1 // SPDX-License-Identifier: GPL-2.0-only 2 /* 3 * linux/net/sunrpc/svc.c 4 * 5 * High-level RPC service routines 6 * 7 * Copyright (C) 1995, 1996 Olaf Kirch <okir@monad.swb.de> 8 * 9 * Multiple threads pools and NUMAisation 10 * Copyright (c) 2006 Silicon Graphics, Inc. 11 * by Greg Banks <gnb@melbourne.sgi.com> 12 */ 13 14 #include <linux/linkage.h> 15 #include <linux/sched/signal.h> 16 #include <linux/errno.h> 17 #include <linux/net.h> 18 #include <linux/in.h> 19 #include <linux/mm.h> 20 #include <linux/interrupt.h> 21 #include <linux/module.h> 22 #include <linux/kthread.h> 23 #include <linux/slab.h> 24 25 #include <linux/sunrpc/types.h> 26 #include <linux/sunrpc/xdr.h> 27 #include <linux/sunrpc/stats.h> 28 #include <linux/sunrpc/svcsock.h> 29 #include <linux/sunrpc/clnt.h> 30 #include <linux/sunrpc/bc_xprt.h> 31 32 #include <trace/events/sunrpc.h> 33 34 #include "fail.h" 35 #include "sunrpc.h" 36 37 #define RPCDBG_FACILITY RPCDBG_SVCDSP 38 39 static void svc_unregister(const struct svc_serv *serv, struct net *net); 40 41 #define SVC_POOL_DEFAULT SVC_POOL_GLOBAL 42 43 /* 44 * Mode for mapping cpus to pools. 45 */ 46 enum { 47 SVC_POOL_AUTO = -1, /* choose one of the others */ 48 SVC_POOL_GLOBAL, /* no mapping, just a single global pool 49 * (legacy & UP mode) */ 50 SVC_POOL_PERCPU, /* one pool per cpu */ 51 SVC_POOL_PERNODE /* one pool per numa node */ 52 }; 53 54 /* 55 * Structure for mapping cpus to pools and vice versa. 56 * Setup once during sunrpc initialisation. 57 */ 58 59 struct svc_pool_map { 60 int count; /* How many svc_servs use us */ 61 int mode; /* Note: int not enum to avoid 62 * warnings about "enumeration value 63 * not handled in switch" */ 64 unsigned int npools; 65 unsigned int *pool_to; /* maps pool id to cpu or node */ 66 unsigned int *to_pool; /* maps cpu or node to pool id */ 67 }; 68 69 static struct svc_pool_map svc_pool_map = { 70 .mode = SVC_POOL_DEFAULT 71 }; 72 73 static DEFINE_MUTEX(svc_pool_map_mutex);/* protects svc_pool_map.count only */ 74 75 static int 76 __param_set_pool_mode(const char *val, struct svc_pool_map *m) 77 { 78 int err, mode; 79 80 mutex_lock(&svc_pool_map_mutex); 81 82 err = 0; 83 if (!strncmp(val, "auto", 4)) 84 mode = SVC_POOL_AUTO; 85 else if (!strncmp(val, "global", 6)) 86 mode = SVC_POOL_GLOBAL; 87 else if (!strncmp(val, "percpu", 6)) 88 mode = SVC_POOL_PERCPU; 89 else if (!strncmp(val, "pernode", 7)) 90 mode = SVC_POOL_PERNODE; 91 else 92 err = -EINVAL; 93 94 if (err) 95 goto out; 96 97 if (m->count == 0) 98 m->mode = mode; 99 else if (mode != m->mode) 100 err = -EBUSY; 101 out: 102 mutex_unlock(&svc_pool_map_mutex); 103 return err; 104 } 105 106 static int 107 param_set_pool_mode(const char *val, const struct kernel_param *kp) 108 { 109 struct svc_pool_map *m = kp->arg; 110 111 return __param_set_pool_mode(val, m); 112 } 113 114 int sunrpc_set_pool_mode(const char *val) 115 { 116 return __param_set_pool_mode(val, &svc_pool_map); 117 } 118 EXPORT_SYMBOL(sunrpc_set_pool_mode); 119 120 /** 121 * sunrpc_get_pool_mode - get the current pool_mode for the host 122 * @buf: where to write the current pool_mode 123 * @size: size of @buf 124 * 125 * Grab the current pool_mode from the svc_pool_map and write 126 * the resulting string to @buf. Returns the number of characters 127 * written to @buf (a'la snprintf()). 128 */ 129 int 130 sunrpc_get_pool_mode(char *buf, size_t size) 131 { 132 struct svc_pool_map *m = &svc_pool_map; 133 134 switch (m->mode) 135 { 136 case SVC_POOL_AUTO: 137 return snprintf(buf, size, "auto"); 138 case SVC_POOL_GLOBAL: 139 return snprintf(buf, size, "global"); 140 case SVC_POOL_PERCPU: 141 return snprintf(buf, size, "percpu"); 142 case SVC_POOL_PERNODE: 143 return snprintf(buf, size, "pernode"); 144 default: 145 return snprintf(buf, size, "%d", m->mode); 146 } 147 } 148 EXPORT_SYMBOL(sunrpc_get_pool_mode); 149 150 static int 151 param_get_pool_mode(char *buf, const struct kernel_param *kp) 152 { 153 char str[16]; 154 int len; 155 156 len = sunrpc_get_pool_mode(str, ARRAY_SIZE(str)); 157 158 /* Ensure we have room for newline and NUL */ 159 len = min_t(int, len, ARRAY_SIZE(str) - 2); 160 161 /* tack on the newline */ 162 str[len] = '\n'; 163 str[len + 1] = '\0'; 164 165 return sysfs_emit(buf, "%s", str); 166 } 167 168 module_param_call(pool_mode, param_set_pool_mode, param_get_pool_mode, 169 &svc_pool_map, 0644); 170 171 /* 172 * Detect best pool mapping mode heuristically, 173 * according to the machine's topology. 174 */ 175 static int 176 svc_pool_map_choose_mode(void) 177 { 178 unsigned int node; 179 180 if (nr_online_nodes > 1) { 181 /* 182 * Actually have multiple NUMA nodes, 183 * so split pools on NUMA node boundaries 184 */ 185 return SVC_POOL_PERNODE; 186 } 187 188 node = first_online_node; 189 if (nr_cpus_node(node) > 2) { 190 /* 191 * Non-trivial SMP, or CONFIG_NUMA on 192 * non-NUMA hardware, e.g. with a generic 193 * x86_64 kernel on Xeons. In this case we 194 * want to divide the pools on cpu boundaries. 195 */ 196 return SVC_POOL_PERCPU; 197 } 198 199 /* default: one global pool */ 200 return SVC_POOL_GLOBAL; 201 } 202 203 /* 204 * Allocate the to_pool[] and pool_to[] arrays. 205 * Returns 0 on success or an errno. 206 */ 207 static int 208 svc_pool_map_alloc_arrays(struct svc_pool_map *m, unsigned int maxpools) 209 { 210 m->to_pool = kcalloc(maxpools, sizeof(unsigned int), GFP_KERNEL); 211 if (!m->to_pool) 212 goto fail; 213 m->pool_to = kcalloc(maxpools, sizeof(unsigned int), GFP_KERNEL); 214 if (!m->pool_to) 215 goto fail_free; 216 217 return 0; 218 219 fail_free: 220 kfree(m->to_pool); 221 m->to_pool = NULL; 222 fail: 223 return -ENOMEM; 224 } 225 226 /* 227 * Initialise the pool map for SVC_POOL_PERCPU mode. 228 * Returns number of pools or <0 on error. 229 */ 230 static int 231 svc_pool_map_init_percpu(struct svc_pool_map *m) 232 { 233 unsigned int maxpools = nr_cpu_ids; 234 unsigned int pidx = 0; 235 unsigned int cpu; 236 int err; 237 238 err = svc_pool_map_alloc_arrays(m, maxpools); 239 if (err) 240 return err; 241 242 for_each_online_cpu(cpu) { 243 BUG_ON(pidx >= maxpools); 244 m->to_pool[cpu] = pidx; 245 m->pool_to[pidx] = cpu; 246 pidx++; 247 } 248 /* cpus brought online later all get mapped to pool0, sorry */ 249 250 return pidx; 251 }; 252 253 254 /* 255 * Initialise the pool map for SVC_POOL_PERNODE mode. 256 * Returns number of pools or <0 on error. 