2 * Copyright (c) 2001 The Regents of the University of Michigan.
5 * Kendrick Smith <kmsmith@umich.edu>
6 * Andy Adamson <kandros@umich.edu>
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
12 * 1. Redistributions of source code must retain the above copyright
13 * notice, this list of conditions and the following disclaimer.
14 * 2. Redistributions in binary form must reproduce the above copyright
15 * notice, this list of conditions and the following disclaimer in the
16 * documentation and/or other materials provided with the distribution.
17 * 3. Neither the name of the University nor the names of its
18 * contributors may be used to endorse or promote products derived
19 * from this software without specific prior written permission.
21 * THIS SOFTWARE IS PROVIDED ``AS IS'' AND ANY EXPRESS OR IMPLIED
22 * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
23 * MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
24 * DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
25 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
26 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
27 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR
28 * BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
29 * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
30 * NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
31 * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
35 #include <linux/file.h>
37 #include <linux/slab.h>
38 #include <linux/namei.h>
39 #include <linux/swap.h>
40 #include <linux/pagemap.h>
41 #include <linux/ratelimit.h>
42 #include <linux/sunrpc/svcauth_gss.h>
43 #include <linux/sunrpc/addr.h>
44 #include <linux/jhash.h>
45 #include <linux/string_helpers.h>
46 #include <linux/fsnotify.h>
47 #include <linux/rhashtable.h>
48 #include <linux/nfs_ssc.h>
53 #include "current_stateid.h"
57 #include "filecache.h"
60 #define NFSDDBG_FACILITY NFSDDBG_PROC
62 #define all_ones {{ ~0, ~0}, ~0}
63 static const stateid_t one_stateid = {
65 .si_opaque = all_ones,
67 static const stateid_t zero_stateid = {
70 static const stateid_t currentstateid = {
73 static const stateid_t close_stateid = {
74 .si_generation = 0xffffffffU,
77 static u64 current_sessionid = 1;
79 #define ZERO_STATEID(stateid) (!memcmp((stateid), &zero_stateid, sizeof(stateid_t)))
80 #define ONE_STATEID(stateid) (!memcmp((stateid), &one_stateid, sizeof(stateid_t)))
81 #define CURRENT_STATEID(stateid) (!memcmp((stateid), ¤tstateid, sizeof(stateid_t)))
82 #define CLOSE_STATEID(stateid) (!memcmp((stateid), &close_stateid, sizeof(stateid_t)))
84 /* forward declarations */
85 static bool check_for_locks(struct nfs4_file *fp, struct nfs4_lockowner *lowner);
86 static void nfs4_free_ol_stateid(struct nfs4_stid *stid);
87 void nfsd4_end_grace(struct nfsd_net *nn);
88 static void _free_cpntf_state_locked(struct nfsd_net *nn, struct nfs4_cpntf_state *cps);
89 static void nfsd4_file_hash_remove(struct nfs4_file *fi);
90 static void deleg_reaper(struct nfsd_net *nn);
95 * Currently used for the del_recall_lru and file hash table. In an
96 * effort to decrease the scope of the client_mutex, this spinlock may
97 * eventually cover more:
99 static DEFINE_SPINLOCK(state_lock);
101 enum nfsd4_st_mutex_lock_subclass {
102 OPEN_STATEID_MUTEX = 0,
103 LOCK_STATEID_MUTEX = 1,
107 * A waitqueue for all in-progress 4.0 CLOSE operations that are waiting for
108 * the refcount on the open stateid to drop.
110 static DECLARE_WAIT_QUEUE_HEAD(close_wq);
113 * A waitqueue where a writer to clients/#/ctl destroying a client can
114 * wait for cl_rpc_users to drop to 0 and then for the client to be
117 static DECLARE_WAIT_QUEUE_HEAD(expiry_wq);
119 static struct kmem_cache *client_slab;
120 static struct kmem_cache *openowner_slab;
121 static struct kmem_cache *lockowner_slab;
122 static struct kmem_cache *file_slab;
123 static struct kmem_cache *stateid_slab;
124 static struct kmem_cache *deleg_slab;
125 static struct kmem_cache *odstate_slab;
127 static void free_session(struct nfsd4_session *);
129 static const struct nfsd4_callback_ops nfsd4_cb_recall_ops;
130 static const struct nfsd4_callback_ops nfsd4_cb_notify_lock_ops;
131 static const struct nfsd4_callback_ops nfsd4_cb_getattr_ops;
133 static struct workqueue_struct *laundry_wq;
135 int nfsd4_create_laundry_wq(void)
139 laundry_wq = alloc_workqueue("%s", WQ_UNBOUND, 0, "nfsd4");
140 if (laundry_wq == NULL)
145 void nfsd4_destroy_laundry_wq(void)
147 destroy_workqueue(laundry_wq);
150 static bool is_session_dead(struct nfsd4_session *ses)
152 return ses->se_flags & NFS4_SESSION_DEAD;
155 static __be32 mark_session_dead_locked(struct nfsd4_session *ses, int ref_held_by_me)
157 if (atomic_read(&ses->se_ref) > ref_held_by_me)
158 return nfserr_jukebox;
159 ses->se_flags |= NFS4_SESSION_DEAD;
163 static bool is_client_expired(struct nfs4_client *clp)
165 return clp->cl_time == 0;
168 static void nfsd4_dec_courtesy_client_count(struct nfsd_net *nn,
169 struct nfs4_client *clp)
171 if (clp->cl_state != NFSD4_ACTIVE)
172 atomic_add_unless(&nn->nfsd_courtesy_clients, -1, 0);
175 static __be32 get_client_locked(struct nfs4_client *clp)
177 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
179 lockdep_assert_held(&nn->client_lock);
181 if (is_client_expired(clp))
182 return nfserr_expired;
183 atomic_inc(&clp->cl_rpc_users);
184 nfsd4_dec_courtesy_client_count(nn, clp);
185 clp->cl_state = NFSD4_ACTIVE;
189 /* must be called under the client_lock */
191 renew_client_locked(struct nfs4_client *clp)
193 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
195 if (is_client_expired(clp)) {
197 printk("%s: client (clientid %08x/%08x) already expired\n",
199 clp->cl_clientid.cl_boot,
200 clp->cl_clientid.cl_id);
204 list_move_tail(&clp->cl_lru, &nn->client_lru);
205 clp->cl_time = ktime_get_boottime_seconds();
206 nfsd4_dec_courtesy_client_count(nn, clp);
207 clp->cl_state = NFSD4_ACTIVE;
210 static void put_client_renew_locked(struct nfs4_client *clp)
212 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
214 lockdep_assert_held(&nn->client_lock);
216 if (!atomic_dec_and_test(&clp->cl_rpc_users))
218 if (!is_client_expired(clp))
219 renew_client_locked(clp);
221 wake_up_all(&expiry_wq);
224 static void put_client_renew(struct nfs4_client *clp)
226 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
228 if (!atomic_dec_and_lock(&clp->cl_rpc_users, &nn->client_lock))
230 if (!is_client_expired(clp))
231 renew_client_locked(clp);
233 wake_up_all(&expiry_wq);
234 spin_unlock(&nn->client_lock);
237 static __be32 nfsd4_get_session_locked(struct nfsd4_session *ses)
241 if (is_session_dead(ses))
242 return nfserr_badsession;
243 status = get_client_locked(ses->se_client);
246 atomic_inc(&ses->se_ref);
250 static void nfsd4_put_session_locked(struct nfsd4_session *ses)
252 struct nfs4_client *clp = ses->se_client;
253 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
255 lockdep_assert_held(&nn->client_lock);
257 if (atomic_dec_and_test(&ses->se_ref) && is_session_dead(ses))
259 put_client_renew_locked(clp);
262 static void nfsd4_put_session(struct nfsd4_session *ses)
264 struct nfs4_client *clp = ses->se_client;
265 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
267 spin_lock(&nn->client_lock);
268 nfsd4_put_session_locked(ses);
269 spin_unlock(&nn->client_lock);
272 static struct nfsd4_blocked_lock *
273 find_blocked_lock(struct nfs4_lockowner *lo, struct knfsd_fh *fh,
276 struct nfsd4_blocked_lock *cur, *found = NULL;
278 spin_lock(&nn->blocked_locks_lock);
279 list_for_each_entry(cur, &lo->lo_blocked, nbl_list) {
280 if (fh_match(fh, &cur->nbl_fh)) {
281 list_del_init(&cur->nbl_list);
282 WARN_ON(list_empty(&cur->nbl_lru));
283 list_del_init(&cur->nbl_lru);
288 spin_unlock(&nn->blocked_locks_lock);
290 locks_delete_block(&found->nbl_lock);
294 static struct nfsd4_blocked_lock *
295 find_or_allocate_block(struct nfs4_lockowner *lo, struct knfsd_fh *fh,
298 struct nfsd4_blocked_lock *nbl;
300 nbl = find_blocked_lock(lo, fh, nn);
302 nbl = kmalloc(sizeof(*nbl), GFP_KERNEL);
304 INIT_LIST_HEAD(&nbl->nbl_list);
305 INIT_LIST_HEAD(&nbl->nbl_lru);
306 fh_copy_shallow(&nbl->nbl_fh, fh);
307 locks_init_lock(&nbl->nbl_lock);
308 kref_init(&nbl->nbl_kref);
309 nfsd4_init_cb(&nbl->nbl_cb, lo->lo_owner.so_client,
310 &nfsd4_cb_notify_lock_ops,
311 NFSPROC4_CLNT_CB_NOTIFY_LOCK);
318 free_nbl(struct kref *kref)
320 struct nfsd4_blocked_lock *nbl;
322 nbl = container_of(kref, struct nfsd4_blocked_lock, nbl_kref);
323 locks_release_private(&nbl->nbl_lock);
328 free_blocked_lock(struct nfsd4_blocked_lock *nbl)
330 locks_delete_block(&nbl->nbl_lock);
331 kref_put(&nbl->nbl_kref, free_nbl);
335 remove_blocked_locks(struct nfs4_lockowner *lo)
337 struct nfs4_client *clp = lo->lo_owner.so_client;
338 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
339 struct nfsd4_blocked_lock *nbl;
342 /* Dequeue all blocked locks */
343 spin_lock(&nn->blocked_locks_lock);
344 while (!list_empty(&lo->lo_blocked)) {
345 nbl = list_first_entry(&lo->lo_blocked,
346 struct nfsd4_blocked_lock,
348 list_del_init(&nbl->nbl_list);
349 WARN_ON(list_empty(&nbl->nbl_lru));
350 list_move(&nbl->nbl_lru, &reaplist);
352 spin_unlock(&nn->blocked_locks_lock);
355 while (!list_empty(&reaplist)) {
356 nbl = list_first_entry(&reaplist, struct nfsd4_blocked_lock,
358 list_del_init(&nbl->nbl_lru);
359 free_blocked_lock(nbl);
364 nfsd4_cb_notify_lock_prepare(struct nfsd4_callback *cb)
366 struct nfsd4_blocked_lock *nbl = container_of(cb,
367 struct nfsd4_blocked_lock, nbl_cb);
368 locks_delete_block(&nbl->nbl_lock);
372 nfsd4_cb_notify_lock_done(struct nfsd4_callback *cb, struct rpc_task *task)
374 trace_nfsd_cb_notify_lock_done(&zero_stateid, task);
377 * Since this is just an optimization, we don't try very hard if it
378 * turns out not to succeed. We'll requeue it on NFS4ERR_DELAY, and
379 * just quit trying on anything else.
381 switch (task->tk_status) {
383 rpc_delay(task, 1 * HZ);
391 nfsd4_cb_notify_lock_release(struct nfsd4_callback *cb)
393 struct nfsd4_blocked_lock *nbl = container_of(cb,
394 struct nfsd4_blocked_lock, nbl_cb);
396 free_blocked_lock(nbl);
399 static const struct nfsd4_callback_ops nfsd4_cb_notify_lock_ops = {
400 .prepare = nfsd4_cb_notify_lock_prepare,
401 .done = nfsd4_cb_notify_lock_done,
402 .release = nfsd4_cb_notify_lock_release,
406 * We store the NONE, READ, WRITE, and BOTH bits separately in the
407 * st_{access,deny}_bmap field of the stateid, in order to track not
408 * only what share bits are currently in force, but also what
409 * combinations of share bits previous opens have used. This allows us
410 * to enforce the recommendation in
411 * https://datatracker.ietf.org/doc/html/rfc7530#section-16.19.4 that
412 * the server return an error if the client attempt to downgrade to a
413 * combination of share bits not explicable by closing some of its
416 * This enforcement is arguably incomplete, since we don't keep
417 * track of access/deny bit combinations; so, e.g., we allow:
419 * OPEN allow read, deny write
420 * OPEN allow both, deny none
421 * DOWNGRADE allow read, deny none
423 * which we should reject.
425 * But you could also argue that our current code is already overkill,
426 * since it only exists to return NFS4ERR_INVAL on incorrect client
430 bmap_to_share_mode(unsigned long bmap)
433 unsigned int access = 0;
435 for (i = 1; i < 4; i++) {
436 if (test_bit(i, &bmap))
442 /* set share access for a given stateid */
444 set_access(u32 access, struct nfs4_ol_stateid *stp)
446 unsigned char mask = 1 << access;
448 WARN_ON_ONCE(access > NFS4_SHARE_ACCESS_BOTH);
449 stp->st_access_bmap |= mask;
452 /* clear share access for a given stateid */
454 clear_access(u32 access, struct nfs4_ol_stateid *stp)
456 unsigned char mask = 1 << access;
458 WARN_ON_ONCE(access > NFS4_SHARE_ACCESS_BOTH);
459 stp->st_access_bmap &= ~mask;
462 /* test whether a given stateid has access */
464 test_access(u32 access, struct nfs4_ol_stateid *stp)
466 unsigned char mask = 1 << access;
468 return (bool)(stp->st_access_bmap & mask);
471 /* set share deny for a given stateid */
473 set_deny(u32 deny, struct nfs4_ol_stateid *stp)
475 unsigned char mask = 1 << deny;
477 WARN_ON_ONCE(deny > NFS4_SHARE_DENY_BOTH);
478 stp->st_deny_bmap |= mask;
481 /* clear share deny for a given stateid */
483 clear_deny(u32 deny, struct nfs4_ol_stateid *stp)
485 unsigned char mask = 1 << deny;
487 WARN_ON_ONCE(deny > NFS4_SHARE_DENY_BOTH);
488 stp->st_deny_bmap &= ~mask;
491 /* test whether a given stateid is denying specific access */
493 test_deny(u32 deny, struct nfs4_ol_stateid *stp)
495 unsigned char mask = 1 << deny;
497 return (bool)(stp->st_deny_bmap & mask);
500 static int nfs4_access_to_omode(u32 access)
502 switch (access & NFS4_SHARE_ACCESS_BOTH) {
503 case NFS4_SHARE_ACCESS_READ:
505 case NFS4_SHARE_ACCESS_WRITE:
507 case NFS4_SHARE_ACCESS_BOTH:
515 access_permit_read(struct nfs4_ol_stateid *stp)
517 return test_access(NFS4_SHARE_ACCESS_READ, stp) ||
518 test_access(NFS4_SHARE_ACCESS_BOTH, stp) ||
519 test_access(NFS4_SHARE_ACCESS_WRITE, stp);
523 access_permit_write(struct nfs4_ol_stateid *stp)
525 return test_access(NFS4_SHARE_ACCESS_WRITE, stp) ||
526 test_access(NFS4_SHARE_ACCESS_BOTH, stp);
529 static inline struct nfs4_stateowner *
530 nfs4_get_stateowner(struct nfs4_stateowner *sop)
532 atomic_inc(&sop->so_count);
537 same_owner_str(struct nfs4_stateowner *sop, struct xdr_netobj *owner)
539 return (sop->so_owner.len == owner->len) &&
540 0 == memcmp(sop->so_owner.data, owner->data, owner->len);
543 static struct nfs4_openowner *
544 find_openstateowner_str_locked(unsigned int hashval, struct nfsd4_open *open,
545 struct nfs4_client *clp)
547 struct nfs4_stateowner *so;
549 lockdep_assert_held(&clp->cl_lock);
551 list_for_each_entry(so, &clp->cl_ownerstr_hashtbl[hashval],
553 if (!so->so_is_open_owner)
555 if (same_owner_str(so, &open->op_owner))
556 return openowner(nfs4_get_stateowner(so));
561 static struct nfs4_openowner *
562 find_openstateowner_str(unsigned int hashval, struct nfsd4_open *open,
563 struct nfs4_client *clp)
565 struct nfs4_openowner *oo;
567 spin_lock(&clp->cl_lock);
568 oo = find_openstateowner_str_locked(hashval, open, clp);
569 spin_unlock(&clp->cl_lock);
574 opaque_hashval(const void *ptr, int nbytes)
576 unsigned char *cptr = (unsigned char *) ptr;
586 static void nfsd4_free_file_rcu(struct rcu_head *rcu)
588 struct nfs4_file *fp = container_of(rcu, struct nfs4_file, fi_rcu);
590 kmem_cache_free(file_slab, fp);
594 put_nfs4_file(struct nfs4_file *fi)
596 if (refcount_dec_and_test(&fi->fi_ref)) {
597 nfsd4_file_hash_remove(fi);
598 WARN_ON_ONCE(!list_empty(&fi->fi_clnt_odstate));
599 WARN_ON_ONCE(!list_empty(&fi->fi_delegations));
600 call_rcu(&fi->fi_rcu, nfsd4_free_file_rcu);
604 static struct nfsd_file *
605 find_writeable_file_locked(struct nfs4_file *f)
607 struct nfsd_file *ret;
609 lockdep_assert_held(&f->fi_lock);
611 ret = nfsd_file_get(f->fi_fds[O_WRONLY]);
613 ret = nfsd_file_get(f->fi_fds[O_RDWR]);
617 static struct nfsd_file *
618 find_writeable_file(struct nfs4_file *f)
620 struct nfsd_file *ret;
622 spin_lock(&f->fi_lock);
623 ret = find_writeable_file_locked(f);
624 spin_unlock(&f->fi_lock);
629 static struct nfsd_file *
630 find_readable_file_locked(struct nfs4_file *f)
632 struct nfsd_file *ret;
634 lockdep_assert_held(&f->fi_lock);
636 ret = nfsd_file_get(f->fi_fds[O_RDONLY]);
638 ret = nfsd_file_get(f->fi_fds[O_RDWR]);
642 static struct nfsd_file *
643 find_readable_file(struct nfs4_file *f)
645 struct nfsd_file *ret;
647 spin_lock(&f->fi_lock);
648 ret = find_readable_file_locked(f);
649 spin_unlock(&f->fi_lock);
654 static struct nfsd_file *
655 find_rw_file(struct nfs4_file *f)
657 struct nfsd_file *ret;
659 spin_lock(&f->fi_lock);
660 ret = nfsd_file_get(f->fi_fds[O_RDWR]);
661 spin_unlock(&f->fi_lock);
667 find_any_file(struct nfs4_file *f)
669 struct nfsd_file *ret;
673 spin_lock(&f->fi_lock);
674 ret = nfsd_file_get(f->fi_fds[O_RDWR]);
676 ret = nfsd_file_get(f->fi_fds[O_WRONLY]);
678 ret = nfsd_file_get(f->fi_fds[O_RDONLY]);
680 spin_unlock(&f->fi_lock);
684 static struct nfsd_file *find_any_file_locked(struct nfs4_file *f)
686 lockdep_assert_held(&f->fi_lock);
688 if (f->fi_fds[O_RDWR])
689 return f->fi_fds[O_RDWR];
690 if (f->fi_fds[O_WRONLY])
691 return f->fi_fds[O_WRONLY];
692 if (f->fi_fds[O_RDONLY])
693 return f->fi_fds[O_RDONLY];
697 static atomic_long_t num_delegations;
698 unsigned long max_delegations;
701 * Open owner state (share locks)
704 /* hash tables for lock and open owners */
705 #define OWNER_HASH_BITS 8
706 #define OWNER_HASH_SIZE (1 << OWNER_HASH_BITS)
707 #define OWNER_HASH_MASK (OWNER_HASH_SIZE - 1)
709 static unsigned int ownerstr_hashval(struct xdr_netobj *ownername)
713 ret = opaque_hashval(ownername->data, ownername->len);
714 return ret & OWNER_HASH_MASK;
717 static struct rhltable nfs4_file_rhltable ____cacheline_aligned_in_smp;
719 static const struct rhashtable_params nfs4_file_rhash_params = {
720 .key_len = sizeof_field(struct nfs4_file, fi_inode),
721 .key_offset = offsetof(struct nfs4_file, fi_inode),
722 .head_offset = offsetof(struct nfs4_file, fi_rlist),
725 * Start with a single page hash table to reduce resizing churn
726 * on light workloads.
729 .automatic_shrinking = true,
733 * Check if courtesy clients have conflicting access and resolve it if possible
735 * access: is op_share_access if share_access is true.
736 * Check if access mode, op_share_access, would conflict with
737 * the current deny mode of the file 'fp'.
738 * access: is op_share_deny if share_access is false.
739 * Check if the deny mode, op_share_deny, would conflict with
740 * current access of the file 'fp'.
741 * stp: skip checking this entry.
742 * new_stp: normal open, not open upgrade.
745 * false - access/deny mode conflict with normal client.
746 * true - no conflict or conflict with courtesy client(s) is resolved.
749 nfs4_resolve_deny_conflicts_locked(struct nfs4_file *fp, bool new_stp,
750 struct nfs4_ol_stateid *stp, u32 access, bool share_access)
752 struct nfs4_ol_stateid *st;
753 bool resolvable = true;
756 struct nfs4_client *clp;
758 lockdep_assert_held(&fp->fi_lock);
759 list_for_each_entry(st, &fp->fi_stateids, st_perfile) {
760 /* ignore lock stateid */
763 if (st == stp && new_stp)
765 /* check file access against deny mode or vice versa */
766 bmap = share_access ? st->st_deny_bmap : st->st_access_bmap;
767 if (!(access & bmap_to_share_mode(bmap)))
769 clp = st->st_stid.sc_client;
770 if (try_to_expire_client(clp))
776 clp = stp->st_stid.sc_client;
777 nn = net_generic(clp->net, nfsd_net_id);
778 mod_delayed_work(laundry_wq, &nn->laundromat_work, 0);
784 __nfs4_file_get_access(struct nfs4_file *fp, u32 access)
786 lockdep_assert_held(&fp->fi_lock);
788 if (access & NFS4_SHARE_ACCESS_WRITE)
789 atomic_inc(&fp->fi_access[O_WRONLY]);
790 if (access & NFS4_SHARE_ACCESS_READ)
791 atomic_inc(&fp->fi_access[O_RDONLY]);
795 nfs4_file_get_access(struct nfs4_file *fp, u32 access)
797 lockdep_assert_held(&fp->fi_lock);
799 /* Does this access mode make sense? */
800 if (access & ~NFS4_SHARE_ACCESS_BOTH)
803 /* Does it conflict with a deny mode already set? */
804 if ((access & fp->fi_share_deny) != 0)
805 return nfserr_share_denied;
807 __nfs4_file_get_access(fp, access);
811 static __be32 nfs4_file_check_deny(struct nfs4_file *fp, u32 deny)
813 /* Common case is that there is no deny mode. */
815 /* Does this deny mode make sense? */
816 if (deny & ~NFS4_SHARE_DENY_BOTH)
819 if ((deny & NFS4_SHARE_DENY_READ) &&
820 atomic_read(&fp->fi_access[O_RDONLY]))
821 return nfserr_share_denied;
823 if ((deny & NFS4_SHARE_DENY_WRITE) &&
824 atomic_read(&fp->fi_access[O_WRONLY]))
825 return nfserr_share_denied;
830 static void __nfs4_file_put_access(struct nfs4_file *fp, int oflag)
832 might_lock(&fp->fi_lock);
834 if (atomic_dec_and_lock(&fp->fi_access[oflag], &fp->fi_lock)) {
835 struct nfsd_file *f1 = NULL;
836 struct nfsd_file *f2 = NULL;
838 swap(f1, fp->fi_fds[oflag]);
839 if (atomic_read(&fp->fi_access[1 - oflag]) == 0)
840 swap(f2, fp->fi_fds[O_RDWR]);
841 spin_unlock(&fp->fi_lock);
849 static void nfs4_file_put_access(struct nfs4_file *fp, u32 access)
851 WARN_ON_ONCE(access & ~NFS4_SHARE_ACCESS_BOTH);
853 if (access & NFS4_SHARE_ACCESS_WRITE)
854 __nfs4_file_put_access(fp, O_WRONLY);
855 if (access & NFS4_SHARE_ACCESS_READ)
856 __nfs4_file_put_access(fp, O_RDONLY);
860 * Allocate a new open/delegation state counter. This is needed for
861 * pNFS for proper return on close semantics.
863 * Note that we only allocate it for pNFS-enabled exports, otherwise
864 * all pointers to struct nfs4_clnt_odstate are always NULL.
866 static struct nfs4_clnt_odstate *
867 alloc_clnt_odstate(struct nfs4_client *clp)
869 struct nfs4_clnt_odstate *co;
871 co = kmem_cache_zalloc(odstate_slab, GFP_KERNEL);
874 refcount_set(&co->co_odcount, 1);
880 hash_clnt_odstate_locked(struct nfs4_clnt_odstate *co)
882 struct nfs4_file *fp = co->co_file;
884 lockdep_assert_held(&fp->fi_lock);
885 list_add(&co->co_perfile, &fp->fi_clnt_odstate);
889 get_clnt_odstate(struct nfs4_clnt_odstate *co)
892 refcount_inc(&co->co_odcount);
896 put_clnt_odstate(struct nfs4_clnt_odstate *co)
898 struct nfs4_file *fp;
904 if (refcount_dec_and_lock(&co->co_odcount, &fp->fi_lock)) {
905 list_del(&co->co_perfile);
906 spin_unlock(&fp->fi_lock);
908 nfsd4_return_all_file_layouts(co->co_client, fp);
909 kmem_cache_free(odstate_slab, co);
913 static struct nfs4_clnt_odstate *
914 find_or_hash_clnt_odstate(struct nfs4_file *fp, struct nfs4_clnt_odstate *new)
916 struct nfs4_clnt_odstate *co;
917 struct nfs4_client *cl;
924 spin_lock(&fp->fi_lock);
925 list_for_each_entry(co, &fp->fi_clnt_odstate, co_perfile) {
926 if (co->co_client == cl) {
927 get_clnt_odstate(co);
933 hash_clnt_odstate_locked(new);
935 spin_unlock(&fp->fi_lock);
939 struct nfs4_stid *nfs4_alloc_stid(struct nfs4_client *cl, struct kmem_cache *slab,
940 void (*sc_free)(struct nfs4_stid *))
942 struct nfs4_stid *stid;
945 stid = kmem_cache_zalloc(slab, GFP_KERNEL);
949 idr_preload(GFP_KERNEL);
950 spin_lock(&cl->cl_lock);
951 /* Reserving 0 for start of file in nfsdfs "states" file: */
952 new_id = idr_alloc_cyclic(&cl->cl_stateids, stid, 1, 0, GFP_NOWAIT);
953 spin_unlock(&cl->cl_lock);
958 stid->sc_free = sc_free;
959 stid->sc_client = cl;
960 stid->sc_stateid.si_opaque.so_id = new_id;
961 stid->sc_stateid.si_opaque.so_clid = cl->cl_clientid;
962 /* Will be incremented before return to client: */
963 refcount_set(&stid->sc_count, 1);
964 spin_lock_init(&stid->sc_lock);
965 INIT_LIST_HEAD(&stid->sc_cp_list);
968 * It shouldn't be a problem to reuse an opaque stateid value.
969 * I don't think it is for 4.1. But with 4.0 I worry that, for
970 * example, a stray write retransmission could be accepted by
971 * the server when it should have been rejected. Therefore,
972 * adopt a trick from the sctp code to attempt to maximize the
973 * amount of time until an id is reused, by ensuring they always
974 * "increase" (mod INT_MAX):
978 kmem_cache_free(slab, stid);
983 * Create a unique stateid_t to represent each COPY.
985 static int nfs4_init_cp_state(struct nfsd_net *nn, copy_stateid_t *stid,
986 unsigned char cs_type)
990 stid->cs_stid.si_opaque.so_clid.cl_boot = (u32)nn->boot_time;
991 stid->cs_stid.si_opaque.so_clid.cl_id = nn->s2s_cp_cl_id;
993 idr_preload(GFP_KERNEL);
994 spin_lock(&nn->s2s_cp_lock);
995 new_id = idr_alloc_cyclic(&nn->s2s_cp_stateids, stid, 0, 0, GFP_NOWAIT);
996 stid->cs_stid.si_opaque.so_id = new_id;
997 stid->cs_stid.si_generation = 1;
998 spin_unlock(&nn->s2s_cp_lock);
1002 stid->cs_type = cs_type;
1006 int nfs4_init_copy_state(struct nfsd_net *nn, struct nfsd4_copy *copy)
1008 return nfs4_init_cp_state(nn, ©->cp_stateid, NFS4_COPY_STID);
1011 struct nfs4_cpntf_state *nfs4_alloc_init_cpntf_state(struct nfsd_net *nn,
1012 struct nfs4_stid *p_stid)
1014 struct nfs4_cpntf_state *cps;
1016 cps = kzalloc(sizeof(struct nfs4_cpntf_state), GFP_KERNEL);
1019 cps->cpntf_time = ktime_get_boottime_seconds();
1020 refcount_set(&cps->cp_stateid.cs_count, 1);
1021 if (!nfs4_init_cp_state(nn, &cps->cp_stateid, NFS4_COPYNOTIFY_STID))
1023 spin_lock(&nn->s2s_cp_lock);
1024 list_add(&cps->cp_list, &p_stid->sc_cp_list);
1025 spin_unlock(&nn->s2s_cp_lock);
1032 void nfs4_free_copy_state(struct nfsd4_copy *copy)
1034 struct nfsd_net *nn;
1036 if (copy->cp_stateid.cs_type != NFS4_COPY_STID)
1038 nn = net_generic(copy->cp_clp->net, nfsd_net_id);
1039 spin_lock(&nn->s2s_cp_lock);
1040 idr_remove(&nn->s2s_cp_stateids,
1041 copy->cp_stateid.cs_stid.si_opaque.so_id);
1042 spin_unlock(&nn->s2s_cp_lock);
1045 static void nfs4_free_cpntf_statelist(struct net *net, struct nfs4_stid *stid)
1047 struct nfs4_cpntf_state *cps;
1048 struct nfsd_net *nn;
1050 nn = net_generic(net, nfsd_net_id);
1051 spin_lock(&nn->s2s_cp_lock);
1052 while (!list_empty(&stid->sc_cp_list)) {
1053 cps = list_first_entry(&stid->sc_cp_list,
1054 struct nfs4_cpntf_state, cp_list);
1055 _free_cpntf_state_locked(nn, cps);
1057 spin_unlock(&nn->s2s_cp_lock);
1060 static struct nfs4_ol_stateid * nfs4_alloc_open_stateid(struct nfs4_client *clp)
1062 struct nfs4_stid *stid;
1064 stid = nfs4_alloc_stid(clp, stateid_slab, nfs4_free_ol_stateid);
1068 return openlockstateid(stid);
1071 static void nfs4_free_deleg(struct nfs4_stid *stid)
1073 struct nfs4_delegation *dp = delegstateid(stid);
1075 WARN_ON_ONCE(!list_empty(&stid->sc_cp_list));
1076 WARN_ON_ONCE(!list_empty(&dp->dl_perfile));
1077 WARN_ON_ONCE(!list_empty(&dp->dl_perclnt));
1078 WARN_ON_ONCE(!list_empty(&dp->dl_recall_lru));
1079 kmem_cache_free(deleg_slab, stid);
1080 atomic_long_dec(&num_delegations);
1084 * When we recall a delegation, we should be careful not to hand it
1085 * out again straight away.
1086 * To ensure this we keep a pair of bloom filters ('new' and 'old')
1087 * in which the filehandles of recalled delegations are "stored".
1088 * If a filehandle appear in either filter, a delegation is blocked.
1089 * When a delegation is recalled, the filehandle is stored in the "new"
1091 * Every 30 seconds we swap the filters and clear the "new" one,
1092 * unless both are empty of course.
1094 * Each filter is 256 bits. We hash the filehandle to 32bit and use the
1095 * low 3 bytes as hash-table indices.
1097 * 'blocked_delegations_lock', which is always taken in block_delegations(),
1098 * is used to manage concurrent access. Testing does not need the lock
1099 * except when swapping the two filters.
