexp(parent), &op_data->op_fid2,
894 GOTO(out, retval = ERR_PTR(rc));
896 rc = pcc_inode_create(parent->i_sb, dataset, &op_data->op_fid2,
899 GOTO(out, retval = ERR_PTR(rc));
901 it->it_flags |= MDS_OPEN_PCC;
904 rc = md_intent_lock(ll_i2mdexp(parent), op_data, it, &req,
905 &ll_md_blocking_ast, 0);
906 /* If the MDS allows the client to chgrp (CFS_SETGRP_PERM), but the
907 * client does not know which suppgid should be sent to the MDS, or
908 * some other(s) changed the target file's GID after this RPC sent
909 * to the MDS with the suppgid as the original GID, then we should
910 * try again with right suppgid. */
911 if (rc == -EACCES && it->it_op & IT_OPEN &&
912 it_disposition(it, DISP_OPEN_DENY)) {
913 struct mdt_body *body;
915 LASSERT(req != NULL);
917 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
918 if (op_data->op_suppgids[0] == body->mbo_gid ||
919 op_data->op_suppgids[1] == body->mbo_gid ||
920 !in_group_p(make_kgid(&init_user_ns, body->mbo_gid)))
921 GOTO(out, retval = ERR_PTR(-EACCES));
923 fid_zero(&op_data->op_fid2);
924 op_data->op_suppgids[1] = body->mbo_gid;
925 ptlrpc_req_finished(req);
927 ll_intent_release(it);
928 rc = md_intent_lock(ll_i2mdexp(parent), op_data, it, &req,
929 &ll_md_blocking_ast, 0);
933 GOTO(out, retval = ERR_PTR(rc));
935 /* dir layout may change */
936 ll_unlock_md_op_lsm(op_data);
937 rc = ll_lookup_it_finish(req, it, parent, &dentry,
938 secctx != NULL ? *secctx : NULL,
939 secctxlen != NULL ? *secctxlen : 0,
940 encctx != NULL ? *encctx : NULL,
941 encctxlen != NULL ? *encctxlen : 0,
944 ll_intent_release(it);
945 GOTO(out, retval = ERR_PTR(rc));
948 if ((it->it_op & IT_OPEN) && dentry->d_inode &&
949 !S_ISREG(dentry->d_inode->i_mode) &&
950 !S_ISDIR(dentry->d_inode->i_mode)) {
951 ll_release_openhandle(dentry, it);
953 ll_lookup_finish_locks(it, dentry);
955 GOTO(out, retval = (dentry == save) ? NULL : dentry);
958 if (op_data != NULL && !IS_ERR(op_data)) {
959 if (secctx != NULL && secctxlen != NULL) {
960 /* caller needs sec ctx info, so reset it in op_data to
961 * prevent it from being freed */
962 op_data->op_file_secctx = NULL;
963 op_data->op_file_secctx_size = 0;
965 if (encctx != NULL && encctxlen != NULL &&
966 it->it_op & IT_CREAT && encrypt) {
967 /* caller needs enc ctx info, so reset it in op_data to
968 * prevent it from being freed
970 op_data->op_file_encctx = NULL;
971 op_data->op_file_encctx_size = 0;
973 ll_finish_md_op_data(op_data);
979 ptlrpc_req_finished(req);
983 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
986 struct lookup_intent *itp, it = { .it_op = IT_GETATTR };
989 CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p), flags=%u\n",
990 dentry, PFID(ll_inode2fid(parent)), parent, flags);
993 * Optimize away (CREATE && !OPEN). Let .create handle the race.
