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 #include <trace/events/sunrpc.h> 32 33 #define RPCDBG_FACILITY RPCDBG_SVCDSP 34 35 static void svc_unregister(const struct svc_serv *serv, struct net *net); 36 37 #define svc_serv_is_pooled(serv) ((serv)->sv_function) 38 39 /* 40 * Mode for mapping cpus to pools. 41 */ 42 enum { 43 SVC_POOL_AUTO = -1, /* choose one of the others */ 44 SVC_POOL_GLOBAL, /* no mapping, just a single global pool 45 * (legacy & UP mode) */ 46 SVC_POOL_PERCPU, /* one pool per cpu */ 47 SVC_POOL_PERNODE /* one pool per numa node */ 48 }; 49 #define SVC_POOL_DEFAULT SVC_POOL_GLOBAL 50 51 /* 52 * Structure for mapping cpus to pools and vice versa. 53 * Setup once during sunrpc initialisation. 54 */ 55 static struct svc_pool_map { 56 int count; /* How many svc_servs use us */ 57 int mode; /* Note: int not enum to avoid 58 * warnings about "enumeration value 59 * not handled in switch" */ 60 unsigned int npools; 61 unsigned int *pool_to; /* maps pool id to cpu or node */ 62 unsigned int *to_pool; /* maps cpu or node to pool id */ 63 } svc_pool_map = { 64 .count = 0, 65 .mode = SVC_POOL_DEFAULT 66 }; 67 static DEFINE_MUTEX(svc_pool_map_mutex);/* protects svc_pool_map.count only */ 68 69 static int 70 param_set_pool_mode(const char *val, struct kernel_param *kp) 71 { 72 int *ip = (int *)kp->arg; 73 struct svc_pool_map *m = &svc_pool_map; 74 int err; 75 76 mutex_lock(&svc_pool_map_mutex); 77 78 err = -EBUSY; 79 if (m->count) 80 goto out; 81 82 err = 0; 83 if (!strncmp(val, "auto", 4)) 84 *ip = SVC_POOL_AUTO; 85 else if (!strncmp(val, "global", 6)) 86 *ip = SVC_POOL_GLOBAL; 87 else if (!strncmp(val, "percpu", 6)) 88 *ip = SVC_POOL_PERCPU; 89 else if (!strncmp(val, "pernode", 7)) 90 *ip = SVC_POOL_PERNODE; 91 else 92 err = -EINVAL; 93 94 out: 95 mutex_unlock(&svc_pool_map_mutex); 96 return err; 97 } 98 99 static int 100 param_get_pool_mode(char *buf, struct kernel_param *kp) 101 { 102 int *ip = (int *)kp->arg; 103 104 switch (*ip) 105 { 106 case SVC_POOL_AUTO: 107 return strlcpy(buf, "auto", 20); 108 case SVC_POOL_GLOBAL: 109 return strlcpy(buf, "global", 20); 110 case SVC_POOL_PERCPU: 111 return strlcpy(buf, "percpu", 20); 112 case SVC_POOL_PERNODE: 113 return strlcpy(buf, "pernode", 20); 114 default: 115 return sprintf(buf, "%d", *ip); 116 } 117 } 118 119 module_param_call(pool_mode, param_set_pool_mode, param_get_pool_mode, 120 &svc_pool_map.mode, 0644); 121 122 /* 123 * Detect best pool mapping mode heuristically, 124 * according to the machine's topology. 125 */ 126 static int 127 svc_pool_map_choose_mode(void) 128 { 129 unsigned int node; 130 131 if (nr_online_nodes > 1) { 132 /* 133 * Actually have multiple NUMA nodes, 134 * so split pools on NUMA node boundaries 135 */ 136 return SVC_POOL_PERNODE; 137 } 138 139 node = first_online_node; 140 if (nr_cpus_node(node) > 2) { 141 /* 142 * Non-trivial SMP, or CONFIG_NUMA on 143 * non-NUMA hardware, e.g. with a generic 144 * x86_64 kernel on Xeons. In this case we 145 * want to divide the pools on cpu boundaries. 146 */ 147 return SVC_POOL_PERCPU; 148 } 149 150 /* default: one global pool */ 151 return SVC_POOL_GLOBAL; 152 } 153 154 /* 155 * Allocate the to_pool[] and pool_to[] arrays. 156 * Returns 0 on success or an errno. 157 */ 158 static int 159 svc_pool_map_alloc_arrays(struct svc_pool_map *m, unsigned int maxpools) 160 { 161 m->to_pool = kcalloc(maxpools, sizeof(unsigned int), GFP_KERNEL); 162 if (!m->to_pool) 163 goto fail; 164 m->pool_to = kcalloc(maxpools, sizeof(unsigned int), GFP_KERNEL); 165 if (!m->pool_to) 166 goto fail_free; 167 168 return 0; 169 170 fail_free: 171 kfree(m->to_pool); 172 m->to_pool = NULL; 173 fail: 174 return -ENOMEM; 175 } 176 177 /* 178 * Initialise the pool map for SVC_POOL_PERCPU mode. 179 * Returns number of pools or <0 on error. 