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/
33 #include <linux/sched.h>
35 #include <linux/file.h>
36 #include <linux/quotaops.h>
37 #include <linux/highmem.h>
38 #include <linux/pagemap.h>
39 #include <linux/user_namespace.h>
40 #include <linux/uidgid.h>
42 #define DEBUG_SUBSYSTEM S_LLITE
44 #include <obd_support.h>
45 #include <lustre_fid.h>
46 #include <lustre_dlm.h>
47 #include "llite_internal.h"
49 static int ll_create_it(struct inode *dir, struct dentry *dentry,
50 struct lookup_intent *it,
51 void *secctx, __u32 secctxlen, bool encrypt,
52 void *encctx, __u32 encctxlen);
54 /* called from iget5_locked->find_inode() under inode_lock spinlock */
55 static int ll_test_inode(struct inode *inode, void *opaque)
57 struct ll_inode_info *lli = ll_i2info(inode);
58 struct lustre_md *md = opaque;
60 if (unlikely(!(md->body->mbo_valid & OBD_MD_FLID))) {
61 CERROR("MDS body missing FID\n");
65 if (!lu_fid_eq(&lli->lli_fid, &md->body->mbo_fid1))
71 static int ll_set_inode(struct inode *inode, void *opaque)
73 struct ll_inode_info *lli = ll_i2info(inode);
74 struct mdt_body *body = ((struct lustre_md *)opaque)->body;
76 if (unlikely(!(body->mbo_valid & OBD_MD_FLID))) {
77 CERROR("MDS body missing FID\n");
81 lli->lli_fid = body->mbo_fid1;
82 if (unlikely(!(body->mbo_valid & OBD_MD_FLTYPE))) {
83 CERROR("Can not initialize inode "DFID" without object type: "
85 PFID(&lli->lli_fid), body->mbo_valid);
89 inode->i_mode = (inode->i_mode & ~S_IFMT) | (body->mbo_mode & S_IFMT);
90 if (unlikely(inode->i_mode == 0)) {
91 CERROR("Invalid inode "DFID" type\n", PFID(&lli->lli_fid));
102 * Get an inode by inode number(@hash), which is already instantiated by
103 * the intent lookup).
105 struct inode *ll_iget(struct super_block *sb, ino_t hash,
106 struct lustre_md *md)
114 inode = iget5_locked(sb, hash, ll_test_inode, ll_set_inode, md);
116 RETURN(ERR_PTR(-ENOMEM));
118 if (inode->i_state & I_NEW) {
119 rc = ll_read_inode2(inode, md);
120 if (rc == 0 && S_ISREG(inode->i_mode) &&
121 ll_i2info(inode)->lli_clob == NULL)
122 rc = cl_file_inode_init(inode, md);
125 /* Let's clear directory lsm here, otherwise
126 * make_bad_inode() will reset the inode mode
127 * to regular, then ll_clear_inode will not
128 * be able to clear lsm_md */
129 if (S_ISDIR(inode->i_mode))
130 ll_dir_clear_lsm_md(inode);
131 make_bad_inode(inode);
132 unlock_new_inode(inode);
136 inode_has_no_xattr(inode);
137 unlock_new_inode(inode);
139 } else if (is_bad_inode(inode)) {
141 inode = ERR_PTR(-ESTALE);
142 } else if (!(inode->i_state & (I_FREEING | I_CLEAR))) {
143 rc = ll_update_inode(inode, md);
144 CDEBUG(D_VFSTRACE, "got inode: "DFID"(%p): rc = %d\n",
145 PFID(&md->body->mbo_fid1), inode, rc);
147 if (S_ISDIR(inode->i_mode))
148 ll_dir_clear_lsm_md(inode);
157 /* mark negative sub file dentries invalid and prune unused dentries */
158 static void ll_prune_negative_children(struct inode *dir)
160 struct dentry *dentry;
161 struct dentry *child;
166 spin_lock(&dir->i_lock);
167 hlist_for_each_entry(dentry, &dir->i_dentry, d_alias) {
168 spin_lock(&dentry->d_lock);
169 list_for_each_entry(child, &dentry->d_subdirs, d_child) {
173 spin_lock_nested(&child->d_lock, DENTRY_D_LOCK_NESTED);
174 __d_lustre_invalidate(child);
175 if (!ll_d_count(child)) {
178 spin_unlock(&child->d_lock);
179 spin_unlock(&dentry->d_lock);
180 spin_unlock(&dir->i_lock);
182 CDEBUG(D_DENTRY, "prune negative dentry %pd\n",
188 spin_unlock(&child->d_lock);
190 spin_unlock(&dentry->d_lock);
192 spin_unlock(&dir->i_lock);
197 int ll_test_inode_by_fid(struct inode *inode, void *opaque)
199 return lu_fid_eq(&ll_i2info(inode)->lli_fid, opaque);
202 static int ll_dom_lock_cancel(struct inode *inode, struct ldlm_lock *lock)
205 struct ll_inode_info *lli = ll_i2info(inode);
210 env = cl_env_get(&refcheck);
212 RETURN(PTR_ERR(env));
214 OBD_FAIL_TIMEOUT(OBD_FAIL_LDLM_REPLAY_PAUSE, cfs_fail_val);
216 /* reach MDC layer to flush data under the DoM ldlm lock */
217 rc = cl_object_flush(env, lli->lli_clob, lock);
218 if (rc == -ENODATA) {
219 CDEBUG(D_INODE, "inode "DFID" layout has no DoM stripe\n",
220 PFID(ll_inode2fid(inode)));
221 /* most likely result of layout change, do nothing */
225 cl_env_put(env, &refcheck);
229 static void ll_lock_cancel_bits(struct ldlm_lock *lock, __u64 to_cancel)
231 struct inode *inode = ll_inode_from_resource_lock(lock);
232 struct ll_inode_info *lli;
233 __u64 bits = to_cancel;
239 /* That means the inode is evicted most likely and may cause
240 * the skipping of lock cleanups below, so print the message
243 if (lock->l_resource->lr_lvb_inode)
245 "can't take inode for the lock (%sevicted)\n",
246 lock->l_resource->lr_lvb_inode->i_state &
247 I_FREEING ? "" : "not ");
251 if (!fid_res_name_eq(ll_inode2fid(inode),
252 &lock->l_resource->lr_name)) {
253 LDLM_ERROR(lock, "data mismatch with object "DFID"(%p)",
254 PFID(ll_inode2fid(inode)), inode);
258 if (bits & MDS_INODELOCK_XATTR) {
259 ll_xattr_cache_destroy(inode);
260 bits &= ~MDS_INODELOCK_XATTR;
263 /* For OPEN locks we differentiate between lock modes
264 * LCK_CR, LCK_CW, LCK_PR - bug 22891 */
265 if (bits & MDS_INODELOCK_OPEN)
266 ll_have_md_lock(inode, &bits, lock->l_req_mode);
268 if (bits & MDS_INODELOCK_OPEN) {
271 switch (lock->l_req_mode) {
282 LDLM_ERROR(lock, "bad lock mode for OPEN lock");
286 ll_md_real_close(inode, fmode);
288 bits &= ~MDS_INODELOCK_OPEN;
291 if (bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
292 MDS_INODELOCK_LAYOUT | MDS_INODELOCK_PERM |
294 ll_have_md_lock(inode, &bits, LCK_MINMODE);
296 if (bits & MDS_INODELOCK_DOM) {
297 rc = ll_dom_lock_cancel(inode, lock);
299 CDEBUG(D_INODE, "cannot flush DoM data "
301 PFID(ll_inode2fid(inode)), rc);
304 if (bits & MDS_INODELOCK_LAYOUT) {
305 struct cl_object_conf conf = {
306 .coc_opc = OBJECT_CONF_INVALIDATE,
310 rc = ll_layout_conf(inode, &conf);
312 CDEBUG(D_INODE, "cannot invalidate layout of "
314 PFID(ll_inode2fid(inode)), rc);
317 lli = ll_i2info(inode);
319 if (bits & MDS_INODELOCK_UPDATE)
320 set_bit(LLIF_UPDATE_ATIME, &lli->lli_flags);
322 if ((bits & MDS_INODELOCK_UPDATE) && S_ISDIR(inode->i_mode)) {
323 CDEBUG(D_INODE, "invalidating inode "DFID" lli = %p, "
324 "pfid = "DFID"\n", PFID(ll_inode2fid(inode)),
325 lli, PFID(&lli->lli_pfid));
326 truncate_inode_pages(inode->i_mapping, 0);
328 if (unlikely(!fid_is_zero(&lli->lli_pfid))) {
329 struct inode *master_inode = NULL;
332 /* This is slave inode, since all of the child dentry
333 * is connected on the master inode, so we have to
334 * invalidate the negative children on master inode */
335 CDEBUG(D_INODE, "Invalidate s"DFID" m"DFID"\n",
336 PFID(ll_inode2fid(inode)), PFID(&lli->lli_pfid));
338 hash = cl_fid_build_ino(&lli->lli_pfid,
339 ll_need_32bit_api(ll_i2sbi(inode)));
341 /* Do not lookup the inode with ilookup5, otherwise
342 * it will cause dead lock,
343 * 1. Client1 send chmod req to the MDT0, then on MDT0,
344 * it enqueues master and all of its slaves lock,
345 * (mdt_attr_set() -> mdt_lock_slaves()), after gets
346 * master and stripe0 lock, it will send the enqueue
347 * req (for stripe1) to MDT1, then MDT1 finds the lock
348 * has been granted to client2. Then MDT1 sends blocking
350 * 2. At the same time, client2 tries to unlink
351 * the striped dir (rm -rf striped_dir), and during
352 * lookup, it will hold the master inode of the striped
353 * directory, whose inode state is NEW, then tries to
354 * revalidate all of its slaves, (ll_prep_inode()->
355 * ll_iget()->ll_read_inode2()-> ll_update_inode().).
