| 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 | |
| 36 | /* |
| 37 | * nfsd_mutex protects nn->nfsd_serv -- both the pointer itself and the members |
| 38 | * of the svc_serv struct. In particular, ->sv_nrthreads but also to some |
| 39 | * extent ->sv_temp_socks and ->sv_permsocks. It also protects nfsdstats.th_cnt |
| 40 | * |
| 41 | * If (out side the lock) nn->nfsd_serv is non-NULL, then it must point to a |
| 42 | * properly initialised 'struct svc_serv' with ->sv_nrthreads > 0. That number |
| 43 | * of nfsd threads must exist and each must listed in ->sp_all_threads in each |
| 44 | * entry of ->sv_pools[]. |
| 45 | * |
| 46 | * Transitions of the thread count between zero and non-zero are of particular |
| 47 | * interest since the svc_serv needs to be created and initialized at that |
| 48 | * point, or freed. |
| 49 | * |
| 50 | * Finally, the nfsd_mutex also protects some of the global variables that are |
| 51 | * accessed when nfsd starts and that are settable via the write_* routines in |
| 52 | * nfsctl.c. In particular: |
| 53 | * |
| 54 | * user_recovery_dirname |
| 55 | * user_lease_time |
| 56 | * nfsd_versions |
| 57 | */ |
| 58 | DEFINE_MUTEX(nfsd_mutex); |
| 59 | |
| 60 | /* |
| 61 | * nfsd_drc_lock protects nfsd_drc_max_pages and nfsd_drc_pages_used. |
| 62 | * nfsd_drc_max_pages limits the total amount of memory available for |
| 63 | * version 4.1 DRC caches. |
| 64 | * nfsd_drc_pages_used tracks the current version 4.1 DRC memory usage. |
| 65 | */ |
| 66 | spinlock_t nfsd_drc_lock; |
| 67 | unsigned long nfsd_drc_max_mem; |
| 68 | unsigned long nfsd_drc_mem_used; |
| 69 | |
| 70 | #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) |
| 71 | static struct svc_stat nfsd_acl_svcstats; |
| 72 | static const struct svc_version *nfsd_acl_version[] = { |
| 73 | [2] = &nfsd_acl_version2, |
| 74 | [3] = &nfsd_acl_version3, |
| 75 | }; |
| 76 | |
| 77 | #define NFSD_ACL_MINVERS 2 |
| 78 | #define NFSD_ACL_NRVERS ARRAY_SIZE(nfsd_acl_version) |
| 79 | static const struct svc_version *nfsd_acl_versions[NFSD_ACL_NRVERS]; |
| 80 | |
| 81 | static struct svc_program nfsd_acl_program = { |
| 82 | .pg_prog = NFS_ACL_PROGRAM, |
| 83 | .pg_nvers = NFSD_ACL_NRVERS, |
| 84 | .pg_vers = nfsd_acl_versions, |
| 85 | .pg_name = "nfsacl", |
| 86 | .pg_class = "nfsd", |
| 87 | .pg_stats = &nfsd_acl_svcstats, |
| 88 | .pg_authenticate = &svc_set_client, |
| 89 | }; |
| 90 | |
| 91 | static struct svc_stat nfsd_acl_svcstats = { |
| 92 | .program = &nfsd_acl_program, |
| 93 | }; |
| 94 | #endif /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */ |
| 95 | |
| 96 | static const struct svc_version *nfsd_version[] = { |
| 97 | [2] = &nfsd_version2, |
| 98 | #if defined(CONFIG_NFSD_V3) |
| 99 | [3] = &nfsd_version3, |
| 100 | #endif |
| 101 | #if defined(CONFIG_NFSD_V4) |
| 102 | [4] = &nfsd_version4, |
| 103 | #endif |
| 104 | }; |
| 105 | |
| 106 | #define NFSD_MINVERS 2 |
| 107 | #define NFSD_NRVERS ARRAY_SIZE(nfsd_version) |
| 108 | static const struct svc_version *nfsd_versions[NFSD_NRVERS]; |
| 109 | |
| 110 | struct svc_program nfsd_program = { |
| 111 | #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) |
| 112 | .pg_next = &nfsd_acl_program, |
| 113 | #endif |
| 114 | .pg_prog = NFS_PROGRAM, /* program number */ |
| 115 | .pg_nvers = NFSD_NRVERS, /* nr of entries in nfsd_version */ |
| 116 | .pg_vers = nfsd_versions, /* version table */ |
| 117 | .pg_name = "nfsd", /* program name */ |
| 118 | .pg_class = "nfsd", /* authentication class */ |
