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 #ifndef HAVE_USER_NAMESPACE_ARG
50 #define ll_create_nd(ns, dir, de, mode, ex) ll_create_nd(dir, de, mode, ex)
51 #define ll_mkdir(ns, dir, dch, mode) ll_mkdir(dir, dch, mode)
52 #define ll_mknod(ns, dir, dch, mode, rd) ll_mknod(dir, dch, mode, rd)
53 #ifdef HAVE_IOPS_RENAME_WITH_FLAGS
54 #define ll_rename(ns, src, sdc, tgt, tdc, fl) ll_rename(src, sdc, tgt, tdc, fl)
56 #define ll_rename(ns, src, sdc, tgt, tdc) ll_rename(src, sdc, tgt, tdc)
57 #endif /* HAVE_IOPS_RENAME_WITH_FLAGS */
58 #define ll_symlink(nd, dir, dch, old) ll_symlink(dir, dch, old)
61 static int ll_create_it(struct inode *dir, struct dentry *dentry,
62 struct lookup_intent *it,
63 void *secctx, __u32 secctxlen, bool encrypt,
64 void *encctx, __u32 encctxlen, unsigned int open_flags);
66 /* called from iget5_locked->find_inode() under inode_lock spinlock */
67 static int ll_test_inode(struct inode *inode, void *opaque)
69 struct ll_inode_info *lli = ll_i2info(inode);
70 struct lustre_md *md = opaque;
72 if (unlikely(!(md->body->mbo_valid & OBD_MD_FLID))) {
73 CERROR("MDS body missing FID\n");
77 if (!lu_fid_eq(&lli->lli_fid, &md->body->mbo_fid1))
83 static int ll_set_inode(struct inode *inode, void *opaque)
85 struct ll_inode_info *lli = ll_i2info(inode);
86 struct mdt_body *body = ((struct lustre_md *)opaque)->body;
88 if (unlikely(!(body->mbo_valid & OBD_MD_FLID))) {
89 CERROR("MDS body missing FID\n");
93 lli->lli_fid = body->mbo_fid1;
94 if (unlikely(!(body->mbo_valid & OBD_MD_FLTYPE))) {
95 CERROR("Can not initialize inode "DFID" without object type: "
97 PFID(&lli->lli_fid), body->mbo_valid);
101 inode->i_mode = (inode->i_mode & ~S_IFMT) | (body->mbo_mode & S_IFMT);
102 if (unlikely(inode->i_mode == 0)) {
103 CERROR("Invalid inode "DFID" type\n", PFID(&lli->lli_fid));
114 * Get an inode by inode number(@hash), which is already instantiated by
115 * the intent lookup).
117 struct inode *ll_iget(struct super_block *sb, ino_t hash,
118 struct lustre_md *md)
126 inode = iget5_locked(sb, hash, ll_test_inode, ll_set_inode, md);
128 RETURN(ERR_PTR(-ENOMEM));
130 if (inode->i_state & I_NEW) {
131 rc = ll_read_inode2(inode, md);
132 if (rc == 0 && S_ISREG(inode->i_mode) &&
133 ll_i2info(inode)->lli_clob == NULL)
134 rc = cl_file_inode_init(inode, md);
137 /* Let's clear directory lsm here, otherwise
138 * make_bad_inode() will reset the inode mode
139 * to regular, then ll_clear_inode will not
140 * be able to clear lsm_md */
141 if (S_ISDIR(inode->i_mode))
142 ll_dir_clear_lsm_md(inode);
143 make_bad_inode(inode);
144 unlock_new_inode(inode);
148 inode_has_no_xattr(inode);
149 unlock_new_inode(inode);
151 } else if (is_bad_inode(inode)) {
153 inode = ERR_PTR(-ESTALE);
154 } else if (!(inode->i_state & (I_FREEING | I_CLEAR))) {
155 rc = ll_update_inode(inode, md);
156 CDEBUG(D_VFSTRACE, "got inode: "DFID"(%p): rc = %d\n",
157 PFID(&md->body->mbo_fid1), inode, rc);
159 if (S_ISDIR(inode->i_mode))
160 ll_dir_clear_lsm_md(inode);
169 /* mark negative sub file dentries invalid and prune unused dentries */
170 static void ll_prune_negative_children(struct inode *dir)
172 struct dentry *dentry;
173 struct dentry *child;
178 spin_lock(&dir->i_lock);
179 hlist_for_each_entry(dentry, &dir->i_dentry, d_alias) {
180 spin_lock(&dentry->d_lock);
181 list_for_each_entry(child, &dentry->d_subdirs, d_child) {
185 spin_lock_nested(&child->d_lock, DENTRY_D_LOCK_NESTED);
186 if (lld_is_init(child))
187 ll_d2d(child)->lld_invalid = 1;
188 if (!ll_d_count(child)) {
191 spin_unlock(&child->d_lock);
192 spin_unlock(&dentry->d_lock);
193 spin_unlock(&dir->i_lock);
195 CDEBUG(D_DENTRY, "prune negative dentry %pd\n",
201 spin_unlock(&child->d_lock);
203 spin_unlock(&dentry->d_lock);
205 spin_unlock(&dir->i_lock);
210 int ll_test_inode_by_fid(struct inode *inode, void *opaque)
212 return lu_fid_eq(&ll_i2info(inode)->lli_fid, opaque);
215 static int ll_dom_lock_cancel(struct inode *inode, struct ldlm_lock *lock)
218 struct ll_inode_info *lli = ll_i2info(inode);
223 env = cl_env_get(&refcheck);
225 RETURN(PTR_ERR(env));
227 CFS_FAIL_TIMEOUT(OBD_FAIL_LDLM_REPLAY_PAUSE, cfs_fail_val);
229 /* reach MDC layer to flush data under the DoM ldlm lock */
230 rc = cl_object_flush(env, lli->lli_clob, lock);
231 if (rc == -ENODATA) {
232 CDEBUG(D_INODE, "inode "DFID" layout has no DoM stripe\n",
233 PFID(ll_inode2fid(inode)));
234 /* most likely result of layout change, do nothing */
238 cl_env_put(env, &refcheck);
242 static void ll_lock_cancel_bits(struct ldlm_lock *lock, __u64 to_cancel)
244 struct inode *inode = ll_inode_from_resource_lock(lock);
245 struct ll_inode_info *lli;
246 __u64 bits = to_cancel;
252 /* That means the inode is evicted most likely and may cause
253 * the skipping of lock cleanups below, so print the message
256 if (lock->l_resource->lr_lvb_inode)
258 "can't take inode for the lock (%sevicted)",
259 lock->l_resource->lr_lvb_inode->i_state &
260 I_FREEING ? "" : "not ");
264 if (!fid_res_name_eq(ll_inode2fid(inode),
265 &lock->l_resource->lr_name)) {
266 LDLM_ERROR(lock, "data mismatch with object "DFID"(%p)",
267 PFID(ll_inode2fid(inode)), inode);
271 if (bits & MDS_INODELOCK_XATTR) {
272 ll_xattr_cache_empty(inode);
273 bits &= ~MDS_INODELOCK_XATTR;
276 /* For OPEN locks we differentiate between lock modes
277 * LCK_CR, LCK_CW, LCK_PR - bug 22891 */
278 if (bits & MDS_INODELOCK_OPEN)
279 ll_have_md_lock(lock->l_conn_export, inode, &bits,
282 if (bits & MDS_INODELOCK_OPEN) {
285 switch (lock->l_req_mode) {
296 LDLM_ERROR(lock, "bad lock mode for OPEN lock");
300 ll_md_real_close(inode, fmode);
302 bits &= ~MDS_INODELOCK_OPEN;
305 if (bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
306 MDS_INODELOCK_LAYOUT | MDS_INODELOCK_PERM |
308 ll_have_md_lock(lock->l_conn_export, inode, &bits, LCK_MINMODE);
310 if (bits & MDS_INODELOCK_DOM) {
311 rc = ll_dom_lock_cancel(inode, lock);
313 CDEBUG(D_INODE, "cannot flush DoM data "
315 PFID(ll_inode2fid(inode)), rc);
318 if (bits & MDS_INODELOCK_LAYOUT) {
319 struct cl_object_conf conf = {
320 .coc_opc = OBJECT_CONF_INVALIDATE,
324 rc = ll_layout_conf(inode, &conf);
326 CDEBUG(D_INODE, "cannot invalidate layout of "
328 PFID(ll_inode2fid(inode)), rc);
331 lli = ll_i2info(inode);
333 if (bits & MDS_INODELOCK_UPDATE)
334 set_bit(LLIF_UPDATE_ATIME, &lli->lli_flags);
336 if ((bits & MDS_INODELOCK_UPDATE) && S_ISDIR(inode->i_mode)) {
337 CDEBUG(D_INODE, "invalidating inode "DFID" lli = %p, "
338 "pfid = "DFID"\n", PFID(ll_inode2fid(inode)),
339 lli, PFID(&lli->lli_pfid));
340 truncate_inode_pages(inode->i_mapping, 0);
342 if (unlikely(!fid_is_zero(&lli->lli_pfid))) {
343 struct inode *master_inode = NULL;
346 /* This is slave inode, since all of the child dentry
347 * is connected on the master inode, so we have to
348 * invalidate the negative children on master inode */
349 CDEBUG(D_INODE, "Invalidate s"DFID" m"DFID"\n",
350 PFID(ll_inode2fid(inode)), PFID(&lli->lli_pfid));
352 hash = cl_fid_build_ino(&lli->lli_pfid,
353 ll_need_32bit_api(ll_i2sbi(inode)));
355 /* Do not lookup the inode with ilookup5, otherwise
356 * it will cause dead lock,
357 * 1. Client1 send chmod req to the MDT0, then on MDT0,
358 * it enqueues master and all of its slaves lock,
359 * (mdt_attr_set() -> mdt_lock_slaves()), after gets
360 * master and stripe0 lock, it will send the enqueue
361 * req (for stripe1) to MDT1, then MDT1 finds the lock
362 * has been granted to client2. Then MDT1 sends blocking
364 * 2. At the same time, client2 tries to unlink
365 * the striped dir (rm -rf striped_dir), and during
366 * lookup, it will hold the master inode of the striped
367 * directory, whose inode state is NEW, then tries to
368 * revalidate all of its slaves, (ll_prep_inode()->
369 * ll_iget()->ll_read_inode2()-> ll_update_inode().).
