1 /* 2 * linux/net/sunrpc/svc.c 3 * 4 * High-level RPC service routines 5 * 6 * Copyright (C) 1995, 1996 Olaf Kirch <okir@monad.swb.de> 7 * 8 * Multiple threads pools and NUMAisation 9 * Copyright (c) 2006 Silicon Graphics, Inc. 10 * by Greg Banks <gnb@melbourne.sgi.com> 11 */ 12 13 #include <linux/linkage.h> 14 #include <linux/sched.h> 15 #include <linux/errno.h> 16 #include <linux/net.h> 17 #include <linux/in.h> 18 #include <linux/mm.h> 19 #include <linux/interrupt.h> 20 #include <linux/module.h> 21 #include <linux/kthread.h> 22 #include <linux/slab.h> 23 24 #include <linux/sunrpc/types.h> 25 #include <linux/sunrpc/xdr.h> 26 #include <linux/sunrpc/stats.h> 27 #include <linux/sunrpc/svcsock.h> 28 #include <linux/sunrpc/clnt.h> 29 #include <linux/sunrpc/bc_xprt.h> 30 31 #define RPCDBG_FACILITY RPCDBG_SVCDSP 32 33 static void svc_unregister(const struct svc_serv *serv, struct net *net); 34 35 #define svc_serv_is_pooled(serv) ((serv)->sv_function) 36 37 /* 38 * Mode for mapping cpus to pools. 39 */ 40 enum { 41 SVC_POOL_AUTO = -1, /* choose one of the others */ 42 SVC_POOL_GLOBAL, /* no mapping, just a single global pool 43 * (legacy & UP mode) */ 44 SVC_POOL_PERCPU, /* one pool per cpu */ 45 SVC_POOL_PERNODE /* one pool per numa node */ 46 }; 47 #define SVC_POOL_DEFAULT SVC_POOL_GLOBAL 48 49 /* 50 * Structure for mapping cpus to pools and vice versa. 51 * Setup once during sunrpc initialisation. 52 */ 53 static struct svc_pool_map { 54 int count; /* How many svc_servs use us */ 55 int mode; /* Note: int not enum to avoid 56 * warnings about "enumeration value 57 * not handled in switch" */ 58 unsigned int npools; 59 unsigned int *pool_to; /* maps pool id to cpu or node */ 60 unsigned int *to_pool; /* maps cpu or node to pool id */ 61 } svc_pool_map = { 62 .count = 0, 63 .mode = SVC_POOL_DEFAULT 64 }; 65 static DEFINE_MUTEX(svc_pool_map_mutex);/* protects svc_pool_map.count only */ 66 67 static int 68 param_set_pool_mode(const char *val, struct kernel_param *kp) 69 { 70 int *ip = (int *)kp->arg; 71 struct svc_pool_map *m = &svc_pool_map; 72 int err; 73 74 mutex_lock(&svc_pool_map_mutex); 75 76 err = -EBUSY; 77 if (m->count) 78 goto out; 79 80 err = 0; 81 if (!strncmp(val, "auto", 4)) 82 *ip = SVC_POOL_AUTO; 83 else if (!strncmp(val, "global", 6)) 84 *ip = SVC_POOL_GLOBAL; 85 else if (!strncmp(val, "percpu", 6)) 86 *ip = SVC_POOL_PERCPU; 87 else if (!strncmp(val, "pernode", 7)) 88 *ip = SVC_POOL_PERNODE; 89 else 90 err = -EINVAL; 91 92 out: 93 mutex_unlock(&svc_pool_map_mutex); 94 return err; 95 } 96 97 static int 98 param_get_pool_mode(char *buf, struct kernel_param *kp) 99 { 100 int *ip = (int *)kp->arg; 101 102 switch (*ip) 103 { 104 case SVC_POOL_AUTO: 105 return strlcpy(buf, "auto", 20); 106 case SVC_POOL_GLOBAL: 107 return strlcpy(buf, "global", 20); 108 case SVC_POOL_PERCPU: 109 return strlcpy(buf, "percpu", 20); 110 case SVC_POOL_PERNODE: 111 return strlcpy(buf, "pernode", 20); 112 default: 113 return sprintf(buf, "%d", *ip); 114 } 115 } 116 117 module_param_call(pool_mode, param_set_pool_mode, param_get_pool_mode, 118 &svc_pool_map.mode, 0644); 119 120 /* 121 * Detect best pool mapping mode heuristically, 122 * according to the machine's topology. 123 */ 124 static int 125 svc_pool_map_choose_mode(void) 126 { 127 unsigned int node; 128 129 if (nr_online_nodes > 1) { 130 /* 131 * Actually have multiple NUMA nodes, 132 * so split pools on NUMA node boundaries 133 */ 134 return SVC_POOL_PERNODE; 135 } 136 137 node = first_online_node; 138 if (nr_cpus_node(node) > 2) { 139 /* 140 * Non-trivial SMP, or CONFIG_NUMA on 141 * non-NUMA hardware, e.g. with a generic 142 * x86_64 kernel on Xeons. In this case we 143 * want to divide the pools on cpu boundaries. 144 */ 145 return SVC_POOL_PERCPU; 146 } 147 148 /* default: one global pool */ 149 return SVC_POOL_GLOBAL; 150 } 151 152 /* 153 * Allocate the to_pool[] and pool_to[] arrays. 154 * Returns 0 on success or an errno. 155 */ 156 static int 157 svc_pool_map_alloc_arrays(struct svc_pool_map *m, unsigned int maxpools) 158 { 159 m->to_pool = kcalloc(maxpools, sizeof(unsigned int), GFP_KERNEL); 160 if (!m->to_pool) 161 goto fail; 162 m->pool_to = kcalloc(maxpools, sizeof(unsigned int), GFP_KERNEL); 163 if (!m->pool_to) 164 goto fail_free; 165 166 return 0; 167 168 fail_free: 169 kfree(m->to_pool); 170 m->to_pool = NULL; 171 fail: 172 return -ENOMEM; 173 } 174 175 /* 176 * Initialise the pool map for SVC_POOL_PERCPU mode. 177 * Returns number of pools or <0 on error. 