nfsd4: remove redundant is_open_owner check
[linux-2.6/libata-dev.git] / fs / nfsd / nfs4state.c
blobd2b637b717c3f617b774b8492bd5a611ff340e23
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/sunrpc/svcauth_gss.h>
42 #include <linux/sunrpc/clnt.h>
43 #include "xdr4.h"
44 #include "vfs.h"
46 #define NFSDDBG_FACILITY NFSDDBG_PROC
48 /* Globals */
49 time_t nfsd4_lease = 90; /* default lease time */
50 time_t nfsd4_grace = 90;
51 static time_t boot_time;
52 static u32 current_ownerid = 1;
53 static u32 current_fileid = 1;
54 static u32 current_delegid = 1;
55 static stateid_t zerostateid; /* bits all 0 */
56 static stateid_t onestateid; /* bits all 1 */
57 static u64 current_sessionid = 1;
59 #define ZERO_STATEID(stateid) (!memcmp((stateid), &zerostateid, sizeof(stateid_t)))
60 #define ONE_STATEID(stateid) (!memcmp((stateid), &onestateid, sizeof(stateid_t)))
62 /* forward declarations */
63 static struct nfs4_stateid * find_stateid(stateid_t *stid, int flags);
64 static struct nfs4_stateid * search_for_stateid(stateid_t *stid);
65 static struct nfs4_delegation * search_for_delegation(stateid_t *stid);
66 static struct nfs4_delegation * find_delegation_stateid(struct inode *ino, stateid_t *stid);
67 static int check_for_locks(struct nfs4_file *filp, struct nfs4_stateowner *lowner);
69 /* Locking: */
71 /* Currently used for almost all code touching nfsv4 state: */
72 static DEFINE_MUTEX(client_mutex);
75 * Currently used for the del_recall_lru and file hash table. In an
76 * effort to decrease the scope of the client_mutex, this spinlock may
77 * eventually cover more:
79 static DEFINE_SPINLOCK(recall_lock);
81 static struct kmem_cache *stateowner_slab = NULL;
82 static struct kmem_cache *file_slab = NULL;
83 static struct kmem_cache *stateid_slab = NULL;
84 static struct kmem_cache *deleg_slab = NULL;
86 void
87 nfs4_lock_state(void)
89 mutex_lock(&client_mutex);
92 void
93 nfs4_unlock_state(void)
95 mutex_unlock(&client_mutex);
98 static inline u32
99 opaque_hashval(const void *ptr, int nbytes)
101 unsigned char *cptr = (unsigned char *) ptr;
103 u32 x = 0;
104 while (nbytes--) {
105 x *= 37;
106 x += *cptr++;
108 return x;
111 static struct list_head del_recall_lru;
113 static inline void
114 put_nfs4_file(struct nfs4_file *fi)
116 if (atomic_dec_and_lock(&fi->fi_ref, &recall_lock)) {
117 list_del(&fi->fi_hash);
118 spin_unlock(&recall_lock);
119 iput(fi->fi_inode);
120 kmem_cache_free(file_slab, fi);
124 static inline void
125 get_nfs4_file(struct nfs4_file *fi)
127 atomic_inc(&fi->fi_ref);
130 static int num_delegations;
131 unsigned int max_delegations;
134 * Open owner state (share locks)
137 /* hash tables for open owners */
138 #define OPEN_OWNER_HASH_BITS 8
139 #define OPEN_OWNER_HASH_SIZE (1 << OPEN_OWNER_HASH_BITS)
140 #define OPEN_OWNER_HASH_MASK (OPEN_OWNER_HASH_SIZE - 1)
142 static unsigned int open_ownerid_hashval(const u32 id)
144 return id & OPEN_OWNER_HASH_MASK;
147 static unsigned int open_ownerstr_hashval(u32 clientid, struct xdr_netobj *ownername)
149 unsigned int ret;
151 ret = opaque_hashval(ownername->data, ownername->len);
152 ret += clientid;
153 return ret & OPEN_OWNER_HASH_MASK;
156 static struct list_head open_ownerid_hashtbl[OPEN_OWNER_HASH_SIZE];
157 static struct list_head open_ownerstr_hashtbl[OPEN_OWNER_HASH_SIZE];
159 /* hash table for nfs4_file */
160 #define FILE_HASH_BITS 8
161 #define FILE_HASH_SIZE (1 << FILE_HASH_BITS)
163 /* hash table for (open)nfs4_stateid */
164 #define STATEID_HASH_BITS 10
165 #define STATEID_HASH_SIZE (1 << STATEID_HASH_BITS)
166 #define STATEID_HASH_MASK (STATEID_HASH_SIZE - 1)
168 static unsigned int file_hashval(struct inode *ino)
170 /* XXX: why are we hashing on inode pointer, anyway? */
171 return hash_ptr(ino, FILE_HASH_BITS);
174 static unsigned int stateid_hashval(u32 owner_id, u32 file_id)
176 return (owner_id + file_id) & STATEID_HASH_MASK;
179 static struct list_head file_hashtbl[FILE_HASH_SIZE];
180 static struct list_head stateid_hashtbl[STATEID_HASH_SIZE];
182 static void __nfs4_file_get_access(struct nfs4_file *fp, int oflag)
184 BUG_ON(!(fp->fi_fds[oflag] || fp->fi_fds[O_RDWR]));
185 atomic_inc(&fp->fi_access[oflag]);
188 static void nfs4_file_get_access(struct nfs4_file *fp, int oflag)
190 if (oflag == O_RDWR) {
191 __nfs4_file_get_access(fp, O_RDONLY);
192 __nfs4_file_get_access(fp, O_WRONLY);
193 } else
194 __nfs4_file_get_access(fp, oflag);
197 static void nfs4_file_put_fd(struct nfs4_file *fp, int oflag)
199 if (fp->fi_fds[oflag]) {
200 fput(fp->fi_fds[oflag]);
201 fp->fi_fds[oflag] = NULL;
205 static void __nfs4_file_put_access(struct nfs4_file *fp, int oflag)
207 if (atomic_dec_and_test(&fp->fi_access[oflag])) {
208 nfs4_file_put_fd(fp, O_RDWR);
209 nfs4_file_put_fd(fp, oflag);
213 static void nfs4_file_put_access(struct nfs4_file *fp, int oflag)
215 if (oflag == O_RDWR) {
216 __nfs4_file_put_access(fp, O_RDONLY);
217 __nfs4_file_put_access(fp, O_WRONLY);
218 } else
219 __nfs4_file_put_access(fp, oflag);
222 static struct nfs4_delegation *
223 alloc_init_deleg(struct nfs4_client *clp, struct nfs4_stateid *stp, struct svc_fh *current_fh, u32 type)
225 struct nfs4_delegation *dp;
226 struct nfs4_file *fp = stp->st_file;
228 dprintk("NFSD alloc_init_deleg\n");
230 * Major work on the lease subsystem (for example, to support
231 * calbacks on stat) will be required before we can support
232 * write delegations properly.
234 if (type != NFS4_OPEN_DELEGATE_READ)
235 return NULL;
236 if (fp->fi_had_conflict)
237 return NULL;
238 if (num_delegations > max_delegations)
239 return NULL;
240 dp = kmem_cache_alloc(deleg_slab, GFP_KERNEL);
241 if (dp == NULL)
242 return dp;
243 num_delegations++;
244 INIT_LIST_HEAD(&dp->dl_perfile);
245 INIT_LIST_HEAD(&dp->dl_perclnt);
246 INIT_LIST_HEAD(&dp->dl_recall_lru);
247 dp->dl_client = clp;
248 get_nfs4_file(fp);
249 dp->dl_file = fp;
250 dp->dl_type = type;
251 dp->dl_stateid.si_boot = boot_time;
252 dp->dl_stateid.si_stateownerid = current_delegid++;
253 dp->dl_stateid.si_fileid = 0;
254 dp->dl_stateid.si_generation = 0;
255 fh_copy_shallow(&dp->dl_fh, &current_fh->fh_handle);
256 dp->dl_time = 0;
257 atomic_set(&dp->dl_count, 1);
258 INIT_WORK(&dp->dl_recall.cb_work, nfsd4_do_callback_rpc);
259 return dp;
262 void
263 nfs4_put_delegation(struct nfs4_delegation *dp)
265 if (atomic_dec_and_test(&dp->dl_count)) {
266 dprintk("NFSD: freeing dp %p\n",dp);
267 put_nfs4_file(dp->dl_file);
268 kmem_cache_free(deleg_slab, dp);
269 num_delegations--;
273 static void nfs4_put_deleg_lease(struct nfs4_file *fp)
275 if (atomic_dec_and_test(&fp->fi_delegees)) {
276 vfs_setlease(fp->fi_deleg_file, F_UNLCK, &fp->fi_lease);
277 fp->fi_lease = NULL;
278 fput(fp->fi_deleg_file);
279 fp->fi_deleg_file = NULL;
283 /* Called under the state lock. */
284 static void
285 unhash_delegation(struct nfs4_delegation *dp)
287 list_del_init(&dp->dl_perclnt);
288 spin_lock(&recall_lock);
289 list_del_init(&dp->dl_perfile);
290 list_del_init(&dp->dl_recall_lru);
291 spin_unlock(&recall_lock);
292 nfs4_put_deleg_lease(dp->dl_file);
293 nfs4_put_delegation(dp);
297 * SETCLIENTID state
300 /* client_lock protects the client lru list and session hash table */
301 static DEFINE_SPINLOCK(client_lock);
303 /* Hash tables for nfs4_clientid state */
304 #define CLIENT_HASH_BITS 4
305 #define CLIENT_HASH_SIZE (1 << CLIENT_HASH_BITS)
306 #define CLIENT_HASH_MASK (CLIENT_HASH_SIZE - 1)
308 static unsigned int clientid_hashval(u32 id)
310 return id & CLIENT_HASH_MASK;
313 static unsigned int clientstr_hashval(const char *name)
315 return opaque_hashval(name, 8) & CLIENT_HASH_MASK;
319 * reclaim_str_hashtbl[] holds known client info from previous reset/reboot
320 * used in reboot/reset lease grace period processing
322 * conf_id_hashtbl[], and conf_str_hashtbl[] hold confirmed
323 * setclientid_confirmed info.
325 * unconf_str_hastbl[] and unconf_id_hashtbl[] hold unconfirmed
326 * setclientid info.
328 * client_lru holds client queue ordered by nfs4_client.cl_time
329 * for lease renewal.
331 * close_lru holds (open) stateowner queue ordered by nfs4_stateowner.so_time
332 * for last close replay.
334 static struct list_head reclaim_str_hashtbl[CLIENT_HASH_SIZE];
335 static int reclaim_str_hashtbl_size = 0;
336 static struct list_head conf_id_hashtbl[CLIENT_HASH_SIZE];
337 static struct list_head conf_str_hashtbl[CLIENT_HASH_SIZE];
338 static struct list_head unconf_str_hashtbl[CLIENT_HASH_SIZE];
339 static struct list_head unconf_id_hashtbl[CLIENT_HASH_SIZE];
340 static struct list_head client_lru;
341 static struct list_head close_lru;
344 * We store the NONE, READ, WRITE, and BOTH bits separately in the
345 * st_{access,deny}_bmap field of the stateid, in order to track not
346 * only what share bits are currently in force, but also what
347 * combinations of share bits previous opens have used. This allows us
348 * to enforce the recommendation of rfc 3530 14.2.19 that the server
349 * return an error if the client attempt to downgrade to a combination
350 * of share bits not explicable by closing some of its previous opens.
352 * XXX: This enforcement is actually incomplete, since we don't keep
353 * track of access/deny bit combinations; so, e.g., we allow:
355 * OPEN allow read, deny write
356 * OPEN allow both, deny none
357 * DOWNGRADE allow read, deny none
359 * which we should reject.
361 static void
362 set_access(unsigned int *access, unsigned long bmap) {
363 int i;
365 *access = 0;
366 for (i = 1; i < 4; i++) {
367 if (test_bit(i, &bmap))
368 *access |= i;
372 static void
373 set_deny(unsigned int *deny, unsigned long bmap) {
374 int i;
376 *deny = 0;
377 for (i = 0; i < 4; i++) {
378 if (test_bit(i, &bmap))
379 *deny |= i ;
383 static int
384 test_share(struct nfs4_stateid *stp, struct nfsd4_open *open) {
385 unsigned int access, deny;
387 set_access(&access, stp->st_access_bmap);
388 set_deny(&deny, stp->st_deny_bmap);
389 if ((access & open->op_share_deny) || (deny & open->op_share_access))
390 return 0;
391 return 1;
394 static int nfs4_access_to_omode(u32 access)
396 switch (access & NFS4_SHARE_ACCESS_BOTH) {
397 case NFS4_SHARE_ACCESS_READ:
398 return O_RDONLY;
399 case NFS4_SHARE_ACCESS_WRITE:
400 return O_WRONLY;
401 case NFS4_SHARE_ACCESS_BOTH:
402 return O_RDWR;
404 BUG();
407 static void unhash_generic_stateid(struct nfs4_stateid *stp)
409 list_del(&stp->st_hash);
410 list_del(&stp->st_perfile);
411 list_del(&stp->st_perstateowner);
414 static void free_generic_stateid(struct nfs4_stateid *stp)
416 int i;
418 if (stp->st_access_bmap) {
419 for (i = 1; i < 4; i++) {
420 if (test_bit(i, &stp->st_access_bmap))
421 nfs4_file_put_access(stp->st_file,
422 nfs4_access_to_omode(i));
425 put_nfs4_file(stp->st_file);
426 kmem_cache_free(stateid_slab, stp);
429 static void release_lock_stateid(struct nfs4_stateid *stp)
431 struct file *file;
433 unhash_generic_stateid(stp);
434 file = find_any_file(stp->st_file);
435 if (file)
436 locks_remove_posix(file, (fl_owner_t)stp->st_stateowner);
437 free_generic_stateid(stp);
440 static void unhash_lockowner(struct nfs4_stateowner *sop)
442 struct nfs4_stateid *stp;
444 list_del(&sop->so_idhash);
445 list_del(&sop->so_strhash);
446 list_del(&sop->so_perstateid);
447 while (!list_empty(&sop->so_stateids)) {
448 stp = list_first_entry(&sop->so_stateids,
449 struct nfs4_stateid, st_perstateowner);
450 release_lock_stateid(stp);
454 static void release_lockowner(struct nfs4_stateowner *sop)
456 unhash_lockowner(sop);
457 nfs4_put_stateowner(sop);
460 static void
461 release_stateid_lockowners(struct nfs4_stateid *open_stp)
463 struct nfs4_stateowner *lock_sop;
465 while (!list_empty(&open_stp->st_lockowners)) {
466 lock_sop = list_entry(open_stp->st_lockowners.next,
467 struct nfs4_stateowner, so_perstateid);
468 /* list_del(&open_stp->st_lockowners); */
469 BUG_ON(lock_sop->so_is_open_owner);
470 release_lockowner(lock_sop);
474 static void release_open_stateid(struct nfs4_stateid *stp)
476 unhash_generic_stateid(stp);
477 release_stateid_lockowners(stp);
478 free_generic_stateid(stp);
481 static void unhash_openowner(struct nfs4_stateowner *sop)
483 struct nfs4_stateid *stp;
485 list_del(&sop->so_idhash);
486 list_del(&sop->so_strhash);
487 list_del(&sop->so_perclient);
488 list_del(&sop->so_perstateid); /* XXX: necessary? */
489 while (!list_empty(&sop->so_stateids)) {
490 stp = list_first_entry(&sop->so_stateids,
491 struct nfs4_stateid, st_perstateowner);
492 release_open_stateid(stp);
496 static void release_openowner(struct nfs4_stateowner *sop)
498 unhash_openowner(sop);
499 list_del(&sop->so_close_lru);
500 nfs4_put_stateowner(sop);
503 #define SESSION_HASH_SIZE 512
504 static struct list_head sessionid_hashtbl[SESSION_HASH_SIZE];
506 static inline int
507 hash_sessionid(struct nfs4_sessionid *sessionid)
509 struct nfsd4_sessionid *sid = (struct nfsd4_sessionid *)sessionid;
511 return sid->sequence % SESSION_HASH_SIZE;
514 static inline void
515 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
517 u32 *ptr = (u32 *)(&sessionid->data[0]);
518 dprintk("%s: %u:%u:%u:%u\n", fn, ptr[0], ptr[1], ptr[2], ptr[3]);
521 static void
522 gen_sessionid(struct nfsd4_session *ses)
524 struct nfs4_client *clp = ses->se_client;
525 struct nfsd4_sessionid *sid;
527 sid = (struct nfsd4_sessionid *)ses->se_sessionid.data;
528 sid->clientid = clp->cl_clientid;
529 sid->sequence = current_sessionid++;
530 sid->reserved = 0;
534 * The protocol defines ca_maxresponssize_cached to include the size of
535 * the rpc header, but all we need to cache is the data starting after
536 * the end of the initial SEQUENCE operation--the rest we regenerate
537 * each time. Therefore we can advertise a ca_maxresponssize_cached
538 * value that is the number of bytes in our cache plus a few additional
539 * bytes. In order to stay on the safe side, and not promise more than
540 * we can cache, those additional bytes must be the minimum possible: 24
541 * bytes of rpc header (xid through accept state, with AUTH_NULL
542 * verifier), 12 for the compound header (with zero-length tag), and 44
543 * for the SEQUENCE op response:
545 #define NFSD_MIN_HDR_SEQ_SZ (24 + 12 + 44)
547 static void
548 free_session_slots(struct nfsd4_session *ses)
550 int i;
552 for (i = 0; i < ses->se_fchannel.maxreqs; i++)
553 kfree(ses->se_slots[i]);
557 * We don't actually need to cache the rpc and session headers, so we
558 * can allocate a little less for each slot:
560 static inline int slot_bytes(struct nfsd4_channel_attrs *ca)
562 return ca->maxresp_cached - NFSD_MIN_HDR_SEQ_SZ;
565 static int nfsd4_sanitize_slot_size(u32 size)
567 size -= NFSD_MIN_HDR_SEQ_SZ; /* We don't cache the rpc header */
568 size = min_t(u32, size, NFSD_SLOT_CACHE_SIZE);
570 return size;
574 * XXX: If we run out of reserved DRC memory we could (up to a point)
575 * re-negotiate active sessions and reduce their slot usage to make
576 * rooom for new connections. For now we just fail the create session.
