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/nfs_ssc.h>
51 #include "current_stateid.h"
55 #include "filecache.h"
58 #define NFSDDBG_FACILITY NFSDDBG_PROC
60 #define all_ones {{~0,~0},~0}
61 static const stateid_t one_stateid = {
63 .si_opaque = all_ones,
65 static const stateid_t zero_stateid = {
68 static const stateid_t currentstateid = {
71 static const stateid_t close_stateid = {
72 .si_generation = 0xffffffffU,
75 static u64 current_sessionid = 1;
77 #define ZERO_STATEID(stateid) (!memcmp((stateid), &zero_stateid, sizeof(stateid_t)))
78 #define ONE_STATEID(stateid) (!memcmp((stateid), &one_stateid, sizeof(stateid_t)))
79 #define CURRENT_STATEID(stateid) (!memcmp((stateid), ¤tstateid, sizeof(stateid_t)))
80 #define CLOSE_STATEID(stateid) (!memcmp((stateid), &close_stateid, sizeof(stateid_t)))
82 /* forward declarations */
83 static bool check_for_locks(struct nfs4_file *fp, struct nfs4_lockowner *lowner);
84 static void nfs4_free_ol_stateid(struct nfs4_stid *stid);
85 void nfsd4_end_grace(struct nfsd_net *nn);
86 static void _free_cpntf_state_locked(struct nfsd_net *nn, struct nfs4_cpntf_state *cps);
91 * Currently used for the del_recall_lru and file hash table. In an
92 * effort to decrease the scope of the client_mutex, this spinlock may
93 * eventually cover more:
95 static DEFINE_SPINLOCK(state_lock);
97 enum nfsd4_st_mutex_lock_subclass {
98 OPEN_STATEID_MUTEX = 0,
99 LOCK_STATEID_MUTEX = 1,
103 * A waitqueue for all in-progress 4.0 CLOSE operations that are waiting for
104 * the refcount on the open stateid to drop.
106 static DECLARE_WAIT_QUEUE_HEAD(close_wq);
109 * A waitqueue where a writer to clients/#/ctl destroying a client can
110 * wait for cl_rpc_users to drop to 0 and then for the client to be
113 static DECLARE_WAIT_QUEUE_HEAD(expiry_wq);
115 static struct kmem_cache *client_slab;
116 static struct kmem_cache *openowner_slab;
117 static struct kmem_cache *lockowner_slab;
118 static struct kmem_cache *file_slab;
119 static struct kmem_cache *stateid_slab;
120 static struct kmem_cache *deleg_slab;
121 static struct kmem_cache *odstate_slab;
123 static void free_session(struct nfsd4_session *);
125 static const struct nfsd4_callback_ops nfsd4_cb_recall_ops;
126 static const struct nfsd4_callback_ops nfsd4_cb_notify_lock_ops;
128 static struct workqueue_struct *laundry_wq;
130 int nfsd4_create_laundry_wq(void)
134 laundry_wq = alloc_workqueue("%s", WQ_UNBOUND, 0, "nfsd4");
135 if (laundry_wq == NULL)
140 void nfsd4_destroy_laundry_wq(void)
142 destroy_workqueue(laundry_wq);
145 static bool is_session_dead(struct nfsd4_session *ses)
147 return ses->se_flags & NFS4_SESSION_DEAD;
150 static __be32 mark_session_dead_locked(struct nfsd4_session *ses, int ref_held_by_me)
152 if (atomic_read(&ses->se_ref) > ref_held_by_me)
153 return nfserr_jukebox;
154 ses->se_flags |= NFS4_SESSION_DEAD;
158 static bool is_client_expired(struct nfs4_client *clp)
160 return clp->cl_time == 0;
163 static void nfsd4_dec_courtesy_client_count(struct nfsd_net *nn,
164 struct nfs4_client *clp)
166 if (clp->cl_state != NFSD4_ACTIVE)
167 atomic_add_unless(&nn->nfsd_courtesy_clients, -1, 0);
170 static __be32 get_client_locked(struct nfs4_client *clp)
172 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
174 lockdep_assert_held(&nn->client_lock);
176 if (is_client_expired(clp))
177 return nfserr_expired;
178 atomic_inc(&clp->cl_rpc_users);
179 nfsd4_dec_courtesy_client_count(nn, clp);
180 clp->cl_state = NFSD4_ACTIVE;
184 /* must be called under the client_lock */
186 renew_client_locked(struct nfs4_client *clp)
188 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
190 if (is_client_expired(clp)) {
192 printk("%s: client (clientid %08x/%08x) already expired\n",
194 clp->cl_clientid.cl_boot,
195 clp->cl_clientid.cl_id);
199 list_move_tail(&clp->cl_lru, &nn->client_lru);
200 clp->cl_time = ktime_get_boottime_seconds();
201 nfsd4_dec_courtesy_client_count(nn, clp);
202 clp->cl_state = NFSD4_ACTIVE;
205 static void put_client_renew_locked(struct nfs4_client *clp)
207 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
209 lockdep_assert_held(&nn->client_lock);
211 if (!atomic_dec_and_test(&clp->cl_rpc_users))
213 if (!is_client_expired(clp))
214 renew_client_locked(clp);
216 wake_up_all(&expiry_wq);
219 static void put_client_renew(struct nfs4_client *clp)
221 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
223 if (!atomic_dec_and_lock(&clp->cl_rpc_users, &nn->client_lock))
225 if (!is_client_expired(clp))
226 renew_client_locked(clp);
228 wake_up_all(&expiry_wq);
229 spin_unlock(&nn->client_lock);
232 static __be32 nfsd4_get_session_locked(struct nfsd4_session *ses)
236 if (is_session_dead(ses))
237 return nfserr_badsession;
238 status = get_client_locked(ses->se_client);
241 atomic_inc(&ses->se_ref);
245 static void nfsd4_put_session_locked(struct nfsd4_session *ses)
247 struct nfs4_client *clp = ses->se_client;
248 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
250 lockdep_assert_held(&nn->client_lock);
252 if (atomic_dec_and_test(&ses->se_ref) && is_session_dead(ses))
254 put_client_renew_locked(clp);
257 static void nfsd4_put_session(struct nfsd4_session *ses)
259 struct nfs4_client *clp = ses->se_client;
260 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
262 spin_lock(&nn->client_lock);
263 nfsd4_put_session_locked(ses);
264 spin_unlock(&nn->client_lock);
267 static struct nfsd4_blocked_lock *
268 find_blocked_lock(struct nfs4_lockowner *lo, struct knfsd_fh *fh,
271 struct nfsd4_blocked_lock *cur, *found = NULL;
273 spin_lock(&nn->blocked_locks_lock);
274 list_for_each_entry(cur, &lo->lo_blocked, nbl_list) {
275 if (fh_match(fh, &cur->nbl_fh)) {
276 list_del_init(&cur->nbl_list);
277 WARN_ON(list_empty(&cur->nbl_lru));
278 list_del_init(&cur->nbl_lru);
283 spin_unlock(&nn->blocked_locks_lock);
285 locks_delete_block(&found->nbl_lock);
289 static struct nfsd4_blocked_lock *
290 find_or_allocate_block(struct nfs4_lockowner *lo, struct knfsd_fh *fh,
293 struct nfsd4_blocked_lock *nbl;
295 nbl = find_blocked_lock(lo, fh, nn);
297 nbl= kmalloc(sizeof(*nbl), GFP_KERNEL);
299 INIT_LIST_HEAD(&nbl->nbl_list);
300 INIT_LIST_HEAD(&nbl->nbl_lru);
301 fh_copy_shallow(&nbl->nbl_fh, fh);
302 locks_init_lock(&nbl->nbl_lock);
303 kref_init(&nbl->nbl_kref);
304 nfsd4_init_cb(&nbl->nbl_cb, lo->lo_owner.so_client,
305 &nfsd4_cb_notify_lock_ops,
306 NFSPROC4_CLNT_CB_NOTIFY_LOCK);
313 free_nbl(struct kref *kref)
315 struct nfsd4_blocked_lock *nbl;
317 nbl = container_of(kref, struct nfsd4_blocked_lock, nbl_kref);
322 free_blocked_lock(struct nfsd4_blocked_lock *nbl)
324 locks_delete_block(&nbl->nbl_lock);
325 locks_release_private(&nbl->nbl_lock);
326 kref_put(&nbl->nbl_kref, free_nbl);
330 remove_blocked_locks(struct nfs4_lockowner *lo)
332 struct nfs4_client *clp = lo->lo_owner.so_client;
333 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
334 struct nfsd4_blocked_lock *nbl;
337 /* Dequeue all blocked locks */
338 spin_lock(&nn->blocked_locks_lock);
339 while (!list_empty(&lo->lo_blocked)) {
340 nbl = list_first_entry(&lo->lo_blocked,
341 struct nfsd4_blocked_lock,
343 list_del_init(&nbl->nbl_list);
344 WARN_ON(list_empty(&nbl->nbl_lru));
345 list_move(&nbl->nbl_lru, &reaplist);
347 spin_unlock(&nn->blocked_locks_lock);
350 while (!list_empty(&reaplist)) {
351 nbl = list_first_entry(&reaplist, struct nfsd4_blocked_lock,
353 list_del_init(&nbl->nbl_lru);
354 free_blocked_lock(nbl);
359 nfsd4_cb_notify_lock_prepare(struct nfsd4_callback *cb)
361 struct nfsd4_blocked_lock *nbl = container_of(cb,
362 struct nfsd4_blocked_lock, nbl_cb);
363 locks_delete_block(&nbl->nbl_lock);
367 nfsd4_cb_notify_lock_done(struct nfsd4_callback *cb, struct rpc_task *task)
369 trace_nfsd_cb_notify_lock_done(&zero_stateid, task);
372 * Since this is just an optimization, we don't try very hard if it
373 * turns out not to succeed. We'll requeue it on NFS4ERR_DELAY, and
374 * just quit trying on anything else.
376 switch (task->tk_status) {
378 rpc_delay(task, 1 * HZ);
386 nfsd4_cb_notify_lock_release(struct nfsd4_callback *cb)
388 struct nfsd4_blocked_lock *nbl = container_of(cb,
389 struct nfsd4_blocked_lock, nbl_cb);
391 free_blocked_lock(nbl);
394 static const struct nfsd4_callback_ops nfsd4_cb_notify_lock_ops = {
395 .prepare = nfsd4_cb_notify_lock_prepare,
396 .done = nfsd4_cb_notify_lock_done,
397 .release = nfsd4_cb_notify_lock_release,
401 * We store the NONE, READ, WRITE, and BOTH bits separately in the
402 * st_{access,deny}_bmap field of the stateid, in order to track not
403 * only what share bits are currently in force, but also what
404 * combinations of share bits previous opens have used. This allows us
405 * to enforce the recommendation in
406 * https://datatracker.ietf.org/doc/html/rfc7530#section-16.19.4 that
407 * the server return an error if the client attempt to downgrade to a
408 * combination of share bits not explicable by closing some of its
411 * This enforcement is arguably incomplete, since we don't keep
412 * track of access/deny bit combinations; so, e.g., we allow:
414 * OPEN allow read, deny write
415 * OPEN allow both, deny none
416 * DOWNGRADE allow read, deny none
418 * which we should reject.
420 * But you could also argue that our current code is already overkill,
421 * since it only exists to return NFS4ERR_INVAL on incorrect client
425 bmap_to_share_mode(unsigned long bmap)
428 unsigned int access = 0;
430 for (i = 1; i < 4; i++) {
431 if (test_bit(i, &bmap))
437 /* set share access for a given stateid */
439 set_access(u32 access, struct nfs4_ol_stateid *stp)
441 unsigned char mask = 1 << access;
443 WARN_ON_ONCE(access > NFS4_SHARE_ACCESS_BOTH);
444 stp->st_access_bmap |= mask;
447 /* clear share access for a given stateid */
449 clear_access(u32 access, struct nfs4_ol_stateid *stp)
451 unsigned char mask = 1 << access;
453 WARN_ON_ONCE(access > NFS4_SHARE_ACCESS_BOTH);
454 stp->st_access_bmap &= ~mask;
457 /* test whether a given stateid has access */
459 test_access(u32 access, struct nfs4_ol_stateid *stp)
461 unsigned char mask = 1 << access;
463 return (bool)(stp->st_access_bmap & mask);
466 /* set share deny for a given stateid */
468 set_deny(u32 deny, struct nfs4_ol_stateid *stp)
470 unsigned char mask = 1 << deny;
472 WARN_ON_ONCE(deny > NFS4_SHARE_DENY_BOTH);
473 stp->st_deny_bmap |= mask;
476 /* clear share deny for a given stateid */
478 clear_deny(u32 deny, struct nfs4_ol_stateid *stp)
480 unsigned char mask = 1 << deny;
482 WARN_ON_ONCE(deny > NFS4_SHARE_DENY_BOTH);
483 stp->st_deny_bmap &= ~mask;
486 /* test whether a given stateid is denying specific access */
488 test_deny(u32 deny, struct nfs4_ol_stateid *stp)
490 unsigned char mask = 1 << deny;
492 return (bool)(stp->st_deny_bmap & mask);
495 static int nfs4_access_to_omode(u32 access)
497 switch (access & NFS4_SHARE_ACCESS_BOTH) {
498 case NFS4_SHARE_ACCESS_READ:
500 case NFS4_SHARE_ACCESS_WRITE:
502 case NFS4_SHARE_ACCESS_BOTH:
510 access_permit_read(struct nfs4_ol_stateid *stp)
512 return test_access(NFS4_SHARE_ACCESS_READ, stp) ||
513 test_access(NFS4_SHARE_ACCESS_BOTH, stp) ||
514 test_access(NFS4_SHARE_ACCESS_WRITE, stp);
518 access_permit_write(struct nfs4_ol_stateid *stp)
520 return test_access(NFS4_SHARE_ACCESS_WRITE, stp) ||
521 test_access(NFS4_SHARE_ACCESS_BOTH, stp);
524 static inline struct nfs4_stateowner *
525 nfs4_get_stateowner(struct nfs4_stateowner *sop)
527 atomic_inc(&sop->so_count);
532 same_owner_str(struct nfs4_stateowner *sop, struct xdr_netobj *owner)
534 return (sop->so_owner.len == owner->len) &&
535 0 == memcmp(sop->so_owner.data, owner->data, owner->len);
538 static struct nfs4_openowner *
539 find_openstateowner_str_locked(unsigned int hashval, struct nfsd4_open *open,
540 struct nfs4_client *clp)
542 struct nfs4_stateowner *so;
544 lockdep_assert_held(&clp->cl_lock);
546 list_for_each_entry(so, &clp->cl_ownerstr_hashtbl[hashval],
548 if (!so->so_is_open_owner)
550 if (same_owner_str(so, &open->op_owner))
551 return openowner(nfs4_get_stateowner(so));
556 static struct nfs4_openowner *
557 find_openstateowner_str(unsigned int hashval, struct nfsd4_open *open,
558 struct nfs4_client *clp)
560 struct nfs4_openowner *oo;
562 spin_lock(&clp->cl_lock);
563 oo = find_openstateowner_str_locked(hashval, open, clp);
564 spin_unlock(&clp->cl_lock);
569 opaque_hashval(const void *ptr, int nbytes)
571 unsigned char *cptr = (unsigned char *) ptr;
581 static void nfsd4_free_file_rcu(struct rcu_head *rcu)
583 struct nfs4_file *fp = container_of(rcu, struct nfs4_file, fi_rcu);
585 kmem_cache_free(file_slab, fp);
589 put_nfs4_file(struct nfs4_file *fi)
591 might_lock(&state_lock);
593 if (refcount_dec_and_lock(&fi->fi_ref, &state_lock)) {
594 hlist_del_rcu(&fi->fi_hash);
595 spin_unlock(&state_lock);
596 WARN_ON_ONCE(!list_empty(&fi->fi_clnt_odstate));
597 WARN_ON_ONCE(!list_empty(&fi->fi_delegations));
598 call_rcu(&fi->fi_rcu, nfsd4_free_file_rcu);
602 static struct nfsd_file *
603 __nfs4_get_fd(struct nfs4_file *f, int oflag)
605 if (f->fi_fds[oflag])
606 return nfsd_file_get(f->fi_fds[oflag]);
610 static struct nfsd_file *
611 find_writeable_file_locked(struct nfs4_file *f)
613 struct nfsd_file *ret;
615 lockdep_assert_held(&f->fi_lock);
617 ret = __nfs4_get_fd(f, O_WRONLY);
619 ret = __nfs4_get_fd(f, O_RDWR);
623 static struct nfsd_file *
624 find_writeable_file(struct nfs4_file *f)
626 struct nfsd_file *ret;
628 spin_lock(&f->fi_lock);
629 ret = find_writeable_file_locked(f);
630 spin_unlock(&f->fi_lock);
635 static struct nfsd_file *
636 find_readable_file_locked(struct nfs4_file *f)
638 struct nfsd_file *ret;
640 lockdep_assert_held(&f->fi_lock);
642 ret = __nfs4_get_fd(f, O_RDONLY);
644 ret = __nfs4_get_fd(f, O_RDWR);
648 static struct nfsd_file *
649 find_readable_file(struct nfs4_file *f)
651 struct nfsd_file *ret;
653 spin_lock(&f->fi_lock);
654 ret = find_readable_file_locked(f);
655 spin_unlock(&f->fi_lock);
661 find_any_file(struct nfs4_file *f)
663 struct nfsd_file *ret;
667 spin_lock(&f->fi_lock);
668 ret = __nfs4_get_fd(f, O_RDWR);
670 ret = __nfs4_get_fd(f, O_WRONLY);
672 ret = __nfs4_get_fd(f, O_RDONLY);
674 spin_unlock(&f->fi_lock);
678 static struct nfsd_file *find_any_file_locked(struct nfs4_file *f)
680 lockdep_assert_held(&f->fi_lock);
682 if (f->fi_fds[O_RDWR])
683 return f->fi_fds[O_RDWR];
684 if (f->fi_fds[O_WRONLY])
685 return f->fi_fds[O_WRONLY];
686 if (f->fi_fds[O_RDONLY])
687 return f->fi_fds[O_RDONLY];
691 static struct nfsd_file *find_deleg_file_locked(struct nfs4_file *f)
693 lockdep_assert_held(&f->fi_lock);
695 if (f->fi_deleg_file)
696 return f->fi_deleg_file;
700 static atomic_long_t num_delegations;
701 unsigned long max_delegations;
704 * Open owner state (share locks)
707 /* hash tables for lock and open owners */
708 #define OWNER_HASH_BITS 8
709 #define OWNER_HASH_SIZE (1 << OWNER_HASH_BITS)
710 #define OWNER_HASH_MASK (OWNER_HASH_SIZE - 1)
712 static unsigned int ownerstr_hashval(struct xdr_netobj *ownername)
716 ret = opaque_hashval(ownername->data, ownername->len);
717 return ret & OWNER_HASH_MASK;
720 /* hash table for nfs4_file */
721 #define FILE_HASH_BITS 8
722 #define FILE_HASH_SIZE (1 << FILE_HASH_BITS)
724 static unsigned int file_hashval(struct svc_fh *fh)
726 struct inode *inode = d_inode(fh->fh_dentry);
728 /* XXX: why not (here & in file cache) use inode? */
729 return (unsigned int)hash_long(inode->i_ino, FILE_HASH_BITS);
732 static struct hlist_head file_hashtbl[FILE_HASH_SIZE];
735 * Check if courtesy clients have conflicting access and resolve it if possible
737 * access: is op_share_access if share_access is true.
738 * Check if access mode, op_share_access, would conflict with
739 * the current deny mode of the file 'fp'.
740 * access: is op_share_deny if share_access is false.
741 * Check if the deny mode, op_share_deny, would conflict with
742 * current access of the file 'fp'.
743 * stp: skip checking this entry.
744 * new_stp: normal open, not open upgrade.
747 * false - access/deny mode conflict with normal client.
748 * true - no conflict or conflict with courtesy client(s) is resolved.
751 nfs4_resolve_deny_conflicts_locked(struct nfs4_file *fp, bool new_stp,
752 struct nfs4_ol_stateid *stp, u32 access, bool share_access)
754 struct nfs4_ol_stateid *st;
755 bool resolvable = true;
758 struct nfs4_client *clp;
760 lockdep_assert_held(&fp->fi_lock);
761 list_for_each_entry(st, &fp->fi_stateids, st_perfile) {
762 /* ignore lock stateid */
765 if (st == stp && new_stp)
767 /* check file access against deny mode or vice versa */
768 bmap = share_access ? st->st_deny_bmap : st->st_access_bmap;
769 if (!(access & bmap_to_share_mode(bmap)))
771 clp = st->st_stid.sc_client;
772 if (try_to_expire_client(clp))
778 clp = stp->st_stid.sc_client;
779 nn = net_generic(clp->net, nfsd_net_id);
780 mod_delayed_work(laundry_wq, &nn->laundromat_work, 0);
786 __nfs4_file_get_access(struct nfs4_file *fp, u32 access)
788 lockdep_assert_held(&fp->fi_lock);
790 if (access & NFS4_SHARE_ACCESS_WRITE)
791 atomic_inc(&fp->fi_access[O_WRONLY]);
792 if (access & NFS4_SHARE_ACCESS_READ)
793 atomic_inc(&fp->fi_access[O_RDONLY]);
797 nfs4_file_get_access(struct nfs4_file *fp, u32 access)
799 lockdep_assert_held(&fp->fi_lock);
801 /* Does this access mode make sense? */
802 if (access & ~NFS4_SHARE_ACCESS_BOTH)
805 /* Does it conflict with a deny mode already set? */
806 if ((access & fp->fi_share_deny) != 0)
807 return nfserr_share_denied;
809 __nfs4_file_get_access(fp, access);
813 static __be32 nfs4_file_check_deny(struct nfs4_file *fp, u32 deny)
815 /* Common case is that there is no deny mode. */
817 /* Does this deny mode make sense? */
818 if (deny & ~NFS4_SHARE_DENY_BOTH)
821 if ((deny & NFS4_SHARE_DENY_READ) &&
822 atomic_read(&fp->fi_access[O_RDONLY]))
823 return nfserr_share_denied;
825 if ((deny & NFS4_SHARE_DENY_WRITE) &&
826 atomic_read(&fp->fi_access[O_WRONLY]))
827 return nfserr_share_denied;
832 static void __nfs4_file_put_access(struct nfs4_file *fp, int oflag)
834 might_lock(&fp->fi_lock);
836 if (atomic_dec_and_lock(&fp->fi_access[oflag], &fp->fi_lock)) {
837 struct nfsd_file *f1 = NULL;
838 struct nfsd_file *f2 = NULL;
840 swap(f1, fp->fi_fds[oflag]);
841 if (atomic_read(&fp->fi_access[1 - oflag]) == 0)
842 swap(f2, fp->fi_fds[O_RDWR]);
843 spin_unlock(&fp->fi_lock);
851 static void nfs4_file_put_access(struct nfs4_file *fp, u32 access)
853 WARN_ON_ONCE(access & ~NFS4_SHARE_ACCESS_BOTH);
855 if (access & NFS4_SHARE_ACCESS_WRITE)
856 __nfs4_file_put_access(fp, O_WRONLY);
857 if (access & NFS4_SHARE_ACCESS_READ)
858 __nfs4_file_put_access(fp, O_RDONLY);
862 * Allocate a new open/delegation state counter. This is needed for
863 * pNFS for proper return on close semantics.
865 * Note that we only allocate it for pNFS-enabled exports, otherwise
866 * all pointers to struct nfs4_clnt_odstate are always NULL.
868 static struct nfs4_clnt_odstate *
869 alloc_clnt_odstate(struct nfs4_client *clp)
871 struct nfs4_clnt_odstate *co;
873 co = kmem_cache_zalloc(odstate_slab, GFP_KERNEL);
876 refcount_set(&co->co_odcount, 1);
882 hash_clnt_odstate_locked(struct nfs4_clnt_odstate *co)
884 struct nfs4_file *fp = co->co_file;
886 lockdep_assert_held(&fp->fi_lock);
887 list_add(&co->co_perfile, &fp->fi_clnt_odstate);
891 get_clnt_odstate(struct nfs4_clnt_odstate *co)
894 refcount_inc(&co->co_odcount);
898 put_clnt_odstate(struct nfs4_clnt_odstate *co)
900 struct nfs4_file *fp;
906 if (refcount_dec_and_lock(&co->co_odcount, &fp->fi_lock)) {
907 list_del(&co->co_perfile);
908 spin_unlock(&fp->fi_lock);
910 nfsd4_return_all_file_layouts(co->co_client, fp);
911 kmem_cache_free(odstate_slab, co);
915 static struct nfs4_clnt_odstate *
916 find_or_hash_clnt_odstate(struct nfs4_file *fp, struct nfs4_clnt_odstate *new)
918 struct nfs4_clnt_odstate *co;
919 struct nfs4_client *cl;
926 spin_lock(&fp->fi_lock);
927 list_for_each_entry(co, &fp->fi_clnt_odstate, co_perfile) {
928 if (co->co_client == cl) {
929 get_clnt_odstate(co);
935 hash_clnt_odstate_locked(new);
937 spin_unlock(&fp->fi_lock);
941 struct nfs4_stid *nfs4_alloc_stid(struct nfs4_client *cl, struct kmem_cache *slab,
942 void (*sc_free)(struct nfs4_stid *))
944 struct nfs4_stid *stid;
947 stid = kmem_cache_zalloc(slab, GFP_KERNEL);
951 idr_preload(GFP_KERNEL);
952 spin_lock(&cl->cl_lock);
953 /* Reserving 0 for start of file in nfsdfs "states" file: */
954 new_id = idr_alloc_cyclic(&cl->cl_stateids, stid, 1, 0, GFP_NOWAIT);
955 spin_unlock(&cl->cl_lock);
960 stid->sc_free = sc_free;
961 stid->sc_client = cl;
962 stid->sc_stateid.si_opaque.so_id = new_id;
963 stid->sc_stateid.si_opaque.so_clid = cl->cl_clientid;
964 /* Will be incremented before return to client: */
965 refcount_set(&stid->sc_count, 1);
966 spin_lock_init(&stid->sc_lock);
967 INIT_LIST_HEAD(&stid->sc_cp_list);
970 * It shouldn't be a problem to reuse an opaque stateid value.
971 * I don't think it is for 4.1. But with 4.0 I worry that, for
972 * example, a stray write retransmission could be accepted by
973 * the server when it should have been rejected. Therefore,
974 * adopt a trick from the sctp code to attempt to maximize the
975 * amount of time until an id is reused, by ensuring they always
976 * "increase" (mod INT_MAX):
980 kmem_cache_free(slab, stid);
985 * Create a unique stateid_t to represent each COPY.
987 static int nfs4_init_cp_state(struct nfsd_net *nn, copy_stateid_t *stid,
988 unsigned char cs_type)
992 stid->cs_stid.si_opaque.so_clid.cl_boot = (u32)nn->boot_time;
993 stid->cs_stid.si_opaque.so_clid.cl_id = nn->s2s_cp_cl_id;
995 idr_preload(GFP_KERNEL);
996 spin_lock(&nn->s2s_cp_lock);
997 new_id = idr_alloc_cyclic(&nn->s2s_cp_stateids, stid, 0, 0, GFP_NOWAIT);
998 stid->cs_stid.si_opaque.so_id = new_id;
999 stid->cs_stid.si_generation = 1;
1000 spin_unlock(&nn->s2s_cp_lock);
1004 stid->cs_type = cs_type;
1008 int nfs4_init_copy_state(struct nfsd_net *nn, struct nfsd4_copy *copy)
1010 return nfs4_init_cp_state(nn, ©->cp_stateid, NFS4_COPY_STID);
1013 struct nfs4_cpntf_state *nfs4_alloc_init_cpntf_state(struct nfsd_net *nn,
1014 struct nfs4_stid *p_stid)
1016 struct nfs4_cpntf_state *cps;
1018 cps = kzalloc(sizeof(struct nfs4_cpntf_state), GFP_KERNEL);
1021 cps->cpntf_time = ktime_get_boottime_seconds();
1022 refcount_set(&cps->cp_stateid.cs_count, 1);
1023 if (!nfs4_init_cp_state(nn, &cps->cp_stateid, NFS4_COPYNOTIFY_STID))
1025 spin_lock(&nn->s2s_cp_lock);
1026 list_add(&cps->cp_list, &p_stid->sc_cp_list);
1027 spin_unlock(&nn->s2s_cp_lock);
1034 void nfs4_free_copy_state(struct nfsd4_copy *copy)
1036 struct nfsd_net *nn;
1038 if (copy->cp_stateid.cs_type != NFS4_COPY_STID)
1040 nn = net_generic(copy->cp_clp->net, nfsd_net_id);
1041 spin_lock(&nn->s2s_cp_lock);
1042 idr_remove(&nn->s2s_cp_stateids,
1043 copy->cp_stateid.cs_stid.si_opaque.so_id);
1044 spin_unlock(&nn->s2s_cp_lock);
1047 static void nfs4_free_cpntf_statelist(struct net *net, struct nfs4_stid *stid)
1049 struct nfs4_cpntf_state *cps;
1050 struct nfsd_net *nn;
1052 nn = net_generic(net, nfsd_net_id);
1053 spin_lock(&nn->s2s_cp_lock);
1054 while (!list_empty(&stid->sc_cp_list)) {
1055 cps = list_first_entry(&stid->sc_cp_list,
1056 struct nfs4_cpntf_state, cp_list);
1057 _free_cpntf_state_locked(nn, cps);
1059 spin_unlock(&nn->s2s_cp_lock);
1062 static struct nfs4_ol_stateid * nfs4_alloc_open_stateid(struct nfs4_client *clp)
1064 struct nfs4_stid *stid;
1066 stid = nfs4_alloc_stid(clp, stateid_slab, nfs4_free_ol_stateid);
1070 return openlockstateid(stid);
1073 static void nfs4_free_deleg(struct nfs4_stid *stid)
1075 struct nfs4_delegation *dp = delegstateid(stid);
1077 WARN_ON_ONCE(!list_empty(&stid->sc_cp_list));
1078 WARN_ON_ONCE(!list_empty(&dp->dl_perfile));
1079 WARN_ON_ONCE(!list_empty(&dp->dl_perclnt));
1080 WARN_ON_ONCE(!list_empty(&dp->dl_recall_lru));
1081 kmem_cache_free(deleg_slab, stid);
1082 atomic_long_dec(&num_delegations);
1086 * When we recall a delegation, we should be careful not to hand it
1087 * out again straight away.
1088 * To ensure this we keep a pair of bloom filters ('new' and 'old')
1089 * in which the filehandles of recalled delegations are "stored".
1090 * If a filehandle appear in either filter, a delegation is blocked.
1091 * When a delegation is recalled, the filehandle is stored in the "new"
1093 * Every 30 seconds we swap the filters and clear the "new" one,
1094 * unless both are empty of course.
1096 * Each filter is 256 bits. We hash the filehandle to 32bit and use the
1097 * low 3 bytes as hash-table indices.
1099 * 'blocked_delegations_lock', which is always taken in block_delegations(),
1100 * is used to manage concurrent access. Testing does not need the lock
1101 * except when swapping the two filters.