994 * but only if we have write permissions there, otherwise we need
995 * to proceed with lookup. LU-4185
997 if ((flags & LOOKUP_CREATE) && !(flags & LOOKUP_OPEN) &&
998 (inode_permission(parent, MAY_WRITE | MAY_EXEC) == 0))
1001 if (flags & (LOOKUP_PARENT|LOOKUP_OPEN|LOOKUP_CREATE))
1005 de = ll_lookup_it(parent, dentry, itp, NULL, NULL, NULL, false,
1009 ll_intent_release(itp);
1014 #ifdef FMODE_CREATED /* added in Linux v4.18-rc1-20-g73a09dd */
1015 # define ll_is_opened(o, f) ((f)->f_mode & FMODE_OPENED)
1016 # define ll_finish_open(f, d, o) finish_open((f), (d), NULL)
1017 # define ll_last_arg
1018 # define ll_set_created(o, f) \
1020 (f)->f_mode |= FMODE_CREATED; \
1024 # define ll_is_opened(o, f) (*(o))
1025 # define ll_finish_open(f, d, o) finish_open((f), (d), NULL, (o))
1026 # define ll_last_arg , int *opened
1027 # define ll_set_created(o, f) \
1029 *(o) |= FILE_CREATED; \
1035 * For cached negative dentry and new dentry, handle lookup/create/open
1038 static int ll_atomic_open(struct inode *dir, struct dentry *dentry,
1039 struct file *file, unsigned open_flags,
1040 umode_t mode ll_last_arg)
1042 struct lookup_intent *it;
1044 long long lookup_flags = LOOKUP_OPEN;
1045 void *secctx = NULL;
1046 __u32 secctxlen = 0;
1047 void *encctx = NULL;
1048 __u32 encctxlen = 0;
1049 struct ll_sb_info *sbi = NULL;
1050 struct pcc_create_attach pca = { NULL, NULL };
1051 bool encrypt = false;
1056 "VFS Op:name=%pd, dir="DFID"(%p), file %p, open_flags %x, mode %x opened %d\n",
1057 dentry, PFID(ll_inode2fid(dir)), dir, file, open_flags, mode,
1058 ll_is_opened(opened, file));
1060 /* Only negative dentries enter here */
1061 LASSERT(dentry->d_inode == NULL);
1063 if (!d_unhashed(dentry)) {
1064 /* A valid negative dentry that just passed revalidation,
1065 * there's little point to try and open it server-side,
1066 * even though there's a minuscule chance it might succeed.
1067 * Either way it's a valid race to just return -ENOENT here.
1069 if (!(open_flags & O_CREAT))
1072 /* Otherwise we just unhash it to be rehashed afresh via
1073 * lookup if necessary
1078 OBD_ALLOC(it, sizeof(*it));
1082 it->it_op = IT_OPEN;
1083 if (open_flags & O_CREAT) {
1084 it->it_op |= IT_CREAT;
1085 lookup_flags |= LOOKUP_CREATE;
1086 sbi = ll_i2sbi(dir);
1087 /* Volatile file is used for HSM restore, so do not use PCC */
1088 if (!filename_is_volatile(dentry->d_name.name,
1089 dentry->d_name.len, NULL)) {
1090 struct pcc_matcher item;
1091 struct pcc_dataset *dataset;
1093 item.pm_uid = from_kuid(&init_user_ns, current_uid());
1094 item.pm_gid = from_kgid(&init_user_ns, current_gid());
1095 item.pm_projid = ll_i2info(dir)->lli_projid;
1096 item.pm_name = &dentry->d_name;
1097 dataset = pcc_dataset_match_get(&sbi->ll_pcc_super,
1099 pca.pca_dataset = dataset;
1102 it->it_create_mode = (mode & S_IALLUGO) | S_IFREG;
1103 it->it_flags = (open_flags & ~O_ACCMODE) | OPEN_FMODE(open_flags);
1104 it->it_flags &= ~MDS_OPEN_FL_INTERNAL;
1106 if (ll_sbi_has_encrypt(ll_i2sbi(dir)) && IS_ENCRYPTED(dir)) {
1107 /* we know that we are going to create a regular file because
1108 * we set S_IFREG bit on it->it_create_mode above
1110 rc = llcrypt_get_encryption_info(dir);
1112 GOTO(out_release, rc);
1113 if (!llcrypt_has_encryption_key(dir))
1114 GOTO(out_release, rc = -ENOKEY);
1119 /* Dentry added to dcache tree in ll_lookup_it */
1120 de = ll_lookup_it(dir, dentry, it, &secctx, &secctxlen, &pca, encrypt,
1121 &encctx, &encctxlen);
1124 else if (de != NULL)
1127 CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_FILE_PAUSE, cfs_fail_val);
1130 if (it_disposition(it, DISP_OPEN_CREATE)) {
1131 /* Dentry instantiated in ll_create_it. */
1132 rc = ll_create_it(dir, dentry, it, secctx, secctxlen,
1133 encrypt, encctx, encctxlen);
1134 security_release_secctx(secctx, secctxlen);
1135 llcrypt_free_ctx(encctx, encctxlen);
1137 /* We dget in ll_splice_alias. */
1143 rc = pcc_inode_create_fini(dentry->d_inode, &pca);
1147 GOTO(out_release, rc);
1150 ll_set_created(opened, file);
1152 /* Open the file with O_CREAT, but the file already
1153 * existed on MDT. This may happend in the case that
1154 * the LOOKUP ibits lock is revoked and the
1155 * corresponding dentry cache is deleted.
