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 void            (*svc_thread_fn)(struct svc_rqst *);
  26
  27 /*
  28  *
  29  * RPC service thread pool.
  30  *
  31  * Pool of threads and temporary sockets.  Generally there is only
  32  * a single one of these per RPC service, but on NUMA machines those
  33  * services that can benefit from it (i.e. nfs but not lockd) will
  34  * have one pool per NUMA node.  This optimisation reduces cross-
  35  * node traffic on multi-node NUMA NFS servers.
  36  */
  37 struct svc_pool {
  38         unsigned int            sp_id;          /* pool id; also node id on NUMA */
  39         spinlock_t              sp_lock;        /* protects all fields */
  40         struct list_head        sp_threads;     /* idle server threads */
  41         struct list_head        sp_sockets;     /* pending sockets */
  42         unsigned int            sp_nrthreads;   /* # of threads in pool */
  43         struct list_head        sp_all_threads; /* all server threads */
  44 } ____cacheline_aligned_in_smp;
  45
  46 /*
  47  * RPC service.
  48  *
  49  * An RPC service is a ``daemon,'' possibly multithreaded, which
  50  * receives and processes incoming RPC messages.
  51  * It has one or more transport sockets associated with it, and maintains
  52  * a list of idle threads waiting for input.
  53  *
  54  * We currently do not support more than one RPC program per daemon.
  55  */
  56 struct svc_serv {
  57         struct svc_program *    sv_program;     /* RPC program */
  58         struct svc_stat *       sv_stats;       /* RPC statistics */
  59         spinlock_t              sv_lock;
  60         unsigned int            sv_nrthreads;   /* # of server threads */
  61         unsigned int            sv_max_payload; /* datagram payload size */
  62         unsigned int            sv_max_mesg;    /* max_payload + 1 page for overheads */
  63         unsigned int            sv_xdrsize;     /* XDR buffer size */
  64
  65         struct list_head        sv_permsocks;   /* all permanent sockets */
  66         struct list_head        sv_tempsocks;   /* all temporary sockets */
  67         int                     sv_tmpcnt;      /* count of temporary sockets */
  68         struct timer_list       sv_temptimer;   /* timer for aging temporary sockets */
  69
  70         char *                  sv_name;        /* service name */
  71
  72         unsigned int            sv_nrpools;     /* number of thread pools */
  73         struct svc_pool *       sv_pools;       /* array of thread pools */
  74
  75         void                    (*sv_shutdown)(struct svc_serv *serv);
  76                                                 /* Callback to use when last thread
  77                                                  * exits.
  78                                                  */
  79
  80         struct module *         sv_module;      /* optional module to count when
  81                                                  * adding threads */
  82         svc_thread_fn           sv_function;    /* main function for threads */
  83         int                     sv_kill_signal; /* signal to kill threads */
  84 };
  85
  86 /*
  87  * We use sv_nrthreads as a reference count.  svc_destroy() drops
  88  * this refcount, so we need to bump it up around operations that
  89  * change the number of threads.  Horrible, but there it is.
  90  * Should be called with the BKL held.
  91  */
  92 static inline void svc_get(struct svc_serv *serv)
  93 {
  94         serv->sv_nrthreads++;
  95 }
  96
  97 /*
  98  * Maximum payload size supported by a kernel RPC server.
  99  * This is use to determine the max number of pages nfsd is
 100  * willing to return in a single READ operation.
 101  *
 102  * These happen to all be powers of 2, which is not strictly
 103  * necessary but helps enforce the real limitation, which is
 104  * that they should be multiples of PAGE_CACHE_SIZE.
 105  *
 106  * For UDP transports, a block plus NFS,RPC, and UDP headers
 107  * has to fit into the IP datagram limit of 64K.  The largest
 108  * feasible number for all known page sizes is probably 48K,
 109  * but we choose 32K here.  This is the same as the historical
 110  * Linux limit; someone who cares more about NFS/UDP performance
 111  * can test a larger number.
 112  *
 113  * For TCP transports we have more freedom.  A size of 1MB is
 114  * chosen to match the client limit.  Other OSes are known to
 115  * have larger limits, but those numbers are probably beyond
 116  * the point of diminishing returns.
