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 /*
  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_maxconn;     /* max connections allowed or
  62                                                  * '0' causing max to be based
  63                                                  * on number of threads. */
  64
  65         unsigned int            sv_max_payload; /* datagram payload size */
  66         unsigned int            sv_max_mesg;    /* max_payload + 1 page for overheads */
  67         unsigned int            sv_xdrsize;     /* XDR buffer size */
  68         struct list_head        sv_permsocks;   /* all permanent sockets */
  69         struct list_head        sv_tempsocks;   /* all temporary sockets */
  70         int                     sv_tmpcnt;      /* count of temporary sockets */
  71         struct timer_list       sv_temptimer;   /* timer for aging temporary sockets */
  72         sa_family_t             sv_family;      /* listener's address family */
  73
  74         char *                  sv_name;        /* service name */
  75
  76         unsigned int            sv_nrpools;     /* number of thread pools */
  77         struct svc_pool *       sv_pools;       /* array of thread pools */
  78
  79         void                    (*sv_shutdown)(struct svc_serv *serv);
  80                                                 /* Callback to use when last thread
  81                                                  * exits.
  82                                                  */
  83
  84         struct module *         sv_module;      /* optional module to count when
  85                                                  * adding threads */
  86         svc_thread_fn           sv_function;    /* main function for threads */
  87 };
  88
  89 /*
  90  * We use sv_nrthreads as a reference count.  svc_destroy() drops
  91  * this refcount, so we need to bump it up around operations that
  92  * change the number of threads.  Horrible, but there it is.
  93  * Should be called with the BKL held.
  94  */
  95 static inline void svc_get(struct svc_serv *serv)
  96 {
  97         serv->sv_nrthreads++;
  98 }
  99
 100 /*
 101  * Maximum payload size supported by a kernel RPC server.
 102  * This is use to determine the max number of pages nfsd is
 103  * willing to return in a single READ operation.
 104  *
 105  * These happen to all be powers of 2, which is not strictly
 106  * necessary but helps enforce the real limitation, which is
 107  * that they should be multiples of PAGE_CACHE_SIZE.
 108  *
 109  * For UDP transports, a block plus NFS,RPC, and UDP headers
 110  * has to fit into the IP datagram limit of 64K.  The largest
 111  * feasible number for all known page sizes is probably 48K,
 112  * but we choose 32K here.  This is the same as the historical
 113  * Linux limit; someone who cares more about NFS/UDP performance
 114  * can test a larger number.
 115  *
 116  * For TCP transports we have more freedom.  A size of 1MB is
 117  * chosen to match the client limit.  Other OSes are known to
 118  * have larger limits, but those numbers are probably beyond
 119  * the point of diminishing returns.
 120  */
 121 #define RPCSVC_MAXPAYLOAD       (1*1024*1024u)
 122 #define RPCSVC_MAXPAYLOAD_TCP   RPCSVC_MAXPAYLOAD
 123 #define RPCSVC_MAXPAYLOAD_UDP   (32*1024u)
 124
 125 extern u32 svc_max_payload(const struct svc_rqst *rqstp);
 126
 127 /*
 128  * RPC Requsts and replies are stored in one or more pages.
 129  * We maintain an array of pages for each server thread.
 130  * Requests are copied into these pages as they arrive.  Remaining
 131  * pages are available to write the reply into.
 132  *
 133  * Pages are sent using ->sendpage so each server thread needs to
 134  * allocate more to replace those used in sending.  To help keep track
 135  * of these pages we have a receive list where all pages initialy live,
 136  * and a send list where pages are moved to when there are to be part
 137  * of a reply.
 138  *
 139  * We use xdr_buf for holding responses as it fits well with NFS
 140  * read responses (that have a header, and some data pages, and possibly
 141  * a tail) and means we can share some client side routines.