178 */ 179 static int 180 svc_pool_map_init_percpu(struct svc_pool_map *m) 181 { 182 unsigned int maxpools = nr_cpu_ids; 183 unsigned int pidx = 0; 184 unsigned int cpu; 185 int err; 186 187 err = svc_pool_map_alloc_arrays(m, maxpools); 188 if (err) 189 return err; 190 191 for_each_online_cpu(cpu) { 192 BUG_ON(pidx > maxpools); 193 m->to_pool[cpu] = pidx; 194 m->pool_to[pidx] = cpu; 195 pidx++; 196 } 197 /* cpus brought online later all get mapped to pool0, sorry */ 198 199 return pidx; 200 }; 201 202 203 /* 204 * Initialise the pool map for SVC_POOL_PERNODE mode. 205 * Returns number of pools or <0 on error. 206 */ 207 static int 208 svc_pool_map_init_pernode(struct svc_pool_map *m) 209 { 210 unsigned int maxpools = nr_node_ids; 211 unsigned int pidx = 0; 212 unsigned int node; 213 int err; 214 215 err = svc_pool_map_alloc_arrays(m, maxpools); 216 if (err) 217 return err; 218 219 for_each_node_with_cpus(node) { 220 /* some architectures (e.g. SN2) have cpuless nodes */ 221 BUG_ON(pidx > maxpools); 222 m->to_pool[node] = pidx; 223 m->pool_to[pidx] = node; 224 pidx++; 225 } 226 /* nodes brought online later all get mapped to pool0, sorry */ 227 228 return pidx; 229 } 230 231 232 /* 233 * Add a reference to the global map of cpus to pools (and 234 * vice versa). Initialise the map if we're the first user. 235 * Returns the number of pools. 236 */ 237 static unsigned int 238 svc_pool_map_get(void) 239 { 240 struct svc_pool_map *m = &svc_pool_map; 241 int npools = -1; 242 243 mutex_lock(&svc_pool_map_mutex); 244 245 if (m->count++) { 246 mutex_unlock(&svc_pool_map_mutex); 247 return m->npools; 248 } 249 250 if (m->mode == SVC_POOL_AUTO) 251 m->mode = svc_pool_map_choose_mode(); 252 253 switch (m->mode) { 254 case SVC_POOL_PERCPU: 255 npools = svc_pool_map_init_percpu(m); 256 break; 257 case SVC_POOL_PERNODE: 258 npools = svc_pool_map_init_pernode(m); 259 break; 260 } 261 262 if (npools < 0) { 263 /* default, or memory allocation failure */ 264 npools = 1; 265 m->mode = SVC_POOL_GLOBAL; 266 } 267 m->npools = npools; 268 269 mutex_unlock(&svc_pool_map_mutex); 270 return m->npools; 271 } 272 273 274 /* 275 * Drop a reference to the global map of cpus to pools. 276 * When the last reference is dropped, the map data is 277 * freed; this allows the sysadmin to change the pool 278 * mode using the pool_mode module option without 279 * rebooting or re-loading sunrpc.ko. 280 */ 281 static void 282 svc_pool_map_put(void) 283 { 284 struct svc_pool_map *m = &svc_pool_map; 285 286 mutex_lock(&svc_pool_map_mutex); 287 288 if (!--m->count) { 289 kfree(m->to_pool); 290 m->to_pool = NULL; 291 kfree(m->pool_to); 292 m->pool_to = NULL; 293 m->npools = 0; 294 } 295 296 mutex_unlock(&svc_pool_map_mutex); 297 } 298 299 300 static int svc_pool_map_get_node(unsigned int pidx) 301 { 302 const struct svc_pool_map *m = &svc_pool_map; 303 304 if (m->count) { 305 if (m->mode == SVC_POOL_PERCPU) 306 return cpu_to_node(m->pool_to[pidx]); 307 if (m->mode == SVC_POOL_PERNODE) 308 return m->pool_to[pidx]; 309 } 310 return NUMA_NO_NODE; 311 } 312 /* 313 * Set the given thread's cpus_allowed mask so that it 314 * will only run on cpus in the given pool. 315 */ 316 static inline void 317 svc_pool_map_set_cpumask(struct task_struct *task, unsigned int pidx) 318 { 319 struct svc_pool_map *m = &svc_pool_map; 320 unsigned int node = m->pool_to[pidx]; 321 322 /* 323 * The caller checks for sv_nrpools > 1, which 324 * implies that we've been initialized. 325 */ 326 WARN_ON_ONCE(m->count == 0); 327 if (m->count == 0) 328 return; 329 330 switch (m->mode) { 331 case SVC_POOL_PERCPU: 332 { 333 set_cpus_allowed_ptr(task, cpumask_of(node)); 334 break; 335 } 336 case SVC_POOL_PERNODE: 337 { 338 set_cpus_allowed_ptr(task, cpumask_of_node(node)); 339 break; 340 } 341 } 342 } 343 344 /* 345 * Use the mapping mode to choose a pool for a given CPU. 346 * Used when enqueueing an incoming RPC. Always returns 347 * a non-NULL pool pointer. 348 */ 349 struct svc_pool * 350 svc_pool_for_cpu(struct svc_serv *serv, int cpu) 351 { 352 struct svc_pool_map *m = &svc_pool_map; 353 unsigned int pidx = 0; 354 355 /* 356 * An uninitialised map happens in a pure client when 357 * lockd is brought up, so silently treat it the 358 * same as SVC_POOL_GLOBAL. 