180 */ 181 static int 182 svc_pool_map_init_percpu(struct svc_pool_map *m) 183 { 184 unsigned int maxpools = nr_cpu_ids; 185 unsigned int pidx = 0; 186 unsigned int cpu; 187 int err; 188 189 err = svc_pool_map_alloc_arrays(m, maxpools); 190 if (err) 191 return err; 192 193 for_each_online_cpu(cpu) { 194 BUG_ON(pidx >= maxpools); 195 m->to_pool[cpu] = pidx; 196 m->pool_to[pidx] = cpu; 197 pidx++; 198 } 199 /* cpus brought online later all get mapped to pool0, sorry */ 200 201 return pidx; 202 }; 203 204 205 /* 206 * Initialise the pool map for SVC_POOL_PERNODE mode. 207 * Returns number of pools or <0 on error. 208 */ 209 static int 210 svc_pool_map_init_pernode(struct svc_pool_map *m) 211 { 212 unsigned int maxpools = nr_node_ids; 213 unsigned int pidx = 0; 214 unsigned int node; 215 int err; 216 217 err = svc_pool_map_alloc_arrays(m, maxpools); 218 if (err) 219 return err; 220 221 for_each_node_with_cpus(node) { 222 /* some architectures (e.g. SN2) have cpuless nodes */ 223 BUG_ON(pidx > maxpools); 224 m->to_pool[node] = pidx; 225 m->pool_to[pidx] = node; 226 pidx++; 227 } 228 /* nodes brought online later all get mapped to pool0, sorry */ 229 230 return pidx; 231 } 232 233 234 /* 235 * Add a reference to the global map of cpus to pools (and 236 * vice versa). Initialise the map if we're the first user. 237 * Returns the number of pools. 238 */ 239 static unsigned int 240 svc_pool_map_get(void) 241 { 242 struct svc_pool_map *m = &svc_pool_map; 243 int npools = -1; 244 245 mutex_lock(&svc_pool_map_mutex); 246 247 if (m->count++) { 248 mutex_unlock(&svc_pool_map_mutex); 249 return m->npools; 250 } 251 252 if (m->mode == SVC_POOL_AUTO) 253 m->mode = svc_pool_map_choose_mode(); 254 255 switch (m->mode) { 256 case SVC_POOL_PERCPU: 257 npools = svc_pool_map_init_percpu(m); 258 break; 259 case SVC_POOL_PERNODE: 260 npools = svc_pool_map_init_pernode(m); 261 break; 262 } 263 264 if (npools < 0) { 265 /* default, or memory allocation failure */ 266 npools = 1; 267 m->mode = SVC_POOL_GLOBAL; 268 } 269 m->npools = npools; 270 271 mutex_unlock(&svc_pool_map_mutex); 272 return m->npools; 273 } 274 275 276 /* 277 * Drop a reference to the global map of cpus to pools. 278 * When the last reference is dropped, the map data is 279 * freed; this allows the sysadmin to change the pool 280 * mode using the pool_mode module option without 281 * rebooting or re-loading sunrpc.ko. 282 */ 283 static void 284 svc_pool_map_put(void) 285 { 286 struct svc_pool_map *m = &svc_pool_map; 287 288 mutex_lock(&svc_pool_map_mutex); 289 290 if (!--m->count) { 291 kfree(m->to_pool); 292 m->to_pool = NULL; 293 kfree(m->pool_to); 294 m->pool_to = NULL; 295 m->npools = 0; 296 } 297 298 mutex_unlock(&svc_pool_map_mutex); 299 } 300 301 302 static int svc_pool_map_get_node(unsigned int pidx) 303 { 304 const struct svc_pool_map *m = &svc_pool_map; 305 306 if (m->count) { 307 if (m->mode == SVC_POOL_PERCPU) 308 return cpu_to_node(m->pool_to[pidx]); 309 if (m->mode == SVC_POOL_PERNODE) 310 return m->pool_to[pidx]; 311 } 312 return NUMA_NO_NODE; 313 } 314 /* 315 * Set the given thread's cpus_allowed mask so that it 316 * will only run on cpus in the given pool. 317 */ 318 static inline void 319 svc_pool_map_set_cpumask(struct task_struct *task, unsigned int pidx) 320 { 321 struct svc_pool_map *m = &svc_pool_map; 322 unsigned int node = m->pool_to[pidx]; 323 324 /* 325 * The caller checks for sv_nrpools > 1, which 326 * implies that we've been initialized. 327 */ 328 WARN_ON_ONCE(m->count == 0); 329 if (m->count == 0) 330 return; 331 332 switch (m->mode) { 333 case SVC_POOL_PERCPU: 334 { 335 set_cpus_allowed_ptr(task, cpumask_of(node)); 336 break; 337 } 338 case SVC_POOL_PERNODE: 339 { 340 set_cpus_allowed_ptr(task, cpumask_of_node(node)); 341 break; 342 } 343 } 344 } 345 346 /* 347 * Use the mapping mode to choose a pool for a given CPU. 348 * Used when enqueueing an incoming RPC. Always returns 349 * a non-NULL pool pointer. 350 */ 351 struct svc_pool * 352 svc_pool_for_cpu(struct svc_serv *serv, int cpu) 353 { 354 struct svc_pool_map *m = &svc_pool_map; 355 unsigned int pidx = 0; 356 357 /* 358 * An uninitialised map happens in a pure client when 359 * lockd is brought up, so silently treat it the 360 * same as SVC_POOL_GLOBAL. 