4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2011, 2017, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
34 #include <linux/sched.h>
36 #include <linux/quotaops.h>
37 #include <linux/highmem.h>
38 #include <linux/pagemap.h>
39 #include <linux/security.h>
40 #include <linux/user_namespace.h>
41 #include <linux/uidgid.h>
43 #define DEBUG_SUBSYSTEM S_LLITE
45 #include <obd_support.h>
46 #include <lustre_fid.h>
47 #include <lustre_dlm.h>
48 #include "llite_internal.h"
50 static int ll_create_it(struct inode *dir, struct dentry *dentry,
51 struct lookup_intent *it,
52 void *secctx, __u32 secctxlen, bool encrypt,
53 void *encctx, __u32 encctxlen);
55 /* called from iget5_locked->find_inode() under inode_lock spinlock */
56 static int ll_test_inode(struct inode *inode, void *opaque)
58 struct ll_inode_info *lli = ll_i2info(inode);
59 struct lustre_md *md = opaque;
61 if (unlikely(!(md->body->mbo_valid & OBD_MD_FLID))) {
62 CERROR("MDS body missing FID\n");
66 if (!lu_fid_eq(&lli->lli_fid, &md->body->mbo_fid1))
72 static int ll_set_inode(struct inode *inode, void *opaque)
74 struct ll_inode_info *lli = ll_i2info(inode);
75 struct mdt_body *body = ((struct lustre_md *)opaque)->body;
77 if (unlikely(!(body->mbo_valid & OBD_MD_FLID))) {
78 CERROR("MDS body missing FID\n");
82 lli->lli_fid = body->mbo_fid1;
83 if (unlikely(!(body->mbo_valid & OBD_MD_FLTYPE))) {
84 CERROR("Can not initialize inode "DFID" without object type: "
86 PFID(&lli->lli_fid), body->mbo_valid);
90 inode->i_mode = (inode->i_mode & ~S_IFMT) | (body->mbo_mode & S_IFMT);
91 if (unlikely(inode->i_mode == 0)) {
92 CERROR("Invalid inode "DFID" type\n", PFID(&lli->lli_fid));
103 * Get an inode by inode number(@hash), which is already instantiated by
104 * the intent lookup).
106 struct inode *ll_iget(struct super_block *sb, ino_t hash,
107 struct lustre_md *md)
115 inode = iget5_locked(sb, hash, ll_test_inode, ll_set_inode, md);
117 RETURN(ERR_PTR(-ENOMEM));
119 if (inode->i_state & I_NEW) {
120 rc = ll_read_inode2(inode, md);
121 if (rc == 0 && S_ISREG(inode->i_mode) &&
122 ll_i2info(inode)->lli_clob == NULL)
123 rc = cl_file_inode_init(inode, md);
126 /* Let's clear directory lsm here, otherwise
127 * make_bad_inode() will reset the inode mode
128 * to regular, then ll_clear_inode will not
129 * be able to clear lsm_md */
130 if (S_ISDIR(inode->i_mode))
131 ll_dir_clear_lsm_md(inode);
132 make_bad_inode(inode);
133 unlock_new_inode(inode);
137 inode_has_no_xattr(inode);
138 unlock_new_inode(inode);
140 } else if (is_bad_inode(inode)) {
142 inode = ERR_PTR(-ESTALE);
143 } else if (!(inode->i_state & (I_FREEING | I_CLEAR))) {
144 rc = ll_update_inode(inode, md);
145 CDEBUG(D_VFSTRACE, "got inode: "DFID"(%p): rc = %d\n",
146 PFID(&md->body->mbo_fid1), inode, rc);
148 if (S_ISDIR(inode->i_mode))
149 ll_dir_clear_lsm_md(inode);
158 static void ll_invalidate_negative_children(struct inode *dir)
160 struct dentry *dentry, *tmp_subdir;
162 spin_lock(&dir->i_lock);
163 hlist_for_each_entry(dentry, &dir->i_dentry, d_alias) {
164 spin_lock(&dentry->d_lock);
165 if (!list_empty(&dentry->d_subdirs)) {
166 struct dentry *child;
168 list_for_each_entry_safe(child, tmp_subdir,
171 if (child->d_inode == NULL)
172 d_lustre_invalidate(child, 1);
175 spin_unlock(&dentry->d_lock);
177 spin_unlock(&dir->i_lock);
180 int ll_test_inode_by_fid(struct inode *inode, void *opaque)
182 return lu_fid_eq(&ll_i2info(inode)->lli_fid, opaque);
185 static int ll_dom_lock_cancel(struct inode *inode, struct ldlm_lock *lock)
188 struct ll_inode_info *lli = ll_i2info(inode);
193 if (!lli->lli_clob) {
194 /* due to DoM read on open, there may exist pages for Lustre
195 * regular file even though cl_object is not set up yet. */
196 truncate_inode_pages(inode->i_mapping, 0);
200 env = cl_env_get(&refcheck);
202 RETURN(PTR_ERR(env));
204 /* reach MDC layer to flush data under the DoM ldlm lock */
205 rc = cl_object_flush(env, lli->lli_clob, lock);
206 if (rc == -ENODATA) {
207 CDEBUG(D_INODE, "inode "DFID" layout has no DoM stripe\n",
208 PFID(ll_inode2fid(inode)));
209 /* most likely result of layout change, do nothing */
213 cl_env_put(env, &refcheck);
217 static void ll_lock_cancel_bits(struct ldlm_lock *lock, __u64 to_cancel)
219 struct inode *inode = ll_inode_from_resource_lock(lock);
220 struct ll_inode_info *lli;
221 __u64 bits = to_cancel;
227 /* That means the inode is evicted most likely and may cause
228 * the skipping of lock cleanups below, so print the message
231 if (lock->l_resource->lr_lvb_inode)
233 "can't take inode for the lock (%sevicted)\n",
234 lock->l_resource->lr_lvb_inode->i_state &
235 I_FREEING ? "" : "not ");
239 if (!fid_res_name_eq(ll_inode2fid(inode),
240 &lock->l_resource->lr_name)) {
241 LDLM_ERROR(lock, "data mismatch with object "DFID"(%p)",
242 PFID(ll_inode2fid(inode)), inode);
246 if (bits & MDS_INODELOCK_XATTR) {
247 ll_xattr_cache_destroy(inode);
248 bits &= ~MDS_INODELOCK_XATTR;
251 /* For OPEN locks we differentiate between lock modes
252 * LCK_CR, LCK_CW, LCK_PR - bug 22891 */
253 if (bits & MDS_INODELOCK_OPEN)
254 ll_have_md_lock(inode, &bits, lock->l_req_mode);
256 if (bits & MDS_INODELOCK_OPEN) {
259 switch (lock->l_req_mode) {
270 LDLM_ERROR(lock, "bad lock mode for OPEN lock");
274 ll_md_real_close(inode, fmode);
276 bits &= ~MDS_INODELOCK_OPEN;
279 if (bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
280 MDS_INODELOCK_LAYOUT | MDS_INODELOCK_PERM |
282 ll_have_md_lock(inode, &bits, LCK_MINMODE);
284 if (bits & MDS_INODELOCK_DOM) {
285 rc = ll_dom_lock_cancel(inode, lock);
287 CDEBUG(D_INODE, "cannot flush DoM data "
289 PFID(ll_inode2fid(inode)), rc);
292 if (bits & MDS_INODELOCK_LAYOUT) {
