1 /* SPDX-License-Identifier: GPL-2.0 */ 2 /* 3 * linux/include/linux/sunrpc/svc.h 4 * 5 * RPC server declarations. 6 * 7 * Copyright (C) 1995, 1996 Olaf Kirch <[email protected]> 8 */ 9 10 11 #ifndef SUNRPC_SVC_H 12 #define SUNRPC_SVC_H 13 14 #include <linux/in.h> 15 #include <linux/in6.h> 16 #include <linux/sunrpc/types.h> 17 #include <linux/sunrpc/xdr.h> 18 #include <linux/sunrpc/auth.h> 19 #include <linux/sunrpc/svcauth.h> 20 #include <linux/lwq.h> 21 #include <linux/wait.h> 22 #include <linux/mm.h> 23 #include <linux/pagevec.h> 24 #include <linux/kthread.h> 25 26 /* 27 * 28 * RPC service thread pool. 29 * 30 * Pool of threads and temporary sockets. Generally there is only 31 * a single one of these per RPC service, but on NUMA machines those 32 * services that can benefit from it (i.e. nfs but not lockd) will 33 * have one pool per NUMA node. This optimisation reduces cross- 34 * node traffic on multi-node NUMA NFS servers. 35 */ 36 struct svc_pool { 37 unsigned int sp_id; /* pool id; also node id on NUMA */ 38 struct lwq sp_xprts; /* pending transports */ 39 unsigned int sp_nrthreads; /* # of threads in pool */ 40 struct list_head sp_all_threads; /* all server threads */ 41 struct llist_head sp_idle_threads; /* idle server threads */ 42 43 /* statistics on pool operation */ 44 struct percpu_counter sp_messages_arrived; 45 struct percpu_counter sp_sockets_queued; 46 struct percpu_counter sp_threads_woken; 47 48 unsigned long sp_flags; 49 } ____cacheline_aligned_in_smp; 50 51 /* bits for sp_flags */ 52 enum { 53 SP_TASK_PENDING, /* still work to do even if no xprt is queued */ 54 SP_NEED_VICTIM, /* One thread needs to agree to exit */ 55 SP_VICTIM_REMAINS, /* One thread needs to actually exit */ 56 }; 57 58 59 /* 60 * RPC service. 61 * 62 * An RPC service is a ``daemon,'' possibly multithreaded, which 63 * receives and processes incoming RPC messages. 64 * It has one or more transport sockets associated with it, and maintains 65 * a list of idle threads waiting for input. 66 * 67 * We currently do not support more than one RPC program per daemon. 68 */ 69 struct svc_serv { 70 struct svc_program * sv_program; /* RPC program */ 71 struct svc_stat * sv_stats; /* RPC statistics */ 72 spinlock_t sv_lock; 73 unsigned int sv_nrthreads; /* # of server threads */ 74 unsigned int sv_maxconn; /* max connections allowed or 75 * '0' causing max to be based 76 * on number of threads. */ 77 78 unsigned int sv_max_payload; /* datagram payload size */ 79 unsigned int sv_max_mesg; /* max_payload + 1 page for overheads */ 80 unsigned int sv_xdrsize; /* XDR buffer size */ 81 struct list_head sv_permsocks; /* all permanent sockets */ 82 struct list_head sv_tempsocks; /* all temporary sockets */ 83 int sv_tmpcnt; /* count of temporary sockets */ 84 struct timer_list sv_temptimer; /* timer for aging temporary sockets */ 85 86 char * sv_name; /* service name */ 87 88 unsigned int sv_nrpools; /* number of thread pools */ 89 bool sv_is_pooled; /* is this a pooled service? */ 90 struct svc_pool * sv_pools; /* array of thread pools */ 91 int (*sv_threadfn)(void *data); 92 93 #if defined(CONFIG_SUNRPC_BACKCHANNEL) 94 struct lwq sv_cb_list; /* queue for callback requests 95 * that arrive over the same 96 * connection */ 97 bool sv_bc_enabled; /* service uses backchannel */ 98 #endif /* CONFIG_SUNRPC_BACKCHANNEL */ 99 }; 100 101 /* This is used by pool_stats to find and lock an svc */ 102 struct svc_info { 103 struct svc_serv *serv; 104 struct mutex *mutex; 105 }; 106 107 void svc_destroy(struct svc_serv **svcp); 108 109 /* 110 * Maximum payload size supported by a kernel RPC server. 