svc.h (18781B)
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/wait.h> 21#include <linux/mm.h> 22#include <linux/pagevec.h> 23 24/* statistics for svc_pool structures */ 25struct svc_pool_stats { 26 atomic_long_t packets; 27 unsigned long sockets_queued; 28 atomic_long_t threads_woken; 29 atomic_long_t threads_timedout; 30}; 31 32/* 33 * 34 * RPC service thread pool. 35 * 36 * Pool of threads and temporary sockets. Generally there is only 37 * a single one of these per RPC service, but on NUMA machines those 38 * services that can benefit from it (i.e. nfs but not lockd) will 39 * have one pool per NUMA node. This optimisation reduces cross- 40 * node traffic on multi-node NUMA NFS servers. 41 */ 42struct svc_pool { 43 unsigned int sp_id; /* pool id; also node id on NUMA */ 44 spinlock_t sp_lock; /* protects all fields */ 45 struct list_head sp_sockets; /* pending sockets */ 46 unsigned int sp_nrthreads; /* # of threads in pool */ 47 struct list_head sp_all_threads; /* all server threads */ 48 struct svc_pool_stats sp_stats; /* statistics on pool operation */ 49#define SP_TASK_PENDING (0) /* still work to do even if no 50 * xprt is queued. */ 51#define SP_CONGESTED (1) 52 unsigned long sp_flags; 53} ____cacheline_aligned_in_smp; 54 55/* 56 * RPC service. 57 * 58 * An RPC service is a ``daemon,'' possibly multithreaded, which 59 * receives and processes incoming RPC messages. 60 * It has one or more transport sockets associated with it, and maintains 61 * a list of idle threads waiting for input. 62 * 63 * We currently do not support more than one RPC program per daemon. 64 */ 65struct svc_serv { 66 struct svc_program * sv_program; /* RPC program */ 67 struct svc_stat * sv_stats; /* RPC statistics */ 68 spinlock_t sv_lock; 69 struct kref sv_refcnt; 70 unsigned int sv_nrthreads; /* # of server threads */ 71 unsigned int sv_maxconn; /* max connections allowed or 72 * '0' causing max to be based 73 * on number of threads. */ 74 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 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 struct svc_pool * sv_pools; /* array of thread pools */ 87 int (*sv_threadfn)(void *data); 88 89#if defined(CONFIG_SUNRPC_BACKCHANNEL) 90 struct list_head sv_cb_list; /* queue for callback requests 91 * that arrive over the same 92 * connection */ 93 spinlock_t sv_cb_lock; /* protects the svc_cb_list */ 94 wait_queue_head_t sv_cb_waitq; /* sleep here if there are no 95 * entries in the svc_cb_list */ 96 bool sv_bc_enabled; /* service uses backchannel */ 97#endif /* CONFIG_SUNRPC_BACKCHANNEL */ 98}; 99 100/** 101 * svc_get() - increment reference count on a SUNRPC serv 102 * @serv: the svc_serv to have count incremented 103 * 104 * Returns: the svc_serv that was passed in. 105 */ 106static inline struct svc_serv *svc_get(struct svc_serv *serv) 107{ 108 kref_get(&serv->sv_refcnt); 109 return serv; 110} 111 112void svc_destroy(struct kref *); 113 114/** 115 * svc_put - decrement reference count on a SUNRPC serv 116 * @serv: the svc_serv to have count decremented 117 * 118 * When the reference count reaches zero, svc_destroy() 119 * is called to clean up and free the serv. 120 */ 121static inline void svc_put(struct svc_serv *serv) 122{ 123 kref_put(&serv->sv_refcnt, svc_destroy); 124} 125 126/** 127 * svc_put_not_last - decrement non-final reference count on SUNRPC serv 128 * @serv: the svc_serv to have count decremented 129 * 130 * Returns: %true is refcount was decremented. 131 * 132 * If the refcount is 1, it is not decremented and instead failure is reported. 133 */ 134static inline bool svc_put_not_last(struct svc_serv *serv) 135{ 136 return refcount_dec_not_one(&serv->sv_refcnt.refcount); 137} 138 139/* 140 * Maximum payload size supported by a kernel RPC server. 