1 /* 2 * Copyright (c) 2001 The Regents of the University of Michigan. 3 * All rights reserved. 4 * 5 * Kendrick Smith <kmsmith@umich.edu> 6 * Andy Adamson <kandros@umich.edu> 7 * 8 * Redistribution and use in source and binary forms, with or without 9 * modification, are permitted provided that the following conditions 10 * are met: 11 * 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. 20 * 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. 32 * 33 */ 34 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/sunrpc/svcauth_gss.h> 41 #include <linux/sunrpc/clnt.h> 42 #include "xdr4.h" 43 #include "vfs.h" 44 45 #define NFSDDBG_FACILITY NFSDDBG_PROC 46 47 /* Globals */ 48 time_t nfsd4_lease = 90; /* default lease time */ 49 time_t nfsd4_grace = 90; 50 static time_t boot_time; 51 static u32 current_ownerid = 1; 52 static u32 current_fileid = 1; 53 static u32 current_delegid = 1; 54 static stateid_t zerostateid; /* bits all 0 */ 55 static stateid_t onestateid; /* bits all 1 */ 56 static u64 current_sessionid = 1; 57 58 #define ZERO_STATEID(stateid) (!memcmp((stateid), &zerostateid, sizeof(stateid_t))) 59 #define ONE_STATEID(stateid) (!memcmp((stateid), &onestateid, sizeof(stateid_t))) 60 61 /* forward declarations */ 62 static struct nfs4_stateid * find_stateid(stateid_t *stid, int flags); 63 static struct nfs4_delegation * find_delegation_stateid(struct inode *ino, stateid_t *stid); 64 static char user_recovery_dirname[PATH_MAX] = "/var/lib/nfs/v4recovery"; 65 static void nfs4_set_recdir(char *recdir); 66 67 /* Locking: */ 68 69 /* Currently used for almost all code touching nfsv4 state: */ 70 static DEFINE_MUTEX(client_mutex); 71 72 /* 73 * Currently used for the del_recall_lru and file hash table. In an 74 * effort to decrease the scope of the client_mutex, this spinlock may 75 * eventually cover more: 76 */ 77 static DEFINE_SPINLOCK(recall_lock); 78 79 static struct kmem_cache *stateowner_slab = NULL; 80 static struct kmem_cache *file_slab = NULL; 81 static struct kmem_cache *stateid_slab = NULL; 82 static struct kmem_cache *deleg_slab = NULL; 83 84 void 85 nfs4_lock_state(void) 86 { 87 mutex_lock(&client_mutex); 88 } 89 90 void 91 nfs4_unlock_state(void) 92 { 93 mutex_unlock(&client_mutex); 94 } 95 96 static inline u32 97 opaque_hashval(const void *ptr, int nbytes) 98 { 99 unsigned char *cptr = (unsigned char *) ptr; 100 101 u32 x = 0; 102 while (nbytes--) { 103 x *= 37; 104 x += *cptr++; 105 } 106 return x; 107 } 108 109 static struct list_head del_recall_lru; 110 111 static inline void 112 put_nfs4_file(struct nfs4_file *fi) 113 { 114 if (atomic_dec_and_lock(&fi->fi_ref, &recall_lock)) { 115 list_del(&fi->fi_hash); 116 spin_unlock(&recall_lock); 117 iput(fi->fi_inode); 118 kmem_cache_free(file_slab, fi); 119 } 120 } 121 122 static inline void 123 get_nfs4_file(struct nfs4_file *fi) 124 { 125 atomic_inc(&fi->fi_ref); 126 } 127 128 static int num_delegations; 129 unsigned int max_delegations; 130 131 /* 132 * Open owner state (share locks) 133 */ 134 135 /* hash tables for nfs4_stateowner */ 136 #define OWNER_HASH_BITS 8 137 #define OWNER_HASH_SIZE (1 << OWNER_HASH_BITS) 138 #define OWNER_HASH_MASK (OWNER_HASH_SIZE - 1) 139 140 #define ownerid_hashval(id) \ 141 ((id) & OWNER_HASH_MASK) 142 #define ownerstr_hashval(clientid, ownername) \ 143 (((clientid) + opaque_hashval((ownername.data), (ownername.len))) & OWNER_HASH_MASK) 144 145 static struct list_head ownerid_hashtbl[OWNER_HASH_SIZE]; 146 static struct list_head ownerstr_hashtbl[OWNER_HASH_SIZE]; 147 148 /* hash table for nfs4_file */ 149 #define FILE_HASH_BITS 8 150 #define FILE_HASH_SIZE (1 << FILE_HASH_BITS) 151 152 /* hash table for (open)nfs4_stateid */ 153 #define STATEID_HASH_BITS 10 154 #define STATEID_HASH_SIZE (1 << STATEID_HASH_BITS) 155 #define STATEID_HASH_MASK (STATEID_HASH_SIZE - 1) 156 157 #define file_hashval(x) \ 158 hash_ptr(x, FILE_HASH_BITS) 159 #define stateid_hashval(owner_id, file_id) \ 160 (((owner_id) + (file_id)) & STATEID_HASH_MASK) 161 162 static struct list_head file_hashtbl[FILE_HASH_SIZE]; 163 static struct list_head stateid_hashtbl[STATEID_HASH_SIZE]; 164 165 static void __nfs4_file_get_access(struct nfs4_file *fp, int oflag) 166 { 167 BUG_ON(!(fp->fi_fds[oflag] || fp->fi_fds[O_RDWR])); 168 atomic_inc(&fp->fi_access[oflag]); 169 } 170 171 static void nfs4_file_get_access(struct nfs4_file *fp, int oflag) 172 { 173 if (oflag == O_RDWR) { 174 __nfs4_file_get_access(fp, O_RDONLY); 175 __nfs4_file_get_access(fp, O_WRONLY); 176 } else 177 __nfs4_file_get_access(fp, oflag); 178 } 179 180 static void nfs4_file_put_fd(struct nfs4_file *fp, int oflag) 181 { 182 if (fp->fi_fds[oflag]) { 183 fput(fp->fi_fds[oflag]); 184 fp->fi_fds[oflag] = NULL; 185 } 186 } 187 188 static void __nfs4_file_put_access(struct nfs4_file *fp, int oflag) 189 { 190 if (atomic_dec_and_test(&fp->fi_access[oflag])) { 191 nfs4_file_put_fd(fp, O_RDWR); 192 nfs4_file_put_fd(fp, oflag); 193 } 194 } 195 196 static void nfs4_file_put_access(struct nfs4_file *fp, int oflag) 197 { 198 if (oflag == O_RDWR) { 199 __nfs4_file_put_access(fp, O_RDONLY); 200 __nfs4_file_put_access(fp, O_WRONLY); 201 } else 202 __nfs4_file_put_access(fp, oflag); 203 } 204 205 static struct nfs4_delegation * 206 alloc_init_deleg(struct nfs4_client *clp, struct nfs4_stateid *stp, struct svc_fh *current_fh, u32 type) 207 { 208 struct nfs4_delegation *dp; 209 struct nfs4_file *fp = stp->st_file; 210 211 dprintk("NFSD alloc_init_deleg\n"); 212 /* 213 * Major work on the lease subsystem (for example, to support 214 * calbacks on stat) will be required before we can support 215 * write delegations properly. 216 */ 217 if (type != NFS4_OPEN_DELEGATE_READ) 218 return NULL; 219 if (fp->fi_had_conflict) 220 return NULL; 221 if (num_delegations > max_delegations) 222 return NULL; 223 dp = kmem_cache_alloc(deleg_slab, GFP_KERNEL); 224 if (dp == NULL) 225 return dp; 226 num_delegations++; 227 INIT_LIST_HEAD(&dp->dl_perfile); 228 INIT_LIST_HEAD(&dp->dl_perclnt); 229 INIT_LIST_HEAD(&dp->dl_recall_lru); 230 dp->dl_client = clp; 231 get_nfs4_file(fp); 232 dp->dl_file = fp; 233 dp->dl_type = type; 234 dp->dl_stateid.si_boot = boot_time; 235 dp->dl_stateid.si_stateownerid = current_delegid++; 236 dp->dl_stateid.si_fileid = 0; 237 dp->dl_stateid.si_generation = 0; 238 fh_copy_shallow(&dp->dl_fh, ¤t_fh->fh_handle); 239 dp->dl_time = 0; 240 atomic_set(&dp->dl_count, 1); 241 INIT_WORK(&dp->dl_recall.cb_work, nfsd4_do_callback_rpc); 242 return dp; 243 } 244 245 void 246 nfs4_put_delegation(struct nfs4_delegation *dp) 247 { 248 if (atomic_dec_and_test(&dp->dl_count)) { 249 dprintk("NFSD: freeing dp %p\n",dp); 250 put_nfs4_file(dp->dl_file); 251 kmem_cache_free(deleg_slab, dp); 252 num_delegations--; 253 } 254 } 255 256 static void nfs4_put_deleg_lease(struct nfs4_file *fp) 257 { 258 if (atomic_dec_and_test(&fp->fi_delegees)) { 259 vfs_setlease(fp->fi_deleg_file, F_UNLCK, &fp->fi_lease); 260 fp->fi_lease = NULL; 261 fp->fi_deleg_file = NULL; 262 } 263 } 264 265 /* Called under the state lock. */ 266 static void 267 unhash_delegation(struct nfs4_delegation *dp) 268 { 269 list_del_init(&dp->dl_perclnt); 270 spin_lock(&recall_lock); 271 list_del_init(&dp->dl_perfile); 272 list_del_init(&dp->dl_recall_lru); 273 spin_unlock(&recall_lock); 274 nfs4_put_deleg_lease(dp->dl_file); 275 nfs4_put_delegation(dp); 276 } 277 278 /* 279 * SETCLIENTID state 280 */ 281 282 /* client_lock protects the client lru list and session hash table */ 283 static DEFINE_SPINLOCK(client_lock); 284 285 /* Hash tables for nfs4_clientid state */ 286 #define CLIENT_HASH_BITS 4 287 #define CLIENT_HASH_SIZE (1 << CLIENT_HASH_BITS) 288 #define CLIENT_HASH_MASK (CLIENT_HASH_SIZE - 1) 289 290 #define clientid_hashval(id) \ 291 ((id) & CLIENT_HASH_MASK) 292 #define clientstr_hashval(name) \ 293 (opaque_hashval((name), 8) & CLIENT_HASH_MASK) 294 /* 295 * reclaim_str_hashtbl[] holds known client info from previous reset/reboot 296 * used in reboot/reset lease grace period processing 297 * 298 * conf_id_hashtbl[], and conf_str_hashtbl[] hold confirmed 299 * setclientid_confirmed info. 300 * 301 * unconf_str_hastbl[] and unconf_id_hashtbl[] hold unconfirmed 302 * setclientid info. 303 * 304 * client_lru holds client queue ordered by nfs4_client.cl_time 305 * for lease renewal. 306 * 307 * close_lru holds (open) stateowner queue ordered by nfs4_stateowner.so_time 308 * for last close replay. 309 */ 310 static struct list_head reclaim_str_hashtbl[CLIENT_HASH_SIZE]; 311 static int reclaim_str_hashtbl_size = 0; 312 static struct list_head conf_id_hashtbl[CLIENT_HASH_SIZE]; 313 static struct list_head conf_str_hashtbl[CLIENT_HASH_SIZE]; 314 static struct list_head unconf_str_hashtbl[CLIENT_HASH_SIZE]; 315 static struct list_head unconf_id_hashtbl[CLIENT_HASH_SIZE]; 316 static struct list_head client_lru; 317 static struct list_head close_lru; 318 319 /* 320 * We store the NONE, READ, WRITE, and BOTH bits separately in the 321 * st_{access,deny}_bmap field of the stateid, in order to track not 322 * only what share bits are currently in force, but also what 323 * combinations of share bits previous opens have used. This allows us 324 * to enforce the recommendation of rfc 3530 14.2.19 that the server 325 * return an error if the client attempt to downgrade to a combination 326 * of share bits not explicable by closing some of its previous opens. 327 * 328 * XXX: This enforcement is actually incomplete, since we don't keep 329 * track of access/deny bit combinations; so, e.g., we allow: 330 * 331 * OPEN allow read, deny write 332 * OPEN allow both, deny none 333 * DOWNGRADE allow read, deny none 334 * 335 * which we should reject. 336 */ 337 static void 338 set_access(unsigned int *access, unsigned long bmap) { 339 int i; 340 341 *access = 0; 342 for (i = 1; i < 4; i++) { 343 if (test_bit(i, &bmap)) 344 *access |= i; 345 } 346 } 347 348 static void 349 set_deny(unsigned int *deny, unsigned long bmap) { 350 int i; 351 352 *deny = 0; 353 for (i = 0; i < 4; i++) { 354 if (test_bit(i, &bmap)) 355 *deny |= i ; 356 } 357 } 358 359 static int 360 test_share(struct nfs4_stateid *stp, struct nfsd4_open *open) { 361 unsigned int access, deny; 362 363 set_access(&access, stp->st_access_bmap); 364 set_deny(&deny, stp->st_deny_bmap); 365 if ((access & open->op_share_deny) || (deny & open->op_share_access)) 366 return 0; 367 return 1; 368 } 369 370 static int nfs4_access_to_omode(u32 access) 371 { 372 switch (access & NFS4_SHARE_ACCESS_BOTH) { 373 case NFS4_SHARE_ACCESS_READ: 374 return O_RDONLY; 375 case NFS4_SHARE_ACCESS_WRITE: 376 return O_WRONLY; 377 case NFS4_SHARE_ACCESS_BOTH: 378 return O_RDWR; 379 } 380 BUG(); 381 } 382 383 static int nfs4_access_bmap_to_omode(struct nfs4_stateid *stp) 384 { 385 unsigned int access; 386 387 set_access(&access, stp->st_access_bmap); 388 return nfs4_access_to_omode(access); 389 } 390 391 static void unhash_generic_stateid(struct nfs4_stateid *stp) 392 { 393 list_del(&stp->st_hash); 394 list_del(&stp->st_perfile); 395 list_del(&stp->st_perstateowner); 396 } 397 398 static void free_generic_stateid(struct nfs4_stateid *stp) 399 { 400 int oflag = nfs4_access_bmap_to_omode(stp); 401 402 nfs4_file_put_access(stp->st_file, oflag); 403 put_nfs4_file(stp->st_file); 404 kmem_cache_free(stateid_slab, stp); 405 } 406 407 static void release_lock_stateid(struct nfs4_stateid *stp) 408 { 409 struct file *file; 410 411 unhash_generic_stateid(stp); 412 file = find_any_file(stp->st_file); 413 if (file) 414 locks_remove_posix(file, (fl_owner_t)stp->st_stateowner); 415 free_generic_stateid(stp); 416 } 417 418 static void unhash_lockowner(struct nfs4_stateowner *sop) 419 { 420 struct nfs4_stateid *stp; 421 422 list_del(&sop->so_idhash); 423 list_del(&sop->so_strhash); 424 list_del(&sop->so_perstateid); 425 while (!list_empty(&sop->so_stateids)) { 426 stp = list_first_entry(&sop->so_stateids, 427 struct nfs4_stateid, st_perstateowner); 428 release_lock_stateid(stp); 429 } 430 } 431 432 static void release_lockowner(struct nfs4_stateowner *sop) 433 { 434 unhash_lockowner(sop); 435 nfs4_put_stateowner(sop); 436 } 437 438 static void 439 release_stateid_lockowners(struct nfs4_stateid *open_stp) 440 { 441 struct nfs4_stateowner *lock_sop; 442 443 while (!list_empty(&open_stp->st_lockowners)) { 444 lock_sop = list_entry(open_stp->st_lockowners.next, 445 struct nfs4_stateowner, so_perstateid); 446 /* list_del(&open_stp->st_lockowners); */ 447 BUG_ON(lock_sop->so_is_open_owner); 448 release_lockowner(lock_sop); 449 } 450 } 451 452 static void release_open_stateid(struct nfs4_stateid *stp) 453 { 454 unhash_generic_stateid(stp); 455 release_stateid_lockowners(stp); 456 free_generic_stateid(stp); 457 } 458 459 static void unhash_openowner(struct nfs4_stateowner *sop) 460 { 461 struct nfs4_stateid *stp; 462 463 list_del(&sop->so_idhash); 464 list_del(&sop->so_strhash); 465 list_del(&sop->so_perclient); 466 list_del(&sop->so_perstateid); /* XXX: necessary? */ 467 while (!list_empty(&sop->so_stateids)) { 468 stp = list_first_entry(&sop->so_stateids, 469 struct nfs4_stateid, st_perstateowner); 470 release_open_stateid(stp); 471 } 472 } 473 474 static void release_openowner(struct nfs4_stateowner *sop) 475 { 476 unhash_openowner(sop); 477 list_del(&sop->so_close_lru); 478 nfs4_put_stateowner(sop); 479 } 480 481 #define SESSION_HASH_SIZE 512 482 static struct list_head sessionid_hashtbl[SESSION_HASH_SIZE]; 483 484 static inline int 485 hash_sessionid(struct nfs4_sessionid *sessionid) 486 { 487 struct nfsd4_sessionid *sid = (struct nfsd4_sessionid *)sessionid; 488 489 return sid->sequence % SESSION_HASH_SIZE; 490 } 491 492 static inline void 493 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid) 494 { 495 u32 *ptr = (u32 *)(&sessionid->data[0]); 496 dprintk("%s: %u:%u:%u:%u\n", fn, ptr[0], ptr[1], ptr[2], ptr[3]); 497 } 498 499 static void 500 gen_sessionid(struct nfsd4_session *ses) 501 { 502 struct nfs4_client *clp = ses->se_client; 503 struct nfsd4_sessionid *sid; 504 505 sid = (struct nfsd4_sessionid *)ses->se_sessionid.data; 506 sid->clientid = clp->cl_clientid; 507 sid->sequence = current_sessionid++; 508 sid->reserved = 0; 509 } 510 511 /* 512 * The protocol defines ca_maxresponssize_cached to include the size of 513 * the rpc header, but all we need to cache is the data starting after 514 * the end of the initial SEQUENCE operation--the rest we regenerate 515 * each time. Therefore we can advertise a ca_maxresponssize_cached 516 * value that is the number of bytes in our cache plus a few additional 517 * bytes. In order to stay on the safe side, and not promise more than 518 * we can cache, those additional bytes must be the minimum possible: 24 519 * bytes of rpc header (xid through accept state, with AUTH_NULL 520 * verifier), 12 for the compound header (with zero-length tag), and 44 521 * for the SEQUENCE op response: 522 */ 523 #define NFSD_MIN_HDR_SEQ_SZ (24 + 12 + 44) 524 525 static void 526 free_session_slots(struct nfsd4_session *ses) 527 { 528 int i; 529 530 for (i = 0; i < ses->se_fchannel.maxreqs; i++) 531 kfree(ses->se_slots[i]); 532 } 533 534 /* 535 * We don't actually need to cache the rpc and session headers, so we 536 * can allocate a little less for each slot: 537 */ 538 static inline int slot_bytes(struct nfsd4_channel_attrs *ca) 539 { 540 return ca->maxresp_cached - NFSD_MIN_HDR_SEQ_SZ; 541 } 542 543 static int nfsd4_sanitize_slot_size(u32 size) 544 { 545 size -= NFSD_MIN_HDR_SEQ_SZ; /* We don't cache the rpc header */ 546 size = min_t(u32, size, NFSD_SLOT_CACHE_SIZE); 547 548 return size; 549 } 550 551 /* 552 * XXX: If we run out of reserved DRC memory we could (up to a point) 553 * re-negotiate active sessions and reduce their slot usage to make 554 * rooom for new connections. For now we just fail the create session. 