1 /* 2 * linux/fs/nfsd/nfssvc.c 3 * 4 * Central processing for nfsd. 5 * 6 * Authors: Olaf Kirch (okir@monad.swb.de) 7 * 8 * Copyright (C) 1995, 1996, 1997 Olaf Kirch <okir@monad.swb.de> 9 */ 10 11 #include <linux/module.h> 12 #include <linux/sched.h> 13 #include <linux/time.h> 14 #include <linux/errno.h> 15 #include <linux/nfs.h> 16 #include <linux/in.h> 17 #include <linux/uio.h> 18 #include <linux/unistd.h> 19 #include <linux/slab.h> 20 #include <linux/smp.h> 21 #include <linux/smp_lock.h> 22 #include <linux/freezer.h> 23 #include <linux/fs_struct.h> 24 25 #include <linux/sunrpc/types.h> 26 #include <linux/sunrpc/stats.h> 27 #include <linux/sunrpc/svc.h> 28 #include <linux/sunrpc/svcsock.h> 29 #include <linux/sunrpc/cache.h> 30 #include <linux/nfsd/nfsd.h> 31 #include <linux/nfsd/stats.h> 32 #include <linux/nfsd/cache.h> 33 #include <linux/nfsd/syscall.h> 34 #include <linux/lockd/bind.h> 35 #include <linux/nfsacl.h> 36 37 #define NFSDDBG_FACILITY NFSDDBG_SVC 38 39 /* these signals will be delivered to an nfsd thread 40 * when handling a request 41 */ 42 #define ALLOWED_SIGS (sigmask(SIGKILL)) 43 /* these signals will be delivered to an nfsd thread 44 * when not handling a request. i.e. when waiting 45 */ 46 #define SHUTDOWN_SIGS (sigmask(SIGKILL) | sigmask(SIGHUP) | sigmask(SIGINT) | sigmask(SIGQUIT)) 47 /* if the last thread dies with SIGHUP, then the exports table is 48 * left unchanged ( like 2.4-{0-9} ). Any other signal will clear 49 * the exports table (like 2.2). 50 */ 51 #define SIG_NOCLEAN SIGHUP 52 53 extern struct svc_program nfsd_program; 54 static void nfsd(struct svc_rqst *rqstp); 55 struct timeval nfssvc_boot; 56 struct svc_serv *nfsd_serv; 57 static atomic_t nfsd_busy; 58 static unsigned long nfsd_last_call; 59 static DEFINE_SPINLOCK(nfsd_call_lock); 60 61 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) 62 static struct svc_stat nfsd_acl_svcstats; 63 static struct svc_version * nfsd_acl_version[] = { 64 [2] = &nfsd_acl_version2, 65 [3] = &nfsd_acl_version3, 66 }; 67 68 #define NFSD_ACL_MINVERS 2 69 #define NFSD_ACL_NRVERS ARRAY_SIZE(nfsd_acl_version) 70 static struct svc_version *nfsd_acl_versions[NFSD_ACL_NRVERS]; 71 72 static struct svc_program nfsd_acl_program = { 73 .pg_prog = NFS_ACL_PROGRAM, 74 .pg_nvers = NFSD_ACL_NRVERS, 75 .pg_vers = nfsd_acl_versions, 76 .pg_name = "nfsacl", 77 .pg_class = "nfsd", 78 .pg_stats = &nfsd_acl_svcstats, 79 .pg_authenticate = &svc_set_client, 80 }; 81 82 static struct svc_stat nfsd_acl_svcstats = { 83 .program = &nfsd_acl_program, 84 }; 85 #endif /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */ 86 87 static struct svc_version * nfsd_version[] = { 88 [2] = &nfsd_version2, 89 #if defined(CONFIG_NFSD_V3) 90 [3] = &nfsd_version3, 91 #endif 92 #if defined(CONFIG_NFSD_V4) 93 [4] = &nfsd_version4, 94 #endif 95 }; 96 97 #define NFSD_MINVERS 2 98 #define NFSD_NRVERS ARRAY_SIZE(nfsd_version) 99 static struct svc_version *nfsd_versions[NFSD_NRVERS]; 100 101 