1101 static DEFINE_SPINLOCK(blocked_delegations_lock);
1102 static struct bloom_pair {
1103 int entries, old_entries;
1105 int new; /* index into 'set' */
1106 DECLARE_BITMAP(set[2], 256);
1107 } blocked_delegations;
1109 static int delegation_blocked(struct knfsd_fh *fh)
1112 struct bloom_pair *bd = &blocked_delegations;
1114 if (bd->entries == 0)
1116 if (ktime_get_seconds() - bd->swap_time > 30) {
1117 spin_lock(&blocked_delegations_lock);
1118 if (ktime_get_seconds() - bd->swap_time > 30) {
1119 bd->entries -= bd->old_entries;
1120 bd->old_entries = bd->entries;
1121 memset(bd->set[bd->new], 0,
1122 sizeof(bd->set[0]));
1123 bd->new = 1-bd->new;
1124 bd->swap_time = ktime_get_seconds();
1126 spin_unlock(&blocked_delegations_lock);
1128 hash = jhash(&fh->fh_raw, fh->fh_size, 0);
1129 if (test_bit(hash&255, bd->set[0]) &&
1130 test_bit((hash>>8)&255, bd->set[0]) &&
1131 test_bit((hash>>16)&255, bd->set[0]))
1134 if (test_bit(hash&255, bd->set[1]) &&
1135 test_bit((hash>>8)&255, bd->set[1]) &&
1136 test_bit((hash>>16)&255, bd->set[1]))
1142 static void block_delegations(struct knfsd_fh *fh)
1145 struct bloom_pair *bd = &blocked_delegations;
1147 hash = jhash(&fh->fh_raw, fh->fh_size, 0);
1149 spin_lock(&blocked_delegations_lock);
1150 __set_bit(hash&255, bd->set[bd->new]);
1151 __set_bit((hash>>8)&255, bd->set[bd->new]);
1152 __set_bit((hash>>16)&255, bd->set[bd->new]);
1153 if (bd->entries == 0)
1154 bd->swap_time = ktime_get_seconds();
1156 spin_unlock(&blocked_delegations_lock);
1159 static struct nfs4_delegation *
1160 alloc_init_deleg(struct nfs4_client *clp, struct nfs4_file *fp,
1161 struct nfs4_clnt_odstate *odstate, u32 dl_type)
1163 struct nfs4_delegation *dp;
1164 struct nfs4_stid *stid;
1167 dprintk("NFSD alloc_init_deleg\n");
1168 n = atomic_long_inc_return(&num_delegations);
1169 if (n < 0 || n > max_delegations)
1171 if (delegation_blocked(&fp->fi_fhandle))
1173 stid = nfs4_alloc_stid(clp, deleg_slab, nfs4_free_deleg);
1176 dp = delegstateid(stid);
1179 * delegation seqid's are never incremented. The 4.1 special
1180 * meaning of seqid 0 isn't meaningful, really, but let's avoid
1181 * 0 anyway just for consistency and use 1:
1183 dp->dl_stid.sc_stateid.si_generation = 1;
1184 INIT_LIST_HEAD(&dp->dl_perfile);
1185 INIT_LIST_HEAD(&dp->dl_perclnt);
1186 INIT_LIST_HEAD(&dp->dl_recall_lru);
1187 dp->dl_clnt_odstate = odstate;
1188 get_clnt_odstate(odstate);
1189 dp->dl_type = dl_type;
1191 dp->dl_recalled = false;
1192 nfsd4_init_cb(&dp->dl_recall, dp->dl_stid.sc_client,
1193 &nfsd4_cb_recall_ops, NFSPROC4_CLNT_CB_RECALL);
1194 nfsd4_init_cb(&dp->dl_cb_fattr.ncf_getattr, dp->dl_stid.sc_client,
1195 &nfsd4_cb_getattr_ops, NFSPROC4_CLNT_CB_GETATTR);
1196 dp->dl_cb_fattr.ncf_file_modified = false;
1197 dp->dl_cb_fattr.ncf_cb_bmap[0] = FATTR4_WORD0_CHANGE | FATTR4_WORD0_SIZE;
1199 dp->dl_stid.sc_file = fp;
1202 atomic_long_dec(&num_delegations);
1207 nfs4_put_stid(struct nfs4_stid *s)
1209 struct nfs4_file *fp = s->sc_file;
1210 struct nfs4_client *clp = s->sc_client;
1212 might_lock(&clp->cl_lock);
1214 if (!refcount_dec_and_lock(&s->sc_count, &clp->cl_lock)) {
1215 wake_up_all(&close_wq);
1218 idr_remove(&clp->cl_stateids, s->sc_stateid.si_opaque.so_id);
1219 if (s->sc_status & SC_STATUS_ADMIN_REVOKED)
1220 atomic_dec(&s->sc_client->cl_admin_revoked);
1221 nfs4_free_cpntf_statelist(clp->net, s);
1222 spin_unlock(&clp->cl_lock);
1229 nfs4_inc_and_copy_stateid(stateid_t *dst, struct nfs4_stid *stid)
1231 stateid_t *src = &stid->sc_stateid;
1233 spin_lock(&stid->sc_lock);
1234 if (unlikely(++src->si_generation == 0))
1235 src->si_generation = 1;
1236 memcpy(dst, src, sizeof(*dst));
1237 spin_unlock(&stid->sc_lock);
1240 static void put_deleg_file(struct nfs4_file *fp)
1242 struct nfsd_file *nf = NULL;
1244 spin_lock(&fp->fi_lock);
1245 if (--fp->fi_delegees == 0)
1246 swap(nf, fp->fi_deleg_file);
1247 spin_unlock(&fp->fi_lock);
1253 static void nfs4_unlock_deleg_lease(struct nfs4_delegation *dp)
1255 struct nfs4_file *fp = dp->dl_stid.sc_file;
1256 struct nfsd_file *nf = fp->fi_deleg_file;
1258 WARN_ON_ONCE(!fp->fi_delegees);
1260 kernel_setlease(nf->nf_file, F_UNLCK, NULL, (void **)&dp);
1264 static void destroy_unhashed_deleg(struct nfs4_delegation *dp)
1266 put_clnt_odstate(dp->dl_clnt_odstate);
1267 nfs4_unlock_deleg_lease(dp);
1268 nfs4_put_stid(&dp->dl_stid);
1272 * nfs4_delegation_exists - Discover if this delegation already exists
1273 * @clp: a pointer to the nfs4_client we're granting a delegation to
1274 * @fp: a pointer to the nfs4_file we're granting a delegation on
1277 * On success: true iff an existing delegation is found
1281 nfs4_delegation_exists(struct nfs4_client *clp, struct nfs4_file *fp)
1283 struct nfs4_delegation *searchdp = NULL;
1284 struct nfs4_client *searchclp = NULL;
1286 lockdep_assert_held(&state_lock);
1287 lockdep_assert_held(&fp->fi_lock);
1289 list_for_each_entry(searchdp, &fp->fi_delegations, dl_perfile) {
1290 searchclp = searchdp->dl_stid.sc_client;
1291 if (clp == searchclp) {
1299 * hash_delegation_locked - Add a delegation to the appropriate lists
1300 * @dp: a pointer to the nfs4_delegation we are adding.
1301 * @fp: a pointer to the nfs4_file we're granting a delegation on
1304 * On success: NULL if the delegation was successfully hashed.
1306 * On error: -EAGAIN if one was previously granted to this
1307 * nfs4_client for this nfs4_file. Delegation is not hashed.
1312 hash_delegation_locked(struct nfs4_delegation *dp, struct nfs4_file *fp)
1314 struct nfs4_client *clp = dp->dl_stid.sc_client;
1316 lockdep_assert_held(&state_lock);
1317 lockdep_assert_held(&fp->fi_lock);
1318 lockdep_assert_held(&clp->cl_lock);
1320 if (nfs4_delegation_exists(clp, fp))
1322 refcount_inc(&dp->dl_stid.sc_count);
1323 dp->dl_stid.sc_type = SC_TYPE_DELEG;
1324 list_add(&dp->dl_perfile, &fp->fi_delegations);
1325 list_add(&dp->dl_perclnt, &clp->cl_delegations);
1329 static bool delegation_hashed(struct nfs4_delegation *dp)
1331 return !(list_empty(&dp->dl_perfile));
1335 unhash_delegation_locked(struct nfs4_delegation *dp, unsigned short statusmask)
1337 struct nfs4_file *fp = dp->dl_stid.sc_file;
1339 lockdep_assert_held(&state_lock);
1341 if (!delegation_hashed(dp))
1344 if (statusmask == SC_STATUS_REVOKED &&
1345 dp->dl_stid.sc_client->cl_minorversion == 0)
1346 statusmask = SC_STATUS_CLOSED;
1347 dp->dl_stid.sc_status |= statusmask;
1348 if (statusmask & SC_STATUS_ADMIN_REVOKED)
1349 atomic_inc(&dp->dl_stid.sc_client->cl_admin_revoked);
1351 /* Ensure that deleg break won't try to requeue it */
1353 spin_lock(&fp->fi_lock);
1354 list_del_init(&dp->dl_perclnt);
1355 list_del_init(&dp->dl_recall_lru);
1356 list_del_init(&dp->dl_perfile);
1357 spin_unlock(&fp->fi_lock);
1361 static void destroy_delegation(struct nfs4_delegation *dp)
1365 spin_lock(&state_lock);
1366 unhashed = unhash_delegation_locked(dp, SC_STATUS_CLOSED);
1367 spin_unlock(&state_lock);
1369 destroy_unhashed_deleg(dp);
1372 static void revoke_delegation(struct nfs4_delegation *dp)
1374 struct nfs4_client *clp = dp->dl_stid.sc_client;
1376 WARN_ON(!list_empty(&dp->dl_recall_lru));
1378 trace_nfsd_stid_revoke(&dp->dl_stid);
1380 if (dp->dl_stid.sc_status &
1381 (SC_STATUS_REVOKED | SC_STATUS_ADMIN_REVOKED)) {
1382 spin_lock(&clp->cl_lock);
1383 refcount_inc(&dp->dl_stid.sc_count);
1384 list_add(&dp->dl_recall_lru, &clp->cl_revoked);
1385 spin_unlock(&clp->cl_lock);
1387 destroy_unhashed_deleg(dp);
1394 static unsigned int clientid_hashval(u32 id)
1396 return id & CLIENT_HASH_MASK;
1399 static unsigned int clientstr_hashval(struct xdr_netobj name)
1401 return opaque_hashval(name.data, 8) & CLIENT_HASH_MASK;
1405 * A stateid that had a deny mode associated with it is being released
1406 * or downgraded. Recalculate the deny mode on the file.
1409 recalculate_deny_mode(struct nfs4_file *fp)
1411 struct nfs4_ol_stateid *stp;
1413 spin_lock(&fp->fi_lock);
1414 fp->fi_share_deny = 0;
1415 list_for_each_entry(stp, &fp->fi_stateids, st_perfile)
1416 fp->fi_share_deny |= bmap_to_share_mode(stp->st_deny_bmap);
1417 spin_unlock(&fp->fi_lock);
1421 reset_union_bmap_deny(u32 deny, struct nfs4_ol_stateid *stp)
1424 bool change = false;
1426 for (i = 1; i < 4; i++) {
1427 if ((i & deny) != i) {
1433 /* Recalculate per-file deny mode if there was a change */
1435 recalculate_deny_mode(stp->st_stid.sc_file);
1438 /* release all access and file references for a given stateid */
1440 release_all_access(struct nfs4_ol_stateid *stp)
1443 struct nfs4_file *fp = stp->st_stid.sc_file;
1445 if (fp && stp->st_deny_bmap != 0)
1446 recalculate_deny_mode(fp);
1448 for (i = 1; i < 4; i++) {
1449 if (test_access(i, stp))
1450 nfs4_file_put_access(stp->st_stid.sc_file, i);
1451 clear_access(i, stp);
1455 static inline void nfs4_free_stateowner(struct nfs4_stateowner *sop)
1457 kfree(sop->so_owner.data);
1458 sop->so_ops->so_free(sop);
1461 static void nfs4_put_stateowner(struct nfs4_stateowner *sop)
1463 struct nfs4_client *clp = sop->so_client;
1465 might_lock(&clp->cl_lock);
1467 if (!atomic_dec_and_lock(&sop->so_count, &clp->cl_lock))
1469 sop->so_ops->so_unhash(sop);
1470 spin_unlock(&clp->cl_lock);
1471 nfs4_free_stateowner(sop);
1475 nfs4_ol_stateid_unhashed(const struct nfs4_ol_stateid *stp)
1477 return list_empty(&stp->st_perfile);
1480 static bool unhash_ol_stateid(struct nfs4_ol_stateid *stp)
1482 struct nfs4_file *fp = stp->st_stid.sc_file;
1484 lockdep_assert_held(&stp->st_stateowner->so_client->cl_lock);
1486 if (list_empty(&stp->st_perfile))
1489 spin_lock(&fp->fi_lock);
1490 list_del_init(&stp->st_perfile);
1491 spin_unlock(&fp->fi_lock);
1492 list_del(&stp->st_perstateowner);
1496 static void nfs4_free_ol_stateid(struct nfs4_stid *stid)
1498 struct nfs4_ol_stateid *stp = openlockstateid(stid);
1500 put_clnt_odstate(stp->st_clnt_odstate);
1501 release_all_access(stp);
1502 if (stp->st_stateowner)
1503 nfs4_put_stateowner(stp->st_stateowner);
1504 WARN_ON(!list_empty(&stid->sc_cp_list));
1505 kmem_cache_free(stateid_slab, stid);
1508 static void nfs4_free_lock_stateid(struct nfs4_stid *stid)
1510 struct nfs4_ol_stateid *stp = openlockstateid(stid);
1511 struct nfs4_lockowner *lo = lockowner(stp->st_stateowner);
1512 struct nfsd_file *nf;
1514 nf = find_any_file(stp->st_stid.sc_file);
1516 get_file(nf->nf_file);
1517 filp_close(nf->nf_file, (fl_owner_t)lo);
1520 nfs4_free_ol_stateid(stid);
1524 * Put the persistent reference to an already unhashed generic stateid, while
1525 * holding the cl_lock. If it's the last reference, then put it onto the
1526 * reaplist for later destruction.
1528 static void put_ol_stateid_locked(struct nfs4_ol_stateid *stp,
1529 struct list_head *reaplist)
1531 struct nfs4_stid *s = &stp->st_stid;
1532 struct nfs4_client *clp = s->sc_client;
1534 lockdep_assert_held(&clp->cl_lock);
1536 WARN_ON_ONCE(!list_empty(&stp->st_locks));
1538 if (!refcount_dec_and_test(&s->sc_count)) {
1539 wake_up_all(&close_wq);
1543 idr_remove(&clp->cl_stateids, s->sc_stateid.si_opaque.so_id);
1544 if (s->sc_status & SC_STATUS_ADMIN_REVOKED)
1545 atomic_dec(&s->sc_client->cl_admin_revoked);
1546 list_add(&stp->st_locks, reaplist);
1549 static bool unhash_lock_stateid(struct nfs4_ol_stateid *stp)
1551 lockdep_assert_held(&stp->st_stid.sc_client->cl_lock);
1553 if (!unhash_ol_stateid(stp))
1555 list_del_init(&stp->st_locks);
1556 stp->st_stid.sc_status |= SC_STATUS_CLOSED;
1560 static void release_lock_stateid(struct nfs4_ol_stateid *stp)
1562 struct nfs4_client *clp = stp->st_stid.sc_client;
1565 spin_lock(&clp->cl_lock);
1566 unhashed = unhash_lock_stateid(stp);
1567 spin_unlock(&clp->cl_lock);
1569 nfs4_put_stid(&stp->st_stid);
1572 static void unhash_lockowner_locked(struct nfs4_lockowner *lo)
1574 struct nfs4_client *clp = lo->lo_owner.so_client;
1576 lockdep_assert_held(&clp->cl_lock);
1578 list_del_init(&lo->lo_owner.so_strhash);
1582 * Free a list of generic stateids that were collected earlier after being
1586 free_ol_stateid_reaplist(struct list_head *reaplist)
1588 struct nfs4_ol_stateid *stp;
1589 struct nfs4_file *fp;
1593 while (!list_empty(reaplist)) {
1594 stp = list_first_entry(reaplist, struct nfs4_ol_stateid,
1596 list_del(&stp->st_locks);
1597 fp = stp->st_stid.sc_file;
1598 stp->st_stid.sc_free(&stp->st_stid);
1604 static void release_open_stateid_locks(struct nfs4_ol_stateid *open_stp,
1605 struct list_head *reaplist)
1607 struct nfs4_ol_stateid *stp;
1609 lockdep_assert_held(&open_stp->st_stid.sc_client->cl_lock);
1611 while (!list_empty(&open_stp->st_locks)) {
1612 stp = list_entry(open_stp->st_locks.next,
1613 struct nfs4_ol_stateid, st_locks);
1614 unhash_lock_stateid(stp);
1615 put_ol_stateid_locked(stp, reaplist);
1619 static bool unhash_open_stateid(struct nfs4_ol_stateid *stp,
1620 struct list_head *reaplist)
1622 lockdep_assert_held(&stp->st_stid.sc_client->cl_lock);
1624 if (!unhash_ol_stateid(stp))
1626 release_open_stateid_locks(stp, reaplist);
1630 static void release_open_stateid(struct nfs4_ol_stateid *stp)
1632 LIST_HEAD(reaplist);
1634 spin_lock(&stp->st_stid.sc_client->cl_lock);
1635 stp->st_stid.sc_status |= SC_STATUS_CLOSED;
1636 if (unhash_open_stateid(stp, &reaplist))
1637 put_ol_stateid_locked(stp, &reaplist);
1638 spin_unlock(&stp->st_stid.sc_client->cl_lock);
1639 free_ol_stateid_reaplist(&reaplist);
1642 static void unhash_openowner_locked(struct nfs4_openowner *oo)
1644 struct nfs4_client *clp = oo->oo_owner.so_client;
1646 lockdep_assert_held(&clp->cl_lock);
1648 list_del_init(&oo->oo_owner.so_strhash);
1649 list_del_init(&oo->oo_perclient);
1652 static void release_last_closed_stateid(struct nfs4_openowner *oo)
1654 struct nfsd_net *nn = net_generic(oo->oo_owner.so_client->net,
1656 struct nfs4_ol_stateid *s;
1658 spin_lock(&nn->client_lock);
1659 s = oo->oo_last_closed_stid;
1661 list_del_init(&oo->oo_close_lru);
1662 oo->oo_last_closed_stid = NULL;
1664 spin_unlock(&nn->client_lock);
1666 nfs4_put_stid(&s->st_stid);
1669 static void release_openowner(struct nfs4_openowner *oo)
1671 struct nfs4_ol_stateid *stp;
1672 struct nfs4_client *clp = oo->oo_owner.so_client;
1673 struct list_head reaplist;
1675 INIT_LIST_HEAD(&reaplist);
1677 spin_lock(&clp->cl_lock);
1678 unhash_openowner_locked(oo);
1679 while (!list_empty(&oo->oo_owner.so_stateids)) {
1680 stp = list_first_entry(&oo->oo_owner.so_stateids,
1681 struct nfs4_ol_stateid, st_perstateowner);
1682 if (unhash_open_stateid(stp, &reaplist))
1683 put_ol_stateid_locked(stp, &reaplist);
1685 spin_unlock(&clp->cl_lock);
1686 free_ol_stateid_reaplist(&reaplist);
1687 release_last_closed_stateid(oo);
1688 nfs4_put_stateowner(&oo->oo_owner);
1691 static struct nfs4_stid *find_one_sb_stid(struct nfs4_client *clp,
1692 struct super_block *sb,
1693 unsigned int sc_types)
1695 unsigned long id, tmp;
1696 struct nfs4_stid *stid;
1698 spin_lock(&clp->cl_lock);
1699 idr_for_each_entry_ul(&clp->cl_stateids, stid, tmp, id)
1700 if ((stid->sc_type & sc_types) &&
1701 stid->sc_status == 0 &&
1702 stid->sc_file->fi_inode->i_sb == sb) {
1703 refcount_inc(&stid->sc_count);
1706 spin_unlock(&clp->cl_lock);
1711 * nfsd4_revoke_states - revoke all nfsv4 states associated with given filesystem
1712 * @net: used to identify instance of nfsd (there is one per net namespace)
1713 * @sb: super_block used to identify target filesystem
1715 * All nfs4 states (open, lock, delegation, layout) held by the server instance
1716 * and associated with a file on the given filesystem will be revoked resulting
1717 * in any files being closed and so all references from nfsd to the filesystem
1718 * being released. Thus nfsd will no longer prevent the filesystem from being
1721 * The clients which own the states will subsequently being notified that the
1722 * states have been "admin-revoked".
1724 void nfsd4_revoke_states(struct net *net, struct super_block *sb)
1726 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
1727 unsigned int idhashval;
1728 unsigned int sc_types;
1730 sc_types = SC_TYPE_OPEN | SC_TYPE_LOCK | SC_TYPE_DELEG | SC_TYPE_LAYOUT;
1732 spin_lock(&nn->client_lock);
1733 for (idhashval = 0; idhashval < CLIENT_HASH_MASK; idhashval++) {
1734 struct list_head *head = &nn->conf_id_hashtbl[idhashval];
1735 struct nfs4_client *clp;
1737 list_for_each_entry(clp, head, cl_idhash) {
1738 struct nfs4_stid *stid = find_one_sb_stid(clp, sb,
1741 struct nfs4_ol_stateid *stp;
1742 struct nfs4_delegation *dp;
1743 struct nfs4_layout_stateid *ls;
1745 spin_unlock(&nn->client_lock);
1746 switch (stid->sc_type) {
1748 stp = openlockstateid(stid);
1749 mutex_lock_nested(&stp->st_mutex,
1750 OPEN_STATEID_MUTEX);
1752 spin_lock(&clp->cl_lock);
1753 if (stid->sc_status == 0) {
1755 SC_STATUS_ADMIN_REVOKED;
1756 atomic_inc(&clp->cl_admin_revoked);
1757 spin_unlock(&clp->cl_lock);
1758 release_all_access(stp);
1760 spin_unlock(&clp->cl_lock);
1761 mutex_unlock(&stp->st_mutex);
1764 stp = openlockstateid(stid);
1765 mutex_lock_nested(&stp->st_mutex,
1766 LOCK_STATEID_MUTEX);
1767 spin_lock(&clp->cl_lock);
1768 if (stid->sc_status == 0) {
1769 struct nfs4_lockowner *lo =
1770 lockowner(stp->st_stateowner);
1771 struct nfsd_file *nf;
1774 SC_STATUS_ADMIN_REVOKED;
1775 atomic_inc(&clp->cl_admin_revoked);
1776 spin_unlock(&clp->cl_lock);
1777 nf = find_any_file(stp->st_stid.sc_file);
1779 get_file(nf->nf_file);
1780 filp_close(nf->nf_file,
1784 release_all_access(stp);
1786 spin_unlock(&clp->cl_lock);
1787 mutex_unlock(&stp->st_mutex);
1790 dp = delegstateid(stid);
1791 spin_lock(&state_lock);
1792 if (!unhash_delegation_locked(
1793 dp, SC_STATUS_ADMIN_REVOKED))
1795 spin_unlock(&state_lock);
1797 revoke_delegation(dp);
1799 case SC_TYPE_LAYOUT:
1800 ls = layoutstateid(stid);
1801 nfsd4_close_layout(ls);
1804 nfs4_put_stid(stid);
1805 spin_lock(&nn->client_lock);
1806 if (clp->cl_minorversion == 0)
1807 /* Allow cleanup after a lease period.
1808 * store_release ensures cleanup will
1809 * see any newly revoked states if it
1810 * sees the time updated.
1812 nn->nfs40_last_revoke =
1813 ktime_get_boottime_seconds();
1818 spin_unlock(&nn->client_lock);
1822 hash_sessionid(struct nfs4_sessionid *sessionid)
1824 struct nfsd4_sessionid *sid = (struct nfsd4_sessionid *)sessionid;
1826 return sid->sequence % SESSION_HASH_SIZE;
1829 #ifdef CONFIG_SUNRPC_DEBUG
1831 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
1833 u32 *ptr = (u32 *)(&sessionid->data[0]);
1834 dprintk("%s: %u:%u:%u:%u\n", fn, ptr[0], ptr[1], ptr[2], ptr[3]);
1838 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
1844 * Bump the seqid on cstate->replay_owner, and clear replay_owner if it
1845 * won't be used for replay.
1847 void nfsd4_bump_seqid(struct nfsd4_compound_state *cstate, __be32 nfserr)
1849 struct nfs4_stateowner *so = cstate->replay_owner;
1851 if (nfserr == nfserr_replay_me)
1854 if (!seqid_mutating_err(ntohl(nfserr))) {
1855 nfsd4_cstate_clear_replay(cstate);
1860 if (so->so_is_open_owner)
1861 release_last_closed_stateid(openowner(so));
1867 gen_sessionid(struct nfsd4_session *ses)
1869 struct nfs4_client *clp = ses->se_client;
1870 struct nfsd4_sessionid *sid;
1872 sid = (struct nfsd4_sessionid *)ses->se_sessionid.data;
1873 sid->clientid = clp->cl_clientid;
1874 sid->sequence = current_sessionid++;
1879 * The protocol defines ca_maxresponssize_cached to include the size of
1880 * the rpc header, but all we need to cache is the data starting after
1881 * the end of the initial SEQUENCE operation--the rest we regenerate
1882 * each time. Therefore we can advertise a ca_maxresponssize_cached
1883 * value that is the number of bytes in our cache plus a few additional
1884 * bytes. In order to stay on the safe side, and not promise more than
1885 * we can cache, those additional bytes must be the minimum possible: 24
1886 * bytes of rpc header (xid through accept state, with AUTH_NULL
1887 * verifier), 12 for the compound header (with zero-length tag), and 44
1888 * for the SEQUENCE op response:
1890 #define NFSD_MIN_HDR_SEQ_SZ (24 + 12 + 44)
1893 free_session_slots(struct nfsd4_session *ses)
1897 for (i = 0; i < ses->se_fchannel.maxreqs; i++) {
1898 free_svc_cred(&ses->se_slots[i]->sl_cred);
1899 kfree(ses->se_slots[i]);
1904 * We don't actually need to cache the rpc and session headers, so we
1905 * can allocate a little less for each slot:
1907 static inline u32 slot_bytes(struct nfsd4_channel_attrs *ca)
1911 if (ca->maxresp_cached < NFSD_MIN_HDR_SEQ_SZ)
1914 size = ca->maxresp_cached - NFSD_MIN_HDR_SEQ_SZ;
1915 return size + sizeof(struct nfsd4_slot);
1919 * XXX: If we run out of reserved DRC memory we could (up to a point)
1920 * re-negotiate active sessions and reduce their slot usage to make
1921 * room for new connections. For now we just fail the create session.
1923 static u32 nfsd4_get_drc_mem(struct nfsd4_channel_attrs *ca, struct nfsd_net *nn)
1925 u32 slotsize = slot_bytes(ca);
1926 u32 num = ca->maxreqs;
1927 unsigned long avail, total_avail;
1928 unsigned int scale_factor;
1930 spin_lock(&nfsd_drc_lock);
1931 if (nfsd_drc_max_mem > nfsd_drc_mem_used)
1932 total_avail = nfsd_drc_max_mem - nfsd_drc_mem_used;
1934 /* We have handed out more space than we chose in
1935 * set_max_drc() to allow. That isn't really a
1936 * problem as long as that doesn't make us think we
1937 * have lots more due to integer overflow.
1940 avail = min((unsigned long)NFSD_MAX_MEM_PER_SESSION, total_avail);
1942 * Never use more than a fraction of the remaining memory,
1943 * unless it's the only way to give this client a slot.
1944 * The chosen fraction is either 1/8 or 1/number of threads,
1945 * whichever is smaller. This ensures there are adequate
1946 * slots to support multiple clients per thread.
1947 * Give the client one slot even if that would require
1948 * over-allocation--it is better than failure.
1950 scale_factor = max_t(unsigned int, 8, nn->nfsd_serv->sv_nrthreads);
1952 avail = clamp_t(unsigned long, avail, slotsize,
1953 total_avail/scale_factor);
1954 num = min_t(int, num, avail / slotsize);
1955 num = max_t(int, num, 1);
1956 nfsd_drc_mem_used += num * slotsize;
1957 spin_unlock(&nfsd_drc_lock);
1962 static void nfsd4_put_drc_mem(struct nfsd4_channel_attrs *ca)
1964 int slotsize = slot_bytes(ca);
1966 spin_lock(&nfsd_drc_lock);
1967 nfsd_drc_mem_used -= slotsize * ca->maxreqs;
1968 spin_unlock(&nfsd_drc_lock);
1971 static struct nfsd4_session *alloc_session(struct nfsd4_channel_attrs *fattrs,
1972 struct nfsd4_channel_attrs *battrs)
1974 int numslots = fattrs->maxreqs;
1975 int slotsize = slot_bytes(fattrs);
1976 struct nfsd4_session *new;
1979 BUILD_BUG_ON(struct_size(new, se_slots, NFSD_MAX_SLOTS_PER_SESSION)
1982 new = kzalloc(struct_size(new, se_slots, numslots), GFP_KERNEL);
1985 /* allocate each struct nfsd4_slot and data cache in one piece */
1986 for (i = 0; i < numslots; i++) {
1987 new->se_slots[i] = kzalloc(slotsize, GFP_KERNEL);
1988 if (!new->se_slots[i])
1992 memcpy(&new->se_fchannel, fattrs, sizeof(struct nfsd4_channel_attrs));
1993 memcpy(&new->se_bchannel, battrs, sizeof(struct nfsd4_channel_attrs));
1998 kfree(new->se_slots[i]);
2003 static void free_conn(struct nfsd4_conn *c)
2005 svc_xprt_put(c->cn_xprt);
2009 static void nfsd4_conn_lost(struct svc_xpt_user *u)
2011 struct nfsd4_conn *c = container_of(u, struct nfsd4_conn, cn_xpt_user);
2012 struct nfs4_client *clp = c->cn_session->se_client;
2014 trace_nfsd_cb_lost(clp);
2016 spin_lock(&clp->cl_lock);
2017 if (!list_empty(&c->cn_persession)) {
2018 list_del(&c->cn_persession);
2021 nfsd4_probe_callback(clp);
2022 spin_unlock(&clp->cl_lock);
2025 static struct nfsd4_conn *alloc_conn(struct svc_rqst *rqstp, u32 flags)
2027 struct nfsd4_conn *conn;
2029 conn = kmalloc(sizeof(struct nfsd4_conn), GFP_KERNEL);
2032 svc_xprt_get(rqstp->rq_xprt);
2033 conn->cn_xprt = rqstp->rq_xprt;
2034 conn->cn_flags = flags;
2035 INIT_LIST_HEAD(&conn->cn_xpt_user.list);
2039 static void __nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
2041 conn->cn_session = ses;
2042 list_add(&conn->cn_persession, &ses->se_conns);
2045 static void nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
2047 struct nfs4_client *clp = ses->se_client;
2049 spin_lock(&clp->cl_lock);
2050 __nfsd4_hash_conn(conn, ses);
2051 spin_unlock(&clp->cl_lock);
2054 static int nfsd4_register_conn(struct nfsd4_conn *conn)
2056 conn->cn_xpt_user.callback = nfsd4_conn_lost;
2057 return register_xpt_user(conn->cn_xprt, &conn->cn_xpt_user);
2060 static void nfsd4_init_conn(struct svc_rqst *rqstp, struct nfsd4_conn *conn, struct nfsd4_session *ses)
2064 nfsd4_hash_conn(conn, ses);
2065 ret = nfsd4_register_conn(conn);
2067 /* oops; xprt is already down: */
2068 nfsd4_conn_lost(&conn->cn_xpt_user);
2069 /* We may have gained or lost a callback channel: */
2070 nfsd4_probe_callback_sync(ses->se_client);
2073 static struct nfsd4_conn *alloc_conn_from_crses(struct svc_rqst *rqstp, struct nfsd4_create_session *cses)
2075 u32 dir = NFS4_CDFC4_FORE;
2077 if (cses->flags & SESSION4_BACK_CHAN)
2078 dir |= NFS4_CDFC4_BACK;
2079 return alloc_conn(rqstp, dir);
2082 /* must be called under client_lock */
2083 static void nfsd4_del_conns(struct nfsd4_session *s)
2085 struct nfs4_client *clp = s->se_client;
2086 struct nfsd4_conn *c;
2088 spin_lock(&clp->cl_lock);
2089 while (!list_empty(&s->se_conns)) {
2090 c = list_first_entry(&s->se_conns, struct nfsd4_conn, cn_persession);
2091 list_del_init(&c->cn_persession);
2092 spin_unlock(&clp->cl_lock);
2094 unregister_xpt_user(c->cn_xprt, &c->cn_xpt_user);
2097 spin_lock(&clp->cl_lock);
2099 spin_unlock(&clp->cl_lock);
2102 static void __free_session(struct nfsd4_session *ses)
2104 free_session_slots(ses);
2108 static void free_session(struct nfsd4_session *ses)
2110 nfsd4_del_conns(ses);
2111 nfsd4_put_drc_mem(&ses->se_fchannel);
2112 __free_session(ses);
2115 static void init_session(struct svc_rqst *rqstp, struct nfsd4_session *new, struct nfs4_client *clp, struct nfsd4_create_session *cses)
2118 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2120 new->se_client = clp;
2123 INIT_LIST_HEAD(&new->se_conns);
2125 new->se_cb_seq_nr = 1;
2126 new->se_flags = cses->flags;
2127 new->se_cb_prog = cses->callback_prog;
2128 new->se_cb_sec = cses->cb_sec;
2129 atomic_set(&new->se_ref, 0);
2130 idx = hash_sessionid(&new->se_sessionid);
2131 list_add(&new->se_hash, &nn->sessionid_hashtbl[idx]);
2132 spin_lock(&clp->cl_lock);
2133 list_add(&new->se_perclnt, &clp->cl_sessions);
2134 spin_unlock(&clp->cl_lock);
2137 struct sockaddr *sa = svc_addr(rqstp);
2139 * This is a little silly; with sessions there's no real
2140 * use for the callback address. Use the peer address
2141 * as a reasonable default for now, but consider fixing
2142 * the rpc client not to require an address in the
2145 rpc_copy_addr((struct sockaddr *)&clp->cl_cb_conn.cb_addr, sa);
2146 clp->cl_cb_conn.cb_addrlen = svc_addr_len(sa);
2150 /* caller must hold client_lock */
2151 static struct nfsd4_session *
2152 __find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid, struct net *net)
2154 struct nfsd4_session *elem;
2156 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
2158 lockdep_assert_held(&nn->client_lock);
2160 dump_sessionid(__func__, sessionid);
2161 idx = hash_sessionid(sessionid);
2162 /* Search in the appropriate list */
2163 list_for_each_entry(elem, &nn->sessionid_hashtbl[idx], se_hash) {
2164 if (!memcmp(elem->se_sessionid.data, sessionid->data,
2165 NFS4_MAX_SESSIONID_LEN)) {
2170 dprintk("%s: session not found\n", __func__);
2174 static struct nfsd4_session *
2175 find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid, struct net *net,
2178 struct nfsd4_session *session;
2179 __be32 status = nfserr_badsession;
2181 session = __find_in_sessionid_hashtbl(sessionid, net);
2184 status = nfsd4_get_session_locked(session);
2192 /* caller must hold client_lock */
2194 unhash_session(struct nfsd4_session *ses)
2196 struct nfs4_client *clp = ses->se_client;
2197 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2199 lockdep_assert_held(&nn->client_lock);
2201 list_del(&ses->se_hash);
2202 spin_lock(&ses->se_client->cl_lock);
2203 list_del(&ses->se_perclnt);
2204 spin_unlock(&ses->se_client->cl_lock);
2207 /* SETCLIENTID and SETCLIENTID_CONFIRM Helper functions */
2209 STALE_CLIENTID(clientid_t *clid, struct nfsd_net *nn)
2212 * We're assuming the clid was not given out from a boot
2213 * precisely 2^32 (about 136 years) before this one. That seems
2214 * a safe assumption:
2216 if (clid->cl_boot == (u32)nn->boot_time)
2218 trace_nfsd_clid_stale(clid);
2223 * XXX Should we use a slab cache ?
2224 * This type of memory management is somewhat inefficient, but we use it
2225 * anyway since SETCLIENTID is not a common operation.