257 */ 258 static int 259 svc_pool_map_init_pernode(struct svc_pool_map *m) 260 { 261 unsigned int maxpools = nr_node_ids; 262 unsigned int pidx = 0; 263 unsigned int node; 264 int err; 265 266 err = svc_pool_map_alloc_arrays(m, maxpools); 267 if (err) 268 return err; 269 270 for_each_node_with_cpus(node) { 271 /* some architectures (e.g. SN2) have cpuless nodes */ 272 BUG_ON(pidx > maxpools); 273 m->to_pool[node] = pidx; 274 m->pool_to[pidx] = node; 275 pidx++; 276 } 277 /* nodes brought online later all get mapped to pool0, sorry */ 278 279 return pidx; 280 } 281 282 283 /* 284 * Add a reference to the global map of cpus to pools (and 285 * vice versa) if pools are in use. 286 * Initialise the map if we're the first user. 287 * Returns the number of pools. If this is '1', no reference 288 * was taken. 289 */ 290 static unsigned int 291 svc_pool_map_get(void) 292 { 293 struct svc_pool_map *m = &svc_pool_map; 294 int npools = -1; 295 296 mutex_lock(&svc_pool_map_mutex); 297 if (m->count++) { 298 mutex_unlock(&svc_pool_map_mutex); 299 return m->npools; 300 } 301 302 if (m->mode == SVC_POOL_AUTO) 303 m->mode = svc_pool_map_choose_mode(); 304 305 switch (m->mode) { 306 case SVC_POOL_PERCPU: 307 npools = svc_pool_map_init_percpu(m); 308 break; 309 case SVC_POOL_PERNODE: 310 npools = svc_pool_map_init_pernode(m); 311 break; 312 } 313 314 if (npools <= 0) { 315 /* default, or memory allocation failure */ 316 npools = 1; 317 m->mode = SVC_POOL_GLOBAL; 318 } 319 m->npools = npools; 320 mutex_unlock(&svc_pool_map_mutex); 321 return npools; 322 } 323 324 /* 325 * Drop a reference to the global map of cpus to pools. 326 * When the last reference is dropped, the map data is 327 * freed; this allows the sysadmin to change the pool. 328 */ 329 static void 330 svc_pool_map_put(void) 331 { 332 struct svc_pool_map *m = &svc_pool_map; 333 334 mutex_lock(&svc_pool_map_mutex); 335 if (!--m->count) { 336 kfree(m->to_pool); 337 m->to_pool = NULL; 338 kfree(m->pool_to); 339 m->pool_to = NULL; 340 m->npools = 0; 341 } 342 mutex_unlock(&svc_pool_map_mutex); 343 } 344 345 static int svc_pool_map_get_node(unsigned int pidx) 346 { 347 const struct svc_pool_map *m = &svc_pool_map; 348 349 if (m->count) { 350 if (m->mode == SVC_POOL_PERCPU) 351 return cpu_to_node(m->pool_to[pidx]); 352 if (m->mode == SVC_POOL_PERNODE) 353 return m->pool_to[pidx]; 354 } 355 return numa_mem_id(); 356 } 357 /* 358 * Set the given thread's cpus_allowed mask so that it 359 * will only run on cpus in the given pool. 360 */ 361 static inline void 362 svc_pool_map_set_cpumask(struct task_struct *task, unsigned int pidx) 363 { 364 struct svc_pool_map *m = &svc_pool_map; 365 unsigned int node = m->pool_to[pidx]; 366 367 /* 368 * The caller checks for sv_nrpools > 1, which 369 * implies that we've been initialized. 370 */ 371 WARN_ON_ONCE(m->count == 0); 372 if (m->count == 0) 373 return; 374 375 switch (m->mode) { 376 case SVC_POOL_PERCPU: 377 { 378 set_cpus_allowed_ptr(task, cpumask_of(node)); 379 break; 380 } 381 case SVC_POOL_PERNODE: 382 { 383 set_cpus_allowed_ptr(task, cpumask_of_node(node)); 384 break; 385 } 386 } 387 } 388 389 /** 390 * svc_pool_for_cpu - Select pool to run a thread on this cpu 391 * @serv: An RPC service 392 * 393 * Use the active CPU and the svc_pool_map's mode setting to 394 * select the svc thread pool to use. Once initialized, the 395 * svc_pool_map does not change. 396 * 397 * Return value: 398 * A pointer to an svc_pool 399 */ 400 struct svc_pool *svc_pool_for_cpu(struct svc_serv *serv) 401 { 402 struct svc_pool_map *m = &svc_pool_map; 403 int cpu = raw_smp_processor_id(); 404 unsigned int pidx = 0; 405 406 if (serv->sv_nrpools <= 1) 407 return serv->sv_pools; 408 409 switch (m->mode) { 410 case SVC_POOL_PERCPU: 411 pidx = m->to_pool[cpu]; 412 break; 413 case SVC_POOL_PERNODE: 414 pidx = m->to_pool[cpu_to_node(cpu)]; 415 break; 416 } 417 418 return &serv->sv_pools[pidx % serv->sv_nrpools]; 419 } 420 421 static int svc_rpcb_setup(struct svc_serv *serv, struct net *net) 422 { 423 int err; 424 425 err = rpcb_create_local(net); 426 if (err) 427 return err; 428 429 /* Remove any stale portmap registrations */ 430 svc_unregister(serv, net); 431 return 0; 432 } 433 434 void svc_rpcb_cleanup(struct svc_serv *serv, struct net *net) 435 { 436 svc_unregister(serv, net); 437 rpcb_put_local(net); 438 } 439 440 static int svc_uses_rpcbind(struct svc_serv *serv) 441 { 442 unsigned int p, i; 443 444 for (p = 0; p < serv->sv_nprogs; p++) { 445 struct svc_program *progp = &serv->sv_programs[p]; 446 447 for (i = 0; i < progp->pg_nvers; i++) { 448 if (progp->pg_vers[i] == NULL) 449 continue; 450 if (!progp->pg_vers[i]->vs_hidden) 451 return 1; 452 } 453 } 454 455 return 0; 456 } 457 458 int svc_bind(struct svc_serv *serv, struct net *net) 459 { 460 if (!svc_uses_rpcbind(serv)) 461 return 0; 462 return svc_rpcb_setup(serv, net); 463 } 464 EXPORT_SYMBOL_GPL(svc_bind); 465 466 #if defined(CONFIG_SUNRPC_BACKCHANNEL) 467 static void 468 __svc_init_bc(struct svc_serv *serv) 469 { 470 lwq_init(&serv->sv_cb_list); 471 } 472 #else 473 static void 474 __svc_init_bc(struct svc_serv *serv) 475 { 476 } 477 #endif 478 479 /* 480 * Create an RPC service 481 */ 482 static struct svc_serv * 483 __svc_create(struct svc_program *prog, int nprogs, struct svc_stat *stats, 484 unsigned int bufsize, int npools, int (*threadfn)(void *data)) 485 { 486 struct svc_serv *serv; 487 unsigned int vers; 488 unsigned int xdrsize; 489 unsigned int i; 490 491 if (!