1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * Central processing for nfsd. 4 * 5 * Authors: Olaf Kirch (okir@monad.swb.de) 6 * 7 * Copyright (C) 1995, 1996, 1997 Olaf Kirch <okir@monad.swb.de> 8 */ 9 10 #include <linux/sched/signal.h> 11 #include <linux/freezer.h> 12 #include <linux/module.h> 13 #include <linux/fs_struct.h> 14 #include <linux/swap.h> 15 #include <linux/siphash.h> 16 17 #include <linux/sunrpc/stats.h> 18 #include <linux/sunrpc/svcsock.h> 19 #include <linux/sunrpc/svc_xprt.h> 20 #include <linux/lockd/bind.h> 21 #include <linux/nfsacl.h> 22 #include <linux/nfslocalio.h> 23 #include <linux/seq_file.h> 24 #include <linux/inetdevice.h> 25 #include <net/addrconf.h> 26 #include <net/ipv6.h> 27 #include <net/net_namespace.h> 28 #include "nfsd.h" 29 #include "cache.h" 30 #include "vfs.h" 31 #include "netns.h" 32 #include "filecache.h" 33 34 #include "trace.h" 35 36 #define NFSDDBG_FACILITY NFSDDBG_SVC 37 38 atomic_t nfsd_th_cnt = ATOMIC_INIT(0); 39 static int nfsd(void *vrqstp); 40 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) 41 static int nfsd_acl_rpcbind_set(struct net *, 42 const struct svc_program *, 43 u32, int, 44 unsigned short, 45 unsigned short); 46 static __be32 nfsd_acl_init_request(struct svc_rqst *, 47 const struct svc_program *, 48 struct svc_process_info *); 49 #endif 50 static int nfsd_rpcbind_set(struct net *, 51 const struct svc_program *, 52 u32, int, 53 unsigned short, 54 unsigned short); 55 static __be32 nfsd_init_request(struct svc_rqst *, 56 const struct svc_program *, 57 struct svc_process_info *); 58 59 /* 60 * nfsd_mutex protects nn->nfsd_serv -- both the pointer itself and some members 61 * of the svc_serv struct such as ->sv_temp_socks and ->sv_permsocks. 62 * 63 * Finally, the nfsd_mutex also protects some of the global variables that are 64 * accessed when nfsd starts and that are settable via the write_* routines in 65 * nfsctl.c. In particular: 66 * 67 * user_recovery_dirname 68 * user_lease_time 69 * nfsd_versions 70 */ 71 DEFINE_MUTEX(nfsd_mutex); 72 73 #if IS_ENABLED(CONFIG_NFS_LOCALIO) 74 static const struct svc_version *localio_versions[] = { 75 [1] = &localio_version1, 76 }; 77 78 #define NFSD_LOCALIO_NRVERS ARRAY_SIZE(localio_versions) 79 80 #endif /* CONFIG_NFS_LOCALIO */ 81 82 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) 83 static const struct svc_version *nfsd_acl_version[] = { 84 # if defined(CONFIG_NFSD_V2_ACL) 85 [2] = &nfsd_acl_version2, 86 # endif 87 # if defined(CONFIG_NFSD_V3_ACL) 88 [3] = &nfsd_acl_version3, 89 # endif 90 }; 91 92 #define NFSD_ACL_MINVERS 2 93 #define NFSD_ACL_NRVERS ARRAY_SIZE(nfsd_acl_version) 94 95 #endif /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */ 96 97 static const struct svc_version *nfsd_version[NFSD_MAXVERS+1] = { 98 #if defined(CONFIG_NFSD_V2) 99 [2] = &nfsd_version2, 100 #endif 101 [3] = &nfsd_version3, 102 #if defined(CONFIG_NFSD_V4) 103 [4] = &nfsd_version4, 104 #endif 105 }; 106 107 struct svc_program nfsd_programs[] = { 108 { 109 .pg_prog = NFS_PROGRAM, /* program number */ 110 .pg_nvers = NFSD_MAXVERS+1, /* nr of entries in nfsd_version */ 111 .pg_vers = nfsd_version, /* version table */ 112 .pg_name = "nfsd", /* program name */ 113 .pg_class = "nfsd", /* authentication class */ 114 .pg_authenticate = svc_set_client, /* export authentication */ 115 .pg_init_request = nfsd_init_request, 116 .pg_rpcbind_set = nfsd_rpcbind_set, 117 }, 118 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) 119 { 120 .pg_prog = NFS_ACL_PROGRAM, 121 .pg_nvers = NFSD_ACL_NRVERS, 122 .pg_vers = nfsd_acl_version, 123 .pg_name = "nfsacl", 124 .pg_class = "nfsd", 125 .pg_authenticate = svc_set_client, 126 .pg_init_request = nfsd_acl_init_request, 127 .pg_rpcbind_set = nfsd_acl_rpcbind_set, 128 }, 129 #endif /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */ 130 #if IS_ENABLED(CONFIG_NFS_LOCALIO) 131 { 132 .pg_prog = NFS_LOCALIO_PROGRAM, 133 .pg_nvers = NFSD_LOCALIO_NRVERS, 134 .pg_vers = localio_versions, 135 .pg_name = "nfslocalio", 136 .pg_class = "nfsd", 137 .pg_authenticate = svc_set_client, 138 .pg_init_request = svc_generic_init_request, 139 .pg_rpcbind_set = svc_generic_rpcbind_set, 140 } 141 #endif /* CONFIG_NFS_LOCALIO */ 142 }; 143 144 bool nfsd_support_version(int vers) 145 { 146 if (vers >= NFSD_MINVERS && vers <= NFSD_MAXVERS) 147 return nfsd_version[vers] != NULL; 148 return false; 149 } 150 151 int nfsd_vers(struct nfsd_net *nn, int vers, enum vers_op change) 152 { 153 if (vers < NFSD_MINVERS || vers > NFSD_MAXVERS) 154 return 0; 155 switch(change) { 156 case NFSD_SET: 157 nn->nfsd_versions[vers] = nfsd_support_version(vers); 158 break; 159 case NFSD_CLEAR: 160 nn->nfsd_versions[vers] = false; 161 break; 162 case NFSD_TEST: 163 return nn->nfsd_versions[vers]; 164 case NFSD_AVAIL: 165 return nfsd_support_version(vers); 166 } 167 return 0; 168 } 169 170 static void 171 nfsd_adjust_nfsd_versions4(struct nfsd_net *nn) 172 { 173 unsigned i; 174 175 for (i = 0; i <= NFSD_SUPPORTED_MINOR_VERSION; i++) { 176 if (nn->nfsd4_minorversions[i]) 177 return; 178 } 179 nfsd_vers(nn, 4, NFSD_CLEAR); 180 } 181 182 int nfsd_minorversion(struct nfsd_net *nn, u32 minorversion, enum vers_op change) 183 { 184 if (minorversion > NFSD_SUPPORTED_MINOR_VERSION && 185 change != NFSD_AVAIL) 186 return -1; 187 188 switch(change) { 189 case NFSD_SET: 190 nfsd_vers(nn, 4, NFSD_SET); 191 nn->nfsd4_minorversions[minorversion] = 192 nfsd_vers(nn, 4, NFSD_TEST); 193 break; 194 case NFSD_CLEAR: 195 nn->nfsd4_minorversions[minorversion] = false; 196 nfsd_adjust_nfsd_versions4(nn); 197 break; 198 case NFSD_TEST: 199 return nn->nfsd4_minorversions[minorversion]; 200 case NFSD_AVAIL: 201 return minorversion <= NFSD_SUPPORTED_MINOR_VERSION && 202 nfsd_vers(nn, 4, NFSD_AVAIL); 203 } 204 return 0; 205 } 206 207 bool nfsd_net_try_get(struct net *net) __must_hold(rcu) 208 { 209 struct nfsd_net *nn = net_generic(net, nfsd_net_id); 210 211 return (nn && percpu_ref_tryget_live(&nn->nfsd_net_ref)); 212 } 213 214 void nfsd_net_put(struct net *net) __must_hold(rcu) 215 { 216 struct nfsd_net *nn = net_generic(net, nfsd_net_id); 217 218 percpu_ref_put(&nn->nfsd_net_ref); 219 } 220 221 static void nfsd_net_done(struct percpu_ref *ref) 222 { 223 struct nfsd_net *nn = container_of(ref, struct nfsd_net, nfsd_net_ref); 224 225 complete(&nn->nfsd_net_confirm_done); 226 } 227 228 static void nfsd_net_free(struct percpu_ref *ref) 229 { 230 struct nfsd_net *nn = container_of(ref, struct nfsd_net, nfsd_net_ref); 231 232 complete(&nn->nfsd_net_free_done); 233 } 234 235 /* 236 * Maximum number of nfsd processes 237 */ 238 #define NFSD_MAXSERVS 8192 239 240 int nfsd_nrthreads(struct net *net) 241 { 242 int i, rv = 0; 243 struct nfsd_net *nn = net_generic(net, nfsd_net_id); 244 245 mutex_lock(&nfsd_mutex); 246 if (nn->nfsd_serv) 247 for (i = 0; i < nn->nfsd_serv->sv_nrpools; ++i) 248 rv += nn->nfsd_serv->sv_pools[i].sp_nrthrmax; 249 mutex_unlock(&nfsd_mutex); 250 return rv; 251 } 252 253 static int nfsd_users = 0; 254 255 static int nfsd_startup_generic(void) 256 { 257 int ret; 258 259 if (nfsd_users++) 260 return 0; 261 262 ret = nfsd_file_cache_init(); 263 if (ret) 264 goto dec_users; 265 266 ret = nfs4_state_start(); 267 if (ret) 268 goto out_file_cache; 269 return 0; 270 271 out_file_cache: 272 nfsd_file_cache_shutdown(); 273 dec_users: 274 nfsd_users--; 275 return ret; 276 } 277 278 static void nfsd_shutdown_generic(void) 279 { 280 if (--nfsd_users) 281 return; 282 283 nfs4_state_shutdown(); 