1156 * i.e. In the current Lustre, the truncate operation
1157 * will revoke the LOOKUP ibits lock, and the file
1158 * dentry cache will be invalidated. The following open
1159 * with O_CREAT flag will call into ->atomic_open, the
1160 * file was wrongly though as newly created file and
1161 * try to auto cache the file. So after client knows it
1162 * is not a DISP_OPEN_CREATE, it should cleanup the
1163 * already created PCC copy.
1165 pcc_create_attach_cleanup(dir->i_sb, &pca);
1167 if (open_flags & O_CREAT && encrypt &&
1169 rc = ll_set_encflags(dentry->d_inode, encctx,
1171 llcrypt_free_ctx(encctx, encctxlen);
1173 GOTO(out_release, rc);
1177 if (dentry->d_inode && it_disposition(it, DISP_OPEN_OPEN)) {
1179 if (S_ISFIFO(dentry->d_inode->i_mode)) {
1180 /* We cannot call open here as it might
1181 * deadlock. This case is unreachable in
1182 * practice because of OBD_CONNECT_NODEVOH. */
1183 rc = finish_no_open(file, de);
1185 file->private_data = it;
1186 rc = ll_finish_open(file, dentry, opened);
1187 /* We dget in ll_splice_alias. finish_open takes
1188 * care of dget for fd open.
1194 rc = finish_no_open(file, de);
1197 pcc_create_attach_cleanup(dir->i_sb, &pca);
1201 ll_intent_release(it);
1202 OBD_FREE(it, sizeof(*it));
1207 /* We depend on "mode" being set with the proper file type/umask by now */
1208 static struct inode *ll_create_node(struct inode *dir, struct lookup_intent *it)
1210 struct inode *inode = NULL;
1211 struct ptlrpc_request *request = NULL;
1212 struct ll_sb_info *sbi = ll_i2sbi(dir);
1216 LASSERT(it && it->it_disposition);
1218 LASSERT(it_disposition(it, DISP_ENQ_CREATE_REF));
1219 request = it->it_request;
1220 it_clear_disposition(it, DISP_ENQ_CREATE_REF);
1221 rc = ll_prep_inode(&inode, request, dir->i_sb, it);
1223 GOTO(out, inode = ERR_PTR(rc));
1225 /* Pause to allow for a race with concurrent access by fid */
1226 OBD_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_NODE_PAUSE, cfs_fail_val);
1228 /* We asked for a lock on the directory, but were granted a
1229 * lock on the inode. Since we finally have an inode pointer,
1230 * stuff it in the lock. */
1231 CDEBUG(D_DLMTRACE, "setting l_ast_data to inode "DFID"(%p)\n",
1232 PFID(ll_inode2fid(inode)), inode);
1233 ll_set_lock_data(sbi->ll_md_exp, inode, it, NULL);
1236 ptlrpc_req_finished(request);
1241 * By the time this is called, we already have created the directory cache
1242 * entry for the new file, but it is so far negative - it has no inode.
