include/linux/sunrpc/svc.h

   1 /*
   2  * linux/include/linux/sunrpc/svc.h
   3  *
   4  * RPC server declarations.
   5  *
   6  * Copyright (C) 1995, 1996 Olaf Kirch <okir@monad.swb.de>
   7  */
   8
   9
  10 #ifndef SUNRPC_SVC_H
  11 #define SUNRPC_SVC_H
  12
  13 #include <linux/in.h>
  14 #include <linux/in6.h>
  15 #include <linux/sunrpc/types.h>
  16 #include <linux/sunrpc/xdr.h>
  17 #include <linux/sunrpc/auth.h>
  18 #include <linux/sunrpc/svcauth.h>
  19 #include <linux/wait.h>
  20 #include <linux/mm.h>
  21
  22 /*
  23  * This is the RPC server thread function prototype
  24  */
  25 typedef int             (*svc_thread_fn)(void *);
  26
  27 /* statistics for svc_pool structures */
  28 struct svc_pool_stats {
  29         atomic_long_t   packets;
  30         unsigned long   sockets_queued;
  31         atomic_long_t   threads_woken;
  32         atomic_long_t   threads_timedout;
  33 };
  34
  35 /*
  36  *
  37  * RPC service thread pool.
  38  *
  39  * Pool of threads and temporary sockets.  Generally there is only
  40  * a single one of these per RPC service, but on NUMA machines those
  41  * services that can benefit from it (i.e. nfs but not lockd) will
  42  * have one pool per NUMA node.  This optimisation reduces cross-
  43  * node traffic on multi-node NUMA NFS servers.
  44  */
  45 struct svc_pool {
  46         unsigned int            sp_id;          /* pool id; also node id on NUMA */
  47         spinlock_t              sp_lock;        /* protects all fields */
  48         struct list_head        sp_threads;     /* idle server threads */
  49         struct list_head        sp_sockets;     /* pending sockets */
  50         unsigned int            sp_nrthreads;   /* # of threads in pool */
  51         struct list_head        sp_all_threads; /* all server threads */
  52         struct svc_pool_stats   sp_stats;       /* statistics on pool operation */
  53 #define SP_TASK_PENDING         (0)             /* still work to do even if no
  54                                                  * xprt is queued. */
  55         unsigned long           sp_flags;
  56 } ____cacheline_aligned_in_smp;
  57
  58 /*
  59  * RPC service.
  60  *
  61  * An RPC service is a ``daemon,'' possibly multithreaded, which
  62  * receives and processes incoming RPC messages.
  63  * It has one or more transport sockets associated with it, and maintains
  64  * a list of idle threads waiting for input.
  65  *
  66  * We currently do not support more than one RPC program per daemon.
  67  */
  68 struct svc_serv {
  69         struct svc_program *    sv_program;     /* RPC program */
  70         struct svc_stat *       sv_stats;       /* RPC statistics */
  71         spinlock_t              sv_lock;
  72         unsigned int            sv_nrthreads;   /* # of server threads */
  73         unsigned int            sv_maxconn;     /* max connections allowed or
  74                                                  * '0' causing max to be based
  75                                                  * on number of threads. */
  76
  77         unsigned int            sv_max_payload; /* datagram payload size */
  78         unsigned int            sv_max_mesg;    /* max_payload + 1 page for overheads */
  79         unsigned int            sv_xdrsize;     /* XDR buffer size */
  80         struct list_head        sv_permsocks;   /* all permanent sockets */
  81         struct list_head        sv_tempsocks;   /* all temporary sockets */
  82         int                     sv_tmpcnt;      /* count of temporary sockets */
  83         struct timer_list       sv_temptimer;   /* timer for aging temporary sockets */
  84
  85         char *                  sv_name;        /* service name */
  86
  87         unsigned int            sv_nrpools;     /* number of thread pools */
  88         struct svc_pool *       sv_pools;       /* array of thread pools */
  89
  90         void                    (*sv_shutdown)(struct svc_serv *serv,
  91                                                struct net *net);
  92                                                 /* Callback to use when last thread
  93                                                  * exits.