1103 static DEFINE_SPINLOCK(blocked_delegations_lock);
1104 static struct bloom_pair {
1105 int entries, old_entries;
1107 int new; /* index into 'set' */
1108 DECLARE_BITMAP(set[2], 256);
1109 } blocked_delegations;
1111 static int delegation_blocked(struct knfsd_fh *fh)
1114 struct bloom_pair *bd = &blocked_delegations;
1116 if (bd->entries == 0)
1118 if (ktime_get_seconds() - bd->swap_time > 30) {
1119 spin_lock(&blocked_delegations_lock);
1120 if (ktime_get_seconds() - bd->swap_time > 30) {
1121 bd->entries -= bd->old_entries;
1122 bd->old_entries = bd->entries;
1123 memset(bd->set[bd->new], 0,
1124 sizeof(bd->set[0]));
1125 bd->new = 1-bd->new;
1126 bd->swap_time = ktime_get_seconds();
1128 spin_unlock(&blocked_delegations_lock);
1130 hash = jhash(&fh->fh_raw, fh->fh_size, 0);
1131 if (test_bit(hash&255, bd->set[0]) &&
1132 test_bit((hash>>8)&255, bd->set[0]) &&
1133 test_bit((hash>>16)&255, bd->set[0]))
1136 if (test_bit(hash&255, bd->set[1]) &&
1137 test_bit((hash>>8)&255, bd->set[1]) &&
1138 test_bit((hash>>16)&255, bd->set[1]))
1144 static void block_delegations(struct knfsd_fh *fh)
1147 struct bloom_pair *bd = &blocked_delegations;
1149 hash = jhash(&fh->fh_raw, fh->fh_size, 0);
1151 spin_lock(&blocked_delegations_lock);
1152 __set_bit(hash&255, bd->set[bd->new]);
1153 __set_bit((hash>>8)&255, bd->set[bd->new]);
1154 __set_bit((hash>>16)&255, bd->set[bd->new]);
1155 if (bd->entries == 0)
1156 bd->swap_time = ktime_get_seconds();
1158 spin_unlock(&blocked_delegations_lock);
1161 static struct nfs4_delegation *
1162 alloc_init_deleg(struct nfs4_client *clp, struct nfs4_file *fp,
1163 struct nfs4_clnt_odstate *odstate)
1165 struct nfs4_delegation *dp;
1168 dprintk("NFSD alloc_init_deleg\n");
1169 n = atomic_long_inc_return(&num_delegations);
1170 if (n < 0 || n > max_delegations)
1172 if (delegation_blocked(&fp->fi_fhandle))
1174 dp = delegstateid(nfs4_alloc_stid(clp, deleg_slab, nfs4_free_deleg));
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 = NFS4_OPEN_DELEGATE_READ;
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);
1195 dp->dl_stid.sc_file = fp;
1198 atomic_long_dec(&num_delegations);
1203 nfs4_put_stid(struct nfs4_stid *s)
1205 struct nfs4_file *fp = s->sc_file;
1206 struct nfs4_client *clp = s->sc_client;
1208 might_lock(&clp->cl_lock);
1210 if (!refcount_dec_and_lock(&s->sc_count, &clp->cl_lock)) {
1211 wake_up_all(&close_wq);
1214 idr_remove(&clp->cl_stateids, s->sc_stateid.si_opaque.so_id);
1215 nfs4_free_cpntf_statelist(clp->net, s);
1216 spin_unlock(&clp->cl_lock);
1223 nfs4_inc_and_copy_stateid(stateid_t *dst, struct nfs4_stid *stid)
1225 stateid_t *src = &stid->sc_stateid;
1227 spin_lock(&stid->sc_lock);
1228 if (unlikely(++src->si_generation == 0))
1229 src->si_generation = 1;
1230 memcpy(dst, src, sizeof(*dst));
1231 spin_unlock(&stid->sc_lock);
1234 static void put_deleg_file(struct nfs4_file *fp)
1236 struct nfsd_file *nf = NULL;
1238 spin_lock(&fp->fi_lock);
1239 if (--fp->fi_delegees == 0)
1240 swap(nf, fp->fi_deleg_file);
1241 spin_unlock(&fp->fi_lock);
1247 static void nfs4_unlock_deleg_lease(struct nfs4_delegation *dp)
1249 struct nfs4_file *fp = dp->dl_stid.sc_file;
1250 struct nfsd_file *nf = fp->fi_deleg_file;
1252 WARN_ON_ONCE(!fp->fi_delegees);
1254 vfs_setlease(nf->nf_file, F_UNLCK, NULL, (void **)&dp);
1258 static void destroy_unhashed_deleg(struct nfs4_delegation *dp)
1260 put_clnt_odstate(dp->dl_clnt_odstate);
1261 nfs4_unlock_deleg_lease(dp);
1262 nfs4_put_stid(&dp->dl_stid);
1265 void nfs4_unhash_stid(struct nfs4_stid *s)
1271 * nfs4_delegation_exists - Discover if this delegation already exists
1272 * @clp: a pointer to the nfs4_client we're granting a delegation to
1273 * @fp: a pointer to the nfs4_file we're granting a delegation on
1276 * On success: true iff an existing delegation is found
1280 nfs4_delegation_exists(struct nfs4_client *clp, struct nfs4_file *fp)
1282 struct nfs4_delegation *searchdp = NULL;
1283 struct nfs4_client *searchclp = NULL;
1285 lockdep_assert_held(&state_lock);
1286 lockdep_assert_held(&fp->fi_lock);
1288 list_for_each_entry(searchdp, &fp->fi_delegations, dl_perfile) {
1289 searchclp = searchdp->dl_stid.sc_client;
1290 if (clp == searchclp) {
1298 * hash_delegation_locked - Add a delegation to the appropriate lists
1299 * @dp: a pointer to the nfs4_delegation we are adding.
1300 * @fp: a pointer to the nfs4_file we're granting a delegation on
1303 * On success: NULL if the delegation was successfully hashed.
1305 * On error: -EAGAIN if one was previously granted to this
1306 * nfs4_client for this nfs4_file. Delegation is not hashed.
1311 hash_delegation_locked(struct nfs4_delegation *dp, struct nfs4_file *fp)
1313 struct nfs4_client *clp = dp->dl_stid.sc_client;
1315 lockdep_assert_held(&state_lock);
1316 lockdep_assert_held(&fp->fi_lock);
1318 if (nfs4_delegation_exists(clp, fp))
1320 refcount_inc(&dp->dl_stid.sc_count);
1321 dp->dl_stid.sc_type = NFS4_DELEG_STID;
1322 list_add(&dp->dl_perfile, &fp->fi_delegations);
1323 list_add(&dp->dl_perclnt, &clp->cl_delegations);
1327 static bool delegation_hashed(struct nfs4_delegation *dp)
1329 return !(list_empty(&dp->dl_perfile));
1333 unhash_delegation_locked(struct nfs4_delegation *dp)
1335 struct nfs4_file *fp = dp->dl_stid.sc_file;
1337 lockdep_assert_held(&state_lock);
1339 if (!delegation_hashed(dp))
1342 dp->dl_stid.sc_type = NFS4_CLOSED_DELEG_STID;
1343 /* Ensure that deleg break won't try to requeue it */
1345 spin_lock(&fp->fi_lock);
1346 list_del_init(&dp->dl_perclnt);
1347 list_del_init(&dp->dl_recall_lru);
1348 list_del_init(&dp->dl_perfile);
1349 spin_unlock(&fp->fi_lock);
1353 static void destroy_delegation(struct nfs4_delegation *dp)
1357 spin_lock(&state_lock);
1358 unhashed = unhash_delegation_locked(dp);
1359 spin_unlock(&state_lock);
1361 destroy_unhashed_deleg(dp);
1364 static void revoke_delegation(struct nfs4_delegation *dp)
1366 struct nfs4_client *clp = dp->dl_stid.sc_client;
1368 WARN_ON(!list_empty(&dp->dl_recall_lru));
1370 if (clp->cl_minorversion) {
1371 dp->dl_stid.sc_type = NFS4_REVOKED_DELEG_STID;
1372 refcount_inc(&dp->dl_stid.sc_count);
1373 spin_lock(&clp->cl_lock);
1374 list_add(&dp->dl_recall_lru, &clp->cl_revoked);
1375 spin_unlock(&clp->cl_lock);
1377 destroy_unhashed_deleg(dp);
1384 static unsigned int clientid_hashval(u32 id)
1386 return id & CLIENT_HASH_MASK;
1389 static unsigned int clientstr_hashval(struct xdr_netobj name)
1391 return opaque_hashval(name.data, 8) & CLIENT_HASH_MASK;
1395 * A stateid that had a deny mode associated with it is being released
1396 * or downgraded. Recalculate the deny mode on the file.
1399 recalculate_deny_mode(struct nfs4_file *fp)
1401 struct nfs4_ol_stateid *stp;
1403 spin_lock(&fp->fi_lock);
1404 fp->fi_share_deny = 0;
1405 list_for_each_entry(stp, &fp->fi_stateids, st_perfile)
1406 fp->fi_share_deny |= bmap_to_share_mode(stp->st_deny_bmap);
1407 spin_unlock(&fp->fi_lock);
1411 reset_union_bmap_deny(u32 deny, struct nfs4_ol_stateid *stp)
1414 bool change = false;
1416 for (i = 1; i < 4; i++) {
1417 if ((i & deny) != i) {
1423 /* Recalculate per-file deny mode if there was a change */
1425 recalculate_deny_mode(stp->st_stid.sc_file);
1428 /* release all access and file references for a given stateid */
1430 release_all_access(struct nfs4_ol_stateid *stp)
1433 struct nfs4_file *fp = stp->st_stid.sc_file;
1435 if (fp && stp->st_deny_bmap != 0)
1436 recalculate_deny_mode(fp);
1438 for (i = 1; i < 4; i++) {
1439 if (test_access(i, stp))
1440 nfs4_file_put_access(stp->st_stid.sc_file, i);
1441 clear_access(i, stp);
1445 static inline void nfs4_free_stateowner(struct nfs4_stateowner *sop)
1447 kfree(sop->so_owner.data);
1448 sop->so_ops->so_free(sop);
1451 static void nfs4_put_stateowner(struct nfs4_stateowner *sop)
1453 struct nfs4_client *clp = sop->so_client;
1455 might_lock(&clp->cl_lock);
1457 if (!atomic_dec_and_lock(&sop->so_count, &clp->cl_lock))
1459 sop->so_ops->so_unhash(sop);
1460 spin_unlock(&clp->cl_lock);
1461 nfs4_free_stateowner(sop);
1465 nfs4_ol_stateid_unhashed(const struct nfs4_ol_stateid *stp)
1467 return list_empty(&stp->st_perfile);
1470 static bool unhash_ol_stateid(struct nfs4_ol_stateid *stp)
1472 struct nfs4_file *fp = stp->st_stid.sc_file;
1474 lockdep_assert_held(&stp->st_stateowner->so_client->cl_lock);
1476 if (list_empty(&stp->st_perfile))
1479 spin_lock(&fp->fi_lock);
1480 list_del_init(&stp->st_perfile);
1481 spin_unlock(&fp->fi_lock);
1482 list_del(&stp->st_perstateowner);
1486 static void nfs4_free_ol_stateid(struct nfs4_stid *stid)
1488 struct nfs4_ol_stateid *stp = openlockstateid(stid);
1490 put_clnt_odstate(stp->st_clnt_odstate);
1491 release_all_access(stp);
1492 if (stp->st_stateowner)
1493 nfs4_put_stateowner(stp->st_stateowner);
1494 WARN_ON(!list_empty(&stid->sc_cp_list));
1495 kmem_cache_free(stateid_slab, stid);
1498 static void nfs4_free_lock_stateid(struct nfs4_stid *stid)
1500 struct nfs4_ol_stateid *stp = openlockstateid(stid);
1501 struct nfs4_lockowner *lo = lockowner(stp->st_stateowner);
1502 struct nfsd_file *nf;
1504 nf = find_any_file(stp->st_stid.sc_file);
1506 get_file(nf->nf_file);
1507 filp_close(nf->nf_file, (fl_owner_t)lo);
1510 nfs4_free_ol_stateid(stid);
1514 * Put the persistent reference to an already unhashed generic stateid, while
1515 * holding the cl_lock. If it's the last reference, then put it onto the
1516 * reaplist for later destruction.
1518 static void put_ol_stateid_locked(struct nfs4_ol_stateid *stp,
1519 struct list_head *reaplist)
1521 struct nfs4_stid *s = &stp->st_stid;
1522 struct nfs4_client *clp = s->sc_client;
1524 lockdep_assert_held(&clp->cl_lock);
1526 WARN_ON_ONCE(!list_empty(&stp->st_locks));
1528 if (!refcount_dec_and_test(&s->sc_count)) {
1529 wake_up_all(&close_wq);
1533 idr_remove(&clp->cl_stateids, s->sc_stateid.si_opaque.so_id);
1534 list_add(&stp->st_locks, reaplist);
1537 static bool unhash_lock_stateid(struct nfs4_ol_stateid *stp)
1539 lockdep_assert_held(&stp->st_stid.sc_client->cl_lock);
1541 if (!unhash_ol_stateid(stp))
1543 list_del_init(&stp->st_locks);
1544 nfs4_unhash_stid(&stp->st_stid);
1548 static void release_lock_stateid(struct nfs4_ol_stateid *stp)
1550 struct nfs4_client *clp = stp->st_stid.sc_client;
1553 spin_lock(&clp->cl_lock);
1554 unhashed = unhash_lock_stateid(stp);
1555 spin_unlock(&clp->cl_lock);
1557 nfs4_put_stid(&stp->st_stid);
1560 static void unhash_lockowner_locked(struct nfs4_lockowner *lo)
1562 struct nfs4_client *clp = lo->lo_owner.so_client;
1564 lockdep_assert_held(&clp->cl_lock);
1566 list_del_init(&lo->lo_owner.so_strhash);
1570 * Free a list of generic stateids that were collected earlier after being
1574 free_ol_stateid_reaplist(struct list_head *reaplist)
1576 struct nfs4_ol_stateid *stp;
1577 struct nfs4_file *fp;
1581 while (!list_empty(reaplist)) {
1582 stp = list_first_entry(reaplist, struct nfs4_ol_stateid,
1584 list_del(&stp->st_locks);
1585 fp = stp->st_stid.sc_file;
1586 stp->st_stid.sc_free(&stp->st_stid);
1592 static void release_open_stateid_locks(struct nfs4_ol_stateid *open_stp,
1593 struct list_head *reaplist)
1595 struct nfs4_ol_stateid *stp;
1597 lockdep_assert_held(&open_stp->st_stid.sc_client->cl_lock);
1599 while (!list_empty(&open_stp->st_locks)) {
1600 stp = list_entry(open_stp->st_locks.next,
1601 struct nfs4_ol_stateid, st_locks);
1602 WARN_ON(!unhash_lock_stateid(stp));
1603 put_ol_stateid_locked(stp, reaplist);
1607 static bool unhash_open_stateid(struct nfs4_ol_stateid *stp,
1608 struct list_head *reaplist)
1610 lockdep_assert_held(&stp->st_stid.sc_client->cl_lock);
1612 if (!unhash_ol_stateid(stp))
1614 release_open_stateid_locks(stp, reaplist);
1618 static void release_open_stateid(struct nfs4_ol_stateid *stp)
1620 LIST_HEAD(reaplist);
1622 spin_lock(&stp->st_stid.sc_client->cl_lock);
1623 if (unhash_open_stateid(stp, &reaplist))
1624 put_ol_stateid_locked(stp, &reaplist);
1625 spin_unlock(&stp->st_stid.sc_client->cl_lock);
1626 free_ol_stateid_reaplist(&reaplist);
1629 static void unhash_openowner_locked(struct nfs4_openowner *oo)
1631 struct nfs4_client *clp = oo->oo_owner.so_client;
1633 lockdep_assert_held(&clp->cl_lock);
1635 list_del_init(&oo->oo_owner.so_strhash);
1636 list_del_init(&oo->oo_perclient);
1639 static void release_last_closed_stateid(struct nfs4_openowner *oo)
1641 struct nfsd_net *nn = net_generic(oo->oo_owner.so_client->net,
1643 struct nfs4_ol_stateid *s;
1645 spin_lock(&nn->client_lock);
1646 s = oo->oo_last_closed_stid;
1648 list_del_init(&oo->oo_close_lru);
1649 oo->oo_last_closed_stid = NULL;
1651 spin_unlock(&nn->client_lock);
1653 nfs4_put_stid(&s->st_stid);
1656 static void release_openowner(struct nfs4_openowner *oo)
1658 struct nfs4_ol_stateid *stp;
1659 struct nfs4_client *clp = oo->oo_owner.so_client;
1660 struct list_head reaplist;
1662 INIT_LIST_HEAD(&reaplist);
1664 spin_lock(&clp->cl_lock);
1665 unhash_openowner_locked(oo);
1666 while (!list_empty(&oo->oo_owner.so_stateids)) {
1667 stp = list_first_entry(&oo->oo_owner.so_stateids,
1668 struct nfs4_ol_stateid, st_perstateowner);
1669 if (unhash_open_stateid(stp, &reaplist))
1670 put_ol_stateid_locked(stp, &reaplist);
1672 spin_unlock(&clp->cl_lock);
1673 free_ol_stateid_reaplist(&reaplist);
1674 release_last_closed_stateid(oo);
1675 nfs4_put_stateowner(&oo->oo_owner);
1679 hash_sessionid(struct nfs4_sessionid *sessionid)
1681 struct nfsd4_sessionid *sid = (struct nfsd4_sessionid *)sessionid;
1683 return sid->sequence % SESSION_HASH_SIZE;
1686 #ifdef CONFIG_SUNRPC_DEBUG
1688 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
1690 u32 *ptr = (u32 *)(&sessionid->data[0]);
1691 dprintk("%s: %u:%u:%u:%u\n", fn, ptr[0], ptr[1], ptr[2], ptr[3]);
1695 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
1701 * Bump the seqid on cstate->replay_owner, and clear replay_owner if it
1702 * won't be used for replay.
1704 void nfsd4_bump_seqid(struct nfsd4_compound_state *cstate, __be32 nfserr)
1706 struct nfs4_stateowner *so = cstate->replay_owner;
1708 if (nfserr == nfserr_replay_me)
1711 if (!seqid_mutating_err(ntohl(nfserr))) {
1712 nfsd4_cstate_clear_replay(cstate);
1717 if (so->so_is_open_owner)
1718 release_last_closed_stateid(openowner(so));
1724 gen_sessionid(struct nfsd4_session *ses)
1726 struct nfs4_client *clp = ses->se_client;
1727 struct nfsd4_sessionid *sid;
1729 sid = (struct nfsd4_sessionid *)ses->se_sessionid.data;
1730 sid->clientid = clp->cl_clientid;
1731 sid->sequence = current_sessionid++;
1736 * The protocol defines ca_maxresponssize_cached to include the size of
1737 * the rpc header, but all we need to cache is the data starting after
1738 * the end of the initial SEQUENCE operation--the rest we regenerate
1739 * each time. Therefore we can advertise a ca_maxresponssize_cached
1740 * value that is the number of bytes in our cache plus a few additional
1741 * bytes. In order to stay on the safe side, and not promise more than
1742 * we can cache, those additional bytes must be the minimum possible: 24
1743 * bytes of rpc header (xid through accept state, with AUTH_NULL
1744 * verifier), 12 for the compound header (with zero-length tag), and 44
1745 * for the SEQUENCE op response:
1747 #define NFSD_MIN_HDR_SEQ_SZ (24 + 12 + 44)
1750 free_session_slots(struct nfsd4_session *ses)
1754 for (i = 0; i < ses->se_fchannel.maxreqs; i++) {
1755 free_svc_cred(&ses->se_slots[i]->sl_cred);
1756 kfree(ses->se_slots[i]);
1761 * We don't actually need to cache the rpc and session headers, so we
1762 * can allocate a little less for each slot:
1764 static inline u32 slot_bytes(struct nfsd4_channel_attrs *ca)
1768 if (ca->maxresp_cached < NFSD_MIN_HDR_SEQ_SZ)
1771 size = ca->maxresp_cached - NFSD_MIN_HDR_SEQ_SZ;
1772 return size + sizeof(struct nfsd4_slot);
1776 * XXX: If we run out of reserved DRC memory we could (up to a point)
1777 * re-negotiate active sessions and reduce their slot usage to make
1778 * room for new connections. For now we just fail the create session.
1780 static u32 nfsd4_get_drc_mem(struct nfsd4_channel_attrs *ca, struct nfsd_net *nn)
1782 u32 slotsize = slot_bytes(ca);
1783 u32 num = ca->maxreqs;
1784 unsigned long avail, total_avail;
1785 unsigned int scale_factor;
1787 spin_lock(&nfsd_drc_lock);
1788 if (nfsd_drc_max_mem > nfsd_drc_mem_used)
1789 total_avail = nfsd_drc_max_mem - nfsd_drc_mem_used;
1791 /* We have handed out more space than we chose in
1792 * set_max_drc() to allow. That isn't really a
1793 * problem as long as that doesn't make us think we
1794 * have lots more due to integer overflow.
1797 avail = min((unsigned long)NFSD_MAX_MEM_PER_SESSION, total_avail);
1799 * Never use more than a fraction of the remaining memory,
1800 * unless it's the only way to give this client a slot.
1801 * The chosen fraction is either 1/8 or 1/number of threads,
1802 * whichever is smaller. This ensures there are adequate
1803 * slots to support multiple clients per thread.
1804 * Give the client one slot even if that would require
1805 * over-allocation--it is better than failure.
1807 scale_factor = max_t(unsigned int, 8, nn->nfsd_serv->sv_nrthreads);
1809 avail = clamp_t(unsigned long, avail, slotsize,
1810 total_avail/scale_factor);
1811 num = min_t(int, num, avail / slotsize);
1812 num = max_t(int, num, 1);
1813 nfsd_drc_mem_used += num * slotsize;
1814 spin_unlock(&nfsd_drc_lock);
1819 static void nfsd4_put_drc_mem(struct nfsd4_channel_attrs *ca)
1821 int slotsize = slot_bytes(ca);
1823 spin_lock(&nfsd_drc_lock);
1824 nfsd_drc_mem_used -= slotsize * ca->maxreqs;
1825 spin_unlock(&nfsd_drc_lock);
1828 static struct nfsd4_session *alloc_session(struct nfsd4_channel_attrs *fattrs,
1829 struct nfsd4_channel_attrs *battrs)
1831 int numslots = fattrs->maxreqs;
1832 int slotsize = slot_bytes(fattrs);
1833 struct nfsd4_session *new;
1836 BUILD_BUG_ON(NFSD_MAX_SLOTS_PER_SESSION * sizeof(struct nfsd4_slot *)
1837 + sizeof(struct nfsd4_session) > PAGE_SIZE);
1838 mem = numslots * sizeof(struct nfsd4_slot *);
1840 new = kzalloc(sizeof(*new) + mem, GFP_KERNEL);
1843 /* allocate each struct nfsd4_slot and data cache in one piece */
1844 for (i = 0; i < numslots; i++) {
1845 new->se_slots[i] = kzalloc(slotsize, GFP_KERNEL);
1846 if (!new->se_slots[i])
1850 memcpy(&new->se_fchannel, fattrs, sizeof(struct nfsd4_channel_attrs));
1851 memcpy(&new->se_bchannel, battrs, sizeof(struct nfsd4_channel_attrs));
1856 kfree(new->se_slots[i]);
1861 static void free_conn(struct nfsd4_conn *c)
1863 svc_xprt_put(c->cn_xprt);
1867 static void nfsd4_conn_lost(struct svc_xpt_user *u)
1869 struct nfsd4_conn *c = container_of(u, struct nfsd4_conn, cn_xpt_user);
1870 struct nfs4_client *clp = c->cn_session->se_client;
1872 trace_nfsd_cb_lost(clp);
1874 spin_lock(&clp->cl_lock);
1875 if (!list_empty(&c->cn_persession)) {
1876 list_del(&c->cn_persession);
1879 nfsd4_probe_callback(clp);
1880 spin_unlock(&clp->cl_lock);
1883 static struct nfsd4_conn *alloc_conn(struct svc_rqst *rqstp, u32 flags)
1885 struct nfsd4_conn *conn;
1887 conn = kmalloc(sizeof(struct nfsd4_conn), GFP_KERNEL);
1890 svc_xprt_get(rqstp->rq_xprt);
1891 conn->cn_xprt = rqstp->rq_xprt;
1892 conn->cn_flags = flags;
1893 INIT_LIST_HEAD(&conn->cn_xpt_user.list);
1897 static void __nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
1899 conn->cn_session = ses;
1900 list_add(&conn->cn_persession, &ses->se_conns);
1903 static void nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
1905 struct nfs4_client *clp = ses->se_client;
1907 spin_lock(&clp->cl_lock);
1908 __nfsd4_hash_conn(conn, ses);
1909 spin_unlock(&clp->cl_lock);
1912 static int nfsd4_register_conn(struct nfsd4_conn *conn)
1914 conn->cn_xpt_user.callback = nfsd4_conn_lost;
1915 return register_xpt_user(conn->cn_xprt, &conn->cn_xpt_user);
1918 static void nfsd4_init_conn(struct svc_rqst *rqstp, struct nfsd4_conn *conn, struct nfsd4_session *ses)
1922 nfsd4_hash_conn(conn, ses);
1923 ret = nfsd4_register_conn(conn);
1925 /* oops; xprt is already down: */
1926 nfsd4_conn_lost(&conn->cn_xpt_user);
1927 /* We may have gained or lost a callback channel: */
1928 nfsd4_probe_callback_sync(ses->se_client);
1931 static struct nfsd4_conn *alloc_conn_from_crses(struct svc_rqst *rqstp, struct nfsd4_create_session *cses)
1933 u32 dir = NFS4_CDFC4_FORE;
1935 if (cses->flags & SESSION4_BACK_CHAN)
1936 dir |= NFS4_CDFC4_BACK;
1937 return alloc_conn(rqstp, dir);
1940 /* must be called under client_lock */
1941 static void nfsd4_del_conns(struct nfsd4_session *s)
1943 struct nfs4_client *clp = s->se_client;
1944 struct nfsd4_conn *c;
1946 spin_lock(&clp->cl_lock);
1947 while (!list_empty(&s->se_conns)) {
1948 c = list_first_entry(&s->se_conns, struct nfsd4_conn, cn_persession);
1949 list_del_init(&c->cn_persession);
1950 spin_unlock(&clp->cl_lock);
1952 unregister_xpt_user(c->cn_xprt, &c->cn_xpt_user);
1955 spin_lock(&clp->cl_lock);
1957 spin_unlock(&clp->cl_lock);
1960 static void __free_session(struct nfsd4_session *ses)
1962 free_session_slots(ses);
1966 static void free_session(struct nfsd4_session *ses)
1968 nfsd4_del_conns(ses);
1969 nfsd4_put_drc_mem(&ses->se_fchannel);
1970 __free_session(ses);
1973 static void init_session(struct svc_rqst *rqstp, struct nfsd4_session *new, struct nfs4_client *clp, struct nfsd4_create_session *cses)
1976 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
1978 new->se_client = clp;
1981 INIT_LIST_HEAD(&new->se_conns);
1983 new->se_cb_seq_nr = 1;
1984 new->se_flags = cses->flags;
1985 new->se_cb_prog = cses->callback_prog;
1986 new->se_cb_sec = cses->cb_sec;
1987 atomic_set(&new->se_ref, 0);
1988 idx = hash_sessionid(&new->se_sessionid);
1989 list_add(&new->se_hash, &nn->sessionid_hashtbl[idx]);
1990 spin_lock(&clp->cl_lock);
1991 list_add(&new->se_perclnt, &clp->cl_sessions);
1992 spin_unlock(&clp->cl_lock);
1995 struct sockaddr *sa = svc_addr(rqstp);
1997 * This is a little silly; with sessions there's no real
1998 * use for the callback address. Use the peer address
1999 * as a reasonable default for now, but consider fixing
2000 * the rpc client not to require an address in the
2003 rpc_copy_addr((struct sockaddr *)&clp->cl_cb_conn.cb_addr, sa);
2004 clp->cl_cb_conn.cb_addrlen = svc_addr_len(sa);
2008 /* caller must hold client_lock */
2009 static struct nfsd4_session *
2010 __find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid, struct net *net)
2012 struct nfsd4_session *elem;
2014 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
2016 lockdep_assert_held(&nn->client_lock);
2018 dump_sessionid(__func__, sessionid);
2019 idx = hash_sessionid(sessionid);
2020 /* Search in the appropriate list */
2021 list_for_each_entry(elem, &nn->sessionid_hashtbl[idx], se_hash) {
2022 if (!memcmp(elem->se_sessionid.data, sessionid->data,
2023 NFS4_MAX_SESSIONID_LEN)) {
2028 dprintk("%s: session not found\n", __func__);
2032 static struct nfsd4_session *
2033 find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid, struct net *net,
2036 struct nfsd4_session *session;
2037 __be32 status = nfserr_badsession;
2039 session = __find_in_sessionid_hashtbl(sessionid, net);
2042 status = nfsd4_get_session_locked(session);
2050 /* caller must hold client_lock */
2052 unhash_session(struct nfsd4_session *ses)
2054 struct nfs4_client *clp = ses->se_client;
2055 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2057 lockdep_assert_held(&nn->client_lock);
2059 list_del(&ses->se_hash);
2060 spin_lock(&ses->se_client->cl_lock);
2061 list_del(&ses->se_perclnt);
2062 spin_unlock(&ses->se_client->cl_lock);
2065 /* SETCLIENTID and SETCLIENTID_CONFIRM Helper functions */
2067 STALE_CLIENTID(clientid_t *clid, struct nfsd_net *nn)
2070 * We're assuming the clid was not given out from a boot
2071 * precisely 2^32 (about 136 years) before this one. That seems
2072 * a safe assumption:
2074 if (clid->cl_boot == (u32)nn->boot_time)
2076 trace_nfsd_clid_stale(clid);
2081 * XXX Should we use a slab cache ?
2082 * This type of memory management is somewhat inefficient, but we use it
2083 * anyway since SETCLIENTID is not a common operation.
