ARM: shmobile: r8a7791 dtsi: Describe DMA for the serial ports
[linux-2.6/btrfs-unstable.git] / fs / nfsd / nfs4state.c
blob38f2d7abe3a707061c3f860ce8042649bf5109eb
1 /*
2 * Copyright (c) 2001 The Regents of the University of Michigan.
3 * All rights reserved.
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
10 * are met:
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>
36 #include <linux/fs.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 "xdr4.h"
46 #include "xdr4cb.h"
47 #include "vfs.h"
48 #include "current_stateid.h"
50 #include "netns.h"
51 #include "pnfs.h"
53 #define NFSDDBG_FACILITY NFSDDBG_PROC
55 #define all_ones {{~0,~0},~0}
56 static const stateid_t one_stateid = {
57 .si_generation = ~0,
58 .si_opaque = all_ones,
60 static const stateid_t zero_stateid = {
61 /* all fields zero */
63 static const stateid_t currentstateid = {
64 .si_generation = 1,
67 static u64 current_sessionid = 1;
69 #define ZERO_STATEID(stateid) (!memcmp((stateid), &zero_stateid, sizeof(stateid_t)))
70 #define ONE_STATEID(stateid) (!memcmp((stateid), &one_stateid, sizeof(stateid_t)))
71 #define CURRENT_STATEID(stateid) (!memcmp((stateid), &currentstateid, sizeof(stateid_t)))
73 /* forward declarations */
74 static bool check_for_locks(struct nfs4_file *fp, struct nfs4_lockowner *lowner);
75 static void nfs4_free_ol_stateid(struct nfs4_stid *stid);
77 /* Locking: */
80 * Currently used for the del_recall_lru and file hash table. In an
81 * effort to decrease the scope of the client_mutex, this spinlock may
82 * eventually cover more:
84 static DEFINE_SPINLOCK(state_lock);
87 * A waitqueue for all in-progress 4.0 CLOSE operations that are waiting for
88 * the refcount on the open stateid to drop.
90 static DECLARE_WAIT_QUEUE_HEAD(close_wq);
92 static struct kmem_cache *openowner_slab;
93 static struct kmem_cache *lockowner_slab;
94 static struct kmem_cache *file_slab;
95 static struct kmem_cache *stateid_slab;
96 static struct kmem_cache *deleg_slab;
98 static void free_session(struct nfsd4_session *);
100 static struct nfsd4_callback_ops nfsd4_cb_recall_ops;
102 static bool is_session_dead(struct nfsd4_session *ses)
104 return ses->se_flags & NFS4_SESSION_DEAD;
107 static __be32 mark_session_dead_locked(struct nfsd4_session *ses, int ref_held_by_me)
109 if (atomic_read(&ses->se_ref) > ref_held_by_me)
110 return nfserr_jukebox;
111 ses->se_flags |= NFS4_SESSION_DEAD;
112 return nfs_ok;
115 static bool is_client_expired(struct nfs4_client *clp)
117 return clp->cl_time == 0;
120 static __be32 get_client_locked(struct nfs4_client *clp)
122 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
124 lockdep_assert_held(&nn->client_lock);
126 if (is_client_expired(clp))
127 return nfserr_expired;
128 atomic_inc(&clp->cl_refcount);
129 return nfs_ok;
132 /* must be called under the client_lock */
133 static inline void
134 renew_client_locked(struct nfs4_client *clp)
136 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
138 if (is_client_expired(clp)) {
139 WARN_ON(1);
140 printk("%s: client (clientid %08x/%08x) already expired\n",
141 __func__,
142 clp->cl_clientid.cl_boot,
143 clp->cl_clientid.cl_id);
144 return;
147 dprintk("renewing client (clientid %08x/%08x)\n",
148 clp->cl_clientid.cl_boot,
149 clp->cl_clientid.cl_id);
150 list_move_tail(&clp->cl_lru, &nn->client_lru);
151 clp->cl_time = get_seconds();
154 static void put_client_renew_locked(struct nfs4_client *clp)
156 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
158 lockdep_assert_held(&nn->client_lock);
160 if (!atomic_dec_and_test(&clp->cl_refcount))
161 return;
162 if (!is_client_expired(clp))
163 renew_client_locked(clp);
166 static void put_client_renew(struct nfs4_client *clp)
168 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
170 if (!atomic_dec_and_lock(&clp->cl_refcount, &nn->client_lock))
171 return;
172 if (!is_client_expired(clp))
173 renew_client_locked(clp);
174 spin_unlock(&nn->client_lock);
177 static __be32 nfsd4_get_session_locked(struct nfsd4_session *ses)
179 __be32 status;
181 if (is_session_dead(ses))
182 return nfserr_badsession;
183 status = get_client_locked(ses->se_client);
184 if (status)
185 return status;
186 atomic_inc(&ses->se_ref);
187 return nfs_ok;
190 static void nfsd4_put_session_locked(struct nfsd4_session *ses)
192 struct nfs4_client *clp = ses->se_client;
193 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
195 lockdep_assert_held(&nn->client_lock);
197 if (atomic_dec_and_test(&ses->se_ref) && is_session_dead(ses))
198 free_session(ses);
199 put_client_renew_locked(clp);
202 static void nfsd4_put_session(struct nfsd4_session *ses)
204 struct nfs4_client *clp = ses->se_client;
205 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
207 spin_lock(&nn->client_lock);
208 nfsd4_put_session_locked(ses);
209 spin_unlock(&nn->client_lock);
212 static inline struct nfs4_stateowner *
213 nfs4_get_stateowner(struct nfs4_stateowner *sop)
215 atomic_inc(&sop->so_count);
216 return sop;
219 static int
220 same_owner_str(struct nfs4_stateowner *sop, struct xdr_netobj *owner)
222 return (sop->so_owner.len == owner->len) &&
223 0 == memcmp(sop->so_owner.data, owner->data, owner->len);
226 static struct nfs4_openowner *
227 find_openstateowner_str_locked(unsigned int hashval, struct nfsd4_open *open,
228 struct nfs4_client *clp)
230 struct nfs4_stateowner *so;
232 lockdep_assert_held(&clp->cl_lock);
234 list_for_each_entry(so, &clp->cl_ownerstr_hashtbl[hashval],
235 so_strhash) {
236 if (!so->so_is_open_owner)
237 continue;
238 if (same_owner_str(so, &open->op_owner))
239 return openowner(nfs4_get_stateowner(so));
241 return NULL;
244 static struct nfs4_openowner *
245 find_openstateowner_str(unsigned int hashval, struct nfsd4_open *open,
246 struct nfs4_client *clp)
248 struct nfs4_openowner *oo;
250 spin_lock(&clp->cl_lock);
251 oo = find_openstateowner_str_locked(hashval, open, clp);
252 spin_unlock(&clp->cl_lock);
253 return oo;
256 static inline u32
257 opaque_hashval(const void *ptr, int nbytes)
259 unsigned char *cptr = (unsigned char *) ptr;
261 u32 x = 0;
262 while (nbytes--) {
263 x *= 37;
264 x += *cptr++;
266 return x;
269 static void nfsd4_free_file_rcu(struct rcu_head *rcu)
271 struct nfs4_file *fp = container_of(rcu, struct nfs4_file, fi_rcu);
273 kmem_cache_free(file_slab, fp);
276 void
277 put_nfs4_file(struct nfs4_file *fi)
279 might_lock(&state_lock);
281 if (atomic_dec_and_lock(&fi->fi_ref, &state_lock)) {
282 hlist_del_rcu(&fi->fi_hash);
283 spin_unlock(&state_lock);
284 WARN_ON_ONCE(!list_empty(&fi->fi_delegations));
285 call_rcu(&fi->fi_rcu, nfsd4_free_file_rcu);
289 static struct file *
290 __nfs4_get_fd(struct nfs4_file *f, int oflag)
292 if (f->fi_fds[oflag])
293 return get_file(f->fi_fds[oflag]);
294 return NULL;
297 static struct file *
298 find_writeable_file_locked(struct nfs4_file *f)
300 struct file *ret;
302 lockdep_assert_held(&f->fi_lock);
304 ret = __nfs4_get_fd(f, O_WRONLY);
305 if (!ret)
306 ret = __nfs4_get_fd(f, O_RDWR);
307 return ret;
310 static struct file *
311 find_writeable_file(struct nfs4_file *f)
313 struct file *ret;
315 spin_lock(&f->fi_lock);
316 ret = find_writeable_file_locked(f);
317 spin_unlock(&f->fi_lock);
319 return ret;
322 static struct file *find_readable_file_locked(struct nfs4_file *f)
324 struct file *ret;
326 lockdep_assert_held(&f->fi_lock);
328 ret = __nfs4_get_fd(f, O_RDONLY);
329 if (!ret)
330 ret = __nfs4_get_fd(f, O_RDWR);
331 return ret;
334 static struct file *
335 find_readable_file(struct nfs4_file *f)
337 struct file *ret;
339 spin_lock(&f->fi_lock);
340 ret = find_readable_file_locked(f);
341 spin_unlock(&f->fi_lock);
343 return ret;
346 struct file *
347 find_any_file(struct nfs4_file *f)
349 struct file *ret;
351 spin_lock(&f->fi_lock);
352 ret = __nfs4_get_fd(f, O_RDWR);
353 if (!ret) {
354 ret = __nfs4_get_fd(f, O_WRONLY);
355 if (!ret)
356 ret = __nfs4_get_fd(f, O_RDONLY);
358 spin_unlock(&f->fi_lock);
359 return ret;
362 static atomic_long_t num_delegations;
363 unsigned long max_delegations;
366 * Open owner state (share locks)
369 /* hash tables for lock and open owners */
370 #define OWNER_HASH_BITS 8
371 #define OWNER_HASH_SIZE (1 << OWNER_HASH_BITS)
372 #define OWNER_HASH_MASK (OWNER_HASH_SIZE - 1)
374 static unsigned int ownerstr_hashval(struct xdr_netobj *ownername)
376 unsigned int ret;
378 ret = opaque_hashval(ownername->data, ownername->len);
379 return ret & OWNER_HASH_MASK;
382 /* hash table for nfs4_file */
383 #define FILE_HASH_BITS 8
384 #define FILE_HASH_SIZE (1 << FILE_HASH_BITS)
386 static unsigned int nfsd_fh_hashval(struct knfsd_fh *fh)
388 return jhash2(fh->fh_base.fh_pad, XDR_QUADLEN(fh->fh_size), 0);
391 static unsigned int file_hashval(struct knfsd_fh *fh)
393 return nfsd_fh_hashval(fh) & (FILE_HASH_SIZE - 1);
396 static struct hlist_head file_hashtbl[FILE_HASH_SIZE];
398 static void
399 __nfs4_file_get_access(struct nfs4_file *fp, u32 access)
401 lockdep_assert_held(&fp->fi_lock);
403 if (access & NFS4_SHARE_ACCESS_WRITE)
404 atomic_inc(&fp->fi_access[O_WRONLY]);
405 if (access & NFS4_SHARE_ACCESS_READ)
406 atomic_inc(&fp->fi_access[O_RDONLY]);
409 static __be32
410 nfs4_file_get_access(struct nfs4_file *fp, u32 access)
412 lockdep_assert_held(&fp->fi_lock);
414 /* Does this access mode make sense? */
415 if (access & ~NFS4_SHARE_ACCESS_BOTH)
416 return nfserr_inval;
418 /* Does it conflict with a deny mode already set? */
419 if ((access & fp->fi_share_deny) != 0)
420 return nfserr_share_denied;
422 __nfs4_file_get_access(fp, access);
423 return nfs_ok;
426 static __be32 nfs4_file_check_deny(struct nfs4_file *fp, u32 deny)
428 /* Common case is that there is no deny mode. */
429 if (deny) {
430 /* Does this deny mode make sense? */
431 if (deny & ~NFS4_SHARE_DENY_BOTH)
432 return nfserr_inval;
434 if ((deny & NFS4_SHARE_DENY_READ) &&
435 atomic_read(&fp->fi_access[O_RDONLY]))
436 return nfserr_share_denied;
438 if ((deny & NFS4_SHARE_DENY_WRITE) &&
439 atomic_read(&fp->fi_access[O_WRONLY]))
440 return nfserr_share_denied;
442 return nfs_ok;
445 static void __nfs4_file_put_access(struct nfs4_file *fp, int oflag)
447 might_lock(&fp->fi_lock);
449 if (atomic_dec_and_lock(&fp->fi_access[oflag], &fp->fi_lock)) {
450 struct file *f1 = NULL;
451 struct file *f2 = NULL;
453 swap(f1, fp->fi_fds[oflag]);
454 if (atomic_read(&fp->fi_access[1 - oflag]) == 0)
455 swap(f2, fp->fi_fds[O_RDWR]);
456 spin_unlock(&fp->fi_lock);
457 if (f1)
458 fput(f1);
459 if (f2)
460 fput(f2);
464 static void nfs4_file_put_access(struct nfs4_file *fp, u32 access)
466 WARN_ON_ONCE(access & ~NFS4_SHARE_ACCESS_BOTH);
468 if (access & NFS4_SHARE_ACCESS_WRITE)
469 __nfs4_file_put_access(fp, O_WRONLY);
470 if (access & NFS4_SHARE_ACCESS_READ)
471 __nfs4_file_put_access(fp, O_RDONLY);
474 struct nfs4_stid *nfs4_alloc_stid(struct nfs4_client *cl,
475 struct kmem_cache *slab)
477 struct nfs4_stid *stid;
478 int new_id;
480 stid = kmem_cache_zalloc(slab, GFP_KERNEL);
481 if (!stid)
482 return NULL;
484 idr_preload(GFP_KERNEL);
485 spin_lock(&cl->cl_lock);
486 new_id = idr_alloc_cyclic(&cl->cl_stateids, stid, 0, 0, GFP_NOWAIT);
487 spin_unlock(&cl->cl_lock);
488 idr_preload_end();
489 if (new_id < 0)
490 goto out_free;
491 stid->sc_client = cl;
492 stid->sc_stateid.si_opaque.so_id = new_id;
493 stid->sc_stateid.si_opaque.so_clid = cl->cl_clientid;
494 /* Will be incremented before return to client: */
495 atomic_set(&stid->sc_count, 1);
498 * It shouldn't be a problem to reuse an opaque stateid value.
499 * I don't think it is for 4.1. But with 4.0 I worry that, for
500 * example, a stray write retransmission could be accepted by
501 * the server when it should have been rejected. Therefore,
502 * adopt a trick from the sctp code to attempt to maximize the
503 * amount of time until an id is reused, by ensuring they always
504 * "increase" (mod INT_MAX):
506 return stid;
507 out_free:
508 kmem_cache_free(slab, stid);
509 return NULL;
512 static struct nfs4_ol_stateid * nfs4_alloc_open_stateid(struct nfs4_client *clp)
514 struct nfs4_stid *stid;
515 struct nfs4_ol_stateid *stp;
517 stid = nfs4_alloc_stid(clp, stateid_slab);
518 if (!stid)
519 return NULL;
521 stp = openlockstateid(stid);
522 stp->st_stid.sc_free = nfs4_free_ol_stateid;
523 return stp;
526 static void nfs4_free_deleg(struct nfs4_stid *stid)
528 kmem_cache_free(deleg_slab, stid);
529 atomic_long_dec(&num_delegations);
533 * When we recall a delegation, we should be careful not to hand it
534 * out again straight away.
535 * To ensure this we keep a pair of bloom filters ('new' and 'old')
536 * in which the filehandles of recalled delegations are "stored".
537 * If a filehandle appear in either filter, a delegation is blocked.
538 * When a delegation is recalled, the filehandle is stored in the "new"
539 * filter.
540 * Every 30 seconds we swap the filters and clear the "new" one,
541 * unless both are empty of course.
543 * Each filter is 256 bits. We hash the filehandle to 32bit and use the
544 * low 3 bytes as hash-table indices.
546 * 'blocked_delegations_lock', which is always taken in block_delegations(),
547 * is used to manage concurrent access. Testing does not need the lock
548 * except when swapping the two filters.
550 static DEFINE_SPINLOCK(blocked_delegations_lock);
551 static struct bloom_pair {
552 int entries, old_entries;
553 time_t swap_time;
554 int new; /* index into 'set' */
555 DECLARE_BITMAP(set[2], 256);
556 } blocked_delegations;
558 static int delegation_blocked(struct knfsd_fh *fh)
560 u32 hash;
561 struct bloom_pair *bd = &blocked_delegations;
563 if (bd->entries == 0)
564 return 0;
565 if (seconds_since_boot() - bd->swap_time > 30) {
566 spin_lock(&blocked_delegations_lock);
567 if (seconds_since_boot() - bd->swap_time > 30) {
568 bd->entries -= bd->old_entries;
569 bd->old_entries = bd->entries;
570 memset(bd->set[bd->new], 0,
571 sizeof(bd->set[0]));
572 bd->new = 1-bd->new;
573 bd->swap_time = seconds_since_boot();
575 spin_unlock(&blocked_delegations_lock);
577 hash = jhash(&fh->fh_base, fh->fh_size, 0);
578 if (test_bit(hash&255, bd->set[0]) &&
579 test_bit((hash>>8)&255, bd->set[0]) &&
580 test_bit((hash>>16)&255, bd->set[0]))
581 return 1;
583 if (test_bit(hash&255, bd->set[1]) &&
584 test_bit((hash>>8)&255, bd->set[1]) &&
585 test_bit((hash>>16)&255, bd->set[1]))
586 return 1;
588 return 0;
591 static void block_delegations(struct knfsd_fh *fh)
593 u32 hash;
594 struct bloom_pair *bd = &blocked_delegations;
596 hash = jhash(&fh->fh_base, fh->fh_size, 0);
598 spin_lock(&blocked_delegations_lock);
599 __set_bit(hash&255, bd->set[bd->new]);
600 __set_bit((hash>>8)&255, bd->set[bd->new]);
601 __set_bit((hash>>16)&255, bd->set[bd->new]);
602 if (bd->entries == 0)
603 bd->swap_time = seconds_since_boot();
604 bd->entries += 1;
605 spin_unlock(&blocked_delegations_lock);
608 static struct nfs4_delegation *
609 alloc_init_deleg(struct nfs4_client *clp, struct svc_fh *current_fh)
611 struct nfs4_delegation *dp;
612 long n;
614 dprintk("NFSD alloc_init_deleg\n");
615 n = atomic_long_inc_return(&num_delegations);
616 if (n < 0 || n > max_delegations)
617 goto out_dec;
618 if (delegation_blocked(&current_fh->fh_handle))
619 goto out_dec;
620 dp = delegstateid(nfs4_alloc_stid(clp, deleg_slab));
621 if (dp == NULL)
622 goto out_dec;
624 dp->dl_stid.sc_free = nfs4_free_deleg;
626 * delegation seqid's are never incremented. The 4.1 special
627 * meaning of seqid 0 isn't meaningful, really, but let's avoid
628 * 0 anyway just for consistency and use 1:
630 dp->dl_stid.sc_stateid.si_generation = 1;
631 INIT_LIST_HEAD(&dp->dl_perfile);
632 INIT_LIST_HEAD(&dp->dl_perclnt);
633 INIT_LIST_HEAD(&dp->dl_recall_lru);
634 dp->dl_type = NFS4_OPEN_DELEGATE_READ;
635 dp->dl_retries = 1;
636 nfsd4_init_cb(&dp->dl_recall, dp->dl_stid.sc_client,
637 &nfsd4_cb_recall_ops, NFSPROC4_CLNT_CB_RECALL);
638 return dp;
639 out_dec:
640 atomic_long_dec(&num_delegations);
641 return NULL;
644 void
645 nfs4_put_stid(struct nfs4_stid *s)
647 struct nfs4_file *fp = s->sc_file;
648 struct nfs4_client *clp = s->sc_client;
650 might_lock(&clp->cl_lock);
652 if (!atomic_dec_and_lock(&s->sc_count, &clp->cl_lock)) {
653 wake_up_all(&close_wq);
654 return;
656 idr_remove(&clp->cl_stateids, s->sc_stateid.si_opaque.so_id);
657 spin_unlock(&clp->cl_lock);
658 s->sc_free(s);
659 if (fp)
660 put_nfs4_file(fp);
663 static void nfs4_put_deleg_lease(struct nfs4_file *fp)
665 struct file *filp = NULL;
667 spin_lock(&fp->fi_lock);
668 if (fp->fi_deleg_file && --fp->fi_delegees == 0)
669 swap(filp, fp->fi_deleg_file);
670 spin_unlock(&fp->fi_lock);
672 if (filp) {
673 vfs_setlease(filp, F_UNLCK, NULL, (void **)&fp);
674 fput(filp);
678 void nfs4_unhash_stid(struct nfs4_stid *s)
680 s->sc_type = 0;
683 static void
684 hash_delegation_locked(struct nfs4_delegation *dp, struct nfs4_file *fp)
686 lockdep_assert_held(&state_lock);
687 lockdep_assert_held(&fp->fi_lock);
689 atomic_inc(&dp->dl_stid.sc_count);
690 dp->dl_stid.sc_type = NFS4_DELEG_STID;
691 list_add(&dp->dl_perfile, &fp->fi_delegations);
692 list_add(&dp->dl_perclnt, &dp->dl_stid.sc_client->cl_delegations);
695 static void
696 unhash_delegation_locked(struct nfs4_delegation *dp)
698 struct nfs4_file *fp = dp->dl_stid.sc_file;
700 lockdep_assert_held(&state_lock);
702 dp->dl_stid.sc_type = NFS4_CLOSED_DELEG_STID;
703 /* Ensure that deleg break won't try to requeue it */
704 ++dp->dl_time;
705 spin_lock(&fp->fi_lock);
706 list_del_init(&dp->dl_perclnt);
707 list_del_init(&dp->dl_recall_lru);
708 list_del_init(&dp->dl_perfile);
709 spin_unlock(&fp->fi_lock);
712 static void destroy_delegation(struct nfs4_delegation *dp)
714 spin_lock(&state_lock);
715 unhash_delegation_locked(dp);
716 spin_unlock(&state_lock);
717 nfs4_put_deleg_lease(dp->dl_stid.sc_file);
718 nfs4_put_stid(&dp->dl_stid);
721 static void revoke_delegation(struct nfs4_delegation *dp)
723 struct nfs4_client *clp = dp->dl_stid.sc_client;
725 WARN_ON(!list_empty(&dp->dl_recall_lru));
727 nfs4_put_deleg_lease(dp->dl_stid.sc_file);
729 if (clp->cl_minorversion == 0)
730 nfs4_put_stid(&dp->dl_stid);
731 else {
732 dp->dl_stid.sc_type = NFS4_REVOKED_DELEG_STID;
733 spin_lock(&clp->cl_lock);
734 list_add(&dp->dl_recall_lru, &clp->cl_revoked);
735 spin_unlock(&clp->cl_lock);
740 * SETCLIENTID state
743 static unsigned int clientid_hashval(u32 id)
745 return id & CLIENT_HASH_MASK;
748 static unsigned int clientstr_hashval(const char *name)
750 return opaque_hashval(name, 8) & CLIENT_HASH_MASK;
754 * We store the NONE, READ, WRITE, and BOTH bits separately in the
755 * st_{access,deny}_bmap field of the stateid, in order to track not
756 * only what share bits are currently in force, but also what
757 * combinations of share bits previous opens have used. This allows us
758 * to enforce the recommendation of rfc 3530 14.2.19 that the server
759 * return an error if the client attempt to downgrade to a combination
760 * of share bits not explicable by closing some of its previous opens.
762 * XXX: This enforcement is actually incomplete, since we don't keep
763 * track of access/deny bit combinations; so, e.g., we allow:
765 * OPEN allow read, deny write
766 * OPEN allow both, deny none
767 * DOWNGRADE allow read, deny none
769 * which we should reject.
771 static unsigned int
772 bmap_to_share_mode(unsigned long bmap) {
773 int i;
774 unsigned int access = 0;
776 for (i = 1; i < 4; i++) {
777 if (test_bit(i, &bmap))
778 access |= i;
780 return access;
783 /* set share access for a given stateid */
784 static inline void
785 set_access(u32 access, struct nfs4_ol_stateid *stp)
787 unsigned char mask = 1 << access;
789 WARN_ON_ONCE(access > NFS4_SHARE_ACCESS_BOTH);
790 stp->st_access_bmap |= mask;
793 /* clear share access for a given stateid */
794 static inline void
795 clear_access(u32 access, struct nfs4_ol_stateid *stp)
797 unsigned char mask = 1 << access;
799 WARN_ON_ONCE(access > NFS4_SHARE_ACCESS_BOTH);
800 stp->st_access_bmap &= ~mask;
803 /* test whether a given stateid has access */
804 static inline bool
805 test_access(u32 access, struct nfs4_ol_stateid *stp)
807 unsigned char mask = 1 << access;
809 return (bool)(stp->st_access_bmap & mask);
812 /* set share deny for a given stateid */
813 static inline void
814 set_deny(u32 deny, struct nfs4_ol_stateid *stp)
816 unsigned char mask = 1 << deny;
818 WARN_ON_ONCE(deny > NFS4_SHARE_DENY_BOTH);
819 stp->st_deny_bmap |= mask;
822 /* clear share deny for a given stateid */
823 static inline void
824 clear_deny(u32 deny, struct nfs4_ol_stateid *stp)
826 unsigned char mask = 1 << deny;
828 WARN_ON_ONCE(deny > NFS4_SHARE_DENY_BOTH);
829 stp->st_deny_bmap &= ~mask;
832 /* test whether a given stateid is denying specific access */
833 static inline bool
834 test_deny(u32 deny, struct nfs4_ol_stateid *stp)
836 unsigned char mask = 1 << deny;
838 return (bool)(stp->st_deny_bmap & mask);
841 static int nfs4_access_to_omode(u32 access)
843 switch (access & NFS4_SHARE_ACCESS_BOTH) {
844 case NFS4_SHARE_ACCESS_READ:
845 return O_RDONLY;
846 case NFS4_SHARE_ACCESS_WRITE:
847 return O_WRONLY;
848 case NFS4_SHARE_ACCESS_BOTH:
849 return O_RDWR;
851 WARN_ON_ONCE(1);
852 return O_RDONLY;
856 * A stateid that had a deny mode associated with it is being released
857 * or downgraded. Recalculate the deny mode on the file.
859 static void
860 recalculate_deny_mode(struct nfs4_file *fp)
862 struct nfs4_ol_stateid *stp;
864 spin_lock(&fp->fi_lock);
865 fp->fi_share_deny = 0;
866 list_for_each_entry(stp, &fp->fi_stateids, st_perfile)
867 fp->fi_share_deny |= bmap_to_share_mode(stp->st_deny_bmap);
868 spin_unlock(&fp->fi_lock);
871 static void
872 reset_union_bmap_deny(u32 deny, struct nfs4_ol_stateid *stp)
874 int i;
875 bool change = false;
877 for (i = 1; i < 4; i++) {
878 if ((i & deny) != i) {
879 change = true;
880 clear_deny(i, stp);
884 /* Recalculate per-file deny mode if there was a change */
885 if (change)
886 recalculate_deny_mode(stp->st_stid.sc_file);
889 /* release all access and file references for a given stateid */
890 static void
891 release_all_access(struct nfs4_ol_stateid *stp)
893 int i;
894 struct nfs4_file *fp = stp->st_stid.sc_file;
896 if (fp && stp->st_deny_bmap != 0)
897 recalculate_deny_mode(fp);
899 for (i = 1; i < 4; i++) {
900 if (test_access(i, stp))
901 nfs4_file_put_access(stp->st_stid.sc_file, i);
902 clear_access(i, stp);
906 static void nfs4_put_stateowner(struct nfs4_stateowner *sop)
908 struct nfs4_client *clp = sop->so_client;
910 might_lock(&clp->cl_lock);
912 if (!atomic_dec_and_lock(&sop->so_count, &clp->cl_lock))
913 return;
914 sop->so_ops->so_unhash(sop);
915 spin_unlock(&clp->cl_lock);
916 kfree(sop->so_owner.data);
917 sop->so_ops->so_free(sop);
920 static void unhash_ol_stateid(struct nfs4_ol_stateid *stp)
922 struct nfs4_file *fp = stp->st_stid.sc_file;
924 lockdep_assert_held(&stp->st_stateowner->so_client->cl_lock);
926 spin_lock(&fp->fi_lock);
927 list_del(&stp->st_perfile);
928 spin_unlock(&fp->fi_lock);
929 list_del(&stp->st_perstateowner);
932 static void nfs4_free_ol_stateid(struct nfs4_stid *stid)
934 struct nfs4_ol_stateid *stp = openlockstateid(stid);
936 release_all_access(stp);
937 if (stp->st_stateowner)
938 nfs4_put_stateowner(stp->st_stateowner);
939 kmem_cache_free(stateid_slab, stid);
942 static void nfs4_free_lock_stateid(struct nfs4_stid *stid)
944 struct nfs4_ol_stateid *stp = openlockstateid(stid);
945 struct nfs4_lockowner *lo = lockowner(stp->st_stateowner);
946 struct file *file;
948 file = find_any_file(stp->st_stid.sc_file);
949 if (file)
950 filp_close(file, (fl_owner_t)lo);
951 nfs4_free_ol_stateid(stid);
955 * Put the persistent reference to an already unhashed generic stateid, while
956 * holding the cl_lock. If it's the last reference, then put it onto the
957 * reaplist for later destruction.