370 * And it will be blocked on the server side because
372 * 3. Then the client get the blocking_ast req, cancel
373 * the lock, but being blocked if using ->ilookup5()),
374 * because master inode state is NEW. */
375 master_inode = ilookup5_nowait(inode->i_sb, hash,
376 ll_test_inode_by_fid,
377 (void *)&lli->lli_pfid);
379 ll_prune_negative_children(master_inode);
383 ll_prune_negative_children(inode);
387 /* at umount s_root becomes NULL */
388 if ((bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM)) &&
389 inode->i_sb->s_root && !is_root_inode(inode))
390 ll_prune_aliases(inode);
392 if (bits & (MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM))
393 forget_all_cached_acls(inode);
399 /* Check if the given lock may be downgraded instead of canceling and
400 * that convert is really needed. */
401 int ll_md_need_convert(struct ldlm_lock *lock)
403 struct ldlm_namespace *ns = ldlm_lock_to_ns(lock);
405 __u64 wanted = lock->l_policy_data.l_inodebits.cancel_bits;
406 __u64 bits = lock->l_policy_data.l_inodebits.bits & ~wanted;
407 enum ldlm_mode mode = LCK_MINMODE;
409 if (!lock->l_conn_export ||
410 !exp_connect_lock_convert(lock->l_conn_export))
413 if (!wanted || !bits || ldlm_is_cancel(lock))
416 /* do not convert locks other than DOM for now */
417 if (!((bits | wanted) & MDS_INODELOCK_DOM))
420 /* We may have already remaining bits in some other lock so
421 * lock convert will leave us just extra lock for the same bit.
422 * Check if client has other lock with the same bits and the same
423 * or lower mode and don't convert if any.
425 switch (lock->l_req_mode) {
439 /* do not convert other modes */
443 /* is lock is too old to be converted? */
444 lock_res_and_lock(lock);
445 if (ktime_after(ktime_get(),
446 ktime_add(lock->l_last_used, ns->ns_dirty_age_limit))) {
447 unlock_res_and_lock(lock);
450 unlock_res_and_lock(lock);
452 inode = ll_inode_from_resource_lock(lock);
453 ll_have_md_lock(lock->l_conn_export, inode, &bits, mode);
458 int ll_md_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *ld,
459 void *data, int flag)
461 struct lustre_handle lockh;
467 case LDLM_CB_BLOCKING:
469 __u64 cancel_flags = LCF_ASYNC;
471 /* if lock convert is not needed then still have to
472 * pass lock via ldlm_cli_convert() to keep all states
473 * correct, set cancel_bits to full lock bits to cause
474 * full cancel to happen.
476 if (!ll_md_need_convert(lock)) {
477 lock_res_and_lock(lock);
478 lock->l_policy_data.l_inodebits.cancel_bits =
479 lock->l_policy_data.l_inodebits.bits;
480 unlock_res_and_lock(lock);
482 rc = ldlm_cli_convert(lock, cancel_flags);
485 /* continue with cancel otherwise */
486 ldlm_lock2handle(lock, &lockh);
487 rc = ldlm_cli_cancel(&lockh, cancel_flags);
489 CDEBUG(D_INODE, "ldlm_cli_cancel: rc = %d\n", rc);
494 case LDLM_CB_CANCELING:
496 __u64 to_cancel = lock->l_policy_data.l_inodebits.bits;
498 /* Nothing to do for non-granted locks */
499 if (!ldlm_is_granted(lock))
502 /* If 'ld' is supplied then bits to be cancelled are passed
503 * implicitly by lock converting and cancel_bits from 'ld'
504 * should be used. Otherwise full cancel is being performed
505 * and lock inodebits are used.
507 * Note: we cannot rely on cancel_bits in lock itself at this
508 * moment because they can be changed by concurrent thread,
509 * so ldlm_cli_inodebits_convert() pass cancel bits implicitly
513 /* partial bits cancel allowed only during convert */
514 LASSERT(ldlm_is_converting(lock));
515 /* mask cancel bits by lock bits so only no any unused
516 * bits are passed to ll_lock_cancel_bits()
518 to_cancel &= ld->l_policy_data.l_inodebits.cancel_bits;
520 ll_lock_cancel_bits(lock, to_cancel);
530 __u32 ll_i2suppgid(struct inode *i)
532 if (in_group_p(i->i_gid))
533 return (__u32)from_kgid(&init_user_ns, i->i_gid);
535 return (__u32) __kgid_val(INVALID_GID);
538 /* Pack the required supplementary groups into the supplied groups array. */
539 void ll_i2gids(__u32 *suppgids, struct inode *i1, struct inode *i2)
542 LASSERT(suppgids != NULL);
544 suppgids[0] = ll_i2suppgid(i1);
547 suppgids[1] = ll_i2suppgid(i2);
553 * try to reuse three types of dentry:
554 * 1. unhashed alias, this one is unhashed by d_invalidate (but it may be valid
555 * by concurrent .revalidate).
556 * 2. INVALID alias (common case for no valid ldlm lock held, but this flag may
557 * be cleared by others calling d_lustre_revalidate).
558 * 3. DISCONNECTED alias.
560 static struct dentry *ll_find_alias(struct inode *inode, struct dentry *dentry)
562 struct dentry *alias, *discon_alias, *invalid_alias;
564 if (hlist_empty(&inode->i_dentry))
567 discon_alias = invalid_alias = NULL;
569 spin_lock(&inode->i_lock);
570 hlist_for_each_entry(alias, &inode->i_dentry, d_alias) {
571 LASSERT(alias != dentry);
573 spin_lock(&alias->d_lock);
574 if ((alias->d_flags & DCACHE_DISCONNECTED) &&
575 S_ISDIR(inode->i_mode))
576 /* LASSERT(last_discon == NULL); LU-405, bz 20055 */
577 discon_alias = alias;
578 else if (alias->d_parent == dentry->d_parent &&
579 alias->d_name.hash == dentry->d_name.hash &&
580 alias->d_name.len == dentry->d_name.len &&
581 memcmp(alias->d_name.name, dentry->d_name.name,
582 dentry->d_name.len) == 0)
583 invalid_alias = alias;
584 spin_unlock(&alias->d_lock);
589 alias = invalid_alias ?: discon_alias ?: NULL;
591 spin_lock(&alias->d_lock);
593 spin_unlock(&alias->d_lock);
595 spin_unlock(&inode->i_lock);
601 * Similar to d_splice_alias(), but lustre treats invalid alias
602 * similar to DCACHE_DISCONNECTED, and tries to use it anyway.