| 119 | .pg_stats = &nfsd_svcstats, /* version table */ |
| 120 | .pg_authenticate = &svc_set_client, /* export authentication */ |
| 121 | |
| 122 | }; |
| 123 | |
| 124 | static bool nfsd_supported_minorversions[NFSD_SUPPORTED_MINOR_VERSION + 1] = { |
| 125 | [0] = 1, |
| 126 | [1] = 1, |
| 127 | [2] = 1, |
| 128 | }; |
| 129 | |
| 130 | int nfsd_vers(int vers, enum vers_op change) |
| 131 | { |
| 132 | if (vers < NFSD_MINVERS || vers >= NFSD_NRVERS) |
| 133 | return 0; |
| 134 | switch(change) { |
| 135 | case NFSD_SET: |
| 136 | nfsd_versions[vers] = nfsd_version[vers]; |
| 137 | #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) |
| 138 | if (vers < NFSD_ACL_NRVERS) |
| 139 | nfsd_acl_versions[vers] = nfsd_acl_version[vers]; |
| 140 | #endif |
| 141 | break; |
| 142 | case NFSD_CLEAR: |
| 143 | nfsd_versions[vers] = NULL; |
| 144 | #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) |
| 145 | if (vers < NFSD_ACL_NRVERS) |
| 146 | nfsd_acl_versions[vers] = NULL; |
| 147 | #endif |
| 148 | break; |
| 149 | case NFSD_TEST: |
| 150 | return nfsd_versions[vers] != NULL; |
| 151 | case NFSD_AVAIL: |
| 152 | return nfsd_version[vers] != NULL; |
| 153 | } |
| 154 | return 0; |
| 155 | } |
| 156 | |
| 157 | static void |
| 158 | nfsd_adjust_nfsd_versions4(void) |
| 159 | { |
| 160 | unsigned i; |
| 161 | |
| 162 | for (i = 0; i <= NFSD_SUPPORTED_MINOR_VERSION; i++) { |
| 163 | if (nfsd_supported_minorversions[i]) |
| 164 | return; |
| 165 | } |
| 166 | nfsd_vers(4, NFSD_CLEAR); |
| 167 | } |
| 168 | |
| 169 | int nfsd_minorversion(u32 minorversion, enum vers_op change) |
| 170 | { |
| 171 | if (minorversion > NFSD_SUPPORTED_MINOR_VERSION && |
| 172 | change != NFSD_AVAIL) |
| 173 | return -1; |
| 174 | switch(change) { |
| 175 | case NFSD_SET: |
| 176 | nfsd_supported_minorversions[minorversion] = true; |
| 177 | nfsd_vers(4, NFSD_SET); |
| 178 | break; |
| 179 | case NFSD_CLEAR: |
| 180 | nfsd_supported_minorversions[minorversion] = false; |
| 181 | nfsd_adjust_nfsd_versions4(); |
| 182 | break; |
| 183 | case NFSD_TEST: |
| 184 | return nfsd_supported_minorversions[minorversion]; |
| 185 | case NFSD_AVAIL: |
| 186 | return minorversion <= NFSD_SUPPORTED_MINOR_VERSION; |
| 187 | } |
| 188 | return 0; |
| 189 | } |
| 190 | |
| 191 | /* |
| 192 | * Maximum number of nfsd processes |
| 193 | */ |
| 194 | #define NFSD_MAXSERVS 8192 |
| 195 | |
| 196 | int nfsd_nrthreads(struct net *net) |
| 197 | { |
| 198 | int rv = 0; |
| 199 | struct nfsd_net *nn = net_generic(net, nfsd_net_id); |
| 200 | |
| 201 | mutex_lock(&nfsd_mutex); |
| 202 | if (nn->nfsd_serv) |
| 203 | rv = nn->nfsd_serv->sv_nrthreads; |
| 204 | mutex_unlock(&nfsd_mutex); |
| 205 | return rv; |
| 206 | } |
| 207 | |
| 208 | static int nfsd_init_socks(struct net *net) |
| 209 | { |
| 210 | int error; |
| 211 | struct nfsd_net *nn = net_generic(net, nfsd_net_id); |
| 212 | |
| 213 | if (!list_empty(&nn->nfsd_serv->sv_permsocks)) |
| 214 | return 0; |
| 215 | |
| 216 | error = svc_create_xprt(nn->nfsd_serv, "udp", net, PF_INET, NFS_PORT, |
| 217 | SVC_SOCK_DEFAULTS); |
| 218 | if (error < 0) |
| 219 | return error; |
| 220 | |
| 221 | error = svc_create_xprt(nn->nfsd_serv, "tcp", net, PF_INET, NFS_PORT, |
| 222 | SVC_SOCK_DEFAULTS); |
| 223 | if (error < 0) |
| 224 | return error; |
| 225 | |
| 226 | return 0; |
| 227 | } |
| 228 | |
| 229 | static int nfsd_users = 0; |
| 230 | |
| 231 | static int nfsd_startup_generic(int nrservs) |
| 232 | { |
| 233 | int ret; |
| 234 | |
| 235 | if (nfsd_users++) |
| 236 | return 0; |
| 237 | |
| 238 | /* |