141 * This is use to determine the max number of pages nfsd is 142 * willing to return in a single READ operation. 143 * 144 * These happen to all be powers of 2, which is not strictly 145 * necessary but helps enforce the real limitation, which is 146 * that they should be multiples of PAGE_SIZE. 147 * 148 * For UDP transports, a block plus NFS,RPC, and UDP headers 149 * has to fit into the IP datagram limit of 64K. The largest 150 * feasible number for all known page sizes is probably 48K, 151 * but we choose 32K here. This is the same as the historical 152 * Linux limit; someone who cares more about NFS/UDP performance 153 * can test a larger number. 154 * 155 * For TCP transports we have more freedom. A size of 1MB is 156 * chosen to match the client limit. Other OSes are known to 157 * have larger limits, but those numbers are probably beyond 158 * the point of diminishing returns. 159 */ 160#define RPCSVC_MAXPAYLOAD (1*1024*1024u) 161#define RPCSVC_MAXPAYLOAD_TCP RPCSVC_MAXPAYLOAD 162#define RPCSVC_MAXPAYLOAD_UDP (32*1024u) 163 164extern u32 svc_max_payload(const struct svc_rqst *rqstp); 165 166/* 167 * RPC Requsts and replies are stored in one or more pages. 168 * We maintain an array of pages for each server thread. 169 * Requests are copied into these pages as they arrive. Remaining 170 * pages are available to write the reply into. 171 * 172 * Pages are sent using ->sendpage so each server thread needs to 173 * allocate more to replace those used in sending. To help keep track 174 * of these pages we have a receive list where all pages initialy live, 175 * and a send list where pages are moved to when there are to be part 176 * of a reply. 177 * 178 * We use xdr_buf for holding responses as it fits well with NFS 179 * read responses (that have a header, and some data pages, and possibly 180 * a tail) and means we can share some client side routines. 181 * 182 * The xdr_buf.head kvec always points to the first page in the rq_*pages 183 * list. The xdr_buf.pages pointer points to the second page on that 184 * list. xdr_buf.tail points to the end of the first page. 185 * This assumes that the non-page part of an rpc reply will fit 186 * in a page - NFSd ensures this. lockd also has no trouble. 187 * 188 * Each request/reply pair can have at most one "payload", plus two pages, 189 * one for the request, and one for the reply. 190 * We using ->sendfile to return read data, we might need one extra page 191 * if the request is not page-aligned. So add another '1'. 192 */ 193#define RPCSVC_MAXPAGES ((RPCSVC_MAXPAYLOAD+PAGE_SIZE-1)/PAGE_SIZE \ 194 + 2 + 1) 195 196static inline u32 svc_getnl(struct kvec *iov) 197{ 198 __be32 val, *vp; 199 vp = iov->iov_base; 200 val = *vp++; 201 iov->iov_base = (void*)vp; 202 iov->iov_len -= sizeof(__be32); 203 return ntohl(val); 204} 205 206static inline void svc_putnl(struct kvec *iov, u32 val) 207{ 208 __be32 *vp = iov->iov_base + iov->iov_len; 209 *vp = htonl(val); 210 iov->iov_len += sizeof(__be32); 211} 212 213static inline __be32 svc_getu32(struct kvec *iov) 214{ 215 __be32 val, *vp; 216 vp = iov->iov_base; 217 val = *vp++; 218 iov->iov_base = (void*)vp; 219 iov->iov_len -= sizeof(__be32); 220 return val; 221} 222 223static inline void svc_ungetu32(struct kvec *iov) 224{ 225 __be32 *vp = (__be32 *)iov->iov_base; 226 iov->iov_base = (void *)(vp - 1); 227 iov->iov_len += sizeof(*vp); 228} 229 230static inline void svc_putu32(struct kvec *iov, __be32 val) 231{ 232 __be32 *vp = iov->iov_base + iov->iov_len; 233 *vp = val; 234 iov->iov_len += sizeof(__be32); 235} 236 237/* 238 * The context of a single thread, including the request currently being 239 * processed. 