578 static int nfsd4_get_drc_mem(int slotsize, u32 num)
580 int avail;
582 num = min_t(u32, num, NFSD_MAX_SLOTS_PER_SESSION);
584 spin_lock(&nfsd_drc_lock);
585 avail = min_t(int, NFSD_MAX_MEM_PER_SESSION,
586 nfsd_drc_max_mem - nfsd_drc_mem_used);
587 num = min_t(int, num, avail / slotsize);
588 nfsd_drc_mem_used += num * slotsize;
589 spin_unlock(&nfsd_drc_lock);
591 return num;
594 static void nfsd4_put_drc_mem(int slotsize, int num)
596 spin_lock(&nfsd_drc_lock);
597 nfsd_drc_mem_used -= slotsize * num;
598 spin_unlock(&nfsd_drc_lock);
601 static struct nfsd4_session *alloc_session(int slotsize, int numslots)
603 struct nfsd4_session *new;
604 int mem, i;
606 BUILD_BUG_ON(NFSD_MAX_SLOTS_PER_SESSION * sizeof(struct nfsd4_slot *)
607 + sizeof(struct nfsd4_session) > PAGE_SIZE);
608 mem = numslots * sizeof(struct nfsd4_slot *);
610 new = kzalloc(sizeof(*new) + mem, GFP_KERNEL);
611 if (!new)
612 return NULL;
613 /* allocate each struct nfsd4_slot and data cache in one piece */
614 for (i = 0; i < numslots; i++) {
615 mem = sizeof(struct nfsd4_slot) + slotsize;
616 new->se_slots[i] = kzalloc(mem, GFP_KERNEL);
617 if (!new->se_slots[i])
618 goto out_free;
620 return new;
621 out_free:
622 while (i--)
623 kfree(new->se_slots[i]);
624 kfree(new);
625 return NULL;
628 static void init_forechannel_attrs(struct nfsd4_channel_attrs *new, struct nfsd4_channel_attrs *req, int numslots, int slotsize)
630 u32 maxrpc = nfsd_serv->sv_max_mesg;
632 new->maxreqs = numslots;
633 new->maxresp_cached = min_t(u32, req->maxresp_cached,
634 slotsize + NFSD_MIN_HDR_SEQ_SZ);
635 new->maxreq_sz = min_t(u32, req->maxreq_sz, maxrpc);
636 new->maxresp_sz = min_t(u32, req->maxresp_sz, maxrpc);
637 new->maxops = min_t(u32, req->maxops, NFSD_MAX_OPS_PER_COMPOUND);
640 static void free_conn(struct nfsd4_conn *c)
642 svc_xprt_put(c->cn_xprt);
643 kfree(c);
646 static void nfsd4_conn_lost(struct svc_xpt_user *u)
648 struct nfsd4_conn *c = container_of(u, struct nfsd4_conn, cn_xpt_user);
649 struct nfs4_client *clp = c->cn_session->se_client;
651 spin_lock(&clp->cl_lock);
652 if (!list_empty(&c->cn_persession)) {
653 list_del(&c->cn_persession);
654 free_conn(c);
656 spin_unlock(&clp->cl_lock);
657 nfsd4_probe_callback(clp);
660 static struct nfsd4_conn *alloc_conn(struct svc_rqst *rqstp, u32 flags)
662 struct nfsd4_conn *conn;
664 conn = kmalloc(sizeof(struct nfsd4_conn), GFP_KERNEL);
665 if (!conn)
666 return NULL;
667 svc_xprt_get(rqstp->rq_xprt);
668 conn->cn_xprt = rqstp->rq_xprt;
669 conn->cn_flags = flags;
670 INIT_LIST_HEAD(&conn->cn_xpt_user.list);
671 return conn;
674 static void __nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
676 conn->cn_session = ses;
677 list_add(&conn->cn_persession, &ses->se_conns);
680 static void nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
682 struct nfs4_client *clp = ses->se_client;
684 spin_lock(&clp->cl_lock);
685 __nfsd4_hash_conn(conn, ses);
686 spin_unlock(&clp->cl_lock);
689 static int nfsd4_register_conn(struct nfsd4_conn *conn)
691 conn->cn_xpt_user.callback = nfsd4_conn_lost;
692 return register_xpt_user(conn->cn_xprt, &conn->cn_xpt_user);
695 static __be32 nfsd4_new_conn(struct svc_rqst *rqstp, struct nfsd4_session *ses, u32 dir)
697 struct nfsd4_conn *conn;
698 int ret;
700 conn = alloc_conn(rqstp, dir);
701 if (!conn)
702 return nfserr_jukebox;
703 nfsd4_hash_conn(conn, ses);
704 ret = nfsd4_register_conn(conn);
705 if (ret)
706 /* oops; xprt is already down: */
707 nfsd4_conn_lost(&conn->cn_xpt_user);
708 return nfs_ok;
711 static __be32 nfsd4_new_conn_from_crses(struct svc_rqst *rqstp, struct nfsd4_session *ses)
713 u32 dir = NFS4_CDFC4_FORE;
715 if (ses->se_flags & SESSION4_BACK_CHAN)
716 dir |= NFS4_CDFC4_BACK;
718 return nfsd4_new_conn(rqstp, ses, dir);
721 /* must be called under client_lock */
722 static void nfsd4_del_conns(struct nfsd4_session *s)
724 struct nfs4_client *clp = s->se_client;
725 struct nfsd4_conn *c;
727 spin_lock(&clp->cl_lock);
728 while (!list_empty(&s->se_conns)) {
729 c = list_first_entry(&s->se_conns, struct nfsd4_conn, cn_persession);
730 list_del_init(&c->cn_persession);
731 spin_unlock(&clp->cl_lock);
733 unregister_xpt_user(c->cn_xprt, &c->cn_xpt_user);
734 free_conn(c);
736 spin_lock(&clp->cl_lock);
738 spin_unlock(&clp->cl_lock);
741 void free_session(struct kref *kref)
743 struct nfsd4_session *ses;
744 int mem;
746 ses = container_of(kref, struct nfsd4_session, se_ref);
747 nfsd4_del_conns(ses);
748 spin_lock(&nfsd_drc_lock);
749 mem = ses->se_fchannel.maxreqs * slot_bytes(&ses->se_fchannel);
750 nfsd_drc_mem_used -= mem;
751 spin_unlock(&nfsd_drc_lock);
752 free_session_slots(ses);
753 kfree(ses);
756 static struct nfsd4_session *alloc_init_session(struct svc_rqst *rqstp, struct nfs4_client *clp, struct nfsd4_create_session *cses)
758 struct nfsd4_session *new;
759 struct nfsd4_channel_attrs *fchan = &cses->fore_channel;
760 int numslots, slotsize;
761 int status;
762 int idx;
765 * Note decreasing slot size below client's request may
766 * make it difficult for client to function correctly, whereas
767 * decreasing the number of slots will (just?) affect
768 * performance. When short on memory we therefore prefer to
769 * decrease number of slots instead of their size.
771 slotsize = nfsd4_sanitize_slot_size(fchan->maxresp_cached);
772 numslots = nfsd4_get_drc_mem(slotsize, fchan->maxreqs);
773 if (numslots < 1)
774 return NULL;
776 new = alloc_session(slotsize, numslots);
777 if (!new) {
778 nfsd4_put_drc_mem(slotsize, fchan->maxreqs);
779 return NULL;
781 init_forechannel_attrs(&new->se_fchannel, fchan, numslots, slotsize);
783 new->se_client = clp;
784 gen_sessionid(new);
786 INIT_LIST_HEAD(&new->se_conns);
788 new->se_cb_seq_nr = 1;
789 new->se_flags = cses->flags;
790 new->se_cb_prog = cses->callback_prog;
791 kref_init(&new->se_ref);
792 idx = hash_sessionid(&new->se_sessionid);
793 spin_lock(&client_lock);
794 list_add(&new->se_hash, &sessionid_hashtbl[idx]);
795 spin_lock(&clp->cl_lock);
796 list_add(&new->se_perclnt, &clp->cl_sessions);
797 spin_unlock(&clp->cl_lock);
798 spin_unlock(&client_lock);
800 status = nfsd4_new_conn_from_crses(rqstp, new);
801 /* whoops: benny points out, status is ignored! (err, or bogus) */
802 if (status) {
803 free_session(&new->se_ref);
804 return NULL;
806 if (cses->flags & SESSION4_BACK_CHAN) {
807 struct sockaddr *sa = svc_addr(rqstp);
809 * This is a little silly; with sessions there's no real
810 * use for the callback address. Use the peer address
811 * as a reasonable default for now, but consider fixing
812 * the rpc client not to require an address in the
813 * future:
815 rpc_copy_addr((struct sockaddr *)&clp->cl_cb_conn.cb_addr, sa);
816 clp->cl_cb_conn.cb_addrlen = svc_addr_len(sa);
818 nfsd4_probe_callback(clp);
819 return new;
822 /* caller must hold client_lock */
823 static struct nfsd4_session *
824 find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid)
826 struct nfsd4_session *elem;
827 int idx;
829 dump_sessionid(__func__, sessionid);
830 idx = hash_sessionid(sessionid);
831 /* Search in the appropriate list */
832 list_for_each_entry(elem, &sessionid_hashtbl[idx], se_hash) {
833 if (!memcmp(elem->se_sessionid.data, sessionid->data,
834 NFS4_MAX_SESSIONID_LEN)) {
835 return elem;
839 dprintk("%s: session not found\n", __func__);
840 return NULL;
843 /* caller must hold client_lock */
844 static void
845 unhash_session(struct nfsd4_session *ses)
847 list_del(&ses->se_hash);
848 spin_lock(&ses->se_client->cl_lock);
849 list_del(&ses->se_perclnt);
850 spin_unlock(&ses->se_client->cl_lock);
853 /* must be called under the client_lock */
854 static inline void
855 renew_client_locked(struct nfs4_client *clp)
857 if (is_client_expired(clp)) {
858 dprintk("%s: client (clientid %08x/%08x) already expired\n",
859 __func__,
860 clp->cl_clientid.cl_boot,
861 clp->cl_clientid.cl_id);
862 return;
866 * Move client to the end to the LRU list.
868 dprintk("renewing client (clientid %08x/%08x)\n",
869 clp->cl_clientid.cl_boot,
870 clp->cl_clientid.cl_id);
871 list_move_tail(&clp->cl_lru, &client_lru);
872 clp->cl_time = get_seconds();
875 static inline void
876 renew_client(struct nfs4_client *clp)
878 spin_lock(&client_lock);
879 renew_client_locked(clp);
880 spin_unlock(&client_lock);
883 /* SETCLIENTID and SETCLIENTID_CONFIRM Helper functions */
884 static int
885 STALE_CLIENTID(clientid_t *clid)
887 if (clid->cl_boot == boot_time)
888 return 0;
889 dprintk("NFSD stale clientid (%08x/%08x) boot_time %08lx\n",
890 clid->cl_boot, clid->cl_id, boot_time);
891 return 1;
895 * XXX Should we use a slab cache ?
896 * This type of memory management is somewhat inefficient, but we use it
897 * anyway since SETCLIENTID is not a common operation.
899 static struct nfs4_client *alloc_client(struct xdr_netobj name)
901 struct nfs4_client *clp;
903 clp = kzalloc(sizeof(struct nfs4_client), GFP_KERNEL);
904 if (clp == NULL)
905 return NULL;
906 clp->cl_name.data = kmalloc(name.len, GFP_KERNEL);
907 if (clp->cl_name.data == NULL) {
908 kfree(clp);
909 return NULL;
911 memcpy(clp->cl_name.data, name.data, name.len);
912 clp->cl_name.len = name.len;
913 return clp;
916 static inline void
917 free_client(struct nfs4_client *clp)
919 while (!list_empty(&clp->cl_sessions)) {
920 struct nfsd4_session *ses;
921 ses = list_entry(clp->cl_sessions.next, struct nfsd4_session,
922 se_perclnt);
923 list_del(&ses->se_perclnt);
924 nfsd4_put_session(ses);
926 if (clp->cl_cred.cr_group_info)
927 put_group_info(clp->cl_cred.cr_group_info);
928 kfree(clp->cl_principal);
929 kfree(clp->cl_name.data);
930 kfree(clp);
933 void
934 release_session_client(struct nfsd4_session *session)
936 struct nfs4_client *clp = session->se_client;
938 if (!atomic_dec_and_lock(&clp->cl_refcount, &client_lock))
939 return;
940 if (is_client_expired(clp)) {
941 free_client(clp);
942 session->se_client = NULL;
943 } else
944 renew_client_locked(clp);
945 spin_unlock(&client_lock);
948 /* must be called under the client_lock */
949 static inline void
950 unhash_client_locked(struct nfs4_client *clp)
952 struct nfsd4_session *ses;
954 mark_client_expired(clp);
955 list_del(&clp->cl_lru);
956 spin_lock(&clp->cl_lock);
957 list_for_each_entry(ses, &clp->cl_sessions, se_perclnt)
958 list_del_init(&ses->se_hash);
959 spin_unlock(&clp->cl_lock);
962 static void
963 expire_client(struct nfs4_client *clp)
965 struct nfs4_stateowner *sop;
966 struct nfs4_delegation *dp;
967 struct list_head reaplist;
969 INIT_LIST_HEAD(&reaplist);
970 spin_lock(&recall_lock);
971 while (!list_empty(&clp->cl_delegations)) {
972 dp = list_entry(clp->cl_delegations.next, struct nfs4_delegation, dl_perclnt);
973 list_del_init(&dp->dl_perclnt);
974 list_move(&dp->dl_recall_lru, &reaplist);
976 spin_unlock(&recall_lock);
977 while (!list_empty(&reaplist)) {
978 dp = list_entry(reaplist.next, struct nfs4_delegation, dl_recall_lru);
979 list_del_init(&dp->dl_recall_lru);
980 unhash_delegation(dp);
982 while (!list_empty(&clp->cl_openowners)) {
983 sop = list_entry(clp->cl_openowners.next, struct nfs4_stateowner, so_perclient);
984 release_openowner(sop);
986 nfsd4_shutdown_callback(clp);
987 if (clp->cl_cb_conn.cb_xprt)
988 svc_xprt_put(clp->cl_cb_conn.cb_xprt);
989 list_del(&clp->cl_idhash);
990 list_del(&clp->cl_strhash);
991 spin_lock(&client_lock);
992 unhash_client_locked(clp);
993 if (atomic_read(&clp->cl_refcount) == 0)
994 free_client(clp);
995 spin_unlock(&client_lock);
998 static void copy_verf(struct nfs4_client *target, nfs4_verifier *source)
1000 memcpy(target->cl_verifier.data, source->data,
1001 sizeof(target->cl_verifier.data));
1004 static void copy_clid(struct nfs4_client *target, struct nfs4_client *source)
1006 target->cl_clientid.cl_boot = source->cl_clientid.cl_boot;
1007 target->cl_clientid.cl_id = source->cl_clientid.cl_id;
1010 static void copy_cred(struct svc_cred *target, struct svc_cred *source)
1012 target->cr_uid = source->cr_uid;
1013 target->cr_gid = source->cr_gid;
1014 target->cr_group_info = source->cr_group_info;
1015 get_group_info(target->cr_group_info);
1018 static int same_name(const char *n1, const char *n2)
1020 return 0 == memcmp(n1, n2, HEXDIR_LEN);
1023 static int
1024 same_verf(nfs4_verifier *v1, nfs4_verifier *v2)
1026 return 0 == memcmp(v1->data, v2->data, sizeof(v1->data));
1029 static int
1030 same_clid(clientid_t *cl1, clientid_t *cl2)
1032 return (cl1->cl_boot == cl2->cl_boot) && (cl1->cl_id == cl2->cl_id);
1035 /* XXX what about NGROUP */
1036 static int
1037 same_creds(struct svc_cred *cr1, struct svc_cred *cr2)
1039 return cr1->cr_uid == cr2->cr_uid;
1042 static void gen_clid(struct nfs4_client *clp)
1044 static u32 current_clientid = 1;
1046 clp->cl_clientid.cl_boot = boot_time;
1047 clp->cl_clientid.cl_id = current_clientid++;
1050 static void gen_confirm(struct nfs4_client *clp)
1052 static u32 i;
1053 u32 *p;
1055 p = (u32 *)clp->cl_confirm.data;
1056 *p++ = get_seconds();
1057 *p++ = i++;
1060 static struct nfs4_client *create_client(struct xdr_netobj name, char *recdir,
1061 struct svc_rqst *rqstp, nfs4_verifier *verf)
1063 struct nfs4_client *clp;
1064 struct sockaddr *sa = svc_addr(rqstp);
1065 char *princ;
1067 clp = alloc_client(name);
1068 if (clp == NULL)
1069 return NULL;
1071 INIT_LIST_HEAD(&clp->cl_sessions);
1073 princ = svc_gss_principal(rqstp);
1074 if (princ) {
1075 clp->cl_principal = kstrdup(princ, GFP_KERNEL);
1076 if (clp->cl_principal == NULL) {
1077 free_client(clp);
1078 return NULL;
1082 memcpy(clp->cl_recdir, recdir, HEXDIR_LEN);
1083 atomic_set(&clp->cl_refcount, 0);
1084 clp->cl_cb_state = NFSD4_CB_UNKNOWN;
1085 INIT_LIST_HEAD(&clp->cl_idhash);
1086 INIT_LIST_HEAD(&clp->cl_strhash);
1087 INIT_LIST_HEAD(&clp->cl_openowners);
1088 INIT_LIST_HEAD(&clp->cl_delegations);
1089 INIT_LIST_HEAD(&clp->cl_lru);
1090 INIT_LIST_HEAD(&clp->cl_callbacks);
1091 spin_lock_init(&clp->cl_lock);
1092 INIT_WORK(&clp->cl_cb_null.cb_work, nfsd4_do_callback_rpc);
1093 clp->cl_time = get_seconds();
1094 clear_bit(0, &clp->cl_cb_slot_busy);
1095 rpc_init_wait_queue(&clp->cl_cb_waitq, "Backchannel slot table");
1096 copy_verf(clp, verf);
1097 rpc_copy_addr((struct sockaddr *) &clp->cl_addr, sa);
1098 clp->cl_flavor = rqstp->rq_flavor;
1099 copy_cred(&clp->cl_cred, &rqstp->rq_cred);
1100 gen_confirm(clp);
1101 clp->cl_cb_session = NULL;
1102 return clp;
1105 static int check_name(struct xdr_netobj name)
1107 if (name.len == 0)
1108 return 0;
1109 if (name.len > NFS4_OPAQUE_LIMIT) {
1110 dprintk("NFSD: check_name: name too long(%d)!\n", name.len);
1111 return 0;
1113 return 1;
1116 static void
1117 add_to_unconfirmed(struct nfs4_client *clp, unsigned int strhashval)
1119 unsigned int idhashval;
1121 list_add(&clp->cl_strhash, &unconf_str_hashtbl[strhashval]);
1122 idhashval = clientid_hashval(clp->cl_clientid.cl_id);
1123 list_add(&clp->cl_idhash, &unconf_id_hashtbl[idhashval]);
1124 renew_client(clp);
1127 static void
1128 move_to_confirmed(struct nfs4_client *clp)
1130 unsigned int idhashval = clientid_hashval(clp->cl_clientid.cl_id);
1131 unsigned int strhashval;
1133 dprintk("NFSD: move_to_confirm nfs4_client %p\n", clp);
1134 list_move(&clp->cl_idhash, &conf_id_hashtbl[idhashval]);
1135 strhashval = clientstr_hashval(clp->cl_recdir);
1136 list_move(&clp->cl_strhash, &conf_str_hashtbl[strhashval]);
1137 renew_client(clp);
1140 static struct nfs4_client *
1141 find_confirmed_client(clientid_t *clid)
1143 struct nfs4_client *clp;
1144 unsigned int idhashval = clientid_hashval(clid->cl_id);
1146 list_for_each_entry(clp, &conf_id_hashtbl[idhashval], cl_idhash) {
1147 if (same_clid(&clp->cl_clientid, clid))
1148 return clp;
1150 return NULL;
1153 static struct nfs4_client *
1154 find_unconfirmed_client(clientid_t *clid)
1156 struct nfs4_client *clp;
1157 unsigned int idhashval = clientid_hashval(clid->cl_id);
1159 list_for_each_entry(clp, &unconf_id_hashtbl[idhashval], cl_idhash) {
1160 if (same_clid(&clp->cl_clientid, clid))
1161 return clp;
1163 return NULL;
1166 static bool clp_used_exchangeid(struct nfs4_client *clp)
1168 return clp->cl_exchange_flags != 0;
1171 static struct nfs4_client *
1172 find_confirmed_client_by_str(const char *dname, unsigned int hashval)
1174 struct nfs4_client *clp;
1176 list_for_each_entry(clp, &conf_str_hashtbl[hashval], cl_strhash) {
1177 if (same_name(clp->cl_recdir, dname))
1178 return clp;
1180 return NULL;
1183 static struct nfs4_client *
1184 find_unconfirmed_client_by_str(const char *dname, unsigned int hashval)
1186 struct nfs4_client *clp;
1188 list_for_each_entry(clp, &unconf_str_hashtbl[hashval], cl_strhash) {
1189 if (same_name(clp->cl_recdir, dname))
1190 return clp;
1192 return NULL;
1195 static void rpc_svcaddr2sockaddr(struct sockaddr *sa, unsigned short family, union svc_addr_u *svcaddr)
1197 switch (family) {
1198 case AF_INET:
1199 ((struct sockaddr_in *)sa)->sin_family = AF_INET;
1200 ((struct sockaddr_in *)sa)->sin_addr = svcaddr->addr;
1201 return;
1202 case AF_INET6:
1203 ((struct sockaddr_in6 *)sa)->sin6_family = AF_INET6;
1204 ((struct sockaddr_in6 *)sa)->sin6_addr = svcaddr->addr6;
1205 return;
1209 static void
1210 gen_callback(struct nfs4_client *clp, struct nfsd4_setclientid *se, struct svc_rqst *rqstp)
1212 struct nfs4_cb_conn *conn = &clp->cl_cb_conn;
1213 struct sockaddr *sa = svc_addr(rqstp);
1214 u32 scopeid = rpc_get_scope_id(sa);
1215 unsigned short expected_family;
1217 /* Currently, we only support tcp and tcp6 for the callback channel */
1218 if (se->se_callback_netid_len == 3 &&
1219 !memcmp(se->se_callback_netid_val, "tcp", 3))
1220 expected_family = AF_INET;
1221 else if (se->se_callback_netid_len == 4 &&
1222 !memcmp(se->se_callback_netid_val, "tcp6", 4))
1223 expected_family = AF_INET6;
1224 else
1225 goto out_err;
1227 conn->cb_addrlen = rpc_uaddr2sockaddr(se->se_callback_addr_val,
1228 se->se_callback_addr_len,
1229 (struct sockaddr *)&conn->cb_addr,
1230 sizeof(conn->cb_addr));
1232 if (!conn->cb_addrlen || conn->cb_addr.ss_family != expected_family)
1233 goto out_err;
1235 if (conn->cb_addr.ss_family == AF_INET6)
1236 ((struct sockaddr_in6 *)&conn->cb_addr)->sin6_scope_id = scopeid;
1238 conn->cb_prog = se->se_callback_prog;
1239 conn->cb_ident = se->se_callback_ident;
1240 rpc_svcaddr2sockaddr((struct sockaddr *)&conn->cb_saddr, expected_family, &rqstp->rq_daddr);
1241 return;
1242 out_err:
1243 conn->cb_addr.ss_family = AF_UNSPEC;
1244 conn->cb_addrlen = 0;
1245 dprintk(KERN_INFO "NFSD: this client (clientid %08x/%08x) "
1246 "will not receive delegations\n",
1247 clp->cl_clientid.cl_boot, clp->cl_clientid.cl_id);
1249 return;
1253 * Cache a reply. nfsd4_check_drc_limit() has bounded the cache size.