361 */ 362 if (svc_serv_is_pooled(serv)) { 363 switch (m->mode) { 364 case SVC_POOL_PERCPU: 365 pidx = m->to_pool[cpu]; 366 break; 367 case SVC_POOL_PERNODE: 368 pidx = m->to_pool[cpu_to_node(cpu)]; 369 break; 370 } 371 } 372 return &serv->sv_pools[pidx % serv->sv_nrpools]; 373 } 374 375 int svc_rpcb_setup(struct svc_serv *serv, struct net *net) 376 { 377 int err; 378 379 err = rpcb_create_local(net); 380 if (err) 381 return err; 382 383 /* Remove any stale portmap registrations */ 384 svc_unregister(serv, net); 385 return 0; 386 } 387 EXPORT_SYMBOL_GPL(svc_rpcb_setup); 388 389 void svc_rpcb_cleanup(struct svc_serv *serv, struct net *net) 390 { 391 svc_unregister(serv, net); 392 rpcb_put_local(net); 393 } 394 EXPORT_SYMBOL_GPL(svc_rpcb_cleanup); 395 396 static int svc_uses_rpcbind(struct svc_serv *serv) 397 { 398 struct svc_program *progp; 399 unsigned int i; 400 401 for (progp = serv->sv_program; progp; progp = progp->pg_next) { 402 for (i = 0; i < progp->pg_nvers; i++) { 403 if (progp->pg_vers[i] == NULL) 404 continue; 405 if (progp->pg_vers[i]->vs_hidden == 0) 406 return 1; 407 } 408 } 409 410 return 0; 411 } 412 413 int svc_bind(struct svc_serv *serv, struct net *net) 414 { 415 if (!svc_uses_rpcbind(serv)) 416 return 0; 417 return svc_rpcb_setup(serv, net); 418 } 419 EXPORT_SYMBOL_GPL(svc_bind); 420 421 /* 422 * Create an RPC service 423 */ 424 static struct svc_serv * 425 __svc_create(struct svc_program *prog, unsigned int bufsize, int npools, 426 void (*shutdown)(struct svc_serv *serv, struct net *net)) 427 { 428 struct svc_serv *serv; 429 unsigned int vers; 430 unsigned int xdrsize; 431 unsigned int i; 432 433 if (!(serv = kzalloc(sizeof(*serv), GFP_KERNEL))) 434 return NULL; 435 serv->sv_name = prog->pg_name; 436 serv->sv_program = prog; 437 serv->sv_nrthreads = 1; 438 serv->sv_stats = prog->pg_stats; 439 if (bufsize > RPCSVC_MAXPAYLOAD) 440 bufsize = RPCSVC_MAXPAYLOAD; 441 serv->sv_max_payload = bufsize? bufsize : 4096; 442 serv->sv_max_mesg = roundup(serv->sv_max_payload + PAGE_SIZE, PAGE_SIZE); 443 serv->sv_shutdown = shutdown; 444 xdrsize = 0; 445 while (prog) { 446 prog->pg_lovers = prog->pg_nvers-1; 447 for (vers=0; vers<prog->pg_nvers ; vers++) 448 if (prog->pg_vers[vers]) { 449 prog->pg_hivers = vers; 450 if (prog->pg_lovers > vers) 451 prog->pg_lovers = vers; 452 if (prog->pg_vers[vers]->vs_xdrsize > xdrsize) 453 xdrsize = prog->pg_vers[vers]->vs_xdrsize; 454 } 455 prog = prog->pg_next; 456 } 457 serv->sv_xdrsize = xdrsize; 458 INIT_LIST_HEAD(&serv->sv_tempsocks); 459 INIT_LIST_HEAD(&serv->sv_permsocks); 460 init_timer(&serv->sv_temptimer); 461 spin_lock_init(&serv->sv_lock); 462 463 serv->sv_nrpools = npools; 464 serv->sv_pools = 465 kcalloc(serv->sv_nrpools, sizeof(struct svc_pool), 466 GFP_KERNEL); 467 if (!serv->sv_pools) { 468 kfree(serv); 469 return NULL; 470 } 471 472 for (i = 0; i < serv->sv_nrpools; i++) { 473 struct svc_pool *pool = &serv->sv_pools[i]; 474 475 dprintk("svc: initialising pool %u for %s\n", 476 i, serv->sv_name); 477 478 pool->sp_id = i; 479 INIT_LIST_HEAD(&pool->sp_sockets); 480 INIT_LIST_HEAD(&pool->sp_all_threads); 481 spin_lock_init(&pool->sp_lock); 482 } 483 484 return serv; 485 } 486 487 struct svc_serv * 488 svc_create(struct svc_program *prog, unsigned int bufsize, 489 void (*shutdown)(struct svc_serv *serv, struct net *net)) 490 { 491 return __svc_create(prog, bufsize, /*npools*/1, shutdown); 492 } 493 EXPORT_SYMBOL_GPL(svc_create); 494 495 struct svc_serv * 496 svc_create_pooled(struct svc_program *prog, unsigned int bufsize, 497 void (*shutdown)(struct svc_serv *serv, struct net *net), 498 svc_thread_fn func, struct module *mod) 499 { 500 struct svc_serv *serv; 501 unsigned int npools = svc_pool_map_get(); 502 503 serv = __svc_create(prog, bufsize, npools, shutdown); 504 if (!serv) 505 goto out_err; 506 507 serv->sv_function = func; 508 serv->sv_module = mod; 509 return serv; 510 out_err: 511 svc_pool_map_put(); 512 return NULL; 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 serv->sv_nrthreads++; 616 __set_bit(RQ_BUSY, &rqstp->rq_flags); 617 spin_lock_init(&rqstp->rq_lock); 618 rqstp->rq_server = serv; 619 rqstp->rq_pool = pool; 620 spin_lock_bh(&pool->sp_lock); 621 