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 16 #include <linux/sunrpc/stats.h> 17 #include <linux/sunrpc/svcsock.h> 18 #include <linux/sunrpc/svc_xprt.h> 19 #include <linux/lockd/bind.h> 20 #include <linux/nfsacl.h> 21 #include <linux/seq_file.h> 22 #include <linux/inetdevice.h> 23 #include <net/addrconf.h> 24 #include <net/ipv6.h> 25 #include <net/net_namespace.h> 26 #include "nfsd.h" 27 #include "cache.h" 28 #include "vfs.h" 29 #include "netns.h" 30 31 #define NFSDDBG_FACILITY NFSDDBG_SVC 32 33 extern struct svc_program nfsd_program; 34 static int nfsd(void *vrqstp); 35 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) 36 static int nfsd_acl_rpcbind_set(struct net *, 37 const struct svc_program *, 38 u32, int, 39 unsigned short, 40 unsigned short); 41 static __be32 nfsd_acl_init_request(struct svc_rqst *, 42 const struct svc_program *, 43 struct svc_process_info *); 44 #endif 45 static int nfsd_rpcbind_set(struct net *, 46 const struct svc_program *, 47 u32, int, 48 unsigned short, 49 unsigned short); 50 static __be32 nfsd_init_request(struct svc_rqst *, 51 const struct svc_program *, 52 struct svc_process_info *); 53 54 /* 55 * nfsd_mutex protects nn->nfsd_serv -- both the pointer itself and the members 56 * of the svc_serv struct. In particular, ->sv_nrthreads but also to some 57 * extent ->sv_temp_socks and ->sv_permsocks. It also protects nfsdstats.th_cnt 58 * 59 * If (out side the lock) nn->nfsd_serv is non-NULL, then it must point to a 60 * properly initialised 'struct svc_serv' with ->sv_nrthreads > 0. That number 61 * of nfsd threads must exist and each must listed in ->sp_all_threads in each 62 * entry of ->sv_pools[]. 63 * 64 * Transitions of the thread count between zero and non-zero are of particular 65 * interest since the svc_serv needs to be created and initialized at that 66 * point, or freed. 67 * 68 * Finally, the nfsd_mutex also protects some of the global variables that are 69 * accessed when nfsd starts and that are settable via the write_* routines in 70 * nfsctl.c. In particular: 71 * 72 * user_recovery_dirname 73 * user_lease_time 74 * nfsd_versions 75 */ 76 DEFINE_MUTEX(nfsd_mutex); 77 78 /* 79 * nfsd_drc_lock protects nfsd_drc_max_pages and nfsd_drc_pages_used. 80 * nfsd_drc_max_pages limits the total amount of memory available for 81 * version 4.1 DRC caches. 82 * nfsd_drc_pages_used tracks the current version 4.1 DRC memory usage. 83 */ 84 spinlock_t nfsd_drc_lock; 85 unsigned long nfsd_drc_max_mem; 86 unsigned long nfsd_drc_mem_used; 87 88 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) 89 static struct svc_stat nfsd_acl_svcstats; 90 static const struct svc_version *nfsd_acl_version[] = { 91 [2] = &nfsd_acl_version2, 92 [3] = &nfsd_acl_version3, 93 }; 94 95 #define NFSD_ACL_MINVERS 2 96 #define NFSD_ACL_NRVERS ARRAY_SIZE(nfsd_acl_version) 97 static const struct svc_version *nfsd_acl_versions[NFSD_ACL_NRVERS]; 98 99 static struct svc_program nfsd_acl_program = { 100 .pg_prog = NFS_ACL_PROGRAM, 101 .pg_nvers = NFSD_ACL_NRVERS, 102 .pg_vers = nfsd_acl_versions, 103 .pg_name = "nfsacl", 104 .pg_class = "nfsd", 105 .pg_stats = &nfsd_acl_svcstats, 106 .pg_authenticate = &svc_set_client, 107 .pg_init_request = nfsd_acl_init_request, 108 .pg_rpcbind_set = nfsd_acl_rpcbind_set, 109 }; 110 111 static struct svc_stat nfsd_acl_svcstats = { 112 .program = &nfsd_acl_program, 113 }; 114 #endif /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */ 115 116 static const struct svc_version *nfsd_version[] = { 117 [2] = &nfsd_version2, 118 #if defined(CONFIG_NFSD_V3) 119 [3] = &nfsd_version3, 120 #endif 121 #if defined(CONFIG_NFSD_V4) 122 [4] = &nfsd_version4, 