nfsd: make delegations shutdown network namespace aware
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / fs / nfsd / nfs4state.c
blobbc2fc9f076fc8b00f0d9a7ad2386c497380757ad
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/clnt.h>
44 #include "xdr4.h"
45 #include "vfs.h"
46 #include "current_stateid.h"
47 #include "fault_inject.h"
49 #include "netns.h"
51 #define NFSDDBG_FACILITY NFSDDBG_PROC
53 /* Globals */
54 time_t nfsd4_lease = 90; /* default lease time */
55 time_t nfsd4_grace = 90;
57 #define all_ones {{~0,~0},~0}
58 static const stateid_t one_stateid = {
59 .si_generation = ~0,
60 .si_opaque = all_ones,
62 static const stateid_t zero_stateid = {
63 /* all fields zero */
65 static const stateid_t currentstateid = {
66 .si_generation = 1,
69 static u64 current_sessionid = 1;
71 #define ZERO_STATEID(stateid) (!memcmp((stateid), &zero_stateid, sizeof(stateid_t)))
72 #define ONE_STATEID(stateid) (!memcmp((stateid), &one_stateid, sizeof(stateid_t)))
73 #define CURRENT_STATEID(stateid) (!memcmp((stateid), &currentstateid, sizeof(stateid_t)))
75 /* forward declarations */
76 static int check_for_locks(struct nfs4_file *filp, struct nfs4_lockowner *lowner);
78 /* Locking: */
80 /* Currently used for almost all code touching nfsv4 state: */
81 static DEFINE_MUTEX(client_mutex);
84 * Currently used for the del_recall_lru and file hash table. In an
85 * effort to decrease the scope of the client_mutex, this spinlock may
86 * eventually cover more:
88 static DEFINE_SPINLOCK(recall_lock);
90 static struct kmem_cache *openowner_slab = NULL;
91 static struct kmem_cache *lockowner_slab = NULL;
92 static struct kmem_cache *file_slab = NULL;
93 static struct kmem_cache *stateid_slab = NULL;
94 static struct kmem_cache *deleg_slab = NULL;
96 void
97 nfs4_lock_state(void)
99 mutex_lock(&client_mutex);
102 static void free_session(struct kref *);
104 /* Must be called under the client_lock */
105 static void nfsd4_put_session_locked(struct nfsd4_session *ses)
107 kref_put(&ses->se_ref, free_session);
110 static void nfsd4_get_session(struct nfsd4_session *ses)
112 kref_get(&ses->se_ref);
115 void
116 nfs4_unlock_state(void)
118 mutex_unlock(&client_mutex);
121 static inline u32
122 opaque_hashval(const void *ptr, int nbytes)
124 unsigned char *cptr = (unsigned char *) ptr;
126 u32 x = 0;
127 while (nbytes--) {
128 x *= 37;
129 x += *cptr++;
131 return x;
134 static struct list_head del_recall_lru;
136 static void nfsd4_free_file(struct nfs4_file *f)
138 kmem_cache_free(file_slab, f);
141 static inline void
142 put_nfs4_file(struct nfs4_file *fi)
144 if (atomic_dec_and_lock(&fi->fi_ref, &recall_lock)) {
145 list_del(&fi->fi_hash);
146 spin_unlock(&recall_lock);
147 iput(fi->fi_inode);
148 nfsd4_free_file(fi);
152 static inline void
153 get_nfs4_file(struct nfs4_file *fi)
155 atomic_inc(&fi->fi_ref);
158 static int num_delegations;
159 unsigned int max_delegations;
162 * Open owner state (share locks)
165 /* hash tables for lock and open owners */
166 #define OWNER_HASH_BITS 8
167 #define OWNER_HASH_SIZE (1 << OWNER_HASH_BITS)
168 #define OWNER_HASH_MASK (OWNER_HASH_SIZE - 1)
170 static unsigned int ownerstr_hashval(u32 clientid, struct xdr_netobj *ownername)
172 unsigned int ret;
174 ret = opaque_hashval(ownername->data, ownername->len);
175 ret += clientid;
176 return ret & OWNER_HASH_MASK;
179 /* hash table for nfs4_file */
180 #define FILE_HASH_BITS 8
181 #define FILE_HASH_SIZE (1 << FILE_HASH_BITS)
183 static unsigned int file_hashval(struct inode *ino)
185 /* XXX: why are we hashing on inode pointer, anyway? */
186 return hash_ptr(ino, FILE_HASH_BITS);
189 static struct list_head file_hashtbl[FILE_HASH_SIZE];
191 static void __nfs4_file_get_access(struct nfs4_file *fp, int oflag)
193 WARN_ON_ONCE(!(fp->fi_fds[oflag] || fp->fi_fds[O_RDWR]));
194 atomic_inc(&fp->fi_access[oflag]);
197 static void nfs4_file_get_access(struct nfs4_file *fp, int oflag)
199 if (oflag == O_RDWR) {
200 __nfs4_file_get_access(fp, O_RDONLY);
201 __nfs4_file_get_access(fp, O_WRONLY);
202 } else
203 __nfs4_file_get_access(fp, oflag);
206 static void nfs4_file_put_fd(struct nfs4_file *fp, int oflag)
208 if (fp->fi_fds[oflag]) {
209 fput(fp->fi_fds[oflag]);
210 fp->fi_fds[oflag] = NULL;
214 static void __nfs4_file_put_access(struct nfs4_file *fp, int oflag)
216 if (atomic_dec_and_test(&fp->fi_access[oflag])) {
217 nfs4_file_put_fd(fp, oflag);
219 * It's also safe to get rid of the RDWR open *if*
220 * we no longer have need of the other kind of access
221 * or if we already have the other kind of open:
223 if (fp->fi_fds[1-oflag]
224 || atomic_read(&fp->fi_access[1 - oflag]) == 0)
225 nfs4_file_put_fd(fp, O_RDWR);
229 static void nfs4_file_put_access(struct nfs4_file *fp, int oflag)
231 if (oflag == O_RDWR) {
232 __nfs4_file_put_access(fp, O_RDONLY);
233 __nfs4_file_put_access(fp, O_WRONLY);
234 } else
235 __nfs4_file_put_access(fp, oflag);
238 static inline int get_new_stid(struct nfs4_stid *stid)
240 static int min_stateid = 0;
241 struct idr *stateids = &stid->sc_client->cl_stateids;
242 int new_stid;
243 int error;
245 error = idr_get_new_above(stateids, stid, min_stateid, &new_stid);
247 * Note: the necessary preallocation was done in
248 * nfs4_alloc_stateid(). The idr code caps the number of
249 * preallocations that can exist at a time, but the state lock
250 * prevents anyone from using ours before we get here:
252 WARN_ON_ONCE(error);
254 * It shouldn't be a problem to reuse an opaque stateid value.
255 * I don't think it is for 4.1. But with 4.0 I worry that, for
256 * example, a stray write retransmission could be accepted by
257 * the server when it should have been rejected. Therefore,
258 * adopt a trick from the sctp code to attempt to maximize the
259 * amount of time until an id is reused, by ensuring they always
260 * "increase" (mod INT_MAX):
263 min_stateid = new_stid+1;
264 if (min_stateid == INT_MAX)
265 min_stateid = 0;
266 return new_stid;
269 static void init_stid(struct nfs4_stid *stid, struct nfs4_client *cl, unsigned char type)
271 stateid_t *s = &stid->sc_stateid;
272 int new_id;
274 stid->sc_type = type;
275 stid->sc_client = cl;
276 s->si_opaque.so_clid = cl->cl_clientid;
277 new_id = get_new_stid(stid);
278 s->si_opaque.so_id = (u32)new_id;
279 /* Will be incremented before return to client: */
280 s->si_generation = 0;
283 static struct nfs4_stid *nfs4_alloc_stid(struct nfs4_client *cl, struct kmem_cache *slab)
285 struct idr *stateids = &cl->cl_stateids;
287 if (!idr_pre_get(stateids, GFP_KERNEL))
288 return NULL;
290 * Note: if we fail here (or any time between now and the time
291 * we actually get the new idr), we won't need to undo the idr
292 * preallocation, since the idr code caps the number of
293 * preallocated entries.
295 return kmem_cache_alloc(slab, GFP_KERNEL);
298 static struct nfs4_ol_stateid * nfs4_alloc_stateid(struct nfs4_client *clp)
300 return openlockstateid(nfs4_alloc_stid(clp, stateid_slab));
303 static struct nfs4_delegation *
304 alloc_init_deleg(struct nfs4_client *clp, struct nfs4_ol_stateid *stp, struct svc_fh *current_fh, u32 type)
306 struct nfs4_delegation *dp;
307 struct nfs4_file *fp = stp->st_file;
309 dprintk("NFSD alloc_init_deleg\n");
311 * Major work on the lease subsystem (for example, to support
312 * calbacks on stat) will be required before we can support
313 * write delegations properly.
315 if (type != NFS4_OPEN_DELEGATE_READ)
316 return NULL;
317 if (fp->fi_had_conflict)
318 return NULL;
319 if (num_delegations > max_delegations)
320 return NULL;
321 dp = delegstateid(nfs4_alloc_stid(clp, deleg_slab));
322 if (dp == NULL)
323 return dp;
324 init_stid(&dp->dl_stid, clp, NFS4_DELEG_STID);
326 * delegation seqid's are never incremented. The 4.1 special
327 * meaning of seqid 0 isn't meaningful, really, but let's avoid
328 * 0 anyway just for consistency and use 1:
330 dp->dl_stid.sc_stateid.si_generation = 1;
331 num_delegations++;
332 INIT_LIST_HEAD(&dp->dl_perfile);
333 INIT_LIST_HEAD(&dp->dl_perclnt);
334 INIT_LIST_HEAD(&dp->dl_recall_lru);
335 get_nfs4_file(fp);
336 dp->dl_file = fp;
337 dp->dl_type = type;
338 fh_copy_shallow(&dp->dl_fh, &current_fh->fh_handle);
339 dp->dl_time = 0;
340 atomic_set(&dp->dl_count, 1);
341 nfsd4_init_callback(&dp->dl_recall);
342 return dp;
345 void
346 nfs4_put_delegation(struct nfs4_delegation *dp)
348 if (atomic_dec_and_test(&dp->dl_count)) {
349 dprintk("NFSD: freeing dp %p\n",dp);
350 put_nfs4_file(dp->dl_file);
351 kmem_cache_free(deleg_slab, dp);
352 num_delegations--;
356 static void nfs4_put_deleg_lease(struct nfs4_file *fp)
358 if (atomic_dec_and_test(&fp->fi_delegees)) {
359 vfs_setlease(fp->fi_deleg_file, F_UNLCK, &fp->fi_lease);
360 fp->fi_lease = NULL;
361 fput(fp->fi_deleg_file);
362 fp->fi_deleg_file = NULL;
366 static void unhash_stid(struct nfs4_stid *s)
368 struct idr *stateids = &s->sc_client->cl_stateids;
370 idr_remove(stateids, s->sc_stateid.si_opaque.so_id);
373 /* Called under the state lock. */
374 static void
375 unhash_delegation(struct nfs4_delegation *dp)
377 unhash_stid(&dp->dl_stid);
378 list_del_init(&dp->dl_perclnt);
379 spin_lock(&recall_lock);
380 list_del_init(&dp->dl_perfile);
381 list_del_init(&dp->dl_recall_lru);
382 spin_unlock(&recall_lock);
383 nfs4_put_deleg_lease(dp->dl_file);
384 nfs4_put_delegation(dp);
388 * SETCLIENTID state
391 static unsigned int clientid_hashval(u32 id)
393 return id & CLIENT_HASH_MASK;
396 static unsigned int clientstr_hashval(const char *name)
398 return opaque_hashval(name, 8) & CLIENT_HASH_MASK;
402 * We store the NONE, READ, WRITE, and BOTH bits separately in the
403 * st_{access,deny}_bmap field of the stateid, in order to track not
404 * only what share bits are currently in force, but also what
405 * combinations of share bits previous opens have used. This allows us
406 * to enforce the recommendation of rfc 3530 14.2.19 that the server
407 * return an error if the client attempt to downgrade to a combination
408 * of share bits not explicable by closing some of its previous opens.
410 * XXX: This enforcement is actually incomplete, since we don't keep
411 * track of access/deny bit combinations; so, e.g., we allow:
413 * OPEN allow read, deny write
414 * OPEN allow both, deny none
415 * DOWNGRADE allow read, deny none
417 * which we should reject.
419 static unsigned int
420 bmap_to_share_mode(unsigned long bmap) {
421 int i;
422 unsigned int access = 0;
424 for (i = 1; i < 4; i++) {
425 if (test_bit(i, &bmap))
426 access |= i;
428 return access;
431 static bool
432 test_share(struct nfs4_ol_stateid *stp, struct nfsd4_open *open) {
433 unsigned int access, deny;
435 access = bmap_to_share_mode(stp->st_access_bmap);
436 deny = bmap_to_share_mode(stp->st_deny_bmap);
437 if ((access & open->op_share_deny) || (deny & open->op_share_access))
438 return false;
439 return true;
442 /* set share access for a given stateid */
443 static inline void
444 set_access(u32 access, struct nfs4_ol_stateid *stp)
446 __set_bit(access, &stp->st_access_bmap);
449 /* clear share access for a given stateid */
450 static inline void
451 clear_access(u32 access, struct nfs4_ol_stateid *stp)
453 __clear_bit(access, &stp->st_access_bmap);
456 /* test whether a given stateid has access */
457 static inline bool
458 test_access(u32 access, struct nfs4_ol_stateid *stp)
460 return test_bit(access, &stp->st_access_bmap);
463 /* set share deny for a given stateid */
464 static inline void
465 set_deny(u32 access, struct nfs4_ol_stateid *stp)
467 __set_bit(access, &stp->st_deny_bmap);
470 /* clear share deny for a given stateid */
471 static inline void
472 clear_deny(u32 access, struct nfs4_ol_stateid *stp)
474 __clear_bit(access, &stp->st_deny_bmap);
477 /* test whether a given stateid is denying specific access */
478 static inline bool
479 test_deny(u32 access, struct nfs4_ol_stateid *stp)
481 return test_bit(access, &stp->st_deny_bmap);
484 static int nfs4_access_to_omode(u32 access)
486 switch (access & NFS4_SHARE_ACCESS_BOTH) {
487 case NFS4_SHARE_ACCESS_READ:
488 return O_RDONLY;
489 case NFS4_SHARE_ACCESS_WRITE:
490 return O_WRONLY;
491 case NFS4_SHARE_ACCESS_BOTH:
492 return O_RDWR;
494 WARN_ON_ONCE(1);
495 return O_RDONLY;
498 /* release all access and file references for a given stateid */
499 static void
500 release_all_access(struct nfs4_ol_stateid *stp)
502 int i;
504 for (i = 1; i < 4; i++) {
505 if (test_access(i, stp))
506 nfs4_file_put_access(stp->st_file,
507 nfs4_access_to_omode(i));
508 clear_access(i, stp);
512 static void unhash_generic_stateid(struct nfs4_ol_stateid *stp)
514 list_del(&stp->st_perfile);
515 list_del(&stp->st_perstateowner);
518 static void close_generic_stateid(struct nfs4_ol_stateid *stp)
520 release_all_access(stp);
521 put_nfs4_file(stp->st_file);
522 stp->st_file = NULL;
525 static void free_generic_stateid(struct nfs4_ol_stateid *stp)
527 kmem_cache_free(stateid_slab, stp);
530 static void release_lock_stateid(struct nfs4_ol_stateid *stp)
532 struct file *file;
534 unhash_generic_stateid(stp);
535 unhash_stid(&stp->st_stid);
536 file = find_any_file(stp->st_file);
537 if (file)
538 locks_remove_posix(file, (fl_owner_t)lockowner(stp->st_stateowner));
539 close_generic_stateid(stp);
540 free_generic_stateid(stp);
543 static void unhash_lockowner(struct nfs4_lockowner *lo)
545 struct nfs4_ol_stateid *stp;
547 list_del(&lo->lo_owner.so_strhash);
548 list_del(&lo->lo_perstateid);
549 list_del(&lo->lo_owner_ino_hash);
550 while (!list_empty(&lo->lo_owner.so_stateids)) {
551 stp = list_first_entry(&lo->lo_owner.so_stateids,
552 struct nfs4_ol_stateid, st_perstateowner);
553 release_lock_stateid(stp);
557 static void release_lockowner(struct nfs4_lockowner *lo)
559 unhash_lockowner(lo);
560 nfs4_free_lockowner(lo);
563 static void
564 release_stateid_lockowners(struct nfs4_ol_stateid *open_stp)
566 struct nfs4_lockowner *lo;
568 while (!list_empty(&open_stp->st_lockowners)) {
569 lo = list_entry(open_stp->st_lockowners.next,
570 struct nfs4_lockowner, lo_perstateid);
571 release_lockowner(lo);
575 static void unhash_open_stateid(struct nfs4_ol_stateid *stp)
577 unhash_generic_stateid(stp);
578 release_stateid_lockowners(stp);
579 close_generic_stateid(stp);
582 static void release_open_stateid(struct nfs4_ol_stateid *stp)
584 unhash_open_stateid(stp);
585 unhash_stid(&stp->st_stid);
586 free_generic_stateid(stp);
589 static void unhash_openowner(struct nfs4_openowner *oo)
591 struct nfs4_ol_stateid *stp;
593 list_del(&oo->oo_owner.so_strhash);
594 list_del(&oo->oo_perclient);
595 while (!list_empty(&oo->oo_owner.so_stateids)) {
596 stp = list_first_entry(&oo->oo_owner.so_stateids,
597 struct nfs4_ol_stateid, st_perstateowner);
598 release_open_stateid(stp);
602 static void release_last_closed_stateid(struct nfs4_openowner *oo)
604 struct nfs4_ol_stateid *s = oo->oo_last_closed_stid;
606 if (s) {
607 unhash_stid(&s->st_stid);
608 free_generic_stateid(s);
609 oo->oo_last_closed_stid = NULL;
613 static void release_openowner(struct nfs4_openowner *oo)
615 unhash_openowner(oo);
616 list_del(&oo->oo_close_lru);
617 release_last_closed_stateid(oo);
618 nfs4_free_openowner(oo);
621 static inline int
622 hash_sessionid(struct nfs4_sessionid *sessionid)
624 struct nfsd4_sessionid *sid = (struct nfsd4_sessionid *)sessionid;
626 return sid->sequence % SESSION_HASH_SIZE;
629 #ifdef NFSD_DEBUG
630 static inline void
631 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
633 u32 *ptr = (u32 *)(&sessionid->data[0]);
634 dprintk("%s: %u:%u:%u:%u\n", fn, ptr[0], ptr[1], ptr[2], ptr[3]);
636 #else
637 static inline void
638 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
641 #endif
644 static void
645 gen_sessionid(struct nfsd4_session *ses)
647 struct nfs4_client *clp = ses->se_client;
648 struct nfsd4_sessionid *sid;
650 sid = (struct nfsd4_sessionid *)ses->se_sessionid.data;
651 sid->clientid = clp->cl_clientid;
652 sid->sequence = current_sessionid++;
653 sid->reserved = 0;
657 * The protocol defines ca_maxresponssize_cached to include the size of
658 * the rpc header, but all we need to cache is the data starting after
659 * the end of the initial SEQUENCE operation--the rest we regenerate
660 * each time. Therefore we can advertise a ca_maxresponssize_cached
661 * value that is the number of bytes in our cache plus a few additional
662 * bytes. In order to stay on the safe side, and not promise more than
663 * we can cache, those additional bytes must be the minimum possible: 24
664 * bytes of rpc header (xid through accept state, with AUTH_NULL
665 * verifier), 12 for the compound header (with zero-length tag), and 44
666 * for the SEQUENCE op response:
668 #define NFSD_MIN_HDR_SEQ_SZ (24 + 12 + 44)
670 static void
671 free_session_slots(struct nfsd4_session *ses)
673 int i;
675 for (i = 0; i < ses->se_fchannel.maxreqs; i++)
676 kfree(ses->se_slots[i]);
680 * We don't actually need to cache the rpc and session headers, so we
681 * can allocate a little less for each slot:
683 static inline int slot_bytes(struct nfsd4_channel_attrs *ca)
685 return ca->maxresp_cached - NFSD_MIN_HDR_SEQ_SZ;
688 static int nfsd4_sanitize_slot_size(u32 size)
690 size -= NFSD_MIN_HDR_SEQ_SZ; /* We don't cache the rpc header */
691 size = min_t(u32, size, NFSD_SLOT_CACHE_SIZE);
693 return size;
697 * XXX: If we run out of reserved DRC memory we could (up to a point)
698 * re-negotiate active sessions and reduce their slot usage to make
699 * room for new connections. For now we just fail the create session.