| 239 | * Readahead param cache - will no-op if it already exists. |
| 240 | * (Note therefore results will be suboptimal if number of |
| 241 | * threads is modified after nfsd start.) |
| 242 | */ |
| 243 | ret = nfsd_racache_init(2*nrservs); |
| 244 | if (ret) |
| 245 | goto dec_users; |
| 246 | |
| 247 | ret = nfs4_state_start(); |
| 248 | if (ret) |
| 249 | goto out_racache; |
| 250 | return 0; |
| 251 | |
| 252 | out_racache: |
| 253 | nfsd_racache_shutdown(); |
| 254 | dec_users: |
| 255 | nfsd_users--; |
| 256 | return ret; |
| 257 | } |
| 258 | |
| 259 | static void nfsd_shutdown_generic(void) |
| 260 | { |
| 261 | if (--nfsd_users) |
| 262 | return; |
| 263 | |
| 264 | nfs4_state_shutdown(); |
| 265 | nfsd_racache_shutdown(); |
| 266 | } |
| 267 | |
| 268 | static bool nfsd_needs_lockd(void) |
| 269 | { |
| 270 | #if defined(CONFIG_NFSD_V3) |
| 271 | return (nfsd_versions[2] != NULL) || (nfsd_versions[3] != NULL); |
| 272 | #else |
| 273 | return (nfsd_versions[2] != NULL); |
| 274 | #endif |
| 275 | } |
| 276 | |
| 277 | static int nfsd_startup_net(int nrservs, struct net *net) |
| 278 | { |
| 279 | struct nfsd_net *nn = net_generic(net, nfsd_net_id); |
| 280 | int ret; |
| 281 | |
| 282 | if (nn->nfsd_net_up) |
| 283 | return 0; |
| 284 | |
| 285 | ret = nfsd_startup_generic(nrservs); |
| 286 | if (ret) |
| 287 | return ret; |
| 288 | ret = nfsd_init_socks(net); |
| 289 | if (ret) |
| 290 | goto out_socks; |
| 291 | |
| 292 | if (nfsd_needs_lockd() && !nn->lockd_up) { |
| 293 | ret = lockd_up(net); |
| 294 | if (ret) |
| 295 | goto out_socks; |
| 296 | nn->lockd_up = 1; |
| 297 | } |
| 298 | |
| 299 | ret = nfs4_state_start_net(net); |
| 300 | if (ret) |
| 301 | goto out_lockd; |
| 302 | |
| 303 | nn->nfsd_net_up = true; |
| 304 | return 0; |
| 305 | |
| 306 | out_lockd: |
| 307 | if (nn->lockd_up) { |
| 308 | lockd_down(net); |
| 309 | nn->lockd_up = 0; |
| 310 | } |
| 311 | out_socks: |
| 312 | nfsd_shutdown_generic(); |
| 313 | return ret; |
| 314 | } |
| 315 | |
| 316 | static void nfsd_shutdown_net(struct net *net) |
| 317 | { |
| 318 | struct nfsd_net *nn = net_generic(net, nfsd_net_id); |
| 319 | |
| 320 | nfs4_state_shutdown_net(net); |
| 321 | if (nn->lockd_up) { |
| 322 | lockd_down(net); |
| 323 | nn->lockd_up = 0; |
| 324 | } |
| 325 | nn->nfsd_net_up = false; |
| 326 | nfsd_shutdown_generic(); |
| 327 | } |
| 328 | |
| 329 | static int nfsd_inetaddr_event(struct notifier_block *this, unsigned long event, |
| 330 | void *ptr) |
| 331 | { |
| 332 | struct in_ifaddr *ifa = (struct in_ifaddr *)ptr; |
| 333 | struct net_device *dev = ifa->ifa_dev->dev; |
| 334 | struct net *net = dev_net(dev); |
| 335 | struct nfsd_net *nn = net_generic(net, nfsd_net_id); |
| 336 | struct sockaddr_in sin; |
| 337 | |
| 338 | if (event != NETDEV_DOWN) |
| 339 | goto out; |
| 340 | |
| 341 | if (nn->nfsd_serv) { |
| 342 | dprintk("nfsd_inetaddr_event: removed %pI4\n", &ifa->ifa_local); |
| 343 | sin.sin_family = AF_INET; |
| 344 | sin.sin_addr.s_addr = ifa->ifa_local; |
| 345 | svc_age_temp_xprts_now(nn->nfsd_serv, (struct sockaddr *)&sin); |
| 346 | } |
| 347 | |
| 348 | out: |
| 349 | return NOTIFY_DONE; |
| 350 | } |
| 351 | |
| 352 | static struct notifier_block nfsd_inetaddr_notifier = { |
| 353 | .notifier_call = nfsd_inetaddr_event, |
| 354 | }; |
| 355 | |
| 356 | #if IS_ENABLED(CONFIG_IPV6) |
| 357 | static int nfsd_inet6addr_event(struct notifier_block *this, |
| 358 | unsigned long event, void *ptr) |
| 359 | { |
| 360 | struct inet6_ifaddr *ifa = (struct inet6_ifaddr *)ptr; |