359 */ 360 if (svc_serv_is_pooled(serv)) { 361 switch (m->mode) { 362 case SVC_POOL_PERCPU: 363 pidx = m->to_pool[cpu]; 364 break; 365 case SVC_POOL_PERNODE: 366 pidx = m->to_pool[cpu_to_node(cpu)]; 367 break; 368 } 369 } 370 return &serv->sv_pools[pidx % serv->sv_nrpools]; 371 } 372 373 int svc_rpcb_setup(struct svc_serv *serv, struct net *net) 374 { 375 int err; 376 377 err = rpcb_create_local(net); 378 if (err) 379 return err; 380 381 /* Remove any stale portmap registrations */ 382 svc_unregister(serv, net); 383 return 0; 384 } 385 EXPORT_SYMBOL_GPL(svc_rpcb_setup); 386 387 void svc_rpcb_cleanup(struct svc_serv *serv, struct net *net) 388 { 389 svc_unregister(serv, net); 390 rpcb_put_local(net); 391 } 392 EXPORT_SYMBOL_GPL(svc_rpcb_cleanup); 393 394 static int svc_uses_rpcbind(struct svc_serv *serv) 395 { 396 struct svc_program *progp; 397 unsigned int i; 398 399 for (progp = serv->sv_program; progp; progp = progp->pg_next) { 400 for (i = 0; i < progp->pg_nvers; i++) { 401 if (progp->pg_vers[i] == NULL) 402 continue; 403 if (progp->pg_vers[i]->vs_hidden == 0) 404 return 1; 405 } 406 } 407 408 return 0; 409 } 410 411 int svc_bind(struct svc_serv *serv, struct net *net) 412 { 413 if (!svc_uses_rpcbind(serv)) 414 return 0; 415 return svc_rpcb_setup(serv, net); 416 } 417 EXPORT_SYMBOL_GPL(svc_bind); 418 419 /* 420 * Create an RPC service 421 */ 422 static struct svc_serv * 423 __svc_create(struct svc_program *prog, unsigned int bufsize, int npools, 424 void (*shutdown)(struct svc_serv *serv, struct net *net)) 425 { 426 struct svc_serv *serv; 427 unsigned int vers; 428 unsigned int xdrsize; 429 unsigned int i; 430 431 if (!(serv = kzalloc(sizeof(*serv), GFP_KERNEL))) 432 return NULL; 433 serv->sv_name = prog->pg_name; 434 serv->sv_program = prog; 435 serv->sv_nrthreads = 1; 436 serv->sv_stats = prog->pg_stats; 437 if (bufsize > RPCSVC_MAXPAYLOAD) 438 bufsize = RPCSVC_MAXPAYLOAD; 439 serv->sv_max_payload = bufsize? bufsize : 4096; 440 serv->sv_max_mesg = roundup(serv->sv_max_payload + PAGE_SIZE, PAGE_SIZE); 441 serv->sv_shutdown = shutdown; 442 xdrsize = 0; 443 while (prog) { 444 prog->pg_lovers = prog->pg_nvers-1; 445 for (vers=0; vers<prog->pg_nvers ; vers++) 446 if (prog->pg_vers[vers]) { 447 prog->pg_hivers = vers; 448 if (prog->pg_lovers > vers) 449 prog->pg_lovers = vers; 450 if (prog->pg_vers[vers]->vs_xdrsize > xdrsize) 451 xdrsize = prog->pg_vers[vers]->vs_xdrsize; 452 } 453 prog = prog->pg_next; 454 } 455 serv->sv_xdrsize = xdrsize; 456 INIT_LIST_HEAD(&serv->sv_tempsocks); 457 INIT_LIST_HEAD(&serv->sv_permsocks); 458 init_timer(&serv->sv_temptimer); 459 spin_lock_init(&serv->sv_lock); 460 461 serv->sv_nrpools = npools; 462 serv->sv_pools = 463 kcalloc(serv->sv_nrpools, sizeof(struct svc_pool), 464 GFP_KERNEL); 465 if (!serv->sv_pools) { 466 kfree(serv); 467 return NULL; 468 } 469 470 for (i = 0; i < serv->sv_nrpools; i++) { 471 struct svc_pool *pool = &serv->sv_pools[i]; 472 473 dprintk("svc: initialising pool %u for %s\n", 474 i, serv->sv_name); 475 476 pool->sp_id = i; 477 INIT_LIST_HEAD(&pool->sp_threads); 478 INIT_LIST_HEAD(&pool->sp_sockets); 479 INIT_LIST_HEAD(&pool->sp_all_threads); 480 spin_lock_init(&pool->sp_lock); 481 } 482 483 if (svc_uses_rpcbind(serv) && (!serv->sv_shutdown)) 484 serv->sv_shutdown = svc_rpcb_cleanup; 485 486 return serv; 487 } 488 489 struct svc_serv * 490 svc_create(struct svc_program *prog, unsigned int bufsize, 491 void (*shutdown)(struct svc_serv *serv, struct net *net)) 492 { 493 return __svc_create(prog, bufsize, /*npools*/1, shutdown); 494 } 495 EXPORT_SYMBOL_GPL(svc_create); 496 497 struct svc_serv * 498 svc_create_pooled(struct svc_program *prog, unsigned int bufsize, 499 void (*shutdown)(struct svc_serv *serv, struct net *net), 500 svc_thread_fn func, struct module *mod) 501 { 502 struct svc_serv *serv; 503 unsigned int npools = svc_pool_map_get(); 504 505 serv = __svc_create(prog, bufsize, npools, shutdown); 506 507 if (serv != NULL) { 508 serv->sv_function = func; 509 serv->sv_module = mod; 510 } 511 512 return serv; 513 } 514 EXPORT_SYMBOL_GPL(svc_create_pooled); 515 516 void svc_shutdown_net(struct svc_serv *serv, struct net *net) 517 { 518 svc_close_net(serv, net); 519 520 if (serv->sv_shutdown) 521 serv->sv_shutdown(serv, net); 522 } 523 EXPORT_SYMBOL_GPL(svc_shutdown_net); 524 525 /* 526 * Destroy an RPC service. Should be called with appropriate locking to 527 * protect the sv_nrthreads, sv_permsocks and sv_tempsocks. 528 */ 529 void 530 svc_destroy(struct svc_serv *serv) 531 { 532 dprintk("svc: svc_destroy(%s, %d)\n", 533 serv->sv_program->pg_name, 534 serv->sv_nrthreads); 535 536 if (serv->sv_nrthreads) { 537 if (--(serv->sv_nrthreads) != 0) { 538 svc_sock_update_bufs(serv); 539 return; 540 } 541 } else 542 printk("svc_destroy: no threads for serv=%p!\n", serv); 543 544 del_timer_sync(&serv->sv_temptimer); 545 546 /* 547 * The last user is gone and thus all sockets have to be destroyed to 548 * the point. Check this. 549 */ 550 BUG_ON(!list_empty(&serv->sv_permsocks)); 551 BUG_ON(!list_empty(&serv->sv_tempsocks)); 552 553 cache_clean_deferred(serv); 554 555 if (svc_serv_is_pooled(serv)) 556 svc_pool_map_put(); 557 558 kfree(serv->sv_pools); 559 kfree(serv); 560 } 561 EXPORT_SYMBOL_GPL(svc_destroy); 562 563 /* 564 * Allocate an RPC server's buffer space. 