struct svc_program nfsd_program = { 102 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) 103 .pg_next = &nfsd_acl_program, 104 #endif 105 .pg_prog = NFS_PROGRAM, /* program number */ 106 .pg_nvers = NFSD_NRVERS, /* nr of entries in nfsd_version */ 107 .pg_vers = nfsd_versions, /* version table */ 108 .pg_name = "nfsd", /* program name */ 109 .pg_class = "nfsd", /* authentication class */ 110 .pg_stats = &nfsd_svcstats, /* version table */ 111 .pg_authenticate = &svc_set_client, /* export authentication */ 112 113 }; 114 115 int nfsd_vers(int vers, enum vers_op change) 116 { 117 if (vers < NFSD_MINVERS || vers >= NFSD_NRVERS) 118 return -1; 119 switch(change) { 120 case NFSD_SET: 121 nfsd_versions[vers] = nfsd_version[vers]; 122 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) 123 if (vers < NFSD_ACL_NRVERS) 124 nfsd_acl_versions[vers] = nfsd_acl_version[vers]; 125 #endif 126 break; 127 case NFSD_CLEAR: 128 nfsd_versions[vers] = NULL; 129 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) 130 if (vers < NFSD_ACL_NRVERS) 131 nfsd_acl_versions[vers] = NULL; 132 #endif 133 break; 134 case NFSD_TEST: 135 return nfsd_versions[vers] != NULL; 136 case NFSD_AVAIL: 137 return nfsd_version[vers] != NULL; 138 } 139 return 0; 140 } 141 /* 142 * Maximum number of nfsd processes 143 */ 144 #define NFSD_MAXSERVS 8192 145 146 int nfsd_nrthreads(void) 147 { 148 if (nfsd_serv == NULL) 149 return 0; 150 else 151 return nfsd_serv->sv_nrthreads; 152 } 153 154 static int killsig; /* signal that was used to kill last nfsd */ 155 static void nfsd_last_thread(struct svc_serv *serv) 156 { 157 /* When last nfsd thread exits we need to do some clean-up */ 158 struct svc_xprt *xprt; 159 list_for_each_entry(xprt, &serv->sv_permsocks, xpt_list) 160 lockd_down(); 161 nfsd_serv = NULL; 162 nfsd_racache_shutdown(); 163 nfs4_state_shutdown(); 164 165 printk(KERN_WARNING "nfsd: last server has exited\n"); 166 if (killsig != SIG_NOCLEAN) { 167 printk(KERN_WARNING "nfsd: unexporting all filesystems\n"); 168 nfsd_export_flush(); 169 } 170 } 171 172 void nfsd_reset_versions(void) 173 { 174 int found_one = 0; 175 int i; 176 177 for (i = NFSD_MINVERS; i < NFSD_NRVERS; i++) { 178 if (nfsd_program.pg_vers[i]) 179 found_one = 1; 180 } 181 182 if (!found_one) { 183 for (i = NFSD_MINVERS; i < NFSD_NRVERS; i++) 184 nfsd_program.pg_vers[i] = nfsd_version[i]; 185 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) 186 for (i = NFSD_ACL_MINVERS; i < NFSD_ACL_NRVERS; i++) 187 nfsd_acl_program.pg_vers[i] = 188 nfsd_acl_version[i]; 189 #endif 190 } 191 } 192 193 int nfsd_create_serv(void) 194 { 195 int err = 0; 196 lock_kernel(); 197 if (nfsd_serv) { 198 svc_get(nfsd_serv); 199 unlock_kernel(); 200 return 0; 201 } 202 if (nfsd_max_blksize == 0) { 203 /* choose a suitable default */ 204 struct sysinfo i; 205 si_meminfo(&i); 206 /* Aim for 1/4096 of memory per thread 207 * This gives 1MB on 4Gig machines 208 * But only uses 32K on 128M machines. 