  94                                                  */
  95
  96         struct module *         sv_module;      /* optional module to count when
  97                                                  * adding threads */
  98         svc_thread_fn           sv_function;    /* main function for threads */
  99 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
 100         struct list_head        sv_cb_list;     /* queue for callback requests
 101                                                  * that arrive over the same
 102                                                  * connection */
 103         spinlock_t              sv_cb_lock;     /* protects the svc_cb_list */
 104         wait_queue_head_t       sv_cb_waitq;    /* sleep here if there are no
 105                                                  * entries in the svc_cb_list */
 106         struct svc_xprt         *sv_bc_xprt;    /* callback on fore channel */
 107 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
 108 };
 109
 110 /*
 111  * We use sv_nrthreads as a reference count.  svc_destroy() drops
 112  * this refcount, so we need to bump it up around operations that
 113  * change the number of threads.  Horrible, but there it is.
 114  * Should be called with the BKL held.
 115  */
 116 static inline void svc_get(struct svc_serv *serv)
 117 {
 118         serv->sv_nrthreads++;
 119 }
 120
 121 /*
 122  * Maximum payload size supported by a kernel RPC server.
 123  * This is use to determine the max number of pages nfsd is
 124  * willing to return in a single READ operation.
 125  *
 126  * These happen to all be powers of 2, which is not strictly
 127  * necessary but helps enforce the real limitation, which is
 128  * that they should be multiples of PAGE_CACHE_SIZE.
 129  *
 130  * For UDP transports, a block plus NFS,RPC, and UDP headers
 131  * has to fit into the IP datagram limit of 64K.  The largest
 132  * feasible number for all known page sizes is probably 48K,
 133  * but we choose 32K here.  This is the same as the historical
 134  * Linux limit; someone who cares more about NFS/UDP performance
 135  * can test a larger number.
 136  *
 137  * For TCP transports we have more freedom.  A size of 1MB is
 138  * chosen to match the client limit.  Other OSes are known to
 139  * have larger limits, but those numbers are probably beyond
 140  * the point of diminishing returns.
 141  */
 142 #define RPCSVC_MAXPAYLOAD       (1*1024*1024u)
 143 #define RPCSVC_MAXPAYLOAD_TCP   RPCSVC_MAXPAYLOAD
 144 #define RPCSVC_MAXPAYLOAD_UDP   (32*1024u)
 145
 146 extern u32 svc_max_payload(const struct svc_rqst *rqstp);
 147
 148 /*
 149  * RPC Requsts and replies are stored in one or more pages.
 150  * We maintain an array of pages for each server thread.
 151  * Requests are copied into these pages as they arrive.  Remaining
 152  * pages are available to write the reply into.
 153  *
 154  * Pages are sent using ->sendpage so each server thread needs to
 155  * allocate more to replace those used in sending.  To help keep track
 156  * of these pages we have a receive list where all pages initialy live,
 157  * and a send list where pages are moved to when there are to be part
 158  * of a reply.
 159  *
 160  * We use xdr_buf for holding responses as it fits well with NFS
 161  * read responses (that have a header, and some data pages, and possibly
 162  * a tail) and means we can share some client side routines.
 163  *
 164  * The xdr_buf.head kvec always points to the first page in the rq_*pages
 165  * list.  The xdr_buf.pages pointer points to the second page on that
 166  * list.  xdr_buf.tail points to the end of the first page.
 167  * This assumes that the non-page part of an rpc reply will fit
 168  * in a page - NFSd ensures this.  lockd also has no trouble.
 169  *
 170  * Each request/reply pair can have at most one "payload", plus two pages,
 171  * one for the request, and one for the reply.
 172  * We using ->sendfile to return read data, we might need one extra page
 173  * if the request is not page-aligned.  So add another '1'.