293 struct cl_object_conf conf = {
294 .coc_opc = OBJECT_CONF_INVALIDATE,
298 rc = ll_layout_conf(inode, &conf);
300 CDEBUG(D_INODE, "cannot invalidate layout of "
302 PFID(ll_inode2fid(inode)), rc);
305 lli = ll_i2info(inode);
307 if (bits & MDS_INODELOCK_UPDATE)
308 ll_file_set_flag(lli, LLIF_UPDATE_ATIME);
310 if ((bits & MDS_INODELOCK_UPDATE) && S_ISDIR(inode->i_mode)) {
311 CDEBUG(D_INODE, "invalidating inode "DFID" lli = %p, "
312 "pfid = "DFID"\n", PFID(ll_inode2fid(inode)),
313 lli, PFID(&lli->lli_pfid));
314 truncate_inode_pages(inode->i_mapping, 0);
316 if (unlikely(!fid_is_zero(&lli->lli_pfid))) {
317 struct inode *master_inode = NULL;
320 /* This is slave inode, since all of the child dentry
321 * is connected on the master inode, so we have to
322 * invalidate the negative children on master inode */
323 CDEBUG(D_INODE, "Invalidate s"DFID" m"DFID"\n",
324 PFID(ll_inode2fid(inode)), PFID(&lli->lli_pfid));
326 hash = cl_fid_build_ino(&lli->lli_pfid,
327 ll_need_32bit_api(ll_i2sbi(inode)));
329 /* Do not lookup the inode with ilookup5, otherwise
330 * it will cause dead lock,
331 * 1. Client1 send chmod req to the MDT0, then on MDT0,
332 * it enqueues master and all of its slaves lock,
333 * (mdt_attr_set() -> mdt_lock_slaves()), after gets
334 * master and stripe0 lock, it will send the enqueue
335 * req (for stripe1) to MDT1, then MDT1 finds the lock
336 * has been granted to client2. Then MDT1 sends blocking
338 * 2. At the same time, client2 tries to unlink
339 * the striped dir (rm -rf striped_dir), and during
340 * lookup, it will hold the master inode of the striped
341 * directory, whose inode state is NEW, then tries to
342 * revalidate all of its slaves, (ll_prep_inode()->
343 * ll_iget()->ll_read_inode2()-> ll_update_inode().).
344 * And it will be blocked on the server side because
346 * 3. Then the client get the blocking_ast req, cancel
347 * the lock, but being blocked if using ->ilookup5()),
348 * because master inode state is NEW. */
349 master_inode = ilookup5_nowait(inode->i_sb, hash,
350 ll_test_inode_by_fid,
351 (void *)&lli->lli_pfid);
353 ll_invalidate_negative_children(master_inode);
357 ll_invalidate_negative_children(inode);
361 if ((bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM)) &&
362 inode->i_sb->s_root != NULL &&
363 inode != inode->i_sb->s_root->d_inode)
364 ll_invalidate_aliases(inode);
366 if (bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM))
367 forget_all_cached_acls(inode);
373 /* Check if the given lock may be downgraded instead of canceling and
374 * that convert is really needed. */
375 int ll_md_need_convert(struct ldlm_lock *lock)
377 struct ldlm_namespace *ns = ldlm_lock_to_ns(lock);
379 __u64 wanted = lock->l_policy_data.l_inodebits.cancel_bits;
380 __u64 bits = lock->l_policy_data.l_inodebits.bits & ~wanted;
381 enum ldlm_mode mode = LCK_MINMODE;
383 if (!lock->l_conn_export ||
384 !exp_connect_lock_convert(lock->l_conn_export))
387 if (!wanted || !bits || ldlm_is_cancel(lock))
390 /* do not convert locks other than DOM for now */
391 if (!((bits | wanted) & MDS_INODELOCK_DOM))
394 /* We may have already remaining bits in some other lock so
395 * lock convert will leave us just extra lock for the same bit.
396 * Check if client has other lock with the same bits and the same
397 * or lower mode and don't convert if any.
399 switch (lock->l_req_mode) {
413 /* do not convert other modes */
417 /* is lock is too old to be converted? */
418 lock_res_and_lock(lock);
419 if (ktime_after(ktime_get(),
420 ktime_add(lock->l_last_used, ns->ns_dirty_age_limit))) {
421 unlock_res_and_lock(lock);
424 unlock_res_and_lock(lock);
426 inode = ll_inode_from_resource_lock(lock);
427 ll_have_md_lock(inode, &bits, mode);
432 int ll_md_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *ld,
433 void *data, int flag)
435 struct lustre_handle lockh;
441 case LDLM_CB_BLOCKING:
443 __u64 cancel_flags = LCF_ASYNC;
445 /* if lock convert is not needed then still have to
446 * pass lock via ldlm_cli_convert() to keep all states
447 * correct, set cancel_bits to full lock bits to cause
448 * full cancel to happen.
450 if (!ll_md_need_convert(lock)) {
451 lock_res_and_lock(lock);
452 lock->l_policy_data.l_inodebits.cancel_bits =
453 lock->l_policy_data.l_inodebits.bits;
454 unlock_res_and_lock(lock);
456 rc = ldlm_cli_convert(lock, cancel_flags);
459 /* continue with cancel otherwise */
460 ldlm_lock2handle(lock, &lockh);
461 rc = ldlm_cli_cancel(&lockh, cancel_flags);
463 CDEBUG(D_INODE, "ldlm_cli_cancel: rc = %d\n", rc);
468 case LDLM_CB_CANCELING:
470 __u64 to_cancel = lock->l_policy_data.l_inodebits.bits;
472 /* Nothing to do for non-granted locks */
473 if (!ldlm_is_granted(lock))
476 /* If 'ld' is supplied then bits to be cancelled are passed
477 * implicitly by lock converting and cancel_bits from 'ld'
478 * should be used. Otherwise full cancel is being performed
479 * and lock inodebits are used.
481 * Note: we cannot rely on cancel_bits in lock itself at this
482 * moment because they can be changed by concurrent thread,
483 * so ldlm_cli_inodebits_convert() pass cancel bits implicitly
487 /* partial bits cancel allowed only during convert */
488 LASSERT(ldlm_is_converting(lock));
489 /* mask cancel bits by lock bits so only no any unused
490 * bits are passed to ll_lock_cancel_bits()
492 to_cancel &= ld->l_policy_data.l_inodebits.cancel_bits;
494 ll_lock_cancel_bits(lock, to_cancel);
504 __u32 ll_i2suppgid(struct inode *i)
506 if (in_group_p(i->i_gid))
507 return (__u32)from_kgid(&init_user_ns, i->i_gid);
509 return (__u32) __kgid_val(INVALID_GID);
512 /* Pack the required supplementary groups into the supplied groups array.
