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 #ifdef HAVE_UIDGID_HEADER
42 # include <linux/uidgid.h>
45 #define DEBUG_SUBSYSTEM S_LLITE
47 #include <obd_support.h>
48 #include <lustre_fid.h>
49 #include <lustre_dlm.h>
50 #include "llite_internal.h"
52 static int ll_create_it(struct inode *dir, struct dentry *dentry,
53 struct lookup_intent *it,
54 void *secctx, __u32 secctxlen);
56 /* called from iget5_locked->find_inode() under inode_lock spinlock */
57 static int ll_test_inode(struct inode *inode, void *opaque)
59 struct ll_inode_info *lli = ll_i2info(inode);
60 struct lustre_md *md = opaque;
62 if (unlikely(!(md->body->mbo_valid & OBD_MD_FLID))) {
63 CERROR("MDS body missing FID\n");
67 if (!lu_fid_eq(&lli->lli_fid, &md->body->mbo_fid1))
73 static int ll_set_inode(struct inode *inode, void *opaque)
75 struct ll_inode_info *lli = ll_i2info(inode);
76 struct mdt_body *body = ((struct lustre_md *)opaque)->body;
78 if (unlikely(!(body->mbo_valid & OBD_MD_FLID))) {
79 CERROR("MDS body missing FID\n");
83 lli->lli_fid = body->mbo_fid1;
84 if (unlikely(!(body->mbo_valid & OBD_MD_FLTYPE))) {
85 CERROR("Can not initialize inode "DFID" without object type: "
87 PFID(&lli->lli_fid), body->mbo_valid);
91 inode->i_mode = (inode->i_mode & ~S_IFMT) | (body->mbo_mode & S_IFMT);
92 if (unlikely(inode->i_mode == 0)) {
93 CERROR("Invalid inode "DFID" type\n", PFID(&lli->lli_fid));
104 * Get an inode by inode number(@hash), which is already instantiated by
105 * the intent lookup).
107 struct inode *ll_iget(struct super_block *sb, ino_t hash,
108 struct lustre_md *md)
116 inode = iget5_locked(sb, hash, ll_test_inode, ll_set_inode, md);
118 RETURN(ERR_PTR(-ENOMEM));
120 if (inode->i_state & I_NEW) {
121 rc = ll_read_inode2(inode, md);
122 if (rc == 0 && S_ISREG(inode->i_mode) &&
123 ll_i2info(inode)->lli_clob == NULL)
124 rc = cl_file_inode_init(inode, md);
127 /* Let's clear directory lsm here, otherwise
128 * make_bad_inode() will reset the inode mode
129 * to regular, then ll_clear_inode will not
130 * be able to clear lsm_md */
131 if (S_ISDIR(inode->i_mode))
132 ll_dir_clear_lsm_md(inode);
133 make_bad_inode(inode);
134 unlock_new_inode(inode);
138 inode_has_no_xattr(inode);
139 unlock_new_inode(inode);
141 } else if (is_bad_inode(inode)) {
143 inode = ERR_PTR(-ESTALE);
144 } else if (!(inode->i_state & (I_FREEING | I_CLEAR))) {
145 rc = ll_update_inode(inode, md);
146 CDEBUG(D_VFSTRACE, "got inode: "DFID"(%p): rc = %d\n",
147 PFID(&md->body->mbo_fid1), inode, rc);
149 if (S_ISDIR(inode->i_mode))
150 ll_dir_clear_lsm_md(inode);
159 static void ll_invalidate_negative_children(struct inode *dir)
161 struct dentry *dentry, *tmp_subdir;
162 DECLARE_LL_D_HLIST_NODE_PTR(p);
164 spin_lock(&dir->i_lock);
165 ll_d_hlist_for_each_entry(dentry, p, &dir->i_dentry) {
166 spin_lock(&dentry->d_lock);
167 if (!list_empty(&dentry->d_subdirs)) {
168 struct dentry *child;
170 list_for_each_entry_safe(child, tmp_subdir,
173 if (child->d_inode == NULL)
174 d_lustre_invalidate(child, 1);
177 spin_unlock(&dentry->d_lock);
179 spin_unlock(&dir->i_lock);
182 int ll_test_inode_by_fid(struct inode *inode, void *opaque)
184 return lu_fid_eq(&ll_i2info(inode)->lli_fid, opaque);
187 static int ll_dom_lock_cancel(struct inode *inode, struct ldlm_lock *lock)
190 struct ll_inode_info *lli = ll_i2info(inode);
195 if (!lli->lli_clob) {
196 /* due to DoM read on open, there may exist pages for Lustre
197 * regular file even though cl_object is not set up yet. */
198 truncate_inode_pages(inode->i_mapping, 0);
202 env = cl_env_get(&refcheck);
204 RETURN(PTR_ERR(env));
206 /* reach MDC layer to flush data under the DoM ldlm lock */
207 rc = cl_object_flush(env, lli->lli_clob, lock);
208 if (rc == -ENODATA) {
209 CDEBUG(D_INODE, "inode "DFID" layout has no DoM stripe\n",
210 PFID(ll_inode2fid(inode)));
211 /* most likely result of layout change, do nothing */
215 cl_env_put(env, &refcheck);
219 static void ll_lock_cancel_bits(struct ldlm_lock *lock, __u64 to_cancel)
221 struct inode *inode = ll_inode_from_resource_lock(lock);
222 struct ll_inode_info *lli;
223 __u64 bits = to_cancel;
229 /* That means the inode is evicted most likely and may cause
230 * the skipping of lock cleanups below, so print the message
233 if (lock->l_resource->lr_lvb_inode)
235 "can't take inode for the lock (%sevicted)\n",
236 lock->l_resource->lr_lvb_inode->i_state &
237 I_FREEING ? "" : "not ");
241 if (!fid_res_name_eq(ll_inode2fid(inode),
242 &lock->l_resource->lr_name)) {
243 LDLM_ERROR(lock, "data mismatch with object "DFID"(%p)",
244 PFID(ll_inode2fid(inode)), inode);
248 if (bits & MDS_INODELOCK_XATTR) {
249 ll_xattr_cache_destroy(inode);
250 bits &= ~MDS_INODELOCK_XATTR;
253 /* For OPEN locks we differentiate between lock modes
254 * LCK_CR, LCK_CW, LCK_PR - bug 22891 */
255 if (bits & MDS_INODELOCK_OPEN)
256 ll_have_md_lock(inode, &bits, lock->l_req_mode);
258 if (bits & MDS_INODELOCK_OPEN) {
261 switch (lock->l_req_mode) {
272 LDLM_ERROR(lock, "bad lock mode for OPEN lock");
276 ll_md_real_close(inode, fmode);
278 bits &= ~MDS_INODELOCK_OPEN;
281 if (bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
282 MDS_INODELOCK_LAYOUT | MDS_INODELOCK_PERM |
284 ll_have_md_lock(inode, &bits, LCK_MINMODE);
286 if (bits & MDS_INODELOCK_DOM) {
287 rc = ll_dom_lock_cancel(inode, lock);
289 CDEBUG(D_INODE, "cannot flush DoM data "
291 PFID(ll_inode2fid(inode)), rc);
294 if (bits & MDS_INODELOCK_LAYOUT) {
295 struct cl_object_conf conf = {
296 .coc_opc = OBJECT_CONF_INVALIDATE,
300 rc = ll_layout_conf(inode, &conf);
302 CDEBUG(D_INODE, "cannot invalidate layout of "
304 PFID(ll_inode2fid(inode)), rc);
307 lli = ll_i2info(inode);