1255 void
1256 nfsd4_store_cache_entry(struct nfsd4_compoundres *resp)
1258 struct nfsd4_slot *slot = resp->cstate.slot;
1259 unsigned int base;
1261 dprintk("--> %s slot %p\n", __func__, slot);
1263 slot->sl_opcnt = resp->opcnt;
1264 slot->sl_status = resp->cstate.status;
1266 if (nfsd4_not_cached(resp)) {
1267 slot->sl_datalen = 0;
1268 return;
1270 slot->sl_datalen = (char *)resp->p - (char *)resp->cstate.datap;
1271 base = (char *)resp->cstate.datap -
1272 (char *)resp->xbuf->head[0].iov_base;
1273 if (read_bytes_from_xdr_buf(resp->xbuf, base, slot->sl_data,
1274 slot->sl_datalen))
1275 WARN("%s: sessions DRC could not cache compound\n", __func__);
1276 return;
1280 * Encode the replay sequence operation from the slot values.
1281 * If cachethis is FALSE encode the uncached rep error on the next
1282 * operation which sets resp->p and increments resp->opcnt for
1283 * nfs4svc_encode_compoundres.
1286 static __be32
1287 nfsd4_enc_sequence_replay(struct nfsd4_compoundargs *args,
1288 struct nfsd4_compoundres *resp)
1290 struct nfsd4_op *op;
1291 struct nfsd4_slot *slot = resp->cstate.slot;
1293 dprintk("--> %s resp->opcnt %d cachethis %u \n", __func__,
1294 resp->opcnt, resp->cstate.slot->sl_cachethis);
1296 /* Encode the replayed sequence operation */
1297 op = &args->ops[resp->opcnt - 1];
1298 nfsd4_encode_operation(resp, op);
1300 /* Return nfserr_retry_uncached_rep in next operation. */
1301 if (args->opcnt > 1 && slot->sl_cachethis == 0) {
1302 op = &args->ops[resp->opcnt++];
1303 op->status = nfserr_retry_uncached_rep;
1304 nfsd4_encode_operation(resp, op);
1306 return op->status;
1310 * The sequence operation is not cached because we can use the slot and
1311 * session values.
1313 __be32
1314 nfsd4_replay_cache_entry(struct nfsd4_compoundres *resp,
1315 struct nfsd4_sequence *seq)
1317 struct nfsd4_slot *slot = resp->cstate.slot;
1318 __be32 status;
1320 dprintk("--> %s slot %p\n", __func__, slot);
1322 /* Either returns 0 or nfserr_retry_uncached */
1323 status = nfsd4_enc_sequence_replay(resp->rqstp->rq_argp, resp);
1324 if (status == nfserr_retry_uncached_rep)
1325 return status;
1327 /* The sequence operation has been encoded, cstate->datap set. */
1328 memcpy(resp->cstate.datap, slot->sl_data, slot->sl_datalen);
1330 resp->opcnt = slot->sl_opcnt;
1331 resp->p = resp->cstate.datap + XDR_QUADLEN(slot->sl_datalen);
1332 status = slot->sl_status;
1334 return status;
1338 * Set the exchange_id flags returned by the server.
1340 static void
1341 nfsd4_set_ex_flags(struct nfs4_client *new, struct nfsd4_exchange_id *clid)
1343 /* pNFS is not supported */
1344 new->cl_exchange_flags |= EXCHGID4_FLAG_USE_NON_PNFS;
1346 /* Referrals are supported, Migration is not. */
1347 new->cl_exchange_flags |= EXCHGID4_FLAG_SUPP_MOVED_REFER;
1349 /* set the wire flags to return to client. */
1350 clid->flags = new->cl_exchange_flags;
1353 __be32
1354 nfsd4_exchange_id(struct svc_rqst *rqstp,
1355 struct nfsd4_compound_state *cstate,
1356 struct nfsd4_exchange_id *exid)
1358 struct nfs4_client *unconf, *conf, *new;
1359 int status;
1360 unsigned int strhashval;
1361 char dname[HEXDIR_LEN];
1362 char addr_str[INET6_ADDRSTRLEN];
1363 nfs4_verifier verf = exid->verifier;
1364 struct sockaddr *sa = svc_addr(rqstp);
1366 rpc_ntop(sa, addr_str, sizeof(addr_str));
1367 dprintk("%s rqstp=%p exid=%p clname.len=%u clname.data=%p "
1368 "ip_addr=%s flags %x, spa_how %d\n",
1369 __func__, rqstp, exid, exid->clname.len, exid->clname.data,
1370 addr_str, exid->flags, exid->spa_how);
1372 if (!check_name(exid->clname) || (exid->flags & ~EXCHGID4_FLAG_MASK_A))
1373 return nfserr_inval;
1375 /* Currently only support SP4_NONE */
1376 switch (exid->spa_how) {
1377 case SP4_NONE:
1378 break;
1379 case SP4_SSV:
1380 return nfserr_serverfault;
1381 default:
1382 BUG(); /* checked by xdr code */
1383 case SP4_MACH_CRED:
1384 return nfserr_serverfault; /* no excuse :-/ */
1387 status = nfs4_make_rec_clidname(dname, &exid->clname);
1389 if (status)
1390 goto error;
1392 strhashval = clientstr_hashval(dname);
1394 nfs4_lock_state();
1395 status = nfs_ok;
1397 conf = find_confirmed_client_by_str(dname, strhashval);
1398 if (conf) {
1399 if (!clp_used_exchangeid(conf)) {
1400 status = nfserr_clid_inuse; /* XXX: ? */
1401 goto out;
1403 if (!same_verf(&verf, &conf->cl_verifier)) {
1404 /* 18.35.4 case 8 */
1405 if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
1406 status = nfserr_not_same;
1407 goto out;
1409 /* Client reboot: destroy old state */
1410 expire_client(conf);
1411 goto out_new;
1413 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
1414 /* 18.35.4 case 9 */
1415 if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
1416 status = nfserr_perm;
1417 goto out;
1419 expire_client(conf);
1420 goto out_new;
1423 * Set bit when the owner id and verifier map to an already
1424 * confirmed client id (18.35.3).
1426 exid->flags |= EXCHGID4_FLAG_CONFIRMED_R;
1429 * Falling into 18.35.4 case 2, possible router replay.
1430 * Leave confirmed record intact and return same result.
1432 copy_verf(conf, &verf);
1433 new = conf;
1434 goto out_copy;
1437 /* 18.35.4 case 7 */
1438 if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
1439 status = nfserr_noent;
1440 goto out;
1443 unconf = find_unconfirmed_client_by_str(dname, strhashval);
1444 if (unconf) {
1446 * Possible retry or client restart. Per 18.35.4 case 4,
1447 * a new unconfirmed record should be generated regardless
1448 * of whether any properties have changed.
1450 expire_client(unconf);
1453 out_new:
1454 /* Normal case */
1455 new = create_client(exid->clname, dname, rqstp, &verf);
1456 if (new == NULL) {
1457 status = nfserr_jukebox;
1458 goto out;
1461 gen_clid(new);
1462 add_to_unconfirmed(new, strhashval);
1463 out_copy:
1464 exid->clientid.cl_boot = new->cl_clientid.cl_boot;
1465 exid->clientid.cl_id = new->cl_clientid.cl_id;
1467 exid->seqid = 1;
1468 nfsd4_set_ex_flags(new, exid);
1470 dprintk("nfsd4_exchange_id seqid %d flags %x\n",
1471 new->cl_cs_slot.sl_seqid, new->cl_exchange_flags);
1472 status = nfs_ok;
1474 out:
1475 nfs4_unlock_state();
1476 error:
1477 dprintk("nfsd4_exchange_id returns %d\n", ntohl(status));
1478 return status;
1481 static int
1482 check_slot_seqid(u32 seqid, u32 slot_seqid, int slot_inuse)
1484 dprintk("%s enter. seqid %d slot_seqid %d\n", __func__, seqid,
1485 slot_seqid);
1487 /* The slot is in use, and no response has been sent. */
1488 if (slot_inuse) {
1489 if (seqid == slot_seqid)
1490 return nfserr_jukebox;
1491 else
1492 return nfserr_seq_misordered;
1494 /* Normal */
1495 if (likely(seqid == slot_seqid + 1))
1496 return nfs_ok;
1497 /* Replay */
1498 if (seqid == slot_seqid)
1499 return nfserr_replay_cache;
1500 /* Wraparound */
1501 if (seqid == 1 && (slot_seqid + 1) == 0)
1502 return nfs_ok;
1503 /* Misordered replay or misordered new request */
1504 return nfserr_seq_misordered;
1508 * Cache the create session result into the create session single DRC
1509 * slot cache by saving the xdr structure. sl_seqid has been set.
1510 * Do this for solo or embedded create session operations.
1512 static void
1513 nfsd4_cache_create_session(struct nfsd4_create_session *cr_ses,
1514 struct nfsd4_clid_slot *slot, int nfserr)
1516 slot->sl_status = nfserr;
1517 memcpy(&slot->sl_cr_ses, cr_ses, sizeof(*cr_ses));
1520 static __be32
1521 nfsd4_replay_create_session(struct nfsd4_create_session *cr_ses,
1522 struct nfsd4_clid_slot *slot)
1524 memcpy(cr_ses, &slot->sl_cr_ses, sizeof(*cr_ses));
1525 return slot->sl_status;
1528 #define NFSD_MIN_REQ_HDR_SEQ_SZ ((\
1529 2 * 2 + /* credential,verifier: AUTH_NULL, length 0 */ \
1530 1 + /* MIN tag is length with zero, only length */ \
1531 3 + /* version, opcount, opcode */ \
1532 XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
1533 /* seqid, slotID, slotID, cache */ \
1534 4 ) * sizeof(__be32))
1536 #define NFSD_MIN_RESP_HDR_SEQ_SZ ((\
1537 2 + /* verifier: AUTH_NULL, length 0 */\
1538 1 + /* status */ \
1539 1 + /* MIN tag is length with zero, only length */ \
1540 3 + /* opcount, opcode, opstatus*/ \
1541 XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
1542 /* seqid, slotID, slotID, slotID, status */ \
1543 5 ) * sizeof(__be32))
1545 static __be32 check_forechannel_attrs(struct nfsd4_channel_attrs fchannel)
1547 return fchannel.maxreq_sz < NFSD_MIN_REQ_HDR_SEQ_SZ
1548 || fchannel.maxresp_sz < NFSD_MIN_RESP_HDR_SEQ_SZ;
1551 __be32
1552 nfsd4_create_session(struct svc_rqst *rqstp,
1553 struct nfsd4_compound_state *cstate,
1554 struct nfsd4_create_session *cr_ses)
1556 struct sockaddr *sa = svc_addr(rqstp);
1557 struct nfs4_client *conf, *unconf;
1558 struct nfsd4_session *new;
1559 struct nfsd4_clid_slot *cs_slot = NULL;
1560 bool confirm_me = false;
1561 int status = 0;
1563 if (cr_ses->flags & ~SESSION4_FLAG_MASK_A)
1564 return nfserr_inval;
1566 nfs4_lock_state();
1567 unconf = find_unconfirmed_client(&cr_ses->clientid);
1568 conf = find_confirmed_client(&cr_ses->clientid);
1570 if (conf) {
1571 cs_slot = &conf->cl_cs_slot;
1572 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
1573 if (status == nfserr_replay_cache) {
1574 dprintk("Got a create_session replay! seqid= %d\n",
1575 cs_slot->sl_seqid);
1576 /* Return the cached reply status */
1577 status = nfsd4_replay_create_session(cr_ses, cs_slot);
1578 goto out;
1579 } else if (cr_ses->seqid != cs_slot->sl_seqid + 1) {
1580 status = nfserr_seq_misordered;
1581 dprintk("Sequence misordered!\n");
1582 dprintk("Expected seqid= %d but got seqid= %d\n",
1583 cs_slot->sl_seqid, cr_ses->seqid);
1584 goto out;
1586 } else if (unconf) {
1587 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred) ||
1588 !rpc_cmp_addr(sa, (struct sockaddr *) &unconf->cl_addr)) {
1589 status = nfserr_clid_inuse;
1590 goto out;
1593 cs_slot = &unconf->cl_cs_slot;
1594 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
1595 if (status) {
1596 /* an unconfirmed replay returns misordered */
1597 status = nfserr_seq_misordered;
1598 goto out;
1601 confirm_me = true;
1602 conf = unconf;
1603 } else {
1604 status = nfserr_stale_clientid;
1605 goto out;
1609 * XXX: we should probably set this at creation time, and check
1610 * for consistent minorversion use throughout:
1612 conf->cl_minorversion = 1;
1614 * We do not support RDMA or persistent sessions
1616 cr_ses->flags &= ~SESSION4_PERSIST;
1617 cr_ses->flags &= ~SESSION4_RDMA;
1619 status = nfserr_toosmall;
1620 if (check_forechannel_attrs(cr_ses->fore_channel))
1621 goto out;
1623 status = nfserr_jukebox;
1624 new = alloc_init_session(rqstp, conf, cr_ses);
1625 if (!new)
1626 goto out;
1627 status = nfs_ok;
1628 memcpy(cr_ses->sessionid.data, new->se_sessionid.data,
1629 NFS4_MAX_SESSIONID_LEN);
1630 memcpy(&cr_ses->fore_channel, &new->se_fchannel,
1631 sizeof(struct nfsd4_channel_attrs));
1632 cs_slot->sl_seqid++;
1633 cr_ses->seqid = cs_slot->sl_seqid;
1635 /* cache solo and embedded create sessions under the state lock */
1636 nfsd4_cache_create_session(cr_ses, cs_slot, status);
1637 if (confirm_me)
1638 move_to_confirmed(conf);
1639 out:
1640 nfs4_unlock_state();
1641 dprintk("%s returns %d\n", __func__, ntohl(status));
1642 return status;
1645 static bool nfsd4_last_compound_op(struct svc_rqst *rqstp)
1647 struct nfsd4_compoundres *resp = rqstp->rq_resp;
1648 struct nfsd4_compoundargs *argp = rqstp->rq_argp;
1650 return argp->opcnt == resp->opcnt;
1653 static __be32 nfsd4_map_bcts_dir(u32 *dir)
1655 switch (*dir) {
1656 case NFS4_CDFC4_FORE:
1657 case NFS4_CDFC4_BACK:
1658 return nfs_ok;
1659 case NFS4_CDFC4_FORE_OR_BOTH:
1660 case NFS4_CDFC4_BACK_OR_BOTH:
1661 *dir = NFS4_CDFC4_BOTH;
1662 return nfs_ok;
1664 return nfserr_inval;
1667 __be32 nfsd4_bind_conn_to_session(struct svc_rqst *rqstp,
1668 struct nfsd4_compound_state *cstate,
1669 struct nfsd4_bind_conn_to_session *bcts)
1671 __be32 status;
1673 if (!nfsd4_last_compound_op(rqstp))
1674 return nfserr_not_only_op;
1675 spin_lock(&client_lock);
1676 cstate->session = find_in_sessionid_hashtbl(&bcts->sessionid);
1677 /* Sorta weird: we only need the refcnt'ing because new_conn acquires
1678 * client_lock iself: */
1679 if (cstate->session) {
1680 nfsd4_get_session(cstate->session);
1681 atomic_inc(&cstate->session->se_client->cl_refcount);
1683 spin_unlock(&client_lock);
1684 if (!cstate->session)
1685 return nfserr_badsession;
1687 status = nfsd4_map_bcts_dir(&bcts->dir);
1688 if (!status)
1689 nfsd4_new_conn(rqstp, cstate->session, bcts->dir);
1690 return status;
1693 static bool nfsd4_compound_in_session(struct nfsd4_session *session, struct nfs4_sessionid *sid)
1695 if (!session)
1696 return 0;
1697 return !memcmp(sid, &session->se_sessionid, sizeof(*sid));
1700 __be32
1701 nfsd4_destroy_session(struct svc_rqst *r,
1702 struct nfsd4_compound_state *cstate,
1703 struct nfsd4_destroy_session *sessionid)
1705 struct nfsd4_session *ses;
1706 u32 status = nfserr_badsession;
1708 /* Notes:
1709 * - The confirmed nfs4_client->cl_sessionid holds destroyed sessinid
1710 * - Should we return nfserr_back_chan_busy if waiting for
1711 * callbacks on to-be-destroyed session?
1712 * - Do we need to clear any callback info from previous session?
1715 if (nfsd4_compound_in_session(cstate->session, &sessionid->sessionid)) {
1716 if (!nfsd4_last_compound_op(r))
1717 return nfserr_not_only_op;
1719 dump_sessionid(__func__, &sessionid->sessionid);
1720 spin_lock(&client_lock);
1721 ses = find_in_sessionid_hashtbl(&sessionid->sessionid);
1722 if (!ses) {
1723 spin_unlock(&client_lock);
1724 goto out;
1727 unhash_session(ses);
1728 spin_unlock(&client_lock);
1730 nfs4_lock_state();
1731 nfsd4_probe_callback_sync(ses->se_client);
1732 nfs4_unlock_state();
1734 nfsd4_del_conns(ses);
1736 nfsd4_put_session(ses);
1737 status = nfs_ok;
1738 out:
1739 dprintk("%s returns %d\n", __func__, ntohl(status));
1740 return status;
1743 static struct nfsd4_conn *__nfsd4_find_conn(struct svc_xprt *xpt, struct nfsd4_session *s)
1745 struct nfsd4_conn *c;
1747 list_for_each_entry(c, &s->se_conns, cn_persession) {
1748 if (c->cn_xprt == xpt) {
1749 return c;
1752 return NULL;
1755 static void nfsd4_sequence_check_conn(struct nfsd4_conn *new, struct nfsd4_session *ses)
1757 struct nfs4_client *clp = ses->se_client;
1758 struct nfsd4_conn *c;
1759 int ret;
1761 spin_lock(&clp->cl_lock);
1762 c = __nfsd4_find_conn(new->cn_xprt, ses);
1763 if (c) {
1764 spin_unlock(&clp->cl_lock);
1765 free_conn(new);
1766 return;
1768 __nfsd4_hash_conn(new, ses);
1769 spin_unlock(&clp->cl_lock);
1770 ret = nfsd4_register_conn(new);
1771 if (ret)
1772 /* oops; xprt is already down: */
1773 nfsd4_conn_lost(&new->cn_xpt_user);
1774 return;
1777 static bool nfsd4_session_too_many_ops(struct svc_rqst *rqstp, struct nfsd4_session *session)
1779 struct nfsd4_compoundargs *args = rqstp->rq_argp;
1781 return args->opcnt > session->se_fchannel.maxops;
1784 static bool nfsd4_request_too_big(struct svc_rqst *rqstp,
1785 struct nfsd4_session *session)
1787 struct xdr_buf *xb = &rqstp->rq_arg;
1789 return xb->len > session->se_fchannel.maxreq_sz;
1792 __be32
1793 nfsd4_sequence(struct svc_rqst *rqstp,
1794 struct nfsd4_compound_state *cstate,
1795 struct nfsd4_sequence *seq)
1797 struct nfsd4_compoundres *resp = rqstp->rq_resp;
1798 struct nfsd4_session *session;
1799 struct nfsd4_slot *slot;
1800 struct nfsd4_conn *conn;
1801 int status;
1803 if (resp->opcnt != 1)
1804 return nfserr_sequence_pos;
1807 * Will be either used or freed by nfsd4_sequence_check_conn
1808 * below.