701 static int nfsd4_get_drc_mem(int slotsize, u32 num)
703 int avail;
705 num = min_t(u32, num, NFSD_MAX_SLOTS_PER_SESSION);
707 spin_lock(&nfsd_drc_lock);
708 avail = min_t(int, NFSD_MAX_MEM_PER_SESSION,
709 nfsd_drc_max_mem - nfsd_drc_mem_used);
710 num = min_t(int, num, avail / slotsize);
711 nfsd_drc_mem_used += num * slotsize;
712 spin_unlock(&nfsd_drc_lock);
714 return num;
717 static void nfsd4_put_drc_mem(int slotsize, int num)
719 spin_lock(&nfsd_drc_lock);
720 nfsd_drc_mem_used -= slotsize * num;
721 spin_unlock(&nfsd_drc_lock);
724 static struct nfsd4_session *__alloc_session(int slotsize, int numslots)
726 struct nfsd4_session *new;
727 int mem, i;
729 BUILD_BUG_ON(NFSD_MAX_SLOTS_PER_SESSION * sizeof(struct nfsd4_slot *)
730 + sizeof(struct nfsd4_session) > PAGE_SIZE);
731 mem = numslots * sizeof(struct nfsd4_slot *);
733 new = kzalloc(sizeof(*new) + mem, GFP_KERNEL);
734 if (!new)
735 return NULL;
736 /* allocate each struct nfsd4_slot and data cache in one piece */
737 for (i = 0; i < numslots; i++) {
738 mem = sizeof(struct nfsd4_slot) + slotsize;
739 new->se_slots[i] = kzalloc(mem, GFP_KERNEL);
740 if (!new->se_slots[i])
741 goto out_free;
743 return new;
744 out_free:
745 while (i--)
746 kfree(new->se_slots[i]);
747 kfree(new);
748 return NULL;
751 static void init_forechannel_attrs(struct nfsd4_channel_attrs *new, struct nfsd4_channel_attrs *req, int numslots, int slotsize)
753 u32 maxrpc = nfsd_serv->sv_max_mesg;
755 new->maxreqs = numslots;
756 new->maxresp_cached = min_t(u32, req->maxresp_cached,
757 slotsize + NFSD_MIN_HDR_SEQ_SZ);
758 new->maxreq_sz = min_t(u32, req->maxreq_sz, maxrpc);
759 new->maxresp_sz = min_t(u32, req->maxresp_sz, maxrpc);
760 new->maxops = min_t(u32, req->maxops, NFSD_MAX_OPS_PER_COMPOUND);
763 static void free_conn(struct nfsd4_conn *c)
765 svc_xprt_put(c->cn_xprt);
766 kfree(c);
769 static void nfsd4_conn_lost(struct svc_xpt_user *u)
771 struct nfsd4_conn *c = container_of(u, struct nfsd4_conn, cn_xpt_user);
772 struct nfs4_client *clp = c->cn_session->se_client;
774 spin_lock(&clp->cl_lock);
775 if (!list_empty(&c->cn_persession)) {
776 list_del(&c->cn_persession);
777 free_conn(c);
779 spin_unlock(&clp->cl_lock);
780 nfsd4_probe_callback(clp);
783 static struct nfsd4_conn *alloc_conn(struct svc_rqst *rqstp, u32 flags)
785 struct nfsd4_conn *conn;
787 conn = kmalloc(sizeof(struct nfsd4_conn), GFP_KERNEL);
788 if (!conn)
789 return NULL;
790 svc_xprt_get(rqstp->rq_xprt);
791 conn->cn_xprt = rqstp->rq_xprt;
792 conn->cn_flags = flags;
793 INIT_LIST_HEAD(&conn->cn_xpt_user.list);
794 return conn;
797 static void __nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
799 conn->cn_session = ses;
800 list_add(&conn->cn_persession, &ses->se_conns);
803 static void nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
805 struct nfs4_client *clp = ses->se_client;
807 spin_lock(&clp->cl_lock);
808 __nfsd4_hash_conn(conn, ses);
809 spin_unlock(&clp->cl_lock);
812 static int nfsd4_register_conn(struct nfsd4_conn *conn)
814 conn->cn_xpt_user.callback = nfsd4_conn_lost;
815 return register_xpt_user(conn->cn_xprt, &conn->cn_xpt_user);
818 static void nfsd4_init_conn(struct svc_rqst *rqstp, struct nfsd4_conn *conn, struct nfsd4_session *ses)
820 int ret;
822 nfsd4_hash_conn(conn, ses);
823 ret = nfsd4_register_conn(conn);
824 if (ret)
825 /* oops; xprt is already down: */
826 nfsd4_conn_lost(&conn->cn_xpt_user);
827 if (conn->cn_flags & NFS4_CDFC4_BACK) {
828 /* callback channel may be back up */
829 nfsd4_probe_callback(ses->se_client);
833 static struct nfsd4_conn *alloc_conn_from_crses(struct svc_rqst *rqstp, struct nfsd4_create_session *cses)
835 u32 dir = NFS4_CDFC4_FORE;
837 if (cses->flags & SESSION4_BACK_CHAN)
838 dir |= NFS4_CDFC4_BACK;
839 return alloc_conn(rqstp, dir);
842 /* must be called under client_lock */
843 static void nfsd4_del_conns(struct nfsd4_session *s)
845 struct nfs4_client *clp = s->se_client;
846 struct nfsd4_conn *c;
848 spin_lock(&clp->cl_lock);
849 while (!list_empty(&s->se_conns)) {
850 c = list_first_entry(&s->se_conns, struct nfsd4_conn, cn_persession);
851 list_del_init(&c->cn_persession);
852 spin_unlock(&clp->cl_lock);
854 unregister_xpt_user(c->cn_xprt, &c->cn_xpt_user);
855 free_conn(c);
857 spin_lock(&clp->cl_lock);
859 spin_unlock(&clp->cl_lock);
862 static void __free_session(struct nfsd4_session *ses)
864 nfsd4_put_drc_mem(slot_bytes(&ses->se_fchannel), ses->se_fchannel.maxreqs);
865 free_session_slots(ses);
866 kfree(ses);
869 static void free_session(struct kref *kref)
871 struct nfsd4_session *ses;
872 struct nfsd_net *nn;
874 ses = container_of(kref, struct nfsd4_session, se_ref);
875 nn = net_generic(ses->se_client->net, nfsd_net_id);
877 lockdep_assert_held(&nn->client_lock);
878 nfsd4_del_conns(ses);
879 __free_session(ses);
882 void nfsd4_put_session(struct nfsd4_session *ses)
884 struct nfsd_net *nn = net_generic(ses->se_client->net, nfsd_net_id);
886 spin_lock(&nn->client_lock);
887 nfsd4_put_session_locked(ses);
888 spin_unlock(&nn->client_lock);
891 static struct nfsd4_session *alloc_session(struct nfsd4_channel_attrs *fchan)
893 struct nfsd4_session *new;
894 int numslots, slotsize;
896 * Note decreasing slot size below client's request may
897 * make it difficult for client to function correctly, whereas
898 * decreasing the number of slots will (just?) affect
899 * performance. When short on memory we therefore prefer to
900 * decrease number of slots instead of their size.
902 slotsize = nfsd4_sanitize_slot_size(fchan->maxresp_cached);
903 numslots = nfsd4_get_drc_mem(slotsize, fchan->maxreqs);
904 if (numslots < 1)
905 return NULL;
907 new = __alloc_session(slotsize, numslots);
908 if (!new) {
909 nfsd4_put_drc_mem(slotsize, fchan->maxreqs);
910 return NULL;
912 init_forechannel_attrs(&new->se_fchannel, fchan, numslots, slotsize);
913 return new;
916 static void init_session(struct svc_rqst *rqstp, struct nfsd4_session *new, struct nfs4_client *clp, struct nfsd4_create_session *cses)
918 int idx;
919 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
921 new->se_client = clp;
922 gen_sessionid(new);
924 INIT_LIST_HEAD(&new->se_conns);
926 new->se_cb_seq_nr = 1;
927 new->se_flags = cses->flags;
928 new->se_cb_prog = cses->callback_prog;
929 new->se_cb_sec = cses->cb_sec;
930 kref_init(&new->se_ref);
931 idx = hash_sessionid(&new->se_sessionid);
932 spin_lock(&nn->client_lock);
933 list_add(&new->se_hash, &nn->sessionid_hashtbl[idx]);
934 spin_lock(&clp->cl_lock);
935 list_add(&new->se_perclnt, &clp->cl_sessions);
936 spin_unlock(&clp->cl_lock);
937 spin_unlock(&nn->client_lock);
939 if (cses->flags & SESSION4_BACK_CHAN) {
940 struct sockaddr *sa = svc_addr(rqstp);
942 * This is a little silly; with sessions there's no real
943 * use for the callback address. Use the peer address
944 * as a reasonable default for now, but consider fixing
945 * the rpc client not to require an address in the
946 * future:
948 rpc_copy_addr((struct sockaddr *)&clp->cl_cb_conn.cb_addr, sa);
949 clp->cl_cb_conn.cb_addrlen = svc_addr_len(sa);
953 /* caller must hold client_lock */
954 static struct nfsd4_session *
955 find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid, struct net *net)
957 struct nfsd4_session *elem;
958 int idx;
959 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
961 dump_sessionid(__func__, sessionid);
962 idx = hash_sessionid(sessionid);
963 /* Search in the appropriate list */
964 list_for_each_entry(elem, &nn->sessionid_hashtbl[idx], se_hash) {
965 if (!memcmp(elem->se_sessionid.data, sessionid->data,
966 NFS4_MAX_SESSIONID_LEN)) {
967 return elem;
971 dprintk("%s: session not found\n", __func__);
972 return NULL;
975 /* caller must hold client_lock */
976 static void
977 unhash_session(struct nfsd4_session *ses)
979 list_del(&ses->se_hash);
980 spin_lock(&ses->se_client->cl_lock);
981 list_del(&ses->se_perclnt);
982 spin_unlock(&ses->se_client->cl_lock);
985 /* must be called under the client_lock */
986 static inline void
987 renew_client_locked(struct nfs4_client *clp)
989 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
991 if (is_client_expired(clp)) {
992 WARN_ON(1);
993 printk("%s: client (clientid %08x/%08x) already expired\n",
994 __func__,
995 clp->cl_clientid.cl_boot,
996 clp->cl_clientid.cl_id);
997 return;
1000 dprintk("renewing client (clientid %08x/%08x)\n",
1001 clp->cl_clientid.cl_boot,
1002 clp->cl_clientid.cl_id);
1003 list_move_tail(&clp->cl_lru, &nn->client_lru);
1004 clp->cl_time = get_seconds();
1007 static inline void
1008 renew_client(struct nfs4_client *clp)
1010 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1012 spin_lock(&nn->client_lock);
1013 renew_client_locked(clp);
1014 spin_unlock(&nn->client_lock);
1017 /* SETCLIENTID and SETCLIENTID_CONFIRM Helper functions */
1018 static int
1019 STALE_CLIENTID(clientid_t *clid, struct nfsd_net *nn)
1021 if (clid->cl_boot == nn->boot_time)
1022 return 0;
1023 dprintk("NFSD stale clientid (%08x/%08x) boot_time %08lx\n",
1024 clid->cl_boot, clid->cl_id, nn->boot_time);
1025 return 1;
1029 * XXX Should we use a slab cache ?
1030 * This type of memory management is somewhat inefficient, but we use it
1031 * anyway since SETCLIENTID is not a common operation.
1033 static struct nfs4_client *alloc_client(struct xdr_netobj name)
1035 struct nfs4_client *clp;
1037 clp = kzalloc(sizeof(struct nfs4_client), GFP_KERNEL);
1038 if (clp == NULL)
1039 return NULL;
1040 clp->cl_name.data = kmemdup(name.data, name.len, GFP_KERNEL);
1041 if (clp->cl_name.data == NULL) {
1042 kfree(clp);
1043 return NULL;
1045 clp->cl_name.len = name.len;
1046 return clp;
1049 static inline void
1050 free_client(struct nfs4_client *clp)
1052 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1054 lockdep_assert_held(&nn->client_lock);
1055 while (!list_empty(&clp->cl_sessions)) {
1056 struct nfsd4_session *ses;
1057 ses = list_entry(clp->cl_sessions.next, struct nfsd4_session,
1058 se_perclnt);
1059 list_del(&ses->se_perclnt);
1060 nfsd4_put_session_locked(ses);
1062 free_svc_cred(&clp->cl_cred);
1063 kfree(clp->cl_name.data);
1064 kfree(clp);
1067 void
1068 release_session_client(struct nfsd4_session *session)
1070 struct nfs4_client *clp = session->se_client;
1071 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1073 if (!atomic_dec_and_lock(&clp->cl_refcount, &nn->client_lock))
1074 return;
1075 if (is_client_expired(clp)) {
1076 free_client(clp);
1077 session->se_client = NULL;
1078 } else
1079 renew_client_locked(clp);
1080 spin_unlock(&nn->client_lock);
1083 /* must be called under the client_lock */
1084 static inline void
1085 unhash_client_locked(struct nfs4_client *clp)
1087 struct nfsd4_session *ses;
1089 mark_client_expired(clp);
1090 list_del(&clp->cl_lru);
1091 spin_lock(&clp->cl_lock);
1092 list_for_each_entry(ses, &clp->cl_sessions, se_perclnt)
1093 list_del_init(&ses->se_hash);
1094 spin_unlock(&clp->cl_lock);
1097 static void
1098 destroy_client(struct nfs4_client *clp)
1100 struct nfs4_openowner *oo;
1101 struct nfs4_delegation *dp;
1102 struct list_head reaplist;
1103 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1105 INIT_LIST_HEAD(&reaplist);
1106 spin_lock(&recall_lock);
1107 while (!list_empty(&clp->cl_delegations)) {
1108 dp = list_entry(clp->cl_delegations.next, struct nfs4_delegation, dl_perclnt);
1109 list_del_init(&dp->dl_perclnt);
1110 list_move(&dp->dl_recall_lru, &reaplist);
1112 spin_unlock(&recall_lock);
1113 while (!list_empty(&reaplist)) {
1114 dp = list_entry(reaplist.next, struct nfs4_delegation, dl_recall_lru);
1115 unhash_delegation(dp);
1117 while (!list_empty(&clp->cl_openowners)) {
1118 oo = list_entry(clp->cl_openowners.next, struct nfs4_openowner, oo_perclient);
1119 release_openowner(oo);
1121 nfsd4_shutdown_callback(clp);
1122 if (clp->cl_cb_conn.cb_xprt)
1123 svc_xprt_put(clp->cl_cb_conn.cb_xprt);
1124 list_del(&clp->cl_idhash);
1125 if (test_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags))
1126 rb_erase(&clp->cl_namenode, &nn->conf_name_tree);
1127 else
1128 rb_erase(&clp->cl_namenode, &nn->unconf_name_tree);
1129 spin_lock(&nn->client_lock);
1130 unhash_client_locked(clp);
1131 if (atomic_read(&clp->cl_refcount) == 0)
1132 free_client(clp);
1133 spin_unlock(&nn->client_lock);
1136 static void expire_client(struct nfs4_client *clp)
1138 nfsd4_client_record_remove(clp);
1139 destroy_client(clp);
1142 static void copy_verf(struct nfs4_client *target, nfs4_verifier *source)
1144 memcpy(target->cl_verifier.data, source->data,
1145 sizeof(target->cl_verifier.data));
1148 static void copy_clid(struct nfs4_client *target, struct nfs4_client *source)
1150 target->cl_clientid.cl_boot = source->cl_clientid.cl_boot;
1151 target->cl_clientid.cl_id = source->cl_clientid.cl_id;
1154 static int copy_cred(struct svc_cred *target, struct svc_cred *source)
1156 if (source->cr_principal) {
1157 target->cr_principal =
1158 kstrdup(source->cr_principal, GFP_KERNEL);
1159 if (target->cr_principal == NULL)
1160 return -ENOMEM;
1161 } else
1162 target->cr_principal = NULL;
1163 target->cr_flavor = source->cr_flavor;
1164 target->cr_uid = source->cr_uid;
1165 target->cr_gid = source->cr_gid;
1166 target->cr_group_info = source->cr_group_info;
1167 get_group_info(target->cr_group_info);
1168 return 0;
1171 static long long
1172 compare_blob(const struct xdr_netobj *o1, const struct xdr_netobj *o2)
1174 long long res;
1176 res = o1->len - o2->len;
1177 if (res)
1178 return res;
1179 return (long long)memcmp(o1->data, o2->data, o1->len);
1182 static int same_name(const char *n1, const char *n2)
1184 return 0 == memcmp(n1, n2, HEXDIR_LEN);
1187 static int
1188 same_verf(nfs4_verifier *v1, nfs4_verifier *v2)
1190 return 0 == memcmp(v1->data, v2->data, sizeof(v1->data));
1193 static int
1194 same_clid(clientid_t *cl1, clientid_t *cl2)
1196 return (cl1->cl_boot == cl2->cl_boot) && (cl1->cl_id == cl2->cl_id);
1199 static bool groups_equal(struct group_info *g1, struct group_info *g2)
1201 int i;
1203 if (g1->ngroups != g2->ngroups)
1204 return false;
1205 for (i=0; i<g1->ngroups; i++)
1206 if (GROUP_AT(g1, i) != GROUP_AT(g2, i))
1207 return false;
1208 return true;
1212 * RFC 3530 language requires clid_inuse be returned when the
1213 * "principal" associated with a requests differs from that previously
1214 * used. We use uid, gid's, and gss principal string as our best
1215 * approximation. We also don't want to allow non-gss use of a client
1216 * established using gss: in theory cr_principal should catch that
1217 * change, but in practice cr_principal can be null even in the gss case
1218 * since gssd doesn't always pass down a principal string.
1220 static bool is_gss_cred(struct svc_cred *cr)
1222 /* Is cr_flavor one of the gss "pseudoflavors"?: */
1223 return (cr->cr_flavor > RPC_AUTH_MAXFLAVOR);
1227 static bool
1228 same_creds(struct svc_cred *cr1, struct svc_cred *cr2)
1230 if ((is_gss_cred(cr1) != is_gss_cred(cr2))
1231 || (cr1->cr_uid != cr2->cr_uid)
1232 || (cr1->cr_gid != cr2->cr_gid)
1233 || !groups_equal(cr1->cr_group_info, cr2->cr_group_info))
1234 return false;
1235 if (cr1->cr_principal == cr2->cr_principal)
1236 return true;
1237 if (!cr1->cr_principal || !cr2->cr_principal)
1238 return false;
1239 return 0 == strcmp(cr1->cr_principal, cr2->cr_principal);
1242 static void gen_clid(struct nfs4_client *clp, struct nfsd_net *nn)
1244 static u32 current_clientid = 1;
1246 clp->cl_clientid.cl_boot = nn->boot_time;
1247 clp->cl_clientid.cl_id = current_clientid++;
1250 static void gen_confirm(struct nfs4_client *clp)
1252 __be32 verf[2];
1253 static u32 i;
1255 verf[0] = (__be32)get_seconds();
1256 verf[1] = (__be32)i++;
1257 memcpy(clp->cl_confirm.data, verf, sizeof(clp->cl_confirm.data));
1260 static struct nfs4_stid *find_stateid(struct nfs4_client *cl, stateid_t *t)
1262 return idr_find(&cl->cl_stateids, t->si_opaque.so_id);
1265 static struct nfs4_stid *find_stateid_by_type(struct nfs4_client *cl, stateid_t *t, char typemask)
1267 struct nfs4_stid *s;
1269 s = find_stateid(cl, t);
1270 if (!s)
1271 return NULL;
1272 if (typemask & s->sc_type)
1273 return s;
1274 return NULL;
1277 static struct nfs4_client *create_client(struct xdr_netobj name,
1278 struct svc_rqst *rqstp, nfs4_verifier *verf)
1280 struct nfs4_client *clp;
1281 struct sockaddr *sa = svc_addr(rqstp);
1282 int ret;
1283 struct net *net = SVC_NET(rqstp);
1284 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
1286 clp = alloc_client(name);
1287 if (clp == NULL)
1288 return NULL;
1290 INIT_LIST_HEAD(&clp->cl_sessions);
1291 ret = copy_cred(&clp->cl_cred, &rqstp->rq_cred);
1292 if (ret) {
1293 spin_lock(&nn->client_lock);
1294 free_client(clp);
1295 spin_unlock(&nn->client_lock);
1296 return NULL;
1298 idr_init(&clp->cl_stateids);
1299 atomic_set(&clp->cl_refcount, 0);
1300 clp->cl_cb_state = NFSD4_CB_UNKNOWN;
1301 INIT_LIST_HEAD(&clp->cl_idhash);
1302 INIT_LIST_HEAD(&clp->cl_openowners);
1303 INIT_LIST_HEAD(&clp->cl_delegations);
1304 INIT_LIST_HEAD(&clp->cl_lru);
1305 INIT_LIST_HEAD(&clp->cl_callbacks);
1306 spin_lock_init(&clp->cl_lock);
1307 nfsd4_init_callback(&clp->cl_cb_null);
1308 clp->cl_time = get_seconds();
1309 clear_bit(0, &clp->cl_cb_slot_busy);
1310 rpc_init_wait_queue(&clp->cl_cb_waitq, "Backchannel slot table");
1311 copy_verf(clp, verf);
1312 rpc_copy_addr((struct sockaddr *) &clp->cl_addr, sa);
1313 gen_confirm(clp);
1314 clp->cl_cb_session = NULL;
1315 clp->net = net;
1316 return clp;
1319 static void
1320 add_clp_to_name_tree(struct nfs4_client *new_clp, struct rb_root *root)
1322 struct rb_node **new = &(root->rb_node), *parent = NULL;
1323 struct nfs4_client *clp;
1325 while (*new) {
1326 clp = rb_entry(*new, struct nfs4_client, cl_namenode);
1327 parent = *new;
1329 if (compare_blob(&clp->cl_name, &new_clp->cl_name) > 0)
1330 new = &((*new)->rb_left);
1331 else
1332 new = &((*new)->rb_right);
1335 rb_link_node(&new_clp->cl_namenode, parent, new);
1336 rb_insert_color(&new_clp->cl_namenode, root);
1339 static struct nfs4_client *
1340 find_clp_in_name_tree(struct xdr_netobj *name, struct rb_root *root)
1342 long long cmp;
1343 struct rb_node *node = root->rb_node;
1344 struct nfs4_client *clp;
1346 while (node) {
1347 clp = rb_entry(node, struct nfs4_client, cl_namenode);
1348 cmp = compare_blob(&clp->cl_name, name);
1349 if (cmp > 0)
1350 node = node->rb_left;
1351 else if (cmp < 0)
1352 node = node->rb_right;
1353 else
1354 return clp;
1356 return NULL;
1359 static void
1360 add_to_unconfirmed(struct nfs4_client *clp)
1362 unsigned int idhashval;
1363 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1365 clear_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags);
1366 add_clp_to_name_tree(clp, &nn->unconf_name_tree);
1367 idhashval = clientid_hashval(clp->cl_clientid.cl_id);
1368 list_add(&clp->cl_idhash, &nn->unconf_id_hashtbl[idhashval]);
1369 renew_client(clp);
1372 static void
1373 move_to_confirmed(struct nfs4_client *clp)
1375 unsigned int idhashval = clientid_hashval(clp->cl_clientid.cl_id);
1376 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1378 dprintk("NFSD: move_to_confirm nfs4_client %p\n", clp);
1379 list_move(&clp->cl_idhash, &nn->conf_id_hashtbl[idhashval]);
1380 rb_erase(&clp->cl_namenode, &nn->unconf_name_tree);
1381 add_clp_to_name_tree(clp, &nn->conf_name_tree);
1382 set_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags);
1383 renew_client(clp);
1386 static struct nfs4_client *
1387 find_confirmed_client(clientid_t *clid, bool sessions, struct nfsd_net *nn)
1389 struct nfs4_client *clp;
1390 unsigned int idhashval = clientid_hashval(clid->cl_id);
1392 list_for_each_entry(clp, &nn->conf_id_hashtbl[idhashval], cl_idhash) {
1393 if (same_clid(&clp->cl_clientid, clid)) {
1394 if ((bool)clp->cl_minorversion != sessions)
1395 return NULL;
1396 renew_client(clp);
1397 return clp;
1400 return NULL;
1403 static struct nfs4_client *
1404 find_unconfirmed_client(clientid_t *clid, bool sessions, struct nfsd_net *nn)
1406 struct nfs4_client *clp;
1407 unsigned int idhashval = clientid_hashval(clid->cl_id);
1409 list_for_each_entry(clp, &nn->unconf_id_hashtbl[idhashval], cl_idhash) {
1410 if (same_clid(&clp->cl_clientid, clid)) {
1411 if ((bool)clp->cl_minorversion != sessions)
1412 return NULL;
1413 return clp;
1416 return NULL;
1419 static bool clp_used_exchangeid(struct nfs4_client *clp)
1421 return clp->cl_exchange_flags != 0;
1424 static struct nfs4_client *
1425 find_confirmed_client_by_name(struct xdr_netobj *name, struct nfsd_net *nn)
1427 return find_clp_in_name_tree(name, &nn->conf_name_tree);
1430 static struct nfs4_client *
1431 find_unconfirmed_client_by_name(struct xdr_netobj *name, struct nfsd_net *nn)
1433 return find_clp_in_name_tree(name, &nn->unconf_name_tree);
1436 static void
1437 gen_callback(struct nfs4_client *clp, struct nfsd4_setclientid *se, struct svc_rqst *rqstp)
1439 struct nfs4_cb_conn *conn = &clp->cl_cb_conn;
1440 struct sockaddr *sa = svc_addr(rqstp);
1441 u32 scopeid = rpc_get_scope_id(sa);
1442 unsigned short expected_family;
1444 /* Currently, we only support tcp and tcp6 for the callback channel */
1445 if (se->se_callback_netid_len == 3 &&
1446 !memcmp(se->se_callback_netid_val, "tcp", 3))
1447 expected_family = AF_INET;
1448 else if (se->se_callback_netid_len == 4 &&
1449 !memcmp(se->se_callback_netid_val, "tcp6", 4))
1450 expected_family = AF_INET6;
1451 else
1452 goto out_err;
1454 conn->cb_addrlen = rpc_uaddr2sockaddr(clp->net, se->se_callback_addr_val,
1455 se->se_callback_addr_len,
1456 (struct sockaddr *)&conn->cb_addr,
1457 sizeof(conn->cb_addr));
1459 if (!conn->cb_addrlen || conn->cb_addr.ss_family != expected_family)
1460 goto out_err;
1462 if (conn->cb_addr.ss_family == AF_INET6)
1463 ((struct sockaddr_in6 *)&conn->cb_addr)->sin6_scope_id = scopeid;
1465 conn->cb_prog = se->se_callback_prog;
1466 conn->cb_ident = se->se_callback_ident;
1467 memcpy(&conn->cb_saddr, &rqstp->rq_daddr, rqstp->rq_daddrlen);
1468 return;
1469 out_err:
1470 conn->cb_addr.ss_family = AF_UNSPEC;
1471 conn->cb_addrlen = 0;
1472 dprintk(KERN_INFO "NFSD: this client (clientid %08x/%08x) "
1473 "will not receive delegations\n",
1474 clp->cl_clientid.cl_boot, clp->cl_clientid.cl_id);
1476 return;
1480 * Cache a reply. nfsd4_check_drc_limit() has bounded the cache size.
