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 nfsd_shutdown(); 258 259 svc_rpcb_cleanup(serv, net); 260 261 printk(KERN_WARNING "nfsd: last server has exited, flushing export " 262 "cache\n"); 263 nfsd_export_flush(net); 264 } 265 266 void nfsd_reset_versions(void) 267 { 268 int found_one = 0; 269 int i; 270 271 for (i = NFSD_MINVERS; i < NFSD_NRVERS; i++) { 272 if (nfsd_program.pg_vers[i]) 273 found_one = 1; 274 } 275 276 if (!found_one) { 277 for (i = NFSD_MINVERS; i < NFSD_NRVERS; i++) 278 nfsd_program.pg_vers[i] = nfsd_version[i]; 279 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) 280 for (i = NFSD_ACL_MINVERS; i < NFSD_ACL_NRVERS; i++) 281 nfsd_acl_program.pg_vers[i] = 282 nfsd_acl_version[i]; 283 #endif 284 } 285 } 286 287 /* 288 * Each session guarantees a negotiated per slot memory cache for replies 289 * which in turn consumes memory beyond the v2/v3/v4.0 server. A dedicated 290 * NFSv4.1 server might want to use more memory for a DRC than a machine 291 * with mutiple services. 292 * 293 * Impose a hard limit on the number of pages for the DRC which varies 294 * according to the machines free pages. This is of course only a default. 295 * 296 * For now this is a #defined shift which could be under admin control 297 * in the future. 298 */ 299 static void set_max_drc(void) 300 { 301 #define NFSD_DRC_SIZE_SHIFT 10 302 nfsd_drc_max_mem = (nr_free_buffer_pages() 303 >> NFSD_DRC_SIZE_SHIFT) * PAGE_SIZE; 304 nfsd_drc_mem_used = 0; 305 spin_lock_init(&nfsd_drc_lock); 306 dprintk("%s nfsd_drc_max_mem %u \n", __func__, nfsd_drc_max_mem); 307 } 308 309 static int nfsd_get_default_max_blksize(void) 310 { 311 struct sysinfo i; 312 unsigned long long target; 313 unsigned long ret; 314 315 si_meminfo(&i); 316 target = (i.totalram - i.totalhigh) << PAGE_SHIFT; 317 /* 318 * Aim for 1/4096 of memory per thread This gives 1MB on 4Gig 319 * machines, but only uses 32K on 128M machines. Bottom out at 320 * 8K on 32M and smaller. Of course, this is only a default. 321 */ 322 target >>= 12; 323 324 ret = NFSSVC_MAXBLKSIZE; 325 while (ret > target && ret >= 8*1024*2) 326 ret /= 2; 327 return ret; 328 } 329 330 int nfsd_create_serv(void) 331 { 332 int error; 333 struct net *net = current->nsproxy->net_ns; 334 335 WARN_ON(!mutex_is_locked(&nfsd_mutex)); 336 if (nfsd_serv) { 337 svc_get(nfsd_serv); 338 return 0; 339 } 340 if (nfsd_max_blksize == 0) 341 nfsd_max_blksize = nfsd_get_default_max_blksize(); 342 nfsd_reset_versions(); 343 nfsd_serv = svc_create_pooled(&nfsd_program, nfsd_max_blksize, 344 nfsd_last_thread, nfsd, THIS_MODULE); 345 if (nfsd_serv == NULL) 346 return -ENOMEM; 347 348 error = svc_bind(nfsd_serv, net); 349 if (error < 0) { 350 svc_destroy(nfsd_serv); 351 return error; 352 } 353 354 set_max_drc(); 355 do_gettimeofday(&nfssvc_boot); /* record boot time */ 356 return 0; 357 } 358 359 int nfsd_nrpools(void) 360 { 361 if (nfsd_serv == NULL) 362 return 0; 363 else 364 return nfsd_serv->sv_nrpools; 365 } 366 367 int nfsd_get_nrthreads(int n, int *nthreads) 368 { 369 int i = 0; 370 371 if (nfsd_serv != NULL) { 372 for (i = 0; i < nfsd_serv->sv_nrpools && i < n; i++) 373 nthreads[i] = nfsd_serv->sv_pools[i].sp_nrthreads; 374 } 375 376 return 0; 377 } 378 379 int nfsd_set_nrthreads(int n, int *nthreads) 380 { 381 int i = 0; 382 int tot = 0; 383 int err = 0; 384 struct net *net = &init_net; 385 386 WARN_ON(!mutex_is_locked(&nfsd_mutex)); 387 388 if (nfsd_serv == NULL || n <= 0) 389 return 0; 390 391 if (n > nfsd_serv->sv_nrpools) 392 n = nfsd_serv->sv_nrpools; 393 394 /* enforce a global maximum number of threads */ 395 tot = 0; 396 for (i = 0; i < n; i++) { 397 if (nthreads[i] > NFSD_MAXSERVS) 398 nthreads[i] = NFSD_MAXSERVS; 399 tot += nthreads[i]; 400 } 401 if (tot > NFSD_MAXSERVS) { 402 /* total too large: scale down requested numbers */ 403 for (i = 0; i < n && tot > 0; i++) { 404 int new = nthreads[i] * NFSD_MAXSERVS / tot; 405 tot -= (nthreads[i] - new); 406 nthreads[i] = new; 407 } 408 for (i = 0; i < n && tot > 0; i++) { 409 nthreads[i]--; 410 tot--; 411 } 412 } 413 414 /* 415 * There must always be a thread in pool 0; the admin 416 * can't shut down NFS completely using pool_threads. 