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);
209 cl_env_put(env, &refcheck);
213 static void ll_lock_cancel_bits(struct ldlm_lock *lock, __u64 to_cancel)
215 struct inode *inode = ll_inode_from_resource_lock(lock);
216 struct ll_inode_info *lli;
217 __u64 bits = to_cancel;
223 /* That means the inode is evicted most likely and may cause
224 * the skipping of lock cleanups below, so print the message
227 if (lock->l_resource->lr_lvb_inode)
229 "can't take inode for the lock (%sevicted)\n",
230 lock->l_resource->lr_lvb_inode->i_state &
231 I_FREEING ? "" : "not ");
235 if (!fid_res_name_eq(ll_inode2fid(inode),
236 &lock->l_resource->lr_name)) {
237 LDLM_ERROR(lock, "data mismatch with object "DFID"(%p)",
238 PFID(ll_inode2fid(inode)), inode);
242 if (bits & MDS_INODELOCK_XATTR) {
243 ll_xattr_cache_destroy(inode);
244 bits &= ~MDS_INODELOCK_XATTR;
247 /* For OPEN locks we differentiate between lock modes
248 * LCK_CR, LCK_CW, LCK_PR - bug 22891 */
249 if (bits & MDS_INODELOCK_OPEN)
250 ll_have_md_lock(inode, &bits, lock->l_req_mode);
252 if (bits & MDS_INODELOCK_OPEN) {
255 switch (lock->l_req_mode) {
266 LDLM_ERROR(lock, "bad lock mode for OPEN lock");
270 ll_md_real_close(inode, fmode);
272 bits &= ~MDS_INODELOCK_OPEN;
275 if (bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
276 MDS_INODELOCK_LAYOUT | MDS_INODELOCK_PERM |
278 ll_have_md_lock(inode, &bits, LCK_MINMODE);
280 if (bits & MDS_INODELOCK_DOM) {
281 rc = ll_dom_lock_cancel(inode, lock);
283 CDEBUG(D_INODE, "cannot flush DoM data "
285 PFID(ll_inode2fid(inode)), rc);
286 lock_res_and_lock(lock);
287 ldlm_set_kms_ignore(lock);
288 unlock_res_and_lock(lock);
291 if (bits & MDS_INODELOCK_LAYOUT) {
292 struct cl_object_conf conf = {
293 .coc_opc = OBJECT_CONF_INVALIDATE,
297 rc = ll_layout_conf(inode, &conf);
299 CDEBUG(D_INODE, "cannot invalidate layout of "
301 PFID(ll_inode2fid(inode)), rc);
304 lli = ll_i2info(inode);
306 if (bits & MDS_INODELOCK_UPDATE)
307 lli->lli_update_atime = 1;
309 if ((bits & MDS_INODELOCK_UPDATE) && S_ISDIR(inode->i_mode)) {
310 CDEBUG(D_INODE, "invalidating inode "DFID" lli = %p, "
311 "pfid = "DFID"\n", PFID(ll_inode2fid(inode)),
312 lli, PFID(&lli->lli_pfid));
313 truncate_inode_pages(inode->i_mapping, 0);
315 if (unlikely(!fid_is_zero(&lli->lli_pfid))) {
316 struct inode *master_inode = NULL;
319 /* This is slave inode, since all of the child dentry
320 * is connected on the master inode, so we have to
321 * invalidate the negative children on master inode */
322 CDEBUG(D_INODE, "Invalidate s"DFID" m"DFID"\n",
323 PFID(ll_inode2fid(inode)), PFID(&lli->lli_pfid));
325 hash = cl_fid_build_ino(&lli->lli_pfid,
326 ll_need_32bit_api(ll_i2sbi(inode)));
328 /* Do not lookup the inode with ilookup5, otherwise
329 * it will cause dead lock,
330 * 1. Client1 send chmod req to the MDT0, then on MDT0,
331 * it enqueues master and all of its slaves lock,
332 * (mdt_attr_set() -> mdt_lock_slaves()), after gets
333 * master and stripe0 lock, it will send the enqueue
334 * req (for stripe1) to MDT1, then MDT1 finds the lock
335 * has been granted to client2. Then MDT1 sends blocking
337 * 2. At the same time, client2 tries to unlink
338 * the striped dir (rm -rf striped_dir), and during
339 * lookup, it will hold the master inode of the striped
340 * directory, whose inode state is NEW, then tries to
341 * revalidate all of its slaves, (ll_prep_inode()->
342 * ll_iget()->ll_read_inode2()-> ll_update_inode().).
343 * And it will be blocked on the server side because
345 * 3. Then the client get the blocking_ast req, cancel
346 * the lock, but being blocked if using ->ilookup5()),
347 * because master inode state is NEW. */
348 master_inode = ilookup5_nowait(inode->i_sb, hash,
349 ll_test_inode_by_fid,
350 (void *)&lli->lli_pfid);
352 ll_invalidate_negative_children(master_inode);
356 ll_invalidate_negative_children(inode);
360 if ((bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM)) &&
361 inode->i_sb->s_root != NULL &&
362 inode != inode->i_sb->s_root->d_inode)
363 ll_invalidate_aliases(inode);
365 if (bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM))
366 forget_all_cached_acls(inode);
372 /* Check if the given lock may be downgraded instead of canceling and
373 * that convert is really needed. */
374 int ll_md_need_convert(struct ldlm_lock *lock)
376 struct ldlm_namespace *ns = ldlm_lock_to_ns(lock);
378 __u64 wanted = lock->l_policy_data.l_inodebits.cancel_bits;
379 __u64 bits = lock->l_policy_data.l_inodebits.bits & ~wanted;
380 enum ldlm_mode mode = LCK_MINMODE;
382 if (!lock->l_conn_export ||
383 !exp_connect_lock_convert(lock->l_conn_export))
386 if (!wanted || !bits || ldlm_is_cancel(lock))
389 /* do not convert locks other than DOM for now */
390 if (!((bits | wanted) & MDS_INODELOCK_DOM))
393 /* We may have already remaining bits in some other lock so
394 * lock convert will leave us just extra lock for the same bit.
395 * Check if client has other lock with the same bits and the same
396 * or lower mode and don't convert if any.
