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