513 * If we don't need to use the groups from the target inode(s) then we
514 * instead pack one or more groups from the user's supplementary group
515 * array in case it might be useful. Not needed if doing an MDS-side upcall. */
516 void ll_i2gids(__u32 *suppgids, struct inode *i1, struct inode *i2)
519 LASSERT(suppgids != NULL);
521 suppgids[0] = ll_i2suppgid(i1);
524 suppgids[1] = ll_i2suppgid(i2);
530 * try to reuse three types of dentry:
531 * 1. unhashed alias, this one is unhashed by d_invalidate (but it may be valid
532 * by concurrent .revalidate).
533 * 2. INVALID alias (common case for no valid ldlm lock held, but this flag may
534 * be cleared by others calling d_lustre_revalidate).
535 * 3. DISCONNECTED alias.
537 static struct dentry *ll_find_alias(struct inode *inode, struct dentry *dentry)
539 struct dentry *alias, *discon_alias, *invalid_alias;
541 if (hlist_empty(&inode->i_dentry))
544 discon_alias = invalid_alias = NULL;
546 spin_lock(&inode->i_lock);
547 hlist_for_each_entry(alias, &inode->i_dentry, d_alias) {
548 LASSERT(alias != dentry);
550 spin_lock(&alias->d_lock);
551 if ((alias->d_flags & DCACHE_DISCONNECTED) &&
552 S_ISDIR(inode->i_mode))
553 /* LASSERT(last_discon == NULL); LU-405, bz 20055 */
554 discon_alias = alias;
555 else if (alias->d_parent == dentry->d_parent &&
556 alias->d_name.hash == dentry->d_name.hash &&
557 alias->d_name.len == dentry->d_name.len &&
558 memcmp(alias->d_name.name, dentry->d_name.name,
559 dentry->d_name.len) == 0)
560 invalid_alias = alias;
561 spin_unlock(&alias->d_lock);
566 alias = invalid_alias ?: discon_alias ?: NULL;
568 spin_lock(&alias->d_lock);
570 spin_unlock(&alias->d_lock);
572 spin_unlock(&inode->i_lock);
578 * Similar to d_splice_alias(), but lustre treats invalid alias
579 * similar to DCACHE_DISCONNECTED, and tries to use it anyway.
581 struct dentry *ll_splice_alias(struct inode *inode, struct dentry *de)
587 new = ll_find_alias(inode, de);
597 "Reuse dentry %p inode %p refc %d flags %#x\n",
598 new, new->d_inode, ll_d_count(new), new->d_flags);
606 CDEBUG(D_DENTRY, "Add dentry %p inode %p refc %d flags %#x\n",
607 de, de->d_inode, ll_d_count(de), de->d_flags);
611 static int ll_lookup_it_finish(struct ptlrpc_request *request,
612 struct lookup_intent *it,
613 struct inode *parent, struct dentry **de,
614 void *secctx, __u32 secctxlen,
615 void *encctx, __u32 encctxlen,
616 ktime_t kstart, bool encrypt)
618 struct inode *inode = NULL;
621 struct dentry *alias;
624 /* NB 1 request reference will be taken away by ll_intent_lock()
626 CDEBUG(D_DENTRY, "it %p it_disposition %x\n", it,
628 if (!it_disposition(it, DISP_LOOKUP_NEG)) {
629 struct req_capsule *pill = &request->rq_pill;
630 struct mdt_body *body = req_capsule_server_get(pill,
633 rc = ll_prep_inode(&inode, request, (*de)->d_sb, it);
637 if (it->it_op & IT_OPEN)
638 ll_dom_finish_open(inode, request, it);
640 ll_set_lock_data(ll_i2sbi(parent)->ll_md_exp, inode, it, &bits);
642 /* We used to query real size from OSTs here, but actually
643 * this is not needed. For stat() calls size would be updated
644 * from subsequent do_revalidate()->ll_inode_revalidate_it() in
646 * vfs_getattr_it->ll_getattr()->ll_inode_revalidate_it() in 2.6
647 * Everybody else who needs correct file size would call
648 * ll_glimpse_size or some equivalent themselves anyway.
652 /* If security context was returned by MDT, put it in
653 * inode now to save an extra getxattr from security hooks,
654 * and avoid deadlock.
656 if (body->mbo_valid & OBD_MD_SECCTX) {
657 secctx = req_capsule_server_get(pill, &RMF_FILE_SECCTX);
658 secctxlen = req_capsule_get_size(pill,
663 CDEBUG(D_SEC, "server returned security context"
665 PFID(ll_inode2fid(inode)));
668 if (secctx != NULL && secctxlen != 0) {
669 /* no need to protect selinux_inode_setsecurity() by
670 * inode_lock. Taking it would lead to a client deadlock
673 rc = security_inode_notifysecctx(inode, secctx,
676 CWARN("%s: cannot set security context for "DFID": rc = %d\n",
677 ll_i2sbi(inode)->ll_fsname,
678 PFID(ll_inode2fid(inode)),
682 /* If encryption context was returned by MDT, put it in
683 * inode now to save an extra getxattr and avoid deadlock.
685 if (body->mbo_valid & OBD_MD_ENCCTX) {
686 encctx = req_capsule_server_get(pill, &RMF_FILE_ENCCTX);
687 encctxlen = req_capsule_get_size(pill,
693 "server returned encryption ctx for "DFID"\n",
694 PFID(ll_inode2fid(inode)));
695 rc = ll_xattr_cache_insert(inode,
696 LL_XATTR_NAME_ENCRYPTION_CONTEXT,
699 CWARN("%s: cannot set enc ctx for "DFID": rc = %d\n",
700 ll_i2sbi(inode)->ll_fsname,
701 PFID(ll_inode2fid(inode)), rc);
706 /* Only hash *de if it is unhashed (new dentry).
707 * Atoimc_open may passin hashed dentries for open.
709 alias = ll_splice_alias(inode, *de);
711 GOTO(out, rc = PTR_ERR(alias));
715 if (!it_disposition(it, DISP_LOOKUP_NEG)) {
716 /* we have lookup look - unhide dentry */
717 if (bits & MDS_INODELOCK_LOOKUP)
718 d_lustre_revalidate(*de);
721 rc = llcrypt_get_encryption_info(inode);
724 if (!llcrypt_has_encryption_key(inode))
725 GOTO(out, rc = -ENOKEY);
727 } else if (!it_disposition(it, DISP_OPEN_CREATE)) {
729 * If file was created on the server, the dentry is revalidated
730 * in ll_create_it if the lock allows for it.