(serv = kzalloc_obj(*serv))) 492 return NULL; 493 serv->sv_name = prog->pg_name; 494 serv->sv_programs = prog; 495 serv->sv_nprogs = nprogs; 496 serv->sv_stats = stats; 497 if (bufsize > RPCSVC_MAXPAYLOAD) 498 bufsize = RPCSVC_MAXPAYLOAD; 499 serv->sv_max_payload = bufsize? bufsize : 4096; 500 serv->sv_max_mesg = roundup(serv->sv_max_payload + PAGE_SIZE, PAGE_SIZE); 501 serv->sv_threadfn = threadfn; 502 xdrsize = 0; 503 for (i = 0; i < nprogs; i++) { 504 struct svc_program *progp = &prog[i]; 505 506 progp->pg_lovers = progp->pg_nvers-1; 507 for (vers = 0; vers < progp->pg_nvers ; vers++) 508 if (progp->pg_vers[vers]) { 509 progp->pg_hivers = vers; 510 if (progp->pg_lovers > vers) 511 progp->pg_lovers = vers; 512 if (progp->pg_vers[vers]->vs_xdrsize > xdrsize) 513 xdrsize = progp->pg_vers[vers]->vs_xdrsize; 514 } 515 } 516 serv->sv_xdrsize = xdrsize; 517 INIT_LIST_HEAD(&serv->sv_tempsocks); 518 INIT_LIST_HEAD(&serv->sv_permsocks); 519 timer_setup(&serv->sv_temptimer, NULL, 0); 520 spin_lock_init(&serv->sv_lock); 521 522 __svc_init_bc(serv); 523 524 serv->sv_nrpools = npools; 525 serv->sv_pools = 526 kzalloc_objs(struct svc_pool, serv->sv_nrpools); 527 if (!serv->sv_pools) { 528 kfree(serv); 529 return NULL; 530 } 531 532 for (i = 0; i < serv->sv_nrpools; i++) { 533 struct svc_pool *pool = &serv->sv_pools[i]; 534 535 dprintk("svc: initialising pool %u for %s\n", 536 i, serv->sv_name); 537 538 pool->sp_id = i; 539 lwq_init(&pool->sp_xprts); 540 INIT_LIST_HEAD(&pool->sp_all_threads); 541 init_llist_head(&pool->sp_idle_threads); 542 543 percpu_counter_init(&pool->sp_messages_arrived, 0, GFP_KERNEL); 544 percpu_counter_init(&pool->sp_sockets_queued, 0, GFP_KERNEL); 545 percpu_counter_init(&pool->sp_threads_woken, 0, GFP_KERNEL); 546 } 547 548 return serv; 549 } 550 551 /** 552 * svc_create - Create an RPC service 553 * @prog: the RPC program the new service will handle 554 * @bufsize: maximum message size for @prog 555 * @threadfn: a function to service RPC requests for @prog 556 * 557 * Returns an instantiated struct svc_serv object or NULL. 558 */ 559 struct svc_serv *svc_create(struct svc_program *prog, unsigned int bufsize, 560 int (*threadfn)(void *data)) 561 { 562 return __svc_create(prog, 1, NULL, bufsize, 1, threadfn); 563 } 564 EXPORT_SYMBOL_GPL(svc_create); 565 566 /** 567 * svc_create_pooled - Create an RPC service with pooled threads 568 * @prog: Array of RPC programs the new service will handle 569 * @nprogs: Number of programs in the array 570 * @stats: the stats struct if desired 571 * @bufsize: maximum message size for @prog 572 * @threadfn: a function to service RPC requests for @prog 573 * 574 * Returns an instantiated struct svc_serv object or NULL. 575 */ 576 struct svc_serv *svc_create_pooled(struct svc_program *prog, 577 unsigned int nprogs, 578 struct svc_stat *stats, 579 unsigned int bufsize, 580 int (*threadfn)(void *data)) 581 { 582 struct svc_serv *serv; 583 unsigned int npools = svc_pool_map_get(); 584 585 serv = __svc_create(prog, nprogs, stats, bufsize, npools, threadfn); 586 if (!serv) 587 goto out_err; 588 serv->sv_is_pooled = true; 589 return serv; 590 out_err: 591 svc_pool_map_put(); 592 return NULL; 593 } 594 EXPORT_SYMBOL_GPL(svc_create_pooled); 595 596 /* 597 * Destroy an RPC service. Should be called with appropriate locking to 598 * protect sv_permsocks and sv_tempsocks. 599 */ 600 void 601 svc_destroy(struct svc_serv **servp) 602 { 603 struct svc_serv *serv = *servp; 604 unsigned int i; 605 606 *servp = NULL; 607 608 dprintk("svc: svc_destroy(%s)\n", serv->sv_programs->pg_name); 609 timer_shutdown_sync(&serv->sv_temptimer); 610 611 /* 612 * Remaining transports at this point are not expected. 613 */ 614 WARN_ONCE(!list_empty(&serv->sv_permsocks), 615 "SVC: permsocks remain for %s\n", serv->sv_programs->pg_name); 616 WARN_ONCE(!list_empty(&serv->sv_tempsocks), 617 "SVC: tempsocks remain for %s\n", serv->sv_programs->pg_name); 618 619 cache_clean_deferred(serv); 620 621 if (serv->sv_is_pooled) 622 svc_pool_map_put(); 623 624 for (i = 0; i < serv->sv_nrpools; i++) { 625 struct svc_pool *pool = &serv->sv_pools[i]; 626 627 percpu_counter_destroy(&pool->sp_messages_arrived); 628 percpu_counter_destroy(&pool->sp_sockets_queued); 629 percpu_counter_destroy(&pool->sp_threads_woken); 630 } 631 kfree(serv->sv_pools); 632 kfree(serv); 633 } 634 EXPORT_SYMBOL_GPL(svc_destroy); 635 636 static bool 637 svc_init_buffer(struct svc_rqst *rqstp, const struct svc_serv *serv, int node) 638 { 639 rqstp->rq_maxpages = svc_serv_maxpages(serv); 640 641 /* rq_pages' last entry is NULL for historical reasons. */ 642 rqstp->rq_pages = kcalloc_node(rqstp->rq_maxpages + 1, 643 sizeof(struct page *), 644 GFP_KERNEL, node); 645 if (!rqstp->rq_pages) 646 return false; 647 648 return true; 649 } 650 651 /* 652 * Release an RPC server buffer 653 */ 654 static void 655 svc_release_buffer(struct svc_rqst *rqstp) 656 { 657 unsigned long i; 658 659 for (i = 0; i < rqstp->rq_maxpages; i++) 660 if (rqstp->rq_pages[i]) 661 put_page(rqstp->rq_pages[i]); 662 kfree(rqstp->rq_pages); 663 } 664 665 static void 666 svc_rqst_free(struct svc_rqst *rqstp) 667 { 668 folio_batch_release(&rqstp->rq_fbatch); 669 kfree(rqstp->rq_bvec); 670 svc_release_buffer(rqstp); 671 if (rqstp->rq_scratch_folio) 672 folio_put(rqstp->rq_scratch_folio); 673 kfree(rqstp->rq_resp); 674 kfree(rqstp->rq_argp); 675 kfree(rqstp->rq_auth_data); 676 kfree_rcu(rqstp, rq_rcu_head); 677 } 678 679 static struct svc_rqst * 680 svc_prepare_thread(struct svc_serv *serv, struct svc_pool *pool, int node) 681 { 682 struct svc_rqst *rqstp; 683 684 rqstp = kzalloc_node(sizeof(*rqstp), GFP_KERNEL, node); 685 if (!rqstp) 686 return rqstp; 687 688 folio_batch_init(&rqstp->rq_fbatch); 689 690 rqstp->rq_server = serv; 691 rqstp->rq_pool = pool; 692 693 rqstp->rq_scratch_folio = __folio_alloc_node(GFP_KERNEL, 0, node); 694 if (!rqstp->rq_scratch_folio) 695 goto out_enomem; 696 697 rqstp->rq_argp = kmalloc_node(serv->sv_xdrsize, GFP_KERNEL, node); 698 if (!rqstp->rq_argp) 699 goto out_enomem; 700 701 rqstp->rq_resp = kmalloc_node(serv->sv_xdrsize, GFP_KERNEL, node); 702 if (!rqstp->rq_resp) 703 goto out_enomem; 704 705 if (!svc_init_buffer(rqstp, serv, node)) 706 goto out_enomem; 707 708 rqstp->rq_bvec = kcalloc_node(rqstp->rq_maxpages, 709 sizeof(struct bio_vec), 710 GFP_KERNEL, node); 711 if (!rqstp->rq_bvec) 712 goto out_enomem; 713 714 rqstp->rq_err = -EAGAIN; /* No error yet */ 715 716 serv->sv_nrthreads += 1; 717 pool->sp_nrthreads += 1; 718 719 /* Protected by whatever lock the service uses when calling 720 * svc_set_num_threads() 721 */ 722 list_add_rcu(&rqstp->rq_all, &pool->sp_all_threads); 723 724 return rqstp; 725 726 out_enomem: 727 svc_rqst_free(rqstp); 728 return NULL; 729 } 730 731 /** 732 * svc_pool_wake_idle_thread - Awaken an idle thread in @pool 733 * @pool: service thread pool 734 * 735 * Can be called from soft IRQ or process context. Finding an idle 736 * service thread and marking it BUSY is atomic with respect to 737 * other calls to svc_pool_wake_idle_thread(). 