240 */ 241struct svc_rqst { 242 struct list_head rq_all; /* all threads list */ 243 struct rcu_head rq_rcu_head; /* for RCU deferred kfree */ 244 struct svc_xprt * rq_xprt; /* transport ptr */ 245 246 struct sockaddr_storage rq_addr; /* peer address */ 247 size_t rq_addrlen; 248 struct sockaddr_storage rq_daddr; /* dest addr of request 249 * - reply from here */ 250 size_t rq_daddrlen; 251 252 struct svc_serv * rq_server; /* RPC service definition */ 253 struct svc_pool * rq_pool; /* thread pool */ 254 const struct svc_procedure *rq_procinfo;/* procedure info */ 255 struct auth_ops * rq_authop; /* authentication flavour */ 256 struct svc_cred rq_cred; /* auth info */ 257 void * rq_xprt_ctxt; /* transport specific context ptr */ 258 struct svc_deferred_req*rq_deferred; /* deferred request we are replaying */ 259 260 struct xdr_buf rq_arg; 261 struct xdr_stream rq_arg_stream; 262 struct xdr_stream rq_res_stream; 263 struct page *rq_scratch_page; 264 struct xdr_buf rq_res; 265 struct page *rq_pages[RPCSVC_MAXPAGES + 1]; 266 struct page * *rq_respages; /* points into rq_pages */ 267 struct page * *rq_next_page; /* next reply page to use */ 268 struct page * *rq_page_end; /* one past the last page */ 269 270 struct pagevec rq_pvec; 271 struct kvec rq_vec[RPCSVC_MAXPAGES]; /* generally useful.. */ 272 struct bio_vec rq_bvec[RPCSVC_MAXPAGES]; 273 274 __be32 rq_xid; /* transmission id */ 275 u32 rq_prog; /* program number */ 276 u32 rq_vers; /* program version */ 277 u32 rq_proc; /* procedure number */ 278 u32 rq_prot; /* IP protocol */ 279 int rq_cachetype; /* catering to nfsd */ 280#define RQ_SECURE (0) /* secure port */ 281#define RQ_LOCAL (1) /* local request */ 282#define RQ_USEDEFERRAL (2) /* use deferral */ 283#define RQ_DROPME (3) /* drop current reply */ 284#define RQ_SPLICE_OK (4) /* turned off in gss privacy 285 * to prevent encrypting page 286 * cache pages */ 287#define RQ_VICTIM (5) /* about to be shut down */ 288#define RQ_BUSY (6) /* request is busy */ 289#define RQ_DATA (7) /* request has data */ 290 unsigned long rq_flags; /* flags field */ 291 ktime_t rq_qtime; /* enqueue time */ 292 293 void * rq_argp; /* decoded arguments */ 294 void * rq_resp; /* xdr'd results */ 295 void * rq_auth_data; /* flavor-specific data */ 296 __be32 rq_auth_stat; /* authentication status */ 297 int rq_auth_slack; /* extra space xdr code 298 * should leave in head 299 * for krb5i, krb5p. 300 */ 301 int rq_reserved; /* space on socket outq 302 * reserved for this request 303 */ 304 ktime_t rq_stime; /* start time */ 305 306 struct cache_req rq_chandle; /* handle passed to caches for 307 * request delaying 308 */ 309 /* Catering to nfsd */ 310 struct auth_domain * rq_client; /* RPC peer info */ 311 struct auth_domain * rq_gssclient; /* "gss/"-style peer info */ 312 struct svc_cacherep * rq_cacherep; /* cache info */ 313 struct task_struct *rq_task; /* service thread */ 314 spinlock_t rq_lock; /* per-request lock */ 315 struct net *rq_bc_net; /* pointer to backchannel's 316 * net namespace 317 */ 318 void ** rq_lease_breaker; /* The v4 client breaking a lease */ 319}; 320 321#define SVC_NET(rqst) (rqst->rq_xprt ? rqst->rq_xprt->xpt_net : rqst->rq_bc_net) 322 323/* 324 * Rigorous type checking on sockaddr type conversions 325 */ 326static inline struct sockaddr_in *svc_addr_in(const struct svc_rqst *rqst) 327{ 328 return (struct sockaddr_in *) &rqst->rq_addr; 329} 330 331static inline struct sockaddr_in6 *svc_addr_in6(const struct svc_rqst *rqst) 332{ 333 return (struct sockaddr_in6 *) &rqst->rq_addr; 334} 335 336static inline struct sockaddr *svc_addr(const struct svc_rqst *rqst) 337{ 338 return (struct sockaddr *) &rqst->rq_addr; 339} 340 341static inline struct sockaddr_in *svc_daddr_in(const struct svc_rqst *rqst) 342{ 343 return (struct sockaddr_in *) &rqst->rq_daddr; 344} 345 346static inline struct sockaddr_in6 *svc_daddr_in6(const struct svc_rqst *rqst) 347{ 348 return (struct sockaddr_in6 *) &rqst->rq_daddr; 349} 350 351static