959 static void put_ol_stateid_locked(struct nfs4_ol_stateid *stp,
960 struct list_head *reaplist)
962 struct nfs4_stid *s = &stp->st_stid;
963 struct nfs4_client *clp = s->sc_client;
965 lockdep_assert_held(&clp->cl_lock);
967 WARN_ON_ONCE(!list_empty(&stp->st_locks));
969 if (!atomic_dec_and_test(&s->sc_count)) {
970 wake_up_all(&close_wq);
971 return;
974 idr_remove(&clp->cl_stateids, s->sc_stateid.si_opaque.so_id);
975 list_add(&stp->st_locks, reaplist);
978 static void unhash_lock_stateid(struct nfs4_ol_stateid *stp)
980 struct nfs4_openowner *oo = openowner(stp->st_openstp->st_stateowner);
982 lockdep_assert_held(&oo->oo_owner.so_client->cl_lock);
984 list_del_init(&stp->st_locks);
985 unhash_ol_stateid(stp);
986 nfs4_unhash_stid(&stp->st_stid);
989 static void release_lock_stateid(struct nfs4_ol_stateid *stp)
991 struct nfs4_openowner *oo = openowner(stp->st_openstp->st_stateowner);
993 spin_lock(&oo->oo_owner.so_client->cl_lock);
994 unhash_lock_stateid(stp);
995 spin_unlock(&oo->oo_owner.so_client->cl_lock);
996 nfs4_put_stid(&stp->st_stid);
999 static void unhash_lockowner_locked(struct nfs4_lockowner *lo)
1001 struct nfs4_client *clp = lo->lo_owner.so_client;
1003 lockdep_assert_held(&clp->cl_lock);
1005 list_del_init(&lo->lo_owner.so_strhash);
1009 * Free a list of generic stateids that were collected earlier after being
1010 * fully unhashed.
1012 static void
1013 free_ol_stateid_reaplist(struct list_head *reaplist)
1015 struct nfs4_ol_stateid *stp;
1016 struct nfs4_file *fp;
1018 might_sleep();
1020 while (!list_empty(reaplist)) {
1021 stp = list_first_entry(reaplist, struct nfs4_ol_stateid,
1022 st_locks);
1023 list_del(&stp->st_locks);
1024 fp = stp->st_stid.sc_file;
1025 stp->st_stid.sc_free(&stp->st_stid);
1026 if (fp)
1027 put_nfs4_file(fp);
1031 static void release_lockowner(struct nfs4_lockowner *lo)
1033 struct nfs4_client *clp = lo->lo_owner.so_client;
1034 struct nfs4_ol_stateid *stp;
1035 struct list_head reaplist;
1037 INIT_LIST_HEAD(&reaplist);
1039 spin_lock(&clp->cl_lock);
1040 unhash_lockowner_locked(lo);
1041 while (!list_empty(&lo->lo_owner.so_stateids)) {
1042 stp = list_first_entry(&lo->lo_owner.so_stateids,
1043 struct nfs4_ol_stateid, st_perstateowner);
1044 unhash_lock_stateid(stp);
1045 put_ol_stateid_locked(stp, &reaplist);
1047 spin_unlock(&clp->cl_lock);
1048 free_ol_stateid_reaplist(&reaplist);
1049 nfs4_put_stateowner(&lo->lo_owner);
1052 static void release_open_stateid_locks(struct nfs4_ol_stateid *open_stp,
1053 struct list_head *reaplist)
1055 struct nfs4_ol_stateid *stp;
1057 while (!list_empty(&open_stp->st_locks)) {
1058 stp = list_entry(open_stp->st_locks.next,
1059 struct nfs4_ol_stateid, st_locks);
1060 unhash_lock_stateid(stp);
1061 put_ol_stateid_locked(stp, reaplist);
1065 static void unhash_open_stateid(struct nfs4_ol_stateid *stp,
1066 struct list_head *reaplist)
1068 lockdep_assert_held(&stp->st_stid.sc_client->cl_lock);
1070 unhash_ol_stateid(stp);
1071 release_open_stateid_locks(stp, reaplist);
1074 static void release_open_stateid(struct nfs4_ol_stateid *stp)
1076 LIST_HEAD(reaplist);
1078 spin_lock(&stp->st_stid.sc_client->cl_lock);
1079 unhash_open_stateid(stp, &reaplist);
1080 put_ol_stateid_locked(stp, &reaplist);
1081 spin_unlock(&stp->st_stid.sc_client->cl_lock);
1082 free_ol_stateid_reaplist(&reaplist);
1085 static void unhash_openowner_locked(struct nfs4_openowner *oo)
1087 struct nfs4_client *clp = oo->oo_owner.so_client;
1089 lockdep_assert_held(&clp->cl_lock);
1091 list_del_init(&oo->oo_owner.so_strhash);
1092 list_del_init(&oo->oo_perclient);
1095 static void release_last_closed_stateid(struct nfs4_openowner *oo)
1097 struct nfsd_net *nn = net_generic(oo->oo_owner.so_client->net,
1098 nfsd_net_id);
1099 struct nfs4_ol_stateid *s;
1101 spin_lock(&nn->client_lock);
1102 s = oo->oo_last_closed_stid;
1103 if (s) {
1104 list_del_init(&oo->oo_close_lru);
1105 oo->oo_last_closed_stid = NULL;
1107 spin_unlock(&nn->client_lock);
1108 if (s)
1109 nfs4_put_stid(&s->st_stid);
1112 static void release_openowner(struct nfs4_openowner *oo)
1114 struct nfs4_ol_stateid *stp;
1115 struct nfs4_client *clp = oo->oo_owner.so_client;
1116 struct list_head reaplist;
1118 INIT_LIST_HEAD(&reaplist);
1120 spin_lock(&clp->cl_lock);
1121 unhash_openowner_locked(oo);
1122 while (!list_empty(&oo->oo_owner.so_stateids)) {
1123 stp = list_first_entry(&oo->oo_owner.so_stateids,
1124 struct nfs4_ol_stateid, st_perstateowner);
1125 unhash_open_stateid(stp, &reaplist);
1126 put_ol_stateid_locked(stp, &reaplist);
1128 spin_unlock(&clp->cl_lock);
1129 free_ol_stateid_reaplist(&reaplist);
1130 release_last_closed_stateid(oo);
1131 nfs4_put_stateowner(&oo->oo_owner);
1134 static inline int
1135 hash_sessionid(struct nfs4_sessionid *sessionid)
1137 struct nfsd4_sessionid *sid = (struct nfsd4_sessionid *)sessionid;
1139 return sid->sequence % SESSION_HASH_SIZE;
1142 #ifdef CONFIG_SUNRPC_DEBUG
1143 static inline void
1144 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
1146 u32 *ptr = (u32 *)(&sessionid->data[0]);
1147 dprintk("%s: %u:%u:%u:%u\n", fn, ptr[0], ptr[1], ptr[2], ptr[3]);
1149 #else
1150 static inline void
1151 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
1154 #endif
1157 * Bump the seqid on cstate->replay_owner, and clear replay_owner if it
1158 * won't be used for replay.
1160 void nfsd4_bump_seqid(struct nfsd4_compound_state *cstate, __be32 nfserr)
1162 struct nfs4_stateowner *so = cstate->replay_owner;
1164 if (nfserr == nfserr_replay_me)
1165 return;
1167 if (!seqid_mutating_err(ntohl(nfserr))) {
1168 nfsd4_cstate_clear_replay(cstate);
1169 return;
1171 if (!so)
1172 return;
1173 if (so->so_is_open_owner)
1174 release_last_closed_stateid(openowner(so));
1175 so->so_seqid++;
1176 return;
1179 static void
1180 gen_sessionid(struct nfsd4_session *ses)
1182 struct nfs4_client *clp = ses->se_client;
1183 struct nfsd4_sessionid *sid;
1185 sid = (struct nfsd4_sessionid *)ses->se_sessionid.data;
1186 sid->clientid = clp->cl_clientid;
1187 sid->sequence = current_sessionid++;
1188 sid->reserved = 0;
1192 * The protocol defines ca_maxresponssize_cached to include the size of
1193 * the rpc header, but all we need to cache is the data starting after
1194 * the end of the initial SEQUENCE operation--the rest we regenerate
1195 * each time. Therefore we can advertise a ca_maxresponssize_cached
1196 * value that is the number of bytes in our cache plus a few additional
1197 * bytes. In order to stay on the safe side, and not promise more than
1198 * we can cache, those additional bytes must be the minimum possible: 24
1199 * bytes of rpc header (xid through accept state, with AUTH_NULL
1200 * verifier), 12 for the compound header (with zero-length tag), and 44
1201 * for the SEQUENCE op response:
1203 #define NFSD_MIN_HDR_SEQ_SZ (24 + 12 + 44)
1205 static void
1206 free_session_slots(struct nfsd4_session *ses)
1208 int i;
1210 for (i = 0; i < ses->se_fchannel.maxreqs; i++)
1211 kfree(ses->se_slots[i]);
1215 * We don't actually need to cache the rpc and session headers, so we
1216 * can allocate a little less for each slot:
1218 static inline u32 slot_bytes(struct nfsd4_channel_attrs *ca)
1220 u32 size;
1222 if (ca->maxresp_cached < NFSD_MIN_HDR_SEQ_SZ)
1223 size = 0;
1224 else
1225 size = ca->maxresp_cached - NFSD_MIN_HDR_SEQ_SZ;
1226 return size + sizeof(struct nfsd4_slot);
1230 * XXX: If we run out of reserved DRC memory we could (up to a point)
1231 * re-negotiate active sessions and reduce their slot usage to make
1232 * room for new connections. For now we just fail the create session.
1234 static u32 nfsd4_get_drc_mem(struct nfsd4_channel_attrs *ca)
1236 u32 slotsize = slot_bytes(ca);
1237 u32 num = ca->maxreqs;
1238 int avail;
1240 spin_lock(&nfsd_drc_lock);
1241 avail = min((unsigned long)NFSD_MAX_MEM_PER_SESSION,
1242 nfsd_drc_max_mem - nfsd_drc_mem_used);
1243 num = min_t(int, num, avail / slotsize);
1244 nfsd_drc_mem_used += num * slotsize;
1245 spin_unlock(&nfsd_drc_lock);
1247 return num;
1250 static void nfsd4_put_drc_mem(struct nfsd4_channel_attrs *ca)
1252 int slotsize = slot_bytes(ca);
1254 spin_lock(&nfsd_drc_lock);
1255 nfsd_drc_mem_used -= slotsize * ca->maxreqs;
1256 spin_unlock(&nfsd_drc_lock);
1259 static struct nfsd4_session *alloc_session(struct nfsd4_channel_attrs *fattrs,
1260 struct nfsd4_channel_attrs *battrs)
1262 int numslots = fattrs->maxreqs;
1263 int slotsize = slot_bytes(fattrs);
1264 struct nfsd4_session *new;
1265 int mem, i;
1267 BUILD_BUG_ON(NFSD_MAX_SLOTS_PER_SESSION * sizeof(struct nfsd4_slot *)
1268 + sizeof(struct nfsd4_session) > PAGE_SIZE);
1269 mem = numslots * sizeof(struct nfsd4_slot *);
1271 new = kzalloc(sizeof(*new) + mem, GFP_KERNEL);
1272 if (!new)
1273 return NULL;
1274 /* allocate each struct nfsd4_slot and data cache in one piece */
1275 for (i = 0; i < numslots; i++) {
1276 new->se_slots[i] = kzalloc(slotsize, GFP_KERNEL);
1277 if (!new->se_slots[i])
1278 goto out_free;
1281 memcpy(&new->se_fchannel, fattrs, sizeof(struct nfsd4_channel_attrs));
1282 memcpy(&new->se_bchannel, battrs, sizeof(struct nfsd4_channel_attrs));
1284 return new;
1285 out_free:
1286 while (i--)
1287 kfree(new->se_slots[i]);
1288 kfree(new);
1289 return NULL;
1292 static void free_conn(struct nfsd4_conn *c)
1294 svc_xprt_put(c->cn_xprt);
1295 kfree(c);
1298 static void nfsd4_conn_lost(struct svc_xpt_user *u)
1300 struct nfsd4_conn *c = container_of(u, struct nfsd4_conn, cn_xpt_user);
1301 struct nfs4_client *clp = c->cn_session->se_client;
1303 spin_lock(&clp->cl_lock);
1304 if (!list_empty(&c->cn_persession)) {
1305 list_del(&c->cn_persession);
1306 free_conn(c);
1308 nfsd4_probe_callback(clp);
1309 spin_unlock(&clp->cl_lock);
1312 static struct nfsd4_conn *alloc_conn(struct svc_rqst *rqstp, u32 flags)
1314 struct nfsd4_conn *conn;
1316 conn = kmalloc(sizeof(struct nfsd4_conn), GFP_KERNEL);
1317 if (!conn)
1318 return NULL;
1319 svc_xprt_get(rqstp->rq_xprt);
1320 conn->cn_xprt = rqstp->rq_xprt;
1321 conn->cn_flags = flags;
1322 INIT_LIST_HEAD(&conn->cn_xpt_user.list);
1323 return conn;
1326 static void __nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
1328 conn->cn_session = ses;
1329 list_add(&conn->cn_persession, &ses->se_conns);
1332 static void nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
1334 struct nfs4_client *clp = ses->se_client;
1336 spin_lock(&clp->cl_lock);
1337 __nfsd4_hash_conn(conn, ses);
1338 spin_unlock(&clp->cl_lock);
1341 static int nfsd4_register_conn(struct nfsd4_conn *conn)
1343 conn->cn_xpt_user.callback = nfsd4_conn_lost;
1344 return register_xpt_user(conn->cn_xprt, &conn->cn_xpt_user);
1347 static void nfsd4_init_conn(struct svc_rqst *rqstp, struct nfsd4_conn *conn, struct nfsd4_session *ses)
1349 int ret;
1351 nfsd4_hash_conn(conn, ses);
1352 ret = nfsd4_register_conn(conn);
1353 if (ret)
1354 /* oops; xprt is already down: */
1355 nfsd4_conn_lost(&conn->cn_xpt_user);
1356 /* We may have gained or lost a callback channel: */
1357 nfsd4_probe_callback_sync(ses->se_client);
1360 static struct nfsd4_conn *alloc_conn_from_crses(struct svc_rqst *rqstp, struct nfsd4_create_session *cses)
1362 u32 dir = NFS4_CDFC4_FORE;
1364 if (cses->flags & SESSION4_BACK_CHAN)
1365 dir |= NFS4_CDFC4_BACK;
1366 return alloc_conn(rqstp, dir);
1369 /* must be called under client_lock */
1370 static void nfsd4_del_conns(struct nfsd4_session *s)
1372 struct nfs4_client *clp = s->se_client;
1373 struct nfsd4_conn *c;
1375 spin_lock(&clp->cl_lock);
1376 while (!list_empty(&s->se_conns)) {
1377 c = list_first_entry(&s->se_conns, struct nfsd4_conn, cn_persession);
1378 list_del_init(&c->cn_persession);
1379 spin_unlock(&clp->cl_lock);
1381 unregister_xpt_user(c->cn_xprt, &c->cn_xpt_user);
1382 free_conn(c);
1384 spin_lock(&clp->cl_lock);
1386 spin_unlock(&clp->cl_lock);
1389 static void __free_session(struct nfsd4_session *ses)
1391 free_session_slots(ses);
1392 kfree(ses);
1395 static void free_session(struct nfsd4_session *ses)
1397 nfsd4_del_conns(ses);
1398 nfsd4_put_drc_mem(&ses->se_fchannel);
1399 __free_session(ses);
1402 static void init_session(struct svc_rqst *rqstp, struct nfsd4_session *new, struct nfs4_client *clp, struct nfsd4_create_session *cses)
1404 int idx;
1405 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
1407 new->se_client = clp;
1408 gen_sessionid(new);
1410 INIT_LIST_HEAD(&new->se_conns);
1412 new->se_cb_seq_nr = 1;
1413 new->se_flags = cses->flags;
1414 new->se_cb_prog = cses->callback_prog;
1415 new->se_cb_sec = cses->cb_sec;
1416 atomic_set(&new->se_ref, 0);
1417 idx = hash_sessionid(&new->se_sessionid);
1418 list_add(&new->se_hash, &nn->sessionid_hashtbl[idx]);
1419 spin_lock(&clp->cl_lock);
1420 list_add(&new->se_perclnt, &clp->cl_sessions);
1421 spin_unlock(&clp->cl_lock);
1424 struct sockaddr *sa = svc_addr(rqstp);
1426 * This is a little silly; with sessions there's no real
1427 * use for the callback address. Use the peer address
1428 * as a reasonable default for now, but consider fixing
1429 * the rpc client not to require an address in the
1430 * future:
1432 rpc_copy_addr((struct sockaddr *)&clp->cl_cb_conn.cb_addr, sa);
1433 clp->cl_cb_conn.cb_addrlen = svc_addr_len(sa);
1437 /* caller must hold client_lock */
1438 static struct nfsd4_session *
1439 __find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid, struct net *net)
1441 struct nfsd4_session *elem;
1442 int idx;
1443 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
1445 lockdep_assert_held(&nn->client_lock);
1447 dump_sessionid(__func__, sessionid);
1448 idx = hash_sessionid(sessionid);
1449 /* Search in the appropriate list */
1450 list_for_each_entry(elem, &nn->sessionid_hashtbl[idx], se_hash) {
1451 if (!memcmp(elem->se_sessionid.data, sessionid->data,
1452 NFS4_MAX_SESSIONID_LEN)) {
1453 return elem;
1457 dprintk("%s: session not found\n", __func__);
1458 return NULL;
1461 static struct nfsd4_session *
1462 find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid, struct net *net,
1463 __be32 *ret)
1465 struct nfsd4_session *session;
1466 __be32 status = nfserr_badsession;
1468 session = __find_in_sessionid_hashtbl(sessionid, net);
1469 if (!session)
1470 goto out;
1471 status = nfsd4_get_session_locked(session);
1472 if (status)
1473 session = NULL;
1474 out:
1475 *ret = status;
1476 return session;
1479 /* caller must hold client_lock */
1480 static void
1481 unhash_session(struct nfsd4_session *ses)
1483 struct nfs4_client *clp = ses->se_client;
1484 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1486 lockdep_assert_held(&nn->client_lock);
1488 list_del(&ses->se_hash);
1489 spin_lock(&ses->se_client->cl_lock);
1490 list_del(&ses->se_perclnt);
1491 spin_unlock(&ses->se_client->cl_lock);
1494 /* SETCLIENTID and SETCLIENTID_CONFIRM Helper functions */
1495 static int
1496 STALE_CLIENTID(clientid_t *clid, struct nfsd_net *nn)
1499 * We're assuming the clid was not given out from a boot
1500 * precisely 2^32 (about 136 years) before this one. That seems
1501 * a safe assumption:
1503 if (clid->cl_boot == (u32)nn->boot_time)
1504 return 0;
1505 dprintk("NFSD stale clientid (%08x/%08x) boot_time %08lx\n",
1506 clid->cl_boot, clid->cl_id, nn->boot_time);
1507 return 1;
1511 * XXX Should we use a slab cache ?
1512 * This type of memory management is somewhat inefficient, but we use it
1513 * anyway since SETCLIENTID is not a common operation.
1515 static struct nfs4_client *alloc_client(struct xdr_netobj name)
1517 struct nfs4_client *clp;
1518 int i;
1520 clp = kzalloc(sizeof(struct nfs4_client), GFP_KERNEL);
1521 if (clp == NULL)
1522 return NULL;
1523 clp->cl_name.data = kmemdup(name.data, name.len, GFP_KERNEL);
1524 if (clp->cl_name.data == NULL)
1525 goto err_no_name;
1526 clp->cl_ownerstr_hashtbl = kmalloc(sizeof(struct list_head) *
1527 OWNER_HASH_SIZE, GFP_KERNEL);
1528 if (!clp->cl_ownerstr_hashtbl)
1529 goto err_no_hashtbl;
1530 for (i = 0; i < OWNER_HASH_SIZE; i++)
1531 INIT_LIST_HEAD(&clp->cl_ownerstr_hashtbl[i]);
1532 clp->cl_name.len = name.len;
1533 INIT_LIST_HEAD(&clp->cl_sessions);
1534 idr_init(&clp->cl_stateids);
1535 atomic_set(&clp->cl_refcount, 0);
1536 clp->cl_cb_state = NFSD4_CB_UNKNOWN;
1537 INIT_LIST_HEAD(&clp->cl_idhash);
1538 INIT_LIST_HEAD(&clp->cl_openowners);
1539 INIT_LIST_HEAD(&clp->cl_delegations);
1540 INIT_LIST_HEAD(&clp->cl_lru);
1541 INIT_LIST_HEAD(&clp->cl_callbacks);
1542 INIT_LIST_HEAD(&clp->cl_revoked);
1543 #ifdef CONFIG_NFSD_PNFS
1544 INIT_LIST_HEAD(&clp->cl_lo_states);
1545 #endif
1546 spin_lock_init(&clp->cl_lock);
1547 rpc_init_wait_queue(&clp->cl_cb_waitq, "Backchannel slot table");
1548 return clp;
1549 err_no_hashtbl:
1550 kfree(clp->cl_name.data);
1551 err_no_name:
1552 kfree(clp);
1553 return NULL;
1556 static void
1557 free_client(struct nfs4_client *clp)
1559 while (!list_empty(&clp->cl_sessions)) {
1560 struct nfsd4_session *ses;
1561 ses = list_entry(clp->cl_sessions.next, struct nfsd4_session,
1562 se_perclnt);
1563 list_del(&ses->se_perclnt);
1564 WARN_ON_ONCE(atomic_read(&ses->se_ref));
1565 free_session(ses);
1567 rpc_destroy_wait_queue(&clp->cl_cb_waitq);
1568 free_svc_cred(&clp->cl_cred);
1569 kfree(clp->cl_ownerstr_hashtbl);
1570 kfree(clp->cl_name.data);
1571 idr_destroy(&clp->cl_stateids);
1572 kfree(clp);
1575 /* must be called under the client_lock */
1576 static void
1577 unhash_client_locked(struct nfs4_client *clp)
1579 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1580 struct nfsd4_session *ses;
1582 lockdep_assert_held(&nn->client_lock);
1584 /* Mark the client as expired! */
1585 clp->cl_time = 0;
1586 /* Make it invisible */
1587 if (!list_empty(&clp->cl_idhash)) {
1588 list_del_init(&clp->cl_idhash);
1589 if (test_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags))
1590 rb_erase(&clp->cl_namenode, &nn->conf_name_tree);
1591 else
1592 rb_erase(&clp->cl_namenode, &nn->unconf_name_tree);
1594 list_del_init(&clp->cl_lru);
1595 spin_lock(&clp->cl_lock);
1596 list_for_each_entry(ses, &clp->cl_sessions, se_perclnt)
1597 list_del_init(&ses->se_hash);
1598 spin_unlock(&clp->cl_lock);
1601 static void
1602 unhash_client(struct nfs4_client *clp)
1604 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1606 spin_lock(&nn->client_lock);
1607 unhash_client_locked(clp);
1608 spin_unlock(&nn->client_lock);
1611 static __be32 mark_client_expired_locked(struct nfs4_client *clp)
1613 if (atomic_read(&clp->cl_refcount))
1614 return nfserr_jukebox;
1615 unhash_client_locked(clp);
1616 return nfs_ok;
1619 static void
1620 __destroy_client(struct nfs4_client *clp)
1622 struct nfs4_openowner *oo;
1623 struct nfs4_delegation *dp;
1624 struct list_head reaplist;
1626 INIT_LIST_HEAD(&reaplist);
1627 spin_lock(&state_lock);
1628 while (!list_empty(&clp->cl_delegations)) {
1629 dp = list_entry(clp->cl_delegations.next, struct nfs4_delegation, dl_perclnt);
1630 unhash_delegation_locked(dp);
1631 list_add(&dp->dl_recall_lru, &reaplist);
1633 spin_unlock(&state_lock);
1634 while (!list_empty(&reaplist)) {
1635 dp = list_entry(reaplist.next, struct nfs4_delegation, dl_recall_lru);
1636 list_del_init(&dp->dl_recall_lru);
1637 nfs4_put_deleg_lease(dp->dl_stid.sc_file);
1638 nfs4_put_stid(&dp->dl_stid);
1640 while (!list_empty(&clp->cl_revoked)) {
1641 dp = list_entry(clp->cl_revoked.next, struct nfs4_delegation, dl_recall_lru);
1642 list_del_init(&dp->dl_recall_lru);
1643 nfs4_put_stid(&dp->dl_stid);
1645 while (!list_empty(&clp->cl_openowners)) {
1646 oo = list_entry(clp->cl_openowners.next, struct nfs4_openowner, oo_perclient);
1647 nfs4_get_stateowner(&oo->oo_owner);
1648 release_openowner(oo);
1650 nfsd4_return_all_client_layouts(clp);
1651 nfsd4_shutdown_callback(clp);
1652 if (clp->cl_cb_conn.cb_xprt)
1653 svc_xprt_put(clp->cl_cb_conn.cb_xprt);
1654 free_client(clp);
1657 static void
1658 destroy_client(struct nfs4_client *clp)
1660 unhash_client(clp);
1661 __destroy_client(clp);
1664 static void expire_client(struct nfs4_client *clp)
1666 unhash_client(clp);
1667 nfsd4_client_record_remove(clp);
1668 __destroy_client(clp);
1671 static void copy_verf(struct nfs4_client *target, nfs4_verifier *source)
1673 memcpy(target->cl_verifier.data, source->data,
1674 sizeof(target->cl_verifier.data));
1677 static void copy_clid(struct nfs4_client *target, struct nfs4_client *source)
1679 target->cl_clientid.cl_boot = source->cl_clientid.cl_boot;
1680 target->cl_clientid.cl_id = source->cl_clientid.cl_id;
1683 static int copy_cred(struct svc_cred *target, struct svc_cred *source)
1685 if (source->cr_principal) {
1686 target->cr_principal =
1687 kstrdup(source->cr_principal, GFP_KERNEL);
1688 if (target->cr_principal == NULL)
1689 return -ENOMEM;
1690 } else
1691 target->cr_principal = NULL;
1692 target->cr_flavor = source->cr_flavor;
1693 target->cr_uid = source->cr_uid;
1694 target->cr_gid = source->cr_gid;
1695 target->cr_group_info = source->cr_group_info;
1696 get_group_info(target->cr_group_info);
1697 target->cr_gss_mech = source->cr_gss_mech;
1698 if (source->cr_gss_mech)
1699 gss_mech_get(source->cr_gss_mech);
1700 return 0;
1703 static int
1704 compare_blob(const struct xdr_netobj *o1, const struct xdr_netobj *o2)
1706 if (o1->len < o2->len)
1707 return -1;
1708 if (o1->len > o2->len)
1709 return 1;
1710 return memcmp(o1->data, o2->data, o1->len);
1713 static int same_name(const char *n1, const char *n2)
1715 return 0 == memcmp(n1, n2, HEXDIR_LEN);
1718 static int
1719 same_verf(nfs4_verifier *v1, nfs4_verifier *v2)
1721 return 0 == memcmp(v1->data, v2->data, sizeof(v1->data));
1724 static int
1725 same_clid(clientid_t *cl1, clientid_t *cl2)
1727 return (cl1->cl_boot == cl2->cl_boot) && (cl1->cl_id == cl2->cl_id);
1730 static bool groups_equal(struct group_info *g1, struct group_info *g2)
1732 int i;
1734 if (g1->ngroups != g2->ngroups)
1735 return false;
1736 for (i=0; i<g1->ngroups; i++)
1737 if (!gid_eq(GROUP_AT(g1, i), GROUP_AT(g2, i)))
1738 return false;
1739 return true;
1743 * RFC 3530 language requires clid_inuse be returned when the
1744 * "principal" associated with a requests differs from that previously
1745 * used. We use uid, gid's, and gss principal string as our best
1746 * approximation. We also don't want to allow non-gss use of a client
1747 * established using gss: in theory cr_principal should catch that
1748 * change, but in practice cr_principal can be null even in the gss case
1749 * since gssd doesn't always pass down a principal string.