398 switch (lock->l_req_mode) {
412 /* do not convert other modes */
416 /* is lock is too old to be converted? */
417 lock_res_and_lock(lock);
418 if (ktime_after(ktime_get(),
419 ktime_add(lock->l_last_used,
420 ktime_set(ns->ns_dirty_age_limit, 0)))) {
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 *desc,
433 void *data, int flag)
435 struct lustre_handle lockh;
436 __u64 bits = lock->l_policy_data.l_inodebits.bits;
442 case LDLM_CB_BLOCKING:
444 __u64 cancel_flags = LCF_ASYNC;
446 if (ll_md_need_convert(lock)) {
447 cancel_flags |= LCF_CONVERT;
448 /* For lock convert some cancel actions may require
449 * this lock with non-dropped canceled bits, e.g. page
450 * flush for DOM lock. So call ll_lock_cancel_bits()
451 * here while canceled bits are still set.
453 bits = lock->l_policy_data.l_inodebits.cancel_bits;
454 if (bits & MDS_INODELOCK_DOM)
455 ll_lock_cancel_bits(lock, MDS_INODELOCK_DOM);
457 ldlm_lock2handle(lock, &lockh);
458 rc = ldlm_cli_cancel(&lockh, cancel_flags);
460 CDEBUG(D_INODE, "ldlm_cli_cancel: rc = %d\n", rc);
465 case LDLM_CB_CANCELING:
466 /* Nothing to do for non-granted locks */
467 if (!ldlm_is_granted(lock))
470 if (ldlm_is_converting(lock)) {
471 /* this is called on already converted lock, so
472 * ibits has remained bits only and cancel_bits
473 * are bits that were dropped.
474 * Note that DOM lock is handled prior lock convert
475 * and is excluded here.
477 bits = lock->l_policy_data.l_inodebits.cancel_bits &
480 LASSERT(ldlm_is_canceling(lock));
482 ll_lock_cancel_bits(lock, bits);
491 __u32 ll_i2suppgid(struct inode *i)
493 if (in_group_p(i->i_gid))
494 return (__u32)from_kgid(&init_user_ns, i->i_gid);
496 return (__u32) __kgid_val(INVALID_GID);
499 /* Pack the required supplementary groups into the supplied groups array.
500 * If we don't need to use the groups from the target inode(s) then we
501 * instead pack one or more groups from the user's supplementary group
502 * array in case it might be useful. Not needed if doing an MDS-side upcall. */
503 void ll_i2gids(__u32 *suppgids, struct inode *i1, struct inode *i2)
506 LASSERT(suppgids != NULL);
508 suppgids[0] = ll_i2suppgid(i1);
511 suppgids[1] = ll_i2suppgid(i2);
517 * try to reuse three types of dentry:
518 * 1. unhashed alias, this one is unhashed by d_invalidate (but it may be valid
519 * by concurrent .revalidate).
520 * 2. INVALID alias (common case for no valid ldlm lock held, but this flag may
521 * be cleared by others calling d_lustre_revalidate).
522 * 3. DISCONNECTED alias.
524 static struct dentry *ll_find_alias(struct inode *inode, struct dentry *dentry)
526 struct dentry *alias, *discon_alias, *invalid_alias;
527 DECLARE_LL_D_HLIST_NODE_PTR(p);
529 if (ll_d_hlist_empty(&inode->i_dentry))
532 discon_alias = invalid_alias = NULL;
534 spin_lock(&inode->i_lock);
535 ll_d_hlist_for_each_entry(alias, p, &inode->i_dentry) {
536 LASSERT(alias != dentry);
538 spin_lock(&alias->d_lock);
539 if ((alias->d_flags & DCACHE_DISCONNECTED) &&
540 S_ISDIR(inode->i_mode))
541 /* LASSERT(last_discon == NULL); LU-405, bz 20055 */
542 discon_alias = alias;
543 else if (alias->d_parent == dentry->d_parent &&
544 alias->d_name.hash == dentry->d_name.hash &&
545 alias->d_name.len == dentry->d_name.len &&
546 memcmp(alias->d_name.name, dentry->d_name.name,
547 dentry->d_name.len) == 0)
548 invalid_alias = alias;
549 spin_unlock(&alias->d_lock);
554 alias = invalid_alias ?: discon_alias ?: NULL;
556 spin_lock(&alias->d_lock);
558 spin_unlock(&alias->d_lock);
560 spin_unlock(&inode->i_lock);
566 * Similar to d_splice_alias(), but lustre treats invalid alias
567 * similar to DCACHE_DISCONNECTED, and tries to use it anyway.
569 struct dentry *ll_splice_alias(struct inode *inode, struct dentry *de)
575 new = ll_find_alias(inode, de);
585 "Reuse dentry %p inode %p refc %d flags %#x\n",
586 new, new->d_inode, ll_d_count(new), new->d_flags);
594 CDEBUG(D_DENTRY, "Add dentry %p inode %p refc %d flags %#x\n",
595 de, de->d_inode, ll_d_count(de), de->d_flags);
599 static int ll_lookup_it_finish(struct ptlrpc_request *request,
600 struct lookup_intent *it,
601 struct inode *parent, struct dentry **de,
602 void *secctx, __u32 secctxlen)
604 struct inode *inode = NULL;
607 struct dentry *alias;
610 /* NB 1 request reference will be taken away by ll_intent_lock()
612 CDEBUG(D_DENTRY, "it %p it_disposition %x\n", it,
614 if (!it_disposition(it, DISP_LOOKUP_NEG)) {
615 struct req_capsule *pill = &request->rq_pill;
616 struct mdt_body *body = req_capsule_server_get(pill,
619 rc = ll_prep_inode(&inode, request, (*de)->d_sb, it);
623 if (it->it_op & IT_OPEN)
624 ll_dom_finish_open(inode, request, it);
626 ll_set_lock_data(ll_i2sbi(parent)->ll_md_exp, inode, it, &bits);
628 /* We used to query real size from OSTs here, but actually
629 * this is not needed. For stat() calls size would be updated
630 * from subsequent do_revalidate()->ll_inode_revalidate_it() in
632 * vfs_getattr_it->ll_getattr()->ll_inode_revalidate_it() in 2.6
633 * Everybody else who needs correct file size would call
634 * ll_glimpse_size or some equivalent themselves anyway.
638 /* If security context was returned by MDT, put it in
639 * inode now to save an extra getxattr from security hooks,
640 * and avoid deadlock.