284 nfsd_file_cache_shutdown(); 285 } 286 287 static bool nfsd_needs_lockd(struct nfsd_net *nn) 288 { 289 return nfsd_vers(nn, 2, NFSD_TEST) || nfsd_vers(nn, 3, NFSD_TEST); 290 } 291 292 /** 293 * nfsd_copy_write_verifier - Atomically copy a write verifier 294 * @verf: buffer in which to receive the verifier cookie 295 * @nn: NFS net namespace 296 * 297 * This function provides a wait-free mechanism for copying the 298 * namespace's write verifier without tearing it. 299 */ 300 void nfsd_copy_write_verifier(__be32 verf[2], struct nfsd_net *nn) 301 { 302 unsigned int seq; 303 304 do { 305 seq = read_seqbegin(&nn->writeverf_lock); 306 memcpy(verf, nn->writeverf, sizeof(nn->writeverf)); 307 } while (read_seqretry(&nn->writeverf_lock, seq)); 308 } 309 310 static void nfsd_reset_write_verifier_locked(struct nfsd_net *nn) 311 { 312 struct timespec64 now; 313 u64 verf; 314 315 /* 316 * Because the time value is hashed, y2038 time_t overflow 317 * is irrelevant in this usage. 318 */ 319 ktime_get_raw_ts64(&now); 320 verf = siphash_2u64(now.tv_sec, now.tv_nsec, &nn->siphash_key); 321 memcpy(nn->writeverf, &verf, sizeof(nn->writeverf)); 322 } 323 324 /** 325 * nfsd_reset_write_verifier - Generate a new write verifier 326 * @nn: NFS net namespace 327 * 328 * This function updates the ->writeverf field of @nn. This field 329 * contains an opaque cookie that, according to Section 18.32.3 of 330 * RFC 8881, "the client can use to determine whether a server has 331 * changed instance state (e.g., server restart) between a call to 332 * WRITE and a subsequent call to either WRITE or COMMIT. This 333 * cookie MUST be unchanged during a single instance of the NFSv4.1 334 * server and MUST be unique between instances of the NFSv4.1 335 * server." 336 */ 337 void nfsd_reset_write_verifier(struct nfsd_net *nn) 338 { 339 write_seqlock(&nn->writeverf_lock); 340 nfsd_reset_write_verifier_locked(nn); 341 write_sequnlock(&nn->writeverf_lock); 342 } 343 344 /* 345 * Crank up a set of per-namespace resources for a new NFSD instance, 346 * including lockd, a duplicate reply cache, an open file cache 347 * instance, and a cache of NFSv4 state objects. 348 */ 349 static int nfsd_startup_net(struct net *net, const struct cred *cred) 350 { 351 struct nfsd_net *nn = net_generic(net, nfsd_net_id); 352 int ret; 353 354 if (nn->nfsd_net_up) 355 return 0; 356 357 ret = nfsd_startup_generic(); 358 if (ret) 359 return ret; 360 361 if (list_empty(&nn->nfsd_serv->sv_permsocks)) { 362 pr_warn("NFSD: Failed to start, no listeners configured.\n"); 363 ret = -EIO; 364 goto out_socks; 365 } 366 367 if (nfsd_needs_lockd(nn) && !nn->lockd_up) { 368 ret = lockd_up(net, cred); 369 if (ret) 370 goto out_socks; 371 nn->lockd_up = true; 372 } 373 374 ret = nfsd_file_cache_start_net(net); 375 if (ret) 376 goto out_lockd; 377 378 ret = nfsd_reply_cache_init(nn); 379 if (ret) 380 goto out_filecache; 381 382 #ifdef CONFIG_NFSD_V4_2_INTER_SSC 383 nfsd4_ssc_init_umount_work(nn); 384 #endif 385 ret = nfs4_state_start_net(net); 386 if (ret) 387 goto out_reply_cache; 388 389 nn->nfsd_net_up = true; 390 return 0; 391 392 out_reply_cache: 393 nfsd_reply_cache_shutdown(nn); 394 out_filecache: 395 nfsd_file_cache_shutdown_net(net); 396 out_lockd: 397 if (nn->lockd_up) { 398 lockd_down(net); 399 nn->lockd_up = false; 400 } 401 out_socks: 402 nfsd_shutdown_generic(); 403 return ret; 404 } 405 406 static void nfsd_shutdown_net(struct net *net) 407 { 408 struct nfsd_net *nn = net_generic(net, nfsd_net_id); 409 410 if (nn->nfsd_net_up) { 411 percpu_ref_kill_and_confirm(&nn->nfsd_net_ref, nfsd_net_done); 412 wait_for_completion(&nn->nfsd_net_confirm_done); 413 414 nfsd_export_flush(net); 415 nfs4_state_shutdown_net(net); 416 nfsd_reply_cache_shutdown(nn); 