2227 static struct nfs4_client *alloc_client(struct xdr_netobj name,
2228 struct nfsd_net *nn)
2230 struct nfs4_client *clp;
2233 if (atomic_read(&nn->nfs4_client_count) >= nn->nfs4_max_clients) {
2234 mod_delayed_work(laundry_wq, &nn->laundromat_work, 0);
2237 clp = kmem_cache_zalloc(client_slab, GFP_KERNEL);
2240 xdr_netobj_dup(&clp->cl_name, &name, GFP_KERNEL);
2241 if (clp->cl_name.data == NULL)
2243 clp->cl_ownerstr_hashtbl = kmalloc_array(OWNER_HASH_SIZE,
2244 sizeof(struct list_head),
2246 if (!clp->cl_ownerstr_hashtbl)
2247 goto err_no_hashtbl;
2248 for (i = 0; i < OWNER_HASH_SIZE; i++)
2249 INIT_LIST_HEAD(&clp->cl_ownerstr_hashtbl[i]);
2250 INIT_LIST_HEAD(&clp->cl_sessions);
2251 idr_init(&clp->cl_stateids);
2252 atomic_set(&clp->cl_rpc_users, 0);
2253 clp->cl_cb_state = NFSD4_CB_UNKNOWN;
2254 clp->cl_state = NFSD4_ACTIVE;
2255 atomic_inc(&nn->nfs4_client_count);
2256 atomic_set(&clp->cl_delegs_in_recall, 0);
2257 INIT_LIST_HEAD(&clp->cl_idhash);
2258 INIT_LIST_HEAD(&clp->cl_openowners);
2259 INIT_LIST_HEAD(&clp->cl_delegations);
2260 INIT_LIST_HEAD(&clp->cl_lru);
2261 INIT_LIST_HEAD(&clp->cl_revoked);
2262 #ifdef CONFIG_NFSD_PNFS
2263 INIT_LIST_HEAD(&clp->cl_lo_states);
2265 INIT_LIST_HEAD(&clp->async_copies);
2266 spin_lock_init(&clp->async_lock);
2267 spin_lock_init(&clp->cl_lock);
2268 rpc_init_wait_queue(&clp->cl_cb_waitq, "Backchannel slot table");
2271 kfree(clp->cl_name.data);
2273 kmem_cache_free(client_slab, clp);
2277 static void __free_client(struct kref *k)
2279 struct nfsdfs_client *c = container_of(k, struct nfsdfs_client, cl_ref);
2280 struct nfs4_client *clp = container_of(c, struct nfs4_client, cl_nfsdfs);
2282 free_svc_cred(&clp->cl_cred);
2283 kfree(clp->cl_ownerstr_hashtbl);
2284 kfree(clp->cl_name.data);
2285 kfree(clp->cl_nii_domain.data);
2286 kfree(clp->cl_nii_name.data);
2287 idr_destroy(&clp->cl_stateids);
2289 kmem_cache_free(client_slab, clp);
2292 static void drop_client(struct nfs4_client *clp)
2294 kref_put(&clp->cl_nfsdfs.cl_ref, __free_client);
2298 free_client(struct nfs4_client *clp)
2300 while (!list_empty(&clp->cl_sessions)) {
2301 struct nfsd4_session *ses;
2302 ses = list_entry(clp->cl_sessions.next, struct nfsd4_session,
2304 list_del(&ses->se_perclnt);
2305 WARN_ON_ONCE(atomic_read(&ses->se_ref));
2308 rpc_destroy_wait_queue(&clp->cl_cb_waitq);
2309 if (clp->cl_nfsd_dentry) {
2310 nfsd_client_rmdir(clp->cl_nfsd_dentry);
2311 clp->cl_nfsd_dentry = NULL;
2312 wake_up_all(&expiry_wq);
2317 /* must be called under the client_lock */
2319 unhash_client_locked(struct nfs4_client *clp)
2321 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2322 struct nfsd4_session *ses;
2324 lockdep_assert_held(&nn->client_lock);
2326 /* Mark the client as expired! */
2328 /* Make it invisible */
2329 if (!list_empty(&clp->cl_idhash)) {
2330 list_del_init(&clp->cl_idhash);
2331 if (test_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags))
2332 rb_erase(&clp->cl_namenode, &nn->conf_name_tree);
2334 rb_erase(&clp->cl_namenode, &nn->unconf_name_tree);
2336 list_del_init(&clp->cl_lru);
2337 spin_lock(&clp->cl_lock);
2338 list_for_each_entry(ses, &clp->cl_sessions, se_perclnt)
2339 list_del_init(&ses->se_hash);
2340 spin_unlock(&clp->cl_lock);
2344 unhash_client(struct nfs4_client *clp)
2346 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2348 spin_lock(&nn->client_lock);
2349 unhash_client_locked(clp);
2350 spin_unlock(&nn->client_lock);
2353 static __be32 mark_client_expired_locked(struct nfs4_client *clp)
2355 if (atomic_read(&clp->cl_rpc_users))
2356 return nfserr_jukebox;
2357 unhash_client_locked(clp);
2362 __destroy_client(struct nfs4_client *clp)
2364 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2366 struct nfs4_openowner *oo;
2367 struct nfs4_delegation *dp;
2368 struct list_head reaplist;
2370 INIT_LIST_HEAD(&reaplist);
2371 spin_lock(&state_lock);
2372 while (!list_empty(&clp->cl_delegations)) {
2373 dp = list_entry(clp->cl_delegations.next, struct nfs4_delegation, dl_perclnt);
2374 unhash_delegation_locked(dp, SC_STATUS_CLOSED);
2375 list_add(&dp->dl_recall_lru, &reaplist);
2377 spin_unlock(&state_lock);
2378 while (!list_empty(&reaplist)) {
2379 dp = list_entry(reaplist.next, struct nfs4_delegation, dl_recall_lru);
2380 list_del_init(&dp->dl_recall_lru);
2381 destroy_unhashed_deleg(dp);
2383 while (!list_empty(&clp->cl_revoked)) {
2384 dp = list_entry(clp->cl_revoked.next, struct nfs4_delegation, dl_recall_lru);
2385 list_del_init(&dp->dl_recall_lru);
2386 nfs4_put_stid(&dp->dl_stid);
2388 while (!list_empty(&clp->cl_openowners)) {
2389 oo = list_entry(clp->cl_openowners.next, struct nfs4_openowner, oo_perclient);
2390 nfs4_get_stateowner(&oo->oo_owner);
2391 release_openowner(oo);
2393 for (i = 0; i < OWNER_HASH_SIZE; i++) {
2394 struct nfs4_stateowner *so, *tmp;
2396 list_for_each_entry_safe(so, tmp, &clp->cl_ownerstr_hashtbl[i],
2398 /* Should be no openowners at this point */
2399 WARN_ON_ONCE(so->so_is_open_owner);
2400 remove_blocked_locks(lockowner(so));
2403 nfsd4_return_all_client_layouts(clp);
2404 nfsd4_shutdown_copy(clp);
2405 nfsd4_shutdown_callback(clp);
2406 if (clp->cl_cb_conn.cb_xprt)
2407 svc_xprt_put(clp->cl_cb_conn.cb_xprt);
2408 atomic_add_unless(&nn->nfs4_client_count, -1, 0);
2409 nfsd4_dec_courtesy_client_count(nn, clp);
2411 wake_up_all(&expiry_wq);
2415 destroy_client(struct nfs4_client *clp)
2418 __destroy_client(clp);
2421 static void inc_reclaim_complete(struct nfs4_client *clp)
2423 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2425 if (!nn->track_reclaim_completes)
2427 if (!nfsd4_find_reclaim_client(clp->cl_name, nn))
2429 if (atomic_inc_return(&nn->nr_reclaim_complete) ==
2430 nn->reclaim_str_hashtbl_size) {
2431 printk(KERN_INFO "NFSD: all clients done reclaiming, ending NFSv4 grace period (net %x)\n",
2433 nfsd4_end_grace(nn);
2437 static void expire_client(struct nfs4_client *clp)
2440 nfsd4_client_record_remove(clp);
2441 __destroy_client(clp);
2444 static void copy_verf(struct nfs4_client *target, nfs4_verifier *source)
2446 memcpy(target->cl_verifier.data, source->data,
2447 sizeof(target->cl_verifier.data));
2450 static void copy_clid(struct nfs4_client *target, struct nfs4_client *source)
2452 target->cl_clientid.cl_boot = source->cl_clientid.cl_boot;
2453 target->cl_clientid.cl_id = source->cl_clientid.cl_id;
2456 static int copy_cred(struct svc_cred *target, struct svc_cred *source)
2458 target->cr_principal = kstrdup(source->cr_principal, GFP_KERNEL);
2459 target->cr_raw_principal = kstrdup(source->cr_raw_principal,
2461 target->cr_targ_princ = kstrdup(source->cr_targ_princ, GFP_KERNEL);
2462 if ((source->cr_principal && !target->cr_principal) ||
2463 (source->cr_raw_principal && !target->cr_raw_principal) ||
2464 (source->cr_targ_princ && !target->cr_targ_princ))
2467 target->cr_flavor = source->cr_flavor;
2468 target->cr_uid = source->cr_uid;
2469 target->cr_gid = source->cr_gid;
2470 target->cr_group_info = source->cr_group_info;
2471 get_group_info(target->cr_group_info);
2472 target->cr_gss_mech = source->cr_gss_mech;
2473 if (source->cr_gss_mech)
2474 gss_mech_get(source->cr_gss_mech);
2479 compare_blob(const struct xdr_netobj *o1, const struct xdr_netobj *o2)
2481 if (o1->len < o2->len)
2483 if (o1->len > o2->len)
2485 return memcmp(o1->data, o2->data, o1->len);
2489 same_verf(nfs4_verifier *v1, nfs4_verifier *v2)
2491 return 0 == memcmp(v1->data, v2->data, sizeof(v1->data));
2495 same_clid(clientid_t *cl1, clientid_t *cl2)
2497 return (cl1->cl_boot == cl2->cl_boot) && (cl1->cl_id == cl2->cl_id);
2500 static bool groups_equal(struct group_info *g1, struct group_info *g2)
2504 if (g1->ngroups != g2->ngroups)
2506 for (i=0; i<g1->ngroups; i++)
2507 if (!gid_eq(g1->gid[i], g2->gid[i]))
2513 * RFC 3530 language requires clid_inuse be returned when the
2514 * "principal" associated with a requests differs from that previously
2515 * used. We use uid, gid's, and gss principal string as our best
2516 * approximation. We also don't want to allow non-gss use of a client
2517 * established using gss: in theory cr_principal should catch that
2518 * change, but in practice cr_principal can be null even in the gss case
2519 * since gssd doesn't always pass down a principal string.
2521 static bool is_gss_cred(struct svc_cred *cr)
2523 /* Is cr_flavor one of the gss "pseudoflavors"?: */
2524 return (cr->cr_flavor > RPC_AUTH_MAXFLAVOR);
2529 same_creds(struct svc_cred *cr1, struct svc_cred *cr2)
2531 if ((is_gss_cred(cr1) != is_gss_cred(cr2))
2532 || (!uid_eq(cr1->cr_uid, cr2->cr_uid))
2533 || (!gid_eq(cr1->cr_gid, cr2->cr_gid))
2534 || !groups_equal(cr1->cr_group_info, cr2->cr_group_info))
2536 /* XXX: check that cr_targ_princ fields match ? */
2537 if (cr1->cr_principal == cr2->cr_principal)
2539 if (!cr1->cr_principal || !cr2->cr_principal)
2541 return 0 == strcmp(cr1->cr_principal, cr2->cr_principal);
2544 static bool svc_rqst_integrity_protected(struct svc_rqst *rqstp)
2546 struct svc_cred *cr = &rqstp->rq_cred;
2549 if (!cr->cr_gss_mech)
2551 service = gss_pseudoflavor_to_service(cr->cr_gss_mech, cr->cr_flavor);
2552 return service == RPC_GSS_SVC_INTEGRITY ||
2553 service == RPC_GSS_SVC_PRIVACY;
2556 bool nfsd4_mach_creds_match(struct nfs4_client *cl, struct svc_rqst *rqstp)
2558 struct svc_cred *cr = &rqstp->rq_cred;
2560 if (!cl->cl_mach_cred)
2562 if (cl->cl_cred.cr_gss_mech != cr->cr_gss_mech)
2564 if (!svc_rqst_integrity_protected(rqstp))
2566 if (cl->cl_cred.cr_raw_principal)
2567 return 0 == strcmp(cl->cl_cred.cr_raw_principal,
2568 cr->cr_raw_principal);
2569 if (!cr->cr_principal)
2571 return 0 == strcmp(cl->cl_cred.cr_principal, cr->cr_principal);
2574 static void gen_confirm(struct nfs4_client *clp, struct nfsd_net *nn)
2579 * This is opaque to client, so no need to byte-swap. Use
2580 * __force to keep sparse happy
2582 verf[0] = (__force __be32)(u32)ktime_get_real_seconds();
2583 verf[1] = (__force __be32)nn->clverifier_counter++;
2584 memcpy(clp->cl_confirm.data, verf, sizeof(clp->cl_confirm.data));
2587 static void gen_clid(struct nfs4_client *clp, struct nfsd_net *nn)
2589 clp->cl_clientid.cl_boot = (u32)nn->boot_time;
2590 clp->cl_clientid.cl_id = nn->clientid_counter++;
2591 gen_confirm(clp, nn);
2594 static struct nfs4_stid *
2595 find_stateid_locked(struct nfs4_client *cl, stateid_t *t)
2597 struct nfs4_stid *ret;
2599 ret = idr_find(&cl->cl_stateids, t->si_opaque.so_id);
2600 if (!ret || !ret->sc_type)
2605 static struct nfs4_stid *
2606 find_stateid_by_type(struct nfs4_client *cl, stateid_t *t,
2607 unsigned short typemask, unsigned short ok_states)
2609 struct nfs4_stid *s;
2611 spin_lock(&cl->cl_lock);
2612 s = find_stateid_locked(cl, t);
2614 if ((s->sc_status & ~ok_states) == 0 &&
2615 (typemask & s->sc_type))
2616 refcount_inc(&s->sc_count);
2620 spin_unlock(&cl->cl_lock);
2624 static struct nfs4_client *get_nfsdfs_clp(struct inode *inode)
2626 struct nfsdfs_client *nc;
2627 nc = get_nfsdfs_client(inode);
2630 return container_of(nc, struct nfs4_client, cl_nfsdfs);
2633 static void seq_quote_mem(struct seq_file *m, char *data, int len)
2636 seq_escape_mem(m, data, len, ESCAPE_HEX | ESCAPE_NAP | ESCAPE_APPEND, "\"\\");
2640 static const char *cb_state2str(int state)
2645 case NFSD4_CB_UNKNOWN:
2649 case NFSD4_CB_FAULT:
2655 static int client_info_show(struct seq_file *m, void *v)
2657 struct inode *inode = file_inode(m->file);
2658 struct nfs4_client *clp;
2661 clp = get_nfsdfs_clp(inode);
2664 memcpy(&clid, &clp->cl_clientid, sizeof(clid));
2665 seq_printf(m, "clientid: 0x%llx\n", clid);
2666 seq_printf(m, "address: \"%pISpc\"\n", (struct sockaddr *)&clp->cl_addr);
2668 if (clp->cl_state == NFSD4_COURTESY)
2669 seq_puts(m, "status: courtesy\n");
2670 else if (clp->cl_state == NFSD4_EXPIRABLE)
2671 seq_puts(m, "status: expirable\n");
2672 else if (test_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags))
2673 seq_puts(m, "status: confirmed\n");
2675 seq_puts(m, "status: unconfirmed\n");
2676 seq_printf(m, "seconds from last renew: %lld\n",
2677 ktime_get_boottime_seconds() - clp->cl_time);
2678 seq_puts(m, "name: ");
2679 seq_quote_mem(m, clp->cl_name.data, clp->cl_name.len);
2680 seq_printf(m, "\nminor version: %d\n", clp->cl_minorversion);
2681 if (clp->cl_nii_domain.data) {
2682 seq_puts(m, "Implementation domain: ");
2683 seq_quote_mem(m, clp->cl_nii_domain.data,
2684 clp->cl_nii_domain.len);
2685 seq_puts(m, "\nImplementation name: ");
2686 seq_quote_mem(m, clp->cl_nii_name.data, clp->cl_nii_name.len);
2687 seq_printf(m, "\nImplementation time: [%lld, %ld]\n",
2688 clp->cl_nii_time.tv_sec, clp->cl_nii_time.tv_nsec);
2690 seq_printf(m, "callback state: %s\n", cb_state2str(clp->cl_cb_state));
2691 seq_printf(m, "callback address: %pISpc\n", &clp->cl_cb_conn.cb_addr);
2692 seq_printf(m, "admin-revoked states: %d\n",
2693 atomic_read(&clp->cl_admin_revoked));
2699 DEFINE_SHOW_ATTRIBUTE(client_info);
2701 static void *states_start(struct seq_file *s, loff_t *pos)
2702 __acquires(&clp->cl_lock)
2704 struct nfs4_client *clp = s->private;
2705 unsigned long id = *pos;
2708 spin_lock(&clp->cl_lock);
2709 ret = idr_get_next_ul(&clp->cl_stateids, &id);
2714 static void *states_next(struct seq_file *s, void *v, loff_t *pos)
2716 struct nfs4_client *clp = s->private;
2717 unsigned long id = *pos;
2722 ret = idr_get_next_ul(&clp->cl_stateids, &id);
2727 static void states_stop(struct seq_file *s, void *v)
2728 __releases(&clp->cl_lock)
2730 struct nfs4_client *clp = s->private;
2732 spin_unlock(&clp->cl_lock);
2735 static void nfs4_show_fname(struct seq_file *s, struct nfsd_file *f)
2737 seq_printf(s, "filename: \"%pD2\"", f->nf_file);
2740 static void nfs4_show_superblock(struct seq_file *s, struct nfsd_file *f)
2742 struct inode *inode = file_inode(f->nf_file);
2744 seq_printf(s, "superblock: \"%02x:%02x:%ld\"",
2745 MAJOR(inode->i_sb->s_dev),
2746 MINOR(inode->i_sb->s_dev),
2750 static void nfs4_show_owner(struct seq_file *s, struct nfs4_stateowner *oo)
2752 seq_puts(s, "owner: ");
2753 seq_quote_mem(s, oo->so_owner.data, oo->so_owner.len);
2756 static void nfs4_show_stateid(struct seq_file *s, stateid_t *stid)
2758 seq_printf(s, "0x%.8x", stid->si_generation);
2759 seq_printf(s, "%12phN", &stid->si_opaque);
2762 static int nfs4_show_open(struct seq_file *s, struct nfs4_stid *st)
2764 struct nfs4_ol_stateid *ols;
2765 struct nfs4_file *nf;
2766 struct nfsd_file *file;
2767 struct nfs4_stateowner *oo;
2768 unsigned int access, deny;
2770 ols = openlockstateid(st);
2771 oo = ols->st_stateowner;
2775 nfs4_show_stateid(s, &st->sc_stateid);
2776 seq_puts(s, ": { type: open, ");
2778 access = bmap_to_share_mode(ols->st_access_bmap);
2779 deny = bmap_to_share_mode(ols->st_deny_bmap);
2781 seq_printf(s, "access: %s%s, ",
2782 access & NFS4_SHARE_ACCESS_READ ? "r" : "-",
2783 access & NFS4_SHARE_ACCESS_WRITE ? "w" : "-");
2784 seq_printf(s, "deny: %s%s, ",
2785 deny & NFS4_SHARE_ACCESS_READ ? "r" : "-",
2786 deny & NFS4_SHARE_ACCESS_WRITE ? "w" : "-");
2788 spin_lock(&nf->fi_lock);
2789 file = find_any_file_locked(nf);
2791 nfs4_show_superblock(s, file);
2793 nfs4_show_fname(s, file);
2796 spin_unlock(&nf->fi_lock);
2797 nfs4_show_owner(s, oo);
2798 if (st->sc_status & SC_STATUS_ADMIN_REVOKED)
2799 seq_puts(s, ", admin-revoked");
2800 seq_puts(s, " }\n");
2804 static int nfs4_show_lock(struct seq_file *s, struct nfs4_stid *st)
2806 struct nfs4_ol_stateid *ols;
2807 struct nfs4_file *nf;
2808 struct nfsd_file *file;
2809 struct nfs4_stateowner *oo;
2811 ols = openlockstateid(st);
2812 oo = ols->st_stateowner;
2816 nfs4_show_stateid(s, &st->sc_stateid);
2817 seq_puts(s, ": { type: lock, ");
2819 spin_lock(&nf->fi_lock);
2820 file = find_any_file_locked(nf);
2823 * Note: a lock stateid isn't really the same thing as a lock,
2824 * it's the locking state held by one owner on a file, and there
2825 * may be multiple (or no) lock ranges associated with it.
2826 * (Same for the matter is true of open stateids.)
2829 nfs4_show_superblock(s, file);
2830 /* XXX: open stateid? */
2832 nfs4_show_fname(s, file);
2835 nfs4_show_owner(s, oo);
2836 if (st->sc_status & SC_STATUS_ADMIN_REVOKED)
2837 seq_puts(s, ", admin-revoked");
2838 seq_puts(s, " }\n");
2839 spin_unlock(&nf->fi_lock);
2843 static int nfs4_show_deleg(struct seq_file *s, struct nfs4_stid *st)
2845 struct nfs4_delegation *ds;
2846 struct nfs4_file *nf;
2847 struct nfsd_file *file;
2849 ds = delegstateid(st);
2853 nfs4_show_stateid(s, &st->sc_stateid);
2854 seq_puts(s, ": { type: deleg, ");
2856 seq_printf(s, "access: %s",
2857 ds->dl_type == NFS4_OPEN_DELEGATE_READ ? "r" : "w");
2859 /* XXX: lease time, whether it's being recalled. */
2861 spin_lock(&nf->fi_lock);
2862 file = nf->fi_deleg_file;
2865 nfs4_show_superblock(s, file);
2867 nfs4_show_fname(s, file);
2869 spin_unlock(&nf->fi_lock);
2870 if (st->sc_status & SC_STATUS_ADMIN_REVOKED)
2871 seq_puts(s, ", admin-revoked");
2872 seq_puts(s, " }\n");
2876 static int nfs4_show_layout(struct seq_file *s, struct nfs4_stid *st)
2878 struct nfs4_layout_stateid *ls;
2879 struct nfsd_file *file;
2881 ls = container_of(st, struct nfs4_layout_stateid, ls_stid);
2884 nfs4_show_stateid(s, &st->sc_stateid);
2885 seq_puts(s, ": { type: layout");
2887 /* XXX: What else would be useful? */
2889 spin_lock(&ls->ls_stid.sc_file->fi_lock);
2893 nfs4_show_superblock(s, file);
2895 nfs4_show_fname(s, file);
2897 spin_unlock(&ls->ls_stid.sc_file->fi_lock);
2898 if (st->sc_status & SC_STATUS_ADMIN_REVOKED)
2899 seq_puts(s, ", admin-revoked");
2900 seq_puts(s, " }\n");
2905 static int states_show(struct seq_file *s, void *v)
2907 struct nfs4_stid *st = v;
2909 switch (st->sc_type) {
2911 return nfs4_show_open(s, st);
2913 return nfs4_show_lock(s, st);
2915 return nfs4_show_deleg(s, st);
2916 case SC_TYPE_LAYOUT:
2917 return nfs4_show_layout(s, st);
2919 return 0; /* XXX: or SEQ_SKIP? */
2921 /* XXX: copy stateids? */
2924 static struct seq_operations states_seq_ops = {
2925 .start = states_start,
2926 .next = states_next,
2927 .stop = states_stop,
2931 static int client_states_open(struct inode *inode, struct file *file)
2934 struct nfs4_client *clp;
2937 clp = get_nfsdfs_clp(inode);
2941 ret = seq_open(file, &states_seq_ops);
2944 s = file->private_data;
2949 static int client_opens_release(struct inode *inode, struct file *file)
2951 struct seq_file *m = file->private_data;
2952 struct nfs4_client *clp = m->private;
2954 /* XXX: alternatively, we could get/drop in seq start/stop */
2956 return seq_release(inode, file);
2959 static const struct file_operations client_states_fops = {
2960 .open = client_states_open,
2962 .llseek = seq_lseek,
2963 .release = client_opens_release,
2967 * Normally we refuse to destroy clients that are in use, but here the
2968 * administrator is telling us to just do it. We also want to wait
2969 * so the caller has a guarantee that the client's locks are gone by
2970 * the time the write returns:
2972 static void force_expire_client(struct nfs4_client *clp)
2974 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2975 bool already_expired;
2977 trace_nfsd_clid_admin_expired(&clp->cl_clientid);
2979 spin_lock(&nn->client_lock);
2981 spin_unlock(&nn->client_lock);
2983 wait_event(expiry_wq, atomic_read(&clp->cl_rpc_users) == 0);
2984 spin_lock(&nn->client_lock);
2985 already_expired = list_empty(&clp->cl_lru);
2986 if (!already_expired)
2987 unhash_client_locked(clp);
2988 spin_unlock(&nn->client_lock);
2990 if (!already_expired)
2993 wait_event(expiry_wq, clp->cl_nfsd_dentry == NULL);
2996 static ssize_t client_ctl_write(struct file *file, const char __user *buf,
2997 size_t size, loff_t *pos)
3000 struct nfs4_client *clp;
3002 data = simple_transaction_get(file, buf, size);
3004 return PTR_ERR(data);
3005 if (size != 7 || 0 != memcmp(data, "expire\n", 7))
3007 clp = get_nfsdfs_clp(file_inode(file));
3010 force_expire_client(clp);
3015 static const struct file_operations client_ctl_fops = {
3016 .write = client_ctl_write,
3017 .release = simple_transaction_release,
3020 static const struct tree_descr client_files[] = {
3021 [0] = {"info", &client_info_fops, S_IRUSR},
3022 [1] = {"states", &client_states_fops, S_IRUSR},
3023 [2] = {"ctl", &client_ctl_fops, S_IWUSR},
3028 nfsd4_cb_recall_any_done(struct nfsd4_callback *cb,
3029 struct rpc_task *task)
3031 trace_nfsd_cb_recall_any_done(cb, task);
3032 switch (task->tk_status) {
3033 case -NFS4ERR_DELAY:
3034 rpc_delay(task, 2 * HZ);
3042 nfsd4_cb_recall_any_release(struct nfsd4_callback *cb)
3044 struct nfs4_client *clp = cb->cb_clp;
3045 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
3047 spin_lock(&nn->client_lock);
3048 clear_bit(NFSD4_CLIENT_CB_RECALL_ANY, &clp->cl_flags);
3049 put_client_renew_locked(clp);
3050 spin_unlock(&nn->client_lock);
3054 nfsd4_cb_getattr_done(struct nfsd4_callback *cb, struct rpc_task *task)
3056 struct nfs4_cb_fattr *ncf =
3057 container_of(cb, struct nfs4_cb_fattr, ncf_getattr);
3059 ncf->ncf_cb_status = task->tk_status;
3060 switch (task->tk_status) {
3061 case -NFS4ERR_DELAY:
3062 rpc_delay(task, 2 * HZ);
3070 nfsd4_cb_getattr_release(struct nfsd4_callback *cb)
3072 struct nfs4_cb_fattr *ncf =
3073 container_of(cb, struct nfs4_cb_fattr, ncf_getattr);
3074 struct nfs4_delegation *dp =
3075 container_of(ncf, struct nfs4_delegation, dl_cb_fattr);
3077 nfs4_put_stid(&dp->dl_stid);
3078 clear_bit(CB_GETATTR_BUSY, &ncf->ncf_cb_flags);
3079 wake_up_bit(&ncf->ncf_cb_flags, CB_GETATTR_BUSY);
3082 static const struct nfsd4_callback_ops nfsd4_cb_recall_any_ops = {
3083 .done = nfsd4_cb_recall_any_done,
3084 .release = nfsd4_cb_recall_any_release,
3087 static const struct nfsd4_callback_ops nfsd4_cb_getattr_ops = {
3088 .done = nfsd4_cb_getattr_done,
3089 .release = nfsd4_cb_getattr_release,
3092 static void nfs4_cb_getattr(struct nfs4_cb_fattr *ncf)
3094 struct nfs4_delegation *dp =
3095 container_of(ncf, struct nfs4_delegation, dl_cb_fattr);
3097 if (test_and_set_bit(CB_GETATTR_BUSY, &ncf->ncf_cb_flags))
3099 /* set to proper status when nfsd4_cb_getattr_done runs */
3100 ncf->ncf_cb_status = NFS4ERR_IO;
3102 refcount_inc(&dp->dl_stid.sc_count);
3103 nfsd4_run_cb(&ncf->ncf_getattr);
3106 static struct nfs4_client *create_client(struct xdr_netobj name,
3107 struct svc_rqst *rqstp, nfs4_verifier *verf)
3109 struct nfs4_client *clp;
3110 struct sockaddr *sa = svc_addr(rqstp);
3112 struct net *net = SVC_NET(rqstp);
3113 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
3114 struct dentry *dentries[ARRAY_SIZE(client_files)];
3116 clp = alloc_client(name, nn);
3120 ret = copy_cred(&clp->cl_cred, &rqstp->rq_cred);
3126 kref_init(&clp->cl_nfsdfs.cl_ref);
3127 nfsd4_init_cb(&clp->cl_cb_null, clp, NULL, NFSPROC4_CLNT_CB_NULL);
3128 clp->cl_time = ktime_get_boottime_seconds();
3129 clear_bit(0, &clp->cl_cb_slot_busy);
3130 copy_verf(clp, verf);
3131 memcpy(&clp->cl_addr, sa, sizeof(struct sockaddr_storage));
3132 clp->cl_cb_session = NULL;
3134 clp->cl_nfsd_dentry = nfsd_client_mkdir(
3135 nn, &clp->cl_nfsdfs,
3136 clp->cl_clientid.cl_id - nn->clientid_base,
3137 client_files, dentries);
3138 clp->cl_nfsd_info_dentry = dentries[0];
3139 if (!clp->cl_nfsd_dentry) {
3143 clp->cl_ra = kzalloc(sizeof(*clp->cl_ra), GFP_KERNEL);
3148 clp->cl_ra_time = 0;
3149 nfsd4_init_cb(&clp->cl_ra->ra_cb, clp, &nfsd4_cb_recall_any_ops,
3150 NFSPROC4_CLNT_CB_RECALL_ANY);
3155 add_clp_to_name_tree(struct nfs4_client *new_clp, struct rb_root *root)
3157 struct rb_node **new = &(root->rb_node), *parent = NULL;
3158 struct nfs4_client *clp;
3161 clp = rb_entry(*new, struct nfs4_client, cl_namenode);
3164 if (compare_blob(&clp->cl_name, &new_clp->cl_name) > 0)
3165 new = &((*new)->rb_left);
3167 new = &((*new)->rb_right);
3170 rb_link_node(&new_clp->cl_namenode, parent, new);
3171 rb_insert_color(&new_clp->cl_namenode, root);
3174 static struct nfs4_client *
3175 find_clp_in_name_tree(struct xdr_netobj *name, struct rb_root *root)
3178 struct rb_node *node = root->rb_node;
3179 struct nfs4_client *clp;
3182 clp = rb_entry(node, struct nfs4_client, cl_namenode);
3183 cmp = compare_blob(&clp->cl_name, name);
3185 node = node->rb_left;
3187 node = node->rb_right;
3195 add_to_unconfirmed(struct nfs4_client *clp)
3197 unsigned int idhashval;
3198 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
3200 lockdep_assert_held(&nn->client_lock);
3202 clear_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags);
3203 add_clp_to_name_tree(clp, &nn->unconf_name_tree);
3204 idhashval = clientid_hashval(clp->cl_clientid.cl_id);
3205 list_add(&clp->cl_idhash, &nn->unconf_id_hashtbl[idhashval]);
3206 renew_client_locked(clp);
3210 move_to_confirmed(struct nfs4_client *clp)
3212 unsigned int idhashval = clientid_hashval(clp->cl_clientid.cl_id);
3213 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
3215 lockdep_assert_held(&nn->client_lock);
3217 list_move(&clp->cl_idhash, &nn->conf_id_hashtbl[idhashval]);
3218 rb_erase(&clp->cl_namenode, &nn->unconf_name_tree);
3219 add_clp_to_name_tree(clp, &nn->conf_name_tree);
3220 set_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags);
3221 trace_nfsd_clid_confirmed(&clp->cl_clientid);
3222 renew_client_locked(clp);
3225 static struct nfs4_client *
3226 find_client_in_id_table(struct list_head *tbl, clientid_t *clid, bool sessions)
3228 struct nfs4_client *clp;
3229 unsigned int idhashval = clientid_hashval(clid->cl_id);
3231 list_for_each_entry(clp, &tbl[idhashval], cl_idhash) {
3232 if (same_clid(&clp->cl_clientid, clid)) {
3233 if ((bool)clp->cl_minorversion != sessions)
3235 renew_client_locked(clp);
3242 static struct nfs4_client *
3243 find_confirmed_client(clientid_t *clid, bool sessions, struct nfsd_net *nn)
3245 struct list_head *tbl = nn->conf_id_hashtbl;
3247 lockdep_assert_held(&nn->client_lock);
3248 return find_client_in_id_table(tbl, clid, sessions);
3251 static struct nfs4_client *
3252 find_unconfirmed_client(clientid_t *clid, bool sessions, struct nfsd_net *nn)
3254 struct list_head *tbl = nn->unconf_id_hashtbl;
3256 lockdep_assert_held(&nn->client_lock);
3257 return find_client_in_id_table(tbl, clid, sessions);
3260 static bool clp_used_exchangeid(struct nfs4_client *clp)
3262 return clp->cl_exchange_flags != 0;
3265 static struct nfs4_client *
3266 find_confirmed_client_by_name(struct xdr_netobj *name, struct nfsd_net *nn)
3268 lockdep_assert_held(&nn->client_lock);
3269 return find_clp_in_name_tree(name, &nn->conf_name_tree);
3272 static struct nfs4_client *
3273 find_unconfirmed_client_by_name(struct xdr_netobj *name, struct nfsd_net *nn)
3275 lockdep_assert_held(&nn->client_lock);
3276 return find_clp_in_name_tree(name, &nn->unconf_name_tree);
3280 gen_callback(struct nfs4_client *clp, struct nfsd4_setclientid *se, struct svc_rqst *rqstp)
3282 struct nfs4_cb_conn *conn = &clp->cl_cb_conn;
3283 struct sockaddr *sa = svc_addr(rqstp);
3284 u32 scopeid = rpc_get_scope_id(sa);
3285 unsigned short expected_family;
3287 /* Currently, we only support tcp and tcp6 for the callback channel */
3288 if (se->se_callback_netid_len == 3 &&
3289 !memcmp(se->se_callback_netid_val, "tcp", 3))
3290 expected_family = AF_INET;
3291 else if (se->se_callback_netid_len == 4 &&
3292 !memcmp(se->se_callback_netid_val, "tcp6", 4))
3293 expected_family = AF_INET6;
3297 conn->cb_addrlen = rpc_uaddr2sockaddr(clp->net, se->se_callback_addr_val,
3298 se->se_callback_addr_len,
3299 (struct sockaddr *)&conn->cb_addr,
3300 sizeof(conn->cb_addr));
3302 if (!conn->cb_addrlen || conn->cb_addr.ss_family != expected_family)
3305 if (conn->cb_addr.ss_family == AF_INET6)
3306 ((struct sockaddr_in6 *)&conn->cb_addr)->sin6_scope_id = scopeid;
3308 conn->cb_prog = se->se_callback_prog;
3309 conn->cb_ident = se->se_callback_ident;
3310 memcpy(&conn->cb_saddr, &rqstp->rq_daddr, rqstp->rq_daddrlen);
3311 trace_nfsd_cb_args(clp, conn);
3314 conn->cb_addr.ss_family = AF_UNSPEC;
3315 conn->cb_addrlen = 0;
3316 trace_nfsd_cb_nodelegs(clp);
3321 * Cache a reply. nfsd4_check_resp_size() has bounded the cache size.
3324 nfsd4_store_cache_entry(struct nfsd4_compoundres *resp)
3326 struct xdr_buf *buf = resp->xdr->buf;
3327 struct nfsd4_slot *slot = resp->cstate.slot;
3330 dprintk("--> %s slot %p\n", __func__, slot);
3332 slot->sl_flags |= NFSD4_SLOT_INITIALIZED;
3333 slot->sl_opcnt = resp->opcnt;
3334 slot->sl_status = resp->cstate.status;
3335 free_svc_cred(&slot->sl_cred);
3336 copy_cred(&slot->sl_cred, &resp->rqstp->rq_cred);
3338 if (!nfsd4_cache_this(resp)) {
3339 slot->sl_flags &= ~NFSD4_SLOT_CACHED;
3342 slot->sl_flags |= NFSD4_SLOT_CACHED;
3344 base = resp->cstate.data_offset;
3345 slot->sl_datalen = buf->len - base;
3346 if (read_bytes_from_xdr_buf(buf, base, slot->sl_data, slot->sl_datalen))
3347 WARN(1, "%s: sessions DRC could not cache compound\n",
3353 * Encode the replay sequence operation from the slot values.