642 if (body->mbo_valid & OBD_MD_SECCTX) {
643 secctx = req_capsule_server_get(pill, &RMF_FILE_SECCTX);
644 secctxlen = req_capsule_get_size(pill,
649 CDEBUG(D_SEC, "server returned security context"
651 PFID(ll_inode2fid(inode)));
654 if (secctx != NULL && secctxlen != 0) {
656 rc = security_inode_notifysecctx(inode, secctx,
660 CWARN("cannot set security context for "
662 PFID(ll_inode2fid(inode)), rc);
666 /* Only hash *de if it is unhashed (new dentry).
667 * Atoimc_open may passin hashed dentries for open.
669 alias = ll_splice_alias(inode, *de);
671 GOTO(out, rc = PTR_ERR(alias));
675 if (!it_disposition(it, DISP_LOOKUP_NEG)) {
676 /* we have lookup look - unhide dentry */
677 if (bits & MDS_INODELOCK_LOOKUP)
678 d_lustre_revalidate(*de);
679 } else if (!it_disposition(it, DISP_OPEN_CREATE)) {
681 * If file was created on the server, the dentry is revalidated
682 * in ll_create_it if the lock allows for it.
684 /* Check that parent has UPDATE lock. */
685 struct lookup_intent parent_it = {
687 .it_lock_handle = 0 };
688 struct lu_fid fid = ll_i2info(parent)->lli_fid;
690 /* If it is striped directory, get the real stripe parent */
691 if (unlikely(ll_dir_striped(parent))) {
692 rc = md_get_fid_from_lsm(ll_i2mdexp(parent),
693 ll_i2info(parent)->lli_lsm_md,
695 (*de)->d_name.len, &fid);
700 if (md_revalidate_lock(ll_i2mdexp(parent), &parent_it, &fid,
702 d_lustre_revalidate(*de);
703 ll_intent_release(&parent_it);
710 if (rc != 0 && it->it_op & IT_OPEN)
711 ll_open_cleanup((*de)->d_sb, request);
716 struct pcc_create_attach {
717 struct pcc_dataset *pca_dataset;
718 struct dentry *pca_dentry;
721 static struct dentry *ll_lookup_it(struct inode *parent, struct dentry *dentry,
722 struct lookup_intent *it,
723 void **secctx, __u32 *secctxlen,
724 struct pcc_create_attach *pca)
726 struct lookup_intent lookup_it = { .it_op = IT_LOOKUP };
727 struct dentry *save = dentry, *retval;
728 struct ptlrpc_request *req = NULL;
729 struct md_op_data *op_data = NULL;
730 struct lov_user_md *lum = NULL;
733 char secctx_name[XATTR_NAME_MAX + 1];
737 if (dentry->d_name.len > ll_i2sbi(parent)->ll_namelen)
738 RETURN(ERR_PTR(-ENAMETOOLONG));
740 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), intent=%s\n",
741 dentry->d_name.len, dentry->d_name.name,
742 PFID(ll_inode2fid(parent)), parent, LL_IT2STR(it));
744 if (d_mountpoint(dentry))
745 CERROR("Tell Peter, lookup on mtpt, it %s\n", LL_IT2STR(it));
747 if (it == NULL || it->it_op == IT_GETXATTR)
750 if (it->it_op == IT_GETATTR && dentry_may_statahead(parent, dentry)) {
751 rc = ll_statahead(parent, &dentry, 0);
753 RETURN(dentry == save ? NULL : dentry);
756 if (it->it_op & IT_OPEN && it->it_flags & FMODE_WRITE &&
757 dentry->d_sb->s_flags & SB_RDONLY)
758 RETURN(ERR_PTR(-EROFS));
760 if (it->it_op & IT_CREAT)
761 opc = LUSTRE_OPC_CREATE;
763 opc = LUSTRE_OPC_ANY;
765 op_data = ll_prep_md_op_data(NULL, parent, NULL, dentry->d_name.name,
766 dentry->d_name.len, 0, opc, NULL);
768 GOTO(out, retval = ERR_CAST(op_data));
770 /* enforce umask if acl disabled or MDS doesn't support umask */
771 if (!IS_POSIXACL(parent) || !exp_connect_umask(ll_i2mdexp(parent)))
772 it->it_create_mode &= ~current_umask();
774 if (it->it_op & IT_CREAT &&
775 ll_i2sbi(parent)->ll_flags & LL_SBI_FILE_SECCTX) {
776 rc = ll_dentry_init_security(dentry, it->it_create_mode,
778 &op_data->op_file_secctx_name,
779 &op_data->op_file_secctx,
780 &op_data->op_file_secctx_size);
782 GOTO(out, retval = ERR_PTR(rc));
784 *secctx = op_data->op_file_secctx;
785 if (secctxlen != NULL)
786 *secctxlen = op_data->op_file_secctx_size;
790 if (secctxlen != NULL)
794 /* ask for security context upon intent */
795 if (it->it_op & (IT_LOOKUP | IT_GETATTR | IT_OPEN)) {
796 /* get name of security xattr to request to server */
797 rc = ll_listsecurity(parent, secctx_name,
798 sizeof(secctx_name));
800 CDEBUG(D_SEC, "cannot get security xattr name for "
802 PFID(ll_inode2fid(parent)), rc);
804 op_data->op_file_secctx_name = secctx_name;
805 op_data->op_file_secctx_name_size = rc;
806 CDEBUG(D_SEC, "'%.*s' is security xattr for "DFID"\n",
807 rc, secctx_name, PFID(ll_inode2fid(parent)));
811 if (pca && pca->pca_dataset) {
812 struct pcc_dataset *dataset = pca->pca_dataset;
816 GOTO(out, retval = ERR_PTR(-ENOMEM));
818 lum->lmm_magic = LOV_USER_MAGIC_V1;
819 lum->lmm_pattern = LOV_PATTERN_F_RELEASED | LOV_PATTERN_RAID0;
820 op_data->op_data = lum;
821 op_data->op_data_size = sizeof(*lum);