356 * And it will be blocked on the server side because
358 * 3. Then the client get the blocking_ast req, cancel
359 * the lock, but being blocked if using ->ilookup5()),
360 * because master inode state is NEW. */
361 master_inode = ilookup5_nowait(inode->i_sb, hash,
362 ll_test_inode_by_fid,
363 (void *)&lli->lli_pfid);
365 ll_prune_negative_children(master_inode);
369 ll_prune_negative_children(inode);
373 if ((bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM)) &&
374 !is_root_inode(inode))
375 ll_prune_aliases(inode);
377 if (bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM))
378 forget_all_cached_acls(inode);
384 /* Check if the given lock may be downgraded instead of canceling and
385 * that convert is really needed. */
386 int ll_md_need_convert(struct ldlm_lock *lock)
388 struct ldlm_namespace *ns = ldlm_lock_to_ns(lock);
390 __u64 wanted = lock->l_policy_data.l_inodebits.cancel_bits;
391 __u64 bits = lock->l_policy_data.l_inodebits.bits & ~wanted;
392 enum ldlm_mode mode = LCK_MINMODE;
394 if (!lock->l_conn_export ||
395 !exp_connect_lock_convert(lock->l_conn_export))
398 if (!wanted || !bits || ldlm_is_cancel(lock))
401 /* do not convert locks other than DOM for now */
402 if (!((bits | wanted) & MDS_INODELOCK_DOM))
405 /* We may have already remaining bits in some other lock so
406 * lock convert will leave us just extra lock for the same bit.
407 * Check if client has other lock with the same bits and the same
408 * or lower mode and don't convert if any.
410 switch (lock->l_req_mode) {
424 /* do not convert other modes */
428 /* is lock is too old to be converted? */
429 lock_res_and_lock(lock);
430 if (ktime_after(ktime_get(),
431 ktime_add(lock->l_last_used, ns->ns_dirty_age_limit))) {
432 unlock_res_and_lock(lock);
435 unlock_res_and_lock(lock);
437 inode = ll_inode_from_resource_lock(lock);
438 ll_have_md_lock(inode, &bits, mode);
443 int ll_md_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *ld,
444 void *data, int flag)
446 struct lustre_handle lockh;
452 case LDLM_CB_BLOCKING:
454 __u64 cancel_flags = LCF_ASYNC;
456 /* if lock convert is not needed then still have to
457 * pass lock via ldlm_cli_convert() to keep all states
458 * correct, set cancel_bits to full lock bits to cause
459 * full cancel to happen.
461 if (!ll_md_need_convert(lock)) {
462 lock_res_and_lock(lock);
463 lock->l_policy_data.l_inodebits.cancel_bits =
464 lock->l_policy_data.l_inodebits.bits;
465 unlock_res_and_lock(lock);
467 rc = ldlm_cli_convert(lock, cancel_flags);
470 /* continue with cancel otherwise */
471 ldlm_lock2handle(lock, &lockh);
472 rc = ldlm_cli_cancel(&lockh, cancel_flags);
474 CDEBUG(D_INODE, "ldlm_cli_cancel: rc = %d\n", rc);
479 case LDLM_CB_CANCELING:
481 __u64 to_cancel = lock->l_policy_data.l_inodebits.bits;
483 /* Nothing to do for non-granted locks */
484 if (!ldlm_is_granted(lock))
487 /* If 'ld' is supplied then bits to be cancelled are passed
488 * implicitly by lock converting and cancel_bits from 'ld'
489 * should be used. Otherwise full cancel is being performed
490 * and lock inodebits are used.
492 * Note: we cannot rely on cancel_bits in lock itself at this
493 * moment because they can be changed by concurrent thread,
494 * so ldlm_cli_inodebits_convert() pass cancel bits implicitly
498 /* partial bits cancel allowed only during convert */
499 LASSERT(ldlm_is_converting(lock));
500 /* mask cancel bits by lock bits so only no any unused
501 * bits are passed to ll_lock_cancel_bits()
503 to_cancel &= ld->l_policy_data.l_inodebits.cancel_bits;
505 ll_lock_cancel_bits(lock, to_cancel);
515 __u32 ll_i2suppgid(struct inode *i)
517 if (in_group_p(i->i_gid))
518 return (__u32)from_kgid(&init_user_ns, i->i_gid);
520 return (__u32) __kgid_val(INVALID_GID);
523 /* Pack the required supplementary groups into the supplied groups array.
524 * If we don't need to use the groups from the target inode(s) then we
525 * instead pack one or more groups from the user's supplementary group
526 * array in case it might be useful. Not needed if doing an MDS-side upcall. */
527 void ll_i2gids(__u32 *suppgids, struct inode *i1, struct inode *i2)
530 LASSERT(suppgids != NULL);
532 suppgids[0] = ll_i2suppgid(i1);
535 suppgids[1] = ll_i2suppgid(i2);
541 * try to reuse three types of dentry:
542 * 1. unhashed alias, this one is unhashed by d_invalidate (but it may be valid
543 * by concurrent .revalidate).
544 * 2. INVALID alias (common case for no valid ldlm lock held, but this flag may
545 * be cleared by others calling d_lustre_revalidate).
546 * 3. DISCONNECTED alias.
548 static struct dentry *ll_find_alias(struct inode *inode, struct dentry *dentry)
550 struct dentry *alias, *discon_alias, *invalid_alias;
552 if (hlist_empty(&inode->i_dentry))
555 discon_alias = invalid_alias = NULL;
557 spin_lock(&inode->i_lock);
558 hlist_for_each_entry(alias, &inode->i_dentry, d_alias) {
559 LASSERT(alias != dentry);
561 spin_lock(&alias->d_lock);
562 if ((alias->d_flags & DCACHE_DISCONNECTED) &&
563 S_ISDIR(inode->i_mode))
564 /* LASSERT(last_discon == NULL); LU-405, bz 20055 */
565 discon_alias = alias;
566 else if (alias->d_parent == dentry->d_parent &&
567 alias->d_name.hash == dentry->d_name.hash &&
568 alias->d_name.len == dentry->d_name.len &&
569 memcmp(alias->d_name.name, dentry->d_name.name,
570 dentry->d_name.len) == 0)
571 invalid_alias = alias;
572 spin_unlock(&alias->d_lock);
577 alias = invalid_alias ?: discon_alias ?: NULL;
579 spin_lock(&alias->d_lock);
581 spin_unlock(&alias->d_lock);
583 spin_unlock(&inode->i_lock);
589 * Similar to d_splice_alias(), but lustre treats invalid alias
590 * similar to DCACHE_DISCONNECTED, and tries to use it anyway.