3354 * If cachethis is FALSE encode the uncached rep error on the next
3355 * operation which sets resp->p and increments resp->opcnt for
3356 * nfs4svc_encode_compoundres.
3360 nfsd4_enc_sequence_replay(struct nfsd4_compoundargs *args,
3361 struct nfsd4_compoundres *resp)
3363 struct nfsd4_op *op;
3364 struct nfsd4_slot *slot = resp->cstate.slot;
3366 /* Encode the replayed sequence operation */
3367 op = &args->ops[resp->opcnt - 1];
3368 nfsd4_encode_operation(resp, op);
3370 if (slot->sl_flags & NFSD4_SLOT_CACHED)
3372 if (args->opcnt == 1) {
3374 * The original operation wasn't a solo sequence--we
3375 * always cache those--so this retry must not match the
3378 op->status = nfserr_seq_false_retry;
3380 op = &args->ops[resp->opcnt++];
3381 op->status = nfserr_retry_uncached_rep;
3382 nfsd4_encode_operation(resp, op);
3388 * The sequence operation is not cached because we can use the slot and
3392 nfsd4_replay_cache_entry(struct nfsd4_compoundres *resp,
3393 struct nfsd4_sequence *seq)
3395 struct nfsd4_slot *slot = resp->cstate.slot;
3396 struct xdr_stream *xdr = resp->xdr;
3400 dprintk("--> %s slot %p\n", __func__, slot);
3402 status = nfsd4_enc_sequence_replay(resp->rqstp->rq_argp, resp);
3406 p = xdr_reserve_space(xdr, slot->sl_datalen);
3409 return nfserr_serverfault;
3411 xdr_encode_opaque_fixed(p, slot->sl_data, slot->sl_datalen);
3412 xdr_commit_encode(xdr);
3414 resp->opcnt = slot->sl_opcnt;
3415 return slot->sl_status;
3419 * Set the exchange_id flags returned by the server.
3422 nfsd4_set_ex_flags(struct nfs4_client *new, struct nfsd4_exchange_id *clid)
3424 #ifdef CONFIG_NFSD_PNFS
3425 new->cl_exchange_flags |= EXCHGID4_FLAG_USE_PNFS_MDS;
3427 new->cl_exchange_flags |= EXCHGID4_FLAG_USE_NON_PNFS;
3430 /* Referrals are supported, Migration is not. */
3431 new->cl_exchange_flags |= EXCHGID4_FLAG_SUPP_MOVED_REFER;
3433 /* set the wire flags to return to client. */
3434 clid->flags = new->cl_exchange_flags;
3437 static bool client_has_openowners(struct nfs4_client *clp)
3439 struct nfs4_openowner *oo;
3441 list_for_each_entry(oo, &clp->cl_openowners, oo_perclient) {
3442 if (!list_empty(&oo->oo_owner.so_stateids))
3448 static bool client_has_state(struct nfs4_client *clp)
3450 return client_has_openowners(clp)
3451 #ifdef CONFIG_NFSD_PNFS
3452 || !list_empty(&clp->cl_lo_states)
3454 || !list_empty(&clp->cl_delegations)
3455 || !list_empty(&clp->cl_sessions)
3456 || !list_empty(&clp->async_copies);
3459 static __be32 copy_impl_id(struct nfs4_client *clp,
3460 struct nfsd4_exchange_id *exid)
3462 if (!exid->nii_domain.data)
3464 xdr_netobj_dup(&clp->cl_nii_domain, &exid->nii_domain, GFP_KERNEL);
3465 if (!clp->cl_nii_domain.data)
3466 return nfserr_jukebox;
3467 xdr_netobj_dup(&clp->cl_nii_name, &exid->nii_name, GFP_KERNEL);
3468 if (!clp->cl_nii_name.data)
3469 return nfserr_jukebox;
3470 clp->cl_nii_time = exid->nii_time;
3475 nfsd4_exchange_id(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3476 union nfsd4_op_u *u)
3478 struct nfsd4_exchange_id *exid = &u->exchange_id;
3479 struct nfs4_client *conf, *new;
3480 struct nfs4_client *unconf = NULL;
3482 char addr_str[INET6_ADDRSTRLEN];
3483 nfs4_verifier verf = exid->verifier;
3484 struct sockaddr *sa = svc_addr(rqstp);
3485 bool update = exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A;
3486 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3488 rpc_ntop(sa, addr_str, sizeof(addr_str));
3489 dprintk("%s rqstp=%p exid=%p clname.len=%u clname.data=%p "
3490 "ip_addr=%s flags %x, spa_how %u\n",
3491 __func__, rqstp, exid, exid->clname.len, exid->clname.data,
3492 addr_str, exid->flags, exid->spa_how);
3494 if (exid->flags & ~EXCHGID4_FLAG_MASK_A)
3495 return nfserr_inval;
3497 new = create_client(exid->clname, rqstp, &verf);
3499 return nfserr_jukebox;
3500 status = copy_impl_id(new, exid);
3504 switch (exid->spa_how) {
3506 exid->spo_must_enforce[0] = 0;
3507 exid->spo_must_enforce[1] = (
3508 1 << (OP_BIND_CONN_TO_SESSION - 32) |
3509 1 << (OP_EXCHANGE_ID - 32) |
3510 1 << (OP_CREATE_SESSION - 32) |
3511 1 << (OP_DESTROY_SESSION - 32) |
3512 1 << (OP_DESTROY_CLIENTID - 32));
3514 exid->spo_must_allow[0] &= (1 << (OP_CLOSE) |
3515 1 << (OP_OPEN_DOWNGRADE) |
3517 1 << (OP_DELEGRETURN));
3519 exid->spo_must_allow[1] &= (
3520 1 << (OP_TEST_STATEID - 32) |
3521 1 << (OP_FREE_STATEID - 32));
3522 if (!svc_rqst_integrity_protected(rqstp)) {
3523 status = nfserr_inval;
3527 * Sometimes userspace doesn't give us a principal.
3528 * Which is a bug, really. Anyway, we can't enforce
3529 * MACH_CRED in that case, better to give up now:
3531 if (!new->cl_cred.cr_principal &&
3532 !new->cl_cred.cr_raw_principal) {
3533 status = nfserr_serverfault;
3536 new->cl_mach_cred = true;
3540 default: /* checked by xdr code */
3544 status = nfserr_encr_alg_unsupp;
3548 /* Cases below refer to rfc 5661 section 18.35.4: */
3549 spin_lock(&nn->client_lock);
3550 conf = find_confirmed_client_by_name(&exid->clname, nn);
3552 bool creds_match = same_creds(&conf->cl_cred, &rqstp->rq_cred);
3553 bool verfs_match = same_verf(&verf, &conf->cl_verifier);
3556 if (!clp_used_exchangeid(conf)) { /* buggy client */
3557 status = nfserr_inval;
3560 if (!nfsd4_mach_creds_match(conf, rqstp)) {
3561 status = nfserr_wrong_cred;
3564 if (!creds_match) { /* case 9 */
3565 status = nfserr_perm;
3568 if (!verfs_match) { /* case 8 */
3569 status = nfserr_not_same;
3573 exid->flags |= EXCHGID4_FLAG_CONFIRMED_R;
3574 trace_nfsd_clid_confirmed_r(conf);
3577 if (!creds_match) { /* case 3 */
3578 if (client_has_state(conf)) {
3579 status = nfserr_clid_inuse;
3580 trace_nfsd_clid_cred_mismatch(conf, rqstp);
3585 if (verfs_match) { /* case 2 */
3586 conf->cl_exchange_flags |= EXCHGID4_FLAG_CONFIRMED_R;
3587 trace_nfsd_clid_confirmed_r(conf);
3590 /* case 5, client reboot */
3591 trace_nfsd_clid_verf_mismatch(conf, rqstp, &verf);
3596 if (update) { /* case 7 */
3597 status = nfserr_noent;
3601 unconf = find_unconfirmed_client_by_name(&exid->clname, nn);
3602 if (unconf) /* case 4, possible retry or client restart */
3603 unhash_client_locked(unconf);
3605 /* case 1, new owner ID */
3606 trace_nfsd_clid_fresh(new);
3610 status = mark_client_expired_locked(conf);
3613 trace_nfsd_clid_replaced(&conf->cl_clientid);
3615 new->cl_minorversion = cstate->minorversion;
3616 new->cl_spo_must_allow.u.words[0] = exid->spo_must_allow[0];
3617 new->cl_spo_must_allow.u.words[1] = exid->spo_must_allow[1];
3619 /* Contrived initial CREATE_SESSION response */
3620 new->cl_cs_slot.sl_status = nfserr_seq_misordered;
3622 add_to_unconfirmed(new);
3625 exid->clientid.cl_boot = conf->cl_clientid.cl_boot;
3626 exid->clientid.cl_id = conf->cl_clientid.cl_id;
3628 exid->seqid = conf->cl_cs_slot.sl_seqid + 1;
3629 nfsd4_set_ex_flags(conf, exid);
3631 dprintk("nfsd4_exchange_id seqid %d flags %x\n",
3632 conf->cl_cs_slot.sl_seqid, conf->cl_exchange_flags);
3636 spin_unlock(&nn->client_lock);
3641 trace_nfsd_clid_expire_unconf(&unconf->cl_clientid);
3642 expire_client(unconf);
3648 check_slot_seqid(u32 seqid, u32 slot_seqid, int slot_inuse)
3650 dprintk("%s enter. seqid %d slot_seqid %d\n", __func__, seqid,
3653 /* The slot is in use, and no response has been sent. */
3655 if (seqid == slot_seqid)
3656 return nfserr_jukebox;
3658 return nfserr_seq_misordered;
3660 /* Note unsigned 32-bit arithmetic handles wraparound: */
3661 if (likely(seqid == slot_seqid + 1))
3663 if (seqid == slot_seqid)
3664 return nfserr_replay_cache;
3665 return nfserr_seq_misordered;
3669 * Cache the create session result into the create session single DRC
3670 * slot cache by saving the xdr structure. sl_seqid has been set.
3671 * Do this for solo or embedded create session operations.
3674 nfsd4_cache_create_session(struct nfsd4_create_session *cr_ses,
3675 struct nfsd4_clid_slot *slot, __be32 nfserr)
3677 slot->sl_status = nfserr;
3678 memcpy(&slot->sl_cr_ses, cr_ses, sizeof(*cr_ses));
3682 nfsd4_replay_create_session(struct nfsd4_create_session *cr_ses,
3683 struct nfsd4_clid_slot *slot)
3685 memcpy(cr_ses, &slot->sl_cr_ses, sizeof(*cr_ses));
3686 return slot->sl_status;
3689 #define NFSD_MIN_REQ_HDR_SEQ_SZ ((\
3690 2 * 2 + /* credential,verifier: AUTH_NULL, length 0 */ \
3691 1 + /* MIN tag is length with zero, only length */ \
3692 3 + /* version, opcount, opcode */ \
3693 XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
3694 /* seqid, slotID, slotID, cache */ \
3695 4 ) * sizeof(__be32))
3697 #define NFSD_MIN_RESP_HDR_SEQ_SZ ((\
3698 2 + /* verifier: AUTH_NULL, length 0 */\
3700 1 + /* MIN tag is length with zero, only length */ \
3701 3 + /* opcount, opcode, opstatus*/ \
3702 XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
3703 /* seqid, slotID, slotID, slotID, status */ \
3704 5 ) * sizeof(__be32))
3706 static __be32 check_forechannel_attrs(struct nfsd4_channel_attrs *ca, struct nfsd_net *nn)
3708 u32 maxrpc = nn->nfsd_serv->sv_max_mesg;
3710 if (ca->maxreq_sz < NFSD_MIN_REQ_HDR_SEQ_SZ)
3711 return nfserr_toosmall;
3712 if (ca->maxresp_sz < NFSD_MIN_RESP_HDR_SEQ_SZ)
3713 return nfserr_toosmall;
3714 ca->headerpadsz = 0;
3715 ca->maxreq_sz = min_t(u32, ca->maxreq_sz, maxrpc);
3716 ca->maxresp_sz = min_t(u32, ca->maxresp_sz, maxrpc);
3717 ca->maxops = min_t(u32, ca->maxops, NFSD_MAX_OPS_PER_COMPOUND);
3718 ca->maxresp_cached = min_t(u32, ca->maxresp_cached,
3719 NFSD_SLOT_CACHE_SIZE + NFSD_MIN_HDR_SEQ_SZ);
3720 ca->maxreqs = min_t(u32, ca->maxreqs, NFSD_MAX_SLOTS_PER_SESSION);
3722 * Note decreasing slot size below client's request may make it
3723 * difficult for client to function correctly, whereas
3724 * decreasing the number of slots will (just?) affect
3725 * performance. When short on memory we therefore prefer to
3726 * decrease number of slots instead of their size. Clients that
3727 * request larger slots than they need will get poor results:
3728 * Note that we always allow at least one slot, because our
3729 * accounting is soft and provides no guarantees either way.
3731 ca->maxreqs = nfsd4_get_drc_mem(ca, nn);
3737 * Server's NFSv4.1 backchannel support is AUTH_SYS-only for now.
3738 * These are based on similar macros in linux/sunrpc/msg_prot.h .
3740 #define RPC_MAX_HEADER_WITH_AUTH_SYS \
3741 (RPC_CALLHDRSIZE + 2 * (2 + UNX_CALLSLACK))
3743 #define RPC_MAX_REPHEADER_WITH_AUTH_SYS \
3744 (RPC_REPHDRSIZE + (2 + NUL_REPLYSLACK))
3746 #define NFSD_CB_MAX_REQ_SZ ((NFS4_enc_cb_recall_sz + \
3747 RPC_MAX_HEADER_WITH_AUTH_SYS) * sizeof(__be32))
3748 #define NFSD_CB_MAX_RESP_SZ ((NFS4_dec_cb_recall_sz + \
3749 RPC_MAX_REPHEADER_WITH_AUTH_SYS) * \
3752 static __be32 check_backchannel_attrs(struct nfsd4_channel_attrs *ca)
3754 ca->headerpadsz = 0;
3756 if (ca->maxreq_sz < NFSD_CB_MAX_REQ_SZ)
3757 return nfserr_toosmall;
3758 if (ca->maxresp_sz < NFSD_CB_MAX_RESP_SZ)
3759 return nfserr_toosmall;
3760 ca->maxresp_cached = 0;
3762 return nfserr_toosmall;
3767 static __be32 nfsd4_check_cb_sec(struct nfsd4_cb_sec *cbs)
3769 switch (cbs->flavor) {
3775 * GSS case: the spec doesn't allow us to return this
3776 * error. But it also doesn't allow us not to support
3778 * I'd rather this fail hard than return some error the
3779 * client might think it can already handle:
3781 return nfserr_encr_alg_unsupp;
3786 nfsd4_create_session(struct svc_rqst *rqstp,
3787 struct nfsd4_compound_state *cstate, union nfsd4_op_u *u)
3789 struct nfsd4_create_session *cr_ses = &u->create_session;
3790 struct sockaddr *sa = svc_addr(rqstp);
3791 struct nfs4_client *conf, *unconf;
3792 struct nfsd4_clid_slot *cs_slot;
3793 struct nfs4_client *old = NULL;
3794 struct nfsd4_session *new;
3795 struct nfsd4_conn *conn;
3797 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3799 if (cr_ses->flags & ~SESSION4_FLAG_MASK_A)
3800 return nfserr_inval;
3801 status = nfsd4_check_cb_sec(&cr_ses->cb_sec);
3804 status = check_forechannel_attrs(&cr_ses->fore_channel, nn);
3807 status = check_backchannel_attrs(&cr_ses->back_channel);
3809 goto out_release_drc_mem;
3810 status = nfserr_jukebox;
3811 new = alloc_session(&cr_ses->fore_channel, &cr_ses->back_channel);
3813 goto out_release_drc_mem;
3814 conn = alloc_conn_from_crses(rqstp, cr_ses);
3816 goto out_free_session;
3818 spin_lock(&nn->client_lock);
3820 /* RFC 8881 Section 18.36.4 Phase 1: Client record look-up. */
3821 unconf = find_unconfirmed_client(&cr_ses->clientid, true, nn);
3822 conf = find_confirmed_client(&cr_ses->clientid, true, nn);
3823 if (!conf && !unconf) {
3824 status = nfserr_stale_clientid;
3828 /* RFC 8881 Section 18.36.4 Phase 2: Sequence ID processing. */
3830 cs_slot = &conf->cl_cs_slot;
3832 cs_slot = &unconf->cl_cs_slot;
3833 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
3835 if (status == nfserr_replay_cache) {
3836 status = nfsd4_replay_create_session(cr_ses, cs_slot);
3839 goto out_cache_error;
3841 cs_slot->sl_seqid++;
3842 cr_ses->seqid = cs_slot->sl_seqid;
3844 /* RFC 8881 Section 18.36.4 Phase 3: Client ID confirmation. */
3846 status = nfserr_wrong_cred;
3847 if (!nfsd4_mach_creds_match(conf, rqstp))
3848 goto out_cache_error;
3850 status = nfserr_clid_inuse;
3851 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred) ||
3852 !rpc_cmp_addr(sa, (struct sockaddr *) &unconf->cl_addr)) {
3853 trace_nfsd_clid_cred_mismatch(unconf, rqstp);
3854 goto out_cache_error;
3856 status = nfserr_wrong_cred;
3857 if (!nfsd4_mach_creds_match(unconf, rqstp))
3858 goto out_cache_error;
3859 old = find_confirmed_client_by_name(&unconf->cl_name, nn);
3861 status = mark_client_expired_locked(old);
3864 goto out_cache_error;
3866 trace_nfsd_clid_replaced(&old->cl_clientid);
3868 move_to_confirmed(unconf);
3872 /* RFC 8881 Section 18.36.4 Phase 4: Session creation. */
3874 /* Persistent sessions are not supported */
3875 cr_ses->flags &= ~SESSION4_PERSIST;
3876 /* Upshifting from TCP to RDMA is not supported */
3877 cr_ses->flags &= ~SESSION4_RDMA;
3879 init_session(rqstp, new, conf, cr_ses);
3880 nfsd4_get_session_locked(new);
3882 memcpy(cr_ses->sessionid.data, new->se_sessionid.data,
3883 NFS4_MAX_SESSIONID_LEN);
3885 /* cache solo and embedded create sessions under the client_lock */
3886 nfsd4_cache_create_session(cr_ses, cs_slot, status);
3887 spin_unlock(&nn->client_lock);
3889 fsnotify_dentry(conf->cl_nfsd_info_dentry, FS_MODIFY);
3890 /* init connection and backchannel */
3891 nfsd4_init_conn(rqstp, conn, new);
3892 nfsd4_put_session(new);
3898 nfsd4_cache_create_session(cr_ses, cs_slot, status);
3900 spin_unlock(&nn->client_lock);
3905 __free_session(new);
3906 out_release_drc_mem:
3907 nfsd4_put_drc_mem(&cr_ses->fore_channel);
3911 static __be32 nfsd4_map_bcts_dir(u32 *dir)
3914 case NFS4_CDFC4_FORE:
3915 case NFS4_CDFC4_BACK:
3917 case NFS4_CDFC4_FORE_OR_BOTH:
3918 case NFS4_CDFC4_BACK_OR_BOTH:
3919 *dir = NFS4_CDFC4_BOTH;
3922 return nfserr_inval;
3925 __be32 nfsd4_backchannel_ctl(struct svc_rqst *rqstp,
3926 struct nfsd4_compound_state *cstate,
3927 union nfsd4_op_u *u)
3929 struct nfsd4_backchannel_ctl *bc = &u->backchannel_ctl;
3930 struct nfsd4_session *session = cstate->session;
3931 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3934 status = nfsd4_check_cb_sec(&bc->bc_cb_sec);
3937 spin_lock(&nn->client_lock);
3938 session->se_cb_prog = bc->bc_cb_program;
3939 session->se_cb_sec = bc->bc_cb_sec;
3940 spin_unlock(&nn->client_lock);
3942 nfsd4_probe_callback(session->se_client);
3947 static struct nfsd4_conn *__nfsd4_find_conn(struct svc_xprt *xpt, struct nfsd4_session *s)
3949 struct nfsd4_conn *c;
3951 list_for_each_entry(c, &s->se_conns, cn_persession) {
3952 if (c->cn_xprt == xpt) {
3959 static __be32 nfsd4_match_existing_connection(struct svc_rqst *rqst,
3960 struct nfsd4_session *session, u32 req, struct nfsd4_conn **conn)
3962 struct nfs4_client *clp = session->se_client;
3963 struct svc_xprt *xpt = rqst->rq_xprt;
3964 struct nfsd4_conn *c;
3967 /* Following the last paragraph of RFC 5661 Section 18.34.3: */
3968 spin_lock(&clp->cl_lock);
3969 c = __nfsd4_find_conn(xpt, session);
3971 status = nfserr_noent;
3972 else if (req == c->cn_flags)
3974 else if (req == NFS4_CDFC4_FORE_OR_BOTH &&
3975 c->cn_flags != NFS4_CDFC4_BACK)
3977 else if (req == NFS4_CDFC4_BACK_OR_BOTH &&
3978 c->cn_flags != NFS4_CDFC4_FORE)
3981 status = nfserr_inval;
3982 spin_unlock(&clp->cl_lock);
3983 if (status == nfs_ok && conn)
3988 __be32 nfsd4_bind_conn_to_session(struct svc_rqst *rqstp,
3989 struct nfsd4_compound_state *cstate,
3990 union nfsd4_op_u *u)
3992 struct nfsd4_bind_conn_to_session *bcts = &u->bind_conn_to_session;
3994 struct nfsd4_conn *conn;
3995 struct nfsd4_session *session;
3996 struct net *net = SVC_NET(rqstp);
3997 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
3999 if (!nfsd4_last_compound_op(rqstp))
4000 return nfserr_not_only_op;
4001 spin_lock(&nn->client_lock);
4002 session = find_in_sessionid_hashtbl(&bcts->sessionid, net, &status);
4003 spin_unlock(&nn->client_lock);
4005 goto out_no_session;
4006 status = nfserr_wrong_cred;
4007 if (!nfsd4_mach_creds_match(session->se_client, rqstp))
4009 status = nfsd4_match_existing_connection(rqstp, session,
4011 if (status == nfs_ok) {
4012 if (bcts->dir == NFS4_CDFC4_FORE_OR_BOTH ||
4013 bcts->dir == NFS4_CDFC4_BACK)
4014 conn->cn_flags |= NFS4_CDFC4_BACK;
4015 nfsd4_probe_callback(session->se_client);
4018 if (status == nfserr_inval)
4020 status = nfsd4_map_bcts_dir(&bcts->dir);
4023 conn = alloc_conn(rqstp, bcts->dir);
4024 status = nfserr_jukebox;
4027 nfsd4_init_conn(rqstp, conn, session);
4030 nfsd4_put_session(session);
4035 static bool nfsd4_compound_in_session(struct nfsd4_compound_state *cstate, struct nfs4_sessionid *sid)
4037 if (!cstate->session)
4039 return !memcmp(sid, &cstate->session->se_sessionid, sizeof(*sid));
4043 nfsd4_destroy_session(struct svc_rqst *r, struct nfsd4_compound_state *cstate,
4044 union nfsd4_op_u *u)
4046 struct nfs4_sessionid *sessionid = &u->destroy_session.sessionid;
4047 struct nfsd4_session *ses;
4049 int ref_held_by_me = 0;
4050 struct net *net = SVC_NET(r);
4051 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
4053 status = nfserr_not_only_op;
4054 if (nfsd4_compound_in_session(cstate, sessionid)) {
4055 if (!nfsd4_last_compound_op(r))
4059 dump_sessionid(__func__, sessionid);
4060 spin_lock(&nn->client_lock);
4061 ses = find_in_sessionid_hashtbl(sessionid, net, &status);
4063 goto out_client_lock;
4064 status = nfserr_wrong_cred;
4065 if (!nfsd4_mach_creds_match(ses->se_client, r))
4066 goto out_put_session;
4067 status = mark_session_dead_locked(ses, 1 + ref_held_by_me);
4069 goto out_put_session;
4070 unhash_session(ses);
4071 spin_unlock(&nn->client_lock);
4073 nfsd4_probe_callback_sync(ses->se_client);
4075 spin_lock(&nn->client_lock);
4078 nfsd4_put_session_locked(ses);
4080 spin_unlock(&nn->client_lock);
4085 static __be32 nfsd4_sequence_check_conn(struct nfsd4_conn *new, struct nfsd4_session *ses)
4087 struct nfs4_client *clp = ses->se_client;
4088 struct nfsd4_conn *c;
4089 __be32 status = nfs_ok;
4092 spin_lock(&clp->cl_lock);
4093 c = __nfsd4_find_conn(new->cn_xprt, ses);
4096 status = nfserr_conn_not_bound_to_session;
4097 if (clp->cl_mach_cred)
4099 __nfsd4_hash_conn(new, ses);
4100 spin_unlock(&clp->cl_lock);
4101 ret = nfsd4_register_conn(new);
4103 /* oops; xprt is already down: */
4104 nfsd4_conn_lost(&new->cn_xpt_user);
4107 spin_unlock(&clp->cl_lock);
4112 static bool nfsd4_session_too_many_ops(struct svc_rqst *rqstp, struct nfsd4_session *session)
4114 struct nfsd4_compoundargs *args = rqstp->rq_argp;
4116 return args->opcnt > session->se_fchannel.maxops;
4119 static bool nfsd4_request_too_big(struct svc_rqst *rqstp,
4120 struct nfsd4_session *session)
4122 struct xdr_buf *xb = &rqstp->rq_arg;
4124 return xb->len > session->se_fchannel.maxreq_sz;
4127 static bool replay_matches_cache(struct svc_rqst *rqstp,
4128 struct nfsd4_sequence *seq, struct nfsd4_slot *slot)
4130 struct nfsd4_compoundargs *argp = rqstp->rq_argp;
4132 if ((bool)(slot->sl_flags & NFSD4_SLOT_CACHETHIS) !=
4133 (bool)seq->cachethis)
4136 * If there's an error then the reply can have fewer ops than
4139 if (slot->sl_opcnt < argp->opcnt && !slot->sl_status)
4142 * But if we cached a reply with *more* ops than the call you're
4143 * sending us now, then this new call is clearly not really a
4144 * replay of the old one:
4146 if (slot->sl_opcnt > argp->opcnt)
4148 /* This is the only check explicitly called by spec: */
4149 if (!same_creds(&rqstp->rq_cred, &slot->sl_cred))
4152 * There may be more comparisons we could actually do, but the
4153 * spec doesn't require us to catch every case where the calls
4154 * don't match (that would require caching the call as well as
4155 * the reply), so we don't bother.
4161 nfsd4_sequence(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4162 union nfsd4_op_u *u)
4164 struct nfsd4_sequence *seq = &u->sequence;
4165 struct nfsd4_compoundres *resp = rqstp->rq_resp;
4166 struct xdr_stream *xdr = resp->xdr;
4167 struct nfsd4_session *session;
4168 struct nfs4_client *clp;
4169 struct nfsd4_slot *slot;
4170 struct nfsd4_conn *conn;
4173 struct net *net = SVC_NET(rqstp);
4174 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
4176 if (resp->opcnt != 1)
4177 return nfserr_sequence_pos;
4180 * Will be either used or freed by nfsd4_sequence_check_conn
4183 conn = alloc_conn(rqstp, NFS4_CDFC4_FORE);
4185 return nfserr_jukebox;
4187 spin_lock(&nn->client_lock);
4188 session = find_in_sessionid_hashtbl(&seq->sessionid, net, &status);
4190 goto out_no_session;
4191 clp = session->se_client;
4193 status = nfserr_too_many_ops;
4194 if (nfsd4_session_too_many_ops(rqstp, session))
4195 goto out_put_session;
4197 status = nfserr_req_too_big;
4198 if (nfsd4_request_too_big(rqstp, session))
4199 goto out_put_session;
4201 status = nfserr_badslot;
4202 if (seq->slotid >= session->se_fchannel.maxreqs)
4203 goto out_put_session;
4205 slot = session->se_slots[seq->slotid];
4206 dprintk("%s: slotid %d\n", __func__, seq->slotid);
4208 /* We do not negotiate the number of slots yet, so set the
4209 * maxslots to the session maxreqs which is used to encode
4210 * sr_highest_slotid and the sr_target_slot id to maxslots */
4211 seq->maxslots = session->se_fchannel.maxreqs;
4213 status = check_slot_seqid(seq->seqid, slot->sl_seqid,
4214 slot->sl_flags & NFSD4_SLOT_INUSE);
4215 if (status == nfserr_replay_cache) {
4216 status = nfserr_seq_misordered;
4217 if (!(slot->sl_flags & NFSD4_SLOT_INITIALIZED))
4218 goto out_put_session;
4219 status = nfserr_seq_false_retry;
4220 if (!replay_matches_cache(rqstp, seq, slot))
4221 goto out_put_session;
4222 cstate->slot = slot;
4223 cstate->session = session;
4225 /* Return the cached reply status and set cstate->status
4226 * for nfsd4_proc_compound processing */
4227 status = nfsd4_replay_cache_entry(resp, seq);
4228 cstate->status = nfserr_replay_cache;
4232 goto out_put_session;
4234 status = nfsd4_sequence_check_conn(conn, session);
4237 goto out_put_session;
4239 buflen = (seq->cachethis) ?
4240 session->se_fchannel.maxresp_cached :
4241 session->se_fchannel.maxresp_sz;
4242 status = (seq->cachethis) ? nfserr_rep_too_big_to_cache :
4244 if (xdr_restrict_buflen(xdr, buflen - rqstp->rq_auth_slack))
4245 goto out_put_session;
4246 svc_reserve(rqstp, buflen);
4249 /* Success! bump slot seqid */
4250 slot->sl_seqid = seq->seqid;
4251 slot->sl_flags |= NFSD4_SLOT_INUSE;
4253 slot->sl_flags |= NFSD4_SLOT_CACHETHIS;
4255 slot->sl_flags &= ~NFSD4_SLOT_CACHETHIS;
4257 cstate->slot = slot;
4258 cstate->session = session;
4262 switch (clp->cl_cb_state) {
4264 seq->status_flags = SEQ4_STATUS_CB_PATH_DOWN;
4266 case NFSD4_CB_FAULT:
4267 seq->status_flags = SEQ4_STATUS_BACKCHANNEL_FAULT;
4270 seq->status_flags = 0;
4272 if (!list_empty(&clp->cl_revoked))
4273 seq->status_flags |= SEQ4_STATUS_RECALLABLE_STATE_REVOKED;
4274 if (atomic_read(&clp->cl_admin_revoked))
4275 seq->status_flags |= SEQ4_STATUS_ADMIN_STATE_REVOKED;
4276 trace_nfsd_seq4_status(rqstp, seq);
4280 spin_unlock(&nn->client_lock);
4283 nfsd4_put_session_locked(session);
4284 goto out_no_session;
4288 nfsd4_sequence_done(struct nfsd4_compoundres *resp)
4290 struct nfsd4_compound_state *cs = &resp->cstate;
4292 if (nfsd4_has_session(cs)) {
4293 if (cs->status != nfserr_replay_cache) {
4294 nfsd4_store_cache_entry(resp);
4295 cs->slot->sl_flags &= ~NFSD4_SLOT_INUSE;
4297 /* Drop session reference that was taken in nfsd4_sequence() */
4298 nfsd4_put_session(cs->session);
4300 put_client_renew(cs->clp);
4304 nfsd4_destroy_clientid(struct svc_rqst *rqstp,
4305 struct nfsd4_compound_state *cstate,
4306 union nfsd4_op_u *u)
4308 struct nfsd4_destroy_clientid *dc = &u->destroy_clientid;
4309 struct nfs4_client *conf, *unconf;
4310 struct nfs4_client *clp = NULL;
4312 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
4314 spin_lock(&nn->client_lock);
4315 unconf = find_unconfirmed_client(&dc->clientid, true, nn);
4316 conf = find_confirmed_client(&dc->clientid, true, nn);
4317 WARN_ON_ONCE(conf && unconf);
4320 if (client_has_state(conf)) {
4321 status = nfserr_clientid_busy;
4324 status = mark_client_expired_locked(conf);
4331 status = nfserr_stale_clientid;
4334 if (!nfsd4_mach_creds_match(clp, rqstp)) {
4336 status = nfserr_wrong_cred;
4339 trace_nfsd_clid_destroyed(&clp->cl_clientid);
4340 unhash_client_locked(clp);
4342 spin_unlock(&nn->client_lock);
4349 nfsd4_reclaim_complete(struct svc_rqst *rqstp,
4350 struct nfsd4_compound_state *cstate, union nfsd4_op_u *u)
4352 struct nfsd4_reclaim_complete *rc = &u->reclaim_complete;
4353 struct nfs4_client *clp = cstate->clp;
4356 if (rc->rca_one_fs) {
4357 if (!cstate->current_fh.fh_dentry)
4358 return nfserr_nofilehandle;
4360 * We don't take advantage of the rca_one_fs case.
4361 * That's OK, it's optional, we can safely ignore it.
4366 status = nfserr_complete_already;
4367 if (test_and_set_bit(NFSD4_CLIENT_RECLAIM_COMPLETE, &clp->cl_flags))
4370 status = nfserr_stale_clientid;
4371 if (is_client_expired(clp))
4373 * The following error isn't really legal.