209 * Bottom out at 8K on 32M and smaller. 210 * Of course, this is only a default. 211 */ 212 nfsd_max_blksize = NFSSVC_MAXBLKSIZE; 213 i.totalram <<= PAGE_SHIFT - 12; 214 while (nfsd_max_blksize > i.totalram && 215 nfsd_max_blksize >= 8*1024*2) 216 nfsd_max_blksize /= 2; 217 } 218 219 atomic_set(&nfsd_busy, 0); 220 nfsd_serv = svc_create_pooled(&nfsd_program, 221 nfsd_max_blksize, 222 nfsd_last_thread, 223 nfsd, SIG_NOCLEAN, THIS_MODULE); 224 if (nfsd_serv == NULL) 225 err = -ENOMEM; 226 unlock_kernel(); 227 do_gettimeofday(&nfssvc_boot); /* record boot time */ 228 return err; 229 } 230 231 static int nfsd_init_socks(int port) 232 { 233 int error; 234 if (!list_empty(&nfsd_serv->sv_permsocks)) 235 return 0; 236 237 error = lockd_up(IPPROTO_UDP); 238 if (error >= 0) { 239 error = svc_create_xprt(nfsd_serv, "udp", port, 240 SVC_SOCK_DEFAULTS); 241 if (error < 0) 242 lockd_down(); 243 } 244 if (error < 0) 245 return error; 246 247 error = lockd_up(IPPROTO_TCP); 248 if (error >= 0) { 249 error = svc_create_xprt(nfsd_serv, "tcp", port, 250 SVC_SOCK_DEFAULTS); 251 if (error < 0) 252 lockd_down(); 253 } 254 if (error < 0) 255 return error; 256 return 0; 257 } 258 259 int nfsd_nrpools(void) 260 { 261 if (nfsd_serv == NULL) 262 return 0; 263 else 264 return nfsd_serv->sv_nrpools; 265 } 266 267 int nfsd_get_nrthreads(int n, int *nthreads) 268 { 269 int i = 0; 270 271 if (nfsd_serv != NULL) { 272 for (i = 0; i < nfsd_serv->sv_nrpools && i < n; i++) 273 nthreads[i] = nfsd_serv->sv_pools[i].sp_nrthreads; 274 } 275 276 return 0; 277 } 278 279 int nfsd_set_nrthreads(int n, int *nthreads) 280 { 281 int i = 0; 282 int tot = 0; 283 int err = 0; 284 285 if (nfsd_serv == NULL || n <= 0) 286 return 0; 287 288 if (n > nfsd_serv->sv_nrpools) 289 n = nfsd_serv->sv_nrpools; 290 291 /* enforce a global maximum number of threads */ 292 tot = 0; 293 for (i = 0; i < n; i++) { 294 if (nthreads[i] > NFSD_MAXSERVS) 295 nthreads[i] = NFSD_MAXSERVS; 296 tot += nthreads[i]; 297 } 298 if (tot > NFSD_MAXSERVS) { 299 /* total too large: scale down requested numbers */ 300 for (i = 0; i < n && tot > 0; i++) { 301 int new = nthreads[i] * NFSD_MAXSERVS / tot; 302 tot -= (nthreads[i] - new); 303 nthreads[i] = new; 304 } 305 for (i = 0; i < n && tot > 0; i++) { 306 nthreads[i]--; 307 tot--; 308 } 309 } 310 311 /* 312 * There must always be a thread in pool 0; the admin 313 * can't shut down NFS completely using pool_threads. 314 */ 315 if (nthreads[0] == 0) 316 nthreads[0] = 1; 317 318 /* apply the new numbers */ 319 lock_kernel(); 320 svc_get(nfsd_serv); 321 for (i = 0; i < n; i++) { 322 err = svc_set_num_threads(nfsd_serv, &nfsd_serv->sv_pools[i], 323 nthreads[i]); 324 if (err) 325 break; 326 } 327 svc_destroy(nfsd_serv); 328 unlock_kernel(); 329 330 return err; 331 } 332 333 int 334 nfsd_svc(unsigned short port, int nrservs) 335 { 336 int error; 337 338 lock_kernel(); 339 dprintk("nfsd: creating service\n"); 340 error = -EINVAL; 341 if (nrservs <= 0) 342 