592 struct dentry *ll_splice_alias(struct inode *inode, struct dentry *de)
598 new = ll_find_alias(inode, de);
608 "Reuse dentry %p inode %p refc %d flags %#x\n",
609 new, new->d_inode, ll_d_count(new), new->d_flags);
618 /* this needs only to be done for foreign symlink dirs as
619 * DCACHE_SYMLINK_TYPE is already set by d_flags_for_inode()
620 * kernel routine for files with symlink ops (ie, real symlink)
622 if (inode && S_ISDIR(inode->i_mode) &&
623 ll_sbi_has_foreign_symlink(ll_i2sbi(inode)) &&
624 #ifdef HAVE_IOP_GET_LINK
625 inode->i_op->get_link) {
627 inode->i_op->follow_link) {
629 CDEBUG(D_INFO, "%s: inode "DFID": faking foreign dir as a symlink\n",
630 ll_i2sbi(inode)->ll_fsname, PFID(ll_inode2fid(inode)));
631 spin_lock(&de->d_lock);
632 /* like d_flags_for_inode() already does for files */
633 de->d_flags = (de->d_flags & ~DCACHE_ENTRY_TYPE) |
635 spin_unlock(&de->d_lock);
638 CDEBUG(D_DENTRY, "Add dentry %p inode %p refc %d flags %#x\n",
639 de, de->d_inode, ll_d_count(de), de->d_flags);
643 static int ll_lookup_it_finish(struct ptlrpc_request *request,
644 struct lookup_intent *it,
645 struct inode *parent, struct dentry **de,
646 void *secctx, __u32 secctxlen,
647 void *encctx, __u32 encctxlen,
648 ktime_t kstart, bool encrypt)
650 struct inode *inode = NULL;
653 struct dentry *alias;
656 /* NB 1 request reference will be taken away by ll_intent_lock()
658 CDEBUG(D_DENTRY, "it %p it_disposition %x\n", it,
660 if (!it_disposition(it, DISP_LOOKUP_NEG)) {
661 struct req_capsule *pill = &request->rq_pill;
662 struct mdt_body *body = req_capsule_server_get(pill,
665 rc = ll_prep_inode(&inode, &request->rq_pill, (*de)->d_sb, it);
669 /* If encryption context was returned by MDT, put it in
670 * inode now to save an extra getxattr and avoid deadlock.
672 if (body->mbo_valid & OBD_MD_ENCCTX) {
673 encctx = req_capsule_server_get(pill, &RMF_FILE_ENCCTX);
674 encctxlen = req_capsule_get_size(pill,
680 "server returned encryption ctx for "DFID"\n",
681 PFID(ll_inode2fid(inode)));
682 rc = ll_xattr_cache_insert(inode,
683 LL_XATTR_NAME_ENCRYPTION_CONTEXT,
686 CWARN("%s: cannot set enc ctx for "DFID": rc = %d\n",
687 ll_i2sbi(inode)->ll_fsname,
688 PFID(ll_inode2fid(inode)), rc);
690 rc = llcrypt_get_encryption_info(inode);
693 "cannot get enc info for "DFID": rc = %d\n",
694 PFID(ll_inode2fid(inode)), rc);
699 ll_set_lock_data(ll_i2sbi(parent)->ll_md_exp, inode, it, &bits);
700 /* OPEN can return data if lock has DoM+LAYOUT bits set */
701 if (it->it_op & IT_OPEN &&
702 bits & MDS_INODELOCK_DOM && bits & MDS_INODELOCK_LAYOUT)
703 ll_dom_finish_open(inode, request);
705 /* We used to query real size from OSTs here, but actually
706 * this is not needed. For stat() calls size would be updated
707 * from subsequent do_revalidate()->ll_inode_revalidate_it() in
709 * vfs_getattr_it->ll_getattr()->ll_inode_revalidate_it() in 2.6
710 * Everybody else who needs correct file size would call
711 * ll_glimpse_size or some equivalent themselves anyway.
715 /* If security context was returned by MDT, put it in
716 * inode now to save an extra getxattr from security hooks,
717 * and avoid deadlock.
719 if (body->mbo_valid & OBD_MD_SECCTX) {
720 secctx = req_capsule_server_get(pill, &RMF_FILE_SECCTX);
721 secctxlen = req_capsule_get_size(pill,
726 CDEBUG(D_SEC, "server returned security context"
728 PFID(ll_inode2fid(inode)));
731 if (secctx != NULL && secctxlen != 0) {
732 /* no need to protect selinux_inode_setsecurity() by
733 * inode_lock. Taking it would lead to a client deadlock
736 rc = security_inode_notifysecctx(inode, secctx,
739 CWARN("%s: cannot set security context for "DFID": rc = %d\n",
740 ll_i2sbi(inode)->ll_fsname,
741 PFID(ll_inode2fid(inode)),
746 /* Only hash *de if it is unhashed (new dentry).
747 * Atoimc_open may passin hashed dentries for open.
749 alias = ll_splice_alias(inode, *de);
751 GOTO(out, rc = PTR_ERR(alias));
755 if (!it_disposition(it, DISP_LOOKUP_NEG)) {
756 /* we have lookup look - unhide dentry */
757 if (bits & MDS_INODELOCK_LOOKUP) {
758 d_lustre_revalidate(*de);
759 ll_update_dir_depth(parent, (*de)->d_inode);
763 rc = llcrypt_get_encryption_info(inode);
766 if (!llcrypt_has_encryption_key(inode))
767 GOTO(out, rc = -ENOKEY);
769 } else if (!it_disposition(it, DISP_OPEN_CREATE)) {
771 * If file was created on the server, the dentry is revalidated
772 * in ll_create_it if the lock allows for it.
774 /* Check that parent has UPDATE lock. */
775 struct lookup_intent parent_it = {
777 .it_lock_handle = 0 };
778 struct lu_fid fid = ll_i2info(parent)->lli_fid;
780 /* If it is striped directory, get the real stripe parent */
781 if (unlikely(ll_dir_striped(parent))) {
782 rc = md_get_fid_from_lsm(ll_i2mdexp(parent),
783 ll_i2info(parent)->lli_lsm_md,
785 (*de)->d_name.len, &fid);
790 if (md_revalidate_lock(ll_i2mdexp(parent), &parent_it, &fid,
792 d_lustre_revalidate(*de);
793 ll_intent_release(&parent_it);
797 if (it_disposition(it, DISP_OPEN_CREATE)) {
798 ll_stats_ops_tally(ll_i2sbi(parent), LPROC_LL_MKNOD,
799 ktime_us_delta(ktime_get(), kstart));
805 if (rc != 0 && it->it_op & IT_OPEN) {
806 ll_intent_drop_lock(it);
807 ll_open_cleanup((*de)->d_sb, &request->rq_pill);
813 static struct dentry *ll_lookup_it(struct inode *parent, struct dentry *dentry,
814 struct lookup_intent *it,
815 void **secctx, __u32 *secctxlen,
816 struct pcc_create_attach *pca,
818 void **encctx, __u32 *encctxlen)
820 ktime_t kstart = ktime_get();
821 struct lookup_intent lookup_it = { .it_op = IT_LOOKUP };
822 struct dentry *save = dentry, *retval;
823 struct ptlrpc_request *req = NULL;
824 struct md_op_data *op_data = NULL;
825 struct lov_user_md *lum = NULL;
828 char secctx_name[XATTR_NAME_MAX + 1];
832 if (dentry->d_name.len > ll_i2sbi(parent)->ll_namelen)
833 RETURN(ERR_PTR(-ENAMETOOLONG));
835 CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p), intent=%s\n",
836 dentry, PFID(ll_inode2fid(parent)), parent, LL_IT2STR(it));
838 if (d_mountpoint(dentry))
839 CERROR("Tell Peter, lookup on mtpt, it %s\n", LL_IT2STR(it));
841 if (it == NULL || it->it_op == IT_GETXATTR)
844 if (it->it_op == IT_GETATTR && dentry_may_statahead(parent, dentry)) {
845 rc = ll_revalidate_statahead(parent, &dentry, 0);
847 RETURN(dentry == save ? NULL : dentry);
850 if (it->it_op & IT_OPEN && it->it_flags & FMODE_WRITE &&
851 dentry->d_sb->s_flags & SB_RDONLY)
852 RETURN(ERR_PTR(-EROFS));
854 if (it->it_op & IT_CREAT)
855 opc = LUSTRE_OPC_CREATE;
857 opc = LUSTRE_OPC_ANY;
859 op_data = ll_prep_md_op_data(NULL, parent, NULL, dentry->d_name.name,
860 dentry->d_name.len, 0, opc, NULL);
862 GOTO(out, retval = ERR_CAST(op_data));
864 /* enforce umask if acl disabled or MDS doesn't support umask */
865 if (!IS_POSIXACL(parent) || !exp_connect_umask(ll_i2mdexp(parent)))
866 it->it_create_mode &= ~current_umask();
868 if (it->it_op & IT_CREAT &&
869 ll_i2sbi(parent)->ll_flags & LL_SBI_FILE_SECCTX) {
870 rc = ll_dentry_init_security(dentry, it->it_create_mode,
872 &op_data->op_file_secctx_name,
873 &op_data->op_file_secctx,
874 &op_data->op_file_secctx_size);
876 GOTO(out, retval = ERR_PTR(rc));
878 *secctx = op_data->op_file_secctx;
879 if (secctxlen != NULL)
880 *secctxlen = op_data->op_file_secctx_size;
884 if (secctxlen != NULL)
887 if (it->it_op & IT_CREAT && encrypt) {
888 /* Volatile file name may look like:
889 * <parent>/LUSTRE_VOLATILE_HDR:<mdt_index>:<random>:fd=<fd>
890 * where fd is opened descriptor of reference file.