732 /* Check that parent has UPDATE lock. */
733 struct lookup_intent parent_it = {
735 .it_lock_handle = 0 };
736 struct lu_fid fid = ll_i2info(parent)->lli_fid;
738 /* If it is striped directory, get the real stripe parent */
739 if (unlikely(ll_dir_striped(parent))) {
740 rc = md_get_fid_from_lsm(ll_i2mdexp(parent),
741 ll_i2info(parent)->lli_lsm_md,
743 (*de)->d_name.len, &fid);
748 if (md_revalidate_lock(ll_i2mdexp(parent), &parent_it, &fid,
750 d_lustre_revalidate(*de);
751 ll_intent_release(&parent_it);
755 if (it_disposition(it, DISP_OPEN_CREATE)) {
756 ll_stats_ops_tally(ll_i2sbi(parent), LPROC_LL_MKNOD,
757 ktime_us_delta(ktime_get(), kstart));
763 if (rc != 0 && it->it_op & IT_OPEN) {
764 ll_intent_drop_lock(it);
765 ll_open_cleanup((*de)->d_sb, request);
771 static struct dentry *ll_lookup_it(struct inode *parent, struct dentry *dentry,
772 struct lookup_intent *it,
773 void **secctx, __u32 *secctxlen,
774 struct pcc_create_attach *pca,
776 void **encctx, __u32 *encctxlen)
778 ktime_t kstart = ktime_get();
779 struct lookup_intent lookup_it = { .it_op = IT_LOOKUP };
780 struct dentry *save = dentry, *retval;
781 struct ptlrpc_request *req = NULL;
782 struct md_op_data *op_data = NULL;
783 struct lov_user_md *lum = NULL;
786 char secctx_name[XATTR_NAME_MAX + 1];
790 if (dentry->d_name.len > ll_i2sbi(parent)->ll_namelen)
791 RETURN(ERR_PTR(-ENAMETOOLONG));
793 CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p), intent=%s\n",
794 dentry, PFID(ll_inode2fid(parent)), parent, LL_IT2STR(it));
796 if (d_mountpoint(dentry))
797 CERROR("Tell Peter, lookup on mtpt, it %s\n", LL_IT2STR(it));
799 if (it == NULL || it->it_op == IT_GETXATTR)
802 if (it->it_op == IT_GETATTR && dentry_may_statahead(parent, dentry)) {
803 rc = ll_revalidate_statahead(parent, &dentry, 0);
805 RETURN(dentry == save ? NULL : dentry);
808 if (it->it_op & IT_OPEN && it->it_flags & FMODE_WRITE &&
809 dentry->d_sb->s_flags & SB_RDONLY)
810 RETURN(ERR_PTR(-EROFS));
812 if (it->it_op & IT_CREAT)
813 opc = LUSTRE_OPC_CREATE;
815 opc = LUSTRE_OPC_ANY;
817 op_data = ll_prep_md_op_data(NULL, parent, NULL, dentry->d_name.name,
818 dentry->d_name.len, 0, opc, NULL);
820 GOTO(out, retval = ERR_CAST(op_data));
822 /* enforce umask if acl disabled or MDS doesn't support umask */
823 if (!IS_POSIXACL(parent) || !exp_connect_umask(ll_i2mdexp(parent)))
824 it->it_create_mode &= ~current_umask();
826 if (it->it_op & IT_CREAT &&
827 ll_i2sbi(parent)->ll_flags & LL_SBI_FILE_SECCTX) {
828 rc = ll_dentry_init_security(dentry, it->it_create_mode,
830 &op_data->op_file_secctx_name,
831 &op_data->op_file_secctx,
832 &op_data->op_file_secctx_size);
834 GOTO(out, retval = ERR_PTR(rc));
836 *secctx = op_data->op_file_secctx;
837 if (secctxlen != NULL)
838 *secctxlen = op_data->op_file_secctx_size;
842 if (secctxlen != NULL)
845 if (it->it_op & IT_CREAT && encrypt) {
846 rc = llcrypt_inherit_context(parent, NULL, op_data, false);
848 GOTO(out, retval = ERR_PTR(rc));
850 *encctx = op_data->op_file_encctx;
851 if (encctxlen != NULL)
852 *encctxlen = op_data->op_file_encctx_size;
856 if (encctxlen != NULL)
860 /* ask for security context upon intent */
861 if (it->it_op & (IT_LOOKUP | IT_GETATTR | IT_OPEN)) {
862 /* get name of security xattr to request to server */
863 rc = ll_listsecurity(parent, secctx_name,
864 sizeof(secctx_name));
866 CDEBUG(D_SEC, "cannot get security xattr name for "
868 PFID(ll_inode2fid(parent)), rc);
870 op_data->op_file_secctx_name = secctx_name;
871 op_data->op_file_secctx_name_size = rc;
872 CDEBUG(D_SEC, "'%.*s' is security xattr for "DFID"\n",
873 rc, secctx_name, PFID(ll_inode2fid(parent)));
877 if (pca && pca->pca_dataset) {
878 struct pcc_dataset *dataset = pca->pca_dataset;
882 GOTO(out, retval = ERR_PTR(-ENOMEM));
884 lum->lmm_magic = LOV_USER_MAGIC_V1;
885 lum->lmm_pattern = LOV_PATTERN_F_RELEASED | LOV_PATTERN_RAID0;
886 op_data->op_data = lum;
887 op_data->op_data_size = sizeof(*lum);
888 op_data->op_archive_id = dataset->pccd_rwid;
890 rc = obd_fid_alloc(NULL, ll_i2mdexp(parent), &op_data->op_fid2,
893 GOTO(out, retval = ERR_PTR(rc));
895 rc = pcc_inode_create(parent->i_sb, dataset, &op_data->op_fid2,
898 GOTO(out, retval = ERR_PTR(rc));
900 it->it_flags |= MDS_OPEN_PCC;
903 rc = md_intent_lock(ll_i2mdexp(parent), op_data, it, &req,
904 &ll_md_blocking_ast, 0);
905 /* If the MDS allows the client to chgrp (CFS_SETGRP_PERM), but the
906 * client does not know which suppgid should be sent to the MDS, or
907 * some other(s) changed the target file's GID after this RPC sent
908 * to the MDS with the suppgid as the original GID, then we should
909 * try again with right suppgid. */
910 if (rc == -EACCES && it->it_op & IT_OPEN &&
911 it_disposition(it, DISP_OPEN_DENY)) {
912 struct mdt_body *body;
914 LASSERT(req != NULL);
916 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
917 if (op_data->op_suppgids[0] == body->mbo_gid ||
918 op_data->op_suppgids[1] == body->mbo_gid ||
919 !in_group_p(make_kgid(&init_user_ns, body->mbo_gid)))
920 GOTO(out, retval = ERR_PTR(-EACCES));
922 fid_zero(&op_data->op_fid2);
923 op_data->op_suppgids[1] = body->mbo_gid;
924 ptlrpc_req_finished(req);
926 ll_intent_release(it);
927 rc = md_intent_lock(ll_i2mdexp(parent), op_data, it, &req,
928 &ll_md_blocking_ast, 0);
932 GOTO(out, retval = ERR_PTR(rc));
934 /* dir layout may change */
935 ll_unlock_md_op_lsm(op_data);
936 rc = ll_lookup_it_finish(req, it, parent, &dentry,
937 secctx != NULL ? *secctx : NULL,
938 secctxlen != NULL ? *secctxlen : 0,
939 encctx != NULL ? *encctx : NULL,
940 encctxlen != NULL ? *encctxlen : 0,
943 ll_intent_release(it);
944 GOTO(out, retval = ERR_PTR(rc));
947 if ((it->it_op & IT_OPEN) && dentry->d_inode &&
948 !S_ISREG(dentry->d_inode->i_mode) &&
949 !S_ISDIR(dentry->d_inode->i_mode)) {
950 ll_release_openhandle(dentry, it);
952 ll_lookup_finish_locks(it, dentry);
954 GOTO(out, retval = (dentry == save) ? NULL : dentry);
957 if (op_data != NULL && !IS_ERR(op_data)) {
958 if (secctx != NULL && secctxlen != NULL) {
959 /* caller needs sec ctx info, so reset it in op_data to
960 * prevent it from being freed */
961 op_data->op_file_secctx = NULL;
962 op_data->op_file_secctx_size = 0;
964 if (encctx != NULL && encctxlen != NULL &&
965 it->it_op & IT_CREAT && encrypt) {
966 /* caller needs enc ctx info, so reset it in op_data to
967 * prevent it from being freed
969 op_data->op_file_encctx = NULL;
970 op_data->op_file_encctx_size = 0;
972 ll_finish_md_op_data(op_data);
978 ptlrpc_req_finished(req);
982 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
985 struct lookup_intent *itp, it = { .it_op = IT_GETATTR };
988 CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p), flags=%u\n",
989 dentry, PFID(ll_inode2fid(parent)), parent, flags);
992 * Optimize away (CREATE && !OPEN). Let .create handle the race.