4374 * But we only get here if the client just explicitly
4375 * destroyed the client. Surely it no longer cares what
4376 * error it gets back on an operation for the dead
4382 trace_nfsd_clid_reclaim_complete(&clp->cl_clientid);
4383 nfsd4_client_record_create(clp);
4384 inc_reclaim_complete(clp);
4390 nfsd4_setclientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4391 union nfsd4_op_u *u)
4393 struct nfsd4_setclientid *setclid = &u->setclientid;
4394 struct xdr_netobj clname = setclid->se_name;
4395 nfs4_verifier clverifier = setclid->se_verf;
4396 struct nfs4_client *conf, *new;
4397 struct nfs4_client *unconf = NULL;
4399 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
4401 new = create_client(clname, rqstp, &clverifier);
4403 return nfserr_jukebox;
4404 spin_lock(&nn->client_lock);
4405 conf = find_confirmed_client_by_name(&clname, nn);
4406 if (conf && client_has_state(conf)) {
4407 status = nfserr_clid_inuse;
4408 if (clp_used_exchangeid(conf))
4410 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
4411 trace_nfsd_clid_cred_mismatch(conf, rqstp);
4415 unconf = find_unconfirmed_client_by_name(&clname, nn);
4417 unhash_client_locked(unconf);
4419 if (same_verf(&conf->cl_verifier, &clverifier)) {
4420 copy_clid(new, conf);
4421 gen_confirm(new, nn);
4423 trace_nfsd_clid_verf_mismatch(conf, rqstp,
4426 trace_nfsd_clid_fresh(new);
4427 new->cl_minorversion = 0;
4428 gen_callback(new, setclid, rqstp);
4429 add_to_unconfirmed(new);
4430 setclid->se_clientid.cl_boot = new->cl_clientid.cl_boot;
4431 setclid->se_clientid.cl_id = new->cl_clientid.cl_id;
4432 memcpy(setclid->se_confirm.data, new->cl_confirm.data, sizeof(setclid->se_confirm.data));
4436 spin_unlock(&nn->client_lock);
4440 trace_nfsd_clid_expire_unconf(&unconf->cl_clientid);
4441 expire_client(unconf);
4447 nfsd4_setclientid_confirm(struct svc_rqst *rqstp,
4448 struct nfsd4_compound_state *cstate,
4449 union nfsd4_op_u *u)
4451 struct nfsd4_setclientid_confirm *setclientid_confirm =
4452 &u->setclientid_confirm;
4453 struct nfs4_client *conf, *unconf;
4454 struct nfs4_client *old = NULL;
4455 nfs4_verifier confirm = setclientid_confirm->sc_confirm;
4456 clientid_t * clid = &setclientid_confirm->sc_clientid;
4458 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
4460 if (STALE_CLIENTID(clid, nn))
4461 return nfserr_stale_clientid;
4463 spin_lock(&nn->client_lock);
4464 conf = find_confirmed_client(clid, false, nn);
4465 unconf = find_unconfirmed_client(clid, false, nn);
4467 * We try hard to give out unique clientid's, so if we get an
4468 * attempt to confirm the same clientid with a different cred,
4469 * the client may be buggy; this should never happen.
4471 * Nevertheless, RFC 7530 recommends INUSE for this case:
4473 status = nfserr_clid_inuse;
4474 if (unconf && !same_creds(&unconf->cl_cred, &rqstp->rq_cred)) {
4475 trace_nfsd_clid_cred_mismatch(unconf, rqstp);
4478 if (conf && !same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
4479 trace_nfsd_clid_cred_mismatch(conf, rqstp);
4482 if (!unconf || !same_verf(&confirm, &unconf->cl_confirm)) {
4483 if (conf && same_verf(&confirm, &conf->cl_confirm)) {
4486 status = nfserr_stale_clientid;
4492 unhash_client_locked(old);
4493 nfsd4_change_callback(conf, &unconf->cl_cb_conn);
4495 old = find_confirmed_client_by_name(&unconf->cl_name, nn);
4497 status = nfserr_clid_inuse;
4498 if (client_has_state(old)
4499 && !same_creds(&unconf->cl_cred,
4504 status = mark_client_expired_locked(old);
4509 trace_nfsd_clid_replaced(&old->cl_clientid);
4511 move_to_confirmed(unconf);
4514 get_client_locked(conf);
4515 spin_unlock(&nn->client_lock);
4517 fsnotify_dentry(conf->cl_nfsd_info_dentry, FS_MODIFY);
4518 nfsd4_probe_callback(conf);
4519 spin_lock(&nn->client_lock);
4520 put_client_renew_locked(conf);
4522 spin_unlock(&nn->client_lock);
4528 static struct nfs4_file *nfsd4_alloc_file(void)
4530 return kmem_cache_alloc(file_slab, GFP_KERNEL);
4533 /* OPEN Share state helper functions */
4535 static void nfsd4_file_init(const struct svc_fh *fh, struct nfs4_file *fp)
4537 refcount_set(&fp->fi_ref, 1);
4538 spin_lock_init(&fp->fi_lock);
4539 INIT_LIST_HEAD(&fp->fi_stateids);
4540 INIT_LIST_HEAD(&fp->fi_delegations);
4541 INIT_LIST_HEAD(&fp->fi_clnt_odstate);
4542 fh_copy_shallow(&fp->fi_fhandle, &fh->fh_handle);
4543 fp->fi_deleg_file = NULL;
4544 fp->fi_had_conflict = false;
4545 fp->fi_share_deny = 0;
4546 memset(fp->fi_fds, 0, sizeof(fp->fi_fds));
4547 memset(fp->fi_access, 0, sizeof(fp->fi_access));
4548 fp->fi_aliased = false;
4549 fp->fi_inode = d_inode(fh->fh_dentry);
4550 #ifdef CONFIG_NFSD_PNFS
4551 INIT_LIST_HEAD(&fp->fi_lo_states);
4552 atomic_set(&fp->fi_lo_recalls, 0);
4557 nfsd4_free_slabs(void)
4559 kmem_cache_destroy(client_slab);
4560 kmem_cache_destroy(openowner_slab);
4561 kmem_cache_destroy(lockowner_slab);
4562 kmem_cache_destroy(file_slab);
4563 kmem_cache_destroy(stateid_slab);
4564 kmem_cache_destroy(deleg_slab);
4565 kmem_cache_destroy(odstate_slab);
4569 nfsd4_init_slabs(void)
4571 client_slab = KMEM_CACHE(nfs4_client, 0);
4572 if (client_slab == NULL)
4574 openowner_slab = KMEM_CACHE(nfs4_openowner, 0);
4575 if (openowner_slab == NULL)
4576 goto out_free_client_slab;
4577 lockowner_slab = KMEM_CACHE(nfs4_lockowner, 0);
4578 if (lockowner_slab == NULL)
4579 goto out_free_openowner_slab;
4580 file_slab = KMEM_CACHE(nfs4_file, 0);
4581 if (file_slab == NULL)
4582 goto out_free_lockowner_slab;
4583 stateid_slab = KMEM_CACHE(nfs4_ol_stateid, 0);
4584 if (stateid_slab == NULL)
4585 goto out_free_file_slab;
4586 deleg_slab = KMEM_CACHE(nfs4_delegation, 0);
4587 if (deleg_slab == NULL)
4588 goto out_free_stateid_slab;
4589 odstate_slab = KMEM_CACHE(nfs4_clnt_odstate, 0);
4590 if (odstate_slab == NULL)
4591 goto out_free_deleg_slab;
4594 out_free_deleg_slab:
4595 kmem_cache_destroy(deleg_slab);
4596 out_free_stateid_slab:
4597 kmem_cache_destroy(stateid_slab);
4599 kmem_cache_destroy(file_slab);
4600 out_free_lockowner_slab:
4601 kmem_cache_destroy(lockowner_slab);
4602 out_free_openowner_slab:
4603 kmem_cache_destroy(openowner_slab);
4604 out_free_client_slab:
4605 kmem_cache_destroy(client_slab);
4610 static unsigned long
4611 nfsd4_state_shrinker_count(struct shrinker *shrink, struct shrink_control *sc)
4614 struct nfsd_net *nn = shrink->private_data;
4616 count = atomic_read(&nn->nfsd_courtesy_clients);
4618 count = atomic_long_read(&num_delegations);
4620 queue_work(laundry_wq, &nn->nfsd_shrinker_work);
4621 return (unsigned long)count;
4624 static unsigned long
4625 nfsd4_state_shrinker_scan(struct shrinker *shrink, struct shrink_control *sc)
4631 nfsd4_init_leases_net(struct nfsd_net *nn)
4636 nn->nfsd4_lease = 90; /* default lease time */
4637 nn->nfsd4_grace = 90;
4638 nn->somebody_reclaimed = false;
4639 nn->track_reclaim_completes = false;
4640 nn->clverifier_counter = get_random_u32();
4641 nn->clientid_base = get_random_u32();
4642 nn->clientid_counter = nn->clientid_base + 1;
4643 nn->s2s_cp_cl_id = nn->clientid_counter++;
4645 atomic_set(&nn->nfs4_client_count, 0);
4647 max_clients = (u64)si.totalram * si.mem_unit / (1024 * 1024 * 1024);
4648 max_clients *= NFS4_CLIENTS_PER_GB;
4649 nn->nfs4_max_clients = max_t(int, max_clients, NFS4_CLIENTS_PER_GB);
4651 atomic_set(&nn->nfsd_courtesy_clients, 0);
4654 static void init_nfs4_replay(struct nfs4_replay *rp)
4656 rp->rp_status = nfserr_serverfault;
4658 rp->rp_buf = rp->rp_ibuf;
4659 mutex_init(&rp->rp_mutex);
4662 static void nfsd4_cstate_assign_replay(struct nfsd4_compound_state *cstate,
4663 struct nfs4_stateowner *so)
4665 if (!nfsd4_has_session(cstate)) {
4666 mutex_lock(&so->so_replay.rp_mutex);
4667 cstate->replay_owner = nfs4_get_stateowner(so);
4671 void nfsd4_cstate_clear_replay(struct nfsd4_compound_state *cstate)
4673 struct nfs4_stateowner *so = cstate->replay_owner;
4676 cstate->replay_owner = NULL;
4677 mutex_unlock(&so->so_replay.rp_mutex);
4678 nfs4_put_stateowner(so);
4682 static inline void *alloc_stateowner(struct kmem_cache *slab, struct xdr_netobj *owner, struct nfs4_client *clp)
4684 struct nfs4_stateowner *sop;
4686 sop = kmem_cache_alloc(slab, GFP_KERNEL);
4690 xdr_netobj_dup(&sop->so_owner, owner, GFP_KERNEL);
4691 if (!sop->so_owner.data) {
4692 kmem_cache_free(slab, sop);
4696 INIT_LIST_HEAD(&sop->so_stateids);
4697 sop->so_client = clp;
4698 init_nfs4_replay(&sop->so_replay);
4699 atomic_set(&sop->so_count, 1);
4703 static void hash_openowner(struct nfs4_openowner *oo, struct nfs4_client *clp, unsigned int strhashval)
4705 lockdep_assert_held(&clp->cl_lock);
4707 list_add(&oo->oo_owner.so_strhash,
4708 &clp->cl_ownerstr_hashtbl[strhashval]);
4709 list_add(&oo->oo_perclient, &clp->cl_openowners);
4712 static void nfs4_unhash_openowner(struct nfs4_stateowner *so)
4714 unhash_openowner_locked(openowner(so));
4717 static void nfs4_free_openowner(struct nfs4_stateowner *so)
4719 struct nfs4_openowner *oo = openowner(so);
4721 kmem_cache_free(openowner_slab, oo);
4724 static const struct nfs4_stateowner_operations openowner_ops = {
4725 .so_unhash = nfs4_unhash_openowner,
4726 .so_free = nfs4_free_openowner,
4729 static struct nfs4_ol_stateid *
4730 nfsd4_find_existing_open(struct nfs4_file *fp, struct nfsd4_open *open)
4732 struct nfs4_ol_stateid *local, *ret = NULL;
4733 struct nfs4_openowner *oo = open->op_openowner;
4735 lockdep_assert_held(&fp->fi_lock);
4737 list_for_each_entry(local, &fp->fi_stateids, st_perfile) {
4738 /* ignore lock owners */
4739 if (local->st_stateowner->so_is_open_owner == 0)
4741 if (local->st_stateowner != &oo->oo_owner)
4743 if (local->st_stid.sc_type == SC_TYPE_OPEN &&
4744 !local->st_stid.sc_status) {
4746 refcount_inc(&ret->st_stid.sc_count);
4753 static void nfsd4_drop_revoked_stid(struct nfs4_stid *s)
4754 __releases(&s->sc_client->cl_lock)
4756 struct nfs4_client *cl = s->sc_client;
4757 LIST_HEAD(reaplist);
4758 struct nfs4_ol_stateid *stp;
4759 struct nfs4_delegation *dp;
4762 switch (s->sc_type) {
4764 stp = openlockstateid(s);
4765 if (unhash_open_stateid(stp, &reaplist))
4766 put_ol_stateid_locked(stp, &reaplist);
4767 spin_unlock(&cl->cl_lock);
4768 free_ol_stateid_reaplist(&reaplist);
4771 stp = openlockstateid(s);
4772 unhashed = unhash_lock_stateid(stp);
4773 spin_unlock(&cl->cl_lock);
4778 dp = delegstateid(s);
4779 list_del_init(&dp->dl_recall_lru);
4780 spin_unlock(&cl->cl_lock);
4784 spin_unlock(&cl->cl_lock);
4788 static void nfsd40_drop_revoked_stid(struct nfs4_client *cl,
4791 /* NFSv4.0 has no way for the client to tell the server
4792 * that it can forget an admin-revoked stateid.
4793 * So we keep it around until the first time that the
4794 * client uses it, and drop it the first time
4795 * nfserr_admin_revoked is returned.
4796 * For v4.1 and later we wait until explicitly told
4797 * to free the stateid.
4799 if (cl->cl_minorversion == 0) {
4800 struct nfs4_stid *st;
4802 spin_lock(&cl->cl_lock);
4803 st = find_stateid_locked(cl, stid);
4805 nfsd4_drop_revoked_stid(st);
4807 spin_unlock(&cl->cl_lock);
4812 nfsd4_verify_open_stid(struct nfs4_stid *s)
4814 __be32 ret = nfs_ok;
4816 if (s->sc_status & SC_STATUS_ADMIN_REVOKED)
4817 ret = nfserr_admin_revoked;
4818 else if (s->sc_status & SC_STATUS_REVOKED)
4819 ret = nfserr_deleg_revoked;
4820 else if (s->sc_status & SC_STATUS_CLOSED)
4821 ret = nfserr_bad_stateid;
4825 /* Lock the stateid st_mutex, and deal with races with CLOSE */
4827 nfsd4_lock_ol_stateid(struct nfs4_ol_stateid *stp)
4831 mutex_lock_nested(&stp->st_mutex, LOCK_STATEID_MUTEX);
4832 ret = nfsd4_verify_open_stid(&stp->st_stid);
4833 if (ret == nfserr_admin_revoked)
4834 nfsd40_drop_revoked_stid(stp->st_stid.sc_client,
4835 &stp->st_stid.sc_stateid);
4838 mutex_unlock(&stp->st_mutex);
4842 static struct nfs4_ol_stateid *
4843 nfsd4_find_and_lock_existing_open(struct nfs4_file *fp, struct nfsd4_open *open)
4845 struct nfs4_ol_stateid *stp;
4847 spin_lock(&fp->fi_lock);
4848 stp = nfsd4_find_existing_open(fp, open);
4849 spin_unlock(&fp->fi_lock);
4850 if (!stp || nfsd4_lock_ol_stateid(stp) == nfs_ok)
4852 nfs4_put_stid(&stp->st_stid);
4857 static struct nfs4_openowner *
4858 alloc_init_open_stateowner(unsigned int strhashval, struct nfsd4_open *open,
4859 struct nfsd4_compound_state *cstate)
4861 struct nfs4_client *clp = cstate->clp;
4862 struct nfs4_openowner *oo, *ret;
4864 oo = alloc_stateowner(openowner_slab, &open->op_owner, clp);
4867 oo->oo_owner.so_ops = &openowner_ops;
4868 oo->oo_owner.so_is_open_owner = 1;
4869 oo->oo_owner.so_seqid = open->op_seqid;
4871 if (nfsd4_has_session(cstate))
4872 oo->oo_flags |= NFS4_OO_CONFIRMED;
4874 oo->oo_last_closed_stid = NULL;
4875 INIT_LIST_HEAD(&oo->oo_close_lru);
4876 spin_lock(&clp->cl_lock);
4877 ret = find_openstateowner_str_locked(strhashval, open, clp);
4879 hash_openowner(oo, clp, strhashval);
4882 nfs4_free_stateowner(&oo->oo_owner);
4884 spin_unlock(&clp->cl_lock);
4888 static struct nfs4_ol_stateid *
4889 init_open_stateid(struct nfs4_file *fp, struct nfsd4_open *open)
4892 struct nfs4_openowner *oo = open->op_openowner;
4893 struct nfs4_ol_stateid *retstp = NULL;
4894 struct nfs4_ol_stateid *stp;
4897 /* We are moving these outside of the spinlocks to avoid the warnings */
4898 mutex_init(&stp->st_mutex);
4899 mutex_lock_nested(&stp->st_mutex, OPEN_STATEID_MUTEX);
4902 spin_lock(&oo->oo_owner.so_client->cl_lock);
4903 spin_lock(&fp->fi_lock);
4905 retstp = nfsd4_find_existing_open(fp, open);
4909 open->op_stp = NULL;
4910 refcount_inc(&stp->st_stid.sc_count);
4911 stp->st_stid.sc_type = SC_TYPE_OPEN;
4912 INIT_LIST_HEAD(&stp->st_locks);
4913 stp->st_stateowner = nfs4_get_stateowner(&oo->oo_owner);
4915 stp->st_stid.sc_file = fp;
4916 stp->st_access_bmap = 0;
4917 stp->st_deny_bmap = 0;
4918 stp->st_openstp = NULL;
4919 list_add(&stp->st_perstateowner, &oo->oo_owner.so_stateids);
4920 list_add(&stp->st_perfile, &fp->fi_stateids);
4923 spin_unlock(&fp->fi_lock);
4924 spin_unlock(&oo->oo_owner.so_client->cl_lock);
4926 /* Handle races with CLOSE */
4927 if (nfsd4_lock_ol_stateid(retstp) != nfs_ok) {
4928 nfs4_put_stid(&retstp->st_stid);
4931 /* To keep mutex tracking happy */
4932 mutex_unlock(&stp->st_mutex);
4939 * In the 4.0 case we need to keep the owners around a little while to handle
4940 * CLOSE replay. We still do need to release any file access that is held by
4941 * them before returning however.
4944 move_to_close_lru(struct nfs4_ol_stateid *s, struct net *net)
4946 struct nfs4_ol_stateid *last;
4947 struct nfs4_openowner *oo = openowner(s->st_stateowner);
4948 struct nfsd_net *nn = net_generic(s->st_stid.sc_client->net,
4951 dprintk("NFSD: move_to_close_lru nfs4_openowner %p\n", oo);
4954 * We know that we hold one reference via nfsd4_close, and another
4955 * "persistent" reference for the client. If the refcount is higher
4956 * than 2, then there are still calls in progress that are using this
4957 * stateid. We can't put the sc_file reference until they are finished.
4958 * Wait for the refcount to drop to 2. Since it has been unhashed,
4959 * there should be no danger of the refcount going back up again at
4962 wait_event(close_wq, refcount_read(&s->st_stid.sc_count) == 2);
4964 release_all_access(s);
4965 if (s->st_stid.sc_file) {
4966 put_nfs4_file(s->st_stid.sc_file);
4967 s->st_stid.sc_file = NULL;
4970 spin_lock(&nn->client_lock);
4971 last = oo->oo_last_closed_stid;
4972 oo->oo_last_closed_stid = s;
4973 list_move_tail(&oo->oo_close_lru, &nn->close_lru);
4974 oo->oo_time = ktime_get_boottime_seconds();
4975 spin_unlock(&nn->client_lock);
4977 nfs4_put_stid(&last->st_stid);
4980 static noinline_for_stack struct nfs4_file *
4981 nfsd4_file_hash_lookup(const struct svc_fh *fhp)
4983 struct inode *inode = d_inode(fhp->fh_dentry);
4984 struct rhlist_head *tmp, *list;
4985 struct nfs4_file *fi;
4988 list = rhltable_lookup(&nfs4_file_rhltable, &inode,
4989 nfs4_file_rhash_params);
4990 rhl_for_each_entry_rcu(fi, tmp, list, fi_rlist) {
4991 if (fh_match(&fi->fi_fhandle, &fhp->fh_handle)) {
4992 if (refcount_inc_not_zero(&fi->fi_ref)) {
5003 * On hash insertion, identify entries with the same inode but
5004 * distinct filehandles. They will all be on the list returned
5005 * by rhltable_lookup().
5007 * inode->i_lock prevents racing insertions from adding an entry
5008 * for the same inode/fhp pair twice.
5010 static noinline_for_stack struct nfs4_file *
5011 nfsd4_file_hash_insert(struct nfs4_file *new, const struct svc_fh *fhp)
5013 struct inode *inode = d_inode(fhp->fh_dentry);
5014 struct rhlist_head *tmp, *list;
5015 struct nfs4_file *ret = NULL;
5016 bool alias_found = false;
5017 struct nfs4_file *fi;
5021 spin_lock(&inode->i_lock);
5023 list = rhltable_lookup(&nfs4_file_rhltable, &inode,
5024 nfs4_file_rhash_params);
5025 rhl_for_each_entry_rcu(fi, tmp, list, fi_rlist) {
5026 if (fh_match(&fi->fi_fhandle, &fhp->fh_handle)) {
5027 if (refcount_inc_not_zero(&fi->fi_ref))
5030 fi->fi_aliased = alias_found = true;
5035 nfsd4_file_init(fhp, new);
5036 err = rhltable_insert(&nfs4_file_rhltable, &new->fi_rlist,
5037 nfs4_file_rhash_params);
5041 new->fi_aliased = alias_found;
5045 spin_unlock(&inode->i_lock);
5050 static noinline_for_stack void nfsd4_file_hash_remove(struct nfs4_file *fi)
5052 rhltable_remove(&nfs4_file_rhltable, &fi->fi_rlist,
5053 nfs4_file_rhash_params);
5057 * Called to check deny when READ with all zero stateid or
5058 * WRITE with all zero or all one stateid
5061 nfs4_share_conflict(struct svc_fh *current_fh, unsigned int deny_type)
5063 struct nfs4_file *fp;
5064 __be32 ret = nfs_ok;
5066 fp = nfsd4_file_hash_lookup(current_fh);
5070 /* Check for conflicting share reservations */
5071 spin_lock(&fp->fi_lock);
5072 if (fp->fi_share_deny & deny_type)
5073 ret = nfserr_locked;
5074 spin_unlock(&fp->fi_lock);
5079 static bool nfsd4_deleg_present(const struct inode *inode)
5081 struct file_lock_context *ctx = locks_inode_context(inode);
5083 return ctx && !list_empty_careful(&ctx->flc_lease);
5087 * nfsd_wait_for_delegreturn - wait for delegations to be returned
5088 * @rqstp: the RPC transaction being executed
5089 * @inode: in-core inode of the file being waited for
5091 * The timeout prevents deadlock if all nfsd threads happen to be
5092 * tied up waiting for returning delegations.
5095 * %true: delegation was returned
5096 * %false: timed out waiting for delegreturn
5098 bool nfsd_wait_for_delegreturn(struct svc_rqst *rqstp, struct inode *inode)
5100 long __maybe_unused timeo;
5102 timeo = wait_var_event_timeout(inode, !nfsd4_deleg_present(inode),
5103 NFSD_DELEGRETURN_TIMEOUT);
5104 trace_nfsd_delegret_wakeup(rqstp, inode, timeo);
5108 static void nfsd4_cb_recall_prepare(struct nfsd4_callback *cb)
5110 struct nfs4_delegation *dp = cb_to_delegation(cb);
5111 struct nfsd_net *nn = net_generic(dp->dl_stid.sc_client->net,
5114 block_delegations(&dp->dl_stid.sc_file->fi_fhandle);
5117 * We can't do this in nfsd_break_deleg_cb because it is
5118 * already holding inode->i_lock.
5120 * If the dl_time != 0, then we know that it has already been
5121 * queued for a lease break. Don't queue it again.
5123 spin_lock(&state_lock);
5124 if (delegation_hashed(dp) && dp->dl_time == 0) {
5125 dp->dl_time = ktime_get_boottime_seconds();
5126 list_add_tail(&dp->dl_recall_lru, &nn->del_recall_lru);
5128 spin_unlock(&state_lock);
5131 static int nfsd4_cb_recall_done(struct nfsd4_callback *cb,
5132 struct rpc_task *task)
5134 struct nfs4_delegation *dp = cb_to_delegation(cb);
5136 trace_nfsd_cb_recall_done(&dp->dl_stid.sc_stateid, task);
5138 if (dp->dl_stid.sc_status)
5139 /* CLOSED or REVOKED */
5142 switch (task->tk_status) {
5145 case -NFS4ERR_DELAY:
5146 rpc_delay(task, 2 * HZ);
5149 case -NFS4ERR_BAD_STATEID:
5151 * Race: client probably got cb_recall before open reply
5152 * granting delegation.
5154 if (dp->dl_retries--) {
5155 rpc_delay(task, 2 * HZ);
5164 static void nfsd4_cb_recall_release(struct nfsd4_callback *cb)
5166 struct nfs4_delegation *dp = cb_to_delegation(cb);
5168 nfs4_put_stid(&dp->dl_stid);
5171 static const struct nfsd4_callback_ops nfsd4_cb_recall_ops = {
5172 .prepare = nfsd4_cb_recall_prepare,
5173 .done = nfsd4_cb_recall_done,
5174 .release = nfsd4_cb_recall_release,
5177 static void nfsd_break_one_deleg(struct nfs4_delegation *dp)
5180 * We're assuming the state code never drops its reference
5181 * without first removing the lease. Since we're in this lease
5182 * callback (and since the lease code is serialized by the
5183 * flc_lock) we know the server hasn't removed the lease yet, and
5184 * we know it's safe to take a reference.
5186 refcount_inc(&dp->dl_stid.sc_count);
5187 WARN_ON_ONCE(!nfsd4_run_cb(&dp->dl_recall));
5190 /* Called from break_lease() with flc_lock held. */
5192 nfsd_break_deleg_cb(struct file_lease *fl)
5194 struct nfs4_delegation *dp = (struct nfs4_delegation *) fl->c.flc_owner;
5195 struct nfs4_file *fp = dp->dl_stid.sc_file;
5196 struct nfs4_client *clp = dp->dl_stid.sc_client;
5197 struct nfsd_net *nn;
5199 trace_nfsd_cb_recall(&dp->dl_stid);
5201 dp->dl_recalled = true;
5202 atomic_inc(&clp->cl_delegs_in_recall);
5203 if (try_to_expire_client(clp)) {
5204 nn = net_generic(clp->net, nfsd_net_id);
5205 mod_delayed_work(laundry_wq, &nn->laundromat_work, 0);
5209 * We don't want the locks code to timeout the lease for us;
5210 * we'll remove it ourself if a delegation isn't returned
5213 fl->fl_break_time = 0;
5215 fp->fi_had_conflict = true;
5216 nfsd_break_one_deleg(dp);
5221 * nfsd_breaker_owns_lease - Check if lease conflict was resolved
5222 * @fl: Lock state to check
5225 * %true: Lease conflict was resolved
5226 * %false: Lease conflict was not resolved.
5228 static bool nfsd_breaker_owns_lease(struct file_lease *fl)
5230 struct nfs4_delegation *dl = fl->c.flc_owner;
5231 struct svc_rqst *rqst;
5232 struct nfs4_client *clp;
5236 rqst = kthread_data(current);
5237 /* Note rq_prog == NFS_ACL_PROGRAM is also possible: */
5238 if (rqst->rq_prog != NFS_PROGRAM || rqst->rq_vers < 4)
5240 clp = *(rqst->rq_lease_breaker);
5241 return dl->dl_stid.sc_client == clp;
5245 nfsd_change_deleg_cb(struct file_lease *onlist, int arg,
5246 struct list_head *dispose)
5248 struct nfs4_delegation *dp = (struct nfs4_delegation *) onlist->c.flc_owner;
5249 struct nfs4_client *clp = dp->dl_stid.sc_client;
5251 if (arg & F_UNLCK) {
5252 if (dp->dl_recalled)
5253 atomic_dec(&clp->cl_delegs_in_recall);
5254 return lease_modify(onlist, arg, dispose);
5259 static const struct lease_manager_operations nfsd_lease_mng_ops = {
5260 .lm_breaker_owns_lease = nfsd_breaker_owns_lease,
5261 .lm_break = nfsd_break_deleg_cb,
5262 .lm_change = nfsd_change_deleg_cb,
5265 static __be32 nfsd4_check_seqid(struct nfsd4_compound_state *cstate, struct nfs4_stateowner *so, u32 seqid)
5267 if (nfsd4_has_session(cstate))
5269 if (seqid == so->so_seqid - 1)
5270 return nfserr_replay_me;
5271 if (seqid == so->so_seqid)
5273 return nfserr_bad_seqid;
5276 static struct nfs4_client *lookup_clientid(clientid_t *clid, bool sessions,
5277 struct nfsd_net *nn)
5279 struct nfs4_client *found;
5281 spin_lock(&nn->client_lock);
5282 found = find_confirmed_client(clid, sessions, nn);
5284 atomic_inc(&found->cl_rpc_users);
5285 spin_unlock(&nn->client_lock);
5289 static __be32 set_client(clientid_t *clid,
5290 struct nfsd4_compound_state *cstate,
5291 struct nfsd_net *nn)
5294 if (!same_clid(&cstate->clp->cl_clientid, clid))
5295 return nfserr_stale_clientid;
5298 if (STALE_CLIENTID(clid, nn))
5299 return nfserr_stale_clientid;
5301 * We're in the 4.0 case (otherwise the SEQUENCE op would have
5302 * set cstate->clp), so session = false:
5304 cstate->clp = lookup_clientid(clid, false, nn);
5306 return nfserr_expired;
5311 nfsd4_process_open1(struct nfsd4_compound_state *cstate,
5312 struct nfsd4_open *open, struct nfsd_net *nn)
5314 clientid_t *clientid = &open->op_clientid;
5315 struct nfs4_client *clp = NULL;
5316 unsigned int strhashval;
5317 struct nfs4_openowner *oo = NULL;
5321 * In case we need it later, after we've already created the
5322 * file and don't want to risk a further failure:
5324 open->op_file = nfsd4_alloc_file();
5325 if (open->op_file == NULL)
5326 return nfserr_jukebox;
5328 status = set_client(clientid, cstate, nn);
5333 strhashval = ownerstr_hashval(&open->op_owner);
5334 oo = find_openstateowner_str(strhashval, open, clp);
5335 open->op_openowner = oo;
5339 if (!(oo->oo_flags & NFS4_OO_CONFIRMED)) {
5340 /* Replace unconfirmed owners without checking for replay. */
5341 release_openowner(oo);
5342 open->op_openowner = NULL;
5345 status = nfsd4_check_seqid(cstate, &oo->oo_owner, open->op_seqid);
5350 oo = alloc_init_open_stateowner(strhashval, open, cstate);
5352 return nfserr_jukebox;
5353 open->op_openowner = oo;
5355 open->op_stp = nfs4_alloc_open_stateid(clp);
5357 return nfserr_jukebox;
5359 if (nfsd4_has_session(cstate) &&
5360 (cstate->current_fh.fh_export->ex_flags & NFSEXP_PNFS)) {
5361 open->op_odstate = alloc_clnt_odstate(clp);
5362 if (!open->op_odstate)
5363 return nfserr_jukebox;
5369 static inline __be32
5370 nfs4_check_delegmode(struct nfs4_delegation *dp, int flags)
5372 if ((flags & WR_STATE) && (dp->dl_type == NFS4_OPEN_DELEGATE_READ))
5373 return nfserr_openmode;
5378 static int share_access_to_flags(u32 share_access)
5380 return share_access == NFS4_SHARE_ACCESS_READ ? RD_STATE : WR_STATE;
5383 static struct nfs4_delegation *find_deleg_stateid(struct nfs4_client *cl,
5386 struct nfs4_stid *ret;
5388 ret = find_stateid_by_type(cl, s, SC_TYPE_DELEG, SC_STATUS_REVOKED);
5391 return delegstateid(ret);
5394 static bool nfsd4_is_deleg_cur(struct nfsd4_open *open)
5396 return open->op_claim_type == NFS4_OPEN_CLAIM_DELEGATE_CUR ||
5397 open->op_claim_type == NFS4_OPEN_CLAIM_DELEG_CUR_FH;
5401 nfs4_check_deleg(struct nfs4_client *cl, struct nfsd4_open *open,
5402 struct nfs4_delegation **dp)
5405 __be32 status = nfserr_bad_stateid;
5406 struct nfs4_delegation *deleg;
5408 deleg = find_deleg_stateid(cl, &open->op_delegate_stateid);
5411 if (deleg->dl_stid.sc_status & SC_STATUS_ADMIN_REVOKED) {
5412 nfs4_put_stid(&deleg->dl_stid);
5413 status = nfserr_admin_revoked;
5416 if (deleg->dl_stid.sc_status & SC_STATUS_REVOKED) {
5417 nfs4_put_stid(&deleg->dl_stid);
5418 nfsd40_drop_revoked_stid(cl, &open->op_delegate_stateid);
5419 status = nfserr_deleg_revoked;
5422 flags = share_access_to_flags(open->op_share_access);
5423 status = nfs4_check_delegmode(deleg, flags);
5425 nfs4_put_stid(&deleg->dl_stid);
5430 if (!nfsd4_is_deleg_cur(open))
5434 open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
5438 static inline int nfs4_access_to_access(u32 nfs4_access)
5442 if (nfs4_access & NFS4_SHARE_ACCESS_READ)
5443 flags |= NFSD_MAY_READ;
5444 if (nfs4_access & NFS4_SHARE_ACCESS_WRITE)
5445 flags |= NFSD_MAY_WRITE;
5449 static inline __be32
5450 nfsd4_truncate(struct svc_rqst *rqstp, struct svc_fh *fh,
5451 struct nfsd4_open *open)
5453 struct iattr iattr = {
5454 .ia_valid = ATTR_SIZE,
5457 struct nfsd_attrs attrs = {
5460 if (!open->op_truncate)
5462 if (!(open->op_share_access & NFS4_SHARE_ACCESS_WRITE))
5463 return nfserr_inval;
5464 return nfsd_setattr(rqstp, fh, &attrs, NULL);
5467 static __be32 nfs4_get_vfs_file(struct svc_rqst *rqstp, struct nfs4_file *fp,
5468 struct svc_fh *cur_fh, struct nfs4_ol_stateid *stp,
5469 struct nfsd4_open *open, bool new_stp)
5471 struct nfsd_file *nf = NULL;
5473 int oflag = nfs4_access_to_omode(open->op_share_access);
5474 int access = nfs4_access_to_access(open->op_share_access);
5475 unsigned char old_access_bmap, old_deny_bmap;
5477 spin_lock(&fp->fi_lock);
5480 * Are we trying to set a deny mode that would conflict with
5483 status = nfs4_file_check_deny(fp, open->op_share_deny);
5484 if (status != nfs_ok) {
5485 if (status != nfserr_share_denied) {
5486 spin_unlock(&fp->fi_lock);
5489 if (nfs4_resolve_deny_conflicts_locked(fp, new_stp,
5490 stp, open->op_share_deny, false))
5491 status = nfserr_jukebox;
5492 spin_unlock(&fp->fi_lock);
5496 /* set access to the file */
5497 status = nfs4_file_get_access(fp, open->op_share_access);
5498 if (status != nfs_ok) {
5499 if (status != nfserr_share_denied) {
5500 spin_unlock(&fp->fi_lock);
5503 if (nfs4_resolve_deny_conflicts_locked(fp, new_stp,
5504 stp, open->op_share_access, true))
5505 status = nfserr_jukebox;
5506 spin_unlock(&fp->fi_lock);
5510 /* Set access bits in stateid */
5511 old_access_bmap = stp->st_access_bmap;
5512 set_access(open->op_share_access, stp);
5514 /* Set new deny mask */
5515 old_deny_bmap = stp->st_deny_bmap;
5516 set_deny(open->op_share_deny, stp);
5517 fp->fi_share_deny |= (open->op_share_deny & NFS4_SHARE_DENY_BOTH);
5519 if (!fp->fi_fds[oflag]) {
5520 spin_unlock(&fp->fi_lock);
5522 status = nfsd_file_acquire_opened(rqstp, cur_fh, access,
5523 open->op_filp, &nf);
5524 if (status != nfs_ok)
5525 goto out_put_access;
5527 spin_lock(&fp->fi_lock);
5528 if (!fp->fi_fds[oflag]) {
5529 fp->fi_fds[oflag] = nf;
5533 spin_unlock(&fp->fi_lock);
5537 status = nfserrno(nfsd_open_break_lease(cur_fh->fh_dentry->d_inode,
5540 goto out_put_access;
5542 status = nfsd4_truncate(rqstp, cur_fh, open);
5544 goto out_put_access;
5548 stp->st_access_bmap = old_access_bmap;
5549 nfs4_file_put_access(fp, open->op_share_access);
5550 reset_union_bmap_deny(bmap_to_share_mode(old_deny_bmap), stp);
5555 nfs4_upgrade_open(struct svc_rqst *rqstp, struct nfs4_file *fp,
5556 struct svc_fh *cur_fh, struct nfs4_ol_stateid *stp,
5557 struct nfsd4_open *open)
5560 unsigned char old_deny_bmap = stp->st_deny_bmap;
5562 if (!test_access(open->op_share_access, stp))
5563 return nfs4_get_vfs_file(rqstp, fp, cur_fh, stp, open, false);
5565 /* test and set deny mode */
5566 spin_lock(&fp->fi_lock);
5567 status = nfs4_file_check_deny(fp, open->op_share_deny);
5570 set_deny(open->op_share_deny, stp);
5571 fp->fi_share_deny |=
5572 (open->op_share_deny & NFS4_SHARE_DENY_BOTH);
5574 case nfserr_share_denied:
5575 if (nfs4_resolve_deny_conflicts_locked(fp, false,
5576 stp, open->op_share_deny, false))
5577 status = nfserr_jukebox;
5580 spin_unlock(&fp->fi_lock);
5582 if (status != nfs_ok)
5585 status = nfsd4_truncate(rqstp, cur_fh, open);
5586 if (status != nfs_ok)
5587 reset_union_bmap_deny(old_deny_bmap, stp);
5591 /* Should we give out recallable state?: */
5592 static bool nfsd4_cb_channel_good(struct nfs4_client *clp)
5594 if (clp->cl_cb_state == NFSD4_CB_UP)
5597 * In the sessions case, since we don't have to establish a
5598 * separate connection for callbacks, we assume it's OK
5599 * until we hear otherwise:
5601 return clp->cl_minorversion && clp->cl_cb_state == NFSD4_CB_UNKNOWN;
5604 static struct file_lease *nfs4_alloc_init_lease(struct nfs4_delegation *dp,
5607 struct file_lease *fl;
5609 fl = locks_alloc_lease();
5612 fl->fl_lmops = &nfsd_lease_mng_ops;
5613 fl->c.flc_flags = FL_DELEG;
5614 fl->c.flc_type = flag == NFS4_OPEN_DELEGATE_READ? F_RDLCK: F_WRLCK;
5615 fl->c.flc_owner = (fl_owner_t)dp;
5616 fl->c.flc_pid = current->tgid;
5617 fl->c.flc_file = dp->dl_stid.sc_file->fi_deleg_file->nf_file;
5621 static int nfsd4_check_conflicting_opens(struct nfs4_client *clp,
5622 struct nfs4_file *fp)
5624 struct nfs4_ol_stateid *st;
5625 struct file *f = fp->fi_deleg_file->nf_file;
5626 struct inode *ino = file_inode(f);
5629 writes = atomic_read(&ino->i_writecount);
5633 * There could be multiple filehandles (hence multiple
5634 * nfs4_files) referencing this file, but that's not too
5635 * common; let's just give up in that case rather than
5636 * trying to go look up all the clients using that other
5637 * nfs4_file as well:
5642 * If there's a close in progress, make sure that we see it
5643 * clear any fi_fds[] entries before we see it decrement
5646 smp_mb__after_atomic();
5648 if (fp->fi_fds[O_WRONLY])
5650 if (fp->fi_fds[O_RDWR])
5653 return -EAGAIN; /* There may be non-NFSv4 writers */
5655 * It's possible there are non-NFSv4 write opens in progress,
5656 * but if they haven't incremented i_writecount yet then they
5657 * also haven't called break lease yet; so, they'll break this
5658 * lease soon enough. So, all that's left to check for is NFSv4
5661 spin_lock(&fp->fi_lock);
5662 list_for_each_entry(st, &fp->fi_stateids, st_perfile) {
5663 if (st->st_openstp == NULL /* it's an open */ &&
5664 access_permit_write(st) &&
5665 st->st_stid.sc_client != clp) {
5666 spin_unlock(&fp->fi_lock);
5670 spin_unlock(&fp->fi_lock);
5672 * There's a small chance that we could be racing with another
5673 * NFSv4 open. However, any open that hasn't added itself to
5674 * the fi_stateids list also hasn't called break_lease yet; so,
5675 * they'll break this lease soon enough.