604 struct dentry *ll_splice_alias(struct inode *inode, struct dentry *de)
609 new = ll_find_alias(inode, de);
611 if (!ll_d_setup(new, true))
612 return ERR_PTR(-ENOMEM);
616 "Reuse dentry %p inode %p refc %d flags %#x\n",
617 new, new->d_inode, ll_d_count(new), new->d_flags);
621 if (!ll_d_setup(de, false))
622 return ERR_PTR(-ENOMEM);
625 /* this needs only to be done for foreign symlink dirs as
626 * DCACHE_SYMLINK_TYPE is already set by d_flags_for_inode()
627 * kernel routine for files with symlink ops (ie, real symlink)
629 if (inode && S_ISDIR(inode->i_mode) &&
630 ll_sbi_has_foreign_symlink(ll_i2sbi(inode)) &&
631 #ifdef HAVE_IOP_GET_LINK
632 inode->i_op->get_link) {
634 inode->i_op->follow_link) {
636 CDEBUG(D_INFO, "%s: inode "DFID": faking foreign dir as a symlink\n",
637 ll_i2sbi(inode)->ll_fsname, PFID(ll_inode2fid(inode)));
638 spin_lock(&de->d_lock);
639 /* like d_flags_for_inode() already does for files */
640 de->d_flags = (de->d_flags & ~DCACHE_ENTRY_TYPE) |
642 spin_unlock(&de->d_lock);
645 CDEBUG(D_DENTRY, "Add dentry %p inode %p refc %d flags %#x\n",
646 de, de->d_inode, ll_d_count(de), de->d_flags);
650 static int ll_lookup_it_finish(struct ptlrpc_request *request,
651 struct lookup_intent *it,
652 struct inode *parent, struct dentry **de,
653 void *secctx, __u32 secctxlen,
654 void *encctx, __u32 encctxlen,
655 ktime_t kstart, bool encrypt)
657 struct inode *inode = NULL;
660 struct dentry *alias;
663 /* NB 1 request reference will be taken away by ll_intent_lock()
665 CDEBUG(D_DENTRY, "it %p it_disposition %x\n", it,
667 if (!it_disposition(it, DISP_LOOKUP_NEG)) {
668 struct req_capsule *pill = &request->rq_pill;
669 struct mdt_body *body = req_capsule_server_get(pill,
672 rc = ll_prep_inode(&inode, &request->rq_pill, (*de)->d_sb, it);
676 /* If encryption context was returned by MDT, put it in
677 * inode now to save an extra getxattr and avoid deadlock.
679 if (body->mbo_valid & OBD_MD_ENCCTX) {
680 encctx = req_capsule_server_get(pill, &RMF_FILE_ENCCTX);
681 encctxlen = req_capsule_get_size(pill,
687 "server returned encryption ctx for "DFID"\n",
688 PFID(ll_inode2fid(inode)));
689 rc = ll_xattr_cache_insert(inode,
690 xattr_for_enc(inode),
693 CWARN("%s: cannot set enc ctx for "DFID": rc = %d\n",
694 ll_i2sbi(inode)->ll_fsname,
695 PFID(ll_inode2fid(inode)), rc);
697 rc = llcrypt_prepare_readdir(inode);
700 "cannot get enc info for "DFID": rc = %d\n",
701 PFID(ll_inode2fid(inode)), rc);
706 ll_set_lock_data(ll_i2sbi(parent)->ll_md_exp, inode, it, &bits);
707 /* OPEN can return data if lock has DoM+LAYOUT bits set */
708 if (it->it_op & IT_OPEN &&
709 bits & MDS_INODELOCK_DOM && bits & MDS_INODELOCK_LAYOUT)
710 ll_dom_finish_open(inode, request);
712 /* We used to query real size from OSTs here, but actually
713 * this is not needed. For stat() calls size would be updated
714 * from subsequent do_revalidate()->ll_inode_revalidate_it() in
716 * vfs_getattr_it->ll_getattr()->ll_inode_revalidate_it() in 2.6
717 * Everybody else who needs correct file size would call
718 * ll_glimpse_size or some equivalent themselves anyway.
722 /* If security context was returned by MDT, put it in
723 * inode now to save an extra getxattr from security hooks,
724 * and avoid deadlock.
726 if (body->mbo_valid & OBD_MD_SECCTX) {
727 secctx = req_capsule_server_get(pill, &RMF_FILE_SECCTX);
728 secctxlen = req_capsule_get_size(pill,
733 CDEBUG(D_SEC, "server returned security context"
735 PFID(ll_inode2fid(inode)));
738 /* resume normally on error */
739 ll_inode_notifysecctx(inode, secctx, secctxlen);
742 /* Only hash *de if it is unhashed (new dentry).
743 * Atoimc_open may passin hashed dentries for open.
745 alias = ll_splice_alias(inode, *de);
747 GOTO(out, rc = PTR_ERR(alias));
751 if (!it_disposition(it, DISP_LOOKUP_NEG)) {
752 /* We have the "lookup" lock, so unhide dentry */
753 if (bits & MDS_INODELOCK_LOOKUP) {
754 d_lustre_revalidate(*de);
755 ll_update_dir_depth(parent, (*de)->d_inode);
759 rc = llcrypt_prepare_readdir(inode);
763 } else if (!it_disposition(it, DISP_OPEN_CREATE)) {
765 * If file was created on the server, the dentry is revalidated
766 * in ll_create_it if the lock allows for it.
768 /* Check that parent has UPDATE lock. */
769 struct lookup_intent parent_it = {
771 .it_lock_handle = 0 };
772 struct ll_inode_info *lli = ll_i2info(parent);
773 struct lu_fid fid = lli->lli_fid;
775 /* If it is striped directory, get the real stripe parent */
776 if (unlikely(ll_dir_striped(parent))) {
777 down_read(&lli->lli_lsm_sem);
778 rc = md_get_fid_from_lsm(ll_i2mdexp(parent),
781 (*de)->d_name.len, &fid);
782 up_read(&lli->lli_lsm_sem);
787 if (md_revalidate_lock(ll_i2mdexp(parent), &parent_it, &fid,
789 d_lustre_revalidate(*de);
790 ll_intent_release(&parent_it);
794 if (it_disposition(it, DISP_OPEN_CREATE)) {
795 ll_stats_ops_tally(ll_i2sbi(parent), LPROC_LL_MKNOD,
796 ktime_us_delta(ktime_get(), kstart));
802 if (rc != 0 && it->it_op & IT_OPEN) {
803 ll_intent_drop_lock(it);
804 ll_open_cleanup((*de)->d_sb, &request->rq_pill);
810 static struct dentry *ll_lookup_it(struct inode *parent, struct dentry *dentry,
811 struct lookup_intent *it,
812 void **secctx, __u32 *secctxlen,
814 struct pcc_create_attach *pca,
816 void **encctx, __u32 *encctxlen)
818 ktime_t kstart = ktime_get();
819 struct lookup_intent lookup_it = { .it_op = IT_LOOKUP };
820 struct dentry *save = dentry, *retval;
821 struct ptlrpc_request *req = NULL;
822 struct md_op_data *op_data = NULL;
823 struct lov_user_md *lum = NULL;
824 struct ll_sb_info *sbi = ll_i2sbi(parent);
827 struct llcrypt_name fname;
831 if (dentry->d_name.len > sbi->ll_namelen)
832 RETURN(ERR_PTR(-ENAMETOOLONG));
834 CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p), intent=%s\n",
835 dentry, PFID(ll_inode2fid(parent)), parent, LL_IT2STR(it));
837 if (d_mountpoint(dentry))
838 CERROR("Tell Peter, lookup on mtpt, it %s\n", LL_IT2STR(it));
840 if (it == NULL || it->it_op == IT_GETXATTR)
843 if (it->it_op == IT_GETATTR && dentry_may_statahead(parent, dentry)) {
844 rc = ll_revalidate_statahead(parent, &dentry, 0);
846 RETURN(dentry == save ? NULL : dentry);
849 if (it->it_op & IT_OPEN && it->it_flags & FMODE_WRITE &&
850 dentry->d_sb->s_flags & SB_RDONLY)
851 RETURN(ERR_PTR(-EROFS));
853 if (it->it_op & IT_CREAT)
854 opc = LUSTRE_OPC_CREATE;
856 opc = LUSTRE_OPC_LOOKUP;
858 rc = ll_prepare_lookup(parent, dentry, &fname, &fid);
860 RETURN(rc != -ENOENT ? ERR_PTR(rc) : NULL);
862 op_data = ll_prep_md_op_data(NULL, parent, NULL, fname.disk_name.name,
863 fname.disk_name.len, 0, opc, NULL);
864 if (IS_ERR(op_data)) {
865 llcrypt_free_filename(&fname);
866 RETURN(ERR_CAST(op_data));
868 if (!fid_is_zero(&fid)) {
869 op_data->op_fid2 = fid;
870 op_data->op_bias = MDS_FID_OP;
871 if (it->it_op & IT_OPEN)
872 it->it_flags |= MDS_OPEN_BY_FID;
875 /* enforce umask if acl disabled or MDS doesn't support umask */
876 if (!IS_POSIXACL(parent) || !exp_connect_umask(ll_i2mdexp(parent)))
877 it->it_create_mode &= ~current_umask();
879 if (it->it_op & IT_CREAT &&
880 test_bit(LL_SBI_FILE_SECCTX, sbi->ll_flags)) {