inline struct sockaddr *svc_daddr(const struct svc_rqst *rqst) 352{ 353 return (struct sockaddr *) &rqst->rq_daddr; 354} 355 356/* 357 * Check buffer bounds after decoding arguments 358 */ 359static inline int 360xdr_argsize_check(struct svc_rqst *rqstp, __be32 *p) 361{ 362 char *cp = (char *)p; 363 struct kvec *vec = &rqstp->rq_arg.head[0]; 364 return cp >= (char*)vec->iov_base 365 && cp <= (char*)vec->iov_base + vec->iov_len; 366} 367 368static inline int 369xdr_ressize_check(struct svc_rqst *rqstp, __be32 *p) 370{ 371 struct kvec *vec = &rqstp->rq_res.head[0]; 372 char *cp = (char*)p; 373 374 vec->iov_len = cp - (char*)vec->iov_base; 375 376 return vec->iov_len <= PAGE_SIZE; 377} 378 379static inline void svc_free_res_pages(struct svc_rqst *rqstp) 380{ 381 while (rqstp->rq_next_page != rqstp->rq_respages) { 382 struct page **pp = --rqstp->rq_next_page; 383 if (*pp) { 384 put_page(*pp); 385 *pp = NULL; 386 } 387 } 388} 389 390struct svc_deferred_req { 391 u32 prot; /* protocol (UDP or TCP) */ 392 struct svc_xprt *xprt; 393 struct sockaddr_storage addr; /* where reply must go */ 394 size_t addrlen; 395 struct sockaddr_storage daddr; /* where reply must come from */ 396 size_t daddrlen; 397 void *xprt_ctxt; 398 struct cache_deferred_req handle; 399 int argslen; 400 __be32 args[]; 401}; 402 403struct svc_process_info { 404 union { 405 int (*dispatch)(struct svc_rqst *, __be32 *); 406 struct { 407 unsigned int lovers; 408 unsigned int hivers; 409 } mismatch; 410 }; 411}; 412 413/* 414 * List of RPC programs on the same transport endpoint 415 */ 416struct svc_program { 417 struct svc_program * pg_next; /* other programs (same xprt) */ 418 u32 pg_prog; /* program number */ 419 unsigned int pg_lovers; /* lowest version */ 420 unsigned int pg_hivers; /* highest version */ 421 unsigned int pg_nvers; /* number of versions */ 422 const struct svc_version **pg_vers; /* version array */ 423 char * pg_name; /* service name */ 424 char * pg_class; /* class name: services sharing authentication */ 425 struct svc_stat * pg_stats; /* rpc statistics */ 426 int (*pg_authenticate)(struct svc_rqst *); 427 __be32 (*pg_init_request)(struct svc_rqst *, 428 const struct svc_program *, 429 struct svc_process_info *); 430 int (*pg_rpcbind_set)(struct net *net, 431 const struct svc_program *, 432 u32 version, int family, 433 unsigned short proto, 434 unsigned short port); 435}; 436 437/* 438 * RPC program version 439 */ 440struct svc_version { 441 u32 vs_vers; /* version number */ 442 u32 vs_nproc; /* number of procedures */ 443 const struct svc_procedure *vs_proc; /* per-procedure info */ 444 unsigned int *vs_count; /* call counts */ 445 u32 vs_xdrsize; /* xdrsize needed for this version */ 446 447 /* Don't register with rpcbind */ 448 bool vs_hidden; 449 450 /* Don't care if the rpcbind registration fails */ 451 bool vs_rpcb_optnl; 452 453 /* Need xprt with congestion control */ 454 bool vs_need_cong_ctrl; 455 456 /* Dispatch function */ 457 int (*vs_dispatch)(struct svc_rqst *, __be32 *); 458}; 459 460/* 461 * RPC procedure info 462 */ 463struct svc_procedure { 464 /* process the request: */ 465 __be32 (*pc_func)(struct svc_rqst *); 466 /* XDR decode args: */ 467 bool (*pc_decode)(struct svc_rqst *rqstp, 468 struct xdr_stream *xdr); 469 /* XDR encode result: */ 470 bool (*pc_encode)(struct svc_rqst *rqstp, 471 struct xdr_stream *xdr); 472 /* XDR free result: */ 473 void (*pc_release)(struct svc_rqst *); 474 unsigned int pc_argsize; /* argument struct size */ 475 unsigned int pc_ressize; /* result struct size */ 476 unsigned int pc_cachetype; /* cache info (NFS) */ 477 unsigned int pc_xdrressize; /* maximum size of XDR reply */ 478 const char * pc_name; /* for display */ 479}; 480 481/* 482 * Function prototypes. 