5681 * It's possible that between opening the dentry and setting the delegation,
5682 * that it has been renamed or unlinked. Redo the lookup to verify that this
5686 nfsd4_verify_deleg_dentry(struct nfsd4_open *open, struct nfs4_file *fp,
5687 struct svc_fh *parent)
5689 struct svc_export *exp;
5690 struct dentry *child;
5693 err = nfsd_lookup_dentry(open->op_rqstp, parent,
5694 open->op_fname, open->op_fnamelen,
5702 if (child != file_dentry(fp->fi_deleg_file->nf_file))
5709 * We avoid breaking delegations held by a client due to its own activity, but
5710 * clearing setuid/setgid bits on a write is an implicit activity and the client
5711 * may not notice and continue using the old mode. Avoid giving out a delegation
5712 * on setuid/setgid files when the client is requesting an open for write.
5715 nfsd4_verify_setuid_write(struct nfsd4_open *open, struct nfsd_file *nf)
5717 struct inode *inode = file_inode(nf->nf_file);
5719 if ((open->op_share_access & NFS4_SHARE_ACCESS_WRITE) &&
5720 (inode->i_mode & (S_ISUID|S_ISGID)))
5725 static struct nfs4_delegation *
5726 nfs4_set_delegation(struct nfsd4_open *open, struct nfs4_ol_stateid *stp,
5727 struct svc_fh *parent)
5730 struct nfs4_client *clp = stp->st_stid.sc_client;
5731 struct nfs4_file *fp = stp->st_stid.sc_file;
5732 struct nfs4_clnt_odstate *odstate = stp->st_clnt_odstate;
5733 struct nfs4_delegation *dp;
5734 struct nfsd_file *nf = NULL;
5735 struct file_lease *fl;
5739 * The fi_had_conflict and nfs_get_existing_delegation checks
5740 * here are just optimizations; we'll need to recheck them at
5743 if (fp->fi_had_conflict)
5744 return ERR_PTR(-EAGAIN);
5747 * Try for a write delegation first. RFC8881 section 10.4 says:
5749 * "An OPEN_DELEGATE_WRITE delegation allows the client to handle,
5750 * on its own, all opens."
5752 * Furthermore the client can use a write delegation for most READ
5753 * operations as well, so we require a O_RDWR file here.
5755 * Offer a write delegation in the case of a BOTH open, and ensure
5756 * we get the O_RDWR descriptor.
5758 if ((open->op_share_access & NFS4_SHARE_ACCESS_BOTH) == NFS4_SHARE_ACCESS_BOTH) {
5759 nf = find_rw_file(fp);
5760 dl_type = NFS4_OPEN_DELEGATE_WRITE;
5764 * If the file is being opened O_RDONLY or we couldn't get a O_RDWR
5765 * file for some reason, then try for a read delegation instead.
5767 if (!nf && (open->op_share_access & NFS4_SHARE_ACCESS_READ)) {
5768 nf = find_readable_file(fp);
5769 dl_type = NFS4_OPEN_DELEGATE_READ;
5773 return ERR_PTR(-EAGAIN);
5775 spin_lock(&state_lock);
5776 spin_lock(&fp->fi_lock);
5777 if (nfs4_delegation_exists(clp, fp))
5779 else if (nfsd4_verify_setuid_write(open, nf))
5781 else if (!fp->fi_deleg_file) {
5782 fp->fi_deleg_file = nf;
5783 /* increment early to prevent fi_deleg_file from being
5785 fp->fi_delegees = 1;
5789 spin_unlock(&fp->fi_lock);
5790 spin_unlock(&state_lock);
5794 return ERR_PTR(status);
5797 dp = alloc_init_deleg(clp, fp, odstate, dl_type);
5801 fl = nfs4_alloc_init_lease(dp, dl_type);
5803 goto out_clnt_odstate;
5805 status = kernel_setlease(fp->fi_deleg_file->nf_file,
5806 fl->c.flc_type, &fl, NULL);
5808 locks_free_lease(fl);
5810 goto out_clnt_odstate;
5813 status = nfsd4_verify_deleg_dentry(open, fp, parent);
5818 status = nfsd4_check_conflicting_opens(clp, fp);
5823 * Now that the deleg is set, check again to ensure that nothing
5824 * raced in and changed the mode while we weren't lookng.
5826 status = nfsd4_verify_setuid_write(open, fp->fi_deleg_file);
5831 if (fp->fi_had_conflict)
5834 spin_lock(&state_lock);
5835 spin_lock(&clp->cl_lock);
5836 spin_lock(&fp->fi_lock);
5837 status = hash_delegation_locked(dp, fp);
5838 spin_unlock(&fp->fi_lock);
5839 spin_unlock(&clp->cl_lock);
5840 spin_unlock(&state_lock);
5847 kernel_setlease(fp->fi_deleg_file->nf_file, F_UNLCK, NULL, (void **)&dp);
5849 put_clnt_odstate(dp->dl_clnt_odstate);
5850 nfs4_put_stid(&dp->dl_stid);
5853 return ERR_PTR(status);
5856 static void nfsd4_open_deleg_none_ext(struct nfsd4_open *open, int status)
5858 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
5859 if (status == -EAGAIN)
5860 open->op_why_no_deleg = WND4_CONTENTION;
5862 open->op_why_no_deleg = WND4_RESOURCE;
5863 switch (open->op_deleg_want) {
5864 case NFS4_SHARE_WANT_READ_DELEG:
5865 case NFS4_SHARE_WANT_WRITE_DELEG:
5866 case NFS4_SHARE_WANT_ANY_DELEG:
5868 case NFS4_SHARE_WANT_CANCEL:
5869 open->op_why_no_deleg = WND4_CANCELLED;
5871 case NFS4_SHARE_WANT_NO_DELEG:
5878 * The Linux NFS server does not offer write delegations to NFSv4.0
5879 * clients in order to avoid conflicts between write delegations and
5880 * GETATTRs requesting CHANGE or SIZE attributes.
5882 * With NFSv4.1 and later minorversions, the SEQUENCE operation that
5883 * begins each COMPOUND contains a client ID. Delegation recall can
5884 * be avoided when the server recognizes the client sending a
5885 * GETATTR also holds write delegation it conflicts with.
5887 * However, the NFSv4.0 protocol does not enable a server to
5888 * determine that a GETATTR originated from the client holding the
5889 * conflicting delegation versus coming from some other client. Per
5890 * RFC 7530 Section 16.7.5, the server must recall or send a
5891 * CB_GETATTR even when the GETATTR originates from the client that
5892 * holds the conflicting delegation.
5894 * An NFSv4.0 client can trigger a pathological situation if it
5895 * always sends a DELEGRETURN preceded by a conflicting GETATTR in
5896 * the same COMPOUND. COMPOUND execution will always stop at the
5897 * GETATTR and the DELEGRETURN will never get executed. The server
5898 * eventually revokes the delegation, which can result in loss of
5899 * open or lock state.
5902 nfs4_open_delegation(struct nfsd4_open *open, struct nfs4_ol_stateid *stp,
5903 struct svc_fh *currentfh)
5905 struct nfs4_delegation *dp;
5906 struct nfs4_openowner *oo = openowner(stp->st_stateowner);
5907 struct nfs4_client *clp = stp->st_stid.sc_client;
5908 struct svc_fh *parent = NULL;
5914 cb_up = nfsd4_cb_channel_good(oo->oo_owner.so_client);
5915 open->op_recall = false;
5916 switch (open->op_claim_type) {
5917 case NFS4_OPEN_CLAIM_PREVIOUS:
5919 open->op_recall = true;
5921 case NFS4_OPEN_CLAIM_NULL:
5924 case NFS4_OPEN_CLAIM_FH:
5926 * Let's not give out any delegations till everyone's
5927 * had the chance to reclaim theirs, *and* until
5928 * NLM locks have all been reclaimed:
5930 if (locks_in_grace(clp->net))
5932 if (!cb_up || !(oo->oo_flags & NFS4_OO_CONFIRMED))
5934 if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE &&
5935 !clp->cl_minorversion)
5941 dp = nfs4_set_delegation(open, stp, parent);
5945 memcpy(&open->op_delegate_stateid, &dp->dl_stid.sc_stateid, sizeof(dp->dl_stid.sc_stateid));
5947 if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE) {
5948 open->op_delegate_type = NFS4_OPEN_DELEGATE_WRITE;
5949 trace_nfsd_deleg_write(&dp->dl_stid.sc_stateid);
5950 path.mnt = currentfh->fh_export->ex_path.mnt;
5951 path.dentry = currentfh->fh_dentry;
5952 if (vfs_getattr(&path, &stat,
5953 (STATX_SIZE | STATX_CTIME | STATX_CHANGE_COOKIE),
5954 AT_STATX_SYNC_AS_STAT)) {
5955 nfs4_put_stid(&dp->dl_stid);
5956 destroy_delegation(dp);
5959 dp->dl_cb_fattr.ncf_cur_fsize = stat.size;
5960 dp->dl_cb_fattr.ncf_initial_cinfo =
5961 nfsd4_change_attribute(&stat, d_inode(currentfh->fh_dentry));
5963 open->op_delegate_type = NFS4_OPEN_DELEGATE_READ;
5964 trace_nfsd_deleg_read(&dp->dl_stid.sc_stateid);
5966 nfs4_put_stid(&dp->dl_stid);
5969 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE;
5970 if (open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS &&
5971 open->op_delegate_type != NFS4_OPEN_DELEGATE_NONE) {
5972 dprintk("NFSD: WARNING: refusing delegation reclaim\n");
5973 open->op_recall = true;
5976 /* 4.1 client asking for a delegation? */
5977 if (open->op_deleg_want)
5978 nfsd4_open_deleg_none_ext(open, status);
5982 static void nfsd4_deleg_xgrade_none_ext(struct nfsd4_open *open,
5983 struct nfs4_delegation *dp)
5985 if (open->op_deleg_want == NFS4_SHARE_WANT_READ_DELEG &&
5986 dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
5987 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
5988 open->op_why_no_deleg = WND4_NOT_SUPP_DOWNGRADE;
5989 } else if (open->op_deleg_want == NFS4_SHARE_WANT_WRITE_DELEG &&
5990 dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
5991 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
5992 open->op_why_no_deleg = WND4_NOT_SUPP_UPGRADE;
5994 /* Otherwise the client must be confused wanting a delegation
5995 * it already has, therefore we don't return
5996 * NFS4_OPEN_DELEGATE_NONE_EXT and reason.
6001 * nfsd4_process_open2 - finish open processing
6002 * @rqstp: the RPC transaction being executed
6003 * @current_fh: NFSv4 COMPOUND's current filehandle
6004 * @open: OPEN arguments
6006 * If successful, (1) truncate the file if open->op_truncate was
6007 * set, (2) set open->op_stateid, (3) set open->op_delegation.
6009 * Returns %nfs_ok on success; otherwise an nfs4stat value in
6010 * network byte order is returned.
6013 nfsd4_process_open2(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_open *open)
6015 struct nfsd4_compoundres *resp = rqstp->rq_resp;
6016 struct nfs4_client *cl = open->op_openowner->oo_owner.so_client;
6017 struct nfs4_file *fp = NULL;
6018 struct nfs4_ol_stateid *stp = NULL;
6019 struct nfs4_delegation *dp = NULL;
6021 bool new_stp = false;
6024 * Lookup file; if found, lookup stateid and check open request,
6025 * and check for delegations in the process of being recalled.
6026 * If not found, create the nfs4_file struct
6028 fp = nfsd4_file_hash_insert(open->op_file, current_fh);
6030 return nfserr_jukebox;
6031 if (fp != open->op_file) {
6032 status = nfs4_check_deleg(cl, open, &dp);
6035 stp = nfsd4_find_and_lock_existing_open(fp, open);
6037 open->op_file = NULL;
6038 status = nfserr_bad_stateid;
6039 if (nfsd4_is_deleg_cur(open))
6044 stp = init_open_stateid(fp, open);
6050 * OPEN the file, or upgrade an existing OPEN.
6051 * If truncate fails, the OPEN fails.
6053 * stp is already locked.
6056 /* Stateid was found, this is an OPEN upgrade */
6057 status = nfs4_upgrade_open(rqstp, fp, current_fh, stp, open);
6059 mutex_unlock(&stp->st_mutex);
6063 status = nfs4_get_vfs_file(rqstp, fp, current_fh, stp, open, true);
6065 release_open_stateid(stp);
6066 mutex_unlock(&stp->st_mutex);
6070 stp->st_clnt_odstate = find_or_hash_clnt_odstate(fp,
6072 if (stp->st_clnt_odstate == open->op_odstate)
6073 open->op_odstate = NULL;
6076 nfs4_inc_and_copy_stateid(&open->op_stateid, &stp->st_stid);
6077 mutex_unlock(&stp->st_mutex);
6079 if (nfsd4_has_session(&resp->cstate)) {
6080 if (open->op_deleg_want & NFS4_SHARE_WANT_NO_DELEG) {
6081 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
6082 open->op_why_no_deleg = WND4_NOT_WANTED;
6088 * Attempt to hand out a delegation. No error return, because the
6089 * OPEN succeeds even if we fail.
6091 nfs4_open_delegation(open, stp, &resp->cstate.current_fh);
6094 trace_nfsd_open(&stp->st_stid.sc_stateid);
6096 /* 4.1 client trying to upgrade/downgrade delegation? */
6097 if (open->op_delegate_type == NFS4_OPEN_DELEGATE_NONE && dp &&
6098 open->op_deleg_want)
6099 nfsd4_deleg_xgrade_none_ext(open, dp);
6103 if (status == 0 && open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS)
6104 open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
6106 * To finish the open response, we just need to set the rflags.
6108 open->op_rflags = NFS4_OPEN_RESULT_LOCKTYPE_POSIX;
6109 if (nfsd4_has_session(&resp->cstate))
6110 open->op_rflags |= NFS4_OPEN_RESULT_MAY_NOTIFY_LOCK;
6111 else if (!(open->op_openowner->oo_flags & NFS4_OO_CONFIRMED))
6112 open->op_rflags |= NFS4_OPEN_RESULT_CONFIRM;
6115 nfs4_put_stid(&dp->dl_stid);
6117 nfs4_put_stid(&stp->st_stid);
6122 void nfsd4_cleanup_open_state(struct nfsd4_compound_state *cstate,
6123 struct nfsd4_open *open)
6125 if (open->op_openowner) {
6126 struct nfs4_stateowner *so = &open->op_openowner->oo_owner;
6128 nfsd4_cstate_assign_replay(cstate, so);
6129 nfs4_put_stateowner(so);
6132 kmem_cache_free(file_slab, open->op_file);
6134 nfs4_put_stid(&open->op_stp->st_stid);
6135 if (open->op_odstate)
6136 kmem_cache_free(odstate_slab, open->op_odstate);
6140 nfsd4_renew(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
6141 union nfsd4_op_u *u)
6143 clientid_t *clid = &u->renew;
6144 struct nfs4_client *clp;
6146 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
6148 trace_nfsd_clid_renew(clid);
6149 status = set_client(clid, cstate, nn);
6153 if (!list_empty(&clp->cl_delegations)
6154 && clp->cl_cb_state != NFSD4_CB_UP)
6155 return nfserr_cb_path_down;
6160 nfsd4_end_grace(struct nfsd_net *nn)
6162 /* do nothing if grace period already ended */
6163 if (nn->grace_ended)
6166 trace_nfsd_grace_complete(nn);
6167 nn->grace_ended = true;
6169 * If the server goes down again right now, an NFSv4
6170 * client will still be allowed to reclaim after it comes back up,
6171 * even if it hasn't yet had a chance to reclaim state this time.
6174 nfsd4_record_grace_done(nn);
6176 * At this point, NFSv4 clients can still reclaim. But if the
6177 * server crashes, any that have not yet reclaimed will be out
6178 * of luck on the next boot.
6180 * (NFSv4.1+ clients are considered to have reclaimed once they
6181 * call RECLAIM_COMPLETE. NFSv4.0 clients are considered to
6182 * have reclaimed after their first OPEN.)
6184 locks_end_grace(&nn->nfsd4_manager);
6186 * At this point, and once lockd and/or any other containers
6187 * exit their grace period, further reclaims will fail and
6188 * regular locking can resume.
6193 * If we've waited a lease period but there are still clients trying to
6194 * reclaim, wait a little longer to give them a chance to finish.
6196 static bool clients_still_reclaiming(struct nfsd_net *nn)
6198 time64_t double_grace_period_end = nn->boot_time +
6199 2 * nn->nfsd4_lease;
6201 if (nn->track_reclaim_completes &&
6202 atomic_read(&nn->nr_reclaim_complete) ==
6203 nn->reclaim_str_hashtbl_size)
6205 if (!nn->somebody_reclaimed)
6207 nn->somebody_reclaimed = false;
6209 * If we've given them *two* lease times to reclaim, and they're
6210 * still not done, give up:
6212 if (ktime_get_boottime_seconds() > double_grace_period_end)
6217 struct laundry_time {
6222 static bool state_expired(struct laundry_time *lt, time64_t last_refresh)
6224 time64_t time_remaining;
6226 if (last_refresh < lt->cutoff)
6228 time_remaining = last_refresh - lt->cutoff;
6229 lt->new_timeo = min(lt->new_timeo, time_remaining);
6233 #ifdef CONFIG_NFSD_V4_2_INTER_SSC
6234 void nfsd4_ssc_init_umount_work(struct nfsd_net *nn)
6236 spin_lock_init(&nn->nfsd_ssc_lock);
6237 INIT_LIST_HEAD(&nn->nfsd_ssc_mount_list);
6238 init_waitqueue_head(&nn->nfsd_ssc_waitq);
6240 EXPORT_SYMBOL_GPL(nfsd4_ssc_init_umount_work);
6243 * This is called when nfsd is being shutdown, after all inter_ssc
6244 * cleanup were done, to destroy the ssc delayed unmount list.
6246 static void nfsd4_ssc_shutdown_umount(struct nfsd_net *nn)
6248 struct nfsd4_ssc_umount_item *ni = NULL;
6249 struct nfsd4_ssc_umount_item *tmp;
6251 spin_lock(&nn->nfsd_ssc_lock);
6252 list_for_each_entry_safe(ni, tmp, &nn->nfsd_ssc_mount_list, nsui_list) {
6253 list_del(&ni->nsui_list);
6254 spin_unlock(&nn->nfsd_ssc_lock);
6255 mntput(ni->nsui_vfsmount);
6257 spin_lock(&nn->nfsd_ssc_lock);
6259 spin_unlock(&nn->nfsd_ssc_lock);
6262 static void nfsd4_ssc_expire_umount(struct nfsd_net *nn)
6264 bool do_wakeup = false;
6265 struct nfsd4_ssc_umount_item *ni = NULL;
6266 struct nfsd4_ssc_umount_item *tmp;
6268 spin_lock(&nn->nfsd_ssc_lock);
6269 list_for_each_entry_safe(ni, tmp, &nn->nfsd_ssc_mount_list, nsui_list) {
6270 if (time_after(jiffies, ni->nsui_expire)) {
6271 if (refcount_read(&ni->nsui_refcnt) > 1)
6274 /* mark being unmount */
6275 ni->nsui_busy = true;
6276 spin_unlock(&nn->nfsd_ssc_lock);
6277 mntput(ni->nsui_vfsmount);
6278 spin_lock(&nn->nfsd_ssc_lock);
6280 /* waiters need to start from begin of list */
6281 list_del(&ni->nsui_list);
6284 /* wakeup ssc_connect waiters */
6291 wake_up_all(&nn->nfsd_ssc_waitq);
6292 spin_unlock(&nn->nfsd_ssc_lock);
6296 /* Check if any lock belonging to this lockowner has any blockers */
6298 nfs4_lockowner_has_blockers(struct nfs4_lockowner *lo)
6300 struct file_lock_context *ctx;
6301 struct nfs4_ol_stateid *stp;
6302 struct nfs4_file *nf;
6304 list_for_each_entry(stp, &lo->lo_owner.so_stateids, st_perstateowner) {
6305 nf = stp->st_stid.sc_file;
6306 ctx = locks_inode_context(nf->fi_inode);
6309 if (locks_owner_has_blockers(ctx, lo))
6316 nfs4_anylock_blockers(struct nfs4_client *clp)
6319 struct nfs4_stateowner *so;
6320 struct nfs4_lockowner *lo;
6322 if (atomic_read(&clp->cl_delegs_in_recall))
6324 spin_lock(&clp->cl_lock);
6325 for (i = 0; i < OWNER_HASH_SIZE; i++) {
6326 list_for_each_entry(so, &clp->cl_ownerstr_hashtbl[i],
6328 if (so->so_is_open_owner)
6331 if (nfs4_lockowner_has_blockers(lo)) {
6332 spin_unlock(&clp->cl_lock);
6337 spin_unlock(&clp->cl_lock);
6342 nfs4_get_client_reaplist(struct nfsd_net *nn, struct list_head *reaplist,
6343 struct laundry_time *lt)
6345 unsigned int maxreap, reapcnt = 0;
6346 struct list_head *pos, *next;
6347 struct nfs4_client *clp;
6349 maxreap = (atomic_read(&nn->nfs4_client_count) >= nn->nfs4_max_clients) ?
6350 NFSD_CLIENT_MAX_TRIM_PER_RUN : 0;
6351 INIT_LIST_HEAD(reaplist);
6352 spin_lock(&nn->client_lock);
6353 list_for_each_safe(pos, next, &nn->client_lru) {
6354 clp = list_entry(pos, struct nfs4_client, cl_lru);
6355 if (clp->cl_state == NFSD4_EXPIRABLE)
6357 if (!state_expired(lt, clp->cl_time))
6359 if (!atomic_read(&clp->cl_rpc_users)) {
6360 if (clp->cl_state == NFSD4_ACTIVE)
6361 atomic_inc(&nn->nfsd_courtesy_clients);
6362 clp->cl_state = NFSD4_COURTESY;
6364 if (!client_has_state(clp))
6366 if (!nfs4_anylock_blockers(clp))
6367 if (reapcnt >= maxreap)
6370 if (!mark_client_expired_locked(clp)) {
6371 list_add(&clp->cl_lru, reaplist);
6375 spin_unlock(&nn->client_lock);
6379 nfs4_get_courtesy_client_reaplist(struct nfsd_net *nn,
6380 struct list_head *reaplist)
6382 unsigned int maxreap = 0, reapcnt = 0;
6383 struct list_head *pos, *next;
6384 struct nfs4_client *clp;
6386 maxreap = NFSD_CLIENT_MAX_TRIM_PER_RUN;
6387 INIT_LIST_HEAD(reaplist);
6389 spin_lock(&nn->client_lock);
6390 list_for_each_safe(pos, next, &nn->client_lru) {
6391 clp = list_entry(pos, struct nfs4_client, cl_lru);
6392 if (clp->cl_state == NFSD4_ACTIVE)
6394 if (reapcnt >= maxreap)
6396 if (!mark_client_expired_locked(clp)) {
6397 list_add(&clp->cl_lru, reaplist);
6401 spin_unlock(&nn->client_lock);
6405 nfs4_process_client_reaplist(struct list_head *reaplist)
6407 struct list_head *pos, *next;
6408 struct nfs4_client *clp;
6410 list_for_each_safe(pos, next, reaplist) {
6411 clp = list_entry(pos, struct nfs4_client, cl_lru);
6412 trace_nfsd_clid_purged(&clp->cl_clientid);
6413 list_del_init(&clp->cl_lru);
6418 static void nfs40_clean_admin_revoked(struct nfsd_net *nn,
6419 struct laundry_time *lt)
6421 struct nfs4_client *clp;
6423 spin_lock(&nn->client_lock);
6424 if (nn->nfs40_last_revoke == 0 ||
6425 nn->nfs40_last_revoke > lt->cutoff) {
6426 spin_unlock(&nn->client_lock);
6429 nn->nfs40_last_revoke = 0;
6432 list_for_each_entry(clp, &nn->client_lru, cl_lru) {
6433 unsigned long id, tmp;
6434 struct nfs4_stid *stid;
6436 if (atomic_read(&clp->cl_admin_revoked) == 0)
6439 spin_lock(&clp->cl_lock);
6440 idr_for_each_entry_ul(&clp->cl_stateids, stid, tmp, id)
6441 if (stid->sc_status & SC_STATUS_ADMIN_REVOKED) {
6442 refcount_inc(&stid->sc_count);
6443 spin_unlock(&nn->client_lock);
6444 /* this function drops ->cl_lock */
6445 nfsd4_drop_revoked_stid(stid);
6446 nfs4_put_stid(stid);
6447 spin_lock(&nn->client_lock);
6450 spin_unlock(&clp->cl_lock);
6452 spin_unlock(&nn->client_lock);
6456 nfs4_laundromat(struct nfsd_net *nn)
6458 struct nfs4_openowner *oo;
6459 struct nfs4_delegation *dp;
6460 struct nfs4_ol_stateid *stp;
6461 struct nfsd4_blocked_lock *nbl;
6462 struct list_head *pos, *next, reaplist;
6463 struct laundry_time lt = {
6464 .cutoff = ktime_get_boottime_seconds() - nn->nfsd4_lease,
6465 .new_timeo = nn->nfsd4_lease
6467 struct nfs4_cpntf_state *cps;
6468 copy_stateid_t *cps_t;
6471 if (clients_still_reclaiming(nn)) {
6475 nfsd4_end_grace(nn);
6477 spin_lock(&nn->s2s_cp_lock);
6478 idr_for_each_entry(&nn->s2s_cp_stateids, cps_t, i) {
6479 cps = container_of(cps_t, struct nfs4_cpntf_state, cp_stateid);
6480 if (cps->cp_stateid.cs_type == NFS4_COPYNOTIFY_STID &&
6481 state_expired(<, cps->cpntf_time))
6482 _free_cpntf_state_locked(nn, cps);
6484 spin_unlock(&nn->s2s_cp_lock);
6485 nfs4_get_client_reaplist(nn, &reaplist, <);
6486 nfs4_process_client_reaplist(&reaplist);
6488 nfs40_clean_admin_revoked(nn, <);
6490 spin_lock(&state_lock);
6491 list_for_each_safe(pos, next, &nn->del_recall_lru) {
6492 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
6493 if (!state_expired(<, dp->dl_time))
6495 unhash_delegation_locked(dp, SC_STATUS_REVOKED);
6496 list_add(&dp->dl_recall_lru, &reaplist);
6498 spin_unlock(&state_lock);
6499 while (!list_empty(&reaplist)) {
6500 dp = list_first_entry(&reaplist, struct nfs4_delegation,
6502 list_del_init(&dp->dl_recall_lru);
6503 revoke_delegation(dp);
6506 spin_lock(&nn->client_lock);
6507 while (!list_empty(&nn->close_lru)) {
6508 oo = list_first_entry(&nn->close_lru, struct nfs4_openowner,
6510 if (!state_expired(<, oo->oo_time))
6512 list_del_init(&oo->oo_close_lru);
6513 stp = oo->oo_last_closed_stid;
6514 oo->oo_last_closed_stid = NULL;
6515 spin_unlock(&nn->client_lock);
6516 nfs4_put_stid(&stp->st_stid);
6517 spin_lock(&nn->client_lock);
6519 spin_unlock(&nn->client_lock);
6522 * It's possible for a client to try and acquire an already held lock
6523 * that is being held for a long time, and then lose interest in it.
6524 * So, we clean out any un-revisited request after a lease period
6525 * under the assumption that the client is no longer interested.
6527 * RFC5661, sec. 9.6 states that the client must not rely on getting
6528 * notifications and must continue to poll for locks, even when the
6529 * server supports them. Thus this shouldn't lead to clients blocking
6530 * indefinitely once the lock does become free.