123 #endif 124 }; 125 126 #define NFSD_MINVERS 2 127 #define NFSD_NRVERS ARRAY_SIZE(nfsd_version) 128 129 struct svc_program nfsd_program = { 130 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) 131 .pg_next = &nfsd_acl_program, 132 #endif 133 .pg_prog = NFS_PROGRAM, /* program number */ 134 .pg_nvers = NFSD_NRVERS, /* nr of entries in nfsd_version */ 135 .pg_vers = nfsd_version, /* version table */ 136 .pg_name = "nfsd", /* program name */ 137 .pg_class = "nfsd", /* authentication class */ 138 .pg_stats = &nfsd_svcstats, /* version table */ 139 .pg_authenticate = &svc_set_client, /* export authentication */ 140 .pg_init_request = nfsd_init_request, 141 .pg_rpcbind_set = nfsd_rpcbind_set, 142 }; 143 144 static bool 145 nfsd_support_version(int vers) 146 { 147 if (vers >= NFSD_MINVERS && vers < NFSD_NRVERS) 148 return nfsd_version[vers] != NULL; 149 return false; 150 } 151 152 static bool * 153 nfsd_alloc_versions(void) 154 { 155 bool *vers = kmalloc_array(NFSD_NRVERS, sizeof(bool), GFP_KERNEL); 156 unsigned i; 157 158 if (vers) { 159 /* All compiled versions are enabled by default */ 160 for (i = 0; i < NFSD_NRVERS; i++) 161 vers[i] = nfsd_support_version(i); 162 } 163 return vers; 164 } 165 166 static bool * 167 nfsd_alloc_minorversions(void) 168 { 169 bool *vers = kmalloc_array(NFSD_SUPPORTED_MINOR_VERSION + 1, 170 sizeof(bool), GFP_KERNEL); 171 unsigned i; 172 173 if (vers) { 174 /* All minor versions are enabled by default */ 175 for (i = 0; i <= NFSD_SUPPORTED_MINOR_VERSION; i++) 176 vers[i] = nfsd_support_version(4); 177 } 178 return vers; 179 } 180 181 void 182 nfsd_netns_free_versions(struct nfsd_net *nn) 183 { 184 kfree(nn->nfsd_versions); 185 kfree(nn->nfsd4_minorversions); 186 nn->nfsd_versions = NULL; 187 nn->nfsd4_minorversions = NULL; 188 } 189 190 static void 191 nfsd_netns_init_versions(struct nfsd_net *nn) 192 { 193 if (!nn->nfsd_versions) { 194 nn->nfsd_versions = nfsd_alloc_versions(); 195 nn->nfsd4_minorversions = nfsd_alloc_minorversions(); 196 if (!nn->nfsd_versions || !nn->nfsd4_minorversions) 197 nfsd_netns_free_versions(nn); 198 } 199 } 200 201 int nfsd_vers(struct nfsd_net *nn, int vers, enum vers_op change) 202 { 203 if (vers < NFSD_MINVERS || vers >= NFSD_NRVERS) 204 return 0; 205 switch(change) { 206 case NFSD_SET: 207 if (nn->nfsd_versions) 208 nn->nfsd_versions[vers] = nfsd_support_version(vers); 209 break; 210 case NFSD_CLEAR: 211 nfsd_netns_init_versions(nn); 212 if (nn->nfsd_versions) 213 nn->nfsd_versions[vers] = false; 214 break; 215 case NFSD_TEST: 216 if (nn->nfsd_versions) 217 return nn->nfsd_versions[vers]; 218 /* Fallthrough */ 219 case NFSD_AVAIL: 220 return nfsd_support_version(vers); 221 } 222 return 0; 223 } 224 225 static void 226 nfsd_adjust_nfsd_versions4(struct nfsd_net *nn) 227 { 228 unsigned i; 229 230 for (i = 0; i <= NFSD_SUPPORTED_MINOR_VERSION; i++) { 231 if (nn->nfsd4_minorversions[i]) 232 return; 233 } 234 nfsd_vers(nn, 4, NFSD_CLEAR); 235 } 236 237 int nfsd_minorversion(struct nfsd_net *nn, u32 minorversion, enum vers_op change) 238 { 239 if (minorversion > NFSD_SUPPORTED_MINOR_VERSION && 240 change != NFSD_AVAIL) 241 return -1; 242 243 switch(change) { 244 case NFSD_SET: 245 if (nn->nfsd4_minorversions) { 246 nfsd_vers(nn, 4, NFSD_SET); 247 nn->nfsd4_minorversions[minorversion] = 248 nfsd_vers(nn, 4, NFSD_TEST); 249 } 250 break; 251 case NFSD_CLEAR: 252 nfsd_netns_init_versions(nn); 253 if (nn->nfsd4_minorversions) { 254 nn->nfsd4_minorversions[minorversion] = false; 255 nfsd_adjust_nfsd_versions4(nn); 256 } 257 break; 258 case NFSD_TEST: 259 if (nn->nfsd4_minorversions) 260 return nn->nfsd4_minorversions[minorversion]; 261 