309 if (bits & MDS_INODELOCK_UPDATE)
310 lli->lli_update_atime = 1;
312 if ((bits & MDS_INODELOCK_UPDATE) && S_ISDIR(inode->i_mode)) {
313 CDEBUG(D_INODE, "invalidating inode "DFID" lli = %p, "
314 "pfid = "DFID"\n", PFID(ll_inode2fid(inode)),
315 lli, PFID(&lli->lli_pfid));
316 truncate_inode_pages(inode->i_mapping, 0);
318 if (unlikely(!fid_is_zero(&lli->lli_pfid))) {
319 struct inode *master_inode = NULL;
322 /* This is slave inode, since all of the child dentry
323 * is connected on the master inode, so we have to
324 * invalidate the negative children on master inode */
325 CDEBUG(D_INODE, "Invalidate s"DFID" m"DFID"\n",
326 PFID(ll_inode2fid(inode)), PFID(&lli->lli_pfid));
328 hash = cl_fid_build_ino(&lli->lli_pfid,
329 ll_need_32bit_api(ll_i2sbi(inode)));
331 /* Do not lookup the inode with ilookup5, otherwise
332 * it will cause dead lock,
333 * 1. Client1 send chmod req to the MDT0, then on MDT0,
334 * it enqueues master and all of its slaves lock,
335 * (mdt_attr_set() -> mdt_lock_slaves()), after gets
336 * master and stripe0 lock, it will send the enqueue
337 * req (for stripe1) to MDT1, then MDT1 finds the lock
338 * has been granted to client2. Then MDT1 sends blocking
340 * 2. At the same time, client2 tries to unlink
341 * the striped dir (rm -rf striped_dir), and during
342 * lookup, it will hold the master inode of the striped
343 * directory, whose inode state is NEW, then tries to
344 * revalidate all of its slaves, (ll_prep_inode()->
345 * ll_iget()->ll_read_inode2()-> ll_update_inode().).
346 * And it will be blocked on the server side because
348 * 3. Then the client get the blocking_ast req, cancel
349 * the lock, but being blocked if using ->ilookup5()),
350 * because master inode state is NEW. */
351 master_inode = ilookup5_nowait(inode->i_sb, hash,
352 ll_test_inode_by_fid,
353 (void *)&lli->lli_pfid);
355 ll_invalidate_negative_children(master_inode);
359 ll_invalidate_negative_children(inode);
363 if ((bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM)) &&
364 inode->i_sb->s_root != NULL &&
365 inode != inode->i_sb->s_root->d_inode)
366 ll_invalidate_aliases(inode);
368 if (bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM))
369 forget_all_cached_acls(inode);
375 /* Check if the given lock may be downgraded instead of canceling and
376 * that convert is really needed. */
377 int ll_md_need_convert(struct ldlm_lock *lock)
379 struct ldlm_namespace *ns = ldlm_lock_to_ns(lock);
381 __u64 wanted = lock->l_policy_data.l_inodebits.cancel_bits;
382 __u64 bits = lock->l_policy_data.l_inodebits.bits & ~wanted;
383 enum ldlm_mode mode = LCK_MINMODE;
385 if (!lock->l_conn_export ||
386 !exp_connect_lock_convert(lock->l_conn_export))
389 if (!wanted || !bits || ldlm_is_cancel(lock))
392 /* do not convert locks other than DOM for now */
393 if (!((bits | wanted) & MDS_INODELOCK_DOM))
396 /* We may have already remaining bits in some other lock so
397 * lock convert will leave us just extra lock for the same bit.
398 * Check if client has other lock with the same bits and the same
399 * or lower mode and don't convert if any.
401 switch (lock->l_req_mode) {
415 /* do not convert other modes */
419 /* is lock is too old to be converted? */
420 lock_res_and_lock(lock);
421 if (ktime_after(ktime_get(),
422 ktime_add(lock->l_last_used,
423 ktime_set(ns->ns_dirty_age_limit, 0)))) {
424 unlock_res_and_lock(lock);
427 unlock_res_and_lock(lock);
429 inode = ll_inode_from_resource_lock(lock);
430 ll_have_md_lock(inode, &bits, mode);
435 int ll_md_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
436 void *data, int flag)
438 struct lustre_handle lockh;
439 __u64 bits = lock->l_policy_data.l_inodebits.bits;
445 case LDLM_CB_BLOCKING:
447 __u64 cancel_flags = LCF_ASYNC;
449 if (ll_md_need_convert(lock)) {
450 cancel_flags |= LCF_CONVERT;
451 /* For lock convert some cancel actions may require
452 * this lock with non-dropped canceled bits, e.g. page
453 * flush for DOM lock. So call ll_lock_cancel_bits()
454 * here while canceled bits are still set.
456 bits = lock->l_policy_data.l_inodebits.cancel_bits;
457 if (bits & MDS_INODELOCK_DOM)
458 ll_lock_cancel_bits(lock, MDS_INODELOCK_DOM);
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:
469 /* Nothing to do for non-granted locks */
470 if (!ldlm_is_granted(lock))
473 if (ldlm_is_converting(lock)) {
474 /* this is called on already converted lock, so
475 * ibits has remained bits only and cancel_bits
476 * are bits that were dropped.
477 * Note that DOM lock is handled prior lock convert
478 * and is excluded here.
480 bits = lock->l_policy_data.l_inodebits.cancel_bits &
483 LASSERT(ldlm_is_canceling(lock));
485 ll_lock_cancel_bits(lock, bits);
494 __u32 ll_i2suppgid(struct inode *i)
496 if (in_group_p(i->i_gid))
497 return (__u32)from_kgid(&init_user_ns, i->i_gid);
499 return (__u32) __kgid_val(INVALID_GID);
502 /* Pack the required supplementary groups into the supplied groups array.
503 * If we don't need to use the groups from the target inode(s) then we
504 * instead pack one or more groups from the user's supplementary group
505 * array in case it might be useful. Not needed if doing an MDS-side upcall. */
506 void ll_i2gids(__u32 *suppgids, struct inode *i1, struct inode *i2)
509 LASSERT(suppgids != NULL);
511 suppgids[0] = ll_i2suppgid(i1);
514 suppgids[1] = ll_i2suppgid(i2);
520 * try to reuse three types of dentry:
521 * 1. unhashed alias, this one is unhashed by d_invalidate (but it may be valid
522 * by concurrent .revalidate).
523 * 2. INVALID alias (common case for no valid ldlm lock held, but this flag may
524 * be cleared by others calling d_lustre_revalidate).
525 * 3. DISCONNECTED alias.