1810 conn = alloc_conn(rqstp, NFS4_CDFC4_FORE);
1811 if (!conn)
1812 return nfserr_jukebox;
1814 spin_lock(&client_lock);
1815 status = nfserr_badsession;
1816 session = find_in_sessionid_hashtbl(&seq->sessionid);
1817 if (!session)
1818 goto out;
1820 status = nfserr_too_many_ops;
1821 if (nfsd4_session_too_many_ops(rqstp, session))
1822 goto out;
1824 status = nfserr_req_too_big;
1825 if (nfsd4_request_too_big(rqstp, session))
1826 goto out;
1828 status = nfserr_badslot;
1829 if (seq->slotid >= session->se_fchannel.maxreqs)
1830 goto out;
1832 slot = session->se_slots[seq->slotid];
1833 dprintk("%s: slotid %d\n", __func__, seq->slotid);
1835 /* We do not negotiate the number of slots yet, so set the
1836 * maxslots to the session maxreqs which is used to encode
1837 * sr_highest_slotid and the sr_target_slot id to maxslots */
1838 seq->maxslots = session->se_fchannel.maxreqs;
1840 status = check_slot_seqid(seq->seqid, slot->sl_seqid, slot->sl_inuse);
1841 if (status == nfserr_replay_cache) {
1842 cstate->slot = slot;
1843 cstate->session = session;
1844 /* Return the cached reply status and set cstate->status
1845 * for nfsd4_proc_compound processing */
1846 status = nfsd4_replay_cache_entry(resp, seq);
1847 cstate->status = nfserr_replay_cache;
1848 goto out;
1850 if (status)
1851 goto out;
1853 nfsd4_sequence_check_conn(conn, session);
1854 conn = NULL;
1856 /* Success! bump slot seqid */
1857 slot->sl_inuse = true;
1858 slot->sl_seqid = seq->seqid;
1859 slot->sl_cachethis = seq->cachethis;
1861 cstate->slot = slot;
1862 cstate->session = session;
1864 out:
1865 /* Hold a session reference until done processing the compound. */
1866 if (cstate->session) {
1867 struct nfs4_client *clp = session->se_client;
1869 nfsd4_get_session(cstate->session);
1870 atomic_inc(&clp->cl_refcount);
1871 if (clp->cl_cb_state == NFSD4_CB_DOWN)
1872 seq->status_flags |= SEQ4_STATUS_CB_PATH_DOWN;
1874 kfree(conn);
1875 spin_unlock(&client_lock);
1876 dprintk("%s: return %d\n", __func__, ntohl(status));
1877 return status;
1880 __be32
1881 nfsd4_reclaim_complete(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate, struct nfsd4_reclaim_complete *rc)
1883 int status = 0;
1885 if (rc->rca_one_fs) {
1886 if (!cstate->current_fh.fh_dentry)
1887 return nfserr_nofilehandle;
1889 * We don't take advantage of the rca_one_fs case.
1890 * That's OK, it's optional, we can safely ignore it.
1892 return nfs_ok;
1895 nfs4_lock_state();
1896 status = nfserr_complete_already;
1897 if (cstate->session->se_client->cl_firststate)
1898 goto out;
1900 status = nfserr_stale_clientid;
1901 if (is_client_expired(cstate->session->se_client))
1903 * The following error isn't really legal.
1904 * But we only get here if the client just explicitly
1905 * destroyed the client. Surely it no longer cares what
1906 * error it gets back on an operation for the dead
1907 * client.
1909 goto out;
1911 status = nfs_ok;
1912 nfsd4_create_clid_dir(cstate->session->se_client);
1913 out:
1914 nfs4_unlock_state();
1915 return status;
1918 __be32
1919 nfsd4_setclientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
1920 struct nfsd4_setclientid *setclid)
1922 struct xdr_netobj clname = {
1923 .len = setclid->se_namelen,
1924 .data = setclid->se_name,
1926 nfs4_verifier clverifier = setclid->se_verf;
1927 unsigned int strhashval;
1928 struct nfs4_client *conf, *unconf, *new;
1929 __be32 status;
1930 char dname[HEXDIR_LEN];
1932 if (!check_name(clname))
1933 return nfserr_inval;
1935 status = nfs4_make_rec_clidname(dname, &clname);
1936 if (status)
1937 return status;
1940 * XXX The Duplicate Request Cache (DRC) has been checked (??)
1941 * We get here on a DRC miss.
1944 strhashval = clientstr_hashval(dname);
1946 nfs4_lock_state();
1947 conf = find_confirmed_client_by_str(dname, strhashval);
1948 if (conf) {
1949 /* RFC 3530 14.2.33 CASE 0: */
1950 status = nfserr_clid_inuse;
1951 if (clp_used_exchangeid(conf))
1952 goto out;
1953 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
1954 char addr_str[INET6_ADDRSTRLEN];
1955 rpc_ntop((struct sockaddr *) &conf->cl_addr, addr_str,
1956 sizeof(addr_str));
1957 dprintk("NFSD: setclientid: string in use by client "
1958 "at %s\n", addr_str);
1959 goto out;
1963 * section 14.2.33 of RFC 3530 (under the heading "IMPLEMENTATION")
1964 * has a description of SETCLIENTID request processing consisting
1965 * of 5 bullet points, labeled as CASE0 - CASE4 below.
1967 unconf = find_unconfirmed_client_by_str(dname, strhashval);
1968 status = nfserr_jukebox;
1969 if (!conf) {
1971 * RFC 3530 14.2.33 CASE 4:
1972 * placed first, because it is the normal case
1974 if (unconf)
1975 expire_client(unconf);
1976 new = create_client(clname, dname, rqstp, &clverifier);
1977 if (new == NULL)
1978 goto out;
1979 gen_clid(new);
1980 } else if (same_verf(&conf->cl_verifier, &clverifier)) {
1982 * RFC 3530 14.2.33 CASE 1:
1983 * probable callback update
1985 if (unconf) {
1986 /* Note this is removing unconfirmed {*x***},
1987 * which is stronger than RFC recommended {vxc**}.
1988 * This has the advantage that there is at most
1989 * one {*x***} in either list at any time.
1991 expire_client(unconf);
1993 new = create_client(clname, dname, rqstp, &clverifier);
1994 if (new == NULL)
1995 goto out;
1996 copy_clid(new, conf);
1997 } else if (!unconf) {
1999 * RFC 3530 14.2.33 CASE 2:
2000 * probable client reboot; state will be removed if
2001 * confirmed.
2003 new = create_client(clname, dname, rqstp, &clverifier);
2004 if (new == NULL)
2005 goto out;
2006 gen_clid(new);
2007 } else {
2009 * RFC 3530 14.2.33 CASE 3:
2010 * probable client reboot; state will be removed if
2011 * confirmed.
2013 expire_client(unconf);
2014 new = create_client(clname, dname, rqstp, &clverifier);
2015 if (new == NULL)
2016 goto out;
2017 gen_clid(new);
2020 * XXX: we should probably set this at creation time, and check
2021 * for consistent minorversion use throughout:
2023 new->cl_minorversion = 0;
2024 gen_callback(new, setclid, rqstp);
2025 add_to_unconfirmed(new, strhashval);
2026 setclid->se_clientid.cl_boot = new->cl_clientid.cl_boot;
2027 setclid->se_clientid.cl_id = new->cl_clientid.cl_id;
2028 memcpy(setclid->se_confirm.data, new->cl_confirm.data, sizeof(setclid->se_confirm.data));
2029 status = nfs_ok;
2030 out:
2031 nfs4_unlock_state();
2032 return status;
2037 * Section 14.2.34 of RFC 3530 (under the heading "IMPLEMENTATION") has
2038 * a description of SETCLIENTID_CONFIRM request processing consisting of 4
2039 * bullets, labeled as CASE1 - CASE4 below.
2041 __be32
2042 nfsd4_setclientid_confirm(struct svc_rqst *rqstp,
2043 struct nfsd4_compound_state *cstate,
2044 struct nfsd4_setclientid_confirm *setclientid_confirm)
2046 struct sockaddr *sa = svc_addr(rqstp);
2047 struct nfs4_client *conf, *unconf;
2048 nfs4_verifier confirm = setclientid_confirm->sc_confirm;
2049 clientid_t * clid = &setclientid_confirm->sc_clientid;
2050 __be32 status;
2052 if (STALE_CLIENTID(clid))
2053 return nfserr_stale_clientid;
2055 * XXX The Duplicate Request Cache (DRC) has been checked (??)
2056 * We get here on a DRC miss.
2059 nfs4_lock_state();
2061 conf = find_confirmed_client(clid);
2062 unconf = find_unconfirmed_client(clid);
2064 status = nfserr_clid_inuse;
2065 if (conf && !rpc_cmp_addr((struct sockaddr *) &conf->cl_addr, sa))
2066 goto out;
2067 if (unconf && !rpc_cmp_addr((struct sockaddr *) &unconf->cl_addr, sa))
2068 goto out;
2071 * section 14.2.34 of RFC 3530 has a description of
2072 * SETCLIENTID_CONFIRM request processing consisting
2073 * of 4 bullet points, labeled as CASE1 - CASE4 below.
2075 if (conf && unconf && same_verf(&confirm, &unconf->cl_confirm)) {
2077 * RFC 3530 14.2.34 CASE 1:
2078 * callback update
2080 if (!same_creds(&conf->cl_cred, &unconf->cl_cred))
2081 status = nfserr_clid_inuse;
2082 else {
2083 nfsd4_change_callback(conf, &unconf->cl_cb_conn);
2084 nfsd4_probe_callback(conf);
2085 expire_client(unconf);
2086 status = nfs_ok;
2089 } else if (conf && !unconf) {
2091 * RFC 3530 14.2.34 CASE 2:
2092 * probable retransmitted request; play it safe and
2093 * do nothing.
2095 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred))
2096 status = nfserr_clid_inuse;
2097 else
2098 status = nfs_ok;
2099 } else if (!conf && unconf
2100 && same_verf(&unconf->cl_confirm, &confirm)) {
2102 * RFC 3530 14.2.34 CASE 3:
2103 * Normal case; new or rebooted client:
2105 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred)) {
2106 status = nfserr_clid_inuse;
2107 } else {
2108 unsigned int hash =
2109 clientstr_hashval(unconf->cl_recdir);
2110 conf = find_confirmed_client_by_str(unconf->cl_recdir,
2111 hash);
2112 if (conf) {
2113 nfsd4_remove_clid_dir(conf);
2114 expire_client(conf);
2116 move_to_confirmed(unconf);
2117 conf = unconf;
2118 nfsd4_probe_callback(conf);
2119 status = nfs_ok;
2121 } else if ((!conf || (conf && !same_verf(&conf->cl_confirm, &confirm)))
2122 && (!unconf || (unconf && !same_verf(&unconf->cl_confirm,
2123 &confirm)))) {
2125 * RFC 3530 14.2.34 CASE 4:
2126 * Client probably hasn't noticed that we rebooted yet.
2128 status = nfserr_stale_clientid;
2129 } else {
2130 /* check that we have hit one of the cases...*/
2131 status = nfserr_clid_inuse;
2133 out:
2134 nfs4_unlock_state();
2135 return status;
2138 /* OPEN Share state helper functions */
2139 static inline struct nfs4_file *
2140 alloc_init_file(struct inode *ino)
2142 struct nfs4_file *fp;
2143 unsigned int hashval = file_hashval(ino);
2145 fp = kmem_cache_alloc(file_slab, GFP_KERNEL);
2146 if (fp) {
2147 atomic_set(&fp->fi_ref, 1);
2148 INIT_LIST_HEAD(&fp->fi_hash);
2149 INIT_LIST_HEAD(&fp->fi_stateids);
2150 INIT_LIST_HEAD(&fp->fi_delegations);
2151 fp->fi_inode = igrab(ino);
2152 fp->fi_id = current_fileid++;
2153 fp->fi_had_conflict = false;
2154 fp->fi_lease = NULL;
2155 memset(fp->fi_fds, 0, sizeof(fp->fi_fds));
2156 memset(fp->fi_access, 0, sizeof(fp->fi_access));
2157 spin_lock(&recall_lock);
2158 list_add(&fp->fi_hash, &file_hashtbl[hashval]);
2159 spin_unlock(&recall_lock);
2160 return fp;
2162 return NULL;
2165 static void
2166 nfsd4_free_slab(struct kmem_cache **slab)
2168 if (*slab == NULL)
2169 return;
2170 kmem_cache_destroy(*slab);
2171 *slab = NULL;
2174 void
2175 nfsd4_free_slabs(void)
2177 nfsd4_free_slab(&stateowner_slab);
2178 nfsd4_free_slab(&file_slab);
2179 nfsd4_free_slab(&stateid_slab);
2180 nfsd4_free_slab(&deleg_slab);
2183 static int
2184 nfsd4_init_slabs(void)
2186 stateowner_slab = kmem_cache_create("nfsd4_stateowners",
2187 sizeof(struct nfs4_stateowner), 0, 0, NULL);
2188 if (stateowner_slab == NULL)
2189 goto out_nomem;
2190 file_slab = kmem_cache_create("nfsd4_files",
2191 sizeof(struct nfs4_file), 0, 0, NULL);
2192 if (file_slab == NULL)
2193 goto out_nomem;
2194 stateid_slab = kmem_cache_create("nfsd4_stateids",
2195 sizeof(struct nfs4_stateid), 0, 0, NULL);
2196 if (stateid_slab == NULL)
2197 goto out_nomem;
2198 deleg_slab = kmem_cache_create("nfsd4_delegations",
2199 sizeof(struct nfs4_delegation), 0, 0, NULL);
2200 if (deleg_slab == NULL)
2201 goto out_nomem;
2202 return 0;
2203 out_nomem:
2204 nfsd4_free_slabs();
2205 dprintk("nfsd4: out of memory while initializing nfsv4\n");
2206 return -ENOMEM;
2209 void
2210 nfs4_free_stateowner(struct kref *kref)
2212 struct nfs4_stateowner *sop =
2213 container_of(kref, struct nfs4_stateowner, so_ref);
2214 kfree(sop->so_owner.data);
2215 kmem_cache_free(stateowner_slab, sop);
2218 static void init_nfs4_replay(struct nfs4_replay *rp)
2220 rp->rp_status = nfserr_serverfault;
2221 rp->rp_buflen = 0;
2222 rp->rp_buf = rp->rp_ibuf;
2225 static inline struct nfs4_stateowner *alloc_stateowner(struct xdr_netobj *owner, struct nfs4_client *clp)
2227 struct nfs4_stateowner *sop;
2229 sop = kmem_cache_alloc(stateowner_slab, GFP_KERNEL);
2230 if (!sop)
2231 return NULL;
2233 sop->so_owner.data = kmemdup(owner->data, owner->len, GFP_KERNEL);
2234 if (!sop->so_owner.data) {
2235 kmem_cache_free(stateowner_slab, sop);
2236 return NULL;
2238 sop->so_owner.len = owner->len;
2240 kref_init(&sop->so_ref);
2241 INIT_LIST_HEAD(&sop->so_perclient);
2242 INIT_LIST_HEAD(&sop->so_stateids);
2243 INIT_LIST_HEAD(&sop->so_perstateid);
2244 INIT_LIST_HEAD(&sop->so_close_lru);
2245 sop->so_id = current_ownerid++;
2246 sop->so_time = 0;
2247 sop->so_client = clp;
2248 init_nfs4_replay(&sop->so_replay);
2249 return sop;
2252 static void hash_openowner(struct nfs4_stateowner *sop, struct nfs4_client *clp, unsigned int strhashval)
2254 unsigned int idhashval;
2256 idhashval = open_ownerid_hashval(sop->so_id);
2257 list_add(&sop->so_idhash, &open_ownerid_hashtbl[idhashval]);
2258 list_add(&sop->so_strhash, &open_ownerstr_hashtbl[strhashval]);
2259 list_add(&sop->so_perclient, &clp->cl_openowners);
2262 static struct nfs4_stateowner *
2263 alloc_init_open_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfsd4_open *open) {
2264 struct nfs4_stateowner *sop;
2266 sop = alloc_stateowner(&open->op_owner, clp);
2267 if (!sop)
2268 return NULL;
2269 sop->so_is_open_owner = 1;
2270 sop->so_seqid = open->op_seqid;
2271 sop->so_confirmed = 0;
2272 hash_openowner(sop, clp, strhashval);
2273 return sop;
2276 static inline void
2277 init_stateid(struct nfs4_stateid *stp, struct nfs4_file *fp, struct nfsd4_open *open) {
2278 struct nfs4_stateowner *sop = open->op_stateowner;
2279 unsigned int hashval = stateid_hashval(sop->so_id, fp->fi_id);
2281 INIT_LIST_HEAD(&stp->st_hash);
2282 INIT_LIST_HEAD(&stp->st_perstateowner);
2283 INIT_LIST_HEAD(&stp->st_lockowners);
2284 INIT_LIST_HEAD(&stp->st_perfile);
2285 list_add(&stp->st_hash, &stateid_hashtbl[hashval]);
2286 list_add(&stp->st_perstateowner, &sop->so_stateids);
2287 list_add(&stp->st_perfile, &fp->fi_stateids);
2288 stp->st_stateowner = sop;
2289 get_nfs4_file(fp);
2290 stp->st_file = fp;
2291 stp->st_stateid.si_boot = boot_time;
2292 stp->st_stateid.si_stateownerid = sop->so_id;
2293 stp->st_stateid.si_fileid = fp->fi_id;
2294 stp->st_stateid.si_generation = 0;
2295 stp->st_access_bmap = 0;
2296 stp->st_deny_bmap = 0;
2297 __set_bit(open->op_share_access & ~NFS4_SHARE_WANT_MASK,
2298 &stp->st_access_bmap);
2299 __set_bit(open->op_share_deny, &stp->st_deny_bmap);
2300 stp->st_openstp = NULL;
2303 static void
2304 move_to_close_lru(struct nfs4_stateowner *sop)
2306 dprintk("NFSD: move_to_close_lru nfs4_stateowner %p\n", sop);
2308 list_move_tail(&sop->so_close_lru, &close_lru);
2309 sop->so_time = get_seconds();
2312 static int
2313 same_owner_str(struct nfs4_stateowner *sop, struct xdr_netobj *owner,
2314 clientid_t *clid)
2316 return (sop->so_owner.len == owner->len) &&
2317 0 == memcmp(sop->so_owner.data, owner->data, owner->len) &&
2318 (sop->so_client->cl_clientid.cl_id == clid->cl_id);
2321 static struct nfs4_stateowner *
2322 find_openstateowner_str(unsigned int hashval, struct nfsd4_open *open)
2324 struct nfs4_stateowner *so = NULL;
2326 list_for_each_entry(so, &open_ownerstr_hashtbl[hashval], so_strhash) {
2327 if (same_owner_str(so, &open->op_owner, &open->op_clientid))
2328 return so;
2330 return NULL;
2333 /* search file_hashtbl[] for file */
2334 static struct nfs4_file *
2335 find_file(struct inode *ino)
2337 unsigned int hashval = file_hashval(ino);
2338 struct nfs4_file *fp;
2340 spin_lock(&recall_lock);
2341 list_for_each_entry(fp, &file_hashtbl[hashval], fi_hash) {