993 * but only if we have write permissions there, otherwise we need
994 * to proceed with lookup. LU-4185
996 if ((flags & LOOKUP_CREATE) && !(flags & LOOKUP_OPEN) &&
997 (inode_permission(parent, MAY_WRITE | MAY_EXEC) == 0))
1000 if (flags & (LOOKUP_PARENT|LOOKUP_OPEN|LOOKUP_CREATE))
1004 de = ll_lookup_it(parent, dentry, itp, NULL, NULL, NULL, false,
1008 ll_intent_release(itp);
1013 #ifdef FMODE_CREATED /* added in Linux v4.18-rc1-20-g73a09dd */
1014 # define ll_is_opened(o, f) ((f)->f_mode & FMODE_OPENED)
1015 # define ll_finish_open(f, d, o) finish_open((f), (d), NULL)
1016 # define ll_last_arg
1017 # define ll_set_created(o, f) \
1019 (f)->f_mode |= FMODE_CREATED; \
1023 # define ll_is_opened(o, f) (*(o))
1024 # define ll_finish_open(f, d, o) finish_open((f), (d), NULL, (o))
1025 # define ll_last_arg , int *opened
1026 # define ll_set_created(o, f) \
1028 *(o) |= FILE_CREATED; \
1034 * For cached negative dentry and new dentry, handle lookup/create/open
1037 static int ll_atomic_open(struct inode *dir, struct dentry *dentry,
1038 struct file *file, unsigned open_flags,
1039 umode_t mode ll_last_arg)
1041 struct lookup_intent *it;
1043 long long lookup_flags = LOOKUP_OPEN;
1044 void *secctx = NULL;
1045 __u32 secctxlen = 0;
1046 void *encctx = NULL;
1047 __u32 encctxlen = 0;
1048 struct ll_sb_info *sbi = NULL;
1049 struct pcc_create_attach pca = { NULL, NULL };
1050 bool encrypt = false;
1055 "VFS Op:name=%pd, dir="DFID"(%p), file %p, open_flags %x, mode %x opened %d\n",
1056 dentry, PFID(ll_inode2fid(dir)), dir, file, open_flags, mode,
1057 ll_is_opened(opened, file));
1059 /* Only negative dentries enter here */
1060 LASSERT(dentry->d_inode == NULL);
1062 if (!d_unhashed(dentry)) {
1063 /* A valid negative dentry that just passed revalidation,
1064 * there's little point to try and open it server-side,
1065 * even though there's a minuscule chance it might succeed.
1066 * Either way it's a valid race to just return -ENOENT here.
1068 if (!(open_flags & O_CREAT))
1071 /* Otherwise we just unhash it to be rehashed afresh via
1072 * lookup if necessary
1077 OBD_ALLOC(it, sizeof(*it));
1081 it->it_op = IT_OPEN;
1082 if (open_flags & O_CREAT) {
1083 it->it_op |= IT_CREAT;
1084 lookup_flags |= LOOKUP_CREATE;
1085 sbi = ll_i2sbi(dir);
1086 /* Volatile file is used for HSM restore, so do not use PCC */
1087 if (!filename_is_volatile(dentry->d_name.name,
1088 dentry->d_name.len, NULL)) {
1089 struct pcc_matcher item;
1090 struct pcc_dataset *dataset;
1092 item.pm_uid = from_kuid(&init_user_ns, current_uid());
1093 item.pm_gid = from_kgid(&init_user_ns, current_gid());
1094 item.pm_projid = ll_i2info(dir)->lli_projid;
1095 item.pm_name = &dentry->d_name;
1096 dataset = pcc_dataset_match_get(&sbi->ll_pcc_super,
1098 pca.pca_dataset = dataset;
1101 it->it_create_mode = (mode & S_IALLUGO) | S_IFREG;
1102 it->it_flags = (open_flags & ~O_ACCMODE) | OPEN_FMODE(open_flags);
1103 it->it_flags &= ~MDS_OPEN_FL_INTERNAL;
1105 if (ll_sbi_has_encrypt(ll_i2sbi(dir)) && IS_ENCRYPTED(dir)) {
1106 /* we know that we are going to create a regular file because
1107 * we set S_IFREG bit on it->it_create_mode above
1109 rc = llcrypt_get_encryption_info(dir);
1111 GOTO(out_release, rc);
1112 if (!llcrypt_has_encryption_key(dir))
1113 GOTO(out_release, rc = -ENOKEY);
1118 /* Dentry added to dcache tree in ll_lookup_it */
1119 de = ll_lookup_it(dir, dentry, it, &secctx, &secctxlen, &pca, encrypt,
1120 &encctx, &encctxlen);
1123 else if (de != NULL)
1126 CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_FILE_PAUSE, cfs_fail_val);
1129 if (it_disposition(it, DISP_OPEN_CREATE)) {
1130 /* Dentry instantiated in ll_create_it. */
1131 rc = ll_create_it(dir, dentry, it, secctx, secctxlen,
1132 encrypt, encctx, encctxlen);
1133 security_release_secctx(secctx, secctxlen);
1134 llcrypt_free_ctx(encctx, encctxlen);
1136 /* We dget in ll_splice_alias. */
1142 rc = pcc_inode_create_fini(dentry->d_inode, &pca);
1146 GOTO(out_release, rc);
1149 ll_set_created(opened, file);
1151 /* Open the file with O_CREAT, but the file already
1152 * existed on MDT. This may happend in the case that
1153 * the LOOKUP ibits lock is revoked and the
1154 * corresponding dentry cache is deleted.