527 static struct dentry *ll_find_alias(struct inode *inode, struct dentry *dentry)
529 struct dentry *alias, *discon_alias, *invalid_alias;
530 DECLARE_LL_D_HLIST_NODE_PTR(p);
532 if (ll_d_hlist_empty(&inode->i_dentry))
535 discon_alias = invalid_alias = NULL;
537 spin_lock(&inode->i_lock);
538 ll_d_hlist_for_each_entry(alias, p, &inode->i_dentry) {
539 LASSERT(alias != dentry);
541 spin_lock(&alias->d_lock);
542 if ((alias->d_flags & DCACHE_DISCONNECTED) &&
543 S_ISDIR(inode->i_mode))
544 /* LASSERT(last_discon == NULL); LU-405, bz 20055 */
545 discon_alias = alias;
546 else if (alias->d_parent == dentry->d_parent &&
547 alias->d_name.hash == dentry->d_name.hash &&
548 alias->d_name.len == dentry->d_name.len &&
549 memcmp(alias->d_name.name, dentry->d_name.name,
550 dentry->d_name.len) == 0)
551 invalid_alias = alias;
552 spin_unlock(&alias->d_lock);
557 alias = invalid_alias ?: discon_alias ?: NULL;
559 spin_lock(&alias->d_lock);
561 spin_unlock(&alias->d_lock);
563 spin_unlock(&inode->i_lock);
569 * Similar to d_splice_alias(), but lustre treats invalid alias
570 * similar to DCACHE_DISCONNECTED, and tries to use it anyway.
572 struct dentry *ll_splice_alias(struct inode *inode, struct dentry *de)
578 new = ll_find_alias(inode, de);
588 "Reuse dentry %p inode %p refc %d flags %#x\n",
589 new, new->d_inode, ll_d_count(new), new->d_flags);
597 CDEBUG(D_DENTRY, "Add dentry %p inode %p refc %d flags %#x\n",
598 de, de->d_inode, ll_d_count(de), de->d_flags);
602 static int ll_lookup_it_finish(struct ptlrpc_request *request,
603 struct lookup_intent *it,
604 struct inode *parent, struct dentry **de,
605 void *secctx, __u32 secctxlen)
607 struct inode *inode = NULL;
610 struct dentry *alias;
613 /* NB 1 request reference will be taken away by ll_intent_lock()
615 CDEBUG(D_DENTRY, "it %p it_disposition %x\n", it,
617 if (!it_disposition(it, DISP_LOOKUP_NEG)) {
618 struct req_capsule *pill = &request->rq_pill;
619 struct mdt_body *body = req_capsule_server_get(pill,
622 rc = ll_prep_inode(&inode, request, (*de)->d_sb, it);
626 if (it->it_op & IT_OPEN)
627 ll_dom_finish_open(inode, request, it);
629 ll_set_lock_data(ll_i2sbi(parent)->ll_md_exp, inode, it, &bits);
631 /* We used to query real size from OSTs here, but actually
632 * this is not needed. For stat() calls size would be updated
633 * from subsequent do_revalidate()->ll_inode_revalidate_it() in
635 * vfs_getattr_it->ll_getattr()->ll_inode_revalidate_it() in 2.6
636 * Everybody else who needs correct file size would call
637 * ll_glimpse_size or some equivalent themselves anyway.
641 /* If security context was returned by MDT, put it in
642 * inode now to save an extra getxattr from security hooks,
643 * and avoid deadlock.
645 if (body->mbo_valid & OBD_MD_SECCTX) {
646 secctx = req_capsule_server_get(pill, &RMF_FILE_SECCTX);
647 secctxlen = req_capsule_get_size(pill,
652 CDEBUG(D_SEC, "server returned security context"
654 PFID(ll_inode2fid(inode)));
657 if (secctx != NULL && secctxlen != 0) {
659 rc = security_inode_notifysecctx(inode, secctx,
663 CWARN("cannot set security context for "
665 PFID(ll_inode2fid(inode)), rc);
669 /* Only hash *de if it is unhashed (new dentry).
670 * Atoimc_open may passin hashed dentries for open.
672 alias = ll_splice_alias(inode, *de);
674 GOTO(out, rc = PTR_ERR(alias));
678 if (!it_disposition(it, DISP_LOOKUP_NEG)) {
679 /* we have lookup look - unhide dentry */
680 if (bits & MDS_INODELOCK_LOOKUP)
681 d_lustre_revalidate(*de);
682 } else if (!it_disposition(it, DISP_OPEN_CREATE)) {
684 * If file was created on the server, the dentry is revalidated
685 * in ll_create_it if the lock allows for it.
687 /* Check that parent has UPDATE lock. */
688 struct lookup_intent parent_it = {
690 .it_lock_handle = 0 };
691 struct lu_fid fid = ll_i2info(parent)->lli_fid;
693 /* If it is striped directory, get the real stripe parent */
694 if (unlikely(ll_dir_striped(parent))) {
695 rc = md_get_fid_from_lsm(ll_i2mdexp(parent),
696 ll_i2info(parent)->lli_lsm_md,
698 (*de)->d_name.len, &fid);
703 if (md_revalidate_lock(ll_i2mdexp(parent), &parent_it, &fid,
705 d_lustre_revalidate(*de);
706 ll_intent_release(&parent_it);
713 if (rc != 0 && it->it_op & IT_OPEN)
714 ll_open_cleanup((*de)->d_sb, request);
719 static struct dentry *ll_lookup_it(struct inode *parent, struct dentry *dentry,
720 struct lookup_intent *it,
721 void **secctx, __u32 *secctxlen,
722 struct pcc_create_attach *pca)
724 struct lookup_intent lookup_it = { .it_op = IT_LOOKUP };
725 struct dentry *save = dentry, *retval;
726 struct ptlrpc_request *req = NULL;
727 struct md_op_data *op_data = NULL;
728 struct lov_user_md *lum = NULL;
731 char secctx_name[XATTR_NAME_MAX + 1];
735 if (dentry->d_name.len > ll_i2sbi(parent)->ll_namelen)
736 RETURN(ERR_PTR(-ENAMETOOLONG));
738 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), intent=%s\n",
739 dentry->d_name.len, dentry->d_name.name,
740 PFID(ll_inode2fid(parent)), parent, LL_IT2STR(it));
742 if (d_mountpoint(dentry))
743 CERROR("Tell Peter, lookup on mtpt, it %s\n", LL_IT2STR(it));
745 if (it == NULL || it->it_op == IT_GETXATTR)
748 if (it->it_op == IT_GETATTR && dentry_may_statahead(parent, dentry)) {
749 rc = ll_statahead(parent, &dentry, 0);
751 RETURN(dentry == save ? NULL : dentry);
754 if (it->it_op & IT_OPEN && it->it_flags & FMODE_WRITE &&
755 dentry->d_sb->s_flags & SB_RDONLY)
756 RETURN(ERR_PTR(-EROFS));
758 if (it->it_op & IT_CREAT)
759 opc = LUSTRE_OPC_CREATE;
761 opc = LUSTRE_OPC_ANY;
763 op_data = ll_prep_md_op_data(NULL, parent, NULL, dentry->d_name.name,
764 dentry->d_name.len, 0, opc, NULL);
766 GOTO(out, retval = ERR_CAST(op_data));
768 /* enforce umask if acl disabled or MDS doesn't support umask */
