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 if (!lli->lli_clob) {
194 /* due to DoM read on open, there may exist pages for Lustre
195 * regular file even though cl_object is not set up yet. */
196 truncate_inode_pages(inode->i_mapping, 0);
200 env = cl_env_get(&refcheck);
202 RETURN(PTR_ERR(env));
204 /* reach MDC layer to flush data under the DoM ldlm lock */
205 rc = cl_object_flush(env, lli->lli_clob, lock);
206 if (rc == -ENODATA) {
207 CDEBUG(D_INODE, "inode "DFID" layout has no DoM stripe\n",
208 PFID(ll_inode2fid(inode)));
209 /* most likely result of layout change, do nothing */
213 cl_env_put(env, &refcheck);
217 static void ll_lock_cancel_bits(struct ldlm_lock *lock, __u64 to_cancel)
219 struct inode *inode = ll_inode_from_resource_lock(lock);
220 struct ll_inode_info *lli;
221 __u64 bits = to_cancel;
227 /* That means the inode is evicted most likely and may cause
228 * the skipping of lock cleanups below, so print the message
231 if (lock->l_resource->lr_lvb_inode)
233 "can't take inode for the lock (%sevicted)\n",
234 lock->l_resource->lr_lvb_inode->i_state &
235 I_FREEING ? "" : "not ");
239 if (!fid_res_name_eq(ll_inode2fid(inode),
240 &lock->l_resource->lr_name)) {
241 LDLM_ERROR(lock, "data mismatch with object "DFID"(%p)",
242 PFID(ll_inode2fid(inode)), inode);
246 if (bits & MDS_INODELOCK_XATTR) {
247 ll_xattr_cache_destroy(inode);
248 bits &= ~MDS_INODELOCK_XATTR;
251 /* For OPEN locks we differentiate between lock modes
252 * LCK_CR, LCK_CW, LCK_PR - bug 22891 */
253 if (bits & MDS_INODELOCK_OPEN)
254 ll_have_md_lock(inode, &bits, lock->l_req_mode);
256 if (bits & MDS_INODELOCK_OPEN) {
259 switch (lock->l_req_mode) {
270 LDLM_ERROR(lock, "bad lock mode for OPEN lock");
274 ll_md_real_close(inode, fmode);
276 bits &= ~MDS_INODELOCK_OPEN;
279 if (bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
280 MDS_INODELOCK_LAYOUT | MDS_INODELOCK_PERM |
282 ll_have_md_lock(inode, &bits, LCK_MINMODE);
284 if (bits & MDS_INODELOCK_DOM) {
285 rc = ll_dom_lock_cancel(inode, lock);
287 CDEBUG(D_INODE, "cannot flush DoM data "
289 PFID(ll_inode2fid(inode)), rc);
292 if (bits & MDS_INODELOCK_LAYOUT) {
293 struct cl_object_conf conf = {
294 .coc_opc = OBJECT_CONF_INVALIDATE,
298 rc = ll_layout_conf(inode, &conf);
300 CDEBUG(D_INODE, "cannot invalidate layout of "
302 PFID(ll_inode2fid(inode)), rc);
305 lli = ll_i2info(inode);
307 if (bits & MDS_INODELOCK_UPDATE)
308 ll_file_set_flag(lli, LLIF_UPDATE_ATIME);
310 if ((bits & MDS_INODELOCK_UPDATE) && S_ISDIR(inode->i_mode)) {
311 CDEBUG(D_INODE, "invalidating inode "DFID" lli = %p, "
312 "pfid = "DFID"\n", PFID(ll_inode2fid(inode)),
313 lli, PFID(&lli->lli_pfid));
314 truncate_inode_pages(inode->i_mapping, 0);
316 if (unlikely(!fid_is_zero(&lli->lli_pfid))) {
317 struct inode *master_inode = NULL;
320 /* This is slave inode, since all of the child dentry
321 * is connected on the master inode, so we have to
322 * invalidate the negative children on master inode */
323 CDEBUG(D_INODE, "Invalidate s"DFID" m"DFID"\n",
324 PFID(ll_inode2fid(inode)), PFID(&lli->lli_pfid));
326 hash = cl_fid_build_ino(&lli->lli_pfid,
327 ll_need_32bit_api(ll_i2sbi(inode)));
329 /* Do not lookup the inode with ilookup5, otherwise
330 * it will cause dead lock,
331 * 1. Client1 send chmod req to the MDT0, then on MDT0,
332 * it enqueues master and all of its slaves lock,
333 * (mdt_attr_set() -> mdt_lock_slaves()), after gets
334 * master and stripe0 lock, it will send the enqueue
335 * req (for stripe1) to MDT1, then MDT1 finds the lock
336 * has been granted to client2. Then MDT1 sends blocking
338 * 2. At the same time, client2 tries to unlink
339 * the striped dir (rm -rf striped_dir), and during
340 * lookup, it will hold the master inode of the striped
341 * directory, whose inode state is NEW, then tries to
342 * revalidate all of its slaves, (ll_prep_inode()->
343 * ll_iget()->ll_read_inode2()-> ll_update_inode().).
344 * And it will be blocked on the server side because
346 * 3. Then the client get the blocking_ast req, cancel
347 * the lock, but being blocked if using ->ilookup5()),
348 * because master inode state is NEW. */
349 master_inode = ilookup5_nowait(inode->i_sb, hash,
350 ll_test_inode_by_fid,
351 (void *)&lli->lli_pfid);
353 ll_invalidate_negative_children(master_inode);
357 ll_invalidate_negative_children(inode);
361 if ((bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM)) &&
362 inode->i_sb->s_root != NULL &&
363 inode != inode->i_sb->s_root->d_inode)
364 ll_invalidate_aliases(inode);
366 if (bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM))
367 forget_all_cached_acls(inode);
373 /* Check if the given lock may be downgraded instead of canceling and
374 * that convert is really needed. */
375 int ll_md_need_convert(struct ldlm_lock *lock)
377 struct ldlm_namespace *ns = ldlm_lock_to_ns(lock);
379 __u64 wanted = lock->l_policy_data.l_inodebits.cancel_bits;
380 __u64 bits = lock->l_policy_data.l_inodebits.bits & ~wanted;
381 enum ldlm_mode mode = LCK_MINMODE;
383 if (!lock->l_conn_export ||
384 !exp_connect_lock_convert(lock->l_conn_export))
387 if (!wanted || !bits || ldlm_is_cancel(lock))
390 /* do not convert locks other than DOM for now */
391 if (!((bits | wanted) & MDS_INODELOCK_DOM))
394 /* We may have already remaining bits in some other lock so
395 * lock convert will leave us just extra lock for the same bit.