return nfsd_vers(nn, 4, NFSD_TEST); 262 case NFSD_AVAIL: 263 return minorversion <= NFSD_SUPPORTED_MINOR_VERSION && 264 nfsd_vers(nn, 4, NFSD_AVAIL); 265 } 266 return 0; 267 } 268 269 /* 270 * Maximum number of nfsd processes 271 */ 272 #define NFSD_MAXSERVS 8192 273 274 int nfsd_nrthreads(struct net *net) 275 { 276 int rv = 0; 277 struct nfsd_net *nn = net_generic(net, nfsd_net_id); 278 279 mutex_lock(&nfsd_mutex); 280 if (nn->nfsd_serv) 281 rv = nn->nfsd_serv->sv_nrthreads; 282 mutex_unlock(&nfsd_mutex); 283 return rv; 284 } 285 286 static int nfsd_init_socks(struct net *net, const struct cred *cred) 287 { 288 int error; 289 struct nfsd_net *nn = net_generic(net, nfsd_net_id); 290 291 if (!list_empty(&nn->nfsd_serv->sv_permsocks)) 292 return 0; 293 294 error = svc_create_xprt(nn->nfsd_serv, "udp", net, PF_INET, NFS_PORT, 295 SVC_SOCK_DEFAULTS, cred); 296 if (error < 0) 297 return error; 298 299 error = svc_create_xprt(nn->nfsd_serv, "tcp", net, PF_INET, NFS_PORT, 300 SVC_SOCK_DEFAULTS, cred); 301 if (error < 0) 302 return error; 303 304 return 0; 305 } 306 307 static int nfsd_users = 0; 308 309 static int nfsd_startup_generic(int nrservs) 310 { 311 int ret; 312 313 if (nfsd_users++) 314 return 0; 315 316 /* 317 * Readahead param cache - will no-op if it already exists. 318 * (Note therefore results will be suboptimal if number of 319 * threads is modified after nfsd start.) 320 */ 321 ret = nfsd_racache_init(2*nrservs); 322 if (ret) 323 goto dec_users; 324 325 ret = nfs4_state_start(); 326 if (ret) 327 goto out_racache; 328 return 0; 329 330 out_racache: 331 nfsd_racache_shutdown(); 332 dec_users: 333 nfsd_users--; 334 return ret; 335 } 336 337 static void nfsd_shutdown_generic(void) 338 { 339 if (--nfsd_users) 340 return; 341 342 nfs4_state_shutdown(); 343 nfsd_racache_shutdown(); 344 } 345 346 static bool nfsd_needs_lockd(struct nfsd_net *nn) 347 { 348 return nfsd_vers(nn, 2, NFSD_TEST) || nfsd_vers(nn, 3, NFSD_TEST); 349 } 350 351 static int nfsd_startup_net(int nrservs, struct net *net, const struct cred *cred) 352 { 353 struct nfsd_net *nn = net_generic(net, nfsd_net_id); 354 int ret; 355 356 if (nn->nfsd_net_up) 357 return 0; 358 359 ret = nfsd_startup_generic(nrservs); 360 if (ret) 361 return ret; 362 ret = nfsd_init_socks(net, cred); 363 if (ret) 364 goto out_socks; 365 366 if (nfsd_needs_lockd(nn) && !nn->lockd_up) { 367 ret = lockd_up(net, cred); 368 if (ret) 369 goto out_socks; 370 nn->lockd_up = 1; 371 } 372 373 ret = nfs4_state_start_net(net); 374 if (ret) 375 goto out_lockd; 376 377 nn->nfsd_net_up = true; 378 return 0; 379 380 out_lockd: 381 if (nn->lockd_up) { 382 lockd_down(net); 383 nn->lockd_up = 0; 384 } 385 out_socks: 386 nfsd_shutdown_generic(); 387 return ret; 388 } 389 390 static void nfsd_shutdown_net(struct net *net) 391 { 392 struct nfsd_net *nn = net_generic(net, nfsd_net_id); 393 394 nfs4_state_shutdown_net(net); 395 if (nn->lockd_up) { 396 lockd_down(net); 397 nn->lockd_up = 0; 398 } 399 nn->nfsd_net_up = false; 400 nfsd_shutdown_generic(); 401 } 402 403 static int nfsd_inetaddr_event(struct notifier_block *this, unsigned long event, 404 void *ptr) 405 { 406 struct in_ifaddr *ifa = (struct in_ifaddr *)ptr; 407 struct net_device *dev = ifa->ifa_dev->dev; 408 struct net *net = dev_net(dev); 409 struct nfsd_net *nn = net_generic(net, nfsd_net_id); 410 struct sockaddr_in sin; 411 412 if ((event != NETDEV_DOWN) || 413 !atomic_inc_not_zero(&nn->ntf_refcnt)) 414 goto out; 415 416 if (nn->nfsd_serv) { 417 dprintk("nfsd_inetaddr_event: removed %pI4\n", &ifa->ifa_local); 418 sin.sin_family = AF_INET; 419 sin.sin_addr.s_addr = ifa->ifa_local; 420 