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