 117  */
 118 #define RPCSVC_MAXPAYLOAD       (1*1024*1024u)
 119 #define RPCSVC_MAXPAYLOAD_TCP   RPCSVC_MAXPAYLOAD
 120 #define RPCSVC_MAXPAYLOAD_UDP   (32*1024u)
 121
 122 extern u32 svc_max_payload(const struct svc_rqst *rqstp);
 123
 124 /*
 125  * RPC Requsts and replies are stored in one or more pages.
 126  * We maintain an array of pages for each server thread.
 127  * Requests are copied into these pages as they arrive.  Remaining
 128  * pages are available to write the reply into.
 129  *
 130  * Pages are sent using ->sendpage so each server thread needs to
 131  * allocate more to replace those used in sending.  To help keep track
 132  * of these pages we have a receive list where all pages initialy live,
 133  * and a send list where pages are moved to when there are to be part
 134  * of a reply.
 135  *
 136  * We use xdr_buf for holding responses as it fits well with NFS
 137  * read responses (that have a header, and some data pages, and possibly
 138  * a tail) and means we can share some client side routines.
 139  *
 140  * The xdr_buf.head kvec always points to the first page in the rq_*pages
 141  * list.  The xdr_buf.pages pointer points to the second page on that
 142  * list.  xdr_buf.tail points to the end of the first page.
 143  * This assumes that the non-page part of an rpc reply will fit
 144  * in a page - NFSd ensures this.  lockd also has no trouble.
 145  *
 146  * Each request/reply pair can have at most one "payload", plus two pages,
 147  * one for the request, and one for the reply.
 148  * We using ->sendfile to return read data, we might need one extra page
 149  * if the request is not page-aligned.  So add another '1'.
 150  */
 151 #define RPCSVC_MAXPAGES         ((RPCSVC_MAXPAYLOAD+PAGE_SIZE-1)/PAGE_SIZE \
 152                                 + 2 + 1)
 153
 154 static inline u32 svc_getnl(struct kvec *iov)
 155 {
 156         __be32 val, *vp;
 157         vp = iov->iov_base;
 158         val = *vp++;
 159         iov->iov_base = (void*)vp;
 160         iov->iov_len -= sizeof(__be32);
 161         return ntohl(val);
 162 }
 163
 164 static inline void svc_putnl(struct kvec *iov, u32 val)
 165 {
 166         __be32 *vp = iov->iov_base + iov->iov_len;
 167         *vp = htonl(val);
 168         iov->iov_len += sizeof(__be32);
 169 }
 170
 171 static inline __be32 svc_getu32(struct kvec *iov)
 172 {
 173         __be32 val, *vp;
 174         vp = iov->iov_base;
 175         val = *vp++;
 176         iov->iov_base = (void*)vp;
 177         iov->iov_len -= sizeof(__be32);
 178         return val;
 179 }
 180
 181 static inline void svc_ungetu32(struct kvec *iov)
 182 {
 183         __be32 *vp = (__be32 *)iov->iov_base;
 184         iov->iov_base = (void *)(vp - 1);
 185         iov->iov_len += sizeof(*vp);
 186 }
 187
 188 static inline void svc_putu32(struct kvec *iov, __be32 val)
 189 {
 190         __be32 *vp = iov->iov_base + iov->iov_len;
 191         *vp = val;
 192         iov->iov_len += sizeof(__be32);
 193 }
 194
 195 union svc_addr_u {
 196     struct in_addr      addr;
 197     struct in6_addr     addr6;
 198 };
 199
 200 /*
 201  * The context of a single thread, including the request currently being
 202  * processed.