2342 if (fp->fi_inode == ino) {
2343 get_nfs4_file(fp);
2344 spin_unlock(&recall_lock);
2345 return fp;
2348 spin_unlock(&recall_lock);
2349 return NULL;
2352 static inline int access_valid(u32 x, u32 minorversion)
2354 if ((x & NFS4_SHARE_ACCESS_MASK) < NFS4_SHARE_ACCESS_READ)
2355 return 0;
2356 if ((x & NFS4_SHARE_ACCESS_MASK) > NFS4_SHARE_ACCESS_BOTH)
2357 return 0;
2358 x &= ~NFS4_SHARE_ACCESS_MASK;
2359 if (minorversion && x) {
2360 if ((x & NFS4_SHARE_WANT_MASK) > NFS4_SHARE_WANT_CANCEL)
2361 return 0;
2362 if ((x & NFS4_SHARE_WHEN_MASK) > NFS4_SHARE_PUSH_DELEG_WHEN_UNCONTENDED)
2363 return 0;
2364 x &= ~(NFS4_SHARE_WANT_MASK | NFS4_SHARE_WHEN_MASK);
2366 if (x)
2367 return 0;
2368 return 1;
2371 static inline int deny_valid(u32 x)
2373 /* Note: unlike access bits, deny bits may be zero. */
2374 return x <= NFS4_SHARE_DENY_BOTH;
2378 * Called to check deny when READ with all zero stateid or
2379 * WRITE with all zero or all one stateid
2381 static __be32
2382 nfs4_share_conflict(struct svc_fh *current_fh, unsigned int deny_type)
2384 struct inode *ino = current_fh->fh_dentry->d_inode;
2385 struct nfs4_file *fp;
2386 struct nfs4_stateid *stp;
2387 __be32 ret;
2389 dprintk("NFSD: nfs4_share_conflict\n");
2391 fp = find_file(ino);
2392 if (!fp)
2393 return nfs_ok;
2394 ret = nfserr_locked;
2395 /* Search for conflicting share reservations */
2396 list_for_each_entry(stp, &fp->fi_stateids, st_perfile) {
2397 if (test_bit(deny_type, &stp->st_deny_bmap) ||
2398 test_bit(NFS4_SHARE_DENY_BOTH, &stp->st_deny_bmap))
2399 goto out;
2401 ret = nfs_ok;
2402 out:
2403 put_nfs4_file(fp);
2404 return ret;
2407 static void nfsd_break_one_deleg(struct nfs4_delegation *dp)
2409 /* We're assuming the state code never drops its reference
2410 * without first removing the lease. Since we're in this lease
2411 * callback (and since the lease code is serialized by the kernel
2412 * lock) we know the server hasn't removed the lease yet, we know
2413 * it's safe to take a reference: */
2414 atomic_inc(&dp->dl_count);
2416 list_add_tail(&dp->dl_recall_lru, &del_recall_lru);
2418 /* only place dl_time is set. protected by lock_flocks*/
2419 dp->dl_time = get_seconds();
2421 nfsd4_cb_recall(dp);
2424 /* Called from break_lease() with lock_flocks() held. */
2425 static void nfsd_break_deleg_cb(struct file_lock *fl)
2427 struct nfs4_file *fp = (struct nfs4_file *)fl->fl_owner;
2428 struct nfs4_delegation *dp;
2430 BUG_ON(!fp);
2431 /* We assume break_lease is only called once per lease: */
2432 BUG_ON(fp->fi_had_conflict);
2434 * We don't want the locks code to timeout the lease for us;
2435 * we'll remove it ourself if a delegation isn't returned
2436 * in time:
2438 fl->fl_break_time = 0;
2440 spin_lock(&recall_lock);
2441 fp->fi_had_conflict = true;
2442 list_for_each_entry(dp, &fp->fi_delegations, dl_perfile)
2443 nfsd_break_one_deleg(dp);
2444 spin_unlock(&recall_lock);
2447 static
2448 int nfsd_change_deleg_cb(struct file_lock **onlist, int arg)
2450 if (arg & F_UNLCK)
2451 return lease_modify(onlist, arg);
2452 else
2453 return -EAGAIN;
2456 static const struct lock_manager_operations nfsd_lease_mng_ops = {
2457 .lm_break = nfsd_break_deleg_cb,
2458 .lm_change = nfsd_change_deleg_cb,
2462 __be32
2463 nfsd4_process_open1(struct nfsd4_compound_state *cstate,
2464 struct nfsd4_open *open)
2466 clientid_t *clientid = &open->op_clientid;
2467 struct nfs4_client *clp = NULL;
2468 unsigned int strhashval;
2469 struct nfs4_stateowner *sop = NULL;
2471 if (!check_name(open->op_owner))
2472 return nfserr_inval;
2474 if (STALE_CLIENTID(&open->op_clientid))
2475 return nfserr_stale_clientid;
2477 strhashval = open_ownerstr_hashval(clientid->cl_id, &open->op_owner);
2478 sop = find_openstateowner_str(strhashval, open);
2479 open->op_stateowner = sop;
2480 if (!sop) {
2481 /* Make sure the client's lease hasn't expired. */
2482 clp = find_confirmed_client(clientid);
2483 if (clp == NULL)
2484 return nfserr_expired;
2485 goto renew;
2487 /* When sessions are used, skip open sequenceid processing */
2488 if (nfsd4_has_session(cstate))
2489 goto renew;
2490 if (!sop->so_confirmed) {
2491 /* Replace unconfirmed owners without checking for replay. */
2492 clp = sop->so_client;
2493 release_openowner(sop);
2494 open->op_stateowner = NULL;
2495 goto renew;
2497 if (open->op_seqid == sop->so_seqid - 1) {
2498 if (sop->so_replay.rp_buflen)
2499 return nfserr_replay_me;
2500 /* The original OPEN failed so spectacularly
2501 * that we don't even have replay data saved!
2502 * Therefore, we have no choice but to continue
2503 * processing this OPEN; presumably, we'll
2504 * fail again for the same reason.
2506 dprintk("nfsd4_process_open1: replay with no replay cache\n");
2507 goto renew;
2509 if (open->op_seqid != sop->so_seqid)
2510 return nfserr_bad_seqid;
2511 renew:
2512 if (open->op_stateowner == NULL) {
2513 sop = alloc_init_open_stateowner(strhashval, clp, open);
2514 if (sop == NULL)
2515 return nfserr_jukebox;
2516 open->op_stateowner = sop;
2518 list_del_init(&sop->so_close_lru);
2519 renew_client(sop->so_client);
2520 return nfs_ok;
2523 static inline __be32
2524 nfs4_check_delegmode(struct nfs4_delegation *dp, int flags)
2526 if ((flags & WR_STATE) && (dp->dl_type == NFS4_OPEN_DELEGATE_READ))
2527 return nfserr_openmode;
2528 else
2529 return nfs_ok;
2532 static struct nfs4_delegation *
2533 find_delegation_file(struct nfs4_file *fp, stateid_t *stid)
2535 struct nfs4_delegation *dp;
2537 spin_lock(&recall_lock);
2538 list_for_each_entry(dp, &fp->fi_delegations, dl_perfile)
2539 if (dp->dl_stateid.si_stateownerid == stid->si_stateownerid) {
2540 spin_unlock(&recall_lock);
2541 return dp;
2543 spin_unlock(&recall_lock);
2544 return NULL;
2547 static int share_access_to_flags(u32 share_access)
2549 share_access &= ~NFS4_SHARE_WANT_MASK;
2551 return share_access == NFS4_SHARE_ACCESS_READ ? RD_STATE : WR_STATE;
2554 static __be32
2555 nfs4_check_deleg(struct nfs4_file *fp, struct nfsd4_open *open,
2556 struct nfs4_delegation **dp)
2558 int flags;
2559 __be32 status = nfserr_bad_stateid;
2561 *dp = find_delegation_file(fp, &open->op_delegate_stateid);
2562 if (*dp == NULL)
2563 goto out;
2564 flags = share_access_to_flags(open->op_share_access);
2565 status = nfs4_check_delegmode(*dp, flags);
2566 if (status)
2567 *dp = NULL;
2568 out:
2569 if (open->op_claim_type != NFS4_OPEN_CLAIM_DELEGATE_CUR)
2570 return nfs_ok;
2571 if (status)
2572 return status;
2573 open->op_stateowner->so_confirmed = 1;
2574 return nfs_ok;
2577 static __be32
2578 nfs4_check_open(struct nfs4_file *fp, struct nfsd4_open *open, struct nfs4_stateid **stpp)
2580 struct nfs4_stateid *local;
2581 __be32 status = nfserr_share_denied;
2582 struct nfs4_stateowner *sop = open->op_stateowner;
2584 list_for_each_entry(local, &fp->fi_stateids, st_perfile) {
2585 /* ignore lock owners */
2586 if (local->st_stateowner->so_is_open_owner == 0)
2587 continue;
2588 /* remember if we have seen this open owner */
2589 if (local->st_stateowner == sop)
2590 *stpp = local;
2591 /* check for conflicting share reservations */
2592 if (!test_share(local, open))
2593 goto out;
2595 status = 0;
2596 out:
2597 return status;
2600 static inline struct nfs4_stateid *
2601 nfs4_alloc_stateid(void)
2603 return kmem_cache_alloc(stateid_slab, GFP_KERNEL);
2606 static inline int nfs4_access_to_access(u32 nfs4_access)
2608 int flags = 0;
2610 if (nfs4_access & NFS4_SHARE_ACCESS_READ)
2611 flags |= NFSD_MAY_READ;
2612 if (nfs4_access & NFS4_SHARE_ACCESS_WRITE)
2613 flags |= NFSD_MAY_WRITE;
2614 return flags;
2617 static __be32 nfs4_get_vfs_file(struct svc_rqst *rqstp, struct nfs4_file *fp,
2618 struct svc_fh *cur_fh, struct nfsd4_open *open)
2620 __be32 status;
2621 int oflag = nfs4_access_to_omode(open->op_share_access);
2622 int access = nfs4_access_to_access(open->op_share_access);
2624 /* CLAIM_DELEGATE_CUR is used in response to a broken lease;
2625 * allowing it to break the lease and return EAGAIN leaves the
2626 * client unable to make progress in returning the delegation */
2627 if (open->op_claim_type == NFS4_OPEN_CLAIM_DELEGATE_CUR)
2628 access |= NFSD_MAY_NOT_BREAK_LEASE;
2630 if (!fp->fi_fds[oflag]) {
2631 status = nfsd_open(rqstp, cur_fh, S_IFREG, access,
2632 &fp->fi_fds[oflag]);
2633 if (status)
2634 return status;
2636 nfs4_file_get_access(fp, oflag);
2638 return nfs_ok;
2641 static __be32
2642 nfs4_new_open(struct svc_rqst *rqstp, struct nfs4_stateid **stpp,
2643 struct nfs4_file *fp, struct svc_fh *cur_fh,
2644 struct nfsd4_open *open)
2646 struct nfs4_stateid *stp;
2647 __be32 status;
2649 stp = nfs4_alloc_stateid();
2650 if (stp == NULL)
2651 return nfserr_jukebox;
2653 status = nfs4_get_vfs_file(rqstp, fp, cur_fh, open);
2654 if (status) {
2655 kmem_cache_free(stateid_slab, stp);
2656 return status;
2658 *stpp = stp;
2659 return 0;
2662 static inline __be32
2663 nfsd4_truncate(struct svc_rqst *rqstp, struct svc_fh *fh,
2664 struct nfsd4_open *open)
2666 struct iattr iattr = {
2667 .ia_valid = ATTR_SIZE,
2668 .ia_size = 0,
2670 if (!open->op_truncate)
2671 return 0;
2672 if (!(open->op_share_access & NFS4_SHARE_ACCESS_WRITE))
2673 return nfserr_inval;
2674 return nfsd_setattr(rqstp, fh, &iattr, 0, (time_t)0);
2677 static __be32
2678 nfs4_upgrade_open(struct svc_rqst *rqstp, struct nfs4_file *fp, struct svc_fh *cur_fh, struct nfs4_stateid *stp, struct nfsd4_open *open)
2680 u32 op_share_access = open->op_share_access & ~NFS4_SHARE_WANT_MASK;
2681 bool new_access;
2682 __be32 status;
2684 new_access = !test_bit(op_share_access, &stp->st_access_bmap);
2685 if (new_access) {
2686 status = nfs4_get_vfs_file(rqstp, fp, cur_fh, open);
2687 if (status)
2688 return status;
2690 status = nfsd4_truncate(rqstp, cur_fh, open);
2691 if (status) {
2692 if (new_access) {
2693 int oflag = nfs4_access_to_omode(op_share_access);
2694 nfs4_file_put_access(fp, oflag);
2696 return status;
2698 /* remember the open */
2699 __set_bit(op_share_access, &stp->st_access_bmap);
2700 __set_bit(open->op_share_deny, &stp->st_deny_bmap);
2702 return nfs_ok;
2706 static void
2707 nfs4_set_claim_prev(struct nfsd4_open *open)
2709 open->op_stateowner->so_confirmed = 1;
2710 open->op_stateowner->so_client->cl_firststate = 1;
2713 /* Should we give out recallable state?: */
2714 static bool nfsd4_cb_channel_good(struct nfs4_client *clp)
2716 if (clp->cl_cb_state == NFSD4_CB_UP)
2717 return true;
2719 * In the sessions case, since we don't have to establish a
2720 * separate connection for callbacks, we assume it's OK
2721 * until we hear otherwise:
2723 return clp->cl_minorversion && clp->cl_cb_state == NFSD4_CB_UNKNOWN;
2726 static struct file_lock *nfs4_alloc_init_lease(struct nfs4_delegation *dp, int flag)
2728 struct file_lock *fl;
2730 fl = locks_alloc_lock();
2731 if (!fl)
2732 return NULL;
2733 locks_init_lock(fl);
2734 fl->fl_lmops = &nfsd_lease_mng_ops;
2735 fl->fl_flags = FL_LEASE;
2736 fl->fl_type = flag == NFS4_OPEN_DELEGATE_READ? F_RDLCK: F_WRLCK;
2737 fl->fl_end = OFFSET_MAX;
2738 fl->fl_owner = (fl_owner_t)(dp->dl_file);
2739 fl->fl_pid = current->tgid;
2740 return fl;
2743 static int nfs4_setlease(struct nfs4_delegation *dp, int flag)
2745 struct nfs4_file *fp = dp->dl_file;
2746 struct file_lock *fl;
2747 int status;
2749 fl = nfs4_alloc_init_lease(dp, flag);
2750 if (!fl)
2751 return -ENOMEM;
2752 fl->fl_file = find_readable_file(fp);
2753 list_add(&dp->dl_perclnt, &dp->dl_client->cl_delegations);
2754 status = vfs_setlease(fl->fl_file, fl->fl_type, &fl);
2755 if (status) {
2756 list_del_init(&dp->dl_perclnt);
2757 locks_free_lock(fl);
2758 return -ENOMEM;
2760 fp->fi_lease = fl;
2761 fp->fi_deleg_file = fl->fl_file;
2762 get_file(fp->fi_deleg_file);
2763 atomic_set(&fp->fi_delegees, 1);
2764 list_add(&dp->dl_perfile, &fp->fi_delegations);
2765 return 0;
2768 static int nfs4_set_delegation(struct nfs4_delegation *dp, int flag)
2770 struct nfs4_file *fp = dp->dl_file;
2772 if (!fp->fi_lease)
2773 return nfs4_setlease(dp, flag);
2774 spin_lock(&recall_lock);
2775 if (fp->fi_had_conflict) {
2776 spin_unlock(&recall_lock);
2777 return -EAGAIN;
2779 atomic_inc(&fp->fi_delegees);
2780 list_add(&dp->dl_perfile, &fp->fi_delegations);
2781 spin_unlock(&recall_lock);
2782 list_add(&dp->dl_perclnt, &dp->dl_client->cl_delegations);
2783 return 0;
2787 * Attempt to hand out a delegation.
2789 static void
2790 nfs4_open_delegation(struct svc_fh *fh, struct nfsd4_open *open, struct nfs4_stateid *stp)
2792 struct nfs4_delegation *dp;
2793 struct nfs4_stateowner *sop = stp->st_stateowner;
2794 int cb_up;
2795 int status, flag = 0;
2797 cb_up = nfsd4_cb_channel_good(sop->so_client);
2798 flag = NFS4_OPEN_DELEGATE_NONE;
2799 open->op_recall = 0;
2800 switch (open->op_claim_type) {
2801 case NFS4_OPEN_CLAIM_PREVIOUS:
2802 if (!cb_up)
2803 open->op_recall = 1;
2804 flag = open->op_delegate_type;
2805 if (flag == NFS4_OPEN_DELEGATE_NONE)
2806 goto out;
2807 break;
2808 case NFS4_OPEN_CLAIM_NULL:
2809 /* Let's not give out any delegations till everyone's
2810 * had the chance to reclaim theirs.... */
2811 if (locks_in_grace())
2812 goto out;
2813 if (!cb_up || !sop->so_confirmed)
2814 goto out;
2815 if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
2816 flag = NFS4_OPEN_DELEGATE_WRITE;
2817 else
2818 flag = NFS4_OPEN_DELEGATE_READ;
2819 break;
2820 default:
2821 goto out;
2824 dp = alloc_init_deleg(sop->so_client, stp, fh, flag);
2825 if (dp == NULL)
2826 goto out_no_deleg;
2827 status = nfs4_set_delegation(dp, flag);
2828 if (status)
2829 goto out_free;
2831 memcpy(&open->op_delegate_stateid, &dp->dl_stateid, sizeof(dp->dl_stateid));
2833 dprintk("NFSD: delegation stateid=" STATEID_FMT "\n",
2834 STATEID_VAL(&dp->dl_stateid));
2835 out:
2836 if (open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS
2837 && flag == NFS4_OPEN_DELEGATE_NONE
2838 && open->op_delegate_type != NFS4_OPEN_DELEGATE_NONE)
2839 dprintk("NFSD: WARNING: refusing delegation reclaim\n");
2840 open->op_delegate_type = flag;
2841 return;
2842 out_free:
2843 nfs4_put_delegation(dp);
2844 out_no_deleg:
2845 flag = NFS4_OPEN_DELEGATE_NONE;
2846 goto out;
2850 * called with nfs4_lock_state() held.
2852 __be32
2853 nfsd4_process_open2(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_open *open)
2855 struct nfsd4_compoundres *resp = rqstp->rq_resp;
2856 struct nfs4_file *fp = NULL;
2857 struct inode *ino = current_fh->fh_dentry->d_inode;
2858 struct nfs4_stateid *stp = NULL;
2859 struct nfs4_delegation *dp = NULL;
2860 __be32 status;
2862 status = nfserr_inval;
2863 if (!access_valid(open->op_share_access, resp->cstate.minorversion)
2864 || !deny_valid(open->op_share_deny))
2865 goto out;
2867 * Lookup file; if found, lookup stateid and check open request,
2868 * and check for delegations in the process of being recalled.
2869 * If not found, create the nfs4_file struct
2871 fp = find_file(ino);
2872 if (fp) {
2873 if ((status = nfs4_check_open(fp, open, &stp)))
2874 goto out;
2875 status = nfs4_check_deleg(fp, open, &dp);
2876 if (status)
2877 goto out;
2878 } else {
2879 status = nfserr_bad_stateid;
2880 if (open->op_claim_type == NFS4_OPEN_CLAIM_DELEGATE_CUR)
2881 goto out;
2882 status = nfserr_jukebox;
2883 fp = alloc_init_file(ino);
2884 if (fp == NULL)
2885 goto out;
2889 * OPEN the file, or upgrade an existing OPEN.
