1 /* 2 * Central processing for nfsd. 3 * 4 * Authors: Olaf Kirch (okir@monad.swb.de) 5 * 6 * Copyright (C) 1995, 1996, 1997 Olaf Kirch <okir@monad.swb.de> 7 */ 8 9 #include <linux/sched.h> 10 #include <linux/freezer.h> 11 #include <linux/module.h> 12 #include <linux/fs_struct.h> 13 #include <linux/swap.h> 14 #include <linux/nsproxy.h> 15 16 #include <linux/sunrpc/stats.h> 17 #include <linux/sunrpc/svcsock.h> 18 #include <linux/lockd/bind.h> 19 #include <linux/nfsacl.h> 20 #include <linux/seq_file.h> 21 #include <net/net_namespace.h> 22 #include "nfsd.h" 23 #include "cache.h" 24 #include "vfs.h" 25 26 #define NFSDDBG_FACILITY NFSDDBG_SVC 27 28 extern struct svc_program nfsd_program; 29 static int nfsd(void *vrqstp); 30 struct timeval nfssvc_boot; 31 32 /* 33 * nfsd_mutex protects nfsd_serv -- both the pointer itself and the members 34 * of the svc_serv struct. In particular, ->sv_nrthreads but also to some 35 * extent ->sv_temp_socks and ->sv_permsocks. It also protects nfsdstats.th_cnt 36 * 37 * If (out side the lock) nfsd_serv is non-NULL, then it must point to a 38 * properly initialised 'struct svc_serv' with ->sv_nrthreads > 0. That number 39 * of nfsd threads must exist and each must listed in ->sp_all_threads in each 40 * entry of ->sv_pools[]. 41 * 42 * Transitions of the thread count between zero and non-zero are of particular 43 * interest since the svc_serv needs to be created and initialized at that 44 * point, or freed. 45 * 46 * Finally, the nfsd_mutex also protects some of the global variables that are 47 * accessed when nfsd starts and that are settable via the write_* routines in 48 * nfsctl.c. In particular: 49 * 50 * user_recovery_dirname 51 * user_lease_time 52 * nfsd_versions 53 */ 54 DEFINE_MUTEX(nfsd_mutex); 55 struct svc_serv *nfsd_serv; 56 57 /* 58 * nfsd_drc_lock protects nfsd_drc_max_pages and nfsd_drc_pages_used. 59 * nfsd_drc_max_pages limits the total amount of memory available for 60 * version 4.1 DRC caches. 61 * nfsd_drc_pages_used tracks the current version 4.1 DRC memory usage. 62 */ 63 spinlock_t nfsd_drc_lock; 64 unsigned int nfsd_drc_max_mem; 65 unsigned int nfsd_drc_mem_used; 66 67 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) 68 static struct svc_stat nfsd_acl_svcstats; 69 static struct svc_version * nfsd_acl_version[] = { 70 [2] = &nfsd_acl_version2, 71 [3] = &nfsd_acl_version3, 72 }; 73 74 #define NFSD_ACL_MINVERS 2 75 #define NFSD_ACL_NRVERS ARRAY_SIZE(nfsd_acl_version) 76 static struct svc_version *nfsd_acl_versions[NFSD_ACL_NRVERS]; 77 78 static struct svc_program nfsd_acl_program = { 79 .pg_prog = NFS_ACL_PROGRAM, 80 .pg_nvers = NFSD_ACL_NRVERS, 81 .pg_vers = nfsd_acl_versions, 82 .pg_name = "nfsacl", 83 .pg_class = "nfsd", 84 .pg_stats = &nfsd_acl_svcstats, 85 .pg_authenticate = &svc_set_client, 86 }; 87 88 static struct svc_stat nfsd_acl_svcstats = { 89 .program = &nfsd_acl_program, 90 }; 91 #endif /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */ 92 93 static struct svc_version * nfsd_version[] = { 94 [2] = &nfsd_version2, 95 #if defined(CONFIG_NFSD_V3) 