555 */ 556 static int nfsd4_get_drc_mem(int slotsize, u32 num) 557 { 558 int avail; 559 560 num = min_t(u32, num, NFSD_MAX_SLOTS_PER_SESSION); 561 562 spin_lock(&nfsd_drc_lock); 563 avail = min_t(int, NFSD_MAX_MEM_PER_SESSION, 564 nfsd_drc_max_mem - nfsd_drc_mem_used); 565 num = min_t(int, num, avail / slotsize); 566 nfsd_drc_mem_used += num * slotsize; 567 spin_unlock(&nfsd_drc_lock); 568 569 return num; 570 } 571 572 static void nfsd4_put_drc_mem(int slotsize, int num) 573 { 574 spin_lock(&nfsd_drc_lock); 575 nfsd_drc_mem_used -= slotsize * num; 576 spin_unlock(&nfsd_drc_lock); 577 } 578 579 static struct nfsd4_session *alloc_session(int slotsize, int numslots) 580 { 581 struct nfsd4_session *new; 582 int mem, i; 583 584 BUILD_BUG_ON(NFSD_MAX_SLOTS_PER_SESSION * sizeof(struct nfsd4_slot *) 585 + sizeof(struct nfsd4_session) > PAGE_SIZE); 586 mem = numslots * sizeof(struct nfsd4_slot *); 587 588 new = kzalloc(sizeof(*new) + mem, GFP_KERNEL); 589 if (!new) 590 return NULL; 591 /* allocate each struct nfsd4_slot and data cache in one piece */ 592 for (i = 0; i < numslots; i++) { 593 mem = sizeof(struct nfsd4_slot) + slotsize; 594 new->se_slots[i] = kzalloc(mem, GFP_KERNEL); 595 if (!new->se_slots[i]) 596 goto out_free; 597 } 598 return new; 599 out_free: 600 while (i--) 601 kfree(new->se_slots[i]); 602 kfree(new); 603 return NULL; 604 } 605 606 static void init_forechannel_attrs(struct nfsd4_channel_attrs *new, struct nfsd4_channel_attrs *req, int numslots, int slotsize) 607 { 608 u32 maxrpc = nfsd_serv->sv_max_mesg; 609 610 new->maxreqs = numslots; 611 new->maxresp_cached = min_t(u32, req->maxresp_cached, 612 slotsize + NFSD_MIN_HDR_SEQ_SZ); 613 new->maxreq_sz = min_t(u32, req->maxreq_sz, maxrpc); 614 new->maxresp_sz = min_t(u32, req->maxresp_sz, maxrpc); 615 new->maxops = min_t(u32, req->maxops, NFSD_MAX_OPS_PER_COMPOUND); 616 } 617 618 static void free_conn(struct nfsd4_conn *c) 619 { 620 svc_xprt_put(c->cn_xprt); 621 kfree(c); 622 } 623 624 static void nfsd4_conn_lost(struct svc_xpt_user *u) 625 { 626 struct nfsd4_conn *c = container_of(u, struct nfsd4_conn, cn_xpt_user); 627 struct nfs4_client *clp = c->cn_session->se_client; 628 629 spin_lock(&clp->cl_lock); 630 if (!list_empty(&c->cn_persession)) { 631 list_del(&c->cn_persession); 632 free_conn(c); 633 } 634 spin_unlock(&clp->cl_lock); 635 nfsd4_probe_callback(clp); 636 } 637 638 static struct nfsd4_conn *alloc_conn(struct svc_rqst *rqstp, u32 flags) 639 { 640 struct nfsd4_conn *conn; 641 642 conn = kmalloc(sizeof(struct nfsd4_conn), GFP_KERNEL); 643 if (!conn) 644 return NULL; 645 svc_xprt_get(rqstp->rq_xprt); 646 conn->cn_xprt = rqstp->rq_xprt; 647 conn->cn_flags = flags; 648 INIT_LIST_HEAD(&conn->cn_xpt_user.list); 649 return conn; 650 } 651 652 static void __nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses) 653 { 654 conn->cn_session = ses; 655 list_add(&conn->cn_persession, &ses->se_conns); 656 } 657 658 static void nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses) 659 { 660 struct nfs4_client *clp = ses->se_client; 661 662 spin_lock(&clp->cl_lock); 663 __nfsd4_hash_conn(conn, ses); 664 spin_unlock(&clp->cl_lock); 665 } 666 667 static int nfsd4_register_conn(struct nfsd4_conn *conn) 668 { 669 conn->cn_xpt_user.callback = nfsd4_conn_lost; 670 return register_xpt_user(conn->cn_xprt, &conn->cn_xpt_user); 671 } 672 673 static __be32 nfsd4_new_conn(struct svc_rqst *rqstp, struct nfsd4_session *ses, u32 dir) 674 { 675 struct nfsd4_conn *conn; 676 int ret; 677 678 conn = alloc_conn(rqstp, dir); 679 if (!conn) 680 return nfserr_jukebox; 681 nfsd4_hash_conn(conn, ses); 682 ret = nfsd4_register_conn(conn); 683 if (ret) 684 /* oops; xprt is already down: */ 685 nfsd4_conn_lost(&conn->cn_xpt_user); 686 return nfs_ok; 687 } 688 689 static __be32 nfsd4_new_conn_from_crses(struct svc_rqst *rqstp, struct nfsd4_session *ses) 690 { 691 u32 dir = NFS4_CDFC4_FORE; 692 693 if (ses->se_flags & SESSION4_BACK_CHAN) 694 dir |= NFS4_CDFC4_BACK; 695 696 return nfsd4_new_conn(rqstp, ses, dir); 697 } 698 699 /* must be called under client_lock */ 700 static void nfsd4_del_conns(struct nfsd4_session *s) 701 { 702 struct nfs4_client *clp = s->se_client; 703 struct nfsd4_conn *c; 704 705 spin_lock(&clp->cl_lock); 706 while (!list_empty(&s->se_conns)) { 707 c = list_first_entry(&s->se_conns, struct nfsd4_conn, cn_persession); 708 list_del_init(&c->cn_persession); 709 spin_unlock(&clp->cl_lock); 710 711 unregister_xpt_user(c->cn_xprt, &c->cn_xpt_user); 712 free_conn(c); 713 714 spin_lock(&clp->cl_lock); 715 } 716 spin_unlock(&clp->cl_lock); 717 } 718 719 void free_session(struct kref *kref) 720 { 721 struct nfsd4_session *ses; 722 int mem; 723 724 ses = container_of(kref, struct nfsd4_session, se_ref); 725 nfsd4_del_conns(ses); 726 spin_lock(&nfsd_drc_lock); 727 mem = ses->se_fchannel.maxreqs * slot_bytes(&ses->se_fchannel); 728 nfsd_drc_mem_used -= mem; 729 spin_unlock(&nfsd_drc_lock); 730 free_session_slots(ses); 731 kfree(ses); 732 } 733 734 static struct nfsd4_session *alloc_init_session(struct svc_rqst *rqstp, struct nfs4_client *clp, struct nfsd4_create_session *cses) 735 { 736 struct nfsd4_session *new; 737 struct nfsd4_channel_attrs *fchan = &cses->fore_channel; 738 int numslots, slotsize; 739 int status; 740 int idx; 741 742 /* 743 * Note decreasing slot size below client's request may 744 * make it difficult for client to function correctly, whereas 745 * decreasing the number of slots will (just?) affect 746 * performance. When short on memory we therefore prefer to 747 * decrease number of slots instead of their size. 748 */ 749 slotsize = nfsd4_sanitize_slot_size(fchan->maxresp_cached); 750 numslots = nfsd4_get_drc_mem(slotsize, fchan->maxreqs); 751 if (numslots < 1) 752 return NULL; 753 754 new = alloc_session(slotsize, numslots); 755 if (!new) { 756 nfsd4_put_drc_mem(slotsize, fchan->maxreqs); 757 return NULL; 758 } 759 init_forechannel_attrs(&new->se_fchannel, fchan, numslots, slotsize); 760 761 new->se_client = clp; 762 gen_sessionid(new); 763 764 INIT_LIST_HEAD(&new->se_conns); 765 766 new->se_cb_seq_nr = 1; 767 new->se_flags = cses->flags; 768 new->se_cb_prog = cses->callback_prog; 769 kref_init(&new->se_ref); 770 idx = hash_sessionid(&new->se_sessionid); 771 spin_lock(&client_lock); 772 list_add(&new->se_hash, &sessionid_hashtbl[idx]); 773 spin_lock(&clp->cl_lock); 774 list_add(&new->se_perclnt, &clp->cl_sessions); 775 spin_unlock(&clp->cl_lock); 776 spin_unlock(&client_lock); 777 778 status = nfsd4_new_conn_from_crses(rqstp, new); 779 /* whoops: benny points out, status is ignored! (err, or bogus) */ 780 if (status) { 781 free_session(&new->se_ref); 782 return NULL; 783 } 784 if (cses->flags & SESSION4_BACK_CHAN) { 785 struct sockaddr *sa = svc_addr(rqstp); 786 /* 787 * This is a little silly; with sessions there's no real 788 * use for the callback address. Use the peer address 789 * as a reasonable default for now, but consider fixing 790 * the rpc client not to require an address in the 791 * future: 792 */ 793 rpc_copy_addr((struct sockaddr *)&clp->cl_cb_conn.cb_addr, sa); 794 clp->cl_cb_conn.cb_addrlen = svc_addr_len(sa); 795 } 796 nfsd4_probe_callback(clp); 797 return new; 798 } 799 800 /* caller must hold client_lock */ 801 static struct nfsd4_session * 802 find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid) 803 { 804 struct nfsd4_session *elem; 805 int idx; 806 807 dump_sessionid(__func__, sessionid); 808 idx = hash_sessionid(sessionid); 809 /* Search in the appropriate list */ 810 list_for_each_entry(elem, &sessionid_hashtbl[idx], se_hash) { 811 if (!memcmp(elem->se_sessionid.data, sessionid->data, 812 NFS4_MAX_SESSIONID_LEN)) { 813 return elem; 814 } 815 } 816 817 dprintk("%s: session not found\n", __func__); 818 return NULL; 819 } 820 821 /* caller must hold client_lock */ 822 static void 823 unhash_session(struct nfsd4_session *ses) 824 { 825 list_del(&ses->se_hash); 826 spin_lock(&ses->se_client->cl_lock); 827 list_del(&ses->se_perclnt); 828 spin_unlock(&ses->se_client->cl_lock); 829 } 830 831 /* must be called under the client_lock */ 832 static inline void 833 renew_client_locked(struct nfs4_client *clp) 834 { 835 if (is_client_expired(clp)) { 836 dprintk("%s: client (clientid %08x/%08x) already expired\n", 837 __func__, 838 clp->cl_clientid.cl_boot, 839 clp->cl_clientid.cl_id); 840 return; 841 } 842 843 /* 844 * Move client to the end to the LRU list. 845 */ 846 dprintk("renewing client (clientid %08x/%08x)\n", 847 clp->cl_clientid.cl_boot, 848 clp->cl_clientid.cl_id); 849 list_move_tail(&clp->cl_lru, &client_lru); 850 clp->cl_time = get_seconds(); 851 } 852 853 static inline void 854 renew_client(struct nfs4_client *clp) 855 { 856 spin_lock(&client_lock); 857 renew_client_locked(clp); 858 spin_unlock(&client_lock); 859 } 860 861 /* SETCLIENTID and SETCLIENTID_CONFIRM Helper functions */ 862 static int 863 STALE_CLIENTID(clientid_t *clid) 864 { 865 if (clid->cl_boot == boot_time) 866 return 0; 867 dprintk("NFSD stale clientid (%08x/%08x) boot_time %08lx\n", 868 clid->cl_boot, clid->cl_id, boot_time); 869 return 1; 870 } 871 872 /* 873 * XXX Should we use a slab cache ? 874 * This type of memory management is somewhat inefficient, but we use it 875 * anyway since SETCLIENTID is not a common operation. 876 */ 877 static struct nfs4_client *alloc_client(struct xdr_netobj name) 878 { 879 struct nfs4_client *clp; 880 881 clp = kzalloc(sizeof(struct nfs4_client), GFP_KERNEL); 882 if (clp == NULL) 883 return NULL; 884 clp->cl_name.data = kmalloc(name.len, GFP_KERNEL); 885 if (clp->cl_name.data == NULL) { 886 kfree(clp); 887 return NULL; 888 } 889 memcpy(clp->cl_name.data, name.data, name.len); 890 clp->cl_name.len = name.len; 891 return clp; 892 } 893 894 static inline void 895 free_client(struct nfs4_client *clp) 896 { 897 while (!list_empty(&clp->cl_sessions)) { 898 struct nfsd4_session *ses; 899 ses = list_entry(clp->cl_sessions.next, struct nfsd4_session, 900 se_perclnt); 901 list_del(&ses->se_perclnt); 902 nfsd4_put_session(ses); 903 } 904 if (clp->cl_cred.cr_group_info) 905 put_group_info(clp->cl_cred.cr_group_info); 906 kfree(clp->cl_principal); 907 kfree(clp->cl_name.data); 908 kfree(clp); 909 } 910 911 void 912 release_session_client(struct nfsd4_session *session) 913 { 914 struct nfs4_client *clp = session->se_client; 915 916 if (!atomic_dec_and_lock(&clp->cl_refcount, &client_lock)) 917 return; 918 if (is_client_expired(clp)) { 919 free_client(clp); 920 session->se_client = NULL; 921 } else 922 renew_client_locked(clp); 923 spin_unlock(&client_lock); 924 } 925 926 /* must be called under the client_lock */ 927 static inline void 928 unhash_client_locked(struct nfs4_client *clp) 929 { 930 struct nfsd4_session *ses; 931 932 mark_client_expired(clp); 933 list_del(&clp->cl_lru); 934 spin_lock(&clp->cl_lock); 935 list_for_each_entry(ses, &clp->cl_sessions, se_perclnt) 936 list_del_init(&ses->se_hash); 937 spin_unlock(&clp->cl_lock); 938 } 939 940 static void 941 expire_client(struct nfs4_client *clp) 942 { 943 struct nfs4_stateowner *sop; 944 struct nfs4_delegation *dp; 945 struct list_head reaplist; 946 947 INIT_LIST_HEAD(&reaplist); 948 spin_lock(&recall_lock); 949 while (!list_empty(&clp->cl_delegations)) { 950 dp = list_entry(clp->cl_delegations.next, struct nfs4_delegation, dl_perclnt); 951 list_del_init(&dp->dl_perclnt); 952 list_move(&dp->dl_recall_lru, &reaplist); 953 } 954 spin_unlock(&recall_lock); 955 while (!list_empty(&reaplist)) { 956 dp = list_entry(reaplist.next, struct nfs4_delegation, dl_recall_lru); 957 list_del_init(&dp->dl_recall_lru); 958 unhash_delegation(dp); 959 } 960 while (!list_empty(&clp->cl_openowners)) { 961 sop = list_entry(clp->cl_openowners.next, struct nfs4_stateowner, so_perclient); 962 release_openowner(sop); 963 } 964 nfsd4_shutdown_callback(clp); 965 if (clp->cl_cb_conn.cb_xprt) 966 svc_xprt_put(clp->cl_cb_conn.cb_xprt); 967 list_del(&clp->cl_idhash); 968 list_del(&clp->cl_strhash); 969 spin_lock(&client_lock); 970 unhash_client_locked(clp); 971 if (atomic_read(&clp->cl_refcount) == 0) 972 free_client(clp); 973 spin_unlock(&client_lock); 974 } 975 976 static void copy_verf(struct nfs4_client *target, nfs4_verifier *source) 977 { 978 memcpy(target->cl_verifier.data, source->data, 979 sizeof(target->cl_verifier.data)); 980 } 981 982 static void copy_clid(struct nfs4_client *target, struct nfs4_client *source) 983 { 984 target->cl_clientid.cl_boot = source->cl_clientid.cl_boot; 985 target->cl_clientid.cl_id = source->cl_clientid.cl_id; 986 } 987 988 static void copy_cred(struct svc_cred *target, struct svc_cred *source) 989 { 990 target->cr_uid = source->cr_uid; 991 target->cr_gid = source->cr_gid; 992 target->cr_group_info = source->cr_group_info; 993 get_group_info(target->cr_group_info); 994 } 995 996 static int same_name(const char *n1, const char *n2) 997 { 998 return 0 == memcmp(n1, n2, HEXDIR_LEN); 999 } 1000 1001 static int 1002 same_verf(nfs4_verifier *v1, nfs4_verifier *v2) 1003 { 1004 return 0 == memcmp(v1->data, v2->data, sizeof(v1->data)); 1005 } 1006 1007 static int 1008 same_clid(clientid_t *cl1, clientid_t *cl2) 1009 { 1010 return (cl1->cl_boot == cl2->cl_boot) && (cl1->cl_id == cl2->cl_id); 1011 } 1012 1013 /* XXX what about NGROUP */ 1014 static int 1015 same_creds(struct svc_cred *cr1, struct svc_cred *cr2) 1016 { 1017 return cr1->cr_uid == cr2->cr_uid; 1018 } 1019 1020 static void gen_clid(struct nfs4_client *clp) 1021 { 1022 static u32 current_clientid = 1; 1023 1024 clp->cl_clientid.cl_boot = boot_time; 1025 clp->cl_clientid.cl_id = current_clientid++; 1026 } 1027 1028 static void gen_confirm(struct nfs4_client *clp) 1029 { 1030 static u32 i; 1031 u32 *p; 1032 1033 p = (u32 *)clp->cl_confirm.data; 1034 *p++ = get_seconds(); 1035 *p++ = i++; 1036 } 1037 1038 static struct nfs4_client *create_client(struct xdr_netobj name, char *recdir, 1039 struct svc_rqst *rqstp, nfs4_verifier *verf) 1040 { 1041 struct nfs4_client *clp; 1042 struct sockaddr *sa = svc_addr(rqstp); 1043 char *princ; 1044 1045 clp = alloc_client(name); 1046 if (clp == NULL) 1047 return NULL; 1048 1049 INIT_LIST_HEAD(&clp->cl_sessions); 1050 1051 princ = svc_gss_principal(rqstp); 1052 if (princ) { 1053 clp->cl_principal = kstrdup(princ, GFP_KERNEL); 1054 if (clp->cl_principal == NULL) { 1055 free_client(clp); 1056 return NULL; 1057 } 1058 } 1059 1060 memcpy(clp->cl_recdir, recdir, HEXDIR_LEN); 1061 atomic_set(&clp->cl_refcount, 0); 1062 clp->cl_cb_state = NFSD4_CB_UNKNOWN; 1063 INIT_LIST_HEAD(&clp->cl_idhash); 1064 INIT_LIST_HEAD(&clp->cl_strhash); 1065 INIT_LIST_HEAD(&clp->cl_openowners); 1066 INIT_LIST_HEAD(&clp->cl_delegations); 1067 INIT_LIST_HEAD(&clp->cl_lru); 1068 INIT_LIST_HEAD(&clp->cl_callbacks); 1069 spin_lock_init(&clp->cl_lock); 1070 INIT_WORK(&clp->cl_cb_null.cb_work, nfsd4_do_callback_rpc); 1071 clp->cl_time = get_seconds(); 1072 clear_bit(0, &clp->cl_cb_slot_busy); 1073 rpc_init_wait_queue(&clp->cl_cb_waitq, "Backchannel slot table"); 1074 copy_verf(clp, verf); 1075 rpc_copy_addr((struct sockaddr *) &clp->cl_addr, sa); 1076 clp->cl_flavor = rqstp->rq_flavor; 1077 copy_cred(&clp->cl_cred, &rqstp->rq_cred); 1078 gen_confirm(clp); 1079 clp->cl_cb_session = NULL; 1080 return clp; 1081 } 1082 1083 static int check_name(struct xdr_netobj name) 1084 { 1085 if (name.len == 0) 1086 return 0; 1087 if (name.len > NFS4_OPAQUE_LIMIT) { 1088 dprintk("NFSD: check_name: name too long(%d)!