822 op_data->op_archive_id = dataset->pccd_rwid;
824 rc = obd_fid_alloc(NULL, ll_i2mdexp(parent), &op_data->op_fid2,
827 GOTO(out, retval = ERR_PTR(rc));
829 rc = pcc_inode_create(parent->i_sb, dataset, &op_data->op_fid2,
832 GOTO(out, retval = ERR_PTR(rc));
834 it->it_flags |= MDS_OPEN_PCC;
837 rc = md_intent_lock(ll_i2mdexp(parent), op_data, it, &req,
838 &ll_md_blocking_ast, 0);
839 /* If the MDS allows the client to chgrp (CFS_SETGRP_PERM), but the
840 * client does not know which suppgid should be sent to the MDS, or
841 * some other(s) changed the target file's GID after this RPC sent
842 * to the MDS with the suppgid as the original GID, then we should
843 * try again with right suppgid. */
844 if (rc == -EACCES && it->it_op & IT_OPEN &&
845 it_disposition(it, DISP_OPEN_DENY)) {
846 struct mdt_body *body;
848 LASSERT(req != NULL);
850 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
851 if (op_data->op_suppgids[0] == body->mbo_gid ||
852 op_data->op_suppgids[1] == body->mbo_gid ||
853 !in_group_p(make_kgid(&init_user_ns, body->mbo_gid)))
854 GOTO(out, retval = ERR_PTR(-EACCES));
856 fid_zero(&op_data->op_fid2);
857 op_data->op_suppgids[1] = body->mbo_gid;
858 ptlrpc_req_finished(req);
860 ll_intent_release(it);
861 rc = md_intent_lock(ll_i2mdexp(parent), op_data, it, &req,
862 &ll_md_blocking_ast, 0);
866 GOTO(out, retval = ERR_PTR(rc));
868 /* dir layout may change */
869 ll_unlock_md_op_lsm(op_data);
870 rc = ll_lookup_it_finish(req, it, parent, &dentry,
871 secctx != NULL ? *secctx : NULL,
872 secctxlen != NULL ? *secctxlen : 0);
874 ll_intent_release(it);
875 GOTO(out, retval = ERR_PTR(rc));
878 if ((it->it_op & IT_OPEN) && dentry->d_inode &&
879 !S_ISREG(dentry->d_inode->i_mode) &&
880 !S_ISDIR(dentry->d_inode->i_mode)) {
881 ll_release_openhandle(dentry, it);
883 ll_lookup_finish_locks(it, dentry);
885 GOTO(out, retval = (dentry == save) ? NULL : dentry);
888 if (op_data != NULL && !IS_ERR(op_data)) {
889 if (secctx != NULL && secctxlen != NULL) {
890 /* caller needs sec ctx info, so reset it in op_data to
891 * prevent it from being freed */
892 op_data->op_file_secctx = NULL;
893 op_data->op_file_secctx_size = 0;
895 ll_finish_md_op_data(op_data);
901 ptlrpc_req_finished(req);
905 #ifdef HAVE_IOP_ATOMIC_OPEN
906 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
909 struct lookup_intent *itp, it = { .it_op = IT_GETATTR };
912 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), flags=%u\n",
913 dentry->d_name.len, dentry->d_name.name,
914 PFID(ll_inode2fid(parent)), parent, flags);
917 * Optimize away (CREATE && !OPEN). Let .create handle the race.
918 * but only if we have write permissions there, otherwise we need
919 * to proceed with lookup. LU-4185
921 if ((flags & LOOKUP_CREATE) && !(flags & LOOKUP_OPEN) &&
922 (inode_permission(parent, MAY_WRITE | MAY_EXEC) == 0))
925 if (flags & (LOOKUP_PARENT|LOOKUP_OPEN|LOOKUP_CREATE))
929 de = ll_lookup_it(parent, dentry, itp, NULL, NULL, NULL);
932 ll_intent_release(itp);
937 #ifdef FMODE_CREATED /* added in Linux v4.18-rc1-20-g73a09dd */
938 # define ll_is_opened(o, f) ((f)->f_mode & FMODE_OPENED)
939 # define ll_finish_open(f, d, o) finish_open((f), (d), NULL)
941 # define ll_set_created(o, f) \
943 (f)->f_mode |= FMODE_CREATED; \
947 # define ll_is_opened(o, f) (*(o))
948 # define ll_finish_open(f, d, o) finish_open((f), (d), NULL, (o))
949 # define ll_last_arg , int *opened
950 # define ll_set_created(o, f) \
952 *(o) |= FILE_CREATED; \
958 * For cached negative dentry and new dentry, handle lookup/create/open
961 static int ll_atomic_open(struct inode *dir, struct dentry *dentry,
962 struct file *file, unsigned open_flags,
963 umode_t mode ll_last_arg)
965 struct lookup_intent *it;
967 long long lookup_flags = LOOKUP_OPEN;
970 struct ll_sb_info *sbi;
971 struct pcc_create_attach pca = {NULL, NULL};
972 struct pcc_dataset *dataset = NULL;
976 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), file %p,"
977 "open_flags %x, mode %x opened %d\n",
978 dentry->d_name.len, dentry->d_name.name,
979 PFID(ll_inode2fid(dir)), dir, file, open_flags, mode,
980 ll_is_opened(opened, file));
982 /* Only negative dentries enter here */
983 LASSERT(dentry->d_inode == NULL);
985 if (!d_unhashed(dentry)) {
986 /* A valid negative dentry that just passed revalidation,
987 * there's little point to try and open it server-side,
988 * even though there's a minuscule chance it might succeed.
989 * Either way it's a valid race to just return -ENOENT here.