881 rc = ll_dentry_init_security(dentry, it->it_create_mode,
883 &op_data->op_file_secctx_name,
884 &op_data->op_file_secctx_name_size,
885 &op_data->op_file_secctx,
886 &op_data->op_file_secctx_size,
887 &op_data->op_file_secctx_slot);
889 GOTO(out, retval = ERR_PTR(rc));
891 *secctx = op_data->op_file_secctx;
892 if (secctxlen != NULL)
893 *secctxlen = op_data->op_file_secctx_size;
894 if (secctxslot != NULL)
895 *secctxslot = op_data->op_file_secctx_slot;
899 if (secctxlen != NULL)
901 if (secctxslot != NULL)
904 if (it->it_op & IT_CREAT && encrypt) {
905 if (unlikely(filename_is_volatile(dentry->d_name.name,
906 dentry->d_name.len, NULL))) {
907 /* get encryption context from reference file */
908 int ctx_size = LLCRYPT_ENC_CTX_SIZE;
909 struct lustre_sb_info *lsi;
910 struct file *ref_file;
911 struct inode *ref_inode;
914 rc = volatile_ref_file(dentry->d_name.name,
918 GOTO(out, retval = ERR_PTR(rc));
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,
936 #elif defined(HAVE_LUSTRE_CRYPTO)
937 rc = ref_inode->i_sb->s_cop->get_context(ref_inode,
943 OBD_FREE(ctx, ctx_size);
949 OBD_FREE(ctx, ctx_size);
953 op_data->op_file_encctx_size = rc;
954 if (rc == ctx_size) {
955 op_data->op_file_encctx = ctx;
957 OBD_ALLOC(op_data->op_file_encctx,
958 op_data->op_file_encctx_size);
959 if (!op_data->op_file_encctx) {
960 OBD_FREE(ctx, ctx_size);
961 GOTO(out, retval = ERR_PTR(-ENOMEM));
963 memcpy(op_data->op_file_encctx, ctx,
964 op_data->op_file_encctx_size);
965 OBD_FREE(ctx, ctx_size);
969 rc = llcrypt_inherit_context(parent, NULL, op_data,
972 GOTO(out, retval = ERR_PTR(rc));
975 *encctx = op_data->op_file_encctx;
976 if (encctxlen != NULL)
977 *encctxlen = op_data->op_file_encctx_size;
981 if (encctxlen != NULL)
985 /* ask for security context upon intent:
986 * get name of security xattr to request to server
988 if (it->it_op & (IT_LOOKUP | IT_GETATTR | IT_OPEN))
989 op_data->op_file_secctx_name_size =
990 ll_secctx_name_get(sbi, &op_data->op_file_secctx_name);
992 if (pca && pca->pca_dataset) {
995 GOTO(out, retval = ERR_PTR(-ENOMEM));
997 lum->lmm_magic = LOV_USER_MAGIC_V1;
998 lum->lmm_pattern = LOV_PATTERN_F_RELEASED | LOV_PATTERN_RAID0;
999 op_data->op_data = lum;
1000 op_data->op_data_size = sizeof(*lum);
1001 op_data->op_archive_id = pca->pca_dataset->pccd_rwid;
1002 it->it_flags |= MDS_OPEN_PCC;
1005 rc = md_intent_lock(ll_i2mdexp(parent), op_data, it, &req,
1006 &ll_md_blocking_ast, 0);
1007 /* If the MDS allows the client to chgrp (CFS_SETGRP_PERM), but the
1008 * client does not know which suppgid should be sent to the MDS, or
1009 * some other(s) changed the target file's GID after this RPC sent
1010 * to the MDS with the suppgid as the original GID, then we should
1011 * try again with right suppgid. */
1012 if (rc == -EACCES && it->it_op & IT_OPEN &&
1013 it_disposition(it, DISP_OPEN_DENY)) {
1014 struct mdt_body *body;
1016 LASSERT(req != NULL);
1018 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1019 if (op_data->op_suppgids[0] == body->mbo_gid ||
1020 op_data->op_suppgids[1] == body->mbo_gid ||
1021 !in_group_p(make_kgid(&init_user_ns, body->mbo_gid)))
1022 GOTO(out, retval = ERR_PTR(-EACCES));
1024 fid_zero(&op_data->op_fid2);
1025 op_data->op_suppgids[1] = body->mbo_gid;
1026 ptlrpc_req_finished(req);
1028 ll_intent_release(it);
1029 rc = md_intent_lock(ll_i2mdexp(parent), op_data, it, &req,
1030 &ll_md_blocking_ast, 0);
1034 GOTO(out, retval = ERR_PTR(rc));
1036 if (pca && pca->pca_dataset) {
1037 rc = pcc_inode_create(parent->i_sb, pca->pca_dataset,
1041 GOTO(out, retval = ERR_PTR(rc));
1044 /* dir layout may change */
1045 ll_unlock_md_op_lsm(op_data);
1046 rc = ll_lookup_it_finish(req, it, parent, &dentry,
1047 secctx != NULL ? *secctx : NULL,
1048 secctxlen != NULL ? *secctxlen : 0,
1049 encctx != NULL ? *encctx : NULL,
1050 encctxlen != NULL ? *encctxlen : 0,
1053 ll_intent_release(it);
1054 GOTO(out, retval = ERR_PTR(rc));
1057 if ((it->it_op & IT_OPEN) && dentry->d_inode &&
1058 !S_ISREG(dentry->d_inode->i_mode) &&
1059 !S_ISDIR(dentry->d_inode->i_mode)) {
1060 ll_release_openhandle(dentry, it);
1062 ll_lookup_finish_locks(it, dentry);
1064 GOTO(out, retval = (dentry == save) ? NULL : dentry);
1067 if (op_data != NULL && !IS_ERR(op_data)) {
1068 if (secctx != NULL && secctxlen != NULL) {
1069 /* caller needs sec ctx info, so reset it in op_data to
1070 * prevent it from being freed */
1071 op_data->op_file_secctx = NULL;
1072 op_data->op_file_secctx_size = 0;
1074 if (encctx != NULL && encctxlen != NULL &&
1075 it->it_op & IT_CREAT && encrypt) {
1076 /* caller needs enc ctx info, so reset it in op_data to
1077 * prevent it from being freed
1079 op_data->op_file_encctx = NULL;
1080 op_data->op_file_encctx_size = 0;
1082 llcrypt_free_filename(&fname);
1083 ll_finish_md_op_data(op_data);
1089 ptlrpc_req_finished(req);
1093 static struct dentry *ll_lookup_nd(struct inode *parent, struct dentry *dentry,
1096 struct lookup_intent *itp, it = { .it_op = IT_GETATTR };
1099 CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p), flags=%u\n",
1100 dentry, PFID(ll_inode2fid(parent)), parent, flags);
1103 * Optimize away (CREATE && !OPEN). Let .create handle the race.
1104 * but only if we have write permissions there, otherwise we need
1105 * to proceed with lookup. LU-4185
1107 if ((flags & LOOKUP_CREATE) && !(flags & LOOKUP_OPEN) &&
1108 (inode_permission(&init_user_ns,
1109 parent, MAY_WRITE | MAY_EXEC) == 0))
1112 if (flags & (LOOKUP_PARENT|LOOKUP_OPEN|LOOKUP_CREATE))
1116 de = ll_lookup_it(parent, dentry, itp, NULL, NULL, NULL, NULL, false,
1120 ll_intent_release(itp);
1125 #ifdef FMODE_CREATED /* added in Linux v4.18-rc1-20-g73a09dd */
1126 # define ll_is_opened(o, f) ((f)->f_mode & FMODE_OPENED)
1127 # define ll_finish_open(f, d, o) finish_open((f), (d), NULL)
1128 # define ll_last_arg
1129 # define ll_set_created(o, f) \
1131 (f)->f_mode |= FMODE_CREATED; \
1135 # define ll_is_opened(o, f) (*(o))
1136 # define ll_finish_open(f, d, o) finish_open((f), (d), NULL, (o))
1137 # define ll_last_arg , int *opened
1138 # define ll_set_created(o, f) \
1140 *(o) |= FILE_CREATED; \
1146 * For cached negative dentry and new dentry, handle lookup/create/open
1149 static int ll_atomic_open(struct inode *dir, struct dentry *dentry,
1150 struct file *file, unsigned open_flags,
1151 umode_t mode ll_last_arg)
1153 struct lookup_intent *it;
1155 long long lookup_flags = LOOKUP_OPEN;
1156 void *secctx = NULL;
1157 __u32 secctxlen = 0;
1159 void *encctx = NULL;
1160 __u32 encctxlen = 0;
1161 struct ll_sb_info *sbi = NULL;
1162 struct pcc_create_attach pca = { NULL, NULL };
1163 bool encrypt = false;
1169 "VFS Op:name=%pd, dir="DFID"(%p), file %p, open_flags %x, mode %x opened %d\n",
1170 dentry, PFID(ll_inode2fid(dir)), dir, file, open_flags, mode,
1171 ll_is_opened(opened, file));
1173 /* Only negative dentries enter here */
1174 LASSERT(dentry->d_inode == NULL);
1176 if (!d_unhashed(dentry)) {
1177 /* A valid negative dentry that just passed revalidation,
1178 * there's little point to try and open it server-side,
1179 * even though there's a minuscule chance it might succeed.