| 361 | struct net_device *dev = ifa->idev->dev; |
| 362 | struct net *net = dev_net(dev); |
| 363 | struct nfsd_net *nn = net_generic(net, nfsd_net_id); |
| 364 | struct sockaddr_in6 sin6; |
| 365 | |
| 366 | if (event != NETDEV_DOWN) |
| 367 | goto out; |
| 368 | |
| 369 | if (nn->nfsd_serv) { |
| 370 | dprintk("nfsd_inet6addr_event: removed %pI6\n", &ifa->addr); |
| 371 | sin6.sin6_family = AF_INET6; |
| 372 | sin6.sin6_addr = ifa->addr; |
| 373 | if (ipv6_addr_type(&sin6.sin6_addr) & IPV6_ADDR_LINKLOCAL) |
| 374 | sin6.sin6_scope_id = ifa->idev->dev->ifindex; |
| 375 | svc_age_temp_xprts_now(nn->nfsd_serv, (struct sockaddr *)&sin6); |
| 376 | } |
| 377 | |
| 378 | out: |
| 379 | return NOTIFY_DONE; |
| 380 | } |
| 381 | |
| 382 | static struct notifier_block nfsd_inet6addr_notifier = { |
| 383 | .notifier_call = nfsd_inet6addr_event, |
| 384 | }; |
| 385 | #endif |
| 386 | |
| 387 | /* Only used under nfsd_mutex, so this atomic may be overkill: */ |
| 388 | static atomic_t nfsd_notifier_refcount = ATOMIC_INIT(0); |
| 389 | |
| 390 | static void nfsd_last_thread(struct svc_serv *serv, struct net *net) |
| 391 | { |
| 392 | struct nfsd_net *nn = net_generic(net, nfsd_net_id); |
| 393 | |
| 394 | /* check if the notifier still has clients */ |
| 395 | if (atomic_dec_return(&nfsd_notifier_refcount) == 0) { |
| 396 | unregister_inetaddr_notifier(&nfsd_inetaddr_notifier); |
| 397 | #if IS_ENABLED(CONFIG_IPV6) |
| 398 | unregister_inet6addr_notifier(&nfsd_inet6addr_notifier); |
| 399 | #endif |
| 400 | } |
| 401 | |
| 402 | /* |
| 403 | * write_ports can create the server without actually starting |
| 404 | * any threads--if we get shut down before any threads are |
| 405 | * started, then nfsd_last_thread will be run before any of this |
| 406 | * other initialization has been done except the rpcb information. |
| 407 | */ |
| 408 | svc_rpcb_cleanup(serv, net); |
| 409 | if (!nn->nfsd_net_up) |
| 410 | return; |
| 411 | |
| 412 | nfsd_shutdown_net(net); |
| 413 | printk(KERN_WARNING "nfsd: last server has exited, flushing export " |
| 414 | "cache\n"); |
| 415 | nfsd_export_flush(net); |
| 416 | } |
| 417 | |
| 418 | void nfsd_reset_versions(void) |
| 419 | { |
| 420 | int i; |
| 421 | |
| 422 | for (i = 0; i < NFSD_NRVERS; i++) |
| 423 | if (nfsd_vers(i, NFSD_TEST)) |
| 424 | return; |
| 425 | |
| 426 | for (i = 0; i < NFSD_NRVERS; i++) |
| 427 | if (i != 4) |
| 428 | nfsd_vers(i, NFSD_SET); |
| 429 | else { |
| 430 | int minor = 0; |
| 431 | while (nfsd_minorversion(minor, NFSD_SET) >= 0) |
| 432 | minor++; |
| 433 | } |
| 434 | } |
| 435 | |
| 436 | /* |
| 437 | * Each session guarantees a negotiated per slot memory cache for replies |
| 438 | * which in turn consumes memory beyond the v2/v3/v4.0 server. A dedicated |
| 439 | * NFSv4.1 server might want to use more memory for a DRC than a machine |
| 440 | * with mutiple services. |
| 441 | * |
| 442 | * Impose a hard limit on the number of pages for the DRC which varies |
| 443 | * according to the machines free pages. This is of course only a default. |
| 444 | * |
| 445 | * For now this is a #defined shift which could be under admin control |
| 446 | * in the future. |
| 447 | */ |
| 448 | static void set_max_drc(void) |
| 449 | { |
| 450 | #define NFSD_DRC_SIZE_SHIFT 10 |
| 451 | nfsd_drc_max_mem = (nr_free_buffer_pages() |
| 452 | >> NFSD_DRC_SIZE_SHIFT) * PAGE_SIZE; |
| 453 | nfsd_drc_mem_used = 0; |
| 454 | spin_lock_init(&nfsd_drc_lock); |
| 455 | dprintk("%s nfsd_drc_max_mem %lu \n", __func__, nfsd_drc_max_mem); |
| 456 | } |
| 457 | |
