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 OBD_FAIL_TIMEOUT(OBD_FAIL_LDLM_REPLAY_PAUSE, cfs_fail_val);
217 /* reach MDC layer to flush data under the DoM ldlm lock */
218 rc = cl_object_flush(env, lli->lli_clob, lock);
219 if (rc == -ENODATA) {
220 CDEBUG(D_INODE, "inode "DFID" layout has no DoM stripe\n",
221 PFID(ll_inode2fid(inode)));
222 /* most likely result of layout change, do nothing */
226 cl_env_put(env, &refcheck);
230 static void ll_lock_cancel_bits(struct ldlm_lock *lock, __u64 to_cancel)
232 struct inode *inode = ll_inode_from_resource_lock(lock);
233 struct ll_inode_info *lli;
234 __u64 bits = to_cancel;
240 /* That means the inode is evicted most likely and may cause
241 * the skipping of lock cleanups below, so print the message
244 if (lock->l_resource->lr_lvb_inode)
246 "can't take inode for the lock (%sevicted)\n",
247 lock->l_resource->lr_lvb_inode->i_state &
248 I_FREEING ? "" : "not ");
252 if (!fid_res_name_eq(ll_inode2fid(inode),
253 &lock->l_resource->lr_name)) {
254 LDLM_ERROR(lock, "data mismatch with object "DFID"(%p)",
255 PFID(ll_inode2fid(inode)), inode);
259 if (bits & MDS_INODELOCK_XATTR) {
260 ll_xattr_cache_destroy(inode);
261 bits &= ~MDS_INODELOCK_XATTR;
264 /* For OPEN locks we differentiate between lock modes
265 * LCK_CR, LCK_CW, LCK_PR - bug 22891 */
266 if (bits & MDS_INODELOCK_OPEN)
267 ll_have_md_lock(inode, &bits, lock->l_req_mode);
269 if (bits & MDS_INODELOCK_OPEN) {
272 switch (lock->l_req_mode) {
283 LDLM_ERROR(lock, "bad lock mode for OPEN lock");
287 ll_md_real_close(inode, fmode);
289 bits &= ~MDS_INODELOCK_OPEN;
292 if (bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
293 MDS_INODELOCK_LAYOUT | MDS_INODELOCK_PERM |
295 ll_have_md_lock(inode, &bits, LCK_MINMODE);
297 if (bits & MDS_INODELOCK_DOM) {
298 rc = ll_dom_lock_cancel(inode, lock);
300 CDEBUG(D_INODE, "cannot flush DoM data "
302 PFID(ll_inode2fid(inode)), rc);
305 if (bits & MDS_INODELOCK_LAYOUT) {
306 struct cl_object_conf conf = {
307 .coc_opc = OBJECT_CONF_INVALIDATE,
311 rc = ll_layout_conf(inode, &conf);
313 CDEBUG(D_INODE, "cannot invalidate layout of "
315 PFID(ll_inode2fid(inode)), rc);
318 lli = ll_i2info(inode);
320 if (bits & MDS_INODELOCK_UPDATE)
321 set_bit(LLIF_UPDATE_ATIME, &lli->lli_flags);
323 if ((bits & MDS_INODELOCK_UPDATE) && S_ISDIR(inode->i_mode)) {
324 CDEBUG(D_INODE, "invalidating inode "DFID" lli = %p, "
325 "pfid = "DFID"\n", PFID(ll_inode2fid(inode)),
326 lli, PFID(&lli->lli_pfid));
327 truncate_inode_pages(inode->i_mapping, 0);
329 if (unlikely(!fid_is_zero(&lli->lli_pfid))) {
330 struct inode *master_inode = NULL;
333 /* This is slave inode, since all of the child dentry
334 * is connected on the master inode, so we have to
335 * invalidate the negative children on master inode */
336 CDEBUG(D_INODE, "Invalidate s"DFID" m"DFID"\n",
337 PFID(ll_inode2fid(inode)), PFID(&lli->lli_pfid));
339 hash = cl_fid_build_ino(&lli->lli_pfid,
340 ll_need_32bit_api(ll_i2sbi(inode)));
342 /* Do not lookup the inode with ilookup5, otherwise
343 * it will cause dead lock,
344 * 1. Client1 send chmod req to the MDT0, then on MDT0,
345 * it enqueues master and all of its slaves lock,
346 * (mdt_attr_set() -> mdt_lock_slaves()), after gets
347 * master and stripe0 lock, it will send the enqueue
348 * req (for stripe1) to MDT1, then MDT1 finds the lock
349 * has been granted to client2. Then MDT1 sends blocking
351 * 2. At the same time, client2 tries to unlink
352 * the striped dir (rm -rf striped_dir), and during
353 * lookup, it will hold the master inode of the striped
354 * directory, whose inode state is NEW, then tries to
355 * revalidate all of its slaves, (ll_prep_inode()->
356 * ll_iget()->ll_read_inode2()-> ll_update_inode().).
357 * And it will be blocked on the server side because
359 * 3. Then the client get the blocking_ast req, cancel
360 * the lock, but being blocked if using ->ilookup5()),
361 * because master inode state is NEW. */
362 master_inode = ilookup5_nowait(inode->i_sb, hash,
363 ll_test_inode_by_fid,
364 (void *)&lli->lli_pfid);
366 ll_prune_negative_children(master_inode);
370 ll_prune_negative_children(inode);
374 if ((bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM)) &&
375 !is_root_inode(inode))
376 ll_prune_aliases(inode);
378 if (bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM))
379 forget_all_cached_acls(inode);
385 /* Check if the given lock may be downgraded instead of canceling and
386 * that convert is really needed. */
387 int ll_md_need_convert(struct ldlm_lock *lock)
389 struct ldlm_namespace *ns = ldlm_lock_to_ns(lock);
391 __u64 wanted = lock->l_policy_data.l_inodebits.cancel_bits;
392 __u64 bits = lock->l_policy_data.l_inodebits.bits & ~wanted;
393 enum ldlm_mode mode = LCK_MINMODE;
395 if (!lock->l_conn_export ||
396 !exp_connect_lock_convert(lock->l_conn_export))
399 if (!wanted || !bits || ldlm_is_cancel(lock))
402 /* do not convert locks other than DOM for now */
403 if (!((bits | wanted) & MDS_INODELOCK_DOM))
406 /* We may have already remaining bits in some other lock so
407 * lock convert will leave us just extra lock for the same bit.
408 * Check if client has other lock with the same bits and the same
409 * or lower mode and don't convert if any.