738 * 739 */ 740 void svc_pool_wake_idle_thread(struct svc_pool *pool) 741 { 742 struct svc_rqst *rqstp; 743 struct llist_node *ln; 744 745 rcu_read_lock(); 746 ln = READ_ONCE(pool->sp_idle_threads.first); 747 if (ln) { 748 rqstp = llist_entry(ln, struct svc_rqst, rq_idle); 749 WRITE_ONCE(rqstp->rq_qtime, ktime_get()); 750 if (!task_is_running(rqstp->rq_task)) { 751 wake_up_process(rqstp->rq_task); 752 trace_svc_pool_thread_wake(pool, rqstp->rq_task->pid); 753 percpu_counter_inc(&pool->sp_threads_woken); 754 } else { 755 trace_svc_pool_thread_running(pool, rqstp->rq_task->pid); 756 } 757 rcu_read_unlock(); 758 return; 759 } 760 rcu_read_unlock(); 761 trace_svc_pool_thread_noidle(pool, 0); 762 } 763 EXPORT_SYMBOL_GPL(svc_pool_wake_idle_thread); 764 765 /** 766 * svc_new_thread - spawn a new thread in the given pool 767 * @serv: the serv to which the pool belongs 768 * @pool: pool in which thread should be spawned 769 * 770 * Create a new thread inside @pool, which is a part of @serv. 771 * Caller must hold the service mutex. 772 * 773 * Returns 0 on success, or -errno on failure. 774 */ 775 int svc_new_thread(struct svc_serv *serv, struct svc_pool *pool) 776 { 777 struct svc_rqst *rqstp; 778 struct task_struct *task; 779 int node; 780 int err = 0; 781 782 node = svc_pool_map_get_node(pool->sp_id); 783 784 rqstp = svc_prepare_thread(serv, pool, node); 785 if (!rqstp) 786 return -ENOMEM; 787 task = kthread_create_on_node(serv->sv_threadfn, rqstp, 788 node, "%s", serv->sv_name); 789 if (IS_ERR(task)) { 790 err = PTR_ERR(task); 791 goto out; 792 } 793 794 rqstp->rq_task = task; 795 if (serv->sv_nrpools > 1) 796 svc_pool_map_set_cpumask(task, pool->sp_id); 797 798 svc_sock_update_bufs(serv); 799 wake_up_process(task); 800 801 /* Wait for the thread to signal initialization status */ 802 wait_var_event(&rqstp->rq_err, rqstp->rq_err != -EAGAIN); 803 err = rqstp->rq_err; 804 out: 805 if (err) 806 svc_exit_thread(rqstp); 807 return err; 808 } 809 EXPORT_SYMBOL_GPL(svc_new_thread); 810 811 static int 812 svc_start_kthreads(struct svc_serv *serv, struct svc_pool *pool, int nrservs) 813 { 814 int err = 0; 815 816 while (!err && nrservs--) 817 err = svc_new_thread(serv, pool); 818 819 return err; 820 } 821 822 static int 823 svc_stop_kthreads(struct svc_serv *serv, struct svc_pool *pool, int nrservs) 824 { 825 do { 826 set_bit(SP_VICTIM_REMAINS, &pool->sp_flags); 827 set_bit(SP_NEED_VICTIM, &pool->sp_flags); 828 svc_pool_wake_idle_thread(pool); 829 wait_on_bit(&pool->sp_flags, SP_VICTIM_REMAINS, TASK_IDLE); 830 nrservs++; 831 } while (nrservs < 0); 832 return 0; 833 } 834 835 /** 836 * svc_set_pool_threads - adjust number of threads per pool 837 * @serv: RPC service to adjust 838 * @pool: Specific pool from which to choose threads 839 * @min_threads: min number of threads to run in @pool 840 * @max_threads: max number of threads in @pool (0 means kill all threads) 841 * 842 * Create or destroy threads in @pool to bring it into an acceptable range 843 * between @min_threads and @max_threads. 844 * 845 * If @min_threads is 0 or larger than @max_threads, then it is ignored and 846 * the pool will be set to run a static @max_threads number of threads. 847 * 848 * Caller must ensure mutual exclusion between this and server startup or 849 * shutdown. 850 * 851 * Returns zero on success or a negative errno if an error occurred while 852 * starting a thread. 853 */ 854 int 855 svc_set_pool_threads(struct svc_serv *serv, struct svc_pool *pool, 856 unsigned int min_threads, unsigned int max_threads) 857 { 858 int delta; 859 860 if (!pool) 861 return -EINVAL; 862 863 /* clamp min threads to the max */ 864 if (min_threads > max_threads) 865 min_threads = max_threads; 866 867 pool->sp_nrthrmin = min_threads; 868 pool->sp_nrthrmax = max_threads; 869 870 /* 871 * When min_threads is set, then only change the number of 872 * threads to bring it within an acceptable range. 873 */ 874 if (min_threads) { 875 if (pool->sp_nrthreads > max_threads) 876 delta = max_threads; 877 else if (pool->sp_nrthreads < min_threads) 878 delta = min_threads; 879 else 880 return 0; 881 } else { 882 delta = max_threads; 883 } 884 885 delta -= pool->sp_nrthreads; 886 if (delta > 0) 887 return svc_start_kthreads(serv, pool, delta); 888 if (delta < 0) 889 return svc_stop_kthreads(serv, pool, delta); 890 return 0; 891 } 892 EXPORT_SYMBOL_GPL(svc_set_pool_threads); 893 894 /** 895 * svc_set_num_threads - adjust number of threads in serv 896 * @serv: RPC service to adjust 897 * @min_threads: min number of threads to run per pool 898 * @nrservs: New number of threads for @serv (0 means kill all threads) 899 * 900 * Create or destroy threads in @serv to bring it to @nrservs. If there 901 * are multiple pools then the new threads or victims will be distributed 902 * evenly among them. 903 * 904 * Caller must ensure mutual exclusion between this and server startup or 905 * shutdown. 906 * 907 * Returns zero on success or a negative errno if an error occurred while 908 * starting a thread. On failure, some pools may have already been 909 * adjusted; the caller is responsible for recovery. 910 */ 911 int 912 svc_set_num_threads(struct svc_serv *serv, unsigned int min_threads, 913 unsigned int nrservs) 914 { 915 unsigned int base = nrservs / serv->sv_nrpools; 916 unsigned int remain = nrservs % serv->sv_nrpools; 917 int i, err = 0; 918 919 for (i = 0; i < serv->sv_nrpools; ++i) { 920 struct svc_pool *pool = &serv->sv_pools[i]; 921 int threads = base; 922 923 if (remain) { 924 ++threads; 925 --remain; 926 } 927 928 err = svc_set_pool_threads(serv, pool, min_threads, threads); 929 if (err) 930 break; 931 } 932 return err; 933 } 934 EXPORT_SYMBOL_GPL(svc_set_num_threads); 935 936 /** 937 * svc_rqst_replace_page - Replace one page in rq_pages[] 938 * @rqstp: svc_rqst with pages to replace 939 * @page: replacement page 940 * 941 * When replacing a page in rq_pages, batch the release of the 942 * replaced pages to avoid hammering the page allocator. 943 * 944 * Return values: 945 * %true: page replaced 946 * %false: array bounds checking failed 947 */ 948 bool svc_rqst_replace_page(struct svc_rqst *rqstp, struct page *page) 949 { 950 struct page **begin = rqstp->rq_pages; 951 struct page **end = &rqstp->rq_pages[rqstp->rq_maxpages]; 952 953 if (unlikely(rqstp->rq_next_page < begin || rqstp->rq_next_page > end)) { 954 trace_svc_replace_page_err(rqstp); 955 return false; 956 } 957 958 if (*rqstp->rq_next_page) { 959 if (!folio_batch_add(&rqstp->rq_fbatch, 960 page_folio(*rqstp->rq_next_page))) 961 __folio_batch_release(&rqstp->rq_fbatch); 962 } 963 964 get_page(page); 965 *(rqstp->rq_next_page++) = page; 966 return true; 967 } 968 EXPORT_SYMBOL_GPL(svc_rqst_replace_page); 969 970 /** 971 * svc_rqst_release_pages - Release Reply buffer pages 972 * @rqstp: RPC transaction context 973 * 974 * Release response pages that might still be in flight after 975 * svc_send, and any spliced filesystem-owned pages. 