 203  */
 204 struct svc_rqst {
 205         struct list_head        rq_list;        /* idle list */
 206         struct list_head        rq_all;         /* all threads list */
 207         union {
 208                 struct svc_xprt *       rq_xprt;        /* transport ptr */
 209                 struct svc_sock *       rq_sock;        /* socket ptr */
 210         };
 211         struct sockaddr_storage rq_addr;        /* peer address */
 212         size_t                  rq_addrlen;
 213
 214         struct svc_serv *       rq_server;      /* RPC service definition */
 215         struct svc_pool *       rq_pool;        /* thread pool */
 216         struct svc_procedure *  rq_procinfo;    /* procedure info */
 217         struct auth_ops *       rq_authop;      /* authentication flavour */
 218         u32                     rq_flavor;      /* pseudoflavor */
 219         struct svc_cred         rq_cred;        /* auth info */
 220         void *                  rq_xprt_ctxt;   /* transport specific context ptr */
 221         struct svc_deferred_req*rq_deferred;    /* deferred request we are replaying */
 222
 223         struct xdr_buf          rq_arg;
 224         struct xdr_buf          rq_res;
 225         struct page *           rq_pages[RPCSVC_MAXPAGES];
 226         struct page *           *rq_respages;   /* points into rq_pages */
 227         int                     rq_resused;     /* number of pages used for result */
 228
 229         struct kvec             rq_vec[RPCSVC_MAXPAGES]; /* generally useful.. */
 230
 231         __be32                  rq_xid;         /* transmission id */
 232         u32                     rq_prog;        /* program number */
 233         u32                     rq_vers;        /* program version */
 234         u32                     rq_proc;        /* procedure number */
 235         u32                     rq_prot;        /* IP protocol */
 236         unsigned short
 237                                 rq_secure  : 1; /* secure port */
 238
 239         union svc_addr_u        rq_daddr;       /* dest addr of request
 240                                                  *  - reply from here */
 241
 242         void *                  rq_argp;        /* decoded arguments */
 243         void *                  rq_resp;        /* xdr'd results */
 244         void *                  rq_auth_data;   /* flavor-specific data */
 245
 246         int                     rq_reserved;    /* space on socket outq
 247                                                  * reserved for this request
 248                                                  */
 249
 250         struct cache_req        rq_chandle;     /* handle passed to caches for
 251                                                  * request delaying
 252                                                  */
 253         /* Catering to nfsd */
 254         struct auth_domain *    rq_client;      /* RPC peer info */
 255         struct auth_domain *    rq_gssclient;   /* "gss/"-style peer info */
 256         struct svc_cacherep *   rq_cacherep;    /* cache info */
 257         struct knfsd_fh *       rq_reffh;       /* Referrence filehandle, used to
 258                                                  * determine what device number
 259                                                  * to report (real or virtual)
 260                                                  */
 261         int                     rq_splice_ok;   /* turned off in gss privacy
 262                                                  * to prevent encrypting page
 263                                                  * cache pages */
 264         wait_queue_head_t       rq_wait;        /* synchronization */
 265         struct task_struct      *rq_task;       /* service thread */
 266 };
 267
 268 /*
 269  * Rigorous type checking on sockaddr type conversions
 270  */
 271 static inline struct sockaddr_in *svc_addr_in(struct svc_rqst *rqst)
 272 {
 273         return (struct sockaddr_in *) &rqst->rq_addr;
 274 }
 275
 276 static inline struct sockaddr_in6 *svc_addr_in6(struct svc_rqst *rqst)
 277 {
 278         return (struct sockaddr_in6 *) &rqst->rq_addr;
 279 }
 280
 281 static inline struct sockaddr *svc_addr(struct svc_rqst *rqst)
 282 {
 283         return (struct sockaddr *) &rqst->rq_addr;
 284 }
 285
 286 /*
 287  * Check buffer bounds after decoding arguments
 288  */
 289 static inline int
 290 xdr_argsize_check(struct svc_rqst *rqstp, __be32 *p)
 291 {
 292         char *cp = (char *)p;
 293         struct kvec *vec = &rqstp->rq_arg.head[0];
 294         return cp >= (char*)vec->iov_base
 295                 && cp <= (char*)vec->iov_base + vec->iov_len;
 296 }
 297
 298 static inline int
 299 xdr_ressize_check(struct svc_rqst *rqstp, __be32 *p)
 300 {
 301         struct kvec *vec = &rqstp->rq_res.head[0];
 302         char *cp = (char*)p;
 303
 304         vec->iov_len = cp - (char*)vec->iov_base;
 305
 306         return vec->iov_len <= PAGE_SIZE;
 307 }
 308
 309 static inline void svc_free_res_pages(struct svc_rqst *rqstp)
 310 {
 311         while (rqstp->rq_resused) {
 312                 struct page **pp = (rqstp->rq_respages +
 313                                     --rqstp->rq_resused);
 314                 if (*pp) {
 315                         put_page(*pp);
 316                         *pp = NULL;
 317                 }
 318         }
 319 }
 320
 321 struct svc_deferred_req {
 322         u32                     prot;   /* protocol (UDP or TCP) */
 323         struct svc_xprt         *xprt;
 324         struct sockaddr_storage addr;   /* where reply must go */
 325         size_t                  addrlen;
 326         union svc_addr_u        daddr;  /* where reply must come from */
 327         struct cache_deferred_req handle;
 328         int                     argslen;
 329         __be32                  args[0];
 330 };
 331
 332 /*
 333  * List of RPC programs on the same transport endpoint
 334  */
 335 struct svc_program {
 336         struct svc_program *    pg_next;        /* other programs (same xprt) */
 337         u32                     pg_prog;        /* program number */
 338         unsigned int            pg_lovers;      /* lowest version */
 339         unsigned int            pg_hivers;      /* lowest version */
 340         unsigned int            pg_nvers;       /* number of versions */
 341         struct svc_version **   pg_vers;        /* version array */
 342         char *                  pg_name;        /* service name */
 343         char *                  pg_class;       /* class name: services sharing authentication */
 344         struct svc_stat *       pg_stats;       /* rpc statistics */
 345         int                     (*pg_authenticate)(struct svc_rqst *);
 346 };
 347
 348 /*
 349  * RPC program version
 350  */
 351 struct svc_version {
 352         u32                     vs_vers;        /* version number */
 353         u32                     vs_nproc;       /* number of procedures */
 354         struct svc_procedure *  vs_proc;        /* per-procedure info */
 355         u32                     vs_xdrsize;     /* xdrsize needed for this version */
 356
 357         unsigned int            vs_hidden : 1;  /* Don't register with portmapper.