411 switch (lock->l_req_mode) {
425 /* do not convert other modes */
429 /* is lock is too old to be converted? */
430 lock_res_and_lock(lock);
431 if (ktime_after(ktime_get(),
432 ktime_add(lock->l_last_used, ns->ns_dirty_age_limit))) {
433 unlock_res_and_lock(lock);
436 unlock_res_and_lock(lock);
438 inode = ll_inode_from_resource_lock(lock);
439 ll_have_md_lock(inode, &bits, mode);
444 int ll_md_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *ld,
445 void *data, int flag)
447 struct lustre_handle lockh;
453 case LDLM_CB_BLOCKING:
455 __u64 cancel_flags = LCF_ASYNC;
457 /* if lock convert is not needed then still have to
458 * pass lock via ldlm_cli_convert() to keep all states
459 * correct, set cancel_bits to full lock bits to cause
460 * full cancel to happen.
462 if (!ll_md_need_convert(lock)) {
463 lock_res_and_lock(lock);
464 lock->l_policy_data.l_inodebits.cancel_bits =
465 lock->l_policy_data.l_inodebits.bits;
466 unlock_res_and_lock(lock);
468 rc = ldlm_cli_convert(lock, cancel_flags);
471 /* continue with cancel otherwise */
472 ldlm_lock2handle(lock, &lockh);
473 rc = ldlm_cli_cancel(&lockh, cancel_flags);
475 CDEBUG(D_INODE, "ldlm_cli_cancel: rc = %d\n", rc);
480 case LDLM_CB_CANCELING:
482 __u64 to_cancel = lock->l_policy_data.l_inodebits.bits;
484 /* Nothing to do for non-granted locks */
485 if (!ldlm_is_granted(lock))
488 /* If 'ld' is supplied then bits to be cancelled are passed
489 * implicitly by lock converting and cancel_bits from 'ld'
490 * should be used. Otherwise full cancel is being performed
491 * and lock inodebits are used.
493 * Note: we cannot rely on cancel_bits in lock itself at this
494 * moment because they can be changed by concurrent thread,
495 * so ldlm_cli_inodebits_convert() pass cancel bits implicitly
499 /* partial bits cancel allowed only during convert */
500 LASSERT(ldlm_is_converting(lock));
501 /* mask cancel bits by lock bits so only no any unused
502 * bits are passed to ll_lock_cancel_bits()
504 to_cancel &= ld->l_policy_data.l_inodebits.cancel_bits;
506 ll_lock_cancel_bits(lock, to_cancel);
516 __u32 ll_i2suppgid(struct inode *i)
518 if (in_group_p(i->i_gid))
519 return (__u32)from_kgid(&init_user_ns, i->i_gid);
521 return (__u32) __kgid_val(INVALID_GID);
524 /* Pack the required supplementary groups into the supplied groups array.
525 * If we don't need to use the groups from the target inode(s) then we
526 * instead pack one or more groups from the user's supplementary group
527 * array in case it might be useful. Not needed if doing an MDS-side upcall. */
528 void ll_i2gids(__u32 *suppgids, struct inode *i1, struct inode *i2)
531 LASSERT(suppgids != NULL);
533 suppgids[0] = ll_i2suppgid(i1);
536 suppgids[1] = ll_i2suppgid(i2);
542 * try to reuse three types of dentry:
543 * 1. unhashed alias, this one is unhashed by d_invalidate (but it may be valid
544 * by concurrent .revalidate).
545 * 2. INVALID alias (common case for no valid ldlm lock held, but this flag may
546 * be cleared by others calling d_lustre_revalidate).
547 * 3. DISCONNECTED alias.
549 static struct dentry *ll_find_alias(struct inode *inode, struct dentry *dentry)
551 struct dentry *alias, *discon_alias, *invalid_alias;
553 if (hlist_empty(&inode->i_dentry))
556 discon_alias = invalid_alias = NULL;
558 spin_lock(&inode->i_lock);
559 hlist_for_each_entry(alias, &inode->i_dentry, d_alias) {
560 LASSERT(alias != dentry);
562 spin_lock(&alias->d_lock);
563 if ((alias->d_flags & DCACHE_DISCONNECTED) &&
564 S_ISDIR(inode->i_mode))
565 /* LASSERT(last_discon == NULL); LU-405, bz 20055 */
566 discon_alias = alias;
567 else if (alias->d_parent == dentry->d_parent &&
568 alias->d_name.hash == dentry->d_name.hash &&
569 alias->d_name.len == dentry->d_name.len &&
570 memcmp(alias->d_name.name, dentry->d_name.name,
571 dentry->d_name.len) == 0)
572 invalid_alias = alias;
573 spin_unlock(&alias->d_lock);
578 alias = invalid_alias ?: discon_alias ?: NULL;
580 spin_lock(&alias->d_lock);
582 spin_unlock(&alias->d_lock);
584 spin_unlock(&inode->i_lock);
590 * Similar to d_splice_alias(), but lustre treats invalid alias
591 * similar to DCACHE_DISCONNECTED, and tries to use it anyway.
593 struct dentry *ll_splice_alias(struct inode *inode, struct dentry *de)
599 new = ll_find_alias(inode, de);
609 "Reuse dentry %p inode %p refc %d flags %#x\n",
610 new, new->d_inode, ll_d_count(new), new->d_flags);
619 /* this needs only to be done for foreign symlink dirs as
620 * DCACHE_SYMLINK_TYPE is already set by d_flags_for_inode()
621 * kernel routine for files with symlink ops (ie, real symlink)
623 if (inode && S_ISDIR(inode->i_mode) &&
624 ll_sbi_has_foreign_symlink(ll_i2sbi(inode)) &&
625 #ifdef HAVE_IOP_GET_LINK
626 inode->i_op->get_link) {
628 inode->i_op->follow_link) {
630 CDEBUG(D_INFO, "%s: inode "DFID": faking foreign dir as a symlink\n",
631 ll_i2sbi(inode)->ll_fsname, PFID(ll_inode2fid(inode)));
632 spin_lock(&de->d_lock);
633 /* like d_flags_for_inode() already does for files */
634 de->d_flags = (de->d_flags & ~DCACHE_ENTRY_TYPE) |
636 spin_unlock(&de->d_lock);
639 CDEBUG(D_DENTRY, "Add dentry %p inode %p refc %d flags %#x\n",
640 de, de->d_inode, ll_d_count(de), de->d_flags);
644 static int ll_lookup_it_finish(struct ptlrpc_request *request,
645 struct lookup_intent *it,
646 struct inode *parent, struct dentry **de,
647 void *secctx, __u32 secctxlen,
648 void *encctx, __u32 encctxlen,
649 ktime_t kstart, bool encrypt)
651 struct inode *inode = NULL;
654 struct dentry *alias;
657 /* NB 1 request reference will be taken away by ll_intent_lock()
659 CDEBUG(D_DENTRY, "it %p it_disposition %x\n", it,
661 if (!it_disposition(it, DISP_LOOKUP_NEG)) {
662 struct req_capsule *pill = &request->rq_pill;
663 struct mdt_body *body = req_capsule_server_get(pill,
666 rc = ll_prep_inode(&inode, request, (*de)->d_sb, it);
670 /* If encryption context was returned by MDT, put it in
671 * inode now to save an extra getxattr and avoid deadlock.