976 */ 977 void svc_rqst_release_pages(struct svc_rqst *rqstp) 978 { 979 int i, count = rqstp->rq_next_page - rqstp->rq_respages; 980 981 if (count) { 982 release_pages(rqstp->rq_respages, count); 983 for (i = 0; i < count; i++) 984 rqstp->rq_respages[i] = NULL; 985 } 986 } 987 988 /** 989 * svc_exit_thread - finalise the termination of a sunrpc server thread 990 * @rqstp: the svc_rqst which represents the thread. 991 * 992 * When a thread started with svc_new_thread() exits it must call 993 * svc_exit_thread() as its last act. This must be done with the 994 * service mutex held. Normally this is held by a DIFFERENT thread, the 995 * one that is calling svc_set_num_threads() and which will wait for 996 * SP_VICTIM_REMAINS to be cleared before dropping the mutex. If the 997 * thread exits for any reason other than svc_thread_should_stop() 998 * returning %true (which indicated that svc_set_num_threads() is 999 * waiting for it to exit), then it must take the service mutex itself, 1000 * which can only safely be done using mutex_try_lock(). 1001 */ 1002 void 1003 svc_exit_thread(struct svc_rqst *rqstp) 1004 { 1005 struct svc_serv *serv = rqstp->rq_server; 1006 struct svc_pool *pool = rqstp->rq_pool; 1007 1008 list_del_rcu(&rqstp->rq_all); 1009 1010 pool->sp_nrthreads -= 1; 1011 serv->sv_nrthreads -= 1; 1012 svc_sock_update_bufs(serv); 1013 1014 svc_rqst_free(rqstp); 1015 1016 clear_and_wake_up_bit(SP_VICTIM_REMAINS, &pool->sp_flags); 1017 } 1018 EXPORT_SYMBOL_GPL(svc_exit_thread); 1019 1020 /* 1021 * Register an "inet" protocol family netid with the local 1022 * rpcbind daemon via an rpcbind v4 SET request. 1023 * 1024 * No netconfig infrastructure is available in the kernel, so 1025 * we map IP_ protocol numbers to netids by hand. 1026 * 1027 * Returns zero on success; a negative errno value is returned 1028 * if any error occurs. 1029 */ 1030 static int __svc_rpcb_register4(struct net *net, const u32 program, 1031 const u32 version, 1032 const unsigned short protocol, 1033 const unsigned short port) 1034 { 1035 const struct sockaddr_in sin = { 1036 .sin_family = AF_INET, 1037 .sin_addr.s_addr = htonl(INADDR_ANY), 1038 .sin_port = htons(port), 1039 }; 1040 const char *netid; 1041 int error; 1042 1043 switch (protocol) { 1044 case IPPROTO_UDP: 1045 netid = RPCBIND_NETID_UDP; 1046 break; 1047 case IPPROTO_TCP: 1048 netid = RPCBIND_NETID_TCP; 1049 break; 1050 default: 1051 return -ENOPROTOOPT; 1052 } 1053 1054 error = rpcb_v4_register(net, program, version, 1055 (const struct sockaddr *)&sin, netid); 1056 1057 /* 1058 * User space didn't support rpcbind v4, so retry this 1059 * registration request with the legacy rpcbind v2 protocol. 1060 */ 1061 if (error == -EPROTONOSUPPORT) 1062 error = rpcb_register(net, program, version, protocol, port); 1063 1064 return error; 1065 } 1066 1067 #if IS_ENABLED(CONFIG_IPV6) 1068 /* 1069 * Register an "inet6" protocol family netid with the local 1070 * rpcbind daemon via an rpcbind v4 SET request. 1071 * 1072 * No netconfig infrastructure is available in the kernel, so 1073 * we map IP_ protocol numbers to netids by hand. 1074 * 1075 * Returns zero on success; a negative errno value is returned 1076 * if any error occurs. 1077 */ 1078 static int __svc_rpcb_register6(struct net *net, const u32 program, 1079 const u32 version, 1080 const unsigned short protocol, 1081 const unsigned short port) 1082 { 1083 const struct sockaddr_in6 sin6 = { 1084 .sin6_family = AF_INET6, 1085 .sin6_addr = IN6ADDR_ANY_INIT, 1086 .sin6_port = htons(port), 1087 }; 1088 const char *netid; 1089 int error; 1090 1091 switch (protocol) { 1092 case IPPROTO_UDP: 1093 netid = RPCBIND_NETID_UDP6; 1094 break; 1095 case IPPROTO_TCP: 1096 netid = RPCBIND_NETID_TCP6; 1097 break; 1098 default: 1099 return -ENOPROTOOPT; 1100 } 1101 1102 error = rpcb_v4_register(net, program, version, 1103 (const struct sockaddr *)&sin6, netid); 1104 1105 /* 1106 * User space didn't support rpcbind version 4, so we won't 1107 * use a PF_INET6 listener. 1108 */ 1109 if (error == -EPROTONOSUPPORT) 1110 error = -EAFNOSUPPORT; 1111 1112 return error; 1113 } 1114 #endif /* IS_ENABLED(CONFIG_IPV6) */ 1115 1116 /* 1117 * Register a kernel RPC service via rpcbind version 4. 1118 * 1119 * Returns zero on success; a negative errno value is returned 1120 * if any error occurs. 1121 */ 1122 static int __svc_register(struct net *net, const char *progname, 1123 const u32 program, const u32 version, 1124 const int family, 1125 const unsigned short protocol, 1126 const unsigned short port) 1127 { 1128 int error = -EAFNOSUPPORT; 1129 1130 switch (family) { 1131 case PF_INET: 1132 error = __svc_rpcb_register4(net, program, version, 1133 protocol, port); 1134 break; 1135 #if IS_ENABLED(CONFIG_IPV6) 1136 case PF_INET6: 1137 error = __svc_rpcb_register6(net, program, version, 1138 protocol, port); 1139 #endif 1140 } 1141 1142 trace_svc_register(progname, version, family, protocol, port, error); 1143 return error; 1144 } 1145 1146 static 1147 int svc_rpcbind_set_version(struct net *net, 1148 const struct svc_program *progp, 1149 u32 version, int family, 1150 unsigned short proto, 1151 unsigned short port) 1152 { 1153 return __svc_register(net, progp->pg_name, progp->pg_prog, 1154 version, family, proto, port); 1155 1156 } 1157 1158 int svc_generic_rpcbind_set(struct net *net, 1159 const struct svc_program *progp, 1160 u32 version, int family, 1161 unsigned short proto, 1162 unsigned short port) 1163 { 1164 const struct svc_version *vers = progp->pg_vers[version]; 1165 int error; 1166 1167 if (vers == NULL) 1168 return 0; 1169 1170 if (vers->vs_hidden) { 1171 trace_svc_noregister(progp->pg_name, version, proto, 1172 port, family, 0); 1173 return 0; 1174 } 1175 1176 /* 1177 * Don't register a UDP port if we need congestion 1178 * control. 1179 */ 1180 if (vers->vs_need_cong_ctrl && proto == IPPROTO_UDP) 1181 return 0; 1182 1183 error = svc_rpcbind_set_version(net, progp, version, 1184 family, proto, port); 1185 1186 return (vers->vs_rpcb_optnl) ? 