769 if (!IS_POSIXACL(parent) || !exp_connect_umask(ll_i2mdexp(parent)))
770 it->it_create_mode &= ~current_umask();
772 if (it->it_op & IT_CREAT &&
773 ll_i2sbi(parent)->ll_flags & LL_SBI_FILE_SECCTX) {
774 rc = ll_dentry_init_security(dentry, it->it_create_mode,
776 &op_data->op_file_secctx_name,
777 &op_data->op_file_secctx,
778 &op_data->op_file_secctx_size);
780 GOTO(out, retval = ERR_PTR(rc));
782 *secctx = op_data->op_file_secctx;
783 if (secctxlen != NULL)
784 *secctxlen = op_data->op_file_secctx_size;
788 if (secctxlen != NULL)
792 /* ask for security context upon intent */
793 if (it->it_op & (IT_LOOKUP | IT_GETATTR | IT_OPEN)) {
794 /* get name of security xattr to request to server */
795 rc = ll_listsecurity(parent, secctx_name,
796 sizeof(secctx_name));
798 CDEBUG(D_SEC, "cannot get security xattr name for "
800 PFID(ll_inode2fid(parent)), rc);
802 op_data->op_file_secctx_name = secctx_name;
803 op_data->op_file_secctx_name_size = rc;
804 CDEBUG(D_SEC, "'%.*s' is security xattr for "DFID"\n",
805 rc, secctx_name, PFID(ll_inode2fid(parent)));
809 if (pca && pca->pca_dataset) {
810 struct pcc_dataset *dataset = pca->pca_dataset;
814 GOTO(out, retval = ERR_PTR(-ENOMEM));
816 lum->lmm_magic = LOV_USER_MAGIC_V1;
817 lum->lmm_pattern = LOV_PATTERN_F_RELEASED | LOV_PATTERN_RAID0;
818 op_data->op_data = lum;
819 op_data->op_data_size = sizeof(*lum);
820 op_data->op_archive_id = dataset->pccd_rwid;
822 rc = obd_fid_alloc(NULL, ll_i2mdexp(parent), &op_data->op_fid2,
825 GOTO(out, retval = ERR_PTR(rc));
827 rc = pcc_inode_create(parent->i_sb, dataset, &op_data->op_fid2,
830 GOTO(out, retval = ERR_PTR(rc));
832 it->it_flags |= MDS_OPEN_PCC;
835 rc = md_intent_lock(ll_i2mdexp(parent), op_data, it, &req,
836 &ll_md_blocking_ast, 0);
837 /* If the MDS allows the client to chgrp (CFS_SETGRP_PERM), but the
838 * client does not know which suppgid should be sent to the MDS, or
839 * some other(s) changed the target file's GID after this RPC sent
840 * to the MDS with the suppgid as the original GID, then we should
841 * try again with right suppgid. */
842 if (rc == -EACCES && it->it_op & IT_OPEN &&
843 it_disposition(it, DISP_OPEN_DENY)) {
844 struct mdt_body *body;
846 LASSERT(req != NULL);
848 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
849 if (op_data->op_suppgids[0] == body->mbo_gid ||
850 op_data->op_suppgids[1] == body->mbo_gid ||
851 !in_group_p(make_kgid(&init_user_ns, body->mbo_gid)))
852 GOTO(out, retval = ERR_PTR(-EACCES));
854 fid_zero(&op_data->op_fid2);
855 op_data->op_suppgids[1] = body->mbo_gid;
856 ptlrpc_req_finished(req);
858 ll_intent_release(it);
859 rc = md_intent_lock(ll_i2mdexp(parent), op_data, it, &req,
860 &ll_md_blocking_ast, 0);
864 GOTO(out, retval = ERR_PTR(rc));
866 /* dir layout may change */
867 ll_unlock_md_op_lsm(op_data);
868 rc = ll_lookup_it_finish(req, it, parent, &dentry,
869 secctx != NULL ? *secctx : NULL,
870 secctxlen != NULL ? *secctxlen : 0);
872 ll_intent_release(it);
873 GOTO(out, retval = ERR_PTR(rc));
876 if ((it->it_op & IT_OPEN) && dentry->d_inode &&
877 !S_ISREG(dentry->d_inode->i_mode) &&
878 !S_ISDIR(dentry->d_inode->i_mode)) {
879 ll_release_openhandle(dentry, it);
881 ll_lookup_finish_locks(it, dentry);
883 GOTO(out, retval = (dentry == save) ? NULL : dentry);
886 if (op_data != NULL && !IS_ERR(op_data)) {
887 if (secctx != NULL && secctxlen != NULL) {
888 /* caller needs sec ctx info, so reset it in op_data to
889 * prevent it from being freed */
890 op_data->op_file_secctx = NULL;
891 op_data->op_file_secctx_size = 0;
893 ll_finish_md_op_data(op_data);
899 ptlrpc_req_finished(req);
903 #ifdef HAVE_IOP_ATOMIC_OPEN
904 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
907 struct lookup_intent *itp, it = { .it_op = IT_GETATTR };
910 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), flags=%u\n",
911 dentry->d_name.len, dentry->d_name.name,
912 PFID(ll_inode2fid(parent)), parent, flags);
915 * Optimize away (CREATE && !OPEN). Let .create handle the race.
916 * but only if we have write permissions there, otherwise we need
917 * to proceed with lookup. LU-4185
919 if ((flags & LOOKUP_CREATE) && !(flags & LOOKUP_OPEN) &&
920 (inode_permission(parent, MAY_WRITE | MAY_EXEC) == 0))
923 if (flags & (LOOKUP_PARENT|LOOKUP_OPEN|LOOKUP_CREATE))
927 de = ll_lookup_it(parent, dentry, itp, NULL, NULL, NULL);
930 ll_intent_release(itp);
935 #ifdef FMODE_CREATED /* added in Linux v4.18-rc1-20-g73a09dd */
936 # define ll_is_opened(o, f) ((f)->f_mode & FMODE_OPENED)
937 # define ll_finish_open(f, d, o) finish_open((f), (d), NULL)
939 # define ll_set_created(o, f) \
941 (f)->f_mode |= FMODE_CREATED; \
945 # define ll_is_opened(o, f) (*(o))
946 # define ll_finish_open(f, d, o) finish_open((f), (d), NULL, (o))
947 # define ll_last_arg , int *opened
948 # define ll_set_created(o, f) \
950 *(o) |= FILE_CREATED; \
956 * For cached negative dentry and new dentry, handle lookup/create/open
959 static int ll_atomic_open(struct inode *dir, struct dentry *dentry,
960 struct file *file, unsigned open_flags,
961 umode_t mode ll_last_arg)
963 struct lookup_intent *it;
965 long long lookup_flags = LOOKUP_OPEN;
968 struct ll_sb_info *sbi;
969 struct pcc_create_attach pca = { NULL, NULL };
973 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), file %p,"
974 "open_flags %x, mode %x opened %d\n",
975 dentry->d_name.len, dentry->d_name.name,
976 PFID(ll_inode2fid(dir)), dir, file, open_flags, mode,
977 ll_is_opened(opened, file));
979 /* Only negative dentries enter here */
980 LASSERT(dentry->d_inode == NULL);
982 if (!d_unhashed(dentry)) {
983 /* A valid negative dentry that just passed revalidation,
984 * there's little point to try and open it server-side,
985 * even though there's a minuscule chance it might succeed.