\n", name.len); 1089 return 0; 1090 } 1091 return 1; 1092 } 1093 1094 static void 1095 add_to_unconfirmed(struct nfs4_client *clp, unsigned int strhashval) 1096 { 1097 unsigned int idhashval; 1098 1099 list_add(&clp->cl_strhash, &unconf_str_hashtbl[strhashval]); 1100 idhashval = clientid_hashval(clp->cl_clientid.cl_id); 1101 list_add(&clp->cl_idhash, &unconf_id_hashtbl[idhashval]); 1102 renew_client(clp); 1103 } 1104 1105 static void 1106 move_to_confirmed(struct nfs4_client *clp) 1107 { 1108 unsigned int idhashval = clientid_hashval(clp->cl_clientid.cl_id); 1109 unsigned int strhashval; 1110 1111 dprintk("NFSD: move_to_confirm nfs4_client %p\n", clp); 1112 list_move(&clp->cl_idhash, &conf_id_hashtbl[idhashval]); 1113 strhashval = clientstr_hashval(clp->cl_recdir); 1114 list_move(&clp->cl_strhash, &conf_str_hashtbl[strhashval]); 1115 renew_client(clp); 1116 } 1117 1118 static struct nfs4_client * 1119 find_confirmed_client(clientid_t *clid) 1120 { 1121 struct nfs4_client *clp; 1122 unsigned int idhashval = clientid_hashval(clid->cl_id); 1123 1124 list_for_each_entry(clp, &conf_id_hashtbl[idhashval], cl_idhash) { 1125 if (same_clid(&clp->cl_clientid, clid)) 1126 return clp; 1127 } 1128 return NULL; 1129 } 1130 1131 static struct nfs4_client * 1132 find_unconfirmed_client(clientid_t *clid) 1133 { 1134 struct nfs4_client *clp; 1135 unsigned int idhashval = clientid_hashval(clid->cl_id); 1136 1137 list_for_each_entry(clp, &unconf_id_hashtbl[idhashval], cl_idhash) { 1138 if (same_clid(&clp->cl_clientid, clid)) 1139 return clp; 1140 } 1141 return NULL; 1142 } 1143 1144 static bool clp_used_exchangeid(struct nfs4_client *clp) 1145 { 1146 return clp->cl_exchange_flags != 0; 1147 } 1148 1149 static struct nfs4_client * 1150 find_confirmed_client_by_str(const char *dname, unsigned int hashval) 1151 { 1152 struct nfs4_client *clp; 1153 1154 list_for_each_entry(clp, &conf_str_hashtbl[hashval], cl_strhash) { 1155 if (same_name(clp->cl_recdir, dname)) 1156 return clp; 1157 } 1158 return NULL; 1159 } 1160 1161 static struct nfs4_client * 1162 find_unconfirmed_client_by_str(const char *dname, unsigned int hashval) 1163 { 1164 struct nfs4_client *clp; 1165 1166 list_for_each_entry(clp, &unconf_str_hashtbl[hashval], cl_strhash) { 1167 if (same_name(clp->cl_recdir, dname)) 1168 return clp; 1169 } 1170 return NULL; 1171 } 1172 1173 static void rpc_svcaddr2sockaddr(struct sockaddr *sa, unsigned short family, union svc_addr_u *svcaddr) 1174 { 1175 switch (family) { 1176 case AF_INET: 1177 ((struct sockaddr_in *)sa)->sin_family = AF_INET; 1178 ((struct sockaddr_in *)sa)->sin_addr = svcaddr->addr; 1179 return; 1180 case AF_INET6: 1181 ((struct sockaddr_in6 *)sa)->sin6_family = AF_INET6; 1182 ((struct sockaddr_in6 *)sa)->sin6_addr = svcaddr->addr6; 1183 return; 1184 } 1185 } 1186 1187 static void 1188 gen_callback(struct nfs4_client *clp, struct nfsd4_setclientid *se, struct svc_rqst *rqstp) 1189 { 1190 struct nfs4_cb_conn *conn = &clp->cl_cb_conn; 1191 struct sockaddr *sa = svc_addr(rqstp); 1192 u32 scopeid = rpc_get_scope_id(sa); 1193 unsigned short expected_family; 1194 1195 /* Currently, we only support tcp and tcp6 for the callback channel */ 1196 if (se->se_callback_netid_len == 3 && 1197 !memcmp(se->se_callback_netid_val, "tcp", 3)) 1198 expected_family = AF_INET; 1199 else if (se->se_callback_netid_len == 4 && 1200 !memcmp(se->se_callback_netid_val, "tcp6", 4)) 1201 expected_family = AF_INET6; 1202 else 1203 goto out_err; 1204 1205 conn->cb_addrlen = rpc_uaddr2sockaddr(se->se_callback_addr_val, 1206 se->se_callback_addr_len, 1207 (struct sockaddr *)&conn->cb_addr, 1208 sizeof(conn->cb_addr)); 1209 1210 if (!conn->cb_addrlen || conn->cb_addr.ss_family != expected_family) 1211 goto out_err; 1212 1213 if (conn->cb_addr.ss_family == AF_INET6) 1214 ((struct sockaddr_in6 *)&conn->cb_addr)->sin6_scope_id = scopeid; 1215 1216 conn->cb_prog = se->se_callback_prog; 1217 conn->cb_ident = se->se_callback_ident; 1218 rpc_svcaddr2sockaddr((struct sockaddr *)&conn->cb_saddr, expected_family, &rqstp->rq_daddr); 1219 return; 1220 out_err: 1221 conn->cb_addr.ss_family = AF_UNSPEC; 1222 conn->cb_addrlen = 0; 1223 dprintk(KERN_INFO "NFSD: this client (clientid %08x/%08x) " 1224 "will not receive delegations\n", 1225 clp->cl_clientid.cl_boot, clp->cl_clientid.cl_id); 1226 1227 return; 1228 } 1229 1230 /* 1231 * Cache a reply. nfsd4_check_drc_limit() has bounded the cache size. 1232 */ 1233 void 1234 nfsd4_store_cache_entry(struct nfsd4_compoundres *resp) 1235 { 1236 struct nfsd4_slot *slot = resp->cstate.slot; 1237 unsigned int base; 1238 1239 dprintk("--> %s slot %p\n", __func__, slot); 1240 1241 slot->sl_opcnt = resp->opcnt; 1242 slot->sl_status = resp->cstate.status; 1243 1244 if (nfsd4_not_cached(resp)) { 1245 slot->sl_datalen = 0; 1246 return; 1247 } 1248 slot->sl_datalen = (char *)resp->p - (char *)resp->cstate.datap; 1249 base = (char *)resp->cstate.datap - 1250 (char *)resp->xbuf->head[0].iov_base; 1251 if (read_bytes_from_xdr_buf(resp->xbuf, base, slot->sl_data, 1252 slot->sl_datalen)) 1253 WARN("%s: sessions DRC could not cache compound\n", __func__); 1254 return; 1255 } 1256 1257 /* 1258 * Encode the replay sequence operation from the slot values. 1259 * If cachethis is FALSE encode the uncached rep error on the next 1260 * operation which sets resp->p and increments resp->opcnt for 1261 * nfs4svc_encode_compoundres. 1262 * 1263 */ 1264 static __be32 1265 nfsd4_enc_sequence_replay(struct nfsd4_compoundargs *args, 1266 struct nfsd4_compoundres *resp) 1267 { 1268 struct nfsd4_op *op; 1269 struct nfsd4_slot *slot = resp->cstate.slot; 1270 1271 dprintk("--> %s resp->opcnt %d cachethis %u \n", __func__, 1272 resp->opcnt, resp->cstate.slot->sl_cachethis); 1273 1274 /* Encode the replayed sequence operation */ 1275 op = &args->ops[resp->opcnt - 1]; 1276 nfsd4_encode_operation(resp, op); 1277 1278 /* Return nfserr_retry_uncached_rep in next operation. */ 1279 if (args->opcnt > 1 && slot->sl_cachethis == 0) { 1280 op = &args->ops[resp->opcnt++]; 1281 op->status = nfserr_retry_uncached_rep; 1282 nfsd4_encode_operation(resp, op); 1283 } 1284 return op->status; 1285 } 1286 1287 /* 1288 * The sequence operation is not cached because we can use the slot and 1289 * session values. 1290 */ 1291 __be32 1292 nfsd4_replay_cache_entry(struct nfsd4_compoundres *resp, 1293 struct nfsd4_sequence *seq) 1294 { 1295 struct nfsd4_slot *slot = resp->cstate.slot; 1296 __be32 status; 1297 1298 dprintk("--> %s slot %p\n", __func__, slot); 1299 1300 /* Either returns 0 or nfserr_retry_uncached */ 1301 status = nfsd4_enc_sequence_replay(resp->rqstp->rq_argp, resp); 1302 if (status == nfserr_retry_uncached_rep) 1303 return status; 1304 1305 /* The sequence operation has been encoded, cstate->datap set. */ 1306 memcpy(resp->cstate.datap, slot->sl_data, slot->sl_datalen); 1307 1308 resp->opcnt = slot->sl_opcnt; 1309 resp->p = resp->cstate.datap + XDR_QUADLEN(slot->sl_datalen); 1310 status = slot->sl_status; 1311 1312 return status; 1313 } 1314 1315 /* 1316 * Set the exchange_id flags returned by the server. 1317 */ 1318 static void 1319 nfsd4_set_ex_flags(struct nfs4_client *new, struct nfsd4_exchange_id *clid) 1320 { 1321 /* pNFS is not supported */ 1322 new->cl_exchange_flags |= EXCHGID4_FLAG_USE_NON_PNFS; 1323 1324 /* Referrals are supported, Migration is not. */ 1325 new->cl_exchange_flags |= EXCHGID4_FLAG_SUPP_MOVED_REFER; 1326 1327 /* set the wire flags to return to client. */ 1328 clid->flags = new->cl_exchange_flags; 1329 } 1330 1331 __be32 1332 nfsd4_exchange_id(struct svc_rqst *rqstp, 1333 struct nfsd4_compound_state *cstate, 1334 struct nfsd4_exchange_id *exid) 1335 { 1336 struct nfs4_client *unconf, *conf, *new; 1337 int status; 1338 unsigned int strhashval; 1339 char dname[HEXDIR_LEN]; 1340 char addr_str[INET6_ADDRSTRLEN]; 1341 nfs4_verifier verf = exid->verifier; 1342 struct sockaddr *sa = svc_addr(rqstp); 1343 1344 rpc_ntop(sa, addr_str, sizeof(addr_str)); 1345 dprintk("%s rqstp=%p exid=%p clname.len=%u clname.data=%p " 1346 "ip_addr=%s flags %x, spa_how %d\n", 1347 __func__, rqstp, exid, exid->clname.len, exid->clname.data, 1348 addr_str, exid->flags, exid->spa_how); 1349 1350 if (!check_name(exid->clname) || (exid->flags & ~EXCHGID4_FLAG_MASK_A)) 1351 return nfserr_inval; 1352 1353 /* Currently only support SP4_NONE */ 1354 switch (exid->spa_how) { 1355 case SP4_NONE: 1356 break; 1357 case SP4_SSV: 1358 return nfserr_serverfault; 1359 default: 1360 BUG(); /* checked by xdr code */ 1361 case SP4_MACH_CRED: 1362 return nfserr_serverfault; /* no excuse :-/ */ 1363 } 1364 1365 status = nfs4_make_rec_clidname(dname, &exid->clname); 1366 1367 if (status) 1368 goto error; 1369 1370 strhashval = clientstr_hashval(dname); 1371 1372 nfs4_lock_state(); 1373 status = nfs_ok; 1374 1375 conf = find_confirmed_client_by_str(dname, strhashval); 1376 if (conf) { 1377 if (!clp_used_exchangeid(conf)) { 1378 status = nfserr_clid_inuse; /* XXX: ? */ 1379 goto out; 1380 } 1381 if (!same_verf(&verf, &conf->cl_verifier)) { 1382 /* 18.35.4 case 8 */ 1383 if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) { 1384 status = nfserr_not_same; 1385 goto out; 1386 } 1387 /* Client reboot: destroy old state */ 1388 expire_client(conf); 1389 goto out_new; 1390 } 1391 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) { 1392 /* 18.35.4 case 9 */ 1393 if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) { 1394 status = nfserr_perm; 1395 goto out; 1396 } 1397 expire_client(conf); 1398 goto out_new; 1399 } 1400 /* 1401 * Set bit when the owner id and verifier map to an already 1402 * confirmed client id (18.35.3). 1403 */ 1404 exid->flags |= EXCHGID4_FLAG_CONFIRMED_R; 1405 1406 /* 1407 * Falling into 18.35.4 case 2, possible router replay. 1408 * Leave confirmed record intact and return same result. 1409 */ 1410 copy_verf(conf, &verf); 1411 new = conf; 1412 goto out_copy; 1413 } 1414 1415 /* 18.35.4 case 7 */ 1416 if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) { 1417 status = nfserr_noent; 1418 goto out; 1419 } 1420 1421 unconf = find_unconfirmed_client_by_str(dname, strhashval); 1422 if (unconf) { 1423 /* 1424 * Possible retry or client restart. Per 18.35.4 case 4, 1425 * a new unconfirmed record should be generated regardless 1426 * of whether any properties have changed. 1427 */ 1428 expire_client(unconf); 1429 } 1430 1431 out_new: 1432 /* Normal case */ 1433 new = create_client(exid->clname, dname, rqstp, &verf); 1434 if (new == NULL) { 1435 status = nfserr_jukebox; 1436 goto out; 1437 } 1438 1439 gen_clid(new); 1440 add_to_unconfirmed(new, strhashval); 1441 out_copy: 1442 exid->clientid.cl_boot = new->cl_clientid.cl_boot; 1443 exid->clientid.cl_id = new->cl_clientid.cl_id; 1444 1445 exid->seqid = 1; 1446 nfsd4_set_ex_flags(new, exid); 1447 1448 dprintk("nfsd4_exchange_id seqid %d flags %x\n", 1449 new->cl_cs_slot.sl_seqid, new->cl_exchange_flags); 1450 status = nfs_ok; 1451 1452 out: 1453 nfs4_unlock_state(); 1454 error: 1455 dprintk("nfsd4_exchange_id returns %d\n", ntohl(status)); 1456 return status; 1457 } 1458 1459 static int 1460 check_slot_seqid(u32 seqid, u32 slot_seqid, int slot_inuse) 1461 { 1462 dprintk("%s enter. seqid %d slot_seqid %d\n", __func__, seqid, 1463 slot_seqid); 1464 1465 /* The slot is in use, and no response has been sent. */ 1466 if (slot_inuse) { 1467 if (seqid == slot_seqid) 1468 return nfserr_jukebox; 1469 else 1470 return nfserr_seq_misordered; 1471 } 1472 /* Normal */ 1473 if (likely(seqid == slot_seqid + 1)) 1474 return nfs_ok; 1475 /* Replay */ 1476 if (seqid == slot_seqid) 1477 return nfserr_replay_cache; 1478 /* Wraparound */ 1479 if (seqid == 1 && (slot_seqid + 1) == 0) 1480 return nfs_ok; 1481 /* Misordered replay or misordered new request */ 1482 return nfserr_seq_misordered; 1483 } 1484 1485 /* 1486 * Cache the create session result into the create session single DRC 1487 * slot cache by saving the xdr structure. sl_seqid has been set. 1488 * Do this for solo or embedded create session operations. 1489 */ 1490 static void 1491 nfsd4_cache_create_session(struct nfsd4_create_session *cr_ses, 1492 struct nfsd4_clid_slot *slot, int nfserr) 1493 { 1494 slot->sl_status = nfserr; 1495 memcpy(&slot->sl_cr_ses, cr_ses, sizeof(*cr_ses)); 1496 } 1497 1498 static __be32 1499 nfsd4_replay_create_session(struct nfsd4_create_session *cr_ses, 1500 struct nfsd4_clid_slot *slot) 1501 { 1502 memcpy(cr_ses, &slot->sl_cr_ses, sizeof(*cr_ses)); 1503 return slot->sl_status; 1504 } 1505 1506 __be32 1507 nfsd4_create_session(struct svc_rqst *rqstp, 1508 struct nfsd4_compound_state *cstate, 1509 struct nfsd4_create_session *cr_ses) 1510 { 1511 struct sockaddr *sa = svc_addr(rqstp); 1512 struct nfs4_client *conf, *unconf; 1513 struct nfsd4_session *new; 1514 struct nfsd4_clid_slot *cs_slot = NULL; 1515 bool confirm_me = false; 1516 int status = 0; 1517 1518 nfs4_lock_state(); 1519 unconf = find_unconfirmed_client(&cr_ses->clientid); 1520 conf = find_confirmed_client(&cr_ses->clientid); 1521 1522 if (conf) { 1523 cs_slot = &conf->cl_cs_slot; 1524 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0); 1525 if (status == nfserr_replay_cache) { 1526 dprintk("Got a create_session replay! seqid= %d\n", 1527 cs_slot->sl_seqid); 1528 /* Return the cached reply status */ 1529 status = nfsd4_replay_create_session(cr_ses, cs_slot); 1530 goto out; 1531 } else if (cr_ses->seqid != cs_slot->sl_seqid + 1) { 1532 status = nfserr_seq_misordered; 1533 dprintk("Sequence misordered!\n"); 1534 dprintk("Expected seqid= %d but got seqid= %d\n", 1535 cs_slot->sl_seqid, cr_ses->seqid); 1536 goto out; 1537 } 1538 } else if (unconf) { 1539 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred) || 1540 !rpc_cmp_addr(sa, (struct sockaddr *) &unconf->cl_addr)) { 1541 status = nfserr_clid_inuse; 1542 goto out; 1543 } 1544 1545 cs_slot = &unconf->cl_cs_slot; 1546 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0); 1547 if (status) { 1548 /* an unconfirmed replay returns misordered */ 1549 status = nfserr_seq_misordered; 1550 goto out; 1551 } 1552 1553 confirm_me = true; 1554 conf = unconf; 1555 } else { 1556 status = nfserr_stale_clientid; 1557 goto out; 1558 } 1559 1560 /* 1561 * XXX: we should probably set this at creation time, and check 1562 * for consistent minorversion use throughout: 1563 */ 1564 conf->cl_minorversion = 1; 1565 /* 1566 * We do not support RDMA or persistent sessions 1567 */ 1568 cr_ses->flags &= ~SESSION4_PERSIST; 1569 cr_ses->flags &= ~SESSION4_RDMA; 1570 1571 status = nfserr_jukebox; 1572 new = alloc_init_session(rqstp, conf, cr_ses); 1573 if (!new) 1574 goto out; 1575 status = nfs_ok; 1576 memcpy(cr_ses->sessionid.data, new->se_sessionid.data, 1577 NFS4_MAX_SESSIONID_LEN); 1578 memcpy(&cr_ses->fore_channel, &new->se_fchannel, 1579 sizeof(struct nfsd4_channel_attrs)); 1580 cs_slot->sl_seqid++; 1581 cr_ses->seqid = cs_slot->sl_seqid; 1582 1583 /* cache solo and embedded create sessions under the state lock */ 1584 nfsd4_cache_create_session(cr_ses, cs_slot, status); 1585 if (confirm_me) 1586 move_to_confirmed(conf); 1587 out: 1588 nfs4_unlock_state(); 1589 dprintk("%s returns %d\n", __func__, ntohl(status)); 1590 return status; 1591 } 1592 1593 static bool nfsd4_last_compound_op(struct svc_rqst *rqstp) 1594 { 1595 struct nfsd4_compoundres *resp = rqstp->rq_resp; 1596 struct nfsd4_compoundargs *argp = rqstp->rq_argp; 1597 1598 return argp->opcnt == resp->opcnt; 1599 } 1600 1601 static __be32 nfsd4_map_bcts_dir(u32 *dir) 1602 { 1603 switch (*dir) { 1604 case NFS4_CDFC4_FORE: 1605 case NFS4_CDFC4_BACK: 1606 return nfs_ok; 1607 case NFS4_CDFC4_FORE_OR_BOTH: 1608 case NFS4_CDFC4_BACK_OR_BOTH: 1609 *dir = NFS4_CDFC4_BOTH; 1610 return nfs_ok; 1611 }; 1612 return nfserr_inval; 1613 } 1614 1615 __be32 nfsd4_bind_conn_to_session(struct svc_rqst *rqstp, 1616 struct nfsd4_compound_state *cstate, 1617 struct nfsd4_bind_conn_to_session *bcts) 1618 { 1619 __be32 status; 1620 1621 if (!nfsd4_last_compound_op(rqstp)) 1622 return nfserr_not_only_op; 1623 spin_lock(&client_lock); 1624 cstate->session = find_in_sessionid_hashtbl(&bcts->sessionid); 1625 /* Sorta weird: we only need the refcnt'ing because new_conn acquires 1626 * client_lock iself: */ 1627 if (cstate->session) { 1628 nfsd4_get_session(cstate->session); 1629 atomic_inc(&cstate->session->se_client->cl_refcount); 1630 } 1631 spin_unlock(&client_lock); 1632 if (!cstate->session) 1633 return nfserr_badsession; 1634 1635 status = nfsd4_map_bcts_dir(&bcts->dir); 1636 nfsd4_new_conn(rqstp, cstate->session, bcts->dir); 1637 return nfs_ok; 1638 } 1639 1640 static bool nfsd4_compound_in_session(struct nfsd4_session *session, struct nfs4_sessionid *sid) 1641 { 1642 if (!session) 1643 return 0; 1644 return !memcmp(sid, &session->se_sessionid, sizeof(*sid)); 1645 } 1646 1647 __be32 1648 nfsd4_destroy_session(struct svc_rqst *r, 1649 struct nfsd4_compound_state *cstate, 1650 struct nfsd4_destroy_session *sessionid) 1651 { 1652 struct nfsd4_session *ses; 1653 u32 status = nfserr_badsession; 1654 1655 /* Notes: 1656 * - The confirmed nfs4_client->cl_sessionid holds destroyed sessinid 1657 * - Should we return nfserr_back_chan_busy if waiting for 1658 * callbacks on to-be-destroyed session? 1659 * - Do we need to clear any callback info from previous session? 1660 */ 1661 1662 if (nfsd4_compound_in_session(cstate->session, &sessionid->sessionid)) { 1663 if (!nfsd4_last_compound_op(r)) 1664 return nfserr_not_only_op; 1665 } 1666 dump_sessionid(__func__, &sessionid->sessionid); 1667 spin_lock(&client_lock); 1668 ses = find_in_sessionid_hashtbl(&sessionid->sessionid); 1669 if (!ses) { 1670 spin_unlock(&client_lock); 1671 goto out; 1672 } 1673 1674 unhash_session(ses); 1675 spin_unlock(&client_lock); 1676 1677 nfs4_lock_state(); 1678 nfsd4_probe_callback_sync(ses->se_client); 1679 nfs4_unlock_state(); 1680 1681 nfsd4_del_conns(ses); 1682 1683 nfsd4_put_session(ses); 1684 status = nfs_ok; 1685 out: 1686 dprintk("%s returns %d\n", __func__, ntohl(status)); 1687 return status; 1688 } 1689 1690 static struct nfsd4_conn *__nfsd4_find_conn(struct svc_xprt *xpt, struct nfsd4_session *s) 1691 { 1692 struct nfsd4_conn *c; 1693 1694 list_for_each_entry(c, &s->se_conns, cn_persession) { 1695 if (c->cn_xprt == xpt) { 1696 return c; 1697 } 1698 } 1699 return NULL; 1700 } 1701 1702 static void nfsd4_sequence_check_conn(struct nfsd4_conn *new, struct nfsd4_session *ses) 1703 { 1704 struct nfs4_client *clp = ses->se_client; 1705 struct nfsd4_conn *c; 1706 int ret; 1707 1708 spin_lock(&clp->cl_lock); 1709 c = __nfsd4_find_conn(new->cn_xprt, ses); 1710 if (c) { 1711 spin_unlock(&clp->cl_lock); 1712 free_conn(new); 1713 return; 1714 } 1715 __nfsd4_hash_conn(new, ses); 1716 spin_unlock(&clp->cl_lock); 1717 ret = nfsd4_register_conn(new); 1718 if (ret) 1719 /* oops; xprt is already down: */ 1720 nfsd4_conn_lost(&new->cn_xpt_user); 1721 return; 1722 } 1723 1724 __be32 1725 nfsd4_sequence(struct svc_rqst *rqstp, 1726 struct nfsd4_compound_state *cstate, 1727 struct nfsd4_sequence *seq) 1728 { 1729 struct nfsd4_compoundres *resp = rqstp->rq_resp; 1730 struct nfsd4_session *session; 1731 struct nfsd4_slot *slot; 1732 struct nfsd4_conn *conn; 1733 int status; 1734 1735 if (resp->opcnt != 1) 1736 return nfserr_sequence_pos; 1737 1738 /* 1739 * Will be either used or freed by nfsd4_sequence_check_conn 1740 * below. 