991 if (!(open_flags & O_CREAT))
994 /* Otherwise we just unhash it to be rehashed afresh via
995 * lookup if necessary
1000 OBD_ALLOC(it, sizeof(*it));
1004 it->it_op = IT_OPEN;
1005 if (open_flags & O_CREAT) {
1006 it->it_op |= IT_CREAT;
1007 lookup_flags |= LOOKUP_CREATE;
1008 sbi = ll_i2sbi(dir);
1009 /* Volatile file is used for HSM restore, so do not use PCC */
1010 if (!filename_is_volatile(dentry->d_name.name,
1011 dentry->d_name.len, NULL)) {
1012 struct pcc_matcher item;
1014 item.pm_uid = from_kuid(&init_user_ns, current_uid());
1015 item.pm_gid = from_kgid(&init_user_ns, current_gid());
1016 item.pm_projid = ll_i2info(dir)->lli_projid;
1017 item.pm_name = &dentry->d_name;
1018 dataset = pcc_dataset_match_get(&sbi->ll_pcc_super,
1020 pca.pca_dataset = dataset;
1023 it->it_create_mode = (mode & S_IALLUGO) | S_IFREG;
1024 it->it_flags = (open_flags & ~O_ACCMODE) | OPEN_FMODE(open_flags);
1025 it->it_flags &= ~MDS_OPEN_FL_INTERNAL;
1027 /* Dentry added to dcache tree in ll_lookup_it */
1028 de = ll_lookup_it(dir, dentry, it, &secctx, &secctxlen, &pca);
1031 else if (de != NULL)
1034 CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_FILE_PAUSE, cfs_fail_val);
1037 if (it_disposition(it, DISP_OPEN_CREATE)) {
1038 /* Dentry instantiated in ll_create_it. */
1039 rc = ll_create_it(dir, dentry, it, secctx, secctxlen);
1040 security_release_secctx(secctx, secctxlen);
1042 /* We dget in ll_splice_alias. */
1047 if (dataset != NULL && dentry->d_inode) {
1048 rc = pcc_inode_create_fini(dataset,
1054 GOTO(out_release, rc);
1057 ll_set_created(opened, file);
1060 if (dentry->d_inode && it_disposition(it, DISP_OPEN_OPEN)) {
1062 if (S_ISFIFO(dentry->d_inode->i_mode)) {
1063 /* We cannot call open here as it might
1064 * deadlock. This case is unreachable in
1065 * practice because of OBD_CONNECT_NODEVOH. */
1066 rc = finish_no_open(file, de);
1068 file->private_data = it;
1069 rc = ll_finish_open(file, dentry, opened);
1070 /* We dget in ll_splice_alias. finish_open takes
1071 * care of dget for fd open.
1077 rc = finish_no_open(file, de);
1082 if (dataset != NULL)
1083 pcc_dataset_put(dataset);
1084 ll_intent_release(it);
1085 OBD_FREE(it, sizeof(*it));
1090 #else /* !HAVE_IOP_ATOMIC_OPEN */
1091 static struct lookup_intent *
1092 ll_convert_intent(struct open_intent *oit, int lookup_flags, bool is_readonly)
1094 struct lookup_intent *it;
1098 return ERR_PTR(-ENOMEM);
1100 if (lookup_flags & LOOKUP_OPEN) {
1101 it->it_op = IT_OPEN;
1102 /* Avoid file creation for ro bind mount point(is_readonly) */
1103 if ((lookup_flags & LOOKUP_CREATE) && !is_readonly)
1104 it->it_op |= IT_CREAT;
1105 it->it_create_mode = (oit->create_mode & S_IALLUGO) | S_IFREG;
1106 it->it_flags = ll_namei_to_lookup_intent_flag(oit->flags &
1107 ~(is_readonly ? O_CREAT : 0));
1108 it->it_flags &= ~MDS_OPEN_FL_INTERNAL;
1110 it->it_op = IT_GETATTR;
1116 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
1117 struct nameidata *nd)
1122 if (nd && !(nd->flags & (LOOKUP_CONTINUE|LOOKUP_PARENT))) {
1123 struct lookup_intent *it;
1125 if (ll_d2d(dentry) && ll_d2d(dentry)->lld_it) {
1126 it = ll_d2d(dentry)->lld_it;
1127 ll_d2d(dentry)->lld_it = NULL;
1130 * Optimize away (CREATE && !OPEN). Let .create handle
1131 * the race. But only if we have write permissions
1132 * there, otherwise we need to proceed with lookup.
1135 if ((nd->flags & LOOKUP_CREATE) &&
1136 !(nd->flags & LOOKUP_OPEN) &&
1137 (inode_permission(parent,
1138 MAY_WRITE | MAY_EXEC) == 0))
1141 it = ll_convert_intent(&nd->intent.open, nd->flags,
1142 (nd->path.mnt->mnt_flags & MNT_READONLY) ||
1143 (nd->path.mnt->mnt_sb->s_flags & SB_RDONLY));
1145 RETURN((struct dentry *)it);
1148 de = ll_lookup_it(parent, dentry, it, NULL, NULL, NULL);
1151 if ((nd->flags & LOOKUP_OPEN) && !IS_ERR(dentry)) { /* Open */
1152 if (dentry->d_inode &&
1153 it_disposition(it, DISP_OPEN_OPEN)) { /* nocreate */
1154 if (S_ISFIFO(dentry->d_inode->i_mode)) {
1155 /* We cannot call open here as it might
1156 * deadlock. This case is unreachable in
1157 * practice because of
1158 * OBD_CONNECT_NODEVOH. */
1162 nd->intent.open.file->private_data = it;
1163 filp = lookup_instantiate_filp(nd,
1169 de = (struct dentry *)filp;
1172 } else if (it_disposition(it, DISP_OPEN_CREATE)) {
1173 /* XXX This can only reliably work on assumption
1174 * that there are NO hashed negative dentries.*/
1175 ll_d2d(dentry)->lld_it = it;
1176 it = NULL; /* Will be freed in ll_create_nd */
1177 /* We absolutely depend on ll_create_nd to be
1178 * called to not leak this intent and possible
1179 * data attached to it */
1184 ll_intent_release(it);
1185 OBD_FREE(it, sizeof(*it));
1188 de = ll_lookup_it(parent, dentry, NULL, NULL, NULL, NULL);
1193 #endif /* HAVE_IOP_ATOMIC_OPEN */
1195 /* We depend on "mode" being set with the proper file type/umask by now */
1196 static struct inode *ll_create_node(struct inode *dir, struct lookup_intent *it)
1198 struct inode *inode = NULL;
1199 struct ptlrpc_request *request = NULL;
1200 struct ll_sb_info *sbi = ll_i2sbi(dir);
1204 LASSERT(it && it->it_disposition);
1206 LASSERT(it_disposition(it, DISP_ENQ_CREATE_REF));
1207 request = it->it_request;
1208 it_clear_disposition(it, DISP_ENQ_CREATE_REF);
1209 rc = ll_prep_inode(&inode, request, dir->i_sb, it);
1211 GOTO(out, inode = ERR_PTR(rc));
1213 /* Pause to allow for a race with concurrent access by fid */
1214 OBD_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_NODE_PAUSE, cfs_fail_val);
1216 /* We asked for a lock on the directory, but were granted a
1217 * lock on the inode. Since we finally have an inode pointer,
1218 * stuff it in the lock. */
1219 CDEBUG(D_DLMTRACE, "setting l_ast_data to inode "DFID"(%p)\n",
1220 PFID(ll_inode2fid(inode)), inode);
1221 ll_set_lock_data(sbi->ll_md_exp, inode, it, NULL);
1224 ptlrpc_req_finished(request);
1229 * By the time this is called, we already have created the directory cache
1230 * entry for the new file, but it is so far negative - it has no inode.