svc_age_temp_xprts_now(nn->nfsd_serv, (struct sockaddr *)&sin); 421 } 422 atomic_dec(&nn->ntf_refcnt); 423 wake_up(&nn->ntf_wq); 424 425 out: 426 return NOTIFY_DONE; 427 } 428 429 static struct notifier_block nfsd_inetaddr_notifier = { 430 .notifier_call = nfsd_inetaddr_event, 431 }; 432 433 #if IS_ENABLED(CONFIG_IPV6) 434 static int nfsd_inet6addr_event(struct notifier_block *this, 435 unsigned long event, void *ptr) 436 { 437 struct inet6_ifaddr *ifa = (struct inet6_ifaddr *)ptr; 438 struct net_device *dev = ifa->idev->dev; 439 struct net *net = dev_net(dev); 440 struct nfsd_net *nn = net_generic(net, nfsd_net_id); 441 struct sockaddr_in6 sin6; 442 443 if ((event != NETDEV_DOWN) || 444 !atomic_inc_not_zero(&nn->ntf_refcnt)) 445 goto out; 446 447 if (nn->nfsd_serv) { 448 dprintk("nfsd_inet6addr_event: removed %pI6\n", &ifa->addr); 449 sin6.sin6_family = AF_INET6; 450 sin6.sin6_addr = ifa->addr; 451 if (ipv6_addr_type(&sin6.sin6_addr) & IPV6_ADDR_LINKLOCAL) 452 sin6.sin6_scope_id = ifa->idev->dev->ifindex; 453 svc_age_temp_xprts_now(nn->nfsd_serv, (struct sockaddr *)&sin6); 454 } 455 atomic_dec(&nn->ntf_refcnt); 456 wake_up(&nn->ntf_wq); 457 out: 458 return NOTIFY_DONE; 459 } 460 461 static struct notifier_block nfsd_inet6addr_notifier = { 462 .notifier_call = nfsd_inet6addr_event, 463 }; 464 #endif 465 466 /* Only used under nfsd_mutex, so this atomic may be overkill: */ 467 static atomic_t nfsd_notifier_refcount = ATOMIC_INIT(0); 468 469 static void nfsd_last_thread(struct svc_serv *serv, struct net *net) 470 { 471 struct nfsd_net *nn = net_generic(net, nfsd_net_id); 472 473 atomic_dec(&nn->ntf_refcnt); 474 /* check if the notifier still has clients */ 475 if (atomic_dec_return(&nfsd_notifier_refcount) == 0) { 476 unregister_inetaddr_notifier(&nfsd_inetaddr_notifier); 477 #if IS_ENABLED(CONFIG_IPV6) 478 unregister_inet6addr_notifier(&nfsd_inet6addr_notifier); 479 #endif 480 } 481 wait_event(nn->ntf_wq, atomic_read(&nn->ntf_refcnt) == 0); 482 483 /* 484 * write_ports can create the server without actually starting 485 * any threads--if we get shut down before any threads are 486 * started, then nfsd_last_thread will be run before any of this 487 * other initialization has been done except the rpcb information. 488 */ 489 svc_rpcb_cleanup(serv, net); 490 if (!nn->nfsd_net_up) 491 return; 492 493 nfsd_shutdown_net(net); 494 printk(KERN_WARNING "nfsd: last server has exited, flushing export " 495 "cache\n"); 496 nfsd_export_flush(net); 497 } 498 499 void nfsd_reset_versions(struct nfsd_net *nn) 500 { 501 int i; 502 503 for (i = 0; i < NFSD_NRVERS; i++) 504 if (nfsd_vers(nn, i, NFSD_TEST)) 505 return; 506 507 for (i = 0; i < NFSD_NRVERS; i++) 508 if (i != 4) 509 nfsd_vers(nn, i, NFSD_SET); 510 else { 511 int minor = 0; 512 while (nfsd_minorversion(nn, minor, NFSD_SET) >= 0) 513 minor++; 514 } 515 } 516 517 /* 518 * Each session guarantees a negotiated per slot memory cache for replies 519 * which in turn consumes memory beyond the v2/v3/v4.0 server. A dedicated 520 * NFSv4.1 server might want to use more memory for a DRC than a machine 521 * with mutiple services. 522 * 523 * Impose a hard limit on the number of pages for the DRC which varies 524 * according to the machines free pages. This is of course only a default. 525 * 526 * For now this is a #defined shift which could be under admin control 527 * in the future. 