1751 static bool is_gss_cred(struct svc_cred *cr)
1753 /* Is cr_flavor one of the gss "pseudoflavors"?: */
1754 return (cr->cr_flavor > RPC_AUTH_MAXFLAVOR);
1758 static bool
1759 same_creds(struct svc_cred *cr1, struct svc_cred *cr2)
1761 if ((is_gss_cred(cr1) != is_gss_cred(cr2))
1762 || (!uid_eq(cr1->cr_uid, cr2->cr_uid))
1763 || (!gid_eq(cr1->cr_gid, cr2->cr_gid))
1764 || !groups_equal(cr1->cr_group_info, cr2->cr_group_info))
1765 return false;
1766 if (cr1->cr_principal == cr2->cr_principal)
1767 return true;
1768 if (!cr1->cr_principal || !cr2->cr_principal)
1769 return false;
1770 return 0 == strcmp(cr1->cr_principal, cr2->cr_principal);
1773 static bool svc_rqst_integrity_protected(struct svc_rqst *rqstp)
1775 struct svc_cred *cr = &rqstp->rq_cred;
1776 u32 service;
1778 if (!cr->cr_gss_mech)
1779 return false;
1780 service = gss_pseudoflavor_to_service(cr->cr_gss_mech, cr->cr_flavor);
1781 return service == RPC_GSS_SVC_INTEGRITY ||
1782 service == RPC_GSS_SVC_PRIVACY;
1785 static bool mach_creds_match(struct nfs4_client *cl, struct svc_rqst *rqstp)
1787 struct svc_cred *cr = &rqstp->rq_cred;
1789 if (!cl->cl_mach_cred)
1790 return true;
1791 if (cl->cl_cred.cr_gss_mech != cr->cr_gss_mech)
1792 return false;
1793 if (!svc_rqst_integrity_protected(rqstp))
1794 return false;
1795 if (!cr->cr_principal)
1796 return false;
1797 return 0 == strcmp(cl->cl_cred.cr_principal, cr->cr_principal);
1800 static void gen_confirm(struct nfs4_client *clp, struct nfsd_net *nn)
1802 __be32 verf[2];
1805 * This is opaque to client, so no need to byte-swap. Use
1806 * __force to keep sparse happy
1808 verf[0] = (__force __be32)get_seconds();
1809 verf[1] = (__force __be32)nn->clientid_counter;
1810 memcpy(clp->cl_confirm.data, verf, sizeof(clp->cl_confirm.data));
1813 static void gen_clid(struct nfs4_client *clp, struct nfsd_net *nn)
1815 clp->cl_clientid.cl_boot = nn->boot_time;
1816 clp->cl_clientid.cl_id = nn->clientid_counter++;
1817 gen_confirm(clp, nn);
1820 static struct nfs4_stid *
1821 find_stateid_locked(struct nfs4_client *cl, stateid_t *t)
1823 struct nfs4_stid *ret;
1825 ret = idr_find(&cl->cl_stateids, t->si_opaque.so_id);
1826 if (!ret || !ret->sc_type)
1827 return NULL;
1828 return ret;
1831 static struct nfs4_stid *
1832 find_stateid_by_type(struct nfs4_client *cl, stateid_t *t, char typemask)
1834 struct nfs4_stid *s;
1836 spin_lock(&cl->cl_lock);
1837 s = find_stateid_locked(cl, t);
1838 if (s != NULL) {
1839 if (typemask & s->sc_type)
1840 atomic_inc(&s->sc_count);
1841 else
1842 s = NULL;
1844 spin_unlock(&cl->cl_lock);
1845 return s;
1848 static struct nfs4_client *create_client(struct xdr_netobj name,
1849 struct svc_rqst *rqstp, nfs4_verifier *verf)
1851 struct nfs4_client *clp;
1852 struct sockaddr *sa = svc_addr(rqstp);
1853 int ret;
1854 struct net *net = SVC_NET(rqstp);
1856 clp = alloc_client(name);
1857 if (clp == NULL)
1858 return NULL;
1860 ret = copy_cred(&clp->cl_cred, &rqstp->rq_cred);
1861 if (ret) {
1862 free_client(clp);
1863 return NULL;
1865 nfsd4_init_cb(&clp->cl_cb_null, clp, NULL, NFSPROC4_CLNT_CB_NULL);
1866 clp->cl_time = get_seconds();
1867 clear_bit(0, &clp->cl_cb_slot_busy);
1868 copy_verf(clp, verf);
1869 rpc_copy_addr((struct sockaddr *) &clp->cl_addr, sa);
1870 clp->cl_cb_session = NULL;
1871 clp->net = net;
1872 return clp;
1875 static void
1876 add_clp_to_name_tree(struct nfs4_client *new_clp, struct rb_root *root)
1878 struct rb_node **new = &(root->rb_node), *parent = NULL;
1879 struct nfs4_client *clp;
1881 while (*new) {
1882 clp = rb_entry(*new, struct nfs4_client, cl_namenode);
1883 parent = *new;
1885 if (compare_blob(&clp->cl_name, &new_clp->cl_name) > 0)
1886 new = &((*new)->rb_left);
1887 else
1888 new = &((*new)->rb_right);
1891 rb_link_node(&new_clp->cl_namenode, parent, new);
1892 rb_insert_color(&new_clp->cl_namenode, root);
1895 static struct nfs4_client *
1896 find_clp_in_name_tree(struct xdr_netobj *name, struct rb_root *root)
1898 int cmp;
1899 struct rb_node *node = root->rb_node;
1900 struct nfs4_client *clp;
1902 while (node) {
1903 clp = rb_entry(node, struct nfs4_client, cl_namenode);
1904 cmp = compare_blob(&clp->cl_name, name);
1905 if (cmp > 0)
1906 node = node->rb_left;
1907 else if (cmp < 0)
1908 node = node->rb_right;
1909 else
1910 return clp;
1912 return NULL;
1915 static void
1916 add_to_unconfirmed(struct nfs4_client *clp)
1918 unsigned int idhashval;
1919 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1921 lockdep_assert_held(&nn->client_lock);
1923 clear_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags);
1924 add_clp_to_name_tree(clp, &nn->unconf_name_tree);
1925 idhashval = clientid_hashval(clp->cl_clientid.cl_id);
1926 list_add(&clp->cl_idhash, &nn->unconf_id_hashtbl[idhashval]);
1927 renew_client_locked(clp);
1930 static void
1931 move_to_confirmed(struct nfs4_client *clp)
1933 unsigned int idhashval = clientid_hashval(clp->cl_clientid.cl_id);
1934 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1936 lockdep_assert_held(&nn->client_lock);
1938 dprintk("NFSD: move_to_confirm nfs4_client %p\n", clp);
1939 list_move(&clp->cl_idhash, &nn->conf_id_hashtbl[idhashval]);
1940 rb_erase(&clp->cl_namenode, &nn->unconf_name_tree);
1941 add_clp_to_name_tree(clp, &nn->conf_name_tree);
1942 set_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags);
1943 renew_client_locked(clp);
1946 static struct nfs4_client *
1947 find_client_in_id_table(struct list_head *tbl, clientid_t *clid, bool sessions)
1949 struct nfs4_client *clp;
1950 unsigned int idhashval = clientid_hashval(clid->cl_id);
1952 list_for_each_entry(clp, &tbl[idhashval], cl_idhash) {
1953 if (same_clid(&clp->cl_clientid, clid)) {
1954 if ((bool)clp->cl_minorversion != sessions)
1955 return NULL;
1956 renew_client_locked(clp);
1957 return clp;
1960 return NULL;
1963 static struct nfs4_client *
1964 find_confirmed_client(clientid_t *clid, bool sessions, struct nfsd_net *nn)
1966 struct list_head *tbl = nn->conf_id_hashtbl;
1968 lockdep_assert_held(&nn->client_lock);
1969 return find_client_in_id_table(tbl, clid, sessions);
1972 static struct nfs4_client *
1973 find_unconfirmed_client(clientid_t *clid, bool sessions, struct nfsd_net *nn)
1975 struct list_head *tbl = nn->unconf_id_hashtbl;
1977 lockdep_assert_held(&nn->client_lock);
1978 return find_client_in_id_table(tbl, clid, sessions);
1981 static bool clp_used_exchangeid(struct nfs4_client *clp)
1983 return clp->cl_exchange_flags != 0;
1986 static struct nfs4_client *
1987 find_confirmed_client_by_name(struct xdr_netobj *name, struct nfsd_net *nn)
1989 lockdep_assert_held(&nn->client_lock);
1990 return find_clp_in_name_tree(name, &nn->conf_name_tree);
1993 static struct nfs4_client *
1994 find_unconfirmed_client_by_name(struct xdr_netobj *name, struct nfsd_net *nn)
1996 lockdep_assert_held(&nn->client_lock);
1997 return find_clp_in_name_tree(name, &nn->unconf_name_tree);
2000 static void
2001 gen_callback(struct nfs4_client *clp, struct nfsd4_setclientid *se, struct svc_rqst *rqstp)
2003 struct nfs4_cb_conn *conn = &clp->cl_cb_conn;
2004 struct sockaddr *sa = svc_addr(rqstp);
2005 u32 scopeid = rpc_get_scope_id(sa);
2006 unsigned short expected_family;
2008 /* Currently, we only support tcp and tcp6 for the callback channel */
2009 if (se->se_callback_netid_len == 3 &&
2010 !memcmp(se->se_callback_netid_val, "tcp", 3))
2011 expected_family = AF_INET;
2012 else if (se->se_callback_netid_len == 4 &&
2013 !memcmp(se->se_callback_netid_val, "tcp6", 4))
2014 expected_family = AF_INET6;
2015 else
2016 goto out_err;
2018 conn->cb_addrlen = rpc_uaddr2sockaddr(clp->net, se->se_callback_addr_val,
2019 se->se_callback_addr_len,
2020 (struct sockaddr *)&conn->cb_addr,
2021 sizeof(conn->cb_addr));
2023 if (!conn->cb_addrlen || conn->cb_addr.ss_family != expected_family)
2024 goto out_err;
2026 if (conn->cb_addr.ss_family == AF_INET6)
2027 ((struct sockaddr_in6 *)&conn->cb_addr)->sin6_scope_id = scopeid;
2029 conn->cb_prog = se->se_callback_prog;
2030 conn->cb_ident = se->se_callback_ident;
2031 memcpy(&conn->cb_saddr, &rqstp->rq_daddr, rqstp->rq_daddrlen);
2032 return;
2033 out_err:
2034 conn->cb_addr.ss_family = AF_UNSPEC;
2035 conn->cb_addrlen = 0;
2036 dprintk(KERN_INFO "NFSD: this client (clientid %08x/%08x) "
2037 "will not receive delegations\n",
2038 clp->cl_clientid.cl_boot, clp->cl_clientid.cl_id);
2040 return;
2044 * Cache a reply. nfsd4_check_resp_size() has bounded the cache size.
2046 static void
2047 nfsd4_store_cache_entry(struct nfsd4_compoundres *resp)
2049 struct xdr_buf *buf = resp->xdr.buf;
2050 struct nfsd4_slot *slot = resp->cstate.slot;
2051 unsigned int base;
2053 dprintk("--> %s slot %p\n", __func__, slot);
2055 slot->sl_opcnt = resp->opcnt;
2056 slot->sl_status = resp->cstate.status;
2058 slot->sl_flags |= NFSD4_SLOT_INITIALIZED;
2059 if (nfsd4_not_cached(resp)) {
2060 slot->sl_datalen = 0;
2061 return;
2063 base = resp->cstate.data_offset;
2064 slot->sl_datalen = buf->len - base;
2065 if (read_bytes_from_xdr_buf(buf, base, slot->sl_data, slot->sl_datalen))
2066 WARN("%s: sessions DRC could not cache compound\n", __func__);
2067 return;
2071 * Encode the replay sequence operation from the slot values.
2072 * If cachethis is FALSE encode the uncached rep error on the next
2073 * operation which sets resp->p and increments resp->opcnt for
2074 * nfs4svc_encode_compoundres.
2077 static __be32
2078 nfsd4_enc_sequence_replay(struct nfsd4_compoundargs *args,
2079 struct nfsd4_compoundres *resp)
2081 struct nfsd4_op *op;
2082 struct nfsd4_slot *slot = resp->cstate.slot;
2084 /* Encode the replayed sequence operation */
2085 op = &args->ops[resp->opcnt - 1];
2086 nfsd4_encode_operation(resp, op);
2088 /* Return nfserr_retry_uncached_rep in next operation. */
2089 if (args->opcnt > 1 && !(slot->sl_flags & NFSD4_SLOT_CACHETHIS)) {
2090 op = &args->ops[resp->opcnt++];
2091 op->status = nfserr_retry_uncached_rep;
2092 nfsd4_encode_operation(resp, op);
2094 return op->status;
2098 * The sequence operation is not cached because we can use the slot and
2099 * session values.
2101 static __be32
2102 nfsd4_replay_cache_entry(struct nfsd4_compoundres *resp,
2103 struct nfsd4_sequence *seq)
2105 struct nfsd4_slot *slot = resp->cstate.slot;
2106 struct xdr_stream *xdr = &resp->xdr;
2107 __be32 *p;
2108 __be32 status;
2110 dprintk("--> %s slot %p\n", __func__, slot);
2112 status = nfsd4_enc_sequence_replay(resp->rqstp->rq_argp, resp);
2113 if (status)
2114 return status;
2116 p = xdr_reserve_space(xdr, slot->sl_datalen);
2117 if (!p) {
2118 WARN_ON_ONCE(1);
2119 return nfserr_serverfault;
2121 xdr_encode_opaque_fixed(p, slot->sl_data, slot->sl_datalen);
2122 xdr_commit_encode(xdr);
2124 resp->opcnt = slot->sl_opcnt;
2125 return slot->sl_status;
2129 * Set the exchange_id flags returned by the server.
2131 static void
2132 nfsd4_set_ex_flags(struct nfs4_client *new, struct nfsd4_exchange_id *clid)
2134 #ifdef CONFIG_NFSD_PNFS
2135 new->cl_exchange_flags |= EXCHGID4_FLAG_USE_PNFS_MDS;
2136 #else
2137 new->cl_exchange_flags |= EXCHGID4_FLAG_USE_NON_PNFS;
2138 #endif
2140 /* Referrals are supported, Migration is not. */
2141 new->cl_exchange_flags |= EXCHGID4_FLAG_SUPP_MOVED_REFER;
2143 /* set the wire flags to return to client. */
2144 clid->flags = new->cl_exchange_flags;
2147 static bool client_has_state(struct nfs4_client *clp)
2150 * Note clp->cl_openowners check isn't quite right: there's no
2151 * need to count owners without stateid's.
2153 * Also note we should probably be using this in 4.0 case too.
2155 return !list_empty(&clp->cl_openowners)
2156 || !list_empty(&clp->cl_delegations)
2157 || !list_empty(&clp->cl_sessions);
2160 __be32
2161 nfsd4_exchange_id(struct svc_rqst *rqstp,
2162 struct nfsd4_compound_state *cstate,
2163 struct nfsd4_exchange_id *exid)
2165 struct nfs4_client *conf, *new;
2166 struct nfs4_client *unconf = NULL;
2167 __be32 status;
2168 char addr_str[INET6_ADDRSTRLEN];
2169 nfs4_verifier verf = exid->verifier;
2170 struct sockaddr *sa = svc_addr(rqstp);
2171 bool update = exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A;
2172 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2174 rpc_ntop(sa, addr_str, sizeof(addr_str));
2175 dprintk("%s rqstp=%p exid=%p clname.len=%u clname.data=%p "
2176 "ip_addr=%s flags %x, spa_how %d\n",
2177 __func__, rqstp, exid, exid->clname.len, exid->clname.data,
2178 addr_str, exid->flags, exid->spa_how);
2180 if (exid->flags & ~EXCHGID4_FLAG_MASK_A)
2181 return nfserr_inval;
2183 switch (exid->spa_how) {
2184 case SP4_MACH_CRED:
2185 if (!svc_rqst_integrity_protected(rqstp))
2186 return nfserr_inval;
2187 case SP4_NONE:
2188 break;
2189 default: /* checked by xdr code */
2190 WARN_ON_ONCE(1);
2191 case SP4_SSV:
2192 return nfserr_encr_alg_unsupp;
2195 new = create_client(exid->clname, rqstp, &verf);
2196 if (new == NULL)
2197 return nfserr_jukebox;
2199 /* Cases below refer to rfc 5661 section 18.35.4: */
2200 spin_lock(&nn->client_lock);
2201 conf = find_confirmed_client_by_name(&exid->clname, nn);
2202 if (conf) {
2203 bool creds_match = same_creds(&conf->cl_cred, &rqstp->rq_cred);
2204 bool verfs_match = same_verf(&verf, &conf->cl_verifier);
2206 if (update) {
2207 if (!clp_used_exchangeid(conf)) { /* buggy client */
2208 status = nfserr_inval;
2209 goto out;
2211 if (!mach_creds_match(conf, rqstp)) {
2212 status = nfserr_wrong_cred;
2213 goto out;
2215 if (!creds_match) { /* case 9 */
2216 status = nfserr_perm;
2217 goto out;
2219 if (!verfs_match) { /* case 8 */
2220 status = nfserr_not_same;
2221 goto out;
2223 /* case 6 */
2224 exid->flags |= EXCHGID4_FLAG_CONFIRMED_R;
2225 goto out_copy;
2227 if (!creds_match) { /* case 3 */
2228 if (client_has_state(conf)) {
2229 status = nfserr_clid_inuse;
2230 goto out;
2232 goto out_new;
2234 if (verfs_match) { /* case 2 */
2235 conf->cl_exchange_flags |= EXCHGID4_FLAG_CONFIRMED_R;
2236 goto out_copy;
2238 /* case 5, client reboot */
2239 conf = NULL;
2240 goto out_new;
2243 if (update) { /* case 7 */
2244 status = nfserr_noent;
2245 goto out;
2248 unconf = find_unconfirmed_client_by_name(&exid->clname, nn);
2249 if (unconf) /* case 4, possible retry or client restart */
2250 unhash_client_locked(unconf);
2252 /* case 1 (normal case) */
2253 out_new:
2254 if (conf) {
2255 status = mark_client_expired_locked(conf);
2256 if (status)
2257 goto out;
2259 new->cl_minorversion = cstate->minorversion;
2260 new->cl_mach_cred = (exid->spa_how == SP4_MACH_CRED);
2262 gen_clid(new, nn);
2263 add_to_unconfirmed(new);
2264 swap(new, conf);
2265 out_copy:
2266 exid->clientid.cl_boot = conf->cl_clientid.cl_boot;
2267 exid->clientid.cl_id = conf->cl_clientid.cl_id;
2269 exid->seqid = conf->cl_cs_slot.sl_seqid + 1;
2270 nfsd4_set_ex_flags(conf, exid);
2272 dprintk("nfsd4_exchange_id seqid %d flags %x\n",
2273 conf->cl_cs_slot.sl_seqid, conf->cl_exchange_flags);
2274 status = nfs_ok;
2276 out:
2277 spin_unlock(&nn->client_lock);
2278 if (new)
2279 expire_client(new);
2280 if (unconf)
2281 expire_client(unconf);
2282 return status;
2285 static __be32
2286 check_slot_seqid(u32 seqid, u32 slot_seqid, int slot_inuse)
2288 dprintk("%s enter. seqid %d slot_seqid %d\n", __func__, seqid,
2289 slot_seqid);
2291 /* The slot is in use, and no response has been sent. */
2292 if (slot_inuse) {
2293 if (seqid == slot_seqid)
2294 return nfserr_jukebox;
2295 else
2296 return nfserr_seq_misordered;
2298 /* Note unsigned 32-bit arithmetic handles wraparound: */
2299 if (likely(seqid == slot_seqid + 1))
2300 return nfs_ok;
2301 if (seqid == slot_seqid)
2302 return nfserr_replay_cache;
2303 return nfserr_seq_misordered;
2307 * Cache the create session result into the create session single DRC
2308 * slot cache by saving the xdr structure. sl_seqid has been set.
2309 * Do this for solo or embedded create session operations.
2311 static void
2312 nfsd4_cache_create_session(struct nfsd4_create_session *cr_ses,
2313 struct nfsd4_clid_slot *slot, __be32 nfserr)
2315 slot->sl_status = nfserr;
2316 memcpy(&slot->sl_cr_ses, cr_ses, sizeof(*cr_ses));
2319 static __be32
2320 nfsd4_replay_create_session(struct nfsd4_create_session *cr_ses,
2321 struct nfsd4_clid_slot *slot)
2323 memcpy(cr_ses, &slot->sl_cr_ses, sizeof(*cr_ses));
2324 return slot->sl_status;
2327 #define NFSD_MIN_REQ_HDR_SEQ_SZ ((\
2328 2 * 2 + /* credential,verifier: AUTH_NULL, length 0 */ \
2329 1 + /* MIN tag is length with zero, only length */ \
2330 3 + /* version, opcount, opcode */ \
2331 XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
2332 /* seqid, slotID, slotID, cache */ \
2333 4 ) * sizeof(__be32))
2335 #define NFSD_MIN_RESP_HDR_SEQ_SZ ((\
2336 2 + /* verifier: AUTH_NULL, length 0 */\
2337 1 + /* status */ \
2338 1 + /* MIN tag is length with zero, only length */ \
2339 3 + /* opcount, opcode, opstatus*/ \
2340 XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
2341 /* seqid, slotID, slotID, slotID, status */ \
2342 5 ) * sizeof(__be32))
2344 static __be32 check_forechannel_attrs(struct nfsd4_channel_attrs *ca, struct nfsd_net *nn)
2346 u32 maxrpc = nn->nfsd_serv->sv_max_mesg;
2348 if (ca->maxreq_sz < NFSD_MIN_REQ_HDR_SEQ_SZ)
2349 return nfserr_toosmall;
2350 if (ca->maxresp_sz < NFSD_MIN_RESP_HDR_SEQ_SZ)
2351 return nfserr_toosmall;
2352 ca->headerpadsz = 0;
2353 ca->maxreq_sz = min_t(u32, ca->maxreq_sz, maxrpc);
2354 ca->maxresp_sz = min_t(u32, ca->maxresp_sz, maxrpc);
2355 ca->maxops = min_t(u32, ca->maxops, NFSD_MAX_OPS_PER_COMPOUND);
2356 ca->maxresp_cached = min_t(u32, ca->maxresp_cached,
2357 NFSD_SLOT_CACHE_SIZE + NFSD_MIN_HDR_SEQ_SZ);
2358 ca->maxreqs = min_t(u32, ca->maxreqs, NFSD_MAX_SLOTS_PER_SESSION);
2360 * Note decreasing slot size below client's request may make it
2361 * difficult for client to function correctly, whereas
2362 * decreasing the number of slots will (just?) affect
2363 * performance. When short on memory we therefore prefer to
2364 * decrease number of slots instead of their size. Clients that
2365 * request larger slots than they need will get poor results:
2367 ca->maxreqs = nfsd4_get_drc_mem(ca);
2368 if (!ca->maxreqs)
2369 return nfserr_jukebox;
2371 return nfs_ok;
2374 #define NFSD_CB_MAX_REQ_SZ ((NFS4_enc_cb_recall_sz + \
2375 RPC_MAX_HEADER_WITH_AUTH) * sizeof(__be32))
2376 #define NFSD_CB_MAX_RESP_SZ ((NFS4_dec_cb_recall_sz + \
2377 RPC_MAX_REPHEADER_WITH_AUTH) * sizeof(__be32))
2379 static __be32 check_backchannel_attrs(struct nfsd4_channel_attrs *ca)
2381 ca->headerpadsz = 0;
2384 * These RPC_MAX_HEADER macros are overkill, especially since we
2385 * don't even do gss on the backchannel yet. But this is still
2386 * less than 1k. Tighten up this estimate in the unlikely event
2387 * it turns out to be a problem for some client:
2389 if (ca->maxreq_sz < NFSD_CB_MAX_REQ_SZ)
2390 return nfserr_toosmall;
2391 if (ca->maxresp_sz < NFSD_CB_MAX_RESP_SZ)
2392 return nfserr_toosmall;
2393 ca->maxresp_cached = 0;
2394 if (ca->maxops < 2)
2395 return nfserr_toosmall;
2397 return nfs_ok;
2400 static __be32 nfsd4_check_cb_sec(struct nfsd4_cb_sec *cbs)
2402 switch (cbs->flavor) {
2403 case RPC_AUTH_NULL:
2404 case RPC_AUTH_UNIX:
2405 return nfs_ok;
2406 default:
2408 * GSS case: the spec doesn't allow us to return this
2409 * error. But it also doesn't allow us not to support
2410 * GSS.