2085 static struct nfs4_client *alloc_client(struct xdr_netobj name,
2086 struct nfsd_net *nn)
2088 struct nfs4_client *clp;
2091 if (atomic_read(&nn->nfs4_client_count) >= nn->nfs4_max_clients) {
2092 mod_delayed_work(laundry_wq, &nn->laundromat_work, 0);
2095 clp = kmem_cache_zalloc(client_slab, GFP_KERNEL);
2098 xdr_netobj_dup(&clp->cl_name, &name, GFP_KERNEL);
2099 if (clp->cl_name.data == NULL)
2101 clp->cl_ownerstr_hashtbl = kmalloc_array(OWNER_HASH_SIZE,
2102 sizeof(struct list_head),
2104 if (!clp->cl_ownerstr_hashtbl)
2105 goto err_no_hashtbl;
2106 for (i = 0; i < OWNER_HASH_SIZE; i++)
2107 INIT_LIST_HEAD(&clp->cl_ownerstr_hashtbl[i]);
2108 INIT_LIST_HEAD(&clp->cl_sessions);
2109 idr_init(&clp->cl_stateids);
2110 atomic_set(&clp->cl_rpc_users, 0);
2111 clp->cl_cb_state = NFSD4_CB_UNKNOWN;
2112 clp->cl_state = NFSD4_ACTIVE;
2113 atomic_inc(&nn->nfs4_client_count);
2114 atomic_set(&clp->cl_delegs_in_recall, 0);
2115 INIT_LIST_HEAD(&clp->cl_idhash);
2116 INIT_LIST_HEAD(&clp->cl_openowners);
2117 INIT_LIST_HEAD(&clp->cl_delegations);
2118 INIT_LIST_HEAD(&clp->cl_lru);
2119 INIT_LIST_HEAD(&clp->cl_revoked);
2120 #ifdef CONFIG_NFSD_PNFS
2121 INIT_LIST_HEAD(&clp->cl_lo_states);
2123 INIT_LIST_HEAD(&clp->async_copies);
2124 spin_lock_init(&clp->async_lock);
2125 spin_lock_init(&clp->cl_lock);
2126 rpc_init_wait_queue(&clp->cl_cb_waitq, "Backchannel slot table");
2129 kfree(clp->cl_name.data);
2131 kmem_cache_free(client_slab, clp);
2135 static void __free_client(struct kref *k)
2137 struct nfsdfs_client *c = container_of(k, struct nfsdfs_client, cl_ref);
2138 struct nfs4_client *clp = container_of(c, struct nfs4_client, cl_nfsdfs);
2140 free_svc_cred(&clp->cl_cred);
2141 kfree(clp->cl_ownerstr_hashtbl);
2142 kfree(clp->cl_name.data);
2143 kfree(clp->cl_nii_domain.data);
2144 kfree(clp->cl_nii_name.data);
2145 idr_destroy(&clp->cl_stateids);
2146 kmem_cache_free(client_slab, clp);
2149 static void drop_client(struct nfs4_client *clp)
2151 kref_put(&clp->cl_nfsdfs.cl_ref, __free_client);
2155 free_client(struct nfs4_client *clp)
2157 while (!list_empty(&clp->cl_sessions)) {
2158 struct nfsd4_session *ses;
2159 ses = list_entry(clp->cl_sessions.next, struct nfsd4_session,
2161 list_del(&ses->se_perclnt);
2162 WARN_ON_ONCE(atomic_read(&ses->se_ref));
2165 rpc_destroy_wait_queue(&clp->cl_cb_waitq);
2166 if (clp->cl_nfsd_dentry) {
2167 nfsd_client_rmdir(clp->cl_nfsd_dentry);
2168 clp->cl_nfsd_dentry = NULL;
2169 wake_up_all(&expiry_wq);
2174 /* must be called under the client_lock */
2176 unhash_client_locked(struct nfs4_client *clp)
2178 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2179 struct nfsd4_session *ses;
2181 lockdep_assert_held(&nn->client_lock);
2183 /* Mark the client as expired! */
2185 /* Make it invisible */
2186 if (!list_empty(&clp->cl_idhash)) {
2187 list_del_init(&clp->cl_idhash);
2188 if (test_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags))
2189 rb_erase(&clp->cl_namenode, &nn->conf_name_tree);
2191 rb_erase(&clp->cl_namenode, &nn->unconf_name_tree);
2193 list_del_init(&clp->cl_lru);
2194 spin_lock(&clp->cl_lock);
2195 list_for_each_entry(ses, &clp->cl_sessions, se_perclnt)
2196 list_del_init(&ses->se_hash);
2197 spin_unlock(&clp->cl_lock);
2201 unhash_client(struct nfs4_client *clp)
2203 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2205 spin_lock(&nn->client_lock);
2206 unhash_client_locked(clp);
2207 spin_unlock(&nn->client_lock);
2210 static __be32 mark_client_expired_locked(struct nfs4_client *clp)
2212 if (atomic_read(&clp->cl_rpc_users))
2213 return nfserr_jukebox;
2214 unhash_client_locked(clp);
2219 __destroy_client(struct nfs4_client *clp)
2221 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2223 struct nfs4_openowner *oo;
2224 struct nfs4_delegation *dp;
2225 struct list_head reaplist;
2227 INIT_LIST_HEAD(&reaplist);
2228 spin_lock(&state_lock);
2229 while (!list_empty(&clp->cl_delegations)) {
2230 dp = list_entry(clp->cl_delegations.next, struct nfs4_delegation, dl_perclnt);
2231 WARN_ON(!unhash_delegation_locked(dp));
2232 list_add(&dp->dl_recall_lru, &reaplist);
2234 spin_unlock(&state_lock);
2235 while (!list_empty(&reaplist)) {
2236 dp = list_entry(reaplist.next, struct nfs4_delegation, dl_recall_lru);
2237 list_del_init(&dp->dl_recall_lru);
2238 destroy_unhashed_deleg(dp);
2240 while (!list_empty(&clp->cl_revoked)) {
2241 dp = list_entry(clp->cl_revoked.next, struct nfs4_delegation, dl_recall_lru);
2242 list_del_init(&dp->dl_recall_lru);
2243 nfs4_put_stid(&dp->dl_stid);
2245 while (!list_empty(&clp->cl_openowners)) {
2246 oo = list_entry(clp->cl_openowners.next, struct nfs4_openowner, oo_perclient);
2247 nfs4_get_stateowner(&oo->oo_owner);
2248 release_openowner(oo);
2250 for (i = 0; i < OWNER_HASH_SIZE; i++) {
2251 struct nfs4_stateowner *so, *tmp;
2253 list_for_each_entry_safe(so, tmp, &clp->cl_ownerstr_hashtbl[i],
2255 /* Should be no openowners at this point */
2256 WARN_ON_ONCE(so->so_is_open_owner);
2257 remove_blocked_locks(lockowner(so));
2260 nfsd4_return_all_client_layouts(clp);
2261 nfsd4_shutdown_copy(clp);
2262 nfsd4_shutdown_callback(clp);
2263 if (clp->cl_cb_conn.cb_xprt)
2264 svc_xprt_put(clp->cl_cb_conn.cb_xprt);
2265 atomic_add_unless(&nn->nfs4_client_count, -1, 0);
2266 nfsd4_dec_courtesy_client_count(nn, clp);
2268 wake_up_all(&expiry_wq);
2272 destroy_client(struct nfs4_client *clp)
2275 __destroy_client(clp);
2278 static void inc_reclaim_complete(struct nfs4_client *clp)
2280 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2282 if (!nn->track_reclaim_completes)
2284 if (!nfsd4_find_reclaim_client(clp->cl_name, nn))
2286 if (atomic_inc_return(&nn->nr_reclaim_complete) ==
2287 nn->reclaim_str_hashtbl_size) {
2288 printk(KERN_INFO "NFSD: all clients done reclaiming, ending NFSv4 grace period (net %x)\n",
2290 nfsd4_end_grace(nn);
2294 static void expire_client(struct nfs4_client *clp)
2297 nfsd4_client_record_remove(clp);
2298 __destroy_client(clp);
2301 static void copy_verf(struct nfs4_client *target, nfs4_verifier *source)
2303 memcpy(target->cl_verifier.data, source->data,
2304 sizeof(target->cl_verifier.data));
2307 static void copy_clid(struct nfs4_client *target, struct nfs4_client *source)
2309 target->cl_clientid.cl_boot = source->cl_clientid.cl_boot;
2310 target->cl_clientid.cl_id = source->cl_clientid.cl_id;
2313 static int copy_cred(struct svc_cred *target, struct svc_cred *source)
2315 target->cr_principal = kstrdup(source->cr_principal, GFP_KERNEL);
2316 target->cr_raw_principal = kstrdup(source->cr_raw_principal,
2318 target->cr_targ_princ = kstrdup(source->cr_targ_princ, GFP_KERNEL);
2319 if ((source->cr_principal && !target->cr_principal) ||
2320 (source->cr_raw_principal && !target->cr_raw_principal) ||
2321 (source->cr_targ_princ && !target->cr_targ_princ))
2324 target->cr_flavor = source->cr_flavor;
2325 target->cr_uid = source->cr_uid;
2326 target->cr_gid = source->cr_gid;
2327 target->cr_group_info = source->cr_group_info;
2328 get_group_info(target->cr_group_info);
2329 target->cr_gss_mech = source->cr_gss_mech;
2330 if (source->cr_gss_mech)
2331 gss_mech_get(source->cr_gss_mech);
2336 compare_blob(const struct xdr_netobj *o1, const struct xdr_netobj *o2)
2338 if (o1->len < o2->len)
2340 if (o1->len > o2->len)
2342 return memcmp(o1->data, o2->data, o1->len);
2346 same_verf(nfs4_verifier *v1, nfs4_verifier *v2)
2348 return 0 == memcmp(v1->data, v2->data, sizeof(v1->data));
2352 same_clid(clientid_t *cl1, clientid_t *cl2)
2354 return (cl1->cl_boot == cl2->cl_boot) && (cl1->cl_id == cl2->cl_id);
2357 static bool groups_equal(struct group_info *g1, struct group_info *g2)
2361 if (g1->ngroups != g2->ngroups)
2363 for (i=0; i<g1->ngroups; i++)
2364 if (!gid_eq(g1->gid[i], g2->gid[i]))
2370 * RFC 3530 language requires clid_inuse be returned when the
2371 * "principal" associated with a requests differs from that previously
2372 * used. We use uid, gid's, and gss principal string as our best
2373 * approximation. We also don't want to allow non-gss use of a client
2374 * established using gss: in theory cr_principal should catch that
2375 * change, but in practice cr_principal can be null even in the gss case
2376 * since gssd doesn't always pass down a principal string.
2378 static bool is_gss_cred(struct svc_cred *cr)
2380 /* Is cr_flavor one of the gss "pseudoflavors"?: */
2381 return (cr->cr_flavor > RPC_AUTH_MAXFLAVOR);
2386 same_creds(struct svc_cred *cr1, struct svc_cred *cr2)
2388 if ((is_gss_cred(cr1) != is_gss_cred(cr2))
2389 || (!uid_eq(cr1->cr_uid, cr2->cr_uid))
2390 || (!gid_eq(cr1->cr_gid, cr2->cr_gid))
2391 || !groups_equal(cr1->cr_group_info, cr2->cr_group_info))
2393 /* XXX: check that cr_targ_princ fields match ? */
2394 if (cr1->cr_principal == cr2->cr_principal)
2396 if (!cr1->cr_principal || !cr2->cr_principal)
2398 return 0 == strcmp(cr1->cr_principal, cr2->cr_principal);
2401 static bool svc_rqst_integrity_protected(struct svc_rqst *rqstp)
2403 struct svc_cred *cr = &rqstp->rq_cred;
2406 if (!cr->cr_gss_mech)
2408 service = gss_pseudoflavor_to_service(cr->cr_gss_mech, cr->cr_flavor);
2409 return service == RPC_GSS_SVC_INTEGRITY ||
2410 service == RPC_GSS_SVC_PRIVACY;
2413 bool nfsd4_mach_creds_match(struct nfs4_client *cl, struct svc_rqst *rqstp)
2415 struct svc_cred *cr = &rqstp->rq_cred;
2417 if (!cl->cl_mach_cred)
2419 if (cl->cl_cred.cr_gss_mech != cr->cr_gss_mech)
2421 if (!svc_rqst_integrity_protected(rqstp))
2423 if (cl->cl_cred.cr_raw_principal)
2424 return 0 == strcmp(cl->cl_cred.cr_raw_principal,
2425 cr->cr_raw_principal);
2426 if (!cr->cr_principal)
2428 return 0 == strcmp(cl->cl_cred.cr_principal, cr->cr_principal);
2431 static void gen_confirm(struct nfs4_client *clp, struct nfsd_net *nn)
2436 * This is opaque to client, so no need to byte-swap. Use
2437 * __force to keep sparse happy
2439 verf[0] = (__force __be32)(u32)ktime_get_real_seconds();
2440 verf[1] = (__force __be32)nn->clverifier_counter++;
2441 memcpy(clp->cl_confirm.data, verf, sizeof(clp->cl_confirm.data));
2444 static void gen_clid(struct nfs4_client *clp, struct nfsd_net *nn)
2446 clp->cl_clientid.cl_boot = (u32)nn->boot_time;
2447 clp->cl_clientid.cl_id = nn->clientid_counter++;
2448 gen_confirm(clp, nn);
2451 static struct nfs4_stid *
2452 find_stateid_locked(struct nfs4_client *cl, stateid_t *t)
2454 struct nfs4_stid *ret;
2456 ret = idr_find(&cl->cl_stateids, t->si_opaque.so_id);
2457 if (!ret || !ret->sc_type)
2462 static struct nfs4_stid *
2463 find_stateid_by_type(struct nfs4_client *cl, stateid_t *t, char typemask)
2465 struct nfs4_stid *s;
2467 spin_lock(&cl->cl_lock);
2468 s = find_stateid_locked(cl, t);
2470 if (typemask & s->sc_type)
2471 refcount_inc(&s->sc_count);
2475 spin_unlock(&cl->cl_lock);
2479 static struct nfs4_client *get_nfsdfs_clp(struct inode *inode)
2481 struct nfsdfs_client *nc;
2482 nc = get_nfsdfs_client(inode);
2485 return container_of(nc, struct nfs4_client, cl_nfsdfs);
2488 static void seq_quote_mem(struct seq_file *m, char *data, int len)
2490 seq_printf(m, "\"");
2491 seq_escape_mem(m, data, len, ESCAPE_HEX | ESCAPE_NAP | ESCAPE_APPEND, "\"\\");
2492 seq_printf(m, "\"");
2495 static const char *cb_state2str(int state)
2500 case NFSD4_CB_UNKNOWN:
2504 case NFSD4_CB_FAULT:
2510 static int client_info_show(struct seq_file *m, void *v)
2512 struct inode *inode = file_inode(m->file);
2513 struct nfs4_client *clp;
2516 clp = get_nfsdfs_clp(inode);
2519 memcpy(&clid, &clp->cl_clientid, sizeof(clid));
2520 seq_printf(m, "clientid: 0x%llx\n", clid);
2521 seq_printf(m, "address: \"%pISpc\"\n", (struct sockaddr *)&clp->cl_addr);
2523 if (clp->cl_state == NFSD4_COURTESY)
2524 seq_puts(m, "status: courtesy\n");
2525 else if (clp->cl_state == NFSD4_EXPIRABLE)
2526 seq_puts(m, "status: expirable\n");
2527 else if (test_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags))
2528 seq_puts(m, "status: confirmed\n");
2530 seq_puts(m, "status: unconfirmed\n");
2531 seq_printf(m, "seconds from last renew: %lld\n",
2532 ktime_get_boottime_seconds() - clp->cl_time);
2533 seq_printf(m, "name: ");
2534 seq_quote_mem(m, clp->cl_name.data, clp->cl_name.len);
2535 seq_printf(m, "\nminor version: %d\n", clp->cl_minorversion);
2536 if (clp->cl_nii_domain.data) {
2537 seq_printf(m, "Implementation domain: ");
2538 seq_quote_mem(m, clp->cl_nii_domain.data,
2539 clp->cl_nii_domain.len);
2540 seq_printf(m, "\nImplementation name: ");
2541 seq_quote_mem(m, clp->cl_nii_name.data, clp->cl_nii_name.len);
2542 seq_printf(m, "\nImplementation time: [%lld, %ld]\n",
2543 clp->cl_nii_time.tv_sec, clp->cl_nii_time.tv_nsec);
2545 seq_printf(m, "callback state: %s\n", cb_state2str(clp->cl_cb_state));
2546 seq_printf(m, "callback address: %pISpc\n", &clp->cl_cb_conn.cb_addr);
2552 DEFINE_SHOW_ATTRIBUTE(client_info);
2554 static void *states_start(struct seq_file *s, loff_t *pos)
2555 __acquires(&clp->cl_lock)
2557 struct nfs4_client *clp = s->private;
2558 unsigned long id = *pos;
2561 spin_lock(&clp->cl_lock);
2562 ret = idr_get_next_ul(&clp->cl_stateids, &id);
2567 static void *states_next(struct seq_file *s, void *v, loff_t *pos)
2569 struct nfs4_client *clp = s->private;
2570 unsigned long id = *pos;
2575 ret = idr_get_next_ul(&clp->cl_stateids, &id);
2580 static void states_stop(struct seq_file *s, void *v)
2581 __releases(&clp->cl_lock)
2583 struct nfs4_client *clp = s->private;
2585 spin_unlock(&clp->cl_lock);
2588 static void nfs4_show_fname(struct seq_file *s, struct nfsd_file *f)
2590 seq_printf(s, "filename: \"%pD2\"", f->nf_file);
2593 static void nfs4_show_superblock(struct seq_file *s, struct nfsd_file *f)
2595 struct inode *inode = file_inode(f->nf_file);
2597 seq_printf(s, "superblock: \"%02x:%02x:%ld\"",
2598 MAJOR(inode->i_sb->s_dev),
2599 MINOR(inode->i_sb->s_dev),
2603 static void nfs4_show_owner(struct seq_file *s, struct nfs4_stateowner *oo)
2605 seq_printf(s, "owner: ");
2606 seq_quote_mem(s, oo->so_owner.data, oo->so_owner.len);
2609 static void nfs4_show_stateid(struct seq_file *s, stateid_t *stid)
2611 seq_printf(s, "0x%.8x", stid->si_generation);
2612 seq_printf(s, "%12phN", &stid->si_opaque);
2615 static int nfs4_show_open(struct seq_file *s, struct nfs4_stid *st)
2617 struct nfs4_ol_stateid *ols;
2618 struct nfs4_file *nf;
2619 struct nfsd_file *file;
2620 struct nfs4_stateowner *oo;
2621 unsigned int access, deny;
2623 if (st->sc_type != NFS4_OPEN_STID && st->sc_type != NFS4_LOCK_STID)
2624 return 0; /* XXX: or SEQ_SKIP? */
2625 ols = openlockstateid(st);
2626 oo = ols->st_stateowner;
2629 spin_lock(&nf->fi_lock);
2630 file = find_any_file_locked(nf);
2634 seq_printf(s, "- ");
2635 nfs4_show_stateid(s, &st->sc_stateid);
2636 seq_printf(s, ": { type: open, ");
2638 access = bmap_to_share_mode(ols->st_access_bmap);
2639 deny = bmap_to_share_mode(ols->st_deny_bmap);
2641 seq_printf(s, "access: %s%s, ",
2642 access & NFS4_SHARE_ACCESS_READ ? "r" : "-",
2643 access & NFS4_SHARE_ACCESS_WRITE ? "w" : "-");
2644 seq_printf(s, "deny: %s%s, ",
2645 deny & NFS4_SHARE_ACCESS_READ ? "r" : "-",
2646 deny & NFS4_SHARE_ACCESS_WRITE ? "w" : "-");
2648 nfs4_show_superblock(s, file);
2649 seq_printf(s, ", ");
2650 nfs4_show_fname(s, file);
2651 seq_printf(s, ", ");
2652 nfs4_show_owner(s, oo);
2653 seq_printf(s, " }\n");
2655 spin_unlock(&nf->fi_lock);
2659 static int nfs4_show_lock(struct seq_file *s, struct nfs4_stid *st)
2661 struct nfs4_ol_stateid *ols;
2662 struct nfs4_file *nf;
2663 struct nfsd_file *file;
2664 struct nfs4_stateowner *oo;
2666 ols = openlockstateid(st);
2667 oo = ols->st_stateowner;
2669 spin_lock(&nf->fi_lock);
2670 file = find_any_file_locked(nf);
2674 seq_printf(s, "- ");
2675 nfs4_show_stateid(s, &st->sc_stateid);
2676 seq_printf(s, ": { type: lock, ");
2679 * Note: a lock stateid isn't really the same thing as a lock,
2680 * it's the locking state held by one owner on a file, and there
2681 * may be multiple (or no) lock ranges associated with it.
2682 * (Same for the matter is true of open stateids.)
2685 nfs4_show_superblock(s, file);
2686 /* XXX: open stateid? */
2687 seq_printf(s, ", ");
2688 nfs4_show_fname(s, file);
2689 seq_printf(s, ", ");
2690 nfs4_show_owner(s, oo);
2691 seq_printf(s, " }\n");
2693 spin_unlock(&nf->fi_lock);
2697 static int nfs4_show_deleg(struct seq_file *s, struct nfs4_stid *st)
2699 struct nfs4_delegation *ds;
2700 struct nfs4_file *nf;
2701 struct nfsd_file *file;
2703 ds = delegstateid(st);
2705 spin_lock(&nf->fi_lock);
2706 file = find_deleg_file_locked(nf);
2710 seq_printf(s, "- ");
2711 nfs4_show_stateid(s, &st->sc_stateid);
2712 seq_printf(s, ": { type: deleg, ");
2714 /* Kinda dead code as long as we only support read delegs: */
2715 seq_printf(s, "access: %s, ",
2716 ds->dl_type == NFS4_OPEN_DELEGATE_READ ? "r" : "w");
2718 /* XXX: lease time, whether it's being recalled. */
2720 nfs4_show_superblock(s, file);
2721 seq_printf(s, ", ");
2722 nfs4_show_fname(s, file);
2723 seq_printf(s, " }\n");
2725 spin_unlock(&nf->fi_lock);
2729 static int nfs4_show_layout(struct seq_file *s, struct nfs4_stid *st)
2731 struct nfs4_layout_stateid *ls;
2732 struct nfsd_file *file;
2734 ls = container_of(st, struct nfs4_layout_stateid, ls_stid);
2737 seq_printf(s, "- ");
2738 nfs4_show_stateid(s, &st->sc_stateid);
2739 seq_printf(s, ": { type: layout, ");
2741 /* XXX: What else would be useful? */
2743 nfs4_show_superblock(s, file);
2744 seq_printf(s, ", ");
2745 nfs4_show_fname(s, file);
2746 seq_printf(s, " }\n");
2751 static int states_show(struct seq_file *s, void *v)
2753 struct nfs4_stid *st = v;
2755 switch (st->sc_type) {
2756 case NFS4_OPEN_STID:
2757 return nfs4_show_open(s, st);
2758 case NFS4_LOCK_STID:
2759 return nfs4_show_lock(s, st);
2760 case NFS4_DELEG_STID:
2761 return nfs4_show_deleg(s, st);
2762 case NFS4_LAYOUT_STID:
2763 return nfs4_show_layout(s, st);
2765 return 0; /* XXX: or SEQ_SKIP? */
2767 /* XXX: copy stateids? */
2770 static struct seq_operations states_seq_ops = {
2771 .start = states_start,
2772 .next = states_next,
2773 .stop = states_stop,
2777 static int client_states_open(struct inode *inode, struct file *file)
2780 struct nfs4_client *clp;
2783 clp = get_nfsdfs_clp(inode);
2787 ret = seq_open(file, &states_seq_ops);
2790 s = file->private_data;
2795 static int client_opens_release(struct inode *inode, struct file *file)
2797 struct seq_file *m = file->private_data;
2798 struct nfs4_client *clp = m->private;
2800 /* XXX: alternatively, we could get/drop in seq start/stop */
2805 static const struct file_operations client_states_fops = {
2806 .open = client_states_open,
2808 .llseek = seq_lseek,
2809 .release = client_opens_release,
2813 * Normally we refuse to destroy clients that are in use, but here the
2814 * administrator is telling us to just do it. We also want to wait
2815 * so the caller has a guarantee that the client's locks are gone by
2816 * the time the write returns:
2818 static void force_expire_client(struct nfs4_client *clp)
2820 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2821 bool already_expired;
2823 trace_nfsd_clid_admin_expired(&clp->cl_clientid);
2825 spin_lock(&nn->client_lock);
2827 spin_unlock(&nn->client_lock);
2829 wait_event(expiry_wq, atomic_read(&clp->cl_rpc_users) == 0);
2830 spin_lock(&nn->client_lock);
2831 already_expired = list_empty(&clp->cl_lru);
2832 if (!already_expired)
2833 unhash_client_locked(clp);
2834 spin_unlock(&nn->client_lock);
2836 if (!already_expired)
2839 wait_event(expiry_wq, clp->cl_nfsd_dentry == NULL);
2842 static ssize_t client_ctl_write(struct file *file, const char __user *buf,
2843 size_t size, loff_t *pos)
2846 struct nfs4_client *clp;
2848 data = simple_transaction_get(file, buf, size);
2850 return PTR_ERR(data);
2851 if (size != 7 || 0 != memcmp(data, "expire\n", 7))
2853 clp = get_nfsdfs_clp(file_inode(file));
2856 force_expire_client(clp);
2861 static const struct file_operations client_ctl_fops = {
2862 .write = client_ctl_write,
2863 .release = simple_transaction_release,
2866 static const struct tree_descr client_files[] = {
2867 [0] = {"info", &client_info_fops, S_IRUSR},
2868 [1] = {"states", &client_states_fops, S_IRUSR},
2869 [2] = {"ctl", &client_ctl_fops, S_IWUSR},
2873 static struct nfs4_client *create_client(struct xdr_netobj name,
2874 struct svc_rqst *rqstp, nfs4_verifier *verf)
2876 struct nfs4_client *clp;
2877 struct sockaddr *sa = svc_addr(rqstp);
2879 struct net *net = SVC_NET(rqstp);
2880 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
2881 struct dentry *dentries[ARRAY_SIZE(client_files)];
2883 clp = alloc_client(name, nn);
2887 ret = copy_cred(&clp->cl_cred, &rqstp->rq_cred);
2893 kref_init(&clp->cl_nfsdfs.cl_ref);
2894 nfsd4_init_cb(&clp->cl_cb_null, clp, NULL, NFSPROC4_CLNT_CB_NULL);
2895 clp->cl_time = ktime_get_boottime_seconds();
2896 clear_bit(0, &clp->cl_cb_slot_busy);
2897 copy_verf(clp, verf);
2898 memcpy(&clp->cl_addr, sa, sizeof(struct sockaddr_storage));
2899 clp->cl_cb_session = NULL;
2901 clp->cl_nfsd_dentry = nfsd_client_mkdir(
2902 nn, &clp->cl_nfsdfs,
2903 clp->cl_clientid.cl_id - nn->clientid_base,
2904 client_files, dentries);
2905 clp->cl_nfsd_info_dentry = dentries[0];
2906 if (!clp->cl_nfsd_dentry) {
2914 add_clp_to_name_tree(struct nfs4_client *new_clp, struct rb_root *root)
2916 struct rb_node **new = &(root->rb_node), *parent = NULL;
2917 struct nfs4_client *clp;
2920 clp = rb_entry(*new, struct nfs4_client, cl_namenode);
2923 if (compare_blob(&clp->cl_name, &new_clp->cl_name) > 0)
2924 new = &((*new)->rb_left);
2926 new = &((*new)->rb_right);
2929 rb_link_node(&new_clp->cl_namenode, parent, new);
2930 rb_insert_color(&new_clp->cl_namenode, root);
2933 static struct nfs4_client *
2934 find_clp_in_name_tree(struct xdr_netobj *name, struct rb_root *root)
2937 struct rb_node *node = root->rb_node;
2938 struct nfs4_client *clp;
2941 clp = rb_entry(node, struct nfs4_client, cl_namenode);
2942 cmp = compare_blob(&clp->cl_name, name);
2944 node = node->rb_left;
2946 node = node->rb_right;
2954 add_to_unconfirmed(struct nfs4_client *clp)
2956 unsigned int idhashval;
2957 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2959 lockdep_assert_held(&nn->client_lock);
2961 clear_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags);
2962 add_clp_to_name_tree(clp, &nn->unconf_name_tree);
2963 idhashval = clientid_hashval(clp->cl_clientid.cl_id);
2964 list_add(&clp->cl_idhash, &nn->unconf_id_hashtbl[idhashval]);
2965 renew_client_locked(clp);
2969 move_to_confirmed(struct nfs4_client *clp)
2971 unsigned int idhashval = clientid_hashval(clp->cl_clientid.cl_id);
2972 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2974 lockdep_assert_held(&nn->client_lock);
2976 list_move(&clp->cl_idhash, &nn->conf_id_hashtbl[idhashval]);
2977 rb_erase(&clp->cl_namenode, &nn->unconf_name_tree);
2978 add_clp_to_name_tree(clp, &nn->conf_name_tree);
2979 set_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags);
2980 trace_nfsd_clid_confirmed(&clp->cl_clientid);
2981 renew_client_locked(clp);
2984 static struct nfs4_client *
2985 find_client_in_id_table(struct list_head *tbl, clientid_t *clid, bool sessions)
2987 struct nfs4_client *clp;
2988 unsigned int idhashval = clientid_hashval(clid->cl_id);
2990 list_for_each_entry(clp, &tbl[idhashval], cl_idhash) {
2991 if (same_clid(&clp->cl_clientid, clid)) {
2992 if ((bool)clp->cl_minorversion != sessions)
2994 renew_client_locked(clp);
3001 static struct nfs4_client *
3002 find_confirmed_client(clientid_t *clid, bool sessions, struct nfsd_net *nn)
3004 struct list_head *tbl = nn->conf_id_hashtbl;
3006 lockdep_assert_held(&nn->client_lock);
3007 return find_client_in_id_table(tbl, clid, sessions);
3010 static struct nfs4_client *
3011 find_unconfirmed_client(clientid_t *clid, bool sessions, struct nfsd_net *nn)
3013 struct list_head *tbl = nn->unconf_id_hashtbl;
3015 lockdep_assert_held(&nn->client_lock);
3016 return find_client_in_id_table(tbl, clid, sessions);
3019 static bool clp_used_exchangeid(struct nfs4_client *clp)
3021 return clp->cl_exchange_flags != 0;
3024 static struct nfs4_client *
3025 find_confirmed_client_by_name(struct xdr_netobj *name, struct nfsd_net *nn)
3027 lockdep_assert_held(&nn->client_lock);
3028 return find_clp_in_name_tree(name, &nn->conf_name_tree);
3031 static struct nfs4_client *
3032 find_unconfirmed_client_by_name(struct xdr_netobj *name, struct nfsd_net *nn)
3034 lockdep_assert_held(&nn->client_lock);
3035 return find_clp_in_name_tree(name, &nn->unconf_name_tree);
3039 gen_callback(struct nfs4_client *clp, struct nfsd4_setclientid *se, struct svc_rqst *rqstp)
3041 struct nfs4_cb_conn *conn = &clp->cl_cb_conn;
3042 struct sockaddr *sa = svc_addr(rqstp);
3043 u32 scopeid = rpc_get_scope_id(sa);
3044 unsigned short expected_family;
3046 /* Currently, we only support tcp and tcp6 for the callback channel */
3047 if (se->se_callback_netid_len == 3 &&
3048 !memcmp(se->se_callback_netid_val, "tcp", 3))
3049 expected_family = AF_INET;
3050 else if (se->se_callback_netid_len == 4 &&
3051 !memcmp(se->se_callback_netid_val, "tcp6", 4))
3052 expected_family = AF_INET6;
3056 conn->cb_addrlen = rpc_uaddr2sockaddr(clp->net, se->se_callback_addr_val,
3057 se->se_callback_addr_len,
3058 (struct sockaddr *)&conn->cb_addr,
3059 sizeof(conn->cb_addr));
3061 if (!conn->cb_addrlen || conn->cb_addr.ss_family != expected_family)
3064 if (conn->cb_addr.ss_family == AF_INET6)
3065 ((struct sockaddr_in6 *)&conn->cb_addr)->sin6_scope_id = scopeid;
3067 conn->cb_prog = se->se_callback_prog;
3068 conn->cb_ident = se->se_callback_ident;
3069 memcpy(&conn->cb_saddr, &rqstp->rq_daddr, rqstp->rq_daddrlen);
3070 trace_nfsd_cb_args(clp, conn);
3073 conn->cb_addr.ss_family = AF_UNSPEC;
3074 conn->cb_addrlen = 0;
3075 trace_nfsd_cb_nodelegs(clp);
3080 * Cache a reply. nfsd4_check_resp_size() has bounded the cache size.
3083 nfsd4_store_cache_entry(struct nfsd4_compoundres *resp)
3085 struct xdr_buf *buf = resp->xdr->buf;
3086 struct nfsd4_slot *slot = resp->cstate.slot;
3089 dprintk("--> %s slot %p\n", __func__, slot);
3091 slot->sl_flags |= NFSD4_SLOT_INITIALIZED;
3092 slot->sl_opcnt = resp->opcnt;
3093 slot->sl_status = resp->cstate.status;
3094 free_svc_cred(&slot->sl_cred);
3095 copy_cred(&slot->sl_cred, &resp->rqstp->rq_cred);
3097 if (!nfsd4_cache_this(resp)) {
3098 slot->sl_flags &= ~NFSD4_SLOT_CACHED;
3101 slot->sl_flags |= NFSD4_SLOT_CACHED;
3103 base = resp->cstate.data_offset;
3104 slot->sl_datalen = buf->len - base;
3105 if (read_bytes_from_xdr_buf(buf, base, slot->sl_data, slot->sl_datalen))
3106 WARN(1, "%s: sessions DRC could not cache compound\n",
3112 * Encode the replay sequence operation from the slot values.
3113 * If cachethis is FALSE encode the uncached rep error on the next
3114 * operation which sets resp->p and increments resp->opcnt for
3115 * nfs4svc_encode_compoundres.