1741 */ 1742 conn = alloc_conn(rqstp, NFS4_CDFC4_FORE); 1743 if (!conn) 1744 return nfserr_jukebox; 1745 1746 spin_lock(&client_lock); 1747 status = nfserr_badsession; 1748 session = find_in_sessionid_hashtbl(&seq->sessionid); 1749 if (!session) 1750 goto out; 1751 1752 status = nfserr_badslot; 1753 if (seq->slotid >= session->se_fchannel.maxreqs) 1754 goto out; 1755 1756 slot = session->se_slots[seq->slotid]; 1757 dprintk("%s: slotid %d\n", __func__, seq->slotid); 1758 1759 /* We do not negotiate the number of slots yet, so set the 1760 * maxslots to the session maxreqs which is used to encode 1761 * sr_highest_slotid and the sr_target_slot id to maxslots */ 1762 seq->maxslots = session->se_fchannel.maxreqs; 1763 1764 status = check_slot_seqid(seq->seqid, slot->sl_seqid, slot->sl_inuse); 1765 if (status == nfserr_replay_cache) { 1766 cstate->slot = slot; 1767 cstate->session = session; 1768 /* Return the cached reply status and set cstate->status 1769 * for nfsd4_proc_compound processing */ 1770 status = nfsd4_replay_cache_entry(resp, seq); 1771 cstate->status = nfserr_replay_cache; 1772 goto out; 1773 } 1774 if (status) 1775 goto out; 1776 1777 nfsd4_sequence_check_conn(conn, session); 1778 conn = NULL; 1779 1780 /* Success! bump slot seqid */ 1781 slot->sl_inuse = true; 1782 slot->sl_seqid = seq->seqid; 1783 slot->sl_cachethis = seq->cachethis; 1784 1785 cstate->slot = slot; 1786 cstate->session = session; 1787 1788 out: 1789 /* Hold a session reference until done processing the compound. */ 1790 if (cstate->session) { 1791 struct nfs4_client *clp = session->se_client; 1792 1793 nfsd4_get_session(cstate->session); 1794 atomic_inc(&clp->cl_refcount); 1795 if (clp->cl_cb_state == NFSD4_CB_DOWN) 1796 seq->status_flags |= SEQ4_STATUS_CB_PATH_DOWN; 1797 } 1798 kfree(conn); 1799 spin_unlock(&client_lock); 1800 dprintk("%s: return %d\n", __func__, ntohl(status)); 1801 return status; 1802 } 1803 1804 __be32 1805 nfsd4_reclaim_complete(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate, struct nfsd4_reclaim_complete *rc) 1806 { 1807 if (rc->rca_one_fs) { 1808 if (!cstate->current_fh.fh_dentry) 1809 return nfserr_nofilehandle; 1810 /* 1811 * We don't take advantage of the rca_one_fs case. 1812 * That's OK, it's optional, we can safely ignore it. 1813 */ 1814 return nfs_ok; 1815 } 1816 nfs4_lock_state(); 1817 if (is_client_expired(cstate->session->se_client)) { 1818 nfs4_unlock_state(); 1819 /* 1820 * The following error isn't really legal. 1821 * But we only get here if the client just explicitly 1822 * destroyed the client. Surely it no longer cares what 1823 * error it gets back on an operation for the dead 1824 * client. 1825 */ 1826 return nfserr_stale_clientid; 1827 } 1828 nfsd4_create_clid_dir(cstate->session->se_client); 1829 nfs4_unlock_state(); 1830 return nfs_ok; 1831 } 1832 1833 __be32 1834 nfsd4_setclientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate, 1835 struct nfsd4_setclientid *setclid) 1836 { 1837 struct xdr_netobj clname = { 1838 .len = setclid->se_namelen, 1839 .data = setclid->se_name, 1840 }; 1841 nfs4_verifier clverifier = setclid->se_verf; 1842 unsigned int strhashval; 1843 struct nfs4_client *conf, *unconf, *new; 1844 __be32 status; 1845 char dname[HEXDIR_LEN]; 1846 1847 if (!check_name(clname)) 1848 return nfserr_inval; 1849 1850 status = nfs4_make_rec_clidname(dname, &clname); 1851 if (status) 1852 return status; 1853 1854 /* 1855 * XXX The Duplicate Request Cache (DRC) has been checked (??) 1856 * We get here on a DRC miss. 1857 */ 1858 1859 strhashval = clientstr_hashval(dname); 1860 1861 nfs4_lock_state(); 1862 conf = find_confirmed_client_by_str(dname, strhashval); 1863 if (conf) { 1864 /* RFC 3530 14.2.33 CASE 0: */ 1865 status = nfserr_clid_inuse; 1866 if (clp_used_exchangeid(conf)) 1867 goto out; 1868 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) { 1869 char addr_str[INET6_ADDRSTRLEN]; 1870 rpc_ntop((struct sockaddr *) &conf->cl_addr, addr_str, 1871 sizeof(addr_str)); 1872 dprintk("NFSD: setclientid: string in use by client " 1873 "at %s\n", addr_str); 1874 goto out; 1875 } 1876 } 1877 /* 1878 * section 14.2.33 of RFC 3530 (under the heading "IMPLEMENTATION") 1879 * has a description of SETCLIENTID request processing consisting 1880 * of 5 bullet points, labeled as CASE0 - CASE4 below. 1881 */ 1882 unconf = find_unconfirmed_client_by_str(dname, strhashval); 1883 status = nfserr_resource; 1884 if (!conf) { 1885 /* 1886 * RFC 3530 14.2.33 CASE 4: 1887 * placed first, because it is the normal case 1888 */ 1889 if (unconf) 1890 expire_client(unconf); 1891 new = create_client(clname, dname, rqstp, &clverifier); 1892 if (new == NULL) 1893 goto out; 1894 gen_clid(new); 1895 } else if (same_verf(&conf->cl_verifier, &clverifier)) { 1896 /* 1897 * RFC 3530 14.2.33 CASE 1: 1898 * probable callback update 1899 */ 1900 if (unconf) { 1901 /* Note this is removing unconfirmed {*x***}, 1902 * which is stronger than RFC recommended {vxc**}. 1903 * This has the advantage that there is at most 1904 * one {*x***} in either list at any time. 1905 */ 1906 expire_client(unconf); 1907 } 1908 new = create_client(clname, dname, rqstp, &clverifier); 1909 if (new == NULL) 1910 goto out; 1911 copy_clid(new, conf); 1912 } else if (!unconf) { 1913 /* 1914 * RFC 3530 14.2.33 CASE 2: 1915 * probable client reboot; state will be removed if 1916 * confirmed. 1917 */ 1918 new = create_client(clname, dname, rqstp, &clverifier); 1919 if (new == NULL) 1920 goto out; 1921 gen_clid(new); 1922 } else { 1923 /* 1924 * RFC 3530 14.2.33 CASE 3: 1925 * probable client reboot; state will be removed if 1926 * confirmed. 1927 */ 1928 expire_client(unconf); 1929 new = create_client(clname, dname, rqstp, &clverifier); 1930 if (new == NULL) 1931 goto out; 1932 gen_clid(new); 1933 } 1934 /* 1935 * XXX: we should probably set this at creation time, and check 1936 * for consistent minorversion use throughout: 1937 */ 1938 new->cl_minorversion = 0; 1939 gen_callback(new, setclid, rqstp); 1940 add_to_unconfirmed(new, strhashval); 1941 setclid->se_clientid.cl_boot = new->cl_clientid.cl_boot; 1942 setclid->se_clientid.cl_id = new->cl_clientid.cl_id; 1943 memcpy(setclid->se_confirm.data, new->cl_confirm.data, sizeof(setclid->se_confirm.data)); 1944 status = nfs_ok; 1945 out: 1946 nfs4_unlock_state(); 1947 return status; 1948 } 1949 1950 1951 /* 1952 * Section 14.2.34 of RFC 3530 (under the heading "IMPLEMENTATION") has 1953 * a description of SETCLIENTID_CONFIRM request processing consisting of 4 1954 * bullets, labeled as CASE1 - CASE4 below. 1955 */ 1956 __be32 1957 nfsd4_setclientid_confirm(struct svc_rqst *rqstp, 1958 struct nfsd4_compound_state *cstate, 1959 struct nfsd4_setclientid_confirm *setclientid_confirm) 1960 { 1961 struct sockaddr *sa = svc_addr(rqstp); 1962 struct nfs4_client *conf, *unconf; 1963 nfs4_verifier confirm = setclientid_confirm->sc_confirm; 1964 clientid_t * clid = &setclientid_confirm->sc_clientid; 1965 __be32 status; 1966 1967 if (STALE_CLIENTID(clid)) 1968 return nfserr_stale_clientid; 1969 /* 1970 * XXX The Duplicate Request Cache (DRC) has been checked (??) 1971 * We get here on a DRC miss. 1972 */ 1973 1974 nfs4_lock_state(); 1975 1976 conf = find_confirmed_client(clid); 1977 unconf = find_unconfirmed_client(clid); 1978 1979 status = nfserr_clid_inuse; 1980 if (conf && !rpc_cmp_addr((struct sockaddr *) &conf->cl_addr, sa)) 1981 goto out; 1982 if (unconf && !rpc_cmp_addr((struct sockaddr *) &unconf->cl_addr, sa)) 1983 goto out; 1984 1985 /* 1986 * section 14.2.34 of RFC 3530 has a description of 1987 * SETCLIENTID_CONFIRM request processing consisting 1988 * of 4 bullet points, labeled as CASE1 - CASE4 below. 1989 */ 1990 if (conf && unconf && same_verf(&confirm, &unconf->cl_confirm)) { 1991 /* 1992 * RFC 3530 14.2.34 CASE 1: 1993 * callback update 1994 */ 1995 if (!same_creds(&conf->cl_cred, &unconf->cl_cred)) 1996 status = nfserr_clid_inuse; 1997 else { 1998 nfsd4_change_callback(conf, &unconf->cl_cb_conn); 1999 nfsd4_probe_callback(conf); 2000 expire_client(unconf); 2001 status = nfs_ok; 2002 2003 } 2004 } else if (conf && !unconf) { 2005 /* 2006 * RFC 3530 14.2.34 CASE 2: 2007 * probable retransmitted request; play it safe and 2008 * do nothing. 2009 */ 2010 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) 2011 status = nfserr_clid_inuse; 2012 else 2013 status = nfs_ok; 2014 } else if (!conf && unconf 2015 && same_verf(&unconf->cl_confirm, &confirm)) { 2016 /* 2017 * RFC 3530 14.2.34 CASE 3: 2018 * Normal case; new or rebooted client: 2019 */ 2020 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred)) { 2021 status = nfserr_clid_inuse; 2022 } else { 2023 unsigned int hash = 2024 clientstr_hashval(unconf->cl_recdir); 2025 conf = find_confirmed_client_by_str(unconf->cl_recdir, 2026 hash); 2027 if (conf) { 2028 nfsd4_remove_clid_dir(conf); 2029 expire_client(conf); 2030 } 2031 move_to_confirmed(unconf); 2032 conf = unconf; 2033 nfsd4_probe_callback(conf); 2034 status = nfs_ok; 2035 } 2036 } else if ((!conf || (conf && !same_verf(&conf->cl_confirm, &confirm))) 2037 && (!unconf || (unconf && !same_verf(&unconf->cl_confirm, 2038 &confirm)))) { 2039 /* 2040 * RFC 3530 14.2.34 CASE 4: 2041 * Client probably hasn't noticed that we rebooted yet. 2042 */ 2043 status = nfserr_stale_clientid; 2044 } else { 2045 /* check that we have hit one of the cases...*/ 2046 status = nfserr_clid_inuse; 2047 } 2048 out: 2049 nfs4_unlock_state(); 2050 return status; 2051 } 2052 2053 /* OPEN Share state helper functions */ 2054 static inline struct nfs4_file * 2055 alloc_init_file(struct inode *ino) 2056 { 2057 struct nfs4_file *fp; 2058 unsigned int hashval = file_hashval(ino); 2059 2060 fp = kmem_cache_alloc(file_slab, GFP_KERNEL); 2061 if (fp) { 2062 atomic_set(&fp->fi_ref, 1); 2063 INIT_LIST_HEAD(&fp->fi_hash); 2064 INIT_LIST_HEAD(&fp->fi_stateids); 2065 INIT_LIST_HEAD(&fp->fi_delegations); 2066 fp->fi_inode = igrab(ino); 2067 fp->fi_id = current_fileid++; 2068 fp->fi_had_conflict = false; 2069 fp->fi_lease = NULL; 2070 memset(fp->fi_fds, 0, sizeof(fp->fi_fds)); 2071 memset(fp->fi_access, 0, sizeof(fp->fi_access)); 2072 spin_lock(&recall_lock); 2073 list_add(&fp->fi_hash, &file_hashtbl[hashval]); 2074 spin_unlock(&recall_lock); 2075 return fp; 2076 } 2077 return NULL; 2078 } 2079 2080 static void 2081 nfsd4_free_slab(struct kmem_cache **slab) 2082 { 2083 if (*slab == NULL) 2084 return; 2085 kmem_cache_destroy(*slab); 2086 *slab = NULL; 2087 } 2088 2089 void 2090 nfsd4_free_slabs(void) 2091 { 2092 nfsd4_free_slab(&stateowner_slab); 2093 nfsd4_free_slab(&file_slab); 2094 nfsd4_free_slab(&stateid_slab); 2095 nfsd4_free_slab(&deleg_slab); 2096 } 2097 2098 static int 2099 nfsd4_init_slabs(void) 2100 { 2101 stateowner_slab = kmem_cache_create("nfsd4_stateowners", 2102 sizeof(struct nfs4_stateowner), 0, 0, NULL); 2103 if (stateowner_slab == NULL) 2104 goto out_nomem; 2105 file_slab = kmem_cache_create("nfsd4_files", 2106 sizeof(struct nfs4_file), 0, 0, NULL); 2107 if (file_slab == NULL) 2108 goto out_nomem; 2109 stateid_slab = kmem_cache_create("nfsd4_stateids", 2110 sizeof(struct nfs4_stateid), 0, 0, NULL); 2111 if (stateid_slab == NULL) 2112 goto out_nomem; 2113 deleg_slab = kmem_cache_create("nfsd4_delegations", 2114 sizeof(struct nfs4_delegation), 0, 0, NULL); 2115 if (deleg_slab == NULL) 2116 goto out_nomem; 2117 return 0; 2118 out_nomem: 2119 nfsd4_free_slabs(); 2120 dprintk("nfsd4: out of memory while initializing nfsv4\n"); 2121 return -ENOMEM; 2122 } 2123 2124 void 2125 nfs4_free_stateowner(struct kref *kref) 2126 { 2127 struct nfs4_stateowner *sop = 2128 container_of(kref, struct nfs4_stateowner, so_ref); 2129 kfree(sop->so_owner.data); 2130 kmem_cache_free(stateowner_slab, sop); 2131 } 2132 2133 static inline struct nfs4_stateowner * 2134 alloc_stateowner(struct xdr_netobj *owner) 2135 { 2136 struct nfs4_stateowner *sop; 2137 2138 if ((sop = kmem_cache_alloc(stateowner_slab, GFP_KERNEL))) { 2139 if ((sop->so_owner.data = kmalloc(owner->len, GFP_KERNEL))) { 2140 memcpy(sop->so_owner.data, owner->data, owner->len); 2141 sop->so_owner.len = owner->len; 2142 kref_init(&sop->so_ref); 2143 return sop; 2144 } 2145 kmem_cache_free(stateowner_slab, sop); 2146 } 2147 return NULL; 2148 } 2149 2150 static struct nfs4_stateowner * 2151 alloc_init_open_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfsd4_open *open) { 2152 struct nfs4_stateowner *sop; 2153 struct nfs4_replay *rp; 2154 unsigned int idhashval; 2155 2156 if (!(sop = alloc_stateowner(&open->op_owner))) 2157 return NULL; 2158 idhashval = ownerid_hashval(current_ownerid); 2159 INIT_LIST_HEAD(&sop->so_idhash); 2160 INIT_LIST_HEAD(&sop->so_strhash); 2161 INIT_LIST_HEAD(&sop->so_perclient); 2162 INIT_LIST_HEAD(&sop->so_stateids); 2163 INIT_LIST_HEAD(&sop->so_perstateid); /* not used */ 2164 INIT_LIST_HEAD(&sop->so_close_lru); 2165 sop->so_time = 0; 2166 list_add(&sop->so_idhash, &ownerid_hashtbl[idhashval]); 2167 list_add(&sop->so_strhash, &ownerstr_hashtbl[strhashval]); 2168 list_add(&sop->so_perclient, &clp->cl_openowners); 2169 sop->so_is_open_owner = 1; 2170 sop->so_id = current_ownerid++; 2171 sop->so_client = clp; 2172 sop->so_seqid = open->op_seqid; 2173 sop->so_confirmed = 0; 2174 rp = &sop->so_replay; 2175 rp->rp_status = nfserr_serverfault; 2176 rp->rp_buflen = 0; 2177 rp->rp_buf = rp->rp_ibuf; 2178 return sop; 2179 } 2180 2181 static inline void 2182 init_stateid(struct nfs4_stateid *stp, struct nfs4_file *fp, struct nfsd4_open *open) { 2183 struct nfs4_stateowner *sop = open->op_stateowner; 2184 unsigned int hashval = stateid_hashval(sop->so_id, fp->fi_id); 2185 2186 INIT_LIST_HEAD(&stp->st_hash); 2187 INIT_LIST_HEAD(&stp->st_perstateowner); 2188 INIT_LIST_HEAD(&stp->st_lockowners); 2189 INIT_LIST_HEAD(&stp->st_perfile); 2190 list_add(&stp->st_hash, &stateid_hashtbl[hashval]); 2191 list_add(&stp->st_perstateowner, &sop->so_stateids); 2192 list_add(&stp->st_perfile, &fp->fi_stateids); 2193 stp->st_stateowner = sop; 2194 get_nfs4_file(fp); 2195 stp->st_file = fp; 2196 stp->st_stateid.si_boot = boot_time; 2197 stp->st_stateid.si_stateownerid = sop->so_id; 2198 stp->st_stateid.si_fileid = fp->fi_id; 2199 stp->st_stateid.si_generation = 0; 2200 stp->st_access_bmap = 0; 2201 stp->st_deny_bmap = 0; 2202 __set_bit(open->op_share_access & ~NFS4_SHARE_WANT_MASK, 2203 &stp->st_access_bmap); 2204 __set_bit(open->op_share_deny, &stp->st_deny_bmap); 2205 stp->st_openstp = NULL; 2206 } 2207 2208 static void 2209 move_to_close_lru(struct nfs4_stateowner *sop) 2210 { 2211 dprintk("NFSD: move_to_close_lru nfs4_stateowner %p\n", sop); 2212 2213 list_move_tail(&sop->so_close_lru, &close_lru); 2214 sop->so_time = get_seconds(); 2215 } 2216 2217 static int 2218 same_owner_str(struct nfs4_stateowner *sop, struct xdr_netobj *owner, 2219 clientid_t *clid) 2220 { 2221 return (sop->so_owner.len == owner->len) && 2222 0 == memcmp(sop->so_owner.data, owner->data, owner->len) && 2223 (sop->so_client->cl_clientid.cl_id == clid->cl_id); 2224 } 2225 2226 static struct nfs4_stateowner * 2227 find_openstateowner_str(unsigned int hashval, struct nfsd4_open *open) 2228 { 2229 struct nfs4_stateowner *so = NULL; 2230 2231 list_for_each_entry(so, &ownerstr_hashtbl[hashval], so_strhash) { 2232 if (same_owner_str(so, &open->op_owner, &open->op_clientid)) 2233 return so; 2234 } 2235 return NULL; 2236 } 2237 2238 /* search file_hashtbl[] for file */ 2239 static struct nfs4_file * 2240 find_file(struct inode *ino) 2241 { 2242 unsigned int hashval = file_hashval(ino); 2243 struct nfs4_file *fp; 2244 2245 spin_lock(&recall_lock); 2246 list_for_each_entry(fp, &file_hashtbl[hashval], fi_hash) { 2247 if (fp->fi_inode == ino) { 2248 get_nfs4_file(fp); 2249 spin_unlock(&recall_lock); 2250 return fp; 2251 } 2252 } 2253 spin_unlock(&recall_lock); 2254 return NULL; 2255 } 2256 2257 static inline int access_valid(u32 x, u32 minorversion) 2258 { 2259 if ((x & NFS4_SHARE_ACCESS_MASK) < NFS4_SHARE_ACCESS_READ) 2260 return 0; 2261 if ((x & NFS4_SHARE_ACCESS_MASK) > NFS4_SHARE_ACCESS_BOTH) 2262 return 0; 2263 x &= ~NFS4_SHARE_ACCESS_MASK; 2264 if (minorversion && x) { 2265 if ((x & NFS4_SHARE_WANT_MASK) > NFS4_SHARE_WANT_CANCEL) 2266 return 0; 2267 if ((x & NFS4_SHARE_WHEN_MASK) > NFS4_SHARE_PUSH_DELEG_WHEN_UNCONTENDED) 2268 return 0; 2269 x &= ~(NFS4_SHARE_WANT_MASK | NFS4_SHARE_WHEN_MASK); 2270 } 2271 if (x) 2272 return 0; 2273 return 1; 2274 } 2275 2276 static inline int deny_valid(u32 x) 2277 { 2278 /* Note: unlike access bits, deny bits may be zero. */ 2279 return x <= NFS4_SHARE_DENY_BOTH; 2280 } 2281 2282 /* 2283 * Called to check deny when READ with all zero stateid or 2284 * WRITE with all zero or all one stateid 2285 */ 2286 static __be32 2287 nfs4_share_conflict(struct svc_fh *current_fh, unsigned int deny_type) 2288 { 2289 struct inode *ino = current_fh->fh_dentry->d_inode; 2290 struct nfs4_file *fp; 2291 struct nfs4_stateid *stp; 2292 __be32 ret; 2293 2294 dprintk("NFSD: nfs4_share_conflict\n"); 2295 2296 fp = find_file(ino); 2297 if (!fp) 2298 return nfs_ok; 2299 ret = nfserr_locked; 2300 /* Search for conflicting share reservations */ 2301 list_for_each_entry(stp, &fp->fi_stateids, st_perfile) { 2302 if (test_bit(deny_type, &stp->st_deny_bmap) || 2303 test_bit(NFS4_SHARE_DENY_BOTH, &stp->st_deny_bmap)) 2304 goto out; 2305 } 2306 ret = nfs_ok; 2307 out: 2308 put_nfs4_file(fp); 2309 return ret; 2310 } 2311 2312 static inline void 2313 nfs4_file_downgrade(struct nfs4_file *fp, unsigned int share_access) 2314 { 2315 if (share_access & NFS4_SHARE_ACCESS_WRITE) 2316 nfs4_file_put_access(fp, O_WRONLY); 2317 if (share_access & NFS4_SHARE_ACCESS_READ) 2318 nfs4_file_put_access(fp, O_RDONLY); 2319 } 2320 2321 static void nfsd_break_one_deleg(struct nfs4_delegation *dp) 2322 { 2323 /* We're assuming the state code never drops its reference 2324 * without first removing the lease. Since we're in this lease 2325 * callback (and since the lease code is serialized by the kernel 2326 * lock) we know the server hasn't removed the lease yet, we know 2327 * it's safe to take a reference: */ 2328 atomic_inc(&dp->dl_count); 2329 2330 list_add_tail(&dp->dl_recall_lru, &del_recall_lru); 2331 2332 /* only place dl_time is set. protected by lock_flocks*/ 2333 dp->dl_time = get_seconds(); 2334 2335 nfsd4_cb_recall(dp); 2336 } 2337 2338 /* Called from break_lease() with lock_flocks() held. */ 2339 static void nfsd_break_deleg_cb(struct file_lock *fl) 2340 { 2341 struct nfs4_file *fp = (struct nfs4_file *)fl->fl_owner; 2342 struct nfs4_delegation *dp; 2343 2344 BUG_ON(!fp); 2345 /* We assume break_lease is only called once per lease: */ 2346 BUG_ON(fp->fi_had_conflict); 2347 /* 2348 * We don't want the locks code to timeout the lease for us; 2349 * we'll remove it ourself if a delegation isn't returned 2350 * in time: 2351 */ 2352 fl->fl_break_time = 0; 2353 2354 spin_lock(&recall_lock); 2355 fp->fi_had_conflict = true; 2356 list_for_each_entry(dp, &fp->fi_delegations, dl_perfile) 2357 nfsd_break_one_deleg(dp); 2358 spin_unlock(&recall_lock); 2359 } 2360 2361 static 2362 int nfsd_change_deleg_cb(struct file_lock **onlist, int arg) 2363 { 2364 if (arg & F_UNLCK) 2365 return lease_modify(onlist, arg); 2366 else 2367 return -EAGAIN; 2368 } 2369 2370 static const struct lock_manager_operations nfsd_lease_mng_ops = { 2371 .fl_break = nfsd_break_deleg_cb, 2372 .fl_change = nfsd_change_deleg_cb, 2373 }; 2374 2375 2376 __be32 2377 nfsd4_process_open1(struct nfsd4_compound_state *cstate, 2378 struct nfsd4_open *open) 2379 { 2380 clientid_t *clientid = &open->op_clientid; 2381 struct nfs4_client *clp = NULL; 2382 unsigned int strhashval; 2383 struct nfs4_stateowner *sop = NULL; 2384 2385 if (!check_name(open->op_owner)) 2386 return nfserr_inval; 2387 2388 if (STALE_CLIENTID(&open->op_clientid)) 2389 return nfserr_stale_clientid; 2390 2391 strhashval = ownerstr_hashval(clientid->cl_id, open->op_owner); 2392 sop = find_openstateowner_str(strhashval, open); 2393 open->op_stateowner = sop; 2394 if (!sop) { 2395 /* Make sure the client's lease hasn't expired. */ 2396 clp = find_confirmed_client(clientid); 2397 if (clp == NULL) 2398 return nfserr_expired; 2399 goto renew; 2400 } 2401 /* When sessions are used, skip open sequenceid processing */ 2402 if (nfsd4_has_session(cstate)) 2403 goto renew; 2404 if (!sop->so_confirmed) { 2405 /* Replace unconfirmed owners without checking for replay. */ 2406 clp = sop->so_client; 2407 release_openowner(sop); 2408 open->op_stateowner = NULL; 2409 goto renew; 2410 } 2411 if (open->op_seqid == sop->so_seqid - 1) { 2412 if (sop->so_replay.rp_buflen) 2413 return nfserr_replay_me; 2414 /* The original OPEN failed so spectacularly 2415 * that we don't even have replay data saved! 2416 * Therefore, we have no choice but to continue 2417 * processing this OPEN; presumably, we'll 2418 * fail again for the same reason. 2419 */ 2420 dprintk("nfsd4_process_open1: replay with no replay cache\n"); 2421 goto renew; 2422 } 2423 if (open->op_seqid != sop->so_seqid) 2424 return nfserr_bad_seqid; 2425 renew: 2426 if (open->op_stateowner == NULL) { 2427 sop = alloc_init_open_stateowner(strhashval, clp, open); 2428 if (sop == NULL) 2429 return nfserr_resource; 2430 open->op_stateowner = sop; 2431 } 2432 list_del_init(&sop->so_close_lru); 2433 renew_client(sop->so_client); 2434 return nfs_ok; 2435 } 2436 2437 static inline __be32 2438 nfs4_check_delegmode(struct nfs4_delegation *dp, int flags) 2439 { 2440 if ((flags & WR_STATE) && (dp->dl_type == NFS4_OPEN_DELEGATE_READ)) 2441 return nfserr_openmode; 2442 else 2443 return nfs_ok; 2444 } 2445 2446 static struct nfs4_delegation * 2447 find_delegation_file(struct nfs4_file *fp, stateid_t *stid) 2448 { 2449 struct nfs4_delegation *dp; 2450 2451 spin_lock(&recall_lock); 2452 list_for_each_entry(dp, &fp->fi_delegations, dl_perfile) 2453 if (dp->dl_stateid.si_stateownerid == stid->si_stateownerid) { 2454 spin_unlock(&recall_lock); 2455 return dp; 2456 } 2457 spin_unlock(&recall_lock); 2458 return NULL; 2459 } 2460 2461 int share_access_to_flags(u32 share_access) 2462 { 2463 share_access &= ~NFS4_SHARE_WANT_MASK; 2464 2465 return share_access == NFS4_SHARE_ACCESS_READ ? RD_STATE : WR_STATE; 2466 } 2467 2468 static __be32 2469 nfs4_check_deleg(struct nfs4_file *fp, struct nfsd4_open *open, 2470 struct nfs4_delegation **dp) 2471 { 2472 int flags; 2473 __be32 status = nfserr_bad_stateid; 2474 2475 *dp = find_delegation_file(fp, &open->op_delegate_stateid); 2476 if (*dp == NULL) 2477 goto out; 2478 flags = share_access_to_flags(open->op_share_access); 2479 status = nfs4_check_delegmode(*dp, flags); 2480 if (status) 2481 *dp = NULL; 2482 out: 2483 if (open->op_claim_type != NFS4_OPEN_CLAIM_DELEGATE_CUR) 2484 return nfs_ok; 2485 if (status) 2486 return status; 2487 open->op_stateowner->so_confirmed = 1; 2488 return nfs_ok; 2489 } 2490 2491 static __be32 2492 nfs4_check_open(struct nfs4_file *fp, struct nfsd4_open *open, struct nfs4_stateid **stpp) 2493 { 2494 struct nfs4_stateid *local; 2495 __be32 status = nfserr_share_denied; 2496 struct nfs4_stateowner *sop = open->op_stateowner; 2497 2498 list_for_each_entry(local, &fp->fi_stateids, st_perfile) { 2499 /* ignore lock owners */ 2500 if (local->st_stateowner->so_is_open_owner == 0) 2501 continue; 2502 /* remember if we have seen this open owner */ 2503 if (local->st_stateowner == sop) 2504 *stpp = local; 2505 /* check for conflicting share reservations */ 2506 if (!test_share(local, open)) 2507 goto out; 2508 } 2509 status = 0; 2510 out: 2511 return status; 2512 } 2513 2514 static inline struct nfs4_stateid * 2515 nfs4_alloc_stateid(void) 2516 { 2517 return kmem_cache_alloc(stateid_slab, GFP_KERNEL); 2518 } 2519 2520 static inline int nfs4_access_to_access(u32 nfs4_access) 2521 { 2522 int flags = 0; 2523 2524 if (nfs4_access & NFS4_SHARE_ACCESS_READ) 2525 flags |= NFSD_MAY_READ; 2526 if (nfs4_access & NFS4_SHARE_ACCESS_WRITE) 2527 flags |= NFSD_MAY_WRITE; 2528 return flags; 2529 } 2530 2531 static __be32 nfs4_get_vfs_file(struct svc_rqst *rqstp, struct nfs4_file 2532 *fp, struct svc_fh *cur_fh, u32 nfs4_access) 2533 { 2534 __be32 status; 2535 int oflag = nfs4_access_to_omode(nfs4_access); 2536 int access = nfs4_access_to_access(nfs4_access); 2537 2538 if (!fp->fi_fds[oflag]) { 2539 status = nfsd_open(rqstp, cur_fh, S_IFREG, access, 2540 &fp->fi_fds[oflag]); 2541 if (status) 2542 return status; 2543 } 2544 nfs4_file_get_access(fp, oflag); 2545 2546 return nfs_ok; 2547 } 2548 2549 static __be32 2550 nfs4_new_open(struct svc_rqst *rqstp, struct nfs4_stateid **stpp, 2551 struct nfs4_file *fp, struct svc_fh *cur_fh, 2552 struct nfsd4_open *open) 2553 { 2554 struct nfs4_stateid *stp; 2555 __be32 status; 2556 2557 stp = nfs4_alloc_stateid(); 2558 if (stp == NULL) 2559 return nfserr_resource; 2560 2561 status = nfs4_get_vfs_file(rqstp, fp, cur_fh, open->op_share_access); 2562 if (status) { 2563 kmem_cache_free(stateid_slab, stp); 2564 return status; 2565 } 2566 *stpp = stp; 2567 return 0; 2568 } 2569 2570 static inline __be32 2571 nfsd4_truncate(struct svc_rqst *rqstp, struct svc_fh *fh, 2572 struct nfsd4_open *open) 2573 { 2574 struct iattr iattr = { 2575 .ia_valid = ATTR_SIZE, 2576 .ia_size = 0, 2577 }; 2578 if (!open->op_truncate) 2579 return 0; 2580 if (!(open->op_share_access & NFS4_SHARE_ACCESS_WRITE)) 2581 return nfserr_inval; 2582 return nfsd_setattr(rqstp, fh, &iattr, 0, (time_t)0); 2583 } 2584 2585 static __be32 2586 nfs4_upgrade_open(struct svc_rqst *rqstp, struct nfs4_file *fp, struct svc_fh *cur_fh, struct nfs4_stateid *stp, struct nfsd4_open *open) 2587 { 2588 u32 op_share_access = open->op_share_access & ~NFS4_SHARE_WANT_MASK; 2589 bool new_access; 2590 __be32 status; 2591 2592 new_access = !test_bit(op_share_access, &stp->st_access_bmap); 2593 if (new_access) { 2594 status = nfs4_get_vfs_file(rqstp, fp, cur_fh, op_share_access); 2595 if (status) 2596 return status; 2597 } 2598 status = nfsd4_truncate(rqstp, cur_fh, open); 2599 if (status) { 2600 if (new_access) { 2601 int oflag = nfs4_access_to_omode(new_access); 2602 nfs4_file_put_access(fp, oflag); 2603 } 2604 return status; 2605 } 2606 /* remember the open */ 2607 __set_bit(op_share_access, &stp->st_access_bmap); 2608 __set_bit(open->op_share_deny, &stp->st_deny_bmap); 2609 2610 return nfs_ok; 2611 } 2612 2613 2614 static void 2615 nfs4_set_claim_prev(struct nfsd4_open *open) 2616 { 2617 open->op_stateowner->so_confirmed = 1; 2618 open->op_stateowner->so_client->cl_firststate = 1; 2619 } 2620 2621 /* Should we give out recallable state?: */ 2622 static bool nfsd4_cb_channel_good(struct nfs4_client *clp) 2623 { 2624 if (clp->cl_cb_state == NFSD4_CB_UP) 2625 return true; 2626 /* 2627 * In the sessions case, since we don't have to establish a 2628 * separate connection for callbacks, we assume it's OK 2629 * until we hear otherwise: 2630 */ 2631 return clp->cl_minorversion && clp->cl_cb_state == NFSD4_CB_UNKNOWN; 2632 } 2633 2634 static struct file_lock *nfs4_alloc_init_lease(struct nfs4_delegation *dp, int flag) 2635 { 2636 struct file_lock *fl; 2637 2638 fl = locks_alloc_lock(); 2639 if (!fl) 2640 return NULL; 2641 locks_init_lock(fl); 2642 fl->fl_lmops = &nfsd_lease_mng_ops; 2643 fl->fl_flags = FL_LEASE; 2644 fl->fl_type = flag == NFS4_OPEN_DELEGATE_READ? F_RDLCK: F_WRLCK; 2645 fl->fl_end = OFFSET_MAX; 2646 fl->fl_owner = (fl_owner_t)(dp->dl_file); 2647 fl->fl_pid = current->tgid; 2648 return fl; 2649 } 2650 2651 static int nfs4_setlease(struct nfs4_delegation *dp, int flag) 2652 { 2653 struct nfs4_file *fp = dp->dl_file; 2654 struct file_lock *fl; 2655 int status; 2656 2657 fl = nfs4_alloc_init_lease(dp, flag); 2658 if (!fl) 2659 return -ENOMEM; 2660 fl->fl_file = find_readable_file(fp); 2661 list_add(&dp->dl_perclnt, &dp->dl_client->cl_delegations); 2662 status = vfs_setlease(fl->fl_file, fl->fl_type, &fl); 2663 if (status) { 2664 list_del_init(&dp->dl_perclnt); 2665 locks_free_lock(fl); 2666 return -ENOMEM; 2667 } 2668 fp->fi_lease = fl; 2669 fp->fi_deleg_file = fl->fl_file; 2670 get_file(fp->fi_deleg_file); 2671 atomic_set(&fp->fi_delegees, 1); 2672 list_add(&dp->dl_perfile, &fp->fi_delegations); 2673 return 0; 2674 } 2675 2676 static int nfs4_set_delegation(struct nfs4_delegation *dp, int flag) 2677 { 2678 struct nfs4_file *fp = dp->dl_file; 2679 2680 if (!fp->fi_lease) 2681 return nfs4_setlease(dp, flag); 2682 spin_lock(&recall_lock); 2683 if (fp->fi_had_conflict) { 2684 spin_unlock(&recall_lock); 2685 return -EAGAIN; 2686 } 2687 atomic_inc(&fp->fi_delegees); 2688 list_add(&dp->dl_perfile, &fp->fi_delegations); 2689 spin_unlock(&recall_lock); 2690 list_add(&dp->dl_perclnt, &dp->dl_client->cl_delegations); 2691 return 0; 2692 } 2693 2694 /* 2695 * Attempt to hand out a delegation. 2696 */ 2697 static void 2698 nfs4_open_delegation(struct svc_fh *fh, struct nfsd4_open *open, struct nfs4_stateid *stp) 2699 { 2700 struct nfs4_delegation *dp; 2701 struct nfs4_stateowner *sop = stp->st_stateowner; 2702 int cb_up; 2703 int status, flag = 0; 2704 2705 cb_up = nfsd4_cb_channel_good(sop->so_client); 2706 flag = NFS4_OPEN_DELEGATE_NONE; 2707 open->op_recall = 0; 2708 switch (open->op_claim_type) { 2709 case NFS4_OPEN_CLAIM_PREVIOUS: 2710 if (!cb_up) 2711 open->op_recall = 1; 2712 flag = open->op_delegate_type; 2713 if (flag == NFS4_OPEN_DELEGATE_NONE) 2714 goto out; 2715 break; 2716 case NFS4_OPEN_CLAIM_NULL: 2717 /* Let's not give out any delegations till everyone's 2718 * had the chance to reclaim theirs.... */ 2719 if (locks_in_grace()) 2720 goto out; 2721 if (!cb_up || !sop->so_confirmed) 2722 goto out; 2723 if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE) 2724 flag = NFS4_OPEN_DELEGATE_WRITE; 2725 else 2726 flag = NFS4_OPEN_DELEGATE_READ; 2727 break; 2728 default: 2729 goto out; 2730 } 2731 2732 dp = alloc_init_deleg(sop->so_client, stp, fh, flag); 2733 if (dp == NULL) 2734 goto out_no_deleg; 2735 status = nfs4_set_delegation(dp, flag); 2736 if (status) 2737 goto out_free; 2738 2739 memcpy(&open->op_delegate_stateid, &dp->dl_stateid, sizeof(dp->dl_stateid)); 2740 2741 dprintk("NFSD: delegation stateid=" STATEID_FMT "\n", 2742 STATEID_VAL(&dp->dl_stateid)); 2743 out: 2744 if (open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS 2745 && flag == NFS4_OPEN_DELEGATE_NONE 2746 && open->op_delegate_type != NFS4_OPEN_DELEGATE_NONE) 2747 dprintk("NFSD: WARNING: refusing delegation reclaim\n"); 2748 open->op_delegate_type = flag; 2749 return; 2750 out_free: 2751 nfs4_put_delegation(dp); 2752 out_no_deleg: 2753 flag = NFS4_OPEN_DELEGATE_NONE; 2754 goto out; 2755 } 2756 2757 /* 2758 * called with nfs4_lock_state() held. 2759 */ 2760 __be32 2761 nfsd4_process_open2(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_open *open) 2762 { 2763 struct nfsd4_compoundres *resp = rqstp->rq_resp; 2764 struct nfs4_file *fp = NULL; 2765 struct inode *ino = current_fh->fh_dentry->d_inode; 2766 struct nfs4_stateid *stp = NULL; 2767 struct nfs4_delegation *dp = NULL; 2768 __be32 status; 2769 2770 status = nfserr_inval; 2771 if (!access_valid(open->op_share_access, resp->cstate.minorversion) 2772 || !deny_valid(open->op_share_deny)) 2773 goto out; 2774 /* 2775 * Lookup file; if found, lookup stateid and check open request, 2776 * and check for delegations in the process of being recalled. 2777 * If not found, create the nfs4_file struct 2778 */ 2779 fp = find_file(ino); 2780 if (fp) { 2781 if ((status = nfs4_check_open(fp, open, &stp))) 2782 goto out; 2783 status = nfs4_check_deleg(fp, open, &dp); 2784 if (status) 2785 goto out; 2786 } else { 2787 status = nfserr_bad_stateid; 2788 if (open->op_claim_type == NFS4_OPEN_CLAIM_DELEGATE_CUR) 2789 goto out; 2790 status = nfserr_resource; 2791 fp = alloc_init_file(ino); 2792 if (fp == NULL) 2793 goto out; 2794 } 2795 2796 /* 2797 * OPEN the file, or upgrade an existing OPEN. 2798 * If truncate fails, the OPEN fails. 