1180 * Either way it's a valid race to just return -ENOENT here.
1182 if (!(open_flags & O_CREAT))
1185 /* Otherwise we just unhash it to be rehashed afresh via
1186 * lookup if necessary
1191 OBD_ALLOC(it, sizeof(*it));
1195 it->it_op = IT_OPEN;
1196 if (open_flags & O_CREAT) {
1197 it->it_op |= IT_CREAT;
1198 lookup_flags |= LOOKUP_CREATE;
1199 sbi = ll_i2sbi(dir);
1200 /* Volatile file is used for HSM restore, so do not use PCC */
1201 if (!filename_is_volatile(dentry->d_name.name,
1202 dentry->d_name.len, NULL)) {
1203 struct pcc_matcher item;
1204 struct pcc_dataset *dataset;
1206 item.pm_uid = from_kuid(&init_user_ns, current_uid());
1207 item.pm_gid = from_kgid(&init_user_ns, current_gid());
1208 item.pm_projid = ll_i2info(dir)->lli_projid;
1209 item.pm_name = &dentry->d_name;
1210 dataset = pcc_dataset_match_get(&sbi->ll_pcc_super,
1212 pca.pca_dataset = dataset;
1215 it->it_create_mode = (mode & S_IALLUGO) | S_IFREG;
1216 it->it_flags = (open_flags & ~O_ACCMODE) | OPEN_FMODE(open_flags);
1217 it->it_flags &= ~MDS_OPEN_FL_INTERNAL;
1219 if (ll_sbi_has_encrypt(ll_i2sbi(dir)) && IS_ENCRYPTED(dir)) {
1220 /* in case of create, this is going to be a regular file because
1221 * we set S_IFREG bit on it->it_create_mode above
1223 rc = llcrypt_prepare_readdir(dir);
1225 GOTO(out_release, rc);
1227 if (open_flags & O_CREAT) {
1228 /* For migration or mirroring without enc key, we still
1229 * need to be able to create a volatile file.
1231 if (!llcrypt_has_encryption_key(dir) &&
1232 (!filename_is_volatile(dentry->d_name.name,
1233 dentry->d_name.len, NULL) ||
1234 (open_flags & O_FILE_ENC) != O_FILE_ENC ||
1235 !(open_flags & O_DIRECT)))
1236 GOTO(out_release, rc = -ENOKEY);
1240 CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_FILE_PAUSE2, cfs_fail_val);
1242 /* We can only arrive at this path when we have no inode, so
1243 * we only need to request open lock if it was requested
1246 if (ll_i2info(dir)->lli_open_thrsh_count != UINT_MAX)
1247 open_threshold = ll_i2info(dir)->lli_open_thrsh_count;
1249 open_threshold = ll_i2sbi(dir)->ll_oc_thrsh_count;
1251 if (open_threshold == 1 &&
1252 exp_connect_flags2(ll_i2mdexp(dir)) &
1253 OBD_CONNECT2_ATOMIC_OPEN_LOCK)
1254 it->it_flags |= MDS_OPEN_LOCK;
1256 /* Dentry added to dcache tree in ll_lookup_it */
1257 de = ll_lookup_it(dir, dentry, it, &secctx, &secctxlen, &secctxslot,
1258 &pca, encrypt, &encctx, &encctxlen);
1261 else if (de != NULL)
1264 CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_FILE_PAUSE, cfs_fail_val);
1267 if (it_disposition(it, DISP_OPEN_CREATE)) {
1268 /* Dentry instantiated in ll_create_it. */
1269 rc = ll_create_it(dir, dentry, it, secctx, secctxlen,
1270 encrypt, encctx, encctxlen,
1272 ll_security_release_secctx(secctx, secctxlen,
1274 llcrypt_free_ctx(encctx, encctxlen);
1276 /* We dget in ll_splice_alias. */
1282 rc = pcc_inode_create_fini(dentry->d_inode, &pca);
1286 GOTO(out_release, rc);
1289 ll_set_created(opened, file);
1291 /* Open the file with O_CREAT, but the file already
1292 * existed on MDT. This may happend in the case that
1293 * the LOOKUP ibits lock is revoked and the
1294 * corresponding dentry cache is deleted.
1295 * i.e. In the current Lustre, the truncate operation
1296 * will revoke the LOOKUP ibits lock, and the file
1297 * dentry cache will be invalidated. The following open
1298 * with O_CREAT flag will call into ->atomic_open, the
1299 * file was wrongly though as newly created file and
1300 * try to auto cache the file. So after client knows it
1301 * is not a DISP_OPEN_CREATE, it should cleanup the
1302 * already created PCC copy.
1304 pcc_create_attach_cleanup(dir->i_sb, &pca);
1306 if (open_flags & O_CREAT && encrypt &&
1308 rc = ll_set_encflags(dentry->d_inode, encctx,
1310 llcrypt_free_ctx(encctx, encctxlen);
1312 GOTO(out_release, rc);
1316 /* check also if a foreign file is openable */
1317 if (dentry->d_inode && it_disposition(it, DISP_OPEN_OPEN) &&
1318 ll_foreign_is_openable(dentry, open_flags)) {
1320 if (S_ISFIFO(dentry->d_inode->i_mode)) {
1321 /* We cannot call open here as it might
1322 * deadlock. This case is unreachable in
1323 * practice because of OBD_CONNECT_NODEVOH. */
1324 rc = finish_no_open(file, de);
1326 file->private_data = it;
1327 rc = ll_finish_open(file, dentry, opened);
1328 /* We dget in ll_splice_alias. finish_open takes
1329 * care of dget for fd open.
1335 rc = finish_no_open(file, de);
1338 pcc_create_attach_cleanup(dir->i_sb, &pca);
1342 ll_intent_release(it);
1343 OBD_FREE(it, sizeof(*it));
1348 /* We depend on "mode" being set with the proper file type/umask by now */
1349 static struct inode *ll_create_node(struct inode *dir, struct lookup_intent *it)
1351 struct inode *inode = NULL;
1352 struct ptlrpc_request *request = NULL;
1353 struct ll_sb_info *sbi = ll_i2sbi(dir);
1357 LASSERT(it && it->it_disposition);
1359 LASSERT(it_disposition(it, DISP_ENQ_CREATE_REF));
1360 request = it->it_request;
1361 it_clear_disposition(it, DISP_ENQ_CREATE_REF);
1362 rc = ll_prep_inode(&inode, &request->rq_pill, dir->i_sb, it);
1364 GOTO(out, inode = ERR_PTR(rc));
1366 /* Pause to allow for a race with concurrent access by fid */
1367 CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_NODE_PAUSE, cfs_fail_val);
1369 /* We asked for a lock on the directory, but were granted a
1370 * lock on the inode. Since we finally have an inode pointer,
1371 * stuff it in the lock. */
1372 CDEBUG(D_DLMTRACE, "setting l_ast_data to inode "DFID"(%p)\n",
1373 PFID(ll_inode2fid(inode)), inode);
1374 ll_set_lock_data(sbi->ll_md_exp, inode, it, NULL);
1377 ptlrpc_req_finished(request);
1382 * By the time this is called, we already have created the directory cache
1383 * entry for the new file, but it is so far negative - it has no inode.
1385 * We defer creating the OBD object(s) until open, to keep the intent and
1386 * non-intent code paths similar, and also because we do not have the MDS
1387 * inode number before calling ll_create_node() (which is needed for LOV),
1388 * so we would need to do yet another RPC to the MDS to store the LOV EA
1389 * data on the MDS. If needed, we would pass the PACKED lmm as data and
1390 * lmm_size in datalen (the MDS still has code which will handle that).
1392 * If the create succeeds, we fill in the inode information
1393 * with d_instantiate().
1395 static int ll_create_it(struct inode *dir, struct dentry *dentry,
1396 struct lookup_intent *it,
1397 void *secctx, __u32 secctxlen, bool encrypt,
1398 void *encctx, __u32 encctxlen, unsigned int open_flags)
1400 struct inode *inode;
1405 CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p), intent=%s\n",
1406 dentry, PFID(ll_inode2fid(dir)), dir, LL_IT2STR(it));
1408 rc = it_open_error(DISP_OPEN_CREATE, it);
1412 inode = ll_create_node(dir, it);
1414 RETURN(PTR_ERR(inode));
1416 /* must be done before d_instantiate, because it calls
1417 * security_d_instantiate, which means a getxattr if security
1418 * context is not set yet
1420 rc = ll_inode_notifysecctx(inode, secctx, secctxlen);
1424 d_instantiate(dentry, inode);
1427 bool preload = true;
1429 /* For migration or mirroring without enc key, we
1430 * create a volatile file without enc context.