3119 nfsd4_enc_sequence_replay(struct nfsd4_compoundargs *args,
3120 struct nfsd4_compoundres *resp)
3122 struct nfsd4_op *op;
3123 struct nfsd4_slot *slot = resp->cstate.slot;
3125 /* Encode the replayed sequence operation */
3126 op = &args->ops[resp->opcnt - 1];
3127 nfsd4_encode_operation(resp, op);
3129 if (slot->sl_flags & NFSD4_SLOT_CACHED)
3131 if (args->opcnt == 1) {
3133 * The original operation wasn't a solo sequence--we
3134 * always cache those--so this retry must not match the
3137 op->status = nfserr_seq_false_retry;
3139 op = &args->ops[resp->opcnt++];
3140 op->status = nfserr_retry_uncached_rep;
3141 nfsd4_encode_operation(resp, op);
3147 * The sequence operation is not cached because we can use the slot and
3151 nfsd4_replay_cache_entry(struct nfsd4_compoundres *resp,
3152 struct nfsd4_sequence *seq)
3154 struct nfsd4_slot *slot = resp->cstate.slot;
3155 struct xdr_stream *xdr = resp->xdr;
3159 dprintk("--> %s slot %p\n", __func__, slot);
3161 status = nfsd4_enc_sequence_replay(resp->rqstp->rq_argp, resp);
3165 p = xdr_reserve_space(xdr, slot->sl_datalen);
3168 return nfserr_serverfault;
3170 xdr_encode_opaque_fixed(p, slot->sl_data, slot->sl_datalen);
3171 xdr_commit_encode(xdr);
3173 resp->opcnt = slot->sl_opcnt;
3174 return slot->sl_status;
3178 * Set the exchange_id flags returned by the server.
3181 nfsd4_set_ex_flags(struct nfs4_client *new, struct nfsd4_exchange_id *clid)
3183 #ifdef CONFIG_NFSD_PNFS
3184 new->cl_exchange_flags |= EXCHGID4_FLAG_USE_PNFS_MDS;
3186 new->cl_exchange_flags |= EXCHGID4_FLAG_USE_NON_PNFS;
3189 /* Referrals are supported, Migration is not. */
3190 new->cl_exchange_flags |= EXCHGID4_FLAG_SUPP_MOVED_REFER;
3192 /* set the wire flags to return to client. */
3193 clid->flags = new->cl_exchange_flags;
3196 static bool client_has_openowners(struct nfs4_client *clp)
3198 struct nfs4_openowner *oo;
3200 list_for_each_entry(oo, &clp->cl_openowners, oo_perclient) {
3201 if (!list_empty(&oo->oo_owner.so_stateids))
3207 static bool client_has_state(struct nfs4_client *clp)
3209 return client_has_openowners(clp)
3210 #ifdef CONFIG_NFSD_PNFS
3211 || !list_empty(&clp->cl_lo_states)
3213 || !list_empty(&clp->cl_delegations)
3214 || !list_empty(&clp->cl_sessions)
3215 || !list_empty(&clp->async_copies);
3218 static __be32 copy_impl_id(struct nfs4_client *clp,
3219 struct nfsd4_exchange_id *exid)
3221 if (!exid->nii_domain.data)
3223 xdr_netobj_dup(&clp->cl_nii_domain, &exid->nii_domain, GFP_KERNEL);
3224 if (!clp->cl_nii_domain.data)
3225 return nfserr_jukebox;
3226 xdr_netobj_dup(&clp->cl_nii_name, &exid->nii_name, GFP_KERNEL);
3227 if (!clp->cl_nii_name.data)
3228 return nfserr_jukebox;
3229 clp->cl_nii_time = exid->nii_time;
3234 nfsd4_exchange_id(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3235 union nfsd4_op_u *u)
3237 struct nfsd4_exchange_id *exid = &u->exchange_id;
3238 struct nfs4_client *conf, *new;
3239 struct nfs4_client *unconf = NULL;
3241 char addr_str[INET6_ADDRSTRLEN];
3242 nfs4_verifier verf = exid->verifier;
3243 struct sockaddr *sa = svc_addr(rqstp);
3244 bool update = exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A;
3245 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3247 rpc_ntop(sa, addr_str, sizeof(addr_str));
3248 dprintk("%s rqstp=%p exid=%p clname.len=%u clname.data=%p "
3249 "ip_addr=%s flags %x, spa_how %u\n",
3250 __func__, rqstp, exid, exid->clname.len, exid->clname.data,
3251 addr_str, exid->flags, exid->spa_how);
3253 if (exid->flags & ~EXCHGID4_FLAG_MASK_A)
3254 return nfserr_inval;
3256 new = create_client(exid->clname, rqstp, &verf);
3258 return nfserr_jukebox;
3259 status = copy_impl_id(new, exid);
3263 switch (exid->spa_how) {
3265 exid->spo_must_enforce[0] = 0;
3266 exid->spo_must_enforce[1] = (
3267 1 << (OP_BIND_CONN_TO_SESSION - 32) |
3268 1 << (OP_EXCHANGE_ID - 32) |
3269 1 << (OP_CREATE_SESSION - 32) |
3270 1 << (OP_DESTROY_SESSION - 32) |
3271 1 << (OP_DESTROY_CLIENTID - 32));
3273 exid->spo_must_allow[0] &= (1 << (OP_CLOSE) |
3274 1 << (OP_OPEN_DOWNGRADE) |
3276 1 << (OP_DELEGRETURN));
3278 exid->spo_must_allow[1] &= (
3279 1 << (OP_TEST_STATEID - 32) |
3280 1 << (OP_FREE_STATEID - 32));
3281 if (!svc_rqst_integrity_protected(rqstp)) {
3282 status = nfserr_inval;
3286 * Sometimes userspace doesn't give us a principal.
3287 * Which is a bug, really. Anyway, we can't enforce
3288 * MACH_CRED in that case, better to give up now:
3290 if (!new->cl_cred.cr_principal &&
3291 !new->cl_cred.cr_raw_principal) {
3292 status = nfserr_serverfault;
3295 new->cl_mach_cred = true;
3299 default: /* checked by xdr code */
3303 status = nfserr_encr_alg_unsupp;
3307 /* Cases below refer to rfc 5661 section 18.35.4: */
3308 spin_lock(&nn->client_lock);
3309 conf = find_confirmed_client_by_name(&exid->clname, nn);
3311 bool creds_match = same_creds(&conf->cl_cred, &rqstp->rq_cred);
3312 bool verfs_match = same_verf(&verf, &conf->cl_verifier);
3315 if (!clp_used_exchangeid(conf)) { /* buggy client */
3316 status = nfserr_inval;
3319 if (!nfsd4_mach_creds_match(conf, rqstp)) {
3320 status = nfserr_wrong_cred;
3323 if (!creds_match) { /* case 9 */
3324 status = nfserr_perm;
3327 if (!verfs_match) { /* case 8 */
3328 status = nfserr_not_same;
3332 exid->flags |= EXCHGID4_FLAG_CONFIRMED_R;
3333 trace_nfsd_clid_confirmed_r(conf);
3336 if (!creds_match) { /* case 3 */
3337 if (client_has_state(conf)) {
3338 status = nfserr_clid_inuse;
3339 trace_nfsd_clid_cred_mismatch(conf, rqstp);
3344 if (verfs_match) { /* case 2 */
3345 conf->cl_exchange_flags |= EXCHGID4_FLAG_CONFIRMED_R;
3346 trace_nfsd_clid_confirmed_r(conf);
3349 /* case 5, client reboot */
3350 trace_nfsd_clid_verf_mismatch(conf, rqstp, &verf);
3355 if (update) { /* case 7 */
3356 status = nfserr_noent;
3360 unconf = find_unconfirmed_client_by_name(&exid->clname, nn);
3361 if (unconf) /* case 4, possible retry or client restart */
3362 unhash_client_locked(unconf);
3364 /* case 1, new owner ID */
3365 trace_nfsd_clid_fresh(new);
3369 status = mark_client_expired_locked(conf);
3372 trace_nfsd_clid_replaced(&conf->cl_clientid);
3374 new->cl_minorversion = cstate->minorversion;
3375 new->cl_spo_must_allow.u.words[0] = exid->spo_must_allow[0];
3376 new->cl_spo_must_allow.u.words[1] = exid->spo_must_allow[1];
3378 add_to_unconfirmed(new);
3381 exid->clientid.cl_boot = conf->cl_clientid.cl_boot;
3382 exid->clientid.cl_id = conf->cl_clientid.cl_id;
3384 exid->seqid = conf->cl_cs_slot.sl_seqid + 1;
3385 nfsd4_set_ex_flags(conf, exid);
3387 dprintk("nfsd4_exchange_id seqid %d flags %x\n",
3388 conf->cl_cs_slot.sl_seqid, conf->cl_exchange_flags);
3392 spin_unlock(&nn->client_lock);
3397 trace_nfsd_clid_expire_unconf(&unconf->cl_clientid);
3398 expire_client(unconf);
3404 check_slot_seqid(u32 seqid, u32 slot_seqid, int slot_inuse)
3406 dprintk("%s enter. seqid %d slot_seqid %d\n", __func__, seqid,
3409 /* The slot is in use, and no response has been sent. */
3411 if (seqid == slot_seqid)
3412 return nfserr_jukebox;
3414 return nfserr_seq_misordered;
3416 /* Note unsigned 32-bit arithmetic handles wraparound: */
3417 if (likely(seqid == slot_seqid + 1))
3419 if (seqid == slot_seqid)
3420 return nfserr_replay_cache;
3421 return nfserr_seq_misordered;
3425 * Cache the create session result into the create session single DRC
3426 * slot cache by saving the xdr structure. sl_seqid has been set.
3427 * Do this for solo or embedded create session operations.
3430 nfsd4_cache_create_session(struct nfsd4_create_session *cr_ses,
3431 struct nfsd4_clid_slot *slot, __be32 nfserr)
3433 slot->sl_status = nfserr;
3434 memcpy(&slot->sl_cr_ses, cr_ses, sizeof(*cr_ses));
3438 nfsd4_replay_create_session(struct nfsd4_create_session *cr_ses,
3439 struct nfsd4_clid_slot *slot)
3441 memcpy(cr_ses, &slot->sl_cr_ses, sizeof(*cr_ses));
3442 return slot->sl_status;
3445 #define NFSD_MIN_REQ_HDR_SEQ_SZ ((\
3446 2 * 2 + /* credential,verifier: AUTH_NULL, length 0 */ \
3447 1 + /* MIN tag is length with zero, only length */ \
3448 3 + /* version, opcount, opcode */ \
3449 XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
3450 /* seqid, slotID, slotID, cache */ \
3451 4 ) * sizeof(__be32))
3453 #define NFSD_MIN_RESP_HDR_SEQ_SZ ((\
3454 2 + /* verifier: AUTH_NULL, length 0 */\
3456 1 + /* MIN tag is length with zero, only length */ \
3457 3 + /* opcount, opcode, opstatus*/ \
3458 XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
3459 /* seqid, slotID, slotID, slotID, status */ \
3460 5 ) * sizeof(__be32))
3462 static __be32 check_forechannel_attrs(struct nfsd4_channel_attrs *ca, struct nfsd_net *nn)
3464 u32 maxrpc = nn->nfsd_serv->sv_max_mesg;
3466 if (ca->maxreq_sz < NFSD_MIN_REQ_HDR_SEQ_SZ)
3467 return nfserr_toosmall;
3468 if (ca->maxresp_sz < NFSD_MIN_RESP_HDR_SEQ_SZ)
3469 return nfserr_toosmall;
3470 ca->headerpadsz = 0;
3471 ca->maxreq_sz = min_t(u32, ca->maxreq_sz, maxrpc);
3472 ca->maxresp_sz = min_t(u32, ca->maxresp_sz, maxrpc);
3473 ca->maxops = min_t(u32, ca->maxops, NFSD_MAX_OPS_PER_COMPOUND);
3474 ca->maxresp_cached = min_t(u32, ca->maxresp_cached,
3475 NFSD_SLOT_CACHE_SIZE + NFSD_MIN_HDR_SEQ_SZ);
3476 ca->maxreqs = min_t(u32, ca->maxreqs, NFSD_MAX_SLOTS_PER_SESSION);
3478 * Note decreasing slot size below client's request may make it
3479 * difficult for client to function correctly, whereas
3480 * decreasing the number of slots will (just?) affect
3481 * performance. When short on memory we therefore prefer to
3482 * decrease number of slots instead of their size. Clients that
3483 * request larger slots than they need will get poor results:
3484 * Note that we always allow at least one slot, because our
3485 * accounting is soft and provides no guarantees either way.
3487 ca->maxreqs = nfsd4_get_drc_mem(ca, nn);
3493 * Server's NFSv4.1 backchannel support is AUTH_SYS-only for now.
3494 * These are based on similar macros in linux/sunrpc/msg_prot.h .
3496 #define RPC_MAX_HEADER_WITH_AUTH_SYS \
3497 (RPC_CALLHDRSIZE + 2 * (2 + UNX_CALLSLACK))
3499 #define RPC_MAX_REPHEADER_WITH_AUTH_SYS \
3500 (RPC_REPHDRSIZE + (2 + NUL_REPLYSLACK))
3502 #define NFSD_CB_MAX_REQ_SZ ((NFS4_enc_cb_recall_sz + \
3503 RPC_MAX_HEADER_WITH_AUTH_SYS) * sizeof(__be32))
3504 #define NFSD_CB_MAX_RESP_SZ ((NFS4_dec_cb_recall_sz + \
3505 RPC_MAX_REPHEADER_WITH_AUTH_SYS) * \
3508 static __be32 check_backchannel_attrs(struct nfsd4_channel_attrs *ca)
3510 ca->headerpadsz = 0;
3512 if (ca->maxreq_sz < NFSD_CB_MAX_REQ_SZ)
3513 return nfserr_toosmall;
3514 if (ca->maxresp_sz < NFSD_CB_MAX_RESP_SZ)
3515 return nfserr_toosmall;
3516 ca->maxresp_cached = 0;
3518 return nfserr_toosmall;
3523 static __be32 nfsd4_check_cb_sec(struct nfsd4_cb_sec *cbs)
3525 switch (cbs->flavor) {
3531 * GSS case: the spec doesn't allow us to return this
3532 * error. But it also doesn't allow us not to support
3534 * I'd rather this fail hard than return some error the
3535 * client might think it can already handle:
3537 return nfserr_encr_alg_unsupp;
3542 nfsd4_create_session(struct svc_rqst *rqstp,
3543 struct nfsd4_compound_state *cstate, union nfsd4_op_u *u)
3545 struct nfsd4_create_session *cr_ses = &u->create_session;
3546 struct sockaddr *sa = svc_addr(rqstp);
3547 struct nfs4_client *conf, *unconf;
3548 struct nfs4_client *old = NULL;
3549 struct nfsd4_session *new;
3550 struct nfsd4_conn *conn;
3551 struct nfsd4_clid_slot *cs_slot = NULL;
3553 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3555 if (cr_ses->flags & ~SESSION4_FLAG_MASK_A)
3556 return nfserr_inval;
3557 status = nfsd4_check_cb_sec(&cr_ses->cb_sec);
3560 status = check_forechannel_attrs(&cr_ses->fore_channel, nn);
3563 status = check_backchannel_attrs(&cr_ses->back_channel);
3565 goto out_release_drc_mem;
3566 status = nfserr_jukebox;
3567 new = alloc_session(&cr_ses->fore_channel, &cr_ses->back_channel);
3569 goto out_release_drc_mem;
3570 conn = alloc_conn_from_crses(rqstp, cr_ses);
3572 goto out_free_session;
3574 spin_lock(&nn->client_lock);
3575 unconf = find_unconfirmed_client(&cr_ses->clientid, true, nn);
3576 conf = find_confirmed_client(&cr_ses->clientid, true, nn);
3577 WARN_ON_ONCE(conf && unconf);
3580 status = nfserr_wrong_cred;
3581 if (!nfsd4_mach_creds_match(conf, rqstp))
3583 cs_slot = &conf->cl_cs_slot;
3584 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
3586 if (status == nfserr_replay_cache)
3587 status = nfsd4_replay_create_session(cr_ses, cs_slot);
3590 } else if (unconf) {
3591 status = nfserr_clid_inuse;
3592 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred) ||
3593 !rpc_cmp_addr(sa, (struct sockaddr *) &unconf->cl_addr)) {
3594 trace_nfsd_clid_cred_mismatch(unconf, rqstp);
3597 status = nfserr_wrong_cred;
3598 if (!nfsd4_mach_creds_match(unconf, rqstp))
3600 cs_slot = &unconf->cl_cs_slot;
3601 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
3603 /* an unconfirmed replay returns misordered */
3604 status = nfserr_seq_misordered;
3607 old = find_confirmed_client_by_name(&unconf->cl_name, nn);
3609 status = mark_client_expired_locked(old);
3614 trace_nfsd_clid_replaced(&old->cl_clientid);
3616 move_to_confirmed(unconf);
3619 status = nfserr_stale_clientid;
3623 /* Persistent sessions are not supported */
3624 cr_ses->flags &= ~SESSION4_PERSIST;
3625 /* Upshifting from TCP to RDMA is not supported */
3626 cr_ses->flags &= ~SESSION4_RDMA;
3628 init_session(rqstp, new, conf, cr_ses);
3629 nfsd4_get_session_locked(new);
3631 memcpy(cr_ses->sessionid.data, new->se_sessionid.data,
3632 NFS4_MAX_SESSIONID_LEN);
3633 cs_slot->sl_seqid++;
3634 cr_ses->seqid = cs_slot->sl_seqid;
3636 /* cache solo and embedded create sessions under the client_lock */
3637 nfsd4_cache_create_session(cr_ses, cs_slot, status);
3638 spin_unlock(&nn->client_lock);
3640 fsnotify_dentry(conf->cl_nfsd_info_dentry, FS_MODIFY);
3641 /* init connection and backchannel */
3642 nfsd4_init_conn(rqstp, conn, new);
3643 nfsd4_put_session(new);
3648 spin_unlock(&nn->client_lock);
3653 __free_session(new);
3654 out_release_drc_mem:
3655 nfsd4_put_drc_mem(&cr_ses->fore_channel);
3659 static __be32 nfsd4_map_bcts_dir(u32 *dir)
3662 case NFS4_CDFC4_FORE:
3663 case NFS4_CDFC4_BACK:
3665 case NFS4_CDFC4_FORE_OR_BOTH:
3666 case NFS4_CDFC4_BACK_OR_BOTH:
3667 *dir = NFS4_CDFC4_BOTH;
3670 return nfserr_inval;
3673 __be32 nfsd4_backchannel_ctl(struct svc_rqst *rqstp,
3674 struct nfsd4_compound_state *cstate,
3675 union nfsd4_op_u *u)
3677 struct nfsd4_backchannel_ctl *bc = &u->backchannel_ctl;
3678 struct nfsd4_session *session = cstate->session;
3679 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3682 status = nfsd4_check_cb_sec(&bc->bc_cb_sec);
3685 spin_lock(&nn->client_lock);
3686 session->se_cb_prog = bc->bc_cb_program;
3687 session->se_cb_sec = bc->bc_cb_sec;
3688 spin_unlock(&nn->client_lock);
3690 nfsd4_probe_callback(session->se_client);
3695 static struct nfsd4_conn *__nfsd4_find_conn(struct svc_xprt *xpt, struct nfsd4_session *s)
3697 struct nfsd4_conn *c;
3699 list_for_each_entry(c, &s->se_conns, cn_persession) {
3700 if (c->cn_xprt == xpt) {
3707 static __be32 nfsd4_match_existing_connection(struct svc_rqst *rqst,
3708 struct nfsd4_session *session, u32 req, struct nfsd4_conn **conn)
3710 struct nfs4_client *clp = session->se_client;
3711 struct svc_xprt *xpt = rqst->rq_xprt;
3712 struct nfsd4_conn *c;
3715 /* Following the last paragraph of RFC 5661 Section 18.34.3: */
3716 spin_lock(&clp->cl_lock);
3717 c = __nfsd4_find_conn(xpt, session);
3719 status = nfserr_noent;
3720 else if (req == c->cn_flags)
3722 else if (req == NFS4_CDFC4_FORE_OR_BOTH &&
3723 c->cn_flags != NFS4_CDFC4_BACK)
3725 else if (req == NFS4_CDFC4_BACK_OR_BOTH &&
3726 c->cn_flags != NFS4_CDFC4_FORE)
3729 status = nfserr_inval;
3730 spin_unlock(&clp->cl_lock);
3731 if (status == nfs_ok && conn)
3736 __be32 nfsd4_bind_conn_to_session(struct svc_rqst *rqstp,
3737 struct nfsd4_compound_state *cstate,
3738 union nfsd4_op_u *u)
3740 struct nfsd4_bind_conn_to_session *bcts = &u->bind_conn_to_session;
3742 struct nfsd4_conn *conn;
3743 struct nfsd4_session *session;
3744 struct net *net = SVC_NET(rqstp);
3745 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
3747 if (!nfsd4_last_compound_op(rqstp))
3748 return nfserr_not_only_op;
3749 spin_lock(&nn->client_lock);
3750 session = find_in_sessionid_hashtbl(&bcts->sessionid, net, &status);
3751 spin_unlock(&nn->client_lock);
3753 goto out_no_session;
3754 status = nfserr_wrong_cred;
3755 if (!nfsd4_mach_creds_match(session->se_client, rqstp))
3757 status = nfsd4_match_existing_connection(rqstp, session,
3759 if (status == nfs_ok) {
3760 if (bcts->dir == NFS4_CDFC4_FORE_OR_BOTH ||
3761 bcts->dir == NFS4_CDFC4_BACK)
3762 conn->cn_flags |= NFS4_CDFC4_BACK;
3763 nfsd4_probe_callback(session->se_client);
3766 if (status == nfserr_inval)
3768 status = nfsd4_map_bcts_dir(&bcts->dir);
3771 conn = alloc_conn(rqstp, bcts->dir);
3772 status = nfserr_jukebox;
3775 nfsd4_init_conn(rqstp, conn, session);
3778 nfsd4_put_session(session);
3783 static bool nfsd4_compound_in_session(struct nfsd4_compound_state *cstate, struct nfs4_sessionid *sid)
3785 if (!cstate->session)
3787 return !memcmp(sid, &cstate->session->se_sessionid, sizeof(*sid));
3791 nfsd4_destroy_session(struct svc_rqst *r, struct nfsd4_compound_state *cstate,
3792 union nfsd4_op_u *u)
3794 struct nfs4_sessionid *sessionid = &u->destroy_session.sessionid;
3795 struct nfsd4_session *ses;
3797 int ref_held_by_me = 0;
3798 struct net *net = SVC_NET(r);
3799 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
3801 status = nfserr_not_only_op;
3802 if (nfsd4_compound_in_session(cstate, sessionid)) {
3803 if (!nfsd4_last_compound_op(r))
3807 dump_sessionid(__func__, sessionid);
3808 spin_lock(&nn->client_lock);
3809 ses = find_in_sessionid_hashtbl(sessionid, net, &status);
3811 goto out_client_lock;
3812 status = nfserr_wrong_cred;
3813 if (!nfsd4_mach_creds_match(ses->se_client, r))
3814 goto out_put_session;
3815 status = mark_session_dead_locked(ses, 1 + ref_held_by_me);
3817 goto out_put_session;
3818 unhash_session(ses);
3819 spin_unlock(&nn->client_lock);
3821 nfsd4_probe_callback_sync(ses->se_client);
3823 spin_lock(&nn->client_lock);
3826 nfsd4_put_session_locked(ses);
3828 spin_unlock(&nn->client_lock);
3833 static __be32 nfsd4_sequence_check_conn(struct nfsd4_conn *new, struct nfsd4_session *ses)
3835 struct nfs4_client *clp = ses->se_client;
3836 struct nfsd4_conn *c;
3837 __be32 status = nfs_ok;
3840 spin_lock(&clp->cl_lock);
3841 c = __nfsd4_find_conn(new->cn_xprt, ses);
3844 status = nfserr_conn_not_bound_to_session;
3845 if (clp->cl_mach_cred)
3847 __nfsd4_hash_conn(new, ses);
3848 spin_unlock(&clp->cl_lock);
3849 ret = nfsd4_register_conn(new);
3851 /* oops; xprt is already down: */
3852 nfsd4_conn_lost(&new->cn_xpt_user);
3855 spin_unlock(&clp->cl_lock);
3860 static bool nfsd4_session_too_many_ops(struct svc_rqst *rqstp, struct nfsd4_session *session)
3862 struct nfsd4_compoundargs *args = rqstp->rq_argp;
3864 return args->opcnt > session->se_fchannel.maxops;
3867 static bool nfsd4_request_too_big(struct svc_rqst *rqstp,
3868 struct nfsd4_session *session)
3870 struct xdr_buf *xb = &rqstp->rq_arg;
3872 return xb->len > session->se_fchannel.maxreq_sz;
3875 static bool replay_matches_cache(struct svc_rqst *rqstp,
3876 struct nfsd4_sequence *seq, struct nfsd4_slot *slot)
3878 struct nfsd4_compoundargs *argp = rqstp->rq_argp;
3880 if ((bool)(slot->sl_flags & NFSD4_SLOT_CACHETHIS) !=
3881 (bool)seq->cachethis)
3884 * If there's an error then the reply can have fewer ops than
3887 if (slot->sl_opcnt < argp->opcnt && !slot->sl_status)
3890 * But if we cached a reply with *more* ops than the call you're
3891 * sending us now, then this new call is clearly not really a
3892 * replay of the old one:
3894 if (slot->sl_opcnt > argp->opcnt)
3896 /* This is the only check explicitly called by spec: */
3897 if (!same_creds(&rqstp->rq_cred, &slot->sl_cred))
3900 * There may be more comparisons we could actually do, but the
3901 * spec doesn't require us to catch every case where the calls
3902 * don't match (that would require caching the call as well as
3903 * the reply), so we don't bother.
3909 nfsd4_sequence(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3910 union nfsd4_op_u *u)
3912 struct nfsd4_sequence *seq = &u->sequence;
3913 struct nfsd4_compoundres *resp = rqstp->rq_resp;
3914 struct xdr_stream *xdr = resp->xdr;
3915 struct nfsd4_session *session;
3916 struct nfs4_client *clp;
3917 struct nfsd4_slot *slot;
3918 struct nfsd4_conn *conn;
3921 struct net *net = SVC_NET(rqstp);
3922 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
3924 if (resp->opcnt != 1)
3925 return nfserr_sequence_pos;
3928 * Will be either used or freed by nfsd4_sequence_check_conn
3931 conn = alloc_conn(rqstp, NFS4_CDFC4_FORE);
3933 return nfserr_jukebox;
3935 spin_lock(&nn->client_lock);
3936 session = find_in_sessionid_hashtbl(&seq->sessionid, net, &status);
3938 goto out_no_session;
3939 clp = session->se_client;
3941 status = nfserr_too_many_ops;
3942 if (nfsd4_session_too_many_ops(rqstp, session))
3943 goto out_put_session;
3945 status = nfserr_req_too_big;
3946 if (nfsd4_request_too_big(rqstp, session))
3947 goto out_put_session;
3949 status = nfserr_badslot;
3950 if (seq->slotid >= session->se_fchannel.maxreqs)
3951 goto out_put_session;
3953 slot = session->se_slots[seq->slotid];
3954 dprintk("%s: slotid %d\n", __func__, seq->slotid);
3956 /* We do not negotiate the number of slots yet, so set the
3957 * maxslots to the session maxreqs which is used to encode
3958 * sr_highest_slotid and the sr_target_slot id to maxslots */
3959 seq->maxslots = session->se_fchannel.maxreqs;
3961 status = check_slot_seqid(seq->seqid, slot->sl_seqid,
3962 slot->sl_flags & NFSD4_SLOT_INUSE);
3963 if (status == nfserr_replay_cache) {
3964 status = nfserr_seq_misordered;
3965 if (!(slot->sl_flags & NFSD4_SLOT_INITIALIZED))
3966 goto out_put_session;
3967 status = nfserr_seq_false_retry;
3968 if (!replay_matches_cache(rqstp, seq, slot))
3969 goto out_put_session;
3970 cstate->slot = slot;
3971 cstate->session = session;
3973 /* Return the cached reply status and set cstate->status
3974 * for nfsd4_proc_compound processing */
3975 status = nfsd4_replay_cache_entry(resp, seq);
3976 cstate->status = nfserr_replay_cache;
3980 goto out_put_session;
3982 status = nfsd4_sequence_check_conn(conn, session);
3985 goto out_put_session;
3987 buflen = (seq->cachethis) ?
3988 session->se_fchannel.maxresp_cached :
3989 session->se_fchannel.maxresp_sz;
3990 status = (seq->cachethis) ? nfserr_rep_too_big_to_cache :
3992 if (xdr_restrict_buflen(xdr, buflen - rqstp->rq_auth_slack))
3993 goto out_put_session;
3994 svc_reserve(rqstp, buflen);
3997 /* Success! bump slot seqid */
3998 slot->sl_seqid = seq->seqid;
3999 slot->sl_flags |= NFSD4_SLOT_INUSE;
4001 slot->sl_flags |= NFSD4_SLOT_CACHETHIS;
4003 slot->sl_flags &= ~NFSD4_SLOT_CACHETHIS;
4005 cstate->slot = slot;
4006 cstate->session = session;
4010 switch (clp->cl_cb_state) {
4012 seq->status_flags = SEQ4_STATUS_CB_PATH_DOWN;
4014 case NFSD4_CB_FAULT:
4015 seq->status_flags = SEQ4_STATUS_BACKCHANNEL_FAULT;
4018 seq->status_flags = 0;
4020 if (!list_empty(&clp->cl_revoked))
4021 seq->status_flags |= SEQ4_STATUS_RECALLABLE_STATE_REVOKED;
4025 spin_unlock(&nn->client_lock);
4028 nfsd4_put_session_locked(session);
4029 goto out_no_session;
4033 nfsd4_sequence_done(struct nfsd4_compoundres *resp)
4035 struct nfsd4_compound_state *cs = &resp->cstate;
4037 if (nfsd4_has_session(cs)) {
4038 if (cs->status != nfserr_replay_cache) {
4039 nfsd4_store_cache_entry(resp);
4040 cs->slot->sl_flags &= ~NFSD4_SLOT_INUSE;
4042 /* Drop session reference that was taken in nfsd4_sequence() */
4043 nfsd4_put_session(cs->session);
4045 put_client_renew(cs->clp);
4049 nfsd4_destroy_clientid(struct svc_rqst *rqstp,
4050 struct nfsd4_compound_state *cstate,
4051 union nfsd4_op_u *u)
4053 struct nfsd4_destroy_clientid *dc = &u->destroy_clientid;
4054 struct nfs4_client *conf, *unconf;
4055 struct nfs4_client *clp = NULL;
4057 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
4059 spin_lock(&nn->client_lock);
4060 unconf = find_unconfirmed_client(&dc->clientid, true, nn);
4061 conf = find_confirmed_client(&dc->clientid, true, nn);
4062 WARN_ON_ONCE(conf && unconf);
4065 if (client_has_state(conf)) {
4066 status = nfserr_clientid_busy;
4069 status = mark_client_expired_locked(conf);
4076 status = nfserr_stale_clientid;
4079 if (!nfsd4_mach_creds_match(clp, rqstp)) {
4081 status = nfserr_wrong_cred;
4084 trace_nfsd_clid_destroyed(&clp->cl_clientid);
4085 unhash_client_locked(clp);
4087 spin_unlock(&nn->client_lock);
4094 nfsd4_reclaim_complete(struct svc_rqst *rqstp,
4095 struct nfsd4_compound_state *cstate, union nfsd4_op_u *u)
4097 struct nfsd4_reclaim_complete *rc = &u->reclaim_complete;
4098 struct nfs4_client *clp = cstate->clp;
4101 if (rc->rca_one_fs) {
4102 if (!cstate->current_fh.fh_dentry)
4103 return nfserr_nofilehandle;
4105 * We don't take advantage of the rca_one_fs case.
4106 * That's OK, it's optional, we can safely ignore it.
4111 status = nfserr_complete_already;
4112 if (test_and_set_bit(NFSD4_CLIENT_RECLAIM_COMPLETE, &clp->cl_flags))
4115 status = nfserr_stale_clientid;
4116 if (is_client_expired(clp))
4118 * The following error isn't really legal.