2890 * If truncate fails, the OPEN fails.
2892 if (stp) {
2893 /* Stateid was found, this is an OPEN upgrade */
2894 status = nfs4_upgrade_open(rqstp, fp, current_fh, stp, open);
2895 if (status)
2896 goto out;
2897 update_stateid(&stp->st_stateid);
2898 } else {
2899 status = nfs4_new_open(rqstp, &stp, fp, current_fh, open);
2900 if (status)
2901 goto out;
2902 init_stateid(stp, fp, open);
2903 status = nfsd4_truncate(rqstp, current_fh, open);
2904 if (status) {
2905 release_open_stateid(stp);
2906 goto out;
2908 if (nfsd4_has_session(&resp->cstate))
2909 update_stateid(&stp->st_stateid);
2911 memcpy(&open->op_stateid, &stp->st_stateid, sizeof(stateid_t));
2913 if (nfsd4_has_session(&resp->cstate))
2914 open->op_stateowner->so_confirmed = 1;
2917 * Attempt to hand out a delegation. No error return, because the
2918 * OPEN succeeds even if we fail.
2920 nfs4_open_delegation(current_fh, open, stp);
2922 status = nfs_ok;
2924 dprintk("%s: stateid=" STATEID_FMT "\n", __func__,
2925 STATEID_VAL(&stp->st_stateid));
2926 out:
2927 if (fp)
2928 put_nfs4_file(fp);
2929 if (status == 0 && open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS)
2930 nfs4_set_claim_prev(open);
2932 * To finish the open response, we just need to set the rflags.
2934 open->op_rflags = NFS4_OPEN_RESULT_LOCKTYPE_POSIX;
2935 if (!open->op_stateowner->so_confirmed &&
2936 !nfsd4_has_session(&resp->cstate))
2937 open->op_rflags |= NFS4_OPEN_RESULT_CONFIRM;
2939 return status;
2942 __be32
2943 nfsd4_renew(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2944 clientid_t *clid)
2946 struct nfs4_client *clp;
2947 __be32 status;
2949 nfs4_lock_state();
2950 dprintk("process_renew(%08x/%08x): starting\n",
2951 clid->cl_boot, clid->cl_id);
2952 status = nfserr_stale_clientid;
2953 if (STALE_CLIENTID(clid))
2954 goto out;
2955 clp = find_confirmed_client(clid);
2956 status = nfserr_expired;
2957 if (clp == NULL) {
2958 /* We assume the client took too long to RENEW. */
2959 dprintk("nfsd4_renew: clientid not found!\n");
2960 goto out;
2962 renew_client(clp);
2963 status = nfserr_cb_path_down;
2964 if (!list_empty(&clp->cl_delegations)
2965 && clp->cl_cb_state != NFSD4_CB_UP)
2966 goto out;
2967 status = nfs_ok;
2968 out:
2969 nfs4_unlock_state();
2970 return status;
2973 static struct lock_manager nfsd4_manager = {
2976 static void
2977 nfsd4_end_grace(void)
2979 dprintk("NFSD: end of grace period\n");
2980 nfsd4_recdir_purge_old();
2981 locks_end_grace(&nfsd4_manager);
2983 * Now that every NFSv4 client has had the chance to recover and
2984 * to see the (possibly new, possibly shorter) lease time, we
2985 * can safely set the next grace time to the current lease time:
2987 nfsd4_grace = nfsd4_lease;
2990 static time_t
2991 nfs4_laundromat(void)
2993 struct nfs4_client *clp;
2994 struct nfs4_stateowner *sop;
2995 struct nfs4_delegation *dp;
2996 struct list_head *pos, *next, reaplist;
2997 time_t cutoff = get_seconds() - nfsd4_lease;
2998 time_t t, clientid_val = nfsd4_lease;
2999 time_t u, test_val = nfsd4_lease;
3001 nfs4_lock_state();
3003 dprintk("NFSD: laundromat service - starting\n");
3004 if (locks_in_grace())
3005 nfsd4_end_grace();
3006 INIT_LIST_HEAD(&reaplist);
3007 spin_lock(&client_lock);
3008 list_for_each_safe(pos, next, &client_lru) {
3009 clp = list_entry(pos, struct nfs4_client, cl_lru);
3010 if (time_after((unsigned long)clp->cl_time, (unsigned long)cutoff)) {
3011 t = clp->cl_time - cutoff;
3012 if (clientid_val > t)
3013 clientid_val = t;
3014 break;
3016 if (atomic_read(&clp->cl_refcount)) {
3017 dprintk("NFSD: client in use (clientid %08x)\n",
3018 clp->cl_clientid.cl_id);
3019 continue;
3021 unhash_client_locked(clp);
3022 list_add(&clp->cl_lru, &reaplist);
3024 spin_unlock(&client_lock);
3025 list_for_each_safe(pos, next, &reaplist) {
3026 clp = list_entry(pos, struct nfs4_client, cl_lru);
3027 dprintk("NFSD: purging unused client (clientid %08x)\n",
3028 clp->cl_clientid.cl_id);
3029 nfsd4_remove_clid_dir(clp);
3030 expire_client(clp);
3032 spin_lock(&recall_lock);
3033 list_for_each_safe(pos, next, &del_recall_lru) {
3034 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
3035 if (time_after((unsigned long)dp->dl_time, (unsigned long)cutoff)) {
3036 u = dp->dl_time - cutoff;
3037 if (test_val > u)
3038 test_val = u;
3039 break;
3041 list_move(&dp->dl_recall_lru, &reaplist);
3043 spin_unlock(&recall_lock);
3044 list_for_each_safe(pos, next, &reaplist) {
3045 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
3046 list_del_init(&dp->dl_recall_lru);
3047 unhash_delegation(dp);
3049 test_val = nfsd4_lease;
3050 list_for_each_safe(pos, next, &close_lru) {
3051 sop = list_entry(pos, struct nfs4_stateowner, so_close_lru);
3052 if (time_after((unsigned long)sop->so_time, (unsigned long)cutoff)) {
3053 u = sop->so_time - cutoff;
3054 if (test_val > u)
3055 test_val = u;
3056 break;
3058 dprintk("NFSD: purging unused open stateowner (so_id %d)\n",
3059 sop->so_id);
3060 release_openowner(sop);
3062 if (clientid_val < NFSD_LAUNDROMAT_MINTIMEOUT)
3063 clientid_val = NFSD_LAUNDROMAT_MINTIMEOUT;
3064 nfs4_unlock_state();
3065 return clientid_val;
3068 static struct workqueue_struct *laundry_wq;
3069 static void laundromat_main(struct work_struct *);
3070 static DECLARE_DELAYED_WORK(laundromat_work, laundromat_main);
3072 static void
3073 laundromat_main(struct work_struct *not_used)
3075 time_t t;
3077 t = nfs4_laundromat();
3078 dprintk("NFSD: laundromat_main - sleeping for %ld seconds\n", t);
3079 queue_delayed_work(laundry_wq, &laundromat_work, t*HZ);
3082 static struct nfs4_stateowner *
3083 search_close_lru(u32 st_id, int flags)
3085 struct nfs4_stateowner *local = NULL;
3087 if (flags & CLOSE_STATE) {
3088 list_for_each_entry(local, &close_lru, so_close_lru) {
3089 if (local->so_id == st_id)
3090 return local;
3093 return NULL;
3096 static inline int
3097 nfs4_check_fh(struct svc_fh *fhp, struct nfs4_stateid *stp)
3099 return fhp->fh_dentry->d_inode != stp->st_file->fi_inode;
3102 static int
3103 STALE_STATEID(stateid_t *stateid)
3105 if (stateid->si_boot == boot_time)
3106 return 0;
3107 dprintk("NFSD: stale stateid " STATEID_FMT "!\n",
3108 STATEID_VAL(stateid));
3109 return 1;
3112 static inline int
3113 access_permit_read(unsigned long access_bmap)
3115 return test_bit(NFS4_SHARE_ACCESS_READ, &access_bmap) ||
3116 test_bit(NFS4_SHARE_ACCESS_BOTH, &access_bmap) ||
3117 test_bit(NFS4_SHARE_ACCESS_WRITE, &access_bmap);
3120 static inline int
3121 access_permit_write(unsigned long access_bmap)
3123 return test_bit(NFS4_SHARE_ACCESS_WRITE, &access_bmap) ||
3124 test_bit(NFS4_SHARE_ACCESS_BOTH, &access_bmap);
3127 static
3128 __be32 nfs4_check_openmode(struct nfs4_stateid *stp, int flags)
3130 __be32 status = nfserr_openmode;
3132 /* For lock stateid's, we test the parent open, not the lock: */
3133 if (stp->st_openstp)
3134 stp = stp->st_openstp;
3135 if ((flags & WR_STATE) && (!access_permit_write(stp->st_access_bmap)))
3136 goto out;
3137 if ((flags & RD_STATE) && (!access_permit_read(stp->st_access_bmap)))
3138 goto out;
3139 status = nfs_ok;
3140 out:
3141 return status;
3144 static inline __be32
3145 check_special_stateids(svc_fh *current_fh, stateid_t *stateid, int flags)
3147 if (ONE_STATEID(stateid) && (flags & RD_STATE))
3148 return nfs_ok;
3149 else if (locks_in_grace()) {
3150 /* Answer in remaining cases depends on existence of
3151 * conflicting state; so we must wait out the grace period. */
3152 return nfserr_grace;
3153 } else if (flags & WR_STATE)
3154 return nfs4_share_conflict(current_fh,
3155 NFS4_SHARE_DENY_WRITE);
3156 else /* (flags & RD_STATE) && ZERO_STATEID(stateid) */
3157 return nfs4_share_conflict(current_fh,
3158 NFS4_SHARE_DENY_READ);
3162 * Allow READ/WRITE during grace period on recovered state only for files
3163 * that are not able to provide mandatory locking.
3165 static inline int
3166 grace_disallows_io(struct inode *inode)
3168 return locks_in_grace() && mandatory_lock(inode);
3171 static int check_stateid_generation(stateid_t *in, stateid_t *ref, bool has_session)
3174 * When sessions are used the stateid generation number is ignored
3175 * when it is zero.
3177 if (has_session && in->si_generation == 0)
3178 goto out;
3180 /* If the client sends us a stateid from the future, it's buggy: */
3181 if (in->si_generation > ref->si_generation)
3182 return nfserr_bad_stateid;
3184 * The following, however, can happen. For example, if the
3185 * client sends an open and some IO at the same time, the open
3186 * may bump si_generation while the IO is still in flight.
3187 * Thanks to hard links and renames, the client never knows what
3188 * file an open will affect. So it could avoid that situation
3189 * only by serializing all opens and IO from the same open
3190 * owner. To recover from the old_stateid error, the client
3191 * will just have to retry the IO:
3193 if (in->si_generation < ref->si_generation)
3194 return nfserr_old_stateid;
3195 out:
3196 return nfs_ok;
3199 static int is_delegation_stateid(stateid_t *stateid)
3201 return stateid->si_fileid == 0;
3204 static int is_open_stateid(struct nfs4_stateid *stateid)
3206 return stateid->st_openstp == NULL;
3209 __be32 nfs4_validate_stateid(stateid_t *stateid, bool has_session)
3211 struct nfs4_stateid *stp = NULL;
3212 __be32 status = nfserr_stale_stateid;
3214 if (STALE_STATEID(stateid))
3215 goto out;
3217 status = nfserr_expired;
3218 stp = search_for_stateid(stateid);
3219 if (!stp)
3220 goto out;
3221 status = nfserr_bad_stateid;
3223 if (!stp->st_stateowner->so_confirmed)
3224 goto out;
3226 status = check_stateid_generation(stateid, &stp->st_stateid, has_session);
3227 if (status)
3228 goto out;
3230 status = nfs_ok;
3231 out:
3232 return status;
3236 * Checks for stateid operations
3238 __be32
3239 nfs4_preprocess_stateid_op(struct nfsd4_compound_state *cstate,
3240 stateid_t *stateid, int flags, struct file **filpp)
3242 struct nfs4_stateid *stp = NULL;
3243 struct nfs4_delegation *dp = NULL;
3244 struct svc_fh *current_fh = &cstate->current_fh;
3245 struct inode *ino = current_fh->fh_dentry->d_inode;
3246 __be32 status;
3248 if (filpp)
3249 *filpp = NULL;
3251 if (grace_disallows_io(ino))
3252 return nfserr_grace;
3254 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
3255 return check_special_stateids(current_fh, stateid, flags);
3257 status = nfserr_stale_stateid;
3258 if (STALE_STATEID(stateid))
3259 goto out;
3262 * We assume that any stateid that has the current boot time,
3263 * but that we can't find, is expired:
3265 status = nfserr_expired;
3266 if (is_delegation_stateid(stateid)) {
3267 dp = find_delegation_stateid(ino, stateid);
3268 if (!dp)
3269 goto out;
3270 status = check_stateid_generation(stateid, &dp->dl_stateid, nfsd4_has_session(cstate));
3271 if (status)
3272 goto out;
3273 status = nfs4_check_delegmode(dp, flags);
3274 if (status)
3275 goto out;
3276 renew_client(dp->dl_client);
3277 if (filpp) {
3278 *filpp = dp->dl_file->fi_deleg_file;
3279 BUG_ON(!*filpp);
3281 } else { /* open or lock stateid */
3282 stp = find_stateid(stateid, flags);
3283 if (!stp)
3284 goto out;
3285 status = nfserr_bad_stateid;
3286 if (nfs4_check_fh(current_fh, stp))
3287 goto out;
3288 if (!stp->st_stateowner->so_confirmed)
3289 goto out;
3290 status = check_stateid_generation(stateid, &stp->st_stateid,
3291 nfsd4_has_session(cstate));
3292 if (status)
3293 goto out;
3294 status = nfs4_check_openmode(stp, flags);
3295 if (status)
3296 goto out;
3297 renew_client(stp->st_stateowner->so_client);
3298 if (filpp) {
3299 if (flags & RD_STATE)
3300 *filpp = find_readable_file(stp->st_file);
3301 else
3302 *filpp = find_writeable_file(stp->st_file);
3305 status = nfs_ok;
3306 out:
3307 return status;
3310 static __be32
3311 nfsd4_free_delegation_stateid(stateid_t *stateid)
3313 struct nfs4_delegation *dp = search_for_delegation(stateid);
3314 if (dp)
3315 return nfserr_locks_held;
3316 return nfserr_bad_stateid;
3319 static __be32
3320 nfsd4_free_lock_stateid(struct nfs4_stateid *stp)
3322 if (check_for_locks(stp->st_file, stp->st_stateowner))
3323 return nfserr_locks_held;
3324 release_lock_stateid(stp);
3325 return nfs_ok;
3329 * Test if the stateid is valid
3331 __be32
3332 nfsd4_test_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3333 struct nfsd4_test_stateid *test_stateid)
3335 test_stateid->ts_has_session = nfsd4_has_session(cstate);
3336 return nfs_ok;
3340 * Free a state id
3342 __be32
3343 nfsd4_free_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3344 struct nfsd4_free_stateid *free_stateid)
3346 stateid_t *stateid = &free_stateid->fr_stateid;
3347 struct nfs4_stateid *stp;
3348 __be32 ret;
3350 nfs4_lock_state();
3351 if (is_delegation_stateid(stateid)) {
3352 ret = nfsd4_free_delegation_stateid(stateid);
3353 goto out;
3356 stp = search_for_stateid(stateid);
3357 if (!stp) {
3358 ret = nfserr_bad_stateid;
3359 goto out;
3361 if (stateid->si_generation != 0) {
3362 if (stateid->si_generation < stp->st_stateid.si_generation) {
3363 ret = nfserr_old_stateid;
3364 goto out;
3366 if (stateid->si_generation > stp->st_stateid.si_generation) {
3367 ret = nfserr_bad_stateid;
3368 goto out;
3372 if (is_open_stateid(stp)) {
3373 ret = nfserr_locks_held;
3374 goto out;
3375 } else {
3376 ret = nfsd4_free_lock_stateid(stp);
3377 goto out;
3380 out:
3381 nfs4_unlock_state();
3382 return ret;
3385 static inline int
3386 setlkflg (int type)
3388 return (type == NFS4_READW_LT || type == NFS4_READ_LT) ?
3389 RD_STATE : WR_STATE;
3393 * Checks for sequence id mutating operations.
3395 static __be32
3396 nfs4_preprocess_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
3397 stateid_t *stateid, int flags,
3398 struct nfs4_stateowner **sopp,
3399 struct nfs4_stateid **stpp)
3401 struct nfs4_stateid *stp;
3402 struct nfs4_stateowner *sop;
3403 struct svc_fh *current_fh = &cstate->current_fh;
3404 __be32 status;
3406 dprintk("NFSD: %s: seqid=%d stateid = " STATEID_FMT "\n", __func__,
3407 seqid, STATEID_VAL(stateid));
3409 *stpp = NULL;
3410 *sopp = NULL;
3412 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid)) {
3413 dprintk("NFSD: preprocess_seqid_op: magic stateid!\n");
3414 return nfserr_bad_stateid;
3417 if (STALE_STATEID(stateid))
3418 return nfserr_stale_stateid;
3421 * We return BAD_STATEID if filehandle doesn't match stateid,
3422 * the confirmed flag is incorrecly set, or the generation
3423 * number is incorrect.
3425 stp = find_stateid(stateid, flags);
3426 if (stp == NULL) {
3428 * Also, we should make sure this isn't just the result of
3429 * a replayed close:
3431 sop = search_close_lru(stateid->si_stateownerid, flags);
3432 /* It's not stale; let's assume it's expired: */
3433 if (sop == NULL)
3434 return nfserr_expired;
3435 *sopp = sop;
3436 goto check_replay;
3439 *stpp = stp;
3440 *sopp = sop = stp->st_stateowner;
3442 if (nfs4_check_fh(current_fh, stp)) {
3443 dprintk("NFSD: preprocess_seqid_op: fh-stateid mismatch!\n");
3444 return nfserr_bad_stateid;
3448 * We now validate the seqid and stateid generation numbers.
3449 * For the moment, we ignore the possibility of
3450 * generation number wraparound.