986 * Either way it's a valid race to just return -ENOENT here.
988 if (!(open_flags & O_CREAT))
991 /* Otherwise we just unhash it to be rehashed afresh via
992 * lookup if necessary
997 OBD_ALLOC(it, sizeof(*it));
1001 it->it_op = IT_OPEN;
1002 if (open_flags & O_CREAT) {
1003 it->it_op |= IT_CREAT;
1004 lookup_flags |= LOOKUP_CREATE;
1005 sbi = ll_i2sbi(dir);
1006 /* Volatile file is used for HSM restore, so do not use PCC */
1007 if (!filename_is_volatile(dentry->d_name.name,
1008 dentry->d_name.len, NULL)) {
1009 struct pcc_matcher item;
1010 struct pcc_dataset *dataset;
1012 item.pm_uid = from_kuid(&init_user_ns, current_uid());
1013 item.pm_gid = from_kgid(&init_user_ns, current_gid());
1014 item.pm_projid = ll_i2info(dir)->lli_projid;
1015 item.pm_name = &dentry->d_name;
1016 dataset = pcc_dataset_match_get(&sbi->ll_pcc_super,
1018 pca.pca_dataset = dataset;
1021 it->it_create_mode = (mode & S_IALLUGO) | S_IFREG;
1022 it->it_flags = (open_flags & ~O_ACCMODE) | OPEN_FMODE(open_flags);
1023 it->it_flags &= ~MDS_OPEN_FL_INTERNAL;
1025 /* Dentry added to dcache tree in ll_lookup_it */
1026 de = ll_lookup_it(dir, dentry, it, &secctx, &secctxlen, &pca);
1029 else if (de != NULL)
1032 CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_FILE_PAUSE, cfs_fail_val);
1035 if (it_disposition(it, DISP_OPEN_CREATE)) {
1036 /* Dentry instantiated in ll_create_it. */
1037 rc = ll_create_it(dir, dentry, it, secctx, secctxlen);
1038 security_release_secctx(secctx, secctxlen);
1040 /* We dget in ll_splice_alias. */
1046 rc = pcc_inode_create_fini(dentry->d_inode, &pca);
1050 GOTO(out_release, rc);
1053 ll_set_created(opened, file);
1055 /* Open the file with O_CREAT, but the file already
1056 * existed on MDT. This may happend in the case that
1057 * the LOOKUP ibits lock is revoked and the
1058 * corresponding dentry cache is deleted.
1059 * i.e. In the current Lustre, the truncate operation
1060 * will revoke the LOOKUP ibits lock, and the file
1061 * dentry cache will be invalidated. The following open
1062 * with O_CREAT flag will call into ->atomic_open, the
1063 * file was wrongly though as newly created file and
1064 * try to auto cache the file. So after client knows it
1065 * is not a DISP_OPEN_CREATE, it should cleanup the
1066 * already created PCC copy.
1068 pcc_create_attach_cleanup(dir->i_sb, &pca);
1071 if (dentry->d_inode && it_disposition(it, DISP_OPEN_OPEN)) {
1073 if (S_ISFIFO(dentry->d_inode->i_mode)) {
1074 /* We cannot call open here as it might
1075 * deadlock. This case is unreachable in
1076 * practice because of OBD_CONNECT_NODEVOH. */
1077 rc = finish_no_open(file, de);
1079 file->private_data = it;
1080 rc = ll_finish_open(file, dentry, opened);
1081 /* We dget in ll_splice_alias. finish_open takes
1082 * care of dget for fd open.
1088 rc = finish_no_open(file, de);
1091 pcc_create_attach_cleanup(dir->i_sb, &pca);
1095 ll_intent_release(it);
1096 OBD_FREE(it, sizeof(*it));
1101 #else /* !HAVE_IOP_ATOMIC_OPEN */
1102 static struct lookup_intent *
1103 ll_convert_intent(struct open_intent *oit, int lookup_flags, bool is_readonly)
1105 struct lookup_intent *it;
1109 return ERR_PTR(-ENOMEM);
1111 if (lookup_flags & LOOKUP_OPEN) {
1112 it->it_op = IT_OPEN;
1113 /* Avoid file creation for ro bind mount point(is_readonly) */
1114 if ((lookup_flags & LOOKUP_CREATE) && !is_readonly)
1115 it->it_op |= IT_CREAT;
1116 it->it_create_mode = (oit->create_mode & S_IALLUGO) | S_IFREG;
1117 it->it_flags = ll_namei_to_lookup_intent_flag(oit->flags &
1118 ~(is_readonly ? O_CREAT : 0));
1119 it->it_flags &= ~MDS_OPEN_FL_INTERNAL;
1121 it->it_op = IT_GETATTR;
1127 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
1128 struct nameidata *nd)
1133 if (nd && !(nd->flags & (LOOKUP_CONTINUE|LOOKUP_PARENT))) {
1134 struct lookup_intent *it;
1136 if (ll_d2d(dentry) && ll_d2d(dentry)->lld_it) {
1137 it = ll_d2d(dentry)->lld_it;
1138 ll_d2d(dentry)->lld_it = NULL;
1141 * Optimize away (CREATE && !OPEN). Let .create handle
1142 * the race. But only if we have write permissions
1143 * there, otherwise we need to proceed with lookup.
1146 if ((nd->flags & LOOKUP_CREATE) &&
1147 !(nd->flags & LOOKUP_OPEN) &&
1148 (inode_permission(parent,
1149 MAY_WRITE | MAY_EXEC) == 0))
1152 it = ll_convert_intent(&nd->intent.open, nd->flags,
1153 (nd->path.mnt->mnt_flags & MNT_READONLY) ||
1154 (nd->path.mnt->mnt_sb->s_flags & SB_RDONLY));
1156 RETURN((struct dentry *)it);
1159 de = ll_lookup_it(parent, dentry, it, NULL, NULL, NULL);
1162 if ((nd->flags & LOOKUP_OPEN) && !IS_ERR(dentry)) { /* Open */
1163 if (dentry->d_inode &&
1164 it_disposition(it, DISP_OPEN_OPEN)) { /* nocreate */
1165 if (S_ISFIFO(dentry->d_inode->i_mode)) {
1166 /* We cannot call open here as it might
1167 * deadlock. This case is unreachable in
1168 * practice because of
1169 * OBD_CONNECT_NODEVOH. */
1173 nd->intent.open.file->private_data = it;
1174 filp = lookup_instantiate_filp(nd,
1180 de = (struct dentry *)filp;
1183 } else if (it_disposition(it, DISP_OPEN_CREATE)) {
1184 /* XXX This can only reliably work on assumption
1185 * that there are NO hashed negative dentries.*/
1186 ll_d2d(dentry)->lld_it = it;
1187 it = NULL; /* Will be freed in ll_create_nd */
1188 /* We absolutely depend on ll_create_nd to be
1189 * called to not leak this intent and possible
1190 * data attached to it */
1195 ll_intent_release(it);
1196 OBD_FREE(it, sizeof(*it));
1199 de = ll_lookup_it(parent, dentry, NULL, NULL, NULL, NULL);
1204 #endif /* HAVE_IOP_ATOMIC_OPEN */
1206 /* We depend on "mode" being set with the proper file type/umask by now */
1207 static struct inode *ll_create_node(struct inode *dir, struct lookup_intent *it)
1209 struct inode *inode = NULL;
1210 struct ptlrpc_request *request = NULL;
1211 struct ll_sb_info *sbi = ll_i2sbi(dir);
1215 LASSERT(it && it->it_disposition);
1217 LASSERT(it_disposition(it, DISP_ENQ_CREATE_REF));
1218 request = it->it_request;
1219 it_clear_disposition(it, DISP_ENQ_CREATE_REF);
1220 rc = ll_prep_inode(&inode, request, dir->i_sb, it);
1222 GOTO(out, inode = ERR_PTR(rc));
1224 /* Pause to allow for a race with concurrent access by fid */
1225 OBD_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_NODE_PAUSE, cfs_fail_val);
1227 /* We asked for a lock on the directory, but were granted a
1228 * lock on the inode. Since we finally have an inode pointer,
1229 * stuff it in the lock. */
1230 CDEBUG(D_DLMTRACE, "setting l_ast_data to inode "DFID"(%p)\n",
1231 PFID(ll_inode2fid(inode)), inode);
1232 ll_set_lock_data(sbi->ll_md_exp, inode, it, NULL);
1235 ptlrpc_req_finished(request);
1240 * By the time this is called, we already have created the directory cache
1241 * entry for the new file, but it is so far negative - it has no inode.