2799 */ 2800 if (stp) { 2801 /* Stateid was found, this is an OPEN upgrade */ 2802 status = nfs4_upgrade_open(rqstp, fp, current_fh, stp, open); 2803 if (status) 2804 goto out; 2805 update_stateid(&stp->st_stateid); 2806 } else { 2807 status = nfs4_new_open(rqstp, &stp, fp, current_fh, open); 2808 if (status) 2809 goto out; 2810 init_stateid(stp, fp, open); 2811 status = nfsd4_truncate(rqstp, current_fh, open); 2812 if (status) { 2813 release_open_stateid(stp); 2814 goto out; 2815 } 2816 if (nfsd4_has_session(&resp->cstate)) 2817 update_stateid(&stp->st_stateid); 2818 } 2819 memcpy(&open->op_stateid, &stp->st_stateid, sizeof(stateid_t)); 2820 2821 if (nfsd4_has_session(&resp->cstate)) 2822 open->op_stateowner->so_confirmed = 1; 2823 2824 /* 2825 * Attempt to hand out a delegation. No error return, because the 2826 * OPEN succeeds even if we fail. 2827 */ 2828 nfs4_open_delegation(current_fh, open, stp); 2829 2830 status = nfs_ok; 2831 2832 dprintk("%s: stateid=" STATEID_FMT "\n", __func__, 2833 STATEID_VAL(&stp->st_stateid)); 2834 out: 2835 if (fp) 2836 put_nfs4_file(fp); 2837 if (status == 0 && open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS) 2838 nfs4_set_claim_prev(open); 2839 /* 2840 * To finish the open response, we just need to set the rflags. 2841 */ 2842 open->op_rflags = NFS4_OPEN_RESULT_LOCKTYPE_POSIX; 2843 if (!open->op_stateowner->so_confirmed && 2844 !nfsd4_has_session(&resp->cstate)) 2845 open->op_rflags |= NFS4_OPEN_RESULT_CONFIRM; 2846 2847 return status; 2848 } 2849 2850 __be32 2851 nfsd4_renew(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate, 2852 clientid_t *clid) 2853 { 2854 struct nfs4_client *clp; 2855 __be32 status; 2856 2857 nfs4_lock_state(); 2858 dprintk("process_renew(%08x/%08x): starting\n", 2859 clid->cl_boot, clid->cl_id); 2860 status = nfserr_stale_clientid; 2861 if (STALE_CLIENTID(clid)) 2862 goto out; 2863 clp = find_confirmed_client(clid); 2864 status = nfserr_expired; 2865 if (clp == NULL) { 2866 /* We assume the client took too long to RENEW. */ 2867 dprintk("nfsd4_renew: clientid not found!\n"); 2868 goto out; 2869 } 2870 renew_client(clp); 2871 status = nfserr_cb_path_down; 2872 if (!list_empty(&clp->cl_delegations) 2873 && clp->cl_cb_state != NFSD4_CB_UP) 2874 goto out; 2875 status = nfs_ok; 2876 out: 2877 nfs4_unlock_state(); 2878 return status; 2879 } 2880 2881 struct lock_manager nfsd4_manager = { 2882 }; 2883 2884 static void 2885 nfsd4_end_grace(void) 2886 { 2887 dprintk("NFSD: end of grace period\n"); 2888 nfsd4_recdir_purge_old(); 2889 locks_end_grace(&nfsd4_manager); 2890 /* 2891 * Now that every NFSv4 client has had the chance to recover and 2892 * to see the (possibly new, possibly shorter) lease time, we 2893 * can safely set the next grace time to the current lease time: 2894 */ 2895 nfsd4_grace = nfsd4_lease; 2896 } 2897 2898 static time_t 2899 nfs4_laundromat(void) 2900 { 2901 struct nfs4_client *clp; 2902 struct nfs4_stateowner *sop; 2903 struct nfs4_delegation *dp; 2904 struct list_head *pos, *next, reaplist; 2905 time_t cutoff = get_seconds() - nfsd4_lease; 2906 time_t t, clientid_val = nfsd4_lease; 2907 time_t u, test_val = nfsd4_lease; 2908 2909 nfs4_lock_state(); 2910 2911 dprintk("NFSD: laundromat service - starting\n"); 2912 if (locks_in_grace()) 2913 nfsd4_end_grace(); 2914 INIT_LIST_HEAD(&reaplist); 2915 spin_lock(&client_lock); 2916 list_for_each_safe(pos, next, &client_lru) { 2917 clp = list_entry(pos, struct nfs4_client, cl_lru); 2918 if (time_after((unsigned long)clp->cl_time, (unsigned long)cutoff)) { 2919 t = clp->cl_time - cutoff; 2920 if (clientid_val > t) 2921 clientid_val = t; 2922 break; 2923 } 2924 if (atomic_read(&clp->cl_refcount)) { 2925 dprintk("NFSD: client in use (clientid %08x)\n", 2926 clp->cl_clientid.cl_id); 2927 continue; 2928 } 2929 unhash_client_locked(clp); 2930 list_add(&clp->cl_lru, &reaplist); 2931 } 2932 spin_unlock(&client_lock); 2933 list_for_each_safe(pos, next, &reaplist) { 2934 clp = list_entry(pos, struct nfs4_client, cl_lru); 2935 dprintk("NFSD: purging unused client (clientid %08x)\n", 2936 clp->cl_clientid.cl_id); 2937 nfsd4_remove_clid_dir(clp); 2938 expire_client(clp); 2939 } 2940 spin_lock(&recall_lock); 2941 list_for_each_safe(pos, next, &del_recall_lru) { 2942 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru); 2943 if (time_after((unsigned long)dp->dl_time, (unsigned long)cutoff)) { 2944 u = dp->dl_time - cutoff; 2945 if (test_val > u) 2946 test_val = u; 2947 break; 2948 } 2949 list_move(&dp->dl_recall_lru, &reaplist); 2950 } 2951 spin_unlock(&recall_lock); 2952 list_for_each_safe(pos, next, &reaplist) { 2953 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru); 2954 list_del_init(&dp->dl_recall_lru); 2955 unhash_delegation(dp); 2956 } 2957 test_val = nfsd4_lease; 2958 list_for_each_safe(pos, next, &close_lru) { 2959 sop = list_entry(pos, struct nfs4_stateowner, so_close_lru); 2960 if (time_after((unsigned long)sop->so_time, (unsigned long)cutoff)) { 2961 u = sop->so_time - cutoff; 2962 if (test_val > u) 2963 test_val = u; 2964 break; 2965 } 2966 dprintk("NFSD: purging unused open stateowner (so_id %d)\n", 2967 sop->so_id); 2968 release_openowner(sop); 2969 } 2970 if (clientid_val < NFSD_LAUNDROMAT_MINTIMEOUT) 2971 clientid_val = NFSD_LAUNDROMAT_MINTIMEOUT; 2972 nfs4_unlock_state(); 2973 return clientid_val; 2974 } 2975 2976 static struct workqueue_struct *laundry_wq; 2977 static void laundromat_main(struct work_struct *); 2978 static DECLARE_DELAYED_WORK(laundromat_work, laundromat_main); 2979 2980 static void 2981 laundromat_main(struct work_struct *not_used) 2982 { 2983 time_t t; 2984 2985 t = nfs4_laundromat(); 2986 dprintk("NFSD: laundromat_main - sleeping for %ld seconds\n", t); 2987 queue_delayed_work(laundry_wq, &laundromat_work, t*HZ); 2988 } 2989 2990 static struct nfs4_stateowner * 2991 search_close_lru(u32 st_id, int flags) 2992 { 2993 struct nfs4_stateowner *local = NULL; 2994 2995 if (flags & CLOSE_STATE) { 2996 list_for_each_entry(local, &close_lru, so_close_lru) { 2997 if (local->so_id == st_id) 2998 return local; 2999 } 3000 } 3001 return NULL; 3002 } 3003 3004 static inline int 3005 nfs4_check_fh(struct svc_fh *fhp, struct nfs4_stateid *stp) 3006 { 3007 return fhp->fh_dentry->d_inode != stp->st_file->fi_inode; 3008 } 3009 3010 static int 3011 STALE_STATEID(stateid_t *stateid) 3012 { 3013 if (stateid->si_boot == boot_time) 3014 return 0; 3015 dprintk("NFSD: stale stateid " STATEID_FMT "!\n", 3016 STATEID_VAL(stateid)); 3017 return 1; 3018 } 3019 3020 static inline int 3021 access_permit_read(unsigned long access_bmap) 3022 { 3023 return test_bit(NFS4_SHARE_ACCESS_READ, &access_bmap) || 3024 test_bit(NFS4_SHARE_ACCESS_BOTH, &access_bmap) || 3025 test_bit(NFS4_SHARE_ACCESS_WRITE, &access_bmap); 3026 } 3027 3028 static inline int 3029 access_permit_write(unsigned long access_bmap) 3030 { 3031 return test_bit(NFS4_SHARE_ACCESS_WRITE, &access_bmap) || 3032 test_bit(NFS4_SHARE_ACCESS_BOTH, &access_bmap); 3033 } 3034 3035 static 3036 __be32 nfs4_check_openmode(struct nfs4_stateid *stp, int flags) 3037 { 3038 __be32 status = nfserr_openmode; 3039 3040 /* For lock stateid's, we test the parent open, not the lock: */ 3041 if (stp->st_openstp) 3042 stp = stp->st_openstp; 3043 if ((flags & WR_STATE) && (!access_permit_write(stp->st_access_bmap))) 3044 goto out; 3045 if ((flags & RD_STATE) && (!access_permit_read(stp->st_access_bmap))) 3046 goto out; 3047 status = nfs_ok; 3048 out: 3049 return status; 3050 } 3051 3052 static inline __be32 3053 check_special_stateids(svc_fh *current_fh, stateid_t *stateid, int flags) 3054 { 3055 if (ONE_STATEID(stateid) && (flags & RD_STATE)) 3056 return nfs_ok; 3057 else if (locks_in_grace()) { 3058 /* Answer in remaining cases depends on existance of 3059 * conflicting state; so we must wait out the grace period. */ 3060 return nfserr_grace; 3061 } else if (flags & WR_STATE) 3062 return nfs4_share_conflict(current_fh, 3063 NFS4_SHARE_DENY_WRITE); 3064 else /* (flags & RD_STATE) && ZERO_STATEID(stateid) */ 3065 return nfs4_share_conflict(current_fh, 3066 NFS4_SHARE_DENY_READ); 3067 } 3068 3069 /* 3070 * Allow READ/WRITE during grace period on recovered state only for files 3071 * that are not able to provide mandatory locking. 3072 */ 3073 static inline int 3074 grace_disallows_io(struct inode *inode) 3075 { 3076 return locks_in_grace() && mandatory_lock(inode); 3077 } 3078 3079 static int check_stateid_generation(stateid_t *in, stateid_t *ref, int flags) 3080 { 3081 /* 3082 * When sessions are used the stateid generation number is ignored 3083 * when it is zero. 3084 */ 3085 if ((flags & HAS_SESSION) && in->si_generation == 0) 3086 goto out; 3087 3088 /* If the client sends us a stateid from the future, it's buggy: */ 3089 if (in->si_generation > ref->si_generation) 3090 return nfserr_bad_stateid; 3091 /* 3092 * The following, however, can happen. For example, if the 3093 * client sends an open and some IO at the same time, the open 3094 * may bump si_generation while the IO is still in flight. 3095 * Thanks to hard links and renames, the client never knows what 3096 * file an open will affect. So it could avoid that situation 3097 * only by serializing all opens and IO from the same open 3098 * owner. To recover from the old_stateid error, the client 3099 * will just have to retry the IO: 3100 */ 3101 if (in->si_generation < ref->si_generation) 3102 return nfserr_old_stateid; 3103 out: 3104 return nfs_ok; 3105 } 3106 3107 static int is_delegation_stateid(stateid_t *stateid) 3108 { 3109 return stateid->si_fileid == 0; 3110 } 3111 3112 /* 3113 * Checks for stateid operations 3114 */ 3115 __be32 3116 nfs4_preprocess_stateid_op(struct nfsd4_compound_state *cstate, 3117 stateid_t *stateid, int flags, struct file **filpp) 3118 { 3119 struct nfs4_stateid *stp = NULL; 3120 struct nfs4_delegation *dp = NULL; 3121 struct svc_fh *current_fh = &cstate->current_fh; 3122 struct inode *ino = current_fh->fh_dentry->d_inode; 3123 __be32 status; 3124 3125 if (filpp) 3126 *filpp = NULL; 3127 3128 if (grace_disallows_io(ino)) 3129 return nfserr_grace; 3130 3131 if (nfsd4_has_session(cstate)) 3132 flags |= HAS_SESSION; 3133 3134 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid)) 3135 return check_special_stateids(current_fh, stateid, flags); 3136 3137 status = nfserr_stale_stateid; 3138 if (STALE_STATEID(stateid)) 3139 goto out; 3140 3141 /* 3142 * We assume that any stateid that has the current boot time, 3143 * but that we can't find, is expired: 3144 */ 3145 status = nfserr_expired; 3146 if (is_delegation_stateid(stateid)) { 3147 dp = find_delegation_stateid(ino, stateid); 3148 if (!dp) 3149 goto out; 3150 status = check_stateid_generation(stateid, &dp->dl_stateid, 3151 flags); 3152 if (status) 3153 goto out; 3154 status = nfs4_check_delegmode(dp, flags); 3155 if (status) 3156 goto out; 3157 renew_client(dp->dl_client); 3158 if (filpp) { 3159 *filpp = dp->dl_file->fi_deleg_file; 3160 BUG_ON(!*filpp); 3161 } 3162 } else { /* open or lock stateid */ 3163 stp = find_stateid(stateid, flags); 3164 if (!stp) 3165 goto out; 3166 status = nfserr_bad_stateid; 3167 if (nfs4_check_fh(current_fh, stp)) 3168 goto out; 3169 if (!stp->st_stateowner->so_confirmed) 3170 goto out; 3171 status = check_stateid_generation(stateid, &stp->st_stateid, 3172 flags); 3173 if (status) 3174 goto out; 3175 status = nfs4_check_openmode(stp, flags); 3176 if (status) 3177 goto out; 3178 renew_client(stp->st_stateowner->so_client); 3179 if (filpp) { 3180 if (flags & RD_STATE) 3181 *filpp = find_readable_file(stp->st_file); 3182 else 3183 *filpp = find_writeable_file(stp->st_file); 3184 } 3185 } 3186 status = nfs_ok; 3187 out: 3188 return status; 3189 } 3190 3191 static inline int 3192 setlkflg (int type) 3193 { 3194 return (type == NFS4_READW_LT || type == NFS4_READ_LT) ? 3195 RD_STATE : WR_STATE; 3196 } 3197 3198 /* 3199 * Checks for sequence id mutating operations. 3200 */ 3201 static __be32 3202 nfs4_preprocess_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid, 3203 stateid_t *stateid, int flags, 3204 struct nfs4_stateowner **sopp, 3205 struct nfs4_stateid **stpp, struct nfsd4_lock *lock) 3206 { 3207 struct nfs4_stateid *stp; 3208 struct nfs4_stateowner *sop; 3209 struct svc_fh *current_fh = &cstate->current_fh; 3210 __be32 status; 3211 3212 dprintk("NFSD: %s: seqid=%d stateid = " STATEID_FMT "\n", __func__, 3213 seqid, STATEID_VAL(stateid)); 3214 3215 *stpp = NULL; 3216 *sopp = NULL; 3217 3218 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid)) { 3219 dprintk("NFSD: preprocess_seqid_op: magic stateid!\n"); 3220 return nfserr_bad_stateid; 3221 } 3222 3223 if (STALE_STATEID(stateid)) 3224 return nfserr_stale_stateid; 3225 3226 if (nfsd4_has_session(cstate)) 3227 flags |= HAS_SESSION; 3228 3229 /* 3230 * We return BAD_STATEID if filehandle doesn't match stateid, 3231 * the confirmed flag is incorrecly set, or the generation 3232 * number is incorrect. 3233 */ 3234 stp = find_stateid(stateid, flags); 3235 if (stp == NULL) { 3236 /* 3237 * Also, we should make sure this isn't just the result of 3238 * a replayed close: 3239 */ 3240 sop = search_close_lru(stateid->si_stateownerid, flags); 3241 /* It's not stale; let's assume it's expired: */ 3242 if (sop == NULL) 3243 return nfserr_expired; 3244 *sopp = sop; 3245 goto check_replay; 3246 } 3247 3248 *stpp = stp; 3249 *sopp = sop = stp->st_stateowner; 3250 3251 if (lock) { 3252 clientid_t *lockclid = &lock->v.new.clientid; 3253 struct nfs4_client *clp = sop->so_client; 3254 int lkflg = 0; 3255 __be32 status; 3256 3257 lkflg = setlkflg(lock->lk_type); 3258 3259 if (lock->lk_is_new) { 3260 if (!sop->so_is_open_owner) 3261 return nfserr_bad_stateid; 3262 if (!(flags & HAS_SESSION) && 3263 !same_clid(&clp->cl_clientid, lockclid)) 3264 return nfserr_bad_stateid; 3265 /* stp is the open stateid */ 3266 status = nfs4_check_openmode(stp, lkflg); 3267 if (status) 3268 return status; 3269 } else { 3270 /* stp is the lock stateid */ 3271 status = nfs4_check_openmode(stp->st_openstp, lkflg); 3272 if (status) 3273 return status; 3274 } 3275 } 3276 3277 if (nfs4_check_fh(current_fh, stp)) { 3278 dprintk("NFSD: preprocess_seqid_op: fh-stateid mismatch!\n"); 3279 return nfserr_bad_stateid; 3280 } 3281 3282 /* 3283 * We now validate the seqid and stateid generation numbers. 3284 * For the moment, we ignore the possibility of 3285 * generation number wraparound. 3286 */ 3287 if (!(flags & HAS_SESSION) && seqid != sop->so_seqid) 3288 goto check_replay; 3289 3290 if (sop->so_confirmed && flags & CONFIRM) { 3291 dprintk("NFSD: preprocess_seqid_op: expected" 3292 " unconfirmed stateowner!\n"); 3293 return nfserr_bad_stateid; 3294 } 3295 if (!sop->so_confirmed && !(flags & CONFIRM)) { 3296 dprintk("NFSD: preprocess_seqid_op: stateowner not" 3297 " confirmed yet!\n"); 3298 return nfserr_bad_stateid; 3299 } 3300 status = check_stateid_generation(stateid, &stp->st_stateid, flags); 3301 if (status) 3302 return status; 3303 renew_client(sop->so_client); 3304 return nfs_ok; 3305 3306 check_replay: 3307 if (seqid == sop->so_seqid - 1) { 3308 dprintk("NFSD: preprocess_seqid_op: retransmission?\n"); 3309 /* indicate replay to calling function */ 3310 return nfserr_replay_me; 3311 } 3312 dprintk("NFSD: preprocess_seqid_op: bad seqid (expected %d, got %d)\n", 3313 sop->so_seqid, seqid); 3314 *sopp = NULL; 3315 return nfserr_bad_seqid; 3316 } 3317 3318 __be32 3319 nfsd4_open_confirm(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate, 3320 struct nfsd4_open_confirm *oc) 3321 { 3322 __be32 status; 3323 struct nfs4_stateowner *sop; 3324 struct nfs4_stateid *stp; 3325 3326 dprintk("NFSD: nfsd4_open_confirm on file %.*s\n", 3327 (int)cstate->current_fh.fh_dentry->d_name.len, 3328 cstate->current_fh.fh_dentry->d_name.name); 3329 3330 status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0); 3331 if (status) 3332 return status; 3333 3334 nfs4_lock_state(); 3335 3336 if ((status = nfs4_preprocess_seqid_op(cstate, 3337 oc->oc_seqid, &oc->oc_req_stateid, 3338 CONFIRM | OPEN_STATE, 3339 &oc->oc_stateowner, &stp, NULL))) 3340 goto out; 3341 3342 sop = oc->oc_stateowner; 3343 sop->so_confirmed = 1; 3344 update_stateid(&stp->st_stateid); 3345 memcpy(&oc->oc_resp_stateid, &stp->st_stateid, sizeof(stateid_t)); 3346 dprintk("NFSD: %s: success, seqid=%d stateid=" STATEID_FMT "\n", 3347 __func__, oc->oc_seqid, STATEID_VAL(&stp->st_stateid)); 3348 3349 nfsd4_create_clid_dir(sop->so_client); 3350 out: 3351 if (oc->oc_stateowner) { 3352 nfs4_get_stateowner(oc->oc_stateowner); 3353 cstate->replay_owner = oc->oc_stateowner; 3354 } 3355 nfs4_unlock_state(); 3356 return status; 3357 } 3358 3359 3360 /* 3361 * unset all bits in union bitmap (bmap) that 3362 * do not exist in share (from successful OPEN_DOWNGRADE) 3363 */ 3364 static void 3365 reset_union_bmap_access(unsigned long access, unsigned long *bmap) 3366 { 3367 int i; 3368 for (i = 1; i < 4; i++) { 3369 if ((i & access) != i) 3370 __clear_bit(i, bmap); 3371 } 3372 } 3373 3374 static void 3375 reset_union_bmap_deny(unsigned long deny, unsigned long *bmap) 3376 { 3377 int i; 3378 for (i = 0; i < 4; i++) { 3379 if ((i & deny) != i) 3380 __clear_bit(i, bmap); 3381 } 3382 } 3383 3384 __be32 3385 nfsd4_open_downgrade(struct svc_rqst *rqstp, 3386 struct nfsd4_compound_state *cstate, 3387 struct nfsd4_open_downgrade *od) 3388 { 3389 __be32 status; 3390 struct nfs4_stateid *stp; 3391 unsigned int share_access; 3392 3393 dprintk("NFSD: nfsd4_open_downgrade on file %.