1482 void
1483 nfsd4_store_cache_entry(struct nfsd4_compoundres *resp)
1485 struct nfsd4_slot *slot = resp->cstate.slot;
1486 unsigned int base;
1488 dprintk("--> %s slot %p\n", __func__, slot);
1490 slot->sl_opcnt = resp->opcnt;
1491 slot->sl_status = resp->cstate.status;
1493 slot->sl_flags |= NFSD4_SLOT_INITIALIZED;
1494 if (nfsd4_not_cached(resp)) {
1495 slot->sl_datalen = 0;
1496 return;
1498 slot->sl_datalen = (char *)resp->p - (char *)resp->cstate.datap;
1499 base = (char *)resp->cstate.datap -
1500 (char *)resp->xbuf->head[0].iov_base;
1501 if (read_bytes_from_xdr_buf(resp->xbuf, base, slot->sl_data,
1502 slot->sl_datalen))
1503 WARN("%s: sessions DRC could not cache compound\n", __func__);
1504 return;
1508 * Encode the replay sequence operation from the slot values.
1509 * If cachethis is FALSE encode the uncached rep error on the next
1510 * operation which sets resp->p and increments resp->opcnt for
1511 * nfs4svc_encode_compoundres.
1514 static __be32
1515 nfsd4_enc_sequence_replay(struct nfsd4_compoundargs *args,
1516 struct nfsd4_compoundres *resp)
1518 struct nfsd4_op *op;
1519 struct nfsd4_slot *slot = resp->cstate.slot;
1521 /* Encode the replayed sequence operation */
1522 op = &args->ops[resp->opcnt - 1];
1523 nfsd4_encode_operation(resp, op);
1525 /* Return nfserr_retry_uncached_rep in next operation. */
1526 if (args->opcnt > 1 && !(slot->sl_flags & NFSD4_SLOT_CACHETHIS)) {
1527 op = &args->ops[resp->opcnt++];
1528 op->status = nfserr_retry_uncached_rep;
1529 nfsd4_encode_operation(resp, op);
1531 return op->status;
1535 * The sequence operation is not cached because we can use the slot and
1536 * session values.
1538 __be32
1539 nfsd4_replay_cache_entry(struct nfsd4_compoundres *resp,
1540 struct nfsd4_sequence *seq)
1542 struct nfsd4_slot *slot = resp->cstate.slot;
1543 __be32 status;
1545 dprintk("--> %s slot %p\n", __func__, slot);
1547 /* Either returns 0 or nfserr_retry_uncached */
1548 status = nfsd4_enc_sequence_replay(resp->rqstp->rq_argp, resp);
1549 if (status == nfserr_retry_uncached_rep)
1550 return status;
1552 /* The sequence operation has been encoded, cstate->datap set. */
1553 memcpy(resp->cstate.datap, slot->sl_data, slot->sl_datalen);
1555 resp->opcnt = slot->sl_opcnt;
1556 resp->p = resp->cstate.datap + XDR_QUADLEN(slot->sl_datalen);
1557 status = slot->sl_status;
1559 return status;
1563 * Set the exchange_id flags returned by the server.
1565 static void
1566 nfsd4_set_ex_flags(struct nfs4_client *new, struct nfsd4_exchange_id *clid)
1568 /* pNFS is not supported */
1569 new->cl_exchange_flags |= EXCHGID4_FLAG_USE_NON_PNFS;
1571 /* Referrals are supported, Migration is not. */
1572 new->cl_exchange_flags |= EXCHGID4_FLAG_SUPP_MOVED_REFER;
1574 /* set the wire flags to return to client. */
1575 clid->flags = new->cl_exchange_flags;
1578 static bool client_has_state(struct nfs4_client *clp)
1581 * Note clp->cl_openowners check isn't quite right: there's no
1582 * need to count owners without stateid's.
1584 * Also note we should probably be using this in 4.0 case too.
1586 return !list_empty(&clp->cl_openowners)
1587 || !list_empty(&clp->cl_delegations)
1588 || !list_empty(&clp->cl_sessions);
1591 __be32
1592 nfsd4_exchange_id(struct svc_rqst *rqstp,
1593 struct nfsd4_compound_state *cstate,
1594 struct nfsd4_exchange_id *exid)
1596 struct nfs4_client *unconf, *conf, *new;
1597 __be32 status;
1598 char addr_str[INET6_ADDRSTRLEN];
1599 nfs4_verifier verf = exid->verifier;
1600 struct sockaddr *sa = svc_addr(rqstp);
1601 bool update = exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A;
1602 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
1604 rpc_ntop(sa, addr_str, sizeof(addr_str));
1605 dprintk("%s rqstp=%p exid=%p clname.len=%u clname.data=%p "
1606 "ip_addr=%s flags %x, spa_how %d\n",
1607 __func__, rqstp, exid, exid->clname.len, exid->clname.data,
1608 addr_str, exid->flags, exid->spa_how);
1610 if (exid->flags & ~EXCHGID4_FLAG_MASK_A)
1611 return nfserr_inval;
1613 /* Currently only support SP4_NONE */
1614 switch (exid->spa_how) {
1615 case SP4_NONE:
1616 break;
1617 default: /* checked by xdr code */
1618 WARN_ON_ONCE(1);
1619 case SP4_SSV:
1620 case SP4_MACH_CRED:
1621 return nfserr_serverfault; /* no excuse :-/ */
1624 /* Cases below refer to rfc 5661 section 18.35.4: */
1625 nfs4_lock_state();
1626 conf = find_confirmed_client_by_name(&exid->clname, nn);
1627 if (conf) {
1628 bool creds_match = same_creds(&conf->cl_cred, &rqstp->rq_cred);
1629 bool verfs_match = same_verf(&verf, &conf->cl_verifier);
1631 if (update) {
1632 if (!clp_used_exchangeid(conf)) { /* buggy client */
1633 status = nfserr_inval;
1634 goto out;
1636 if (!creds_match) { /* case 9 */
1637 status = nfserr_perm;
1638 goto out;
1640 if (!verfs_match) { /* case 8 */
1641 status = nfserr_not_same;
1642 goto out;
1644 /* case 6 */
1645 exid->flags |= EXCHGID4_FLAG_CONFIRMED_R;
1646 new = conf;
1647 goto out_copy;
1649 if (!creds_match) { /* case 3 */
1650 if (client_has_state(conf)) {
1651 status = nfserr_clid_inuse;
1652 goto out;
1654 expire_client(conf);
1655 goto out_new;
1657 if (verfs_match) { /* case 2 */
1658 conf->cl_exchange_flags |= EXCHGID4_FLAG_CONFIRMED_R;
1659 new = conf;
1660 goto out_copy;
1662 /* case 5, client reboot */
1663 goto out_new;
1666 if (update) { /* case 7 */
1667 status = nfserr_noent;
1668 goto out;
1671 unconf = find_unconfirmed_client_by_name(&exid->clname, nn);
1672 if (unconf) /* case 4, possible retry or client restart */
1673 expire_client(unconf);
1675 /* case 1 (normal case) */
1676 out_new:
1677 new = create_client(exid->clname, rqstp, &verf);
1678 if (new == NULL) {
1679 status = nfserr_jukebox;
1680 goto out;
1682 new->cl_minorversion = 1;
1684 gen_clid(new, nn);
1685 add_to_unconfirmed(new);
1686 out_copy:
1687 exid->clientid.cl_boot = new->cl_clientid.cl_boot;
1688 exid->clientid.cl_id = new->cl_clientid.cl_id;
1690 exid->seqid = new->cl_cs_slot.sl_seqid + 1;
1691 nfsd4_set_ex_flags(new, exid);
1693 dprintk("nfsd4_exchange_id seqid %d flags %x\n",
1694 new->cl_cs_slot.sl_seqid, new->cl_exchange_flags);
1695 status = nfs_ok;
1697 out:
1698 nfs4_unlock_state();
1699 return status;
1702 static __be32
1703 check_slot_seqid(u32 seqid, u32 slot_seqid, int slot_inuse)
1705 dprintk("%s enter. seqid %d slot_seqid %d\n", __func__, seqid,
1706 slot_seqid);
1708 /* The slot is in use, and no response has been sent. */
1709 if (slot_inuse) {
1710 if (seqid == slot_seqid)
1711 return nfserr_jukebox;
1712 else
1713 return nfserr_seq_misordered;
1715 /* Note unsigned 32-bit arithmetic handles wraparound: */
1716 if (likely(seqid == slot_seqid + 1))
1717 return nfs_ok;
1718 if (seqid == slot_seqid)
1719 return nfserr_replay_cache;
1720 return nfserr_seq_misordered;
1724 * Cache the create session result into the create session single DRC
1725 * slot cache by saving the xdr structure. sl_seqid has been set.
1726 * Do this for solo or embedded create session operations.
1728 static void
1729 nfsd4_cache_create_session(struct nfsd4_create_session *cr_ses,
1730 struct nfsd4_clid_slot *slot, __be32 nfserr)
1732 slot->sl_status = nfserr;
1733 memcpy(&slot->sl_cr_ses, cr_ses, sizeof(*cr_ses));
1736 static __be32
1737 nfsd4_replay_create_session(struct nfsd4_create_session *cr_ses,
1738 struct nfsd4_clid_slot *slot)
1740 memcpy(cr_ses, &slot->sl_cr_ses, sizeof(*cr_ses));
1741 return slot->sl_status;
1744 #define NFSD_MIN_REQ_HDR_SEQ_SZ ((\
1745 2 * 2 + /* credential,verifier: AUTH_NULL, length 0 */ \
1746 1 + /* MIN tag is length with zero, only length */ \
1747 3 + /* version, opcount, opcode */ \
1748 XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
1749 /* seqid, slotID, slotID, cache */ \
1750 4 ) * sizeof(__be32))
1752 #define NFSD_MIN_RESP_HDR_SEQ_SZ ((\
1753 2 + /* verifier: AUTH_NULL, length 0 */\
1754 1 + /* status */ \
1755 1 + /* MIN tag is length with zero, only length */ \
1756 3 + /* opcount, opcode, opstatus*/ \
1757 XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
1758 /* seqid, slotID, slotID, slotID, status */ \
1759 5 ) * sizeof(__be32))
1761 static bool check_forechannel_attrs(struct nfsd4_channel_attrs fchannel)
1763 return fchannel.maxreq_sz < NFSD_MIN_REQ_HDR_SEQ_SZ
1764 || fchannel.maxresp_sz < NFSD_MIN_RESP_HDR_SEQ_SZ;
1767 __be32
1768 nfsd4_create_session(struct svc_rqst *rqstp,
1769 struct nfsd4_compound_state *cstate,
1770 struct nfsd4_create_session *cr_ses)
1772 struct sockaddr *sa = svc_addr(rqstp);
1773 struct nfs4_client *conf, *unconf;
1774 struct nfsd4_session *new;
1775 struct nfsd4_conn *conn;
1776 struct nfsd4_clid_slot *cs_slot = NULL;
1777 __be32 status = 0;
1778 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
1780 if (cr_ses->flags & ~SESSION4_FLAG_MASK_A)
1781 return nfserr_inval;
1782 if (check_forechannel_attrs(cr_ses->fore_channel))
1783 return nfserr_toosmall;
1784 new = alloc_session(&cr_ses->fore_channel);
1785 if (!new)
1786 return nfserr_jukebox;
1787 status = nfserr_jukebox;
1788 conn = alloc_conn_from_crses(rqstp, cr_ses);
1789 if (!conn)
1790 goto out_free_session;
1792 nfs4_lock_state();
1793 unconf = find_unconfirmed_client(&cr_ses->clientid, true, nn);
1794 conf = find_confirmed_client(&cr_ses->clientid, true, nn);
1796 if (conf) {
1797 cs_slot = &conf->cl_cs_slot;
1798 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
1799 if (status == nfserr_replay_cache) {
1800 status = nfsd4_replay_create_session(cr_ses, cs_slot);
1801 goto out_free_conn;
1802 } else if (cr_ses->seqid != cs_slot->sl_seqid + 1) {
1803 status = nfserr_seq_misordered;
1804 goto out_free_conn;
1806 } else if (unconf) {
1807 struct nfs4_client *old;
1808 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred) ||
1809 !rpc_cmp_addr(sa, (struct sockaddr *) &unconf->cl_addr)) {
1810 status = nfserr_clid_inuse;
1811 goto out_free_conn;
1813 cs_slot = &unconf->cl_cs_slot;
1814 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
1815 if (status) {
1816 /* an unconfirmed replay returns misordered */
1817 status = nfserr_seq_misordered;
1818 goto out_free_conn;
1820 old = find_confirmed_client_by_name(&unconf->cl_name, nn);
1821 if (old)
1822 expire_client(old);
1823 move_to_confirmed(unconf);
1824 conf = unconf;
1825 } else {
1826 status = nfserr_stale_clientid;
1827 goto out_free_conn;
1829 status = nfs_ok;
1831 * We do not support RDMA or persistent sessions
1833 cr_ses->flags &= ~SESSION4_PERSIST;
1834 cr_ses->flags &= ~SESSION4_RDMA;
1836 init_session(rqstp, new, conf, cr_ses);
1837 nfsd4_init_conn(rqstp, conn, new);
1839 memcpy(cr_ses->sessionid.data, new->se_sessionid.data,
1840 NFS4_MAX_SESSIONID_LEN);
1841 memcpy(&cr_ses->fore_channel, &new->se_fchannel,
1842 sizeof(struct nfsd4_channel_attrs));
1843 cs_slot->sl_seqid++;
1844 cr_ses->seqid = cs_slot->sl_seqid;
1846 /* cache solo and embedded create sessions under the state lock */
1847 nfsd4_cache_create_session(cr_ses, cs_slot, status);
1848 out:
1849 nfs4_unlock_state();
1850 dprintk("%s returns %d\n", __func__, ntohl(status));
1851 return status;
1852 out_free_conn:
1853 free_conn(conn);
1854 out_free_session:
1855 __free_session(new);
1856 goto out;
1859 static bool nfsd4_last_compound_op(struct svc_rqst *rqstp)
1861 struct nfsd4_compoundres *resp = rqstp->rq_resp;
1862 struct nfsd4_compoundargs *argp = rqstp->rq_argp;
1864 return argp->opcnt == resp->opcnt;
1867 static __be32 nfsd4_map_bcts_dir(u32 *dir)
1869 switch (*dir) {
1870 case NFS4_CDFC4_FORE:
1871 case NFS4_CDFC4_BACK:
1872 return nfs_ok;
1873 case NFS4_CDFC4_FORE_OR_BOTH:
1874 case NFS4_CDFC4_BACK_OR_BOTH:
1875 *dir = NFS4_CDFC4_BOTH;
1876 return nfs_ok;
1878 return nfserr_inval;
1881 __be32 nfsd4_backchannel_ctl(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate, struct nfsd4_backchannel_ctl *bc)
1883 struct nfsd4_session *session = cstate->session;
1884 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
1886 spin_lock(&nn->client_lock);
1887 session->se_cb_prog = bc->bc_cb_program;
1888 session->se_cb_sec = bc->bc_cb_sec;
1889 spin_unlock(&nn->client_lock);
1891 nfsd4_probe_callback(session->se_client);
1893 return nfs_ok;
1896 __be32 nfsd4_bind_conn_to_session(struct svc_rqst *rqstp,
1897 struct nfsd4_compound_state *cstate,
1898 struct nfsd4_bind_conn_to_session *bcts)
1900 __be32 status;
1901 struct nfsd4_conn *conn;
1902 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
1904 if (!nfsd4_last_compound_op(rqstp))
1905 return nfserr_not_only_op;
1906 spin_lock(&nn->client_lock);
1907 cstate->session = find_in_sessionid_hashtbl(&bcts->sessionid, SVC_NET(rqstp));
1908 /* Sorta weird: we only need the refcnt'ing because new_conn acquires
1909 * client_lock iself: */
1910 if (cstate->session) {
1911 nfsd4_get_session(cstate->session);
1912 atomic_inc(&cstate->session->se_client->cl_refcount);
1914 spin_unlock(&nn->client_lock);
1915 if (!cstate->session)
1916 return nfserr_badsession;
1918 status = nfsd4_map_bcts_dir(&bcts->dir);
1919 if (status)
1920 return status;
1921 conn = alloc_conn(rqstp, bcts->dir);
1922 if (!conn)
1923 return nfserr_jukebox;
1924 nfsd4_init_conn(rqstp, conn, cstate->session);
1925 return nfs_ok;
1928 static bool nfsd4_compound_in_session(struct nfsd4_session *session, struct nfs4_sessionid *sid)
1930 if (!session)
1931 return 0;
1932 return !memcmp(sid, &session->se_sessionid, sizeof(*sid));
1935 __be32
1936 nfsd4_destroy_session(struct svc_rqst *r,
1937 struct nfsd4_compound_state *cstate,
1938 struct nfsd4_destroy_session *sessionid)
1940 struct nfsd4_session *ses;
1941 __be32 status = nfserr_badsession;
1942 struct nfsd_net *nn = net_generic(SVC_NET(r), nfsd_net_id);
1944 /* Notes:
1945 * - The confirmed nfs4_client->cl_sessionid holds destroyed sessinid
1946 * - Should we return nfserr_back_chan_busy if waiting for
1947 * callbacks on to-be-destroyed session?
1948 * - Do we need to clear any callback info from previous session?