| 458 | static int nfsd_get_default_max_blksize(void) |
| 459 | { |
| 460 | struct sysinfo i; |
| 461 | unsigned long long target; |
| 462 | unsigned long ret; |
| 463 | |
| 464 | si_meminfo(&i); |
| 465 | target = (i.totalram - i.totalhigh) << PAGE_SHIFT; |
| 466 | /* |
| 467 | * Aim for 1/4096 of memory per thread This gives 1MB on 4Gig |
| 468 | * machines, but only uses 32K on 128M machines. Bottom out at |
| 469 | * 8K on 32M and smaller. Of course, this is only a default. |
| 470 | */ |
| 471 | target >>= 12; |
| 472 | |
| 473 | ret = NFSSVC_MAXBLKSIZE; |
| 474 | while (ret > target && ret >= 8*1024*2) |
| 475 | ret /= 2; |
| 476 | return ret; |
| 477 | } |
| 478 | |
| 479 | static const struct svc_serv_ops nfsd_thread_sv_ops = { |
| 480 | .svo_shutdown = nfsd_last_thread, |
| 481 | .svo_function = nfsd, |
| 482 | .svo_enqueue_xprt = svc_xprt_do_enqueue, |
| 483 | .svo_setup = svc_set_num_threads, |
| 484 | .svo_module = THIS_MODULE, |
| 485 | }; |
| 486 | |
| 487 | int nfsd_create_serv(struct net *net) |
| 488 | { |
| 489 | int error; |
| 490 | struct nfsd_net *nn = net_generic(net, nfsd_net_id); |
| 491 | |
| 492 | WARN_ON(!mutex_is_locked(&nfsd_mutex)); |
| 493 | if (nn->nfsd_serv) { |
| 494 | svc_get(nn->nfsd_serv); |
| 495 | return 0; |
| 496 | } |
| 497 | if (nfsd_max_blksize == 0) |
| 498 | nfsd_max_blksize = nfsd_get_default_max_blksize(); |
| 499 | nfsd_reset_versions(); |
| 500 | nn->nfsd_serv = svc_create_pooled(&nfsd_program, nfsd_max_blksize, |
| 501 | &nfsd_thread_sv_ops); |
| 502 | if (nn->nfsd_serv == NULL) |
| 503 | return -ENOMEM; |
| 504 | |
| 505 | nn->nfsd_serv->sv_maxconn = nn->max_connections; |
| 506 | error = svc_bind(nn->nfsd_serv, net); |
| 507 | if (error < 0) { |
| 508 | svc_destroy(nn->nfsd_serv); |
| 509 | return error; |
| 510 | } |
| 511 | |
| 512 | set_max_drc(); |
| 513 | /* check if the notifier is already set */ |
| 514 | if (atomic_inc_return(&nfsd_notifier_refcount) == 1) { |
| 515 | register_inetaddr_notifier(&nfsd_inetaddr_notifier); |
| 516 | #if IS_ENABLED(CONFIG_IPV6) |
| 517 | register_inet6addr_notifier(&nfsd_inet6addr_notifier); |
| 518 | #endif |
| 519 | } |
| 520 | do_gettimeofday(&nn->nfssvc_boot); /* record boot time */ |
| 521 | return 0; |
| 522 | } |
| 523 | |
| 524 | int nfsd_nrpools(struct net *net) |
| 525 | { |
| 526 | struct nfsd_net *nn = net_generic(net, nfsd_net_id); |
| 527 | |
| 528 | if (nn->nfsd_serv == NULL) |
| 529 | return 0; |
| 530 | else |
| 531 | return nn->nfsd_serv->sv_nrpools; |
| 532 | } |
| 533 | |
| 534 | int nfsd_get_nrthreads(int n, int *nthreads, struct net *net) |
| 535 | { |
| 536 | int i = 0; |
| 537 | struct nfsd_net *nn = net_generic(net, nfsd_net_id); |
| 538 | |
| 539 | if (nn->nfsd_serv != NULL) { |
| 540 | for (i = 0; i < nn->nfsd_serv->sv_nrpools && i < n; i++) |
| 541 | nthreads[i] = nn->nfsd_serv->sv_pools[i].sp_nrthreads; |
| 542 | } |
| 543 | |
| 544 | return 0; |
| 545 | } |
| 546 | |
| 547 | void nfsd_destroy(struct net *net) |
| 548 | { |
| 549 | struct nfsd_net *nn = net_generic(net, nfsd_net_id); |
| 550 | int destroy = (nn->nfsd_serv->sv_nrthreads == 1); |
| 551 | |
| 552 | if (destroy) |
| 553 | svc_shutdown_net(nn->nfsd_serv, net); |
| 554 | svc_destroy(nn->nfsd_serv); |
| 555 | if (destroy) |
| 556 | nn->nfsd_serv = NULL; |
| 557 | } |
| 558 | |
| 559 | int nfsd_set_nrthreads(int n, int *nthreads, struct net *net) |
| 560 | { |
| 561 | int i = 0; |
| 562 | int tot = 0; |
| 563 | int err = 0; |
| 564 | struct nfsd_net *nn = net_generic(net, nfsd_net_id); |
| 565 | |
| 566 | WARN_ON(!mutex_is_locked(&nfsd_mutex)); |