1244 * We defer creating the OBD object(s) until open, to keep the intent and
1245 * non-intent code paths similar, and also because we do not have the MDS
1246 * inode number before calling ll_create_node() (which is needed for LOV),
1247 * so we would need to do yet another RPC to the MDS to store the LOV EA
1248 * data on the MDS. If needed, we would pass the PACKED lmm as data and
1249 * lmm_size in datalen (the MDS still has code which will handle that).
1251 * If the create succeeds, we fill in the inode information
1252 * with d_instantiate().
1254 static int ll_create_it(struct inode *dir, struct dentry *dentry,
1255 struct lookup_intent *it,
1256 void *secctx, __u32 secctxlen, bool encrypt,
1257 void *encctx, __u32 encctxlen)
1259 struct inode *inode;
1264 CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p), intent=%s\n",
1265 dentry, PFID(ll_inode2fid(dir)), dir, LL_IT2STR(it));
1267 rc = it_open_error(DISP_OPEN_CREATE, it);
1271 inode = ll_create_node(dir, it);
1273 RETURN(PTR_ERR(inode));
1275 if ((ll_i2sbi(inode)->ll_flags & LL_SBI_FILE_SECCTX) &&
1277 /* must be done before d_instantiate, because it calls
1278 * security_d_instantiate, which means a getxattr if security
1279 * context is not set yet */
1280 /* no need to protect selinux_inode_setsecurity() by
1281 * inode_lock. Taking it would lead to a client deadlock
1284 rc = security_inode_notifysecctx(inode, secctx, secctxlen);
1289 d_instantiate(dentry, inode);
1292 rc = ll_set_encflags(inode, encctx, encctxlen, true);
1297 if (!(ll_i2sbi(inode)->ll_flags & LL_SBI_FILE_SECCTX)) {
1298 rc = ll_inode_init_security(dentry, inode, dir);
1303 ll_set_lock_data(ll_i2sbi(dir)->ll_md_exp, inode, it, &bits);
1304 if (bits & MDS_INODELOCK_LOOKUP)
1305 d_lustre_revalidate(dentry);
1310 void ll_update_times(struct ptlrpc_request *request, struct inode *inode)
1312 struct mdt_body *body = req_capsule_server_get(&request->rq_pill,
1316 if (body->mbo_valid & OBD_MD_FLMTIME &&
1317 body->mbo_mtime > inode->i_mtime.tv_sec) {
1319 "setting fid " DFID " mtime from %lld to %llu\n",
1320 PFID(ll_inode2fid(inode)),
1321 (s64)inode->i_mtime.tv_sec, body->mbo_mtime);
1322 inode->i_mtime.tv_sec = body->mbo_mtime;
1325 if (body->mbo_valid & OBD_MD_FLCTIME &&
1326 body->mbo_ctime > inode->i_ctime.tv_sec)
1327 inode->i_ctime.tv_sec = body->mbo_ctime;
1330 static int ll_new_node(struct inode *dir, struct dentry *dchild,
1331 const char *tgt, umode_t mode, int rdev, __u32 opc)
1333 struct qstr *name = &dchild->d_name;
1334 struct ptlrpc_request *request = NULL;
1335 struct md_op_data *op_data = NULL;
1336 struct inode *inode = NULL;
1337 struct ll_sb_info *sbi = ll_i2sbi(dir);
1339 bool encrypt = false;
1343 if (unlikely(tgt != NULL))
1344 tgt_len = strlen(tgt) + 1;
1347 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
1348 name->len, 0, opc, NULL);
1349 if (IS_ERR(op_data))
1350 GOTO(err_exit, err = PTR_ERR(op_data));
1352 if (sbi->ll_flags & LL_SBI_FILE_SECCTX) {
1353 err = ll_dentry_init_security(dchild, mode, &dchild->d_name,
1354 &op_data->op_file_secctx_name,
1355 &op_data->op_file_secctx,
1356 &op_data->op_file_secctx_size);
1358 GOTO(err_exit, err);
1361 if (ll_sbi_has_encrypt(sbi) &&
1362 ((IS_ENCRYPTED(dir) &&
1363 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) ||
1364 (unlikely(llcrypt_dummy_context_enabled(dir)) && S_ISDIR(mode)))) {
1365 err = llcrypt_get_encryption_info(dir);
1367 GOTO(err_exit, err);
1368 if (!llcrypt_has_encryption_key(dir))
1369 GOTO(err_exit, err = -ENOKEY);
1374 err = llcrypt_inherit_context(dir, NULL, op_data, false);
1376 GOTO(err_exit, err);
1379 err = md_create(sbi->ll_md_exp, op_data, tgt, tgt_len, mode,
1380 from_kuid(&init_user_ns, current_fsuid()),
1381 from_kgid(&init_user_ns, current_fsgid()),
1382 cfs_curproc_cap_pack(), rdev, &request);
1383 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 14, 58, 0)
1385 * server < 2.12.58 doesn't pack default LMV in intent_getattr reply,
1386 * fetch default LMV here.