1951 if (nfsd4_compound_in_session(cstate->session, &sessionid->sessionid)) {
1952 if (!nfsd4_last_compound_op(r))
1953 return nfserr_not_only_op;
1955 dump_sessionid(__func__, &sessionid->sessionid);
1956 spin_lock(&nn->client_lock);
1957 ses = find_in_sessionid_hashtbl(&sessionid->sessionid, SVC_NET(r));
1958 if (!ses) {
1959 spin_unlock(&nn->client_lock);
1960 goto out;
1963 unhash_session(ses);
1964 spin_unlock(&nn->client_lock);
1966 nfs4_lock_state();
1967 nfsd4_probe_callback_sync(ses->se_client);
1968 nfs4_unlock_state();
1970 spin_lock(&nn->client_lock);
1971 nfsd4_del_conns(ses);
1972 nfsd4_put_session_locked(ses);
1973 spin_unlock(&nn->client_lock);
1974 status = nfs_ok;
1975 out:
1976 dprintk("%s returns %d\n", __func__, ntohl(status));
1977 return status;
1980 static struct nfsd4_conn *__nfsd4_find_conn(struct svc_xprt *xpt, struct nfsd4_session *s)
1982 struct nfsd4_conn *c;
1984 list_for_each_entry(c, &s->se_conns, cn_persession) {
1985 if (c->cn_xprt == xpt) {
1986 return c;
1989 return NULL;
1992 static void nfsd4_sequence_check_conn(struct nfsd4_conn *new, struct nfsd4_session *ses)
1994 struct nfs4_client *clp = ses->se_client;
1995 struct nfsd4_conn *c;
1996 int ret;
1998 spin_lock(&clp->cl_lock);
1999 c = __nfsd4_find_conn(new->cn_xprt, ses);
2000 if (c) {
2001 spin_unlock(&clp->cl_lock);
2002 free_conn(new);
2003 return;
2005 __nfsd4_hash_conn(new, ses);
2006 spin_unlock(&clp->cl_lock);
2007 ret = nfsd4_register_conn(new);
2008 if (ret)
2009 /* oops; xprt is already down: */
2010 nfsd4_conn_lost(&new->cn_xpt_user);
2011 return;
2014 static bool nfsd4_session_too_many_ops(struct svc_rqst *rqstp, struct nfsd4_session *session)
2016 struct nfsd4_compoundargs *args = rqstp->rq_argp;
2018 return args->opcnt > session->se_fchannel.maxops;
2021 static bool nfsd4_request_too_big(struct svc_rqst *rqstp,
2022 struct nfsd4_session *session)
2024 struct xdr_buf *xb = &rqstp->rq_arg;
2026 return xb->len > session->se_fchannel.maxreq_sz;
2029 __be32
2030 nfsd4_sequence(struct svc_rqst *rqstp,
2031 struct nfsd4_compound_state *cstate,
2032 struct nfsd4_sequence *seq)
2034 struct nfsd4_compoundres *resp = rqstp->rq_resp;
2035 struct nfsd4_session *session;
2036 struct nfsd4_slot *slot;
2037 struct nfsd4_conn *conn;
2038 __be32 status;
2039 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2041 if (resp->opcnt != 1)
2042 return nfserr_sequence_pos;
2045 * Will be either used or freed by nfsd4_sequence_check_conn
2046 * below.
2048 conn = alloc_conn(rqstp, NFS4_CDFC4_FORE);
2049 if (!conn)
2050 return nfserr_jukebox;
2052 spin_lock(&nn->client_lock);
2053 status = nfserr_badsession;
2054 session = find_in_sessionid_hashtbl(&seq->sessionid, SVC_NET(rqstp));
2055 if (!session)
2056 goto out;
2058 status = nfserr_too_many_ops;
2059 if (nfsd4_session_too_many_ops(rqstp, session))
2060 goto out;
2062 status = nfserr_req_too_big;
2063 if (nfsd4_request_too_big(rqstp, session))
2064 goto out;
2066 status = nfserr_badslot;
2067 if (seq->slotid >= session->se_fchannel.maxreqs)
2068 goto out;
2070 slot = session->se_slots[seq->slotid];
2071 dprintk("%s: slotid %d\n", __func__, seq->slotid);
2073 /* We do not negotiate the number of slots yet, so set the
2074 * maxslots to the session maxreqs which is used to encode
2075 * sr_highest_slotid and the sr_target_slot id to maxslots */
2076 seq->maxslots = session->se_fchannel.maxreqs;
2078 status = check_slot_seqid(seq->seqid, slot->sl_seqid,
2079 slot->sl_flags & NFSD4_SLOT_INUSE);
2080 if (status == nfserr_replay_cache) {
2081 status = nfserr_seq_misordered;
2082 if (!(slot->sl_flags & NFSD4_SLOT_INITIALIZED))
2083 goto out;
2084 cstate->slot = slot;
2085 cstate->session = session;
2086 /* Return the cached reply status and set cstate->status
2087 * for nfsd4_proc_compound processing */
2088 status = nfsd4_replay_cache_entry(resp, seq);
2089 cstate->status = nfserr_replay_cache;
2090 goto out;
2092 if (status)
2093 goto out;
2095 nfsd4_sequence_check_conn(conn, session);
2096 conn = NULL;
2098 /* Success! bump slot seqid */
2099 slot->sl_seqid = seq->seqid;
2100 slot->sl_flags |= NFSD4_SLOT_INUSE;
2101 if (seq->cachethis)
2102 slot->sl_flags |= NFSD4_SLOT_CACHETHIS;
2103 else
2104 slot->sl_flags &= ~NFSD4_SLOT_CACHETHIS;
2106 cstate->slot = slot;
2107 cstate->session = session;
2109 out:
2110 /* Hold a session reference until done processing the compound. */
2111 if (cstate->session) {
2112 struct nfs4_client *clp = session->se_client;
2114 nfsd4_get_session(cstate->session);
2115 atomic_inc(&clp->cl_refcount);
2116 switch (clp->cl_cb_state) {
2117 case NFSD4_CB_DOWN:
2118 seq->status_flags = SEQ4_STATUS_CB_PATH_DOWN;
2119 break;
2120 case NFSD4_CB_FAULT:
2121 seq->status_flags = SEQ4_STATUS_BACKCHANNEL_FAULT;
2122 break;
2123 default:
2124 seq->status_flags = 0;
2127 kfree(conn);
2128 spin_unlock(&nn->client_lock);
2129 dprintk("%s: return %d\n", __func__, ntohl(status));
2130 return status;
2133 __be32
2134 nfsd4_destroy_clientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate, struct nfsd4_destroy_clientid *dc)
2136 struct nfs4_client *conf, *unconf, *clp;
2137 __be32 status = 0;
2138 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2140 nfs4_lock_state();
2141 unconf = find_unconfirmed_client(&dc->clientid, true, nn);
2142 conf = find_confirmed_client(&dc->clientid, true, nn);
2144 if (conf) {
2145 clp = conf;
2147 if (!is_client_expired(conf) && client_has_state(conf)) {
2148 status = nfserr_clientid_busy;
2149 goto out;
2152 /* rfc5661 18.50.3 */
2153 if (cstate->session && conf == cstate->session->se_client) {
2154 status = nfserr_clientid_busy;
2155 goto out;
2157 } else if (unconf)
2158 clp = unconf;
2159 else {
2160 status = nfserr_stale_clientid;
2161 goto out;
2164 expire_client(clp);
2165 out:
2166 nfs4_unlock_state();
2167 dprintk("%s return %d\n", __func__, ntohl(status));
2168 return status;
2171 __be32
2172 nfsd4_reclaim_complete(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate, struct nfsd4_reclaim_complete *rc)
2174 __be32 status = 0;
2176 if (rc->rca_one_fs) {
2177 if (!cstate->current_fh.fh_dentry)
2178 return nfserr_nofilehandle;
2180 * We don't take advantage of the rca_one_fs case.
2181 * That's OK, it's optional, we can safely ignore it.
2183 return nfs_ok;
2186 nfs4_lock_state();
2187 status = nfserr_complete_already;
2188 if (test_and_set_bit(NFSD4_CLIENT_RECLAIM_COMPLETE,
2189 &cstate->session->se_client->cl_flags))
2190 goto out;
2192 status = nfserr_stale_clientid;
2193 if (is_client_expired(cstate->session->se_client))
2195 * The following error isn't really legal.
2196 * But we only get here if the client just explicitly
2197 * destroyed the client. Surely it no longer cares what
2198 * error it gets back on an operation for the dead
2199 * client.
2201 goto out;
2203 status = nfs_ok;
2204 nfsd4_client_record_create(cstate->session->se_client);
2205 out:
2206 nfs4_unlock_state();
2207 return status;
2210 __be32
2211 nfsd4_setclientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2212 struct nfsd4_setclientid *setclid)
2214 struct xdr_netobj clname = setclid->se_name;
2215 nfs4_verifier clverifier = setclid->se_verf;
2216 struct nfs4_client *conf, *unconf, *new;
2217 __be32 status;
2218 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2220 /* Cases below refer to rfc 3530 section 14.2.33: */
2221 nfs4_lock_state();
2222 conf = find_confirmed_client_by_name(&clname, nn);
2223 if (conf) {
2224 /* case 0: */
2225 status = nfserr_clid_inuse;
2226 if (clp_used_exchangeid(conf))
2227 goto out;
2228 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
2229 char addr_str[INET6_ADDRSTRLEN];
2230 rpc_ntop((struct sockaddr *) &conf->cl_addr, addr_str,
2231 sizeof(addr_str));
2232 dprintk("NFSD: setclientid: string in use by client "
2233 "at %s\n", addr_str);
2234 goto out;
2237 unconf = find_unconfirmed_client_by_name(&clname, nn);
2238 if (unconf)
2239 expire_client(unconf);
2240 status = nfserr_jukebox;
2241 new = create_client(clname, rqstp, &clverifier);
2242 if (new == NULL)
2243 goto out;
2244 if (conf && same_verf(&conf->cl_verifier, &clverifier))
2245 /* case 1: probable callback update */
2246 copy_clid(new, conf);
2247 else /* case 4 (new client) or cases 2, 3 (client reboot): */
2248 gen_clid(new, nn);
2249 new->cl_minorversion = 0;
2250 gen_callback(new, setclid, rqstp);
2251 add_to_unconfirmed(new);
2252 setclid->se_clientid.cl_boot = new->cl_clientid.cl_boot;
2253 setclid->se_clientid.cl_id = new->cl_clientid.cl_id;
2254 memcpy(setclid->se_confirm.data, new->cl_confirm.data, sizeof(setclid->se_confirm.data));
2255 status = nfs_ok;
2256 out:
2257 nfs4_unlock_state();
2258 return status;
2262 __be32
2263 nfsd4_setclientid_confirm(struct svc_rqst *rqstp,
2264 struct nfsd4_compound_state *cstate,
2265 struct nfsd4_setclientid_confirm *setclientid_confirm)
2267 struct nfs4_client *conf, *unconf;
2268 nfs4_verifier confirm = setclientid_confirm->sc_confirm;
2269 clientid_t * clid = &setclientid_confirm->sc_clientid;
2270 __be32 status;
2271 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2273 if (STALE_CLIENTID(clid, nn))
2274 return nfserr_stale_clientid;
2275 nfs4_lock_state();
2277 conf = find_confirmed_client(clid, false, nn);
2278 unconf = find_unconfirmed_client(clid, false, nn);
2280 * We try hard to give out unique clientid's, so if we get an
2281 * attempt to confirm the same clientid with a different cred,
2282 * there's a bug somewhere. Let's charitably assume it's our
2283 * bug.
2285 status = nfserr_serverfault;
2286 if (unconf && !same_creds(&unconf->cl_cred, &rqstp->rq_cred))
2287 goto out;
2288 if (conf && !same_creds(&conf->cl_cred, &rqstp->rq_cred))
2289 goto out;
2290 /* cases below refer to rfc 3530 section 14.2.34: */
2291 if (!unconf || !same_verf(&confirm, &unconf->cl_confirm)) {
2292 if (conf && !unconf) /* case 2: probable retransmit */
2293 status = nfs_ok;
2294 else /* case 4: client hasn't noticed we rebooted yet? */
2295 status = nfserr_stale_clientid;
2296 goto out;
2298 status = nfs_ok;
2299 if (conf) { /* case 1: callback update */
2300 nfsd4_change_callback(conf, &unconf->cl_cb_conn);
2301 nfsd4_probe_callback(conf);
2302 expire_client(unconf);
2303 } else { /* case 3: normal case; new or rebooted client */
2304 conf = find_confirmed_client_by_name(&unconf->cl_name, nn);
2305 if (conf)
2306 expire_client(conf);
2307 move_to_confirmed(unconf);
2308 nfsd4_probe_callback(unconf);
2310 out:
2311 nfs4_unlock_state();
2312 return status;
2315 static struct nfs4_file *nfsd4_alloc_file(void)
2317 return kmem_cache_alloc(file_slab, GFP_KERNEL);
2320 /* OPEN Share state helper functions */
2321 static void nfsd4_init_file(struct nfs4_file *fp, struct inode *ino)
2323 unsigned int hashval = file_hashval(ino);
2325 atomic_set(&fp->fi_ref, 1);
2326 INIT_LIST_HEAD(&fp->fi_hash);
2327 INIT_LIST_HEAD(&fp->fi_stateids);
2328 INIT_LIST_HEAD(&fp->fi_delegations);
2329 fp->fi_inode = igrab(ino);
2330 fp->fi_had_conflict = false;
2331 fp->fi_lease = NULL;
2332 memset(fp->fi_fds, 0, sizeof(fp->fi_fds));
2333 memset(fp->fi_access, 0, sizeof(fp->fi_access));
2334 spin_lock(&recall_lock);
2335 list_add(&fp->fi_hash, &file_hashtbl[hashval]);
2336 spin_unlock(&recall_lock);
2339 static void
2340 nfsd4_free_slab(struct kmem_cache **slab)
2342 if (*slab == NULL)
2343 return;
2344 kmem_cache_destroy(*slab);
2345 *slab = NULL;
2348 void
2349 nfsd4_free_slabs(void)
2351 nfsd4_free_slab(&openowner_slab);
2352 nfsd4_free_slab(&lockowner_slab);
2353 nfsd4_free_slab(&file_slab);
2354 nfsd4_free_slab(&stateid_slab);
2355 nfsd4_free_slab(&deleg_slab);
2359 nfsd4_init_slabs(void)
2361 openowner_slab = kmem_cache_create("nfsd4_openowners",
2362 sizeof(struct nfs4_openowner), 0, 0, NULL);
2363 if (openowner_slab == NULL)
2364 goto out_nomem;
2365 lockowner_slab = kmem_cache_create("nfsd4_lockowners",
2366 sizeof(struct nfs4_lockowner), 0, 0, NULL);
2367 if (lockowner_slab == NULL)
2368 goto out_nomem;
2369 file_slab = kmem_cache_create("nfsd4_files",
2370 sizeof(struct nfs4_file), 0, 0, NULL);
2371 if (file_slab == NULL)
2372 goto out_nomem;
2373 stateid_slab = kmem_cache_create("nfsd4_stateids",
2374 sizeof(struct nfs4_ol_stateid), 0, 0, NULL);
2375 if (stateid_slab == NULL)
2376 goto out_nomem;
2377 deleg_slab = kmem_cache_create("nfsd4_delegations",
2378 sizeof(struct nfs4_delegation), 0, 0, NULL);
2379 if (deleg_slab == NULL)
2380 goto out_nomem;
2381 return 0;
2382 out_nomem:
2383 nfsd4_free_slabs();
2384 dprintk("nfsd4: out of memory while initializing nfsv4\n");
2385 return -ENOMEM;
2388 void nfs4_free_openowner(struct nfs4_openowner *oo)
2390 kfree(oo->oo_owner.so_owner.data);
2391 kmem_cache_free(openowner_slab, oo);
2394 void nfs4_free_lockowner(struct nfs4_lockowner *lo)
2396 kfree(lo->lo_owner.so_owner.data);
2397 kmem_cache_free(lockowner_slab, lo);
2400 static void init_nfs4_replay(struct nfs4_replay *rp)
2402 rp->rp_status = nfserr_serverfault;
2403 rp->rp_buflen = 0;
2404 rp->rp_buf = rp->rp_ibuf;
2407 static inline void *alloc_stateowner(struct kmem_cache *slab, struct xdr_netobj *owner, struct nfs4_client *clp)
2409 struct nfs4_stateowner *sop;
2411 sop = kmem_cache_alloc(slab, GFP_KERNEL);
2412 if (!sop)
2413 return NULL;
2415 sop->so_owner.data = kmemdup(owner->data, owner->len, GFP_KERNEL);
2416 if (!sop->so_owner.data) {
2417 kmem_cache_free(slab, sop);
2418 return NULL;
2420 sop->so_owner.len = owner->len;
2422 INIT_LIST_HEAD(&sop->so_stateids);
2423 sop->so_client = clp;
2424 init_nfs4_replay(&sop->so_replay);
2425 return sop;
2428 static void hash_openowner(struct nfs4_openowner *oo, struct nfs4_client *clp, unsigned int strhashval)
2430 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2432 list_add(&oo->oo_owner.so_strhash, &nn->ownerstr_hashtbl[strhashval]);
2433 list_add(&oo->oo_perclient, &clp->cl_openowners);
2436 static struct nfs4_openowner *
2437 alloc_init_open_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfsd4_open *open) {
2438 struct nfs4_openowner *oo;
2440 oo = alloc_stateowner(openowner_slab, &open->op_owner, clp);
2441 if (!oo)
2442 return NULL;
2443 oo->oo_owner.so_is_open_owner = 1;
2444 oo->oo_owner.so_seqid = open->op_seqid;
2445 oo->oo_flags = NFS4_OO_NEW;
2446 oo->oo_time = 0;
2447 oo->oo_last_closed_stid = NULL;
2448 INIT_LIST_HEAD(&oo->oo_close_lru);
2449 hash_openowner(oo, clp, strhashval);
2450 return oo;
2453 static void init_open_stateid(struct nfs4_ol_stateid *stp, struct nfs4_file *fp, struct nfsd4_open *open) {
2454 struct nfs4_openowner *oo = open->op_openowner;
2455 struct nfs4_client *clp = oo->oo_owner.so_client;
2457 init_stid(&stp->st_stid, clp, NFS4_OPEN_STID);
2458 INIT_LIST_HEAD(&stp->st_lockowners);
2459 list_add(&stp->st_perstateowner, &oo->oo_owner.so_stateids);
2460 list_add(&stp->st_perfile, &fp->fi_stateids);
2461 stp->st_stateowner = &oo->oo_owner;
2462 get_nfs4_file(fp);
2463 stp->st_file = fp;
2464 stp->st_access_bmap = 0;
2465 stp->st_deny_bmap = 0;
2466 set_access(open->op_share_access, stp);
2467 set_deny(open->op_share_deny, stp);
2468 stp->st_openstp = NULL;
2471 static void
2472 move_to_close_lru(struct nfs4_openowner *oo, struct net *net)
2474 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
2476 dprintk("NFSD: move_to_close_lru nfs4_openowner %p\n", oo);
2478 list_move_tail(&oo->oo_close_lru, &nn->close_lru);
2479 oo->oo_time = get_seconds();
2482 static int
2483 same_owner_str(struct nfs4_stateowner *sop, struct xdr_netobj *owner,
2484 clientid_t *clid)
2486 return (sop->so_owner.len == owner->len) &&
2487 0 == memcmp(sop->so_owner.data, owner->data, owner->len) &&
2488 (sop->so_client->cl_clientid.cl_id == clid->cl_id);
2491 static struct nfs4_openowner *
2492 find_openstateowner_str(unsigned int hashval, struct nfsd4_open *open,
2493 bool sessions, struct nfsd_net *nn)
2495 struct nfs4_stateowner *so;
2496 struct nfs4_openowner *oo;
2497 struct nfs4_client *clp;
2499 list_for_each_entry(so, &nn->ownerstr_hashtbl[hashval], so_strhash) {
2500 if (!so->so_is_open_owner)
2501 continue;
2502 if (same_owner_str(so, &open->op_owner, &open->op_clientid)) {
2503 oo = openowner(so);
2504 clp = oo->oo_owner.so_client;
2505 if ((bool)clp->cl_minorversion != sessions)
2506 return NULL;
2507 renew_client(oo->oo_owner.so_client);
2508 return oo;
2511 return NULL;
2514 /* search file_hashtbl[] for file */
2515 static struct nfs4_file *
2516 find_file(struct inode *ino)
2518 unsigned int hashval = file_hashval(ino);
2519 struct nfs4_file *fp;
2521 spin_lock(&recall_lock);
2522 list_for_each_entry(fp, &file_hashtbl[hashval], fi_hash) {
2523 if (fp->fi_inode == ino) {
2524 get_nfs4_file(fp);
2525 spin_unlock(&recall_lock);
2526 return fp;