96 [3] = &nfsd_version3, 97 #endif 98 #if defined(CONFIG_NFSD_V4) 99 [4] = &nfsd_version4, 100 #endif 101 }; 102 103 #define NFSD_MINVERS 2 104 #define NFSD_NRVERS ARRAY_SIZE(nfsd_version) 105 static struct svc_version *nfsd_versions[NFSD_NRVERS]; 106 107 struct svc_program nfsd_program = { 108 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) 109 .pg_next = &nfsd_acl_program, 110 #endif 111 .pg_prog = NFS_PROGRAM, /* program number */ 112 .pg_nvers = NFSD_NRVERS, /* nr of entries in nfsd_version */ 113 .pg_vers = nfsd_versions, /* version table */ 114 .pg_name = "nfsd", /* program name */ 115 .pg_class = "nfsd", /* authentication class */ 116 .pg_stats = &nfsd_svcstats, /* version table */ 117 .pg_authenticate = &svc_set_client, /* export authentication */ 118 119 }; 120 121 u32 nfsd_supported_minorversion; 122 123 int nfsd_vers(int vers, enum vers_op change) 124 { 125 if (vers < NFSD_MINVERS || vers >= NFSD_NRVERS) 126 return 0; 127 switch(change) { 128 case NFSD_SET: 129 nfsd_versions[vers] = nfsd_version[vers]; 130 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) 131 if (vers < NFSD_ACL_NRVERS) 132 nfsd_acl_versions[vers] = nfsd_acl_version[vers]; 133 #endif 134 break; 135 case NFSD_CLEAR: 136 nfsd_versions[vers] = NULL; 137 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) 138 if (vers < NFSD_ACL_NRVERS) 139 nfsd_acl_versions[vers] = NULL; 140 #endif 141 break; 142 case NFSD_TEST: 143 return nfsd_versions[vers] != NULL; 144 case NFSD_AVAIL: 145 return nfsd_version[vers] != NULL; 146 } 147 return 0; 148 } 149 150 int nfsd_minorversion(u32 minorversion, enum vers_op change) 151 { 152 if (minorversion > NFSD_SUPPORTED_MINOR_VERSION) 153 return -1; 154 switch(change) { 155 case NFSD_SET: 156 nfsd_supported_minorversion = minorversion; 157 break; 158 case NFSD_CLEAR: 159 if (minorversion == 0) 160 return -1; 161 nfsd_supported_minorversion = minorversion - 1; 162 break; 163 case NFSD_TEST: 164 return minorversion <= nfsd_supported_minorversion; 165 case NFSD_AVAIL: 166 return minorversion <= NFSD_SUPPORTED_MINOR_VERSION; 167 } 168 return 0; 169 } 170 171 /* 172 * Maximum number of nfsd processes 173 */ 174 #define NFSD_MAXSERVS 8192 175 176 int nfsd_nrthreads(void) 177 { 178 int rv = 0; 179 mutex_lock(&nfsd_mutex); 180 if (nfsd_serv) 181 rv = nfsd_serv->sv_nrthreads; 182 mutex_unlock(&nfsd_mutex); 183 return rv; 184 } 185 186 static int nfsd_init_socks(int port) 187 { 188 int error; 189 if (!list_empty(&nfsd_serv->sv_permsocks)) 190 return 0; 191 192 error = svc_create_xprt(nfsd_serv, "udp", &init_net, PF_INET, port, 193 SVC_SOCK_DEFAULTS); 194 if (error < 0) 195 return error; 196 197 error = svc_create_xprt(nfsd_serv, "tcp", &init_net, PF_INET, port, 198 SVC_SOCK_DEFAULTS); 199 if (error < 0) 200 return error; 201 202 return 0; 203 } 204 205 static bool nfsd_up = false; 206 207 static int nfsd_startup(unsigned short port, int nrservs) 208 { 209 int ret; 210 211 if (nfsd_up) 212 return 0; 213 /* 214 * Readahead param cache - will no-op if it already exists. 