528 */ 529 static void set_max_drc(void) 530 { 531 #define NFSD_DRC_SIZE_SHIFT 7 532 nfsd_drc_max_mem = (nr_free_buffer_pages() 533 >> NFSD_DRC_SIZE_SHIFT) * PAGE_SIZE; 534 nfsd_drc_mem_used = 0; 535 spin_lock_init(&nfsd_drc_lock); 536 dprintk("%s nfsd_drc_max_mem %lu \n", __func__, nfsd_drc_max_mem); 537 } 538 539 static int nfsd_get_default_max_blksize(void) 540 { 541 struct sysinfo i; 542 unsigned long long target; 543 unsigned long ret; 544 545 si_meminfo(&i); 546 target = (i.totalram - i.totalhigh) << PAGE_SHIFT; 547 /* 548 * Aim for 1/4096 of memory per thread This gives 1MB on 4Gig 549 * machines, but only uses 32K on 128M machines. Bottom out at 550 * 8K on 32M and smaller. Of course, this is only a default. 551 */ 552 target >>= 12; 553 554 ret = NFSSVC_MAXBLKSIZE; 555 while (ret > target && ret >= 8*1024*2) 556 ret /= 2; 557 return ret; 558 } 559 560 static const struct svc_serv_ops nfsd_thread_sv_ops = { 561 .svo_shutdown = nfsd_last_thread, 562 .svo_function = nfsd, 563 .svo_enqueue_xprt = svc_xprt_do_enqueue, 564 .svo_setup = svc_set_num_threads, 565 .svo_module = THIS_MODULE, 566 }; 567 568 int nfsd_create_serv(struct net *net) 569 { 570 int error; 571 struct nfsd_net *nn = net_generic(net, nfsd_net_id); 572 573 WARN_ON(!mutex_is_locked(&nfsd_mutex)); 574 if (nn->nfsd_serv) { 575 svc_get(nn->nfsd_serv); 576 return 0; 577 } 578 if (nfsd_max_blksize == 0) 579 nfsd_max_blksize = nfsd_get_default_max_blksize(); 580 nfsd_reset_versions(nn); 581 nn->nfsd_serv = svc_create_pooled(&nfsd_program, nfsd_max_blksize, 582 &nfsd_thread_sv_ops); 583 if (nn->nfsd_serv == NULL) 584 return -ENOMEM; 585 586 nn->nfsd_serv->sv_maxconn = nn->max_connections; 587 error = svc_bind(nn->nfsd_serv, net); 588 if (error < 0) { 589 svc_destroy(nn->nfsd_serv); 590 return error; 591 } 592 593 set_max_drc(); 594 /* check if the notifier is already set */ 595 if (atomic_inc_return(&nfsd_notifier_refcount) == 1) { 596 register_inetaddr_notifier(&nfsd_inetaddr_notifier); 597 #if IS_ENABLED(CONFIG_IPV6) 598 register_inet6addr_notifier(&nfsd_inet6addr_notifier); 599 #endif 600 } 601 atomic_inc(&nn->ntf_refcnt); 602 ktime_get_real_ts64(&nn->nfssvc_boot); /* record boot time */ 603 return 0; 604 } 605 606 int nfsd_nrpools(struct net *net) 607 { 608 struct nfsd_net *nn = net_generic(net, nfsd_net_id); 609 610 if (nn->nfsd_serv == NULL) 611 return 0; 612 else 613 return nn->nfsd_serv->sv_nrpools; 614 } 615 616 int nfsd_get_nrthreads(int n, int *nthreads, struct net *net) 617 { 618 int i = 0; 619 struct nfsd_net *nn = net_generic(net, nfsd_net_id); 620 621 if (nn->nfsd_serv != NULL) { 622 for (i = 0; i < nn->nfsd_serv->sv_nrpools && i < n; i++) 623 nthreads[i] = nn->nfsd_serv->sv_pools[i].sp_nrthreads; 624 } 625 626 return 0; 627 } 628 629 void nfsd_destroy(struct net *net) 630 { 631 struct nfsd_net *nn = net_generic(net, nfsd_net_id); 632 int destroy = (nn->nfsd_serv->sv_nrthreads == 1); 633 634 if (destroy) 635 svc_shutdown_net(nn->nfsd_serv, net); 636 svc_destroy(nn->nfsd_serv); 637 if (destroy) 638 nn->nfsd_serv = NULL; 639 } 640 641 int nfsd_set_nrthreads(int n, int *nthreads, struct net *net) 642 { 643 int i = 0; 644 int tot = 0; 645 int err = 0; 646 struct nfsd_net *nn = net_generic(net, nfsd_net_id); 647 648 WARN_ON(!mutex_is_locked(&nfsd_mutex)); 649 650 if (nn->nfsd_serv == NULL || n <= 0) 651 return 0; 652 653 if (n > nn->nfsd_serv->sv_nrpools) 654 n = nn->nfsd_serv->sv_nrpools; 655 656 /* enforce a global maximum number of threads */ 657 tot = 0; 658 for (i = 0; i < n; i++) { 659 nthreads[i] = min(nthreads[i], NFSD_MAXSERVS); 660 tot += nthreads[i]; 661 } 662 if (tot > NFSD_MAXSERVS) { 663 /* total too large: scale down requested numbers */ 664 for (i = 0; i < n && tot > 0; i++) { 665 int new = nthreads[i] * NFSD_MAXSERVS / tot; 666 tot -= (nthreads[i] - new); 667 nthreads[i] = new; 668 } 669 for (i = 0; i < n && tot > 0; i++) { 670 nthreads[i]--; 671 tot--; 672 } 673 } 674 675 /* 676 * There must always be a thread in pool 0; the admin 677 * can't shut down NFS completely using pool_threads. 