2411 * I'd rather this fail hard than return some error the
2412 * client might think it can already handle:
2414 return nfserr_encr_alg_unsupp;
2418 __be32
2419 nfsd4_create_session(struct svc_rqst *rqstp,
2420 struct nfsd4_compound_state *cstate,
2421 struct nfsd4_create_session *cr_ses)
2423 struct sockaddr *sa = svc_addr(rqstp);
2424 struct nfs4_client *conf, *unconf;
2425 struct nfs4_client *old = NULL;
2426 struct nfsd4_session *new;
2427 struct nfsd4_conn *conn;
2428 struct nfsd4_clid_slot *cs_slot = NULL;
2429 __be32 status = 0;
2430 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2432 if (cr_ses->flags & ~SESSION4_FLAG_MASK_A)
2433 return nfserr_inval;
2434 status = nfsd4_check_cb_sec(&cr_ses->cb_sec);
2435 if (status)
2436 return status;
2437 status = check_forechannel_attrs(&cr_ses->fore_channel, nn);
2438 if (status)
2439 return status;
2440 status = check_backchannel_attrs(&cr_ses->back_channel);
2441 if (status)
2442 goto out_release_drc_mem;
2443 status = nfserr_jukebox;
2444 new = alloc_session(&cr_ses->fore_channel, &cr_ses->back_channel);
2445 if (!new)
2446 goto out_release_drc_mem;
2447 conn = alloc_conn_from_crses(rqstp, cr_ses);
2448 if (!conn)
2449 goto out_free_session;
2451 spin_lock(&nn->client_lock);
2452 unconf = find_unconfirmed_client(&cr_ses->clientid, true, nn);
2453 conf = find_confirmed_client(&cr_ses->clientid, true, nn);
2454 WARN_ON_ONCE(conf && unconf);
2456 if (conf) {
2457 status = nfserr_wrong_cred;
2458 if (!mach_creds_match(conf, rqstp))
2459 goto out_free_conn;
2460 cs_slot = &conf->cl_cs_slot;
2461 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
2462 if (status == nfserr_replay_cache) {
2463 status = nfsd4_replay_create_session(cr_ses, cs_slot);
2464 goto out_free_conn;
2465 } else if (cr_ses->seqid != cs_slot->sl_seqid + 1) {
2466 status = nfserr_seq_misordered;
2467 goto out_free_conn;
2469 } else if (unconf) {
2470 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred) ||
2471 !rpc_cmp_addr(sa, (struct sockaddr *) &unconf->cl_addr)) {
2472 status = nfserr_clid_inuse;
2473 goto out_free_conn;
2475 status = nfserr_wrong_cred;
2476 if (!mach_creds_match(unconf, rqstp))
2477 goto out_free_conn;
2478 cs_slot = &unconf->cl_cs_slot;
2479 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
2480 if (status) {
2481 /* an unconfirmed replay returns misordered */
2482 status = nfserr_seq_misordered;
2483 goto out_free_conn;
2485 old = find_confirmed_client_by_name(&unconf->cl_name, nn);
2486 if (old) {
2487 status = mark_client_expired_locked(old);
2488 if (status) {
2489 old = NULL;
2490 goto out_free_conn;
2493 move_to_confirmed(unconf);
2494 conf = unconf;
2495 } else {
2496 status = nfserr_stale_clientid;
2497 goto out_free_conn;
2499 status = nfs_ok;
2501 * We do not support RDMA or persistent sessions
2503 cr_ses->flags &= ~SESSION4_PERSIST;
2504 cr_ses->flags &= ~SESSION4_RDMA;
2506 init_session(rqstp, new, conf, cr_ses);
2507 nfsd4_get_session_locked(new);
2509 memcpy(cr_ses->sessionid.data, new->se_sessionid.data,
2510 NFS4_MAX_SESSIONID_LEN);
2511 cs_slot->sl_seqid++;
2512 cr_ses->seqid = cs_slot->sl_seqid;
2514 /* cache solo and embedded create sessions under the client_lock */
2515 nfsd4_cache_create_session(cr_ses, cs_slot, status);
2516 spin_unlock(&nn->client_lock);
2517 /* init connection and backchannel */
2518 nfsd4_init_conn(rqstp, conn, new);
2519 nfsd4_put_session(new);
2520 if (old)
2521 expire_client(old);
2522 return status;
2523 out_free_conn:
2524 spin_unlock(&nn->client_lock);
2525 free_conn(conn);
2526 if (old)
2527 expire_client(old);
2528 out_free_session:
2529 __free_session(new);
2530 out_release_drc_mem:
2531 nfsd4_put_drc_mem(&cr_ses->fore_channel);
2532 return status;
2535 static __be32 nfsd4_map_bcts_dir(u32 *dir)
2537 switch (*dir) {
2538 case NFS4_CDFC4_FORE:
2539 case NFS4_CDFC4_BACK:
2540 return nfs_ok;
2541 case NFS4_CDFC4_FORE_OR_BOTH:
2542 case NFS4_CDFC4_BACK_OR_BOTH:
2543 *dir = NFS4_CDFC4_BOTH;
2544 return nfs_ok;
2546 return nfserr_inval;
2549 __be32 nfsd4_backchannel_ctl(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate, struct nfsd4_backchannel_ctl *bc)
2551 struct nfsd4_session *session = cstate->session;
2552 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2553 __be32 status;
2555 status = nfsd4_check_cb_sec(&bc->bc_cb_sec);
2556 if (status)
2557 return status;
2558 spin_lock(&nn->client_lock);
2559 session->se_cb_prog = bc->bc_cb_program;
2560 session->se_cb_sec = bc->bc_cb_sec;
2561 spin_unlock(&nn->client_lock);
2563 nfsd4_probe_callback(session->se_client);
2565 return nfs_ok;
2568 __be32 nfsd4_bind_conn_to_session(struct svc_rqst *rqstp,
2569 struct nfsd4_compound_state *cstate,
2570 struct nfsd4_bind_conn_to_session *bcts)
2572 __be32 status;
2573 struct nfsd4_conn *conn;
2574 struct nfsd4_session *session;
2575 struct net *net = SVC_NET(rqstp);
2576 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
2578 if (!nfsd4_last_compound_op(rqstp))
2579 return nfserr_not_only_op;
2580 spin_lock(&nn->client_lock);
2581 session = find_in_sessionid_hashtbl(&bcts->sessionid, net, &status);
2582 spin_unlock(&nn->client_lock);
2583 if (!session)
2584 goto out_no_session;
2585 status = nfserr_wrong_cred;
2586 if (!mach_creds_match(session->se_client, rqstp))
2587 goto out;
2588 status = nfsd4_map_bcts_dir(&bcts->dir);
2589 if (status)
2590 goto out;
2591 conn = alloc_conn(rqstp, bcts->dir);
2592 status = nfserr_jukebox;
2593 if (!conn)
2594 goto out;
2595 nfsd4_init_conn(rqstp, conn, session);
2596 status = nfs_ok;
2597 out:
2598 nfsd4_put_session(session);
2599 out_no_session:
2600 return status;
2603 static bool nfsd4_compound_in_session(struct nfsd4_session *session, struct nfs4_sessionid *sid)
2605 if (!session)
2606 return 0;
2607 return !memcmp(sid, &session->se_sessionid, sizeof(*sid));
2610 __be32
2611 nfsd4_destroy_session(struct svc_rqst *r,
2612 struct nfsd4_compound_state *cstate,
2613 struct nfsd4_destroy_session *sessionid)
2615 struct nfsd4_session *ses;
2616 __be32 status;
2617 int ref_held_by_me = 0;
2618 struct net *net = SVC_NET(r);
2619 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
2621 status = nfserr_not_only_op;
2622 if (nfsd4_compound_in_session(cstate->session, &sessionid->sessionid)) {
2623 if (!nfsd4_last_compound_op(r))
2624 goto out;
2625 ref_held_by_me++;
2627 dump_sessionid(__func__, &sessionid->sessionid);
2628 spin_lock(&nn->client_lock);
2629 ses = find_in_sessionid_hashtbl(&sessionid->sessionid, net, &status);
2630 if (!ses)
2631 goto out_client_lock;
2632 status = nfserr_wrong_cred;
2633 if (!mach_creds_match(ses->se_client, r))
2634 goto out_put_session;
2635 status = mark_session_dead_locked(ses, 1 + ref_held_by_me);
2636 if (status)
2637 goto out_put_session;
2638 unhash_session(ses);
2639 spin_unlock(&nn->client_lock);
2641 nfsd4_probe_callback_sync(ses->se_client);
2643 spin_lock(&nn->client_lock);
2644 status = nfs_ok;
2645 out_put_session:
2646 nfsd4_put_session_locked(ses);
2647 out_client_lock:
2648 spin_unlock(&nn->client_lock);
2649 out:
2650 return status;
2653 static struct nfsd4_conn *__nfsd4_find_conn(struct svc_xprt *xpt, struct nfsd4_session *s)
2655 struct nfsd4_conn *c;
2657 list_for_each_entry(c, &s->se_conns, cn_persession) {
2658 if (c->cn_xprt == xpt) {
2659 return c;
2662 return NULL;
2665 static __be32 nfsd4_sequence_check_conn(struct nfsd4_conn *new, struct nfsd4_session *ses)
2667 struct nfs4_client *clp = ses->se_client;
2668 struct nfsd4_conn *c;
2669 __be32 status = nfs_ok;
2670 int ret;
2672 spin_lock(&clp->cl_lock);
2673 c = __nfsd4_find_conn(new->cn_xprt, ses);
2674 if (c)
2675 goto out_free;
2676 status = nfserr_conn_not_bound_to_session;
2677 if (clp->cl_mach_cred)
2678 goto out_free;
2679 __nfsd4_hash_conn(new, ses);
2680 spin_unlock(&clp->cl_lock);
2681 ret = nfsd4_register_conn(new);
2682 if (ret)
2683 /* oops; xprt is already down: */
2684 nfsd4_conn_lost(&new->cn_xpt_user);
2685 return nfs_ok;
2686 out_free:
2687 spin_unlock(&clp->cl_lock);
2688 free_conn(new);
2689 return status;
2692 static bool nfsd4_session_too_many_ops(struct svc_rqst *rqstp, struct nfsd4_session *session)
2694 struct nfsd4_compoundargs *args = rqstp->rq_argp;
2696 return args->opcnt > session->se_fchannel.maxops;
2699 static bool nfsd4_request_too_big(struct svc_rqst *rqstp,
2700 struct nfsd4_session *session)
2702 struct xdr_buf *xb = &rqstp->rq_arg;
2704 return xb->len > session->se_fchannel.maxreq_sz;
2707 __be32
2708 nfsd4_sequence(struct svc_rqst *rqstp,
2709 struct nfsd4_compound_state *cstate,
2710 struct nfsd4_sequence *seq)
2712 struct nfsd4_compoundres *resp = rqstp->rq_resp;
2713 struct xdr_stream *xdr = &resp->xdr;
2714 struct nfsd4_session *session;
2715 struct nfs4_client *clp;
2716 struct nfsd4_slot *slot;
2717 struct nfsd4_conn *conn;
2718 __be32 status;
2719 int buflen;
2720 struct net *net = SVC_NET(rqstp);
2721 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
2723 if (resp->opcnt != 1)
2724 return nfserr_sequence_pos;
2727 * Will be either used or freed by nfsd4_sequence_check_conn
2728 * below.
2730 conn = alloc_conn(rqstp, NFS4_CDFC4_FORE);
2731 if (!conn)
2732 return nfserr_jukebox;
2734 spin_lock(&nn->client_lock);
2735 session = find_in_sessionid_hashtbl(&seq->sessionid, net, &status);
2736 if (!session)
2737 goto out_no_session;
2738 clp = session->se_client;
2740 status = nfserr_too_many_ops;
2741 if (nfsd4_session_too_many_ops(rqstp, session))
2742 goto out_put_session;
2744 status = nfserr_req_too_big;
2745 if (nfsd4_request_too_big(rqstp, session))
2746 goto out_put_session;
2748 status = nfserr_badslot;
2749 if (seq->slotid >= session->se_fchannel.maxreqs)
2750 goto out_put_session;
2752 slot = session->se_slots[seq->slotid];
2753 dprintk("%s: slotid %d\n", __func__, seq->slotid);
2755 /* We do not negotiate the number of slots yet, so set the
2756 * maxslots to the session maxreqs which is used to encode
2757 * sr_highest_slotid and the sr_target_slot id to maxslots */
2758 seq->maxslots = session->se_fchannel.maxreqs;
2760 status = check_slot_seqid(seq->seqid, slot->sl_seqid,
2761 slot->sl_flags & NFSD4_SLOT_INUSE);
2762 if (status == nfserr_replay_cache) {
2763 status = nfserr_seq_misordered;
2764 if (!(slot->sl_flags & NFSD4_SLOT_INITIALIZED))
2765 goto out_put_session;
2766 cstate->slot = slot;
2767 cstate->session = session;
2768 cstate->clp = clp;
2769 /* Return the cached reply status and set cstate->status
2770 * for nfsd4_proc_compound processing */
2771 status = nfsd4_replay_cache_entry(resp, seq);
2772 cstate->status = nfserr_replay_cache;
2773 goto out;
2775 if (status)
2776 goto out_put_session;
2778 status = nfsd4_sequence_check_conn(conn, session);
2779 conn = NULL;
2780 if (status)
2781 goto out_put_session;
2783 buflen = (seq->cachethis) ?
2784 session->se_fchannel.maxresp_cached :
2785 session->se_fchannel.maxresp_sz;
2786 status = (seq->cachethis) ? nfserr_rep_too_big_to_cache :
2787 nfserr_rep_too_big;
2788 if (xdr_restrict_buflen(xdr, buflen - rqstp->rq_auth_slack))
2789 goto out_put_session;
2790 svc_reserve(rqstp, buflen);
2792 status = nfs_ok;
2793 /* Success! bump slot seqid */
2794 slot->sl_seqid = seq->seqid;
2795 slot->sl_flags |= NFSD4_SLOT_INUSE;
2796 if (seq->cachethis)
2797 slot->sl_flags |= NFSD4_SLOT_CACHETHIS;
2798 else
2799 slot->sl_flags &= ~NFSD4_SLOT_CACHETHIS;
2801 cstate->slot = slot;
2802 cstate->session = session;
2803 cstate->clp = clp;
2805 out:
2806 switch (clp->cl_cb_state) {
2807 case NFSD4_CB_DOWN:
2808 seq->status_flags = SEQ4_STATUS_CB_PATH_DOWN;
2809 break;
2810 case NFSD4_CB_FAULT:
2811 seq->status_flags = SEQ4_STATUS_BACKCHANNEL_FAULT;
2812 break;
2813 default:
2814 seq->status_flags = 0;
2816 if (!list_empty(&clp->cl_revoked))
2817 seq->status_flags |= SEQ4_STATUS_RECALLABLE_STATE_REVOKED;
2818 out_no_session:
2819 if (conn)
2820 free_conn(conn);
2821 spin_unlock(&nn->client_lock);
2822 return status;
2823 out_put_session:
2824 nfsd4_put_session_locked(session);
2825 goto out_no_session;
2828 void
2829 nfsd4_sequence_done(struct nfsd4_compoundres *resp)
2831 struct nfsd4_compound_state *cs = &resp->cstate;
2833 if (nfsd4_has_session(cs)) {
2834 if (cs->status != nfserr_replay_cache) {
2835 nfsd4_store_cache_entry(resp);
2836 cs->slot->sl_flags &= ~NFSD4_SLOT_INUSE;
2838 /* Drop session reference that was taken in nfsd4_sequence() */
2839 nfsd4_put_session(cs->session);
2840 } else if (cs->clp)
2841 put_client_renew(cs->clp);
2844 __be32
2845 nfsd4_destroy_clientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate, struct nfsd4_destroy_clientid *dc)
2847 struct nfs4_client *conf, *unconf;
2848 struct nfs4_client *clp = NULL;
2849 __be32 status = 0;
2850 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2852 spin_lock(&nn->client_lock);
2853 unconf = find_unconfirmed_client(&dc->clientid, true, nn);
2854 conf = find_confirmed_client(&dc->clientid, true, nn);
2855 WARN_ON_ONCE(conf && unconf);
2857 if (conf) {
2858 if (client_has_state(conf)) {
2859 status = nfserr_clientid_busy;
2860 goto out;
2862 status = mark_client_expired_locked(conf);
2863 if (status)
2864 goto out;
2865 clp = conf;
2866 } else if (unconf)
2867 clp = unconf;
2868 else {
2869 status = nfserr_stale_clientid;
2870 goto out;
2872 if (!mach_creds_match(clp, rqstp)) {
2873 clp = NULL;
2874 status = nfserr_wrong_cred;
2875 goto out;
2877 unhash_client_locked(clp);
2878 out:
2879 spin_unlock(&nn->client_lock);
2880 if (clp)
2881 expire_client(clp);
2882 return status;
2885 __be32
2886 nfsd4_reclaim_complete(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate, struct nfsd4_reclaim_complete *rc)
2888 __be32 status = 0;
2890 if (rc->rca_one_fs) {
2891 if (!cstate->current_fh.fh_dentry)
2892 return nfserr_nofilehandle;
2894 * We don't take advantage of the rca_one_fs case.
2895 * That's OK, it's optional, we can safely ignore it.
2897 return nfs_ok;
2900 status = nfserr_complete_already;
2901 if (test_and_set_bit(NFSD4_CLIENT_RECLAIM_COMPLETE,
2902 &cstate->session->se_client->cl_flags))
2903 goto out;
2905 status = nfserr_stale_clientid;
2906 if (is_client_expired(cstate->session->se_client))
2908 * The following error isn't really legal.
2909 * But we only get here if the client just explicitly
2910 * destroyed the client. Surely it no longer cares what
2911 * error it gets back on an operation for the dead
2912 * client.
2914 goto out;
2916 status = nfs_ok;
2917 nfsd4_client_record_create(cstate->session->se_client);
2918 out:
2919 return status;
2922 __be32
2923 nfsd4_setclientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2924 struct nfsd4_setclientid *setclid)
2926 struct xdr_netobj clname = setclid->se_name;
2927 nfs4_verifier clverifier = setclid->se_verf;
2928 struct nfs4_client *conf, *new;
2929 struct nfs4_client *unconf = NULL;
2930 __be32 status;
2931 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2933 new = create_client(clname, rqstp, &clverifier);
2934 if (new == NULL)
2935 return nfserr_jukebox;
2936 /* Cases below refer to rfc 3530 section 14.2.33: */
2937 spin_lock(&nn->client_lock);
2938 conf = find_confirmed_client_by_name(&clname, nn);
2939 if (conf) {
2940 /* case 0: */
2941 status = nfserr_clid_inuse;
2942 if (clp_used_exchangeid(conf))
2943 goto out;
2944 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
2945 char addr_str[INET6_ADDRSTRLEN];
2946 rpc_ntop((struct sockaddr *) &conf->cl_addr, addr_str,
2947 sizeof(addr_str));
2948 dprintk("NFSD: setclientid: string in use by client "
2949 "at %s\n", addr_str);
2950 goto out;
2953 unconf = find_unconfirmed_client_by_name(&clname, nn);
2954 if (unconf)
2955 unhash_client_locked(unconf);
2956 if (conf && same_verf(&conf->cl_verifier, &clverifier))
2957 /* case 1: probable callback update */
2958 copy_clid(new, conf);
2959 else /* case 4 (new client) or cases 2, 3 (client reboot): */
2960 gen_clid(new, nn);
2961 new->cl_minorversion = 0;
2962 gen_callback(new, setclid, rqstp);
2963 add_to_unconfirmed(new);
2964 setclid->se_clientid.cl_boot = new->cl_clientid.cl_boot;
2965 setclid->se_clientid.cl_id = new->cl_clientid.cl_id;
2966 memcpy(setclid->se_confirm.data, new->cl_confirm.data, sizeof(setclid->se_confirm.data));
2967 new = NULL;
2968 status = nfs_ok;
2969 out:
2970 spin_unlock(&nn->client_lock);
2971 if (new)
2972 free_client(new);
2973 if (unconf)
2974 expire_client(unconf);
2975 return status;
2979 __be32
2980 nfsd4_setclientid_confirm(struct svc_rqst *rqstp,
2981 struct nfsd4_compound_state *cstate,
2982 struct nfsd4_setclientid_confirm *setclientid_confirm)
2984 struct nfs4_client *conf, *unconf;
2985 struct nfs4_client *old = NULL;
2986 nfs4_verifier confirm = setclientid_confirm->sc_confirm;
2987 clientid_t * clid = &setclientid_confirm->sc_clientid;
2988 __be32 status;
2989 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2991 if (STALE_CLIENTID(clid, nn))
2992 return nfserr_stale_clientid;
2994 spin_lock(&nn->client_lock);
2995 conf = find_confirmed_client(clid, false, nn);
2996 unconf = find_unconfirmed_client(clid, false, nn);
2998 * We try hard to give out unique clientid's, so if we get an
2999 * attempt to confirm the same clientid with a different cred,
3000 * there's a bug somewhere. Let's charitably assume it's our
3001 * bug.
3003 status = nfserr_serverfault;
3004 if (unconf && !same_creds(&unconf->cl_cred, &rqstp->rq_cred))
3005 goto out;
3006 if (conf && !same_creds(&conf->cl_cred, &rqstp->rq_cred))
3007 goto out;
3008 /* cases below refer to rfc 3530 section 14.2.34: */
3009 if (!unconf || !same_verf(&confirm, &unconf->cl_confirm)) {
3010 if (conf && !unconf) /* case 2: probable retransmit */
3011 status = nfs_ok;
3012 else /* case 4: client hasn't noticed we rebooted yet? */
3013 status = nfserr_stale_clientid;
3014 goto out;
3016 status = nfs_ok;
3017 if (conf) { /* case 1: callback update */
3018 old = unconf;
3019 unhash_client_locked(old);
3020 nfsd4_change_callback(conf, &unconf->cl_cb_conn);
3021 } else { /* case 3: normal case; new or rebooted client */
3022 old = find_confirmed_client_by_name(&unconf->cl_name, nn);
3023 if (old) {
3024 status = mark_client_expired_locked(old);
3025 if (status) {
3026 old = NULL;
3027 goto out;
3030 move_to_confirmed(unconf);
3031 conf = unconf;
3033 get_client_locked(conf);
3034 spin_unlock(&nn->client_lock);
3035 nfsd4_probe_callback(conf);
3036 spin_lock(&nn->client_lock);
3037 put_client_renew_locked(conf);
3038 out:
3039 spin_unlock(&nn->client_lock);
3040 if (old)
3041 expire_client(old);
3042 return status;
3045 static struct nfs4_file *nfsd4_alloc_file(void)
3047 return kmem_cache_alloc(file_slab, GFP_KERNEL);
3050 /* OPEN Share state helper functions */
3051 static void nfsd4_init_file(struct knfsd_fh *fh, unsigned int hashval,
3052 struct nfs4_file *fp)
3054 lockdep_assert_held(&state_lock);
3056 atomic_set(&fp->fi_ref, 1);
3057 spin_lock_init(&fp->fi_lock);
3058 INIT_LIST_HEAD(&fp->fi_stateids);
3059 INIT_LIST_HEAD(&fp->fi_delegations);
3060 fh_copy_shallow(&fp->fi_fhandle, fh);
3061 fp->fi_deleg_file = NULL;
3062 fp->fi_had_conflict = false;
3063 fp->fi_share_deny = 0;
3064 memset(fp->fi_fds, 0, sizeof(fp->fi_fds));
3065 memset(fp->fi_access, 0, sizeof(fp->fi_access));
3066 #ifdef CONFIG_NFSD_PNFS
3067 INIT_LIST_HEAD(&fp->fi_lo_states);
3068 atomic_set(&fp->fi_lo_recalls, 0);
3069 #endif
3070 hlist_add_head_rcu(&fp->fi_hash, &file_hashtbl[hashval]);
3073 void
3074 nfsd4_free_slabs(void)
3076 kmem_cache_destroy(openowner_slab);
3077 kmem_cache_destroy(lockowner_slab);
3078 kmem_cache_destroy(file_slab);
3079 kmem_cache_destroy(stateid_slab);
3080 kmem_cache_destroy(deleg_slab);
3084 nfsd4_init_slabs(void)
3086 openowner_slab = kmem_cache_create("nfsd4_openowners",
3087 sizeof(struct nfs4_openowner), 0, 0, NULL);
3088 if (openowner_slab == NULL)
3089 goto out;
3090 lockowner_slab = kmem_cache_create("nfsd4_lockowners",
3091 sizeof(struct nfs4_lockowner), 0, 0, NULL);
3092 if (lockowner_slab == NULL)
3093 goto out_free_openowner_slab;
3094 file_slab = kmem_cache_create("nfsd4_files",
3095 sizeof(struct nfs4_file), 0, 0, NULL);
3096 if (file_slab == NULL)
3097 goto out_free_lockowner_slab;
3098 stateid_slab = kmem_cache_create("nfsd4_stateids",
3099 sizeof(struct nfs4_ol_stateid), 0, 0, NULL);
3100 if (stateid_slab == NULL)
3101 goto out_free_file_slab;
3102 deleg_slab = kmem_cache_create("nfsd4_delegations",
3103 sizeof(struct nfs4_delegation), 0, 0, NULL);
3104 if (deleg_slab == NULL)
3105 goto out_free_stateid_slab;
3106 return 0;
3108 out_free_stateid_slab:
3109 kmem_cache_destroy(stateid_slab);
3110 out_free_file_slab:
3111 kmem_cache_destroy(file_slab);
3112 out_free_lockowner_slab:
3113 kmem_cache_destroy(lockowner_slab);
3114 out_free_openowner_slab:
3115 kmem_cache_destroy(openowner_slab);
3116 out:
3117 dprintk("nfsd4: out of memory while initializing nfsv4\n");
3118 return -ENOMEM;
3121 static void init_nfs4_replay(struct nfs4_replay *rp)
3123 rp->rp_status = nfserr_serverfault;
3124 rp->rp_buflen = 0;
3125 rp->rp_buf = rp->rp_ibuf;
3126 mutex_init(&rp->rp_mutex);
3129 static void nfsd4_cstate_assign_replay(struct nfsd4_compound_state *cstate,
3130 struct nfs4_stateowner *so)
3132 if (!nfsd4_has_session(cstate)) {
3133 mutex_lock(&so->so_replay.rp_mutex);
3134 cstate->replay_owner = nfs4_get_stateowner(so);
3138 void nfsd4_cstate_clear_replay(struct nfsd4_compound_state *cstate)
3140 struct nfs4_stateowner *so = cstate->replay_owner;
3142 if (so != NULL) {
3143 cstate->replay_owner = NULL;
3144 mutex_unlock(&so->so_replay.rp_mutex);
3145 nfs4_put_stateowner(so);
3149 static inline void *alloc_stateowner(struct kmem_cache *slab, struct xdr_netobj *owner, struct nfs4_client *clp)
3151 struct nfs4_stateowner *sop;
3153 sop = kmem_cache_alloc(slab, GFP_KERNEL);
3154 if (!sop)
3155 return NULL;
3157 sop->so_owner.data = kmemdup(owner->data, owner->len, GFP_KERNEL);
3158 if (!sop->so_owner.data) {
3159 kmem_cache_free(slab, sop);
3160 return NULL;
3162 sop->so_owner.len = owner->len;
3164 INIT_LIST_HEAD(&sop->so_stateids);
3165 sop->so_client = clp;
3166 init_nfs4_replay(&sop->so_replay);
3167 atomic_set(&sop->so_count, 1);
3168 return sop;
3171 static void hash_openowner(struct nfs4_openowner *oo, struct nfs4_client *clp, unsigned int strhashval)
3173 lockdep_assert_held(&clp->cl_lock);
3175 list_add(&oo->oo_owner.so_strhash,
3176 &clp->cl_ownerstr_hashtbl[strhashval]);
3177 list_add(&oo->oo_perclient, &clp->cl_openowners);
3180 static void nfs4_unhash_openowner(struct nfs4_stateowner *so)
3182 unhash_openowner_locked(openowner(so));
3185 static void nfs4_free_openowner(struct nfs4_stateowner *so)
3187 struct nfs4_openowner *oo = openowner(so);
3189 kmem_cache_free(openowner_slab, oo);
3192 static const struct nfs4_stateowner_operations openowner_ops = {
3193 .so_unhash = nfs4_unhash_openowner,
3194 .so_free = nfs4_free_openowner,
3197 static struct nfs4_openowner *
3198 alloc_init_open_stateowner(unsigned int strhashval, struct nfsd4_open *open,
3199 struct nfsd4_compound_state *cstate)
3201 struct nfs4_client *clp = cstate->clp;
3202 struct nfs4_openowner *oo, *ret;
3204 oo = alloc_stateowner(openowner_slab, &open->op_owner, clp);
3205 if (!oo)
3206 return NULL;
3207 oo->oo_owner.so_ops = &openowner_ops;
3208 oo->oo_owner.so_is_open_owner = 1;
3209 oo->oo_owner.so_seqid = open->op_seqid;
3210 oo->oo_flags = 0;
3211 if (nfsd4_has_session(cstate))
3212 oo->oo_flags |= NFS4_OO_CONFIRMED;
3213 oo->oo_time = 0;
3214 oo->oo_last_closed_stid = NULL;
3215 INIT_LIST_HEAD(&oo->oo_close_lru);
3216 spin_lock(&clp->cl_lock);
3217 ret = find_openstateowner_str_locked(strhashval, open, clp);
3218 if (ret == NULL) {
3219 hash_openowner(oo, clp, strhashval);
3220 ret = oo;
3221 } else
3222 nfs4_free_openowner(&oo->oo_owner);
3223 spin_unlock(&clp->cl_lock);
3224 return ret;
3227 static void init_open_stateid(struct nfs4_ol_stateid *stp, struct nfs4_file *fp, struct nfsd4_open *open) {
3228 struct nfs4_openowner *oo = open->op_openowner;
3230 atomic_inc(&stp->st_stid.sc_count);
3231 stp->st_stid.sc_type = NFS4_OPEN_STID;
3232 INIT_LIST_HEAD(&stp->st_locks);
3233 stp->st_stateowner = nfs4_get_stateowner(&oo->oo_owner);
3234 get_nfs4_file(fp);
3235 stp->st_stid.sc_file = fp;
3236 stp->st_access_bmap = 0;
3237 stp->st_deny_bmap = 0;
3238 stp->st_openstp = NULL;
3239 spin_lock(&oo->oo_owner.so_client->cl_lock);
3240 list_add(&stp->st_perstateowner, &oo->oo_owner.so_stateids);
3241 spin_lock(&fp->fi_lock);
3242 list_add(&stp->st_perfile, &fp->fi_stateids);
3243 spin_unlock(&fp->fi_lock);
3244 spin_unlock(&oo->oo_owner.so_client->cl_lock);
3248 * In the 4.0 case we need to keep the owners around a little while to handle
3249 * CLOSE replay. We still do need to release any file access that is held by
3250 * them before returning however.
3252 static void
3253 move_to_close_lru(struct nfs4_ol_stateid *s, struct net *net)
3255 struct nfs4_ol_stateid *last;
3256 struct nfs4_openowner *oo = openowner(s->st_stateowner);
3257 struct nfsd_net *nn = net_generic(s->st_stid.sc_client->net,
3258 nfsd_net_id);
3260 dprintk("NFSD: move_to_close_lru nfs4_openowner %p\n", oo);
3263 * We know that we hold one reference via nfsd4_close, and another
3264 * "persistent" reference for the client. If the refcount is higher
3265 * than 2, then there are still calls in progress that are using this
3266 * stateid. We can't put the sc_file reference until they are finished.