565 * We allocate pages and place them in rq_argpages. 566 */ 567 static int 568 svc_init_buffer(struct svc_rqst *rqstp, unsigned int size, int node) 569 { 570 unsigned int pages, arghi; 571 572 /* bc_xprt uses fore channel allocated buffers */ 573 if (svc_is_backchannel(rqstp)) 574 return 1; 575 576 pages = size / PAGE_SIZE + 1; /* extra page as we hold both request and reply. 577 * We assume one is at most one page 578 */ 579 arghi = 0; 580 WARN_ON_ONCE(pages > RPCSVC_MAXPAGES); 581 if (pages > RPCSVC_MAXPAGES) 582 pages = RPCSVC_MAXPAGES; 583 while (pages) { 584 struct page *p = alloc_pages_node(node, GFP_KERNEL, 0); 585 if (!p) 586 break; 587 rqstp->rq_pages[arghi++] = p; 588 pages--; 589 } 590 return pages == 0; 591 } 592 593 /* 594 * Release an RPC server buffer 595 */ 596 static void 597 svc_release_buffer(struct svc_rqst *rqstp) 598 { 599 unsigned int i; 600 601 for (i = 0; i < ARRAY_SIZE(rqstp->rq_pages); i++) 602 if (rqstp->rq_pages[i]) 603 put_page(rqstp->rq_pages[i]); 604 } 605 606 struct svc_rqst * 607 svc_prepare_thread(struct svc_serv *serv, struct svc_pool *pool, int node) 608 { 609 struct svc_rqst *rqstp; 610 611 rqstp = kzalloc_node(sizeof(*rqstp), GFP_KERNEL, node); 612 if (!rqstp) 613 goto out_enomem; 614 615 init_waitqueue_head(&rqstp->rq_wait); 616 617 serv->sv_nrthreads++; 618 spin_lock_bh(&pool->sp_lock); 619 pool->sp_nrthreads++; 620 list_add(&rqstp->rq_all, &pool->sp_all_threads); 621 spin_unlock_bh(&pool->sp_lock); 622 rqstp->rq_server = serv; 623 rqstp->rq_pool = pool; 624 625 rqstp->rq_argp = kmalloc_node(serv->sv_xdrsize, GFP_KERNEL, node); 626 if (!rqstp->rq_argp) 627 goto out_thread; 628 629 rqstp->rq_resp = kmalloc_node(serv->sv_xdrsize, GFP_KERNEL, node); 630 if (!rqstp->rq_resp) 631 goto out_thread; 632 633 if (!svc_init_buffer(rqstp, serv->sv_max_mesg, node)) 634 goto out_thread; 635 636 return rqstp; 637 out_thread: 638 svc_exit_thread(rqstp); 639 out_enomem: 640 return ERR_PTR(-ENOMEM); 641 } 642 EXPORT_SYMBOL_GPL(svc_prepare_thread); 643 644 /* 645 * Choose a pool in which to create a new thread, for svc_set_num_threads 646 */ 647 static inline struct svc_pool * 648 choose_pool(struct svc_serv *serv, struct svc_pool *pool, unsigned int *state) 649 { 650 if (pool != NULL) 651 return pool; 652 653 return &serv->sv_pools[(*state)++ % serv->sv_nrpools]; 654 } 655 656 /* 657 * Choose a thread to kill, for svc_set_num_threads 658 */ 659 static inline struct task_struct * 660 choose_victim(struct svc_serv *serv, struct svc_pool *pool, unsigned int *state) 661 { 662 unsigned int i; 663 struct task_struct *task = NULL; 664 665 if (pool != NULL) { 666 spin_lock_bh(&pool->sp_lock); 667 } else { 668 /* choose a pool in round-robin fashion */ 669 for (i = 0; i < serv->sv_nrpools; i++) { 670 pool = &serv->sv_pools[--(*state) % serv->sv_nrpools]; 671 spin_lock_bh(&pool->sp_lock); 672 if (!list_empty(&pool->sp_all_threads)) 673 goto found_pool; 674 spin_unlock_bh(&pool->sp_lock); 675 } 676 return NULL; 677 } 678 679 found_pool: 680 if (!list_empty(&pool->sp_all_threads)) { 681 struct svc_rqst *rqstp; 682 683 /* 684 * Remove from the pool->sp_all_threads list 685 * so we don't try to kill it again. 686 */ 687 rqstp = list_entry(pool->sp_all_threads.next, struct svc_rqst, rq_all); 688 list_del_init(&rqstp->rq_all); 689 task = rqstp->rq_task; 690 } 691 spin_unlock_bh(&pool->sp_lock); 692 693 return task; 694 } 695 696 /* 697 * Create or destroy enough new threads to make the number 698 * of threads the given number. If `pool' is non-NULL, applies 699 * only to threads in that pool, otherwise round-robins between 700 * all pools. Caller must ensure that mutual exclusion between this and 701 * server startup or shutdown. 702 * 703 * Destroying threads relies on the service threads filling in 704 * rqstp->rq_task, which only the nfs ones do. Assumes the serv 705 * has been created using svc_create_pooled(). 706 * 707 * Based on code that used to be in nfsd_svc() but tweaked 708 * to be pool-aware. 