 142  *
 143  * The xdr_buf.head kvec always points to the first page in the rq_*pages
 144  * list.  The xdr_buf.pages pointer points to the second page on that
 145  * list.  xdr_buf.tail points to the end of the first page.
 146  * This assumes that the non-page part of an rpc reply will fit
 147  * in a page - NFSd ensures this.  lockd also has no trouble.
 148  *
 149  * Each request/reply pair can have at most one "payload", plus two pages,
 150  * one for the request, and one for the reply.
 151  * We using ->sendfile to return read data, we might need one extra page
 152  * if the request is not page-aligned.  So add another '1'.
 153  */
 154 #define RPCSVC_MAXPAGES         ((RPCSVC_MAXPAYLOAD+PAGE_SIZE-1)/PAGE_SIZE \
 155                                 + 2 + 1)
 156
 157 static inline u32 svc_getnl(struct kvec *iov)
 158 {
 159         __be32 val, *vp;
 160         vp = iov->iov_base;
 161         val = *vp++;
 162         iov->iov_base = (void*)vp;
 163         iov->iov_len -= sizeof(__be32);
 164         return ntohl(val);
 165 }
 166
 167 static inline void svc_putnl(struct kvec *iov, u32 val)
 168 {
 169         __be32 *vp = iov->iov_base + iov->iov_len;
 170         *vp = htonl(val);
 171         iov->iov_len += sizeof(__be32);
 172 }
 173
 174 static inline __be32 svc_getu32(struct kvec *iov)
 175 {
 176         __be32 val, *vp;
 177         vp = iov->iov_base;
 178         val = *vp++;
 179         iov->iov_base = (void*)vp;
 180         iov->iov_len -= sizeof(__be32);
 181         return val;
 182 }
 183
 184 static inline void svc_ungetu32(struct kvec *iov)
 185 {
 186         __be32 *vp = (__be32 *)iov->iov_base;
 187         iov->iov_base = (void *)(vp - 1);
 188         iov->iov_len += sizeof(*vp);
 189 }
 190
 191 static inline void svc_putu32(struct kvec *iov, __be32 val)
 192 {
 193         __be32 *vp = iov->iov_base + iov->iov_len;
 194         *vp = val;
 195         iov->iov_len += sizeof(__be32);
 196 }
 197
 198 union svc_addr_u {
 199     struct in_addr      addr;
 200     struct in6_addr     addr6;
 201 };
 202
 203 /*
 204  * The context of a single thread, including the request currently being
 205  * processed.
 206  */
 207 struct svc_rqst {
 208         struct list_head        rq_list;        /* idle list */
 209         struct list_head        rq_all;         /* all threads list */
 210         struct svc_xprt *       rq_xprt;        /* transport ptr */
 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         size_t                  rq_xprt_hlen;   /* xprt header len */
 224         struct xdr_buf          rq_arg;
 225         struct xdr_buf          rq_res;
 226         struct page *           rq_pages[RPCSVC_MAXPAGES];
 227         struct page *           *rq_respages;   /* points into rq_pages */
 228         int                     rq_resused;     /* number of pages used for result */
 229
 230         struct kvec             rq_vec[RPCSVC_MAXPAGES]; /* generally useful.. */
 231
 232         __be32                  rq_xid;         /* transmission id */
 233         u32                     rq_prog;        /* program number */
 234         u32                     rq_vers;        /* program version */
 235         u32                     rq_proc;        /* procedure number */
 236         u32                     rq_prot;        /* IP protocol */
 237         unsigned short
 238                                 rq_secure  : 1; /* secure port */
 239
 240         union svc_addr_u        rq_daddr;       /* dest addr of request
 241                                                  *  - reply from here */
 242
 243         void *                  rq_argp;        /* decoded arguments */
 244         void *                  rq_resp;        /* xdr'd results */
 245         void *                  rq_auth_data;   /* flavor-specific data */
 246
 247         int                     rq_reserved;    /* space on socket outq
 248                                                  * reserved for this request
 249                                                  */
 250
 251         struct cache_req        rq_chandle;     /* handle passed to caches for
 252                                                  * request delaying
 253                                                  */
 254         /* Catering to nfsd */
 255         struct auth_domain *    rq_client;      /* RPC peer info */