396 * Check if client has other lock with the same bits and the same
397 * or lower mode and don't convert if any.
399 switch (lock->l_req_mode) {
413 /* do not convert other modes */
417 /* is lock is too old to be converted? */
418 lock_res_and_lock(lock);
419 if (ktime_after(ktime_get(),
420 ktime_add(lock->l_last_used, ns->ns_dirty_age_limit))) {
421 unlock_res_and_lock(lock);
424 unlock_res_and_lock(lock);
426 inode = ll_inode_from_resource_lock(lock);
427 ll_have_md_lock(inode, &bits, mode);
432 int ll_md_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *ld,
433 void *data, int flag)
435 struct lustre_handle lockh;
441 case LDLM_CB_BLOCKING:
443 __u64 cancel_flags = LCF_ASYNC;
445 /* if lock convert is not needed then still have to
446 * pass lock via ldlm_cli_convert() to keep all states
447 * correct, set cancel_bits to full lock bits to cause
448 * full cancel to happen.
450 if (!ll_md_need_convert(lock)) {
451 lock_res_and_lock(lock);
452 lock->l_policy_data.l_inodebits.cancel_bits =
453 lock->l_policy_data.l_inodebits.bits;
454 unlock_res_and_lock(lock);
456 rc = ldlm_cli_convert(lock, cancel_flags);
459 /* continue with cancel otherwise */
460 ldlm_lock2handle(lock, &lockh);
461 rc = ldlm_cli_cancel(&lockh, cancel_flags);
463 CDEBUG(D_INODE, "ldlm_cli_cancel: rc = %d\n", rc);
468 case LDLM_CB_CANCELING:
470 __u64 to_cancel = lock->l_policy_data.l_inodebits.bits;
472 /* Nothing to do for non-granted locks */
473 if (!ldlm_is_granted(lock))
476 /* If 'ld' is supplied then bits to be cancelled are passed
477 * implicitly by lock converting and cancel_bits from 'ld'
478 * should be used. Otherwise full cancel is being performed
479 * and lock inodebits are used.
481 * Note: we cannot rely on cancel_bits in lock itself at this
482 * moment because they can be changed by concurrent thread,
483 * so ldlm_cli_inodebits_convert() pass cancel bits implicitly
487 /* partial bits cancel allowed only during convert */
488 LASSERT(ldlm_is_converting(lock));
489 /* mask cancel bits by lock bits so only no any unused
490 * bits are passed to ll_lock_cancel_bits()
492 to_cancel &= ld->l_policy_data.l_inodebits.cancel_bits;
494 ll_lock_cancel_bits(lock, to_cancel);
504 __u32 ll_i2suppgid(struct inode *i)
506 if (in_group_p(i->i_gid))
507 return (__u32)from_kgid(&init_user_ns, i->i_gid);
509 return (__u32) __kgid_val(INVALID_GID);
512 /* Pack the required supplementary groups into the supplied groups array.
513 * If we don't need to use the groups from the target inode(s) then we
514 * instead pack one or more groups from the user's supplementary group
515 * array in case it might be useful. Not needed if doing an MDS-side upcall. */
516 void ll_i2gids(__u32 *suppgids, struct inode *i1, struct inode *i2)
519 LASSERT(suppgids != NULL);
521 suppgids[0] = ll_i2suppgid(i1);
524 suppgids[1] = ll_i2suppgid(i2);
530 * try to reuse three types of dentry:
531 * 1. unhashed alias, this one is unhashed by d_invalidate (but it may be valid
532 * by concurrent .revalidate).
533 * 2. INVALID alias (common case for no valid ldlm lock held, but this flag may
534 * be cleared by others calling d_lustre_revalidate).
535 * 3. DISCONNECTED alias.
537 static struct dentry *ll_find_alias(struct inode *inode, struct dentry *dentry)
539 struct dentry *alias, *discon_alias, *invalid_alias;
541 if (hlist_empty(&inode->i_dentry))
544 discon_alias = invalid_alias = NULL;
546 spin_lock(&inode->i_lock);
547 hlist_for_each_entry(alias, &inode->i_dentry, d_alias) {
548 LASSERT(alias != dentry);
550 spin_lock(&alias->d_lock);
551 if ((alias->d_flags & DCACHE_DISCONNECTED) &&
552 S_ISDIR(inode->i_mode))
553 /* LASSERT(last_discon == NULL); LU-405, bz 20055 */
554 discon_alias = alias;
555 else if (alias->d_parent == dentry->d_parent &&
556 alias->d_name.hash == dentry->d_name.hash &&
557 alias->d_name.len == dentry->d_name.len &&
558 memcmp(alias->d_name.name, dentry->d_name.name,
559 dentry->d_name.len) == 0)
560 invalid_alias = alias;
561 spin_unlock(&alias->d_lock);
566 alias = invalid_alias ?: discon_alias ?: NULL;
568 spin_lock(&alias->d_lock);
570 spin_unlock(&alias->d_lock);
572 spin_unlock(&inode->i_lock);
578 * Similar to d_splice_alias(), but lustre treats invalid alias
579 * similar to DCACHE_DISCONNECTED, and tries to use it anyway.
581 struct dentry *ll_splice_alias(struct inode *inode, struct dentry *de)
587 new = ll_find_alias(inode, de);
597 "Reuse dentry %p inode %p refc %d flags %#x\n",
598 new, new->d_inode, ll_d_count(new), new->d_flags);
606 CDEBUG(D_DENTRY, "Add dentry %p inode %p refc %d flags %#x\n",
607 de, de->d_inode, ll_d_count(de), de->d_flags);
611 static int ll_lookup_it_finish(struct ptlrpc_request *request,
612 struct lookup_intent *it,
613 struct inode *parent, struct dentry **de,
614 void *secctx, __u32 secctxlen,
615 void *encctx, __u32 encctxlen,
616 ktime_t kstart, bool encrypt)
618 struct inode *inode = NULL;
621 struct dentry *alias;
624 /* NB 1 request reference will be taken away by ll_intent_lock()
626 CDEBUG(D_DENTRY, "it %p it_disposition %x\n", it,
628 if (!it_disposition(it, DISP_LOOKUP_NEG)) {
629 struct req_capsule *pill = &request->rq_pill;
630 struct mdt_body *body = req_capsule_server_get(pill,
633 rc = ll_prep_inode(&inode, request, (*de)->d_sb, it);
637 /* If encryption context was returned by MDT, put it in
638 * inode now to save an extra getxattr and avoid deadlock.