2529 spin_unlock(&recall_lock);
2530 return NULL;
2534 * Called to check deny when READ with all zero stateid or
2535 * WRITE with all zero or all one stateid
2537 static __be32
2538 nfs4_share_conflict(struct svc_fh *current_fh, unsigned int deny_type)
2540 struct inode *ino = current_fh->fh_dentry->d_inode;
2541 struct nfs4_file *fp;
2542 struct nfs4_ol_stateid *stp;
2543 __be32 ret;
2545 dprintk("NFSD: nfs4_share_conflict\n");
2547 fp = find_file(ino);
2548 if (!fp)
2549 return nfs_ok;
2550 ret = nfserr_locked;
2551 /* Search for conflicting share reservations */
2552 list_for_each_entry(stp, &fp->fi_stateids, st_perfile) {
2553 if (test_deny(deny_type, stp) ||
2554 test_deny(NFS4_SHARE_DENY_BOTH, stp))
2555 goto out;
2557 ret = nfs_ok;
2558 out:
2559 put_nfs4_file(fp);
2560 return ret;
2563 static void nfsd_break_one_deleg(struct nfs4_delegation *dp)
2565 /* We're assuming the state code never drops its reference
2566 * without first removing the lease. Since we're in this lease
2567 * callback (and since the lease code is serialized by the kernel
2568 * lock) we know the server hasn't removed the lease yet, we know
2569 * it's safe to take a reference: */
2570 atomic_inc(&dp->dl_count);
2572 list_add_tail(&dp->dl_recall_lru, &del_recall_lru);
2574 /* only place dl_time is set. protected by lock_flocks*/
2575 dp->dl_time = get_seconds();
2577 nfsd4_cb_recall(dp);
2580 /* Called from break_lease() with lock_flocks() held. */
2581 static void nfsd_break_deleg_cb(struct file_lock *fl)
2583 struct nfs4_file *fp = (struct nfs4_file *)fl->fl_owner;
2584 struct nfs4_delegation *dp;
2586 if (!fp) {
2587 WARN(1, "(%p)->fl_owner NULL\n", fl);
2588 return;
2590 if (fp->fi_had_conflict) {
2591 WARN(1, "duplicate break on %p\n", fp);
2592 return;
2595 * We don't want the locks code to timeout the lease for us;
2596 * we'll remove it ourself if a delegation isn't returned
2597 * in time:
2599 fl->fl_break_time = 0;
2601 spin_lock(&recall_lock);
2602 fp->fi_had_conflict = true;
2603 list_for_each_entry(dp, &fp->fi_delegations, dl_perfile)
2604 nfsd_break_one_deleg(dp);
2605 spin_unlock(&recall_lock);
2608 static
2609 int nfsd_change_deleg_cb(struct file_lock **onlist, int arg)
2611 if (arg & F_UNLCK)
2612 return lease_modify(onlist, arg);
2613 else
2614 return -EAGAIN;
2617 static const struct lock_manager_operations nfsd_lease_mng_ops = {
2618 .lm_break = nfsd_break_deleg_cb,
2619 .lm_change = nfsd_change_deleg_cb,
2622 static __be32 nfsd4_check_seqid(struct nfsd4_compound_state *cstate, struct nfs4_stateowner *so, u32 seqid)
2624 if (nfsd4_has_session(cstate))
2625 return nfs_ok;
2626 if (seqid == so->so_seqid - 1)
2627 return nfserr_replay_me;
2628 if (seqid == so->so_seqid)
2629 return nfs_ok;
2630 return nfserr_bad_seqid;
2633 __be32
2634 nfsd4_process_open1(struct nfsd4_compound_state *cstate,
2635 struct nfsd4_open *open, struct nfsd_net *nn)
2637 clientid_t *clientid = &open->op_clientid;
2638 struct nfs4_client *clp = NULL;
2639 unsigned int strhashval;
2640 struct nfs4_openowner *oo = NULL;
2641 __be32 status;
2643 if (STALE_CLIENTID(&open->op_clientid, nn))
2644 return nfserr_stale_clientid;
2646 * In case we need it later, after we've already created the
2647 * file and don't want to risk a further failure:
2649 open->op_file = nfsd4_alloc_file();
2650 if (open->op_file == NULL)
2651 return nfserr_jukebox;
2653 strhashval = ownerstr_hashval(clientid->cl_id, &open->op_owner);
2654 oo = find_openstateowner_str(strhashval, open, cstate->minorversion, nn);
2655 open->op_openowner = oo;
2656 if (!oo) {
2657 clp = find_confirmed_client(clientid, cstate->minorversion,
2658 nn);
2659 if (clp == NULL)
2660 return nfserr_expired;
2661 goto new_owner;
2663 if (!(oo->oo_flags & NFS4_OO_CONFIRMED)) {
2664 /* Replace unconfirmed owners without checking for replay. */
2665 clp = oo->oo_owner.so_client;
2666 release_openowner(oo);
2667 open->op_openowner = NULL;
2668 goto new_owner;
2670 status = nfsd4_check_seqid(cstate, &oo->oo_owner, open->op_seqid);
2671 if (status)
2672 return status;
2673 clp = oo->oo_owner.so_client;
2674 goto alloc_stateid;
2675 new_owner:
2676 oo = alloc_init_open_stateowner(strhashval, clp, open);
2677 if (oo == NULL)
2678 return nfserr_jukebox;
2679 open->op_openowner = oo;
2680 alloc_stateid:
2681 open->op_stp = nfs4_alloc_stateid(clp);
2682 if (!open->op_stp)
2683 return nfserr_jukebox;
2684 return nfs_ok;
2687 static inline __be32
2688 nfs4_check_delegmode(struct nfs4_delegation *dp, int flags)
2690 if ((flags & WR_STATE) && (dp->dl_type == NFS4_OPEN_DELEGATE_READ))
2691 return nfserr_openmode;
2692 else
2693 return nfs_ok;
2696 static int share_access_to_flags(u32 share_access)
2698 return share_access == NFS4_SHARE_ACCESS_READ ? RD_STATE : WR_STATE;
2701 static struct nfs4_delegation *find_deleg_stateid(struct nfs4_client *cl, stateid_t *s)
2703 struct nfs4_stid *ret;
2705 ret = find_stateid_by_type(cl, s, NFS4_DELEG_STID);
2706 if (!ret)
2707 return NULL;
2708 return delegstateid(ret);
2711 static bool nfsd4_is_deleg_cur(struct nfsd4_open *open)
2713 return open->op_claim_type == NFS4_OPEN_CLAIM_DELEGATE_CUR ||
2714 open->op_claim_type == NFS4_OPEN_CLAIM_DELEG_CUR_FH;
2717 static __be32
2718 nfs4_check_deleg(struct nfs4_client *cl, struct nfs4_file *fp, struct nfsd4_open *open,
2719 struct nfs4_delegation **dp)
2721 int flags;
2722 __be32 status = nfserr_bad_stateid;
2724 *dp = find_deleg_stateid(cl, &open->op_delegate_stateid);
2725 if (*dp == NULL)
2726 goto out;
2727 flags = share_access_to_flags(open->op_share_access);
2728 status = nfs4_check_delegmode(*dp, flags);
2729 if (status)
2730 *dp = NULL;
2731 out:
2732 if (!nfsd4_is_deleg_cur(open))
2733 return nfs_ok;
2734 if (status)
2735 return status;
2736 open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
2737 return nfs_ok;
2740 static __be32
2741 nfs4_check_open(struct nfs4_file *fp, struct nfsd4_open *open, struct nfs4_ol_stateid **stpp)
2743 struct nfs4_ol_stateid *local;
2744 struct nfs4_openowner *oo = open->op_openowner;
2746 list_for_each_entry(local, &fp->fi_stateids, st_perfile) {
2747 /* ignore lock owners */
2748 if (local->st_stateowner->so_is_open_owner == 0)
2749 continue;
2750 /* remember if we have seen this open owner */
2751 if (local->st_stateowner == &oo->oo_owner)
2752 *stpp = local;
2753 /* check for conflicting share reservations */
2754 if (!test_share(local, open))
2755 return nfserr_share_denied;
2757 return nfs_ok;
2760 static inline int nfs4_access_to_access(u32 nfs4_access)
2762 int flags = 0;
2764 if (nfs4_access & NFS4_SHARE_ACCESS_READ)
2765 flags |= NFSD_MAY_READ;
2766 if (nfs4_access & NFS4_SHARE_ACCESS_WRITE)
2767 flags |= NFSD_MAY_WRITE;
2768 return flags;
2771 static __be32 nfs4_get_vfs_file(struct svc_rqst *rqstp, struct nfs4_file *fp,
2772 struct svc_fh *cur_fh, struct nfsd4_open *open)
2774 __be32 status;
2775 int oflag = nfs4_access_to_omode(open->op_share_access);
2776 int access = nfs4_access_to_access(open->op_share_access);
2778 if (!fp->fi_fds[oflag]) {
2779 status = nfsd_open(rqstp, cur_fh, S_IFREG, access,
2780 &fp->fi_fds[oflag]);
2781 if (status)
2782 return status;
2784 nfs4_file_get_access(fp, oflag);
2786 return nfs_ok;
2789 static inline __be32
2790 nfsd4_truncate(struct svc_rqst *rqstp, struct svc_fh *fh,
2791 struct nfsd4_open *open)
2793 struct iattr iattr = {
2794 .ia_valid = ATTR_SIZE,
2795 .ia_size = 0,
2797 if (!open->op_truncate)
2798 return 0;
2799 if (!(open->op_share_access & NFS4_SHARE_ACCESS_WRITE))
2800 return nfserr_inval;
2801 return nfsd_setattr(rqstp, fh, &iattr, 0, (time_t)0);
2804 static __be32
2805 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)
2807 u32 op_share_access = open->op_share_access;
2808 bool new_access;
2809 __be32 status;
2811 new_access = !test_access(op_share_access, stp);
2812 if (new_access) {
2813 status = nfs4_get_vfs_file(rqstp, fp, cur_fh, open);
2814 if (status)
2815 return status;
2817 status = nfsd4_truncate(rqstp, cur_fh, open);
2818 if (status) {
2819 if (new_access) {
2820 int oflag = nfs4_access_to_omode(op_share_access);
2821 nfs4_file_put_access(fp, oflag);
2823 return status;
2825 /* remember the open */
2826 set_access(op_share_access, stp);
2827 set_deny(open->op_share_deny, stp);
2829 return nfs_ok;
2833 static void
2834 nfs4_set_claim_prev(struct nfsd4_open *open, bool has_session)
2836 open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
2839 /* Should we give out recallable state?: */
2840 static bool nfsd4_cb_channel_good(struct nfs4_client *clp)
2842 if (clp->cl_cb_state == NFSD4_CB_UP)
2843 return true;
2845 * In the sessions case, since we don't have to establish a
2846 * separate connection for callbacks, we assume it's OK
2847 * until we hear otherwise:
2849 return clp->cl_minorversion && clp->cl_cb_state == NFSD4_CB_UNKNOWN;
2852 static struct file_lock *nfs4_alloc_init_lease(struct nfs4_delegation *dp, int flag)
2854 struct file_lock *fl;
2856 fl = locks_alloc_lock();
2857 if (!fl)
2858 return NULL;
2859 locks_init_lock(fl);
2860 fl->fl_lmops = &nfsd_lease_mng_ops;
2861 fl->fl_flags = FL_LEASE;
2862 fl->fl_type = flag == NFS4_OPEN_DELEGATE_READ? F_RDLCK: F_WRLCK;
2863 fl->fl_end = OFFSET_MAX;
2864 fl->fl_owner = (fl_owner_t)(dp->dl_file);
2865 fl->fl_pid = current->tgid;
2866 return fl;
2869 static int nfs4_setlease(struct nfs4_delegation *dp, int flag)
2871 struct nfs4_file *fp = dp->dl_file;
2872 struct file_lock *fl;
2873 int status;
2875 fl = nfs4_alloc_init_lease(dp, flag);
2876 if (!fl)
2877 return -ENOMEM;
2878 fl->fl_file = find_readable_file(fp);
2879 list_add(&dp->dl_perclnt, &dp->dl_stid.sc_client->cl_delegations);
2880 status = vfs_setlease(fl->fl_file, fl->fl_type, &fl);
2881 if (status) {
2882 list_del_init(&dp->dl_perclnt);
2883 locks_free_lock(fl);
2884 return -ENOMEM;
2886 fp->fi_lease = fl;
2887 fp->fi_deleg_file = get_file(fl->fl_file);
2888 atomic_set(&fp->fi_delegees, 1);
2889 list_add(&dp->dl_perfile, &fp->fi_delegations);
2890 return 0;
2893 static int nfs4_set_delegation(struct nfs4_delegation *dp, int flag)
2895 struct nfs4_file *fp = dp->dl_file;
2897 if (!fp->fi_lease)
2898 return nfs4_setlease(dp, flag);
2899 spin_lock(&recall_lock);
2900 if (fp->fi_had_conflict) {
2901 spin_unlock(&recall_lock);
2902 return -EAGAIN;
2904 atomic_inc(&fp->fi_delegees);
2905 list_add(&dp->dl_perfile, &fp->fi_delegations);
2906 spin_unlock(&recall_lock);
2907 list_add(&dp->dl_perclnt, &dp->dl_stid.sc_client->cl_delegations);
2908 return 0;
2911 static void nfsd4_open_deleg_none_ext(struct nfsd4_open *open, int status)
2913 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
2914 if (status == -EAGAIN)
2915 open->op_why_no_deleg = WND4_CONTENTION;
2916 else {
2917 open->op_why_no_deleg = WND4_RESOURCE;
2918 switch (open->op_deleg_want) {
2919 case NFS4_SHARE_WANT_READ_DELEG:
2920 case NFS4_SHARE_WANT_WRITE_DELEG:
2921 case NFS4_SHARE_WANT_ANY_DELEG:
2922 break;
2923 case NFS4_SHARE_WANT_CANCEL:
2924 open->op_why_no_deleg = WND4_CANCELLED;
2925 break;
2926 case NFS4_SHARE_WANT_NO_DELEG:
2927 WARN_ON_ONCE(1);
2933 * Attempt to hand out a delegation.
2935 static void
2936 nfs4_open_delegation(struct net *net, struct svc_fh *fh,
2937 struct nfsd4_open *open, struct nfs4_ol_stateid *stp)
2939 struct nfs4_delegation *dp;
2940 struct nfs4_openowner *oo = container_of(stp->st_stateowner, struct nfs4_openowner, oo_owner);
2941 int cb_up;
2942 int status = 0, flag = 0;
2944 cb_up = nfsd4_cb_channel_good(oo->oo_owner.so_client);
2945 flag = NFS4_OPEN_DELEGATE_NONE;
2946 open->op_recall = 0;
2947 switch (open->op_claim_type) {
2948 case NFS4_OPEN_CLAIM_PREVIOUS:
2949 if (!cb_up)
2950 open->op_recall = 1;
2951 flag = open->op_delegate_type;
2952 if (flag == NFS4_OPEN_DELEGATE_NONE)
2953 goto out;
2954 break;
2955 case NFS4_OPEN_CLAIM_NULL:
2956 /* Let's not give out any delegations till everyone's
2957 * had the chance to reclaim theirs.... */
2958 if (locks_in_grace(net))
2959 goto out;
2960 if (!cb_up || !(oo->oo_flags & NFS4_OO_CONFIRMED))
2961 goto out;
2962 if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
2963 flag = NFS4_OPEN_DELEGATE_WRITE;
2964 else
2965 flag = NFS4_OPEN_DELEGATE_READ;
2966 break;
2967 default:
2968 goto out;
2971 dp = alloc_init_deleg(oo->oo_owner.so_client, stp, fh, flag);
2972 if (dp == NULL)
2973 goto out_no_deleg;
2974 status = nfs4_set_delegation(dp, flag);
2975 if (status)
2976 goto out_free;
2978 memcpy(&open->op_delegate_stateid, &dp->dl_stid.sc_stateid, sizeof(dp->dl_stid.sc_stateid));
2980 dprintk("NFSD: delegation stateid=" STATEID_FMT "\n",
2981 STATEID_VAL(&dp->dl_stid.sc_stateid));
2982 out:
2983 open->op_delegate_type = flag;
2984 if (flag == NFS4_OPEN_DELEGATE_NONE) {
2985 if (open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS &&
2986 open->op_delegate_type != NFS4_OPEN_DELEGATE_NONE)
2987 dprintk("NFSD: WARNING: refusing delegation reclaim\n");
2989 /* 4.1 client asking for a delegation? */
2990 if (open->op_deleg_want)
2991 nfsd4_open_deleg_none_ext(open, status);
2993 return;
2994 out_free:
2995 nfs4_put_delegation(dp);
2996 out_no_deleg:
2997 flag = NFS4_OPEN_DELEGATE_NONE;
2998 goto out;
3001 static void nfsd4_deleg_xgrade_none_ext(struct nfsd4_open *open,
3002 struct nfs4_delegation *dp)
3004 if (open->op_deleg_want == NFS4_SHARE_WANT_READ_DELEG &&
3005 dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
3006 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
3007 open->op_why_no_deleg = WND4_NOT_SUPP_DOWNGRADE;
3008 } else if (open->op_deleg_want == NFS4_SHARE_WANT_WRITE_DELEG &&
3009 dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
3010 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
3011 open->op_why_no_deleg = WND4_NOT_SUPP_UPGRADE;
3013 /* Otherwise the client must be confused wanting a delegation
3014 * it already has, therefore we don't return
3015 * NFS4_OPEN_DELEGATE_NONE_EXT and reason.
3020 * called with nfs4_lock_state() held.
3022 __be32
3023 nfsd4_process_open2(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_open *open)
3025 struct nfsd4_compoundres *resp = rqstp->rq_resp;
3026 struct nfs4_client *cl = open->op_openowner->oo_owner.so_client;
3027 struct nfs4_file *fp = NULL;
3028 struct inode *ino = current_fh->fh_dentry->d_inode;
3029 struct nfs4_ol_stateid *stp = NULL;
3030 struct nfs4_delegation *dp = NULL;
3031 __be32 status;
3034 * Lookup file; if found, lookup stateid and check open request,
3035 * and check for delegations in the process of being recalled.
3036 * If not found, create the nfs4_file struct
3038 fp = find_file(ino);
3039 if (fp) {
3040 if ((status = nfs4_check_open(fp, open, &stp)))
3041 goto out;
3042 status = nfs4_check_deleg(cl, fp, open, &dp);
3043 if (status)
3044 goto out;
3045 } else {
3046 status = nfserr_bad_stateid;
3047 if (nfsd4_is_deleg_cur(open))
3048 goto out;
3049 status = nfserr_jukebox;
3050 fp = open->op_file;
3051 open->op_file = NULL;
3052 nfsd4_init_file(fp, ino);
3056 * OPEN the file, or upgrade an existing OPEN.
3057 * If truncate fails, the OPEN fails.
3059 if (stp) {
3060 /* Stateid was found, this is an OPEN upgrade */
3061 status = nfs4_upgrade_open(rqstp, fp, current_fh, stp, open);
3062 if (status)
3063 goto out;
3064 } else {
3065 status = nfs4_get_vfs_file(rqstp, fp, current_fh, open);
3066 if (status)
3067 goto out;
3068 status = nfsd4_truncate(rqstp, current_fh, open);
3069 if (status)
3070 goto out;
3071 stp = open->op_stp;
3072 open->op_stp = NULL;
3073 init_open_stateid(stp, fp, open);
3075 update_stateid(&stp->st_stid.sc_stateid);
3076 memcpy(&open->op_stateid, &stp->st_stid.sc_stateid, sizeof(stateid_t));
3078 if (nfsd4_has_session(&resp->cstate)) {
3079 open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
3081 if (open->op_deleg_want & NFS4_SHARE_WANT_NO_DELEG) {
3082 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
3083 open->op_why_no_deleg = WND4_NOT_WANTED;
3084 goto nodeleg;
3089 * Attempt to hand out a delegation. No error return, because the
3090 * OPEN succeeds even if we fail.
3092 nfs4_open_delegation(SVC_NET(rqstp), current_fh, open, stp);
3093 nodeleg:
3094 status = nfs_ok;
3096 dprintk("%s: stateid=" STATEID_FMT "\n", __func__,
3097 STATEID_VAL(&stp->st_stid.sc_stateid));
3098 out:
3099 /* 4.1 client trying to upgrade/downgrade delegation? */
3100 if (open->op_delegate_type == NFS4_OPEN_DELEGATE_NONE && dp &&
3101 open->op_deleg_want)
3102 nfsd4_deleg_xgrade_none_ext(open, dp);
3104 if (fp)
3105 put_nfs4_file(fp);
3106 if (status == 0 && open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS)
3107 nfs4_set_claim_prev(open, nfsd4_has_session(&resp->cstate));
3109 * To finish the open response, we just need to set the rflags.