 256         struct auth_domain *    rq_gssclient;   /* "gss/"-style peer info */
 257         struct svc_cacherep *   rq_cacherep;    /* cache info */
 258         struct knfsd_fh *       rq_reffh;       /* Referrence filehandle, used to
 259                                                  * determine what device number
 260                                                  * to report (real or virtual)
 261                                                  */
 262         int                     rq_splice_ok;   /* turned off in gss privacy
 263                                                  * to prevent encrypting page
 264                                                  * cache pages */
 265         wait_queue_head_t       rq_wait;        /* synchronization */
 266         struct task_struct      *rq_task;       /* service thread */
 267 };
 268
 269 /*
 270  * Rigorous type checking on sockaddr type conversions
 271  */
 272 static inline struct sockaddr_in *svc_addr_in(const struct svc_rqst *rqst)
 273 {
 274         return (struct sockaddr_in *) &rqst->rq_addr;
 275 }
 276
 277 static inline struct sockaddr_in6 *svc_addr_in6(const struct svc_rqst *rqst)
 278 {
 279         return (struct sockaddr_in6 *) &rqst->rq_addr;
 280 }
 281
 282 static inline struct sockaddr *svc_addr(const struct svc_rqst *rqst)
 283 {
 284         return (struct sockaddr *) &rqst->rq_addr;
 285 }
 286
 287 /*
 288  * Check buffer bounds after decoding arguments
 289  */
 290 static inline int
 291 xdr_argsize_check(struct svc_rqst *rqstp, __be32 *p)
 292 {
 293         char *cp = (char *)p;
 294         struct kvec *vec = &rqstp->rq_arg.head[0];
 295         return cp >= (char*)vec->iov_base
 296                 && cp <= (char*)vec->iov_base + vec->iov_len;
 297 }
 298
 299 static inline int
 300 xdr_ressize_check(struct svc_rqst *rqstp, __be32 *p)
 301 {
 302         struct kvec *vec = &rqstp->rq_res.head[0];
 303         char *cp = (char*)p;
 304
 305         vec->iov_len = cp - (char*)vec->iov_base;
 306
 307         return vec->iov_len <= PAGE_SIZE;
 308 }
 309
 310 static inline void svc_free_res_pages(struct svc_rqst *rqstp)
 311 {
 312         while (rqstp->rq_resused) {
 313                 struct page **pp = (rqstp->rq_respages +
 314                                     --rqstp->rq_resused);
 315                 if (*pp) {
 316                         put_page(*pp);
 317                         *pp = NULL;
 318                 }
 319         }
 320 }
 321
 322 struct svc_deferred_req {
 323         u32                     prot;   /* protocol (UDP or TCP) */
 324         struct svc_xprt         *xprt;
 325         struct sockaddr_storage addr;   /* where reply must go */
 326         size_t                  addrlen;
 327         union svc_addr_u        daddr;  /* where reply must come from */
 328         struct cache_deferred_req handle;
 329         size_t                  xprt_hlen;
 330         int                     argslen;
 331         __be32                  args[0];
 332 };
 333
 334 /*
 335  * List of RPC programs on the same transport endpoint
 336  */
 337 struct svc_program {
 338         struct svc_program *    pg_next;        /* other programs (same xprt) */
 339         u32                     pg_prog;        /* program number */
 340         unsigned int            pg_lovers;      /* lowest version */
 341         unsigned int            pg_hivers;      /* lowest version */
 342         unsigned int            pg_nvers;       /* number of versions */
 343         struct svc_version **   pg_vers;        /* version array */
 344         char *                  pg_name;        /* service name */
 345         char *                  pg_class;       /* class name: services sharing authentication */
 346         struct svc_stat *       pg_stats;       /* rpc statistics */
 347         int                     (*pg_authenticate)(struct svc_rqst *);
 348 };
 349
 350 /*
 351  * RPC program version
 352  */
 353 struct svc_version {
 354         u32                     vs_vers;        /* version number */
 355         u32                     vs_nproc;       /* number of procedures */
 356         struct svc_procedure *  vs_proc;        /* per-procedure info */
 357         u32                     vs_xdrsize;     /* xdrsize needed for this version */
 358
 359         unsigned int            vs_hidden : 1;  /* Don't register with portmapper.