pool->sp_nrthreads++; 622 list_add_rcu(&rqstp->rq_all, &pool->sp_all_threads); 623 spin_unlock_bh(&pool->sp_lock); 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 set_bit(RQ_VICTIM, &rqstp->rq_flags); 689 list_del_rcu(&rqstp->rq_all); 690 task = rqstp->rq_task; 691 } 692 spin_unlock_bh(&pool->sp_lock); 693 694 return task; 695 } 696 697 /* 698 * Create or destroy enough new threads to make the number 699 * of threads the given number. If `pool' is non-NULL, applies 700 * only to threads in that pool, otherwise round-robins between 701 * all pools. Caller must ensure that mutual exclusion between this and 702 * server startup or shutdown. 703 * 704 * Destroying threads relies on the service threads filling in 705 * rqstp->rq_task, which only the nfs ones do. Assumes the serv 706 * has been created using svc_create_pooled(). 707 * 708 * Based on code that used to be in nfsd_svc() but tweaked 709 * to be pool-aware. 710 */ 711 int 712 svc_set_num_threads(struct svc_serv *serv, struct svc_pool *pool, int nrservs) 713 { 714 struct svc_rqst *rqstp; 715 struct task_struct *task; 716 struct svc_pool *chosen_pool; 717 int error = 0; 718 unsigned int state = serv->sv_nrthreads-1; 719 int node; 720 721 if (pool == NULL) { 722 /* The -1 assumes caller has done a svc_get() */ 723 nrservs -= (serv->sv_nrthreads-1); 724 } else { 725 spin_lock_bh(&pool->sp_lock); 726 nrservs -= pool->sp_nrthreads; 727 spin_unlock_bh(&pool->sp_lock); 728 } 729 730 /* create new threads */ 731 while (nrservs > 0) { 732 nrservs--; 733 chosen_pool = choose_pool(serv, pool, &state); 734 735 node = svc_pool_map_get_node(chosen_pool->sp_id); 736 rqstp = svc_prepare_thread(serv, chosen_pool, node); 737 if (IS_ERR(rqstp)) { 738 error = PTR_ERR(rqstp); 739 break; 740 } 741 742 __module_get(serv->sv_module); 743 task = kthread_create_on_node(serv->sv_function, rqstp, 744 node, "%s", serv->sv_name); 745 if (IS_ERR(task)) { 746 error = PTR_ERR(task); 747 module_put(serv->sv_module); 748 svc_exit_thread(rqstp); 749 break; 750 } 751 752 rqstp->rq_task = task; 753 if (serv->sv_nrpools > 1) 754 svc_pool_map_set_cpumask(task, chosen_pool->sp_id); 755 756 svc_sock_update_bufs(serv); 757 wake_up_process(task); 758 } 759 /* destroy old threads */ 760 while (nrservs < 0 && 761 (task = choose_victim(serv, pool, &state)) != NULL) { 762 send_sig(SIGINT, task, 1); 763 nrservs++; 764 } 765 766 return error; 767 } 768 EXPORT_SYMBOL_GPL(svc_set_num_threads); 769 770 /* 771 * Called from a server thread as it's exiting. Caller must hold the BKL or 772 * the "service mutex", whichever is appropriate for the service. 773 */ 774 void 775 svc_exit_thread(struct svc_rqst *rqstp) 776 { 777 struct svc_serv *serv = rqstp->rq_server; 778 struct svc_pool *pool = rqstp->rq_pool; 779 780 svc_release_buffer(rqstp); 781 kfree(rqstp->rq_resp); 782 kfree(rqstp->rq_argp); 783 kfree(rqstp->rq_auth_data); 784 785 spin_lock_bh(&pool->sp_lock); 786 pool->sp_nrthreads--; 787 if (!test_and_set_bit(RQ_VICTIM, &rqstp->rq_flags)) 788 list_del_rcu(&rqstp->rq_all); 789 spin_unlock_bh(&pool->sp_lock); 790 791 kfree_rcu(rqstp, rq_rcu_head); 792 793 /* Release the server */ 794 if (serv) 795 svc_destroy(serv); 796 } 797 EXPORT_SYMBOL_GPL(svc_exit_thread); 798 799 /* 800 * Register an "inet" protocol family netid with the local 801 * rpcbind daemon via an rpcbind v4 SET request. 802 * 803 * No netconfig infrastructure is available in the kernel, so 804 * we map IP_ protocol numbers to netids by hand. 805 * 806 * Returns zero on success; a negative errno value is returned 807 * if any error occurs. 808 */ 809 static int __svc_rpcb_register4(struct net *net, const u32 program, 810 const u32 version, 811 const unsigned short protocol, 812 const unsigned short port) 813 { 814 const struct sockaddr_in sin = { 815 .sin_family = AF_INET, 816 .sin_addr.s_addr = htonl(INADDR_ANY), 817 .sin_port = htons(port), 818 }; 819 const char *netid; 820 int error; 821 822 switch (protocol) { 823 case IPPROTO_UDP: 824 netid = RPCBIND_NETID_UDP; 825 break; 826 case IPPROTO_TCP: 827 netid = RPCBIND_NETID_TCP; 828 break; 829 default: 830 return -ENOPROTOOPT; 831 } 832 833 error = rpcb_v4_register(net, program, version, 834 (const struct sockaddr *)&sin, netid); 835 836 /* 837 * User space didn't support rpcbind v4, so retry this 838 * registration request with the legacy rpcbind v2 protocol. 