215 * (Note therefore results will be suboptimal if number of 216 * threads is modified after nfsd start.) 217 */ 218 ret = nfsd_racache_init(2*nrservs); 219 if (ret) 220 return ret; 221 ret = nfsd_init_socks(port); 222 if (ret) 223 goto out_racache; 224 ret = lockd_up(&init_net); 225 if (ret) 226 goto out_racache; 227 ret = nfs4_state_start(); 228 if (ret) 229 goto out_lockd; 230 nfsd_up = true; 231 return 0; 232 out_lockd: 233 lockd_down(&init_net); 234 out_racache: 235 nfsd_racache_shutdown(); 236 return ret; 237 } 238 239 static void nfsd_shutdown(void) 240 { 241 /* 242 * write_ports can create the server without actually starting 243 * any threads--if we get shut down before any threads are 244 * started, then nfsd_last_thread will be run before any of this 245 * other initialization has been done. 246 */ 247 if (!nfsd_up) 248 return; 249 nfs4_state_shutdown(); 250 lockd_down(&init_net); 251 nfsd_racache_shutdown(); 252 nfsd_up = false; 253 } 254 255 static void nfsd_last_thread(struct svc_serv *serv, struct net *net) 256 { 257 /* When last nfsd thread exits we need to do some clean-up */ 258 nfsd_serv = NULL; 259 nfsd_shutdown(); 260 261 svc_rpcb_cleanup(serv, net); 262 263 printk(KERN_WARNING "nfsd: last server has exited, flushing export " 264 "cache\n"); 265 nfsd_export_flush(net); 266 } 267 268 void nfsd_reset_versions(void) 269 { 270 int found_one = 0; 271 int i; 272 273 for (i = NFSD_MINVERS; i < NFSD_NRVERS; i++) { 274 if (nfsd_program.pg_vers[i]) 275 found_one = 1; 276 } 277 278 if (!found_one) { 279 for (i = NFSD_MINVERS; i < NFSD_NRVERS; i++) 280 nfsd_program.pg_vers[i] = nfsd_version[i]; 281 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) 282 for (i = NFSD_ACL_MINVERS; i < NFSD_ACL_NRVERS; i++) 283 nfsd_acl_program.pg_vers[i] = 284 nfsd_acl_version[i]; 285 #endif 286 } 287 } 288 289 /* 290 * Each session guarantees a negotiated per slot memory cache for replies 291 * which in turn consumes memory beyond the v2/v3/v4.0 server. A dedicated 292 * NFSv4.1 server might want to use more memory for a DRC than a machine 293 * with mutiple services. 294 * 295 * Impose a hard limit on the number of pages for the DRC which varies 296 * according to the machines free pages. This is of course only a default. 297 * 298 * For now this is a #defined shift which could be under admin control 299 * in the future. 300 */ 301 static void set_max_drc(void) 302 { 303 #define NFSD_DRC_SIZE_SHIFT 10 304 nfsd_drc_max_mem = (nr_free_buffer_pages() 305 >> NFSD_DRC_SIZE_SHIFT) * PAGE_SIZE; 306 nfsd_drc_mem_used = 0; 307 spin_lock_init(&nfsd_drc_lock); 308 dprintk("%s nfsd_drc_max_mem %u \n", __func__, nfsd_drc_max_mem); 309 } 310 311 static int nfsd_get_default_max_blksize(void) 312 { 313 struct sysinfo i; 314 unsigned long long target; 315 unsigned long ret; 316 317 si_meminfo(&i); 318 target = (i.totalram - i.totalhigh) << PAGE_SHIFT; 319 /* 320 * Aim for 1/4096 of memory per thread This gives 1MB on 4Gig 321 * machines, but only uses 32K on 128M machines. Bottom out at 322 * 8K on 32M and smaller. Of course, this is only a default. 