1388 if (unlikely(err == -EREMOTE)) {
1389 struct ll_inode_info *lli = ll_i2info(dir);
1390 struct lmv_user_md *lum;
1394 ptlrpc_req_finished(request);
1397 err2 = ll_dir_getstripe(dir, (void **)&lum, &lumsize, &request,
1398 OBD_MD_DEFAULT_MEA);
1399 ll_finish_md_op_data(op_data);
1402 struct lustre_md md = { NULL };
1404 md.body = req_capsule_server_get(&request->rq_pill,
1407 GOTO(err_exit, err = -EPROTO);
1409 OBD_ALLOC_PTR(md.default_lmv);
1410 if (!md.default_lmv)
1411 GOTO(err_exit, err = -ENOMEM);
1413 md.default_lmv->lsm_md_magic = lum->lum_magic;
1414 md.default_lmv->lsm_md_stripe_count =
1415 lum->lum_stripe_count;
1416 md.default_lmv->lsm_md_master_mdt_index =
1417 lum->lum_stripe_offset;
1418 md.default_lmv->lsm_md_hash_type = lum->lum_hash_type;
1420 err = ll_update_inode(dir, &md);
1421 md_free_lustre_md(sbi->ll_md_exp, &md);
1423 GOTO(err_exit, err);
1424 } else if (err2 == -ENODATA && lli->lli_default_lsm_md) {
1426 * If there are no default stripe EA on the MDT, but the
1427 * client has default stripe, then it probably means
1428 * default stripe EA has just been deleted.
1430 down_write(&lli->lli_lsm_sem);
1431 if (lli->lli_default_lsm_md)
1432 OBD_FREE_PTR(lli->lli_default_lsm_md);
1433 lli->lli_default_lsm_md = NULL;
1434 up_write(&lli->lli_lsm_sem);
1436 GOTO(err_exit, err);
1439 ptlrpc_req_finished(request);
1446 GOTO(err_exit, err);
1448 ll_update_times(request, dir);
1450 CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_NEWNODE_PAUSE, cfs_fail_val);
1452 err = ll_prep_inode(&inode, request, dchild->d_sb, NULL);
1454 GOTO(err_exit, err);
1456 if (sbi->ll_flags & LL_SBI_FILE_SECCTX) {
1457 /* must be done before d_instantiate, because it calls
1458 * security_d_instantiate, which means a getxattr if security
1459 * context is not set yet */
1460 /* no need to protect selinux_inode_setsecurity() by
1461 * inode_lock. Taking it would lead to a client deadlock
1464 err = security_inode_notifysecctx(inode,
1465 op_data->op_file_secctx,
1466 op_data->op_file_secctx_size);
1468 GOTO(err_exit, err);
1471 d_instantiate(dchild, inode);
1474 err = ll_set_encflags(inode, op_data->op_file_encctx,
1475 op_data->op_file_encctx_size, true);
1477 GOTO(err_exit, err);
1480 if (!(sbi->ll_flags & LL_SBI_FILE_SECCTX)) {
1481 err = ll_inode_init_security(dchild, inode, dir);
1483 GOTO(err_exit, err);
1488 if (request != NULL)
1489 ptlrpc_req_finished(request);
1491 if (!IS_ERR_OR_NULL(op_data))
1492 ll_finish_md_op_data(op_data);
1497 static int ll_mknod(struct inode *dir, struct dentry *dchild, umode_t mode,
1500 ktime_t kstart = ktime_get();
1504 CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p) mode %o dev %x\n",
1505 dchild, PFID(ll_inode2fid(dir)), dir, mode, rdev);
1507 if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
1508 mode &= ~current_umask();
1510 switch (mode & S_IFMT) {
1519 err = ll_new_node(dir, dchild, NULL, mode, old_encode_dev(rdev),
1530 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKNOD,
1531 ktime_us_delta(ktime_get(), kstart));
1537 * Plain create. Intent create is handled in atomic_open.