1243 * We defer creating the OBD object(s) until open, to keep the intent and
1244 * non-intent code paths similar, and also because we do not have the MDS
1245 * inode number before calling ll_create_node() (which is needed for LOV),
1246 * so we would need to do yet another RPC to the MDS to store the LOV EA
1247 * data on the MDS. If needed, we would pass the PACKED lmm as data and
1248 * lmm_size in datalen (the MDS still has code which will handle that).
1250 * If the create succeeds, we fill in the inode information
1251 * with d_instantiate().
1253 static int ll_create_it(struct inode *dir, struct dentry *dentry,
1254 struct lookup_intent *it,
1255 void *secctx, __u32 secctxlen)
1257 struct inode *inode;
1262 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), intent=%s\n",
1263 dentry->d_name.len, dentry->d_name.name,
1264 PFID(ll_inode2fid(dir)), dir, LL_IT2STR(it));
1266 rc = it_open_error(DISP_OPEN_CREATE, it);
1270 inode = ll_create_node(dir, it);
1272 RETURN(PTR_ERR(inode));
1274 if ((ll_i2sbi(inode)->ll_flags & LL_SBI_FILE_SECCTX) &&
1277 /* must be done before d_instantiate, because it calls
1278 * security_d_instantiate, which means a getxattr if security
1279 * context is not set yet */
1280 rc = security_inode_notifysecctx(inode, secctx, secctxlen);
1281 inode_unlock(inode);
1286 d_instantiate(dentry, inode);
1288 if (!(ll_i2sbi(inode)->ll_flags & LL_SBI_FILE_SECCTX)) {
1289 rc = ll_inode_init_security(dentry, inode, dir);
1294 ll_set_lock_data(ll_i2sbi(dir)->ll_md_exp, inode, it, &bits);
1295 if (bits & MDS_INODELOCK_LOOKUP)
1296 d_lustre_revalidate(dentry);
1301 void ll_update_times(struct ptlrpc_request *request, struct inode *inode)
1303 struct mdt_body *body = req_capsule_server_get(&request->rq_pill,
1307 if (body->mbo_valid & OBD_MD_FLMTIME &&
1308 body->mbo_mtime > inode->i_mtime.tv_sec) {
1310 "setting fid " DFID " mtime from %lld to %llu\n",
1311 PFID(ll_inode2fid(inode)),
1312 (s64)inode->i_mtime.tv_sec, body->mbo_mtime);
1313 inode->i_mtime.tv_sec = body->mbo_mtime;
1316 if (body->mbo_valid & OBD_MD_FLCTIME &&
1317 body->mbo_ctime > inode->i_ctime.tv_sec)
1318 inode->i_ctime.tv_sec = body->mbo_ctime;
1321 static int ll_new_node(struct inode *dir, struct dentry *dchild,
1322 const char *tgt, umode_t mode, int rdev, __u32 opc)
1324 struct qstr *name = &dchild->d_name;
1325 struct ptlrpc_request *request = NULL;
1326 struct md_op_data *op_data;
1327 struct inode *inode = NULL;
1328 struct ll_sb_info *sbi = ll_i2sbi(dir);
1333 if (unlikely(tgt != NULL))
1334 tgt_len = strlen(tgt) + 1;
1337 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
1338 name->len, 0, opc, NULL);
1339 if (IS_ERR(op_data))
1340 GOTO(err_exit, err = PTR_ERR(op_data));
1342 if (sbi->ll_flags & LL_SBI_FILE_SECCTX) {
1343 err = ll_dentry_init_security(dchild, mode, &dchild->d_name,
1344 &op_data->op_file_secctx_name,
1345 &op_data->op_file_secctx,
1346 &op_data->op_file_secctx_size);
1348 GOTO(err_exit, err);
1351 err = md_create(sbi->ll_md_exp, op_data, tgt, tgt_len, mode,
1352 from_kuid(&init_user_ns, current_fsuid()),
1353 from_kgid(&init_user_ns, current_fsgid()),
1354 cfs_curproc_cap_pack(), rdev, &request);
1355 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 14, 58, 0)
1357 * server < 2.12.58 doesn't pack default LMV in intent_getattr reply,
1358 * fetch default LMV here.
1360 if (unlikely(err == -EREMOTE)) {
1361 struct ll_inode_info *lli = ll_i2info(dir);
1362 struct lmv_user_md *lum;
1366 ptlrpc_req_finished(request);
1369 err2 = ll_dir_getstripe(dir, (void **)&lum, &lumsize, &request,
1370 OBD_MD_DEFAULT_MEA);
1371 ll_finish_md_op_data(op_data);
1374 struct lustre_md md = { NULL };
1376 md.body = req_capsule_server_get(&request->rq_pill,
1379 GOTO(err_exit, err = -EPROTO);
1381 OBD_ALLOC_PTR(md.default_lmv);
1382 if (!md.default_lmv)
1383 GOTO(err_exit, err = -ENOMEM);
1385 md.default_lmv->lsm_md_magic = lum->lum_magic;
1386 md.default_lmv->lsm_md_stripe_count =
1387 lum->lum_stripe_count;
1388 md.default_lmv->lsm_md_master_mdt_index =
1389 lum->lum_stripe_offset;
1390 md.default_lmv->lsm_md_hash_type = lum->lum_hash_type;
1392 err = ll_update_inode(dir, &md);
1393 md_free_lustre_md(sbi->ll_md_exp, &md);
1395 GOTO(err_exit, err);
1396 } else if (err2 == -ENODATA && lli->lli_default_lsm_md) {
1398 * If there are no default stripe EA on the MDT, but the
1399 * client has default stripe, then it probably means
1400 * default stripe EA has just been deleted.