3452 if (!nfsd4_has_session(cstate) && seqid != sop->so_seqid)
3453 goto check_replay;
3455 if (sop->so_confirmed && flags & CONFIRM) {
3456 dprintk("NFSD: preprocess_seqid_op: expected"
3457 " unconfirmed stateowner!\n");
3458 return nfserr_bad_stateid;
3460 if (!sop->so_confirmed && !(flags & CONFIRM)) {
3461 dprintk("NFSD: preprocess_seqid_op: stateowner not"
3462 " confirmed yet!\n");
3463 return nfserr_bad_stateid;
3465 status = check_stateid_generation(stateid, &stp->st_stateid, nfsd4_has_session(cstate));
3466 if (status)
3467 return status;
3468 renew_client(sop->so_client);
3469 return nfs_ok;
3471 check_replay:
3472 if (seqid == sop->so_seqid - 1) {
3473 dprintk("NFSD: preprocess_seqid_op: retransmission?\n");
3474 /* indicate replay to calling function */
3475 return nfserr_replay_me;
3477 dprintk("NFSD: preprocess_seqid_op: bad seqid (expected %d, got %d)\n",
3478 sop->so_seqid, seqid);
3479 *sopp = NULL;
3480 return nfserr_bad_seqid;
3483 __be32
3484 nfsd4_open_confirm(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3485 struct nfsd4_open_confirm *oc)
3487 __be32 status;
3488 struct nfs4_stateowner *sop;
3489 struct nfs4_stateid *stp;
3491 dprintk("NFSD: nfsd4_open_confirm on file %.*s\n",
3492 (int)cstate->current_fh.fh_dentry->d_name.len,
3493 cstate->current_fh.fh_dentry->d_name.name);
3495 status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0);
3496 if (status)
3497 return status;
3499 nfs4_lock_state();
3501 if ((status = nfs4_preprocess_seqid_op(cstate,
3502 oc->oc_seqid, &oc->oc_req_stateid,
3503 CONFIRM | OPEN_STATE,
3504 &oc->oc_stateowner, &stp)))
3505 goto out;
3507 sop = oc->oc_stateowner;
3508 sop->so_confirmed = 1;
3509 update_stateid(&stp->st_stateid);
3510 memcpy(&oc->oc_resp_stateid, &stp->st_stateid, sizeof(stateid_t));
3511 dprintk("NFSD: %s: success, seqid=%d stateid=" STATEID_FMT "\n",
3512 __func__, oc->oc_seqid, STATEID_VAL(&stp->st_stateid));
3514 nfsd4_create_clid_dir(sop->so_client);
3515 out:
3516 if (oc->oc_stateowner) {
3517 nfs4_get_stateowner(oc->oc_stateowner);
3518 cstate->replay_owner = oc->oc_stateowner;
3520 nfs4_unlock_state();
3521 return status;
3524 static inline void nfs4_file_downgrade(struct nfs4_stateid *stp, unsigned int to_access)
3526 int i;
3528 for (i = 1; i < 4; i++) {
3529 if (test_bit(i, &stp->st_access_bmap) && !(i & to_access)) {
3530 nfs4_file_put_access(stp->st_file, i);
3531 __clear_bit(i, &stp->st_access_bmap);
3536 static void
3537 reset_union_bmap_deny(unsigned long deny, unsigned long *bmap)
3539 int i;
3540 for (i = 0; i < 4; i++) {
3541 if ((i & deny) != i)
3542 __clear_bit(i, bmap);
3546 __be32
3547 nfsd4_open_downgrade(struct svc_rqst *rqstp,
3548 struct nfsd4_compound_state *cstate,
3549 struct nfsd4_open_downgrade *od)
3551 __be32 status;
3552 struct nfs4_stateid *stp;
3554 dprintk("NFSD: nfsd4_open_downgrade on file %.*s\n",
3555 (int)cstate->current_fh.fh_dentry->d_name.len,
3556 cstate->current_fh.fh_dentry->d_name.name);
3558 if (!access_valid(od->od_share_access, cstate->minorversion)
3559 || !deny_valid(od->od_share_deny))
3560 return nfserr_inval;
3562 nfs4_lock_state();
3563 if ((status = nfs4_preprocess_seqid_op(cstate,
3564 od->od_seqid,
3565 &od->od_stateid,
3566 OPEN_STATE,
3567 &od->od_stateowner, &stp)))
3568 goto out;
3570 status = nfserr_inval;
3571 if (!test_bit(od->od_share_access, &stp->st_access_bmap)) {
3572 dprintk("NFSD:access not a subset current bitmap: 0x%lx, input access=%08x\n",
3573 stp->st_access_bmap, od->od_share_access);
3574 goto out;
3576 if (!test_bit(od->od_share_deny, &stp->st_deny_bmap)) {
3577 dprintk("NFSD:deny not a subset current bitmap: 0x%lx, input deny=%08x\n",
3578 stp->st_deny_bmap, od->od_share_deny);
3579 goto out;
3581 nfs4_file_downgrade(stp, od->od_share_access);
3583 reset_union_bmap_deny(od->od_share_deny, &stp->st_deny_bmap);
3585 update_stateid(&stp->st_stateid);
3586 memcpy(&od->od_stateid, &stp->st_stateid, sizeof(stateid_t));
3587 status = nfs_ok;
3588 out:
3589 if (od->od_stateowner) {
3590 nfs4_get_stateowner(od->od_stateowner);
3591 cstate->replay_owner = od->od_stateowner;
3593 nfs4_unlock_state();
3594 return status;
3598 * nfs4_unlock_state() called after encode
3600 __be32
3601 nfsd4_close(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3602 struct nfsd4_close *close)
3604 __be32 status;
3605 struct nfs4_stateid *stp;
3607 dprintk("NFSD: nfsd4_close on file %.*s\n",
3608 (int)cstate->current_fh.fh_dentry->d_name.len,
3609 cstate->current_fh.fh_dentry->d_name.name);
3611 nfs4_lock_state();
3612 /* check close_lru for replay */
3613 if ((status = nfs4_preprocess_seqid_op(cstate,
3614 close->cl_seqid,
3615 &close->cl_stateid,
3616 OPEN_STATE | CLOSE_STATE,
3617 &close->cl_stateowner, &stp)))
3618 goto out;
3619 status = nfs_ok;
3620 update_stateid(&stp->st_stateid);
3621 memcpy(&close->cl_stateid, &stp->st_stateid, sizeof(stateid_t));
3623 /* release_stateid() calls nfsd_close() if needed */
3624 release_open_stateid(stp);
3626 /* place unused nfs4_stateowners on so_close_lru list to be
3627 * released by the laundromat service after the lease period
3628 * to enable us to handle CLOSE replay
3630 if (list_empty(&close->cl_stateowner->so_stateids))
3631 move_to_close_lru(close->cl_stateowner);
3632 out:
3633 if (close->cl_stateowner) {
3634 nfs4_get_stateowner(close->cl_stateowner);
3635 cstate->replay_owner = close->cl_stateowner;
3637 nfs4_unlock_state();
3638 return status;
3641 __be32
3642 nfsd4_delegreturn(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3643 struct nfsd4_delegreturn *dr)
3645 struct nfs4_delegation *dp;
3646 stateid_t *stateid = &dr->dr_stateid;
3647 struct inode *inode;
3648 __be32 status;
3650 if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
3651 return status;
3652 inode = cstate->current_fh.fh_dentry->d_inode;
3654 nfs4_lock_state();
3655 status = nfserr_bad_stateid;
3656 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
3657 goto out;
3658 status = nfserr_stale_stateid;
3659 if (STALE_STATEID(stateid))
3660 goto out;
3661 status = nfserr_bad_stateid;
3662 if (!is_delegation_stateid(stateid))
3663 goto out;
3664 status = nfserr_expired;
3665 dp = find_delegation_stateid(inode, stateid);
3666 if (!dp)
3667 goto out;
3668 status = check_stateid_generation(stateid, &dp->dl_stateid, nfsd4_has_session(cstate));
3669 if (status)
3670 goto out;
3671 renew_client(dp->dl_client);
3673 unhash_delegation(dp);
3674 out:
3675 nfs4_unlock_state();
3677 return status;
3682 * Lock owner state (byte-range locks)
3684 #define LOFF_OVERFLOW(start, len) ((u64)(len) > ~(u64)(start))
3685 #define LOCK_HASH_BITS 8
3686 #define LOCK_HASH_SIZE (1 << LOCK_HASH_BITS)
3687 #define LOCK_HASH_MASK (LOCK_HASH_SIZE - 1)
3689 static inline u64
3690 end_offset(u64 start, u64 len)
3692 u64 end;
3694 end = start + len;
3695 return end >= start ? end: NFS4_MAX_UINT64;
3698 /* last octet in a range */
3699 static inline u64
3700 last_byte_offset(u64 start, u64 len)
3702 u64 end;
3704 BUG_ON(!len);
3705 end = start + len;
3706 return end > start ? end - 1: NFS4_MAX_UINT64;
3709 static unsigned int lockownerid_hashval(u32 id)
3711 return id & LOCK_HASH_MASK;
3714 static inline unsigned int
3715 lock_ownerstr_hashval(struct inode *inode, u32 cl_id,
3716 struct xdr_netobj *ownername)
3718 return (file_hashval(inode) + cl_id
3719 + opaque_hashval(ownername->data, ownername->len))
3720 & LOCK_HASH_MASK;
3723 static struct list_head lock_ownerid_hashtbl[LOCK_HASH_SIZE];
3724 static struct list_head lock_ownerstr_hashtbl[LOCK_HASH_SIZE];
3725 static struct list_head lockstateid_hashtbl[STATEID_HASH_SIZE];
3727 static int
3728 same_stateid(stateid_t *id_one, stateid_t *id_two)
3730 if (id_one->si_stateownerid != id_two->si_stateownerid)
3731 return 0;
3732 return id_one->si_fileid == id_two->si_fileid;
3735 static struct nfs4_stateid *
3736 find_stateid(stateid_t *stid, int flags)
3738 struct nfs4_stateid *local;
3739 u32 st_id = stid->si_stateownerid;
3740 u32 f_id = stid->si_fileid;
3741 unsigned int hashval;
3743 dprintk("NFSD: find_stateid flags 0x%x\n",flags);
3744 if (flags & (LOCK_STATE | RD_STATE | WR_STATE)) {
3745 hashval = stateid_hashval(st_id, f_id);
3746 list_for_each_entry(local, &lockstateid_hashtbl[hashval], st_hash) {
3747 if ((local->st_stateid.si_stateownerid == st_id) &&
3748 (local->st_stateid.si_fileid == f_id))
3749 return local;
3753 if (flags & (OPEN_STATE | RD_STATE | WR_STATE)) {
3754 hashval = stateid_hashval(st_id, f_id);
3755 list_for_each_entry(local, &stateid_hashtbl[hashval], st_hash) {
3756 if ((local->st_stateid.si_stateownerid == st_id) &&
3757 (local->st_stateid.si_fileid == f_id))
3758 return local;
3761 return NULL;
3764 static struct nfs4_stateid *
3765 search_for_stateid(stateid_t *stid)
3767 struct nfs4_stateid *local;
3768 unsigned int hashval = stateid_hashval(stid->si_stateownerid, stid->si_fileid);
3770 list_for_each_entry(local, &lockstateid_hashtbl[hashval], st_hash) {
3771 if (same_stateid(&local->st_stateid, stid))
3772 return local;
3775 list_for_each_entry(local, &stateid_hashtbl[hashval], st_hash) {
3776 if (same_stateid(&local->st_stateid, stid))
3777 return local;
3779 return NULL;
3782 static struct nfs4_delegation *
3783 search_for_delegation(stateid_t *stid)
3785 struct nfs4_file *fp;
3786 struct nfs4_delegation *dp;
3787 struct list_head *pos;
3788 int i;
3790 for (i = 0; i < FILE_HASH_SIZE; i++) {
3791 list_for_each_entry(fp, &file_hashtbl[i], fi_hash) {
3792 list_for_each(pos, &fp->fi_delegations) {
3793 dp = list_entry(pos, struct nfs4_delegation, dl_perfile);
3794 if (same_stateid(&dp->dl_stateid, stid))
3795 return dp;
3799 return NULL;
3802 static struct nfs4_delegation *
3803 find_delegation_stateid(struct inode *ino, stateid_t *stid)
3805 struct nfs4_file *fp;
3806 struct nfs4_delegation *dl;
3808 dprintk("NFSD: %s: stateid=" STATEID_FMT "\n", __func__,
3809 STATEID_VAL(stid));
3811 fp = find_file(ino);
3812 if (!fp)
3813 return NULL;
3814 dl = find_delegation_file(fp, stid);
3815 put_nfs4_file(fp);
3816 return dl;
3820 * TODO: Linux file offsets are _signed_ 64-bit quantities, which means that
3821 * we can't properly handle lock requests that go beyond the (2^63 - 1)-th
3822 * byte, because of sign extension problems. Since NFSv4 calls for 64-bit
3823 * locking, this prevents us from being completely protocol-compliant. The
3824 * real solution to this problem is to start using unsigned file offsets in
3825 * the VFS, but this is a very deep change!
3827 static inline void
3828 nfs4_transform_lock_offset(struct file_lock *lock)
3830 if (lock->fl_start < 0)
3831 lock->fl_start = OFFSET_MAX;
3832 if (lock->fl_end < 0)
3833 lock->fl_end = OFFSET_MAX;
3836 /* Hack!: For now, we're defining this just so we can use a pointer to it
3837 * as a unique cookie to identify our (NFSv4's) posix locks. */
3838 static const struct lock_manager_operations nfsd_posix_mng_ops = {
3841 static inline void
3842 nfs4_set_lock_denied(struct file_lock *fl, struct nfsd4_lock_denied *deny)
3844 struct nfs4_stateowner *sop;
3846 if (fl->fl_lmops == &nfsd_posix_mng_ops) {
3847 sop = (struct nfs4_stateowner *) fl->fl_owner;
3848 kref_get(&sop->so_ref);
3849 deny->ld_sop = sop;
3850 deny->ld_clientid = sop->so_client->cl_clientid;
3851 } else {
3852 deny->ld_sop = NULL;
3853 deny->ld_clientid.cl_boot = 0;
3854 deny->ld_clientid.cl_id = 0;
3856 deny->ld_start = fl->fl_start;
3857 deny->ld_length = NFS4_MAX_UINT64;
3858 if (fl->fl_end != NFS4_MAX_UINT64)
3859 deny->ld_length = fl->fl_end - fl->fl_start + 1;
3860 deny->ld_type = NFS4_READ_LT;
3861 if (fl->fl_type != F_RDLCK)
3862 deny->ld_type = NFS4_WRITE_LT;
3865 static struct nfs4_stateowner *
3866 find_lockstateowner_str(struct inode *inode, clientid_t *clid,
3867 struct xdr_netobj *owner)
3869 unsigned int hashval = lock_ownerstr_hashval(inode, clid->cl_id, owner);
3870 struct nfs4_stateowner *op;
3872 list_for_each_entry(op, &lock_ownerstr_hashtbl[hashval], so_strhash) {
3873 if (same_owner_str(op, owner, clid))
3874 return op;
3876 return NULL;
3879 static void hash_lockowner(struct nfs4_stateowner *sop, unsigned int strhashval, struct nfs4_client *clp, struct nfs4_stateid *open_stp)
3881 unsigned int idhashval;
3883 idhashval = lockownerid_hashval(sop->so_id);
3884 list_add(&sop->so_idhash, &lock_ownerid_hashtbl[idhashval]);
3885 list_add(&sop->so_strhash, &lock_ownerstr_hashtbl[strhashval]);
3886 list_add(&sop->so_perstateid, &open_stp->st_lockowners);
3890 * Alloc a lock owner structure.
3891 * Called in nfsd4_lock - therefore, OPEN and OPEN_CONFIRM (if needed) has
3892 * occurred.
3894 * strhashval = lock_ownerstr_hashval
3897 static struct nfs4_stateowner *
3898 alloc_init_lock_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfs4_stateid *open_stp, struct nfsd4_lock *lock) {
3899 struct nfs4_stateowner *sop;
3901 sop = alloc_stateowner(&lock->lk_new_owner, clp);
3902 if (!sop)
3903 return NULL;
3904 INIT_LIST_HEAD(&sop->so_stateids);
3905 sop->so_is_open_owner = 0;
3906 /* It is the openowner seqid that will be incremented in encode in the
3907 * case of new lockowners; so increment the lock seqid manually: */
3908 sop->so_seqid = lock->lk_new_lock_seqid + 1;
3909 sop->so_confirmed = 1;
3910 hash_lockowner(sop, strhashval, clp, open_stp);
3911 return sop;
3914 static struct nfs4_stateid *
3915 alloc_init_lock_stateid(struct nfs4_stateowner *sop, struct nfs4_file *fp, struct nfs4_stateid *open_stp)
3917 struct nfs4_stateid *stp;
3918 unsigned int hashval = stateid_hashval(sop->so_id, fp->fi_id);
3920 stp = nfs4_alloc_stateid();
3921 if (stp == NULL)
3922 goto out;
3923 INIT_LIST_HEAD(&stp->st_hash);
3924 INIT_LIST_HEAD(&stp->st_perfile);
3925 INIT_LIST_HEAD(&stp->st_perstateowner);
3926 INIT_LIST_HEAD(&stp->st_lockowners); /* not used */
3927 list_add(&stp->st_hash, &lockstateid_hashtbl[hashval]);
3928 list_add(&stp->st_perfile, &fp->fi_stateids);
3929 list_add(&stp->st_perstateowner, &sop->so_stateids);
3930 stp->st_stateowner = sop;
3931 get_nfs4_file(fp);
3932 stp->st_file = fp;
3933 stp->st_stateid.si_boot = boot_time;
3934 stp->st_stateid.si_stateownerid = sop->so_id;
3935 stp->st_stateid.si_fileid = fp->fi_id;
3936 stp->st_stateid.si_generation = 0;
3937 stp->st_access_bmap = 0;
3938 stp->st_deny_bmap = open_stp->st_deny_bmap;
3939 stp->st_openstp = open_stp;
3941 out:
3942 return stp;
3945 static int
3946 check_lock_length(u64 offset, u64 length)
3948 return ((length == 0) || ((length != NFS4_MAX_UINT64) &&
3949 LOFF_OVERFLOW(offset, length)));
3952 static void get_lock_access(struct nfs4_stateid *lock_stp, u32 access)
3954 struct nfs4_file *fp = lock_stp->st_file;
3955 int oflag = nfs4_access_to_omode(access);
3957 if (test_bit(access, &lock_stp->st_access_bmap))
3958 return;
3959 nfs4_file_get_access(fp, oflag);
3960 __set_bit(access, &lock_stp->st_access_bmap);
3964 * LOCK operation
3966 __be32
3967 nfsd4_lock(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3968 struct nfsd4_lock *lock)
3970 struct nfs4_stateowner *open_sop = NULL;
3971 struct nfs4_stateowner *lock_sop = NULL;
3972 struct nfs4_stateid *lock_stp;
3973 struct nfs4_file *fp;
3974 struct file *filp = NULL;
3975 struct file_lock file_lock;
3976 struct file_lock conflock;
3977 __be32 status = 0;
3978 unsigned int strhashval;
3979 int lkflg;
3980 int err;
3982 dprintk("NFSD: nfsd4_lock: start=%Ld length=%Ld\n",
3983 (long long) lock->lk_offset,
3984 (long long) lock->lk_length);
3986 if (check_lock_length(lock->lk_offset, lock->lk_length))
3987 return nfserr_inval;
3989 if ((status = fh_verify(rqstp, &cstate->current_fh,
3990 S_IFREG, NFSD_MAY_LOCK))) {
3991 dprintk("NFSD: nfsd4_lock: permission denied!\n");
3992 return status;
3995 nfs4_lock_state();
3997 if (lock->lk_is_new) {
3999 * Client indicates that this is a new lockowner.
4000 * Use open owner and open stateid to create lock owner and
4001 * lock stateid.