1155 * i.e. In the current Lustre, the truncate operation
1156 * will revoke the LOOKUP ibits lock, and the file
1157 * dentry cache will be invalidated. The following open
1158 * with O_CREAT flag will call into ->atomic_open, the
1159 * file was wrongly though as newly created file and
1160 * try to auto cache the file. So after client knows it
1161 * is not a DISP_OPEN_CREATE, it should cleanup the
1162 * already created PCC copy.
1164 pcc_create_attach_cleanup(dir->i_sb, &pca);
1166 if (open_flags & O_CREAT && encrypt &&
1168 rc = ll_set_encflags(dentry->d_inode, encctx,
1170 llcrypt_free_ctx(encctx, encctxlen);
1172 GOTO(out_release, rc);
1176 if (dentry->d_inode && it_disposition(it, DISP_OPEN_OPEN)) {
1178 if (S_ISFIFO(dentry->d_inode->i_mode)) {
1179 /* We cannot call open here as it might
1180 * deadlock. This case is unreachable in
1181 * practice because of OBD_CONNECT_NODEVOH. */
1182 rc = finish_no_open(file, de);
1184 file->private_data = it;
1185 rc = ll_finish_open(file, dentry, opened);
1186 /* We dget in ll_splice_alias. finish_open takes
1187 * care of dget for fd open.
1193 rc = finish_no_open(file, de);
1196 pcc_create_attach_cleanup(dir->i_sb, &pca);
1200 ll_intent_release(it);
1201 OBD_FREE(it, sizeof(*it));
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, bool encrypt,
1256 void *encctx, __u32 encctxlen)
1258 struct inode *inode;
1263 CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p), intent=%s\n",
1264 dentry, 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) &&
1276 /* must be done before d_instantiate, because it calls
1277 * security_d_instantiate, which means a getxattr if security
1278 * context is not set yet */
1279 /* no need to protect selinux_inode_setsecurity() by
1280 * inode_lock. Taking it would lead to a client deadlock
1283 rc = security_inode_notifysecctx(inode, secctx, secctxlen);
1288 d_instantiate(dentry, inode);
1291 rc = ll_set_encflags(inode, encctx, encctxlen, true);
1296 if (!(ll_i2sbi(inode)->ll_flags & LL_SBI_FILE_SECCTX)) {
1297 rc = ll_inode_init_security(dentry, inode, dir);
1302 ll_set_lock_data(ll_i2sbi(dir)->ll_md_exp, inode, it, &bits);
1303 if (bits & MDS_INODELOCK_LOOKUP)
1304 d_lustre_revalidate(dentry);
1309 void ll_update_times(struct ptlrpc_request *request, struct inode *inode)
1311 struct mdt_body *body = req_capsule_server_get(&request->rq_pill,
1315 if (body->mbo_valid & OBD_MD_FLMTIME &&
1316 body->mbo_mtime > inode->i_mtime.tv_sec) {
1318 "setting fid " DFID " mtime from %lld to %llu\n",
1319 PFID(ll_inode2fid(inode)),
1320 (s64)inode->i_mtime.tv_sec, body->mbo_mtime);
1321 inode->i_mtime.tv_sec = body->mbo_mtime;
1324 if (body->mbo_valid & OBD_MD_FLCTIME &&
1325 body->mbo_ctime > inode->i_ctime.tv_sec)
1326 inode->i_ctime.tv_sec = body->mbo_ctime;
1329 static int ll_new_node(struct inode *dir, struct dentry *dchild,
1330 const char *tgt, umode_t mode, int rdev, __u32 opc)
1332 struct qstr *name = &dchild->d_name;
1333 struct ptlrpc_request *request = NULL;
1334 struct md_op_data *op_data = NULL;
1335 struct inode *inode = NULL;
1336 struct ll_sb_info *sbi = ll_i2sbi(dir);
1338 bool encrypt = false;
1342 if (unlikely(tgt != NULL))
1343 tgt_len = strlen(tgt) + 1;
1346 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
1347 name->len, 0, opc, NULL);
1348 if (IS_ERR(op_data))
1349 GOTO(err_exit, err = PTR_ERR(op_data));
1351 if (sbi->ll_flags & LL_SBI_FILE_SECCTX) {
1352 err = ll_dentry_init_security(dchild, mode, &dchild->d_name,
1353 &op_data->op_file_secctx_name,
1354 &op_data->op_file_secctx,
1355 &op_data->op_file_secctx_size);
1357 GOTO(err_exit, err);
1360 if (ll_sbi_has_encrypt(sbi) &&
1361 ((IS_ENCRYPTED(dir) &&
1362 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) ||
1363 (unlikely(llcrypt_dummy_context_enabled(dir)) && S_ISDIR(mode)))) {
1364 err = llcrypt_get_encryption_info(dir);
1366 GOTO(err_exit, err);
1367 if (!llcrypt_has_encryption_key(dir))
1368 GOTO(err_exit, err = -ENOKEY);
1373 err = llcrypt_inherit_context(dir, NULL, op_data, false);
1375 GOTO(err_exit, err);
1378 err = md_create(sbi->ll_md_exp, op_data, tgt, tgt_len, mode,
1379 from_kuid(&init_user_ns, current_fsuid()),
1380 from_kgid(&init_user_ns, current_fsgid()),
1381 cfs_curproc_cap_pack(), rdev, &request);
1382 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 14, 58, 0)
1384 * server < 2.12.58 doesn't pack default LMV in intent_getattr reply,
1385 * fetch default LMV here.
1387 if (unlikely(err == -EREMOTE)) {
1388 struct ll_inode_info *lli = ll_i2info(dir);
1389 struct lmv_user_md *lum;
1393 ptlrpc_req_finished(request);
1396 err2 = ll_dir_getstripe(dir, (void **)&lum, &lumsize, &request,
1397 OBD_MD_DEFAULT_MEA);
1398 ll_finish_md_op_data(op_data);
1401 struct lustre_md md = { NULL };
1403 md.body = req_capsule_server_get(&request->rq_pill,
1406 GOTO(err_exit, err = -EPROTO);
1408 OBD_ALLOC_PTR(md.default_lmv);
1409 if (!md.default_lmv)
1410 GOTO(err_exit, err = -ENOMEM);
1412 md.default_lmv->lsm_md_magic = lum->lum_magic;
1413 md.default_lmv->lsm_md_stripe_count =
1414 lum->lum_stripe_count;
1415 md.default_lmv->lsm_md_master_mdt_index =
1416 lum->lum_stripe_offset;
1417 md.default_lmv->lsm_md_hash_type = lum->lum_hash_type;
1419 err = ll_update_inode(dir, &md);
1420 md_free_lustre_md(sbi->ll_md_exp, &md);
1422 GOTO(err_exit, err);
1423 } else if (err2 == -ENODATA && lli->lli_default_lsm_md) {
1425 * If there are no default stripe EA on the MDT, but the
1426 * client has default stripe, then it probably means
1427 * default stripe EA has just been deleted.