673 if (body->mbo_valid & OBD_MD_ENCCTX) {
674 encctx = req_capsule_server_get(pill, &RMF_FILE_ENCCTX);
675 encctxlen = req_capsule_get_size(pill,
681 "server returned encryption ctx for "DFID"\n",
682 PFID(ll_inode2fid(inode)));
683 rc = ll_xattr_cache_insert(inode,
684 LL_XATTR_NAME_ENCRYPTION_CONTEXT,
687 CWARN("%s: cannot set enc ctx for "DFID": rc = %d\n",
688 ll_i2sbi(inode)->ll_fsname,
689 PFID(ll_inode2fid(inode)), rc);
691 rc = llcrypt_get_encryption_info(inode);
694 "cannot get enc info for "DFID": rc = %d\n",
695 PFID(ll_inode2fid(inode)), rc);
700 ll_set_lock_data(ll_i2sbi(parent)->ll_md_exp, inode, it, &bits);
701 /* OPEN can return data if lock has DoM+LAYOUT bits set */
702 if (it->it_op & IT_OPEN &&
703 bits & MDS_INODELOCK_DOM && bits & MDS_INODELOCK_LAYOUT)
704 ll_dom_finish_open(inode, request);
706 /* We used to query real size from OSTs here, but actually
707 * this is not needed. For stat() calls size would be updated
708 * from subsequent do_revalidate()->ll_inode_revalidate_it() in
710 * vfs_getattr_it->ll_getattr()->ll_inode_revalidate_it() in 2.6
711 * Everybody else who needs correct file size would call
712 * ll_glimpse_size or some equivalent themselves anyway.
716 /* If security context was returned by MDT, put it in
717 * inode now to save an extra getxattr from security hooks,
718 * and avoid deadlock.
720 if (body->mbo_valid & OBD_MD_SECCTX) {
721 secctx = req_capsule_server_get(pill, &RMF_FILE_SECCTX);
722 secctxlen = req_capsule_get_size(pill,
727 CDEBUG(D_SEC, "server returned security context"
729 PFID(ll_inode2fid(inode)));
732 if (secctx != NULL && secctxlen != 0) {
733 /* no need to protect selinux_inode_setsecurity() by
734 * inode_lock. Taking it would lead to a client deadlock
737 rc = security_inode_notifysecctx(inode, secctx,
740 CWARN("%s: cannot set security context for "DFID": rc = %d\n",
741 ll_i2sbi(inode)->ll_fsname,
742 PFID(ll_inode2fid(inode)),
747 /* Only hash *de if it is unhashed (new dentry).
748 * Atoimc_open may passin hashed dentries for open.
750 alias = ll_splice_alias(inode, *de);
752 GOTO(out, rc = PTR_ERR(alias));
756 if (!it_disposition(it, DISP_LOOKUP_NEG)) {
757 /* we have lookup look - unhide dentry */
758 if (bits & MDS_INODELOCK_LOOKUP)
759 d_lustre_revalidate(*de);
762 rc = llcrypt_get_encryption_info(inode);
765 if (!llcrypt_has_encryption_key(inode))
766 GOTO(out, rc = -ENOKEY);
768 } else if (!it_disposition(it, DISP_OPEN_CREATE)) {
770 * If file was created on the server, the dentry is revalidated
771 * in ll_create_it if the lock allows for it.
773 /* Check that parent has UPDATE lock. */
774 struct lookup_intent parent_it = {
776 .it_lock_handle = 0 };
777 struct lu_fid fid = ll_i2info(parent)->lli_fid;
779 /* If it is striped directory, get the real stripe parent */
780 if (unlikely(ll_dir_striped(parent))) {
781 rc = md_get_fid_from_lsm(ll_i2mdexp(parent),
782 ll_i2info(parent)->lli_lsm_md,
784 (*de)->d_name.len, &fid);
789 if (md_revalidate_lock(ll_i2mdexp(parent), &parent_it, &fid,
791 d_lustre_revalidate(*de);
792 ll_intent_release(&parent_it);
796 if (it_disposition(it, DISP_OPEN_CREATE)) {
797 ll_stats_ops_tally(ll_i2sbi(parent), LPROC_LL_MKNOD,
798 ktime_us_delta(ktime_get(), kstart));
804 if (rc != 0 && it->it_op & IT_OPEN) {
805 ll_intent_drop_lock(it);
806 ll_open_cleanup((*de)->d_sb, request);
812 static struct dentry *ll_lookup_it(struct inode *parent, struct dentry *dentry,
813 struct lookup_intent *it,
814 void **secctx, __u32 *secctxlen,
815 struct pcc_create_attach *pca,
817 void **encctx, __u32 *encctxlen)
819 ktime_t kstart = ktime_get();
820 struct lookup_intent lookup_it = { .it_op = IT_LOOKUP };
821 struct dentry *save = dentry, *retval;
822 struct ptlrpc_request *req = NULL;
823 struct md_op_data *op_data = NULL;
824 struct lov_user_md *lum = NULL;
827 char secctx_name[XATTR_NAME_MAX + 1];
831 if (dentry->d_name.len > ll_i2sbi(parent)->ll_namelen)
832 RETURN(ERR_PTR(-ENAMETOOLONG));
834 CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p), intent=%s\n",
835 dentry, PFID(ll_inode2fid(parent)), parent, LL_IT2STR(it));
837 if (d_mountpoint(dentry))
838 CERROR("Tell Peter, lookup on mtpt, it %s\n", LL_IT2STR(it));
840 if (it == NULL || it->it_op == IT_GETXATTR)
843 if (it->it_op == IT_GETATTR && dentry_may_statahead(parent, dentry)) {
844 rc = ll_revalidate_statahead(parent, &dentry, 0);
846 RETURN(dentry == save ? NULL : dentry);
849 if (it->it_op & IT_OPEN && it->it_flags & FMODE_WRITE &&
850 dentry->d_sb->s_flags & SB_RDONLY)
851 RETURN(ERR_PTR(-EROFS));
853 if (it->it_op & IT_CREAT)
854 opc = LUSTRE_OPC_CREATE;
856 opc = LUSTRE_OPC_ANY;
858 op_data = ll_prep_md_op_data(NULL, parent, NULL, dentry->d_name.name,
859 dentry->d_name.len, 0, opc, NULL);
861 GOTO(out, retval = ERR_CAST(op_data));
863 /* enforce umask if acl disabled or MDS doesn't support umask */
864 if (!IS_POSIXACL(parent) || !exp_connect_umask(ll_i2mdexp(parent)))
865 it->it_create_mode &= ~current_umask();
867 if (it->it_op & IT_CREAT &&
868 ll_i2sbi(parent)->ll_flags & LL_SBI_FILE_SECCTX) {
869 rc = ll_dentry_init_security(dentry, it->it_create_mode,
871 &op_data->op_file_secctx_name,
872 &op_data->op_file_secctx,
873 &op_data->op_file_secctx_size);
875 GOTO(out, retval = ERR_PTR(rc));
877 *secctx = op_data->op_file_secctx;
878 if (secctxlen != NULL)
879 *secctxlen = op_data->op_file_secctx_size;
883 if (secctxlen != NULL)
886 if (it->it_op & IT_CREAT && encrypt) {
887 rc = llcrypt_inherit_context(parent, NULL, op_data, false);
889 GOTO(out, retval = ERR_PTR(rc));
891 *encctx = op_data->op_file_encctx;
892 if (encctxlen != NULL)
893 *encctxlen = op_data->op_file_encctx_size;