111 * This is use to determine the max number of pages nfsd is 112 * willing to return in a single READ operation. 113 * 114 * These happen to all be powers of 2, which is not strictly 115 * necessary but helps enforce the real limitation, which is 116 * that they should be multiples of PAGE_SIZE. 117 * 118 * For UDP transports, a block plus NFS,RPC, and UDP headers 119 * has to fit into the IP datagram limit of 64K. The largest 120 * feasible number for all known page sizes is probably 48K, 121 * but we choose 32K here. This is the same as the historical 122 * Linux limit; someone who cares more about NFS/UDP performance 123 * can test a larger number. 124 * 125 * For TCP transports we have more freedom. A size of 1MB is 126 * chosen to match the client limit. Other OSes are known to 127 * have larger limits, but those numbers are probably beyond 128 * the point of diminishing returns. 129 */ 130 #define RPCSVC_MAXPAYLOAD (1*1024*1024u) 131 #define RPCSVC_MAXPAYLOAD_TCP RPCSVC_MAXPAYLOAD 132 #define RPCSVC_MAXPAYLOAD_UDP (32*1024u) 133 134 extern u32 svc_max_payload(const struct svc_rqst *rqstp); 135 136 /* 137 * RPC Requests and replies are stored in one or more pages. 138 * We maintain an array of pages for each server thread. 139 * Requests are copied into these pages as they arrive. Remaining 140 * pages are available to write the reply into. 141 * 142 * Pages are sent using ->sendmsg with MSG_SPLICE_PAGES so each server thread 143 * needs to allocate more to replace those used in sending. To help keep track 144 * of these pages we have a receive list where all pages initialy live, and a 145 * send list where pages are moved to when there are to be part of a reply. 146 * 147 * We use xdr_buf for holding responses as it fits well with NFS 148 * read responses (that have a header, and some data pages, and possibly 149 * a tail) and means we can share some client side routines. 150 * 151 * The xdr_buf.head kvec always points to the first page in the rq_*pages 152 * list. The xdr_buf.pages pointer points to the second page on that 153 * list. xdr_buf.tail points to the end of the first page. 154 * This assumes that the non-page part of an rpc reply will fit 155 * in a page - NFSd ensures this. lockd also has no trouble. 156 * 157 * Each request/reply pair can have at most one "payload", plus two pages, 158 * one for the request, and one for the reply. 159 * We using ->sendfile to return read data, we might need one extra page 160 * if the request is not page-aligned. So add another '1'. 161 */ 162 #define RPCSVC_MAXPAGES ((RPCSVC_MAXPAYLOAD+PAGE_SIZE-1)/PAGE_SIZE \ 163 + 2 + 1) 164 165 /* 166 * The context of a single thread, including the request currently being 167 * processed. 168 */ 169 struct svc_rqst { 170 struct list_head rq_all; /* all threads list */ 171 struct llist_node rq_idle; /* On the idle list */ 172 struct rcu_head rq_rcu_head; /* for RCU deferred kfree */ 173 struct svc_xprt * rq_xprt; /* transport ptr */ 174 175 struct sockaddr_storage rq_addr; /* peer address */ 176 size_t rq_addrlen; 177 struct sockaddr_storage rq_daddr; /* dest addr of request 178 * - reply from here */ 179 size_t rq_daddrlen; 180 181 struct svc_serv * rq_server; /* RPC service definition */ 182 struct svc_pool * rq_pool; /* thread pool */ 183 const struct svc_procedure *rq_procinfo;/* procedure info */ 184 struct auth_ops * rq_authop; /* authentication