3267 * Wait for the refcount to drop to 2. Since it has been unhashed,
3268 * there should be no danger of the refcount going back up again at
3269 * this point.
3271 wait_event(close_wq, atomic_read(&s->st_stid.sc_count) == 2);
3273 release_all_access(s);
3274 if (s->st_stid.sc_file) {
3275 put_nfs4_file(s->st_stid.sc_file);
3276 s->st_stid.sc_file = NULL;
3279 spin_lock(&nn->client_lock);
3280 last = oo->oo_last_closed_stid;
3281 oo->oo_last_closed_stid = s;
3282 list_move_tail(&oo->oo_close_lru, &nn->close_lru);
3283 oo->oo_time = get_seconds();
3284 spin_unlock(&nn->client_lock);
3285 if (last)
3286 nfs4_put_stid(&last->st_stid);
3289 /* search file_hashtbl[] for file */
3290 static struct nfs4_file *
3291 find_file_locked(struct knfsd_fh *fh, unsigned int hashval)
3293 struct nfs4_file *fp;
3295 hlist_for_each_entry_rcu(fp, &file_hashtbl[hashval], fi_hash) {
3296 if (fh_match(&fp->fi_fhandle, fh)) {
3297 if (atomic_inc_not_zero(&fp->fi_ref))
3298 return fp;
3301 return NULL;
3304 struct nfs4_file *
3305 find_file(struct knfsd_fh *fh)
3307 struct nfs4_file *fp;
3308 unsigned int hashval = file_hashval(fh);
3310 rcu_read_lock();
3311 fp = find_file_locked(fh, hashval);
3312 rcu_read_unlock();
3313 return fp;
3316 static struct nfs4_file *
3317 find_or_add_file(struct nfs4_file *new, struct knfsd_fh *fh)
3319 struct nfs4_file *fp;
3320 unsigned int hashval = file_hashval(fh);
3322 rcu_read_lock();
3323 fp = find_file_locked(fh, hashval);
3324 rcu_read_unlock();
3325 if (fp)
3326 return fp;
3328 spin_lock(&state_lock);
3329 fp = find_file_locked(fh, hashval);
3330 if (likely(fp == NULL)) {
3331 nfsd4_init_file(fh, hashval, new);
3332 fp = new;
3334 spin_unlock(&state_lock);
3336 return fp;
3340 * Called to check deny when READ with all zero stateid or
3341 * WRITE with all zero or all one stateid
3343 static __be32
3344 nfs4_share_conflict(struct svc_fh *current_fh, unsigned int deny_type)
3346 struct nfs4_file *fp;
3347 __be32 ret = nfs_ok;
3349 fp = find_file(&current_fh->fh_handle);
3350 if (!fp)
3351 return ret;
3352 /* Check for conflicting share reservations */
3353 spin_lock(&fp->fi_lock);
3354 if (fp->fi_share_deny & deny_type)
3355 ret = nfserr_locked;
3356 spin_unlock(&fp->fi_lock);
3357 put_nfs4_file(fp);
3358 return ret;
3361 static void nfsd4_cb_recall_prepare(struct nfsd4_callback *cb)
3363 struct nfs4_delegation *dp = cb_to_delegation(cb);
3364 struct nfsd_net *nn = net_generic(dp->dl_stid.sc_client->net,
3365 nfsd_net_id);
3367 block_delegations(&dp->dl_stid.sc_file->fi_fhandle);
3370 * We can't do this in nfsd_break_deleg_cb because it is
3371 * already holding inode->i_lock.
3373 * If the dl_time != 0, then we know that it has already been
3374 * queued for a lease break. Don't queue it again.
3376 spin_lock(&state_lock);
3377 if (dp->dl_time == 0) {
3378 dp->dl_time = get_seconds();
3379 list_add_tail(&dp->dl_recall_lru, &nn->del_recall_lru);
3381 spin_unlock(&state_lock);
3384 static int nfsd4_cb_recall_done(struct nfsd4_callback *cb,
3385 struct rpc_task *task)
3387 struct nfs4_delegation *dp = cb_to_delegation(cb);
3389 switch (task->tk_status) {
3390 case 0:
3391 return 1;
3392 case -EBADHANDLE:
3393 case -NFS4ERR_BAD_STATEID:
3395 * Race: client probably got cb_recall before open reply
3396 * granting delegation.
3398 if (dp->dl_retries--) {
3399 rpc_delay(task, 2 * HZ);
3400 return 0;
3402 /*FALLTHRU*/
3403 default:
3404 return -1;
3408 static void nfsd4_cb_recall_release(struct nfsd4_callback *cb)
3410 struct nfs4_delegation *dp = cb_to_delegation(cb);
3412 nfs4_put_stid(&dp->dl_stid);
3415 static struct nfsd4_callback_ops nfsd4_cb_recall_ops = {
3416 .prepare = nfsd4_cb_recall_prepare,
3417 .done = nfsd4_cb_recall_done,
3418 .release = nfsd4_cb_recall_release,
3421 static void nfsd_break_one_deleg(struct nfs4_delegation *dp)
3424 * We're assuming the state code never drops its reference
3425 * without first removing the lease. Since we're in this lease
3426 * callback (and since the lease code is serialized by the kernel
3427 * lock) we know the server hasn't removed the lease yet, we know
3428 * it's safe to take a reference.
3430 atomic_inc(&dp->dl_stid.sc_count);
3431 nfsd4_run_cb(&dp->dl_recall);
3434 /* Called from break_lease() with i_lock held. */
3435 static bool
3436 nfsd_break_deleg_cb(struct file_lock *fl)
3438 bool ret = false;
3439 struct nfs4_file *fp = (struct nfs4_file *)fl->fl_owner;
3440 struct nfs4_delegation *dp;
3442 if (!fp) {
3443 WARN(1, "(%p)->fl_owner NULL\n", fl);
3444 return ret;
3446 if (fp->fi_had_conflict) {
3447 WARN(1, "duplicate break on %p\n", fp);
3448 return ret;
3451 * We don't want the locks code to timeout the lease for us;
3452 * we'll remove it ourself if a delegation isn't returned
3453 * in time:
3455 fl->fl_break_time = 0;
3457 spin_lock(&fp->fi_lock);
3458 fp->fi_had_conflict = true;
3460 * If there are no delegations on the list, then return true
3461 * so that the lease code will go ahead and delete it.
3463 if (list_empty(&fp->fi_delegations))
3464 ret = true;
3465 else
3466 list_for_each_entry(dp, &fp->fi_delegations, dl_perfile)
3467 nfsd_break_one_deleg(dp);
3468 spin_unlock(&fp->fi_lock);
3469 return ret;
3472 static int
3473 nfsd_change_deleg_cb(struct file_lock *onlist, int arg,
3474 struct list_head *dispose)
3476 if (arg & F_UNLCK)
3477 return lease_modify(onlist, arg, dispose);
3478 else
3479 return -EAGAIN;
3482 static const struct lock_manager_operations nfsd_lease_mng_ops = {
3483 .lm_break = nfsd_break_deleg_cb,
3484 .lm_change = nfsd_change_deleg_cb,
3487 static __be32 nfsd4_check_seqid(struct nfsd4_compound_state *cstate, struct nfs4_stateowner *so, u32 seqid)
3489 if (nfsd4_has_session(cstate))
3490 return nfs_ok;
3491 if (seqid == so->so_seqid - 1)
3492 return nfserr_replay_me;
3493 if (seqid == so->so_seqid)
3494 return nfs_ok;
3495 return nfserr_bad_seqid;
3498 static __be32 lookup_clientid(clientid_t *clid,
3499 struct nfsd4_compound_state *cstate,
3500 struct nfsd_net *nn)
3502 struct nfs4_client *found;
3504 if (cstate->clp) {
3505 found = cstate->clp;
3506 if (!same_clid(&found->cl_clientid, clid))
3507 return nfserr_stale_clientid;
3508 return nfs_ok;
3511 if (STALE_CLIENTID(clid, nn))
3512 return nfserr_stale_clientid;
3515 * For v4.1+ we get the client in the SEQUENCE op. If we don't have one
3516 * cached already then we know this is for is for v4.0 and "sessions"
3517 * will be false.
3519 WARN_ON_ONCE(cstate->session);
3520 spin_lock(&nn->client_lock);
3521 found = find_confirmed_client(clid, false, nn);
3522 if (!found) {
3523 spin_unlock(&nn->client_lock);
3524 return nfserr_expired;
3526 atomic_inc(&found->cl_refcount);
3527 spin_unlock(&nn->client_lock);
3529 /* Cache the nfs4_client in cstate! */
3530 cstate->clp = found;
3531 return nfs_ok;
3534 __be32
3535 nfsd4_process_open1(struct nfsd4_compound_state *cstate,
3536 struct nfsd4_open *open, struct nfsd_net *nn)
3538 clientid_t *clientid = &open->op_clientid;
3539 struct nfs4_client *clp = NULL;
3540 unsigned int strhashval;
3541 struct nfs4_openowner *oo = NULL;
3542 __be32 status;
3544 if (STALE_CLIENTID(&open->op_clientid, nn))
3545 return nfserr_stale_clientid;
3547 * In case we need it later, after we've already created the
3548 * file and don't want to risk a further failure:
3550 open->op_file = nfsd4_alloc_file();
3551 if (open->op_file == NULL)
3552 return nfserr_jukebox;
3554 status = lookup_clientid(clientid, cstate, nn);
3555 if (status)
3556 return status;
3557 clp = cstate->clp;
3559 strhashval = ownerstr_hashval(&open->op_owner);
3560 oo = find_openstateowner_str(strhashval, open, clp);
3561 open->op_openowner = oo;
3562 if (!oo) {
3563 goto new_owner;
3565 if (!(oo->oo_flags & NFS4_OO_CONFIRMED)) {
3566 /* Replace unconfirmed owners without checking for replay. */
3567 release_openowner(oo);
3568 open->op_openowner = NULL;
3569 goto new_owner;
3571 status = nfsd4_check_seqid(cstate, &oo->oo_owner, open->op_seqid);
3572 if (status)
3573 return status;
3574 goto alloc_stateid;
3575 new_owner:
3576 oo = alloc_init_open_stateowner(strhashval, open, cstate);
3577 if (oo == NULL)
3578 return nfserr_jukebox;
3579 open->op_openowner = oo;
3580 alloc_stateid:
3581 open->op_stp = nfs4_alloc_open_stateid(clp);
3582 if (!open->op_stp)
3583 return nfserr_jukebox;
3584 return nfs_ok;
3587 static inline __be32
3588 nfs4_check_delegmode(struct nfs4_delegation *dp, int flags)
3590 if ((flags & WR_STATE) && (dp->dl_type == NFS4_OPEN_DELEGATE_READ))
3591 return nfserr_openmode;
3592 else
3593 return nfs_ok;
3596 static int share_access_to_flags(u32 share_access)
3598 return share_access == NFS4_SHARE_ACCESS_READ ? RD_STATE : WR_STATE;
3601 static struct nfs4_delegation *find_deleg_stateid(struct nfs4_client *cl, stateid_t *s)
3603 struct nfs4_stid *ret;
3605 ret = find_stateid_by_type(cl, s, NFS4_DELEG_STID);
3606 if (!ret)
3607 return NULL;
3608 return delegstateid(ret);
3611 static bool nfsd4_is_deleg_cur(struct nfsd4_open *open)
3613 return open->op_claim_type == NFS4_OPEN_CLAIM_DELEGATE_CUR ||
3614 open->op_claim_type == NFS4_OPEN_CLAIM_DELEG_CUR_FH;
3617 static __be32
3618 nfs4_check_deleg(struct nfs4_client *cl, struct nfsd4_open *open,
3619 struct nfs4_delegation **dp)
3621 int flags;
3622 __be32 status = nfserr_bad_stateid;
3623 struct nfs4_delegation *deleg;
3625 deleg = find_deleg_stateid(cl, &open->op_delegate_stateid);
3626 if (deleg == NULL)
3627 goto out;
3628 flags = share_access_to_flags(open->op_share_access);
3629 status = nfs4_check_delegmode(deleg, flags);
3630 if (status) {
3631 nfs4_put_stid(&deleg->dl_stid);
3632 goto out;
3634 *dp = deleg;
3635 out:
3636 if (!nfsd4_is_deleg_cur(open))
3637 return nfs_ok;
3638 if (status)
3639 return status;
3640 open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
3641 return nfs_ok;
3644 static struct nfs4_ol_stateid *
3645 nfsd4_find_existing_open(struct nfs4_file *fp, struct nfsd4_open *open)
3647 struct nfs4_ol_stateid *local, *ret = NULL;
3648 struct nfs4_openowner *oo = open->op_openowner;
3650 spin_lock(&fp->fi_lock);
3651 list_for_each_entry(local, &fp->fi_stateids, st_perfile) {
3652 /* ignore lock owners */
3653 if (local->st_stateowner->so_is_open_owner == 0)
3654 continue;
3655 if (local->st_stateowner == &oo->oo_owner) {
3656 ret = local;
3657 atomic_inc(&ret->st_stid.sc_count);
3658 break;
3661 spin_unlock(&fp->fi_lock);
3662 return ret;
3665 static inline int nfs4_access_to_access(u32 nfs4_access)
3667 int flags = 0;
3669 if (nfs4_access & NFS4_SHARE_ACCESS_READ)
3670 flags |= NFSD_MAY_READ;
3671 if (nfs4_access & NFS4_SHARE_ACCESS_WRITE)
3672 flags |= NFSD_MAY_WRITE;
3673 return flags;
3676 static inline __be32
3677 nfsd4_truncate(struct svc_rqst *rqstp, struct svc_fh *fh,
3678 struct nfsd4_open *open)
3680 struct iattr iattr = {
3681 .ia_valid = ATTR_SIZE,
3682 .ia_size = 0,
3684 if (!open->op_truncate)
3685 return 0;
3686 if (!(open->op_share_access & NFS4_SHARE_ACCESS_WRITE))
3687 return nfserr_inval;
3688 return nfsd_setattr(rqstp, fh, &iattr, 0, (time_t)0);
3691 static __be32 nfs4_get_vfs_file(struct svc_rqst *rqstp, struct nfs4_file *fp,
3692 struct svc_fh *cur_fh, struct nfs4_ol_stateid *stp,
3693 struct nfsd4_open *open)
3695 struct file *filp = NULL;
3696 __be32 status;
3697 int oflag = nfs4_access_to_omode(open->op_share_access);
3698 int access = nfs4_access_to_access(open->op_share_access);
3699 unsigned char old_access_bmap, old_deny_bmap;
3701 spin_lock(&fp->fi_lock);
3704 * Are we trying to set a deny mode that would conflict with
3705 * current access?
3707 status = nfs4_file_check_deny(fp, open->op_share_deny);
3708 if (status != nfs_ok) {
3709 spin_unlock(&fp->fi_lock);
3710 goto out;
3713 /* set access to the file */
3714 status = nfs4_file_get_access(fp, open->op_share_access);
3715 if (status != nfs_ok) {
3716 spin_unlock(&fp->fi_lock);
3717 goto out;
3720 /* Set access bits in stateid */
3721 old_access_bmap = stp->st_access_bmap;
3722 set_access(open->op_share_access, stp);
3724 /* Set new deny mask */
3725 old_deny_bmap = stp->st_deny_bmap;
3726 set_deny(open->op_share_deny, stp);
3727 fp->fi_share_deny |= (open->op_share_deny & NFS4_SHARE_DENY_BOTH);
3729 if (!fp->fi_fds[oflag]) {
3730 spin_unlock(&fp->fi_lock);
3731 status = nfsd_open(rqstp, cur_fh, S_IFREG, access, &filp);
3732 if (status)
3733 goto out_put_access;
3734 spin_lock(&fp->fi_lock);
3735 if (!fp->fi_fds[oflag]) {
3736 fp->fi_fds[oflag] = filp;
3737 filp = NULL;
3740 spin_unlock(&fp->fi_lock);
3741 if (filp)
3742 fput(filp);
3744 status = nfsd4_truncate(rqstp, cur_fh, open);
3745 if (status)
3746 goto out_put_access;
3747 out:
3748 return status;
3749 out_put_access:
3750 stp->st_access_bmap = old_access_bmap;
3751 nfs4_file_put_access(fp, open->op_share_access);
3752 reset_union_bmap_deny(bmap_to_share_mode(old_deny_bmap), stp);
3753 goto out;
3756 static __be32
3757 nfs4_upgrade_open(struct svc_rqst *rqstp, struct nfs4_file *fp, struct svc_fh *cur_fh, struct nfs4_ol_stateid *stp, struct nfsd4_open *open)
3759 __be32 status;
3760 unsigned char old_deny_bmap;
3762 if (!test_access(open->op_share_access, stp))
3763 return nfs4_get_vfs_file(rqstp, fp, cur_fh, stp, open);
3765 /* test and set deny mode */
3766 spin_lock(&fp->fi_lock);
3767 status = nfs4_file_check_deny(fp, open->op_share_deny);
3768 if (status == nfs_ok) {
3769 old_deny_bmap = stp->st_deny_bmap;
3770 set_deny(open->op_share_deny, stp);
3771 fp->fi_share_deny |=
3772 (open->op_share_deny & NFS4_SHARE_DENY_BOTH);
3774 spin_unlock(&fp->fi_lock);
3776 if (status != nfs_ok)
3777 return status;
3779 status = nfsd4_truncate(rqstp, cur_fh, open);
3780 if (status != nfs_ok)
3781 reset_union_bmap_deny(old_deny_bmap, stp);
3782 return status;
3785 static void
3786 nfs4_set_claim_prev(struct nfsd4_open *open, bool has_session)
3788 open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
3791 /* Should we give out recallable state?: */
3792 static bool nfsd4_cb_channel_good(struct nfs4_client *clp)
3794 if (clp->cl_cb_state == NFSD4_CB_UP)
3795 return true;
3797 * In the sessions case, since we don't have to establish a
3798 * separate connection for callbacks, we assume it's OK
3799 * until we hear otherwise:
3801 return clp->cl_minorversion && clp->cl_cb_state == NFSD4_CB_UNKNOWN;
3804 static struct file_lock *nfs4_alloc_init_lease(struct nfs4_file *fp, int flag)
3806 struct file_lock *fl;
3808 fl = locks_alloc_lock();
3809 if (!fl)
3810 return NULL;
3811 fl->fl_lmops = &nfsd_lease_mng_ops;
3812 fl->fl_flags = FL_DELEG;
3813 fl->fl_type = flag == NFS4_OPEN_DELEGATE_READ? F_RDLCK: F_WRLCK;
3814 fl->fl_end = OFFSET_MAX;
3815 fl->fl_owner = (fl_owner_t)fp;
3816 fl->fl_pid = current->tgid;
3817 return fl;
3820 static int nfs4_setlease(struct nfs4_delegation *dp)
3822 struct nfs4_file *fp = dp->dl_stid.sc_file;
3823 struct file_lock *fl, *ret;
3824 struct file *filp;
3825 int status = 0;
3827 fl = nfs4_alloc_init_lease(fp, NFS4_OPEN_DELEGATE_READ);
3828 if (!fl)
3829 return -ENOMEM;
3830 filp = find_readable_file(fp);
3831 if (!filp) {
3832 /* We should always have a readable file here */
3833 WARN_ON_ONCE(1);
3834 return -EBADF;
3836 fl->fl_file = filp;
3837 ret = fl;
3838 status = vfs_setlease(filp, fl->fl_type, &fl, NULL);
3839 if (fl)
3840 locks_free_lock(fl);
3841 if (status)
3842 goto out_fput;
3843 spin_lock(&state_lock);
3844 spin_lock(&fp->fi_lock);
3845 /* Did the lease get broken before we took the lock? */
3846 status = -EAGAIN;
3847 if (fp->fi_had_conflict)
3848 goto out_unlock;
3849 /* Race breaker */
3850 if (fp->fi_deleg_file) {
3851 status = 0;
3852 ++fp->fi_delegees;
3853 hash_delegation_locked(dp, fp);
3854 goto out_unlock;
3856 fp->fi_deleg_file = filp;
3857 fp->fi_delegees = 1;
3858 hash_delegation_locked(dp, fp);
3859 spin_unlock(&fp->fi_lock);
3860 spin_unlock(&state_lock);
3861 return 0;
3862 out_unlock:
3863 spin_unlock(&fp->fi_lock);
3864 spin_unlock(&state_lock);
3865 out_fput:
3866 fput(filp);
3867 return status;
3870 static struct nfs4_delegation *
3871 nfs4_set_delegation(struct nfs4_client *clp, struct svc_fh *fh,
3872 struct nfs4_file *fp)
3874 int status;
3875 struct nfs4_delegation *dp;
3877 if (fp->fi_had_conflict)
3878 return ERR_PTR(-EAGAIN);
3880 dp = alloc_init_deleg(clp, fh);
3881 if (!dp)
3882 return ERR_PTR(-ENOMEM);
3884 get_nfs4_file(fp);
3885 spin_lock(&state_lock);
3886 spin_lock(&fp->fi_lock);
3887 dp->dl_stid.sc_file = fp;
3888 if (!fp->fi_deleg_file) {
3889 spin_unlock(&fp->fi_lock);
3890 spin_unlock(&state_lock);
3891 status = nfs4_setlease(dp);
3892 goto out;
3894 if (fp->fi_had_conflict) {
3895 status = -EAGAIN;
3896 goto out_unlock;
3898 ++fp->fi_delegees;
3899 hash_delegation_locked(dp, fp);
3900 status = 0;
3901 out_unlock:
3902 spin_unlock(&fp->fi_lock);
3903 spin_unlock(&state_lock);
3904 out:
3905 if (status) {
3906 nfs4_put_stid(&dp->dl_stid);
3907 return ERR_PTR(status);
3909 return dp;
3912 static void nfsd4_open_deleg_none_ext(struct nfsd4_open *open, int status)
3914 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
3915 if (status == -EAGAIN)
3916 open->op_why_no_deleg = WND4_CONTENTION;
3917 else {
3918 open->op_why_no_deleg = WND4_RESOURCE;
3919 switch (open->op_deleg_want) {
3920 case NFS4_SHARE_WANT_READ_DELEG:
3921 case NFS4_SHARE_WANT_WRITE_DELEG:
3922 case NFS4_SHARE_WANT_ANY_DELEG:
3923 break;
3924 case NFS4_SHARE_WANT_CANCEL:
3925 open->op_why_no_deleg = WND4_CANCELLED;
3926 break;
3927 case NFS4_SHARE_WANT_NO_DELEG:
3928 WARN_ON_ONCE(1);
3934 * Attempt to hand out a delegation.
3936 * Note we don't support write delegations, and won't until the vfs has
3937 * proper support for them.
3939 static void
3940 nfs4_open_delegation(struct svc_fh *fh, struct nfsd4_open *open,
3941 struct nfs4_ol_stateid *stp)
3943 struct nfs4_delegation *dp;
3944 struct nfs4_openowner *oo = openowner(stp->st_stateowner);
3945 struct nfs4_client *clp = stp->st_stid.sc_client;
3946 int cb_up;
3947 int status = 0;
3949 cb_up = nfsd4_cb_channel_good(oo->oo_owner.so_client);
3950 open->op_recall = 0;
3951 switch (open->op_claim_type) {
3952 case NFS4_OPEN_CLAIM_PREVIOUS:
3953 if (!cb_up)
3954 open->op_recall = 1;
3955 if (open->op_delegate_type != NFS4_OPEN_DELEGATE_READ)
3956 goto out_no_deleg;
3957 break;
3958 case NFS4_OPEN_CLAIM_NULL:
3959 case NFS4_OPEN_CLAIM_FH:
3961 * Let's not give out any delegations till everyone's
3962 * had the chance to reclaim theirs....
3964 if (locks_in_grace(clp->net))
3965 goto out_no_deleg;
3966 if (!cb_up || !(oo->oo_flags & NFS4_OO_CONFIRMED))
3967 goto out_no_deleg;
3969 * Also, if the file was opened for write or
3970 * create, there's a good chance the client's
3971 * about to write to it, resulting in an
3972 * immediate recall (since we don't support
3973 * write delegations):
3975 if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
3976 goto out_no_deleg;
3977 if (open->op_create == NFS4_OPEN_CREATE)
3978 goto out_no_deleg;
3979 break;
3980 default:
3981 goto out_no_deleg;
3983 dp = nfs4_set_delegation(clp, fh, stp->st_stid.sc_file);
3984 if (IS_ERR(dp))
3985 goto out_no_deleg;
3987 memcpy(&open->op_delegate_stateid, &dp->dl_stid.sc_stateid, sizeof(dp->dl_stid.sc_stateid));
3989 dprintk("NFSD: delegation stateid=" STATEID_FMT "\n",
3990 STATEID_VAL(&dp->dl_stid.sc_stateid));
3991 open->op_delegate_type = NFS4_OPEN_DELEGATE_READ;
3992 nfs4_put_stid(&dp->dl_stid);
3993 return;
3994 out_no_deleg:
3995 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE;
3996 if (open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS &&
3997 open->op_delegate_type != NFS4_OPEN_DELEGATE_NONE) {
3998 dprintk("NFSD: WARNING: refusing delegation reclaim\n");
3999 open->op_recall = 1;
4002 /* 4.1 client asking for a delegation? */
4003 if (open->op_deleg_want)
4004 nfsd4_open_deleg_none_ext(open, status);
4005 return;
4008 static void nfsd4_deleg_xgrade_none_ext(struct nfsd4_open *open,
4009 struct nfs4_delegation *dp)
4011 if (open->op_deleg_want == NFS4_SHARE_WANT_READ_DELEG &&
4012 dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
4013 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
4014 open->op_why_no_deleg = WND4_NOT_SUPP_DOWNGRADE;
4015 } else if (open->op_deleg_want == NFS4_SHARE_WANT_WRITE_DELEG &&
4016 dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
4017 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
4018 open->op_why_no_deleg = WND4_NOT_SUPP_UPGRADE;
4020 /* Otherwise the client must be confused wanting a delegation
4021 * it already has, therefore we don't return
4022 * NFS4_OPEN_DELEGATE_NONE_EXT and reason.
4026 __be32
4027 nfsd4_process_open2(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_open *open)
4029 struct nfsd4_compoundres *resp = rqstp->rq_resp;
4030 struct nfs4_client *cl = open->op_openowner->oo_owner.so_client;
4031 struct nfs4_file *fp = NULL;
4032 struct nfs4_ol_stateid *stp = NULL;
4033 struct nfs4_delegation *dp = NULL;
4034 __be32 status;
4037 * Lookup file; if found, lookup stateid and check open request,
4038 * and check for delegations in the process of being recalled.
4039 * If not found, create the nfs4_file struct
4041 fp = find_or_add_file(open->op_file, &current_fh->fh_handle);
4042 if (fp != open->op_file) {
4043 status = nfs4_check_deleg(cl, open, &dp);
4044 if (status)
4045 goto out;
4046 stp = nfsd4_find_existing_open(fp, open);
4047 } else {
4048 open->op_file = NULL;
4049 status = nfserr_bad_stateid;
4050 if (nfsd4_is_deleg_cur(open))
4051 goto out;
4055 * OPEN the file, or upgrade an existing OPEN.
4056 * If truncate fails, the OPEN fails.
4058 if (stp) {
4059 /* Stateid was found, this is an OPEN upgrade */
4060 status = nfs4_upgrade_open(rqstp, fp, current_fh, stp, open);
4061 if (status)
4062 goto out;
4063 } else {
4064 stp = open->op_stp;
4065 open->op_stp = NULL;
4066 init_open_stateid(stp, fp, open);
4067 status = nfs4_get_vfs_file(rqstp, fp, current_fh, stp, open);
4068 if (status) {
4069 release_open_stateid(stp);
4070 goto out;
4073 update_stateid(&stp->st_stid.sc_stateid);
4074 memcpy(&open->op_stateid, &stp->st_stid.sc_stateid, sizeof(stateid_t));
4076 if (nfsd4_has_session(&resp->cstate)) {
4077 if (open->op_deleg_want & NFS4_SHARE_WANT_NO_DELEG) {
4078 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
4079 open->op_why_no_deleg = WND4_NOT_WANTED;
4080 goto nodeleg;
4085 * Attempt to hand out a delegation. No error return, because the
4086 * OPEN succeeds even if we fail.