839 */ 840 if (error == -EPROTONOSUPPORT) 841 error = rpcb_register(net, program, version, protocol, port); 842 843 return error; 844 } 845 846 #if IS_ENABLED(CONFIG_IPV6) 847 /* 848 * Register an "inet6" protocol family netid with the local 849 * rpcbind daemon via an rpcbind v4 SET request. 850 * 851 * No netconfig infrastructure is available in the kernel, so 852 * we map IP_ protocol numbers to netids by hand. 853 * 854 * Returns zero on success; a negative errno value is returned 855 * if any error occurs. 856 */ 857 static int __svc_rpcb_register6(struct net *net, const u32 program, 858 const u32 version, 859 const unsigned short protocol, 860 const unsigned short port) 861 { 862 const struct sockaddr_in6 sin6 = { 863 .sin6_family = AF_INET6, 864 .sin6_addr = IN6ADDR_ANY_INIT, 865 .sin6_port = htons(port), 866 }; 867 const char *netid; 868 int error; 869 870 switch (protocol) { 871 case IPPROTO_UDP: 872 netid = RPCBIND_NETID_UDP6; 873 break; 874 case IPPROTO_TCP: 875 netid = RPCBIND_NETID_TCP6; 876 break; 877 default: 878 return -ENOPROTOOPT; 879 } 880 881 error = rpcb_v4_register(net, program, version, 882 (const struct sockaddr *)&sin6, netid); 883 884 /* 885 * User space didn't support rpcbind version 4, so we won't 886 * use a PF_INET6 listener. 887 */ 888 if (error == -EPROTONOSUPPORT) 889 error = -EAFNOSUPPORT; 890 891 return error; 892 } 893 #endif /* IS_ENABLED(CONFIG_IPV6) */ 894 895 /* 896 * Register a kernel RPC service via rpcbind version 4. 897 * 898 * Returns zero on success; a negative errno value is returned 899 * if any error occurs. 900 */ 901 static int __svc_register(struct net *net, const char *progname, 902 const u32 program, const u32 version, 903 const int family, 904 const unsigned short protocol, 905 const unsigned short port) 906 { 907 int error = -EAFNOSUPPORT; 908 909 switch (family) { 910 case PF_INET: 911 error = __svc_rpcb_register4(net, program, version, 912 protocol, port); 913 break; 914 #if IS_ENABLED(CONFIG_IPV6) 915 case PF_INET6: 916 error = __svc_rpcb_register6(net, program, version, 917 protocol, port); 918 #endif 919 } 920 921 return error; 922 } 923 924 /** 925 * svc_register - register an RPC service with the local portmapper 926 * @serv: svc_serv struct for the service to register 927 * @net: net namespace for the service to register 928 * @family: protocol family of service's listener socket 929 * @proto: transport protocol number to advertise 930 * @port: port to advertise 931 * 932 * Service is registered for any address in the passed-in protocol family 933 */ 934 int svc_register(const struct svc_serv *serv, struct net *net, 935 const int family, const unsigned short proto, 936 const unsigned short port) 937 { 938 struct svc_program *progp; 939 struct svc_version *vers; 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 vers = progp->pg_vers[i]; 950 if (vers == NULL) 951 continue; 952 953 dprintk("svc: svc_register(%sv%d, %s, %u, %u)%s\n", 954 progp->pg_name, 955 i, 956 proto == IPPROTO_UDP? "udp" : "tcp", 957 port, 958 family, 959 vers->vs_hidden ? 960 " (but not telling portmap)" : ""); 961 962 if (vers->vs_hidden) 963 continue; 964 965 error = __svc_register(net, progp->pg_name, progp->pg_prog, 966 i, family, proto, port); 967 968 if (vers->vs_rpcb_optnl) { 969 error = 0; 970 continue; 971 } 972 973 if (error < 0) { 974 printk(KERN_WARNING "svc: failed to register " 975 "%sv%u RPC service (errno %d).\n", 976 progp->pg_name, i, -error); 977 break; 978 } 979 } 980 } 981 982 return error; 983 } 984 985 /* 986 * If user space is running rpcbind, it should take the v4 UNSET 987 * and clear everything for this [program, version]. If user space 988 * is running portmap, it will reject the v4 UNSET, but won't have 989 * any "inet6" entries anyway. So a PMAP_UNSET should be sufficient 990 * in this case to clear all existing entries for [program, version]. 