897 if (encctxlen != NULL)
901 /* ask for security context upon intent */
902 if (it->it_op & (IT_LOOKUP | IT_GETATTR | IT_OPEN)) {
903 /* get name of security xattr to request to server */
904 rc = ll_listsecurity(parent, secctx_name,
905 sizeof(secctx_name));
907 CDEBUG(D_SEC, "cannot get security xattr name for "
909 PFID(ll_inode2fid(parent)), rc);
911 op_data->op_file_secctx_name = secctx_name;
912 op_data->op_file_secctx_name_size = rc;
913 CDEBUG(D_SEC, "'%.*s' is security xattr for "DFID"\n",
914 rc, secctx_name, PFID(ll_inode2fid(parent)));
918 if (pca && pca->pca_dataset) {
919 struct pcc_dataset *dataset = pca->pca_dataset;
923 GOTO(out, retval = ERR_PTR(-ENOMEM));
925 lum->lmm_magic = LOV_USER_MAGIC_V1;
926 lum->lmm_pattern = LOV_PATTERN_F_RELEASED | LOV_PATTERN_RAID0;
927 op_data->op_data = lum;
928 op_data->op_data_size = sizeof(*lum);
929 op_data->op_archive_id = dataset->pccd_rwid;
931 rc = obd_fid_alloc(NULL, ll_i2mdexp(parent), &op_data->op_fid2,
934 GOTO(out, retval = ERR_PTR(rc));
936 rc = pcc_inode_create(parent->i_sb, dataset, &op_data->op_fid2,
939 GOTO(out, retval = ERR_PTR(rc));
941 it->it_flags |= MDS_OPEN_PCC;
944 rc = md_intent_lock(ll_i2mdexp(parent), op_data, it, &req,
945 &ll_md_blocking_ast, 0);
946 /* If the MDS allows the client to chgrp (CFS_SETGRP_PERM), but the
947 * client does not know which suppgid should be sent to the MDS, or
948 * some other(s) changed the target file's GID after this RPC sent
949 * to the MDS with the suppgid as the original GID, then we should
950 * try again with right suppgid. */
951 if (rc == -EACCES && it->it_op & IT_OPEN &&
952 it_disposition(it, DISP_OPEN_DENY)) {
953 struct mdt_body *body;
955 LASSERT(req != NULL);
957 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
958 if (op_data->op_suppgids[0] == body->mbo_gid ||
959 op_data->op_suppgids[1] == body->mbo_gid ||
960 !in_group_p(make_kgid(&init_user_ns, body->mbo_gid)))
961 GOTO(out, retval = ERR_PTR(-EACCES));
963 fid_zero(&op_data->op_fid2);
964 op_data->op_suppgids[1] = body->mbo_gid;
965 ptlrpc_req_finished(req);
967 ll_intent_release(it);
968 rc = md_intent_lock(ll_i2mdexp(parent), op_data, it, &req,
969 &ll_md_blocking_ast, 0);
973 GOTO(out, retval = ERR_PTR(rc));
975 /* dir layout may change */
976 ll_unlock_md_op_lsm(op_data);
977 rc = ll_lookup_it_finish(req, it, parent, &dentry,
978 secctx != NULL ? *secctx : NULL,
979 secctxlen != NULL ? *secctxlen : 0,
980 encctx != NULL ? *encctx : NULL,
981 encctxlen != NULL ? *encctxlen : 0,
984 ll_intent_release(it);
985 GOTO(out, retval = ERR_PTR(rc));
988 if ((it->it_op & IT_OPEN) && dentry->d_inode &&
989 !S_ISREG(dentry->d_inode->i_mode) &&
990 !S_ISDIR(dentry->d_inode->i_mode)) {
991 ll_release_openhandle(dentry, it);
993 ll_lookup_finish_locks(it, dentry);
995 GOTO(out, retval = (dentry == save) ? NULL : dentry);
998 if (op_data != NULL && !IS_ERR(op_data)) {
999 if (secctx != NULL && secctxlen != NULL) {
1000 /* caller needs sec ctx info, so reset it in op_data to
1001 * prevent it from being freed */
1002 op_data->op_file_secctx = NULL;
1003 op_data->op_file_secctx_size = 0;
1005 if (encctx != NULL && encctxlen != NULL &&
1006 it->it_op & IT_CREAT && encrypt) {
1007 /* caller needs enc ctx info, so reset it in op_data to
1008 * prevent it from being freed
1010 op_data->op_file_encctx = NULL;
1011 op_data->op_file_encctx_size = 0;
1013 ll_finish_md_op_data(op_data);
1019 ptlrpc_req_finished(req);
1023 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
1026 struct lookup_intent *itp, it = { .it_op = IT_GETATTR };
1029 CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p), flags=%u\n",
1030 dentry, PFID(ll_inode2fid(parent)), parent, flags);
1033 * Optimize away (CREATE && !OPEN). Let .create handle the race.
1034 * but only if we have write permissions there, otherwise we need
1035 * to proceed with lookup. LU-4185
1037 if ((flags & LOOKUP_CREATE) && !(flags & LOOKUP_OPEN) &&
1038 (inode_permission(parent, MAY_WRITE | MAY_EXEC) == 0))
1041 if (flags & (LOOKUP_PARENT|LOOKUP_OPEN|LOOKUP_CREATE))
1045 de = ll_lookup_it(parent, dentry, itp, NULL, NULL, NULL, false,
1049 ll_intent_release(itp);
1054 #ifdef FMODE_CREATED /* added in Linux v4.18-rc1-20-g73a09dd */
1055 # define ll_is_opened(o, f) ((f)->f_mode & FMODE_OPENED)
1056 # define ll_finish_open(f, d, o) finish_open((f), (d), NULL)
1057 # define ll_last_arg
1058 # define ll_set_created(o, f) \
1060 (f)->f_mode |= FMODE_CREATED; \
1064 # define ll_is_opened(o, f) (*(o))
1065 # define ll_finish_open(f, d, o) finish_open((f), (d), NULL, (o))
1066 # define ll_last_arg , int *opened
1067 # define ll_set_created(o, f) \
1069 *(o) |= FILE_CREATED; \
1075 * For cached negative dentry and new dentry, handle lookup/create/open
1078 static int ll_atomic_open(struct inode *dir, struct dentry *dentry,
1079 struct file *file, unsigned open_flags,
1080 umode_t mode ll_last_arg)
1082 struct lookup_intent *it;
1084 long long lookup_flags = LOOKUP_OPEN;
1085 void *secctx = NULL;
1086 __u32 secctxlen = 0;
1087 void *encctx = NULL;
1088 __u32 encctxlen = 0;
1089 struct ll_sb_info *sbi = NULL;
1090 struct pcc_create_attach pca = { NULL, NULL };
1091 bool encrypt = false;
1096 "VFS Op:name=%pd, dir="DFID"(%p), file %p, open_flags %x, mode %x opened %d\n",
1097 dentry, PFID(ll_inode2fid(dir)), dir, file, open_flags, mode,
1098 ll_is_opened(opened, file));
1100 /* Only negative dentries enter here */
1101 LASSERT(dentry->d_inode == NULL);
1103 if (!d_unhashed(dentry)) {
1104 /* A valid negative dentry that just passed revalidation,
1105 * there's little point to try and open it server-side,
1106 * even though there's a minuscule chance it might succeed.