417 nfsd_file_cache_shutdown_net(net); 418 if (nn->lockd_up) { 419 lockd_down(net); 420 nn->lockd_up = false; 421 } 422 wait_for_completion(&nn->nfsd_net_free_done); 423 } 424 425 percpu_ref_exit(&nn->nfsd_net_ref); 426 427 if (nn->nfsd_net_up) 428 nfsd_shutdown_generic(); 429 nn->nfsd_net_up = false; 430 } 431 432 static DEFINE_SPINLOCK(nfsd_notifier_lock); 433 static int nfsd_inetaddr_event(struct notifier_block *this, unsigned long event, 434 void *ptr) 435 { 436 struct in_ifaddr *ifa = (struct in_ifaddr *)ptr; 437 struct net_device *dev = ifa->ifa_dev->dev; 438 struct net *net = dev_net(dev); 439 struct nfsd_net *nn = net_generic(net, nfsd_net_id); 440 struct sockaddr_in sin; 441 442 if (event != NETDEV_DOWN || !nn->nfsd_serv) 443 goto out; 444 445 spin_lock(&nfsd_notifier_lock); 446 if (nn->nfsd_serv) { 447 dprintk("nfsd_inetaddr_event: removed %pI4\n", &ifa->ifa_local); 448 sin.sin_family = AF_INET; 449 sin.sin_addr.s_addr = ifa->ifa_local; 450 svc_age_temp_xprts_now(nn->nfsd_serv, (struct sockaddr *)&sin); 451 } 452 spin_unlock(&nfsd_notifier_lock); 453 454 out: 455 return NOTIFY_DONE; 456 } 457 458 static struct notifier_block nfsd_inetaddr_notifier = { 459 .notifier_call = nfsd_inetaddr_event, 460 }; 461 462 #if IS_ENABLED(CONFIG_IPV6) 463 static int nfsd_inet6addr_event(struct notifier_block *this, 464 unsigned long event, void *ptr) 465 { 466 struct inet6_ifaddr *ifa = (struct inet6_ifaddr *)ptr; 467 struct net_device *dev = ifa->idev->dev; 468 struct net *net = dev_net(dev); 469 struct nfsd_net *nn = net_generic(net, nfsd_net_id); 470 struct sockaddr_in6 sin6; 471 472 if (event != NETDEV_DOWN || !nn->nfsd_serv) 473 goto out; 474 475 spin_lock(&nfsd_notifier_lock); 476 if (nn->nfsd_serv) { 477 dprintk("nfsd_inet6addr_event: removed %pI6\n", &ifa->addr); 478 sin6.sin6_family = AF_INET6; 479 sin6.sin6_addr = ifa->addr; 480 if (ipv6_addr_type(&sin6.sin6_addr) & IPV6_ADDR_LINKLOCAL) 481 sin6.sin6_scope_id = ifa->idev->dev->ifindex; 482 svc_age_temp_xprts_now(nn->nfsd_serv, (struct sockaddr *)&sin6); 483 } 484 spin_unlock(&nfsd_notifier_lock); 485 486 out: 487 return NOTIFY_DONE; 488 } 489 490 static struct notifier_block nfsd_inet6addr_notifier = { 491 .notifier_call = nfsd_inet6addr_event, 492 }; 493 #endif 494 495 /* Only used under nfsd_mutex, so this atomic may be overkill: */ 496 static atomic_t nfsd_notifier_refcount = ATOMIC_INIT(0); 497 498 /** 499 * nfsd_destroy_serv - tear down NFSD's svc_serv for a namespace 500 * @net: network namespace the NFS service is associated with 501 */ 502 void nfsd_destroy_serv(struct net *net) 503 { 504 struct nfsd_net *nn = net_generic(net, nfsd_net_id); 505 struct svc_serv *serv = nn->nfsd_serv; 506 507 lockdep_assert_held(&nfsd_mutex); 508 509 spin_lock(&nfsd_notifier_lock); 510 nn->nfsd_serv = NULL; 511 spin_unlock(&nfsd_notifier_lock); 512 513 /* check if the notifier still has clients */ 514 if (atomic_dec_return(&nfsd_notifier_refcount) == 0) { 515 unregister_inetaddr_notifier(&nfsd_inetaddr_notifier); 516 #if IS_ENABLED(CONFIG_IPV6) 517 unregister_inet6addr_notifier(&nfsd_inet6addr_notifier); 518 #endif 519 } 520 521 /* 522 * write_ports can create the server without actually starting 523 * any threads. If we get shut down before any threads are 524 * started, then nfsd_destroy_serv will be run before any of this 525 * other initialization has been done except the rpcb information. 