991 */ 992 static void __svc_unregister(struct net *net, const u32 program, const u32 version, 993 const char *progname) 994 { 995 int error; 996 997 error = rpcb_v4_register(net, program, version, NULL, ""); 998 999 /* 1000 * User space didn't support rpcbind v4, so retry this 1001 * request with the legacy rpcbind v2 protocol. 1002 */ 1003 if (error == -EPROTONOSUPPORT) 1004 error = rpcb_register(net, program, version, 0, 0); 1005 1006 dprintk("svc: %s(%sv%u), error %d\n", 1007 __func__, progname, version, error); 1008 } 1009 1010 /* 1011 * All netids, bind addresses and ports registered for [program, version] 1012 * are removed from the local rpcbind database (if the service is not 1013 * hidden) to make way for a new instance of the service. 1014 * 1015 * The result of unregistration is reported via dprintk for those who want 1016 * verification of the result, but is otherwise not important. 1017 */ 1018 static void svc_unregister(const struct svc_serv *serv, struct net *net) 1019 { 1020 struct svc_program *progp; 1021 unsigned long flags; 1022 unsigned int i; 1023 1024 clear_thread_flag(TIF_SIGPENDING); 1025 1026 for (progp = serv->sv_program; progp; progp = progp->pg_next) { 1027 for (i = 0; i < progp->pg_nvers; i++) { 1028 if (progp->pg_vers[i] == NULL) 1029 continue; 1030 if (progp->pg_vers[i]->vs_hidden) 1031 continue; 1032 1033 dprintk("svc: attempting to unregister %sv%u\n", 1034 progp->pg_name, i); 1035 __svc_unregister(net, progp->pg_prog, i, progp->pg_name); 1036 } 1037 } 1038 1039 spin_lock_irqsave(¤t->sighand->siglock, flags); 1040 recalc_sigpending(); 1041 spin_unlock_irqrestore(¤t->sighand->siglock, flags); 1042 } 1043 1044 /* 1045 * dprintk the given error with the address of the client that caused it. 1046 */ 1047 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG) 1048 static __printf(2, 3) 1049 void svc_printk(struct svc_rqst *rqstp, const char *fmt, ...) 1050 { 1051 struct va_format vaf; 1052 va_list args; 1053 char buf[RPC_MAX_ADDRBUFLEN]; 1054 1055 va_start(args, fmt); 1056 1057 vaf.fmt = fmt; 1058 vaf.va = &args; 1059 1060 dprintk("svc: %s: %pV", svc_print_addr(rqstp, buf, sizeof(buf)), &vaf); 1061 1062 va_end(args); 1063 } 1064 #else 1065 static __printf(2,3) void svc_printk(struct svc_rqst *rqstp, const char *fmt, ...) {} 1066 #endif 1067 1068 /* 1069 * Common routine for processing the RPC request. 1070 */ 1071 static int 1072 svc_process_common(struct svc_rqst *rqstp, struct kvec *argv, struct kvec *resv) 1073 { 1074 struct svc_program *progp; 1075 struct svc_version *versp = NULL; /* compiler food */ 1076 struct svc_procedure *procp = NULL; 1077 struct svc_serv *serv = rqstp->rq_server; 1078 kxdrproc_t xdr; 1079 __be32 *statp; 1080 u32 prog, vers, proc; 1081 __be32 auth_stat, rpc_stat; 1082 int auth_res; 1083 __be32 *reply_statp; 1084 1085 rpc_stat = rpc_success; 1086 1087 if (argv->iov_len < 6*4) 1088 goto err_short_len; 1089 1090 /* Will be turned off only in gss privacy case: */ 1091 set_bit(RQ_SPLICE_OK, &rqstp->rq_flags); 1092 /* Will be turned off only when NFSv4 Sessions are used */ 1093 set_bit(RQ_USEDEFERRAL, &rqstp->rq_flags); 1094 clear_bit(RQ_DROPME, &rqstp->rq_flags); 1095 1096 /* Setup reply header */ 1097 rqstp->rq_xprt->xpt_ops->xpo_prep_reply_hdr(rqstp); 1098 1099 svc_putu32(resv, rqstp->rq_xid); 1100 1101 vers = svc_getnl(argv); 1102 1103 /* First words of reply: */ 1104 svc_putnl(resv, 1); /* REPLY */ 1105 1106 if (vers != 2) /* RPC version number */ 1107 goto err_bad_rpc; 1108 1109 /* Save position in case we later decide to reject: */ 1110 reply_statp = resv->iov_base + resv->iov_len; 1111 1112 svc_putnl(resv, 0); /* ACCEPT */ 1113 1114 rqstp->rq_prog = prog = svc_getnl(argv); /* program number */ 1115 rqstp->rq_vers = vers = svc_getnl(argv); /* version number */ 1116 rqstp->rq_proc = proc = svc_getnl(argv); /* procedure number */ 1117 1118 for (progp = serv->sv_program; progp; progp = progp->pg_next) 1119 if (prog == progp->pg_prog) 1120 break; 1121 1122 /* 1123 * Decode auth data, and add verifier to reply buffer. 