640 if (body->mbo_valid & OBD_MD_ENCCTX) {
641 encctx = req_capsule_server_get(pill, &RMF_FILE_ENCCTX);
642 encctxlen = req_capsule_get_size(pill,
648 "server returned encryption ctx for "DFID"\n",
649 PFID(ll_inode2fid(inode)));
650 rc = ll_xattr_cache_insert(inode,
651 LL_XATTR_NAME_ENCRYPTION_CONTEXT,
654 CWARN("%s: cannot set enc ctx for "DFID": rc = %d\n",
655 ll_i2sbi(inode)->ll_fsname,
656 PFID(ll_inode2fid(inode)), rc);
658 rc = llcrypt_get_encryption_info(inode);
661 "cannot get enc info for "DFID": rc = %d\n",
662 PFID(ll_inode2fid(inode)), rc);
667 if (it->it_op & IT_OPEN)
668 ll_dom_finish_open(inode, request, it);
670 ll_set_lock_data(ll_i2sbi(parent)->ll_md_exp, inode, it, &bits);
672 /* We used to query real size from OSTs here, but actually
673 * this is not needed. For stat() calls size would be updated
674 * from subsequent do_revalidate()->ll_inode_revalidate_it() in
676 * vfs_getattr_it->ll_getattr()->ll_inode_revalidate_it() in 2.6
677 * Everybody else who needs correct file size would call
678 * ll_glimpse_size or some equivalent themselves anyway.
682 /* If security context was returned by MDT, put it in
683 * inode now to save an extra getxattr from security hooks,
684 * and avoid deadlock.
686 if (body->mbo_valid & OBD_MD_SECCTX) {
687 secctx = req_capsule_server_get(pill, &RMF_FILE_SECCTX);
688 secctxlen = req_capsule_get_size(pill,
693 CDEBUG(D_SEC, "server returned security context"
695 PFID(ll_inode2fid(inode)));
698 if (secctx != NULL && secctxlen != 0) {
699 /* no need to protect selinux_inode_setsecurity() by
700 * inode_lock. Taking it would lead to a client deadlock
703 rc = security_inode_notifysecctx(inode, secctx,
706 CWARN("%s: cannot set security context for "DFID": rc = %d\n",
707 ll_i2sbi(inode)->ll_fsname,
708 PFID(ll_inode2fid(inode)),
713 /* Only hash *de if it is unhashed (new dentry).
714 * Atoimc_open may passin hashed dentries for open.
716 alias = ll_splice_alias(inode, *de);
718 GOTO(out, rc = PTR_ERR(alias));
722 if (!it_disposition(it, DISP_LOOKUP_NEG)) {
723 /* we have lookup look - unhide dentry */
724 if (bits & MDS_INODELOCK_LOOKUP)
725 d_lustre_revalidate(*de);
728 rc = llcrypt_get_encryption_info(inode);
731 if (!llcrypt_has_encryption_key(inode))
732 GOTO(out, rc = -ENOKEY);
734 } else if (!it_disposition(it, DISP_OPEN_CREATE)) {
736 * If file was created on the server, the dentry is revalidated
737 * in ll_create_it if the lock allows for it.
739 /* Check that parent has UPDATE lock. */
740 struct lookup_intent parent_it = {
742 .it_lock_handle = 0 };
743 struct lu_fid fid = ll_i2info(parent)->lli_fid;
745 /* If it is striped directory, get the real stripe parent */
746 if (unlikely(ll_dir_striped(parent))) {
747 rc = md_get_fid_from_lsm(ll_i2mdexp(parent),
748 ll_i2info(parent)->lli_lsm_md,
750 (*de)->d_name.len, &fid);
755 if (md_revalidate_lock(ll_i2mdexp(parent), &parent_it, &fid,
757 d_lustre_revalidate(*de);
758 ll_intent_release(&parent_it);
762 if (it_disposition(it, DISP_OPEN_CREATE)) {
763 ll_stats_ops_tally(ll_i2sbi(parent), LPROC_LL_MKNOD,
764 ktime_us_delta(ktime_get(), kstart));
770 if (rc != 0 && it->it_op & IT_OPEN) {
771 ll_intent_drop_lock(it);
772 ll_open_cleanup((*de)->d_sb, request);
778 static struct dentry *ll_lookup_it(struct inode *parent, struct dentry *dentry,
779 struct lookup_intent *it,
780 void **secctx, __u32 *secctxlen,
781 struct pcc_create_attach *pca,
783 void **encctx, __u32 *encctxlen)
785 ktime_t kstart = ktime_get();
786 struct lookup_intent lookup_it = { .it_op = IT_LOOKUP };
787 struct dentry *save = dentry, *retval;
788 struct ptlrpc_request *req = NULL;
789 struct md_op_data *op_data = NULL;
790 struct lov_user_md *lum = NULL;
793 char secctx_name[XATTR_NAME_MAX + 1];
797 if (dentry->d_name.len > ll_i2sbi(parent)->ll_namelen)
798 RETURN(ERR_PTR(-ENAMETOOLONG));
800 CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p), intent=%s\n",
801 dentry, PFID(ll_inode2fid(parent)), parent, LL_IT2STR(it));
803 if (d_mountpoint(dentry))
804 CERROR("Tell Peter, lookup on mtpt, it %s\n", LL_IT2STR(it));
806 if (it == NULL || it->it_op == IT_GETXATTR)
809 if (it->it_op == IT_GETATTR && dentry_may_statahead(parent, dentry)) {
810 rc = ll_revalidate_statahead(parent, &dentry, 0);
812 RETURN(dentry == save ? NULL : dentry);
815 if (it->it_op & IT_OPEN && it->it_flags & FMODE_WRITE &&
816 dentry->d_sb->s_flags & SB_RDONLY)
817 RETURN(ERR_PTR(-EROFS));
819 if (it->it_op & IT_CREAT)
820 opc = LUSTRE_OPC_CREATE;
822 opc = LUSTRE_OPC_ANY;
824 op_data = ll_prep_md_op_data(NULL, parent, NULL, dentry->d_name.name,
825 dentry->d_name.len, 0, opc, NULL);
827 GOTO(out, retval = ERR_CAST(op_data));
829 /* enforce umask if acl disabled or MDS doesn't support umask */
830 if (!IS_POSIXACL(parent) || !exp_connect_umask(ll_i2mdexp(parent)))
831 it->it_create_mode &= ~current_umask();
833 if (it->it_op & IT_CREAT &&
834 ll_i2sbi(parent)->ll_flags & LL_SBI_FILE_SECCTX) {
835 rc = ll_dentry_init_security(dentry, it->it_create_mode,
837 &op_data->op_file_secctx_name,
838 &op_data->op_file_secctx,
839 &op_data->op_file_secctx_size);
841 GOTO(out, retval = ERR_PTR(rc));
843 *secctx = op_data->op_file_secctx;
844 if (secctxlen != NULL)
845 *secctxlen = op_data->op_file_secctx_size;
849 if (secctxlen != NULL)
852 if (it->it_op & IT_CREAT && encrypt) {
853 rc = llcrypt_inherit_context(parent, NULL, op_data, false);
855 GOTO(out, retval = ERR_PTR(rc));
857 *encctx = op_data->op_file_encctx;
858 if (encctxlen != NULL)
859 *encctxlen = op_data->op_file_encctx_size;
863 if (encctxlen != NULL)
867 /* ask for security context upon intent */