flavour */ 185 struct svc_cred rq_cred; /* auth info */ 186 void * rq_xprt_ctxt; /* transport specific context ptr */ 187 struct svc_deferred_req*rq_deferred; /* deferred request we are replaying */ 188 189 struct xdr_buf rq_arg; 190 struct xdr_stream rq_arg_stream; 191 struct xdr_stream rq_res_stream; 192 struct page *rq_scratch_page; 193 struct xdr_buf rq_res; 194 struct page *rq_pages[RPCSVC_MAXPAGES + 1]; 195 struct page * *rq_respages; /* points into rq_pages */ 196 struct page * *rq_next_page; /* next reply page to use */ 197 struct page * *rq_page_end; /* one past the last page */ 198 199 struct folio_batch rq_fbatch; 200 struct kvec rq_vec[RPCSVC_MAXPAGES]; /* generally useful.. */ 201 struct bio_vec rq_bvec[RPCSVC_MAXPAGES]; 202 203 __be32 rq_xid; /* transmission id */ 204 u32 rq_prog; /* program number */ 205 u32 rq_vers; /* program version */ 206 u32 rq_proc; /* procedure number */ 207 u32 rq_prot; /* IP protocol */ 208 int rq_cachetype; /* catering to nfsd */ 209 unsigned long rq_flags; /* flags field */ 210 ktime_t rq_qtime; /* enqueue time */ 211 212 void * rq_argp; /* decoded arguments */ 213 void * rq_resp; /* xdr'd results */ 214 __be32 *rq_accept_statp; 215 void * rq_auth_data; /* flavor-specific data */ 216 __be32 rq_auth_stat; /* authentication status */ 217 int rq_auth_slack; /* extra space xdr code 218 * should leave in head 219 * for krb5i, krb5p. 220 */ 221 int rq_reserved; /* space on socket outq 222 * reserved for this request 223 */ 224 ktime_t rq_stime; /* start time */ 225 226 struct cache_req rq_chandle; /* handle passed to caches for 227 * request delaying 228 */ 229 /* Catering to nfsd */ 230 struct auth_domain * rq_client; /* RPC peer info */ 231 struct auth_domain * rq_gssclient; /* "gss/"-style peer info */ 232 struct task_struct *rq_task; /* service thread */ 233 struct net *rq_bc_net; /* pointer to backchannel's 234 * net namespace 235 */ 236 237 int rq_err; /* Thread sets this to inidicate 238 * initialisation success. 239 */ 240 241 unsigned long bc_to_initval; 242 unsigned int bc_to_retries; 243 void ** rq_lease_breaker; /* The v4 client breaking a lease */ 244 unsigned int rq_status_counter; /* RPC processing counter */ 245 }; 246 247 /* bits for rq_flags */ 248 enum { 249 RQ_SECURE, /* secure port */ 250 RQ_LOCAL, /* local request */ 251 RQ_USEDEFERRAL, /* use deferral */ 252 RQ_DROPME, /* drop current reply */ 253 RQ_VICTIM, /* Have agreed to shut down */ 254 RQ_DATA, /* request has data */ 255 }; 256 257 #define SVC_NET(rqst) (rqst->rq_xprt ? rqst->rq_xprt->xpt_net : rqst->rq_bc_net) 258 259 /* 260 * Rigorous type checking on sockaddr type conversions 261 */ 262 static inline struct sockaddr_in *svc_addr_in(const struct svc_rqst *rqst) 263 { 264 return (struct sockaddr_in *) &rqst->rq_addr; 265 } 266 267 static inline struct sockaddr_in6 *svc_addr_in6(const struct svc_rqst *rqst) 268 { 269 return (struct sockaddr_in6 *) &rqst->rq_addr; 270 } 271 272 static inline struct sockaddr *svc_addr(const struct svc_rqst *rqst) 273 { 274 return (struct sockaddr *) &rqst->rq_addr; 275 } 276 277 static inline struct sockaddr_in *svc_daddr_in(const struct svc_rqst *rqst) 278 { 279 return (struct sockaddr_in *) &rqst->rq_daddr; 280 } 281 282 static inline struct sockaddr_in6 *svc_daddr_in6(const struct svc_rqst *rqst) 283 { 284 return (struct sockaddr_in6 *) &rqst->rq_daddr; 285 } 286 287 static inline struct sockaddr *svc_daddr(const struct svc_rqst *rqst) 288 { 289 return (struct sockaddr *) &rqst->rq_daddr; 290 } 291 292 /** 293 * svc_thread_should_stop - check if this thread should stop 294 * @rqstp: the thread that might need to stop 295 * 296 * To stop an svc thread, the pool flags SP_NEED_VICTIM and SP_VICTIM_REMAINS 297 * are set. The first thread which sees SP_NEED_VICTIM clears it, becoming 298 * the victim using this function. It should then promptly call 299 * svc_exit_thread() to complete the process, clearing SP_VICTIM_REMAINS 300 * so the task waiting for a thread to exit can wake and continue. 