678 */ 679 if (nthreads[0] == 0) 680 nthreads[0] = 1; 681 682 /* apply the new numbers */ 683 svc_get(nn->nfsd_serv); 684 for (i = 0; i < n; i++) { 685 err = nn->nfsd_serv->sv_ops->svo_setup(nn->nfsd_serv, 686 &nn->nfsd_serv->sv_pools[i], nthreads[i]); 687 if (err) 688 break; 689 } 690 nfsd_destroy(net); 691 return err; 692 } 693 694 /* 695 * Adjust the number of threads and return the new number of threads. 696 * This is also the function that starts the server if necessary, if 697 * this is the first time nrservs is nonzero. 698 */ 699 int 700 nfsd_svc(int nrservs, struct net *net, const struct cred *cred) 701 { 702 int error; 703 bool nfsd_up_before; 704 struct nfsd_net *nn = net_generic(net, nfsd_net_id); 705 706 mutex_lock(&nfsd_mutex); 707 dprintk("nfsd: creating service\n"); 708 709 nrservs = max(nrservs, 0); 710 nrservs = min(nrservs, NFSD_MAXSERVS); 711 error = 0; 712 713 if (nrservs == 0 && nn->nfsd_serv == NULL) 714 goto out; 715 716 error = nfsd_create_serv(net); 717 if (error) 718 goto out; 719 720 nfsd_up_before = nn->nfsd_net_up; 721 722 error = nfsd_startup_net(nrservs, net, cred); 723 if (error) 724 goto out_destroy; 725 error = nn->nfsd_serv->sv_ops->svo_setup(nn->nfsd_serv, 726 NULL, nrservs); 727 if (error) 728 goto out_shutdown; 729 /* We are holding a reference to nn->nfsd_serv which 730 * we don't want to count in the return value, 731 * so subtract 1 732 */ 733 error = nn->nfsd_serv->sv_nrthreads - 1; 734 out_shutdown: 735 if (error < 0 && !nfsd_up_before) 736 nfsd_shutdown_net(net); 737 out_destroy: 738 nfsd_destroy(net); /* Release server */ 739 out: 740 mutex_unlock(&nfsd_mutex); 741 return error; 742 } 743 744 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) 745 static bool 746 nfsd_support_acl_version(int vers) 747 { 748 if (vers >= NFSD_ACL_MINVERS && vers < NFSD_ACL_NRVERS) 749 return nfsd_acl_version[vers] != NULL; 750 return false; 751 } 752 753 static int 754 nfsd_acl_rpcbind_set(struct net *net, const struct svc_program *progp, 755 u32 version, int family, unsigned short proto, 756 unsigned short port) 757 { 758 if (!nfsd_support_acl_version(version) || 759 !nfsd_vers(net_generic(net, nfsd_net_id), version, NFSD_TEST)) 760 return 0; 761 return svc_generic_rpcbind_set(net, progp, version, family, 762 proto, port); 763 } 764 765 static __be32 766 nfsd_acl_init_request(struct svc_rqst *rqstp, 767 const struct svc_program *progp, 768 struct svc_process_info *ret) 769 { 770 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id); 771 int i; 772 773 if (likely(nfsd_support_acl_version(rqstp->rq_vers) && 774 nfsd_vers(nn, rqstp->rq_vers, NFSD_TEST))) 775 return svc_generic_init_request(rqstp, progp, ret); 776 777 ret->mismatch.lovers = NFSD_ACL_NRVERS; 778 for (i = NFSD_ACL_MINVERS; i < NFSD_ACL_NRVERS; i++) { 779 if (nfsd_support_acl_version(rqstp->rq_vers) && 780 nfsd_vers(nn, i, NFSD_TEST)) { 781 ret->mismatch.lovers = i; 782 break; 783 } 784 } 785 if (ret->mismatch.lovers == NFSD_ACL_NRVERS) 786 return rpc_prog_unavail; 787 ret->mismatch.hivers = NFSD_ACL_MINVERS; 788 for (i = NFSD_ACL_NRVERS - 1; i >= NFSD_ACL_MINVERS; i--) { 789 if (nfsd_support_acl_version(rqstp->rq_vers) && 790 nfsd_vers(nn, i, NFSD_TEST)) { 791 ret->mismatch.hivers = i; 792 break; 793 } 794 } 795 return rpc_prog_mismatch; 796 } 797 #endif 798 799 static int 800 nfsd_rpcbind_set(struct net *net, const struct svc_program *progp, 801 u32 version, int family, unsigned short proto, 802 unsigned short port) 803 { 804 if (!nfsd_vers(net_generic(net, nfsd_net_id), version, NFSD_TEST)) 805 return 0; 806 