 174  */
 175 #define RPCSVC_MAXPAGES         ((RPCSVC_MAXPAYLOAD+PAGE_SIZE-1)/PAGE_SIZE \
 176                                 + 2 + 1)
 177
 178 static inline u32 svc_getnl(struct kvec *iov)
 179 {
 180         __be32 val, *vp;
 181         vp = iov->iov_base;
 182         val = *vp++;
 183         iov->iov_base = (void*)vp;
 184         iov->iov_len -= sizeof(__be32);
 185         return ntohl(val);
 186 }
 187
 188 static inline void svc_putnl(struct kvec *iov, u32 val)
 189 {
 190         __be32 *vp = iov->iov_base + iov->iov_len;
 191         *vp = htonl(val);
 192         iov->iov_len += sizeof(__be32);
 193 }
 194
 195 static inline __be32 svc_getu32(struct kvec *iov)
 196 {
 197         __be32 val, *vp;
 198         vp = iov->iov_base;
 199         val = *vp++;
 200         iov->iov_base = (void*)vp;
 201         iov->iov_len -= sizeof(__be32);
 202         return val;
 203 }
 204
 205 static inline void svc_ungetu32(struct kvec *iov)
 206 {
 207         __be32 *vp = (__be32 *)iov->iov_base;
 208         iov->iov_base = (void *)(vp - 1);
 209         iov->iov_len += sizeof(*vp);
 210 }
 211
 212 static inline void svc_putu32(struct kvec *iov, __be32 val)
 213 {
 214         __be32 *vp = iov->iov_base + iov->iov_len;
 215         *vp = val;
 216         iov->iov_len += sizeof(__be32);
 217 }
 218
 219 /*
 220  * The context of a single thread, including the request currently being
 221  * processed.
 222  */
 223 struct svc_rqst {
 224         struct list_head        rq_list;        /* idle list */
 225         struct list_head        rq_all;         /* all threads list */
 226         struct rcu_head         rq_rcu_head;    /* for RCU deferred kfree */
 227         struct svc_xprt *       rq_xprt;        /* transport ptr */
 228
 229         struct sockaddr_storage rq_addr;        /* peer address */
 230         size_t                  rq_addrlen;
 231         struct sockaddr_storage rq_daddr;       /* dest addr of request
 232                                                  *  - reply from here */
 233         size_t                  rq_daddrlen;
 234
 235         struct svc_serv *       rq_server;      /* RPC service definition */
 236         struct svc_pool *       rq_pool;        /* thread pool */
 237         struct svc_procedure *  rq_procinfo;    /* procedure info */
 238         struct auth_ops *       rq_authop;      /* authentication flavour */
 239         struct svc_cred         rq_cred;        /* auth info */
 240         void *                  rq_xprt_ctxt;   /* transport specific context ptr */
 241         struct svc_deferred_req*rq_deferred;    /* deferred request we are replaying */
 242
 243         size_t                  rq_xprt_hlen;   /* xprt header len */
 244         struct xdr_buf          rq_arg;
 245         struct xdr_buf          rq_res;
 246         struct page *           rq_pages[RPCSVC_MAXPAGES];
 247         struct page *           *rq_respages;   /* points into rq_pages */
 248         struct page *           *rq_next_page; /* next reply page to use */
 249         struct page *           *rq_page_end;  /* one past the last page */
 250
 251         struct kvec             rq_vec[RPCSVC_MAXPAGES]; /* generally useful.. */
 252
 253         __be32                  rq_xid;         /* transmission id */
 254         u32                     rq_prog;        /* program number */
 255         u32                     rq_vers;        /* program version */
 256         u32                     rq_proc;        /* procedure number */
 257         u32                     rq_prot;        /* IP protocol */
 258         int                     rq_cachetype;   /* catering to nfsd */
 259 #define RQ_SECURE       (0)                     /* secure port */
 260 #define RQ_LOCAL        (1)                     /* local request */
 261 #define RQ_USEDEFERRAL  (2)                     /* use deferral */
 262 #define RQ_DROPME       (3)                     /* drop current reply */
 263 #define RQ_SPLICE_OK    (4)                     /* turned off in gss privacy
 264                                                  * to prevent encrypting page
 265                                                  * cache pages */
 266 #define RQ_VICTIM       (5)                     /* about to be shut down */
 267         unsigned long           rq_flags;       /* flags field */
 268
 269         void *                  rq_argp;        /* decoded arguments */
 270         void *                  rq_resp;        /* xdr'd results */
 271         void *                  rq_auth_data;   /* flavor-specific data */
 272         int                     rq_auth_slack;  /* extra space xdr code
 273                                                  * should leave in head
 274                                                  * for krb5i, krb5p.