323 */ 324 target >>= 12; 325 326 ret = NFSSVC_MAXBLKSIZE; 327 while (ret > target && ret >= 8*1024*2) 328 ret /= 2; 329 return ret; 330 } 331 332 int nfsd_create_serv(void) 333 { 334 int error; 335 336 WARN_ON(!mutex_is_locked(&nfsd_mutex)); 337 if (nfsd_serv) { 338 svc_get(nfsd_serv); 339 return 0; 340 } 341 if (nfsd_max_blksize == 0) 342 nfsd_max_blksize = nfsd_get_default_max_blksize(); 343 nfsd_reset_versions(); 344 nfsd_serv = svc_create_pooled(&nfsd_program, nfsd_max_blksize, 345 nfsd_last_thread, nfsd, THIS_MODULE); 346 if (nfsd_serv == NULL) 347 return -ENOMEM; 348 349 error = svc_bind(nfsd_serv, current->nsproxy->net_ns); 350 if (error < 0) { 351 svc_destroy(nfsd_serv); 352 return error; 353 } 354 355 set_max_drc(); 356 do_gettimeofday(&nfssvc_boot); /* record boot time */ 357 return 0; 358 } 359 360 int nfsd_nrpools(void) 361 { 362 if (nfsd_serv == NULL) 363 return 0; 364 else 365 return nfsd_serv->sv_nrpools; 366 } 367 368 int nfsd_get_nrthreads(int n, int *nthreads) 369 { 370 int i = 0; 371 372 if (nfsd_serv != NULL) { 373 for (i = 0; i < nfsd_serv->sv_nrpools && i < n; i++) 374 nthreads[i] = nfsd_serv->sv_pools[i].sp_nrthreads; 375 } 376 377 return 0; 378 } 379 380 int nfsd_set_nrthreads(int n, int *nthreads) 381 { 382 int i = 0; 383 int tot = 0; 384 int err = 0; 385 struct net *net = &init_net; 386 387 WARN_ON(!mutex_is_locked(&nfsd_mutex)); 388 389 if (nfsd_serv == NULL || n <= 0) 390 return 0; 391 392 if (n > nfsd_serv->sv_nrpools) 393 n = nfsd_serv->sv_nrpools; 394 395 /* enforce a global maximum number of threads */ 396 tot = 0; 397 for (i = 0; i < n; i++) { 398 if (nthreads[i] > NFSD_MAXSERVS) 399 nthreads[i] = NFSD_MAXSERVS; 400 tot += nthreads[i]; 401 } 402 if (tot > NFSD_MAXSERVS) { 403 /* total too large: scale down requested numbers */ 404 for (i = 0; i < n && tot > 0; i++) { 405 int new = nthreads[i] * NFSD_MAXSERVS / tot; 406 tot -= (nthreads[i] - new); 407 nthreads[i] = new; 408 } 409 for (i = 0; i < n && tot > 0; i++) { 410 nthreads[i]--; 411 tot--; 412 } 413 } 414 415 /* 416 * There must always be a thread in pool 0; the admin 417 * can't shut down NFS completely using pool_threads. 418 */ 419 if (nthreads[0] == 0) 420 nthreads[0] = 1; 421 422 /* apply the new numbers */ 423 svc_get(nfsd_serv); 424 for (i = 0; i < n; i++) { 425 err = svc_set_num_threads(nfsd_serv, &nfsd_serv->sv_pools[i], 426 nthreads[i]); 427 if (err) 428 break; 429 } 430 431 if (nfsd_serv->sv_nrthreads == 1) 432 svc_shutdown_net(nfsd_serv, net); 433 svc_destroy(nfsd_serv); 434 435 return err; 436 } 437 438 /* 439 * Adjust the number of threads and return the new number of threads. 440 * This is also the function that starts the server if necessary, if 441 * this is the first time nrservs is nonzero. 