1124 * We do this before anything else in order to get a decent 1125 * auth verifier. 1126 */ 1127 auth_res = svc_authenticate(rqstp, &auth_stat); 1128 /* Also give the program a chance to reject this call: */ 1129 if (auth_res == SVC_OK && progp) { 1130 auth_stat = rpc_autherr_badcred; 1131 auth_res = progp->pg_authenticate(rqstp); 1132 } 1133 switch (auth_res) { 1134 case SVC_OK: 1135 break; 1136 case SVC_GARBAGE: 1137 goto err_garbage; 1138 case SVC_SYSERR: 1139 rpc_stat = rpc_system_err; 1140 goto err_bad; 1141 case SVC_DENIED: 1142 goto err_bad_auth; 1143 case SVC_CLOSE: 1144 if (test_bit(XPT_TEMP, &rqstp->rq_xprt->xpt_flags)) 1145 svc_close_xprt(rqstp->rq_xprt); 1146 case SVC_DROP: 1147 goto dropit; 1148 case SVC_COMPLETE: 1149 goto sendit; 1150 } 1151 1152 if (progp == NULL) 1153 goto err_bad_prog; 1154 1155 if (vers >= progp->pg_nvers || 1156 !(versp = progp->pg_vers[vers])) 1157 goto err_bad_vers; 1158 1159 procp = versp->vs_proc + proc; 1160 if (proc >= versp->vs_nproc || !procp->pc_func) 1161 goto err_bad_proc; 1162 rqstp->rq_procinfo = procp; 1163 1164 /* Syntactic check complete */ 1165 serv->sv_stats->rpccnt++; 1166 1167 /* Build the reply header. */ 1168 statp = resv->iov_base +resv->iov_len; 1169 svc_putnl(resv, RPC_SUCCESS); 1170 1171 /* Bump per-procedure stats counter */ 1172 procp->pc_count++; 1173 1174 /* Initialize storage for argp and resp */ 1175 memset(rqstp->rq_argp, 0, procp->pc_argsize); 1176 memset(rqstp->rq_resp, 0, procp->pc_ressize); 1177 1178 /* un-reserve some of the out-queue now that we have a 1179 * better idea of reply size 1180 */ 1181 if (procp->pc_xdrressize) 1182 svc_reserve_auth(rqstp, procp->pc_xdrressize<<2); 1183 1184 /* Call the function that processes the request. */ 1185 if (!versp->vs_dispatch) { 1186 /* Decode arguments */ 1187 xdr = procp->pc_decode; 1188 if (xdr && !xdr(rqstp, argv->iov_base, rqstp->rq_argp)) 1189 goto err_garbage; 1190 1191 *statp = procp->pc_func(rqstp, rqstp->rq_argp, rqstp->rq_resp); 1192 1193 /* Encode reply */ 1194 if (test_bit(RQ_DROPME, &rqstp->rq_flags)) { 1195 if (procp->pc_release) 1196 procp->pc_release(rqstp, NULL, rqstp->rq_resp); 1197 goto dropit; 1198 } 1199 if (*statp == rpc_success && 1200 (xdr = procp->pc_encode) && 1201 !xdr(rqstp, resv->iov_base+resv->iov_len, rqstp->rq_resp)) { 1202 dprintk("svc: failed to encode reply\n"); 1203 /* serv->sv_stats->rpcsystemerr++; */ 1204 *statp = rpc_system_err; 1205 } 1206 } else { 1207 dprintk("svc: calling dispatcher\n"); 1208 if (!versp->vs_dispatch(rqstp, statp)) { 1209 /* Release reply info */ 1210 if (procp->pc_release) 1211 procp->pc_release(rqstp, NULL, rqstp->rq_resp); 1212 goto dropit; 1213 } 1214 } 1215 1216 /* Check RPC status result */ 1217 if (*statp != rpc_success) 1218 resv->iov_len = ((void*)statp) - resv->iov_base + 4; 1219 1220 /* Release reply info */ 1221 if (procp->pc_release) 1222 procp->pc_release(rqstp, NULL, rqstp->rq_resp); 1223 1224 if (procp->pc_encode == NULL) 1225 goto dropit; 1226 1227 sendit: 1228 if (svc_authorise(rqstp)) 1229 goto dropit; 1230 return 1; /* Caller can now send it */ 1231 1232 dropit: 1233 svc_authorise(rqstp); /* doesn't hurt to call this twice */ 1234 dprintk("svc: svc_process dropit\n"); 1235 return 0; 1236 1237 err_short_len: 1238 svc_printk(rqstp, "short len %Zd, dropping request\n", 1239 argv->iov_len); 1240 1241 goto dropit; /* drop request */ 1242 1243 err_bad_rpc: 1244 serv->sv_stats->rpcbadfmt++; 1245 svc_putnl(resv, 1); /* REJECT */ 1246 svc_putnl(resv, 0); /* RPC_MISMATCH */ 1247 svc_putnl(resv, 2); /* Only RPCv2 supported */ 1248 svc_putnl(resv, 2); 1249 goto sendit; 1250 1251 err_bad_auth: 1252 dprintk("svc: authentication failed (%d)\n", ntohl(auth_stat)); 1253 serv->sv_stats->rpcbadauth++; 1254 /* Restore write pointer to location of accept status: */ 1255 xdr_ressize_check(rqstp, reply_statp); 1256 svc_putnl(resv, 1); /* REJECT */ 1257 svc_putnl(resv, 1); /* AUTH_ERROR */ 1258 svc_putnl(resv, ntohl(auth_stat)); /* status */ 1259 goto sendit; 1260 1261 err_bad_prog: 1262 dprintk("svc: unknown program %d\n", prog); 1263 