nrservs = 0; 343 if (nrservs > NFSD_MAXSERVS) 344 nrservs = NFSD_MAXSERVS; 345 346 /* Readahead param cache - will no-op if it already exists */ 347 error = nfsd_racache_init(2*nrservs); 348 if (error<0) 349 goto out; 350 nfs4_state_start(); 351 352 nfsd_reset_versions(); 353 354 error = nfsd_create_serv(); 355 356 if (error) 357 goto out; 358 error = nfsd_init_socks(port); 359 if (error) 360 goto failure; 361 362 error = svc_set_num_threads(nfsd_serv, NULL, nrservs); 363 failure: 364 svc_destroy(nfsd_serv); /* Release server */ 365 out: 366 unlock_kernel(); 367 return error; 368 } 369 370 static inline void 371 update_thread_usage(int busy_threads) 372 { 373 unsigned long prev_call; 374 unsigned long diff; 375 int decile; 376 377 spin_lock(&nfsd_call_lock); 378 prev_call = nfsd_last_call; 379 nfsd_last_call = jiffies; 380 decile = busy_threads*10/nfsdstats.th_cnt; 381 if (decile>0 && decile <= 10) { 382 diff = nfsd_last_call - prev_call; 383 if ( (nfsdstats.th_usage[decile-1] += diff) >= NFSD_USAGE_WRAP) 384 nfsdstats.th_usage[decile-1] -= NFSD_USAGE_WRAP; 385 if (decile == 10) 386 nfsdstats.th_fullcnt++; 387 } 388 spin_unlock(&nfsd_call_lock); 389 } 390 391 /* 392 * This is the NFS server kernel thread 393 */ 394 static void 395 nfsd(struct svc_rqst *rqstp) 396 { 397 struct fs_struct *fsp; 398 int err; 399 sigset_t shutdown_mask, allowed_mask; 400 401 /* Lock module and set up kernel thread */ 402 lock_kernel(); 403 daemonize("nfsd"); 404 405 /* After daemonize() this kernel thread shares current->fs 406 * with the init process. We need to create files with a 407 * umask of 0 instead of init's umask. */ 408 fsp = copy_fs_struct(current->fs); 409 if (!fsp) { 410 printk("Unable to start nfsd thread: out of memory\n"); 411 goto out; 412 } 413 exit_fs(current); 414 current->fs = fsp; 415 current->fs->umask = 0; 416 417 siginitsetinv(&shutdown_mask, SHUTDOWN_SIGS); 418 siginitsetinv(&allowed_mask, ALLOWED_SIGS); 419 420 nfsdstats.th_cnt++; 421 422 rqstp->rq_task = current; 423 424 unlock_kernel(); 425 426 /* 427 * We want less throttling in balance_dirty_pages() so that nfs to 428 * localhost doesn't cause nfsd to lock up due to all the client's 429 * dirty pages. 430 */ 431 current->flags |= PF_LESS_THROTTLE; 432 set_freezable(); 433 434 /* 435 * The main request loop 436 */ 437 for (;;) { 438 /* Block all but the shutdown signals */ 439 sigprocmask(SIG_SETMASK, &shutdown_mask, NULL); 440 441 /* 442 * Find a socket with data available and call its 443 * recvfrom routine. 444 */ 445 while ((err = svc_recv(rqstp, 60*60*HZ)) == -EAGAIN) 446 ; 447 if (err < 0) 448 break; 449 update_thread_usage(atomic_read(&nfsd_busy)); 450 atomic_inc(&nfsd_busy); 451 452 /* Lock the export hash tables for reading. */ 453 exp_readlock(); 454 455 /* Process request with signals blocked. */ 456 sigprocmask(SIG_SETMASK, &allowed_mask, NULL); 457 458 svc_process(rqstp); 459 460 /* Unlock export hash tables */ 461 exp_readunlock(); 462 