526 */ 527 svc_xprt_destroy_all(serv, net, true); 528 nfsd_shutdown_net(net); 529 svc_destroy(&serv); 530 } 531 532 void nfsd_reset_versions(struct nfsd_net *nn) 533 { 534 int i; 535 536 for (i = 0; i <= NFSD_MAXVERS; i++) 537 if (nfsd_vers(nn, i, NFSD_TEST)) 538 return; 539 540 for (i = 0; i <= NFSD_MAXVERS; i++) 541 if (i != 4) 542 nfsd_vers(nn, i, NFSD_SET); 543 else { 544 int minor = 0; 545 while (nfsd_minorversion(nn, minor, NFSD_SET) >= 0) 546 minor++; 547 } 548 } 549 550 static int nfsd_get_default_max_blksize(void) 551 { 552 struct sysinfo i; 553 unsigned long long target; 554 unsigned long ret; 555 556 si_meminfo(&i); 557 target = (i.totalram - i.totalhigh) << PAGE_SHIFT; 558 /* 559 * Aim for 1/4096 of memory per thread This gives 1MB on 4Gig 560 * machines, but only uses 32K on 128M machines. Bottom out at 561 * 8K on 32M and smaller. Of course, this is only a default. 562 */ 563 target >>= 12; 564 565 ret = NFSSVC_DEFBLKSIZE; 566 while (ret > target && ret >= 8*1024*2) 567 ret /= 2; 568 return ret; 569 } 570 571 void nfsd_shutdown_threads(struct net *net) 572 { 573 struct nfsd_net *nn = net_generic(net, nfsd_net_id); 574 struct svc_serv *serv; 575 576 mutex_lock(&nfsd_mutex); 577 serv = nn->nfsd_serv; 578 if (serv == NULL) { 579 mutex_unlock(&nfsd_mutex); 580 return; 581 } 582 583 /* Kill outstanding nfsd threads */ 584 svc_set_num_threads(serv, 0, 0); 585 nfsd_destroy_serv(net); 586 mutex_unlock(&nfsd_mutex); 587 } 588 589 struct svc_rqst *nfsd_current_rqst(void) 590 { 591 if (kthread_func(current) == nfsd) 592 return kthread_data(current); 593 return NULL; 594 } 595 596 int nfsd_create_serv(struct net *net) 597 { 598 int error; 599 struct nfsd_net *nn = net_generic(net, nfsd_net_id); 600 struct svc_serv *serv; 601 602 WARN_ON(!mutex_is_locked(&nfsd_mutex)); 603 if (nn->nfsd_serv) 604 return 0; 605 606 error = percpu_ref_init(&nn->nfsd_net_ref, nfsd_net_free, 607 0, GFP_KERNEL); 608 if (error) 609 return error; 610 init_completion(&nn->nfsd_net_free_done); 611 init_completion(&nn->nfsd_net_confirm_done); 612 613 if (nfsd_max_blksize == 0) 614 nfsd_max_blksize = nfsd_get_default_max_blksize(); 615 nfsd_reset_versions(nn); 616 serv = svc_create_pooled(nfsd_programs, ARRAY_SIZE(nfsd_programs), 617 &nn->nfsd_svcstats, 618 nfsd_max_blksize, nfsd); 619 if (serv == NULL) { 620 percpu_ref_exit(&nn->nfsd_net_ref); 621 return -ENOMEM; 622 } 623 624 error = svc_bind(serv, net); 625 if (error < 0) { 626 svc_destroy(&serv); 627 percpu_ref_exit(&nn->nfsd_net_ref); 628 return error; 629 } 630 spin_lock(&nfsd_notifier_lock); 631 nn->nfsd_serv = serv; 632 spin_unlock(&nfsd_notifier_lock); 633 634 /* check if the notifier is already set */ 635 if (atomic_inc_return(&nfsd_notifier_refcount) == 1) { 636 register_inetaddr_notifier(&nfsd_inetaddr_notifier); 637 #if IS_ENABLED(CONFIG_IPV6) 638 register_inet6addr_notifier(&nfsd_inet6addr_notifier); 639 #endif 640 } 641 nfsd_reset_write_verifier(nn); 642 return 0; 643 } 644 645 int nfsd_nrpools(struct net *net) 646 { 647 struct nfsd_net *nn = net_generic(net, nfsd_net_id); 648 649 if (nn->nfsd_serv == NULL) 650 return 0; 651 else 652 return nn->nfsd_serv->sv_nrpools; 653 } 654 655 int nfsd_get_nrthreads(int n, int *nthreads, struct net *net) 656 { 657 struct nfsd_net *nn = net_generic(net, nfsd_net_id); 658 struct svc_serv *serv = nn->nfsd_serv; 659 int i; 660 661 if (serv) 662 for (i = 0; i < serv->sv_nrpools && i < n; i++) 663 nthreads[i] = serv->sv_pools[i].sp_nrthrmax; 664 return 0; 665 } 666 667 /** 668 * nfsd_set_nrthreads - set the number of running threads in the net's service 669 * @n: number of array members in @nthreads 670 * @nthreads: array of thread counts for each pool 671 * @net: network namespace to operate within 672 * 673 * This function alters the number of running threads for the given network 674 * namespace in each pool. If passed an array longer then the number of pools 675 * the extra pool settings are ignored. If passed an array shorter than the 676 * number of pools, the missing values are interpreted as 0's. 677 * 678 * Returns 0 on success or a negative errno on error. 