3111 open->op_rflags = NFS4_OPEN_RESULT_LOCKTYPE_POSIX;
3112 if (!(open->op_openowner->oo_flags & NFS4_OO_CONFIRMED) &&
3113 !nfsd4_has_session(&resp->cstate))
3114 open->op_rflags |= NFS4_OPEN_RESULT_CONFIRM;
3116 return status;
3119 void nfsd4_cleanup_open_state(struct nfsd4_open *open, __be32 status)
3121 if (open->op_openowner) {
3122 struct nfs4_openowner *oo = open->op_openowner;
3124 if (!list_empty(&oo->oo_owner.so_stateids))
3125 list_del_init(&oo->oo_close_lru);
3126 if (oo->oo_flags & NFS4_OO_NEW) {
3127 if (status) {
3128 release_openowner(oo);
3129 open->op_openowner = NULL;
3130 } else
3131 oo->oo_flags &= ~NFS4_OO_NEW;
3134 if (open->op_file)
3135 nfsd4_free_file(open->op_file);
3136 if (open->op_stp)
3137 free_generic_stateid(open->op_stp);
3140 __be32
3141 nfsd4_renew(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3142 clientid_t *clid)
3144 struct nfs4_client *clp;
3145 __be32 status;
3146 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3148 nfs4_lock_state();
3149 dprintk("process_renew(%08x/%08x): starting\n",
3150 clid->cl_boot, clid->cl_id);
3151 status = nfserr_stale_clientid;
3152 if (STALE_CLIENTID(clid, nn))
3153 goto out;
3154 clp = find_confirmed_client(clid, cstate->minorversion, nn);
3155 status = nfserr_expired;
3156 if (clp == NULL) {
3157 /* We assume the client took too long to RENEW. */
3158 dprintk("nfsd4_renew: clientid not found!\n");
3159 goto out;
3161 status = nfserr_cb_path_down;
3162 if (!list_empty(&clp->cl_delegations)
3163 && clp->cl_cb_state != NFSD4_CB_UP)
3164 goto out;
3165 status = nfs_ok;
3166 out:
3167 nfs4_unlock_state();
3168 return status;
3171 static void
3172 nfsd4_end_grace(struct nfsd_net *nn)
3174 /* do nothing if grace period already ended */
3175 if (nn->grace_ended)
3176 return;
3178 dprintk("NFSD: end of grace period\n");
3179 nn->grace_ended = true;
3180 nfsd4_record_grace_done(nn, nn->boot_time);
3181 locks_end_grace(&nn->nfsd4_manager);
3183 * Now that every NFSv4 client has had the chance to recover and
3184 * to see the (possibly new, possibly shorter) lease time, we
3185 * can safely set the next grace time to the current lease time:
3187 nfsd4_grace = nfsd4_lease;
3190 static time_t
3191 nfs4_laundromat(struct nfsd_net *nn)
3193 struct nfs4_client *clp;
3194 struct nfs4_openowner *oo;
3195 struct nfs4_delegation *dp;
3196 struct list_head *pos, *next, reaplist;
3197 time_t cutoff = get_seconds() - nfsd4_lease;
3198 time_t t, clientid_val = nfsd4_lease;
3199 time_t u, test_val = nfsd4_lease;
3201 nfs4_lock_state();
3203 dprintk("NFSD: laundromat service - starting\n");
3204 nfsd4_end_grace(nn);
3205 INIT_LIST_HEAD(&reaplist);
3206 spin_lock(&nn->client_lock);
3207 list_for_each_safe(pos, next, &nn->client_lru) {
3208 clp = list_entry(pos, struct nfs4_client, cl_lru);
3209 if (time_after((unsigned long)clp->cl_time, (unsigned long)cutoff)) {
3210 t = clp->cl_time - cutoff;
3211 if (clientid_val > t)
3212 clientid_val = t;
3213 break;
3215 if (atomic_read(&clp->cl_refcount)) {
3216 dprintk("NFSD: client in use (clientid %08x)\n",
3217 clp->cl_clientid.cl_id);
3218 continue;
3220 unhash_client_locked(clp);
3221 list_add(&clp->cl_lru, &reaplist);
3223 spin_unlock(&nn->client_lock);
3224 list_for_each_safe(pos, next, &reaplist) {
3225 clp = list_entry(pos, struct nfs4_client, cl_lru);
3226 dprintk("NFSD: purging unused client (clientid %08x)\n",
3227 clp->cl_clientid.cl_id);
3228 expire_client(clp);
3230 spin_lock(&recall_lock);
3231 list_for_each_safe(pos, next, &del_recall_lru) {
3232 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
3233 if (net_generic(dp->dl_stid.sc_client->net, nfsd_net_id) != nn)
3234 continue;
3235 if (time_after((unsigned long)dp->dl_time, (unsigned long)cutoff)) {
3236 u = dp->dl_time - cutoff;
3237 if (test_val > u)
3238 test_val = u;
3239 break;
3241 list_move(&dp->dl_recall_lru, &reaplist);
3243 spin_unlock(&recall_lock);
3244 list_for_each_safe(pos, next, &reaplist) {
3245 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
3246 unhash_delegation(dp);
3248 test_val = nfsd4_lease;
3249 list_for_each_safe(pos, next, &nn->close_lru) {
3250 oo = container_of(pos, struct nfs4_openowner, oo_close_lru);
3251 if (time_after((unsigned long)oo->oo_time, (unsigned long)cutoff)) {
3252 u = oo->oo_time - cutoff;
3253 if (test_val > u)
3254 test_val = u;
3255 break;
3257 release_openowner(oo);
3259 if (clientid_val < NFSD_LAUNDROMAT_MINTIMEOUT)
3260 clientid_val = NFSD_LAUNDROMAT_MINTIMEOUT;
3261 nfs4_unlock_state();
3262 return clientid_val;
3265 static struct workqueue_struct *laundry_wq;
3266 static void laundromat_main(struct work_struct *);
3268 static void
3269 laundromat_main(struct work_struct *laundry)
3271 time_t t;
3272 struct delayed_work *dwork = container_of(laundry, struct delayed_work,
3273 work);
3274 struct nfsd_net *nn = container_of(dwork, struct nfsd_net,
3275 laundromat_work);
3277 t = nfs4_laundromat(nn);
3278 dprintk("NFSD: laundromat_main - sleeping for %ld seconds\n", t);
3279 queue_delayed_work(laundry_wq, &nn->laundromat_work, t*HZ);
3282 static inline __be32 nfs4_check_fh(struct svc_fh *fhp, struct nfs4_ol_stateid *stp)
3284 if (fhp->fh_dentry->d_inode != stp->st_file->fi_inode)
3285 return nfserr_bad_stateid;
3286 return nfs_ok;
3289 static int
3290 STALE_STATEID(stateid_t *stateid, struct nfsd_net *nn)
3292 if (stateid->si_opaque.so_clid.cl_boot == nn->boot_time)
3293 return 0;
3294 dprintk("NFSD: stale stateid " STATEID_FMT "!\n",
3295 STATEID_VAL(stateid));
3296 return 1;
3299 static inline int
3300 access_permit_read(struct nfs4_ol_stateid *stp)
3302 return test_access(NFS4_SHARE_ACCESS_READ, stp) ||
3303 test_access(NFS4_SHARE_ACCESS_BOTH, stp) ||
3304 test_access(NFS4_SHARE_ACCESS_WRITE, stp);
3307 static inline int
3308 access_permit_write(struct nfs4_ol_stateid *stp)
3310 return test_access(NFS4_SHARE_ACCESS_WRITE, stp) ||
3311 test_access(NFS4_SHARE_ACCESS_BOTH, stp);
3314 static
3315 __be32 nfs4_check_openmode(struct nfs4_ol_stateid *stp, int flags)
3317 __be32 status = nfserr_openmode;
3319 /* For lock stateid's, we test the parent open, not the lock: */
3320 if (stp->st_openstp)
3321 stp = stp->st_openstp;
3322 if ((flags & WR_STATE) && !access_permit_write(stp))
3323 goto out;
3324 if ((flags & RD_STATE) && !access_permit_read(stp))
3325 goto out;
3326 status = nfs_ok;
3327 out:
3328 return status;
3331 static inline __be32
3332 check_special_stateids(struct net *net, svc_fh *current_fh, stateid_t *stateid, int flags)
3334 if (ONE_STATEID(stateid) && (flags & RD_STATE))
3335 return nfs_ok;
3336 else if (locks_in_grace(net)) {
3337 /* Answer in remaining cases depends on existence of
3338 * conflicting state; so we must wait out the grace period. */
3339 return nfserr_grace;
3340 } else if (flags & WR_STATE)
3341 return nfs4_share_conflict(current_fh,
3342 NFS4_SHARE_DENY_WRITE);
3343 else /* (flags & RD_STATE) && ZERO_STATEID(stateid) */
3344 return nfs4_share_conflict(current_fh,
3345 NFS4_SHARE_DENY_READ);
3349 * Allow READ/WRITE during grace period on recovered state only for files
3350 * that are not able to provide mandatory locking.
3352 static inline int
3353 grace_disallows_io(struct net *net, struct inode *inode)
3355 return locks_in_grace(net) && mandatory_lock(inode);
3358 /* Returns true iff a is later than b: */
3359 static bool stateid_generation_after(stateid_t *a, stateid_t *b)
3361 return (s32)a->si_generation - (s32)b->si_generation > 0;
3364 static __be32 check_stateid_generation(stateid_t *in, stateid_t *ref, bool has_session)
3367 * When sessions are used the stateid generation number is ignored
3368 * when it is zero.
3370 if (has_session && in->si_generation == 0)
3371 return nfs_ok;
3373 if (in->si_generation == ref->si_generation)
3374 return nfs_ok;
3376 /* If the client sends us a stateid from the future, it's buggy: */
3377 if (stateid_generation_after(in, ref))
3378 return nfserr_bad_stateid;
3380 * However, we could see a stateid from the past, even from a
3381 * non-buggy client. For example, if the client sends a lock
3382 * while some IO is outstanding, the lock may bump si_generation
3383 * while the IO is still in flight. The client could avoid that
3384 * situation by waiting for responses on all the IO requests,
3385 * but better performance may result in retrying IO that
3386 * receives an old_stateid error if requests are rarely
3387 * reordered in flight:
3389 return nfserr_old_stateid;
3392 static __be32 nfsd4_validate_stateid(struct nfs4_client *cl, stateid_t *stateid)
3394 struct nfs4_stid *s;
3395 struct nfs4_ol_stateid *ols;
3396 __be32 status;
3398 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
3399 return nfserr_bad_stateid;
3400 /* Client debugging aid. */
3401 if (!same_clid(&stateid->si_opaque.so_clid, &cl->cl_clientid)) {
3402 char addr_str[INET6_ADDRSTRLEN];
3403 rpc_ntop((struct sockaddr *)&cl->cl_addr, addr_str,
3404 sizeof(addr_str));
3405 pr_warn_ratelimited("NFSD: client %s testing state ID "
3406 "with incorrect client ID\n", addr_str);
3407 return nfserr_bad_stateid;
3409 s = find_stateid(cl, stateid);
3410 if (!s)
3411 return nfserr_bad_stateid;
3412 status = check_stateid_generation(stateid, &s->sc_stateid, 1);
3413 if (status)
3414 return status;
3415 if (!(s->sc_type & (NFS4_OPEN_STID | NFS4_LOCK_STID)))
3416 return nfs_ok;
3417 ols = openlockstateid(s);
3418 if (ols->st_stateowner->so_is_open_owner
3419 && !(openowner(ols->st_stateowner)->oo_flags & NFS4_OO_CONFIRMED))
3420 return nfserr_bad_stateid;
3421 return nfs_ok;
3424 static __be32 nfsd4_lookup_stateid(stateid_t *stateid, unsigned char typemask,
3425 struct nfs4_stid **s, bool sessions,
3426 struct nfsd_net *nn)
3428 struct nfs4_client *cl;
3430 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
3431 return nfserr_bad_stateid;
3432 if (STALE_STATEID(stateid, nn))
3433 return nfserr_stale_stateid;
3434 cl = find_confirmed_client(&stateid->si_opaque.so_clid, sessions, nn);
3435 if (!cl)
3436 return nfserr_expired;
3437 *s = find_stateid_by_type(cl, stateid, typemask);
3438 if (!*s)
3439 return nfserr_bad_stateid;
3440 return nfs_ok;
3445 * Checks for stateid operations
3447 __be32
3448 nfs4_preprocess_stateid_op(struct net *net, struct nfsd4_compound_state *cstate,
3449 stateid_t *stateid, int flags, struct file **filpp)
3451 struct nfs4_stid *s;
3452 struct nfs4_ol_stateid *stp = NULL;
3453 struct nfs4_delegation *dp = NULL;
3454 struct svc_fh *current_fh = &cstate->current_fh;
3455 struct inode *ino = current_fh->fh_dentry->d_inode;
3456 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
3457 __be32 status;
3459 if (filpp)
3460 *filpp = NULL;
3462 if (grace_disallows_io(net, ino))
3463 return nfserr_grace;
3465 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
3466 return check_special_stateids(net, current_fh, stateid, flags);
3468 status = nfsd4_lookup_stateid(stateid, NFS4_DELEG_STID|NFS4_OPEN_STID|NFS4_LOCK_STID,
3469 &s, cstate->minorversion, nn);
3470 if (status)
3471 return status;
3472 status = check_stateid_generation(stateid, &s->sc_stateid, nfsd4_has_session(cstate));
3473 if (status)
3474 goto out;
3475 switch (s->sc_type) {
3476 case NFS4_DELEG_STID:
3477 dp = delegstateid(s);
3478 status = nfs4_check_delegmode(dp, flags);
3479 if (status)
3480 goto out;
3481 if (filpp) {
3482 *filpp = dp->dl_file->fi_deleg_file;
3483 if (!*filpp) {
3484 WARN_ON_ONCE(1);
3485 status = nfserr_serverfault;
3486 goto out;
3489 break;
3490 case NFS4_OPEN_STID:
3491 case NFS4_LOCK_STID:
3492 stp = openlockstateid(s);
3493 status = nfs4_check_fh(current_fh, stp);
3494 if (status)
3495 goto out;
3496 if (stp->st_stateowner->so_is_open_owner
3497 && !(openowner(stp->st_stateowner)->oo_flags & NFS4_OO_CONFIRMED))
3498 goto out;
3499 status = nfs4_check_openmode(stp, flags);
3500 if (status)
3501 goto out;
3502 if (filpp) {
3503 if (flags & RD_STATE)
3504 *filpp = find_readable_file(stp->st_file);
3505 else
3506 *filpp = find_writeable_file(stp->st_file);
3508 break;
3509 default:
3510 return nfserr_bad_stateid;
3512 status = nfs_ok;
3513 out:
3514 return status;
3517 static __be32
3518 nfsd4_free_lock_stateid(struct nfs4_ol_stateid *stp)
3520 if (check_for_locks(stp->st_file, lockowner(stp->st_stateowner)))
3521 return nfserr_locks_held;
3522 release_lock_stateid(stp);
3523 return nfs_ok;
3527 * Test if the stateid is valid
3529 __be32
3530 nfsd4_test_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3531 struct nfsd4_test_stateid *test_stateid)
3533 struct nfsd4_test_stateid_id *stateid;
3534 struct nfs4_client *cl = cstate->session->se_client;
3536 nfs4_lock_state();
3537 list_for_each_entry(stateid, &test_stateid->ts_stateid_list, ts_id_list)
3538 stateid->ts_id_status =
3539 nfsd4_validate_stateid(cl, &stateid->ts_id_stateid);
3540 nfs4_unlock_state();
3542 return nfs_ok;
3545 __be32
3546 nfsd4_free_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3547 struct nfsd4_free_stateid *free_stateid)
3549 stateid_t *stateid = &free_stateid->fr_stateid;
3550 struct nfs4_stid *s;
3551 struct nfs4_client *cl = cstate->session->se_client;
3552 __be32 ret = nfserr_bad_stateid;
3554 nfs4_lock_state();
3555 s = find_stateid(cl, stateid);
3556 if (!s)
3557 goto out;
3558 switch (s->sc_type) {
3559 case NFS4_DELEG_STID:
3560 ret = nfserr_locks_held;
3561 goto out;
3562 case NFS4_OPEN_STID:
3563 case NFS4_LOCK_STID:
3564 ret = check_stateid_generation(stateid, &s->sc_stateid, 1);
3565 if (ret)
3566 goto out;
3567 if (s->sc_type == NFS4_LOCK_STID)
3568 ret = nfsd4_free_lock_stateid(openlockstateid(s));
3569 else
3570 ret = nfserr_locks_held;
3571 break;
3572 default:
3573 ret = nfserr_bad_stateid;
3575 out:
3576 nfs4_unlock_state();
3577 return ret;
3580 static inline int
3581 setlkflg (int type)
3583 return (type == NFS4_READW_LT || type == NFS4_READ_LT) ?
3584 RD_STATE : WR_STATE;
3587 static __be32 nfs4_seqid_op_checks(struct nfsd4_compound_state *cstate, stateid_t *stateid, u32 seqid, struct nfs4_ol_stateid *stp)
3589 struct svc_fh *current_fh = &cstate->current_fh;
3590 struct nfs4_stateowner *sop = stp->st_stateowner;
3591 __be32 status;
3593 status = nfsd4_check_seqid(cstate, sop, seqid);
3594 if (status)
3595 return status;
3596 if (stp->st_stid.sc_type == NFS4_CLOSED_STID)
3598 * "Closed" stateid's exist *only* to return
3599 * nfserr_replay_me from the previous step.
3601 return nfserr_bad_stateid;
3602 status = check_stateid_generation(stateid, &stp->st_stid.sc_stateid, nfsd4_has_session(cstate));
3603 if (status)
3604 return status;
3605 return nfs4_check_fh(current_fh, stp);
3609 * Checks for sequence id mutating operations.
3611 static __be32
3612 nfs4_preprocess_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
3613 stateid_t *stateid, char typemask,
3614 struct nfs4_ol_stateid **stpp,
3615 struct nfsd_net *nn)
3617 __be32 status;
3618 struct nfs4_stid *s;
3620 dprintk("NFSD: %s: seqid=%d stateid = " STATEID_FMT "\n", __func__,
3621 seqid, STATEID_VAL(stateid));
3623 *stpp = NULL;
3624 status = nfsd4_lookup_stateid(stateid, typemask, &s,
3625 cstate->minorversion, nn);
3626 if (status)
3627 return status;
3628 *stpp = openlockstateid(s);
3629 cstate->replay_owner = (*stpp)->st_stateowner;
3631 return nfs4_seqid_op_checks(cstate, stateid, seqid, *stpp);
3634 static __be32 nfs4_preprocess_confirmed_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
3635 stateid_t *stateid, struct nfs4_ol_stateid **stpp, struct nfsd_net *nn)
3637 __be32 status;
3638 struct nfs4_openowner *oo;
3640 status = nfs4_preprocess_seqid_op(cstate, seqid, stateid,
3641 NFS4_OPEN_STID, stpp, nn);
3642 if (status)
3643 return status;
3644 oo = openowner((*stpp)->st_stateowner);
3645 if (!(oo->oo_flags & NFS4_OO_CONFIRMED))
3646 return nfserr_bad_stateid;
3647 return nfs_ok;
3650 __be32
3651 nfsd4_open_confirm(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3652 struct nfsd4_open_confirm *oc)
3654 __be32 status;
3655 struct nfs4_openowner *oo;
3656 struct nfs4_ol_stateid *stp;
3657 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3659 dprintk("NFSD: nfsd4_open_confirm on file %.*s\n",
3660 (int)cstate->current_fh.fh_dentry->d_name.len,
3661 cstate->current_fh.fh_dentry->d_name.name);
3663 status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0);
3664 if (status)
3665 return status;
3667 nfs4_lock_state();
3669 status = nfs4_preprocess_seqid_op(cstate,
3670 oc->oc_seqid, &oc->oc_req_stateid,
3671 NFS4_OPEN_STID, &stp, nn);
3672 if (status)
3673 goto out;
3674 oo = openowner(stp->st_stateowner);
3675 status = nfserr_bad_stateid;
3676 if (oo->oo_flags & NFS4_OO_CONFIRMED)
3677 goto out;
3678 oo->oo_flags |= NFS4_OO_CONFIRMED;
3679 update_stateid(&stp->st_stid.sc_stateid);
3680 memcpy(&oc->oc_resp_stateid, &stp->st_stid.sc_stateid, sizeof(stateid_t));
3681 dprintk("NFSD: %s: success, seqid=%d stateid=" STATEID_FMT "\n",
3682 __func__, oc->oc_seqid, STATEID_VAL(&stp->st_stid.sc_stateid));
3684 nfsd4_client_record_create(oo->oo_owner.so_client);
3685 status = nfs_ok;
3686 out:
3687 if (!cstate->replay_owner)
3688 nfs4_unlock_state();
3689 return status;
3692 static inline void nfs4_stateid_downgrade_bit(struct nfs4_ol_stateid *stp, u32 access)
3694 if (!test_access(access, stp))
3695 return;
3696 nfs4_file_put_access(stp->st_file, nfs4_access_to_omode(access));
3697 clear_access(access, stp);
3700 static inline void nfs4_stateid_downgrade(struct nfs4_ol_stateid *stp, u32 to_access)
3702 switch (to_access) {
3703 case NFS4_SHARE_ACCESS_READ:
3704 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_WRITE);
3705 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
3706 break;
3707 case NFS4_SHARE_ACCESS_WRITE:
3708 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_READ);
3709 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
3710 break;
3711 case NFS4_SHARE_ACCESS_BOTH:
3712 break;
3713 default:
3714 WARN_ON_ONCE(1);
3718 static void
3719 reset_union_bmap_deny(unsigned long deny, struct nfs4_ol_stateid *stp)
3721 int i;
3722 for (i = 0; i < 4; i++) {
3723 if ((i & deny) != i)
3724 clear_deny(i, stp);
3728 __be32
3729 nfsd4_open_downgrade(struct svc_rqst *rqstp,
3730 struct nfsd4_compound_state *cstate,
3731 struct nfsd4_open_downgrade *od)
3733 __be32 status;
3734 struct nfs4_ol_stateid *stp;
3735 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3737 dprintk("NFSD: nfsd4_open_downgrade on file %.*s\n",
3738 (int)cstate->current_fh.fh_dentry->d_name.len,
3739 cstate->current_fh.fh_dentry->d_name.name);
3741 /* We don't yet support WANT bits: */
3742 if (od->od_deleg_want)
3743 dprintk("NFSD: %s: od_deleg_want=0x%x ignored\n", __func__,
3744 od->od_deleg_want);
3746 nfs4_lock_state();
3747 status = nfs4_preprocess_confirmed_seqid_op(cstate, od->od_seqid,
3748 &od->od_stateid, &stp, nn);
3749 if (status)
3750 goto out;
3751 status = nfserr_inval;
3752 if (!test_access(od->od_share_access, stp)) {
3753 dprintk("NFSD: access not a subset current bitmap: 0x%lx, input access=%08x\n",
3754 stp->st_access_bmap, od->od_share_access);
3755 goto out;
3757 if (!test_deny(od->od_share_deny, stp)) {
3758 dprintk("NFSD:deny not a subset current bitmap: 0x%lx, input deny=%08x\n",
3759 stp->st_deny_bmap, od->od_share_deny);
3760 goto out;
3762 nfs4_stateid_downgrade(stp, od->od_share_access);
3764 reset_union_bmap_deny(od->od_share_deny, stp);
3766 update_stateid(&stp->st_stid.sc_stateid);
3767 memcpy(&od->od_stateid, &stp->st_stid.sc_stateid, sizeof(stateid_t));
3768 status = nfs_ok;
3769 out:
3770 if (!cstate->replay_owner)
3771 nfs4_unlock_state();
3772 return status;
3775 void nfsd4_purge_closed_stateid(struct nfs4_stateowner *so)
3777 struct nfs4_openowner *oo;
3778 struct nfs4_ol_stateid *s;
3780 if (!so->so_is_open_owner)
3781 return;
3782 oo = openowner(so);
3783 s = oo->oo_last_closed_stid;
3784 if (!s)
3785 return;
3786 if (!(oo->oo_flags & NFS4_OO_PURGE_CLOSE)) {
3787 /* Release the last_closed_stid on the next seqid bump: */
3788 oo->oo_flags |= NFS4_OO_PURGE_CLOSE;
3789 return;
3791 oo->oo_flags &= ~NFS4_OO_PURGE_CLOSE;
3792 release_last_closed_stateid(oo);
3795 static void nfsd4_close_open_stateid(struct nfs4_ol_stateid *s)
3797 unhash_open_stateid(s);
3798 s->st_stid.sc_type = NFS4_CLOSED_STID;
3802 * nfs4_unlock_state() called after encode
3804 __be32
3805 nfsd4_close(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3806 struct nfsd4_close *close)
3808 __be32 status;
3809 struct nfs4_openowner *oo;
3810 struct nfs4_ol_stateid *stp;
3811 struct net *net = SVC_NET(rqstp);
3812 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
3814 dprintk("NFSD: nfsd4_close on file %.*s\n",
3815 (int)cstate->current_fh.fh_dentry->d_name.len,
3816 cstate->current_fh.fh_dentry->d_name.name);
3818 nfs4_lock_state();
3819 status = nfs4_preprocess_seqid_op(cstate, close->cl_seqid,
3820 &close->cl_stateid,
3821 NFS4_OPEN_STID|NFS4_CLOSED_STID,
3822 &stp, nn);
3823 if (status)
3824 goto out;
3825 oo = openowner(stp->st_stateowner);
3826 status = nfs_ok;
3827 update_stateid(&stp->st_stid.sc_stateid);
3828 memcpy(&close->cl_stateid, &stp->st_stid.sc_stateid, sizeof(stateid_t));
3830 nfsd4_close_open_stateid(stp);
3831 release_last_closed_stateid(oo);
3832 oo->oo_last_closed_stid = stp;
3834 if (list_empty(&oo->oo_owner.so_stateids)) {
3835 if (cstate->minorversion) {
3836 release_openowner(oo);
3837 cstate->replay_owner = NULL;
3838 } else {
3840 * In the 4.0 case we need to keep the owners around a
3841 * little while to handle CLOSE replay.