update_thread_usage(atomic_read(&nfsd_busy)); 463 atomic_dec(&nfsd_busy); 464 } 465 466 if (err != -EINTR) { 467 printk(KERN_WARNING "nfsd: terminating on error %d\n", -err); 468 } else { 469 unsigned int signo; 470 471 for (signo = 1; signo <= _NSIG; signo++) 472 if (sigismember(¤t->pending.signal, signo) && 473 !sigismember(¤t->blocked, signo)) 474 break; 475 killsig = signo; 476 } 477 /* Clear signals before calling svc_exit_thread() */ 478 flush_signals(current); 479 480 lock_kernel(); 481 482 nfsdstats.th_cnt --; 483 484 out: 485 /* Release the thread */ 486 svc_exit_thread(rqstp); 487 488 /* Release module */ 489 unlock_kernel(); 490 module_put_and_exit(0); 491 } 492 493 static __be32 map_new_errors(u32 vers, __be32 nfserr) 494 { 495 if (nfserr == nfserr_jukebox && vers == 2) 496 return nfserr_dropit; 497 if (nfserr == nfserr_wrongsec && vers < 4) 498 return nfserr_acces; 499 return nfserr; 500 } 501 502 int 503 nfsd_dispatch(struct svc_rqst *rqstp, __be32 *statp) 504 { 505 struct svc_procedure *proc; 506 kxdrproc_t xdr; 507 __be32 nfserr; 508 __be32 *nfserrp; 509 510 dprintk("nfsd_dispatch: vers %d proc %d\n", 511 rqstp->rq_vers, rqstp->rq_proc); 512 proc = rqstp->rq_procinfo; 513 514 /* Check whether we have this call in the cache. */ 515 switch (nfsd_cache_lookup(rqstp, proc->pc_cachetype)) { 516 case RC_INTR: 517 case RC_DROPIT: 518 return 0; 519 case RC_REPLY: 520 return 1; 521 case RC_DOIT:; 522 /* do it */ 523 } 524 525 /* Decode arguments */ 526 xdr = proc->pc_decode; 527 if (xdr && !xdr(rqstp, (__be32*)rqstp->rq_arg.head[0].iov_base, 528 rqstp->rq_argp)) { 529 dprintk("nfsd: failed to decode arguments!\n"); 530 nfsd_cache_update(rqstp, RC_NOCACHE, NULL); 531 *statp = rpc_garbage_args; 532 return 1; 533 } 534 535 /* need to grab the location to store the status, as 536 * nfsv4 does some encoding while processing 537 */ 538 nfserrp = rqstp->rq_res.head[0].iov_base 539 + rqstp->rq_res.head[0].iov_len; 540 rqstp->rq_res.head[0].iov_len += sizeof(__be32); 541 542 /* Now call the procedure handler, and encode NFS status. */ 543 nfserr = proc->pc_func(rqstp, rqstp->rq_argp, rqstp->rq_resp); 544 nfserr = map_new_errors(rqstp->rq_vers, nfserr); 545 if (nfserr == nfserr_dropit) { 546 dprintk("nfsd: Dropping request; may be revisited later\n"); 547 nfsd_cache_update(rqstp, RC_NOCACHE, NULL); 548 return 0; 549 } 550 551 if (rqstp->rq_proc != 0) 552 *nfserrp++ = nfserr; 553 554 /* Encode result. 555 * For NFSv2, additional info is never returned in case of an error. 556 */ 557 if (!(nfserr && rqstp->rq_vers == 2)) { 558 xdr = proc->pc_encode; 559 if (xdr && !xdr(rqstp, nfserrp, 560 rqstp->rq_resp)) { 561 /* Failed to encode result. Release cache entry */ 562 dprintk("nfsd: failed to encode result!\n"); 563 nfsd_cache_update(rqstp, RC_NOCACHE, NULL); 564 *statp = rpc_system_err; 565 return 1; 566 } 567 } 568 569 /* Store reply in cache. */ 570 nfsd_cache_update(rqstp, proc->pc_cachetype, statp + 1); 571 return 1; 572 } 573