*s\n", 3394 (int)cstate->current_fh.fh_dentry->d_name.len, 3395 cstate->current_fh.fh_dentry->d_name.name); 3396 3397 if (!access_valid(od->od_share_access, cstate->minorversion) 3398 || !deny_valid(od->od_share_deny)) 3399 return nfserr_inval; 3400 3401 nfs4_lock_state(); 3402 if ((status = nfs4_preprocess_seqid_op(cstate, 3403 od->od_seqid, 3404 &od->od_stateid, 3405 OPEN_STATE, 3406 &od->od_stateowner, &stp, NULL))) 3407 goto out; 3408 3409 status = nfserr_inval; 3410 if (!test_bit(od->od_share_access, &stp->st_access_bmap)) { 3411 dprintk("NFSD:access not a subset current bitmap: 0x%lx, input access=%08x\n", 3412 stp->st_access_bmap, od->od_share_access); 3413 goto out; 3414 } 3415 if (!test_bit(od->od_share_deny, &stp->st_deny_bmap)) { 3416 dprintk("NFSD:deny not a subset current bitmap: 0x%lx, input deny=%08x\n", 3417 stp->st_deny_bmap, od->od_share_deny); 3418 goto out; 3419 } 3420 set_access(&share_access, stp->st_access_bmap); 3421 nfs4_file_downgrade(stp->st_file, share_access & ~od->od_share_access); 3422 3423 reset_union_bmap_access(od->od_share_access, &stp->st_access_bmap); 3424 reset_union_bmap_deny(od->od_share_deny, &stp->st_deny_bmap); 3425 3426 update_stateid(&stp->st_stateid); 3427 memcpy(&od->od_stateid, &stp->st_stateid, sizeof(stateid_t)); 3428 status = nfs_ok; 3429 out: 3430 if (od->od_stateowner) { 3431 nfs4_get_stateowner(od->od_stateowner); 3432 cstate->replay_owner = od->od_stateowner; 3433 } 3434 nfs4_unlock_state(); 3435 return status; 3436 } 3437 3438 /* 3439 * nfs4_unlock_state() called after encode 3440 */ 3441 __be32 3442 nfsd4_close(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate, 3443 struct nfsd4_close *close) 3444 { 3445 __be32 status; 3446 struct nfs4_stateid *stp; 3447 3448 dprintk("NFSD: nfsd4_close on file %.*s\n", 3449 (int)cstate->current_fh.fh_dentry->d_name.len, 3450 cstate->current_fh.fh_dentry->d_name.name); 3451 3452 nfs4_lock_state(); 3453 /* check close_lru for replay */ 3454 if ((status = nfs4_preprocess_seqid_op(cstate, 3455 close->cl_seqid, 3456 &close->cl_stateid, 3457 OPEN_STATE | CLOSE_STATE, 3458 &close->cl_stateowner, &stp, NULL))) 3459 goto out; 3460 status = nfs_ok; 3461 update_stateid(&stp->st_stateid); 3462 memcpy(&close->cl_stateid, &stp->st_stateid, sizeof(stateid_t)); 3463 3464 /* release_stateid() calls nfsd_close() if needed */ 3465 release_open_stateid(stp); 3466 3467 /* place unused nfs4_stateowners on so_close_lru list to be 3468 * released by the laundromat service after the lease period 3469 * to enable us to handle CLOSE replay 3470 */ 3471 if (list_empty(&close->cl_stateowner->so_stateids)) 3472 move_to_close_lru(close->cl_stateowner); 3473 out: 3474 if (close->cl_stateowner) { 3475 nfs4_get_stateowner(close->cl_stateowner); 3476 cstate->replay_owner = close->cl_stateowner; 3477 } 3478 nfs4_unlock_state(); 3479 return status; 3480 } 3481 3482 __be32 3483 nfsd4_delegreturn(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate, 3484 struct nfsd4_delegreturn *dr) 3485 { 3486 struct nfs4_delegation *dp; 3487 stateid_t *stateid = &dr->dr_stateid; 3488 struct inode *inode; 3489 __be32 status; 3490 int flags = 0; 3491 3492 if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0))) 3493 return status; 3494 inode = cstate->current_fh.fh_dentry->d_inode; 3495 3496 if (nfsd4_has_session(cstate)) 3497 flags |= HAS_SESSION; 3498 nfs4_lock_state(); 3499 status = nfserr_bad_stateid; 3500 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid)) 3501 goto out; 3502 status = nfserr_stale_stateid; 3503 if (STALE_STATEID(stateid)) 3504 goto out; 3505 status = nfserr_bad_stateid; 3506 if (!is_delegation_stateid(stateid)) 3507 goto out; 3508 status = nfserr_expired; 3509 dp = find_delegation_stateid(inode, stateid); 3510 if (!dp) 3511 goto out; 3512 status = check_stateid_generation(stateid, &dp->dl_stateid, flags); 3513 if (status) 3514 goto out; 3515 renew_client(dp->dl_client); 3516 3517 unhash_delegation(dp); 3518 out: 3519 nfs4_unlock_state(); 3520 3521 return status; 3522 } 3523 3524 3525 /* 3526 * Lock owner state (byte-range locks) 3527 */ 3528 #define LOFF_OVERFLOW(start, len) ((u64)(len) > ~(u64)(start)) 3529 #define LOCK_HASH_BITS 8 3530 #define LOCK_HASH_SIZE (1 << LOCK_HASH_BITS) 3531 #define LOCK_HASH_MASK (LOCK_HASH_SIZE - 1) 3532 3533 static inline u64 3534 end_offset(u64 start, u64 len) 3535 { 3536 u64 end; 3537 3538 end = start + len; 3539 return end >= start ? end: NFS4_MAX_UINT64; 3540 } 3541 3542 /* last octet in a range */ 3543 static inline u64 3544 last_byte_offset(u64 start, u64 len) 3545 { 3546 u64 end; 3547 3548 BUG_ON(!len); 3549 end = start + len; 3550 return end > start ? end - 1: NFS4_MAX_UINT64; 3551 } 3552 3553 #define lockownerid_hashval(id) \ 3554 ((id) & LOCK_HASH_MASK) 3555 3556 static inline unsigned int 3557 lock_ownerstr_hashval(struct inode *inode, u32 cl_id, 3558 struct xdr_netobj *ownername) 3559 { 3560 return (file_hashval(inode) + cl_id 3561 + opaque_hashval(ownername->data, ownername->len)) 3562 & LOCK_HASH_MASK; 3563 } 3564 3565 static struct list_head lock_ownerid_hashtbl[LOCK_HASH_SIZE]; 3566 static struct list_head lock_ownerstr_hashtbl[LOCK_HASH_SIZE]; 3567 static struct list_head lockstateid_hashtbl[STATEID_HASH_SIZE]; 3568 3569 static struct nfs4_stateid * 3570 find_stateid(stateid_t *stid, int flags) 3571 { 3572 struct nfs4_stateid *local; 3573 u32 st_id = stid->si_stateownerid; 3574 u32 f_id = stid->si_fileid; 3575 unsigned int hashval; 3576 3577 dprintk("NFSD: find_stateid flags 0x%x\n",flags); 3578 if (flags & (LOCK_STATE | RD_STATE | WR_STATE)) { 3579 hashval = stateid_hashval(st_id, f_id); 3580 list_for_each_entry(local, &lockstateid_hashtbl[hashval], st_hash) { 3581 if ((local->st_stateid.si_stateownerid == st_id) && 3582 (local->st_stateid.si_fileid == f_id)) 3583 return local; 3584 } 3585 } 3586 3587 if (flags & (OPEN_STATE | RD_STATE | WR_STATE)) { 3588 hashval = stateid_hashval(st_id, f_id); 3589 list_for_each_entry(local, &stateid_hashtbl[hashval], st_hash) { 3590 if ((local->st_stateid.si_stateownerid == st_id) && 3591 (local->st_stateid.si_fileid == f_id)) 3592 return local; 3593 } 3594 } 3595 return NULL; 3596 } 3597 3598 static struct nfs4_delegation * 3599 find_delegation_stateid(struct inode *ino, stateid_t *stid) 3600 { 3601 struct nfs4_file *fp; 3602 struct nfs4_delegation *dl; 3603 3604 dprintk("NFSD: %s: stateid=" STATEID_FMT "\n", __func__, 3605 STATEID_VAL(stid)); 3606 3607 fp = find_file(ino); 3608 if (!fp) 3609 return NULL; 3610 dl = find_delegation_file(fp, stid); 3611 put_nfs4_file(fp); 3612 return dl; 3613 } 3614 3615 /* 3616 * TODO: Linux file offsets are _signed_ 64-bit quantities, which means that 3617 * we can't properly handle lock requests that go beyond the (2^63 - 1)-th 3618 * byte, because of sign extension problems. Since NFSv4 calls for 64-bit 3619 * locking, this prevents us from being completely protocol-compliant. The 3620 * real solution to this problem is to start using unsigned file offsets in 3621 * the VFS, but this is a very deep change! 3622 */ 3623 static inline void 3624 nfs4_transform_lock_offset(struct file_lock *lock) 3625 { 3626 if (lock->fl_start < 0) 3627 lock->fl_start = OFFSET_MAX; 3628 if (lock->fl_end < 0) 3629 lock->fl_end = OFFSET_MAX; 3630 } 3631 3632 /* Hack!: For now, we're defining this just so we can use a pointer to it 3633 * as a unique cookie to identify our (NFSv4's) posix locks. */ 3634 static const struct lock_manager_operations nfsd_posix_mng_ops = { 3635 }; 3636 3637 static inline void 3638 nfs4_set_lock_denied(struct file_lock *fl, struct nfsd4_lock_denied *deny) 3639 { 3640 struct nfs4_stateowner *sop; 3641 3642 if (fl->fl_lmops == &nfsd_posix_mng_ops) { 3643 sop = (struct nfs4_stateowner *) fl->fl_owner; 3644 kref_get(&sop->so_ref); 3645 deny->ld_sop = sop; 3646 deny->ld_clientid = sop->so_client->cl_clientid; 3647 } else { 3648 deny->ld_sop = NULL; 3649 deny->ld_clientid.cl_boot = 0; 3650 deny->ld_clientid.cl_id = 0; 3651 } 3652 deny->ld_start = fl->fl_start; 3653 deny->ld_length = NFS4_MAX_UINT64; 3654 if (fl->fl_end != NFS4_MAX_UINT64) 3655 deny->ld_length = fl->fl_end - fl->fl_start + 1; 3656 deny->ld_type = NFS4_READ_LT; 3657 if (fl->fl_type != F_RDLCK) 3658 deny->ld_type = NFS4_WRITE_LT; 3659 } 3660 3661 static struct nfs4_stateowner * 3662 find_lockstateowner_str(struct inode *inode, clientid_t *clid, 3663 struct xdr_netobj *owner) 3664 { 3665 unsigned int hashval = lock_ownerstr_hashval(inode, clid->cl_id, owner); 3666 struct nfs4_stateowner *op; 3667 3668 list_for_each_entry(op, &lock_ownerstr_hashtbl[hashval], so_strhash) { 3669 if (same_owner_str(op, owner, clid)) 3670 return op; 3671 } 3672 return NULL; 3673 } 3674 3675 /* 3676 * Alloc a lock owner structure. 3677 * Called in nfsd4_lock - therefore, OPEN and OPEN_CONFIRM (if needed) has 3678 * occured. 3679 * 3680 * strhashval = lock_ownerstr_hashval 3681 */ 3682 3683 static struct nfs4_stateowner * 3684 alloc_init_lock_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfs4_stateid *open_stp, struct nfsd4_lock *lock) { 3685 struct nfs4_stateowner *sop; 3686 struct nfs4_replay *rp; 3687 unsigned int idhashval; 3688 3689 if (!(sop = alloc_stateowner(&lock->lk_new_owner))) 3690 return NULL; 3691 idhashval = lockownerid_hashval(current_ownerid); 3692 INIT_LIST_HEAD(&sop->so_idhash); 3693 INIT_LIST_HEAD(&sop->so_strhash); 3694 INIT_LIST_HEAD(&sop->so_perclient); 3695 INIT_LIST_HEAD(&sop->so_stateids); 3696 INIT_LIST_HEAD(&sop->so_perstateid); 3697 INIT_LIST_HEAD(&sop->so_close_lru); /* not used */ 3698 sop->so_time = 0; 3699 list_add(&sop->so_idhash, &lock_ownerid_hashtbl[idhashval]); 3700 list_add(&sop->so_strhash, &lock_ownerstr_hashtbl[strhashval]); 3701 list_add(&sop->so_perstateid, &open_stp->st_lockowners); 3702 sop->so_is_open_owner = 0; 3703 sop->so_id = current_ownerid++; 3704 sop->so_client = clp; 3705 /* It is the openowner seqid that will be incremented in encode in the 3706 * case of new lockowners; so increment the lock seqid manually: */ 3707 sop->so_seqid = lock->lk_new_lock_seqid + 1; 3708 sop->so_confirmed = 1; 3709 rp = &sop->so_replay; 3710 rp->rp_status = nfserr_serverfault; 3711 rp->rp_buflen = 0; 3712 rp->rp_buf = rp->rp_ibuf; 3713 return sop; 3714 } 3715 3716 static struct nfs4_stateid * 3717 alloc_init_lock_stateid(struct nfs4_stateowner *sop, struct nfs4_file *fp, struct nfs4_stateid *open_stp) 3718 { 3719 struct nfs4_stateid *stp; 3720 unsigned int hashval = stateid_hashval(sop->so_id, fp->fi_id); 3721 3722 stp = nfs4_alloc_stateid(); 3723 if (stp == NULL) 3724 goto out; 3725 INIT_LIST_HEAD(&stp->st_hash); 3726 INIT_LIST_HEAD(&stp->st_perfile); 3727 INIT_LIST_HEAD(&stp->st_perstateowner); 3728 INIT_LIST_HEAD(&stp->st_lockowners); /* not used */ 3729 list_add(&stp->st_hash, &lockstateid_hashtbl[hashval]); 3730 list_add(&stp->st_perfile, &fp->fi_stateids); 3731 list_add(&stp->st_perstateowner, &sop->so_stateids); 3732 stp->st_stateowner = sop; 3733 get_nfs4_file(fp); 3734 stp->st_file = fp; 3735 stp->st_stateid.si_boot = boot_time; 3736 stp->st_stateid.si_stateownerid = sop->so_id; 3737 stp->st_stateid.si_fileid = fp->fi_id; 3738 stp->st_stateid.si_generation = 0; 3739 stp->st_access_bmap = 0; 3740 stp->st_deny_bmap = open_stp->st_deny_bmap; 3741 stp->st_openstp = open_stp; 3742 3743 out: 3744 return stp; 3745 } 3746 3747 static int 3748 check_lock_length(u64 offset, u64 length) 3749 { 3750 return ((length == 0) || ((length != NFS4_MAX_UINT64) && 3751 LOFF_OVERFLOW(offset, length))); 3752 } 3753 3754 static void get_lock_access(struct nfs4_stateid *lock_stp, u32 access) 3755 { 3756 struct nfs4_file *fp = lock_stp->st_file; 3757 int oflag = nfs4_access_to_omode(access); 3758 3759 if (test_bit(access, &lock_stp->st_access_bmap)) 3760 return; 3761 nfs4_file_get_access(fp, oflag); 3762 __set_bit(access, &lock_stp->st_access_bmap); 3763 } 3764 3765 /* 3766 * LOCK operation 3767 */ 3768 __be32 3769 nfsd4_lock(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate, 3770 struct nfsd4_lock *lock) 3771 { 3772 struct nfs4_stateowner *open_sop = NULL; 3773 struct nfs4_stateowner *lock_sop = NULL; 3774 struct nfs4_stateid *lock_stp; 3775 struct nfs4_file *fp; 3776 struct file *filp = NULL; 3777 struct file_lock file_lock; 3778 struct file_lock conflock; 3779 __be32 status = 0; 3780 unsigned int strhashval; 3781 int err; 3782 3783 dprintk("NFSD: nfsd4_lock: start=%Ld length=%Ld\n", 3784 (long long) lock->lk_offset, 3785 (long long) lock->lk_length); 3786 3787 if (check_lock_length(lock->lk_offset, lock->lk_length)) 3788 return nfserr_inval; 3789 3790 if ((status = fh_verify(rqstp, &cstate->current_fh, 3791 S_IFREG, NFSD_MAY_LOCK))) { 3792 dprintk("NFSD: nfsd4_lock: permission denied!\n"); 3793 return status; 3794 } 3795 3796 nfs4_lock_state(); 3797 3798 if (lock->lk_is_new) { 3799 /* 3800 * Client indicates that this is a new lockowner. 3801 * Use open owner and open stateid to create lock owner and 3802 * lock stateid. 3803 */ 3804 struct nfs4_stateid *open_stp = NULL; 3805 3806 status = nfserr_stale_clientid; 3807 if (!nfsd4_has_session(cstate) && 3808 STALE_CLIENTID(&lock->lk_new_clientid)) 3809 goto out; 3810 3811 /* validate and update open stateid and open seqid */ 3812 status = nfs4_preprocess_seqid_op(cstate, 3813 lock->lk_new_open_seqid, 3814 &lock->lk_new_open_stateid, 3815 OPEN_STATE, 3816 &lock->lk_replay_owner, &open_stp, 3817 lock); 3818 if (status) 3819 goto out; 3820 open_sop = lock->lk_replay_owner; 3821 /* create lockowner and lock stateid */ 3822 fp = open_stp->st_file; 3823 strhashval = lock_ownerstr_hashval(fp->fi_inode, 3824 open_sop->so_client->cl_clientid.cl_id, 3825 &lock->v.new.owner); 3826 /* XXX: Do we need to check for duplicate stateowners on 3827 * the same file, or should they just be allowed (and 3828 * create new stateids)? */ 3829 status = nfserr_resource; 3830 lock_sop = alloc_init_lock_stateowner(strhashval, 3831 open_sop->so_client, open_stp, lock); 3832 if (lock_sop == NULL) 3833 goto out; 3834 lock_stp = alloc_init_lock_stateid(lock_sop, fp, open_stp); 3835 if (lock_stp == NULL) 3836 goto out; 3837 } else { 3838 /* lock (lock owner + lock stateid) already exists */ 3839 status = nfs4_preprocess_seqid_op(cstate, 3840 lock->lk_old_lock_seqid, 3841 &lock->lk_old_lock_stateid, 3842 LOCK_STATE, 3843 &lock->lk_replay_owner, &lock_stp, lock); 3844 if (status) 3845 goto out; 3846 lock_sop = lock->lk_replay_owner; 3847 fp = lock_stp->st_file; 3848 } 3849 /* lock->lk_replay_owner and lock_stp have been created or found */ 3850 3851 status = nfserr_grace; 3852 if (locks_in_grace() && !lock->lk_reclaim) 3853 goto out; 3854 status = nfserr_no_grace; 3855 if (!locks_in_grace() && lock->lk_reclaim) 3856 goto out; 3857 3858 locks_init_lock(&file_lock); 3859 switch (lock->lk_type) { 3860 case NFS4_READ_LT: 3861 case NFS4_READW_LT: 3862 filp = find_readable_file(lock_stp->st_file); 3863 if (filp) 3864 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_READ); 3865 file_lock.fl_type = F_RDLCK; 3866 break; 3867 case NFS4_WRITE_LT: 3868 case NFS4_WRITEW_LT: 3869 filp = find_writeable_file(lock_stp->st_file); 3870 if (filp) 3871 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_WRITE); 3872 file_lock.fl_type = F_WRLCK; 3873 break; 3874 default: 3875 status = nfserr_inval; 3876 goto out; 3877 } 3878 if (!filp) { 3879 status = nfserr_openmode; 3880 goto out; 3881 } 3882 file_lock.fl_owner = (fl_owner_t)lock_sop; 3883 file_lock.fl_pid = current->tgid; 3884 file_lock.fl_file = filp; 3885 file_lock.fl_flags = FL_POSIX; 3886 file_lock.fl_lmops = &nfsd_posix_mng_ops; 3887 3888 file_lock.fl_start = lock->lk_offset; 3889 file_lock.fl_end = last_byte_offset(lock->lk_offset, lock->lk_length); 3890 nfs4_transform_lock_offset(&file_lock); 3891 3892 /* 3893 * Try to lock the file in the VFS. 3894 * Note: locks.c uses the BKL to protect the inode's lock list. 3895 */ 3896 3897 err = vfs_lock_file(filp, F_SETLK, &file_lock, &conflock); 3898 switch (-err) { 3899 case 0: /* success! */ 3900 update_stateid(&lock_stp->st_stateid); 3901 memcpy(&lock->lk_resp_stateid, &lock_stp->st_stateid, 3902 sizeof(stateid_t)); 3903 status = 0; 3904 break; 3905 case (EAGAIN): /* conflock holds conflicting lock */ 3906 status = nfserr_denied; 3907 dprintk("NFSD: nfsd4_lock: conflicting lock found!