3843 if (list_empty(&oo->oo_owner.so_stateids))
3844 move_to_close_lru(oo, SVC_NET(rqstp));
3847 out:
3848 if (!cstate->replay_owner)
3849 nfs4_unlock_state();
3850 return status;
3853 __be32
3854 nfsd4_delegreturn(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3855 struct nfsd4_delegreturn *dr)
3857 struct nfs4_delegation *dp;
3858 stateid_t *stateid = &dr->dr_stateid;
3859 struct nfs4_stid *s;
3860 __be32 status;
3861 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3863 if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
3864 return status;
3866 nfs4_lock_state();
3867 status = nfsd4_lookup_stateid(stateid, NFS4_DELEG_STID, &s,
3868 cstate->minorversion, nn);
3869 if (status)
3870 goto out;
3871 dp = delegstateid(s);
3872 status = check_stateid_generation(stateid, &dp->dl_stid.sc_stateid, nfsd4_has_session(cstate));
3873 if (status)
3874 goto out;
3876 unhash_delegation(dp);
3877 out:
3878 nfs4_unlock_state();
3880 return status;
3884 #define LOFF_OVERFLOW(start, len) ((u64)(len) > ~(u64)(start))
3886 #define LOCKOWNER_INO_HASH_MASK (LOCKOWNER_INO_HASH_SIZE - 1)
3888 static inline u64
3889 end_offset(u64 start, u64 len)
3891 u64 end;
3893 end = start + len;
3894 return end >= start ? end: NFS4_MAX_UINT64;
3897 /* last octet in a range */
3898 static inline u64
3899 last_byte_offset(u64 start, u64 len)
3901 u64 end;
3903 WARN_ON_ONCE(!len);
3904 end = start + len;
3905 return end > start ? end - 1: NFS4_MAX_UINT64;
3908 static unsigned int lockowner_ino_hashval(struct inode *inode, u32 cl_id, struct xdr_netobj *ownername)
3910 return (file_hashval(inode) + cl_id
3911 + opaque_hashval(ownername->data, ownername->len))
3912 & LOCKOWNER_INO_HASH_MASK;
3916 * TODO: Linux file offsets are _signed_ 64-bit quantities, which means that
3917 * we can't properly handle lock requests that go beyond the (2^63 - 1)-th
3918 * byte, because of sign extension problems. Since NFSv4 calls for 64-bit
3919 * locking, this prevents us from being completely protocol-compliant. The
3920 * real solution to this problem is to start using unsigned file offsets in
3921 * the VFS, but this is a very deep change!
3923 static inline void
3924 nfs4_transform_lock_offset(struct file_lock *lock)
3926 if (lock->fl_start < 0)
3927 lock->fl_start = OFFSET_MAX;
3928 if (lock->fl_end < 0)
3929 lock->fl_end = OFFSET_MAX;
3932 /* Hack!: For now, we're defining this just so we can use a pointer to it
3933 * as a unique cookie to identify our (NFSv4's) posix locks. */
3934 static const struct lock_manager_operations nfsd_posix_mng_ops = {
3937 static inline void
3938 nfs4_set_lock_denied(struct file_lock *fl, struct nfsd4_lock_denied *deny)
3940 struct nfs4_lockowner *lo;
3942 if (fl->fl_lmops == &nfsd_posix_mng_ops) {
3943 lo = (struct nfs4_lockowner *) fl->fl_owner;
3944 deny->ld_owner.data = kmemdup(lo->lo_owner.so_owner.data,
3945 lo->lo_owner.so_owner.len, GFP_KERNEL);
3946 if (!deny->ld_owner.data)
3947 /* We just don't care that much */
3948 goto nevermind;
3949 deny->ld_owner.len = lo->lo_owner.so_owner.len;
3950 deny->ld_clientid = lo->lo_owner.so_client->cl_clientid;
3951 } else {
3952 nevermind:
3953 deny->ld_owner.len = 0;
3954 deny->ld_owner.data = NULL;
3955 deny->ld_clientid.cl_boot = 0;
3956 deny->ld_clientid.cl_id = 0;
3958 deny->ld_start = fl->fl_start;
3959 deny->ld_length = NFS4_MAX_UINT64;
3960 if (fl->fl_end != NFS4_MAX_UINT64)
3961 deny->ld_length = fl->fl_end - fl->fl_start + 1;
3962 deny->ld_type = NFS4_READ_LT;
3963 if (fl->fl_type != F_RDLCK)
3964 deny->ld_type = NFS4_WRITE_LT;
3967 static bool same_lockowner_ino(struct nfs4_lockowner *lo, struct inode *inode, clientid_t *clid, struct xdr_netobj *owner)
3969 struct nfs4_ol_stateid *lst;
3971 if (!same_owner_str(&lo->lo_owner, owner, clid))
3972 return false;
3973 lst = list_first_entry(&lo->lo_owner.so_stateids,
3974 struct nfs4_ol_stateid, st_perstateowner);
3975 return lst->st_file->fi_inode == inode;
3978 static struct nfs4_lockowner *
3979 find_lockowner_str(struct inode *inode, clientid_t *clid,
3980 struct xdr_netobj *owner, struct nfsd_net *nn)
3982 unsigned int hashval = lockowner_ino_hashval(inode, clid->cl_id, owner);
3983 struct nfs4_lockowner *lo;
3985 list_for_each_entry(lo, &nn->lockowner_ino_hashtbl[hashval], lo_owner_ino_hash) {
3986 if (same_lockowner_ino(lo, inode, clid, owner))
3987 return lo;
3989 return NULL;
3992 static void hash_lockowner(struct nfs4_lockowner *lo, unsigned int strhashval, struct nfs4_client *clp, struct nfs4_ol_stateid *open_stp)
3994 struct inode *inode = open_stp->st_file->fi_inode;
3995 unsigned int inohash = lockowner_ino_hashval(inode,
3996 clp->cl_clientid.cl_id, &lo->lo_owner.so_owner);
3997 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
3999 list_add(&lo->lo_owner.so_strhash, &nn->ownerstr_hashtbl[strhashval]);
4000 list_add(&lo->lo_owner_ino_hash, &nn->lockowner_ino_hashtbl[inohash]);
4001 list_add(&lo->lo_perstateid, &open_stp->st_lockowners);
4005 * Alloc a lock owner structure.
4006 * Called in nfsd4_lock - therefore, OPEN and OPEN_CONFIRM (if needed) has
4007 * occurred.
4009 * strhashval = ownerstr_hashval
4012 static struct nfs4_lockowner *
4013 alloc_init_lock_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfs4_ol_stateid *open_stp, struct nfsd4_lock *lock) {
4014 struct nfs4_lockowner *lo;
4016 lo = alloc_stateowner(lockowner_slab, &lock->lk_new_owner, clp);
4017 if (!lo)
4018 return NULL;
4019 INIT_LIST_HEAD(&lo->lo_owner.so_stateids);
4020 lo->lo_owner.so_is_open_owner = 0;
4021 /* It is the openowner seqid that will be incremented in encode in the
4022 * case of new lockowners; so increment the lock seqid manually: */
4023 lo->lo_owner.so_seqid = lock->lk_new_lock_seqid + 1;
4024 hash_lockowner(lo, strhashval, clp, open_stp);
4025 return lo;
4028 static struct nfs4_ol_stateid *
4029 alloc_init_lock_stateid(struct nfs4_lockowner *lo, struct nfs4_file *fp, struct nfs4_ol_stateid *open_stp)
4031 struct nfs4_ol_stateid *stp;
4032 struct nfs4_client *clp = lo->lo_owner.so_client;
4034 stp = nfs4_alloc_stateid(clp);
4035 if (stp == NULL)
4036 return NULL;
4037 init_stid(&stp->st_stid, clp, NFS4_LOCK_STID);
4038 list_add(&stp->st_perfile, &fp->fi_stateids);
4039 list_add(&stp->st_perstateowner, &lo->lo_owner.so_stateids);
4040 stp->st_stateowner = &lo->lo_owner;
4041 get_nfs4_file(fp);
4042 stp->st_file = fp;
4043 stp->st_access_bmap = 0;
4044 stp->st_deny_bmap = open_stp->st_deny_bmap;
4045 stp->st_openstp = open_stp;
4046 return stp;
4049 static int
4050 check_lock_length(u64 offset, u64 length)
4052 return ((length == 0) || ((length != NFS4_MAX_UINT64) &&
4053 LOFF_OVERFLOW(offset, length)));
4056 static void get_lock_access(struct nfs4_ol_stateid *lock_stp, u32 access)
4058 struct nfs4_file *fp = lock_stp->st_file;
4059 int oflag = nfs4_access_to_omode(access);
4061 if (test_access(access, lock_stp))
4062 return;
4063 nfs4_file_get_access(fp, oflag);
4064 set_access(access, lock_stp);
4067 static __be32 lookup_or_create_lock_state(struct nfsd4_compound_state *cstate, struct nfs4_ol_stateid *ost, struct nfsd4_lock *lock, struct nfs4_ol_stateid **lst, bool *new)
4069 struct nfs4_file *fi = ost->st_file;
4070 struct nfs4_openowner *oo = openowner(ost->st_stateowner);
4071 struct nfs4_client *cl = oo->oo_owner.so_client;
4072 struct nfs4_lockowner *lo;
4073 unsigned int strhashval;
4074 struct nfsd_net *nn = net_generic(cl->net, nfsd_net_id);
4076 lo = find_lockowner_str(fi->fi_inode, &cl->cl_clientid,
4077 &lock->v.new.owner, nn);
4078 if (lo) {
4079 if (!cstate->minorversion)
4080 return nfserr_bad_seqid;
4081 /* XXX: a lockowner always has exactly one stateid: */
4082 *lst = list_first_entry(&lo->lo_owner.so_stateids,
4083 struct nfs4_ol_stateid, st_perstateowner);
4084 return nfs_ok;
4086 strhashval = ownerstr_hashval(cl->cl_clientid.cl_id,
4087 &lock->v.new.owner);
4088 lo = alloc_init_lock_stateowner(strhashval, cl, ost, lock);
4089 if (lo == NULL)
4090 return nfserr_jukebox;
4091 *lst = alloc_init_lock_stateid(lo, fi, ost);
4092 if (*lst == NULL) {
4093 release_lockowner(lo);
4094 return nfserr_jukebox;
4096 *new = true;
4097 return nfs_ok;
4101 * LOCK operation
4103 __be32
4104 nfsd4_lock(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4105 struct nfsd4_lock *lock)
4107 struct nfs4_openowner *open_sop = NULL;
4108 struct nfs4_lockowner *lock_sop = NULL;
4109 struct nfs4_ol_stateid *lock_stp;
4110 struct file *filp = NULL;
4111 struct file_lock *file_lock = NULL;
4112 struct file_lock *conflock = NULL;
4113 __be32 status = 0;
4114 bool new_state = false;
4115 int lkflg;
4116 int err;
4117 struct net *net = SVC_NET(rqstp);
4118 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
4120 dprintk("NFSD: nfsd4_lock: start=%Ld length=%Ld\n",
4121 (long long) lock->lk_offset,
4122 (long long) lock->lk_length);
4124 if (check_lock_length(lock->lk_offset, lock->lk_length))
4125 return nfserr_inval;
4127 if ((status = fh_verify(rqstp, &cstate->current_fh,
4128 S_IFREG, NFSD_MAY_LOCK))) {
4129 dprintk("NFSD: nfsd4_lock: permission denied!\n");
4130 return status;
4133 nfs4_lock_state();
4135 if (lock->lk_is_new) {
4136 struct nfs4_ol_stateid *open_stp = NULL;
4138 if (nfsd4_has_session(cstate))
4139 /* See rfc 5661 18.10.3: given clientid is ignored: */
4140 memcpy(&lock->v.new.clientid,
4141 &cstate->session->se_client->cl_clientid,
4142 sizeof(clientid_t));
4144 status = nfserr_stale_clientid;
4145 if (STALE_CLIENTID(&lock->lk_new_clientid, nn))
4146 goto out;
4148 /* validate and update open stateid and open seqid */
4149 status = nfs4_preprocess_confirmed_seqid_op(cstate,
4150 lock->lk_new_open_seqid,
4151 &lock->lk_new_open_stateid,
4152 &open_stp, nn);
4153 if (status)
4154 goto out;
4155 open_sop = openowner(open_stp->st_stateowner);
4156 status = nfserr_bad_stateid;
4157 if (!same_clid(&open_sop->oo_owner.so_client->cl_clientid,
4158 &lock->v.new.clientid))
4159 goto out;
4160 status = lookup_or_create_lock_state(cstate, open_stp, lock,
4161 &lock_stp, &new_state);
4162 } else
4163 status = nfs4_preprocess_seqid_op(cstate,
4164 lock->lk_old_lock_seqid,
4165 &lock->lk_old_lock_stateid,
4166 NFS4_LOCK_STID, &lock_stp, nn);
4167 if (status)
4168 goto out;
4169 lock_sop = lockowner(lock_stp->st_stateowner);
4171 lkflg = setlkflg(lock->lk_type);
4172 status = nfs4_check_openmode(lock_stp, lkflg);
4173 if (status)
4174 goto out;
4176 status = nfserr_grace;
4177 if (locks_in_grace(net) && !lock->lk_reclaim)
4178 goto out;
4179 status = nfserr_no_grace;
4180 if (!locks_in_grace(net) && lock->lk_reclaim)
4181 goto out;
4183 file_lock = locks_alloc_lock();
4184 if (!file_lock) {
4185 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
4186 status = nfserr_jukebox;
4187 goto out;
4190 locks_init_lock(file_lock);
4191 switch (lock->lk_type) {
4192 case NFS4_READ_LT:
4193 case NFS4_READW_LT:
4194 filp = find_readable_file(lock_stp->st_file);
4195 if (filp)
4196 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_READ);
4197 file_lock->fl_type = F_RDLCK;
4198 break;
4199 case NFS4_WRITE_LT:
4200 case NFS4_WRITEW_LT:
4201 filp = find_writeable_file(lock_stp->st_file);
4202 if (filp)
4203 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_WRITE);
4204 file_lock->fl_type = F_WRLCK;
4205 break;
4206 default:
4207 status = nfserr_inval;
4208 goto out;
4210 if (!filp) {
4211 status = nfserr_openmode;
4212 goto out;
4214 file_lock->fl_owner = (fl_owner_t)lock_sop;
4215 file_lock->fl_pid = current->tgid;
4216 file_lock->fl_file = filp;
4217 file_lock->fl_flags = FL_POSIX;
4218 file_lock->fl_lmops = &nfsd_posix_mng_ops;
4219 file_lock->fl_start = lock->lk_offset;
4220 file_lock->fl_end = last_byte_offset(lock->lk_offset, lock->lk_length);
4221 nfs4_transform_lock_offset(file_lock);
4223 conflock = locks_alloc_lock();
4224 if (!conflock) {
4225 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
4226 status = nfserr_jukebox;
4227 goto out;
4230 err = vfs_lock_file(filp, F_SETLK, file_lock, conflock);
4231 switch (-err) {
4232 case 0: /* success! */
4233 update_stateid(&lock_stp->st_stid.sc_stateid);
4234 memcpy(&lock->lk_resp_stateid, &lock_stp->st_stid.sc_stateid,
4235 sizeof(stateid_t));
4236 status = 0;
4237 break;
4238 case (EAGAIN): /* conflock holds conflicting lock */
4239 status = nfserr_denied;
4240 dprintk("NFSD: nfsd4_lock: conflicting lock found!\n");
4241 nfs4_set_lock_denied(conflock, &lock->lk_denied);
4242 break;
4243 case (EDEADLK):
4244 status = nfserr_deadlock;
4245 break;
4246 default:
4247 dprintk("NFSD: nfsd4_lock: vfs_lock_file() failed! status %d\n",err);
4248 status = nfserrno(err);
4249 break;
4251 out:
4252 if (status && new_state)
4253 release_lockowner(lock_sop);
4254 if (!cstate->replay_owner)
4255 nfs4_unlock_state();
4256 if (file_lock)
4257 locks_free_lock(file_lock);
4258 if (conflock)
4259 locks_free_lock(conflock);
4260 return status;
4264 * The NFSv4 spec allows a client to do a LOCKT without holding an OPEN,
4265 * so we do a temporary open here just to get an open file to pass to
4266 * vfs_test_lock. (Arguably perhaps test_lock should be done with an
4267 * inode operation.)
4269 static __be32 nfsd_test_lock(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file_lock *lock)
4271 struct file *file;
4272 __be32 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file);
4273 if (!err) {
4274 err = nfserrno(vfs_test_lock(file, lock));
4275 nfsd_close(file);
4277 return err;
4281 * LOCKT operation
4283 __be32
4284 nfsd4_lockt(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4285 struct nfsd4_lockt *lockt)
4287 struct inode *inode;
4288 struct file_lock *file_lock = NULL;
4289 struct nfs4_lockowner *lo;
4290 __be32 status;
4291 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
4293 if (locks_in_grace(SVC_NET(rqstp)))
4294 return nfserr_grace;
4296 if (check_lock_length(lockt->lt_offset, lockt->lt_length))
4297 return nfserr_inval;
4299 nfs4_lock_state();
4301 status = nfserr_stale_clientid;
4302 if (!nfsd4_has_session(cstate) && STALE_CLIENTID(&lockt->lt_clientid, nn))
4303 goto out;
4305 if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
4306 goto out;
4308 inode = cstate->current_fh.fh_dentry->d_inode;
4309 file_lock = locks_alloc_lock();
4310 if (!file_lock) {
4311 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
4312 status = nfserr_jukebox;
4313 goto out;
4315 locks_init_lock(file_lock);
4316 switch (lockt->lt_type) {
4317 case NFS4_READ_LT:
4318 case NFS4_READW_LT:
4319 file_lock->fl_type = F_RDLCK;
4320 break;
4321 case NFS4_WRITE_LT:
4322 case NFS4_WRITEW_LT:
4323 file_lock->fl_type = F_WRLCK;
4324 break;
4325 default:
4326 dprintk("NFSD: nfs4_lockt: bad lock type!\n");
4327 status = nfserr_inval;
4328 goto out;
4331 lo = find_lockowner_str(inode, &lockt->lt_clientid, &lockt->lt_owner, nn);
4332 if (lo)
4333 file_lock->fl_owner = (fl_owner_t)lo;
4334 file_lock->fl_pid = current->tgid;
4335 file_lock->fl_flags = FL_POSIX;
4337 file_lock->fl_start = lockt->lt_offset;
4338 file_lock->fl_end = last_byte_offset(lockt->lt_offset, lockt->lt_length);
4340 nfs4_transform_lock_offset(file_lock);
4342 status = nfsd_test_lock(rqstp, &cstate->current_fh, file_lock);
4343 if (status)
4344 goto out;
4346 if (file_lock->fl_type != F_UNLCK) {
4347 status = nfserr_denied;
4348 nfs4_set_lock_denied(file_lock, &lockt->lt_denied);
4350 out:
4351 nfs4_unlock_state();
4352 if (file_lock)
4353 locks_free_lock(file_lock);
4354 return status;
4357 __be32
4358 nfsd4_locku(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4359 struct nfsd4_locku *locku)
4361 struct nfs4_ol_stateid *stp;
4362 struct file *filp = NULL;
4363 struct file_lock *file_lock = NULL;
4364 __be32 status;
4365 int err;
4366 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
4368 dprintk("NFSD: nfsd4_locku: start=%Ld length=%Ld\n",
4369 (long long) locku->lu_offset,
4370 (long long) locku->lu_length);
4372 if (check_lock_length(locku->lu_offset, locku->lu_length))
4373 return nfserr_inval;
4375 nfs4_lock_state();
4377 status = nfs4_preprocess_seqid_op(cstate, locku->lu_seqid,
4378 &locku->lu_stateid, NFS4_LOCK_STID,
4379 &stp, nn);
4380 if (status)
4381 goto out;
4382 filp = find_any_file(stp->st_file);
4383 if (!filp) {
4384 status = nfserr_lock_range;
4385 goto out;
4387 file_lock = locks_alloc_lock();
4388 if (!file_lock) {
4389 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
4390 status = nfserr_jukebox;
4391 goto out;
4393 locks_init_lock(file_lock);
4394 file_lock->fl_type = F_UNLCK;
4395 file_lock->fl_owner = (fl_owner_t)lockowner(stp->st_stateowner);
4396 file_lock->fl_pid = current->tgid;
4397 file_lock->fl_file = filp;
4398 file_lock->fl_flags = FL_POSIX;
4399 file_lock->fl_lmops = &nfsd_posix_mng_ops;
4400 file_lock->fl_start = locku->lu_offset;
4402 file_lock->fl_end = last_byte_offset(locku->lu_offset,
4403 locku->lu_length);
4404 nfs4_transform_lock_offset(file_lock);
4407 * Try to unlock the file in the VFS.