679 */ 680 int nfsd_set_nrthreads(int n, int *nthreads, struct net *net) 681 { 682 int i = 0; 683 int tot = 0; 684 int err = 0; 685 struct nfsd_net *nn = net_generic(net, nfsd_net_id); 686 687 lockdep_assert_held(&nfsd_mutex); 688 689 if (nn->nfsd_serv == NULL || n <= 0) 690 return 0; 691 692 /* Special case: When n == 1, distribute threads equally among pools. */ 693 if (n == 1) 694 return svc_set_num_threads(nn->nfsd_serv, nn->min_threads, nthreads[0]); 695 696 if (n > nn->nfsd_serv->sv_nrpools) 697 n = nn->nfsd_serv->sv_nrpools; 698 699 /* enforce a global maximum number of threads */ 700 tot = 0; 701 for (i = 0; i < n; i++) { 702 nthreads[i] = min(nthreads[i], NFSD_MAXSERVS); 703 tot += nthreads[i]; 704 } 705 if (tot > NFSD_MAXSERVS) { 706 /* total too large: scale down requested numbers */ 707 for (i = 0; i < n && tot > 0; i++) { 708 int new = nthreads[i] * NFSD_MAXSERVS / tot; 709 tot -= (nthreads[i] - new); 710 nthreads[i] = new; 711 } 712 for (i = 0; i < n && tot > 0; i++) { 713 nthreads[i]--; 714 tot--; 715 } 716 } 717 718 /* apply the new numbers */ 719 for (i = 0; i < n; i++) { 720 err = svc_set_pool_threads(nn->nfsd_serv, 721 &nn->nfsd_serv->sv_pools[i], 722 nn->min_threads, nthreads[i]); 723 if (err) 724 goto out; 725 } 726 727 /* Anything undefined in array is considered to be 0 */ 728 for (i = n; i < nn->nfsd_serv->sv_nrpools; ++i) { 729 err = svc_set_pool_threads(nn->nfsd_serv, 730 &nn->nfsd_serv->sv_pools[i], 731 0, 0); 732 if (err) 733 goto out; 734 } 735 out: 736 return err; 737 } 738 739 /** 740 * nfsd_svc: start up or shut down the nfsd server 741 * @n: number of array members in @nthreads 742 * @nthreads: array of thread counts for each pool 743 * @net: network namespace to operate within 744 * @cred: credentials to use for xprt creation 745 * @scope: server scope value (defaults to nodename) 746 * 747 * Adjust the number of threads in each pool and return the new 748 * total number of threads in the service. 749 */ 750 int 751 nfsd_svc(int n, int *nthreads, struct net *net, const struct cred *cred, const char *scope) 752 { 753 int error; 754 struct nfsd_net *nn = net_generic(net, nfsd_net_id); 755 struct svc_serv *serv; 756 757 lockdep_assert_held(&nfsd_mutex); 758 759 dprintk("nfsd: creating service\n"); 760 761 strscpy(nn->nfsd_name, scope ? scope : utsname()->nodename, 762 sizeof(nn->nfsd_name)); 763 764 error = nfsd_create_serv(net); 765 if (error) 766 goto out; 767 serv = nn->nfsd_serv; 768 769 error = nfsd_startup_net(net, cred); 770 if (error) 771 goto out_put; 772 error = nfsd_set_nrthreads(n, nthreads, net); 773 if (error) 774 goto out_put; 775 error = serv->sv_nrthreads; 776 out_put: 777 if (serv->sv_nrthreads == 0) 778 nfsd_destroy_serv(net); 779 out: 780 return error; 781 } 782 783 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) 784 static bool 785 nfsd_support_acl_version(int vers) 786 { 787 if (vers >= NFSD_ACL_MINVERS && vers < NFSD_ACL_NRVERS) 788 return nfsd_acl_version[vers] != NULL; 789 return false; 790 } 791 792 static int 793 nfsd_acl_rpcbind_set(struct net *net, const struct svc_program *progp, 794 u32 version, int family, unsigned short proto, 795 unsigned short port) 796 { 797 if (!nfsd_support_acl_version(version) || 798 !nfsd_vers(net_generic(net, nfsd_net_id), version, NFSD_TEST)) 799 return 0; 800 return svc_generic_rpcbind_set(net, progp, version, family, 801 proto, port); 802 } 803 804 static __be32 805 nfsd_acl_init_request(struct svc_rqst *rqstp, 806 const struct svc_program *progp, 807 struct svc_process_info *ret) 808 { 809 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id); 810 int i; 811 812 if (likely(nfsd_support_acl_version(rqstp->rq_vers) && 813 nfsd_vers(nn, rqstp->rq_vers, NFSD_TEST))) 