1539 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
1540 umode_t mode, bool want_excl)
1542 ktime_t kstart = ktime_get();
1545 CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_FILE_PAUSE, cfs_fail_val);
1548 "VFS Op:name=%pd, dir="DFID"(%p), flags=%u, excl=%d\n",
1549 dentry, PFID(ll_inode2fid(dir)), dir, mode, want_excl);
1551 /* Using mknod(2) to create a regular file is designed to not recognize
1552 * volatile file name, so we use ll_mknod() here. */
1553 rc = ll_mknod(dir, dentry, mode, 0);
1555 CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, unhashed %d\n",
1556 dentry, d_unhashed(dentry));
1559 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_CREATE,
1560 ktime_us_delta(ktime_get(), kstart));
1565 static int ll_symlink(struct inode *dir, struct dentry *dchild,
1566 const char *oldpath)
1568 ktime_t kstart = ktime_get();
1572 CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p), target=%.*s\n",
1573 dchild, PFID(ll_inode2fid(dir)), dir, 3000, oldpath);
1575 err = ll_new_node(dir, dchild, oldpath, S_IFLNK | S_IRWXUGO, 0,
1576 LUSTRE_OPC_SYMLINK);
1579 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_SYMLINK,
1580 ktime_us_delta(ktime_get(), kstart));
1585 static int ll_link(struct dentry *old_dentry, struct inode *dir,
1586 struct dentry *new_dentry)
1588 struct inode *src = old_dentry->d_inode;
1589 struct qstr *name = &new_dentry->d_name;
1590 struct ll_sb_info *sbi = ll_i2sbi(dir);
1591 struct ptlrpc_request *request = NULL;
1592 struct md_op_data *op_data;
1593 ktime_t kstart = ktime_get();
1598 "VFS Op: inode="DFID"(%p), dir="DFID"(%p), target=%pd\n",
1599 PFID(ll_inode2fid(src)), src,
1600 PFID(ll_inode2fid(dir)), dir, new_dentry);
1602 err = llcrypt_prepare_link(old_dentry, dir, new_dentry);
1606 op_data = ll_prep_md_op_data(NULL, src, dir, name->name, name->len,
1607 0, LUSTRE_OPC_ANY, NULL);
1608 if (IS_ERR(op_data))
1609 RETURN(PTR_ERR(op_data));
1611 err = md_link(sbi->ll_md_exp, op_data, &request);
1612 ll_finish_md_op_data(op_data);
1616 ll_update_times(request, dir);
1617 ll_stats_ops_tally(sbi, LPROC_LL_LINK,
1618 ktime_us_delta(ktime_get(), kstart));
1621 ptlrpc_req_finished(request);
1625 static int ll_mkdir(struct inode *dir, struct dentry *dchild, umode_t mode)
1627 ktime_t kstart = ktime_get();
1631 CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p)\n",
1632 dchild, PFID(ll_inode2fid(dir)), dir);
1634 if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
1635 mode &= ~current_umask();
1637 mode = (mode & (S_IRWXUGO|S_ISVTX)) | S_IFDIR;
1639 err = ll_new_node(dir, dchild, NULL, mode, 0, LUSTRE_OPC_MKDIR);
1641 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKDIR,
1642 ktime_us_delta(ktime_get(), kstart));
1647 static int ll_rmdir(struct inode *dir, struct dentry *dchild)