4088 nfs4_open_delegation(current_fh, open, stp);
4089 nodeleg:
4090 status = nfs_ok;
4092 dprintk("%s: stateid=" STATEID_FMT "\n", __func__,
4093 STATEID_VAL(&stp->st_stid.sc_stateid));
4094 out:
4095 /* 4.1 client trying to upgrade/downgrade delegation? */
4096 if (open->op_delegate_type == NFS4_OPEN_DELEGATE_NONE && dp &&
4097 open->op_deleg_want)
4098 nfsd4_deleg_xgrade_none_ext(open, dp);
4100 if (fp)
4101 put_nfs4_file(fp);
4102 if (status == 0 && open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS)
4103 nfs4_set_claim_prev(open, nfsd4_has_session(&resp->cstate));
4105 * To finish the open response, we just need to set the rflags.
4107 open->op_rflags = NFS4_OPEN_RESULT_LOCKTYPE_POSIX;
4108 if (!(open->op_openowner->oo_flags & NFS4_OO_CONFIRMED) &&
4109 !nfsd4_has_session(&resp->cstate))
4110 open->op_rflags |= NFS4_OPEN_RESULT_CONFIRM;
4111 if (dp)
4112 nfs4_put_stid(&dp->dl_stid);
4113 if (stp)
4114 nfs4_put_stid(&stp->st_stid);
4116 return status;
4119 void nfsd4_cleanup_open_state(struct nfsd4_compound_state *cstate,
4120 struct nfsd4_open *open)
4122 if (open->op_openowner) {
4123 struct nfs4_stateowner *so = &open->op_openowner->oo_owner;
4125 nfsd4_cstate_assign_replay(cstate, so);
4126 nfs4_put_stateowner(so);
4128 if (open->op_file)
4129 kmem_cache_free(file_slab, open->op_file);
4130 if (open->op_stp)
4131 nfs4_put_stid(&open->op_stp->st_stid);
4134 __be32
4135 nfsd4_renew(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4136 clientid_t *clid)
4138 struct nfs4_client *clp;
4139 __be32 status;
4140 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
4142 dprintk("process_renew(%08x/%08x): starting\n",
4143 clid->cl_boot, clid->cl_id);
4144 status = lookup_clientid(clid, cstate, nn);
4145 if (status)
4146 goto out;
4147 clp = cstate->clp;
4148 status = nfserr_cb_path_down;
4149 if (!list_empty(&clp->cl_delegations)
4150 && clp->cl_cb_state != NFSD4_CB_UP)
4151 goto out;
4152 status = nfs_ok;
4153 out:
4154 return status;
4157 void
4158 nfsd4_end_grace(struct nfsd_net *nn)
4160 /* do nothing if grace period already ended */
4161 if (nn->grace_ended)
4162 return;
4164 dprintk("NFSD: end of grace period\n");
4165 nn->grace_ended = true;
4167 * If the server goes down again right now, an NFSv4
4168 * client will still be allowed to reclaim after it comes back up,
4169 * even if it hasn't yet had a chance to reclaim state this time.
4172 nfsd4_record_grace_done(nn);
4174 * At this point, NFSv4 clients can still reclaim. But if the
4175 * server crashes, any that have not yet reclaimed will be out
4176 * of luck on the next boot.
4178 * (NFSv4.1+ clients are considered to have reclaimed once they
4179 * call RECLAIM_COMPLETE. NFSv4.0 clients are considered to
4180 * have reclaimed after their first OPEN.)
4182 locks_end_grace(&nn->nfsd4_manager);
4184 * At this point, and once lockd and/or any other containers
4185 * exit their grace period, further reclaims will fail and
4186 * regular locking can resume.
4190 static time_t
4191 nfs4_laundromat(struct nfsd_net *nn)
4193 struct nfs4_client *clp;
4194 struct nfs4_openowner *oo;
4195 struct nfs4_delegation *dp;
4196 struct nfs4_ol_stateid *stp;
4197 struct list_head *pos, *next, reaplist;
4198 time_t cutoff = get_seconds() - nn->nfsd4_lease;
4199 time_t t, new_timeo = nn->nfsd4_lease;
4201 dprintk("NFSD: laundromat service - starting\n");
4202 nfsd4_end_grace(nn);
4203 INIT_LIST_HEAD(&reaplist);
4204 spin_lock(&nn->client_lock);
4205 list_for_each_safe(pos, next, &nn->client_lru) {
4206 clp = list_entry(pos, struct nfs4_client, cl_lru);
4207 if (time_after((unsigned long)clp->cl_time, (unsigned long)cutoff)) {
4208 t = clp->cl_time - cutoff;
4209 new_timeo = min(new_timeo, t);
4210 break;
4212 if (mark_client_expired_locked(clp)) {
4213 dprintk("NFSD: client in use (clientid %08x)\n",
4214 clp->cl_clientid.cl_id);
4215 continue;
4217 list_add(&clp->cl_lru, &reaplist);
4219 spin_unlock(&nn->client_lock);
4220 list_for_each_safe(pos, next, &reaplist) {
4221 clp = list_entry(pos, struct nfs4_client, cl_lru);
4222 dprintk("NFSD: purging unused client (clientid %08x)\n",
4223 clp->cl_clientid.cl_id);
4224 list_del_init(&clp->cl_lru);
4225 expire_client(clp);
4227 spin_lock(&state_lock);
4228 list_for_each_safe(pos, next, &nn->del_recall_lru) {
4229 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
4230 if (net_generic(dp->dl_stid.sc_client->net, nfsd_net_id) != nn)
4231 continue;
4232 if (time_after((unsigned long)dp->dl_time, (unsigned long)cutoff)) {
4233 t = dp->dl_time - cutoff;
4234 new_timeo = min(new_timeo, t);
4235 break;
4237 unhash_delegation_locked(dp);
4238 list_add(&dp->dl_recall_lru, &reaplist);
4240 spin_unlock(&state_lock);
4241 while (!list_empty(&reaplist)) {
4242 dp = list_first_entry(&reaplist, struct nfs4_delegation,
4243 dl_recall_lru);
4244 list_del_init(&dp->dl_recall_lru);
4245 revoke_delegation(dp);
4248 spin_lock(&nn->client_lock);
4249 while (!list_empty(&nn->close_lru)) {
4250 oo = list_first_entry(&nn->close_lru, struct nfs4_openowner,
4251 oo_close_lru);
4252 if (time_after((unsigned long)oo->oo_time,
4253 (unsigned long)cutoff)) {
4254 t = oo->oo_time - cutoff;
4255 new_timeo = min(new_timeo, t);
4256 break;
4258 list_del_init(&oo->oo_close_lru);
4259 stp = oo->oo_last_closed_stid;
4260 oo->oo_last_closed_stid = NULL;
4261 spin_unlock(&nn->client_lock);
4262 nfs4_put_stid(&stp->st_stid);
4263 spin_lock(&nn->client_lock);
4265 spin_unlock(&nn->client_lock);
4267 new_timeo = max_t(time_t, new_timeo, NFSD_LAUNDROMAT_MINTIMEOUT);
4268 return new_timeo;
4271 static struct workqueue_struct *laundry_wq;
4272 static void laundromat_main(struct work_struct *);
4274 static void
4275 laundromat_main(struct work_struct *laundry)
4277 time_t t;
4278 struct delayed_work *dwork = container_of(laundry, struct delayed_work,
4279 work);
4280 struct nfsd_net *nn = container_of(dwork, struct nfsd_net,
4281 laundromat_work);
4283 t = nfs4_laundromat(nn);
4284 dprintk("NFSD: laundromat_main - sleeping for %ld seconds\n", t);
4285 queue_delayed_work(laundry_wq, &nn->laundromat_work, t*HZ);
4288 static inline __be32 nfs4_check_fh(struct svc_fh *fhp, struct nfs4_ol_stateid *stp)
4290 if (!fh_match(&fhp->fh_handle, &stp->st_stid.sc_file->fi_fhandle))
4291 return nfserr_bad_stateid;
4292 return nfs_ok;
4295 static inline int
4296 access_permit_read(struct nfs4_ol_stateid *stp)
4298 return test_access(NFS4_SHARE_ACCESS_READ, stp) ||
4299 test_access(NFS4_SHARE_ACCESS_BOTH, stp) ||
4300 test_access(NFS4_SHARE_ACCESS_WRITE, stp);
4303 static inline int
4304 access_permit_write(struct nfs4_ol_stateid *stp)
4306 return test_access(NFS4_SHARE_ACCESS_WRITE, stp) ||
4307 test_access(NFS4_SHARE_ACCESS_BOTH, stp);
4310 static
4311 __be32 nfs4_check_openmode(struct nfs4_ol_stateid *stp, int flags)
4313 __be32 status = nfserr_openmode;
4315 /* For lock stateid's, we test the parent open, not the lock: */
4316 if (stp->st_openstp)
4317 stp = stp->st_openstp;
4318 if ((flags & WR_STATE) && !access_permit_write(stp))
4319 goto out;
4320 if ((flags & RD_STATE) && !access_permit_read(stp))
4321 goto out;
4322 status = nfs_ok;
4323 out:
4324 return status;
4327 static inline __be32
4328 check_special_stateids(struct net *net, svc_fh *current_fh, stateid_t *stateid, int flags)
4330 if (ONE_STATEID(stateid) && (flags & RD_STATE))
4331 return nfs_ok;
4332 else if (locks_in_grace(net)) {
4333 /* Answer in remaining cases depends on existence of
4334 * conflicting state; so we must wait out the grace period. */
4335 return nfserr_grace;
4336 } else if (flags & WR_STATE)
4337 return nfs4_share_conflict(current_fh,
4338 NFS4_SHARE_DENY_WRITE);
4339 else /* (flags & RD_STATE) && ZERO_STATEID(stateid) */
4340 return nfs4_share_conflict(current_fh,
4341 NFS4_SHARE_DENY_READ);
4345 * Allow READ/WRITE during grace period on recovered state only for files
4346 * that are not able to provide mandatory locking.
4348 static inline int
4349 grace_disallows_io(struct net *net, struct inode *inode)
4351 return locks_in_grace(net) && mandatory_lock(inode);
4354 /* Returns true iff a is later than b: */
4355 static bool stateid_generation_after(stateid_t *a, stateid_t *b)
4357 return (s32)(a->si_generation - b->si_generation) > 0;
4360 static __be32 check_stateid_generation(stateid_t *in, stateid_t *ref, bool has_session)
4363 * When sessions are used the stateid generation number is ignored
4364 * when it is zero.
4366 if (has_session && in->si_generation == 0)
4367 return nfs_ok;
4369 if (in->si_generation == ref->si_generation)
4370 return nfs_ok;
4372 /* If the client sends us a stateid from the future, it's buggy: */
4373 if (stateid_generation_after(in, ref))
4374 return nfserr_bad_stateid;
4376 * However, we could see a stateid from the past, even from a
4377 * non-buggy client. For example, if the client sends a lock
4378 * while some IO is outstanding, the lock may bump si_generation
4379 * while the IO is still in flight. The client could avoid that
4380 * situation by waiting for responses on all the IO requests,
4381 * but better performance may result in retrying IO that
4382 * receives an old_stateid error if requests are rarely
4383 * reordered in flight:
4385 return nfserr_old_stateid;
4388 static __be32 nfsd4_validate_stateid(struct nfs4_client *cl, stateid_t *stateid)
4390 struct nfs4_stid *s;
4391 struct nfs4_ol_stateid *ols;
4392 __be32 status = nfserr_bad_stateid;
4394 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
4395 return status;
4396 /* Client debugging aid. */
4397 if (!same_clid(&stateid->si_opaque.so_clid, &cl->cl_clientid)) {
4398 char addr_str[INET6_ADDRSTRLEN];
4399 rpc_ntop((struct sockaddr *)&cl->cl_addr, addr_str,
4400 sizeof(addr_str));
4401 pr_warn_ratelimited("NFSD: client %s testing state ID "
4402 "with incorrect client ID\n", addr_str);
4403 return status;
4405 spin_lock(&cl->cl_lock);
4406 s = find_stateid_locked(cl, stateid);
4407 if (!s)
4408 goto out_unlock;
4409 status = check_stateid_generation(stateid, &s->sc_stateid, 1);
4410 if (status)
4411 goto out_unlock;
4412 switch (s->sc_type) {
4413 case NFS4_DELEG_STID:
4414 status = nfs_ok;
4415 break;
4416 case NFS4_REVOKED_DELEG_STID:
4417 status = nfserr_deleg_revoked;
4418 break;
4419 case NFS4_OPEN_STID:
4420 case NFS4_LOCK_STID:
4421 ols = openlockstateid(s);
4422 if (ols->st_stateowner->so_is_open_owner
4423 && !(openowner(ols->st_stateowner)->oo_flags
4424 & NFS4_OO_CONFIRMED))
4425 status = nfserr_bad_stateid;
4426 else
4427 status = nfs_ok;
4428 break;
4429 default:
4430 printk("unknown stateid type %x\n", s->sc_type);
4431 /* Fallthrough */
4432 case NFS4_CLOSED_STID:
4433 case NFS4_CLOSED_DELEG_STID:
4434 status = nfserr_bad_stateid;
4436 out_unlock:
4437 spin_unlock(&cl->cl_lock);
4438 return status;
4441 __be32
4442 nfsd4_lookup_stateid(struct nfsd4_compound_state *cstate,
4443 stateid_t *stateid, unsigned char typemask,
4444 struct nfs4_stid **s, struct nfsd_net *nn)
4446 __be32 status;
4448 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
4449 return nfserr_bad_stateid;
4450 status = lookup_clientid(&stateid->si_opaque.so_clid, cstate, nn);
4451 if (status == nfserr_stale_clientid) {
4452 if (cstate->session)
4453 return nfserr_bad_stateid;
4454 return nfserr_stale_stateid;
4456 if (status)
4457 return status;
4458 *s = find_stateid_by_type(cstate->clp, stateid, typemask);
4459 if (!*s)
4460 return nfserr_bad_stateid;
4461 return nfs_ok;
4465 * Checks for stateid operations
4467 __be32
4468 nfs4_preprocess_stateid_op(struct net *net, struct nfsd4_compound_state *cstate,
4469 stateid_t *stateid, int flags, struct file **filpp)
4471 struct nfs4_stid *s;
4472 struct nfs4_ol_stateid *stp = NULL;
4473 struct nfs4_delegation *dp = NULL;
4474 struct svc_fh *current_fh = &cstate->current_fh;
4475 struct inode *ino = d_inode(current_fh->fh_dentry);
4476 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
4477 struct file *file = NULL;
4478 __be32 status;
4480 if (filpp)
4481 *filpp = NULL;
4483 if (grace_disallows_io(net, ino))
4484 return nfserr_grace;
4486 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
4487 return check_special_stateids(net, current_fh, stateid, flags);
4489 status = nfsd4_lookup_stateid(cstate, stateid,
4490 NFS4_DELEG_STID|NFS4_OPEN_STID|NFS4_LOCK_STID,
4491 &s, nn);
4492 if (status)
4493 return status;
4494 status = check_stateid_generation(stateid, &s->sc_stateid, nfsd4_has_session(cstate));
4495 if (status)
4496 goto out;
4497 switch (s->sc_type) {
4498 case NFS4_DELEG_STID:
4499 dp = delegstateid(s);
4500 status = nfs4_check_delegmode(dp, flags);
4501 if (status)
4502 goto out;
4503 if (filpp) {
4504 file = dp->dl_stid.sc_file->fi_deleg_file;
4505 if (!file) {
4506 WARN_ON_ONCE(1);
4507 status = nfserr_serverfault;
4508 goto out;
4510 get_file(file);
4512 break;
4513 case NFS4_OPEN_STID:
4514 case NFS4_LOCK_STID:
4515 stp = openlockstateid(s);
4516 status = nfs4_check_fh(current_fh, stp);
4517 if (status)
4518 goto out;
4519 if (stp->st_stateowner->so_is_open_owner
4520 && !(openowner(stp->st_stateowner)->oo_flags & NFS4_OO_CONFIRMED))
4521 goto out;
4522 status = nfs4_check_openmode(stp, flags);
4523 if (status)
4524 goto out;
4525 if (filpp) {
4526 struct nfs4_file *fp = stp->st_stid.sc_file;
4528 if (flags & RD_STATE)
4529 file = find_readable_file(fp);
4530 else
4531 file = find_writeable_file(fp);
4533 break;
4534 default:
4535 status = nfserr_bad_stateid;
4536 goto out;
4538 status = nfs_ok;
4539 if (file)
4540 *filpp = file;
4541 out:
4542 nfs4_put_stid(s);
4543 return status;
4547 * Test if the stateid is valid
4549 __be32
4550 nfsd4_test_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4551 struct nfsd4_test_stateid *test_stateid)
4553 struct nfsd4_test_stateid_id *stateid;
4554 struct nfs4_client *cl = cstate->session->se_client;
4556 list_for_each_entry(stateid, &test_stateid->ts_stateid_list, ts_id_list)
4557 stateid->ts_id_status =
4558 nfsd4_validate_stateid(cl, &stateid->ts_id_stateid);
4560 return nfs_ok;
4563 __be32
4564 nfsd4_free_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4565 struct nfsd4_free_stateid *free_stateid)
4567 stateid_t *stateid = &free_stateid->fr_stateid;
4568 struct nfs4_stid *s;
4569 struct nfs4_delegation *dp;
4570 struct nfs4_ol_stateid *stp;
4571 struct nfs4_client *cl = cstate->session->se_client;
4572 __be32 ret = nfserr_bad_stateid;
4574 spin_lock(&cl->cl_lock);
4575 s = find_stateid_locked(cl, stateid);
4576 if (!s)
4577 goto out_unlock;
4578 switch (s->sc_type) {
4579 case NFS4_DELEG_STID:
4580 ret = nfserr_locks_held;
4581 break;
4582 case NFS4_OPEN_STID:
4583 ret = check_stateid_generation(stateid, &s->sc_stateid, 1);
4584 if (ret)
4585 break;
4586 ret = nfserr_locks_held;
4587 break;
4588 case NFS4_LOCK_STID:
4589 ret = check_stateid_generation(stateid, &s->sc_stateid, 1);
4590 if (ret)
4591 break;
4592 stp = openlockstateid(s);
4593 ret = nfserr_locks_held;
4594 if (check_for_locks(stp->st_stid.sc_file,
4595 lockowner(stp->st_stateowner)))
4596 break;
4597 unhash_lock_stateid(stp);
4598 spin_unlock(&cl->cl_lock);
4599 nfs4_put_stid(s);
4600 ret = nfs_ok;
4601 goto out;
4602 case NFS4_REVOKED_DELEG_STID:
4603 dp = delegstateid(s);
4604 list_del_init(&dp->dl_recall_lru);
4605 spin_unlock(&cl->cl_lock);
4606 nfs4_put_stid(s);
4607 ret = nfs_ok;
4608 goto out;
4609 /* Default falls through and returns nfserr_bad_stateid */
4611 out_unlock:
4612 spin_unlock(&cl->cl_lock);
4613 out:
4614 return ret;
4617 static inline int
4618 setlkflg (int type)
4620 return (type == NFS4_READW_LT || type == NFS4_READ_LT) ?
4621 RD_STATE : WR_STATE;
4624 static __be32 nfs4_seqid_op_checks(struct nfsd4_compound_state *cstate, stateid_t *stateid, u32 seqid, struct nfs4_ol_stateid *stp)
4626 struct svc_fh *current_fh = &cstate->current_fh;
4627 struct nfs4_stateowner *sop = stp->st_stateowner;
4628 __be32 status;
4630 status = nfsd4_check_seqid(cstate, sop, seqid);
4631 if (status)
4632 return status;
4633 if (stp->st_stid.sc_type == NFS4_CLOSED_STID
4634 || stp->st_stid.sc_type == NFS4_REVOKED_DELEG_STID)
4636 * "Closed" stateid's exist *only* to return
4637 * nfserr_replay_me from the previous step, and
4638 * revoked delegations are kept only for free_stateid.
4640 return nfserr_bad_stateid;
4641 status = check_stateid_generation(stateid, &stp->st_stid.sc_stateid, nfsd4_has_session(cstate));
4642 if (status)
4643 return status;
4644 return nfs4_check_fh(current_fh, stp);
4648 * Checks for sequence id mutating operations.
4650 static __be32
4651 nfs4_preprocess_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
4652 stateid_t *stateid, char typemask,
4653 struct nfs4_ol_stateid **stpp,
4654 struct nfsd_net *nn)
4656 __be32 status;
4657 struct nfs4_stid *s;
4658 struct nfs4_ol_stateid *stp = NULL;
4660 dprintk("NFSD: %s: seqid=%d stateid = " STATEID_FMT "\n", __func__,
4661 seqid, STATEID_VAL(stateid));
4663 *stpp = NULL;
4664 status = nfsd4_lookup_stateid(cstate, stateid, typemask, &s, nn);
4665 if (status)
4666 return status;
4667 stp = openlockstateid(s);
4668 nfsd4_cstate_assign_replay(cstate, stp->st_stateowner);
4670 status = nfs4_seqid_op_checks(cstate, stateid, seqid, stp);
4671 if (!status)
4672 *stpp = stp;
4673 else
4674 nfs4_put_stid(&stp->st_stid);
4675 return status;
4678 static __be32 nfs4_preprocess_confirmed_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
4679 stateid_t *stateid, struct nfs4_ol_stateid **stpp, struct nfsd_net *nn)
4681 __be32 status;
4682 struct nfs4_openowner *oo;
4683 struct nfs4_ol_stateid *stp;
4685 status = nfs4_preprocess_seqid_op(cstate, seqid, stateid,
4686 NFS4_OPEN_STID, &stp, nn);
4687 if (status)
4688 return status;
4689 oo = openowner(stp->st_stateowner);
4690 if (!(oo->oo_flags & NFS4_OO_CONFIRMED)) {
4691 nfs4_put_stid(&stp->st_stid);
4692 return nfserr_bad_stateid;
4694 *stpp = stp;
4695 return nfs_ok;
4698 __be32
4699 nfsd4_open_confirm(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4700 struct nfsd4_open_confirm *oc)
4702 __be32 status;
4703 struct nfs4_openowner *oo;
4704 struct nfs4_ol_stateid *stp;
4705 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
4707 dprintk("NFSD: nfsd4_open_confirm on file %pd\n",
4708 cstate->current_fh.fh_dentry);
4710 status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0);
4711 if (status)
4712 return status;
4714 status = nfs4_preprocess_seqid_op(cstate,
4715 oc->oc_seqid, &oc->oc_req_stateid,
4716 NFS4_OPEN_STID, &stp, nn);
4717 if (status)
4718 goto out;
4719 oo = openowner(stp->st_stateowner);
4720 status = nfserr_bad_stateid;
4721 if (oo->oo_flags & NFS4_OO_CONFIRMED)
4722 goto put_stateid;
4723 oo->oo_flags |= NFS4_OO_CONFIRMED;
4724 update_stateid(&stp->st_stid.sc_stateid);
4725 memcpy(&oc->oc_resp_stateid, &stp->st_stid.sc_stateid, sizeof(stateid_t));
4726 dprintk("NFSD: %s: success, seqid=%d stateid=" STATEID_FMT "\n",
4727 __func__, oc->oc_seqid, STATEID_VAL(&stp->st_stid.sc_stateid));
4729 nfsd4_client_record_create(oo->oo_owner.so_client);
4730 status = nfs_ok;
4731 put_stateid:
4732 nfs4_put_stid(&stp->st_stid);
4733 out:
4734 nfsd4_bump_seqid(cstate, status);
4735 return status;
4738 static inline void nfs4_stateid_downgrade_bit(struct nfs4_ol_stateid *stp, u32 access)
4740 if (!test_access(access, stp))
4741 return;
4742 nfs4_file_put_access(stp->st_stid.sc_file, access);
4743 clear_access(access, stp);
4746 static inline void nfs4_stateid_downgrade(struct nfs4_ol_stateid *stp, u32 to_access)
4748 switch (to_access) {
4749 case NFS4_SHARE_ACCESS_READ:
4750 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_WRITE);
4751 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
4752 break;
4753 case NFS4_SHARE_ACCESS_WRITE:
4754 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_READ);
4755 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
4756 break;
4757 case NFS4_SHARE_ACCESS_BOTH:
4758 break;
4759 default:
4760 WARN_ON_ONCE(1);
4764 __be32
4765 nfsd4_open_downgrade(struct svc_rqst *rqstp,
4766 struct nfsd4_compound_state *cstate,
4767 struct nfsd4_open_downgrade *od)
4769 __be32 status;
4770 struct nfs4_ol_stateid *stp;
4771 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
4773 dprintk("NFSD: nfsd4_open_downgrade on file %pd\n",
4774 cstate->current_fh.fh_dentry);
4776 /* We don't yet support WANT bits: */
4777 if (od->od_deleg_want)
4778 dprintk("NFSD: %s: od_deleg_want=0x%x ignored\n", __func__,
4779 od->od_deleg_want);
4781 status = nfs4_preprocess_confirmed_seqid_op(cstate, od->od_seqid,
4782 &od->od_stateid, &stp, nn);
4783 if (status)
4784 goto out;
4785 status = nfserr_inval;
4786 if (!test_access(od->od_share_access, stp)) {
4787 dprintk("NFSD: access not a subset of current bitmap: 0x%hhx, input access=%08x\n",
4788 stp->st_access_bmap, od->od_share_access);
4789 goto put_stateid;
4791 if (!test_deny(od->od_share_deny, stp)) {
4792 dprintk("NFSD: deny not a subset of current bitmap: 0x%hhx, input deny=%08x\n",
4793 stp->st_deny_bmap, od->od_share_deny);
4794 goto put_stateid;
4796 nfs4_stateid_downgrade(stp, od->od_share_access);
4798 reset_union_bmap_deny(od->od_share_deny, stp);
4800 update_stateid(&stp->st_stid.sc_stateid);
4801 memcpy(&od->od_stateid, &stp->st_stid.sc_stateid, sizeof(stateid_t));
4802 status = nfs_ok;
4803 put_stateid:
4804 nfs4_put_stid(&stp->st_stid);
4805 out:
4806 nfsd4_bump_seqid(cstate, status);
4807 return status;
4810 static void nfsd4_close_open_stateid(struct nfs4_ol_stateid *s)
4812 struct nfs4_client *clp = s->st_stid.sc_client;
4813 LIST_HEAD(reaplist);
4815 s->st_stid.sc_type = NFS4_CLOSED_STID;
4816 spin_lock(&clp->cl_lock);
4817 unhash_open_stateid(s, &reaplist);
4819 if (clp->cl_minorversion) {
4820 put_ol_stateid_locked(s, &reaplist);
4821 spin_unlock(&clp->cl_lock);
4822 free_ol_stateid_reaplist(&reaplist);
4823 } else {
4824 spin_unlock(&clp->cl_lock);
4825 free_ol_stateid_reaplist(&reaplist);
4826 move_to_close_lru(s, clp->net);
4831 * nfs4_unlock_state() called after encode
4833 __be32
4834 nfsd4_close(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4835 struct nfsd4_close *close)
4837 __be32 status;
4838 struct nfs4_ol_stateid *stp;
4839 struct net *net = SVC_NET(rqstp);
4840 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
4842 dprintk("NFSD: nfsd4_close on file %pd\n",
4843 cstate->current_fh.fh_dentry);
4845 status = nfs4_preprocess_seqid_op(cstate, close->cl_seqid,
4846 &close->cl_stateid,
4847 NFS4_OPEN_STID|NFS4_CLOSED_STID,
4848 &stp, nn);
4849 nfsd4_bump_seqid(cstate, status);
4850 if (status)
4851 goto out;
4852 update_stateid(&stp->st_stid.sc_stateid);
4853 memcpy(&close->cl_stateid, &stp->st_stid.sc_stateid, sizeof(stateid_t));
4855 nfsd4_return_all_file_layouts(stp->st_stateowner->so_client,
4856 stp->st_stid.sc_file);
4858 nfsd4_close_open_stateid(stp);
4860 /* put reference from nfs4_preprocess_seqid_op */
4861 nfs4_put_stid(&stp->st_stid);
4862 out:
4863 return status;
4866 __be32
4867 nfsd4_delegreturn(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4868 struct nfsd4_delegreturn *dr)
4870 struct nfs4_delegation *dp;
4871 stateid_t *stateid = &dr->dr_stateid;
4872 struct nfs4_stid *s;
4873 __be32 status;
4874 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
4876 if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
4877 return status;
4879 status = nfsd4_lookup_stateid(cstate, stateid, NFS4_DELEG_STID, &s, nn);
4880 if (status)
4881 goto out;
4882 dp = delegstateid(s);
4883 status = check_stateid_generation(stateid, &dp->dl_stid.sc_stateid, nfsd4_has_session(cstate));
4884 if (status)
4885 goto put_stateid;
4887 destroy_delegation(dp);
4888 put_stateid:
4889 nfs4_put_stid(&dp->dl_stid);
4890 out:
4891 return status;
4895 #define LOFF_OVERFLOW(start, len) ((u64)(len) > ~(u64)(start))
4897 static inline u64
4898 end_offset(u64 start, u64 len)
4900 u64 end;
4902 end = start + len;
4903 return end >= start ? end: NFS4_MAX_UINT64;
4906 /* last octet in a range */
4907 static inline u64
4908 last_byte_offset(u64 start, u64 len)
4910 u64 end;
4912 WARN_ON_ONCE(!len);
4913 end = start + len;
4914 return end > start ? end - 1: NFS4_MAX_UINT64;
4918 * TODO: Linux file offsets are _signed_ 64-bit quantities, which means that
4919 * we can't properly handle lock requests that go beyond the (2^63 - 1)-th
4920 * byte, because of sign extension problems. Since NFSv4 calls for 64-bit
4921 * locking, this prevents us from being completely protocol-compliant. The
4922 * real solution to this problem is to start using unsigned file offsets in
4923 * the VFS, but this is a very deep change!