892 if (unlikely(filename_is_volatile(dentry->d_name.name,
893 dentry->d_name.len, NULL))) {
894 int ctx_size = LLCRYPT_ENC_CTX_SIZE;
895 struct lustre_sb_info *lsi;
896 struct file *ref_file;
897 struct inode *ref_inode;
898 char *p, *q, *fd_str;
902 p = strnstr(dentry->d_name.name, ":fd=",
904 if (!p || strlen(p + 4) == 0)
905 GOTO(out, retval = ERR_PTR(-EINVAL));
907 q = strchrnul(p + 4, ':');
908 fd_str = kstrndup(p + 4, q - p - 4, GFP_NOFS);
910 GOTO(out, retval = ERR_PTR(-ENOMEM));
911 rc = kstrtouint(fd_str, 10, &fd);
914 GOTO(inherit, rc = -EINVAL);
918 GOTO(inherit, rc = -EINVAL);
920 ref_inode = file_inode(ref_file);
923 GOTO(inherit, rc = -EINVAL);
926 lsi = s2lsi(ref_inode->i_sb);
929 OBD_ALLOC(ctx, ctx_size);
931 GOTO(out, retval = ERR_PTR(-ENOMEM));
933 #ifdef CONFIG_LL_ENCRYPTION
934 rc = lsi->lsi_cop->get_context(ref_inode,
940 OBD_FREE(ctx, ctx_size);
946 OBD_FREE(ctx, ctx_size);
950 op_data->op_file_encctx_size = rc;
951 if (rc == ctx_size) {
952 op_data->op_file_encctx = ctx;
954 OBD_ALLOC(op_data->op_file_encctx,
955 op_data->op_file_encctx_size);
956 if (!op_data->op_file_encctx) {
957 OBD_FREE(ctx, ctx_size);
958 GOTO(out, retval = ERR_PTR(-ENOMEM));
960 memcpy(op_data->op_file_encctx, ctx,
961 op_data->op_file_encctx_size);
962 OBD_FREE(ctx, ctx_size);
967 rc = llcrypt_inherit_context(parent, NULL, op_data,
970 GOTO(out, retval = ERR_PTR(rc));
973 *encctx = op_data->op_file_encctx;
974 if (encctxlen != NULL)
975 *encctxlen = op_data->op_file_encctx_size;
979 if (encctxlen != NULL)
983 /* ask for security context upon intent */
984 if (it->it_op & (IT_LOOKUP | IT_GETATTR | IT_OPEN)) {
985 /* get name of security xattr to request to server */
986 rc = ll_listsecurity(parent, secctx_name,
987 sizeof(secctx_name));
989 CDEBUG(D_SEC, "cannot get security xattr name for "
991 PFID(ll_inode2fid(parent)), rc);
993 op_data->op_file_secctx_name = secctx_name;
994 op_data->op_file_secctx_name_size = rc;
995 CDEBUG(D_SEC, "'%.*s' is security xattr for "DFID"\n",
996 rc, secctx_name, PFID(ll_inode2fid(parent)));
1000 if (pca && pca->pca_dataset) {
1003 GOTO(out, retval = ERR_PTR(-ENOMEM));
1005 lum->lmm_magic = LOV_USER_MAGIC_V1;
1006 lum->lmm_pattern = LOV_PATTERN_F_RELEASED | LOV_PATTERN_RAID0;
1007 op_data->op_data = lum;
1008 op_data->op_data_size = sizeof(*lum);
1009 op_data->op_archive_id = pca->pca_dataset->pccd_rwid;
1010 it->it_flags |= MDS_OPEN_PCC;
1013 rc = md_intent_lock(ll_i2mdexp(parent), op_data, it, &req,
1014 &ll_md_blocking_ast, 0);
1015 /* If the MDS allows the client to chgrp (CFS_SETGRP_PERM), but the
1016 * client does not know which suppgid should be sent to the MDS, or
1017 * some other(s) changed the target file's GID after this RPC sent
1018 * to the MDS with the suppgid as the original GID, then we should
1019 * try again with right suppgid. */
1020 if (rc == -EACCES && it->it_op & IT_OPEN &&
1021 it_disposition(it, DISP_OPEN_DENY)) {
1022 struct mdt_body *body;
1024 LASSERT(req != NULL);
1026 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1027 if (op_data->op_suppgids[0] == body->mbo_gid ||
1028 op_data->op_suppgids[1] == body->mbo_gid ||
1029 !in_group_p(make_kgid(&init_user_ns, body->mbo_gid)))
1030 GOTO(out, retval = ERR_PTR(-EACCES));
1032 fid_zero(&op_data->op_fid2);
1033 op_data->op_suppgids[1] = body->mbo_gid;
1034 ptlrpc_req_finished(req);
1036 ll_intent_release(it);
1037 rc = md_intent_lock(ll_i2mdexp(parent), op_data, it, &req,
1038 &ll_md_blocking_ast, 0);
1042 GOTO(out, retval = ERR_PTR(rc));
1044 if (pca && pca->pca_dataset) {
1045 rc = pcc_inode_create(parent->i_sb, pca->pca_dataset,
1049 GOTO(out, retval = ERR_PTR(rc));
1052 /* dir layout may change */
1053 ll_unlock_md_op_lsm(op_data);
1054 rc = ll_lookup_it_finish(req, it, parent, &dentry,
1055 secctx != NULL ? *secctx : NULL,
1056 secctxlen != NULL ? *secctxlen : 0,
1057 encctx != NULL ? *encctx : NULL,
1058 encctxlen != NULL ? *encctxlen : 0,
1061 ll_intent_release(it);
1062 GOTO(out, retval = ERR_PTR(rc));
1065 if ((it->it_op & IT_OPEN) && dentry->d_inode &&
1066 !S_ISREG(dentry->d_inode->i_mode) &&
1067 !S_ISDIR(dentry->d_inode->i_mode)) {
1068 ll_release_openhandle(dentry, it);
1070 ll_lookup_finish_locks(it, dentry);
1072 GOTO(out, retval = (dentry == save) ? NULL : dentry);
1075 if (op_data != NULL && !IS_ERR(op_data)) {
1076 if (secctx != NULL && secctxlen != NULL) {
1077 /* caller needs sec ctx info, so reset it in op_data to
1078 * prevent it from being freed */
1079 op_data->op_file_secctx = NULL;
1080 op_data->op_file_secctx_size = 0;
1082 if (encctx != NULL && encctxlen != NULL &&
1083 it->it_op & IT_CREAT && encrypt) {
1084 /* caller needs enc ctx info, so reset it in op_data to
1085 * prevent it from being freed
1087 op_data->op_file_encctx = NULL;
1088 op_data->op_file_encctx_size = 0;
1090 ll_finish_md_op_data(op_data);
1096 ptlrpc_req_finished(req);
1100 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
1103 struct lookup_intent *itp, it = { .it_op = IT_GETATTR };
1106 CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p), flags=%u\n",
1107 dentry, PFID(ll_inode2fid(parent)), parent, flags);
1110 * Optimize away (CREATE && !OPEN). Let .create handle the race.