1107 * Either way it's a valid race to just return -ENOENT here.
1109 if (!(open_flags & O_CREAT))
1112 /* Otherwise we just unhash it to be rehashed afresh via
1113 * lookup if necessary
1118 OBD_ALLOC(it, sizeof(*it));
1122 it->it_op = IT_OPEN;
1123 if (open_flags & O_CREAT) {
1124 it->it_op |= IT_CREAT;
1125 lookup_flags |= LOOKUP_CREATE;
1126 sbi = ll_i2sbi(dir);
1127 /* Volatile file is used for HSM restore, so do not use PCC */
1128 if (!filename_is_volatile(dentry->d_name.name,
1129 dentry->d_name.len, NULL)) {
1130 struct pcc_matcher item;
1131 struct pcc_dataset *dataset;
1133 item.pm_uid = from_kuid(&init_user_ns, current_uid());
1134 item.pm_gid = from_kgid(&init_user_ns, current_gid());
1135 item.pm_projid = ll_i2info(dir)->lli_projid;
1136 item.pm_name = &dentry->d_name;
1137 dataset = pcc_dataset_match_get(&sbi->ll_pcc_super,
1139 pca.pca_dataset = dataset;
1142 it->it_create_mode = (mode & S_IALLUGO) | S_IFREG;
1143 it->it_flags = (open_flags & ~O_ACCMODE) | OPEN_FMODE(open_flags);
1144 it->it_flags &= ~MDS_OPEN_FL_INTERNAL;
1146 if (ll_sbi_has_encrypt(ll_i2sbi(dir)) && IS_ENCRYPTED(dir)) {
1147 /* in case of create, this is going to be a regular file because
1148 * we set S_IFREG bit on it->it_create_mode above
1150 rc = llcrypt_get_encryption_info(dir);
1152 GOTO(out_release, rc);
1153 if (open_flags & O_CREAT) {
1154 if (!llcrypt_has_encryption_key(dir))
1155 GOTO(out_release, rc = -ENOKEY);
1160 OBD_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_FILE_PAUSE2, cfs_fail_val);
1162 /* Dentry added to dcache tree in ll_lookup_it */
1163 de = ll_lookup_it(dir, dentry, it, &secctx, &secctxlen, &pca, encrypt,
1164 &encctx, &encctxlen);
1167 else if (de != NULL)
1170 CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_FILE_PAUSE, cfs_fail_val);
1173 if (it_disposition(it, DISP_OPEN_CREATE)) {
1174 /* Dentry instantiated in ll_create_it. */
1175 rc = ll_create_it(dir, dentry, it, secctx, secctxlen,
1176 encrypt, encctx, encctxlen);
1177 security_release_secctx(secctx, secctxlen);
1178 llcrypt_free_ctx(encctx, encctxlen);
1180 /* We dget in ll_splice_alias. */
1186 rc = pcc_inode_create_fini(dentry->d_inode, &pca);
1190 GOTO(out_release, rc);
1193 ll_set_created(opened, file);
1195 /* Open the file with O_CREAT, but the file already
1196 * existed on MDT. This may happend in the case that
1197 * the LOOKUP ibits lock is revoked and the
1198 * corresponding dentry cache is deleted.
1199 * i.e. In the current Lustre, the truncate operation
1200 * will revoke the LOOKUP ibits lock, and the file
1201 * dentry cache will be invalidated. The following open
1202 * with O_CREAT flag will call into ->atomic_open, the
1203 * file was wrongly though as newly created file and
1204 * try to auto cache the file. So after client knows it
1205 * is not a DISP_OPEN_CREATE, it should cleanup the
1206 * already created PCC copy.
1208 pcc_create_attach_cleanup(dir->i_sb, &pca);
1210 if (open_flags & O_CREAT && encrypt &&
1212 rc = ll_set_encflags(dentry->d_inode, encctx,
1214 llcrypt_free_ctx(encctx, encctxlen);
1216 GOTO(out_release, rc);
1220 /* check also if a foreign file is openable */
1221 if (dentry->d_inode && it_disposition(it, DISP_OPEN_OPEN) &&
1222 ll_foreign_is_openable(dentry, open_flags)) {
1224 if (S_ISFIFO(dentry->d_inode->i_mode)) {
1225 /* We cannot call open here as it might
1226 * deadlock. This case is unreachable in
1227 * practice because of OBD_CONNECT_NODEVOH. */
1228 rc = finish_no_open(file, de);
1230 file->private_data = it;
1231 rc = ll_finish_open(file, dentry, opened);
1232 /* We dget in ll_splice_alias. finish_open takes
1233 * care of dget for fd open.
1239 rc = finish_no_open(file, de);
1242 pcc_create_attach_cleanup(dir->i_sb, &pca);
1246 ll_intent_release(it);
1247 OBD_FREE(it, sizeof(*it));
1252 /* We depend on "mode" being set with the proper file type/umask by now */
1253 static struct inode *ll_create_node(struct inode *dir, struct lookup_intent *it)
1255 struct inode *inode = NULL;
1256 struct ptlrpc_request *request = NULL;
1257 struct ll_sb_info *sbi = ll_i2sbi(dir);
1261 LASSERT(it && it->it_disposition);
1263 LASSERT(it_disposition(it, DISP_ENQ_CREATE_REF));
1264 request = it->it_request;
1265 it_clear_disposition(it, DISP_ENQ_CREATE_REF);
1266 rc = ll_prep_inode(&inode, request, dir->i_sb, it);
1268 GOTO(out, inode = ERR_PTR(rc));
1270 /* Pause to allow for a race with concurrent access by fid */
1271 OBD_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_NODE_PAUSE, cfs_fail_val);
1273 /* We asked for a lock on the directory, but were granted a
1274 * lock on the inode. Since we finally have an inode pointer,
1275 * stuff it in the lock. */
1276 CDEBUG(D_DLMTRACE, "setting l_ast_data to inode "DFID"(%p)\n",
1277 PFID(ll_inode2fid(inode)), inode);
1278 ll_set_lock_data(sbi->ll_md_exp, inode, it, NULL);
1281 ptlrpc_req_finished(request);
1286 * By the time this is called, we already have created the directory cache
1287 * entry for the new file, but it is so far negative - it has no inode.