709 */ 710 int 711 svc_set_num_threads(struct svc_serv *serv, struct svc_pool *pool, int nrservs) 712 { 713 struct svc_rqst *rqstp; 714 struct task_struct *task; 715 struct svc_pool *chosen_pool; 716 int error = 0; 717 unsigned int state = serv->sv_nrthreads-1; 718 int node; 719 720 if (pool == NULL) { 721 /* The -1 assumes caller has done a svc_get() */ 722 nrservs -= (serv->sv_nrthreads-1); 723 } else { 724 spin_lock_bh(&pool->sp_lock); 725 nrservs -= pool->sp_nrthreads; 726 spin_unlock_bh(&pool->sp_lock); 727 } 728 729 /* create new threads */ 730 while (nrservs > 0) { 731 nrservs--; 732 chosen_pool = choose_pool(serv, pool, &state); 733 734 node = svc_pool_map_get_node(chosen_pool->sp_id); 735 rqstp = svc_prepare_thread(serv, chosen_pool, node); 736 if (IS_ERR(rqstp)) { 737 error = PTR_ERR(rqstp); 738 break; 739 } 740 741 __module_get(serv->sv_module); 742 task = kthread_create_on_node(serv->sv_function, rqstp, 743 node, serv->sv_name); 744 if (IS_ERR(task)) { 745 error = PTR_ERR(task); 746 module_put(serv->sv_module); 747 svc_exit_thread(rqstp); 748 break; 749 } 750 751 rqstp->rq_task = task; 752 if (serv->sv_nrpools > 1) 753 svc_pool_map_set_cpumask(task, chosen_pool->sp_id); 754 755 svc_sock_update_bufs(serv); 756 wake_up_process(task); 757 } 758 /* destroy old threads */ 759 while (nrservs < 0 && 760 (task = choose_victim(serv, pool, &state)) != NULL) { 761 send_sig(SIGINT, task, 1); 762 nrservs++; 763 } 764 765 return error; 766 } 767 EXPORT_SYMBOL_GPL(svc_set_num_threads); 768 769 /* 770 * Called from a server thread as it's exiting. Caller must hold the BKL or 771 * the "service mutex", whichever is appropriate for the service. 772 */ 773 void 774 svc_exit_thread(struct svc_rqst *rqstp) 775 { 776 struct svc_serv *serv = rqstp->rq_server; 777 struct svc_pool *pool = rqstp->rq_pool; 778 779 svc_release_buffer(rqstp); 780 kfree(rqstp->rq_resp); 781 kfree(rqstp->rq_argp); 782 kfree(rqstp->rq_auth_data); 783 784 spin_lock_bh(&pool->sp_lock); 785 pool->sp_nrthreads--; 786 list_del(&rqstp->rq_all); 787 spin_unlock_bh(&pool->sp_lock); 788 789 kfree(rqstp); 790 791 /* Release the server */ 792 if (serv) 793 svc_destroy(serv); 794 } 795 EXPORT_SYMBOL_GPL(svc_exit_thread); 796 797 /* 798 * Register an "inet" protocol family netid with the local 799 * rpcbind daemon via an rpcbind v4 SET request. 800 * 801 * No netconfig infrastructure is available in the kernel, so 802 * we map IP_ protocol numbers to netids by hand. 803 * 804 * Returns zero on success; a negative errno value is returned 805 * if any error occurs. 806 */ 807 static int __svc_rpcb_register4(struct net *net, const u32 program, 808 const u32 version, 809 const unsigned short protocol, 810 const unsigned short port) 811 { 812 const struct sockaddr_in sin = { 813 .sin_family = AF_INET, 814 .sin_addr.s_addr = htonl(INADDR_ANY), 815 .sin_port = htons(port), 816 }; 817 const char *netid; 818 int error; 819 820 switch (protocol) { 821 case IPPROTO_UDP: 822 netid = RPCBIND_NETID_UDP; 823 break; 824 case IPPROTO_TCP: 825 netid = RPCBIND_NETID_TCP; 826 break; 827 default: 828 return -ENOPROTOOPT; 829 } 830 831 error = rpcb_v4_register(net, program, version, 832 (const struct sockaddr *)&sin, netid); 833 834 /* 835 * User space didn't support rpcbind v4, so retry this 836 * registration request with the legacy rpcbind v2 protocol. 837 */ 838 if (error == -EPROTONOSUPPORT) 839 error = rpcb_register(net, program, version, protocol, port); 840 841 return error; 842 } 843 844 #if IS_ENABLED(CONFIG_IPV6) 845 /* 846 * Register an "inet6" protocol family netid with the local 847 * rpcbind daemon via an rpcbind v4 SET request. 848 * 849 * No netconfig infrastructure is available in the kernel, so 850 * we map IP_ protocol numbers to netids by hand. 851 * 852 * Returns zero on success; a negative errno value is returned 853 * if any error occurs. 854 */ 855 static int __svc_rpcb_register6(struct net *net, const u32 program, 856 const u32 version, 857 const unsigned short protocol, 858 const unsigned short port) 859 { 860 const struct sockaddr_in6 sin6 = { 861 .sin6_family = AF_INET6, 862 .sin6_addr = IN6ADDR_ANY_INIT, 863 .sin6_port = htons(port), 864 }; 865 const char *netid; 866 int error; 867 868 switch (protocol) { 869 case IPPROTO_UDP: 870 netid = RPCBIND_NETID_UDP6; 871 break; 872 case IPPROTO_TCP: 873 netid = RPCBIND_NETID_TCP6; 874 break; 875 default: 876 return -ENOPROTOOPT; 877 } 878 879 error = rpcb_v4_register(net, program, version, 880 (const struct sockaddr *)&sin6, netid); 881 882 /* 883 * User space didn't support rpcbind version 4, so we won't 884 * use a PF_INET6 listener. 885 */ 886 if (error == -EPROTONOSUPPORT) 887 error = -EAFNOSUPPORT; 888 889 return error; 890 } 891 #endif /* IS_ENABLED(CONFIG_IPV6) */ 892 893 /* 894 * Register a kernel RPC service via rpcbind version 4. 895 * 896 * Returns zero on success; a negative errno value is returned 897 * if any error occurs. 898 */ 899 static int __svc_register(struct net *net, const char *progname, 900 const u32 program, const u32 version, 901 const int family, 902 const unsigned short protocol, 903 const unsigned short port) 904 { 905 int error = -EAFNOSUPPORT; 906 907 switch (family) { 908 case PF_INET: 909 error = __svc_rpcb_register4(net, program, version, 910 protocol, port); 911 break; 912 #if IS_ENABLED(CONFIG_IPV6) 913 case PF_INET6: 914 error = __svc_rpcb_register6(net, program, version, 915 protocol, port); 916 #endif 917 } 918 919 if (error < 0) 920 printk(KERN_WARNING "svc: failed to register %sv%u RPC " 921 "service (errno %d).