0 : error; 1187 } 1188 EXPORT_SYMBOL_GPL(svc_generic_rpcbind_set); 1189 1190 /** 1191 * svc_register - register an RPC service with the local portmapper 1192 * @serv: svc_serv struct for the service to register 1193 * @net: net namespace for the service to register 1194 * @family: protocol family of service's listener socket 1195 * @proto: transport protocol number to advertise 1196 * @port: port to advertise 1197 * 1198 * Service is registered for any address in the passed-in protocol family 1199 */ 1200 int svc_register(const struct svc_serv *serv, struct net *net, 1201 const int family, const unsigned short proto, 1202 const unsigned short port) 1203 { 1204 unsigned int p, i; 1205 int error = 0; 1206 1207 WARN_ON_ONCE(proto == 0 && port == 0); 1208 if (proto == 0 && port == 0) 1209 return -EINVAL; 1210 1211 for (p = 0; p < serv->sv_nprogs; p++) { 1212 struct svc_program *progp = &serv->sv_programs[p]; 1213 1214 for (i = 0; i < progp->pg_nvers; i++) { 1215 1216 error = progp->pg_rpcbind_set(net, progp, i, 1217 family, proto, port); 1218 if (error < 0) { 1219 printk(KERN_WARNING "svc: failed to register " 1220 "%sv%u RPC service (errno %d).\n", 1221 progp->pg_name, i, -error); 1222 break; 1223 } 1224 } 1225 } 1226 1227 return error; 1228 } 1229 1230 /* 1231 * If user space is running rpcbind, it should take the v4 UNSET 1232 * and clear everything for this [program, version]. If user space 1233 * is running portmap, it will reject the v4 UNSET, but won't have 1234 * any "inet6" entries anyway. So a PMAP_UNSET should be sufficient 1235 * in this case to clear all existing entries for [program, version]. 1236 */ 1237 static void __svc_unregister(struct net *net, const u32 program, const u32 version, 1238 const char *progname) 1239 { 1240 int error; 1241 1242 error = rpcb_v4_register(net, program, version, NULL, ""); 1243 1244 /* 1245 * User space didn't support rpcbind v4, so retry this 1246 * request with the legacy rpcbind v2 protocol. 1247 */ 1248 if (error == -EPROTONOSUPPORT) 1249 error = rpcb_register(net, program, version, 0, 0); 1250 1251 trace_svc_unregister(progname, version, error); 1252 } 1253 1254 /* 1255 * All netids, bind addresses and ports registered for [program, version] 1256 * are removed from the local rpcbind database (if the service is not 1257 * hidden) to make way for a new instance of the service. 1258 * 1259 * The result of unregistration is reported via dprintk for those who want 1260 * verification of the result, but is otherwise not important. 1261 */ 1262 static void svc_unregister(const struct svc_serv *serv, struct net *net) 1263 { 1264 struct sighand_struct *sighand; 1265 unsigned long flags; 1266 unsigned int p, i; 1267 1268 clear_thread_flag(TIF_SIGPENDING); 1269 1270 for (p = 0; p < serv->sv_nprogs; p++) { 1271 struct svc_program *progp = &serv->sv_programs[p]; 1272 1273 for (i = 0; i < progp->pg_nvers; i++) { 1274 if (progp->pg_vers[i] == NULL) 1275 continue; 1276 if (progp->pg_vers[i]->vs_hidden) 1277 continue; 1278 __svc_unregister(net, progp->pg_prog, i, progp->pg_name); 1279 } 1280 } 1281 1282 rcu_read_lock(); 1283 sighand = rcu_dereference(current->sighand); 1284 spin_lock_irqsave(&sighand->siglock, flags); 1285 recalc_sigpending(); 1286 spin_unlock_irqrestore(&sighand->siglock, flags); 1287 rcu_read_unlock(); 1288 } 1289 1290 /* 1291 * dprintk the given error with the address of the client that caused it. 1292 */ 1293 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG) 1294 static __printf(2, 3) 1295 void svc_printk(struct svc_rqst *rqstp, const char *fmt, ...) 1296 { 1297 struct va_format vaf; 1298 va_list args; 1299 char buf[RPC_MAX_ADDRBUFLEN]; 1300 1301 va_start(args, fmt); 1302 1303 vaf.fmt = fmt; 1304 vaf.va = &args; 1305 1306 dprintk("svc: %s: %pV", svc_print_addr(rqstp, buf, sizeof(buf)), &vaf); 1307 1308 va_end(args); 1309 } 1310 #else 1311 static __printf(2,3) void svc_printk(struct svc_rqst *rqstp, const char *fmt, ...) {} 1312 #endif 1313 1314 __be32 1315 svc_generic_init_request(struct svc_rqst *rqstp, 1316 const struct svc_program *progp, 1317 struct svc_process_info *ret) 1318 { 1319 const struct svc_version *versp = NULL; /* compiler food */ 1320 const struct svc_procedure *procp = NULL; 1321 1322 if (rqstp->rq_vers >= progp->pg_nvers ) 1323 goto err_bad_vers; 1324 versp = progp->pg_vers[rqstp->rq_vers]; 1325 if (!versp) 1326 goto err_bad_vers; 1327 1328 /* 1329 * Some protocol versions (namely NFSv4) require some form of 1330 * congestion control. (See RFC 7530 section 3.1 paragraph 2) 1331 * In other words, UDP is not allowed. We mark those when setting 1332 * up the svc_xprt, and verify that here. 1333 * 1334 * The spec is not very clear about what error should be returned 1335 * when someone tries to access a server that is listening on UDP 1336 * for lower versions. RPC_PROG_MISMATCH seems to be the closest 1337 * fit. 1338 */ 1339 if (versp->vs_need_cong_ctrl && rqstp->rq_xprt && 1340 !test_bit(XPT_CONG_CTRL, &rqstp->rq_xprt->xpt_flags)) 1341 goto err_bad_vers; 1342 1343 if (rqstp->rq_proc >= versp->vs_nproc) 1344 goto err_bad_proc; 1345 rqstp->rq_procinfo = procp = &versp->vs_proc[rqstp->rq_proc]; 1346 1347 /* Initialize storage for argp and resp */ 1348 memset(rqstp->rq_argp, 0, procp->pc_argzero); 1349 memset(rqstp->rq_resp, 0, procp->pc_ressize); 1350 1351 /* Bump per-procedure stats counter */ 1352 this_cpu_inc(versp->vs_count[rqstp->rq_proc]); 1353 1354 ret->dispatch = versp->vs_dispatch; 1355 return rpc_success; 1356 err_bad_vers: 1357 ret->mismatch.lovers = progp->pg_lovers; 1358 ret->mismatch.hivers = progp->pg_hivers; 1359 return rpc_prog_mismatch; 1360 err_bad_proc: 1361 return rpc_proc_unavail; 1362 } 1363 EXPORT_SYMBOL_GPL(svc_generic_init_request); 1364 1365 /* 1366 * Common routine for processing the RPC request. 