1402 down_write(&lli->lli_lsm_sem);
1403 if (lli->lli_default_lsm_md)
1404 OBD_FREE_PTR(lli->lli_default_lsm_md);
1405 lli->lli_default_lsm_md = NULL;
1406 up_write(&lli->lli_lsm_sem);
1408 GOTO(err_exit, err);
1411 ptlrpc_req_finished(request);
1418 GOTO(err_exit, err);
1420 ll_update_times(request, dir);
1422 CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_NEWNODE_PAUSE, cfs_fail_val);
1424 err = ll_prep_inode(&inode, request, dchild->d_sb, NULL);
1426 GOTO(err_exit, err);
1428 if (sbi->ll_flags & LL_SBI_FILE_SECCTX) {
1430 /* must be done before d_instantiate, because it calls
1431 * security_d_instantiate, which means a getxattr if security
1432 * context is not set yet */
1433 err = security_inode_notifysecctx(inode,
1434 op_data->op_file_secctx,
1435 op_data->op_file_secctx_size);
1436 inode_unlock(inode);
1438 GOTO(err_exit, err);
1441 d_instantiate(dchild, inode);
1443 if (!(sbi->ll_flags & LL_SBI_FILE_SECCTX)) {
1444 err = ll_inode_init_security(dchild, inode, dir);
1446 GOTO(err_exit, err);
1451 if (request != NULL)
1452 ptlrpc_req_finished(request);
1454 if (!IS_ERR_OR_NULL(op_data))
1455 ll_finish_md_op_data(op_data);
1460 static int ll_mknod(struct inode *dir, struct dentry *dchild, ll_umode_t mode,
1463 struct qstr *name = &dchild->d_name;
1464 ktime_t kstart = ktime_get();
1468 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p) mode %o dev %x\n",
1469 name->len, name->name, PFID(ll_inode2fid(dir)), dir,
1472 if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
1473 mode &= ~current_umask();
1475 switch (mode & S_IFMT) {
1484 err = ll_new_node(dir, dchild, NULL, mode, old_encode_dev(rdev),
1495 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKNOD,
1496 ktime_us_delta(ktime_get(), kstart));
1501 #ifdef HAVE_IOP_ATOMIC_OPEN
1503 * Plain create. Intent create is handled in atomic_open.
1505 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
1506 umode_t mode, bool want_excl)
1508 ktime_t kstart = ktime_get();
1511 CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_FILE_PAUSE, cfs_fail_val);
1513 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), "
1514 "flags=%u, excl=%d\n", dentry->d_name.len,
1515 dentry->d_name.name, PFID(ll_inode2fid(dir)),
1516 dir, mode, want_excl);
1518 /* Using mknod(2) to create a regular file is designed to not recognize
1519 * volatile file name, so we use ll_mknod() here. */
1520 rc = ll_mknod(dir, dentry, mode, 0);
1522 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, unhashed %d\n",
1523 dentry->d_name.len, dentry->d_name.name, d_unhashed(dentry));
1526 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_CREATE,
1527 ktime_us_delta(ktime_get(), kstart));
1531 #else /* !HAVE_IOP_ATOMIC_OPEN */
1532 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
1533 ll_umode_t mode, struct nameidata *nd)
1535 struct ll_dentry_data *lld = ll_d2d(dentry);
1536 struct lookup_intent *it = NULL;
1537 ktime_t kstart = ktime_get();
1540 CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_FILE_PAUSE, cfs_fail_val);
1546 /* LU-8559: use LUSTRE_OPC_CREATE for non atomic open case
1547 * so that volatile file name is recoginized.
1548 * Mknod(2), however, is designed to not recognize volatile
1549 * file name to avoid inode leak under orphan directory until
1551 return ll_new_node(dir, dentry, NULL, mode, 0,
1557 /* Was there an error? Propagate it! */
1558 if (it->it_status) {
1563 rc = ll_create_it(dir, dentry, it, NULL, 0);
1564 if (nd && (nd->flags & LOOKUP_OPEN) && dentry->d_inode) { /* Open */
1567 nd->intent.open.file->private_data = it;
1568 filp = lookup_instantiate_filp(nd, dentry, NULL);
1574 ll_intent_release(it);
1575 OBD_FREE(it, sizeof(*it));
1578 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_CREATE,
1579 ktime_us_delta(ktime_get(), kstart));
1583 #endif /* HAVE_IOP_ATOMIC_OPEN */
1585 static int ll_symlink(struct inode *dir, struct dentry *dchild,
1586 const char *oldpath)
1588 struct qstr *name = &dchild->d_name;
1589 ktime_t kstart = ktime_get();
1593 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), target=%.*s\n",
1594 name->len, name->name, PFID(ll_inode2fid(dir)),
1595 dir, 3000, oldpath);
1597 err = ll_new_node(dir, dchild, oldpath, S_IFLNK | S_IRWXUGO, 0,
1598 LUSTRE_OPC_SYMLINK);
1601 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_SYMLINK,
1602 ktime_us_delta(ktime_get(), kstart));
1607 static int ll_link(struct dentry *old_dentry, struct inode *dir,
1608 struct dentry *new_dentry)
1610 struct inode *src = old_dentry->d_inode;
1611 struct qstr *name = &new_dentry->d_name;
1612 struct ll_sb_info *sbi = ll_i2sbi(dir);
1613 struct ptlrpc_request *request = NULL;
1614 struct md_op_data *op_data;
1615 ktime_t kstart = ktime_get();
1619 CDEBUG(D_VFSTRACE, "VFS Op: inode="DFID"(%p), dir="DFID"(%p), "
1620 "target=%.*s\n", PFID(ll_inode2fid(src)), src,
1621 PFID(ll_inode2fid(dir)), dir, name->len, name->name);
1623 op_data = ll_prep_md_op_data(NULL, src, dir, name->name, name->len,
1624 0, LUSTRE_OPC_ANY, NULL);
1625 if (IS_ERR(op_data))
1626 RETURN(PTR_ERR(op_data));
1628 err = md_link(sbi->ll_md_exp, op_data, &request);
1629 ll_finish_md_op_data(op_data);
1633 ll_update_times(request, dir);
1634 ll_stats_ops_tally(sbi, LPROC_LL_LINK,
1635 ktime_us_delta(ktime_get(), kstart));
1638 ptlrpc_req_finished(request);
1642 static int ll_mkdir(struct inode *dir, struct dentry *dchild, ll_umode_t mode)
1644 struct qstr *name = &dchild->d_name;
1645 ktime_t kstart = ktime_get();
1649 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1650 name->len, name->name, PFID(ll_inode2fid(dir)), dir);
1652 if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
1653 mode &= ~current_umask();
1655 mode = (mode & (S_IRWXUGO|S_ISVTX)) | S_IFDIR;
1657 err = ll_new_node(dir, dchild, NULL, mode, 0, LUSTRE_OPC_MKDIR);