1232 * We defer creating the OBD object(s) until open, to keep the intent and
1233 * non-intent code paths similar, and also because we do not have the MDS
1234 * inode number before calling ll_create_node() (which is needed for LOV),
1235 * so we would need to do yet another RPC to the MDS to store the LOV EA
1236 * data on the MDS. If needed, we would pass the PACKED lmm as data and
1237 * lmm_size in datalen (the MDS still has code which will handle that).
1239 * If the create succeeds, we fill in the inode information
1240 * with d_instantiate().
1242 static int ll_create_it(struct inode *dir, struct dentry *dentry,
1243 struct lookup_intent *it,
1244 void *secctx, __u32 secctxlen)
1246 struct inode *inode;
1251 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), intent=%s\n",
1252 dentry->d_name.len, dentry->d_name.name,
1253 PFID(ll_inode2fid(dir)), dir, LL_IT2STR(it));
1255 rc = it_open_error(DISP_OPEN_CREATE, it);
1259 inode = ll_create_node(dir, it);
1261 RETURN(PTR_ERR(inode));
1263 if ((ll_i2sbi(inode)->ll_flags & LL_SBI_FILE_SECCTX) &&
1266 /* must be done before d_instantiate, because it calls
1267 * security_d_instantiate, which means a getxattr if security
1268 * context is not set yet */
1269 rc = security_inode_notifysecctx(inode, secctx, secctxlen);
1270 inode_unlock(inode);
1275 d_instantiate(dentry, inode);
1277 if (!(ll_i2sbi(inode)->ll_flags & LL_SBI_FILE_SECCTX)) {
1278 rc = ll_inode_init_security(dentry, inode, dir);
1283 ll_set_lock_data(ll_i2sbi(dir)->ll_md_exp, inode, it, &bits);
1284 if (bits & MDS_INODELOCK_LOOKUP)
1285 d_lustre_revalidate(dentry);
1290 void ll_update_times(struct ptlrpc_request *request, struct inode *inode)
1292 struct mdt_body *body = req_capsule_server_get(&request->rq_pill,
1296 if (body->mbo_valid & OBD_MD_FLMTIME &&
1297 body->mbo_mtime > inode->i_mtime.tv_sec) {
1299 "setting fid " DFID " mtime from %lld to %llu\n",
1300 PFID(ll_inode2fid(inode)),
1301 (s64)inode->i_mtime.tv_sec, body->mbo_mtime);
1302 inode->i_mtime.tv_sec = body->mbo_mtime;
1305 if (body->mbo_valid & OBD_MD_FLCTIME &&
1306 body->mbo_ctime > inode->i_ctime.tv_sec)
1307 inode->i_ctime.tv_sec = body->mbo_ctime;
1310 static int ll_new_node(struct inode *dir, struct dentry *dchild,
1311 const char *tgt, umode_t mode, int rdev, __u32 opc)
1313 struct qstr *name = &dchild->d_name;
1314 struct ptlrpc_request *request = NULL;
1315 struct md_op_data *op_data;
1316 struct inode *inode = NULL;
1317 struct ll_sb_info *sbi = ll_i2sbi(dir);
1322 if (unlikely(tgt != NULL))
1323 tgt_len = strlen(tgt) + 1;
1326 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
1327 name->len, 0, opc, NULL);
1328 if (IS_ERR(op_data))
1329 GOTO(err_exit, err = PTR_ERR(op_data));
1331 if (sbi->ll_flags & LL_SBI_FILE_SECCTX) {
1332 err = ll_dentry_init_security(dchild, mode, &dchild->d_name,
1333 &op_data->op_file_secctx_name,
1334 &op_data->op_file_secctx,
1335 &op_data->op_file_secctx_size);
1337 GOTO(err_exit, err);
1340 err = md_create(sbi->ll_md_exp, op_data, tgt, tgt_len, mode,
1341 from_kuid(&init_user_ns, current_fsuid()),
1342 from_kgid(&init_user_ns, current_fsgid()),
1343 cfs_curproc_cap_pack(), rdev, &request);
1344 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 14, 58, 0)
1346 * server < 2.12.58 doesn't pack default LMV in intent_getattr reply,
1347 * fetch default LMV here.
1349 if (unlikely(err == -EREMOTE)) {
1350 struct ll_inode_info *lli = ll_i2info(dir);
1351 struct lmv_user_md *lum;
1355 ptlrpc_req_finished(request);
1358 err2 = ll_dir_getstripe(dir, (void **)&lum, &lumsize, &request,
1359 OBD_MD_DEFAULT_MEA);
1360 ll_finish_md_op_data(op_data);
1363 struct lustre_md md = { NULL };
1365 md.body = req_capsule_server_get(&request->rq_pill,
1368 GOTO(err_exit, err = -EPROTO);
1370 OBD_ALLOC_PTR(md.default_lmv);
1371 if (!md.default_lmv)
1372 GOTO(err_exit, err = -ENOMEM);
1374 md.default_lmv->lsm_md_magic = lum->lum_magic;
1375 md.default_lmv->lsm_md_stripe_count =
1376 lum->lum_stripe_count;
1377 md.default_lmv->lsm_md_master_mdt_index =
1378 lum->lum_stripe_offset;
1379 md.default_lmv->lsm_md_hash_type = lum->lum_hash_type;
1381 err = ll_update_inode(dir, &md);
1382 md_free_lustre_md(sbi->ll_md_exp, &md);
1384 GOTO(err_exit, err);
1385 } else if (err2 == -ENODATA && lli->lli_default_lsm_md) {
1387 * If there are no default stripe EA on the MDT, but the
1388 * client has default stripe, then it probably means
1389 * default stripe EA has just been deleted.