| 567 | |
| 568 | if (nn->nfsd_serv == NULL || n <= 0) |
| 569 | return 0; |
| 570 | |
| 571 | if (n > nn->nfsd_serv->sv_nrpools) |
| 572 | n = nn->nfsd_serv->sv_nrpools; |
| 573 | |
| 574 | /* enforce a global maximum number of threads */ |
| 575 | tot = 0; |
| 576 | for (i = 0; i < n; i++) { |
| 577 | nthreads[i] = min(nthreads[i], NFSD_MAXSERVS); |
| 578 | tot += nthreads[i]; |
| 579 | } |
| 580 | if (tot > NFSD_MAXSERVS) { |
| 581 | /* total too large: scale down requested numbers */ |
| 582 | for (i = 0; i < n && tot > 0; i++) { |
| 583 | int new = nthreads[i] * NFSD_MAXSERVS / tot; |
| 584 | tot -= (nthreads[i] - new); |
| 585 | nthreads[i] = new; |
| 586 | } |
| 587 | for (i = 0; i < n && tot > 0; i++) { |
| 588 | nthreads[i]--; |
| 589 | tot--; |
| 590 | } |
| 591 | } |
| 592 | |
| 593 | /* |
| 594 | * There must always be a thread in pool 0; the admin |
| 595 | * can't shut down NFS completely using pool_threads. |
| 596 | */ |
| 597 | if (nthreads[0] == 0) |
| 598 | nthreads[0] = 1; |
| 599 | |
| 600 | /* apply the new numbers */ |
| 601 | svc_get(nn->nfsd_serv); |
| 602 | for (i = 0; i < n; i++) { |
| 603 | err = nn->nfsd_serv->sv_ops->svo_setup(nn->nfsd_serv, |
| 604 | &nn->nfsd_serv->sv_pools[i], nthreads[i]); |
| 605 | if (err) |
| 606 | break; |
| 607 | } |
| 608 | nfsd_destroy(net); |
| 609 | return err; |
| 610 | } |
| 611 | |
| 612 | /* |
| 613 | * Adjust the number of threads and return the new number of threads. |
| 614 | * This is also the function that starts the server if necessary, if |
| 615 | * this is the first time nrservs is nonzero. |
| 616 | */ |
| 617 | int |
| 618 | nfsd_svc(int nrservs, struct net *net) |
| 619 | { |
| 620 | int error; |
| 621 | bool nfsd_up_before; |
| 622 | struct nfsd_net *nn = net_generic(net, nfsd_net_id); |
| 623 | |
| 624 | mutex_lock(&nfsd_mutex); |
| 625 | dprintk("nfsd: creating service\n"); |
| 626 | |
| 627 | nrservs = max(nrservs, 0); |
| 628 | nrservs = min(nrservs, NFSD_MAXSERVS); |
| 629 | error = 0; |
| 630 | |
| 631 | if (nrservs == 0 && nn->nfsd_serv == NULL) |
| 632 | goto out; |
| 633 | |
| 634 | error = nfsd_create_serv(net); |
| 635 | if (error) |
| 636 | goto out; |
| 637 | |
| 638 | nfsd_up_before = nn->nfsd_net_up; |
| 639 | |
| 640 | error = nfsd_startup_net(nrservs, net); |
| 641 | if (error) |
| 642 | goto out_destroy; |
| 643 | error = nn->nfsd_serv->sv_ops->svo_setup(nn->nfsd_serv, |
| 644 | NULL, nrservs); |
| 645 | if (error) |
| 646 | goto out_shutdown; |
| 647 | /* We are holding a reference to nn->nfsd_serv which |
| 648 | * we don't want to count in the return value, |
| 649 | * so subtract 1 |
| 650 | */ |
| 651 | error = nn->nfsd_serv->sv_nrthreads - 1; |
| 652 | out_shutdown: |
| 653 | if (error < 0 && !nfsd_up_before) |
| 654 | nfsd_shutdown_net(net); |
| 655 | out_destroy: |
| 656 | nfsd_destroy(net); /* Release server */ |
| 657 | out: |
| 658 | mutex_unlock(&nfsd_mutex); |
| 659 | return error; |
| 660 | } |
| 661 | |
| 662 | |
| 663 | /* |
| 664 | * This is the NFS server kernel thread |
| 665 | */ |
| 666 | static int |
| 667 | nfsd(void *vrqstp) |
| 668 | { |
| 669 | struct svc_rqst *rqstp = (struct svc_rqst *) vrqstp; |
| 670 | struct svc_xprt *perm_sock = list_entry(rqstp->rq_server->sv_permsocks.next, typeof(struct svc_xprt), xpt_list); |
| 671 | struct net *net = perm_sock->xpt_net; |
| 672 | struct nfsd_net *nn = net_generic(net, nfsd_net_id); |
| 673 | int err; |
| 674 | |
| 675 | /* Lock module and set up kernel thread */ |
| 676 | mutex_lock(&nfsd_mutex); |
| 677 | |