1659 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKDIR,
1660 ktime_us_delta(ktime_get(), kstart));
1665 static int ll_rmdir(struct inode *dir, struct dentry *dchild)
1667 struct qstr *name = &dchild->d_name;
1668 struct ptlrpc_request *request = NULL;
1669 struct md_op_data *op_data;
1670 ktime_t kstart = ktime_get();
1675 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1676 name->len, name->name, PFID(ll_inode2fid(dir)), dir);
1678 if (unlikely(d_mountpoint(dchild)))
1681 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len,
1682 S_IFDIR, LUSTRE_OPC_ANY, NULL);
1683 if (IS_ERR(op_data))
1684 RETURN(PTR_ERR(op_data));
1686 if (dchild->d_inode != NULL)
1687 op_data->op_fid3 = *ll_inode2fid(dchild->d_inode);
1689 op_data->op_fid2 = op_data->op_fid3;
1690 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1691 ll_finish_md_op_data(op_data);
1693 ll_update_times(request, dir);
1695 ptlrpc_req_finished(request);
1697 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR,
1698 ktime_us_delta(ktime_get(), kstart));
1705 int ll_rmdir_entry(struct inode *dir, char *name, int namelen)
1707 struct ptlrpc_request *request = NULL;
1708 struct md_op_data *op_data;
1709 ktime_t kstart = ktime_get();
1713 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1714 namelen, name, PFID(ll_inode2fid(dir)), dir);
1716 op_data = ll_prep_md_op_data(NULL, dir, NULL, name, strlen(name),
1717 S_IFDIR, LUSTRE_OPC_ANY, NULL);
1718 if (IS_ERR(op_data))
1719 RETURN(PTR_ERR(op_data));
1720 op_data->op_cli_flags |= CLI_RM_ENTRY;
1721 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1722 ll_finish_md_op_data(op_data);
1724 ll_update_times(request, dir);
1726 ptlrpc_req_finished(request);
1728 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR,
1729 ktime_us_delta(ktime_get(), kstart));
1733 static int ll_unlink(struct inode *dir, struct dentry *dchild)
1735 struct qstr *name = &dchild->d_name;
1736 struct ptlrpc_request *request = NULL;
1737 struct md_op_data *op_data;
1738 struct mdt_body *body;
1739 ktime_t kstart = ktime_get();
1744 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1745 name->len, name->name, PFID(ll_inode2fid(dir)), dir);
1748 * XXX: unlink bind mountpoint maybe call to here,
1749 * just check it as vfs_unlink does.
1751 if (unlikely(d_mountpoint(dchild)))
1754 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len, 0,
1755 LUSTRE_OPC_ANY, NULL);
1756 if (IS_ERR(op_data))
1757 RETURN(PTR_ERR(op_data));
1759 op_data->op_fid3 = *ll_inode2fid(dchild->d_inode);
1761 op_data->op_fid2 = op_data->op_fid3;
1762 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1763 ll_finish_md_op_data(op_data);
1768 * The server puts attributes in on the last unlink, use them to update
1769 * the link count so the inode can be freed immediately.
1771 body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
1772 if (body->mbo_valid & OBD_MD_FLNLINK)
1773 set_nlink(dchild->d_inode, body->mbo_nlink);
1775 ll_update_times(request, dir);
1778 ptlrpc_req_finished(request);
1780 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_UNLINK,
1781 ktime_us_delta(ktime_get(), kstart));
1785 static int ll_rename(struct inode *src, struct dentry *src_dchild,
1786 struct inode *tgt, struct dentry *tgt_dchild
1787 #ifdef HAVE_IOPS_RENAME_WITH_FLAGS
1788 , unsigned int flags
1792 struct qstr *src_name = &src_dchild->d_name;
1793 struct qstr *tgt_name = &tgt_dchild->d_name;
1794 struct ptlrpc_request *request = NULL;
1795 struct ll_sb_info *sbi = ll_i2sbi(src);
1796 struct md_op_data *op_data;
1797 ktime_t kstart = ktime_get();
1801 #ifdef HAVE_IOPS_RENAME_WITH_FLAGS
1806 CDEBUG(D_VFSTRACE, "VFS Op:oldname=%.*s, src_dir="DFID
1807 "(%p), newname=%.*s, tgt_dir="DFID"(%p)\n",
1808 src_name->len, src_name->name,
1809 PFID(ll_inode2fid(src)), src, tgt_name->len,
1810 tgt_name->name, PFID(ll_inode2fid(tgt)), tgt);
1812 if (unlikely(d_mountpoint(src_dchild) || d_mountpoint(tgt_dchild)))
1815 op_data = ll_prep_md_op_data(NULL, src, tgt, NULL, 0, 0,
1816 LUSTRE_OPC_ANY, NULL);
1817 if (IS_ERR(op_data))
1818 RETURN(PTR_ERR(op_data));
1820 if (src_dchild->d_inode != NULL)
1821 op_data->op_fid3 = *ll_inode2fid(src_dchild->d_inode);
1823 if (tgt_dchild->d_inode != NULL)
1824 op_data->op_fid4 = *ll_inode2fid(tgt_dchild->d_inode);
1826 err = md_rename(sbi->ll_md_exp, op_data,
1827 src_name->name, src_name->len,
1828 tgt_name->name, tgt_name->len, &request);
1829 ll_finish_md_op_data(op_data);
1831 ll_update_times(request, src);
1832 ll_update_times(request, tgt);
1835 ptlrpc_req_finished(request);
1838 d_move(src_dchild, tgt_dchild);
1839 ll_stats_ops_tally(sbi, LPROC_LL_RENAME,
1840 ktime_us_delta(ktime_get(), kstart));
1846 const struct inode_operations ll_dir_inode_operations = {
1848 #ifdef HAVE_IOP_ATOMIC_OPEN
1849 .atomic_open = ll_atomic_open,
1851 .lookup = ll_lookup_nd,
1852 .create = ll_create_nd,
1853 /* We need all these non-raw things for NFSD, to not patch it. */
1854 .unlink = ll_unlink,
1857 .symlink = ll_symlink,
1859 .rename = ll_rename,
1860 .setattr = ll_setattr,
1861 .getattr = ll_getattr,
1862 .permission = ll_inode_permission,
1863 #ifdef HAVE_IOP_XATTR
1864 .setxattr = ll_setxattr,
1865 .getxattr = ll_getxattr,
1866 .removexattr = ll_removexattr,
1868 .listxattr = ll_listxattr,
1869 #ifdef HAVE_IOP_GET_ACL
1870 .get_acl = ll_get_acl,
1872 #ifdef HAVE_IOP_SET_ACL
1873 .set_acl = ll_set_acl,
1877 const struct inode_operations ll_special_inode_operations = {
1878 .setattr = ll_setattr,
1879 .getattr = ll_getattr,
1880 .permission = ll_inode_permission,
1881 #ifdef HAVE_IOP_XATTR
1882 .setxattr = ll_setxattr,
1883 .getxattr = ll_getxattr,
1884 .removexattr = ll_removexattr,
1886 .listxattr = ll_listxattr,
1887 #ifdef HAVE_IOP_GET_ACL
1888 .get_acl = ll_get_acl,
1890 #ifdef HAVE_IOP_SET_ACL
1891 .set_acl = ll_set_acl,