4119 * But we only get here if the client just explicitly
4120 * destroyed the client. Surely it no longer cares what
4121 * error it gets back on an operation for the dead
4127 trace_nfsd_clid_reclaim_complete(&clp->cl_clientid);
4128 nfsd4_client_record_create(clp);
4129 inc_reclaim_complete(clp);
4135 nfsd4_setclientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4136 union nfsd4_op_u *u)
4138 struct nfsd4_setclientid *setclid = &u->setclientid;
4139 struct xdr_netobj clname = setclid->se_name;
4140 nfs4_verifier clverifier = setclid->se_verf;
4141 struct nfs4_client *conf, *new;
4142 struct nfs4_client *unconf = NULL;
4144 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
4146 new = create_client(clname, rqstp, &clverifier);
4148 return nfserr_jukebox;
4149 spin_lock(&nn->client_lock);
4150 conf = find_confirmed_client_by_name(&clname, nn);
4151 if (conf && client_has_state(conf)) {
4152 status = nfserr_clid_inuse;
4153 if (clp_used_exchangeid(conf))
4155 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
4156 trace_nfsd_clid_cred_mismatch(conf, rqstp);
4160 unconf = find_unconfirmed_client_by_name(&clname, nn);
4162 unhash_client_locked(unconf);
4164 if (same_verf(&conf->cl_verifier, &clverifier)) {
4165 copy_clid(new, conf);
4166 gen_confirm(new, nn);
4168 trace_nfsd_clid_verf_mismatch(conf, rqstp,
4171 trace_nfsd_clid_fresh(new);
4172 new->cl_minorversion = 0;
4173 gen_callback(new, setclid, rqstp);
4174 add_to_unconfirmed(new);
4175 setclid->se_clientid.cl_boot = new->cl_clientid.cl_boot;
4176 setclid->se_clientid.cl_id = new->cl_clientid.cl_id;
4177 memcpy(setclid->se_confirm.data, new->cl_confirm.data, sizeof(setclid->se_confirm.data));
4181 spin_unlock(&nn->client_lock);
4185 trace_nfsd_clid_expire_unconf(&unconf->cl_clientid);
4186 expire_client(unconf);
4192 nfsd4_setclientid_confirm(struct svc_rqst *rqstp,
4193 struct nfsd4_compound_state *cstate,
4194 union nfsd4_op_u *u)
4196 struct nfsd4_setclientid_confirm *setclientid_confirm =
4197 &u->setclientid_confirm;
4198 struct nfs4_client *conf, *unconf;
4199 struct nfs4_client *old = NULL;
4200 nfs4_verifier confirm = setclientid_confirm->sc_confirm;
4201 clientid_t * clid = &setclientid_confirm->sc_clientid;
4203 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
4205 if (STALE_CLIENTID(clid, nn))
4206 return nfserr_stale_clientid;
4208 spin_lock(&nn->client_lock);
4209 conf = find_confirmed_client(clid, false, nn);
4210 unconf = find_unconfirmed_client(clid, false, nn);
4212 * We try hard to give out unique clientid's, so if we get an
4213 * attempt to confirm the same clientid with a different cred,
4214 * the client may be buggy; this should never happen.
4216 * Nevertheless, RFC 7530 recommends INUSE for this case:
4218 status = nfserr_clid_inuse;
4219 if (unconf && !same_creds(&unconf->cl_cred, &rqstp->rq_cred)) {
4220 trace_nfsd_clid_cred_mismatch(unconf, rqstp);
4223 if (conf && !same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
4224 trace_nfsd_clid_cred_mismatch(conf, rqstp);
4227 if (!unconf || !same_verf(&confirm, &unconf->cl_confirm)) {
4228 if (conf && same_verf(&confirm, &conf->cl_confirm)) {
4231 status = nfserr_stale_clientid;
4237 unhash_client_locked(old);
4238 nfsd4_change_callback(conf, &unconf->cl_cb_conn);
4240 old = find_confirmed_client_by_name(&unconf->cl_name, nn);
4242 status = nfserr_clid_inuse;
4243 if (client_has_state(old)
4244 && !same_creds(&unconf->cl_cred,
4249 status = mark_client_expired_locked(old);
4254 trace_nfsd_clid_replaced(&old->cl_clientid);
4256 move_to_confirmed(unconf);
4259 get_client_locked(conf);
4260 spin_unlock(&nn->client_lock);
4262 fsnotify_dentry(conf->cl_nfsd_info_dentry, FS_MODIFY);
4263 nfsd4_probe_callback(conf);
4264 spin_lock(&nn->client_lock);
4265 put_client_renew_locked(conf);
4267 spin_unlock(&nn->client_lock);
4273 static struct nfs4_file *nfsd4_alloc_file(void)
4275 return kmem_cache_alloc(file_slab, GFP_KERNEL);
4278 /* OPEN Share state helper functions */
4279 static void nfsd4_init_file(struct svc_fh *fh, unsigned int hashval,
4280 struct nfs4_file *fp)
4282 lockdep_assert_held(&state_lock);
4284 refcount_set(&fp->fi_ref, 1);
4285 spin_lock_init(&fp->fi_lock);
4286 INIT_LIST_HEAD(&fp->fi_stateids);
4287 INIT_LIST_HEAD(&fp->fi_delegations);
4288 INIT_LIST_HEAD(&fp->fi_clnt_odstate);
4289 fh_copy_shallow(&fp->fi_fhandle, &fh->fh_handle);
4290 fp->fi_deleg_file = NULL;
4291 fp->fi_had_conflict = false;
4292 fp->fi_share_deny = 0;
4293 memset(fp->fi_fds, 0, sizeof(fp->fi_fds));
4294 memset(fp->fi_access, 0, sizeof(fp->fi_access));
4295 fp->fi_aliased = false;
4296 fp->fi_inode = d_inode(fh->fh_dentry);
4297 #ifdef CONFIG_NFSD_PNFS
4298 INIT_LIST_HEAD(&fp->fi_lo_states);
4299 atomic_set(&fp->fi_lo_recalls, 0);
4301 hlist_add_head_rcu(&fp->fi_hash, &file_hashtbl[hashval]);
4305 nfsd4_free_slabs(void)
4307 kmem_cache_destroy(client_slab);
4308 kmem_cache_destroy(openowner_slab);
4309 kmem_cache_destroy(lockowner_slab);
4310 kmem_cache_destroy(file_slab);
4311 kmem_cache_destroy(stateid_slab);
4312 kmem_cache_destroy(deleg_slab);
4313 kmem_cache_destroy(odstate_slab);
4317 nfsd4_init_slabs(void)
4319 client_slab = kmem_cache_create("nfsd4_clients",
4320 sizeof(struct nfs4_client), 0, 0, NULL);
4321 if (client_slab == NULL)
4323 openowner_slab = kmem_cache_create("nfsd4_openowners",
4324 sizeof(struct nfs4_openowner), 0, 0, NULL);
4325 if (openowner_slab == NULL)
4326 goto out_free_client_slab;
4327 lockowner_slab = kmem_cache_create("nfsd4_lockowners",
4328 sizeof(struct nfs4_lockowner), 0, 0, NULL);
4329 if (lockowner_slab == NULL)
4330 goto out_free_openowner_slab;
4331 file_slab = kmem_cache_create("nfsd4_files",
4332 sizeof(struct nfs4_file), 0, 0, NULL);
4333 if (file_slab == NULL)
4334 goto out_free_lockowner_slab;
4335 stateid_slab = kmem_cache_create("nfsd4_stateids",
4336 sizeof(struct nfs4_ol_stateid), 0, 0, NULL);
4337 if (stateid_slab == NULL)
4338 goto out_free_file_slab;
4339 deleg_slab = kmem_cache_create("nfsd4_delegations",
4340 sizeof(struct nfs4_delegation), 0, 0, NULL);
4341 if (deleg_slab == NULL)
4342 goto out_free_stateid_slab;
4343 odstate_slab = kmem_cache_create("nfsd4_odstate",
4344 sizeof(struct nfs4_clnt_odstate), 0, 0, NULL);
4345 if (odstate_slab == NULL)
4346 goto out_free_deleg_slab;
4349 out_free_deleg_slab:
4350 kmem_cache_destroy(deleg_slab);
4351 out_free_stateid_slab:
4352 kmem_cache_destroy(stateid_slab);
4354 kmem_cache_destroy(file_slab);
4355 out_free_lockowner_slab:
4356 kmem_cache_destroy(lockowner_slab);
4357 out_free_openowner_slab:
4358 kmem_cache_destroy(openowner_slab);
4359 out_free_client_slab:
4360 kmem_cache_destroy(client_slab);
4365 static unsigned long
4366 nfsd_courtesy_client_count(struct shrinker *shrink, struct shrink_control *sc)
4369 struct nfsd_net *nn = container_of(shrink,
4370 struct nfsd_net, nfsd_client_shrinker);
4372 cnt = atomic_read(&nn->nfsd_courtesy_clients);
4374 mod_delayed_work(laundry_wq, &nn->nfsd_shrinker_work, 0);
4375 return (unsigned long)cnt;
4378 static unsigned long
4379 nfsd_courtesy_client_scan(struct shrinker *shrink, struct shrink_control *sc)
4385 nfsd4_init_leases_net(struct nfsd_net *nn)
4390 nn->nfsd4_lease = 90; /* default lease time */
4391 nn->nfsd4_grace = 90;
4392 nn->somebody_reclaimed = false;
4393 nn->track_reclaim_completes = false;
4394 nn->clverifier_counter = get_random_u32();
4395 nn->clientid_base = get_random_u32();
4396 nn->clientid_counter = nn->clientid_base + 1;
4397 nn->s2s_cp_cl_id = nn->clientid_counter++;
4399 atomic_set(&nn->nfs4_client_count, 0);
4401 max_clients = (u64)si.totalram * si.mem_unit / (1024 * 1024 * 1024);
4402 max_clients *= NFS4_CLIENTS_PER_GB;
4403 nn->nfs4_max_clients = max_t(int, max_clients, NFS4_CLIENTS_PER_GB);
4405 atomic_set(&nn->nfsd_courtesy_clients, 0);
4406 nn->nfsd_client_shrinker.scan_objects = nfsd_courtesy_client_scan;
4407 nn->nfsd_client_shrinker.count_objects = nfsd_courtesy_client_count;
4408 nn->nfsd_client_shrinker.seeks = DEFAULT_SEEKS;
4409 return register_shrinker(&nn->nfsd_client_shrinker, "nfsd-client");
4413 nfsd4_leases_net_shutdown(struct nfsd_net *nn)
4415 unregister_shrinker(&nn->nfsd_client_shrinker);
4418 static void init_nfs4_replay(struct nfs4_replay *rp)
4420 rp->rp_status = nfserr_serverfault;
4422 rp->rp_buf = rp->rp_ibuf;
4423 mutex_init(&rp->rp_mutex);
4426 static void nfsd4_cstate_assign_replay(struct nfsd4_compound_state *cstate,
4427 struct nfs4_stateowner *so)
4429 if (!nfsd4_has_session(cstate)) {
4430 mutex_lock(&so->so_replay.rp_mutex);
4431 cstate->replay_owner = nfs4_get_stateowner(so);
4435 void nfsd4_cstate_clear_replay(struct nfsd4_compound_state *cstate)
4437 struct nfs4_stateowner *so = cstate->replay_owner;
4440 cstate->replay_owner = NULL;
4441 mutex_unlock(&so->so_replay.rp_mutex);
4442 nfs4_put_stateowner(so);
4446 static inline void *alloc_stateowner(struct kmem_cache *slab, struct xdr_netobj *owner, struct nfs4_client *clp)
4448 struct nfs4_stateowner *sop;
4450 sop = kmem_cache_alloc(slab, GFP_KERNEL);
4454 xdr_netobj_dup(&sop->so_owner, owner, GFP_KERNEL);
4455 if (!sop->so_owner.data) {
4456 kmem_cache_free(slab, sop);
4460 INIT_LIST_HEAD(&sop->so_stateids);
4461 sop->so_client = clp;
4462 init_nfs4_replay(&sop->so_replay);
4463 atomic_set(&sop->so_count, 1);
4467 static void hash_openowner(struct nfs4_openowner *oo, struct nfs4_client *clp, unsigned int strhashval)
4469 lockdep_assert_held(&clp->cl_lock);
4471 list_add(&oo->oo_owner.so_strhash,
4472 &clp->cl_ownerstr_hashtbl[strhashval]);
4473 list_add(&oo->oo_perclient, &clp->cl_openowners);
4476 static void nfs4_unhash_openowner(struct nfs4_stateowner *so)
4478 unhash_openowner_locked(openowner(so));
4481 static void nfs4_free_openowner(struct nfs4_stateowner *so)
4483 struct nfs4_openowner *oo = openowner(so);
4485 kmem_cache_free(openowner_slab, oo);
4488 static const struct nfs4_stateowner_operations openowner_ops = {
4489 .so_unhash = nfs4_unhash_openowner,
4490 .so_free = nfs4_free_openowner,
4493 static struct nfs4_ol_stateid *
4494 nfsd4_find_existing_open(struct nfs4_file *fp, struct nfsd4_open *open)
4496 struct nfs4_ol_stateid *local, *ret = NULL;
4497 struct nfs4_openowner *oo = open->op_openowner;
4499 lockdep_assert_held(&fp->fi_lock);
4501 list_for_each_entry(local, &fp->fi_stateids, st_perfile) {
4502 /* ignore lock owners */
4503 if (local->st_stateowner->so_is_open_owner == 0)
4505 if (local->st_stateowner != &oo->oo_owner)
4507 if (local->st_stid.sc_type == NFS4_OPEN_STID) {
4509 refcount_inc(&ret->st_stid.sc_count);
4517 nfsd4_verify_open_stid(struct nfs4_stid *s)
4519 __be32 ret = nfs_ok;
4521 switch (s->sc_type) {
4525 case NFS4_CLOSED_STID:
4526 case NFS4_CLOSED_DELEG_STID:
4527 ret = nfserr_bad_stateid;
4529 case NFS4_REVOKED_DELEG_STID:
4530 ret = nfserr_deleg_revoked;
4535 /* Lock the stateid st_mutex, and deal with races with CLOSE */
4537 nfsd4_lock_ol_stateid(struct nfs4_ol_stateid *stp)
4541 mutex_lock_nested(&stp->st_mutex, LOCK_STATEID_MUTEX);
4542 ret = nfsd4_verify_open_stid(&stp->st_stid);
4544 mutex_unlock(&stp->st_mutex);
4548 static struct nfs4_ol_stateid *
4549 nfsd4_find_and_lock_existing_open(struct nfs4_file *fp, struct nfsd4_open *open)
4551 struct nfs4_ol_stateid *stp;
4553 spin_lock(&fp->fi_lock);
4554 stp = nfsd4_find_existing_open(fp, open);
4555 spin_unlock(&fp->fi_lock);
4556 if (!stp || nfsd4_lock_ol_stateid(stp) == nfs_ok)
4558 nfs4_put_stid(&stp->st_stid);
4563 static struct nfs4_openowner *
4564 alloc_init_open_stateowner(unsigned int strhashval, struct nfsd4_open *open,
4565 struct nfsd4_compound_state *cstate)
4567 struct nfs4_client *clp = cstate->clp;
4568 struct nfs4_openowner *oo, *ret;
4570 oo = alloc_stateowner(openowner_slab, &open->op_owner, clp);
4573 oo->oo_owner.so_ops = &openowner_ops;
4574 oo->oo_owner.so_is_open_owner = 1;
4575 oo->oo_owner.so_seqid = open->op_seqid;
4577 if (nfsd4_has_session(cstate))
4578 oo->oo_flags |= NFS4_OO_CONFIRMED;
4580 oo->oo_last_closed_stid = NULL;
4581 INIT_LIST_HEAD(&oo->oo_close_lru);
4582 spin_lock(&clp->cl_lock);
4583 ret = find_openstateowner_str_locked(strhashval, open, clp);
4585 hash_openowner(oo, clp, strhashval);
4588 nfs4_free_stateowner(&oo->oo_owner);
4590 spin_unlock(&clp->cl_lock);
4594 static struct nfs4_ol_stateid *
4595 init_open_stateid(struct nfs4_file *fp, struct nfsd4_open *open)
4598 struct nfs4_openowner *oo = open->op_openowner;
4599 struct nfs4_ol_stateid *retstp = NULL;
4600 struct nfs4_ol_stateid *stp;
4603 /* We are moving these outside of the spinlocks to avoid the warnings */
4604 mutex_init(&stp->st_mutex);
4605 mutex_lock_nested(&stp->st_mutex, OPEN_STATEID_MUTEX);
4608 spin_lock(&oo->oo_owner.so_client->cl_lock);
4609 spin_lock(&fp->fi_lock);
4611 retstp = nfsd4_find_existing_open(fp, open);
4615 open->op_stp = NULL;
4616 refcount_inc(&stp->st_stid.sc_count);
4617 stp->st_stid.sc_type = NFS4_OPEN_STID;
4618 INIT_LIST_HEAD(&stp->st_locks);
4619 stp->st_stateowner = nfs4_get_stateowner(&oo->oo_owner);
4621 stp->st_stid.sc_file = fp;
4622 stp->st_access_bmap = 0;
4623 stp->st_deny_bmap = 0;
4624 stp->st_openstp = NULL;
4625 list_add(&stp->st_perstateowner, &oo->oo_owner.so_stateids);
4626 list_add(&stp->st_perfile, &fp->fi_stateids);
4629 spin_unlock(&fp->fi_lock);
4630 spin_unlock(&oo->oo_owner.so_client->cl_lock);
4632 /* Handle races with CLOSE */
4633 if (nfsd4_lock_ol_stateid(retstp) != nfs_ok) {
4634 nfs4_put_stid(&retstp->st_stid);
4637 /* To keep mutex tracking happy */
4638 mutex_unlock(&stp->st_mutex);
4645 * In the 4.0 case we need to keep the owners around a little while to handle
4646 * CLOSE replay. We still do need to release any file access that is held by
4647 * them before returning however.
4650 move_to_close_lru(struct nfs4_ol_stateid *s, struct net *net)
4652 struct nfs4_ol_stateid *last;
4653 struct nfs4_openowner *oo = openowner(s->st_stateowner);
4654 struct nfsd_net *nn = net_generic(s->st_stid.sc_client->net,
4657 dprintk("NFSD: move_to_close_lru nfs4_openowner %p\n", oo);
4660 * We know that we hold one reference via nfsd4_close, and another
4661 * "persistent" reference for the client. If the refcount is higher
4662 * than 2, then there are still calls in progress that are using this
4663 * stateid. We can't put the sc_file reference until they are finished.
4664 * Wait for the refcount to drop to 2. Since it has been unhashed,
4665 * there should be no danger of the refcount going back up again at
4668 wait_event(close_wq, refcount_read(&s->st_stid.sc_count) == 2);
4670 release_all_access(s);
4671 if (s->st_stid.sc_file) {
4672 put_nfs4_file(s->st_stid.sc_file);
4673 s->st_stid.sc_file = NULL;
4676 spin_lock(&nn->client_lock);
4677 last = oo->oo_last_closed_stid;
4678 oo->oo_last_closed_stid = s;
4679 list_move_tail(&oo->oo_close_lru, &nn->close_lru);
4680 oo->oo_time = ktime_get_boottime_seconds();
4681 spin_unlock(&nn->client_lock);
4683 nfs4_put_stid(&last->st_stid);
4686 /* search file_hashtbl[] for file */
4687 static struct nfs4_file *
4688 find_file_locked(struct svc_fh *fh, unsigned int hashval)
4690 struct nfs4_file *fp;
4692 hlist_for_each_entry_rcu(fp, &file_hashtbl[hashval], fi_hash,
4693 lockdep_is_held(&state_lock)) {
4694 if (fh_match(&fp->fi_fhandle, &fh->fh_handle)) {
4695 if (refcount_inc_not_zero(&fp->fi_ref))
4702 static struct nfs4_file *insert_file(struct nfs4_file *new, struct svc_fh *fh,
4703 unsigned int hashval)
4705 struct nfs4_file *fp;
4706 struct nfs4_file *ret = NULL;
4707 bool alias_found = false;
4709 spin_lock(&state_lock);
4710 hlist_for_each_entry_rcu(fp, &file_hashtbl[hashval], fi_hash,
4711 lockdep_is_held(&state_lock)) {
4712 if (fh_match(&fp->fi_fhandle, &fh->fh_handle)) {
4713 if (refcount_inc_not_zero(&fp->fi_ref))
4715 } else if (d_inode(fh->fh_dentry) == fp->fi_inode)
4716 fp->fi_aliased = alias_found = true;
4718 if (likely(ret == NULL)) {
4719 nfsd4_init_file(fh, hashval, new);
4720 new->fi_aliased = alias_found;
4723 spin_unlock(&state_lock);
4727 static struct nfs4_file * find_file(struct svc_fh *fh)
4729 struct nfs4_file *fp;
4730 unsigned int hashval = file_hashval(fh);
4733 fp = find_file_locked(fh, hashval);
4738 static struct nfs4_file *
4739 find_or_add_file(struct nfs4_file *new, struct svc_fh *fh)
4741 struct nfs4_file *fp;
4742 unsigned int hashval = file_hashval(fh);
4745 fp = find_file_locked(fh, hashval);
4750 return insert_file(new, fh, hashval);
4754 * Called to check deny when READ with all zero stateid or
4755 * WRITE with all zero or all one stateid
4758 nfs4_share_conflict(struct svc_fh *current_fh, unsigned int deny_type)
4760 struct nfs4_file *fp;
4761 __be32 ret = nfs_ok;
4763 fp = find_file(current_fh);
4766 /* Check for conflicting share reservations */
4767 spin_lock(&fp->fi_lock);
4768 if (fp->fi_share_deny & deny_type)
4769 ret = nfserr_locked;
4770 spin_unlock(&fp->fi_lock);
4775 static bool nfsd4_deleg_present(const struct inode *inode)
4777 struct file_lock_context *ctx = smp_load_acquire(&inode->i_flctx);
4779 return ctx && !list_empty_careful(&ctx->flc_lease);
4783 * nfsd_wait_for_delegreturn - wait for delegations to be returned
4784 * @rqstp: the RPC transaction being executed
4785 * @inode: in-core inode of the file being waited for
4787 * The timeout prevents deadlock if all nfsd threads happen to be
4788 * tied up waiting for returning delegations.
4791 * %true: delegation was returned
4792 * %false: timed out waiting for delegreturn
4794 bool nfsd_wait_for_delegreturn(struct svc_rqst *rqstp, struct inode *inode)
4796 long __maybe_unused timeo;
4798 timeo = wait_var_event_timeout(inode, !nfsd4_deleg_present(inode),
4799 NFSD_DELEGRETURN_TIMEOUT);
4800 trace_nfsd_delegret_wakeup(rqstp, inode, timeo);
4804 static void nfsd4_cb_recall_prepare(struct nfsd4_callback *cb)
4806 struct nfs4_delegation *dp = cb_to_delegation(cb);
4807 struct nfsd_net *nn = net_generic(dp->dl_stid.sc_client->net,
4810 block_delegations(&dp->dl_stid.sc_file->fi_fhandle);
4813 * We can't do this in nfsd_break_deleg_cb because it is
4814 * already holding inode->i_lock.
4816 * If the dl_time != 0, then we know that it has already been
4817 * queued for a lease break. Don't queue it again.
4819 spin_lock(&state_lock);
4820 if (delegation_hashed(dp) && dp->dl_time == 0) {
4821 dp->dl_time = ktime_get_boottime_seconds();
4822 list_add_tail(&dp->dl_recall_lru, &nn->del_recall_lru);
4824 spin_unlock(&state_lock);
4827 static int nfsd4_cb_recall_done(struct nfsd4_callback *cb,
4828 struct rpc_task *task)
4830 struct nfs4_delegation *dp = cb_to_delegation(cb);
4832 trace_nfsd_cb_recall_done(&dp->dl_stid.sc_stateid, task);
4834 if (dp->dl_stid.sc_type == NFS4_CLOSED_DELEG_STID ||
4835 dp->dl_stid.sc_type == NFS4_REVOKED_DELEG_STID)
4838 switch (task->tk_status) {
4841 case -NFS4ERR_DELAY:
4842 rpc_delay(task, 2 * HZ);
4845 case -NFS4ERR_BAD_STATEID:
4847 * Race: client probably got cb_recall before open reply
4848 * granting delegation.
4850 if (dp->dl_retries--) {
4851 rpc_delay(task, 2 * HZ);
4860 static void nfsd4_cb_recall_release(struct nfsd4_callback *cb)
4862 struct nfs4_delegation *dp = cb_to_delegation(cb);
4864 nfs4_put_stid(&dp->dl_stid);
4867 static const struct nfsd4_callback_ops nfsd4_cb_recall_ops = {
4868 .prepare = nfsd4_cb_recall_prepare,
4869 .done = nfsd4_cb_recall_done,
4870 .release = nfsd4_cb_recall_release,
4873 static void nfsd_break_one_deleg(struct nfs4_delegation *dp)
4876 * We're assuming the state code never drops its reference
4877 * without first removing the lease. Since we're in this lease
4878 * callback (and since the lease code is serialized by the
4879 * flc_lock) we know the server hasn't removed the lease yet, and
4880 * we know it's safe to take a reference.
4882 refcount_inc(&dp->dl_stid.sc_count);
4883 WARN_ON_ONCE(!nfsd4_run_cb(&dp->dl_recall));
4886 /* Called from break_lease() with flc_lock held. */
4888 nfsd_break_deleg_cb(struct file_lock *fl)
4890 struct nfs4_delegation *dp = (struct nfs4_delegation *)fl->fl_owner;
4891 struct nfs4_file *fp = dp->dl_stid.sc_file;
4892 struct nfs4_client *clp = dp->dl_stid.sc_client;
4893 struct nfsd_net *nn;
4895 trace_nfsd_cb_recall(&dp->dl_stid);
4897 dp->dl_recalled = true;
4898 atomic_inc(&clp->cl_delegs_in_recall);
4899 if (try_to_expire_client(clp)) {
4900 nn = net_generic(clp->net, nfsd_net_id);
4901 mod_delayed_work(laundry_wq, &nn->laundromat_work, 0);
4905 * We don't want the locks code to timeout the lease for us;
4906 * we'll remove it ourself if a delegation isn't returned
4909 fl->fl_break_time = 0;
4911 spin_lock(&fp->fi_lock);
4912 fp->fi_had_conflict = true;
4913 nfsd_break_one_deleg(dp);
4914 spin_unlock(&fp->fi_lock);
4919 * nfsd_breaker_owns_lease - Check if lease conflict was resolved
4920 * @fl: Lock state to check
4923 * %true: Lease conflict was resolved
4924 * %false: Lease conflict was not resolved.