6532 BUG_ON(!list_empty(&reaplist));
6533 spin_lock(&nn->blocked_locks_lock);
6534 while (!list_empty(&nn->blocked_locks_lru)) {
6535 nbl = list_first_entry(&nn->blocked_locks_lru,
6536 struct nfsd4_blocked_lock, nbl_lru);
6537 if (!state_expired(<, nbl->nbl_time))
6539 list_move(&nbl->nbl_lru, &reaplist);
6540 list_del_init(&nbl->nbl_list);
6542 spin_unlock(&nn->blocked_locks_lock);
6544 while (!list_empty(&reaplist)) {
6545 nbl = list_first_entry(&reaplist,
6546 struct nfsd4_blocked_lock, nbl_lru);
6547 list_del_init(&nbl->nbl_lru);
6548 free_blocked_lock(nbl);
6550 #ifdef CONFIG_NFSD_V4_2_INTER_SSC
6551 /* service the server-to-server copy delayed unmount list */
6552 nfsd4_ssc_expire_umount(nn);
6554 if (atomic_long_read(&num_delegations) >= max_delegations)
6557 return max_t(time64_t, lt.new_timeo, NFSD_LAUNDROMAT_MINTIMEOUT);
6560 static void laundromat_main(struct work_struct *);
6563 laundromat_main(struct work_struct *laundry)
6566 struct delayed_work *dwork = to_delayed_work(laundry);
6567 struct nfsd_net *nn = container_of(dwork, struct nfsd_net,
6570 t = nfs4_laundromat(nn);
6571 queue_delayed_work(laundry_wq, &nn->laundromat_work, t*HZ);
6575 courtesy_client_reaper(struct nfsd_net *nn)
6577 struct list_head reaplist;
6579 nfs4_get_courtesy_client_reaplist(nn, &reaplist);
6580 nfs4_process_client_reaplist(&reaplist);
6584 deleg_reaper(struct nfsd_net *nn)
6586 struct list_head *pos, *next;
6587 struct nfs4_client *clp;
6588 struct list_head cblist;
6590 INIT_LIST_HEAD(&cblist);
6591 spin_lock(&nn->client_lock);
6592 list_for_each_safe(pos, next, &nn->client_lru) {
6593 clp = list_entry(pos, struct nfs4_client, cl_lru);
6594 if (clp->cl_state != NFSD4_ACTIVE ||
6595 list_empty(&clp->cl_delegations) ||
6596 atomic_read(&clp->cl_delegs_in_recall) ||
6597 test_bit(NFSD4_CLIENT_CB_RECALL_ANY, &clp->cl_flags) ||
6598 (ktime_get_boottime_seconds() -
6599 clp->cl_ra_time < 5)) {
6602 list_add(&clp->cl_ra_cblist, &cblist);
6604 /* release in nfsd4_cb_recall_any_release */
6605 atomic_inc(&clp->cl_rpc_users);
6606 set_bit(NFSD4_CLIENT_CB_RECALL_ANY, &clp->cl_flags);
6607 clp->cl_ra_time = ktime_get_boottime_seconds();
6609 spin_unlock(&nn->client_lock);
6611 while (!list_empty(&cblist)) {
6612 clp = list_first_entry(&cblist, struct nfs4_client,
6614 list_del_init(&clp->cl_ra_cblist);
6615 clp->cl_ra->ra_keep = 0;
6616 clp->cl_ra->ra_bmval[0] = BIT(RCA4_TYPE_MASK_RDATA_DLG);
6617 clp->cl_ra->ra_bmval[0] = BIT(RCA4_TYPE_MASK_RDATA_DLG) |
6618 BIT(RCA4_TYPE_MASK_WDATA_DLG);
6619 trace_nfsd_cb_recall_any(clp->cl_ra);
6620 nfsd4_run_cb(&clp->cl_ra->ra_cb);
6625 nfsd4_state_shrinker_worker(struct work_struct *work)
6627 struct nfsd_net *nn = container_of(work, struct nfsd_net,
6628 nfsd_shrinker_work);
6630 courtesy_client_reaper(nn);
6634 static inline __be32 nfs4_check_fh(struct svc_fh *fhp, struct nfs4_stid *stp)
6636 if (!fh_match(&fhp->fh_handle, &stp->sc_file->fi_fhandle))
6637 return nfserr_bad_stateid;
6642 __be32 nfs4_check_openmode(struct nfs4_ol_stateid *stp, int flags)
6644 __be32 status = nfserr_openmode;
6646 /* For lock stateid's, we test the parent open, not the lock: */
6647 if (stp->st_openstp)
6648 stp = stp->st_openstp;
6649 if ((flags & WR_STATE) && !access_permit_write(stp))
6651 if ((flags & RD_STATE) && !access_permit_read(stp))
6658 static inline __be32
6659 check_special_stateids(struct net *net, svc_fh *current_fh, stateid_t *stateid, int flags)
6661 if (ONE_STATEID(stateid) && (flags & RD_STATE))
6663 else if (opens_in_grace(net)) {
6664 /* Answer in remaining cases depends on existence of
6665 * conflicting state; so we must wait out the grace period. */
6666 return nfserr_grace;
6667 } else if (flags & WR_STATE)
6668 return nfs4_share_conflict(current_fh,
6669 NFS4_SHARE_DENY_WRITE);
6670 else /* (flags & RD_STATE) && ZERO_STATEID(stateid) */
6671 return nfs4_share_conflict(current_fh,
6672 NFS4_SHARE_DENY_READ);
6675 static __be32 check_stateid_generation(stateid_t *in, stateid_t *ref, bool has_session)
6678 * When sessions are used the stateid generation number is ignored
6681 if (has_session && in->si_generation == 0)
6684 if (in->si_generation == ref->si_generation)
6687 /* If the client sends us a stateid from the future, it's buggy: */
6688 if (nfsd4_stateid_generation_after(in, ref))
6689 return nfserr_bad_stateid;
6691 * However, we could see a stateid from the past, even from a
6692 * non-buggy client. For example, if the client sends a lock
6693 * while some IO is outstanding, the lock may bump si_generation
6694 * while the IO is still in flight. The client could avoid that
6695 * situation by waiting for responses on all the IO requests,
6696 * but better performance may result in retrying IO that
6697 * receives an old_stateid error if requests are rarely
6698 * reordered in flight:
6700 return nfserr_old_stateid;
6703 static __be32 nfsd4_stid_check_stateid_generation(stateid_t *in, struct nfs4_stid *s, bool has_session)
6707 spin_lock(&s->sc_lock);
6708 ret = nfsd4_verify_open_stid(s);
6710 ret = check_stateid_generation(in, &s->sc_stateid, has_session);
6711 spin_unlock(&s->sc_lock);
6712 if (ret == nfserr_admin_revoked)
6713 nfsd40_drop_revoked_stid(s->sc_client,
6718 static __be32 nfsd4_check_openowner_confirmed(struct nfs4_ol_stateid *ols)
6720 if (ols->st_stateowner->so_is_open_owner &&
6721 !(openowner(ols->st_stateowner)->oo_flags & NFS4_OO_CONFIRMED))
6722 return nfserr_bad_stateid;
6726 static __be32 nfsd4_validate_stateid(struct nfs4_client *cl, stateid_t *stateid)
6728 struct nfs4_stid *s;
6729 __be32 status = nfserr_bad_stateid;
6731 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid) ||
6732 CLOSE_STATEID(stateid))
6734 spin_lock(&cl->cl_lock);
6735 s = find_stateid_locked(cl, stateid);
6738 status = nfsd4_stid_check_stateid_generation(stateid, s, 1);
6741 status = nfsd4_verify_open_stid(s);
6745 switch (s->sc_type) {
6751 status = nfsd4_check_openowner_confirmed(openlockstateid(s));
6754 printk("unknown stateid type %x\n", s->sc_type);
6755 status = nfserr_bad_stateid;
6758 spin_unlock(&cl->cl_lock);
6759 if (status == nfserr_admin_revoked)
6760 nfsd40_drop_revoked_stid(cl, stateid);
6765 nfsd4_lookup_stateid(struct nfsd4_compound_state *cstate,
6767 unsigned short typemask, unsigned short statusmask,
6768 struct nfs4_stid **s, struct nfsd_net *nn)
6771 struct nfs4_stid *stid;
6772 bool return_revoked = false;
6775 * only return revoked delegations if explicitly asked.
6776 * otherwise we report revoked or bad_stateid status.
6778 if (statusmask & SC_STATUS_REVOKED)
6779 return_revoked = true;
6780 if (typemask & SC_TYPE_DELEG)
6781 /* Always allow REVOKED for DELEG so we can
6782 * retturn the appropriate error.
6784 statusmask |= SC_STATUS_REVOKED;
6786 statusmask |= SC_STATUS_ADMIN_REVOKED;
6788 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid) ||
6789 CLOSE_STATEID(stateid))
6790 return nfserr_bad_stateid;
6791 status = set_client(&stateid->si_opaque.so_clid, cstate, nn);
6792 if (status == nfserr_stale_clientid) {
6793 if (cstate->session)
6794 return nfserr_bad_stateid;
6795 return nfserr_stale_stateid;
6799 stid = find_stateid_by_type(cstate->clp, stateid, typemask, statusmask);
6801 return nfserr_bad_stateid;
6802 if ((stid->sc_status & SC_STATUS_REVOKED) && !return_revoked) {
6803 nfs4_put_stid(stid);
6804 return nfserr_deleg_revoked;
6806 if (stid->sc_status & SC_STATUS_ADMIN_REVOKED) {
6807 nfsd40_drop_revoked_stid(cstate->clp, stateid);
6808 nfs4_put_stid(stid);
6809 return nfserr_admin_revoked;
6815 static struct nfsd_file *
6816 nfs4_find_file(struct nfs4_stid *s, int flags)
6818 struct nfsd_file *ret = NULL;
6820 if (!s || s->sc_status)
6823 switch (s->sc_type) {
6825 spin_lock(&s->sc_file->fi_lock);
6826 ret = nfsd_file_get(s->sc_file->fi_deleg_file);
6827 spin_unlock(&s->sc_file->fi_lock);
6831 if (flags & RD_STATE)
6832 ret = find_readable_file(s->sc_file);
6834 ret = find_writeable_file(s->sc_file);
6841 nfs4_check_olstateid(struct nfs4_ol_stateid *ols, int flags)
6845 status = nfsd4_check_openowner_confirmed(ols);
6848 return nfs4_check_openmode(ols, flags);
6852 nfs4_check_file(struct svc_rqst *rqstp, struct svc_fh *fhp, struct nfs4_stid *s,
6853 struct nfsd_file **nfp, int flags)
6855 int acc = (flags & RD_STATE) ? NFSD_MAY_READ : NFSD_MAY_WRITE;
6856 struct nfsd_file *nf;
6859 nf = nfs4_find_file(s, flags);
6861 status = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
6862 acc | NFSD_MAY_OWNER_OVERRIDE);
6868 status = nfsd_file_acquire(rqstp, fhp, acc, &nf);
6877 _free_cpntf_state_locked(struct nfsd_net *nn, struct nfs4_cpntf_state *cps)
6879 WARN_ON_ONCE(cps->cp_stateid.cs_type != NFS4_COPYNOTIFY_STID);
6880 if (!refcount_dec_and_test(&cps->cp_stateid.cs_count))
6882 list_del(&cps->cp_list);
6883 idr_remove(&nn->s2s_cp_stateids,
6884 cps->cp_stateid.cs_stid.si_opaque.so_id);
6888 * A READ from an inter server to server COPY will have a
6889 * copy stateid. Look up the copy notify stateid from the
6890 * idr structure and take a reference on it.
6892 __be32 manage_cpntf_state(struct nfsd_net *nn, stateid_t *st,
6893 struct nfs4_client *clp,
6894 struct nfs4_cpntf_state **cps)
6896 copy_stateid_t *cps_t;
6897 struct nfs4_cpntf_state *state = NULL;
6899 if (st->si_opaque.so_clid.cl_id != nn->s2s_cp_cl_id)
6900 return nfserr_bad_stateid;
6901 spin_lock(&nn->s2s_cp_lock);
6902 cps_t = idr_find(&nn->s2s_cp_stateids, st->si_opaque.so_id);
6904 state = container_of(cps_t, struct nfs4_cpntf_state,
6906 if (state->cp_stateid.cs_type != NFS4_COPYNOTIFY_STID) {
6911 refcount_inc(&state->cp_stateid.cs_count);
6913 _free_cpntf_state_locked(nn, state);
6916 spin_unlock(&nn->s2s_cp_lock);
6918 return nfserr_bad_stateid;
6924 static __be32 find_cpntf_state(struct nfsd_net *nn, stateid_t *st,
6925 struct nfs4_stid **stid)
6928 struct nfs4_cpntf_state *cps = NULL;
6929 struct nfs4_client *found;
6931 status = manage_cpntf_state(nn, st, NULL, &cps);
6935 cps->cpntf_time = ktime_get_boottime_seconds();
6937 status = nfserr_expired;
6938 found = lookup_clientid(&cps->cp_p_clid, true, nn);
6942 *stid = find_stateid_by_type(found, &cps->cp_p_stateid,
6943 SC_TYPE_DELEG|SC_TYPE_OPEN|SC_TYPE_LOCK,
6948 status = nfserr_bad_stateid;
6950 put_client_renew(found);
6952 nfs4_put_cpntf_state(nn, cps);
6956 void nfs4_put_cpntf_state(struct nfsd_net *nn, struct nfs4_cpntf_state *cps)
6958 spin_lock(&nn->s2s_cp_lock);
6959 _free_cpntf_state_locked(nn, cps);
6960 spin_unlock(&nn->s2s_cp_lock);
6964 * nfs4_preprocess_stateid_op - find and prep stateid for an operation
6965 * @rqstp: incoming request from client
6966 * @cstate: current compound state
6967 * @fhp: filehandle associated with requested stateid
6968 * @stateid: stateid (provided by client)
6969 * @flags: flags describing type of operation to be done
6970 * @nfp: optional nfsd_file return pointer (may be NULL)
6971 * @cstid: optional returned nfs4_stid pointer (may be NULL)
6973 * Given info from the client, look up a nfs4_stid for the operation. On
6974 * success, it returns a reference to the nfs4_stid and/or the nfsd_file
6975 * associated with it.
6978 nfs4_preprocess_stateid_op(struct svc_rqst *rqstp,
6979 struct nfsd4_compound_state *cstate, struct svc_fh *fhp,
6980 stateid_t *stateid, int flags, struct nfsd_file **nfp,
6981 struct nfs4_stid **cstid)
6983 struct net *net = SVC_NET(rqstp);
6984 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
6985 struct nfs4_stid *s = NULL;
6991 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid)) {
6993 status = nfserr_bad_stateid;
6995 status = check_special_stateids(net, fhp, stateid,
7000 status = nfsd4_lookup_stateid(cstate, stateid,
7001 SC_TYPE_DELEG|SC_TYPE_OPEN|SC_TYPE_LOCK,
7003 if (status == nfserr_bad_stateid)
7004 status = find_cpntf_state(nn, stateid, &s);
7007 status = nfsd4_stid_check_stateid_generation(stateid, s,
7008 nfsd4_has_session(cstate));
7012 switch (s->sc_type) {
7014 status = nfs4_check_delegmode(delegstateid(s), flags);
7018 status = nfs4_check_olstateid(openlockstateid(s), flags);
7023 status = nfs4_check_fh(fhp, s);
7026 if (status == nfs_ok && nfp)
7027 status = nfs4_check_file(rqstp, fhp, s, nfp, flags);
7030 if (!status && cstid)
7039 * Test if the stateid is valid
7042 nfsd4_test_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
7043 union nfsd4_op_u *u)
7045 struct nfsd4_test_stateid *test_stateid = &u->test_stateid;
7046 struct nfsd4_test_stateid_id *stateid;
7047 struct nfs4_client *cl = cstate->clp;
7049 list_for_each_entry(stateid, &test_stateid->ts_stateid_list, ts_id_list)
7050 stateid->ts_id_status =
7051 nfsd4_validate_stateid(cl, &stateid->ts_id_stateid);
7057 nfsd4_free_lock_stateid(stateid_t *stateid, struct nfs4_stid *s)
7059 struct nfs4_ol_stateid *stp = openlockstateid(s);
7062 ret = nfsd4_lock_ol_stateid(stp);
7066 ret = check_stateid_generation(stateid, &s->sc_stateid, 1);
7070 ret = nfserr_locks_held;
7071 if (check_for_locks(stp->st_stid.sc_file,
7072 lockowner(stp->st_stateowner)))
7075 release_lock_stateid(stp);
7079 mutex_unlock(&stp->st_mutex);
7086 nfsd4_free_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
7087 union nfsd4_op_u *u)
7089 struct nfsd4_free_stateid *free_stateid = &u->free_stateid;
7090 stateid_t *stateid = &free_stateid->fr_stateid;
7091 struct nfs4_stid *s;
7092 struct nfs4_delegation *dp;
7093 struct nfs4_client *cl = cstate->clp;
7094 __be32 ret = nfserr_bad_stateid;
7096 spin_lock(&cl->cl_lock);
7097 s = find_stateid_locked(cl, stateid);
7098 if (!s || s->sc_status & SC_STATUS_CLOSED)
7100 if (s->sc_status & SC_STATUS_ADMIN_REVOKED) {
7101 nfsd4_drop_revoked_stid(s);
7105 spin_lock(&s->sc_lock);
7106 switch (s->sc_type) {
7108 if (s->sc_status & SC_STATUS_REVOKED) {
7109 spin_unlock(&s->sc_lock);
7110 dp = delegstateid(s);
7111 list_del_init(&dp->dl_recall_lru);
7112 spin_unlock(&cl->cl_lock);
7117 ret = nfserr_locks_held;
7120 ret = check_stateid_generation(stateid, &s->sc_stateid, 1);
7123 ret = nfserr_locks_held;
7126 spin_unlock(&s->sc_lock);
7127 refcount_inc(&s->sc_count);
7128 spin_unlock(&cl->cl_lock);
7129 ret = nfsd4_free_lock_stateid(stateid, s);
7132 spin_unlock(&s->sc_lock);
7134 spin_unlock(&cl->cl_lock);
7142 return (type == NFS4_READW_LT || type == NFS4_READ_LT) ?
7143 RD_STATE : WR_STATE;
7146 static __be32 nfs4_seqid_op_checks(struct nfsd4_compound_state *cstate, stateid_t *stateid, u32 seqid, struct nfs4_ol_stateid *stp)
7148 struct svc_fh *current_fh = &cstate->current_fh;
7149 struct nfs4_stateowner *sop = stp->st_stateowner;
7152 status = nfsd4_check_seqid(cstate, sop, seqid);
7155 status = nfsd4_lock_ol_stateid(stp);
7156 if (status != nfs_ok)
7158 status = check_stateid_generation(stateid, &stp->st_stid.sc_stateid, nfsd4_has_session(cstate));
7159 if (status == nfs_ok)
7160 status = nfs4_check_fh(current_fh, &stp->st_stid);
7161 if (status != nfs_ok)
7162 mutex_unlock(&stp->st_mutex);
7167 * nfs4_preprocess_seqid_op - find and prep an ol_stateid for a seqid-morphing op
7168 * @cstate: compund state
7169 * @seqid: seqid (provided by client)
7170 * @stateid: stateid (provided by client)
7171 * @typemask: mask of allowable types for this operation
7172 * @statusmask: mask of allowed states: 0 or STID_CLOSED
7173 * @stpp: return pointer for the stateid found
7174 * @nn: net namespace for request
7176 * Given a stateid+seqid from a client, look up an nfs4_ol_stateid and
7177 * return it in @stpp. On a nfs_ok return, the returned stateid will
7178 * have its st_mutex locked.
7181 nfs4_preprocess_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
7183 unsigned short typemask, unsigned short statusmask,
7184 struct nfs4_ol_stateid **stpp,
7185 struct nfsd_net *nn)
7188 struct nfs4_stid *s;
7189 struct nfs4_ol_stateid *stp = NULL;
7191 trace_nfsd_preprocess(seqid, stateid);
7194 status = nfsd4_lookup_stateid(cstate, stateid,
7195 typemask, statusmask, &s, nn);
7198 stp = openlockstateid(s);
7199 nfsd4_cstate_assign_replay(cstate, stp->st_stateowner);
7201 status = nfs4_seqid_op_checks(cstate, stateid, seqid, stp);
7205 nfs4_put_stid(&stp->st_stid);
7209 static __be32 nfs4_preprocess_confirmed_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
7210 stateid_t *stateid, struct nfs4_ol_stateid **stpp, struct nfsd_net *nn)
7213 struct nfs4_openowner *oo;
7214 struct nfs4_ol_stateid *stp;
7216 status = nfs4_preprocess_seqid_op(cstate, seqid, stateid,
7217 SC_TYPE_OPEN, 0, &stp, nn);
7220 oo = openowner(stp->st_stateowner);
7221 if (!(oo->oo_flags & NFS4_OO_CONFIRMED)) {
7222 mutex_unlock(&stp->st_mutex);
7223 nfs4_put_stid(&stp->st_stid);
7224 return nfserr_bad_stateid;
7231 nfsd4_open_confirm(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
7232 union nfsd4_op_u *u)
7234 struct nfsd4_open_confirm *oc = &u->open_confirm;
7236 struct nfs4_openowner *oo;
7237 struct nfs4_ol_stateid *stp;
7238 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
7240 dprintk("NFSD: nfsd4_open_confirm on file %pd\n",
7241 cstate->current_fh.fh_dentry);
7243 status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0);
7247 status = nfs4_preprocess_seqid_op(cstate,
7248 oc->oc_seqid, &oc->oc_req_stateid,
7249 SC_TYPE_OPEN, 0, &stp, nn);
7252 oo = openowner(stp->st_stateowner);
7253 status = nfserr_bad_stateid;
7254 if (oo->oo_flags & NFS4_OO_CONFIRMED) {
7255 mutex_unlock(&stp->st_mutex);
7258 oo->oo_flags |= NFS4_OO_CONFIRMED;
7259 nfs4_inc_and_copy_stateid(&oc->oc_resp_stateid, &stp->st_stid);
7260 mutex_unlock(&stp->st_mutex);
7261 trace_nfsd_open_confirm(oc->oc_seqid, &stp->st_stid.sc_stateid);
7262 nfsd4_client_record_create(oo->oo_owner.so_client);
7265 nfs4_put_stid(&stp->st_stid);
7267 nfsd4_bump_seqid(cstate, status);
7271 static inline void nfs4_stateid_downgrade_bit(struct nfs4_ol_stateid *stp, u32 access)
7273 if (!test_access(access, stp))
7275 nfs4_file_put_access(stp->st_stid.sc_file, access);
7276 clear_access(access, stp);
7279 static inline void nfs4_stateid_downgrade(struct nfs4_ol_stateid *stp, u32 to_access)
7281 switch (to_access) {
7282 case NFS4_SHARE_ACCESS_READ:
7283 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_WRITE);
7284 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
7286 case NFS4_SHARE_ACCESS_WRITE:
7287 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_READ);
7288 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
7290 case NFS4_SHARE_ACCESS_BOTH:
7298 nfsd4_open_downgrade(struct svc_rqst *rqstp,
7299 struct nfsd4_compound_state *cstate, union nfsd4_op_u *u)
7301 struct nfsd4_open_downgrade *od = &u->open_downgrade;
7303 struct nfs4_ol_stateid *stp;
7304 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
7306 dprintk("NFSD: nfsd4_open_downgrade on file %pd\n",
7307 cstate->current_fh.fh_dentry);
7309 /* We don't yet support WANT bits: */
7310 if (od->od_deleg_want)
7311 dprintk("NFSD: %s: od_deleg_want=0x%x ignored\n", __func__,
7314 status = nfs4_preprocess_confirmed_seqid_op(cstate, od->od_seqid,
7315 &od->od_stateid, &stp, nn);
7318 status = nfserr_inval;
7319 if (!test_access(od->od_share_access, stp)) {
7320 dprintk("NFSD: access not a subset of current bitmap: 0x%hhx, input access=%08x\n",
7321 stp->st_access_bmap, od->od_share_access);
7324 if (!test_deny(od->od_share_deny, stp)) {
7325 dprintk("NFSD: deny not a subset of current bitmap: 0x%hhx, input deny=%08x\n",
7326 stp->st_deny_bmap, od->od_share_deny);
7329 nfs4_stateid_downgrade(stp, od->od_share_access);
7330 reset_union_bmap_deny(od->od_share_deny, stp);
7331 nfs4_inc_and_copy_stateid(&od->od_stateid, &stp->st_stid);
7334 mutex_unlock(&stp->st_mutex);
7335 nfs4_put_stid(&stp->st_stid);
7337 nfsd4_bump_seqid(cstate, status);
7341 static void nfsd4_close_open_stateid(struct nfs4_ol_stateid *s)
7343 struct nfs4_client *clp = s->st_stid.sc_client;
7345 LIST_HEAD(reaplist);
7346 struct nfs4_ol_stateid *stp;
7348 spin_lock(&clp->cl_lock);
7349 unhashed = unhash_open_stateid(s, &reaplist);
7351 if (clp->cl_minorversion) {
7353 put_ol_stateid_locked(s, &reaplist);
7354 spin_unlock(&clp->cl_lock);
7355 list_for_each_entry(stp, &reaplist, st_locks)
7356 nfs4_free_cpntf_statelist(clp->net, &stp->st_stid);
7357 free_ol_stateid_reaplist(&reaplist);
7359 spin_unlock(&clp->cl_lock);
7360 free_ol_stateid_reaplist(&reaplist);
7362 move_to_close_lru(s, clp->net);
7367 * nfs4_unlock_state() called after encode
7370 nfsd4_close(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
7371 union nfsd4_op_u *u)
7373 struct nfsd4_close *close = &u->close;
7375 struct nfs4_ol_stateid *stp;
7376 struct net *net = SVC_NET(rqstp);
7377 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
7379 dprintk("NFSD: nfsd4_close on file %pd\n",
7380 cstate->current_fh.fh_dentry);
7382 status = nfs4_preprocess_seqid_op(cstate, close->cl_seqid,
7384 SC_TYPE_OPEN, SC_STATUS_CLOSED,
7386 nfsd4_bump_seqid(cstate, status);
7390 spin_lock(&stp->st_stid.sc_client->cl_lock);
7391 stp->st_stid.sc_status |= SC_STATUS_CLOSED;
7392 spin_unlock(&stp->st_stid.sc_client->cl_lock);
7395 * Technically we don't _really_ have to increment or copy it, since
7396 * it should just be gone after this operation and we clobber the
7397 * copied value below, but we continue to do so here just to ensure
7398 * that racing ops see that there was a state change.
7400 nfs4_inc_and_copy_stateid(&close->cl_stateid, &stp->st_stid);
7402 nfsd4_close_open_stateid(stp);
7403 mutex_unlock(&stp->st_mutex);
7405 /* v4.1+ suggests that we send a special stateid in here, since the
7406 * clients should just ignore this anyway. Since this is not useful
7407 * for v4.0 clients either, we set it to the special close_stateid
7410 * See RFC5661 section 18.2.4, and RFC7530 section 16.2.5
7412 memcpy(&close->cl_stateid, &close_stateid, sizeof(close->cl_stateid));
7414 /* put reference from nfs4_preprocess_seqid_op */
7415 nfs4_put_stid(&stp->st_stid);
7421 nfsd4_delegreturn(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
7422 union nfsd4_op_u *u)
7424 struct nfsd4_delegreturn *dr = &u->delegreturn;
7425 struct nfs4_delegation *dp;
7426 stateid_t *stateid = &dr->dr_stateid;
7427 struct nfs4_stid *s;
7429 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
7431 if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
7434 status = nfsd4_lookup_stateid(cstate, stateid, SC_TYPE_DELEG, 0, &s, nn);
7437 dp = delegstateid(s);
7438 status = nfsd4_stid_check_stateid_generation(stateid, &dp->dl_stid, nfsd4_has_session(cstate));
7442 trace_nfsd_deleg_return(stateid);
7443 wake_up_var(d_inode(cstate->current_fh.fh_dentry));
7444 destroy_delegation(dp);
7446 nfs4_put_stid(&dp->dl_stid);
7451 /* last octet in a range */
7453 last_byte_offset(u64 start, u64 len)
7459 return end > start ? end - 1: NFS4_MAX_UINT64;
7463 * TODO: Linux file offsets are _signed_ 64-bit quantities, which means that
7464 * we can't properly handle lock requests that go beyond the (2^63 - 1)-th
7465 * byte, because of sign extension problems. Since NFSv4 calls for 64-bit
7466 * locking, this prevents us from being completely protocol-compliant. The
7467 * real solution to this problem is to start using unsigned file offsets in
7468 * the VFS, but this is a very deep change!
7471 nfs4_transform_lock_offset(struct file_lock *lock)
7473 if (lock->fl_start < 0)
7474 lock->fl_start = OFFSET_MAX;
7475 if (lock->fl_end < 0)
7476 lock->fl_end = OFFSET_MAX;
7480 nfsd4_lm_get_owner(fl_owner_t owner)
7482 struct nfs4_lockowner *lo = (struct nfs4_lockowner *)owner;
7484 nfs4_get_stateowner(&lo->lo_owner);
7489 nfsd4_lm_put_owner(fl_owner_t owner)
7491 struct nfs4_lockowner *lo = (struct nfs4_lockowner *)owner;
7494 nfs4_put_stateowner(&lo->lo_owner);
7497 /* return pointer to struct nfs4_client if client is expirable */
7499 nfsd4_lm_lock_expirable(struct file_lock *cfl)
7501 struct nfs4_lockowner *lo = (struct nfs4_lockowner *) cfl->c.flc_owner;
7502 struct nfs4_client *clp = lo->lo_owner.so_client;
7503 struct nfsd_net *nn;
7505 if (try_to_expire_client(clp)) {
7506 nn = net_generic(clp->net, nfsd_net_id);
7507 mod_delayed_work(laundry_wq, &nn->laundromat_work, 0);
7513 /* schedule laundromat to run immediately and wait for it to complete */
7515 nfsd4_lm_expire_lock(void)
7517 flush_workqueue(laundry_wq);
7521 nfsd4_lm_notify(struct file_lock *fl)
7523 struct nfs4_lockowner *lo = (struct nfs4_lockowner *) fl->c.flc_owner;
7524 struct net *net = lo->lo_owner.so_client->net;
7525 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
7526 struct nfsd4_blocked_lock *nbl = container_of(fl,
7527 struct nfsd4_blocked_lock, nbl_lock);
7530 /* An empty list means that something else is going to be using it */
7531 spin_lock(&nn->blocked_locks_lock);
7532 if (!list_empty(&nbl->nbl_list)) {
7533 list_del_init(&nbl->nbl_list);
7534 list_del_init(&nbl->nbl_lru);
7537 spin_unlock(&nn->blocked_locks_lock);
7540 trace_nfsd_cb_notify_lock(lo, nbl);
7541 nfsd4_run_cb(&nbl->nbl_cb);
7545 static const struct lock_manager_operations nfsd_posix_mng_ops = {
7546 .lm_mod_owner = THIS_MODULE,
7547 .lm_notify = nfsd4_lm_notify,
7548 .lm_get_owner = nfsd4_lm_get_owner,
7549 .lm_put_owner = nfsd4_lm_put_owner,
7550 .lm_lock_expirable = nfsd4_lm_lock_expirable,
7551 .lm_expire_lock = nfsd4_lm_expire_lock,
7555 nfs4_set_lock_denied(struct file_lock *fl, struct nfsd4_lock_denied *deny)
7557 struct nfs4_lockowner *lo;
7559 if (fl->fl_lmops == &nfsd_posix_mng_ops) {
7560 lo = (struct nfs4_lockowner *) fl->c.flc_owner;
7561 xdr_netobj_dup(&deny->ld_owner, &lo->lo_owner.so_owner,
7563 if (!deny->ld_owner.data)
7564 /* We just don't care that much */
7566 deny->ld_clientid = lo->lo_owner.so_client->cl_clientid;
7569 deny->ld_owner.len = 0;
7570 deny->ld_owner.data = NULL;
7571 deny->ld_clientid.cl_boot = 0;
7572 deny->ld_clientid.cl_id = 0;
7574 deny->ld_start = fl->fl_start;
7575 deny->ld_length = NFS4_MAX_UINT64;
7576 if (fl->fl_end != NFS4_MAX_UINT64)
7577 deny->ld_length = fl->fl_end - fl->fl_start + 1;
7578 deny->ld_type = NFS4_READ_LT;
7579 if (fl->c.flc_type != F_RDLCK)
7580 deny->ld_type = NFS4_WRITE_LT;
7583 static struct nfs4_lockowner *
7584 find_lockowner_str_locked(struct nfs4_client *clp, struct xdr_netobj *owner)
7586 unsigned int strhashval = ownerstr_hashval(owner);
7587 struct nfs4_stateowner *so;
7589 lockdep_assert_held(&clp->cl_lock);
7591 list_for_each_entry(so, &clp->cl_ownerstr_hashtbl[strhashval],
7593 if (so->so_is_open_owner)
7595 if (same_owner_str(so, owner))
7596 return lockowner(nfs4_get_stateowner(so));
7601 static struct nfs4_lockowner *
7602 find_lockowner_str(struct nfs4_client *clp, struct xdr_netobj *owner)
7604 struct nfs4_lockowner *lo;
7606 spin_lock(&clp->cl_lock);
7607 lo = find_lockowner_str_locked(clp, owner);
7608 spin_unlock(&clp->cl_lock);
7612 static void nfs4_unhash_lockowner(struct nfs4_stateowner *sop)
7614 unhash_lockowner_locked(lockowner(sop));
7617 static void nfs4_free_lockowner(struct nfs4_stateowner *sop)
7619 struct nfs4_lockowner *lo = lockowner(sop);
7621 kmem_cache_free(lockowner_slab, lo);
7624 static const struct nfs4_stateowner_operations lockowner_ops = {
7625 .so_unhash = nfs4_unhash_lockowner,
7626 .so_free = nfs4_free_lockowner,
7630 * Alloc a lock owner structure.