| 678 | /* At this point, the thread shares current->fs |
| 679 | * with the init process. We need to create files with the |
| 680 | * umask as defined by the client instead of init's umask. */ |
| 681 | if (unshare_fs_struct() < 0) { |
| 682 | printk("Unable to start nfsd thread: out of memory\n"); |
| 683 | goto out; |
| 684 | } |
| 685 | |
| 686 | current->fs->umask = 0; |
| 687 | |
| 688 | /* |
| 689 | * thread is spawned with all signals set to SIG_IGN, re-enable |
| 690 | * the ones that will bring down the thread |
| 691 | */ |
| 692 | allow_signal(SIGKILL); |
| 693 | allow_signal(SIGHUP); |
| 694 | allow_signal(SIGINT); |
| 695 | allow_signal(SIGQUIT); |
| 696 | |
| 697 | nfsdstats.th_cnt++; |
| 698 | mutex_unlock(&nfsd_mutex); |
| 699 | |
| 700 | set_freezable(); |
| 701 | |
| 702 | /* |
| 703 | * The main request loop |
| 704 | */ |
| 705 | for (;;) { |
| 706 | /* Update sv_maxconn if it has changed */ |
| 707 | rqstp->rq_server->sv_maxconn = nn->max_connections; |
| 708 | |
| 709 | /* |
| 710 | * Find a socket with data available and call its |
| 711 | * recvfrom routine. |
| 712 | */ |
| 713 | while ((err = svc_recv(rqstp, 60*60*HZ)) == -EAGAIN) |
| 714 | ; |
| 715 | if (err == -EINTR) |
| 716 | break; |
| 717 | validate_process_creds(); |
| 718 | svc_process(rqstp); |
| 719 | validate_process_creds(); |
| 720 | } |
| 721 | |
| 722 | /* Clear signals before calling svc_exit_thread() */ |
| 723 | flush_signals(current); |
| 724 | |
| 725 | mutex_lock(&nfsd_mutex); |
| 726 | nfsdstats.th_cnt --; |
| 727 | |
| 728 | out: |
| 729 | rqstp->rq_server = NULL; |
| 730 | |
| 731 | /* Release the thread */ |
| 732 | svc_exit_thread(rqstp); |
| 733 | |
| 734 | nfsd_destroy(net); |
| 735 | |
| 736 | /* Release module */ |
| 737 | mutex_unlock(&nfsd_mutex); |
| 738 | module_put_and_exit(0); |
| 739 | return 0; |
| 740 | } |
| 741 | |
| 742 | static __be32 map_new_errors(u32 vers, __be32 nfserr) |
| 743 | { |
| 744 | if (nfserr == nfserr_jukebox && vers == 2) |
| 745 | return nfserr_dropit; |
| 746 | if (nfserr == nfserr_wrongsec && vers < 4) |
| 747 | return nfserr_acces; |
| 748 | return nfserr; |
| 749 | } |
| 750 | |
| 751 | /* |
| 752 | * A write procedure can have a large argument, and a read procedure can |
| 753 | * have a large reply, but no NFSv2 or NFSv3 procedure has argument and |
| 754 | * reply that can both be larger than a page. The xdr code has taken |
| 755 | * advantage of this assumption to be a sloppy about bounds checking in |
| 756 | * some cases. Pending a rewrite of the NFSv2/v3 xdr code to fix that |
| 757 | * problem, we enforce these assumptions here: |
| 758 | */ |
| 759 | static bool nfs_request_too_big(struct svc_rqst *rqstp, |
| 760 | const struct svc_procedure *proc) |
| 761 | { |
| 762 | /* |
| 763 | * The ACL code has more careful bounds-checking and is not |
| 764 | * susceptible to this problem: |
| 765 | */ |
| 766 | if (rqstp->rq_prog != NFS_PROGRAM) |
| 767 | return false; |
| 768 | /* |
| 769 | * Ditto NFSv4 (which can in theory have argument and reply both |
| 770 | * more than a page): |
| 771 | */ |
| 772 | if (rqstp->rq_vers >= 4) |
| 773 | return false; |
| 774 | /* The reply will be small, we're OK: */ |
| 775 | if (proc->pc_xdrressize > 0 && |
| 776 | proc->pc_xdrressize < XDR_QUADLEN(PAGE_SIZE)) |
| 777 | return false; |
| 778 | |
| 779 | return rqstp->rq_arg.len > PAGE_SIZE; |
| 780 | } |
| 781 | |
| 782 | int |
| 783 | nfsd_dispatch(struct svc_rqst *rqstp, __be32 *statp) |
| 784 | { |
| 785 | const struct svc_procedure *proc; |
| 786 | __be32 nfserr; |