4925 static inline void
4926 nfs4_transform_lock_offset(struct file_lock *lock)
4928 if (lock->fl_start < 0)
4929 lock->fl_start = OFFSET_MAX;
4930 if (lock->fl_end < 0)
4931 lock->fl_end = OFFSET_MAX;
4934 static fl_owner_t
4935 nfsd4_fl_get_owner(fl_owner_t owner)
4937 struct nfs4_lockowner *lo = (struct nfs4_lockowner *)owner;
4939 nfs4_get_stateowner(&lo->lo_owner);
4940 return owner;
4943 static void
4944 nfsd4_fl_put_owner(fl_owner_t owner)
4946 struct nfs4_lockowner *lo = (struct nfs4_lockowner *)owner;
4948 if (lo)
4949 nfs4_put_stateowner(&lo->lo_owner);
4952 static const struct lock_manager_operations nfsd_posix_mng_ops = {
4953 .lm_get_owner = nfsd4_fl_get_owner,
4954 .lm_put_owner = nfsd4_fl_put_owner,
4957 static inline void
4958 nfs4_set_lock_denied(struct file_lock *fl, struct nfsd4_lock_denied *deny)
4960 struct nfs4_lockowner *lo;
4962 if (fl->fl_lmops == &nfsd_posix_mng_ops) {
4963 lo = (struct nfs4_lockowner *) fl->fl_owner;
4964 deny->ld_owner.data = kmemdup(lo->lo_owner.so_owner.data,
4965 lo->lo_owner.so_owner.len, GFP_KERNEL);
4966 if (!deny->ld_owner.data)
4967 /* We just don't care that much */
4968 goto nevermind;
4969 deny->ld_owner.len = lo->lo_owner.so_owner.len;
4970 deny->ld_clientid = lo->lo_owner.so_client->cl_clientid;
4971 } else {
4972 nevermind:
4973 deny->ld_owner.len = 0;
4974 deny->ld_owner.data = NULL;
4975 deny->ld_clientid.cl_boot = 0;
4976 deny->ld_clientid.cl_id = 0;
4978 deny->ld_start = fl->fl_start;
4979 deny->ld_length = NFS4_MAX_UINT64;
4980 if (fl->fl_end != NFS4_MAX_UINT64)
4981 deny->ld_length = fl->fl_end - fl->fl_start + 1;
4982 deny->ld_type = NFS4_READ_LT;
4983 if (fl->fl_type != F_RDLCK)
4984 deny->ld_type = NFS4_WRITE_LT;
4987 static struct nfs4_lockowner *
4988 find_lockowner_str_locked(clientid_t *clid, struct xdr_netobj *owner,
4989 struct nfs4_client *clp)
4991 unsigned int strhashval = ownerstr_hashval(owner);
4992 struct nfs4_stateowner *so;
4994 lockdep_assert_held(&clp->cl_lock);
4996 list_for_each_entry(so, &clp->cl_ownerstr_hashtbl[strhashval],
4997 so_strhash) {
4998 if (so->so_is_open_owner)
4999 continue;
5000 if (same_owner_str(so, owner))
5001 return lockowner(nfs4_get_stateowner(so));
5003 return NULL;
5006 static struct nfs4_lockowner *
5007 find_lockowner_str(clientid_t *clid, struct xdr_netobj *owner,
5008 struct nfs4_client *clp)
5010 struct nfs4_lockowner *lo;
5012 spin_lock(&clp->cl_lock);
5013 lo = find_lockowner_str_locked(clid, owner, clp);
5014 spin_unlock(&clp->cl_lock);
5015 return lo;
5018 static void nfs4_unhash_lockowner(struct nfs4_stateowner *sop)
5020 unhash_lockowner_locked(lockowner(sop));
5023 static void nfs4_free_lockowner(struct nfs4_stateowner *sop)
5025 struct nfs4_lockowner *lo = lockowner(sop);
5027 kmem_cache_free(lockowner_slab, lo);
5030 static const struct nfs4_stateowner_operations lockowner_ops = {
5031 .so_unhash = nfs4_unhash_lockowner,
5032 .so_free = nfs4_free_lockowner,
5036 * Alloc a lock owner structure.
5037 * Called in nfsd4_lock - therefore, OPEN and OPEN_CONFIRM (if needed) has
5038 * occurred.
5040 * strhashval = ownerstr_hashval
5042 static struct nfs4_lockowner *
5043 alloc_init_lock_stateowner(unsigned int strhashval, struct nfs4_client *clp,
5044 struct nfs4_ol_stateid *open_stp,
5045 struct nfsd4_lock *lock)
5047 struct nfs4_lockowner *lo, *ret;
5049 lo = alloc_stateowner(lockowner_slab, &lock->lk_new_owner, clp);
5050 if (!lo)
5051 return NULL;
5052 INIT_LIST_HEAD(&lo->lo_owner.so_stateids);
5053 lo->lo_owner.so_is_open_owner = 0;
5054 lo->lo_owner.so_seqid = lock->lk_new_lock_seqid;
5055 lo->lo_owner.so_ops = &lockowner_ops;
5056 spin_lock(&clp->cl_lock);
5057 ret = find_lockowner_str_locked(&clp->cl_clientid,
5058 &lock->lk_new_owner, clp);
5059 if (ret == NULL) {
5060 list_add(&lo->lo_owner.so_strhash,
5061 &clp->cl_ownerstr_hashtbl[strhashval]);
5062 ret = lo;
5063 } else
5064 nfs4_free_lockowner(&lo->lo_owner);
5065 spin_unlock(&clp->cl_lock);
5066 return ret;
5069 static void
5070 init_lock_stateid(struct nfs4_ol_stateid *stp, struct nfs4_lockowner *lo,
5071 struct nfs4_file *fp, struct inode *inode,
5072 struct nfs4_ol_stateid *open_stp)
5074 struct nfs4_client *clp = lo->lo_owner.so_client;
5076 lockdep_assert_held(&clp->cl_lock);
5078 atomic_inc(&stp->st_stid.sc_count);
5079 stp->st_stid.sc_type = NFS4_LOCK_STID;
5080 stp->st_stateowner = nfs4_get_stateowner(&lo->lo_owner);
5081 get_nfs4_file(fp);
5082 stp->st_stid.sc_file = fp;
5083 stp->st_stid.sc_free = nfs4_free_lock_stateid;
5084 stp->st_access_bmap = 0;
5085 stp->st_deny_bmap = open_stp->st_deny_bmap;
5086 stp->st_openstp = open_stp;
5087 list_add(&stp->st_locks, &open_stp->st_locks);
5088 list_add(&stp->st_perstateowner, &lo->lo_owner.so_stateids);
5089 spin_lock(&fp->fi_lock);
5090 list_add(&stp->st_perfile, &fp->fi_stateids);
5091 spin_unlock(&fp->fi_lock);
5094 static struct nfs4_ol_stateid *
5095 find_lock_stateid(struct nfs4_lockowner *lo, struct nfs4_file *fp)
5097 struct nfs4_ol_stateid *lst;
5098 struct nfs4_client *clp = lo->lo_owner.so_client;
5100 lockdep_assert_held(&clp->cl_lock);
5102 list_for_each_entry(lst, &lo->lo_owner.so_stateids, st_perstateowner) {
5103 if (lst->st_stid.sc_file == fp) {
5104 atomic_inc(&lst->st_stid.sc_count);
5105 return lst;
5108 return NULL;
5111 static struct nfs4_ol_stateid *
5112 find_or_create_lock_stateid(struct nfs4_lockowner *lo, struct nfs4_file *fi,
5113 struct inode *inode, struct nfs4_ol_stateid *ost,
5114 bool *new)
5116 struct nfs4_stid *ns = NULL;
5117 struct nfs4_ol_stateid *lst;
5118 struct nfs4_openowner *oo = openowner(ost->st_stateowner);
5119 struct nfs4_client *clp = oo->oo_owner.so_client;
5121 spin_lock(&clp->cl_lock);
5122 lst = find_lock_stateid(lo, fi);
5123 if (lst == NULL) {
5124 spin_unlock(&clp->cl_lock);
5125 ns = nfs4_alloc_stid(clp, stateid_slab);
5126 if (ns == NULL)
5127 return NULL;
5129 spin_lock(&clp->cl_lock);
5130 lst = find_lock_stateid(lo, fi);
5131 if (likely(!lst)) {
5132 lst = openlockstateid(ns);
5133 init_lock_stateid(lst, lo, fi, inode, ost);
5134 ns = NULL;
5135 *new = true;
5138 spin_unlock(&clp->cl_lock);
5139 if (ns)
5140 nfs4_put_stid(ns);
5141 return lst;
5144 static int
5145 check_lock_length(u64 offset, u64 length)
5147 return ((length == 0) || ((length != NFS4_MAX_UINT64) &&
5148 LOFF_OVERFLOW(offset, length)));
5151 static void get_lock_access(struct nfs4_ol_stateid *lock_stp, u32 access)
5153 struct nfs4_file *fp = lock_stp->st_stid.sc_file;
5155 lockdep_assert_held(&fp->fi_lock);
5157 if (test_access(access, lock_stp))
5158 return;
5159 __nfs4_file_get_access(fp, access);
5160 set_access(access, lock_stp);
5163 static __be32
5164 lookup_or_create_lock_state(struct nfsd4_compound_state *cstate,
5165 struct nfs4_ol_stateid *ost,
5166 struct nfsd4_lock *lock,
5167 struct nfs4_ol_stateid **lst, bool *new)
5169 __be32 status;
5170 struct nfs4_file *fi = ost->st_stid.sc_file;
5171 struct nfs4_openowner *oo = openowner(ost->st_stateowner);
5172 struct nfs4_client *cl = oo->oo_owner.so_client;
5173 struct inode *inode = d_inode(cstate->current_fh.fh_dentry);
5174 struct nfs4_lockowner *lo;
5175 unsigned int strhashval;
5177 lo = find_lockowner_str(&cl->cl_clientid, &lock->v.new.owner, cl);
5178 if (!lo) {
5179 strhashval = ownerstr_hashval(&lock->v.new.owner);
5180 lo = alloc_init_lock_stateowner(strhashval, cl, ost, lock);
5181 if (lo == NULL)
5182 return nfserr_jukebox;
5183 } else {
5184 /* with an existing lockowner, seqids must be the same */
5185 status = nfserr_bad_seqid;
5186 if (!cstate->minorversion &&
5187 lock->lk_new_lock_seqid != lo->lo_owner.so_seqid)
5188 goto out;
5191 *lst = find_or_create_lock_stateid(lo, fi, inode, ost, new);
5192 if (*lst == NULL) {
5193 status = nfserr_jukebox;
5194 goto out;
5196 status = nfs_ok;
5197 out:
5198 nfs4_put_stateowner(&lo->lo_owner);
5199 return status;
5203 * LOCK operation
5205 __be32
5206 nfsd4_lock(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5207 struct nfsd4_lock *lock)
5209 struct nfs4_openowner *open_sop = NULL;
5210 struct nfs4_lockowner *lock_sop = NULL;
5211 struct nfs4_ol_stateid *lock_stp = NULL;
5212 struct nfs4_ol_stateid *open_stp = NULL;
5213 struct nfs4_file *fp;
5214 struct file *filp = NULL;
5215 struct file_lock *file_lock = NULL;
5216 struct file_lock *conflock = NULL;
5217 __be32 status = 0;
5218 int lkflg;
5219 int err;
5220 bool new = false;
5221 struct net *net = SVC_NET(rqstp);
5222 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
5224 dprintk("NFSD: nfsd4_lock: start=%Ld length=%Ld\n",
5225 (long long) lock->lk_offset,
5226 (long long) lock->lk_length);
5228 if (check_lock_length(lock->lk_offset, lock->lk_length))
5229 return nfserr_inval;
5231 if ((status = fh_verify(rqstp, &cstate->current_fh,
5232 S_IFREG, NFSD_MAY_LOCK))) {
5233 dprintk("NFSD: nfsd4_lock: permission denied!\n");
5234 return status;
5237 if (lock->lk_is_new) {
5238 if (nfsd4_has_session(cstate))
5239 /* See rfc 5661 18.10.3: given clientid is ignored: */
5240 memcpy(&lock->v.new.clientid,
5241 &cstate->session->se_client->cl_clientid,
5242 sizeof(clientid_t));
5244 status = nfserr_stale_clientid;
5245 if (STALE_CLIENTID(&lock->lk_new_clientid, nn))
5246 goto out;
5248 /* validate and update open stateid and open seqid */
5249 status = nfs4_preprocess_confirmed_seqid_op(cstate,
5250 lock->lk_new_open_seqid,
5251 &lock->lk_new_open_stateid,
5252 &open_stp, nn);
5253 if (status)
5254 goto out;
5255 open_sop = openowner(open_stp->st_stateowner);
5256 status = nfserr_bad_stateid;
5257 if (!same_clid(&open_sop->oo_owner.so_client->cl_clientid,
5258 &lock->v.new.clientid))
5259 goto out;
5260 status = lookup_or_create_lock_state(cstate, open_stp, lock,
5261 &lock_stp, &new);
5262 } else {
5263 status = nfs4_preprocess_seqid_op(cstate,
5264 lock->lk_old_lock_seqid,
5265 &lock->lk_old_lock_stateid,
5266 NFS4_LOCK_STID, &lock_stp, nn);
5268 if (status)
5269 goto out;
5270 lock_sop = lockowner(lock_stp->st_stateowner);
5272 lkflg = setlkflg(lock->lk_type);
5273 status = nfs4_check_openmode(lock_stp, lkflg);
5274 if (status)
5275 goto out;
5277 status = nfserr_grace;
5278 if (locks_in_grace(net) && !lock->lk_reclaim)
5279 goto out;
5280 status = nfserr_no_grace;
5281 if (!locks_in_grace(net) && lock->lk_reclaim)
5282 goto out;
5284 file_lock = locks_alloc_lock();
5285 if (!file_lock) {
5286 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
5287 status = nfserr_jukebox;
5288 goto out;
5291 fp = lock_stp->st_stid.sc_file;
5292 switch (lock->lk_type) {
5293 case NFS4_READ_LT:
5294 case NFS4_READW_LT:
5295 spin_lock(&fp->fi_lock);
5296 filp = find_readable_file_locked(fp);
5297 if (filp)
5298 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_READ);
5299 spin_unlock(&fp->fi_lock);
5300 file_lock->fl_type = F_RDLCK;
5301 break;
5302 case NFS4_WRITE_LT:
5303 case NFS4_WRITEW_LT:
5304 spin_lock(&fp->fi_lock);
5305 filp = find_writeable_file_locked(fp);
5306 if (filp)
5307 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_WRITE);
5308 spin_unlock(&fp->fi_lock);
5309 file_lock->fl_type = F_WRLCK;
5310 break;
5311 default:
5312 status = nfserr_inval;
5313 goto out;
5315 if (!filp) {
5316 status = nfserr_openmode;
5317 goto out;
5320 file_lock->fl_owner = (fl_owner_t)lockowner(nfs4_get_stateowner(&lock_sop->lo_owner));
5321 file_lock->fl_pid = current->tgid;
5322 file_lock->fl_file = filp;
5323 file_lock->fl_flags = FL_POSIX;
5324 file_lock->fl_lmops = &nfsd_posix_mng_ops;
5325 file_lock->fl_start = lock->lk_offset;
5326 file_lock->fl_end = last_byte_offset(lock->lk_offset, lock->lk_length);
5327 nfs4_transform_lock_offset(file_lock);
5329 conflock = locks_alloc_lock();
5330 if (!conflock) {
5331 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
5332 status = nfserr_jukebox;
5333 goto out;
5336 err = vfs_lock_file(filp, F_SETLK, file_lock, conflock);
5337 switch (-err) {
5338 case 0: /* success! */
5339 update_stateid(&lock_stp->st_stid.sc_stateid);
5340 memcpy(&lock->lk_resp_stateid, &lock_stp->st_stid.sc_stateid,
5341 sizeof(stateid_t));
5342 status = 0;
5343 break;
5344 case (EAGAIN): /* conflock holds conflicting lock */
5345 status = nfserr_denied;
5346 dprintk("NFSD: nfsd4_lock: conflicting lock found!\n");
5347 nfs4_set_lock_denied(conflock, &lock->lk_denied);
5348 break;
5349 case (EDEADLK):
5350 status = nfserr_deadlock;
5351 break;
5352 default:
5353 dprintk("NFSD: nfsd4_lock: vfs_lock_file() failed! status %d\n",err);
5354 status = nfserrno(err);
5355 break;
5357 out:
5358 if (filp)
5359 fput(filp);
5360 if (lock_stp) {
5361 /* Bump seqid manually if the 4.0 replay owner is openowner */
5362 if (cstate->replay_owner &&
5363 cstate->replay_owner != &lock_sop->lo_owner &&
5364 seqid_mutating_err(ntohl(status)))
5365 lock_sop->lo_owner.so_seqid++;
5368 * If this is a new, never-before-used stateid, and we are
5369 * returning an error, then just go ahead and release it.
5371 if (status && new)
5372 release_lock_stateid(lock_stp);
5374 nfs4_put_stid(&lock_stp->st_stid);
5376 if (open_stp)
5377 nfs4_put_stid(&open_stp->st_stid);
5378 nfsd4_bump_seqid(cstate, status);
5379 if (file_lock)
5380 locks_free_lock(file_lock);
5381 if (conflock)
5382 locks_free_lock(conflock);
5383 return status;
5387 * The NFSv4 spec allows a client to do a LOCKT without holding an OPEN,
5388 * so we do a temporary open here just to get an open file to pass to
5389 * vfs_test_lock. (Arguably perhaps test_lock should be done with an
5390 * inode operation.)
5392 static __be32 nfsd_test_lock(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file_lock *lock)
5394 struct file *file;
5395 __be32 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file);
5396 if (!err) {
5397 err = nfserrno(vfs_test_lock(file, lock));
5398 nfsd_close(file);
5400 return err;
5404 * LOCKT operation
5406 __be32
5407 nfsd4_lockt(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5408 struct nfsd4_lockt *lockt)
5410 struct file_lock *file_lock = NULL;
5411 struct nfs4_lockowner *lo = NULL;
5412 __be32 status;
5413 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
5415 if (locks_in_grace(SVC_NET(rqstp)))
5416 return nfserr_grace;
5418 if (check_lock_length(lockt->lt_offset, lockt->lt_length))
5419 return nfserr_inval;
5421 if (!nfsd4_has_session(cstate)) {
5422 status = lookup_clientid(&lockt->lt_clientid, cstate, nn);
5423 if (status)
5424 goto out;
5427 if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
5428 goto out;
5430 file_lock = locks_alloc_lock();
5431 if (!file_lock) {
5432 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
5433 status = nfserr_jukebox;
5434 goto out;
5437 switch (lockt->lt_type) {
5438 case NFS4_READ_LT:
5439 case NFS4_READW_LT:
5440 file_lock->fl_type = F_RDLCK;
5441 break;
5442 case NFS4_WRITE_LT:
5443 case NFS4_WRITEW_LT:
5444 file_lock->fl_type = F_WRLCK;
5445 break;
5446 default:
5447 dprintk("NFSD: nfs4_lockt: bad lock type!\n");
5448 status = nfserr_inval;
5449 goto out;
5452 lo = find_lockowner_str(&lockt->lt_clientid, &lockt->lt_owner,
5453 cstate->clp);
5454 if (lo)
5455 file_lock->fl_owner = (fl_owner_t)lo;
5456 file_lock->fl_pid = current->tgid;
5457 file_lock->fl_flags = FL_POSIX;
5459 file_lock->fl_start = lockt->lt_offset;
5460 file_lock->fl_end = last_byte_offset(lockt->lt_offset, lockt->lt_length);
5462 nfs4_transform_lock_offset(file_lock);
5464 status = nfsd_test_lock(rqstp, &cstate->current_fh, file_lock);
5465 if (status)
5466 goto out;
5468 if (file_lock->fl_type != F_UNLCK) {
5469 status = nfserr_denied;
5470 nfs4_set_lock_denied(file_lock, &lockt->lt_denied);
5472 out:
5473 if (lo)
5474 nfs4_put_stateowner(&lo->lo_owner);
5475 if (file_lock)
5476 locks_free_lock(file_lock);
5477 return status;
5480 __be32
5481 nfsd4_locku(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5482 struct nfsd4_locku *locku)
5484 struct nfs4_ol_stateid *stp;
5485 struct file *filp = NULL;
5486 struct file_lock *file_lock = NULL;
5487 __be32 status;
5488 int err;
5489 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
5491 dprintk("NFSD: nfsd4_locku: start=%Ld length=%Ld\n",
5492 (long long) locku->lu_offset,
5493 (long long) locku->lu_length);
5495 if (check_lock_length(locku->lu_offset, locku->lu_length))
5496 return nfserr_inval;
5498 status = nfs4_preprocess_seqid_op(cstate, locku->lu_seqid,
5499 &locku->lu_stateid, NFS4_LOCK_STID,
5500 &stp, nn);
5501 if (status)
5502 goto out;
5503 filp = find_any_file(stp->st_stid.sc_file);
5504 if (!filp) {
5505 status = nfserr_lock_range;
5506 goto put_stateid;
5508 file_lock = locks_alloc_lock();
5509 if (!file_lock) {
5510 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
5511 status = nfserr_jukebox;
5512 goto fput;
5515 file_lock->fl_type = F_UNLCK;
5516 file_lock->fl_owner = (fl_owner_t)lockowner(nfs4_get_stateowner(stp->st_stateowner));
5517 file_lock->fl_pid = current->tgid;
5518 file_lock->fl_file = filp;
5519 file_lock->fl_flags = FL_POSIX;
5520 file_lock->fl_lmops = &nfsd_posix_mng_ops;
5521 file_lock->fl_start = locku->lu_offset;
5523 file_lock->fl_end = last_byte_offset(locku->lu_offset,
5524 locku->lu_length);
5525 nfs4_transform_lock_offset(file_lock);
5527 err = vfs_lock_file(filp, F_SETLK, file_lock, NULL);
5528 if (err) {
5529 dprintk("NFSD: nfs4_locku: vfs_lock_file failed!\n");
5530 goto out_nfserr;
5532 update_stateid(&stp->st_stid.sc_stateid);
5533 memcpy(&locku->lu_stateid, &stp->st_stid.sc_stateid, sizeof(stateid_t));
5534 fput:
5535 fput(filp);
5536 put_stateid:
5537 nfs4_put_stid(&stp->st_stid);
5538 out:
5539 nfsd4_bump_seqid(cstate, status);
5540 if (file_lock)
5541 locks_free_lock(file_lock);
5542 return status;
5544 out_nfserr:
5545 status = nfserrno(err);
5546 goto fput;
5550 * returns
5551 * true: locks held by lockowner
5552 * false: no locks held by lockowner
5554 static bool
5555 check_for_locks(struct nfs4_file *fp, struct nfs4_lockowner *lowner)
5557 struct file_lock *fl;
5558 int status = false;
5559 struct file *filp = find_any_file(fp);
5560 struct inode *inode;
5561 struct file_lock_context *flctx;
5563 if (!filp) {
5564 /* Any valid lock stateid should have some sort of access */
5565 WARN_ON_ONCE(1);
5566 return status;
5569 inode = file_inode(filp);
5570 flctx = inode->i_flctx;
5572 if (flctx && !list_empty_careful(&flctx->flc_posix)) {
5573 spin_lock(&flctx->flc_lock);
5574 list_for_each_entry(fl, &flctx->flc_posix, fl_list) {
5575 if (fl->fl_owner == (fl_owner_t)lowner) {
5576 status = true;
5577 break;
5580 spin_unlock(&flctx->flc_lock);
5582 fput(filp);
5583 return status;
5586 __be32
5587 nfsd4_release_lockowner(struct svc_rqst *rqstp,
5588 struct nfsd4_compound_state *cstate,
5589 struct nfsd4_release_lockowner *rlockowner)
5591 clientid_t *clid = &rlockowner->rl_clientid;
5592 struct nfs4_stateowner *sop;
5593 struct nfs4_lockowner *lo = NULL;
5594 struct nfs4_ol_stateid *stp;
5595 struct xdr_netobj *owner = &rlockowner->rl_owner;
5596 unsigned int hashval = ownerstr_hashval(owner);
5597 __be32 status;
5598 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
5599 struct nfs4_client *clp;
5601 dprintk("nfsd4_release_lockowner clientid: (%08x/%08x):\n",
5602 clid->cl_boot, clid->cl_id);
5604 status = lookup_clientid(clid, cstate, nn);
5605 if (status)
5606 return status;
5608 clp = cstate->clp;
5609 /* Find the matching lock stateowner */
5610 spin_lock(&clp->cl_lock);
5611 list_for_each_entry(sop, &clp->cl_ownerstr_hashtbl[hashval],
5612 so_strhash) {
5614 if (sop->so_is_open_owner || !same_owner_str(sop, owner))
5615 continue;
5617 /* see if there are still any locks associated with it */
5618 lo = lockowner(sop);
5619 list_for_each_entry(stp, &sop->so_stateids, st_perstateowner) {
5620 if (check_for_locks(stp->st_stid.sc_file, lo)) {
5621 status = nfserr_locks_held;
5622 spin_unlock(&clp->cl_lock);
5623 return status;
5627 nfs4_get_stateowner(sop);
5628 break;
5630 spin_unlock(&clp->cl_lock);
5631 if (lo)
5632 release_lockowner(lo);
5633 return status;
5636 static inline struct nfs4_client_reclaim *
5637 alloc_reclaim(void)
5639 return kmalloc(sizeof(struct nfs4_client_reclaim), GFP_KERNEL);
5642 bool
5643 nfs4_has_reclaimed_state(const char *name, struct nfsd_net *nn)
5645 struct nfs4_client_reclaim *crp;
5647 crp = nfsd4_find_reclaim_client(name, nn);
5648 return (crp && crp->cr_clp);
5652 * failure => all reset bets are off, nfserr_no_grace...