1432 if (!llcrypt_has_encryption_key(dir) &&
1433 filename_is_volatile(dentry->d_name.name,
1434 dentry->d_name.len, NULL) &&
1435 (open_flags & O_FILE_ENC) == O_FILE_ENC &&
1436 open_flags & O_DIRECT)
1438 rc = ll_set_encflags(inode, encctx, encctxlen, preload);
1443 if (!test_bit(LL_SBI_FILE_SECCTX, ll_i2sbi(inode)->ll_flags)) {
1444 rc = ll_inode_init_security(dentry, inode, dir);
1449 ll_set_lock_data(ll_i2sbi(dir)->ll_md_exp, inode, it, &bits);
1450 if (bits & MDS_INODELOCK_LOOKUP) {
1451 d_lustre_revalidate(dentry);
1452 ll_update_dir_depth(dir, inode);
1458 void ll_update_times(struct ptlrpc_request *request, struct inode *inode)
1460 struct mdt_body *body = req_capsule_server_get(&request->rq_pill,
1464 if (body->mbo_valid & OBD_MD_FLMTIME &&
1465 body->mbo_mtime > inode->i_mtime.tv_sec) {
1467 "setting fid " DFID " mtime from %lld to %llu\n",
1468 PFID(ll_inode2fid(inode)),
1469 (s64)inode->i_mtime.tv_sec, body->mbo_mtime);
1470 inode->i_mtime.tv_sec = body->mbo_mtime;
1473 if (body->mbo_valid & OBD_MD_FLCTIME &&
1474 body->mbo_ctime > inode->i_ctime.tv_sec)
1475 inode->i_ctime.tv_sec = body->mbo_ctime;
1478 /* once default LMV (space balanced) is set on ROOT, it should take effect if
1479 * default LMV is not set on parent directory.
1481 static void ll_qos_mkdir_prep(struct md_op_data *op_data, struct inode *dir)
1483 struct inode *root = dir->i_sb->s_root->d_inode;
1484 struct ll_inode_info *rlli = ll_i2info(root);
1485 struct ll_inode_info *lli = ll_i2info(dir);
1486 struct lmv_stripe_md *lsm;
1487 unsigned short depth;
1489 op_data->op_dir_depth = lli->lli_inherit_depth ?: lli->lli_dir_depth;
1490 depth = lli->lli_dir_depth;
1492 /* parent directory is striped */
1493 if (unlikely(lli->lli_lsm_md))
1496 /* default LMV set on parent directory */
1497 if (unlikely(lli->lli_default_lsm_md))
1500 /* parent is ROOT */
1501 if (unlikely(dir == root))
1504 /* default LMV not set on ROOT */
1505 if (!rlli->lli_default_lsm_md)
1508 down_read(&rlli->lli_lsm_sem);
1509 lsm = rlli->lli_default_lsm_md;
1513 /* not space balanced */
1514 if (lsm->lsm_md_master_mdt_index != LMV_OFFSET_DEFAULT)
1518 * Check if the fs default is to be applied.
1519 * depth == 0 means 'not inited' for not root dir.
1521 if (lsm->lsm_md_max_inherit != LMV_INHERIT_NONE &&
1522 (lsm->lsm_md_max_inherit == LMV_INHERIT_UNLIMITED ||
1523 (depth && lsm->lsm_md_max_inherit > depth))) {
1524 op_data->op_flags |= MF_QOS_MKDIR;
1525 if (lsm->lsm_md_max_inherit_rr != LMV_INHERIT_RR_NONE &&
1526 (lsm->lsm_md_max_inherit_rr == LMV_INHERIT_RR_UNLIMITED ||
1527 (depth && lsm->lsm_md_max_inherit_rr > depth)))
1528 op_data->op_flags |= MF_RR_MKDIR;
1529 CDEBUG(D_INODE, DFID" requests qos mkdir %#x\n",
1530 PFID(&lli->lli_fid), op_data->op_flags);
1533 up_read(&rlli->lli_lsm_sem);
1536 static int ll_new_node(struct inode *dir, struct dentry *dchild,
1537 const char *tgt, umode_t mode, __u64 rdev, __u32 opc)
1539 struct qstr *name = &dchild->d_name;
1540 struct ptlrpc_request *request = NULL;
1541 struct md_op_data *op_data = NULL;
1542 struct inode *inode = NULL;
1543 struct ll_sb_info *sbi = ll_i2sbi(dir);
1544 struct llcrypt_str *disk_link = NULL;
1545 bool encrypt = false;
1549 if (unlikely(tgt != NULL)) {
1550 disk_link = (struct llcrypt_str *)rdev;
1557 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name,
1558 name->len, 0, opc, NULL);
1559 if (IS_ERR(op_data))
1560 GOTO(err_exit, err = PTR_ERR(op_data));
1563 ll_qos_mkdir_prep(op_data, dir);
1565 if (test_bit(LL_SBI_FILE_SECCTX, sbi->ll_flags)) {
1566 err = ll_dentry_init_security(dchild, mode, &dchild->d_name,
1567 &op_data->op_file_secctx_name,
1568 &op_data->op_file_secctx_name_size,
1569 &op_data->op_file_secctx,
1570 &op_data->op_file_secctx_size,
1571 &op_data->op_file_secctx_slot);
1573 GOTO(err_exit, err);
1576 if (ll_sbi_has_encrypt(sbi) &&
1577 ((IS_ENCRYPTED(dir) &&
1578 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) ||
1579 (unlikely(ll_sb_has_test_dummy_encryption(dir->i_sb)) &&
1581 err = llcrypt_prepare_readdir(dir);
1583 GOTO(err_exit, err);
1584 if (!llcrypt_has_encryption_key(dir))
1585 GOTO(err_exit, err = -ENOKEY);
1590 err = llcrypt_inherit_context(dir, NULL, op_data, false);
1592 GOTO(err_exit, err);
1594 if (S_ISLNK(mode)) {
1595 /* llcrypt needs inode to encrypt target name, so create
1596 * a fake inode and associate encryption context got
1597 * from llcrypt_inherit_context.
1599 struct inode *fakeinode =
1600 dchild->d_sb->s_op->alloc_inode(dchild->d_sb);
1603 GOTO(err_exit, err = -ENOMEM);
1604 fakeinode->i_sb = dchild->d_sb;
1605 fakeinode->i_mode |= S_IFLNK;
1607 fakeinode->i_opflags |= IOP_XATTR;
1609 ll_lli_init(ll_i2info(fakeinode));
1610 err = ll_set_encflags(fakeinode,
1611 op_data->op_file_encctx,
1612 op_data->op_file_encctx_size,
1615 err = __llcrypt_encrypt_symlink(fakeinode, tgt,
1619 ll_xattr_cache_destroy(fakeinode);
1620 llcrypt_put_encryption_info(fakeinode);
1621 dchild->d_sb->s_op->destroy_inode(fakeinode);
1623 GOTO(err_exit, err);
1627 err = md_create(sbi->ll_md_exp, op_data, tgt ? disk_link->name : NULL,
1628 tgt ? disk_link->len : 0, mode,
1629 from_kuid(&init_user_ns, current_fsuid()),
1630 from_kgid(&init_user_ns, current_fsgid()),
1631 current_cap(), rdev, &request);
1632 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 17, 58, 0)
1634 * server < 2.12.58 doesn't pack default LMV in intent_getattr reply,
1635 * fetch default LMV here.
1637 if (unlikely(err == -EREMOTE)) {
1638 struct ll_inode_info *lli = ll_i2info(dir);
1639 struct lmv_user_md *lum;
1643 ptlrpc_req_finished(request);
1645 ll_finish_md_op_data(op_data);
1648 err2 = ll_dir_getstripe(dir, (void **)&lum, &lumsize, &request,
1649 OBD_MD_DEFAULT_MEA);
1651 struct lustre_md md = { NULL };
1653 md.body = req_capsule_server_get(&request->rq_pill,
1656 GOTO(err_exit, err = -EPROTO);
1658 OBD_ALLOC_PTR(md.default_lmv);
1659 if (!md.default_lmv)
1660 GOTO(err_exit, err = -ENOMEM);
1662 md.default_lmv->lsm_md_magic = lum->lum_magic;
1663 md.default_lmv->lsm_md_stripe_count =
1664 lum->lum_stripe_count;
1665 md.default_lmv->lsm_md_master_mdt_index =
1666 lum->lum_stripe_offset;
1667 md.default_lmv->lsm_md_hash_type = lum->lum_hash_type;
1668 md.default_lmv->lsm_md_max_inherit =
1669 lum->lum_max_inherit;
1670 md.default_lmv->lsm_md_max_inherit_rr =
1671 lum->lum_max_inherit_rr;
1673 err = ll_update_inode(dir, &md);
1674 md_free_lustre_md(sbi->ll_md_exp, &md);
1676 GOTO(err_exit, err);
1677 } else if (err2 == -ENODATA && lli->lli_default_lsm_md) {
1679 * If there are no default stripe EA on the MDT, but the
1680 * client has default stripe, then it probably means
1681 * default stripe EA has just been deleted.