1289 * We defer creating the OBD object(s) until open, to keep the intent and
1290 * non-intent code paths similar, and also because we do not have the MDS
1291 * inode number before calling ll_create_node() (which is needed for LOV),
1292 * so we would need to do yet another RPC to the MDS to store the LOV EA
1293 * data on the MDS. If needed, we would pass the PACKED lmm as data and
1294 * lmm_size in datalen (the MDS still has code which will handle that).
1296 * If the create succeeds, we fill in the inode information
1297 * with d_instantiate().
1299 static int ll_create_it(struct inode *dir, struct dentry *dentry,
1300 struct lookup_intent *it,
1301 void *secctx, __u32 secctxlen, bool encrypt,
1302 void *encctx, __u32 encctxlen)
1304 struct inode *inode;
1309 CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p), intent=%s\n",
1310 dentry, PFID(ll_inode2fid(dir)), dir, LL_IT2STR(it));
1312 rc = it_open_error(DISP_OPEN_CREATE, it);
1316 inode = ll_create_node(dir, it);
1318 RETURN(PTR_ERR(inode));
1320 if ((ll_i2sbi(inode)->ll_flags & LL_SBI_FILE_SECCTX) &&
1322 /* must be done before d_instantiate, because it calls
1323 * security_d_instantiate, which means a getxattr if security
1324 * context is not set yet */
1325 /* no need to protect selinux_inode_setsecurity() by
1326 * inode_lock. Taking it would lead to a client deadlock
1329 rc = security_inode_notifysecctx(inode, secctx, secctxlen);
1334 d_instantiate(dentry, inode);
1337 rc = ll_set_encflags(inode, encctx, encctxlen, true);
1342 if (!(ll_i2sbi(inode)->ll_flags & LL_SBI_FILE_SECCTX)) {
1343 rc = ll_inode_init_security(dentry, inode, dir);
1348 ll_set_lock_data(ll_i2sbi(dir)->ll_md_exp, inode, it, &bits);
1349 if (bits & MDS_INODELOCK_LOOKUP)
1350 d_lustre_revalidate(dentry);
1355 void ll_update_times(struct ptlrpc_request *request, struct inode *inode)
1357 struct mdt_body *body = req_capsule_server_get(&request->rq_pill,
1361 if (body->mbo_valid & OBD_MD_FLMTIME &&
1362 body->mbo_mtime > inode->i_mtime.tv_sec) {
1364 "setting fid " DFID " mtime from %lld to %llu\n",
1365 PFID(ll_inode2fid(inode)),
1366 (s64)inode->i_mtime.tv_sec, body->mbo_mtime);
1367 inode->i_mtime.tv_sec = body->mbo_mtime;
1370 if (body->mbo_valid & OBD_MD_FLCTIME &&
1371 body->mbo_ctime > inode->i_ctime.tv_sec)
1372 inode->i_ctime.tv_sec = body->mbo_ctime;
1375 static int ll_new_node(struct inode *dir, struct dentry *dchild,
1376 const char *tgt, umode_t mode, int rdev, __u32 opc)
1378 struct qstr *name = &dchild->d_name;
1379 struct ptlrpc_request *request = NULL;
1380 struct md_op_data *op_data = NULL;
1381 struct inode *inode = NULL;
1382 struct ll_sb_info *sbi = ll_i2sbi(dir);
1384 bool encrypt = false;
1388 if (unlikely(tgt != NULL))
1389 tgt_len = strlen(tgt) + 1;
1392 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
1393 name->len, 0, opc, NULL);
1394 if (IS_ERR(op_data))
1395 GOTO(err_exit, err = PTR_ERR(op_data));
1397 if (sbi->ll_flags & LL_SBI_FILE_SECCTX) {
1398 err = ll_dentry_init_security(dchild, mode, &dchild->d_name,
1399 &op_data->op_file_secctx_name,
1400 &op_data->op_file_secctx,
1401 &op_data->op_file_secctx_size);
1403 GOTO(err_exit, err);
1406 if (ll_sbi_has_encrypt(sbi) &&
1407 ((IS_ENCRYPTED(dir) &&
1408 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) ||
1409 (unlikely(llcrypt_dummy_context_enabled(dir)) && S_ISDIR(mode)))) {
1410 err = llcrypt_get_encryption_info(dir);
1412 GOTO(err_exit, err);
1413 if (!llcrypt_has_encryption_key(dir))
1414 GOTO(err_exit, err = -ENOKEY);
1419 err = llcrypt_inherit_context(dir, NULL, op_data, false);
1421 GOTO(err_exit, err);
1424 err = md_create(sbi->ll_md_exp, op_data, tgt, tgt_len, mode,
1425 from_kuid(&init_user_ns, current_fsuid()),
1426 from_kgid(&init_user_ns, current_fsgid()),
1427 cfs_curproc_cap_pack(), rdev, &request);
1428 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 14, 58, 0)
1430 * server < 2.12.58 doesn't pack default LMV in intent_getattr reply,
1431 * fetch default LMV here.
1433 if (unlikely(err == -EREMOTE)) {
1434 struct ll_inode_info *lli = ll_i2info(dir);
1435 struct lmv_user_md *lum;
1439 ptlrpc_req_finished(request);
1442 err2 = ll_dir_getstripe(dir, (void **)&lum, &lumsize, &request,
1443 OBD_MD_DEFAULT_MEA);
1444 ll_finish_md_op_data(op_data);
1447 struct lustre_md md = { NULL };
1449 md.body = req_capsule_server_get(&request->rq_pill,
1452 GOTO(err_exit, err = -EPROTO);
1454 OBD_ALLOC_PTR(md.default_lmv);
1455 if (!md.default_lmv)
1456 GOTO(err_exit, err = -ENOMEM);
1458 md.default_lmv->lsm_md_magic = lum->lum_magic;
1459 md.default_lmv->lsm_md_stripe_count =
1460 lum->lum_stripe_count;
1461 md.default_lmv->lsm_md_master_mdt_index =
1462 lum->lum_stripe_offset;
1463 md.default_lmv->lsm_md_hash_type = lum->lum_hash_type;
1465 err = ll_update_inode(dir, &md);
1466 md_free_lustre_md(sbi->ll_md_exp, &md);
1468 GOTO(err_exit, err);
1469 } else if (err2 == -ENODATA && lli->lli_default_lsm_md) {
1471 * If there are no default stripe EA on the MDT, but the
1472 * client has default stripe, then it probably means
1473 * default stripe EA has just been deleted.