1111 * but only if we have write permissions there, otherwise we need
1112 * to proceed with lookup. LU-4185
1114 if ((flags & LOOKUP_CREATE) && !(flags & LOOKUP_OPEN) &&
1115 (inode_permission(parent, MAY_WRITE | MAY_EXEC) == 0))
1118 if (flags & (LOOKUP_PARENT|LOOKUP_OPEN|LOOKUP_CREATE))
1122 de = ll_lookup_it(parent, dentry, itp, NULL, NULL, NULL, false,
1126 ll_intent_release(itp);
1131 #ifdef FMODE_CREATED /* added in Linux v4.18-rc1-20-g73a09dd */
1132 # define ll_is_opened(o, f) ((f)->f_mode & FMODE_OPENED)
1133 # define ll_finish_open(f, d, o) finish_open((f), (d), NULL)
1134 # define ll_last_arg
1135 # define ll_set_created(o, f) \
1137 (f)->f_mode |= FMODE_CREATED; \
1141 # define ll_is_opened(o, f) (*(o))
1142 # define ll_finish_open(f, d, o) finish_open((f), (d), NULL, (o))
1143 # define ll_last_arg , int *opened
1144 # define ll_set_created(o, f) \
1146 *(o) |= FILE_CREATED; \
1152 * For cached negative dentry and new dentry, handle lookup/create/open
1155 static int ll_atomic_open(struct inode *dir, struct dentry *dentry,
1156 struct file *file, unsigned open_flags,
1157 umode_t mode ll_last_arg)
1159 struct lookup_intent *it;
1161 long long lookup_flags = LOOKUP_OPEN;
1162 void *secctx = NULL;
1163 __u32 secctxlen = 0;
1164 void *encctx = NULL;
1165 __u32 encctxlen = 0;
1166 struct ll_sb_info *sbi = NULL;
1167 struct pcc_create_attach pca = { NULL, NULL };
1168 bool encrypt = false;
1173 "VFS Op:name=%pd, dir="DFID"(%p), file %p, open_flags %x, mode %x opened %d\n",
1174 dentry, PFID(ll_inode2fid(dir)), dir, file, open_flags, mode,
1175 ll_is_opened(opened, file));
1177 /* Only negative dentries enter here */
1178 LASSERT(dentry->d_inode == NULL);
1180 if (!d_unhashed(dentry)) {
1181 /* A valid negative dentry that just passed revalidation,
1182 * there's little point to try and open it server-side,
1183 * even though there's a minuscule chance it might succeed.
1184 * Either way it's a valid race to just return -ENOENT here.
1186 if (!(open_flags & O_CREAT))
1189 /* Otherwise we just unhash it to be rehashed afresh via
1190 * lookup if necessary
1195 OBD_ALLOC(it, sizeof(*it));
1199 it->it_op = IT_OPEN;
1200 if (open_flags & O_CREAT) {
1201 it->it_op |= IT_CREAT;
1202 lookup_flags |= LOOKUP_CREATE;
1203 sbi = ll_i2sbi(dir);
1204 /* Volatile file is used for HSM restore, so do not use PCC */
1205 if (!filename_is_volatile(dentry->d_name.name,
1206 dentry->d_name.len, NULL)) {
1207 struct pcc_matcher item;
1208 struct pcc_dataset *dataset;
1210 item.pm_uid = from_kuid(&init_user_ns, current_uid());
1211 item.pm_gid = from_kgid(&init_user_ns, current_gid());
1212 item.pm_projid = ll_i2info(dir)->lli_projid;
1213 item.pm_name = &dentry->d_name;
1214 dataset = pcc_dataset_match_get(&sbi->ll_pcc_super,
1216 pca.pca_dataset = dataset;
1219 it->it_create_mode = (mode & S_IALLUGO) | S_IFREG;
1220 it->it_flags = (open_flags & ~O_ACCMODE) | OPEN_FMODE(open_flags);
1221 it->it_flags &= ~MDS_OPEN_FL_INTERNAL;
1223 if (ll_sbi_has_encrypt(ll_i2sbi(dir)) && IS_ENCRYPTED(dir)) {
1224 /* in case of create, this is going to be a regular file because
1225 * we set S_IFREG bit on it->it_create_mode above
1227 rc = llcrypt_get_encryption_info(dir);
1229 GOTO(out_release, rc);
1230 if (open_flags & O_CREAT) {
1231 if (!llcrypt_has_encryption_key(dir))
1232 GOTO(out_release, rc = -ENOKEY);
1237 OBD_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_FILE_PAUSE2, cfs_fail_val);
1239 /* We can only arrive at this path when we have no inode, so
1240 * we only need to request open lock if it was requested
1243 if (ll_i2sbi(dir)->ll_oc_thrsh_count == 1 &&
1244 exp_connect_flags2(ll_i2mdexp(dir)) &
1245 OBD_CONNECT2_ATOMIC_OPEN_LOCK)
1246 it->it_flags |= MDS_OPEN_LOCK;
1248 /* Dentry added to dcache tree in ll_lookup_it */
1249 de = ll_lookup_it(dir, dentry, it, &secctx, &secctxlen, &pca, encrypt,
1250 &encctx, &encctxlen);
1253 else if (de != NULL)
1256 CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_FILE_PAUSE, cfs_fail_val);
1259 if (it_disposition(it, DISP_OPEN_CREATE)) {
1260 /* Dentry instantiated in ll_create_it. */
1261 rc = ll_create_it(dir, dentry, it, secctx, secctxlen,
1262 encrypt, encctx, encctxlen);
1263 ll_security_release_secctx(secctx, secctxlen);
1264 llcrypt_free_ctx(encctx, encctxlen);
1266 /* We dget in ll_splice_alias. */
1272 rc = pcc_inode_create_fini(dentry->d_inode, &pca);
1276 GOTO(out_release, rc);
1279 ll_set_created(opened, file);
1281 /* Open the file with O_CREAT, but the file already
1282 * existed on MDT. This may happend in the case that
1283 * the LOOKUP ibits lock is revoked and the
1284 * corresponding dentry cache is deleted.
1285 * i.e. In the current Lustre, the truncate operation
1286 * will revoke the LOOKUP ibits lock, and the file
1287 * dentry cache will be invalidated. The following open
1288 * with O_CREAT flag will call into ->atomic_open, the
1289 * file was wrongly though as newly created file and
1290 * try to auto cache the file. So after client knows it
1291 * is not a DISP_OPEN_CREATE, it should cleanup the
1292 * already created PCC copy.
1294 pcc_create_attach_cleanup(dir->i_sb, &pca);
1296 if (open_flags & O_CREAT && encrypt &&
1298 rc = ll_set_encflags(dentry->d_inode, encctx,
1300 llcrypt_free_ctx(encctx, encctxlen);
1302 GOTO(out_release, rc);
1306 /* check also if a foreign file is openable */
1307 if (dentry->d_inode && it_disposition(it, DISP_OPEN_OPEN) &&
1308 ll_foreign_is_openable(dentry, open_flags)) {
1310 if (S_ISFIFO(dentry->d_inode->i_mode)) {
1311 /* We cannot call open here as it might
1312 * deadlock. This case is unreachable in
1313 * practice because of OBD_CONNECT_NODEVOH. */
1314 rc = finish_no_open(file, de);
1316 file->private_data = it;
1317 rc = ll_finish_open(file, dentry, opened);
1318 /* We dget in ll_splice_alias. finish_open takes
1319 * care of dget for fd open.