4926 static bool nfsd_breaker_owns_lease(struct file_lock *fl)
4928 struct nfs4_delegation *dl = fl->fl_owner;
4929 struct svc_rqst *rqst;
4930 struct nfs4_client *clp;
4934 rqst = kthread_data(current);
4935 /* Note rq_prog == NFS_ACL_PROGRAM is also possible: */
4936 if (rqst->rq_prog != NFS_PROGRAM || rqst->rq_vers < 4)
4938 clp = *(rqst->rq_lease_breaker);
4939 return dl->dl_stid.sc_client == clp;
4943 nfsd_change_deleg_cb(struct file_lock *onlist, int arg,
4944 struct list_head *dispose)
4946 struct nfs4_delegation *dp = (struct nfs4_delegation *)onlist->fl_owner;
4947 struct nfs4_client *clp = dp->dl_stid.sc_client;
4949 if (arg & F_UNLCK) {
4950 if (dp->dl_recalled)
4951 atomic_dec(&clp->cl_delegs_in_recall);
4952 return lease_modify(onlist, arg, dispose);
4957 static const struct lock_manager_operations nfsd_lease_mng_ops = {
4958 .lm_breaker_owns_lease = nfsd_breaker_owns_lease,
4959 .lm_break = nfsd_break_deleg_cb,
4960 .lm_change = nfsd_change_deleg_cb,
4963 static __be32 nfsd4_check_seqid(struct nfsd4_compound_state *cstate, struct nfs4_stateowner *so, u32 seqid)
4965 if (nfsd4_has_session(cstate))
4967 if (seqid == so->so_seqid - 1)
4968 return nfserr_replay_me;
4969 if (seqid == so->so_seqid)
4971 return nfserr_bad_seqid;
4974 static struct nfs4_client *lookup_clientid(clientid_t *clid, bool sessions,
4975 struct nfsd_net *nn)
4977 struct nfs4_client *found;
4979 spin_lock(&nn->client_lock);
4980 found = find_confirmed_client(clid, sessions, nn);
4982 atomic_inc(&found->cl_rpc_users);
4983 spin_unlock(&nn->client_lock);
4987 static __be32 set_client(clientid_t *clid,
4988 struct nfsd4_compound_state *cstate,
4989 struct nfsd_net *nn)
4992 if (!same_clid(&cstate->clp->cl_clientid, clid))
4993 return nfserr_stale_clientid;
4996 if (STALE_CLIENTID(clid, nn))
4997 return nfserr_stale_clientid;
4999 * We're in the 4.0 case (otherwise the SEQUENCE op would have
5000 * set cstate->clp), so session = false:
5002 cstate->clp = lookup_clientid(clid, false, nn);
5004 return nfserr_expired;
5009 nfsd4_process_open1(struct nfsd4_compound_state *cstate,
5010 struct nfsd4_open *open, struct nfsd_net *nn)
5012 clientid_t *clientid = &open->op_clientid;
5013 struct nfs4_client *clp = NULL;
5014 unsigned int strhashval;
5015 struct nfs4_openowner *oo = NULL;
5019 * In case we need it later, after we've already created the
5020 * file and don't want to risk a further failure:
5022 open->op_file = nfsd4_alloc_file();
5023 if (open->op_file == NULL)
5024 return nfserr_jukebox;
5026 status = set_client(clientid, cstate, nn);
5031 strhashval = ownerstr_hashval(&open->op_owner);
5032 oo = find_openstateowner_str(strhashval, open, clp);
5033 open->op_openowner = oo;
5037 if (!(oo->oo_flags & NFS4_OO_CONFIRMED)) {
5038 /* Replace unconfirmed owners without checking for replay. */
5039 release_openowner(oo);
5040 open->op_openowner = NULL;
5043 status = nfsd4_check_seqid(cstate, &oo->oo_owner, open->op_seqid);
5048 oo = alloc_init_open_stateowner(strhashval, open, cstate);
5050 return nfserr_jukebox;
5051 open->op_openowner = oo;
5053 open->op_stp = nfs4_alloc_open_stateid(clp);
5055 return nfserr_jukebox;
5057 if (nfsd4_has_session(cstate) &&
5058 (cstate->current_fh.fh_export->ex_flags & NFSEXP_PNFS)) {
5059 open->op_odstate = alloc_clnt_odstate(clp);
5060 if (!open->op_odstate)
5061 return nfserr_jukebox;
5067 static inline __be32
5068 nfs4_check_delegmode(struct nfs4_delegation *dp, int flags)
5070 if ((flags & WR_STATE) && (dp->dl_type == NFS4_OPEN_DELEGATE_READ))
5071 return nfserr_openmode;
5076 static int share_access_to_flags(u32 share_access)
5078 return share_access == NFS4_SHARE_ACCESS_READ ? RD_STATE : WR_STATE;
5081 static struct nfs4_delegation *find_deleg_stateid(struct nfs4_client *cl, stateid_t *s)
5083 struct nfs4_stid *ret;
5085 ret = find_stateid_by_type(cl, s,
5086 NFS4_DELEG_STID|NFS4_REVOKED_DELEG_STID);
5089 return delegstateid(ret);
5092 static bool nfsd4_is_deleg_cur(struct nfsd4_open *open)
5094 return open->op_claim_type == NFS4_OPEN_CLAIM_DELEGATE_CUR ||
5095 open->op_claim_type == NFS4_OPEN_CLAIM_DELEG_CUR_FH;
5099 nfs4_check_deleg(struct nfs4_client *cl, struct nfsd4_open *open,
5100 struct nfs4_delegation **dp)
5103 __be32 status = nfserr_bad_stateid;
5104 struct nfs4_delegation *deleg;
5106 deleg = find_deleg_stateid(cl, &open->op_delegate_stateid);
5109 if (deleg->dl_stid.sc_type == NFS4_REVOKED_DELEG_STID) {
5110 nfs4_put_stid(&deleg->dl_stid);
5111 if (cl->cl_minorversion)
5112 status = nfserr_deleg_revoked;
5115 flags = share_access_to_flags(open->op_share_access);
5116 status = nfs4_check_delegmode(deleg, flags);
5118 nfs4_put_stid(&deleg->dl_stid);
5123 if (!nfsd4_is_deleg_cur(open))
5127 open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
5131 static inline int nfs4_access_to_access(u32 nfs4_access)
5135 if (nfs4_access & NFS4_SHARE_ACCESS_READ)
5136 flags |= NFSD_MAY_READ;
5137 if (nfs4_access & NFS4_SHARE_ACCESS_WRITE)
5138 flags |= NFSD_MAY_WRITE;
5142 static inline __be32
5143 nfsd4_truncate(struct svc_rqst *rqstp, struct svc_fh *fh,
5144 struct nfsd4_open *open)
5146 struct iattr iattr = {
5147 .ia_valid = ATTR_SIZE,
5150 struct nfsd_attrs attrs = {
5153 if (!open->op_truncate)
5155 if (!(open->op_share_access & NFS4_SHARE_ACCESS_WRITE))
5156 return nfserr_inval;
5157 return nfsd_setattr(rqstp, fh, &attrs, 0, (time64_t)0);
5160 static __be32 nfs4_get_vfs_file(struct svc_rqst *rqstp, struct nfs4_file *fp,
5161 struct svc_fh *cur_fh, struct nfs4_ol_stateid *stp,
5162 struct nfsd4_open *open, bool new_stp)
5164 struct nfsd_file *nf = NULL;
5166 int oflag = nfs4_access_to_omode(open->op_share_access);
5167 int access = nfs4_access_to_access(open->op_share_access);
5168 unsigned char old_access_bmap, old_deny_bmap;
5170 spin_lock(&fp->fi_lock);
5173 * Are we trying to set a deny mode that would conflict with
5176 status = nfs4_file_check_deny(fp, open->op_share_deny);
5177 if (status != nfs_ok) {
5178 if (status != nfserr_share_denied) {
5179 spin_unlock(&fp->fi_lock);
5182 if (nfs4_resolve_deny_conflicts_locked(fp, new_stp,
5183 stp, open->op_share_deny, false))
5184 status = nfserr_jukebox;
5185 spin_unlock(&fp->fi_lock);
5189 /* set access to the file */
5190 status = nfs4_file_get_access(fp, open->op_share_access);
5191 if (status != nfs_ok) {
5192 if (status != nfserr_share_denied) {
5193 spin_unlock(&fp->fi_lock);
5196 if (nfs4_resolve_deny_conflicts_locked(fp, new_stp,
5197 stp, open->op_share_access, true))
5198 status = nfserr_jukebox;
5199 spin_unlock(&fp->fi_lock);
5203 /* Set access bits in stateid */
5204 old_access_bmap = stp->st_access_bmap;
5205 set_access(open->op_share_access, stp);
5207 /* Set new deny mask */
5208 old_deny_bmap = stp->st_deny_bmap;
5209 set_deny(open->op_share_deny, stp);
5210 fp->fi_share_deny |= (open->op_share_deny & NFS4_SHARE_DENY_BOTH);
5212 if (!fp->fi_fds[oflag]) {
5213 spin_unlock(&fp->fi_lock);
5215 status = nfsd_file_acquire_opened(rqstp, cur_fh, access,
5216 open->op_filp, &nf);
5217 if (status != nfs_ok)
5218 goto out_put_access;
5220 spin_lock(&fp->fi_lock);
5221 if (!fp->fi_fds[oflag]) {
5222 fp->fi_fds[oflag] = nf;
5226 spin_unlock(&fp->fi_lock);
5230 status = nfserrno(nfsd_open_break_lease(cur_fh->fh_dentry->d_inode,
5233 goto out_put_access;
5235 status = nfsd4_truncate(rqstp, cur_fh, open);
5237 goto out_put_access;
5241 stp->st_access_bmap = old_access_bmap;
5242 nfs4_file_put_access(fp, open->op_share_access);
5243 reset_union_bmap_deny(bmap_to_share_mode(old_deny_bmap), stp);
5248 nfs4_upgrade_open(struct svc_rqst *rqstp, struct nfs4_file *fp,
5249 struct svc_fh *cur_fh, struct nfs4_ol_stateid *stp,
5250 struct nfsd4_open *open)
5253 unsigned char old_deny_bmap = stp->st_deny_bmap;
5255 if (!test_access(open->op_share_access, stp))
5256 return nfs4_get_vfs_file(rqstp, fp, cur_fh, stp, open, false);
5258 /* test and set deny mode */
5259 spin_lock(&fp->fi_lock);
5260 status = nfs4_file_check_deny(fp, open->op_share_deny);
5261 if (status == nfs_ok) {
5262 if (status != nfserr_share_denied) {
5263 set_deny(open->op_share_deny, stp);
5264 fp->fi_share_deny |=
5265 (open->op_share_deny & NFS4_SHARE_DENY_BOTH);
5267 if (nfs4_resolve_deny_conflicts_locked(fp, false,
5268 stp, open->op_share_deny, false))
5269 status = nfserr_jukebox;
5272 spin_unlock(&fp->fi_lock);
5274 if (status != nfs_ok)
5277 status = nfsd4_truncate(rqstp, cur_fh, open);
5278 if (status != nfs_ok)
5279 reset_union_bmap_deny(old_deny_bmap, stp);
5283 /* Should we give out recallable state?: */
5284 static bool nfsd4_cb_channel_good(struct nfs4_client *clp)
5286 if (clp->cl_cb_state == NFSD4_CB_UP)
5289 * In the sessions case, since we don't have to establish a
5290 * separate connection for callbacks, we assume it's OK
5291 * until we hear otherwise:
5293 return clp->cl_minorversion && clp->cl_cb_state == NFSD4_CB_UNKNOWN;
5296 static struct file_lock *nfs4_alloc_init_lease(struct nfs4_delegation *dp,
5299 struct file_lock *fl;
5301 fl = locks_alloc_lock();
5304 fl->fl_lmops = &nfsd_lease_mng_ops;
5305 fl->fl_flags = FL_DELEG;
5306 fl->fl_type = flag == NFS4_OPEN_DELEGATE_READ? F_RDLCK: F_WRLCK;
5307 fl->fl_end = OFFSET_MAX;
5308 fl->fl_owner = (fl_owner_t)dp;
5309 fl->fl_pid = current->tgid;
5310 fl->fl_file = dp->dl_stid.sc_file->fi_deleg_file->nf_file;
5314 static int nfsd4_check_conflicting_opens(struct nfs4_client *clp,
5315 struct nfs4_file *fp)
5317 struct nfs4_ol_stateid *st;
5318 struct file *f = fp->fi_deleg_file->nf_file;
5319 struct inode *ino = locks_inode(f);
5322 writes = atomic_read(&ino->i_writecount);
5326 * There could be multiple filehandles (hence multiple
5327 * nfs4_files) referencing this file, but that's not too
5328 * common; let's just give up in that case rather than
5329 * trying to go look up all the clients using that other
5330 * nfs4_file as well:
5335 * If there's a close in progress, make sure that we see it
5336 * clear any fi_fds[] entries before we see it decrement
5339 smp_mb__after_atomic();
5341 if (fp->fi_fds[O_WRONLY])
5343 if (fp->fi_fds[O_RDWR])
5346 return -EAGAIN; /* There may be non-NFSv4 writers */
5348 * It's possible there are non-NFSv4 write opens in progress,
5349 * but if they haven't incremented i_writecount yet then they
5350 * also haven't called break lease yet; so, they'll break this
5351 * lease soon enough. So, all that's left to check for is NFSv4
5354 spin_lock(&fp->fi_lock);
5355 list_for_each_entry(st, &fp->fi_stateids, st_perfile) {
5356 if (st->st_openstp == NULL /* it's an open */ &&
5357 access_permit_write(st) &&
5358 st->st_stid.sc_client != clp) {
5359 spin_unlock(&fp->fi_lock);
5363 spin_unlock(&fp->fi_lock);
5365 * There's a small chance that we could be racing with another
5366 * NFSv4 open. However, any open that hasn't added itself to
5367 * the fi_stateids list also hasn't called break_lease yet; so,
5368 * they'll break this lease soon enough.
5374 * It's possible that between opening the dentry and setting the delegation,
5375 * that it has been renamed or unlinked. Redo the lookup to verify that this
5379 nfsd4_verify_deleg_dentry(struct nfsd4_open *open, struct nfs4_file *fp,
5380 struct svc_fh *parent)
5382 struct svc_export *exp;
5383 struct dentry *child;
5386 err = nfsd_lookup_dentry(open->op_rqstp, parent,
5387 open->op_fname, open->op_fnamelen,
5395 if (child != file_dentry(fp->fi_deleg_file->nf_file))
5401 static struct nfs4_delegation *
5402 nfs4_set_delegation(struct nfsd4_open *open, struct nfs4_ol_stateid *stp,
5403 struct svc_fh *parent)
5406 struct nfs4_client *clp = stp->st_stid.sc_client;
5407 struct nfs4_file *fp = stp->st_stid.sc_file;
5408 struct nfs4_clnt_odstate *odstate = stp->st_clnt_odstate;
5409 struct nfs4_delegation *dp;
5410 struct nfsd_file *nf;
5411 struct file_lock *fl;
5414 * The fi_had_conflict and nfs_get_existing_delegation checks
5415 * here are just optimizations; we'll need to recheck them at
5418 if (fp->fi_had_conflict)
5419 return ERR_PTR(-EAGAIN);
5421 nf = find_readable_file(fp);
5424 * We probably could attempt another open and get a read
5425 * delegation, but for now, don't bother until the
5426 * client actually sends us one.
5428 return ERR_PTR(-EAGAIN);
5430 spin_lock(&state_lock);
5431 spin_lock(&fp->fi_lock);
5432 if (nfs4_delegation_exists(clp, fp))
5434 else if (!fp->fi_deleg_file) {
5435 fp->fi_deleg_file = nf;
5436 /* increment early to prevent fi_deleg_file from being
5438 fp->fi_delegees = 1;
5442 spin_unlock(&fp->fi_lock);
5443 spin_unlock(&state_lock);
5447 return ERR_PTR(status);
5450 dp = alloc_init_deleg(clp, fp, odstate);
5454 fl = nfs4_alloc_init_lease(dp, NFS4_OPEN_DELEGATE_READ);
5456 goto out_clnt_odstate;
5458 status = vfs_setlease(fp->fi_deleg_file->nf_file, fl->fl_type, &fl, NULL);
5460 locks_free_lock(fl);
5462 goto out_clnt_odstate;
5465 status = nfsd4_verify_deleg_dentry(open, fp, parent);
5470 status = nfsd4_check_conflicting_opens(clp, fp);
5474 spin_lock(&state_lock);
5475 spin_lock(&fp->fi_lock);
5476 if (fp->fi_had_conflict)
5479 status = hash_delegation_locked(dp, fp);
5480 spin_unlock(&fp->fi_lock);
5481 spin_unlock(&state_lock);
5488 vfs_setlease(fp->fi_deleg_file->nf_file, F_UNLCK, NULL, (void **)&dp);
5490 put_clnt_odstate(dp->dl_clnt_odstate);
5491 nfs4_put_stid(&dp->dl_stid);
5494 return ERR_PTR(status);
5497 static void nfsd4_open_deleg_none_ext(struct nfsd4_open *open, int status)
5499 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
5500 if (status == -EAGAIN)
5501 open->op_why_no_deleg = WND4_CONTENTION;
5503 open->op_why_no_deleg = WND4_RESOURCE;
5504 switch (open->op_deleg_want) {
5505 case NFS4_SHARE_WANT_READ_DELEG:
5506 case NFS4_SHARE_WANT_WRITE_DELEG:
5507 case NFS4_SHARE_WANT_ANY_DELEG:
5509 case NFS4_SHARE_WANT_CANCEL:
5510 open->op_why_no_deleg = WND4_CANCELLED;
5512 case NFS4_SHARE_WANT_NO_DELEG:
5519 * Attempt to hand out a delegation.
5521 * Note we don't support write delegations, and won't until the vfs has
5522 * proper support for them.
5525 nfs4_open_delegation(struct nfsd4_open *open, struct nfs4_ol_stateid *stp,
5526 struct svc_fh *currentfh)
5528 struct nfs4_delegation *dp;
5529 struct nfs4_openowner *oo = openowner(stp->st_stateowner);
5530 struct nfs4_client *clp = stp->st_stid.sc_client;
5531 struct svc_fh *parent = NULL;
5535 cb_up = nfsd4_cb_channel_good(oo->oo_owner.so_client);
5536 open->op_recall = 0;
5537 switch (open->op_claim_type) {
5538 case NFS4_OPEN_CLAIM_PREVIOUS:
5540 open->op_recall = 1;
5541 if (open->op_delegate_type != NFS4_OPEN_DELEGATE_READ)
5544 case NFS4_OPEN_CLAIM_NULL:
5547 case NFS4_OPEN_CLAIM_FH:
5549 * Let's not give out any delegations till everyone's
5550 * had the chance to reclaim theirs, *and* until
5551 * NLM locks have all been reclaimed:
5553 if (locks_in_grace(clp->net))
5555 if (!cb_up || !(oo->oo_flags & NFS4_OO_CONFIRMED))
5561 dp = nfs4_set_delegation(open, stp, parent);
5565 memcpy(&open->op_delegate_stateid, &dp->dl_stid.sc_stateid, sizeof(dp->dl_stid.sc_stateid));
5567 trace_nfsd_deleg_read(&dp->dl_stid.sc_stateid);
5568 open->op_delegate_type = NFS4_OPEN_DELEGATE_READ;
5569 nfs4_put_stid(&dp->dl_stid);
5572 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE;
5573 if (open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS &&
5574 open->op_delegate_type != NFS4_OPEN_DELEGATE_NONE) {
5575 dprintk("NFSD: WARNING: refusing delegation reclaim\n");
5576 open->op_recall = 1;
5579 /* 4.1 client asking for a delegation? */
5580 if (open->op_deleg_want)
5581 nfsd4_open_deleg_none_ext(open, status);
5585 static void nfsd4_deleg_xgrade_none_ext(struct nfsd4_open *open,
5586 struct nfs4_delegation *dp)
5588 if (open->op_deleg_want == NFS4_SHARE_WANT_READ_DELEG &&
5589 dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
5590 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
5591 open->op_why_no_deleg = WND4_NOT_SUPP_DOWNGRADE;
5592 } else if (open->op_deleg_want == NFS4_SHARE_WANT_WRITE_DELEG &&
5593 dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
5594 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
5595 open->op_why_no_deleg = WND4_NOT_SUPP_UPGRADE;
5597 /* Otherwise the client must be confused wanting a delegation
5598 * it already has, therefore we don't return
5599 * NFS4_OPEN_DELEGATE_NONE_EXT and reason.
5604 * nfsd4_process_open2 - finish open processing
5605 * @rqstp: the RPC transaction being executed
5606 * @current_fh: NFSv4 COMPOUND's current filehandle
5607 * @open: OPEN arguments
5609 * If successful, (1) truncate the file if open->op_truncate was
5610 * set, (2) set open->op_stateid, (3) set open->op_delegation.
5612 * Returns %nfs_ok on success; otherwise an nfs4stat value in
5613 * network byte order is returned.
5616 nfsd4_process_open2(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_open *open)
5618 struct nfsd4_compoundres *resp = rqstp->rq_resp;
5619 struct nfs4_client *cl = open->op_openowner->oo_owner.so_client;
5620 struct nfs4_file *fp = NULL;
5621 struct nfs4_ol_stateid *stp = NULL;
5622 struct nfs4_delegation *dp = NULL;
5624 bool new_stp = false;
5627 * Lookup file; if found, lookup stateid and check open request,
5628 * and check for delegations in the process of being recalled.
5629 * If not found, create the nfs4_file struct
5631 fp = find_or_add_file(open->op_file, current_fh);
5632 if (fp != open->op_file) {
5633 status = nfs4_check_deleg(cl, open, &dp);
5636 stp = nfsd4_find_and_lock_existing_open(fp, open);
5638 open->op_file = NULL;
5639 status = nfserr_bad_stateid;
5640 if (nfsd4_is_deleg_cur(open))
5645 stp = init_open_stateid(fp, open);
5651 * OPEN the file, or upgrade an existing OPEN.
5652 * If truncate fails, the OPEN fails.
5654 * stp is already locked.
5657 /* Stateid was found, this is an OPEN upgrade */
5658 status = nfs4_upgrade_open(rqstp, fp, current_fh, stp, open);
5660 mutex_unlock(&stp->st_mutex);
5664 status = nfs4_get_vfs_file(rqstp, fp, current_fh, stp, open, true);
5666 stp->st_stid.sc_type = NFS4_CLOSED_STID;
5667 release_open_stateid(stp);
5668 mutex_unlock(&stp->st_mutex);
5672 stp->st_clnt_odstate = find_or_hash_clnt_odstate(fp,
5674 if (stp->st_clnt_odstate == open->op_odstate)
5675 open->op_odstate = NULL;
5678 nfs4_inc_and_copy_stateid(&open->op_stateid, &stp->st_stid);
5679 mutex_unlock(&stp->st_mutex);
5681 if (nfsd4_has_session(&resp->cstate)) {
5682 if (open->op_deleg_want & NFS4_SHARE_WANT_NO_DELEG) {
5683 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
5684 open->op_why_no_deleg = WND4_NOT_WANTED;
5690 * Attempt to hand out a delegation. No error return, because the
5691 * OPEN succeeds even if we fail.
5693 nfs4_open_delegation(open, stp, &resp->cstate.current_fh);
5696 trace_nfsd_open(&stp->st_stid.sc_stateid);
5698 /* 4.1 client trying to upgrade/downgrade delegation? */
5699 if (open->op_delegate_type == NFS4_OPEN_DELEGATE_NONE && dp &&
5700 open->op_deleg_want)
5701 nfsd4_deleg_xgrade_none_ext(open, dp);
5705 if (status == 0 && open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS)
5706 open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
5708 * To finish the open response, we just need to set the rflags.
5710 open->op_rflags = NFS4_OPEN_RESULT_LOCKTYPE_POSIX;
5711 if (nfsd4_has_session(&resp->cstate))
5712 open->op_rflags |= NFS4_OPEN_RESULT_MAY_NOTIFY_LOCK;
5713 else if (!(open->op_openowner->oo_flags & NFS4_OO_CONFIRMED))
5714 open->op_rflags |= NFS4_OPEN_RESULT_CONFIRM;
5717 nfs4_put_stid(&dp->dl_stid);
5719 nfs4_put_stid(&stp->st_stid);
5724 void nfsd4_cleanup_open_state(struct nfsd4_compound_state *cstate,
5725 struct nfsd4_open *open)
5727 if (open->op_openowner) {
5728 struct nfs4_stateowner *so = &open->op_openowner->oo_owner;
5730 nfsd4_cstate_assign_replay(cstate, so);
5731 nfs4_put_stateowner(so);
5734 kmem_cache_free(file_slab, open->op_file);
5736 nfs4_put_stid(&open->op_stp->st_stid);
5737 if (open->op_odstate)
5738 kmem_cache_free(odstate_slab, open->op_odstate);
5742 nfsd4_renew(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5743 union nfsd4_op_u *u)
5745 clientid_t *clid = &u->renew;
5746 struct nfs4_client *clp;
5748 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
5750 trace_nfsd_clid_renew(clid);
5751 status = set_client(clid, cstate, nn);
5755 if (!list_empty(&clp->cl_delegations)
5756 && clp->cl_cb_state != NFSD4_CB_UP)
5757 return nfserr_cb_path_down;
5762 nfsd4_end_grace(struct nfsd_net *nn)
5764 /* do nothing if grace period already ended */
5765 if (nn->grace_ended)
5768 trace_nfsd_grace_complete(nn);
5769 nn->grace_ended = true;
5771 * If the server goes down again right now, an NFSv4
5772 * client will still be allowed to reclaim after it comes back up,
5773 * even if it hasn't yet had a chance to reclaim state this time.
5776 nfsd4_record_grace_done(nn);
5778 * At this point, NFSv4 clients can still reclaim. But if the
5779 * server crashes, any that have not yet reclaimed will be out
5780 * of luck on the next boot.
5782 * (NFSv4.1+ clients are considered to have reclaimed once they
5783 * call RECLAIM_COMPLETE. NFSv4.0 clients are considered to
5784 * have reclaimed after their first OPEN.)
5786 locks_end_grace(&nn->nfsd4_manager);
5788 * At this point, and once lockd and/or any other containers
5789 * exit their grace period, further reclaims will fail and
5790 * regular locking can resume.
5795 * If we've waited a lease period but there are still clients trying to
5796 * reclaim, wait a little longer to give them a chance to finish.
5798 static bool clients_still_reclaiming(struct nfsd_net *nn)
5800 time64_t double_grace_period_end = nn->boot_time +
5801 2 * nn->nfsd4_lease;
5803 if (nn->track_reclaim_completes &&
5804 atomic_read(&nn->nr_reclaim_complete) ==
5805 nn->reclaim_str_hashtbl_size)
5807 if (!nn->somebody_reclaimed)
5809 nn->somebody_reclaimed = false;
5811 * If we've given them *two* lease times to reclaim, and they're
5812 * still not done, give up:
5814 if (ktime_get_boottime_seconds() > double_grace_period_end)
5819 struct laundry_time {
5824 static bool state_expired(struct laundry_time *lt, time64_t last_refresh)
5826 time64_t time_remaining;
5828 if (last_refresh < lt->cutoff)
5830 time_remaining = last_refresh - lt->cutoff;
5831 lt->new_timeo = min(lt->new_timeo, time_remaining);
5835 #ifdef CONFIG_NFSD_V4_2_INTER_SSC
5836 void nfsd4_ssc_init_umount_work(struct nfsd_net *nn)
5838 spin_lock_init(&nn->nfsd_ssc_lock);
5839 INIT_LIST_HEAD(&nn->nfsd_ssc_mount_list);
5840 init_waitqueue_head(&nn->nfsd_ssc_waitq);
5842 EXPORT_SYMBOL_GPL(nfsd4_ssc_init_umount_work);
5845 * This is called when nfsd is being shutdown, after all inter_ssc
5846 * cleanup were done, to destroy the ssc delayed unmount list.
5848 static void nfsd4_ssc_shutdown_umount(struct nfsd_net *nn)
5850 struct nfsd4_ssc_umount_item *ni = NULL;
5851 struct nfsd4_ssc_umount_item *tmp;
5853 spin_lock(&nn->nfsd_ssc_lock);
5854 list_for_each_entry_safe(ni, tmp, &nn->nfsd_ssc_mount_list, nsui_list) {
5855 list_del(&ni->nsui_list);
5856 spin_unlock(&nn->nfsd_ssc_lock);
5857 mntput(ni->nsui_vfsmount);
5859 spin_lock(&nn->nfsd_ssc_lock);
5861 spin_unlock(&nn->nfsd_ssc_lock);
5864 static void nfsd4_ssc_expire_umount(struct nfsd_net *nn)
5866 bool do_wakeup = false;
5867 struct nfsd4_ssc_umount_item *ni = NULL;
5868 struct nfsd4_ssc_umount_item *tmp;
5870 spin_lock(&nn->nfsd_ssc_lock);
5871 list_for_each_entry_safe(ni, tmp, &nn->nfsd_ssc_mount_list, nsui_list) {
5872 if (time_after(jiffies, ni->nsui_expire)) {
5873 if (refcount_read(&ni->nsui_refcnt) > 1)
5876 /* mark being unmount */
5877 ni->nsui_busy = true;
5878 spin_unlock(&nn->nfsd_ssc_lock);
5879 mntput(ni->nsui_vfsmount);
5880 spin_lock(&nn->nfsd_ssc_lock);
5882 /* waiters need to start from begin of list */
5883 list_del(&ni->nsui_list);
5886 /* wakeup ssc_connect waiters */
5893 wake_up_all(&nn->nfsd_ssc_waitq);
5894 spin_unlock(&nn->nfsd_ssc_lock);
5898 /* Check if any lock belonging to this lockowner has any blockers */
5900 nfs4_lockowner_has_blockers(struct nfs4_lockowner *lo)
5902 struct file_lock_context *ctx;
5903 struct nfs4_ol_stateid *stp;
5904 struct nfs4_file *nf;
5906 list_for_each_entry(stp, &lo->lo_owner.so_stateids, st_perstateowner) {
5907 nf = stp->st_stid.sc_file;
5908 ctx = nf->fi_inode->i_flctx;
5911 if (locks_owner_has_blockers(ctx, lo))
5918 nfs4_anylock_blockers(struct nfs4_client *clp)
5921 struct nfs4_stateowner *so;
5922 struct nfs4_lockowner *lo;
5924 if (atomic_read(&clp->cl_delegs_in_recall))
5926 spin_lock(&clp->cl_lock);
5927 for (i = 0; i < OWNER_HASH_SIZE; i++) {
5928 list_for_each_entry(so, &clp->cl_ownerstr_hashtbl[i],
5930 if (so->so_is_open_owner)
5933 if (nfs4_lockowner_has_blockers(lo)) {
5934 spin_unlock(&clp->cl_lock);
5939 spin_unlock(&clp->cl_lock);
5944 nfs4_get_client_reaplist(struct nfsd_net *nn, struct list_head *reaplist,
5945 struct laundry_time *lt)
5947 unsigned int maxreap, reapcnt = 0;
5948 struct list_head *pos, *next;
5949 struct nfs4_client *clp;
5951 maxreap = (atomic_read(&nn->nfs4_client_count) >= nn->nfs4_max_clients) ?
5952 NFSD_CLIENT_MAX_TRIM_PER_RUN : 0;
5953 INIT_LIST_HEAD(reaplist);
5954 spin_lock(&nn->client_lock);
5955 list_for_each_safe(pos, next, &nn->client_lru) {
5956 clp = list_entry(pos, struct nfs4_client, cl_lru);
5957 if (clp->cl_state == NFSD4_EXPIRABLE)
5959 if (!state_expired(lt, clp->cl_time))
5961 if (!atomic_read(&clp->cl_rpc_users)) {
5962 if (clp->cl_state == NFSD4_ACTIVE)
5963 atomic_inc(&nn->nfsd_courtesy_clients);
5964 clp->cl_state = NFSD4_COURTESY;
5966 if (!client_has_state(clp))
5968 if (!nfs4_anylock_blockers(clp))
5969 if (reapcnt >= maxreap)
5972 if (!mark_client_expired_locked(clp)) {
5973 list_add(&clp->cl_lru, reaplist);
5977 spin_unlock(&nn->client_lock);
5981 nfs4_get_courtesy_client_reaplist(struct nfsd_net *nn,
5982 struct list_head *reaplist)
5984 unsigned int maxreap = 0, reapcnt = 0;
5985 struct list_head *pos, *next;
5986 struct nfs4_client *clp;
5988 maxreap = NFSD_CLIENT_MAX_TRIM_PER_RUN;
5989 INIT_LIST_HEAD(reaplist);
5991 spin_lock(&nn->client_lock);
5992 list_for_each_safe(pos, next, &nn->client_lru) {
5993 clp = list_entry(pos, struct nfs4_client, cl_lru);
5994 if (clp->cl_state == NFSD4_ACTIVE)
5996 if (reapcnt >= maxreap)
5998 if (!mark_client_expired_locked(clp)) {
5999 list_add(&clp->cl_lru, reaplist);
6003 spin_unlock(&nn->client_lock);
6007 nfs4_process_client_reaplist(struct list_head *reaplist)
6009 struct list_head *pos, *next;
6010 struct nfs4_client *clp;
6012 list_for_each_safe(pos, next, reaplist) {
6013 clp = list_entry(pos, struct nfs4_client, cl_lru);
6014 trace_nfsd_clid_purged(&clp->cl_clientid);
6015 list_del_init(&clp->cl_lru);
6021 nfs4_laundromat(struct nfsd_net *nn)
6023 struct nfs4_openowner *oo;
6024 struct nfs4_delegation *dp;
6025 struct nfs4_ol_stateid *stp;
6026 struct nfsd4_blocked_lock *nbl;
6027 struct list_head *pos, *next, reaplist;
6028 struct laundry_time lt = {
6029 .cutoff = ktime_get_boottime_seconds() - nn->nfsd4_lease,
6030 .new_timeo = nn->nfsd4_lease
6032 struct nfs4_cpntf_state *cps;
6033 copy_stateid_t *cps_t;
6036 if (clients_still_reclaiming(nn)) {
6040 nfsd4_end_grace(nn);
6042 spin_lock(&nn->s2s_cp_lock);
6043 idr_for_each_entry(&nn->s2s_cp_stateids, cps_t, i) {
6044 cps = container_of(cps_t, struct nfs4_cpntf_state, cp_stateid);
6045 if (cps->cp_stateid.cs_type == NFS4_COPYNOTIFY_STID &&
6046 state_expired(<, cps->cpntf_time))
6047 _free_cpntf_state_locked(nn, cps);
6049 spin_unlock(&nn->s2s_cp_lock);
6050 nfs4_get_client_reaplist(nn, &reaplist, <);
6051 nfs4_process_client_reaplist(&reaplist);
6053 spin_lock(&state_lock);
6054 list_for_each_safe(pos, next, &nn->del_recall_lru) {
6055 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
6056 if (!state_expired(<, dp->dl_time))
6058 WARN_ON(!unhash_delegation_locked(dp));
6059 list_add(&dp->dl_recall_lru, &reaplist);
6061 spin_unlock(&state_lock);
6062 while (!list_empty(&reaplist)) {
6063 dp = list_first_entry(&reaplist, struct nfs4_delegation,
6065 list_del_init(&dp->dl_recall_lru);
6066 revoke_delegation(dp);
6069 spin_lock(&nn->client_lock);
6070 while (!list_empty(&nn->close_lru)) {
6071 oo = list_first_entry(&nn->close_lru, struct nfs4_openowner,
6073 if (!state_expired(<, oo->oo_time))
6075 list_del_init(&oo->oo_close_lru);
6076 stp = oo->oo_last_closed_stid;
6077 oo->oo_last_closed_stid = NULL;
6078 spin_unlock(&nn->client_lock);
6079 nfs4_put_stid(&stp->st_stid);
6080 spin_lock(&nn->client_lock);
6082 spin_unlock(&nn->client_lock);
6085 * It's possible for a client to try and acquire an already held lock
6086 * that is being held for a long time, and then lose interest in it.
6087 * So, we clean out any un-revisited request after a lease period
6088 * under the assumption that the client is no longer interested.
6090 * RFC5661, sec. 9.6 states that the client must not rely on getting
6091 * notifications and must continue to poll for locks, even when the
6092 * server supports them. Thus this shouldn't lead to clients blocking
6093 * indefinitely once the lock does become free.