| 787 | __be32 *nfserrp; |
| 788 | |
| 789 | dprintk("nfsd_dispatch: vers %d proc %d\n", |
| 790 | rqstp->rq_vers, rqstp->rq_proc); |
| 791 | proc = rqstp->rq_procinfo; |
| 792 | |
| 793 | if (nfs_request_too_big(rqstp, proc)) { |
| 794 | dprintk("nfsd: NFSv%d argument too large\n", rqstp->rq_vers); |
| 795 | *statp = rpc_garbage_args; |
| 796 | return 1; |
| 797 | } |
| 798 | /* |
| 799 | * Give the xdr decoder a chance to change this if it wants |
| 800 | * (necessary in the NFSv4.0 compound case) |
| 801 | */ |
| 802 | rqstp->rq_cachetype = proc->pc_cachetype; |
| 803 | /* Decode arguments */ |
| 804 | if (proc->pc_decode && |
| 805 | !proc->pc_decode(rqstp, (__be32*)rqstp->rq_arg.head[0].iov_base)) { |
| 806 | dprintk("nfsd: failed to decode arguments!\n"); |
| 807 | *statp = rpc_garbage_args; |
| 808 | return 1; |
| 809 | } |
| 810 | |
| 811 | /* Check whether we have this call in the cache. */ |
| 812 | switch (nfsd_cache_lookup(rqstp)) { |
| 813 | case RC_DROPIT: |
| 814 | return 0; |
| 815 | case RC_REPLY: |
| 816 | return 1; |
| 817 | case RC_DOIT:; |
| 818 | /* do it */ |
| 819 | } |
| 820 | |
| 821 | /* need to grab the location to store the status, as |
| 822 | * nfsv4 does some encoding while processing |
| 823 | */ |
| 824 | nfserrp = rqstp->rq_res.head[0].iov_base |
| 825 | + rqstp->rq_res.head[0].iov_len; |
| 826 | rqstp->rq_res.head[0].iov_len += sizeof(__be32); |
| 827 | |
| 828 | /* Now call the procedure handler, and encode NFS status. */ |
| 829 | nfserr = proc->pc_func(rqstp); |
| 830 | nfserr = map_new_errors(rqstp->rq_vers, nfserr); |
| 831 | if (nfserr == nfserr_dropit || test_bit(RQ_DROPME, &rqstp->rq_flags)) { |
| 832 | dprintk("nfsd: Dropping request; may be revisited later\n"); |
| 833 | nfsd_cache_update(rqstp, RC_NOCACHE, NULL); |
| 834 | return 0; |
| 835 | } |
| 836 | |
| 837 | if (rqstp->rq_proc != 0) |
| 838 | *nfserrp++ = nfserr; |
| 839 | |
| 840 | /* Encode result. |
| 841 | * For NFSv2, additional info is never returned in case of an error. |
| 842 | */ |
| 843 | if (!(nfserr && rqstp->rq_vers == 2)) { |
| 844 | if (proc->pc_encode && !proc->pc_encode(rqstp, nfserrp)) { |
| 845 | /* Failed to encode result. Release cache entry */ |
| 846 | dprintk("nfsd: failed to encode result!\n"); |
| 847 | nfsd_cache_update(rqstp, RC_NOCACHE, NULL); |
| 848 | *statp = rpc_system_err; |
| 849 | return 1; |
| 850 | } |
| 851 | } |
| 852 | |
| 853 | /* Store reply in cache. */ |
| 854 | nfsd_cache_update(rqstp, rqstp->rq_cachetype, statp + 1); |
| 855 | return 1; |
| 856 | } |
| 857 | |
| 858 | int nfsd_pool_stats_open(struct inode *inode, struct file *file) |
| 859 | { |
| 860 | int ret; |
| 861 | struct nfsd_net *nn = net_generic(inode->i_sb->s_fs_info, nfsd_net_id); |
| 862 | |
| 863 | mutex_lock(&nfsd_mutex); |
| 864 | if (nn->nfsd_serv == NULL) { |
| 865 | mutex_unlock(&nfsd_mutex); |
| 866 | return -ENODEV; |
| 867 | } |
| 868 | /* bump up the psudo refcount while traversing */ |
| 869 | svc_get(nn->nfsd_serv); |
| 870 | ret = svc_pool_stats_open(nn->nfsd_serv, file); |
| 871 | mutex_unlock(&nfsd_mutex); |
| 872 | return ret; |
| 873 | } |
| 874 | |
| 875 | int nfsd_pool_stats_release(struct inode *inode, struct file *file) |
| 876 | { |
| 877 | int ret = seq_release(inode, file); |
| 878 | struct net *net = inode->i_sb->s_fs_info; |
| 879 | |
| 880 | mutex_lock(&nfsd_mutex); |
| 881 | /* this function really, really should have been called svc_put() */ |
| 882 | nfsd_destroy(net); |
| 883 | mutex_unlock(&nfsd_mutex); |
| 884 | return ret; |
| 885 | } |