868 if (it->it_op & (IT_LOOKUP | IT_GETATTR | IT_OPEN)) {
869 /* get name of security xattr to request to server */
870 rc = ll_listsecurity(parent, secctx_name,
871 sizeof(secctx_name));
873 CDEBUG(D_SEC, "cannot get security xattr name for "
875 PFID(ll_inode2fid(parent)), rc);
877 op_data->op_file_secctx_name = secctx_name;
878 op_data->op_file_secctx_name_size = rc;
879 CDEBUG(D_SEC, "'%.*s' is security xattr for "DFID"\n",
880 rc, secctx_name, PFID(ll_inode2fid(parent)));
884 if (pca && pca->pca_dataset) {
885 struct pcc_dataset *dataset = pca->pca_dataset;
889 GOTO(out, retval = ERR_PTR(-ENOMEM));
891 lum->lmm_magic = LOV_USER_MAGIC_V1;
892 lum->lmm_pattern = LOV_PATTERN_F_RELEASED | LOV_PATTERN_RAID0;
893 op_data->op_data = lum;
894 op_data->op_data_size = sizeof(*lum);
895 op_data->op_archive_id = dataset->pccd_rwid;
897 rc = obd_fid_alloc(NULL, ll_i2mdexp(parent), &op_data->op_fid2,
900 GOTO(out, retval = ERR_PTR(rc));
902 rc = pcc_inode_create(parent->i_sb, dataset, &op_data->op_fid2,
905 GOTO(out, retval = ERR_PTR(rc));
907 it->it_flags |= MDS_OPEN_PCC;
910 rc = md_intent_lock(ll_i2mdexp(parent), op_data, it, &req,
911 &ll_md_blocking_ast, 0);
912 /* If the MDS allows the client to chgrp (CFS_SETGRP_PERM), but the
913 * client does not know which suppgid should be sent to the MDS, or
914 * some other(s) changed the target file's GID after this RPC sent
915 * to the MDS with the suppgid as the original GID, then we should
916 * try again with right suppgid. */
917 if (rc == -EACCES && it->it_op & IT_OPEN &&
918 it_disposition(it, DISP_OPEN_DENY)) {
919 struct mdt_body *body;
921 LASSERT(req != NULL);
923 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
924 if (op_data->op_suppgids[0] == body->mbo_gid ||
925 op_data->op_suppgids[1] == body->mbo_gid ||
926 !in_group_p(make_kgid(&init_user_ns, body->mbo_gid)))
927 GOTO(out, retval = ERR_PTR(-EACCES));
929 fid_zero(&op_data->op_fid2);
930 op_data->op_suppgids[1] = body->mbo_gid;
931 ptlrpc_req_finished(req);
933 ll_intent_release(it);
934 rc = md_intent_lock(ll_i2mdexp(parent), op_data, it, &req,
935 &ll_md_blocking_ast, 0);
939 GOTO(out, retval = ERR_PTR(rc));
941 /* dir layout may change */
942 ll_unlock_md_op_lsm(op_data);
943 rc = ll_lookup_it_finish(req, it, parent, &dentry,
944 secctx != NULL ? *secctx : NULL,
945 secctxlen != NULL ? *secctxlen : 0,
946 encctx != NULL ? *encctx : NULL,
947 encctxlen != NULL ? *encctxlen : 0,
950 ll_intent_release(it);
951 GOTO(out, retval = ERR_PTR(rc));
954 if ((it->it_op & IT_OPEN) && dentry->d_inode &&
955 !S_ISREG(dentry->d_inode->i_mode) &&
956 !S_ISDIR(dentry->d_inode->i_mode)) {
957 ll_release_openhandle(dentry, it);
959 ll_lookup_finish_locks(it, dentry);
961 GOTO(out, retval = (dentry == save) ? NULL : dentry);
964 if (op_data != NULL && !IS_ERR(op_data)) {
965 if (secctx != NULL && secctxlen != NULL) {
966 /* caller needs sec ctx info, so reset it in op_data to
967 * prevent it from being freed */
968 op_data->op_file_secctx = NULL;
969 op_data->op_file_secctx_size = 0;
971 if (encctx != NULL && encctxlen != NULL &&
972 it->it_op & IT_CREAT && encrypt) {
973 /* caller needs enc ctx info, so reset it in op_data to
974 * prevent it from being freed
976 op_data->op_file_encctx = NULL;
977 op_data->op_file_encctx_size = 0;
979 ll_finish_md_op_data(op_data);
985 ptlrpc_req_finished(req);
989 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
992 struct lookup_intent *itp, it = { .it_op = IT_GETATTR };
995 CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p), flags=%u\n",
996 dentry, PFID(ll_inode2fid(parent)), parent, flags);
999 * Optimize away (CREATE && !OPEN). Let .create handle the race.
1000 * but only if we have write permissions there, otherwise we need
1001 * to proceed with lookup. LU-4185
1003 if ((flags & LOOKUP_CREATE) && !(flags & LOOKUP_OPEN) &&
1004 (inode_permission(parent, MAY_WRITE | MAY_EXEC) == 0))
1007 if (flags & (LOOKUP_PARENT|LOOKUP_OPEN|LOOKUP_CREATE))
1011 de = ll_lookup_it(parent, dentry, itp, NULL, NULL, NULL, false,
1015 ll_intent_release(itp);
1020 #ifdef FMODE_CREATED /* added in Linux v4.18-rc1-20-g73a09dd */
1021 # define ll_is_opened(o, f) ((f)->f_mode & FMODE_OPENED)
1022 # define ll_finish_open(f, d, o) finish_open((f), (d), NULL)
1023 # define ll_last_arg
1024 # define ll_set_created(o, f) \
1026 (f)->f_mode |= FMODE_CREATED; \
1030 # define ll_is_opened(o, f) (*(o))
1031 # define ll_finish_open(f, d, o) finish_open((f), (d), NULL, (o))
1032 # define ll_last_arg , int *opened
1033 # define ll_set_created(o, f) \
1035 *(o) |= FILE_CREATED; \
1041 * For cached negative dentry and new dentry, handle lookup/create/open
1044 static int ll_atomic_open(struct inode *dir, struct dentry *dentry,
1045 struct file *file, unsigned open_flags,
1046 umode_t mode ll_last_arg)
1048 struct lookup_intent *it;
1050 long long lookup_flags = LOOKUP_OPEN;
1051 void *secctx = NULL;
1052 __u32 secctxlen = 0;
1053 void *encctx = NULL;
1054 __u32 encctxlen = 0;
1055 struct ll_sb_info *sbi = NULL;
1056 struct pcc_create_attach pca = { NULL, NULL };
1057 bool encrypt = false;
1062 "VFS Op:name=%pd, dir="DFID"(%p), file %p, open_flags %x, mode %x opened %d\n",
1063 dentry, PFID(ll_inode2fid(dir)), dir, file, open_flags, mode,
1064 ll_is_opened(opened, file));
1066 /* Only negative dentries enter here */
1067 LASSERT(dentry->d_inode == NULL);
1069 if (!d_unhashed(dentry)) {
1070 /* A valid negative dentry that just passed revalidation,
1071 * there's little point to try and open it server-side,
1072 * even though there's a minuscule chance it might succeed.