301 * 302 * Return values: 303 * %true: caller should invoke svc_exit_thread() 304 * %false: caller should do nothing 305 */ 306 static inline bool svc_thread_should_stop(struct svc_rqst *rqstp) 307 { 308 if (test_and_clear_bit(SP_NEED_VICTIM, &rqstp->rq_pool->sp_flags)) 309 set_bit(RQ_VICTIM, &rqstp->rq_flags); 310 311 return test_bit(RQ_VICTIM, &rqstp->rq_flags); 312 } 313 314 /** 315 * svc_thread_init_status - report whether thread has initialised successfully 316 * @rqstp: the thread in question 317 * @err: errno code 318 * 319 * After performing any initialisation that could fail, and before starting 320 * normal work, each sunrpc svc_thread must call svc_thread_init_status() 321 * with an appropriate error, or zero. 322 * 323 * If zero is passed, the thread is ready and must continue until 324 * svc_thread_should_stop() returns true. If a non-zero error is passed 325 * the call will not return - the thread will exit. 326 */ 327 static inline void svc_thread_init_status(struct svc_rqst *rqstp, int err) 328 { 329 rqstp->rq_err = err; 330 /* memory barrier ensures assignment to error above is visible before 331 * waitqueue_active() test below completes. 332 */ 333 smp_mb(); 334 wake_up_var(&rqstp->rq_err); 335 if (err) 336 kthread_exit(1); 337 } 338 339 struct svc_deferred_req { 340 u32 prot; /* protocol (UDP or TCP) */ 341 struct svc_xprt *xprt; 342 struct sockaddr_storage addr; /* where reply must go */ 343 size_t addrlen; 344 struct sockaddr_storage daddr; /* where reply must come from */ 345 size_t daddrlen; 346 void *xprt_ctxt; 347 struct cache_deferred_req handle; 348 int argslen; 349 __be32 args[]; 350 }; 351 352 struct svc_process_info { 353 union { 354 int (*dispatch)(struct svc_rqst *rqstp); 355 struct { 356 unsigned int lovers; 357 unsigned int hivers; 358 } mismatch; 359 }; 360 }; 361 362 /* 363 * List of RPC programs on the same transport endpoint 364 */ 365 struct svc_program { 366 struct svc_program * pg_next; /* other programs (same xprt) */ 367 u32 pg_prog; /* program number */ 368 unsigned int pg_lovers; /* lowest version */ 369 unsigned int pg_hivers; /* highest version */ 370 unsigned int pg_nvers; /* number of versions */ 371 const struct svc_version **pg_vers; /* version array */ 372 char * pg_name; /* service name */ 373 char * pg_class; /* class name: services sharing authentication */ 374 enum svc_auth_status (*pg_authenticate)(struct svc_rqst *rqstp); 375 __be32 (*pg_init_request)(struct svc_rqst *, 376 const struct svc_program *, 377 struct svc_process_info *); 378 int (*pg_rpcbind_set)(struct net *net, 379 const struct svc_program *, 380 u32 version, int family, 381 unsigned short proto, 382 unsigned short port); 383 }; 384 385 /* 386 * RPC program version 387 */ 388 struct svc_version { 389 u32 vs_vers; /* version number */ 390 u32 vs_nproc; /* number of procedures */ 391 const struct svc_procedure *vs_proc; /* per-procedure info */ 392 unsigned long __percpu *vs_count; /* call counts */ 393 u32 vs_xdrsize; /* xdrsize needed