1367 */ 1368 static int 1369 svc_process_common(struct svc_rqst *rqstp) 1370 { 1371 struct xdr_stream *xdr = &rqstp->rq_res_stream; 1372 struct svc_program *progp = NULL; 1373 const struct svc_procedure *procp = NULL; 1374 struct svc_serv *serv = rqstp->rq_server; 1375 struct svc_process_info process; 1376 enum svc_auth_status auth_res; 1377 unsigned int aoffset; 1378 int pr, rc; 1379 __be32 *p; 1380 1381 /* Reset the accept_stat for the RPC */ 1382 rqstp->rq_accept_statp = NULL; 1383 1384 /* Will be turned off only when NFSv4 Sessions are used */ 1385 set_bit(RQ_USEDEFERRAL, &rqstp->rq_flags); 1386 clear_bit(RQ_DROPME, &rqstp->rq_flags); 1387 1388 /* Construct the first words of the reply: */ 1389 svcxdr_init_encode(rqstp); 1390 xdr_stream_encode_be32(xdr, rqstp->rq_xid); 1391 xdr_stream_encode_be32(xdr, rpc_reply); 1392 1393 p = xdr_inline_decode(&rqstp->rq_arg_stream, XDR_UNIT * 4); 1394 if (unlikely(!p)) 1395 goto err_short_len; 1396 if (*p++ != cpu_to_be32(RPC_VERSION)) 1397 goto err_bad_rpc; 1398 1399 xdr_stream_encode_be32(xdr, rpc_msg_accepted); 1400 1401 rqstp->rq_prog = be32_to_cpup(p++); 1402 rqstp->rq_vers = be32_to_cpup(p++); 1403 rqstp->rq_proc = be32_to_cpup(p); 1404 1405 for (pr = 0; pr < serv->sv_nprogs; pr++) 1406 if (rqstp->rq_prog == serv->sv_programs[pr].pg_prog) 1407 progp = &serv->sv_programs[pr]; 1408 1409 /* 1410 * Decode auth data, and add verifier to reply buffer. 1411 * We do this before anything else in order to get a decent 1412 * auth verifier. 1413 */ 1414 auth_res = svc_authenticate(rqstp); 1415 /* Also give the program a chance to reject this call: */ 1416 if (auth_res == SVC_OK && progp) 1417 auth_res = progp->pg_authenticate(rqstp); 1418 trace_svc_authenticate(rqstp, auth_res); 1419 switch (auth_res) { 1420 case SVC_OK: 1421 break; 1422 case SVC_GARBAGE: 1423 rqstp->rq_auth_stat = rpc_autherr_badcred; 1424 goto err_bad_auth; 1425 case SVC_DENIED: 1426 goto err_bad_auth; 1427 case SVC_CLOSE: 1428 goto close; 1429 case SVC_DROP: 1430 goto dropit; 1431 case SVC_COMPLETE: 1432 goto sendit; 1433 default: 1434 pr_warn_once("Unexpected svc_auth_status (%d)\n", auth_res); 1435 rqstp->rq_auth_stat = rpc_autherr_failed; 1436 goto err_bad_auth; 1437 } 1438 1439 if (progp == NULL) 1440 goto err_bad_prog; 1441 1442 switch (progp->pg_init_request(rqstp, progp, &process)) { 1443 case rpc_success: 1444 break; 1445 case rpc_prog_unavail: 1446 goto err_bad_prog; 1447 case rpc_prog_mismatch: 1448 goto err_bad_vers; 1449 case rpc_proc_unavail: 1450 goto err_bad_proc; 1451 } 1452 1453 procp = rqstp->rq_procinfo; 1454 /* Should this check go into the dispatcher? */ 1455 if (!procp || !procp->pc_func) 1456 goto err_bad_proc; 1457 1458 /* Syntactic check complete */ 1459 if (serv->sv_stats) 1460 serv->sv_stats->rpccnt++; 1461 trace_svc_process(rqstp, progp->pg_name); 1462 1463 aoffset = xdr_stream_pos(xdr); 1464 1465 /* un-reserve some of the out-queue now that we have a 1466 * better idea of reply size 1467 */ 1468 if (procp->pc_xdrressize) 1469 svc_reserve_auth(rqstp, procp->pc_xdrressize<<2); 1470 1471 /* Call the function that processes the request. */ 1472 rc = process.dispatch(rqstp); 1473 xdr_finish_decode(xdr); 1474 1475 if (!rc) 1476 goto dropit; 1477 if (rqstp->rq_auth_stat != rpc_auth_ok) 1478 goto err_bad_auth; 1479 1480 if (*rqstp->rq_accept_statp != rpc_success) 1481 xdr_truncate_encode(xdr, aoffset); 1482 1483 if (procp->pc_encode == NULL) 1484 goto dropit; 1485 1486 sendit: 1487 if (svc_authorise(rqstp)) 1488 goto close_xprt; 1489 return 1; /* Caller can now send it */ 1490 1491 dropit: 1492 svc_authorise(rqstp); /* doesn't hurt to call this twice */ 1493 dprintk("svc: svc_process dropit\n"); 1494 return 0; 1495 1496 close: 1497 svc_authorise(rqstp); 1498 close_xprt: 1499 if (rqstp->rq_xprt && test_bit(XPT_TEMP, &rqstp->rq_xprt->xpt_flags)) 1500 svc_xprt_close(rqstp->rq_xprt); 1501 dprintk("svc: svc_process close\n"); 1502 return 0; 1503 1504 err_short_len: 1505 svc_printk(rqstp, "short len %u, dropping request\n", 1506 rqstp->rq_arg.len); 1507 goto close_xprt; 1508 1509 err_bad_rpc: 1510 if (serv->sv_stats) 1511 serv->sv_stats->rpcbadfmt++; 1512 xdr_stream_encode_u32(xdr, RPC_MSG_DENIED); 1513 xdr_stream_encode_u32(xdr, RPC_MISMATCH); 1514 /* Only RPCv2 supported */ 1515 xdr_stream_encode_u32(xdr, RPC_VERSION); 1516 xdr_stream_encode_u32(xdr, RPC_VERSION); 1517 return 1; /* don't wrap */ 1518 1519 err_bad_auth: 1520 dprintk("svc: authentication failed (%d)\n", 1521 be32_to_cpu(rqstp->rq_auth_stat)); 1522 if (serv->sv_stats) 1523 serv->sv_stats->rpcbadauth++; 1524 /* Restore write pointer to location of reply status: */ 1525 xdr_truncate_encode(xdr, XDR_UNIT * 2); 1526 xdr_stream_encode_u32(xdr, RPC_MSG_DENIED); 1527 xdr_stream_encode_u32(xdr, RPC_AUTH_ERROR); 1528 xdr_stream_encode_be32(xdr, rqstp->rq_auth_stat); 1529 goto sendit; 1530 1531 err_bad_prog: 1532 dprintk("svc: unknown program %d\n", rqstp->rq_prog); 1533 if (serv->sv_stats) 1534 serv->sv_stats->rpcbadfmt++; 1535 *rqstp->rq_accept_statp = rpc_prog_unavail; 1536 goto sendit; 1537 1538 err_bad_vers: 1539 svc_printk(rqstp, "unknown version (%d for prog %d, %s)\n", 1540 rqstp->rq_vers, rqstp->rq_prog, progp->pg_name); 1541 1542 if (serv->sv_stats) 1543 serv->sv_stats->rpcbadfmt++; 1544 *rqstp->rq_accept_statp = rpc_prog_mismatch; 1545 1546 /* 1547 * svc_authenticate() has already added the verifier and 1548 * advanced the stream just past rq_accept_statp. 1549 */ 1550 xdr_stream_encode_u32(xdr, process.mismatch.lovers); 1551 xdr_stream_encode_u32(xdr, process.mismatch.hivers); 1552 goto sendit; 1553 1554 err_bad_proc: 1555 svc_printk(rqstp, "unknown procedure (%d)\n", rqstp->rq_proc); 1556 1557 if (serv->sv_stats) 1558 serv->sv_stats->rpcbadfmt++; 1559 *rqstp->rq_accept_statp = rpc_proc_unavail; 1560 goto sendit; 1561 } 1562 1563 /* 1564 * Drop request 1565 */ 1566 static void svc_drop(struct svc_rqst *rqstp) 1567 { 1568 trace_svc_drop(rqstp); 1569 } 1570 1571 static void svc_release_rqst(struct svc_rqst *rqstp) 1572 { 1573 const struct svc_procedure *procp = rqstp->rq_procinfo; 1574 1575 if (procp && procp->pc_release) 1576 procp->pc_release(rqstp); 1577 } 1578 1579 /** 1580 * svc_process - Execute one RPC transaction 1581 * @rqstp: RPC transaction context 1582 * 1583 */ 1584 void svc_process(struct svc_rqst *rqstp) 1585 { 1586 struct kvec *resv = &rqstp->rq_res.head[0]; 1587 __be32 *p; 1588 1589 #if IS_ENABLED(CONFIG_FAIL_SUNRPC) 1590 if (!fail_sunrpc.ignore_server_disconnect && 1591 should_fail(&fail_sunrpc.attr, 1)) 1592 svc_xprt_deferred_close(rqstp->rq_xprt); 1593 #endif 1594 1595 /* 1596 * Setup response xdr_buf. 1597 * Initially it has just one page 1598 */ 1599 rqstp->rq_next_page = &rqstp->rq_respages[1]; 1600 resv->iov_base = page_address(rqstp->rq_respages[0]); 1601 resv->iov_len = 0; 1602 rqstp->rq_res.pages = rqstp->rq_next_page; 1603 rqstp->rq_res.len = 0; 1604 rqstp->rq_res.page_base = 0; 1605 rqstp->rq_res.page_len = 0; 1606 