1325 rc = finish_no_open(file, de);
1328 pcc_create_attach_cleanup(dir->i_sb, &pca);
1332 ll_intent_release(it);
1333 OBD_FREE(it, sizeof(*it));
1338 /* We depend on "mode" being set with the proper file type/umask by now */
1339 static struct inode *ll_create_node(struct inode *dir, struct lookup_intent *it)
1341 struct inode *inode = NULL;
1342 struct ptlrpc_request *request = NULL;
1343 struct ll_sb_info *sbi = ll_i2sbi(dir);
1347 LASSERT(it && it->it_disposition);
1349 LASSERT(it_disposition(it, DISP_ENQ_CREATE_REF));
1350 request = it->it_request;
1351 it_clear_disposition(it, DISP_ENQ_CREATE_REF);
1352 rc = ll_prep_inode(&inode, &request->rq_pill, dir->i_sb, it);
1354 GOTO(out, inode = ERR_PTR(rc));
1356 /* Pause to allow for a race with concurrent access by fid */
1357 OBD_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_NODE_PAUSE, cfs_fail_val);
1359 /* We asked for a lock on the directory, but were granted a
1360 * lock on the inode. Since we finally have an inode pointer,
1361 * stuff it in the lock. */
1362 CDEBUG(D_DLMTRACE, "setting l_ast_data to inode "DFID"(%p)\n",
1363 PFID(ll_inode2fid(inode)), inode);
1364 ll_set_lock_data(sbi->ll_md_exp, inode, it, NULL);
1367 ptlrpc_req_finished(request);
1372 * By the time this is called, we already have created the directory cache
1373 * entry for the new file, but it is so far negative - it has no inode.
1375 * We defer creating the OBD object(s) until open, to keep the intent and
1376 * non-intent code paths similar, and also because we do not have the MDS
1377 * inode number before calling ll_create_node() (which is needed for LOV),
1378 * so we would need to do yet another RPC to the MDS to store the LOV EA
1379 * data on the MDS. If needed, we would pass the PACKED lmm as data and
1380 * lmm_size in datalen (the MDS still has code which will handle that).
1382 * If the create succeeds, we fill in the inode information
1383 * with d_instantiate().
1385 static int ll_create_it(struct inode *dir, struct dentry *dentry,
1386 struct lookup_intent *it,
1387 void *secctx, __u32 secctxlen, bool encrypt,
1388 void *encctx, __u32 encctxlen)
1390 struct inode *inode;
1395 CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p), intent=%s\n",
1396 dentry, PFID(ll_inode2fid(dir)), dir, LL_IT2STR(it));
1398 rc = it_open_error(DISP_OPEN_CREATE, it);
1402 inode = ll_create_node(dir, it);
1404 RETURN(PTR_ERR(inode));
1406 if ((ll_i2sbi(inode)->ll_flags & LL_SBI_FILE_SECCTX) &&
1408 /* must be done before d_instantiate, because it calls
1409 * security_d_instantiate, which means a getxattr if security
1410 * context is not set yet */
1411 /* no need to protect selinux_inode_setsecurity() by
1412 * inode_lock. Taking it would lead to a client deadlock
1415 rc = security_inode_notifysecctx(inode, secctx, secctxlen);
1420 d_instantiate(dentry, inode);
1423 rc = ll_set_encflags(inode, encctx, encctxlen, true);
1428 if (!(ll_i2sbi(inode)->ll_flags & LL_SBI_FILE_SECCTX)) {
1429 rc = ll_inode_init_security(dentry, inode, dir);
1434 ll_set_lock_data(ll_i2sbi(dir)->ll_md_exp, inode, it, &bits);
1435 if (bits & MDS_INODELOCK_LOOKUP) {
1436 d_lustre_revalidate(dentry);
1437 ll_update_dir_depth(dir, inode);
1443 void ll_update_times(struct ptlrpc_request *request, struct inode *inode)
1445 struct mdt_body *body = req_capsule_server_get(&request->rq_pill,
1449 if (body->mbo_valid & OBD_MD_FLMTIME &&
1450 body->mbo_mtime > inode->i_mtime.tv_sec) {
1452 "setting fid " DFID " mtime from %lld to %llu\n",
1453 PFID(ll_inode2fid(inode)),
1454 (s64)inode->i_mtime.tv_sec, body->mbo_mtime);
1455 inode->i_mtime.tv_sec = body->mbo_mtime;
1458 if (body->mbo_valid & OBD_MD_FLCTIME &&
1459 body->mbo_ctime > inode->i_ctime.tv_sec)
1460 inode->i_ctime.tv_sec = body->mbo_ctime;
1463 /* once default LMV (space balanced) is set on ROOT, it should take effect if
1464 * default LMV is not set on parent directory.
1466 static void ll_qos_mkdir_prep(struct md_op_data *op_data, struct inode *dir)
1468 struct inode *root = dir->i_sb->s_root->d_inode;
1469 struct ll_inode_info *rlli = ll_i2info(root);
1470 struct ll_inode_info *lli = ll_i2info(dir);
1471 struct lmv_stripe_md *lsm;
1473 op_data->op_dir_depth = lli->lli_depth;
1475 /* parent directory is striped */
1476 if (unlikely(lli->lli_lsm_md))
1479 /* default LMV set on parent directory */
1480 if (unlikely(lli->lli_default_lsm_md))
1483 /* parent is ROOT */
1484 if (unlikely(dir == root))
1487 /* default LMV not set on ROOT */
1488 if (!rlli->lli_default_lsm_md)
1491 down_read(&rlli->lli_lsm_sem);
1492 lsm = rlli->lli_default_lsm_md;
1496 /* not space balanced */
1497 if (lsm->lsm_md_master_mdt_index != LMV_OFFSET_DEFAULT)
1500 if (lsm->lsm_md_max_inherit != LMV_INHERIT_NONE &&
1501 (lsm->lsm_md_max_inherit == LMV_INHERIT_UNLIMITED ||
1502 lsm->lsm_md_max_inherit >= lli->lli_depth)) {
1503 op_data->op_flags |= MF_QOS_MKDIR;
1504 if (lsm->lsm_md_max_inherit_rr != LMV_INHERIT_RR_NONE &&
1505 (lsm->lsm_md_max_inherit_rr == LMV_INHERIT_RR_UNLIMITED ||
1506 lsm->lsm_md_max_inherit_rr >= lli->lli_depth))
1507 op_data->op_flags |= MF_RR_MKDIR;
1508 CDEBUG(D_INODE, DFID" requests qos mkdir %#x\n",
1509 PFID(&lli->lli_fid), op_data->op_flags);
1512 up_read(&rlli->lli_lsm_sem);
1515 static int ll_new_node(struct inode *dir, struct dentry *dchild,
1516 const char *tgt, umode_t mode, int rdev, __u32 opc)
1518 struct qstr *name = &dchild->d_name;
1519 struct ptlrpc_request *request = NULL;
1520 struct md_op_data *op_data = NULL;
1521 struct inode *inode = NULL;
1522 struct ll_sb_info *sbi = ll_i2sbi(dir);
1524 bool encrypt = false;
1528 if (unlikely(tgt != NULL))
1529 tgt_len = strlen(tgt) + 1;
1532 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
1533 name->len, 0, opc, NULL);
1534 if (IS_ERR(op_data))
1535 GOTO(err_exit, err = PTR_ERR(op_data));
1538 ll_qos_mkdir_prep(op_data, dir);
1540 if (sbi->ll_flags & LL_SBI_FILE_SECCTX) {
1541 err = ll_dentry_init_security(dchild, mode, &dchild->d_name,
1542 &op_data->op_file_secctx_name,
1543 &op_data->op_file_secctx,
1544 &op_data->op_file_secctx_size);
1546 GOTO(err_exit, err);
1549 if (ll_sbi_has_encrypt(sbi) &&
1550 ((IS_ENCRYPTED(dir) &&
1551 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) ||
1552 (unlikely(llcrypt_dummy_context_enabled(dir)) && S_ISDIR(mode)))) {
1553 err = llcrypt_get_encryption_info(dir);
1555 GOTO(err_exit, err);
1556 if (!llcrypt_has_encryption_key(dir))
1557 GOTO(err_exit, err = -ENOKEY);
1562 err = llcrypt_inherit_context(dir, NULL, op_data, false);
1564 GOTO(err_exit, err);
1567 err = md_create(sbi->ll_md_exp, op_data, tgt, tgt_len, mode,
1568 from_kuid(&init_user_ns, current_fsuid()),
1569 from_kgid(&init_user_ns, current_fsgid()),
1570 cfs_curproc_cap_pack(), rdev, &request);
1571 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 14, 58, 0)
1573 * server < 2.12.58 doesn't pack default LMV in intent_getattr reply,
1574 * fetch default LMV here.