\n", progname, version, -error); 922 return error; 923 } 924 925 /** 926 * svc_register - register an RPC service with the local portmapper 927 * @serv: svc_serv struct for the service to register 928 * @net: net namespace for the service to register 929 * @family: protocol family of service's listener socket 930 * @proto: transport protocol number to advertise 931 * @port: port to advertise 932 * 933 * Service is registered for any address in the passed-in protocol family 934 */ 935 int svc_register(const struct svc_serv *serv, struct net *net, 936 const int family, const unsigned short proto, 937 const unsigned short port) 938 { 939 struct svc_program *progp; 940 unsigned int i; 941 int error = 0; 942 943 WARN_ON_ONCE(proto == 0 && port == 0); 944 if (proto == 0 && port == 0) 945 return -EINVAL; 946 947 for (progp = serv->sv_program; progp; progp = progp->pg_next) { 948 for (i = 0; i < progp->pg_nvers; i++) { 949 if (progp->pg_vers[i] == NULL) 950 continue; 951 952 dprintk("svc: svc_register(%sv%d, %s, %u, %u)%s\n", 953 progp->pg_name, 954 i, 955 proto == IPPROTO_UDP? "udp" : "tcp", 956 port, 957 family, 958 progp->pg_vers[i]->vs_hidden? 959 " (but not telling portmap)" : ""); 960 961 if (progp->pg_vers[i]->vs_hidden) 962 continue; 963 964 error = __svc_register(net, progp->pg_name, progp->pg_prog, 965 i, family, proto, port); 966 if (error < 0) 967 break; 968 } 969 } 970 971 return error; 972 } 973 974 /* 975 * If user space is running rpcbind, it should take the v4 UNSET 976 * and clear everything for this [program, version]. If user space 977 * is running portmap, it will reject the v4 UNSET, but won't have 978 * any "inet6" entries anyway. So a PMAP_UNSET should be sufficient 979 * in this case to clear all existing entries for [program, version]. 980 */ 981 static void __svc_unregister(struct net *net, const u32 program, const u32 version, 982 const char *progname) 983 { 984 int error; 985 986 error = rpcb_v4_register(net, program, version, NULL, ""); 987 988 /* 989 * User space didn't support rpcbind v4, so retry this 990 * request with the legacy rpcbind v2 protocol. 991 */ 992 if (error == -EPROTONOSUPPORT) 993 error = rpcb_register(net, program, version, 0, 0); 994 995 dprintk("svc: %s(%sv%u), error %d\n", 996 __func__, progname, version, error); 997 } 998 999 /* 1000 * All netids, bind addresses and ports registered for [program, version] 1001 * are removed from the local rpcbind database (if the service is not 1002 * hidden) to make way for a new instance of the service. 1003 * 1004 * The result of unregistration is reported via dprintk for those who want 1005 * verification of the result, but is otherwise not important. 1006 */ 1007 static void svc_unregister(const struct svc_serv *serv, struct net *net) 1008 { 1009 struct svc_program *progp; 1010 unsigned long flags; 1011 unsigned int i; 1012 1013 clear_thread_flag(TIF_SIGPENDING); 1014 1015 for (progp = serv->sv_program; progp; progp = progp->pg_next) { 1016 for (i = 0; i < progp->pg_nvers; i++) { 1017 if (progp->pg_vers[i] == NULL) 1018 continue; 1019 if (progp->pg_vers[i]->vs_hidden) 1020 continue; 1021 1022 dprintk("svc: attempting to unregister %sv%u\n", 1023 progp->pg_name, i); 1024 __svc_unregister(net, progp->pg_prog, i, progp->pg_name); 1025 } 1026 } 1027 1028 spin_lock_irqsave(¤t->sighand->siglock, flags); 1029 recalc_sigpending(); 1030 spin_unlock_irqrestore(¤t->sighand->siglock, flags); 1031 } 1032 1033 /* 1034 * dprintk the given error with the address of the client that caused it. 1035 */ 1036 #ifdef RPC_DEBUG 1037 static __printf(2, 3) 1038 void svc_printk(struct svc_rqst *rqstp, const char *fmt, ...) 1039 { 1040 struct va_format vaf; 1041 va_list args; 1042 char buf[RPC_MAX_ADDRBUFLEN]; 1043 1044 va_start(args, fmt); 1045 1046 vaf.fmt = fmt; 1047 vaf.va = &args; 1048 1049 dprintk("svc: %s: %pV", svc_print_addr(rqstp, buf, sizeof(buf)), &vaf); 1050 1051 va_end(args); 1052 } 1053 #else 1054 static __printf(2,3) void svc_printk(struct svc_rqst *rqstp, const char *fmt, ...) {} 1055 #endif 1056 1057 /* 1058 * Common routine for processing the RPC request. 