814 return svc_generic_init_request(rqstp, progp, ret); 815 816 ret->mismatch.lovers = NFSD_ACL_NRVERS; 817 for (i = NFSD_ACL_MINVERS; i < NFSD_ACL_NRVERS; i++) { 818 if (nfsd_support_acl_version(rqstp->rq_vers) && 819 nfsd_vers(nn, i, NFSD_TEST)) { 820 ret->mismatch.lovers = i; 821 break; 822 } 823 } 824 if (ret->mismatch.lovers == NFSD_ACL_NRVERS) 825 return rpc_prog_unavail; 826 ret->mismatch.hivers = NFSD_ACL_MINVERS; 827 for (i = NFSD_ACL_NRVERS - 1; i >= NFSD_ACL_MINVERS; i--) { 828 if (nfsd_support_acl_version(rqstp->rq_vers) && 829 nfsd_vers(nn, i, NFSD_TEST)) { 830 ret->mismatch.hivers = i; 831 break; 832 } 833 } 834 return rpc_prog_mismatch; 835 } 836 #endif 837 838 static int 839 nfsd_rpcbind_set(struct net *net, const struct svc_program *progp, 840 u32 version, int family, unsigned short proto, 841 unsigned short port) 842 { 843 if (!nfsd_vers(net_generic(net, nfsd_net_id), version, NFSD_TEST)) 844 return 0; 845 return svc_generic_rpcbind_set(net, progp, version, family, 846 proto, port); 847 } 848 849 static __be32 850 nfsd_init_request(struct svc_rqst *rqstp, 851 const struct svc_program *progp, 852 struct svc_process_info *ret) 853 { 854 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id); 855 int i; 856 857 if (likely(nfsd_vers(nn, rqstp->rq_vers, NFSD_TEST))) 858 return svc_generic_init_request(rqstp, progp, ret); 859 860 ret->mismatch.lovers = NFSD_MAXVERS + 1; 861 for (i = NFSD_MINVERS; i <= NFSD_MAXVERS; i++) { 862 if (nfsd_vers(nn, i, NFSD_TEST)) { 863 ret->mismatch.lovers = i; 864 break; 865 } 866 } 867 if (ret->mismatch.lovers > NFSD_MAXVERS) 868 return rpc_prog_unavail; 869 ret->mismatch.hivers = NFSD_MINVERS; 870 for (i = NFSD_MAXVERS; i >= NFSD_MINVERS; i--) { 871 if (nfsd_vers(nn, i, NFSD_TEST)) { 872 ret->mismatch.hivers = i; 873 break; 874 } 875 } 876 return rpc_prog_mismatch; 877 } 878 879 /* 880 * This is the NFS server kernel thread 881 */ 882 static int 883 nfsd(void *vrqstp) 884 { 885 struct svc_rqst *rqstp = (struct svc_rqst *) vrqstp; 886 struct svc_pool *pool = rqstp->rq_pool; 887 struct svc_xprt *perm_sock = list_entry(rqstp->rq_server->sv_permsocks.next, typeof(struct svc_xprt), xpt_list); 888 struct net *net = perm_sock->xpt_net; 889 struct nfsd_net *nn = net_generic(net, nfsd_net_id); 890 bool have_mutex = false; 891 892 /* At this point, the thread shares current->fs 893 * with the init process. We need to create files with the 894 * umask as defined by the client instead of init's umask. 895 */ 896 svc_thread_init_status(rqstp, unshare_fs_struct()); 897 898 current->fs->umask = 0; 899 900 atomic_inc(&nfsd_th_cnt); 901 902 set_freezable(); 903 904 /* 905 * The main request loop 906 */ 907 while (!svc_thread_should_stop(rqstp)) { 908 switch (svc_recv(rqstp, 5 * HZ)) { 909 case -ETIMEDOUT: 910 /* No work arrived within the timeout window */ 911 if (mutex_trylock(&nfsd_mutex)) { 912 if (pool->sp_nrthreads > pool->sp_nrthrmin) { 913 trace_nfsd_dynthread_kill(net, pool); 914 set_bit(RQ_VICTIM, &rqstp->rq_flags); 915 have_mutex = true; 916 } else { 917 mutex_unlock(&nfsd_mutex); 918 } 919 } else { 920 trace_nfsd_dynthread_trylock_fail(net, pool); 921 } 922 break; 923 case -EBUSY: 924 /* No idle threads; consider spawning another */ 925 if (pool->sp_nrthreads < pool->sp_nrthrmax) { 926 if (mutex_trylock(&nfsd_mutex)) { 927 if (pool->sp_nrthreads < pool->sp_nrthrmax) { 928 int ret; 929 930 trace_nfsd_dynthread_start(net, pool); 931 ret = svc_new_thread(rqstp->rq_server, pool); 932 if (ret) 933 pr_notice_ratelimited("%s: unable to spawn new thread: %d\n", 934 __func__, ret); 935 } 936 mutex_unlock(&nfsd_mutex); 937 } else { 938 trace_nfsd_dynthread_trylock_fail(net, pool); 939 } 940 } 941 clear_bit(SP_TASK_STARTING, &pool->sp_flags); 942 break; 943 default: 944 break; 945 } 946 nfsd_file_net_dispose(nn); 947 } 948 949 atomic_dec(&nfsd_th_cnt); 950 951 /* Release the thread */ 952 svc_exit_thread(rqstp); 953 if (have_mutex) 954 mutex_unlock(&nfsd_mutex); 955 return 0; 956 } 957 958 /** 959 * nfsd_dispatch - Process an NFS or NFSACL or LOCALIO Request 960 * @rqstp: incoming request 961 * 962 * This RPC dispatcher integrates the NFS server's duplicate reply cache. 