1576 if (unlikely(err == -EREMOTE)) {
1577 struct ll_inode_info *lli = ll_i2info(dir);
1578 struct lmv_user_md *lum;
1582 ptlrpc_req_finished(request);
1585 err2 = ll_dir_getstripe(dir, (void **)&lum, &lumsize, &request,
1586 OBD_MD_DEFAULT_MEA);
1587 ll_finish_md_op_data(op_data);
1590 struct lustre_md md = { NULL };
1592 md.body = req_capsule_server_get(&request->rq_pill,
1595 GOTO(err_exit, err = -EPROTO);
1597 OBD_ALLOC_PTR(md.default_lmv);
1598 if (!md.default_lmv)
1599 GOTO(err_exit, err = -ENOMEM);
1601 md.default_lmv->lsm_md_magic = lum->lum_magic;
1602 md.default_lmv->lsm_md_stripe_count =
1603 lum->lum_stripe_count;
1604 md.default_lmv->lsm_md_master_mdt_index =
1605 lum->lum_stripe_offset;
1606 md.default_lmv->lsm_md_hash_type = lum->lum_hash_type;
1607 md.default_lmv->lsm_md_max_inherit =
1608 lum->lum_max_inherit;
1609 md.default_lmv->lsm_md_max_inherit_rr =
1610 lum->lum_max_inherit_rr;
1612 err = ll_update_inode(dir, &md);
1613 md_free_lustre_md(sbi->ll_md_exp, &md);
1615 GOTO(err_exit, err);
1616 } else if (err2 == -ENODATA && lli->lli_default_lsm_md) {
1618 * If there are no default stripe EA on the MDT, but the
1619 * client has default stripe, then it probably means
1620 * default stripe EA has just been deleted.
1622 down_write(&lli->lli_lsm_sem);
1623 if (lli->lli_default_lsm_md)
1624 OBD_FREE_PTR(lli->lli_default_lsm_md);
1625 lli->lli_default_lsm_md = NULL;
1626 up_write(&lli->lli_lsm_sem);
1628 GOTO(err_exit, err);
1631 ptlrpc_req_finished(request);
1638 GOTO(err_exit, err);
1640 ll_update_times(request, dir);
1642 CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_NEWNODE_PAUSE, cfs_fail_val);
1644 err = ll_prep_inode(&inode, &request->rq_pill, dchild->d_sb, NULL);
1646 GOTO(err_exit, err);
1648 if (sbi->ll_flags & LL_SBI_FILE_SECCTX) {
1649 /* must be done before d_instantiate, because it calls
1650 * security_d_instantiate, which means a getxattr if security
1651 * context is not set yet */
1652 /* no need to protect selinux_inode_setsecurity() by
1653 * inode_lock. Taking it would lead to a client deadlock
1656 err = security_inode_notifysecctx(inode,
1657 op_data->op_file_secctx,
1658 op_data->op_file_secctx_size);
1660 GOTO(err_exit, err);
1663 d_instantiate(dchild, inode);
1666 err = ll_set_encflags(inode, op_data->op_file_encctx,
1667 op_data->op_file_encctx_size, true);
1669 GOTO(err_exit, err);
1672 if (!(sbi->ll_flags & LL_SBI_FILE_SECCTX)) {
1673 err = ll_inode_init_security(dchild, inode, dir);
1675 GOTO(err_exit, err);
1680 if (request != NULL)
1681 ptlrpc_req_finished(request);
1683 if (!IS_ERR_OR_NULL(op_data))
1684 ll_finish_md_op_data(op_data);
1689 static int ll_mknod(struct inode *dir, struct dentry *dchild, umode_t mode,
1692 ktime_t kstart = ktime_get();
1696 CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p) mode %o dev %x\n",
1697 dchild, PFID(ll_inode2fid(dir)), dir, mode, rdev);
1699 if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
1700 mode &= ~current_umask();
1702 switch (mode & S_IFMT) {
1711 err = ll_new_node(dir, dchild, NULL, mode, old_encode_dev(rdev),
1722 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKNOD,
1723 ktime_us_delta(ktime_get(), kstart));
1729 * Plain create. Intent create is handled in atomic_open.
1731 static int ll_create_nd(struct inode *dir, struct dentry *dentry,
1732 umode_t mode, bool want_excl)
1734 ktime_t kstart = ktime_get();
1737 CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_FILE_PAUSE, cfs_fail_val);
1740 "VFS Op:name=%pd, dir="DFID"(%p), flags=%u, excl=%d\n",
1741 dentry, PFID(ll_inode2fid(dir)), dir, mode, want_excl);
1743 /* Using mknod(2) to create a regular file is designed to not recognize
1744 * volatile file name, so we use ll_mknod() here. */
1745 rc = ll_mknod(dir, dentry, mode, 0);
1747 CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, unhashed %d\n",
1748 dentry, d_unhashed(dentry));
1751 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_CREATE,
1752 ktime_us_delta(ktime_get(), kstart));
1757 static int ll_symlink(struct inode *dir, struct dentry *dchild,
1758 const char *oldpath)
1760 ktime_t kstart = ktime_get();
1764 CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p), target=%.*s\n",
1765 dchild, PFID(ll_inode2fid(dir)), dir, 3000, oldpath);
1767 err = ll_new_node(dir, dchild, oldpath, S_IFLNK | S_IRWXUGO, 0,
1768 LUSTRE_OPC_SYMLINK);
1771 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_SYMLINK,
1772 ktime_us_delta(ktime_get(), kstart));
1777 static int ll_link(struct dentry *old_dentry, struct inode *dir,
1778 struct dentry *new_dentry)
1780 struct inode *src = old_dentry->d_inode;
1781 struct qstr *name = &new_dentry->d_name;
1782 struct ll_sb_info *sbi = ll_i2sbi(dir);
1783 struct ptlrpc_request *request = NULL;
1784 struct md_op_data *op_data;
1785 ktime_t kstart = ktime_get();
1790 "VFS Op: inode="DFID"(%p), dir="DFID"(%p), target=%pd\n",
1791 PFID(ll_inode2fid(src)), src,
1792 PFID(ll_inode2fid(dir)), dir, new_dentry);
1794 err = llcrypt_prepare_link(old_dentry, dir, new_dentry);
1798 op_data = ll_prep_md_op_data(NULL, src, dir, name->name, name->len,
1799 0, LUSTRE_OPC_ANY, NULL);
1800 if (IS_ERR(op_data))
1801 RETURN(PTR_ERR(op_data));
1803 err = md_link(sbi->ll_md_exp, op_data, &request);
1804 ll_finish_md_op_data(op_data);
1808 ll_update_times(request, dir);
1809 ll_stats_ops_tally(sbi, LPROC_LL_LINK,
1810 ktime_us_delta(ktime_get(), kstart));
1813 ptlrpc_req_finished(request);
1817 static int ll_mkdir(struct inode *dir, struct dentry *dchild, umode_t mode)
1819 ktime_t kstart = ktime_get();
1823 CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p)\n",
1824 dchild, PFID(ll_inode2fid(dir)), dir);
1826 if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
1827 mode &= ~current_umask();
1829 mode = (mode & (S_IRWXUGO|S_ISVTX)) | S_IFDIR;
1831 err = ll_new_node(dir, dchild, NULL, mode, 0, LUSTRE_OPC_MKDIR);
1833 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKDIR,
1834 ktime_us_delta(ktime_get(), kstart));
1839 static int ll_rmdir(struct inode *dir, struct dentry *dchild)
1841 struct qstr *name = &dchild->d_name;
1842 struct ptlrpc_request *request = NULL;
1843 struct md_op_data *op_data;
1844 ktime_t kstart = ktime_get();
1849 CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p)\n",
1850 dchild, PFID(ll_inode2fid(dir)), dir);
1852 if (unlikely(d_mountpoint(dchild)))