1059 */ 1060 static int 1061 svc_process_common(struct svc_rqst *rqstp, struct kvec *argv, struct kvec *resv) 1062 { 1063 struct svc_program *progp; 1064 struct svc_version *versp = NULL; /* compiler food */ 1065 struct svc_procedure *procp = NULL; 1066 struct svc_serv *serv = rqstp->rq_server; 1067 kxdrproc_t xdr; 1068 __be32 *statp; 1069 u32 prog, vers, proc; 1070 __be32 auth_stat, rpc_stat; 1071 int auth_res; 1072 __be32 *reply_statp; 1073 1074 rpc_stat = rpc_success; 1075 1076 if (argv->iov_len < 6*4) 1077 goto err_short_len; 1078 1079 /* Will be turned off only in gss privacy case: */ 1080 rqstp->rq_splice_ok = 1; 1081 /* Will be turned off only when NFSv4 Sessions are used */ 1082 rqstp->rq_usedeferral = 1; 1083 rqstp->rq_dropme = false; 1084 1085 /* Setup reply header */ 1086 rqstp->rq_xprt->xpt_ops->xpo_prep_reply_hdr(rqstp); 1087 1088 svc_putu32(resv, rqstp->rq_xid); 1089 1090 vers = svc_getnl(argv); 1091 1092 /* First words of reply: */ 1093 svc_putnl(resv, 1); /* REPLY */ 1094 1095 if (vers != 2) /* RPC version number */ 1096 goto err_bad_rpc; 1097 1098 /* Save position in case we later decide to reject: */ 1099 reply_statp = resv->iov_base + resv->iov_len; 1100 1101 svc_putnl(resv, 0); /* ACCEPT */ 1102 1103 rqstp->rq_prog = prog = svc_getnl(argv); /* program number */ 1104 rqstp->rq_vers = vers = svc_getnl(argv); /* version number */ 1105 rqstp->rq_proc = proc = svc_getnl(argv); /* procedure number */ 1106 1107 progp = serv->sv_program; 1108 1109 for (progp = serv->sv_program; progp; progp = progp->pg_next) 1110 if (prog == progp->pg_prog) 1111 break; 1112 1113 /* 1114 * Decode auth data, and add verifier to reply buffer. 1115 * We do this before anything else in order to get a decent 1116 * auth verifier. 1117 */ 1118 auth_res = svc_authenticate(rqstp, &auth_stat); 1119 /* Also give the program a chance to reject this call: */ 1120 if (auth_res == SVC_OK && progp) { 1121 auth_stat = rpc_autherr_badcred; 1122 auth_res = progp->pg_authenticate(rqstp); 1123 } 1124 switch (auth_res) { 1125 case SVC_OK: 1126 break; 1127 case SVC_GARBAGE: 1128 goto err_garbage; 1129 case SVC_SYSERR: 1130 rpc_stat = rpc_system_err; 1131 goto err_bad; 1132 case SVC_DENIED: 1133 goto err_bad_auth; 1134 case SVC_CLOSE: 1135 if (test_bit(XPT_TEMP, &rqstp->rq_xprt->xpt_flags)) 1136 svc_close_xprt(rqstp->rq_xprt); 1137 case SVC_DROP: 1138 goto dropit; 1139 case SVC_COMPLETE: 1140 goto sendit; 1141 } 1142 1143 if (progp == NULL) 1144 goto err_bad_prog; 1145 1146 if (vers >= progp->pg_nvers || 1147 !(versp = progp->pg_vers[vers])) 1148 goto err_bad_vers; 1149 1150 procp = versp->vs_proc + proc; 1151 if (proc >= versp->vs_nproc || !procp->pc_func) 1152 goto err_bad_proc; 1153 rqstp->rq_procinfo = procp; 1154 1155 /* Syntactic check complete */ 1156 serv->sv_stats->rpccnt++; 1157 1158 /* Build the reply header. */ 1159 statp = resv->iov_base +resv->iov_len; 1160 svc_putnl(resv, RPC_SUCCESS); 1161 1162 /* Bump per-procedure stats counter */ 1163 procp->pc_count++; 1164 1165 /* Initialize storage for argp and resp */ 1166 memset(rqstp->rq_argp, 0, procp->pc_argsize); 1167 memset(rqstp->rq_resp, 0, procp->pc_ressize); 1168 1169 /* un-reserve some of the out-queue now that we have a 1170 * better idea of reply size 1171 */ 1172 if (procp->pc_xdrressize) 1173 svc_reserve_auth(rqstp, procp->pc_xdrressize<<2); 1174 1175 /* Call the function that processes the request. */ 1176 if (!versp->vs_dispatch) { 1177 /* Decode arguments */ 1178 xdr = procp->pc_decode; 1179 if (xdr && !xdr(rqstp, argv->iov_base, rqstp->rq_argp)) 1180 goto err_garbage; 1181 1182 *statp = procp->pc_func(rqstp, rqstp->rq_argp, rqstp->rq_resp); 1183 1184 /* Encode reply */ 1185 if (rqstp->rq_dropme) { 1186 if (procp->pc_release) 1187 procp->pc_release(rqstp, NULL, rqstp->rq_resp); 1188 goto dropit; 1189 } 1190 if (*statp == rpc_success && 1191 (xdr = procp->pc_encode) && 1192 !xdr(rqstp, resv->iov_base+resv->iov_len, rqstp->rq_resp)) { 1193 dprintk("svc: failed to encode reply\n"); 1194 /* serv->sv_stats->rpcsystemerr++; */ 1195 *statp = rpc_system_err; 1196 } 1197 } else { 1198 dprintk("svc: calling dispatcher\n"); 1199 if (!versp->vs_dispatch(rqstp, statp)) { 1200 /* Release reply info */ 1201 if (procp->pc_release) 1202 procp->pc_release(rqstp, NULL, rqstp->rq_resp); 1203 goto dropit; 1204 } 1205 } 1206 1207 /* Check RPC status result */ 1208 if (*statp != rpc_success) 1209 resv->iov_len = ((void*)statp) - resv->iov_base + 4; 1210 1211 /* Release reply info */ 1212 if (procp->pc_release) 1213 procp->pc_release(rqstp, NULL, rqstp->rq_resp); 1214 1215 if (procp->pc_encode == NULL) 1216 goto dropit; 1217 1218 sendit: 1219 if (svc_authorise(rqstp)) 1220 goto dropit; 1221 return 1; /* Caller can now send it */ 1222 1223 dropit: 1224 svc_authorise(rqstp); /* doesn't hurt to call this twice */ 1225 dprintk("svc: svc_process dropit\n"); 1226 return 0; 1227 1228 err_short_len: 1229 svc_printk(rqstp, "short len %Zd, dropping request\n", 1230 argv->iov_len); 1231 1232 goto dropit; /* drop request */ 1233 1234 err_bad_rpc: 1235 serv->sv_stats->rpcbadfmt++; 1236 