1429 down_write(&lli->lli_lsm_sem);
1430 if (lli->lli_default_lsm_md)
1431 OBD_FREE_PTR(lli->lli_default_lsm_md);
1432 lli->lli_default_lsm_md = NULL;
1433 up_write(&lli->lli_lsm_sem);
1435 GOTO(err_exit, err);
1438 ptlrpc_req_finished(request);
1445 GOTO(err_exit, err);
1447 ll_update_times(request, dir);
1449 CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_NEWNODE_PAUSE, cfs_fail_val);
1451 err = ll_prep_inode(&inode, request, dchild->d_sb, NULL);
1453 GOTO(err_exit, err);
1455 if (sbi->ll_flags & LL_SBI_FILE_SECCTX) {
1456 /* must be done before d_instantiate, because it calls
1457 * security_d_instantiate, which means a getxattr if security
1458 * context is not set yet */
1459 /* no need to protect selinux_inode_setsecurity() by
1460 * inode_lock. Taking it would lead to a client deadlock
1463 err = security_inode_notifysecctx(inode,
1464 op_data->op_file_secctx,
1465 op_data->op_file_secctx_size);
1467 GOTO(err_exit, err);
1470 d_instantiate(dchild, inode);
1473 err = ll_set_encflags(inode, op_data->op_file_encctx,
1474 op_data->op_file_encctx_size, true);
1476 GOTO(err_exit, err);
1479 if (!(sbi->ll_flags & LL_SBI_FILE_SECCTX)) {
1480 err = ll_inode_init_security(dchild, inode, dir);
1482 GOTO(err_exit, err);
1487 if (request != NULL)
1488 ptlrpc_req_finished(request);
1490 if (!IS_ERR_OR_NULL(op_data))
1491 ll_finish_md_op_data(op_data);
1496 static int ll_mknod(struct inode *dir, struct dentry *dchild, umode_t mode,
1499 ktime_t kstart = ktime_get();
1503 CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p) mode %o dev %x\n",
1504 dchild, PFID(ll_inode2fid(dir)), dir, mode, rdev);
1506 if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
1507 mode &= ~current_umask();
1509 switch (mode & S_IFMT) {
1518 err = ll_new_node(dir, dchild, NULL, mode, old_encode_dev(rdev),
1529 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKNOD,
1530 ktime_us_delta(ktime_get(), kstart));
1536 * Plain create. Intent create is handled in atomic_open.
1538 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
1539 umode_t mode, bool want_excl)
1541 ktime_t kstart = ktime_get();
1544 CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_FILE_PAUSE, cfs_fail_val);
1547 "VFS Op:name=%pd, dir="DFID"(%p), flags=%u, excl=%d\n",
1548 dentry, PFID(ll_inode2fid(dir)), dir, mode, want_excl);
1550 /* Using mknod(2) to create a regular file is designed to not recognize
1551 * volatile file name, so we use ll_mknod() here. */
1552 rc = ll_mknod(dir, dentry, mode, 0);
1554 CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, unhashed %d\n",
1555 dentry, d_unhashed(dentry));
1558 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_CREATE,
1559 ktime_us_delta(ktime_get(), kstart));
1564 static int ll_symlink(struct inode *dir, struct dentry *dchild,
1565 const char *oldpath)
1567 ktime_t kstart = ktime_get();
1571 CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p), target=%.*s\n",
1572 dchild, PFID(ll_inode2fid(dir)), dir, 3000, oldpath);
1574 err = ll_new_node(dir, dchild, oldpath, S_IFLNK | S_IRWXUGO, 0,
1575 LUSTRE_OPC_SYMLINK);
1578 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_SYMLINK,
1579 ktime_us_delta(ktime_get(), kstart));
1584 static int ll_link(struct dentry *old_dentry, struct inode *dir,
1585 struct dentry *new_dentry)
1587 struct inode *src = old_dentry->d_inode;
1588 struct qstr *name = &new_dentry->d_name;
1589 struct ll_sb_info *sbi = ll_i2sbi(dir);
1590 struct ptlrpc_request *request = NULL;
1591 struct md_op_data *op_data;
1592 ktime_t kstart = ktime_get();
1597 "VFS Op: inode="DFID"(%p), dir="DFID"(%p), target=%pd\n",
1598 PFID(ll_inode2fid(src)), src,
1599 PFID(ll_inode2fid(dir)), dir, new_dentry);
1601 err = llcrypt_prepare_link(old_dentry, dir, new_dentry);
1605 op_data = ll_prep_md_op_data(NULL, src, dir, name->name, name->len,
1606 0, LUSTRE_OPC_ANY, NULL);
1607 if (IS_ERR(op_data))
1608 RETURN(PTR_ERR(op_data));
1610 err = md_link(sbi->ll_md_exp, op_data, &request);
1611 ll_finish_md_op_data(op_data);
1615 ll_update_times(request, dir);
1616 ll_stats_ops_tally(sbi, LPROC_LL_LINK,
1617 ktime_us_delta(ktime_get(), kstart));
1620 ptlrpc_req_finished(request);
1624 static int ll_mkdir(struct inode *dir, struct dentry *dchild, umode_t mode)
1626 ktime_t kstart = ktime_get();
1630 CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p)\n",
1631 dchild, PFID(ll_inode2fid(dir)), dir);
1633 if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
1634 mode &= ~current_umask();
1636 mode = (mode & (S_IRWXUGO|S_ISVTX)) | S_IFDIR;
1638 err = ll_new_node(dir, dchild, NULL, mode, 0, LUSTRE_OPC_MKDIR);
1640 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKDIR,
1641 ktime_us_delta(ktime_get(), kstart));
1646 static int ll_rmdir(struct inode *dir, struct dentry *dchild)
1648 struct qstr *name = &dchild->d_name;
1649 struct ptlrpc_request *request = NULL;
1650 struct md_op_data *op_data;
1651 ktime_t kstart = ktime_get();
1656 CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p)\n",
1657 dchild, PFID(ll_inode2fid(dir)), dir);
1659 if (unlikely(d_mountpoint(dchild)))
1662 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len,
1663 S_IFDIR, LUSTRE_OPC_ANY, NULL);
1664 if (IS_ERR(op_data))
1665 RETURN(PTR_ERR(op_data));
1667 if (dchild->d_inode != NULL)
1668 op_data->op_fid3 = *ll_inode2fid(dchild->d_inode);
1670 op_data->op_fid2 = op_data->op_fid3;
1671 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1672 ll_finish_md_op_data(op_data);
1674 ll_update_times(request, dir);
1676 ptlrpc_req_finished(request);
1678 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR,
1679 ktime_us_delta(ktime_get(), kstart));
1686 int ll_rmdir_entry(struct inode *dir, char *name, int namelen)
1688 struct ptlrpc_request *request = NULL;
1689 struct md_op_data *op_data;
1690 ktime_t kstart = ktime_get();
1694 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1695 namelen, name, PFID(ll_inode2fid(dir)), dir);
1697 op_data = ll_prep_md_op_data(NULL, dir, NULL, name, strlen(name),
1698 S_IFDIR, LUSTRE_OPC_ANY, NULL);
1699 if (IS_ERR(op_data))
1700 RETURN(PTR_ERR(op_data));
1701 op_data->op_cli_flags |= CLI_RM_ENTRY;
1702 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1703 ll_finish_md_op_data(op_data);
1705 ll_update_times(request, dir);
1707 ptlrpc_req_finished(request);
1709 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR,
1710 ktime_us_delta(ktime_get(), kstart));
1714 static int ll_unlink(struct inode *dir, struct dentry *dchild)
1716 struct qstr *name = &dchild->d_name;
1717 struct ptlrpc_request *request = NULL;
1718 struct md_op_data *op_data;
1719 struct mdt_body *body;
1720 ktime_t kstart = ktime_get();
1725 CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p)\n",
1726 dchild, PFID(ll_inode2fid(dir)), dir);
1729 * XXX: unlink bind mountpoint maybe call to here,
1730 * just check it as vfs_unlink does.