\n"); 3908 nfs4_set_lock_denied(&conflock, &lock->lk_denied); 3909 break; 3910 case (EDEADLK): 3911 status = nfserr_deadlock; 3912 break; 3913 default: 3914 dprintk("NFSD: nfsd4_lock: vfs_lock_file() failed! status %d\n",err); 3915 status = nfserr_resource; 3916 break; 3917 } 3918 out: 3919 if (status && lock->lk_is_new && lock_sop) 3920 release_lockowner(lock_sop); 3921 if (lock->lk_replay_owner) { 3922 nfs4_get_stateowner(lock->lk_replay_owner); 3923 cstate->replay_owner = lock->lk_replay_owner; 3924 } 3925 nfs4_unlock_state(); 3926 return status; 3927 } 3928 3929 /* 3930 * The NFSv4 spec allows a client to do a LOCKT without holding an OPEN, 3931 * so we do a temporary open here just to get an open file to pass to 3932 * vfs_test_lock. (Arguably perhaps test_lock should be done with an 3933 * inode operation.) 3934 */ 3935 static int nfsd_test_lock(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file_lock *lock) 3936 { 3937 struct file *file; 3938 int err; 3939 3940 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file); 3941 if (err) 3942 return err; 3943 err = vfs_test_lock(file, lock); 3944 nfsd_close(file); 3945 return err; 3946 } 3947 3948 /* 3949 * LOCKT operation 3950 */ 3951 __be32 3952 nfsd4_lockt(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate, 3953 struct nfsd4_lockt *lockt) 3954 { 3955 struct inode *inode; 3956 struct file_lock file_lock; 3957 int error; 3958 __be32 status; 3959 3960 if (locks_in_grace()) 3961 return nfserr_grace; 3962 3963 if (check_lock_length(lockt->lt_offset, lockt->lt_length)) 3964 return nfserr_inval; 3965 3966 lockt->lt_stateowner = NULL; 3967 nfs4_lock_state(); 3968 3969 status = nfserr_stale_clientid; 3970 if (!nfsd4_has_session(cstate) && STALE_CLIENTID(&lockt->lt_clientid)) 3971 goto out; 3972 3973 if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0))) { 3974 dprintk("NFSD: nfsd4_lockt: fh_verify() failed!\n"); 3975 if (status == nfserr_symlink) 3976 status = nfserr_inval; 3977 goto out; 3978 } 3979 3980 inode = cstate->current_fh.fh_dentry->d_inode; 3981 locks_init_lock(&file_lock); 3982 switch (lockt->lt_type) { 3983 case NFS4_READ_LT: 3984 case NFS4_READW_LT: 3985 file_lock.fl_type = F_RDLCK; 3986 break; 3987 case NFS4_WRITE_LT: 3988 case NFS4_WRITEW_LT: 3989 file_lock.fl_type = F_WRLCK; 3990 break; 3991 default: 3992 dprintk("NFSD: nfs4_lockt: bad lock type!\n"); 3993 status = nfserr_inval; 3994 goto out; 3995 } 3996 3997 lockt->lt_stateowner = find_lockstateowner_str(inode, 3998 &lockt->lt_clientid, &lockt->lt_owner); 3999 if (lockt->lt_stateowner) 4000 file_lock.fl_owner = (fl_owner_t)lockt->lt_stateowner; 4001 file_lock.fl_pid = current->tgid; 4002 file_lock.fl_flags = FL_POSIX; 4003 4004 file_lock.fl_start = lockt->lt_offset; 4005 file_lock.fl_end = last_byte_offset(lockt->lt_offset, lockt->lt_length); 4006 4007 nfs4_transform_lock_offset(&file_lock); 4008 4009 status = nfs_ok; 4010 error = nfsd_test_lock(rqstp, &cstate->current_fh, &file_lock); 4011 if (error) { 4012 status = nfserrno(error); 4013 goto out; 4014 } 4015 if (file_lock.fl_type != F_UNLCK) { 4016 status = nfserr_denied; 4017 nfs4_set_lock_denied(&file_lock, &lockt->lt_denied); 4018 } 4019 out: 4020 nfs4_unlock_state(); 4021 return status; 4022 } 4023 4024 __be32 4025 nfsd4_locku(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate, 4026 struct nfsd4_locku *locku) 4027 { 4028 struct nfs4_stateid *stp; 4029 struct file *filp = NULL; 4030 struct file_lock file_lock; 4031 __be32 status; 4032 int err; 4033 4034 dprintk("NFSD: nfsd4_locku: start=%Ld length=%Ld\n", 4035 (long long) locku->lu_offset, 4036 (long long) locku->lu_length); 4037 4038 if (check_lock_length(locku->lu_offset, locku->lu_length)) 4039 return nfserr_inval; 4040 4041 nfs4_lock_state(); 4042 4043 if ((status = nfs4_preprocess_seqid_op(cstate, 4044 locku->lu_seqid, 4045 &locku->lu_stateid, 4046 LOCK_STATE, 4047 &locku->lu_stateowner, &stp, NULL))) 4048 goto out; 4049 4050 filp = find_any_file(stp->st_file); 4051 if (!filp) { 4052 status = nfserr_lock_range; 4053 goto out; 4054 } 4055 BUG_ON(!filp); 4056 locks_init_lock(&file_lock); 4057 file_lock.fl_type = F_UNLCK; 4058 file_lock.fl_owner = (fl_owner_t) locku->lu_stateowner; 4059 file_lock.fl_pid = current->tgid; 4060 file_lock.fl_file = filp; 4061 file_lock.fl_flags = FL_POSIX; 4062 file_lock.fl_lmops = &nfsd_posix_mng_ops; 4063 file_lock.fl_start = locku->lu_offset; 4064 4065 file_lock.fl_end = last_byte_offset(locku->lu_offset, locku->lu_length); 4066 nfs4_transform_lock_offset(&file_lock); 4067 4068 /* 4069 * Try to unlock the file in the VFS. 4070 */ 4071 err = vfs_lock_file(filp, F_SETLK, &file_lock, NULL); 4072 if (err) { 4073 dprintk("NFSD: nfs4_locku: vfs_lock_file failed!\n"); 4074 goto out_nfserr; 4075 } 4076 /* 4077 * OK, unlock succeeded; the only thing left to do is update the stateid. 4078 */ 4079 update_stateid(&stp->st_stateid); 4080 memcpy(&locku->lu_stateid, &stp->st_stateid, sizeof(stateid_t)); 4081 4082 out: 4083 if (locku->lu_stateowner) { 4084 nfs4_get_stateowner(locku->lu_stateowner); 4085 cstate->replay_owner = locku->lu_stateowner; 4086 } 4087 nfs4_unlock_state(); 4088 return status; 4089 4090 out_nfserr: 4091 status = nfserrno(err); 4092 goto out; 4093 } 4094 4095 /* 4096 * returns 4097 * 1: locks held by lockowner 4098 * 0: no locks held by lockowner 4099 */ 4100 static int 4101 check_for_locks(struct nfs4_file *filp, struct nfs4_stateowner *lowner) 4102 { 4103 struct file_lock **flpp; 4104 struct inode *inode = filp->fi_inode; 4105 int status = 0; 4106 4107 lock_flocks(); 4108 for (flpp = &inode->i_flock; *flpp != NULL; flpp = &(*flpp)->fl_next) { 4109 if ((*flpp)->fl_owner == (fl_owner_t)lowner) { 4110 status = 1; 4111 goto out; 4112 } 4113 } 4114 out: 4115 unlock_flocks(); 4116 return status; 4117 } 4118 4119 __be32 4120 nfsd4_release_lockowner(struct svc_rqst *rqstp, 4121 struct nfsd4_compound_state *cstate, 4122 struct nfsd4_release_lockowner *rlockowner) 4123 { 4124 clientid_t *clid = &rlockowner->rl_clientid; 4125 struct nfs4_stateowner *sop; 4126 struct nfs4_stateid *stp; 4127 struct xdr_netobj *owner = &rlockowner->rl_owner; 4128 struct list_head matches; 4129 int i; 4130 __be32 status; 4131 4132 dprintk("nfsd4_release_lockowner clientid: (%08x/%08x):\n", 4133 clid->cl_boot, clid->cl_id); 4134 4135 /* XXX check for lease expiration */ 4136 4137 status = nfserr_stale_clientid; 4138 if (STALE_CLIENTID(clid)) 4139 return status; 4140 4141 nfs4_lock_state(); 4142 4143 status = nfserr_locks_held; 4144 /* XXX: we're doing a linear search through all the lockowners. 4145 * Yipes! For now we'll just hope clients aren't really using 4146 * release_lockowner much, but eventually we have to fix these 4147 * data structures. */ 4148 INIT_LIST_HEAD(&matches); 4149 for (i = 0; i < LOCK_HASH_SIZE; i++) { 4150 list_for_each_entry(sop, &lock_ownerid_hashtbl[i], so_idhash) { 4151 if (!same_owner_str(sop, owner, clid)) 4152 continue; 4153 list_for_each_entry(stp, &sop->so_stateids, 4154 st_perstateowner) { 4155 if (check_for_locks(stp->st_file, sop)) 4156 goto out; 4157 /* Note: so_perclient unused for lockowners, 4158 * so it's OK to fool with here. */ 4159 list_add(&sop->so_perclient, &matches); 4160 } 4161 } 4162 } 4163 /* Clients probably won't expect us to return with some (but not all) 4164 * of the lockowner state released; so don't release any until all 4165 * have been checked. */ 4166 status = nfs_ok; 4167 while (!list_empty(&matches)) { 4168 sop = list_entry(matches.next, struct nfs4_stateowner, 4169 so_perclient); 4170 /* unhash_stateowner deletes so_perclient only 4171 * for openowners. */ 4172 list_del(&sop->so_perclient); 4173 release_lockowner(sop); 4174 } 4175 out: 4176 nfs4_unlock_state(); 4177 return status; 4178 } 4179 4180 static inline struct nfs4_client_reclaim * 4181 alloc_reclaim(void) 4182 { 4183 return kmalloc(sizeof(struct nfs4_client_reclaim), GFP_KERNEL); 4184 } 4185 4186 int 4187 nfs4_has_reclaimed_state(const char *name, bool use_exchange_id) 4188 { 4189 unsigned int strhashval = clientstr_hashval(name); 4190 struct nfs4_client *clp; 4191 4192 clp = find_confirmed_client_by_str(name, strhashval); 4193 return clp ? 1 : 0; 4194 } 4195 4196 /* 4197 * failure => all reset bets are off, nfserr_no_grace... 4198 */ 4199 int 4200 nfs4_client_to_reclaim(const char *name) 4201 { 4202 unsigned int strhashval; 4203 struct nfs4_client_reclaim *crp = NULL; 4204 4205 dprintk("NFSD nfs4_client_to_reclaim NAME: %.*s\n", HEXDIR_LEN, name); 4206 crp = alloc_reclaim(); 4207 if (!crp) 4208 return 0; 4209 strhashval = clientstr_hashval(name); 4210 INIT_LIST_HEAD(&crp->cr_strhash); 4211 list_add(&crp->cr_strhash, &reclaim_str_hashtbl[strhashval]); 4212 memcpy(crp->cr_recdir, name, HEXDIR_LEN); 4213 reclaim_str_hashtbl_size++; 4214 return 1; 4215 } 4216 4217 static void 4218 nfs4_release_reclaim(void) 4219 { 4220 struct nfs4_client_reclaim *crp = NULL; 4221 int i; 4222 4223 for (i = 0; i < CLIENT_HASH_SIZE; i++) { 4224 while (!list_empty(&reclaim_str_hashtbl[i])) { 4225 crp = list_entry(reclaim_str_hashtbl[i].next, 4226 struct nfs4_client_reclaim, cr_strhash); 4227 list_del(&crp->cr_strhash); 4228 kfree(crp); 4229 reclaim_str_hashtbl_size--; 4230 } 4231 } 4232 BUG_ON(reclaim_str_hashtbl_size); 4233 } 4234 4235 /* 4236 * called from OPEN, CLAIM_PREVIOUS with a new clientid. */ 4237 static struct nfs4_client_reclaim * 4238 nfs4_find_reclaim_client(clientid_t *clid) 4239 { 4240 unsigned int strhashval; 4241 struct nfs4_client *clp; 4242 struct nfs4_client_reclaim *crp = NULL; 4243 4244 4245 /* find clientid in conf_id_hashtbl */ 4246 clp = find_confirmed_client(clid); 4247 if (clp == NULL) 4248 return NULL; 4249 4250 dprintk("NFSD: nfs4_find_reclaim_client for %.*s with recdir %s\n", 4251 clp->cl_name.len, clp->cl_name.data, 4252 clp->cl_recdir); 4253 4254 /* find clp->cl_name in reclaim_str_hashtbl */ 4255 strhashval = clientstr_hashval(clp->cl_recdir); 4256 list_for_each_entry(crp, &reclaim_str_hashtbl[strhashval], cr_strhash) { 4257 if (same_name(crp->cr_recdir, clp->cl_recdir)) { 4258 return crp; 4259 } 4260 } 4261 return NULL; 4262 } 4263 4264 /* 4265 * Called from OPEN. Look for clientid in reclaim list. 4266 */ 4267 __be32 4268 nfs4_check_open_reclaim(clientid_t *clid) 4269 { 4270 return nfs4_find_reclaim_client(clid) ? nfs_ok : nfserr_reclaim_bad; 4271 } 4272 4273 /* initialization to perform at module load time: */ 4274 4275 int 4276 nfs4_state_init(void) 4277 { 4278 int i, status; 4279 4280 status = nfsd4_init_slabs(); 4281 if (status) 4282 return status; 4283 for (i = 0; i < CLIENT_HASH_SIZE; i++) { 4284 INIT_LIST_HEAD(&conf_id_hashtbl[i]); 4285 INIT_LIST_HEAD(&conf_str_hashtbl[i]); 4286 INIT_LIST_HEAD(&unconf_str_hashtbl[i]); 4287 INIT_LIST_HEAD(&unconf_id_hashtbl[i]); 4288 INIT_LIST_HEAD(&reclaim_str_hashtbl[i]); 4289 } 4290 for (i = 0; i < SESSION_HASH_SIZE; i++) 4291 INIT_LIST_HEAD(&sessionid_hashtbl[i]); 4292 for (i = 0; i < FILE_HASH_SIZE; i++) { 4293 INIT_LIST_HEAD(&file_hashtbl[i]); 4294 } 4295 for (i = 0; i < OWNER_HASH_SIZE; i++) { 4296 INIT_LIST_HEAD(&ownerstr_hashtbl[i]); 4297 INIT_LIST_HEAD(&ownerid_hashtbl[i]); 4298 } 4299 for (i = 0; i < STATEID_HASH_SIZE; i++) { 4300 INIT_LIST_HEAD(&stateid_hashtbl[i]); 4301 INIT_LIST_HEAD(&lockstateid_hashtbl[i]); 4302 } 4303 for (i = 0; i < LOCK_HASH_SIZE; i++) { 4304 INIT_LIST_HEAD(&lock_ownerid_hashtbl[i]); 4305 INIT_LIST_HEAD(&lock_ownerstr_hashtbl[i]); 4306 } 4307 memset(&onestateid, ~0, sizeof(stateid_t)); 4308 INIT_LIST_HEAD(&close_lru); 4309 INIT_LIST_HEAD(&client_lru); 4310 INIT_LIST_HEAD(&del_recall_lru); 4311 reclaim_str_hashtbl_size = 0; 4312 return 0; 4313 } 4314 4315 static void 4316 nfsd4_load_reboot_recovery_data(void) 4317 { 4318 int status; 4319 4320 nfs4_lock_state(); 4321 nfsd4_init_recdir(user_recovery_dirname); 4322 status = nfsd4_recdir_load(); 4323 nfs4_unlock_state(); 4324 if (status) 4325 printk("NFSD: Failure reading reboot recovery data\n"); 4326 } 4327 4328 /* 4329 * Since the lifetime of a delegation isn't limited to that of an open, a 4330 * client may quite reasonably hang on to a delegation as long as it has 4331 * the inode cached. This becomes an obvious problem the first time a 4332 * client's inode cache approaches the size of the server's total memory. 4333 * 4334 * For now we avoid this problem by imposing a hard limit on the number 4335 * of delegations, which varies according to the server's memory size. 4336 */ 4337 static void 4338 set_max_delegations(void) 4339 { 4340 /* 4341 * Allow at most 4 delegations per megabyte of RAM. Quick 4342 * estimates suggest that in the worst case (where every delegation 4343 * is for a different inode), a delegation could take about 1.5K, 4344 * giving a worst case usage of about 6% of memory. 4345 */ 4346 max_delegations = nr_free_buffer_pages() >> (20 - 2 - PAGE_SHIFT); 4347 } 4348 4349 /* initialization to perform when the nfsd service is started: */ 4350 4351 static int 4352 __nfs4_state_start(void) 4353 { 4354 int ret; 4355 4356 boot_time = get_seconds(); 4357 locks_start_grace(&nfsd4_manager); 4358 printk(KERN_INFO "NFSD: starting %ld-second grace period\n", 4359 nfsd4_grace); 4360 ret = set_callback_cred(); 4361 if (ret) 4362 return -ENOMEM; 4363 laundry_wq = create_singlethread_workqueue("nfsd4"); 4364 if (laundry_wq == NULL) 4365 return -ENOMEM; 4366 ret = nfsd4_create_callback_queue(); 4367 if (ret) 4368 goto out_free_laundry; 4369 queue_delayed_work(laundry_wq, &laundromat_work, nfsd4_grace * HZ); 4370 set_max_delegations(); 4371 return 0; 4372 out_free_laundry: 4373 destroy_workqueue(laundry_wq); 4374 return ret; 4375 } 4376 4377 int 4378 nfs4_state_start(void) 4379 { 4380 nfsd4_load_reboot_recovery_data(); 4381 return __nfs4_state_start(); 4382 } 4383 4384 static void 4385 __nfs4_state_shutdown(void) 4386 { 4387 int i; 4388 struct nfs4_client *clp = NULL; 4389 struct nfs4_delegation *dp = NULL; 4390 struct list_head *pos, *next, reaplist; 4391 4392 for (i = 0; i < CLIENT_HASH_SIZE; i++) { 4393 while (!list_empty(&conf_id_hashtbl[i])) { 4394 clp = list_entry(conf_id_hashtbl[i].next, struct nfs4_client, cl_idhash); 4395 expire_client(clp); 4396 } 4397 while (!list_empty(&unconf_str_hashtbl[i])) { 4398 clp = list_entry(unconf_str_hashtbl[i].next, struct nfs4_client, cl_strhash); 4399 expire_client(clp); 4400 } 4401 } 4402 INIT_LIST_HEAD(&reaplist); 4403 spin_lock(&recall_lock); 4404 list_for_each_safe(pos, next, &del_recall_lru) { 4405 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru); 4406 list_move(&dp->dl_recall_lru, &reaplist); 4407 } 4408 spin_unlock(&recall_lock); 4409 list_for_each_safe(pos, next, &reaplist) { 4410 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru); 4411 list_del_init(&dp->dl_recall_lru); 4412 unhash_delegation(dp); 4413 } 4414 4415 nfsd4_shutdown_recdir(); 4416 } 4417 4418 void 4419 nfs4_state_shutdown(void) 4420 { 4421 cancel_delayed_work_sync(&laundromat_work); 4422 destroy_workqueue(laundry_wq); 4423 locks_end_grace(&nfsd4_manager); 4424 nfs4_lock_state(); 4425 nfs4_release_reclaim(); 4426 __nfs4_state_shutdown(); 4427 nfs4_unlock_state(); 4428 nfsd4_destroy_callback_queue(); 4429 } 4430 4431 /* 4432 * user_recovery_dirname is protected by the nfsd_mutex since it's only 4433 * accessed when nfsd is starting. 4434 */ 4435 static void 4436 nfs4_set_recdir(char *recdir) 4437 { 4438 strcpy(user_recovery_dirname, recdir); 4439 } 4440 4441 /* 4442 * Change the NFSv4 recovery directory to recdir. 4443 */ 4444 int 4445 nfs4_reset_recoverydir(char *recdir) 4446 { 4447 int status; 4448 struct path path; 4449 4450 status = kern_path(recdir, LOOKUP_FOLLOW, &path); 4451 if (status) 4452 return status; 4453 status = -ENOTDIR; 4454 if (S_ISDIR(path.dentry->d_inode->i_mode)) { 4455 nfs4_set_recdir(recdir); 4456 status = 0; 4457 } 4458 path_put(&path); 4459 return status; 4460 } 4461 4462 char * 4463 nfs4_recoverydir(void) 4464 { 4465 return user_recovery_dirname; 4466 } 4467