483 */ 484int svc_rpcb_setup(struct svc_serv *serv, struct net *net); 485void svc_rpcb_cleanup(struct svc_serv *serv, struct net *net); 486int svc_bind(struct svc_serv *serv, struct net *net); 487struct svc_serv *svc_create(struct svc_program *, unsigned int, 488 int (*threadfn)(void *data)); 489struct svc_rqst *svc_rqst_alloc(struct svc_serv *serv, 490 struct svc_pool *pool, int node); 491void svc_rqst_replace_page(struct svc_rqst *rqstp, 492 struct page *page); 493void svc_rqst_free(struct svc_rqst *); 494void svc_exit_thread(struct svc_rqst *); 495struct svc_serv * svc_create_pooled(struct svc_program *, unsigned int, 496 int (*threadfn)(void *data)); 497int svc_set_num_threads(struct svc_serv *, struct svc_pool *, int); 498int svc_pool_stats_open(struct svc_serv *serv, struct file *file); 499int svc_process(struct svc_rqst *); 500int bc_svc_process(struct svc_serv *, struct rpc_rqst *, 501 struct svc_rqst *); 502int svc_register(const struct svc_serv *, struct net *, const int, 503 const unsigned short, const unsigned short); 504 505void svc_wake_up(struct svc_serv *); 506void svc_reserve(struct svc_rqst *rqstp, int space); 507struct svc_pool *svc_pool_for_cpu(struct svc_serv *serv); 508char * svc_print_addr(struct svc_rqst *, char *, size_t); 509const char * svc_proc_name(const struct svc_rqst *rqstp); 510int svc_encode_result_payload(struct svc_rqst *rqstp, 511 unsigned int offset, 512 unsigned int length); 513unsigned int svc_fill_write_vector(struct svc_rqst *rqstp, 514 struct xdr_buf *payload); 515char *svc_fill_symlink_pathname(struct svc_rqst *rqstp, 516 struct kvec *first, void *p, 517 size_t total); 518__be32 svc_generic_init_request(struct svc_rqst *rqstp, 519 const struct svc_program *progp, 520 struct svc_process_info *procinfo); 521int svc_generic_rpcbind_set(struct net *net, 522 const struct svc_program *progp, 523 u32 version, int family, 524 unsigned short proto, 525 unsigned short port); 526int svc_rpcbind_set_version(struct net *net, 527 const struct svc_program *progp, 528 u32 version, int family, 529 unsigned short proto, 530 unsigned short port); 531 532#define RPC_MAX_ADDRBUFLEN (63U) 533 534/* 535 * When we want to reduce the size of the reserved space in the response 536 * buffer, we need to take into account the size of any checksum data that 537 * may be at the end of the packet. This is difficult to determine exactly 538 * for all cases without actually generating the checksum, so we just use a 539 * static value. 540 */ 541static inline void svc_reserve_auth(struct svc_rqst *rqstp, int space) 542{ 543 svc_reserve(rqstp, space + rqstp->rq_auth_slack); 544} 545 546/** 547 * svcxdr_init_decode - Prepare an xdr_stream for svc Call decoding 548 * @rqstp: controlling server RPC transaction context 549 * 550 */ 551static inline void svcxdr_init_decode(struct svc_rqst *rqstp) 552{ 553 struct xdr_stream *xdr = &rqstp->rq_arg_stream; 554 struct kvec *argv = rqstp->rq_arg.head; 555 556 xdr_init_decode(xdr, &rqstp->rq_arg, argv->iov_base, NULL); 557 xdr_set_scratch_page(xdr, rqstp->rq_scratch_page); 558} 559 560/** 561 * svcxdr_init_encode - Prepare an xdr_stream for svc Reply encoding 562 * @rqstp: controlling server RPC transaction context 563 * 564 */ 565static inline void svcxdr_init_encode(struct svc_rqst *rqstp) 566{ 567 struct xdr_stream *xdr = &rqstp->rq_res_stream; 568 struct xdr_buf *buf = &rqstp->rq_res; 569 struct kvec *resv = buf->head; 570 571 xdr_reset_scratch_buffer(xdr); 572 573 xdr->buf = buf; 574 xdr->iov = resv; 575 xdr->p = resv->iov_base + resv->iov_len; 576 xdr->end = resv->iov_base + PAGE_SIZE - rqstp->rq_auth_slack; 577 buf->len = resv->iov_len; 578 xdr->page_ptr = buf->pages - 1; 579 buf->buflen = PAGE_SIZE * (1 + rqstp->rq_page_end - buf->pages); 580 buf->buflen -= rqstp->rq_auth_slack; 581 xdr->rqst = NULL; 582} 583 584#endif /* SUNRPC_SVC_H */