417 */ 418 if (nthreads[0] == 0) 419 nthreads[0] = 1; 420 421 /* apply the new numbers */ 422 svc_get(nfsd_serv); 423 for (i = 0; i < n; i++) { 424 err = svc_set_num_threads(nfsd_serv, &nfsd_serv->sv_pools[i], 425 nthreads[i]); 426 if (err) 427 break; 428 } 429 nfsd_destroy(net); 430 return err; 431 } 432 433 /* 434 * Adjust the number of threads and return the new number of threads. 435 * This is also the function that starts the server if necessary, if 436 * this is the first time nrservs is nonzero. 437 */ 438 int 439 nfsd_svc(unsigned short port, int nrservs) 440 { 441 int error; 442 bool nfsd_up_before; 443 struct net *net = &init_net; 444 445 mutex_lock(&nfsd_mutex); 446 dprintk("nfsd: creating service\n"); 447 if (nrservs <= 0) 448 nrservs = 0; 449 if (nrservs > NFSD_MAXSERVS) 450 nrservs = NFSD_MAXSERVS; 451 error = 0; 452 if (nrservs == 0 && nfsd_serv == NULL) 453 goto out; 454 455 error = nfsd_create_serv(); 456 if (error) 457 goto out; 458 459 nfsd_up_before = nfsd_up; 460 461 error = nfsd_startup(port, nrservs); 462 if (error) 463 goto out_destroy; 464 error = svc_set_num_threads(nfsd_serv, NULL, nrservs); 465 if (error) 466 goto out_shutdown; 467 /* We are holding a reference to nfsd_serv which 468 * we don't want to count in the return value, 469 * so subtract 1 470 */ 471 error = nfsd_serv->sv_nrthreads - 1; 472 out_shutdown: 473 if (error < 0 && !nfsd_up_before) 474 nfsd_shutdown(); 475 out_destroy: 476 nfsd_destroy(net); /* Release server */ 477 out: 478 mutex_unlock(&nfsd_mutex); 479 return error; 480 } 481 482 483 /* 484 * This is the NFS server kernel thread 485 */ 486 static int 487 nfsd(void *vrqstp) 488 { 489 struct svc_rqst *rqstp = (struct svc_rqst *) vrqstp; 490 int err, preverr = 0; 491 492 /* Lock module and set up kernel thread */ 493 mutex_lock(&nfsd_mutex); 494 495 /* At this point, the thread shares current->fs 496 * with the init process. We need to create files with a 497 * umask of 0 instead of init's umask. */ 498 if (unshare_fs_struct() < 0) { 499 printk("Unable to start nfsd thread: out of memory\n"); 500 goto out; 501 } 502 503 current->fs->umask = 0; 504 505 /* 506 * thread is spawned with all signals set to SIG_IGN, re-enable 507 * the ones that will bring down the thread 508 */ 509 allow_signal(SIGKILL); 510 allow_signal(SIGHUP); 511 allow_signal(SIGINT); 512 allow_signal(SIGQUIT); 513 514 nfsdstats.th_cnt++; 515 mutex_unlock(&nfsd_mutex); 516 517 /* 518 * We want less throttling in balance_dirty_pages() so that nfs to 519 * localhost doesn't cause nfsd to lock up due to all the client's 520 * dirty pages. 521 */ 522 current->flags |= PF_LESS_THROTTLE; 523 set_freezable(); 524 525 /* 526 * The main request loop 527 */ 528 for (;;) { 529 /* 530 * Find a socket with data available and call its 531 * recvfrom routine. 