 360                                                  * Only used for nfsacl so far. */
 361
 362         /* Override dispatch function (e.g. when caching replies).
 363          * A return value of 0 means drop the request.
 364          * vs_dispatch == NULL means use default dispatcher.
 365          */
 366         int                     (*vs_dispatch)(struct svc_rqst *, __be32 *);
 367 };
 368
 369 /*
 370  * RPC procedure info
 371  */
 372 typedef __be32  (*svc_procfunc)(struct svc_rqst *, void *argp, void *resp);
 373 struct svc_procedure {
 374         svc_procfunc            pc_func;        /* process the request */
 375         kxdrproc_t              pc_decode;      /* XDR decode args */
 376         kxdrproc_t              pc_encode;      /* XDR encode result */
 377         kxdrproc_t              pc_release;     /* XDR free result */
 378         unsigned int            pc_argsize;     /* argument struct size */
 379         unsigned int            pc_ressize;     /* result struct size */
 380         unsigned int            pc_count;       /* call count */
 381         unsigned int            pc_cachetype;   /* cache info (NFS) */
 382         unsigned int            pc_xdrressize;  /* maximum size of XDR reply */
 383 };
 384
 385 /*
 386  * Function prototypes.
 387  */
 388 struct svc_serv *svc_create(struct svc_program *, unsigned int, sa_family_t,
 389                             void (*shutdown)(struct svc_serv *));
 390 struct svc_rqst *svc_prepare_thread(struct svc_serv *serv,
 391                                         struct svc_pool *pool);
 392 void               svc_exit_thread(struct svc_rqst *);
 393 struct svc_serv *  svc_create_pooled(struct svc_program *, unsigned int,
 394                         sa_family_t, void (*shutdown)(struct svc_serv *),
 395                         svc_thread_fn, struct module *);
 396 int                svc_set_num_threads(struct svc_serv *, struct svc_pool *, int);
 397 void               svc_destroy(struct svc_serv *);
 398 int                svc_process(struct svc_rqst *);
 399 int                svc_register(const struct svc_serv *, const unsigned short,
 400                                 const unsigned short);
 401
 402 void               svc_wake_up(struct svc_serv *);
 403 void               svc_reserve(struct svc_rqst *rqstp, int space);
 404 struct svc_pool *  svc_pool_for_cpu(struct svc_serv *serv, int cpu);
 405 char *             svc_print_addr(struct svc_rqst *, char *, size_t);
 406
 407 #define RPC_MAX_ADDRBUFLEN      (63U)
 408
 409 /*
 410  * When we want to reduce the size of the reserved space in the response
 411  * buffer, we need to take into account the size of any checksum data that
 412  * may be at the end of the packet. This is difficult to determine exactly
 413  * for all cases without actually generating the checksum, so we just use a
 414  * static value.
 415  */
 416 static inline void svc_reserve_auth(struct svc_rqst *rqstp, int space)
 417 {
 418         int added_space = 0;
 419
 420         if (rqstp->rq_authop->flavour)
 421                 added_space = RPC_MAX_AUTH_SIZE;
 422         svc_reserve(rqstp, space + added_space);
 423 }
 424
 425 #endif /* SUNRPC_SVC_H */