1683 down_write(&lli->lli_lsm_sem);
1684 if (lli->lli_default_lsm_md)
1685 OBD_FREE_PTR(lli->lli_default_lsm_md);
1686 lli->lli_default_lsm_md = NULL;
1687 up_write(&lli->lli_lsm_sem);
1689 GOTO(err_exit, err);
1692 ptlrpc_req_finished(request);
1699 GOTO(err_exit, err);
1701 ll_update_times(request, dir);
1703 CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_NEWNODE_PAUSE, cfs_fail_val);
1705 err = ll_prep_inode(&inode, &request->rq_pill, dchild->d_sb, NULL);
1707 GOTO(err_exit, err);
1709 /* must be done before d_instantiate, because it calls
1710 * security_d_instantiate, which means a getxattr if security
1711 * context is not set yet
1713 err = ll_inode_notifysecctx(inode,
1714 op_data->op_file_secctx,
1715 op_data->op_file_secctx_size);
1717 GOTO(err_exit, err);
1719 d_instantiate(dchild, inode);
1722 err = ll_set_encflags(inode, op_data->op_file_encctx,
1723 op_data->op_file_encctx_size, true);
1725 GOTO(err_exit, err);
1727 if (S_ISLNK(mode)) {
1728 struct ll_inode_info *lli = ll_i2info(inode);
1730 /* Cache the plaintext symlink target
1731 * for later use by get_link()
1733 OBD_ALLOC(lli->lli_symlink_name, strlen(tgt) + 1);
1734 /* do not return an error if we cannot
1735 * cache the symlink locally
1737 if (lli->lli_symlink_name)
1738 memcpy(lli->lli_symlink_name,
1739 tgt, strlen(tgt) + 1);
1743 if (!test_bit(LL_SBI_FILE_SECCTX, sbi->ll_flags)) {
1744 err = ll_inode_init_security(dchild, inode, dir);
1746 GOTO(err_exit, err);
1751 if (request != NULL)
1752 ptlrpc_req_finished(request);
1754 if (!IS_ERR_OR_NULL(op_data))
1755 ll_finish_md_op_data(op_data);
1760 static int ll_mknod(struct user_namespace *mnt_userns, struct inode *dir,
1761 struct dentry *dchild, umode_t mode, dev_t rdev)
1763 ktime_t kstart = ktime_get();
1767 CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p) mode %o dev %x\n",
1768 dchild, PFID(ll_inode2fid(dir)), dir, mode, rdev);
1770 if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
1771 mode &= ~current_umask();
1773 switch (mode & S_IFMT) {
1782 err = ll_new_node(dir, dchild, NULL, mode, old_encode_dev(rdev),
1793 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKNOD,
1794 ktime_us_delta(ktime_get(), kstart));
1800 * Plain create. Intent create is handled in atomic_open.
1802 static int ll_create_nd(struct user_namespace *mnt_userns,
1803 struct inode *dir, struct dentry *dentry,
1804 umode_t mode, bool want_excl)
1806 ktime_t kstart = ktime_get();
1809 CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_CREATE_FILE_PAUSE, cfs_fail_val);
1812 "VFS Op:name=%pd, dir="DFID"(%p), flags=%u, excl=%d\n",
1813 dentry, PFID(ll_inode2fid(dir)), dir, mode, want_excl);
1815 /* Using mknod(2) to create a regular file is designed to not recognize
1816 * volatile file name, so we use ll_mknod() here. */
1817 rc = ll_mknod(mnt_userns, dir, dentry, mode, 0);
1819 CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, unhashed %d\n",
1820 dentry, d_unhashed(dentry));
1823 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_CREATE,
1824 ktime_us_delta(ktime_get(), kstart));
1829 static int ll_symlink(struct user_namespace *mnt_userns, struct inode *dir,
1830 struct dentry *dchild, const char *oldpath)
1832 ktime_t kstart = ktime_get();
1833 int len = strlen(oldpath);
1834 struct llcrypt_str disk_link;
1838 CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p), target=%.*s\n",
1839 dchild, PFID(ll_inode2fid(dir)), dir, 3000, oldpath);
1841 err = llcrypt_prepare_symlink(dir, oldpath, len, dir->i_sb->s_blocksize,
1846 err = ll_new_node(dir, dchild, oldpath, S_IFLNK | S_IRWXUGO,
1847 (__u64)&disk_link, LUSTRE_OPC_SYMLINK);
1849 if (disk_link.name != (unsigned char *)oldpath)
1850 kfree(disk_link.name);
1853 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_SYMLINK,
1854 ktime_us_delta(ktime_get(), kstart));
1859 static int ll_link(struct dentry *old_dentry, struct inode *dir,
1860 struct dentry *new_dentry)
1862 struct inode *src = old_dentry->d_inode;
1863 struct qstr *name = &new_dentry->d_name;
1864 struct ll_sb_info *sbi = ll_i2sbi(dir);
1865 struct ptlrpc_request *request = NULL;
1866 struct md_op_data *op_data;
1867 ktime_t kstart = ktime_get();
1872 "VFS Op: inode="DFID"(%p), dir="DFID"(%p), target=%pd\n",
1873 PFID(ll_inode2fid(src)), src,
1874 PFID(ll_inode2fid(dir)), dir, new_dentry);
1876 err = llcrypt_prepare_link(old_dentry, dir, new_dentry);
1880 op_data = ll_prep_md_op_data(NULL, src, dir, name->name, name->len,
1881 0, LUSTRE_OPC_ANY, NULL);
1882 if (IS_ERR(op_data))
1883 RETURN(PTR_ERR(op_data));
1885 err = md_link(sbi->ll_md_exp, op_data, &request);
1886 ll_finish_md_op_data(op_data);
1890 ll_update_times(request, dir);
1891 ll_stats_ops_tally(sbi, LPROC_LL_LINK,
1892 ktime_us_delta(ktime_get(), kstart));
1895 ptlrpc_req_finished(request);
1899 static int ll_mkdir(struct user_namespace *mnt_userns, struct inode *dir,
1900 struct dentry *dchild, umode_t mode)
1902 ktime_t kstart = ktime_get();
1906 CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p)\n",
1907 dchild, PFID(ll_inode2fid(dir)), dir);
1909 if (!IS_POSIXACL(dir) || !exp_connect_umask(ll_i2mdexp(dir)))
1910 mode &= ~current_umask();
1912 mode = (mode & (S_IRWXUGO|S_ISVTX)) | S_IFDIR;
1914 err = ll_new_node(dir, dchild, NULL, mode, 0, LUSTRE_OPC_MKDIR);
1916 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_MKDIR,
1917 ktime_us_delta(ktime_get(), kstart));
1922 static int ll_rmdir(struct inode *dir, struct dentry *dchild)
1924 struct qstr *name = &dchild->d_name;
1925 struct ptlrpc_request *request = NULL;
1926 struct md_op_data *op_data;
1927 ktime_t kstart = ktime_get();
1932 CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p)\n",
1933 dchild, PFID(ll_inode2fid(dir)), dir);
1935 if (unlikely(d_mountpoint(dchild)))
1938 /* some foreign dir may not be allowed to be removed */
1939 if (!ll_foreign_is_removable(dchild, false))
1942 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len,
1943 S_IFDIR, LUSTRE_OPC_ANY, NULL);
1944 if (IS_ERR(op_data))
1945 RETURN(PTR_ERR(op_data));
1947 if (dchild->d_inode != NULL)
1948 op_data->op_fid3 = *ll_inode2fid(dchild->d_inode);
1950 if (fid_is_zero(&op_data->op_fid2))
1951 op_data->op_fid2 = op_data->op_fid3;
1952 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1953 ll_finish_md_op_data(op_data);
1955 struct mdt_body *body;
1957 ll_update_times(request, dir);
1958 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR,
1959 ktime_us_delta(ktime_get(), kstart));
1962 * The server puts attributes in on the last unlink, use them
1963 * to update the link count so the inode can be freed
1966 body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
1967 if (body->mbo_valid & OBD_MD_FLNLINK) {
1968 spin_lock(&dchild->d_inode->i_lock);
1969 set_nlink(dchild->d_inode, body->mbo_nlink);
1970 spin_unlock(&dchild->d_inode->i_lock);
1974 ptlrpc_req_finished(request);
1982 int ll_rmdir_entry(struct inode *dir, char *name, int namelen)
1984 struct ptlrpc_request *request = NULL;
1985 struct md_op_data *op_data;
1986 ktime_t kstart = ktime_get();
1990 CDEBUG(D_VFSTRACE, "VFS Op:name=%.*s, dir="DFID"(%p)\n",
1991 namelen, name, PFID(ll_inode2fid(dir)), dir);
1993 op_data = ll_prep_md_op_data(NULL, dir, NULL, name, strlen(name),
1994 S_IFDIR, LUSTRE_OPC_ANY, NULL);
1995 if (IS_ERR(op_data))
1996 RETURN(PTR_ERR(op_data));
1997 op_data->op_cli_flags |= CLI_RM_ENTRY;
1998 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
1999 ll_finish_md_op_data(op_data);
2001 ll_update_times(request, dir);
2003 ptlrpc_req_finished(request);
2005 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_RMDIR,
2006 ktime_us_delta(ktime_get(), kstart));
2010 static int ll_unlink(struct inode *dir, struct dentry *dchild)
2012 struct qstr *name = &dchild->d_name;
2013 struct ptlrpc_request *request = NULL;
2014 struct md_op_data *op_data;
2015 struct mdt_body *body;
2016 ktime_t kstart = ktime_get();
2021 CDEBUG(D_VFSTRACE, "VFS Op:name=%pd, dir="DFID"(%p)\n",
2022 dchild, PFID(ll_inode2fid(dir)), dir);
2025 * XXX: unlink bind mountpoint maybe call to here,
2026 * just check it as vfs_unlink does.