1649 struct qstr *name = &dchild->d_name;
1650 struct ptlrpc_request *request = NULL;
1651 struct md_op_data *op_data;
1652 ktime_t kstart = ktime_get();
1657 CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p)\n",
1658 dchild, PFID(ll_inode2fid(dir)), dir);
1660 if (unlikely(d_mountpoint(dchild)))
1663 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len,
1664 S_IFDIR, LUSTRE_OPC_ANY, NULL);
1665 if (IS_ERR(op_data))
1666 RETURN(PTR_ERR(op_data));
1668 if (dchild->d_inode != NULL)
1669 op_data->op_fid3 = *ll_inode2fid(dchild->d_inode);
1671 op_data->op_fid2 = op_data->op_fid3;
1672 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1673 ll_finish_md_op_data(op_data);
1675 ll_update_times(request, dir);
1677 ptlrpc_req_finished(request);
1679 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR,
1680 ktime_us_delta(ktime_get(), kstart));
1687 int ll_rmdir_entry(struct inode *dir, char *name, int namelen)
1689 struct ptlrpc_request *request = NULL;
1690 struct md_op_data *op_data;
1691 ktime_t kstart = ktime_get();
1695 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1696 namelen, name, PFID(ll_inode2fid(dir)), dir);
1698 op_data = ll_prep_md_op_data(NULL, dir, NULL, name, strlen(name),
1699 S_IFDIR, LUSTRE_OPC_ANY, NULL);
1700 if (IS_ERR(op_data))
1701 RETURN(PTR_ERR(op_data));
1702 op_data->op_cli_flags |= CLI_RM_ENTRY;
1703 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1704 ll_finish_md_op_data(op_data);
1706 ll_update_times(request, dir);
1708 ptlrpc_req_finished(request);
1710 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR,
1711 ktime_us_delta(ktime_get(), kstart));
1715 static int ll_unlink(struct inode *dir, struct dentry *dchild)
1717 struct qstr *name = &dchild->d_name;
1718 struct ptlrpc_request *request = NULL;
1719 struct md_op_data *op_data;
1720 struct mdt_body *body;
1721 ktime_t kstart = ktime_get();
1726 CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p)\n",
1727 dchild, PFID(ll_inode2fid(dir)), dir);
1730 * XXX: unlink bind mountpoint maybe call to here,
1731 * just check it as vfs_unlink does.
1733 if (unlikely(d_mountpoint(dchild)))
1736 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len, 0,
1737 LUSTRE_OPC_ANY, NULL);
1738 if (IS_ERR(op_data))
1739 RETURN(PTR_ERR(op_data));
1741 op_data->op_fid3 = *ll_inode2fid(dchild->d_inode);
1742 /* notify lower layer if inode has dirty pages */
1743 if (S_ISREG(dchild->d_inode->i_mode) &&
1744 ll_i2info(dchild->d_inode)->lli_clob &&
1745 dirty_cnt(dchild->d_inode))
1746 op_data->op_cli_flags |= CLI_DIRTY_DATA;
1747 op_data->op_fid2 = op_data->op_fid3;
1748 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1749 ll_finish_md_op_data(op_data);
1754 * The server puts attributes in on the last unlink, use them to update
1755 * the link count so the inode can be freed immediately.