return svc_generic_rpcbind_set(net, progp, version, family, 807 proto, port); 808 } 809 810 static __be32 811 nfsd_init_request(struct svc_rqst *rqstp, 812 const struct svc_program *progp, 813 struct svc_process_info *ret) 814 { 815 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id); 816 int i; 817 818 if (likely(nfsd_vers(nn, rqstp->rq_vers, NFSD_TEST))) 819 return svc_generic_init_request(rqstp, progp, ret); 820 821 ret->mismatch.lovers = NFSD_NRVERS; 822 for (i = NFSD_MINVERS; i < NFSD_NRVERS; i++) { 823 if (nfsd_vers(nn, i, NFSD_TEST)) { 824 ret->mismatch.lovers = i; 825 break; 826 } 827 } 828 if (ret->mismatch.lovers == NFSD_NRVERS) 829 return rpc_prog_unavail; 830 ret->mismatch.hivers = NFSD_MINVERS; 831 for (i = NFSD_NRVERS - 1; i >= NFSD_MINVERS; i--) { 832 if (nfsd_vers(nn, i, NFSD_TEST)) { 833 ret->mismatch.hivers = i; 834 break; 835 } 836 } 837 return rpc_prog_mismatch; 838 } 839 840 /* 841 * This is the NFS server kernel thread 842 */ 843 static int 844 nfsd(void *vrqstp) 845 { 846 struct svc_rqst *rqstp = (struct svc_rqst *) vrqstp; 847 struct svc_xprt *perm_sock = list_entry(rqstp->rq_server->sv_permsocks.next, typeof(struct svc_xprt), xpt_list); 848 struct net *net = perm_sock->xpt_net; 849 struct nfsd_net *nn = net_generic(net, nfsd_net_id); 850 int err; 851 852 /* Lock module and set up kernel thread */ 853 mutex_lock(&nfsd_mutex); 854 855 /* At this point, the thread shares current->fs 856 * with the init process. We need to create files with the 857 * umask as defined by the client instead of init's umask. */ 858 if (unshare_fs_struct() < 0) { 859 printk("Unable to start nfsd thread: out of memory\n"); 860 goto out; 861 } 862 863 current->fs->umask = 0; 864 865 /* 866 * thread is spawned with all signals set to SIG_IGN, re-enable 867 * the ones that will bring down the thread 868 */ 869 allow_signal(SIGKILL); 870 allow_signal(SIGHUP); 871 allow_signal(SIGINT); 872 allow_signal(SIGQUIT); 873 874 nfsdstats.th_cnt++; 875 mutex_unlock(&nfsd_mutex); 876 877 set_freezable(); 878 879 /* 880 * The main request loop 881 */ 882 for (;;) { 883 /* Update sv_maxconn if it has changed */ 884 rqstp->rq_server->sv_maxconn = nn->max_connections; 885 886 /* 887 * Find a socket with data available and call its 888 * recvfrom routine. 889 */ 890 while ((err = svc_recv(rqstp, 60*60*HZ)) == -EAGAIN) 891 ; 892 if (err == -EINTR) 893 break; 894 validate_process_creds(); 895 svc_process(rqstp); 896 validate_process_creds(); 897 } 898 899 /* Clear signals before calling svc_exit_thread() */ 900 flush_signals(current); 901 902 mutex_lock(&nfsd_mutex); 903 nfsdstats.th_cnt --; 904 905 out: 906 rqstp->rq_server = NULL; 907 908 /* Release the thread */ 909 svc_exit_thread(rqstp); 910 911 nfsd_destroy(net); 912 913 /* Release module */ 914 mutex_unlock(&nfsd_mutex); 915 module_put_and_exit(0); 916 return 0; 917 } 918 919 static __be32 map_new_errors(u32 vers, __be32 nfserr) 920 { 921 if (nfserr == nfserr_jukebox && vers == 2) 922 return nfserr_dropit; 923 if (nfserr == nfserr_wrongsec && vers < 4) 924 return nfserr_acces; 925 return nfserr; 926 } 927 928 /* 929 * A write procedure can have a large argument, and a read procedure can 930 * have a large reply, but no NFSv2 or NFSv3 procedure has argument and 931 * reply that can both be larger than a page. The xdr code has taken 932 * advantage of this assumption to be a sloppy about bounds checking in 933 * some cases. Pending a rewrite of the NFSv2/v3 xdr code to fix that 934 * problem, we enforce these assumptions here: 935 */ 936 static bool nfs_request_too_big(struct svc_rqst *rqstp, 937 const struct svc_procedure *proc) 938 { 939 /* 940 * The