2028 if (unlikely(d_mountpoint(dchild)))
2031 /* some foreign file/dir may not be allowed to be unlinked */
2032 if (!ll_foreign_is_removable(dchild, false))
2035 op_data = ll_prep_md_op_data(NULL, dir, NULL, name->name, name->len, 0,
2036 LUSTRE_OPC_ANY, NULL);
2037 if (IS_ERR(op_data))
2038 RETURN(PTR_ERR(op_data));
2040 op_data->op_fid3 = *ll_inode2fid(dchild->d_inode);
2041 /* notify lower layer if inode has dirty pages */
2042 if (S_ISREG(dchild->d_inode->i_mode) &&
2043 ll_i2info(dchild->d_inode)->lli_clob &&
2044 dirty_cnt(dchild->d_inode))
2045 op_data->op_cli_flags |= CLI_DIRTY_DATA;
2046 if (fid_is_zero(&op_data->op_fid2))
2047 op_data->op_fid2 = op_data->op_fid3;
2048 rc = md_unlink(ll_i2sbi(dir)->ll_md_exp, op_data, &request);
2049 ll_finish_md_op_data(op_data);
2054 * The server puts attributes in on the last unlink, use them to update
2055 * the link count so the inode can be freed immediately.
2057 body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
2058 if (body->mbo_valid & OBD_MD_FLNLINK) {
2059 spin_lock(&dchild->d_inode->i_lock);
2060 set_nlink(dchild->d_inode, body->mbo_nlink);
2061 spin_unlock(&dchild->d_inode->i_lock);
2064 ll_update_times(request, dir);
2067 ptlrpc_req_finished(request);
2069 ll_stats_ops_tally(ll_i2sbi(dir), LPROC_LL_UNLINK,
2070 ktime_us_delta(ktime_get(), kstart));
2074 static int ll_rename(struct user_namespace *mnt_userns,
2075 struct inode *src, struct dentry *src_dchild,
2076 struct inode *tgt, struct dentry *tgt_dchild
2077 #if defined(HAVE_USER_NAMESPACE_ARG) || defined(HAVE_IOPS_RENAME_WITH_FLAGS)
2078 , unsigned int flags
2082 struct ptlrpc_request *request = NULL;
2083 struct ll_sb_info *sbi = ll_i2sbi(src);
2084 struct md_op_data *op_data;
2085 ktime_t kstart = ktime_get();
2087 struct llcrypt_name foldname, fnewname;
2091 #if defined(HAVE_USER_NAMESPACE_ARG) || defined(HAVE_IOPS_RENAME_WITH_FLAGS)
2097 "VFS Op:oldname=%pd, src_dir="DFID"(%p), newname=%pd, tgt_dir="DFID"(%p)\n",
2098 src_dchild, PFID(ll_inode2fid(src)), src,
2099 tgt_dchild, PFID(ll_inode2fid(tgt)), tgt);
2101 if (unlikely(d_mountpoint(src_dchild) || d_mountpoint(tgt_dchild)))
2104 #if defined(HAVE_USER_NAMESPACE_ARG) || defined(HAVE_IOPS_RENAME_WITH_FLAGS)
2105 err = llcrypt_prepare_rename(src, src_dchild, tgt, tgt_dchild, flags);
2107 err = llcrypt_prepare_rename(src, src_dchild, tgt, tgt_dchild, 0);
2111 /* we prevent an encrypted file from being renamed
2112 * into an unencrypted dir
2114 if (IS_ENCRYPTED(src) && !IS_ENCRYPTED(tgt))
2117 if (src_dchild->d_inode)
2118 mode = src_dchild->d_inode->i_mode;
2120 if (tgt_dchild->d_inode)
2121 mode = tgt_dchild->d_inode->i_mode;
2123 op_data = ll_prep_md_op_data(NULL, src, tgt, NULL, 0, mode,
2124 LUSTRE_OPC_ANY, NULL);
2125 if (IS_ERR(op_data))
2126 RETURN(PTR_ERR(op_data));
2128 /* If the client is using a subdir mount and does a rename to what it
2129 * sees as /.fscrypt, interpret it as the .fscrypt dir at fs root.
2131 if (unlikely(is_root_inode(tgt) && !fid_is_root(ll_inode2fid(tgt)) &&
2132 tgt_dchild->d_name.len == strlen(dot_fscrypt_name) &&
2133 strncmp(tgt_dchild->d_name.name, dot_fscrypt_name,
2134 tgt_dchild->d_name.len) == 0))
2135 lu_root_fid(&op_data->op_fid2);
2137 if (src_dchild->d_inode)
2138 op_data->op_fid3 = *ll_inode2fid(src_dchild->d_inode);
2140 if (tgt_dchild->d_inode)
2141 op_data->op_fid4 = *ll_inode2fid(tgt_dchild->d_inode);
2143 err = ll_setup_filename(src, &src_dchild->d_name, 1, &foldname, NULL);
2146 err = ll_setup_filename(tgt, &tgt_dchild->d_name, 1, &fnewname, NULL);
2148 llcrypt_free_filename(&foldname);
2151 err = md_rename(sbi->ll_md_exp, op_data,
2152 foldname.disk_name.name, foldname.disk_name.len,
2153 fnewname.disk_name.name, fnewname.disk_name.len,
2155 llcrypt_free_filename(&foldname);
2156 llcrypt_free_filename(&fnewname);
2157 ll_finish_md_op_data(op_data);
2159 ll_update_times(request, src);
2160 ll_update_times(request, tgt);
2163 ptlrpc_req_finished(request);
2166 d_move(src_dchild, tgt_dchild);
2167 ll_stats_ops_tally(sbi, LPROC_LL_RENAME,
2168 ktime_us_delta(ktime_get(), kstart));
2174 const struct inode_operations ll_dir_inode_operations = {
2176 .atomic_open = ll_atomic_open,
2177 .lookup = ll_lookup_nd,
2178 .create = ll_create_nd,
2179 /* We need all these non-raw things for NFSD, to not patch it. */
2180 .unlink = ll_unlink,
2183 .symlink = ll_symlink,
2185 .rename = ll_rename,
2186 .setattr = ll_setattr,
2187 .getattr = ll_getattr,
2188 .permission = ll_inode_permission,
2189 #ifdef HAVE_IOP_XATTR
2190 .setxattr = ll_setxattr,
2191 .getxattr = ll_getxattr,
2192 .removexattr = ll_removexattr,
2194 .listxattr = ll_listxattr,
2195 .get_acl = ll_get_acl,
2196 #ifdef HAVE_IOP_SET_ACL
2197 .set_acl = ll_set_acl,
2201 const struct inode_operations ll_special_inode_operations = {
2202 .setattr = ll_setattr,
2203 .getattr = ll_getattr,
2204 .permission = ll_inode_permission,
2205 #ifdef HAVE_IOP_XATTR
2206 .setxattr = ll_setxattr,
2207 .getxattr = ll_getxattr,
2208 .removexattr = ll_removexattr,
2210 .listxattr = ll_listxattr,
2211 .get_acl = ll_get_acl,
2212 #ifdef HAVE_IOP_SET_ACL
2213 .set_acl = ll_set_acl,