4003 struct nfs4_stateid *open_stp = NULL;
4005 status = nfserr_stale_clientid;
4006 if (!nfsd4_has_session(cstate) &&
4007 STALE_CLIENTID(&lock->lk_new_clientid))
4008 goto out;
4010 /* validate and update open stateid and open seqid */
4011 status = nfs4_preprocess_seqid_op(cstate,
4012 lock->lk_new_open_seqid,
4013 &lock->lk_new_open_stateid,
4014 OPEN_STATE,
4015 &lock->lk_replay_owner, &open_stp);
4016 if (status)
4017 goto out;
4018 status = nfserr_bad_stateid;
4019 open_sop = lock->lk_replay_owner;
4020 if (!nfsd4_has_session(cstate) &&
4021 !same_clid(&open_sop->so_client->cl_clientid,
4022 &lock->v.new.clientid))
4023 goto out;
4024 /* create lockowner and lock stateid */
4025 fp = open_stp->st_file;
4026 strhashval = lock_ownerstr_hashval(fp->fi_inode,
4027 open_sop->so_client->cl_clientid.cl_id,
4028 &lock->v.new.owner);
4029 /* XXX: Do we need to check for duplicate stateowners on
4030 * the same file, or should they just be allowed (and
4031 * create new stateids)? */
4032 status = nfserr_jukebox;
4033 lock_sop = alloc_init_lock_stateowner(strhashval,
4034 open_sop->so_client, open_stp, lock);
4035 if (lock_sop == NULL)
4036 goto out;
4037 lock_stp = alloc_init_lock_stateid(lock_sop, fp, open_stp);
4038 if (lock_stp == NULL)
4039 goto out;
4040 } else {
4041 /* lock (lock owner + lock stateid) already exists */
4042 status = nfs4_preprocess_seqid_op(cstate,
4043 lock->lk_old_lock_seqid,
4044 &lock->lk_old_lock_stateid,
4045 LOCK_STATE,
4046 &lock->lk_replay_owner, &lock_stp);
4047 if (status)
4048 goto out;
4049 lock_sop = lock->lk_replay_owner;
4050 fp = lock_stp->st_file;
4052 /* lock->lk_replay_owner and lock_stp have been created or found */
4054 lkflg = setlkflg(lock->lk_type);
4055 status = nfs4_check_openmode(lock_stp, lkflg);
4056 if (status)
4057 goto out;
4059 status = nfserr_grace;
4060 if (locks_in_grace() && !lock->lk_reclaim)
4061 goto out;
4062 status = nfserr_no_grace;
4063 if (!locks_in_grace() && lock->lk_reclaim)
4064 goto out;
4066 locks_init_lock(&file_lock);
4067 switch (lock->lk_type) {
4068 case NFS4_READ_LT:
4069 case NFS4_READW_LT:
4070 filp = find_readable_file(lock_stp->st_file);
4071 if (filp)
4072 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_READ);
4073 file_lock.fl_type = F_RDLCK;
4074 break;
4075 case NFS4_WRITE_LT:
4076 case NFS4_WRITEW_LT:
4077 filp = find_writeable_file(lock_stp->st_file);
4078 if (filp)
4079 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_WRITE);
4080 file_lock.fl_type = F_WRLCK;
4081 break;
4082 default:
4083 status = nfserr_inval;
4084 goto out;
4086 if (!filp) {
4087 status = nfserr_openmode;
4088 goto out;
4090 file_lock.fl_owner = (fl_owner_t)lock_sop;
4091 file_lock.fl_pid = current->tgid;
4092 file_lock.fl_file = filp;
4093 file_lock.fl_flags = FL_POSIX;
4094 file_lock.fl_lmops = &nfsd_posix_mng_ops;
4096 file_lock.fl_start = lock->lk_offset;
4097 file_lock.fl_end = last_byte_offset(lock->lk_offset, lock->lk_length);
4098 nfs4_transform_lock_offset(&file_lock);
4101 * Try to lock the file in the VFS.
4102 * Note: locks.c uses the BKL to protect the inode's lock list.
4105 err = vfs_lock_file(filp, F_SETLK, &file_lock, &conflock);
4106 switch (-err) {
4107 case 0: /* success! */
4108 update_stateid(&lock_stp->st_stateid);
4109 memcpy(&lock->lk_resp_stateid, &lock_stp->st_stateid,
4110 sizeof(stateid_t));
4111 status = 0;
4112 break;
4113 case (EAGAIN): /* conflock holds conflicting lock */
4114 status = nfserr_denied;
4115 dprintk("NFSD: nfsd4_lock: conflicting lock found!\n");
4116 nfs4_set_lock_denied(&conflock, &lock->lk_denied);
4117 break;
4118 case (EDEADLK):
4119 status = nfserr_deadlock;
4120 break;
4121 default:
4122 dprintk("NFSD: nfsd4_lock: vfs_lock_file() failed! status %d\n",err);
4123 status = nfserrno(err);
4124 break;
4126 out:
4127 if (status && lock->lk_is_new && lock_sop)
4128 release_lockowner(lock_sop);
4129 if (lock->lk_replay_owner) {
4130 nfs4_get_stateowner(lock->lk_replay_owner);
4131 cstate->replay_owner = lock->lk_replay_owner;
4133 nfs4_unlock_state();
4134 return status;
4138 * The NFSv4 spec allows a client to do a LOCKT without holding an OPEN,
4139 * so we do a temporary open here just to get an open file to pass to
4140 * vfs_test_lock. (Arguably perhaps test_lock should be done with an
4141 * inode operation.)
4143 static int nfsd_test_lock(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file_lock *lock)
4145 struct file *file;
4146 int err;
4148 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file);
4149 if (err)
4150 return err;
4151 err = vfs_test_lock(file, lock);
4152 nfsd_close(file);
4153 return err;
4157 * LOCKT operation
4159 __be32
4160 nfsd4_lockt(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4161 struct nfsd4_lockt *lockt)
4163 struct inode *inode;
4164 struct file_lock file_lock;
4165 int error;
4166 __be32 status;
4168 if (locks_in_grace())
4169 return nfserr_grace;
4171 if (check_lock_length(lockt->lt_offset, lockt->lt_length))
4172 return nfserr_inval;
4174 lockt->lt_stateowner = NULL;
4175 nfs4_lock_state();
4177 status = nfserr_stale_clientid;
4178 if (!nfsd4_has_session(cstate) && STALE_CLIENTID(&lockt->lt_clientid))
4179 goto out;
4181 if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
4182 goto out;
4184 inode = cstate->current_fh.fh_dentry->d_inode;
4185 locks_init_lock(&file_lock);
4186 switch (lockt->lt_type) {
4187 case NFS4_READ_LT:
4188 case NFS4_READW_LT:
4189 file_lock.fl_type = F_RDLCK;
4190 break;
4191 case NFS4_WRITE_LT:
4192 case NFS4_WRITEW_LT:
4193 file_lock.fl_type = F_WRLCK;
4194 break;
4195 default:
4196 dprintk("NFSD: nfs4_lockt: bad lock type!\n");
4197 status = nfserr_inval;
4198 goto out;
4201 lockt->lt_stateowner = find_lockstateowner_str(inode,
4202 &lockt->lt_clientid, &lockt->lt_owner);
4203 if (lockt->lt_stateowner)
4204 file_lock.fl_owner = (fl_owner_t)lockt->lt_stateowner;
4205 file_lock.fl_pid = current->tgid;
4206 file_lock.fl_flags = FL_POSIX;
4208 file_lock.fl_start = lockt->lt_offset;
4209 file_lock.fl_end = last_byte_offset(lockt->lt_offset, lockt->lt_length);
4211 nfs4_transform_lock_offset(&file_lock);
4213 status = nfs_ok;
4214 error = nfsd_test_lock(rqstp, &cstate->current_fh, &file_lock);
4215 if (error) {
4216 status = nfserrno(error);
4217 goto out;
4219 if (file_lock.fl_type != F_UNLCK) {
4220 status = nfserr_denied;
4221 nfs4_set_lock_denied(&file_lock, &lockt->lt_denied);
4223 out:
4224 nfs4_unlock_state();
4225 return status;
4228 __be32
4229 nfsd4_locku(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4230 struct nfsd4_locku *locku)
4232 struct nfs4_stateid *stp;
4233 struct file *filp = NULL;
4234 struct file_lock file_lock;
4235 __be32 status;
4236 int err;
4238 dprintk("NFSD: nfsd4_locku: start=%Ld length=%Ld\n",
4239 (long long) locku->lu_offset,
4240 (long long) locku->lu_length);
4242 if (check_lock_length(locku->lu_offset, locku->lu_length))
4243 return nfserr_inval;
4245 nfs4_lock_state();
4247 if ((status = nfs4_preprocess_seqid_op(cstate,
4248 locku->lu_seqid,
4249 &locku->lu_stateid,
4250 LOCK_STATE,
4251 &locku->lu_stateowner, &stp)))
4252 goto out;
4254 filp = find_any_file(stp->st_file);
4255 if (!filp) {
4256 status = nfserr_lock_range;
4257 goto out;
4259 BUG_ON(!filp);
4260 locks_init_lock(&file_lock);
4261 file_lock.fl_type = F_UNLCK;
4262 file_lock.fl_owner = (fl_owner_t) locku->lu_stateowner;
4263 file_lock.fl_pid = current->tgid;
4264 file_lock.fl_file = filp;
4265 file_lock.fl_flags = FL_POSIX;
4266 file_lock.fl_lmops = &nfsd_posix_mng_ops;
4267 file_lock.fl_start = locku->lu_offset;
4269 file_lock.fl_end = last_byte_offset(locku->lu_offset, locku->lu_length);
4270 nfs4_transform_lock_offset(&file_lock);
4273 * Try to unlock the file in the VFS.
4275 err = vfs_lock_file(filp, F_SETLK, &file_lock, NULL);
4276 if (err) {
4277 dprintk("NFSD: nfs4_locku: vfs_lock_file failed!\n");
4278 goto out_nfserr;
4281 * OK, unlock succeeded; the only thing left to do is update the stateid.
4283 update_stateid(&stp->st_stateid);
4284 memcpy(&locku->lu_stateid, &stp->st_stateid, sizeof(stateid_t));
4286 out:
4287 if (locku->lu_stateowner) {
4288 nfs4_get_stateowner(locku->lu_stateowner);
4289 cstate->replay_owner = locku->lu_stateowner;
4291 nfs4_unlock_state();
4292 return status;
4294 out_nfserr:
4295 status = nfserrno(err);
4296 goto out;
4300 * returns
4301 * 1: locks held by lockowner
4302 * 0: no locks held by lockowner
4304 static int
4305 check_for_locks(struct nfs4_file *filp, struct nfs4_stateowner *lowner)
4307 struct file_lock **flpp;
4308 struct inode *inode = filp->fi_inode;
4309 int status = 0;
4311 lock_flocks();
4312 for (flpp = &inode->i_flock; *flpp != NULL; flpp = &(*flpp)->fl_next) {
4313 if ((*flpp)->fl_owner == (fl_owner_t)lowner) {
4314 status = 1;
4315 goto out;
4318 out:
4319 unlock_flocks();
4320 return status;
4323 __be32
4324 nfsd4_release_lockowner(struct svc_rqst *rqstp,
4325 struct nfsd4_compound_state *cstate,
4326 struct nfsd4_release_lockowner *rlockowner)
4328 clientid_t *clid = &rlockowner->rl_clientid;
4329 struct nfs4_stateowner *sop;
4330 struct nfs4_stateid *stp;
4331 struct xdr_netobj *owner = &rlockowner->rl_owner;
4332 struct list_head matches;
4333 int i;
4334 __be32 status;
4336 dprintk("nfsd4_release_lockowner clientid: (%08x/%08x):\n",
4337 clid->cl_boot, clid->cl_id);
4339 /* XXX check for lease expiration */
4341 status = nfserr_stale_clientid;
4342 if (STALE_CLIENTID(clid))
4343 return status;
4345 nfs4_lock_state();
4347 status = nfserr_locks_held;
4348 /* XXX: we're doing a linear search through all the lockowners.
4349 * Yipes! For now we'll just hope clients aren't really using
4350 * release_lockowner much, but eventually we have to fix these
4351 * data structures. */
4352 INIT_LIST_HEAD(&matches);
4353 for (i = 0; i < LOCK_HASH_SIZE; i++) {
4354 list_for_each_entry(sop, &lock_ownerid_hashtbl[i], so_idhash) {
4355 if (!same_owner_str(sop, owner, clid))
4356 continue;
4357 list_for_each_entry(stp, &sop->so_stateids,
4358 st_perstateowner) {
4359 if (check_for_locks(stp->st_file, sop))
4360 goto out;
4361 /* Note: so_perclient unused for lockowners,
4362 * so it's OK to fool with here. */
4363 list_add(&sop->so_perclient, &matches);
4367 /* Clients probably won't expect us to return with some (but not all)
4368 * of the lockowner state released; so don't release any until all
4369 * have been checked. */
4370 status = nfs_ok;
4371 while (!list_empty(&matches)) {
4372 sop = list_entry(matches.next, struct nfs4_stateowner,
4373 so_perclient);
4374 /* unhash_stateowner deletes so_perclient only
4375 * for openowners. */
4376 list_del(&sop->so_perclient);
4377 release_lockowner(sop);
4379 out:
4380 nfs4_unlock_state();
4381 return status;
4384 static inline struct nfs4_client_reclaim *
4385 alloc_reclaim(void)
4387 return kmalloc(sizeof(struct nfs4_client_reclaim), GFP_KERNEL);
4391 nfs4_has_reclaimed_state(const char *name, bool use_exchange_id)
4393 unsigned int strhashval = clientstr_hashval(name);
4394 struct nfs4_client *clp;
4396 clp = find_confirmed_client_by_str(name, strhashval);
4397 return clp ? 1 : 0;
4401 * failure => all reset bets are off, nfserr_no_grace...
4404 nfs4_client_to_reclaim(const char *name)
4406 unsigned int strhashval;
4407 struct nfs4_client_reclaim *crp = NULL;
4409 dprintk("NFSD nfs4_client_to_reclaim NAME: %.*s\n", HEXDIR_LEN, name);
4410 crp = alloc_reclaim();
4411 if (!crp)
4412 return 0;
4413 strhashval = clientstr_hashval(name);
4414 INIT_LIST_HEAD(&crp->cr_strhash);
4415 list_add(&crp->cr_strhash, &reclaim_str_hashtbl[strhashval]);
4416 memcpy(crp->cr_recdir, name, HEXDIR_LEN);
4417 reclaim_str_hashtbl_size++;
4418 return 1;
4421 static void
4422 nfs4_release_reclaim(void)
4424 struct nfs4_client_reclaim *crp = NULL;
4425 int i;
4427 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
4428 while (!list_empty(&reclaim_str_hashtbl[i])) {
4429 crp = list_entry(reclaim_str_hashtbl[i].next,
4430 struct nfs4_client_reclaim, cr_strhash);
4431 list_del(&crp->cr_strhash);
4432 kfree(crp);
4433 reclaim_str_hashtbl_size--;
4436 BUG_ON(reclaim_str_hashtbl_size);
4440 * called from OPEN, CLAIM_PREVIOUS with a new clientid. */
4441 static struct nfs4_client_reclaim *
4442 nfs4_find_reclaim_client(clientid_t *clid)
4444 unsigned int strhashval;
4445 struct nfs4_client *clp;
4446 struct nfs4_client_reclaim *crp = NULL;
4449 /* find clientid in conf_id_hashtbl */
4450 clp = find_confirmed_client(clid);
4451 if (clp == NULL)
4452 return NULL;
4454 dprintk("NFSD: nfs4_find_reclaim_client for %.*s with recdir %s\n",
4455 clp->cl_name.len, clp->cl_name.data,
4456 clp->cl_recdir);
4458 /* find clp->cl_name in reclaim_str_hashtbl */
4459 strhashval = clientstr_hashval(clp->cl_recdir);
4460 list_for_each_entry(crp, &reclaim_str_hashtbl[strhashval], cr_strhash) {
4461 if (same_name(crp->cr_recdir, clp->cl_recdir)) {
4462 return crp;
4465 return NULL;
4469 * Called from OPEN. Look for clientid in reclaim list.
4471 __be32
4472 nfs4_check_open_reclaim(clientid_t *clid)
4474 return nfs4_find_reclaim_client(clid) ? nfs_ok : nfserr_reclaim_bad;
4477 /* initialization to perform at module load time: */
4480 nfs4_state_init(void)
4482 int i, status;
4484 status = nfsd4_init_slabs();
4485 if (status)
4486 return status;
4487 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
4488 INIT_LIST_HEAD(&conf_id_hashtbl[i]);
4489 INIT_LIST_HEAD(&conf_str_hashtbl[i]);
4490 INIT_LIST_HEAD(&unconf_str_hashtbl[i]);
4491 INIT_LIST_HEAD(&unconf_id_hashtbl[i]);
4492 INIT_LIST_HEAD(&reclaim_str_hashtbl[i]);
4494 for (i = 0; i < SESSION_HASH_SIZE; i++)
4495 INIT_LIST_HEAD(&sessionid_hashtbl[i]);
4496 for (i = 0; i < FILE_HASH_SIZE; i++) {
4497 INIT_LIST_HEAD(&file_hashtbl[i]);
4499 for (i = 0; i < OPEN_OWNER_HASH_SIZE; i++) {
4500 INIT_LIST_HEAD(&open_ownerstr_hashtbl[i]);
4501 INIT_LIST_HEAD(&open_ownerid_hashtbl[i]);
4503 for (i = 0; i < STATEID_HASH_SIZE; i++) {
4504 INIT_LIST_HEAD(&stateid_hashtbl[i]);
4505 INIT_LIST_HEAD(&lockstateid_hashtbl[i]);
4507 for (i = 0; i < LOCK_HASH_SIZE; i++) {
4508 INIT_LIST_HEAD(&lock_ownerid_hashtbl[i]);
4509 INIT_LIST_HEAD(&lock_ownerstr_hashtbl[i]);
4511 memset(&onestateid, ~0, sizeof(stateid_t));
4512 INIT_LIST_HEAD(&close_lru);
4513 INIT_LIST_HEAD(&client_lru);
4514 INIT_LIST_HEAD(&del_recall_lru);
4515 reclaim_str_hashtbl_size = 0;
4516 return 0;
4519 static void
4520 nfsd4_load_reboot_recovery_data(void)
4522 int status;
4524 nfs4_lock_state();
4525 nfsd4_init_recdir();
4526 status = nfsd4_recdir_load();
4527 nfs4_unlock_state();
4528 if (status)
4529 printk("NFSD: Failure reading reboot recovery data\n");
4533 * Since the lifetime of a delegation isn't limited to that of an open, a
4534 * client may quite reasonably hang on to a delegation as long as it has
4535 * the inode cached. This becomes an obvious problem the first time a
4536 * client's inode cache approaches the size of the server's total memory.
4538 * For now we avoid this problem by imposing a hard limit on the number
4539 * of delegations, which varies according to the server's memory size.
4541 static void
4542 set_max_delegations(void)
4545 * Allow at most 4 delegations per megabyte of RAM. Quick
4546 * estimates suggest that in the worst case (where every delegation
4547 * is for a different inode), a delegation could take about 1.5K,
4548 * giving a worst case usage of about 6% of memory.
4550 max_delegations = nr_free_buffer_pages() >> (20 - 2 - PAGE_SHIFT);
4553 /* initialization to perform when the nfsd service is started: */
4555 static int
4556 __nfs4_state_start(void)
4558 int ret;
4560 boot_time = get_seconds();
4561 locks_start_grace(&nfsd4_manager);
4562 printk(KERN_INFO "NFSD: starting %ld-second grace period\n",
4563 nfsd4_grace);
4564 ret = set_callback_cred();
4565 if (ret)
4566 return -ENOMEM;
4567 laundry_wq = create_singlethread_workqueue("nfsd4");
4568 if (laundry_wq == NULL)
4569 return -ENOMEM;
4570 ret = nfsd4_create_callback_queue();
4571 if (ret)
4572 goto out_free_laundry;
4573 queue_delayed_work(laundry_wq, &laundromat_work, nfsd4_grace * HZ);
4574 set_max_delegations();
4575 return 0;
4576 out_free_laundry:
4577 destroy_workqueue(laundry_wq);
4578 return ret;
4582 nfs4_state_start(void)
4584 nfsd4_load_reboot_recovery_data();
4585 return __nfs4_state_start();
4588 static void
4589 __nfs4_state_shutdown(void)
4591 int i;
4592 struct nfs4_client *clp = NULL;
4593 struct nfs4_delegation *dp = NULL;
4594 struct list_head *pos, *next, reaplist;
4596 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
4597 while (!list_empty(&conf_id_hashtbl[i])) {
4598 clp = list_entry(conf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
4599 expire_client(clp);
4601 while (!list_empty(&unconf_str_hashtbl[i])) {
4602 clp = list_entry(unconf_str_hashtbl[i].next, struct nfs4_client, cl_strhash);
4603 expire_client(clp);
4606 INIT_LIST_HEAD(&reaplist);
4607 spin_lock(&recall_lock);
4608 list_for_each_safe(pos, next, &del_recall_lru) {
4609 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
4610 list_move(&dp->dl_recall_lru, &reaplist);
4612 spin_unlock(&recall_lock);
4613 list_for_each_safe(pos, next, &reaplist) {
4614 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
4615 list_del_init(&dp->dl_recall_lru);
4616 unhash_delegation(dp);
4619 nfsd4_shutdown_recdir();
4622 void
4623 nfs4_state_shutdown(void)
4625 cancel_delayed_work_sync(&laundromat_work);
4626 destroy_workqueue(laundry_wq);
4627 locks_end_grace(&nfsd4_manager);
4628 nfs4_lock_state();
4629 nfs4_release_reclaim();
4630 __nfs4_state_shutdown();
4631 nfs4_unlock_state();
4632 nfsd4_destroy_callback_queue();