1073 * Either way it's a valid race to just return -ENOENT here.
1075 if (!(open_flags & O_CREAT))
1078 /* Otherwise we just unhash it to be rehashed afresh via
1079 * lookup if necessary
1084 OBD_ALLOC(it, sizeof(*it));
1088 it->it_op = IT_OPEN;
1089 if (open_flags & O_CREAT) {
1090 it->it_op |= IT_CREAT;
1091 lookup_flags |= LOOKUP_CREATE;
1092 sbi = ll_i2sbi(dir);
1093 /* Volatile file is used for HSM restore, so do not use PCC */
1094 if (!filename_is_volatile(dentry->d_name.name,
1095 dentry->d_name.len, NULL)) {
1096 struct pcc_matcher item;
1097 struct pcc_dataset *dataset;
1099 item.pm_uid = from_kuid(&init_user_ns, current_uid());
1100 item.pm_gid = from_kgid(&init_user_ns, current_gid());
1101 item.pm_projid = ll_i2info(dir)->lli_projid;
1102 item.pm_name = &dentry->d_name;
1103 dataset = pcc_dataset_match_get(&sbi->ll_pcc_super,
1105 pca.pca_dataset = dataset;
1108 it->it_create_mode = (mode & S_IALLUGO) | S_IFREG;
1109 it->it_flags = (open_flags & ~O_ACCMODE) | OPEN_FMODE(open_flags);
1110 it->it_flags &= ~MDS_OPEN_FL_INTERNAL;
1112 if (ll_sbi_has_encrypt(ll_i2sbi(dir)) && IS_ENCRYPTED(dir)) {
1113 /* we know that we are going to create a regular file because
1114 * we set S_IFREG bit on it->it_create_mode above
1116 rc = llcrypt_get_encryption_info(dir);
1118 GOTO(out_release, rc);
1119 if (!llcrypt_has_encryption_key(dir))
1120 GOTO(out_release, rc = -ENOKEY);
1125 /* Dentry added to dcache tree in ll_lookup_it */
1126 de = ll_lookup_it(dir, dentry, it, &secctx, &secctxlen, &pca, encrypt,
1127 &encctx, &encctxlen);
1130 else if (de != NULL)
1133 CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_FILE_PAUSE, cfs_fail_val);
1136 if (it_disposition(it, DISP_OPEN_CREATE)) {
1137 /* Dentry instantiated in ll_create_it. */
1138 rc = ll_create_it(dir, dentry, it, secctx, secctxlen,
1139 encrypt, encctx, encctxlen);
1140 security_release_secctx(secctx, secctxlen);
1141 llcrypt_free_ctx(encctx, encctxlen);
1143 /* We dget in ll_splice_alias. */
1149 rc = pcc_inode_create_fini(dentry->d_inode, &pca);
1153 GOTO(out_release, rc);
1156 ll_set_created(opened, file);
1158 /* Open the file with O_CREAT, but the file already
1159 * existed on MDT. This may happend in the case that
1160 * the LOOKUP ibits lock is revoked and the
1161 * corresponding dentry cache is deleted.
1162 * i.e. In the current Lustre, the truncate operation
1163 * will revoke the LOOKUP ibits lock, and the file
1164 * dentry cache will be invalidated. The following open
1165 * with O_CREAT flag will call into ->atomic_open, the
1166 * file was wrongly though as newly created file and
1167 * try to auto cache the file. So after client knows it
1168 * is not a DISP_OPEN_CREATE, it should cleanup the
1169 * already created PCC copy.
1171 pcc_create_attach_cleanup(dir->i_sb, &pca);
1173 if (open_flags & O_CREAT && encrypt &&
1175 rc = ll_set_encflags(dentry->d_inode, encctx,
1177 llcrypt_free_ctx(encctx, encctxlen);
1179 GOTO(out_release, rc);
1183 if (dentry->d_inode && it_disposition(it, DISP_OPEN_OPEN)) {
1185 if (S_ISFIFO(dentry->d_inode->i_mode)) {
1186 /* We cannot call open here as it might
1187 * deadlock. This case is unreachable in
1188 * practice because of OBD_CONNECT_NODEVOH. */
1189 rc = finish_no_open(file, de);
1191 file->private_data = it;
1192 rc = ll_finish_open(file, dentry, opened);
1193 /* We dget in ll_splice_alias. finish_open takes
1194 * care of dget for fd open.
1200 rc = finish_no_open(file, de);
1203 pcc_create_attach_cleanup(dir->i_sb, &pca);
1207 ll_intent_release(it);
1208 OBD_FREE(it, sizeof(*it));
1213 /* We depend on "mode" being set with the proper file type/umask by now */
1214 static struct inode *ll_create_node(struct inode *dir, struct lookup_intent *it)
1216 struct inode *inode = NULL;
1217 struct ptlrpc_request *request = NULL;
1218 struct ll_sb_info *sbi = ll_i2sbi(dir);
1222 LASSERT(it && it->it_disposition);
1224 LASSERT(it_disposition(it, DISP_ENQ_CREATE_REF));
1225 request = it->it_request;
1226 it_clear_disposition(it, DISP_ENQ_CREATE_REF);
1227 rc = ll_prep_inode(&inode, request, dir->i_sb, it);
1229 GOTO(out, inode = ERR_PTR(rc));
1231 /* Pause to allow for a race with concurrent access by fid */
1232 OBD_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_NODE_PAUSE, cfs_fail_val);
1234 /* We asked for a lock on the directory, but were granted a
1235 * lock on the inode. Since we finally have an inode pointer,
1236 * stuff it in the lock. */
1237 CDEBUG(D_DLMTRACE, "setting l_ast_data to inode "DFID"(%p)\n",
1238 PFID(ll_inode2fid(inode)), inode);
1239 ll_set_lock_data(sbi->ll_md_exp, inode, it, NULL);
1242 ptlrpc_req_finished(request);
1247 * By the time this is called, we already have created the directory cache
1248 * entry for the new file, but it is so far negative - it has no inode.