5654 struct nfs4_client_reclaim *
5655 nfs4_client_to_reclaim(const char *name, struct nfsd_net *nn)
5657 unsigned int strhashval;
5658 struct nfs4_client_reclaim *crp;
5660 dprintk("NFSD nfs4_client_to_reclaim NAME: %.*s\n", HEXDIR_LEN, name);
5661 crp = alloc_reclaim();
5662 if (crp) {
5663 strhashval = clientstr_hashval(name);
5664 INIT_LIST_HEAD(&crp->cr_strhash);
5665 list_add(&crp->cr_strhash, &nn->reclaim_str_hashtbl[strhashval]);
5666 memcpy(crp->cr_recdir, name, HEXDIR_LEN);
5667 crp->cr_clp = NULL;
5668 nn->reclaim_str_hashtbl_size++;
5670 return crp;
5673 void
5674 nfs4_remove_reclaim_record(struct nfs4_client_reclaim *crp, struct nfsd_net *nn)
5676 list_del(&crp->cr_strhash);
5677 kfree(crp);
5678 nn->reclaim_str_hashtbl_size--;
5681 void
5682 nfs4_release_reclaim(struct nfsd_net *nn)
5684 struct nfs4_client_reclaim *crp = NULL;
5685 int i;
5687 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
5688 while (!list_empty(&nn->reclaim_str_hashtbl[i])) {
5689 crp = list_entry(nn->reclaim_str_hashtbl[i].next,
5690 struct nfs4_client_reclaim, cr_strhash);
5691 nfs4_remove_reclaim_record(crp, nn);
5694 WARN_ON_ONCE(nn->reclaim_str_hashtbl_size);
5698 * called from OPEN, CLAIM_PREVIOUS with a new clientid. */
5699 struct nfs4_client_reclaim *
5700 nfsd4_find_reclaim_client(const char *recdir, struct nfsd_net *nn)
5702 unsigned int strhashval;
5703 struct nfs4_client_reclaim *crp = NULL;
5705 dprintk("NFSD: nfs4_find_reclaim_client for recdir %s\n", recdir);
5707 strhashval = clientstr_hashval(recdir);
5708 list_for_each_entry(crp, &nn->reclaim_str_hashtbl[strhashval], cr_strhash) {
5709 if (same_name(crp->cr_recdir, recdir)) {
5710 return crp;
5713 return NULL;
5717 * Called from OPEN. Look for clientid in reclaim list.
5719 __be32
5720 nfs4_check_open_reclaim(clientid_t *clid,
5721 struct nfsd4_compound_state *cstate,
5722 struct nfsd_net *nn)
5724 __be32 status;
5726 /* find clientid in conf_id_hashtbl */
5727 status = lookup_clientid(clid, cstate, nn);
5728 if (status)
5729 return nfserr_reclaim_bad;
5731 if (test_bit(NFSD4_CLIENT_RECLAIM_COMPLETE, &cstate->clp->cl_flags))
5732 return nfserr_no_grace;
5734 if (nfsd4_client_record_check(cstate->clp))
5735 return nfserr_reclaim_bad;
5737 return nfs_ok;
5740 #ifdef CONFIG_NFSD_FAULT_INJECTION
5741 static inline void
5742 put_client(struct nfs4_client *clp)
5744 atomic_dec(&clp->cl_refcount);
5747 static struct nfs4_client *
5748 nfsd_find_client(struct sockaddr_storage *addr, size_t addr_size)
5750 struct nfs4_client *clp;
5751 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
5752 nfsd_net_id);
5754 if (!nfsd_netns_ready(nn))
5755 return NULL;
5757 list_for_each_entry(clp, &nn->client_lru, cl_lru) {
5758 if (memcmp(&clp->cl_addr, addr, addr_size) == 0)
5759 return clp;
5761 return NULL;
5765 nfsd_inject_print_clients(void)
5767 struct nfs4_client *clp;
5768 u64 count = 0;
5769 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
5770 nfsd_net_id);
5771 char buf[INET6_ADDRSTRLEN];
5773 if (!nfsd_netns_ready(nn))
5774 return 0;
5776 spin_lock(&nn->client_lock);
5777 list_for_each_entry(clp, &nn->client_lru, cl_lru) {
5778 rpc_ntop((struct sockaddr *)&clp->cl_addr, buf, sizeof(buf));
5779 pr_info("NFS Client: %s\n", buf);
5780 ++count;
5782 spin_unlock(&nn->client_lock);
5784 return count;
5788 nfsd_inject_forget_client(struct sockaddr_storage *addr, size_t addr_size)
5790 u64 count = 0;
5791 struct nfs4_client *clp;
5792 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
5793 nfsd_net_id);
5795 if (!nfsd_netns_ready(nn))
5796 return count;
5798 spin_lock(&nn->client_lock);
5799 clp = nfsd_find_client(addr, addr_size);
5800 if (clp) {
5801 if (mark_client_expired_locked(clp) == nfs_ok)
5802 ++count;
5803 else
5804 clp = NULL;
5806 spin_unlock(&nn->client_lock);
5808 if (clp)
5809 expire_client(clp);
5811 return count;
5815 nfsd_inject_forget_clients(u64 max)
5817 u64 count = 0;
5818 struct nfs4_client *clp, *next;
5819 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
5820 nfsd_net_id);
5821 LIST_HEAD(reaplist);
5823 if (!nfsd_netns_ready(nn))
5824 return count;
5826 spin_lock(&nn->client_lock);
5827 list_for_each_entry_safe(clp, next, &nn->client_lru, cl_lru) {
5828 if (mark_client_expired_locked(clp) == nfs_ok) {
5829 list_add(&clp->cl_lru, &reaplist);
5830 if (max != 0 && ++count >= max)
5831 break;
5834 spin_unlock(&nn->client_lock);
5836 list_for_each_entry_safe(clp, next, &reaplist, cl_lru)
5837 expire_client(clp);
5839 return count;
5842 static void nfsd_print_count(struct nfs4_client *clp, unsigned int count,
5843 const char *type)
5845 char buf[INET6_ADDRSTRLEN];
5846 rpc_ntop((struct sockaddr *)&clp->cl_addr, buf, sizeof(buf));
5847 printk(KERN_INFO "NFS Client: %s has %u %s\n", buf, count, type);
5850 static void
5851 nfsd_inject_add_lock_to_list(struct nfs4_ol_stateid *lst,
5852 struct list_head *collect)
5854 struct nfs4_client *clp = lst->st_stid.sc_client;
5855 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
5856 nfsd_net_id);
5858 if (!collect)
5859 return;
5861 lockdep_assert_held(&nn->client_lock);
5862 atomic_inc(&clp->cl_refcount);
5863 list_add(&lst->st_locks, collect);
5866 static u64 nfsd_foreach_client_lock(struct nfs4_client *clp, u64 max,
5867 struct list_head *collect,
5868 void (*func)(struct nfs4_ol_stateid *))
5870 struct nfs4_openowner *oop;
5871 struct nfs4_ol_stateid *stp, *st_next;
5872 struct nfs4_ol_stateid *lst, *lst_next;
5873 u64 count = 0;
5875 spin_lock(&clp->cl_lock);
5876 list_for_each_entry(oop, &clp->cl_openowners, oo_perclient) {
5877 list_for_each_entry_safe(stp, st_next,
5878 &oop->oo_owner.so_stateids, st_perstateowner) {
5879 list_for_each_entry_safe(lst, lst_next,
5880 &stp->st_locks, st_locks) {
5881 if (func) {
5882 func(lst);
5883 nfsd_inject_add_lock_to_list(lst,
5884 collect);
5886 ++count;
5888 * Despite the fact that these functions deal
5889 * with 64-bit integers for "count", we must
5890 * ensure that it doesn't blow up the
5891 * clp->cl_refcount. Throw a warning if we
5892 * start to approach INT_MAX here.
5894 WARN_ON_ONCE(count == (INT_MAX / 2));
5895 if (count == max)
5896 goto out;
5900 out:
5901 spin_unlock(&clp->cl_lock);
5903 return count;
5906 static u64
5907 nfsd_collect_client_locks(struct nfs4_client *clp, struct list_head *collect,
5908 u64 max)
5910 return nfsd_foreach_client_lock(clp, max, collect, unhash_lock_stateid);
5913 static u64
5914 nfsd_print_client_locks(struct nfs4_client *clp)
5916 u64 count = nfsd_foreach_client_lock(clp, 0, NULL, NULL);
5917 nfsd_print_count(clp, count, "locked files");
5918 return count;
5922 nfsd_inject_print_locks(void)
5924 struct nfs4_client *clp;
5925 u64 count = 0;
5926 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
5927 nfsd_net_id);
5929 if (!nfsd_netns_ready(nn))
5930 return 0;
5932 spin_lock(&nn->client_lock);
5933 list_for_each_entry(clp, &nn->client_lru, cl_lru)
5934 count += nfsd_print_client_locks(clp);
5935 spin_unlock(&nn->client_lock);
5937 return count;
5940 static void
5941 nfsd_reap_locks(struct list_head *reaplist)
5943 struct nfs4_client *clp;
5944 struct nfs4_ol_stateid *stp, *next;
5946 list_for_each_entry_safe(stp, next, reaplist, st_locks) {
5947 list_del_init(&stp->st_locks);
5948 clp = stp->st_stid.sc_client;
5949 nfs4_put_stid(&stp->st_stid);
5950 put_client(clp);
5955 nfsd_inject_forget_client_locks(struct sockaddr_storage *addr, size_t addr_size)
5957 unsigned int count = 0;
5958 struct nfs4_client *clp;
5959 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
5960 nfsd_net_id);
5961 LIST_HEAD(reaplist);
5963 if (!nfsd_netns_ready(nn))
5964 return count;
5966 spin_lock(&nn->client_lock);
5967 clp = nfsd_find_client(addr, addr_size);
5968 if (clp)
5969 count = nfsd_collect_client_locks(clp, &reaplist, 0);
5970 spin_unlock(&nn->client_lock);
5971 nfsd_reap_locks(&reaplist);
5972 return count;
5976 nfsd_inject_forget_locks(u64 max)
5978 u64 count = 0;
5979 struct nfs4_client *clp;
5980 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
5981 nfsd_net_id);
5982 LIST_HEAD(reaplist);
5984 if (!nfsd_netns_ready(nn))
5985 return count;
5987 spin_lock(&nn->client_lock);
5988 list_for_each_entry(clp, &nn->client_lru, cl_lru) {
5989 count += nfsd_collect_client_locks(clp, &reaplist, max - count);
5990 if (max != 0 && count >= max)
5991 break;
5993 spin_unlock(&nn->client_lock);
5994 nfsd_reap_locks(&reaplist);
5995 return count;
5998 static u64
5999 nfsd_foreach_client_openowner(struct nfs4_client *clp, u64 max,
6000 struct list_head *collect,
6001 void (*func)(struct nfs4_openowner *))
6003 struct nfs4_openowner *oop, *next;
6004 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6005 nfsd_net_id);
6006 u64 count = 0;
6008 lockdep_assert_held(&nn->client_lock);
6010 spin_lock(&clp->cl_lock);
6011 list_for_each_entry_safe(oop, next, &clp->cl_openowners, oo_perclient) {
6012 if (func) {
6013 func(oop);
6014 if (collect) {
6015 atomic_inc(&clp->cl_refcount);
6016 list_add(&oop->oo_perclient, collect);
6019 ++count;
6021 * Despite the fact that these functions deal with
6022 * 64-bit integers for "count", we must ensure that
6023 * it doesn't blow up the clp->cl_refcount. Throw a
6024 * warning if we start to approach INT_MAX here.
6026 WARN_ON_ONCE(count == (INT_MAX / 2));
6027 if (count == max)
6028 break;
6030 spin_unlock(&clp->cl_lock);
6032 return count;
6035 static u64
6036 nfsd_print_client_openowners(struct nfs4_client *clp)
6038 u64 count = nfsd_foreach_client_openowner(clp, 0, NULL, NULL);
6040 nfsd_print_count(clp, count, "openowners");
6041 return count;
6044 static u64
6045 nfsd_collect_client_openowners(struct nfs4_client *clp,
6046 struct list_head *collect, u64 max)
6048 return nfsd_foreach_client_openowner(clp, max, collect,
6049 unhash_openowner_locked);
6053 nfsd_inject_print_openowners(void)
6055 struct nfs4_client *clp;
6056 u64 count = 0;
6057 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6058 nfsd_net_id);
6060 if (!nfsd_netns_ready(nn))
6061 return 0;
6063 spin_lock(&nn->client_lock);
6064 list_for_each_entry(clp, &nn->client_lru, cl_lru)
6065 count += nfsd_print_client_openowners(clp);
6066 spin_unlock(&nn->client_lock);
6068 return count;
6071 static void
6072 nfsd_reap_openowners(struct list_head *reaplist)
6074 struct nfs4_client *clp;
6075 struct nfs4_openowner *oop, *next;
6077 list_for_each_entry_safe(oop, next, reaplist, oo_perclient) {
6078 list_del_init(&oop->oo_perclient);
6079 clp = oop->oo_owner.so_client;
6080 release_openowner(oop);
6081 put_client(clp);
6086 nfsd_inject_forget_client_openowners(struct sockaddr_storage *addr,
6087 size_t addr_size)
6089 unsigned int count = 0;
6090 struct nfs4_client *clp;
6091 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6092 nfsd_net_id);
6093 LIST_HEAD(reaplist);
6095 if (!nfsd_netns_ready(nn))
6096 return count;
6098 spin_lock(&nn->client_lock);
6099 clp = nfsd_find_client(addr, addr_size);
6100 if (clp)
6101 count = nfsd_collect_client_openowners(clp, &reaplist, 0);
6102 spin_unlock(&nn->client_lock);
6103 nfsd_reap_openowners(&reaplist);
6104 return count;
6108 nfsd_inject_forget_openowners(u64 max)
6110 u64 count = 0;
6111 struct nfs4_client *clp;
6112 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6113 nfsd_net_id);
6114 LIST_HEAD(reaplist);
6116 if (!nfsd_netns_ready(nn))
6117 return count;
6119 spin_lock(&nn->client_lock);
6120 list_for_each_entry(clp, &nn->client_lru, cl_lru) {
6121 count += nfsd_collect_client_openowners(clp, &reaplist,
6122 max - count);
6123 if (max != 0 && count >= max)
6124 break;
6126 spin_unlock(&nn->client_lock);
6127 nfsd_reap_openowners(&reaplist);
6128 return count;
6131 static u64 nfsd_find_all_delegations(struct nfs4_client *clp, u64 max,
6132 struct list_head *victims)
6134 struct nfs4_delegation *dp, *next;
6135 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6136 nfsd_net_id);
6137 u64 count = 0;
6139 lockdep_assert_held(&nn->client_lock);
6141 spin_lock(&state_lock);
6142 list_for_each_entry_safe(dp, next, &clp->cl_delegations, dl_perclnt) {
6143 if (victims) {
6145 * It's not safe to mess with delegations that have a
6146 * non-zero dl_time. They might have already been broken
6147 * and could be processed by the laundromat outside of
6148 * the state_lock. Just leave them be.
6150 if (dp->dl_time != 0)
6151 continue;
6153 atomic_inc(&clp->cl_refcount);
6154 unhash_delegation_locked(dp);
6155 list_add(&dp->dl_recall_lru, victims);
6157 ++count;
6159 * Despite the fact that these functions deal with
6160 * 64-bit integers for "count", we must ensure that
6161 * it doesn't blow up the clp->cl_refcount. Throw a
6162 * warning if we start to approach INT_MAX here.
6164 WARN_ON_ONCE(count == (INT_MAX / 2));
6165 if (count == max)
6166 break;
6168 spin_unlock(&state_lock);
6169 return count;
6172 static u64
6173 nfsd_print_client_delegations(struct nfs4_client *clp)
6175 u64 count = nfsd_find_all_delegations(clp, 0, NULL);
6177 nfsd_print_count(clp, count, "delegations");
6178 return count;
6182 nfsd_inject_print_delegations(void)
6184 struct nfs4_client *clp;
6185 u64 count = 0;
6186 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6187 nfsd_net_id);
6189 if (!nfsd_netns_ready(nn))
6190 return 0;
6192 spin_lock(&nn->client_lock);
6193 list_for_each_entry(clp, &nn->client_lru, cl_lru)
6194 count += nfsd_print_client_delegations(clp);
6195 spin_unlock(&nn->client_lock);
6197 return count;
6200 static void
6201 nfsd_forget_delegations(struct list_head *reaplist)
6203 struct nfs4_client *clp;
6204 struct nfs4_delegation *dp, *next;
6206 list_for_each_entry_safe(dp, next, reaplist, dl_recall_lru) {
6207 list_del_init(&dp->dl_recall_lru);
6208 clp = dp->dl_stid.sc_client;
6209 revoke_delegation(dp);
6210 put_client(clp);
6215 nfsd_inject_forget_client_delegations(struct sockaddr_storage *addr,
6216 size_t addr_size)
6218 u64 count = 0;
6219 struct nfs4_client *clp;
6220 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6221 nfsd_net_id);
6222 LIST_HEAD(reaplist);
6224 if (!nfsd_netns_ready(nn))
6225 return count;
6227 spin_lock(&nn->client_lock);
6228 clp = nfsd_find_client(addr, addr_size);
6229 if (clp)
6230 count = nfsd_find_all_delegations(clp, 0, &reaplist);
6231 spin_unlock(&nn->client_lock);
6233 nfsd_forget_delegations(&reaplist);
6234 return count;
6238 nfsd_inject_forget_delegations(u64 max)
6240 u64 count = 0;
6241 struct nfs4_client *clp;
6242 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6243 nfsd_net_id);
6244 LIST_HEAD(reaplist);
6246 if (!nfsd_netns_ready(nn))
6247 return count;
6249 spin_lock(&nn->client_lock);
6250 list_for_each_entry(clp, &nn->client_lru, cl_lru) {
6251 count += nfsd_find_all_delegations(clp, max - count, &reaplist);
6252 if (max != 0 && count >= max)
6253 break;
6255 spin_unlock(&nn->client_lock);
6256 nfsd_forget_delegations(&reaplist);
6257 return count;
6260 static void
6261 nfsd_recall_delegations(struct list_head *reaplist)
6263 struct nfs4_client *clp;
6264 struct nfs4_delegation *dp, *next;
6266 list_for_each_entry_safe(dp, next, reaplist, dl_recall_lru) {
6267 list_del_init(&dp->dl_recall_lru);
6268 clp = dp->dl_stid.sc_client;
6270 * We skipped all entries that had a zero dl_time before,
6271 * so we can now reset the dl_time back to 0. If a delegation
6272 * break comes in now, then it won't make any difference since
6273 * we're recalling it either way.
6275 spin_lock(&state_lock);
6276 dp->dl_time = 0;
6277 spin_unlock(&state_lock);
6278 nfsd_break_one_deleg(dp);
6279 put_client(clp);
6284 nfsd_inject_recall_client_delegations(struct sockaddr_storage *addr,
6285 size_t addr_size)
6287 u64 count = 0;
6288 struct nfs4_client *clp;
6289 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6290 nfsd_net_id);
6291 LIST_HEAD(reaplist);
6293 if (!nfsd_netns_ready(nn))
6294 return count;
6296 spin_lock(&nn->client_lock);
6297 clp = nfsd_find_client(addr, addr_size);
6298 if (clp)
6299 count = nfsd_find_all_delegations(clp, 0, &reaplist);
6300 spin_unlock(&nn->client_lock);
6302 nfsd_recall_delegations(&reaplist);
6303 return count;
6307 nfsd_inject_recall_delegations(u64 max)
6309 u64 count = 0;
6310 struct nfs4_client *clp, *next;
6311 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6312 nfsd_net_id);
6313 LIST_HEAD(reaplist);
6315 if (!nfsd_netns_ready(nn))
6316 return count;
6318 spin_lock(&nn->client_lock);
6319 list_for_each_entry_safe(clp, next, &nn->client_lru, cl_lru) {
6320 count += nfsd_find_all_delegations(clp, max - count, &reaplist);
6321 if (max != 0 && ++count >= max)
6322 break;
6324 spin_unlock(&nn->client_lock);
6325 nfsd_recall_delegations(&reaplist);
6326 return count;
6328 #endif /* CONFIG_NFSD_FAULT_INJECTION */
6331 * Since the lifetime of a delegation isn't limited to that of an open, a
6332 * client may quite reasonably hang on to a delegation as long as it has
6333 * the inode cached. This becomes an obvious problem the first time a
6334 * client's inode cache approaches the size of the server's total memory.
6336 * For now we avoid this problem by imposing a hard limit on the number
6337 * of delegations, which varies according to the server's memory size.
6339 static void
6340 set_max_delegations(void)
6343 * Allow at most 4 delegations per megabyte of RAM. Quick
6344 * estimates suggest that in the worst case (where every delegation
6345 * is for a different inode), a delegation could take about 1.5K,
6346 * giving a worst case usage of about 6% of memory.
6348 max_delegations = nr_free_buffer_pages() >> (20 - 2 - PAGE_SHIFT);
6351 static int nfs4_state_create_net(struct net *net)
6353 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
6354 int i;
6356 nn->conf_id_hashtbl = kmalloc(sizeof(struct list_head) *
6357 CLIENT_HASH_SIZE, GFP_KERNEL);
6358 if (!nn->conf_id_hashtbl)
6359 goto err;
6360 nn->unconf_id_hashtbl = kmalloc(sizeof(struct list_head) *
6361 CLIENT_HASH_SIZE, GFP_KERNEL);
6362 if (!nn->unconf_id_hashtbl)
6363 goto err_unconf_id;
6364 nn->sessionid_hashtbl = kmalloc(sizeof(struct list_head) *
6365 SESSION_HASH_SIZE, GFP_KERNEL);
6366 if (!nn->sessionid_hashtbl)
6367 goto err_sessionid;
6369 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
6370 INIT_LIST_HEAD(&nn->conf_id_hashtbl[i]);
6371 INIT_LIST_HEAD(&nn->unconf_id_hashtbl[i]);
6373 for (i = 0; i < SESSION_HASH_SIZE; i++)
6374 INIT_LIST_HEAD(&nn->sessionid_hashtbl[i]);
6375 nn->conf_name_tree = RB_ROOT;
6376 nn->unconf_name_tree = RB_ROOT;
6377 INIT_LIST_HEAD(&nn->client_lru);
6378 INIT_LIST_HEAD(&nn->close_lru);
6379 INIT_LIST_HEAD(&nn->del_recall_lru);
6380 spin_lock_init(&nn->client_lock);
6382 INIT_DELAYED_WORK(&nn->laundromat_work, laundromat_main);
6383 get_net(net);
6385 return 0;
6387 err_sessionid:
6388 kfree(nn->unconf_id_hashtbl);
6389 err_unconf_id:
6390 kfree(nn->conf_id_hashtbl);
6391 err:
6392 return -ENOMEM;
6395 static void
6396 nfs4_state_destroy_net(struct net *net)
6398 int i;
6399 struct nfs4_client *clp = NULL;
6400 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
6402 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
6403 while (!list_empty(&nn->conf_id_hashtbl[i])) {
6404 clp = list_entry(nn->conf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
6405 destroy_client(clp);
6409 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
6410 while (!list_empty(&nn->unconf_id_hashtbl[i])) {
6411 clp = list_entry(nn->unconf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
6412 destroy_client(clp);
6416 kfree(nn->sessionid_hashtbl);
6417 kfree(nn->unconf_id_hashtbl);
6418 kfree(nn->conf_id_hashtbl);
6419 put_net(net);
6423 nfs4_state_start_net(struct net *net)
6425 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
6426 int ret;
6428 ret = nfs4_state_create_net(net);
6429 if (ret)
6430 return ret;
6431 nn->boot_time = get_seconds();
6432 nn->grace_ended = false;
6433 locks_start_grace(net, &nn->nfsd4_manager);
6434 nfsd4_client_tracking_init(net);
6435 printk(KERN_INFO "NFSD: starting %ld-second grace period (net %p)\n",
6436 nn->nfsd4_grace, net);
6437 queue_delayed_work(laundry_wq, &nn->laundromat_work, nn->nfsd4_grace * HZ);
6438 return 0;
6441 /* initialization to perform when the nfsd service is started: */
6444 nfs4_state_start(void)
6446 int ret;
6448 ret = set_callback_cred();
6449 if (ret)
6450 return -ENOMEM;
6451 laundry_wq = create_singlethread_workqueue("nfsd4");
6452 if (laundry_wq == NULL) {
6453 ret = -ENOMEM;
6454 goto out_recovery;
6456 ret = nfsd4_create_callback_queue();
6457 if (ret)
6458 goto out_free_laundry;
6460 set_max_delegations();
6462 return 0;
6464 out_free_laundry:
6465 destroy_workqueue(laundry_wq);
6466 out_recovery:
6467 return ret;
6470 void
6471 nfs4_state_shutdown_net(struct net *net)
6473 struct nfs4_delegation *dp = NULL;
6474 struct list_head *pos, *next, reaplist;
6475 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
6477 cancel_delayed_work_sync(&nn->laundromat_work);
6478 locks_end_grace(&nn->nfsd4_manager);
6480 INIT_LIST_HEAD(&reaplist);
6481 spin_lock(&state_lock);
6482 list_for_each_safe(pos, next, &nn->del_recall_lru) {
6483 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
6484 unhash_delegation_locked(dp);
6485 list_add(&dp->dl_recall_lru, &reaplist);
6487 spin_unlock(&state_lock);
6488 list_for_each_safe(pos, next, &reaplist) {
6489 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
6490 list_del_init(&dp->dl_recall_lru);
6491 nfs4_put_deleg_lease(dp->dl_stid.sc_file);
6492 nfs4_put_stid(&dp->dl_stid);
6495 nfsd4_client_tracking_exit(net);
6496 nfs4_state_destroy_net(net);
6499 void
6500 nfs4_state_shutdown(void)
6502 destroy_workqueue(laundry_wq);
6503 nfsd4_destroy_callback_queue();
6506 static void
6507 get_stateid(struct nfsd4_compound_state *cstate, stateid_t *stateid)
6509 if (HAS_STATE_ID(cstate, CURRENT_STATE_ID_FLAG) && CURRENT_STATEID(stateid))
6510 memcpy(stateid, &cstate->current_stateid, sizeof(stateid_t));
6513 static void
6514 put_stateid(struct nfsd4_compound_state *cstate, stateid_t *stateid)
6516 if (cstate->minorversion) {
6517 memcpy(&cstate->current_stateid, stateid, sizeof(stateid_t));
6518 SET_STATE_ID(cstate, CURRENT_STATE_ID_FLAG);
6522 void
6523 clear_current_stateid(struct nfsd4_compound_state *cstate)
6525 CLEAR_STATE_ID(cstate, CURRENT_STATE_ID_FLAG);
6529 * functions to set current state id
6531 void
6532 nfsd4_set_opendowngradestateid(struct nfsd4_compound_state *cstate, struct nfsd4_open_downgrade *odp)
6534 put_stateid(cstate, &odp->od_stateid);
6537 void
6538 nfsd4_set_openstateid(struct nfsd4_compound_state *cstate, struct nfsd4_open *open)
6540 put_stateid(cstate, &open->op_stateid);
6543 void
6544 nfsd4_set_closestateid(struct nfsd4_compound_state *cstate, struct nfsd4_close *close)
6546 put_stateid(cstate, &close->cl_stateid);
6549 void
6550 nfsd4_set_lockstateid(struct nfsd4_compound_state *cstate, struct nfsd4_lock *lock)
6552 put_stateid(cstate, &lock->lk_resp_stateid);
6556 * functions to consume current state id
6559 void
6560 nfsd4_get_opendowngradestateid(struct nfsd4_compound_state *cstate, struct nfsd4_open_downgrade *odp)
6562 get_stateid(cstate, &odp->od_stateid);
6565 void
6566 nfsd4_get_delegreturnstateid(struct nfsd4_compound_state *cstate, struct nfsd4_delegreturn *drp)
6568 get_stateid(cstate, &drp->dr_stateid);
6571 void
6572 nfsd4_get_freestateid(struct nfsd4_compound_state *cstate, struct nfsd4_free_stateid *fsp)
6574 get_stateid(cstate, &fsp->fr_stateid);
6577 void
6578 nfsd4_get_setattrstateid(struct nfsd4_compound_state *cstate, struct nfsd4_setattr *setattr)
6580 get_stateid(cstate, &setattr->sa_stateid);
6583 void
6584 nfsd4_get_closestateid(struct nfsd4_compound_state *cstate, struct nfsd4_close *close)
6586 get_stateid(cstate, &close->cl_stateid);
6589 void
6590 nfsd4_get_lockustateid(struct nfsd4_compound_state *cstate, struct nfsd4_locku *locku)
6592 get_stateid(cstate, &locku->lu_stateid);
6595 void
6596 nfsd4_get_readstateid(struct nfsd4_compound_state *cstate, struct nfsd4_read *read)
6598 get_stateid(cstate, &read->rd_stateid);
6601 void
6602 nfsd4_get_writestateid(struct nfsd4_compound_state *cstate, struct nfsd4_write *write)
6604 get_stateid(cstate, &write->wr_stateid);