6095 BUG_ON(!list_empty(&reaplist));
6096 spin_lock(&nn->blocked_locks_lock);
6097 while (!list_empty(&nn->blocked_locks_lru)) {
6098 nbl = list_first_entry(&nn->blocked_locks_lru,
6099 struct nfsd4_blocked_lock, nbl_lru);
6100 if (!state_expired(<, nbl->nbl_time))
6102 list_move(&nbl->nbl_lru, &reaplist);
6103 list_del_init(&nbl->nbl_list);
6105 spin_unlock(&nn->blocked_locks_lock);
6107 while (!list_empty(&reaplist)) {
6108 nbl = list_first_entry(&reaplist,
6109 struct nfsd4_blocked_lock, nbl_lru);
6110 list_del_init(&nbl->nbl_lru);
6111 free_blocked_lock(nbl);
6113 #ifdef CONFIG_NFSD_V4_2_INTER_SSC
6114 /* service the server-to-server copy delayed unmount list */
6115 nfsd4_ssc_expire_umount(nn);
6118 return max_t(time64_t, lt.new_timeo, NFSD_LAUNDROMAT_MINTIMEOUT);
6121 static void laundromat_main(struct work_struct *);
6124 laundromat_main(struct work_struct *laundry)
6127 struct delayed_work *dwork = to_delayed_work(laundry);
6128 struct nfsd_net *nn = container_of(dwork, struct nfsd_net,
6131 t = nfs4_laundromat(nn);
6132 queue_delayed_work(laundry_wq, &nn->laundromat_work, t*HZ);
6136 courtesy_client_reaper(struct work_struct *reaper)
6138 struct list_head reaplist;
6139 struct delayed_work *dwork = to_delayed_work(reaper);
6140 struct nfsd_net *nn = container_of(dwork, struct nfsd_net,
6141 nfsd_shrinker_work);
6143 nfs4_get_courtesy_client_reaplist(nn, &reaplist);
6144 nfs4_process_client_reaplist(&reaplist);
6147 static inline __be32 nfs4_check_fh(struct svc_fh *fhp, struct nfs4_stid *stp)
6149 if (!fh_match(&fhp->fh_handle, &stp->sc_file->fi_fhandle))
6150 return nfserr_bad_stateid;
6155 __be32 nfs4_check_openmode(struct nfs4_ol_stateid *stp, int flags)
6157 __be32 status = nfserr_openmode;
6159 /* For lock stateid's, we test the parent open, not the lock: */
6160 if (stp->st_openstp)
6161 stp = stp->st_openstp;
6162 if ((flags & WR_STATE) && !access_permit_write(stp))
6164 if ((flags & RD_STATE) && !access_permit_read(stp))
6171 static inline __be32
6172 check_special_stateids(struct net *net, svc_fh *current_fh, stateid_t *stateid, int flags)
6174 if (ONE_STATEID(stateid) && (flags & RD_STATE))
6176 else if (opens_in_grace(net)) {
6177 /* Answer in remaining cases depends on existence of
6178 * conflicting state; so we must wait out the grace period. */
6179 return nfserr_grace;
6180 } else if (flags & WR_STATE)
6181 return nfs4_share_conflict(current_fh,
6182 NFS4_SHARE_DENY_WRITE);
6183 else /* (flags & RD_STATE) && ZERO_STATEID(stateid) */
6184 return nfs4_share_conflict(current_fh,
6185 NFS4_SHARE_DENY_READ);
6188 static __be32 check_stateid_generation(stateid_t *in, stateid_t *ref, bool has_session)
6191 * When sessions are used the stateid generation number is ignored
6194 if (has_session && in->si_generation == 0)
6197 if (in->si_generation == ref->si_generation)
6200 /* If the client sends us a stateid from the future, it's buggy: */
6201 if (nfsd4_stateid_generation_after(in, ref))
6202 return nfserr_bad_stateid;
6204 * However, we could see a stateid from the past, even from a
6205 * non-buggy client. For example, if the client sends a lock
6206 * while some IO is outstanding, the lock may bump si_generation
6207 * while the IO is still in flight. The client could avoid that
6208 * situation by waiting for responses on all the IO requests,
6209 * but better performance may result in retrying IO that
6210 * receives an old_stateid error if requests are rarely
6211 * reordered in flight:
6213 return nfserr_old_stateid;
6216 static __be32 nfsd4_stid_check_stateid_generation(stateid_t *in, struct nfs4_stid *s, bool has_session)
6220 spin_lock(&s->sc_lock);
6221 ret = nfsd4_verify_open_stid(s);
6223 ret = check_stateid_generation(in, &s->sc_stateid, has_session);
6224 spin_unlock(&s->sc_lock);
6228 static __be32 nfsd4_check_openowner_confirmed(struct nfs4_ol_stateid *ols)
6230 if (ols->st_stateowner->so_is_open_owner &&
6231 !(openowner(ols->st_stateowner)->oo_flags & NFS4_OO_CONFIRMED))
6232 return nfserr_bad_stateid;
6236 static __be32 nfsd4_validate_stateid(struct nfs4_client *cl, stateid_t *stateid)
6238 struct nfs4_stid *s;
6239 __be32 status = nfserr_bad_stateid;
6241 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid) ||
6242 CLOSE_STATEID(stateid))
6244 if (!same_clid(&stateid->si_opaque.so_clid, &cl->cl_clientid))
6246 spin_lock(&cl->cl_lock);
6247 s = find_stateid_locked(cl, stateid);
6250 status = nfsd4_stid_check_stateid_generation(stateid, s, 1);
6253 switch (s->sc_type) {
6254 case NFS4_DELEG_STID:
6257 case NFS4_REVOKED_DELEG_STID:
6258 status = nfserr_deleg_revoked;
6260 case NFS4_OPEN_STID:
6261 case NFS4_LOCK_STID:
6262 status = nfsd4_check_openowner_confirmed(openlockstateid(s));
6265 printk("unknown stateid type %x\n", s->sc_type);
6267 case NFS4_CLOSED_STID:
6268 case NFS4_CLOSED_DELEG_STID:
6269 status = nfserr_bad_stateid;
6272 spin_unlock(&cl->cl_lock);
6277 nfsd4_lookup_stateid(struct nfsd4_compound_state *cstate,
6278 stateid_t *stateid, unsigned char typemask,
6279 struct nfs4_stid **s, struct nfsd_net *nn)
6282 struct nfs4_stid *stid;
6283 bool return_revoked = false;
6286 * only return revoked delegations if explicitly asked.
6287 * otherwise we report revoked or bad_stateid status.
6289 if (typemask & NFS4_REVOKED_DELEG_STID)
6290 return_revoked = true;
6291 else if (typemask & NFS4_DELEG_STID)
6292 typemask |= NFS4_REVOKED_DELEG_STID;
6294 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid) ||
6295 CLOSE_STATEID(stateid))
6296 return nfserr_bad_stateid;
6297 status = set_client(&stateid->si_opaque.so_clid, cstate, nn);
6298 if (status == nfserr_stale_clientid) {
6299 if (cstate->session)
6300 return nfserr_bad_stateid;
6301 return nfserr_stale_stateid;
6305 stid = find_stateid_by_type(cstate->clp, stateid, typemask);
6307 return nfserr_bad_stateid;
6308 if ((stid->sc_type == NFS4_REVOKED_DELEG_STID) && !return_revoked) {
6309 nfs4_put_stid(stid);
6310 if (cstate->minorversion)
6311 return nfserr_deleg_revoked;
6312 return nfserr_bad_stateid;
6318 static struct nfsd_file *
6319 nfs4_find_file(struct nfs4_stid *s, int flags)
6324 switch (s->sc_type) {
6325 case NFS4_DELEG_STID:
6326 if (WARN_ON_ONCE(!s->sc_file->fi_deleg_file))
6328 return nfsd_file_get(s->sc_file->fi_deleg_file);
6329 case NFS4_OPEN_STID:
6330 case NFS4_LOCK_STID:
6331 if (flags & RD_STATE)
6332 return find_readable_file(s->sc_file);
6334 return find_writeable_file(s->sc_file);
6341 nfs4_check_olstateid(struct nfs4_ol_stateid *ols, int flags)
6345 status = nfsd4_check_openowner_confirmed(ols);
6348 return nfs4_check_openmode(ols, flags);
6352 nfs4_check_file(struct svc_rqst *rqstp, struct svc_fh *fhp, struct nfs4_stid *s,
6353 struct nfsd_file **nfp, int flags)
6355 int acc = (flags & RD_STATE) ? NFSD_MAY_READ : NFSD_MAY_WRITE;
6356 struct nfsd_file *nf;
6359 nf = nfs4_find_file(s, flags);
6361 status = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
6362 acc | NFSD_MAY_OWNER_OVERRIDE);
6368 status = nfsd_file_acquire(rqstp, fhp, acc, &nf);
6377 _free_cpntf_state_locked(struct nfsd_net *nn, struct nfs4_cpntf_state *cps)
6379 WARN_ON_ONCE(cps->cp_stateid.cs_type != NFS4_COPYNOTIFY_STID);
6380 if (!refcount_dec_and_test(&cps->cp_stateid.cs_count))
6382 list_del(&cps->cp_list);
6383 idr_remove(&nn->s2s_cp_stateids,
6384 cps->cp_stateid.cs_stid.si_opaque.so_id);
6388 * A READ from an inter server to server COPY will have a
6389 * copy stateid. Look up the copy notify stateid from the
6390 * idr structure and take a reference on it.
6392 __be32 manage_cpntf_state(struct nfsd_net *nn, stateid_t *st,
6393 struct nfs4_client *clp,
6394 struct nfs4_cpntf_state **cps)
6396 copy_stateid_t *cps_t;
6397 struct nfs4_cpntf_state *state = NULL;
6399 if (st->si_opaque.so_clid.cl_id != nn->s2s_cp_cl_id)
6400 return nfserr_bad_stateid;
6401 spin_lock(&nn->s2s_cp_lock);
6402 cps_t = idr_find(&nn->s2s_cp_stateids, st->si_opaque.so_id);
6404 state = container_of(cps_t, struct nfs4_cpntf_state,
6406 if (state->cp_stateid.cs_type != NFS4_COPYNOTIFY_STID) {
6411 refcount_inc(&state->cp_stateid.cs_count);
6413 _free_cpntf_state_locked(nn, state);
6416 spin_unlock(&nn->s2s_cp_lock);
6418 return nfserr_bad_stateid;
6424 static __be32 find_cpntf_state(struct nfsd_net *nn, stateid_t *st,
6425 struct nfs4_stid **stid)
6428 struct nfs4_cpntf_state *cps = NULL;
6429 struct nfs4_client *found;
6431 status = manage_cpntf_state(nn, st, NULL, &cps);
6435 cps->cpntf_time = ktime_get_boottime_seconds();
6437 status = nfserr_expired;
6438 found = lookup_clientid(&cps->cp_p_clid, true, nn);
6442 *stid = find_stateid_by_type(found, &cps->cp_p_stateid,
6443 NFS4_DELEG_STID|NFS4_OPEN_STID|NFS4_LOCK_STID);
6447 status = nfserr_bad_stateid;
6449 put_client_renew(found);
6451 nfs4_put_cpntf_state(nn, cps);
6455 void nfs4_put_cpntf_state(struct nfsd_net *nn, struct nfs4_cpntf_state *cps)
6457 spin_lock(&nn->s2s_cp_lock);
6458 _free_cpntf_state_locked(nn, cps);
6459 spin_unlock(&nn->s2s_cp_lock);
6463 * Checks for stateid operations
6466 nfs4_preprocess_stateid_op(struct svc_rqst *rqstp,
6467 struct nfsd4_compound_state *cstate, struct svc_fh *fhp,
6468 stateid_t *stateid, int flags, struct nfsd_file **nfp,
6469 struct nfs4_stid **cstid)
6471 struct net *net = SVC_NET(rqstp);
6472 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
6473 struct nfs4_stid *s = NULL;
6479 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid)) {
6481 status = nfserr_bad_stateid;
6483 status = check_special_stateids(net, fhp, stateid,
6488 status = nfsd4_lookup_stateid(cstate, stateid,
6489 NFS4_DELEG_STID|NFS4_OPEN_STID|NFS4_LOCK_STID,
6491 if (status == nfserr_bad_stateid)
6492 status = find_cpntf_state(nn, stateid, &s);
6495 status = nfsd4_stid_check_stateid_generation(stateid, s,
6496 nfsd4_has_session(cstate));
6500 switch (s->sc_type) {
6501 case NFS4_DELEG_STID:
6502 status = nfs4_check_delegmode(delegstateid(s), flags);
6504 case NFS4_OPEN_STID:
6505 case NFS4_LOCK_STID:
6506 status = nfs4_check_olstateid(openlockstateid(s), flags);
6509 status = nfserr_bad_stateid;
6514 status = nfs4_check_fh(fhp, s);
6517 if (status == nfs_ok && nfp)
6518 status = nfs4_check_file(rqstp, fhp, s, nfp, flags);
6521 if (!status && cstid)
6530 * Test if the stateid is valid
6533 nfsd4_test_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
6534 union nfsd4_op_u *u)
6536 struct nfsd4_test_stateid *test_stateid = &u->test_stateid;
6537 struct nfsd4_test_stateid_id *stateid;
6538 struct nfs4_client *cl = cstate->clp;
6540 list_for_each_entry(stateid, &test_stateid->ts_stateid_list, ts_id_list)
6541 stateid->ts_id_status =
6542 nfsd4_validate_stateid(cl, &stateid->ts_id_stateid);
6548 nfsd4_free_lock_stateid(stateid_t *stateid, struct nfs4_stid *s)
6550 struct nfs4_ol_stateid *stp = openlockstateid(s);
6553 ret = nfsd4_lock_ol_stateid(stp);
6557 ret = check_stateid_generation(stateid, &s->sc_stateid, 1);
6561 ret = nfserr_locks_held;
6562 if (check_for_locks(stp->st_stid.sc_file,
6563 lockowner(stp->st_stateowner)))
6566 release_lock_stateid(stp);
6570 mutex_unlock(&stp->st_mutex);
6577 nfsd4_free_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
6578 union nfsd4_op_u *u)
6580 struct nfsd4_free_stateid *free_stateid = &u->free_stateid;
6581 stateid_t *stateid = &free_stateid->fr_stateid;
6582 struct nfs4_stid *s;
6583 struct nfs4_delegation *dp;
6584 struct nfs4_client *cl = cstate->clp;
6585 __be32 ret = nfserr_bad_stateid;
6587 spin_lock(&cl->cl_lock);
6588 s = find_stateid_locked(cl, stateid);
6591 spin_lock(&s->sc_lock);
6592 switch (s->sc_type) {
6593 case NFS4_DELEG_STID:
6594 ret = nfserr_locks_held;
6596 case NFS4_OPEN_STID:
6597 ret = check_stateid_generation(stateid, &s->sc_stateid, 1);
6600 ret = nfserr_locks_held;
6602 case NFS4_LOCK_STID:
6603 spin_unlock(&s->sc_lock);
6604 refcount_inc(&s->sc_count);
6605 spin_unlock(&cl->cl_lock);
6606 ret = nfsd4_free_lock_stateid(stateid, s);
6608 case NFS4_REVOKED_DELEG_STID:
6609 spin_unlock(&s->sc_lock);
6610 dp = delegstateid(s);
6611 list_del_init(&dp->dl_recall_lru);
6612 spin_unlock(&cl->cl_lock);
6616 /* Default falls through and returns nfserr_bad_stateid */
6618 spin_unlock(&s->sc_lock);
6620 spin_unlock(&cl->cl_lock);
6628 return (type == NFS4_READW_LT || type == NFS4_READ_LT) ?
6629 RD_STATE : WR_STATE;
6632 static __be32 nfs4_seqid_op_checks(struct nfsd4_compound_state *cstate, stateid_t *stateid, u32 seqid, struct nfs4_ol_stateid *stp)
6634 struct svc_fh *current_fh = &cstate->current_fh;
6635 struct nfs4_stateowner *sop = stp->st_stateowner;
6638 status = nfsd4_check_seqid(cstate, sop, seqid);
6641 status = nfsd4_lock_ol_stateid(stp);
6642 if (status != nfs_ok)
6644 status = check_stateid_generation(stateid, &stp->st_stid.sc_stateid, nfsd4_has_session(cstate));
6645 if (status == nfs_ok)
6646 status = nfs4_check_fh(current_fh, &stp->st_stid);
6647 if (status != nfs_ok)
6648 mutex_unlock(&stp->st_mutex);
6653 * Checks for sequence id mutating operations.
6656 nfs4_preprocess_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
6657 stateid_t *stateid, char typemask,
6658 struct nfs4_ol_stateid **stpp,
6659 struct nfsd_net *nn)
6662 struct nfs4_stid *s;
6663 struct nfs4_ol_stateid *stp = NULL;
6665 trace_nfsd_preprocess(seqid, stateid);
6668 status = nfsd4_lookup_stateid(cstate, stateid, typemask, &s, nn);
6671 stp = openlockstateid(s);
6672 nfsd4_cstate_assign_replay(cstate, stp->st_stateowner);
6674 status = nfs4_seqid_op_checks(cstate, stateid, seqid, stp);
6678 nfs4_put_stid(&stp->st_stid);
6682 static __be32 nfs4_preprocess_confirmed_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
6683 stateid_t *stateid, struct nfs4_ol_stateid **stpp, struct nfsd_net *nn)
6686 struct nfs4_openowner *oo;
6687 struct nfs4_ol_stateid *stp;
6689 status = nfs4_preprocess_seqid_op(cstate, seqid, stateid,
6690 NFS4_OPEN_STID, &stp, nn);
6693 oo = openowner(stp->st_stateowner);
6694 if (!(oo->oo_flags & NFS4_OO_CONFIRMED)) {
6695 mutex_unlock(&stp->st_mutex);
6696 nfs4_put_stid(&stp->st_stid);
6697 return nfserr_bad_stateid;
6704 nfsd4_open_confirm(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
6705 union nfsd4_op_u *u)
6707 struct nfsd4_open_confirm *oc = &u->open_confirm;
6709 struct nfs4_openowner *oo;
6710 struct nfs4_ol_stateid *stp;
6711 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
6713 dprintk("NFSD: nfsd4_open_confirm on file %pd\n",
6714 cstate->current_fh.fh_dentry);
6716 status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0);
6720 status = nfs4_preprocess_seqid_op(cstate,
6721 oc->oc_seqid, &oc->oc_req_stateid,
6722 NFS4_OPEN_STID, &stp, nn);
6725 oo = openowner(stp->st_stateowner);
6726 status = nfserr_bad_stateid;
6727 if (oo->oo_flags & NFS4_OO_CONFIRMED) {
6728 mutex_unlock(&stp->st_mutex);
6731 oo->oo_flags |= NFS4_OO_CONFIRMED;
6732 nfs4_inc_and_copy_stateid(&oc->oc_resp_stateid, &stp->st_stid);
6733 mutex_unlock(&stp->st_mutex);
6734 trace_nfsd_open_confirm(oc->oc_seqid, &stp->st_stid.sc_stateid);
6735 nfsd4_client_record_create(oo->oo_owner.so_client);
6738 nfs4_put_stid(&stp->st_stid);
6740 nfsd4_bump_seqid(cstate, status);
6744 static inline void nfs4_stateid_downgrade_bit(struct nfs4_ol_stateid *stp, u32 access)
6746 if (!test_access(access, stp))
6748 nfs4_file_put_access(stp->st_stid.sc_file, access);
6749 clear_access(access, stp);
6752 static inline void nfs4_stateid_downgrade(struct nfs4_ol_stateid *stp, u32 to_access)
6754 switch (to_access) {
6755 case NFS4_SHARE_ACCESS_READ:
6756 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_WRITE);
6757 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
6759 case NFS4_SHARE_ACCESS_WRITE:
6760 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_READ);
6761 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
6763 case NFS4_SHARE_ACCESS_BOTH:
6771 nfsd4_open_downgrade(struct svc_rqst *rqstp,
6772 struct nfsd4_compound_state *cstate, union nfsd4_op_u *u)
6774 struct nfsd4_open_downgrade *od = &u->open_downgrade;
6776 struct nfs4_ol_stateid *stp;
6777 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
6779 dprintk("NFSD: nfsd4_open_downgrade on file %pd\n",
6780 cstate->current_fh.fh_dentry);
6782 /* We don't yet support WANT bits: */
6783 if (od->od_deleg_want)
6784 dprintk("NFSD: %s: od_deleg_want=0x%x ignored\n", __func__,
6787 status = nfs4_preprocess_confirmed_seqid_op(cstate, od->od_seqid,
6788 &od->od_stateid, &stp, nn);
6791 status = nfserr_inval;
6792 if (!test_access(od->od_share_access, stp)) {
6793 dprintk("NFSD: access not a subset of current bitmap: 0x%hhx, input access=%08x\n",
6794 stp->st_access_bmap, od->od_share_access);
6797 if (!test_deny(od->od_share_deny, stp)) {
6798 dprintk("NFSD: deny not a subset of current bitmap: 0x%hhx, input deny=%08x\n",
6799 stp->st_deny_bmap, od->od_share_deny);
6802 nfs4_stateid_downgrade(stp, od->od_share_access);
6803 reset_union_bmap_deny(od->od_share_deny, stp);
6804 nfs4_inc_and_copy_stateid(&od->od_stateid, &stp->st_stid);
6807 mutex_unlock(&stp->st_mutex);
6808 nfs4_put_stid(&stp->st_stid);
6810 nfsd4_bump_seqid(cstate, status);
6814 static void nfsd4_close_open_stateid(struct nfs4_ol_stateid *s)
6816 struct nfs4_client *clp = s->st_stid.sc_client;
6818 LIST_HEAD(reaplist);
6819 struct nfs4_ol_stateid *stp;
6821 spin_lock(&clp->cl_lock);
6822 unhashed = unhash_open_stateid(s, &reaplist);
6824 if (clp->cl_minorversion) {
6826 put_ol_stateid_locked(s, &reaplist);
6827 spin_unlock(&clp->cl_lock);
6828 list_for_each_entry(stp, &reaplist, st_locks)
6829 nfs4_free_cpntf_statelist(clp->net, &stp->st_stid);
6830 free_ol_stateid_reaplist(&reaplist);
6832 spin_unlock(&clp->cl_lock);
6833 free_ol_stateid_reaplist(&reaplist);
6835 move_to_close_lru(s, clp->net);
6840 * nfs4_unlock_state() called after encode
6843 nfsd4_close(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
6844 union nfsd4_op_u *u)
6846 struct nfsd4_close *close = &u->close;
6848 struct nfs4_ol_stateid *stp;
6849 struct net *net = SVC_NET(rqstp);
6850 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
6852 dprintk("NFSD: nfsd4_close on file %pd\n",
6853 cstate->current_fh.fh_dentry);
6855 status = nfs4_preprocess_seqid_op(cstate, close->cl_seqid,
6857 NFS4_OPEN_STID|NFS4_CLOSED_STID,
6859 nfsd4_bump_seqid(cstate, status);
6863 stp->st_stid.sc_type = NFS4_CLOSED_STID;
6866 * Technically we don't _really_ have to increment or copy it, since
6867 * it should just be gone after this operation and we clobber the
6868 * copied value below, but we continue to do so here just to ensure
6869 * that racing ops see that there was a state change.
6871 nfs4_inc_and_copy_stateid(&close->cl_stateid, &stp->st_stid);
6873 nfsd4_close_open_stateid(stp);
6874 mutex_unlock(&stp->st_mutex);
6876 /* v4.1+ suggests that we send a special stateid in here, since the
6877 * clients should just ignore this anyway. Since this is not useful
6878 * for v4.0 clients either, we set it to the special close_stateid
6881 * See RFC5661 section 18.2.4, and RFC7530 section 16.2.5
6883 memcpy(&close->cl_stateid, &close_stateid, sizeof(close->cl_stateid));
6885 /* put reference from nfs4_preprocess_seqid_op */
6886 nfs4_put_stid(&stp->st_stid);
6892 nfsd4_delegreturn(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
6893 union nfsd4_op_u *u)
6895 struct nfsd4_delegreturn *dr = &u->delegreturn;
6896 struct nfs4_delegation *dp;
6897 stateid_t *stateid = &dr->dr_stateid;
6898 struct nfs4_stid *s;
6900 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
6902 if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
6905 status = nfsd4_lookup_stateid(cstate, stateid, NFS4_DELEG_STID, &s, nn);
6908 dp = delegstateid(s);
6909 status = nfsd4_stid_check_stateid_generation(stateid, &dp->dl_stid, nfsd4_has_session(cstate));
6913 wake_up_var(d_inode(cstate->current_fh.fh_dentry));
6914 destroy_delegation(dp);
6916 nfs4_put_stid(&dp->dl_stid);
6921 /* last octet in a range */
6923 last_byte_offset(u64 start, u64 len)
6929 return end > start ? end - 1: NFS4_MAX_UINT64;
6933 * TODO: Linux file offsets are _signed_ 64-bit quantities, which means that
6934 * we can't properly handle lock requests that go beyond the (2^63 - 1)-th
6935 * byte, because of sign extension problems. Since NFSv4 calls for 64-bit
6936 * locking, this prevents us from being completely protocol-compliant. The
6937 * real solution to this problem is to start using unsigned file offsets in
6938 * the VFS, but this is a very deep change!
6941 nfs4_transform_lock_offset(struct file_lock *lock)
6943 if (lock->fl_start < 0)
6944 lock->fl_start = OFFSET_MAX;
6945 if (lock->fl_end < 0)
6946 lock->fl_end = OFFSET_MAX;
6950 nfsd4_lm_get_owner(fl_owner_t owner)
6952 struct nfs4_lockowner *lo = (struct nfs4_lockowner *)owner;
6954 nfs4_get_stateowner(&lo->lo_owner);
6959 nfsd4_lm_put_owner(fl_owner_t owner)
6961 struct nfs4_lockowner *lo = (struct nfs4_lockowner *)owner;
6964 nfs4_put_stateowner(&lo->lo_owner);
6967 /* return pointer to struct nfs4_client if client is expirable */
6969 nfsd4_lm_lock_expirable(struct file_lock *cfl)
6971 struct nfs4_lockowner *lo = (struct nfs4_lockowner *)cfl->fl_owner;
6972 struct nfs4_client *clp = lo->lo_owner.so_client;
6973 struct nfsd_net *nn;
6975 if (try_to_expire_client(clp)) {
6976 nn = net_generic(clp->net, nfsd_net_id);
6977 mod_delayed_work(laundry_wq, &nn->laundromat_work, 0);
6983 /* schedule laundromat to run immediately and wait for it to complete */
6985 nfsd4_lm_expire_lock(void)
6987 flush_workqueue(laundry_wq);
6991 nfsd4_lm_notify(struct file_lock *fl)
6993 struct nfs4_lockowner *lo = (struct nfs4_lockowner *)fl->fl_owner;
6994 struct net *net = lo->lo_owner.so_client->net;
6995 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
6996 struct nfsd4_blocked_lock *nbl = container_of(fl,
6997 struct nfsd4_blocked_lock, nbl_lock);
7000 /* An empty list means that something else is going to be using it */
7001 spin_lock(&nn->blocked_locks_lock);
7002 if (!list_empty(&nbl->nbl_list)) {
7003 list_del_init(&nbl->nbl_list);
7004 list_del_init(&nbl->nbl_lru);
7007 spin_unlock(&nn->blocked_locks_lock);
7010 trace_nfsd_cb_notify_lock(lo, nbl);
7011 nfsd4_run_cb(&nbl->nbl_cb);
7015 static const struct lock_manager_operations nfsd_posix_mng_ops = {
7016 .lm_mod_owner = THIS_MODULE,
7017 .lm_notify = nfsd4_lm_notify,
7018 .lm_get_owner = nfsd4_lm_get_owner,
7019 .lm_put_owner = nfsd4_lm_put_owner,
7020 .lm_lock_expirable = nfsd4_lm_lock_expirable,
7021 .lm_expire_lock = nfsd4_lm_expire_lock,
7025 nfs4_set_lock_denied(struct file_lock *fl, struct nfsd4_lock_denied *deny)
7027 struct nfs4_lockowner *lo;
7029 if (fl->fl_lmops == &nfsd_posix_mng_ops) {
7030 lo = (struct nfs4_lockowner *) fl->fl_owner;
7031 xdr_netobj_dup(&deny->ld_owner, &lo->lo_owner.so_owner,
7033 if (!deny->ld_owner.data)
7034 /* We just don't care that much */
7036 deny->ld_clientid = lo->lo_owner.so_client->cl_clientid;
7039 deny->ld_owner.len = 0;
7040 deny->ld_owner.data = NULL;
7041 deny->ld_clientid.cl_boot = 0;
7042 deny->ld_clientid.cl_id = 0;
7044 deny->ld_start = fl->fl_start;
7045 deny->ld_length = NFS4_MAX_UINT64;
7046 if (fl->fl_end != NFS4_MAX_UINT64)
7047 deny->ld_length = fl->fl_end - fl->fl_start + 1;
7048 deny->ld_type = NFS4_READ_LT;
7049 if (fl->fl_type != F_RDLCK)
7050 deny->ld_type = NFS4_WRITE_LT;
7053 static struct nfs4_lockowner *
7054 find_lockowner_str_locked(struct nfs4_client *clp, struct xdr_netobj *owner)
7056 unsigned int strhashval = ownerstr_hashval(owner);
7057 struct nfs4_stateowner *so;
7059 lockdep_assert_held(&clp->cl_lock);
7061 list_for_each_entry(so, &clp->cl_ownerstr_hashtbl[strhashval],
7063 if (so->so_is_open_owner)
7065 if (same_owner_str(so, owner))
7066 return lockowner(nfs4_get_stateowner(so));
7071 static struct nfs4_lockowner *
7072 find_lockowner_str(struct nfs4_client *clp, struct xdr_netobj *owner)
7074 struct nfs4_lockowner *lo;
7076 spin_lock(&clp->cl_lock);
7077 lo = find_lockowner_str_locked(clp, owner);
7078 spin_unlock(&clp->cl_lock);
7082 static void nfs4_unhash_lockowner(struct nfs4_stateowner *sop)
7084 unhash_lockowner_locked(lockowner(sop));
7087 static void nfs4_free_lockowner(struct nfs4_stateowner *sop)
7089 struct nfs4_lockowner *lo = lockowner(sop);
7091 kmem_cache_free(lockowner_slab, lo);
7094 static const struct nfs4_stateowner_operations lockowner_ops = {
7095 .so_unhash = nfs4_unhash_lockowner,
7096 .so_free = nfs4_free_lockowner,
7100 * Alloc a lock owner structure.