963 * 964 * Return values: 965 * %0: Processing complete; do not send a Reply 966 * %1: Processing complete; send Reply in rqstp->rq_res 967 */ 968 int nfsd_dispatch(struct svc_rqst *rqstp) 969 { 970 const struct svc_procedure *proc = rqstp->rq_procinfo; 971 __be32 *statp = rqstp->rq_accept_statp; 972 struct nfsd_cacherep *rp; 973 unsigned int start, len; 974 __be32 *nfs_reply; 975 976 /* 977 * Give the xdr decoder a chance to change this if it wants 978 * (necessary in the NFSv4.0 compound case) 979 */ 980 rqstp->rq_cachetype = proc->pc_cachetype; 981 982 /* 983 * ->pc_decode advances the argument stream past the NFS 984 * Call header, so grab the header's starting location and 985 * size now for the call to nfsd_cache_lookup(). 986 */ 987 start = xdr_stream_pos(&rqstp->rq_arg_stream); 988 len = xdr_stream_remaining(&rqstp->rq_arg_stream); 989 if (!proc->pc_decode(rqstp, &rqstp->rq_arg_stream)) 990 goto out_decode_err; 991 992 /* 993 * Release rq_status_counter setting it to an odd value after the rpc 994 * request has been properly parsed. rq_status_counter is used to 995 * notify the consumers if the rqstp fields are stable 996 * (rq_status_counter is odd) or not meaningful (rq_status_counter 997 * is even). 998 */ 999 smp_store_release(&rqstp->rq_status_counter, rqstp->rq_status_counter | 1); 1000 1001 rp = NULL; 1002 switch (nfsd_cache_lookup(rqstp, start, len, &rp)) { 1003 case RC_DOIT: 1004 break; 1005 case RC_REPLY: 1006 goto out_cached_reply; 1007 case RC_DROPIT: 1008 goto out_dropit; 1009 } 1010 1011 nfs_reply = xdr_inline_decode(&rqstp->rq_res_stream, 0); 1012 *statp = proc->pc_func(rqstp); 1013 if (test_bit(RQ_DROPME, &rqstp->rq_flags)) 1014 goto out_update_drop; 1015 1016 if (!proc->pc_encode(rqstp, &rqstp->rq_res_stream)) 1017 goto out_encode_err; 1018 1019 /* 1020 * Release rq_status_counter setting it to an even value after the rpc 1021 * request has been properly processed. 1022 */ 1023 smp_store_release(&rqstp->rq_status_counter, rqstp->rq_status_counter + 1); 1024 1025 nfsd_cache_update(rqstp, rp, rqstp->rq_cachetype, nfs_reply); 1026 out_cached_reply: 1027 return 1; 1028 1029 out_decode_err: 1030 trace_nfsd_garbage_args_err(rqstp); 1031 *statp = rpc_garbage_args; 1032 return 1; 1033 1034 out_update_drop: 1035 nfsd_cache_update(rqstp, rp, RC_NOCACHE, NULL); 1036 out_dropit: 1037 return 0; 1038 1039 out_encode_err: 1040 trace_nfsd_cant_encode_err(rqstp); 1041 nfsd_cache_update(rqstp, rp, RC_NOCACHE, NULL); 1042 *statp = rpc_system_err; 1043 return 1; 1044 } 1045 1046 /** 1047 * nfssvc_decode_voidarg - Decode void arguments 1048 * @rqstp: Server RPC transaction context 1049 * @xdr: XDR stream positioned at arguments to decode 1050 * 1051 * Return values: 1052 * %false: Arguments were not valid 1053 * %true: Decoding was successful 1054 */ 1055 bool nfssvc_decode_voidarg(struct svc_rqst *rqstp, struct xdr_stream *xdr) 1056 { 1057 return true; 1058 } 1059 1060 /** 1061 * nfssvc_encode_voidres - Encode void results 1062 * @rqstp: Server RPC transaction context 1063 * @xdr: XDR stream into which to encode results 1064 * 1065 * Return values: 1066 * %false: Local error while encoding 1067 * %true: Encoding was successful 1068 */ 1069 bool nfssvc_encode_voidres(struct svc_rqst *rqstp, struct xdr_stream *xdr) 1070 { 1071 return true; 1072 } 1073