1250 * We defer creating the OBD object(s) until open, to keep the intent and
1251 * non-intent code paths similar, and also because we do not have the MDS
1252 * inode number before calling ll_create_node() (which is needed for LOV),
1253 * so we would need to do yet another RPC to the MDS to store the LOV EA
1254 * data on the MDS. If needed, we would pass the PACKED lmm as data and
1255 * lmm_size in datalen (the MDS still has code which will handle that).
1257 * If the create succeeds, we fill in the inode information
1258 * with d_instantiate().
1260 static int ll_create_it(struct inode *dir, struct dentry *dentry,
1261 struct lookup_intent *it,
1262 void *secctx, __u32 secctxlen, bool encrypt,
1263 void *encctx, __u32 encctxlen)
1265 struct inode *inode;
1270 CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p), intent=%s\n",
1271 dentry, PFID(ll_inode2fid(dir)), dir, LL_IT2STR(it));
1273 rc = it_open_error(DISP_OPEN_CREATE, it);
1277 inode = ll_create_node(dir, it);
1279 RETURN(PTR_ERR(inode));
1281 if ((ll_i2sbi(inode)->ll_flags & LL_SBI_FILE_SECCTX) &&
1283 /* must be done before d_instantiate, because it calls
1284 * security_d_instantiate, which means a getxattr if security
1285 * context is not set yet */
1286 /* no need to protect selinux_inode_setsecurity() by
1287 * inode_lock. Taking it would lead to a client deadlock
1290 rc = security_inode_notifysecctx(inode, secctx, secctxlen);
1295 d_instantiate(dentry, inode);
1298 rc = ll_set_encflags(inode, encctx, encctxlen, true);
1303 if (!(ll_i2sbi(inode)->ll_flags & LL_SBI_FILE_SECCTX)) {
1304 rc = ll_inode_init_security(dentry, inode, dir);
1309 ll_set_lock_data(ll_i2sbi(dir)->ll_md_exp, inode, it, &bits);
1310 if (bits & MDS_INODELOCK_LOOKUP)
1311 d_lustre_revalidate(dentry);
1316 void ll_update_times(struct ptlrpc_request *request, struct inode *inode)
1318 struct mdt_body *body = req_capsule_server_get(&request->rq_pill,
1322 if (body->mbo_valid & OBD_MD_FLMTIME &&
1323 body->mbo_mtime > inode->i_mtime.tv_sec) {
1325 "setting fid " DFID " mtime from %lld to %llu\n",
1326 PFID(ll_inode2fid(inode)),
1327 (s64)inode->i_mtime.tv_sec, body->mbo_mtime);
1328 inode->i_mtime.tv_sec = body->mbo_mtime;
1331 if (body->mbo_valid & OBD_MD_FLCTIME &&
1332 body->mbo_ctime > inode->i_ctime.tv_sec)
1333 inode->i_ctime.tv_sec = body->mbo_ctime;
1336 static int ll_new_node(struct inode *dir, struct dentry *dchild,
1337 const char *tgt, umode_t mode, int rdev, __u32 opc)
1339 struct qstr *name = &dchild->d_name;
1340 struct ptlrpc_request *request = NULL;
1341 struct md_op_data *op_data = NULL;
1342 struct inode *inode = NULL;
1343 struct ll_sb_info *sbi = ll_i2sbi(dir);
1345 bool encrypt = false;
1349 if (unlikely(tgt != NULL))
1350 tgt_len = strlen(tgt) + 1;
1353 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
1354 name->len, 0, opc, NULL);
1355 if (IS_ERR(op_data))
1356 GOTO(err_exit, err = PTR_ERR(op_data));
1358 if (sbi->ll_flags & LL_SBI_FILE_SECCTX) {
1359 err = ll_dentry_init_security(dchild, mode, &dchild->d_name,
1360 &op_data->op_file_secctx_name,
1361 &op_data->op_file_secctx,
1362 &op_data->op_file_secctx_size);
1364 GOTO(err_exit, err);
1367 if (ll_sbi_has_encrypt(sbi) &&
1368 ((IS_ENCRYPTED(dir) &&
1369 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) ||
1370 (unlikely(llcrypt_dummy_context_enabled(dir)) && S_ISDIR(mode)))) {
1371 err = llcrypt_get_encryption_info(dir);
1373 GOTO(err_exit, err);
1374 if (!llcrypt_has_encryption_key(dir))
1375 GOTO(err_exit, err = -ENOKEY);
1380 err = llcrypt_inherit_context(dir, NULL, op_data, false);
1382 GOTO(err_exit, err);
1385 err = md_create(sbi->ll_md_exp, op_data, tgt, tgt_len, mode,
1386 from_kuid(&init_user_ns, current_fsuid()),
1387 from_kgid(&init_user_ns, current_fsgid()),
1388 cfs_curproc_cap_pack(), rdev, &request);
1389 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 14, 58, 0)
1391 * server < 2.12.58 doesn't pack default LMV in intent_getattr reply,
1392 * fetch default LMV here.
1394 if (unlikely(err == -EREMOTE)) {
1395 struct ll_inode_info *lli = ll_i2info(dir);
1396 struct lmv_user_md *lum;
1400 ptlrpc_req_finished(request);
1403 err2 = ll_dir_getstripe(dir, (void **)&lum, &lumsize, &request,
1404 OBD_MD_DEFAULT_MEA);
1405 ll_finish_md_op_data(op_data);
1408 struct lustre_md md = { NULL };
1410 md.body = req_capsule_server_get(&request->rq_pill,
1413 GOTO(err_exit, err = -EPROTO);
1415 OBD_ALLOC_PTR(md.default_lmv);
1416 if (!md.default_lmv)
1417 GOTO(err_exit, err = -ENOMEM);
1419 md.default_lmv->lsm_md_magic = lum->lum_magic;
1420 md.default_lmv->lsm_md_stripe_count =
1421 lum->lum_stripe_count;
1422 md.default_lmv->lsm_md_master_mdt_index =
1423 lum->lum_stripe_offset;
1424 md.default_lmv->lsm_md_hash_type = lum->lum_hash_type;
1426 err = ll_update_inode(dir, &md);
1427 md_free_lustre_md(sbi->ll_md_exp, &md);
1429 GOTO(err_exit, err);
1430 } else if (err2 == -ENODATA && lli->lli_default_lsm_md) {
1432 * If there are no default stripe EA on the MDT, but the
1433 * client has default stripe, then it probably means
1434 * default stripe EA has just been deleted.