svc_putnl(resv, 1); /* REJECT */ 1237 svc_putnl(resv, 0); /* RPC_MISMATCH */ 1238 svc_putnl(resv, 2); /* Only RPCv2 supported */ 1239 svc_putnl(resv, 2); 1240 goto sendit; 1241 1242 err_bad_auth: 1243 dprintk("svc: authentication failed (%d)\n", ntohl(auth_stat)); 1244 serv->sv_stats->rpcbadauth++; 1245 /* Restore write pointer to location of accept status: */ 1246 xdr_ressize_check(rqstp, reply_statp); 1247 svc_putnl(resv, 1); /* REJECT */ 1248 svc_putnl(resv, 1); /* AUTH_ERROR */ 1249 svc_putnl(resv, ntohl(auth_stat)); /* status */ 1250 goto sendit; 1251 1252 err_bad_prog: 1253 dprintk("svc: unknown program %d\n", prog); 1254 serv->sv_stats->rpcbadfmt++; 1255 svc_putnl(resv, RPC_PROG_UNAVAIL); 1256 goto sendit; 1257 1258 err_bad_vers: 1259 svc_printk(rqstp, "unknown version (%d for prog %d, %s)\n", 1260 vers, prog, progp->pg_name); 1261 1262 serv->sv_stats->rpcbadfmt++; 1263 svc_putnl(resv, RPC_PROG_MISMATCH); 1264 svc_putnl(resv, progp->pg_lovers); 1265 svc_putnl(resv, progp->pg_hivers); 1266 goto sendit; 1267 1268 err_bad_proc: 1269 svc_printk(rqstp, "unknown procedure (%d)\n", proc); 1270 1271 serv->sv_stats->rpcbadfmt++; 1272 svc_putnl(resv, RPC_PROC_UNAVAIL); 1273 goto sendit; 1274 1275 err_garbage: 1276 svc_printk(rqstp, "failed to decode args\n"); 1277 1278 rpc_stat = rpc_garbage_args; 1279 err_bad: 1280 serv->sv_stats->rpcbadfmt++; 1281 svc_putnl(resv, ntohl(rpc_stat)); 1282 goto sendit; 1283 } 1284 EXPORT_SYMBOL_GPL(svc_process); 1285 1286 /* 1287 * Process the RPC request. 1288 */ 1289 int 1290 svc_process(struct svc_rqst *rqstp) 1291 { 1292 struct kvec *argv = &rqstp->rq_arg.head[0]; 1293 struct kvec *resv = &rqstp->rq_res.head[0]; 1294 struct svc_serv *serv = rqstp->rq_server; 1295 u32 dir; 1296 1297 /* 1298 * Setup response xdr_buf. 1299 * Initially it has just one page 1300 */ 1301 rqstp->rq_next_page = &rqstp->rq_respages[1]; 1302 resv->iov_base = page_address(rqstp->rq_respages[0]); 1303 resv->iov_len = 0; 1304 rqstp->rq_res.pages = rqstp->rq_respages + 1; 1305 rqstp->rq_res.len = 0; 1306 rqstp->rq_res.page_base = 0; 1307 rqstp->rq_res.page_len = 0; 1308 rqstp->rq_res.buflen = PAGE_SIZE; 1309 rqstp->rq_res.tail[0].iov_base = NULL; 1310 rqstp->rq_res.tail[0].iov_len = 0; 1311 1312 rqstp->rq_xid = svc_getu32(argv); 1313 1314 dir = svc_getnl(argv); 1315 if (dir != 0) { 1316 /* direction != CALL */ 1317 svc_printk(rqstp, "bad direction %d, dropping request\n", dir); 1318 serv->sv_stats->rpcbadfmt++; 1319 svc_drop(rqstp); 1320 return 0; 1321 } 1322 1323 /* Returns 1 for send, 0 for drop */ 1324 if (svc_process_common(rqstp, argv, resv)) 1325 return svc_send(rqstp); 1326 else { 1327 svc_drop(rqstp); 1328 return 0; 1329 } 1330 } 1331 1332 #if defined(CONFIG_SUNRPC_BACKCHANNEL) 1333 /* 1334 * Process a backchannel RPC request that arrived over an existing 1335 * outbound connection 1336 */ 1337 int 1338 bc_svc_process(struct svc_serv *serv, struct rpc_rqst *req, 1339 struct svc_rqst *rqstp) 1340 { 1341 struct kvec *argv = &rqstp->rq_arg.head[0]; 1342 struct kvec *resv = &rqstp->rq_res.head[0]; 1343 1344 /* Build the svc_rqst used by the common processing routine */ 1345 rqstp->rq_xprt = serv->sv_bc_xprt; 1346 rqstp->rq_xid = req->rq_xid; 1347 rqstp->rq_prot = req->rq_xprt->prot; 1348 rqstp->rq_server = serv; 1349 1350 rqstp->rq_addrlen = sizeof(req->rq_xprt->addr); 1351 memcpy(&rqstp->rq_addr, &req->rq_xprt->addr, rqstp->rq_addrlen); 1352 memcpy(&rqstp->rq_arg, &req->rq_rcv_buf, sizeof(rqstp->rq_arg)); 1353 memcpy(&rqstp->rq_res, &req->rq_snd_buf, sizeof(rqstp->rq_res)); 1354 1355 /* reset result send buffer "put" position */ 1356 resv->iov_len = 0; 1357 1358 if (rqstp->rq_prot != IPPROTO_TCP) { 1359 printk(KERN_ERR "No support for Non-TCP transports!\n"); 1360 BUG(); 1361 } 1362 1363 /* 1364 * Skip the next two words because they've already been 1365 * processed in the trasport 1366 */ 1367 svc_getu32(argv); /* XID */ 1368 svc_getnl(argv); /* CALLDIR */ 1369 1370 /* Returns 1 for send, 0 for drop */ 1371 if (svc_process_common(rqstp, argv, resv)) { 1372 memcpy(&req->rq_snd_buf, &rqstp->rq_res, 1373 sizeof(req->rq_snd_buf)); 1374 return bc_send(req); 1375 } else { 1376 /* drop request */ 1377 xprt_free_bc_request(req); 1378 return 0; 1379 } 1380 } 1381 EXPORT_SYMBOL_GPL(bc_svc_process); 1382 #endif /* CONFIG_SUNRPC_BACKCHANNEL */ 1383 1384 /* 1385 * Return (transport-specific) limit on the rpc payload. 1386 */ 1387 u32 svc_max_payload(const struct svc_rqst *rqstp) 1388 { 1389 u32 max = rqstp->rq_xprt->xpt_class->xcl_max_payload; 1390 1391 if (rqstp->rq_server->sv_max_payload < max) 1392 max = rqstp->rq_server->sv_max_payload; 1393 return max; 1394 } 1395 EXPORT_SYMBOL_GPL(svc_max_payload); 1396