 358                                                  * Only used for nfsacl so far. */
 359
 360         /* Override dispatch function (e.g. when caching replies).
 361          * A return value of 0 means drop the request.
 362          * vs_dispatch == NULL means use default dispatcher.
 363          */
 364         int                     (*vs_dispatch)(struct svc_rqst *, __be32 *);
 365 };
 366
 367 /*
 368  * RPC procedure info
 369  */
 370 typedef __be32  (*svc_procfunc)(struct svc_rqst *, void *argp, void *resp);
 371 struct svc_procedure {
 372         svc_procfunc            pc_func;        /* process the request */
 373         kxdrproc_t              pc_decode;      /* XDR decode args */
 374         kxdrproc_t              pc_encode;      /* XDR encode result */
 375         kxdrproc_t              pc_release;     /* XDR free result */
 376         unsigned int            pc_argsize;     /* argument struct size */
 377         unsigned int            pc_ressize;     /* result struct size */
 378         unsigned int            pc_count;       /* call count */
 379         unsigned int            pc_cachetype;   /* cache info (NFS) */
 380         unsigned int            pc_xdrressize;  /* maximum size of XDR reply */
 381 };
 382
 383 /*
 384  * Function prototypes.
 385  */
 386 struct svc_serv *  svc_create(struct svc_program *, unsigned int,
 387                               void (*shutdown)(struct svc_serv*));
 388 int                svc_create_thread(svc_thread_fn, struct svc_serv *);
 389 void               svc_exit_thread(struct svc_rqst *);
 390 struct svc_serv *  svc_create_pooled(struct svc_program *, unsigned int,
 391                         void (*shutdown)(struct svc_serv*),
 392                         svc_thread_fn, int sig, struct module *);
 393 int                svc_set_num_threads(struct svc_serv *, struct svc_pool *, int);
 394 void               svc_destroy(struct svc_serv *);
 395 int                svc_process(struct svc_rqst *);
 396 int                svc_register(struct svc_serv *, int, unsigned short);
 397 void               svc_wake_up(struct svc_serv *);
 398 void               svc_reserve(struct svc_rqst *rqstp, int space);
 399 struct svc_pool *  svc_pool_for_cpu(struct svc_serv *serv, int cpu);
 400 char *             svc_print_addr(struct svc_rqst *, char *, size_t);
 401
 402 #define RPC_MAX_ADDRBUFLEN      (63U)
 403
 404 /*
 405  * When we want to reduce the size of the reserved space in the response
 406  * buffer, we need to take into account the size of any checksum data that
 407  * may be at the end of the packet. This is difficult to determine exactly
 408  * for all cases without actually generating the checksum, so we just use a
 409  * static value.
 410  */
 411 static inline void
 412 svc_reserve_auth(struct svc_rqst *rqstp, int space)
 413 {
 414         int                     added_space = 0;
 415
 416         switch(rqstp->rq_authop->flavour) {
 417                 case RPC_AUTH_GSS:
 418                         added_space = RPC_MAX_AUTH_SIZE;
 419         }
 420         return svc_reserve(rqstp, space + added_space);
 421 }
 422
 423 #endif /* SUNRPC_SVC_H */