 275                                                  */
 276         int                     rq_reserved;    /* space on socket outq
 277                                                  * reserved for this request
 278                                                  */
 279
 280         struct cache_req        rq_chandle;     /* handle passed to caches for
 281                                                  * request delaying
 282                                                  */
 283         /* Catering to nfsd */
 284         struct auth_domain *    rq_client;      /* RPC peer info */
 285         struct auth_domain *    rq_gssclient;   /* "gss/"-style peer info */
 286         struct svc_cacherep *   rq_cacherep;    /* cache info */
 287         struct task_struct      *rq_task;       /* service thread */
 288 };
 289
 290 #define SVC_NET(svc_rqst)       (svc_rqst->rq_xprt->xpt_net)
 291
 292 /*
 293  * Rigorous type checking on sockaddr type conversions
 294  */
 295 static inline struct sockaddr_in *svc_addr_in(const struct svc_rqst *rqst)
 296 {
 297         return (struct sockaddr_in *) &rqst->rq_addr;
 298 }
 299
 300 static inline struct sockaddr_in6 *svc_addr_in6(const struct svc_rqst *rqst)
 301 {
 302         return (struct sockaddr_in6 *) &rqst->rq_addr;
 303 }
 304
 305 static inline struct sockaddr *svc_addr(const struct svc_rqst *rqst)
 306 {
 307         return (struct sockaddr *) &rqst->rq_addr;
 308 }
 309
 310 static inline struct sockaddr_in *svc_daddr_in(const struct svc_rqst *rqst)
 311 {
 312         return (struct sockaddr_in *) &rqst->rq_daddr;
 313 }
 314
 315 static inline struct sockaddr_in6 *svc_daddr_in6(const struct svc_rqst *rqst)
 316 {
 317         return (struct sockaddr_in6 *) &rqst->rq_daddr;
 318 }
 319
 320 static inline struct sockaddr *svc_daddr(const struct svc_rqst *rqst)
 321 {
 322         return (struct sockaddr *) &rqst->rq_daddr;
 323 }
 324
 325 /*
 326  * Check buffer bounds after decoding arguments
 327  */
 328 static inline int
 329 xdr_argsize_check(struct svc_rqst *rqstp, __be32 *p)
 330 {
 331         char *cp = (char *)p;
 332         struct kvec *vec = &rqstp->rq_arg.head[0];
 333         return cp >= (char*)vec->iov_base
 334                 && cp <= (char*)vec->iov_base + vec->iov_len;
 335 }
 336
 337 static inline int
 338 xdr_ressize_check(struct svc_rqst *rqstp, __be32 *p)
 339 {
 340         struct kvec *vec = &rqstp->rq_res.head[0];
 341         char *cp = (char*)p;
 342
 343         vec->iov_len = cp - (char*)vec->iov_base;
 344
 345         return vec->iov_len <= PAGE_SIZE;
 346 }
 347
 348 static inline void svc_free_res_pages(struct svc_rqst *rqstp)
 349 {
 350         while (rqstp->rq_next_page != rqstp->rq_respages) {
 351                 struct page **pp = --rqstp->rq_next_page;
 352                 if (*pp) {
 353                         put_page(*pp);
 354                         *pp = NULL;
 355                 }
 356         }
 357 }
 358
 359 struct svc_deferred_req {
 360         u32                     prot;   /* protocol (UDP or TCP) */
 361         struct svc_xprt         *xprt;
 362         struct sockaddr_storage addr;   /* where reply must go */
 363         size_t                  addrlen;
 364         struct sockaddr_storage daddr;  /* where reply must come from */
 365         size_t                  daddrlen;
 366         struct cache_deferred_req handle;
 367         size_t                  xprt_hlen;
 368         int                     argslen;
 369         __be32                  args[0];
 370 };
 371
 372 /*
 373  * List of RPC programs on the same transport endpoint
 374  */
 375 struct svc_program {
 376         struct svc_program *    pg_next;        /* other programs (same xprt) */
 377         u32                     pg_prog;        /* program number */
 378         unsigned int            pg_lovers;      /* lowest version */
 379         unsigned int            pg_hivers;      /* highest version */
 380         unsigned int            pg_nvers;       /* number of versions */
 381         struct svc_version **   pg_vers;        /* version array */
 382         char *                  pg_name;        /* service name */
 383         char *                  pg_class;       /* class name: services sharing authentication */
 384         struct svc_stat *       pg_stats;       /* rpc statistics */
 385         int                     (*pg_authenticate)(struct svc_rqst *);
 386 };
 387
 388 /*
 389  * RPC program version
 390  */
 391 struct svc_version {
 392         u32                     vs_vers;        /* version number */
 393         u32                     vs_nproc;       /* number of procedures */
 394         struct svc_procedure *  vs_proc;        /* per-procedure info */
 395         u32                     vs_xdrsize;     /* xdrsize needed for this version */
 396
 397         unsigned int            vs_hidden : 1,  /* Don't register with portmapper.