1475 down_write(&lli->lli_lsm_sem);
1476 if (lli->lli_default_lsm_md)
1477 OBD_FREE_PTR(lli->lli_default_lsm_md);
1478 lli->lli_default_lsm_md = NULL;
1479 up_write(&lli->lli_lsm_sem);
1481 GOTO(err_exit, err);
1484 ptlrpc_req_finished(request);
1491 GOTO(err_exit, err);
1493 ll_update_times(request, dir);
1495 CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_NEWNODE_PAUSE, cfs_fail_val);
1497 err = ll_prep_inode(&inode, request, dchild->d_sb, NULL);
1499 GOTO(err_exit, err);
1501 if (sbi->ll_flags & LL_SBI_FILE_SECCTX) {
1502 /* must be done before d_instantiate, because it calls
1503 * security_d_instantiate, which means a getxattr if security
1504 * context is not set yet */
1505 /* no need to protect selinux_inode_setsecurity() by
1506 * inode_lock. Taking it would lead to a client deadlock
1509 err = security_inode_notifysecctx(inode,
1510 op_data->op_file_secctx,
1511 op_data->op_file_secctx_size);
1513 GOTO(err_exit, err);
1516 d_instantiate(dchild, inode);
1519 err = ll_set_encflags(inode, op_data->op_file_encctx,
1520 op_data->op_file_encctx_size, true);
1522 GOTO(err_exit, err);
1525 if (!(sbi->ll_flags & LL_SBI_FILE_SECCTX)) {
1526 err = ll_inode_init_security(dchild, inode, dir);
1528 GOTO(err_exit, err);
1533 if (request != NULL)
1534 ptlrpc_req_finished(request);
1536 if (!IS_ERR_OR_NULL(op_data))
1537 ll_finish_md_op_data(op_data);
1542 static int ll_mknod(struct inode *dir, struct dentry *dchild, umode_t mode,
1545 ktime_t kstart = ktime_get();
1549 CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p) mode %o dev %x\n",
1550 dchild, PFID(ll_inode2fid(dir)), dir, mode, rdev);
1552 if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
1553 mode &= ~current_umask();
1555 switch (mode & S_IFMT) {
1564 err = ll_new_node(dir, dchild, NULL, mode, old_encode_dev(rdev),
1575 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKNOD,
1576 ktime_us_delta(ktime_get(), kstart));
1582 * Plain create. Intent create is handled in atomic_open.
1584 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
1585 umode_t mode, bool want_excl)
1587 ktime_t kstart = ktime_get();
1590 CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_FILE_PAUSE, cfs_fail_val);
1593 "VFS Op:name=%pd, dir="DFID"(%p), flags=%u, excl=%d\n",
1594 dentry, PFID(ll_inode2fid(dir)), dir, mode, want_excl);
1596 /* Using mknod(2) to create a regular file is designed to not recognize
1597 * volatile file name, so we use ll_mknod() here. */
1598 rc = ll_mknod(dir, dentry, mode, 0);
1600 CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, unhashed %d\n",
1601 dentry, d_unhashed(dentry));
1604 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_CREATE,
1605 ktime_us_delta(ktime_get(), kstart));
1610 static int ll_symlink(struct inode *dir, struct dentry *dchild,
1611 const char *oldpath)
1613 ktime_t kstart = ktime_get();
1617 CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p), target=%.*s\n",
1618 dchild, PFID(ll_inode2fid(dir)), dir, 3000, oldpath);
1620 err = ll_new_node(dir, dchild, oldpath, S_IFLNK | S_IRWXUGO, 0,
1621 LUSTRE_OPC_SYMLINK);
1624 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_SYMLINK,
1625 ktime_us_delta(ktime_get(), kstart));
1630 static int ll_link(struct dentry *old_dentry, struct inode *dir,
1631 struct dentry *new_dentry)
1633 struct inode *src = old_dentry->d_inode;
1634 struct qstr *name = &new_dentry->d_name;
1635 struct ll_sb_info *sbi = ll_i2sbi(dir);
1636 struct ptlrpc_request *request = NULL;
1637 struct md_op_data *op_data;
1638 ktime_t kstart = ktime_get();
1643 "VFS Op: inode="DFID"(%p), dir="DFID"(%p), target=%pd\n",
1644 PFID(ll_inode2fid(src)), src,
1645 PFID(ll_inode2fid(dir)), dir, new_dentry);
1647 err = llcrypt_prepare_link(old_dentry, dir, new_dentry);
1651 op_data = ll_prep_md_op_data(NULL, src, dir, name->name, name->len,
1652 0, LUSTRE_OPC_ANY, NULL);
1653 if (IS_ERR(op_data))
1654 RETURN(PTR_ERR(op_data));
1656 err = md_link(sbi->ll_md_exp, op_data, &request);
1657 ll_finish_md_op_data(op_data);
1661 ll_update_times(request, dir);
1662 ll_stats_ops_tally(sbi, LPROC_LL_LINK,
1663 ktime_us_delta(ktime_get(), kstart));
1666 ptlrpc_req_finished(request);
1670 static int ll_mkdir(struct inode *dir, struct dentry *dchild, umode_t mode)
1672 ktime_t kstart = ktime_get();
1676 CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p)\n",
1677 dchild, PFID(ll_inode2fid(dir)), dir);
1679 if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
1680 mode &= ~current_umask();
1682 mode = (mode & (S_IRWXUGO|S_ISVTX)) | S_IFDIR;
1684 err = ll_new_node(dir, dchild, NULL, mode, 0, LUSTRE_OPC_MKDIR);
1686 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKDIR,
1687 ktime_us_delta(ktime_get(), kstart));
1692 static int ll_rmdir(struct inode *dir, struct dentry *dchild)
1694 struct qstr *name = &dchild->d_name;
1695 struct ptlrpc_request *request = NULL;
1696 struct md_op_data *op_data;
1697 ktime_t kstart = ktime_get();
1702 CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p)\n",
1703 dchild, PFID(ll_inode2fid(dir)), dir);
1705 if (unlikely(d_mountpoint(dchild)))
1708 /* some foreign dir may not be allowed to be removed */
1709 if (!ll_foreign_is_removable(dchild, false))
1712 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len,
1713 S_IFDIR, LUSTRE_OPC_ANY, NULL);
1714 if (IS_ERR(op_data))
1715 RETURN(PTR_ERR(op_data));
1717 if (dchild->d_inode != NULL)
1718 op_data->op_fid3 = *ll_inode2fid(dchild->d_inode);
1720 op_data->op_fid2 = op_data->op_fid3;
1721 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1722 ll_finish_md_op_data(op_data);
1724 struct mdt_body *body;
1726 ll_update_times(request, dir);
1727 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR,
1728 ktime_us_delta(ktime_get(), kstart));
1731 * The server puts attributes in on the last unlink, use them
1732 * to update the link count so the inode can be freed
1735 body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
1736 if (body->mbo_valid & OBD_MD_FLNLINK)
1737 set_nlink(dchild->d_inode, body->mbo_nlink);
1740 ptlrpc_req_finished(request);
1748 int ll_rmdir_entry(struct inode *dir, char *name, int namelen)
1750 struct ptlrpc_request *request = NULL;
1751 struct md_op_data *op_data;
1752 ktime_t kstart = ktime_get();
1756 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1757 namelen, name, PFID(ll_inode2fid(dir)), dir);
1759 op_data = ll_prep_md_op_data(NULL, dir, NULL, name, strlen(name),
1760 S_IFDIR, LUSTRE_OPC_ANY, NULL);
1761 if (IS_ERR(op_data))
1762 RETURN(PTR_ERR(op_data));
1763 op_data->op_cli_flags |= CLI_RM_ENTRY;
1764 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1765 ll_finish_md_op_data(op_data);
1767 ll_update_times(request, dir);
1769 ptlrpc_req_finished(request);
1771 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR,
1772 ktime_us_delta(ktime_get(), kstart));
1776 static int ll_unlink(struct inode *dir, struct dentry *dchild)
1778 struct qstr *name = &dchild->d_name;
1779 struct ptlrpc_request *request = NULL;
1780 struct md_op_data *op_data;
1781 struct mdt_body *body;
1782 ktime_t kstart = ktime_get();
1787 CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p)\n",
1788 dchild, PFID(ll_inode2fid(dir)), dir);
1791 * XXX: unlink bind mountpoint maybe call to here,
1792 * just check it as vfs_unlink does.