442 */ 443 int 444 nfsd_svc(unsigned short port, int nrservs) 445 { 446 int error; 447 bool nfsd_up_before; 448 struct net *net = &init_net; 449 450 mutex_lock(&nfsd_mutex); 451 dprintk("nfsd: creating service\n"); 452 if (nrservs <= 0) 453 nrservs = 0; 454 if (nrservs > NFSD_MAXSERVS) 455 nrservs = NFSD_MAXSERVS; 456 error = 0; 457 if (nrservs == 0 && nfsd_serv == NULL) 458 goto out; 459 460 error = nfsd_create_serv(); 461 if (error) 462 goto out; 463 464 nfsd_up_before = nfsd_up; 465 466 error = nfsd_startup(port, nrservs); 467 if (error) 468 goto out_destroy; 469 error = svc_set_num_threads(nfsd_serv, NULL, nrservs); 470 if (error) 471 goto out_shutdown; 472 /* We are holding a reference to nfsd_serv which 473 * we don't want to count in the return value, 474 * so subtract 1 475 */ 476 error = nfsd_serv->sv_nrthreads - 1; 477 out_shutdown: 478 if (error < 0 && !nfsd_up_before) 479 nfsd_shutdown(); 480 out_destroy: 481 if (nfsd_serv->sv_nrthreads == 1) 482 svc_shutdown_net(nfsd_serv, net); 483 svc_destroy(nfsd_serv); /* Release server */ 484 out: 485 mutex_unlock(&nfsd_mutex); 486 return error; 487 } 488 489 490 /* 491 * This is the NFS server kernel thread 492 */ 493 static int 494 nfsd(void *vrqstp) 495 { 496 struct svc_rqst *rqstp = (struct svc_rqst *) vrqstp; 497 int err, preverr = 0; 498 499 /* Lock module and set up kernel thread */ 500 mutex_lock(&nfsd_mutex); 501 502 /* At this point, the thread shares current->fs 503 * with the init process. We need to create files with a 504 * umask of 0 instead of init's umask. */ 505 if (unshare_fs_struct() < 0) { 506 printk("Unable to start nfsd thread: out of memory\n"); 507 goto out; 508 } 509 510 current->fs->umask = 0; 511 512 /* 513 * thread is spawned with all signals set to SIG_IGN, re-enable 514 * the ones that will bring down the thread 515 */ 516 allow_signal(SIGKILL); 517 allow_signal(SIGHUP); 518 allow_signal(SIGINT); 519 allow_signal(SIGQUIT); 520 521 nfsdstats.th_cnt++; 522 mutex_unlock(&nfsd_mutex); 523 524 /* 525 * We want less throttling in balance_dirty_pages() so that nfs to 526 * localhost doesn't cause nfsd to lock up due to all the client's 527 * dirty pages. 528 */ 529 current->flags |= PF_LESS_THROTTLE; 530 set_freezable(); 531 532 /* 533 * The main request loop 534 */ 535 for (;;) { 536 /* 537 * Find a socket with data available and call its 538 * recvfrom routine. 539 */ 540 while ((err = svc_recv(rqstp, 60*60*HZ)) == -EAGAIN) 541 ; 542 if (err == -EINTR) 543 break; 544 else if (err < 0) { 545 if (err != preverr) { 546 printk(KERN_WARNING "%s: unexpected error " 547 "from svc_recv (%d)\n", __func__, -err); 548 preverr = err; 549 } 550 schedule_timeout_uninterruptible(HZ); 551 continue; 552 } 553 554 validate_process_creds(); 555 svc_process(rqstp); 556 validate_process_creds(); 557 } 558 559 /* Clear signals before calling svc_exit_thread() */ 560 flush_signals(current); 561 562 mutex_lock(&nfsd_mutex); 563 nfsdstats.th_cnt --; 564 565 out: 566 if (rqstp->rq_server->sv_nrthreads == 1) 567 svc_shutdown_net(rqstp->rq_server, &init_net); 568 569 /* Release the thread */ 570 svc_exit_thread(rqstp); 571 572 /* Release module */ 573 mutex_unlock(&nfsd_mutex); 574 module_put_and_exit(0); 575 return 0; 576 } 577 578 static __be32 map_new_errors(u32 vers, __be32 nfserr) 579 { 580 if (nfserr == nfserr_jukebox && vers == 2) 581 return nfserr_dropit; 582 if (nfserr == nfserr_wrongsec && vers < 4) 583 return nfserr_acces; 584 return nfserr; 585 } 586 587 int 588 nfsd_dispatch(struct svc_rqst *rqstp, __be32 *statp) 589 { 590 struct svc_procedure *proc; 591 kxdrproc_t xdr; 592 __be32 nfserr; 593 __be32 *nfserrp; 594 595 dprintk("nfsd_dispatch: vers %d proc %d\n", 596 rqstp->rq_vers, rqstp->rq_proc); 597 proc = rqstp->rq_procinfo; 598 599 /* 600 * Give the xdr decoder a chance to change this if it wants 601 * (necessary in the NFSv4.0 compound case) 602 */ 603 rqstp->rq_cachetype = proc->pc_cachetype; 604 /* Decode arguments */ 605 xdr = proc->pc_decode; 606 if (xdr && !xdr(rqstp, (__be32*)rqstp->rq_arg.head[0].iov_base, 607 rqstp->rq_argp)) { 608 dprintk("nfsd: failed to decode arguments!\n"); 609 *statp = rpc_garbage_args; 610 return 1; 611 } 612 613 /* Check whether we have this call in the cache. */ 614 switch (nfsd_cache_lookup(rqstp)) { 615 case RC_INTR: 616 case RC_DROPIT: 617 return 0; 618 case RC_REPLY: 619 return 1; 620 case RC_DOIT:; 621 /* do it */ 622 } 623 624 /* need to grab the location to store the status, as 625 * nfsv4 does some encoding while processing 626 */ 627 nfserrp = rqstp->rq_res.head[0].iov_base 628 + rqstp->rq_res.head[0].iov_len; 629 rqstp->rq_res.head[0].iov_len += sizeof(__be32); 630 631 /* Now call the procedure handler, and encode NFS status. */ 632 nfserr = proc->pc_func(rqstp, rqstp->rq_argp, rqstp->rq_resp); 633 nfserr = map_new_errors(rqstp->rq_vers, nfserr); 634 if (nfserr == nfserr_dropit || rqstp->rq_dropme) { 635 dprintk("nfsd: Dropping request; may be revisited later\n"); 636 nfsd_cache_update(rqstp, RC_NOCACHE, NULL); 637 return 0; 638 } 639 640 if (rqstp->rq_proc != 0) 641 *nfserrp++ = nfserr; 642 643 /* Encode result. 644 * For NFSv2, additional info is never returned in case of an error. 645 */ 646 if (!(nfserr && rqstp->rq_vers == 2)) { 647 xdr = proc->pc_encode; 648 if (xdr && !xdr(rqstp, nfserrp, 649 rqstp->rq_resp)) { 650 /* Failed to encode result. Release cache entry */ 651 dprintk("nfsd: failed to encode result!\n"); 652 nfsd_cache_update(rqstp, RC_NOCACHE, NULL); 653 *statp = rpc_system_err; 654 return 1; 655 } 656 } 657 658 /* Store reply in cache. */ 659 nfsd_cache_update(rqstp, proc->pc_cachetype, statp + 1); 660 return 1; 661 } 662 663 int nfsd_pool_stats_open(struct inode *inode, struct file *file) 664 { 665 int ret; 666 mutex_lock(&nfsd_mutex); 667 if (nfsd_serv == NULL) { 668 mutex_unlock(&nfsd_mutex); 669 return -ENODEV; 670 } 671 /* bump up the psudo refcount while traversing */ 672 svc_get(nfsd_serv); 673 ret = svc_pool_stats_open(nfsd_serv, file); 674 mutex_unlock(&nfsd_mutex); 675 return ret; 676 } 677 678 int nfsd_pool_stats_release(struct inode *inode, struct file *file) 679 { 680 int ret = seq_release(inode, file); 681 struct net *net = &init_net; 682 683 mutex_lock(&nfsd_mutex); 684 /* this function really, really should have been called svc_put() */ 685 if (nfsd_serv->sv_nrthreads == 1) 686 svc_shutdown_net(nfsd_serv, net); 687 svc_destroy(nfsd_serv); 688 mutex_unlock(&nfsd_mutex); 689 return ret; 690 } 691