 398                                                  * Only used for nfsacl so far. */
 399                                 vs_rpcb_optnl:1;/* Don't care the result of register.
 400                                                  * Only used for nfsv4. */
 401
 402         /* Override dispatch function (e.g. when caching replies).
 403          * A return value of 0 means drop the request.
 404          * vs_dispatch == NULL means use default dispatcher.
 405          */
 406         int                     (*vs_dispatch)(struct svc_rqst *, __be32 *);
 407 };
 408
 409 /*
 410  * RPC procedure info
 411  */
 412 typedef __be32  (*svc_procfunc)(struct svc_rqst *, void *argp, void *resp);
 413 struct svc_procedure {
 414         svc_procfunc            pc_func;        /* process the request */
 415         kxdrproc_t              pc_decode;      /* XDR decode args */
 416         kxdrproc_t              pc_encode;      /* XDR encode result */
 417         kxdrproc_t              pc_release;     /* XDR free result */
 418         unsigned int            pc_argsize;     /* argument struct size */
 419         unsigned int            pc_ressize;     /* result struct size */
 420         unsigned int            pc_count;       /* call count */
 421         unsigned int            pc_cachetype;   /* cache info (NFS) */
 422         unsigned int            pc_xdrressize;  /* maximum size of XDR reply */
 423 };
 424
 425 /*
 426  * Function prototypes.
 427  */
 428 int svc_rpcb_setup(struct svc_serv *serv, struct net *net);
 429 void svc_rpcb_cleanup(struct svc_serv *serv, struct net *net);
 430 int svc_bind(struct svc_serv *serv, struct net *net);
 431 struct svc_serv *svc_create(struct svc_program *, unsigned int,
 432                             void (*shutdown)(struct svc_serv *, struct net *net));
 433 struct svc_rqst *svc_prepare_thread(struct svc_serv *serv,
 434                                         struct svc_pool *pool, int node);
 435 void               svc_exit_thread(struct svc_rqst *);
 436 struct svc_serv *  svc_create_pooled(struct svc_program *, unsigned int,
 437                         void (*shutdown)(struct svc_serv *, struct net *net),
 438                         svc_thread_fn, struct module *);
 439 int                svc_set_num_threads(struct svc_serv *, struct svc_pool *, int);
 440 int                svc_pool_stats_open(struct svc_serv *serv, struct file *file);
 441 void               svc_destroy(struct svc_serv *);
 442 void               svc_shutdown_net(struct svc_serv *, struct net *);
 443 int                svc_process(struct svc_rqst *);
 444 int                bc_svc_process(struct svc_serv *, struct rpc_rqst *,
 445                         struct svc_rqst *);
 446 int                svc_register(const struct svc_serv *, struct net *, const int,
 447                                 const unsigned short, const unsigned short);
 448
 449 void               svc_wake_up(struct svc_serv *);
 450 void               svc_reserve(struct svc_rqst *rqstp, int space);
 451 struct svc_pool *  svc_pool_for_cpu(struct svc_serv *serv, int cpu);
 452 char *             svc_print_addr(struct svc_rqst *, char *, size_t);
 453
 454 #define RPC_MAX_ADDRBUFLEN      (63U)
 455
 456 /*
 457  * When we want to reduce the size of the reserved space in the response
 458  * buffer, we need to take into account the size of any checksum data that
 459  * may be at the end of the packet. This is difficult to determine exactly
 460  * for all cases without actually generating the checksum, so we just use a
 461  * static value.
 462  */
 463 static inline void svc_reserve_auth(struct svc_rqst *rqstp, int space)
 464 {
 465         svc_reserve(rqstp, space + rqstp->rq_auth_slack);
 466 }
 467
 468 #endif /* SUNRPC_SVC_H */