7101 * Called in nfsd4_lock - therefore, OPEN and OPEN_CONFIRM (if needed) has
7104 * strhashval = ownerstr_hashval
7106 static struct nfs4_lockowner *
7107 alloc_init_lock_stateowner(unsigned int strhashval, struct nfs4_client *clp,
7108 struct nfs4_ol_stateid *open_stp,
7109 struct nfsd4_lock *lock)
7111 struct nfs4_lockowner *lo, *ret;
7113 lo = alloc_stateowner(lockowner_slab, &lock->lk_new_owner, clp);
7116 INIT_LIST_HEAD(&lo->lo_blocked);
7117 INIT_LIST_HEAD(&lo->lo_owner.so_stateids);
7118 lo->lo_owner.so_is_open_owner = 0;
7119 lo->lo_owner.so_seqid = lock->lk_new_lock_seqid;
7120 lo->lo_owner.so_ops = &lockowner_ops;
7121 spin_lock(&clp->cl_lock);
7122 ret = find_lockowner_str_locked(clp, &lock->lk_new_owner);
7124 list_add(&lo->lo_owner.so_strhash,
7125 &clp->cl_ownerstr_hashtbl[strhashval]);
7128 nfs4_free_stateowner(&lo->lo_owner);
7130 spin_unlock(&clp->cl_lock);
7134 static struct nfs4_ol_stateid *
7135 find_lock_stateid(const struct nfs4_lockowner *lo,
7136 const struct nfs4_ol_stateid *ost)
7138 struct nfs4_ol_stateid *lst;
7140 lockdep_assert_held(&ost->st_stid.sc_client->cl_lock);
7142 /* If ost is not hashed, ost->st_locks will not be valid */
7143 if (!nfs4_ol_stateid_unhashed(ost))
7144 list_for_each_entry(lst, &ost->st_locks, st_locks) {
7145 if (lst->st_stateowner == &lo->lo_owner) {
7146 refcount_inc(&lst->st_stid.sc_count);
7153 static struct nfs4_ol_stateid *
7154 init_lock_stateid(struct nfs4_ol_stateid *stp, struct nfs4_lockowner *lo,
7155 struct nfs4_file *fp, struct inode *inode,
7156 struct nfs4_ol_stateid *open_stp)
7158 struct nfs4_client *clp = lo->lo_owner.so_client;
7159 struct nfs4_ol_stateid *retstp;
7161 mutex_init(&stp->st_mutex);
7162 mutex_lock_nested(&stp->st_mutex, OPEN_STATEID_MUTEX);
7164 spin_lock(&clp->cl_lock);
7165 if (nfs4_ol_stateid_unhashed(open_stp))
7167 retstp = find_lock_stateid(lo, open_stp);
7170 refcount_inc(&stp->st_stid.sc_count);
7171 stp->st_stid.sc_type = NFS4_LOCK_STID;
7172 stp->st_stateowner = nfs4_get_stateowner(&lo->lo_owner);
7174 stp->st_stid.sc_file = fp;
7175 stp->st_access_bmap = 0;
7176 stp->st_deny_bmap = open_stp->st_deny_bmap;
7177 stp->st_openstp = open_stp;
7178 spin_lock(&fp->fi_lock);
7179 list_add(&stp->st_locks, &open_stp->st_locks);
7180 list_add(&stp->st_perstateowner, &lo->lo_owner.so_stateids);
7181 list_add(&stp->st_perfile, &fp->fi_stateids);
7182 spin_unlock(&fp->fi_lock);
7183 spin_unlock(&clp->cl_lock);
7186 spin_unlock(&clp->cl_lock);
7187 if (nfsd4_lock_ol_stateid(retstp) != nfs_ok) {
7188 nfs4_put_stid(&retstp->st_stid);
7191 /* To keep mutex tracking happy */
7192 mutex_unlock(&stp->st_mutex);
7195 spin_unlock(&clp->cl_lock);
7196 mutex_unlock(&stp->st_mutex);
7200 static struct nfs4_ol_stateid *
7201 find_or_create_lock_stateid(struct nfs4_lockowner *lo, struct nfs4_file *fi,
7202 struct inode *inode, struct nfs4_ol_stateid *ost,
7205 struct nfs4_stid *ns = NULL;
7206 struct nfs4_ol_stateid *lst;
7207 struct nfs4_openowner *oo = openowner(ost->st_stateowner);
7208 struct nfs4_client *clp = oo->oo_owner.so_client;
7211 spin_lock(&clp->cl_lock);
7212 lst = find_lock_stateid(lo, ost);
7213 spin_unlock(&clp->cl_lock);
7215 if (nfsd4_lock_ol_stateid(lst) == nfs_ok)
7217 nfs4_put_stid(&lst->st_stid);
7219 ns = nfs4_alloc_stid(clp, stateid_slab, nfs4_free_lock_stateid);
7223 lst = init_lock_stateid(openlockstateid(ns), lo, fi, inode, ost);
7224 if (lst == openlockstateid(ns))
7233 check_lock_length(u64 offset, u64 length)
7235 return ((length == 0) || ((length != NFS4_MAX_UINT64) &&
7236 (length > ~offset)));
7239 static void get_lock_access(struct nfs4_ol_stateid *lock_stp, u32 access)
7241 struct nfs4_file *fp = lock_stp->st_stid.sc_file;
7243 lockdep_assert_held(&fp->fi_lock);
7245 if (test_access(access, lock_stp))
7247 __nfs4_file_get_access(fp, access);
7248 set_access(access, lock_stp);
7252 lookup_or_create_lock_state(struct nfsd4_compound_state *cstate,
7253 struct nfs4_ol_stateid *ost,
7254 struct nfsd4_lock *lock,
7255 struct nfs4_ol_stateid **plst, bool *new)
7258 struct nfs4_file *fi = ost->st_stid.sc_file;
7259 struct nfs4_openowner *oo = openowner(ost->st_stateowner);
7260 struct nfs4_client *cl = oo->oo_owner.so_client;
7261 struct inode *inode = d_inode(cstate->current_fh.fh_dentry);
7262 struct nfs4_lockowner *lo;
7263 struct nfs4_ol_stateid *lst;
7264 unsigned int strhashval;
7266 lo = find_lockowner_str(cl, &lock->lk_new_owner);
7268 strhashval = ownerstr_hashval(&lock->lk_new_owner);
7269 lo = alloc_init_lock_stateowner(strhashval, cl, ost, lock);
7271 return nfserr_jukebox;
7273 /* with an existing lockowner, seqids must be the same */
7274 status = nfserr_bad_seqid;
7275 if (!cstate->minorversion &&
7276 lock->lk_new_lock_seqid != lo->lo_owner.so_seqid)
7280 lst = find_or_create_lock_stateid(lo, fi, inode, ost, new);
7282 status = nfserr_jukebox;
7289 nfs4_put_stateowner(&lo->lo_owner);
7297 nfsd4_lock(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
7298 union nfsd4_op_u *u)
7300 struct nfsd4_lock *lock = &u->lock;
7301 struct nfs4_openowner *open_sop = NULL;
7302 struct nfs4_lockowner *lock_sop = NULL;
7303 struct nfs4_ol_stateid *lock_stp = NULL;
7304 struct nfs4_ol_stateid *open_stp = NULL;
7305 struct nfs4_file *fp;
7306 struct nfsd_file *nf = NULL;
7307 struct nfsd4_blocked_lock *nbl = NULL;
7308 struct file_lock *file_lock = NULL;
7309 struct file_lock *conflock = NULL;
7314 unsigned char fl_type;
7315 unsigned int fl_flags = FL_POSIX;
7316 struct net *net = SVC_NET(rqstp);
7317 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
7319 dprintk("NFSD: nfsd4_lock: start=%Ld length=%Ld\n",
7320 (long long) lock->lk_offset,
7321 (long long) lock->lk_length);
7323 if (check_lock_length(lock->lk_offset, lock->lk_length))
7324 return nfserr_inval;
7326 if ((status = fh_verify(rqstp, &cstate->current_fh,
7327 S_IFREG, NFSD_MAY_LOCK))) {
7328 dprintk("NFSD: nfsd4_lock: permission denied!\n");
7332 if (lock->lk_is_new) {
7333 if (nfsd4_has_session(cstate))
7334 /* See rfc 5661 18.10.3: given clientid is ignored: */
7335 memcpy(&lock->lk_new_clientid,
7336 &cstate->clp->cl_clientid,
7337 sizeof(clientid_t));
7339 /* validate and update open stateid and open seqid */
7340 status = nfs4_preprocess_confirmed_seqid_op(cstate,
7341 lock->lk_new_open_seqid,
7342 &lock->lk_new_open_stateid,
7346 mutex_unlock(&open_stp->st_mutex);
7347 open_sop = openowner(open_stp->st_stateowner);
7348 status = nfserr_bad_stateid;
7349 if (!same_clid(&open_sop->oo_owner.so_client->cl_clientid,
7350 &lock->lk_new_clientid))
7352 status = lookup_or_create_lock_state(cstate, open_stp, lock,
7355 status = nfs4_preprocess_seqid_op(cstate,
7356 lock->lk_old_lock_seqid,
7357 &lock->lk_old_lock_stateid,
7358 NFS4_LOCK_STID, &lock_stp, nn);
7362 lock_sop = lockowner(lock_stp->st_stateowner);
7364 lkflg = setlkflg(lock->lk_type);
7365 status = nfs4_check_openmode(lock_stp, lkflg);
7369 status = nfserr_grace;
7370 if (locks_in_grace(net) && !lock->lk_reclaim)
7372 status = nfserr_no_grace;
7373 if (!locks_in_grace(net) && lock->lk_reclaim)
7376 if (lock->lk_reclaim)
7377 fl_flags |= FL_RECLAIM;
7379 fp = lock_stp->st_stid.sc_file;
7380 switch (lock->lk_type) {
7382 if (nfsd4_has_session(cstate))
7383 fl_flags |= FL_SLEEP;
7386 spin_lock(&fp->fi_lock);
7387 nf = find_readable_file_locked(fp);
7389 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_READ);
7390 spin_unlock(&fp->fi_lock);
7393 case NFS4_WRITEW_LT:
7394 if (nfsd4_has_session(cstate))
7395 fl_flags |= FL_SLEEP;
7398 spin_lock(&fp->fi_lock);
7399 nf = find_writeable_file_locked(fp);
7401 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_WRITE);
7402 spin_unlock(&fp->fi_lock);
7406 status = nfserr_inval;
7411 status = nfserr_openmode;
7416 * Most filesystems with their own ->lock operations will block
7417 * the nfsd thread waiting to acquire the lock. That leads to
7418 * deadlocks (we don't want every nfsd thread tied up waiting
7419 * for file locks), so don't attempt blocking lock notifications
7420 * on those filesystems:
7422 if (nf->nf_file->f_op->lock)
7423 fl_flags &= ~FL_SLEEP;
7425 nbl = find_or_allocate_block(lock_sop, &fp->fi_fhandle, nn);
7427 dprintk("NFSD: %s: unable to allocate block!\n", __func__);
7428 status = nfserr_jukebox;
7432 file_lock = &nbl->nbl_lock;
7433 file_lock->fl_type = fl_type;
7434 file_lock->fl_owner = (fl_owner_t)lockowner(nfs4_get_stateowner(&lock_sop->lo_owner));
7435 file_lock->fl_pid = current->tgid;
7436 file_lock->fl_file = nf->nf_file;
7437 file_lock->fl_flags = fl_flags;
7438 file_lock->fl_lmops = &nfsd_posix_mng_ops;
7439 file_lock->fl_start = lock->lk_offset;
7440 file_lock->fl_end = last_byte_offset(lock->lk_offset, lock->lk_length);
7441 nfs4_transform_lock_offset(file_lock);
7443 conflock = locks_alloc_lock();
7445 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
7446 status = nfserr_jukebox;
7450 if (fl_flags & FL_SLEEP) {
7451 nbl->nbl_time = ktime_get_boottime_seconds();
7452 spin_lock(&nn->blocked_locks_lock);
7453 list_add_tail(&nbl->nbl_list, &lock_sop->lo_blocked);
7454 list_add_tail(&nbl->nbl_lru, &nn->blocked_locks_lru);
7455 kref_get(&nbl->nbl_kref);
7456 spin_unlock(&nn->blocked_locks_lock);
7459 err = vfs_lock_file(nf->nf_file, F_SETLK, file_lock, conflock);
7461 case 0: /* success! */
7462 nfs4_inc_and_copy_stateid(&lock->lk_resp_stateid, &lock_stp->st_stid);
7464 if (lock->lk_reclaim)
7465 nn->somebody_reclaimed = true;
7467 case FILE_LOCK_DEFERRED:
7468 kref_put(&nbl->nbl_kref, free_nbl);
7471 case -EAGAIN: /* conflock holds conflicting lock */
7472 status = nfserr_denied;
7473 dprintk("NFSD: nfsd4_lock: conflicting lock found!\n");
7474 nfs4_set_lock_denied(conflock, &lock->lk_denied);
7477 status = nfserr_deadlock;
7480 dprintk("NFSD: nfsd4_lock: vfs_lock_file() failed! status %d\n",err);
7481 status = nfserrno(err);
7486 /* dequeue it if we queued it before */
7487 if (fl_flags & FL_SLEEP) {
7488 spin_lock(&nn->blocked_locks_lock);
7489 if (!list_empty(&nbl->nbl_list) &&
7490 !list_empty(&nbl->nbl_lru)) {
7491 list_del_init(&nbl->nbl_list);
7492 list_del_init(&nbl->nbl_lru);
7493 kref_put(&nbl->nbl_kref, free_nbl);
7495 /* nbl can use one of lists to be linked to reaplist */
7496 spin_unlock(&nn->blocked_locks_lock);
7498 free_blocked_lock(nbl);
7503 /* Bump seqid manually if the 4.0 replay owner is openowner */
7504 if (cstate->replay_owner &&
7505 cstate->replay_owner != &lock_sop->lo_owner &&
7506 seqid_mutating_err(ntohl(status)))
7507 lock_sop->lo_owner.so_seqid++;
7510 * If this is a new, never-before-used stateid, and we are
7511 * returning an error, then just go ahead and release it.
7514 release_lock_stateid(lock_stp);
7516 mutex_unlock(&lock_stp->st_mutex);
7518 nfs4_put_stid(&lock_stp->st_stid);
7521 nfs4_put_stid(&open_stp->st_stid);
7522 nfsd4_bump_seqid(cstate, status);
7524 locks_free_lock(conflock);
7529 * The NFSv4 spec allows a client to do a LOCKT without holding an OPEN,
7530 * so we do a temporary open here just to get an open file to pass to
7533 static __be32 nfsd_test_lock(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file_lock *lock)
7535 struct nfsd_file *nf;
7536 struct inode *inode;
7539 err = nfsd_file_acquire(rqstp, fhp, NFSD_MAY_READ, &nf);
7542 inode = fhp->fh_dentry->d_inode;
7543 inode_lock(inode); /* to block new leases till after test_lock: */
7544 err = nfserrno(nfsd_open_break_lease(inode, NFSD_MAY_READ));
7547 lock->fl_file = nf->nf_file;
7548 err = nfserrno(vfs_test_lock(nf->nf_file, lock));
7549 lock->fl_file = NULL;
7551 inode_unlock(inode);
7560 nfsd4_lockt(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
7561 union nfsd4_op_u *u)
7563 struct nfsd4_lockt *lockt = &u->lockt;
7564 struct file_lock *file_lock = NULL;
7565 struct nfs4_lockowner *lo = NULL;
7567 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
7569 if (locks_in_grace(SVC_NET(rqstp)))
7570 return nfserr_grace;
7572 if (check_lock_length(lockt->lt_offset, lockt->lt_length))
7573 return nfserr_inval;
7575 if (!nfsd4_has_session(cstate)) {
7576 status = set_client(&lockt->lt_clientid, cstate, nn);
7581 if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
7584 file_lock = locks_alloc_lock();
7586 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
7587 status = nfserr_jukebox;
7591 switch (lockt->lt_type) {
7594 file_lock->fl_type = F_RDLCK;
7597 case NFS4_WRITEW_LT:
7598 file_lock->fl_type = F_WRLCK;
7601 dprintk("NFSD: nfs4_lockt: bad lock type!\n");
7602 status = nfserr_inval;
7606 lo = find_lockowner_str(cstate->clp, &lockt->lt_owner);
7608 file_lock->fl_owner = (fl_owner_t)lo;
7609 file_lock->fl_pid = current->tgid;
7610 file_lock->fl_flags = FL_POSIX;
7612 file_lock->fl_start = lockt->lt_offset;
7613 file_lock->fl_end = last_byte_offset(lockt->lt_offset, lockt->lt_length);
7615 nfs4_transform_lock_offset(file_lock);
7617 status = nfsd_test_lock(rqstp, &cstate->current_fh, file_lock);
7621 if (file_lock->fl_type != F_UNLCK) {
7622 status = nfserr_denied;
7623 nfs4_set_lock_denied(file_lock, &lockt->lt_denied);
7627 nfs4_put_stateowner(&lo->lo_owner);
7629 locks_free_lock(file_lock);
7634 nfsd4_locku(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
7635 union nfsd4_op_u *u)
7637 struct nfsd4_locku *locku = &u->locku;
7638 struct nfs4_ol_stateid *stp;
7639 struct nfsd_file *nf = NULL;
7640 struct file_lock *file_lock = NULL;
7643 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
7645 dprintk("NFSD: nfsd4_locku: start=%Ld length=%Ld\n",
7646 (long long) locku->lu_offset,
7647 (long long) locku->lu_length);
7649 if (check_lock_length(locku->lu_offset, locku->lu_length))
7650 return nfserr_inval;
7652 status = nfs4_preprocess_seqid_op(cstate, locku->lu_seqid,
7653 &locku->lu_stateid, NFS4_LOCK_STID,
7657 nf = find_any_file(stp->st_stid.sc_file);
7659 status = nfserr_lock_range;
7662 file_lock = locks_alloc_lock();
7664 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
7665 status = nfserr_jukebox;
7669 file_lock->fl_type = F_UNLCK;
7670 file_lock->fl_owner = (fl_owner_t)lockowner(nfs4_get_stateowner(stp->st_stateowner));
7671 file_lock->fl_pid = current->tgid;
7672 file_lock->fl_file = nf->nf_file;
7673 file_lock->fl_flags = FL_POSIX;
7674 file_lock->fl_lmops = &nfsd_posix_mng_ops;
7675 file_lock->fl_start = locku->lu_offset;
7677 file_lock->fl_end = last_byte_offset(locku->lu_offset,
7679 nfs4_transform_lock_offset(file_lock);
7681 err = vfs_lock_file(nf->nf_file, F_SETLK, file_lock, NULL);
7683 dprintk("NFSD: nfs4_locku: vfs_lock_file failed!\n");
7686 nfs4_inc_and_copy_stateid(&locku->lu_stateid, &stp->st_stid);
7690 mutex_unlock(&stp->st_mutex);
7691 nfs4_put_stid(&stp->st_stid);
7693 nfsd4_bump_seqid(cstate, status);
7695 locks_free_lock(file_lock);
7699 status = nfserrno(err);
7705 * true: locks held by lockowner
7706 * false: no locks held by lockowner
7709 check_for_locks(struct nfs4_file *fp, struct nfs4_lockowner *lowner)
7711 struct file_lock *fl;
7713 struct nfsd_file *nf = find_any_file(fp);
7714 struct inode *inode;
7715 struct file_lock_context *flctx;
7718 /* Any valid lock stateid should have some sort of access */
7723 inode = locks_inode(nf->nf_file);
7724 flctx = inode->i_flctx;
7726 if (flctx && !list_empty_careful(&flctx->flc_posix)) {
7727 spin_lock(&flctx->flc_lock);
7728 list_for_each_entry(fl, &flctx->flc_posix, fl_list) {
7729 if (fl->fl_owner == (fl_owner_t)lowner) {
7734 spin_unlock(&flctx->flc_lock);
7741 * nfsd4_release_lockowner - process NFSv4.0 RELEASE_LOCKOWNER operations
7742 * @rqstp: RPC transaction
7743 * @cstate: NFSv4 COMPOUND state
7744 * @u: RELEASE_LOCKOWNER arguments
7746 * The lockowner's so_count is bumped when a lock record is added
7747 * or when copying a conflicting lock. The latter case is brief,
7748 * but can lead to fleeting false positives when looking for
7752 * %nfs_ok: lockowner released or not found
7753 * %nfserr_locks_held: lockowner still in use
7754 * %nfserr_stale_clientid: clientid no longer active
7755 * %nfserr_expired: clientid not recognized
7758 nfsd4_release_lockowner(struct svc_rqst *rqstp,
7759 struct nfsd4_compound_state *cstate,
7760 union nfsd4_op_u *u)
7762 struct nfsd4_release_lockowner *rlockowner = &u->release_lockowner;
7763 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
7764 clientid_t *clid = &rlockowner->rl_clientid;
7765 struct nfs4_ol_stateid *stp;
7766 struct nfs4_lockowner *lo;
7767 struct nfs4_client *clp;
7768 LIST_HEAD(reaplist);
7771 dprintk("nfsd4_release_lockowner clientid: (%08x/%08x):\n",
7772 clid->cl_boot, clid->cl_id);
7774 status = set_client(clid, cstate, nn);
7779 spin_lock(&clp->cl_lock);
7780 lo = find_lockowner_str_locked(clp, &rlockowner->rl_owner);
7782 spin_unlock(&clp->cl_lock);
7785 if (atomic_read(&lo->lo_owner.so_count) != 2) {
7786 spin_unlock(&clp->cl_lock);
7787 nfs4_put_stateowner(&lo->lo_owner);
7788 return nfserr_locks_held;
7790 unhash_lockowner_locked(lo);
7791 while (!list_empty(&lo->lo_owner.so_stateids)) {
7792 stp = list_first_entry(&lo->lo_owner.so_stateids,
7793 struct nfs4_ol_stateid,
7795 WARN_ON(!unhash_lock_stateid(stp));
7796 put_ol_stateid_locked(stp, &reaplist);
7798 spin_unlock(&clp->cl_lock);
7800 free_ol_stateid_reaplist(&reaplist);
7801 remove_blocked_locks(lo);
7802 nfs4_put_stateowner(&lo->lo_owner);
7806 static inline struct nfs4_client_reclaim *
7809 return kmalloc(sizeof(struct nfs4_client_reclaim), GFP_KERNEL);
7813 nfs4_has_reclaimed_state(struct xdr_netobj name, struct nfsd_net *nn)
7815 struct nfs4_client_reclaim *crp;
7817 crp = nfsd4_find_reclaim_client(name, nn);
7818 return (crp && crp->cr_clp);
7822 * failure => all reset bets are off, nfserr_no_grace...
7824 * The caller is responsible for freeing name.data if NULL is returned (it
7825 * will be freed in nfs4_remove_reclaim_record in the normal case).
7827 struct nfs4_client_reclaim *
7828 nfs4_client_to_reclaim(struct xdr_netobj name, struct xdr_netobj princhash,
7829 struct nfsd_net *nn)
7831 unsigned int strhashval;
7832 struct nfs4_client_reclaim *crp;
7834 crp = alloc_reclaim();
7836 strhashval = clientstr_hashval(name);
7837 INIT_LIST_HEAD(&crp->cr_strhash);
7838 list_add(&crp->cr_strhash, &nn->reclaim_str_hashtbl[strhashval]);
7839 crp->cr_name.data = name.data;
7840 crp->cr_name.len = name.len;
7841 crp->cr_princhash.data = princhash.data;
7842 crp->cr_princhash.len = princhash.len;
7844 nn->reclaim_str_hashtbl_size++;
7850 nfs4_remove_reclaim_record(struct nfs4_client_reclaim *crp, struct nfsd_net *nn)
7852 list_del(&crp->cr_strhash);
7853 kfree(crp->cr_name.data);
7854 kfree(crp->cr_princhash.data);
7856 nn->reclaim_str_hashtbl_size--;
7860 nfs4_release_reclaim(struct nfsd_net *nn)
7862 struct nfs4_client_reclaim *crp = NULL;
7865 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
7866 while (!list_empty(&nn->reclaim_str_hashtbl[i])) {
7867 crp = list_entry(nn->reclaim_str_hashtbl[i].next,
7868 struct nfs4_client_reclaim, cr_strhash);
7869 nfs4_remove_reclaim_record(crp, nn);
7872 WARN_ON_ONCE(nn->reclaim_str_hashtbl_size);
7876 * called from OPEN, CLAIM_PREVIOUS with a new clientid. */
7877 struct nfs4_client_reclaim *
7878 nfsd4_find_reclaim_client(struct xdr_netobj name, struct nfsd_net *nn)
7880 unsigned int strhashval;
7881 struct nfs4_client_reclaim *crp = NULL;
7883 strhashval = clientstr_hashval(name);
7884 list_for_each_entry(crp, &nn->reclaim_str_hashtbl[strhashval], cr_strhash) {
7885 if (compare_blob(&crp->cr_name, &name) == 0) {
7893 nfs4_check_open_reclaim(struct nfs4_client *clp)
7895 if (test_bit(NFSD4_CLIENT_RECLAIM_COMPLETE, &clp->cl_flags))
7896 return nfserr_no_grace;
7898 if (nfsd4_client_record_check(clp))
7899 return nfserr_reclaim_bad;
7905 * Since the lifetime of a delegation isn't limited to that of an open, a
7906 * client may quite reasonably hang on to a delegation as long as it has
7907 * the inode cached. This becomes an obvious problem the first time a
7908 * client's inode cache approaches the size of the server's total memory.
7910 * For now we avoid this problem by imposing a hard limit on the number
7911 * of delegations, which varies according to the server's memory size.
7914 set_max_delegations(void)
7917 * Allow at most 4 delegations per megabyte of RAM. Quick
7918 * estimates suggest that in the worst case (where every delegation
7919 * is for a different inode), a delegation could take about 1.5K,
7920 * giving a worst case usage of about 6% of memory.
7922 max_delegations = nr_free_buffer_pages() >> (20 - 2 - PAGE_SHIFT);
7925 static int nfs4_state_create_net(struct net *net)
7927 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
7930 nn->conf_id_hashtbl = kmalloc_array(CLIENT_HASH_SIZE,
7931 sizeof(struct list_head),
7933 if (!nn->conf_id_hashtbl)
7935 nn->unconf_id_hashtbl = kmalloc_array(CLIENT_HASH_SIZE,
7936 sizeof(struct list_head),
7938 if (!nn->unconf_id_hashtbl)
7940 nn->sessionid_hashtbl = kmalloc_array(SESSION_HASH_SIZE,
7941 sizeof(struct list_head),
7943 if (!nn->sessionid_hashtbl)
7946 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
7947 INIT_LIST_HEAD(&nn->conf_id_hashtbl[i]);
7948 INIT_LIST_HEAD(&nn->unconf_id_hashtbl[i]);
7950 for (i = 0; i < SESSION_HASH_SIZE; i++)
7951 INIT_LIST_HEAD(&nn->sessionid_hashtbl[i]);
7952 nn->conf_name_tree = RB_ROOT;
7953 nn->unconf_name_tree = RB_ROOT;
7954 nn->boot_time = ktime_get_real_seconds();
7955 nn->grace_ended = false;
7956 nn->nfsd4_manager.block_opens = true;
7957 INIT_LIST_HEAD(&nn->nfsd4_manager.list);
7958 INIT_LIST_HEAD(&nn->client_lru);
7959 INIT_LIST_HEAD(&nn->close_lru);
7960 INIT_LIST_HEAD(&nn->del_recall_lru);
7961 spin_lock_init(&nn->client_lock);
7962 spin_lock_init(&nn->s2s_cp_lock);
7963 idr_init(&nn->s2s_cp_stateids);
7965 spin_lock_init(&nn->blocked_locks_lock);
7966 INIT_LIST_HEAD(&nn->blocked_locks_lru);
7968 INIT_DELAYED_WORK(&nn->laundromat_work, laundromat_main);
7969 INIT_DELAYED_WORK(&nn->nfsd_shrinker_work, courtesy_client_reaper);
7975 kfree(nn->unconf_id_hashtbl);
7977 kfree(nn->conf_id_hashtbl);
7983 nfs4_state_destroy_net(struct net *net)
7986 struct nfs4_client *clp = NULL;
7987 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
7989 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
7990 while (!list_empty(&nn->conf_id_hashtbl[i])) {
7991 clp = list_entry(nn->conf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
7992 destroy_client(clp);
7996 WARN_ON(!list_empty(&nn->blocked_locks_lru));
7998 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
7999 while (!list_empty(&nn->unconf_id_hashtbl[i])) {
8000 clp = list_entry(nn->unconf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
8001 destroy_client(clp);
8005 kfree(nn->sessionid_hashtbl);
8006 kfree(nn->unconf_id_hashtbl);
8007 kfree(nn->conf_id_hashtbl);
8012 nfs4_state_start_net(struct net *net)
8014 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
8017 ret = nfs4_state_create_net(net);
8020 locks_start_grace(net, &nn->nfsd4_manager);
8021 nfsd4_client_tracking_init(net);
8022 if (nn->track_reclaim_completes && nn->reclaim_str_hashtbl_size == 0)
8024 printk(KERN_INFO "NFSD: starting %lld-second grace period (net %x)\n",
8025 nn->nfsd4_grace, net->ns.inum);
8026 trace_nfsd_grace_start(nn);
8027 queue_delayed_work(laundry_wq, &nn->laundromat_work, nn->nfsd4_grace * HZ);
8031 printk(KERN_INFO "NFSD: no clients to reclaim, skipping NFSv4 grace period (net %x)\n",
8033 queue_delayed_work(laundry_wq, &nn->laundromat_work, nn->nfsd4_lease * HZ);
8034 nfsd4_end_grace(nn);
8038 /* initialization to perform when the nfsd service is started: */
8041 nfs4_state_start(void)
8045 ret = nfsd4_create_callback_queue();
8049 set_max_delegations();
8054 nfs4_state_shutdown_net(struct net *net)
8056 struct nfs4_delegation *dp = NULL;
8057 struct list_head *pos, *next, reaplist;
8058 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
8060 cancel_delayed_work_sync(&nn->laundromat_work);
8061 locks_end_grace(&nn->nfsd4_manager);
8063 INIT_LIST_HEAD(&reaplist);
8064 spin_lock(&state_lock);
8065 list_for_each_safe(pos, next, &nn->del_recall_lru) {
8066 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
8067 WARN_ON(!unhash_delegation_locked(dp));
8068 list_add(&dp->dl_recall_lru, &reaplist);
8070 spin_unlock(&state_lock);
8071 list_for_each_safe(pos, next, &reaplist) {
8072 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
8073 list_del_init(&dp->dl_recall_lru);
8074 destroy_unhashed_deleg(dp);
8077 nfsd4_client_tracking_exit(net);
8078 nfs4_state_destroy_net(net);
8079 #ifdef CONFIG_NFSD_V4_2_INTER_SSC
8080 nfsd4_ssc_shutdown_umount(nn);
8085 nfs4_state_shutdown(void)
8087 nfsd4_destroy_callback_queue();
8091 get_stateid(struct nfsd4_compound_state *cstate, stateid_t *stateid)
8093 if (HAS_CSTATE_FLAG(cstate, CURRENT_STATE_ID_FLAG) &&
8094 CURRENT_STATEID(stateid))
8095 memcpy(stateid, &cstate->current_stateid, sizeof(stateid_t));
8099 put_stateid(struct nfsd4_compound_state *cstate, stateid_t *stateid)
8101 if (cstate->minorversion) {
8102 memcpy(&cstate->current_stateid, stateid, sizeof(stateid_t));
8103 SET_CSTATE_FLAG(cstate, CURRENT_STATE_ID_FLAG);
8108 clear_current_stateid(struct nfsd4_compound_state *cstate)
8110 CLEAR_CSTATE_FLAG(cstate, CURRENT_STATE_ID_FLAG);
8114 * functions to set current state id
8117 nfsd4_set_opendowngradestateid(struct nfsd4_compound_state *cstate,
8118 union nfsd4_op_u *u)
8120 put_stateid(cstate, &u->open_downgrade.od_stateid);
8124 nfsd4_set_openstateid(struct nfsd4_compound_state *cstate,
8125 union nfsd4_op_u *u)
8127 put_stateid(cstate, &u->open.op_stateid);
8131 nfsd4_set_closestateid(struct nfsd4_compound_state *cstate,
8132 union nfsd4_op_u *u)
8134 put_stateid(cstate, &u->close.cl_stateid);
8138 nfsd4_set_lockstateid(struct nfsd4_compound_state *cstate,
8139 union nfsd4_op_u *u)
8141 put_stateid(cstate, &u->lock.lk_resp_stateid);
8145 * functions to consume current state id
8149 nfsd4_get_opendowngradestateid(struct nfsd4_compound_state *cstate,
8150 union nfsd4_op_u *u)
8152 get_stateid(cstate, &u->open_downgrade.od_stateid);
8156 nfsd4_get_delegreturnstateid(struct nfsd4_compound_state *cstate,
8157 union nfsd4_op_u *u)
8159 get_stateid(cstate, &u->delegreturn.dr_stateid);
8163 nfsd4_get_freestateid(struct nfsd4_compound_state *cstate,
8164 union nfsd4_op_u *u)
8166 get_stateid(cstate, &u->free_stateid.fr_stateid);
8170 nfsd4_get_setattrstateid(struct nfsd4_compound_state *cstate,
8171 union nfsd4_op_u *u)
8173 get_stateid(cstate, &u->setattr.sa_stateid);
8177 nfsd4_get_closestateid(struct nfsd4_compound_state *cstate,
8178 union nfsd4_op_u *u)
8180 get_stateid(cstate, &u->close.cl_stateid);
8184 nfsd4_get_lockustateid(struct nfsd4_compound_state *cstate,
8185 union nfsd4_op_u *u)
8187 get_stateid(cstate, &u->locku.lu_stateid);
8191 nfsd4_get_readstateid(struct nfsd4_compound_state *cstate,
8192 union nfsd4_op_u *u)
8194 get_stateid(cstate, &u->read.rd_stateid);
8198 nfsd4_get_writestateid(struct nfsd4_compound_state *cstate,
8199 union nfsd4_op_u *u)
8201 get_stateid(cstate, &u->write.wr_stateid);