532 */ 533 while ((err = svc_recv(rqstp, 60*60*HZ)) == -EAGAIN) 534 ; 535 if (err == -EINTR) 536 break; 537 else if (err < 0) { 538 if (err != preverr) { 539 printk(KERN_WARNING "%s: unexpected error " 540 "from svc_recv (%d)\n", __func__, -err); 541 preverr = err; 542 } 543 schedule_timeout_uninterruptible(HZ); 544 continue; 545 } 546 547 validate_process_creds(); 548 svc_process(rqstp); 549 validate_process_creds(); 550 } 551 552 /* Clear signals before calling svc_exit_thread() */ 553 flush_signals(current); 554 555 mutex_lock(&nfsd_mutex); 556 nfsdstats.th_cnt --; 557 558 out: 559 rqstp->rq_server = NULL; 560 561 /* Release the thread */ 562 svc_exit_thread(rqstp); 563 564 nfsd_destroy(&init_net); 565 566 /* Release module */ 567 mutex_unlock(&nfsd_mutex); 568 module_put_and_exit(0); 569 return 0; 570 } 571 572 static __be32 map_new_errors(u32 vers, __be32 nfserr) 573 { 574 if (nfserr == nfserr_jukebox && vers == 2) 575 return nfserr_dropit; 576 if (nfserr == nfserr_wrongsec && vers < 4) 577 return nfserr_acces; 578 return nfserr; 579 } 580 581 int 582 nfsd_dispatch(struct svc_rqst *rqstp, __be32 *statp) 583 { 584 struct svc_procedure *proc; 585 kxdrproc_t xdr; 586 __be32 nfserr; 587 __be32 *nfserrp; 588 589 dprintk("nfsd_dispatch: vers %d proc %d\n", 590 rqstp->rq_vers, rqstp->rq_proc); 591 proc = rqstp->rq_procinfo; 592 593 /* 594 * Give the xdr decoder a chance to change this if it wants 595 * (necessary in the NFSv4.0 compound case) 596 */ 597 rqstp->rq_cachetype = proc->pc_cachetype; 598 /* Decode arguments */ 599 xdr = proc->pc_decode; 600 if (xdr && !xdr(rqstp, (__be32*)rqstp->rq_arg.head[0].iov_base, 601 rqstp->rq_argp)) { 602 dprintk("nfsd: failed to decode arguments!\n"); 603 *statp = rpc_garbage_args; 604 return 1; 605 } 606 607 /* Check whether we have this call in the cache. */ 608 switch (nfsd_cache_lookup(rqstp)) { 609 case RC_INTR: 610 case RC_DROPIT: 611 return 0; 612 case RC_REPLY: 613 return 1; 614 case RC_DOIT:; 615 /* do it */ 616 } 617 618 /* need to grab the location to store the status, as 619 * nfsv4 does some encoding while processing 620 */ 621 nfserrp = rqstp->rq_res.head[0].iov_base 622 + rqstp->rq_res.head[0].iov_len; 623 rqstp->rq_res.head[0].iov_len += sizeof(__be32); 624 625 /* Now call the procedure handler, and encode NFS status. */ 626 nfserr = proc->pc_func(rqstp, rqstp->rq_argp, rqstp->rq_resp); 627 nfserr = map_new_errors(rqstp->rq_vers, nfserr); 628 if (nfserr == nfserr_dropit || rqstp->rq_dropme) { 629 dprintk("nfsd: Dropping request; may be revisited later\n"); 630 nfsd_cache_update(rqstp, RC_NOCACHE, NULL); 631 return 0; 632 } 633 634 if (rqstp->rq_proc != 0) 635 *nfserrp++ = nfserr; 636 637 /* Encode result. 638 * For NFSv2, additional info is never returned in case of an error. 639 */ 640 if (!(nfserr && rqstp->rq_vers == 2)) { 641 xdr = proc->pc_encode; 642 if (xdr && !xdr(rqstp, nfserrp, 643 rqstp->rq_resp)) { 644 /* Failed to encode result. Release cache entry */ 645 dprintk("nfsd: failed to encode result!\n"); 646 nfsd_cache_update(rqstp, RC_NOCACHE, NULL); 647 *statp = rpc_system_err; 648 return 1; 649 } 650 } 651 652 /* Store reply in cache. */ 653 nfsd_cache_update(rqstp, proc->pc_cachetype, statp + 1); 654 return 1; 655 } 656 657 int nfsd_pool_stats_open(struct inode *inode, struct file *file) 658 { 659 int ret; 660 mutex_lock(&nfsd_mutex); 661 if (nfsd_serv == NULL) { 662 mutex_unlock(&nfsd_mutex); 663 return -ENODEV; 664 } 665 /* bump up the psudo refcount while traversing */ 666 svc_get(nfsd_serv); 667 ret = svc_pool_stats_open(nfsd_serv, file); 668 mutex_unlock(&nfsd_mutex); 669 return ret; 670 } 671 672 int nfsd_pool_stats_release(struct inode *inode, struct file *file) 673 { 674 int ret = seq_release(inode, file); 675 struct net *net = &init_net; 676 677 mutex_lock(&nfsd_mutex); 678 /* this function really, really should have been called svc_put() */ 679 nfsd_destroy(net); 680 mutex_unlock(&nfsd_mutex); 681 return ret; 682 } 683