1391 down_write(&lli->lli_lsm_sem);
1392 if (lli->lli_default_lsm_md)
1393 OBD_FREE_PTR(lli->lli_default_lsm_md);
1394 lli->lli_default_lsm_md = NULL;
1395 up_write(&lli->lli_lsm_sem);
1397 GOTO(err_exit, err);
1400 ptlrpc_req_finished(request);
1407 GOTO(err_exit, err);
1409 ll_update_times(request, dir);
1411 CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_NEWNODE_PAUSE, cfs_fail_val);
1413 err = ll_prep_inode(&inode, request, dchild->d_sb, NULL);
1415 GOTO(err_exit, err);
1417 if (sbi->ll_flags & LL_SBI_FILE_SECCTX) {
1419 /* must be done before d_instantiate, because it calls
1420 * security_d_instantiate, which means a getxattr if security
1421 * context is not set yet */
1422 err = security_inode_notifysecctx(inode,
1423 op_data->op_file_secctx,
1424 op_data->op_file_secctx_size);
1425 inode_unlock(inode);
1427 GOTO(err_exit, err);
1430 d_instantiate(dchild, inode);
1432 if (!(sbi->ll_flags & LL_SBI_FILE_SECCTX)) {
1433 err = ll_inode_init_security(dchild, inode, dir);
1435 GOTO(err_exit, err);
1440 if (request != NULL)
1441 ptlrpc_req_finished(request);
1443 if (!IS_ERR_OR_NULL(op_data))
1444 ll_finish_md_op_data(op_data);
1449 static int ll_mknod(struct inode *dir, struct dentry *dchild, ll_umode_t mode,
1452 struct qstr *name = &dchild->d_name;
1456 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p) mode %o dev %x\n",
1457 name->len, name->name, PFID(ll_inode2fid(dir)), dir,
1460 if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
1461 mode &= ~current_umask();
1463 switch (mode & S_IFMT) {
1472 err = ll_new_node(dir, dchild, NULL, mode, old_encode_dev(rdev),
1483 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKNOD, 1);
1488 #ifdef HAVE_IOP_ATOMIC_OPEN
1490 * Plain create. Intent create is handled in atomic_open.
1492 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
1493 umode_t mode, bool want_excl)
1497 CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_FILE_PAUSE, cfs_fail_val);
1499 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), "
1500 "flags=%u, excl=%d\n", dentry->d_name.len,
1501 dentry->d_name.name, PFID(ll_inode2fid(dir)),
1502 dir, mode, want_excl);
1504 /* Using mknod(2) to create a regular file is designed to not recognize
1505 * volatile file name, so we use ll_mknod() here. */
1506 rc = ll_mknod(dir, dentry, mode, 0);
1508 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_CREATE, 1);
1510 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, unhashed %d\n",
1511 dentry->d_name.len, dentry->d_name.name, d_unhashed(dentry));
1515 #else /* !HAVE_IOP_ATOMIC_OPEN */
1516 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
1517 ll_umode_t mode, struct nameidata *nd)
1519 struct ll_dentry_data *lld = ll_d2d(dentry);
1520 struct lookup_intent *it = NULL;
1523 CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_FILE_PAUSE, cfs_fail_val);
1529 /* LU-8559: use LUSTRE_OPC_CREATE for non atomic open case
1530 * so that volatile file name is recoginized.
1531 * Mknod(2), however, is designed to not recognize volatile
1532 * file name to avoid inode leak under orphan directory until
1534 return ll_new_node(dir, dentry, NULL, mode, 0,
1540 /* Was there an error? Propagate it! */
1541 if (it->it_status) {
1546 rc = ll_create_it(dir, dentry, it, NULL, 0);
1547 if (nd && (nd->flags & LOOKUP_OPEN) && dentry->d_inode) { /* Open */
1550 nd->intent.open.file->private_data = it;
1551 filp = lookup_instantiate_filp(nd, dentry, NULL);
1557 ll_intent_release(it);
1558 OBD_FREE(it, sizeof(*it));
1561 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_CREATE, 1);
1565 #endif /* HAVE_IOP_ATOMIC_OPEN */
1567 static int ll_symlink(struct inode *dir, struct dentry *dchild,
1568 const char *oldpath)
1570 struct qstr *name = &dchild->d_name;
1574 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p), target=%.*s\n",
1575 name->len, name->name, PFID(ll_inode2fid(dir)),
1576 dir, 3000, oldpath);
1578 err = ll_new_node(dir, dchild, oldpath, S_IFLNK | S_IRWXUGO, 0,
1579 LUSTRE_OPC_SYMLINK);
1582 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_SYMLINK, 1);
1587 static int ll_link(struct dentry *old_dentry, struct inode *dir,
1588 struct dentry *new_dentry)
1590 struct inode *src = old_dentry->d_inode;
1591 struct qstr *name = &new_dentry->d_name;
1592 struct ll_sb_info *sbi = ll_i2sbi(dir);
1593 struct ptlrpc_request *request = NULL;
1594 struct md_op_data *op_data;
1598 CDEBUG(D_VFSTRACE, "VFS Op: inode="DFID"(%p), dir="DFID"(%p), "
1599 "target=%.*s\n", PFID(ll_inode2fid(src)), src,
1600 PFID(ll_inode2fid(dir)), dir, name->len, name->name);
1602 op_data = ll_prep_md_op_data(NULL, src, dir, name->name, name->len,
1603 0, LUSTRE_OPC_ANY, NULL);
1604 if (IS_ERR(op_data))
1605 RETURN(PTR_ERR(op_data));
1607 err = md_link(sbi->ll_md_exp, op_data, &request);
1608 ll_finish_md_op_data(op_data);
1612 ll_update_times(request, dir);
1613 ll_stats_ops_tally(sbi, LPROC_LL_LINK, 1);
1616 ptlrpc_req_finished(request);
1620 static int ll_mkdir(struct inode *dir, struct dentry *dchild, ll_umode_t mode)
1622 struct qstr *name = &dchild->d_name;
1626 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1627 name->len, name->name, PFID(ll_inode2fid(dir)), dir);
1629 if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
1630 mode &= ~current_umask();
1632 mode = (mode & (S_IRWXUGO|S_ISVTX)) | S_IFDIR;
1634 err = ll_new_node(dir, dchild, NULL, mode, 0, LUSTRE_OPC_MKDIR);
1636 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKDIR, 1);
1641 static int ll_rmdir(struct inode *dir, struct dentry *dchild)
1643 struct qstr *name = &dchild->d_name;
1644 struct ptlrpc_request *request = NULL;
1645 struct md_op_data *op_data;
1649 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1650 name->len, name->name, PFID(ll_inode2fid(dir)), dir);
1652 if (unlikely(d_mountpoint(dchild)))
1655 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len,
1656 S_IFDIR, LUSTRE_OPC_ANY, NULL);
1657 if (IS_ERR(op_data))
1658 RETURN(PTR_ERR(op_data));
1660 if (dchild->d_inode != NULL)
1661 op_data->op_fid3 = *ll_inode2fid(dchild->d_inode);
1663 op_data->op_fid2 = op_data->op_fid3;
1664 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1665 ll_finish_md_op_data(op_data);
1667 ll_update_times(request, dir);
1668 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR, 1);
1671 ptlrpc_req_finished(request);
1678 int ll_rmdir_entry(struct inode *dir, char *name, int namelen)
1680 struct ptlrpc_request *request = NULL;
1681 struct md_op_data *op_data;
1685 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1686 namelen, name, PFID(ll_inode2fid(dir)), dir);
1688 op_data = ll_prep_md_op_data(NULL, dir, NULL, name, strlen(name),
1689 S_IFDIR, LUSTRE_OPC_ANY, NULL);
1690 if (IS_ERR(op_data))
1691 RETURN(PTR_ERR(op_data));
1692 op_data->op_cli_flags |= CLI_RM_ENTRY;
1693 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1694 ll_finish_md_op_data(op_data);
1696 ll_update_times(request, dir);
1697 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR, 1);
1700 ptlrpc_req_finished(request);
1704 static int ll_unlink(struct inode *dir, struct dentry *dchild)
1706 struct qstr *name = &dchild->d_name;
1707 struct ptlrpc_request *request = NULL;
1708 struct md_op_data *op_data;
1709 struct mdt_body *body;
1712 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1713 name->len, name->name, PFID(ll_inode2fid(dir)), dir);
1716 * XXX: unlink bind mountpoint maybe call to here,
1717 * just check it as vfs_unlink does.