1436 down_write(&lli->lli_lsm_sem);
1437 if (lli->lli_default_lsm_md)
1438 OBD_FREE_PTR(lli->lli_default_lsm_md);
1439 lli->lli_default_lsm_md = NULL;
1440 up_write(&lli->lli_lsm_sem);
1442 GOTO(err_exit, err);
1445 ptlrpc_req_finished(request);
1452 GOTO(err_exit, err);
1454 ll_update_times(request, dir);
1456 CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_NEWNODE_PAUSE, cfs_fail_val);
1458 err = ll_prep_inode(&inode, request, dchild->d_sb, NULL);
1460 GOTO(err_exit, err);
1462 if (sbi->ll_flags & LL_SBI_FILE_SECCTX) {
1463 /* must be done before d_instantiate, because it calls
1464 * security_d_instantiate, which means a getxattr if security
1465 * context is not set yet */
1466 /* no need to protect selinux_inode_setsecurity() by
1467 * inode_lock. Taking it would lead to a client deadlock
1470 err = security_inode_notifysecctx(inode,
1471 op_data->op_file_secctx,
1472 op_data->op_file_secctx_size);
1474 GOTO(err_exit, err);
1477 d_instantiate(dchild, inode);
1480 err = ll_set_encflags(inode, op_data->op_file_encctx,
1481 op_data->op_file_encctx_size, true);
1483 GOTO(err_exit, err);
1486 if (!(sbi->ll_flags & LL_SBI_FILE_SECCTX)) {
1487 err = ll_inode_init_security(dchild, inode, dir);
1489 GOTO(err_exit, err);
1494 if (request != NULL)
1495 ptlrpc_req_finished(request);
1497 if (!IS_ERR_OR_NULL(op_data))
1498 ll_finish_md_op_data(op_data);
1503 static int ll_mknod(struct inode *dir, struct dentry *dchild, umode_t mode,
1506 ktime_t kstart = ktime_get();
1510 CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p) mode %o dev %x\n",
1511 dchild, PFID(ll_inode2fid(dir)), dir, mode, rdev);
1513 if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
1514 mode &= ~current_umask();
1516 switch (mode & S_IFMT) {
1525 err = ll_new_node(dir, dchild, NULL, mode, old_encode_dev(rdev),
1536 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKNOD,
1537 ktime_us_delta(ktime_get(), kstart));
1543 * Plain create. Intent create is handled in atomic_open.
1545 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
1546 umode_t mode, bool want_excl)
1548 ktime_t kstart = ktime_get();
1551 CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_FILE_PAUSE, cfs_fail_val);
1554 "VFS Op:name=%pd, dir="DFID"(%p), flags=%u, excl=%d\n",
1555 dentry, PFID(ll_inode2fid(dir)), dir, mode, want_excl);
1557 /* Using mknod(2) to create a regular file is designed to not recognize
1558 * volatile file name, so we use ll_mknod() here. */
1559 rc = ll_mknod(dir, dentry, mode, 0);
1561 CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, unhashed %d\n",
1562 dentry, d_unhashed(dentry));
1565 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_CREATE,
1566 ktime_us_delta(ktime_get(), kstart));
1571 static int ll_symlink(struct inode *dir, struct dentry *dchild,
1572 const char *oldpath)
1574 ktime_t kstart = ktime_get();
1578 CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p), target=%.*s\n",
1579 dchild, PFID(ll_inode2fid(dir)), dir, 3000, oldpath);
1581 err = ll_new_node(dir, dchild, oldpath, S_IFLNK | S_IRWXUGO, 0,
1582 LUSTRE_OPC_SYMLINK);
1585 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_SYMLINK,
1586 ktime_us_delta(ktime_get(), kstart));
1591 static int ll_link(struct dentry *old_dentry, struct inode *dir,
1592 struct dentry *new_dentry)
1594 struct inode *src = old_dentry->d_inode;
1595 struct qstr *name = &new_dentry->d_name;
1596 struct ll_sb_info *sbi = ll_i2sbi(dir);
1597 struct ptlrpc_request *request = NULL;
1598 struct md_op_data *op_data;
1599 ktime_t kstart = ktime_get();
1604 "VFS Op: inode="DFID"(%p), dir="DFID"(%p), target=%pd\n",
1605 PFID(ll_inode2fid(src)), src,
1606 PFID(ll_inode2fid(dir)), dir, new_dentry);
1608 err = llcrypt_prepare_link(old_dentry, dir, new_dentry);
1612 op_data = ll_prep_md_op_data(NULL, src, dir, name->name, name->len,
1613 0, LUSTRE_OPC_ANY, NULL);
1614 if (IS_ERR(op_data))
1615 RETURN(PTR_ERR(op_data));
1617 err = md_link(sbi->ll_md_exp, op_data, &request);
1618 ll_finish_md_op_data(op_data);
1622 ll_update_times(request, dir);
1623 ll_stats_ops_tally(sbi, LPROC_LL_LINK,
1624 ktime_us_delta(ktime_get(), kstart));
1627 ptlrpc_req_finished(request);
1631 static int ll_mkdir(struct inode *dir, struct dentry *dchild, umode_t mode)
1633 ktime_t kstart = ktime_get();
1637 CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p)\n",
1638 dchild, PFID(ll_inode2fid(dir)), dir);
1640 if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
1641 mode &= ~current_umask();
1643 mode = (mode & (S_IRWXUGO|S_ISVTX)) | S_IFDIR;
1645 err = ll_new_node(dir, dchild, NULL, mode, 0, LUSTRE_OPC_MKDIR);
1647 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKDIR,
1648 ktime_us_delta(ktime_get(), kstart));
1653 static int ll_rmdir(struct inode *dir, struct dentry *dchild)
1655 struct qstr *name = &dchild->d_name;
1656 struct ptlrpc_request *request = NULL;
1657 struct md_op_data *op_data;
1658 ktime_t kstart = ktime_get();
1663 CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p)\n",
1664 dchild, PFID(ll_inode2fid(dir)), dir);
1666 if (unlikely(d_mountpoint(dchild)))
1669 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len,
1670 S_IFDIR, LUSTRE_OPC_ANY, NULL);
1671 if (IS_ERR(op_data))
1672 RETURN(PTR_ERR(op_data));
1674 if (dchild->d_inode != NULL)
1675 op_data->op_fid3 = *ll_inode2fid(dchild->d_inode);
1677 op_data->op_fid2 = op_data->op_fid3;
1678 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1679 ll_finish_md_op_data(op_data);
1681 ll_update_times(request, dir);
1683 ptlrpc_req_finished(request);
1685 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR,
1686 ktime_us_delta(ktime_get(), kstart));
1693 int ll_rmdir_entry(struct inode *dir, char *name, int namelen)
1695 struct ptlrpc_request *request = NULL;
1696 struct md_op_data *op_data;
1697 ktime_t kstart = ktime_get();
1701 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1702 namelen, name, PFID(ll_inode2fid(dir)), dir);
1704 op_data = ll_prep_md_op_data(NULL, dir, NULL, name, strlen(name),
1705 S_IFDIR, LUSTRE_OPC_ANY, NULL);
1706 if (IS_ERR(op_data))
1707 RETURN(PTR_ERR(op_data));
1708 op_data->op_cli_flags |= CLI_RM_ENTRY;
1709 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1710 ll_finish_md_op_data(op_data);
1712 ll_update_times(request, dir);
1714 ptlrpc_req_finished(request);
1716 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR,
1717 ktime_us_delta(ktime_get(), kstart));
1721 static int ll_unlink(struct inode *dir, struct dentry *dchild)
1723 struct qstr *name = &dchild->d_name;
1724 struct ptlrpc_request *request = NULL;
1725 struct md_op_data *op_data;
1726 struct mdt_body *body;
1727 ktime_t kstart = ktime_get();
1732 CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p)\n",
1733 dchild, PFID(ll_inode2fid(dir)), dir);
1736 * XXX: unlink bind mountpoint maybe call to here,
1737 * just check it as vfs_unlink does.