1794 if (unlikely(d_mountpoint(dchild)))
1797 /* some foreign file/dir may not be allowed to be unlinked */
1798 if (!ll_foreign_is_removable(dchild, false))
1801 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len, 0,
1802 LUSTRE_OPC_ANY, NULL);
1803 if (IS_ERR(op_data))
1804 RETURN(PTR_ERR(op_data));
1806 op_data->op_fid3 = *ll_inode2fid(dchild->d_inode);
1807 /* notify lower layer if inode has dirty pages */
1808 if (S_ISREG(dchild->d_inode->i_mode) &&
1809 ll_i2info(dchild->d_inode)->lli_clob &&
1810 dirty_cnt(dchild->d_inode))
1811 op_data->op_cli_flags |= CLI_DIRTY_DATA;
1812 op_data->op_fid2 = op_data->op_fid3;
1813 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1814 ll_finish_md_op_data(op_data);
1819 * The server puts attributes in on the last unlink, use them to update
1820 * the link count so the inode can be freed immediately.
1822 body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
1823 if (body->mbo_valid & OBD_MD_FLNLINK)
1824 set_nlink(dchild->d_inode, body->mbo_nlink);
1826 ll_update_times(request, dir);
1829 ptlrpc_req_finished(request);
1831 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_UNLINK,
1832 ktime_us_delta(ktime_get(), kstart));
1836 static int ll_rename(struct inode *src, struct dentry *src_dchild,
1837 struct inode *tgt, struct dentry *tgt_dchild
1838 #ifdef HAVE_IOPS_RENAME_WITH_FLAGS
1839 , unsigned int flags
1843 struct qstr *src_name = &src_dchild->d_name;
1844 struct qstr *tgt_name = &tgt_dchild->d_name;
1845 struct ptlrpc_request *request = NULL;
1846 struct ll_sb_info *sbi = ll_i2sbi(src);
1847 struct md_op_data *op_data;
1848 ktime_t kstart = ktime_get();
1853 #ifdef HAVE_IOPS_RENAME_WITH_FLAGS
1859 "VFS Op:oldname=%pd, src_dir="DFID"(%p), newname=%pd, tgt_dir="DFID"(%p)\n",
1860 src_dchild, PFID(ll_inode2fid(src)), src,
1861 tgt_dchild, PFID(ll_inode2fid(tgt)), tgt);
1863 if (unlikely(d_mountpoint(src_dchild) || d_mountpoint(tgt_dchild)))
1866 #ifdef HAVE_IOPS_RENAME_WITH_FLAGS
1867 err = llcrypt_prepare_rename(src, src_dchild, tgt, tgt_dchild, flags);
1869 err = llcrypt_prepare_rename(src, src_dchild, tgt, tgt_dchild, 0);
1874 if (src_dchild->d_inode)
1875 mode = src_dchild->d_inode->i_mode;
1877 if (tgt_dchild->d_inode)
1878 mode = tgt_dchild->d_inode->i_mode;
1880 op_data = ll_prep_md_op_data(NULL, src, tgt, NULL, 0, mode,
1881 LUSTRE_OPC_ANY, NULL);
1882 if (IS_ERR(op_data))
1883 RETURN(PTR_ERR(op_data));
1885 if (src_dchild->d_inode)
1886 op_data->op_fid3 = *ll_inode2fid(src_dchild->d_inode);
1888 if (tgt_dchild->d_inode)
1889 op_data->op_fid4 = *ll_inode2fid(tgt_dchild->d_inode);
1891 err = md_rename(sbi->ll_md_exp, op_data,
1892 src_name->name, src_name->len,
1893 tgt_name->name, tgt_name->len, &request);
1894 ll_finish_md_op_data(op_data);
1896 ll_update_times(request, src);
1897 ll_update_times(request, tgt);
1900 ptlrpc_req_finished(request);
1903 d_move(src_dchild, tgt_dchild);
1904 ll_stats_ops_tally(sbi, LPROC_LL_RENAME,
1905 ktime_us_delta(ktime_get(), kstart));
1911 const struct inode_operations ll_dir_inode_operations = {
1913 .atomic_open = ll_atomic_open,
1914 .lookup = ll_lookup_nd,
1915 .create = ll_create_nd,
1916 /* We need all these non-raw things for NFSD, to not patch it. */
1917 .unlink = ll_unlink,
1920 .symlink = ll_symlink,
1922 .rename = ll_rename,
1923 .setattr = ll_setattr,
1924 .getattr = ll_getattr,
1925 .permission = ll_inode_permission,
1926 #ifdef HAVE_IOP_XATTR
1927 .setxattr = ll_setxattr,
1928 .getxattr = ll_getxattr,
1929 .removexattr = ll_removexattr,
1931 .listxattr = ll_listxattr,
1932 .get_acl = ll_get_acl,
1933 #ifdef HAVE_IOP_SET_ACL
1934 .set_acl = ll_set_acl,
1938 const struct inode_operations ll_special_inode_operations = {
1939 .setattr = ll_setattr,
1940 .getattr = ll_getattr,
1941 .permission = ll_inode_permission,
1942 #ifdef HAVE_IOP_XATTR
1943 .setxattr = ll_setxattr,
1944 .getxattr = ll_getxattr,
1945 .removexattr = ll_removexattr,
1947 .listxattr = ll_listxattr,
1948 .get_acl = ll_get_acl,
1949 #ifdef HAVE_IOP_SET_ACL
1950 .set_acl = ll_set_acl,