1855 /* some foreign dir may not be allowed to be removed */
1856 if (!ll_foreign_is_removable(dchild, false))
1859 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len,
1860 S_IFDIR, LUSTRE_OPC_ANY, NULL);
1861 if (IS_ERR(op_data))
1862 RETURN(PTR_ERR(op_data));
1864 if (dchild->d_inode != NULL)
1865 op_data->op_fid3 = *ll_inode2fid(dchild->d_inode);
1867 op_data->op_fid2 = op_data->op_fid3;
1868 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1869 ll_finish_md_op_data(op_data);
1871 struct mdt_body *body;
1873 ll_update_times(request, dir);
1874 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR,
1875 ktime_us_delta(ktime_get(), kstart));
1878 * The server puts attributes in on the last unlink, use them
1879 * to update the link count so the inode can be freed
1882 body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
1883 if (body->mbo_valid & OBD_MD_FLNLINK)
1884 set_nlink(dchild->d_inode, body->mbo_nlink);
1887 ptlrpc_req_finished(request);
1895 int ll_rmdir_entry(struct inode *dir, char *name, int namelen)
1897 struct ptlrpc_request *request = NULL;
1898 struct md_op_data *op_data;
1899 ktime_t kstart = ktime_get();
1903 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1904 namelen, name, PFID(ll_inode2fid(dir)), dir);
1906 op_data = ll_prep_md_op_data(NULL, dir, NULL, name, strlen(name),
1907 S_IFDIR, LUSTRE_OPC_ANY, NULL);
1908 if (IS_ERR(op_data))
1909 RETURN(PTR_ERR(op_data));
1910 op_data->op_cli_flags |= CLI_RM_ENTRY;
1911 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1912 ll_finish_md_op_data(op_data);
1914 ll_update_times(request, dir);
1916 ptlrpc_req_finished(request);
1918 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR,
1919 ktime_us_delta(ktime_get(), kstart));
1923 static int ll_unlink(struct inode *dir, struct dentry *dchild)
1925 struct qstr *name = &dchild->d_name;
1926 struct ptlrpc_request *request = NULL;
1927 struct md_op_data *op_data;
1928 struct mdt_body *body;
1929 ktime_t kstart = ktime_get();
1934 CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p)\n",
1935 dchild, PFID(ll_inode2fid(dir)), dir);
1938 * XXX: unlink bind mountpoint maybe call to here,
1939 * just check it as vfs_unlink does.
1941 if (unlikely(d_mountpoint(dchild)))
1944 /* some foreign file/dir may not be allowed to be unlinked */
1945 if (!ll_foreign_is_removable(dchild, false))
1948 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len, 0,
1949 LUSTRE_OPC_ANY, NULL);
1950 if (IS_ERR(op_data))
1951 RETURN(PTR_ERR(op_data));
1953 op_data->op_fid3 = *ll_inode2fid(dchild->d_inode);
1954 /* notify lower layer if inode has dirty pages */
1955 if (S_ISREG(dchild->d_inode->i_mode) &&
1956 ll_i2info(dchild->d_inode)->lli_clob &&
1957 dirty_cnt(dchild->d_inode))
1958 op_data->op_cli_flags |= CLI_DIRTY_DATA;
1959 op_data->op_fid2 = op_data->op_fid3;
1960 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1961 ll_finish_md_op_data(op_data);
1966 * The server puts attributes in on the last unlink, use them to update
1967 * the link count so the inode can be freed immediately.
1969 body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
1970 if (body->mbo_valid & OBD_MD_FLNLINK)
1971 set_nlink(dchild->d_inode, body->mbo_nlink);
1973 ll_update_times(request, dir);
1976 ptlrpc_req_finished(request);
1978 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_UNLINK,
1979 ktime_us_delta(ktime_get(), kstart));
1983 static int ll_rename(struct inode *src, struct dentry *src_dchild,
1984 struct inode *tgt, struct dentry *tgt_dchild
1985 #ifdef HAVE_IOPS_RENAME_WITH_FLAGS
1986 , unsigned int flags
1990 struct qstr *src_name = &src_dchild->d_name;
1991 struct qstr *tgt_name = &tgt_dchild->d_name;
1992 struct ptlrpc_request *request = NULL;
1993 struct ll_sb_info *sbi = ll_i2sbi(src);
1994 struct md_op_data *op_data;
1995 ktime_t kstart = ktime_get();
2000 #ifdef HAVE_IOPS_RENAME_WITH_FLAGS
2006 "VFS Op:oldname=%pd, src_dir="DFID"(%p), newname=%pd, tgt_dir="DFID"(%p)\n",
2007 src_dchild, PFID(ll_inode2fid(src)), src,
2008 tgt_dchild, PFID(ll_inode2fid(tgt)), tgt);
2010 if (unlikely(d_mountpoint(src_dchild) || d_mountpoint(tgt_dchild)))
2013 #ifdef HAVE_IOPS_RENAME_WITH_FLAGS
2014 err = llcrypt_prepare_rename(src, src_dchild, tgt, tgt_dchild, flags);
2016 err = llcrypt_prepare_rename(src, src_dchild, tgt, tgt_dchild, 0);
2020 /* we prevent an encrypted file from being renamed
2021 * into an unencrypted dir
2023 if (IS_ENCRYPTED(src) && !IS_ENCRYPTED(tgt))
2026 if (src_dchild->d_inode)
2027 mode = src_dchild->d_inode->i_mode;
2029 if (tgt_dchild->d_inode)
2030 mode = tgt_dchild->d_inode->i_mode;
2032 op_data = ll_prep_md_op_data(NULL, src, tgt, NULL, 0, mode,
2033 LUSTRE_OPC_ANY, NULL);
2034 if (IS_ERR(op_data))
2035 RETURN(PTR_ERR(op_data));
2037 if (src_dchild->d_inode)
2038 op_data->op_fid3 = *ll_inode2fid(src_dchild->d_inode);
2040 if (tgt_dchild->d_inode)
2041 op_data->op_fid4 = *ll_inode2fid(tgt_dchild->d_inode);
2043 err = md_rename(sbi->ll_md_exp, op_data,
2044 src_name->name, src_name->len,
2045 tgt_name->name, tgt_name->len, &request);
2046 ll_finish_md_op_data(op_data);
2048 ll_update_times(request, src);
2049 ll_update_times(request, tgt);
2052 ptlrpc_req_finished(request);
2055 d_move(src_dchild, tgt_dchild);
2056 ll_stats_ops_tally(sbi, LPROC_LL_RENAME,
2057 ktime_us_delta(ktime_get(), kstart));
2063 const struct inode_operations ll_dir_inode_operations = {
2065 .atomic_open = ll_atomic_open,
2066 .lookup = ll_lookup_nd,
2067 .create = ll_create_nd,
2068 /* We need all these non-raw things for NFSD, to not patch it. */
2069 .unlink = ll_unlink,
2072 .symlink = ll_symlink,
2074 .rename = ll_rename,
2075 .setattr = ll_setattr,
2076 .getattr = ll_getattr,
2077 .permission = ll_inode_permission,
2078 #ifdef HAVE_IOP_XATTR
2079 .setxattr = ll_setxattr,
2080 .getxattr = ll_getxattr,
2081 .removexattr = ll_removexattr,
2083 .listxattr = ll_listxattr,
2084 .get_acl = ll_get_acl,
2085 #ifdef HAVE_IOP_SET_ACL
2086 .set_acl = ll_set_acl,
2090 const struct inode_operations ll_special_inode_operations = {
2091 .setattr = ll_setattr,
2092 .getattr = ll_getattr,
2093 .permission = ll_inode_permission,
2094 #ifdef HAVE_IOP_XATTR
2095 .setxattr = ll_setxattr,
2096 .getxattr = ll_getxattr,
2097 .removexattr = ll_removexattr,
2099 .listxattr = ll_listxattr,
2100 .get_acl = ll_get_acl,
2101 #ifdef HAVE_IOP_SET_ACL
2102 .set_acl = ll_set_acl,