serv->sv_stats->rpcbadfmt++; 1264 svc_putnl(resv, RPC_PROG_UNAVAIL); 1265 goto sendit; 1266 1267 err_bad_vers: 1268 svc_printk(rqstp, "unknown version (%d for prog %d, %s)\n", 1269 vers, prog, progp->pg_name); 1270 1271 serv->sv_stats->rpcbadfmt++; 1272 svc_putnl(resv, RPC_PROG_MISMATCH); 1273 svc_putnl(resv, progp->pg_lovers); 1274 svc_putnl(resv, progp->pg_hivers); 1275 goto sendit; 1276 1277 err_bad_proc: 1278 svc_printk(rqstp, "unknown procedure (%d)\n", proc); 1279 1280 serv->sv_stats->rpcbadfmt++; 1281 svc_putnl(resv, RPC_PROC_UNAVAIL); 1282 goto sendit; 1283 1284 err_garbage: 1285 svc_printk(rqstp, "failed to decode args\n"); 1286 1287 rpc_stat = rpc_garbage_args; 1288 err_bad: 1289 serv->sv_stats->rpcbadfmt++; 1290 svc_putnl(resv, ntohl(rpc_stat)); 1291 goto sendit; 1292 } 1293 EXPORT_SYMBOL_GPL(svc_process); 1294 1295 /* 1296 * Process the RPC request. 1297 */ 1298 int 1299 svc_process(struct svc_rqst *rqstp) 1300 { 1301 struct kvec *argv = &rqstp->rq_arg.head[0]; 1302 struct kvec *resv = &rqstp->rq_res.head[0]; 1303 struct svc_serv *serv = rqstp->rq_server; 1304 u32 dir; 1305 1306 /* 1307 * Setup response xdr_buf. 1308 * Initially it has just one page 1309 */ 1310 rqstp->rq_next_page = &rqstp->rq_respages[1]; 1311 resv->iov_base = page_address(rqstp->rq_respages[0]); 1312 resv->iov_len = 0; 1313 rqstp->rq_res.pages = rqstp->rq_respages + 1; 1314 rqstp->rq_res.len = 0; 1315 rqstp->rq_res.page_base = 0; 1316 rqstp->rq_res.page_len = 0; 1317 rqstp->rq_res.buflen = PAGE_SIZE; 1318 rqstp->rq_res.tail[0].iov_base = NULL; 1319 rqstp->rq_res.tail[0].iov_len = 0; 1320 1321 dir = svc_getnl(argv); 1322 if (dir != 0) { 1323 /* direction != CALL */ 1324 svc_printk(rqstp, "bad direction %d, dropping request\n", dir); 1325 serv->sv_stats->rpcbadfmt++; 1326 goto out_drop; 1327 } 1328 1329 /* Returns 1 for send, 0 for drop */ 1330 if (likely(svc_process_common(rqstp, argv, resv))) { 1331 int ret = svc_send(rqstp); 1332 1333 trace_svc_process(rqstp, ret); 1334 return ret; 1335 } 1336 out_drop: 1337 trace_svc_process(rqstp, 0); 1338 svc_drop(rqstp); 1339 return 0; 1340 } 1341 1342 #if defined(CONFIG_SUNRPC_BACKCHANNEL) 1343 /* 1344 * Process a backchannel RPC request that arrived over an existing 1345 * outbound connection 1346 */ 1347 int 1348 bc_svc_process(struct svc_serv *serv, struct rpc_rqst *req, 1349 struct svc_rqst *rqstp) 1350 { 1351 struct kvec *argv = &rqstp->rq_arg.head[0]; 1352 struct kvec *resv = &rqstp->rq_res.head[0]; 1353 1354 /* Build the svc_rqst used by the common processing routine */ 1355 rqstp->rq_xprt = serv->sv_bc_xprt; 1356 rqstp->rq_xid = req->rq_xid; 1357 rqstp->rq_prot = req->rq_xprt->prot; 1358 rqstp->rq_server = serv; 1359 1360 rqstp->rq_addrlen = sizeof(req->rq_xprt->addr); 1361 memcpy(&rqstp->rq_addr, &req->rq_xprt->addr, rqstp->rq_addrlen); 1362 memcpy(&rqstp->rq_arg, &req->rq_rcv_buf, sizeof(rqstp->rq_arg)); 1363 memcpy(&rqstp->rq_res, &req->rq_snd_buf, sizeof(rqstp->rq_res)); 1364 1365 /* reset result send buffer "put" position */ 1366 resv->iov_len = 0; 1367 1368 if (rqstp->rq_prot != IPPROTO_TCP) { 1369 printk(KERN_ERR "No support for Non-TCP transports!\n"); 1370 BUG(); 1371 } 1372 1373 /* 1374 * Skip the next two words because they've already been 1375 * processed in the trasport 1376 */ 1377 svc_getu32(argv); /* XID */ 1378 svc_getnl(argv); /* CALLDIR */ 1379 1380 /* Returns 1 for send, 0 for drop */ 1381 if (svc_process_common(rqstp, argv, resv)) { 1382 memcpy(&req->rq_snd_buf, &rqstp->rq_res, 1383 sizeof(req->rq_snd_buf)); 1384 return bc_send(req); 1385 } else { 1386 /* drop request */ 1387 xprt_free_bc_request(req); 1388 return 0; 1389 } 1390 } 1391 EXPORT_SYMBOL_GPL(bc_svc_process); 1392 #endif /* CONFIG_SUNRPC_BACKCHANNEL */ 1393 1394 /* 1395 * Return (transport-specific) limit on the rpc payload. 1396 */ 1397 u32 svc_max_payload(const struct svc_rqst *rqstp) 1398 { 1399 u32 max = rqstp->rq_xprt->xpt_class->xcl_max_payload; 1400 1401 if (rqstp->rq_server->sv_max_payload < max) 1402 max = rqstp->rq_server->sv_max_payload; 1403 return max; 1404 } 1405 EXPORT_SYMBOL_GPL(svc_max_payload); 1406