1732 if (unlikely(d_mountpoint(dchild)))
1735 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len, 0,
1736 LUSTRE_OPC_ANY, NULL);
1737 if (IS_ERR(op_data))
1738 RETURN(PTR_ERR(op_data));
1740 op_data->op_fid3 = *ll_inode2fid(dchild->d_inode);
1741 /* notify lower layer if inode has dirty pages */
1742 if (S_ISREG(dchild->d_inode->i_mode) &&
1743 ll_i2info(dchild->d_inode)->lli_clob &&
1744 dirty_cnt(dchild->d_inode))
1745 op_data->op_cli_flags |= CLI_DIRTY_DATA;
1746 op_data->op_fid2 = op_data->op_fid3;
1747 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1748 ll_finish_md_op_data(op_data);
1753 * The server puts attributes in on the last unlink, use them to update
1754 * the link count so the inode can be freed immediately.
1756 body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
1757 if (body->mbo_valid & OBD_MD_FLNLINK)
1758 set_nlink(dchild->d_inode, body->mbo_nlink);
1760 ll_update_times(request, dir);
1763 ptlrpc_req_finished(request);
1765 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_UNLINK,
1766 ktime_us_delta(ktime_get(), kstart));
1770 static int ll_rename(struct inode *src, struct dentry *src_dchild,
1771 struct inode *tgt, struct dentry *tgt_dchild
1772 #ifdef HAVE_IOPS_RENAME_WITH_FLAGS
1773 , unsigned int flags
1777 struct qstr *src_name = &src_dchild->d_name;
1778 struct qstr *tgt_name = &tgt_dchild->d_name;
1779 struct ptlrpc_request *request = NULL;
1780 struct ll_sb_info *sbi = ll_i2sbi(src);
1781 struct md_op_data *op_data;
1782 ktime_t kstart = ktime_get();
1786 #ifdef HAVE_IOPS_RENAME_WITH_FLAGS
1792 "VFS Op:oldname=%pd, src_dir="DFID"(%p), newname=%pd, tgt_dir="DFID"(%p)\n",
1793 src_dchild, PFID(ll_inode2fid(src)), src,
1794 tgt_dchild, PFID(ll_inode2fid(tgt)), tgt);
1796 if (unlikely(d_mountpoint(src_dchild) || d_mountpoint(tgt_dchild)))
1799 #ifdef HAVE_IOPS_RENAME_WITH_FLAGS
1800 err = llcrypt_prepare_rename(src, src_dchild, tgt, tgt_dchild, flags);
1802 err = llcrypt_prepare_rename(src, src_dchild, tgt, tgt_dchild, 0);
1807 op_data = ll_prep_md_op_data(NULL, src, tgt, NULL, 0, 0,
1808 LUSTRE_OPC_ANY, NULL);
1809 if (IS_ERR(op_data))
1810 RETURN(PTR_ERR(op_data));
1812 if (src_dchild->d_inode != NULL)
1813 op_data->op_fid3 = *ll_inode2fid(src_dchild->d_inode);
1815 if (tgt_dchild->d_inode != NULL)
1816 op_data->op_fid4 = *ll_inode2fid(tgt_dchild->d_inode);
1818 err = md_rename(sbi->ll_md_exp, op_data,
1819 src_name->name, src_name->len,
1820 tgt_name->name, tgt_name->len, &request);
1821 ll_finish_md_op_data(op_data);
1823 ll_update_times(request, src);
1824 ll_update_times(request, tgt);
1827 ptlrpc_req_finished(request);
1830 d_move(src_dchild, tgt_dchild);
1831 ll_stats_ops_tally(sbi, LPROC_LL_RENAME,
1832 ktime_us_delta(ktime_get(), kstart));
1838 const struct inode_operations ll_dir_inode_operations = {
1840 .atomic_open = ll_atomic_open,
1841 .lookup = ll_lookup_nd,
1842 .create = ll_create_nd,
1843 /* We need all these non-raw things for NFSD, to not patch it. */
1844 .unlink = ll_unlink,
1847 .symlink = ll_symlink,
1849 .rename = ll_rename,
1850 .setattr = ll_setattr,
1851 .getattr = ll_getattr,
1852 .permission = ll_inode_permission,
1853 #ifdef HAVE_IOP_XATTR
1854 .setxattr = ll_setxattr,
1855 .getxattr = ll_getxattr,
1856 .removexattr = ll_removexattr,
1858 .listxattr = ll_listxattr,
1859 .get_acl = ll_get_acl,
1860 #ifdef HAVE_IOP_SET_ACL
1861 .set_acl = ll_set_acl,
1865 const struct inode_operations ll_special_inode_operations = {
1866 .setattr = ll_setattr,
1867 .getattr = ll_getattr,
1868 .permission = ll_inode_permission,
1869 #ifdef HAVE_IOP_XATTR
1870 .setxattr = ll_setxattr,
1871 .getxattr = ll_getxattr,
1872 .removexattr = ll_removexattr,
1874 .listxattr = ll_listxattr,
1875 .get_acl = ll_get_acl,
1876 #ifdef HAVE_IOP_SET_ACL
1877 .set_acl = ll_set_acl,