4409 err = vfs_lock_file(filp, F_SETLK, file_lock, NULL);
4410 if (err) {
4411 dprintk("NFSD: nfs4_locku: vfs_lock_file failed!\n");
4412 goto out_nfserr;
4415 * OK, unlock succeeded; the only thing left to do is update the stateid.
4417 update_stateid(&stp->st_stid.sc_stateid);
4418 memcpy(&locku->lu_stateid, &stp->st_stid.sc_stateid, sizeof(stateid_t));
4420 out:
4421 if (!cstate->replay_owner)
4422 nfs4_unlock_state();
4423 if (file_lock)
4424 locks_free_lock(file_lock);
4425 return status;
4427 out_nfserr:
4428 status = nfserrno(err);
4429 goto out;
4433 * returns
4434 * 1: locks held by lockowner
4435 * 0: no locks held by lockowner
4437 static int
4438 check_for_locks(struct nfs4_file *filp, struct nfs4_lockowner *lowner)
4440 struct file_lock **flpp;
4441 struct inode *inode = filp->fi_inode;
4442 int status = 0;
4444 lock_flocks();
4445 for (flpp = &inode->i_flock; *flpp != NULL; flpp = &(*flpp)->fl_next) {
4446 if ((*flpp)->fl_owner == (fl_owner_t)lowner) {
4447 status = 1;
4448 goto out;
4451 out:
4452 unlock_flocks();
4453 return status;
4456 __be32
4457 nfsd4_release_lockowner(struct svc_rqst *rqstp,
4458 struct nfsd4_compound_state *cstate,
4459 struct nfsd4_release_lockowner *rlockowner)
4461 clientid_t *clid = &rlockowner->rl_clientid;
4462 struct nfs4_stateowner *sop;
4463 struct nfs4_lockowner *lo;
4464 struct nfs4_ol_stateid *stp;
4465 struct xdr_netobj *owner = &rlockowner->rl_owner;
4466 struct list_head matches;
4467 unsigned int hashval = ownerstr_hashval(clid->cl_id, owner);
4468 __be32 status;
4469 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
4471 dprintk("nfsd4_release_lockowner clientid: (%08x/%08x):\n",
4472 clid->cl_boot, clid->cl_id);
4474 /* XXX check for lease expiration */
4476 status = nfserr_stale_clientid;
4477 if (STALE_CLIENTID(clid, nn))
4478 return status;
4480 nfs4_lock_state();
4482 status = nfserr_locks_held;
4483 INIT_LIST_HEAD(&matches);
4485 list_for_each_entry(sop, &nn->ownerstr_hashtbl[hashval], so_strhash) {
4486 if (sop->so_is_open_owner)
4487 continue;
4488 if (!same_owner_str(sop, owner, clid))
4489 continue;
4490 list_for_each_entry(stp, &sop->so_stateids,
4491 st_perstateowner) {
4492 lo = lockowner(sop);
4493 if (check_for_locks(stp->st_file, lo))
4494 goto out;
4495 list_add(&lo->lo_list, &matches);
4498 /* Clients probably won't expect us to return with some (but not all)
4499 * of the lockowner state released; so don't release any until all
4500 * have been checked. */
4501 status = nfs_ok;
4502 while (!list_empty(&matches)) {
4503 lo = list_entry(matches.next, struct nfs4_lockowner,
4504 lo_list);
4505 /* unhash_stateowner deletes so_perclient only
4506 * for openowners. */
4507 list_del(&lo->lo_list);
4508 release_lockowner(lo);
4510 out:
4511 nfs4_unlock_state();
4512 return status;
4515 static inline struct nfs4_client_reclaim *
4516 alloc_reclaim(void)
4518 return kmalloc(sizeof(struct nfs4_client_reclaim), GFP_KERNEL);
4521 bool
4522 nfs4_has_reclaimed_state(const char *name, struct nfsd_net *nn)
4524 struct nfs4_client_reclaim *crp;
4526 crp = nfsd4_find_reclaim_client(name, nn);
4527 return (crp && crp->cr_clp);
4531 * failure => all reset bets are off, nfserr_no_grace...
4533 struct nfs4_client_reclaim *
4534 nfs4_client_to_reclaim(const char *name, struct nfsd_net *nn)
4536 unsigned int strhashval;
4537 struct nfs4_client_reclaim *crp;
4539 dprintk("NFSD nfs4_client_to_reclaim NAME: %.*s\n", HEXDIR_LEN, name);
4540 crp = alloc_reclaim();
4541 if (crp) {
4542 strhashval = clientstr_hashval(name);
4543 INIT_LIST_HEAD(&crp->cr_strhash);
4544 list_add(&crp->cr_strhash, &nn->reclaim_str_hashtbl[strhashval]);
4545 memcpy(crp->cr_recdir, name, HEXDIR_LEN);
4546 crp->cr_clp = NULL;
4547 nn->reclaim_str_hashtbl_size++;
4549 return crp;
4552 void
4553 nfs4_remove_reclaim_record(struct nfs4_client_reclaim *crp, struct nfsd_net *nn)
4555 list_del(&crp->cr_strhash);
4556 kfree(crp);
4557 nn->reclaim_str_hashtbl_size--;
4560 void
4561 nfs4_release_reclaim(struct nfsd_net *nn)
4563 struct nfs4_client_reclaim *crp = NULL;
4564 int i;
4566 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
4567 while (!list_empty(&nn->reclaim_str_hashtbl[i])) {
4568 crp = list_entry(nn->reclaim_str_hashtbl[i].next,
4569 struct nfs4_client_reclaim, cr_strhash);
4570 nfs4_remove_reclaim_record(crp, nn);
4573 WARN_ON_ONCE(nn->reclaim_str_hashtbl_size);
4577 * called from OPEN, CLAIM_PREVIOUS with a new clientid. */
4578 struct nfs4_client_reclaim *
4579 nfsd4_find_reclaim_client(const char *recdir, struct nfsd_net *nn)
4581 unsigned int strhashval;
4582 struct nfs4_client_reclaim *crp = NULL;
4584 dprintk("NFSD: nfs4_find_reclaim_client for recdir %s\n", recdir);
4586 strhashval = clientstr_hashval(recdir);
4587 list_for_each_entry(crp, &nn->reclaim_str_hashtbl[strhashval], cr_strhash) {
4588 if (same_name(crp->cr_recdir, recdir)) {
4589 return crp;
4592 return NULL;
4596 * Called from OPEN. Look for clientid in reclaim list.
4598 __be32
4599 nfs4_check_open_reclaim(clientid_t *clid, bool sessions, struct nfsd_net *nn)
4601 struct nfs4_client *clp;
4603 /* find clientid in conf_id_hashtbl */
4604 clp = find_confirmed_client(clid, sessions, nn);
4605 if (clp == NULL)
4606 return nfserr_reclaim_bad;
4608 return nfsd4_client_record_check(clp) ? nfserr_reclaim_bad : nfs_ok;
4611 #ifdef CONFIG_NFSD_FAULT_INJECTION
4613 void nfsd_forget_clients(u64 num)
4615 struct nfs4_client *clp, *next;
4616 int count = 0;
4617 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns, nfsd_net_id);
4619 nfs4_lock_state();
4620 list_for_each_entry_safe(clp, next, &nn->client_lru, cl_lru) {
4621 expire_client(clp);
4622 if (++count == num)
4623 break;
4625 nfs4_unlock_state();
4627 printk(KERN_INFO "NFSD: Forgot %d clients", count);
4630 static void release_lockowner_sop(struct nfs4_stateowner *sop)
4632 release_lockowner(lockowner(sop));
4635 static void release_openowner_sop(struct nfs4_stateowner *sop)
4637 release_openowner(openowner(sop));
4640 static int nfsd_release_n_owners(u64 num, bool is_open_owner,
4641 void (*release_sop)(struct nfs4_stateowner *),
4642 struct nfsd_net *nn)
4644 int i, count = 0;
4645 struct nfs4_stateowner *sop, *next;
4647 for (i = 0; i < OWNER_HASH_SIZE; i++) {
4648 list_for_each_entry_safe(sop, next, &nn->ownerstr_hashtbl[i], so_strhash) {
4649 if (sop->so_is_open_owner != is_open_owner)
4650 continue;
4651 release_sop(sop);
4652 if (++count == num)
4653 return count;
4656 return count;
4659 void nfsd_forget_locks(u64 num)
4661 int count;
4662 struct nfsd_net *nn = net_generic(&init_net, nfsd_net_id);
4664 nfs4_lock_state();
4665 count = nfsd_release_n_owners(num, false, release_lockowner_sop, nn);
4666 nfs4_unlock_state();
4668 printk(KERN_INFO "NFSD: Forgot %d locks", count);
4671 void nfsd_forget_openowners(u64 num)
4673 int count;
4674 struct nfsd_net *nn = net_generic(&init_net, nfsd_net_id);
4676 nfs4_lock_state();
4677 count = nfsd_release_n_owners(num, true, release_openowner_sop, nn);
4678 nfs4_unlock_state();
4680 printk(KERN_INFO "NFSD: Forgot %d open owners", count);
4683 static int nfsd_process_n_delegations(u64 num, struct list_head *list)
4685 int i, count = 0;
4686 struct nfs4_file *fp, *fnext;
4687 struct nfs4_delegation *dp, *dnext;
4689 for (i = 0; i < FILE_HASH_SIZE; i++) {
4690 list_for_each_entry_safe(fp, fnext, &file_hashtbl[i], fi_hash) {
4691 list_for_each_entry_safe(dp, dnext, &fp->fi_delegations, dl_perfile) {
4692 list_move(&dp->dl_recall_lru, list);
4693 if (++count == num)
4694 return count;
4699 return count;
4702 void nfsd_forget_delegations(u64 num)
4704 unsigned int count;
4705 LIST_HEAD(victims);
4706 struct nfs4_delegation *dp, *dnext;
4708 spin_lock(&recall_lock);
4709 count = nfsd_process_n_delegations(num, &victims);
4710 spin_unlock(&recall_lock);
4712 nfs4_lock_state();
4713 list_for_each_entry_safe(dp, dnext, &victims, dl_recall_lru)
4714 unhash_delegation(dp);
4715 nfs4_unlock_state();
4717 printk(KERN_INFO "NFSD: Forgot %d delegations", count);
4720 void nfsd_recall_delegations(u64 num)
4722 unsigned int count;
4723 LIST_HEAD(victims);
4724 struct nfs4_delegation *dp, *dnext;
4726 spin_lock(&recall_lock);
4727 count = nfsd_process_n_delegations(num, &victims);
4728 list_for_each_entry_safe(dp, dnext, &victims, dl_recall_lru) {
4729 list_del(&dp->dl_recall_lru);
4730 nfsd_break_one_deleg(dp);
4732 spin_unlock(&recall_lock);
4734 printk(KERN_INFO "NFSD: Recalled %d delegations", count);
4737 #endif /* CONFIG_NFSD_FAULT_INJECTION */
4739 /* initialization to perform at module load time: */
4741 void
4742 nfs4_state_init(void)
4744 int i;
4746 for (i = 0; i < FILE_HASH_SIZE; i++) {
4747 INIT_LIST_HEAD(&file_hashtbl[i]);
4749 INIT_LIST_HEAD(&del_recall_lru);
4753 * Since the lifetime of a delegation isn't limited to that of an open, a
4754 * client may quite reasonably hang on to a delegation as long as it has
4755 * the inode cached. This becomes an obvious problem the first time a
4756 * client's inode cache approaches the size of the server's total memory.
4758 * For now we avoid this problem by imposing a hard limit on the number
4759 * of delegations, which varies according to the server's memory size.
4761 static void
4762 set_max_delegations(void)
4765 * Allow at most 4 delegations per megabyte of RAM. Quick
4766 * estimates suggest that in the worst case (where every delegation
4767 * is for a different inode), a delegation could take about 1.5K,
4768 * giving a worst case usage of about 6% of memory.
4770 max_delegations = nr_free_buffer_pages() >> (20 - 2 - PAGE_SHIFT);
4773 static int nfs4_state_start_net(struct net *net)
4775 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
4776 int i;
4778 nn->conf_id_hashtbl = kmalloc(sizeof(struct list_head) *
4779 CLIENT_HASH_SIZE, GFP_KERNEL);
4780 if (!nn->conf_id_hashtbl)
4781 goto err;
4782 nn->unconf_id_hashtbl = kmalloc(sizeof(struct list_head) *
4783 CLIENT_HASH_SIZE, GFP_KERNEL);
4784 if (!nn->unconf_id_hashtbl)
4785 goto err_unconf_id;
4786 nn->ownerstr_hashtbl = kmalloc(sizeof(struct list_head) *
4787 OWNER_HASH_SIZE, GFP_KERNEL);
4788 if (!nn->ownerstr_hashtbl)
4789 goto err_ownerstr;
4790 nn->lockowner_ino_hashtbl = kmalloc(sizeof(struct list_head) *
4791 LOCKOWNER_INO_HASH_SIZE, GFP_KERNEL);
4792 if (!nn->lockowner_ino_hashtbl)
4793 goto err_lockowner_ino;
4794 nn->sessionid_hashtbl = kmalloc(sizeof(struct list_head) *
4795 SESSION_HASH_SIZE, GFP_KERNEL);
4796 if (!nn->sessionid_hashtbl)
4797 goto err_sessionid;
4799 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
4800 INIT_LIST_HEAD(&nn->conf_id_hashtbl[i]);
4801 INIT_LIST_HEAD(&nn->unconf_id_hashtbl[i]);
4803 for (i = 0; i < OWNER_HASH_SIZE; i++)
4804 INIT_LIST_HEAD(&nn->ownerstr_hashtbl[i]);
4805 for (i = 0; i < LOCKOWNER_INO_HASH_SIZE; i++)
4806 INIT_LIST_HEAD(&nn->lockowner_ino_hashtbl[i]);
4807 for (i = 0; i < SESSION_HASH_SIZE; i++)
4808 INIT_LIST_HEAD(&nn->sessionid_hashtbl[i]);
4809 nn->conf_name_tree = RB_ROOT;
4810 nn->unconf_name_tree = RB_ROOT;
4811 INIT_LIST_HEAD(&nn->client_lru);
4812 INIT_LIST_HEAD(&nn->close_lru);
4813 spin_lock_init(&nn->client_lock);
4815 INIT_DELAYED_WORK(&nn->laundromat_work, laundromat_main);
4817 return 0;
4819 err_sessionid:
4820 kfree(nn->lockowner_ino_hashtbl);
4821 err_lockowner_ino:
4822 kfree(nn->ownerstr_hashtbl);
4823 err_ownerstr:
4824 kfree(nn->unconf_id_hashtbl);
4825 err_unconf_id:
4826 kfree(nn->conf_id_hashtbl);
4827 err:
4828 return -ENOMEM;
4831 static void
4832 __nfs4_state_shutdown_net(struct net *net)
4834 int i;
4835 struct nfs4_client *clp = NULL;
4836 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
4837 struct rb_node *node, *tmp;
4839 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
4840 while (!list_empty(&nn->conf_id_hashtbl[i])) {
4841 clp = list_entry(nn->conf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
4842 destroy_client(clp);
4846 node = rb_first(&nn->unconf_name_tree);
4847 while (node != NULL) {
4848 tmp = node;
4849 node = rb_next(tmp);
4850 clp = rb_entry(tmp, struct nfs4_client, cl_namenode);
4851 rb_erase(tmp, &nn->unconf_name_tree);
4852 destroy_client(clp);
4855 kfree(nn->sessionid_hashtbl);
4856 kfree(nn->lockowner_ino_hashtbl);
4857 kfree(nn->ownerstr_hashtbl);
4858 kfree(nn->unconf_id_hashtbl);
4859 kfree(nn->conf_id_hashtbl);
4862 /* initialization to perform when the nfsd service is started: */
4865 nfs4_state_start(void)
4867 struct net *net = &init_net;
4868 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
4869 int ret;
4872 * FIXME: For now, we hang most of the pernet global stuff off of
4873 * init_net until nfsd is fully containerized. Eventually, we'll
4874 * need to pass a net pointer into this function, take a reference
4875 * to that instead and then do most of the rest of this on a per-net
4876 * basis.
4878 get_net(net);
4879 ret = nfs4_state_start_net(net);
4880 if (ret)
4881 return ret;
4882 nfsd4_client_tracking_init(net);
4883 nn->boot_time = get_seconds();
4884 locks_start_grace(net, &nn->nfsd4_manager);
4885 nn->grace_ended = false;
4886 printk(KERN_INFO "NFSD: starting %ld-second grace period\n",
4887 nfsd4_grace);
4888 ret = set_callback_cred();
4889 if (ret) {
4890 ret = -ENOMEM;
4891 goto out_recovery;
4893 laundry_wq = create_singlethread_workqueue("nfsd4");
4894 if (laundry_wq == NULL) {
4895 ret = -ENOMEM;
4896 goto out_recovery;
4898 ret = nfsd4_create_callback_queue();
4899 if (ret)
4900 goto out_free_laundry;
4902 queue_delayed_work(laundry_wq, &nn->laundromat_work, nfsd4_grace * HZ);
4903 set_max_delegations();
4904 return 0;
4905 out_free_laundry:
4906 destroy_workqueue(laundry_wq);
4907 out_recovery:
4908 nfsd4_client_tracking_exit(net);
4909 __nfs4_state_shutdown_net(net);
4910 put_net(net);
4911 return ret;
4914 /* should be called with the state lock held */
4915 static void
4916 __nfs4_state_shutdown(struct net *net)
4918 struct nfs4_delegation *dp = NULL;
4919 struct list_head *pos, *next, reaplist;
4921 __nfs4_state_shutdown_net(net);
4923 INIT_LIST_HEAD(&reaplist);
4924 spin_lock(&recall_lock);
4925 list_for_each_safe(pos, next, &del_recall_lru) {
4926 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
4927 if (dp->dl_stid.sc_client->net != net)
4928 continue;
4929 list_move(&dp->dl_recall_lru, &reaplist);
4931 spin_unlock(&recall_lock);
4932 list_for_each_safe(pos, next, &reaplist) {
4933 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
4934 unhash_delegation(dp);
4937 nfsd4_client_tracking_exit(net);
4938 put_net(net);
4941 void
4942 nfs4_state_shutdown(void)
4944 struct net *net = &init_net;
4945 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
4947 cancel_delayed_work_sync(&nn->laundromat_work);
4948 destroy_workqueue(laundry_wq);
4949 locks_end_grace(&nn->nfsd4_manager);
4950 __nfs4_state_shutdown(net);
4951 nfsd4_destroy_callback_queue();
4954 static void
4955 get_stateid(struct nfsd4_compound_state *cstate, stateid_t *stateid)
4957 if (HAS_STATE_ID(cstate, CURRENT_STATE_ID_FLAG) && CURRENT_STATEID(stateid))
4958 memcpy(stateid, &cstate->current_stateid, sizeof(stateid_t));
4961 static void
4962 put_stateid(struct nfsd4_compound_state *cstate, stateid_t *stateid)
4964 if (cstate->minorversion) {
4965 memcpy(&cstate->current_stateid, stateid, sizeof(stateid_t));
4966 SET_STATE_ID(cstate, CURRENT_STATE_ID_FLAG);
4970 void
4971 clear_current_stateid(struct nfsd4_compound_state *cstate)
4973 CLEAR_STATE_ID(cstate, CURRENT_STATE_ID_FLAG);
4977 * functions to set current state id
4979 void
4980 nfsd4_set_opendowngradestateid(struct nfsd4_compound_state *cstate, struct nfsd4_open_downgrade *odp)
4982 put_stateid(cstate, &odp->od_stateid);
4985 void
4986 nfsd4_set_openstateid(struct nfsd4_compound_state *cstate, struct nfsd4_open *open)
4988 put_stateid(cstate, &open->op_stateid);
4991 void
4992 nfsd4_set_closestateid(struct nfsd4_compound_state *cstate, struct nfsd4_close *close)
4994 put_stateid(cstate, &close->cl_stateid);
4997 void
4998 nfsd4_set_lockstateid(struct nfsd4_compound_state *cstate, struct nfsd4_lock *lock)
5000 put_stateid(cstate, &lock->lk_resp_stateid);
5004 * functions to consume current state id
5007 void
5008 nfsd4_get_opendowngradestateid(struct nfsd4_compound_state *cstate, struct nfsd4_open_downgrade *odp)
5010 get_stateid(cstate, &odp->od_stateid);
5013 void
5014 nfsd4_get_delegreturnstateid(struct nfsd4_compound_state *cstate, struct nfsd4_delegreturn *drp)
5016 get_stateid(cstate, &drp->dr_stateid);
5019 void
5020 nfsd4_get_freestateid(struct nfsd4_compound_state *cstate, struct nfsd4_free_stateid *fsp)
5022 get_stateid(cstate, &fsp->fr_stateid);
5025 void
5026 nfsd4_get_setattrstateid(struct nfsd4_compound_state *cstate, struct nfsd4_setattr *setattr)
5028 get_stateid(cstate, &setattr->sa_stateid);
5031 void
5032 nfsd4_get_closestateid(struct nfsd4_compound_state *cstate, struct nfsd4_close *close)
5034 get_stateid(cstate, &close->cl_stateid);
5037 void
5038 nfsd4_get_lockustateid(struct nfsd4_compound_state *cstate, struct nfsd4_locku *locku)
5040 get_stateid(cstate, &locku->lu_stateid);
5043 void
5044 nfsd4_get_readstateid(struct nfsd4_compound_state *cstate, struct nfsd4_read *read)
5046 get_stateid(cstate, &read->rd_stateid);
5049 void
5050 nfsd4_get_writestateid(struct nfsd4_compound_state *cstate, struct nfsd4_write *write)
5052 get_stateid(cstate, &write->wr_stateid);