rqstp->rq_res.buflen = PAGE_SIZE; 1607 rqstp->rq_res.tail[0].iov_base = NULL; 1608 rqstp->rq_res.tail[0].iov_len = 0; 1609 1610 svcxdr_init_decode(rqstp); 1611 p = xdr_inline_decode(&rqstp->rq_arg_stream, XDR_UNIT * 2); 1612 if (unlikely(!p)) 1613 goto out_drop; 1614 rqstp->rq_xid = *p++; 1615 if (unlikely(*p != rpc_call)) 1616 goto out_baddir; 1617 1618 if (!svc_process_common(rqstp)) { 1619 svc_release_rqst(rqstp); 1620 goto out_drop; 1621 } 1622 svc_send(rqstp); 1623 svc_release_rqst(rqstp); 1624 return; 1625 1626 out_baddir: 1627 svc_printk(rqstp, "bad direction 0x%08x, dropping request\n", 1628 be32_to_cpu(*p)); 1629 if (rqstp->rq_server->sv_stats) 1630 rqstp->rq_server->sv_stats->rpcbadfmt++; 1631 out_drop: 1632 svc_drop(rqstp); 1633 } 1634 1635 #if defined(CONFIG_SUNRPC_BACKCHANNEL) 1636 /** 1637 * svc_process_bc - process a reverse-direction RPC request 1638 * @req: RPC request to be used for client-side processing 1639 * @rqstp: server-side execution context 1640 * 1641 */ 1642 void svc_process_bc(struct rpc_rqst *req, struct svc_rqst *rqstp) 1643 { 1644 struct rpc_timeout timeout = { 1645 .to_increment = 0, 1646 }; 1647 struct rpc_task *task; 1648 int proc_error; 1649 1650 /* Build the svc_rqst used by the common processing routine */ 1651 rqstp->rq_xid = req->rq_xid; 1652 rqstp->rq_prot = req->rq_xprt->prot; 1653 rqstp->rq_bc_net = req->rq_xprt->xprt_net; 1654 1655 rqstp->rq_addrlen = sizeof(req->rq_xprt->addr); 1656 memcpy(&rqstp->rq_addr, &req->rq_xprt->addr, rqstp->rq_addrlen); 1657 memcpy(&rqstp->rq_arg, &req->rq_rcv_buf, sizeof(rqstp->rq_arg)); 1658 memcpy(&rqstp->rq_res, &req->rq_snd_buf, sizeof(rqstp->rq_res)); 1659 1660 /* Adjust the argument buffer length */ 1661 rqstp->rq_arg.len = req->rq_private_buf.len; 1662 if (rqstp->rq_arg.len <= rqstp->rq_arg.head[0].iov_len) { 1663 rqstp->rq_arg.head[0].iov_len = rqstp->rq_arg.len; 1664 rqstp->rq_arg.page_len = 0; 1665 } else if (rqstp->rq_arg.len <= rqstp->rq_arg.head[0].iov_len + 1666 rqstp->rq_arg.page_len) 1667 rqstp->rq_arg.page_len = rqstp->rq_arg.len - 1668 rqstp->rq_arg.head[0].iov_len; 1669 else 1670 rqstp->rq_arg.len = rqstp->rq_arg.head[0].iov_len + 1671 rqstp->rq_arg.page_len; 1672 1673 /* Reset the response buffer */ 1674 rqstp->rq_res.head[0].iov_len = 0; 1675 1676 /* 1677 * Skip the XID and calldir fields because they've already 1678 * been processed by the caller. 1679 */ 1680 svcxdr_init_decode(rqstp); 1681 if (!xdr_inline_decode(&rqstp->rq_arg_stream, XDR_UNIT * 2)) 1682 return; 1683 1684 /* Parse and execute the bc call */ 1685 proc_error = svc_process_common(rqstp); 1686 1687 atomic_dec(&req->rq_xprt->bc_slot_count); 1688 if (!proc_error) { 1689 /* Processing error: drop the request */ 1690 xprt_free_bc_request(req); 1691 svc_release_rqst(rqstp); 1692 return; 1693 } 1694 /* Finally, send the reply synchronously */ 1695 if (rqstp->bc_to_initval > 0) { 1696 timeout.to_initval = rqstp->bc_to_initval; 1697 timeout.to_retries = rqstp->bc_to_retries; 1698 } else { 1699 timeout.to_initval = req->rq_xprt->timeout->to_initval; 1700 timeout.to_retries = req->rq_xprt->timeout->to_retries; 1701 } 1702 timeout.to_maxval = timeout.to_initval; 1703 memcpy(&req->rq_snd_buf, &rqstp->rq_res, sizeof(req->rq_snd_buf)); 1704 task = rpc_run_bc_task(req, &timeout); 1705 svc_release_rqst(rqstp); 1706 1707 if (IS_ERR(task)) 1708 return; 1709 1710 WARN_ON_ONCE(atomic_read(&task->tk_count) != 1); 1711 rpc_put_task(task); 1712 } 1713 #endif /* CONFIG_SUNRPC_BACKCHANNEL */ 1714 1715 /** 1716 * svc_max_payload - Return transport-specific limit on the RPC payload 1717 * @rqstp: RPC transaction context 1718 * 1719 * Returns the maximum number of payload bytes the current transport 1720 * allows. 1721 */ 1722 u32 svc_max_payload(const struct svc_rqst *rqstp) 1723 { 1724 u32 max = rqstp->rq_xprt->xpt_class->xcl_max_payload; 1725 1726 if (rqstp->rq_server->sv_max_payload < max) 1727 max = rqstp->rq_server->sv_max_payload; 1728 return max; 1729 } 1730 EXPORT_SYMBOL_GPL(svc_max_payload); 1731 1732 /** 1733 * svc_proc_name - Return RPC procedure name in string form 1734 * @rqstp: svc_rqst to operate on 1735 * 1736 * Return value: 1737 * Pointer to a NUL-terminated string 1738 */ 1739 const char *svc_proc_name(const struct svc_rqst *rqstp) 1740 { 1741 if (rqstp && rqstp->rq_procinfo) 1742 return rqstp->rq_procinfo->pc_name; 1743 return "unknown"; 1744 } 1745 1746 1747 /** 1748 * svc_encode_result_payload - mark a range of bytes as a result payload 1749 * @rqstp: svc_rqst to operate on 1750 * @offset: payload's byte offset in rqstp->rq_res 1751 * @length: size of payload, in bytes 1752 * 1753 * Returns zero on success, or a negative errno if a permanent 1754 * error occurred. 1755 */ 1756 int svc_encode_result_payload(struct svc_rqst *rqstp, unsigned int offset, 1757 unsigned int length) 1758 { 1759 return rqstp->rq_xprt->xpt_ops->xpo_result_payload(rqstp, offset, 1760 length); 1761 } 1762 EXPORT_SYMBOL_GPL(svc_encode_result_payload); 1763 1764 /** 1765 * svc_fill_symlink_pathname - Construct pathname argument for VFS symlink call 1766 * @rqstp: svc_rqst to operate on 1767 * @first: buffer containing first section of pathname 1768 * @p: buffer containing remaining section of pathname 1769 * @total: total length of the pathname argument 1770 * 1771 * The VFS symlink API demands a NUL-terminated pathname in mapped memory. 1772 * Returns pointer to a NUL-terminated string, or an ERR_PTR. Caller must free 1773 * the returned string. 1774 */ 1775 char *svc_fill_symlink_pathname(struct svc_rqst *rqstp, struct kvec *first, 1776 void *p, size_t total) 1777 { 1778 size_t len, remaining; 1779 char *result, *dst; 1780 1781 result = kmalloc(total + 1, GFP_KERNEL); 1782 if (!result) 1783 return ERR_PTR(-ESERVERFAULT); 1784 1785 dst = result; 1786 remaining = total; 1787 1788 len = min_t(size_t, total, first->iov_len); 1789 if (len) { 1790 memcpy(dst, first->iov_base, len); 1791 dst += len; 1792 remaining -= len; 1793 } 1794 1795 if (remaining) { 1796 len = min_t(size_t, remaining, PAGE_SIZE); 1797 memcpy(dst, p, len); 1798 dst += len; 1799 } 1800 1801 *dst = '\0'; 1802 1803 /* Sanity check: Linux doesn't allow the pathname argument to 1804 * contain a NUL byte. 1805 */ 1806 if (strlen(result) != total) { 1807 kfree(result); 1808 return ERR_PTR(-EINVAL); 1809 } 1810 return result; 1811 } 1812 EXPORT_SYMBOL_GPL(svc_fill_symlink_pathname); 1813