7631 * Called in nfsd4_lock - therefore, OPEN and OPEN_CONFIRM (if needed) has
7634 * strhashval = ownerstr_hashval
7636 static struct nfs4_lockowner *
7637 alloc_init_lock_stateowner(unsigned int strhashval, struct nfs4_client *clp,
7638 struct nfs4_ol_stateid *open_stp,
7639 struct nfsd4_lock *lock)
7641 struct nfs4_lockowner *lo, *ret;
7643 lo = alloc_stateowner(lockowner_slab, &lock->lk_new_owner, clp);
7646 INIT_LIST_HEAD(&lo->lo_blocked);
7647 INIT_LIST_HEAD(&lo->lo_owner.so_stateids);
7648 lo->lo_owner.so_is_open_owner = 0;
7649 lo->lo_owner.so_seqid = lock->lk_new_lock_seqid;
7650 lo->lo_owner.so_ops = &lockowner_ops;
7651 spin_lock(&clp->cl_lock);
7652 ret = find_lockowner_str_locked(clp, &lock->lk_new_owner);
7654 list_add(&lo->lo_owner.so_strhash,
7655 &clp->cl_ownerstr_hashtbl[strhashval]);
7658 nfs4_free_stateowner(&lo->lo_owner);
7660 spin_unlock(&clp->cl_lock);
7664 static struct nfs4_ol_stateid *
7665 find_lock_stateid(const struct nfs4_lockowner *lo,
7666 const struct nfs4_ol_stateid *ost)
7668 struct nfs4_ol_stateid *lst;
7670 lockdep_assert_held(&ost->st_stid.sc_client->cl_lock);
7672 /* If ost is not hashed, ost->st_locks will not be valid */
7673 if (!nfs4_ol_stateid_unhashed(ost))
7674 list_for_each_entry(lst, &ost->st_locks, st_locks) {
7675 if (lst->st_stateowner == &lo->lo_owner) {
7676 refcount_inc(&lst->st_stid.sc_count);
7683 static struct nfs4_ol_stateid *
7684 init_lock_stateid(struct nfs4_ol_stateid *stp, struct nfs4_lockowner *lo,
7685 struct nfs4_file *fp, struct inode *inode,
7686 struct nfs4_ol_stateid *open_stp)
7688 struct nfs4_client *clp = lo->lo_owner.so_client;
7689 struct nfs4_ol_stateid *retstp;
7691 mutex_init(&stp->st_mutex);
7692 mutex_lock_nested(&stp->st_mutex, OPEN_STATEID_MUTEX);
7694 spin_lock(&clp->cl_lock);
7695 if (nfs4_ol_stateid_unhashed(open_stp))
7697 retstp = find_lock_stateid(lo, open_stp);
7700 refcount_inc(&stp->st_stid.sc_count);
7701 stp->st_stid.sc_type = SC_TYPE_LOCK;
7702 stp->st_stateowner = nfs4_get_stateowner(&lo->lo_owner);
7704 stp->st_stid.sc_file = fp;
7705 stp->st_access_bmap = 0;
7706 stp->st_deny_bmap = open_stp->st_deny_bmap;
7707 stp->st_openstp = open_stp;
7708 spin_lock(&fp->fi_lock);
7709 list_add(&stp->st_locks, &open_stp->st_locks);
7710 list_add(&stp->st_perstateowner, &lo->lo_owner.so_stateids);
7711 list_add(&stp->st_perfile, &fp->fi_stateids);
7712 spin_unlock(&fp->fi_lock);
7713 spin_unlock(&clp->cl_lock);
7716 spin_unlock(&clp->cl_lock);
7717 if (nfsd4_lock_ol_stateid(retstp) != nfs_ok) {
7718 nfs4_put_stid(&retstp->st_stid);
7721 /* To keep mutex tracking happy */
7722 mutex_unlock(&stp->st_mutex);
7725 spin_unlock(&clp->cl_lock);
7726 mutex_unlock(&stp->st_mutex);
7730 static struct nfs4_ol_stateid *
7731 find_or_create_lock_stateid(struct nfs4_lockowner *lo, struct nfs4_file *fi,
7732 struct inode *inode, struct nfs4_ol_stateid *ost,
7735 struct nfs4_stid *ns = NULL;
7736 struct nfs4_ol_stateid *lst;
7737 struct nfs4_openowner *oo = openowner(ost->st_stateowner);
7738 struct nfs4_client *clp = oo->oo_owner.so_client;
7741 spin_lock(&clp->cl_lock);
7742 lst = find_lock_stateid(lo, ost);
7743 spin_unlock(&clp->cl_lock);
7745 if (nfsd4_lock_ol_stateid(lst) == nfs_ok)
7747 nfs4_put_stid(&lst->st_stid);
7749 ns = nfs4_alloc_stid(clp, stateid_slab, nfs4_free_lock_stateid);
7753 lst = init_lock_stateid(openlockstateid(ns), lo, fi, inode, ost);
7754 if (lst == openlockstateid(ns))
7763 check_lock_length(u64 offset, u64 length)
7765 return ((length == 0) || ((length != NFS4_MAX_UINT64) &&
7766 (length > ~offset)));
7769 static void get_lock_access(struct nfs4_ol_stateid *lock_stp, u32 access)
7771 struct nfs4_file *fp = lock_stp->st_stid.sc_file;
7773 lockdep_assert_held(&fp->fi_lock);
7775 if (test_access(access, lock_stp))
7777 __nfs4_file_get_access(fp, access);
7778 set_access(access, lock_stp);
7782 lookup_or_create_lock_state(struct nfsd4_compound_state *cstate,
7783 struct nfs4_ol_stateid *ost,
7784 struct nfsd4_lock *lock,
7785 struct nfs4_ol_stateid **plst, bool *new)
7788 struct nfs4_file *fi = ost->st_stid.sc_file;
7789 struct nfs4_openowner *oo = openowner(ost->st_stateowner);
7790 struct nfs4_client *cl = oo->oo_owner.so_client;
7791 struct inode *inode = d_inode(cstate->current_fh.fh_dentry);
7792 struct nfs4_lockowner *lo;
7793 struct nfs4_ol_stateid *lst;
7794 unsigned int strhashval;
7796 lo = find_lockowner_str(cl, &lock->lk_new_owner);
7798 strhashval = ownerstr_hashval(&lock->lk_new_owner);
7799 lo = alloc_init_lock_stateowner(strhashval, cl, ost, lock);
7801 return nfserr_jukebox;
7803 /* with an existing lockowner, seqids must be the same */
7804 status = nfserr_bad_seqid;
7805 if (!cstate->minorversion &&
7806 lock->lk_new_lock_seqid != lo->lo_owner.so_seqid)
7810 lst = find_or_create_lock_stateid(lo, fi, inode, ost, new);
7812 status = nfserr_jukebox;
7819 nfs4_put_stateowner(&lo->lo_owner);
7827 nfsd4_lock(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
7828 union nfsd4_op_u *u)
7830 struct nfsd4_lock *lock = &u->lock;
7831 struct nfs4_openowner *open_sop = NULL;
7832 struct nfs4_lockowner *lock_sop = NULL;
7833 struct nfs4_ol_stateid *lock_stp = NULL;
7834 struct nfs4_ol_stateid *open_stp = NULL;
7835 struct nfs4_file *fp;
7836 struct nfsd_file *nf = NULL;
7837 struct nfsd4_blocked_lock *nbl = NULL;
7838 struct file_lock *file_lock = NULL;
7839 struct file_lock *conflock = NULL;
7840 struct super_block *sb;
7846 unsigned int flags = FL_POSIX;
7847 struct net *net = SVC_NET(rqstp);
7848 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
7850 dprintk("NFSD: nfsd4_lock: start=%Ld length=%Ld\n",
7851 (long long) lock->lk_offset,
7852 (long long) lock->lk_length);
7854 if (check_lock_length(lock->lk_offset, lock->lk_length))
7855 return nfserr_inval;
7857 if ((status = fh_verify(rqstp, &cstate->current_fh,
7858 S_IFREG, NFSD_MAY_LOCK))) {
7859 dprintk("NFSD: nfsd4_lock: permission denied!\n");
7862 sb = cstate->current_fh.fh_dentry->d_sb;
7864 if (lock->lk_is_new) {
7865 if (nfsd4_has_session(cstate))
7866 /* See rfc 5661 18.10.3: given clientid is ignored: */
7867 memcpy(&lock->lk_new_clientid,
7868 &cstate->clp->cl_clientid,
7869 sizeof(clientid_t));
7871 /* validate and update open stateid and open seqid */
7872 status = nfs4_preprocess_confirmed_seqid_op(cstate,
7873 lock->lk_new_open_seqid,
7874 &lock->lk_new_open_stateid,
7878 mutex_unlock(&open_stp->st_mutex);
7879 open_sop = openowner(open_stp->st_stateowner);
7880 status = nfserr_bad_stateid;
7881 if (!same_clid(&open_sop->oo_owner.so_client->cl_clientid,
7882 &lock->lk_new_clientid))
7884 status = lookup_or_create_lock_state(cstate, open_stp, lock,
7887 status = nfs4_preprocess_seqid_op(cstate,
7888 lock->lk_old_lock_seqid,
7889 &lock->lk_old_lock_stateid,
7890 SC_TYPE_LOCK, 0, &lock_stp,
7895 lock_sop = lockowner(lock_stp->st_stateowner);
7897 lkflg = setlkflg(lock->lk_type);
7898 status = nfs4_check_openmode(lock_stp, lkflg);
7902 status = nfserr_grace;
7903 if (locks_in_grace(net) && !lock->lk_reclaim)
7905 status = nfserr_no_grace;
7906 if (!locks_in_grace(net) && lock->lk_reclaim)
7909 if (lock->lk_reclaim)
7910 flags |= FL_RECLAIM;
7912 fp = lock_stp->st_stid.sc_file;
7913 switch (lock->lk_type) {
7915 if (nfsd4_has_session(cstate) ||
7916 exportfs_lock_op_is_async(sb->s_export_op))
7920 spin_lock(&fp->fi_lock);
7921 nf = find_readable_file_locked(fp);
7923 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_READ);
7924 spin_unlock(&fp->fi_lock);
7927 case NFS4_WRITEW_LT:
7928 if (nfsd4_has_session(cstate) ||
7929 exportfs_lock_op_is_async(sb->s_export_op))
7933 spin_lock(&fp->fi_lock);
7934 nf = find_writeable_file_locked(fp);
7936 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_WRITE);
7937 spin_unlock(&fp->fi_lock);
7941 status = nfserr_inval;
7946 status = nfserr_openmode;
7951 * Most filesystems with their own ->lock operations will block
7952 * the nfsd thread waiting to acquire the lock. That leads to
7953 * deadlocks (we don't want every nfsd thread tied up waiting
7954 * for file locks), so don't attempt blocking lock notifications
7955 * on those filesystems:
7957 if (!exportfs_lock_op_is_async(sb->s_export_op))
7960 nbl = find_or_allocate_block(lock_sop, &fp->fi_fhandle, nn);
7962 dprintk("NFSD: %s: unable to allocate block!\n", __func__);
7963 status = nfserr_jukebox;
7967 file_lock = &nbl->nbl_lock;
7968 file_lock->c.flc_type = type;
7969 file_lock->c.flc_owner = (fl_owner_t)lockowner(nfs4_get_stateowner(&lock_sop->lo_owner));
7970 file_lock->c.flc_pid = current->tgid;
7971 file_lock->c.flc_file = nf->nf_file;
7972 file_lock->c.flc_flags = flags;
7973 file_lock->fl_lmops = &nfsd_posix_mng_ops;
7974 file_lock->fl_start = lock->lk_offset;
7975 file_lock->fl_end = last_byte_offset(lock->lk_offset, lock->lk_length);
7976 nfs4_transform_lock_offset(file_lock);
7978 conflock = locks_alloc_lock();
7980 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
7981 status = nfserr_jukebox;
7985 if (flags & FL_SLEEP) {
7986 nbl->nbl_time = ktime_get_boottime_seconds();
7987 spin_lock(&nn->blocked_locks_lock);
7988 list_add_tail(&nbl->nbl_list, &lock_sop->lo_blocked);
7989 list_add_tail(&nbl->nbl_lru, &nn->blocked_locks_lru);
7990 kref_get(&nbl->nbl_kref);
7991 spin_unlock(&nn->blocked_locks_lock);
7994 err = vfs_lock_file(nf->nf_file, F_SETLK, file_lock, conflock);
7996 case 0: /* success! */
7997 nfs4_inc_and_copy_stateid(&lock->lk_resp_stateid, &lock_stp->st_stid);
7999 if (lock->lk_reclaim)
8000 nn->somebody_reclaimed = true;
8002 case FILE_LOCK_DEFERRED:
8003 kref_put(&nbl->nbl_kref, free_nbl);
8006 case -EAGAIN: /* conflock holds conflicting lock */
8007 status = nfserr_denied;
8008 dprintk("NFSD: nfsd4_lock: conflicting lock found!\n");
8009 nfs4_set_lock_denied(conflock, &lock->lk_denied);
8012 status = nfserr_deadlock;
8015 dprintk("NFSD: nfsd4_lock: vfs_lock_file() failed! status %d\n",err);
8016 status = nfserrno(err);
8021 /* dequeue it if we queued it before */
8022 if (flags & FL_SLEEP) {
8023 spin_lock(&nn->blocked_locks_lock);
8024 if (!list_empty(&nbl->nbl_list) &&
8025 !list_empty(&nbl->nbl_lru)) {
8026 list_del_init(&nbl->nbl_list);
8027 list_del_init(&nbl->nbl_lru);
8028 kref_put(&nbl->nbl_kref, free_nbl);
8030 /* nbl can use one of lists to be linked to reaplist */
8031 spin_unlock(&nn->blocked_locks_lock);
8033 free_blocked_lock(nbl);
8038 /* Bump seqid manually if the 4.0 replay owner is openowner */
8039 if (cstate->replay_owner &&
8040 cstate->replay_owner != &lock_sop->lo_owner &&
8041 seqid_mutating_err(ntohl(status)))
8042 lock_sop->lo_owner.so_seqid++;
8045 * If this is a new, never-before-used stateid, and we are
8046 * returning an error, then just go ahead and release it.
8049 release_lock_stateid(lock_stp);
8051 mutex_unlock(&lock_stp->st_mutex);
8053 nfs4_put_stid(&lock_stp->st_stid);
8056 nfs4_put_stid(&open_stp->st_stid);
8057 nfsd4_bump_seqid(cstate, status);
8059 locks_free_lock(conflock);
8063 void nfsd4_lock_release(union nfsd4_op_u *u)
8065 struct nfsd4_lock *lock = &u->lock;
8066 struct nfsd4_lock_denied *deny = &lock->lk_denied;
8068 kfree(deny->ld_owner.data);
8072 * The NFSv4 spec allows a client to do a LOCKT without holding an OPEN,
8073 * so we do a temporary open here just to get an open file to pass to
8076 static __be32 nfsd_test_lock(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file_lock *lock)
8078 struct nfsd_file *nf;
8079 struct inode *inode;
8082 err = nfsd_file_acquire(rqstp, fhp, NFSD_MAY_READ, &nf);
8085 inode = fhp->fh_dentry->d_inode;
8086 inode_lock(inode); /* to block new leases till after test_lock: */
8087 err = nfserrno(nfsd_open_break_lease(inode, NFSD_MAY_READ));
8090 lock->c.flc_file = nf->nf_file;
8091 err = nfserrno(vfs_test_lock(nf->nf_file, lock));
8092 lock->c.flc_file = NULL;
8094 inode_unlock(inode);
8103 nfsd4_lockt(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
8104 union nfsd4_op_u *u)
8106 struct nfsd4_lockt *lockt = &u->lockt;
8107 struct file_lock *file_lock = NULL;
8108 struct nfs4_lockowner *lo = NULL;
8110 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
8112 if (locks_in_grace(SVC_NET(rqstp)))
8113 return nfserr_grace;
8115 if (check_lock_length(lockt->lt_offset, lockt->lt_length))
8116 return nfserr_inval;
8118 if (!nfsd4_has_session(cstate)) {
8119 status = set_client(&lockt->lt_clientid, cstate, nn);
8124 if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
8127 file_lock = locks_alloc_lock();
8129 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
8130 status = nfserr_jukebox;
8134 switch (lockt->lt_type) {
8137 file_lock->c.flc_type = F_RDLCK;
8140 case NFS4_WRITEW_LT:
8141 file_lock->c.flc_type = F_WRLCK;
8144 dprintk("NFSD: nfs4_lockt: bad lock type!\n");
8145 status = nfserr_inval;
8149 lo = find_lockowner_str(cstate->clp, &lockt->lt_owner);
8151 file_lock->c.flc_owner = (fl_owner_t)lo;
8152 file_lock->c.flc_pid = current->tgid;
8153 file_lock->c.flc_flags = FL_POSIX;
8155 file_lock->fl_start = lockt->lt_offset;
8156 file_lock->fl_end = last_byte_offset(lockt->lt_offset, lockt->lt_length);
8158 nfs4_transform_lock_offset(file_lock);
8160 status = nfsd_test_lock(rqstp, &cstate->current_fh, file_lock);
8164 if (file_lock->c.flc_type != F_UNLCK) {
8165 status = nfserr_denied;
8166 nfs4_set_lock_denied(file_lock, &lockt->lt_denied);
8170 nfs4_put_stateowner(&lo->lo_owner);
8172 locks_free_lock(file_lock);
8176 void nfsd4_lockt_release(union nfsd4_op_u *u)
8178 struct nfsd4_lockt *lockt = &u->lockt;
8179 struct nfsd4_lock_denied *deny = &lockt->lt_denied;
8181 kfree(deny->ld_owner.data);
8185 nfsd4_locku(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
8186 union nfsd4_op_u *u)
8188 struct nfsd4_locku *locku = &u->locku;
8189 struct nfs4_ol_stateid *stp;
8190 struct nfsd_file *nf = NULL;
8191 struct file_lock *file_lock = NULL;
8194 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
8196 dprintk("NFSD: nfsd4_locku: start=%Ld length=%Ld\n",
8197 (long long) locku->lu_offset,
8198 (long long) locku->lu_length);
8200 if (check_lock_length(locku->lu_offset, locku->lu_length))
8201 return nfserr_inval;
8203 status = nfs4_preprocess_seqid_op(cstate, locku->lu_seqid,
8204 &locku->lu_stateid, SC_TYPE_LOCK, 0,
8208 nf = find_any_file(stp->st_stid.sc_file);
8210 status = nfserr_lock_range;
8213 file_lock = locks_alloc_lock();
8215 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
8216 status = nfserr_jukebox;
8220 file_lock->c.flc_type = F_UNLCK;
8221 file_lock->c.flc_owner = (fl_owner_t)lockowner(nfs4_get_stateowner(stp->st_stateowner));
8222 file_lock->c.flc_pid = current->tgid;
8223 file_lock->c.flc_file = nf->nf_file;
8224 file_lock->c.flc_flags = FL_POSIX;
8225 file_lock->fl_lmops = &nfsd_posix_mng_ops;
8226 file_lock->fl_start = locku->lu_offset;
8228 file_lock->fl_end = last_byte_offset(locku->lu_offset,
8230 nfs4_transform_lock_offset(file_lock);
8232 err = vfs_lock_file(nf->nf_file, F_SETLK, file_lock, NULL);
8234 dprintk("NFSD: nfs4_locku: vfs_lock_file failed!\n");
8237 nfs4_inc_and_copy_stateid(&locku->lu_stateid, &stp->st_stid);
8241 mutex_unlock(&stp->st_mutex);
8242 nfs4_put_stid(&stp->st_stid);
8244 nfsd4_bump_seqid(cstate, status);
8246 locks_free_lock(file_lock);
8250 status = nfserrno(err);
8256 * true: locks held by lockowner
8257 * false: no locks held by lockowner
8260 check_for_locks(struct nfs4_file *fp, struct nfs4_lockowner *lowner)
8262 struct file_lock *fl;
8264 struct nfsd_file *nf;
8265 struct inode *inode;
8266 struct file_lock_context *flctx;
8268 spin_lock(&fp->fi_lock);
8269 nf = find_any_file_locked(fp);
8271 /* Any valid lock stateid should have some sort of access */
8276 inode = file_inode(nf->nf_file);
8277 flctx = locks_inode_context(inode);
8279 if (flctx && !list_empty_careful(&flctx->flc_posix)) {
8280 spin_lock(&flctx->flc_lock);
8281 for_each_file_lock(fl, &flctx->flc_posix) {
8282 if (fl->c.flc_owner == (fl_owner_t)lowner) {
8287 spin_unlock(&flctx->flc_lock);
8290 spin_unlock(&fp->fi_lock);
8295 * nfsd4_release_lockowner - process NFSv4.0 RELEASE_LOCKOWNER operations
8296 * @rqstp: RPC transaction
8297 * @cstate: NFSv4 COMPOUND state
8298 * @u: RELEASE_LOCKOWNER arguments
8300 * Check if theree are any locks still held and if not - free the lockowner
8301 * and any lock state that is owned.
8304 * %nfs_ok: lockowner released or not found
8305 * %nfserr_locks_held: lockowner still in use
8306 * %nfserr_stale_clientid: clientid no longer active
8307 * %nfserr_expired: clientid not recognized
8310 nfsd4_release_lockowner(struct svc_rqst *rqstp,
8311 struct nfsd4_compound_state *cstate,
8312 union nfsd4_op_u *u)
8314 struct nfsd4_release_lockowner *rlockowner = &u->release_lockowner;
8315 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
8316 clientid_t *clid = &rlockowner->rl_clientid;
8317 struct nfs4_ol_stateid *stp;
8318 struct nfs4_lockowner *lo;
8319 struct nfs4_client *clp;
8320 LIST_HEAD(reaplist);
8323 dprintk("nfsd4_release_lockowner clientid: (%08x/%08x):\n",
8324 clid->cl_boot, clid->cl_id);
8326 status = set_client(clid, cstate, nn);
8331 spin_lock(&clp->cl_lock);
8332 lo = find_lockowner_str_locked(clp, &rlockowner->rl_owner);
8334 spin_unlock(&clp->cl_lock);
8338 list_for_each_entry(stp, &lo->lo_owner.so_stateids, st_perstateowner) {
8339 if (check_for_locks(stp->st_stid.sc_file, lo)) {
8340 spin_unlock(&clp->cl_lock);
8341 nfs4_put_stateowner(&lo->lo_owner);
8342 return nfserr_locks_held;
8345 unhash_lockowner_locked(lo);
8346 while (!list_empty(&lo->lo_owner.so_stateids)) {
8347 stp = list_first_entry(&lo->lo_owner.so_stateids,
8348 struct nfs4_ol_stateid,
8350 unhash_lock_stateid(stp);
8351 put_ol_stateid_locked(stp, &reaplist);
8353 spin_unlock(&clp->cl_lock);
8355 free_ol_stateid_reaplist(&reaplist);
8356 remove_blocked_locks(lo);
8357 nfs4_put_stateowner(&lo->lo_owner);
8361 static inline struct nfs4_client_reclaim *
8364 return kmalloc(sizeof(struct nfs4_client_reclaim), GFP_KERNEL);
8368 nfs4_has_reclaimed_state(struct xdr_netobj name, struct nfsd_net *nn)
8370 struct nfs4_client_reclaim *crp;
8372 crp = nfsd4_find_reclaim_client(name, nn);
8373 return (crp && crp->cr_clp);
8377 * failure => all reset bets are off, nfserr_no_grace...
8379 * The caller is responsible for freeing name.data if NULL is returned (it
8380 * will be freed in nfs4_remove_reclaim_record in the normal case).
8382 struct nfs4_client_reclaim *
8383 nfs4_client_to_reclaim(struct xdr_netobj name, struct xdr_netobj princhash,
8384 struct nfsd_net *nn)
8386 unsigned int strhashval;
8387 struct nfs4_client_reclaim *crp;
8389 crp = alloc_reclaim();
8391 strhashval = clientstr_hashval(name);
8392 INIT_LIST_HEAD(&crp->cr_strhash);
8393 list_add(&crp->cr_strhash, &nn->reclaim_str_hashtbl[strhashval]);
8394 crp->cr_name.data = name.data;
8395 crp->cr_name.len = name.len;
8396 crp->cr_princhash.data = princhash.data;
8397 crp->cr_princhash.len = princhash.len;
8399 nn->reclaim_str_hashtbl_size++;
8405 nfs4_remove_reclaim_record(struct nfs4_client_reclaim *crp, struct nfsd_net *nn)
8407 list_del(&crp->cr_strhash);
8408 kfree(crp->cr_name.data);
8409 kfree(crp->cr_princhash.data);
8411 nn->reclaim_str_hashtbl_size--;
8415 nfs4_release_reclaim(struct nfsd_net *nn)
8417 struct nfs4_client_reclaim *crp = NULL;
8420 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
8421 while (!list_empty(&nn->reclaim_str_hashtbl[i])) {
8422 crp = list_entry(nn->reclaim_str_hashtbl[i].next,
8423 struct nfs4_client_reclaim, cr_strhash);
8424 nfs4_remove_reclaim_record(crp, nn);
8427 WARN_ON_ONCE(nn->reclaim_str_hashtbl_size);
8431 * called from OPEN, CLAIM_PREVIOUS with a new clientid. */
8432 struct nfs4_client_reclaim *
8433 nfsd4_find_reclaim_client(struct xdr_netobj name, struct nfsd_net *nn)
8435 unsigned int strhashval;
8436 struct nfs4_client_reclaim *crp = NULL;
8438 strhashval = clientstr_hashval(name);
8439 list_for_each_entry(crp, &nn->reclaim_str_hashtbl[strhashval], cr_strhash) {
8440 if (compare_blob(&crp->cr_name, &name) == 0) {
8448 nfs4_check_open_reclaim(struct nfs4_client *clp)
8450 if (test_bit(NFSD4_CLIENT_RECLAIM_COMPLETE, &clp->cl_flags))
8451 return nfserr_no_grace;
8453 if (nfsd4_client_record_check(clp))
8454 return nfserr_reclaim_bad;
8460 * Since the lifetime of a delegation isn't limited to that of an open, a
8461 * client may quite reasonably hang on to a delegation as long as it has
8462 * the inode cached. This becomes an obvious problem the first time a
8463 * client's inode cache approaches the size of the server's total memory.
8465 * For now we avoid this problem by imposing a hard limit on the number
8466 * of delegations, which varies according to the server's memory size.
8469 set_max_delegations(void)
8472 * Allow at most 4 delegations per megabyte of RAM. Quick
8473 * estimates suggest that in the worst case (where every delegation
8474 * is for a different inode), a delegation could take about 1.5K,
8475 * giving a worst case usage of about 6% of memory.
8477 max_delegations = nr_free_buffer_pages() >> (20 - 2 - PAGE_SHIFT);
8480 static int nfs4_state_create_net(struct net *net)
8482 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
8485 nn->conf_id_hashtbl = kmalloc_array(CLIENT_HASH_SIZE,
8486 sizeof(struct list_head),
8488 if (!nn->conf_id_hashtbl)
8490 nn->unconf_id_hashtbl = kmalloc_array(CLIENT_HASH_SIZE,
8491 sizeof(struct list_head),
8493 if (!nn->unconf_id_hashtbl)
8495 nn->sessionid_hashtbl = kmalloc_array(SESSION_HASH_SIZE,
8496 sizeof(struct list_head),
8498 if (!nn->sessionid_hashtbl)
8501 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
8502 INIT_LIST_HEAD(&nn->conf_id_hashtbl[i]);
8503 INIT_LIST_HEAD(&nn->unconf_id_hashtbl[i]);
8505 for (i = 0; i < SESSION_HASH_SIZE; i++)
8506 INIT_LIST_HEAD(&nn->sessionid_hashtbl[i]);
8507 nn->conf_name_tree = RB_ROOT;
8508 nn->unconf_name_tree = RB_ROOT;
8509 nn->boot_time = ktime_get_real_seconds();
8510 nn->grace_ended = false;
8511 nn->nfsd4_manager.block_opens = true;
8512 INIT_LIST_HEAD(&nn->nfsd4_manager.list);
8513 INIT_LIST_HEAD(&nn->client_lru);
8514 INIT_LIST_HEAD(&nn->close_lru);
8515 INIT_LIST_HEAD(&nn->del_recall_lru);
8516 spin_lock_init(&nn->client_lock);
8517 spin_lock_init(&nn->s2s_cp_lock);
8518 idr_init(&nn->s2s_cp_stateids);
8520 spin_lock_init(&nn->blocked_locks_lock);
8521 INIT_LIST_HEAD(&nn->blocked_locks_lru);
8523 INIT_DELAYED_WORK(&nn->laundromat_work, laundromat_main);
8524 INIT_WORK(&nn->nfsd_shrinker_work, nfsd4_state_shrinker_worker);
8527 nn->nfsd_client_shrinker = shrinker_alloc(0, "nfsd-client");
8528 if (!nn->nfsd_client_shrinker)
8531 nn->nfsd_client_shrinker->scan_objects = nfsd4_state_shrinker_scan;
8532 nn->nfsd_client_shrinker->count_objects = nfsd4_state_shrinker_count;
8533 nn->nfsd_client_shrinker->private_data = nn;
8535 shrinker_register(nn->nfsd_client_shrinker);
8541 kfree(nn->sessionid_hashtbl);
8543 kfree(nn->unconf_id_hashtbl);
8545 kfree(nn->conf_id_hashtbl);
8551 nfs4_state_destroy_net(struct net *net)
8554 struct nfs4_client *clp = NULL;
8555 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
8557 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
8558 while (!list_empty(&nn->conf_id_hashtbl[i])) {
8559 clp = list_entry(nn->conf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
8560 destroy_client(clp);
8564 WARN_ON(!list_empty(&nn->blocked_locks_lru));
8566 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
8567 while (!list_empty(&nn->unconf_id_hashtbl[i])) {
8568 clp = list_entry(nn->unconf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
8569 destroy_client(clp);
8573 kfree(nn->sessionid_hashtbl);
8574 kfree(nn->unconf_id_hashtbl);
8575 kfree(nn->conf_id_hashtbl);
8580 nfs4_state_start_net(struct net *net)
8582 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
8585 ret = nfs4_state_create_net(net);
8588 locks_start_grace(net, &nn->nfsd4_manager);
8589 nfsd4_client_tracking_init(net);
8590 if (nn->track_reclaim_completes && nn->reclaim_str_hashtbl_size == 0)
8592 printk(KERN_INFO "NFSD: starting %lld-second grace period (net %x)\n",
8593 nn->nfsd4_grace, net->ns.inum);
8594 trace_nfsd_grace_start(nn);
8595 queue_delayed_work(laundry_wq, &nn->laundromat_work, nn->nfsd4_grace * HZ);
8599 printk(KERN_INFO "NFSD: no clients to reclaim, skipping NFSv4 grace period (net %x)\n",
8601 queue_delayed_work(laundry_wq, &nn->laundromat_work, nn->nfsd4_lease * HZ);
8602 nfsd4_end_grace(nn);
8606 /* initialization to perform when the nfsd service is started: */
8609 nfs4_state_start(void)
8613 ret = rhltable_init(&nfs4_file_rhltable, &nfs4_file_rhash_params);
8617 ret = nfsd4_create_callback_queue();
8619 rhltable_destroy(&nfs4_file_rhltable);
8623 set_max_delegations();
8628 nfs4_state_shutdown_net(struct net *net)
8630 struct nfs4_delegation *dp = NULL;
8631 struct list_head *pos, *next, reaplist;
8632 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
8634 shrinker_free(nn->nfsd_client_shrinker);
8635 cancel_work(&nn->nfsd_shrinker_work);
8636 cancel_delayed_work_sync(&nn->laundromat_work);
8637 locks_end_grace(&nn->nfsd4_manager);
8639 INIT_LIST_HEAD(&reaplist);
8640 spin_lock(&state_lock);
8641 list_for_each_safe(pos, next, &nn->del_recall_lru) {
8642 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
8643 unhash_delegation_locked(dp, SC_STATUS_CLOSED);
8644 list_add(&dp->dl_recall_lru, &reaplist);
8646 spin_unlock(&state_lock);
8647 list_for_each_safe(pos, next, &reaplist) {
8648 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
8649 list_del_init(&dp->dl_recall_lru);
8650 destroy_unhashed_deleg(dp);
8653 nfsd4_client_tracking_exit(net);
8654 nfs4_state_destroy_net(net);
8655 #ifdef CONFIG_NFSD_V4_2_INTER_SSC
8656 nfsd4_ssc_shutdown_umount(nn);
8661 nfs4_state_shutdown(void)
8663 nfsd4_destroy_callback_queue();
8664 rhltable_destroy(&nfs4_file_rhltable);
8668 get_stateid(struct nfsd4_compound_state *cstate, stateid_t *stateid)
8670 if (HAS_CSTATE_FLAG(cstate, CURRENT_STATE_ID_FLAG) &&
8671 CURRENT_STATEID(stateid))
8672 memcpy(stateid, &cstate->current_stateid, sizeof(stateid_t));
8676 put_stateid(struct nfsd4_compound_state *cstate, stateid_t *stateid)
8678 if (cstate->minorversion) {
8679 memcpy(&cstate->current_stateid, stateid, sizeof(stateid_t));
8680 SET_CSTATE_FLAG(cstate, CURRENT_STATE_ID_FLAG);
8685 clear_current_stateid(struct nfsd4_compound_state *cstate)
8687 CLEAR_CSTATE_FLAG(cstate, CURRENT_STATE_ID_FLAG);
8691 * functions to set current state id
8694 nfsd4_set_opendowngradestateid(struct nfsd4_compound_state *cstate,
8695 union nfsd4_op_u *u)
8697 put_stateid(cstate, &u->open_downgrade.od_stateid);
8701 nfsd4_set_openstateid(struct nfsd4_compound_state *cstate,
8702 union nfsd4_op_u *u)
8704 put_stateid(cstate, &u->open.op_stateid);
8708 nfsd4_set_closestateid(struct nfsd4_compound_state *cstate,
8709 union nfsd4_op_u *u)
8711 put_stateid(cstate, &u->close.cl_stateid);
8715 nfsd4_set_lockstateid(struct nfsd4_compound_state *cstate,
8716 union nfsd4_op_u *u)
8718 put_stateid(cstate, &u->lock.lk_resp_stateid);
8722 * functions to consume current state id
8726 nfsd4_get_opendowngradestateid(struct nfsd4_compound_state *cstate,
8727 union nfsd4_op_u *u)
8729 get_stateid(cstate, &u->open_downgrade.od_stateid);
8733 nfsd4_get_delegreturnstateid(struct nfsd4_compound_state *cstate,
8734 union nfsd4_op_u *u)
8736 get_stateid(cstate, &u->delegreturn.dr_stateid);
8740 nfsd4_get_freestateid(struct nfsd4_compound_state *cstate,
8741 union nfsd4_op_u *u)
8743 get_stateid(cstate, &u->free_stateid.fr_stateid);
8747 nfsd4_get_setattrstateid(struct nfsd4_compound_state *cstate,
8748 union nfsd4_op_u *u)
8750 get_stateid(cstate, &u->setattr.sa_stateid);
8754 nfsd4_get_closestateid(struct nfsd4_compound_state *cstate,
8755 union nfsd4_op_u *u)
8757 get_stateid(cstate, &u->close.cl_stateid);
8761 nfsd4_get_lockustateid(struct nfsd4_compound_state *cstate,
8762 union nfsd4_op_u *u)
8764 get_stateid(cstate, &u->locku.lu_stateid);
8768 nfsd4_get_readstateid(struct nfsd4_compound_state *cstate,
8769 union nfsd4_op_u *u)
8771 get_stateid(cstate, &u->read.rd_stateid);
8775 nfsd4_get_writestateid(struct nfsd4_compound_state *cstate,
8776 union nfsd4_op_u *u)
8778 get_stateid(cstate, &u->write.wr_stateid);
8782 * nfsd4_deleg_getattr_conflict - Recall if GETATTR causes conflict
8783 * @rqstp: RPC transaction context
8784 * @inode: file to be checked for a conflict
8785 * @modified: return true if file was modified
8786 * @size: new size of file if modified is true
8788 * This function is called when there is a conflict between a write
8789 * delegation and a change/size GETATTR from another client. The server
8790 * must either use the CB_GETATTR to get the current values of the
8791 * attributes from the client that holds the delegation or recall the
8792 * delegation before replying to the GETATTR. See RFC 8881 section
8795 * Returns 0 if there is no conflict; otherwise an nfs_stat
8799 nfsd4_deleg_getattr_conflict(struct svc_rqst *rqstp, struct inode *inode,
8800 bool *modified, u64 *size)
8803 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
8804 struct file_lock_context *ctx;
8805 struct file_lease *fl;
8806 struct nfs4_delegation *dp;
8808 struct nfs4_cb_fattr *ncf;
8811 ctx = locks_inode_context(inode);
8814 spin_lock(&ctx->flc_lock);
8815 for_each_file_lock(fl, &ctx->flc_lease) {
8816 unsigned char type = fl->c.flc_type;
8818 if (fl->c.flc_flags == FL_LAYOUT)
8820 if (fl->fl_lmops != &nfsd_lease_mng_ops) {
8822 * non-nfs lease, if it's a lease with F_RDLCK then
8823 * we are done; there isn't any write delegation
8826 if (type == F_RDLCK)
8830 if (type == F_WRLCK) {
8831 dp = fl->c.flc_owner;
8832 if (dp->dl_recall.cb_clp == *(rqstp->rq_lease_breaker)) {
8833 spin_unlock(&ctx->flc_lock);
8837 nfsd_stats_wdeleg_getattr_inc(nn);
8838 dp = fl->c.flc_owner;
8839 ncf = &dp->dl_cb_fattr;
8840 nfs4_cb_getattr(&dp->dl_cb_fattr);
8841 spin_unlock(&ctx->flc_lock);
8842 wait_on_bit_timeout(&ncf->ncf_cb_flags, CB_GETATTR_BUSY,
8843 TASK_INTERRUPTIBLE, NFSD_CB_GETATTR_TIMEOUT);
8844 if (ncf->ncf_cb_status) {
8845 /* Recall delegation only if client didn't respond */
8846 status = nfserrno(nfsd_open_break_lease(inode, NFSD_MAY_READ));
8847 if (status != nfserr_jukebox ||
8848 !nfsd_wait_for_delegreturn(rqstp, inode))
8851 if (!ncf->ncf_file_modified &&
8852 (ncf->ncf_initial_cinfo != ncf->ncf_cb_change ||
8853 ncf->ncf_cur_fsize != ncf->ncf_cb_fsize))
8854 ncf->ncf_file_modified = true;
8855 if (ncf->ncf_file_modified) {
8857 * Per section 10.4.3 of RFC 8881, the server would
8858 * not update the file's metadata with the client's
8861 attrs.ia_mtime = attrs.ia_ctime = current_time(inode);
8862 attrs.ia_valid = ATTR_MTIME | ATTR_CTIME;
8863 setattr_copy(&nop_mnt_idmap, inode, &attrs);
8864 mark_inode_dirty(inode);
8865 ncf->ncf_cur_fsize = ncf->ncf_cb_fsize;
8866 *size = ncf->ncf_cur_fsize;
8873 spin_unlock(&ctx->flc_lock);