for this version */ 394 395 /* Don't register with rpcbind */ 396 bool vs_hidden; 397 398 /* Don't care if the rpcbind registration fails */ 399 bool vs_rpcb_optnl; 400 401 /* Need xprt with congestion control */ 402 bool vs_need_cong_ctrl; 403 404 /* Dispatch function */ 405 int (*vs_dispatch)(struct svc_rqst *rqstp); 406 }; 407 408 /* 409 * RPC procedure info 410 */ 411 struct svc_procedure { 412 /* process the request: */ 413 __be32 (*pc_func)(struct svc_rqst *); 414 /* XDR decode args: */ 415 bool (*pc_decode)(struct svc_rqst *rqstp, 416 struct xdr_stream *xdr); 417 /* XDR encode result: */ 418 bool (*pc_encode)(struct svc_rqst *rqstp, 419 struct xdr_stream *xdr); 420 /* XDR free result: */ 421 void (*pc_release)(struct svc_rqst *); 422 unsigned int pc_argsize; /* argument struct size */ 423 unsigned int pc_argzero; /* how much of argument to clear */ 424 unsigned int pc_ressize; /* result struct size */ 425 unsigned int pc_cachetype; /* cache info (NFS) */ 426 unsigned int pc_xdrressize; /* maximum size of XDR reply */ 427 const char * pc_name; /* for display */ 428 }; 429 430 /* 431 * Function prototypes. 432 */ 433 int sunrpc_set_pool_mode(const char *val); 434 int sunrpc_get_pool_mode(char *val, size_t size); 435 void svc_rpcb_cleanup(struct svc_serv *serv, struct net *net); 436 int svc_bind(struct svc_serv *serv, struct net *net); 437 struct svc_serv *svc_create(struct svc_program *, unsigned int, 438 int (*threadfn)(void *data)); 439 bool svc_rqst_replace_page(struct svc_rqst *rqstp, 440 struct page *page); 441 void svc_rqst_release_pages(struct svc_rqst *rqstp); 442 void svc_exit_thread(struct svc_rqst *); 443 struct svc_serv * svc_create_pooled(struct svc_program *prog, 444 struct svc_stat *stats, 445 unsigned int bufsize, 446 int (*threadfn)(void *data)); 447 int svc_set_num_threads(struct svc_serv *, struct svc_pool *, int); 448 int svc_pool_stats_open(struct svc_info *si, struct file *file); 449 void svc_process(struct svc_rqst *rqstp); 450 void svc_process_bc(struct rpc_rqst *req, struct svc_rqst *rqstp); 451 int svc_register(const struct svc_serv *, struct net *, const int, 452 const unsigned short, const unsigned short); 453 454 void svc_wake_up(struct svc_serv *); 455 void svc_reserve(struct svc_rqst *rqstp, int space); 456 void svc_pool_wake_idle_thread(struct svc_pool *pool); 457 struct svc_pool *svc_pool_for_cpu(struct svc_serv *serv); 458 char * svc_print_addr(struct svc_rqst *, char *, size_t); 459 const char * svc_proc_name(const struct svc_rqst *rqstp); 460 int svc_encode_result_payload(struct svc_rqst *rqstp, 461 unsigned int offset, 462 unsigned int length); 463 unsigned int svc_fill_write_vector(struct svc_rqst *rqstp, 464 struct xdr_buf *payload); 465 char *svc_fill_symlink_pathname(struct svc_rqst *rqstp, 466 struct kvec *first, void *p, 467 size_t total); 468 __be32 svc_generic_init_request(struct svc_rqst *rqstp, 469 const struct svc_program *progp, 470 struct svc_process_info *procinfo); 471 int svc_generic_rpcbind_set(struct net *net, 472 const struct svc_program *progp, 473 u32 version, int family, 474 unsigned short proto, 475 unsigned short port); 476 477 #define RPC_MAX_ADDRBUFLEN (63U) 478 479 /* 480 * When we want to reduce the size of the reserved space in the response 481 * buffer, we need to take into account the size of any checksum data that 482 * may be at the end of the packet. This is difficult to determine exactly 483 * for all cases without actually generating the checksum, so we just use a 484 * static value. 