1757 body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
1758 if (body->mbo_valid & OBD_MD_FLNLINK)
1759 set_nlink(dchild->d_inode, body->mbo_nlink);
1761 ll_update_times(request, dir);
1764 ptlrpc_req_finished(request);
1766 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_UNLINK,
1767 ktime_us_delta(ktime_get(), kstart));
1771 static int ll_rename(struct inode *src, struct dentry *src_dchild,
1772 struct inode *tgt, struct dentry *tgt_dchild
1773 #ifdef HAVE_IOPS_RENAME_WITH_FLAGS
1774 , unsigned int flags
1778 struct qstr *src_name = &src_dchild->d_name;
1779 struct qstr *tgt_name = &tgt_dchild->d_name;
1780 struct ptlrpc_request *request = NULL;
1781 struct ll_sb_info *sbi = ll_i2sbi(src);
1782 struct md_op_data *op_data;
1783 ktime_t kstart = ktime_get();
1787 #ifdef HAVE_IOPS_RENAME_WITH_FLAGS
1793 "VFS Op:oldname=%pd, src_dir="DFID"(%p), newname=%pd, tgt_dir="DFID"(%p)\n",
1794 src_dchild, PFID(ll_inode2fid(src)), src,
1795 tgt_dchild, PFID(ll_inode2fid(tgt)), tgt);
1797 if (unlikely(d_mountpoint(src_dchild) || d_mountpoint(tgt_dchild)))
1800 #ifdef HAVE_IOPS_RENAME_WITH_FLAGS
1801 err = llcrypt_prepare_rename(src, src_dchild, tgt, tgt_dchild, flags);
1803 err = llcrypt_prepare_rename(src, src_dchild, tgt, tgt_dchild, 0);
1808 op_data = ll_prep_md_op_data(NULL, src, tgt, NULL, 0, 0,
1809 LUSTRE_OPC_ANY, NULL);
1810 if (IS_ERR(op_data))
1811 RETURN(PTR_ERR(op_data));
1813 if (src_dchild->d_inode != NULL)
1814 op_data->op_fid3 = *ll_inode2fid(src_dchild->d_inode);
1816 if (tgt_dchild->d_inode != NULL)
1817 op_data->op_fid4 = *ll_inode2fid(tgt_dchild->d_inode);
1819 err = md_rename(sbi->ll_md_exp, op_data,
1820 src_name->name, src_name->len,
1821 tgt_name->name, tgt_name->len, &request);
1822 ll_finish_md_op_data(op_data);
1824 ll_update_times(request, src);
1825 ll_update_times(request, tgt);
1828 ptlrpc_req_finished(request);
1831 d_move(src_dchild, tgt_dchild);
1832 ll_stats_ops_tally(sbi, LPROC_LL_RENAME,
1833 ktime_us_delta(ktime_get(), kstart));
1839 const struct inode_operations ll_dir_inode_operations = {
1841 .atomic_open = ll_atomic_open,
1842 .lookup = ll_lookup_nd,
1843 .create = ll_create_nd,
1844 /* We need all these non-raw things for NFSD, to not patch it. */
1845 .unlink = ll_unlink,
1848 .symlink = ll_symlink,
1850 .rename = ll_rename,
1851 .setattr = ll_setattr,
1852 .getattr = ll_getattr,
1853 .permission = ll_inode_permission,
1854 #ifdef HAVE_IOP_XATTR
1855 .setxattr = ll_setxattr,
1856 .getxattr = ll_getxattr,
1857 .removexattr = ll_removexattr,
1859 .listxattr = ll_listxattr,
1860 .get_acl = ll_get_acl,
1861 #ifdef HAVE_IOP_SET_ACL
1862 .set_acl = ll_set_acl,
1866 const struct inode_operations ll_special_inode_operations = {
1867 .setattr = ll_setattr,
1868 .getattr = ll_getattr,
1869 .permission = ll_inode_permission,
1870 #ifdef HAVE_IOP_XATTR
1871 .setxattr = ll_setxattr,
1872 .getxattr = ll_getxattr,
1873 .removexattr = ll_removexattr,
1875 .listxattr = ll_listxattr,
1876 .get_acl = ll_get_acl,
1877 #ifdef HAVE_IOP_SET_ACL
1878 .set_acl = ll_set_acl,