ACL code has more careful bounds-checking and is not 941 * susceptible to this problem: 942 */ 943 if (rqstp->rq_prog != NFS_PROGRAM) 944 return false; 945 /* 946 * Ditto NFSv4 (which can in theory have argument and reply both 947 * more than a page): 948 */ 949 if (rqstp->rq_vers >= 4) 950 return false; 951 /* The reply will be small, we're OK: */ 952 if (proc->pc_xdrressize > 0 && 953 proc->pc_xdrressize < XDR_QUADLEN(PAGE_SIZE)) 954 return false; 955 956 return rqstp->rq_arg.len > PAGE_SIZE; 957 } 958 959 int 960 nfsd_dispatch(struct svc_rqst *rqstp, __be32 *statp) 961 { 962 const struct svc_procedure *proc; 963 __be32 nfserr; 964 __be32 *nfserrp; 965 966 dprintk("nfsd_dispatch: vers %d proc %d\n", 967 rqstp->rq_vers, rqstp->rq_proc); 968 proc = rqstp->rq_procinfo; 969 970 if (nfs_request_too_big(rqstp, proc)) { 971 dprintk("nfsd: NFSv%d argument too large\n", rqstp->rq_vers); 972 *statp = rpc_garbage_args; 973 return 1; 974 } 975 /* 976 * Give the xdr decoder a chance to change this if it wants 977 * (necessary in the NFSv4.0 compound case) 978 */ 979 rqstp->rq_cachetype = proc->pc_cachetype; 980 /* Decode arguments */ 981 if (proc->pc_decode && 982 !proc->pc_decode(rqstp, (__be32*)rqstp->rq_arg.head[0].iov_base)) { 983 dprintk("nfsd: failed to decode arguments!\n"); 984 *statp = rpc_garbage_args; 985 return 1; 986 } 987 988 /* Check whether we have this call in the cache. */ 989 switch (nfsd_cache_lookup(rqstp)) { 990 case RC_DROPIT: 991 return 0; 992 case RC_REPLY: 993 return 1; 994 case RC_DOIT:; 995 /* do it */ 996 } 997 998 /* need to grab the location to store the status, as 999 * nfsv4 does some encoding while processing 1000 */ 1001 nfserrp = rqstp->rq_res.head[0].iov_base 1002 + rqstp->rq_res.head[0].iov_len; 1003 rqstp->rq_res.head[0].iov_len += sizeof(__be32); 1004 1005 /* Now call the procedure handler, and encode NFS status. */ 1006 nfserr = proc->pc_func(rqstp); 1007 nfserr = map_new_errors(rqstp->rq_vers, nfserr); 1008 if (nfserr == nfserr_dropit || test_bit(RQ_DROPME, &rqstp->rq_flags)) { 1009 dprintk("nfsd: Dropping request; may be revisited later\n"); 1010 nfsd_cache_update(rqstp, RC_NOCACHE, NULL); 1011 return 0; 1012 } 1013 1014 if (rqstp->rq_proc != 0) 1015 *nfserrp++ = nfserr; 1016 1017 /* Encode result. 1018 * For NFSv2, additional info is never returned in case of an error. 1019 */ 1020 if (!(nfserr && rqstp->rq_vers == 2)) { 1021 if (proc->pc_encode && !proc->pc_encode(rqstp, nfserrp)) { 1022 /* Failed to encode result. Release cache entry */ 1023 dprintk("nfsd: failed to encode result!\n"); 1024 nfsd_cache_update(rqstp, RC_NOCACHE, NULL); 1025 *statp = rpc_system_err; 1026 return 1; 1027 } 1028 } 1029 1030 /* Store reply in cache. */ 1031 nfsd_cache_update(rqstp, rqstp->rq_cachetype, statp + 1); 1032 return 1; 1033 } 1034 1035 int nfsd_pool_stats_open(struct inode *inode, struct file *file) 1036 { 1037 int ret; 1038 struct nfsd_net *nn = net_generic(inode->i_sb->s_fs_info, nfsd_net_id); 1039 1040 mutex_lock(&nfsd_mutex); 1041 if (nn->nfsd_serv == NULL) { 1042 mutex_unlock(&nfsd_mutex); 1043 return -ENODEV; 1044 } 1045 /* bump up the psudo refcount while traversing */ 1046 svc_get(nn->nfsd_serv); 1047 ret = svc_pool_stats_open(nn->nfsd_serv, file); 1048 mutex_unlock(&nfsd_mutex); 1049 return ret; 1050 } 1051 1052 int nfsd_pool_stats_release(struct inode *inode, struct file *file) 1053 { 1054 int ret = seq_release(inode, file); 1055 struct net *net = inode->i_sb->s_fs_info; 1056 1057 mutex_lock(&nfsd_mutex); 1058 /* this function really, really should have been called svc_put() */ 1059 nfsd_destroy(net); 1060 mutex_unlock(&nfsd_mutex); 1061 return ret; 1062 } 1063