485 */ 486 static inline void svc_reserve_auth(struct svc_rqst *rqstp, int space) 487 { 488 svc_reserve(rqstp, space + rqstp->rq_auth_slack); 489 } 490 491 /** 492 * svcxdr_init_decode - Prepare an xdr_stream for Call decoding 493 * @rqstp: controlling server RPC transaction context 494 * 495 */ 496 static inline void svcxdr_init_decode(struct svc_rqst *rqstp) 497 { 498 struct xdr_stream *xdr = &rqstp->rq_arg_stream; 499 struct xdr_buf *buf = &rqstp->rq_arg; 500 struct kvec *argv = buf->head; 501 502 WARN_ON(buf->len != buf->head->iov_len + buf->page_len + buf->tail->iov_len); 503 buf->len = buf->head->iov_len + buf->page_len + buf->tail->iov_len; 504 505 xdr_init_decode(xdr, buf, argv->iov_base, NULL); 506 xdr_set_scratch_page(xdr, rqstp->rq_scratch_page); 507 } 508 509 /** 510 * svcxdr_init_encode - Prepare an xdr_stream for svc Reply encoding 511 * @rqstp: controlling server RPC transaction context 512 * 513 */ 514 static inline void svcxdr_init_encode(struct svc_rqst *rqstp) 515 { 516 struct xdr_stream *xdr = &rqstp->rq_res_stream; 517 struct xdr_buf *buf = &rqstp->rq_res; 518 struct kvec *resv = buf->head; 519 520 xdr_reset_scratch_buffer(xdr); 521 522 xdr->buf = buf; 523 xdr->iov = resv; 524 xdr->p = resv->iov_base + resv->iov_len; 525 xdr->end = resv->iov_base + PAGE_SIZE; 526 buf->len = resv->iov_len; 527 xdr->page_ptr = buf->pages - 1; 528 buf->buflen = PAGE_SIZE * (rqstp->rq_page_end - buf->pages); 529 xdr->rqst = NULL; 530 } 531 532 /** 533 * svcxdr_encode_opaque_pages - Insert pages into an xdr_stream 534 * @xdr: xdr_stream to be updated 535 * @pages: array of pages to insert 536 * @base: starting offset of first data byte in @pages 537 * @len: number of data bytes in @pages to insert 538 * 539 * After the @pages are added, the tail iovec is instantiated pointing 540 * to end of the head buffer, and the stream is set up to encode 541 * subsequent items into the tail. 542 */ 543 static inline void svcxdr_encode_opaque_pages(struct svc_rqst *rqstp, 544 struct xdr_stream *xdr, 545 struct page **pages, 546 unsigned int base, 547 unsigned int len) 548 { 549 xdr_write_pages(xdr, pages, base, len); 550 xdr->page_ptr = rqstp->rq_next_page - 1; 551 } 552 553 /** 554 * svcxdr_set_auth_slack - 555 * @rqstp: RPC transaction 556 * @slack: buffer space to reserve for the transaction's security flavor 557 * 558 * Set the request's slack space requirement, and set aside that much 559 * space in the rqstp's rq_res.head for use when the auth wraps the Reply. 560 */ 561 static inline void svcxdr_set_auth_slack(struct svc_rqst *rqstp, int slack) 562 { 563 struct xdr_stream *xdr = &rqstp->rq_res_stream; 564 struct xdr_buf *buf = &rqstp->rq_res; 565 struct kvec *resv = buf->head; 566 567 rqstp->rq_auth_slack = slack; 568 569 xdr->end -= XDR_QUADLEN(slack); 570 buf->buflen -= rqstp->rq_auth_slack; 571 572 WARN_ON(xdr->iov != resv); 573 WARN_ON(xdr->p > xdr->end); 574 } 575 576 /** 577 * svcxdr_set_accept_stat - Reserve space for the accept_stat field 578 * @rqstp: RPC transaction context 579 * 580 * Return values: 581 * %true: Success 582 * %false: No response buffer space was available 583 */ 584 static inline bool svcxdr_set_accept_stat(struct svc_rqst *rqstp) 585 { 586 struct xdr_stream *xdr = &rqstp->rq_res_stream; 587 588 rqstp->rq_accept_statp = xdr_reserve_space(xdr, XDR_UNIT); 589 if (unlikely(!rqstp->rq_accept_statp)) 590 return false; 591 *rqstp->rq_accept_statp = rpc_success; 592 return true; 593 } 594 595 #endif /* SUNRPC_SVC_H */ 596