1719 if (unlikely(d_mountpoint(dchild)))
1722 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len, 0,
1723 LUSTRE_OPC_ANY, NULL);
1724 if (IS_ERR(op_data))
1725 RETURN(PTR_ERR(op_data));
1727 op_data->op_fid3 = *ll_inode2fid(dchild->d_inode);
1729 op_data->op_fid2 = op_data->op_fid3;
1730 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1731 ll_finish_md_op_data(op_data);
1736 * The server puts attributes in on the last unlink, use them to update
1737 * the link count so the inode can be freed immediately.
1739 body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
1740 if (body->mbo_valid & OBD_MD_FLNLINK)
1741 set_nlink(dchild->d_inode, body->mbo_nlink);
1743 ll_update_times(request, dir);
1744 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_UNLINK, 1);
1747 ptlrpc_req_finished(request);
1751 static int ll_rename(struct inode *src, struct dentry *src_dchild,
1752 struct inode *tgt, struct dentry *tgt_dchild
1753 #ifdef HAVE_IOPS_RENAME_WITH_FLAGS
1754 , unsigned int flags
1758 struct qstr *src_name = &src_dchild->d_name;
1759 struct qstr *tgt_name = &tgt_dchild->d_name;
1760 struct ptlrpc_request *request = NULL;
1761 struct ll_sb_info *sbi = ll_i2sbi(src);
1762 struct md_op_data *op_data;
1766 #ifdef HAVE_IOPS_RENAME_WITH_FLAGS
1771 CDEBUG(D_VFSTRACE, "VFS Op:oldname=%.*s, src_dir="DFID
1772 "(%p), newname=%.*s, tgt_dir="DFID"(%p)\n",
1773 src_name->len, src_name->name,
1774 PFID(ll_inode2fid(src)), src, tgt_name->len,
1775 tgt_name->name, PFID(ll_inode2fid(tgt)), tgt);
1777 if (unlikely(d_mountpoint(src_dchild) || d_mountpoint(tgt_dchild)))
1780 op_data = ll_prep_md_op_data(NULL, src, tgt, NULL, 0, 0,
1781 LUSTRE_OPC_ANY, NULL);
1782 if (IS_ERR(op_data))
1783 RETURN(PTR_ERR(op_data));
1785 if (src_dchild->d_inode != NULL)
1786 op_data->op_fid3 = *ll_inode2fid(src_dchild->d_inode);
1788 if (tgt_dchild->d_inode != NULL)
1789 op_data->op_fid4 = *ll_inode2fid(tgt_dchild->d_inode);
1791 err = md_rename(sbi->ll_md_exp, op_data,
1792 src_name->name, src_name->len,
1793 tgt_name->name, tgt_name->len, &request);
1794 ll_finish_md_op_data(op_data);
1796 ll_update_times(request, src);
1797 ll_update_times(request, tgt);
1798 ll_stats_ops_tally(sbi, LPROC_LL_RENAME, 1);
1801 ptlrpc_req_finished(request);
1804 d_move(src_dchild, tgt_dchild);
1809 const struct inode_operations ll_dir_inode_operations = {
1811 #ifdef HAVE_IOP_ATOMIC_OPEN
1812 .atomic_open = ll_atomic_open,
1814 .lookup = ll_lookup_nd,
1815 .create = ll_create_nd,
1816 /* We need all these non-raw things for NFSD, to not patch it. */
1817 .unlink = ll_unlink,
1820 .symlink = ll_symlink,
1822 .rename = ll_rename,
1823 .setattr = ll_setattr,
1824 .getattr = ll_getattr,
1825 .permission = ll_inode_permission,
1826 #ifdef HAVE_IOP_XATTR
1827 .setxattr = ll_setxattr,
1828 .getxattr = ll_getxattr,
1829 .removexattr = ll_removexattr,
1831 .listxattr = ll_listxattr,
1832 #ifdef HAVE_IOP_GET_ACL
1833 .get_acl = ll_get_acl,
1835 #ifdef HAVE_IOP_SET_ACL
1836 .set_acl = ll_set_acl,
1840 const struct inode_operations ll_special_inode_operations = {
1841 .setattr = ll_setattr,
1842 .getattr = ll_getattr,
1843 .permission = ll_inode_permission,
1844 #ifdef HAVE_IOP_XATTR
1845 .setxattr = ll_setxattr,
1846 .getxattr = ll_getxattr,
1847 .removexattr = ll_removexattr,
1849 .listxattr = ll_listxattr,
1850 #ifdef HAVE_IOP_GET_ACL
1851 .get_acl = ll_get_acl,
1853 #ifdef HAVE_IOP_SET_ACL
1854 .set_acl = ll_set_acl,