1739 if (unlikely(d_mountpoint(dchild)))
1742 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len, 0,
1743 LUSTRE_OPC_ANY, NULL);
1744 if (IS_ERR(op_data))
1745 RETURN(PTR_ERR(op_data));
1747 op_data->op_fid3 = *ll_inode2fid(dchild->d_inode);
1748 /* notify lower layer if inode has dirty pages */
1749 if (S_ISREG(dchild->d_inode->i_mode) &&
1750 ll_i2info(dchild->d_inode)->lli_clob &&
1751 dirty_cnt(dchild->d_inode))
1752 op_data->op_cli_flags |= CLI_DIRTY_DATA;
1753 op_data->op_fid2 = op_data->op_fid3;
1754 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1755 ll_finish_md_op_data(op_data);
1760 * The server puts attributes in on the last unlink, use them to update
1761 * the link count so the inode can be freed immediately.
1763 body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
1764 if (body->mbo_valid & OBD_MD_FLNLINK)
1765 set_nlink(dchild->d_inode, body->mbo_nlink);
1767 ll_update_times(request, dir);
1770 ptlrpc_req_finished(request);
1772 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_UNLINK,
1773 ktime_us_delta(ktime_get(), kstart));
1777 static int ll_rename(struct inode *src, struct dentry *src_dchild,
1778 struct inode *tgt, struct dentry *tgt_dchild
1779 #ifdef HAVE_IOPS_RENAME_WITH_FLAGS
1780 , unsigned int flags
1784 struct qstr *src_name = &src_dchild->d_name;
1785 struct qstr *tgt_name = &tgt_dchild->d_name;
1786 struct ptlrpc_request *request = NULL;
1787 struct ll_sb_info *sbi = ll_i2sbi(src);
1788 struct md_op_data *op_data;
1789 ktime_t kstart = ktime_get();
1793 #ifdef HAVE_IOPS_RENAME_WITH_FLAGS
1799 "VFS Op:oldname=%pd, src_dir="DFID"(%p), newname=%pd, tgt_dir="DFID"(%p)\n",
1800 src_dchild, PFID(ll_inode2fid(src)), src,
1801 tgt_dchild, PFID(ll_inode2fid(tgt)), tgt);
1803 if (unlikely(d_mountpoint(src_dchild) || d_mountpoint(tgt_dchild)))
1806 #ifdef HAVE_IOPS_RENAME_WITH_FLAGS
1807 err = llcrypt_prepare_rename(src, src_dchild, tgt, tgt_dchild, flags);
1809 err = llcrypt_prepare_rename(src, src_dchild, tgt, tgt_dchild, 0);
1814 op_data = ll_prep_md_op_data(NULL, src, tgt, NULL, 0, 0,
1815 LUSTRE_OPC_ANY, NULL);
1816 if (IS_ERR(op_data))
1817 RETURN(PTR_ERR(op_data));
1819 if (src_dchild->d_inode != NULL)
1820 op_data->op_fid3 = *ll_inode2fid(src_dchild->d_inode);
1822 if (tgt_dchild->d_inode != NULL)
1823 op_data->op_fid4 = *ll_inode2fid(tgt_dchild->d_inode);
1825 err = md_rename(sbi->ll_md_exp, op_data,
1826 src_name->name, src_name->len,
1827 tgt_name->name, tgt_name->len, &request);
1828 ll_finish_md_op_data(op_data);
1830 ll_update_times(request, src);
1831 ll_update_times(request, tgt);
1834 ptlrpc_req_finished(request);
1837 d_move(src_dchild, tgt_dchild);
1838 ll_stats_ops_tally(sbi, LPROC_LL_RENAME,
1839 ktime_us_delta(ktime_get(), kstart));
1845 const struct inode_operations ll_dir_inode_operations = {
1847 .atomic_open = ll_atomic_open,
1848 .lookup = ll_lookup_nd,
1849 .create = ll_create_nd,
1850 /* We need all these non-raw things for NFSD, to not patch it. */
1851 .unlink = ll_unlink,
1854 .symlink = ll_symlink,
1856 .rename = ll_rename,
1857 .setattr = ll_setattr,
1858 .getattr = ll_getattr,
1859 .permission = ll_inode_permission,
1860 #ifdef HAVE_IOP_XATTR
1861 .setxattr = ll_setxattr,
1862 .getxattr = ll_getxattr,
1863 .removexattr = ll_removexattr,
1865 .listxattr = ll_listxattr,
1866 .get_acl = ll_get_acl,
1867 #ifdef HAVE_IOP_SET_ACL
1868 .set_acl = ll_set_acl,
1872 const struct inode_operations ll_special_inode_operations = {
1873 .setattr = ll_setattr,
1874 .getattr = ll_getattr,
1875 .permission = ll_inode_permission,
1876 #ifdef HAVE_IOP_XATTR
1877 .setxattr = ll_setxattr,
1878 .getxattr = ll_getxattr,
1879 .removexattr = ll_removexattr,
1881 .listxattr = ll_listxattr,
1882 .get_acl = ll_get_acl,
1883 #ifdef HAVE_IOP_SET_ACL
1884 .set_acl = ll_set_acl,